[PATCH] log machine check errors

Linas Vepstas linas at austin.ibm.com
Fri Aug 27 01:52:43 EST 2004


On Thu, Aug 26, 2004 at 07:36:14AM -0500, Jake Moilanen was heard to remark:
>
> On Thu, 26 Aug 2004 15:57:20 +1000
> Paul Mackerras <paulus at samba.org> wrote:
>
> > Jake Moilanen writes:
> >
> > > Somewhere along the line it looks like logging machine check errors
> > > never got put in 2.6.  Machine check error logs were one of the main
> > > reasons for storing logs to nvram.

Huh.  I was wondering what happened to this.  Note also, the 2.4 code
used a hard-coded buffer size, which, when set to 2K will crash in
firmware on power4.  The fix is to do the same rtas-error-log-max
value as in the patch I posted yesterday.  (I chatted with firmware
to confirm that this is what firmware expects).

> > > +char mce_data_buf[RTAS_ERROR_LOG_MAX]__page_aligned;
> >
> > Why is this page-aligned?
>
> The page alignment can probably be taken out.  The reason it was
> originally aligned in 2.4, was that this buffer was also used for
> check-exception (which we don't have in 2.6 or at least aren't
> supporting).  IIRC check-exception expects that the buffer not cross an
> LMB.  One way we can guarantee it is to make sure it also doesn't cross
> a page boundary, hence the page alignment.

In that case, this restriction would propbably apply to allthe other
buffers that use an rtas error log (rtas.c rtasd.c eeh.c ras.c).
I've asked firmware (vikram sethi) about this, no response yet.

--linas


** Sent via the linuxppc64-dev mail list. See http://lists.linuxppc.org/





More information about the Linuxppc64-dev mailing list