#
✕ nick:axboe
#asahi-dev · 2022-02-17 17:07
axboenot sure why I seem to be hitting the segments != 1 path in apple_nvme_map_data() though
#asahi-dev · 2022-02-17 17:06
axboeit worked
#asahi-dev · 2022-02-17 16:51
axboedone, let's see if it works...
#asahi-dev · 2022-02-17 16:40
axboeright?
#asahi-dev · 2022-02-17 16:40
axboelet's find out
#asahi-dev · 2022-02-17 16:40
axboeah screw it, we'll do it live
#asahi-dev · 2022-02-17 16:38
axboethere's no way to test apple-nvme in qemu yet, is there?
#asahi-dev · 2022-02-17 16:37
axboeit was a fun project
#asahi-dev · 2022-02-17 16:35
axboewas part of the "lets see how many iops we can do on a core" experiments, 1 of 2 series that hasn't been posted anywhere yet as it's a bit of a hack
#asahi-dev · 2022-02-17 16:35
axboewas planning on adding support for apple nvme just to test out what it can actually do, if I do that, then I can try the nvmmu inval optimization too as it'd likely make a difference at that point
#asahi-dev · 2022-02-17 16:34
axboeit'd bump the peak 17xK rand iops to substantially more
#asahi-dev · 2022-02-17 16:34
axboeI've got this funky series for io_uring that allows registered buffers to retain dma and iommu mappings, instead of doing them for each io submit and complete
#asahi-dev · 2022-02-17 16:33
axboeprobably not worthwhile to pursue, but something to keep in mind
#asahi-dev · 2022-02-17 16:31
axboesven: though for the batching, we could optimize the nvmmu_inval by writing each tag, then one readl_relaxed at the end...
#asahi-dev · 2022-02-17 16:18
axboeoh forgot, also running that "set aq depth to 2" patch
#asahi-dev · 2022-02-17 16:16
axboeone less to worry about :)
#asahi-dev · 2022-02-17 16:16
axboeirq fix is queued up btw, but I guess you saw that already
#asahi-dev · 2022-02-17 16:16
axboein the commit message too, but more importantly in the code
#asahi-dev · 2022-02-17 16:15
axboeI think we just need a good comment around why we _think_ it's needed
#asahi-dev · 2022-02-17 16:15
axboeif only there was documentation ;)
#asahi-dev · 2022-02-17 16:15
axboeif it is indeed inval vs doorbell write
#asahi-dev · 2022-02-17 16:14
axboeyour osx tracing sounds useful though
#asahi-dev · 2022-02-17 16:14
axboeyeah I know
#asahi-dev · 2022-02-17 16:13
axboemaybe differences in clock between simultaneous issuers? though I don't see how...
#asahi-dev · 2022-02-17 16:13
axboeand didn't see it
#asahi-dev · 2022-02-17 16:13
axboeI ran quite a bit with the cpus at max freq before that
#asahi-dev · 2022-02-17 16:13
axboedefinitely some correlation there, just not sure what!
#asahi-dev · 2022-02-17 16:11
axboeso while that could be improved, it doesn't matter in practice imho
#asahi-dev · 2022-02-17 16:10
axboeit'd be better if we didn't need to share this lock (and particularly since it's irq disabling on the submit path), but at the rates of these drives and the fact that it only has a single queue anyway, it's not going to be an iops monster anyway and hence it doesn't really matter
#asahi-dev · 2022-02-17 16:09
axboeI generally didn't feel comfortable without that lock to begin with, only part I'm a bit puzzled on is why the cpu freq changes makes this so much more likely to trigger
#asahi-dev · 2022-02-17 16:09
axboeas nvmmu invalidation is already under the anv lock
#asahi-dev · 2022-02-17 16:08
axboeso that won't happen with that patch either then
#asahi-dev · 2022-02-17 16:08
axboeok
#asahi-dev · 2022-02-17 16:07
axboetcb clear cleanup, lock around writel, and the flush deferral
#asahi-dev · 2022-02-17 16:07
axboeI'm running the three nvme patches now and it's all dandy
#asahi-dev · 2022-02-17 16:07
axboewould always hang
#asahi-dev · 2022-02-17 16:07
axboesven: reproduces trivially for me, was a bit harder with issues serialized, but could still hit it. consistently just doing a make -j8 kernel compile on it
#asahi-dev · 2022-02-17 16:06
axboesven: just around writel is enough, that's what I've been running and it's been rock solid
#asahi-dev · 2022-02-17 15:30
axboegood to see some noise on this topic
#asahi-dev · 2022-02-17 15:30
axboeI already had 3 separate people send me marcans flush rant this morning ;)
#asahi-dev · 2022-02-17 01:40
axboes/wrong/right, obviously
#asahi-dev · 2022-02-17 01:39
axboeI'm sure there's tons of shitty ones that just nop it :/
#asahi-dev · 2022-02-17 01:38
axboebut it's one of the 4 core primitives, really should get that wrong for any half-way decent device and above
#asahi-dev · 2022-02-17 01:38
axboeheh I know
#asahi-dev · 2022-02-17 00:53
axboeif previously acked writes aren't power loss stable after that, then the hw is defective
#asahi-dev · 2022-02-17 00:52
axboethat's what the flush cache command is for
#asahi-dev · 2022-02-17 00:40
axboeI used to use openbsd for a ppoatm router, when that was my inet connection
#asahi-dev · 2022-02-17 00:40
axboeI mean, all's well unless it crashes or you lose power
#asahi-dev · 2022-02-17 00:39
axboewell I guess any kind of fs perf numbers on openbsd needs to come with that caveat ;)
#asahi-dev · 2022-02-17 00:37
axboewow

plus ›