Re: [PATCH v10 3/9] drm/i915: vgpu pv command buffer support

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Quoting Xiaolin Zhang (2019-09-17 06:48:14)
> based on the shared memory setup between guest and GVT, the simple
> pv command buffer ring was introduced by this patch used to perform
> guest-2-gvt single direction communication.
> 
> v1: initial support, added to address i915 PV v6 patch set comment.
> 
> Signed-off-by: Xiaolin Zhang <xiaolin.zhang@xxxxxxxxx>
> ---
>  drivers/gpu/drm/i915/i915_pvinfo.h |   1 +
>  drivers/gpu/drm/i915/i915_vgpu.c   | 193 +++++++++++++++++++++++++++++++++++++
>  drivers/gpu/drm/i915/i915_vgpu.h   |  66 +++++++++++++
>  3 files changed, 260 insertions(+)
> 
> diff --git a/drivers/gpu/drm/i915/i915_pvinfo.h b/drivers/gpu/drm/i915/i915_pvinfo.h
> index 3c63603..db9eebb 100644
> --- a/drivers/gpu/drm/i915/i915_pvinfo.h
> +++ b/drivers/gpu/drm/i915/i915_pvinfo.h
> @@ -49,6 +49,7 @@ enum vgt_g2v_type {
>         VGT_G2V_EXECLIST_CONTEXT_CREATE,
>         VGT_G2V_EXECLIST_CONTEXT_DESTROY,
>         VGT_G2V_SHARED_PAGE_SETUP,
> +       VGT_G2V_PV_SEND_TRIGGER,
>         VGT_G2V_MAX,
>  };
>  
> diff --git a/drivers/gpu/drm/i915/i915_vgpu.c b/drivers/gpu/drm/i915/i915_vgpu.c
> index 68fecfd..6e29a52 100644
> --- a/drivers/gpu/drm/i915/i915_vgpu.c
> +++ b/drivers/gpu/drm/i915/i915_vgpu.c
> @@ -322,6 +322,187 @@ int intel_vgt_balloon(struct i915_ggtt *ggtt)
>   * i915 vgpu PV support for Linux
>   */
>  
> +/**
> + * wait_for_desc_update - Wait for the command buffer descriptor update.
> + * @desc:      buffer descriptor
> + * @fence:     response fence
> + * @status:    placeholder for status
> + *
> + * GVT will update command buffer descriptor with new fence and status
> + * after processing the command identified by the fence. Wait for
> + * specified fence and then read from the descriptor status of the
> + * command.
> + *
> + * Return:
> + * *   0 response received (status is valid)
> + * *   -ETIMEDOUT no response within hardcoded timeout
> + * *   -EPROTO no response, CT buffer is in error
> + */
> +static int wait_for_desc_update(struct vgpu_pv_ct_buffer_desc *desc,
> +               u32 fence, u32 *status)
> +{
> +       int err;
> +
> +#define done (READ_ONCE(desc->fence) == fence)
> +       err = wait_for_us(done, 5);
> +       if (err)
> +               err = wait_for(done, 10);
> +#undef done
> +
> +       if (unlikely(err)) {
> +               DRM_ERROR("CT: fence %u failed; reported fence=%u\n",
> +                               fence, desc->fence);
> +       }
> +
> +       *status = desc->status;
> +       return err;
> +}
> +
> +/**
> + * DOC: CTB Guest to GVT request
> + *
> + * Format of the CTB Guest to GVT request message is as follows::
> + *
> + *      +------------+---------+---------+---------+---------+
> + *      |   msg[0]   |   [1]   |   [2]   |   ...   |  [n-1]  |
> + *      +------------+---------+---------+---------+---------+
> + *      |   MESSAGE  |       MESSAGE PAYLOAD                 |
> + *      +   HEADER   +---------+---------+---------+---------+
> + *      |            |    0    |    1    |   ...   |    n    |
> + *      +============+=========+=========+=========+=========+
> + *      |  len >= 1  |  FENCE  |     request specific data   |
> + *      +------+-----+---------+---------+---------+---------+
> + *
> + *                   ^-----------------len-------------------^
> + */
> +static int pv_command_buffer_write(struct i915_virtual_gpu_pv *pv,
> +               const u32 *action, u32 len /* in dwords */, u32 fence)
> +{
> +       struct vgpu_pv_ct_buffer_desc *desc = pv->ctb.desc;
> +       u32 head = desc->head / 4;      /* in dwords */
> +       u32 tail = desc->tail / 4;      /* in dwords */
> +       u32 size = desc->size / 4;      /* in dwords */
> +       u32 used;                       /* in dwords */
> +       u32 header;
> +       u32 *cmds = pv->ctb.cmds;
> +       unsigned int i;
> +
> +       GEM_BUG_ON(desc->size % 4);
> +       GEM_BUG_ON(desc->head % 4);
> +       GEM_BUG_ON(desc->tail % 4);
> +       GEM_BUG_ON(tail >= size);
> +
> +       /*
> +        * tail == head condition indicates empty.
> +        */
> +       if (tail < head)
> +               used = (size - head) + tail;
> +       else
> +               used = tail - head;
> +
> +       /* make sure there is a space including extra dw for the fence */
> +       if (unlikely(used + len + 1 >= size))
> +               return -ENOSPC;
> +
> +       /*
> +        * Write the message. The format is the following:
> +        * DW0: header (including action code)
> +        * DW1: fence
> +        * DW2+: action data
> +        */
> +       header = (len << PV_CT_MSG_LEN_SHIFT) |
> +                (PV_CT_MSG_WRITE_FENCE_TO_DESC) |
> +                (action[0] << PV_CT_MSG_ACTION_SHIFT);
> +
> +       cmds[tail] = header;
> +       tail = (tail + 1) % size;
> +
> +       cmds[tail] = fence;
> +       tail = (tail + 1) % size;
> +
> +       for (i = 1; i < len; i++) {
> +               cmds[tail] = action[i];
> +               tail = (tail + 1) % size;
> +       }
> +
> +       /* now update desc tail (back in bytes) */

wmb();

Answers on a postcard as to why this is required and where the
complementary rmb/mb should be.

> +       desc->tail = tail * 4;
> +       GEM_BUG_ON(desc->tail > desc->size);
> +
> +       return 0;
> +}
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx




[Index of Archives]     [AMD Graphics]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux