Mesa 19.3.3 Released With Many Fixes

[tildearrow]

Have you tried the AMD Radeon Software for Linux package (the amdgpu-open)?
I suggest you either use that or update your amdgpu kernel driver...
Would be interested in outcome of those

I haven't yet, but I might try soon. I did install AMDGPU-PRO just to be sure but VA-API still uses the Mesa codepath :<

I don't think the kernel module is the issue. It has been working well with Mesa 19.0.
After updating to 5.4 the freeze still reproduces with 19.3.

[tildearrow]

Typo:

Well, that does it. I've just stepped down as typo reporter at phoronix. These typo-free articles are simply too annoying, I keep opening new articles that contain nothing. It's sad that this is still the case. (just kidding. don't worry.)

[smitty3268]

How in the heck did it break like that since Mesa 19.1?!

Well, that should be easy enough to answer with a git bisect. Pierre-Eric Pelloux-Prayer has been fixing things like this recently, so you might want to comment on any open bug reports so he sees them.

Also, you should probably actually test this release out. It contains a fix here https://cgit.freedesktop.org/mesa/me...91c7a8ec9acd0e that supposedly fixes a bug with VAAPI that was introduced 9 months ago in https://cgit.freedesktop.org/mesa/me...id=69430d7e59e.

It may not be the same problem you're having, but definitely worth a test.

[franglais125]

Well, that does it. I've just stepped down as typo reporter at phoronix. These typo-free articles are simply too annoying, I keep opening new articles that contain nothing. It's sad that this is still the case. (just kidding. don't worry.)

Now I'm wondering if anybody else got that reference :P

[tildearrow]

Well, that should be easy enough to answer with a git bisect. Pierre-Eric Pelloux-Prayer has been fixing things like this recently, so you might want to comment on any open bug reports so he sees them.

Also, you should probably actually test this release out. It contains a fix here https://cgit.freedesktop.org/mesa/me...91c7a8ec9acd0e that supposedly fixes a bug with VAAPI that was introduced 9 months ago in https://cgit.freedesktop.org/mesa/me...id=69430d7e59e.

It may not be the same problem you're having, but definitely worth a test.

OK, that is good. In fact the program in question does use vaRenderPicture for the RGB -> YUV 4:2:0/4:4:4 conversion, so this may make sense.
I'll test later after being done for the day.

Edit: I am beginning to doubt now.... because the reporter's hang didn't cause page faults... but mine did so there is a difference and now I am scared :<

[schmidtbag]

Stability issues with a later point version of 5.0.
I had to fsck my drive due to corruption in 5.0.6 if I remember.

I just was scared to upgrade... but now I am using 5.4. The hang still reproduces though :l

Well.... unless your radeonsi problem is widely acknowledged, Linuxxx is probably right that you should try updating your kernel before you point fingers at AMD's QC. You could be right, but I guess I'm saying let's not jump to conclusions.

[tildearrow]

smitty3268 Apparently Mesa 19.3.3 seems to work well again and it doesn't hang anymore so far. I still gotta do more testing, but otherwise, thank you AMD team for the fix!

Please see below now

[tildearrow]

...NOT! The freaking card still has the same piece of nonsense freeze under Mesa 19.3.3!
What a freaking lie this whole thing was. *sigh*

I was testing the recorder, and only 1 day after it was fairly stable the card turned on again, and it is getting on my nerves!!!
Seriously, who thought it was a good idea to add this "freeze" feature to the VA-API driver?! Like, come on I just want to record my gameplay, be happy and let that be it but no, it freaking crashes and it is lost and I get mad and stuff. It just goes to the point that I just feel like yelling to the computer for its sinful action of interrupting my work!

If only they cared more about REALLY fixing the bugs instead of masking them by disabling warnings, I would be happy today. But no. I am raging and I will keep raging. I could literally rage and rant for one whole hour about this...

Like who thought it was a good idea to hide warnings and pretend "that fixes it"? A warning is a warning! It means something may be wrong and the compiler is clearly telling you that "that piece of code may fail and crash and go haywire at some point", so that you get to fix it before it happens. But no, instead you decided to hide it and just say "hey, that's good". What a failure!!!!

This is what is making me lose trust on AMD. I keep having to downgrade to the latest truly working version, and then I make a freaking mess with new versions here and old versions there and then forced to recompile the WHOLE darn thing myself because some random upgrade breaks the system and then error while loading shared libraries and other crap THAT I DON'T LIKE!

How terrible... I just don't trust anything anymore...... more like Mesa 19.3.3 Released With No Fixes At All!

[tildearrow]

And no, I have no time to bisect. Plus the last time I tried bisecting it was just a disaster-fest.
I tried to bisect but I couldn't find anything on the master branch, and the problem is that since Mesa devs create new branches for every major release, I can't really find what caused the problem because the freezes began exactly on Mesa 19.1! Which means trying to cross-bisect, which I just can't do...... ughhhhhhhhhhhhhhhhhhhhhhhhh