alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - Logs https://freenode.irclog.whitequark.org/panfrost - <daniels> avoiding X is a huge feature
* urjaman bisected the Xorg/glamor hitting that assert i mentioned before
<urjaman> first bad commit: [b8c4fb235ef4055a14a9a2aec07f3f906ef8a841] pan/midgard: Implement SIMD-aware dead code elimination
<urjaman> "Results are meh." indeed xD
rhyskidd has joined #panfrost
rhyskidd has quit [Ping timeout: 276 seconds]
rhyskidd has joined #panfrost
<alyssa> urjaman: Quiet you ;p
<alyssa> urjaman: Mayhaps what HdkR said :)
<HdkR> I'm just crazy old man, I don't know what you're talking about
* alyssa is one of those three things
rhyskidd has quit [Ping timeout: 265 seconds]
_whitelogger has joined #panfrost
rhyskidd has joined #panfrost
* urjaman is maybe 2 of those things :P
<urjaman> anyways what i was trying to say before I updated mesa and crashed Xorg by breathing at it (yeah any interaction with the top bar, and alt-tabbing to another window could cause it)
<urjaman> was that somehow the scrolling-segfault (was basically if you scroll a gtk app, included mousepad, thunar, transmission) has at some point disappeared and instead is an odd graphical glitch on scroll
rhyskidd has quit [Ping timeout: 265 seconds]
rhyskidd has joined #panfrost
rhyskidd has quit [Ping timeout: 265 seconds]
davidlt has joined #panfrost
rhyskidd has joined #panfrost
rhyskidd has quit [Ping timeout: 265 seconds]
rhyskidd has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
fysa has joined #panfrost
fysa has quit [Remote host closed the connection]
rhyskidd has joined #panfrost
megi has quit [Ping timeout: 265 seconds]
davidlt_ has joined #panfrost
rhyskidd has quit [Ping timeout: 265 seconds]
davidlt has quit [Ping timeout: 268 seconds]
davidlt_ has quit [Ping timeout: 240 seconds]
davidlt has joined #panfrost
rhyskidd has joined #panfrost
rhyskidd has quit [Ping timeout: 240 seconds]
fysa has joined #panfrost
fysa has quit [Ping timeout: 245 seconds]
rhyskidd has joined #panfrost
<tomeu> robmur01: for some reason, rc4 isn't working on the veyron: https://gitlab.freedesktop.org/tomeu/mesa/-/jobs/791418
<tomeu> will try to reproduce locally
rhyskidd has quit [Ping timeout: 246 seconds]
fysa has joined #panfrost
fysa has quit [Ping timeout: 265 seconds]
rhyskidd has joined #panfrost
<tomeu> couldn't reproduce with the old mesa I have around here, so it could be related to the parallelization changes
<tomeu> rebuilding mesa now
<tomeu> bbrezillon: any ideas?
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
paulk-leonov has quit [Ping timeout: 240 seconds]
<bbrezillon> tomeu: wow
<bbrezillon> tomeu: 5.4-rc4 + mesa master?
raster has joined #panfrost
<tomeu> bbrezillon: yeah, but I cannot reproduce it in this veyron I have here
<tomeu> have run 7 instances of deqp in parallel and all seems fine
<bbrezillon> tomeu: what's the old kernel version, 5.2?
<bbrezillon> it's definitely a kernel bug, but it might be triggered by my batch parallelization changes, indeed
<tomeu> bbrezillon: master is using 5.3.0-rc8 atm
<bbrezillon> tomeu: and it works fine, right? I guess it's worth looking at the panfrost changes that were merged between 5.3 and 5.4-rc4
<bbrezillon> I don't have time to look at it this week, but I can give it a try after ELCE
paulk-leonov has joined #panfrost
rhyskidd has quit [Ping timeout: 245 seconds]
<tomeu> bbrezillon: will do that, but I guess what would be most important is to have a way for me to reproduce
<tomeu> guess I will try with using the same kernel as gitlab
RCF is now known as rcf
<tomeu> bbrezillon: ok, managed to reproduce with the kernel from gitlab
<tomeu> looks like the GPU isn't powered on
<tomeu> [ 50.643093] panfrost ffa30000.gpu: gpu sched timeout, js=1, config=0x0, status=0x0, head=0x0, tail=0x0, sched_job=a1cce9cf
<tomeu> [ 51.163069] panfrost ffa30000.gpu: gpu sched timeout, js=0, config=0x0, status=0x0, head=0x0, tail=0x0, sched_job=930af481
<tomeu> as we're reading zeroes from the job registers
wens has joined #panfrost
fysa has joined #panfrost
rhyskidd has joined #panfrost
<tomeu> robher: I have a vague recollection of talking about such a regression on veyron at plumbers
Green has joined #panfrost
fysa has quit [Ping timeout: 240 seconds]
<wens> what regression? just did a bisect on veyron-speedy
<wens> looks like the regression I reported awhile back
<bbrezillon> tomeu: I wonder how it can trigger a kernel oops though
<wens> got it down to a handful of commits related to runtime PM
<wens> before commit 635430797d3f drm/panfrost: Rework runtime PM initialization, everything worked fine
<wens> the subsequent commits made X just crash, and after that bunch, the gpu timeout regressions
<wens> that was with some month old master branch of mesa. latest HEAD makes X crash.
<wens> about to drop back to 19.2
raster has quit [Quit: Gettin' stinky!]
<tomeu> wens: ah, it indeed seems to be it
<tomeu> I'm going to dig further on it, thanks for the info
<tomeu> awesome, managed to reproduce locally
<tomeu> seems to be dependent on the kconfig
rhyskidd has quit [Ping timeout: 265 seconds]
<urjaman> wens: if you want to run latest HEAD (or atleast as of yesterday-ish) in Xorg, revert b8c4fb235ef4055a14a9a2aec07f3f906ef8a841
<urjaman> (of mesa that is)
<wens> tomeu: you want my kernel config?
<tomeu> wens: no need, was able to reproduce it
<tomeu> thanks
raster has joined #panfrost
rhyskidd has joined #panfrost
<tomeu> wens: guess you don't build panfrost as a module?
megi has joined #panfrost
<tomeu> looks like, when panfrost is built-in, the pm_runtime callbacks aren't being called
rhyskidd has quit [Ping timeout: 265 seconds]
<wens> tomeu: no, it's easier to have most or all drivers built-in, and just sign the zImage
<wens> I still have chromeos on my device
warpme_ has joined #panfrost
<tomeu> ok, that's why
<urjaman> why built-in or why pm callbacks not being called? :P
<wens> urjaman: why only some people run into it? :)
<tomeu> pm runtime callbacks not being called seems to be related to the driver being built-in
<tomeu> probably because it's probed before something else
rhyskidd has joined #panfrost
BenG83 has quit [Ping timeout: 252 seconds]
BenG83 has joined #panfrost
BenG83 has quit [Ping timeout: 240 seconds]
<tomeu> wens: if you give me a name and email address, I will aatribute you the reporting of the bug
* tomeu has a one-liner to fix the kernel
BenG83 has joined #panfrost
<tomeu> wens: np, I have already found you :)
<wens> my name is probably the easiet to find in the kernel. it's in the maintainer's PGP guide :p
<wens> tomeu: I'll give it a spin tomorrow
<tomeu> nice!
<alyssa> urjaman: FWIW I have the wacky scroll glitch reproduced locally
<wens> tomeu: why is it v2 though?
<tomeu> wens: because I have been using gitlab mrs for too long a time and I botched git-sendmail
<tomeu> and I left your reported-by
<alyssa> I miss mailing lists.
<alyssa> Review on MRs is just so uncomfortable
<alyssa> I did learn about appending .patch to the URL (which is what I do with GitHub) yesterday, so that makes MR review massively less annoying
<alyssa> Can just open that all up locally and review like email
<tomeu> maybe after a while you will miss gitlab when having to send patches :p
<alyssa> Hmmm?
<alyssa> ML is more convenient in both directions
<alyssa> Only thing gitlab wins on is ease of applying other people's patches
<alyssa> and now the CI stuff
<tomeu> just mentioning that convenience is related to habit as well
<alyssa> GitLab is literally slower though
<TheCycoONE> heh
<alyssa> There. I downloaded the patch with .patch
<alyssa> Edited in vim like I was respondong to mail
<alyssa> and pasted my responses in ```tags```
<alyssa> This Is Fine
<tomeu> everybody happy!
<alyssa> \o/
<alyssa> Compromise! :p
fysa has joined #panfrost
fysa has quit [Ping timeout: 276 seconds]
fysa has joined #panfrost
<tomeu> narmstrong: btw, are you planning to re-enable lima CI?
fysa has quit [Ping timeout: 265 seconds]
<narmstrong> tomeu: I should
fysa has joined #panfrost
fysa has quit [Ping timeout: 240 seconds]
enunes has quit [Read error: Connection reset by peer]
enunes has joined #panfrost
vstehle has quit [Quit: WeeChat 2.6]
vstehle has joined #panfrost
fysa has joined #panfrost
fysa has quit [Remote host closed the connection]
fysa has joined #panfrost
<narmstrong> tomeu: i will, when more h3 boards are hooked on the lab
<anarsoul> narmstrong: tomeu: what about these lafrite boards?
<narmstrong> anarsoul: jobs are sent to potato board, which has the same gpu and twice memory
<anarsoul> I thought potato had midgard
<anarsoul> anyway, lafrite is more than enough to run deqp
<narmstrong> We have more potato boards now in the lab, but whatever they share the same die but only the package differs between s905x and s805x
stikonas has joined #panfrost
<anarsoul> narmstrong: I see
raster has quit [Quit: Gettin' stinky!]
raster has joined #panfrost
adjtm has quit [Ping timeout: 252 seconds]
TheKit has quit [Remote host closed the connection]
enunes has quit [Ping timeout: 268 seconds]
enunes has joined #panfrost
adjtm has joined #panfrost
janrinze has quit [Remote host closed the connection]
davidlt has quit [Ping timeout: 276 seconds]
TheKit has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
* urjaman has maybe figured out something about the memory corruption on exit of openscad
<urjaman> the valgrind run effectively said that the block it tries to touch (during context destruction) was freed during a previous context destruction
<urjaman> so i put a breakpoint on the context destruction and yeah openscad opens atleast 3 contexts, and the canary assert happens on the third context destruction...