stikonas has quit [Remote host closed the connection]
fysa has quit [Read error: Connection reset by peer]
<endrift>
I think it's trying to load the firmware for the wrong wifi module, and failing because that firmware doesn't exist (and I can't even find it online)...
<anarsoul>
just comment wifi out for now in dts?
<anarsoul>
btw what firmware it's looking for?
<endrift>
yeah but I want wifi :( and brcmfmac43456-sdio
<endrift>
the dts says it's an ap6354, which doesn't seem to align to that brcm
<endrift>
well, I'll make that symlink and see what happens
<anarsoul>
you still need brcm/brcmfmac43456-sdio.txt
<endrift>
right I know
<HdkR>
https://imgur.com/a/Su7hHVs Anyone have experience with EFI Stub stopping after exiting boot services? :P
<EmilKarlson>
HdkR: are you sure it's just not some other early boot problem?
<EmilKarlson>
I'd guess that maybe it uses EFI services to print the first lines to screen and never gets to a point where it would print more after initializing serial port natively
<HdkR>
Well it is generating an empty DTB as well, so I guess it never found the DTB I generated?
<anarsoul>
I guess
<anarsoul>
I suppose you don't have uart?
<HdkR>
Not that I'm aware of anyway
<EmilKarlson>
oh yes, sorry that was display directly, not serial port
<EmilKarlson>
that is obviously a bit more complex then
<EmilKarlson>
what device is that?
<HdkR>
Surface Pro X
vstehle has quit [Ping timeout: 265 seconds]
<HdkR>
Apparently you're able to do earlyprintk=efi but I get nothign with it
<HdkR>
Is appended_dtb no longer an option?
<HdkR>
I have no idea how to use DTB :D
<HdkR>
Ah, looks like if the kernel can't pull the dtb from uefi then if you have secure boot disabled then you can pass in dtb=<dtb location> which is part of the rootfs
NeuroScr has quit [Quit: NeuroScr]
<HdkR>
EFI's exit_boot_services is hanging, hmmm
chrisf has quit [Ping timeout: 264 seconds]
chrisf has joined #panfrost
<HdkR>
hah! what!? Penguins!
* HdkR
breaks out slow motion camera to figure out wtf is happening
<HdkR>
Looks like arm-smmu kicks up and then the screen vanishes
<HdkR>
robclark: Do you have any experience with a device reseting once smmu comes up? :D
<robclark>
HdkR: if trying to run upstream kernel, and you have some device behind smmu in active dma (ie. display scanout), then pour yourself a beer to cry into ;-)
<robclark>
upstream is completely unprepared to deal with that
<HdkR>
Hm
<robclark>
for drm/msm we really just want the smmu to not touch anything until we are ready to take over (since not using dma api).. I've tried a few approaches to that.. but nothing upstream has been happy with
<robclark>
tbf, clk and genpd issues are there too.. so really arm-smmu is just the hilarious cherry on the fail-sundae
<alyssa>
Tasty
<HdkR>
I see
<HdkR>
I currently have msm disabled even for testing
<robclark>
and an attempt at keeping arm-smmu out of the way that mostly work
<robclark>
it's what I'm using on my yoga c630
<robclark>
which has basically all the same problems you are dealing with
<HdkR>
Nice. I'll grab that branch and see if it gets me any farther
<alyssa>
robclark: Is that your main machine yet? :)
<robclark>
(also jfyi, #aarch64-laptops and for general snapdragon stuff ##linux-msm)
<HdkR>
oo
<robclark>
alyssa: not quite, but I hear progress is being made on wifi.. which is really the only thing that keeps it relegated to "an extra board to test mesa stuff on"
<alyssa>
robclark: Pff :p
* alyssa
looks around
<alyssa>
I don't believe I have any non-Arm devices in this room
<alyssa>
Well, I suppose the thermostat might be some ancient 8-bit micro :)
<robclark>
I do really need to rebase and repost a bunch of non-drm patches for it though.. although in my defence XDC and moving and other things
<robclark>
HdkR: btw, if you boot into windows in the near future on the thing, I'm a bit curious how https://playcanv.as/b/SFIt5Ode/ does.. it completely falls over for qcom android blob.. mostly works with freedreno/a6xx (after some gpu hangs on "draw a quad with insane frag shader".. after that there is some mis-rendering but basically the same probs I have on my intel laptop)
<robclark>
(I guess I can check on my c630 tomorrow but I left it at office)
* HdkR
looks
<HdkR>
pfft. Edge just stalls
<robclark>
actually, I think it was rendering ok on my laptop before upgrading to f31 a few days ago.. so maybe the problems are a more recent nir issue?
<HdkR>
Seems to work in Firefox
<HdkR>
~20FPS on High
<HdkR>
Also ~20FPS on Low. Task Manager claiming 100% GPU utilization
<robclark>
hmm, I didn't try high yet.. wasn't quite getting 20fps, but otoh it was on something that is somewhat less than half of an a680
<robclark>
heh
<HdkR>
Low with MSAA disabled is ~30fps
<robclark>
I tried but failed to get an apitrace of it.. I'm a bit curious what it does.. it's not all "shadertoy" but it does seem like there is some "shadertoy" style draws happening early on for some reason.. where I was seeing gpu hangs was a monster frag shader with a whole bunch of sets of 3-level-deep nested for loops
megi has quit [Ping timeout: 268 seconds]
<robclark>
so I could totally believe it didn't really hang, it just didn't complete in time..
<HdkR>
Interesting
* robclark
not sure if there is some sane way to debug webgl stuff
<HdkR>
I've never done it to know
* robclark
wants renderdoc for webgl :-P
<HdkR>
It feels like everyone does
<anarsoul>
alyssa: thermostat is likely atmega (if it's one from honeywell)
<anarsoul>
I disassembled one before installing it ~2y ago
<robclark>
HdkR: maybe it's just trying to mine some bitcoin :-P
<HdkR>
2TFLOP in such a small power budget is quite nice
<alyssa>
robclark: webgl is itself the insane part.
<robclark>
yeah, I really want me some a680.. the lesser a6xx were already quite smoking in the tflop dept (*if* you can keep the register footprint down and still schedule for ILP)
<HdkR>
Samsung's will probably be the best option when it ever releases
<HdkR>
This ProX is dumb expensive
<robclark>
ping me when sammy ships.. I have 8cx on my christmas wish list (for me to buy myself)
<HdkR>
:D
<HdkR>
Wouldn't be surprised if it ships this month finally
<HdkR>
"Blow for Windows 10 on ARM as Snapdragon Samsung Galaxy Book S rumoured to be cancelled in favour of Intel version"
<HdkR>
Unless that actually ends up being a thing...
<alyssa>
Samsung Chromebook Plus
<HdkR>
alyssa: Gimme one with a top end Samsung chip in it :D
<alyssa>
HdkR: Interested in writing a bifrost compiler? :^)
<HdkR>
lol, why do people keep asking me this?
<HdkR>
alyssa: Are you saying that you have a product in hiding that I could get if I write a bifrost compiler? :P
chewitt has quit [Read error: Connection reset by peer]
<HdkR>
Building this c630 kernel
davidlt has joined #panfrost
<HdkR>
"found a panel-id!"
<HdkR>
Blah, still unable to find the dtb from command line for whatever reason
<robclark>
HdkR: so I guess you need a bunch of dt additions.. as far as panel, maybe look at ' WIP: dt updates for multiple yoga c630 panels' (although there is still not upsteam conscensus on how to deal with that.. but as long as you are the only person trying to run linux on the thing maybe you can ignore other panel options)
<robclark>
the "found a panel-id" comes from uefi variable, fwiw
<robclark>
see "efi/libstub: detect panel-id"
<HdkR>
hm
<HdkR>
Trying to figure out why it fails to find my dtb and is making an empty one
<robclark>
so.. some of that is that we have enough people with c630's that we have to deal with the multi-panel issue.. you probably don't have that problem yet on the surface..
<robclark>
but yeah, I guess you need to sort out getting right dtb loaded first before you worry about these problems
<HdkR>
Right, should only be one panel, they even released a video claiming they are really proud that every panel is calibrated to be the exact same
<robclark>
(don't believe marketing ;-))
<HdkR>
Of course
<robclark>
anyways, sounds like you need to do some dt hacking before you need to worry about multi-panel ;-)
<HdkR>
Seems like it's just ignoring my dtb= argument
<robclark>
hmm.. I was loading dtb w/ devicetree command in grub
<HdkR>
I don't even have grub
<robclark>
I think it needs to be loaded as part of the bootloader->kernel transition
<robclark>
so I don't think kernel cmdline arg helps
<robclark>
probably if kernel starts w/out dt it assumes acpi??
<HdkR>
Yea, it goes through acpi if you don't provide the argument
<HdkR>
and ACPI doesn't give one so that isn't going to work there
<robclark>
we are a bit far from more than basic stuff working with ACPI.. linux needs to learn some new PEP tricks for acpi to happn..
<robclark>
but dt can be loaded before kernel starts and passed to kernel from bootloader
<robclark>
I've done that so far w/ grub
<HdkR>
Okay, I hardcoded my dtb location in to the kernel source
<robclark>
but leiflindholm in #aarch64-laptops had a uefi module thing that would load dt
<robclark>
hmm
<robclark>
ok, I've never tried that path.. no idea what happens there
<robclark>
anyways, probably you can ignore panel and those sort of things and try to get kernel working with no drm/msm but using efifb
<HdkR>
Yea, that was the plan. Just disabling drm entirely for now
<HdkR>
oh, it uses efi_convert_cmdline to get the command line. It doesn't use the kernel options for it
<HdkR>
That's why it wasn't being passed over
<HdkR>
Hm, haven't been able to get any penguins on this kernel at all
<robmur01>
HdkR: yeesh, just take a GRUB binary from one of the existing aarch64-laptops images ;)
<robmur01>
AIUI the EFI stub DTB override thingy is super-dodgy and people would rather it went away entirely
<robmur01>
endrift: 3-minute boot time on RK3399 means you're missing early PLL initialisation - use the upstream SPL/TPL instead of the RK miniloader, or hack it into the main stage to match what the downstream u-boot does
<robmur01>
otherwise, your big cores are at 12MHz, while the little ones are semi-randomly at 200 or 400 (depending mostly on warm vs. cold reset, seemingly)
<alyssa>
/* The less questions you ask, the better */
<alyssa>
really inspiring to read
<alyssa>
especially inspiring when i wrote that comment >.<
kaspter has quit [Remote host closed the connection]
kaspter has joined #panfrost
fysa has joined #panfrost
gcl has joined #panfrost
fysa has quit [Ping timeout: 240 seconds]
maccraft123 has joined #panfrost
<alyssa>
Trying to sort out a bunch of disasm issues prompted by new texturing stuff
<Lyude>
HdkR: let me know how that surface book works out
fysa has joined #panfrost
<alyssa>
kevin tbh
maccraft123 has quit [Quit: WeeChat 2.6]
megi has quit [Quit: WeeChat 2.6]
fysa has quit [Remote host closed the connection]
<alyssa>
Erg
<alyssa>
GNOME regressed on master it looks like, let's bisect this thing :|
<TheCycoONE>
Should I try sddm on master again? (the shader regression you said you were pulling out anyway)
<TheCycoONE>
how does it compare to linear scan ra?
<alyssa>
Presumably better seeing as it can model control flow just fine (linear scan can't)
<alyssa>
I know the Fropessionals (TM) tend to hybrids of graph colouring globally and linear scan locally .. this is a lot simpler and does well.
<TheCycoONE>
Fropessionals :D
* TheCycoONE
misses 4th year algorithms
<TheCycoONE>
10+ years in the industry gluing widgets together and I forget everything
yann has joined #panfrost
stikonas has joined #panfrost
chewitt has quit [Quit: Zzz..]
chewitt has joined #panfrost
<endrift>
robmur01: I'm not entirely sure what that entails, but at this point I'm pretty resigned to waiting for the Manjaro ARM release. Managing this is becoming more trouble than it's worth for me right now
unoccupied has joined #panfrost
unoccupied is now known as leah
leah is now known as _4of7
_4of7 is now known as unoccupied
unoccupied has quit [Client Quit]
chewitt has quit [Quit: Zzz..]
davidlt has quit [Remote host closed the connection]
<HdkR>
robmur01: Problem is that I don't know what images support ARM64+UEFI+Grub there :P
krh has joined #panfrost
maccraft123 has joined #panfrost
NeuroScr has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
kherbst is now known as karolherbst
<HdkR>
Why is building grub2 such a pita?
maccraft123 has quit [Quit: WeeChat 2.6]
<anarsoul>
is it?
<anarsoul>
then you just need some tzatziki for it
<HdkR>
:D
<HdkR>
Stick a falafel in it
* alyssa
is hungry
<HdkR>
Me too now
raster has joined #panfrost
<alyssa>
Oops.
pH5 has quit [Ping timeout: 268 seconds]
pH5 has joined #panfrost
<enunes>
fedora has grub2 working on aarch64 and is pretty much all about uefi (or uboot+efiboot) + grub2
<anarsoul>
enunes: have you got PBP yet? :)
<enunes>
not sure, I'm not familiar with that
<anarsoul>
pinebook pro
<alyssa>
enunes: yas join the frosty side.
<anarsoul>
alyssa: I'm still working on lima :P panfrost will be my side-side project (since lima is side project)
<enunes>
ah, I don't know either and I don't have one to test, though fedora 31 reported to have some improved support for some rk3399 platforms and panfrost
<alyssa>
anarsoul: is your name enunes ;p
<anarsoul>
nope
<anarsoul>
alyssa: you should get a PBP too!
<alyssa>
I mean
<alyssa>
I already have 3 RK3399 devices in this room and 2 of them are laptops :p
<anarsoul>
yeah
<anarsoul>
so?
<alyssa>
\o/
<anarsoul>
you can't have too many RK3399 devices
<enunes>
that was just a drive by comment as I read the backlog and saw something from HdkR about images with grub2 :)
<HdkR>
enunes: Does Fedora provide an image that I can rip their grub2 build off of?
<HdkR>
Nice. I'll nab that image and see if I can see an EFI partition
<anarsoul>
iirc it's iso image
<enunes>
that one is a xz'ed raw image that you can just dd, the iso image is in the similar url but /iso/
<enunes>
the fedora image "installer" will basically dd that image and then dd uboot in a blank space between the mbr and the first partition
<HdkR>
I see an EFI folder
<HdkR>
I'll dump that on to my drive and see if it boots grub
<HdkR>
Once I'm done working though
<HdkR>
mount -o ro,loop,offset=<...> is great stuff :P
<anarsoul>
:)
<anarsoul>
'losetup -P' is also nice
<anarsoul>
(-P for partscan)
<enunes>
I do losetup -f -P <image.raw> and them mount /dev/loopNpX
<HdkR>
Interesting
<HdkR>
Never heard of that one
<stikonas>
isn't dd'ing system onto disk bad practice? I.e. if disk has 4096 byte logical sector, that would fail, you would have to fix partition boundaries...