#asahi-dev · 2026-06-24 14:17
flokliyou might not need HEAD~2 if you're running 26.6 beta2#asahi-dev · 2026-06-24 14:11
floklienverbalalic: https://github.com/flokli/linux/tree/j773s#asahi-dev · 2026-06-24 13:53
jannausofus: messages with "HVLOG:" in the HV uart are passed through to the HV log. see 093f59ea ("hv_vuart: Pass through "HVLOG: ..." messages to the main console")#asahi-dev · 2026-06-24 12:39
enverbalalicthat would be great if you can#asahi-dev · 2026-06-24 12:38
flokliConsidering there's also t602x-common.dtsi, that's how it might end up getting upstreamed anyways#asahi-dev · 2026-06-24 12:37
flokliI can probably post my t6040 one, most of it might be applicable to t6041 as well#asahi-dev · 2026-06-24 12:15
sofuswhat is HVLOG?#asahi-dev · 2026-06-24 12:15
sofusjannau: thanks, https://github.com/sofus13/linux/tree/avd-restore-iommu#asahi-dev · 2026-06-24 12:07
jannausofus: might be an apple-dart quirk. if you push your branch I can take a look. I would start with tracing dart-avd and HVLOG prints around the domain switching#asahi-dev · 2026-06-24 12:03
yukajust m1n1 main and linux 7.1 or linux-next should be a good starting point to test your own device tree if you want to write one#asahi-dev · 2026-06-24 12:03
yukaor published, at least#asahi-dev · 2026-06-24 12:03
yukathere is not really one branch, and I don't think anyone wrote a device tree for t6041 yet#asahi-dev · 2026-06-24 12:02
enverbalalicI have a t6041 that i want to start poking around, who's repo / branch is the most up to date for m4 stuff right now ? yuka ?#asahi-dev · 2026-06-24 11:57
sofusi tried doing the same as https://lore.kernel.org/all/20250508-rkvdec-iommu-reset-v1-1-c46b6efa6e9b@collabora.com/#asahi-dev · 2026-06-24 11:57
sofusjannau: about the iommu restore, it does not restore the mappings for some reason?#asahi-dev · 2026-06-24 11:56
chadmedno none at all, just the dev_info we added and the errors about the fifo slots #asahi-dev · 2026-06-24 11:55
sofuschadmed: no are you getting timeouts? any "no handler for IRQ" logs?#asahi-dev · 2026-06-24 11:42
flokliBut this machine has 12 cores, so fairly high chance it reproduces whatever race there is easily without restricting.#asahi-dev · 2026-06-24 11:41
flokliso booting with maxcpus=6 also worked after a few attempts#asahi-dev · 2026-06-24 11:39
LarstiQflokli: heh, just musing about which thing makes it work after booting and what are the bits that might get scheduled on another core before that#asahi-dev · 2026-06-24 11:39
chadmed(not that it matters, the output is still garbage/uninitialised green) #asahi-dev · 2026-06-24 11:39
chadmedare we not clearing the vpdone interrupt properly#asahi-dev · 2026-06-24 11:38
chadmedsofus: we now get 0x102s but still invalid slots, which eventually become no free slots#asahi-dev · 2026-06-24 11:37
flokliLarstiQ: i'd need to probably do a lot more reboots to give you a somewhat confident answer to that ;-)#asahi-dev · 2026-06-24 11:36
sventhat really sounds like some weird race condition #asahi-dev · 2026-06-24 11:35
LarstiQdoes the probability of crashing scale linearly with the number of cpus?#asahi-dev · 2026-06-24 11:34
flokliyeah, so the higher I set maxcpus=, the less likely the chance for the bootup to succeed. maxcpus=4 is a bit hit and miss, any more cores than that I wasn't able to boot up the kernel without oopsing#asahi-dev · 2026-06-24 11:33
yukawhich either is the piece of code which crashes, or a piece of code which runs on the core before and messes things up#asahi-dev · 2026-06-24 11:32
yukaso this could be related to the chances of some piece of code running on a secondary core#asahi-dev · 2026-06-24 11:31
flokliso it correlates with the number of cores that are running. i wasn't able to successfully boot at all with all 12 cores. but i was able to boot up with maxcpus=2 just now#asahi-dev · 2026-06-24 11:30
jannaumaybe a slightly broken memory map?#asahi-dev · 2026-06-24 11:30
svenhuh#asahi-dev · 2026-06-24 11:28
yukathey only happen during boot, and only sometimes#asahi-dev · 2026-06-24 11:28
floklisorry, was missing the 'unable to handle kernel write' line. more logs of another boot: https://paste.linuxlounge.net/PMTU4BIEOV4UPN55VPS2V6OSBY#asahi-dev · 2026-06-24 11:28
yukaWe were not able to cause these crashes once it booted#asahi-dev · 2026-06-24 11:27
jannauerror looks mmu setup issue#asahi-dev · 2026-06-24 11:26
sofuschadmed: please try https://github.com/sofus13/avd-fw/tree/t8103-fix#asahi-dev · 2026-06-24 11:23
yukaI will try to get them reproduced again on my base M4#asahi-dev · 2026-06-24 11:23
floklihttps://paste.linuxlounge.net/AG72ORIPUUH6KEAN2FXWDVOADI#asahi-dev · 2026-06-24 11:23
yukaI don't think the other crashes are immediately related to wfi, they are just dependent on some race condition (some code running on the secondary core?)#asahi-dev · 2026-06-24 11:22
yukaidle=nop also removes the use of wfit in delay()#asahi-dev · 2026-06-24 11:22
yukathat is included in my patch#asahi-dev · 2026-06-24 11:22
jannauthere is still the wfit in delay() if that's not fixed differently#asahi-dev · 2026-06-24 11:21
flokliyuka: so the less minimal kernel config still crashes somewhat reliably without maxcpus=1. Maybe there's some other place using wfi, which we didn't catch with this patch?#asahi-dev · 2026-06-24 11:19
JamesCalligeros[m]6.6k*#asahi-dev · 2026-06-24 11:17
JamesCalligeros[m]Oh my 66kloc of slop#asahi-dev · 2026-06-24 11:13
yukaI assumed so, just need to figure out which header ;)#asahi-dev · 2026-06-24 11:13
svenwould move the enum to a header and add yield like will suggested, but looks good otherwise #asahi-dev · 2026-06-24 11:12
jannauchadmed: there's https://gitlab.freedesktop.org/sree/mesa/-/commit/5955e6eb0a03fdd0804b9b3ecf98d8681187c189 as well if can ignore the slop (I haven't really looked at it)