[PATCH] Tight check of pfn_valid on sparsemem - v2

[mel@csn.ul.ie (Mel Gorman)]

On Sun, Jul 18, 2010 at 07:18:31PM +0900, Minchan Kim wrote:
> Kukjin reported oops happen while he change min_free_kbytes
> http://www.spinics.net/lists/arm-kernel/msg92894.html
> It happen by memory map on sparsemem.
> 

First off, thanks for working on this.

> The system has a memory map following as. 
>      section 0             section 1              section 2
> 0x20000000-0x25000000, 0x40000000-0x50000000, 0x50000000-0x58000000
> SECTION_SIZE_BITS 28(256M)
> 
> It means section 0 is an incompletely filled section.
> Nontheless, current pfn_valid of sparsemem checks pfn loosely. 
> It checks only mem_section's validation but ARM can free mem_map on hole 
> to save memory space. So in above case, pfn on 0x25000000 can pass pfn_valid's 
> validation check. It's not what we want. 
> 
> We can match section size to smallest valid size.(ex, above case, 16M)
> But Russell doesn't like it due to mem_section's memory overhead with different
> configuration(ex, 512K section).
> 
> I tried to add valid pfn range in mem_section but everyone doesn't like it 
> due to size overhead.

Also IIRC, it was vunerable to a hole being punched in the middle of the
section.

> This patch is suggested by KAMEZAWA-san. 
> I just fixed compile error and change some naming. 
> 
> This patch registers address of mem_section to memmap itself's page struct's
> pg->private field. This means the page is used for memmap of the section.
> Otherwise, the page is used for other purpose and memmap has a hole.
> 
> This patch is based on mmotm-2010-07-01-12-19.
> 
> Signed-off-by: Minchan Kim <minchan.kim@gmail.com>
> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> Reported-by: Kukjin Kim <kgene.kim@samsung.com>
> ---
>  arch/arm/mm/init.c     |    9 ++++++++-
>  include/linux/mmzone.h |   21 ++++++++++++++++++++-
>  mm/Kconfig             |    5 +++++
>  mm/sparse.c            |   41 +++++++++++++++++++++++++++++++++++++++++
>  4 files changed, 74 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm/mm/init.c b/arch/arm/mm/init.c
> index cfe4c5e..4586f40 100644
> --- a/arch/arm/mm/init.c
> +++ b/arch/arm/mm/init.c
> @@ -234,6 +234,11 @@ static void __init arm_bootmem_free(struct meminfo *mi)
>  	arch_adjust_zones(zone_size, zhole_size);
>  
>  	free_area_init_node(0, zone_size, min, zhole_size);
> +
> +	for_each_bank(i, mi) {
> +		mark_memmap_hole(bank_pfn_start(&mi->bank[i]),
> +				bank_pfn_end(&mi->bank[i]), true);
> +	}
>  }

Why do we need to mark banks both valid and invalid? Is it not enough to
just mark the holes in free_memmap() and assume it is valid otherwise?

>  
>  #ifndef CONFIG_SPARSEMEM
> @@ -386,8 +391,10 @@ free_memmap(unsigned long start_pfn, unsigned long end_pfn)
>  	 * If there are free pages between these,
>  	 * free the section of the memmap array.
>  	 */
> -	if (pg < pgend)
> +	if (pg < pgend) {
> +		mark_memmap_hole(pg >> PAGE_SHIFT, pgend >> PAGE_SHIFT, false);
>  		free_bootmem(pg, pgend - pg);
> +	}
>  }
>  
>  /*
> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
> index 9ed9c45..2ed9728 100644
> --- a/include/linux/mmzone.h
> +++ b/include/linux/mmzone.h
> @@ -15,6 +15,7 @@
>  #include <linux/seqlock.h>
>  #include <linux/nodemask.h>
>  #include <linux/pageblock-flags.h>
> +#include <linux/mm_types.h>
>  #include <generated/bounds.h>
>  #include <asm/atomic.h>
>  #include <asm/page.h>
> @@ -1047,11 +1048,29 @@ static inline struct mem_section *__pfn_to_section(unsigned long pfn)
>  	return __nr_to_section(pfn_to_section_nr(pfn));
>  }
>  
> +void mark_memmap_hole(unsigned long start, unsigned long end, bool valid);
> +

The naming is confusing with the "valid" parameter.

What's a "valid hole"? I can see that one being a cause of head
scratching in the future :)

> +#ifdef CONFIG_SPARSEMEM_HAS_HOLE

Why not use CONFIG_ARCH_HAS_HOLES_MEMORYMODEL ?

> +static inline int page_valid(struct mem_section *ms, unsigned long pfn)
> +{
> +	struct page *page = pfn_to_page(pfn);
> +	struct page *__pg = virt_to_page(page);
> +	return __pg->private == (unsigned long)ms;
> +}
> +#else
> +static inline int page_valid(struct mem_section *ms, unsigned long pfn)
> +{
> +	return 1;
> +}
> +#endif
> +
>  static inline int pfn_valid(unsigned long pfn)
>  {
> +	struct mem_section *ms;
>  	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
>  		return 0;
> -	return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
> +	ms = __nr_to_section(pfn_to_section_nr(pfn));
> +	return valid_section(ms) && page_valid(ms, pfn);
>  }

So it appears here that we unconditionally check page_valid() but we know
which sections had holes in them at the time we called mark_memmap_hole(). Can
the sections with holes be tagged so that only some sections need to call
page_valid()? As it is, ARM will be taking a an performance hit just in case
the section has holes but it should only need to take a performance hit
on the corner case where a section is not fully populated.

>  
>  static inline int pfn_present(unsigned long pfn)
> diff --git a/mm/Kconfig b/mm/Kconfig
> index 527136b..959ac1d 100644
> --- a/mm/Kconfig
> +++ b/mm/Kconfig
> @@ -128,6 +128,11 @@ config SPARSEMEM_VMEMMAP
>  	 pfn_to_page and page_to_pfn operations.  This is the most
>  	 efficient option when sufficient kernel resources are available.
>  
> +config SPARSEMEM_HAS_HOLE
> +	bool "allow holes in sparsemem's memmap"
> +	depends on ARM && SPARSEMEM && !SPARSEMEM_VMEMMAP
> +	default n
> +
>  # eventually, we can have this option just 'select SPARSEMEM'
>  config MEMORY_HOTPLUG
>  	bool "Allow for memory hot-add"
> diff --git a/mm/sparse.c b/mm/sparse.c
> index 95ac219..76d5012 100644
> --- a/mm/sparse.c
> +++ b/mm/sparse.c
> @@ -615,6 +615,47 @@ void __init sparse_init(void)
>  	free_bootmem(__pa(usemap_map), size);
>  }
>  
> +#ifdef CONFIG_SPARSEMEM_HAS_HOLE
> +/*
> + * Fill memmap's pg->private with a pointer to mem_section.
> + * pfn_valid() will check this later. (see include/linux/mmzone.h)
> + * Evenry arch should call
> + * 	mark_memmap_hole(start, end, true) # for all allocated mem_map
> + * 	and, after that,
> + * 	mark_memmap_hole(start, end, false) # for all holes in mem_map.
> + * please see usage in ARM.
> + */
> +void mark_memmap_hole(unsigned long start, unsigned long end, bool valid)
> +{
> +	struct mem_section *ms;
> +	unsigned long pos, next;
> +	struct page *pg;
> +	void *memmap, *mapend;
> +
> +	for (pos = start; pos < end; pos = next) {
> +		next = (pos + PAGES_PER_SECTION) & PAGE_SECTION_MASK;
> +		ms = __pfn_to_section(pos);
> +		if (!valid_section(ms))
> +			continue;
> +
> +		for (memmap = (void*)pfn_to_page(pos),
> +			/* The last page in section */
> +			mapend = pfn_to_page(next-1);
> +			memmap < mapend; memmap += PAGE_SIZE) {
> +			pg = virt_to_page(memmap);
> +			if (valid)
> +				pg->private = (unsigned long)ms;
> +			else
> +				pg->private = 0;
> +		}
> +	}
> +}
> +#else
> +void mark_memmap_hole(unsigned long start, unsigned long end, bool valid)
> +{
> +}
> +#endif
> +

The patch should also delete memmap_valid_within() and replace it with a
call to pfn_valid_within(). The reason memmap_valid_within() existed was
because sparsemem had holes punched in it but I'd rather not see use of
that function grow.

>  #ifdef CONFIG_MEMORY_HOTPLUG
>  #ifdef CONFIG_SPARSEMEM_VMEMMAP
>  static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
> -- 
> 1.7.0.5
> 

-- 
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab