Re: 2.6.15-rc4 panic in __nr_to_section() with CONFIG_SPARSEMEM

[Dave Hansen <haveblue@us.ibm.com>]

On Mon, 2005-12-12 at 08:19 -0800, Badari Pulavarty wrote:
> NACK. 
> 
> Like I mentioned to you privately, I still get Oops with this patch for
> largepages. Please let me know, if you have a new version to try.

I'm going to throw this back over the wall at Adam and Dave Gibson.  It
is going to take a bit more than a cosmetic fix.  The complexity comes
because there are two _distinct_ hugetlb cases here.  The first is when
the HPTE is condensed into the PMD like on normal i386 or in the !
64K_PAGES case on ppc64.  The second is when the HPTE is stored like a
normal PTE in a PTE page like on the 64K_PAGES ppc64 case.

These need to be handled in two different places in the smaps pagetable
walk.  Add that into the normal small PTE case and we end up having
_three_ cases to handle for the end of the pagetable walk.

Before we did faulting for these PTEs, it would have been easy to have a
if() to hack it in at the top of the pagetable walk, but we can't do
that any more.

The big question is: do we ever need to deal with large pages in a
normal pagetable walk?  If not, we can probably just hack the two extra
cases in.  How do we tell in generic code whether we're looking at a
real "huge PMD", or one that points to a PTE page with "huge PTE"s?
That seems to be a ppc64-specific question for now.

Badari, appended is a patch that doesn't fix the accounting in smaps,
but will simply skip the huge page vmas.  It will at least keep you from
oopsing.

-- Dave

 smaps-fix-dave/fs/proc/task_mmu.c |    2 +-
 1 files changed, 1 insertion(+), 1 deletion(-)

diff -puN fs/proc/task_mmu.c~smaps_fix fs/proc/task_mmu.c
--- smaps-fix/fs/proc/task_mmu.c~smaps_fix	2005-12-12 13:19:05.000000000 -0800
+++ smaps-fix-dave/fs/proc/task_mmu.c	2005-12-12 13:21:07.000000000 -0800
@@ -289,7 +289,7 @@ static int show_smap(struct seq_file *m,
 	struct mem_size_stats mss;
 
 	memset(&mss, 0, sizeof mss);
-	if (vma->vm_mm)
+	if (vma->vm_mm && !is_vm_hugetlb_page(vma))
 		smaps_pgd_range(vma, vma->vm_start, vma->vm_end, &mss);
 	return show_map_internal(m, v, &mss);
 }
_


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/