#asahi-dev · 2026-07-28 22:01
sven~none#asahi-dev · 2026-07-28 19:51
svenworth a try i guess#asahi-dev · 2026-07-28 19:49
sveni think if we really need userspace jit to work for some reason i'll just patch the msr/mrs in that .dylib to a nop and map that sprr index as rwx#asahi-dev · 2026-07-28 19:40
svenso does that trap from el0 all the way up to el2 even with el1 inbetween?#asahi-dev · 2026-07-28 19:28
sventhere's some weirdness with the bridge#asahi-dev · 2026-07-28 19:27
svenyou should no longer be banned now#asahi-dev · 2026-07-28 19:27
sventhat was supposed to unban him *sigh*#asahi-dev · 2026-07-28 19:26
svenbah#asahi-dev · 2026-07-28 19:26
svenprobably over-eager anti-spam protection#asahi-dev · 2026-07-28 19:01
sventhanks for the offer, but i think there's enough stuff i can make progress on before that thing arrives#asahi-dev · 2026-07-28 19:01
svenand i can probably get started with sptm on my m1.#asahi-dev · 2026-07-28 19:01
svenyuka: probably won't have time to continue before the weekend and there's at least one GL0 inaccuracy i need to fix anyway#asahi-dev · 2026-07-28 18:13
svennext up: time to get a M4 and see what SPTM/TXM needs additionally#asahi-dev · 2026-07-28 18:11
svenhttps://github.com/AsahiLinux/m1n1/pull/643 the cursed thing#asahi-dev · 2026-07-28 14:25
svenhmhm.. no idea where that mask comes from then#asahi-dev · 2026-07-28 14:22
svenno idea, those bits might've just been a guess because the next field starts at 60#asahi-dev · 2026-07-28 08:56
svenwe don't need this to work for tracing, would've just been fun#asahi-dev · 2026-07-28 08:56
svenit's just not important though#asahi-dev · 2026-07-28 08:55
sveni'd just patch the library and make it do what i want#asahi-dev · 2026-07-28 08:50
svenor see if i can mess with the commpage in some way to make it think it has userspace sprr but then not actually try to mrs/msr#asahi-dev · 2026-07-28 08:49
sveni can only trap that register up to EL1 but not to EL2. i'd just patch libsystem_pthread or whatever at that point#asahi-dev · 2026-07-28 08:47
sveni could probably hack this by just mapping the pages as rwx but meh.. this will be good enough for tracing#asahi-dev · 2026-07-28 08:46
svenyeah, that's obviously what happens#asahi-dev · 2026-07-28 08:23
sven\o/#asahi-dev · 2026-07-28 07:23
svenpity, aprr_jit=0 disables the userspace JIT stuff but safari then somehow does millions of genter/exit calls so that it doesn't start anyway :(#asahi-dev · 2026-07-27 22:03
svenisn’t it enabled by default?#asahi-dev · 2026-07-27 21:47
svenxnu under the cursed hv complains about heavy cpu usage from random processes here because I can’t account for the time I steal in the hv but at least it doesn’t kills anything or panic #asahi-dev · 2026-07-27 21:43
svenouch, does it not like us stealing too much time?#asahi-dev · 2026-07-27 21:33
svenyeah. i'm glad he noticed because i probably would've spend a long time auditing my code to find that#asahi-dev · 2026-07-27 21:32
sventhat#asahi-dev · 2026-07-27 21:32
sven<jannau> you could try without "clpc=0" in boot-args if you have it. I suspect that it prevents the use of performance cores#asahi-dev · 2026-07-27 21:23
svenhttps://f.svpe.de/9273c4c464e6f1a9d9ec631c37c9b3f4174eab1b85c9598992f0ffb5e2806c69_fast.jpeg still a bit of overhead but much better :-)#asahi-dev · 2026-07-27 21:11
svenlol, and by not doing dumb shit like taking the bhl for everything this is actually usable now and doesn't lag anymore#asahi-dev · 2026-07-27 20:07
sven 14 files changed, 1412 insertions(+), 42 deletions(-) surprisingly small diff for the core of this cursed thing :D#asahi-dev · 2026-07-27 19:14
sventime to clean this up i guess#asahi-dev · 2026-07-27 19:11
svenand anything like safari that uses userspace jit will just crash#asahi-dev · 2026-07-27 19:10
svenalright, this should probably be fast enough for hw tracing except for maybe gpu.#asahi-dev · 2026-07-27 18:34
sveni actually removed the clpc node from the ADT because i ran into a panic when trying single cpu boot#asahi-dev · 2026-07-27 18:33
svenjannau: yup, that seems to have fixed it. thanks!#asahi-dev · 2026-07-27 18:25
svenyou can see just how inefficient this entire thing is though. most of the red part is probably the HV doing its thing with pagetables#asahi-dev · 2026-07-27 18:22
svenhttps://f.svpe.de/ed300e454de66186311e0059c43330f49532f4020a4c0e1c542f4c9dcace06b9_smp.jpeg no idea why it doesn't want to use the performance cores :(#asahi-dev · 2026-07-27 16:59
svenstill crashing with smp, but in a different way now :D#asahi-dev · 2026-07-27 14:56
svenoh, lol, the corruption is because i've eaten all of m1n1's heap and kept writing into the guest's memory space :D#asahi-dev · 2026-07-27 10:35
svenlooks like my smp issue is some truly weird memory corruption. this is gonna be fun to track down again :(#asahi-dev · 2026-07-27 09:13
svensptm hans't hit m1 yet in any macos release, has it?#asahi-dev · 2026-07-26 21:51
svenunfortunately SMP also dies very soon after xnu brings up the 2nd core :(#asahi-dev · 2026-07-26 21:51
svenlol, as expected safari just dies immediately because i ignore the JIT SPRR crap entirely