Bug 105884

Created attachment 138790 [details]
Another crash

This is definitely a regression in kernel 4.15 because I've never had this issue before.

It is possible that kernel 4.15.9 didn't have this issue, so a regression window is quite small.

Created attachment 138797 [details]
X.org.log

Seemingly there's nothing interesting in the X.org log.

4.15 had a number of things added to it.

One thing is using hugepages in ttm, which a lot of people have reported issues with. Try disabling CONFIG_SWIOTLB (not really needed on modern hw, afaik).

Also nouveau's vm got a rewrite. Some fixes needed to be applied, I think they're all in 4.16, not as sure about 4.15.x.

The start of your errors is

[ 7661.417755] nouveau 0000:01:00.0: fifo: PBDMA0: 00040000 [PBENTRY] ch 7 [017f3cc000 firefox[19007]] subc 0 mthd 0000 data 00000000
[ 7661.417774] nouveau 0000:01:00.0: gr: TRAP ch 7 [017f3cc000 firefox[19007]]
[ 7661.417780] nouveau 0000:01:00.0: gr: DISPATCH 80000002 [CLASS_SUBCH_MISMATCH]
[ 7661.417793] nouveau 0000:01:00.0: fifo: PBDMA0: 00200000 [METHOD] ch 7 [017f3cc000 firefox[19007]] subc 0 mthd 000c data 88888888

which doesn't look extremely familiar. "FIFO is confused", after which everything goes downhill pretty fast.

(In reply to Ilia Mirkin from comment #4)
> 4.15 had a number of things added to it.
> 
> One thing is using hugepages in ttm, which a lot of people have reported
> issues with. Try disabling CONFIG_SWIOTLB (not really needed on modern hw,
> afaik).
> 
> Also nouveau's vm got a rewrite. Some fixes needed to be applied, I think
> they're all in 4.16, not as sure about 4.15.x.
> 

I'd be glad to disable this option but I'm on Fedora 27 and I use the distro kernel, so I'm not sure how I can do that. Aren't you employed by RedHat/Fedora, Ilia? I guess it's in your power to raise this issue internally.

(In reply to Artem S. Tashkinov from comment #5)
> (In reply to Ilia Mirkin from comment #4)
> > 4.15 had a number of things added to it.
> > 
> > One thing is using hugepages in ttm, which a lot of people have reported
> > issues with. Try disabling CONFIG_SWIOTLB (not really needed on modern hw,
> > afaik).
> > 
> > Also nouveau's vm got a rewrite. Some fixes needed to be applied, I think
> > they're all in 4.16, not as sure about 4.15.x.
> > 
> 
> I'd be glad to disable this option but I'm on Fedora 27 and I use the distro
> kernel, so I'm not sure how I can do that. Aren't you employed by
> RedHat/Fedora, Ilia? I guess it's in your power to raise this issue
> internally.

I don't think so. Perhaps you know something I don't?

Either way, this is the upstream bug system. If you have an issue with $distro, bring up the problem with $distro.

(In reply to Ilia Mirkin from comment #6)
> 
> Either way, this is the upstream bug system. If you have an issue with
> $distro, bring up the problem with $distro.

Does "iommu=off" disable SWIOTLB?

This happens under 4.16 as well but a lot less frequently (maybe because I switched to Google Chrome to watch youtube):

May 11 14:17:06 localhost kernel: nouveau 0000:01:00.0: fifo: PBDMA0: 00040000 [PBENTRY] ch 9 [017ed17000 firefox[14167]] subc 0 mthd 0000 data 00000000
May 11 14:17:06 localhost kernel: nouveau 0000:01:00.0: fifo: PBDMA0: 00040000 [PBENTRY] ch 9 [017ed17000 firefox[14167]] subc 0 mthd 0000 data 00000000
May 11 14:17:06 localhost kernel: nouveau 0000:01:00.0: fifo: PBDMA0: 00040000 [PBENTRY] ch 9 [017ed17000 firefox[14167]] subc 0 mthd 0000 data 00000000

Created attachment 139854 [details]
dmesg again

[14443.603553] kauditd_printk_skb: 51 callbacks suppressed

Probably related bug 99202

-- GitLab Migration Automatic Message --

This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/xorg/driver/xf86-video-nouveau/issues/419.