Show memcg information during OOM (v2)
Subsystems (3)
PATCH 1/1 Show memcg information during OOM (v2)
Description: Add RSS and swap to OOM output from memcg
From: Balbir Singh <[email protected]>
Changelog v2..v1:
1. Add more information about task's memcg and the memcg over it's limit
2. Print data in KB
3. Move the print routine outside task_lock()
4. Use rcu_read_lock() around cgroup_path, strictly speaking it is not required, but relying on the current memcg implementation is not a good idea.
This patch displays memcg values like failcnt, usage and limit when an OOM occurs due to memcg.
NOTE: In case the path exceeds 128 bytes, we omit printing the name of the cgroups. It is possible to circumvent this problem by using static arrays of PAGE_SIZE and we know that OOM is serialized when invoked from the memory controller. This did not seem like a good idea, but can be implemented if 128 bytes seems like a severe limitation.
Thanks go out to Johannes Weiner, Li Zefan, David Rientjes, Kamezawa Hiroyuki, Daisuke Nishimura and KOSAKI Motohiro for review.
Sample output
-------------
Task in /a/x killed as a result of limit of /a
memory: usage 1048576kB, limit 1048576kB, failcnt 4183 memory+swap: usage 1400964kB, limit 9007199254740991kB, failcnt 0
---
include/linux/memcontrol.h | 6 ++++ mm/memcontrol.c | 61 ++++++++++++++++++++++++++++++++++++++++++++ mm/oom_kill.c | 1 + 3 files changed, 68 insertions(+), 0 deletions(-)
include/linux/memcontrol.h +6 -0
@@ -104,6 +104,8 @@ struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, struct zone *zone);struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat_from_page (struct page *page); +extern void mem_cgroup_print_mem_info (struct mem_cgroup *memcg, +struct task_struct *p);#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP extern int do_swap_account;
@@ -270,6 +272,10 @@ mem_cgroup_get_reclaim_stat_from_page(struct page *page)return NULL ; } +void mem_cgroup_print_mem_info (struct mem_cgroup *memcg,struct task_struct *p) +{ +} +#endif /* CONFIG_CGROUP_MEM_CONT */ #endif /* _LINUX_MEMCONTROL_H */
mm/memcontrol.c +61 -0
73 lines not on this page · raw message
mm/oom_kill.c +1 -0
@@ -394,6 +394,7 @@ static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,cpuset_print_task_mems_allowed (current);task_unlock (current);dump_stack (); +mem_cgroup_print_mem_info (mem, current);show_mem ();if (sysctl_oom_dump_tasks)dump_tasks (mem);
Signature · 1 line
Balbir
Checkpatch caught an additional space, so here is the patch again
Description: Add RSS and swap to OOM output from memcg
From: Balbir Singh <[email protected]>
Changelog v2..v1:
1. Add more information about task's memcg and the memcg over it's limit
2. Print data in KB
3. Move the print routine outside task_lock()
4. Use rcu_read_lock() around cgroup_path, strictly speaking it is not required, but relying on the current memcg implementation is not a good idea.
This patch displays memcg values like failcnt, usage and limit when an OOM occurs due to memcg.
Thanks go out to Johannes Weiner, Li Zefan, David Rientjes, Kamezawa Hiroyuki, Daisuke Nishimura and KOSAKI Motohiro for review.
Sample output
-------------
Task in /a/x killed as a result of limit of /a
memory: usage 1048576kB, limit 1048576kB, failcnt 4183 memory+swap: usage 1400964kB, limit 9007199254740991kB, failcnt 0
---
include/linux/memcontrol.h | 6 ++++ mm/memcontrol.c | 61 ++++++++++++++++++++++++++++++++++++++++++++ mm/oom_kill.c | 1 + 3 files changed, 68 insertions(+), 0 deletions(-)
include/linux/memcontrol.h +6 -0
@@ -104,6 +104,8 @@ struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, struct zone *zone);struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat_from_page (struct page *page); +extern void mem_cgroup_print_mem_info (struct mem_cgroup *memcg, +struct task_struct *p);#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP extern int do_swap_account;
@@ -270,6 +272,10 @@ mem_cgroup_get_reclaim_stat_from_page(struct page *page)return NULL ; } +void mem_cgroup_print_mem_info (struct mem_cgroup *memcg,struct task_struct *p) +{ +} +#endif /* CONFIG_CGROUP_MEM_CONT */ #endif /* _LINUX_MEMCONTROL_H */
mm/memcontrol.c +61 -0
73 lines not on this page · raw message
mm/oom_kill.c +1 -0
@@ -394,6 +394,7 @@ static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,cpuset_print_task_mems_allowed (current);task_unlock (current);dump_stack (); +mem_cgroup_print_mem_info (mem, current);show_mem ();if (sysctl_oom_dump_tasks)dump_tasks (mem);
Signature · 7 lines
Balbir -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Quoted 18 lines · On Tue, 3 Feb 2009 12:57:01 +0530 Balbir Singh <[email protected]>…
Checkpatch caught an additional space, so here is the patch again
Description: Add RSS and swap to OOM output from memcg
From: Balbir Singh <[email protected]>
Changelog v2..v1:
1. Add more information about task's memcg and the memcg over it's limit
2. Print data in KB
3. Move the print routine outsidetask_lock()
4. Usercu_read_lock()aroundcgroup_path, strictly speaking it is not required, but relying on the current memcg implementation is not a good idea.
Quoted 7 lines
This patch displays memcg values like failcnt, usage and limit when an OOM occurs due to memcg.
Thanks go out to Johannes Weiner, Li Zefan, David Rientjes, Kamezawa Hiroyuki, Daisuke Nishimura and KOSAKI Motohiro for review.
IIUC, this oom_kill is serialized by memcg_tasklist mutex. Then, you don't have to allocate buffer on stack.
mm/memcontrol.c bool mem_cgroup_oom_called(struct task_struct *task) · line 837
Quoted 11 lines
mm/memcontrol.c +11
-0+void mem_cgroup_print_mem_info (struct mem_cgroup *memcg,struct task_struct *p) +{ +struct cgroup *task_cgrp; +struct cgroup *mem_cgrp; +/* +* Need a buffer on stack, can't rely on allocations. +*/ +char task_memcg_name[MEM_CGROUP_OOM_BUF_SIZE ]; +char memcg_name[MEM_CGROUP_OOM_BUF_SIZE ]; +int ret; +
making this as
static char task_memcg_name[PATH_MAX];
static char memcg_name[PATH_MAX];
is ok, I think. and the patch will be more simple.
Thanks,
-kame
Quoted 58 lines
mm/memcontrol.c +39
-0+if (!memcg) +return ; + + mem_cgrp = memcg->css.cgroup; + task_cgrp =mem_cgroup_from_task (p)->css.cgroup; + +rcu_read_lock (); + ret =cgroup_path (task_cgrp, task_memcg_name,MEM_CGROUP_OOM_BUF_SIZE ); +if (ret <0 ) { +/* +* Unfortunately, we are unable to convert to a useful name +* But we'll still print out the usage information +*/ +rcu_read_unlock (); + goto done; + } + ret =cgroup_path (mem_cgrp, memcg_name,MEM_CGROUP_OOM_BUF_SIZE ); +if (ret <0 ) { +rcu_read_unlock (); + goto done; + } + +rcu_read_unlock (); + +printk (KERN_INFO "Task in %s killed as a result of limit of %s\n" , + task_memcg_name, memcg_name); +done: + +printk (KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n" , +res_counter_read_u64 (&memcg->res,RES_USAGE ) >>10 , +res_counter_read_u64 (&memcg->res,RES_LIMIT ) >>10 , +res_counter_read_u64 (&memcg->res,RES_FAILCNT )); +printk (KERN_INFO "memory+swap: usage %llukB, limit %llukB, " +"failcnt %llu\n" , +res_counter_read_u64 (&memcg->memsw,RES_USAGE ) >>10 , +res_counter_read_u64 (&memcg->memsw,RES_LIMIT ) >>10 , +res_counter_read_u64 (&memcg->memsw,RES_FAILCNT )); +} +/* * Unlike exported interface, "oom" parameter is added. if oom==true, * oom-killer can be invoked. mm/oom_kill.c +1
-0@@ -394,6 +394,7 @@ static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,cpuset_print_task_mems_allowed (current);task_unlock (current);dump_stack (); +mem_cgroup_print_mem_info (mem, current);show_mem ();if (sysctl_oom_dump_tasks)dump_tasks (mem);Signature · 2 lines
-- Balbir
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Quoted 47 lines · KAMEZAWA Hiroyuki
Quoted 18 lines · On Tue, 3 Feb 2009 12:57:01 +0530 Balbir Singh <[email protected]>…
Checkpatch caught an additional space, so here is the patch again
Description: Add RSS and swap to OOM output from memcg
From: Balbir Singh <[email protected]>
Changelog v2..v1:
1. Add more information about task's memcg and the memcg over it's limit
2. Print data in KB
3. Move the print routine outsidetask_lock()
4. Usercu_read_lock()aroundcgroup_path, strictly speaking it is not required, but relying on the current memcg implementation is not a good idea.Quoted 7 lines
This patch displays memcg values like failcnt, usage and limit when an OOM occurs due to memcg.
Thanks go out to Johannes Weiner, Li Zefan, David Rientjes, Kamezawa Hiroyuki, Daisuke Nishimura and KOSAKI Motohiro for review.
IIUC, this
oom_killis serialized bymemcg_tasklistmutex. Then, you don't have to allocate buffer on stack.mm/memcontrol.c bool mem_cgroup_oom_called(struct task_struct *task) · line 837
Quoted 11 lines
mm/memcontrol.c +11
-0+void mem_cgroup_print_mem_info (struct mem_cgroup *memcg,struct task_struct *p) +{ +struct cgroup *task_cgrp; +struct cgroup *mem_cgrp; +/* +* Need a buffer on stack, can't rely on allocations. +*/ +char task_memcg_name[MEM_CGROUP_OOM_BUF_SIZE ]; +char memcg_name[MEM_CGROUP_OOM_BUF_SIZE ]; +int ret; +making this as
Quoted 5 lines
static char task_memcg_name[PATH_MAX];
static char memcg_name[PATH_MAX];is ok, I think. and the patch will be more simple.
I've mentioned it in the NOTE section as well, I wanted more opinions before going that route. I'll resend v3.
Signature · 7 lines
Balbir -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Quoted 47 lines · KAMEZAWA Hiroyuki
Quoted 18 lines · On Tue, 3 Feb 2009 12:57:01 +0530 Balbir Singh <[email protected]>…
Checkpatch caught an additional space, so here is the patch again
Description: Add RSS and swap to OOM output from memcg
From: Balbir Singh <[email protected]>
Changelog v2..v1:
1. Add more information about task's memcg and the memcg over it's limit
2. Print data in KB
3. Move the print routine outsidetask_lock()
4. Usercu_read_lock()aroundcgroup_path, strictly speaking it is not required, but relying on the current memcg implementation is not a good idea.Quoted 7 lines
This patch displays memcg values like failcnt, usage and limit when an OOM occurs due to memcg.
Thanks go out to Johannes Weiner, Li Zefan, David Rientjes, Kamezawa Hiroyuki, Daisuke Nishimura and KOSAKI Motohiro for review.
IIUC, this
oom_killis serialized bymemcg_tasklistmutex. Then, you don't have to allocate buffer on stack.mm/memcontrol.c bool mem_cgroup_oom_called(struct task_struct *task) · line 837
Quoted 11 lines
mm/memcontrol.c +11
-0+void mem_cgroup_print_mem_info (struct mem_cgroup *memcg,struct task_struct *p) +{ +struct cgroup *task_cgrp; +struct cgroup *mem_cgrp; +/* +* Need a buffer on stack, can't rely on allocations. +*/ +char task_memcg_name[MEM_CGROUP_OOM_BUF_SIZE ]; +char memcg_name[MEM_CGROUP_OOM_BUF_SIZE ]; +int ret; +making this as
Quoted 5 lines
static char task_memcg_name[PATH_MAX];
static char memcg_name[PATH_MAX];is ok, I think. and the patch will be more simple.
I am having second thoughts about this one. It introduces a standard overhead of 2 pages on x86*, while the first one will work for most cases and all the overhead is on stack, which disappears quickly. That is the reason I did not do it in the first place and put it as a NOTE.
Signature · 7 lines
Balbir -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Quoted 52 lines · On Tue, 3 Feb 2009 15:49:21 +0530 Balbir Singh <[email protected]>…
Quoted 50 lines · KAMEZAWA Hiroyuki
Quoted 18 lines · On Tue, 3 Feb 2009 12:57:01 +0530 Balbir Singh <[email protected]>…
Checkpatch caught an additional space, so here is the patch again
Description: Add RSS and swap to OOM output from memcg
From: Balbir Singh <[email protected]>
Changelog v2..v1:
1. Add more information about task's memcg and the memcg over it's limit
2. Print data in KB
3. Move the print routine outsidetask_lock()
4. Usercu_read_lock()aroundcgroup_path, strictly speaking it is not required, but relying on the current memcg implementation is not a good idea.Quoted 7 lines
This patch displays memcg values like failcnt, usage and limit when an OOM occurs due to memcg.
Thanks go out to Johannes Weiner, Li Zefan, David Rientjes, Kamezawa Hiroyuki, Daisuke Nishimura and KOSAKI Motohiro for review.
IIUC, this
oom_killis serialized bymemcg_tasklistmutex. Then, you don't have to allocate buffer on stack.mm/memcontrol.c bool mem_cgroup_oom_called(struct task_struct *task) · line 837
Quoted 11 lines
mm/memcontrol.c +11
-0+void mem_cgroup_print_mem_info (struct mem_cgroup *memcg,struct task_struct *p) +{ +struct cgroup *task_cgrp; +struct cgroup *mem_cgrp; +/* +* Need a buffer on stack, can't rely on allocations. +*/ +char task_memcg_name[MEM_CGROUP_OOM_BUF_SIZE ]; +char memcg_name[MEM_CGROUP_OOM_BUF_SIZE ]; +int ret; +making this as
static char task_memcg_name[PATH_MAX];
static char memcg_name[PATH_MAX];
Quoted 9 lines
Quoted 2 lines
is ok, I think. and the patch will be more simple.
I am having second thoughts about this one. It introduces a standard overhead of 2 pages on x86*, while the first one will work for most cases and all the overhead is on stack, which disappears quickly. That is the reason I did not do it in the first place and put it as a NOTE.
But *128* is tooooooo short ;)
And, your patch makes "OOM Message Format" unstable.
Quoted 1 line
From system administration view, it's unacceptable.
Not printing name at all is better than "printed out sometimes you lucky"
Thanks,
-Kame
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Quoted 64 lines · KAMEZAWA Hiroyuki
Quoted 52 lines · On Tue, 3 Feb 2009 15:49:21 +0530 Balbir Singh <[email protected]>…
Quoted 50 lines · KAMEZAWA Hiroyuki
Quoted 18 lines · On Tue, 3 Feb 2009 12:57:01 +0530 Balbir Singh <[email protected]>…
Checkpatch caught an additional space, so here is the patch again
Description: Add RSS and swap to OOM output from memcg
From: Balbir Singh <[email protected]>
Changelog v2..v1:
1. Add more information about task's memcg and the memcg over it's limit
2. Print data in KB
3. Move the print routine outsidetask_lock()
4. Usercu_read_lock()aroundcgroup_path, strictly speaking it is not required, but relying on the current memcg implementation is not a good idea.Quoted 7 lines
This patch displays memcg values like failcnt, usage and limit when an OOM occurs due to memcg.
Thanks go out to Johannes Weiner, Li Zefan, David Rientjes, Kamezawa Hiroyuki, Daisuke Nishimura and KOSAKI Motohiro for review.
IIUC, this
oom_killis serialized bymemcg_tasklistmutex. Then, you don't have to allocate buffer on stack.mm/memcontrol.c bool mem_cgroup_oom_called(struct task_struct *task) · line 837
Quoted 11 lines
mm/memcontrol.c +11
-0+void mem_cgroup_print_mem_info (struct mem_cgroup *memcg,struct task_struct *p) +{ +struct cgroup *task_cgrp; +struct cgroup *mem_cgrp; +/* +* Need a buffer on stack, can't rely on allocations. +*/ +char task_memcg_name[MEM_CGROUP_OOM_BUF_SIZE ]; +char memcg_name[MEM_CGROUP_OOM_BUF_SIZE ]; +int ret; +making this as
static char task_memcg_name[PATH_MAX];
static char memcg_name[PATH_MAX];Quoted 9 lines
Quoted 2 lines
is ok, I think. and the patch will be more simple.
I am having second thoughts about this one. It introduces a standard overhead of 2 pages on x86*, while the first one will work for most cases and all the overhead is on stack, which disappears quickly. That is the reason I did not do it in the first place and put it as a NOTE.
Quoted 5 lines
But *128* is tooooooo short ;)
And, your patch makes "OOM Message Format" unstable.Quoted 1 line
From system administration view, it's unacceptable.
Not printing name at all is better than "printed out sometimes you lucky"
OK, I have the code with PATH_MAX ready. I'll send that out.
Signature · 7 lines
Balbir -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"[email protected]"> [email protected] </a>