Ccing Mel.
On Wed, May 25, 2016 at 03:36:48PM -0700, Shi, Yang wrote:
On 5/25/2016 3:23 PM, Andrew Morton wrote:
On Wed, 25 May 2016 14:00:07 -0700 Yang Shi yang.shi@linaro.org wrote:
register_page_bootmem_info_node() is invoked in mem_init(), so it will be called before page_alloc_init_late() if CONFIG_DEFERRED_STRUCT_PAGE_INIT is enabled. But, pfn_to_nid() depends on memmap which won't be fully setup until page_alloc_init_late() is done, so replace pfn_to_nid() by early_pfn_to_nid().
What are the runtime effects of this fix?
I didn't experience any problem without the fix. During working on the page_ext_init() fix (replace to early_pfn_to_nid()), I added printk before each pfn_to_nid() calls to check which one might be called before page_alloc_init_late(), then this one is caught.
From the code perspective, it sounds not right since register_page_bootmem_info_section() may miss some pfns when CONFIG_DEFERRED_STRUCT_PAGE_INIT is enabled, just like the problem happened in page_ext_init().
Hello, Mel.
There was an issue in page_ext [1] due to your deferred struct page init feature. Before your change, we assumed that we can use pfn_to_nid() after memmap init is called. But, after your change, we can use pfn_to_nid() after page_alloc_init_late(). Yang found two call sites that uses pfn_to_nid() before page_alloc_init_late() and they could be fixed by using early_pfn_to_nid(). I guess that there are more problems due to this change so it's better to check it by patch author.
One thing I have noticed is that dirty_limit could be set wrongly. It is intialized by using freepage count. Since it is intialized before page_alloc_init_late(), freepages are not initialized yet and it could be wrong. If my analysis is correct, please fix it.
And, could you check again that there is no more problem?
Thanks.
[1] http://lkml.kernel.org/r/CAAmzW4OUmyPwQjvd7QUfc6W1Aic__TyAuH80MLRZNMxKy0-wPQ...