#asahi-dev · 2022-04-07 11:01
svenah, good#asahi-dev · 2022-04-07 10:58
svenmight be too many differences to the arm pagetable format at that point for robin#asahi-dev · 2022-04-07 10:57
svenand then fix the kernel patch to not set that for everything.. but ugh...#asahi-dev · 2022-04-07 10:57
svenwould still be good to confirm that somehow#asahi-dev · 2022-04-07 10:57
svengiven that "disable subpage protection" no longer works I'm willing to bet that bit means uncachable as well#asahi-dev · 2022-04-07 10:49
svenwe should also figure out if the bit that used to be “disable subpage protection” actually is “mark this page as uncachable”#asahi-dev · 2022-04-07 10:48
svenhaven’t heard anything from the last time I pinged robin#asahi-dev · 2022-04-07 10:43
svenonce people start paying money to me it feels like a day job and I already have one of those :)#asahi-dev · 2022-04-07 10:43
svenno, and that’s quite intentional :)#asahi-dev · 2022-04-07 10:42
svenyes#asahi-dev · 2022-04-07 10:42
svenotherwise I would’ve already written support for it ;)#asahi-dev · 2022-04-07 10:40
sventhunderbolt and some other things on the t600x#asahi-dev · 2022-04-07 10:31
svennice!#asahi-dev · 2022-04-07 08:09
svenand the nvmmu inval stuff is because I forgot to initialize cq_phase when originally allocating the queues#asahi-dev · 2022-04-07 08:09
sventhat's inside nvme_reset_work or however it's called#asahi-dev · 2022-04-07 07:41
svenSMC is still alive though at that point I think, it just doesn't talk to the AP#asahi-dev · 2022-04-07 07:38
svenI don't think the normal SMC message handler does anything fancy but I never looked into if ANS also talks to SMC during boot#asahi-dev · 2022-04-07 07:37
svenoh, true. didn't even think of that#asahi-dev · 2022-04-07 06:14
svenI don’t understand why it would fail to boot though just because DCP does something #asahi-dev · 2022-04-07 06:14
svenother than that, the nvme bug is that ANS fails to boot and then my error handling apparently sucks#asahi-dev · 2022-04-07 06:13
sventhat fixes the usage of that weird function#asahi-dev · 2022-04-07 06:13
svenDCP needs to move to the raw iommu api #asahi-dev · 2022-04-06 17:42
svenplease re-read the discussion, we’ve covered that#asahi-dev · 2022-04-06 17:37
svensomeone should “just” find a bootrom bug so that I can run Linux on those M1 iPads please ;)#asahi-dev · 2022-04-06 17:37
svenso lots of work either way#asahi-dev · 2022-04-06 17:35
sveni don’t think DCP existed then and usb was definitely dwc2 with a different phy#asahi-dev · 2022-04-06 17:35
svenyeah. I know that the last chip with the checkm8 but would already be quite some effort#asahi-dev · 2022-04-06 17:34
svenyeah#asahi-dev · 2022-04-06 17:34
sventrue#asahi-dev · 2022-04-06 17:33
svenyeah, maaaaybe some rtkit stuff. But I think even nvme used to be different#asahi-dev · 2022-04-06 17:33
svenand I think those also used dwc2#asahi-dev · 2022-04-06 17:33
svensome of the later chips with that checkm8 bug might be possible, but I know that at least dart is different #asahi-dev · 2022-04-06 17:32
sventhere might be some stuff that can be reused (i2c) but I’d expect many things to be very different#asahi-dev · 2022-04-06 07:07
svenit wouldn’t surprise me at all if there’s something missing inside atcphy to make it work there as well#asahi-dev · 2022-04-06 07:06
svenyes#asahi-dev · 2022-04-06 06:56
sventry again with atcphy disabled#asahi-dev · 2022-04-06 06:55
svendid you have atcphy enabled?#asahi-dev · 2022-04-06 06:55
svenwouldn’t surprise me if the atcphy init needs to be done slightly different though#asahi-dev · 2022-04-06 06:54
svenrqou_: in theory yes, but I never tested Gadget mode with usb3. It does work with usb2 only though#asahi-dev · 2022-04-05 12:40
svensimilar to how M1 Pro/Max/Ultra (this is as bad as the USB naming *sigh*) is t6000/t6001/t6002 and how they also all share the same kernel#asahi-dev · 2022-04-05 12:38
svenM1 essentially is a variant of A14 and they share the same kernel iirc#asahi-dev · 2022-04-05 10:38
sveni thought it might be something like "re-issue and complete a command with a tag that's already in the CQ but not yet acked with the DB" but that also works#asahi-dev · 2022-04-05 10:37
svenokay, i guess I still don't know what that nvme race is. turns out writing to the CQ DB with head < tail will re-enable the interrupt#asahi-dev · 2022-04-05 07:34
svenso even if you drive the nvme controller using polling and handle everything, disable interrupts at the nvme level and then shut it down again you'll _still_ get that interrupt#asahi-dev · 2022-04-05 07:34
svenand that interrupt sticks until it's actually triggered and handled by the CPU#asahi-dev · 2022-04-05 07:31
svenso that "mystery lock" really just fixes that race and ensures the controller won't see any new commands inbetween the start of the irq handler and the write to the CQ DB#asahi-dev · 2022-04-05 07:31
svenlooks like the nvme irq is just a bit dumb. it'll trigger once when a new completion arrives, and if you send a new command and that command completes before you write the CQ doorbell it won't trigger again :/