<anarsoul>
likely it's mismatch between struct sizes on 32 bit and 64 bit, see mesa/include/drm-uapi/lima_drm.h
<anarsoul>
I can't find obvious problem there by skimming though it, feel free to poke it and submit an MR if you find the issue
<mrfixit2001>
Will be happy to have a look... although many applications don't actually *crash*... they just don't render anything on the screen. The screen goes black, and the backlight lights up like it's about to render something... but then it doesn't. No errors to capture and nothing in the dmesg
<anarsoul>
goes black how?
<mrfixit2001>
Well for example some of these applications specifically skim all the crtc modes and will change the resolution. I've confirmed that works and that the resolution changes. Then the screen goes blank and nothing is on it. But it app doesn't crash at all.
<anarsoul>
can you check if it's actually doing something?
<anarsoul>
check /proc/interrupts
<anarsoul>
there's pp and gp interrupts
<anarsoul>
are the numbers increasing?
<mrfixit2001>
pp_bcast is continuing to increase and so is gp.
<mrfixit2001>
gpmmu stays at 1, and everything else is 0
<anarsoul>
what hw is it?
<mrfixit2001>
Rock64 rk3328
<mrfixit2001>
I compiled the latest bleeding edge commits from 5.4rc1 and mesa from today
<mrfixit2001>
and lima is clearly working, as kmscube runs perfectly too. Very strange.
<anarsoul>
mrfixit2001: so it's rendering something
<anarsoul>
maybe it failed to import bo from display driver, no idea
<mrfixit2001>
any thoughts on how to debug further without any actual error messages?
<mrfixit2001>
retroarch oddly renders *something* to the screen... once again, no errors... it's interface is usually bright blue with a slightly animated ribbon in the background. I get a gray animated ribbon in the middle of the screen and nothing else at all haha
<anarsoul>
likely you have to debug display part
<anarsoul>
I guess you can try dumping framebuffer that lima renders into, see lima_draw.c:_lima_flush()
<anarsoul>
call lima_submit_wait(ctx->pp_submit, PIPE_TIMEOUT_INFINITE); at the end, then you can call lima_bo_map() on framebuffer BOs and then you can actually dump it to file (use regular fopen/fwrite/fclose)
<anarsoul>
see lima_pack_wb_cbuf_reg() on how to get resource for framebuffer
<anarsoul>
likely you'll get a lot of intermediate images if your app renders into FBOs
<mrfixit2001>
Much appreciated. Will tinker a bit further this week. Interestingly, retroarch rendered perfectly when rendered fully aarch64.
<anarsoul>
note that calling lima_submit_wait() on pp submit will kill performance for sure since it serializes jobs
<anarsoul>
but with dumping fb to file that wouldn't be a bottleneck :)
<mrfixit2001>
anarsoul any of this concerning to you during lima commit? Or is it fairly standard / expected: https://pastebin.com/etez3vYk
<mrfixit2001>
*init, not commit
<anarsoul>
is it with extra debug?
<mrfixit2001>
Not sure which config setting specifically you're asking about, but these are all enabled: CONFIG_DEBUG_INFO, CONFIG_DEBUG_KERNEL, CONFIG_DEBUG_MISC, CONFIG_DEBUG_BUGVERBOSE
<anarsoul>
mrfixit2001: it's unlikely to be the issue (you're running the same kernel anyway)
yuq825 has joined #lima
yuq825 has quit [Remote host closed the connection]
<rellla>
i get EGL_BAD_ALLOC during eglInitialize ..
<rellla>
ah... trying to reconfigure kernel now.
monstr has joined #lima
adjtm_ has quit [Ping timeout: 265 seconds]
<MoeIcenowy>
anarsoul: serialized run of piglit shows no this pp error now...
yuq8251 has joined #lima
yuq825 has quit [Remote host closed the connection]
adjtm_ has joined #lima
<rellla>
ok, so no more lima work today since my opipc2 isn't reachable over ssh anymore. thanks mali binary blob!
<MoeIcenowy>
setting up kernel over ssh is really brave...
<rellla>
yeah :)
<rellla>
it booted, after a modprobe mali i got kernel memory errors, lima and mali weren't able to be rmmod'ed and after a reboot it's not reachable anymore :(
<anarsoul>
Danct12: also make sure you don't disable glamor
<anarsoul>
rellla: IIRC you need to modify your demo to work with mali blob
<anarsoul>
it doesn't like ARGB8888 format so you have to use XRGB
<anarsoul>
basically get it working with kmscube
<anarsoul>
and then compare kmscube egl initialization to your demo and make necessary adjustments
<rellla>
anarsoul: i randomly looked into https://github.com/yuq/gfx and adjusted them, but i missed sth. and now i somehow crashed my soc and first have to see what happened :p
<rellla>
i kind of try&error recompiled the kernel with CMA_SIZE and CONFIG_DRM_FBDEV_OVERALLOC adjusted - and built mali with debug enabled.
<rellla>
there was some step included, which my soc did not like :(
<anarsoul>
:(
<rellla>
as MoeIcenowy said, hacking the kernel over ssh is .... bad :)
<anarsoul>
yeah
<anarsoul>
rellla: hacking kernel when all you have is LCD and usb is even worse :) I did that back in 2006 though
<Danct12>
anarsoul: -Dglx=dri in the apkbuild, however dri3 is not mentioned anywhere in the compile flags, will add that and try
<anarsoul>
Danct12: sounds good
<Danct12>
anarsoul: nope, doesn't seem to work for me
<Danct12>
[ 84.203] (EE) AIGLX error: sun4i-drm does not export required DRI extension
<Danct12>
[ 84.203] (EE) AIGLX error: dlopen of /usr/lib/xorg/modules/dri/swrast_dri.so failed (Error loading shared library /usr/lib/xorg/modules/dri/swrast_dri.so: No such file or directory)
<Danct12>
still trying to load this swrast thing
<warpme_>
hmm: I'm also getting (EE) AIGLX error: sun4i-drm does not export required DRI extension on h6 (t720). Just curious what is root cause of this error. As exactly that same kernel/mesa/app works ok on s905 and rk3328 - this must be allwinner drm related thing. has anybody idea how to move forward with this error?