Messages in this thread | | | Date | Wed, 08 May 2013 23:50:47 +0800 | From | Jiang Liu <> | Subject | Re: [PATCH v5, part3 11/15] mm: use a dedicated lock to protect totalram_pages and zone->managed_pages |
| |
On 05/08/2013 11:27 PM, Rik van Riel wrote: > On 05/08/2013 11:17 AM, Jiang Liu wrote: > >> @@ -5186,6 +5189,15 @@ early_param("movablecore", cmdline_parse_movablecore); >> >> #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ >> >> +void adjust_managed_page_count(struct page *page, long count) >> +{ >> + spin_lock(&managed_page_count_lock); >> + page_zone(page)->managed_pages += count; >> + totalram_pages += count; >> + spin_unlock(&managed_page_count_lock); >> +} >> +EXPORT_SYMBOL(adjust_managed_page_count); >> + > > Something I should have thought of when I reviewed the patch > last time, but forgot... > > What happens when the hotplug event adds more pages than fit > in this zone, and some of the pages should go in the next > zone? > > For example, think about a 3GB x86_64 machine, which gets > 2GB of memory hot-added. Roughly half may get added to the > DMA32 zone, the rest to the NORMAL zone. > > Do the callers of adjust_managed_page_count correctly make > one call for each zone, or does the above code open up a > window for a bug? Hi Rik, Thanks for review! Yes, the caller will make one call for each zone. Actually it will call adjust_managed_page_count() for each page. Regards! Gerry
| |