Re: 3.18-rc regression: drm/nouveau: use shared fences for readable objects

From: Maarten Lankhorst
Date: Thu Nov 20 2014 - 03:41:29 EST


Op 20-11-14 om 00:08 schreef Tobias Klausmann:
> On 19.11.2014 09:10, Maarten Lankhorst wrote:
>> ...
>> On the EDITED patch from fixed-fences-for-bisect, can you do the following:
>>
>> In nouveau/nv84_fence.c function nv84_fence_context_new, remove
>>
>> fctx->base.sequence = nv84_fence_read(chan);
>>
>> and add back
>>
>> nouveau_bo_wr32(priv->bo, chan->chid * 16/4, 0x00000000);
>>
>> ...
>
> Added the above on top of your "fixed-fences-for-bisect" branch and guessed it would work, but did not :/
> Anyway, as this "initializes" the fence to a known state, maybe you should consider pushing that.
Hey,

There is a reason I don't set it to a known state on nv84+.

Channel 2 is created, fence seqno ends up being 100, other channel waits on seqno reaching 100.
Channel 2 is destroyed, and immediately recreated. Seqno is reset to 0.
Other channel waits for channel 2's seqno being 100.

The other channel can keep waiting indefinitely.

I guess it might be useful to reset the contents of the bo to zero on allocation, but it should not be done in fence_context_new.

~Maarten

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/