#
✕ nick:maz
#asahi · 2021-03-19 15:06
mazoh, excellent.
#asahi · 2021-03-19 15:04
mazkettenis2: I take it that your u-boot doesn't boot from m1n1?
#asahi · 2021-03-19 15:02
mazgreat. I didn't fancy reinventing the wheel.
#asahi · 2021-03-19 15:02
mazkettenis2: ARM page table format? or their own invention?
#asahi · 2021-03-19 15:00
mazthat's not much... :D
#asahi · 2021-03-19 15:00
mazkettenis2: I guess the M1 dart has nothing to do with the PPC one?
#asahi · 2021-03-19 14:59
mazkettenis2: yeah, that's what I was trying to gauge.
#asahi · 2021-03-19 14:50
maz(now that KVM is mostly out of the way, I need storage and networking...)
#asahi · 2021-03-19 14:50
mazexcellent, thanks a lot.
#asahi · 2021-03-19 14:48
mazthanks a bunch.
#asahi · 2021-03-19 14:48
maz(it is already baked)
#asahi · 2021-03-19 14:48
mazmake*
#asahi · 2021-03-19 14:48
mazsven: any known tool to bake this human readable?
#asahi · 2021-03-19 14:47
mazsven: magic.
#asahi · 2021-03-19 14:42
mazdoes anyone have an ADT dump handy? or a simple way to extract it from m1n1?
#asahi · 2021-03-08 16:42
maznever_released: ICH_HCR_EL2.T{ALL0,ALL1,C}.
#asahi · 2021-03-08 16:02
maznever_released: but as long as the traps work, we can deal with that in the kernel, where we have all the required emulation already.
#asahi · 2021-03-08 16:01
maznever_released: as agraf said, it's possible that it is totally broken.
#asahi · 2021-03-08 16:00
maznever_released: you get enough to *virtualise* a GICv3.
#asahi · 2021-03-08 16:00
mazregisters*
#asahi · 2021-03-08 16:00
maznever_released: no. what's available is the ICH_*_EL2 and the ICV__EL1 registerse.
#asahi · 2021-03-08 15:30
mazagraf: as for bugs, I'm planning to taint the kernel with CPU_OUT_OF_SPEC at KVM-init time.
#asahi · 2021-03-08 15:29
mazoverhead compared to running on bare-metal.
#asahi · 2021-03-08 15:29
mazagraf: running hackbench (as IPI-heavy as it gets) in a guest, I get a 6/7% overhead.
#asahi · 2021-03-08 15:27
mazagraf: they of course miss all the linked deactivation, but hey, sub-standard HW... ;-)
#asahi · 2021-03-08 15:26
mazagraf: just enough to have a guest that doesn't suck too much.
#asahi · 2021-03-08 15:25
mazagraf: all the standard GICv3 stuff is there and seems functionnal.
#asahi · 2021-03-08 15:24
mazagraf: KVM directly uses all the vgic registers without any trapping.
#asahi · 2021-03-08 15:05
mazyup.
#asahi · 2021-03-08 15:05
mazkvmtool and crossvm are equially broken.
#asahi · 2021-03-08 15:04
mazreduced*
#asahi · 2021-03-08 15:04
mazmarcan: qemu needs fixing, there is a stupid but with the reduce IPA space.
#asahi · 2021-03-08 15:00
mazgreat for performance... I'll stay with my in-kernel, non-trapping handling, thank you very much.,
#asahi · 2021-03-08 14:59
mazthen I don't understand how this stuff works.
#asahi · 2021-03-08 14:58
mazwhat is rawValue then?
#asahi · 2021-03-08 14:56
mazand that explains why userspace needs to mediate the vtimer as well.
#asahi · 2021-03-08 14:56
mazI think they have the distributor in userspace, while KVM has it in the kernel.
#asahi · 2021-03-08 14:55
maz(grouping in the GIC sense)
#asahi · 2021-03-08 14:55
mazIRQ?FIQ is just the grouping here.
#asahi · 2021-03-08 14:54
mazhv_vcpu_set_pending_interrupt clearly injects a GIC interrupt, because there is an interrupt value provided to it.
#asahi · 2021-03-08 14:54
mazI think you are confusing two different things: injecting a IRQ/FIQ directly, and injecting a GIC interrupt.
#asahi · 2021-03-08 14:53
mazthey don't have any exit reason for a trapped sysreg.
#asahi · 2021-03-08 14:52
mazwhy would they build a vgic the first place if they aren't using it...
#asahi · 2021-03-08 14:51
mazthat'd seem... silly.
#asahi · 2021-03-08 14:51
mazare they actually doing that?
#asahi · 2021-03-08 14:51
mazplus the distributor and RD MMIO as well.
#asahi · 2021-03-08 14:51
mazmarcan: to do the GIC in userspace, you'd need to pass the trapping of all ICV_* regs to userspace.
#asahi · 2021-03-08 14:50
mazmarcan: it has to do *some* vgic in the kernel, as the LRs cannot be populated from EL0. from what I see, their interface is exactly the same as KVM, except for the timer (and I don't really understand why).
#asahi · 2021-03-08 13:14
mazarnd: I doubt it. performance would be absolutely terrible, similar to what we end-up with running KVM on a RPi3...
#asahi · 2021-03-08 13:13
mazarnd: from what I understand, HVF gives you a prety high-level interface to the vcpu, and can do a lot of stuff in the hypervisor layer.

plus ›