#asahi · 2022-03-29 13:16
mazanyway, I'll go back to my cave.#asahi · 2022-03-29 13:15
mazpovik: and that's bad enough. this is Android again, just with another name.#asahi · 2022-03-29 13:14
mazGlanzmann: don't you understand that by making these available to people, you are actually distributing these patches and condoning their use as if they were "upstream quality"?#asahi · 2022-03-29 13:10
mazGlanzmann: the whole point of the Linux review process is to catch things *before* they reach the users. what you are doing is to bypass the review process and hoping that things will be fine. most of the time, they are. except when they are not.#asahi · 2022-03-29 13:09
mazGlanzmann: and I'm glad you did. but there is a *major* difference between you running my no-quite-debugged code vs the 1000+ people having installed Debian on M1 and pulling an update blindly.#asahi · 2022-03-29 13:01
maz-rc3 is usually the sweet spot indeed.#asahi · 2022-03-29 13:00
mazj`ey: yes, Sunday.#asahi · 2022-03-29 12:59
mazthis is true of anything that *accesses memory*. mess with a pointer, corrupt a filesystem.#asahi · 2022-03-29 12:57
mazGlanzmann: while you're at it, why don't you just pick random, unreviewed patches from LKML? #asahi · 2022-03-29 12:57
mazGlanzmann: I'm really glad it works for you, but that's not nearly enough to feed this to unsuspecting users. for all we know, this patch would silently eat your data.#asahi · 2022-03-29 12:54
mazGlanzmann: as you said, works for you. which means nothing at scale.#asahi · 2022-03-24 16:43
mazbut someone would shout at you for not writing them otherwise -- it's a loosing game.#asahi · 2022-03-24 16:43
mazthe fate of most docs...#asahi · 2022-03-23 14:44
mazcertainly, using a variable page size approach would be better. more hardcoding is just kicking the same bug down the road.#asahi · 2022-03-22 09:44
mazI'm not condoning the use of 64k pages on arm64 (I wish we never had the option), but there *are* workloads that benefit (a bit) from them on some implementations.#asahi · 2022-03-22 09:42
mazi.e. waste of memory is not a concern if my workload is about squeezing out every bit of performance.#asahi · 2022-03-22 09:42
mazAdryzzOLEDEdition[m]: with 64kB, I want 512M blocks. it's not about individual page allocation, but efficiency of the TLBs.#asahi · 2022-03-22 09:39
mazAdryzzOLEDEdition[m]: it has nothing to do with the memory capacity, but with the way you use that memory.#asahi · 2022-03-22 09:38
mazjn: I guess support in the PTW, for example.#asahi · 2022-03-22 09:37
mazno, and no.#asahi · 2022-03-21 15:58
mazj`ey: *the* slackware user!#asahi · 2022-03-21 11:44
mazLuEvers[m]: no, because of *stupidity* and *cupidity*. nothing to do with security.#asahi · 2022-03-21 11:43
mazanyway, life is too short to give a fsck about libvirt.#asahi · 2022-03-21 11:42
mazthe only ones that don't are QC devices that boot the kernel at EL1.#asahi · 2022-03-21 11:42
mazmost arm64 systems support KVM.#asahi · 2022-03-21 11:38
mazOK, so that's a libvirt issue then. -EWORKSFINE.#asahi · 2022-03-21 11:38
mazenchiladasconpinguino[m]: please try with QEMU only, and dump the output here.#asahi · 2022-03-21 11:29
mazGavinP[m]: that wasn't completely apparent to me, but hey, glad it works for you.#asahi · 2022-03-21 11:26
mazmake sure you use a very recent qemu (i.e. the unreleased 7.0). earlier releases can't work on M1.#asahi · 2022-03-21 11:23
mazj`ey: no, just being able to use the GICv3 CPU interface without trapping and emulating each and every sysreg access...#asahi · 2022-03-21 11:22
mazimagine how well VMs would run if the CPU wasn't buggy! :D#asahi · 2022-03-18 16:12
mazA compose key sorts this out pretty easily. I write in French often enough, and yet have standardised on a UK keyboard for... too long.#asahi · 2022-03-17 12:01
maznot got anything yet, but I'm not scheduled before mid-May anyway...#asahi · 2022-03-11 10:11
mazI remember running axboe's io_uring test, and consistently getting a 20% difference between a 4k guest and a 16k guest, the host system using 16k.#asahi · 2022-03-11 10:05
mazmarcan: it is the sort of speedup you'd see if 4kB resulted in only using a quarter of the cache...#asahi · 2022-03-10 10:58
mazyeah, I was able to boot an initramfs at about that time a year ago.#asahi · 2022-03-09 18:25
mazarnd: says 9 May - 13 May#asahi · 2022-03-09 18:24
mazjust in time for my birthday!#asahi · 2022-03-09 18:23
mazah, nice!#asahi · 2022-03-09 18:22
mazarnd: ah, there *is* a version with a smaller nvme?#asahi · 2022-03-09 18:13
mazarnd: what model did you go for?#asahi · 2022-03-09 16:10
mazmarcan: awesome, thanks.#asahi · 2022-03-09 16:09
mazmarcan: any update on AICv2? I'm hopeful that BHB will buy us an extra week, but I can't promise it either...\#asahi · 2022-03-07 16:31
mazyou're trying to load the kernel directly, which isn't the same thing.#asahi · 2022-03-07 16:30
mazbut obsd has a loader, similar to grub#asahi · 2022-03-07 16:29
mazon my VM, about 14M.#asahi · 2022-03-07 14:31
mazEDK2*#asahi · 2022-03-07 14:31
mazor use the DEI2 shell if you are really perverse! :D#asahi · 2022-03-07 14:26
maznot using EFI on arm64 is a bit backward, tbh.#asahi · 2022-03-07 14:26
mazyup