Bug 74784

Please attach your xorg log and dmesg output.  Are using using the llvm backend in r600g?  If so, does disabling it help?  Also, if this is a regression, can you bisect?

Yes, I am using the llvm backend (you mean the --enable-r600-llvm-compiler configure option, don't you? )

I will compile a mesa package without that option, and tonight I will see if it continues to happend and I will attach the xorg and dmesg logs.

And yes, it is a regression, although the best date I can give as not happening is the 29th of December.

If disabling the llvm backend doesn't help, I will start bisecting.

Thanks for all!

Created attachment 93795 [details]
dmesg log

Theres like 440 lines that look like this:

radeon 0000:01:00.0: evergreen_cs_track_validate_cb:413 cb invalid format 0 for 1 (0x00000000)
radeon 0000:01:00.0: evergreen_packet3_check:1918 invalid cmd s[drm:radeon_cs_ib_chunk] *ERROR* Invalid command stream !

Created attachment 93796 [details]
xorg log

This may be a duplicate of bug 68451.  Make sure your kernel has this patch:

drm/radeon: skip colorbuffer checking if COLOR_INFO.FORMAT is set to INVALID

No, Arch doesn't seem to apply that patch. But I am still using kernel 3.12.9, should it be there too?

Also, that bug report was reopened on the 19 Dec, but I think (I may be wrong) I recompiled mesa after that and didn't see anything wrong.

And compiling without --enable-r600-llvm-compiler doesn't help either.

If I can help with anything else, just ask.

Thanks

With mesa at commit 6d1cecb (Friday evening) this problem is fixed.

Thanks!

P.D: Maybe it was related to the hyperz problem of r600g drivers, as I am using a revision that is after this commit http://cgit.freedesktop.org/mesa/mesa/commit/?id=01e637114914453451becc0dc8afe60faff48d84

(In reply to comment #7)
> With mesa at commit 6d1cecb (Friday evening) this problem is fixed.
> 
> Thanks!
> 
> P.D: Maybe it was related to the hyperz problem of r600g drivers, as I am
> using a revision that is after this commit
> http://cgit.freedesktop.org/mesa/mesa/commit/
> ?id=01e637114914453451becc0dc8afe60faff48d84

Could you set the following environment variable to re-enable hyperz?

R600_DEBUG=hyperz

With 'R600_DEBUG=hyperz' the problem happens again, so I'm afraid that it's something related to the hyperz.

If I can help with anything else, just ask.

The bug is not fixed, HyperZ has only been disabled by default. Please, keep opened to track the problem until HyperZ is fixed.

Could you please test if this is fixed with current Mesa git?

Hi,

I have just tried with 'R600_DEBUG=hyperz' to enable hyperz with mesa built from yesterday master.

I have played a match with a hero that used to cause failures with some of his abilities, but as far as I can tell, there has been no problem.

So it seems that this bug has been resolved (^_^)

If you need anything else just ask.

Thanks!