#asahi-re · 2026-08-23 17:01
sventhe t6040 kext from the 27 beta i think#asahi-re · 2026-08-23 17:01
chaos_princesswhat did you run strings on?#asahi-re · 2026-08-23 17:00
chaos_princessoh, nvm, i can't read#asahi-re · 2026-08-23 17:00
sveni only knew about the one inside acio#asahi-re · 2026-08-23 17:00
svenhuh, never noticed that#asahi-re · 2026-08-23 17:00
chaos_princessno, the m3 coproc has been called "embedded cpu" since m1#asahi-re · 2026-08-23 16:59
sven%s@%s: %s::%s: couldn't set up embedded CPU#asahi-re · 2026-08-23 16:57
svenlolwhat, does atcphy on m4 pro now have a cpu in there?!#asahi-re · 2026-08-22 21:38
djanyone tried if smp issues are still there on t6050 on latest betas? there's been some iboot revs. not sure if apple would randomly drop a fix but one can hope#asahi-re · 2026-08-22 21:14
svenso maybe really just one SID for each rx/tx ring then#asahi-re · 2026-08-22 21:12
svenand we've seen weird bypass SIDs since t8103 and don't set them up on linux anyway#asahi-re · 2026-08-22 21:05
jannau12, 13 don't have valid ttbr so they are probably usnused#asahi-re · 2026-08-22 21:03
jannauhas anyone experimented with FLUSH_VA_RANGE on hw_rev 2 t8110 darts?#asahi-re · 2026-08-22 21:03
svenweird, i'm not aware of anything in the NHI that does DMA outside the rx/tx rings. and rtkit on t8103/t6000 uses sram mapped to the SoC bus#asahi-re · 2026-08-22 21:02
jannaunothing mapped there#asahi-re · 2026-08-22 21:01
svenwonder what 12 and 13 are then#asahi-re · 2026-08-22 20:59
chaos_princessas removing 16 made it not able to communicate#asahi-re · 2026-08-22 20:59
chaos_princessi suspect 1-11 and 16-27 are per-ring#asahi-re · 2026-08-22 20:58
jannau1-11 are remapped to 0, 12 and 13 are translated with separate ttbr, 14 and 15 are bypass and 16-27 are remapped to 0#asahi-re · 2026-08-22 20:33
jannauit was t8122#asahi-re · 2026-08-22 20:30
jannaubut that would still have multiple darts in iommus#asahi-re · 2026-08-22 20:30
jannauI haven't seen 27 remapped sids on t8122 or t6030#asahi-re · 2026-08-22 20:27
svenno dt changes required#asahi-re · 2026-08-22 20:26
svenright now it does a foreach_set_bit over the sidmask, but i bet we could make that use remap instead and just program a single sid#asahi-re · 2026-08-22 20:25
svenhttps://github.com/AsahiLinux/linux/blob/e2e1930a9595bffafad92cec2b5504525efb9cd4/drivers/iommu/apple-dart.c#L257 <-- that thing here#asahi-re · 2026-08-22 20:25
svenwe already have a sidmask per dart in some struct in there, the hack i had in mind was to just setup the first SID and enable remap to that one for all the others#asahi-re · 2026-08-22 20:23
jannaubut I don't want to write proper dt-bindings for it and since invalidation is not faster with remap on t6020 the ugly dts is a weak reason for it#asahi-re · 2026-08-22 20:20
jannauI have a hacked-up remap implementation in dart_probe(). just checks if remap exists and then programs it for t60[02]x acio#asahi-re · 2026-08-22 20:02
svenyeah, that would be bad#asahi-re · 2026-08-22 20:01
chaos_princessi was mostly concerned about the extra memory use since i incorrectly assumed it was duplicating the page table#asahi-re · 2026-08-22 20:00
svenand if we wanted to use remap for this because the invalidation is faster we should be able to set that up entirely in apple-dart.c I think#asahi-re · 2026-08-22 19:59
svenif you trace it you should see the same pagetable in all TTBRs#asahi-re · 2026-08-22 19:58
svenone of those :D#asahi-re · 2026-08-22 19:58
svenor was it iommu group?#asahi-re · 2026-08-22 19:58
svenit stuffs all those 25 dart instances into a single iommu domain and each iommu domain has a pagetable#asahi-re · 2026-08-22 19:58
chaos_princessi assumed it was the former#asahi-re · 2026-08-22 19:58
chaos_princessah, in that case yes, we don't care about remap#asahi-re · 2026-08-22 19:57
svenone table programmed 25 times#asahi-re · 2026-08-22 19:57
chaos_princessis it 25 identical copies of the page table, or one table programmed 25 times?#asahi-re · 2026-08-22 19:57
sventhe dt will just look a bit ugly, but oh well#asahi-re · 2026-08-22 19:56
sveni don't think we need remap for this, only difference to macos will be that it will invalidate all 24 streams at once instead of just setting a single stream id#asahi-re · 2026-08-22 19:56
svenah, it's 3 dart instances but there's no limit on the streams within each dart#asahi-re · 2026-08-22 19:55
chaos_princessmacos also appears to smash all of it down to stream 0 and only program that#asahi-re · 2026-08-22 19:55
svenit's gonna program all of them to the same pagetable then#asahi-re · 2026-08-22 19:55
svenguess that fixed at some point then#asahi-re · 2026-08-22 19:54
chaos_princessidk, worked on a m3max#asahi-re · 2026-08-22 19:54
chaos_princessexcept there is 25 stream ids#asahi-re · 2026-08-22 19:54
svenor did we fix that at some point?#asahi-re · 2026-08-22 19:54
svendoes that even work with the current dart driver? doesn't it support at most 3(?) darts?#asahi-re · 2026-08-22 19:54
chaos_princessthat is my suspicion too