2004-11-17 03:10:15

by Shaohua Li

Subject: [PATCH]Teach drivers don't call may-sleep routines in resume code

Hi,
We already found one driver (PCI link device driver) does the odd thing, so alert other drivers.

Thanks,
Shaohua

--- 2.6/Documentation/power/devices.txt.orig 2004-11-17 10:42:25.160212832 +0800
+++ 2.6/Documentation/power/devices.txt 2004-11-17 10:46:11.070869192 +0800
@@ -70,6 +70,9 @@ System devices will only be suspended wi
after all other devices have been suspended. On resume, they will be
resumed before any other devices, and also with interrupts disabled.

+*CAUTION*: The resume methods of drivers (normal devices and system devices)
+should never use any may-sleep methods, since when resume from memory (S3),
+no task is running.

Runtime Power Management

2004-11-19 01:55:21

by Shaohua Li

Subject: Re: [PATCH]Teach drivers don't call may-sleep routines in resume code

On Wed, 2004-11-17 at 11:04, Li Shaohua wrote:
> Hi,
> We already found one driver (PCI link device driver) does the odd thing, so alert other drivers.
>
> Thanks,
> Shaohua
>
> --- 2.6/Documentation/power/devices.txt.orig 2004-11-17 10:42:25.160212832 +0800
> +++ 2.6/Documentation/power/devices.txt 2004-11-17 10:46:11.070869192 +0800
> @@ -70,6 +70,9 @@ System devices will only be suspended wi
> after all other devices have been suspended. On resume, they will be
> resumed before any other devices, and also with interrupts disabled.
>
> +*CAUTION*: The resume methods of drivers (normal devices and system devices)
> +should never use any may-sleep methods, since when resume from memory (S3),
> +no task is running.
Oops, the description isn't precise. Some may-sleep routines such as
waiting a sempahore are ok in resume routines, but some (kmalloc) are
bad. Please ignore it till I get a precise description.

Thanks,
Shaohua