#
✕ nick:maz
#asahi · 2022-06-21 12:43
mazfor example, an SMC will result in an EL1 exception, and won't be seen by KVM at all.
#asahi · 2022-06-21 12:42
mazj`ey: this is only valid of TCG, I think. I can't see how this would work for KVM.
#asahi · 2022-06-21 12:26
mazsame thing here. it seems they are trying to use features that don't exist on the apple HW, such as SMC or the architected PMU.
#asahi · 2022-06-21 12:25
mazno 64k pages though, the HW doesn't support them.
#asahi · 2022-06-21 12:24
mazno
#asahi · 2022-06-20 15:31
mazthis is called job security.
#asahi · 2022-06-20 15:30
mazno need. there is already an x86 emulator written in js.
#asahi · 2022-06-17 14:21
mazso much for me waiting for older M1s to drop in price! ;-)
#asahi · 2022-06-13 12:38
mazkaprests: the discussion hasn't really started yet.
#asahi · 2022-06-10 17:39
mazmps: I think you have the wrong concept. speculation is about hypothesis. risk is about potential danger.
#asahi · 2022-06-10 17:37
mazs/you/your/
#asahi · 2022-06-10 17:37
mazmps: again, it is only a security risk if you environment is risk adverse.
#asahi · 2022-06-10 17:30
mazdefine risk. what risky in a given context may not be tyhat much of a problem in another. and after almost 5 years of spectre security theatre, it really shows that the whole industry doesn't care.
#asahi · 2022-06-10 17:19
mazspeculation introduces leakage, news at 11.
#asahi · 2022-06-06 13:47
mazRy_Darcy: no. oyu need to specify a *device*, most likely a cdrom over virtio-scsi.
#asahi · 2022-06-06 13:43
mazRy_Darcy: the PXE loop is because you didn't specify a block device that EFI understands for your CD image. it thus fallback to PXE.
#asahi · 2022-06-06 13:42
mazj`ey: well, funnily enough, it knows about the instruction if you flip a bit in some IMPDEF register. but it then convey the trap with an UNKNOWN EC.
#asahi · 2022-06-06 13:40
mazj`ey: last time I checked, win10 really didn't like the M1 at all. I suspect they try to use SMC...
#asahi · 2022-06-06 07:37
mazamarioguy: that's about 60% of the whole KVM code.
#asahi · 2022-06-06 07:37
mazamarioguy: the vgic relies on a full SW emulation of the distributor, the redistributors, and to feed the virtual interrupts into a very thin pipe that is (badly) implemented in HW.
#asahi · 2022-05-30 08:08
maz(patch will be in 5.19).
#asahi · 2022-05-30 08:07
mazaep: you need a small patch to KVM to avoid a HW bug if running on anything but the original M1 (the patch enables the same workaround for max/pro).
#asahi · 2022-05-15 11:49
mazcmalvi[m]: you can resize partitions, but resizing the FS has to happen from within Linux.
#asahi · 2022-05-15 11:47
mazs/in/is/
#asahi · 2022-05-15 11:47
mazkettenis: and good luck finding a definition of the memory model for C and Rust interactions. what seems to work on x86 in very unlikely to work on arm64...
#asahi · 2022-05-12 12:17
mazwith LTO, all bets are off.
#asahi · 2022-05-12 12:17
mazas long as you don't use LTO, linking requires very little RAM. barely more than the size of the object you are trying to link.
#asahi · 2022-05-05 15:08
mazit's only a hint though. there is no guarantee that the HW actually supports it (but since it supports 4k as a base granule, odds are they also support 2M).
#asahi · 2022-05-04 17:29
mazyeah, I solely rely on the watchdog for that.
#asahi · 2022-05-04 17:27
mazI think somehow u-boot advertises RTS, but obviously fails to implement it (and I still run an ancient version of u-boot).
#asahi · 2022-05-04 17:27
mazI have similar problems with rebooting. if I don't disable runtime services from the kernel command line, it goes mad.
#asahi · 2022-05-03 16:32
maznp
#asahi · 2022-05-03 16:31
mazthere is *NO*
#asahi · 2022-05-03 16:31
mazPhyx: this is a big-little system. there is "right" PMU...
#asahi · 2022-05-03 16:19
mazPhyx: unless you have a *very* old perf, you don't need to recompile it. you just need to specify which PMUs you want to count the events on.
#asahi · 2022-04-19 10:30
mazar: timers are just the same as other implementations. it is interrupts and memory attributes that have been difficult to shove into the code.
#asahi · 2022-04-18 13:04
mazsteps away from the computer...
#asahi · 2022-04-18 13:03
mazI fixed QEMU, kvmtool and CrosVM. someone else can pick that sucker.
#asahi · 2022-04-18 13:02
mazprobably because GEM5 is buggy and creates a VM with the default IPA size, which the M1 cannot support by default./
#asahi · 2022-04-18 13:00
maz'dmesg| grep kvm' should say something if KVM is compiled in.
#asahi · 2022-04-18 13:00
mazKVM is not a module.
#asahi · 2022-04-18 09:24
maztg3 also seem to be 32bit only.
#asahi · 2022-04-18 08:53
mazdevice mappings don't allow non-aligned accesses. as for non-32bit accesses, it isn't clear to me whether that's a specific device problem, or something more general. Does USB or TG3 use any sub-32bit access?
#asahi · 2022-04-16 09:40
mazst3g4n0[m]: you will need a brand new QEMU (7.0), as it is other wise broken on M1 (an older version may be OK if you have a Pro or Max).
#asahi · 2022-04-16 09:35
mazno. this isn't different from any other arm64 KVM host.
#asahi · 2022-04-16 09:35
mazno idea, I don't use libvirt.
#asahi · 2022-04-16 09:34
mazst3g4n0[m]: about a year ago.
#asahi · 2022-03-29 16:29
mazdottedmag: Sun Fire v490.
#asahi · 2022-03-29 16:28
mazjannau: nah... some of my machines are over 50kg!
#asahi · 2022-03-29 16:26
mazj`ey: and you didn't run away with the box?

plus ›