ChanServ changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - Logs https://oftc.irclog.whitequark.org/panfrost - <macc24> i have been here before it was popular
<anarsoul> alyssa: so you're actually doing development on macos?
<anarsoul> nevermind, still reading your latest blog post
<alyssa> anarsoul: nah, Linux on M1
<alyssa> bare metal Linux, no VMs or anything
<alyssa> admittedly I /am/ using macOS for reversing, but that has more to do with owning 0 x86 devices and macOS having Rosetta
<alyssa> HdkR: ...sorry ;p
<HdkR> Don't worry, I am using an M1 laptop as well
<HdkR> Asahi Mesa is more important
<HdkR> Just make sure to support VR once that laptop runs mesa :P
<HdkR> M1 Max running Half-Life Alyx would be pretty cool
karolherbst has quit [Read error: Connection reset by peer]
karolherbst has joined #panfrost
<Dylanger> <alyssa> "bare metal Linux, no VMs or..." <- In your opinion, how well is the M1 running Linux?
* alyssa shrugs
<alyssa> hello from my M1 Linux box?
<alyssa> if that answers your question :-p
<HdkR> It's fun seeing all the kernel patches coming
nlhowell is now known as Guest833
nlhowell has joined #panfrost
JulianGro has quit [Remote host closed the connection]
Guest833 has quit [Ping timeout: 480 seconds]
macc24_ has joined #panfrost
macc24 has quit [Ping timeout: 480 seconds]
macc24 has joined #panfrost
macc24_ has quit [Ping timeout: 480 seconds]
<robclark> alyssa: ironically last time I checked, drm-shim didn't actually work on arm64 linux
<alyssa> robclark: works for me
<robclark> hmm, maybe something was fixed in the mean time..
<alyssa> robclark: what wasn't working?
<robclark> I think it might even even just not compiled.. you'd have to ask six-month-ago-me :-P
robmur01 has quit [Read error: Connection reset by peer]
robmur01 has joined #panfrost
<alyssa> lol ok
<robclark> at the time I decided that debugging whatever I was debugging at the time was more important than my pride, so I just did it on x86 :-P
<alyssa> ;-p
<alyssa> My only issue was the 16K page thing
<alyssa> which I actually fixed in July or August but shhhh
nlhowell has quit [Remote host closed the connection]
nlhowell has joined #panfrost
chewitt has quit [Quit: Zzz..]
JulianGro has joined #panfrost
nlhowell has quit [Ping timeout: 480 seconds]
MajorBiscuit has joined #panfrost
tolszak has joined #panfrost
rasterman has joined #panfrost
chewitt has joined #panfrost
nlhowell has joined #panfrost
nlhowell is now known as Guest878
nlhowell has joined #panfrost
Guest878 has quit [Ping timeout: 480 seconds]
nlhowell is now known as Guest882
nlhowell has joined #panfrost
Guest882 has quit [Ping timeout: 480 seconds]
mps has joined #panfrost
nlhowell is now known as Guest891
nlhowell has joined #panfrost
<mps> does anynone how to force panfrost for xorg on armv7 exynos-5800 peach pi
<mps> module is loaded but xorg didn't tried to use it but use exynos-drm
<mps> works fine on gru-kevin (rk3399)
<mps> s/how/know how/
Guest891 has quit [Ping timeout: 480 seconds]
<robmur01> mps: I believe exynos-drm is right either way (the way Mesa works for separate 3d/display is a bit funky), but Panfrost will be bailing out because that particular GPU is not properly supported
<robmur01> you can bypass the allowlist in Mesa to get it to run, but rendering will be about 50% wrong and it might lock up the hardware
<mps> robmur01: aha, thanks for info. I thought to test it on peach pi because it works fine on gru-kevin, but if it is buggy then is not worth to try
<alyssa> mali-t720 is the oldest GPU we support upstream
<robmur01> technically T760 is older :P
<mps> alyssa: so kernel config mislead me saying it support 'DRM driver for ARM Mali Midgard (T6xx, T7xx, T8xx)'
<mps> peach pi has T628
<robmur01> yeah, the kernel driver supports it, but Mesa doesn't ;)
<mps> eh, pity, this chromebook has fine anti glare display
<alyssa> Panfrost kernel != Panfrost mesa
<mps> anyway, it is useful with exynos-drm fine, except for some apps which need gpu to work fine
<robmur01> (and that's "supports" as in works fine for T624, but is actually missing some of the horrible stuff to deal with larger T628 configurations, which doesn't currently matter as long as there's no supported userspace)
<alyssa> alas
<robmur01> let's hope Steve's hack is enough for now
<alyssa> robmur01: It's my understanding it's enough for r1p0 (the BE thing) but not for r0p1 (exynos)
<robmur01> except AFAIK r1p0 and r0p1 are actually the same thing
<alyssa> kbase disagrees
<alyssa> given the extensive list of ISSUE_* that only apply to r0p1
<alyssa> many of which with nontrivial workarounds
<robmur01> certainly the "r1p0" in Juno reports "major 0x0 minor 0x1"
<alyssa> huh
<alyssa> dunno then
<robmur01> although I've now just turned up an errata doc that implies they *are* distinct
<robmur01> also I know about T620 errata now
<alyssa> Womp womp
<alyssa> robmur01: I mean T760 and earlier errata was well documented in early kbase
macc24 has quit []
nlhowell has quit [Ping timeout: 480 seconds]
<alyssa> oh.. I understand how to read Bifrost GPU IDs now
<alyssa> arch major, arch minor, arch revision, project
<alyssa> arch major, arch minor, arch revision, product
<alyssa> (nibbles from left to right)
<alyssa> G52: 0x7212
<alyssa> G52 r1: 0x7402
<alyssa> both are v7
<alyssa> both are the G52 product (0x2)
<alyssa> but original G52 is architecture version 7.2, while new G52 is version 7.4
<alyssa> do we know what's new in 7.4? uhh
<alyssa> mostly this means we can expect to see a /lot/ more GPU IDs..
<alyssa> and that we haven't says more about the diversity of Bifrost hardware we've seen
nlhowell has joined #panfrost
<alyssa> it seems kbase only lists (arch major, product) to identify a product, and then uses revision to identify revisions
<alyssa> i.e. ignoring arch minor and arch revision altogether
<alyssa> I think what we're doing now is fine for now, and if it ceases to be so we'll deal with that then
tolszak has quit [Ping timeout: 480 seconds]
<alyssa> robmur01: I think the model table series is good to go now
<robmur01> IIRC the main justification for that minor arch bump in was that the VA size shrank
<robmur01> s/in//
<alyssa> Womp womp
<alyssa> 57 files changed, 4488 insertions(+), 451 deletions(-)
<alyssa> Well that really didn't make a dent in my upstreaming "fun".
<cphealy> Is the current PF code supporting both versions of G52? (7.2 and 7.4)
macc24 has joined #panfrost
<alyssa> yes
<cphealy> tnx
nlhowell has quit [Ping timeout: 480 seconds]
<alyssa> Umm did Panfrost CI get disabled when I wasn't looking
<alyssa> oh nvm the pipelines just got shorter and i looked in the wrong place?
<alyssa> or nothing changed and i'm losing my mind
<robmur01> hmm, I guess if I no longer need to carry my allowlist hack, there's little reason not to upstream the HDLCD patch for kmsro either...
<alyssa> robmur01: You now need to carry a model line
<robmur01> or rather, try to convince rasterman to do so :D
<rasterman> robmur01: i'm happy to do the mr for you
<alyssa> happy to merge the HDLCD patch even if I'm not so happy to merge the T6xx patch :-p
<rasterman> i actually never tried panfrost + hdlcd as the mali on juno was in the "urgh.. not the 6xx" list that everyone stopped caring about :)
<alyssa> Yep..
<rasterman> alyssa: hahahah they go together like peas in a pod
<rasterman> :)
<robmur01> oh, I missed that in the new version - I see G71 is still there, but I guess we don't have anyone using those so my comment is moot
<alyssa> oh, right, G71 is there. um. right
<alyssa> Meh, it's.. fine. Hidden feature.
<alyssa> G71 is working until proven otherwise....
<rasterman> thats probably the way to go
<rasterman> deal with the "there's a problem" cases when they happen
<alyssa> G71 r0p0 does have a bunch of errata that G72 r0p0 doesn'
<alyssa> 7940, 8138, 8206, 8343, 8463, 8456, 8438
<alyssa> 8463/8438 are kernel issues
<alyssa> (Need kernel workaroudn)
<alyssa> 8343 relates to YUV so we won't hit
<alyssa> 7940, 8138, 8206, and 8456 do legitimately affect mesa+g71
<alyssa> though none seem catastrophic
MajorBiscuit has quit [Quit: WeeChat 3.3]
<alyssa> Valhall upstreaming tracker: 57 files changed, 4483 insertions(+), 449 deletions(-)
<alyssa> (when those numbers all get to 0, valhall is upstream)
mps has left #panfrost [#panfrost]
macc24 has quit []
macc24 has joined #panfrost
lerrc has joined #panfrost
nlhowell has joined #panfrost
lerrc has quit [Remote host closed the connection]
Daanct12 has joined #panfrost
Danct12 has quit [Ping timeout: 480 seconds]
tolszak has joined #panfrost
<HdkR> alyssa: How many open PRs in that tracker so far? :P
<alyssa> HdkR: too many
<alyssa> and too few
<HdkR> :D
tolszak has quit [Remote host closed the connection]
tolszak has joined #panfrost
Guest762 has quit [Remote host closed the connection]
urja has joined #panfrost
jambalaya has joined #panfrost
nlhowell is now known as Guest932
nlhowell has joined #panfrost
floof58 has joined #panfrost
floof58_ has quit []
<alyssa> hm.. LD_TILE is causing a GPU sched timeout...
<alyssa> no fault
<alyssa> meanwhile, EXT_framebuffer_fetch just straight up doesn't work, no time outs tho
nlhowell is now known as Guest934
nlhowell has joined #panfrost