#
✕ nick:maz
#asahi · 2022-08-09 16:37
mazUSB-A doesn't work yet.
#asahi · 2022-08-07 15:46
mazthere is also a huge number of comments from Alex that I absolutely need to address.
#asahi · 2022-08-07 15:45
mazand hopefully, a M2-mini will appear shortly after, which should speed up the development (watching paint dry in the model is only so much fun...)
#asahi · 2022-08-07 15:43
mazjannau: no, I don't at the moment. I'm hoping to find a bit of time for that after September, as I'm pretty busy until then...
#asahi · 2022-08-07 14:50
maznah, seems to happen long before the sigsetjmp(). interestingly, this is a per-thread variable, so relative to tpidr_el0. maybe worth checking if this gets clobbered one way or another.
#asahi · 2022-08-07 14:23
mazit'd be interesting to disassemble this function and work out where the 'current_cpu = cpu;' assignment actually occurs.
#asahi · 2022-08-07 14:19
mazwhat if you taskset it on *one* CPU?
#asahi · 2022-08-07 14:14
mazit seems to be a consequence of a longjmp
#asahi · 2022-08-05 11:39
mazexactly. power control is the PDU's job! :D
#asahi · 2022-08-05 11:38
mazwho does?
#asahi · 2022-08-05 11:38
maz+1. my mini runs upstream only.
#asahi · 2022-08-05 11:37
mazyou can't easily reset the affinity of processes that already exist.
#asahi · 2022-08-05 11:37
mazprobie: only if you taskset from PID1, before it has a chance to fork.
#asahi · 2022-08-05 11:31
mazthe affinity is usually inherited by children,.
#asahi · 2022-08-04 07:55
mazj`ey: yeah, it is absolutely pointless, because this is not a synchronous signal. it means that you absolutely need to inspect the vgic state on every exit. so we simply enable the interrupt at boot time (because the GIC otherwise gates it), and forget about it. on M1, we do nothing, which is the bestest.
#asahi · 2022-08-04 07:44
mazhowever, the signal very much exists, and is extensively used to manage the vgic.
#asahi · 2022-08-04 07:43
mazj`ey: yeah, this only indicates the lack of a *Linux* maintenance IRQ, because the traditional way to represent it is to expose it from the GIC itself. here, the signal comes directly from the CPU, unfiltered by AIC, and we simply don't represent it in the DT. Hence, no Linux view of that signal.
#asahi · 2022-08-04 07:26
mazas the M1 implements SEIS, it will deliver an SError. But instead of delivering a *virtual* SError, it delivers a physical one. please see the Linux commit message for the gory details.
#asahi · 2022-08-04 07:24
mazand the M1 bug only shows if you violate the GICv3 state machine, such as EOI a non active interrupt.
#asahi · 2022-08-04 07:23
mazall we need is for the FIQ to fire and get us out of the guest. the source of the interrupt will be disabled long before there is anything to handle.
#asahi · 2022-08-04 07:23
mazamarioguy: maintenance interrupts have always been supported, even on M1, and I've always seen them firing. They are simply not represented as a Linux IRQ, but we don't care about that for KVM.
#asahi · 2022-08-02 14:34
maznice...
#asahi · 2022-08-02 14:33
mazit is quite funny to see them adding their MIDRs to the errata lists each time there is a workaround...
#asahi · 2022-08-02 14:32
mazand of course the MIDR, which changes everything!
#asahi · 2022-08-02 14:31
mazusually, larger caches...
#asahi · 2022-08-02 14:31
maztheir Krio cores are nothing but rebadged Cortex.
#asahi · 2022-08-02 14:30
mazdon't they? I don't think they are shipping any internal design at the moment.
#asahi · 2022-08-02 14:30
maznot really, I meant "cores designed by ARM".
#asahi · 2022-08-02 14:28
mazat this point I'll be all over it.
#asahi · 2022-08-02 14:27
mazno M2 yet, waiting for a mini to land.
#asahi · 2022-08-02 14:27
maz"Make arm ARM again".
#asahi · 2022-08-02 14:26
mazI'm quite curious to see what they will do with the Nuvia design though.
#asahi · 2022-08-02 14:26
mazthat, and a couple of other things! :D
#asahi · 2022-08-02 14:25
mazif you restrict it to "ARM Ltd" based laptops, they may actually be right. but otherwise, yeah, right...
#asahi · 2022-08-02 11:29
mazone always need one more machine anyway.
#asahi · 2022-08-01 10:30
mazivar: just tried (no -L), and no problem.
#asahi · 2022-07-28 07:18
mazI wonder how many other libs have handcrafted prologue/epilogue code...
#asahi · 2022-07-27 13:59
maz(this is starting to sound like project management... ;-)
#asahi · 2022-07-27 13:29
mazplus I'm still running some old 5.18-rc.
#asahi · 2022-07-27 13:28
mazjanrinze: not really. I cherry-picked what I needed, and hacked the rest myself.
#asahi · 2022-07-27 13:02
mazjanrinze: in one of sven's branches on the Asahi repo.
#asahi · 2022-07-27 13:01
mazI guess that this makes the FW loader for the PCIe USB slightly more urgent that we'd like it.
#asahi · 2022-07-27 13:00
mazfair enough.
#asahi · 2022-07-27 12:57
mazI've been using sven's patches on the Ultra (together with a bunch of u-boot and DT changes) for about 2 months now, and USB3 on the USBC connectors is doing fine. I wish these patches made it upstream.
#asahi · 2022-07-27 09:32
maz(they should call them red and blue, really...)
#asahi · 2022-07-27 09:31
maz8 fast cores, 2 "slow" cores.
#asahi · 2022-07-27 08:27
mazfunnily, someone is shouting about the same thing on #qemu. I guess some package must have been updated for macos...
#asahi · 2022-07-27 08:23
mazmps: highmem=off was a workaround to deal with some internal qemu nonsense. I fixed it for both KVM and hvf. shout if it doesn't work with 7.0, and ping agraf in the process.
#asahi · 2022-07-27 08:19
mazthen get rid of it.
#asahi · 2022-07-27 08:16
mazdo you have highmem=off on the command line?

plus ›