Re: [RFC][PATCH 1/2] memcg: oom notifier

From: Kirill A. Shutemov
Date: Mon Mar 08 2010 - 03:33:13 EST

Next message: KAMEZAWA Hiroyuki: "Re: [PATCH -mmotm 3/4] memcg: dirty pages accounting and limitinginfrastructure"
Previous message: Cyril Hrubis: "Re: bit errors on spitz"
In reply to: KAMEZAWA Hiroyuki: "[RFC][PATCH 1/2] memcg: oom notifier"
Next in thread: KAMEZAWA Hiroyuki: "Re: [RFC][PATCH 1/2] memcg: oom notifier"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Mon, Mar 8, 2010 at 9:25 AM, KAMEZAWA Hiroyuki
<kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote:
> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
>
> Considering containers or other resource management softwares in userland,
> event notification of OOM in memcg should be implemented.
> Now, memcg has "threshold" notifier which uses eventfd, we can make
> use of it for oom notification.
>
> This patch adds oom notification eventfd callback for memcg. The usage
> is very similar to threshold notifier, but control file is
> memory.oom_control and no arguments other than eventfd is required.
>
> Â Â Â Â% cgroup_event_notifier /cgroup/A/memory.oom_control dummy
> Â Â Â Â(About cgroup_event_notifier, see Documentation/cgroup/)

Nice idea!

But I don't think that sharing mem_cgroup_(un)register_event()
with thresholds is a good idea. There are too many
"if (type != _OOM_TYPE)". Probably, it's cleaner to create separate
register/unregister for oom events, since oom event is quite different
from threshold. We, also, don't need RCU for oom events. It's not
a critical path.

> TODO:
> Â- add a knob to disable oom-kill under a memcg.
> Â- add read/write function to oom_control
>
> Changelog: 20100304
> Â- renewed implemnation.
>
> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
> ---
> ÂDocumentation/cgroups/memory.txt | Â 20 ++++-
> Âmm/memcontrol.c Â Â Â Â Â Â Â Â Â| Â155 ++++++++++++++++++++++++++++-----------
> Â2 files changed, 131 insertions(+), 44 deletions(-)
>
> Index: mmotm-2.6.33-Mar5/mm/memcontrol.c
> ===================================================================
> --- mmotm-2.6.33-Mar5.orig/mm/memcontrol.c
> +++ mmotm-2.6.33-Mar5/mm/memcontrol.c
> @@ -159,6 +159,7 @@ struct mem_cgroup_threshold_ary {
> Â};
>
> Âstatic void mem_cgroup_threshold(struct mem_cgroup *mem);
> +static void mem_cgroup_oom_notify(struct mem_cgroup *mem);
>
> Â/*
> Â* The memory controller data structure. The memory controller controls both
> @@ -220,6 +221,9 @@ struct mem_cgroup {
> Â Â Â Â/* thresholds for mem+swap usage. RCU-protected */
> Â Â Â Âstruct mem_cgroup_threshold_ary *memsw_thresholds;
>
> + Â Â Â /* For oom notifier event fd */
> + Â Â Â struct mem_cgroup_threshold_ary *oom_notify;
> +
> Â Â Â Â/*
> Â Â Â Â * Should we move charges of a task when a task is moved into this
> Â Â Â Â * mem_cgroup ? And what type of charges should we move ?
> @@ -282,9 +286,12 @@ enum charge_type {
> Â/* for encoding cft->private value on file */
> Â#define _MEM Â Â Â Â Â Â Â Â Â (0)
> Â#define _MEMSWAP Â Â Â Â Â Â Â (1)
> +#define _OOM_TYPE Â Â Â Â Â Â Â(2)
> Â#define MEMFILE_PRIVATE(x, val) Â Â Â Â(((x) << 16) | (val))
> Â#define MEMFILE_TYPE(val) Â Â Â(((val) >> 16) & 0xffff)
> Â#define MEMFILE_ATTR(val) Â Â Â((val) & 0xffff)
> +/* Used for OOM nofiier */
> +#define OOM_CONTROL Â Â Â Â Â Â(0)
>
> Â/*
> Â* Reclaim flags for mem_cgroup_hierarchical_reclaim
> @@ -1313,9 +1320,10 @@ bool mem_cgroup_handle_oom(struct mem_cg
> Â Â Â Â Â Â Â Âprepare_to_wait(&memcg_oom_waitq, &wait, TASK_KILLABLE);
> Â Â Â Âmutex_unlock(&memcg_oom_mutex);
>
> - Â Â Â if (locked)
> + Â Â Â if (locked) {
> + Â Â Â Â Â Â Â mem_cgroup_oom_notify(mem);
> Â Â Â Â Â Â Â Âmem_cgroup_out_of_memory(mem, mask);
> - Â Â Â else {
> + Â Â Â } else {
> Â Â Â Â Â Â Â Âschedule();
> Â Â Â Â Â Â Â Âfinish_wait(&memcg_oom_waitq, &wait);
> Â Â Â Â}
> @@ -3363,33 +3371,65 @@ static int compare_thresholds(const void
> Â Â Â Âreturn _a->threshold - _b->threshold;
> Â}
>
> +static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem, void *data)
> +{
> + Â Â Â struct mem_cgroup_threshold_ary *x;
> + Â Â Â int i;
> +
> + Â Â Â rcu_read_lock();
> + Â Â Â x = rcu_dereference(mem->oom_notify);
> + Â Â Â for (i = 0; x && i < x->size; i++)
> + Â Â Â Â Â Â Â eventfd_signal(x->entries[i].eventfd, 1);
> + Â Â Â rcu_read_unlock();
> + Â Â Â return 0;
> +}
> +
> +static void mem_cgroup_oom_notify(struct mem_cgroup *mem)
> +{
> + Â Â Â mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_notify_cb);
> +}
> +
> Âstatic int mem_cgroup_register_event(struct cgroup *cgrp, struct cftype *cft,
> Â Â Â Â Â Â Â Âstruct eventfd_ctx *eventfd, const char *args)
> Â{
> Â Â Â Âstruct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
> Â Â Â Âstruct mem_cgroup_threshold_ary *thresholds, *thresholds_new;
> Â Â Â Âint type = MEMFILE_TYPE(cft->private);
> - Â Â Â u64 threshold, usage;
> + Â Â Â u64 threshold;
> + Â Â Â u64 usage = 0;
> Â Â Â Âint size;
> Â Â Â Âint i, ret;
>
> - Â Â Â ret = res_counter_memparse_write_strategy(args, &threshold);
> - Â Â Â if (ret)
> - Â Â Â Â Â Â Â return ret;
> + Â Â Â if (type != _OOM_TYPE) {
> + Â Â Â Â Â Â Â ret = res_counter_memparse_write_strategy(args, &threshold);
> + Â Â Â Â Â Â Â if (ret)
> + Â Â Â Â Â Â Â Â Â Â Â return ret;
> + Â Â Â } else if (mem_cgroup_is_root(memcg)) /* root cgroup ? */
> + Â Â Â Â Â Â Â return -ENOTSUPP;
>
> Â Â Â Âmutex_lock(&memcg->thresholds_lock);
> - Â Â Â if (type == _MEM)
> + Â Â Â /* For waiting OOM notify, "-1" is passed */
> +
> + Â Â Â switch (type) {
> + Â Â Â case _MEM:
> Â Â Â Â Â Â Â Âthresholds = memcg->thresholds;
> - Â Â Â else if (type == _MEMSWAP)
> + Â Â Â Â Â Â Â break;
> + Â Â Â case _MEMSWAP:
> Â Â Â Â Â Â Â Âthresholds = memcg->memsw_thresholds;
> - Â Â Â else
> + Â Â Â Â Â Â Â break;
> + Â Â Â case _OOM_TYPE:
> + Â Â Â Â Â Â Â thresholds = memcg->oom_notify;
> + Â Â Â Â Â Â Â break;
> + Â Â Â default:
> Â Â Â Â Â Â Â ÂBUG();
> + Â Â Â }
>
> - Â Â Â usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
> -
> - Â Â Â /* Check if a threshold crossed before adding a new one */
> - Â Â Â if (thresholds)
> - Â Â Â Â Â Â Â __mem_cgroup_threshold(memcg, type == _MEMSWAP);
> + Â Â Â if (type != _OOM_TYPE) {
> + Â Â Â Â Â Â Â usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
> + Â Â Â Â Â Â Â /* Check if a threshold crossed before adding a new one */
> + Â Â Â Â Â Â Â if (thresholds)
> + Â Â Â Â Â Â Â Â Â Â Â __mem_cgroup_threshold(memcg, type == _MEMSWAP);
> + Â Â Â }
>
> Â Â Â Âif (thresholds)
> Â Â Â Â Â Â Â Âsize = thresholds->size + 1;
> @@ -3416,27 +3456,34 @@ static int mem_cgroup_register_event(str
> Â Â Â Âthresholds_new->entries[size - 1].threshold = threshold;
>
> Â Â Â Â/* Sort thresholds. Registering of new threshold isn't time-critical */
> - Â Â Â sort(thresholds_new->entries, size,
> + Â Â Â if (type != _OOM_TYPE) {
> + Â Â Â Â Â Â Â sort(thresholds_new->entries, size,
> Â Â Â Â Â Â Â Â Â Â Â Âsizeof(struct mem_cgroup_threshold),
> Â Â Â Â Â Â Â Â Â Â Â Âcompare_thresholds, NULL);
> -
> - Â Â Â /* Find current threshold */
> - Â Â Â atomic_set(&thresholds_new->current_threshold, -1);
> - Â Â Â for (i = 0; i < size; i++) {
> - Â Â Â Â Â Â Â if (thresholds_new->entries[i].threshold < usage) {
> - Â Â Â Â Â Â Â Â Â Â Â /*
> - Â Â Â Â Â Â Â Â Â Â Â Â* thresholds_new->current_threshold will not be used
> - Â Â Â Â Â Â Â Â Â Â Â Â* until rcu_assign_pointer(), so it's safe to increment
> - Â Â Â Â Â Â Â Â Â Â Â Â* it here.
> - Â Â Â Â Â Â Â Â Â Â Â Â*/
> - Â Â Â Â Â Â Â Â Â Â Â atomic_inc(&thresholds_new->current_threshold);
> + Â Â Â Â Â Â Â /* Find current threshold */
> + Â Â Â Â Â Â Â atomic_set(&thresholds_new->current_threshold, -1);
> + Â Â Â Â Â Â Â for (i = 0; i < size; i++) {
> + Â Â Â Â Â Â Â Â Â Â Â if (thresholds_new->entries[i].threshold < usage) {
> + Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â /*
> + Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â* thresholds_new->current_threshold will not
> + Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â* be used until rcu_assign_pointer(), so it's
> + Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â* safe to increment it here.
> + Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â*/
> + Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â atomic_inc(&thresholds_new->current_threshold);
> + Â Â Â Â Â Â Â Â Â Â Â }
> Â Â Â Â Â Â Â Â}
> Â Â Â Â}
> -
> - Â Â Â if (type == _MEM)
> + Â Â Â switch (type) {
> + Â Â Â case _MEM:
> Â Â Â Â Â Â Â Ârcu_assign_pointer(memcg->thresholds, thresholds_new);
> - Â Â Â else
> + Â Â Â Â Â Â Â break;
> + Â Â Â case _MEMSWAP:
> Â Â Â Â Â Â Â Ârcu_assign_pointer(memcg->memsw_thresholds, thresholds_new);
> + Â Â Â Â Â Â Â break;
> + Â Â Â case _OOM_TYPE:
> + Â Â Â Â Â Â Â rcu_assign_pointer(memcg->oom_notify, thresholds_new);
> + Â Â Â Â Â Â Â break;
> + Â Â Â }
>
> Â Â Â Â/* To be sure that nobody uses thresholds before freeing it */
> Â Â Â Âsynchronize_rcu();
> @@ -3454,17 +3501,25 @@ static int mem_cgroup_unregister_event(s
> Â Â Â Âstruct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
> Â Â Â Âstruct mem_cgroup_threshold_ary *thresholds, *thresholds_new;
> Â Â Â Âint type = MEMFILE_TYPE(cft->private);
> - Â Â Â u64 usage;
> + Â Â Â u64 usage = 0;
> Â Â Â Âint size = 0;
> Â Â Â Âint i, j, ret;
>
> Â Â Â Âmutex_lock(&memcg->thresholds_lock);
> - Â Â Â if (type == _MEM)
> + Â Â Â /* check eventfd is for OOM check or not */
> + Â Â Â switch (type) {
> + Â Â Â case _MEM:
> Â Â Â Â Â Â Â Âthresholds = memcg->thresholds;
> - Â Â Â else if (type == _MEMSWAP)
> + Â Â Â Â Â Â Â break;
> + Â Â Â case _MEMSWAP:
> Â Â Â Â Â Â Â Âthresholds = memcg->memsw_thresholds;
> - Â Â Â else
> + Â Â Â Â Â Â Â break;
> + Â Â Â case _OOM_TYPE:
> + Â Â Â Â Â Â Â thresholds = memcg->oom_notify;
> + Â Â Â Â Â Â Â break;
> + Â Â Â default:
> Â Â Â Â Â Â Â ÂBUG();
> + Â Â Â }
>
> Â Â Â Â/*
> Â Â Â Â * Something went wrong if we trying to unregister a threshold
> @@ -3472,11 +3527,11 @@ static int mem_cgroup_unregister_event(s
> Â Â Â Â */
> Â Â Â ÂBUG_ON(!thresholds);
>
> - Â Â Â usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
> -
> - Â Â Â /* Check if a threshold crossed before removing */
> - Â Â Â __mem_cgroup_threshold(memcg, type == _MEMSWAP);
> -
> + Â Â Â if (type != _OOM_TYPE) {
> + Â Â Â Â Â Â Â usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
> + Â Â Â Â Â Â Â /* Check if a threshold crossed before removing */
> + Â Â Â Â Â Â Â __mem_cgroup_threshold(memcg, type == _MEMSWAP);
> + Â Â Â }
> Â Â Â Â/* Calculate new number of threshold */
> Â Â Â Âfor (i = 0; i < thresholds->size; i++) {
> Â Â Â Â Â Â Â Âif (thresholds->entries[i].eventfd != eventfd)
> @@ -3500,13 +3555,15 @@ static int mem_cgroup_unregister_event(s
> Â Â Â Âthresholds_new->size = size;
>
> Â Â Â Â/* Copy thresholds and find current threshold */
> - Â Â Â atomic_set(&thresholds_new->current_threshold, -1);
> + Â Â Â if (type != _OOM_TYPE)
> + Â Â Â Â Â Â Â atomic_set(&thresholds_new->current_threshold, -1);
> Â Â Â Âfor (i = 0, j = 0; i < thresholds->size; i++) {
> Â Â Â Â Â Â Â Âif (thresholds->entries[i].eventfd == eventfd)
> Â Â Â Â Â Â Â Â Â Â Â Âcontinue;
>
> Â Â Â Â Â Â Â Âthresholds_new->entries[j] = thresholds->entries[i];
> - Â Â Â Â Â Â Â if (thresholds_new->entries[j].threshold < usage) {
> + Â Â Â Â Â Â Â if (type != _OOM_TYPE &&
> + Â Â Â Â Â Â Â Â Â Â Â thresholds_new->entries[j].threshold < usage) {
> Â Â Â Â Â Â Â Â Â Â Â Â/*
> Â Â Â Â Â Â Â Â Â Â Â Â * thresholds_new->current_threshold will not be used
> Â Â Â Â Â Â Â Â Â Â Â Â * until rcu_assign_pointer(), so it's safe to increment
> @@ -3518,11 +3575,17 @@ static int mem_cgroup_unregister_event(s
> Â Â Â Â}
>
> Âassign:
> - Â Â Â if (type == _MEM)
> + Â Â Â switch (type) {
> + Â Â Â case _MEM:
> Â Â Â Â Â Â Â Ârcu_assign_pointer(memcg->thresholds, thresholds_new);
> - Â Â Â else
> + Â Â Â Â Â Â Â break;
> + Â Â Â case _MEMSWAP:
> Â Â Â Â Â Â Â Ârcu_assign_pointer(memcg->memsw_thresholds, thresholds_new);
> -
> + Â Â Â Â Â Â Â break;
> + Â Â Â case _OOM_TYPE:
> + Â Â Â Â Â Â Â rcu_assign_pointer(memcg->oom_notify, thresholds_new);
> + Â Â Â Â Â Â Â break;
> + Â Â Â }
> Â Â Â Â/* To be sure that nobody uses thresholds before freeing it */
> Â Â Â Âsynchronize_rcu();
>
> @@ -3588,6 +3651,12 @@ static struct cftype mem_cgroup_files[]
> Â Â Â Â Â Â Â Â.read_u64 = mem_cgroup_move_charge_read,
> Â Â Â Â Â Â Â Â.write_u64 = mem_cgroup_move_charge_write,
> Â Â Â Â},
> + Â Â Â {
> + Â Â Â Â Â Â Â .name = "oom_control",
> + Â Â Â Â Â Â Â .register_event = mem_cgroup_register_event,
> + Â Â Â Â Â Â Â .unregister_event = mem_cgroup_unregister_event,
> + Â Â Â Â Â Â Â .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
> + Â Â Â },
> Â};
>
> Â#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
> Index: mmotm-2.6.33-Mar5/Documentation/cgroups/memory.txt
> ===================================================================
> --- mmotm-2.6.33-Mar5.orig/Documentation/cgroups/memory.txt
> +++ mmotm-2.6.33-Mar5/Documentation/cgroups/memory.txt
> @@ -184,6 +184,9 @@ limits on the root cgroup.
>
> ÂNote2: When panic_on_oom is set to "2", the whole system will panic.
>
> +When oom event notifier is registered, event will be delivered.
> +(See oom_control section)
> +
> Â2. Locking
>
> ÂThe memory controller uses the following hierarchy
> @@ -486,7 +489,22 @@ threshold in any direction.
>
> ÂIt's applicable for root and non-root cgroup.
>
> -10. TODO
> +10. OOM Control
> +
> +Memory controler implements oom notifier using cgroup notification
> +API (See cgroups.txt). It allows to register multiple oom notification
> +delivery and gets notification when oom happens.
> +
> +To register a notifier, application need:
> + - create an eventfd using eventfd(2)
> + - open memory.oom_control file
> + - write string like "<event_fd> <memory.oom_control>" to cgroup.event_control
> +
> +Application will be notifier through eventfd when oom happens.
> +OOM notification doesn't work for root cgroup.
> +
> +
> +11. TODO
>
> Â1. Add support for accounting huge pages (as a separate controller)
> Â2. Make per-cgroup scanner reclaim not-shared pages first
>
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@xxxxxxxxxx ÂFor more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Next message: KAMEZAWA Hiroyuki: "Re: [PATCH -mmotm 3/4] memcg: dirty pages accounting and limitinginfrastructure"
Previous message: Cyril Hrubis: "Re: bit errors on spitz"
In reply to: KAMEZAWA Hiroyuki: "[RFC][PATCH 1/2] memcg: oom notifier"
Next in thread: KAMEZAWA Hiroyuki: "Re: [RFC][PATCH 1/2] memcg: oom notifier"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]