suprothunderbolt has joined #lima
yuq825 has joined #lima
jrmuizel has quit [Remote host closed the connection]
jrmuizel has joined #lima
jrmuizel has quit [Remote host closed the connection]
dddddd has quit [Remote host closed the connection]
<
MoeIcenowy>
anarsoul: for throttling is work at both kernel and userspace side needed?
_whitelogger has joined #lima
_whitelogger has joined #lima
Barada has joined #lima
<
anarsoul>
MoeIcenowy: I think it'd be better to either fix X11
<
anarsoul>
or try to optimize lima
<
anarsoul>
yuq825: hi
<
anarsoul>
is there a way to optimize fb reload so it reloads only part of fb is we have scissor test enabled?
<
yuq825>
sounds like partial update
<
anarsoul>
but without EGL_KHR_partial_update
<
anarsoul>
X11 seems to need it
<
yuq825>
method is same
<
yuq825>
skip pp stream does not change
<
yuq825>
when flush, union scissor region
<
anarsoul>
OK, thanks
<
yuq825>
generate a pp stream which only contains scissor regions
<
yuq825>
but the region has to be small enough, otherwise regenerate the pp stream is also not cheap
<
anarsoul>
it's cursor
<
anarsoul>
and it's expensive to reload whole fb to just repaint a cursor
<
MoeIcenowy>
anarsoul: there could be other programs that have the same program
<
MoeIcenowy>
so fixing our driver is better than fixing user program
<
anarsoul>
MoeIcenowy: not really
<
anarsoul>
X11 is a really bad example
<
MoeIcenowy>
here sometimes an area of content directly disappeared
<
MoeIcenowy>
this happens when running gtk3-demo
<
anarsoul>
that's some other issue
<
MoeIcenowy>
and after some operations in gtk3-demo, a gp error raised
<
MoeIcenowy>
and the screen gets freezed
<
anarsoul>
0x400000?
<
anarsoul>
well, someone needs to debug it
<
MoeIcenowy>
BTW gdm on wayland also triggers this gp error
<
anarsoul>
I know very little about command stream
<
MoeIcenowy>
anarsoul: how to?
<
anarsoul>
and sway triggers it
<
anarsoul>
MoeIcenowy: capture command stream dump and analyze it
<
MoeIcenowy>
I remember that Android may also trigger it
<
anarsoul>
also it'd be nice to capture job that causes an error in kernel
<
anarsoul>
and expose it via debugfs
<
anarsoul>
akin to i915 driver
<
MoeIcenowy>
is there any other useful driver internals?
<
MoeIcenowy>
anarsoul: BTW why a single GP error freezes the display?
<
MoeIcenowy>
should we be able to recover?
<
anarsoul>
MoeIcenowy: because context marked as bad and won't execute any jobs
<
anarsoul>
if it happens to be X server or wayland compositor it means everything freezes
<
MoeIcenowy>
what prevents continuing to execute jobs?
<
anarsoul>
MoeIcenowy: see log from yesterday
<
anarsoul>
yuq825 explained it to me :)
<
MoeIcenowy>
yesterday in which TZ?
<
MoeIcenowy>
okay find it
<
MoeIcenowy>
but looks not so well explained
<
MoeIcenowy>
anarsoul: BTW could we start a new kernel context?
<
MoeIcenowy>
and drop the polluted one
<
anarsoul>
MoeIcenowy: you can kill the process
<
yuq825>
unless we know why the error happens and can recover the error, keep running the bad context is not a good idea
<
anarsoul>
and restart it
<
MoeIcenowy>
too violent.
<
anarsoul>
I tried it yesterday
<
anarsoul>
it doesn't recover and keep throwing the same GP error
<
MoeIcenowy>
do you mean kernel thread?
<
anarsoul>
s/keep/keeps
<
MoeIcenowy>
or user process?
<
anarsoul>
user process
<
anarsoul>
kernel is fine
<
anarsoul>
I don't think that we start any kernel threads
<
MoeIcenowy>
at least I think there're two named gp and pp
<
yuq825>
that's task scheduler
<
anarsoul>
MoeIcenowy: anyway, someone should figure out why we get this GP error
<
MoeIcenowy>
I think there's currently no way to reproduce it w/o user interaction?
<
anarsoul>
MoeIcenowy: start sway, open any window, move cursor
<
anarsoul>
that's like 100% reproducible
<
MoeIcenowy>
anarsoul: oh sorry just now I did it...
<
MoeIcenowy>
and it works...
<
anarsoul>
open another window
<
MoeIcenowy>
ooooooooooooops.
<
MoeIcenowy>
BTW does apitrace works on GBM?
<
anarsoul>
probably
<
MoeIcenowy>
found a way to reproduce here: press the session select key on Wayland-based GDM
<
anarsoul>
hope you can figure out why it happens :)
<
anarsoul>
I just pushed new version of BO cache
<
anarsoul>
now it frees BOs that haven't been in use for longer than 5 seconds
<
MoeIcenowy>
anarsoul: what window did you use for sway triggering
<
yuq825>
does enlarge PLB size or GP heap size solve the GP error?
<
anarsoul|c>
There's dedicated IRQ to signal if we don't have enough heap
<
anarsoul|c>
MoeIcenowy: it's termite
<
anarsoul|c>
However weston-terminal works as well
<
anarsoul|c>
Well, breaks as well
<
anarsoul|c>
IIRC not matching PLB size breaks everything
<
anarsoul|c>
OK, bed time here. Good night
griffinp has quit [Read error: Connection reset by peer]
bhoj has quit [Read error: Connection reset by peer]
griffinp has joined #lima
bhoj has joined #lima
<
MoeIcenowy>
oh looks like finally the cause is found?
suprothunderbolt has quit [Ping timeout: 276 seconds]
suprothunderbolt has joined #lima
yuq825 has quit [Quit: Leaving.]
dddddd has joined #lima
megi has joined #lima
jrmuizel has joined #lima
jrmuizel has quit [Remote host closed the connection]
jrmuizel has joined #lima
adjtm has quit [Ping timeout: 276 seconds]
jrmuizel has quit [Remote host closed the connection]
adjtm has joined #lima
kaspter has quit [Read error: Connection reset by peer]
kaspter has joined #lima
Barada has quit [Quit: Barada]
suprothunderbolt has quit [Read error: Connection reset by peer]
<
anarsoul>
MoeIcenowy: what cause?
jrmuizel has joined #lima
adjtm has quit [Remote host closed the connection]
adjtm has joined #lima
adjtm has quit [Remote host closed the connection]
adjtm has joined #lima
kaspter has quit [Read error: Connection reset by peer]
kaspter has joined #lima
<
MoeIcenowy>
GP error
<
MoeIcenowy>
anarsoul: ^
<
anarsoul>
and what's the cause?
<
MoeIcenowy>
`not matching PLB size breaks everything` ?
<
anarsoul>
it's not the cause of this 0x400000 error
<
anarsoul>
if you change PLB_BLK_SIZE from 512 on Mali400 to something else it'll start failing everywhere
<
anarsoul>
it's 2048 for Mali450 on H5 and 4096 for any other Mali450
<
MoeIcenowy>
oh why is H5 so special?
<
anarsoul>
thanks again, allwinner :)
<
anarsoul>
MoeIcenowy: allwinner changed something in it (probably to reduce die size?)
<
anarsoul>
and it's not compatible with mali blobs that aren't specific to H5 anymore :)
<
jernej>
I imagine this PLB size can't be determined at runtime?
<
anarsoul>
there's a hack for mali kernel driver that patches pp stream on fly
<
anarsoul>
to allow H5 working with available blobs
<
jernej>
I know, but I prefer waiting on lima to be in good shape :)
<
anarsoul>
it's not in bad shape atm
<
anarsoul>
IIRC mythtv works fine with git master
<
jernej>
well, there is still issue with one shader used by Kodi, it was reported a while ago, let me find it
<
enunes>
jernej: if you actually use kodi it might be helpful to test the old issue again and check if it still happens
<
enunes>
and close it if it's fixed
<
enunes>
I don't have a kodi setup
<
jernej>
I think it was just a note in some issue not directly related to Kodi
<
enunes>
searching for kodi I can only find two issues which are respectively 5 and 6 months old
<
enunes>
even before ppir cf which is implemented now
<
jernej>
and chewitt tested lima with Kodi this week again and same issue was present
<
enunes>
so one of them can probably be closed
<
anarsoul>
even before we fixed int/float mess
<
enunes>
chewitt's issue should be fixed
<
jernej>
as I said, this issue I'm talking about was not directly reported
<
jernej>
IIRC issue is with the shader which does color space conversion
<
anarsoul>
should work now
<
jernej>
and it works for HD videos but not for SD
<
jernej>
as I said, it was tested this week with mesa master
<
anarsoul>
then please open new issue
<
jernej>
yeah, once I make whole setup
<
jernej>
it was not tested by me, so I can't give any useful info
<
anarsoul>
are you sure it's still broken?
forkbomb has quit [Ping timeout: 264 seconds]
forkbomb has joined #lima
<
jernej>
that's what I've been told
<
jernej>
but it might not be this particular shader
<
jernej>
I'll do some tests and open separate issue
<
jernej>
but it will take a while
<
anarsoul>
shader itself compiles just fine
<
anarsoul>
and I don't see anything fishy in disassembly
<
jernej>
maybe precision or range issue? It works for HD but not SD videos IIRC
<
jernej>
or viceversa? I don't remember exactly
<
anarsoul>
it would be otherwise if it was precision
<
anarsoul>
(also should be fixed now)
drod has joined #lima
adjtm has quit [Ping timeout: 276 seconds]
adjtm has joined #lima
drod has quit [Ping timeout: 240 seconds]
drod has joined #lima
jrmuizel has quit [Remote host closed the connection]
jrmuizel has joined #lima
jrmuizel has quit [Remote host closed the connection]
jrmuizel has joined #lima
drod has quit [Read error: Connection reset by peer]
drod has joined #lima
jrmuizel has quit [Remote host closed the connection]
jrmuizel has joined #lima
nerdboy has quit [Excess Flood]
nerdboy has joined #lima
nerdboy has quit [Changing host]
nerdboy has joined #lima
jrmuizel has quit [Remote host closed the connection]
drod has quit [Remote host closed the connection]
jrmuizel has joined #lima
jrmuizel has quit [Remote host closed the connection]