[RFC V3 0/2] Intel IFC VF driver for VDPA

[RFC V3 0/2] Intel IFC VF driver for VDPA

[Zhu Lingshan]

Hi all:
  This series intends to introduce Intel IFC VF NIC driver for Vhost
Data Plane Acceleration(VDPA).

Here comes two main parts, one is ifcvf_base layer, which handles
hardware operations. The other is ifcvf_main layer handles VF
initialization, configuration and removal, which depends on
and complys to:
virtio_mdev https://lkml.org/lkml/2019/11/7/650
vhost_mdev https://lkml.org/lkml/2019/11/7/62

This patchset passed netperf tests.

This is RFC V3, plese help review.

Changes from V2:
removed some unnecessary code.
using a struct ifcvf_lm_cfg to address hw->lm_cfg and operations on it.
set CONFIG_S_FAILED  when ifcvf_start_hw() fail.
removed VIRTIO_NET_F_CTRL_VQ and VIRTIO_NET_F_STATUS
replace ifcvf_net_config with virtio_net_config.
minor changes

Changes from V1:
using le32_to_cpu() to convert PCI capabilities.
some set /get  operations will sync with the hardware, eg get_status
and get_generation.
remove feature bit VHOST_F_LOG_ALL, add VIRTIO_F_ORDERED_PLATFORM
add get/set_config functions.
split mdev type group into mdev_type_group_virtio and mdev_type_group_vhost
add ifcvf_mdev_get_mdev_features()
coding stype changes.

Zhu Lingshan (2):
  This commit introduced ifcvf_base layer, which handles hardware    
    operations and configurations.
  This commit introduced IFC operations for vdpa, which complys to    
    virtio_mdev and vhost_mdev interfaces, handles IFC VF    
    initialization, configuration and removal.

 drivers/vhost/ifcvf/ifcvf_base.c | 326 ++++++++++++++++++++++
 drivers/vhost/ifcvf/ifcvf_base.h | 129 +++++++++
 drivers/vhost/ifcvf/ifcvf_main.c | 579 +++++++++++++++++++++++++++++++++++++++
 3 files changed, 1034 insertions(+)
 create mode 100644 drivers/vhost/ifcvf/ifcvf_base.c
 create mode 100644 drivers/vhost/ifcvf/ifcvf_base.h
 create mode 100644 drivers/vhost/ifcvf/ifcvf_main.c

-- 
1.8.3.1

[RFC V3 1/2] This commit introduced ifcvf_base layer, which handles hardware operations and configurations.

[Zhu Lingshan]

Signed-off-by: Zhu Lingshan <lingshan.zhu@intel.com>
---
 drivers/vhost/ifcvf/ifcvf_base.c | 326 +++++++++++++++++++++++++++++++++++++++
 drivers/vhost/ifcvf/ifcvf_base.h | 129 ++++++++++++++++
 2 files changed, 455 insertions(+)
 create mode 100644 drivers/vhost/ifcvf/ifcvf_base.c
 create mode 100644 drivers/vhost/ifcvf/ifcvf_base.h

diff --git a/drivers/vhost/ifcvf/ifcvf_base.c b/drivers/vhost/ifcvf/ifcvf_base.c
new file mode 100644
index 0000000..6bf2ddd
--- /dev/null
+++ b/drivers/vhost/ifcvf/ifcvf_base.c
@@ -0,0 +1,326 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2019 Intel Corporation.
+ */
+
+#include "ifcvf_base.h"
+
+static void *get_cap_addr(struct ifcvf_hw *hw, struct virtio_pci_cap *cap)
+{
+	struct ifcvf_adapter *ifcvf;
+	u32 length, offset;
+	u8 bar;
+
+	length = le32_to_cpu(cap->length);
+	offset = le32_to_cpu(cap->offset);
+	bar = le32_to_cpu(cap->bar);
+
+	ifcvf = container_of(hw, struct ifcvf_adapter, vf);
+
+	if (bar >= IFCVF_PCI_MAX_RESOURCE) {
+		IFC_DBG(ifcvf->dev,
+			"Invalid bar number %u to get capabilities\n", bar);
+		return NULL;
+	}
+
+	if (offset + length > hw->mem_resource[cap->bar].len) {
+		IFC_DBG(ifcvf->dev,
+			"offset(%u) + len(%u) overflows bar%u to get capabilities\n",
+			offset, length, bar);
+		return NULL;
+	}
+
+	return hw->mem_resource[bar].addr + offset;
+}
+
+int ifcvf_read_config_range(struct pci_dev *dev,
+			uint32_t *val, int size, int where)
+{
+	int ret, i;
+
+	for (i = 0; i < size; i += 4) {
+		ret = pci_read_config_dword(dev, where + i, val + i / 4);
+		if (ret < 0)
+			return ret;
+	}
+
+	return 0;
+}
+
+int ifcvf_init_hw(struct ifcvf_hw *hw, struct pci_dev *dev)
+{
+	struct virtio_pci_cap cap;
+	u16 notify_off;
+	int ret;
+	u8 pos;
+	u32 i;
+
+	ret = pci_read_config_byte(dev, PCI_CAPABILITY_LIST, &pos);
+
+	if (ret < 0) {
+		IFC_ERR(&dev->dev, "Failed to read PCI capability list\n");
+		return -EIO;
+	}
+
+	while (pos) {
+		ret = ifcvf_read_config_range(dev, (u32 *)&cap,
+					      sizeof(cap), pos);
+
+		if (ret < 0) {
+			IFC_ERR(&dev->dev, "Failed to get PCI capability at %x\n",
+				pos);
+			break;
+		}
+
+		if (cap.cap_vndr != PCI_CAP_ID_VNDR)
+			goto next;
+
+		IFC_DBG(&dev->dev, "read PCI config: config type: %u, PCI bar: %u, PCI bar offset: %u, PCI config len: %u\n",
+			cap.cfg_type, cap.bar, cap.offset, cap.length);
+
+		switch (cap.cfg_type) {
+		case VIRTIO_PCI_CAP_COMMON_CFG:
+			hw->common_cfg = get_cap_addr(hw, &cap);
+			IFC_DBG(&dev->dev, "hw->common_cfg = %p\n",
+				hw->common_cfg);
+			break;
+		case VIRTIO_PCI_CAP_NOTIFY_CFG:
+			pci_read_config_dword(dev, pos + sizeof(cap),
+					      &hw->notify_off_multiplier);
+			hw->notify_bar = cap.bar;
+			hw->notify_base = get_cap_addr(hw, &cap);
+			IFC_DBG(&dev->dev, "hw->notify_base = %p\n",
+				hw->notify_base);
+			break;
+		case VIRTIO_PCI_CAP_ISR_CFG:
+			hw->isr = get_cap_addr(hw, &cap);
+			IFC_DBG(&dev->dev, "hw->isr = %p\n", hw->isr);
+			break;
+		case VIRTIO_PCI_CAP_DEVICE_CFG:
+			hw->net_cfg = get_cap_addr(hw, &cap);
+			IFC_DBG(&dev->dev, "hw->net_cfg = %p\n", hw->net_cfg);
+			break;
+		}
+next:
+		pos = cap.cap_next;
+	}
+
+	if (hw->common_cfg == NULL || hw->notify_base == NULL ||
+	    hw->isr == NULL || hw->net_cfg == NULL) {
+		IFC_DBG(&dev->dev, "Incomplete PCI capabilities\n");
+		return -1;
+	}
+
+	for (i = 0; i < IFCVF_MAX_QUEUE_PAIRS * 2; i++) {
+		iowrite16(i, &hw->common_cfg->queue_select);
+		notify_off = ioread16(&hw->common_cfg->queue_notify_off);
+		hw->vring->notify_addr[i] = (void *)((u8 *)hw->notify_base +
+				     notify_off * hw->notify_off_multiplier);
+	}
+
+	hw->lm_cfg = hw->mem_resource[IFCVF_LM_BAR].addr;
+
+	IFC_DBG(&dev->dev, "PCI capability mapping: common cfg: %p, notify base: %p\n, isr cfg: %p, device cfg: %p, multiplier: %u\n",
+		hw->common_cfg, hw->notify_base, hw->isr,
+		hw->net_cfg, hw->notify_off_multiplier);
+
+	return 0;
+}
+
+u8 ifcvf_get_status(struct ifcvf_hw *hw)
+{
+	return ioread8(&hw->common_cfg->device_status);
+}
+
+void ifcvf_set_status(struct ifcvf_hw *hw, u8 status)
+{
+	iowrite8(status, &hw->common_cfg->device_status);
+}
+
+void ifcvf_reset(struct ifcvf_hw *hw)
+{
+	ifcvf_set_status(hw, 0);
+	ifcvf_get_status(hw);
+}
+
+static void ifcvf_add_status(struct ifcvf_hw *hw, u8 status)
+{
+	if (status != 0)
+		status |= ifcvf_get_status(hw);
+
+	ifcvf_set_status(hw, status);
+	ifcvf_get_status(hw);
+}
+
+u64 ifcvf_get_features(struct ifcvf_hw *hw)
+{
+	struct virtio_pci_common_cfg *cfg = hw->common_cfg;
+	u32 features_lo, features_hi;
+
+	iowrite32(0, &cfg->device_feature_select);
+	features_lo = ioread32(&cfg->device_feature);
+
+	iowrite32(1, &cfg->device_feature_select);
+	features_hi = ioread32(&cfg->device_feature);
+
+	return ((u64)features_hi << 32) | features_lo;
+}
+
+void ifcvf_read_net_config(struct ifcvf_hw *hw, u64 offset,
+		       void *dst, int length)
+{
+	u8 old_gen, new_gen, *p;
+	int i;
+
+	WARN_ON(offset + length > sizeof (struct virtio_net_config));
+
+	do {
+		old_gen = ioread8(&hw->common_cfg->config_generation);
+		p = dst;
+
+		for (i = 0; i < length; i++)
+			*p++ = ioread8((u8 *)hw->net_cfg + offset + i);
+
+		new_gen = ioread8(&hw->common_cfg->config_generation);
+	} while (old_gen != new_gen);
+}
+
+void ifcvf_write_net_config(struct ifcvf_hw *hw, u64 offset,
+			    const void *src, int length)
+{
+	const u8 *p;
+	int i;
+
+	p = src;
+	WARN_ON(offset + length > sizeof(struct virtio_net_config));
+
+	for (i = 0; i < length; i++)
+		iowrite8(*p++, (u8 *)hw->net_cfg + offset + i);
+}
+
+static void ifcvf_set_features(struct ifcvf_hw *hw, u64 features)
+{
+	struct virtio_pci_common_cfg *cfg = hw->common_cfg;
+
+	iowrite32(0, &cfg->guest_feature_select);
+	iowrite32((u32)features, &cfg->guest_feature);
+
+	iowrite32(1, &cfg->guest_feature_select);
+	iowrite32(features >> 32, &cfg->guest_feature);
+}
+
+static int ifcvf_config_features(struct ifcvf_hw *hw)
+{
+	struct ifcvf_adapter *ifcvf;
+
+	ifcvf = container_of(hw, struct ifcvf_adapter, vf);
+	ifcvf_set_features(hw, hw->req_features);
+	ifcvf_add_status(hw, VIRTIO_CONFIG_S_FEATURES_OK);
+
+	if (!(ifcvf_get_status(hw) & VIRTIO_CONFIG_S_FEATURES_OK)) {
+		IFC_ERR(ifcvf->dev, "Failed to set FEATURES_OK status\n");
+		return -EIO;
+	}
+
+	return 0;
+}
+
+void io_write64_twopart(u64 val, u32 *lo, u32 *hi)
+{
+	iowrite32(val & ((1ULL << 32) - 1), lo);
+	iowrite32(val >> 32, hi);
+}
+
+static int ifcvf_hw_enable(struct ifcvf_hw *hw)
+{
+	struct ifcvf_lm_cfg __iomem *ifcvf_lm;
+	struct virtio_pci_common_cfg *cfg;
+	struct ifcvf_adapter *ifcvf;
+	u32 __iomem* idx_addr;
+	u32 i, val;
+
+	ifcvf_lm = (struct ifcvf_lm_cfg*)hw->lm_cfg;
+	ifcvf = container_of(hw, struct ifcvf_adapter, vf);
+	cfg = hw->common_cfg;
+	iowrite16(IFCVF_MSI_CONFIG_OFF, &cfg->msix_config);
+
+	if (ioread16(&cfg->msix_config) == VIRTIO_MSI_NO_VECTOR) {
+		IFC_ERR(ifcvf->dev, "No msix vector for device config\n");
+		return -1;
+	}
+
+	for (i = 0; i < hw->nr_vring; i++) {
+		if (!hw->vring[i].ready)
+			break;
+
+		iowrite16(i, &cfg->queue_select);
+		io_write64_twopart(hw->vring[i].desc, &cfg->queue_desc_lo,
+				&cfg->queue_desc_hi);
+		io_write64_twopart(hw->vring[i].avail, &cfg->queue_avail_lo,
+				&cfg->queue_avail_hi);
+		io_write64_twopart(hw->vring[i].used, &cfg->queue_used_lo,
+				&cfg->queue_used_hi);
+		iowrite16(hw->vring[i].size, &cfg->queue_size);
+		idx_addr = &(ifcvf_lm->vring_lm_cfg[i/IFCVF_MAX_QUEUES].idx_addr[i%IFCVF_MAX_QUEUES]);
+		val = (u32)hw->vring[i].last_avail_idx |
+			((u32)hw->vring[i].last_used_idx << 16);
+		iowrite32(val, idx_addr);
+		iowrite16(i + IFCVF_MSI_QUEUE_OFF, &cfg->queue_msix_vector);
+
+		if (ioread16(&cfg->queue_msix_vector) ==
+		    VIRTIO_MSI_NO_VECTOR) {
+			IFC_ERR(ifcvf->dev,
+				"No msix vector for queue %u\n", i);
+			return -1;
+		}
+
+		iowrite16(1, &cfg->queue_enable);
+	}
+
+	return 0;
+}
+
+static void ifcvf_hw_disable(struct ifcvf_hw *hw)
+{
+	struct virtio_pci_common_cfg *cfg;
+	u32 i;
+
+	cfg = hw->common_cfg;
+	iowrite16(VIRTIO_MSI_NO_VECTOR, &cfg->msix_config);
+
+	for (i = 0; i < hw->nr_vring; i++) {
+		iowrite16(i, &cfg->queue_select);
+		iowrite16(0, &cfg->queue_enable);
+		iowrite16(VIRTIO_MSI_NO_VECTOR, &cfg->queue_msix_vector);
+	}
+
+	ioread16(&cfg->queue_msix_vector);
+}
+
+int ifcvf_start_hw(struct ifcvf_hw *hw)
+{
+	ifcvf_reset(hw);
+	ifcvf_add_status(hw, VIRTIO_CONFIG_S_ACKNOWLEDGE);
+	ifcvf_add_status(hw, VIRTIO_CONFIG_S_DRIVER);
+
+	if (ifcvf_config_features(hw) < 0)
+		return -1;
+
+	if (ifcvf_hw_enable(hw) < 0)
+		return -1;
+
+	ifcvf_add_status(hw, VIRTIO_CONFIG_S_DRIVER_OK);
+
+	return 0;
+}
+
+void ifcvf_stop_hw(struct ifcvf_hw *hw)
+{
+	ifcvf_hw_disable(hw);
+	ifcvf_reset(hw);
+}
+
+void ifcvf_notify_queue(struct ifcvf_hw *hw, u16 qid)
+{
+	iowrite16(qid, hw->vring->notify_addr[qid]);
+}
diff --git a/drivers/vhost/ifcvf/ifcvf_base.h b/drivers/vhost/ifcvf/ifcvf_base.h
new file mode 100644
index 0000000..587c114
--- /dev/null
+++ b/drivers/vhost/ifcvf/ifcvf_base.h
@@ -0,0 +1,129 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2019 Intel Corporation.
+ */
+
+#ifndef _IFCVF_H_
+#define _IFCVF_H_
+
+#include <linux/mdev_virtio_ops.h>
+#include <linux/mdev.h>
+#include <linux/pci.h>
+#include <linux/pci_regs.h>
+#include <uapi/linux/virtio_net.h>
+#include <uapi/linux/virtio_config.h>
+#include <uapi/linux/virtio_pci.h>
+
+#define IFCVF_VENDOR_ID         0x1AF4
+#define IFCVF_DEVICE_ID         0x1041
+#define IFCVF_SUBSYS_VENDOR_ID  0x8086
+#define IFCVF_SUBSYS_DEVICE_ID  0x001A
+
+#define IFCVF_MDEV_LIMIT	1
+
+#define IFC_SUPPORTED_FEATURES \
+		((1ULL << VIRTIO_NET_F_MAC)			| \
+		 (1ULL << VIRTIO_F_ANY_LAYOUT)			| \
+		 (1ULL << VIRTIO_F_VERSION_1)			| \
+		 (1ULL << VIRTIO_F_ORDER_PLATFORM)		| \
+		 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE)		| \
+		 (1ULL << VIRTIO_F_IOMMU_PLATFORM)		| \
+		 (1ULL << VIRTIO_NET_F_MRG_RXBUF)		| \
+
+/* Not support MQ, only one queue pair for now. */
+#define IFCVF_MAX_QUEUE_PAIRS		1
+#define IFCVF_MAX_QUEUES		2
+
+#define IFCVF_QUEUE_ALIGNMENT		PAGE_SIZE
+#define IFCVF_QUEUE_MAX		32 * 1024
+#define IFCVF_MSI_CONFIG_OFF	0
+#define IFCVF_MSI_QUEUE_OFF	1
+#define IFCVF_PCI_MAX_RESOURCE	6
+
+#define IFCVF_LM_CFG_SIZE		0x40
+#define IFCVF_LM_RING_STATE_OFFSET	0x20
+#define IFCVF_LM_BAR	4
+
+#define IFCVF_32_BIT_MASK		0xffffffff
+
+#define IFC_ERR(dev, fmt, ...)	dev_err(dev, fmt, ##__VA_ARGS__)
+#define IFC_DBG(dev, fmt, ...)	dev_dbg(dev, fmt, ##__VA_ARGS__)
+#define IFC_INFO(dev, fmt, ...)	dev_info(dev, fmt, ##__VA_ARGS__)
+
+#define IFC_PRIVATE_TO_VF(adapter) \
+	(&((struct ifcvf_adapter *)adapter)->vf)
+
+#define IFCVF_MAX_INTR (IFCVF_MAX_QUEUE_PAIRS * 2 + 1)
+
+struct ifcvf_pci_mem_resource {
+	/* Physical address, 0 if not resource. */
+	u64      phys_addr;
+	/* Length of the resource. */
+	u64      len;
+	/* Virtual address, NULL when not mapped. */
+	u8       *addr;
+};
+
+struct vring_info {
+	u64 desc;
+	u64 avail;
+	u64 used;
+	u16 size;
+	u16 last_avail_idx;
+	u16 last_used_idx;
+	bool ready;
+	char msix_name[256];
+	u16 __iomem *notify_addr[IFCVF_MAX_QUEUE_PAIRS * 2];
+	struct virtio_mdev_callback cb;
+};
+
+struct ifcvf_hw {
+	u8	*isr;
+	/* Bar to nofity the VF (queue) */
+	u8	notify_bar;
+	/* live migration */
+	u8	__iomem	*lm_cfg;
+	u8	nr_vring;
+	/* Notificaiton bar address*/
+	u16	*notify_base;
+	u32	notify_off_multiplier;
+	u64	req_features;
+	struct	virtio_pci_common_cfg __iomem	*common_cfg;
+	struct	virtio_net_config __iomem	*net_cfg;
+	struct	vring_info vring[IFCVF_MAX_QUEUE_PAIRS * 2];
+	struct	ifcvf_pci_mem_resource mem_resource[IFCVF_PCI_MAX_RESOURCE];
+};
+
+struct ifcvf_adapter {
+	struct	device *dev;
+	struct	mutex mdev_lock;
+	int	mdev_count;
+	int	vectors;
+	struct	ifcvf_hw vf;
+};
+
+struct ifcvf_vring_lm_cfg {
+	u32 idx_addr[2];
+	u8 reserved[IFCVF_LM_CFG_SIZE - 8];
+};
+
+struct ifcvf_lm_cfg {
+	u8 reserved[IFCVF_LM_RING_STATE_OFFSET];
+	struct ifcvf_vring_lm_cfg vring_lm_cfg[IFCVF_MAX_QUEUE_PAIRS];
+};
+
+int ifcvf_init_hw(struct ifcvf_hw *hw, struct pci_dev *dev);
+int ifcvf_start_hw(struct ifcvf_hw *hw);
+void ifcvf_stop_hw(struct ifcvf_hw *hw);
+void ifcvf_notify_queue(struct ifcvf_hw *hw, u16 qid);
+void ifcvf_read_net_config(struct ifcvf_hw *hw, u64 offset,
+			   void *dst, int length);
+void ifcvf_write_net_config(struct ifcvf_hw *hw, u64 offset,
+			    const void *src, int length);
+u8 ifcvf_get_status(struct ifcvf_hw *hw);
+void ifcvf_set_status(struct ifcvf_hw *hw, u8 status);
+void io_write64_twopart(u64 val, u32 *lo, u32 *hi);
+void ifcvf_reset(struct ifcvf_hw *hw);
+u64 ifcvf_get_features(struct ifcvf_hw *hw);
+
+#endif /* _IFCVF_H_ */
-- 
1.8.3.1

[RFC V3 2/2] This commit introduced IFC operations for vdpa, which complys to virtio_mdev and vhost_mdev interfaces, handles IFC VF initialization, configuration and removal.

[Zhu Lingshan]

Signed-off-by: Zhu Lingshan <lingshan.zhu@intel.com>
---
 drivers/vhost/ifcvf/ifcvf_main.c | 579 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 579 insertions(+)
 create mode 100644 drivers/vhost/ifcvf/ifcvf_main.c

diff --git a/drivers/vhost/ifcvf/ifcvf_main.c b/drivers/vhost/ifcvf/ifcvf_main.c
new file mode 100644
index 0000000..0b33410
--- /dev/null
+++ b/drivers/vhost/ifcvf/ifcvf_main.c
@@ -0,0 +1,579 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2019 Intel Corporation.
+ */
+
+#include <linux/interrupt.h>
+#include <linux/module.h>
+#include <linux/mdev.h>
+#include <linux/pci.h>
+#include <linux/sysfs.h>
+#include "ifcvf_base.h"
+
+#define VERSION_STRING	"0.1"
+#define DRIVER_AUTHOR	"Intel Corporation"
+#define IFCVF_DRIVER_NAME	"ifcvf"
+
+static struct ifcvf_hw *mdev_to_vf(struct mdev_device *mdev)
+{
+	struct ifcvf_adapter *adapter = mdev_get_drvdata(mdev);
+	struct ifcvf_hw *vf = IFC_PRIVATE_TO_VF(adapter);
+
+	return vf;
+}
+
+static irqreturn_t ifcvf_intr_handler(int irq, void *arg)
+{
+	struct vring_info *vring = arg;
+
+	if (vring->cb.callback)
+		return vring->cb.callback(vring->cb.private);
+
+	return IRQ_HANDLED;
+}
+
+static u64 ifcvf_mdev_get_features(struct mdev_device *mdev)
+{
+	struct ifcvf_hw *vf = mdev_to_vf(mdev);
+
+	return ifcvf_get_features(vf);
+}
+
+static int ifcvf_mdev_set_features(struct mdev_device *mdev, u64 features)
+{
+	struct ifcvf_hw *vf = mdev_to_vf(mdev);
+
+	vf->req_features = features;
+
+	return 0;
+}
+
+static u64 ifcvf_mdev_get_vq_state(struct mdev_device *mdev, u16 qid)
+{
+	struct ifcvf_hw *vf = mdev_to_vf(mdev);
+	u16 last_avail_idx;
+	u16 __iomem *idx_addr;
+
+	idx_addr = (u16 __iomem*)(vf->lm_cfg + IFCVF_LM_RING_STATE_OFFSET +
+			(qid / 2) * IFCVF_LM_CFG_SIZE + (qid % 2) * 4);
+
+	last_avail_idx = ioread16(idx_addr);
+
+	return last_avail_idx;
+}
+
+static int ifcvf_mdev_set_vq_state(struct mdev_device *mdev, u16 qid, u64 num)
+{
+	struct ifcvf_hw *vf = mdev_to_vf(mdev);
+
+	vf->vring[qid].last_avail_idx = num;
+
+	return 0;
+}
+
+static int ifcvf_mdev_set_vq_address(struct mdev_device *mdev, u16 idx,
+				     u64 desc_area, u64 driver_area,
+				     u64 device_area)
+{
+	struct ifcvf_hw *vf = mdev_to_vf(mdev);
+
+	vf->vring[idx].desc = desc_area;
+	vf->vring[idx].avail = driver_area;
+	vf->vring[idx].used = device_area;
+
+	return 0;
+}
+
+static void ifcvf_mdev_set_vq_num(struct mdev_device *mdev, u16 qid, u32 num)
+{
+	struct ifcvf_hw *vf = mdev_to_vf(mdev);
+
+	vf->vring[qid].size = num;
+}
+
+static void ifcvf_mdev_set_vq_ready(struct mdev_device *mdev,
+				    u16 qid, bool ready)
+{
+	struct ifcvf_hw *vf = mdev_to_vf(mdev);
+
+	vf->vring[qid].ready = ready;
+}
+
+static bool ifcvf_mdev_get_vq_ready(struct mdev_device *mdev, u16 qid)
+{
+	struct ifcvf_hw *vf = mdev_to_vf(mdev);
+
+	return vf->vring[qid].ready;
+}
+
+static void ifcvf_mdev_set_vq_cb(struct mdev_device *mdev, u16 idx,
+				 struct virtio_mdev_callback *cb)
+{
+	struct ifcvf_hw *vf = mdev_to_vf(mdev);
+
+	vf->vring[idx].cb = *cb;
+}
+
+static void ifcvf_mdev_kick_vq(struct mdev_device *mdev, u16 idx)
+{
+	struct ifcvf_hw *vf = mdev_to_vf(mdev);
+
+	ifcvf_notify_queue(vf, idx);
+}
+
+static u8 ifcvf_mdev_get_status(struct mdev_device *mdev)
+{
+	struct ifcvf_hw *vf = mdev_to_vf(mdev);
+
+	return ifcvf_get_status(vf);
+}
+
+static u32 ifcvf_mdev_get_generation(struct mdev_device *mdev)
+{
+	struct ifcvf_hw *vf = mdev_to_vf(mdev);
+
+	return ioread8(&vf->common_cfg->config_generation);
+}
+
+static u32 ifcvf_mdev_get_device_id(struct mdev_device *mdev)
+{
+	return VIRTIO_ID_NET;
+}
+
+static u32 ifcvf_mdev_get_vendor_id(struct mdev_device *mdev)
+{
+	return IFCVF_VENDOR_ID;
+}
+
+static u16 ifcvf_mdev_get_vq_align(struct mdev_device *mdev)
+{
+	return IFCVF_QUEUE_ALIGNMENT;
+}
+
+static int ifcvf_start_datapath(void *private)
+{
+	struct ifcvf_hw *vf = IFC_PRIVATE_TO_VF(private);
+	struct ifcvf_adapter *ifcvf;
+	int ret = 0;
+	u8 status;
+
+	ifcvf = container_of(vf, struct ifcvf_adapter, vf);
+	vf->nr_vring = IFCVF_MAX_QUEUE_PAIRS * 2;
+	ret = ifcvf_start_hw(vf);
+
+	if (ret) {
+		status = ifcvf_get_status(vf);
+		status |= VIRTIO_CONFIG_S_FAILED;
+		ifcvf_set_status(vf, status);
+	}
+
+	return ret;
+}
+
+static int ifcvf_stop_datapath(void *private)
+{
+	struct ifcvf_hw *vf = IFC_PRIVATE_TO_VF(private);
+	int i;
+
+	for (i = 0; i < IFCVF_MAX_QUEUES; i++)
+		vf->vring[i].cb.callback = NULL;
+
+	ifcvf_stop_hw(vf);
+
+	return 0;
+}
+
+static void ifcvf_reset_vring(struct ifcvf_adapter *adapter)
+{
+	struct ifcvf_hw *vf = IFC_PRIVATE_TO_VF(adapter);
+	int i;
+
+	for (i = 0; i < IFCVF_MAX_QUEUE_PAIRS * 2; i++) {
+		vf->vring[i].last_used_idx = 0;
+		vf->vring[i].last_avail_idx = 0;
+		vf->vring[i].desc = 0;
+		vf->vring[i].avail = 0;
+		vf->vring[i].used = 0;
+		vf->vring[i].ready = 0;
+		vf->vring->cb.callback = NULL;
+		vf->vring->cb.private = NULL;
+
+	}
+
+	ifcvf_reset(vf);
+}
+
+static void ifcvf_mdev_set_status(struct mdev_device *mdev, u8 status)
+{
+	struct ifcvf_adapter *adapter = mdev_get_drvdata(mdev);
+	struct ifcvf_hw *vf = IFC_PRIVATE_TO_VF(adapter);
+	int ret = 0;
+
+	if (status == 0) {
+		ifcvf_stop_datapath(adapter);
+		ifcvf_reset_vring(adapter);
+		return;
+	}
+
+	if (status & VIRTIO_CONFIG_S_DRIVER_OK) {
+		ret = ifcvf_start_datapath(adapter);
+
+		if (ret)
+			IFC_ERR(adapter->dev, "Failed to set mdev status %u\n",
+				status);
+	}
+
+	ifcvf_set_status(vf, status);
+}
+
+static u16 ifcvf_mdev_get_vq_num_max(struct mdev_device *mdev)
+{
+
+	return (u16)IFCVF_QUEUE_MAX;
+}
+static void ifcvf_mdev_get_config(struct mdev_device *mdev, unsigned int offset,
+			     void *buf, unsigned int len)
+{
+	struct ifcvf_hw *vf = mdev_to_vf(mdev);
+
+	WARN_ON(offset + len > sizeof(struct virtio_net_config));
+	ifcvf_read_net_config(vf, offset, buf, len);
+}
+
+static void ifcvf_mdev_set_config(struct mdev_device *mdev, unsigned int offset,
+			     const void *buf, unsigned int len)
+{
+	struct ifcvf_hw *vf = mdev_to_vf(mdev);
+
+	WARN_ON(offset + len > sizeof(struct virtio_net_config));
+	ifcvf_write_net_config(vf, offset, buf, len);
+}
+
+static struct mdev_virtio_device_ops ifc_mdev_ops = {
+	.get_features  = ifcvf_mdev_get_features,
+	.set_features  = ifcvf_mdev_set_features,
+	.get_status    = ifcvf_mdev_get_status,
+	.set_status    = ifcvf_mdev_set_status,
+	.get_vq_num_max = ifcvf_mdev_get_vq_num_max,
+	.get_vq_state   = ifcvf_mdev_get_vq_state,
+	.set_vq_state   = ifcvf_mdev_set_vq_state,
+	.set_vq_cb      = ifcvf_mdev_set_vq_cb,
+	.set_vq_ready   = ifcvf_mdev_set_vq_ready,
+	.get_vq_ready	= ifcvf_mdev_get_vq_ready,
+	.set_vq_num     = ifcvf_mdev_set_vq_num,
+	.set_vq_address = ifcvf_mdev_set_vq_address,
+	.kick_vq        = ifcvf_mdev_kick_vq,
+	.get_generation	= ifcvf_mdev_get_generation,
+	.get_device_id	= ifcvf_mdev_get_device_id,
+	.get_vendor_id	= ifcvf_mdev_get_vendor_id,
+	.get_vq_align	= ifcvf_mdev_get_vq_align,
+	.get_config	= ifcvf_mdev_get_config,
+	.set_config	= ifcvf_mdev_set_config,
+};
+
+static int ifcvf_init_msix(struct ifcvf_adapter *adapter)
+{
+	struct pci_dev *pdev = to_pci_dev(adapter->dev);
+	struct ifcvf_hw *vf = &adapter->vf;
+	int vector, i, ret, irq;
+
+	ret = pci_alloc_irq_vectors(pdev, IFCVF_MAX_INTR,
+				    IFCVF_MAX_INTR, PCI_IRQ_MSIX);
+	if (ret < 0) {
+		IFC_ERR(adapter->dev, "Failed to alloc irq vectors\n");
+		return ret;
+	}
+
+	for (i = 0; i < IFCVF_MAX_QUEUE_PAIRS * 2; i++) {
+		vector = i + IFCVF_MSI_QUEUE_OFF;
+		irq = pci_irq_vector(pdev, vector);
+		ret = request_irq(irq, ifcvf_intr_handler, 0,
+				pci_name(pdev), &vf->vring[i]);
+		if (ret) {
+			IFC_ERR(adapter->dev,
+				"Failed to request irq for vq %d\n", i);
+			return ret;
+		}
+	}
+
+	return 0;
+}
+
+static void ifcvf_destroy_adapter(struct ifcvf_adapter *adapter)
+{
+	struct ifcvf_hw *vf = IFC_PRIVATE_TO_VF(adapter);
+	struct pci_dev *pdev = to_pci_dev(adapter->dev);
+	int i, vector, irq;
+
+	for (i = 0; i < IFCVF_MAX_QUEUE_PAIRS * 2; i++) {
+		vector = i + IFCVF_MSI_QUEUE_OFF;
+		irq = pci_irq_vector(pdev, vector);
+		free_irq(irq, &vf->vring[i]);
+	}
+}
+
+static ssize_t name_show(struct kobject *kobj, struct device *dev, char *buf)
+{
+	const char *name = "IFC VF virtio/vhost accelerator (virtio ring compatible)";
+
+	return sprintf(buf, "%s\n", name);
+}
+MDEV_TYPE_ATTR_RO(name);
+
+static ssize_t device_api_show(struct kobject *kobj, struct device *dev,
+			       char *buf)
+{
+	return sprintf(buf, "%s\n", VIRTIO_MDEV_DEVICE_API_STRING);
+}
+MDEV_TYPE_ATTR_RO(device_api);
+
+static ssize_t available_instances_show(struct kobject *kobj,
+					struct device *dev, char *buf)
+{
+	struct pci_dev *pdev;
+	struct ifcvf_adapter *adapter;
+
+	pdev = to_pci_dev(dev);
+	adapter = pci_get_drvdata(pdev);
+
+	return sprintf(buf, "%d\n", adapter->mdev_count);
+}
+
+MDEV_TYPE_ATTR_RO(available_instances);
+
+static ssize_t type_show(struct kobject *kobj,
+			struct device *dev, char *buf)
+{
+	return sprintf(buf, "%s\n", "net");
+}
+
+MDEV_TYPE_ATTR_RO(type);
+
+
+static struct attribute *mdev_types_attrs[] = {
+	&mdev_type_attr_name.attr,
+	&mdev_type_attr_device_api.attr,
+	&mdev_type_attr_available_instances.attr,
+	&mdev_type_attr_type.attr,
+	NULL,
+};
+
+static struct attribute_group mdev_type_group_virtio = {
+	.name  = "virtio_mdev",
+	.attrs = mdev_types_attrs,
+};
+
+static struct attribute_group mdev_type_group_vhost = {
+	.name  = "vhost_mdev",
+	.attrs = mdev_types_attrs,
+};
+
+static struct attribute_group *mdev_type_groups[] = {
+	&mdev_type_group_virtio,
+	&mdev_type_group_vhost,
+	NULL,
+};
+
+const struct attribute_group *mdev_dev_groups[] = {
+	NULL,
+};
+
+static int ifcvf_mdev_create(struct kobject *kobj, struct mdev_device *mdev)
+{
+	struct device *dev = mdev_parent_dev(mdev);
+	struct pci_dev *pdev = to_pci_dev(dev);
+	struct ifcvf_adapter *adapter = pci_get_drvdata(pdev);
+	int ret = 0;
+
+	mutex_lock(&adapter->mdev_lock);
+
+	if (adapter->mdev_count < IFCVF_MDEV_LIMIT) {
+		IFC_ERR(&pdev->dev,
+			"Can not create mdev, reached limitation %d\n",
+			IFCVF_MDEV_LIMIT);
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (!strcmp(kobj->name, "ifcvf-virtio_mdev"))
+		mdev_set_virtio_ops(mdev, &ifc_mdev_ops);
+
+	if (!strcmp(kobj->name, "ifcvf-vhost_mdev"))
+		mdev_set_vhost_ops(mdev, &ifc_mdev_ops);
+
+	mdev_set_drvdata(mdev, adapter);
+	mdev_set_iommu_device(mdev_dev(mdev), dev);
+	adapter->mdev_count--;
+
+out:
+	mutex_unlock(&adapter->mdev_lock);
+	return ret;
+}
+
+static int ifcvf_mdev_remove(struct mdev_device *mdev)
+{
+	struct device *dev = mdev_parent_dev(mdev);
+	struct pci_dev *pdev = to_pci_dev(dev);
+	struct ifcvf_adapter *adapter = pci_get_drvdata(pdev);
+
+	mutex_lock(&adapter->mdev_lock);
+	adapter->mdev_count++;
+	mutex_unlock(&adapter->mdev_lock);
+
+	return 0;
+}
+
+static struct mdev_parent_ops ifcvf_mdev_fops = {
+	.owner			= THIS_MODULE,
+	.supported_type_groups	= mdev_type_groups,
+	.mdev_attr_groups	= mdev_dev_groups,
+	.create			= ifcvf_mdev_create,
+	.remove			= ifcvf_mdev_remove,
+};
+
+static int ifcvf_probe(struct pci_dev *pdev, const struct pci_device_id *id)
+{
+	struct device *dev = &pdev->dev;
+	struct ifcvf_adapter *adapter;
+	struct ifcvf_hw *vf;
+	int ret, i;
+
+	adapter = kzalloc(sizeof(struct ifcvf_adapter), GFP_KERNEL);
+
+	if (adapter == NULL) {
+		ret = -ENOMEM;
+		goto fail;
+	}
+
+	mutex_init(&adapter->mdev_lock);
+	adapter->mdev_count = IFCVF_MDEV_LIMIT;
+	adapter->dev = dev;
+	pci_set_drvdata(pdev, adapter);
+	ret = pci_enable_device(pdev);
+
+	if (ret) {
+		IFC_ERR(adapter->dev, "Failed to enable device\n");
+		goto free_adapter;
+	}
+
+	ret = pci_request_regions(pdev, IFCVF_DRIVER_NAME);
+
+	if (ret) {
+		IFC_ERR(adapter->dev, "Failed to request MMIO region\n");
+		goto disable_device;
+	}
+
+	pci_set_master(pdev);
+	ret = ifcvf_init_msix(adapter);
+
+	if (ret) {
+		IFC_ERR(adapter->dev, "Failed to initialize MSI-X\n");
+		goto free_msix;
+	}
+
+	vf = &adapter->vf;
+	
+	for (i = 0; i < IFCVF_PCI_MAX_RESOURCE; i++) {
+		vf->mem_resource[i].phys_addr = pci_resource_start(pdev, i);
+		vf->mem_resource[i].len = pci_resource_len(pdev, i);
+		if (!vf->mem_resource[i].len) {
+			vf->mem_resource[i].addr = NULL;
+			continue;
+		}
+
+		vf->mem_resource[i].addr = pci_iomap_range(pdev, i, 0,
+				vf->mem_resource[i].len);
+		if (!vf->mem_resource[i].addr) {
+			IFC_ERR(adapter->dev, "Failed to map IO resource %d\n",
+				i);
+			ret = -1;
+			goto free_msix;
+		}
+	}
+
+	if (ifcvf_init_hw(vf, pdev) < 0) {
+		ret = -1;
+		goto destroy_adapter;
+	}
+
+	ret = mdev_register_device(dev, &ifcvf_mdev_fops);
+
+	if (ret) {
+		IFC_ERR(adapter->dev,  "Failed to register mdev device\n");
+		goto destroy_adapter;
+	}
+
+	return 0;
+
+destroy_adapter:
+	ifcvf_destroy_adapter(adapter);
+free_msix:
+	pci_free_irq_vectors(pdev);
+	pci_release_regions(pdev);
+disable_device:
+	pci_disable_device(pdev);
+free_adapter:
+	kfree(adapter);
+fail:
+	return ret;
+}
+
+static void ifcvf_remove(struct pci_dev *pdev)
+{
+	struct ifcvf_adapter *adapter = pci_get_drvdata(pdev);
+	struct device *dev = &pdev->dev;
+	struct ifcvf_hw *vf;
+	int i;
+
+	mdev_unregister_device(dev);
+
+	vf = &adapter->vf;
+	for (i = 0; i < IFCVF_PCI_MAX_RESOURCE; i++) {
+		if (vf->mem_resource[i].addr) {
+			pci_iounmap(pdev, vf->mem_resource[i].addr);
+			vf->mem_resource[i].addr = NULL;
+		}
+	}
+
+	ifcvf_destroy_adapter(adapter);
+	pci_free_irq_vectors(pdev);
+	pci_release_regions(pdev);
+	pci_disable_device(pdev);
+	kfree(adapter);
+}
+
+static struct pci_device_id ifcvf_pci_ids[] = {
+	{ PCI_DEVICE_SUB(IFCVF_VENDOR_ID,
+			IFCVF_DEVICE_ID,
+			IFCVF_SUBSYS_VENDOR_ID,
+			IFCVF_SUBSYS_DEVICE_ID) },
+	{ 0 },
+};
+MODULE_DEVICE_TABLE(pci, ifcvf_pci_ids);
+
+static struct pci_driver ifcvf_driver = {
+	.name     = IFCVF_DRIVER_NAME,
+	.id_table = ifcvf_pci_ids,
+	.probe    = ifcvf_probe,
+	.remove   = ifcvf_remove,
+};
+
+static int __init ifcvf_init_module(void)
+{
+	int ret;
+
+	ret = pci_register_driver(&ifcvf_driver);
+	return ret;
+}
+
+static void __exit ifcvf_exit_module(void)
+{
+	pci_unregister_driver(&ifcvf_driver);
+}
+
+module_init(ifcvf_init_module);
+module_exit(ifcvf_exit_module);
+
+MODULE_LICENSE("GPL v2");
+MODULE_VERSION(VERSION_STRING);
+MODULE_AUTHOR(DRIVER_AUTHOR);
-- 
1.8.3.1