Re: 2.6.16-rc1-mm2 (mm5 too) panics

[Ed Tomlinson <edt@aei.ca>]

Hi,

I need some help figuring this one out.  I cannot use git bisect and applying the 2.6.15-1 reiserfs4
patch does not work with newer 2.6.16 levels - looks like the mutex conversion hits it hard.  Looking
in mm there are about 50 reiser4 patches...  

To reproduce this problem I need reiser4 though the issue maybe in libata or scsi.  Does anyone
have an idea how I can proceed to find what makes newer kernels get io errors which trigger a reiser4
panic?   The panic is a zam-597 (fs/reiser4/txnmgr.c) with an rc=-5 

Anyone have a git tree tracking linus with reiser4 patches applied staring before 2.6.15 - if so git 
bisect could be used to find the change causing the problem.

One datapoint.  In 2.6.15 + reiser4 2.6.15-1 works fine - the resiser4 filesystem is heavily used with
no errors.  Smart report no problems with the drive being used.  The libata driver used is sata_nv.

Please reply to my email - I am only subscribed to lkml.

Help!
Ed Tomlinson

ret = reiser4_write_logs(nr_submitted);
if (ret < 0)
    reiser4_panic("zam-597", "write log failed (%ld)\n", ret);

On Saturday 28 January 2006 08:46, Ed Tomlinson wrote:
> On Friday 27 January 2006 09:53, you wrote:
> > Ed Tomlinson wrote:
> > > Summarizing all this.  There are two problems here.
> > > 
> > > 1. reserifs4 panics when it gets io errors - I remember this was an issue that
> > > needed to be fixed in the R4 code before it moves to mainline...
> > > 
> > > 2. Why does a drive which is fine with 2.6.15-rc5-mm3, return a -5 with 2.6.16-mm3
> > > and above?  Smart reports no problems with the drive hardware.  What has changed 
> > > in the libata/scsi stacks?
> > 
> > That's a long answer.  Could you assist in narrowing down the versions 
> > which are affected?
> > 
> > It would also be useful if you could try vanilla kernels, and help us 
> > discover whether problems surfaces in 2.6.15, 2.6.15-git[1234], 
> > 2.6.16-rc1, etc.
> 
> Jeff,
> 
> I'll see what I can do with git bisect.  Given that reserifs4 is in the picture this may be
> fun...   I expect it will be a slow process (kernels take 40min to build here).
> 
> Thanks
> Ed Tomlinson
> 
>