I suggest to rather do this in amdgpu_ib_get() instead.
As soon as the size is specified we get some memory from the SA
allocator and should also set the flag to be on the save side.
Thanks,
Christian.
Am 20.04.21 um 08:55 schrieb Su, Jinzhou (Joe):
[AMD Official Use Only - Internal Distribution Only]
Hello Christian,
Thanks for your information. We add this due to sdma0 timeout during S3 stress test.
Will update this for other cases in v2.
Regards,
Joe
-----Original Message-----
From: Christian König <ckoenig.leichtzumerken@xxxxxxxxx>
Sent: Tuesday, April 20, 2021 2:39 PM
To: Su, Jinzhou (Joe) <Jinzhou.Su@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx
Cc: Huang, Ray <Ray.Huang@xxxxxxx>
Subject: Re: [PATCH] drm/amdgpu: Add mem sync flag for SDMA IB test
Am 20.04.21 um 04:23 schrieb Jinzhou Su:
The buffer for SDMA IB test is allocated by sa bo which may be used by
other purpose. Better to flush the cache before commit the IB.
Good point, but shouldn't we do this for a lot of other cases as well?
I think the only place where we should not set the flag is in the CS IOCTL.
Regards,
Christian.
Signed-off-by: Jinzhou Su <Jinzhou.Su@xxxxxxx>
---
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c
b/drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c
index b1ad9e52b234..da67f440b102 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c
@@ -1000,6 +1000,7 @@ static int sdma_v5_2_ring_test_ib(struct amdgpu_ring *ring, long timeout)
ib.ptr[7] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
ib.length_dw = 8;
+ ib.flags = AMDGPU_IB_FLAG_EMIT_MEM_SYNC;
r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
if (r)
goto err1;
_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
