* [PATCH] virtio: vdpa: new SolidNET DPU driver.
@ 2022-11-08 10:09 Alvaro Karsz
2022-11-09 6:47 ` Jason Wang
0 siblings, 1 reply; 5+ messages in thread
From: Alvaro Karsz @ 2022-11-08 10:09 UTC (permalink / raw)
To: virtualization; +Cc: Jean Delvare, Guenter Roeck, Michael S. Tsirkin
This commit includes:
1) The driver to manage the controlplane over vDPA bus.
2) A HW monitor device to read health values from the DPU.
Signed-off-by: Alvaro Karsz <alvaro.karsz@solid-run.com>
---
MAINTAINERS | 4 +
drivers/vdpa/Kconfig | 11 +
drivers/vdpa/Makefile | 1 +
drivers/vdpa/solidrun/Makefile | 3 +
drivers/vdpa/solidrun/snet_hwmon.c | 191 +++++++
drivers/vdpa/solidrun/snet_main.c | 858 +++++++++++++++++++++++++++++
drivers/vdpa/solidrun/snet_vdpa.h | 192 +++++++
7 files changed, 1260 insertions(+)
create mode 100644 drivers/vdpa/solidrun/Makefile
create mode 100644 drivers/vdpa/solidrun/snet_hwmon.c
create mode 100644 drivers/vdpa/solidrun/snet_main.c
create mode 100644 drivers/vdpa/solidrun/snet_vdpa.h
diff --git a/MAINTAINERS b/MAINTAINERS
index 046ff06ff97..ae425b3bed4 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -21725,6 +21725,10 @@ IFCVF VIRTIO DATA PATH ACCELERATOR
R: Zhu Lingshan <lingshan.zhu@intel.com>
F: drivers/vdpa/ifcvf/
+SNET DPU VIRTIO DATA PATH ACCELERATOR
+R: Alvaro Karsz <alvaro.karsz@solid-run.com>
+F: drivers/vdpa/solidrun/
+
VIRTIO BALLOON
M: "Michael S. Tsirkin" <mst@redhat.com>
M: David Hildenbrand <david@redhat.com>
diff --git a/drivers/vdpa/Kconfig b/drivers/vdpa/Kconfig
index 50f45d03761..cd880d1af7f 100644
--- a/drivers/vdpa/Kconfig
+++ b/drivers/vdpa/Kconfig
@@ -86,4 +86,15 @@ config ALIBABA_ENI_VDPA
VDPA driver for Alibaba ENI (Elastic Network Interface) which is built upon
virtio 0.9.5 specification.
+ config SNET_VDPA
+ tristate "SolidRun's vDPA driver for SolidNET"
+ select HWMON
+ depends on PCI_MSI
+ help
+ vDPA driver for SolidNET DPU.
+ With this driver, the VirtIO dataplane can be
+ offloaded to a SolidNET DPU.
+ This driver includes a HW monitor device that
+ reads health values from the DPU.
+
endif # VDPA
diff --git a/drivers/vdpa/Makefile b/drivers/vdpa/Makefile
index 15665563a7f..59396ff2a31 100644
--- a/drivers/vdpa/Makefile
+++ b/drivers/vdpa/Makefile
@@ -6,3 +6,4 @@ obj-$(CONFIG_IFCVF) += ifcvf/
obj-$(CONFIG_MLX5_VDPA) += mlx5/
obj-$(CONFIG_VP_VDPA) += virtio_pci/
obj-$(CONFIG_ALIBABA_ENI_VDPA) += alibaba/
+obj-$(CONFIG_SNET_VDPA) += solidrun/
diff --git a/drivers/vdpa/solidrun/Makefile b/drivers/vdpa/solidrun/Makefile
new file mode 100644
index 00000000000..0acabd040fd
--- /dev/null
+++ b/drivers/vdpa/solidrun/Makefile
@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: GPL-2.0
+obj-$(CONFIG_SNET_VDPA) += snet_vdpa.o
+snet_vdpa-$(CONFIG_SNET_VDPA) += snet_main.o snet_hwmon.o
diff --git a/drivers/vdpa/solidrun/snet_hwmon.c b/drivers/vdpa/solidrun/snet_hwmon.c
new file mode 100644
index 00000000000..50e5743b8ee
--- /dev/null
+++ b/drivers/vdpa/solidrun/snet_hwmon.c
@@ -0,0 +1,191 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * SolidRun DPU driver for control plane
+ *
+ * Copyright (C) 2022 SolidRun
+ *
+ * Author: Alvaro Karsz <alvaro.karsz@solid-run.com>
+ *
+ */
+
+#include <linux/hwmon.h>
+#include <linux/hwmon-sysfs.h>
+
+#include "snet_vdpa.h"
+
+/* Monitor offsets */
+#define SNET_MON_TMP0_IN_OFF 0x00
+#define SNET_MON_TMP0_MAX_OFF 0x08
+#define SNET_MON_TMP0_CRIT_OFF 0x10
+#define SNET_MON_TMP1_IN_OFF 0x18
+#define SNET_MON_TMP1_CRIT_OFF 0x20
+#define SNET_MON_CURR_IN_OFF 0x28
+#define SNET_MON_CURR_MAX_OFF 0x30
+#define SNET_MON_CURR_CRIT_OFF 0x38
+#define SNET_MON_PWR_IN_OFF 0x40
+#define SNET_MON_VOLT_IN_OFF 0x48
+#define SNET_MON_VOLT_CRIT_OFF 0x50
+#define SNET_MON_VOLT_LCRIT_OFF 0x58
+
+static void snet_hwmon_read_reg(struct psnet *psnet, u32 reg, long *out)
+{
+ *out = (long)psnet_read64(psnet, psnet->cfg.hwmon_off + reg);
+}
+
+static umode_t snet_howmon_is_visible(const void *data,
+ enum hwmon_sensor_types type,
+ u32 attr, int channel)
+{
+ /* Files are read only */
+ return 0444;
+}
+
+static int snet_howmon_read(struct device *dev, enum hwmon_sensor_types type,
+ u32 attr, int channel, long *val)
+{
+ struct psnet *psnet = dev_get_drvdata(dev);
+ int ret = 0;
+
+ switch (type) {
+ case hwmon_in:
+ switch (attr) {
+ case hwmon_in_lcrit:
+ snet_hwmon_read_reg(psnet, SNET_MON_VOLT_LCRIT_OFF, val);
+ break;
+ case hwmon_in_crit:
+ snet_hwmon_read_reg(psnet, SNET_MON_VOLT_CRIT_OFF, val);
+ break;
+ case hwmon_in_input:
+ snet_hwmon_read_reg(psnet, SNET_MON_VOLT_IN_OFF, val);
+ break;
+ default:
+ ret = -EOPNOTSUPP;
+ break;
+ }
+ break;
+
+ case hwmon_power:
+ switch (attr) {
+ case hwmon_power_input:
+ snet_hwmon_read_reg(psnet, SNET_MON_PWR_IN_OFF, val);
+ break;
+
+ default:
+ ret = -EOPNOTSUPP;
+ break;
+ }
+ break;
+
+ case hwmon_curr:
+ switch (attr) {
+ case hwmon_curr_input:
+ snet_hwmon_read_reg(psnet, SNET_MON_CURR_IN_OFF, val);
+ break;
+ case hwmon_curr_max:
+ snet_hwmon_read_reg(psnet, SNET_MON_CURR_MAX_OFF, val);
+ break;
+ case hwmon_curr_crit:
+ snet_hwmon_read_reg(psnet, SNET_MON_CURR_CRIT_OFF, val);
+ break;
+ default:
+ ret = -EOPNOTSUPP;
+ break;
+ }
+ break;
+
+ case hwmon_temp:
+ switch (attr) {
+ case hwmon_temp_input:
+ if (channel == 0)
+ snet_hwmon_read_reg(psnet, SNET_MON_TMP0_IN_OFF, val);
+ else
+ snet_hwmon_read_reg(psnet, SNET_MON_TMP1_IN_OFF, val);
+ break;
+ case hwmon_temp_max:
+ if (channel == 0)
+ snet_hwmon_read_reg(psnet, SNET_MON_TMP0_MAX_OFF, val);
+ else
+ ret = -EOPNOTSUPP;
+ break;
+ case hwmon_temp_crit:
+ if (channel == 0)
+ snet_hwmon_read_reg(psnet, SNET_MON_TMP0_CRIT_OFF, val);
+ else
+ snet_hwmon_read_reg(psnet, SNET_MON_TMP1_CRIT_OFF, val);
+ break;
+
+ default:
+ ret = -EOPNOTSUPP;
+ break;
+ }
+ break;
+
+ default:
+ ret = -EOPNOTSUPP;
+ break;
+ }
+ return ret;
+}
+
+static int snet_hwmon_read_string(struct device *dev,
+ enum hwmon_sensor_types type, u32 attr,
+ int channel, const char **str)
+{
+ int ret = 0;
+
+ switch (type) {
+ case hwmon_in:
+ *str = "main_vin";
+ break;
+ case hwmon_power:
+ *str = "soc_pin";
+ break;
+ case hwmon_curr:
+ *str = "soc_iin";
+ break;
+ case hwmon_temp:
+ if (channel == 0)
+ *str = "power_stage_temp";
+ else
+ *str = "ic_junction_temp";
+ break;
+ default:
+ ret = -EOPNOTSUPP;
+ break;
+ }
+ return ret;
+}
+
+static const struct hwmon_ops snet_hwmon_ops = {
+ .is_visible = snet_howmon_is_visible,
+ .read = snet_howmon_read,
+ .read_string = snet_hwmon_read_string
+};
+
+static const struct hwmon_channel_info *snet_hwmon_info[] = {
+ HWMON_CHANNEL_INFO(temp, HWMON_T_INPUT | HWMON_T_MAX | HWMON_T_CRIT | HWMON_T_LABEL,
+ HWMON_T_INPUT | HWMON_T_CRIT | HWMON_T_LABEL),
+ HWMON_CHANNEL_INFO(power, HWMON_P_INPUT | HWMON_P_LABEL),
+ HWMON_CHANNEL_INFO(curr, HWMON_C_INPUT | HWMON_C_MAX | HWMON_C_CRIT | HWMON_C_LABEL),
+ HWMON_CHANNEL_INFO(in, HWMON_I_INPUT | HWMON_I_CRIT | HWMON_I_LCRIT | HWMON_I_LABEL),
+ NULL
+};
+
+static const struct hwmon_chip_info snet_hwmono_info = {
+ .ops = &snet_hwmon_ops,
+ .info = snet_hwmon_info,
+};
+
+/* Create an HW monitor device */
+void snet_create_hwmon(struct pci_dev *pdev)
+{
+ struct device *hwmon;
+ struct psnet *psnet = pci_get_drvdata(pdev);
+
+ snprintf(psnet->hwmon_name, SNET_HWMON_NAME_SIZE, "snet%d_hwmon", psnet->id);
+ hwmon = devm_hwmon_device_register_with_info(&pdev->dev, psnet->hwmon_name, psnet,
+ &snet_hwmono_info, NULL);
+ /* The monitor is not mandatory, Just alert user in case of an error */
+ if (!hwmon)
+ SNET_WARN(pdev, "Failed to create SNET hwmon\n");
+}
diff --git a/drivers/vdpa/solidrun/snet_main.c b/drivers/vdpa/solidrun/snet_main.c
new file mode 100644
index 00000000000..c27e345b3bb
--- /dev/null
+++ b/drivers/vdpa/solidrun/snet_main.c
@@ -0,0 +1,858 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * SolidRun DPU driver for control plane
+ *
+ * Copyright (C) 2022 SolidRun
+ *
+ * Author: Alvaro Karsz <alvaro.karsz@solid-run.com>
+ *
+ */
+
+#include "snet_vdpa.h"
+
+/* SolidRun's PCI vendor ID */
+#define SOLIDRUN_PCI_VENDOR_ID 0xD063
+#define SNET_DEVICE_ID 0x1000
+/* PCI Bar used to access DPU */
+#define SNET_ACCESS_BAR 0x2
+/* SNET signature */
+#define SNET_SIGNATURE 0xD0D06363
+/* Max. config version that we can work with */
+#define SNET_CFG_VERSION 0x1
+/* Queue align */
+#define SNET_QUEUE_ALIGNMENT PAGE_SIZE
+/* Kick value to notify that new data is available */
+#define SNET_KICK_VAL 0x1
+#define SNET_CONFIG_OFF 0x0
+
+enum snet_msg {
+ SNET_MSG_CLEAR = 0,
+ SNET_MSG_DESTROY = 1,
+};
+
+/* Global variable to hold the next psnet unique id to be used */
+static u32 psnet_next_uid;
+
+static struct snet *vdpa_to_snet(struct vdpa_device *vdpa)
+{
+ return container_of(vdpa, struct snet, vdpa);
+}
+
+static int snet_set_vq_address(struct vdpa_device *vdev, u16 idx, u64 desc_area,
+ u64 driver_area, u64 device_area)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+ /* save received parameters in vqueue sturct */
+ snet->vqs[idx]->desc_area = desc_area;
+ snet->vqs[idx]->driver_area = driver_area;
+ snet->vqs[idx]->device_area = device_area;
+ /* Set serial ID */
+ snet->vqs[idx]->sid = snet->psnet->next_vq++;
+ /* Set kick address - every VQ gets 4B */
+ snet->vqs[idx]->kick_offset = snet->psnet->cfg.kick_off + snet->vqs[idx]->sid * 4;
+ /* Save kick ptr*/
+ snet->vqs[idx]->kick_ptr = snet->psnet->access_bar + snet->vqs[idx]->kick_offset;
+ /* Clear kick address for this VQ */
+ iowrite32(0, snet->vqs[idx]->kick_ptr);
+
+ return 0;
+}
+
+static void snet_set_vq_num(struct vdpa_device *vdev, u16 idx, u32 num)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+ /* save num in vqueue */
+ snet->vqs[idx]->num = num;
+}
+
+static void snet_kick_vq(struct vdpa_device *vdev, u16 idx)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+ /* not ready - ignore */
+ if (!snet->vqs[idx]->ready)
+ return;
+
+ iowrite32(SNET_KICK_VAL, snet->vqs[idx]->kick_ptr);
+}
+
+static void snet_set_vq_cb(struct vdpa_device *vdev, u16 idx, struct vdpa_callback *cb)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+ snet->vqs[idx]->cb.callback = cb->callback;
+ snet->vqs[idx]->cb.private = cb->private;
+}
+
+static void snet_set_vq_ready(struct vdpa_device *vdev, u16 idx, bool ready)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+
+ snet->vqs[idx]->ready = ready;
+}
+
+static bool snet_get_vq_ready(struct vdpa_device *vdev, u16 idx)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+ return snet->vqs[idx]->ready;
+}
+
+static int snet_set_vq_state(struct vdpa_device *vdev, u16 idx, const struct vdpa_vq_state *state)
+{
+ /* Not supported */
+ return 0;
+}
+
+static int snet_get_vq_state(struct vdpa_device *vdev, u16 idx, struct vdpa_vq_state *state)
+{
+ /* Not supported */
+ return -EOPNOTSUPP;
+}
+
+static int snet_get_vq_irq(struct vdpa_device *vdev, u16 idx)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+ return snet->vqs[idx]->irq;
+}
+
+static u32 snet_get_vq_align(struct vdpa_device *vdev)
+{
+ return (u32)SNET_QUEUE_ALIGNMENT;
+}
+
+static void snet_reset(struct snet *snet)
+{
+ /* no reset yet - just update generation */
+ snet->generation++;
+}
+
+static int snet_main_reset(struct vdpa_device *vdev)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+ snet_reset(snet);
+ return 0;
+}
+
+static size_t snet_get_config_size(struct vdpa_device *vdev)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+ return (size_t)snet->cfg->cfg_size;
+}
+
+static u64 snet_get_features(struct vdpa_device *vdev)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+ return snet->cfg->features;
+}
+
+static int snet_set_drv_features(struct vdpa_device *vdev, u64 features)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+ snet->used_features = snet->cfg->features & features;
+ return 0;
+}
+
+static u64 snet_get_drv_features(struct vdpa_device *vdev)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+ return snet->used_features;
+}
+
+static u16 snet_get_vq_num_max(struct vdpa_device *vdev)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+ return (u16)snet->cfg->vq_size;
+}
+
+static void snet_set_config_cb(struct vdpa_device *vdev, struct vdpa_callback *cb)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+ snet->cb.callback = cb->callback;
+ snet->cb.private = cb->private;
+}
+
+static u32 snet_get_device_id(struct vdpa_device *vdev)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+ return snet->cfg->virtio_id;
+}
+
+static u32 snet_get_vendor_id(struct vdpa_device *vdev)
+{
+ return (u32)SOLIDRUN_PCI_VENDOR_ID;
+}
+
+static u8 snet_get_status(struct vdpa_device *vdev)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+ return snet->status;
+}
+
+static bool is_last_snet_in_pdev(struct snet *snet)
+{
+ /* Check if this is the last device (count starts from 0) */
+ return snet->sid == snet->psnet->total_devs - 1;
+}
+
+static void snet_send_msg(struct snet *snet, u32 msg)
+{
+ psnet_write32(snet->psnet, snet->cfg->snet_msg_off, msg);
+ /* Wait until the msg is written.
+ * We need to be sure that the DPU gets the msg.
+ */
+ smp_wmb();
+}
+
+static void snet_write_conf(struct snet *snet)
+{
+ struct psnet *psnet = snet->psnet;
+ u32 off, magic_off, next_off;
+ int i;
+
+ /* No need to notify twice for the same SNET device */
+ if (snet->dpu_ready)
+ return;
+
+ /* Get offset from snet pdev */
+ off = snet->psnet->next_dpu_addr;
+
+ /* Snet data :
+ * 0x0 0x4 0x8 0xC 0x10 0x14 0x18
+ * | MAGIC NUMBER | CFG VER | SNET SID | SNET TYPE | NUMBER OF QUEUES | CFG IRQ IDX |
+ *
+ * 0x18 0x1C 0x24 0x28 0x50
+ * | CONFIG OFF | FEATURES | MSG OFF | RESERVED |
+ *
+ * For every VQ:
+ * 0x0 0x4 0x8 0xC 0x20
+ * | VQ SID | QUEUE SIZE | IRQ Index |
+ * | DESC AREA |
+ * | DEVICE AREA |
+ * | DRIVER AREA |
+ * | KICK AREA |
+ * | RESERVED |
+ *
+ * Footer:
+ * 0x0 0x8
+ * | Next Device Address |
+ *
+ * If this is the last device, footer is set to 0
+ *
+ * Magic number should be written last, this is the DPU indication that the data is ready
+ */
+ magic_off = off;
+ off += 4;
+ psnet_write32(psnet, off, psnet->negotiated_cfg_ver);
+ off += 4;
+ psnet_write32(psnet, off, snet->sid);
+ off += 4;
+ psnet_write32(psnet, off, snet->cfg->virtio_id);
+ off += 4;
+ psnet_write32(psnet, off, snet->cfg->vq_num);
+ off += 4;
+ psnet_write32(psnet, off, snet->cfg_irq_idx);
+ off += 4;
+ psnet_write32(psnet, off, snet->cfg->cfg_off);
+ off += 4;
+ psnet_write64(psnet, off, snet->used_features);
+ off += 8;
+ psnet_write32(psnet, off, snet->cfg->snet_msg_off);
+ off += 4;
+ /* Ignore reserved */
+ off += 40;
+ /* Write VQs */
+ for (i = 0 ; i < snet->cfg->vq_num ; i++) {
+ psnet_write32(psnet, off, i);
+ off += 4;
+ psnet_write32(psnet, off, snet->vqs[i]->num);
+ off += 4;
+ psnet_write32(psnet, off, snet->vqs[i]->irq_idx);
+ off += 4;
+ psnet_write64(psnet, off, snet->vqs[i]->desc_area);
+ off += 8;
+ psnet_write64(psnet, off, snet->vqs[i]->device_area);
+ off += 8;
+ psnet_write64(psnet, off, snet->vqs[i]->driver_area);
+ off += 8;
+ psnet_write64(psnet, off, snet->vqs[i]->kick_offset);
+ off += 8;
+ /* Ignore reserved */
+ off += 32;
+ }
+ /* Calculate and write the next address */
+ if (is_last_snet_in_pdev(snet))
+ next_off = 0;
+ else
+ next_off = off + 8;
+
+ psnet_write64(psnet, off, next_off);
+ off += 8;
+ /* Save next address in snet pdev */
+ snet->psnet->next_dpu_addr = next_off;
+ /* Clear snet messages address */
+ snet_send_msg(snet, SNET_MSG_CLEAR);
+ /* Write magic number - data is ready */
+ psnet_write32(psnet, magic_off, SNET_SIGNATURE);
+ /* set DPU flag */
+ snet->dpu_ready = true;
+}
+
+static void snet_set_status(struct vdpa_device *vdev, u8 status)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+ snet->status = status;
+ /*if status is 0, do a reset*/
+ if (!status)
+ snet_reset(snet);
+
+ /* If device is ready - write device configuration to the DPU */
+ if (status & VIRTIO_CONFIG_S_DRIVER_OK)
+ snet_write_conf(snet);
+}
+
+static void snet_get_config(struct vdpa_device *vdev, unsigned int offset,
+ void *buf, unsigned int len)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+ u32 i;
+ /* check for offset error */
+ if (offset + len > snet->cfg->cfg_size)
+ return;
+
+ /* Write into buffer */
+ for (i = 0; i < len; i++)
+ *(u8 *)((u64)buf + i) = ioread8(snet->cfg->virtio_cfg + offset + i);
+}
+
+static void snet_set_config(struct vdpa_device *vdev, unsigned int offset,
+ const void *buf, unsigned int len)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+ u8 i;
+ /* check for offset error */
+ if (offset + len > snet->cfg->cfg_size)
+ return;
+
+ /* Write into PCIe BAR */
+ for (i = 0; i < len; i++)
+ iowrite8(*(u8 *)((u64)buf + i), snet->cfg->virtio_cfg + offset + i);
+}
+
+static u32 snet_get_generation(struct vdpa_device *vdev)
+{
+ struct snet *snet = vdpa_to_snet(vdev);
+
+ return snet->generation;
+}
+
+static struct vdpa_iova_range snet_get_iova_range(struct vdpa_device *vdev)
+{
+ struct vdpa_iova_range range = {
+ .first = 0ULL,
+ .last = ULLONG_MAX,
+ };
+
+ return range;
+}
+
+static void snet_free(struct vdpa_device *vdev)
+{
+ /* Resources are freed when pci's remove callback is called */
+}
+
+static const struct vdpa_config_ops snet_config_ops = {
+ .set_vq_address = snet_set_vq_address,
+ .set_vq_num = snet_set_vq_num,
+ .kick_vq = snet_kick_vq,
+ .set_vq_cb = snet_set_vq_cb,
+ .set_vq_ready = snet_set_vq_ready,
+ .get_vq_ready = snet_get_vq_ready,
+ .set_vq_state = snet_set_vq_state,
+ .get_vq_state = snet_get_vq_state,
+ .get_vq_irq = snet_get_vq_irq,
+ .get_vq_align = snet_get_vq_align,
+ .reset = snet_main_reset,
+ .get_config_size = snet_get_config_size,
+ .get_device_features = snet_get_features,
+ .set_driver_features = snet_set_drv_features,
+ .get_driver_features = snet_get_drv_features,
+ .get_vq_num_min = snet_get_vq_num_max,
+ .get_vq_num_max = snet_get_vq_num_max,
+ .set_config_cb = snet_set_config_cb,
+ .get_device_id = snet_get_device_id,
+ .get_vendor_id = snet_get_vendor_id,
+ .get_status = snet_get_status,
+ .set_status = snet_set_status,
+ .get_config = snet_get_config,
+ .set_config = snet_set_config,
+ .get_generation = snet_get_generation,
+ .get_iova_range = snet_get_iova_range,
+ .free = snet_free,
+};
+
+static int snet_open_access_bar(struct pci_dev *pdev, struct psnet *psnet)
+{
+ int ret;
+ /* Request BAR */
+ ret = pci_request_region(pdev, SNET_ACCESS_BAR, "snet-notify");
+ if (ret) {
+ SNET_ERR(pdev, "Failed to request PCI BAR\n");
+ return ret;
+ }
+
+ /* Map BAR */
+ psnet->access_bar = pci_iomap(pdev, SNET_ACCESS_BAR, 0);
+ if (!psnet->access_bar) {
+ SNET_ERR(pdev, "Failed to map PCI BAR\n");
+ ret = -ENOMEM;
+ goto release;
+ }
+ return 0;
+release:
+ pci_release_region(pdev, SNET_ACCESS_BAR);
+ return ret;
+}
+
+static void snet_free_cfg(struct snet_cfg *cfg)
+{
+ u32 i;
+
+ if (!cfg->devs)
+ return;
+
+ /* Free devices */
+ for (i = 0; i < cfg->devices_num; i++) {
+ if (!cfg->devs[i])
+ break;
+
+ kfree(cfg->devs[i]);
+ }
+ /* Free pointers to devices */
+ kfree(cfg->devs);
+}
+
+/* Read SNET config from PCIe BAR */
+static int snet_read_cfg(struct pci_dev *pdev, struct psnet *psnet)
+{
+ struct snet_cfg *cfg = &psnet->cfg;
+ u32 i, off;
+ /* Move to where the config starts */
+ off = SNET_CONFIG_OFF;
+
+ /* SNET DPU will write SNET's signature when the config is ready */
+ cfg->key = psnet_read32(psnet, off);
+ off += 4;
+ if (cfg->key != SNET_SIGNATURE) {
+ SNET_INFO(pdev, "No valid config found in PCI BAR.\n");
+ return -ENODEV;
+ }
+
+ /* Load general config data */
+ cfg->cfg_size = psnet_read32(psnet, off);
+ off += 4;
+ cfg->cfg_ver = psnet_read32(psnet, off);
+ off += 4;
+ /* The negotiated config version is the lower one between this driver's config
+ * and the DPU's.
+ */
+ psnet->negotiated_cfg_ver = min_t(u32, cfg->cfg_ver, SNET_CFG_VERSION);
+ SNET_DBG(pdev, "SNET config version %u\n", psnet->negotiated_cfg_ver);
+
+ cfg->irqs_num = psnet_read32(psnet, off);
+ off += 4;
+ cfg->dpu_off = psnet_read32(psnet, off);
+ off += 4;
+ cfg->kick_off = psnet_read32(psnet, off);
+ off += 4;
+ cfg->hwmon_off = psnet_read32(psnet, off);
+ off += 4;
+ cfg->flags = psnet_read32(psnet, off);
+ off += 4;
+ /* Ignore Reserved */
+ off += sizeof(cfg->rsvd);
+
+ cfg->devices_num = psnet_read32(psnet, off);
+ off += 4;
+ /* Allocate memory to hold pointer to the devices */
+ cfg->devs = kcalloc(cfg->devices_num, sizeof(void *), GFP_KERNEL);
+ if (!cfg->devs) {
+ SNET_ERR(pdev, "Failed to allocate memory for SNET config\n");
+ return -ENOMEM;
+ }
+
+ /* Load device configuration from BAR */
+ for (i = 0; i < cfg->devices_num; i++) {
+ cfg->devs[i] = kzalloc(sizeof(struct snet_dev_cfg), GFP_KERNEL);
+ if (!cfg->devs[i]) {
+ SNET_ERR(pdev, "Failed to allocate memory for SNET config\n");
+ snet_free_cfg(cfg);
+ return -ENOMEM;
+ }
+ /* Read device config */
+ cfg->devs[i]->virtio_id = psnet_read32(psnet, off);
+ off += 4;
+ cfg->devs[i]->vq_num = psnet_read32(psnet, off);
+ off += 4;
+ cfg->devs[i]->vq_size = psnet_read32(psnet, off);
+ off += 4;
+ cfg->devs[i]->features = psnet_read64(psnet, off);
+ off += 8;
+ cfg->devs[i]->snet_msg_off = psnet_read32(psnet, off);
+ off += 4;
+ /* Ignore Reserved */
+ off += sizeof(cfg->devs[i]->rsvd);
+
+ cfg->devs[i]->cfg_size = psnet_read32(psnet, off);
+ off += 4;
+ cfg->devs[i]->cfg_off = psnet_read32(psnet, off);
+ off += 4;
+ /* Save VirtIO config address in BAR */
+ cfg->devs[i]->virtio_cfg = psnet->access_bar + off;
+ /* Move to next device, we don't want to load the VirtIO config */
+ off += cfg->devs[i]->cfg_size;
+ }
+ return 0;
+}
+
+static int snet_alloc_irq_vector(struct pci_dev *pdev, struct psnet *psnet)
+{
+ int ret = 0;
+
+ ret = pci_alloc_irq_vectors(pdev, psnet->cfg.irqs_num, psnet->cfg.irqs_num, PCI_IRQ_MSIX);
+ if (ret <= 0) {
+ SNET_ERR(pdev, "Failed to allocate IRQ vectors\n");
+ return ret;
+ } else if (ret != psnet->cfg.irqs_num) {
+ SNET_WARN(pdev, "Allocated %u irq vectors instead of %u\n", ret,
+ psnet->cfg.irqs_num);
+ }
+ return 0;
+}
+
+static void snet_free_vqs(struct snet *snet)
+{
+ u32 i;
+
+ if (!snet->vqs)
+ return;
+
+ for (i = 0 ; i < snet->cfg->vq_num ; i++) {
+ if (!snet->vqs[i])
+ break;
+
+ kfree(snet->vqs[i]);
+ }
+ kfree(snet->vqs);
+}
+
+static int snet_build_vqs(struct pci_dev *pdev, struct snet *snet)
+{
+ u32 i;
+ /* Allocate the array of VQ pointers */
+ snet->vqs = kcalloc(snet->cfg->vq_num, sizeof(void *), GFP_KERNEL);
+ if (!snet->vqs) {
+ SNET_ERR(pdev, "Failed to allocate memory for SNET VQs\n");
+ return -ENOMEM;
+ }
+
+ /* Allocate VQs */
+ for (i = 0; i < snet->cfg->vq_num; i++) {
+ /* Allocate VQ memory */
+ snet->vqs[i] = kzalloc(sizeof(struct snet_vq), GFP_KERNEL);
+ if (!snet->vqs[i]) {
+ SNET_ERR(pdev, "Failed to allocate memory for a SNET VQ\n");
+ snet_free_vqs(snet);
+ return -ENOMEM;
+ }
+ }
+ return 0;
+}
+
+static irqreturn_t snet_cfg_irq_hndlr(int irq, void *data)
+{
+ struct snet *snet = data;
+ /* Call callback if any */
+ if (snet->cb.callback)
+ return snet->cb.callback(snet->cb.private);
+
+ return IRQ_HANDLED;
+}
+
+static irqreturn_t snet_vq_irq_hndlr(int irq, void *data)
+{
+ struct snet_vq *vq = data;
+ /* Call callback if any */
+ if (vq->cb.callback)
+ return vq->cb.callback(vq->cb.private);
+
+ return IRQ_HANDLED;
+}
+
+static int snet_request_irqs(struct pci_dev *pdev, struct snet *snet)
+{
+ int ret, i;
+ /* Request one IRQ for every VQ, and one for the config change */
+ /* Get next free IRQ ID */
+ snet->cfg_irq_idx = snet->psnet->next_irq++;
+ snprintf(snet->cfg_irq_name, SNET_IRQ_NAME_SIZE, "snet-cfg-%d", snet->cfg_irq_idx);
+
+ /* Request config IRQ */
+ snet->cfg_irq = pci_irq_vector(pdev, snet->cfg_irq_idx);
+ ret = devm_request_irq(&pdev->dev, snet->cfg_irq, snet_cfg_irq_hndlr, 0,
+ snet->cfg_irq_name, snet);
+ if (ret) {
+ SNET_ERR(pdev, "Failed to request IRQ\n");
+ return ret;
+ }
+
+ /* Request IRQ for every VQ */
+ for (i = 0; i < snet->cfg->vq_num; i++) {
+ /* Get next free IRQ ID */
+ snet->vqs[i]->irq_idx = snet->psnet->next_irq++;
+ /* Write IRQ name */
+ snprintf(snet->vqs[i]->irq_name, SNET_IRQ_NAME_SIZE, "snet[%u]-vq[%u]",
+ snet->sid, i);
+ /* Request IRQ */
+ snet->vqs[i]->irq = pci_irq_vector(pdev, snet->vqs[i]->irq_idx);
+ ret = devm_request_irq(&pdev->dev, snet->vqs[i]->irq, snet_vq_irq_hndlr, 0,
+ snet->vqs[i]->irq_name, snet->vqs[i]);
+ if (ret) {
+ SNET_ERR(pdev, "Failed to request IRQ\n");
+ /* Free all IRQs requested so far and exit */
+ devm_free_irq(&pdev->dev, snet->cfg_irq, snet);
+ i--;
+ for (; i != -1; i--)
+ devm_free_irq(&pdev->dev, snet->vqs[i]->irq, snet->vqs[i]);
+ return ret;
+ }
+ }
+ return 0;
+}
+
+static void free_snet_irqs(struct pci_dev *pdev, struct snet *snet)
+{
+ u32 i;
+ /* Free config's IRQ */
+ devm_free_irq(&pdev->dev, snet->cfg_irq, snet);
+ /* Free VQ IRQs */
+ for (i = 0; i < snet->cfg->vq_num; i++)
+ devm_free_irq(&pdev->dev, snet->vqs[i]->irq, snet->vqs[i]);
+}
+
+static int snet_alloc_dev(struct pci_dev *pdev, struct psnet *psnet, struct snet_dev_cfg *dev_cfg)
+{
+ struct snet *snet;
+ int ret;
+ /* Allocate vdpa device */
+ snet = vdpa_alloc_device(struct snet, vdpa, &pdev->dev, &snet_config_ops, 1, 1, NULL,
+ false);
+ if (!snet) {
+ SNET_ERR(pdev, "Failed to allocate a vdpa device\n");
+ return -ENOMEM;
+ }
+ snet->psnet = psnet;
+ snet->cfg = dev_cfg;
+ snet->dpu_ready = false;
+ snet->sid = psnet->next_snet++;
+
+ /* build snet VQs */
+ ret = snet_build_vqs(pdev, snet);
+ if (ret)
+ goto put_device;
+
+ /* Request IRQs */
+ ret = snet_request_irqs(pdev, snet);
+ if (ret)
+ goto free_vqs;
+
+ /*set DMA device*/
+ snet->vdpa.dma_dev = &pdev->dev;
+
+ /* Register VDPA device */
+ ret = vdpa_register_device(&snet->vdpa, snet->cfg->vq_num);
+ if (ret) {
+ SNET_ERR(pdev, "Failed to register vdpa device\n");
+ goto free_irqs;
+ }
+
+ /* Save snet in psnet */
+ psnet->devs[psnet->dev_num++] = snet;
+ return 0;
+
+free_irqs:
+ free_snet_irqs(pdev, snet);
+free_vqs:
+ snet_free_vqs(snet);
+put_device:
+ put_device(&snet->vdpa.dev);
+ return ret;
+}
+
+/* Free a single snet device */
+static void snet_free_dev(struct pci_dev *pdev, struct snet *snet)
+{
+ vdpa_unregister_device(&snet->vdpa);
+ snet_free_vqs(snet);
+ free_snet_irqs(pdev, snet);
+}
+
+/* Free all snet devices */
+static void snet_free_devs(struct pci_dev *pdev, struct psnet *psnet)
+{
+ u32 i;
+
+ for (i = 0 ; i < psnet->total_devs ; i++) {
+ if (psnet->devs[i])
+ snet_free_dev(pdev, psnet->devs[i]);
+ else
+ break;
+ }
+ kfree(psnet->devs);
+}
+
+static int snet_create_devs(struct pci_dev *pdev, struct psnet *psnet)
+{
+ int ret, i;
+ /* Allocate memory to hold the device pointers */
+ psnet->total_devs = psnet->cfg.devices_num;
+ psnet->devs = kcalloc(psnet->total_devs, sizeof(void *), GFP_KERNEL);
+ if (!psnet->devs) {
+ SNET_ERR(pdev, "Failed to allocate memory for SNET devices\n");
+ return -ENOMEM;
+ }
+ /* init next address */
+ psnet->next_dpu_addr = psnet->cfg.dpu_off;
+ /* Give the device an unique ID */
+ psnet->id = psnet_next_uid++;
+ /* Allocate all devices */
+ for (i = 0; i < psnet->total_devs; i++) {
+ ret = snet_alloc_dev(pdev, psnet, psnet->cfg.devs[i]);
+ if (ret) {
+ snet_free_devs(pdev, psnet);
+ return ret;
+ }
+ }
+ SNET_INFO(pdev, "Created %u SNET devices\n", psnet->total_devs);
+ return 0;
+}
+
+static int snet_vdpa_probe(struct pci_dev *pdev, const struct pci_device_id *id)
+{
+ struct psnet *psnet;
+ int ret = 0;
+
+ ret = pcim_enable_device(pdev);
+ if (ret) {
+ SNET_ERR(pdev, "Failed to enable PCI device\n");
+ return ret;
+ }
+
+ psnet = kzalloc(sizeof(struct psnet), GFP_KERNEL);
+ if (!psnet) {
+ SNET_ERR(pdev, "Failed to allocate memory for PSNET device\n");
+ ret = -ENOMEM;
+ goto disable_dev;
+ }
+
+ /* Open SNET BAR */
+ ret = snet_open_access_bar(pdev, psnet);
+ if (ret)
+ goto free_psnet;
+
+ /* Try to read SNET's config from PCIe BAR */
+ ret = snet_read_cfg(pdev, psnet);
+ if (ret)
+ goto release_bar;
+
+ /* Request for MSI-X IRQs */
+ ret = snet_alloc_irq_vector(pdev, psnet);
+ if (ret)
+ goto free_cfg;
+
+ /* Create all devices from config */
+ ret = snet_create_devs(pdev, psnet);
+ if (ret)
+ goto free_irqs;
+
+ pci_set_master(pdev);
+ pci_set_drvdata(pdev, psnet);
+
+ /* Create an HW monitor device */
+ if (SNET_CFG_FLAG(psnet, SNET_CFG_FLAG_HWMON))
+ snet_create_hwmon(pdev);
+
+ return 0;
+
+free_irqs:
+ pci_free_irq_vectors(pdev);
+free_cfg:
+ snet_free_cfg(&psnet->cfg);
+release_bar:
+ pci_iounmap(pdev, psnet->access_bar);
+ pci_release_region(pdev, SNET_ACCESS_BAR);
+free_psnet:
+ kfree(psnet);
+disable_dev:
+ pci_disable_device(pdev);
+ return ret;
+}
+
+static void snet_vdpa_remove(struct pci_dev *pdev)
+{
+ struct psnet *psnet = pci_get_drvdata(pdev);
+ struct snet *snet;
+ u32 i;
+
+ /* Iterate through all SNET devices and send a "destroy" message */
+ for (i = 0; i < psnet->cfg.devices_num; i++) {
+ snet = psnet->devs[i];
+ snet_send_msg(snet, SNET_MSG_DESTROY);
+ /* Reset status */
+ snet->status = 0;
+ }
+ /* Free all SNET devices */
+ snet_free_devs(pdev, psnet);
+ /* Free IRQ vectors */
+ pci_free_irq_vectors(pdev);
+ /* Free SNET config read from DPU */
+ snet_free_cfg(&psnet->cfg);
+ /* Unmap and release PCI BAR */
+ pci_iounmap(pdev, psnet->access_bar);
+ pci_release_region(pdev, SNET_ACCESS_BAR);
+ /* Free PSNET device */
+ SNET_INFO(pdev, "Removed %u SNET devices\n", psnet->total_devs);
+ kfree(psnet);
+}
+
+static struct pci_device_id snet_driver_pci_ids[] = {
+ { PCI_DEVICE_SUB(SOLIDRUN_PCI_VENDOR_ID, SNET_DEVICE_ID,
+ SOLIDRUN_PCI_VENDOR_ID, SNET_DEVICE_ID) },
+ { 0 },
+};
+
+MODULE_DEVICE_TABLE(pci, snet_driver_pci_ids);
+
+static struct pci_driver snet_vdpa_driver = {
+ .name = "snet-vdpa-driver",
+ .id_table = snet_driver_pci_ids,
+ .probe = snet_vdpa_probe,
+ .remove = snet_vdpa_remove,
+};
+
+module_pci_driver(snet_vdpa_driver);
+
+MODULE_AUTHOR("Alvaro Karsz <alvaro.karsz@solid-run.com>");
+MODULE_DESCRIPTION("SolidRun vDPA driver");
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/vdpa/solidrun/snet_vdpa.h b/drivers/vdpa/solidrun/snet_vdpa.h
new file mode 100644
index 00000000000..e44c900832a
--- /dev/null
+++ b/drivers/vdpa/solidrun/snet_vdpa.h
@@ -0,0 +1,192 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * SolidRun DPU driver for control plane
+ *
+ * Copyright (C) 2022 SolidRun
+ *
+ * Author: Alvaro Karsz <alvaro.karsz@solid-run.com>
+ *
+ */
+#ifndef _SNET_VDPA_H_
+#define _SNET_VDPA_H_
+
+#include <linux/vdpa.h>
+#include <linux/pci.h>
+
+#define SNET_IRQ_NAME_SIZE 20
+#define SNET_HWMON_NAME_SIZE 20
+
+#define SNET_ERR(pdev, fmt, ...) dev_err(&pdev->dev, "%s"fmt, "snet_vdpa: ", ##__VA_ARGS__)
+#define SNET_WARN(pdev, fmt, ...) dev_warn(&pdev->dev, "%s"fmt, "snet_vdpa: ", ##__VA_ARGS__)
+#define SNET_INFO(pdev, fmt, ...) dev_info(&pdev->dev, "%s"fmt, "snet_vdpa: ", ##__VA_ARGS__)
+#define SNET_DBG(pdev, fmt, ...) dev_dbg(&pdev->dev, "%s"fmt, "snet_vdpa: ", ##__VA_ARGS__)
+
+/* VQ struct */
+struct snet_vq {
+ /* VQ callback */
+ struct vdpa_callback cb;
+ /* desc base address */
+ u64 desc_area;
+ /* device base address */
+ u64 device_area;
+ /* driver base address */
+ u64 driver_area;
+ /* Queue size */
+ u32 num;
+ /* Kick address offset for this VQ */
+ u64 kick_offset;
+ /* Serial ID for VQ */
+ u32 sid;
+ /* is ready flag */
+ bool ready;
+ /* IRQ number */
+ u32 irq;
+ /* IRQ index, DPU uses this to parse data from MSI-X table */
+ u32 irq_idx;
+ /* IRQ name */
+ char irq_name[SNET_IRQ_NAME_SIZE];
+ /* pointer to mapped PCI BAR register used by this VQ to kick */
+ void __iomem *kick_ptr;
+};
+
+struct snet {
+ /* vdpa device */
+ struct vdpa_device vdpa;
+ /* Config callback */
+ struct vdpa_callback cb;
+ /* array of virqueues */
+ struct snet_vq **vqs;
+ /* Used features */
+ u64 used_features;
+ /* driver features */
+ u64 drv_features;
+ /* device generation */
+ u32 generation;
+ /* Device serial ID */
+ u32 sid;
+ /* device status */
+ u8 status;
+ /* boolean indicating if VQ data was dumped to DPU */
+ bool dpu_ready;
+ /* IRQ number */
+ u32 cfg_irq;
+ /* IRQ index, DPU uses this to parse data from MSI-X table */
+ u32 cfg_irq_idx;
+ /* IRQ name */
+ char cfg_irq_name[SNET_IRQ_NAME_SIZE];
+ /* Pointer to snet pdev parent device */
+ struct psnet *psnet;
+ /* Pointer to snet config device */
+ struct snet_dev_cfg *cfg;
+};
+
+struct snet_dev_cfg {
+ /* Device ID following VirtIO spec. */
+ u32 virtio_id;
+ /* Number of VQs for this device (NOT PAIRS!) */
+ u32 vq_num;
+ /* Size of every VQ */
+ u32 vq_size;
+ /* Device features, following VirtIO spec */
+ u64 features;
+ /* Offset to write messages to the DPU SNET_MSG_* */
+ u32 snet_msg_off;
+ /* Reserved for future usage */
+ u32 rsvd[6];
+ /* VirtIO device specific config size */
+ u32 cfg_size;
+ /* VirtIO device specific config offset in BAR */
+ u32 cfg_off;
+ /* VirtIO device specific config address */
+ void __iomem *virtio_cfg;
+} __packed;
+
+struct snet_cfg {
+ /* Magic key */
+ u32 key;
+ /* Size of total config in bytes */
+ u32 cfg_size;
+ /* Config version */
+ u32 cfg_ver;
+ /* Number of IRQs to be asked */
+ u32 irqs_num;
+ /* Offset in PCIe BAR for writing device data to DPU */
+ u32 dpu_off;
+ /* Offset in PCIe BAR for VQ kicks */
+ u32 kick_off;
+ /* Offset in PCIe BAR for HW monitoring */
+ u32 hwmon_off;
+ /* Config general flags - enum snet_cfg_flags */
+ u32 flags;
+ /* Reserved for future usage */
+ u32 rsvd[6];
+ /* Number of snet devices */
+ u32 devices_num;
+ /* The actual devices */
+ struct snet_dev_cfg **devs;
+} __packed;
+
+/* SolidNET PCIe device, one device per PCIe physical function */
+struct psnet {
+ /* Unique snet pdev id */
+ u32 id;
+ /* Number of snet devices */
+ u32 dev_num;
+ /* Control IRQ IDs */
+ u32 next_irq;
+ /* Control device IDs */
+ u32 next_snet;
+ /* Control VQ IDs */
+ u32 next_vq;
+ /* Total number of devices (including future devices) */
+ u32 total_devs;
+ /* Negotiated config version */
+ u32 negotiated_cfg_ver;
+ /* Next address to write snet info to DPU */
+ u64 next_dpu_addr;
+ /* devices list */
+ struct snet **devs;
+ /* pointer to mapped PCI BAR */
+ void __iomem *access_bar;
+ /* Pointer to the device's config read from BAR */
+ struct snet_cfg cfg;
+ /* Name of monitor device */
+ char hwmon_name[SNET_HWMON_NAME_SIZE];
+};
+
+enum snet_cfg_flags {
+ /* Create a HWMON device */
+ SNET_CFG_FLAG_HWMON = BIT(0),
+};
+
+/* Check if a config flag is set */
+#define SNET_CFG_FLAG(p, f) ((p)->cfg.flags & (f))
+
+static inline u32 psnet_read32(struct psnet *psnet, u32 off)
+{
+ return ioread32(psnet->access_bar + off);
+}
+
+static inline void psnet_write32(struct psnet *psnet, u32 off, u32 val)
+{
+ iowrite32(val, psnet->access_bar + off);
+}
+
+static inline u64 psnet_read64(struct psnet *psnet, u32 off)
+{
+ u64 val;
+
+ val = (u64)ioread32(psnet->access_bar + off);
+ val |= ((u64)ioread32(psnet->access_bar + off + 4) << 32);
+ return val;
+}
+
+static inline void psnet_write64(struct psnet *psnet, u32 off, u64 val)
+{
+ psnet_write32(psnet, off, val >> 32);
+ psnet_write32(psnet, off + 4, val & 0xFFFFFFFF);
+}
+
+void snet_create_hwmon(struct pci_dev *pdev);
+
+#endif //_SNET_VDPA_H_
--
2.32.0
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
^ permalink raw reply related [flat|nested] 5+ messages in thread
* Re: [PATCH] virtio: vdpa: new SolidNET DPU driver.
2022-11-08 10:09 [PATCH] virtio: vdpa: new SolidNET DPU driver Alvaro Karsz
@ 2022-11-09 6:47 ` Jason Wang
2022-11-09 9:44 ` Alvaro Karsz
0 siblings, 1 reply; 5+ messages in thread
From: Jason Wang @ 2022-11-09 6:47 UTC (permalink / raw)
To: Alvaro Karsz, virtualization
Cc: Jean Delvare, Guenter Roeck, Michael S. Tsirkin
在 2022/11/8 18:09, Alvaro Karsz 写道:
> This commit includes:
> 1) The driver to manage the controlplane over vDPA bus.
> 2) A HW monitor device to read health values from the DPU.
>
> Signed-off-by: Alvaro Karsz <alvaro.karsz@solid-run.com>
> ---
> MAINTAINERS | 4 +
> drivers/vdpa/Kconfig | 11 +
> drivers/vdpa/Makefile | 1 +
> drivers/vdpa/solidrun/Makefile | 3 +
> drivers/vdpa/solidrun/snet_hwmon.c | 191 +++++++
> drivers/vdpa/solidrun/snet_main.c | 858 +++++++++++++++++++++++++++++
> drivers/vdpa/solidrun/snet_vdpa.h | 192 +++++++
> 7 files changed, 1260 insertions(+)
> create mode 100644 drivers/vdpa/solidrun/Makefile
> create mode 100644 drivers/vdpa/solidrun/snet_hwmon.c
> create mode 100644 drivers/vdpa/solidrun/snet_main.c
> create mode 100644 drivers/vdpa/solidrun/snet_vdpa.h
>
> diff --git a/MAINTAINERS b/MAINTAINERS
> index 046ff06ff97..ae425b3bed4 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -21725,6 +21725,10 @@ IFCVF VIRTIO DATA PATH ACCELERATOR
> R: Zhu Lingshan <lingshan.zhu@intel.com>
> F: drivers/vdpa/ifcvf/
>
> +SNET DPU VIRTIO DATA PATH ACCELERATOR
> +R: Alvaro Karsz <alvaro.karsz@solid-run.com>
> +F: drivers/vdpa/solidrun/
> +
> VIRTIO BALLOON
> M: "Michael S. Tsirkin" <mst@redhat.com>
> M: David Hildenbrand <david@redhat.com>
> diff --git a/drivers/vdpa/Kconfig b/drivers/vdpa/Kconfig
> index 50f45d03761..cd880d1af7f 100644
> --- a/drivers/vdpa/Kconfig
> +++ b/drivers/vdpa/Kconfig
> @@ -86,4 +86,15 @@ config ALIBABA_ENI_VDPA
> VDPA driver for Alibaba ENI (Elastic Network Interface) which is built upon
> virtio 0.9.5 specification.
>
> + config SNET_VDPA
> + tristate "SolidRun's vDPA driver for SolidNET"
> + select HWMON
> + depends on PCI_MSI
> + help
> + vDPA driver for SolidNET DPU.
> + With this driver, the VirtIO dataplane can be
> + offloaded to a SolidNET DPU.
> + This driver includes a HW monitor device that
> + reads health values from the DPU.
> +
> endif # VDPA
> diff --git a/drivers/vdpa/Makefile b/drivers/vdpa/Makefile
> index 15665563a7f..59396ff2a31 100644
> --- a/drivers/vdpa/Makefile
> +++ b/drivers/vdpa/Makefile
> @@ -6,3 +6,4 @@ obj-$(CONFIG_IFCVF) += ifcvf/
> obj-$(CONFIG_MLX5_VDPA) += mlx5/
> obj-$(CONFIG_VP_VDPA) += virtio_pci/
> obj-$(CONFIG_ALIBABA_ENI_VDPA) += alibaba/
> +obj-$(CONFIG_SNET_VDPA) += solidrun/
> diff --git a/drivers/vdpa/solidrun/Makefile b/drivers/vdpa/solidrun/Makefile
> new file mode 100644
> index 00000000000..0acabd040fd
> --- /dev/null
> +++ b/drivers/vdpa/solidrun/Makefile
> @@ -0,0 +1,3 @@
> +# SPDX-License-Identifier: GPL-2.0
> +obj-$(CONFIG_SNET_VDPA) += snet_vdpa.o
> +snet_vdpa-$(CONFIG_SNET_VDPA) += snet_main.o snet_hwmon.o
> diff --git a/drivers/vdpa/solidrun/snet_hwmon.c b/drivers/vdpa/solidrun/snet_hwmon.c
> new file mode 100644
> index 00000000000..50e5743b8ee
> --- /dev/null
> +++ b/drivers/vdpa/solidrun/snet_hwmon.c
> @@ -0,0 +1,191 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * SolidRun DPU driver for control plane
> + *
> + * Copyright (C) 2022 SolidRun
> + *
> + * Author: Alvaro Karsz <alvaro.karsz@solid-run.com>
> + *
> + */
> +
> +#include <linux/hwmon.h>
> +#include <linux/hwmon-sysfs.h>
> +
> +#include "snet_vdpa.h"
> +
> +/* Monitor offsets */
> +#define SNET_MON_TMP0_IN_OFF 0x00
> +#define SNET_MON_TMP0_MAX_OFF 0x08
> +#define SNET_MON_TMP0_CRIT_OFF 0x10
> +#define SNET_MON_TMP1_IN_OFF 0x18
> +#define SNET_MON_TMP1_CRIT_OFF 0x20
> +#define SNET_MON_CURR_IN_OFF 0x28
> +#define SNET_MON_CURR_MAX_OFF 0x30
> +#define SNET_MON_CURR_CRIT_OFF 0x38
> +#define SNET_MON_PWR_IN_OFF 0x40
> +#define SNET_MON_VOLT_IN_OFF 0x48
> +#define SNET_MON_VOLT_CRIT_OFF 0x50
> +#define SNET_MON_VOLT_LCRIT_OFF 0x58
> +
> +static void snet_hwmon_read_reg(struct psnet *psnet, u32 reg, long *out)
> +{
> + *out = (long)psnet_read64(psnet, psnet->cfg.hwmon_off + reg);
> +}
> +
> +static umode_t snet_howmon_is_visible(const void *data,
> + enum hwmon_sensor_types type,
> + u32 attr, int channel)
> +{
> + /* Files are read only */
> + return 0444;
> +}
> +
> +static int snet_howmon_read(struct device *dev, enum hwmon_sensor_types type,
> + u32 attr, int channel, long *val)
> +{
> + struct psnet *psnet = dev_get_drvdata(dev);
> + int ret = 0;
> +
> + switch (type) {
> + case hwmon_in:
> + switch (attr) {
> + case hwmon_in_lcrit:
> + snet_hwmon_read_reg(psnet, SNET_MON_VOLT_LCRIT_OFF, val);
> + break;
> + case hwmon_in_crit:
> + snet_hwmon_read_reg(psnet, SNET_MON_VOLT_CRIT_OFF, val);
> + break;
> + case hwmon_in_input:
> + snet_hwmon_read_reg(psnet, SNET_MON_VOLT_IN_OFF, val);
> + break;
> + default:
> + ret = -EOPNOTSUPP;
> + break;
> + }
> + break;
> +
> + case hwmon_power:
> + switch (attr) {
> + case hwmon_power_input:
> + snet_hwmon_read_reg(psnet, SNET_MON_PWR_IN_OFF, val);
> + break;
> +
> + default:
> + ret = -EOPNOTSUPP;
> + break;
> + }
> + break;
> +
> + case hwmon_curr:
> + switch (attr) {
> + case hwmon_curr_input:
> + snet_hwmon_read_reg(psnet, SNET_MON_CURR_IN_OFF, val);
> + break;
> + case hwmon_curr_max:
> + snet_hwmon_read_reg(psnet, SNET_MON_CURR_MAX_OFF, val);
> + break;
> + case hwmon_curr_crit:
> + snet_hwmon_read_reg(psnet, SNET_MON_CURR_CRIT_OFF, val);
> + break;
> + default:
> + ret = -EOPNOTSUPP;
> + break;
> + }
> + break;
> +
> + case hwmon_temp:
> + switch (attr) {
> + case hwmon_temp_input:
> + if (channel == 0)
> + snet_hwmon_read_reg(psnet, SNET_MON_TMP0_IN_OFF, val);
> + else
> + snet_hwmon_read_reg(psnet, SNET_MON_TMP1_IN_OFF, val);
> + break;
> + case hwmon_temp_max:
> + if (channel == 0)
> + snet_hwmon_read_reg(psnet, SNET_MON_TMP0_MAX_OFF, val);
> + else
> + ret = -EOPNOTSUPP;
> + break;
> + case hwmon_temp_crit:
> + if (channel == 0)
> + snet_hwmon_read_reg(psnet, SNET_MON_TMP0_CRIT_OFF, val);
> + else
> + snet_hwmon_read_reg(psnet, SNET_MON_TMP1_CRIT_OFF, val);
> + break;
> +
> + default:
> + ret = -EOPNOTSUPP;
> + break;
> + }
> + break;
> +
> + default:
> + ret = -EOPNOTSUPP;
> + break;
> + }
> + return ret;
> +}
> +
> +static int snet_hwmon_read_string(struct device *dev,
> + enum hwmon_sensor_types type, u32 attr,
> + int channel, const char **str)
> +{
> + int ret = 0;
> +
> + switch (type) {
> + case hwmon_in:
> + *str = "main_vin";
> + break;
> + case hwmon_power:
> + *str = "soc_pin";
> + break;
> + case hwmon_curr:
> + *str = "soc_iin";
> + break;
> + case hwmon_temp:
> + if (channel == 0)
> + *str = "power_stage_temp";
> + else
> + *str = "ic_junction_temp";
> + break;
> + default:
> + ret = -EOPNOTSUPP;
> + break;
> + }
> + return ret;
> +}
> +
> +static const struct hwmon_ops snet_hwmon_ops = {
> + .is_visible = snet_howmon_is_visible,
> + .read = snet_howmon_read,
> + .read_string = snet_hwmon_read_string
> +};
> +
> +static const struct hwmon_channel_info *snet_hwmon_info[] = {
> + HWMON_CHANNEL_INFO(temp, HWMON_T_INPUT | HWMON_T_MAX | HWMON_T_CRIT | HWMON_T_LABEL,
> + HWMON_T_INPUT | HWMON_T_CRIT | HWMON_T_LABEL),
> + HWMON_CHANNEL_INFO(power, HWMON_P_INPUT | HWMON_P_LABEL),
> + HWMON_CHANNEL_INFO(curr, HWMON_C_INPUT | HWMON_C_MAX | HWMON_C_CRIT | HWMON_C_LABEL),
> + HWMON_CHANNEL_INFO(in, HWMON_I_INPUT | HWMON_I_CRIT | HWMON_I_LCRIT | HWMON_I_LABEL),
> + NULL
> +};
> +
> +static const struct hwmon_chip_info snet_hwmono_info = {
> + .ops = &snet_hwmon_ops,
> + .info = snet_hwmon_info,
> +};
> +
> +/* Create an HW monitor device */
> +void snet_create_hwmon(struct pci_dev *pdev)
> +{
> + struct device *hwmon;
> + struct psnet *psnet = pci_get_drvdata(pdev);
> +
> + snprintf(psnet->hwmon_name, SNET_HWMON_NAME_SIZE, "snet%d_hwmon", psnet->id);
> + hwmon = devm_hwmon_device_register_with_info(&pdev->dev, psnet->hwmon_name, psnet,
> + &snet_hwmono_info, NULL);
> + /* The monitor is not mandatory, Just alert user in case of an error */
> + if (!hwmon)
> + SNET_WARN(pdev, "Failed to create SNET hwmon\n");
> +}
> diff --git a/drivers/vdpa/solidrun/snet_main.c b/drivers/vdpa/solidrun/snet_main.c
> new file mode 100644
> index 00000000000..c27e345b3bb
> --- /dev/null
> +++ b/drivers/vdpa/solidrun/snet_main.c
> @@ -0,0 +1,858 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * SolidRun DPU driver for control plane
> + *
> + * Copyright (C) 2022 SolidRun
> + *
> + * Author: Alvaro Karsz <alvaro.karsz@solid-run.com>
> + *
> + */
> +
> +#include "snet_vdpa.h"
> +
> +/* SolidRun's PCI vendor ID */
> +#define SOLIDRUN_PCI_VENDOR_ID 0xD063
> +#define SNET_DEVICE_ID 0x1000
> +/* PCI Bar used to access DPU */
> +#define SNET_ACCESS_BAR 0x2
> +/* SNET signature */
> +#define SNET_SIGNATURE 0xD0D06363
> +/* Max. config version that we can work with */
> +#define SNET_CFG_VERSION 0x1
> +/* Queue align */
> +#define SNET_QUEUE_ALIGNMENT PAGE_SIZE
> +/* Kick value to notify that new data is available */
> +#define SNET_KICK_VAL 0x1
> +#define SNET_CONFIG_OFF 0x0
> +
> +enum snet_msg {
> + SNET_MSG_CLEAR = 0,
> + SNET_MSG_DESTROY = 1,
> +};
> +
> +/* Global variable to hold the next psnet unique id to be used */
> +static u32 psnet_next_uid;
> +
> +static struct snet *vdpa_to_snet(struct vdpa_device *vdpa)
> +{
> + return container_of(vdpa, struct snet, vdpa);
> +}
> +
> +static int snet_set_vq_address(struct vdpa_device *vdev, u16 idx, u64 desc_area,
> + u64 driver_area, u64 device_area)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> + /* save received parameters in vqueue sturct */
> + snet->vqs[idx]->desc_area = desc_area;
> + snet->vqs[idx]->driver_area = driver_area;
> + snet->vqs[idx]->device_area = device_area;
> + /* Set serial ID */
> + snet->vqs[idx]->sid = snet->psnet->next_vq++;
> + /* Set kick address - every VQ gets 4B */
> + snet->vqs[idx]->kick_offset = snet->psnet->cfg.kick_off + snet->vqs[idx]->sid * 4;
If I was not wrong, this value seems doesn't change. If yes, I wonder we
can move the kick_offset logic to snet_alloc_dev()?
> + /* Save kick ptr*/
> + snet->vqs[idx]->kick_ptr = snet->psnet->access_bar + snet->vqs[idx]->kick_offset;
> + /* Clear kick address for this VQ */
> + iowrite32(0, snet->vqs[idx]->kick_ptr);
> +
> + return 0;
> +}
> +
> +static void snet_set_vq_num(struct vdpa_device *vdev, u16 idx, u32 num)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> + /* save num in vqueue */
> + snet->vqs[idx]->num = num;
> +}
> +
> +static void snet_kick_vq(struct vdpa_device *vdev, u16 idx)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> + /* not ready - ignore */
> + if (!snet->vqs[idx]->ready)
> + return;
> +
> + iowrite32(SNET_KICK_VAL, snet->vqs[idx]->kick_ptr);
> +}
> +
> +static void snet_set_vq_cb(struct vdpa_device *vdev, u16 idx, struct vdpa_callback *cb)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> + snet->vqs[idx]->cb.callback = cb->callback;
> + snet->vqs[idx]->cb.private = cb->private;
> +}
> +
> +static void snet_set_vq_ready(struct vdpa_device *vdev, u16 idx, bool ready)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> +
> + snet->vqs[idx]->ready = ready;
> +}
> +
> +static bool snet_get_vq_ready(struct vdpa_device *vdev, u16 idx)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> + return snet->vqs[idx]->ready;
> +}
> +
> +static int snet_set_vq_state(struct vdpa_device *vdev, u16 idx, const struct vdpa_vq_state *state)
> +{
> + /* Not supported */
> + return 0;
-EOPNOTSUPP?
> +}
> +
> +static int snet_get_vq_state(struct vdpa_device *vdev, u16 idx, struct vdpa_vq_state *state)
> +{
> + /* Not supported */
> + return -EOPNOTSUPP;
> +}
> +
> +static int snet_get_vq_irq(struct vdpa_device *vdev, u16 idx)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> + return snet->vqs[idx]->irq;
> +}
> +
> +static u32 snet_get_vq_align(struct vdpa_device *vdev)
> +{
> + return (u32)SNET_QUEUE_ALIGNMENT;
> +}
> +
> +static void snet_reset(struct snet *snet)
> +{
> + /* no reset yet - just update generation */
> + snet->generation++;
I fail to understand how can this work. E.g after reset there should be
no interaction from the device like DMA, otherwise there could have
security implications.
> +}
> +
> +static int snet_main_reset(struct vdpa_device *vdev)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> + snet_reset(snet);
> + return 0;
> +}
> +
> +static size_t snet_get_config_size(struct vdpa_device *vdev)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> + return (size_t)snet->cfg->cfg_size;
> +}
> +
> +static u64 snet_get_features(struct vdpa_device *vdev)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> + return snet->cfg->features;
> +}
> +
> +static int snet_set_drv_features(struct vdpa_device *vdev, u64 features)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> + snet->used_features = snet->cfg->features & features;
> + return 0;
> +}
> +
> +static u64 snet_get_drv_features(struct vdpa_device *vdev)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> + return snet->used_features;
> +}
> +
> +static u16 snet_get_vq_num_max(struct vdpa_device *vdev)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> + return (u16)snet->cfg->vq_size;
> +}
> +
> +static void snet_set_config_cb(struct vdpa_device *vdev, struct vdpa_callback *cb)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> + snet->cb.callback = cb->callback;
> + snet->cb.private = cb->private;
> +}
> +
> +static u32 snet_get_device_id(struct vdpa_device *vdev)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> + return snet->cfg->virtio_id;
> +}
> +
> +static u32 snet_get_vendor_id(struct vdpa_device *vdev)
> +{
> + return (u32)SOLIDRUN_PCI_VENDOR_ID;
> +}
> +
> +static u8 snet_get_status(struct vdpa_device *vdev)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> + return snet->status;
> +}
> +
> +static bool is_last_snet_in_pdev(struct snet *snet)
> +{
> + /* Check if this is the last device (count starts from 0) */
> + return snet->sid == snet->psnet->total_devs - 1;
> +}
> +
> +static void snet_send_msg(struct snet *snet, u32 msg)
> +{
> + psnet_write32(snet->psnet, snet->cfg->snet_msg_off, msg);
> + /* Wait until the msg is written.
Since read is ordered with write, a more easy way is to perform a ioread
here.
> + * We need to be sure that the DPU gets the msg.
> + */
> + smp_wmb();
Interesting, which barrier is this paired?
> +}
> +
> +static void snet_write_conf(struct snet *snet)
> +{
> + struct psnet *psnet = snet->psnet;
> + u32 off, magic_off, next_off;
> + int i;
> +
> + /* No need to notify twice for the same SNET device */
> + if (snet->dpu_ready)
> + return;
> +
> + /* Get offset from snet pdev */
> + off = snet->psnet->next_dpu_addr;
> +
> + /* Snet data :
> + * 0x0 0x4 0x8 0xC 0x10 0x14 0x18
> + * | MAGIC NUMBER | CFG VER | SNET SID | SNET TYPE | NUMBER OF QUEUES | CFG IRQ IDX |
> + *
> + * 0x18 0x1C 0x24 0x28 0x50
> + * | CONFIG OFF | FEATURES | MSG OFF | RESERVED |
> + *
> + * For every VQ:
> + * 0x0 0x4 0x8 0xC 0x20
> + * | VQ SID | QUEUE SIZE | IRQ Index |
> + * | DESC AREA |
> + * | DEVICE AREA |
> + * | DRIVER AREA |
> + * | KICK AREA |
> + * | RESERVED |
> + *
> + * Footer:
> + * 0x0 0x8
> + * | Next Device Address |
> + *
> + * If this is the last device, footer is set to 0
> + *
> + * Magic number should be written last, this is the DPU indication that the data is ready
> + */
> + magic_off = off;
> + off += 4;
> + psnet_write32(psnet, off, psnet->negotiated_cfg_ver);
> + off += 4;
> + psnet_write32(psnet, off, snet->sid);
> + off += 4;
> + psnet_write32(psnet, off, snet->cfg->virtio_id);
> + off += 4;
> + psnet_write32(psnet, off, snet->cfg->vq_num);
> + off += 4;
> + psnet_write32(psnet, off, snet->cfg_irq_idx);
> + off += 4;
> + psnet_write32(psnet, off, snet->cfg->cfg_off);
> + off += 4;
> + psnet_write64(psnet, off, snet->used_features);
> + off += 8;
> + psnet_write32(psnet, off, snet->cfg->snet_msg_off);
> + off += 4;
> + /* Ignore reserved */
> + off += 40;
> + /* Write VQs */
> + for (i = 0 ; i < snet->cfg->vq_num ; i++) {
> + psnet_write32(psnet, off, i);
> + off += 4;
> + psnet_write32(psnet, off, snet->vqs[i]->num);
> + off += 4;
> + psnet_write32(psnet, off, snet->vqs[i]->irq_idx);
> + off += 4;
> + psnet_write64(psnet, off, snet->vqs[i]->desc_area);
Do we need to do endian conversion here (cpu_to_le64())?
> + off += 8;
> + psnet_write64(psnet, off, snet->vqs[i]->device_area);
> + off += 8;
> + psnet_write64(psnet, off, snet->vqs[i]->driver_area);
> + off += 8;
> + psnet_write64(psnet, off, snet->vqs[i]->kick_offset);
> + off += 8;
> + /* Ignore reserved */
> + off += 32;
> + }
> + /* Calculate and write the next address */
> + if (is_last_snet_in_pdev(snet))
> + next_off = 0;
> + else
> + next_off = off + 8;
> +
> + psnet_write64(psnet, off, next_off);
> + off += 8;
> + /* Save next address in snet pdev */
> + snet->psnet->next_dpu_addr = next_off;
> + /* Clear snet messages address */
> + snet_send_msg(snet, SNET_MSG_CLEAR);
> + /* Write magic number - data is ready */
> + psnet_write32(psnet, magic_off, SNET_SIGNATURE);
> + /* set DPU flag */
> + snet->dpu_ready = true;
> +}
> +
> +static void snet_set_status(struct vdpa_device *vdev, u8 status)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> + snet->status = status;
> + /*if status is 0, do a reset*/
> + if (!status)
> + snet_reset(snet);
> +
> + /* If device is ready - write device configuration to the DPU */
> + if (status & VIRTIO_CONFIG_S_DRIVER_OK)
> + snet_write_conf(snet);
> +}
> +
> +static void snet_get_config(struct vdpa_device *vdev, unsigned int offset,
> + void *buf, unsigned int len)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> + u32 i;
> + /* check for offset error */
> + if (offset + len > snet->cfg->cfg_size)
> + return;
> +
> + /* Write into buffer */
> + for (i = 0; i < len; i++)
> + *(u8 *)((u64)buf + i) = ioread8(snet->cfg->virtio_cfg + offset + i);
> +}
> +
> +static void snet_set_config(struct vdpa_device *vdev, unsigned int offset,
> + const void *buf, unsigned int len)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> + u8 i;
> + /* check for offset error */
> + if (offset + len > snet->cfg->cfg_size)
> + return;
> +
> + /* Write into PCIe BAR */
> + for (i = 0; i < len; i++)
> + iowrite8(*(u8 *)((u64)buf + i), snet->cfg->virtio_cfg + offset + i);
Need to take endianess into account.
> +}
> +
> +static u32 snet_get_generation(struct vdpa_device *vdev)
> +{
> + struct snet *snet = vdpa_to_snet(vdev);
> +
> + return snet->generation;
> +}
> +
> +static struct vdpa_iova_range snet_get_iova_range(struct vdpa_device *vdev)
> +{
> + struct vdpa_iova_range range = {
> + .first = 0ULL,
> + .last = ULLONG_MAX,
> + };
> +
> + return range;
If I was not wrong, the device depends on the platform IOMMU to work. So
unless device have a more strict iova range than what platform IOMMU can
provide, we can simply not advertise this and use the one that is
provided by the IOMMU:
static void vhost_vdpa_set_iova_range(struct vhost_vdpa *v)
{
struct vdpa_iova_range *range = &v->range;
struct vdpa_device *vdpa = v->vdpa;
const struct vdpa_config_ops *ops = vdpa->config;
if (ops->get_iova_range) {
*range = ops->get_iova_range(vdpa);
} else if (v->domain && v->domain->geometry.force_aperture) {
range->first = v->domain->geometry.aperture_start;
range->last = v->domain->geometry.aperture_end;
} else {
range->first = 0;
range->last = ULLONG_MAX;
}
}
> +}
> +
> +static void snet_free(struct vdpa_device *vdev)
> +{
> + /* Resources are freed when pci's remove callback is called */
> +}
> +
> +static const struct vdpa_config_ops snet_config_ops = {
> + .set_vq_address = snet_set_vq_address,
> + .set_vq_num = snet_set_vq_num,
> + .kick_vq = snet_kick_vq,
> + .set_vq_cb = snet_set_vq_cb,
> + .set_vq_ready = snet_set_vq_ready,
> + .get_vq_ready = snet_get_vq_ready,
> + .set_vq_state = snet_set_vq_state,
> + .get_vq_state = snet_get_vq_state,
> + .get_vq_irq = snet_get_vq_irq,
> + .get_vq_align = snet_get_vq_align,
> + .reset = snet_main_reset,
> + .get_config_size = snet_get_config_size,
> + .get_device_features = snet_get_features,
> + .set_driver_features = snet_set_drv_features,
> + .get_driver_features = snet_get_drv_features,
> + .get_vq_num_min = snet_get_vq_num_max,
> + .get_vq_num_max = snet_get_vq_num_max,
> + .set_config_cb = snet_set_config_cb,
> + .get_device_id = snet_get_device_id,
> + .get_vendor_id = snet_get_vendor_id,
> + .get_status = snet_get_status,
> + .set_status = snet_set_status,
> + .get_config = snet_get_config,
> + .set_config = snet_set_config,
> + .get_generation = snet_get_generation,
> + .get_iova_range = snet_get_iova_range,
> + .free = snet_free,
> +};
> +
> +static int snet_open_access_bar(struct pci_dev *pdev, struct psnet *psnet)
> +{
> + int ret;
> + /* Request BAR */
> + ret = pci_request_region(pdev, SNET_ACCESS_BAR, "snet-notify");
Any chance to use device managed region helper here? It seems to
simplify the codes (e.g the cleanup stuffs).
> + if (ret) {
> + SNET_ERR(pdev, "Failed to request PCI BAR\n");
> + return ret;
> + }
> +
> + /* Map BAR */
> + psnet->access_bar = pci_iomap(pdev, SNET_ACCESS_BAR, 0);
> + if (!psnet->access_bar) {
> + SNET_ERR(pdev, "Failed to map PCI BAR\n");
> + ret = -ENOMEM;
> + goto release;
> + }
> + return 0;
> +release:
> + pci_release_region(pdev, SNET_ACCESS_BAR);
> + return ret;
> +}
> +
> +static void snet_free_cfg(struct snet_cfg *cfg)
> +{
> + u32 i;
> +
> + if (!cfg->devs)
> + return;
> +
> + /* Free devices */
> + for (i = 0; i < cfg->devices_num; i++) {
> + if (!cfg->devs[i])
> + break;
> +
> + kfree(cfg->devs[i]);
> + }
> + /* Free pointers to devices */
> + kfree(cfg->devs);
> +}
> +
> +/* Read SNET config from PCIe BAR */
> +static int snet_read_cfg(struct pci_dev *pdev, struct psnet *psnet)
> +{
> + struct snet_cfg *cfg = &psnet->cfg;
> + u32 i, off;
> + /* Move to where the config starts */
> + off = SNET_CONFIG_OFF;
> +
> + /* SNET DPU will write SNET's signature when the config is ready */
> + cfg->key = psnet_read32(psnet, off);
> + off += 4;
> + if (cfg->key != SNET_SIGNATURE) {
> + SNET_INFO(pdev, "No valid config found in PCI BAR.\n");
Is this better to say "config is not ready"? Btw, I wonder if it makes
more sense to wait until the confg is ready with a timeout?
> + return -ENODEV;
> + }
> +
> + /* Load general config data */
> + cfg->cfg_size = psnet_read32(psnet, off);
> + off += 4;
> + cfg->cfg_ver = psnet_read32(psnet, off);
> + off += 4;
> + /* The negotiated config version is the lower one between this driver's config
> + * and the DPU's.
> + */
> + psnet->negotiated_cfg_ver = min_t(u32, cfg->cfg_ver, SNET_CFG_VERSION);
> + SNET_DBG(pdev, "SNET config version %u\n", psnet->negotiated_cfg_ver);
> +
> + cfg->irqs_num = psnet_read32(psnet, off);
> + off += 4;
> + cfg->dpu_off = psnet_read32(psnet, off);
> + off += 4;
> + cfg->kick_off = psnet_read32(psnet, off);
> + off += 4;
> + cfg->hwmon_off = psnet_read32(psnet, off);
> + off += 4;
> + cfg->flags = psnet_read32(psnet, off);
> + off += 4;
> + /* Ignore Reserved */
> + off += sizeof(cfg->rsvd);
> +
> + cfg->devices_num = psnet_read32(psnet, off);
> + off += 4;
> + /* Allocate memory to hold pointer to the devices */
> + cfg->devs = kcalloc(cfg->devices_num, sizeof(void *), GFP_KERNEL);
> + if (!cfg->devs) {
> + SNET_ERR(pdev, "Failed to allocate memory for SNET config\n");
> + return -ENOMEM;
> + }
> +
> + /* Load device configuration from BAR */
> + for (i = 0; i < cfg->devices_num; i++) {
> + cfg->devs[i] = kzalloc(sizeof(struct snet_dev_cfg), GFP_KERNEL);
> + if (!cfg->devs[i]) {
> + SNET_ERR(pdev, "Failed to allocate memory for SNET config\n");
> + snet_free_cfg(cfg);
> + return -ENOMEM;
> + }
> + /* Read device config */
> + cfg->devs[i]->virtio_id = psnet_read32(psnet, off);
> + off += 4;
> + cfg->devs[i]->vq_num = psnet_read32(psnet, off);
> + off += 4;
> + cfg->devs[i]->vq_size = psnet_read32(psnet, off);
> + off += 4;
> + cfg->devs[i]->features = psnet_read64(psnet, off);
> + off += 8;
> + cfg->devs[i]->snet_msg_off = psnet_read32(psnet, off);
> + off += 4;
> + /* Ignore Reserved */
> + off += sizeof(cfg->devs[i]->rsvd);
> +
> + cfg->devs[i]->cfg_size = psnet_read32(psnet, off);
> + off += 4;
> + cfg->devs[i]->cfg_off = psnet_read32(psnet, off);
> + off += 4;
> + /* Save VirtIO config address in BAR */
> + cfg->devs[i]->virtio_cfg = psnet->access_bar + off;
> + /* Move to next device, we don't want to load the VirtIO config */
> + off += cfg->devs[i]->cfg_size;
> + }
> + return 0;
> +}
> +
> +static int snet_alloc_irq_vector(struct pci_dev *pdev, struct psnet *psnet)
> +{
> + int ret = 0;
> +
> + ret = pci_alloc_irq_vectors(pdev, psnet->cfg.irqs_num, psnet->cfg.irqs_num, PCI_IRQ_MSIX);
> + if (ret <= 0) {
> + SNET_ERR(pdev, "Failed to allocate IRQ vectors\n");
> + return ret;
> + } else if (ret != psnet->cfg.irqs_num) {
> + SNET_WARN(pdev, "Allocated %u irq vectors instead of %u\n", ret,
> + psnet->cfg.irqs_num);
> + }
> + return 0;
> +}
> +
> +static void snet_free_vqs(struct snet *snet)
> +{
> + u32 i;
> +
> + if (!snet->vqs)
> + return;
> +
> + for (i = 0 ; i < snet->cfg->vq_num ; i++) {
> + if (!snet->vqs[i])
> + break;
> +
> + kfree(snet->vqs[i]);
> + }
> + kfree(snet->vqs);
> +}
> +
> +static int snet_build_vqs(struct pci_dev *pdev, struct snet *snet)
> +{
> + u32 i;
> + /* Allocate the array of VQ pointers */
> + snet->vqs = kcalloc(snet->cfg->vq_num, sizeof(void *), GFP_KERNEL);
> + if (!snet->vqs) {
> + SNET_ERR(pdev, "Failed to allocate memory for SNET VQs\n");
> + return -ENOMEM;
> + }
> +
> + /* Allocate VQs */
> + for (i = 0; i < snet->cfg->vq_num; i++) {
> + /* Allocate VQ memory */
> + snet->vqs[i] = kzalloc(sizeof(struct snet_vq), GFP_KERNEL);
> + if (!snet->vqs[i]) {
> + SNET_ERR(pdev, "Failed to allocate memory for a SNET VQ\n");
> + snet_free_vqs(snet);
> + return -ENOMEM;
> + }
> + }
> + return 0;
> +}
> +
> +static irqreturn_t snet_cfg_irq_hndlr(int irq, void *data)
> +{
> + struct snet *snet = data;
> + /* Call callback if any */
> + if (snet->cb.callback)
> + return snet->cb.callback(snet->cb.private);
> +
> + return IRQ_HANDLED;
> +}
> +
> +static irqreturn_t snet_vq_irq_hndlr(int irq, void *data)
> +{
> + struct snet_vq *vq = data;
> + /* Call callback if any */
> + if (vq->cb.callback)
> + return vq->cb.callback(vq->cb.private);
> +
> + return IRQ_HANDLED;
> +}
> +
> +static int snet_request_irqs(struct pci_dev *pdev, struct snet *snet)
> +{
> + int ret, i;
> + /* Request one IRQ for every VQ, and one for the config change */
> + /* Get next free IRQ ID */
> + snet->cfg_irq_idx = snet->psnet->next_irq++;
> + snprintf(snet->cfg_irq_name, SNET_IRQ_NAME_SIZE, "snet-cfg-%d", snet->cfg_irq_idx);
> +
> + /* Request config IRQ */
> + snet->cfg_irq = pci_irq_vector(pdev, snet->cfg_irq_idx);
> + ret = devm_request_irq(&pdev->dev, snet->cfg_irq, snet_cfg_irq_hndlr, 0,
> + snet->cfg_irq_name, snet);
> + if (ret) {
> + SNET_ERR(pdev, "Failed to request IRQ\n");
> + return ret;
> + }
> +
> + /* Request IRQ for every VQ */
> + for (i = 0; i < snet->cfg->vq_num; i++) {
> + /* Get next free IRQ ID */
> + snet->vqs[i]->irq_idx = snet->psnet->next_irq++;
> + /* Write IRQ name */
> + snprintf(snet->vqs[i]->irq_name, SNET_IRQ_NAME_SIZE, "snet[%u]-vq[%u]",
> + snet->sid, i);
> + /* Request IRQ */
> + snet->vqs[i]->irq = pci_irq_vector(pdev, snet->vqs[i]->irq_idx);
> + ret = devm_request_irq(&pdev->dev, snet->vqs[i]->irq, snet_vq_irq_hndlr, 0,
> + snet->vqs[i]->irq_name, snet->vqs[i]);
> + if (ret) {
> + SNET_ERR(pdev, "Failed to request IRQ\n");
> + /* Free all IRQs requested so far and exit */
> + devm_free_irq(&pdev->dev, snet->cfg_irq, snet);
> + i--;
> + for (; i != -1; i--)
> + devm_free_irq(&pdev->dev, snet->vqs[i]->irq, snet->vqs[i]);
I wonder if it's worth to bother consider we're using devres to manage irqs.
> + return ret;
> + }
> + }
> + return 0;
> +}
> +
> +static void free_snet_irqs(struct pci_dev *pdev, struct snet *snet)
> +{
> + u32 i;
> + /* Free config's IRQ */
> + devm_free_irq(&pdev->dev, snet->cfg_irq, snet);
> + /* Free VQ IRQs */
> + for (i = 0; i < snet->cfg->vq_num; i++)
> + devm_free_irq(&pdev->dev, snet->vqs[i]->irq, snet->vqs[i]);
> +}
> +
> +static int snet_alloc_dev(struct pci_dev *pdev, struct psnet *psnet, struct snet_dev_cfg *dev_cfg)
> +{
> + struct snet *snet;
> + int ret;
> + /* Allocate vdpa device */
> + snet = vdpa_alloc_device(struct snet, vdpa, &pdev->dev, &snet_config_ops, 1, 1, NULL,
> + false);
> + if (!snet) {
> + SNET_ERR(pdev, "Failed to allocate a vdpa device\n");
> + return -ENOMEM;
> + }
> + snet->psnet = psnet;
> + snet->cfg = dev_cfg;
> + snet->dpu_ready = false;
> + snet->sid = psnet->next_snet++;
> +
> + /* build snet VQs */
> + ret = snet_build_vqs(pdev, snet);
> + if (ret)
> + goto put_device;
> +
> + /* Request IRQs */
> + ret = snet_request_irqs(pdev, snet);
> + if (ret)
> + goto free_vqs;
> +
> + /*set DMA device*/
> + snet->vdpa.dma_dev = &pdev->dev;
> +
> + /* Register VDPA device */
> + ret = vdpa_register_device(&snet->vdpa, snet->cfg->vq_num);
> + if (ret) {
> + SNET_ERR(pdev, "Failed to register vdpa device\n");
> + goto free_irqs;
> + }
> +
> + /* Save snet in psnet */
> + psnet->devs[psnet->dev_num++] = snet;
> + return 0;
> +
> +free_irqs:
> + free_snet_irqs(pdev, snet);
> +free_vqs:
> + snet_free_vqs(snet);
> +put_device:
> + put_device(&snet->vdpa.dev);
> + return ret;
> +}
> +
> +/* Free a single snet device */
> +static void snet_free_dev(struct pci_dev *pdev, struct snet *snet)
> +{
> + vdpa_unregister_device(&snet->vdpa);
> + snet_free_vqs(snet);
> + free_snet_irqs(pdev, snet);
> +}
> +
> +/* Free all snet devices */
> +static void snet_free_devs(struct pci_dev *pdev, struct psnet *psnet)
> +{
> + u32 i;
> +
> + for (i = 0 ; i < psnet->total_devs ; i++) {
> + if (psnet->devs[i])
> + snet_free_dev(pdev, psnet->devs[i]);
> + else
> + break;
> + }
> + kfree(psnet->devs);
> +}
> +
> +static int snet_create_devs(struct pci_dev *pdev, struct psnet *psnet)
> +{
> + int ret, i;
> + /* Allocate memory to hold the device pointers */
> + psnet->total_devs = psnet->cfg.devices_num;
> + psnet->devs = kcalloc(psnet->total_devs, sizeof(void *), GFP_KERNEL);
> + if (!psnet->devs) {
> + SNET_ERR(pdev, "Failed to allocate memory for SNET devices\n");
> + return -ENOMEM;
> + }
> + /* init next address */
> + psnet->next_dpu_addr = psnet->cfg.dpu_off;
> + /* Give the device an unique ID */
> + psnet->id = psnet_next_uid++;
> + /* Allocate all devices */
> + for (i = 0; i < psnet->total_devs; i++) {
> + ret = snet_alloc_dev(pdev, psnet, psnet->cfg.devs[i]);
It looks to me all the devices created here use the same dma_dev (the
PCI device), I wonder how the DMA is isolated among the vDPA devices
created here.
Btw, the vDPA has been switched to use vDPA tool to create devices, it
is suggested to implement the mgmt devices as what other parents did.
Then the snet_alloc_dev() could be used for dev_add().
> + if (ret) {
> + snet_free_devs(pdev, psnet);
> + return ret;
> + }
> + }
> + SNET_INFO(pdev, "Created %u SNET devices\n", psnet->total_devs);
> + return 0;
> +}
> +
> +static int snet_vdpa_probe(struct pci_dev *pdev, const struct pci_device_id *id)
> +{
> + struct psnet *psnet;
> + int ret = 0;
> +
> + ret = pcim_enable_device(pdev);
> + if (ret) {
> + SNET_ERR(pdev, "Failed to enable PCI device\n");
> + return ret;
> + }
> +
> + psnet = kzalloc(sizeof(struct psnet), GFP_KERNEL);
> + if (!psnet) {
> + SNET_ERR(pdev, "Failed to allocate memory for PSNET device\n");
> + ret = -ENOMEM;
> + goto disable_dev;
> + }
> +
> + /* Open SNET BAR */
> + ret = snet_open_access_bar(pdev, psnet);
> + if (ret)
> + goto free_psnet;
> +
> + /* Try to read SNET's config from PCIe BAR */
> + ret = snet_read_cfg(pdev, psnet);
> + if (ret)
> + goto release_bar;
> +
> + /* Request for MSI-X IRQs */
> + ret = snet_alloc_irq_vector(pdev, psnet);
> + if (ret)
> + goto free_cfg;
> +
> + /* Create all devices from config */
> + ret = snet_create_devs(pdev, psnet);
> + if (ret)
> + goto free_irqs;
> +
There looks like a race, the vDPA device could be registered to the bus
and used by userspace by bus master/drvdata is set.
> + pci_set_master(pdev);
> + pci_set_drvdata(pdev, psnet);
> +
> + /* Create an HW monitor device */
> + if (SNET_CFG_FLAG(psnet, SNET_CFG_FLAG_HWMON))
> + snet_create_hwmon(pdev);
> +
> + return 0;
> +
> +free_irqs:
> + pci_free_irq_vectors(pdev);
> +free_cfg:
> + snet_free_cfg(&psnet->cfg);
> +release_bar:
> + pci_iounmap(pdev, psnet->access_bar);
> + pci_release_region(pdev, SNET_ACCESS_BAR);
> +free_psnet:
> + kfree(psnet);
> +disable_dev:
> + pci_disable_device(pdev);
I think devres should take care of this since we're using
pcim_enable_device()?
> + return ret;
> +}
> +
> +static void snet_vdpa_remove(struct pci_dev *pdev)
> +{
> + struct psnet *psnet = pci_get_drvdata(pdev);
> + struct snet *snet;
> + u32 i;
> +
> + /* Iterate through all SNET devices and send a "destroy" message */
> + for (i = 0; i < psnet->cfg.devices_num; i++) {
> + snet = psnet->devs[i];
> + snet_send_msg(snet, SNET_MSG_DESTROY);
> + /* Reset status */
> + snet->status = 0;
> + }
> + /* Free all SNET devices */
> + snet_free_devs(pdev, psnet);
> + /* Free IRQ vectors */
> + pci_free_irq_vectors(pdev);
> + /* Free SNET config read from DPU */
> + snet_free_cfg(&psnet->cfg);
> + /* Unmap and release PCI BAR */
> + pci_iounmap(pdev, psnet->access_bar);
> + pci_release_region(pdev, SNET_ACCESS_BAR);
> + /* Free PSNET device */
> + SNET_INFO(pdev, "Removed %u SNET devices\n", psnet->total_devs);
> + kfree(psnet);
> +}
> +
> +static struct pci_device_id snet_driver_pci_ids[] = {
> + { PCI_DEVICE_SUB(SOLIDRUN_PCI_VENDOR_ID, SNET_DEVICE_ID,
> + SOLIDRUN_PCI_VENDOR_ID, SNET_DEVICE_ID) },
> + { 0 },
> +};
> +
> +MODULE_DEVICE_TABLE(pci, snet_driver_pci_ids);
> +
> +static struct pci_driver snet_vdpa_driver = {
> + .name = "snet-vdpa-driver",
> + .id_table = snet_driver_pci_ids,
> + .probe = snet_vdpa_probe,
> + .remove = snet_vdpa_remove,
> +};
> +
> +module_pci_driver(snet_vdpa_driver);
> +
> +MODULE_AUTHOR("Alvaro Karsz <alvaro.karsz@solid-run.com>");
> +MODULE_DESCRIPTION("SolidRun vDPA driver");
> +MODULE_LICENSE("GPL v2");
> diff --git a/drivers/vdpa/solidrun/snet_vdpa.h b/drivers/vdpa/solidrun/snet_vdpa.h
> new file mode 100644
> index 00000000000..e44c900832a
> --- /dev/null
> +++ b/drivers/vdpa/solidrun/snet_vdpa.h
> @@ -0,0 +1,192 @@
> +/* SPDX-License-Identifier: GPL-2.0-only */
> +/*
> + * SolidRun DPU driver for control plane
> + *
> + * Copyright (C) 2022 SolidRun
> + *
> + * Author: Alvaro Karsz <alvaro.karsz@solid-run.com>
> + *
> + */
> +#ifndef _SNET_VDPA_H_
> +#define _SNET_VDPA_H_
> +
> +#include <linux/vdpa.h>
> +#include <linux/pci.h>
> +
> +#define SNET_IRQ_NAME_SIZE 20
> +#define SNET_HWMON_NAME_SIZE 20
> +
> +#define SNET_ERR(pdev, fmt, ...) dev_err(&pdev->dev, "%s"fmt, "snet_vdpa: ", ##__VA_ARGS__)
> +#define SNET_WARN(pdev, fmt, ...) dev_warn(&pdev->dev, "%s"fmt, "snet_vdpa: ", ##__VA_ARGS__)
> +#define SNET_INFO(pdev, fmt, ...) dev_info(&pdev->dev, "%s"fmt, "snet_vdpa: ", ##__VA_ARGS__)
> +#define SNET_DBG(pdev, fmt, ...) dev_dbg(&pdev->dev, "%s"fmt, "snet_vdpa: ", ##__VA_ARGS__)
> +
> +/* VQ struct */
> +struct snet_vq {
> + /* VQ callback */
> + struct vdpa_callback cb;
> + /* desc base address */
> + u64 desc_area;
> + /* device base address */
> + u64 device_area;
> + /* driver base address */
> + u64 driver_area;
> + /* Queue size */
> + u32 num;
> + /* Kick address offset for this VQ */
> + u64 kick_offset;
> + /* Serial ID for VQ */
> + u32 sid;
> + /* is ready flag */
> + bool ready;
> + /* IRQ number */
> + u32 irq;
> + /* IRQ index, DPU uses this to parse data from MSI-X table */
> + u32 irq_idx;
> + /* IRQ name */
> + char irq_name[SNET_IRQ_NAME_SIZE];
> + /* pointer to mapped PCI BAR register used by this VQ to kick */
> + void __iomem *kick_ptr;
> +};
> +
> +struct snet {
> + /* vdpa device */
> + struct vdpa_device vdpa;
> + /* Config callback */
> + struct vdpa_callback cb;
> + /* array of virqueues */
> + struct snet_vq **vqs;
> + /* Used features */
> + u64 used_features;
According to the code, this seems to be the driver features and
> + /* driver features */
> + u64 drv_features;
This seems to be unused.
Thanks
> + /* device generation */
> + u32 generation;
> + /* Device serial ID */
> + u32 sid;
> + /* device status */
> + u8 status;
> + /* boolean indicating if VQ data was dumped to DPU */
> + bool dpu_ready;
> + /* IRQ number */
> + u32 cfg_irq;
> + /* IRQ index, DPU uses this to parse data from MSI-X table */
> + u32 cfg_irq_idx;
> + /* IRQ name */
> + char cfg_irq_name[SNET_IRQ_NAME_SIZE];
> + /* Pointer to snet pdev parent device */
> + struct psnet *psnet;
> + /* Pointer to snet config device */
> + struct snet_dev_cfg *cfg;
> +};
> +
> +struct snet_dev_cfg {
> + /* Device ID following VirtIO spec. */
> + u32 virtio_id;
> + /* Number of VQs for this device (NOT PAIRS!) */
> + u32 vq_num;
> + /* Size of every VQ */
> + u32 vq_size;
> + /* Device features, following VirtIO spec */
> + u64 features;
> + /* Offset to write messages to the DPU SNET_MSG_* */
> + u32 snet_msg_off;
> + /* Reserved for future usage */
> + u32 rsvd[6];
> + /* VirtIO device specific config size */
> + u32 cfg_size;
> + /* VirtIO device specific config offset in BAR */
> + u32 cfg_off;
> + /* VirtIO device specific config address */
> + void __iomem *virtio_cfg;
> +} __packed;
> +
> +struct snet_cfg {
> + /* Magic key */
> + u32 key;
> + /* Size of total config in bytes */
> + u32 cfg_size;
> + /* Config version */
> + u32 cfg_ver;
> + /* Number of IRQs to be asked */
> + u32 irqs_num;
> + /* Offset in PCIe BAR for writing device data to DPU */
> + u32 dpu_off;
> + /* Offset in PCIe BAR for VQ kicks */
> + u32 kick_off;
> + /* Offset in PCIe BAR for HW monitoring */
> + u32 hwmon_off;
> + /* Config general flags - enum snet_cfg_flags */
> + u32 flags;
> + /* Reserved for future usage */
> + u32 rsvd[6];
> + /* Number of snet devices */
> + u32 devices_num;
> + /* The actual devices */
> + struct snet_dev_cfg **devs;
> +} __packed;
> +
> +/* SolidNET PCIe device, one device per PCIe physical function */
> +struct psnet {
> + /* Unique snet pdev id */
> + u32 id;
> + /* Number of snet devices */
> + u32 dev_num;
> + /* Control IRQ IDs */
> + u32 next_irq;
> + /* Control device IDs */
> + u32 next_snet;
> + /* Control VQ IDs */
> + u32 next_vq;
> + /* Total number of devices (including future devices) */
> + u32 total_devs;
> + /* Negotiated config version */
> + u32 negotiated_cfg_ver;
> + /* Next address to write snet info to DPU */
> + u64 next_dpu_addr;
> + /* devices list */
> + struct snet **devs;
> + /* pointer to mapped PCI BAR */
> + void __iomem *access_bar;
> + /* Pointer to the device's config read from BAR */
> + struct snet_cfg cfg;
> + /* Name of monitor device */
> + char hwmon_name[SNET_HWMON_NAME_SIZE];
> +};
> +
> +enum snet_cfg_flags {
> + /* Create a HWMON device */
> + SNET_CFG_FLAG_HWMON = BIT(0),
> +};
> +
> +/* Check if a config flag is set */
> +#define SNET_CFG_FLAG(p, f) ((p)->cfg.flags & (f))
> +
> +static inline u32 psnet_read32(struct psnet *psnet, u32 off)
> +{
> + return ioread32(psnet->access_bar + off);
> +}
> +
> +static inline void psnet_write32(struct psnet *psnet, u32 off, u32 val)
> +{
> + iowrite32(val, psnet->access_bar + off);
> +}
> +
> +static inline u64 psnet_read64(struct psnet *psnet, u32 off)
> +{
> + u64 val;
> +
> + val = (u64)ioread32(psnet->access_bar + off);
> + val |= ((u64)ioread32(psnet->access_bar + off + 4) << 32);
> + return val;
> +}
> +
> +static inline void psnet_write64(struct psnet *psnet, u32 off, u64 val)
> +{
> + psnet_write32(psnet, off, val >> 32);
> + psnet_write32(psnet, off + 4, val & 0xFFFFFFFF);
> +}
> +
> +void snet_create_hwmon(struct pci_dev *pdev);
> +
> +#endif //_SNET_VDPA_H_
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] virtio: vdpa: new SolidNET DPU driver.
2022-11-09 6:47 ` Jason Wang
@ 2022-11-09 9:44 ` Alvaro Karsz
2022-11-10 1:54 ` Jason Wang
0 siblings, 1 reply; 5+ messages in thread
From: Alvaro Karsz @ 2022-11-09 9:44 UTC (permalink / raw)
To: Jason Wang
Cc: Jean Delvare, Michael S. Tsirkin, Guenter Roeck, virtualization
Hi Jason,
> If I was not wrong, this value seems doesn't change. If yes, I wonder we
> can move the kick_offset logic to snet_alloc_dev()?
There is not a real reason to have this logic in snet_set_vq_address,
so it could be moved to snet_build_vqs (which is called within
snet_alloc_dev).
> -EOPNOTSUPP?
Returning an error from the set_vq_state callback leads to probe failure.
This code is taken from drivers/virtio/virtio_vdpa.c,
virtio_vdpa_setup_vq function:
> struct vdpa_vq_state state = {0};
>
> ......
>
> err = ops->set_vq_state(vdpa, index, &state);
> if (err)
> goto err_vq;
I could check struct vdpa_vq_state, and return 0 if struct
vdpa_vq_state value is 0, -EOPNOTSUPP otherwise.
What do you think?
> I fail to understand how can this work. E.g after reset there should be
> no interaction from the device like DMA, otherwise there could have
> security implications.
You're right, I'll add a proper reset callback.
> Since read is ordered with write, a more easy way is to perform a ioread
> here.
> Interesting, which barrier is this paired?
Usually reads are slow, but we don't care about speed when sending
messages (since we only send a "destroy" message so far, meaning that
the pci remove callback was called), so the memory barrier can be
replaced with a read operation.
>
> Do we need to do endian conversion here (cpu_to_le64())?
Yes, I'll add it.
>
> Need to take endianess into account.
I'm not sure about that.
The endianness appears to be handled by the caller.
Function from include/linux/virtio_config.h
> static inline void virtio_cwrite16(struct virtio_device *vdev,
> unsigned int offset, u16 val)
> {
> __virtio16 v;
>
>
> might_sleep();
> v = cpu_to_virtio16(vdev, val);
> vdev->config->set(vdev, offset, &v, sizeof(v));
> }
> static inline void virtio_cwrite32(struct virtio_device *vdev,
> unsigned int offset, u32 val)
> {
> __virtio32 v;
>
>
> might_sleep();
> v = cpu_to_virtio32(vdev, val);
> vdev->config->set(vdev, offset, &v, sizeof(v));
> }
>
> static inline void virtio_cwrite64(struct virtio_device *vdev,
> unsigned int offset, u64 val)
> {
> __virtio64 v;
>
>
> might_sleep();
> v = cpu_to_virtio64(vdev, val);
> vdev->config->set(vdev, offset, &v, sizeof(v));
> }
I'm not sure how the endianness can be handled by the vDPA driver.
This function may be called for a 8bit, 16bit, 32bit or 64bit variables.
It theoretically may be called to change multiple variables at once.
It may be called to change part of a variable.
> If I was not wrong, the device depends on the platform IOMMU to work. So
> unless device have a more strict iova range than what platform IOMMU can
> provide, we can simply not advertise this and use the one that is
> provided by the IOMMU:
>
>
> static void vhost_vdpa_set_iova_range(struct vhost_vdpa *v)
> {
> struct vdpa_iova_range *range = &v->range;
> struct vdpa_device *vdpa = v->vdpa;
> const struct vdpa_config_ops *ops = vdpa->config;
>
>
> if (ops->get_iova_range) {
> *range = ops->get_iova_range(vdpa);
> } else if (v->domain && v->domain->geometry.force_aperture) {
> range->first = v->domain->geometry.aperture_start;
> range->last = v->domain->geometry.aperture_end;
> } else {
> range->first = 0;
> range->last = ULLONG_MAX;
> }
> }
I'll delete the snet_get_iova_range function.
> Any chance to use device managed region helper here? It seems to
> simplify the codes (e.g the cleanup stuffs).
Ok, I'll do it.
> Is this better to say "config is not ready"? Btw, I wonder if it makes
> more sense to wait until the confg is ready with a timeout?
Good idea, I'll implement the wait & timeout.
> I wonder if it's worth to bother consider we're using devres to manage irqs.
You're right, this isn't required, I'll remove it.
>
> It looks to me all the devices created here use the same dma_dev (the
> PCI device), I wonder how the DMA is isolated among the vDPA devices
> created here.
All vDPA devices indeed use the same DMA device, there is no isolation
between the devices.
I'm not sure why there should be isolation in this case.
> Btw, the vDPA has been switched to use vDPA tool to create devices, it
> is suggested to implement the mgmt devices as what other parents did.
> Then the snet_alloc_dev() could be used for dev_add().
We want to leave control to the DPU at the moment, the number/type of
devices is determined by the DPU's config, and can't be controlled
from userspace.
> There looks like a race, the vDPA device could be registered to the bus
> and used by userspace by bus master/drvdata is set.
You're right, the vdpa registration should be done after the
master/drvdata is set.
> I think devres should take care of this since we're using
> pcim_enable_device()?
You're right, this isn't required, I'll remove it.
> According to the code, this seems to be the driver features and
These are the negotiated features
We're not keeping record of the driver features, when
set_driver_features is called, we just logic AND the driver features
with the supported features received from the DPU.
I'll rename it to be 'negotiated_features", this seems more accurate.
> static int snet_set_drv_features(struct vdpa_device *vdev, u64 features)
> {
> struct snet *snet = vdpa_to_snet(vdev);
>
>
> snet->used_features = snet->cfg->features & features;
> return 0;
> }
> This seems to be unused.
You're right, I'll remove it.
Thank you for your comments.
I'll send a new version once I finish working on the comments.
Alvaro
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] virtio: vdpa: new SolidNET DPU driver.
2022-11-09 9:44 ` Alvaro Karsz
@ 2022-11-10 1:54 ` Jason Wang
2022-11-13 12:24 ` Alvaro Karsz
0 siblings, 1 reply; 5+ messages in thread
From: Jason Wang @ 2022-11-10 1:54 UTC (permalink / raw)
To: Alvaro Karsz
Cc: Jean Delvare, Michael S. Tsirkin, Guenter Roeck, virtualization
在 2022/11/9 17:44, Alvaro Karsz 写道:
> Hi Jason,
>
>> If I was not wrong, this value seems doesn't change. If yes, I wonder we
>> can move the kick_offset logic to snet_alloc_dev()?
>
> There is not a real reason to have this logic in snet_set_vq_address,
> so it could be moved to snet_build_vqs (which is called within
> snet_alloc_dev).
>
>> -EOPNOTSUPP?
>
> Returning an error from the set_vq_state callback leads to probe failure.
> This code is taken from drivers/virtio/virtio_vdpa.c,
> virtio_vdpa_setup_vq function:
>
>> struct vdpa_vq_state state = {0};
>>
>> ......
>>
>> err = ops->set_vq_state(vdpa, index, &state);
>> if (err)
>> goto err_vq;
>
> I could check struct vdpa_vq_state, and return 0 if struct
> vdpa_vq_state value is 0, -EOPNOTSUPP otherwise.
> What do you think?
That would be fine. (It could be something similar to
vp_vdpa_set_vq_state_split()).
>
>> I fail to understand how can this work. E.g after reset there should be
>> no interaction from the device like DMA, otherwise there could have
>> security implications.
>
> You're right, I'll add a proper reset callback.
>
>> Since read is ordered with write, a more easy way is to perform a ioread
>> here.
>> Interesting, which barrier is this paired?
>
> Usually reads are slow, but we don't care about speed when sending
> messages (since we only send a "destroy" message so far, meaning that
> the pci remove callback was called), so the memory barrier can be
> replaced with a read operation.
Yes, actually, the memory barrier can't guarantee the write has been
processed by the device.
>
>> Do we need to do endian conversion here (cpu_to_le64())?
>
> Yes, I'll add it.
>
>> Need to take endianess into account.
>
> I'm not sure about that.
> The endianness appears to be handled by the caller.
> Function from include/linux/virtio_config.h
>
>> static inline void virtio_cwrite16(struct virtio_device *vdev,
>> unsigned int offset, u16 val)
>> {
>> __virtio16 v;
>>
>>
>> might_sleep();
>> v = cpu_to_virtio16(vdev, val);
>> vdev->config->set(vdev, offset, &v, sizeof(v));
>> }
You are right.
>
>> static inline void virtio_cwrite32(struct virtio_device *vdev,
>> unsigned int offset, u32 val)
>> {
>> __virtio32 v;
>>
>>
>> might_sleep();
>> v = cpu_to_virtio32(vdev, val);
>> vdev->config->set(vdev, offset, &v, sizeof(v));
>> }
>>
>
>> static inline void virtio_cwrite64(struct virtio_device *vdev,
>> unsigned int offset, u64 val)
>> {
>> __virtio64 v;
>>
>>
>> might_sleep();
>> v = cpu_to_virtio64(vdev, val);
>> vdev->config->set(vdev, offset, &v, sizeof(v));
>> }
>
> I'm not sure how the endianness can be handled by the vDPA driver.
> This function may be called for a 8bit, 16bit, 32bit or 64bit variables.
> It theoretically may be called to change multiple variables at once.
> It may be called to change part of a variable.
>
>
>> If I was not wrong, the device depends on the platform IOMMU to work. So
>> unless device have a more strict iova range than what platform IOMMU can
>> provide, we can simply not advertise this and use the one that is
>> provided by the IOMMU:
>>
>>
>> static void vhost_vdpa_set_iova_range(struct vhost_vdpa *v)
>> {
>> struct vdpa_iova_range *range = &v->range;
>> struct vdpa_device *vdpa = v->vdpa;
>> const struct vdpa_config_ops *ops = vdpa->config;
>>
>>
>> if (ops->get_iova_range) {
>> *range = ops->get_iova_range(vdpa);
>> } else if (v->domain && v->domain->geometry.force_aperture) {
>> range->first = v->domain->geometry.aperture_start;
>> range->last = v->domain->geometry.aperture_end;
>> } else {
>> range->first = 0;
>> range->last = ULLONG_MAX;
>> }
>> }
>
> I'll delete the snet_get_iova_range function.
>
>> Any chance to use device managed region helper here? It seems to
>> simplify the codes (e.g the cleanup stuffs).
>
> Ok, I'll do it.
>
>> Is this better to say "config is not ready"? Btw, I wonder if it makes
>> more sense to wait until the confg is ready with a timeout?
>
> Good idea, I'll implement the wait & timeout.
>
>> I wonder if it's worth to bother consider we're using devres to manage irqs.
>
> You're right, this isn't required, I'll remove it.
>
>> It looks to me all the devices created here use the same dma_dev (the
>> PCI device), I wonder how the DMA is isolated among the vDPA devices
>> created here.
>
> All vDPA devices indeed use the same DMA device, there is no isolation
> between the devices.
> I'm not sure why there should be isolation in this case.
Each vDPA needs to be able to assigned to a userspace(VM) independently
when bound to vhost-vDPA. If they share the same IOMMU domain, there
will be security issues.
>
>> Btw, the vDPA has been switched to use vDPA tool to create devices, it
>> is suggested to implement the mgmt devices as what other parents did.
>> Then the snet_alloc_dev() could be used for dev_add().
>
> We want to leave control to the DPU at the moment, the number/type of
> devices is determined by the DPU's config, and can't be controlled
> from userspace.
That's fine, it might be better to state this in the changelog or as a
comment somewhere in the code.
>
>> There looks like a race, the vDPA device could be registered to the bus
>> and used by userspace by bus master/drvdata is set.
>
> You're right, the vdpa registration should be done after the
> master/drvdata is set.
>
>> I think devres should take care of this since we're using
>> pcim_enable_device()?
>
> You're right, this isn't required, I'll remove it.
>
>> According to the code, this seems to be the driver features and
>
> These are the negotiated features
> We're not keeping record of the driver features, when
> set_driver_features is called, we just logic AND the driver features
> with the supported features received from the DPU.
> I'll rename it to be 'negotiated_features", this seems more accurate.
Ok.
>
>> static int snet_set_drv_features(struct vdpa_device *vdev, u64 features)
>> {
>> struct snet *snet = vdpa_to_snet(vdev);
>>
>>
>> snet->used_features = snet->cfg->features & features;
>> return 0;
>> }
>
>
>> This seems to be unused.
>
> You're right, I'll remove it.
>
>
> Thank you for your comments.
> I'll send a new version once I finish working on the comments.
>
> Alvaro
Thanks
>
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] virtio: vdpa: new SolidNET DPU driver.
2022-11-10 1:54 ` Jason Wang
@ 2022-11-13 12:24 ` Alvaro Karsz
0 siblings, 0 replies; 5+ messages in thread
From: Alvaro Karsz @ 2022-11-13 12:24 UTC (permalink / raw)
To: Jason Wang
Cc: Jean Delvare, Michael S. Tsirkin, Guenter Roeck, virtualization
> Yes, actually, the memory barrier can't guarantee the write has been
> processed by the device.
Yes, of course, I meant that we need to wait until the message is
sent, not until the DPU sees the message.
> Each vDPA needs to be able to assigned to a userspace(VM) independently
> when bound to vhost-vDPA. If they share the same IOMMU domain, there
> will be security issues.
We were planning on using the SR-IOV feature for isolation in the future.
We will add the feature and create a new version.
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2022-11-13 12:24 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-11-08 10:09 [PATCH] virtio: vdpa: new SolidNET DPU driver Alvaro Karsz
2022-11-09 6:47 ` Jason Wang
2022-11-09 9:44 ` Alvaro Karsz
2022-11-10 1:54 ` Jason Wang
2022-11-13 12:24 ` Alvaro Karsz
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.