On 04/15/2013 12:56 PM, Paul Mackerras wrote:
> On Sat, Apr 13, 2013 at 11:36:20PM +0800, Jiang Liu wrote:
>> Commit 600cc5b7f6 "mm: Kill NO_BOOTMEM version free_all_bootmem_node()"
>> has kill free_all_bootmem_node() for NO_BOOTMEM.
>>
>> Currently the usage pattern for free_all_bootmem_node() is like:
>> for_each_online_pgdat(pgdat)
>> 	free_all_bootmem_node(pgdat);
>>
>> It's equivalent to free_all_bootmem(), so this patchset goes one
>> step further to kill free_all_bootmem_node() for BOOTMEM too.
>>
>> This patchset also tries to clean up code and comments related to
>> VALID_PAGE() because it has been removed from kernel long time ago.
>>
>> Patch 1-11:
>> 	Kill free_all_bootmem_node()
>> Patch 12-16:
>> 	Clean up code and comments related to VALID_PAGE()
>> Patch 17:
>> 	Fix a minor build warning for m68k
>> Patch 18:
>> 	merge Alpha's mem_init() for UMA and NUMA.
>> Patch 19:
>> 	call register_page_bootmem_info_node() from mm core
> 
> How does this not break bisection?  Will a kernel still boot with
> patches 1-11 applied but not patch 19?  As far as I can see such a
> kernel would have no memory available to it after mem_init() time
> and would therefore crash early in boot.
Hi Paul,
	Thanks for review!
	Patch 1-11 replace free_all_bootmem_node() with free_all_bootmem(),
so all normal pages will be freed into the buddy system as before. And
patch 1-11 are independent with patch 19.
	Gerry
> 
> Paul.
>