Hi Christian，
With the series patch set , amdgpu_vm_validate_pt_bos occasionally evicted amdgpu BOs failed and can’t
find the valid first busy bo . Another problem is that during the first BOs get lock period will run into deadlock .
/* check if other user occupy memory too long time */
if (!first_bo
|| !request_resv || !request_resv->lock.ctx) {
if (first_bo)
ttm_bo_put(first_bo);
return -EBUSY;
}
if (first_bo->resv == request_resv) {
ttm_bo_put(first_bo);
return -EBUSY;
}
if (ctx->interruptible)
ret = ww_mutex_lock_interruptible(&first_bo->resv->lock,
request_resv->lock.ctx);
else
ret = ww_mutex_lock(&first_bo->resv->lock, request_resv->lock.ctx);
if (ret) {
ttm_bo_put(first_bo);
if (ret == -EDEADLK) {
ret = -EAGAIN;
}
return ret;
}
Thanks
Prike
From: Christian König <ckoenig.leichtzumerken@xxxxxxxxx>
Sent: Wednesday, May 15, 2019 3:05 PM
To: Liang, Prike <Prike.Liang@xxxxxxx>; Marek Olšák <maraeo@xxxxxxxxx>
Cc: Zhou, David(ChunMing) <David1.Zhou@xxxxxxx>; dri-devel <dri-devel@xxxxxxxxxxxxxxxxxxxxx>; amd-gfx mailing list <amd-gfx@xxxxxxxxxxxxxxxxxxxxx>
Subject: Re: [PATCH 11/11] drm/amdgpu: stop removing BOs from the LRU during CS
[CAUTION: External Email]
Hi Prike,
no, that can lead to massive problems in a real OOM situation and is not something we can do here.
Christian.
Am 15.05.19 um 04:00 schrieb Liang, Prike:
Hi Christian ,
I just wonder when encounter ENOMEM error during pin amdgpu BOs can we retry validate again as below.
With the following simply patch the Abaqus pinned issue not observed.
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 11cbf63..72a32f5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -902,11 +902,15 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
bo->placements[i].lpfn = lpfn;
bo->placements[i].flags |= TTM_PL_FLAG_NO_EVICT;
}
-
+retry:
r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
if (unlikely(r)) {
- dev_err(adev->dev, "%p pin failed\n", bo);
- goto error;
+ if (r == -ENOMEM){
+ goto retry;
+ } else {
+ dev_err(adev->dev, "%p pin failed\n", bo);
+ goto error;
+ }
}
bo->pin_count = 1;
Thanks,
Prike
[CAUTION: External Email]
This series fixes the OOM errors. However, if I torture the kernel driver more, I can get it to deadlock and end up with unkillable processes. I can also get an OOM error. I just ran the test 5 times:
AMD_DEBUG=testgdsmm glxgears & AMD_DEBUG=testgdsmm glxgears & AMD_DEBUG=testgdsmm glxgears & AMD_DEBUG=testgdsmm glxgears & AMD_DEBUG=testgdsmm glxgears
This avoids OOM situations when we have lots of threads
submitting at the same time.
Signed-off-by: Christian König <christian.koenig@xxxxxxx>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index fff558cf385b..f9240a94217b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -648,7 +648,7 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
}
r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true,
- &duplicates, true);
+ &duplicates, false);
if (unlikely(r != 0)) {
if (r != -ERESTARTSYS)
DRM_ERROR("ttm_eu_reserve_buffers failed.\n");
--
2.17.1
_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
|
