#
✕ nick:maz
#asahi · 2021-05-26 08:14
mazmarcan: I don't see how that's a DoS though.
#asahi · 2021-05-26 08:13
mazmarcan: as far as a KVM guest is concerned, IMDEP encoding are never implemented, so at least we're consistent.
#asahi · 2021-05-26 08:11
mazmarcan: we just inject an UNDEF into the guest.
#asahi · 2021-05-14 11:41
mazpipcet[m]: among other things. but it implicitly assumes that memory isn't badly fragmented (which isn't usually true)
#asahi · 2021-05-14 11:36
mazsven: a poor-man stage-2.
#asahi · 2021-05-14 11:36
mazsven: unless dart and DAPF can be controlled by separate entities (kernel vs hypervsor)
#asahi · 2021-05-12 14:16
mazI'm using a vintage AX88772, which works well enough... r8152 is awfully modern compared to it.
#asahi · 2021-05-12 14:16
mazhuh...
#asahi · 2021-05-12 14:15
mazhow slow?
#asahi · 2021-05-12 13:57
mazkettenis: I'll try and trace things a bit next week (that's what holidays are for...)
#asahi · 2021-05-12 13:36
mazmy bet is on a screaming interrupt.
#asahi · 2021-05-12 13:36
mazand I don't see this with other systems that use u-boot.
#asahi · 2021-05-12 13:35
mazkettenis: the slowdown happens if running an EFI app under u-boot.
#asahi · 2021-05-12 13:35
mazkettenis: I don't think this is related to the USB stack. it is definitely good enough to load stuff from the network pretty quickly.
#asahi · 2021-05-12 12:25
mazkettenis: I parked that for the moment, specially given that I'm out of cycles...
#asahi · 2021-05-12 12:25
mazkettenis: I've seen the same issues as _andy_t_. booting the kernel as an EFI application from the kernel is fine. booting an intermediate bootloader (such as ... err... grub) results is something horribly slow and unreliable.
#asahi · 2021-05-07 17:06
mazsven: in your SPRR blog, you refer to HCR_EL2.THE. Did you mean TGE instead?
#asahi · 2021-05-05 11:55
mazthat's the patch hacked on to get it a bootloader on kevin.
#asahi · 2021-05-05 11:55
mazhttps://build.opensuse.org/package/view_file/SUSE:SLE-12:GA/grub2/grub2-s390x-02-kexec-module-added-to-emu.patch?expand=1
#asahi · 2021-05-05 11:52
mazhttps://share.confex.com/share/122/webprogram/Handout/Session14802/SHARE-Anaheim-2014-zBoot-14802-20140304.pdf
#asahi · 2021-05-05 10:49
mazoddly enough, I really like Forth. I'll have my head checked...
#asahi · 2021-05-05 08:21
mazthere is also a set of patches that allow grub to run in *cough* userspace and use kexec as a backend. pretty neat for distro kernels.
#asahi · 2021-04-29 13:53
mazas I mentioned earlier, CPU isolation does exactly that to avoid any disruption to userspace performing a timing-critical task. if you're careful enough, you don't even get IPIs. of course, that comes at the expense of moving all ancillary tasks to worker CPUs which have to deal with everything the kernel normally does in the background.
#asahi · 2021-04-29 13:50
mazpipcet[m]: that's fine as a special case, and you can do that from userspace by moving all the IRQs to a single core. But in general, you want to be able to place interrupts where they matter most, specially if you have to deal with devices doing back-to-back interrupts (wild guess, a router).
#asahi · 2021-04-29 13:28
mazI guess you could use CPU isolation which comes very close to that.
#asahi · 2021-04-29 13:28
mazerr... no.
#asahi · 2021-04-29 13:26
mazopticron: I'm sure it is extensively documented in the spec.
#asahi · 2021-04-29 13:25
mazyup. different bit in PSTATE.
#asahi · 2021-04-29 13:17
mazthat'd mean the interrupt controller can peek into PSTATE for each CPU. doesn't scale to big systems, but certainly a good optimisation on a small system.
#asahi · 2021-04-29 12:22
mazit's called secure computing.
#asahi · 2021-04-29 12:09
mazproblem solved! :D
#asahi · 2021-04-29 12:06
mazthings may be different on M1, but a change would need to be backed by hard numbers.
#asahi · 2021-04-29 12:05
mazthat's at least what the x86 and GICv2 experience has taught us over the years.
#asahi · 2021-04-29 12:04
mazin general, dispatching on multiple CPUs has an adverse effect on the interrupt service latency.
#asahi · 2021-04-29 08:06
mazpipcet[m]: the default affinity is to CPU0. you can play with /proc/irq/*/smp_affinity* to set it to any random CPU.
#asahi · 2021-04-12 14:18
mazdisclaimer: I don't give a rat's arse about x86 emulation.
#asahi · 2021-04-12 14:16
mazmarcan: what ARM comes up with or not is only one side of the problem. adding non-architectural stuff in a userspace visible way is not something anyone will consider lightly. If you want to run TSO, flip the bit from the bootloader and be done with it.
#asahi · 2021-04-12 13:53
mazmarcan: you'd have to abstract it behind some "strongly ordered" API. overall , this doesn't fit the architecture, nor its direction of travel.
#asahi · 2021-04-12 13:51
mazmarcan: I'm pretty sure you will get a huge push-back on exposing TSO to userspace in any shape or form. it's like SVE, only far worse because there is no architecture to write against.
#asahi · 2021-04-07 15:09
mazarnd: yup, that's the ugly link in the chain.
#asahi · 2021-04-07 15:00
mazarnd: I don't really see what that buys us. it's not like we're merging a fully working system. At the current state of things, it only gives you a serial console...
#asahi · 2021-04-07 14:57
mazmarcan: as you pointed out, that's not really a blocker. these patches can land independently, and nothing breaks compilation wise.
#asahi · 2021-04-07 14:46
mazneither Will nor Catalin have even Acked the second patch, which is the critical one.
#asahi · 2021-04-07 14:46
mazno, they must go via the arm64 tree.
#asahi · 2021-04-07 08:59
mazmarcan: always happy to help.
#asahi · 2021-04-07 08:56
mazthe CPUs can't do non-VHE, which is the way Linux boots now.
#asahi · 2021-04-07 08:51
mazI'm trying to get Will to get the VHE crap a once over.
#asahi · 2021-04-07 08:50
mazmarcan: I expect arnd to drag it.
#asahi · 2021-04-07 08:49
mazmarcan: actually, the FIQ stuff is in the arm64 tree.
#asahi · 2021-04-07 08:49
mazmarcan: nah, I'm preparing the branch at the moment.

plus ›