UFS related panic (daily <-> find)

[John Baldwin]

On Monday, October 21, 2013 9:30:36 am rank1seeker at gmail.com wrote:
> > > Same drill as before, see what instruction this is.  Actually, this 
> looks
> > > to
> > > be in the same location as your last panic, so a NULL pointer is 0x1
> > > instead
> > > of 0x0 again.  In my experience, this would still indicate failing RAM 
> to
> > > me,
> > > memtest86+ notwithstanding (memtest86+ is single threaded AFAIK, so it 
> may
> > > not stress the hardware quite the same, e.g. if the error is heat 
> related,
> > > etc.).
> > 
> > 
> > memtest* cannot conclusively diagnose a dimm as good.  Usually the only
> > practical solution is to swap modules with known good ones.
> > 
> 
> 
> 0xc082c552 <inodedep_find+13>:  cmp    %ecx,0x24(%eax)
>     PREVIOUS we talked about
> 0xc083bd42 <inodedep_find+13>:  cmp    %ecx,0x24(%eax)
>     CURRENT ONE

Different instruction pointer doesn't matter.  The error is in the memory
that %eax is loaded from in a prior instruction.

> Now, after all this I recompiled kernel and world and there was no crash.
> How can it be, when it is far more stresing dan daily's 'find'?!

Because it might have shuffled where the bad memory cell now lives by having
the kernel text + data laid out differently in RAM?

> I see addresses 0xc08* and 0xc06* appearing each time, so as I have four 
> DDR1 (400) modules, each of 256 MB = 1GB, can those addresses aid me in 
> targeting failing module?

The virtual addresses (0xc*) do not matter.  They are not physical addresses
which are what you would need.

> If I can't use memtest86+-4.20, to determine failing module, then what is a 
> use of it at all?
> Test RAM speed perhaps?

Swap out your dimms.  That's really the only test, esp. if you have a
reproducible crash.

-- 
John Baldwin