#asahi · 2021-03-04 14:36
mazAIDR_EL1: "Provides IMPLEMENTATION DEFINED identification information."#asahi · 2021-03-04 14:28
mazI wonder if this would also trap stuff that is outside of the IMPDEF sysreg range...#asahi · 2021-03-04 14:14
mazmarcan: that would make some sense. the whole point of the "Auxiliary" stuff is to provide IMPDEF functionality. do these bits override HCR_EL2.TPIDR? or replace it?#asahi · 2021-03-04 10:14
mazkettenis: no, that's totally u-arch specific. the architecture only talks about what is permitted to be cached in a TLB.#asahi · 2021-03-04 10:12
mazit is interesting how they are trying to limit the size of their TLBs.#asahi · 2021-03-04 10:10
mazit's just that all other implementations are sane...#asahi · 2021-03-04 10:09
mazyou're right, it was never forbidden.#asahi · 2021-03-04 10:08
mazLinux handles rollover, but we also reserve half of the space to deal with KPTI.#asahi · 2021-03-04 10:06
mazit's mandatory in ARMv8...#asahi · 2021-03-04 10:05
mazfsck me.#asahi · 2021-03-04 10:03
mazkettenis: please don't tell me they are stuck on 8bit...#asahi · 2021-03-04 10:03
mazkettenis: don't they have 16bit ASIDS?#asahi · 2021-03-04 09:03
mazmarcan: we're all getting it.#asahi · 2021-03-03 18:26
mazI'm vaguely partial to (1), but that's only me. I bet that performance wise, they are strictly equivalent.#asahi · 2021-03-03 16:53
mazhe's great fun to watch.#asahi · 2021-03-03 16:52
mazyup. now imagine trying to get both HW and SW people to agree on the meaning of these things, with different interpretations of the English text. Fun. Not.#asahi · 2021-03-03 16:49
mazmarcan: the architecture is shifting to a "rule system" to describe all that, but I'm not sure it is any better.#asahi · 2021-03-03 16:47
mazmarcan: definitely. it is a maddening part of the architecture... someone even said that it is an adversarial architecture, and I can't really say otherwise.#asahi · 2021-03-03 16:45
mazmarcan: SGTM!#asahi · 2021-03-03 16:18
maz(and it gives me more time to sort out the KVM changes!)#asahi · 2021-03-03 16:17
mazj`ey: I'm planning to take the first two patches of that series for -rc2 (because they are obvious fixes), but the rest needs to simmer in -next for a bit.#asahi · 2021-03-03 16:15
mazj`ey: no much will go in 5.12, only small cleanups. the core FIQ stuff definitely is 5.13 material.#asahi · 2021-03-03 14:07
mazmarcan: yup, interesting litmus test.#asahi · 2021-03-03 12:13
mazmarcan: I'm pretty sure herd7 does support LSE atomics.#asahi · 2021-03-03 12:12
mazmarcan: but hey, bugs everywhere at the moment! :D#asahi · 2021-03-03 12:11
mazmarcan: I'd be surprised, given that atomic.txt, the LSE atomics and the ARMv8 memory model have the same author...#asahi · 2021-03-01 14:24
mazmarcan: also, SYS_APL_VM_TMR_MASK behaves like an enable, and no a mask (bit set to allow the interrupt). it'd be good if that was reflected in the name of the sysreg.#asahi · 2021-03-01 14:23
mazmarcan: just realised another couple of nits: it'd be good if your APL-specific definitions were suffixed with the minimum EL at which they are accessible, just like any other AArch64 sysreg.#asahi · 2021-03-01 11:00
mazif there is any valid level at all (not necessary leaves), you could end-up with intermediate walk entries being present, and that's no good.#asahi · 2021-03-01 10:55
mazwhat does your TTBR0 contain at the point where you invalidate on ASID reallocation?#asahi · 2021-03-01 10:50
mazkettenis: i.e. the PTs have been removed and there is no way the core can parse these page tables again?#asahi · 2021-03-01 10:49
mazkettenis: I guess you have a strong guarantee that once you marked this ASID as "invalid", you never make any use of this address space until it is reallocated?#asahi · 2021-03-01 10:44
mazkettenis: are you using ASIDs? or just global TLBs like FreeBSD?#asahi · 2021-03-01 10:28
mazcool.#asahi · 2021-03-01 10:28
mazmarcan: either way is fine by me, I'll leave it up to you.#asahi · 2021-03-01 09:10
mazso you could have all the CPUs in one IS domain, the interrupt controller and the IOMMU in another IS domain, and you'd still have to order things at the OS level... :-/#asahi · 2021-03-01 09:09
mazinterestingly, there is no way to discover whether two blocks are in the *same* IS shareability domain.#asahi · 2021-03-01 08:58
mazmarcan: btw, I think the patch that moves all the ICC_* stuff into sysreg.h breaks 32bit builds (the GIC driver ls shared between the two architectures). Consider only moving the ICH_* definition instead.#asahi · 2021-03-01 08:57
mazmarcan: ah, fair point about writel.#asahi · 2021-02-24 11:46
mazanyway. one is dead, and the other won't be available for another 2 years, I expect.#asahi · 2021-02-24 11:45
mazthough none of the two were actually v8.4 due to the lack of SEL2.#asahi · 2021-02-24 11:45
mazsame as Zeus.#asahi · 2021-02-24 11:45
mazno, it had the full v8.4-NV.#asahi · 2021-02-24 11:43
mazto my knowledge, only 2 CPUs implement NV: Neoverse V1, and TX3. and the latter is dead.#asahi · 2021-02-24 11:39
maznever_released: the _EL123 only gives you some save/restore facility. being out of context, exceptions doen't apply to them.#asahi · 2021-02-24 11:39
mazyou will need some sane state at EL2 though.#asahi · 2021-02-24 11:38
mazEL1 will magically appear.#asahi · 2021-02-24 11:38
mazmarcan: just clear TGE, which should be enough.#asahi · 2021-02-24 11:28
mazOK, that's somewhat reasuring.#asahi · 2021-02-24 10:57
mazbut we definitely want that on for KVM.