#asahi-dev · 2024-09-05 13:42
nickchanmaz: cleaner than adding a gross a10 midr check in id_aa64pfr0_32bit_el1 in cpufeature.h#asahi-dev · 2024-09-05 13:38
nickchanhttps://lore.kernel.org/linux-arm-kernel/20231127114559.990314-11-maz@kernel.org/#asahi-dev · 2024-09-05 13:37
nickchanmaz: am looking for some cleaner ways to disable 32-bit EL0 for A10, and I think I need to use your MIDR-based override patch#asahi-dev · 2024-09-02 03:21
nickchanjust don't init when it's not set#asahi-dev · 2024-09-02 03:21
nickchanAIDR_EL1<2> is the feature bit for state retention#asahi-dev · 2024-09-02 03:20
nickchanjannau: for cpuidle after checking apple,arm-platform#asahi-dev · 2024-09-01 16:05
nickchanstaged for#asahi-dev · 2024-09-01 16:05
nickchansven: perhaps the aic patches can be staged linux-next#asahi-dev · 2024-09-01 12:58
nickchanfrom adt mmio and interrupts are also different#asahi-dev · 2024-09-01 12:55
nickchanlikely need to do some pinmux just like atv4k#asahi-dev · 2024-09-01 12:54
nickchanpongoOS serial on broken on t2 too, and that probably should be fixed first#asahi-dev · 2024-09-01 12:54
nickchant2 linux or m1n1 confirmed to not work will need to investigate#asahi-dev · 2024-08-31 19:02
nickchanfor tracing on a7-a11 currently im thinking of hijacking a core and use the debug registers to debug the other cores#asahi-dev · 2024-08-31 15:58
nickchanokay it's res0 at least on the iPhone X#asahi-dev · 2024-08-31 15:13
nickchanim actually not super sure need to check#asahi-dev · 2024-08-31 15:10
nickchanremember to clear the bit after testing! otherwise you will get the a7-a11 behavior even on m1#asahi-dev · 2024-08-31 15:09
nickchanon a7-a11 it is res0#asahi-dev · 2024-08-31 15:09
nickchanon the asi mac this can be set#asahi-dev · 2024-08-31 15:09
nickchanthis is the disable wfi retention bits#asahi-dev · 2024-08-31 15:08
nickchanjannau: you can also do a runtime check by checking whether bit 0 of s3_5_c15_c5_0 is RES0#asahi-dev · 2024-08-31 15:07
nickchannow because this is a downstream driver#asahi-dev · 2024-08-31 14:57
nickchanfor difference between s8000 and s8003: midr is different and the chicken bits to be applied is also different#asahi-dev · 2024-08-31 14:55
nickchanapple,s5l8960x || apple,t7000 || apple,t7001 || apple,s8000 || apple,s8001 || apple,s8003 || apple,t8010 || apple,t8011 || apple,t8012 || apple,t8015#asahi-dev · 2024-08-31 14:54
nickchanso full list to exclude:#asahi-dev · 2024-08-31 14:54
nickchanall skus will only been given its own soc-specific compatible and not some compatible of a similar soc as far as soc-based compatibles go#asahi-dev · 2024-08-31 14:53
nickchangiven that the t6001 machines were not given the t6000 compatibles#asahi-dev · 2024-08-31 14:50
nickchanand finally apple,t8012 would be for the T2#asahi-dev · 2024-08-31 14:50
nickchanand then there's TSMC A9 which is apple,s8003#asahi-dev · 2024-08-31 14:50
nickchanso that's apple,t7001, apple,s8001, apple,t8011#asahi-dev · 2024-08-31 14:49
nickchannow the devicetree stuff is very much downstream and things could change but we currently do not make the skus based on the AX-series SoC to have the base A-series SoC compatible#asahi-dev · 2024-08-31 14:48
nickchanat the very least apple,s5l8960x also needed to be excluded#asahi-dev · 2024-08-31 14:48
nickchanjannau: you missed quite a lot of compatibles #asahi-dev · 2024-08-31 11:57
nickchanA10 "Fusion"#asahi-dev · 2024-08-31 11:57
nickchanchadmed: I think those parts are just fused#asahi-dev · 2024-08-31 11:54
nickchanit doesn't seem like it's a good patch but I do not have other sensible solutions#asahi-dev · 2024-08-31 11:53
nickchanhttps://lore.kernel.org/asahi/20240829093636.100840-2-towinchenmi@gmail.com#asahi-dev · 2024-08-31 11:53
nickchanand due to this one of the patches that I have is a gross patch that pretend a10 does not support 32 bit el0#asahi-dev · 2024-08-31 11:51
nickchan(and ability to execute aarch32)#asahi-dev · 2024-08-31 11:51
nickchanit is cores that suddenly changes performance when crossing a certain p-state#asahi-dev · 2024-08-31 11:51
nickchanthere isn't a pcore and ecore to software#asahi-dev · 2024-08-31 11:49
nickchandividing all e-core frequencies by 2.65 seems to do the trick#asahi-dev · 2024-08-31 11:49
nickchanso it has been adjusted so that performance scales roughly linearly#asahi-dev · 2024-08-31 11:49
nickchandue to the core type#asahi-dev · 2024-08-31 11:49
nickchanuserspace absolutely does not expect a lower frequency to have more performance than a higher frequency#asahi-dev · 2024-08-31 11:48
nickchanand the higher p-states are the hurricane cores#asahi-dev · 2024-08-31 11:48
nickchansince the lower p-states are the zephyr core#asahi-dev · 2024-08-31 11:48
nickchanalso has the lower p-states frequencies artifically adjusted#asahi-dev · 2024-08-31 11:47
nickchana10(x) is directly from corellium#asahi-dev · 2024-08-31 11:47
nickchanhaven't have luck locating the performance controller from the adt though#asahi-dev · 2024-08-31 11:47
nickchanthough other SoCs im dealing with like A7 are also like that