15:23 <rellla> uahhh. no more deqp tests left that pass with the blob and fail with lima ;)

16:08 <enunes> rellla: cool, I didnt know we were even this close when I mentioned that in the talk

16:09 <enunes> rellla: you can keep the r-b in the second patch for your clip MR

17:11 <tlwoerner> anarsoul: thanks! tweaking the kernel config seems to have improved things. i can run the entire glmark2-es2 benchmark in fullscreen and it completes without crashing

17:12 <tlwoerner> i'm still seeing 2 issues

17:12 <tlwoerner> rock64 kernel: lima ff300000.gpu: fail to save task state from <process>

17:12 <tlwoerner> where <process> is either glmark2-es2 or Xorg

17:13 <tlwoerner> if <process> == glmark2-es2 then i will see a wicked flashing, and almost always occurs in one or the other of the [effects2d] tests

17:13 <anarsoul> tlwoerner: you should hav an error prior to that

17:14 <tlwoerner> if <process> == Xorg then the test continues on, but the screen is frozen at whatever it was showing when the dmesg message appeared (which is, again, almost always in the 2nd [effect2d] test)

17:14 <tlwoerner> in any case, this is a huge improvement! thanks :-)

17:17 <anarsoul> tlwoerner: can you share complete dmesg after reproducing the issue?

17:19 <tlwoerner> anarsoul: the full dmesg from start of boot? or the relevant lima parts when the issue occurs?

17:22 <anarsoul> from start of boot

17:29 <tlwoerner> https://paste.ee/p/CeCDw

17:37 <anarsoul> job timeout

17:38 <anarsoul> you can either increase timeout with sched_timeout_ms=1000 argument to lima module

17:39 <tlwoerner> what's the default?

17:40 <anarsoul> 500

17:41 <anarsoul> or add operating points to gpu node in dts

17:41 <anarsoul> likely it's running not at the highest freq

17:42 <anarsoul> [ 0.379711] lima ff300000.gpu: mod rate = 163840000

17:43 <anarsoul> 163.84mhz, that's pretty slow

17:53 <anarsoul> tlwoerner: it can go up to 500mhz on rk3328, see dtsi from bsp kernel: https://github.com/rockchip-linux/kernel/blob/develop-4.4/arch/arm64/boot/dts/rockchip/rk3328.dtsi#L729

17:55 <anarsoul> so if you're interested in running lima on this board you can get 3x performance boost if you add operating points

18:29 <tlwoerner> interesting! i'm using an upstream kernel, shouldn't this be in the upstream kernel?

18:30 <anarsoul> tlwoerner: no one bothered to adapt it from bsp to upstream

18:31 <anarsoul> it's in dts, we obviously can't cover every board, it's up to board or soc maintainers

18:32 <anarsoul> but code to handle it is there, see sun8i-h3.dtsi as example

18:34 <tlwoerner> ah interesting, i see

18:35 <tlwoerner> thankfully i'm the meta-rockchip maintainer, so i'll add it in there ;-)

18:35 <anarsoul> tlwoerner: please submit it upstream

18:36 <anarsoul> keeping it downstream (i.e. in meta-rockchip) isn't a good idea

18:36 <tlwoerner> i'm a kernel nobody, aren't there people in the project better-positioned to submit this?

18:37 <anarsoul> tlwoerner: so what? you can patch the dtsi, you can test it, so once it's done you can submit it upstream

18:38 <anarsoul> tlwoerner: there's tens of different SoCs with mali4x0, we just don't have every board in the world with every SoC to test it

18:38 <tlwoerner> ah, true :-)

18:38 <anarsoul> mali4x0 can have different frequences and voltages in different SoCs

18:39 <anarsoul> IIRC it goes up to 800mhz in amlogic

18:39 <anarsoul> but on allwinner A64 it's only 432mhz