01:34 <Lyude> oh hell yeah, I've got the day after thx gives off as well

01:34 <Lyude> guess i've got even more time for panfrost this weekend then I thought :)

02:05 <NeuroScr> :)

10:19 <tomeu> Lyude: have you needed to change anything in ARM's driver already?

17:25 <Lyude> tomeu: not in their driver-but I will need to add some platform code for GPU resetting on meson most likely

17:25 <Lyude> currently the only solution I have for that is rebooting the whole system, which takes a few minutes :(

19:24 <NeuroScr> ok I got mesa compiled

19:26 <NeuroScr> looks like my system is still use /usr/lib set tho

20:00 <narmstrong> Lyude: you haven’t found anything useful in the driver modified by amlogic ?

20:23 <NeuroScr> ok I think I got glxgears working on it

20:25 <NeuroScr> 90fps in the initial window, 18fps maximized

20:47 <NeuroScr> pan_nondrm.c:79 pandev_open: Assertion ‘fl != -1’ failed

20:47 <NeuroScr> *fd

20:47 <NeuroScr> can’t get any of these other demos to run

21:25 <Lyude> narmstrong: I have, it's just the actual gpu reset that isn't implemented in the amlogic modified version

21:25 <Lyude> NeuroScr: means it's probably using llvmpipe

21:25 <Lyude> fd != -1 means it didn't manage to get a kernel context, what branch of panfrost are you building and against what mali_kbase version

21:26 <Lyude> narmstrong: anyway-the code for that should be pretty simple to write; it's just i've been doing other things today since thanksgiving :p

21:26 <Lyude> but either tonight or tommorow is when I start working on that again, until monday

21:30 <narmstrong> Lyude: there is reset line available for Mali : RESET_MALI in https://github.com/torvalds/linux/blob/master/include/dt-bindings/reset/amlogic%2Cmeson-gxbb-reset.h

21:32 <Lyude> narmstrong: oooh, that is probably what we want, I guess then the question is how do we actually know if mali_kbase is using that and if not, how do we make it use it

21:32 <Lyude> narmstrong: would it help at all if I showed you where in kbase it fails to reset the gpu?

21:32 <Lyude> or a backtrace rather

21:45 <narmstrong> Not sure, does kbase have a reset handler to trigger an hw reset ?

21:46 <narmstrong> Going to sleep, but you can show me the code ! I’ll have a look next morning

22:09 <Lyude> narmstrong: I'm not sure; I assume so but figuring that out is also part of what I'm planning on doing this weekend :)

22:16 <Lyude> narmstrong: for when you get back, what happens is a GPU error gets printed from mali_kbase into dmesg, then it looks like mali_kbase hangs in kbase_pm_gpu_poweroff_wait_wq

22:16 <Lyude> https://lyude.net/~lyudess/amethyst-hang-narmstrong.txt

22:16 <Lyude> trying to unload mali_kbase at that point fails obv; the rmmod command just hangs

22:16 <Lyude> going to try this with a lockdep kernel next chance I get