#asahi-dev · 2022-03-02 16:38
svenbut m1n1 should’ve already taken it out of reset #asahi-dev · 2022-03-02 16:38
svenall regs read back as 0 then and the dwc driver can’t find some magic value #asahi-dev · 2022-03-02 16:37
svenmaybe it’s still in reset #asahi-dev · 2022-03-02 16:37
svenlol#asahi-dev · 2022-03-02 16:31
svenbut there it was a minute or so before it decided to reboot#asahi-dev · 2022-03-02 16:31
svensame thing happen a while ago when I tried to bring up cio to test my mailbox patches there#asahi-dev · 2022-03-02 16:30
svenI’ve only been poking some registers there and it just rebooted my system a few seconds later#asahi-dev · 2022-03-02 16:28
svenlol, so atcphy definitely has a few “reset the whole system if you do something wrong” modes #asahi-dev · 2022-03-02 16:00
svenmy bet for breakage would’ve been SART or rtkit though#asahi-dev · 2022-03-02 15:48
sveni don’t think I ever tried anything as a module fwiw#asahi-dev · 2022-03-02 12:07
svenouch#asahi-dev · 2022-03-02 11:45
svenit's the same for this weird thing where the linux driver sometimes misses irqs or commands for other people. i could reproduce it once (but I even doubt that at this point) but since then it's been working fine for me even without that additional mystery spinlock around command submission#asahi-dev · 2022-03-02 11:44
sven<3 ANS#asahi-dev · 2022-03-02 11:44
svenlol#asahi-dev · 2022-03-02 11:35
svenand then wait for NVME_CSTS.SHST#asahi-dev · 2022-03-02 11:35
svenyeah, you can set NVME_CC.SHN to trigger the shutdown#asahi-dev · 2022-03-02 11:34
svenbut let me check again#asahi-dev · 2022-03-02 11:34
sveni kinda take the big hammer approach by clearing NVME_CC.EN#asahi-dev · 2022-03-02 11:34
svenso i think there's another way to shut down nvme a bit more cleanly#asahi-dev · 2022-03-02 11:32
svencan you check if nvme_ctrl_disable was actually successful?#asahi-dev · 2022-03-02 11:27
svenno idea why it panics then though :/#asahi-dev · 2022-03-02 11:24
sveniirc it’s supposed to delete the queues once NVME_CC.EN has been cleared #asahi-dev · 2022-03-02 11:24
svenhuh. can you also remove the disable_ctrl call?#asahi-dev · 2022-03-02 11:22
svensounds like the queues are deleted then though and the print is just misleading#asahi-dev · 2022-03-02 11:21
svenat least on the cq. the sq doesn’t have a head/tail because it’s not really a queue when the nvmmu is used #asahi-dev · 2022-03-02 11:20
sven+to#asahi-dev · 2022-03-02 11:20
svencan you try send a few commands? It should show something different from 0/0 then#asahi-dev · 2022-03-02 11:17
sveni just tested it on my t8103 and it still works there#asahi-dev · 2022-03-02 11:17
svenyeah#asahi-dev · 2022-03-02 11:16
svenyeah#asahi-dev · 2022-03-02 11:13
sventags can be reused once they appear on the completion queue and have been written to NVMMU_TCB_INVAL#asahi-dev · 2022-03-02 11:12
svenI only realized later that I could've also decreases the queue size then but forgot to fix that#asahi-dev · 2022-03-02 11:11
svenyeah, I only queue one command at a time there anyway#asahi-dev · 2022-03-02 11:09
svenhrm, iirc those things should disappear (or at least have size: 0)#asahi-dev · 2022-03-02 11:03
svenand i guess the panic message isn't all that helpful?#asahi-dev · 2022-03-02 10:58
svenbut i think you already merged the PR for that#asahi-dev · 2022-03-02 10:58
svenhrm, it did that after i updated to 12.x when I didn't deleted the SQ/CQ#asahi-dev · 2022-03-02 10:57
svenuh. so it never completes the command?#asahi-dev · 2022-03-02 10:49
svenbut i think it'll complain in syslog anyway if you don't#asahi-dev · 2022-03-02 10:48
svendo you use namespace id 1?#asahi-dev · 2022-03-02 10:35
svennever saw that on my t8103#asahi-dev · 2022-03-02 10:34
sventhat sounds weird#asahi-dev · 2022-03-02 10:34
sveni don't have a t6k#asahi-dev · 2022-03-01 18:07
svenah, yes, and min_align_mask is why patching swiotlb is going to be a bit annoying#asahi-dev · 2022-03-01 17:58
svendomains#asahi-dev · 2022-03-01 17:58
svenah... right. i remember the issue with that 4k patch again. swiotlb couldn't align to more than PAGE_SIZE (at least in october) and upstream prefers to just disallow allocating unmanaged domain if there's a page size mismatch (but allow dma domains) instead of my workaround in attach_dev where I could actually reject untrusted devices for dma#asahi-dev · 2022-03-01 17:20
svenmaz: i think at least some issue in emacs is already fixed, bug#47125