07:58pq: linyaa, are you assuming that the VK implementation has access to read the KMS "hardware clock" at will when saying VK can convert the timestamps to a common time base better than the kernel driver?
08:00pq: Shouldn't time base conversion happen as close in time to the time instant of the timestamp as possible? Who would keep track of the drift between the two clocks?
08:32sima: emersion, MrCooper btw did any doc patch come out of the recent-ish discussions on how/why/when compositors should retry atomic test to find a good config?
08:39emersion: i don't think so
09:00mripard: sima: hey, could you have a look at the locking issue I mentionned https://lore.kernel.org/r/20250129-test-kunit-v2-0-fe59c43805d5@kernel.org ?
09:00sima: mripard, will try, bit a backlog of tricky stuff I need to get to atm
09:01sima: but I think I already promised you to look and failed :-/
09:01MrCooper: sima: haven't seen any (though I haven't caught up from being sick at the end of last week), and no bandwidth to work on that myself
09:05mripard: sima: you didn't :)
09:05mripard: but ack, take your time :)
11:10sima: mripard, I think you need a kunit_add_synchronous_actio_or_reset or something like that
11:11sima: and always enable CONFIG_DEBUG_WW_MUTEX_SLOWPATH when touching any gem/dma_resv/modeset locking
11:12sima: it's pretty much guaranteed broken otherwise
11:13sima: (also counts for anything CI)
11:14sima: daniels, hm helen koike not around here? for the above
11:15daniels: sima: vignesh is handling DRM CI these days
11:15daniels: vignesh: ^
11:15sima: ah cool
11:18sima: mripard, I guess interim the only real fix is to drop the kunit_add_action_or_reset from drm_kunit_helper_acquire_ctx_alloc and hand-roll the cleanup
11:20sima: also the EDEADLCK issue is kinda absolutely everywhere in kms tests
11:25sima: mripard, probably needs a bit a coordinated effort to sort this out and stop more fail from spreading
11:27sima: mripard, cleanup.h might help to make this not too terrible, at least for the ctx destruction
11:48vignesh: sima, daniels: will enable CONFIG_DEBUG_WW_MUTEX_SLOWPATH in drm-ci and send a patch
11:49daniels: vignesh: thanks a lot!
11:49sima: vignesh, it will be a lot more than just one patch unfortunately, from a quick look a lot of the kms tests are broken :-/
11:49sima: but yeah a patch to see what all breaks is step 1
11:49sima: also annoying lockdep shuts down after the first splat, so you need to fix them all before the amount of work is clear
11:50sima:apologizes for being the harbinger of bad news here
11:52vignesh: sima: I'm going to resend this soon - https://lore.kernel.org/dri-devel/20240812112030.81774-1-vignesh.raman@collabora.com/T/#u. We can add enable CONFIG_DEBUG_WW_MUTEX_SLOWPATH in this. I will test and see what all is broken with the recent kernel
11:57sima: ah yeah that's the prereq, and sounds like a plan
12:37mripard: sima: ack, thanks for looking into it.
12:38mripard: sima: about EDEADLCK, which functions are affected in addition to drm_atomic_set_crtc_for_connector() ?
12:47sima: mripard, the various ones that get a state you don't yet have
12:47sima: unless I'm really confused
12:49neatparlanx: I was more than ten hours straight in computer , so my eyes got swollen, so i am not responsible as to what i typed:D exposed perhaps some core functionality, but it seems i can withstand long hours without so much effect on sanity degration. It's trainable skill however yeah eye needs to be taken care of now. I am unsure if this last red indian stands for after even more sets get tested,
12:49neatparlanx: it was really trial and run discovery. i think i only gave it three repetitions, or four.
12:51neatparlanx: i remember as to how the elmination tasks were handled in math, and tried this approach and it got the delta shifted which was surprising overall to me.
12:55neatparlanx: Well the macbook from intel is the best experience i have ever gotten, intel drivers are fun, i wrote one mouse driver which needs to be ported to libinput and then it should be nearly perfect.
12:56neatparlanx: i lost the adapter but it has a powerful hub, to charge my phone as well as headsets.
12:56neatparlanx: Your teams write drivers ok, i think, so i hence try my best to modernize the compilers too.
13:06neatparlanx: I think that no work was wasted effort, it's the system bring up over drivers was quite complex but it happened to succeed well, it's only the base buffers we can work very well with to scale the systems up and higher.
13:10neatparlanx: I have not got that experience i mainly think that the base needs no changes once you stabilize it, cause of human needs to have the base in certain frequencies to work with less issues, i.e stuff that computer does in the base needs to meet the prequisites of human brain and eye.
13:56mripard: sima: ack
16:23neatparlanx: So it's that i have an office in the countriside now, with computers runnable from your past driver works that i could pimp up and use. I use snb i5 for my main works, but have box full of harddrives laptops pc's etc. something very knew i can not yet afford, dream was to get some FPGA boards, bwidawsk a great driver architect has been drifting towards FPGA development era as i
16:23neatparlanx: understood, lots of hw predesign can be simulated on FPGA first , if i was getting better along with my dad's friends i would suggest making full in house chip factory to asia with hotel near by to acommodate workers or and guests. but it would be mtj and fpga fabric only, ASML machines for very latest transistor sizes are surprisingly expensive. It will take years until i get to
16:23neatparlanx: designing chips though, i work on asics sw atm. then programmable fabric of hw then design later.
18:36hamstringnetuse: I actually expect this to go through all the tests as it appears so based of the logics i now see. I have new other things to do atm. filesystem experts can later jump into discussion where to store the metrics of drives and metadata of fs, i myself design simpler things for my own little works. I say that some X info can be added to every hash in fixed indexes. those are my last
18:36hamstringnetuse: comments about data structures or fs-s based of hashsums.