ChanServ changed the topic of #lima to: Development channel for open source lima driver for ARM Mali4** GPUs - Kernel has landed in mainline, userspace driver is part of mesa - Logs at https://people.freedesktop.org/~cbrill/dri-log/index.php?channel=lima and https://freenode.irclog.whitequark.org/lima - Contact ARM for binary driver support!
machinehum has joined #lima
<Marex> anarsoul: hey, you mentioned 5.4.y is not good enough for testing Lima, right ?
<Marex> seems like Xilinx totally botched ABI compatibility with their TFA/PMUFW, so I cannot get anything newer than 5.4 working, lovely
yuq825 has joined #lima
<Marex> ah look, if I revert the broken xilinx patches from next, then I can boot next, good
<anarsoul> Marex: you can always backport latest lima patches into 5.4
<Marex> oooh, and the GALLIUM_HUD also works with next, unlike 5.4.y
<Marex> anarsoul: I thought you mentioned there's some DRM scheduling problem in 5.4
<anarsoul> oh, right
<anarsoul> and drm_sched patches
<Marex> anarsoul: do you know which drm sched patches are relevant ?
<anarsoul> no, ask MoeIcenowy
<Marex> anarsoul: I think it's drm/scheduler: rework entity creation
<anarsoul> Marex: I haven't looked into it. Feel free to experiment :)
<Marex> anarsoul: already done, I'll run dEQPs with next and then this patched 5.4.32
<anarsoul> you'll likely get slightly different result since next has dynamic tile heap support while 5.4 uses fixed 1mb tile heap
<Marex> anarsoul: I backported that too
<anarsoul> good
<MoeIcenowy> Marex: I backported "drm/sched: Fix passing zero to 'PTR_ERR' warning v2", "drm/scheduler: Avoid accessing freed bad job.", "drm/lima: use drm_sched_fault for error task handling"
<anarsoul> MoeIcenowy: thanks
<Marex> MoeIcenowy: I have all those backported too, thanks
dddddd has quit [Ping timeout: 240 seconds]
<bshah> re: https://github.com/yuq/linux-lima/commits/topic/error-5, where would I find the frequency table for a64's gpu? I looked into dts, but dts just defines the cooling points and not the clock frequency
<bshah> oh, so state defined in gpu_cooling are actually frequency?
<bshah> but then 384MHz value is not there, so I am confused :)
<bshah> anarsoul: do you have any ref for from where the value of 384MHz was taken?
<anarsoul> IIRC 384 is default
<anarsoul> there's no frequency table for A64, BSP runs it constantly at 432
<bshah> hm
<bshah> okay, well... I will at least test out suspend/resume part, sinc eI do occasionally hit gp error after resume in kwin
<anarsoul> update mesa
<bshah> should be roughly ~3 days old
<anarsoul> OK
<anarsoul> enunes: did you try enabling const and uniforms cloning for gpir? it should benefit from it as well
<bshah> so this patch seems to work fine
<anarsoul> no more gp errors?
<bshah> nope, kwin no longer crashes... but I will give it bit more testing of 1 day before sending "Tested-by"
<bshah> (mostly becaue this was not consistent error)
<bshah> it always used to happen like 1/10 times
<anarsoul> yuq825: ^^
<anarsoul> bshah: suspend/resume 100 times in a row? :)
<bshah> xD I can script it probably with rtcwake or something xD
_whitelogger has joined #lima
<Marex> anarsoul: how many dEQP errors should I expect ?
<Marex> anarsoul: on dEQP-GLES2.functional.* that is
<anarsoul> Marex: see mesa/.gitlab-ci/deqp-lima-fails.txt
<anarsoul> and deqp-lima-skips.txt
<Marex> so around 60 ?
warpme_ has quit [Quit: Connection closed for inactivity]
buzzmarshall has quit [Remote host closed the connection]
<anarsoul> likely more since there're flakes in skips
<anarsoul> should be under 100 though
<Marex> I had 500 on next/master + mesa/master, so I suspect I am doing something wrong
<Marex> I'll keep digging
<anarsoul> Marex: skip skips :)
<anarsoul> once you hit GPU error context is considered tainted and jobs from this context are not executed
<anarsoul> so if some tests hits gpu error all the following tests will fail if deqp reuses context
Barada has joined #lima
<bshah> some more manual testing and seems completely stable, no gpu crashes on resume, I will find patch on ML and send Tested-by
<anarsoul> as far as I remember it's not on ML
<yuq825> sorry, I haven't send it to ML as I don't have a way to test it
<anarsoul> apparently you've got it tested :)
<yuq825> yeah, I'll send it out latter
<bshah> Feel free to add "Tested-by: Bhushan Shah <bshah at kde.org>"
<yuq825> thanks
dddddd has joined #lima
machinehum has quit [Ping timeout: 246 seconds]
<kittehuwu> anarsoul: in glxinfo it doesnt even list half of the FBO extensions
warpme_ has joined #lima
<enunes> anarsoul: I didn't try cloning in gp, I'll give it a try
<enunes> I had another small set of small gp changes from a long time ago, mostly moving more stuff to nir passes, but that changed behavior of some tests that started hitting the gpir unknown bugs... so i preferred to not change it at all
<enunes> if it's more stable now, I can also try that again
Barada has quit [Quit: Barada]
buzzmarshall has joined #lima
yuq825 has quit [Quit: Leaving.]
<bshah> agh
<bshah> just when yuq sent the patchset for suspend resume, my lima crashed :S
<bshah> [ 2414.378652] lima 1c40000.gpu: gp error irq state=4 status=2b
<bshah> [ 2414.387932] lima 1c40000.gpu: fail to save task state: error task list is full
<bshah> [ 2414.387951] lima 1c40000.gpu: gp task error int_state=0 status=aa
<bshah> but this looks like potentially different error than what I used to get?
<kittehuwu> yeah i just, cant get any fbo working on my phone
cwabbott has quit [Ping timeout: 246 seconds]
cwabbott has joined #lima
tbueno has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
machinehum has joined #lima
Putti has quit [Remote host closed the connection]
Putti has joined #lima
<anarsoul> bshah: 4 is LIMA_GP_IRQ_PLBU_OUT_OF_MEM
<anarsoul> you need dynamic tile heap patches for kernel, it's not related to suspend
Ntemis has joined #lima
<anarsoul> kittehuwu: IIRC this demo works fine on lima: https://github.com/yuq/gfx/blob/master/gbm-surface-fbo/main.c
<anarsoul> kittehuwu: also a bunch of deqp fbo tests pass
<anarsoul> so I'm not sure what's your issue since you're not telling what it is besides that fbos aren't working
<anarsoul> bshah: however it's pretty surprising to see that you hit it
Ntemis has quit [Remote host closed the connection]
machinehum has quit [Ping timeout: 256 seconds]
hell__ has quit [Quit: CPU triple-faulted.]
hell__ has joined #lima