#asahi-dev · 2026-07-23 07:42
nickchanhaving that bit set should use the 24 mhz clock#asahi-dev · 2026-07-23 07:41
nickchanlinux says that bit is responsible for clock selection#asahi-dev · 2026-07-22 18:44
nickchanm1n1 pr#asahi-dev · 2026-07-22 18:44
nickchancreated pr for 4 level paging, which is pratically a bunch of copy pasting#asahi-dev · 2026-07-21 07:04
nickchanthe notion of being able to generate impdef errors with no way to handle them also doesn't sound right#asahi-dev · 2026-07-21 07:03
nickchansome L2C registers to sweep mmio without generating an SError storm too i think#asahi-dev · 2026-07-19 08:39
nickchanwhere they come from#asahi-dev · 2026-07-19 08:38
nickchanwhat are the words in the apple logo though#asahi-dev · 2026-07-19 08:27
nickchanch more likely for changes to be pulled. since it is still a fork what upstream do should not matter, but not everyone knows that.#asahi-dev · 2026-07-19 08:27
nickchan based on my understanding of linux kernel development trees like soc/apple is not materially different from <random linux fork>, but being an established tree to pull changes from it is mu#asahi-dev · 2026-07-19 08:22
nickchana statement of "still no slop regardless of what upstream does" (but more formal) is fine too#asahi-dev · 2026-07-19 08:21
nickchancation especially in regards to soc/apple.#asahi-dev · 2026-07-19 08:21
nickchanI do feel like the slop policy is due for a revision, not because of stance changed but rather because upstream linux has plenty of ai content now, so the policy could use some clarifi#asahi-dev · 2026-07-18 10:07
nickchanw#asahi-dev · 2026-07-18 08:47
nickchanthe cache properties in dt is only used for sysfs so that makes zero sense whatsoever#asahi-dev · 2026-07-18 06:48
nickchanwhatever the same crashes yuka has in m4 even with idle=nop#asahi-dev · 2026-07-18 06:47
nickchanmaybe it's something to do with wfi#asahi-dev · 2026-07-18 06:40
nickchanand the memory reg doesn't matter because m1n1 overwrites it#asahi-dev · 2026-07-18 06:39
nickchanprobably some bad assumption in m1n1#asahi-dev · 2026-07-18 06:38
nickchanand yes it should be disabled in the dtsi because it doesn't work before filled by loader#asahi-dev · 2026-07-18 06:37
nickchanyuka: m1n1 enables the framebuffer in dt_set_fb()#asahi-dev · 2026-07-18 06:36
nickchanm1n1 enables the framebuffer#asahi-dev · 2026-07-18 06:29
nickchanyeah m1n1 really wants /reserved-memory#asahi-dev · 2026-07-18 06:29
nickchanother things look either OK or pointless#asahi-dev · 2026-07-18 06:28
nickchankeep it as is filled by loader#asahi-dev · 2026-07-18 06:28
nickchanrhlvista[m]: you do want the reserved-memory node#asahi-dev · 2026-07-18 06:25
nickchanyou could also check the example#asahi-dev · 2026-07-18 06:25
nickchan"Nodes without unit addresses shall be ordered alpha-numerically by the node name."#asahi-dev · 2026-07-18 06:25
nickchanand if a node is not on a bus then it goes to 2#asahi-dev · 2026-07-18 06:24
nickchanthe bus is relvant for the fact that it uses unit addresses for its children#asahi-dev · 2026-07-18 06:23
nickchan"Nodes on any bus, thus using unit addresses"#asahi-dev · 2026-07-18 06:19
nickchanerm, recommend upstream#asahi-dev · 2026-07-18 06:19
nickchanso at least i would not recommend that until there's more information on sotra_d#asahi-dev · 2026-07-18 06:18
nickchanand the most pedantic way of doing the t6050 dtsi is probably have 36 cpus but that doesn't matter until sorta_d or whatever comes out#asahi-dev · 2026-07-18 06:16
nickchanfor pcluster and ecluster respectively#asahi-dev · 2026-07-18 06:16
nickchanhw.perflevel1.l1icachesize, hw.perflevel1.l1dcachesize, hw.perflevel1.l2cachesize, hw.perflevel0.l1icachesize, hw.perflevel0.l1dcachesize, hw.perflevel0.l2cachesize#asahi-dev · 2026-07-18 06:15
nickchani think people query it from macos sysctl#asahi-dev · 2026-07-18 06:14
nickchanyou could just not include them now#asahi-dev · 2026-07-18 06:12
nickchanhttps://docs.kernel.org/devicetree/bindings/dts-coding-style.html#asahi-dev · 2026-07-18 06:11
nickchanat least timer should be after soc#asahi-dev · 2026-07-18 06:11
nickchana nit that device tree does violate the usual ordering guidelines#asahi-dev · 2026-07-18 06:10
nickchani don't think you need affinities unless unless one plans on using perf#asahi-dev · 2026-07-18 05:35
nickchanyou could hardcode it in the kconfig or use m1n1's linux.py#asahi-dev · 2026-07-18 05:25
nickchanwith how sloppy apple has been recently wouldn't be surprised#asahi-dev · 2026-07-18 05:25
nickchanon m5 #asahi-dev · 2026-07-18 05:25
nickchanthough... i wonder if some *other* random register is unlocked instead#asahi-dev · 2026-07-18 05:24
nickchanremove access to that msr which would work as long as kvm is not enabled#asahi-dev · 2026-07-18 05:22
nickchanappeared over a year ago inside some ane firmware#asahi-dev · 2026-07-18 05:21
nickchanthis page is up to date afaik#asahi-dev · 2026-07-18 05:21
nickchanhttps://asahilinux.org/docs/hw/soc/soc-codenames/