Re: sata_sil24 broken since 2.6.23-rc4-mm1

[Jeff Garzik]

Torsten Kaiser wrote:
> On 9/27/07, Tejun Heo <htejun@xxxxxxxxx> wrote:
> > Tejun Heo wrote:
> > > Torsten Kaiser wrote:
> > > > Comparing the driver/ata directory from rc3-mm1 and rc4-mm1 the
> > > > following change looked the most suspicions to me:
> > > > http://git.kernel.org/?p=linux/kernel/git/jgarzik/libata-dev.git;a=blobdiff;f=drivers/ata/sata_sil24.c;h=3dcb223117be9739ee04d70b6bfc776a4b839a3f;hp=e0cd31aa8002350add53ba6ff07493e503275244;hb=020bc1bd8d369a77bd9379cd9763ac0057651753;hpb=8d4bdf8087e682df98bdb856f6ad451bf6d597e7
> > > >
> > > > That after rc4-mm1 the sata_sil24.c did not change anymore also
> > > > matches the occurrence of the error.
> > > >
> > > > To confirm my theorie I exchanged the sata_sil24.c from rc8-mm1 with
> > > > the version from rc3-mm1.
> > > > I was able to boot the resulting kernel successfully 5 times, without
> > > > the error happening again.
> > > Thanks a lot for chasing down the problem.  The changed code is address
> > > initialization path and it's weird that it causes intermittent failures,
> > > not a consistent one.
> > >
> > > Anyways, does the attached patch fix the problem?
>
> I'm starting to *really* hate that bug.
> My analysis was wrong, as I booted to modified 2.6.23-rc8-mm1 this
> morning, that failed too. (Same error messages as -rc7-mm1 from the
> first mail in this thread.)
> So it's not that change that causes the breakage.
>
> And I'm not really finding a good pattern to what boots fail and what work.
> It seems to only fail, if I completely power off the system for
> several hours. (Using the physical switch at the backside of the
> powersupply, not the normal soft-off)
>
> One of the five boots I tried yesterday, I also powered the system
> completely off that way, but only leaving it off ~10..20 seconds
> seemed not to trigger the bug.
>
> But I still think that is not a hardware failure, as the -rc3-mm1
> kernel never showed that error, even when I used it several times
> after the first -rc4-mm1 failures.
>
> > If not, can you add printk of iomap[SIL24_PORT_BAR], offset, initialized
> > cmd_addr and scr_addr in the loop and see whether anything is different
> > between when the driver works and fails.
>
> Should I do this anyway?
>
> I compared the dmesg form good and bad boots with -rc7-mm1 but could
> not see any difference, so do you think that these additional
> diagnostics could show a difference?
> Or could you suggest any other debugging options I should try?

I think since its a reproducible problem, I think it's easiest to get 
you straight to git-bisect.  In this case, that would be

1. Clone branch "upstream" of 
git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev.git

2. Test.  If bug persists, you have narrowed down the problem to the -mm 
changes from the SATA developers, that are to be sent for 2.6.24.  If 
the problem does not persist, then it's a problem added in the -mm 
patchset alone, which carries few ATA patches outside of libata-dev.git.

3. If the problem is in libata-dev.git#upstream (likely), you can now 
use git-bisect to find the specific commit that causes the problems. 
Read the git-bisect man page for full details, but the basics are

	a) start with a known good point (v2.6.22? v2.6.23?) and
	known bad point (HEAD, aka the most recent commit in
	libata-dev.git#upstream)

	b) build and boot kernels, marking each as known-good or
	known-bad.

	c) This process will systematically narrow down the problem
	to a single git commit.

Regards,

	Jeff



-
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/