#
✕ nick:nickchan
#asahi-dev · 2024-09-13 04:16
nickchansupposedly
#asahi-dev · 2024-09-13 04:16
nickchanby a12 it supposed already isn't like this
#asahi-dev · 2024-09-13 04:16
nickchannicolas17: on a more serious note a11 seems like a giant hack to get all the cores active at the same time
#asahi-dev · 2024-09-13 04:02
nickchanat this point this needs a seperate apple_t8015_cpufreq driver
#asahi-dev · 2024-09-13 04:01
nickchanhttps://paste.debian.net/1329232/
#asahi-dev · 2024-09-13 04:01
nickchani don't have any words for it just see the results
#asahi-dev · 2024-09-13 03:45
nickchanto the point where I am considering *not* use apple_soc_cpufreq
#asahi-dev · 2024-09-13 03:45
nickchanokay a11 cpufreq is even more cursed than i thought
#asahi-dev · 2024-09-13 03:06
nickchanand the driver does not get frequency from hw
#asahi-dev · 2024-09-13 03:06
nickchanbut the comments are not very clear on how exactly to get the frequency
#asahi-dev · 2024-09-13 03:05
nickchanokay APPLE_DVFS_PLL_STATUS and APPLE_DVFS_PLL_FACTOR seems to be related to getting frequency from hw
#asahi-dev · 2024-09-13 02:31
nickchanwhat the upstream kernel cannot currently handle is something like the arm32 situation on a10
#asahi-dev · 2024-09-13 02:31
nickchanand the kernel only expose a safe value
#asahi-dev · 2024-09-13 02:30
nickchanpeople mix and match cpus all the time in arm socs
#asahi-dev · 2024-09-13 02:30
nickchannicolas17: the kernel should be able to handle that
#asahi-dev · 2024-09-13 02:30
nickchanaarch32 linux binaries
#asahi-dev · 2024-09-13 02:29
nickchanso it's not a big loss
#asahi-dev · 2024-09-13 02:29
nickchanI mean it's okay at the page size that a10 supports most aarch32 binaries won't run anyways
#asahi-dev · 2024-09-13 02:28
nickchanDon't think so
#asahi-dev · 2024-09-13 02:28
nickchanonly hurricane does arm32
#asahi-dev · 2024-09-13 02:28
nickchanarm32
#asahi-dev · 2024-09-13 02:28
nickchanthey in fact are not
#asahi-dev · 2024-09-13 02:27
nickchannicolas17: a10 also has that
#asahi-dev · 2024-09-13 02:20
nickchana11 is probably similarly cursed at the physical level but can't be abstrated to allow all cores to be active at the same time
#asahi-dev · 2024-09-13 02:19
nickchannicolas17: a10 is cursed on the physicl level but it is mostly abstracted away from software
#asahi-dev · 2024-09-13 02:09
nickchanI expect G10P AGX DRM support to be even more cursed but currently I am not ready to go into that rabbit hole yet
#asahi-dev · 2024-09-13 02:06
nickchanthere's two other ways that i think a11 can work, one more cursed one less cursed, time to go experiment
#asahi-dev · 2024-09-13 01:47
nickchanand have a11 continue to get the t8103 compatible
#asahi-dev · 2024-09-13 01:46
nickchanwell, unless I apply the a11 logic whenever the new property in cluster cpufreq is present
#asahi-dev · 2024-09-13 01:45
nickchanthis also means that a11 *cannot* get apple,cluster-cpufreq compatible
#asahi-dev · 2024-09-13 01:43
nickchanpossibly missing some tunables if some pcore tunables are missing issues would not show up till the cpu frequency is attempted to be adjusted due to the fusion arch
#asahi-dev · 2024-09-13 01:43
nickchanhowever there are some issues that locks the whole system up
#asahi-dev · 2024-09-13 01:42
nickchanI know the pstates are different from t8010 and the performance controller looks like it is at the same address
#asahi-dev · 2024-09-13 01:41
nickchanas for cpufreq on t8011 we have not got it to work yet
#asahi-dev · 2024-09-13 01:40
nickchanprobably gonna add a new property to the cluster cpufreq node specifying the highest ecore pstate
#asahi-dev · 2024-09-13 01:40
nickchanthis may be tricky too because this means the cpu across cluster are related_cpus but not completely related
#asahi-dev · 2024-09-13 01:36
nickchanthere might be some other stuff too when I am 100% sure I will modify the cpufreq driver to account for this
#asahi-dev · 2024-09-13 01:34
nickchanwhen the pcores are not at their lowest frequency, the ecores can only be at their highest frequency
#asahi-dev · 2024-09-13 01:33
nickchanwhen the ecore is not at its highest frequency, the pcores can only be at their lowest frequency
#asahi-dev · 2024-09-13 01:33
nickchanthis means that:
#asahi-dev · 2024-09-13 01:33
nickchanbut all cores can be active at the same time
#asahi-dev · 2024-09-13 01:33
nickchan- thirdly the performance controller is still like a10 where low states represent ecore and high states represent pcore
#asahi-dev · 2024-09-13 01:32
nickchan- secondly the mpidr for cpu4 and cpu5 are 0x10004 and 0x10005 respectively
#asahi-dev · 2024-09-13 01:32
nickchan- first the aic does not support cluster fast ipi
#asahi-dev · 2024-09-13 01:32
nickchana11 acts like as if it has like 1.5 clusters in all sorts of ways
#asahi-dev · 2024-09-12 14:56
nickchanthis will fix the issue https://paste.debian.net/hidden/8c917f99/
#asahi-dev · 2024-09-12 14:56
nickchanjannau: your cpuidle disable on a7-a11 generates a null deref because pdev is not initialized
#asahi-dev · 2024-09-12 13:38
nickchanj`ey: maybe the comments can help but that gets the frequencies from the dt
#asahi-dev · 2024-09-12 13:35
nickchanis it possible to get the cpu frequency from the hw?
#asahi-dev · 2024-09-12 11:39
nickchangoing to include an expanded version of ordering explaination from jannau's series to add m2 device trees

plus ›