virtio: vdpa: vDPA driver for Marvell OCTEON DPU devices
authorSrujana Challa <schalla@marvell.com>
Fri, 14 Jun 2024 14:46:59 +0000 (20:16 +0530)
committerMichael S. Tsirkin <mst@redhat.com>
Tue, 9 Jul 2024 12:42:39 +0000 (08:42 -0400)
This commit introduces a new vDPA driver specifically designed for
managing the virtio control plane over the vDPA bus for OCTEON DPU
devices. The driver consists of two layers:

1. Octep HW Layer (Octeon Endpoint): Responsible for handling hardware
operations and configurations related to the DPU device.

2. Octep Main Layer: Compliant with the vDPA bus framework, this layer
implements device operations for the vDPA bus. It handles device
probing, bus attachment, vring operations, and other relevant tasks.

Signed-off-by: Srujana Challa <schalla@marvell.com>
Signed-off-by: Vamsi Attunuru <vattunuru@marvell.com>
Signed-off-by: Shijith Thotton <sthotton@marvell.com>
Signed-off-by: Nithin Dabilpuram <ndabilpuram@marvell.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20240614144659.1776067-1-schalla@marvell.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
MAINTAINERS
drivers/vdpa/Kconfig
drivers/vdpa/Makefile
drivers/vdpa/octeon_ep/Makefile [new file with mode: 0644]
drivers/vdpa/octeon_ep/octep_vdpa.h [new file with mode: 0644]
drivers/vdpa/octeon_ep/octep_vdpa_hw.c [new file with mode: 0644]
drivers/vdpa/octeon_ep/octep_vdpa_main.c [new file with mode: 0644]

index cd2ca0c3158e068c5e789258e29b17271bbc7464..d0249d7538e13b13e2ba5a21e528bf6f67dbe413 100644 (file)
@@ -13459,6 +13459,11 @@ S:     Supported
 F:     Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.yaml
 F:     drivers/mmc/host/sdhci-xenon*
 
+MARVELL OCTEON ENDPOINT VIRTIO DATA PATH ACCELERATOR
+R:     schalla@marvell.com
+R:     vattunuru@marvell.com
+F:     drivers/vdpa/octeon_ep/
+
 MATROX FRAMEBUFFER DRIVER
 L:     linux-fbdev@vger.kernel.org
 S:     Orphan
index 656c1cb541deb7c51c31b5304ff5f5d5b3395da7..5265d09fc1c40979a49b9d7f9a3fbfcabe87726c 100644 (file)
@@ -126,4 +126,15 @@ config PDS_VDPA
          With this driver, the VirtIO dataplane can be
          offloaded to an AMD/Pensando DSC device.
 
+config OCTEONEP_VDPA
+       tristate "vDPA driver for Octeon DPU devices"
+       depends on m
+       depends on PCI_MSI
+       help
+         This is a vDPA driver designed for Marvell's Octeon DPU devices.
+         This driver enables the offloading of the VirtIO dataplane to an
+         Octeon DPU device.
+         Please note that this driver must be built as a module and it
+         cannot be loaded until the Octeon emulation software is running.
+
 endif # VDPA
index 8f53c6f3cca7c5e5dacc726f2d8f9d94390e04bb..5654d36707af23ac66e2c904e5bbb319a9dfde8e 100644 (file)
@@ -8,3 +8,4 @@ obj-$(CONFIG_VP_VDPA)    += virtio_pci/
 obj-$(CONFIG_ALIBABA_ENI_VDPA) += alibaba/
 obj-$(CONFIG_SNET_VDPA) += solidrun/
 obj-$(CONFIG_PDS_VDPA) += pds/
+obj-$(CONFIG_OCTEONEP_VDPA) += octeon_ep/
diff --git a/drivers/vdpa/octeon_ep/Makefile b/drivers/vdpa/octeon_ep/Makefile
new file mode 100644 (file)
index 0000000..e23e2ff
--- /dev/null
@@ -0,0 +1,4 @@
+# SPDX-License-Identifier: GPL-2.0
+obj-$(CONFIG_OCTEONEP_VDPA) += octep_vdpa.o
+octep_vdpa-$(CONFIG_OCTEONEP_VDPA) += octep_vdpa_main.o
+octep_vdpa-$(CONFIG_OCTEONEP_VDPA) += octep_vdpa_hw.o
diff --git a/drivers/vdpa/octeon_ep/octep_vdpa.h b/drivers/vdpa/octeon_ep/octep_vdpa.h
new file mode 100644 (file)
index 0000000..046710e
--- /dev/null
@@ -0,0 +1,94 @@
+/* SPDX-License-Identifier: GPL-2.0-only
+ * Copyright (C) 2024 Marvell.
+ */
+#ifndef __OCTEP_VDPA_H__
+#define __OCTEP_VDPA_H__
+
+#include <linux/pci.h>
+#include <linux/pci_regs.h>
+#include <linux/vdpa.h>
+#include <linux/virtio_pci_modern.h>
+#include <uapi/linux/virtio_net.h>
+#include <uapi/linux/virtio_blk.h>
+#include <uapi/linux/virtio_config.h>
+#include <uapi/linux/virtio_pci.h>
+#include <uapi/linux/vdpa.h>
+
+#define OCTEP_VDPA_DEVID_CN106K_PF 0xb900
+#define OCTEP_VDPA_DEVID_CN106K_VF 0xb903
+#define OCTEP_VDPA_DEVID_CN105K_PF 0xba00
+#define OCTEP_VDPA_DEVID_CN105K_VF 0xba03
+#define OCTEP_VDPA_DEVID_CN103K_PF 0xbd00
+#define OCTEP_VDPA_DEVID_CN103K_VF 0xbd03
+
+#define OCTEP_HW_MBOX_BAR 0
+#define OCTEP_HW_CAPS_BAR 4
+
+#define OCTEP_DEV_READY_SIGNATURE 0xBABABABA
+
+#define OCTEP_EPF_RINFO(x) (0x000209f0 | ((x) << 25))
+#define OCTEP_VF_MBOX_DATA(x) (0x00010210 | ((x) << 17))
+#define OCTEP_PF_MBOX_DATA(x) (0x00022000 | ((x) << 4))
+
+#define OCTEP_EPF_RINFO_RPVF(val) (((val) >> 32) & 0xF)
+#define OCTEP_EPF_RINFO_NVFS(val) (((val) >> 48) & 0x7F)
+
+#define OCTEP_FW_READY_SIGNATURE0  0xFEEDFEED
+#define OCTEP_FW_READY_SIGNATURE1  0x3355ffaa
+
+enum octep_vdpa_dev_status {
+       OCTEP_VDPA_DEV_STATUS_INVALID,
+       OCTEP_VDPA_DEV_STATUS_ALLOC,
+       OCTEP_VDPA_DEV_STATUS_WAIT_FOR_BAR_INIT,
+       OCTEP_VDPA_DEV_STATUS_INIT,
+       OCTEP_VDPA_DEV_STATUS_READY,
+       OCTEP_VDPA_DEV_STATUS_UNINIT
+};
+
+struct octep_vring_info {
+       struct vdpa_callback cb;
+       void __iomem *notify_addr;
+       u32 __iomem *cb_notify_addr;
+       phys_addr_t notify_pa;
+       char msix_name[256];
+};
+
+struct octep_hw {
+       struct pci_dev *pdev;
+       u8 __iomem *base[PCI_STD_NUM_BARS];
+       struct virtio_pci_common_cfg __iomem *common_cfg;
+       u8 __iomem *dev_cfg;
+       u8 __iomem *isr;
+       void __iomem *notify_base;
+       phys_addr_t notify_base_pa;
+       u32 notify_off_multiplier;
+       u8 notify_bar;
+       struct octep_vring_info *vqs;
+       struct vdpa_callback config_cb;
+       u64 features;
+       u16 nr_vring;
+       u32 config_size;
+       int irq;
+};
+
+u8 octep_hw_get_status(struct octep_hw *oct_hw);
+void octep_hw_set_status(struct octep_hw *dev, uint8_t status);
+void octep_hw_reset(struct octep_hw *oct_hw);
+void octep_write_queue_select(struct octep_hw *oct_hw, u16 queue_id);
+void octep_notify_queue(struct octep_hw *oct_hw, u16 qid);
+void octep_read_dev_config(struct octep_hw *oct_hw, u64 offset, void *dst, int length);
+int octep_set_vq_address(struct octep_hw *oct_hw, u16 qid, u64 desc_area, u64 driver_area,
+                        u64 device_area);
+void octep_set_vq_num(struct octep_hw *oct_hw, u16 qid, u32 num);
+void octep_set_vq_ready(struct octep_hw *oct_hw, u16 qid, bool ready);
+bool octep_get_vq_ready(struct octep_hw *oct_hw, u16 qid);
+int octep_set_vq_state(struct octep_hw *oct_hw, u16 qid, const struct vdpa_vq_state *state);
+int octep_get_vq_state(struct octep_hw *oct_hw, u16 qid, struct vdpa_vq_state *state);
+u16 octep_get_vq_size(struct octep_hw *oct_hw);
+int octep_hw_caps_read(struct octep_hw *oct_hw, struct pci_dev *pdev);
+u64 octep_hw_get_dev_features(struct octep_hw *oct_hw);
+void octep_hw_set_drv_features(struct octep_hw *oct_hw, u64 features);
+u64 octep_hw_get_drv_features(struct octep_hw *oct_hw);
+int octep_verify_features(u64 features);
+
+#endif /* __OCTEP_VDPA_H__ */
diff --git a/drivers/vdpa/octeon_ep/octep_vdpa_hw.c b/drivers/vdpa/octeon_ep/octep_vdpa_hw.c
new file mode 100644 (file)
index 0000000..7fa0491
--- /dev/null
@@ -0,0 +1,517 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Copyright (C) 2024 Marvell. */
+
+#include <linux/iopoll.h>
+
+#include "octep_vdpa.h"
+
+enum octep_mbox_ids {
+       OCTEP_MBOX_MSG_SET_VQ_STATE = 1,
+       OCTEP_MBOX_MSG_GET_VQ_STATE,
+};
+
+#define OCTEP_HW_TIMEOUT       10000000
+
+#define MBOX_OFFSET            64
+#define MBOX_RSP_MASK          0x00000001
+#define MBOX_RC_MASK           0x0000FFFE
+
+#define MBOX_RSP_TO_ERR(val)   (-(((val) & MBOX_RC_MASK) >> 2))
+#define MBOX_AVAIL(val)        (((val) & MBOX_RSP_MASK))
+#define MBOX_RSP(val)          ((val) & (MBOX_RC_MASK | MBOX_RSP_MASK))
+
+#define DEV_RST_ACK_BIT        7
+#define FEATURE_SEL_ACK_BIT    15
+#define QUEUE_SEL_ACK_BIT      15
+
+struct octep_mbox_hdr {
+       u8 ver;
+       u8 rsvd1;
+       u16 id;
+       u16 rsvd2;
+#define MBOX_REQ_SIG (0xdead)
+#define MBOX_RSP_SIG (0xbeef)
+       u16 sig;
+};
+
+struct octep_mbox_sts {
+       u16 rsp:1;
+       u16 rc:15;
+       u16 rsvd;
+};
+
+struct octep_mbox {
+       struct octep_mbox_hdr hdr;
+       struct octep_mbox_sts sts;
+       u64 rsvd;
+       u32 data[];
+};
+
+static inline struct octep_mbox __iomem *octep_get_mbox(struct octep_hw *oct_hw)
+{
+       return (struct octep_mbox __iomem *)(oct_hw->dev_cfg + MBOX_OFFSET);
+}
+
+static inline int octep_wait_for_mbox_avail(struct octep_mbox __iomem *mbox)
+{
+       u32 val;
+
+       return readx_poll_timeout(ioread32, &mbox->sts, val, MBOX_AVAIL(val), 10,
+                                 OCTEP_HW_TIMEOUT);
+}
+
+static inline int octep_wait_for_mbox_rsp(struct octep_mbox __iomem *mbox)
+{
+       u32 val;
+
+       return readx_poll_timeout(ioread32, &mbox->sts, val, MBOX_RSP(val), 10,
+                                 OCTEP_HW_TIMEOUT);
+}
+
+static inline void octep_write_hdr(struct octep_mbox __iomem *mbox, u16 id, u16 sig)
+{
+       iowrite16(id, &mbox->hdr.id);
+       iowrite16(sig, &mbox->hdr.sig);
+}
+
+static inline u32 octep_read_sig(struct octep_mbox __iomem *mbox)
+{
+       return ioread16(&mbox->hdr.sig);
+}
+
+static inline void octep_write_sts(struct octep_mbox __iomem *mbox, u32 sts)
+{
+       iowrite32(sts, &mbox->sts);
+}
+
+static inline u32 octep_read_sts(struct octep_mbox __iomem *mbox)
+{
+       return ioread32(&mbox->sts);
+}
+
+static inline u32 octep_read32_word(struct octep_mbox __iomem *mbox, u16 word_idx)
+{
+       return ioread32(&mbox->data[word_idx]);
+}
+
+static inline void octep_write32_word(struct octep_mbox __iomem *mbox, u16 word_idx, u32 word)
+{
+       return iowrite32(word, &mbox->data[word_idx]);
+}
+
+static int octep_process_mbox(struct octep_hw *oct_hw, u16 id, u16 qid, void *buffer,
+                             u32 buf_size, bool write)
+{
+       struct octep_mbox __iomem *mbox = octep_get_mbox(oct_hw);
+       struct pci_dev *pdev = oct_hw->pdev;
+       u32 *p = (u32 *)buffer;
+       u16 data_wds;
+       int ret, i;
+       u32 val;
+
+       if (!IS_ALIGNED(buf_size, 4))
+               return -EINVAL;
+
+       /* Make sure mbox space is available */
+       ret = octep_wait_for_mbox_avail(mbox);
+       if (ret) {
+               dev_warn(&pdev->dev, "Timeout waiting for previous mbox data to be consumed\n");
+               return ret;
+       }
+       data_wds = buf_size / 4;
+
+       if (write) {
+               for (i = 1; i <= data_wds; i++) {
+                       octep_write32_word(mbox, i, *p);
+                       p++;
+               }
+       }
+       octep_write32_word(mbox, 0, (u32)qid);
+       octep_write_sts(mbox, 0);
+
+       octep_write_hdr(mbox, id, MBOX_REQ_SIG);
+
+       ret = octep_wait_for_mbox_rsp(mbox);
+       if (ret) {
+               dev_warn(&pdev->dev, "Timeout waiting for mbox : %d response\n", id);
+               return ret;
+       }
+
+       val = octep_read_sig(mbox);
+       if ((val & 0xFFFF) != MBOX_RSP_SIG) {
+               dev_warn(&pdev->dev, "Invalid Signature from mbox : %d response\n", id);
+               return ret;
+       }
+
+       val = octep_read_sts(mbox);
+       if (val & MBOX_RC_MASK) {
+               ret = MBOX_RSP_TO_ERR(val);
+               dev_warn(&pdev->dev, "Error while processing mbox : %d, err %d\n", id, ret);
+               return ret;
+       }
+
+       if (!write)
+               for (i = 1; i <= data_wds; i++)
+                       *p++ = octep_read32_word(mbox, i);
+
+       return 0;
+}
+
+static void octep_mbox_init(struct octep_mbox __iomem *mbox)
+{
+       iowrite32(1, &mbox->sts);
+}
+
+int octep_verify_features(u64 features)
+{
+       /* Minimum features to expect */
+       if (!(features & BIT_ULL(VIRTIO_F_VERSION_1)))
+               return -EOPNOTSUPP;
+
+       if (!(features & BIT_ULL(VIRTIO_F_NOTIFICATION_DATA)))
+               return -EOPNOTSUPP;
+
+       if (!(features & BIT_ULL(VIRTIO_F_RING_PACKED)))
+               return -EOPNOTSUPP;
+
+       return 0;
+}
+
+u8 octep_hw_get_status(struct octep_hw *oct_hw)
+{
+       return ioread8(&oct_hw->common_cfg->device_status);
+}
+
+void octep_hw_set_status(struct octep_hw *oct_hw, u8 status)
+{
+       iowrite8(status, &oct_hw->common_cfg->device_status);
+}
+
+void octep_hw_reset(struct octep_hw *oct_hw)
+{
+       u8 val;
+
+       octep_hw_set_status(oct_hw, 0 | BIT(DEV_RST_ACK_BIT));
+       if (readx_poll_timeout(ioread8, &oct_hw->common_cfg->device_status, val, !val, 10,
+                              OCTEP_HW_TIMEOUT)) {
+               dev_warn(&oct_hw->pdev->dev, "Octeon device reset timeout\n");
+               return;
+       }
+}
+
+static int feature_sel_write_with_timeout(struct octep_hw *oct_hw, u32 select, void __iomem *addr)
+{
+       u32 val;
+
+       iowrite32(select | BIT(FEATURE_SEL_ACK_BIT), addr);
+
+       if (readx_poll_timeout(ioread32, addr, val, val == select, 10, OCTEP_HW_TIMEOUT)) {
+               dev_warn(&oct_hw->pdev->dev, "Feature select%d write timeout\n", select);
+               return -1;
+       }
+       return 0;
+}
+
+u64 octep_hw_get_dev_features(struct octep_hw *oct_hw)
+{
+       u32 features_lo, features_hi;
+
+       if (feature_sel_write_with_timeout(oct_hw, 0, &oct_hw->common_cfg->device_feature_select))
+               return 0;
+
+       features_lo = ioread32(&oct_hw->common_cfg->device_feature);
+
+       if (feature_sel_write_with_timeout(oct_hw, 1, &oct_hw->common_cfg->device_feature_select))
+               return 0;
+
+       features_hi = ioread32(&oct_hw->common_cfg->device_feature);
+
+       return ((u64)features_hi << 32) | features_lo;
+}
+
+u64 octep_hw_get_drv_features(struct octep_hw *oct_hw)
+{
+       u32 features_lo, features_hi;
+
+       if (feature_sel_write_with_timeout(oct_hw, 0, &oct_hw->common_cfg->guest_feature_select))
+               return 0;
+
+       features_lo = ioread32(&oct_hw->common_cfg->guest_feature);
+
+       if (feature_sel_write_with_timeout(oct_hw, 1, &oct_hw->common_cfg->guest_feature_select))
+               return 0;
+
+       features_hi = ioread32(&oct_hw->common_cfg->guest_feature);
+
+       return ((u64)features_hi << 32) | features_lo;
+}
+
+void octep_hw_set_drv_features(struct octep_hw *oct_hw, u64 features)
+{
+       if (feature_sel_write_with_timeout(oct_hw, 0, &oct_hw->common_cfg->guest_feature_select))
+               return;
+
+       iowrite32(features & (BIT_ULL(32) - 1), &oct_hw->common_cfg->guest_feature);
+
+       if (feature_sel_write_with_timeout(oct_hw, 1, &oct_hw->common_cfg->guest_feature_select))
+               return;
+
+       iowrite32(features >> 32, &oct_hw->common_cfg->guest_feature);
+}
+
+void octep_write_queue_select(struct octep_hw *oct_hw, u16 queue_id)
+{
+       u16 val;
+
+       iowrite16(queue_id | BIT(QUEUE_SEL_ACK_BIT), &oct_hw->common_cfg->queue_select);
+
+       if (readx_poll_timeout(ioread16, &oct_hw->common_cfg->queue_select, val, val == queue_id,
+                              10, OCTEP_HW_TIMEOUT)) {
+               dev_warn(&oct_hw->pdev->dev, "Queue select write timeout\n");
+               return;
+       }
+}
+
+void octep_notify_queue(struct octep_hw *oct_hw, u16 qid)
+{
+       iowrite16(qid, oct_hw->vqs[qid].notify_addr);
+}
+
+void octep_read_dev_config(struct octep_hw *oct_hw, u64 offset, void *dst, int length)
+{
+       u8 old_gen, new_gen, *p;
+       int i;
+
+       if (WARN_ON(offset + length > oct_hw->config_size))
+               return;
+
+       do {
+               old_gen = ioread8(&oct_hw->common_cfg->config_generation);
+               p = dst;
+               for (i = 0; i < length; i++)
+                       *p++ = ioread8(oct_hw->dev_cfg + offset + i);
+
+               new_gen = ioread8(&oct_hw->common_cfg->config_generation);
+       } while (old_gen != new_gen);
+}
+
+int octep_set_vq_address(struct octep_hw *oct_hw, u16 qid, u64 desc_area, u64 driver_area,
+                        u64 device_area)
+{
+       struct virtio_pci_common_cfg __iomem *cfg = oct_hw->common_cfg;
+
+       octep_write_queue_select(oct_hw, qid);
+       vp_iowrite64_twopart(desc_area, &cfg->queue_desc_lo,
+                            &cfg->queue_desc_hi);
+       vp_iowrite64_twopart(driver_area, &cfg->queue_avail_lo,
+                            &cfg->queue_avail_hi);
+       vp_iowrite64_twopart(device_area, &cfg->queue_used_lo,
+                            &cfg->queue_used_hi);
+
+       return 0;
+}
+
+int octep_get_vq_state(struct octep_hw *oct_hw, u16 qid, struct vdpa_vq_state *state)
+{
+       return octep_process_mbox(oct_hw, OCTEP_MBOX_MSG_GET_VQ_STATE, qid, state,
+                                 sizeof(*state), 0);
+}
+
+int octep_set_vq_state(struct octep_hw *oct_hw, u16 qid, const struct vdpa_vq_state *state)
+{
+       struct vdpa_vq_state q_state;
+
+       memcpy(&q_state, state, sizeof(struct vdpa_vq_state));
+       return octep_process_mbox(oct_hw, OCTEP_MBOX_MSG_SET_VQ_STATE, qid, &q_state,
+                                 sizeof(*state), 1);
+}
+
+void octep_set_vq_num(struct octep_hw *oct_hw, u16 qid, u32 num)
+{
+       struct virtio_pci_common_cfg __iomem *cfg = oct_hw->common_cfg;
+
+       octep_write_queue_select(oct_hw, qid);
+       iowrite16(num, &cfg->queue_size);
+}
+
+void octep_set_vq_ready(struct octep_hw *oct_hw, u16 qid, bool ready)
+{
+       struct virtio_pci_common_cfg __iomem *cfg = oct_hw->common_cfg;
+
+       octep_write_queue_select(oct_hw, qid);
+       iowrite16(ready, &cfg->queue_enable);
+}
+
+bool octep_get_vq_ready(struct octep_hw *oct_hw, u16 qid)
+{
+       struct virtio_pci_common_cfg __iomem *cfg = oct_hw->common_cfg;
+
+       octep_write_queue_select(oct_hw, qid);
+       return ioread16(&cfg->queue_enable);
+}
+
+u16 octep_get_vq_size(struct octep_hw *oct_hw)
+{
+       octep_write_queue_select(oct_hw, 0);
+       return ioread16(&oct_hw->common_cfg->queue_size);
+}
+
+static u32 octep_get_config_size(struct octep_hw *oct_hw)
+{
+       return sizeof(struct virtio_net_config);
+}
+
+static void __iomem *octep_get_cap_addr(struct octep_hw *oct_hw, struct virtio_pci_cap *cap)
+{
+       struct device *dev = &oct_hw->pdev->dev;
+       u32 length = le32_to_cpu(cap->length);
+       u32 offset = le32_to_cpu(cap->offset);
+       u8  bar    = cap->bar;
+       u32 len;
+
+       if (bar != OCTEP_HW_CAPS_BAR) {
+               dev_err(dev, "Invalid bar: %u\n", bar);
+               return NULL;
+       }
+       if (offset + length < offset) {
+               dev_err(dev, "offset(%u) + length(%u) overflows\n",
+                       offset, length);
+               return NULL;
+       }
+       len = pci_resource_len(oct_hw->pdev, bar);
+       if (offset + length > len) {
+               dev_err(dev, "invalid cap: overflows bar space: %u > %u\n",
+                       offset + length, len);
+               return NULL;
+       }
+       return oct_hw->base[bar] + offset;
+}
+
+/* In Octeon DPU device, the virtio config space is completely
+ * emulated by the device's firmware. So, the standard pci config
+ * read apis can't be used for reading the virtio capability.
+ */
+static void octep_pci_caps_read(struct octep_hw *oct_hw, void *buf, size_t len, off_t offset)
+{
+       u8 __iomem *bar = oct_hw->base[OCTEP_HW_CAPS_BAR];
+       u8 *p = buf;
+       size_t i;
+
+       for (i = 0; i < len; i++)
+               *p++ = ioread8(bar + offset + i);
+}
+
+static int octep_pci_signature_verify(struct octep_hw *oct_hw)
+{
+       u32 signature[2];
+
+       octep_pci_caps_read(oct_hw, &signature, sizeof(signature), 0);
+
+       if (signature[0] != OCTEP_FW_READY_SIGNATURE0)
+               return -1;
+
+       if (signature[1] != OCTEP_FW_READY_SIGNATURE1)
+               return -1;
+
+       return 0;
+}
+
+int octep_hw_caps_read(struct octep_hw *oct_hw, struct pci_dev *pdev)
+{
+       struct octep_mbox __iomem *mbox;
+       struct device *dev = &pdev->dev;
+       struct virtio_pci_cap cap;
+       u16 notify_off;
+       int i, ret;
+       u8 pos;
+
+       oct_hw->pdev = pdev;
+       ret = octep_pci_signature_verify(oct_hw);
+       if (ret) {
+               dev_err(dev, "Octeon Virtio FW is not initialized\n");
+               return -EIO;
+       }
+
+       octep_pci_caps_read(oct_hw, &pos, 1, PCI_CAPABILITY_LIST);
+
+       while (pos) {
+               octep_pci_caps_read(oct_hw, &cap, 2, pos);
+
+               if (cap.cap_vndr != PCI_CAP_ID_VNDR) {
+                       dev_err(dev, "Found invalid capability vndr id: %d\n", cap.cap_vndr);
+                       break;
+               }
+
+               octep_pci_caps_read(oct_hw, &cap, sizeof(cap), pos);
+
+               dev_info(dev, "[%2x] cfg type: %u, bar: %u, offset: %04x, len: %u\n",
+                        pos, cap.cfg_type, cap.bar, cap.offset, cap.length);
+
+               switch (cap.cfg_type) {
+               case VIRTIO_PCI_CAP_COMMON_CFG:
+                       oct_hw->common_cfg = octep_get_cap_addr(oct_hw, &cap);
+                       break;
+               case VIRTIO_PCI_CAP_NOTIFY_CFG:
+                       octep_pci_caps_read(oct_hw, &oct_hw->notify_off_multiplier,
+                                           4, pos + sizeof(cap));
+
+                       oct_hw->notify_base = octep_get_cap_addr(oct_hw, &cap);
+                       oct_hw->notify_bar = cap.bar;
+                       oct_hw->notify_base_pa = pci_resource_start(pdev, cap.bar) +
+                                                le32_to_cpu(cap.offset);
+                       break;
+               case VIRTIO_PCI_CAP_DEVICE_CFG:
+                       oct_hw->dev_cfg = octep_get_cap_addr(oct_hw, &cap);
+                       break;
+               case VIRTIO_PCI_CAP_ISR_CFG:
+                       oct_hw->isr = octep_get_cap_addr(oct_hw, &cap);
+                       break;
+               }
+
+               pos = cap.cap_next;
+       }
+       if (!oct_hw->common_cfg || !oct_hw->notify_base ||
+           !oct_hw->dev_cfg    || !oct_hw->isr) {
+               dev_err(dev, "Incomplete PCI capabilities");
+               return -EIO;
+       }
+       dev_info(dev, "common cfg mapped at: 0x%016llx\n", (u64)(uintptr_t)oct_hw->common_cfg);
+       dev_info(dev, "device cfg mapped at: 0x%016llx\n", (u64)(uintptr_t)oct_hw->dev_cfg);
+       dev_info(dev, "isr cfg mapped at: 0x%016llx\n", (u64)(uintptr_t)oct_hw->isr);
+       dev_info(dev, "notify base: 0x%016llx, notify off multiplier: %u\n",
+                (u64)(uintptr_t)oct_hw->notify_base, oct_hw->notify_off_multiplier);
+
+       oct_hw->config_size = octep_get_config_size(oct_hw);
+       oct_hw->features = octep_hw_get_dev_features(oct_hw);
+
+       ret = octep_verify_features(oct_hw->features);
+       if (ret) {
+               dev_err(&pdev->dev, "Couldn't read features from the device FW\n");
+               return ret;
+       }
+       oct_hw->nr_vring = vp_ioread16(&oct_hw->common_cfg->num_queues);
+
+       oct_hw->vqs = devm_kcalloc(&pdev->dev, oct_hw->nr_vring, sizeof(*oct_hw->vqs), GFP_KERNEL);
+       if (!oct_hw->vqs)
+               return -ENOMEM;
+
+       oct_hw->irq = -1;
+
+       dev_info(&pdev->dev, "Device features : %llx\n", oct_hw->features);
+       dev_info(&pdev->dev, "Maximum queues : %u\n", oct_hw->nr_vring);
+
+       for (i = 0; i < oct_hw->nr_vring; i++) {
+               octep_write_queue_select(oct_hw, i);
+               notify_off = vp_ioread16(&oct_hw->common_cfg->queue_notify_off);
+               oct_hw->vqs[i].notify_addr = oct_hw->notify_base +
+                       notify_off * oct_hw->notify_off_multiplier;
+               oct_hw->vqs[i].cb_notify_addr = (u32 __iomem *)oct_hw->vqs[i].notify_addr + 1;
+               oct_hw->vqs[i].notify_pa = oct_hw->notify_base_pa +
+                       notify_off * oct_hw->notify_off_multiplier;
+       }
+       mbox = octep_get_mbox(oct_hw);
+       octep_mbox_init(mbox);
+       dev_info(dev, "mbox mapped at: 0x%016llx\n", (u64)(uintptr_t)mbox);
+
+       return 0;
+}
diff --git a/drivers/vdpa/octeon_ep/octep_vdpa_main.c b/drivers/vdpa/octeon_ep/octep_vdpa_main.c
new file mode 100644 (file)
index 0000000..cd55b1a
--- /dev/null
@@ -0,0 +1,857 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Copyright (C) 2024 Marvell. */
+
+#include <linux/interrupt.h>
+#include <linux/io-64-nonatomic-lo-hi.h>
+#include <linux/module.h>
+#include <linux/iommu.h>
+#include "octep_vdpa.h"
+
+#define OCTEP_VDPA_DRIVER_NAME "octep_vdpa"
+
+struct octep_pf {
+       u8 __iomem *base[PCI_STD_NUM_BARS];
+       struct pci_dev *pdev;
+       struct resource res;
+       u64 vf_base;
+       int enabled_vfs;
+       u32 vf_stride;
+       u16 vf_devid;
+};
+
+struct octep_vdpa {
+       struct vdpa_device vdpa;
+       struct octep_hw *oct_hw;
+       struct pci_dev *pdev;
+};
+
+struct octep_vdpa_mgmt_dev {
+       struct vdpa_mgmt_dev mdev;
+       struct octep_hw oct_hw;
+       struct pci_dev *pdev;
+       /* Work entry to handle device setup */
+       struct work_struct setup_task;
+       /* Device status */
+       atomic_t status;
+};
+
+static struct octep_hw *vdpa_to_octep_hw(struct vdpa_device *vdpa_dev)
+{
+       struct octep_vdpa *oct_vdpa;
+
+       oct_vdpa = container_of(vdpa_dev, struct octep_vdpa, vdpa);
+
+       return oct_vdpa->oct_hw;
+}
+
+static irqreturn_t octep_vdpa_intr_handler(int irq, void *data)
+{
+       struct octep_hw *oct_hw = data;
+       int i;
+
+       for (i = 0; i < oct_hw->nr_vring; i++) {
+               if (oct_hw->vqs[i].cb.callback && ioread32(oct_hw->vqs[i].cb_notify_addr)) {
+                       /* Acknowledge the per queue notification to the device */
+                       iowrite32(0, oct_hw->vqs[i].cb_notify_addr);
+                       oct_hw->vqs[i].cb.callback(oct_hw->vqs[i].cb.private);
+               }
+       }
+
+       return IRQ_HANDLED;
+}
+
+static void octep_free_irqs(struct octep_hw *oct_hw)
+{
+       struct pci_dev *pdev = oct_hw->pdev;
+
+       if (oct_hw->irq != -1) {
+               devm_free_irq(&pdev->dev, oct_hw->irq, oct_hw);
+               oct_hw->irq = -1;
+       }
+       pci_free_irq_vectors(pdev);
+}
+
+static int octep_request_irqs(struct octep_hw *oct_hw)
+{
+       struct pci_dev *pdev = oct_hw->pdev;
+       int ret, irq;
+
+       /* Currently HW device provisions one IRQ per VF, hence
+        * allocate one IRQ for all virtqueues call interface.
+        */
+       ret = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_MSIX);
+       if (ret < 0) {
+               dev_err(&pdev->dev, "Failed to alloc msix vector");
+               return ret;
+       }
+
+       snprintf(oct_hw->vqs->msix_name, sizeof(oct_hw->vqs->msix_name),
+                OCTEP_VDPA_DRIVER_NAME "-vf-%d", pci_iov_vf_id(pdev));
+
+       irq = pci_irq_vector(pdev, 0);
+       ret = devm_request_irq(&pdev->dev, irq, octep_vdpa_intr_handler, 0,
+                              oct_hw->vqs->msix_name, oct_hw);
+       if (ret) {
+               dev_err(&pdev->dev, "Failed to register interrupt handler\n");
+               goto free_irq_vec;
+       }
+       oct_hw->irq = irq;
+
+       return 0;
+
+free_irq_vec:
+       pci_free_irq_vectors(pdev);
+       return ret;
+}
+
+static u64 octep_vdpa_get_device_features(struct vdpa_device *vdpa_dev)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       return oct_hw->features;
+}
+
+static int octep_vdpa_set_driver_features(struct vdpa_device *vdpa_dev, u64 features)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+       int ret;
+
+       pr_debug("Driver Features: %llx\n", features);
+
+       ret = octep_verify_features(features);
+       if (ret) {
+               dev_warn(&oct_hw->pdev->dev,
+                        "Must negotiate minimum features 0x%llx for this device",
+                        BIT_ULL(VIRTIO_F_VERSION_1) | BIT_ULL(VIRTIO_F_NOTIFICATION_DATA) |
+                        BIT_ULL(VIRTIO_F_RING_PACKED));
+               return ret;
+       }
+       octep_hw_set_drv_features(oct_hw, features);
+
+       return 0;
+}
+
+static u64 octep_vdpa_get_driver_features(struct vdpa_device *vdpa_dev)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       return octep_hw_get_drv_features(oct_hw);
+}
+
+static u8 octep_vdpa_get_status(struct vdpa_device *vdpa_dev)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       return octep_hw_get_status(oct_hw);
+}
+
+static void octep_vdpa_set_status(struct vdpa_device *vdpa_dev, u8 status)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+       u8 status_old;
+
+       status_old = octep_hw_get_status(oct_hw);
+
+       if (status_old == status)
+               return;
+
+       if ((status & VIRTIO_CONFIG_S_DRIVER_OK) &&
+           !(status_old & VIRTIO_CONFIG_S_DRIVER_OK)) {
+               if (octep_request_irqs(oct_hw))
+                       status = status_old | VIRTIO_CONFIG_S_FAILED;
+       }
+       octep_hw_set_status(oct_hw, status);
+}
+
+static int octep_vdpa_reset(struct vdpa_device *vdpa_dev)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+       u8 status = octep_hw_get_status(oct_hw);
+       u16 qid;
+
+       if (status == 0)
+               return 0;
+
+       for (qid = 0; qid < oct_hw->nr_vring; qid++) {
+               oct_hw->vqs[qid].cb.callback = NULL;
+               oct_hw->vqs[qid].cb.private = NULL;
+               oct_hw->config_cb.callback = NULL;
+               oct_hw->config_cb.private = NULL;
+       }
+       octep_hw_reset(oct_hw);
+
+       if (status & VIRTIO_CONFIG_S_DRIVER_OK)
+               octep_free_irqs(oct_hw);
+
+       return 0;
+}
+
+static u16 octep_vdpa_get_vq_num_max(struct vdpa_device *vdpa_dev)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       return octep_get_vq_size(oct_hw);
+}
+
+static int octep_vdpa_get_vq_state(struct vdpa_device *vdpa_dev, u16 qid,
+                                  struct vdpa_vq_state *state)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       return octep_get_vq_state(oct_hw, qid, state);
+}
+
+static int octep_vdpa_set_vq_state(struct vdpa_device *vdpa_dev, u16 qid,
+                                  const struct vdpa_vq_state *state)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       return octep_set_vq_state(oct_hw, qid, state);
+}
+
+static void octep_vdpa_set_vq_cb(struct vdpa_device *vdpa_dev, u16 qid, struct vdpa_callback *cb)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       oct_hw->vqs[qid].cb = *cb;
+}
+
+static void octep_vdpa_set_vq_ready(struct vdpa_device *vdpa_dev, u16 qid, bool ready)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       octep_set_vq_ready(oct_hw, qid, ready);
+}
+
+static bool octep_vdpa_get_vq_ready(struct vdpa_device *vdpa_dev, u16 qid)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       return octep_get_vq_ready(oct_hw, qid);
+}
+
+static void octep_vdpa_set_vq_num(struct vdpa_device *vdpa_dev, u16 qid, u32 num)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       octep_set_vq_num(oct_hw, qid, num);
+}
+
+static int octep_vdpa_set_vq_address(struct vdpa_device *vdpa_dev, u16 qid, u64 desc_area,
+                                    u64 driver_area, u64 device_area)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       pr_debug("qid[%d]: desc_area: %llx\n", qid, desc_area);
+       pr_debug("qid[%d]: driver_area: %llx\n", qid, driver_area);
+       pr_debug("qid[%d]: device_area: %llx\n\n", qid, device_area);
+
+       return octep_set_vq_address(oct_hw, qid, desc_area, driver_area, device_area);
+}
+
+static void octep_vdpa_kick_vq(struct vdpa_device *vdpa_dev, u16 qid)
+{
+       /* Not supported */
+}
+
+static void octep_vdpa_kick_vq_with_data(struct vdpa_device *vdpa_dev, u32 data)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+       u16 idx = data & 0xFFFF;
+
+       vp_iowrite32(data, oct_hw->vqs[idx].notify_addr);
+}
+
+static u32 octep_vdpa_get_generation(struct vdpa_device *vdpa_dev)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       return vp_ioread8(&oct_hw->common_cfg->config_generation);
+}
+
+static u32 octep_vdpa_get_device_id(struct vdpa_device *vdpa_dev)
+{
+       return VIRTIO_ID_NET;
+}
+
+static u32 octep_vdpa_get_vendor_id(struct vdpa_device *vdpa_dev)
+{
+       return PCI_VENDOR_ID_CAVIUM;
+}
+
+static u32 octep_vdpa_get_vq_align(struct vdpa_device *vdpa_dev)
+{
+       return PAGE_SIZE;
+}
+
+static size_t octep_vdpa_get_config_size(struct vdpa_device *vdpa_dev)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       return oct_hw->config_size;
+}
+
+static void octep_vdpa_get_config(struct vdpa_device *vdpa_dev, unsigned int offset, void *buf,
+                                 unsigned int len)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       octep_read_dev_config(oct_hw, offset, buf, len);
+}
+
+static void octep_vdpa_set_config(struct vdpa_device *vdpa_dev, unsigned int offset,
+                                 const void *buf, unsigned int len)
+{
+       /* Not supported */
+}
+
+static void octep_vdpa_set_config_cb(struct vdpa_device *vdpa_dev, struct vdpa_callback *cb)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+
+       oct_hw->config_cb.callback = cb->callback;
+       oct_hw->config_cb.private = cb->private;
+}
+
+static struct vdpa_notification_area octep_get_vq_notification(struct vdpa_device *vdpa_dev,
+                                                              u16 idx)
+{
+       struct octep_hw *oct_hw = vdpa_to_octep_hw(vdpa_dev);
+       struct vdpa_notification_area area;
+
+       area.addr = oct_hw->vqs[idx].notify_pa;
+       area.size = PAGE_SIZE;
+
+       return area;
+}
+
+static struct vdpa_config_ops octep_vdpa_ops = {
+       .get_device_features = octep_vdpa_get_device_features,
+       .set_driver_features = octep_vdpa_set_driver_features,
+       .get_driver_features = octep_vdpa_get_driver_features,
+       .get_status     = octep_vdpa_get_status,
+       .set_status     = octep_vdpa_set_status,
+       .reset          = octep_vdpa_reset,
+       .get_vq_num_max = octep_vdpa_get_vq_num_max,
+       .get_vq_state   = octep_vdpa_get_vq_state,
+       .set_vq_state   = octep_vdpa_set_vq_state,
+       .set_vq_cb      = octep_vdpa_set_vq_cb,
+       .set_vq_ready   = octep_vdpa_set_vq_ready,
+       .get_vq_ready   = octep_vdpa_get_vq_ready,
+       .set_vq_num     = octep_vdpa_set_vq_num,
+       .set_vq_address = octep_vdpa_set_vq_address,
+       .get_vq_irq     = NULL,
+       .kick_vq        = octep_vdpa_kick_vq,
+       .kick_vq_with_data      = octep_vdpa_kick_vq_with_data,
+       .get_generation = octep_vdpa_get_generation,
+       .get_device_id  = octep_vdpa_get_device_id,
+       .get_vendor_id  = octep_vdpa_get_vendor_id,
+       .get_vq_align   = octep_vdpa_get_vq_align,
+       .get_config_size        = octep_vdpa_get_config_size,
+       .get_config     = octep_vdpa_get_config,
+       .set_config     = octep_vdpa_set_config,
+       .set_config_cb  = octep_vdpa_set_config_cb,
+       .get_vq_notification = octep_get_vq_notification,
+};
+
+static int octep_iomap_region(struct pci_dev *pdev, u8 __iomem **tbl, u8 bar)
+{
+       int ret;
+
+       ret = pci_request_region(pdev, bar, OCTEP_VDPA_DRIVER_NAME);
+       if (ret) {
+               dev_err(&pdev->dev, "Failed to request BAR:%u region\n", bar);
+               return ret;
+       }
+
+       tbl[bar] = pci_iomap(pdev, bar, pci_resource_len(pdev, bar));
+       if (!tbl[bar]) {
+               dev_err(&pdev->dev, "Failed to iomap BAR:%u\n", bar);
+               pci_release_region(pdev, bar);
+               ret = -ENOMEM;
+       }
+
+       return ret;
+}
+
+static void octep_iounmap_region(struct pci_dev *pdev, u8 __iomem **tbl, u8 bar)
+{
+       pci_iounmap(pdev, tbl[bar]);
+       pci_release_region(pdev, bar);
+}
+
+static void octep_vdpa_pf_bar_shrink(struct octep_pf *octpf)
+{
+       struct pci_dev *pf_dev = octpf->pdev;
+       struct resource *res = pf_dev->resource + PCI_STD_RESOURCES + 4;
+       struct pci_bus_region bus_region;
+
+       octpf->res.start = res->start;
+       octpf->res.end = res->end;
+       octpf->vf_base = res->start;
+
+       bus_region.start = res->start;
+       bus_region.end = res->start - 1;
+
+       pcibios_bus_to_resource(pf_dev->bus, res, &bus_region);
+}
+
+static void octep_vdpa_pf_bar_expand(struct octep_pf *octpf)
+{
+       struct pci_dev *pf_dev = octpf->pdev;
+       struct resource *res = pf_dev->resource + PCI_STD_RESOURCES + 4;
+       struct pci_bus_region bus_region;
+
+       bus_region.start = octpf->res.start;
+       bus_region.end = octpf->res.end;
+
+       pcibios_bus_to_resource(pf_dev->bus, res, &bus_region);
+}
+
+static void octep_vdpa_remove_pf(struct pci_dev *pdev)
+{
+       struct octep_pf *octpf = pci_get_drvdata(pdev);
+
+       pci_disable_sriov(pdev);
+
+       if (octpf->base[OCTEP_HW_CAPS_BAR])
+               octep_iounmap_region(pdev, octpf->base, OCTEP_HW_CAPS_BAR);
+
+       if (octpf->base[OCTEP_HW_MBOX_BAR])
+               octep_iounmap_region(pdev, octpf->base, OCTEP_HW_MBOX_BAR);
+
+       octep_vdpa_pf_bar_expand(octpf);
+}
+
+static void octep_vdpa_vf_bar_shrink(struct pci_dev *pdev)
+{
+       struct resource *vf_res = pdev->resource + PCI_STD_RESOURCES + 4;
+
+       memset(vf_res, 0, sizeof(*vf_res));
+}
+
+static void octep_vdpa_remove_vf(struct pci_dev *pdev)
+{
+       struct octep_vdpa_mgmt_dev *mgmt_dev = pci_get_drvdata(pdev);
+       struct octep_hw *oct_hw;
+       int status;
+
+       oct_hw = &mgmt_dev->oct_hw;
+       status = atomic_read(&mgmt_dev->status);
+       atomic_set(&mgmt_dev->status, OCTEP_VDPA_DEV_STATUS_UNINIT);
+
+       cancel_work_sync(&mgmt_dev->setup_task);
+       if (status == OCTEP_VDPA_DEV_STATUS_READY)
+               vdpa_mgmtdev_unregister(&mgmt_dev->mdev);
+
+       if (oct_hw->base[OCTEP_HW_CAPS_BAR])
+               octep_iounmap_region(pdev, oct_hw->base, OCTEP_HW_CAPS_BAR);
+
+       if (oct_hw->base[OCTEP_HW_MBOX_BAR])
+               octep_iounmap_region(pdev, oct_hw->base, OCTEP_HW_MBOX_BAR);
+
+       octep_vdpa_vf_bar_shrink(pdev);
+}
+
+static void octep_vdpa_remove(struct pci_dev *pdev)
+{
+       if (pdev->is_virtfn)
+               octep_vdpa_remove_vf(pdev);
+       else
+               octep_vdpa_remove_pf(pdev);
+}
+
+static int octep_vdpa_dev_add(struct vdpa_mgmt_dev *mdev, const char *name,
+                             const struct vdpa_dev_set_config *config)
+{
+       struct octep_vdpa_mgmt_dev *mgmt_dev = container_of(mdev, struct octep_vdpa_mgmt_dev, mdev);
+       struct octep_hw *oct_hw = &mgmt_dev->oct_hw;
+       struct pci_dev *pdev = oct_hw->pdev;
+       struct vdpa_device *vdpa_dev;
+       struct octep_vdpa *oct_vdpa;
+       u64 device_features;
+       int ret;
+
+       oct_vdpa = vdpa_alloc_device(struct octep_vdpa, vdpa, &pdev->dev, &octep_vdpa_ops, 1, 1,
+                                    NULL, false);
+       if (IS_ERR(oct_vdpa)) {
+               dev_err(&pdev->dev, "Failed to allocate vDPA structure for octep vdpa device");
+               return PTR_ERR(oct_vdpa);
+       }
+
+       oct_vdpa->pdev = pdev;
+       oct_vdpa->vdpa.dma_dev = &pdev->dev;
+       oct_vdpa->vdpa.mdev = mdev;
+       oct_vdpa->oct_hw = oct_hw;
+       vdpa_dev = &oct_vdpa->vdpa;
+
+       device_features = oct_hw->features;
+       if (config->mask & BIT_ULL(VDPA_ATTR_DEV_FEATURES)) {
+               if (config->device_features & ~device_features) {
+                       dev_err(&pdev->dev, "The provisioned features 0x%llx are not supported by this device with features 0x%llx\n",
+                               config->device_features, device_features);
+                       ret = -EINVAL;
+                       goto vdpa_dev_put;
+               }
+               device_features &= config->device_features;
+       }
+
+       oct_hw->features = device_features;
+       dev_info(&pdev->dev, "Vdpa management device features : %llx\n", device_features);
+
+       ret = octep_verify_features(device_features);
+       if (ret) {
+               dev_warn(mdev->device,
+                        "Must provision minimum features 0x%llx for this device",
+                        BIT_ULL(VIRTIO_F_VERSION_1) | BIT_ULL(VIRTIO_F_ACCESS_PLATFORM) |
+                        BIT_ULL(VIRTIO_F_NOTIFICATION_DATA) | BIT_ULL(VIRTIO_F_RING_PACKED));
+               goto vdpa_dev_put;
+       }
+       if (name)
+               ret = dev_set_name(&vdpa_dev->dev, "%s", name);
+       else
+               ret = dev_set_name(&vdpa_dev->dev, "vdpa%u", vdpa_dev->index);
+
+       ret = _vdpa_register_device(&oct_vdpa->vdpa, oct_hw->nr_vring);
+       if (ret) {
+               dev_err(&pdev->dev, "Failed to register to vDPA bus");
+               goto vdpa_dev_put;
+       }
+       return 0;
+
+vdpa_dev_put:
+       put_device(&oct_vdpa->vdpa.dev);
+       return ret;
+}
+
+static void octep_vdpa_dev_del(struct vdpa_mgmt_dev *mdev, struct vdpa_device *vdpa_dev)
+{
+       _vdpa_unregister_device(vdpa_dev);
+}
+
+static const struct vdpa_mgmtdev_ops octep_vdpa_mgmt_dev_ops = {
+       .dev_add = octep_vdpa_dev_add,
+       .dev_del = octep_vdpa_dev_del
+};
+
+static bool get_device_ready_status(u8 __iomem *addr)
+{
+       u64 signature = readq(addr + OCTEP_VF_MBOX_DATA(0));
+
+       if (signature == OCTEP_DEV_READY_SIGNATURE) {
+               writeq(0, addr + OCTEP_VF_MBOX_DATA(0));
+               return true;
+       }
+
+       return false;
+}
+
+static struct virtio_device_id id_table[] = {
+       { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
+       { 0 },
+};
+
+static void octep_vdpa_setup_task(struct work_struct *work)
+{
+       struct octep_vdpa_mgmt_dev *mgmt_dev = container_of(work, struct octep_vdpa_mgmt_dev,
+                                                           setup_task);
+       struct pci_dev *pdev = mgmt_dev->pdev;
+       struct device *dev = &pdev->dev;
+       struct octep_hw *oct_hw;
+       unsigned long timeout;
+       int ret;
+
+       oct_hw = &mgmt_dev->oct_hw;
+
+       atomic_set(&mgmt_dev->status, OCTEP_VDPA_DEV_STATUS_WAIT_FOR_BAR_INIT);
+
+       /* Wait for a maximum of 5 sec */
+       timeout = jiffies + msecs_to_jiffies(5000);
+       while (!time_after(jiffies, timeout)) {
+               if (get_device_ready_status(oct_hw->base[OCTEP_HW_MBOX_BAR])) {
+                       atomic_set(&mgmt_dev->status, OCTEP_VDPA_DEV_STATUS_INIT);
+                       break;
+               }
+
+               if (atomic_read(&mgmt_dev->status) >= OCTEP_VDPA_DEV_STATUS_READY) {
+                       dev_info(dev, "Stopping vDPA setup task.\n");
+                       return;
+               }
+
+               usleep_range(1000, 1500);
+       }
+
+       if (atomic_read(&mgmt_dev->status) != OCTEP_VDPA_DEV_STATUS_INIT) {
+               dev_err(dev, "BAR initialization is timed out\n");
+               return;
+       }
+
+       ret = octep_iomap_region(pdev, oct_hw->base, OCTEP_HW_CAPS_BAR);
+       if (ret)
+               return;
+
+       ret = octep_hw_caps_read(oct_hw, pdev);
+       if (ret < 0)
+               goto unmap_region;
+
+       mgmt_dev->mdev.ops = &octep_vdpa_mgmt_dev_ops;
+       mgmt_dev->mdev.id_table = id_table;
+       mgmt_dev->mdev.max_supported_vqs = oct_hw->nr_vring;
+       mgmt_dev->mdev.supported_features = oct_hw->features;
+       mgmt_dev->mdev.config_attr_mask = (1 << VDPA_ATTR_DEV_FEATURES);
+       mgmt_dev->mdev.device = dev;
+
+       ret = vdpa_mgmtdev_register(&mgmt_dev->mdev);
+       if (ret) {
+               dev_err(dev, "Failed to register vdpa management interface\n");
+               goto unmap_region;
+       }
+
+       atomic_set(&mgmt_dev->status, OCTEP_VDPA_DEV_STATUS_READY);
+
+       return;
+
+unmap_region:
+       octep_iounmap_region(pdev, oct_hw->base, OCTEP_HW_CAPS_BAR);
+       oct_hw->base[OCTEP_HW_CAPS_BAR] = NULL;
+}
+
+static int octep_vdpa_probe_vf(struct pci_dev *pdev)
+{
+       struct octep_vdpa_mgmt_dev *mgmt_dev;
+       struct device *dev = &pdev->dev;
+       int ret;
+
+       ret = pcim_enable_device(pdev);
+       if (ret) {
+               dev_err(dev, "Failed to enable device\n");
+               return ret;
+       }
+
+       ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64));
+       if (ret) {
+               dev_err(dev, "No usable DMA configuration\n");
+               return ret;
+       }
+       pci_set_master(pdev);
+
+       mgmt_dev = devm_kzalloc(dev, sizeof(struct octep_vdpa_mgmt_dev), GFP_KERNEL);
+       if (!mgmt_dev)
+               return -ENOMEM;
+
+       ret = octep_iomap_region(pdev, mgmt_dev->oct_hw.base, OCTEP_HW_MBOX_BAR);
+       if (ret)
+               return ret;
+
+       mgmt_dev->pdev = pdev;
+       pci_set_drvdata(pdev, mgmt_dev);
+
+       atomic_set(&mgmt_dev->status, OCTEP_VDPA_DEV_STATUS_ALLOC);
+       INIT_WORK(&mgmt_dev->setup_task, octep_vdpa_setup_task);
+       schedule_work(&mgmt_dev->setup_task);
+       dev_info(&pdev->dev, "octep vdpa mgmt device setup task is queued\n");
+
+       return 0;
+}
+
+static void octep_vdpa_assign_barspace(struct pci_dev *vf_dev, struct pci_dev *pf_dev, u8 idx)
+{
+       struct resource *vf_res = vf_dev->resource + PCI_STD_RESOURCES + 4;
+       struct resource *pf_res = pf_dev->resource + PCI_STD_RESOURCES + 4;
+       struct octep_pf *pf = pci_get_drvdata(pf_dev);
+       struct pci_bus_region bus_region;
+
+       vf_res->name = pci_name(vf_dev);
+       vf_res->flags = pf_res->flags;
+       vf_res->parent = (pf_dev->resource + PCI_STD_RESOURCES)->parent;
+
+       bus_region.start = pf->vf_base + idx * pf->vf_stride;
+       bus_region.end = bus_region.start + pf->vf_stride - 1;
+       pcibios_bus_to_resource(vf_dev->bus, vf_res, &bus_region);
+}
+
+static int octep_sriov_enable(struct pci_dev *pdev, int num_vfs)
+{
+       struct octep_pf *pf = pci_get_drvdata(pdev);
+       u8 __iomem *addr = pf->base[OCTEP_HW_MBOX_BAR];
+       struct pci_dev *vf_pdev = NULL;
+       bool done = false;
+       int index = 0;
+       int ret, i;
+
+       ret = pci_enable_sriov(pdev, num_vfs);
+       if (ret)
+               return ret;
+
+       pf->enabled_vfs = num_vfs;
+
+       while ((vf_pdev = pci_get_device(PCI_VENDOR_ID_CAVIUM, PCI_ANY_ID, vf_pdev))) {
+               if (vf_pdev->device != pf->vf_devid)
+                       continue;
+
+               octep_vdpa_assign_barspace(vf_pdev, pdev, index);
+               if (++index == num_vfs) {
+                       done = true;
+                       break;
+               }
+       }
+
+       if (done) {
+               for (i = 0; i < pf->enabled_vfs; i++)
+                       writeq(OCTEP_DEV_READY_SIGNATURE, addr + OCTEP_PF_MBOX_DATA(i));
+       }
+
+       return num_vfs;
+}
+
+static int octep_sriov_disable(struct pci_dev *pdev)
+{
+       struct octep_pf *pf = pci_get_drvdata(pdev);
+
+       if (!pci_num_vf(pdev))
+               return 0;
+
+       pci_disable_sriov(pdev);
+       pf->enabled_vfs = 0;
+
+       return 0;
+}
+
+static int octep_vdpa_sriov_configure(struct pci_dev *pdev, int num_vfs)
+{
+       if (num_vfs > 0)
+               return octep_sriov_enable(pdev, num_vfs);
+       else
+               return octep_sriov_disable(pdev);
+}
+
+static u16 octep_get_vf_devid(struct pci_dev *pdev)
+{
+       u16 did;
+
+       switch (pdev->device) {
+       case OCTEP_VDPA_DEVID_CN106K_PF:
+               did = OCTEP_VDPA_DEVID_CN106K_VF;
+               break;
+       case OCTEP_VDPA_DEVID_CN105K_PF:
+               did = OCTEP_VDPA_DEVID_CN105K_VF;
+               break;
+       case OCTEP_VDPA_DEVID_CN103K_PF:
+               did = OCTEP_VDPA_DEVID_CN103K_VF;
+               break;
+       default:
+               did = 0xFFFF;
+               break;
+       }
+
+       return did;
+}
+
+static int octep_vdpa_pf_setup(struct octep_pf *octpf)
+{
+       u8 __iomem *addr = octpf->base[OCTEP_HW_MBOX_BAR];
+       struct pci_dev *pdev = octpf->pdev;
+       int totalvfs;
+       size_t len;
+       u64 val;
+
+       totalvfs = pci_sriov_get_totalvfs(pdev);
+       if (unlikely(!totalvfs)) {
+               dev_info(&pdev->dev, "Total VFs are %d in PF sriov configuration\n", totalvfs);
+               return 0;
+       }
+
+       addr = octpf->base[OCTEP_HW_MBOX_BAR];
+       val = readq(addr + OCTEP_EPF_RINFO(0));
+       if (val == 0) {
+               dev_err(&pdev->dev, "Invalid device configuration\n");
+               return -EINVAL;
+       }
+
+       if (OCTEP_EPF_RINFO_RPVF(val) != BIT_ULL(0)) {
+               val &= ~GENMASK_ULL(35, 32);
+               val |= BIT_ULL(32);
+               writeq(val, addr + OCTEP_EPF_RINFO(0));
+       }
+
+       len = pci_resource_len(pdev, OCTEP_HW_CAPS_BAR);
+
+       octpf->vf_stride = len / totalvfs;
+       octpf->vf_devid = octep_get_vf_devid(pdev);
+
+       octep_vdpa_pf_bar_shrink(octpf);
+
+       return 0;
+}
+
+static int octep_vdpa_probe_pf(struct pci_dev *pdev)
+{
+       struct device *dev = &pdev->dev;
+       struct octep_pf *octpf;
+       int ret;
+
+       ret = pcim_enable_device(pdev);
+       if (ret) {
+               dev_err(dev, "Failed to enable device\n");
+               return ret;
+       }
+
+       ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64));
+       if (ret) {
+               dev_err(dev, "No usable DMA configuration\n");
+               return ret;
+       }
+       octpf = devm_kzalloc(dev, sizeof(*octpf), GFP_KERNEL);
+       if (!octpf)
+               return -ENOMEM;
+
+       ret = octep_iomap_region(pdev, octpf->base, OCTEP_HW_MBOX_BAR);
+       if (ret)
+               return ret;
+
+       pci_set_master(pdev);
+       pci_set_drvdata(pdev, octpf);
+       octpf->pdev = pdev;
+
+       ret = octep_vdpa_pf_setup(octpf);
+       if (ret)
+               goto unmap_region;
+
+       return 0;
+
+unmap_region:
+       octep_iounmap_region(pdev, octpf->base, OCTEP_HW_MBOX_BAR);
+       return ret;
+}
+
+static int octep_vdpa_probe(struct pci_dev *pdev, const struct pci_device_id *id)
+{
+       if (pdev->is_virtfn)
+               return octep_vdpa_probe_vf(pdev);
+       else
+               return octep_vdpa_probe_pf(pdev);
+}
+
+static struct pci_device_id octep_pci_vdpa_map[] = {
+       { PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, OCTEP_VDPA_DEVID_CN106K_PF) },
+       { PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, OCTEP_VDPA_DEVID_CN106K_VF) },
+       { PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, OCTEP_VDPA_DEVID_CN105K_PF) },
+       { PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, OCTEP_VDPA_DEVID_CN105K_VF) },
+       { PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, OCTEP_VDPA_DEVID_CN103K_PF) },
+       { PCI_DEVICE(PCI_VENDOR_ID_CAVIUM, OCTEP_VDPA_DEVID_CN103K_VF) },
+       { 0 },
+};
+
+static struct pci_driver octep_pci_vdpa = {
+       .name     = OCTEP_VDPA_DRIVER_NAME,
+       .id_table = octep_pci_vdpa_map,
+       .probe    = octep_vdpa_probe,
+       .remove   = octep_vdpa_remove,
+       .sriov_configure = octep_vdpa_sriov_configure
+};
+
+module_pci_driver(octep_pci_vdpa);
+
+MODULE_AUTHOR("Marvell");
+MODULE_DESCRIPTION("Marvell Octeon PCIe endpoint vDPA driver");
+MODULE_LICENSE("GPL");