#asahi-dev · 2024-10-24 14:53
nickchanI think it's like 230 MB in size at least for M4 iPad#asahi-dev · 2024-10-24 14:53
nickchanexclave userland is called exclaveOS, has a rootfs dmg, dyld, libsystem all that#asahi-dev · 2024-10-24 14:52
nickchancL4 kernel, runs in GL1, called exclaveCore#asahi-dev · 2024-10-24 14:52
nickchanchaos_princess: Exclave is its own little OS#asahi-dev · 2024-10-24 14:46
nickchanI expect m4 macs to get it so we should see soon#asahi-dev · 2024-10-24 14:46
nickchancan do nested on m2 at least but m1, uh idk#asahi-dev · 2024-10-24 14:45
nickchanbecause with exclaves, sptm and txm macOS by default would have stuff in EL2, EL0, GL2, GL1(!) and GL0#asahi-dev · 2024-10-24 14:44
nickchanespecially on m1#asahi-dev · 2024-10-24 14:44
nickchansven: Don't think the exclaves stuff is deployed on macOS yet but after that happens running macOS in hypervisor is going to be... interesting to say at least#asahi-dev · 2024-10-24 14:37
nickchanwithout demoting#asahi-dev · 2024-10-24 14:37
nickchanbecause ktrw on a11 *has* to work somehow#asahi-dev · 2024-10-24 14:37
nickchansven: that must be a recent addition#asahi-dev · 2024-10-24 14:29
nickchanwhen not demoted I think coresight also doesn't work correctly#asahi-dev · 2024-10-24 14:26
nickchanhowever, I think it has some compatibility problems and does not work for latest macOS#asahi-dev · 2024-10-24 14:26
nickchanthe workaround for apple silicon mac would be to run to run macOS in hypervisor on m1n1#asahi-dev · 2024-10-24 14:25
nickchandavid: it's a software restriction where some demotion registers (needed to enable debugging) disabled by boot ROM #asahi-dev · 2024-10-15 15:58
nickchansomething like 2400 MHz is a lot more stable though#asahi-dev · 2024-10-15 14:51
nickchanprobably because i didn't mess with the other fields#asahi-dev · 2024-10-15 14:51
nickchanthough i must say that state is very unstable#asahi-dev · 2024-10-15 14:42
nickchanand see if those states are the same now#asahi-dev · 2024-10-15 14:41
nickchana safe test would be to say, overwrite the definition of pstate 3 with defintion of pstate 4#asahi-dev · 2024-10-15 14:41
nickchannot sure if they are locked down on m1#asahi-dev · 2024-10-15 14:36
nickchanthe cpu ratio is multiplied with the 24 MHz clock to get the final frequency#asahi-dev · 2024-10-15 14:35
nickchanhaven't figured other fields and psinfo2 yet but presumable has to do with core and sram voltage#asahi-dev · 2024-10-15 14:35
nickchanon a10 the cpu ratio is (((psinfo1 & 0xffffffff) >> 0xe) & 7) / 5 + ((psinfo1 >> 4) & 0xff) / ((psinfo1 & 0xf) * 2 + 2)#asahi-dev · 2024-10-15 14:34
nickchanso yeah that's an overclock primitive#asahi-dev · 2024-10-15 14:33
nickchanin the example above I am modifying the definition of pstate 1#asahi-dev · 2024-10-15 14:33
nickchanso like what frequency and voltage pstate N is#asahi-dev · 2024-10-15 14:32
nickchanwhich allows arbitrary changes in meaning of the pstates#asahi-dev · 2024-10-15 14:31
nickchanthat register (at a different offset) on a10 seems to be writable#asahi-dev · 2024-10-15 14:31
nickchanthe register that corellium called CLUSTER_PSINFO1_GET() and CLUSTER_PSINFO2_GET() in linux-m1#asahi-dev · 2024-10-15 14:30
nickchanusing modified proxyclient/experiments/cpu_pstates.py#asahi-dev · 2024-10-15 14:29
nickchanmeasured cpu frequency reached 2508 MHz#asahi-dev · 2024-10-15 14:27
nickchanresult:#asahi-dev · 2024-10-15 14:27
nickchan>>> write64(0x202f00000 + 0x20020, 1 << 25 | 1 | 1 << 12)#asahi-dev · 2024-10-15 14:26
nickchan>>> write64(0x202f00000 + 0x80000 + 0x20 * 1, 0xae00000404a20dc0)#asahi-dev · 2024-10-15 14:26
nickchan>>> write64(0x202f00008 + 0x80000 + 0x20 * 1, 0x50000f8c0a000606)#asahi-dev · 2024-10-15 14:26
nickchan>>> write64(0x202f00000 + 0x20020, 1 << 25 | 2 | 2 << 12)#asahi-dev · 2024-10-15 14:26
nickchanon a10:#asahi-dev · 2024-10-14 14:18
nickchanelse it won't work (so linux needs to be modified)#asahi-dev · 2024-10-14 14:18
nickchancommand is 64-bit while status is 32-bit#asahi-dev · 2024-10-14 14:18
nickchancommand = 0x202220020, status = 0x202220050#asahi-dev · 2024-10-14 14:17
nickchanfound a9 cpufreq#asahi-dev · 2024-10-14 12:57
nickchanCan\'t find a PCore frequency which is lower than the Ecore frequency we are trying to achieve\n#asahi-dev · 2024-10-14 12:57
nickchan*strings*#asahi-dev · 2024-10-14 12:57
nickchana11 cpufreq confirmed to be cursed#asahi-dev · 2024-10-12 12:00
nickchanpmgr adt on older socs seems to have a slightly different struct i will have to fix that#asahi-dev · 2024-10-10 14:02
nickchanonly needs to be 8 big though, aic1 will never be bigger than that#asahi-dev · 2024-10-10 13:37
nickchani will just add a logical map then#asahi-dev · 2024-10-10 13:34
nickchanin aic 1