#asahi-dev · 2022-01-04 13:27
mazI guess the bit of glue logic that 'implements' PCI is slightly less than perfect...#asahi-dev · 2021-12-22 10:00
mazkettenis: agreed on both fronts. the pwren-gpios in the sifive binding looks strikingly similar, and Linux should definitely wiggle this pin if present in the DT. we need to work out timings (I doubt power-on is instantaneous).#asahi-dev · 2021-12-22 08:49
mazkettenis: ah. no power, no link, no love.#asahi-dev · 2021-12-22 08:45
maz(not going to be online much in the coming days, but trying to keep an eye on this)#asahi-dev · 2021-12-22 08:44
mazkettenis: marcan: what service does the SMC gpio provide to the RC?#asahi-dev · 2021-12-21 15:47
mazmarcan: gave the SMC trap change a go, works as expected. thanks!#asahi-dev · 2021-12-21 10:50
mazwell, that's what I do already. just need to rebuild the test m1n1/u-boot bundle... once I have sorted the NV branch...#asahi-dev · 2021-12-21 10:47
mazmarcan: now you are forcing me to update the box... ;-)#asahi-dev · 2021-12-21 10:44
mazto taste*#asahi-dev · 2021-12-21 10:44
mazand maybe leave that as an option for people who really want the taste the madness!#asahi-dev · 2021-12-21 10:44
mazI'd rather we turn that off by default, because the behaviour really isn't compliant with the architecture.#asahi-dev · 2021-12-21 10:42
mazmarcan: I think they use it in hvf to emulate SMC, but that's just gross.#asahi-dev · 2021-12-21 09:43
mazj`ey: FWIW, SMC correctly UNDEFs at EL2 (there is nowhere else to report it anyway)#asahi-dev · 2021-12-21 09:18
mazwe have no idea what the guest wants to do anyway.#asahi-dev · 2021-12-21 09:18
mazj`ey: yup, that's the only sensible thing to do.#asahi-dev · 2021-12-21 09:14
mazj`ey: that's an SMC in a KVM guest (EC=0b111111).#asahi-dev · 2021-12-21 09:13
mazj`ey: kvm [13218]: Unknown exception class: esr: 0xfe000020 -- UNRECOGNIZED EC#asahi-dev · 2021-12-21 09:08
mazthey are above that.#asahi-dev · 2021-12-21 09:06
mazi had some good fun tracing that one in KVM.#asahi-dev · 2021-12-21 09:06
mazah, and of course it generates it to EL2, instead of staying at EL1.#asahi-dev · 2021-12-21 09:05
mazfrom EL1, it generates a exception with an EC that is reserved...#asahi-dev · 2021-12-21 09:04
mazit doesn't, unfortunately.#asahi-dev · 2021-12-15 14:41
mazmarcan: I came to that conclusion after a number of failed attempts at making TB work. eventually, I abandoned TB altogether, but git-send-email stayed...#asahi-dev · 2021-12-15 13:41
mazmarcan: the CPU PMU will not make it into 5.17, the branch is already closed.#asahi-dev · 2021-12-14 15:44
mazmarcan: this is going to hurt with devices with multiple queues where the driver really expects a given queue to be associated with a given CPU and doesn't expect the interrupt anywhere else.#asahi-dev · 2021-12-14 15:40
mazmarcan: AICv2: is there really no way to control the affinity at all?#asahi-dev · 2021-12-13 13:44
mazsuggests making it a configuration option (I want the opening chord sequence to Zoot Allures)#asahi-dev · 2021-12-12 11:13
mazmarcan: yeah, that's on my long list of things to do. let me shake off this hangover first! ;-)#asahi-dev · 2021-12-07 16:20
mazmy laptop has a dedicated 'print screen' key. I use it to locate the mouse pointer...#asahi-dev · 2021-12-07 16:18
mazthe positive thing is that someone found a use for this key!#asahi-dev · 2021-12-07 16:12
mazmarcan: big-little is a terrible hack overall. shame it has spread...#asahi-dev · 2021-12-07 16:09
mazotherwise, you can't tell. they all have the same programming model, and count events similarly numbered. but the semantics of the events are can be radically different (they are on ARM cores, less so on Apple).#asahi-dev · 2021-12-07 16:07
mazmarcan: that's the standard big-little horror. the core PMU code uses the interrupt affinity to find out about the PMUs that are compatible. E/P PMUs are obviously incompatible (they count different things), so you have to segregate them.#asahi-dev · 2021-12-07 13:29
mazso that the PMU interrupts can be discriminated.#asahi-dev · 2021-12-07 13:29
mazmarcan: we need to find a way to expose the E/P affinities in DT one way or another.#asahi-dev · 2021-12-07 12:03
mazworse case, make it explicit in DT.#asahi-dev · 2021-12-02 08:48
mazmarcan: does this require any change in the u-boot/Linux/OpenBSD drivers, or can you keep that in m1n1?#asahi-dev · 2021-12-01 15:10
mazmarcan: ah, sorry about that.#asahi-dev · 2021-12-01 14:50
mazarnd_: thanks.#asahi-dev · 2021-12-01 14:47
maznow, the really cool kids use notmuch and whatever front-end works for them.#asahi-dev · 2021-12-01 14:46
mazGnus is the other possible choice, but it is slow as hell.#asahi-dev · 2021-12-01 14:45
mazEmantor: it's an acquired taste, but if you like spending most of your life in Emacs, it can fit the bill.#asahi-dev · 2021-12-01 14:44
mazj_ey: can't always win! :D#asahi-dev · 2021-12-01 14:42
mazWanderlust.#asahi-dev · 2021-12-01 14:36
maz(duh, that's what kettenis was saying... 4 lines above)#asahi-dev · 2021-12-01 14:35
mazarnd_: did you spot that email about picking one of the DT changes for PCIe? the PCI folks are scared of it. or of you. either way, it'd be good to know how to take that into 5.16.#asahi-dev · 2021-11-24 16:06
mazmarcan: it's the bestest. just burn it.#asahi-dev · 2021-11-24 09:46
mazthe only solution I found was to let the fault happen, and then to reset it. only then you can get a sane state.#asahi-dev · 2021-11-24 09:45
mazno amount of resetting can fix it.