05:09 <sternez> jannau: sorry, yesterday I spent too much time fixing m1n1 on another computer. dcp_iboot_m2pro.py's patch and log: https://paste.centos.org/view/8f638370

05:11 <sternez> I also modified the line: # [cpu0] [0xfffffe0014c9678c] MMIO: W.4 0x39c000010 (lpdptx-phy0[1], offset 0x10) = 0x0

05:11 <sternez> p.write32(self.phy_regs[1] + 0x10, 0x1)

06:46 <jannau> please test https://github.com/AsahiLinux/m1n1/pull/329 - HDMI output support for m2+ desktops, requires proxy mode to power the dp2hdmi converter on

13:22 <sternez> jannau: log after upgrading the m1n1: https://paste.centos.org/view/1a666458

13:31 <jannau> sternez: please try to add 'pmgr_adt_power_enable(cfg->dptx_phy);' next to the other one in dcp_init()

13:47 <pb17> I created a PR with a fix for wifi AP mode: https://github.com/AsahiLinux/linux/pull/184

13:47 <pb17> Since I only have a j316s it would be cool if someone else could check if it fixes AP mode on other devices as well

13:49 <sternez> jannau: no crash now, https://paste.centos.org/view/21143949

13:50 <jannau> but no picture?

13:53 <jannau> can you run m1n1 under the hypervisor `tools/run_guest.py -m hv/trace_dcp.py -m hv/trace_dptx.py -l dcp_dptx.log -r ../build/m1n1.bin`

13:54 <jannau> edit hv/trace_dcp.py and set dcp_name to "dcpext4"

13:54 <sternez> jannau: when I run it again, different message: https://paste.centos.org/view/fd461aa1

13:55 <jannau> that looks like it worked. did it?

13:56 <sternez> jannau: actually I have no monitor, but I have a usb hdmi capture, when I play the hdmi output with ffplay, it showed "unsupported video format", before the message is "no video output"

13:57 <sternez> jannau: I think so, I'll try to run the hypervisor now

13:57 <jannau> does the hdmi capture work with macos?

13:57 <sternez> yes

14:00 <jannau> that sounds like m1n1 is unlucky in selecting a color mode

14:01 <sternez> https://paste.centos.org/view/6772bfb8

14:05 <jannau> that looks truncated, which hdmi capture device is this? a ms2109/ms2130 one?

14:08 <sternez> jannau: I borrowed a monitor, now I do see the Asahi logo, I think it works on m2 ultra studio now. great job!

14:20 <jannau> thanks for testing, pushed the power on change to the PR

14:22 <jannau> you should be able to boot linux now, probably needs pd_ignore_unused to avoid turning the display out off

14:55 <sternez``> jannau: you are welcome, I would like to learn something to make some help. still some problems here: 1. I have to run `python experiments/dp2hdmi.py` three times for the asahi to show up. 2. then I started a hypervisor with command `make -C ../ -j10 && sh tools/run_guest_kernel.sh ../../linux/build/edge 'debug rootwait pd_ignore_unused root=/dev/nvme0n1p6 rootflags=subvol=root' firmware.cpio.gz`, I saw the boot logs printed on the

14:55 <sternez``> monitor, about 20 seconds later everything disappeared except the asahi logo. I traced the hypervisor: https://paste.centos.org/view/c458f59b, and the dmesg https://paste.centos.org/view/a9fc183d , forget the kernel version, I switched to your branch: https://github.com/jannau/linux/commits/asahi-6.4-9-dcpext_dp_t602x

15:02 <sternez``> sorry, it's two times, not three

16:19 <marcan> pushed m1n1 v1.3.4 to asahi-dev with a really, really stupid bugfix that should improve performance but needs testing on all major chip variants to make sure I didn't break the world

16:31 <jannau> kcbench is still at the value I remember from m1 ultra bringup which should predate the the bug

16:40 <marcan> ah so this at least came after ultra, less likely to break the world then

16:40 <marcan> might break t8112/t602x though

16:40 <jannau> t6002 and t8112 appear ok. no noticable perf change on kcbench

16:46 <jannau> sternez: do not use that branch, it's know broken and requires different m1n1. it is just old test branch which was used for GPU tests which "required" a display

16:48 <jannau> marcan: fails on t6020: https://gist.github.com/jannau/691b0bfb604191673650665ee331d842

16:49 <marcan> whoops

16:49 <marcan> no surprise there, I just winged that

16:49 <marcan> sigh

16:55 <marcan> found another bug on t600x that would make it not really work, heh

17:13 <mps> with mesa asahi 20230903 and Xorg firefox crashes X server on start. Alpine linux with kernel 6.4-10

17:13 <mps> with 20230902 it works fine

17:14 <mps> not important for me, just reverted back to mesa asahi 20230902

17:16 <mps> ah there is 20230903b now

17:59 <marcan> jannau: try 1.3.5 on t602x

18:07 <marcan> jannau: try https://github.com/clamchowder/Microbenchmarks/tree/master/MemoryLatency , that shows a definite improvement on t6000 around the point where it spills from L2 to SLC

18:18 <jannau> that survices at least m1n1 but I have trouble booting into linux. systemd apparently decides to reboot immediately

18:18 <marcan> that sounds... weird

18:23 <jannau> I think it's for selinux relabeling

18:25 <marcan> oh yeah that would happen

18:25 <marcan> did it work the second time?

18:26 <marcan> I pushed a fix for t8112 that may or may not matter

18:27 <jannau> no, it constantly failed but succeded now with selinux disabled

18:28 <marcan> I'm very confused as to how that m1n1 change could break selinux

18:30 <jannau> seems to work at least well enough to boot fedora. I blame booting via hypervisor/chainload on breaking selinux. I'd assume I'll see the same issues with m1n1-1.3.3

18:30 <marcan> tagged v1.3.6 with the t8112 fix then

18:31 <jannau> did you push main as well? for v1.3.5 it was just the tag

18:32 <marcan> yes

18:46 <marcan> I'll push 1.3.6 to prod, seems to be ok

18:49 <jannau> yes, works here on t6020 and the selinux behavior is the same with m1n1 1.3.3

18:53 <jannau> the memorylatency tests shows a clear improvement starting at 16384 on t6020 as well

18:54 <marcan> excellent

18:54 <marcan> I should get some sleep, but if you get a chance t8112 tests would be good too :)

18:54 <marcan> t8103 is tested and improved, so just t8112 left

19:10 <jannau> also clear improvement on t8112 between 12288 and 32768

20:34 <jannau> sternez: are you just running into proxy timeouts? try if `M1N1TIMEOUT=12 ./experiments/dp2hdmi.py` works at the first try