[PATCH for-next 00/20] RDMA/hns: Add hip08 RoCE driver support

[PATCH for-next 00/20] RDMA/hns: Add hip08 RoCE driver support

[Wei Hu (Xavier)]

This patchset adds support for the HiSilicon RoCE engine in the hip08
SoC.

The driver is refactored so it can support both platform and pci device.
And hip08 RoCE engine is a pci device, hip06 RoCE engine is a platform
device.

New hardware layer file hns_roce_hw_v2.c is added for hw v2 support,
which also includes pci device probing and initialization.

Common functionality is still in hns-roce driver, along with device
initialization.

This patchset depends on hns3 NIC driver which had been accepted by David,
and is based on for-next branch of the repo:
https://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma.git

Shaobo Xu (3):
  RDMA/hns: Add the interfaces to support multi hop addressing for the
    contexts in hip08
  RDMA/hns: Update the interfaces for MTT/CQE multi hop addressing in
    hip08
  RDMA/hns: Split CQE from MTT in hip08

Wei Hu (Xavier) (17):
  RDMA/hns: Split hw v1 driver from hns roce driver
  RDMA/hns: Move priv in order to add multiple hns_roce support
  RDMA/hns: Initialize the PCI device for hip08 RoCE
  RDMA/hns: Modify assignment device variable to support both PCI device
    and platform device
  RDMA/hns: Add command queue support for hip08 RoCE driver
  RDMA/hns: Add profile support for hip08 driver
  RDMA/hns: Add mailbox's implementation for hip08 RoCE driver
  RDMA/hns: Configure BT BA and BT attribute for the contexts in hip08
  RDMA/hns: Support multi hop addressing for PBL in hip08
  RDMA/hns: Configure mac&gid and user access region for hip08 RoCE
    driver
  RDMA/hns: Add CQ operations support for hip08 RoCE driver
  RDMA/hns: Add QP operations support for hip08 SoC
  RDMA/hns: Add support for processing send wr and receive wr
  RDMA/hns: Configure the MTPT in hip08
  RDMA/hns: Add releasing resource operation in error branch
  RDMA/hns: Replace condition statement using hardware version
    information
  RDMA/hns: Fix inconsistent warning

 drivers/infiniband/hw/hns/Kconfig           |   25 +-
 drivers/infiniband/hw/hns/Makefile          |    8 +-
 drivers/infiniband/hw/hns/hns_roce_ah.c     |    2 +-
 drivers/infiniband/hw/hns/hns_roce_alloc.c  |    8 +-
 drivers/infiniband/hw/hns/hns_roce_cmd.c    |  107 +-
 drivers/infiniband/hw/hns/hns_roce_cmd.h    |   50 +
 drivers/infiniband/hw/hns/hns_roce_common.h |   23 +
 drivers/infiniband/hw/hns/hns_roce_cq.c     |   72 +-
 drivers/infiniband/hw/hns/hns_roce_device.h |   98 +-
 drivers/infiniband/hw/hns/hns_roce_hem.c    |  699 +++++-
 drivers/infiniband/hw/hns/hns_roce_hem.h    |   32 +-
 drivers/infiniband/hw/hns/hns_roce_hw_v1.c  |  401 +++-
 drivers/infiniband/hw/hns/hns_roce_hw_v1.h  |    5 +
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 3128 +++++++++++++++++++++++++++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  | 1165 ++++++++++
 drivers/infiniband/hw/hns/hns_roce_main.c   |  324 +--
 drivers/infiniband/hw/hns/hns_roce_mr.c     |  513 ++++-
 drivers/infiniband/hw/hns/hns_roce_pd.c     |   20 +-
 drivers/infiniband/hw/hns/hns_roce_qp.c     |  179 +-
 19 files changed, 6318 insertions(+), 541 deletions(-)
 create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.c
 create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.h

-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 01/20] RDMA/hns: Split hw v1 driver from hns roce driver

[Wei Hu (Xavier)]

The hardware relevant definitions and operations are implemented
in hns_roce_hw_v* file. According to the diversity chips, the file
is named as hns_roce_hw_v1.c or hns_roce_hw_v2.c etc.

The general software process flow, common structures and allocated
algorithms are implemented in other files located in hns roce driver.

In order to support the scalability of the hardware version, the
common driver features are in the hns-roce.ko, and the hardware
relevant operations are in hns_roce_hw_v1.ko or hns_roce_hw_v2.ko
based on the series chips.

Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/Kconfig           |  15 +-
 drivers/infiniband/hw/hns/Makefile          |   4 +-
 drivers/infiniband/hw/hns/hns_roce_alloc.c  |   2 +
 drivers/infiniband/hw/hns/hns_roce_cmd.c    |   3 +
 drivers/infiniband/hw/hns/hns_roce_cq.c     |   3 +
 drivers/infiniband/hw/hns/hns_roce_device.h |   6 +-
 drivers/infiniband/hw/hns/hns_roce_hem.c    |   1 +
 drivers/infiniband/hw/hns/hns_roce_hw_v1.c  | 244 +++++++++++++++++++++++-
 drivers/infiniband/hw/hns/hns_roce_main.c   | 282 +++++-----------------------
 drivers/infiniband/hw/hns/hns_roce_mr.c     |   3 +
 drivers/infiniband/hw/hns/hns_roce_pd.c     |   2 +
 drivers/infiniband/hw/hns/hns_roce_qp.c     |  12 ++
 12 files changed, 330 insertions(+), 247 deletions(-)

diff --git a/drivers/infiniband/hw/hns/Kconfig b/drivers/infiniband/hw/hns/Kconfig
index 61c93bb..37e33e5 100644
--- a/drivers/infiniband/hw/hns/Kconfig
+++ b/drivers/infiniband/hw/hns/Kconfig
@@ -1,10 +1,21 @@
 config INFINIBAND_HNS
 	tristate "HNS RoCE Driver"
 	depends on NET_VENDOR_HISILICON
-	depends on (ARM64 || (COMPILE_TEST && 64BIT)) && HNS && HNS_DSAF && HNS_ENET
+	depends on ARM64 || (COMPILE_TEST && 64BIT)
 	---help---
 	  This is a RoCE/RDMA driver for the Hisilicon RoCE engine. The engine
-	  is used in Hisilicon Hi1610 and more further ICT SoC.
+	  is used in Hisilicon Hip06 and more further ICT SoC based on
+	  platform device.
 
 	  To compile this driver as a module, choose M here: the module
 	  will be called hns-roce.
+
+config INFINIBAND_HNS_HIP06
+	tristate "Hisilicon Hip06 Family RoCE support"
+	depends on INFINIBAND_HNS && HNS && HNS_DSAF && HNS_ENET
+	---help---
+	  RoCE driver support for Hisilicon RoCE engine in Hisilicon Hip06 and
+	  Hip07 SoC. These RoCE engines are platform devices.
+
+	  To compile this driver as a module, choose M here: the module
+	  will be called hns-roce-hw-v1.
diff --git a/drivers/infiniband/hw/hns/Makefile b/drivers/infiniband/hw/hns/Makefile
index 7e8ebd2..806ac4e 100644
--- a/drivers/infiniband/hw/hns/Makefile
+++ b/drivers/infiniband/hw/hns/Makefile
@@ -5,4 +5,6 @@
 obj-$(CONFIG_INFINIBAND_HNS) += hns-roce.o
 hns-roce-objs := hns_roce_main.o hns_roce_cmd.o hns_roce_eq.o hns_roce_pd.o \
 	hns_roce_ah.o hns_roce_hem.o hns_roce_mr.o hns_roce_qp.o \
-	hns_roce_cq.o hns_roce_alloc.o hns_roce_hw_v1.o
+	hns_roce_cq.o hns_roce_alloc.o
+obj-$(CONFIG_INFINIBAND_HNS_HIP06) += hns-roce-hw-v1.o
+hns-roce-hw-v1-objs := hns_roce_hw_v1.o
diff --git a/drivers/infiniband/hw/hns/hns_roce_alloc.c b/drivers/infiniband/hw/hns/hns_roce_alloc.c
index e1b433c..71a5098 100644
--- a/drivers/infiniband/hw/hns/hns_roce_alloc.c
+++ b/drivers/infiniband/hw/hns/hns_roce_alloc.c
@@ -67,6 +67,7 @@ void hns_roce_bitmap_free(struct hns_roce_bitmap *bitmap, unsigned long obj,
 {
 	hns_roce_bitmap_free_range(bitmap, obj, 1, rr);
 }
+EXPORT_SYMBOL_GPL(hns_roce_bitmap_free);
 
 int hns_roce_bitmap_alloc_range(struct hns_roce_bitmap *bitmap, int cnt,
 				int align, unsigned long *obj)
@@ -177,6 +178,7 @@ void hns_roce_buf_free(struct hns_roce_dev *hr_dev, u32 size,
 		kfree(buf->page_list);
 	}
 }
+EXPORT_SYMBOL_GPL(hns_roce_buf_free);
 
 int hns_roce_buf_alloc(struct hns_roce_dev *hr_dev, u32 size, u32 max_direct,
 		       struct hns_roce_buf *buf)
diff --git a/drivers/infiniband/hw/hns/hns_roce_cmd.c b/drivers/infiniband/hw/hns/hns_roce_cmd.c
index b94dcd8..c981d9d 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cmd.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cmd.c
@@ -269,6 +269,7 @@ int hns_roce_cmd_mbox(struct hns_roce_dev *hr_dev, u64 in_param, u64 out_param,
 					      in_modifier, op_modifier, op,
 					      timeout);
 }
+EXPORT_SYMBOL_GPL(hns_roce_cmd_mbox);
 
 int hns_roce_cmd_init(struct hns_roce_dev *hr_dev)
 {
@@ -356,6 +357,7 @@ struct hns_roce_cmd_mailbox
 
 	return mailbox;
 }
+EXPORT_SYMBOL_GPL(hns_roce_alloc_cmd_mailbox);
 
 void hns_roce_free_cmd_mailbox(struct hns_roce_dev *hr_dev,
 			       struct hns_roce_cmd_mailbox *mailbox)
@@ -366,3 +368,4 @@ void hns_roce_free_cmd_mailbox(struct hns_roce_dev *hr_dev,
 	dma_pool_free(hr_dev->cmd.pool, mailbox->buf, mailbox->dma);
 	kfree(mailbox);
 }
+EXPORT_SYMBOL_GPL(hns_roce_free_cmd_mailbox);
diff --git a/drivers/infiniband/hw/hns/hns_roce_cq.c b/drivers/infiniband/hw/hns/hns_roce_cq.c
index b89fd71..1d57df7 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cq.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cq.c
@@ -205,6 +205,7 @@ void hns_roce_free_cq(struct hns_roce_dev *hr_dev, struct hns_roce_cq *hr_cq)
 	hns_roce_table_put(hr_dev, &cq_table->table, hr_cq->cqn);
 	hns_roce_bitmap_free(&cq_table->bitmap, hr_cq->cqn, BITMAP_NO_RR);
 }
+EXPORT_SYMBOL_GPL(hns_roce_free_cq);
 
 static int hns_roce_ib_get_cq_umem(struct hns_roce_dev *hr_dev,
 				   struct ib_ucontext *context,
@@ -385,6 +386,7 @@ struct ib_cq *hns_roce_ib_create_cq(struct ib_device *ib_dev,
 	kfree(hr_cq);
 	return ERR_PTR(ret);
 }
+EXPORT_SYMBOL_GPL(hns_roce_ib_create_cq);
 
 int hns_roce_ib_destroy_cq(struct ib_cq *ib_cq)
 {
@@ -410,6 +412,7 @@ int hns_roce_ib_destroy_cq(struct ib_cq *ib_cq)
 
 	return ret;
 }
+EXPORT_SYMBOL_GPL(hns_roce_ib_destroy_cq);
 
 void hns_roce_cq_completion(struct hns_roce_dev *hr_dev, u32 cqn)
 {
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index e493a61..fc2b82d 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -571,7 +571,7 @@ struct hns_roce_dev {
 	int			loop_idc;
 	dma_addr_t		tptr_dma_addr; /*only for hw v1*/
 	u32			tptr_size; /*only for hw v1*/
-	struct hns_roce_hw	*hw;
+	const struct hns_roce_hw *hw;
 };
 
 static inline struct hns_roce_dev *to_hr_dev(struct ib_device *ib_dev)
@@ -749,7 +749,7 @@ struct ib_cq *hns_roce_ib_create_cq(struct ib_device *ib_dev,
 void hns_roce_cq_event(struct hns_roce_dev *hr_dev, u32 cqn, int event_type);
 void hns_roce_qp_event(struct hns_roce_dev *hr_dev, u32 qpn, int event_type);
 int hns_get_gid_index(struct hns_roce_dev *hr_dev, u8 port, int gid_index);
-
-extern struct hns_roce_hw hns_roce_hw_v1;
+int hns_roce_init(struct hns_roce_dev *hr_dev);
+void hns_roce_exit(struct hns_roce_dev *hr_dev);
 
 #endif /* _HNS_ROCE_DEVICE_H */
diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c
index c5104e0..06aed4a 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hem.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hem.c
@@ -314,6 +314,7 @@ void *hns_roce_table_find(struct hns_roce_hem_table *table, unsigned long obj,
 	mutex_unlock(&table->mutex);
 	return page ? lowmem_page_address(page) + offset : NULL;
 }
+EXPORT_SYMBOL_GPL(hns_roce_table_find);
 
 int hns_roce_table_get_range(struct hns_roce_dev *hr_dev,
 			     struct hns_roce_hem_table *table,
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 747efd1..917d003 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -34,6 +34,7 @@
 #include <linux/acpi.h>
 #include <linux/etherdevice.h>
 #include <linux/of.h>
+#include <linux/of_platform.h>
 #include <rdma/ib_umem.h>
 #include "hns_roce_common.h"
 #include "hns_roce_device.h"
@@ -3843,7 +3844,7 @@ int hns_roce_v1_destroy_cq(struct ib_cq *ibcq)
 
 struct hns_roce_v1_priv hr_v1_priv;
 
-struct hns_roce_hw hns_roce_hw_v1 = {
+static const struct hns_roce_hw hns_roce_hw_v1 = {
 	.reset = hns_roce_v1_reset,
 	.hw_profile = hns_roce_v1_profile,
 	.hw_init = hns_roce_v1_init,
@@ -3865,3 +3866,244 @@ struct hns_roce_hw hns_roce_hw_v1 = {
 	.destroy_cq = hns_roce_v1_destroy_cq,
 	.priv = &hr_v1_priv,
 };
+
+static const struct of_device_id hns_roce_of_match[] = {
+	{ .compatible = "hisilicon,hns-roce-v1", .data = &hns_roce_hw_v1, },
+	{},
+};
+MODULE_DEVICE_TABLE(of, hns_roce_of_match);
+
+static const struct acpi_device_id hns_roce_acpi_match[] = {
+	{ "HISI00D1", (kernel_ulong_t)&hns_roce_hw_v1 },
+	{},
+};
+MODULE_DEVICE_TABLE(acpi, hns_roce_acpi_match);
+
+static int hns_roce_node_match(struct device *dev, void *fwnode)
+{
+	return dev->fwnode == fwnode;
+}
+
+static struct
+platform_device *hns_roce_find_pdev(struct fwnode_handle *fwnode)
+{
+	struct device *dev;
+
+	/* get the 'device' corresponding to the matching 'fwnode' */
+	dev = bus_find_device(&platform_bus_type, NULL,
+			      fwnode, hns_roce_node_match);
+	/* get the platform device */
+	return dev ? to_platform_device(dev) : NULL;
+}
+
+static int hns_roce_get_cfg(struct hns_roce_dev *hr_dev)
+{
+	struct device *dev = &hr_dev->pdev->dev;
+	struct platform_device *pdev = NULL;
+	struct net_device *netdev = NULL;
+	struct device_node *net_node;
+	struct resource *res;
+	int port_cnt = 0;
+	u8 phy_port;
+	int ret;
+	int i;
+
+	/* check if we are compatible with the underlying SoC */
+	if (dev_of_node(dev)) {
+		const struct of_device_id *of_id;
+
+		of_id = of_match_node(hns_roce_of_match, dev->of_node);
+		if (!of_id) {
+			dev_err(dev, "device is not compatible!\n");
+			return -ENXIO;
+		}
+		hr_dev->hw = (const struct hns_roce_hw *)of_id->data;
+		if (!hr_dev->hw) {
+			dev_err(dev, "couldn't get H/W specific DT data!\n");
+			return -ENXIO;
+		}
+	} else if (is_acpi_device_node(dev->fwnode)) {
+		const struct acpi_device_id *acpi_id;
+
+		acpi_id = acpi_match_device(hns_roce_acpi_match, dev);
+		if (!acpi_id) {
+			dev_err(dev, "device is not compatible!\n");
+			return -ENXIO;
+		}
+		hr_dev->hw = (const struct hns_roce_hw *) acpi_id->driver_data;
+		if (!hr_dev->hw) {
+			dev_err(dev, "couldn't get H/W specific ACPI data!\n");
+			return -ENXIO;
+		}
+	} else {
+		dev_err(dev, "can't read compatibility data from DT or ACPI\n");
+		return -ENXIO;
+	}
+
+	/* get the mapped register base address */
+	res = platform_get_resource(hr_dev->pdev, IORESOURCE_MEM, 0);
+	if (!res) {
+		dev_err(dev, "memory resource not found!\n");
+		return -EINVAL;
+	}
+	hr_dev->reg_base = devm_ioremap_resource(dev, res);
+	if (IS_ERR(hr_dev->reg_base))
+		return PTR_ERR(hr_dev->reg_base);
+
+	/* read the node_guid of IB device from the DT or ACPI */
+	ret = device_property_read_u8_array(dev, "node-guid",
+					    (u8 *)&hr_dev->ib_dev.node_guid,
+					    GUID_LEN);
+	if (ret) {
+		dev_err(dev, "couldn't get node_guid from DT or ACPI!\n");
+		return ret;
+	}
+
+	/* get the RoCE associated ethernet ports or netdevices */
+	for (i = 0; i < HNS_ROCE_MAX_PORTS; i++) {
+		if (dev_of_node(dev)) {
+			net_node = of_parse_phandle(dev->of_node, "eth-handle",
+						    i);
+			if (!net_node)
+				continue;
+			pdev = of_find_device_by_node(net_node);
+		} else if (is_acpi_device_node(dev->fwnode)) {
+			struct acpi_reference_args args;
+			struct fwnode_handle *fwnode;
+
+			ret = acpi_node_get_property_reference(dev->fwnode,
+							       "eth-handle",
+							       i, &args);
+			if (ret)
+				continue;
+			fwnode = acpi_fwnode_handle(args.adev);
+			pdev = hns_roce_find_pdev(fwnode);
+		} else {
+			dev_err(dev, "cannot read data from DT or ACPI\n");
+			return -ENXIO;
+		}
+
+		if (pdev) {
+			netdev = platform_get_drvdata(pdev);
+			phy_port = (u8)i;
+			if (netdev) {
+				hr_dev->iboe.netdevs[port_cnt] = netdev;
+				hr_dev->iboe.phy_port[port_cnt] = phy_port;
+			} else {
+				dev_err(dev, "no netdev found with pdev %s\n",
+					pdev->name);
+				return -ENODEV;
+			}
+			port_cnt++;
+		}
+	}
+
+	if (port_cnt == 0) {
+		dev_err(dev, "unable to get eth-handle for available ports!\n");
+		return -EINVAL;
+	}
+
+	hr_dev->caps.num_ports = port_cnt;
+
+	/* cmd issue mode: 0 is poll, 1 is event */
+	hr_dev->cmd_mod = 1;
+	hr_dev->loop_idc = 0;
+
+	/* read the interrupt names from the DT or ACPI */
+	ret = device_property_read_string_array(dev, "interrupt-names",
+						hr_dev->irq_names,
+						HNS_ROCE_MAX_IRQ_NUM);
+	if (ret < 0) {
+		dev_err(dev, "couldn't get interrupt names from DT or ACPI!\n");
+		return ret;
+	}
+
+	/* fetch the interrupt numbers */
+	for (i = 0; i < HNS_ROCE_MAX_IRQ_NUM; i++) {
+		hr_dev->irq[i] = platform_get_irq(hr_dev->pdev, i);
+		if (hr_dev->irq[i] <= 0) {
+			dev_err(dev, "platform get of irq[=%d] failed!\n", i);
+			return -EINVAL;
+		}
+	}
+
+	return 0;
+}
+
+/**
+ * hns_roce_probe - RoCE driver entrance
+ * @pdev: pointer to platform device
+ * Return : int
+ *
+ */
+static int hns_roce_probe(struct platform_device *pdev)
+{
+	int ret;
+	struct hns_roce_dev *hr_dev;
+	struct device *dev = &pdev->dev;
+
+	hr_dev = (struct hns_roce_dev *)ib_alloc_device(sizeof(*hr_dev));
+	if (!hr_dev)
+		return -ENOMEM;
+
+	hr_dev->pdev = pdev;
+	platform_set_drvdata(pdev, hr_dev);
+
+	if (dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64ULL)) &&
+	    dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32ULL))) {
+		dev_err(dev, "Not usable DMA addressing mode\n");
+		ret = -EIO;
+		goto error_failed_get_cfg;
+	}
+
+	ret = hns_roce_get_cfg(hr_dev);
+	if (ret) {
+		dev_err(dev, "Get Configuration failed!\n");
+		goto error_failed_get_cfg;
+	}
+
+	ret = hns_roce_init(hr_dev);
+	if (ret) {
+		dev_err(dev, "RoCE engine init failed!\n");
+		goto error_failed_get_cfg;
+	}
+
+	return 0;
+
+error_failed_get_cfg:
+	ib_dealloc_device(&hr_dev->ib_dev);
+
+	return ret;
+}
+
+/**
+ * hns_roce_remove - remove RoCE device
+ * @pdev: pointer to platform device
+ */
+static int hns_roce_remove(struct platform_device *pdev)
+{
+	struct hns_roce_dev *hr_dev = platform_get_drvdata(pdev);
+
+	hns_roce_exit(hr_dev);
+	ib_dealloc_device(&hr_dev->ib_dev);
+
+	return 0;
+}
+
+static struct platform_driver hns_roce_driver = {
+	.probe = hns_roce_probe,
+	.remove = hns_roce_remove,
+	.driver = {
+		.name = DRV_NAME,
+		.of_match_table = hns_roce_of_match,
+		.acpi_match_table = ACPI_PTR(hns_roce_acpi_match),
+	},
+};
+
+module_platform_driver(hns_roce_driver);
+
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_AUTHOR("Wei Hu <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>");
+MODULE_AUTHOR("Nenglong Zhao <zhaonenglong-C8/M+/jPZTeaMJb+Lgu22Q@public.gmane.org>");
+MODULE_AUTHOR("Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>");
+MODULE_DESCRIPTION("Hisilicon Hip06 Family RoCE Driver");
diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c
index d9777b6..7afa0c8 100644
--- a/drivers/infiniband/hw/hns/hns_roce_main.c
+++ b/drivers/infiniband/hw/hns/hns_roce_main.c
@@ -57,6 +57,7 @@ int hns_get_gid_index(struct hns_roce_dev *hr_dev, u8 port, int gid_index)
 {
 	return gid_index * hr_dev->caps.num_ports + port;
 }
+EXPORT_SYMBOL_GPL(hns_get_gid_index);
 
 static void hns_roce_set_mac(struct hns_roce_dev *hr_dev, u8 port, u8 *addr)
 {
@@ -174,8 +175,9 @@ static int hns_roce_setup_mtu_mac(struct hns_roce_dev *hr_dev)
 	u8 i;
 
 	for (i = 0; i < hr_dev->caps.num_ports; i++) {
-		hr_dev->hw->set_mtu(hr_dev, hr_dev->iboe.phy_port[i],
-				    hr_dev->caps.max_mtu);
+		if (hr_dev->hw->set_mtu)
+			hr_dev->hw->set_mtu(hr_dev, hr_dev->iboe.phy_port[i],
+					    hr_dev->caps.max_mtu);
 		hns_roce_set_mac(hr_dev, i, hr_dev->iboe.netdevs[i]->dev_addr);
 	}
 
@@ -531,169 +533,6 @@ static int hns_roce_register_device(struct hns_roce_dev *hr_dev)
 	return ret;
 }
 
-static const struct of_device_id hns_roce_of_match[] = {
-	{ .compatible = "hisilicon,hns-roce-v1", .data = &hns_roce_hw_v1, },
-	{},
-};
-MODULE_DEVICE_TABLE(of, hns_roce_of_match);
-
-static const struct acpi_device_id hns_roce_acpi_match[] = {
-	{ "HISI00D1", (kernel_ulong_t)&hns_roce_hw_v1 },
-	{},
-};
-MODULE_DEVICE_TABLE(acpi, hns_roce_acpi_match);
-
-static int hns_roce_node_match(struct device *dev, void *fwnode)
-{
-	return dev->fwnode == fwnode;
-}
-
-static struct
-platform_device *hns_roce_find_pdev(struct fwnode_handle *fwnode)
-{
-	struct device *dev;
-
-	/* get the 'device'corresponding to matching 'fwnode' */
-	dev = bus_find_device(&platform_bus_type, NULL,
-			      fwnode, hns_roce_node_match);
-	/* get the platform device */
-	return dev ? to_platform_device(dev) : NULL;
-}
-
-static int hns_roce_get_cfg(struct hns_roce_dev *hr_dev)
-{
-	int i;
-	int ret;
-	u8 phy_port;
-	int port_cnt = 0;
-	struct device *dev = &hr_dev->pdev->dev;
-	struct device_node *net_node;
-	struct net_device *netdev = NULL;
-	struct platform_device *pdev = NULL;
-	struct resource *res;
-
-	/* check if we are compatible with the underlying SoC */
-	if (dev_of_node(dev)) {
-		const struct of_device_id *of_id;
-
-		of_id = of_match_node(hns_roce_of_match, dev->of_node);
-		if (!of_id) {
-			dev_err(dev, "device is not compatible!\n");
-			return -ENXIO;
-		}
-		hr_dev->hw = (struct hns_roce_hw *)of_id->data;
-		if (!hr_dev->hw) {
-			dev_err(dev, "couldn't get H/W specific DT data!\n");
-			return -ENXIO;
-		}
-	} else if (is_acpi_device_node(dev->fwnode)) {
-		const struct acpi_device_id *acpi_id;
-
-		acpi_id = acpi_match_device(hns_roce_acpi_match, dev);
-		if (!acpi_id) {
-			dev_err(dev, "device is not compatible!\n");
-			return -ENXIO;
-		}
-		hr_dev->hw = (struct hns_roce_hw *) acpi_id->driver_data;
-		if (!hr_dev->hw) {
-			dev_err(dev, "couldn't get H/W specific ACPI data!\n");
-			return -ENXIO;
-		}
-	} else {
-		dev_err(dev, "can't read compatibility data from DT or ACPI\n");
-		return -ENXIO;
-	}
-
-	/* get the mapped register base address */
-	res = platform_get_resource(hr_dev->pdev, IORESOURCE_MEM, 0);
-	if (!res) {
-		dev_err(dev, "memory resource not found!\n");
-		return -EINVAL;
-	}
-	hr_dev->reg_base = devm_ioremap_resource(dev, res);
-	if (IS_ERR(hr_dev->reg_base))
-		return PTR_ERR(hr_dev->reg_base);
-
-	/* read the node_guid of IB device from the DT or ACPI */
-	ret = device_property_read_u8_array(dev, "node-guid",
-					    (u8 *)&hr_dev->ib_dev.node_guid,
-					    GUID_LEN);
-	if (ret) {
-		dev_err(dev, "couldn't get node_guid from DT or ACPI!\n");
-		return ret;
-	}
-
-	/* get the RoCE associated ethernet ports or netdevices */
-	for (i = 0; i < HNS_ROCE_MAX_PORTS; i++) {
-		if (dev_of_node(dev)) {
-			net_node = of_parse_phandle(dev->of_node, "eth-handle",
-						    i);
-			if (!net_node)
-				continue;
-			pdev = of_find_device_by_node(net_node);
-		} else if (is_acpi_device_node(dev->fwnode)) {
-			struct acpi_reference_args args;
-			struct fwnode_handle *fwnode;
-
-			ret = acpi_node_get_property_reference(dev->fwnode,
-							       "eth-handle",
-							       i, &args);
-			if (ret)
-				continue;
-			fwnode = acpi_fwnode_handle(args.adev);
-			pdev = hns_roce_find_pdev(fwnode);
-		} else {
-			dev_err(dev, "cannot read data from DT or ACPI\n");
-			return -ENXIO;
-		}
-
-		if (pdev) {
-			netdev = platform_get_drvdata(pdev);
-			phy_port = (u8)i;
-			if (netdev) {
-				hr_dev->iboe.netdevs[port_cnt] = netdev;
-				hr_dev->iboe.phy_port[port_cnt] = phy_port;
-			} else {
-				dev_err(dev, "no netdev found with pdev %s\n",
-					pdev->name);
-				return -ENODEV;
-			}
-			port_cnt++;
-		}
-	}
-
-	if (port_cnt == 0) {
-		dev_err(dev, "unable to get eth-handle for available ports!\n");
-		return -EINVAL;
-	}
-
-	hr_dev->caps.num_ports = port_cnt;
-
-	/* cmd issue mode: 0 is poll, 1 is event */
-	hr_dev->cmd_mod = 1;
-	hr_dev->loop_idc = 0;
-
-	/* read the interrupt names from the DT or ACPI */
-	ret = device_property_read_string_array(dev, "interrupt-names",
-						hr_dev->irq_names,
-						HNS_ROCE_MAX_IRQ_NUM);
-	if (ret < 0) {
-		dev_err(dev, "couldn't get interrupt names from DT or ACPI!\n");
-		return ret;
-	}
-
-	/* fetch the interrupt numbers */
-	for (i = 0; i < HNS_ROCE_MAX_IRQ_NUM; i++) {
-		hr_dev->irq[i] = platform_get_irq(hr_dev->pdev, i);
-		if (hr_dev->irq[i] <= 0) {
-			dev_err(dev, "platform get of irq[=%d] failed!\n", i);
-			return -EINVAL;
-		}
-	}
-
-	return 0;
-}
-
 static int hns_roce_init_hem(struct hns_roce_dev *hr_dev)
 {
 	int ret;
@@ -826,42 +665,17 @@ static int hns_roce_setup_hca(struct hns_roce_dev *hr_dev)
 	return ret;
 }
 
-/**
- * hns_roce_probe - RoCE driver entrance
- * @pdev: pointer to platform device
- * Return : int
- *
- */
-static int hns_roce_probe(struct platform_device *pdev)
+int hns_roce_init(struct hns_roce_dev *hr_dev)
 {
 	int ret;
-	struct hns_roce_dev *hr_dev;
-	struct device *dev = &pdev->dev;
-
-	hr_dev = (struct hns_roce_dev *)ib_alloc_device(sizeof(*hr_dev));
-	if (!hr_dev)
-		return -ENOMEM;
-
-	hr_dev->pdev = pdev;
-	platform_set_drvdata(pdev, hr_dev);
-
-	if (dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64ULL)) &&
-	    dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32ULL))) {
-		dev_err(dev, "Not usable DMA addressing mode\n");
-		ret = -EIO;
-		goto error_failed_get_cfg;
-	}
-
-	ret = hns_roce_get_cfg(hr_dev);
-	if (ret) {
-		dev_err(dev, "Get Configuration failed!\n");
-		goto error_failed_get_cfg;
-	}
+	struct device *dev = &hr_dev->pdev->dev;
 
-	ret = hr_dev->hw->reset(hr_dev, true);
-	if (ret) {
-		dev_err(dev, "Reset RoCE engine failed!\n");
-		goto error_failed_get_cfg;
+	if (hr_dev->hw->reset) {
+		ret = hr_dev->hw->reset(hr_dev, true);
+		if (ret) {
+			dev_err(dev, "Reset RoCE engine failed!\n");
+			return ret;
+		}
 	}
 
 	hr_dev->hw->hw_profile(hr_dev);
@@ -872,10 +686,12 @@ static int hns_roce_probe(struct platform_device *pdev)
 		goto error_failed_cmd_init;
 	}
 
-	ret = hns_roce_init_eq_table(hr_dev);
-	if (ret) {
-		dev_err(dev, "eq init failed!\n");
-		goto error_failed_eq_table;
+	if (hr_dev->cmd_mod) {
+		ret = hns_roce_init_eq_table(hr_dev);
+		if (ret) {
+			dev_err(dev, "eq init failed!\n");
+			goto error_failed_eq_table;
+		}
 	}
 
 	if (hr_dev->cmd_mod) {
@@ -898,10 +714,12 @@ static int hns_roce_probe(struct platform_device *pdev)
 		goto error_failed_setup_hca;
 	}
 
-	ret = hr_dev->hw->hw_init(hr_dev);
-	if (ret) {
-		dev_err(dev, "hw_init failed!\n");
-		goto error_failed_engine_init;
+	if (hr_dev->hw->hw_init) {
+		ret = hr_dev->hw->hw_init(hr_dev);
+		if (ret) {
+			dev_err(dev, "hw_init failed!\n");
+			goto error_failed_engine_init;
+		}
 	}
 
 	ret = hns_roce_register_device(hr_dev);
@@ -911,7 +729,8 @@ static int hns_roce_probe(struct platform_device *pdev)
 	return 0;
 
 error_failed_register_device:
-	hr_dev->hw->hw_exit(hr_dev);
+	if (hr_dev->hw->hw_exit)
+		hr_dev->hw->hw_exit(hr_dev);
 
 error_failed_engine_init:
 	hns_roce_cleanup_bitmap(hr_dev);
@@ -924,58 +743,41 @@ static int hns_roce_probe(struct platform_device *pdev)
 		hns_roce_cmd_use_polling(hr_dev);
 
 error_failed_use_event:
-	hns_roce_cleanup_eq_table(hr_dev);
+	if (hr_dev->cmd_mod)
+		hns_roce_cleanup_eq_table(hr_dev);
 
 error_failed_eq_table:
 	hns_roce_cmd_cleanup(hr_dev);
 
 error_failed_cmd_init:
-	ret = hr_dev->hw->reset(hr_dev, false);
-	if (ret)
-		dev_err(&hr_dev->pdev->dev, "roce_engine reset fail\n");
-
-error_failed_get_cfg:
-	ib_dealloc_device(&hr_dev->ib_dev);
+	if (hr_dev->hw->reset) {
+		ret = hr_dev->hw->reset(hr_dev, false);
+		if (ret)
+			dev_err(dev, "Dereset RoCE engine failed!\n");
+	}
 
 	return ret;
 }
+EXPORT_SYMBOL_GPL(hns_roce_init);
 
-/**
- * hns_roce_remove - remove RoCE device
- * @pdev: pointer to platform device
- */
-static int hns_roce_remove(struct platform_device *pdev)
+void hns_roce_exit(struct hns_roce_dev *hr_dev)
 {
-	struct hns_roce_dev *hr_dev = platform_get_drvdata(pdev);
-
 	hns_roce_unregister_device(hr_dev);
-	hr_dev->hw->hw_exit(hr_dev);
+	if (hr_dev->hw->hw_exit)
+		hr_dev->hw->hw_exit(hr_dev);
 	hns_roce_cleanup_bitmap(hr_dev);
 	hns_roce_cleanup_hem(hr_dev);
 
 	if (hr_dev->cmd_mod)
 		hns_roce_cmd_use_polling(hr_dev);
 
-	hns_roce_cleanup_eq_table(hr_dev);
+	if (hr_dev->cmd_mod)
+		hns_roce_cleanup_eq_table(hr_dev);
 	hns_roce_cmd_cleanup(hr_dev);
-	hr_dev->hw->reset(hr_dev, false);
-
-	ib_dealloc_device(&hr_dev->ib_dev);
-
-	return 0;
+	if (hr_dev->hw->reset)
+		hr_dev->hw->reset(hr_dev, false);
 }
-
-static struct platform_driver hns_roce_driver = {
-	.probe = hns_roce_probe,
-	.remove = hns_roce_remove,
-	.driver = {
-		.name = DRV_NAME,
-		.of_match_table = hns_roce_of_match,
-		.acpi_match_table = ACPI_PTR(hns_roce_acpi_match),
-	},
-};
-
-module_platform_driver(hns_roce_driver);
+EXPORT_SYMBOL_GPL(hns_roce_exit);
 
 MODULE_LICENSE("Dual BSD/GPL");
 MODULE_AUTHOR("Wei Hu <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>");
diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c b/drivers/infiniband/hw/hns/hns_roce_mr.c
index e387360..ed35713 100644
--- a/drivers/infiniband/hw/hns/hns_roce_mr.c
+++ b/drivers/infiniband/hw/hns/hns_roce_mr.c
@@ -47,6 +47,7 @@ unsigned long key_to_hw_index(u32 key)
 {
 	return (key << 24) | (key >> 8);
 }
+EXPORT_SYMBOL_GPL(key_to_hw_index);
 
 static int hns_roce_sw2hw_mpt(struct hns_roce_dev *hr_dev,
 			      struct hns_roce_cmd_mailbox *mailbox,
@@ -65,6 +66,7 @@ int hns_roce_hw2sw_mpt(struct hns_roce_dev *hr_dev,
 				 mpt_index, !mailbox, HNS_ROCE_CMD_HW2SW_MPT,
 				 HNS_ROCE_CMD_TIMEOUT_MSECS);
 }
+EXPORT_SYMBOL_GPL(hns_roce_hw2sw_mpt);
 
 static int hns_roce_buddy_alloc(struct hns_roce_buddy *buddy, int order,
 				unsigned long *seg)
@@ -233,6 +235,7 @@ void hns_roce_mtt_cleanup(struct hns_roce_dev *hr_dev, struct hns_roce_mtt *mtt)
 	hns_roce_table_put_range(hr_dev, &mr_table->mtt_table, mtt->first_seg,
 				 mtt->first_seg + (1 << mtt->order) - 1);
 }
+EXPORT_SYMBOL_GPL(hns_roce_mtt_cleanup);
 
 static int hns_roce_mr_alloc(struct hns_roce_dev *hr_dev, u32 pd, u64 iova,
 			     u64 size, u32 access, int npages,
diff --git a/drivers/infiniband/hw/hns/hns_roce_pd.c b/drivers/infiniband/hw/hns/hns_roce_pd.c
index a64500f..ffbf249 100644
--- a/drivers/infiniband/hw/hns/hns_roce_pd.c
+++ b/drivers/infiniband/hw/hns/hns_roce_pd.c
@@ -86,6 +86,7 @@ struct ib_pd *hns_roce_alloc_pd(struct ib_device *ib_dev,
 
 	return &pd->ibpd;
 }
+EXPORT_SYMBOL_GPL(hns_roce_alloc_pd);
 
 int hns_roce_dealloc_pd(struct ib_pd *pd)
 {
@@ -94,6 +95,7 @@ int hns_roce_dealloc_pd(struct ib_pd *pd)
 
 	return 0;
 }
+EXPORT_SYMBOL_GPL(hns_roce_dealloc_pd);
 
 int hns_roce_uar_alloc(struct hns_roce_dev *hr_dev, struct hns_roce_uar *uar)
 {
diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c
index f5dd21c..ad13d05 100644
--- a/drivers/infiniband/hw/hns/hns_roce_qp.c
+++ b/drivers/infiniband/hw/hns/hns_roce_qp.c
@@ -136,6 +136,7 @@ enum hns_roce_qp_state to_hns_roce_state(enum ib_qp_state state)
 		return HNS_ROCE_QP_NUM_STATE;
 	}
 }
+EXPORT_SYMBOL_GPL(to_hns_roce_state);
 
 static int hns_roce_gsi_qp_alloc(struct hns_roce_dev *hr_dev, unsigned long qpn,
 				 struct hns_roce_qp *hr_qp)
@@ -227,6 +228,7 @@ void hns_roce_qp_remove(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp)
 			  hr_qp->qpn & (hr_dev->caps.num_qps - 1));
 	spin_unlock_irqrestore(&qp_table->lock, flags);
 }
+EXPORT_SYMBOL_GPL(hns_roce_qp_remove);
 
 void hns_roce_qp_free(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp)
 {
@@ -241,6 +243,7 @@ void hns_roce_qp_free(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp)
 		hns_roce_table_put(hr_dev, &qp_table->qp_table, hr_qp->qpn);
 	}
 }
+EXPORT_SYMBOL_GPL(hns_roce_qp_free);
 
 void hns_roce_release_range_qp(struct hns_roce_dev *hr_dev, int base_qpn,
 			       int cnt)
@@ -252,6 +255,7 @@ void hns_roce_release_range_qp(struct hns_roce_dev *hr_dev, int base_qpn,
 
 	hns_roce_bitmap_free_range(&qp_table->bitmap, base_qpn, cnt, BITMAP_RR);
 }
+EXPORT_SYMBOL_GPL(hns_roce_release_range_qp);
 
 static int hns_roce_set_rq_size(struct hns_roce_dev *hr_dev,
 				struct ib_qp_cap *cap, int is_user, int has_srq,
@@ -629,6 +633,7 @@ struct ib_qp *hns_roce_create_qp(struct ib_pd *pd,
 
 	return &hr_qp->ibqp;
 }
+EXPORT_SYMBOL_GPL(hns_roce_create_qp);
 
 int to_hr_qp_type(int qp_type)
 {
@@ -647,6 +652,7 @@ int to_hr_qp_type(int qp_type)
 
 	return transport_type;
 }
+EXPORT_SYMBOL_GPL(to_hr_qp_type);
 
 int hns_roce_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 		       int attr_mask, struct ib_udata *udata)
@@ -745,6 +751,7 @@ void hns_roce_lock_cqs(struct hns_roce_cq *send_cq, struct hns_roce_cq *recv_cq)
 		spin_lock_nested(&send_cq->lock, SINGLE_DEPTH_NESTING);
 	}
 }
+EXPORT_SYMBOL_GPL(hns_roce_lock_cqs);
 
 void hns_roce_unlock_cqs(struct hns_roce_cq *send_cq,
 			 struct hns_roce_cq *recv_cq) __releases(&send_cq->lock)
@@ -761,6 +768,7 @@ void hns_roce_unlock_cqs(struct hns_roce_cq *send_cq,
 		spin_unlock_irq(&recv_cq->lock);
 	}
 }
+EXPORT_SYMBOL_GPL(hns_roce_unlock_cqs);
 
 __be32 send_ieth(struct ib_send_wr *wr)
 {
@@ -774,6 +782,7 @@ __be32 send_ieth(struct ib_send_wr *wr)
 		return 0;
 	}
 }
+EXPORT_SYMBOL_GPL(send_ieth);
 
 static void *get_wqe(struct hns_roce_qp *hr_qp, int offset)
 {
@@ -785,11 +794,13 @@ void *get_recv_wqe(struct hns_roce_qp *hr_qp, int n)
 {
 	return get_wqe(hr_qp, hr_qp->rq.offset + (n << hr_qp->rq.wqe_shift));
 }
+EXPORT_SYMBOL_GPL(get_recv_wqe);
 
 void *get_send_wqe(struct hns_roce_qp *hr_qp, int n)
 {
 	return get_wqe(hr_qp, hr_qp->sq.offset + (n << hr_qp->sq.wqe_shift));
 }
+EXPORT_SYMBOL_GPL(get_send_wqe);
 
 bool hns_roce_wq_overflow(struct hns_roce_wq *hr_wq, int nreq,
 			  struct ib_cq *ib_cq)
@@ -808,6 +819,7 @@ bool hns_roce_wq_overflow(struct hns_roce_wq *hr_wq, int nreq,
 
 	return cur + nreq >= hr_wq->max_post;
 }
+EXPORT_SYMBOL_GPL(hns_roce_wq_overflow);
 
 int hns_roce_init_qp_table(struct hns_roce_dev *hr_dev)
 {
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 02/20] RDMA/hns: Move priv in order to add multiple hns_roce support

[Wei Hu (Xavier)]

Move the data member called priv from hns_roce_hw to hns_roce_dev
structure in order to support multiple hns_roce devices in one
system at the same time. For example, there are two hip06 engines
in the system.

Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_device.h |  2 +-
 drivers/infiniband/hw/hns/hns_roce_hw_v1.c  | 61 ++++++++++++++++-------------
 2 files changed, 35 insertions(+), 28 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index fc2b82d..aa9281e 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -535,7 +535,6 @@ struct hns_roce_hw {
 	int (*poll_cq)(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
 	int (*dereg_mr)(struct hns_roce_dev *hr_dev, struct hns_roce_mr *mr);
 	int (*destroy_cq)(struct ib_cq *ibcq);
-	void	*priv;
 };
 
 struct hns_roce_dev {
@@ -572,6 +571,7 @@ struct hns_roce_dev {
 	dma_addr_t		tptr_dma_addr; /*only for hw v1*/
 	u32			tptr_size; /*only for hw v1*/
 	const struct hns_roce_hw *hw;
+	void			*priv;
 };
 
 static inline struct hns_roce_dev *to_hr_dev(struct ib_device *ib_dev)
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 917d003..e4d13a9 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -473,7 +473,7 @@ static void hns_roce_set_sdb_ext(struct hns_roce_dev *hr_dev, u32 ext_sdb_alept,
 	dma_addr_t sdb_dma_addr;
 	u32 val;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	db = &priv->db_table;
 
 	/* Configure extend SDB threshold */
@@ -512,7 +512,7 @@ static void hns_roce_set_odb_ext(struct hns_roce_dev *hr_dev, u32 ext_odb_alept,
 	dma_addr_t odb_dma_addr;
 	u32 val;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	db = &priv->db_table;
 
 	/* Configure extend ODB threshold */
@@ -548,7 +548,7 @@ static int hns_roce_db_ext_init(struct hns_roce_dev *hr_dev, u32 sdb_ext_mod,
 	dma_addr_t odb_dma_addr;
 	int ret = 0;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	db = &priv->db_table;
 
 	db->ext_db = kmalloc(sizeof(*db->ext_db), GFP_KERNEL);
@@ -669,7 +669,7 @@ static int hns_roce_v1_rsv_lp_qp(struct hns_roce_dev *hr_dev)
 	u8 port = 0;
 	u8 sl;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	free_mr = &priv->free_mr;
 
 	/* Reserved cq for loop qp */
@@ -817,7 +817,7 @@ static void hns_roce_v1_release_lp_qp(struct hns_roce_dev *hr_dev)
 	int ret;
 	int i;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	free_mr = &priv->free_mr;
 
 	for (i = 0; i < HNS_ROCE_V1_RESV_QP; i++) {
@@ -851,7 +851,7 @@ static int hns_roce_db_init(struct hns_roce_dev *hr_dev)
 	u32 odb_evt_mod;
 	int ret = 0;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	db = &priv->db_table;
 
 	memset(db, 0, sizeof(*db));
@@ -907,7 +907,7 @@ static int hns_roce_v1_recreate_lp_qp(struct hns_roce_dev *hr_dev)
 	unsigned long end =
 	  msecs_to_jiffies(HNS_ROCE_V1_RECREATE_LP_QP_TIMEOUT_MSECS) + jiffies;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	free_mr = &priv->free_mr;
 
 	lp_qp_work = kzalloc(sizeof(struct hns_roce_recreate_lp_qp_work),
@@ -983,7 +983,7 @@ static void hns_roce_v1_mr_free_work_fn(struct work_struct *work)
 	hr_dev = to_hr_dev(mr_work->ib_dev);
 	dev = &hr_dev->pdev->dev;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	free_mr = &priv->free_mr;
 	mr_free_cq = free_mr->mr_free_cq;
 
@@ -1039,7 +1039,7 @@ int hns_roce_v1_dereg_mr(struct hns_roce_dev *hr_dev, struct hns_roce_mr *mr)
 	int npages;
 	int ret = 0;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	free_mr = &priv->free_mr;
 
 	if (mr->enabled) {
@@ -1104,7 +1104,7 @@ static void hns_roce_db_free(struct hns_roce_dev *hr_dev)
 	struct hns_roce_v1_priv *priv;
 	struct hns_roce_db_table *db;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	db = &priv->db_table;
 
 	if (db->sdb_ext_mod) {
@@ -1134,7 +1134,7 @@ static int hns_roce_raq_init(struct hns_roce_dev *hr_dev)
 	struct hns_roce_raq_table *raq;
 	struct device *dev = &hr_dev->pdev->dev;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	raq = &priv->raq_table;
 
 	raq->e_raq_buf = kzalloc(sizeof(*(raq->e_raq_buf)), GFP_KERNEL);
@@ -1211,7 +1211,7 @@ static void hns_roce_raq_free(struct hns_roce_dev *hr_dev)
 	struct hns_roce_v1_priv *priv;
 	struct hns_roce_raq_table *raq;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	raq = &priv->raq_table;
 
 	dma_free_coherent(dev, HNS_ROCE_V1_RAQ_SIZE, raq->e_raq_buf->buf,
@@ -1245,7 +1245,7 @@ static int hns_roce_bt_init(struct hns_roce_dev *hr_dev)
 	struct hns_roce_v1_priv *priv;
 	int ret;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 
 	priv->bt_table.qpc_buf.buf = dma_alloc_coherent(dev,
 		HNS_ROCE_BT_RSV_BUF_SIZE, &priv->bt_table.qpc_buf.map,
@@ -1287,7 +1287,7 @@ static void hns_roce_bt_free(struct hns_roce_dev *hr_dev)
 	struct device *dev = &hr_dev->pdev->dev;
 	struct hns_roce_v1_priv *priv;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 
 	dma_free_coherent(dev, HNS_ROCE_BT_RSV_BUF_SIZE,
 		priv->bt_table.cqc_buf.buf, priv->bt_table.cqc_buf.map);
@@ -1305,7 +1305,7 @@ static int hns_roce_tptr_init(struct hns_roce_dev *hr_dev)
 	struct hns_roce_buf_list *tptr_buf;
 	struct hns_roce_v1_priv *priv;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	tptr_buf = &priv->tptr_table.tptr_buf;
 
 	/*
@@ -1331,7 +1331,7 @@ static void hns_roce_tptr_free(struct hns_roce_dev *hr_dev)
 	struct hns_roce_buf_list *tptr_buf;
 	struct hns_roce_v1_priv *priv;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	tptr_buf = &priv->tptr_table.tptr_buf;
 
 	dma_free_coherent(dev, HNS_ROCE_V1_TPTR_BUF_SIZE,
@@ -1345,7 +1345,7 @@ static int hns_roce_free_mr_init(struct hns_roce_dev *hr_dev)
 	struct hns_roce_v1_priv *priv;
 	int ret = 0;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	free_mr = &priv->free_mr;
 
 	free_mr->free_mr_wq = create_singlethread_workqueue("hns_roce_free_mr");
@@ -1369,7 +1369,7 @@ static void hns_roce_free_mr_free(struct hns_roce_dev *hr_dev)
 	struct hns_roce_free_mr *free_mr;
 	struct hns_roce_v1_priv *priv;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	free_mr = &priv->free_mr;
 
 	flush_workqueue(free_mr->free_mr_wq);
@@ -1433,7 +1433,7 @@ static int hns_roce_des_qp_init(struct hns_roce_dev *hr_dev)
 	struct hns_roce_v1_priv *priv;
 	struct hns_roce_des_qp *des_qp;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	des_qp = &priv->des_qp;
 
 	des_qp->requeue_flag = 1;
@@ -1451,7 +1451,7 @@ static void hns_roce_des_qp_free(struct hns_roce_dev *hr_dev)
 	struct hns_roce_v1_priv *priv;
 	struct hns_roce_des_qp *des_qp;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	des_qp = &priv->des_qp;
 
 	des_qp->requeue_flag = 0;
@@ -1942,7 +1942,7 @@ void hns_roce_v1_write_cqc(struct hns_roce_dev *hr_dev,
 	dma_addr_t tptr_dma_addr;
 	int offset;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	tptr_buf = &priv->tptr_table.tptr_buf;
 
 	cq_context = mb_buf;
@@ -2290,7 +2290,7 @@ int hns_roce_v1_clear_hem(struct hns_roce_dev *hr_dev,
 	void __iomem *bt_cmd;
 	u64 bt_ba = 0;
 
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 
 	switch (table->type) {
 	case HEM_TYPE_QPC:
@@ -3665,7 +3665,7 @@ static void hns_roce_v1_destroy_qp_work_fn(struct work_struct *work)
 	qp_work_entry = container_of(work, struct hns_roce_qp_work, work);
 	hr_dev = to_hr_dev(qp_work_entry->ib_dev);
 	dev = &hr_dev->pdev->dev;
-	priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+	priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 	hr_qp = qp_work_entry->qp;
 	qpn = hr_qp->qpn;
 
@@ -3782,7 +3782,7 @@ int hns_roce_v1_destroy_qp(struct ib_qp *ibqp)
 		qp_work->sdb_inv_cnt	= qp_work_entry.sdb_inv_cnt;
 		qp_work->sche_cnt	= qp_work_entry.sche_cnt;
 
-		priv = (struct hns_roce_v1_priv *)hr_dev->hw->priv;
+		priv = (struct hns_roce_v1_priv *)hr_dev->priv;
 		queue_work(priv->des_qp.qp_wq, &qp_work->work);
 		dev_dbg(dev, "Begin destroy QP(0x%lx) work.\n", hr_qp->qpn);
 	}
@@ -3842,8 +3842,6 @@ int hns_roce_v1_destroy_cq(struct ib_cq *ibcq)
 	return ret;
 }
 
-struct hns_roce_v1_priv hr_v1_priv;
-
 static const struct hns_roce_hw hns_roce_hw_v1 = {
 	.reset = hns_roce_v1_reset,
 	.hw_profile = hns_roce_v1_profile,
@@ -3864,7 +3862,6 @@ int hns_roce_v1_destroy_cq(struct ib_cq *ibcq)
 	.poll_cq = hns_roce_v1_poll_cq,
 	.dereg_mr = hns_roce_v1_dereg_mr,
 	.destroy_cq = hns_roce_v1_destroy_cq,
-	.priv = &hr_v1_priv,
 };
 
 static const struct of_device_id hns_roce_of_match[] = {
@@ -4046,6 +4043,12 @@ static int hns_roce_probe(struct platform_device *pdev)
 	if (!hr_dev)
 		return -ENOMEM;
 
+	hr_dev->priv = kzalloc(sizeof(struct hns_roce_v1_priv), GFP_KERNEL);
+	if (!hr_dev->priv) {
+		ret = -ENOMEM;
+		goto error_failed_kzalloc;
+	}
+
 	hr_dev->pdev = pdev;
 	platform_set_drvdata(pdev, hr_dev);
 
@@ -4071,6 +4074,9 @@ static int hns_roce_probe(struct platform_device *pdev)
 	return 0;
 
 error_failed_get_cfg:
+	kfree(hr_dev->priv);
+
+error_failed_kzalloc:
 	ib_dealloc_device(&hr_dev->ib_dev);
 
 	return ret;
@@ -4085,6 +4091,7 @@ static int hns_roce_remove(struct platform_device *pdev)
 	struct hns_roce_dev *hr_dev = platform_get_drvdata(pdev);
 
 	hns_roce_exit(hr_dev);
+	kfree(hr_dev->priv);
 	ib_dealloc_device(&hr_dev->ib_dev);
 
 	return 0;
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 03/20] RDMA/hns: Initialize the PCI device for hip08 RoCE

[Wei Hu (Xavier)]

The hip08 RoCE engine is a PCI device. This patch mainly
obtains some information of PCI device for hip08 RoCE from
hns3 NIC driver in the initialization.

Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/Kconfig           |  10 ++
 drivers/infiniband/hw/hns/Makefile          |   4 +
 drivers/infiniband/hw/hns/hns_roce_device.h |   2 +
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 151 ++++++++++++++++++++++++++++
 4 files changed, 167 insertions(+)
 create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.c

diff --git a/drivers/infiniband/hw/hns/Kconfig b/drivers/infiniband/hw/hns/Kconfig
index 37e33e5..fddb5fd 100644
--- a/drivers/infiniband/hw/hns/Kconfig
+++ b/drivers/infiniband/hw/hns/Kconfig
@@ -19,3 +19,13 @@ config INFINIBAND_HNS_HIP06
 
 	  To compile this driver as a module, choose M here: the module
 	  will be called hns-roce-hw-v1.
+
+config INFINIBAND_HNS_HIP08
+	tristate "Hisilicon Hip08 Family RoCE support"
+	depends on INFINIBAND_HNS && PCI && HNS3
+	---help---
+	  RoCE driver support for Hisilicon RoCE engine in Hisilicon Hip08 SoC.
+	  The RoCE engine is a PCI device.
+
+	  To compile this driver as a module, choose M here: the module
+	  will be called hns-roce-hw-v2.
diff --git a/drivers/infiniband/hw/hns/Makefile b/drivers/infiniband/hw/hns/Makefile
index 806ac4e..ff426a6 100644
--- a/drivers/infiniband/hw/hns/Makefile
+++ b/drivers/infiniband/hw/hns/Makefile
@@ -2,9 +2,13 @@
 # Makefile for the Hisilicon RoCE drivers.
 #
 
+ccflags-y :=  -Idrivers/net/ethernet/hisilicon/hns3
+
 obj-$(CONFIG_INFINIBAND_HNS) += hns-roce.o
 hns-roce-objs := hns_roce_main.o hns_roce_cmd.o hns_roce_eq.o hns_roce_pd.o \
 	hns_roce_ah.o hns_roce_hem.o hns_roce_mr.o hns_roce_qp.o \
 	hns_roce_cq.o hns_roce_alloc.o
 obj-$(CONFIG_INFINIBAND_HNS_HIP06) += hns-roce-hw-v1.o
 hns-roce-hw-v1-objs := hns_roce_hw_v1.o
+obj-$(CONFIG_INFINIBAND_HNS_HIP08) += hns-roce-hw-v2.o
+hns-roce-hw-v2-objs := hns_roce_hw_v2.o
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index aa9281e..3132b68 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -540,6 +540,8 @@ struct hns_roce_hw {
 struct hns_roce_dev {
 	struct ib_device	ib_dev;
 	struct platform_device  *pdev;
+	struct pci_dev		*pci_dev;
+	struct device		*dev;
 	struct hns_roce_uar     priv_uar;
 	const char		*irq_names[HNS_ROCE_MAX_IRQ_NUM];
 	spinlock_t		sm_lock;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
new file mode 100644
index 0000000..b35c72d
--- /dev/null
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -0,0 +1,151 @@
+/*
+ * Copyright (c) 2016-2017 Hisilicon Limited.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/acpi.h>
+#include <linux/etherdevice.h>
+#include <linux/interrupt.h>
+#include <linux/kernel.h>
+#include <rdma/ib_umem.h>
+
+#include "hnae3.h"
+#include "hns_roce_common.h"
+#include "hns_roce_device.h"
+#include "hns_roce_cmd.h"
+#include "hns_roce_hem.h"
+
+static const struct hns_roce_hw hns_roce_hw_v2;
+
+static const struct pci_device_id hns_roce_hw_v2_pci_tbl[] = {
+	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA), 0},
+	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC), 0},
+	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC), 0},
+	/* required last entry */
+	{0, }
+};
+
+static int hns_roce_hw_v2_get_cfg(struct hns_roce_dev *hr_dev,
+				  struct hnae3_handle *handle)
+{
+	const struct pci_device_id *id;
+
+	id = pci_match_id(hns_roce_hw_v2_pci_tbl, hr_dev->pci_dev);
+	if (!id) {
+		dev_err(hr_dev->dev, "device is not compatible!\n");
+		return -ENXIO;
+	}
+
+	hr_dev->hw = &hns_roce_hw_v2;
+
+	/* Get info from NIC driver. */
+	hr_dev->reg_base = handle->rinfo.roce_io_base;
+	hr_dev->caps.num_ports = 1;
+	hr_dev->iboe.netdevs[0] = handle->rinfo.netdev;
+	hr_dev->iboe.phy_port[0] = 0;
+
+	/* cmd issue mode: 0 is poll, 1 is event */
+	hr_dev->cmd_mod = 0;
+	hr_dev->loop_idc = 0;
+
+	return 0;
+}
+
+static int hns_roce_hw_v2_init_instance(struct hnae3_handle *handle)
+{
+	struct hns_roce_dev *hr_dev;
+	int ret;
+
+	hr_dev = (struct hns_roce_dev *)ib_alloc_device(sizeof(*hr_dev));
+	if (!hr_dev)
+		return -ENOMEM;
+
+	hr_dev->pci_dev = handle->pdev;
+	hr_dev->dev = &handle->pdev->dev;
+	handle->priv = hr_dev;
+
+	ret = hns_roce_hw_v2_get_cfg(hr_dev, handle);
+	if (ret) {
+		dev_err(hr_dev->dev, "Get Configuration failed!\n");
+		goto error_failed_get_cfg;
+	}
+
+	ret = hns_roce_init(hr_dev);
+	if (ret) {
+		dev_err(hr_dev->dev, "RoCE Engine init failed!\n");
+		goto error_failed_get_cfg;
+	}
+
+	return 0;
+
+error_failed_get_cfg:
+	ib_dealloc_device(&hr_dev->ib_dev);
+
+	return ret;
+}
+
+static void hns_roce_hw_v2_uninit_instance(struct hnae3_handle *handle,
+					   bool reset)
+{
+	struct hns_roce_dev *hr_dev = (struct hns_roce_dev *)handle->priv;
+
+	hns_roce_exit(hr_dev);
+	ib_dealloc_device(&hr_dev->ib_dev);
+}
+
+static const struct hnae3_client_ops hns_roce_hw_v2_ops = {
+	.init_instance = hns_roce_hw_v2_init_instance,
+	.uninit_instance = hns_roce_hw_v2_uninit_instance,
+};
+
+static struct hnae3_client hns_roce_hw_v2_client = {
+	.name = "hns_roce_hw_v2",
+	.type = HNAE3_CLIENT_ROCE,
+	.ops = &hns_roce_hw_v2_ops,
+};
+
+static int __init hns_roce_hw_v2_init(void)
+{
+	return hnae3_register_client(&hns_roce_hw_v2_client);
+}
+
+static void __exit hns_roce_hw_v2_exit(void)
+{
+	hnae3_unregister_client(&hns_roce_hw_v2_client);
+}
+
+module_init(hns_roce_hw_v2_init);
+module_exit(hns_roce_hw_v2_exit);
+
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_AUTHOR("Wei Hu <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>");
+MODULE_AUTHOR("Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>");
+MODULE_AUTHOR("Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>");
+MODULE_DESCRIPTION("Hisilicon Hip08 Family RoCE Driver");
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 04/20] RDMA/hns: Modify assignment device variable to support both PCI device and platform device

[Wei Hu (Xavier)]

In order to support the scalability of the hardware version, the
features irrelevant to the hardware will be located in the hns-roce.ko,
and the hardware relevant operations will be located in hns_roce_hw_v1.ko
or hns_roce_hw_v2.ko based on the series chips.

The hip08 RoCE engine is a PCI device, hip06 RoCE engine is a platform
device. In order to support both platform device and PCI device, We
replace &hr_dev->pdev->dev with hr_dev->dev in hns-roce.ko as belows:
	Before modification:
		struct device *dev = hr_dev->dev;
	After modification:
		struct device *dev = &hr_dev->pdev->dev;

	The related structure:
	struct hns_roce_dev {
		...
		struct platform_device  *pdev;
		struct pci_dev		*pci_dev;
		struct device		*dev;
		...
	}

Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_ah.c    |  2 +-
 drivers/infiniband/hw/hns/hns_roce_alloc.c |  6 +++---
 drivers/infiniband/hw/hns/hns_roce_cmd.c   |  6 +++---
 drivers/infiniband/hw/hns/hns_roce_cq.c    | 12 ++++++------
 drivers/infiniband/hw/hns/hns_roce_hem.c   | 12 ++++++------
 drivers/infiniband/hw/hns/hns_roce_hw_v1.c |  1 +
 drivers/infiniband/hw/hns/hns_roce_main.c  | 12 ++++++------
 drivers/infiniband/hw/hns/hns_roce_mr.c    |  8 ++++----
 drivers/infiniband/hw/hns/hns_roce_pd.c    |  2 +-
 drivers/infiniband/hw/hns/hns_roce_qp.c    | 20 ++++++++++----------
 10 files changed, 41 insertions(+), 40 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_ah.c b/drivers/infiniband/hw/hns/hns_roce_ah.c
index d545302..b30bc88 100644
--- a/drivers/infiniband/hw/hns/hns_roce_ah.c
+++ b/drivers/infiniband/hw/hns/hns_roce_ah.c
@@ -44,7 +44,7 @@ struct ib_ah *hns_roce_create_ah(struct ib_pd *ibpd,
 				 struct ib_udata *udata)
 {
 	struct hns_roce_dev *hr_dev = to_hr_dev(ibpd->device);
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	struct ib_gid_attr gid_attr;
 	struct hns_roce_ah *ah;
 	u16 vlan_tag = 0xffff;
diff --git a/drivers/infiniband/hw/hns/hns_roce_alloc.c b/drivers/infiniband/hw/hns/hns_roce_alloc.c
index 71a5098..8c9a33f 100644
--- a/drivers/infiniband/hw/hns/hns_roce_alloc.c
+++ b/drivers/infiniband/hw/hns/hns_roce_alloc.c
@@ -161,7 +161,7 @@ void hns_roce_buf_free(struct hns_roce_dev *hr_dev, u32 size,
 		       struct hns_roce_buf *buf)
 {
 	int i;
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	u32 bits_per_long = BITS_PER_LONG;
 
 	if (buf->nbufs == 1) {
@@ -172,7 +172,7 @@ void hns_roce_buf_free(struct hns_roce_dev *hr_dev, u32 size,
 
 		for (i = 0; i < buf->nbufs; ++i)
 			if (buf->page_list[i].buf)
-				dma_free_coherent(&hr_dev->pdev->dev, PAGE_SIZE,
+				dma_free_coherent(dev, PAGE_SIZE,
 						  buf->page_list[i].buf,
 						  buf->page_list[i].map);
 		kfree(buf->page_list);
@@ -186,7 +186,7 @@ int hns_roce_buf_alloc(struct hns_roce_dev *hr_dev, u32 size, u32 max_direct,
 	int i = 0;
 	dma_addr_t t;
 	struct page **pages;
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	u32 bits_per_long = BITS_PER_LONG;
 
 	/* SQ/RQ buf lease than one page, SQ + RQ = 8K */
diff --git a/drivers/infiniband/hw/hns/hns_roce_cmd.c b/drivers/infiniband/hw/hns/hns_roce_cmd.c
index c981d9d..67ad3e9 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cmd.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cmd.c
@@ -125,7 +125,7 @@ static int __hns_roce_cmd_mbox_poll(struct hns_roce_dev *hr_dev, u64 in_param,
 				    u8 op_modifier, u16 op,
 				    unsigned long timeout)
 {
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	u8 __iomem *hcr = hr_dev->cmd.hcr;
 	unsigned long end = 0;
 	u32 status = 0;
@@ -196,8 +196,8 @@ static int __hns_roce_cmd_mbox_wait(struct hns_roce_dev *hr_dev, u64 in_param,
 				    unsigned long timeout)
 {
 	struct hns_roce_cmdq *cmd = &hr_dev->cmd;
-	struct device *dev = &hr_dev->pdev->dev;
 	struct hns_roce_cmd_context *context;
+	struct device *dev = hr_dev->dev;
 	int ret = 0;
 
 	spin_lock(&cmd->context_lock);
@@ -273,7 +273,7 @@ int hns_roce_cmd_mbox(struct hns_roce_dev *hr_dev, u64 in_param, u64 out_param,
 
 int hns_roce_cmd_init(struct hns_roce_dev *hr_dev)
 {
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 
 	mutex_init(&hr_dev->cmd.hcr_mutex);
 	sema_init(&hr_dev->cmd.poll_sem, 1);
diff --git a/drivers/infiniband/hw/hns/hns_roce_cq.c b/drivers/infiniband/hw/hns/hns_roce_cq.c
index 1d57df7..5f51242 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cq.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cq.c
@@ -58,7 +58,7 @@ static void hns_roce_ib_cq_event(struct hns_roce_cq *hr_cq,
 	if (event_type != HNS_ROCE_EVENT_TYPE_CQ_ID_INVALID &&
 	    event_type != HNS_ROCE_EVENT_TYPE_CQ_ACCESS_ERROR &&
 	    event_type != HNS_ROCE_EVENT_TYPE_CQ_OVERFLOW) {
-		dev_err(&hr_dev->pdev->dev,
+		dev_err(hr_dev->dev,
 			"hns_roce_ib: Unexpected event type 0x%x on CQ %06lx\n",
 			event_type, hr_cq->cqn);
 		return;
@@ -87,7 +87,7 @@ static int hns_roce_cq_alloc(struct hns_roce_dev *hr_dev, int nent,
 {
 	struct hns_roce_cmd_mailbox *mailbox = NULL;
 	struct hns_roce_cq_table *cq_table = NULL;
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	dma_addr_t dma_handle;
 	u64 *mtts = NULL;
 	int ret = 0;
@@ -182,7 +182,7 @@ static int hns_roce_hw2sw_cq(struct hns_roce_dev *dev,
 void hns_roce_free_cq(struct hns_roce_dev *hr_dev, struct hns_roce_cq *hr_cq)
 {
 	struct hns_roce_cq_table *cq_table = &hr_dev->cq_table;
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	int ret;
 
 	ret = hns_roce_hw2sw_cq(hr_dev, NULL, hr_cq->cqn);
@@ -282,7 +282,7 @@ struct ib_cq *hns_roce_ib_create_cq(struct ib_device *ib_dev,
 				    struct ib_udata *udata)
 {
 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	struct hns_roce_ib_create_cq ucmd;
 	struct hns_roce_cq *hr_cq = NULL;
 	struct hns_roce_uar *uar = NULL;
@@ -416,7 +416,7 @@ int hns_roce_ib_destroy_cq(struct ib_cq *ib_cq)
 
 void hns_roce_cq_completion(struct hns_roce_dev *hr_dev, u32 cqn)
 {
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	struct hns_roce_cq *cq;
 
 	cq = radix_tree_lookup(&hr_dev->cq_table.tree,
@@ -432,7 +432,7 @@ void hns_roce_cq_completion(struct hns_roce_dev *hr_dev, u32 cqn)
 void hns_roce_cq_event(struct hns_roce_dev *hr_dev, u32 cqn, int event_type)
 {
 	struct hns_roce_cq_table *cq_table = &hr_dev->cq_table;
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	struct hns_roce_cq *cq;
 
 	cq = radix_tree_lookup(&cq_table->tree,
diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c
index 06aed4a..0ab4912 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hem.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hem.c
@@ -84,7 +84,7 @@ struct hns_roce_hem *hns_roce_alloc_hem(struct hns_roce_dev *hr_dev, int npages,
 		 * memory, directly return fail.
 		 */
 		mem = &chunk->mem[chunk->npages];
-		buf = dma_alloc_coherent(&hr_dev->pdev->dev, PAGE_SIZE << order,
+		buf = dma_alloc_coherent(hr_dev->dev, PAGE_SIZE << order,
 				&sg_dma_address(mem), gfp_mask);
 		if (!buf)
 			goto fail;
@@ -115,7 +115,7 @@ void hns_roce_free_hem(struct hns_roce_dev *hr_dev, struct hns_roce_hem *hem)
 
 	list_for_each_entry_safe(chunk, tmp, &hem->chunk_list, list) {
 		for (i = 0; i < chunk->npages; ++i)
-			dma_free_coherent(&hr_dev->pdev->dev,
+			dma_free_coherent(hr_dev->dev,
 				   chunk->mem[i].length,
 				   lowmem_page_address(sg_page(&chunk->mem[i])),
 				   sg_dma_address(&chunk->mem[i]));
@@ -128,8 +128,8 @@ void hns_roce_free_hem(struct hns_roce_dev *hr_dev, struct hns_roce_hem *hem)
 static int hns_roce_set_hem(struct hns_roce_dev *hr_dev,
 			    struct hns_roce_hem_table *table, unsigned long obj)
 {
-	struct device *dev = &hr_dev->pdev->dev;
 	spinlock_t *lock = &hr_dev->bt_cmd_lock;
+	struct device *dev = hr_dev->dev;
 	unsigned long end = 0;
 	unsigned long flags;
 	struct hns_roce_hem_iter iter;
@@ -212,7 +212,7 @@ static int hns_roce_set_hem(struct hns_roce_dev *hr_dev,
 int hns_roce_table_get(struct hns_roce_dev *hr_dev,
 		       struct hns_roce_hem_table *table, unsigned long obj)
 {
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	int ret = 0;
 	unsigned long i;
 
@@ -251,7 +251,7 @@ int hns_roce_table_get(struct hns_roce_dev *hr_dev,
 void hns_roce_table_put(struct hns_roce_dev *hr_dev,
 			struct hns_roce_hem_table *table, unsigned long obj)
 {
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	unsigned long i;
 
 	i = (obj & (table->num_obj - 1)) /
@@ -380,7 +380,7 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev,
 void hns_roce_cleanup_hem_table(struct hns_roce_dev *hr_dev,
 				struct hns_roce_hem_table *table)
 {
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	unsigned long i;
 
 	for (i = 0; i < table->num_hem; ++i)
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index e4d13a9..22cc3fb 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -4050,6 +4050,7 @@ static int hns_roce_probe(struct platform_device *pdev)
 	}
 
 	hr_dev->pdev = pdev;
+	hr_dev->dev = dev;
 	platform_set_drvdata(pdev, hr_dev);
 
 	if (dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64ULL)) &&
diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c
index 7afa0c8..b07d437 100644
--- a/drivers/infiniband/hw/hns/hns_roce_main.c
+++ b/drivers/infiniband/hw/hns/hns_roce_main.c
@@ -117,7 +117,7 @@ static int hns_roce_del_gid(struct ib_device *device, u8 port_num,
 static int handle_en_event(struct hns_roce_dev *hr_dev, u8 port,
 			   unsigned long event)
 {
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	struct net_device *netdev;
 
 	netdev = hr_dev->iboe.netdevs[port];
@@ -240,7 +240,7 @@ static int hns_roce_query_port(struct ib_device *ib_dev, u8 port_num,
 			       struct ib_port_attr *props)
 {
 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	struct net_device *net_dev;
 	unsigned long flags;
 	enum ib_mtu mtu;
@@ -428,7 +428,7 @@ static int hns_roce_register_device(struct hns_roce_dev *hr_dev)
 	int ret;
 	struct hns_roce_ib_iboe *iboe = NULL;
 	struct ib_device *ib_dev = NULL;
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 
 	iboe = &hr_dev->iboe;
 	spin_lock_init(&iboe->lock);
@@ -536,7 +536,7 @@ static int hns_roce_register_device(struct hns_roce_dev *hr_dev)
 static int hns_roce_init_hem(struct hns_roce_dev *hr_dev)
 {
 	int ret;
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 
 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->mr_table.mtt_table,
 				      HEM_TYPE_MTT, hr_dev->caps.mtt_entry_sz,
@@ -605,7 +605,7 @@ static int hns_roce_init_hem(struct hns_roce_dev *hr_dev)
 static int hns_roce_setup_hca(struct hns_roce_dev *hr_dev)
 {
 	int ret;
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 
 	spin_lock_init(&hr_dev->sm_lock);
 	spin_lock_init(&hr_dev->bt_cmd_lock);
@@ -668,7 +668,7 @@ static int hns_roce_setup_hca(struct hns_roce_dev *hr_dev)
 int hns_roce_init(struct hns_roce_dev *hr_dev)
 {
 	int ret;
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 
 	if (hr_dev->hw->reset) {
 		ret = hr_dev->hw->reset(hr_dev, true);
diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c b/drivers/infiniband/hw/hns/hns_roce_mr.c
index ed35713..923d2b4 100644
--- a/drivers/infiniband/hw/hns/hns_roce_mr.c
+++ b/drivers/infiniband/hw/hns/hns_roce_mr.c
@@ -241,9 +241,9 @@ static int hns_roce_mr_alloc(struct hns_roce_dev *hr_dev, u32 pd, u64 iova,
 			     u64 size, u32 access, int npages,
 			     struct hns_roce_mr *mr)
 {
+	struct device *dev = hr_dev->dev;
 	unsigned long index = 0;
 	int ret = 0;
-	struct device *dev = &hr_dev->pdev->dev;
 
 	/* Allocate a key for mr from mr_table */
 	ret = hns_roce_bitmap_alloc(&hr_dev->mr_table.mtpt_bitmap, &index);
@@ -276,7 +276,7 @@ static int hns_roce_mr_alloc(struct hns_roce_dev *hr_dev, u32 pd, u64 iova,
 static void hns_roce_mr_free(struct hns_roce_dev *hr_dev,
 			     struct hns_roce_mr *mr)
 {
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	int npages = 0;
 	int ret;
 
@@ -302,7 +302,7 @@ static int hns_roce_mr_enable(struct hns_roce_dev *hr_dev,
 {
 	int ret;
 	unsigned long mtpt_idx = key_to_hw_index(mr->key);
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	struct hns_roce_cmd_mailbox *mailbox;
 	struct hns_roce_mr_table *mr_table = &hr_dev->mr_table;
 
@@ -552,7 +552,7 @@ struct ib_mr *hns_roce_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 				   struct ib_udata *udata)
 {
 	struct hns_roce_dev *hr_dev = to_hr_dev(pd->device);
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	struct hns_roce_mr *mr = NULL;
 	int ret = 0;
 	int n = 0;
diff --git a/drivers/infiniband/hw/hns/hns_roce_pd.c b/drivers/infiniband/hw/hns/hns_roce_pd.c
index ffbf249..079bb10 100644
--- a/drivers/infiniband/hw/hns/hns_roce_pd.c
+++ b/drivers/infiniband/hw/hns/hns_roce_pd.c
@@ -60,7 +60,7 @@ struct ib_pd *hns_roce_alloc_pd(struct ib_device *ib_dev,
 				struct ib_udata *udata)
 {
 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	struct hns_roce_pd *pd;
 	int ret;
 
diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c
index ad13d05..b512ecc 100644
--- a/drivers/infiniband/hw/hns/hns_roce_qp.c
+++ b/drivers/infiniband/hw/hns/hns_roce_qp.c
@@ -44,7 +44,7 @@
 void hns_roce_qp_event(struct hns_roce_dev *hr_dev, u32 qpn, int event_type)
 {
 	struct hns_roce_qp_table *qp_table = &hr_dev->qp_table;
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	struct hns_roce_qp *qp;
 
 	spin_lock(&qp_table->lock);
@@ -154,7 +154,7 @@ static int hns_roce_gsi_qp_alloc(struct hns_roce_dev *hr_dev, unsigned long qpn,
 				hr_qp->qpn & (hr_dev->caps.num_qps - 1), hr_qp);
 	spin_unlock_irq(&qp_table->lock);
 	if (ret) {
-		dev_err(&hr_dev->pdev->dev, "QPC radix_tree_insert failed\n");
+		dev_err(hr_dev->dev, "QPC radix_tree_insert failed\n");
 		goto err_put_irrl;
 	}
 
@@ -172,7 +172,7 @@ static int hns_roce_qp_alloc(struct hns_roce_dev *hr_dev, unsigned long qpn,
 			     struct hns_roce_qp *hr_qp)
 {
 	struct hns_roce_qp_table *qp_table = &hr_dev->qp_table;
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	int ret;
 
 	if (!qpn)
@@ -261,8 +261,8 @@ static int hns_roce_set_rq_size(struct hns_roce_dev *hr_dev,
 				struct ib_qp_cap *cap, int is_user, int has_srq,
 				struct hns_roce_qp *hr_qp)
 {
+	struct device *dev = hr_dev->dev;
 	u32 max_cnt;
-	struct device *dev = &hr_dev->pdev->dev;
 
 	/* Check the validity of QP support capacity */
 	if (cap->max_recv_wr > hr_dev->caps.max_wqes ||
@@ -319,7 +319,7 @@ static int hns_roce_set_user_sq_size(struct hns_roce_dev *hr_dev,
 	if ((u32)(1 << ucmd->log_sq_bb_count) > hr_dev->caps.max_wqes ||
 	     ucmd->log_sq_stride > max_sq_stride ||
 	     ucmd->log_sq_stride < HNS_ROCE_IB_MIN_SQ_STRIDE) {
-		dev_err(&hr_dev->pdev->dev, "check SQ size error!\n");
+		dev_err(hr_dev->dev, "check SQ size error!\n");
 		return -EINVAL;
 	}
 
@@ -343,7 +343,7 @@ static int hns_roce_set_kernel_sq_size(struct hns_roce_dev *hr_dev,
 				       struct ib_qp_cap *cap,
 				       struct hns_roce_qp *hr_qp)
 {
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	u32 max_cnt;
 
 	if (cap->max_send_wr  > hr_dev->caps.max_wqes  ||
@@ -395,7 +395,7 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev,
 				     struct ib_udata *udata, unsigned long sqpn,
 				     struct hns_roce_qp *hr_qp)
 {
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	struct hns_roce_ib_create_qp ucmd;
 	unsigned long qpn = 0;
 	int ret = 0;
@@ -575,7 +575,7 @@ struct ib_qp *hns_roce_create_qp(struct ib_pd *pd,
 				 struct ib_udata *udata)
 {
 	struct hns_roce_dev *hr_dev = to_hr_dev(pd->device);
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	struct hns_roce_sqp *hr_sqp;
 	struct hns_roce_qp *hr_qp;
 	int ret;
@@ -660,7 +660,7 @@ int hns_roce_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
 	enum ib_qp_state cur_state, new_state;
-	struct device *dev = &hr_dev->pdev->dev;
+	struct device *dev = hr_dev->dev;
 	int ret = -EINVAL;
 	int p;
 	enum ib_mtu active_mtu;
@@ -835,7 +835,7 @@ int hns_roce_init_qp_table(struct hns_roce_dev *hr_dev)
 				   hr_dev->caps.num_qps - 1, SQP_NUM,
 				   reserved_from_top);
 	if (ret) {
-		dev_err(&hr_dev->pdev->dev, "qp bitmap init failed!error=%d\n",
+		dev_err(hr_dev->dev, "qp bitmap init failed!error=%d\n",
 			ret);
 		return ret;
 	}
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Wei Hu (Xavier)]

The command queue is the configuration queue. The software
configures hardware by filling the commands into command
queues. It includes command send queue and receive queue.

In hip08 RoCE engine, It supports to configure and query
registers by command queue.

Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_common.h |  13 ++
 drivers/infiniband/hw/hns/hns_roce_device.h |   2 +
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 281 +++++++++++++++++++++++++++-
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  | 111 +++++++++++
 drivers/infiniband/hw/hns/hns_roce_main.c   |  14 ++
 5 files changed, 420 insertions(+), 1 deletion(-)
 create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.h

diff --git a/drivers/infiniband/hw/hns/hns_roce_common.h b/drivers/infiniband/hw/hns/hns_roce_common.h
index 4af403e..94381c2 100644
--- a/drivers/infiniband/hw/hns/hns_roce_common.h
+++ b/drivers/infiniband/hw/hns/hns_roce_common.h
@@ -362,4 +362,17 @@
 #define ROCEE_ECC_UCERR_ALM0_REG		0xB34
 #define ROCEE_ECC_CERR_ALM0_REG			0xB40
 
+/* V2 ROCEE REG */
+#define ROCEE_TX_CMQ_BASEADDR_L_REG		0x07000
+#define ROCEE_TX_CMQ_BASEADDR_H_REG		0x07004
+#define ROCEE_TX_CMQ_DEPTH_REG			0x07008
+#define ROCEE_TX_CMQ_TAIL_REG			0x07010
+#define ROCEE_TX_CMQ_HEAD_REG			0x07014
+
+#define ROCEE_RX_CMQ_BASEADDR_L_REG		0x07018
+#define ROCEE_RX_CMQ_BASEADDR_H_REG		0x0701c
+#define ROCEE_RX_CMQ_DEPTH_REG			0x07020
+#define ROCEE_RX_CMQ_TAIL_REG			0x07024
+#define ROCEE_RX_CMQ_HEAD_REG			0x07028
+
 #endif /* _HNS_ROCE_COMMON_H */
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 3132b68..87c2e07 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -506,6 +506,8 @@ struct hns_roce_caps {
 
 struct hns_roce_hw {
 	int (*reset)(struct hns_roce_dev *hr_dev, bool enable);
+	int (*cmq_init)(struct hns_roce_dev *hr_dev);
+	void (*cmq_exit)(struct hns_roce_dev *hr_dev);
 	void (*hw_profile)(struct hns_roce_dev *hr_dev);
 	int (*hw_init)(struct hns_roce_dev *hr_dev);
 	void (*hw_exit)(struct hns_roce_dev *hr_dev);
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index b35c72d..b610f3a 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -41,8 +41,277 @@
 #include "hns_roce_device.h"
 #include "hns_roce_cmd.h"
 #include "hns_roce_hem.h"
+#include "hns_roce_hw_v2.h"
 
-static const struct hns_roce_hw hns_roce_hw_v2;
+static int hns_roce_cmq_space(struct hns_roce_v2_cmq_ring *ring)
+{
+	int ntu = ring->next_to_use;
+	int ntc = ring->next_to_clean;
+	int used = (ntu - ntc + ring->desc_num) % ring->desc_num;
+
+	return ring->desc_num - used - 1;
+}
+
+static int hns_roce_alloc_cmq_desc(struct hns_roce_dev *hr_dev,
+				   struct hns_roce_v2_cmq_ring *ring)
+{
+	int size = ring->desc_num * sizeof(struct hns_roce_cmq_desc);
+
+	ring->desc = kzalloc(size, GFP_KERNEL);
+	if (!ring->desc)
+		return -ENOMEM;
+
+	ring->desc_dma_addr = dma_map_single(hr_dev->dev, ring->desc, size,
+					     DMA_BIDIRECTIONAL);
+	if (dma_mapping_error(hr_dev->dev, ring->desc_dma_addr)) {
+		ring->desc_dma_addr = 0;
+		kfree(ring->desc);
+		ring->desc = NULL;
+		return -ENOMEM;
+	}
+
+	return 0;
+}
+
+static void hns_roce_free_cmq_desc(struct hns_roce_dev *hr_dev,
+				   struct hns_roce_v2_cmq_ring *ring)
+{
+	dma_unmap_single(hr_dev->dev, ring->desc_dma_addr,
+			 ring->desc_num * sizeof(struct hns_roce_cmq_desc),
+			 DMA_BIDIRECTIONAL);
+	kfree(ring->desc);
+}
+
+static int hns_roce_init_cmq_ring(struct hns_roce_dev *hr_dev, bool ring_type)
+{
+	struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv;
+	struct hns_roce_v2_cmq_ring *ring = (ring_type == TYPE_CSQ) ?
+					    &priv->cmq.csq : &priv->cmq.crq;
+
+	ring->flag = ring_type;
+	ring->next_to_clean = 0;
+	ring->next_to_use = 0;
+
+	return hns_roce_alloc_cmq_desc(hr_dev, ring);
+}
+
+static void hns_roce_cmq_init_regs(struct hns_roce_dev *hr_dev, bool ring_type)
+{
+	struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv;
+	struct hns_roce_v2_cmq_ring *ring = (ring_type == TYPE_CSQ) ?
+					    &priv->cmq.csq : &priv->cmq.crq;
+	dma_addr_t dma = ring->desc_dma_addr;
+
+	if (ring_type == TYPE_CSQ) {
+		roce_write(hr_dev, ROCEE_TX_CMQ_BASEADDR_L_REG, (u32)dma);
+		roce_write(hr_dev, ROCEE_TX_CMQ_BASEADDR_H_REG,
+			   upper_32_bits(dma));
+		roce_write(hr_dev, ROCEE_TX_CMQ_DEPTH_REG,
+			  (ring->desc_num >> HNS_ROCE_CMQ_DESC_NUM_S) |
+			   HNS_ROCE_CMQ_ENABLE);
+		roce_write(hr_dev, ROCEE_TX_CMQ_HEAD_REG, 0);
+		roce_write(hr_dev, ROCEE_TX_CMQ_TAIL_REG, 0);
+	} else {
+		roce_write(hr_dev, ROCEE_RX_CMQ_BASEADDR_L_REG, (u32)dma);
+		roce_write(hr_dev, ROCEE_RX_CMQ_BASEADDR_H_REG,
+			   upper_32_bits(dma));
+		roce_write(hr_dev, ROCEE_RX_CMQ_DEPTH_REG,
+			  (ring->desc_num >> HNS_ROCE_CMQ_DESC_NUM_S) |
+			   HNS_ROCE_CMQ_ENABLE);
+		roce_write(hr_dev, ROCEE_RX_CMQ_HEAD_REG, 0);
+		roce_write(hr_dev, ROCEE_RX_CMQ_TAIL_REG, 0);
+	}
+}
+
+static int hns_roce_v2_cmq_init(struct hns_roce_dev *hr_dev)
+{
+	struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv;
+	int ret;
+
+	/* Setup the queue entries for command queue */
+	priv->cmq.csq.desc_num = 1024;
+	priv->cmq.crq.desc_num = 1024;
+
+	/* Setup the lock for command queue */
+	spin_lock_init(&priv->cmq.csq.lock);
+	spin_lock_init(&priv->cmq.crq.lock);
+
+	/* Setup Tx write back timeout */
+	priv->cmq.tx_timeout = HNS_ROCE_CMQ_TX_TIMEOUT;
+
+	/* Init CSQ */
+	ret = hns_roce_init_cmq_ring(hr_dev, TYPE_CSQ);
+	if (ret) {
+		dev_err(hr_dev->dev, "Init CSQ error, ret = %d.\n", ret);
+		return ret;
+	}
+
+	/* Init CRQ */
+	ret = hns_roce_init_cmq_ring(hr_dev, TYPE_CRQ);
+	if (ret) {
+		dev_err(hr_dev->dev, "Init CRQ error, ret = %d.\n", ret);
+		goto err_crq;
+	}
+
+	/* Init CSQ REG */
+	hns_roce_cmq_init_regs(hr_dev, TYPE_CSQ);
+
+	/* Init CRQ REG */
+	hns_roce_cmq_init_regs(hr_dev, TYPE_CRQ);
+
+	return 0;
+
+err_crq:
+	hns_roce_free_cmq_desc(hr_dev, &priv->cmq.csq);
+
+	return ret;
+}
+
+static void hns_roce_v2_cmq_exit(struct hns_roce_dev *hr_dev)
+{
+	struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv;
+
+	hns_roce_free_cmq_desc(hr_dev, &priv->cmq.csq);
+	hns_roce_free_cmq_desc(hr_dev, &priv->cmq.crq);
+}
+
+void hns_roce_cmq_setup_basic_desc(struct hns_roce_cmq_desc *desc,
+				   enum hns_roce_opcode_type opcode,
+				   bool is_read)
+{
+	memset((void *)desc, 0, sizeof(struct hns_roce_cmq_desc));
+	desc->opcode = cpu_to_le16(opcode);
+	desc->flag =
+		cpu_to_le16(HNS_ROCE_CMD_FLAG_NO_INTR | HNS_ROCE_CMD_FLAG_IN);
+	if (is_read)
+		desc->flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_WR);
+	else
+		desc->flag &= cpu_to_le16(~HNS_ROCE_CMD_FLAG_WR);
+}
+
+static int hns_roce_cmq_csq_done(struct hns_roce_dev *hr_dev)
+{
+	struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv;
+	u32 head = roce_read(hr_dev, ROCEE_TX_CMQ_HEAD_REG);
+
+	return head == priv->cmq.csq.next_to_use;
+}
+
+static int hns_roce_cmq_csq_clean(struct hns_roce_dev *hr_dev)
+{
+	struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv;
+	struct hns_roce_v2_cmq_ring *csq = &priv->cmq.csq;
+	struct hns_roce_cmq_desc *desc;
+	u16 ntc = csq->next_to_clean;
+	u32 head;
+	int clean = 0;
+
+	desc = &csq->desc[ntc];
+	head = roce_read(hr_dev, ROCEE_TX_CMQ_HEAD_REG);
+	while (head != ntc) {
+		memset(desc, 0, sizeof(*desc));
+		ntc++;
+		if (ntc == csq->desc_num)
+			ntc = 0;
+		desc = &csq->desc[ntc];
+		clean++;
+	}
+	csq->next_to_clean = ntc;
+
+	return clean;
+}
+
+int hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
+		      struct hns_roce_cmq_desc *desc, int num)
+{
+	struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv;
+	struct hns_roce_v2_cmq_ring *csq = &priv->cmq.csq;
+	struct hns_roce_cmq_desc *desc_to_use;
+	bool complete = false;
+	u32 timeout = 0;
+	int handle = 0;
+	u16 desc_ret;
+	int ret = 0;
+	int ntc;
+
+	spin_lock_bh(&csq->lock);
+
+	if (num > hns_roce_cmq_space(csq)) {
+		spin_unlock_bh(&csq->lock);
+		return -EBUSY;
+	}
+
+	/*
+	 * Record the location of desc in the cmq for this time
+	 * which will be use for hardware to write back
+	 */
+	ntc = csq->next_to_use;
+
+	while (handle < num) {
+		desc_to_use = &csq->desc[csq->next_to_use];
+		*desc_to_use = desc[handle];
+		dev_dbg(hr_dev->dev, "set cmq desc:\n");
+		csq->next_to_use++;
+		if (csq->next_to_use == csq->desc_num)
+			csq->next_to_use = 0;
+		handle++;
+	}
+
+	/* Write to hardware */
+	roce_write(hr_dev, ROCEE_TX_CMQ_TAIL_REG, csq->next_to_use);
+
+	/*
+	 * If the command is sync, wait for the firmware to write back,
+	 * if multi descriptors to be sent, use the first one to check
+	 */
+	if ((desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
+		do {
+			if (hns_roce_cmq_csq_done(hr_dev))
+				break;
+			usleep_range(1000, 2000);
+			timeout++;
+		} while (timeout < priv->cmq.tx_timeout);
+	}
+
+	if (hns_roce_cmq_csq_done(hr_dev)) {
+		complete = true;
+		handle = 0;
+		while (handle < num) {
+			/* get the result of hardware write back */
+			desc_to_use = &csq->desc[ntc];
+			desc[handle] = *desc_to_use;
+			dev_dbg(hr_dev->dev, "Get cmq desc:\n");
+			desc_ret = desc[handle].retval;
+			if (desc_ret == CMD_EXEC_SUCCESS)
+				ret = 0;
+			else
+				ret = -EIO;
+			priv->cmq.last_status = desc_ret;
+			ntc++;
+			handle++;
+			if (ntc == csq->desc_num)
+				ntc = 0;
+		}
+	}
+
+	if (!complete)
+		ret = -EAGAIN;
+
+	/* clean the command send queue */
+	handle = hns_roce_cmq_csq_clean(hr_dev);
+	if (handle != num)
+		dev_warn(hr_dev->dev, "Cleaned %d, need to clean %d\n",
+			 handle, num);
+
+	spin_unlock_bh(&csq->lock);
+
+	return ret;
+}
+
+static const struct hns_roce_hw hns_roce_hw_v2 = {
+	.cmq_init = hns_roce_v2_cmq_init,
+	.cmq_exit = hns_roce_v2_cmq_exit,
+};
 
 static const struct pci_device_id hns_roce_hw_v2_pci_tbl[] = {
 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA), 0},
@@ -87,6 +356,12 @@ static int hns_roce_hw_v2_init_instance(struct hnae3_handle *handle)
 	if (!hr_dev)
 		return -ENOMEM;
 
+	hr_dev->priv = kzalloc(sizeof(struct hns_roce_v2_priv), GFP_KERNEL);
+	if (!hr_dev->priv) {
+		ret = -ENOMEM;
+		goto error_failed_kzalloc;
+	}
+
 	hr_dev->pci_dev = handle->pdev;
 	hr_dev->dev = &handle->pdev->dev;
 	handle->priv = hr_dev;
@@ -106,6 +381,9 @@ static int hns_roce_hw_v2_init_instance(struct hnae3_handle *handle)
 	return 0;
 
 error_failed_get_cfg:
+	kfree(hr_dev->priv);
+
+error_failed_kzalloc:
 	ib_dealloc_device(&hr_dev->ib_dev);
 
 	return ret;
@@ -117,6 +395,7 @@ static void hns_roce_hw_v2_uninit_instance(struct hnae3_handle *handle,
 	struct hns_roce_dev *hr_dev = (struct hns_roce_dev *)handle->priv;
 
 	hns_roce_exit(hr_dev);
+	kfree(hr_dev->priv);
 	ib_dealloc_device(&hr_dev->ib_dev);
 }
 
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
new file mode 100644
index 0000000..781ae74
--- /dev/null
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -0,0 +1,111 @@
+/*
+ * Copyright (c) 2016-2017 Hisilicon Limited.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef _HNS_ROCE_HW_V2_H
+#define _HNS_ROCE_HW_V2_H
+
+#define HNS_ROCE_CMQ_TX_TIMEOUT			200
+
+#define HNS_ROCE_CMD_FLAG_IN_VALID_SHIFT	0
+#define HNS_ROCE_CMD_FLAG_OUT_VALID_SHIFT	1
+#define HNS_ROCE_CMD_FLAG_NEXT_SHIFT		2
+#define HNS_ROCE_CMD_FLAG_WR_OR_RD_SHIFT	3
+#define HNS_ROCE_CMD_FLAG_NO_INTR_SHIFT		4
+#define HNS_ROCE_CMD_FLAG_ERR_INTR_SHIFT	5
+
+#define HNS_ROCE_CMD_FLAG_IN		BIT(HNS_ROCE_CMD_FLAG_IN_VALID_SHIFT)
+#define HNS_ROCE_CMD_FLAG_OUT		BIT(HNS_ROCE_CMD_FLAG_OUT_VALID_SHIFT)
+#define HNS_ROCE_CMD_FLAG_NEXT		BIT(HNS_ROCE_CMD_FLAG_NEXT_SHIFT)
+#define HNS_ROCE_CMD_FLAG_WR		BIT(HNS_ROCE_CMD_FLAG_WR_OR_RD_SHIFT)
+#define HNS_ROCE_CMD_FLAG_NO_INTR	BIT(HNS_ROCE_CMD_FLAG_NO_INTR_SHIFT)
+#define HNS_ROCE_CMD_FLAG_ERR_INTR	BIT(HNS_ROCE_CMD_FLAG_ERR_INTR_SHIFT)
+
+#define HNS_ROCE_CMQ_DESC_NUM_S		3
+#define HNS_ROCE_CMQ_EN_B		16
+#define HNS_ROCE_CMQ_ENABLE		BIT(HNS_ROCE_CMQ_EN_B)
+
+/* CMQ command */
+enum hns_roce_opcode_type {
+	HNS_ROCE_OPC_QUERY_HW_VER			= 0x8000,
+	HNS_ROCE_OPC_CFG_GLOBAL_PARAM			= 0x8001,
+	HNS_ROCE_OPC_ALLOC_PF_RES			= 0x8004,
+	HNS_ROCE_OPC_QUERY_PF_RES			= 0x8400,
+	HNS_ROCE_OPC_ALLOC_VF_RES			= 0x8401,
+	HNS_ROCE_OPC_CFG_BT_ATTR			= 0x8506,
+};
+
+enum {
+	TYPE_CRQ,
+	TYPE_CSQ,
+};
+
+enum hns_roce_cmd_return_status {
+	CMD_EXEC_SUCCESS	= 0,
+	CMD_NO_AUTH		= 1,
+	CMD_NOT_EXEC		= 2,
+	CMD_QUEUE_FULL		= 3,
+};
+
+struct hns_roce_cmq_desc {
+	u16 opcode;
+	u16 flag;
+	u16 retval;
+	u16 rsv;
+	u32 data[6];
+};
+
+struct hns_roce_v2_cmq_ring {
+	dma_addr_t desc_dma_addr;
+	struct hns_roce_cmq_desc *desc;
+	u32 head;
+	u32 tail;
+
+	u16 buf_size;
+	u16 desc_num;
+	int next_to_use;
+	int next_to_clean;
+	u8 flag;
+	spinlock_t lock; /* command queue lock */
+};
+
+struct hns_roce_v2_cmq {
+	struct hns_roce_v2_cmq_ring csq;
+	struct hns_roce_v2_cmq_ring crq;
+	u16 tx_timeout;
+	u16 last_status;
+};
+
+struct hns_roce_v2_priv {
+	struct hns_roce_v2_cmq cmq;
+};
+
+#endif
diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c
index b07d437..4f5a6fd 100644
--- a/drivers/infiniband/hw/hns/hns_roce_main.c
+++ b/drivers/infiniband/hw/hns/hns_roce_main.c
@@ -678,6 +678,14 @@ int hns_roce_init(struct hns_roce_dev *hr_dev)
 		}
 	}
 
+	if (hr_dev->hw->cmq_init) {
+		ret = hr_dev->hw->cmq_init(hr_dev);
+		if (ret) {
+			dev_err(dev, "Init RoCE Command Queue failed!\n");
+			goto error_failed_cmq_init;
+		}
+	}
+
 	hr_dev->hw->hw_profile(hr_dev);
 
 	ret = hns_roce_cmd_init(hr_dev);
@@ -750,6 +758,10 @@ int hns_roce_init(struct hns_roce_dev *hr_dev)
 	hns_roce_cmd_cleanup(hr_dev);
 
 error_failed_cmd_init:
+	if (hr_dev->hw->cmq_exit)
+		hr_dev->hw->cmq_exit(hr_dev);
+
+error_failed_cmq_init:
 	if (hr_dev->hw->reset) {
 		ret = hr_dev->hw->reset(hr_dev, false);
 		if (ret)
@@ -774,6 +786,8 @@ void hns_roce_exit(struct hns_roce_dev *hr_dev)
 	if (hr_dev->cmd_mod)
 		hns_roce_cleanup_eq_table(hr_dev);
 	hns_roce_cmd_cleanup(hr_dev);
+	if (hr_dev->hw->cmq_exit)
+		hr_dev->hw->cmq_exit(hr_dev);
 	if (hr_dev->hw->reset)
 		hr_dev->hw->reset(hr_dev, false);
 }
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 06/20] RDMA/hns: Add profile support for hip08 driver

[Wei Hu (Xavier)]

The profile's content mainly set some specifications and obtain
some hardware resources by implementing the relative commands.
Because max sge num of send queue is not the same with receive
queue in hip08, we modified the calculation of props->max_sge
in query_device ops.

Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_device.h |   8 +-
 drivers/infiniband/hw/hns/hns_roce_hw_v1.c  |   4 +-
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 246 ++++++++++++++++++++++++++++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  | 161 ++++++++++++++++++
 drivers/infiniband/hw/hns/hns_roce_main.c   |   8 +-
 5 files changed, 423 insertions(+), 4 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 87c2e07..8ec6d6e 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -477,6 +477,7 @@ struct hns_roce_caps {
 	u32		max_wqes;	/* 16k */
 	u32		max_sq_desc_sz;	/* 64 */
 	u32		max_rq_desc_sz;	/* 64 */
+	u32		max_srq_desc_sz;
 	int		max_qp_init_rdma;
 	int		max_qp_dest_rdma;
 	int		num_cqs;
@@ -487,6 +488,7 @@ struct hns_roce_caps {
 	int		num_other_vectors;
 	int		num_mtpts;
 	u32		num_mtt_segs;
+	u32		num_cqe_segs;
 	int		reserved_mrws;
 	int		reserved_uars;
 	int		num_pds;
@@ -502,13 +504,17 @@ struct hns_roce_caps {
 	int		aeqe_depth;
 	int		ceqe_depth[HNS_ROCE_COMP_VEC_NUM];
 	enum ib_mtu	max_mtu;
+	u32		qpc_bt_num;
+	u32		srqc_bt_num;
+	u32		cqc_bt_num;
+	u32		mpt_bt_num;
 };
 
 struct hns_roce_hw {
 	int (*reset)(struct hns_roce_dev *hr_dev, bool enable);
 	int (*cmq_init)(struct hns_roce_dev *hr_dev);
 	void (*cmq_exit)(struct hns_roce_dev *hr_dev);
-	void (*hw_profile)(struct hns_roce_dev *hr_dev);
+	int (*hw_profile)(struct hns_roce_dev *hr_dev);
 	int (*hw_init)(struct hns_roce_dev *hr_dev);
 	void (*hw_exit)(struct hns_roce_dev *hr_dev);
 	void (*set_gid)(struct hns_roce_dev *hr_dev, u8 port, int gid_index,
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 22cc3fb..267e400 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -1459,7 +1459,7 @@ static void hns_roce_des_qp_free(struct hns_roce_dev *hr_dev)
 	destroy_workqueue(des_qp->qp_wq);
 }
 
-void hns_roce_v1_profile(struct hns_roce_dev *hr_dev)
+int hns_roce_v1_profile(struct hns_roce_dev *hr_dev)
 {
 	int i = 0;
 	struct hns_roce_caps *caps = &hr_dev->caps;
@@ -1525,6 +1525,8 @@ void hns_roce_v1_profile(struct hns_roce_dev *hr_dev)
 	caps->local_ca_ack_delay = le32_to_cpu(roce_read(hr_dev,
 							 ROCEE_ACK_DELAY_REG));
 	caps->max_mtu = IB_MTU_2048;
+
+	return 0;
 }
 
 int hns_roce_v1_init(struct hns_roce_dev *hr_dev)
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index b610f3a..af3cb30 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -308,9 +308,255 @@ int hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 	return ret;
 }
 
+int hns_roce_cmq_query_hw_info(struct hns_roce_dev *hr_dev)
+{
+	struct hns_roce_query_version *resp;
+	struct hns_roce_cmq_desc desc;
+	int ret;
+
+	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_QUERY_HW_VER, true);
+	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
+	if (ret)
+		return ret;
+
+	resp = (struct hns_roce_query_version *)desc.data;
+	hr_dev->hw_rev = le32_to_cpu(resp->rocee_hw_version);
+	hr_dev->vendor_id = le32_to_cpu(resp->rocee_vendor_id);
+
+	return 0;
+}
+
+static int hns_roce_config_global_param(struct hns_roce_dev *hr_dev)
+{
+	struct hns_roce_cfg_global_param *req;
+	struct hns_roce_cmq_desc desc;
+
+	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_GLOBAL_PARAM,
+				      false);
+
+	req = (struct hns_roce_cfg_global_param *)desc.data;
+	memset(req, 0, sizeof(*req));
+	roce_set_field(req->time_cfg_udp_port,
+		       CFG_GLOBAL_PARAM_DATA_0_ROCEE_TIME_1US_CFG_M,
+		       CFG_GLOBAL_PARAM_DATA_0_ROCEE_TIME_1US_CFG_S, 0x3e8);
+	roce_set_field(req->time_cfg_udp_port,
+		       CFG_GLOBAL_PARAM_DATA_0_ROCEE_UDP_PORT_M,
+		       CFG_GLOBAL_PARAM_DATA_0_ROCEE_UDP_PORT_S, 0x12b7);
+
+	return hns_roce_cmq_send(hr_dev, &desc, 1);
+}
+
+static int hns_roce_query_pf_resource(struct hns_roce_dev *hr_dev)
+{
+	struct hns_roce_cmq_desc desc[2];
+	struct hns_roce_pf_res *res;
+	int ret;
+	int i;
+
+	for (i = 0; i < 2; i++) {
+		hns_roce_cmq_setup_basic_desc(&desc[i],
+					      HNS_ROCE_OPC_QUERY_PF_RES, true);
+
+		if (i == 0)
+			desc[i].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
+		else
+			desc[i].flag &= ~cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
+	}
+
+	ret = hns_roce_cmq_send(hr_dev, desc, 2);
+	if (ret)
+		return ret;
+
+	res = (struct hns_roce_pf_res *)desc[0].data;
+
+	hr_dev->caps.qpc_bt_num = roce_get_field(res->qpc_bt_idx_num,
+						 PF_RES_DATA_1_PF_QPC_BT_NUM_M,
+						 PF_RES_DATA_1_PF_QPC_BT_NUM_S);
+	hr_dev->caps.srqc_bt_num = roce_get_field(res->srqc_bt_idx_num,
+						PF_RES_DATA_2_PF_SRQC_BT_NUM_M,
+						PF_RES_DATA_2_PF_SRQC_BT_NUM_S);
+	hr_dev->caps.cqc_bt_num = roce_get_field(res->cqc_bt_idx_num,
+						 PF_RES_DATA_3_PF_CQC_BT_NUM_M,
+						 PF_RES_DATA_3_PF_CQC_BT_NUM_S);
+	hr_dev->caps.mpt_bt_num = roce_get_field(res->mpt_bt_idx_num,
+						 PF_RES_DATA_4_PF_MPT_BT_NUM_M,
+						 PF_RES_DATA_4_PF_MPT_BT_NUM_S);
+
+	return 0;
+}
+
+static int hns_roce_alloc_vf_resource(struct hns_roce_dev *hr_dev)
+{
+	struct hns_roce_cmq_desc desc[2];
+	struct hns_roce_vf_res_a *req_a;
+	struct hns_roce_vf_res_b *req_b;
+	int i;
+
+	req_a = (struct hns_roce_vf_res_a *)desc[0].data;
+	req_b = (struct hns_roce_vf_res_b *)desc[1].data;
+	memset(req_a, 0, sizeof(*req_a));
+	memset(req_b, 0, sizeof(*req_b));
+	for (i = 0; i < 2; i++) {
+		hns_roce_cmq_setup_basic_desc(&desc[i],
+					      HNS_ROCE_OPC_ALLOC_VF_RES, false);
+
+		if (i == 0)
+			desc[i].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
+		else
+			desc[i].flag &= ~cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
+
+		if (i == 0) {
+			roce_set_field(req_a->vf_qpc_bt_idx_num,
+				       VF_RES_A_DATA_1_VF_QPC_BT_IDX_M,
+				       VF_RES_A_DATA_1_VF_QPC_BT_IDX_S, 0);
+			roce_set_field(req_a->vf_qpc_bt_idx_num,
+				       VF_RES_A_DATA_1_VF_QPC_BT_NUM_M,
+				       VF_RES_A_DATA_1_VF_QPC_BT_NUM_S,
+				       HNS_ROCE_VF_QPC_BT_NUM);
+
+			roce_set_field(req_a->vf_srqc_bt_idx_num,
+				       VF_RES_A_DATA_2_VF_SRQC_BT_IDX_M,
+				       VF_RES_A_DATA_2_VF_SRQC_BT_IDX_S, 0);
+			roce_set_field(req_a->vf_srqc_bt_idx_num,
+				       VF_RES_A_DATA_2_VF_SRQC_BT_NUM_M,
+				       VF_RES_A_DATA_2_VF_SRQC_BT_NUM_S,
+				       HNS_ROCE_VF_SRQC_BT_NUM);
+
+			roce_set_field(req_a->vf_cqc_bt_idx_num,
+				       VF_RES_A_DATA_3_VF_CQC_BT_IDX_M,
+				       VF_RES_A_DATA_3_VF_CQC_BT_IDX_S, 0);
+			roce_set_field(req_a->vf_cqc_bt_idx_num,
+				       VF_RES_A_DATA_3_VF_CQC_BT_NUM_M,
+				       VF_RES_A_DATA_3_VF_CQC_BT_NUM_S,
+				       HNS_ROCE_VF_CQC_BT_NUM);
+
+			roce_set_field(req_a->vf_mpt_bt_idx_num,
+				       VF_RES_A_DATA_4_VF_MPT_BT_IDX_M,
+				       VF_RES_A_DATA_4_VF_MPT_BT_IDX_S, 0);
+			roce_set_field(req_a->vf_mpt_bt_idx_num,
+				       VF_RES_A_DATA_4_VF_MPT_BT_NUM_M,
+				       VF_RES_A_DATA_4_VF_MPT_BT_NUM_S,
+				       HNS_ROCE_VF_MPT_BT_NUM);
+
+			roce_set_field(req_a->vf_eqc_bt_idx_num,
+				       VF_RES_A_DATA_5_VF_EQC_IDX_M,
+				       VF_RES_A_DATA_5_VF_EQC_IDX_S, 0);
+			roce_set_field(req_a->vf_eqc_bt_idx_num,
+				       VF_RES_A_DATA_5_VF_EQC_NUM_M,
+				       VF_RES_A_DATA_5_VF_EQC_NUM_S,
+				       HNS_ROCE_VF_EQC_NUM);
+		} else {
+			roce_set_field(req_b->vf_smac_idx_num,
+				       VF_RES_B_DATA_1_VF_SMAC_IDX_M,
+				       VF_RES_B_DATA_1_VF_SMAC_IDX_S, 0);
+			roce_set_field(req_b->vf_smac_idx_num,
+				       VF_RES_B_DATA_1_VF_SMAC_NUM_M,
+				       VF_RES_B_DATA_1_VF_SMAC_NUM_S,
+				       HNS_ROCE_VF_SMAC_NUM);
+
+			roce_set_field(req_b->vf_sgid_idx_num,
+				       VF_RES_B_DATA_2_VF_SGID_IDX_M,
+				       VF_RES_B_DATA_2_VF_SGID_IDX_S, 0);
+			roce_set_field(req_b->vf_sgid_idx_num,
+				       VF_RES_B_DATA_2_VF_SGID_NUM_M,
+				       VF_RES_B_DATA_2_VF_SGID_NUM_S,
+				       HNS_ROCE_VF_SGID_NUM);
+
+			roce_set_field(req_b->vf_qid_idx_sl_num,
+				       VF_RES_B_DATA_3_VF_QID_IDX_M,
+				       VF_RES_B_DATA_3_VF_QID_IDX_S, 0);
+			roce_set_field(req_b->vf_qid_idx_sl_num,
+				       VF_RES_B_DATA_3_VF_SL_NUM_M,
+				       VF_RES_B_DATA_3_VF_SL_NUM_S,
+				       HNS_ROCE_VF_SL_NUM);
+		}
+	}
+
+	return hns_roce_cmq_send(hr_dev, desc, 2);
+}
+
+static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev)
+{
+	struct hns_roce_caps *caps = &hr_dev->caps;
+	int ret;
+
+	ret = hns_roce_cmq_query_hw_info(hr_dev);
+	if (ret) {
+		dev_err(hr_dev->dev, "Query firmware version fail, ret = %d.\n",
+			ret);
+		return ret;
+	}
+
+	ret = hns_roce_config_global_param(hr_dev);
+	if (ret) {
+		dev_err(hr_dev->dev, "Configure global param fail, ret = %d.\n",
+			ret);
+	}
+
+	/* Get pf resource owned by every pf */
+	ret = hns_roce_query_pf_resource(hr_dev);
+	if (ret) {
+		dev_err(hr_dev->dev, "Query pf resource fail, ret = %d.\n",
+			ret);
+		return ret;
+	}
+
+	ret = hns_roce_alloc_vf_resource(hr_dev);
+	if (ret) {
+		dev_err(hr_dev->dev, "Allocate vf resource fail, ret = %d.\n",
+			ret);
+		return ret;
+	}
+
+	hr_dev->vendor_part_id = 0;
+	hr_dev->sys_image_guid = 0;
+
+	caps->num_qps		= HNS_ROCE_V2_MAX_QP_NUM;
+	caps->max_wqes		= HNS_ROCE_V2_MAX_WQE_NUM;
+	caps->num_cqs		= HNS_ROCE_V2_MAX_CQ_NUM;
+	caps->max_cqes		= HNS_ROCE_V2_MAX_CQE_NUM;
+	caps->max_sq_sg		= HNS_ROCE_V2_MAX_SQ_SGE_NUM;
+	caps->max_rq_sg		= HNS_ROCE_V2_MAX_RQ_SGE_NUM;
+	caps->max_sq_inline	= HNS_ROCE_V2_MAX_SQ_INLINE;
+	caps->num_uars		= HNS_ROCE_V2_UAR_NUM;
+	caps->phy_num_uars	= HNS_ROCE_V2_PHY_UAR_NUM;
+	caps->num_aeq_vectors	= 1;
+	caps->num_comp_vectors	= 63;
+	caps->num_other_vectors	= 0;
+	caps->num_mtpts		= HNS_ROCE_V2_MAX_MTPT_NUM;
+	caps->num_mtt_segs	= HNS_ROCE_V2_MAX_MTT_SEGS;
+	caps->num_cqe_segs	= HNS_ROCE_V2_MAX_CQE_SEGS;
+	caps->num_pds		= HNS_ROCE_V2_MAX_PD_NUM;
+	caps->max_qp_init_rdma	= HNS_ROCE_V2_MAX_QP_INIT_RDMA;
+	caps->max_qp_dest_rdma	= HNS_ROCE_V2_MAX_QP_DEST_RDMA;
+	caps->max_sq_desc_sz	= HNS_ROCE_V2_MAX_SQ_DESC_SZ;
+	caps->max_rq_desc_sz	= HNS_ROCE_V2_MAX_RQ_DESC_SZ;
+	caps->max_srq_desc_sz	= HNS_ROCE_V2_MAX_SRQ_DESC_SZ;
+	caps->qpc_entry_sz	= HNS_ROCE_V2_QPC_ENTRY_SZ;
+	caps->irrl_entry_sz	= HNS_ROCE_V2_IRRL_ENTRY_SZ;
+	caps->cqc_entry_sz	= HNS_ROCE_V2_CQC_ENTRY_SZ;
+	caps->mtpt_entry_sz	= HNS_ROCE_V2_MTPT_ENTRY_SZ;
+	caps->mtt_entry_sz	= HNS_ROCE_V2_MTT_ENTRY_SZ;
+	caps->cq_entry_sz	= HNS_ROCE_V2_CQE_ENTRY_SIZE;
+	caps->page_size_cap	= HNS_ROCE_V2_PAGE_SIZE_SUPPORTED;
+	caps->reserved_lkey	= 0;
+	caps->reserved_pds	= 0;
+	caps->reserved_mrws	= 1;
+	caps->reserved_uars	= 0;
+	caps->reserved_cqs	= 0;
+
+	caps->pkey_table_len[0] = 1;
+	caps->gid_table_len[0] = 2;
+	caps->local_ca_ack_delay = 0;
+	caps->max_mtu = IB_MTU_4096;
+
+	return 0;
+}
+
 static const struct hns_roce_hw hns_roce_hw_v2 = {
 	.cmq_init = hns_roce_v2_cmq_init,
 	.cmq_exit = hns_roce_v2_cmq_exit,
+	.hw_profile = hns_roce_v2_profile,
 };
 
 static const struct pci_device_id hns_roce_hw_v2_pci_tbl[] = {
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index 781ae74..3dcc800 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -33,6 +33,43 @@
 #ifndef _HNS_ROCE_HW_V2_H
 #define _HNS_ROCE_HW_V2_H
 
+#include <linux/bitops.h>
+
+#define HNS_ROCE_VF_QPC_BT_NUM			256
+#define HNS_ROCE_VF_SRQC_BT_NUM			64
+#define HNS_ROCE_VF_CQC_BT_NUM			64
+#define HNS_ROCE_VF_MPT_BT_NUM			64
+#define HNS_ROCE_VF_EQC_NUM			64
+#define HNS_ROCE_VF_SMAC_NUM			32
+#define HNS_ROCE_VF_SGID_NUM			32
+#define HNS_ROCE_VF_SL_NUM			8
+
+#define HNS_ROCE_V2_MAX_QP_NUM			0x2000
+#define HNS_ROCE_V2_MAX_WQE_NUM			0x8000
+#define HNS_ROCE_V2_MAX_CQ_NUM			0x8000
+#define HNS_ROCE_V2_MAX_CQE_NUM			0x400000
+#define HNS_ROCE_V2_MAX_RQ_SGE_NUM		0x100
+#define HNS_ROCE_V2_MAX_SQ_SGE_NUM		0xff
+#define HNS_ROCE_V2_MAX_SQ_INLINE		0x20
+#define HNS_ROCE_V2_UAR_NUM			256
+#define HNS_ROCE_V2_PHY_UAR_NUM			1
+#define HNS_ROCE_V2_MAX_MTPT_NUM		0x8000
+#define HNS_ROCE_V2_MAX_MTT_SEGS		0x100000
+#define HNS_ROCE_V2_MAX_CQE_SEGS		0x10000
+#define HNS_ROCE_V2_MAX_PD_NUM			0x400000
+#define HNS_ROCE_V2_MAX_QP_INIT_RDMA		128
+#define HNS_ROCE_V2_MAX_QP_DEST_RDMA		128
+#define HNS_ROCE_V2_MAX_SQ_DESC_SZ		64
+#define HNS_ROCE_V2_MAX_RQ_DESC_SZ		16
+#define HNS_ROCE_V2_MAX_SRQ_DESC_SZ		64
+#define HNS_ROCE_V2_QPC_ENTRY_SZ		256
+#define HNS_ROCE_V2_IRRL_ENTRY_SZ		64
+#define HNS_ROCE_V2_CQC_ENTRY_SZ		64
+#define HNS_ROCE_V2_MTPT_ENTRY_SZ		64
+#define HNS_ROCE_V2_MTT_ENTRY_SZ		64
+#define HNS_ROCE_V2_CQE_ENTRY_SIZE		32
+#define HNS_ROCE_V2_PAGE_SIZE_SUPPORTED		0xFFFFF000
+#define HNS_ROCE_V2_MAX_INNER_MTPT_NUM		2
 #define HNS_ROCE_CMQ_TX_TIMEOUT			200
 
 #define HNS_ROCE_CMD_FLAG_IN_VALID_SHIFT	0
@@ -75,6 +112,130 @@ enum hns_roce_cmd_return_status {
 	CMD_QUEUE_FULL		= 3,
 };
 
+struct hns_roce_query_version {
+	__le16 rocee_vendor_id;
+	__le16 rocee_hw_version;
+	__le32 rsv[5];
+};
+
+struct hns_roce_cfg_global_param {
+	__le32 time_cfg_udp_port;
+	__le32 rsv[5];
+};
+
+#define CFG_GLOBAL_PARAM_DATA_0_ROCEE_TIME_1US_CFG_S 0
+#define CFG_GLOBAL_PARAM_DATA_0_ROCEE_TIME_1US_CFG_M GENMASK(9, 0)
+
+#define CFG_GLOBAL_PARAM_DATA_0_ROCEE_UDP_PORT_S 16
+#define CFG_GLOBAL_PARAM_DATA_0_ROCEE_UDP_PORT_M GENMASK(31, 16)
+
+struct hns_roce_pf_res {
+	__le32	rsv;
+	__le32	qpc_bt_idx_num;
+	__le32	srqc_bt_idx_num;
+	__le32	cqc_bt_idx_num;
+	__le32	mpt_bt_idx_num;
+	__le32	eqc_bt_idx_num;
+};
+
+#define PF_RES_DATA_1_PF_QPC_BT_IDX_S 0
+#define PF_RES_DATA_1_PF_QPC_BT_IDX_M GENMASK(10, 0)
+
+#define PF_RES_DATA_1_PF_QPC_BT_NUM_S 16
+#define PF_RES_DATA_1_PF_QPC_BT_NUM_M GENMASK(27, 16)
+
+#define PF_RES_DATA_2_PF_SRQC_BT_IDX_S 0
+#define PF_RES_DATA_2_PF_SRQC_BT_IDX_M GENMASK(8, 0)
+
+#define PF_RES_DATA_2_PF_SRQC_BT_NUM_S 16
+#define PF_RES_DATA_2_PF_SRQC_BT_NUM_M GENMASK(25, 16)
+
+#define PF_RES_DATA_3_PF_CQC_BT_IDX_S 0
+#define PF_RES_DATA_3_PF_CQC_BT_IDX_M GENMASK(8, 0)
+
+#define PF_RES_DATA_3_PF_CQC_BT_NUM_S 16
+#define PF_RES_DATA_3_PF_CQC_BT_NUM_M GENMASK(25, 16)
+
+#define PF_RES_DATA_4_PF_MPT_BT_IDX_S 0
+#define PF_RES_DATA_4_PF_MPT_BT_IDX_M GENMASK(8, 0)
+
+#define PF_RES_DATA_4_PF_MPT_BT_NUM_S 16
+#define PF_RES_DATA_4_PF_MPT_BT_NUM_M GENMASK(25, 16)
+
+#define PF_RES_DATA_5_PF_EQC_BT_IDX_S 0
+#define PF_RES_DATA_5_PF_EQC_BT_IDX_M GENMASK(8, 0)
+
+#define PF_RES_DATA_5_PF_EQC_BT_NUM_S 16
+#define PF_RES_DATA_5_PF_EQC_BT_NUM_M GENMASK(25, 16)
+
+struct hns_roce_vf_res_a {
+	u32 vf_id;
+	u32 vf_qpc_bt_idx_num;
+	u32 vf_srqc_bt_idx_num;
+	u32 vf_cqc_bt_idx_num;
+	u32 vf_mpt_bt_idx_num;
+	u32 vf_eqc_bt_idx_num;
+};
+
+#define VF_RES_A_DATA_1_VF_QPC_BT_IDX_S 0
+#define VF_RES_A_DATA_1_VF_QPC_BT_IDX_M GENMASK(10, 0)
+
+#define VF_RES_A_DATA_1_VF_QPC_BT_NUM_S 16
+#define VF_RES_A_DATA_1_VF_QPC_BT_NUM_M GENMASK(27, 16)
+
+#define VF_RES_A_DATA_2_VF_SRQC_BT_IDX_S 0
+#define VF_RES_A_DATA_2_VF_SRQC_BT_IDX_M GENMASK(8, 0)
+
+#define VF_RES_A_DATA_2_VF_SRQC_BT_NUM_S 16
+#define VF_RES_A_DATA_2_VF_SRQC_BT_NUM_M GENMASK(25, 16)
+
+#define VF_RES_A_DATA_3_VF_CQC_BT_IDX_S 0
+#define VF_RES_A_DATA_3_VF_CQC_BT_IDX_M GENMASK(8, 0)
+
+#define VF_RES_A_DATA_3_VF_CQC_BT_NUM_S 16
+#define VF_RES_A_DATA_3_VF_CQC_BT_NUM_M GENMASK(25, 16)
+
+#define VF_RES_A_DATA_4_VF_MPT_BT_IDX_S 0
+#define VF_RES_A_DATA_4_VF_MPT_BT_IDX_M GENMASK(8, 0)
+
+#define VF_RES_A_DATA_4_VF_MPT_BT_NUM_S 16
+#define VF_RES_A_DATA_4_VF_MPT_BT_NUM_M GENMASK(25, 16)
+
+#define VF_RES_A_DATA_5_VF_EQC_IDX_S 0
+#define VF_RES_A_DATA_5_VF_EQC_IDX_M GENMASK(8, 0)
+
+#define VF_RES_A_DATA_5_VF_EQC_NUM_S 16
+#define VF_RES_A_DATA_5_VF_EQC_NUM_M GENMASK(25, 16)
+
+struct hns_roce_vf_res_b {
+	u32 rsv0;
+	u32 vf_smac_idx_num;
+	u32 vf_sgid_idx_num;
+	u32 vf_qid_idx_sl_num;
+	u32 rsv[2];
+};
+
+#define VF_RES_B_DATA_0_VF_ID_S 0
+#define VF_RES_B_DATA_0_VF_ID_M GENMASK(7, 0)
+
+#define VF_RES_B_DATA_1_VF_SMAC_IDX_S 0
+#define VF_RES_B_DATA_1_VF_SMAC_IDX_M GENMASK(7, 0)
+
+#define VF_RES_B_DATA_1_VF_SMAC_NUM_S 8
+#define VF_RES_B_DATA_1_VF_SMAC_NUM_M GENMASK(16, 8)
+
+#define VF_RES_B_DATA_2_VF_SGID_IDX_S 0
+#define VF_RES_B_DATA_2_VF_SGID_IDX_M GENMASK(7, 0)
+
+#define VF_RES_B_DATA_2_VF_SGID_NUM_S 8
+#define VF_RES_B_DATA_2_VF_SGID_NUM_M GENMASK(16, 8)
+
+#define VF_RES_B_DATA_3_VF_QID_IDX_S 0
+#define VF_RES_B_DATA_3_VF_QID_IDX_M GENMASK(9, 0)
+
+#define VF_RES_B_DATA_3_VF_SL_NUM_S 16
+#define VF_RES_B_DATA_3_VF_SL_NUM_M GENMASK(19, 16)
+
 struct hns_roce_cmq_desc {
 	u16 opcode;
 	u16 flag;
diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c
index 4f5a6fd..db6593e 100644
--- a/drivers/infiniband/hw/hns/hns_roce_main.c
+++ b/drivers/infiniband/hw/hns/hns_roce_main.c
@@ -202,7 +202,7 @@ static int hns_roce_query_device(struct ib_device *ib_dev,
 	props->max_qp_wr = hr_dev->caps.max_wqes;
 	props->device_cap_flags = IB_DEVICE_PORT_ACTIVE_EVENT |
 				  IB_DEVICE_RC_RNR_NAK_GEN;
-	props->max_sge = hr_dev->caps.max_sq_sg;
+	props->max_sge = max(hr_dev->caps.max_sq_sg, hr_dev->caps.max_rq_sg);
 	props->max_sge_rd = 1;
 	props->max_cq = hr_dev->caps.num_cqs;
 	props->max_cqe = hr_dev->caps.max_cqes;
@@ -686,7 +686,11 @@ int hns_roce_init(struct hns_roce_dev *hr_dev)
 		}
 	}
 
-	hr_dev->hw->hw_profile(hr_dev);
+	ret = hr_dev->hw->hw_profile(hr_dev);
+	if (ret) {
+		dev_err(dev, "Get RoCE engine profile failed!\n");
+		goto error_failed_cmd_init;
+	}
 
 	ret = hns_roce_cmd_init(hr_dev);
 	if (ret) {
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 07/20] RDMA/hns: Add mailbox's implementation for hip08 RoCE driver

[Wei Hu (Xavier)]

In hip08 SoC, the hardware implementation of mailbox command
has changed with hip06 SoC. As a result, it adjusts the
architecture of the command code and implements the interfaces
of mailbox for hip08 SoC.

Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_cmd.c    | 98 ++---------------------------
 drivers/infiniband/hw/hns/hns_roce_common.h |  1 +
 drivers/infiniband/hw/hns/hns_roce_device.h |  5 +-
 drivers/infiniband/hw/hns/hns_roce_hw_v1.c  | 75 ++++++++++++++++++++++
 drivers/infiniband/hw/hns/hns_roce_hw_v1.h  |  5 ++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 84 +++++++++++++++++++++++++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  | 20 ++++++
 7 files changed, 195 insertions(+), 93 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_cmd.c b/drivers/infiniband/hw/hns/hns_roce_cmd.c
index 67ad3e9..1085cb2 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cmd.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cmd.c
@@ -38,69 +38,7 @@
 
 #define CMD_POLL_TOKEN		0xffff
 #define CMD_MAX_NUM		32
-#define STATUS_MASK		0xff
 #define CMD_TOKEN_MASK		0x1f
-#define GO_BIT_TIMEOUT_MSECS	10000
-
-enum {
-	HCR_TOKEN_OFFSET	= 0x14,
-	HCR_STATUS_OFFSET	= 0x18,
-	HCR_GO_BIT		= 15,
-};
-
-static int cmd_pending(struct hns_roce_dev *hr_dev)
-{
-	u32 status = readl(hr_dev->cmd.hcr + HCR_TOKEN_OFFSET);
-
-	return (!!(status & (1 << HCR_GO_BIT)));
-}
-
-/* this function should be serialized with "hcr_mutex" */
-static int __hns_roce_cmd_mbox_post_hw(struct hns_roce_dev *hr_dev,
-				       u64 in_param, u64 out_param,
-				       u32 in_modifier, u8 op_modifier, u16 op,
-				       u16 token, int event)
-{
-	struct hns_roce_cmdq *cmd = &hr_dev->cmd;
-	struct device *dev = &hr_dev->pdev->dev;
-	u32 __iomem *hcr = (u32 *)cmd->hcr;
-	int ret = -EAGAIN;
-	unsigned long end;
-	u32 val = 0;
-
-	end = msecs_to_jiffies(GO_BIT_TIMEOUT_MSECS) + jiffies;
-	while (cmd_pending(hr_dev)) {
-		if (time_after(jiffies, end)) {
-			dev_dbg(dev, "jiffies=%d end=%d\n", (int)jiffies,
-				(int)end);
-			goto out;
-		}
-		cond_resched();
-	}
-
-	roce_set_field(val, ROCEE_MB6_ROCEE_MB_CMD_M, ROCEE_MB6_ROCEE_MB_CMD_S,
-		       op);
-	roce_set_field(val, ROCEE_MB6_ROCEE_MB_CMD_MDF_M,
-		       ROCEE_MB6_ROCEE_MB_CMD_MDF_S, op_modifier);
-	roce_set_bit(val, ROCEE_MB6_ROCEE_MB_EVENT_S, event);
-	roce_set_bit(val, ROCEE_MB6_ROCEE_MB_HW_RUN_S, 1);
-	roce_set_field(val, ROCEE_MB6_ROCEE_MB_TOKEN_M,
-		       ROCEE_MB6_ROCEE_MB_TOKEN_S, token);
-
-	__raw_writeq(cpu_to_le64(in_param), hcr + 0);
-	__raw_writeq(cpu_to_le64(out_param), hcr + 2);
-	__raw_writel(cpu_to_le32(in_modifier), hcr + 4);
-	/* Memory barrier */
-	wmb();
-
-	__raw_writel(cpu_to_le32(val), hcr + 5);
-
-	mmiowb();
-	ret = 0;
-
-out:
-	return ret;
-}
 
 static int hns_roce_cmd_mbox_post_hw(struct hns_roce_dev *hr_dev, u64 in_param,
 				     u64 out_param, u32 in_modifier,
@@ -108,12 +46,11 @@ static int hns_roce_cmd_mbox_post_hw(struct hns_roce_dev *hr_dev, u64 in_param,
 				     int event)
 {
 	struct hns_roce_cmdq *cmd = &hr_dev->cmd;
-	int ret = -EAGAIN;
+	int ret;
 
 	mutex_lock(&cmd->hcr_mutex);
-	ret = __hns_roce_cmd_mbox_post_hw(hr_dev, in_param, out_param,
-					  in_modifier, op_modifier, op, token,
-					  event);
+	ret = hr_dev->hw->post_mbox(hr_dev, in_param, out_param, in_modifier,
+				    op_modifier, op, token, event);
 	mutex_unlock(&cmd->hcr_mutex);
 
 	return ret;
@@ -126,9 +63,6 @@ static int __hns_roce_cmd_mbox_poll(struct hns_roce_dev *hr_dev, u64 in_param,
 				    unsigned long timeout)
 {
 	struct device *dev = hr_dev->dev;
-	u8 __iomem *hcr = hr_dev->cmd.hcr;
-	unsigned long end = 0;
-	u32 status = 0;
 	int ret;
 
 	ret = hns_roce_cmd_mbox_post_hw(hr_dev, in_param, out_param,
@@ -136,29 +70,10 @@ static int __hns_roce_cmd_mbox_poll(struct hns_roce_dev *hr_dev, u64 in_param,
 					CMD_POLL_TOKEN, 0);
 	if (ret) {
 		dev_err(dev, "[cmd_poll]hns_roce_cmd_mbox_post_hw failed\n");
-		goto out;
-	}
-
-	end = msecs_to_jiffies(timeout) + jiffies;
-	while (cmd_pending(hr_dev) && time_before(jiffies, end))
-		cond_resched();
-
-	if (cmd_pending(hr_dev)) {
-		dev_err(dev, "[cmd_poll]hw run cmd TIMEDOUT!\n");
-		ret = -ETIMEDOUT;
-		goto out;
+		return ret;
 	}
 
-	status = le32_to_cpu((__force __be32)
-			      __raw_readl(hcr + HCR_STATUS_OFFSET));
-	if ((status & STATUS_MASK) != 0x1) {
-		dev_err(dev, "mailbox status 0x%x!\n", status);
-		ret = -EBUSY;
-		goto out;
-	}
-
-out:
-	return ret;
+	return hr_dev->hw->chk_mbox(hr_dev, timeout);
 }
 
 static int hns_roce_cmd_mbox_poll(struct hns_roce_dev *hr_dev, u64 in_param,
@@ -198,7 +113,7 @@ static int __hns_roce_cmd_mbox_wait(struct hns_roce_dev *hr_dev, u64 in_param,
 	struct hns_roce_cmdq *cmd = &hr_dev->cmd;
 	struct hns_roce_cmd_context *context;
 	struct device *dev = hr_dev->dev;
-	int ret = 0;
+	int ret;
 
 	spin_lock(&cmd->context_lock);
 	WARN_ON(cmd->free_head < 0);
@@ -280,7 +195,6 @@ int hns_roce_cmd_init(struct hns_roce_dev *hr_dev)
 	hr_dev->cmd.use_events = 0;
 	hr_dev->cmd.toggle = 1;
 	hr_dev->cmd.max_cmds = CMD_MAX_NUM;
-	hr_dev->cmd.hcr = hr_dev->reg_base + ROCEE_MB1_REG;
 	hr_dev->cmd.pool = dma_pool_create("hns_roce_cmd", dev,
 					   HNS_ROCE_MAILBOX_SIZE,
 					   HNS_ROCE_MAILBOX_SIZE, 0);
diff --git a/drivers/infiniband/hw/hns/hns_roce_common.h b/drivers/infiniband/hw/hns/hns_roce_common.h
index 94381c2..0c950f8 100644
--- a/drivers/infiniband/hw/hns/hns_roce_common.h
+++ b/drivers/infiniband/hw/hns/hns_roce_common.h
@@ -341,6 +341,7 @@
 #define ROCEE_BT_CMD_L_REG			0x200
 
 #define ROCEE_MB1_REG				0x210
+#define ROCEE_MB6_REG				0x224
 #define ROCEE_DB_SQ_L_0_REG			0x230
 #define ROCEE_DB_OTHERS_L_0_REG			0x238
 #define ROCEE_QP1C_CFG0_0_REG			0x270
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 8ec6d6e..7c58750 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -367,7 +367,6 @@ struct hns_roce_cmd_context {
 
 struct hns_roce_cmdq {
 	struct dma_pool		*pool;
-	u8 __iomem		*hcr;
 	struct mutex		hcr_mutex;
 	struct semaphore	poll_sem;
 	/*
@@ -517,6 +516,10 @@ struct hns_roce_hw {
 	int (*hw_profile)(struct hns_roce_dev *hr_dev);
 	int (*hw_init)(struct hns_roce_dev *hr_dev);
 	void (*hw_exit)(struct hns_roce_dev *hr_dev);
+	int (*post_mbox)(struct hns_roce_dev *hr_dev, u64 in_param,
+			 u64 out_param, u32 in_modifier, u8 op_modifier, u16 op,
+			 u16 token, int event);
+	int (*chk_mbox)(struct hns_roce_dev *hr_dev, unsigned long timeout);
 	void (*set_gid)(struct hns_roce_dev *hr_dev, u8 port, int gid_index,
 			union ib_gid *gid);
 	void (*set_mac)(struct hns_roce_dev *hr_dev, u8 phy_port, u8 *addr);
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 267e400..2bf2816 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -1619,6 +1619,79 @@ void hns_roce_v1_exit(struct hns_roce_dev *hr_dev)
 	hns_roce_db_free(hr_dev);
 }
 
+static int hns_roce_v1_cmd_pending(struct hns_roce_dev *hr_dev)
+{
+	u32 status = readl(hr_dev->reg_base + ROCEE_MB6_REG);
+
+	return (!!(status & (1 << HCR_GO_BIT)));
+}
+
+int hns_roce_v1_post_mbox(struct hns_roce_dev *hr_dev, u64 in_param,
+			  u64 out_param, u32 in_modifier, u8 op_modifier,
+			  u16 op, u16 token, int event)
+{
+	u32 *hcr = (u32 *)(hr_dev->reg_base + ROCEE_MB1_REG);
+	unsigned long end;
+	u32 val = 0;
+
+	end = msecs_to_jiffies(GO_BIT_TIMEOUT_MSECS) + jiffies;
+	while (hns_roce_v1_cmd_pending(hr_dev)) {
+		if (time_after(jiffies, end)) {
+			dev_err(hr_dev->dev, "jiffies=%d end=%d\n",
+				(int)jiffies, (int)end);
+			return -EAGAIN;
+		}
+		cond_resched();
+	}
+
+	roce_set_field(val, ROCEE_MB6_ROCEE_MB_CMD_M, ROCEE_MB6_ROCEE_MB_CMD_S,
+		       op);
+	roce_set_field(val, ROCEE_MB6_ROCEE_MB_CMD_MDF_M,
+		       ROCEE_MB6_ROCEE_MB_CMD_MDF_S, op_modifier);
+	roce_set_bit(val, ROCEE_MB6_ROCEE_MB_EVENT_S, event);
+	roce_set_bit(val, ROCEE_MB6_ROCEE_MB_HW_RUN_S, 1);
+	roce_set_field(val, ROCEE_MB6_ROCEE_MB_TOKEN_M,
+		       ROCEE_MB6_ROCEE_MB_TOKEN_S, token);
+
+	__raw_writeq(cpu_to_le64(in_param), hcr + 0);
+	__raw_writeq(cpu_to_le64(out_param), hcr + 2);
+	__raw_writel(cpu_to_le32(in_modifier), hcr + 4);
+	/* Memory barrier */
+	wmb();
+
+	__raw_writel(cpu_to_le32(val), hcr + 5);
+
+	mmiowb();
+
+	return 0;
+}
+
+static int hns_roce_v1_chk_mbox(struct hns_roce_dev *hr_dev,
+				unsigned long timeout)
+{
+	u8 __iomem *hcr = hr_dev->reg_base + ROCEE_MB1_REG;
+	unsigned long end = 0;
+	u32 status = 0;
+
+	end = msecs_to_jiffies(timeout) + jiffies;
+	while (hns_roce_v1_cmd_pending(hr_dev) && time_before(jiffies, end))
+		cond_resched();
+
+	if (hns_roce_v1_cmd_pending(hr_dev)) {
+		dev_err(hr_dev->dev, "[cmd_poll]hw run cmd TIMEDOUT!\n");
+		return -ETIMEDOUT;
+	}
+
+	status = le32_to_cpu((__force __be32)
+			      __raw_readl(hcr + HCR_STATUS_OFFSET));
+	if ((status & STATUS_MASK) != 0x1) {
+		dev_err(hr_dev->dev, "mailbox status 0x%x!\n", status);
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
 void hns_roce_v1_set_gid(struct hns_roce_dev *hr_dev, u8 port, int gid_index,
 			 union ib_gid *gid)
 {
@@ -3849,6 +3922,8 @@ int hns_roce_v1_destroy_cq(struct ib_cq *ibcq)
 	.hw_profile = hns_roce_v1_profile,
 	.hw_init = hns_roce_v1_init,
 	.hw_exit = hns_roce_v1_exit,
+	.post_mbox = hns_roce_v1_post_mbox,
+	.chk_mbox = hns_roce_v1_chk_mbox,
 	.set_gid = hns_roce_v1_set_gid,
 	.set_mac = hns_roce_v1_set_mac,
 	.set_mtu = hns_roce_v1_set_mtu,
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.h b/drivers/infiniband/hw/hns/hns_roce_hw_v1.h
index b213b5e..eb83ff3 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.h
@@ -948,6 +948,11 @@ struct hns_roce_qp_context {
 #define QP_CONTEXT_QPC_BYTES_188_TX_RETRY_CUR_INDEX_M   \
 	(((1UL << 15) - 1) << QP_CONTEXT_QPC_BYTES_188_TX_RETRY_CUR_INDEX_S)
 
+#define STATUS_MASK		0xff
+#define GO_BIT_TIMEOUT_MSECS	10000
+#define HCR_STATUS_OFFSET	0x18
+#define HCR_GO_BIT		15
+
 struct hns_roce_rq_db {
 	u32    u32_4;
 	u32    u32_8;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index af3cb30..259af2e 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -553,10 +553,94 @@ static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev)
 	return 0;
 }
 
+static int hns_roce_v2_cmd_pending(struct hns_roce_dev *hr_dev)
+{
+	u32 status = readl(hr_dev->reg_base + ROCEE_VF_MB_STATUS_REG);
+
+	return status >> HNS_ROCE_HW_RUN_BIT_SHIFT;
+}
+
+static int hns_roce_v2_cmd_complete(struct hns_roce_dev *hr_dev)
+{
+	u32 status = readl(hr_dev->reg_base + ROCEE_VF_MB_STATUS_REG);
+
+	return status & HNS_ROCE_HW_MB_STATUS_MASK;
+}
+
+static int hns_roce_v2_post_mbox(struct hns_roce_dev *hr_dev, u64 in_param,
+				 u64 out_param, u32 in_modifier, u8 op_modifier,
+				 u16 op, u16 token, int event)
+{
+	struct device *dev = hr_dev->dev;
+	u32 *hcr = (u32 *)(hr_dev->reg_base + ROCEE_VF_MB_CFG0_REG);
+	unsigned long end;
+	u32 val0 = 0;
+	u32 val1 = 0;
+
+	end = msecs_to_jiffies(HNS_ROCE_V2_GO_BIT_TIMEOUT_MSECS) + jiffies;
+	while (hns_roce_v2_cmd_pending(hr_dev)) {
+		if (time_after(jiffies, end)) {
+			dev_dbg(dev, "jiffies=%d end=%d\n", (int)jiffies,
+				(int)end);
+			return -EAGAIN;
+		}
+		cond_resched();
+	}
+
+	roce_set_field(val0, HNS_ROCE_VF_MB4_TAG_MASK,
+		       HNS_ROCE_VF_MB4_TAG_SHIFT, in_modifier);
+	roce_set_field(val0, HNS_ROCE_VF_MB4_CMD_MASK,
+		       HNS_ROCE_VF_MB4_CMD_SHIFT, op);
+	roce_set_field(val1, HNS_ROCE_VF_MB5_EVENT_MASK,
+		       HNS_ROCE_VF_MB5_EVENT_SHIFT, event);
+	roce_set_field(val1, HNS_ROCE_VF_MB5_TOKEN_MASK,
+		       HNS_ROCE_VF_MB5_TOKEN_SHIFT, token);
+
+	__raw_writeq(cpu_to_le64(in_param), hcr + 0);
+	__raw_writeq(cpu_to_le64(out_param), hcr + 2);
+
+	/* Memory barrier */
+	wmb();
+
+	__raw_writel(cpu_to_le32(val0), hcr + 4);
+	__raw_writel(cpu_to_le32(val1), hcr + 5);
+
+	mmiowb();
+
+	return 0;
+}
+
+static int hns_roce_v2_chk_mbox(struct hns_roce_dev *hr_dev,
+				unsigned long timeout)
+{
+	struct device *dev = hr_dev->dev;
+	unsigned long end = 0;
+	u32 status;
+
+	end = msecs_to_jiffies(timeout) + jiffies;
+	while (hns_roce_v2_cmd_pending(hr_dev) && time_before(jiffies, end))
+		cond_resched();
+
+	if (hns_roce_v2_cmd_pending(hr_dev)) {
+		dev_err(dev, "[cmd_poll]hw run cmd TIMEDOUT!\n");
+		return -ETIMEDOUT;
+	}
+
+	status = hns_roce_v2_cmd_complete(hr_dev);
+	if (status != 0x1) {
+		dev_err(dev, "mailbox status 0x%x!\n", status);
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
 static const struct hns_roce_hw hns_roce_hw_v2 = {
 	.cmq_init = hns_roce_v2_cmq_init,
 	.cmq_exit = hns_roce_v2_cmq_exit,
 	.hw_profile = hns_roce_v2_profile,
+	.post_mbox = hns_roce_v2_post_mbox,
+	.chk_mbox = hns_roce_v2_chk_mbox,
 };
 
 static const struct pci_device_id hns_roce_hw_v2_pci_tbl[] = {
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index 3dcc800..b420ea9 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -244,6 +244,26 @@ struct hns_roce_cmq_desc {
 	u32 data[6];
 };
 
+#define ROCEE_VF_MB_CFG0_REG		0x40
+#define ROCEE_VF_MB_STATUS_REG		0x58
+
+#define HNS_ROCE_V2_GO_BIT_TIMEOUT_MSECS	10000
+
+#define HNS_ROCE_HW_RUN_BIT_SHIFT	31
+#define HNS_ROCE_HW_MB_STATUS_MASK	0xFF
+
+#define HNS_ROCE_VF_MB4_TAG_MASK	0xFFFFFF00
+#define HNS_ROCE_VF_MB4_TAG_SHIFT	8
+
+#define HNS_ROCE_VF_MB4_CMD_MASK	0xFF
+#define HNS_ROCE_VF_MB4_CMD_SHIFT	0
+
+#define HNS_ROCE_VF_MB5_EVENT_MASK	0x10000
+#define HNS_ROCE_VF_MB5_EVENT_SHIFT	16
+
+#define HNS_ROCE_VF_MB5_TOKEN_MASK	0xFFFF
+#define HNS_ROCE_VF_MB5_TOKEN_SHIFT	0
+
 struct hns_roce_v2_cmq_ring {
 	dma_addr_t desc_dma_addr;
 	struct hns_roce_cmq_desc *desc;
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 08/20] RDMA/hns: Add the interfaces to support multi hop addressing for the contexts in hip08

[Wei Hu (Xavier)]

From: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>

The contexts (QPC/MTPT/CQC/SRQC) in hip08 can support multi hop
addressing. The address of context can be retrieved by the
BT (Base Address Table) with multi hop addressing. The first hop
BT BA can be retrieved from the RAM in the chip by the bt_idx and
bt_num.

This patch is to add the interfaces in HEM to support multi hop
addressing for the contexts.

Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_device.h |  23 +-
 drivers/infiniband/hw/hns/hns_roce_hem.c    | 578 +++++++++++++++++++++++++++-
 drivers/infiniband/hw/hns/hns_roce_hem.h    |  23 ++
 drivers/infiniband/hw/hns/hns_roce_hw_v1.c  |   2 +-
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  |  13 +
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  |   3 +
 6 files changed, 630 insertions(+), 12 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 7c58750..4574a7b 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -78,6 +78,8 @@
 #define HNS_ROCE_MAX_GID_NUM			16
 #define HNS_ROCE_GID_SIZE			16
 
+#define HNS_ROCE_HOP_NUM_0			0xff
+
 #define BITMAP_NO_RR				0
 #define BITMAP_RR				1
 
@@ -232,6 +234,10 @@ struct hns_roce_hem_table {
 	int		lowmem;
 	struct mutex	mutex;
 	struct hns_roce_hem **hem;
+	u64		**bt_l1;
+	dma_addr_t	*bt_l1_dma_addr;
+	u64		**bt_l0;
+	dma_addr_t	*bt_l0_dma_addr;
 };
 
 struct hns_roce_mtt {
@@ -507,6 +513,18 @@ struct hns_roce_caps {
 	u32		srqc_bt_num;
 	u32		cqc_bt_num;
 	u32		mpt_bt_num;
+	u32		qpc_ba_pg_sz;
+	u32		qpc_buf_pg_sz;
+	u32		qpc_hop_num;
+	u32		srqc_ba_pg_sz;
+	u32		srqc_buf_pg_sz;
+	u32		srqc_hop_num;
+	u32		cqc_ba_pg_sz;
+	u32		cqc_buf_pg_sz;
+	u32		cqc_hop_num;
+	u32		mpt_ba_pg_sz;
+	u32		mpt_buf_pg_sz;
+	u32		mpt_hop_num;
 };
 
 struct hns_roce_hw {
@@ -530,8 +548,11 @@ struct hns_roce_hw {
 	void (*write_cqc)(struct hns_roce_dev *hr_dev,
 			  struct hns_roce_cq *hr_cq, void *mb_buf, u64 *mtts,
 			  dma_addr_t dma_handle, int nent, u32 vector);
+	int (*set_hem)(struct hns_roce_dev *hr_dev,
+		       struct hns_roce_hem_table *table, int obj, int step_idx);
 	int (*clear_hem)(struct hns_roce_dev *hr_dev,
-			 struct hns_roce_hem_table *table, int obj);
+			 struct hns_roce_hem_table *table, int obj,
+			 int step_idx);
 	int (*query_qp)(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
 	int (*modify_qp)(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c
index 0ab4912..9bc8c6b 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hem.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hem.c
@@ -42,8 +42,140 @@
 #define DMA_ADDR_T_SHIFT		12
 #define BT_BA_SHIFT			32
 
-struct hns_roce_hem *hns_roce_alloc_hem(struct hns_roce_dev *hr_dev, int npages,
-					gfp_t gfp_mask)
+bool hns_roce_check_whether_mhop(struct hns_roce_dev *hr_dev, u32 type)
+{
+	if ((hr_dev->caps.qpc_hop_num && type == HEM_TYPE_QPC) ||
+	    (hr_dev->caps.mpt_hop_num && type == HEM_TYPE_MTPT) ||
+	    (hr_dev->caps.cqc_hop_num && type == HEM_TYPE_CQC) ||
+	    (hr_dev->caps.srqc_hop_num && type == HEM_TYPE_SRQC))
+		return true;
+
+	return false;
+}
+EXPORT_SYMBOL_GPL(hns_roce_check_whether_mhop);
+
+static bool hns_roce_check_hem_null(struct hns_roce_hem **hem, u64 start_idx,
+			    u32 bt_chunk_num)
+{
+	int i;
+
+	for (i = 0; i < bt_chunk_num; i++)
+		if (hem[start_idx + i])
+			return false;
+
+	return true;
+}
+
+static bool hns_roce_check_bt_null(u64 **bt, u64 start_idx, u32 bt_chunk_num)
+{
+	int i;
+
+	for (i = 0; i < bt_chunk_num; i++)
+		if (bt[start_idx + i])
+			return false;
+
+	return true;
+}
+
+static int hns_roce_get_bt_num(u32 table_type, u32 hop_num)
+{
+	if (check_whether_bt_num_3(table_type, hop_num))
+		return 3;
+	else if (check_whether_bt_num_2(table_type, hop_num))
+		return 2;
+	else if (check_whether_bt_num_1(table_type, hop_num))
+		return 1;
+	else
+		return 0;
+}
+
+int hns_roce_calc_hem_mhop(struct hns_roce_dev *hr_dev,
+			   struct hns_roce_hem_table *table, unsigned long *obj,
+			   struct hns_roce_hem_mhop *mhop)
+{
+	struct device *dev = hr_dev->dev;
+	u32 chunk_ba_num;
+	u32 table_idx;
+	u32 bt_num;
+	u32 chunk_size;
+
+	switch (table->type) {
+	case HEM_TYPE_QPC:
+		mhop->buf_chunk_size = 1 << (hr_dev->caps.qpc_buf_pg_sz
+					     + PAGE_SHIFT);
+		mhop->bt_chunk_size = 1 << (hr_dev->caps.qpc_ba_pg_sz
+					     + PAGE_SHIFT);
+		mhop->ba_l0_num = hr_dev->caps.qpc_bt_num;
+		mhop->hop_num = hr_dev->caps.qpc_hop_num;
+		break;
+	case HEM_TYPE_MTPT:
+		mhop->buf_chunk_size = 1 << (hr_dev->caps.mpt_buf_pg_sz
+					     + PAGE_SHIFT);
+		mhop->bt_chunk_size = 1 << (hr_dev->caps.mpt_ba_pg_sz
+					     + PAGE_SHIFT);
+		mhop->ba_l0_num = hr_dev->caps.mpt_bt_num;
+		mhop->hop_num = hr_dev->caps.mpt_hop_num;
+		break;
+	case HEM_TYPE_CQC:
+		mhop->buf_chunk_size = 1 << (hr_dev->caps.cqc_buf_pg_sz
+					     + PAGE_SHIFT);
+		mhop->bt_chunk_size = 1 << (hr_dev->caps.cqc_ba_pg_sz
+					    + PAGE_SHIFT);
+		mhop->ba_l0_num = hr_dev->caps.cqc_bt_num;
+		mhop->hop_num = hr_dev->caps.cqc_hop_num;
+		break;
+	case HEM_TYPE_SRQC:
+		mhop->buf_chunk_size = 1 << (hr_dev->caps.srqc_buf_pg_sz
+					     + PAGE_SHIFT);
+		mhop->bt_chunk_size = 1 << (hr_dev->caps.srqc_ba_pg_sz
+					     + PAGE_SHIFT);
+		mhop->ba_l0_num = hr_dev->caps.srqc_bt_num;
+		mhop->hop_num = hr_dev->caps.srqc_hop_num;
+		break;
+	default:
+		dev_err(dev, "Table %d not support multi-hop addressing!\n",
+			 table->type);
+		return -EINVAL;
+	}
+
+	if (!obj)
+		return 0;
+
+	/* QPC/MTPT/CQC/SRQC alloc hem for buffer pages. */
+	bt_num = hns_roce_get_bt_num(table->type, mhop->hop_num);
+	chunk_ba_num = mhop->bt_chunk_size / 8;
+	chunk_size = mhop->buf_chunk_size;
+	table_idx = (*obj & (table->num_obj - 1)) /
+		     (chunk_size / table->obj_size);
+	switch (bt_num) {
+	case 3:
+		mhop->l2_idx = table_idx & (chunk_ba_num - 1);
+		mhop->l1_idx = table_idx / chunk_ba_num & (chunk_ba_num - 1);
+		mhop->l0_idx = table_idx / chunk_ba_num / chunk_ba_num;
+		break;
+	case 2:
+		mhop->l1_idx = table_idx & (chunk_ba_num - 1);
+		mhop->l0_idx = table_idx / chunk_ba_num;
+		break;
+	case 1:
+		mhop->l0_idx = table_idx;
+		break;
+	default:
+		dev_err(dev, "Table %d not support hop_num = %d!\n",
+			     table->type, mhop->hop_num);
+		return -EINVAL;
+	}
+	if (mhop->l0_idx >= mhop->ba_l0_num)
+		mhop->l0_idx %= mhop->ba_l0_num;
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(hns_roce_calc_hem_mhop);
+
+static struct hns_roce_hem *hns_roce_alloc_hem(struct hns_roce_dev *hr_dev,
+					       int npages,
+					       unsigned long hem_alloc_size,
+					       gfp_t gfp_mask)
 {
 	struct hns_roce_hem_chunk *chunk = NULL;
 	struct hns_roce_hem *hem;
@@ -61,7 +193,7 @@ struct hns_roce_hem *hns_roce_alloc_hem(struct hns_roce_dev *hr_dev, int npages,
 	hem->refcount = 0;
 	INIT_LIST_HEAD(&hem->chunk_list);
 
-	order = get_order(HNS_ROCE_HEM_ALLOC_SIZE);
+	order = get_order(hem_alloc_size);
 
 	while (npages > 0) {
 		if (!chunk) {
@@ -209,6 +341,169 @@ static int hns_roce_set_hem(struct hns_roce_dev *hr_dev,
 	return ret;
 }
 
+int hns_roce_table_mhop_get(struct hns_roce_dev *hr_dev,
+			    struct hns_roce_hem_table *table,
+			    unsigned long obj)
+{
+	struct device *dev = hr_dev->dev;
+	struct hns_roce_hem_mhop mhop;
+	struct hns_roce_hem_iter iter;
+	u32 buf_chunk_size;
+	u32 bt_chunk_size;
+	u32 chunk_ba_num;
+	u32 hop_num;
+	u32 size;
+	u32 bt_num;
+	u64 hem_idx;
+	u64 bt_l1_idx = 0;
+	u64 bt_l0_idx = 0;
+	u64 bt_ba;
+	unsigned long mhop_obj = obj;
+	int bt_l1_allocated = 0;
+	int bt_l0_allocated = 0;
+	int step_idx;
+	int ret;
+
+	ret = hns_roce_calc_hem_mhop(hr_dev, table, &mhop_obj, &mhop);
+	if (ret)
+		return ret;
+
+	buf_chunk_size = mhop.buf_chunk_size;
+	bt_chunk_size = mhop.bt_chunk_size;
+	hop_num = mhop.hop_num;
+	chunk_ba_num = bt_chunk_size / 8;
+
+	bt_num = hns_roce_get_bt_num(table->type, hop_num);
+	switch (bt_num) {
+	case 3:
+		hem_idx = mhop.l0_idx * chunk_ba_num * chunk_ba_num +
+			  mhop.l1_idx * chunk_ba_num + mhop.l2_idx;
+		bt_l1_idx = mhop.l0_idx * chunk_ba_num + mhop.l1_idx;
+		bt_l0_idx = mhop.l0_idx;
+		break;
+	case 2:
+		hem_idx = mhop.l0_idx * chunk_ba_num + mhop.l1_idx;
+		bt_l0_idx = mhop.l0_idx;
+		break;
+	case 1:
+		hem_idx = mhop.l0_idx;
+		break;
+	default:
+		dev_err(dev, "Table %d not support hop_num = %d!\n",
+			     table->type, hop_num);
+		return -EINVAL;
+	}
+
+	mutex_lock(&table->mutex);
+
+	if (table->hem[hem_idx]) {
+		++table->hem[hem_idx]->refcount;
+		goto out;
+	}
+
+	/* alloc L1 BA's chunk */
+	if ((check_whether_bt_num_3(table->type, hop_num) ||
+		check_whether_bt_num_2(table->type, hop_num)) &&
+		!table->bt_l0[bt_l0_idx]) {
+		table->bt_l0[bt_l0_idx] = dma_alloc_coherent(dev, bt_chunk_size,
+					    &(table->bt_l0_dma_addr[bt_l0_idx]),
+					    GFP_KERNEL);
+		if (!table->bt_l0[bt_l0_idx]) {
+			ret = -ENOMEM;
+			goto out;
+		}
+		bt_l0_allocated = 1;
+
+		/* set base address to hardware */
+		if (table->type < HEM_TYPE_MTT) {
+			step_idx = 0;
+			if (hr_dev->hw->set_hem(hr_dev, table, obj, step_idx)) {
+				ret = -ENODEV;
+				dev_err(dev, "set HEM base address to HW failed!\n");
+				goto err_dma_alloc_l1;
+			}
+		}
+	}
+
+	/* alloc L2 BA's chunk */
+	if (check_whether_bt_num_3(table->type, hop_num) &&
+	    !table->bt_l1[bt_l1_idx])  {
+		table->bt_l1[bt_l1_idx] = dma_alloc_coherent(dev, bt_chunk_size,
+					    &(table->bt_l1_dma_addr[bt_l1_idx]),
+					    GFP_KERNEL);
+		if (!table->bt_l1[bt_l1_idx]) {
+			ret = -ENOMEM;
+			goto err_dma_alloc_l1;
+		}
+		bt_l1_allocated = 1;
+		*(table->bt_l0[bt_l0_idx] + mhop.l1_idx) =
+					       table->bt_l1_dma_addr[bt_l1_idx];
+
+		/* set base address to hardware */
+		step_idx = 1;
+		if (hr_dev->hw->set_hem(hr_dev, table, obj, step_idx)) {
+			ret = -ENODEV;
+			dev_err(dev, "set HEM base address to HW failed!\n");
+			goto err_alloc_hem_buf;
+		}
+	}
+
+	/* alloc buffer space chunk for QPC/MTPT/CQC/SRQC. */
+	size = buf_chunk_size;
+	table->hem[hem_idx] = hns_roce_alloc_hem(hr_dev,
+						size >> PAGE_SHIFT,
+						size,
+						(table->lowmem ? GFP_KERNEL :
+						GFP_HIGHUSER) | __GFP_NOWARN);
+	if (!table->hem[hem_idx]) {
+		ret = -ENOMEM;
+		goto err_alloc_hem_buf;
+	}
+
+	hns_roce_hem_first(table->hem[hem_idx], &iter);
+	bt_ba = hns_roce_hem_addr(&iter);
+
+	if (table->type < HEM_TYPE_MTT) {
+		if (hop_num == 2) {
+			*(table->bt_l1[bt_l1_idx] + mhop.l2_idx) = bt_ba;
+			step_idx = 2;
+		} else if (hop_num == 1) {
+			*(table->bt_l0[bt_l0_idx] + mhop.l1_idx) = bt_ba;
+			step_idx = 1;
+		} else if (hop_num == HNS_ROCE_HOP_NUM_0) {
+			step_idx = 0;
+		}
+
+		/* set HEM base address to hardware */
+		if (hr_dev->hw->set_hem(hr_dev, table, obj, step_idx)) {
+			ret = -ENODEV;
+			dev_err(dev, "set HEM base address to HW failed!\n");
+			goto err_alloc_hem_buf;
+		}
+	}
+
+	++table->hem[hem_idx]->refcount;
+	goto out;
+
+err_alloc_hem_buf:
+	if (bt_l1_allocated) {
+		dma_free_coherent(dev, bt_chunk_size, table->bt_l1[bt_l1_idx],
+				  table->bt_l1_dma_addr[bt_l1_idx]);
+		table->bt_l1[bt_l1_idx] = NULL;
+	}
+
+err_dma_alloc_l1:
+	if (bt_l0_allocated) {
+		dma_free_coherent(dev, bt_chunk_size, table->bt_l0[bt_l0_idx],
+				  table->bt_l0_dma_addr[bt_l0_idx]);
+		table->bt_l0[bt_l0_idx] = NULL;
+	}
+
+out:
+	mutex_unlock(&table->mutex);
+	return ret;
+}
+
 int hns_roce_table_get(struct hns_roce_dev *hr_dev,
 		       struct hns_roce_hem_table *table, unsigned long obj)
 {
@@ -216,6 +511,9 @@ int hns_roce_table_get(struct hns_roce_dev *hr_dev,
 	int ret = 0;
 	unsigned long i;
 
+	if (hns_roce_check_whether_mhop(hr_dev, table->type))
+		return hns_roce_table_mhop_get(hr_dev, table, obj);
+
 	i = (obj & (table->num_obj - 1)) / (HNS_ROCE_TABLE_CHUNK_SIZE /
 	     table->obj_size);
 
@@ -228,6 +526,7 @@ int hns_roce_table_get(struct hns_roce_dev *hr_dev,
 
 	table->hem[i] = hns_roce_alloc_hem(hr_dev,
 				       HNS_ROCE_TABLE_CHUNK_SIZE >> PAGE_SHIFT,
+				       HNS_ROCE_HEM_ALLOC_SIZE,
 				       (table->lowmem ? GFP_KERNEL :
 					GFP_HIGHUSER) | __GFP_NOWARN);
 	if (!table->hem[i]) {
@@ -248,12 +547,128 @@ int hns_roce_table_get(struct hns_roce_dev *hr_dev,
 	return ret;
 }
 
+void hns_roce_table_mhop_put(struct hns_roce_dev *hr_dev,
+			     struct hns_roce_hem_table *table,
+			     unsigned long obj,
+			     int check_refcount)
+{
+	struct device *dev = hr_dev->dev;
+	struct hns_roce_hem_mhop mhop;
+	unsigned long mhop_obj = obj;
+	u32 bt_chunk_size;
+	u32 chunk_ba_num;
+	u32 hop_num;
+	u32 start_idx;
+	u32 bt_num;
+	u64 hem_idx;
+	u64 bt_l1_idx = 0;
+	int ret;
+
+	ret = hns_roce_calc_hem_mhop(hr_dev, table, &mhop_obj, &mhop);
+	if (ret)
+		return;
+
+	bt_chunk_size = mhop.bt_chunk_size;
+	hop_num = mhop.hop_num;
+	chunk_ba_num = bt_chunk_size / 8;
+
+	bt_num = hns_roce_get_bt_num(table->type, hop_num);
+	switch (bt_num) {
+	case 3:
+		hem_idx = mhop.l0_idx * chunk_ba_num * chunk_ba_num +
+			  mhop.l1_idx * chunk_ba_num + mhop.l2_idx;
+		bt_l1_idx = mhop.l0_idx * chunk_ba_num + mhop.l1_idx;
+		break;
+	case 2:
+		hem_idx = mhop.l0_idx * chunk_ba_num + mhop.l1_idx;
+		break;
+	case 1:
+		hem_idx = mhop.l0_idx;
+		break;
+	default:
+		dev_err(dev, "Table %d not support hop_num = %d!\n",
+			     table->type, hop_num);
+		return;
+	}
+
+	mutex_lock(&table->mutex);
+
+	if (check_refcount && (--table->hem[hem_idx]->refcount > 0)) {
+		mutex_unlock(&table->mutex);
+		return;
+	}
+
+	if (table->type < HEM_TYPE_MTT && hop_num == 1) {
+		if (hr_dev->hw->clear_hem(hr_dev, table, obj, 1))
+			dev_warn(dev, "Clear HEM base address failed.\n");
+	} else if (table->type < HEM_TYPE_MTT && hop_num == 2) {
+		if (hr_dev->hw->clear_hem(hr_dev, table, obj, 2))
+			dev_warn(dev, "Clear HEM base address failed.\n");
+	} else if (table->type < HEM_TYPE_MTT &&
+		   hop_num == HNS_ROCE_HOP_NUM_0) {
+		if (hr_dev->hw->clear_hem(hr_dev, table, obj, 0))
+			dev_warn(dev, "Clear HEM base address failed.\n");
+	}
+
+	/* free buffer space chunk for QPC/MTPT/CQC/SRQC. */
+	hns_roce_free_hem(hr_dev, table->hem[hem_idx]);
+	table->hem[hem_idx] = NULL;
+
+	if (check_whether_bt_num_2(table->type, hop_num)) {
+		start_idx = mhop.l0_idx * chunk_ba_num;
+		if (hns_roce_check_hem_null(table->hem, start_idx,
+					    chunk_ba_num)) {
+			if (table->type < HEM_TYPE_MTT &&
+			    hr_dev->hw->clear_hem(hr_dev, table, obj, 0))
+				dev_warn(dev, "Clear HEM base address failed.\n");
+
+			dma_free_coherent(dev, bt_chunk_size,
+					  table->bt_l0[mhop.l0_idx],
+					  table->bt_l0_dma_addr[mhop.l0_idx]);
+			table->bt_l0[mhop.l0_idx] = NULL;
+		}
+	} else if (check_whether_bt_num_3(table->type, hop_num)) {
+		start_idx = mhop.l0_idx * chunk_ba_num * chunk_ba_num +
+			    mhop.l1_idx * chunk_ba_num;
+		if (hns_roce_check_hem_null(table->hem, start_idx,
+					    chunk_ba_num)) {
+			if (hr_dev->hw->clear_hem(hr_dev, table, obj, 1))
+				dev_warn(dev, "Clear HEM base address failed.\n");
+
+			dma_free_coherent(dev, bt_chunk_size,
+					  table->bt_l1[bt_l1_idx],
+					  table->bt_l1_dma_addr[bt_l1_idx]);
+			table->bt_l1[bt_l1_idx] = NULL;
+
+			start_idx = mhop.l0_idx * chunk_ba_num;
+			if (hns_roce_check_bt_null(table->bt_l1, start_idx,
+						   chunk_ba_num)) {
+				if (hr_dev->hw->clear_hem(hr_dev, table, obj,
+							  0))
+					dev_warn(dev, "Clear HEM base address failed.\n");
+
+				dma_free_coherent(dev, bt_chunk_size,
+					    table->bt_l0[mhop.l0_idx],
+					    table->bt_l0_dma_addr[mhop.l0_idx]);
+				table->bt_l0[mhop.l0_idx] = NULL;
+			}
+		}
+	}
+
+	mutex_unlock(&table->mutex);
+}
+
 void hns_roce_table_put(struct hns_roce_dev *hr_dev,
 			struct hns_roce_hem_table *table, unsigned long obj)
 {
 	struct device *dev = hr_dev->dev;
 	unsigned long i;
 
+	if (hns_roce_check_whether_mhop(hr_dev, table->type)) {
+		hns_roce_table_mhop_put(hr_dev, table, obj, 1);
+		return;
+	}
+
 	i = (obj & (table->num_obj - 1)) /
 	    (HNS_ROCE_TABLE_CHUNK_SIZE / table->obj_size);
 
@@ -261,7 +676,7 @@ void hns_roce_table_put(struct hns_roce_dev *hr_dev,
 
 	if (--table->hem[i]->refcount == 0) {
 		/* Clear HEM base address */
-		if (hr_dev->hw->clear_hem(hr_dev, table, obj))
+		if (hr_dev->hw->clear_hem(hr_dev, table, obj, 0))
 			dev_warn(dev, "Clear HEM base address failed.\n");
 
 		hns_roce_free_hem(hr_dev, table->hem[i]);
@@ -357,15 +772,105 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev,
 			    unsigned long obj_size, unsigned long nobj,
 			    int use_lowmem)
 {
+	struct device *dev = hr_dev->dev;
 	unsigned long obj_per_chunk;
 	unsigned long num_hem;
 
-	obj_per_chunk = HNS_ROCE_TABLE_CHUNK_SIZE / obj_size;
-	num_hem = (nobj + obj_per_chunk - 1) / obj_per_chunk;
+	if (!hns_roce_check_whether_mhop(hr_dev, type)) {
+		obj_per_chunk = HNS_ROCE_TABLE_CHUNK_SIZE / obj_size;
+		num_hem = (nobj + obj_per_chunk - 1) / obj_per_chunk;
+
+		table->hem = kcalloc(num_hem, sizeof(*table->hem), GFP_KERNEL);
+		if (!table->hem)
+			return -ENOMEM;
+	} else {
+		unsigned long buf_chunk_size;
+		unsigned long bt_chunk_size;
+		unsigned long bt_chunk_num;
+		unsigned long num_bt_l0;
+		u32 hop_num;
+
+		switch (type) {
+		case HEM_TYPE_QPC:
+			buf_chunk_size = 1 << (hr_dev->caps.qpc_buf_pg_sz
+					+ PAGE_SHIFT);
+			bt_chunk_size = 1 << (hr_dev->caps.qpc_ba_pg_sz
+					+ PAGE_SHIFT);
+			num_bt_l0 = hr_dev->caps.qpc_bt_num;
+			hop_num = hr_dev->caps.qpc_hop_num;
+			break;
+		case HEM_TYPE_MTPT:
+			buf_chunk_size = 1 << (hr_dev->caps.mpt_buf_pg_sz
+					+ PAGE_SHIFT);
+			bt_chunk_size = 1 << (hr_dev->caps.mpt_ba_pg_sz
+					+ PAGE_SHIFT);
+			num_bt_l0 = hr_dev->caps.mpt_bt_num;
+			hop_num = hr_dev->caps.mpt_hop_num;
+			break;
+		case HEM_TYPE_CQC:
+			buf_chunk_size = 1 << (hr_dev->caps.cqc_buf_pg_sz
+					+ PAGE_SHIFT);
+			bt_chunk_size = 1 << (hr_dev->caps.cqc_ba_pg_sz
+					+ PAGE_SHIFT);
+			num_bt_l0 = hr_dev->caps.cqc_bt_num;
+			hop_num = hr_dev->caps.cqc_hop_num;
+			break;
+		case HEM_TYPE_SRQC:
+			buf_chunk_size = 1 << (hr_dev->caps.srqc_buf_pg_sz
+					+ PAGE_SHIFT);
+			bt_chunk_size = 1 << (hr_dev->caps.srqc_ba_pg_sz
+					+ PAGE_SHIFT);
+			num_bt_l0 = hr_dev->caps.srqc_bt_num;
+			hop_num = hr_dev->caps.srqc_hop_num;
+			break;
+		default:
+			dev_err(dev,
+			  "Table %d not support to init hem table here!\n",
+			  type);
+			return -EINVAL;
+		}
+		obj_per_chunk = buf_chunk_size / obj_size;
+		num_hem = (nobj + obj_per_chunk - 1) / obj_per_chunk;
+		bt_chunk_num = bt_chunk_size / 8;
+
+		table->hem = kcalloc(num_hem, sizeof(*table->hem),
+					 GFP_KERNEL);
+		if (!table->hem)
+			goto err_kcalloc_hem_buf;
+
+		if (check_whether_bt_num_3(table->type, hop_num)) {
+			unsigned long num_bt_l1;
+
+			num_bt_l1 = (num_hem + bt_chunk_num - 1) /
+					     bt_chunk_num;
+			table->bt_l1 = kcalloc(num_bt_l1,
+					       sizeof(*table->bt_l1),
+					       GFP_KERNEL);
+			if (!table->bt_l1)
+				goto err_kcalloc_bt_l1;
+
+			table->bt_l1_dma_addr = kcalloc(num_bt_l1,
+						 sizeof(*table->bt_l1_dma_addr),
+						 GFP_KERNEL);
+
+			if (!table->bt_l1_dma_addr)
+				goto err_kcalloc_l1_dma;
+		}
 
-	table->hem = kcalloc(num_hem, sizeof(*table->hem), GFP_KERNEL);
-	if (!table->hem)
-		return -ENOMEM;
+		if (check_whether_bt_num_2(table->type, hop_num) ||
+			check_whether_bt_num_3(table->type, hop_num)) {
+			table->bt_l0 = kcalloc(num_bt_l0, sizeof(*table->bt_l0),
+					       GFP_KERNEL);
+			if (!table->bt_l0)
+				goto err_kcalloc_bt_l0;
+
+			table->bt_l0_dma_addr = kcalloc(num_bt_l0,
+						 sizeof(*table->bt_l0_dma_addr),
+						 GFP_KERNEL);
+			if (!table->bt_l0_dma_addr)
+				goto err_kcalloc_l0_dma;
+		}
+	}
 
 	table->type = type;
 	table->num_hem = num_hem;
@@ -375,6 +880,54 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev,
 	mutex_init(&table->mutex);
 
 	return 0;
+
+err_kcalloc_l0_dma:
+	kfree(table->bt_l0);
+	table->bt_l0 = NULL;
+
+err_kcalloc_bt_l0:
+	kfree(table->bt_l1_dma_addr);
+	table->bt_l1_dma_addr = NULL;
+
+err_kcalloc_l1_dma:
+	kfree(table->bt_l1);
+	table->bt_l1 = NULL;
+
+err_kcalloc_bt_l1:
+	kfree(table->hem);
+	table->hem = NULL;
+
+err_kcalloc_hem_buf:
+	return -ENOMEM;
+}
+
+void hns_roce_cleanup_mhop_hem_table(struct hns_roce_dev *hr_dev,
+				     struct hns_roce_hem_table *table)
+{
+	struct hns_roce_hem_mhop mhop;
+	u32 buf_chunk_size;
+	int i;
+	u64 obj;
+
+	hns_roce_calc_hem_mhop(hr_dev, table, NULL, &mhop);
+	buf_chunk_size = mhop.buf_chunk_size;
+
+	for (i = 0; i < table->num_hem; ++i) {
+		obj = i * buf_chunk_size / table->obj_size;
+		if (table->hem[i])
+			hns_roce_table_mhop_put(hr_dev, table, obj, 0);
+	}
+
+	kfree(table->hem);
+	table->hem = NULL;
+	kfree(table->bt_l1);
+	table->bt_l1 = NULL;
+	kfree(table->bt_l1_dma_addr);
+	table->bt_l1_dma_addr = NULL;
+	kfree(table->bt_l0);
+	table->bt_l0 = NULL;
+	kfree(table->bt_l0_dma_addr);
+	table->bt_l0_dma_addr = NULL;
 }
 
 void hns_roce_cleanup_hem_table(struct hns_roce_dev *hr_dev,
@@ -383,10 +936,15 @@ void hns_roce_cleanup_hem_table(struct hns_roce_dev *hr_dev,
 	struct device *dev = hr_dev->dev;
 	unsigned long i;
 
+	if (hns_roce_check_whether_mhop(hr_dev, table->type)) {
+		hns_roce_cleanup_mhop_hem_table(hr_dev, table);
+		return;
+	}
+
 	for (i = 0; i < table->num_hem; ++i)
 		if (table->hem[i]) {
 			if (hr_dev->hw->clear_hem(hr_dev, table,
-			    i * HNS_ROCE_TABLE_CHUNK_SIZE / table->obj_size))
+			    i * HNS_ROCE_TABLE_CHUNK_SIZE / table->obj_size, 0))
 				dev_err(dev, "Clear HEM base address failed.\n");
 
 			hns_roce_free_hem(hr_dev, table->hem[i]);
diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.h b/drivers/infiniband/hw/hns/hns_roce_hem.h
index 4357488..b064ef5 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hem.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hem.h
@@ -54,6 +54,15 @@ enum {
 	 ((256 - sizeof(struct list_head) - 2 * sizeof(int)) /	 \
 	 (sizeof(struct scatterlist)))
 
+#define check_whether_bt_num_3(type, hop_num) \
+	(type < HEM_TYPE_MTT && hop_num == 2)
+
+#define check_whether_bt_num_2(type, hop_num) \
+	(type < HEM_TYPE_MTT && hop_num == 1)
+
+#define check_whether_bt_num_1(type, hop_num) \
+	(type < HEM_TYPE_MTT && hop_num == HNS_ROCE_HOP_NUM_0)
+
 enum {
 	 HNS_ROCE_HEM_PAGE_SHIFT = 12,
 	 HNS_ROCE_HEM_PAGE_SIZE  = 1 << HNS_ROCE_HEM_PAGE_SHIFT,
@@ -77,6 +86,16 @@ struct hns_roce_hem_iter {
 	int				 page_idx;
 };
 
+struct hns_roce_hem_mhop {
+	u32	hop_num;
+	u32	buf_chunk_size;
+	u32	bt_chunk_size;
+	u32	ba_l0_num;
+	u32	l0_idx;/* level 0 base address table index */
+	u32	l1_idx;/* level 1 base address table index */
+	u32	l2_idx;/* level 2 base address table index */
+};
+
 void hns_roce_free_hem(struct hns_roce_dev *hr_dev, struct hns_roce_hem *hem);
 int hns_roce_table_get(struct hns_roce_dev *hr_dev,
 		       struct hns_roce_hem_table *table, unsigned long obj);
@@ -97,6 +116,10 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev,
 void hns_roce_cleanup_hem_table(struct hns_roce_dev *hr_dev,
 				struct hns_roce_hem_table *table);
 void hns_roce_cleanup_hem(struct hns_roce_dev *hr_dev);
+int hns_roce_calc_hem_mhop(struct hns_roce_dev *hr_dev,
+			   struct hns_roce_hem_table *table, unsigned long *obj,
+			   struct hns_roce_hem_mhop *mhop);
+bool hns_roce_check_whether_mhop(struct hns_roce_dev *hr_dev, u32 type);
 
 static inline void hns_roce_hem_first(struct hns_roce_hem *hem,
 				      struct hns_roce_hem_iter *iter)
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 2bf2816..50dee13 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -2356,7 +2356,7 @@ int hns_roce_v1_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
 }
 
 int hns_roce_v1_clear_hem(struct hns_roce_dev *hr_dev,
-		struct hns_roce_hem_table *table, int obj)
+		struct hns_roce_hem_table *table, int obj, int step_idx)
 {
 	struct device *dev = &hr_dev->pdev->dev;
 	struct hns_roce_v1_priv *priv;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 259af2e..5127edc 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -545,6 +545,19 @@ static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev)
 	caps->reserved_uars	= 0;
 	caps->reserved_cqs	= 0;
 
+	caps->qpc_ba_pg_sz	= 0;
+	caps->qpc_buf_pg_sz	= 0;
+	caps->qpc_hop_num	= HNS_ROCE_CONTEXT_HOP_NUM;
+	caps->srqc_ba_pg_sz	= 0;
+	caps->srqc_buf_pg_sz	= 0;
+	caps->srqc_hop_num	= HNS_ROCE_HOP_NUM_0;
+	caps->cqc_ba_pg_sz	= 0;
+	caps->cqc_buf_pg_sz	= 0;
+	caps->cqc_hop_num	= HNS_ROCE_CONTEXT_HOP_NUM;
+	caps->mpt_ba_pg_sz	= 0;
+	caps->mpt_buf_pg_sz	= 0;
+	caps->mpt_hop_num	= HNS_ROCE_CONTEXT_HOP_NUM;
+
 	caps->pkey_table_len[0] = 1;
 	caps->gid_table_len[0] = 2;
 	caps->local_ca_ack_delay = 0;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index b420ea9..91f5161 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -72,6 +72,9 @@
 #define HNS_ROCE_V2_MAX_INNER_MTPT_NUM		2
 #define HNS_ROCE_CMQ_TX_TIMEOUT			200
 
+#define HNS_ROCE_CONTEXT_HOP_NUM		1
+#define HNS_ROCE_MTT_HOP_NUM			1
+
 #define HNS_ROCE_CMD_FLAG_IN_VALID_SHIFT	0
 #define HNS_ROCE_CMD_FLAG_OUT_VALID_SHIFT	1
 #define HNS_ROCE_CMD_FLAG_NEXT_SHIFT		2
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 09/20] RDMA/hns: Configure BT BA and BT attribute for the contexts in hip08

[Wei Hu (Xavier)]

BT is used to retrieve the addresses of the contexts(QPC/MPT/CQC/SRQC)
in memory. In order to support multi hop addressing for the contexts,
the BT BA should be configured by mailbox, and the BT attribution will
be set by command.

This patch is to configure the BT BA and BT attribution for the
contexts.

Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_cmd.h   |  46 +++++++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 198 ++++++++++++++++++++++++++++-
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h |  49 +++++++
 3 files changed, 292 insertions(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_cmd.h b/drivers/infiniband/hw/hns/hns_roce_cmd.h
index f5a9ee2..fc40069 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cmd.h
+++ b/drivers/infiniband/hw/hns/hns_roce_cmd.h
@@ -37,6 +37,52 @@
 #define HNS_ROCE_CMD_TIMEOUT_MSECS	10000
 
 enum {
+	/* QPC BT commands */
+	HNS_ROCE_CMD_WRITE_QPC_BT0	= 0x0,
+	HNS_ROCE_CMD_WRITE_QPC_BT1	= 0x1,
+	HNS_ROCE_CMD_WRITE_QPC_BT2	= 0x2,
+	HNS_ROCE_CMD_READ_QPC_BT0	= 0x4,
+	HNS_ROCE_CMD_READ_QPC_BT1	= 0x5,
+	HNS_ROCE_CMD_READ_QPC_BT2	= 0x6,
+	HNS_ROCE_CMD_DESTROY_QPC_BT0	= 0x8,
+	HNS_ROCE_CMD_DESTROY_QPC_BT1	= 0x9,
+	HNS_ROCE_CMD_DESTROY_QPC_BT2	= 0xa,
+
+	/* CQC BT commands */
+	HNS_ROCE_CMD_WRITE_CQC_BT0	= 0x10,
+	HNS_ROCE_CMD_WRITE_CQC_BT1	= 0x11,
+	HNS_ROCE_CMD_WRITE_CQC_BT2	= 0x12,
+	HNS_ROCE_CMD_READ_CQC_BT0	= 0x14,
+	HNS_ROCE_CMD_READ_CQC_BT1	= 0x15,
+	HNS_ROCE_CMD_READ_CQC_BT2	= 0x1b,
+	HNS_ROCE_CMD_DESTROY_CQC_BT0	= 0x18,
+	HNS_ROCE_CMD_DESTROY_CQC_BT1	= 0x19,
+	HNS_ROCE_CMD_DESTROY_CQC_BT2	= 0x1a,
+
+	/* MPT BT commands */
+	HNS_ROCE_CMD_WRITE_MPT_BT0	= 0x20,
+	HNS_ROCE_CMD_WRITE_MPT_BT1	= 0x21,
+	HNS_ROCE_CMD_WRITE_MPT_BT2	= 0x22,
+	HNS_ROCE_CMD_READ_MPT_BT0	= 0x24,
+	HNS_ROCE_CMD_READ_MPT_BT1	= 0x25,
+	HNS_ROCE_CMD_READ_MPT_BT2	= 0x26,
+	HNS_ROCE_CMD_DESTROY_MPT_BT0	= 0x28,
+	HNS_ROCE_CMD_DESTROY_MPT_BT1	= 0x29,
+	HNS_ROCE_CMD_DESTROY_MPT_BT2	= 0x2a,
+
+	/* SRQC BT commands */
+	HNS_ROCE_CMD_WRITE_SRQC_BT0	= 0x30,
+	HNS_ROCE_CMD_WRITE_SRQC_BT1	= 0x31,
+	HNS_ROCE_CMD_WRITE_SRQC_BT2	= 0x32,
+	HNS_ROCE_CMD_READ_SRQC_BT0	= 0x34,
+	HNS_ROCE_CMD_READ_SRQC_BT1	= 0x35,
+	HNS_ROCE_CMD_READ_SRQC_BT2	= 0x36,
+	HNS_ROCE_CMD_DESTROY_SRQC_BT0	= 0x38,
+	HNS_ROCE_CMD_DESTROY_SRQC_BT1	= 0x39,
+	HNS_ROCE_CMD_DESTROY_SRQC_BT2	= 0x3a,
+};
+
+enum {
 	/* TPT commands */
 	HNS_ROCE_CMD_SW2HW_MPT		= 0xd,
 	HNS_ROCE_CMD_HW2SW_MPT		= 0xf,
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 5127edc..77cf476 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -475,6 +475,62 @@ static int hns_roce_alloc_vf_resource(struct hns_roce_dev *hr_dev)
 	return hns_roce_cmq_send(hr_dev, desc, 2);
 }
 
+static int hns_roce_v2_set_bt(struct hns_roce_dev *hr_dev)
+{
+	u8 srqc_hop_num = hr_dev->caps.srqc_hop_num;
+	u8 qpc_hop_num = hr_dev->caps.qpc_hop_num;
+	u8 cqc_hop_num = hr_dev->caps.cqc_hop_num;
+	u8 mpt_hop_num = hr_dev->caps.mpt_hop_num;
+	struct hns_roce_cfg_bt_attr *req;
+	struct hns_roce_cmq_desc desc;
+
+	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_BT_ATTR, false);
+	req = (struct hns_roce_cfg_bt_attr *)desc.data;
+	memset(req, 0, sizeof(*req));
+
+	roce_set_field(req->vf_qpc_cfg, CFG_BT_ATTR_DATA_0_VF_QPC_BA_PGSZ_M,
+		       CFG_BT_ATTR_DATA_0_VF_QPC_BA_PGSZ_S,
+		       hr_dev->caps.qpc_ba_pg_sz);
+	roce_set_field(req->vf_qpc_cfg, CFG_BT_ATTR_DATA_0_VF_QPC_BUF_PGSZ_M,
+		       CFG_BT_ATTR_DATA_0_VF_QPC_BUF_PGSZ_S,
+		       hr_dev->caps.qpc_buf_pg_sz);
+	roce_set_field(req->vf_qpc_cfg, CFG_BT_ATTR_DATA_0_VF_QPC_HOPNUM_M,
+		       CFG_BT_ATTR_DATA_0_VF_QPC_HOPNUM_S,
+		       qpc_hop_num == HNS_ROCE_HOP_NUM_0 ? 0 : qpc_hop_num);
+
+	roce_set_field(req->vf_srqc_cfg, CFG_BT_ATTR_DATA_1_VF_SRQC_BA_PGSZ_M,
+		       CFG_BT_ATTR_DATA_1_VF_SRQC_BA_PGSZ_S,
+		       hr_dev->caps.srqc_ba_pg_sz);
+	roce_set_field(req->vf_srqc_cfg, CFG_BT_ATTR_DATA_1_VF_SRQC_BUF_PGSZ_M,
+		       CFG_BT_ATTR_DATA_1_VF_SRQC_BUF_PGSZ_S,
+		       hr_dev->caps.srqc_buf_pg_sz);
+	roce_set_field(req->vf_srqc_cfg, CFG_BT_ATTR_DATA_1_VF_SRQC_HOPNUM_M,
+		       CFG_BT_ATTR_DATA_1_VF_SRQC_HOPNUM_S,
+		       srqc_hop_num == HNS_ROCE_HOP_NUM_0 ? 0 : srqc_hop_num);
+
+	roce_set_field(req->vf_cqc_cfg, CFG_BT_ATTR_DATA_2_VF_CQC_BA_PGSZ_M,
+		       CFG_BT_ATTR_DATA_2_VF_CQC_BA_PGSZ_S,
+		       hr_dev->caps.cqc_ba_pg_sz);
+	roce_set_field(req->vf_cqc_cfg, CFG_BT_ATTR_DATA_2_VF_CQC_BUF_PGSZ_M,
+		       CFG_BT_ATTR_DATA_2_VF_CQC_BUF_PGSZ_S,
+		       hr_dev->caps.cqc_buf_pg_sz);
+	roce_set_field(req->vf_cqc_cfg, CFG_BT_ATTR_DATA_2_VF_CQC_HOPNUM_M,
+		       CFG_BT_ATTR_DATA_2_VF_CQC_HOPNUM_S,
+		       cqc_hop_num == HNS_ROCE_HOP_NUM_0 ? 0 : cqc_hop_num);
+
+	roce_set_field(req->vf_mpt_cfg, CFG_BT_ATTR_DATA_3_VF_MPT_BA_PGSZ_M,
+		       CFG_BT_ATTR_DATA_3_VF_MPT_BA_PGSZ_S,
+		       hr_dev->caps.mpt_ba_pg_sz);
+	roce_set_field(req->vf_mpt_cfg, CFG_BT_ATTR_DATA_3_VF_MPT_BUF_PGSZ_M,
+		       CFG_BT_ATTR_DATA_3_VF_MPT_BUF_PGSZ_S,
+		       hr_dev->caps.mpt_buf_pg_sz);
+	roce_set_field(req->vf_mpt_cfg, CFG_BT_ATTR_DATA_3_VF_MPT_HOPNUM_M,
+		       CFG_BT_ATTR_DATA_3_VF_MPT_HOPNUM_S,
+		       mpt_hop_num == HNS_ROCE_HOP_NUM_0 ? 0 : mpt_hop_num);
+
+	return hns_roce_cmq_send(hr_dev, &desc, 1);
+}
+
 static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev)
 {
 	struct hns_roce_caps *caps = &hr_dev->caps;
@@ -563,7 +619,12 @@ static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev)
 	caps->local_ca_ack_delay = 0;
 	caps->max_mtu = IB_MTU_4096;
 
-	return 0;
+	ret = hns_roce_v2_set_bt(hr_dev);
+	if (ret)
+		dev_err(hr_dev->dev, "Configure bt attribute fail, ret = %d.\n",
+			ret);
+
+	return ret;
 }
 
 static int hns_roce_v2_cmd_pending(struct hns_roce_dev *hr_dev)
@@ -648,12 +709,147 @@ static int hns_roce_v2_chk_mbox(struct hns_roce_dev *hr_dev,
 	return 0;
 }
 
+static int hns_roce_v2_set_hem(struct hns_roce_dev *hr_dev,
+			       struct hns_roce_hem_table *table, int obj,
+			       int step_idx)
+{
+	struct device *dev = hr_dev->dev;
+	struct hns_roce_cmd_mailbox *mailbox;
+	struct hns_roce_hem_iter iter;
+	struct hns_roce_hem_mhop mhop;
+	struct hns_roce_hem *hem;
+	unsigned long mhop_obj = obj;
+	int i, j, k;
+	int ret = 0;
+	u64 hem_idx = 0;
+	u64 l1_idx = 0;
+	u64 bt_ba = 0;
+	u32 chunk_ba_num;
+	u32 hop_num;
+	u16 op = 0xff;
+
+	if (!hns_roce_check_whether_mhop(hr_dev, table->type))
+		return 0;
+
+	hns_roce_calc_hem_mhop(hr_dev, table, &mhop_obj, &mhop);
+	i = mhop.l0_idx;
+	j = mhop.l1_idx;
+	k = mhop.l2_idx;
+	hop_num = mhop.hop_num;
+	chunk_ba_num = mhop.bt_chunk_size / 8;
+
+	if (hop_num == 2) {
+		hem_idx = i * chunk_ba_num * chunk_ba_num + j * chunk_ba_num +
+			  k;
+		l1_idx = i * chunk_ba_num + j;
+	} else if (hop_num == 1) {
+		hem_idx = i * chunk_ba_num + j;
+	} else if (hop_num == HNS_ROCE_HOP_NUM_0) {
+		hem_idx = i;
+	}
+
+	switch (table->type) {
+	case HEM_TYPE_QPC:
+		op = HNS_ROCE_CMD_WRITE_QPC_BT0;
+		break;
+	case HEM_TYPE_MTPT:
+		op = HNS_ROCE_CMD_WRITE_MPT_BT0;
+		break;
+	case HEM_TYPE_CQC:
+		op = HNS_ROCE_CMD_WRITE_CQC_BT0;
+		break;
+	case HEM_TYPE_SRQC:
+		op = HNS_ROCE_CMD_WRITE_SRQC_BT0;
+		break;
+	default:
+		dev_warn(dev, "Table %d not to be written by mailbox!\n",
+			 table->type);
+		return 0;
+	}
+	op += step_idx;
+
+	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
+	if (IS_ERR(mailbox))
+		return PTR_ERR(mailbox);
+
+	if (check_whether_last_step(hop_num, step_idx)) {
+		hem = table->hem[hem_idx];
+		for (hns_roce_hem_first(hem, &iter);
+		     !hns_roce_hem_last(&iter); hns_roce_hem_next(&iter)) {
+			bt_ba = hns_roce_hem_addr(&iter);
+
+			/* configure the ba, tag, and op */
+			ret = hns_roce_cmd_mbox(hr_dev, bt_ba, mailbox->dma,
+						obj, 0, op,
+						HNS_ROCE_CMD_TIMEOUT_MSECS);
+		}
+	} else {
+		if (step_idx == 0)
+			bt_ba = table->bt_l0_dma_addr[i];
+		else if (step_idx == 1 && hop_num == 2)
+			bt_ba = table->bt_l1_dma_addr[l1_idx];
+
+		/* configure the ba, tag, and op */
+		ret = hns_roce_cmd_mbox(hr_dev, bt_ba, mailbox->dma, obj,
+					0, op, HNS_ROCE_CMD_TIMEOUT_MSECS);
+	}
+
+	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
+	return ret;
+}
+
+static int hns_roce_v2_clear_hem(struct hns_roce_dev *hr_dev,
+				 struct hns_roce_hem_table *table, int obj,
+				 int step_idx)
+{
+	struct device *dev = hr_dev->dev;
+	struct hns_roce_cmd_mailbox *mailbox;
+	int ret = 0;
+	u16 op = 0xff;
+
+	if (!hns_roce_check_whether_mhop(hr_dev, table->type))
+		return 0;
+
+	switch (table->type) {
+	case HEM_TYPE_QPC:
+		op = HNS_ROCE_CMD_DESTROY_QPC_BT0;
+		break;
+	case HEM_TYPE_MTPT:
+		op = HNS_ROCE_CMD_DESTROY_MPT_BT0;
+		break;
+	case HEM_TYPE_CQC:
+		op = HNS_ROCE_CMD_DESTROY_CQC_BT0;
+		break;
+	case HEM_TYPE_SRQC:
+		op = HNS_ROCE_CMD_DESTROY_SRQC_BT0;
+		break;
+	default:
+		dev_warn(dev, "Table %d not to be destroyed by mailbox!\n",
+			 table->type);
+		return 0;
+	}
+	op += step_idx;
+
+	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
+	if (IS_ERR(mailbox))
+		return PTR_ERR(mailbox);
+
+	/* configure the tag and op */
+	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, obj, 0, op,
+				HNS_ROCE_CMD_TIMEOUT_MSECS);
+
+	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
+	return ret;
+}
+
 static const struct hns_roce_hw hns_roce_hw_v2 = {
 	.cmq_init = hns_roce_v2_cmq_init,
 	.cmq_exit = hns_roce_v2_cmq_exit,
 	.hw_profile = hns_roce_v2_profile,
 	.post_mbox = hns_roce_v2_post_mbox,
 	.chk_mbox = hns_roce_v2_chk_mbox,
+	.set_hem = hns_roce_v2_set_hem,
+	.clear_hem = hns_roce_v2_clear_hem,
 };
 
 static const struct pci_device_id hns_roce_hw_v2_pci_tbl[] = {
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index 91f5161..cc8c8d0 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -93,6 +93,11 @@
 #define HNS_ROCE_CMQ_EN_B		16
 #define HNS_ROCE_CMQ_ENABLE		BIT(HNS_ROCE_CMQ_EN_B)
 
+#define check_whether_last_step(hop_num, step_idx) \
+	((step_idx == 0 && hop_num == HNS_ROCE_HOP_NUM_0) || \
+	(step_idx == 1 && hop_num == 1) || \
+	(step_idx == 2 && hop_num == 2))
+
 /* CMQ command */
 enum hns_roce_opcode_type {
 	HNS_ROCE_OPC_QUERY_HW_VER			= 0x8000,
@@ -239,6 +244,50 @@ struct hns_roce_vf_res_b {
 #define VF_RES_B_DATA_3_VF_SL_NUM_S 16
 #define VF_RES_B_DATA_3_VF_SL_NUM_M GENMASK(19, 16)
 
+struct hns_roce_cfg_bt_attr {
+	u32 vf_qpc_cfg;
+	u32 vf_srqc_cfg;
+	u32 vf_cqc_cfg;
+	u32 vf_mpt_cfg;
+	u32 rsv[2];
+};
+
+#define CFG_BT_ATTR_DATA_0_VF_QPC_BA_PGSZ_S 0
+#define CFG_BT_ATTR_DATA_0_VF_QPC_BA_PGSZ_M GENMASK(3, 0)
+
+#define CFG_BT_ATTR_DATA_0_VF_QPC_BUF_PGSZ_S 4
+#define CFG_BT_ATTR_DATA_0_VF_QPC_BUF_PGSZ_M GENMASK(7, 4)
+
+#define CFG_BT_ATTR_DATA_0_VF_QPC_HOPNUM_S 8
+#define CFG_BT_ATTR_DATA_0_VF_QPC_HOPNUM_M GENMASK(9, 8)
+
+#define CFG_BT_ATTR_DATA_1_VF_SRQC_BA_PGSZ_S 0
+#define CFG_BT_ATTR_DATA_1_VF_SRQC_BA_PGSZ_M GENMASK(3, 0)
+
+#define CFG_BT_ATTR_DATA_1_VF_SRQC_BUF_PGSZ_S 4
+#define CFG_BT_ATTR_DATA_1_VF_SRQC_BUF_PGSZ_M GENMASK(7, 4)
+
+#define CFG_BT_ATTR_DATA_1_VF_SRQC_HOPNUM_S 8
+#define CFG_BT_ATTR_DATA_1_VF_SRQC_HOPNUM_M GENMASK(9, 8)
+
+#define CFG_BT_ATTR_DATA_2_VF_CQC_BA_PGSZ_S 0
+#define CFG_BT_ATTR_DATA_2_VF_CQC_BA_PGSZ_M GENMASK(3, 0)
+
+#define CFG_BT_ATTR_DATA_2_VF_CQC_BUF_PGSZ_S 4
+#define CFG_BT_ATTR_DATA_2_VF_CQC_BUF_PGSZ_M GENMASK(7, 4)
+
+#define CFG_BT_ATTR_DATA_2_VF_CQC_HOPNUM_S 8
+#define CFG_BT_ATTR_DATA_2_VF_CQC_HOPNUM_M GENMASK(9, 8)
+
+#define CFG_BT_ATTR_DATA_3_VF_MPT_BA_PGSZ_S 0
+#define CFG_BT_ATTR_DATA_3_VF_MPT_BA_PGSZ_M GENMASK(3, 0)
+
+#define CFG_BT_ATTR_DATA_3_VF_MPT_BUF_PGSZ_S 4
+#define CFG_BT_ATTR_DATA_3_VF_MPT_BUF_PGSZ_M GENMASK(7, 4)
+
+#define CFG_BT_ATTR_DATA_3_VF_MPT_HOPNUM_S 8
+#define CFG_BT_ATTR_DATA_3_VF_MPT_HOPNUM_M GENMASK(9, 8)
+
 struct hns_roce_cmq_desc {
 	u16 opcode;
 	u16 flag;
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 10/20] RDMA/hns: Update the interfaces for MTT/CQE multi hop addressing in hip08

[Wei Hu (Xavier)]

From: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>

The MTT(SQWQE/SGE/RQWQE) and CQE in hip08 can support multi hop
addressing. The address of MTT/CQE can be retrieved by the BT
(Base Address Table) with multi hop addressing.

This patch is to update the interfaces in HEM to support multi hop
addressing for the MTT/CQE.

Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_cq.c     |  12 +--
 drivers/infiniband/hw/hns/hns_roce_device.h |   6 ++
 drivers/infiniband/hw/hns/hns_roce_hem.c    | 119 +++++++++++++++++++++++-----
 drivers/infiniband/hw/hns/hns_roce_hem.h    |  11 ++-
 drivers/infiniband/hw/hns/hns_roce_hw_v1.c  |   8 +-
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  |   6 ++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  |   1 +
 drivers/infiniband/hw/hns/hns_roce_mr.c     |  16 ++--
 8 files changed, 143 insertions(+), 36 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_cq.c b/drivers/infiniband/hw/hns/hns_roce_cq.c
index 5f51242..362aeec 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cq.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cq.c
@@ -85,17 +85,19 @@ static int hns_roce_cq_alloc(struct hns_roce_dev *hr_dev, int nent,
 			     struct hns_roce_uar *hr_uar,
 			     struct hns_roce_cq *hr_cq, int vector)
 {
-	struct hns_roce_cmd_mailbox *mailbox = NULL;
-	struct hns_roce_cq_table *cq_table = NULL;
+	struct hns_roce_cmd_mailbox *mailbox;
+	struct hns_roce_hem_table *mtt_table;
+	struct hns_roce_cq_table *cq_table;
 	struct device *dev = hr_dev->dev;
 	dma_addr_t dma_handle;
-	u64 *mtts = NULL;
-	int ret = 0;
+	u64 *mtts;
+	int ret;
 
 	cq_table = &hr_dev->cq_table;
 
 	/* Get the physical address of cq buf */
-	mtts = hns_roce_table_find(&hr_dev->mr_table.mtt_table,
+	mtt_table = &hr_dev->mr_table.mtt_table;
+	mtts = hns_roce_table_find(hr_dev, mtt_table,
 				   hr_mtt->first_seg, &dma_handle);
 	if (!mtts) {
 		dev_err(dev, "CQ alloc.Failed to find cq buf addr.\n");
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 4574a7b..7dd8fbb2 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -525,6 +525,12 @@ struct hns_roce_caps {
 	u32		mpt_ba_pg_sz;
 	u32		mpt_buf_pg_sz;
 	u32		mpt_hop_num;
+	u32		mtt_ba_pg_sz;
+	u32		mtt_buf_pg_sz;
+	u32		mtt_hop_num;
+	u32		cqe_ba_pg_sz;
+	u32		cqe_buf_pg_sz;
+	u32		cqe_hop_num;
 };
 
 struct hns_roce_hw {
diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c
index 9bc8c6b..ac2d671 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hem.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hem.c
@@ -47,7 +47,9 @@ bool hns_roce_check_whether_mhop(struct hns_roce_dev *hr_dev, u32 type)
 	if ((hr_dev->caps.qpc_hop_num && type == HEM_TYPE_QPC) ||
 	    (hr_dev->caps.mpt_hop_num && type == HEM_TYPE_MTPT) ||
 	    (hr_dev->caps.cqc_hop_num && type == HEM_TYPE_CQC) ||
-	    (hr_dev->caps.srqc_hop_num && type == HEM_TYPE_SRQC))
+	    (hr_dev->caps.srqc_hop_num && type == HEM_TYPE_SRQC) ||
+	    (hr_dev->caps.cqe_hop_num && type == HEM_TYPE_CQE) ||
+	    (hr_dev->caps.mtt_hop_num && type == HEM_TYPE_MTT))
 		return true;
 
 	return false;
@@ -132,6 +134,22 @@ int hns_roce_calc_hem_mhop(struct hns_roce_dev *hr_dev,
 		mhop->ba_l0_num = hr_dev->caps.srqc_bt_num;
 		mhop->hop_num = hr_dev->caps.srqc_hop_num;
 		break;
+	case HEM_TYPE_MTT:
+		mhop->buf_chunk_size = 1 << (hr_dev->caps.mtt_buf_pg_sz
+					     + PAGE_SHIFT);
+		mhop->bt_chunk_size = 1 << (hr_dev->caps.mtt_ba_pg_sz
+					     + PAGE_SHIFT);
+		mhop->ba_l0_num = mhop->bt_chunk_size / 8;
+		mhop->hop_num = hr_dev->caps.mtt_hop_num;
+		break;
+	case HEM_TYPE_CQE:
+		mhop->buf_chunk_size = 1 << (hr_dev->caps.cqe_buf_pg_sz
+					     + PAGE_SHIFT);
+		mhop->bt_chunk_size = 1 << (hr_dev->caps.cqe_ba_pg_sz
+					     + PAGE_SHIFT);
+		mhop->ba_l0_num = mhop->bt_chunk_size / 8;
+		mhop->hop_num = hr_dev->caps.cqe_hop_num;
+		break;
 	default:
 		dev_err(dev, "Table %d not support multi-hop addressing!\n",
 			 table->type);
@@ -141,10 +159,14 @@ int hns_roce_calc_hem_mhop(struct hns_roce_dev *hr_dev,
 	if (!obj)
 		return 0;
 
-	/* QPC/MTPT/CQC/SRQC alloc hem for buffer pages. */
+	/*
+	 * QPC/MTPT/CQC/SRQC alloc hem for buffer pages.
+	 * MTT/CQE alloc hem for bt pages.
+	 */
 	bt_num = hns_roce_get_bt_num(table->type, mhop->hop_num);
 	chunk_ba_num = mhop->bt_chunk_size / 8;
-	chunk_size = mhop->buf_chunk_size;
+	chunk_size = table->type < HEM_TYPE_MTT ? mhop->buf_chunk_size :
+			      mhop->bt_chunk_size;
 	table_idx = (*obj & (table->num_obj - 1)) /
 		     (chunk_size / table->obj_size);
 	switch (bt_num) {
@@ -448,8 +470,11 @@ int hns_roce_table_mhop_get(struct hns_roce_dev *hr_dev,
 		}
 	}
 
-	/* alloc buffer space chunk for QPC/MTPT/CQC/SRQC. */
-	size = buf_chunk_size;
+	/*
+	 * alloc buffer space chunk for QPC/MTPT/CQC/SRQC.
+	 * alloc bt space chunk for MTT/CQE.
+	 */
+	size = table->type < HEM_TYPE_MTT ? buf_chunk_size : bt_chunk_size;
 	table->hem[hem_idx] = hns_roce_alloc_hem(hr_dev,
 						size >> PAGE_SHIFT,
 						size,
@@ -480,6 +505,8 @@ int hns_roce_table_mhop_get(struct hns_roce_dev *hr_dev,
 			dev_err(dev, "set HEM base address to HW failed!\n");
 			goto err_alloc_hem_buf;
 		}
+	} else if (hop_num == 2) {
+		*(table->bt_l0[bt_l0_idx] + mhop.l1_idx) = bt_ba;
 	}
 
 	++table->hem[hem_idx]->refcount;
@@ -610,7 +637,10 @@ void hns_roce_table_mhop_put(struct hns_roce_dev *hr_dev,
 			dev_warn(dev, "Clear HEM base address failed.\n");
 	}
 
-	/* free buffer space chunk for QPC/MTPT/CQC/SRQC. */
+	/*
+	 * free buffer space chunk for QPC/MTPT/CQC/SRQC.
+	 * free bt space chunk for MTT/CQE.
+	 */
 	hns_roce_free_hem(hr_dev, table->hem[hem_idx]);
 	table->hem[hem_idx] = NULL;
 
@@ -686,23 +716,46 @@ void hns_roce_table_put(struct hns_roce_dev *hr_dev,
 	mutex_unlock(&table->mutex);
 }
 
-void *hns_roce_table_find(struct hns_roce_hem_table *table, unsigned long obj,
-			  dma_addr_t *dma_handle)
+void *hns_roce_table_find(struct hns_roce_dev *hr_dev,
+			  struct hns_roce_hem_table *table,
+			  unsigned long obj, dma_addr_t *dma_handle)
 {
 	struct hns_roce_hem_chunk *chunk;
-	unsigned long idx;
-	int i;
-	int offset, dma_offset;
+	struct hns_roce_hem_mhop mhop;
 	struct hns_roce_hem *hem;
 	struct page *page = NULL;
+	unsigned long mhop_obj = obj;
+	unsigned long idx;
+	int offset, dma_offset;
+	int i, j;
+	u32 hem_idx = 0;
 
 	if (!table->lowmem)
 		return NULL;
 
 	mutex_lock(&table->mutex);
-	idx = (obj & (table->num_obj - 1)) * table->obj_size;
-	hem = table->hem[idx / HNS_ROCE_TABLE_CHUNK_SIZE];
-	dma_offset = offset = idx % HNS_ROCE_TABLE_CHUNK_SIZE;
+
+	if (!hns_roce_check_whether_mhop(hr_dev, table->type)) {
+		idx = (obj & (table->num_obj - 1)) * table->obj_size;
+		hem = table->hem[idx / HNS_ROCE_TABLE_CHUNK_SIZE];
+		dma_offset = offset = idx % HNS_ROCE_TABLE_CHUNK_SIZE;
+	} else {
+		hns_roce_calc_hem_mhop(hr_dev, table, &mhop_obj, &mhop);
+		/* mtt mhop */
+		i = mhop.l0_idx;
+		j = mhop.l1_idx;
+		if (mhop.hop_num == 2)
+			hem_idx = i * (mhop.bt_chunk_size / 8) + j;
+		else if (mhop.hop_num == 1 ||
+			 mhop.hop_num == HNS_ROCE_HOP_NUM_0)
+			hem_idx = i;
+
+		hem = table->hem[hem_idx];
+		dma_offset = offset = (obj & (table->num_obj - 1)) *
+				       table->obj_size % mhop.bt_chunk_size;
+		if (mhop.hop_num == 2)
+			dma_offset = offset = 0;
+	}
 
 	if (!hem)
 		goto out;
@@ -735,9 +788,15 @@ int hns_roce_table_get_range(struct hns_roce_dev *hr_dev,
 			     struct hns_roce_hem_table *table,
 			     unsigned long start, unsigned long end)
 {
+	struct hns_roce_hem_mhop mhop;
 	unsigned long inc = HNS_ROCE_TABLE_CHUNK_SIZE / table->obj_size;
-	unsigned long i = 0;
-	int ret = 0;
+	unsigned long i;
+	int ret;
+
+	if (hns_roce_check_whether_mhop(hr_dev, table->type)) {
+		hns_roce_calc_hem_mhop(hr_dev, table, NULL, &mhop);
+		inc = mhop.bt_chunk_size / table->obj_size;
+	}
 
 	/* Allocate MTT entry memory according to chunk(128K) */
 	for (i = start; i <= end; i += inc) {
@@ -760,10 +819,17 @@ void hns_roce_table_put_range(struct hns_roce_dev *hr_dev,
 			      struct hns_roce_hem_table *table,
 			      unsigned long start, unsigned long end)
 {
+	struct hns_roce_hem_mhop mhop;
+	unsigned long inc = HNS_ROCE_TABLE_CHUNK_SIZE / table->obj_size;
 	unsigned long i;
 
+	if (hns_roce_check_whether_mhop(hr_dev, table->type)) {
+		hns_roce_calc_hem_mhop(hr_dev, table, NULL, &mhop);
+		inc = mhop.bt_chunk_size / table->obj_size;
+	}
+
 	for (i = start; i <= end;
-		i += HNS_ROCE_TABLE_CHUNK_SIZE / table->obj_size)
+		i += inc)
 		hns_roce_table_put(hr_dev, table, i);
 }
 
@@ -787,7 +853,7 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev,
 		unsigned long buf_chunk_size;
 		unsigned long bt_chunk_size;
 		unsigned long bt_chunk_num;
-		unsigned long num_bt_l0;
+		unsigned long num_bt_l0 = 0;
 		u32 hop_num;
 
 		switch (type) {
@@ -823,6 +889,18 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev,
 			num_bt_l0 = hr_dev->caps.srqc_bt_num;
 			hop_num = hr_dev->caps.srqc_hop_num;
 			break;
+		case HEM_TYPE_MTT:
+			buf_chunk_size = 1 << (hr_dev->caps.mtt_ba_pg_sz
+					+ PAGE_SHIFT);
+			bt_chunk_size = buf_chunk_size;
+			hop_num = hr_dev->caps.mtt_hop_num;
+			break;
+		case HEM_TYPE_CQE:
+			buf_chunk_size = 1 << (hr_dev->caps.cqe_ba_pg_sz
+					+ PAGE_SHIFT);
+			bt_chunk_size = buf_chunk_size;
+			hop_num = hr_dev->caps.cqe_hop_num;
+			break;
 		default:
 			dev_err(dev,
 			  "Table %d not support to init hem table here!\n",
@@ -832,6 +910,8 @@ int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev,
 		obj_per_chunk = buf_chunk_size / obj_size;
 		num_hem = (nobj + obj_per_chunk - 1) / obj_per_chunk;
 		bt_chunk_num = bt_chunk_size / 8;
+		if (table->type >= HEM_TYPE_MTT)
+			num_bt_l0 = bt_chunk_num;
 
 		table->hem = kcalloc(num_hem, sizeof(*table->hem),
 					 GFP_KERNEL);
@@ -910,7 +990,8 @@ void hns_roce_cleanup_mhop_hem_table(struct hns_roce_dev *hr_dev,
 	u64 obj;
 
 	hns_roce_calc_hem_mhop(hr_dev, table, NULL, &mhop);
-	buf_chunk_size = mhop.buf_chunk_size;
+	buf_chunk_size = table->type < HEM_TYPE_MTT ? mhop.buf_chunk_size :
+					mhop.bt_chunk_size;
 
 	for (i = 0; i < table->num_hem; ++i) {
 		obj = i * buf_chunk_size / table->obj_size;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.h b/drivers/infiniband/hw/hns/hns_roce_hem.h
index b064ef5..1aa54b6 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hem.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hem.h
@@ -47,6 +47,7 @@ enum {
 
 	 /* UNMAP HEM */
 	HEM_TYPE_MTT,
+	HEM_TYPE_CQE,
 	HEM_TYPE_IRRL,
 };
 
@@ -58,10 +59,13 @@ enum {
 	(type < HEM_TYPE_MTT && hop_num == 2)
 
 #define check_whether_bt_num_2(type, hop_num) \
-	(type < HEM_TYPE_MTT && hop_num == 1)
+	((type < HEM_TYPE_MTT && hop_num == 1) || \
+	(type >= HEM_TYPE_MTT && hop_num == 2))
 
 #define check_whether_bt_num_1(type, hop_num) \
-	(type < HEM_TYPE_MTT && hop_num == HNS_ROCE_HOP_NUM_0)
+	((type < HEM_TYPE_MTT && hop_num == HNS_ROCE_HOP_NUM_0) || \
+	(type >= HEM_TYPE_MTT && hop_num == 1) || \
+	(type >= HEM_TYPE_MTT && hop_num == HNS_ROCE_HOP_NUM_0))
 
 enum {
 	 HNS_ROCE_HEM_PAGE_SHIFT = 12,
@@ -101,7 +105,8 @@ int hns_roce_table_get(struct hns_roce_dev *hr_dev,
 		       struct hns_roce_hem_table *table, unsigned long obj);
 void hns_roce_table_put(struct hns_roce_dev *hr_dev,
 			struct hns_roce_hem_table *table, unsigned long obj);
-void *hns_roce_table_find(struct hns_roce_hem_table *table, unsigned long obj,
+void *hns_roce_table_find(struct hns_roce_dev *hr_dev,
+			  struct hns_roce_hem_table *table, unsigned long obj,
 			  dma_addr_t *dma_handle);
 int hns_roce_table_get_range(struct hns_roce_dev *hr_dev,
 			     struct hns_roce_hem_table *table,
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 50dee13..0a634c5 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -2524,7 +2524,7 @@ static int hns_roce_v1_m_sqp(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
 		return -ENOMEM;
 
 	/* Search QP buf's MTTs */
-	mtts = hns_roce_table_find(&hr_dev->mr_table.mtt_table,
+	mtts = hns_roce_table_find(hr_dev, &hr_dev->mr_table.mtt_table,
 				   hr_qp->mtt.first_seg, &dma_handle);
 	if (!mtts) {
 		dev_err(dev, "qp buf pa find failed\n");
@@ -2671,7 +2671,7 @@ static int hns_roce_v1_m_qp(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
 		return -ENOMEM;
 
 	/* Search qp buf's mtts */
-	mtts = hns_roce_table_find(&hr_dev->mr_table.mtt_table,
+	mtts = hns_roce_table_find(hr_dev, &hr_dev->mr_table.mtt_table,
 				   hr_qp->mtt.first_seg, &dma_handle);
 	if (mtts == NULL) {
 		dev_err(dev, "qp buf pa find failed\n");
@@ -2679,8 +2679,8 @@ static int hns_roce_v1_m_qp(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
 	}
 
 	/* Search IRRL's mtts */
-	mtts_2 = hns_roce_table_find(&hr_dev->qp_table.irrl_table, hr_qp->qpn,
-				     &dma_handle_2);
+	mtts_2 = hns_roce_table_find(hr_dev, &hr_dev->qp_table.irrl_table,
+				     hr_qp->qpn, &dma_handle_2);
 	if (mtts_2 == NULL) {
 		dev_err(dev, "qp irrl_table find failed\n");
 		goto out;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 77cf476..c59d15e 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -613,6 +613,12 @@ static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev)
 	caps->mpt_ba_pg_sz	= 0;
 	caps->mpt_buf_pg_sz	= 0;
 	caps->mpt_hop_num	= HNS_ROCE_CONTEXT_HOP_NUM;
+	caps->mtt_ba_pg_sz	= 0;
+	caps->mtt_buf_pg_sz	= 0;
+	caps->mtt_hop_num	= HNS_ROCE_MTT_HOP_NUM;
+	caps->cqe_ba_pg_sz	= 0;
+	caps->cqe_buf_pg_sz	= 0;
+	caps->cqe_hop_num	= HNS_ROCE_CQE_HOP_NUM;
 
 	caps->pkey_table_len[0] = 1;
 	caps->gid_table_len[0] = 2;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index cc8c8d0..212d511 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -74,6 +74,7 @@
 
 #define HNS_ROCE_CONTEXT_HOP_NUM		1
 #define HNS_ROCE_MTT_HOP_NUM			1
+#define HNS_ROCE_CQE_HOP_NUM			1
 
 #define HNS_ROCE_CMD_FLAG_IN_VALID_SHIFT	0
 #define HNS_ROCE_CMD_FLAG_OUT_VALID_SHIFT	1
diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c b/drivers/infiniband/hw/hns/hns_roce_mr.c
index 923d2b4..d8f7b41 100644
--- a/drivers/infiniband/hw/hns/hns_roce_mr.c
+++ b/drivers/infiniband/hw/hns/hns_roce_mr.c
@@ -348,10 +348,11 @@ static int hns_roce_write_mtt_chunk(struct hns_roce_dev *hr_dev,
 				    struct hns_roce_mtt *mtt, u32 start_index,
 				    u32 npages, u64 *page_list)
 {
-	u32 i = 0;
-	__le64 *mtts = NULL;
+	struct hns_roce_hem_table *table;
 	dma_addr_t dma_handle;
+	__le64 *mtts;
 	u32 s = start_index * sizeof(u64);
+	u32 i;
 
 	/* All MTTs must fit in the same page */
 	if (start_index / (PAGE_SIZE / sizeof(u64)) !=
@@ -361,15 +362,20 @@ static int hns_roce_write_mtt_chunk(struct hns_roce_dev *hr_dev,
 	if (start_index & (HNS_ROCE_MTT_ENTRY_PER_SEG - 1))
 		return -EINVAL;
 
-	mtts = hns_roce_table_find(&hr_dev->mr_table.mtt_table,
+	table = &hr_dev->mr_table.mtt_table;
+	mtts = hns_roce_table_find(hr_dev, table,
 				mtt->first_seg + s / hr_dev->caps.mtt_entry_sz,
 				&dma_handle);
 	if (!mtts)
 		return -ENOMEM;
 
 	/* Save page addr, low 12 bits : 0 */
-	for (i = 0; i < npages; ++i)
-		mtts[i] = (cpu_to_le64(page_list[i])) >> PAGE_ADDR_SHIFT;
+	for (i = 0; i < npages; ++i) {
+		if (!hr_dev->caps.mtt_hop_num)
+			mtts[i] = cpu_to_le64(page_list[i] >> PAGE_ADDR_SHIFT);
+		else
+			mtts[i] = cpu_to_le64(page_list[i]);
+	}
 
 	return 0;
 }
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 11/20] RDMA/hns: Split CQE from MTT in hip08

[Wei Hu (Xavier)]

From: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>

In hip08, the SQWQE/SGE/RQWQE and CQE have different hop num and
page size, so we need to manage the base address table of the
SQWQE/SGE/RQWQE and CQE separately.

This patch is to split CQE from MTT(SQWQE/SGE/RQWQE).

Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_cq.c     | 15 ++++++-
 drivers/infiniband/hw/hns/hns_roce_device.h | 14 ++++--
 drivers/infiniband/hw/hns/hns_roce_hem.c    |  3 ++
 drivers/infiniband/hw/hns/hns_roce_main.c   | 17 ++++++++
 drivers/infiniband/hw/hns/hns_roce_mr.c     | 66 ++++++++++++++++++++++-------
 drivers/infiniband/hw/hns/hns_roce_qp.c     |  2 +
 6 files changed, 98 insertions(+), 19 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_cq.c b/drivers/infiniband/hw/hns/hns_roce_cq.c
index 362aeec..9cf213c 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cq.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cq.c
@@ -96,7 +96,11 @@ static int hns_roce_cq_alloc(struct hns_roce_dev *hr_dev, int nent,
 	cq_table = &hr_dev->cq_table;
 
 	/* Get the physical address of cq buf */
-	mtt_table = &hr_dev->mr_table.mtt_table;
+	if (hns_roce_check_whether_mhop(hr_dev, HEM_TYPE_CQE))
+		mtt_table = &hr_dev->mr_table.mtt_cqe_table;
+	else
+		mtt_table = &hr_dev->mr_table.mtt_table;
+
 	mtts = hns_roce_table_find(hr_dev, mtt_table,
 				   hr_mtt->first_seg, &dma_handle);
 	if (!mtts) {
@@ -221,6 +225,10 @@ static int hns_roce_ib_get_cq_umem(struct hns_roce_dev *hr_dev,
 	if (IS_ERR(*umem))
 		return PTR_ERR(*umem);
 
+	if (hns_roce_check_whether_mhop(hr_dev, HEM_TYPE_CQE))
+		buf->hr_mtt.mtt_type = MTT_TYPE_CQE;
+	else
+		buf->hr_mtt.mtt_type = MTT_TYPE_WQE;
 	ret = hns_roce_mtt_init(hr_dev, ib_umem_page_count(*umem),
 				(*umem)->page_shift, &buf->hr_mtt);
 	if (ret)
@@ -250,6 +258,11 @@ static int hns_roce_ib_alloc_cq_buf(struct hns_roce_dev *hr_dev,
 	if (ret)
 		goto out;
 
+	if (hns_roce_check_whether_mhop(hr_dev, HEM_TYPE_CQE))
+		buf->hr_mtt.mtt_type = MTT_TYPE_CQE;
+	else
+		buf->hr_mtt.mtt_type = MTT_TYPE_WQE;
+
 	ret = hns_roce_mtt_init(hr_dev, buf->hr_buf.npages,
 				buf->hr_buf.page_shift, &buf->hr_mtt);
 	if (ret)
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 7dd8fbb2..319c7054 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -170,6 +170,11 @@ enum {
 	HNS_ROCE_OPCODE_RDMA_WITH_IMM_RECEIVE	= 0x07,
 };
 
+enum hns_roce_mtt_type {
+	MTT_TYPE_WQE = 0,
+	MTT_TYPE_CQE,
+};
+
 #define HNS_ROCE_CMD_SUCCESS			1
 
 #define HNS_ROCE_PORT_DOWN			0
@@ -241,9 +246,10 @@ struct hns_roce_hem_table {
 };
 
 struct hns_roce_mtt {
-	unsigned long	first_seg;
-	int		order;
-	int		page_shift;
+	unsigned long		first_seg;
+	int			order;
+	int			page_shift;
+	enum hns_roce_mtt_type	mtt_type;
 };
 
 /* Only support 4K page size for mr register */
@@ -268,6 +274,8 @@ struct hns_roce_mr_table {
 	struct hns_roce_buddy		mtt_buddy;
 	struct hns_roce_hem_table	mtt_table;
 	struct hns_roce_hem_table	mtpt_table;
+	struct hns_roce_buddy		mtt_cqe_buddy;
+	struct hns_roce_hem_table	mtt_cqe_table;
 };
 
 struct hns_roce_wq {
diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c
index ac2d671..8a3e174 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hem.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hem.c
@@ -1041,4 +1041,7 @@ void hns_roce_cleanup_hem(struct hns_roce_dev *hr_dev)
 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.qp_table);
 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table);
 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtt_table);
+	if (hns_roce_check_whether_mhop(hr_dev, HEM_TYPE_CQE))
+		hns_roce_cleanup_hem_table(hr_dev,
+					   &hr_dev->mr_table.mtt_cqe_table);
 }
diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c
index db6593e..a110f96 100644
--- a/drivers/infiniband/hw/hns/hns_roce_main.c
+++ b/drivers/infiniband/hw/hns/hns_roce_main.c
@@ -546,6 +546,17 @@ static int hns_roce_init_hem(struct hns_roce_dev *hr_dev)
 		return ret;
 	}
 
+	if (hns_roce_check_whether_mhop(hr_dev, HEM_TYPE_CQE)) {
+		ret = hns_roce_init_hem_table(hr_dev,
+				      &hr_dev->mr_table.mtt_cqe_table,
+				      HEM_TYPE_CQE, hr_dev->caps.mtt_entry_sz,
+				      hr_dev->caps.num_cqe_segs, 1);
+		if (ret) {
+			dev_err(dev, "Failed to init MTT CQE context memory, aborting.\n");
+			goto err_unmap_cqe;
+		}
+	}
+
 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table,
 				      HEM_TYPE_MTPT, hr_dev->caps.mtpt_entry_sz,
 				      hr_dev->caps.num_mtpts, 1);
@@ -593,6 +604,12 @@ static int hns_roce_init_hem(struct hns_roce_dev *hr_dev)
 
 err_unmap_mtt:
 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtt_table);
+	if (hns_roce_check_whether_mhop(hr_dev, HEM_TYPE_CQE))
+		hns_roce_cleanup_hem_table(hr_dev,
+					   &hr_dev->mr_table.mtt_cqe_table);
+
+err_unmap_cqe:
+	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtt_table);
 
 	return ret;
 }
diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c b/drivers/infiniband/hw/hns/hns_roce_mr.c
index d8f7b41..88b436f 100644
--- a/drivers/infiniband/hw/hns/hns_roce_mr.c
+++ b/drivers/infiniband/hw/hns/hns_roce_mr.c
@@ -177,18 +177,28 @@ static void hns_roce_buddy_cleanup(struct hns_roce_buddy *buddy)
 }
 
 static int hns_roce_alloc_mtt_range(struct hns_roce_dev *hr_dev, int order,
-				    unsigned long *seg)
+				    unsigned long *seg, u32 mtt_type)
 {
 	struct hns_roce_mr_table *mr_table = &hr_dev->mr_table;
-	int ret = 0;
+	struct hns_roce_hem_table *table;
+	struct hns_roce_buddy *buddy;
+	int ret;
+
+	if (mtt_type == MTT_TYPE_WQE) {
+		buddy = &mr_table->mtt_buddy;
+		table = &mr_table->mtt_table;
+	} else {
+		buddy = &mr_table->mtt_cqe_buddy;
+		table = &mr_table->mtt_cqe_table;
+	}
 
-	ret = hns_roce_buddy_alloc(&mr_table->mtt_buddy, order, seg);
+	ret = hns_roce_buddy_alloc(buddy, order, seg);
 	if (ret == -1)
 		return -1;
 
-	if (hns_roce_table_get_range(hr_dev, &mr_table->mtt_table, *seg,
+	if (hns_roce_table_get_range(hr_dev, table, *seg,
 				     *seg + (1 << order) - 1)) {
-		hns_roce_buddy_free(&mr_table->mtt_buddy, *seg, order);
+		hns_roce_buddy_free(buddy, *seg, order);
 		return -1;
 	}
 
@@ -198,7 +208,7 @@ static int hns_roce_alloc_mtt_range(struct hns_roce_dev *hr_dev, int order,
 int hns_roce_mtt_init(struct hns_roce_dev *hr_dev, int npages, int page_shift,
 		      struct hns_roce_mtt *mtt)
 {
-	int ret = 0;
+	int ret;
 	int i;
 
 	/* Page num is zero, correspond to DMA memory register */
@@ -217,7 +227,8 @@ int hns_roce_mtt_init(struct hns_roce_dev *hr_dev, int npages, int page_shift,
 		++mtt->order;
 
 	/* Allocate MTT entry */
-	ret = hns_roce_alloc_mtt_range(hr_dev, mtt->order, &mtt->first_seg);
+	ret = hns_roce_alloc_mtt_range(hr_dev, mtt->order, &mtt->first_seg,
+				       mtt->mtt_type);
 	if (ret == -1)
 		return -ENOMEM;
 
@@ -231,9 +242,19 @@ void hns_roce_mtt_cleanup(struct hns_roce_dev *hr_dev, struct hns_roce_mtt *mtt)
 	if (mtt->order < 0)
 		return;
 
-	hns_roce_buddy_free(&mr_table->mtt_buddy, mtt->first_seg, mtt->order);
-	hns_roce_table_put_range(hr_dev, &mr_table->mtt_table, mtt->first_seg,
-				 mtt->first_seg + (1 << mtt->order) - 1);
+	if (mtt->mtt_type == MTT_TYPE_WQE) {
+		hns_roce_buddy_free(&mr_table->mtt_buddy, mtt->first_seg,
+				    mtt->order);
+		hns_roce_table_put_range(hr_dev, &mr_table->mtt_table,
+					mtt->first_seg,
+					mtt->first_seg + (1 << mtt->order) - 1);
+	} else {
+		hns_roce_buddy_free(&mr_table->mtt_cqe_buddy, mtt->first_seg,
+				    mtt->order);
+		hns_roce_table_put_range(hr_dev, &mr_table->mtt_cqe_table,
+					mtt->first_seg,
+					mtt->first_seg + (1 << mtt->order) - 1);
+	}
 }
 EXPORT_SYMBOL_GPL(hns_roce_mtt_cleanup);
 
@@ -362,7 +383,11 @@ static int hns_roce_write_mtt_chunk(struct hns_roce_dev *hr_dev,
 	if (start_index & (HNS_ROCE_MTT_ENTRY_PER_SEG - 1))
 		return -EINVAL;
 
-	table = &hr_dev->mr_table.mtt_table;
+	if (mtt->mtt_type == MTT_TYPE_WQE)
+		table = &hr_dev->mr_table.mtt_table;
+	else
+		table = &hr_dev->mr_table.mtt_cqe_table;
+
 	mtts = hns_roce_table_find(hr_dev, table,
 				mtt->first_seg + s / hr_dev->caps.mtt_entry_sz,
 				&dma_handle);
@@ -409,9 +434,9 @@ static int hns_roce_write_mtt(struct hns_roce_dev *hr_dev,
 int hns_roce_buf_write_mtt(struct hns_roce_dev *hr_dev,
 			   struct hns_roce_mtt *mtt, struct hns_roce_buf *buf)
 {
-	u32 i = 0;
-	int ret = 0;
-	u64 *page_list = NULL;
+	u64 *page_list;
+	int ret;
+	u32 i;
 
 	page_list = kmalloc_array(buf->npages, sizeof(*page_list), GFP_KERNEL);
 	if (!page_list)
@@ -434,7 +459,7 @@ int hns_roce_buf_write_mtt(struct hns_roce_dev *hr_dev,
 int hns_roce_init_mr_table(struct hns_roce_dev *hr_dev)
 {
 	struct hns_roce_mr_table *mr_table = &hr_dev->mr_table;
-	int ret = 0;
+	int ret;
 
 	ret = hns_roce_bitmap_init(&mr_table->mtpt_bitmap,
 				   hr_dev->caps.num_mtpts,
@@ -448,8 +473,17 @@ int hns_roce_init_mr_table(struct hns_roce_dev *hr_dev)
 	if (ret)
 		goto err_buddy;
 
+	if (hns_roce_check_whether_mhop(hr_dev, HEM_TYPE_CQE)) {
+		ret = hns_roce_buddy_init(&mr_table->mtt_cqe_buddy,
+					  ilog2(hr_dev->caps.num_cqe_segs));
+		if (ret)
+			goto err_buddy_cqe;
+	}
 	return 0;
 
+err_buddy_cqe:
+	hns_roce_buddy_cleanup(&mr_table->mtt_buddy);
+
 err_buddy:
 	hns_roce_bitmap_cleanup(&mr_table->mtpt_bitmap);
 	return ret;
@@ -460,6 +494,8 @@ void hns_roce_cleanup_mr_table(struct hns_roce_dev *hr_dev)
 	struct hns_roce_mr_table *mr_table = &hr_dev->mr_table;
 
 	hns_roce_buddy_cleanup(&mr_table->mtt_buddy);
+	if (hns_roce_check_whether_mhop(hr_dev, HEM_TYPE_CQE))
+		hns_roce_buddy_cleanup(&mr_table->mtt_cqe_buddy);
 	hns_roce_bitmap_cleanup(&mr_table->mtpt_bitmap);
 }
 
diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c
index b512ecc..881ea67 100644
--- a/drivers/infiniband/hw/hns/hns_roce_qp.c
+++ b/drivers/infiniband/hw/hns/hns_roce_qp.c
@@ -440,6 +440,7 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev,
 			goto err_out;
 		}
 
+		hr_qp->mtt.mtt_type = MTT_TYPE_WQE;
 		ret = hns_roce_mtt_init(hr_dev, ib_umem_page_count(hr_qp->umem),
 					hr_qp->umem->page_shift, &hr_qp->mtt);
 		if (ret) {
@@ -490,6 +491,7 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev,
 			goto err_out;
 		}
 
+		hr_qp->mtt.mtt_type = MTT_TYPE_WQE;
 		/* Write MTT */
 		ret = hns_roce_mtt_init(hr_dev, hr_qp->hr_buf.npages,
 					hr_qp->hr_buf.page_shift, &hr_qp->mtt);
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 12/20] RDMA/hns: Support multi hop addressing for PBL in hip08

[Wei Hu (Xavier)]

The block base address in the MR can be retrieved by the block number
which is calculated with the VA in the SGE and MTPT. In hip08, the PBL
supports multi hop addressing to retrieve the block base address by
the block number.

This patch is to add the interfaces in the MR to support multi hop
addressing for the PBL.

Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_device.h |  16 ++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  |   3 +
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  |   1 +
 drivers/infiniband/hw/hns/hns_roce_mr.c     | 422 ++++++++++++++++++++++++++--
 4 files changed, 418 insertions(+), 24 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 319c7054..7815e4e 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -267,6 +267,19 @@ struct hns_roce_mr {
 	int			type;	/* MR's register type */
 	u64			*pbl_buf;/* MR's PBL space */
 	dma_addr_t		pbl_dma_addr;	/* MR's PBL space PA */
+	u32			pbl_size;/* PA number in the PBL */
+	u64			pbl_ba;/* page table address */
+	u32			l0_chunk_last_num;/* L0 last number */
+	u32			l1_chunk_last_num;/* L1 last number */
+	u64			**pbl_bt_l2;/* PBL BT L2 */
+	u64			**pbl_bt_l1;/* PBL BT L1 */
+	u64			*pbl_bt_l0;/* PBL BT L0 */
+	dma_addr_t		*pbl_l2_dma_addr;/* PBL BT L2 dma addr */
+	dma_addr_t		*pbl_l1_dma_addr;/* PBL BT L1 dma addr */
+	dma_addr_t		pbl_l0_dma_addr;/* PBL BT L0 dma addr */
+	u32			pbl_ba_pg_sz;/* BT chunk page size */
+	u32			pbl_buf_pg_sz;/* buf chunk page size */
+	u32			pbl_hop_num;/* multi-hop number */
 };
 
 struct hns_roce_mr_table {
@@ -514,6 +527,9 @@ struct hns_roce_caps {
 	int		qpc_entry_sz;
 	int		irrl_entry_sz;
 	int		cqc_entry_sz;
+	u32		pbl_ba_pg_sz;
+	u32		pbl_buf_pg_sz;
+	u32		pbl_hop_num;
 	int		aeqe_depth;
 	int		ceqe_depth[HNS_ROCE_COMP_VEC_NUM];
 	enum ib_mtu	max_mtu;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index c59d15e..542540d 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -613,6 +613,9 @@ static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev)
 	caps->mpt_ba_pg_sz	= 0;
 	caps->mpt_buf_pg_sz	= 0;
 	caps->mpt_hop_num	= HNS_ROCE_CONTEXT_HOP_NUM;
+	caps->pbl_ba_pg_sz	= 0;
+	caps->pbl_buf_pg_sz	= 0;
+	caps->pbl_hop_num	= HNS_ROCE_PBL_HOP_NUM;
 	caps->mtt_ba_pg_sz	= 0;
 	caps->mtt_buf_pg_sz	= 0;
 	caps->mtt_hop_num	= HNS_ROCE_MTT_HOP_NUM;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index 212d511..593d886 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -75,6 +75,7 @@
 #define HNS_ROCE_CONTEXT_HOP_NUM		1
 #define HNS_ROCE_MTT_HOP_NUM			1
 #define HNS_ROCE_CQE_HOP_NUM			1
+#define HNS_ROCE_PBL_HOP_NUM			2
 
 #define HNS_ROCE_CMD_FLAG_IN_VALID_SHIFT	0
 #define HNS_ROCE_CMD_FLAG_OUT_VALID_SHIFT	1
diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c b/drivers/infiniband/hw/hns/hns_roce_mr.c
index 88b436f..7e6ce76 100644
--- a/drivers/infiniband/hw/hns/hns_roce_mr.c
+++ b/drivers/infiniband/hw/hns/hns_roce_mr.c
@@ -258,6 +258,239 @@ void hns_roce_mtt_cleanup(struct hns_roce_dev *hr_dev, struct hns_roce_mtt *mtt)
 }
 EXPORT_SYMBOL_GPL(hns_roce_mtt_cleanup);
 
+static void hns_roce_loop_free(struct hns_roce_dev *hr_dev,
+			       struct hns_roce_mr *mr, int err_loop_index,
+			       int loop_i, int loop_j)
+{
+	struct device *dev = hr_dev->dev;
+	u32 mhop_num;
+	u32 pbl_bt_sz;
+	u64 bt_idx;
+	int i, j;
+
+	pbl_bt_sz = 1 << (hr_dev->caps.pbl_ba_pg_sz + PAGE_SHIFT);
+	mhop_num = hr_dev->caps.pbl_hop_num;
+
+	i = loop_i;
+	j = loop_j;
+	if (mhop_num == 3 && err_loop_index == 2) {
+		for (; i >= 0; i--) {
+			dma_free_coherent(dev, pbl_bt_sz, mr->pbl_bt_l1[i],
+					  mr->pbl_l1_dma_addr[i]);
+
+			for (j = 0; j < pbl_bt_sz / 8; j++) {
+				if (i == loop_i && j >= loop_j)
+					break;
+
+				bt_idx = i * pbl_bt_sz / 8 + j;
+				dma_free_coherent(dev, pbl_bt_sz,
+						  mr->pbl_bt_l2[bt_idx],
+						  mr->pbl_l2_dma_addr[bt_idx]);
+			}
+		}
+	} else if (mhop_num == 3 && err_loop_index == 1) {
+		for (i -= 1; i >= 0; i--) {
+			dma_free_coherent(dev, pbl_bt_sz, mr->pbl_bt_l1[i],
+					  mr->pbl_l1_dma_addr[i]);
+
+			for (j = 0; j < pbl_bt_sz / 8; j++) {
+				bt_idx = i * pbl_bt_sz / 8 + j;
+				dma_free_coherent(dev, pbl_bt_sz,
+						  mr->pbl_bt_l2[bt_idx],
+						  mr->pbl_l2_dma_addr[bt_idx]);
+			}
+		}
+	} else if (mhop_num == 2 && err_loop_index == 1) {
+		for (i -= 1; i >= 0; i--)
+			dma_free_coherent(dev, pbl_bt_sz, mr->pbl_bt_l1[i],
+					  mr->pbl_l1_dma_addr[i]);
+	} else {
+		dev_warn(dev, "not support: mhop_num=%d, err_loop_index=%d.",
+			 mhop_num, err_loop_index);
+		return;
+	}
+
+	dma_free_coherent(dev, pbl_bt_sz, mr->pbl_bt_l0, mr->pbl_l0_dma_addr);
+	mr->pbl_bt_l0 = NULL;
+	mr->pbl_l0_dma_addr = 0;
+}
+
+/* PBL multi hop addressing */
+static int hns_roce_mhop_alloc(struct hns_roce_dev *hr_dev, int npages,
+			       struct hns_roce_mr *mr)
+{
+	struct device *dev = hr_dev->dev;
+	int mr_alloc_done = 0;
+	int npages_allocated;
+	int i = 0, j = 0;
+	u32 pbl_bt_sz;
+	u32 mhop_num;
+	u64 pbl_last_bt_num;
+	u64 pbl_bt_cnt = 0;
+	u64 bt_idx;
+	u64 size;
+
+	mhop_num = hr_dev->caps.pbl_hop_num;
+	pbl_bt_sz = 1 << (hr_dev->caps.pbl_ba_pg_sz + PAGE_SHIFT);
+	pbl_last_bt_num = (npages + pbl_bt_sz / 8 - 1) / (pbl_bt_sz / 8);
+
+	if (mhop_num == HNS_ROCE_HOP_NUM_0)
+		return 0;
+
+	/* hop_num = 1 */
+	if (mhop_num == 1) {
+		if (npages > pbl_bt_sz / 8) {
+			dev_err(dev, "npages %d is larger than buf_pg_sz!",
+				npages);
+			return -EINVAL;
+		}
+		mr->pbl_buf = dma_alloc_coherent(dev, npages * 8,
+						 &(mr->pbl_dma_addr),
+						 GFP_KERNEL);
+		if (!mr->pbl_buf)
+			return -ENOMEM;
+
+		mr->pbl_size = npages;
+		mr->pbl_ba = mr->pbl_dma_addr;
+		mr->pbl_hop_num = hr_dev->caps.pbl_hop_num;
+		mr->pbl_ba_pg_sz = hr_dev->caps.pbl_ba_pg_sz;
+		mr->pbl_buf_pg_sz = hr_dev->caps.pbl_buf_pg_sz;
+		return 0;
+	}
+
+	mr->pbl_l1_dma_addr = kcalloc(pbl_bt_sz / 8,
+				      sizeof(*mr->pbl_l1_dma_addr),
+				      GFP_KERNEL);
+	if (!mr->pbl_l1_dma_addr)
+		return -ENOMEM;
+
+	mr->pbl_bt_l1 = kcalloc(pbl_bt_sz / 8, sizeof(*mr->pbl_bt_l1),
+				GFP_KERNEL);
+	if (!mr->pbl_bt_l1)
+		goto err_kcalloc_bt_l1;
+
+	if (mhop_num == 3) {
+		mr->pbl_l2_dma_addr = kcalloc(pbl_last_bt_num,
+					      sizeof(*mr->pbl_l2_dma_addr),
+					      GFP_KERNEL);
+		if (!mr->pbl_l2_dma_addr)
+			goto err_kcalloc_l2_dma;
+
+		mr->pbl_bt_l2 = kcalloc(pbl_last_bt_num,
+					sizeof(*mr->pbl_bt_l2),
+					GFP_KERNEL);
+		if (!mr->pbl_bt_l2)
+			goto err_kcalloc_bt_l2;
+	}
+
+	/* alloc L0 BT */
+	mr->pbl_bt_l0 = dma_alloc_coherent(dev, pbl_bt_sz,
+					   &(mr->pbl_l0_dma_addr),
+					   GFP_KERNEL);
+	if (!mr->pbl_bt_l0)
+		goto err_dma_alloc_l0;
+
+	if (mhop_num == 2) {
+		/* alloc L1 BT */
+		for (i = 0; i < pbl_bt_sz / 8; i++) {
+			if (pbl_bt_cnt + 1 < pbl_last_bt_num) {
+				size = pbl_bt_sz;
+			} else {
+				npages_allocated = i * (pbl_bt_sz / 8);
+				size = (npages - npages_allocated) * 8;
+			}
+			mr->pbl_bt_l1[i] = dma_alloc_coherent(dev, size,
+						    &(mr->pbl_l1_dma_addr[i]),
+						    GFP_KERNEL);
+			if (!mr->pbl_bt_l1[i]) {
+				hns_roce_loop_free(hr_dev, mr, 1, i, 0);
+				goto err_dma_alloc_l0;
+			}
+
+			*(mr->pbl_bt_l0 + i) = mr->pbl_l1_dma_addr[i];
+
+			pbl_bt_cnt++;
+			if (pbl_bt_cnt >= pbl_last_bt_num)
+				break;
+		}
+	} else if (mhop_num == 3) {
+		/* alloc L1, L2 BT */
+		for (i = 0; i < pbl_bt_sz / 8; i++) {
+			mr->pbl_bt_l1[i] = dma_alloc_coherent(dev, pbl_bt_sz,
+						    &(mr->pbl_l1_dma_addr[i]),
+						    GFP_KERNEL);
+			if (!mr->pbl_bt_l1[i]) {
+				hns_roce_loop_free(hr_dev, mr, 1, i, 0);
+				goto err_dma_alloc_l0;
+			}
+
+			*(mr->pbl_bt_l0 + i) = mr->pbl_l1_dma_addr[i];
+
+			for (j = 0; j < pbl_bt_sz / 8; j++) {
+				bt_idx = i * pbl_bt_sz / 8 + j;
+
+				if (pbl_bt_cnt + 1 < pbl_last_bt_num) {
+					size = pbl_bt_sz;
+				} else {
+					npages_allocated = bt_idx *
+							   (pbl_bt_sz / 8);
+					size = (npages - npages_allocated) * 8;
+				}
+				mr->pbl_bt_l2[bt_idx] = dma_alloc_coherent(
+					      dev, size,
+					      &(mr->pbl_l2_dma_addr[bt_idx]),
+					      GFP_KERNEL);
+				if (!mr->pbl_bt_l2[bt_idx]) {
+					hns_roce_loop_free(hr_dev, mr, 2, i, j);
+					goto err_dma_alloc_l0;
+				}
+
+				*(mr->pbl_bt_l1[i] + j) =
+						mr->pbl_l2_dma_addr[bt_idx];
+
+				pbl_bt_cnt++;
+				if (pbl_bt_cnt >= pbl_last_bt_num) {
+					mr_alloc_done = 1;
+					break;
+				}
+			}
+
+			if (mr_alloc_done)
+				break;
+		}
+	}
+
+	mr->l0_chunk_last_num = i + 1;
+	if (mhop_num == 3)
+		mr->l1_chunk_last_num = j + 1;
+
+	mr->pbl_size = npages;
+	mr->pbl_ba = mr->pbl_l0_dma_addr;
+	mr->pbl_hop_num = hr_dev->caps.pbl_hop_num;
+	mr->pbl_ba_pg_sz = hr_dev->caps.pbl_ba_pg_sz;
+	mr->pbl_buf_pg_sz = hr_dev->caps.pbl_buf_pg_sz;
+
+	return 0;
+
+err_dma_alloc_l0:
+	kfree(mr->pbl_bt_l2);
+	mr->pbl_bt_l2 = NULL;
+
+err_kcalloc_bt_l2:
+	kfree(mr->pbl_l2_dma_addr);
+	mr->pbl_l2_dma_addr = NULL;
+
+err_kcalloc_l2_dma:
+	kfree(mr->pbl_bt_l1);
+	mr->pbl_bt_l1 = NULL;
+
+err_kcalloc_bt_l1:
+	kfree(mr->pbl_l1_dma_addr);
+	mr->pbl_l1_dma_addr = NULL;
+
+	return -ENOMEM;
+}
+
 static int hns_roce_mr_alloc(struct hns_roce_dev *hr_dev, u32 pd, u64 iova,
 			     u64 size, u32 access, int npages,
 			     struct hns_roce_mr *mr)
@@ -282,16 +515,111 @@ static int hns_roce_mr_alloc(struct hns_roce_dev *hr_dev, u32 pd, u64 iova,
 		mr->type = MR_TYPE_DMA;
 		mr->pbl_buf = NULL;
 		mr->pbl_dma_addr = 0;
+		/* PBL multi-hop addressing parameters */
+		mr->pbl_bt_l2 = NULL;
+		mr->pbl_bt_l1 = NULL;
+		mr->pbl_bt_l0 = NULL;
+		mr->pbl_l2_dma_addr = NULL;
+		mr->pbl_l1_dma_addr = NULL;
+		mr->pbl_l0_dma_addr = 0;
 	} else {
 		mr->type = MR_TYPE_MR;
-		mr->pbl_buf = dma_alloc_coherent(dev, npages * 8,
-						 &(mr->pbl_dma_addr),
-						 GFP_KERNEL);
-		if (!mr->pbl_buf)
-			return -ENOMEM;
+		if (!hr_dev->caps.pbl_hop_num) {
+			mr->pbl_buf = dma_alloc_coherent(dev, npages * 8,
+							 &(mr->pbl_dma_addr),
+							 GFP_KERNEL);
+			if (!mr->pbl_buf)
+				return -ENOMEM;
+		} else {
+			ret = hns_roce_mhop_alloc(hr_dev, npages, mr);
+		}
 	}
 
-	return 0;
+	return ret;
+}
+
+static void hns_roce_mhop_free(struct hns_roce_dev *hr_dev,
+			       struct hns_roce_mr *mr)
+{
+	struct device *dev = hr_dev->dev;
+	int npages_allocated;
+	int npages;
+	int i, j;
+	u32 pbl_bt_sz;
+	u32 mhop_num;
+	u64 bt_idx;
+
+	npages = ib_umem_page_count(mr->umem);
+	pbl_bt_sz = 1 << (hr_dev->caps.pbl_ba_pg_sz + PAGE_SHIFT);
+	mhop_num = hr_dev->caps.pbl_hop_num;
+
+	if (mhop_num == HNS_ROCE_HOP_NUM_0)
+		return;
+
+	/* hop_num = 1 */
+	if (mhop_num == 1) {
+		dma_free_coherent(dev, (unsigned int)(npages * 8),
+				  mr->pbl_buf, mr->pbl_dma_addr);
+		return;
+	}
+
+	dma_free_coherent(dev, pbl_bt_sz, mr->pbl_bt_l0,
+			  mr->pbl_l0_dma_addr);
+
+	if (mhop_num == 2) {
+		for (i = 0; i < mr->l0_chunk_last_num; i++) {
+			if (i == mr->l0_chunk_last_num - 1) {
+				npages_allocated = i * (pbl_bt_sz / 8);
+
+				dma_free_coherent(dev,
+					      (npages - npages_allocated) * 8,
+					      mr->pbl_bt_l1[i],
+					      mr->pbl_l1_dma_addr[i]);
+
+				break;
+			}
+
+			dma_free_coherent(dev, pbl_bt_sz, mr->pbl_bt_l1[i],
+					  mr->pbl_l1_dma_addr[i]);
+		}
+	} else if (mhop_num == 3) {
+		for (i = 0; i < mr->l0_chunk_last_num; i++) {
+			dma_free_coherent(dev, pbl_bt_sz, mr->pbl_bt_l1[i],
+					  mr->pbl_l1_dma_addr[i]);
+
+			for (j = 0; j < pbl_bt_sz / 8; j++) {
+				bt_idx = i * (pbl_bt_sz / 8) + j;
+
+				if ((i == mr->l0_chunk_last_num - 1)
+				    && j == mr->l1_chunk_last_num - 1) {
+					npages_allocated = bt_idx *
+							   (pbl_bt_sz / 8);
+
+					dma_free_coherent(dev,
+					      (npages - npages_allocated) * 8,
+					      mr->pbl_bt_l2[bt_idx],
+					      mr->pbl_l2_dma_addr[bt_idx]);
+
+					break;
+				}
+
+				dma_free_coherent(dev, pbl_bt_sz,
+						mr->pbl_bt_l2[bt_idx],
+						mr->pbl_l2_dma_addr[bt_idx]);
+			}
+		}
+	}
+
+	kfree(mr->pbl_bt_l1);
+	kfree(mr->pbl_l1_dma_addr);
+	mr->pbl_bt_l1 = NULL;
+	mr->pbl_l1_dma_addr = NULL;
+	if (mhop_num == 3) {
+		kfree(mr->pbl_bt_l2);
+		kfree(mr->pbl_l2_dma_addr);
+		mr->pbl_bt_l2 = NULL;
+		mr->pbl_l2_dma_addr = NULL;
+	}
 }
 
 static void hns_roce_mr_free(struct hns_roce_dev *hr_dev,
@@ -310,10 +638,18 @@ static void hns_roce_mr_free(struct hns_roce_dev *hr_dev,
 
 	if (mr->size != ~0ULL) {
 		npages = ib_umem_page_count(mr->umem);
-		dma_free_coherent(dev, (unsigned int)(npages * 8), mr->pbl_buf,
-				  mr->pbl_dma_addr);
+
+		if (!hr_dev->caps.pbl_hop_num)
+			dma_free_coherent(dev, (unsigned int)(npages * 8),
+					  mr->pbl_buf, mr->pbl_dma_addr);
+		else
+			hns_roce_mhop_free(hr_dev, mr);
 	}
 
+	if (mr->enabled)
+		hns_roce_table_put(hr_dev, &hr_dev->mr_table.mtpt_table,
+				   key_to_hw_index(mr->key));
+
 	hns_roce_bitmap_free(&hr_dev->mr_table.mtpt_bitmap,
 			     key_to_hw_index(mr->key), BITMAP_NO_RR);
 }
@@ -501,8 +837,8 @@ void hns_roce_cleanup_mr_table(struct hns_roce_dev *hr_dev)
 
 struct ib_mr *hns_roce_get_dma_mr(struct ib_pd *pd, int acc)
 {
-	int ret = 0;
-	struct hns_roce_mr *mr = NULL;
+	struct hns_roce_mr *mr;
+	int ret;
 
 	mr = kmalloc(sizeof(*mr), GFP_KERNEL);
 	if (mr == NULL)
@@ -571,16 +907,36 @@ int hns_roce_ib_umem_write_mtt(struct hns_roce_dev *hr_dev,
 	return ret;
 }
 
-static int hns_roce_ib_umem_write_mr(struct hns_roce_mr *mr,
+static int hns_roce_ib_umem_write_mr(struct hns_roce_dev *hr_dev,
+				     struct hns_roce_mr *mr,
 				     struct ib_umem *umem)
 {
-	int i = 0;
-	int entry;
 	struct scatterlist *sg;
+	int i = 0, j = 0;
+	int entry;
+
+	if (hr_dev->caps.pbl_hop_num == HNS_ROCE_HOP_NUM_0)
+		return 0;
 
 	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
-		mr->pbl_buf[i] = ((u64)sg_dma_address(sg)) >> 12;
-		i++;
+		if (!hr_dev->caps.pbl_hop_num) {
+			mr->pbl_buf[i] = ((u64)sg_dma_address(sg)) >> 12;
+			i++;
+		} else if (hr_dev->caps.pbl_hop_num == 1) {
+			mr->pbl_buf[i] = sg_dma_address(sg);
+			i++;
+		} else {
+			if (hr_dev->caps.pbl_hop_num == 2)
+				mr->pbl_bt_l1[i][j] = sg_dma_address(sg);
+			else if (hr_dev->caps.pbl_hop_num == 3)
+				mr->pbl_bt_l2[i][j] = sg_dma_address(sg);
+
+			j++;
+			if (j >= (PAGE_SIZE / 8)) {
+				i++;
+				j = 0;
+			}
+		}
 	}
 
 	/* Memory barrier */
@@ -595,9 +951,11 @@ struct ib_mr *hns_roce_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 {
 	struct hns_roce_dev *hr_dev = to_hr_dev(pd->device);
 	struct device *dev = hr_dev->dev;
-	struct hns_roce_mr *mr = NULL;
-	int ret = 0;
-	int n = 0;
+	struct hns_roce_mr *mr;
+	int bt_size;
+	int ret;
+	int n;
+	int i;
 
 	mr = kmalloc(sizeof(*mr), GFP_KERNEL);
 	if (!mr)
@@ -618,11 +976,27 @@ struct ib_mr *hns_roce_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 		goto err_umem;
 	}
 
-	if (n > HNS_ROCE_MAX_MTPT_PBL_NUM) {
-		dev_err(dev, " MR len %lld err. MR is limited to 4G at most!\n",
-			length);
-		ret = -EINVAL;
-		goto err_umem;
+	if (!hr_dev->caps.pbl_hop_num) {
+		if (n > HNS_ROCE_MAX_MTPT_PBL_NUM) {
+			dev_err(dev,
+			     " MR len %lld err. MR is limited to 4G at most!\n",
+			     length);
+			ret = -EINVAL;
+			goto err_umem;
+		}
+	} else {
+		int pbl_size = 1;
+
+		bt_size = (1 << PAGE_SHIFT) / 8;
+		for (i = 0; i < hr_dev->caps.pbl_hop_num; i++)
+			pbl_size *= bt_size;
+		if (n > pbl_size) {
+			dev_err(dev,
+			    " MR len %lld err. MR page num is limited to %d!\n",
+			    length, pbl_size);
+			ret = -EINVAL;
+			goto err_umem;
+		}
 	}
 
 	ret = hns_roce_mr_alloc(hr_dev, to_hr_pd(pd)->pdn, virt_addr, length,
@@ -630,7 +1004,7 @@ struct ib_mr *hns_roce_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	if (ret)
 		goto err_umem;
 
-	ret = hns_roce_ib_umem_write_mr(mr, mr->umem);
+	ret = hns_roce_ib_umem_write_mr(hr_dev, mr, mr->umem);
 	if (ret)
 		goto err_mr;
 
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 13/20] RDMA/hns: Configure mac&gid and user access region for hip08 RoCE driver

[Wei Hu (Xavier)]

In hip08, the user access region(UAR) pfn is calculated
from pci device memory resource.

This patch mainly sets mac and gid table by configuring
the relevant registers and updates the uar pfn for hip08 SoC.

Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_common.h |  9 ++++++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 49 +++++++++++++++++++++++++++++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  |  7 +++++
 drivers/infiniband/hw/hns/hns_roce_pd.c     | 16 +++++++---
 4 files changed, 76 insertions(+), 5 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_common.h b/drivers/infiniband/hw/hns/hns_roce_common.h
index 0c950f8..7ecb7a4 100644
--- a/drivers/infiniband/hw/hns/hns_roce_common.h
+++ b/drivers/infiniband/hw/hns/hns_roce_common.h
@@ -376,4 +376,13 @@
 #define ROCEE_RX_CMQ_TAIL_REG			0x07024
 #define ROCEE_RX_CMQ_HEAD_REG			0x07028
 
+#define ROCEE_VF_SMAC_CFG0_REG			0x12000
+#define ROCEE_VF_SMAC_CFG1_REG			0x12004
+
+#define ROCEE_VF_SGID_CFG0_REG			0x10000
+#define ROCEE_VF_SGID_CFG1_REG			0x10004
+#define ROCEE_VF_SGID_CFG2_REG			0x10008
+#define ROCEE_VF_SGID_CFG3_REG			0x1000c
+#define ROCEE_VF_SGID_CFG4_REG			0x10010
+
 #endif /* _HNS_ROCE_COMMON_H */
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 542540d..e7aeb53 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -718,6 +718,53 @@ static int hns_roce_v2_chk_mbox(struct hns_roce_dev *hr_dev,
 	return 0;
 }
 
+static void hns_roce_v2_set_gid(struct hns_roce_dev *hr_dev, u8 port,
+				int gid_index, union ib_gid *gid)
+{
+	u32 *p;
+	u32 val;
+
+	p = (u32 *)&gid->raw[0];
+	roce_raw_write(*p, hr_dev->reg_base + ROCEE_VF_SGID_CFG0_REG +
+		       0x20 * gid_index);
+
+	p = (u32 *)&gid->raw[4];
+	roce_raw_write(*p, hr_dev->reg_base + ROCEE_VF_SGID_CFG1_REG +
+		       0x20 * gid_index);
+
+	p = (u32 *)&gid->raw[8];
+	roce_raw_write(*p, hr_dev->reg_base + ROCEE_VF_SGID_CFG2_REG +
+		       0x20 * gid_index);
+
+	p = (u32 *)&gid->raw[0xc];
+	roce_raw_write(*p, hr_dev->reg_base + ROCEE_VF_SGID_CFG3_REG +
+		       0x20 * gid_index);
+
+	val = roce_read(hr_dev, ROCEE_VF_SGID_CFG4_REG + 0x20 * gid_index);
+	roce_set_field(val, ROCEE_VF_SGID_CFG4_SGID_TYPE_M,
+		       ROCEE_VF_SGID_CFG4_SGID_TYPE_S, 0);
+
+	roce_write(hr_dev, ROCEE_VF_SGID_CFG4_REG + 0x20 * gid_index, val);
+}
+
+static void hns_roce_v2_set_mac(struct hns_roce_dev *hr_dev, u8 phy_port,
+				u8 *addr)
+{
+	u16 reg_smac_h;
+	u32 reg_smac_l;
+	u32 val;
+
+	reg_smac_l = *(u32 *)(&addr[0]);
+	roce_raw_write(reg_smac_l, hr_dev->reg_base + ROCEE_VF_SMAC_CFG0_REG +
+		       0x08 * phy_port);
+	val = roce_read(hr_dev, ROCEE_VF_SMAC_CFG1_REG + 0x08 * phy_port);
+
+	reg_smac_h  = *(u16 *)(&addr[4]);
+	roce_set_field(val, ROCEE_VF_SMAC_CFG1_VF_SMAC_H_M,
+		       ROCEE_VF_SMAC_CFG1_VF_SMAC_H_S, reg_smac_h);
+	roce_write(hr_dev, ROCEE_VF_SMAC_CFG1_REG + 0x08 * phy_port, val);
+}
+
 static int hns_roce_v2_set_hem(struct hns_roce_dev *hr_dev,
 			       struct hns_roce_hem_table *table, int obj,
 			       int step_idx)
@@ -857,6 +904,8 @@ static int hns_roce_v2_clear_hem(struct hns_roce_dev *hr_dev,
 	.hw_profile = hns_roce_v2_profile,
 	.post_mbox = hns_roce_v2_post_mbox,
 	.chk_mbox = hns_roce_v2_chk_mbox,
+	.set_gid = hns_roce_v2_set_gid,
+	.set_mac = hns_roce_v2_set_mac,
 	.set_hem = hns_roce_v2_set_hem,
 	.clear_hem = hns_roce_v2_clear_hem,
 };
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index 593d886..aee8f34 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -246,6 +246,13 @@ struct hns_roce_vf_res_b {
 #define VF_RES_B_DATA_3_VF_SL_NUM_S 16
 #define VF_RES_B_DATA_3_VF_SL_NUM_M GENMASK(19, 16)
 
+/* Reg field definition */
+#define ROCEE_VF_SMAC_CFG1_VF_SMAC_H_S 0
+#define ROCEE_VF_SMAC_CFG1_VF_SMAC_H_M GENMASK(15, 0)
+
+#define ROCEE_VF_SGID_CFG4_SGID_TYPE_S 0
+#define ROCEE_VF_SGID_CFG4_SGID_TYPE_M GENMASK(1, 0)
+
 struct hns_roce_cfg_bt_attr {
 	u32 vf_qpc_cfg;
 	u32 vf_srqc_cfg;
diff --git a/drivers/infiniband/hw/hns/hns_roce_pd.c b/drivers/infiniband/hw/hns/hns_roce_pd.c
index 079bb10..bdab218 100644
--- a/drivers/infiniband/hw/hns/hns_roce_pd.c
+++ b/drivers/infiniband/hw/hns/hns_roce_pd.c
@@ -31,6 +31,7 @@
  */
 
 #include <linux/platform_device.h>
+#include <linux/pci.h>
 #include "hns_roce_device.h"
 
 static int hns_roce_pd_alloc(struct hns_roce_dev *hr_dev, unsigned long *pdn)
@@ -111,12 +112,17 @@ int hns_roce_uar_alloc(struct hns_roce_dev *hr_dev, struct hns_roce_uar *uar)
 		uar->index = (uar->index - 1) %
 			     (hr_dev->caps.phy_num_uars - 1) + 1;
 
-	res = platform_get_resource(hr_dev->pdev, IORESOURCE_MEM, 0);
-	if (!res) {
-		dev_err(&hr_dev->pdev->dev, "memory resource not found!\n");
-		return -EINVAL;
+	if (!dev_is_pci(hr_dev->dev)) {
+		res = platform_get_resource(hr_dev->pdev, IORESOURCE_MEM, 0);
+		if (!res) {
+			dev_err(&hr_dev->pdev->dev, "memory resource not found!\n");
+			return -EINVAL;
+		}
+		uar->pfn = ((res->start) >> PAGE_SHIFT) + uar->index;
+	} else {
+		uar->pfn = ((pci_resource_start(hr_dev->pci_dev, 2))
+			   >> PAGE_SHIFT);
 	}
-	uar->pfn = ((res->start) >> PAGE_SHIFT) + uar->index;
 
 	return 0;
 }
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 14/20] RDMA/hns: Add CQ operations support for hip08 RoCE driver

[Wei Hu (Xavier)]

This patch adds CQ relevant operations for hip08 RoCE driver,
such as create CQ, destroy CQ, poll CQ and Request Completion
Notification(req_notify_cq).

Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_cq.c     |  28 +--
 drivers/infiniband/hw/hns/hns_roce_device.h |   1 +
 drivers/infiniband/hw/hns/hns_roce_hw_v1.c  |   1 +
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 376 ++++++++++++++++++++++++++++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  | 222 ++++++++++++++++
 5 files changed, 614 insertions(+), 14 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_cq.c b/drivers/infiniband/hw/hns/hns_roce_cq.c
index 9cf213c..fa17dbc 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cq.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cq.c
@@ -195,14 +195,15 @@ void hns_roce_free_cq(struct hns_roce_dev *hr_dev, struct hns_roce_cq *hr_cq)
 	if (ret)
 		dev_err(dev, "HW2SW_CQ failed (%d) for CQN %06lx\n", ret,
 			hr_cq->cqn);
-
-	/* Waiting interrupt process procedure carried out */
-	synchronize_irq(hr_dev->eq_table.eq[hr_cq->vector].irq);
-
-	/* wait for all interrupt processed */
-	if (atomic_dec_and_test(&hr_cq->refcount))
-		complete(&hr_cq->free);
-	wait_for_completion(&hr_cq->free);
+	if (hr_dev->eq_table.eq) {
+		/* Waiting interrupt process procedure carried out */
+		synchronize_irq(hr_dev->eq_table.eq[hr_cq->vector].irq);
+
+		/* wait for all interrupt processed */
+		if (atomic_dec_and_test(&hr_cq->refcount))
+			complete(&hr_cq->free);
+		wait_for_completion(&hr_cq->free);
+	}
 
 	spin_lock_irq(&cq_table->lock);
 	radix_tree_delete(&cq_table->tree, hr_cq->cqn);
@@ -303,7 +304,7 @@ struct ib_cq *hns_roce_ib_create_cq(struct ib_device *ib_dev,
 	struct hns_roce_uar *uar = NULL;
 	int vector = attr->comp_vector;
 	int cq_entries = attr->cqe;
-	int ret = 0;
+	int ret;
 
 	if (cq_entries < 1 || cq_entries > hr_dev->caps.max_cqes) {
 		dev_err(dev, "Creat CQ failed. entries=%d, max=%d\n",
@@ -311,13 +312,12 @@ struct ib_cq *hns_roce_ib_create_cq(struct ib_device *ib_dev,
 		return ERR_PTR(-EINVAL);
 	}
 
-	hr_cq = kmalloc(sizeof(*hr_cq), GFP_KERNEL);
+	hr_cq = kzalloc(sizeof(*hr_cq), GFP_KERNEL);
 	if (!hr_cq)
 		return ERR_PTR(-ENOMEM);
 
-	/* In v1 engine, parameter verification */
-	if (cq_entries < HNS_ROCE_MIN_CQE_NUM)
-		cq_entries = HNS_ROCE_MIN_CQE_NUM;
+	if (hr_dev->caps.min_cqes)
+		cq_entries = max(cq_entries, hr_dev->caps.min_cqes);
 
 	cq_entries = roundup_pow_of_two((unsigned int)cq_entries);
 	hr_cq->ib_cq.cqe = cq_entries - 1;
@@ -369,7 +369,7 @@ struct ib_cq *hns_roce_ib_create_cq(struct ib_device *ib_dev,
 	 * problems if tptr is set to zero here, so we initialze it in user
 	 * space.
 	 */
-	if (!context)
+	if (!context && hr_cq->tptr_addr)
 		*hr_cq->tptr_addr = 0;
 
 	/* Get created cq handler and carry out event */
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 7815e4e..bee2088 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -508,6 +508,7 @@ struct hns_roce_caps {
 	int		max_qp_dest_rdma;
 	int		num_cqs;
 	int		max_cqes;
+	int		min_cqes;
 	int		reserved_cqs;
 	int		num_aeq_vectors;	/* 1 */
 	int		num_comp_vectors;	/* 32 ceq */
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 0a634c5..1aa2be6 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -1476,6 +1476,7 @@ int hns_roce_v1_profile(struct hns_roce_dev *hr_dev)
 	caps->num_qps		= HNS_ROCE_V1_MAX_QP_NUM;
 	caps->max_wqes		= HNS_ROCE_V1_MAX_WQE_NUM;
 	caps->num_cqs		= HNS_ROCE_V1_MAX_CQ_NUM;
+	caps->min_cqes		= HNS_ROCE_MIN_CQE_NUM;
 	caps->max_cqes		= HNS_ROCE_V1_MAX_CQE_NUM;
 	caps->max_sq_sg		= HNS_ROCE_V1_SG_NUM;
 	caps->max_rq_sg		= HNS_ROCE_V1_SG_NUM;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index e7aeb53..94339af 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -765,6 +765,379 @@ static void hns_roce_v2_set_mac(struct hns_roce_dev *hr_dev, u8 phy_port,
 	roce_write(hr_dev, ROCEE_VF_SMAC_CFG1_REG + 0x08 * phy_port, val);
 }
 
+static void *get_cqe_v2(struct hns_roce_cq *hr_cq, int n)
+{
+	return hns_roce_buf_offset(&hr_cq->hr_buf.hr_buf,
+				   n * HNS_ROCE_V2_CQE_ENTRY_SIZE);
+}
+
+static void *get_sw_cqe_v2(struct hns_roce_cq *hr_cq, int n)
+{
+	struct hns_roce_v2_cqe *cqe = get_cqe_v2(hr_cq, n & hr_cq->ib_cq.cqe);
+
+	/* Get cqe when Owner bit is Conversely with the MSB of cons_idx */
+	return (roce_get_bit(cqe->byte_4, V2_CQE_BYTE_4_OWNER_S) ^
+		!!(n & (hr_cq->ib_cq.cqe + 1))) ? cqe : NULL;
+}
+
+static struct hns_roce_v2_cqe *next_cqe_sw_v2(struct hns_roce_cq *hr_cq)
+{
+	return get_sw_cqe_v2(hr_cq, hr_cq->cons_index);
+}
+
+static void hns_roce_v2_cq_set_ci(struct hns_roce_cq *hr_cq, u32 cons_index)
+{
+	struct hns_roce_v2_cq_db cq_db;
+
+	cq_db.byte_4 = 0;
+	cq_db.parameter = 0;
+
+	roce_set_field(cq_db.byte_4, V2_CQ_DB_BYTE_4_TAG_M,
+		       V2_CQ_DB_BYTE_4_TAG_S, hr_cq->cqn);
+	roce_set_field(cq_db.byte_4, V2_CQ_DB_BYTE_4_CMD_M,
+		       V2_CQ_DB_BYTE_4_CMD_S, HNS_ROCE_V2_CQ_DB_PTR);
+
+	roce_set_field(cq_db.parameter, V2_CQ_DB_PARAMETER_CONS_IDX_M,
+		       V2_CQ_DB_PARAMETER_CONS_IDX_S,
+		       cons_index & ((hr_cq->cq_depth << 1) - 1));
+	roce_set_field(cq_db.parameter, V2_CQ_DB_PARAMETER_CMD_SN_M,
+		       V2_CQ_DB_PARAMETER_CMD_SN_S, 1);
+
+	hns_roce_write64_k((__be32 *)&cq_db, hr_cq->cq_db_l);
+
+}
+
+static void hns_roce_v2_write_cqc(struct hns_roce_dev *hr_dev,
+				  struct hns_roce_cq *hr_cq, void *mb_buf,
+				  u64 *mtts, dma_addr_t dma_handle, int nent,
+				  u32 vector)
+{
+	struct hns_roce_v2_cq_context *cq_context;
+
+	cq_context = mb_buf;
+	memset(cq_context, 0, sizeof(*cq_context));
+
+	roce_set_field(cq_context->byte_4_pg_ceqn, V2_CQC_BYTE_4_CQ_ST_M,
+		       V2_CQC_BYTE_4_CQ_ST_S, V2_CQ_STATE_VALID);
+	roce_set_field(cq_context->byte_4_pg_ceqn, V2_CQC_BYTE_4_SHIFT_M,
+		       V2_CQC_BYTE_4_SHIFT_S, ilog2((unsigned int)nent));
+	roce_set_field(cq_context->byte_4_pg_ceqn, V2_CQC_BYTE_4_CEQN_M,
+		       V2_CQC_BYTE_4_CEQN_S, vector);
+	cq_context->byte_4_pg_ceqn = cpu_to_le32(cq_context->byte_4_pg_ceqn);
+
+	roce_set_field(cq_context->byte_8_cqn, V2_CQC_BYTE_8_CQN_M,
+		       V2_CQC_BYTE_8_CQN_S, hr_cq->cqn);
+
+	cq_context->cqe_cur_blk_addr = (u32)(mtts[0] >> PAGE_ADDR_SHIFT);
+	cq_context->cqe_cur_blk_addr =
+				cpu_to_le32(cq_context->cqe_cur_blk_addr);
+
+	roce_set_field(cq_context->byte_16_hop_addr,
+		       V2_CQC_BYTE_16_CQE_CUR_BLK_ADDR_M,
+		       V2_CQC_BYTE_16_CQE_CUR_BLK_ADDR_S,
+		       cpu_to_le32((mtts[0]) >> (32 + PAGE_ADDR_SHIFT)));
+	roce_set_field(cq_context->byte_16_hop_addr,
+		       V2_CQC_BYTE_16_CQE_HOP_NUM_M,
+		       V2_CQC_BYTE_16_CQE_HOP_NUM_S, hr_dev->caps.cqe_hop_num ==
+		       HNS_ROCE_HOP_NUM_0 ? 0 : hr_dev->caps.cqe_hop_num);
+
+	cq_context->cqe_nxt_blk_addr = (u32)(mtts[1] >> PAGE_ADDR_SHIFT);
+	roce_set_field(cq_context->byte_24_pgsz_addr,
+		       V2_CQC_BYTE_24_CQE_NXT_BLK_ADDR_M,
+		       V2_CQC_BYTE_24_CQE_NXT_BLK_ADDR_S,
+		       cpu_to_le32((mtts[1]) >> (32 + PAGE_ADDR_SHIFT)));
+	roce_set_field(cq_context->byte_24_pgsz_addr,
+		       V2_CQC_BYTE_24_CQE_BA_PG_SZ_M,
+		       V2_CQC_BYTE_24_CQE_BA_PG_SZ_S,
+		       hr_dev->caps.cqe_ba_pg_sz);
+	roce_set_field(cq_context->byte_24_pgsz_addr,
+		       V2_CQC_BYTE_24_CQE_BUF_PG_SZ_M,
+		       V2_CQC_BYTE_24_CQE_BUF_PG_SZ_S,
+		       hr_dev->caps.cqe_buf_pg_sz);
+
+	cq_context->cqe_ba = (u32)(dma_handle >> 3);
+
+	roce_set_field(cq_context->byte_40_cqe_ba, V2_CQC_BYTE_40_CQE_BA_M,
+		       V2_CQC_BYTE_40_CQE_BA_S, (dma_handle >> (32 + 3)));
+}
+
+static int hns_roce_v2_req_notify_cq(struct ib_cq *ibcq,
+				     enum ib_cq_notify_flags flags)
+{
+	struct hns_roce_cq *hr_cq = to_hr_cq(ibcq);
+	u32 notification_flag;
+	u32 doorbell[2];
+
+	doorbell[0] = 0;
+	doorbell[1] = 0;
+
+	notification_flag = (flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED ?
+			     V2_CQ_DB_REQ_NOT : V2_CQ_DB_REQ_NOT_SOL;
+	/*
+	 * flags = 0; Notification Flag = 1, next
+	 * flags = 1; Notification Flag = 0, solocited
+	 */
+	roce_set_field(doorbell[0], V2_CQ_DB_BYTE_4_TAG_M, V2_DB_BYTE_4_TAG_S,
+		       hr_cq->cqn);
+	roce_set_field(doorbell[0], V2_CQ_DB_BYTE_4_CMD_M, V2_DB_BYTE_4_CMD_S,
+		       HNS_ROCE_V2_CQ_DB_NTR);
+	roce_set_field(doorbell[1], V2_CQ_DB_PARAMETER_CONS_IDX_M,
+		       V2_CQ_DB_PARAMETER_CONS_IDX_S,
+		       hr_cq->cons_index & ((hr_cq->cq_depth << 1) - 1));
+	roce_set_field(doorbell[1], V2_CQ_DB_PARAMETER_CMD_SN_M,
+		       V2_CQ_DB_PARAMETER_CMD_SN_S, 1);
+	roce_set_bit(doorbell[1], V2_CQ_DB_PARAMETER_NOTIFY_S,
+		     notification_flag);
+
+	hns_roce_write64_k(doorbell, hr_cq->cq_db_l);
+
+	return 0;
+}
+
+static int hns_roce_v2_poll_one(struct hns_roce_cq *hr_cq,
+				struct hns_roce_qp **cur_qp, struct ib_wc *wc)
+{
+	struct hns_roce_dev *hr_dev;
+	struct hns_roce_v2_cqe *cqe;
+	struct hns_roce_qp *hr_qp;
+	struct hns_roce_wq *wq;
+	int is_send;
+	u16 wqe_ctr;
+	u32 opcode;
+	u32 status;
+	int qpn;
+
+	/* Find cqe according to consumer index */
+	cqe = next_cqe_sw_v2(hr_cq);
+	if (!cqe)
+		return -EAGAIN;
+
+	++hr_cq->cons_index;
+	/* Memory barrier */
+	rmb();
+
+	/* 0->SQ, 1->RQ */
+	is_send = !roce_get_bit(cqe->byte_4, V2_CQE_BYTE_4_S_R_S);
+
+	qpn = roce_get_field(cqe->byte_16, V2_CQE_BYTE_16_LCL_QPN_M,
+				V2_CQE_BYTE_16_LCL_QPN_S);
+
+	if (!*cur_qp || (qpn & HNS_ROCE_V2_CQE_QPN_MASK) != (*cur_qp)->qpn) {
+		hr_dev = to_hr_dev(hr_cq->ib_cq.device);
+		hr_qp = __hns_roce_qp_lookup(hr_dev, qpn);
+		if (unlikely(!hr_qp)) {
+			dev_err(hr_dev->dev, "CQ %06lx with entry for unknown QPN %06x\n",
+				hr_cq->cqn, (qpn & HNS_ROCE_V2_CQE_QPN_MASK));
+			return -EINVAL;
+		}
+		*cur_qp = hr_qp;
+	}
+
+	wc->qp = &(*cur_qp)->ibqp;
+	wc->vendor_err = 0;
+
+	status = roce_get_field(cqe->byte_4, V2_CQE_BYTE_4_STATUS_M,
+				V2_CQE_BYTE_4_STATUS_S);
+	switch (status & HNS_ROCE_V2_CQE_STATUS_MASK) {
+	case HNS_ROCE_CQE_V2_SUCCESS:
+		wc->status = IB_WC_SUCCESS;
+		break;
+	case HNS_ROCE_CQE_V2_LOCAL_LENGTH_ERR:
+		wc->status = IB_WC_LOC_LEN_ERR;
+		break;
+	case HNS_ROCE_CQE_V2_LOCAL_QP_OP_ERR:
+		wc->status = IB_WC_LOC_QP_OP_ERR;
+		break;
+	case HNS_ROCE_CQE_V2_LOCAL_PROT_ERR:
+		wc->status = IB_WC_LOC_PROT_ERR;
+		break;
+	case HNS_ROCE_CQE_V2_WR_FLUSH_ERR:
+		wc->status = IB_WC_WR_FLUSH_ERR;
+		break;
+	case HNS_ROCE_CQE_V2_MW_BIND_ERR:
+		wc->status = IB_WC_MW_BIND_ERR;
+		break;
+	case HNS_ROCE_CQE_V2_BAD_RESP_ERR:
+		wc->status = IB_WC_BAD_RESP_ERR;
+		break;
+	case HNS_ROCE_CQE_V2_LOCAL_ACCESS_ERR:
+		wc->status = IB_WC_LOC_ACCESS_ERR;
+		break;
+	case HNS_ROCE_CQE_V2_REMOTE_INVAL_REQ_ERR:
+		wc->status = IB_WC_REM_INV_REQ_ERR;
+		break;
+	case HNS_ROCE_CQE_V2_REMOTE_ACCESS_ERR:
+		wc->status = IB_WC_REM_ACCESS_ERR;
+		break;
+	case HNS_ROCE_CQE_V2_REMOTE_OP_ERR:
+		wc->status = IB_WC_REM_OP_ERR;
+		break;
+	case HNS_ROCE_CQE_V2_TRANSPORT_RETRY_EXC_ERR:
+		wc->status = IB_WC_RETRY_EXC_ERR;
+		break;
+	case HNS_ROCE_CQE_V2_RNR_RETRY_EXC_ERR:
+		wc->status = IB_WC_RNR_RETRY_EXC_ERR;
+		break;
+	case HNS_ROCE_CQE_V2_REMOTE_ABORT_ERR:
+		wc->status = IB_WC_REM_ABORT_ERR;
+		break;
+	default:
+		wc->status = IB_WC_GENERAL_ERR;
+		break;
+	}
+
+	/* CQE status error, directly return */
+	if (wc->status != IB_WC_SUCCESS)
+		return 0;
+
+	if (is_send) {
+		wc->wc_flags = 0;
+		/* SQ corresponding to CQE */
+		switch (roce_get_field(cqe->byte_4, V2_CQE_BYTE_4_OPCODE_M,
+				       V2_CQE_BYTE_4_OPCODE_S) & 0x1f) {
+		case HNS_ROCE_SQ_OPCODE_SEND:
+			wc->opcode = IB_WC_SEND;
+			break;
+		case HNS_ROCE_SQ_OPCODE_SEND_WITH_INV:
+			wc->opcode = IB_WC_SEND;
+			break;
+		case HNS_ROCE_SQ_OPCODE_SEND_WITH_IMM:
+			wc->opcode = IB_WC_SEND;
+			wc->wc_flags |= IB_WC_WITH_IMM;
+			break;
+		case HNS_ROCE_SQ_OPCODE_RDMA_READ:
+			wc->opcode = IB_WC_RDMA_READ;
+			wc->byte_len = le32_to_cpu(cqe->byte_cnt);
+			break;
+		case HNS_ROCE_SQ_OPCODE_RDMA_WRITE:
+			wc->opcode = IB_WC_RDMA_WRITE;
+			break;
+		case HNS_ROCE_SQ_OPCODE_RDMA_WRITE_WITH_IMM:
+			wc->opcode = IB_WC_RDMA_WRITE;
+			wc->wc_flags |= IB_WC_WITH_IMM;
+			break;
+		case HNS_ROCE_SQ_OPCODE_LOCAL_INV:
+			wc->opcode = IB_WC_LOCAL_INV;
+			wc->wc_flags |= IB_WC_WITH_INVALIDATE;
+			break;
+		case HNS_ROCE_SQ_OPCODE_ATOMIC_COMP_AND_SWAP:
+			wc->opcode = IB_WC_COMP_SWAP;
+			wc->byte_len  = 8;
+			break;
+		case HNS_ROCE_SQ_OPCODE_ATOMIC_FETCH_AND_ADD:
+			wc->opcode = IB_WC_FETCH_ADD;
+			wc->byte_len  = 8;
+			break;
+		case HNS_ROCE_SQ_OPCODE_ATOMIC_MASK_COMP_AND_SWAP:
+			wc->opcode = IB_WC_MASKED_COMP_SWAP;
+			wc->byte_len  = 8;
+			break;
+		case HNS_ROCE_SQ_OPCODE_ATOMIC_MASK_FETCH_AND_ADD:
+			wc->opcode = IB_WC_MASKED_FETCH_ADD;
+			wc->byte_len  = 8;
+			break;
+		case HNS_ROCE_SQ_OPCODE_FAST_REG_WR:
+			wc->opcode = IB_WC_REG_MR;
+			break;
+		case HNS_ROCE_SQ_OPCODE_BIND_MW:
+			wc->opcode = IB_WC_REG_MR;
+			break;
+		default:
+			wc->status = IB_WC_GENERAL_ERR;
+			break;
+		}
+
+		wq = &(*cur_qp)->sq;
+		if ((*cur_qp)->sq_signal_bits) {
+			/*
+			 * If sg_signal_bit is 1,
+			 * firstly tail pointer updated to wqe
+			 * which current cqe correspond to
+			 */
+			wqe_ctr = (u16)roce_get_field(cqe->byte_4,
+						      V2_CQE_BYTE_4_WQE_INDX_M,
+						      V2_CQE_BYTE_4_WQE_INDX_S);
+			wq->tail += (wqe_ctr - (u16)wq->tail) &
+				    (wq->wqe_cnt - 1);
+		}
+
+		wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
+		++wq->tail;
+	} else {
+		/* RQ correspond to CQE */
+		wc->byte_len = le32_to_cpu(cqe->byte_cnt);
+
+		opcode = roce_get_field(cqe->byte_4, V2_CQE_BYTE_4_OPCODE_M,
+					V2_CQE_BYTE_4_OPCODE_S);
+		switch (opcode & 0x1f) {
+		case HNS_ROCE_V2_OPCODE_RDMA_WRITE_IMM:
+			wc->opcode = IB_WC_RECV_RDMA_WITH_IMM;
+			wc->wc_flags = IB_WC_WITH_IMM;
+			wc->ex.imm_data = le32_to_cpu(cqe->rkey_immtdata);
+			break;
+		case HNS_ROCE_V2_OPCODE_SEND:
+			wc->opcode = IB_WC_RECV;
+			wc->wc_flags = 0;
+			break;
+		case HNS_ROCE_V2_OPCODE_SEND_WITH_IMM:
+			wc->opcode = IB_WC_RECV;
+			wc->wc_flags = IB_WC_WITH_IMM;
+			wc->ex.imm_data = le32_to_cpu(cqe->rkey_immtdata);
+			break;
+		case HNS_ROCE_V2_OPCODE_SEND_WITH_INV:
+			wc->opcode = IB_WC_RECV;
+			wc->wc_flags = IB_WC_WITH_INVALIDATE;
+			wc->ex.invalidate_rkey = cqe->rkey_immtdata;
+			break;
+		default:
+			wc->status = IB_WC_GENERAL_ERR;
+			break;
+		}
+
+		/* Update tail pointer, record wr_id */
+		wq = &(*cur_qp)->rq;
+		wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
+		++wq->tail;
+
+		wc->sl = (u8)roce_get_field(cqe->byte_32, V2_CQE_BYTE_32_SL_M,
+					    V2_CQE_BYTE_32_SL_S);
+		wc->src_qp = (u8)roce_get_field(cqe->byte_32,
+						V2_CQE_BYTE_32_RMT_QPN_M,
+						V2_CQE_BYTE_32_RMT_QPN_S);
+		wc->wc_flags |= (roce_get_bit(cqe->byte_32,
+					      V2_CQE_BYTE_32_GRH_S) ?
+					      IB_WC_GRH : 0);
+	}
+
+	return 0;
+}
+
+static int hns_roce_v2_poll_cq(struct ib_cq *ibcq, int num_entries,
+			       struct ib_wc *wc)
+{
+	struct hns_roce_cq *hr_cq = to_hr_cq(ibcq);
+	struct hns_roce_qp *cur_qp = NULL;
+	unsigned long flags;
+	int npolled;
+
+	spin_lock_irqsave(&hr_cq->lock, flags);
+
+	for (npolled = 0; npolled < num_entries; ++npolled) {
+		if (hns_roce_v2_poll_one(hr_cq, &cur_qp, wc + npolled))
+			break;
+	}
+
+	if (npolled) {
+		/* Memory barrier */
+		wmb();
+		hns_roce_v2_cq_set_ci(hr_cq, hr_cq->cons_index);
+	}
+
+	spin_unlock_irqrestore(&hr_cq->lock, flags);
+
+	return npolled;
+}
+
 static int hns_roce_v2_set_hem(struct hns_roce_dev *hr_dev,
 			       struct hns_roce_hem_table *table, int obj,
 			       int step_idx)
@@ -906,8 +1279,11 @@ static int hns_roce_v2_clear_hem(struct hns_roce_dev *hr_dev,
 	.chk_mbox = hns_roce_v2_chk_mbox,
 	.set_gid = hns_roce_v2_set_gid,
 	.set_mac = hns_roce_v2_set_mac,
+	.write_cqc = hns_roce_v2_write_cqc,
 	.set_hem = hns_roce_v2_set_hem,
 	.clear_hem = hns_roce_v2_clear_hem,
+	.req_notify_cq = hns_roce_v2_req_notify_cq,
+	.poll_cq = hns_roce_v2_poll_cq,
 };
 
 static const struct pci_device_id hns_roce_hw_v2_pci_tbl[] = {
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index aee8f34..8b69875 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -100,6 +100,61 @@
 	(step_idx == 1 && hop_num == 1) || \
 	(step_idx == 2 && hop_num == 2))
 
+#define V2_CQ_DB_REQ_NOT_SOL			0
+#define V2_CQ_DB_REQ_NOT			1
+
+#define V2_CQ_STATE_VALID			1
+
+#define HNS_ROCE_V2_CQE_QPN_MASK		0x3ffff
+
+enum {
+	HNS_ROCE_SQ_OPCODE_SEND = 0x0,
+	HNS_ROCE_SQ_OPCODE_SEND_WITH_INV = 0x1,
+	HNS_ROCE_SQ_OPCODE_SEND_WITH_IMM = 0x2,
+	HNS_ROCE_SQ_OPCODE_RDMA_WRITE = 0x3,
+	HNS_ROCE_SQ_OPCODE_RDMA_WRITE_WITH_IMM = 0x4,
+	HNS_ROCE_SQ_OPCODE_RDMA_READ = 0x5,
+	HNS_ROCE_SQ_OPCODE_ATOMIC_COMP_AND_SWAP = 0x6,
+	HNS_ROCE_SQ_OPCODE_ATOMIC_FETCH_AND_ADD = 0x7,
+	HNS_ROCE_SQ_OPCODE_ATOMIC_MASK_COMP_AND_SWAP = 0x8,
+	HNS_ROCE_SQ_OPCODE_ATOMIC_MASK_FETCH_AND_ADD = 0x9,
+	HNS_ROCE_SQ_OPCODE_FAST_REG_WR = 0xa,
+	HNS_ROCE_SQ_OPCODE_LOCAL_INV = 0xb,
+	HNS_ROCE_SQ_OPCODE_BIND_MW = 0xc,
+};
+
+enum {
+	/* rq operations */
+	HNS_ROCE_V2_OPCODE_RDMA_WRITE_IMM = 0x0,
+	HNS_ROCE_V2_OPCODE_SEND = 0x1,
+	HNS_ROCE_V2_OPCODE_SEND_WITH_IMM = 0x2,
+	HNS_ROCE_V2_OPCODE_SEND_WITH_INV = 0x3,
+};
+
+enum {
+	HNS_ROCE_V2_CQ_DB_PTR	= 0x3,
+	HNS_ROCE_V2_CQ_DB_NTR	= 0x4,
+};
+
+enum {
+	HNS_ROCE_CQE_V2_SUCCESS				= 0x00,
+	HNS_ROCE_CQE_V2_LOCAL_LENGTH_ERR		= 0x01,
+	HNS_ROCE_CQE_V2_LOCAL_QP_OP_ERR			= 0x02,
+	HNS_ROCE_CQE_V2_LOCAL_PROT_ERR			= 0x04,
+	HNS_ROCE_CQE_V2_WR_FLUSH_ERR			= 0x05,
+	HNS_ROCE_CQE_V2_MW_BIND_ERR			= 0x06,
+	HNS_ROCE_CQE_V2_BAD_RESP_ERR			= 0x10,
+	HNS_ROCE_CQE_V2_LOCAL_ACCESS_ERR		= 0x11,
+	HNS_ROCE_CQE_V2_REMOTE_INVAL_REQ_ERR		= 0x12,
+	HNS_ROCE_CQE_V2_REMOTE_ACCESS_ERR		= 0x13,
+	HNS_ROCE_CQE_V2_REMOTE_OP_ERR			= 0x14,
+	HNS_ROCE_CQE_V2_TRANSPORT_RETRY_EXC_ERR		= 0x15,
+	HNS_ROCE_CQE_V2_RNR_RETRY_EXC_ERR		= 0x16,
+	HNS_ROCE_CQE_V2_REMOTE_ABORT_ERR		= 0x22,
+
+	HNS_ROCE_V2_CQE_STATUS_MASK			= 0xff,
+};
+
 /* CMQ command */
 enum hns_roce_opcode_type {
 	HNS_ROCE_OPC_QUERY_HW_VER			= 0x8000,
@@ -122,6 +177,173 @@ enum hns_roce_cmd_return_status {
 	CMD_QUEUE_FULL		= 3,
 };
 
+struct hns_roce_v2_cq_context {
+	u32	byte_4_pg_ceqn;
+	u32	byte_8_cqn;
+	u32	cqe_cur_blk_addr;
+	u32	byte_16_hop_addr;
+	u32	cqe_nxt_blk_addr;
+	u32	byte_24_pgsz_addr;
+	u32	byte_28_cq_pi;
+	u32	byte_32_cq_ci;
+	u32	cqe_ba;
+	u32	byte_40_cqe_ba;
+	u32	byte_44_db_record;
+	u32	db_record_addr;
+	u32	byte_52_cqe_cnt;
+	u32	byte_56_cqe_period_maxcnt;
+	u32	cqe_report_timer;
+	u32	byte_64_se_cqe_idx;
+};
+#define	V2_CQC_BYTE_4_CQ_ST_S 0
+#define V2_CQC_BYTE_4_CQ_ST_M GENMASK(1, 0)
+
+#define	V2_CQC_BYTE_4_POLL_S 2
+
+#define	V2_CQC_BYTE_4_SE_S 3
+
+#define	V2_CQC_BYTE_4_OVER_IGNORE_S 4
+
+#define	V2_CQC_BYTE_4_COALESCE_S 5
+
+#define	V2_CQC_BYTE_4_ARM_ST_S 6
+#define V2_CQC_BYTE_4_ARM_ST_M GENMASK(7, 6)
+
+#define	V2_CQC_BYTE_4_SHIFT_S 8
+#define V2_CQC_BYTE_4_SHIFT_M GENMASK(12, 8)
+
+#define	V2_CQC_BYTE_4_CMD_SN_S 13
+#define V2_CQC_BYTE_4_CMD_SN_M GENMASK(14, 13)
+
+#define	V2_CQC_BYTE_4_CEQN_S 15
+#define V2_CQC_BYTE_4_CEQN_M GENMASK(23, 15)
+
+#define	V2_CQC_BYTE_4_PAGE_OFFSET_S 24
+#define V2_CQC_BYTE_4_PAGE_OFFSET_M GENMASK(31, 24)
+
+#define	V2_CQC_BYTE_8_CQN_S 0
+#define V2_CQC_BYTE_8_CQN_M GENMASK(23, 0)
+
+#define	V2_CQC_BYTE_16_CQE_CUR_BLK_ADDR_S 0
+#define V2_CQC_BYTE_16_CQE_CUR_BLK_ADDR_M GENMASK(19, 0)
+
+#define	V2_CQC_BYTE_16_CQE_HOP_NUM_S 30
+#define V2_CQC_BYTE_16_CQE_HOP_NUM_M GENMASK(31, 30)
+
+#define	V2_CQC_BYTE_24_CQE_NXT_BLK_ADDR_S 0
+#define V2_CQC_BYTE_24_CQE_NXT_BLK_ADDR_M GENMASK(19, 0)
+
+#define	V2_CQC_BYTE_24_CQE_BA_PG_SZ_S 24
+#define V2_CQC_BYTE_24_CQE_BA_PG_SZ_M GENMASK(27, 24)
+
+#define	V2_CQC_BYTE_24_CQE_BUF_PG_SZ_S 28
+#define V2_CQC_BYTE_24_CQE_BUF_PG_SZ_M GENMASK(31, 28)
+
+#define	V2_CQC_BYTE_28_CQ_PRODUCER_IDX_S 0
+#define V2_CQC_BYTE_28_CQ_PRODUCER_IDX_M GENMASK(23, 0)
+
+#define	V2_CQC_BYTE_32_CQ_CONSUMER_IDX_S 0
+#define V2_CQC_BYTE_32_CQ_CONSUMER_IDX_M GENMASK(23, 0)
+
+#define	V2_CQC_BYTE_40_CQE_BA_S 0
+#define V2_CQC_BYTE_40_CQE_BA_M GENMASK(28, 0)
+
+#define	V2_CQC_BYTE_44_DB_RECORD_EN_S 0
+
+#define	V2_CQC_BYTE_52_CQE_CNT_S 0
+#define	V2_CQC_BYTE_52_CQE_CNT_M GENMASK(23, 0)
+
+#define	V2_CQC_BYTE_56_CQ_MAX_CNT_S 0
+#define V2_CQC_BYTE_56_CQ_MAX_CNT_M GENMASK(15, 0)
+
+#define	V2_CQC_BYTE_56_CQ_PERIOD_S 16
+#define V2_CQC_BYTE_56_CQ_PERIOD_M GENMASK(31, 16)
+
+#define	V2_CQC_BYTE_64_SE_CQE_IDX_S 0
+#define	V2_CQC_BYTE_64_SE_CQE_IDX_M GENMASK(23, 0)
+
+struct hns_roce_v2_cqe {
+	u32	byte_4;
+	u32	rkey_immtdata;
+	u32	byte_12;
+	u32	byte_16;
+	u32	byte_cnt;
+	u32	smac;
+	u32	byte_28;
+	u32	byte_32;
+};
+
+#define	V2_CQE_BYTE_4_OPCODE_S 0
+#define V2_CQE_BYTE_4_OPCODE_M GENMASK(4, 0)
+
+#define	V2_CQE_BYTE_4_RQ_INLINE_S 5
+
+#define	V2_CQE_BYTE_4_S_R_S 6
+
+#define	V2_CQE_BYTE_4_OWNER_S 7
+
+#define	V2_CQE_BYTE_4_STATUS_S 8
+#define V2_CQE_BYTE_4_STATUS_M GENMASK(15, 8)
+
+#define	V2_CQE_BYTE_4_WQE_INDX_S 16
+#define V2_CQE_BYTE_4_WQE_INDX_M GENMASK(31, 16)
+
+#define	V2_CQE_BYTE_12_XRC_SRQN_S 0
+#define V2_CQE_BYTE_12_XRC_SRQN_M GENMASK(23, 0)
+
+#define	V2_CQE_BYTE_16_LCL_QPN_S 0
+#define V2_CQE_BYTE_16_LCL_QPN_M GENMASK(23, 0)
+
+#define	V2_CQE_BYTE_16_SUB_STATUS_S 24
+#define V2_CQE_BYTE_16_SUB_STATUS_M GENMASK(31, 24)
+
+#define	V2_CQE_BYTE_28_SMAC_4_S 0
+#define V2_CQE_BYTE_28_SMAC_4_M	GENMASK(7, 0)
+
+#define	V2_CQE_BYTE_28_SMAC_5_S 8
+#define V2_CQE_BYTE_28_SMAC_5_M	GENMASK(15, 8)
+
+#define	V2_CQE_BYTE_28_PORT_TYPE_S 16
+#define V2_CQE_BYTE_28_PORT_TYPE_M GENMASK(17, 16)
+
+#define	V2_CQE_BYTE_32_RMT_QPN_S 0
+#define V2_CQE_BYTE_32_RMT_QPN_M GENMASK(23, 0)
+
+#define	V2_CQE_BYTE_32_SL_S 24
+#define V2_CQE_BYTE_32_SL_M GENMASK(26, 24)
+
+#define	V2_CQE_BYTE_32_PORTN_S 27
+#define V2_CQE_BYTE_32_PORTN_M GENMASK(29, 27)
+
+#define	V2_CQE_BYTE_32_GRH_S 30
+
+#define	V2_CQE_BYTE_32_LPK_S 31
+
+#define	V2_DB_BYTE_4_TAG_S 0
+#define V2_DB_BYTE_4_TAG_M GENMASK(23, 0)
+
+#define	V2_DB_BYTE_4_CMD_S 24
+#define V2_DB_BYTE_4_CMD_M GENMASK(27, 24)
+
+struct hns_roce_v2_cq_db {
+	u32	byte_4;
+	u32	parameter;
+};
+
+#define	V2_CQ_DB_BYTE_4_TAG_S 0
+#define V2_CQ_DB_BYTE_4_TAG_M GENMASK(23, 0)
+
+#define	V2_CQ_DB_BYTE_4_CMD_S 24
+#define V2_CQ_DB_BYTE_4_CMD_M GENMASK(27, 24)
+
+#define V2_CQ_DB_PARAMETER_CONS_IDX_S 0
+#define V2_CQ_DB_PARAMETER_CONS_IDX_M GENMASK(23, 0)
+
+#define V2_CQ_DB_PARAMETER_CMD_SN_S 25
+#define V2_CQ_DB_PARAMETER_CMD_SN_M GENMASK(26, 25)
+
+#define V2_CQ_DB_PARAMETER_NOTIFY_S 24
+
 struct hns_roce_query_version {
 	__le16 rocee_vendor_id;
 	__le16 rocee_hw_version;
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 15/20] RDMA/hns: Add QP operations support for hip08 SoC

[Wei Hu (Xavier)]

This patch implements QP operations for hip08 RoCE driver and
fixes some checkpatch warning about print message in QP function.
The QP operations includes create QP, query QP, modify QP and
destroy QP.

Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_cmd.h    |    4 +
 drivers/infiniband/hw/hns/hns_roce_device.h |   11 +
 drivers/infiniband/hw/hns/hns_roce_hw_v1.c  |    1 +
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 1306 +++++++++++++++++++++++++++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  |  450 +++++++++
 drivers/infiniband/hw/hns/hns_roce_qp.c     |  140 ++-
 6 files changed, 1883 insertions(+), 29 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_cmd.h b/drivers/infiniband/hw/hns/hns_roce_cmd.h
index fc40069..77299ae5 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cmd.h
+++ b/drivers/infiniband/hw/hns/hns_roce_cmd.h
@@ -48,6 +48,10 @@ enum {
 	HNS_ROCE_CMD_DESTROY_QPC_BT1	= 0x9,
 	HNS_ROCE_CMD_DESTROY_QPC_BT2	= 0xa,
 
+	/* QPC operation */
+	HNS_ROCE_CMD_MODIFY_QPC		= 0x41,
+	HNS_ROCE_CMD_QUERY_QPC		= 0x42,
+
 	/* CQC BT commands */
 	HNS_ROCE_CMD_WRITE_CQC_BT0	= 0x10,
 	HNS_ROCE_CMD_WRITE_CQC_BT1	= 0x11,
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index bee2088..22f2079 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -304,6 +304,12 @@ struct hns_roce_wq {
 	void __iomem	*db_reg_l;
 };
 
+struct hns_roce_sge {
+	int		sge_cnt;  /* SGE num */
+	int		offset;
+	int		sge_shift;/* SGE size */
+};
+
 struct hns_roce_buf_list {
 	void		*buf;
 	dma_addr_t	map;
@@ -455,6 +461,9 @@ struct hns_roce_qp {
 
 	atomic_t		refcount;
 	struct completion	free;
+
+	struct hns_roce_sge	sge;
+	u32			next_sge;
 };
 
 struct hns_roce_sqp {
@@ -509,6 +518,7 @@ struct hns_roce_caps {
 	int		num_cqs;
 	int		max_cqes;
 	int		min_cqes;
+	u32		min_wqes;
 	int		reserved_cqs;
 	int		num_aeq_vectors;	/* 1 */
 	int		num_comp_vectors;	/* 32 ceq */
@@ -788,6 +798,7 @@ int hns_roce_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 		       int attr_mask, struct ib_udata *udata);
 void *get_recv_wqe(struct hns_roce_qp *hr_qp, int n);
 void *get_send_wqe(struct hns_roce_qp *hr_qp, int n);
+void *get_send_extend_sge(struct hns_roce_qp *hr_qp, int n);
 bool hns_roce_wq_overflow(struct hns_roce_wq *hr_wq, int nreq,
 			  struct ib_cq *ib_cq);
 enum hns_roce_qp_state to_hns_roce_state(enum ib_qp_state state);
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 1aa2be6..8ed0538 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -1475,6 +1475,7 @@ int hns_roce_v1_profile(struct hns_roce_dev *hr_dev)
 
 	caps->num_qps		= HNS_ROCE_V1_MAX_QP_NUM;
 	caps->max_wqes		= HNS_ROCE_V1_MAX_WQE_NUM;
+	caps->min_wqes		= HNS_ROCE_MIN_WQE_NUM;
 	caps->num_cqs		= HNS_ROCE_V1_MAX_CQ_NUM;
 	caps->min_cqes		= HNS_ROCE_MIN_CQE_NUM;
 	caps->max_cqes		= HNS_ROCE_V1_MAX_CQE_NUM;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 94339af..f2425ee 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -807,6 +807,61 @@ static void hns_roce_v2_cq_set_ci(struct hns_roce_cq *hr_cq, u32 cons_index)
 
 }
 
+static void __hns_roce_v2_cq_clean(struct hns_roce_cq *hr_cq, u32 qpn,
+				   struct hns_roce_srq *srq)
+{
+	struct hns_roce_v2_cqe *cqe, *dest;
+	u32 prod_index;
+	int nfreed = 0;
+	u8 owner_bit;
+
+	for (prod_index = hr_cq->cons_index; get_sw_cqe_v2(hr_cq, prod_index);
+	     ++prod_index) {
+		if (prod_index == hr_cq->cons_index + hr_cq->ib_cq.cqe)
+			break;
+	}
+
+	/*
+	 * Now backwards through the CQ, removing CQ entries
+	 * that match our QP by overwriting them with next entries.
+	 */
+	while ((int) --prod_index - (int) hr_cq->cons_index >= 0) {
+		cqe = get_cqe_v2(hr_cq, prod_index & hr_cq->ib_cq.cqe);
+		if ((roce_get_field(cqe->byte_16, V2_CQE_BYTE_16_LCL_QPN_M,
+				    V2_CQE_BYTE_16_LCL_QPN_S) &
+				    HNS_ROCE_V2_CQE_QPN_MASK) == qpn) {
+			/* In v1 engine, not support SRQ */
+			++nfreed;
+		} else if (nfreed) {
+			dest = get_cqe_v2(hr_cq, (prod_index + nfreed) &
+					  hr_cq->ib_cq.cqe);
+			owner_bit = roce_get_bit(dest->byte_4,
+						 V2_CQE_BYTE_4_OWNER_S);
+			memcpy(dest, cqe, sizeof(*cqe));
+			roce_set_bit(dest->byte_4, V2_CQE_BYTE_4_OWNER_S,
+				     owner_bit);
+		}
+	}
+
+	if (nfreed) {
+		hr_cq->cons_index += nfreed;
+		/*
+		 * Make sure update of buffer contents is done before
+		 * updating consumer index.
+		 */
+		wmb();
+		hns_roce_v2_cq_set_ci(hr_cq, hr_cq->cons_index);
+	}
+}
+
+static void hns_roce_v2_cq_clean(struct hns_roce_cq *hr_cq, u32 qpn,
+				 struct hns_roce_srq *srq)
+{
+	spin_lock_irq(&hr_cq->lock);
+	__hns_roce_v2_cq_clean(hr_cq, qpn, srq);
+	spin_unlock_irq(&hr_cq->lock);
+}
+
 static void hns_roce_v2_write_cqc(struct hns_roce_dev *hr_dev,
 				  struct hns_roce_cq *hr_cq, void *mb_buf,
 				  u64 *mtts, dma_addr_t dma_handle, int nent,
@@ -1271,6 +1326,1254 @@ static int hns_roce_v2_clear_hem(struct hns_roce_dev *hr_dev,
 	return ret;
 }
 
+static int hns_roce_v2_qp_modify(struct hns_roce_dev *hr_dev,
+				 struct hns_roce_mtt *mtt,
+				 enum ib_qp_state cur_state,
+				 enum ib_qp_state new_state,
+				 struct hns_roce_v2_qp_context *context,
+				 struct hns_roce_qp *hr_qp)
+{
+	struct hns_roce_cmd_mailbox *mailbox;
+	int ret;
+
+	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
+	if (IS_ERR(mailbox))
+		return PTR_ERR(mailbox);
+
+	memcpy(mailbox->buf, context, sizeof(*context) * 2);
+
+	ret = hns_roce_cmd_mbox(hr_dev, mailbox->dma, 0, hr_qp->qpn, 0,
+				HNS_ROCE_CMD_MODIFY_QPC,
+				HNS_ROCE_CMD_TIMEOUT_MSECS);
+
+	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
+
+	return ret;
+}
+
+static void modify_qp_reset_to_init(struct ib_qp *ibqp,
+				    const struct ib_qp_attr *attr,
+				    struct hns_roce_v2_qp_context *context,
+				    struct hns_roce_v2_qp_context *qpc_mask)
+{
+	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
+
+	/*
+	 * In v2 engine, software pass context and context mask to hardware
+	 * when modifying qp. If software need modify some fields in context,
+	 * we should set all bits of the relevant fields in context mask to
+	 * 0 at the same time, else set them to 0x1.
+	 */
+	roce_set_field(context->byte_4_sqpn_tst, V2_QPC_BYTE_4_TST_M,
+		       V2_QPC_BYTE_4_TST_S, to_hr_qp_type(hr_qp->ibqp.qp_type));
+	roce_set_field(qpc_mask->byte_4_sqpn_tst, V2_QPC_BYTE_4_TST_M,
+		       V2_QPC_BYTE_4_TST_S, 0);
+
+	roce_set_field(context->byte_4_sqpn_tst, V2_QPC_BYTE_4_SGE_SHIFT_M,
+		       V2_QPC_BYTE_4_SGE_SHIFT_S, hr_qp->sq.max_gs > 2 ?
+		       ilog2((unsigned int)hr_qp->sge.sge_cnt) : 0);
+	roce_set_field(qpc_mask->byte_4_sqpn_tst, V2_QPC_BYTE_4_SGE_SHIFT_M,
+		       V2_QPC_BYTE_4_SGE_SHIFT_S, 0);
+
+	roce_set_field(context->byte_4_sqpn_tst, V2_QPC_BYTE_4_SQPN_M,
+		       V2_QPC_BYTE_4_SQPN_S, hr_qp->qpn);
+	roce_set_field(qpc_mask->byte_4_sqpn_tst, V2_QPC_BYTE_4_SQPN_M,
+		       V2_QPC_BYTE_4_SQPN_S, 0);
+
+	roce_set_field(context->byte_16_buf_ba_pg_sz, V2_QPC_BYTE_16_PD_M,
+		       V2_QPC_BYTE_16_PD_S, to_hr_pd(ibqp->pd)->pdn);
+	roce_set_field(qpc_mask->byte_16_buf_ba_pg_sz, V2_QPC_BYTE_16_PD_M,
+		       V2_QPC_BYTE_16_PD_S, 0);
+
+	roce_set_field(context->byte_20_smac_sgid_idx, V2_QPC_BYTE_20_RQWS_M,
+		       V2_QPC_BYTE_20_RQWS_S, ilog2(hr_qp->rq.max_gs));
+	roce_set_field(qpc_mask->byte_20_smac_sgid_idx, V2_QPC_BYTE_20_RQWS_M,
+		       V2_QPC_BYTE_20_RQWS_S, 0);
+
+	roce_set_field(context->byte_20_smac_sgid_idx,
+		       V2_QPC_BYTE_20_SQ_SHIFT_M, V2_QPC_BYTE_20_SQ_SHIFT_S,
+		       ilog2((unsigned int)hr_qp->sq.wqe_cnt));
+	roce_set_field(qpc_mask->byte_20_smac_sgid_idx,
+		       V2_QPC_BYTE_20_SQ_SHIFT_M, V2_QPC_BYTE_20_SQ_SHIFT_S, 0);
+
+	roce_set_field(context->byte_20_smac_sgid_idx,
+		       V2_QPC_BYTE_20_RQ_SHIFT_M, V2_QPC_BYTE_20_RQ_SHIFT_S,
+		       ilog2((unsigned int)hr_qp->rq.wqe_cnt));
+	roce_set_field(qpc_mask->byte_20_smac_sgid_idx,
+		       V2_QPC_BYTE_20_RQ_SHIFT_M, V2_QPC_BYTE_20_RQ_SHIFT_S, 0);
+
+	/* No VLAN need to set 0xFFF */
+	roce_set_field(context->byte_24_mtu_tc, V2_QPC_BYTE_24_VLAN_IDX_M,
+		       V2_QPC_BYTE_24_VLAN_IDX_S, 0xfff);
+	roce_set_field(qpc_mask->byte_24_mtu_tc, V2_QPC_BYTE_24_VLAN_IDX_M,
+		       V2_QPC_BYTE_24_VLAN_IDX_S, 0);
+
+	/*
+	 * Set some fields in context to zero, Because the default values
+	 * of all fields in context are zero, we need not set them to 0 again.
+	 * but we should set the relevant fields of context mask to 0.
+	 */
+	roce_set_bit(qpc_mask->byte_56_dqpn_err, V2_QPC_BYTE_56_SQ_TX_ERR_S, 0);
+	roce_set_bit(qpc_mask->byte_56_dqpn_err, V2_QPC_BYTE_56_SQ_RX_ERR_S, 0);
+	roce_set_bit(qpc_mask->byte_56_dqpn_err, V2_QPC_BYTE_56_RQ_TX_ERR_S, 0);
+	roce_set_bit(qpc_mask->byte_56_dqpn_err, V2_QPC_BYTE_56_RQ_RX_ERR_S, 0);
+
+	roce_set_field(qpc_mask->byte_60_qpst_mapid, V2_QPC_BYTE_60_MAPID_M,
+		       V2_QPC_BYTE_60_MAPID_S, 0);
+
+	roce_set_bit(qpc_mask->byte_60_qpst_mapid,
+		     V2_QPC_BYTE_60_INNER_MAP_IND_S, 0);
+	roce_set_bit(qpc_mask->byte_60_qpst_mapid, V2_QPC_BYTE_60_SQ_MAP_IND_S,
+		     0);
+	roce_set_bit(qpc_mask->byte_60_qpst_mapid, V2_QPC_BYTE_60_RQ_MAP_IND_S,
+		     0);
+	roce_set_bit(qpc_mask->byte_60_qpst_mapid, V2_QPC_BYTE_60_EXT_MAP_IND_S,
+		     0);
+	roce_set_bit(qpc_mask->byte_60_qpst_mapid, V2_QPC_BYTE_60_SQ_RLS_IND_S,
+		     0);
+	roce_set_bit(qpc_mask->byte_60_qpst_mapid, V2_QPC_BYTE_60_SQ_EXT_IND_S,
+		     0);
+	roce_set_bit(qpc_mask->byte_28_at_fl, V2_QPC_BYTE_28_CNP_TX_FLAG_S, 0);
+	roce_set_bit(qpc_mask->byte_28_at_fl, V2_QPC_BYTE_28_CE_FLAG_S, 0);
+
+	roce_set_bit(context->byte_76_srqn_op_en, V2_QPC_BYTE_76_RRE_S,
+		     !!(attr->qp_access_flags & IB_ACCESS_REMOTE_READ));
+	roce_set_bit(qpc_mask->byte_76_srqn_op_en, V2_QPC_BYTE_76_RRE_S, 0);
+
+	roce_set_bit(context->byte_76_srqn_op_en, V2_QPC_BYTE_76_RWE_S,
+		     !!(attr->qp_access_flags & IB_ACCESS_REMOTE_WRITE));
+	roce_set_bit(qpc_mask->byte_76_srqn_op_en, V2_QPC_BYTE_76_RWE_S, 0);
+
+	roce_set_bit(context->byte_76_srqn_op_en, V2_QPC_BYTE_76_ATE_S,
+		     !!(attr->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC));
+	roce_set_bit(qpc_mask->byte_76_srqn_op_en, V2_QPC_BYTE_76_ATE_S, 0);
+
+	roce_set_bit(qpc_mask->byte_76_srqn_op_en, V2_QPC_BYTE_76_RQIE_S, 0);
+
+	roce_set_field(context->byte_80_rnr_rx_cqn, V2_QPC_BYTE_80_RX_CQN_M,
+		       V2_QPC_BYTE_80_RX_CQN_S, to_hr_cq(ibqp->recv_cq)->cqn);
+	roce_set_field(qpc_mask->byte_80_rnr_rx_cqn, V2_QPC_BYTE_80_RX_CQN_M,
+		       V2_QPC_BYTE_80_RX_CQN_S, 0);
+	if (ibqp->srq) {
+		roce_set_field(context->byte_76_srqn_op_en,
+			       V2_QPC_BYTE_76_SRQN_M, V2_QPC_BYTE_76_SRQN_S,
+			       to_hr_srq(ibqp->srq)->srqn);
+		roce_set_field(qpc_mask->byte_76_srqn_op_en,
+			       V2_QPC_BYTE_76_SRQN_M, V2_QPC_BYTE_76_SRQN_S, 0);
+		roce_set_bit(context->byte_76_srqn_op_en,
+			     V2_QPC_BYTE_76_SRQ_EN_S, 1);
+		roce_set_bit(qpc_mask->byte_76_srqn_op_en,
+			     V2_QPC_BYTE_76_SRQ_EN_S, 0);
+	}
+
+	roce_set_field(qpc_mask->byte_84_rq_ci_pi,
+		       V2_QPC_BYTE_84_RQ_PRODUCER_IDX_M,
+		       V2_QPC_BYTE_84_RQ_PRODUCER_IDX_S, 0);
+	roce_set_field(qpc_mask->byte_84_rq_ci_pi,
+		       V2_QPC_BYTE_84_RQ_CONSUMER_IDX_M,
+		       V2_QPC_BYTE_84_RQ_CONSUMER_IDX_S, 0);
+
+	roce_set_field(qpc_mask->byte_92_srq_info, V2_QPC_BYTE_92_SRQ_INFO_M,
+		       V2_QPC_BYTE_92_SRQ_INFO_S, 0);
+
+	roce_set_field(qpc_mask->byte_96_rx_reqmsn, V2_QPC_BYTE_96_RX_REQ_MSN_M,
+		       V2_QPC_BYTE_96_RX_REQ_MSN_S, 0);
+
+	roce_set_field(qpc_mask->byte_104_rq_sge,
+		       V2_QPC_BYTE_104_RQ_CUR_WQE_SGE_NUM_M,
+		       V2_QPC_BYTE_104_RQ_CUR_WQE_SGE_NUM_S, 0);
+
+	roce_set_bit(qpc_mask->byte_108_rx_reqepsn,
+		     V2_QPC_BYTE_108_RX_REQ_PSN_ERR_S, 0);
+	roce_set_field(qpc_mask->byte_108_rx_reqepsn,
+		       V2_QPC_BYTE_108_RX_REQ_LAST_OPTYPE_M,
+		       V2_QPC_BYTE_108_RX_REQ_LAST_OPTYPE_S, 0);
+	roce_set_bit(qpc_mask->byte_108_rx_reqepsn,
+		     V2_QPC_BYTE_108_RX_REQ_RNR_S, 0);
+
+	qpc_mask->rq_rnr_timer = 0;
+	qpc_mask->rx_msg_len = 0;
+	qpc_mask->rx_rkey_pkt_info = 0;
+	qpc_mask->rx_va = 0;
+
+	roce_set_field(qpc_mask->byte_132_trrl, V2_QPC_BYTE_132_TRRL_HEAD_MAX_M,
+		       V2_QPC_BYTE_132_TRRL_HEAD_MAX_S, 0);
+	roce_set_field(qpc_mask->byte_132_trrl, V2_QPC_BYTE_132_TRRL_TAIL_MAX_M,
+		       V2_QPC_BYTE_132_TRRL_TAIL_MAX_S, 0);
+
+	roce_set_bit(qpc_mask->byte_140_raq, V2_QPC_BYTE_140_RSVD_RAQ_MAP_S, 0);
+	roce_set_field(qpc_mask->byte_140_raq, V2_QPC_BYTE_140_RAQ_TRRL_HEAD_M,
+		       V2_QPC_BYTE_140_RAQ_TRRL_HEAD_S, 0);
+	roce_set_field(qpc_mask->byte_140_raq, V2_QPC_BYTE_140_RAQ_TRRL_TAIL_M,
+		       V2_QPC_BYTE_140_RAQ_TRRL_TAIL_S, 0);
+
+	roce_set_field(qpc_mask->byte_144_raq,
+		       V2_QPC_BYTE_144_RAQ_RTY_INI_PSN_M,
+		       V2_QPC_BYTE_144_RAQ_RTY_INI_PSN_S, 0);
+	roce_set_bit(qpc_mask->byte_144_raq, V2_QPC_BYTE_144_RAQ_RTY_INI_IND_S,
+		     0);
+	roce_set_field(qpc_mask->byte_144_raq, V2_QPC_BYTE_144_RAQ_CREDIT_M,
+		       V2_QPC_BYTE_144_RAQ_CREDIT_S, 0);
+	roce_set_bit(qpc_mask->byte_144_raq, V2_QPC_BYTE_144_RESP_RTY_FLG_S, 0);
+
+	roce_set_field(qpc_mask->byte_148_raq, V2_QPC_BYTE_148_RQ_MSN_M,
+		       V2_QPC_BYTE_148_RQ_MSN_S, 0);
+	roce_set_field(qpc_mask->byte_148_raq, V2_QPC_BYTE_148_RAQ_SYNDROME_M,
+		       V2_QPC_BYTE_148_RAQ_SYNDROME_S, 0);
+
+	roce_set_field(qpc_mask->byte_152_raq, V2_QPC_BYTE_152_RAQ_PSN_M,
+		       V2_QPC_BYTE_152_RAQ_PSN_S, 0);
+	roce_set_field(qpc_mask->byte_152_raq,
+		       V2_QPC_BYTE_152_RAQ_TRRL_RTY_HEAD_M,
+		       V2_QPC_BYTE_152_RAQ_TRRL_RTY_HEAD_S, 0);
+
+	roce_set_field(qpc_mask->byte_156_raq, V2_QPC_BYTE_156_RAQ_USE_PKTN_M,
+		       V2_QPC_BYTE_156_RAQ_USE_PKTN_S, 0);
+
+	roce_set_field(qpc_mask->byte_160_sq_ci_pi,
+		       V2_QPC_BYTE_160_SQ_PRODUCER_IDX_M,
+		       V2_QPC_BYTE_160_SQ_PRODUCER_IDX_S, 0);
+	roce_set_field(qpc_mask->byte_160_sq_ci_pi,
+		       V2_QPC_BYTE_160_SQ_CONSUMER_IDX_M,
+		       V2_QPC_BYTE_160_SQ_CONSUMER_IDX_S, 0);
+
+	roce_set_field(context->byte_168_irrl_idx,
+		       V2_QPC_BYTE_168_SQ_SHIFT_BAK_M,
+		       V2_QPC_BYTE_168_SQ_SHIFT_BAK_S,
+		       ilog2((unsigned int)hr_qp->sq.wqe_cnt));
+	roce_set_field(qpc_mask->byte_168_irrl_idx,
+		       V2_QPC_BYTE_168_SQ_SHIFT_BAK_M,
+		       V2_QPC_BYTE_168_SQ_SHIFT_BAK_S, 0);
+
+	roce_set_bit(qpc_mask->byte_168_irrl_idx,
+		     V2_QPC_BYTE_168_MSG_RTY_LP_FLG_S, 0);
+	roce_set_field(qpc_mask->byte_168_irrl_idx,
+		       V2_QPC_BYTE_168_IRRL_IDX_LSB_M,
+		       V2_QPC_BYTE_168_IRRL_IDX_LSB_S, 0);
+
+	roce_set_field(context->byte_172_sq_psn, V2_QPC_BYTE_172_ACK_REQ_FREQ_M,
+		       V2_QPC_BYTE_172_ACK_REQ_FREQ_S, 4);
+	roce_set_field(qpc_mask->byte_172_sq_psn,
+		       V2_QPC_BYTE_172_ACK_REQ_FREQ_M,
+		       V2_QPC_BYTE_172_ACK_REQ_FREQ_S, 0);
+
+	roce_set_bit(qpc_mask->byte_172_sq_psn, V2_QPC_BYTE_172_MSG_RNR_FLG_S,
+		     0);
+
+	roce_set_field(qpc_mask->byte_176_msg_pktn,
+		       V2_QPC_BYTE_176_MSG_USE_PKTN_M,
+		       V2_QPC_BYTE_176_MSG_USE_PKTN_S, 0);
+	roce_set_field(qpc_mask->byte_176_msg_pktn,
+		       V2_QPC_BYTE_176_IRRL_HEAD_PRE_M,
+		       V2_QPC_BYTE_176_IRRL_HEAD_PRE_S, 0);
+
+	roce_set_field(qpc_mask->byte_184_irrl_idx,
+		       V2_QPC_BYTE_184_IRRL_IDX_MSB_M,
+		       V2_QPC_BYTE_184_IRRL_IDX_MSB_S, 0);
+
+	qpc_mask->cur_sge_offset = 0;
+
+	roce_set_field(qpc_mask->byte_192_ext_sge,
+		       V2_QPC_BYTE_192_CUR_SGE_IDX_M,
+		       V2_QPC_BYTE_192_CUR_SGE_IDX_S, 0);
+	roce_set_field(qpc_mask->byte_192_ext_sge,
+		       V2_QPC_BYTE_192_EXT_SGE_NUM_LEFT_M,
+		       V2_QPC_BYTE_192_EXT_SGE_NUM_LEFT_S, 0);
+
+	roce_set_field(qpc_mask->byte_196_sq_psn, V2_QPC_BYTE_196_IRRL_HEAD_M,
+		       V2_QPC_BYTE_196_IRRL_HEAD_S, 0);
+
+	roce_set_field(qpc_mask->byte_200_sq_max, V2_QPC_BYTE_200_SQ_MAX_IDX_M,
+		       V2_QPC_BYTE_200_SQ_MAX_IDX_S, 0);
+	roce_set_field(qpc_mask->byte_200_sq_max,
+		       V2_QPC_BYTE_200_LCL_OPERATED_CNT_M,
+		       V2_QPC_BYTE_200_LCL_OPERATED_CNT_S, 0);
+
+	roce_set_bit(qpc_mask->byte_208_irrl, V2_QPC_BYTE_208_PKT_RNR_FLG_S, 0);
+	roce_set_bit(qpc_mask->byte_208_irrl, V2_QPC_BYTE_208_PKT_RTY_FLG_S, 0);
+
+	roce_set_field(qpc_mask->byte_212_lsn, V2_QPC_BYTE_212_CHECK_FLG_M,
+		       V2_QPC_BYTE_212_CHECK_FLG_S, 0);
+
+	qpc_mask->sq_timer = 0;
+
+	roce_set_field(qpc_mask->byte_220_retry_psn_msn,
+		       V2_QPC_BYTE_220_RETRY_MSG_MSN_M,
+		       V2_QPC_BYTE_220_RETRY_MSG_MSN_S, 0);
+	roce_set_field(qpc_mask->byte_232_irrl_sge,
+		       V2_QPC_BYTE_232_IRRL_SGE_IDX_M,
+		       V2_QPC_BYTE_232_IRRL_SGE_IDX_S, 0);
+
+	qpc_mask->irrl_cur_sge_offset = 0;
+
+	roce_set_field(qpc_mask->byte_240_irrl_tail,
+		       V2_QPC_BYTE_240_IRRL_TAIL_REAL_M,
+		       V2_QPC_BYTE_240_IRRL_TAIL_REAL_S, 0);
+	roce_set_field(qpc_mask->byte_240_irrl_tail,
+		       V2_QPC_BYTE_240_IRRL_TAIL_RD_M,
+		       V2_QPC_BYTE_240_IRRL_TAIL_RD_S, 0);
+	roce_set_field(qpc_mask->byte_240_irrl_tail,
+		       V2_QPC_BYTE_240_RX_ACK_MSN_M,
+		       V2_QPC_BYTE_240_RX_ACK_MSN_S, 0);
+
+	roce_set_field(qpc_mask->byte_248_ack_psn, V2_QPC_BYTE_248_IRRL_PSN_M,
+		       V2_QPC_BYTE_248_IRRL_PSN_S, 0);
+	roce_set_bit(qpc_mask->byte_248_ack_psn, V2_QPC_BYTE_248_ACK_PSN_ERR_S,
+		     0);
+	roce_set_field(qpc_mask->byte_248_ack_psn,
+		       V2_QPC_BYTE_248_ACK_LAST_OPTYPE_M,
+		       V2_QPC_BYTE_248_ACK_LAST_OPTYPE_S, 0);
+	roce_set_bit(qpc_mask->byte_248_ack_psn, V2_QPC_BYTE_248_IRRL_PSN_VLD_S,
+		     0);
+	roce_set_bit(qpc_mask->byte_248_ack_psn,
+		     V2_QPC_BYTE_248_RNR_RETRY_FLAG_S, 0);
+	roce_set_bit(qpc_mask->byte_248_ack_psn, V2_QPC_BYTE_248_CQ_ERR_IND_S,
+		     0);
+
+	hr_qp->access_flags = attr->qp_access_flags;
+	hr_qp->pkey_index = attr->pkey_index;
+	roce_set_field(context->byte_252_err_txcqn, V2_QPC_BYTE_252_TX_CQN_M,
+		       V2_QPC_BYTE_252_TX_CQN_S, to_hr_cq(ibqp->send_cq)->cqn);
+	roce_set_field(qpc_mask->byte_252_err_txcqn, V2_QPC_BYTE_252_TX_CQN_M,
+		       V2_QPC_BYTE_252_TX_CQN_S, 0);
+
+	roce_set_field(qpc_mask->byte_252_err_txcqn, V2_QPC_BYTE_252_ERR_TYPE_M,
+		       V2_QPC_BYTE_252_ERR_TYPE_S, 0);
+
+	roce_set_field(qpc_mask->byte_256_sqflush_rqcqe,
+		       V2_QPC_BYTE_256_RQ_CQE_IDX_M,
+		       V2_QPC_BYTE_256_RQ_CQE_IDX_S, 0);
+	roce_set_field(qpc_mask->byte_256_sqflush_rqcqe,
+		       V2_QPC_BYTE_256_SQ_FLUSH_IDX_M,
+		       V2_QPC_BYTE_256_SQ_FLUSH_IDX_S, 0);
+}
+
+static void modify_qp_init_to_init(struct ib_qp *ibqp,
+				   const struct ib_qp_attr *attr, int attr_mask,
+				   struct hns_roce_v2_qp_context *context,
+				   struct hns_roce_v2_qp_context *qpc_mask)
+{
+	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
+
+	/*
+	 * In v2 engine, software pass context and context mask to hardware
+	 * when modifying qp. If software need modify some fields in context,
+	 * we should set all bits of the relevant fields in context mask to
+	 * 0 at the same time, else set them to 0x1.
+	 */
+	roce_set_field(context->byte_4_sqpn_tst, V2_QPC_BYTE_4_TST_M,
+		       V2_QPC_BYTE_4_TST_S, to_hr_qp_type(hr_qp->ibqp.qp_type));
+	roce_set_field(qpc_mask->byte_4_sqpn_tst, V2_QPC_BYTE_4_TST_M,
+		       V2_QPC_BYTE_4_TST_S, 0);
+
+	roce_set_field(context->byte_4_sqpn_tst, V2_QPC_BYTE_4_SGE_SHIFT_M,
+		       V2_QPC_BYTE_4_SGE_SHIFT_S, hr_qp->sq.max_gs > 2 ?
+		       ilog2((unsigned int)hr_qp->sge.sge_cnt) : 0);
+	roce_set_field(qpc_mask->byte_4_sqpn_tst, V2_QPC_BYTE_4_SGE_SHIFT_M,
+		       V2_QPC_BYTE_4_SGE_SHIFT_S, 0);
+
+	if (attr_mask & IB_QP_ACCESS_FLAGS) {
+		roce_set_bit(context->byte_76_srqn_op_en, V2_QPC_BYTE_76_RRE_S,
+			     !!(attr->qp_access_flags & IB_ACCESS_REMOTE_READ));
+		roce_set_bit(qpc_mask->byte_76_srqn_op_en, V2_QPC_BYTE_76_RRE_S,
+			     0);
+
+		roce_set_bit(context->byte_76_srqn_op_en, V2_QPC_BYTE_76_RWE_S,
+			     !!(attr->qp_access_flags &
+			     IB_ACCESS_REMOTE_WRITE));
+		roce_set_bit(qpc_mask->byte_76_srqn_op_en, V2_QPC_BYTE_76_RWE_S,
+			     0);
+
+		roce_set_bit(context->byte_76_srqn_op_en, V2_QPC_BYTE_76_ATE_S,
+			     !!(attr->qp_access_flags &
+			     IB_ACCESS_REMOTE_ATOMIC));
+		roce_set_bit(qpc_mask->byte_76_srqn_op_en, V2_QPC_BYTE_76_ATE_S,
+			     0);
+	} else {
+		roce_set_bit(context->byte_76_srqn_op_en, V2_QPC_BYTE_76_RRE_S,
+			     !!(hr_qp->access_flags & IB_ACCESS_REMOTE_READ));
+		roce_set_bit(qpc_mask->byte_76_srqn_op_en, V2_QPC_BYTE_76_RRE_S,
+			     0);
+
+		roce_set_bit(context->byte_76_srqn_op_en, V2_QPC_BYTE_76_RWE_S,
+			     !!(hr_qp->access_flags & IB_ACCESS_REMOTE_WRITE));
+		roce_set_bit(qpc_mask->byte_76_srqn_op_en, V2_QPC_BYTE_76_RWE_S,
+			     0);
+
+		roce_set_bit(context->byte_76_srqn_op_en, V2_QPC_BYTE_76_ATE_S,
+			     !!(hr_qp->access_flags & IB_ACCESS_REMOTE_ATOMIC));
+		roce_set_bit(qpc_mask->byte_76_srqn_op_en, V2_QPC_BYTE_76_ATE_S,
+			     0);
+	}
+
+	roce_set_field(context->byte_20_smac_sgid_idx,
+		       V2_QPC_BYTE_20_SQ_SHIFT_M, V2_QPC_BYTE_20_SQ_SHIFT_S,
+		       ilog2((unsigned int)hr_qp->sq.wqe_cnt));
+	roce_set_field(qpc_mask->byte_20_smac_sgid_idx,
+		       V2_QPC_BYTE_20_SQ_SHIFT_M, V2_QPC_BYTE_20_SQ_SHIFT_S, 0);
+
+	roce_set_field(context->byte_20_smac_sgid_idx,
+		       V2_QPC_BYTE_20_RQ_SHIFT_M, V2_QPC_BYTE_20_RQ_SHIFT_S,
+		       ilog2((unsigned int)hr_qp->rq.wqe_cnt));
+	roce_set_field(qpc_mask->byte_20_smac_sgid_idx,
+		       V2_QPC_BYTE_20_RQ_SHIFT_M, V2_QPC_BYTE_20_RQ_SHIFT_S, 0);
+
+	roce_set_field(context->byte_16_buf_ba_pg_sz, V2_QPC_BYTE_16_PD_M,
+		       V2_QPC_BYTE_16_PD_S, to_hr_pd(ibqp->pd)->pdn);
+	roce_set_field(qpc_mask->byte_16_buf_ba_pg_sz, V2_QPC_BYTE_16_PD_M,
+		       V2_QPC_BYTE_16_PD_S, 0);
+
+	roce_set_field(context->byte_80_rnr_rx_cqn, V2_QPC_BYTE_80_RX_CQN_M,
+		       V2_QPC_BYTE_80_RX_CQN_S, to_hr_cq(ibqp->recv_cq)->cqn);
+	roce_set_field(qpc_mask->byte_80_rnr_rx_cqn, V2_QPC_BYTE_80_RX_CQN_M,
+		       V2_QPC_BYTE_80_RX_CQN_S, 0);
+
+	roce_set_field(context->byte_252_err_txcqn, V2_QPC_BYTE_252_TX_CQN_M,
+		       V2_QPC_BYTE_252_TX_CQN_S, to_hr_cq(ibqp->recv_cq)->cqn);
+	roce_set_field(qpc_mask->byte_252_err_txcqn, V2_QPC_BYTE_252_TX_CQN_M,
+		       V2_QPC_BYTE_252_TX_CQN_S, 0);
+
+	if (ibqp->srq) {
+		roce_set_bit(context->byte_76_srqn_op_en,
+			     V2_QPC_BYTE_76_SRQ_EN_S, 1);
+		roce_set_bit(qpc_mask->byte_76_srqn_op_en,
+			     V2_QPC_BYTE_76_SRQ_EN_S, 0);
+		roce_set_field(context->byte_76_srqn_op_en,
+			       V2_QPC_BYTE_76_SRQN_M, V2_QPC_BYTE_76_SRQN_S,
+			       to_hr_srq(ibqp->srq)->srqn);
+		roce_set_field(qpc_mask->byte_76_srqn_op_en,
+			       V2_QPC_BYTE_76_SRQN_M, V2_QPC_BYTE_76_SRQN_S, 0);
+	}
+
+	if (attr_mask & IB_QP_PKEY_INDEX)
+		context->qkey_xrcd = attr->pkey_index;
+	else
+		context->qkey_xrcd = hr_qp->pkey_index;
+
+	roce_set_field(context->byte_4_sqpn_tst, V2_QPC_BYTE_4_SQPN_M,
+		       V2_QPC_BYTE_4_SQPN_S, hr_qp->qpn);
+	roce_set_field(qpc_mask->byte_4_sqpn_tst, V2_QPC_BYTE_4_SQPN_M,
+		       V2_QPC_BYTE_4_SQPN_S, 0);
+
+	roce_set_field(context->byte_56_dqpn_err, V2_QPC_BYTE_56_DQPN_M,
+		       V2_QPC_BYTE_56_DQPN_S, hr_qp->qpn);
+	roce_set_field(qpc_mask->byte_56_dqpn_err, V2_QPC_BYTE_56_DQPN_M,
+		       V2_QPC_BYTE_56_DQPN_S, 0);
+	roce_set_field(context->byte_168_irrl_idx,
+		       V2_QPC_BYTE_168_SQ_SHIFT_BAK_M,
+		       V2_QPC_BYTE_168_SQ_SHIFT_BAK_S,
+		       ilog2((unsigned int)hr_qp->sq.wqe_cnt));
+	roce_set_field(qpc_mask->byte_168_irrl_idx,
+		       V2_QPC_BYTE_168_SQ_SHIFT_BAK_M,
+		       V2_QPC_BYTE_168_SQ_SHIFT_BAK_S, 0);
+}
+
+static int modify_qp_init_to_rtr(struct ib_qp *ibqp,
+				 const struct ib_qp_attr *attr, int attr_mask,
+				 struct hns_roce_v2_qp_context *context,
+				 struct hns_roce_v2_qp_context *qpc_mask)
+{
+	const struct ib_global_route *grh = rdma_ah_read_grh(&attr->ah_attr);
+	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
+	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
+	struct device *dev = hr_dev->dev;
+	dma_addr_t dma_handle_2;
+	dma_addr_t dma_handle;
+	u32 page_size;
+	u8 port_num;
+	u64 *mtts_2;
+	u64 *mtts;
+	u8 *dmac;
+	u8 *smac;
+	int port;
+
+	/* Search qp buf's mtts */
+	mtts = hns_roce_table_find(hr_dev, &hr_dev->mr_table.mtt_table,
+				   hr_qp->mtt.first_seg, &dma_handle);
+	if (!mtts) {
+		dev_err(dev, "qp buf pa find failed\n");
+		return -EINVAL;
+	}
+
+	/* Search IRRL's mtts */
+	mtts_2 = hns_roce_table_find(hr_dev, &hr_dev->qp_table.irrl_table,
+				     hr_qp->qpn, &dma_handle_2);
+	if (!mtts_2) {
+		dev_err(dev, "qp irrl_table find failed\n");
+		return -EINVAL;
+	}
+
+	if ((attr_mask & IB_QP_ALT_PATH) || (attr_mask & IB_QP_ACCESS_FLAGS) ||
+	    (attr_mask & IB_QP_PKEY_INDEX) || (attr_mask & IB_QP_QKEY)) {
+		dev_err(dev, "INIT2RTR attr_mask (0x%x) error\n", attr_mask);
+		return -EINVAL;
+	}
+
+	dmac = (u8 *)attr->ah_attr.roce.dmac;
+	context->wqe_sge_ba = (u32)(dma_handle >> 3);
+	qpc_mask->wqe_sge_ba = 0;
+
+	/*
+	 * In v2 engine, software pass context and context mask to hardware
+	 * when modifying qp. If software need modify some fields in context,
+	 * we should set all bits of the relevant fields in context mask to
+	 * 0 at the same time, else set them to 0x1.
+	 */
+	roce_set_field(context->byte_12_sq_hop, V2_QPC_BYTE_12_WQE_SGE_BA_M,
+		       V2_QPC_BYTE_12_WQE_SGE_BA_S, dma_handle >> (32 + 3));
+	roce_set_field(qpc_mask->byte_12_sq_hop, V2_QPC_BYTE_12_WQE_SGE_BA_M,
+		       V2_QPC_BYTE_12_WQE_SGE_BA_S, 0);
+
+	roce_set_field(context->byte_12_sq_hop, V2_QPC_BYTE_12_SQ_HOP_NUM_M,
+		       V2_QPC_BYTE_12_SQ_HOP_NUM_S,
+		       hr_dev->caps.mtt_hop_num == HNS_ROCE_HOP_NUM_0 ?
+		       0 : hr_dev->caps.mtt_hop_num);
+	roce_set_field(qpc_mask->byte_12_sq_hop, V2_QPC_BYTE_12_SQ_HOP_NUM_M,
+		       V2_QPC_BYTE_12_SQ_HOP_NUM_S, 0);
+
+	roce_set_field(context->byte_20_smac_sgid_idx,
+		       V2_QPC_BYTE_20_SGE_HOP_NUM_M,
+		       V2_QPC_BYTE_20_SGE_HOP_NUM_S,
+		       hr_qp->sq.max_gs > 2 ? hr_dev->caps.mtt_hop_num : 0);
+	roce_set_field(qpc_mask->byte_20_smac_sgid_idx,
+		       V2_QPC_BYTE_20_SGE_HOP_NUM_M,
+		       V2_QPC_BYTE_20_SGE_HOP_NUM_S, 0);
+
+	roce_set_field(context->byte_20_smac_sgid_idx,
+		       V2_QPC_BYTE_20_RQ_HOP_NUM_M,
+		       V2_QPC_BYTE_20_RQ_HOP_NUM_S,
+		       hr_dev->caps.mtt_hop_num == HNS_ROCE_HOP_NUM_0 ?
+		       0 : hr_dev->caps.mtt_hop_num);
+	roce_set_field(qpc_mask->byte_20_smac_sgid_idx,
+		       V2_QPC_BYTE_20_RQ_HOP_NUM_M,
+		       V2_QPC_BYTE_20_RQ_HOP_NUM_S, 0);
+
+	roce_set_field(context->byte_16_buf_ba_pg_sz,
+		       V2_QPC_BYTE_16_WQE_SGE_BA_PG_SZ_M,
+		       V2_QPC_BYTE_16_WQE_SGE_BA_PG_SZ_S,
+		       hr_dev->caps.mtt_ba_pg_sz);
+	roce_set_field(qpc_mask->byte_16_buf_ba_pg_sz,
+		       V2_QPC_BYTE_16_WQE_SGE_BA_PG_SZ_M,
+		       V2_QPC_BYTE_16_WQE_SGE_BA_PG_SZ_S, 0);
+
+	roce_set_field(context->byte_16_buf_ba_pg_sz,
+		       V2_QPC_BYTE_16_WQE_SGE_BUF_PG_SZ_M,
+		       V2_QPC_BYTE_16_WQE_SGE_BUF_PG_SZ_S,
+		       hr_dev->caps.mtt_buf_pg_sz);
+	roce_set_field(qpc_mask->byte_16_buf_ba_pg_sz,
+		       V2_QPC_BYTE_16_WQE_SGE_BUF_PG_SZ_M,
+		       V2_QPC_BYTE_16_WQE_SGE_BUF_PG_SZ_S, 0);
+
+	roce_set_field(context->byte_80_rnr_rx_cqn,
+		       V2_QPC_BYTE_80_MIN_RNR_TIME_M,
+		       V2_QPC_BYTE_80_MIN_RNR_TIME_S, attr->min_rnr_timer);
+	roce_set_field(qpc_mask->byte_80_rnr_rx_cqn,
+		       V2_QPC_BYTE_80_MIN_RNR_TIME_M,
+		       V2_QPC_BYTE_80_MIN_RNR_TIME_S, 0);
+
+	page_size = 1 << (hr_dev->caps.mtt_buf_pg_sz + PAGE_SHIFT);
+	context->rq_cur_blk_addr = (u32)(mtts[hr_qp->rq.offset / page_size]
+				    >> PAGE_ADDR_SHIFT);
+	qpc_mask->rq_cur_blk_addr = 0;
+
+	roce_set_field(context->byte_92_srq_info,
+		       V2_QPC_BYTE_92_RQ_CUR_BLK_ADDR_M,
+		       V2_QPC_BYTE_92_RQ_CUR_BLK_ADDR_S,
+		       mtts[hr_qp->rq.offset / page_size]
+		       >> (32 + PAGE_ADDR_SHIFT));
+	roce_set_field(qpc_mask->byte_92_srq_info,
+		       V2_QPC_BYTE_92_RQ_CUR_BLK_ADDR_M,
+		       V2_QPC_BYTE_92_RQ_CUR_BLK_ADDR_S, 0);
+
+	context->rq_nxt_blk_addr = (u32)(mtts[hr_qp->rq.offset / page_size + 1]
+				    >> PAGE_ADDR_SHIFT);
+	qpc_mask->rq_nxt_blk_addr = 0;
+
+	roce_set_field(context->byte_104_rq_sge,
+		       V2_QPC_BYTE_104_RQ_NXT_BLK_ADDR_M,
+		       V2_QPC_BYTE_104_RQ_NXT_BLK_ADDR_S,
+		       mtts[hr_qp->rq.offset / page_size + 1]
+		       >> (32 + PAGE_ADDR_SHIFT));
+	roce_set_field(qpc_mask->byte_104_rq_sge,
+		       V2_QPC_BYTE_104_RQ_NXT_BLK_ADDR_M,
+		       V2_QPC_BYTE_104_RQ_NXT_BLK_ADDR_S, 0);
+
+	roce_set_field(context->byte_108_rx_reqepsn,
+		       V2_QPC_BYTE_108_RX_REQ_EPSN_M,
+		       V2_QPC_BYTE_108_RX_REQ_EPSN_S, attr->rq_psn);
+	roce_set_field(qpc_mask->byte_108_rx_reqepsn,
+		       V2_QPC_BYTE_108_RX_REQ_EPSN_M,
+		       V2_QPC_BYTE_108_RX_REQ_EPSN_S, 0);
+
+	context->irrl_ba = (u32)dma_handle_2;
+	qpc_mask->irrl_ba = 0;
+	roce_set_field(context->byte_208_irrl, V2_QPC_BYTE_208_IRRL_BA_M,
+		       V2_QPC_BYTE_208_IRRL_BA_S,
+		      (dma_handle_2 >> 32) & V2_QPC_BYTE_208_IRRL_BA_M);
+	roce_set_field(qpc_mask->byte_208_irrl, V2_QPC_BYTE_208_IRRL_BA_M,
+		       V2_QPC_BYTE_208_IRRL_BA_S, 0);
+
+	roce_set_bit(context->byte_208_irrl, V2_QPC_BYTE_208_RMT_E2E_S, 1);
+	roce_set_bit(qpc_mask->byte_208_irrl, V2_QPC_BYTE_208_RMT_E2E_S, 0);
+
+	roce_set_bit(context->byte_252_err_txcqn, V2_QPC_BYTE_252_SIG_TYPE_S,
+		     hr_qp->sq_signal_bits);
+	roce_set_bit(qpc_mask->byte_252_err_txcqn, V2_QPC_BYTE_252_SIG_TYPE_S,
+		     0);
+
+	port = (attr_mask & IB_QP_PORT) ? (attr->port_num - 1) : hr_qp->port;
+
+	smac = (u8 *)hr_dev->dev_addr[port];
+	/* when dmac equals smac or loop_idc is 1, it should loopback */
+	if (ether_addr_equal_unaligned(dmac, smac) ||
+	    hr_dev->loop_idc == 0x1) {
+		roce_set_bit(context->byte_28_at_fl, V2_QPC_BYTE_28_LBI_S, 1);
+		roce_set_bit(qpc_mask->byte_28_at_fl, V2_QPC_BYTE_28_LBI_S, 0);
+	}
+
+	roce_set_field(context->byte_140_raq, V2_QPC_BYTE_140_RR_MAX_M,
+		       V2_QPC_BYTE_140_RR_MAX_S,
+		       ilog2((unsigned int)attr->max_dest_rd_atomic));
+	roce_set_field(qpc_mask->byte_140_raq, V2_QPC_BYTE_140_RR_MAX_M,
+		       V2_QPC_BYTE_140_RR_MAX_S, 0);
+
+	roce_set_field(context->byte_56_dqpn_err, V2_QPC_BYTE_56_DQPN_M,
+		       V2_QPC_BYTE_56_DQPN_S, attr->dest_qp_num);
+	roce_set_field(qpc_mask->byte_56_dqpn_err, V2_QPC_BYTE_56_DQPN_M,
+		       V2_QPC_BYTE_56_DQPN_S, 0);
+
+	/* Configure GID index */
+	port_num = rdma_ah_get_port_num(&attr->ah_attr);
+	roce_set_field(context->byte_20_smac_sgid_idx,
+		       V2_QPC_BYTE_20_SGID_IDX_M,
+		       V2_QPC_BYTE_20_SGID_IDX_S,
+		       hns_get_gid_index(hr_dev, port_num - 1,
+					 grh->sgid_index));
+	roce_set_field(qpc_mask->byte_20_smac_sgid_idx,
+		       V2_QPC_BYTE_20_SGID_IDX_M,
+		       V2_QPC_BYTE_20_SGID_IDX_S, 0);
+	memcpy(&(context->dmac), dmac, 4);
+	roce_set_field(context->byte_52_udpspn_dmac, V2_QPC_BYTE_52_DMAC_M,
+		       V2_QPC_BYTE_52_DMAC_S, *((u16 *)(&dmac[4])));
+	qpc_mask->dmac = 0;
+	roce_set_field(qpc_mask->byte_52_udpspn_dmac, V2_QPC_BYTE_52_DMAC_M,
+		       V2_QPC_BYTE_52_DMAC_S, 0);
+
+	roce_set_field(context->byte_56_dqpn_err, V2_QPC_BYTE_56_LP_PKTN_INI_M,
+		       V2_QPC_BYTE_56_LP_PKTN_INI_S, 4);
+	roce_set_field(qpc_mask->byte_56_dqpn_err, V2_QPC_BYTE_56_LP_PKTN_INI_M,
+		       V2_QPC_BYTE_56_LP_PKTN_INI_S, 0);
+
+	roce_set_field(context->byte_24_mtu_tc, V2_QPC_BYTE_24_HOP_LIMIT_M,
+		       V2_QPC_BYTE_24_HOP_LIMIT_S, grh->hop_limit);
+	roce_set_field(qpc_mask->byte_24_mtu_tc, V2_QPC_BYTE_24_HOP_LIMIT_M,
+		       V2_QPC_BYTE_24_HOP_LIMIT_S, 0);
+
+	roce_set_field(context->byte_28_at_fl, V2_QPC_BYTE_28_FL_M,
+		       V2_QPC_BYTE_28_FL_S, grh->flow_label);
+	roce_set_field(qpc_mask->byte_28_at_fl, V2_QPC_BYTE_28_FL_M,
+		       V2_QPC_BYTE_28_FL_S, 0);
+
+	roce_set_field(context->byte_24_mtu_tc, V2_QPC_BYTE_24_TC_M,
+		       V2_QPC_BYTE_24_TC_S, grh->traffic_class);
+	roce_set_field(qpc_mask->byte_24_mtu_tc, V2_QPC_BYTE_24_TC_M,
+		       V2_QPC_BYTE_24_TC_S, 0);
+
+	roce_set_field(context->byte_24_mtu_tc, V2_QPC_BYTE_24_MTU_M,
+		       V2_QPC_BYTE_24_MTU_S, attr->path_mtu);
+	roce_set_field(qpc_mask->byte_24_mtu_tc, V2_QPC_BYTE_24_MTU_M,
+		       V2_QPC_BYTE_24_MTU_S, 0);
+
+	memcpy(context->dgid, grh->dgid.raw, sizeof(grh->dgid.raw));
+	memset(qpc_mask->dgid, 0, sizeof(grh->dgid.raw));
+
+	roce_set_field(context->byte_84_rq_ci_pi,
+		       V2_QPC_BYTE_84_RQ_PRODUCER_IDX_M,
+		       V2_QPC_BYTE_84_RQ_PRODUCER_IDX_S, hr_qp->rq.head);
+	roce_set_field(qpc_mask->byte_84_rq_ci_pi,
+		       V2_QPC_BYTE_84_RQ_PRODUCER_IDX_M,
+		       V2_QPC_BYTE_84_RQ_PRODUCER_IDX_S, 0);
+
+	roce_set_field(qpc_mask->byte_84_rq_ci_pi,
+		       V2_QPC_BYTE_84_RQ_CONSUMER_IDX_M,
+		       V2_QPC_BYTE_84_RQ_CONSUMER_IDX_S, 0);
+	roce_set_bit(qpc_mask->byte_108_rx_reqepsn,
+		     V2_QPC_BYTE_108_RX_REQ_PSN_ERR_S, 0);
+	roce_set_field(qpc_mask->byte_96_rx_reqmsn, V2_QPC_BYTE_96_RX_REQ_MSN_M,
+		       V2_QPC_BYTE_96_RX_REQ_MSN_S, 0);
+	roce_set_field(qpc_mask->byte_108_rx_reqepsn,
+		       V2_QPC_BYTE_108_RX_REQ_LAST_OPTYPE_M,
+		       V2_QPC_BYTE_108_RX_REQ_LAST_OPTYPE_S, 0);
+
+	context->rq_rnr_timer = 0;
+	qpc_mask->rq_rnr_timer = 0;
+
+	roce_set_field(context->byte_152_raq, V2_QPC_BYTE_152_RAQ_PSN_M,
+		       V2_QPC_BYTE_152_RAQ_PSN_S, attr->rq_psn - 1);
+	roce_set_field(qpc_mask->byte_152_raq, V2_QPC_BYTE_152_RAQ_PSN_M,
+		       V2_QPC_BYTE_152_RAQ_PSN_S, 0);
+
+	roce_set_field(qpc_mask->byte_132_trrl, V2_QPC_BYTE_132_TRRL_HEAD_MAX_M,
+		       V2_QPC_BYTE_132_TRRL_HEAD_MAX_S, 0);
+	roce_set_field(qpc_mask->byte_132_trrl, V2_QPC_BYTE_132_TRRL_TAIL_MAX_M,
+		       V2_QPC_BYTE_132_TRRL_TAIL_MAX_S, 0);
+
+	roce_set_field(context->byte_168_irrl_idx,
+		       V2_QPC_BYTE_168_LP_SGEN_INI_M,
+		       V2_QPC_BYTE_168_LP_SGEN_INI_S, 3);
+	roce_set_field(qpc_mask->byte_168_irrl_idx,
+		       V2_QPC_BYTE_168_LP_SGEN_INI_M,
+		       V2_QPC_BYTE_168_LP_SGEN_INI_S, 0);
+
+	roce_set_field(context->byte_208_irrl, V2_QPC_BYTE_208_SR_MAX_M,
+		       V2_QPC_BYTE_208_SR_MAX_S,
+		       ilog2((unsigned int)attr->max_rd_atomic));
+	roce_set_field(qpc_mask->byte_208_irrl, V2_QPC_BYTE_208_SR_MAX_M,
+		       V2_QPC_BYTE_208_SR_MAX_S, 0);
+
+	roce_set_field(context->byte_28_at_fl, V2_QPC_BYTE_28_SL_M,
+		       V2_QPC_BYTE_28_SL_S, rdma_ah_get_sl(&attr->ah_attr));
+	roce_set_field(qpc_mask->byte_28_at_fl, V2_QPC_BYTE_28_SL_M,
+		       V2_QPC_BYTE_28_SL_S, 0);
+	hr_qp->sl = rdma_ah_get_sl(&attr->ah_attr);
+
+	return 0;
+}
+
+static int modify_qp_rtr_to_rts(struct ib_qp *ibqp,
+				const struct ib_qp_attr *attr, int attr_mask,
+				struct hns_roce_v2_qp_context *context,
+				struct hns_roce_v2_qp_context *qpc_mask)
+{
+	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
+	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
+	struct device *dev = hr_dev->dev;
+	dma_addr_t dma_handle;
+	u64 *mtts;
+
+	/* Search qp buf's mtts */
+	mtts = hns_roce_table_find(hr_dev, &hr_dev->mr_table.mtt_table,
+				   hr_qp->mtt.first_seg, &dma_handle);
+	if (!mtts) {
+		dev_err(dev, "qp buf pa find failed\n");
+		return -EINVAL;
+	}
+
+	/* If exist optional param, return error */
+	if ((attr_mask & IB_QP_ALT_PATH) || (attr_mask & IB_QP_ACCESS_FLAGS) ||
+	    (attr_mask & IB_QP_QKEY) || (attr_mask & IB_QP_PATH_MIG_STATE) ||
+	    (attr_mask & IB_QP_CUR_STATE) ||
+	    (attr_mask & IB_QP_MIN_RNR_TIMER)) {
+		dev_err(dev, "RTR2RTS attr_mask (0x%x)error\n", attr_mask);
+		return -EINVAL;
+	}
+
+	/*
+	 * In v2 engine, software pass context and context mask to hardware
+	 * when modifying qp. If software need modify some fields in context,
+	 * we should set all bits of the relevant fields in context mask to
+	 * 0 at the same time, else set them to 0x1.
+	 */
+	roce_set_field(context->byte_60_qpst_mapid,
+		       V2_QPC_BYTE_60_RTY_NUM_INI_BAK_M,
+		       V2_QPC_BYTE_60_RTY_NUM_INI_BAK_S, attr->retry_cnt);
+	roce_set_field(qpc_mask->byte_60_qpst_mapid,
+		       V2_QPC_BYTE_60_RTY_NUM_INI_BAK_M,
+		       V2_QPC_BYTE_60_RTY_NUM_INI_BAK_S, 0);
+
+	context->sq_cur_blk_addr = (u32)(mtts[0] >> PAGE_ADDR_SHIFT);
+	roce_set_field(context->byte_168_irrl_idx,
+		       V2_QPC_BYTE_168_SQ_CUR_BLK_ADDR_M,
+		       V2_QPC_BYTE_168_SQ_CUR_BLK_ADDR_S,
+		       mtts[0] >> (32 + PAGE_ADDR_SHIFT));
+	qpc_mask->sq_cur_blk_addr = 0;
+	roce_set_field(qpc_mask->byte_168_irrl_idx,
+		       V2_QPC_BYTE_168_SQ_CUR_BLK_ADDR_M,
+		       V2_QPC_BYTE_168_SQ_CUR_BLK_ADDR_S, 0);
+
+	context->rx_sq_cur_blk_addr = (u32)(mtts[0] >> PAGE_ADDR_SHIFT);
+	roce_set_field(context->byte_232_irrl_sge,
+		       V2_QPC_BYTE_232_RX_SQ_CUR_BLK_ADDR_M,
+		       V2_QPC_BYTE_232_RX_SQ_CUR_BLK_ADDR_S,
+		       mtts[0] >> (32 + PAGE_ADDR_SHIFT));
+	qpc_mask->rx_sq_cur_blk_addr = 0;
+	roce_set_field(qpc_mask->byte_232_irrl_sge,
+		       V2_QPC_BYTE_232_RX_SQ_CUR_BLK_ADDR_M,
+		       V2_QPC_BYTE_232_RX_SQ_CUR_BLK_ADDR_S, 0);
+
+	/*
+	 * Set some fields in context to zero, Because the default values
+	 * of all fields in context are zero, we need not set them to 0 again.
+	 * but we should set the relevant fields of context mask to 0.
+	 */
+	roce_set_field(qpc_mask->byte_232_irrl_sge,
+		       V2_QPC_BYTE_232_IRRL_SGE_IDX_M,
+		       V2_QPC_BYTE_232_IRRL_SGE_IDX_S, 0);
+
+	roce_set_field(qpc_mask->byte_240_irrl_tail,
+		       V2_QPC_BYTE_240_RX_ACK_MSN_M,
+		       V2_QPC_BYTE_240_RX_ACK_MSN_S, 0);
+
+	roce_set_field(context->byte_244_rnr_rxack,
+		       V2_QPC_BYTE_244_RX_ACK_EPSN_M,
+		       V2_QPC_BYTE_244_RX_ACK_EPSN_S, attr->sq_psn);
+	roce_set_field(qpc_mask->byte_244_rnr_rxack,
+		       V2_QPC_BYTE_244_RX_ACK_EPSN_M,
+		       V2_QPC_BYTE_244_RX_ACK_EPSN_S, 0);
+
+	roce_set_field(qpc_mask->byte_248_ack_psn,
+		       V2_QPC_BYTE_248_ACK_LAST_OPTYPE_M,
+		       V2_QPC_BYTE_248_ACK_LAST_OPTYPE_S, 0);
+	roce_set_bit(qpc_mask->byte_248_ack_psn,
+		     V2_QPC_BYTE_248_IRRL_PSN_VLD_S, 0);
+	roce_set_field(qpc_mask->byte_248_ack_psn,
+		       V2_QPC_BYTE_248_IRRL_PSN_M,
+		       V2_QPC_BYTE_248_IRRL_PSN_S, 0);
+
+	roce_set_field(qpc_mask->byte_240_irrl_tail,
+		       V2_QPC_BYTE_240_IRRL_TAIL_REAL_M,
+		       V2_QPC_BYTE_240_IRRL_TAIL_REAL_S, 0);
+
+	roce_set_field(context->byte_220_retry_psn_msn,
+		       V2_QPC_BYTE_220_RETRY_MSG_PSN_M,
+		       V2_QPC_BYTE_220_RETRY_MSG_PSN_S, attr->sq_psn);
+	roce_set_field(qpc_mask->byte_220_retry_psn_msn,
+		       V2_QPC_BYTE_220_RETRY_MSG_PSN_M,
+		       V2_QPC_BYTE_220_RETRY_MSG_PSN_S, 0);
+
+	roce_set_field(context->byte_224_retry_msg,
+		       V2_QPC_BYTE_224_RETRY_MSG_PSN_M,
+		       V2_QPC_BYTE_224_RETRY_MSG_PSN_S, attr->sq_psn >> 16);
+	roce_set_field(qpc_mask->byte_224_retry_msg,
+		       V2_QPC_BYTE_224_RETRY_MSG_PSN_M,
+		       V2_QPC_BYTE_224_RETRY_MSG_PSN_S, 0);
+
+	roce_set_field(context->byte_224_retry_msg,
+		       V2_QPC_BYTE_224_RETRY_MSG_FPKT_PSN_M,
+		       V2_QPC_BYTE_224_RETRY_MSG_FPKT_PSN_S, attr->sq_psn);
+	roce_set_field(qpc_mask->byte_224_retry_msg,
+		       V2_QPC_BYTE_224_RETRY_MSG_FPKT_PSN_M,
+		       V2_QPC_BYTE_224_RETRY_MSG_FPKT_PSN_S, 0);
+
+	roce_set_field(qpc_mask->byte_220_retry_psn_msn,
+		       V2_QPC_BYTE_220_RETRY_MSG_MSN_M,
+		       V2_QPC_BYTE_220_RETRY_MSG_MSN_S, 0);
+
+	roce_set_bit(qpc_mask->byte_248_ack_psn,
+		     V2_QPC_BYTE_248_RNR_RETRY_FLAG_S, 0);
+
+	roce_set_field(qpc_mask->byte_212_lsn, V2_QPC_BYTE_212_CHECK_FLG_M,
+		       V2_QPC_BYTE_212_CHECK_FLG_S, 0);
+
+	roce_set_field(context->byte_212_lsn, V2_QPC_BYTE_212_RETRY_CNT_M,
+		       V2_QPC_BYTE_212_RETRY_CNT_S, attr->retry_cnt);
+	roce_set_field(qpc_mask->byte_212_lsn, V2_QPC_BYTE_212_RETRY_CNT_M,
+		       V2_QPC_BYTE_212_RETRY_CNT_S, 0);
+
+	roce_set_field(context->byte_212_lsn, V2_QPC_BYTE_212_RETRY_NUM_INIT_M,
+		       V2_QPC_BYTE_212_RETRY_NUM_INIT_S, attr->retry_cnt);
+	roce_set_field(qpc_mask->byte_212_lsn, V2_QPC_BYTE_212_RETRY_NUM_INIT_M,
+		       V2_QPC_BYTE_212_RETRY_NUM_INIT_S, 0);
+
+	roce_set_field(context->byte_244_rnr_rxack,
+		       V2_QPC_BYTE_244_RNR_NUM_INIT_M,
+		       V2_QPC_BYTE_244_RNR_NUM_INIT_S, attr->rnr_retry);
+	roce_set_field(qpc_mask->byte_244_rnr_rxack,
+		       V2_QPC_BYTE_244_RNR_NUM_INIT_M,
+		       V2_QPC_BYTE_244_RNR_NUM_INIT_S, 0);
+
+	roce_set_field(context->byte_244_rnr_rxack, V2_QPC_BYTE_244_RNR_CNT_M,
+		       V2_QPC_BYTE_244_RNR_CNT_S, attr->rnr_retry);
+	roce_set_field(qpc_mask->byte_244_rnr_rxack, V2_QPC_BYTE_244_RNR_CNT_M,
+		       V2_QPC_BYTE_244_RNR_CNT_S, 0);
+
+	roce_set_field(context->byte_212_lsn, V2_QPC_BYTE_212_LSN_M,
+		       V2_QPC_BYTE_212_LSN_S, 0x100);
+	roce_set_field(qpc_mask->byte_212_lsn, V2_QPC_BYTE_212_LSN_M,
+		       V2_QPC_BYTE_212_LSN_S, 0);
+
+	if (attr->timeout < 0xf)
+		roce_set_field(context->byte_28_at_fl, V2_QPC_BYTE_28_AT_M,
+			       V2_QPC_BYTE_28_AT_S, 0xf);
+	else
+		roce_set_field(context->byte_28_at_fl, V2_QPC_BYTE_28_AT_M,
+			       V2_QPC_BYTE_28_AT_S, attr->timeout);
+
+	roce_set_field(qpc_mask->byte_28_at_fl, V2_QPC_BYTE_28_AT_M,
+		       V2_QPC_BYTE_28_AT_S, 0);
+
+	roce_set_field(context->byte_28_at_fl, V2_QPC_BYTE_28_SL_M,
+		       V2_QPC_BYTE_28_SL_S,
+		       rdma_ah_get_sl(&attr->ah_attr));
+	roce_set_field(qpc_mask->byte_28_at_fl, V2_QPC_BYTE_28_SL_M,
+		       V2_QPC_BYTE_28_SL_S, 0);
+	hr_qp->sl = rdma_ah_get_sl(&attr->ah_attr);
+
+	roce_set_field(context->byte_172_sq_psn, V2_QPC_BYTE_172_SQ_CUR_PSN_M,
+		       V2_QPC_BYTE_172_SQ_CUR_PSN_S, attr->sq_psn);
+	roce_set_field(qpc_mask->byte_172_sq_psn, V2_QPC_BYTE_172_SQ_CUR_PSN_M,
+		       V2_QPC_BYTE_172_SQ_CUR_PSN_S, 0);
+
+	roce_set_field(qpc_mask->byte_196_sq_psn, V2_QPC_BYTE_196_IRRL_HEAD_M,
+		       V2_QPC_BYTE_196_IRRL_HEAD_S, 0);
+	roce_set_field(context->byte_196_sq_psn, V2_QPC_BYTE_196_SQ_MAX_PSN_M,
+		       V2_QPC_BYTE_196_SQ_MAX_PSN_S, attr->sq_psn);
+	roce_set_field(qpc_mask->byte_196_sq_psn, V2_QPC_BYTE_196_SQ_MAX_PSN_M,
+		       V2_QPC_BYTE_196_SQ_MAX_PSN_S, 0);
+
+	return 0;
+}
+
+static int hns_roce_v2_modify_qp(struct ib_qp *ibqp,
+				 const struct ib_qp_attr *attr,
+				 int attr_mask, enum ib_qp_state cur_state,
+				 enum ib_qp_state new_state)
+{
+	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
+	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
+	struct hns_roce_v2_qp_context *context;
+	struct hns_roce_v2_qp_context *qpc_mask;
+	struct device *dev = hr_dev->dev;
+	int ret = -EINVAL;
+
+	context = kzalloc(2 * sizeof(*context), GFP_KERNEL);
+	if (!context)
+		return -ENOMEM;
+
+	qpc_mask = context + 1;
+	/*
+	 * In v2 engine, software pass context and context mask to hardware
+	 * when modifying qp. If software need modify some fields in context,
+	 * we should set all bits of the relevant fields in context mask to
+	 * 0 at the same time, else set them to 0x1.
+	 */
+	memset(qpc_mask, 0xff, sizeof(*qpc_mask));
+	if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) {
+		modify_qp_reset_to_init(ibqp, attr, context, qpc_mask);
+	} else if (cur_state == IB_QPS_INIT && new_state == IB_QPS_INIT) {
+		modify_qp_init_to_init(ibqp, attr, attr_mask, context,
+				       qpc_mask);
+	} else if (cur_state == IB_QPS_INIT && new_state == IB_QPS_RTR) {
+		ret = modify_qp_init_to_rtr(ibqp, attr, attr_mask, context,
+					    qpc_mask);
+		if (ret)
+			goto out;
+	} else if (cur_state == IB_QPS_RTR && new_state == IB_QPS_RTS) {
+		ret = modify_qp_rtr_to_rts(ibqp, attr, attr_mask, context,
+					   qpc_mask);
+		if (ret)
+			goto out;
+	} else if ((cur_state == IB_QPS_RTS && new_state == IB_QPS_RTS) ||
+		   (cur_state == IB_QPS_SQE && new_state == IB_QPS_RTS) ||
+		   (cur_state == IB_QPS_RTS && new_state == IB_QPS_SQD) ||
+		   (cur_state == IB_QPS_SQD && new_state == IB_QPS_SQD) ||
+		   (cur_state == IB_QPS_SQD && new_state == IB_QPS_RTS) ||
+		   (cur_state == IB_QPS_INIT && new_state == IB_QPS_RESET) ||
+		   (cur_state == IB_QPS_RTR && new_state == IB_QPS_RESET) ||
+		   (cur_state == IB_QPS_RTS && new_state == IB_QPS_RESET) ||
+		   (cur_state == IB_QPS_ERR && new_state == IB_QPS_RESET) ||
+		   (cur_state == IB_QPS_INIT && new_state == IB_QPS_ERR) ||
+		   (cur_state == IB_QPS_RTR && new_state == IB_QPS_ERR) ||
+		   (cur_state == IB_QPS_RTS && new_state == IB_QPS_ERR) ||
+		   (cur_state == IB_QPS_SQD && new_state == IB_QPS_ERR) ||
+		   (cur_state == IB_QPS_SQE && new_state == IB_QPS_ERR)) {
+		/* Nothing */
+		;
+	} else {
+		dev_err(dev, "Illegal state for QP!\n");
+		goto out;
+	}
+
+	/* Every status migrate must change state */
+	roce_set_field(context->byte_60_qpst_mapid, V2_QPC_BYTE_60_QP_ST_M,
+		       V2_QPC_BYTE_60_QP_ST_S, new_state);
+	roce_set_field(qpc_mask->byte_60_qpst_mapid, V2_QPC_BYTE_60_QP_ST_M,
+		       V2_QPC_BYTE_60_QP_ST_S, 0);
+
+	/* SW pass context to HW */
+	ret = hns_roce_v2_qp_modify(hr_dev, &hr_qp->mtt, cur_state, new_state,
+				    context, hr_qp);
+	if (ret) {
+		dev_err(dev, "hns_roce_qp_modify failed(%d)\n", ret);
+		goto out;
+	}
+
+	hr_qp->state = new_state;
+
+	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
+		hr_qp->resp_depth = attr->max_dest_rd_atomic;
+	if (attr_mask & IB_QP_PORT) {
+		hr_qp->port = attr->port_num - 1;
+		hr_qp->phy_port = hr_dev->iboe.phy_port[hr_qp->port];
+	}
+
+	if (new_state == IB_QPS_RESET && !ibqp->uobject) {
+		hns_roce_v2_cq_clean(to_hr_cq(ibqp->recv_cq), hr_qp->qpn,
+				     ibqp->srq ? to_hr_srq(ibqp->srq) : NULL);
+		if (ibqp->send_cq != ibqp->recv_cq)
+			hns_roce_v2_cq_clean(to_hr_cq(ibqp->send_cq),
+					     hr_qp->qpn, NULL);
+
+		hr_qp->rq.head = 0;
+		hr_qp->rq.tail = 0;
+		hr_qp->sq.head = 0;
+		hr_qp->sq.tail = 0;
+		hr_qp->sq_next_wqe = 0;
+		hr_qp->next_sge = 0;
+	}
+
+out:
+	kfree(context);
+	return ret;
+}
+
+static inline enum ib_qp_state to_ib_qp_st(enum hns_roce_v2_qp_state state)
+{
+	switch (state) {
+	case HNS_ROCE_QP_ST_RST:	return IB_QPS_RESET;
+	case HNS_ROCE_QP_ST_INIT:	return IB_QPS_INIT;
+	case HNS_ROCE_QP_ST_RTR:	return IB_QPS_RTR;
+	case HNS_ROCE_QP_ST_RTS:	return IB_QPS_RTS;
+	case HNS_ROCE_QP_ST_SQ_DRAINING:
+	case HNS_ROCE_QP_ST_SQD:	return IB_QPS_SQD;
+	case HNS_ROCE_QP_ST_SQER:	return IB_QPS_SQE;
+	case HNS_ROCE_QP_ST_ERR:	return IB_QPS_ERR;
+	default:			return -1;
+	}
+}
+
+static int hns_roce_v2_query_qpc(struct hns_roce_dev *hr_dev,
+				 struct hns_roce_qp *hr_qp,
+				 struct hns_roce_v2_qp_context *hr_context)
+{
+	struct hns_roce_cmd_mailbox *mailbox;
+	int ret;
+
+	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
+	if (IS_ERR(mailbox))
+		return PTR_ERR(mailbox);
+
+	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, hr_qp->qpn, 0,
+				HNS_ROCE_CMD_QUERY_QPC,
+				HNS_ROCE_CMD_TIMEOUT_MSECS);
+	if (ret) {
+		dev_err(hr_dev->dev, "QUERY QP cmd process error\n");
+		goto out;
+	}
+
+	memcpy(hr_context, mailbox->buf, sizeof(*hr_context));
+
+out:
+	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
+	return ret;
+}
+
+static int hns_roce_v2_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
+				int qp_attr_mask,
+				struct ib_qp_init_attr *qp_init_attr)
+{
+	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
+	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
+	struct hns_roce_v2_qp_context *context;
+	struct device *dev = hr_dev->dev;
+	int tmp_qp_state;
+	int state;
+	int ret;
+
+	context = kzalloc(sizeof(*context), GFP_KERNEL);
+	if (!context)
+		return -ENOMEM;
+
+	memset(qp_attr, 0, sizeof(*qp_attr));
+	memset(qp_init_attr, 0, sizeof(*qp_init_attr));
+
+	mutex_lock(&hr_qp->mutex);
+
+	if (hr_qp->state == IB_QPS_RESET) {
+		qp_attr->qp_state = IB_QPS_RESET;
+		goto done;
+	}
+
+	ret = hns_roce_v2_query_qpc(hr_dev, hr_qp, context);
+	if (ret) {
+		dev_err(dev, "query qpc error\n");
+		ret = -EINVAL;
+		goto out;
+	}
+
+	state = roce_get_field(context->byte_60_qpst_mapid,
+			       V2_QPC_BYTE_60_QP_ST_M, V2_QPC_BYTE_60_QP_ST_S);
+	tmp_qp_state = to_ib_qp_st((enum hns_roce_v2_qp_state)state);
+	if (tmp_qp_state == -1) {
+		dev_err(dev, "Illegal ib_qp_state\n");
+		ret = -EINVAL;
+		goto out;
+	}
+	hr_qp->state = (u8)tmp_qp_state;
+	qp_attr->qp_state = (enum ib_qp_state)hr_qp->state;
+	qp_attr->path_mtu = (enum ib_mtu)roce_get_field(context->byte_24_mtu_tc,
+							V2_QPC_BYTE_24_MTU_M,
+							V2_QPC_BYTE_24_MTU_S);
+	qp_attr->path_mig_state = IB_MIG_ARMED;
+	if (hr_qp->ibqp.qp_type == IB_QPT_UD)
+		qp_attr->qkey = V2_QKEY_VAL;
+
+	qp_attr->rq_psn = roce_get_field(context->byte_108_rx_reqepsn,
+					 V2_QPC_BYTE_108_RX_REQ_EPSN_M,
+					 V2_QPC_BYTE_108_RX_REQ_EPSN_S);
+	qp_attr->sq_psn = (u32)roce_get_field(context->byte_172_sq_psn,
+					      V2_QPC_BYTE_172_SQ_CUR_PSN_M,
+					      V2_QPC_BYTE_172_SQ_CUR_PSN_S);
+	qp_attr->dest_qp_num = (u8)roce_get_field(context->byte_56_dqpn_err,
+						  V2_QPC_BYTE_56_DQPN_M,
+						  V2_QPC_BYTE_56_DQPN_S);
+	qp_attr->qp_access_flags = ((roce_get_bit(context->byte_76_srqn_op_en,
+						  V2_QPC_BYTE_76_RRE_S)) << 2) |
+				   ((roce_get_bit(context->byte_76_srqn_op_en,
+						  V2_QPC_BYTE_76_RWE_S)) << 1) |
+				   ((roce_get_bit(context->byte_76_srqn_op_en,
+						  V2_QPC_BYTE_76_ATE_S)) << 3);
+	if (hr_qp->ibqp.qp_type == IB_QPT_RC ||
+	    hr_qp->ibqp.qp_type == IB_QPT_UC) {
+		struct ib_global_route *grh =
+				rdma_ah_retrieve_grh(&qp_attr->ah_attr);
+
+		rdma_ah_set_sl(&qp_attr->ah_attr,
+			       roce_get_field(context->byte_28_at_fl,
+					      V2_QPC_BYTE_28_SL_M,
+					      V2_QPC_BYTE_28_SL_S));
+		grh->flow_label = roce_get_field(context->byte_28_at_fl,
+						 V2_QPC_BYTE_28_FL_M,
+						 V2_QPC_BYTE_28_FL_S);
+		grh->sgid_index = roce_get_field(context->byte_20_smac_sgid_idx,
+						 V2_QPC_BYTE_20_SGID_IDX_M,
+						 V2_QPC_BYTE_20_SGID_IDX_S);
+		grh->hop_limit = roce_get_field(context->byte_24_mtu_tc,
+						V2_QPC_BYTE_24_HOP_LIMIT_M,
+						V2_QPC_BYTE_24_HOP_LIMIT_S);
+		grh->traffic_class = roce_get_field(context->byte_24_mtu_tc,
+						    V2_QPC_BYTE_24_TC_M,
+						    V2_QPC_BYTE_24_TC_S);
+
+		memcpy(grh->dgid.raw, context->dgid, sizeof(grh->dgid.raw));
+	}
+
+	qp_attr->port_num = hr_qp->port + 1;
+	qp_attr->sq_draining = 0;
+	qp_attr->max_rd_atomic = 1 << roce_get_field(context->byte_208_irrl,
+						     V2_QPC_BYTE_208_SR_MAX_M,
+						     V2_QPC_BYTE_208_SR_MAX_S);
+	qp_attr->max_dest_rd_atomic = 1 << roce_get_field(context->byte_140_raq,
+						     V2_QPC_BYTE_140_RR_MAX_M,
+						     V2_QPC_BYTE_140_RR_MAX_S);
+	qp_attr->min_rnr_timer = (u8)roce_get_field(context->byte_80_rnr_rx_cqn,
+						 V2_QPC_BYTE_80_MIN_RNR_TIME_M,
+						 V2_QPC_BYTE_80_MIN_RNR_TIME_S);
+	qp_attr->timeout = (u8)roce_get_field(context->byte_28_at_fl,
+					      V2_QPC_BYTE_28_AT_M,
+					      V2_QPC_BYTE_28_AT_S);
+	qp_attr->retry_cnt = roce_get_field(context->byte_212_lsn,
+					    V2_QPC_BYTE_212_RETRY_CNT_M,
+					    V2_QPC_BYTE_212_RETRY_CNT_S);
+	qp_attr->rnr_retry = context->rq_rnr_timer;
+
+done:
+	qp_attr->cur_qp_state = qp_attr->qp_state;
+	qp_attr->cap.max_recv_wr = hr_qp->rq.wqe_cnt;
+	qp_attr->cap.max_recv_sge = hr_qp->rq.max_gs;
+
+	if (!ibqp->uobject) {
+		qp_attr->cap.max_send_wr = hr_qp->sq.wqe_cnt;
+		qp_attr->cap.max_send_sge = hr_qp->sq.max_gs;
+	} else {
+		qp_attr->cap.max_send_wr = 0;
+		qp_attr->cap.max_send_sge = 0;
+	}
+
+	qp_init_attr->cap = qp_attr->cap;
+
+out:
+	mutex_unlock(&hr_qp->mutex);
+	kfree(context);
+	return ret;
+}
+
+static int hns_roce_v2_destroy_qp_common(struct hns_roce_dev *hr_dev,
+					 struct hns_roce_qp *hr_qp,
+					 int is_user)
+{
+	struct hns_roce_cq *send_cq, *recv_cq;
+	struct device *dev = hr_dev->dev;
+	int ret;
+
+	if (hr_qp->ibqp.qp_type == IB_QPT_RC && hr_qp->state != IB_QPS_RESET) {
+		/* Modify qp to reset before destroying qp */
+		ret = hns_roce_v2_modify_qp(&hr_qp->ibqp, NULL, 0,
+					    hr_qp->state, IB_QPS_RESET);
+		if (ret) {
+			dev_err(dev, "modify QP %06lx to ERR failed.\n",
+				hr_qp->qpn);
+			return ret;
+		}
+	}
+
+	send_cq = to_hr_cq(hr_qp->ibqp.send_cq);
+	recv_cq = to_hr_cq(hr_qp->ibqp.recv_cq);
+
+	hns_roce_lock_cqs(send_cq, recv_cq);
+
+	if (!is_user) {
+		__hns_roce_v2_cq_clean(recv_cq, hr_qp->qpn, hr_qp->ibqp.srq ?
+				       to_hr_srq(hr_qp->ibqp.srq) : NULL);
+		if (send_cq != recv_cq)
+			__hns_roce_v2_cq_clean(send_cq, hr_qp->qpn, NULL);
+	}
+
+	hns_roce_qp_remove(hr_dev, hr_qp);
+
+	hns_roce_unlock_cqs(send_cq, recv_cq);
+
+	hns_roce_qp_free(hr_dev, hr_qp);
+
+	/* Not special_QP, free their QPN */
+	if ((hr_qp->ibqp.qp_type == IB_QPT_RC) ||
+	    (hr_qp->ibqp.qp_type == IB_QPT_UC) ||
+	    (hr_qp->ibqp.qp_type == IB_QPT_UD))
+		hns_roce_release_range_qp(hr_dev, hr_qp->qpn, 1);
+
+	hns_roce_mtt_cleanup(hr_dev, &hr_qp->mtt);
+
+	if (is_user) {
+		ib_umem_release(hr_qp->umem);
+	} else {
+		kfree(hr_qp->sq.wrid);
+		kfree(hr_qp->rq.wrid);
+		hns_roce_buf_free(hr_dev, hr_qp->buff_size, &hr_qp->hr_buf);
+	}
+
+	return 0;
+}
+
+static int hns_roce_v2_destroy_qp(struct ib_qp *ibqp)
+{
+	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
+	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
+	int ret;
+
+	ret = hns_roce_v2_destroy_qp_common(hr_dev, hr_qp, !!ibqp->pd->uobject);
+	if (ret) {
+		dev_err(hr_dev->dev, "Destroy qp failed(%d)\n", ret);
+		return ret;
+	}
+
+	if (hr_qp->ibqp.qp_type == IB_QPT_GSI)
+		kfree(hr_to_hr_sqp(hr_qp));
+	else
+		kfree(hr_qp);
+
+	return 0;
+}
+
 static const struct hns_roce_hw hns_roce_hw_v2 = {
 	.cmq_init = hns_roce_v2_cmq_init,
 	.cmq_exit = hns_roce_v2_cmq_exit,
@@ -1282,6 +2585,9 @@ static int hns_roce_v2_clear_hem(struct hns_roce_dev *hr_dev,
 	.write_cqc = hns_roce_v2_write_cqc,
 	.set_hem = hns_roce_v2_set_hem,
 	.clear_hem = hns_roce_v2_clear_hem,
+	.modify_qp = hns_roce_v2_modify_qp,
+	.query_qp = hns_roce_v2_query_qp,
+	.destroy_qp = hns_roce_v2_destroy_qp,
 	.req_notify_cq = hns_roce_v2_req_notify_cq,
 	.poll_cq = hns_roce_v2_poll_cq,
 };
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index 8b69875..0360df0 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -104,6 +104,9 @@
 #define V2_CQ_DB_REQ_NOT			1
 
 #define V2_CQ_STATE_VALID			1
+#define V2_QKEY_VAL				0x80010000
+
+#define	GID_LEN_V2				16
 
 #define HNS_ROCE_V2_CQE_QPN_MASK		0x3ffff
 
@@ -262,6 +265,453 @@ struct hns_roce_v2_cq_context {
 #define	V2_CQC_BYTE_64_SE_CQE_IDX_S 0
 #define	V2_CQC_BYTE_64_SE_CQE_IDX_M GENMASK(23, 0)
 
+enum{
+	V2_MPT_ST_VALID = 0x1,
+};
+
+enum hns_roce_v2_qp_state {
+	HNS_ROCE_QP_ST_RST,
+	HNS_ROCE_QP_ST_INIT,
+	HNS_ROCE_QP_ST_RTR,
+	HNS_ROCE_QP_ST_RTS,
+	HNS_ROCE_QP_ST_SQER,
+	HNS_ROCE_QP_ST_SQD,
+	HNS_ROCE_QP_ST_ERR,
+	HNS_ROCE_QP_ST_SQ_DRAINING,
+	HNS_ROCE_QP_NUM_ST
+};
+
+struct hns_roce_v2_qp_context {
+	u32	byte_4_sqpn_tst;
+	u32	wqe_sge_ba;
+	u32	byte_12_sq_hop;
+	u32	byte_16_buf_ba_pg_sz;
+	u32	byte_20_smac_sgid_idx;
+	u32	byte_24_mtu_tc;
+	u32	byte_28_at_fl;
+	u8	dgid[GID_LEN_V2];
+	u32	dmac;
+	u32	byte_52_udpspn_dmac;
+	u32	byte_56_dqpn_err;
+	u32	byte_60_qpst_mapid;
+	u32	qkey_xrcd;
+	u32	byte_68_rq_db;
+	u32	rq_db_record_addr;
+	u32	byte_76_srqn_op_en;
+	u32	byte_80_rnr_rx_cqn;
+	u32	byte_84_rq_ci_pi;
+	u32	rq_cur_blk_addr;
+	u32	byte_92_srq_info;
+	u32	byte_96_rx_reqmsn;
+	u32	rq_nxt_blk_addr;
+	u32	byte_104_rq_sge;
+	u32	byte_108_rx_reqepsn;
+	u32	rq_rnr_timer;
+	u32	rx_msg_len;
+	u32	rx_rkey_pkt_info;
+	u64	rx_va;
+	u32	byte_132_trrl;
+	u32	trrl_ba;
+	u32	byte_140_raq;
+	u32	byte_144_raq;
+	u32	byte_148_raq;
+	u32	byte_152_raq;
+	u32	byte_156_raq;
+	u32	byte_160_sq_ci_pi;
+	u32	sq_cur_blk_addr;
+	u32	byte_168_irrl_idx;
+	u32	byte_172_sq_psn;
+	u32	byte_176_msg_pktn;
+	u32	sq_cur_sqe_blk_addr;
+	u32	byte_184_irrl_idx;
+	u32	cur_sge_offset;
+	u32	byte_192_ext_sge;
+	u32	byte_196_sq_psn;
+	u32	byte_200_sq_max;
+	u32	irrl_ba;
+	u32	byte_208_irrl;
+	u32	byte_212_lsn;
+	u32	sq_timer;
+	u32	byte_220_retry_psn_msn;
+	u32	byte_224_retry_msg;
+	u32	rx_sq_cur_blk_addr;
+	u32	byte_232_irrl_sge;
+	u32	irrl_cur_sge_offset;
+	u32	byte_240_irrl_tail;
+	u32	byte_244_rnr_rxack;
+	u32	byte_248_ack_psn;
+	u32	byte_252_err_txcqn;
+	u32	byte_256_sqflush_rqcqe;
+};
+
+#define	V2_QPC_BYTE_4_TST_S 0
+#define V2_QPC_BYTE_4_TST_M GENMASK(2, 0)
+
+#define	V2_QPC_BYTE_4_SGE_SHIFT_S 3
+#define V2_QPC_BYTE_4_SGE_SHIFT_M GENMASK(7, 3)
+
+#define	V2_QPC_BYTE_4_SQPN_S 8
+#define V2_QPC_BYTE_4_SQPN_M  GENMASK(31, 8)
+
+#define	V2_QPC_BYTE_12_WQE_SGE_BA_S 0
+#define V2_QPC_BYTE_12_WQE_SGE_BA_M GENMASK(28, 0)
+
+#define	V2_QPC_BYTE_12_SQ_HOP_NUM_S 29
+#define V2_QPC_BYTE_12_SQ_HOP_NUM_M GENMASK(30, 29)
+
+#define V2_QPC_BYTE_12_RSVD_LKEY_EN_S 31
+
+#define	V2_QPC_BYTE_16_WQE_SGE_BA_PG_SZ_S 0
+#define V2_QPC_BYTE_16_WQE_SGE_BA_PG_SZ_M GENMASK(3, 0)
+
+#define	V2_QPC_BYTE_16_WQE_SGE_BUF_PG_SZ_S 4
+#define V2_QPC_BYTE_16_WQE_SGE_BUF_PG_SZ_M GENMASK(7, 4)
+
+#define	V2_QPC_BYTE_16_PD_S 8
+#define V2_QPC_BYTE_16_PD_M GENMASK(31, 8)
+
+#define	V2_QPC_BYTE_20_RQ_HOP_NUM_S 0
+#define V2_QPC_BYTE_20_RQ_HOP_NUM_M GENMASK(1, 0)
+
+#define	V2_QPC_BYTE_20_SGE_HOP_NUM_S 2
+#define V2_QPC_BYTE_20_SGE_HOP_NUM_M GENMASK(3, 2)
+
+#define	V2_QPC_BYTE_20_RQWS_S 4
+#define V2_QPC_BYTE_20_RQWS_M GENMASK(7, 4)
+
+#define	V2_QPC_BYTE_20_SQ_SHIFT_S 8
+#define V2_QPC_BYTE_20_SQ_SHIFT_M GENMASK(11, 8)
+
+#define	V2_QPC_BYTE_20_RQ_SHIFT_S 12
+#define V2_QPC_BYTE_20_RQ_SHIFT_M GENMASK(15, 12)
+
+#define	V2_QPC_BYTE_20_SGID_IDX_S 16
+#define V2_QPC_BYTE_20_SGID_IDX_M GENMASK(23, 16)
+
+#define	V2_QPC_BYTE_20_SMAC_IDX_S 24
+#define V2_QPC_BYTE_20_SMAC_IDX_M GENMASK(31, 24)
+
+#define	V2_QPC_BYTE_24_HOP_LIMIT_S 0
+#define V2_QPC_BYTE_24_HOP_LIMIT_M GENMASK(7, 0)
+
+#define	V2_QPC_BYTE_24_TC_S 8
+#define V2_QPC_BYTE_24_TC_M GENMASK(15, 8)
+
+#define	V2_QPC_BYTE_24_VLAN_IDX_S 16
+#define V2_QPC_BYTE_24_VLAN_IDX_M GENMASK(27, 16)
+
+#define	V2_QPC_BYTE_24_MTU_S 28
+#define V2_QPC_BYTE_24_MTU_M GENMASK(31, 28)
+
+#define	V2_QPC_BYTE_28_FL_S 0
+#define V2_QPC_BYTE_28_FL_M GENMASK(19, 0)
+
+#define	V2_QPC_BYTE_28_SL_S 20
+#define V2_QPC_BYTE_28_SL_M GENMASK(23, 20)
+
+#define V2_QPC_BYTE_28_CNP_TX_FLAG_S 24
+
+#define V2_QPC_BYTE_28_CE_FLAG_S 25
+
+#define V2_QPC_BYTE_28_LBI_S 26
+
+#define	V2_QPC_BYTE_28_AT_S 27
+#define V2_QPC_BYTE_28_AT_M GENMASK(31, 27)
+
+#define	V2_QPC_BYTE_52_DMAC_S 0
+#define V2_QPC_BYTE_52_DMAC_M GENMASK(15, 0)
+
+#define V2_QPC_BYTE_52_UDPSPN_S 16
+#define V2_QPC_BYTE_52_UDPSPN_M GENMASK(31, 16)
+
+#define	V2_QPC_BYTE_56_DQPN_S 0
+#define V2_QPC_BYTE_56_DQPN_M GENMASK(23, 0)
+
+#define	V2_QPC_BYTE_56_SQ_TX_ERR_S 24
+#define	V2_QPC_BYTE_56_SQ_RX_ERR_S 25
+#define	V2_QPC_BYTE_56_RQ_TX_ERR_S 26
+#define	V2_QPC_BYTE_56_RQ_RX_ERR_S 27
+
+#define	V2_QPC_BYTE_56_LP_PKTN_INI_S 28
+#define V2_QPC_BYTE_56_LP_PKTN_INI_M GENMASK(31, 28)
+
+#define	V2_QPC_BYTE_60_MAPID_S 0
+#define V2_QPC_BYTE_60_MAPID_M GENMASK(12, 0)
+
+#define	V2_QPC_BYTE_60_INNER_MAP_IND_S 13
+
+#define	V2_QPC_BYTE_60_SQ_MAP_IND_S 14
+
+#define	V2_QPC_BYTE_60_RQ_MAP_IND_S 15
+
+#define	V2_QPC_BYTE_60_TEMPID_S 16
+#define V2_QPC_BYTE_60_TEMPID_M  GENMASK(22, 16)
+
+#define	V2_QPC_BYTE_60_EXT_MAP_IND_S 23
+
+#define	V2_QPC_BYTE_60_RTY_NUM_INI_BAK_S 24
+#define V2_QPC_BYTE_60_RTY_NUM_INI_BAK_M GENMASK(26, 24)
+
+#define V2_QPC_BYTE_60_SQ_RLS_IND_S 27
+
+#define	V2_QPC_BYTE_60_SQ_EXT_IND_S 28
+
+#define	V2_QPC_BYTE_60_QP_ST_S 29
+#define V2_QPC_BYTE_60_QP_ST_M GENMASK(31, 29)
+
+#define	V2_QPC_BYTE_68_RQ_RECORD_EN_S 0
+
+#define	V2_QPC_BYTE_68_RQ_DB_RECORD_ADDR_S 1
+#define V2_QPC_BYTE_68_RQ_DB_RECORD_ADDR_M GENMASK(31, 1)
+
+#define	V2_QPC_BYTE_76_SRQN_S 0
+#define V2_QPC_BYTE_76_SRQN_M GENMASK(23, 0)
+
+#define	V2_QPC_BYTE_76_SRQ_EN_S 24
+
+#define	V2_QPC_BYTE_76_RRE_S 25
+
+#define	V2_QPC_BYTE_76_RWE_S 26
+
+#define	V2_QPC_BYTE_76_ATE_S 27
+
+#define	V2_QPC_BYTE_76_RQIE_S 28
+
+#define	V2_QPC_BYTE_80_RX_CQN_S 0
+#define V2_QPC_BYTE_80_RX_CQN_M GENMASK(23, 0)
+
+#define	V2_QPC_BYTE_80_MIN_RNR_TIME_S 27
+#define V2_QPC_BYTE_80_MIN_RNR_TIME_M GENMASK(31, 27)
+
+#define	V2_QPC_BYTE_84_RQ_PRODUCER_IDX_S 0
+#define V2_QPC_BYTE_84_RQ_PRODUCER_IDX_M GENMASK(15, 0)
+
+#define	V2_QPC_BYTE_84_RQ_CONSUMER_IDX_S 16
+#define V2_QPC_BYTE_84_RQ_CONSUMER_IDX_M GENMASK(31, 16)
+
+#define	V2_QPC_BYTE_92_RQ_CUR_BLK_ADDR_S 0
+#define V2_QPC_BYTE_92_RQ_CUR_BLK_ADDR_M GENMASK(19, 0)
+
+#define	V2_QPC_BYTE_92_SRQ_INFO_S 20
+#define V2_QPC_BYTE_92_SRQ_INFO_M GENMASK(31, 20)
+
+#define	V2_QPC_BYTE_96_RX_REQ_MSN_S 0
+#define V2_QPC_BYTE_96_RX_REQ_MSN_M GENMASK(23, 0)
+
+#define	V2_QPC_BYTE_104_RQ_NXT_BLK_ADDR_S 0
+#define V2_QPC_BYTE_104_RQ_NXT_BLK_ADDR_M GENMASK(19, 0)
+
+#define	V2_QPC_BYTE_104_RQ_CUR_WQE_SGE_NUM_S 24
+#define V2_QPC_BYTE_104_RQ_CUR_WQE_SGE_NUM_M GENMASK(31, 24)
+
+#define V2_QPC_BYTE_108_INV_CREDIT_S 0
+
+#define V2_QPC_BYTE_108_RX_REQ_PSN_ERR_S 3
+
+#define	V2_QPC_BYTE_108_RX_REQ_LAST_OPTYPE_S 4
+#define V2_QPC_BYTE_108_RX_REQ_LAST_OPTYPE_M GENMASK(6, 4)
+
+#define V2_QPC_BYTE_108_RX_REQ_RNR_S 7
+
+#define	V2_QPC_BYTE_108_RX_REQ_EPSN_S 8
+#define V2_QPC_BYTE_108_RX_REQ_EPSN_M GENMASK(31, 8)
+
+#define	V2_QPC_BYTE_132_TRRL_HEAD_MAX_S 0
+#define V2_QPC_BYTE_132_TRRL_HEAD_MAX_M GENMASK(7, 0)
+
+#define	V2_QPC_BYTE_132_TRRL_TAIL_MAX_S 8
+#define V2_QPC_BYTE_132_TRRL_TAIL_MAX_M GENMASK(15, 8)
+
+#define	V2_QPC_BYTE_132_TRRL_BA_S 16
+#define V2_QPC_BYTE_132_TRRL_BA_M GENMASK(31, 16)
+
+#define	V2_QPC_BYTE_140_TRRL_BA_S 0
+#define V2_QPC_BYTE_140_TRRL_BA_M GENMASK(11, 0)
+
+#define	V2_QPC_BYTE_140_RR_MAX_S 12
+#define V2_QPC_BYTE_140_RR_MAX_M GENMASK(14, 12)
+
+#define	V2_QPC_BYTE_140_RSVD_RAQ_MAP_S 15
+
+#define	V2_QPC_BYTE_140_RAQ_TRRL_HEAD_S 16
+#define V2_QPC_BYTE_140_RAQ_TRRL_HEAD_M GENMASK(23, 16)
+
+#define	V2_QPC_BYTE_140_RAQ_TRRL_TAIL_S 24
+#define V2_QPC_BYTE_140_RAQ_TRRL_TAIL_M GENMASK(31, 24)
+
+#define	V2_QPC_BYTE_144_RAQ_RTY_INI_PSN_S 0
+#define V2_QPC_BYTE_144_RAQ_RTY_INI_PSN_M GENMASK(23, 0)
+
+#define V2_QPC_BYTE_144_RAQ_RTY_INI_IND_S 24
+
+#define V2_QPC_BYTE_144_RAQ_CREDIT_S 25
+#define V2_QPC_BYTE_144_RAQ_CREDIT_M GENMASK(29, 25)
+
+#define V2_QPC_BYTE_144_RESP_RTY_FLG_S 31
+
+#define	V2_QPC_BYTE_148_RQ_MSN_S 0
+#define V2_QPC_BYTE_148_RQ_MSN_M GENMASK(23, 0)
+
+#define	V2_QPC_BYTE_148_RAQ_SYNDROME_S 24
+#define V2_QPC_BYTE_148_RAQ_SYNDROME_M GENMASK(31, 24)
+
+#define	V2_QPC_BYTE_152_RAQ_PSN_S 8
+#define V2_QPC_BYTE_152_RAQ_PSN_M GENMASK(31, 8)
+
+#define	V2_QPC_BYTE_152_RAQ_TRRL_RTY_HEAD_S 24
+#define V2_QPC_BYTE_152_RAQ_TRRL_RTY_HEAD_M GENMASK(31, 24)
+
+#define	V2_QPC_BYTE_156_RAQ_USE_PKTN_S 0
+#define V2_QPC_BYTE_156_RAQ_USE_PKTN_M GENMASK(23, 0)
+
+#define	V2_QPC_BYTE_160_SQ_PRODUCER_IDX_S 0
+#define V2_QPC_BYTE_160_SQ_PRODUCER_IDX_M GENMASK(15, 0)
+
+#define	V2_QPC_BYTE_160_SQ_CONSUMER_IDX_S 16
+#define V2_QPC_BYTE_160_SQ_CONSUMER_IDX_M GENMASK(31, 16)
+
+#define	V2_QPC_BYTE_168_SQ_CUR_BLK_ADDR_S 0
+#define V2_QPC_BYTE_168_SQ_CUR_BLK_ADDR_M GENMASK(19, 0)
+
+#define V2_QPC_BYTE_168_MSG_RTY_LP_FLG_S 20
+
+#define	V2_QPC_BYTE_168_LP_SGEN_INI_S 21
+#define V2_QPC_BYTE_168_LP_SGEN_INI_M GENMASK(23, 21)
+
+#define	V2_QPC_BYTE_168_SQ_SHIFT_BAK_S 24
+#define V2_QPC_BYTE_168_SQ_SHIFT_BAK_M GENMASK(27, 24)
+
+#define	V2_QPC_BYTE_168_IRRL_IDX_LSB_S 28
+#define V2_QPC_BYTE_168_IRRL_IDX_LSB_M GENMASK(31, 28)
+
+#define	V2_QPC_BYTE_172_ACK_REQ_FREQ_S 0
+#define V2_QPC_BYTE_172_ACK_REQ_FREQ_M GENMASK(5, 0)
+
+#define V2_QPC_BYTE_172_MSG_RNR_FLG_S 6
+
+#define V2_QPC_BYTE_172_FRE_S 7
+
+#define	V2_QPC_BYTE_172_SQ_CUR_PSN_S 8
+#define V2_QPC_BYTE_172_SQ_CUR_PSN_M GENMASK(31, 8)
+
+#define	V2_QPC_BYTE_176_MSG_USE_PKTN_S 0
+#define V2_QPC_BYTE_176_MSG_USE_PKTN_M GENMASK(23, 0)
+
+#define	V2_QPC_BYTE_176_IRRL_HEAD_PRE_S 24
+#define V2_QPC_BYTE_176_IRRL_HEAD_PRE_M GENMASK(31, 24)
+
+#define	V2_QPC_BYTE_184_SQ_CUR_SGE_BLK_ADDR_S 0
+#define V2_QPC_BYTE_184_SQ_CUR_SGE_BLK_ADDR_M GENMASK(19, 0)
+
+#define	V2_QPC_BYTE_184_IRRL_IDX_MSB_S 20
+#define V2_QPC_BYTE_184_IRRL_IDX_MSB_M GENMASK(31, 20)
+
+#define	V2_QPC_BYTE_192_CUR_SGE_IDX_S 0
+#define V2_QPC_BYTE_192_CUR_SGE_IDX_M GENMASK(23, 0)
+
+#define	V2_QPC_BYTE_192_EXT_SGE_NUM_LEFT_S 24
+#define V2_QPC_BYTE_192_EXT_SGE_NUM_LEFT_M GENMASK(31, 24)
+
+#define	V2_QPC_BYTE_196_IRRL_HEAD_S 0
+#define V2_QPC_BYTE_196_IRRL_HEAD_M GENMASK(7, 0)
+
+#define	V2_QPC_BYTE_196_SQ_MAX_PSN_S 8
+#define V2_QPC_BYTE_196_SQ_MAX_PSN_M GENMASK(31, 8)
+
+#define	V2_QPC_BYTE_200_SQ_MAX_IDX_S 0
+#define V2_QPC_BYTE_200_SQ_MAX_IDX_M GENMASK(15, 0)
+
+#define	V2_QPC_BYTE_200_LCL_OPERATED_CNT_S 16
+#define V2_QPC_BYTE_200_LCL_OPERATED_CNT_M GENMASK(31, 16)
+
+#define	V2_QPC_BYTE_208_IRRL_BA_S 0
+#define V2_QPC_BYTE_208_IRRL_BA_M GENMASK(25, 0)
+
+#define V2_QPC_BYTE_208_PKT_RNR_FLG_S 26
+
+#define V2_QPC_BYTE_208_PKT_RTY_FLG_S 27
+
+#define V2_QPC_BYTE_208_RMT_E2E_S 28
+
+#define	V2_QPC_BYTE_208_SR_MAX_S 29
+#define V2_QPC_BYTE_208_SR_MAX_M GENMASK(31, 29)
+
+#define	V2_QPC_BYTE_212_LSN_S 0
+#define V2_QPC_BYTE_212_LSN_M GENMASK(23, 0)
+
+#define	V2_QPC_BYTE_212_RETRY_NUM_INIT_S 24
+#define V2_QPC_BYTE_212_RETRY_NUM_INIT_M GENMASK(26, 24)
+
+#define	V2_QPC_BYTE_212_CHECK_FLG_S 27
+#define V2_QPC_BYTE_212_CHECK_FLG_M GENMASK(28, 27)
+
+#define	V2_QPC_BYTE_212_RETRY_CNT_S 29
+#define V2_QPC_BYTE_212_RETRY_CNT_M GENMASK(31, 29)
+
+#define	V2_QPC_BYTE_220_RETRY_MSG_MSN_S 0
+#define V2_QPC_BYTE_220_RETRY_MSG_MSN_M GENMASK(15, 0)
+
+#define	V2_QPC_BYTE_220_RETRY_MSG_PSN_S 16
+#define V2_QPC_BYTE_220_RETRY_MSG_PSN_M GENMASK(31, 16)
+
+#define	V2_QPC_BYTE_224_RETRY_MSG_PSN_S 0
+#define V2_QPC_BYTE_224_RETRY_MSG_PSN_M GENMASK(7, 0)
+
+#define	V2_QPC_BYTE_224_RETRY_MSG_FPKT_PSN_S 8
+#define V2_QPC_BYTE_224_RETRY_MSG_FPKT_PSN_M GENMASK(31, 8)
+
+#define	V2_QPC_BYTE_232_RX_SQ_CUR_BLK_ADDR_S 0
+#define V2_QPC_BYTE_232_RX_SQ_CUR_BLK_ADDR_M GENMASK(19, 0)
+
+#define	V2_QPC_BYTE_232_IRRL_SGE_IDX_S 20
+#define V2_QPC_BYTE_232_IRRL_SGE_IDX_M GENMASK(28, 20)
+
+#define	V2_QPC_BYTE_240_IRRL_TAIL_REAL_S 0
+#define V2_QPC_BYTE_240_IRRL_TAIL_REAL_M GENMASK(7, 0)
+
+#define	V2_QPC_BYTE_240_IRRL_TAIL_RD_S 8
+#define V2_QPC_BYTE_240_IRRL_TAIL_RD_M GENMASK(15, 8)
+
+#define	V2_QPC_BYTE_240_RX_ACK_MSN_S 16
+#define V2_QPC_BYTE_240_RX_ACK_MSN_M GENMASK(31, 16)
+
+#define	V2_QPC_BYTE_244_RX_ACK_EPSN_S 0
+#define V2_QPC_BYTE_244_RX_ACK_EPSN_M GENMASK(23, 0)
+
+#define	V2_QPC_BYTE_244_RNR_NUM_INIT_S 24
+#define V2_QPC_BYTE_244_RNR_NUM_INIT_M GENMASK(26, 24)
+
+#define	V2_QPC_BYTE_244_RNR_CNT_S 27
+#define V2_QPC_BYTE_244_RNR_CNT_M GENMASK(29, 27)
+
+#define	V2_QPC_BYTE_248_IRRL_PSN_S 0
+#define V2_QPC_BYTE_248_IRRL_PSN_M GENMASK(23, 0)
+
+#define V2_QPC_BYTE_248_ACK_PSN_ERR_S 24
+
+#define	V2_QPC_BYTE_248_ACK_LAST_OPTYPE_S 25
+#define V2_QPC_BYTE_248_ACK_LAST_OPTYPE_M GENMASK(26, 25)
+
+#define V2_QPC_BYTE_248_IRRL_PSN_VLD_S 27
+
+#define V2_QPC_BYTE_248_RNR_RETRY_FLAG_S 28
+
+#define V2_QPC_BYTE_248_CQ_ERR_IND_S 31
+
+#define	V2_QPC_BYTE_252_TX_CQN_S 0
+#define V2_QPC_BYTE_252_TX_CQN_M GENMASK(23, 0)
+
+#define	V2_QPC_BYTE_252_SIG_TYPE_S 24
+
+#define	V2_QPC_BYTE_252_ERR_TYPE_S 25
+#define V2_QPC_BYTE_252_ERR_TYPE_M GENMASK(31, 25)
+
+#define	V2_QPC_BYTE_256_RQ_CQE_IDX_S 0
+#define V2_QPC_BYTE_256_RQ_CQE_IDX_M GENMASK(15, 0)
+
+#define	V2_QPC_BYTE_256_SQ_FLUSH_IDX_S 16
+#define V2_QPC_BYTE_256_SQ_FLUSH_IDX_M GENMASK(31, 16)
+
 struct hns_roce_v2_cqe {
 	u32	byte_4;
 	u32	rkey_immtdata;
diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c
index 881ea67..c04aa81 100644
--- a/drivers/infiniband/hw/hns/hns_roce_qp.c
+++ b/drivers/infiniband/hw/hns/hns_roce_qp.c
@@ -286,20 +286,27 @@ static int hns_roce_set_rq_size(struct hns_roce_dev *hr_dev,
 			return -EINVAL;
 		}
 
-		/* In v1 engine, parameter verification procession */
-		max_cnt = cap->max_recv_wr > HNS_ROCE_MIN_WQE_NUM ?
-			  cap->max_recv_wr : HNS_ROCE_MIN_WQE_NUM;
+		if (hr_dev->caps.min_wqes)
+			max_cnt = max(cap->max_recv_wr, hr_dev->caps.min_wqes);
+		else
+			max_cnt = cap->max_recv_wr;
+
 		hr_qp->rq.wqe_cnt = roundup_pow_of_two(max_cnt);
 
 		if ((u32)hr_qp->rq.wqe_cnt > hr_dev->caps.max_wqes) {
-			dev_err(dev, "hns_roce_set_rq_size rq.wqe_cnt too large\n");
+			dev_err(dev, "while setting rq size, rq.wqe_cnt too large\n");
 			return -EINVAL;
 		}
 
 		max_cnt = max(1U, cap->max_recv_sge);
 		hr_qp->rq.max_gs = roundup_pow_of_two(max_cnt);
-		/* WQE is fixed for 64B */
-		hr_qp->rq.wqe_shift = ilog2(hr_dev->caps.max_rq_desc_sz);
+		if (hr_dev->caps.max_rq_sg <= 2)
+			hr_qp->rq.wqe_shift =
+					ilog2(hr_dev->caps.max_rq_desc_sz);
+		else
+			hr_qp->rq.wqe_shift =
+					ilog2(hr_dev->caps.max_rq_desc_sz
+					      * hr_qp->rq.max_gs);
 	}
 
 	cap->max_recv_wr = hr_qp->rq.max_post = hr_qp->rq.wqe_cnt;
@@ -309,11 +316,13 @@ static int hns_roce_set_rq_size(struct hns_roce_dev *hr_dev,
 }
 
 static int hns_roce_set_user_sq_size(struct hns_roce_dev *hr_dev,
+				     struct ib_qp_cap *cap,
 				     struct hns_roce_qp *hr_qp,
 				     struct hns_roce_ib_create_qp *ucmd)
 {
 	u32 roundup_sq_stride = roundup_pow_of_two(hr_dev->caps.max_sq_desc_sz);
 	u8 max_sq_stride = ilog2(roundup_sq_stride);
+	u32 max_cnt;
 
 	/* Sanity check SQ size before proceeding */
 	if ((u32)(1 << ucmd->log_sq_bb_count) > hr_dev->caps.max_wqes ||
@@ -323,18 +332,61 @@ static int hns_roce_set_user_sq_size(struct hns_roce_dev *hr_dev,
 		return -EINVAL;
 	}
 
+	if (cap->max_send_sge > hr_dev->caps.max_sq_sg) {
+		dev_err(hr_dev->dev, "SQ sge error! max_send_sge=%d\n",
+			cap->max_send_sge);
+		return -EINVAL;
+	}
+
 	hr_qp->sq.wqe_cnt = 1 << ucmd->log_sq_bb_count;
 	hr_qp->sq.wqe_shift = ucmd->log_sq_stride;
 
+	max_cnt = max(1U, cap->max_send_sge);
+	if (hr_dev->caps.max_sq_sg <= 2)
+		hr_qp->sq.max_gs = roundup_pow_of_two(max_cnt);
+	else
+		hr_qp->sq.max_gs = max_cnt;
+
+	if (hr_qp->sq.max_gs > 2)
+		hr_qp->sge.sge_cnt = roundup_pow_of_two(hr_qp->sq.wqe_cnt *
+							(hr_qp->sq.max_gs - 2));
+	hr_qp->sge.sge_shift = 4;
+
 	/* Get buf size, SQ and RQ  are aligned to page_szie */
-	hr_qp->buff_size = HNS_ROCE_ALOGN_UP((hr_qp->rq.wqe_cnt <<
+	if (hr_dev->caps.max_sq_sg <= 2) {
+		hr_qp->buff_size = HNS_ROCE_ALOGN_UP((hr_qp->rq.wqe_cnt <<
 					     hr_qp->rq.wqe_shift), PAGE_SIZE) +
-			   HNS_ROCE_ALOGN_UP((hr_qp->sq.wqe_cnt <<
+				   HNS_ROCE_ALOGN_UP((hr_qp->sq.wqe_cnt <<
 					     hr_qp->sq.wqe_shift), PAGE_SIZE);
 
-	hr_qp->sq.offset = 0;
-	hr_qp->rq.offset = HNS_ROCE_ALOGN_UP((hr_qp->sq.wqe_cnt <<
+		hr_qp->sq.offset = 0;
+		hr_qp->rq.offset = HNS_ROCE_ALOGN_UP((hr_qp->sq.wqe_cnt <<
 					     hr_qp->sq.wqe_shift), PAGE_SIZE);
+	} else {
+		hr_qp->buff_size = HNS_ROCE_ALOGN_UP((hr_qp->rq.wqe_cnt <<
+					     hr_qp->rq.wqe_shift), PAGE_SIZE) +
+				   HNS_ROCE_ALOGN_UP((hr_qp->sge.sge_cnt <<
+					     hr_qp->sge.sge_shift), PAGE_SIZE) +
+				   HNS_ROCE_ALOGN_UP((hr_qp->sq.wqe_cnt <<
+					     hr_qp->sq.wqe_shift), PAGE_SIZE);
+
+		hr_qp->sq.offset = 0;
+		if (hr_qp->sge.sge_cnt) {
+			hr_qp->sge.offset = HNS_ROCE_ALOGN_UP(
+							(hr_qp->sq.wqe_cnt <<
+							hr_qp->sq.wqe_shift),
+							PAGE_SIZE);
+			hr_qp->rq.offset = hr_qp->sge.offset +
+					HNS_ROCE_ALOGN_UP((hr_qp->sge.sge_cnt <<
+						hr_qp->sge.sge_shift),
+						PAGE_SIZE);
+		} else {
+			hr_qp->rq.offset = HNS_ROCE_ALOGN_UP(
+							(hr_qp->sq.wqe_cnt <<
+							hr_qp->sq.wqe_shift),
+							PAGE_SIZE);
+		}
+	}
 
 	return 0;
 }
@@ -345,11 +397,12 @@ static int hns_roce_set_kernel_sq_size(struct hns_roce_dev *hr_dev,
 {
 	struct device *dev = hr_dev->dev;
 	u32 max_cnt;
+	int size;
 
 	if (cap->max_send_wr  > hr_dev->caps.max_wqes  ||
 	    cap->max_send_sge > hr_dev->caps.max_sq_sg ||
 	    cap->max_inline_data > hr_dev->caps.max_sq_inline) {
-		dev_err(dev, "hns_roce_set_kernel_sq_size error1\n");
+		dev_err(dev, "SQ WR or sge or inline data error!\n");
 		return -EINVAL;
 	}
 
@@ -357,27 +410,45 @@ static int hns_roce_set_kernel_sq_size(struct hns_roce_dev *hr_dev,
 	hr_qp->sq_max_wqes_per_wr = 1;
 	hr_qp->sq_spare_wqes = 0;
 
-	/* In v1 engine, parameter verification procession */
-	max_cnt = cap->max_send_wr > HNS_ROCE_MIN_WQE_NUM ?
-		  cap->max_send_wr : HNS_ROCE_MIN_WQE_NUM;
+	if (hr_dev->caps.min_wqes)
+		max_cnt = max(cap->max_send_wr, hr_dev->caps.min_wqes);
+	else
+		max_cnt = cap->max_send_wr;
+
 	hr_qp->sq.wqe_cnt = roundup_pow_of_two(max_cnt);
 	if ((u32)hr_qp->sq.wqe_cnt > hr_dev->caps.max_wqes) {
-		dev_err(dev, "hns_roce_set_kernel_sq_size sq.wqe_cnt too large\n");
+		dev_err(dev, "while setting kernel sq size, sq.wqe_cnt too large\n");
 		return -EINVAL;
 	}
 
 	/* Get data_seg numbers */
 	max_cnt = max(1U, cap->max_send_sge);
-	hr_qp->sq.max_gs = roundup_pow_of_two(max_cnt);
+	if (hr_dev->caps.max_sq_sg <= 2)
+		hr_qp->sq.max_gs = roundup_pow_of_two(max_cnt);
+	else
+		hr_qp->sq.max_gs = max_cnt;
 
-	/* Get buf size, SQ and RQ  are aligned to page_szie */
-	hr_qp->buff_size = HNS_ROCE_ALOGN_UP((hr_qp->rq.wqe_cnt <<
-					     hr_qp->rq.wqe_shift), PAGE_SIZE) +
-			   HNS_ROCE_ALOGN_UP((hr_qp->sq.wqe_cnt <<
-					     hr_qp->sq.wqe_shift), PAGE_SIZE);
+	if (hr_qp->sq.max_gs > 2) {
+		hr_qp->sge.sge_cnt = roundup_pow_of_two(hr_qp->sq.wqe_cnt *
+				     (hr_qp->sq.max_gs - 2));
+		hr_qp->sge.sge_shift = 4;
+	}
+
+	/* Get buf size, SQ and RQ are aligned to PAGE_SIZE */
 	hr_qp->sq.offset = 0;
-	hr_qp->rq.offset = HNS_ROCE_ALOGN_UP((hr_qp->sq.wqe_cnt <<
-					      hr_qp->sq.wqe_shift), PAGE_SIZE);
+	size = HNS_ROCE_ALOGN_UP(hr_qp->sq.wqe_cnt << hr_qp->sq.wqe_shift,
+				 PAGE_SIZE);
+
+	if (hr_dev->caps.max_sq_sg > 2 && hr_qp->sge.sge_cnt) {
+		hr_qp->sge.offset = size;
+		size += HNS_ROCE_ALOGN_UP(hr_qp->sge.sge_cnt <<
+					  hr_qp->sge.sge_shift, PAGE_SIZE);
+	}
+
+	hr_qp->rq.offset = size;
+	size += HNS_ROCE_ALOGN_UP((hr_qp->rq.wqe_cnt << hr_qp->rq.wqe_shift),
+				  PAGE_SIZE);
+	hr_qp->buff_size = size;
 
 	/* Get wr and sge number which send */
 	cap->max_send_wr = hr_qp->sq.max_post = hr_qp->sq.wqe_cnt;
@@ -425,7 +496,8 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev,
 			goto err_out;
 		}
 
-		ret = hns_roce_set_user_sq_size(hr_dev, hr_qp, &ucmd);
+		ret = hns_roce_set_user_sq_size(hr_dev, &init_attr->cap, hr_qp,
+						&ucmd);
 		if (ret) {
 			dev_err(dev, "hns_roce_set_user_sq_size error for create qp\n");
 			goto err_out;
@@ -528,7 +600,9 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev,
 		}
 	}
 
-	if ((init_attr->qp_type) == IB_QPT_GSI) {
+	if (init_attr->qp_type == IB_QPT_GSI &&
+	    hr_dev->hw_rev == HNS_ROCE_HW_VER1) {
+		/* In v1 engine, GSI QP context in RoCE engine's register */
 		ret = hns_roce_gsi_qp_alloc(hr_dev, qpn, hr_qp);
 		if (ret) {
 			dev_err(dev, "hns_roce_qp_alloc failed!\n");
@@ -700,7 +774,10 @@ int hns_roce_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 		p = attr_mask & IB_QP_PORT ? (attr->port_num - 1) : hr_qp->port;
 		active_mtu = iboe_get_mtu(hr_dev->iboe.netdevs[p]->mtu);
 
-		if (attr->path_mtu > IB_MTU_2048 ||
+		if ((hr_dev->caps.max_mtu == IB_MTU_4096 &&
+		    attr->path_mtu > IB_MTU_4096) ||
+		    (hr_dev->caps.max_mtu == IB_MTU_2048 &&
+		    attr->path_mtu > IB_MTU_2048) ||
 		    attr->path_mtu < IB_MTU_256 ||
 		    attr->path_mtu > active_mtu) {
 			dev_err(dev, "attr path_mtu(%d)invalid while modify qp",
@@ -724,9 +801,7 @@ int hns_roce_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	}
 
 	if (cur_state == new_state && cur_state == IB_QPS_RESET) {
-		ret = -EPERM;
-		dev_err(dev, "cur_state=%d new_state=%d\n", cur_state,
-			new_state);
+		ret = 0;
 		goto out;
 	}
 
@@ -804,6 +879,13 @@ void *get_send_wqe(struct hns_roce_qp *hr_qp, int n)
 }
 EXPORT_SYMBOL_GPL(get_send_wqe);
 
+void *get_send_extend_sge(struct hns_roce_qp *hr_qp, int n)
+{
+	return hns_roce_buf_offset(&hr_qp->hr_buf, hr_qp->sge.offset +
+					(n << hr_qp->sge.sge_shift));
+}
+EXPORT_SYMBOL_GPL(get_send_extend_sge);
+
 bool hns_roce_wq_overflow(struct hns_roce_wq *hr_wq, int nreq,
 			  struct ib_cq *ib_cq)
 {
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 16/20] RDMA/hns: Add support for processing send wr and receive wr

[Wei Hu (Xavier)]

This patch is implementing for posting send request and
receiving request for hip08 RoCE driver. such as
post send verbs and post recv verbs.

Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_cq.c     |   4 +-
 drivers/infiniband/hw/hns/hns_roce_device.h |   2 +
 drivers/infiniband/hw/hns/hns_roce_hw_v1.c  |   4 +-
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 325 ++++++++++++++++++++++++++++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  |  72 ++++++
 drivers/infiniband/hw/hns/hns_roce_qp.c     |   5 +-
 6 files changed, 406 insertions(+), 6 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_cq.c b/drivers/infiniband/hw/hns/hns_roce_cq.c
index fa17dbc..88cdf6f 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cq.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cq.c
@@ -351,8 +351,8 @@ struct ib_cq *hns_roce_ib_create_cq(struct ib_device *ib_dev,
 		}
 
 		uar = &hr_dev->priv_uar;
-		hr_cq->cq_db_l = hr_dev->reg_base + ROCEE_DB_OTHERS_L_0_REG +
-				 0x1000 * uar->index;
+		hr_cq->cq_db_l = hr_dev->reg_base + hr_dev->odb_offset +
+				DB_REG_OFFSET * uar->index;
 	}
 
 	/* Allocate cq index, fill cq_context */
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 22f2079..b45dba5 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -643,6 +643,8 @@ struct hns_roce_dev {
 
 	int			cmd_mod;
 	int			loop_idc;
+	u32			sdb_offset;
+	u32			odb_offset;
 	dma_addr_t		tptr_dma_addr; /*only for hw v1*/
 	u32			tptr_size; /*only for hw v1*/
 	const struct hns_roce_hw *hw;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index 8ed0538..426f55a 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -3221,7 +3221,7 @@ static int hns_roce_v1_m_qp(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
 
 		if (ibqp->uobject) {
 			hr_qp->rq.db_reg_l = hr_dev->reg_base +
-				     ROCEE_DB_OTHERS_L_0_REG +
+				     hr_dev->odb_offset +
 				     DB_REG_OFFSET * hr_dev->priv_uar.index;
 		}
 
@@ -4084,6 +4084,8 @@ static int hns_roce_get_cfg(struct hns_roce_dev *hr_dev)
 	/* cmd issue mode: 0 is poll, 1 is event */
 	hr_dev->cmd_mod = 1;
 	hr_dev->loop_idc = 0;
+	hr_dev->sdb_offset = ROCEE_DB_SQ_L_0_REG;
+	hr_dev->odb_offset = ROCEE_DB_OTHERS_L_0_REG;
 
 	/* read the interrupt names from the DT or ACPI */
 	ret = device_property_read_string_array(dev, "interrupt-names",
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index f2425ee..b614177 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -43,6 +43,327 @@
 #include "hns_roce_hem.h"
 #include "hns_roce_hw_v2.h"
 
+static void set_data_seg_v2(struct hns_roce_v2_wqe_data_seg *dseg,
+			    struct ib_sge *sg)
+{
+	dseg->lkey = cpu_to_le32(sg->lkey);
+	dseg->addr = cpu_to_le64(sg->addr);
+	dseg->len  = cpu_to_le32(sg->length);
+}
+
+static int hns_roce_v2_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+				 struct ib_send_wr **bad_wr)
+{
+	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
+	struct hns_roce_v2_rc_send_wqe *rc_sq_wqe;
+	struct hns_roce_qp *qp = to_hr_qp(ibqp);
+	struct hns_roce_v2_wqe_data_seg *dseg;
+	struct device *dev = hr_dev->dev;
+	struct hns_roce_v2_db sq_db;
+	unsigned int sge_ind = 0;
+	unsigned int wqe_sz = 0;
+	unsigned long flags;
+	unsigned int ind;
+	void *wqe = NULL;
+	int ret = 0;
+	int nreq;
+	int i;
+
+	if (unlikely(ibqp->qp_type != IB_QPT_RC)) {
+		dev_err(dev, "Not supported QP(0x%x)type!\n", ibqp->qp_type);
+		*bad_wr = NULL;
+		return -EOPNOTSUPP;
+	}
+
+	if (unlikely(qp->state != IB_QPS_RTS && qp->state != IB_QPS_SQD)) {
+		dev_err(dev, "Post WQE fail, QP state %d err!\n", qp->state);
+		*bad_wr = wr;
+		return -EINVAL;
+	}
+
+	spin_lock_irqsave(&qp->sq.lock, flags);
+	ind = qp->sq_next_wqe;
+	sge_ind = qp->next_sge;
+
+	for (nreq = 0; wr; ++nreq, wr = wr->next) {
+		if (hns_roce_wq_overflow(&qp->sq, nreq, qp->ibqp.send_cq)) {
+			ret = -ENOMEM;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		if (unlikely(wr->num_sge > qp->sq.max_gs)) {
+			dev_err(dev, "num_sge=%d > qp->sq.max_gs=%d\n",
+				wr->num_sge, qp->sq.max_gs);
+			ret = -EINVAL;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		wqe = get_send_wqe(qp, ind & (qp->sq.wqe_cnt - 1));
+		qp->sq.wrid[(qp->sq.head + nreq) & (qp->sq.wqe_cnt - 1)] =
+								      wr->wr_id;
+
+
+		rc_sq_wqe = wqe;
+		memset(rc_sq_wqe, 0, sizeof(*rc_sq_wqe));
+		for (i = 0; i < wr->num_sge; i++)
+			rc_sq_wqe->msg_len += wr->sg_list[i].length;
+
+		rc_sq_wqe->inv_key_immtdata = send_ieth(wr);
+
+		switch (wr->opcode) {
+		case IB_WR_RDMA_READ:
+			roce_set_field(rc_sq_wqe->byte_4,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
+				       HNS_ROCE_V2_WQE_OP_RDMA_READ);
+			rc_sq_wqe->rkey = cpu_to_le32(rdma_wr(wr)->rkey);
+			rc_sq_wqe->va = cpu_to_le64(rdma_wr(wr)->remote_addr);
+			break;
+		case IB_WR_RDMA_WRITE:
+			roce_set_field(rc_sq_wqe->byte_4,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
+				       HNS_ROCE_V2_WQE_OP_RDMA_WRITE);
+			rc_sq_wqe->rkey = cpu_to_le32(rdma_wr(wr)->rkey);
+			rc_sq_wqe->va = cpu_to_le64(rdma_wr(wr)->remote_addr);
+			break;
+		case IB_WR_RDMA_WRITE_WITH_IMM:
+			roce_set_field(rc_sq_wqe->byte_4,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
+				       HNS_ROCE_V2_WQE_OP_RDMA_WRITE_WITH_IMM);
+			rc_sq_wqe->rkey = cpu_to_le32(rdma_wr(wr)->rkey);
+			rc_sq_wqe->va = cpu_to_le64(rdma_wr(wr)->remote_addr);
+			break;
+		case IB_WR_SEND:
+			roce_set_field(rc_sq_wqe->byte_4,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
+				       HNS_ROCE_V2_WQE_OP_SEND);
+			break;
+		case IB_WR_SEND_WITH_INV:
+			roce_set_field(rc_sq_wqe->byte_4,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
+				       HNS_ROCE_V2_WQE_OP_SEND_WITH_INV);
+			break;
+		case IB_WR_SEND_WITH_IMM:
+			roce_set_field(rc_sq_wqe->byte_4,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
+				       HNS_ROCE_V2_WQE_OP_SEND_WITH_IMM);
+			break;
+		case IB_WR_LOCAL_INV:
+			roce_set_field(rc_sq_wqe->byte_4,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
+				       HNS_ROCE_V2_WQE_OP_LOCAL_INV);
+			break;
+		case IB_WR_ATOMIC_CMP_AND_SWP:
+			roce_set_field(rc_sq_wqe->byte_4,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
+				       HNS_ROCE_V2_WQE_OP_ATOM_CMP_AND_SWAP);
+			break;
+		case IB_WR_ATOMIC_FETCH_AND_ADD:
+			roce_set_field(rc_sq_wqe->byte_4,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
+				       HNS_ROCE_V2_WQE_OP_ATOM_FETCH_AND_ADD);
+			break;
+		case IB_WR_MASKED_ATOMIC_CMP_AND_SWP:
+			roce_set_field(rc_sq_wqe->byte_4,
+				      V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
+				      V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
+				      HNS_ROCE_V2_WQE_OP_ATOM_MSK_CMP_AND_SWAP);
+			break;
+		case IB_WR_MASKED_ATOMIC_FETCH_AND_ADD:
+			roce_set_field(rc_sq_wqe->byte_4,
+				     V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
+				     V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
+				     HNS_ROCE_V2_WQE_OP_ATOM_MSK_FETCH_AND_ADD);
+			break;
+		default:
+			roce_set_field(rc_sq_wqe->byte_4,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
+				       V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
+				       HNS_ROCE_V2_WQE_OP_MASK);
+			break;
+		}
+
+		roce_set_bit(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_CQE_S, 1);
+
+		wqe += sizeof(struct hns_roce_v2_rc_send_wqe);
+		dseg = wqe;
+		if (wr->send_flags & IB_SEND_INLINE && wr->num_sge) {
+			if (rc_sq_wqe->msg_len >
+				hr_dev->caps.max_sq_inline) {
+				ret = -EINVAL;
+				*bad_wr = wr;
+				dev_err(dev, "inline len(1-%d)=%d, illegal",
+					rc_sq_wqe->msg_len,
+					hr_dev->caps.max_sq_inline);
+				goto out;
+			}
+
+			for (i = 0; i < wr->num_sge; i++) {
+				memcpy(wqe, ((void *)wr->sg_list[i].addr),
+				       wr->sg_list[i].length);
+				wqe += wr->sg_list[i].length;
+				wqe_sz += wr->sg_list[i].length;
+			}
+
+			roce_set_bit(rc_sq_wqe->byte_4,
+				     V2_RC_SEND_WQE_BYTE_4_INLINE_S, 1);
+		} else {
+			if (wr->num_sge <= 2) {
+				for (i = 0; i < wr->num_sge; i++)
+					set_data_seg_v2(dseg + i,
+							wr->sg_list + i);
+			} else {
+				roce_set_field(rc_sq_wqe->byte_20,
+				V2_RC_SEND_WQE_BYTE_20_MSG_START_SGE_IDX_M,
+				V2_RC_SEND_WQE_BYTE_20_MSG_START_SGE_IDX_S,
+				sge_ind & (qp->sge.sge_cnt - 1));
+
+				for (i = 0; i < 2; i++)
+					set_data_seg_v2(dseg + i,
+							wr->sg_list + i);
+
+				dseg = get_send_extend_sge(qp,
+					sge_ind & (qp->sge.sge_cnt - 1));
+
+				for (i = 0; i < wr->num_sge - 2; i++) {
+					set_data_seg_v2(dseg + i,
+							wr->sg_list + 2 + i);
+					sge_ind++;
+				}
+			}
+
+			roce_set_field(rc_sq_wqe->byte_16,
+				       V2_RC_SEND_WQE_BYTE_16_SGE_NUM_M,
+				       V2_RC_SEND_WQE_BYTE_16_SGE_NUM_S,
+				       wr->num_sge);
+			wqe_sz += wr->num_sge *
+				  sizeof(struct hns_roce_v2_wqe_data_seg);
+		}
+		ind++;
+	}
+
+out:
+	if (likely(nreq)) {
+		qp->sq.head += nreq;
+		/* Memory barrier */
+		wmb();
+
+		sq_db.byte_4 = 0;
+		sq_db.parameter = 0;
+
+		roce_set_field(sq_db.byte_4, V2_DB_BYTE_4_TAG_M,
+			       V2_DB_BYTE_4_TAG_S, qp->doorbell_qpn);
+		roce_set_field(sq_db.byte_4, V2_DB_BYTE_4_CMD_M,
+			       V2_DB_BYTE_4_CMD_S, HNS_ROCE_V2_SQ_DB);
+		roce_set_field(sq_db.parameter, V2_DB_PARAMETER_CONS_IDX_M,
+			       V2_DB_PARAMETER_CONS_IDX_S,
+			       qp->sq.head & ((qp->sq.wqe_cnt << 1) - 1));
+		roce_set_field(sq_db.parameter, V2_DB_PARAMETER_SL_M,
+			       V2_DB_PARAMETER_SL_S, qp->sl);
+
+		hns_roce_write64_k((__be32 *)&sq_db, qp->sq.db_reg_l);
+
+		qp->sq_next_wqe = ind;
+		qp->next_sge = sge_ind;
+	}
+
+	spin_unlock_irqrestore(&qp->sq.lock, flags);
+
+	return ret;
+}
+
+static int hns_roce_v2_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+				 struct ib_recv_wr **bad_wr)
+{
+	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
+	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
+	struct hns_roce_v2_wqe_data_seg *dseg;
+	struct device *dev = hr_dev->dev;
+	struct hns_roce_v2_db rq_db;
+	unsigned long flags;
+	void *wqe = NULL;
+	int ret = 0;
+	int nreq;
+	int ind;
+	int i;
+
+	spin_lock_irqsave(&hr_qp->rq.lock, flags);
+	ind = hr_qp->rq.head & (hr_qp->rq.wqe_cnt - 1);
+
+	if (hr_qp->state == IB_QPS_RESET || hr_qp->state == IB_QPS_ERR) {
+		spin_unlock_irqrestore(&hr_qp->rq.lock, flags);
+		*bad_wr = wr;
+		return -EINVAL;
+	}
+
+	for (nreq = 0; wr; ++nreq, wr = wr->next) {
+		if (hns_roce_wq_overflow(&hr_qp->rq, nreq,
+			hr_qp->ibqp.recv_cq)) {
+			ret = -ENOMEM;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		if (unlikely(wr->num_sge > hr_qp->rq.max_gs)) {
+			dev_err(dev, "rq:num_sge=%d > qp->sq.max_gs=%d\n",
+				wr->num_sge, hr_qp->rq.max_gs);
+			ret = -EINVAL;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		wqe = get_recv_wqe(hr_qp, ind);
+		dseg = (struct hns_roce_v2_wqe_data_seg *)wqe;
+		for (i = 0; i < wr->num_sge; i++) {
+			if (!wr->sg_list[i].length)
+				continue;
+			set_data_seg_v2(dseg, wr->sg_list + i);
+			dseg++;
+		}
+
+		if (i < hr_qp->rq.max_gs) {
+			dseg[i].lkey = cpu_to_be32(HNS_ROCE_INVALID_LKEY);
+			dseg[i].addr = 0;
+		}
+
+		hr_qp->rq.wrid[ind] = wr->wr_id;
+
+		ind = (ind + 1) & (hr_qp->rq.wqe_cnt - 1);
+	}
+
+out:
+	if (likely(nreq)) {
+		hr_qp->rq.head += nreq;
+		/* Memory barrier */
+		wmb();
+
+		rq_db.byte_4 = 0;
+		rq_db.parameter = 0;
+
+		roce_set_field(rq_db.byte_4, V2_DB_BYTE_4_TAG_M,
+			       V2_DB_BYTE_4_TAG_S, hr_qp->qpn);
+		roce_set_field(rq_db.byte_4, V2_DB_BYTE_4_CMD_M,
+			       V2_DB_BYTE_4_CMD_S, HNS_ROCE_V2_RQ_DB);
+		roce_set_field(rq_db.parameter, V2_DB_PARAMETER_CONS_IDX_M,
+			       V2_DB_PARAMETER_CONS_IDX_S, hr_qp->rq.head);
+
+		hns_roce_write64_k((__be32 *)&rq_db, hr_qp->rq.db_reg_l);
+	}
+	spin_unlock_irqrestore(&hr_qp->rq.lock, flags);
+
+	return ret;
+}
+
 static int hns_roce_cmq_space(struct hns_roce_v2_cmq_ring *ring)
 {
 	int ntu = ring->next_to_use;
@@ -2588,6 +2909,8 @@ static int hns_roce_v2_destroy_qp(struct ib_qp *ibqp)
 	.modify_qp = hns_roce_v2_modify_qp,
 	.query_qp = hns_roce_v2_query_qp,
 	.destroy_qp = hns_roce_v2_destroy_qp,
+	.post_send = hns_roce_v2_post_send,
+	.post_recv = hns_roce_v2_post_recv,
 	.req_notify_cq = hns_roce_v2_req_notify_cq,
 	.poll_cq = hns_roce_v2_poll_cq,
 };
@@ -2612,6 +2935,8 @@ static int hns_roce_hw_v2_get_cfg(struct hns_roce_dev *hr_dev,
 	}
 
 	hr_dev->hw = &hns_roce_hw_v2;
+	hr_dev->sdb_offset = ROCEE_DB_SQ_L_0_REG;
+	hr_dev->odb_offset = hr_dev->sdb_offset;
 
 	/* Get info from NIC driver. */
 	hr_dev->reg_base = handle->rinfo.roce_io_base;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index 0360df0..6172146 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -70,6 +70,7 @@
 #define HNS_ROCE_V2_CQE_ENTRY_SIZE		32
 #define HNS_ROCE_V2_PAGE_SIZE_SUPPORTED		0xFFFFF000
 #define HNS_ROCE_V2_MAX_INNER_MTPT_NUM		2
+#define HNS_ROCE_INVALID_LKEY			0x100
 #define HNS_ROCE_CMQ_TX_TIMEOUT			200
 
 #define HNS_ROCE_CONTEXT_HOP_NUM		1
@@ -111,6 +112,23 @@
 #define HNS_ROCE_V2_CQE_QPN_MASK		0x3ffff
 
 enum {
+	HNS_ROCE_V2_WQE_OP_SEND				= 0x0,
+	HNS_ROCE_V2_WQE_OP_SEND_WITH_INV		= 0x1,
+	HNS_ROCE_V2_WQE_OP_SEND_WITH_IMM		= 0x2,
+	HNS_ROCE_V2_WQE_OP_RDMA_WRITE			= 0x3,
+	HNS_ROCE_V2_WQE_OP_RDMA_WRITE_WITH_IMM		= 0x4,
+	HNS_ROCE_V2_WQE_OP_RDMA_READ			= 0x5,
+	HNS_ROCE_V2_WQE_OP_ATOM_CMP_AND_SWAP		= 0x6,
+	HNS_ROCE_V2_WQE_OP_ATOM_FETCH_AND_ADD		= 0x7,
+	HNS_ROCE_V2_WQE_OP_ATOM_MSK_CMP_AND_SWAP	= 0x8,
+	HNS_ROCE_V2_WQE_OP_ATOM_MSK_FETCH_AND_ADD	= 0x9,
+	HNS_ROCE_V2_WQE_OP_FAST_REG_PMR			= 0xa,
+	HNS_ROCE_V2_WQE_OP_LOCAL_INV			= 0xb,
+	HNS_ROCE_V2_WQE_OP_BIND_MW_TYPE			= 0xc,
+	HNS_ROCE_V2_WQE_OP_MASK				= 0x1f,
+};
+
+enum {
 	HNS_ROCE_SQ_OPCODE_SEND = 0x0,
 	HNS_ROCE_SQ_OPCODE_SEND_WITH_INV = 0x1,
 	HNS_ROCE_SQ_OPCODE_SEND_WITH_IMM = 0x2,
@@ -135,6 +153,9 @@ enum {
 };
 
 enum {
+	HNS_ROCE_V2_SQ_DB	= 0x0,
+	HNS_ROCE_V2_RQ_DB	= 0x1,
+	HNS_ROCE_V2_SRQ_DB	= 0x2,
 	HNS_ROCE_V2_CQ_DB_PTR	= 0x3,
 	HNS_ROCE_V2_CQ_DB_NTR	= 0x4,
 };
@@ -775,6 +796,12 @@ struct hns_roce_v2_cqe {
 #define	V2_DB_BYTE_4_CMD_S 24
 #define V2_DB_BYTE_4_CMD_M GENMASK(27, 24)
 
+#define V2_DB_PARAMETER_CONS_IDX_S 0
+#define V2_DB_PARAMETER_CONS_IDX_M GENMASK(15, 0)
+
+#define V2_DB_PARAMETER_SL_S 16
+#define V2_DB_PARAMETER_SL_M GENMASK(18, 16)
+
 struct hns_roce_v2_cq_db {
 	u32	byte_4;
 	u32	parameter;
@@ -794,6 +821,51 @@ struct hns_roce_v2_cq_db {
 
 #define V2_CQ_DB_PARAMETER_NOTIFY_S 24
 
+struct hns_roce_v2_rc_send_wqe {
+	u32		byte_4;
+	u32		msg_len;
+	u32		inv_key_immtdata;
+	u32		byte_16;
+	u32		byte_20;
+	u32		rkey;
+	u64		va;
+};
+
+#define	V2_RC_SEND_WQE_BYTE_4_OPCODE_S 0
+#define V2_RC_SEND_WQE_BYTE_4_OPCODE_M GENMASK(4, 0)
+
+#define V2_RC_SEND_WQE_BYTE_4_OWNER_S 7
+
+#define V2_RC_SEND_WQE_BYTE_4_CQE_S 8
+
+#define V2_RC_SEND_WQE_BYTE_4_FENCE_S 9
+
+#define V2_RC_SEND_WQE_BYTE_4_SO_S 10
+
+#define V2_RC_SEND_WQE_BYTE_4_SE_S 11
+
+#define V2_RC_SEND_WQE_BYTE_4_INLINE_S 12
+
+#define	V2_RC_SEND_WQE_BYTE_16_XRC_SRQN_S 0
+#define V2_RC_SEND_WQE_BYTE_16_XRC_SRQN_M GENMASK(23, 0)
+
+#define	V2_RC_SEND_WQE_BYTE_16_SGE_NUM_S 24
+#define V2_RC_SEND_WQE_BYTE_16_SGE_NUM_M GENMASK(31, 24)
+
+#define V2_RC_SEND_WQE_BYTE_20_MSG_START_SGE_IDX_S 0
+#define V2_RC_SEND_WQE_BYTE_20_MSG_START_SGE_IDX_M GENMASK(23, 0)
+
+struct hns_roce_v2_wqe_data_seg {
+	__be32    len;
+	__be32    lkey;
+	__be64    addr;
+};
+
+struct hns_roce_v2_db {
+	u32	byte_4;
+	u32	parameter;
+};
+
 struct hns_roce_query_version {
 	__le16 rocee_vendor_id;
 	__le16 rocee_hw_version;
diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c
index c04aa81..e6d1115 100644
--- a/drivers/infiniband/hw/hns/hns_roce_qp.c
+++ b/drivers/infiniband/hw/hns/hns_roce_qp.c
@@ -549,10 +549,9 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev,
 		}
 
 		/* QP doorbell register address */
-		hr_qp->sq.db_reg_l = hr_dev->reg_base + ROCEE_DB_SQ_L_0_REG +
+		hr_qp->sq.db_reg_l = hr_dev->reg_base + hr_dev->sdb_offset +
 				     DB_REG_OFFSET * hr_dev->priv_uar.index;
-		hr_qp->rq.db_reg_l = hr_dev->reg_base +
-				     ROCEE_DB_OTHERS_L_0_REG +
+		hr_qp->rq.db_reg_l = hr_dev->reg_base + hr_dev->odb_offset +
 				     DB_REG_OFFSET * hr_dev->priv_uar.index;
 
 		/* Allocate QP buf */
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 17/20] RDMA/hns: Configure the MTPT in hip08

[Wei Hu (Xavier)]

The MTPT records the attribute of the registered MR. The MTPT format
will be updated in hip08, and the MTPT should be configured.

This patch is to configure the MTPT for the registered MR in hip08.

Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 94 ++++++++++++++++++++++++++++++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h | 68 +++++++++++++++++++++
 2 files changed, 162 insertions(+)

diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index b614177..4171f73 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -1086,6 +1086,99 @@ static void hns_roce_v2_set_mac(struct hns_roce_dev *hr_dev, u8 phy_port,
 	roce_write(hr_dev, ROCEE_VF_SMAC_CFG1_REG + 0x08 * phy_port, val);
 }
 
+static int hns_roce_v2_write_mtpt(void *mb_buf, struct hns_roce_mr *mr,
+				  unsigned long mtpt_idx)
+{
+	struct hns_roce_v2_mpt_entry *mpt_entry;
+	struct scatterlist *sg;
+	u64 *pages;
+	int entry;
+	int i;
+
+	mpt_entry = mb_buf;
+	memset(mpt_entry, 0, sizeof(*mpt_entry));
+
+	roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_MPT_ST_M,
+		       V2_MPT_BYTE_4_MPT_ST_S, V2_MPT_ST_VALID);
+	roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_PBL_HOP_NUM_M,
+		       V2_MPT_BYTE_4_PBL_HOP_NUM_S, mr->pbl_hop_num ==
+		       HNS_ROCE_HOP_NUM_0 ? 0 : mr->pbl_hop_num);
+	roce_set_field(mpt_entry->byte_4_pd_hop_st,
+		       V2_MPT_BYTE_4_PBL_BA_PG_SZ_M,
+		       V2_MPT_BYTE_4_PBL_BA_PG_SZ_S, mr->pbl_ba_pg_sz);
+	roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_PD_M,
+		       V2_MPT_BYTE_4_PD_S, mr->pd);
+	mpt_entry->byte_4_pd_hop_st = cpu_to_le32(mpt_entry->byte_4_pd_hop_st);
+
+	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_RA_EN_S, 0);
+	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_R_INV_EN_S, 1);
+	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_L_INV_EN_S, 0);
+	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_BIND_EN_S,
+		     (mr->access & IB_ACCESS_MW_BIND ? 1 : 0));
+	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_ATOMIC_EN_S, 0);
+	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_RR_EN_S,
+		     (mr->access & IB_ACCESS_REMOTE_READ ? 1 : 0));
+	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_RW_EN_S,
+		     (mr->access & IB_ACCESS_REMOTE_WRITE ? 1 : 0));
+	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_LW_EN_S,
+		     (mr->access & IB_ACCESS_LOCAL_WRITE ? 1 : 0));
+	mpt_entry->byte_8_mw_cnt_en = cpu_to_le32(mpt_entry->byte_8_mw_cnt_en);
+
+	roce_set_bit(mpt_entry->byte_12_mw_pa, V2_MPT_BYTE_12_PA_S,
+		     mr->type == MR_TYPE_MR ? 0 : 1);
+	mpt_entry->byte_12_mw_pa = cpu_to_le32(mpt_entry->byte_12_mw_pa);
+
+	mpt_entry->len_l = cpu_to_le32(lower_32_bits(mr->size));
+	mpt_entry->len_h = cpu_to_le32(upper_32_bits(mr->size));
+	mpt_entry->lkey = cpu_to_le32(mr->key);
+	mpt_entry->va_l = cpu_to_le32(lower_32_bits(mr->iova));
+	mpt_entry->va_h = cpu_to_le32(upper_32_bits(mr->iova));
+
+	if (mr->type == MR_TYPE_DMA)
+		return 0;
+
+	mpt_entry->pbl_size = cpu_to_le32(mr->pbl_size);
+
+	mpt_entry->pbl_ba_l = cpu_to_le32(lower_32_bits(mr->pbl_ba >> 3));
+	roce_set_field(mpt_entry->byte_48_mode_ba, V2_MPT_BYTE_48_PBL_BA_H_M,
+		       V2_MPT_BYTE_48_PBL_BA_H_S,
+		       upper_32_bits(mr->pbl_ba >> 3));
+	mpt_entry->byte_48_mode_ba = cpu_to_le32(mpt_entry->byte_48_mode_ba);
+
+	pages = (u64 *)__get_free_page(GFP_KERNEL);
+	if (!pages)
+		return -ENOMEM;
+
+	i = 0;
+	for_each_sg(mr->umem->sg_head.sgl, sg, mr->umem->nmap, entry) {
+		pages[i] = ((u64)sg_dma_address(sg)) >> 6;
+
+		/* Record the first 2 entry directly to MTPT table */
+		if (i >= HNS_ROCE_V2_MAX_INNER_MTPT_NUM - 1)
+			break;
+		i++;
+	}
+
+	mpt_entry->pa0_l = cpu_to_le32(lower_32_bits(pages[0]));
+	roce_set_field(mpt_entry->byte_56_pa0_h, V2_MPT_BYTE_56_PA0_H_M,
+		       V2_MPT_BYTE_56_PA0_H_S,
+		       upper_32_bits(pages[0]));
+	mpt_entry->byte_56_pa0_h = cpu_to_le32(mpt_entry->byte_56_pa0_h);
+
+	mpt_entry->pa1_l = cpu_to_le32(lower_32_bits(pages[1]));
+	roce_set_field(mpt_entry->byte_64_buf_pa1, V2_MPT_BYTE_64_PA1_H_M,
+		       V2_MPT_BYTE_64_PA1_H_S, upper_32_bits(pages[1]));
+
+	free_page((unsigned long)pages);
+
+	roce_set_field(mpt_entry->byte_64_buf_pa1,
+		       V2_MPT_BYTE_64_PBL_BUF_PG_SZ_M,
+		       V2_MPT_BYTE_64_PBL_BUF_PG_SZ_S, mr->pbl_buf_pg_sz);
+	mpt_entry->byte_64_buf_pa1 = cpu_to_le32(mpt_entry->byte_64_buf_pa1);
+
+	return 0;
+}
+
 static void *get_cqe_v2(struct hns_roce_cq *hr_cq, int n)
 {
 	return hns_roce_buf_offset(&hr_cq->hr_buf.hr_buf,
@@ -2903,6 +2996,7 @@ static int hns_roce_v2_destroy_qp(struct ib_qp *ibqp)
 	.chk_mbox = hns_roce_v2_chk_mbox,
 	.set_gid = hns_roce_v2_set_gid,
 	.set_mac = hns_roce_v2_set_mac,
+	.write_mtpt = hns_roce_v2_write_mtpt,
 	.write_cqc = hns_roce_v2_write_cqc,
 	.set_hem = hns_roce_v2_set_hem,
 	.clear_hem = hns_roce_v2_clear_hem,
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index 6172146..4fc4acd 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -790,6 +790,74 @@ struct hns_roce_v2_cqe {
 
 #define	V2_CQE_BYTE_32_LPK_S 31
 
+struct hns_roce_v2_mpt_entry {
+	__le32	byte_4_pd_hop_st;
+	__le32	byte_8_mw_cnt_en;
+	__le32	byte_12_mw_pa;
+	__le32	bound_lkey;
+	__le32	len_l;
+	__le32	len_h;
+	__le32	lkey;
+	__le32	va_l;
+	__le32	va_h;
+	__le32	pbl_size;
+	__le32	pbl_ba_l;
+	__le32	byte_48_mode_ba;
+	__le32	pa0_l;
+	__le32	byte_56_pa0_h;
+	__le32	pa1_l;
+	__le32	byte_64_buf_pa1;
+};
+
+#define V2_MPT_BYTE_4_MPT_ST_S 0
+#define V2_MPT_BYTE_4_MPT_ST_M GENMASK(1, 0)
+
+#define V2_MPT_BYTE_4_PBL_HOP_NUM_S 2
+#define V2_MPT_BYTE_4_PBL_HOP_NUM_M GENMASK(3, 2)
+
+#define V2_MPT_BYTE_4_PBL_BA_PG_SZ_S 4
+#define V2_MPT_BYTE_4_PBL_BA_PG_SZ_M GENMASK(7, 4)
+
+#define V2_MPT_BYTE_4_PD_S 8
+#define V2_MPT_BYTE_4_PD_M GENMASK(31, 8)
+
+#define V2_MPT_BYTE_8_RA_EN_S 0
+
+#define V2_MPT_BYTE_8_R_INV_EN_S 1
+
+#define V2_MPT_BYTE_8_L_INV_EN_S 2
+
+#define V2_MPT_BYTE_8_BIND_EN_S 3
+
+#define V2_MPT_BYTE_8_ATOMIC_EN_S 4
+
+#define V2_MPT_BYTE_8_RR_EN_S 5
+
+#define V2_MPT_BYTE_8_RW_EN_S 6
+
+#define V2_MPT_BYTE_8_LW_EN_S 7
+
+#define V2_MPT_BYTE_12_PA_S 1
+
+#define V2_MPT_BYTE_12_INNER_PA_VLD_S 7
+
+#define V2_MPT_BYTE_12_MW_BIND_QPN_S 8
+#define V2_MPT_BYTE_12_MW_BIND_QPN_M GENMASK(31, 8)
+
+#define V2_MPT_BYTE_48_PBL_BA_H_S 0
+#define V2_MPT_BYTE_48_PBL_BA_H_M GENMASK(28, 0)
+
+#define V2_MPT_BYTE_48_BLK_MODE_S 29
+
+#define V2_MPT_BYTE_56_PA0_H_S 0
+#define V2_MPT_BYTE_56_PA0_H_M GENMASK(25, 0)
+
+#define V2_MPT_BYTE_64_PA1_H_S 0
+#define V2_MPT_BYTE_64_PA1_H_M GENMASK(25, 0)
+
+#define V2_MPT_BYTE_64_PBL_BUF_PG_SZ_S 28
+#define V2_MPT_BYTE_64_PBL_BUF_PG_SZ_M GENMASK(31, 28)
+
 #define	V2_DB_BYTE_4_TAG_S 0
 #define V2_DB_BYTE_4_TAG_M GENMASK(23, 0)
 
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 18/20] RDMA/hns: Add releasing resource operation in error branch

[Wei Hu (Xavier)]

This patch adds releasing resource operation in error branch
of the function named hns_roce_table_get.

Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_hem.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c
index 8a3e174..125e26e 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hem.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hem.c
@@ -563,6 +563,8 @@ int hns_roce_table_get(struct hns_roce_dev *hr_dev,
 
 	/* Set HEM base address(128K/page, pa) to Hardware */
 	if (hns_roce_set_hem(hr_dev, table, obj)) {
+		hns_roce_free_hem(hr_dev, table->hem[i]);
+		table->hem[i] = NULL;
 		ret = -ENODEV;
 		dev_err(dev, "set HEM base address to HW failed.\n");
 		goto out;
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 19/20] RDMA/hns: Replace condition statement using hardware version information

[Wei Hu (Xavier)]

This patch replaces condition statement to reduce usage of hardware version
information in common driver.

Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_main.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c
index a110f96..7a0c1e8 100644
--- a/drivers/infiniband/hw/hns/hns_roce_main.c
+++ b/drivers/infiniband/hw/hns/hns_roce_main.c
@@ -381,7 +381,8 @@ static int hns_roce_mmap(struct ib_ucontext *context,
 				       to_hr_ucontext(context)->uar.pfn,
 				       PAGE_SIZE, vma->vm_page_prot))
 			return -EAGAIN;
-	} else if (vma->vm_pgoff == 1 && hr_dev->hw_rev == HNS_ROCE_HW_VER1) {
+	} else if (vma->vm_pgoff == 1 && hr_dev->tptr_dma_addr &&
+		   hr_dev->tptr_size) {
 		/* vm_pgoff: 1 -- TPTR */
 		if (io_remap_pfn_range(vma, vma->vm_start,
 				       hr_dev->tptr_dma_addr >> PAGE_SHIFT,
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH for-next 20/20] RDMA/hns: Fix inconsistent warning

[Wei Hu (Xavier)]

This patch fixes smatch inconsistent warning as below:
drivers/infiniband/hw/hns/hns_roce_hem.h:136 hns_roce_hem_first() warn: inconsistent indenting

Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 drivers/infiniband/hw/hns/hns_roce_hem.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.h b/drivers/infiniband/hw/hns/hns_roce_hem.h
index 1aa54b6..af28bbf 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hem.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hem.h
@@ -133,7 +133,7 @@ static inline void hns_roce_hem_first(struct hns_roce_hem *hem,
 	iter->chunk = list_empty(&hem->chunk_list) ? NULL :
 				 list_entry(hem->chunk_list.next,
 					    struct hns_roce_hem_chunk, list);
-	 iter->page_idx = 0;
+	iter->page_idx = 0;
 }
 
 static inline int hns_roce_hem_last(struct hns_roce_hem_iter *iter)
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 11/20] RDMA/hns: Split CQE from MTT in hip08

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 1360 bytes --]

On Wed, Aug 30, 2017 at 05:23:09PM +0800, Wei Hu (Xavier) wrote:
> From: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
>
> In hip08, the SQWQE/SGE/RQWQE and CQE have different hop num and
> page size, so we need to manage the base address table of the
> SQWQE/SGE/RQWQE and CQE separately.
>
> This patch is to split CQE from MTT(SQWQE/SGE/RQWQE).
>
> Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
> ---
>  drivers/infiniband/hw/hns/hns_roce_cq.c     | 15 ++++++-
>  drivers/infiniband/hw/hns/hns_roce_device.h | 14 ++++--
>  drivers/infiniband/hw/hns/hns_roce_hem.c    |  3 ++
>  drivers/infiniband/hw/hns/hns_roce_main.c   | 17 ++++++++
>  drivers/infiniband/hw/hns/hns_roce_mr.c     | 66 ++++++++++++++++++++++-------
>  drivers/infiniband/hw/hns/hns_roce_qp.c     |  2 +
>  6 files changed, 98 insertions(+), 19 deletions(-)

<...>

>
> --- a/drivers/infiniband/hw/hns/hns_roce_device.h
> +++ b/drivers/infiniband/hw/hns/hns_roce_device.h
> @@ -170,6 +170,11 @@ enum {
>  	HNS_ROCE_OPCODE_RDMA_WITH_IMM_RECEIVE	= 0x07,
>  };
>
> +enum hns_roce_mtt_type {
> +	MTT_TYPE_WQE = 0,
> +	MTT_TYPE_CQE,
> +};

No need to initialize enum to zero.
Thanks

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH for-next 00/20] RDMA/hns: Add hip08 RoCE driver support

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 3444 bytes --]

On Wed, Aug 30, 2017 at 05:22:58PM +0800, Wei Hu (Xavier) wrote:
> This patchset adds support for the HiSilicon RoCE engine in the hip08
> SoC.
>
> The driver is refactored so it can support both platform and pci device.
> And hip08 RoCE engine is a pci device, hip06 RoCE engine is a platform
> device.
>
> New hardware layer file hns_roce_hw_v2.c is added for hw v2 support,
> which also includes pci device probing and initialization.
>
> Common functionality is still in hns-roce driver, along with device
> initialization.
>
> This patchset depends on hns3 NIC driver which had been accepted by David,
> and is based on for-next branch of the repo:
> https://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma.git
>
> Shaobo Xu (3):
>   RDMA/hns: Add the interfaces to support multi hop addressing for the
>     contexts in hip08
>   RDMA/hns: Update the interfaces for MTT/CQE multi hop addressing in
>     hip08
>   RDMA/hns: Split CQE from MTT in hip08
>
> Wei Hu (Xavier) (17):
>   RDMA/hns: Split hw v1 driver from hns roce driver
>   RDMA/hns: Move priv in order to add multiple hns_roce support
>   RDMA/hns: Initialize the PCI device for hip08 RoCE
>   RDMA/hns: Modify assignment device variable to support both PCI device
>     and platform device
>   RDMA/hns: Add command queue support for hip08 RoCE driver
>   RDMA/hns: Add profile support for hip08 driver
>   RDMA/hns: Add mailbox's implementation for hip08 RoCE driver
>   RDMA/hns: Configure BT BA and BT attribute for the contexts in hip08
>   RDMA/hns: Support multi hop addressing for PBL in hip08
>   RDMA/hns: Configure mac&gid and user access region for hip08 RoCE
>     driver
>   RDMA/hns: Add CQ operations support for hip08 RoCE driver
>   RDMA/hns: Add QP operations support for hip08 SoC
>   RDMA/hns: Add support for processing send wr and receive wr
>   RDMA/hns: Configure the MTPT in hip08
>   RDMA/hns: Add releasing resource operation in error branch
>   RDMA/hns: Replace condition statement using hardware version
>     information
>   RDMA/hns: Fix inconsistent warning
>
>  drivers/infiniband/hw/hns/Kconfig           |   25 +-
>  drivers/infiniband/hw/hns/Makefile          |    8 +-
>  drivers/infiniband/hw/hns/hns_roce_ah.c     |    2 +-
>  drivers/infiniband/hw/hns/hns_roce_alloc.c  |    8 +-
>  drivers/infiniband/hw/hns/hns_roce_cmd.c    |  107 +-
>  drivers/infiniband/hw/hns/hns_roce_cmd.h    |   50 +
>  drivers/infiniband/hw/hns/hns_roce_common.h |   23 +
>  drivers/infiniband/hw/hns/hns_roce_cq.c     |   72 +-
>  drivers/infiniband/hw/hns/hns_roce_device.h |   98 +-
>  drivers/infiniband/hw/hns/hns_roce_hem.c    |  699 +++++-
>  drivers/infiniband/hw/hns/hns_roce_hem.h    |   32 +-
>  drivers/infiniband/hw/hns/hns_roce_hw_v1.c  |  401 +++-
>  drivers/infiniband/hw/hns/hns_roce_hw_v1.h  |    5 +
>  drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 3128 +++++++++++++++++++++++++++
>  drivers/infiniband/hw/hns/hns_roce_hw_v2.h  | 1165 ++++++++++
>  drivers/infiniband/hw/hns/hns_roce_main.c   |  324 +--
>  drivers/infiniband/hw/hns/hns_roce_mr.c     |  513 ++++-
>  drivers/infiniband/hw/hns/hns_roce_pd.c     |   20 +-
>  drivers/infiniband/hw/hns/hns_roce_qp.c     |  179 +-
>  19 files changed, 6318 insertions(+), 541 deletions(-)
>  create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>  create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.h
>
> --

As far as I can see, it is ready to be merged.

Thanks,

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH for-next 00/20] RDMA/hns: Add hip08 RoCE driver support

[Wei Hu (Xavier)]

On 2017/9/14 1:55, Leon Romanovsky wrote:
> On Wed, Aug 30, 2017 at 05:22:58PM +0800, Wei Hu (Xavier) wrote:
>> This patchset adds support for the HiSilicon RoCE engine in the hip08
>> SoC.
>>
>> The driver is refactored so it can support both platform and pci device.
>> And hip08 RoCE engine is a pci device, hip06 RoCE engine is a platform
>> device.
>>
>> New hardware layer file hns_roce_hw_v2.c is added for hw v2 support,
>> which also includes pci device probing and initialization.
>>
>> Common functionality is still in hns-roce driver, along with device
>> initialization.
>>
>> This patchset depends on hns3 NIC driver which had been accepted by David,
>> and is based on for-next branch of the repo:
>> https://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma.git
>>
>> Shaobo Xu (3):
>>    RDMA/hns: Add the interfaces to support multi hop addressing for the
>>      contexts in hip08
>>    RDMA/hns: Update the interfaces for MTT/CQE multi hop addressing in
>>      hip08
>>    RDMA/hns: Split CQE from MTT in hip08
>>
>> Wei Hu (Xavier) (17):
>>    RDMA/hns: Split hw v1 driver from hns roce driver
>>    RDMA/hns: Move priv in order to add multiple hns_roce support
>>    RDMA/hns: Initialize the PCI device for hip08 RoCE
>>    RDMA/hns: Modify assignment device variable to support both PCI device
>>      and platform device
>>    RDMA/hns: Add command queue support for hip08 RoCE driver
>>    RDMA/hns: Add profile support for hip08 driver
>>    RDMA/hns: Add mailbox's implementation for hip08 RoCE driver
>>    RDMA/hns: Configure BT BA and BT attribute for the contexts in hip08
>>    RDMA/hns: Support multi hop addressing for PBL in hip08
>>    RDMA/hns: Configure mac&gid and user access region for hip08 RoCE
>>      driver
>>    RDMA/hns: Add CQ operations support for hip08 RoCE driver
>>    RDMA/hns: Add QP operations support for hip08 SoC
>>    RDMA/hns: Add support for processing send wr and receive wr
>>    RDMA/hns: Configure the MTPT in hip08
>>    RDMA/hns: Add releasing resource operation in error branch
>>    RDMA/hns: Replace condition statement using hardware version
>>      information
>>    RDMA/hns: Fix inconsistent warning
>>
>>   drivers/infiniband/hw/hns/Kconfig           |   25 +-
>>   drivers/infiniband/hw/hns/Makefile          |    8 +-
>>   drivers/infiniband/hw/hns/hns_roce_ah.c     |    2 +-
>>   drivers/infiniband/hw/hns/hns_roce_alloc.c  |    8 +-
>>   drivers/infiniband/hw/hns/hns_roce_cmd.c    |  107 +-
>>   drivers/infiniband/hw/hns/hns_roce_cmd.h    |   50 +
>>   drivers/infiniband/hw/hns/hns_roce_common.h |   23 +
>>   drivers/infiniband/hw/hns/hns_roce_cq.c     |   72 +-
>>   drivers/infiniband/hw/hns/hns_roce_device.h |   98 +-
>>   drivers/infiniband/hw/hns/hns_roce_hem.c    |  699 +++++-
>>   drivers/infiniband/hw/hns/hns_roce_hem.h    |   32 +-
>>   drivers/infiniband/hw/hns/hns_roce_hw_v1.c  |  401 +++-
>>   drivers/infiniband/hw/hns/hns_roce_hw_v1.h  |    5 +
>>   drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 3128 +++++++++++++++++++++++++++
>>   drivers/infiniband/hw/hns/hns_roce_hw_v2.h  | 1165 ++++++++++
>>   drivers/infiniband/hw/hns/hns_roce_main.c   |  324 +--
>>   drivers/infiniband/hw/hns/hns_roce_mr.c     |  513 ++++-
>>   drivers/infiniband/hw/hns/hns_roce_pd.c     |   20 +-
>>   drivers/infiniband/hw/hns/hns_roce_qp.c     |  179 +-
>>   19 files changed, 6318 insertions(+), 541 deletions(-)
>>   create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>>   create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.h
>>
>> --
> As far as I can see, it is ready to be merged.
>
> Thanks,
Hi, Leon
     Thanks for your comments.

     And about the comment on the No.11 patch of this patchset,
     Which option we should do?  Can you give us some suggestions?
     1. send PATCH V2 to fix it.
     2. After this patchset will be merged, send a new patch to fix it.

     The link about the comment: 
https://www.spinics.net/lists/linux-rdma/msg54392.html

     Regards
Wei Hu


--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 00/20] RDMA/hns: Add hip08 RoCE driver support

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 4780 bytes --]

On Thu, Sep 14, 2017 at 03:45:53PM +0800, Wei Hu (Xavier) wrote:
>
>
> On 2017/9/14 1:55, Leon Romanovsky wrote:
> > On Wed, Aug 30, 2017 at 05:22:58PM +0800, Wei Hu (Xavier) wrote:
> > > This patchset adds support for the HiSilicon RoCE engine in the hip08
> > > SoC.
> > >
> > > The driver is refactored so it can support both platform and pci device.
> > > And hip08 RoCE engine is a pci device, hip06 RoCE engine is a platform
> > > device.
> > >
> > > New hardware layer file hns_roce_hw_v2.c is added for hw v2 support,
> > > which also includes pci device probing and initialization.
> > >
> > > Common functionality is still in hns-roce driver, along with device
> > > initialization.
> > >
> > > This patchset depends on hns3 NIC driver which had been accepted by David,
> > > and is based on for-next branch of the repo:
> > > https://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma.git
> > >
> > > Shaobo Xu (3):
> > >    RDMA/hns: Add the interfaces to support multi hop addressing for the
> > >      contexts in hip08
> > >    RDMA/hns: Update the interfaces for MTT/CQE multi hop addressing in
> > >      hip08
> > >    RDMA/hns: Split CQE from MTT in hip08
> > >
> > > Wei Hu (Xavier) (17):
> > >    RDMA/hns: Split hw v1 driver from hns roce driver
> > >    RDMA/hns: Move priv in order to add multiple hns_roce support
> > >    RDMA/hns: Initialize the PCI device for hip08 RoCE
> > >    RDMA/hns: Modify assignment device variable to support both PCI device
> > >      and platform device
> > >    RDMA/hns: Add command queue support for hip08 RoCE driver
> > >    RDMA/hns: Add profile support for hip08 driver
> > >    RDMA/hns: Add mailbox's implementation for hip08 RoCE driver
> > >    RDMA/hns: Configure BT BA and BT attribute for the contexts in hip08
> > >    RDMA/hns: Support multi hop addressing for PBL in hip08
> > >    RDMA/hns: Configure mac&gid and user access region for hip08 RoCE
> > >      driver
> > >    RDMA/hns: Add CQ operations support for hip08 RoCE driver
> > >    RDMA/hns: Add QP operations support for hip08 SoC
> > >    RDMA/hns: Add support for processing send wr and receive wr
> > >    RDMA/hns: Configure the MTPT in hip08
> > >    RDMA/hns: Add releasing resource operation in error branch
> > >    RDMA/hns: Replace condition statement using hardware version
> > >      information
> > >    RDMA/hns: Fix inconsistent warning
> > >
> > >   drivers/infiniband/hw/hns/Kconfig           |   25 +-
> > >   drivers/infiniband/hw/hns/Makefile          |    8 +-
> > >   drivers/infiniband/hw/hns/hns_roce_ah.c     |    2 +-
> > >   drivers/infiniband/hw/hns/hns_roce_alloc.c  |    8 +-
> > >   drivers/infiniband/hw/hns/hns_roce_cmd.c    |  107 +-
> > >   drivers/infiniband/hw/hns/hns_roce_cmd.h    |   50 +
> > >   drivers/infiniband/hw/hns/hns_roce_common.h |   23 +
> > >   drivers/infiniband/hw/hns/hns_roce_cq.c     |   72 +-
> > >   drivers/infiniband/hw/hns/hns_roce_device.h |   98 +-
> > >   drivers/infiniband/hw/hns/hns_roce_hem.c    |  699 +++++-
> > >   drivers/infiniband/hw/hns/hns_roce_hem.h    |   32 +-
> > >   drivers/infiniband/hw/hns/hns_roce_hw_v1.c  |  401 +++-
> > >   drivers/infiniband/hw/hns/hns_roce_hw_v1.h  |    5 +
> > >   drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 3128 +++++++++++++++++++++++++++
> > >   drivers/infiniband/hw/hns/hns_roce_hw_v2.h  | 1165 ++++++++++
> > >   drivers/infiniband/hw/hns/hns_roce_main.c   |  324 +--
> > >   drivers/infiniband/hw/hns/hns_roce_mr.c     |  513 ++++-
> > >   drivers/infiniband/hw/hns/hns_roce_pd.c     |   20 +-
> > >   drivers/infiniband/hw/hns/hns_roce_qp.c     |  179 +-
> > >   19 files changed, 6318 insertions(+), 541 deletions(-)
> > >   create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.c
> > >   create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.h
> > >
> > > --
> > As far as I can see, it is ready to be merged.
> >
> > Thanks,
> Hi, Leon
>     Thanks for your comments.
>
>     And about the comment on the No.11 patch of this patchset,
>     Which option we should do?  Can you give us some suggestions?
>     1. send PATCH V2 to fix it.
>     2. After this patchset will be merged, send a new patch to fix it.

In your case (20 patches, many LOCs), it will be enough to send a followup patch
to remove it, however you don't need to wait till it is merged and you
can send it now with note placed under "---" markings that it is based
on this series.

Thanks

>
>     The link about the comment:
> https://www.spinics.net/lists/linux-rdma/msg54392.html
>
>     Regards
> Wei Hu
>
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH for-next 11/20] RDMA/hns: Split CQE from MTT in hip08

[Wei Hu (Xavier)]

On 2017/9/14 1:52, Leon Romanovsky wrote:
> On Wed, Aug 30, 2017 at 05:23:09PM +0800, Wei Hu (Xavier) wrote:
>> From: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
>>
>> In hip08, the SQWQE/SGE/RQWQE and CQE have different hop num and
>> page size, so we need to manage the base address table of the
>> SQWQE/SGE/RQWQE and CQE separately.
>>
>> This patch is to split CQE from MTT(SQWQE/SGE/RQWQE).
>>
>> Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
>> Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
>> Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
>> ---
>>   drivers/infiniband/hw/hns/hns_roce_cq.c     | 15 ++++++-
>>   drivers/infiniband/hw/hns/hns_roce_device.h | 14 ++++--
>>   drivers/infiniband/hw/hns/hns_roce_hem.c    |  3 ++
>>   drivers/infiniband/hw/hns/hns_roce_main.c   | 17 ++++++++
>>   drivers/infiniband/hw/hns/hns_roce_mr.c     | 66 ++++++++++++++++++++++-------
>>   drivers/infiniband/hw/hns/hns_roce_qp.c     |  2 +
>>   6 files changed, 98 insertions(+), 19 deletions(-)
> <...>
>
>> --- a/drivers/infiniband/hw/hns/hns_roce_device.h
>> +++ b/drivers/infiniband/hw/hns/hns_roce_device.h
>> @@ -170,6 +170,11 @@ enum {
>>   	HNS_ROCE_OPCODE_RDMA_WITH_IMM_RECEIVE	= 0x07,
>>   };
>>
>> +enum hns_roce_mtt_type {
>> +	MTT_TYPE_WQE = 0,
>> +	MTT_TYPE_CQE,
>> +};
> No need to initialize enum to zero.
> Thanks
Hi, Leon
     We will send a followup patch to fix it at your suggestion.
     Thanks

     Regards
Wei Hu


--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 00/20] RDMA/hns: Add hip08 RoCE driver support

[Wei Hu (Xavier)]

On 2017/9/14 20:43, Leon Romanovsky wrote:
> On Thu, Sep 14, 2017 at 03:45:53PM +0800, Wei Hu (Xavier) wrote:
>>
>> On 2017/9/14 1:55, Leon Romanovsky wrote:
>>> On Wed, Aug 30, 2017 at 05:22:58PM +0800, Wei Hu (Xavier) wrote:
>>>> This patchset adds support for the HiSilicon RoCE engine in the hip08
>>>> SoC.
>>>>
>>>> The driver is refactored so it can support both platform and pci device.
>>>> And hip08 RoCE engine is a pci device, hip06 RoCE engine is a platform
>>>> device.
>>>>
>>>> New hardware layer file hns_roce_hw_v2.c is added for hw v2 support,
>>>> which also includes pci device probing and initialization.
>>>>
>>>> Common functionality is still in hns-roce driver, along with device
>>>> initialization.
>>>>
>>>> This patchset depends on hns3 NIC driver which had been accepted by David,
>>>> and is based on for-next branch of the repo:
>>>> https://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma.git
>>>>
>>>> Shaobo Xu (3):
>>>>     RDMA/hns: Add the interfaces to support multi hop addressing for the
>>>>       contexts in hip08
>>>>     RDMA/hns: Update the interfaces for MTT/CQE multi hop addressing in
>>>>       hip08
>>>>     RDMA/hns: Split CQE from MTT in hip08
>>>>
>>>> Wei Hu (Xavier) (17):
>>>>     RDMA/hns: Split hw v1 driver from hns roce driver
>>>>     RDMA/hns: Move priv in order to add multiple hns_roce support
>>>>     RDMA/hns: Initialize the PCI device for hip08 RoCE
>>>>     RDMA/hns: Modify assignment device variable to support both PCI device
>>>>       and platform device
>>>>     RDMA/hns: Add command queue support for hip08 RoCE driver
>>>>     RDMA/hns: Add profile support for hip08 driver
>>>>     RDMA/hns: Add mailbox's implementation for hip08 RoCE driver
>>>>     RDMA/hns: Configure BT BA and BT attribute for the contexts in hip08
>>>>     RDMA/hns: Support multi hop addressing for PBL in hip08
>>>>     RDMA/hns: Configure mac&gid and user access region for hip08 RoCE
>>>>       driver
>>>>     RDMA/hns: Add CQ operations support for hip08 RoCE driver
>>>>     RDMA/hns: Add QP operations support for hip08 SoC
>>>>     RDMA/hns: Add support for processing send wr and receive wr
>>>>     RDMA/hns: Configure the MTPT in hip08
>>>>     RDMA/hns: Add releasing resource operation in error branch
>>>>     RDMA/hns: Replace condition statement using hardware version
>>>>       information
>>>>     RDMA/hns: Fix inconsistent warning
>>>>
>>>>    drivers/infiniband/hw/hns/Kconfig           |   25 +-
>>>>    drivers/infiniband/hw/hns/Makefile          |    8 +-
>>>>    drivers/infiniband/hw/hns/hns_roce_ah.c     |    2 +-
>>>>    drivers/infiniband/hw/hns/hns_roce_alloc.c  |    8 +-
>>>>    drivers/infiniband/hw/hns/hns_roce_cmd.c    |  107 +-
>>>>    drivers/infiniband/hw/hns/hns_roce_cmd.h    |   50 +
>>>>    drivers/infiniband/hw/hns/hns_roce_common.h |   23 +
>>>>    drivers/infiniband/hw/hns/hns_roce_cq.c     |   72 +-
>>>>    drivers/infiniband/hw/hns/hns_roce_device.h |   98 +-
>>>>    drivers/infiniband/hw/hns/hns_roce_hem.c    |  699 +++++-
>>>>    drivers/infiniband/hw/hns/hns_roce_hem.h    |   32 +-
>>>>    drivers/infiniband/hw/hns/hns_roce_hw_v1.c  |  401 +++-
>>>>    drivers/infiniband/hw/hns/hns_roce_hw_v1.h  |    5 +
>>>>    drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 3128 +++++++++++++++++++++++++++
>>>>    drivers/infiniband/hw/hns/hns_roce_hw_v2.h  | 1165 ++++++++++
>>>>    drivers/infiniband/hw/hns/hns_roce_main.c   |  324 +--
>>>>    drivers/infiniband/hw/hns/hns_roce_mr.c     |  513 ++++-
>>>>    drivers/infiniband/hw/hns/hns_roce_pd.c     |   20 +-
>>>>    drivers/infiniband/hw/hns/hns_roce_qp.c     |  179 +-
>>>>    19 files changed, 6318 insertions(+), 541 deletions(-)
>>>>    create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>>>>    create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.h
>>>>
>>>> --
>>> As far as I can see, it is ready to be merged.
>>>
>>> Thanks,
>> Hi, Leon
>>      Thanks for your comments.
>>
>>      And about the comment on the No.11 patch of this patchset,
>>      Which option we should do?  Can you give us some suggestions?
>>      1. send PATCH V2 to fix it.
>>      2. After this patchset will be merged, send a new patch to fix it.
> In your case (20 patches, many LOCs), it will be enough to send a followup patch
> to remove it, however you don't need to wait till it is merged and you
> can send it now with note placed under "---" markings that it is based
> on this series.
>
> Thanks
Thanks , Leon
We will send a followup patch to fix it at your suggestion.

Regards
Wei Hu
>>      The link about the comment:
>> https://www.spinics.net/lists/linux-rdma/msg54392.html
>>
>>      Regards
>> Wei Hu
>>
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
>> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html


--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 11/20] RDMA/hns: Split CQE from MTT in hip08

[Wei Hu (Xavier)]

Hi, Doug
     We resend the replied email, because it is in the window:
         From 20:30 UTC Sept 14 and 20:30 UTC Sept 15
     Thanks.

    Regards
Wei Hu

On 2017/9/15 9:09, Wei Hu (Xavier) wrote:
>
>
> On 2017/9/14 1:52, Leon Romanovsky wrote:
>> On Wed, Aug 30, 2017 at 05:23:09PM +0800, Wei Hu (Xavier) wrote:
>>> From: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
>>>
>>> In hip08, the SQWQE/SGE/RQWQE and CQE have different hop num and
>>> page size, so we need to manage the base address table of the
>>> SQWQE/SGE/RQWQE and CQE separately.
>>>
>>> This patch is to split CQE from MTT(SQWQE/SGE/RQWQE).
>>>
>>> Signed-off-by: Shaobo Xu <xushaobo2-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
>>> Signed-off-by: Lijun Ou <oulijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
>>> Signed-off-by: Wei Hu (Xavier) <xavier.huwei-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
>>> ---
>>>   drivers/infiniband/hw/hns/hns_roce_cq.c     | 15 ++++++-
>>>   drivers/infiniband/hw/hns/hns_roce_device.h | 14 ++++--
>>>   drivers/infiniband/hw/hns/hns_roce_hem.c    |  3 ++
>>>   drivers/infiniband/hw/hns/hns_roce_main.c   | 17 ++++++++
>>>   drivers/infiniband/hw/hns/hns_roce_mr.c     | 66 
>>> ++++++++++++++++++++++-------
>>>   drivers/infiniband/hw/hns/hns_roce_qp.c     |  2 +
>>>   6 files changed, 98 insertions(+), 19 deletions(-)
>> <...>
>>
>>> --- a/drivers/infiniband/hw/hns/hns_roce_device.h
>>> +++ b/drivers/infiniband/hw/hns/hns_roce_device.h
>>> @@ -170,6 +170,11 @@ enum {
>>>       HNS_ROCE_OPCODE_RDMA_WITH_IMM_RECEIVE    = 0x07,
>>>   };
>>>
>>> +enum hns_roce_mtt_type {
>>> +    MTT_TYPE_WQE = 0,
>>> +    MTT_TYPE_CQE,
>>> +};
>> No need to initialize enum to zero.
>> Thanks
> Hi, Leon
>     We will send a followup patch to fix it at your suggestion.
>     Thanks
>
>     Regards
> Wei Hu
>
>
> -- 
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
> .
>


--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 00/20] RDMA/hns: Add hip08 RoCE driver support

[Wei Hu (Xavier)]

Hi, Doug
     We resend the replied email, because it is in the window:
         From 20:30 UTC Sept 14 and 20:30 UTC Sept 15
     Thanks.

    Regards
Wei Hu

On 2017/9/15 9:12, Wei Hu (Xavier) wrote:
>
>
> On 2017/9/14 20:43, Leon Romanovsky wrote:
>> On Thu, Sep 14, 2017 at 03:45:53PM +0800, Wei Hu (Xavier) wrote:
>>>
>>> On 2017/9/14 1:55, Leon Romanovsky wrote:
>>>> On Wed, Aug 30, 2017 at 05:22:58PM +0800, Wei Hu (Xavier) wrote:
>>>>> This patchset adds support for the HiSilicon RoCE engine in the hip08
>>>>> SoC.
>>>>>
>>>>> The driver is refactored so it can support both platform and pci 
>>>>> device.
>>>>> And hip08 RoCE engine is a pci device, hip06 RoCE engine is a 
>>>>> platform
>>>>> device.
>>>>>
>>>>> New hardware layer file hns_roce_hw_v2.c is added for hw v2 support,
>>>>> which also includes pci device probing and initialization.
>>>>>
>>>>> Common functionality is still in hns-roce driver, along with device
>>>>> initialization.
>>>>>
>>>>> This patchset depends on hns3 NIC driver which had been accepted 
>>>>> by David,
>>>>> and is based on for-next branch of the repo:
>>>>> https://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma.git
>>>>>
>>>>> Shaobo Xu (3):
>>>>>     RDMA/hns: Add the interfaces to support multi hop addressing 
>>>>> for the
>>>>>       contexts in hip08
>>>>>     RDMA/hns: Update the interfaces for MTT/CQE multi hop 
>>>>> addressing in
>>>>>       hip08
>>>>>     RDMA/hns: Split CQE from MTT in hip08
>>>>>
>>>>> Wei Hu (Xavier) (17):
>>>>>     RDMA/hns: Split hw v1 driver from hns roce driver
>>>>>     RDMA/hns: Move priv in order to add multiple hns_roce support
>>>>>     RDMA/hns: Initialize the PCI device for hip08 RoCE
>>>>>     RDMA/hns: Modify assignment device variable to support both 
>>>>> PCI device
>>>>>       and platform device
>>>>>     RDMA/hns: Add command queue support for hip08 RoCE driver
>>>>>     RDMA/hns: Add profile support for hip08 driver
>>>>>     RDMA/hns: Add mailbox's implementation for hip08 RoCE driver
>>>>>     RDMA/hns: Configure BT BA and BT attribute for the contexts in 
>>>>> hip08
>>>>>     RDMA/hns: Support multi hop addressing for PBL in hip08
>>>>>     RDMA/hns: Configure mac&gid and user access region for hip08 RoCE
>>>>>       driver
>>>>>     RDMA/hns: Add CQ operations support for hip08 RoCE driver
>>>>>     RDMA/hns: Add QP operations support for hip08 SoC
>>>>>     RDMA/hns: Add support for processing send wr and receive wr
>>>>>     RDMA/hns: Configure the MTPT in hip08
>>>>>     RDMA/hns: Add releasing resource operation in error branch
>>>>>     RDMA/hns: Replace condition statement using hardware version
>>>>>       information
>>>>>     RDMA/hns: Fix inconsistent warning
>>>>>
>>>>>    drivers/infiniband/hw/hns/Kconfig           |   25 +-
>>>>>    drivers/infiniband/hw/hns/Makefile          |    8 +-
>>>>>    drivers/infiniband/hw/hns/hns_roce_ah.c     |    2 +-
>>>>>    drivers/infiniband/hw/hns/hns_roce_alloc.c  |    8 +-
>>>>>    drivers/infiniband/hw/hns/hns_roce_cmd.c    |  107 +-
>>>>>    drivers/infiniband/hw/hns/hns_roce_cmd.h    |   50 +
>>>>>    drivers/infiniband/hw/hns/hns_roce_common.h |   23 +
>>>>>    drivers/infiniband/hw/hns/hns_roce_cq.c     |   72 +-
>>>>>    drivers/infiniband/hw/hns/hns_roce_device.h |   98 +-
>>>>>    drivers/infiniband/hw/hns/hns_roce_hem.c    |  699 +++++-
>>>>>    drivers/infiniband/hw/hns/hns_roce_hem.h    |   32 +-
>>>>>    drivers/infiniband/hw/hns/hns_roce_hw_v1.c  |  401 +++-
>>>>>    drivers/infiniband/hw/hns/hns_roce_hw_v1.h  |    5 +
>>>>>    drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 3128 
>>>>> +++++++++++++++++++++++++++
>>>>>    drivers/infiniband/hw/hns/hns_roce_hw_v2.h  | 1165 ++++++++++
>>>>>    drivers/infiniband/hw/hns/hns_roce_main.c   |  324 +--
>>>>>    drivers/infiniband/hw/hns/hns_roce_mr.c     |  513 ++++-
>>>>>    drivers/infiniband/hw/hns/hns_roce_pd.c     |   20 +-
>>>>>    drivers/infiniband/hw/hns/hns_roce_qp.c     |  179 +-
>>>>>    19 files changed, 6318 insertions(+), 541 deletions(-)
>>>>>    create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>>>>>    create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.h
>>>>>
>>>>> -- 
>>>> As far as I can see, it is ready to be merged.
>>>>
>>>> Thanks,
>>> Hi, Leon
>>>      Thanks for your comments.
>>>
>>>      And about the comment on the No.11 patch of this patchset,
>>>      Which option we should do?  Can you give us some suggestions?
>>>      1. send PATCH V2 to fix it.
>>>      2. After this patchset will be merged, send a new patch to fix it.
>> In your case (20 patches, many LOCs), it will be enough to send a 
>> followup patch
>> to remove it, however you don't need to wait till it is merged and you
>> can send it now with note placed under "---" markings that it is based
>> on this series.
>>
>> Thanks
> Thanks , Leon
> We will send a followup patch to fix it at your suggestion.
>
> Regards
> Wei Hu
>>>      The link about the comment:
>>> https://www.spinics.net/lists/linux-rdma/msg54392.html
>>>
>>>      Regards
>>> Wei Hu
>>>
>>>
>>> -- 
>>> To unsubscribe from this list: send the line "unsubscribe 
>>> linux-rdma" in
>>> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>
>
> _______________________________________________
> linuxarm mailing list
> linuxarm-hv44wF8Li93QT0dZR+AlfA@public.gmane.org
> http://rnd-openeuler.huawei.com/mailman/listinfo/linuxarm
>
> .
>


--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 00/20] RDMA/hns: Add hip08 RoCE driver support

[Wei Hu (Xavier)]

On 2017/8/30 17:22, Wei Hu (Xavier) wrote:
> This patchset adds support for the HiSilicon RoCE engine in the hip08
> SoC.
>
> The driver is refactored so it can support both platform and pci device.
> And hip08 RoCE engine is a pci device, hip06 RoCE engine is a platform
> device.
>
> New hardware layer file hns_roce_hw_v2.c is added for hw v2 support,
> which also includes pci device probing and initialization.
>
> Common functionality is still in hns-roce driver, along with device
> initialization.
>
> This patchset depends on hns3 NIC driver which had been accepted by David,
> and is based on for-next branch of the repo:
> https://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma.git
Hi, Doug
     The dependent hns3 NIC driver had been merged into Linus' v4.14-rc1.
     Thanks.

     Regards
Wei Hu
> Shaobo Xu (3):
>    RDMA/hns: Add the interfaces to support multi hop addressing for the
>      contexts in hip08
>    RDMA/hns: Update the interfaces for MTT/CQE multi hop addressing in
>      hip08
>    RDMA/hns: Split CQE from MTT in hip08
>
> Wei Hu (Xavier) (17):
>    RDMA/hns: Split hw v1 driver from hns roce driver
>    RDMA/hns: Move priv in order to add multiple hns_roce support
>    RDMA/hns: Initialize the PCI device for hip08 RoCE
>    RDMA/hns: Modify assignment device variable to support both PCI device
>      and platform device
>    RDMA/hns: Add command queue support for hip08 RoCE driver
>    RDMA/hns: Add profile support for hip08 driver
>    RDMA/hns: Add mailbox's implementation for hip08 RoCE driver
>    RDMA/hns: Configure BT BA and BT attribute for the contexts in hip08
>    RDMA/hns: Support multi hop addressing for PBL in hip08
>    RDMA/hns: Configure mac&gid and user access region for hip08 RoCE
>      driver
>    RDMA/hns: Add CQ operations support for hip08 RoCE driver
>    RDMA/hns: Add QP operations support for hip08 SoC
>    RDMA/hns: Add support for processing send wr and receive wr
>    RDMA/hns: Configure the MTPT in hip08
>    RDMA/hns: Add releasing resource operation in error branch
>    RDMA/hns: Replace condition statement using hardware version
>      information
>    RDMA/hns: Fix inconsistent warning
>
>   drivers/infiniband/hw/hns/Kconfig           |   25 +-
>   drivers/infiniband/hw/hns/Makefile          |    8 +-
>   drivers/infiniband/hw/hns/hns_roce_ah.c     |    2 +-
>   drivers/infiniband/hw/hns/hns_roce_alloc.c  |    8 +-
>   drivers/infiniband/hw/hns/hns_roce_cmd.c    |  107 +-
>   drivers/infiniband/hw/hns/hns_roce_cmd.h    |   50 +
>   drivers/infiniband/hw/hns/hns_roce_common.h |   23 +
>   drivers/infiniband/hw/hns/hns_roce_cq.c     |   72 +-
>   drivers/infiniband/hw/hns/hns_roce_device.h |   98 +-
>   drivers/infiniband/hw/hns/hns_roce_hem.c    |  699 +++++-
>   drivers/infiniband/hw/hns/hns_roce_hem.h    |   32 +-
>   drivers/infiniband/hw/hns/hns_roce_hw_v1.c  |  401 +++-
>   drivers/infiniband/hw/hns/hns_roce_hw_v1.h  |    5 +
>   drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 3128 +++++++++++++++++++++++++++
>   drivers/infiniband/hw/hns/hns_roce_hw_v2.h  | 1165 ++++++++++
>   drivers/infiniband/hw/hns/hns_roce_main.c   |  324 +--
>   drivers/infiniband/hw/hns/hns_roce_mr.c     |  513 ++++-
>   drivers/infiniband/hw/hns/hns_roce_pd.c     |   20 +-
>   drivers/infiniband/hw/hns/hns_roce_qp.c     |  179 +-
>   19 files changed, 6318 insertions(+), 541 deletions(-)
>   create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>   create mode 100644 drivers/infiniband/hw/hns/hns_roce_hw_v2.h
>


--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 00/20] RDMA/hns: Add hip08 RoCE driver support

[Doug Ledford]

On Mon, 2017-09-25 at 14:18 +0800, Wei Hu (Xavier) wrote:
> 
> On 2017/8/30 17:22, Wei Hu (Xavier) wrote:
> > This patchset adds support for the HiSilicon RoCE engine in the
> > hip08
> > SoC.
> > 
> > The driver is refactored so it can support both platform and pci
> > device.
> > And hip08 RoCE engine is a pci device, hip06 RoCE engine is a
> > platform
> > device.
> > 
> > New hardware layer file hns_roce_hw_v2.c is added for hw v2
> > support,
> > which also includes pci device probing and initialization.
> > 
> > Common functionality is still in hns-roce driver, along with device
> > initialization.
> > 
> > This patchset depends on hns3 NIC driver which had been accepted by
> > David,
> > and is based on for-next branch of the repo:
> > https://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma.git
> 
> Hi, Doug
>      The dependent hns3 NIC driver had been merged into Linus' v4.14-
> rc1.

Thanks, I'm processing this series now.

-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
    GPG KeyID: B826A3330E572FDD
    Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Doug Ledford]

On Wed, 2017-08-30 at 17:23 +0800, Wei Hu (Xavier) wrote:

So, we set the send timeout limit here:

> +	/* Setup Tx write back timeout */
> +	priv->cmq.tx_timeout = HNS_ROCE_CMQ_TX_TIMEOUT;
> +

[ snip ]

> +int hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
> +		      struct hns_roce_cmq_desc *desc, int num)
> +{
> +	struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv
> *)hr_dev->priv;
> +	struct hns_roce_v2_cmq_ring *csq = &priv->cmq.csq;
> +	struct hns_roce_cmq_desc *desc_to_use;
> +	bool complete = false;
> +	u32 timeout = 0;
> +	int handle = 0;
> +	u16 desc_ret;
> +	int ret = 0;
> +	int ntc;
> +
> +	spin_lock_bh(&csq->lock);

and we take a bh lock here...

> +	if (num > hns_roce_cmq_space(csq)) {
> +		spin_unlock_bh(&csq->lock);
> +		return -EBUSY;
> +	}
> +
> +	/*
> +	 * Record the location of desc in the cmq for this time
> +	 * which will be use for hardware to write back
> +	 */
> +	ntc = csq->next_to_use;
> +
> +	while (handle < num) {
> +		desc_to_use = &csq->desc[csq->next_to_use];
> +		*desc_to_use = desc[handle];
> +		dev_dbg(hr_dev->dev, "set cmq desc:\n");
> +		csq->next_to_use++;
> +		if (csq->next_to_use == csq->desc_num)
> +			csq->next_to_use = 0;
> +		handle++;
> +	}
> +
> +	/* Write to hardware */
> +	roce_write(hr_dev, ROCEE_TX_CMQ_TAIL_REG, csq->next_to_use);
> +
> +	/*
> +	 * If the command is sync, wait for the firmware to write
> back,
> +	 * if multi descriptors to be sent, use the first one to
> check
> +	 */
> +	if ((desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
> +		do {
> +			if (hns_roce_cmq_csq_done(hr_dev))
> +				break;
> +			usleep_range(1000, 2000);
> +			timeout++;
> +		} while (timeout < priv->cmq.tx_timeout);
> +	}

then we spin here for a maximum amount of time between 200 and 400ms,
so 1/4 to 1/2 a second.  All the time we are holding the bh lock on
this CPU.  That seems excessive to me.  If we are going to spin that
long, can you find a way to allocate/reserve your resources, send the
command, then drop the bh lock while you spin, and retake it before you
complete once the spinning is done?

> +#define HNS_ROCE_CMQ_TX_TIMEOUT			200

or you could reduce the size of this define...

-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
    GPG KeyID: B826A3330E572FDD
    Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 2891 bytes --]

On Mon, Sep 25, 2017 at 01:06:53PM -0400, Doug Ledford wrote:
> On Wed, 2017-08-30 at 17:23 +0800, Wei Hu (Xavier) wrote:
>
> So, we set the send timeout limit here:
>
> > +	/* Setup Tx write back timeout */
> > +	priv->cmq.tx_timeout = HNS_ROCE_CMQ_TX_TIMEOUT;
> > +
>
> [ snip ]
>
> > +int hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
> > +		      struct hns_roce_cmq_desc *desc, int num)
> > +{
> > +	struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv
> > *)hr_dev->priv;
> > +	struct hns_roce_v2_cmq_ring *csq = &priv->cmq.csq;
> > +	struct hns_roce_cmq_desc *desc_to_use;
> > +	bool complete = false;
> > +	u32 timeout = 0;
> > +	int handle = 0;
> > +	u16 desc_ret;
> > +	int ret = 0;
> > +	int ntc;
> > +
> > +	spin_lock_bh(&csq->lock);
>
> and we take a bh lock here...
>
> > +	if (num > hns_roce_cmq_space(csq)) {
> > +		spin_unlock_bh(&csq->lock);
> > +		return -EBUSY;
> > +	}
> > +
> > +	/*
> > +	 * Record the location of desc in the cmq for this time
> > +	 * which will be use for hardware to write back
> > +	 */
> > +	ntc = csq->next_to_use;
> > +
> > +	while (handle < num) {
> > +		desc_to_use = &csq->desc[csq->next_to_use];
> > +		*desc_to_use = desc[handle];
> > +		dev_dbg(hr_dev->dev, "set cmq desc:\n");
> > +		csq->next_to_use++;
> > +		if (csq->next_to_use == csq->desc_num)
> > +			csq->next_to_use = 0;
> > +		handle++;
> > +	}
> > +
> > +	/* Write to hardware */
> > +	roce_write(hr_dev, ROCEE_TX_CMQ_TAIL_REG, csq->next_to_use);
> > +
> > +	/*
> > +	 * If the command is sync, wait for the firmware to write
> > back,
> > +	 * if multi descriptors to be sent, use the first one to
> > check
> > +	 */
> > +	if ((desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
> > +		do {
> > +			if (hns_roce_cmq_csq_done(hr_dev))
> > +				break;
> > +			usleep_range(1000, 2000);
> > +			timeout++;
> > +		} while (timeout < priv->cmq.tx_timeout);
> > +	}
>
> then we spin here for a maximum amount of time between 200 and 400ms,
> so 1/4 to 1/2 a second.  All the time we are holding the bh lock on
> this CPU.  That seems excessive to me.  If we are going to spin that
> long, can you find a way to allocate/reserve your resources, send the
> command, then drop the bh lock while you spin, and retake it before you
> complete once the spinning is done?

They don't allocate anything in this loop, but checking the pointers are
the same, see hns_roce_cmq_csq_done.

>
> > +#define HNS_ROCE_CMQ_TX_TIMEOUT			200
>
> or you could reduce the size of this define...
>
> --
> Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
>     GPG KeyID: B826A3330E572FDD
>     Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Doug Ledford]

On Mon, 2017-09-25 at 20:18 +0300, Leon Romanovsky wrote:
> On Mon, Sep 25, 2017 at 01:06:53PM -0400, Doug Ledford wrote:
> > On Wed, 2017-08-30 at 17:23 +0800, Wei Hu (Xavier) wrote:
> > 
> > > +	/*
> > > +	 * If the command is sync, wait for the firmware to
> > > write
> > > back,
> > > +	 * if multi descriptors to be sent, use the first one to
> > > check
> > > +	 */
> > > +	if ((desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
> > > +		do {
> > > +			if (hns_roce_cmq_csq_done(hr_dev))
> > > +				break;
> > > +			usleep_range(1000, 2000);
> > > +			timeout++;
> > > +		} while (timeout < priv->cmq.tx_timeout);
> > > +	}
> > 
> > then we spin here for a maximum amount of time between 200 and
> > 400ms,
> > so 1/4 to 1/2 a second.  All the time we are holding the bh lock on
> > this CPU.  That seems excessive to me.  If we are going to spin
> > that
> > long, can you find a way to allocate/reserve your resources, send
> > the
> > command, then drop the bh lock while you spin, and retake it before
> > you
> > complete once the spinning is done?
> 
> They don't allocate anything in this loop, but checking the pointers
> are
> the same, see hns_roce_cmq_csq_done.

I'm not sure I understand your intended implication of your comment.  I
wasn't concerned about them allocating anything, only that if the
hardware is hung, then this loop will hang out for 1/4 to 1/2 a second
and hold up all bottom half processing on this CPU in the meantime. 
That's the sort of things that provides poor overall system behavior.

Now, since they are really only checking to see if the hardware has
gotten around to their particular command, and their command is part of
a ring structure, it's possible to record the original head command,
and our new head command, and then release the spin_lock_bh around the
entire do{ }while construct, and in hns_roce_cmd_csq_done() you could
check that head is not in the range old_head:new_head.  That would
protect you in case something in the bottom half processing queued up
some more commands and from one sleep to the next the head jumped from
something other than the new_head to something past new_head, so that
head == priv->cmq.csq.next_to_use ends up being perpetually false. 
But, that's just from a quick read of the code, I could easily be
missing something here...

> > 
> > > +#define HNS_ROCE_CMQ_TX_TIMEOUT			200
> > 
> > or you could reduce the size of this define...
> > 
> > --
> > Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
> >     GPG KeyID: B826A3330E572FDD
> >     Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57
> > 2FDD
> > 
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-
> > rdma" in
> > the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
    GPG KeyID: B826A3330E572FDD
    Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 00/20] RDMA/hns: Add hip08 RoCE driver support

[Doug Ledford]

On Mon, 2017-09-25 at 11:57 -0400, Doug Ledford wrote:
> On Mon, 2017-09-25 at 14:18 +0800, Wei Hu (Xavier) wrote:
> > 
> > On 2017/8/30 17:22, Wei Hu (Xavier) wrote:
> > > This patchset adds support for the HiSilicon RoCE engine in the
> > > hip08
> > > SoC.
> > > 
> > > The driver is refactored so it can support both platform and pci
> > > device.
> > > And hip08 RoCE engine is a pci device, hip06 RoCE engine is a
> > > platform
> > > device.
> > > 
> > > New hardware layer file hns_roce_hw_v2.c is added for hw v2
> > > support,
> > > which also includes pci device probing and initialization.
> > > 
> > > Common functionality is still in hns-roce driver, along with
> > > device
> > > initialization.
> > > 
> > > This patchset depends on hns3 NIC driver which had been accepted
> > > by
> > > David,
> > > and is based on for-next branch of the repo:
> > > https://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma.git
> > 
> > Hi, Doug
> >      The dependent hns3 NIC driver had been merged into Linus'
> > v4.14-
> > rc1.
> 
> Thanks, I'm processing this series now.

OK, I've gone through the entire series, and pending resolution to my
comments on patch 5, the rest looks OK to me.

-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
    GPG KeyID: B826A3330E572FDD
    Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 3350 bytes --]

On Mon, Sep 25, 2017 at 01:36:55PM -0400, Doug Ledford wrote:
> On Mon, 2017-09-25 at 20:18 +0300, Leon Romanovsky wrote:
> > On Mon, Sep 25, 2017 at 01:06:53PM -0400, Doug Ledford wrote:
> > > On Wed, 2017-08-30 at 17:23 +0800, Wei Hu (Xavier) wrote:
> > >
> > > > +	/*
> > > > +	 * If the command is sync, wait for the firmware to
> > > > write
> > > > back,
> > > > +	 * if multi descriptors to be sent, use the first one to
> > > > check
> > > > +	 */
> > > > +	if ((desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
> > > > +		do {
> > > > +			if (hns_roce_cmq_csq_done(hr_dev))
> > > > +				break;
> > > > +			usleep_range(1000, 2000);
> > > > +			timeout++;
> > > > +		} while (timeout < priv->cmq.tx_timeout);
> > > > +	}
> > >
> > > then we spin here for a maximum amount of time between 200 and
> > > 400ms,
> > > so 1/4 to 1/2 a second.  All the time we are holding the bh lock on
> > > this CPU.  That seems excessive to me.  If we are going to spin
> > > that
> > > long, can you find a way to allocate/reserve your resources, send
> > > the
> > > command, then drop the bh lock while you spin, and retake it before
> > > you
> > > complete once the spinning is done?
> >
> > They don't allocate anything in this loop, but checking the pointers
> > are
> > the same, see hns_roce_cmq_csq_done.
>
> I'm not sure I understand your intended implication of your comment.  I
> wasn't concerned about them allocating anything, only that if the
> hardware is hung, then this loop will hang out for 1/4 to 1/2 a second
> and hold up all bottom half processing on this CPU in the meantime.
> That's the sort of things that provides poor overall system behavior.
>
> Now, since they are really only checking to see if the hardware has
> gotten around to their particular command, and their command is part of
> a ring structure, it's possible to record the original head command,
> and our new head command, and then release the spin_lock_bh around the
> entire do{ }while construct, and in hns_roce_cmd_csq_done() you could
> check that head is not in the range old_head:new_head.  That would
> protect you in case something in the bottom half processing queued up
> some more commands and from one sleep to the next the head jumped from
> something other than the new_head to something past new_head, so that
> head == priv->cmq.csq.next_to_use ends up being perpetually false.
> But, that's just from a quick read of the code, I could easily be
> missing something here...

Thanks,
I got your point, the phrase "can you find a way to allocate/reserve
your resources," confused me back then.

>
> > >
> > > > +#define HNS_ROCE_CMQ_TX_TIMEOUT			200
> > >
> > > or you could reduce the size of this define...
> > >
> > > --
> > > Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
> > >     GPG KeyID: B826A3330E572FDD
> > >     Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57
> > > 2FDD
> > >
> > > --
> > > To unsubscribe from this list: send the line "unsubscribe linux-
> > > rdma" in
> > > the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> > > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> --
> Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
>     GPG KeyID: B826A3330E572FDD
>     Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD
>

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Wei Hu (Xavier)]

On 2017/9/26 1:36, Doug Ledford wrote:
> On Mon, 2017-09-25 at 20:18 +0300, Leon Romanovsky wrote:
>> On Mon, Sep 25, 2017 at 01:06:53PM -0400, Doug Ledford wrote:
>>> On Wed, 2017-08-30 at 17:23 +0800, Wei Hu (Xavier) wrote:
>>>
>>>> +	/*
>>>> +	 * If the command is sync, wait for the firmware to
>>>> write
>>>> back,
>>>> +	 * if multi descriptors to be sent, use the first one to
>>>> check
>>>> +	 */
>>>> +	if ((desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
>>>> +		do {
>>>> +			if (hns_roce_cmq_csq_done(hr_dev))
>>>> +				break;
>>>> +			usleep_range(1000, 2000);
>>>> +			timeout++;
>>>> +		} while (timeout < priv->cmq.tx_timeout);
>>>> +	}
>>> then we spin here for a maximum amount of time between 200 and
>>> 400ms,
>>> so 1/4 to 1/2 a second.  All the time we are holding the bh lock on
>>> this CPU.  That seems excessive to me.  If we are going to spin
>>> that
>>> long, can you find a way to allocate/reserve your resources, send
>>> the
>>> command, then drop the bh lock while you spin, and retake it before
>>> you
>>> complete once the spinning is done?
>> They don't allocate anything in this loop, but checking the pointers
>> are
>> the same, see hns_roce_cmq_csq_done.
> I'm not sure I understand your intended implication of your comment.  I
> wasn't concerned about them allocating anything, only that if the
> hardware is hung, then this loop will hang out for 1/4 to 1/2 a second
> and hold up all bottom half processing on this CPU in the meantime.
> That's the sort of things that provides poor overall system behavior.
>
> Now, since they are really only checking to see if the hardware has
> gotten around to their particular command, and their command is part of
> a ring structure, it's possible to record the original head command,
> and our new head command, and then release the spin_lock_bh around the
> entire do{ }while construct, and in hns_roce_cmd_csq_done() you could
> check that head is not in the range old_head:new_head.  That would
> protect you in case something in the bottom half processing queued up
> some more commands and from one sleep to the next the head jumped from
> something other than the new_head to something past new_head, so that
> head == priv->cmq.csq.next_to_use ends up being perpetually false.
> But, that's just from a quick read of the code, I could easily be
> missing something here...
Hi, Doug
     Driver issues the cmds in cmq, and firmware gets and processes them.
     The firmware process only one cmd at the same time, and it will take
     about serveral to 200 us in one cmd currently, so the driver need
     not to use stream mode to issue cmd.

     Regards
Wei Hu
>>>> +#define HNS_ROCE_CMQ_TX_TIMEOUT			200
>>> or you could reduce the size of this define...
>>>
>>> --
>>> Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
>>>      GPG KeyID: B826A3330E572FDD
>>>      Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57
>>> 2FDD
>>>
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-
>>> rdma" in
>>> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html


--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Wei Hu (Xavier)]

On 2017/9/26 21:13, Wei Hu (Xavier) wrote:
>
>
> On 2017/9/26 1:36, Doug Ledford wrote:
>> On Mon, 2017-09-25 at 20:18 +0300, Leon Romanovsky wrote:
>>> On Mon, Sep 25, 2017 at 01:06:53PM -0400, Doug Ledford wrote:
>>>> On Wed, 2017-08-30 at 17:23 +0800, Wei Hu (Xavier) wrote:
>>>>
>>>>> +    /*
>>>>> +     * If the command is sync, wait for the firmware to
>>>>> write
>>>>> back,
>>>>> +     * if multi descriptors to be sent, use the first one to
>>>>> check
>>>>> +     */
>>>>> +    if ((desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
>>>>> +        do {
>>>>> +            if (hns_roce_cmq_csq_done(hr_dev))
>>>>> +                break;
>>>>> +            usleep_range(1000, 2000);
>>>>> +            timeout++;
>>>>> +        } while (timeout < priv->cmq.tx_timeout);
>>>>> +    }
>>>> then we spin here for a maximum amount of time between 200 and
>>>> 400ms,
>>>> so 1/4 to 1/2 a second.  All the time we are holding the bh lock on
>>>> this CPU.  That seems excessive to me.  If we are going to spin
>>>> that
>>>> long, can you find a way to allocate/reserve your resources, send
>>>> the
>>>> command, then drop the bh lock while you spin, and retake it before
>>>> you
>>>> complete once the spinning is done?
>>> They don't allocate anything in this loop, but checking the pointers
>>> are
>>> the same, see hns_roce_cmq_csq_done.
>> I'm not sure I understand your intended implication of your comment.  I
>> wasn't concerned about them allocating anything, only that if the
>> hardware is hung, then this loop will hang out for 1/4 to 1/2 a second
>> and hold up all bottom half processing on this CPU in the meantime.
>> That's the sort of things that provides poor overall system behavior.
>>
>> Now, since they are really only checking to see if the hardware has
>> gotten around to their particular command, and their command is part of
>> a ring structure, it's possible to record the original head command,
>> and our new head command, and then release the spin_lock_bh around the
>> entire do{ }while construct, and in hns_roce_cmd_csq_done() you could
>> check that head is not in the range old_head:new_head.  That would
>> protect you in case something in the bottom half processing queued up
>> some more commands and from one sleep to the next the head jumped from
>> something other than the new_head to something past new_head, so that
>> head == priv->cmq.csq.next_to_use ends up being perpetually false.
>> But, that's just from a quick read of the code, I could easily be
>> missing something here...
> Hi, Doug
>     Driver issues the cmds in cmq, and firmware gets and processes them.
>     The firmware process only one cmd at the same time, and it will take
>     about serveral to 200 us in one cmd currently, so the driver need
>     not to use stream mode to issue cmd.
>
>     Regards
> Wei Hu
Hi, Doug
     We can replace usleep_range(1000, 2000); with the following statement:
         udelay(1);
     to avoid spinning too long time and using usleep_range function in
     spin_lock_bh locked region.

     Can you give some suggestions?

     Regards
Wei Hu
>>>>> +#define HNS_ROCE_CMQ_TX_TIMEOUT            200
>>>> or you could reduce the size of this define...
>>>>
>>>> -- 
>>>> Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
>>>>      GPG KeyID: B826A3330E572FDD
>>>>      Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57
>>>> 2FDD
>>>>
>>>> -- 
>>>> To unsubscribe from this list: send the line "unsubscribe linux-
>>>> rdma" in
>>>> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>
>
> -- 
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 4457 bytes --]

On Tue, Sep 26, 2017 at 11:24:41PM +0800, Wei Hu (Xavier) wrote:
>
>
> On 2017/9/26 21:13, Wei Hu (Xavier) wrote:
> >
> >
> > On 2017/9/26 1:36, Doug Ledford wrote:
> > > On Mon, 2017-09-25 at 20:18 +0300, Leon Romanovsky wrote:
> > > > On Mon, Sep 25, 2017 at 01:06:53PM -0400, Doug Ledford wrote:
> > > > > On Wed, 2017-08-30 at 17:23 +0800, Wei Hu (Xavier) wrote:
> > > > >
> > > > > > +    /*
> > > > > > +     * If the command is sync, wait for the firmware to
> > > > > > write
> > > > > > back,
> > > > > > +     * if multi descriptors to be sent, use the first one to
> > > > > > check
> > > > > > +     */
> > > > > > +    if ((desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
> > > > > > +        do {
> > > > > > +            if (hns_roce_cmq_csq_done(hr_dev))
> > > > > > +                break;
> > > > > > +            usleep_range(1000, 2000);
> > > > > > +            timeout++;
> > > > > > +        } while (timeout < priv->cmq.tx_timeout);
> > > > > > +    }
> > > > > then we spin here for a maximum amount of time between 200 and
> > > > > 400ms,
> > > > > so 1/4 to 1/2 a second.  All the time we are holding the bh lock on
> > > > > this CPU.  That seems excessive to me.  If we are going to spin
> > > > > that
> > > > > long, can you find a way to allocate/reserve your resources, send
> > > > > the
> > > > > command, then drop the bh lock while you spin, and retake it before
> > > > > you
> > > > > complete once the spinning is done?
> > > > They don't allocate anything in this loop, but checking the pointers
> > > > are
> > > > the same, see hns_roce_cmq_csq_done.
> > > I'm not sure I understand your intended implication of your comment.  I
> > > wasn't concerned about them allocating anything, only that if the
> > > hardware is hung, then this loop will hang out for 1/4 to 1/2 a second
> > > and hold up all bottom half processing on this CPU in the meantime.
> > > That's the sort of things that provides poor overall system behavior.
> > >
> > > Now, since they are really only checking to see if the hardware has
> > > gotten around to their particular command, and their command is part of
> > > a ring structure, it's possible to record the original head command,
> > > and our new head command, and then release the spin_lock_bh around the
> > > entire do{ }while construct, and in hns_roce_cmd_csq_done() you could
> > > check that head is not in the range old_head:new_head.  That would
> > > protect you in case something in the bottom half processing queued up
> > > some more commands and from one sleep to the next the head jumped from
> > > something other than the new_head to something past new_head, so that
> > > head == priv->cmq.csq.next_to_use ends up being perpetually false.
> > > But, that's just from a quick read of the code, I could easily be
> > > missing something here...
> > Hi, Doug
> >     Driver issues the cmds in cmq, and firmware gets and processes them.
> >     The firmware process only one cmd at the same time, and it will take
> >     about serveral to 200 us in one cmd currently, so the driver need
> >     not to use stream mode to issue cmd.
> >
> >     Regards
> > Wei Hu
> Hi, Doug
>     We can replace usleep_range(1000, 2000); with the following statement:
>         udelay(1);
>     to avoid spinning too long time and using usleep_range function in
>     spin_lock_bh locked region.
>
>     Can you give some suggestions?

He already suggested number of options, while the simple one was to reduce
HNS_ROCE_CMQ_TX_TIMEOUT from 200 to be something sane (for example 5).

Thanks
>
>     Regards
> Wei Hu
> > > > > > +#define HNS_ROCE_CMQ_TX_TIMEOUT            200
> > > > > or you could reduce the size of this define...
> > > > >
> > > > > --
> > > > > Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
> > > > >      GPG KeyID: B826A3330E572FDD
> > > > >      Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57
> > > > > 2FDD
> > > > >
> > > > > --
> > > > > To unsubscribe from this list: send the line "unsubscribe linux-
> > > > > rdma" in
> > > > > the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> > > > > More majordomo info at http://vger.kernel.org/majordomo-info.html
> >
> >
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> > the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Wei Hu (Xavier)]

On 2017/9/26 23:51, Leon Romanovsky wrote:
> On Tue, Sep 26, 2017 at 11:24:41PM +0800, Wei Hu (Xavier) wrote:
>>
>> On 2017/9/26 21:13, Wei Hu (Xavier) wrote:
>>>
>>> On 2017/9/26 1:36, Doug Ledford wrote:
>>>> On Mon, 2017-09-25 at 20:18 +0300, Leon Romanovsky wrote:
>>>>> On Mon, Sep 25, 2017 at 01:06:53PM -0400, Doug Ledford wrote:
>>>>>> On Wed, 2017-08-30 at 17:23 +0800, Wei Hu (Xavier) wrote:
>>>>>>
>>>>>>> +    /*
>>>>>>> +     * If the command is sync, wait for the firmware to
>>>>>>> write
>>>>>>> back,
>>>>>>> +     * if multi descriptors to be sent, use the first one to
>>>>>>> check
>>>>>>> +     */
>>>>>>> +    if ((desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
>>>>>>> +        do {
>>>>>>> +            if (hns_roce_cmq_csq_done(hr_dev))
>>>>>>> +                break;
>>>>>>> +            usleep_range(1000, 2000);
>>>>>>> +            timeout++;
>>>>>>> +        } while (timeout < priv->cmq.tx_timeout);
>>>>>>> +    }
>>>>>> then we spin here for a maximum amount of time between 200 and
>>>>>> 400ms,
>>>>>> so 1/4 to 1/2 a second.  All the time we are holding the bh lock on
>>>>>> this CPU.  That seems excessive to me.  If we are going to spin
>>>>>> that
>>>>>> long, can you find a way to allocate/reserve your resources, send
>>>>>> the
>>>>>> command, then drop the bh lock while you spin, and retake it before
>>>>>> you
>>>>>> complete once the spinning is done?
>>>>> They don't allocate anything in this loop, but checking the pointers
>>>>> are
>>>>> the same, see hns_roce_cmq_csq_done.
>>>> I'm not sure I understand your intended implication of your comment.  I
>>>> wasn't concerned about them allocating anything, only that if the
>>>> hardware is hung, then this loop will hang out for 1/4 to 1/2 a second
>>>> and hold up all bottom half processing on this CPU in the meantime.
>>>> That's the sort of things that provides poor overall system behavior.
>>>>
>>>> Now, since they are really only checking to see if the hardware has
>>>> gotten around to their particular command, and their command is part of
>>>> a ring structure, it's possible to record the original head command,
>>>> and our new head command, and then release the spin_lock_bh around the
>>>> entire do{ }while construct, and in hns_roce_cmd_csq_done() you could
>>>> check that head is not in the range old_head:new_head.  That would
>>>> protect you in case something in the bottom half processing queued up
>>>> some more commands and from one sleep to the next the head jumped from
>>>> something other than the new_head to something past new_head, so that
>>>> head == priv->cmq.csq.next_to_use ends up being perpetually false.
>>>> But, that's just from a quick read of the code, I could easily be
>>>> missing something here...
>>> Hi, Doug
>>>      Driver issues the cmds in cmq, and firmware gets and processes them.
>>>      The firmware process only one cmd at the same time, and it will take
>>>      about serveral to 200 us in one cmd currently, so the driver need
>>>      not to use stream mode to issue cmd.
>>>
>>>      Regards
>>> Wei Hu
>> Hi, Doug
>>      We can replace usleep_range(1000, 2000); with the following statement:
>>          udelay(1);
>>      to avoid spinning too long time and using usleep_range function in
>>      spin_lock_bh locked region.
>>
>>      Can you give som=e suggestions?
> He already suggested number of options, while the simple one was to reduce
> HNS_ROCE_CMQ_TX_TIMEOUT from 200 to be something sane (for example 5).
>
> Thanks
Thanks, Leon
     Actually We want to use another option:
     replace usleep_range(1000, 2000); with the following statement:
          udelay(1);
     And if so, we can avoid using usleep_range function in spin_lock_bh 
spin region,
     because it probally cause calltrace by rcu.

     Best regards
Wei Hu
>>      Regards
>> Wei Hu
>>>>>>> +#define HNS_ROCE_CMQ_TX_TIMEOUT            200
>>>>>> or you could reduce the size of this define...
>>>>>>
>>>>>> --
>>>>>> Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
>>>>>>       GPG KeyID: B826A3330E572FDD
>>>>>>       Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57
>>>>>> 2FDD
>>>>>>
>>>>>> --
>>>>>> To unsubscribe from this list: send the line "unsubscribe linux-
>>>>>> rdma" in
>>>>>> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>>>>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>>>
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
>>> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Doug Ledford]

[-- Attachment #1.1: Type: text/plain, Size: 5149 bytes --]

On 9/26/2017 9:13 AM, Wei Hu (Xavier) wrote:
> 
> 
> On 2017/9/26 1:36, Doug Ledford wrote:
>> On Mon, 2017-09-25 at 20:18 +0300, Leon Romanovsky wrote:
>>> On Mon, Sep 25, 2017 at 01:06:53PM -0400, Doug Ledford wrote:
>>>> On Wed, 2017-08-30 at 17:23 +0800, Wei Hu (Xavier) wrote:
>>>>
>>>>> +    /*
>>>>> +     * If the command is sync, wait for the firmware to
>>>>> write
>>>>> back,
>>>>> +     * if multi descriptors to be sent, use the first one to
>>>>> check
>>>>> +     */
>>>>> +    if ((desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
>>>>> +        do {
>>>>> +            if (hns_roce_cmq_csq_done(hr_dev))
>>>>> +                break;
>>>>> +            usleep_range(1000, 2000);
>>>>> +            timeout++;
>>>>> +        } while (timeout < priv->cmq.tx_timeout);
>>>>> +    }
>>>> then we spin here for a maximum amount of time between 200 and
>>>> 400ms,
>>>> so 1/4 to 1/2 a second.  All the time we are holding the bh lock on
>>>> this CPU.  That seems excessive to me.  If we are going to spin
>>>> that
>>>> long, can you find a way to allocate/reserve your resources, send
>>>> the
>>>> command, then drop the bh lock while you spin, and retake it before
>>>> you
>>>> complete once the spinning is done?
>>> They don't allocate anything in this loop, but checking the pointers
>>> are
>>> the same, see hns_roce_cmq_csq_done.
>> I'm not sure I understand your intended implication of your comment.  I
>> wasn't concerned about them allocating anything, only that if the
>> hardware is hung, then this loop will hang out for 1/4 to 1/2 a second
>> and hold up all bottom half processing on this CPU in the meantime.
>> That's the sort of things that provides poor overall system behavior.
>>
>> Now, since they are really only checking to see if the hardware has
>> gotten around to their particular command, and their command is part of
>> a ring structure, it's possible to record the original head command,
>> and our new head command, and then release the spin_lock_bh around the
>> entire do{ }while construct, and in hns_roce_cmd_csq_done() you could
>> check that head is not in the range old_head:new_head.  That would
>> protect you in case something in the bottom half processing queued up
>> some more commands and from one sleep to the next the head jumped from
>> something other than the new_head to something past new_head, so that
>> head == priv->cmq.csq.next_to_use ends up being perpetually false.
>> But, that's just from a quick read of the code, I could easily be
>> missing something here...
> Hi, Doug
>     Driver issues the cmds in cmq, and firmware gets and processes them.
>     The firmware process only one cmd at the same time, and it will take
>     about serveral to 200 us in one cmd currently, so the driver need
>     not to use stream mode to issue cmd.

I'm not sure I understand your response here.

I get that the driver issues cmds in the cmq, and that the firmware gets
them and processes them.

I get that the firmware will only work on one command at a time and only
move to the next one once the current one is complete.

I get that commands take anywhere from a few usec to a couple hundred usec.

I also get that because you are sleeping for somewhere in between 1000
and 2000 usecs, that the driver could easily finish a whole slew of
commands.  It could do 10 slow commands, or 100 or more fast commands.
What this tells me is that the only reason your current implementation
of hns_roce_cmq_csq_done() works at all is because you keep the device
locked out from any other commands being put on the queue.  As far as I
can tell, that's the only way you can guarantee that at some point you
will wake up and the head pointer will be exactly at csq->next_to_use.
Otherwise, if you didn't block them out, then you could sleep with the
head pointer before csq->next_to_use and wake up the next time with it
already well past csq->next_to_use.  Am I right about that?  While you
are waiting on this command queue, any other commands are blocked from
being placed on the command queue?

I don't understand what you mean by "so the driver need not to use
stream mode to issue cmd".

>     Regards
> Wei Hu
>>>>> +#define HNS_ROCE_CMQ_TX_TIMEOUT            200
>>>> or you could reduce the size of this define...
>>>>
>>>> -- 
>>>> Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
>>>>      GPG KeyID: B826A3330E572FDD
>>>>      Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57
>>>> 2FDD
>>>>
>>>> -- 
>>>> To unsubscribe from this list: send the line "unsubscribe linux-
>>>> rdma" in
>>>> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 
> 


-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
    GPG Key ID: B826A3330E572FDD
    Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 884 bytes --]

Re: [PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Wei Hu (Xavier)]

On 2017/9/27 0:13, Wei Hu (Xavier) wrote:
>
>
> On 2017/9/26 23:51, Leon Romanovsky wrote:
>> On Tue, Sep 26, 2017 at 11:24:41PM +0800, Wei Hu (Xavier) wrote:
>>>
>>> On 2017/9/26 21:13, Wei Hu (Xavier) wrote:
>>>>
>>>> On 2017/9/26 1:36, Doug Ledford wrote:
>>>>> On Mon, 2017-09-25 at 20:18 +0300, Leon Romanovsky wrote:
>>>>>> On Mon, Sep 25, 2017 at 01:06:53PM -0400, Doug Ledford wrote:
>>>>>>> On Wed, 2017-08-30 at 17:23 +0800, Wei Hu (Xavier) wrote:
>>>>>>>
>>>>>>>> +    /*
>>>>>>>> +     * If the command is sync, wait for the firmware to
>>>>>>>> write
>>>>>>>> back,
>>>>>>>> +     * if multi descriptors to be sent, use the first one to
>>>>>>>> check
>>>>>>>> +     */
>>>>>>>> +    if ((desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
>>>>>>>> +        do {
>>>>>>>> +            if (hns_roce_cmq_csq_done(hr_dev))
>>>>>>>> +                break;
>>>>>>>> +            usleep_range(1000, 2000);
>>>>>>>> +            timeout++;
>>>>>>>> +        } while (timeout < priv->cmq.tx_timeout);
>>>>>>>> +    }
>>>>>>> then we spin here for a maximum amount of time between 200 and
>>>>>>> 400ms,
>>>>>>> so 1/4 to 1/2 a second.  All the time we are holding the bh lock on
>>>>>>> this CPU.  That seems excessive to me.  If we are going to spin
>>>>>>> that
>>>>>>> long, can you find a way to allocate/reserve your resources, send
>>>>>>> the
>>>>>>> command, then drop the bh lock while you spin, and retake it before
>>>>>>> you
>>>>>>> complete once the spinning is done?
>>>>>> They don't allocate anything in this loop, but checking the pointers
>>>>>> are
>>>>>> the same, see hns_roce_cmq_csq_done.
>>>>> I'm not sure I understand your intended implication of your 
>>>>> comment.  I
>>>>> wasn't concerned about them allocating anything, only that if the
>>>>> hardware is hung, then this loop will hang out for 1/4 to 1/2 a 
>>>>> second
>>>>> and hold up all bottom half processing on this CPU in the meantime.
>>>>> That's the sort of things that provides poor overall system behavior.
>>>>>
>>>>> Now, since they are really only checking to see if the hardware has
>>>>> gotten around to their particular command, and their command is 
>>>>> part of
>>>>> a ring structure, it's possible to record the original head command,
>>>>> and our new head command, and then release the spin_lock_bh around 
>>>>> the
>>>>> entire do{ }while construct, and in hns_roce_cmd_csq_done() you could
>>>>> check that head is not in the range old_head:new_head. That would
>>>>> protect you in case something in the bottom half processing queued up
>>>>> some more commands and from one sleep to the next the head jumped 
>>>>> from
>>>>> something other than the new_head to something past new_head, so that
>>>>> head == priv->cmq.csq.next_to_use ends up being perpetually false.
>>>>> But, that's just from a quick read of the code, I could easily be
>>>>> missing something here...
>>>> Hi, Doug
>>>>      Driver issues the cmds in cmq, and firmware gets and processes 
>>>> them.
>>>>      The firmware process only one cmd at the same time, and it 
>>>> will take
>>>>      about serveral to 200 us in one cmd currently, so the driver need
>>>>      not to use stream mode to issue cmd.
>>>>
>>>>      Regards
>>>> Wei Hu
>>> Hi, Doug
>>>      We can replace usleep_range(1000, 2000); with the following 
>>> statement:
>>>          udelay(1);
>>>      to avoid spinning too long time and using usleep_range function in
>>>      spin_lock_bh locked region.
>>>
>>>      Can you give som=e suggestions?
>> He already suggested number of options, while the simple one was to 
>> reduce
>> HNS_ROCE_CMQ_TX_TIMEOUT from 200 to be something sane (for example 5).
>>
>> Thanks
> Thanks, Leon
>     Actually We want to use another option:
>     replace usleep_range(1000, 2000); with the following statement:
>          udelay(1);
>     And if so, we can avoid using usleep_range function in 
> spin_lock_bh spin region,
>     because it probally cause calltrace by rcu.
>
Can we send a following up patch to fix this problem? or send patch v2?
Thanks
>     Best regards
> Wei Hu
>>>     Regards
>>> Wei Hu
>>>>>>>> +#define HNS_ROCE_CMQ_TX_TIMEOUT            200
>>>>>>> or you could reduce the size of this define...
>>>>>>>
>>>>>>> -- 
>>>>>>> Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
>>>>>>>       GPG KeyID: B826A3330E572FDD
>>>>>>>       Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 
>>>>>>> 0E57
>>>>>>> 2FDD
>>>>>>>
>>>>>>> -- 
>>>>>>> To unsubscribe from this list: send the line "unsubscribe linux-
>>>>>>> rdma" in
>>>>>>> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>>>>>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>>>>
>>>> -- 
>>>> To unsubscribe from this list: send the line "unsubscribe 
>>>> linux-rdma" in
>>>> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Wei Hu (Xavier)]

On 2017/9/27 0:18, Doug Ledford wrote:
> On 9/26/2017 9:13 AM, Wei Hu (Xavier) wrote:
>>
>> On 2017/9/26 1:36, Doug Ledford wrote:
>>> On Mon, 2017-09-25 at 20:18 +0300, Leon Romanovsky wrote:
>>>> On Mon, Sep 25, 2017 at 01:06:53PM -0400, Doug Ledford wrote:
>>>>> On Wed, 2017-08-30 at 17:23 +0800, Wei Hu (Xavier) wrote:
>>>>>
>>>>>> +    /*
>>>>>> +     * If the command is sync, wait for the firmware to
>>>>>> write
>>>>>> back,
>>>>>> +     * if multi descriptors to be sent, use the first one to
>>>>>> check
>>>>>> +     */
>>>>>> +    if ((desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
>>>>>> +        do {
>>>>>> +            if (hns_roce_cmq_csq_done(hr_dev))
>>>>>> +                break;
>>>>>> +            usleep_range(1000, 2000);
>>>>>> +            timeout++;
>>>>>> +        } while (timeout < priv->cmq.tx_timeout);
>>>>>> +    }
>>>>> then we spin here for a maximum amount of time between 200 and
>>>>> 400ms,
>>>>> so 1/4 to 1/2 a second.  All the time we are holding the bh lock on
>>>>> this CPU.  That seems excessive to me.  If we are going to spin
>>>>> that
>>>>> long, can you find a way to allocate/reserve your resources, send
>>>>> the
>>>>> command, then drop the bh lock while you spin, and retake it before
>>>>> you
>>>>> complete once the spinning is done?
>>>> They don't allocate anything in this loop, but checking the pointers
>>>> are
>>>> the same, see hns_roce_cmq_csq_done.
>>> I'm not sure I understand your intended implication of your comment.  I
>>> wasn't concerned about them allocating anything, only that if the
>>> hardware is hung, then this loop will hang out for 1/4 to 1/2 a second
>>> and hold up all bottom half processing on this CPU in the meantime.
>>> That's the sort of things that provides poor overall system behavior.
>>>
>>> Now, since they are really only checking to see if the hardware has
>>> gotten around to their particular command, and their command is part of
>>> a ring structure, it's possible to record the original head command,
>>> and our new head command, and then release the spin_lock_bh around the
>>> entire do{ }while construct, and in hns_roce_cmd_csq_done() you could
>>> check that head is not in the range old_head:new_head.  That would
>>> protect you in case something in the bottom half processing queued up
>>> some more commands and from one sleep to the next the head jumped from
>>> something other than the new_head to something past new_head, so that
>>> head == priv->cmq.csq.next_to_use ends up being perpetually false.
>>> But, that's just from a quick read of the code, I could easily be
>>> missing something here...
>> Hi, Doug
>>      Driver issues the cmds in cmq, and firmware gets and processes them.
>>      The firmware process only one cmd at the same time, and it will take
>>      about serveral to 200 us in one cmd currently, so the driver need
>>      not to use stream mode to issue cmd.
> I'm not sure I understand your response here.
>
> I get that the driver issues cmds in the cmq, and that the firmware gets
> them and processes them.
>
> I get that the firmware will only work on one command at a time and only
> move to the next one once the current one is complete.
>
> I get that commands take anywhere from a few usec to a couple hundred usec.
>
> I also get that because you are sleeping for somewhere in between 1000
> and 2000 usecs, that the driver could easily finish a whole slew of
> commands.  It could do 10 slow commands, or 100 or more fast commands.
> What this tells me is that the only reason your current implementation
> of hns_roce_cmq_csq_done() works at all is because you keep the device
> locked out from any other commands being put on the queue.  As far as I
> can tell, that's the only way you can guarantee that at some point you
> will wake up and the head pointer will be exactly at csq->next_to_use.
> Otherwise, if you didn't block them out, then you could sleep with the
> head pointer before csq->next_to_use and wake up the next time with it
> already well past csq->next_to_use.  Am I right about that?  While you
> are waiting on this command queue, any other commands are blocked from
> being placed on the command queue?
Hi, Doug,
you are right.
And one "hns_x" ib device only has one command queue in hip08,
other commands will be blocked when waiting on the command queue.
>
> I don't understand what you mean by "so the driver need not to use
> stream mode to issue cmd".
Sorry, my expression error.
stream -> pipeline

And if you argee, after this patchset has been accepted we will send a 
following up patch :
     In hns_roce_cmq_send function, replace
         usleep_range(1000, 2000);
     with the following statement:
          udelay(1);
     And if so, we can avoid using usleep_range function in spin_lock_bh 
spin region,
     because it probally cause calltrace.

     Best regards
Wei Hu
>>      Regards
>> Wei Hu
>>>>>> +#define HNS_ROCE_CMQ_TX_TIMEOUT            200
>>>>> or you could reduce the size of this define...
>>>>>
>>>>> -- 
>>>>> Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
>>>>>       GPG KeyID: B826A3330E572FDD
>>>>>       Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57
>>>>> 2FDD
>>>>>
>>>>> -- 
>>>>> To unsubscribe from this list: send the line "unsubscribe linux-
>>>>> rdma" in
>>>>> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
>

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Doug Ledford]

[-- Attachment #1.1: Type: text/plain, Size: 5496 bytes --]

On 9/26/2017 10:46 PM, Wei Hu (Xavier) wrote:
> 
> 
> On 2017/9/27 0:18, Doug Ledford wrote:
>> On 9/26/2017 9:13 AM, Wei Hu (Xavier) wrote:
>>>
>>> On 2017/9/26 1:36, Doug Ledford wrote:
>>>> On Mon, 2017-09-25 at 20:18 +0300, Leon Romanovsky wrote:
>>>>> On Mon, Sep 25, 2017 at 01:06:53PM -0400, Doug Ledford wrote:
>>>>>> On Wed, 2017-08-30 at 17:23 +0800, Wei Hu (Xavier) wrote:
>>>>>>
>>>>>>> +    /*
>>>>>>> +     * If the command is sync, wait for the firmware to
>>>>>>> write
>>>>>>> back,
>>>>>>> +     * if multi descriptors to be sent, use the first one to
>>>>>>> check
>>>>>>> +     */
>>>>>>> +    if ((desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
>>>>>>> +        do {
>>>>>>> +            if (hns_roce_cmq_csq_done(hr_dev))
>>>>>>> +                break;
>>>>>>> +            usleep_range(1000, 2000);
>>>>>>> +            timeout++;
>>>>>>> +        } while (timeout < priv->cmq.tx_timeout);
>>>>>>> +    }
>>>>>> then we spin here for a maximum amount of time between 200 and
>>>>>> 400ms,
>>>>>> so 1/4 to 1/2 a second.  All the time we are holding the bh lock on
>>>>>> this CPU.  That seems excessive to me.  If we are going to spin
>>>>>> that
>>>>>> long, can you find a way to allocate/reserve your resources, send
>>>>>> the
>>>>>> command, then drop the bh lock while you spin, and retake it before
>>>>>> you
>>>>>> complete once the spinning is done?
>>>>> They don't allocate anything in this loop, but checking the pointers
>>>>> are
>>>>> the same, see hns_roce_cmq_csq_done.
>>>> I'm not sure I understand your intended implication of your comment.  I
>>>> wasn't concerned about them allocating anything, only that if the
>>>> hardware is hung, then this loop will hang out for 1/4 to 1/2 a second
>>>> and hold up all bottom half processing on this CPU in the meantime.
>>>> That's the sort of things that provides poor overall system behavior.
>>>>
>>>> Now, since they are really only checking to see if the hardware has
>>>> gotten around to their particular command, and their command is part of
>>>> a ring structure, it's possible to record the original head command,
>>>> and our new head command, and then release the spin_lock_bh around the
>>>> entire do{ }while construct, and in hns_roce_cmd_csq_done() you could
>>>> check that head is not in the range old_head:new_head.  That would
>>>> protect you in case something in the bottom half processing queued up
>>>> some more commands and from one sleep to the next the head jumped from
>>>> something other than the new_head to something past new_head, so that
>>>> head == priv->cmq.csq.next_to_use ends up being perpetually false.
>>>> But, that's just from a quick read of the code, I could easily be
>>>> missing something here...
>>> Hi, Doug
>>>      Driver issues the cmds in cmq, and firmware gets and processes
>>> them.
>>>      The firmware process only one cmd at the same time, and it will
>>> take
>>>      about serveral to 200 us in one cmd currently, so the driver need
>>>      not to use stream mode to issue cmd.
>> I'm not sure I understand your response here.
>>
>> I get that the driver issues cmds in the cmq, and that the firmware gets
>> them and processes them.
>>
>> I get that the firmware will only work on one command at a time and only
>> move to the next one once the current one is complete.
>>
>> I get that commands take anywhere from a few usec to a couple hundred
>> usec.
>>
>> I also get that because you are sleeping for somewhere in between 1000
>> and 2000 usecs, that the driver could easily finish a whole slew of
>> commands.  It could do 10 slow commands, or 100 or more fast commands.
>> What this tells me is that the only reason your current implementation
>> of hns_roce_cmq_csq_done() works at all is because you keep the device
>> locked out from any other commands being put on the queue.  As far as I
>> can tell, that's the only way you can guarantee that at some point you
>> will wake up and the head pointer will be exactly at csq->next_to_use.
>> Otherwise, if you didn't block them out, then you could sleep with the
>> head pointer before csq->next_to_use and wake up the next time with it
>> already well past csq->next_to_use.  Am I right about that?  While you
>> are waiting on this command queue, any other commands are blocked from
>> being placed on the command queue?
> Hi, Doug,
> you are right.
> And one "hns_x" ib device only has one command queue in hip08,
> other commands will be blocked when waiting on the command queue.
>>
>> I don't understand what you mean by "so the driver need not to use
>> stream mode to issue cmd".
> Sorry, my expression error.
> stream -> pipeline
> 
> And if you argee, after this patchset has been accepted we will send a
> following up patch :
>     In hns_roce_cmq_send function, replace
>         usleep_range(1000, 2000);
>     with the following statement:
>          udelay(1);
>     And if so, we can avoid using usleep_range function in spin_lock_bh
> spin region,
>     because it probally cause calltrace.

Ok, I'm fine with that.  I'll pull these in.



-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
    GPG Key ID: B826A3330E572FDD
    Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 884 bytes --]

Re: [PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Doug Ledford]

On Wed, 2017-09-27 at 08:21 -0400, Doug Ledford wrote:
> > And if you argee, after this patchset has been accepted we will
> send a
> > following up patch :
> >     In hns_roce_cmq_send function, replace
> >         usleep_range(1000, 2000);
> >     with the following statement:
> >          udelay(1);
> >     And if so, we can avoid using usleep_range function in
> spin_lock_bh
> > spin region,
> >     because it probally cause calltrace.
> 
> Ok, I'm fine with that.  I'll pull these in.

OK, these are in my tree, thanks.

-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
    GPG KeyID: B826A3330E572FDD
    Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH for-next 05/20] RDMA/hns: Add command queue support for hip08 RoCE driver

[Wei Hu (Xavier)]

On 2017/9/27 20:41, Doug Ledford wrote:
> On Wed, 2017-09-27 at 08:21 -0400, Doug Ledford wrote:
>>> And if you argee, after this patchset has been accepted we will
>> send a
>>> following up patch :
>>>      In hns_roce_cmq_send function, replace
>>>          usleep_range(1000, 2000);
>>>      with the following statement:
>>>           udelay(1);
>>>      And if so, we can avoid using usleep_range function in
>> spin_lock_bh
>>> spin region,
>>>      because it probally cause calltrace.
>> Ok, I'm fine with that.  I'll pull these in.
> OK, these are in my tree, thanks.
Thanks a lot.

     Regards
Wei Hu