#asahi · 2023-09-14 14:55
agrafi509vcb: watch the video above :)#asahi · 2023-09-14 14:54
agrafsven: https://www.youtube.com/watch?v=oZqFYJVOUQo :)#asahi · 2023-09-14 14:48
agrafIt's also hvf only. With TCG, it doesn't work out of the box. IIRC Apple has some implicit PAC initialization. I don't remember if there was more :) #asahi · 2023-09-14 14:47
agrafsven: It's macOS-on-macOS only right now. GPU virtualization is a binary only framework#asahi · 2023-09-14 14:47
agrafsven: There's a video about that :)#asahi · 2023-03-09 16:34
agrafsven: I'd love to have a modern day Mac-on-Linux :D#asahi · 2023-03-09 16:33
agrafsven: bummer :)#asahi · 2023-03-09 14:57
agrafsven: Do you think anyone would be up to RE the pv gpu protocol? Once things calmed down a bit at work I can polish my QEMU branch that can run full vmapple2 VMs with HVF :)#asahi · 2022-09-10 16:23
agrafOther than that SMCCC call and the fact that you need to expose a vPMU, Windows should work out of the box on anything that runs KVM#asahi · 2022-09-10 16:22
agrafmarcan: re Windows in KVM: Windows always emits an SMC based SMCCC call on boot. If that doesn't return -1 in x0, it fails to boot. So to have Windows work in KVM, we need to have SMC handling in KVM. For that we need the weird apple specific bit set to make SMC calls trap into EL2 rather than EL1 and then handle the Apple specific exit to deal with the call#asahi · 2022-06-23 18:52
agrafElvishJerricco[m]: All the hv_* functions are provided by Apple#asahi · 2022-06-23 18:52
agrafElvishJerricco[m]: QEMU code that calls into hvf: https://github.com/qemu/qemu/blob/master/target/arm/hvf/hvf.c#asahi · 2022-06-23 18:51
agrafElvishJerricco[m]: docs: https://developer.apple.com/documentation/hypervisor#asahi · 2022-06-23 18:51
agrafElvishJerricco[m]: Not sure what you mean. You mean a link to the docs?#asahi · 2022-06-23 18:50
agrafElvishJerricco[m]: Not sure I follow? QEMU uses the official Hvf API the way it's intended :)#asahi · 2022-03-11 14:03
agrafPoint taken :)#asahi · 2022-03-11 14:01
agrafEspecially since 16kb doesn't give you any functional benefit. It's not like you get more PA bits like with 64kb#asahi · 2022-03-11 14:01
agrafarnd: and then you would do alternatives for pte removal/modification and tlb invalidations. I agree that it's probably going to become very messy very quickly and is probably not worth it#asahi · 2022-03-11 14:00
agrafarnd: So my thinking was that the Linux internal page tables would still be the 16kb format basically#asahi · 2022-03-11 13:59
agrafarnd: I wasn't thinking of actually making anything a runtime thing. You would basically create alternatives for the fault handler and PTE modification code. I guess a lot of that is super deeply engrained on arm64 though :/#asahi · 2022-03-11 13:58
agrafj`ey: Hah :)#asahi · 2022-03-11 13:57
agrafarnd: so that you could build a kernel that has PAGE_SIZE=16kb, but still runs with 4kb granule configs#asahi · 2022-03-11 13:56
agrafarnd: How hard would it be to have 4kb granule fallback code in Linux if PAGE_SIZE is 16kb?#asahi · 2022-03-11 13:56
agrafarnd: wrt <arnd> the flip side is that a lot of hardware just doesn't support 16KB#asahi · 2022-01-31 10:24
agrafsure#asahi · 2022-01-31 10:24
agrafpeac_: what are you trying to do? :)#asahi · 2022-01-31 10:19
agrafpeac_: your average distros should "just work". What problems did you run into when booting?#asahi · 2022-01-31 10:19
agrafpeac_: when running in QEMU, you really don't need any of the Asahi enablement#asahi · 2021-10-26 15:06
agrafIt's one of those bits in the spec where based on what functionality your CPU has, the hypervisor's ability to trap it changes#asahi · 2021-10-26 15:05
agrafjmr2: Windows is running an SMC call on boot. The QEMU arm maintainer considers the default virt machine to have "no EL3 implemented" which means the SMC instruction is supposed to trigger a #UDEF according to the spec#asahi · 2021-10-26 15:04
agrafjmr2: (well, you can, with a trick: Use -M virt,virtualization=on. That will switch internal QEMU logic into treating SMC calls as PSCI calls)#asahi · 2021-10-26 15:03
agrafjmr2: yeah, it will be in the next release. Also you can't boot Windows :)#asahi · 2021-10-26 10:46
agrafmaz: thanks, my hvf patch to QEMU does the same now :)#asahi · 2021-10-25 19:08
agrafmaz: I'm mostly trying to understand where in the abstraction levels this would usually be cut off. Theoretically, a DATA exit to QEMU is on a pretty similar level to an AXI message on a real core.#asahi · 2021-10-25 18:55
agraf(which is good news, it means in this context it's just a nop)#asahi · 2021-10-25 18:54
agrafmaz: Sounds like it's the core / cache's responsibility to deal with this instruction then rather than the device. I don't remember any AXI message that would even remote know anything about cache ops.#asahi · 2021-10-25 18:34
agrafOr is it supposed to just be a NOP?#asahi · 2021-10-25 18:33
agrafmaz: what is it supposed to do on device memory? Or in other words: What AXI message does it translate to?#asahi · 2021-10-25 16:23
agrafkettenis: you do OpenBSD things, right? https://github.com/utmapp/UTM/issues/3197#asahi · 2021-05-31 21:32
agrafThat said, I haven't tried myself :)#asahi · 2021-05-31 21:32
agrafarnd: but normal macOS kernelcaches should just work#asahi · 2021-05-31 21:31
agrafarnd: because it wouldn't pass any signature verification :)#asahi · 2021-05-31 21:31
agrafarnd: So m1n1 on the external drive I can absolutely see how that would fail#asahi · 2021-05-31 21:28
agrafarnd: like look at the LocalPolicy files before/after, validate signatures, see if maybe the OIK ownership was changed/revoked#asahi · 2021-05-31 21:28
agrafarnd: Did you manage to dig down into why it's breaking?#asahi · 2021-05-31 21:26
agrafarnd: that sounds .... very surprising :)#asahi · 2021-05-31 21:24
agrafj_ey: they do, the normal one is for macOS, the iSC one is for the firmware#asahi · 2021-05-31 21:24
agrafj_ey: sorry, it should be /System/Volumes/iSCPreboot IIRC#asahi · 2021-05-31 21:23
agrafarnd: those were 2 separate installations on the same external drive, but different partitions?#asahi · 2021-05-31 21:22
agrafarnd: but there's a good chance the "boot selector" thing doesn't set the hash that allows you to run 3rd party binaries