[Cbe-oss-dev] [PATCH] Fix sparsemem on Cell

Dave Hansen haveblue at us.ibm.com
Sat Dec 16 04:57:50 EST 2006


On Fri, 2006-12-15 at 18:22 +0100, Michael Buesch wrote:
> On Friday 15 December 2006 17:53, Dave Hansen wrote:
> >  lxc-dave/init/main.c     |    4 ++++
> >  lxc-dave/mm/page_alloc.c |   28 +++++++++++++++++++++++++---
> >  2 files changed, 29 insertions(+), 3 deletions(-)
> >
> > diff -puN init/main.c~sparsemem-fix init/main.c
> > --- lxc/init/main.c~sparsemem-fix	2006-12-15 08:49:53.000000000 -0800
> > +++ lxc-dave/init/main.c	2006-12-15 08:49:53.000000000 -0800
> > @@ -770,6 +770,10 @@ static int init(void * unused)
> >  	free_initmem();
> >  	unlock_kernel();
> >  	mark_rodata_ro();
> > +	/*
> > +	 * Memory hotplug requires that this system_state transition
> > +	 * happer after free_initmem().  (see memmap_init_zone())
> 
> s/happer/happens/
> 
> Other than that, can't this possibly race and crash here?
> I mean, it's not a big race window, but it can happen, no?

That's a good point.  Nice eye.

There are three routes in here: boot-time init, an ACPI call, and a
write to a sysfs file.  Bootmem is taken care of.  The write to a sysfs
file can't happen yet because userspace isn't up.

The only question would be about ACPI.  I _guess_ an ACPI event could
come in at any time, and could hit this race window.  

One other thought I had was to add an argument to memmap_init_zone() to
indicate that the memory being fed to it was contiguous and did not need
the validation checks.

Anybody have thoughts on that?

-- Dave




More information about the cbe-oss-dev mailing list