On Fri, 2013-03-08 at 13:31 -0800, Andrew Morton wrote: > On Fri, 8 Mar 2013 08:41:41 -0700 Toshi Kani <toshi.kani@xxxxxx> wrote: > > > remove_memory() calls walk_memory_range() with [start_pfn, end_pfn), > > where end_pfn is exclusive in this range. Therefore, end_pfn needs > > to be set to the next page of the end address. > > > > ... > > > > --- a/mm/memory_hotplug.c > > +++ b/mm/memory_hotplug.c > > @@ -1801,7 +1801,7 @@ int __ref remove_memory(int nid, u64 start, u64 size) > > int retry = 1; > > > > start_pfn = PFN_DOWN(start); > > - end_pfn = start_pfn + PFN_DOWN(size); > > + end_pfn = PFN_UP(start + size - 1); > > > > /* > > * When CONFIG_MEMCG is on, one memory block may be used by other > > That looks right, although these rounding/boundary things are always > hard. I wonder if `start' and `size' are ever not multiples of > PAGE_SIZE.. > > How did you discover this? Code inspection, or some runtime > malfunction? Please always include this info when fixing bugs. It was found in code inspection. For ACPI memory hot-delete, both start and size values are obtained from ACPI, and should always be page-aligned. So, this issue is not exposed at this point. That said, it should handle the boundary condition correctly since it might be called from other path in future. Yes, I will include such info when fixing bugs. Thanks, -Toshi -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>