#asahi-dev · 2022-03-08 17:09
mazplease base it on -rc3 though. I'll take care of conflicts.#asahi-dev · 2022-03-08 17:08
mazmarcan: ideally, I'd like an ack from robher on the binding. if he can rubber stamp that, I'll take the series in.#asahi-dev · 2022-03-08 14:15
mazmarcan: perf stuff now queued. feel free to also merge my irq/aic-pmu branch in your tree if you need to resolve conflicts.#asahi-dev · 2022-03-07 08:35
mazonce this is upstream, I'll be able to reason about. until then, I can't do much.#asahi-dev · 2022-03-07 08:35
mazalso, there is zero indication of how this kernel was build the first place, with what configuration and what source code...#asahi-dev · 2022-03-07 08:33
mazin this case, it looks like the device wasn't really there.#asahi-dev · 2022-03-07 08:32
mazthey could try and make the timeout longer, but thatt's just papering over the core problem.#asahi-dev · 2022-03-07 08:23
mazj`ey: there is something in the spec for link-up, but this is not necessarily the right value for the HW at hand. once we have the SMC driver up and running independently of the rest of the stack, I'll have a look at just handling hotplug events without ever waiting for anything.#asahi-dev · 2022-03-03 16:20
mazalso, it is pretty hard to emulate a GIC on top of AIC without resorting to virtualisation. the life cycle of the interrupts on AIC really doesn't match the GIC's.#asahi-dev · 2022-03-03 14:49
mazmarcan: code can wait forever. your health can't. just rest.#asahi-dev · 2022-03-01 18:03
mazconfirmed: emacs-28 is OK (building with one page size and running with another, in both directions).#asahi-dev · 2022-03-01 17:27
mazthe not-yet-released version, yes (still building).#asahi-dev · 2022-03-01 17:22
mazsven: yup, looks like the same thing I'm hitting.#asahi-dev · 2022-03-01 17:16
mazdives in the emacs source code...#asahi-dev · 2022-03-01 17:08
mazI already run Chromium like that...#asahi-dev · 2022-03-01 17:08
mazworse case, I can run emacs in a VM, but that's pushing it a bit...#asahi-dev · 2022-03-01 17:01
mazj`ey: it's not an editor. it's my operating system.#asahi-dev · 2022-03-01 17:01
mazsven: I *really* need your 4k page stuff now.#asahi-dev · 2022-03-01 17:00
mazapparently so.#asahi-dev · 2022-03-01 16:45
mazemacs, you are so shit.#asahi-dev · 2022-03-01 16:45
mazmmap(0x7ffeea7c5000, 3246840, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED, 3, 0x70d000) = -1 EINVAL (Invalid argument)#asahi-dev · 2022-02-25 15:15
mazHmmm. haven't managed to convince $boss yet, but once a bigger, meaner desktop machine is available...#asahi-dev · 2022-02-25 15:02
mazkettenis: new toys! :D#asahi-dev · 2022-02-25 15:02
mazamusingly, the extra DSB doesn't hurt that much.#asahi-dev · 2022-02-25 14:57
mazaxboe: if oyu're using a Pro/Max, you already have this, as there is no MMIO version on that HW.#asahi-dev · 2022-02-25 11:46
mazI guess something like hackbench would be more significant.#asahi-dev · 2022-02-25 11:45
mazyup. minor, but still observable.#asahi-dev · 2022-02-25 11:44
mazmarcan: so fast-ipis are shaving another ~1s off my automatic Debian VM install benchmark (now 2m18.637s).#asahi-dev · 2022-02-24 13:22
maz(only 276 emails behind...)#asahi-dev · 2022-02-24 13:21
mazmarcan: cheers.#asahi-dev · 2022-02-24 11:48
mazyup, we can always work something out.#asahi-dev · 2022-02-24 11:46
mazif the PMU goes in, it goes directly via the perf tree. we can ask Will for a stable branch for the prefix that may conflict with your stuff, but that's pretty minor anyway.#asahi-dev · 2022-02-21 13:57
mazactually, S1 beng enabled or not doesn't matter. the combination of the two levels of translations should apply irrespective of that. only FWB is allows to change that.#asahi-dev · 2022-02-21 13:56
mazmarcan: with S1 disabled, the strongest memory attribute wins. if that's not what you are seeing, then that's a violation of the architecture. that's c;early described in the VMSA spec.#asahi-dev · 2022-02-21 13:35
mazmarcan: MMU off at S1 gives you device memory, which is not speculated. But that relies on S2 not using FWB.#asahi-dev · 2022-02-21 08:59
mazmarcan: I miss some of the context, but a guest can always restrict speculation on some memory range by mapping it as device, provided thatt the hypervisor doesn't set HCR_EL2.FWB.#asahi-dev · 2022-02-21 08:47
mazmarcan: the timeout stuff looks good to me.#asahi-dev · 2022-02-17 11:24
mazkettenis: if it does, the latter probably isn't very well tested...#asahi-dev · 2022-02-17 11:11
mazsven: contrary to what I said, the AIC doesn't seem to have a configuration for edge/level, and "knows" which line in which. and the fasteoi flow doesn't distinguish them either.#asahi-dev · 2022-02-16 13:30
mazsven: well, it certainly has a very different behaviour at the interrupt controller level.#asahi-dev · 2022-02-16 13:23
mazwritel has a DMB, so that part is safe.#asahi-dev · 2022-02-16 13:21
mazsven: without other synchronisation, there is indeed no ordering.#asahi-dev · 2022-02-09 18:52
mazmarcan: I'm starting queuing stuff for 5.18, so if you wanted to repost the fast-IPI stuff to reduce the queue, now would be the time! ;-)#asahi-dev · 2022-02-03 18:01
mazalyssa: small scale, very low visibility,and I personally know each and every one of the members. Limited risks.#asahi-dev · 2022-02-03 16:30
mazyeah, I had to severely tweak the MLs I host to be DMARC compliant and avoid the traffic ending up as spam.#asahi-dev · 2022-02-03 08:40
mazmarcan: ;-)#asahi-dev · 2022-02-02 15:02
mazGlanzmann: I stay away from the internal prod stuff. otherwise, I'll get roped in for ridiculous stuff.#asahi-dev · 2022-02-02 14:52
mazone thing for sure, I'm not going to pay my employer to allow me to do my job... ;-)#asahi-dev · 2022-02-02 14:50
mazyahoo has been the worse over the years.