Am 03.05.2017 um 05:48 schrieb Monk Liu:
> if sriov gpu reset is invoked by job timeout, it is run
> in a global work-queue which is very slow and better not call
> msleep ortherwise it takes long time to get back CPU.
>
> so make below changes:
>
> 1: Change msleep 1 to mdelay 5
> 2: Ignore the ack fail from pf after time out,
> because VF FLR will clear ack, sometime VF FLR is done
> prior to the beginning of poll_ack so we can ignore this ack
>
> TODO:
> Put job_timedout (and the following gpu reset) in a driver thread,
> instead of the global work_struct.
>
> Change-Id: I4608c67b55c67c88597e03eee35a126d7e850839
> Signed-off-by: Monk Liu <Monk.Liu at amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c | 18 ++++++++++--------
> drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c | 10 +++++-----
> 2 files changed, 15 insertions(+), 13 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c b/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c
> index 712f36e..e967a7b 100644
> --- a/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c
> +++ b/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c
> @@ -124,8 +124,8 @@ static int xgpu_ai_poll_ack(struct amdgpu_device *adev)
> r = -ETIME;
> break;
> }
> - msleep(1);
> - timeout -= 1;
> + mdelay(5);
> + timeout -= 5;
>
> reg = RREG32_NO_KIQ(SOC15_REG_OFFSET(NBIO, 0,
> mmBIF_BX_PF0_MAILBOX_CONTROL));
> @@ -141,12 +141,12 @@ static int xgpu_ai_poll_msg(struct amdgpu_device *adev, enum idh_event event)
> r = xgpu_ai_mailbox_rcv_msg(adev, event);
> while (r) {
> if (timeout <= 0) {
> - pr_err("Doesn't get ack from pf.\n");
> + pr_err("Doesn't get msg:%d from pf.\n", event);
> r = -ETIME;
> break;
> }
> - msleep(1);
> - timeout -= 1;
> + mdelay(5);
> + timeout -= 5;
>
> r = xgpu_ai_mailbox_rcv_msg(adev, event);
> }
> @@ -165,7 +165,7 @@ static int xgpu_ai_send_access_requests(struct amdgpu_device *adev,
> /* start to poll ack */
> r = xgpu_ai_poll_ack(adev);
> if (r)
> - return r;
> + pr_err("Doesn't get ack from pf, continue\n");
>
> xgpu_ai_mailbox_set_valid(adev, false);
>
> @@ -174,8 +174,10 @@ static int xgpu_ai_send_access_requests(struct amdgpu_device *adev,
> req == IDH_REQ_GPU_FINI_ACCESS ||
> req == IDH_REQ_GPU_RESET_ACCESS) {
> r = xgpu_ai_poll_msg(adev, IDH_READY_TO_ACCESS_GPU);
> - if (r)
> + if (r) {
> + pr_err("Doesn't get READY_TO_ACCESS_GPU from pf, give up\n");
> return r;
> + }
> }
>
> return 0;
> @@ -211,7 +213,7 @@ static int xgpu_ai_mailbox_ack_irq(struct amdgpu_device *adev,
> struct amdgpu_irq_src *source,
> struct amdgpu_iv_entry *entry)
> {
> - DRM_DEBUG("get ack intr and do nothing.\n");
> + printk("get ack intr and do nothing.\n");
Changing a DRM_DEBUG to a printk looks odd. How about using pr_warn or
pr_info instead?
Apart from that patch looks good to me, but I don't deeply into that stuff.
So with the printk fixed feel free to add an Acked-by: Christian König
<christian.koenig at amd.com> to it.
Regards,
Christian.
> return 0;
> }
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c b/drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c
> index 7bdc51b..f0d64f1 100644
> --- a/drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c
> +++ b/drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c
> @@ -398,8 +398,8 @@ static int xgpu_vi_poll_ack(struct amdgpu_device *adev)
> r = -ETIME;
> break;
> }
> - msleep(1);
> - timeout -= 1;
> + mdelay(5);
> + timeout -= 5;
>
> reg = RREG32_NO_KIQ(mmMAILBOX_CONTROL);
> }
> @@ -418,8 +418,8 @@ static int xgpu_vi_poll_msg(struct amdgpu_device *adev, enum idh_event event)
> r = -ETIME;
> break;
> }
> - msleep(1);
> - timeout -= 1;
> + mdelay(5);
> + timeout -= 5;
>
> r = xgpu_vi_mailbox_rcv_msg(adev, event);
> }
> @@ -447,7 +447,7 @@ static int xgpu_vi_send_access_requests(struct amdgpu_device *adev,
> request == IDH_REQ_GPU_RESET_ACCESS) {
> r = xgpu_vi_poll_msg(adev, IDH_READY_TO_ACCESS_GPU);
> if (r)
> - return r;
> + pr_err("Doesn't get ack from pf, continue\n");
> }
>
> return 0;
