#asahi · 2022-02-09 11:54
sveni haven't submitted anything for sart/rtkit/nvme yet because we wanted to wait until SMC kinda-sorta works#asahi · 2022-02-09 11:53
svennot so sure about that#asahi · 2022-02-09 11:52
svenit's totally fine if people are aware of that ofc, but i feel like most end users aren't#asahi · 2022-02-09 11:51
sventhe rest isn't as bad, but if we have users who install m1n1/u-boot right now and then the bindings we made up change later that's going to be painful#asahi · 2022-02-09 11:50
svenespecially for things where the dt bindings aren't decided yet#asahi · 2022-02-09 11:50
svenyeah, i kinda hope we can reduce the patches we still have to carry in linux as soon as possible#asahi · 2022-02-09 09:18
svenI don’t think upstream has the nodes in the dt yet#asahi · 2022-02-08 21:45
svenno, that should’ve been enough #asahi · 2022-02-08 21:44
svenjust to make sure, did you reboot the m1 between trying each cable?#asahi · 2022-02-08 21:40
svenIt would’ve be usb if there was a pattern ;)#asahi · 2022-02-08 21:36
svenfor me a usb c/usb c cable with all lines present (i.e. it claims 10 gbps support) works just fine #asahi · 2022-02-08 21:35
svenknowing usb it’s probably both sides :P#asahi · 2022-02-08 21:31
svenif the M1 is in gadget mode (which it always is inside m1n1) and the usb connection is flakey and drops at the wrong moment the entire controller needs a hard reset to work again#asahi · 2022-02-08 21:30
svenas usual, usb is cursed and all that. but this specific dwc3 controllers deserves a special place in hell#asahi · 2022-02-08 21:28
svenget a better usb cable :-(#asahi · 2022-02-08 20:23
sventhe way it’s meant to be!#asahi · 2022-02-08 20:18
svenunless you just disable modules and make everything built-in :P#asahi · 2022-02-08 18:42
svenbut good to hear you guys didn't manage to break it either ;)#asahi · 2022-02-08 18:41
svenyeah, i've been using it for a while now. i'm confident it does what it's supposed to do. it just needs a more work to (hopefully) be accepted upstream#asahi · 2022-02-08 18:29
svenwell, "works". it won't for untrusted devices but it'll be a while before we have thunderbolt anyway#asahi · 2022-02-08 18:28
sventpw_rules: it works but I need to work on the review comments I got last time#asahi · 2022-02-08 18:27
svenhrm, i should also repost that ugly usb hack. and maybe work on that 4k iommu thing again. #asahi · 2022-02-08 18:15
svenyup#asahi · 2022-02-08 18:12
sveni only saw how to share tags across queues inside a single tagset but not how to share tags between ctrl.admin_tagset and ctrl.tagset#asahi · 2022-02-08 18:12
svensure! :)#asahi · 2022-02-08 18:11
svenaxboe: because i couldn't figure out how to do that#asahi · 2022-02-08 18:10
svenaxboe: w.r.t nvme: the controller is indeed limited to 1 admin and 1 io queue, both with 64 entries at most and the tags have to be shared across them because the controller FW uses the tag as an index for a shared data structure#asahi · 2022-02-08 12:25
sventhere's already stuff inside mainline and we'll upstream the rest as well#asahi · 2022-02-08 12:25
svenuh, no#asahi · 2022-02-08 07:17
svenaxboe: yeah, single io and a single admin queue only. I'll take a look into using shared tags, the current approach is just copied from the pci driver#asahi · 2022-02-07 21:42
sven☹️#asahi · 2022-02-07 20:11
sventhe T2 macs have the exact same controller fwiw. those just have that T2 chip that proxies it to the x86 core over pcie. i can actually flip a bit in a undocumented register and then the M1 controller disables its weird IOMMU and instead also needs that quirk to increase the sq entry size to 128 bytes#asahi · 2022-02-07 20:06
svenbut doing those in linux didn't change any behavior #asahi · 2022-02-07 20:06
svenyes, it issues two or three vendor specific commands. one of them sends the current time and the others read some information about the actual flash storage layout#asahi · 2022-02-07 20:04
sveni've brought up the controller inside m1n1 as well and it also happens there#asahi · 2022-02-07 20:03
svenyes, but much less frequent when using fio with the same settings#asahi · 2022-02-07 20:02
svenconsistently if you first write random blocks and then issue the flush#asahi · 2022-02-07 20:01
svenand it looks like it's indeed the controller which just hates flushes :(#asahi · 2022-02-07 19:52
svenugh.. the random-write-flush problem. someone confirmed the same thing also happens on the T2 macs fwiw (i.e. those who also need NVME_QUIRK_SKIP_CID_GEN)#asahi · 2022-02-06 18:53
svenoh god. I remember doing that maybe 15 years ago. After that gentoo was almost fun and simple :D#asahi · 2022-02-06 15:22
sven:D#asahi · 2022-02-04 14:12
svenj`ey: :D#asahi · 2022-02-04 14:12
svensee what j`ey said#asahi · 2022-02-04 14:11
sveni'll share whatever whenever i see fit and I won't change that#asahi · 2022-02-04 14:11
svenno, the point is that i don't want to be asked about progress#asahi · 2022-02-04 14:09
sven<sven> this is not a company, if you want weekly progress reports from me you better pay me damn well for it.#asahi · 2022-02-04 14:08
svenyes, i do understand.#asahi · 2022-02-04 14:08
sventhis is not a company, if you want weekly progress reports from me you better pay me damn well for it.#asahi · 2022-02-04 10:46
svenit's just yet another co-processor running rtkit#asahi · 2022-02-04 10:46
svenno