On Thu, Oct 30, 2014 at 09:32:14AM -0400, Sasha Levin wrote: > On 10/30/2014 04:27 AM, Michal Hocko wrote: > > On Wed 29-10-14 21:44:24, Sasha Levin wrote: > >> > Commit "mm: memcontrol: fix missed end-writeback page accounting" has changed > >> > the behaviour of mem_cgroup_begin_page_stat() to not always set the "locked" > >> > parameter. > >> > > >> > We should initialize it at the callers to prevent garbage being used in a > >> > later call to mem_cgroup_end_page_stat(). > > The contract is that if the returned memcg is non-NULL then the locked > > is always initialized. Nobody but mem_cgroup_end_page_stat should touch > > this variable and this function makes sure it uses it properly. Similar > > applies to flags which is initialized only if we really take the slow > > path (has a meaning only if locked == true). > > > > So this is not really needed. Was this triggered by a compiler warning? > > The problem is that you are attempting to read 'locked' when you call > mem_cgroup_end_page_stat(), so it gets used even before you enter the > function - and using uninitialized variables is undefined. We are not using that value anywhere if !memcg. What path are you referring to? > Yes, it's a compiler warning. Could you provide that please, including arch, and gcc version? Thanks -- 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>