| Message ID | 1604470210-124827-2-git-send-email-feng.tang@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | mm: fix OOMs for binding workloads to movable zone only node | expand |
On Wed 04-11-20 14:10:09, Feng Tang wrote: > In some OOM cases, if there is memory node binding(current->mems_allowed > is not NULL), system may only print the meminfo for these bound nodes, > while other nodes' info could still be important for debugging. > > For example on a platform with one normal node (has DMA/DMA32/NORMAL... > zones) and one node which only has movable zone (either for memory hotplug > case or a persistent memory node), some user will run docker while binding > memory to the movable node. many memory allocations originated from the > docker instance will fall back to the other node, and when a OOM happens, > meminfo for both nodes are needed. > > So extend the show_mem() to cover all memory nodes. I do not like this change. The reason why we print only relevant numa nodes is the size of the oom report. Also all other numa nodes are not really relevant to the allocation so there is no real reason to print their info. We used to do that in the past and decided that this is more than suboptimal. I do understand that this is a preliminary work for your later patch which tweaks the node binding and so more numa nodes are eligible but then I would propose to merge the two patches. > Signed-off-by: Feng Tang <feng.tang@intel.com> > --- > mm/oom_kill.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/mm/oom_kill.c b/mm/oom_kill.c > index 8b84661..601476cc 100644 > --- a/mm/oom_kill.c > +++ b/mm/oom_kill.c > @@ -462,7 +462,7 @@ static void dump_header(struct oom_control *oc, struct task_struct *p) > if (is_memcg_oom(oc)) > mem_cgroup_print_oom_meminfo(oc->memcg); > else { > - show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask); > + show_mem(SHOW_MEM_FILTER_NODES, &node_states[N_MEMORY]); > if (is_dump_unreclaim_slabs()) > dump_unreclaimable_slab(); > } > -- > 2.7.4
diff --git a/mm/oom_kill.c b/mm/oom_kill.c index 8b84661..601476cc 100644 --- a/mm/oom_kill.c +++ b/mm/oom_kill.c @@ -462,7 +462,7 @@ static void dump_header(struct oom_control *oc, struct task_struct *p) if (is_memcg_oom(oc)) mem_cgroup_print_oom_meminfo(oc->memcg); else { - show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask); + show_mem(SHOW_MEM_FILTER_NODES, &node_states[N_MEMORY]); if (is_dump_unreclaim_slabs()) dump_unreclaimable_slab(); }
In some OOM cases, if there is memory node binding(current->mems_allowed is not NULL), system may only print the meminfo for these bound nodes, while other nodes' info could still be important for debugging. For example on a platform with one normal node (has DMA/DMA32/NORMAL... zones) and one node which only has movable zone (either for memory hotplug case or a persistent memory node), some user will run docker while binding memory to the movable node. many memory allocations originated from the docker instance will fall back to the other node, and when a OOM happens, meminfo for both nodes are needed. So extend the show_mem() to cover all memory nodes. Signed-off-by: Feng Tang <feng.tang@intel.com> --- mm/oom_kill.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)