<BlueMatt>
fwiw I built the bl301 bin separately and that worked fine with gcc from debian's repos
<BlueMatt>
no need to download ancient gcc-cross
<BlueMatt>
though it failed for the acs bin :(
<chewitt>
^ that repo has everything you want
<BlueMatt>
right.....may make sense to replace the instructions that include downloading two toolchains and building the ancient amlogic u-boot with just downloading those bins :p
<chewitt>
it's always best to reflect the 'true' method of building things from HK sources
<BlueMatt>
did you manage to rebuild the acs bin with a modern compiler?
<BlueMatt>
it at least works for bl301 and its trivial to pull the bl301 build folder and use that makefile without building the whole thing
<BlueMatt>
is it really the 'true' method if its some hacked up makefile from arm to generate the tiny binary? :/
<chewitt>
I have a VM with all the different u-boot sources inside with the 'legacy' toolchains setup .. it's simple to generate things
<chewitt>
blame Amlogic for using 2015 sources :)
<BlueMatt>
:shrug: just seems a bit roundabout to build a full u-boot to pull out 2kb of binary
<BlueMatt>
which is shoved in and only tangentially built with the u-boot
<chewitt>
hence the LE repo exists :)
<BlueMatt>
lol, fair
<BlueMatt>
final quest...why the hell is my odroid-c4 a ton slower than my c2 :(
<BlueMatt>
any hints on reading clockrate off the damn thing?
<chewitt>
I'm using 5.11 to reduce the size of the patchset I hoard on-top of the base kernel, but it's stable
warpme_ has quit [Quit: Connection closed for inactivity]
<chewitt>
my defconfig is a bit minimal for desktop use, but has everything needed for Amlogic boards
ballerburg9005 has joined #linux-amlogic
<ballerburg9005>
I am making system images by simply combining the Libreelec kernel with any distribution
<ballerburg9005>
I got VPU running on Ubuntu, although it tears quite unusually
<ballerburg9005>
but cpufreq is not working
<ballerburg9005>
support is compiled into the kernel, I thought it might be a firmware issue
<ballerburg9005>
it was working on 4.9 coreelec but now its 5.10 something from libreelec
<BlueMatt>
chewitt: hmm, you sure the cpufreq/cpuinfo isn't just in 5.11/your patchset?
<ballerburg9005>
BlueMatt: you mean, if the kernel was compiled without it?
<BlueMatt>
no, as in i think i turned on all the relevant-ish configs and still dont see what he saw on his 5.11++-based kernel
<ballerburg9005>
oh you don't mean me, it was the same topic it seemed to me
<ballerburg9005>
Could you tell me what the general situation is of video decoder acceleration? On my box it has weird blocky tearing. What will it take for it to work from Firefox?
<ballerburg9005>
is it rather that its buggy in most situations, or that it works on most boxes?
<ballerburg9005>
also is there a double buffering option to avoid the normal Xorg tearing?
<chewitt>
use mesa master .. panfrost is still under rapid development
<chewitt>
no idea what X support is like as I never use it
<chewitt>
the vdec is usable with H264 but there are glitches and ffmpeg needs work
<chewitt>
HEVC support is only partially implemented (and not upstream)
<ballerburg9005>
thanks for the info
<chewitt>
right now the vdec is IMHO the weakest part of the upstream codebase
<chewitt>
pi foundation are currently doing work to improve the stateful v4l2 code path in ffmpeg
<chewitt>
but .. they reached a point where vdec tweaks were needed to get seeking working properly
<chewitt>
so for a while ffmpeg support was improving, but beyond this point it regresses again - similar tweaks are needed in the Amlogic vdec
<ballerburg9005>
but vdec hardware acceleration doesn't work for the browser?
<ballerburg9005>
I am mainly concerned with getting videos in the browser to work, because that's what sells it to people in the end
<chewitt>
i'm not sure what's missing there .. I don't really track desktop things
<ballerburg9005>
oh ok
<chewitt>
which distro are you working on?
<ballerburg9005>
I am using Ubuntu MATE 20.10 right now
<chewitt>
BlueMatt there's nothing related to cpufreq in my patchset
<chewitt>
those stats are from an N2+ but it's the same kernel
ballerburg9005 has quit [Quit: Lost terminal]
ballerburg9005 has joined #linux-amlogic
hexdump0815 has joined #linux-amlogic
<hexdump0815>
BlueMatt: c4 (s905x3) can in some situations be slower than c2(s905) - i also have a use case where this is the case - one thing i found out so far is that the s905x3 seems to have only half of the l1 cache (16k) of the other amlogic socs (32k) which might result in such behaviour
hexdump0815 has quit [Quit: Connection closed]
sputnik_ has joined #linux-amlogic
sputnik_ has quit [Remote host closed the connection]
sputnik_ has joined #linux-amlogic
buzzmarshall has quit [Read error: Connection reset by peer]
buzzmarshall has joined #linux-amlogic
sputnik_ has quit [Ping timeout: 260 seconds]
sputnik_ has joined #linux-amlogic
buzzmarshall has quit [Quit: Leaving]
_whitelogger has joined #linux-amlogic
sputnik_ has quit [Read error: Connection reset by peer]
<BlueMatt>
chewitt: hum...I definitely have all those configs (and more from your defconfig you pasted previously, and have nothing for cpufreq at all in sysfs :/
ballerburg9005 has joined #linux-amlogic
<BlueMatt>
and, fwiw, it seems the acs.bin defaults to 1.2GHz which if I hex-edit the 1.2 to 1.6 performance matches odroid-c2 exactly....obviously linux should be frequency-scaling but at least i can hand-edit it
<BlueMatt>
chewitt: hum, yea, thats got a trans_table with transitions and a cur_freq :/
<BlueMatt>
LibreELEC:/ # [ 79.264243] reboot: Restarting system
<BlueMatt>
wait, what, is it supposed to crash after a minute
<chewitt>
sounds like the installer
<BlueMatt>
ah, i dont have a monitor or keyboard hooked up so wasnt sure what it even was lol
<chewitt>
it boots, expands storage to 100% of SD card, then reboots into Kodi
<BlueMatt>
anyway, yea, so (a) somehow your image supports reboot (which others have reported, and i also dont have working)?, (b) it supports cpufreq :(
<chewitt>
compare whether you have modules =y or =m
<BlueMatt>
is it possible i need to build with graphics or something crazy for that
<BlueMatt>
I dont even have module support on mine
<chewitt>
???
<BlueMatt>
as in - why compile with loadable module support when I'm compiling for a single board and dont need graphics support
<BlueMatt>
there's no modules I'd ever want to be optionally-loadable
<chewitt>
I build most things with =y to give faster boot
<chewitt>
also because historically it solves some timing/race issues
<chewitt>
I think most have been resolved now, but..
<BlueMatt>
right, I also just go ahead and set CONFIG_MODULES=n :)
<chewitt>
feel free to pastebin the boot log
<chewitt>
I'm heading to Zzz but will look in the morning
<xdarklight>
BlueMatt: VDDCPU is missing, check if you have CONFIG_REGULATOR_PWM=y in your .config
<xdarklight>
(unrelated to this: someone with an Odroid-C4 should check how the SAR ADC voltage reference is wired up and assign that in the .dts)
<BlueMatt>
yep, that gave me a trans_table and everything jumped up to 1.9G
<BlueMatt>
though strangely its listed as 2.016GHz on the manufacturer site....i guess upstream decided that wasnt stable?
Tonymac32 has quit [Read error: Connection reset by peer]
Tonymac32 has joined #linux-amlogic
buzzmarshall has quit [Remote host closed the connection]
<xdarklight>
BlueMatt: as far as I understand there's different binning versions of the SoC. 1.9GHz is the maximum for the "lower grade" while >2GHz is the maximum for the "higher grade"
<BlueMatt>
xdarklight: ah, and I presume linux doesnt detect the binning, but hardkernel pays more for the binned ones
<BlueMatt>
?
<xdarklight>
BlueMatt: I am not sure if there's a way to detect it. at least I couldn't find any references in Khadas/Hardkernel's Linux sources so far
<BlueMatt>
damn :/
<xdarklight>
that said, it could be hidden somewhere in u-boot though. but I haven't checked that yet