Re: [PATCH] vdpa/mlx5: Add support for doorbell bypassing

From: Jason Wang
Date: Thu Apr 22 2021 - 00:52:27 EST



在 2021/4/22 上午11:27, Mika Penttilä 写道:


On 22.4.2021 5.37, Jason Wang wrote:

在 2021/4/21 下午6:41, Eli Cohen 写道:
Implement mlx5_get_vq_notification() to return the doorbell address.
Size is set to one system page as required.

Signed-off-by: Eli Cohen <elic@xxxxxxxxxx>
---
  drivers/vdpa/mlx5/core/mlx5_vdpa.h | 1 +
  drivers/vdpa/mlx5/core/resources.c | 1 +
  drivers/vdpa/mlx5/net/mlx5_vnet.c  | 6 ++++++
  3 files changed, 8 insertions(+)

diff --git a/drivers/vdpa/mlx5/core/mlx5_vdpa.h b/drivers/vdpa/mlx5/core/mlx5_vdpa.h
index b6cc53ba980c..49de62cda598 100644
--- a/drivers/vdpa/mlx5/core/mlx5_vdpa.h
+++ b/drivers/vdpa/mlx5/core/mlx5_vdpa.h
@@ -41,6 +41,7 @@ struct mlx5_vdpa_resources {
      u32 pdn;
      struct mlx5_uars_page *uar;
      void __iomem *kick_addr;
+    u64 phys_kick_addr;
      u16 uid;
      u32 null_mkey;
      bool valid;
diff --git a/drivers/vdpa/mlx5/core/resources.c b/drivers/vdpa/mlx5/core/resources.c
index 6521cbd0f5c2..665f8fc1710f 100644
--- a/drivers/vdpa/mlx5/core/resources.c
+++ b/drivers/vdpa/mlx5/core/resources.c
@@ -247,6 +247,7 @@ int mlx5_vdpa_alloc_resources(struct mlx5_vdpa_dev *mvdev)
          goto err_key;
        kick_addr = mdev->bar_addr + offset;
+    res->phys_kick_addr = kick_addr;
        res->kick_addr = ioremap(kick_addr, PAGE_SIZE);
      if (!res->kick_addr) {
diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
index 10c5fef3c020..680751074d2a 100644
--- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
+++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
@@ -1865,8 +1865,14 @@ static void mlx5_vdpa_free(struct vdpa_device *vdev)
    static struct vdpa_notification_area mlx5_get_vq_notification(struct vdpa_device *vdev, u16 idx)
  {
+    struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
      struct vdpa_notification_area ret = {};
+    struct mlx5_vdpa_net *ndev;
+
+    ndev = to_mlx5_vdpa_ndev(mvdev);
  +    ret.addr = (phys_addr_t)ndev->mvdev.res.phys_kick_addr;
+    ret.size = PAGE_SIZE;


Note that the page will be mapped in to guest, so it's only safe if the doorbeel exclusively own the page. This means if there're other registers in the page, we can not let the doorbell bypass to work.

So this is suspicious at least in the case of subfunction where we calculate the bar length in mlx5_sf_dev_table_create() as:

table->sf_bar_length = 1 << (MLX5_CAP_GEN(dev, log_min_sf_size) + 12);

It looks to me this can only work for the arch with PAGE_SIZE = 4096, otherwise we can map more into the userspace(guest).

Thanks


Is there support as of today (in qemu  or elsewhere) to use this mmap doorbell instead of the traditional kick.


Vhost-user had already had this support in Qemu (see set_host_notifier_mr).

And I've posted patches to support this for qemu[1] and vp_vdpa[2] driver.

Thanks

[1] https://www.mail-archive.com/qemu-devel@xxxxxxxxxx/msg798935.html

[2] https://lore.kernel.org/virtualization/20210415073147.19331-5-jasowang@xxxxxxxxxx/T/



--Mika




      return ret;
  }