Thread (1 message) 1 message, 1 author, 2021-09-16

Re: [PATCH v2 4/5] vdpa: add new vdpa attribute VDPA_ATTR_DEV_F_VERSION_1

From: Jason Wang <hidden>
Date: 2021-09-16 01:11:36
Also in: lkml

On Wed, Sep 15, 2021 at 8:12 PM Wu Zongyong
[off-list ref] wrote:
On Wed, Sep 15, 2021 at 07:08:51AM -0400, Michael S. Tsirkin wrote:
quoted
On Wed, Sep 15, 2021 at 04:06:57PM +0800, Jason Wang wrote:
quoted
On Wed, Sep 15, 2021 at 3:38 PM Michael S. Tsirkin [off-list ref] wrote:
quoted
On Wed, Sep 15, 2021 at 11:18:06AM +0800, Jason Wang wrote:
quoted
On Tue, Sep 14, 2021 at 8:58 PM Michael S. Tsirkin [off-list ref] wrote:
quoted
On Tue, Sep 14, 2021 at 08:24:51PM +0800, Wu Zongyong wrote:
quoted
This new attribute advertises whether the vdpa device is legacy or not.
Users can pick right virtqueue size if the vdpa device is legacy which
doesn't support to change virtqueue size.

Signed-off-by: Wu Zongyong <redacted>
So if we are bothering with legacy,
I think we'd better not. I guess the following may work:

1) disable the driver on BE host
2) present VERSION_1 with ACCESS_PLATFORM in get_features()
3) extend the management to advertise max_queue_size and
min_queue_size, for ENI they are the same so management layer knows it
needs to set the queue_size correctly during launching qemu

Thoughts?

Thanks
There are other subtle differences such as header size without
mergeable buffers for net.
This can be solved by mandating the feature of a mergeable buffer?

Thanks
PXE and some dpdk versions are only some of the guests that
disable mergeable buffers feature.
So what about this:

1) disable the driver on BE host
   AFAIK, there are no use cases for ENI to be used in a BE machine. So
   just disable the driver on BE machine, it will make things simper.
2) present ACCESS_PLATFORM but not VERSION_1 in get_features()
This sounds like a violation of the virtio spec. ACCESS_PLATFORM
depends on VERSION_1.
3) extend the management to advertise min_queue_size
   min_queue_size is the same as with max_queue_size for ENI.

 Another choice for 3):
   extend the management to advertise the flag F_VERSION_1 just like
   this patch
This will bring a lot of trouble, notice that the legacy/transitional
device doesn't work in several layers (both vdpa kernel and qemu).

If we can afford mandating mergeable rx buffers in the driver, it's
the most simple way.

I guess you should have the plan for the next generation ENI which
should support VERSION_1 and RING_PACKED.

Thanks
quoted
quoted
quoted
quoted
quoted
I think there are
several things to do when building the interface
- support transitional devices, that is allow userspace
  to tell device it's in legacy mode
- support reporting/setting supporting endian-ness
quoted
---
 drivers/vdpa/vdpa.c          | 6 ++++++
 drivers/virtio/virtio_vdpa.c | 7 ++++++-
 include/uapi/linux/vdpa.h    | 1 +
 3 files changed, 13 insertions(+), 1 deletion(-)
diff --git a/drivers/vdpa/vdpa.c b/drivers/vdpa/vdpa.c
index 1dc121a07a93..533d7f589eee 100644
--- a/drivers/vdpa/vdpa.c
+++ b/drivers/vdpa/vdpa.c
@@ -12,6 +12,7 @@
 #include <linux/slab.h>
 #include <linux/vdpa.h>
 #include <uapi/linux/vdpa.h>
+#include <uapi/linux/virtio_config.h>
 #include <net/genetlink.h>
 #include <linux/mod_devicetable.h>
@@ -494,6 +495,7 @@ vdpa_dev_fill(struct vdpa_device *vdev, struct sk_buff *msg, u32 portid, u32 seq
      u16 max_vq_size;
      u32 device_id;
      u32 vendor_id;
+     u64 features;
      void *hdr;
      int err;
@@ -508,6 +510,7 @@ vdpa_dev_fill(struct vdpa_device *vdev, struct sk_buff *msg, u32 portid, u32 seq
      device_id = vdev->config->get_device_id(vdev);
      vendor_id = vdev->config->get_vendor_id(vdev);
      max_vq_size = vdev->config->get_vq_num_max(vdev);
+     features = vdev->config->get_features(vdev);

      err = -EMSGSIZE;
      if (nla_put_string(msg, VDPA_ATTR_DEV_NAME, dev_name(&vdev->dev)))
@@ -520,6 +523,9 @@ vdpa_dev_fill(struct vdpa_device *vdev, struct sk_buff *msg, u32 portid, u32 seq
              goto msg_err;
      if (nla_put_u16(msg, VDPA_ATTR_DEV_MAX_VQ_SIZE, max_vq_size))
              goto msg_err;
+     if (features & BIT_ULL(VIRTIO_F_VERSION_1) &&
+         nla_put_flag(msg, VDPA_ATTR_DEV_VERSION_1))
+             goto msg_err;

      genlmsg_end(msg, hdr);
      return 0;
diff --git a/drivers/virtio/virtio_vdpa.c b/drivers/virtio/virtio_vdpa.c
index 72eaef2caeb1..1cba957c4cdc 100644
--- a/drivers/virtio/virtio_vdpa.c
+++ b/drivers/virtio/virtio_vdpa.c
@@ -7,6 +7,7 @@
  *
  */

+#include "linux/virtio_config.h"
 #include <linux/init.h>
 #include <linux/module.h>
 #include <linux/device.h>
@@ -145,6 +146,7 @@ virtio_vdpa_setup_vq(struct virtio_device *vdev, unsigned int index,
      /* Assume split virtqueue, switch to packed if necessary */
      struct vdpa_vq_state state = {0};
      unsigned long flags;
+     bool may_reduce_num = false;
      u32 align, num;
      int err;
@@ -169,10 +171,13 @@ virtio_vdpa_setup_vq(struct virtio_device *vdev, unsigned int index,
              goto error_new_virtqueue;
      }

+     if (ops->get_features(vdpa) & BIT_ULL(VIRTIO_F_VERSION_1))
+             may_reduce_num = true;
+
      /* Create the vring */
      align = ops->get_vq_align(vdpa);
      vq = vring_create_virtqueue(index, num, align, vdev,
-                                 true, true, ctx,
+                                 true, may_reduce_num, ctx,
                                  virtio_vdpa_notify, callback, name);
      if (!vq) {
              err = -ENOMEM;
diff --git a/include/uapi/linux/vdpa.h b/include/uapi/linux/vdpa.h
index 66a41e4ec163..ce0b74276a5b 100644
--- a/include/uapi/linux/vdpa.h
+++ b/include/uapi/linux/vdpa.h
@@ -32,6 +32,7 @@ enum vdpa_attr {
      VDPA_ATTR_DEV_VENDOR_ID,                /* u32 */
      VDPA_ATTR_DEV_MAX_VQS,                  /* u32 */
      VDPA_ATTR_DEV_MAX_VQ_SIZE,              /* u16 */
+     VDPA_ATTR_DEV_VERSION_1,                /* flag */

      /* new attributes must be added above here */
      VDPA_ATTR_MAX,
--
2.31.1
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help