nouveau regression: read fault PAGE_NOT_PRESENT with new fence interface

nouveau regression: read fault PAGE_NOT_PRESENT with new fence interface

[Ted Percival]

Hi, I noticed a regression in the next-20140903 kernel that was not
present in next-20140902. When Xorg starts up, the display is garbled
(or contains old image bits) and I see a page fault in the kernel log. X
is not usable in this state - there is no pointer and gdm doesn't come
up. It works fine after Xorg (gdm3) is restarted. The fault is still
present in next-20140918.

[drm] Initialized nouveau 1.2.0 20120801 for 0000:01:00.0 on minor 0
nouveau E[   PFIFO][0000:01:00.0] read fault at 0x0001260000
[PAGE_NOT_PRESENT] from PGRAPH/GPC0/TEX on channel 0x001fcd1000 [Xorg[3874]]
nouveau E[   PFIFO][0000:01:00.0] PGRAPH engine fault on channel 2,
recovering...
nouveau E[  PGRAPH][0000:01:00.0] TRAP ch 2 [0x001fcd1000 Xorg[3874]]
nouveau E[  PGRAPH][0000:01:00.0] GPC0/TPC0/TEX: 0x80000041

My hardware is: NVIDIA Corporation GF119 [Quadro NVS 4200M] (rev a1)

I bisected it down to this commit:

  commit 29ba89b2371d466ca68973525816cf10debc2655
  Author: Maarten Lankhorst <maarten.lankhorst@ubuntu.com>
  Date:   Thu Jan 9 11:03:11 2014 +0100

      drm/nouveau: rework to new fence interface

      Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
      Acked-by: Ben Skeggs <bskeggs@redhat.com>

I tried tracing the DEVICE, PGRAPH, PFIFO & PCE0 engines but nothing
obvious stood out before the read fault shown above.

Is there any other information I can provide to help track this down?
paranoia or spam level logging, for example?

Re: nouveau regression: read fault PAGE_NOT_PRESENT with new fence interface

[Maarten Lankhorst]

Hey,

Op 18-09-14 om 22:07 schreef Ted Percival:
> Hi, I noticed a regression in the next-20140903 kernel that was not
> present in next-20140902. When Xorg starts up, the display is garbled
> (or contains old image bits) and I see a page fault in the kernel log. X
> is not usable in this state - there is no pointer and gdm doesn't come
> up. It works fine after Xorg (gdm3) is restarted. The fault is still
> present in next-20140918.
>
> [drm] Initialized nouveau 1.2.0 20120801 for 0000:01:00.0 on minor 0
> nouveau E[   PFIFO][0000:01:00.0] read fault at 0x0001260000
> [PAGE_NOT_PRESENT] from PGRAPH/GPC0/TEX on channel 0x001fcd1000 [Xorg[3874]]
> nouveau E[   PFIFO][0000:01:00.0] PGRAPH engine fault on channel 2,
> recovering...
> nouveau E[  PGRAPH][0000:01:00.0] TRAP ch 2 [0x001fcd1000 Xorg[3874]]
> nouveau E[  PGRAPH][0000:01:00.0] GPC0/TPC0/TEX: 0x80000041
>
> My hardware is: NVIDIA Corporation GF119 [Quadro NVS 4200M] (rev a1)
>
> I bisected it down to this commit:
>
>   commit 29ba89b2371d466ca68973525816cf10debc2655
>   Author: Maarten Lankhorst <maarten.lankhorst@ubuntu.com>
>   Date:   Thu Jan 9 11:03:11 2014 +0100
>
>       drm/nouveau: rework to new fence interface
>
>       Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
>       Acked-by: Ben Skeggs <bskeggs@redhat.com>
>
> I tried tracing the DEVICE, PGRAPH, PFIFO & PCE0 engines but nothing
> obvious stood out before the read fault shown above.
>
> Is there any other information I can provide to help track this down?
> paranoia or spam level logging, for example?

Does this patch fix things?

diff --git a/drivers/gpu/drm/nouveau/nv84_fence.c b/drivers/gpu/drm/nouveau/nv84_fence.c
index 7b372a68aa4e..2a372b08ae44 100644
--- a/drivers/gpu/drm/nouveau/nv84_fence.c
+++ b/drivers/gpu/drm/nouveau/nv84_fence.c
@@ -159,8 +159,6 @@ nv84_fence_context_new(struct nouveau_channel *chan)
 		ret = nouveau_bo_vma_add(bo, cli->vm, &fctx->dispc_vma[i]);
 	}
 
-	nouveau_bo_wr32(priv->bo, chan->chid * 16/4, 0x00000000);
-
 	if (ret)
 		nv84_fence_context_del(chan);
 	return ret;

Re: nouveau regression: read fault PAGE_NOT_PRESENT with new fence interface

[Ted Percival]

On 09/22/2014 02:38 AM, Maarten Lankhorst wrote:
> Op 18-09-14 om 22:07 schreef Ted Percival:
>> Hi, I noticed a regression in the next-20140903 kernel that was not
>> present in next-20140902. When Xorg starts up, the display is garbled
>> (or contains old image bits) and I see a page fault in the kernel log. X
>> is not usable in this state - there is no pointer and gdm doesn't come
>> up. It works fine after Xorg (gdm3) is restarted. The fault is still
>> present in next-20140918.
>>
>> [drm] Initialized nouveau 1.2.0 20120801 for 0000:01:00.0 on minor 0
>> nouveau E[   PFIFO][0000:01:00.0] read fault at 0x0001260000
>> [PAGE_NOT_PRESENT] from PGRAPH/GPC0/TEX on channel 0x001fcd1000 [Xorg[3874]]
>> nouveau E[   PFIFO][0000:01:00.0] PGRAPH engine fault on channel 2,
>> recovering...
>> nouveau E[  PGRAPH][0000:01:00.0] TRAP ch 2 [0x001fcd1000 Xorg[3874]]
>> nouveau E[  PGRAPH][0000:01:00.0] GPC0/TPC0/TEX: 0x80000041
>>
>> My hardware is: NVIDIA Corporation GF119 [Quadro NVS 4200M] (rev a1)
> 
> Does this patch fix things?
> 
> diff --git a/drivers/gpu/drm/nouveau/nv84_fence.c b/drivers/gpu/drm/nouveau/nv84_fence.c
> index 7b372a68aa4e..2a372b08ae44 100644
> --- a/drivers/gpu/drm/nouveau/nv84_fence.c
> +++ b/drivers/gpu/drm/nouveau/nv84_fence.c
> @@ -159,8 +159,6 @@ nv84_fence_context_new(struct nouveau_channel *chan)
>  		ret = nouveau_bo_vma_add(bo, cli->vm, &fctx->dispc_vma[i]);
>  	}
>  
> -	nouveau_bo_wr32(priv->bo, chan->chid * 16/4, 0x00000000);
> -
>  	if (ret)
>  		nv84_fence_context_del(chan);
>  	return ret;
> 

Yes, this fixes the problem.