[BUG RESEND] unsuspend failure under load

[Sean Nyekjaer]

On Tue, Jul 06, 2021 at 11:35:44AM +0200, Sean Nyekjaer wrote:
> On Mon, Jul 05, 2021 at 08:58:38AM +0200, Sean Nyekjaer wrote:
> > Hi Miquel and Sascha,
> + Richard + linux-pm
> > 
> > We are having some trouble when our i.MX6 unsuspends while writes to
> > ubifs is in progess. In the log it looks like it syncing the filesystem
> > before suspend.
> > 
> > The SoC a i.MX6ul/ull, the issue is (lucky for us) quite easy to reproduce.
> > The reproduce script: [0]
> > Kernel log when it happens: [1]
> > 
> > I have bisected the bug to: ef347c0cfd61 ("mtd: rawnand: gpmi: Implement exec_op")
> > 
> > Any idea to where I should start looking? or to what happens?
> > 
> > Esben have posted to patches that relates to suspend/unsuspend but it
> > doesn't seem to releated to this issue.
> > 5bc6bb603b4d ("mtd: rawnand: gpmi: Fix suspend/resume problem")
> > d70486668cdf ("mtd: rawnand: gpmi: Restore nfc timing setup after suspend/resume")
> > 
> > /Sean
> 
> nand_resume() is called some time after ubi_io_write tries to write. Thats why
> mtd_write() is returning -EBUSY.
> 
> I have tried patch [3], and it seems to fix it.
> I think it would be okay to add the retry option, but the mdelay is not
> obviously a nogo.
> 
> Any idea to how we could wait here for the nand_resume() to be called?
> 
> @linux-pm:
> I have noticed "Filsystems sync" happens before "Freezing user space
> processes".
> If I apply patch [4] (without [3]), it would also fix our issue. But I
> don't have en insight in to what impact the change might have.
> 
> /Sean
> 

Ping Miquel + Sascha,

Have you had time to take a look at this?

/Sean