#asahi-dev · 2025-09-15 09:37
mazthey could also be choosing to have a PIPT behaviour by halving the cache size to avoid aliasing. that's a pretty common trick.#asahi-dev · 2025-09-15 09:34
maz*macos's problem.#asahi-dev · 2025-09-15 09:34
maznickchan: possible. but that gives *more* guarantees to the guest, which means the hypervisor has to work harder to avoid aliasing. in any case, that's macos'#asahi-dev · 2025-09-15 08:27
mazPretends that the I$ is PIPT, which allows SW to elide CMOs in a number of cases. they must have ways to handle that in the hypervisor somehow to avoid aliasing.#asahi-dev · 2025-09-08 10:48
maznickchan: no, that's not something I can do. Will is the maintainer for this, and I cannot bypass him that way.#asahi-dev · 2025-08-13 11:54
mazI'd be happy to help reviving this effort if at all possible.#asahi-dev · 2025-08-13 11:53
mazkettenis: in my case, I'm dependent on the FW loader in u-boot, as all my stuff lives on external storage connected to the USB-A port.#asahi-dev · 2025-08-12 17:46
mazjannau: hasn't the ASM FW loader been ripped out of u-boot? I haven't updated mine in ages because of that...#asahi-dev · 2025-08-12 10:29
mazjust checked on my box, and all ports are indeed enabled.#asahi-dev · 2025-08-12 10:24
mazhow comes this doesn't show? my M2 pro mini seems fine (but I'm running with a 2023-vintage DT, so I may be missing a lot of things...)#asahi-dev · 2025-08-12 10:23
mazhuh.#asahi-dev · 2025-08-12 10:21
mazbreak?*#asahi-dev · 2025-08-12 10:21
mazdoes anything actually rbeak?#asahi-dev · 2025-07-24 09:16
mazdart and iopgtables would be good as well -- they tend to badly clash on each rebase...#asahi-dev · 2025-07-24 09:11
mazsven: party!#asahi-dev · 2025-07-16 10:07
mazno worries -- I know the feeling.#asahi-dev · 2025-07-16 10:06
mazsven: is SMC on track to be picked up this time around? I don't see it in -next just yet.#asahi-dev · 2025-07-15 14:27
mazgot basic 6.16 running on M2-pro mini with just dart and SMC added. Ugly conflicts in io-pgtables-dart.c, but it has booted -- must be perfect! ;-)#asahi-dev · 2025-07-07 10:36
mazfl0_id: the IPA reported by the HW should tell you where this lands, assuming you have a 1:1 mapping at S2. You could then compare that with the HW memory map and work out what this is about.#asahi-dev · 2025-06-19 08:26
maznickchan: I don't think Will has any issue with that (we speak f2f often enough).#asahi-dev · 2025-06-17 16:56
mazI don't have a lot of bandwidth left for it though ($work is keeping me busy enough).#asahi-dev · 2025-06-17 16:55
maznickchan: there is no need for a particular maintainer entry, but I'm happy to look after the m1/m2 parts of it.#asahi-dev · 2025-06-16 08:20
mazI'm sorry for your loss ;-)#asahi-dev · 2025-06-16 08:19
mazchaos_princess: 6.16-rc1 is your friend. YMMV.#asahi-dev · 2025-06-10 17:22
mazso its not that LPIs are more general, it's mostly that level interrupts as SPIs are a pretty dumb way of injecting virtual events in a VM.#asahi-dev · 2025-06-10 17:21
mazonce you start having virtual devices, it's a lot simpler to have MSIs for most of the devices. they scale, and they are simpler to handle.#asahi-dev · 2025-06-10 17:18
mazand GICv3 is soooo last year anyway... ;-)#asahi-dev · 2025-06-10 17:17
mazsame thing for PPIs, we're quite happy with only a handful.#asahi-dev · 2025-06-10 17:17
mazI never bothered emulating it in KVM because our VMs tends to only have a small number of SPIs, and use LPIs much more liberally.#asahi-dev · 2025-06-10 17:14
mazit should be fine to expose extended ranges, as long as your dist supports it. in general, the cpuif shouldn't really care about the INTID it reports to the guest (they all have the same state machine), so supporting it should be rather simple (and by that, I mean simple for a GIC feature...).#asahi-dev · 2025-06-10 17:10
mazamarioguy: well, the *emulation* you present to the guest must be consistent. Either you fully support extended ranges (both in the cpuif and dist/rdist), or you don't.#asahi-dev · 2025-06-10 17:02
mazamarioguy: which patch set?#asahi-dev · 2025-06-08 08:37
mazyou could implement the emulation in userspace if you wanted, it wouldn't be a massive change to the KVM code.#asahi-dev · 2025-06-08 08:35
maznickchan: oh, KVM supports trapping the IMPDEF stuff alright. It actually sets the trap bit (HCR_EL2.TIDCP), and inject an UNDEF as a result.#asahi-dev · 2025-05-30 09:09
mazoh, and there is Documentation/admin-guide/dynamic-debug-howto.rst (can't quite believe we have actual documentation!)#asahi-dev · 2025-05-30 09:07
maz /sys/kernel/debug/dynamic_debug/control has all the traces, and you can pick the ones you want to see.#asahi-dev · 2025-05-30 09:02
mazIIRC, with CONFIG_DYNAMIC_DEBUG, you can enable debug on a call-site basis.#asahi-dev · 2025-05-29 10:14
mazcheers! :)#asahi-dev · 2025-05-29 10:13
mazchadmed: I like the analogy! mind if I quote you? :)#asahi-dev · 2025-05-29 07:53
mazand since AIC2 is pretty retarded on that front (no direct injection), it's not like we're losing much.#asahi-dev · 2025-05-29 07:52
mazyou "just" leave some performance on the table if you can't control it.#asahi-dev · 2025-05-29 07:52
mazKVM doesn't care about the routing of the interrupts as long as they are purely virtual#asahi-dev · 2025-05-29 07:52
mazKVm doesn'#asahi-dev · 2025-05-29 07:51
mazamarioguy: I'm not quite dead yet! :)#asahi-dev · 2025-05-11 18:33
mazfeels like a very late (or even too late) change, given how many systems have been deployed with this particular name.#asahi-dev · 2025-05-01 12:18
maznickchan: funny that. we had the same bug in KVM, fixed in 2015. Qualcomm had it in their SMMU emulation code, never fixed. this architecture makes a point in making things difficult.#asahi-dev · 2025-05-01 10:52
maznickchan: let me guess: it used SP instead...#asahi-dev · 2025-04-13 09:42
mazalyssa: the PCIe series is 6.16 material, I expect. I'll ping Bjorn around -rc4 to make sure it hasn't got off his radar.#asahi-dev · 2025-04-04 09:08
mazrobher: any clue?#asahi-dev · 2025-04-04 09:08
mazI guess this should be part of describing the slots past the switch, but how to express that numbering is a mystery. wonder if there's prior art around that.