Re: [PATCH] do not count pages in holes with sparsemem

[Franck Bui-Huu <vagabon.xyz@gmail.com>]

Atsushi Nemoto wrote:
> On Wed, 26 Jul 2006 16:33:45 +0200, Franck Bui-Huu <vagabon.xyz@gmail.com> 
> wrote:
>> I don't think that's correct to mark them as "reserved". Basicaly
>> "reserved" means that it belongs to the kernel (code or data), these
>> holes are not and we will end up to have wrong value as you pointed
>> out.
>>
>> Having quick look at sparsemem code, I don't think that it expects
>> to have holes inside a section, do it ? If so you probably have to
>> fix up your section size...
> 
> Yes, for such small holes, sparsemem and flatmem is same.  We can use
> smaller section size to save more memory, but I suppose it will be a
> bit slower.
> 

I'm suprised that sparsemem code doens't check for holes inside
sections. I would feel really more confortable to use sparsemem if a
check like the following patch exists. We could safely use pfn_valid()
in _any_ cases and if holes exist inside sections then the user have
to fix up its section sizes.

what do you think ?

                Franck

-- >8 --

diff --git a/mm/sparse.c b/mm/sparse.c
index 86c52ab..4c29a13 100644
--- a/mm/sparse.c
+++ b/mm/sparse.c
@@ -119,6 +119,13 @@ void memory_present(int nid, unsigned lo
 {
        unsigned long pfn;
 
+       if (start & (PAGES_PER_SECTION-1) || end & (PAGES_PER_SECTION-1)) {
+               printk(KERN_ERR "SPARSEMEM: memory area (%lx-%lx) creates a "
+                      "hole inside a section, fix your SECTION_SIZE_BITS "
+                      "value...\n", start, end);
+               BUG();
+       }
+       
        start &= PAGE_SECTION_MASK;
        for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
                unsigned long section = pfn_to_section_nr(pfn);