Re: [Resend][PATCH] ACPI / EC: Remove race between EC driver and suspend process (rev. 3)
From: Alan Jenkins
Date: Thu Mar 04 2010 - 04:32:16 EST
On 3/4/10, Len Brown <lenb@xxxxxxxxxx> wrote:
> Isn't the problem at hand that the boot-kernel stops
> in mid-transaction, and the resumed hibernate image
> then blunders forward with its first transaction
> only to find the EC in an intermediate state?
> If yes, I then making the transaction atomic WRT
> the boot kernel stopping should fix the problem.
> The patch sets a FROZEN bit, and subsequent
> requests for EC transactions simply fail.
> How do we know that we set the bit at the right time?
> What transactions or parts of transactions will fail,
> and what are the consequences of those failures?
The problem transactions on my machine were probably started by
ec_check_sci(). I.e. QUERY transactions which read the value of an
event raised by the EC. (I triggered the problem by pressing ACPI
hotkeys, although I haven't been able to reproduce it since).
That might explain why the transactions can be asynchronous with the
s2disk task - because these transactions are run in the ACPI
workqueue. Perhaps the workqueue needs to be flushed after the EC GPE
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/