Re: 2.6.17-rc5-git1: regression: resume from suspend(RAM) fails: libata issue

[Jens Axboe <axboe@suse.de>]

On Fri, May 26 2006, Mark Lord wrote:
> Mark Lord wrote:
> >Mark Lord wrote:
> >>My ata_piix based Notebook (Dell i9300) suspends/resumes perfectly (RAM 
> >>or disk)
> >>with 2.6.16.xx kernels, but fails resume on 2.6.17-rc5-git1 (the first 
> >>2.6.17-*
> >>I've attempted on this machine).
> >>
> >>On resume from RAM, after a 30-second-ish timeout, the screen comes on
> >>but the hard disk is NOT accessible.  "dmesg" in an already-open window
> >>shows this (typed in from handwritten notes):
> >>
> >>sd 0:0:0:0: SCSI error: return code = 0x40000
> >>end_request: I/O error, /dev/sda, sector nnnnnnn
> >...
> >
> >Ahh.. the fix for this was posted earlier today by Forrest Zhao.
> ..
> >Here is a modified version of Forrest's original patch, for 
> >2.6.17-rc5-git1.
> >It seems to have fixed the resume issue on my machine here,
> >so that things are now working as they were in the unpatched 2.6.16 
> >kernels.
> 
> >Jens Axboe wrote:
> >This has the problem that it introduces scsi specific knowledge into
> >ata_piix, something that is both a violation and a problem because we
> >are moving libata away from scsi. I think the best way to currently do
> >this is to introduce a ata_port_ops hook (pre_resume()?) that waits for
> >busy clear and gets called in ata_device_resume is probably the way to go.
> 
> Well, this problem has been with us all for a year now,
> and at this point it impacts practically *every* new "centrino"
> notebook out there.
> 
> We have a very simple workaround (previous post) that addresses it
> for 2.6.17, and it's about damn time it got fixed.
> 
> If there's a better solution for *2.6.17*, then *please* post it.
> Otherwise, we have a fix.  Maybe Linus or Andrew should just apply it?

Don't get me wrong, I could not agree more. I rely on suspend/resume all
the time, and the fact that we didn't get this fully working _in kernel_
years ago is really embarassing. So I fully want the ata_piix busy clear
patch to be in 2.6.17, my objection was merely that shoving scsi
knowledge into ata_piix is not the way to do it.

-- 
Jens Axboe