<alyssa>
I'm assuming I've botched something with my frankendebian
<macromorgan>
I can send you the binaries I'm using if you want, I'm on mainline now
<alyssa>
appreciate it
<alyssa>
binaries for u-boot or..?
<macromorgan>
or you can just build it yourself... note that you need to do a bit of jiggery-pokery if you want to use the SPI flash, as it's still WIP (but mostly upstreamed now)
<alyssa>
jiggery-pokery. oh boy.
<macromorgan>
had to figure out the hard way that 1) the SPI flash boots from a different location than the SD card and 2) that the SPI flash doesn't like using DMA in SPL, otherwise it crashes A-TF and causes Linux to fail to boot
<alyssa>
Oh boy.
<macromorgan>
let me finish testing the latest patches and I'll upload my bins
<macromorgan>
yeah, it's all figured out though and working fine now
<alyssa>
I have mainline kernel torvalds/linux built, this same issue as happening with debian kernel etc
<alyssa>
so I'm assuming I just have something botched on the u-boot end
<macromorgan>
most definitely
<macromorgan>
worst case have you tried reflashing the U-Boot? it's possible the BMPs are corrupted
<alyssa>
where do I eevn get a u-boot to flash..?
<macromorgan>
that's a really old version
<alyssa>
I hadn't touched this machine in a year mind you
<macromorgan>
try with that, in the mean time I'll get upstream binaries built
<alyssa>
Thanks
<alyssa>
(FWIW, my goal is to run the opengl es 3.1 conformance tests on G31. They passed on G52 last night.)
<macromorgan>
actually just do that with the odroid image... turns out unless you apply a series of patches to upstream (which are also in mainline review) you won't be able to see the chip from upstream anyway
<alyssa>
ruh roh
<alyssa>
odroid image doesn't have the panfrost kernel though last I checked..
<macromorgan>
that image should just be for U-Boot
<macromorgan>
you can use downstream U-boot with an upstream kernel
<alyssa>
Awesome
<alyssa>
(do I just dd that .img to the mmcblk1?)
<macromorgan>
yeah, then when you boot hold down the reset button on the back to force it to boot off of the mmc and bypass the SFC (which is first in the boot order)
<macromorgan>
or I can just send you my kernel image if you want (doesn't use an initramfs, no modules, just for testing)
<macromorgan>
you can use it to at least flash upstream U-Boot
<alyssa>
alright, trying on my own first but may need it let's see
<alyssa>
what does your partition layout look like?
<macromorgan>
okay. I'll test upstream again booting from SPI, have only tested MMC lately
<alyssa>
(i.e. do I still need that recovery image now that it's flashed allegedly?)
<macromorgan>
I have a GPT partition with the boot partition (VFAT) starting at 16M, a root partition (ext4) going until -2G, and then a swap partition at the end of the MMC. In the first 16MB I have upstream u-boot-rockchip.bin flashed at offset 64 (assuming 512 block size)
<macromorgan>
no you don't
<macromorgan>
forgot to mention you could also just entirely erase the SPI chip and instead use U-Boot directly from the MMC
<alyssa>
Too many choices 😆
<macromorgan>
upstream works fine so long as you apply that one patch above (for the 2 clock fixes)
<macromorgan>
yeah, macc24 and myself have been working hard on mainline support
<macromorgan>
audio is available as of 5.14, sfc might be coming in 5.15, and battery has been submitted but still under review. Crypto notwithstanding everything else is fully supported in mainline.
<alyssa>
if your graphics performance sucks it wasn't me X:
<macromorgan>
at this point I just want it to work :-)
<macromorgan>
getting that funky -1 shift error every time I try
<alyssa>
oh!
<alyssa>
sorry I didn't connect the two nicks, I feel like such a doofus
<macromorgan>
tried with both 20.3 and 21.1 mesa
<macromorgan>
no worries
<alyssa>
it's not a mesa issue
<macromorgan>
right
<alyssa>
now you know I'm procrastinating on that bugfix to get my ogo adv setup again *whistles*
<macromorgan>
thought I'd try just the same, but it's that line in the kernel I think
<alyssa>
/dev/mmcblk1p1 32768 534527 501760 245M c W95 FAT32 (LBA)
<alyssa>
/dev/mmcblk1p1 32M 32M 0 100% /mnt
* alyssa
blinks
<alyssa>
maybe 33mb is excessive for a kernel ...
* alyssa
just wants this to work 😢
<macromorgan>
I can send you my stuff, hang on
<macromorgan>
first I'm going to test the SPI image to make sure it works
<alyssa>
I think I'm going to take a break for a little bit. Maybe fix that kernel bug for you in the mean time ...
<alyssa>
Mostly hoping to get the CTS run started before I leave for the weekend tomorrow night (and then I should have results on Monday morning)
<macromorgan>
okay
<macromorgan>
I'll get you a working upstream U-Boot if you need it
<robmur01>
moar space for kernels! Now's the time to admit that my Juno filesystem got minced in a partition resizing incident after my 100MB /boot became too small...
<robmur01>
sadly 33MB is not at all excessive for an uncompressed arm64 Image these days. GRUB with decompression FTW!
<maz_>
robmur01: I recently found that 32GB is on the tiny size for /boot, as I had all most kernel versions installed since... 4.2?
<maz_>
s/size/side/
<robmur01>
maz_: but I bet you were booting them all simultaneously in VMs, right? :P
<maz_>
robmur01: how did you know? :D
<alyssa>
robmur01: Speaking of Juno, has T604 support in panfrost broken again? :-p
maz_ is now known as maz
vagrantc has joined #linux-rockchip
<robmur01>
broken gets a whole new meaning when `file /usr/local/bin/glmark2-es2-drm` says "Git packfile"
<robmur01>
and hey, Juno is T624, not that broken 1st-gen rubbish!
<robmur01>
still need to find a few hours to build a new FS and try to rescue scraps from the old one (I repair-bodged it far enough to run the tests I needed for my IOMMU stuff, but it's too far gone really)
<robmur01>
I still have tabs open on those T720/T820 issues to repro when I finally get round to it...
<macromorgan>
U-boot tested out fine, building a kernel now
<macromorgan>
you need the kernel or just want the U-boot bins?
<alyssa>
robmur01: Lovely
<macromorgan>
(or did that flashing fix yours alyssa)?
<alyssa>
macromorgan: Flashing destroyed my partition table and rootfs and everything so haven't tried booting again
<alyssa>
and gnome disks sdid not like it.
<macromorgan>
fuck
<alyssa>
It was a pristine rootfs, don't worry
<macromorgan>
I have a debian image I can send you too if you need?
<alyssa>
i have run debootstrap like 6 times in the past 2 weeks
<macromorgan>
a pristine debian with my kernel and u-boot on it
<alyssa>
Ooh, yes please in that case :)
<macromorgan>
k, let me finish building the kernel, test it, then I'll send you the image with everything
<alyssa>
thank you :>
<alyssa>
I guess in that case I really ought to fix your panfrost kernel bug then
<alyssa>
(Not sure why nobody else hits it. I guess the current code works by chance and you only found the issue with ASan?)
<CounterPillow>
is someone here by any chance very intimately familiar with device tree schemas?
<CounterPillow>
Specifically, I have two boolean properties and want to do an "if" to check for the absence of both of them. Can't find any documentation on the "if" stuff
* robmur01
hides
<CounterPillow>
imma try if:\nproperties:\nfoo: false\nbar:false
<alyssa>
robmur01: lol
<robmur01>
I know enough to ask the questions, just not to have the answers :)
<CounterPillow>
ok hell yeah this works
<robmur01>
hmm, apparently my schema version is too old... now if only I had any idea where or how I installed it (given that I don't even have pip3) :/
<CounterPillow>
they do require pretty much git master these days, it seems.
<macromorgan>
uploading my image now
<macromorgan>
also I guess if I can just disable asan I will do that for now :-)
<macromorgan>
user/pass is odroid
<macromorgan>
note that it's fully upstream U-boot, so you may have issues booting it. For some reason I have to hold the button down right until the light turns on. If I hold the button down too long or not long enough it fails to boot (never figured out why)
<macromorgan>
once you flash upstream U-boot to the SPI chip it's no longer an issue though
<macromorgan>
upstream U-boot (for the SPI chip) is located in the /home/odroid folder
<macromorgan>
sudo dd if=idbloader.img of=/dev/mtdblock0 bs=512 seek=128 && sudo dd if=u-boot.itb of=/dev/mtdblock0 bs=512 seek=2048 should flash it for you
<macromorgan>
or if you don't want to do that just erase your SPI chip and it should boot off of the SD card without any tricks
<alyssa>
ooh
<macromorgan>
also I have kaslr enabled too... I find that you catch crazy memory bugs when you enable that
<alyssa>
Nod
<alyssa>
macromorgan: email address? for rpeorted-by
<macromorgan>
macromorgan@hotmail.com
<alyssa>
and name?
<macromorgan>
Chris Morgan
<alyssa>
thx
<macromorgan>
you'd be shocked how much shit you get on the kernel mailing list for having a hotmail address
<alyssa>
hotmail blackholes all my mail so..
<macromorgan>
still, that address pre-dates Microsoft owning Hotmail
<alyssa>
I still remember rocketmail :-p
<macromorgan>
you're old-school like me
<alyssa>
(but only as Yahoo :V)
<alyssa>
wikipedia says that was 2008-2013 timeframe
<macromorgan>
I remember Rocketmail was better than hotmail for use with lynx back in the day
<macromorgan>
since my first internet connection was a dial-up shell account
<macromorgan>
(early/mid 90s)
<macromorgan>
anyway I'm going to go get some food, let me know if you have any issues with that disk image
<alyssa>
right now the odroid seems more interesting :-p
<macromorgan>
I'll test it too... unfortunately I think my issue is something else in Wayland and not necessarily panfrost related
<macromorgan>
might need to switch to Xorg or something
<alyssa>
does fbcon work?
<alyssa>
if no, it's not any of the three..
<alyssa>
macromorgan: trying to boot your .img
<macromorgan>
yeah framebuffer console is working
<macromorgan>
digging through logs now
<alyssa>
not getting anything on screenwith your .img
<alyssa>
not even the uboot splash
<alyssa>
and unfortunately lack a uart cable
repk has quit [Quit: WeeChat 3.2]
<macromorgan>
there is no u-boot splash on upstream :-(
<macromorgan>
the screen is blank until you hit the kernel
<macromorgan>
it's likely U-boot failed, again for some reason upstream has an issue when you're holding down the button, if not timed perfectly it gets stuck jumping from SPL stage to U-boot stage
<macromorgan>
sadly that's a hard thing to see if you don't have a serial console :-(
<macromorgan>
only unsupported combos are downstream idbloader with upstream U-boot
<macromorgan>
note that for that image you'll need a proper boot.ini
<macromorgan>
(also downstream can't turn on kaslr, but that's no big deal I guess)
<macromorgan>
upstream can't run Optee and doesn't have video... it's tradeoffs all the way down
repk has joined #linux-rockchip
<alyssa>
Uhmmmmmmmmmmm
<alyssa>
ijustwanttobootdebian
<macromorgan>
take my image, write zeroes over the U-boot portion (dd if=/dev/zero of=linux_oga_bullseye.img bs=512 seek=64 count=2048), and create a boot.ini file and put it in the first partition
<macromorgan>
that should get you using that image with the stock U-Boot
<macromorgan>
alternatively, I think I know the reason why it has issues when you hold down the button... I can look at that later and figure it out...
phh has quit [*.net *.split]
macromorgan has quit [*.net *.split]
phh has joined #linux-rockchip
phh has joined #linux-rockchip
phh has quit [Changing host]
macromorgan has joined #linux-rockchip
<macromorgan>
nope, that wasn't it
<macromorgan>
I'll try to figure something else out
<macromorgan>
for now you can just write zeroes over offset 64x512 for about 2MB and it should boot with the onboard SPI... if you put a proper boot.ini in place
<macromorgan>
note that you'll have to delete all the random and chosen commands from the boot.txt, but you can do that and rename it to boot.ini (and put some special first line in place that I forget) and it will be accepted
<alyssa>
ack
<alyssa>
system failure, weee
<alyssa>
at least it's on jan 2021 u-boot now
<alyssa>
ok, now it's not failing, it's just hanging at the splash. Is that progress? Maybe? :|
<macromorgan>
probably due to an incorrect boot.ini maybe?
<macromorgan>
now if you're feeling adventurous flash mainline U-boot :-)
<alyssa>
macromorgan: thank you! and not adventurous at all, I just want to run my test suite so I can claim Mali G31 is conformant :-p
<alyssa>
mount: /mnt: mount(2) system call failed: File exists.
<alyssa>
uh, I should hope so
<alyssa>
you seem to have a frankendebian, debootstrapping fresh
<macromorgan>
which tree do you have posted as your "next" tree?
<macromorgan>
yeah, my Debian is pretty franken... I just debootstrapped into a disk image and then use that each time I need to write a new dev environment
<macromorgan>
also it doesn't help that I do all my kernel development on Windows
<macromorgan>
*ducks*
<alyssa>
..........never mind
<alyssa>
how does that even work?
<macromorgan>
WSL2
<alyssa>
like, WSL i get but
<alyssa>
oh, okay.
<macromorgan>
then scp it to my device and reboot
<macromorgan>
okay, tested your patch series and it seems to work (at least for kmscube until I can figure out how to get phoc/phosh working)
<macromorgan>
I'm experimenting with shells for the OGA
warpme_ has joined #linux-rockchip
<alyssa>
how would phosh work without a touch screen?
<vagrantc>
with annoying mouse movements?
<macromorgan>
probably
<macromorgan>
maybe use a joystick as a mouse? not sure
<macromorgan>
I'm just messing with that, plasma mobile, plasma TV, and ludo (have to figure out how to add external apps to it which would make it great)
<alyssa>
plasma TV of the worst seems least bad but idk
<alyssa>
i guess kodi on this thing would be fine with the joystick hooked up
<alyssa>
Aside - does anyone know why on some of my systems i need to type `sudo dhclient` explicitly to get internet up?
<alyssa>
hm install network manager would help wouldn't it :-p
<diederik>
I've seen it sometimes too, assuming you meant you need that command to get an IP address. IIRC, the NIC itself was up
<alyssa>
yeah
<diederik>
do you use /etc/network/interfaces[.d/*]?
<alyssa>
not intentionally
<alyssa>
I just installed network-manager and it works fine now
<diederik>
that can work too ;)
<alyssa>
version: "OpenGL ES 3.1 Mesa 21.3.0-devel (git-8e6e8243f2)"
<alyssa>
renderer: "Mali G31 (Panfrost)"
<alyssa>
that's a good sign, isn't it? :')
<alyssa>
libEGL warning: MESA-LOADER: failed to open ��Ά: /home/alyssa/lib/dri/��Ά_dri.so: cannot open shared object file: No such file or directory (search paths /home/alyssa/lib/dri)
<alyssa>
that is not a good sign.
WaluigiWare64 has quit [Read error: Connection reset by peer]
WaluigiWare642 has joined #linux-rockchip
<CounterPillow>
��Ά is perhaps my favourite gpu vendor
<robmur01>
macromorgan: heh, nowt wrong with WSL - most of my patches for the last year or so have been developed on an Envy X2 and Surface Pro X (even the benchmarking for my arm64 memchr() implementation), and I work on the kernel for a living :D
<robmur01>
although on that note, IT policy does requires me to install this month's Windows update by tomorrow, hey ho...
robmur01 has quit [Quit: Leaving]
Rathann has joined #linux-rockchip
robmur01 has joined #linux-rockchip
<alyssa>
robmur01: you don't get exempt from the windows policy being in OSS?
<robmur01>
nah, remember I'm the weirdo who'd rather join the Windows-on-Arm pilot than have Ubuntu on an x86 Thinkpad ;)
<robmur01>
what's great is that WSL works, VSCode works, Office works, Zoom and Teams work, the Cisco VPN works (even if it breaks networking in WSL2), I don't have to mess around playing sysadmin, and I get to feel smug about not using x86
<robmur01>
I didn't want to bring my giganto Linux-on-Arm lump home since I don't really have the space, and it can get kinda loud
<robmur01>
Shame it means I'm on Adreno rather than Mali, but you can't have everything :D
<chewitt>
about 50% the surface area of a credit card, and has a G31 inside
<alyssa>
chewitt: Speaking of G31, I just hit a stupid bug
<alyssa>
I'd call it a hardware bug but Arm would claim it's actually an implementation difference.. uh huh..
<chewitt>
errata?
<alyssa>
No... it was intended by someone...
<alyssa>
robmur01: CLPER.i32 encoding on G31 matches G72, not G76.
<alyssa>
Despite that G31 is v7, not v6.
<alyssa>
I guess it makes sense -- CLPER gained a bunch of useless features in v7. Save a tiny bit of silicon this way.
<CounterPillow>
"it's actually an implementation difference" is what rockchip is currently going with for their new messed up GIC in the rk356x :(
<alyssa>
gotta say, interrupt controllers are really not the place to have "implementation differences" ......
<robmur01>
if only they hadn't implemented it in a way that's objectively broken
<CounterPillow>
If only they'd publish an errata
<alyssa>
CounterPillow: oh no when it's an implementation difference you don't have to publish it ;-P
<robmur01>
alyssa: well, in fairness G31 is older than G76 - there's a minor version bump between them AFAICS
<alyssa>
robmur01: Even so... you don't change the ISA with only a minor version bump :|
<alyssa>
I mean. Arm does all the time but.
<robmur01>
encodings != ISA, at least if you're Midgard :P
<alyssa>
robmur01: Speaking of, should I allowlist G51 and G76 despite no linux hw existing for them? :-p
<alyssa>
maybe postmarketOS could make use of it
<alyssa>
And another failure. Wonder if this is also an "implementation difference" 🙄
<macromorgan>
your VPN also breaks WSL2? thought that was just me...
<robmur01>
alyssa: maybe if you feel confident enough - all I really have is a wiki page with high-level details and history of what was based on what, so I wouldn't trust my judgement on the practical idiosyncrasies
archetyp has quit [Quit: Leaving]
<robmur01>
I have a G51 bitfile I could play with on the FPGA, but no G76
<alyssa>
next question, why is this test timing out. Is that a driver bug?
<alyssa>
Or is G31 just so dog slow and the shader is just so cat fast?
<alyssa>
s/fast/complicated/
<alyssa>
Guess I should compile a kernel for this with the timeout bumped up
<alyssa>
Or I can just disable the extension and skip it 😇
WaluigiWare642 is now known as WaluigiWare64
<alyssa>
(on Mali G31 only. other systems unaffected.)
<robmur01>
it is only wee (even though the PX30 TRM implies it's big by G31 standards)
<robmur01>
does it at least have AFBC?
<alyssa>
The one in PX30 has AFBC, yes.
<alyssa>
Though from a bug report I learned there are AFBC-less G31...
<robmur01>
that was Amlogic G12A (S905something) wasn't it?
<alyssa>
maybe?
<alyssa>
Uh oh... am I drawing more current than USB supplies?
<alyssa>
maybe if I turn the display off it'd be happier
warpme_ has quit [Quit: Connection closed for inactivity]
hipboi has quit [Remote host closed the connection]