[Bug 215880] Resume process hangs for 5-6 seconds starting sometime in 5.16

[bugzilla-daemon@kernel.org]

https://bugzilla.kernel.org/show_bug.cgi?id=215880

--- Comment #52 from Damien Le Moal (damien.lemoal@wdc.com) ---
(In reply to Paul Ausbeck from comment #51)
> The PCIe BAR (Base address register) is usually GPU related as lspci
> verifies. BAR mapping failures are pretty routine on my older machines. My
> assessment is that they aren't relevant except to show that at least some
> GPU activity is deferred until after ATA bring up.
> 
> I think more important is that Restarting tasks ...  and even PM: suspend
> exit are happening before ATA bring up on 5.10 but after ATA bring up on 6.1.
> 
> This seems like a pretty serious change to me.

The "serious change" is a bug fix. "PM: suspend exit" sets the system out of
suspend mode but before commit 6aa0365a3c85, libata was "lying" about the state
of its devices. They were not really resumed and that was causing issues: some
users reported inability to use storage because of a deadlock between ata
resume and scsi resume. So we are not going back on that. This commit fixes a
serious issue and makes libata compliant with PM semantic.

I can try to revisit that fix to see if I can shorten the time to signaling a
"resumed" state for ata devices, but I doubt much is possible.

In any case, for now, having resume wait for the hdd spinup is safer than the
alternative. Your other issue about the GPU and mouse is I think different and
not related to libata resume. Not sure exactly what is happening. As I said, I
think it is worth signaling this to the pci, drm and or pm dev lists.

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

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