#asahi-dev · 2026-04-18 08:10
yukaof rebooting after 120s regardless of the known watchdog being disabled#asahi-dev · 2026-04-18 08:10
yukacan you remind me again, do you have the same problem on the macbook neo?#asahi-dev · 2026-04-18 08:05
yukabut once it's booted into userspace it always lasts for the 2 minutes until it gets killed by the watchdog#asahi-dev · 2026-04-18 08:05
yukaSo like what seems to be least likely to run into this crash is adding save & restore around the wfi and replacing wfit with nop in barrier.h. But again very possible that that's just a timing thing and not actually causality.#asahi-dev · 2026-04-18 08:00
yukaAh no it's just flaky, now got the same crash with pd_ignore_unused#asahi-dev · 2026-04-18 07:59
yukaI'll take it#asahi-dev · 2026-04-18 07:59
yukabut the pd_ignore_unused seems to help anyways#asahi-dev · 2026-04-18 07:58
yukaand thought that was the start of the crash#asahi-dev · 2026-04-18 07:58
yukaI was looking at "Unable to handle kernel write to read-only memory at virtual address ffff800003a0c000" way before#asahi-dev · 2026-04-18 07:57
yukaOhh#asahi-dev · 2026-04-18 07:57
yukapd?#asahi-dev · 2026-04-18 07:51
yukaLike with many combos which theoretically should work I get crashes like this: https://termbin.com/e0uq#asahi-dev · 2026-04-18 07:44
yukabut I'm not sure if that's a race condition#asahi-dev · 2026-04-18 07:44
yukanop also does not work btw#asahi-dev · 2026-04-18 07:42
yukawfi->wfe gets stuck#asahi-dev · 2026-04-18 07:41
yukanickchan does not work#asahi-dev · 2026-04-17 20:38
yukaOk got it working#asahi-dev · 2026-04-17 19:30
yuka(source: https://lists.infradead.org/pipermail/linux-arm-kernel/2022-April/735274.html)#asahi-dev · 2026-04-17 19:30
yukaso this msr _is_ the wfit instruction apparently#asahi-dev · 2026-04-17 19:29
yuka> Note that the encoding is that reported by objdump, as no currrent toolchain knows about these instructions yet.#asahi-dev · 2026-04-17 19:27
yukathough I do not see how wfit instructions are generated by these defines#asahi-dev · 2026-04-17 19:26
yukawfit is still a thing#asahi-dev · 2026-04-17 19:26
yukaOh!#asahi-dev · 2026-04-17 19:22
yukabut I can not figure out which wfi causes this#asahi-dev · 2026-04-17 19:22
yukato me it looks again like losing register state, because most registers are zeroed#asahi-dev · 2026-04-17 19:22
yukaNow I'm somewhat consistently running into the following: https://termbin.com/cdky#asahi-dev · 2026-04-17 19:21
yukaI could not identify any other relevant wfis in inline assembly#asahi-dev · 2026-04-17 19:21
yukaSo I replaced the wfi in arch/arm64/include/asm/barrier.h with a nop#asahi-dev · 2026-04-17 08:35
yukasven: doesn't the find_regs already check if I can write to the regs, and the result was that all are readonly? or is there possibly still some registers missed by find_regs#asahi-dev · 2026-04-17 08:32
yukayep#asahi-dev · 2026-04-17 08:19
yukawhat is NV?#asahi-dev · 2026-04-17 08:18
yukaI guess I could start m1n1 as macho boot object and experiment in there?#asahi-dev · 2026-04-17 06:22
yukadear fruit overlord, please let me use the computer I bought from you#asahi-dev · 2026-04-17 06:21
yukaOr for 👉️👈️ at 🍏 for letting us use them#asahi-dev · 2026-04-17 06:20
yukaOh for example for detecting easily whether saving and restoring registers is necessary#asahi-dev · 2026-04-16 18:23
yuka*used#asahi-dev · 2026-04-16 18:23
yukaAnd if it could be uses to find out the meaning of the impl regs#asahi-dev · 2026-04-16 18:23
yukaI wonder if kexts loaded on mac os after csmutil disable have more access to impl regs on M4#asahi-dev · 2026-04-15 08:15
yukaeven though they are both from the boot core#asahi-dev · 2026-04-15 08:15
yukaI also noticed my regs_boot_core and regs_full dumps have some different values in the s3_6_c15_c{12,13,14}_* range#asahi-dev · 2026-04-15 07:32
yukaBut if the chickens are initialized by iboot on the boot core but not on the other cores that would in general make sense that the other cores behave weirdly#asahi-dev · 2026-04-15 07:31
yukaI'm not sure if I didn't find all wfis but I'm definitely still getting some register corruption unless I compile with pretty much all drivers and features disabled#asahi-dev · 2026-04-15 07:25
yukaSo lots of s3_0_c11_c*_* which differ between core 6 (boot core) and 7, but are the same between 7 and 8. but all are readonly.#asahi-dev · 2026-04-15 07:22
yukahttp://yuyu.yuka.dev/public/regs_core0#asahi-dev · 2026-04-15 07:22
yukaAnd also the regs of an e-core:#asahi-dev · 2026-04-15 07:19
yukahttp://yuyu.yuka.dev/public/regs_full2#asahi-dev · 2026-04-15 07:19
yukaalso did the readonly check on a secondary core to be sure#asahi-dev · 2026-04-15 07:18
yukathe others are just the values on different cores#asahi-dev · 2026-04-15 07:18
yukaregs_full is with the readonly / el2-only checks