Messages in this thread | | | Date | Tue, 31 Jan 2012 05:39:00 +0530 | From | "Srivatsa S. Bhat" <> | Subject | Re: [PATCH v2 2/4] PM/Freezer: Make thaw_processes() thaw only userspace tasks |
| |
On 01/31/2012 05:00 AM, Tejun Heo wrote:
> On Tue, Jan 31, 2012 at 04:44:48AM +0530, Srivatsa S. Bhat wrote: >> Currently the situation is: >> >> freeze_processes() - freezes only userspace tasks >> freeze_kernel_threads() - freezes only kernel threads >> thaw_kernel_threads() - thaws only kernel threads >> thaw_processes() - thaws *everything* (both userspace tasks and kernel threads) > > Umm... I don't really get this. Why is this a problem? The list is > not even correct. freeze_kernel_threads() doesn't freeze "only" > kernel threads. It freezes all threads "including" kernel threads and
Oh, you are are right - the list is incorrect. I guess I got carried away thinking about thaw_kernel_threads().
> that's only natural because you can't freeze kernel threads without > freezing userland threads and of course you can't thaw userland > threads without thawing kernel threads. >
> The system simply won't work if you do it otherwise and making them
> disjoint operations increases the chance of bugs. These operations > are naturally enclosed within each other and trying to break them > apart isn't a good idea. >
Yeah, I get it now.. Thanks for the explanation!
> What's the problem you're trying to solve here? I don't really see > code clean up. Code is different but not necessarily cleaner and FWIW > it seems more unnatural and brittle to me. >
The thing is that, I wanted to avoid a bug in the patch posted at https://lkml.org/lkml/2012/1/29/47 as explained in the link.
So I guess I should have simply done:
freeze_kernel_threads() calls thaw_kernel_threads() upon error. The caller of freeze_kernel_threads() will call thaw_processes() if necessary.
This way even the SNAPSHOT_CREATE_IMAGE ioctl would remain safe.
I'll think it through again and post an updated patch. Thank you very much for the review!
Regards, Srivatsa S. Bhat
| |