#
✕ nick:pitust
#asahi · 2026-04-11 12:20
pitusthmm okay. i was just confused because the numbers were suspiciously close
#asahi · 2026-04-11 11:44
pitusthmm, maybe it's not quite as close as i thought it was, although it's still very strange
#asahi · 2026-04-11 11:42
pitustdoes the m1 have anything like dynamic frequency scaling? i've been messing around with performance counters (i wrote a stupid lkm to make them readable from usermode) and with CNTVCT_EL0 (which self-reports to be a fixed 24 MHz clock) and it's very very close to a 3 GHz fixed rate which is very suspicious
#asahi · 2026-04-09 12:31
pitustyeah, true...
#asahi · 2026-04-09 12:25
pitust(although obviously i haven't looked at the actual impl, maybe (and probably) it isn't that simple)
#asahi · 2026-04-09 12:24
pitustwell, you want to lock access to the SPTM registers either way, right? because you don't want to allow disabling SPTM at runtime. so the code could literally be `if (should_use_sptm) { load_sptm(); } lock_sptm_regs();`...
#asahi · 2026-04-09 12:19
pitustidk, imo its more probable that it was removed because it made sptm easier to implement
#asahi · 2026-04-09 12:16
pitustalso, you could make this sort of "fake xnu" work while a huge amount of the available hardware to a mostly-normal linux env, by running everything in a VM (which isn't ideal for asahi since it breaks KVM and requires using SPTM APIs, but is almost certainly viable)
#asahi · 2026-04-09 12:12
pitustbut that's completely separate problem from a legal one, anyway; and my understanding of the arm license is that this kind of extension *is* allowed (although i could be completely wrong)
#asahi · 2026-04-09 12:11
pitustah.
#asahi · 2026-04-09 12:09
pitustinteresting, then
#asahi · 2026-04-09 12:09
pitustahh okay
#asahi · 2026-04-09 12:06
pitustwhich patch? the one that was rejected because linux didn't want to have features intended only for one CPU vendor?
#asahi · 2026-04-09 12:03
pitustand it doesnt make a huge amount of sense
#asahi · 2026-04-09 12:03
pitustthats pure speculation
#asahi · 2026-04-09 11:59
pitustand it doesn't really make sense, either: you can still access a fair bit of the features by making a fake XNU
#asahi · 2026-04-09 11:58
pitusti can't find any sources on this being a legal thing
#asahi · 2026-04-09 11:51
pitustapple has made changes to their chips for a very long time, including usermode-visible changes, and they have always been the only company allowed to do that afaiu
#asahi · 2026-04-09 11:50
pitustJamesCalligeros[m]: do you have a source on the legal dispute thing? or is it just speculation?
#asahi · 2026-01-31 19:11
pitustbut you can't move the linux space back (at least, not easily)
#asahi · 2026-01-31 19:11
pitustyou can only (easily) move the end, not the start, so the installer moves the end of the macos partition by however much space you set for linux, and puts linux there
#asahi · 2026-01-31 19:10
pitustand you usually can also only shrink them in one direction
#asahi · 2026-01-31 19:10
pitustbecause you can only make partitions bigger in one direction
#asahi · 2026-01-31 19:09
pitustoxbar: there are two parts to this question. first, how do you free up space on macOS (which is relatively easy), and the second is how do you give it to fedora (to which the answer is basically "you probably can't")
#asahi · 2026-01-21 11:10
pitusti don't think this behavior is *incorrect*, really, sooo
#asahi · 2026-01-21 11:08
pitustmy guess is that it doesnt stop the watchdog timer
#asahi · 2026-01-15 17:37
pitustah hm
#asahi · 2026-01-15 17:35
pitusthow do you invoke swayidle?
#asahi · 2026-01-15 17:35
pitustno more like swayidle config issue
#asahi · 2026-01-15 17:34
pitustGlanzmann: feels like a config issue, tbh
#asahi · 2026-01-14 15:32
pitustoh hmm then its different i think
#asahi · 2026-01-14 15:32
pitusti have had similar issues and running `(modprobe -r brcmfmac_wcc; modprobe -r brcmfmac; modprobe -r brcmutil); modprobe brcmutil && modprobe brcmfmac && modprobe brcmfmac_wcc && sudo systemctl restart wpa_supplicant NetworkManager` helped for me
#asahi · 2026-01-14 15:31
pitustdoes the issue go away on its own?
#asahi · 2026-01-13 20:24
pitustwell, it needs to be done anyway
#asahi · 2026-01-13 20:23
pitustcouldn't you make the mmio tracer work with only one core?
#asahi · 2025-12-18 09:16
pitusti mean personally i would give it 64 kilobytes, but yeah makes sense
#asahi · 2025-12-18 09:14
pitustthe kasan reports i saw suggested it got allocated in an 8k arena (although that might just be kasan padding, not sure)
#asahi · 2025-12-18 09:13
pitustalso, the buffer size of 4000 bytes doesn't stop it from being allocated 8k of data it seems
#asahi · 2025-12-18 09:13
pitustbecause if there is *something else* which can cause the data to be wrong, it would be nice to at least get a warning
#asahi · 2025-12-18 09:12
pitustchaos_princess: maybe you should also add a WARN_ON() if the packet length is over the buffer, as well?
#asahi · 2025-12-17 23:23
pitustah okay
#asahi · 2025-12-17 23:22
pitustyou may want to poll the device until it gives you a valid report
#asahi · 2025-12-17 23:22
pitustalso, that change doesn't feel great
#asahi · 2025-12-17 23:22
pitustso, yeah
#asahi · 2025-12-17 23:22
pitustreproducing a race condition reliably is kinda difficult
#asahi · 2025-12-17 23:21
pitustm1 mbp
#asahi · 2025-12-17 23:17
pitusti could repro it quite reliably
#asahi · 2025-12-17 23:17
pitustinteresting
#asahi · 2025-12-17 23:08
pitusti'd be happy to test it (and i don't really love the idea of an msleep(100) being load-bearing...)
#asahi · 2025-12-17 23:07
pitustdo you have a real fix that isn't just msleep(100) figured out?

plus ›