[Bug 221073] xHCI host controller dies on resume from s2idle on AMD Strix Halo [1022:1587]

[bugzilla-daemon@kernel.org]

https://bugzilla.kernel.org/show_bug.cgi?id=221073

--- Comment #14 from Michał Pecio (michal.pecio@gmail.com) ---
Interesting. Your new dmesg shows that the quirk works: all driver state is
discarded and the xHC is reset and configured again.

Also, with or without the quirk, USBSTS on command timeout has the EINT bit
set, which means that IRQ is pending according to the xHC and our handler
hasn't cleared the bit yet, which is the first thing it should do when it runs.

Obviously the HC could be broken and ignoring requests to clear the bit, while
we also have a bug which causes pending events to be missed by the handler
without even logging an error, but I think the most likely and straightforward
explanation is that IRQs really aren't being delivered for some reason.

Unbind/bind possibly helps because it frees and allocates IRQs again. Maybe
disabling MSIs (xhci_hcd.quirks=0x40) would make a difference.

I think at this point I would try emailing linux-pci to see what they think
(I'm not sure if that subsystem monitors bugzilla). Because it looks like the
IRQ simply isn't being reactivated on resume.

-- 
You may reply to this email to add a comment.

You are receiving this mail because:
You are watching the assignee of the bug.