#
✕ nick:marcan
#asahi-dev · 2023-06-01 16:23
marcanthis is why I want to move away from ALARM...
#asahi-dev · 2023-06-01 16:20
marcancherry-picked the XFS fix into our tree (probably easier than a rebase on stable, especially since I think I saw most of the WiFi series get AUTOSEL'd and I really don't feel like wrangling that rebase right now)
#asahi-dev · 2023-06-01 16:12
marcansomething big enough to be immediate cause for concern would make the news and I'd cherry pick it in of course
#asahi-dev · 2023-06-01 16:11
marcanpeople who need near 0-day security response are probably building their own kernels and have a team dedicated to that anyway :)
#asahi-dev · 2023-06-01 16:10
marcan(I don't expect anyone to be using Asahi unmodified and not separately maintained to, like, run a cloud hosting platform or something where security is absolutely paramount)
#asahi-dev · 2023-06-01 16:09
marcanwe reasonably closely track kernel releases so we get the security fixes in a reasonable timeframe by default
#asahi-dev · 2023-06-01 16:08
marcanyou do not want to be stuck on some ancient kernel without security updates
#asahi-dev · 2023-06-01 16:08
marcansid-linux: security fixes on stable kernels are more relevant to *long-term stable* which we do not maintain
#asahi-dev · 2023-06-01 16:07
marcan(I regularly get patch merged for stable notifications for stuff I wrote that is absolutely not a bugfix and useless to stable kernel users and I didn't tag for stable and I have *no* idea why)
#asahi-dev · 2023-06-01 16:07
marcan(though TBH, in my opinion, the stable kernel folks these days are *way* too eager to merge random crap)
#asahi-dev · 2023-06-01 16:07
marcanit's pretty easy for anyone downstream to rebase our patchset on stable point releases though, rarely do you get bad conflicts there
#asahi-dev · 2023-06-01 16:05
marcanin rare cases when they do we cherry-pick
#asahi-dev · 2023-06-01 16:05
marcanmost bugs and issues and stable fixes are in random drivers people are unlikely to use on this platform
#asahi-dev · 2023-06-01 16:05
marcanbut then again our use cases are usually a small fraction of the codebase
#asahi-dev · 2023-06-01 16:05
marcanour downstream kernel tree does not claim to be stable
#asahi-dev · 2023-06-01 16:05
marcanI mean we release kernels based on RCs sometimes
#asahi-dev · 2023-06-01 16:04
marcanwe haven't ever seen that and I hard reboot my machines all the damn time, so that makes me pretty confident we don't have a major problem, but worth checking
#asahi-dev · 2023-06-01 16:03
marcando a filesystem check, I'm curious whether anything bad happened at the metadata level
#asahi-dev · 2023-06-01 16:03
marcaneric_engestrom: if it's the watchdog timing out then we have a problem and that will not do a clean NVMe shutdown
#asahi-dev · 2023-06-01 16:02
marcanI normally don't care for stable point releases but corruption is kinda bad
#asahi-dev · 2023-06-01 16:01
marcanyeah I guess I should probably rebase on 6.3.4 this once
#asahi-dev · 2023-06-01 10:11
marcan(the power button sends multiple warning messages before hard shutdown, we use the last one to trigger a kernel shutdown)
#asahi-dev · 2023-06-01 10:11
marcanif you get the same kind of issue, it's FS/app-level
#asahi-dev · 2023-06-01 10:11
marcaneric_engestrom: if you want to test the "clean NVMe shutdown, hard FS-level shutdown" case (i.e. the "this is totally not Apple's fault" case), hold down the power button. I'm pretty sure we have the NVMe driver doing a clean shutdown in that case.
#asahi-dev · 2023-06-01 10:08
marcanso that means the data will wait around 30 seconds before it gets flushed if nothing else flushes it
#asahi-dev · 2023-06-01 10:08
marcan/proc/sys/vm/dirty_expire_centisecs defaults to 30s
#asahi-dev · 2023-06-01 10:08
marcanah but wait
#asahi-dev · 2023-06-01 10:07
marcanso unless you have a high/saturating write workload, you'd expect it to stay caught up to within that time give or take
#asahi-dev · 2023-06-01 10:06
marcanbut like if your machine is otherwise idle, dirty pages should start to be written out within 5 seconds
#asahi-dev · 2023-06-01 10:06
marcanit means the kernel tries to write out data every 5 seconds. it doesn't mean it won't fall further behind than that though.
#asahi-dev · 2023-06-01 10:05
marcanbut yeah it just looks like you had a bunch of stuff in page cache and not flushed...
#asahi-dev · 2023-06-01 10:04
marcanI would normally not expect much data loss along those lines though, I think /proc/sys/vm/dirty_writeback_centisecs defaults to 5 seconds?
#asahi-dev · 2023-06-01 10:00
marcaneric_engestrom: those all sound like normal corruption issues you'd expect on ext4 if apps aren't doing the requisite flushes, and I doubt git flushes anything because that would probably kill performance
#asahi-dev · 2023-05-30 13:41
marcanplus that whole asmedia thing needs hardening anyway, it was never fully reliable
#asahi-dev · 2023-05-30 13:41
marcanI think I'll look at this one tomorrow or something (along with the firmware extraction since it kind of goes with it)
#asahi-dev · 2023-05-30 13:34
marcanhm, then that's weird
#asahi-dev · 2023-05-30 13:34
marcanbut does it leave the PCIe stuff up?
#asahi-dev · 2023-05-30 13:33
marcanwhat does u-boot do when it doesn't have that?
#asahi-dev · 2023-05-30 13:33
marcanprobably because no asmedia firmware yet in u-boot?
#asahi-dev · 2023-05-30 13:31
marcanjannau: gpio thing should be fixed
#asahi-dev · 2023-05-30 13:30
marcanbut anyway
#asahi-dev · 2023-05-30 13:30
marcanKDE bug in that case
#asahi-dev · 2023-05-30 13:30
marcanI think the backend just rejects settings with start > stop or something
#asahi-dev · 2023-05-30 13:30
marcanactually it seems you need to change both anyway, lol
#asahi-dev · 2023-05-30 13:29
marcanwhat I mean is using stop_threshold to drive the change is better
#asahi-dev · 2023-05-30 13:29
marcanI'm not hiding it, otherwise KDE's UI won't work, it wants both
#asahi-dev · 2023-05-30 13:26
marcanwhich means you need to change both to flip
#asahi-dev · 2023-05-30 13:26
marcankettenis: ah yeah, that actually behaves better with KDE, because it clamps the start threshold to stop - 1
#asahi-dev · 2023-05-30 13:19
marcanwe could probably implement that one as readonly, but at least KDE won't like that
#asahi-dev · 2023-05-30 13:18
marcanah yeah, so then we want the same thing and that driver didn't even bother to implement start_threshold

plus ›