lkml.org 
[lkml]   [2002]   [Sep]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Fix 2.5.34+swsusp data corruption on IDE
Hi!

> > Jens, where is the problem? This should have absolutely zero impact on
> > "interesting" code, making only changes to initialization and
> > suspend/resume...
>
> I think we are looking to settle current problems before adding anything
> else in the mix. It will go in somehow, okay.

Good.

> > > Why are we not blocking read/write requests in the mainloop regardless?
> > > If the request gets to the subdriver, ide-disk, has it not gotten to far
> > > down the pipes?
> >
> > All processes capable of generating requests are safely stopped, so no
> > request should get down to drivers. That's why I simply BUG_ON(), not
> > block or anything more sophiscated. It should never ever happen.
>
> You have to qualify it to R/W, and that has to be done high than where it
> is now. What you can not block is the command to reset; regardless, if it
> is soft or hard. This is how you wake a device, then you have to clean up
> the mess resulting from the reset.
>
> > > Specifically ls120's and zips.
> > >
> > > I understand you are address disk but suspend is more than disk in the
> > > power management picture. Can you walk me through your process of sole
> > > concern with platter media? Remember microdrvies are platters too, as are
> > > flash drives, and memory drives.
> >
> > High levels are stopped, so there are no new requests coming.
>
> Where?

kernel/suspend.c; it puts all processes to refrigerator so there's
noone who can do request.

> > What is needed in idedisk_suspend is to make sure that no requests are
> > "in flight". DMA scribbling over random memory is not good thing.
> > idedisk_suspend then sends drive to standby to make sure writeback
> > caches are flushed.
>
> Well then you need to block the queue and the hold until the device goes
> idle via check-power. Then Flush-Cache, and repeat the host idle check
> until valid. Otherwise you have no way to insure all data is down.
>
> Do you disagree with this point?

No.. I guess you are right. Trouble is that I can reproduce data
corruption without patch and can no longer reproduce anything with it.

> Here are two types of media which use DMA and are read/write native.

Too bad I do not have any of these :-(. Yes bad things are going to
happen when suspending with these active.
Pavel
--
Casualities in World Trade Center: ~3k dead inside the building,
cryptography in U.S.A. and free speech in Czech Republic.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:28    [W:0.040 / U:9.296 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site