#asahi-dev · 2021-11-03 10:42
mazmarcan: +1 for a tree. it would even make sense today, where we only use a handful of interrupts, and yet allocate a fixed-size domain.#asahi-dev · 2021-11-03 10:36
mazmarcan: looking good! if you want to split the fast IPI from the t6k bring up, that could trickle into upstream quickly (5.17).#asahi-dev · 2021-11-03 09:36
maz(stupid big-little strikes again)#asahi-dev · 2021-11-03 09:36
mazI end-up having do know that for the PMU...#asahi-dev · 2021-11-03 09:35
mazmarcan: do max/pro have a similar cluster layout as m1, where MPIDR_EL1.Aff2 indicates whether the CPU is E or P?#asahi-dev · 2021-11-02 17:24
mazyeah, I don't really uses laptops for anything other than a glorified remote desktop. so +1 for a boosted mini or something that can live in the rack...#asahi-dev · 2021-11-02 17:14
mazsven: do you have a dump of the max/pro id registers somewhere?#asahi-dev · 2021-11-02 16:23
mazmarcan: my experience with GICv1/2 where we shoved the PPI affinity in one of the cells makes me want to use an extra cell.#asahi-dev · 2021-11-02 16:17
maz(transparent on the GIC, not on AIC)#asahi-dev · 2021-11-02 16:17
mazOK, so that's not that different from the multi-chip GIC implementations, except that it is transparent to SW.#asahi-dev · 2021-11-02 16:16
mazor is that still possible, and the die is just an extra routing hop?#asahi-dev · 2021-11-02 16:15
mazmarcan: that's odd. it means that an interrupt is strictly routed to a particular die? I would have expected SW to be able to route the interrupt to any CPUs...#asahi-dev · 2021-11-02 16:00
mazsounds like a plan. do people actually work on enabling Linux on pre-M1?#asahi-dev · 2021-11-02 15:58
mazI though as much. I guess the affinity management would be different, but we can also probably migrate M1 over to sysreg-based IPIs, right?#asahi-dev · 2021-11-02 15:57
mazmarcan: if we can keep a single driver, I'm all for it. if you feel that the programming model has diverged too much, we write a new one. no big deal.#asahi-dev · 2021-11-02 15:46
mazare they still presented as FIQ on the receiving end?#asahi-dev · 2021-11-02 15:44
mazmarcan: eh, if we need them, we implement them!#asahi-dev · 2021-11-02 14:36
mazI guess clearing the overflow is done by clearing the interrupt...#asahi-dev · 2021-11-02 14:35
mazas it turns out, SYS_APL_PMSR_EL1 cannot be written to (it UNDEFs).#asahi-dev · 2021-11-02 11:46
mazhuh...#asahi-dev · 2021-11-02 11:45
mazmarcan: now have a skeletal PMU driver. has anyone documented the PMU events, apart from the two fixed counters? I've sort of worked out that event 2 is most likely cycles too, but that's about it.#asahi-dev · 2021-11-01 19:41
mazI expect they moved away from the strict 50% split between ram and IO.#asahi-dev · 2021-11-01 19:40
mazthat's a lot more than I anticipated. probably means that they expect to have systems with a lot more than the current 64GB.#asahi-dev · 2021-11-01 19:38
mazmarcan: what's the PA range on this new toy?#asahi-dev · 2021-10-29 13:15
mazI'm all for silly things, today...#asahi-dev · 2021-10-29 13:15
mazmarcan: kernel_neon_begin()/end(). the crypto code is full of it! :-)#asahi-dev · 2021-10-27 15:21
mazjust like emacs is the real operating system.#asahi-dev · 2021-10-27 15:21
mazjn: it's not really an interrupt controller. it is the CPU, the OS, and more. what you call the CPU is just a toy on the side.. :D#asahi-dev · 2021-10-27 15:20
mazj_ey: as the VMM? yes.#asahi-dev · 2021-10-27 15:14
maz(800 pages, will keep you warm for a while...)#asahi-dev · 2021-10-27 15:14
mazyou can't be nice to the GICv3 spec. you just want to burn it.#asahi-dev · 2021-10-27 15:13
mazyou should have seen the engineering spec!#asahi-dev · 2021-10-27 15:13
mazbaroque doesn't quite describe it. more like a sick, twisted fsck.#asahi-dev · 2021-10-27 15:05
mazkettenis: does openbsd have support for GICv3 ITS? it says "virtio33[0]: couldn't establish msix interrupt", and falls back to INTx instead...#asahi-dev · 2021-10-27 14:47
mazthis is PCIe, so hotplug is more or less baked in the root port. rfkill should be fundamentally independent.#asahi-dev · 2021-10-27 14:35
mazalyssa: good stuff. I guess that in order of priority, I have booting from u-boot, wifi, and PMU. fun... ;-)#asahi-dev · 2021-10-27 14:33
mazkettenis: ACPI didn't quite work, but switching to a DT instead gave me a working guest (on the M1, btw..)#asahi-dev · 2021-10-27 14:32
mazalyssa: right. I'll have a look next week if I have a bit of time ($work is ultra busy atm).#asahi-dev · 2021-10-27 14:31
mazkettenis: I'm installing openbsd in a guest for the first time. surprisingly painless so far!#asahi-dev · 2021-10-27 14:30
mazI was thinking of injecting a PCIe hotplug event, but I need to try and see how that work. which means I need to look into the wifi ugliness...#asahi-dev · 2021-10-27 14:29
mazalyssa: you need to go and tell the core to perform the rescan. effectively, we need to treat that as a hotplug.#asahi-dev · 2021-10-27 14:05
mazthere was at least a report yesterday of PCIe needing some additional love and care to issue a PERST#.#asahi-dev · 2021-10-27 14:04
mazkettenis: yup, that's one of the reasons why I want to have a look.#asahi-dev · 2021-10-27 13:57
maztrying to move over to u-boot...#asahi-dev · 2021-10-27 13:56
mazdebian creates a 'debian' directory, for example.#asahi-dev · 2021-10-26 15:06
mazpovik: I suspect we'll need something similar when booting from u-boot anyway. #asahi-dev · 2021-10-26 14:45
mazpovik: seems to match this: https://community.intel.com/cipcp26785/attachments/cipcp26785/programmable-devices/56074/1/cpci_power_on_timing.pdf#asahi-dev · 2021-10-26 14:32
mazpovik: interested in that pcie patch. is that to make sure the link is brought down before we reinit everything else?#asahi-dev · 2021-10-22 08:23
mazyeah, and they have ugly hooks in things like GPU drivers.