Re: [PATCH] mm: fix a regression with HIGHMEM introduced by changeset 7f1290f2f2a4d

[Jiang Liu <liuj97@gmail.com>]

On 11/07/2012 04:43 AM, Andrew Morton wrote:
> On Tue, 6 Nov 2012 09:31:57 +0800
> Jiang Liu <jiang.liu@huawei.com> wrote:
> 
>> Changeset 7f1290f2f2 tries to fix a issue when calculating
>> zone->present_pages, but it causes a regression to 32bit systems with
>> HIGHMEM. With that changeset, function reset_zone_present_pages()
>> resets all zone->present_pages to zero, and fixup_zone_present_pages()
>> is called to recalculate zone->present_pages when boot allocator frees
>> core memory pages into buddy allocator. Because highmem pages are not
>> freed by bootmem allocator, all highmem zones' present_pages becomes
>> zero.
>>
>> Actually there's no need to recalculate present_pages for highmem zone
>> because bootmem allocator never allocates pages from them. So fix the
>> regression by skipping highmem in function reset_zone_present_pages()
>> and fixup_zone_present_pages().
>>
>> ...
>>
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -6108,7 +6108,8 @@ void reset_zone_present_pages(void)
>>  	for_each_node_state(nid, N_HIGH_MEMORY) {
>>  		for (i = 0; i < MAX_NR_ZONES; i++) {
>>  			z = NODE_DATA(nid)->node_zones + i;
>> -			z->present_pages = 0;
>> +			if (!is_highmem(z))
>> +				z->present_pages = 0;
>>  		}
>>  	}
>>  }
>> @@ -6123,10 +6124,11 @@ void fixup_zone_present_pages(int nid, unsigned long start_pfn,
>>  
>>  	for (i = 0; i < MAX_NR_ZONES; i++) {
>>  		z = NODE_DATA(nid)->node_zones + i;
>> +		if (is_highmem(z))
>> +			continue;
>> +
>>  		zone_start_pfn = z->zone_start_pfn;
>>  		zone_end_pfn = zone_start_pfn + z->spanned_pages;
>> -
>> -		/* if the two regions intersect */
>>  		if (!(zone_start_pfn >= end_pfn	|| zone_end_pfn <= start_pfn))
>>  			z->present_pages += min(end_pfn, zone_end_pfn) -
>>  					    max(start_pfn, zone_start_pfn);
> 
> This ...  isn't very nice.  It is embeds within
> reset_zone_present_pages() and fixup_zone_present_pages() knowledge
> about their caller's state.  Or, more specifically, it is emebedding
> knowledge about the overall state of the system when these functions
> are called.
> 
> I mean, a function called "reset_zone_present_pages" should reset
> ->present_pages!
> 
> The fact that fixup_zone_present_page() has multiple call sites makes
> this all even more risky.  And what are the interactions between this
> and memory hotplug?
> 
> Can we find a cleaner fix?
> 
> Please tell us more about what's happening here.  Is it the case that
> reset_zone_present_pages() is being called *after* highmem has been
> populated?  If so, then fixup_zone_present_pages() should work
> correctly for highmem?  Or is it the case that highmem hasn't yet been
> setup?  IOW, what is the sequence of operations here?
> 
> Is the problem that we're *missing* a call to
> fixup_zone_present_pages(), perhaps?  If we call
> fixup_zone_present_pages() after highmem has been populated,
> fixup_zone_present_pages() should correctly fill in the highmem zone's
> ->present_pages?
Hi Andrew,
	Sorry for the late response:(
	I have done more investigations according to your suggestions. Currently
we have only called fixup_zone_present_pages() for memory freed by bootmem
allocator and missed HIGHMEM pages. We could also call fixup_zone_present_pages()
for HIGHMEM pages, but that will need to change arch specific code for x86, powerpc,
sparc, microblaze, arm, mips, um and tile etc. Seems a little overhead.
	And sadly enough, I found the quick fix is still incomplete. The original
patch still have another issue that, reset_zone_present_pages() is only called
for IA64, so it will cause trouble for other arches which make use of "bootmem.c".
	Then I feel a little guilty and tried to find a cleaner solution without
touching arch specific code. But things are more complex than my expectation and
I'm still working on that.
	So how about totally reverting the changeset 7f1290f2f2a4d2c3f1b7ce8e87256e052ca23125
and I will post another version once I found a cleaner way?
	Thanks!
	Gerry

> 
> 
> --
> 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/
> 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>