#asahi-dev · 2022-11-22 09:25
mazthe thing I really dislike about this is that you are creating a Linux-specific API at the hypervisor level. if the hypervisor needs to interpret this, that's a pretty large attack surface. mujltiply this by the number of CPU architectures out there...#asahi-dev · 2022-11-22 09:20
mazwell, that's the portability vs performance trade-off.#asahi-dev · 2022-11-22 09:19
mazthis is starting to look a lot like virtio-gpu#asahi-dev · 2022-11-22 09:18
mazhmmm. there is HW out there that doesn't require a trap on submission (the HW implements VFs to can be directly controlled by the guest).#asahi-dev · 2022-11-22 09:17
mazas long as you don't trap too much, there is no reason why a VM would be too slow. device assignment solves that. you just get extra SW complexity to manage a stage-2 on the IOMMU side.#asahi-dev · 2022-11-22 09:13
mazall the 4k-only crap in a guest, problem solved.that's where I run emacs these days.#asahi-dev · 2022-11-21 09:12
mazmba-m2 up and running (more or less)#asahi-dev · 2022-11-12 11:15
mazand I'll be positively annoyed if they announce the M2 mini this side of Xmas! ;-)#asahi-dev · 2022-11-12 11:14
mazjannau: fancy indeed. but indeed I don't intend to face the rack (it is behind me when I'm home, and rather far away when at work...)#asahi-dev · 2022-11-12 10:58
maznow the question is how to rack the damn computer. any advise?#asahi-dev · 2022-11-12 10:52
mazjannau: mba-M2 on order (tricked $boss into the "lab equipment" argument). I guess I'll revive these NV patches after all...#asahi-dev · 2022-11-02 14:57
mazprobably. the ports are definitely only doing standard USB3 at the moment.#asahi-dev · 2022-11-02 14:55
mazgiven that I *have*#asahi-dev · 2022-11-02 14:55
mazChaosPrincess: given that I no TB support whatsoever in the kernel, I doubt that the USB hub I see has *anything* to do with tunelling.#asahi-dev · 2022-11-02 14:53
mazright. it was worth checking anyway.#asahi-dev · 2022-11-02 14:50
mazany interest in that?#asahi-dev · 2022-11-02 14:50
mazsven: I have a TB3 NVMe device that has a USB hub inside (the kernel says so when I plug it).#asahi-dev · 2022-11-02 14:47
mazsven: how do you identify such a beast?#asahi-dev · 2022-11-02 14:33
mazj`ey: I though we'd have a discussion on the list first? did I miss anything?#asahi-dev · 2022-10-27 15:10
mazhuh, that's tomorrow's job for me... guess I'll have a fine weekend...#asahi-dev · 2022-10-24 16:50
mazI'm not on asahi@ (too many MLs...)#asahi-dev · 2022-10-24 16:47
mazmarcan: yup, count me in.#asahi-dev · 2022-10-19 14:48
mazor maybe they just default to enabled... odd.#asahi-dev · 2022-10-19 14:44
mazmarcan: does m1n1 always enable the guest timers before entering the payload?#asahi-dev · 2022-10-18 11:29
mazmarcan: ah, yes, that's how it's actually called. yeah, I expect it to be a two-way thing. I bet this stuff only exists in Android otherwise...#asahi-dev · 2022-10-18 11:25
mazis btcohab part of this?#asahi-dev · 2022-10-17 08:36
mazdevmem is, in general, a recipe for disaster...#asahi-dev · 2022-10-11 11:25
maz(I sort of miss those!)#asahi-dev · 2022-10-11 11:25
mazI'll raise you with a PREP box!#asahi-dev · 2022-10-11 11:25
mazyeah, CHRP strikes back! :D#asahi-dev · 2022-09-29 17:08
mazeverything that's on the Internet is true, by definition.#asahi-dev · 2022-09-29 15:25
mazthat's awesome. I'll take delivery of the TB patches next week! :D#asahi-dev · 2022-09-29 15:01
mazwow. congrats! :D#asahi-dev · 2022-09-29 15:01
mazHabemus DP?#asahi-dev · 2022-09-26 13:43
mazyeah, I'm probably badly tainted here...#asahi-dev · 2022-09-26 13:42
mazdamn. I'm losing the plot.#asahi-dev · 2022-09-26 13:41
mazmarcan: yeah, I got confused with TLBI by VA, which uses the ASID present in the TTBR.#asahi-dev · 2022-09-26 13:34
maztakes the asid as a parameter (in the same field as TTBR)#asahi-dev · 2022-09-26 13:33
mazthere is TLBI ASIDE1#asahi-dev · 2022-09-26 13:33
mazah, maybe not#asahi-dev · 2022-09-26 13:32
mazme checks again#asahi-dev · 2022-09-26 13:31
mazmy earlier divagation was based on the hope that userspace and GPU could use the same context, making it easy to perform the invalidation.#asahi-dev · 2022-09-26 13:30
mazempty*#asahi-dev · 2022-09-26 13:29
mazinvalidation by ASID relies on the ASID being present in TTBR. if you want to use a GPU-private ASID, you need to switch that TTBR to have the ASID and an enpty set of PTs.#asahi-dev · 2022-09-26 13:23
mazjust invalidating the ASID on allocation is hopefully enough.#asahi-dev · 2022-09-26 13:22
mazIdeally, we wouldn't have to track GPU-specific ASIDs in the arm64 code at all.#asahi-dev · 2022-09-26 13:20
mazsomething like that indeed. #asahi-dev · 2022-09-26 13:13
mazif that's not the case, then we may need something else.