Messages in this thread | | | Date | Fri, 15 May 2009 12:37:03 -0700 (PDT) | From | David Rientjes <> | Subject | Re: Misleading OOM messages |
| |
On Fri, 15 May 2009, Dave Hansen wrote:
> We get a wee bit of info out for the cgroups case at least: > > void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask) > { > ... > if (oom_kill_process(p, gfp_mask, 0, points, mem, > "Memory cgroup out of memory")) > goto retry; > > That can surely be improved, but it's a decent start. >
Cpusets are also cgroups and have their own oom handling logic (CONSTRAINT_CPUSET and the penalization of the badness score for not sharing memory with current's set of allowed nodes). In this case, we're interested in only the nodes set in cpuset_current_mems_allowed, for instance, and not the entire state of the machine for exclusive cpusets.
| |