Atsushi Nemoto wrote: > With SPARSEMEM, the single node can contains some holes so there might > be many invalid pages. For example, with two 256M memory and one 256M Does SPARSEMEM is the only memory model where we can have memory holes ? > hole, some variables (num_physpage, totalpages, nr_kernel_pages, > nr_all_pages, etc.) will indicate that there are 768MB on this system. > This is not desired because, for example, alloc_large_system_hash() > allocates too many entries. > > Use free_area_init_node() with counted zholes_size[] instead of > free_area_init(). > > For num_physpages, use number of ram pages instead of max_low_pfn. > > Signed-off-by: Atsushi Nemoto <anemo@xxxxxxxxxxxxx> > > diff --git a/arch/mips/mm/init.c b/arch/mips/mm/init.c > index 802bdd3..d41dee5 100644 > --- a/arch/mips/mm/init.c > +++ b/arch/mips/mm/init.c [snip] > > @@ -174,29 +200,17 @@ #ifdef CONFIG_HIGHMEM > zones_size[ZONE_HIGHMEM] = high - low; > #endif > > +#ifdef CONFIG_SPARSEMEM > + pfn = 0; > + for (i = 0; i < MAX_NR_ZONES; i++) > + for (j = 0; j < zones_size[i]; j++, pfn++) > + if (!page_is_ram(pfn)) > + zholes_size[i]++; > + free_area_init_node(0, NODE_DATA(0), zones_size, > + __pa(PAGE_OFFSET), zholes_size); Does this code really need the ifdef CONFIG_SPARSEMEM ? Can't we make it generic instead. Only zholes_size[] initialisation really depends on the memory model. Of course FLATMEM will let zholes_size as is... If I remember correctly free_area_init_node() takes a pfn number as fourth parameter: __pa(PAGE_OFFSET) results in a physical address... BTW why using __pa(OFFSET) ? isn't it going to yield always into 0 ? At least on MIPS, it's defined as #define __pa(x) ((unsigned long) (x) - PAGE_OFFSET) why not using ARCH_PFN_OFFSET instead ? thanks Franck
