- 07 Feb, 2008 40 commits
-
-
Paul Jackson authored
- Narrow the scope of callback_mutex in scan_for_empty_cpusets(). - Avoid rewriting the cpus, mems of cpusets except when it is likely that we'll be changing them. - Have remove_tasks_in_empty_cpuset() also check for empty mems. Signed-off-by:
Paul Jackson <pj@sgi.com> Acked-by:
Cliff Wickman <cpw@sgi.com> Cc: David Rientjes <rientjes@google.com> Cc: Paul Menage <menage@google.com> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Paul Jackson authored
Various minor formatting and comment tweaks to Cliff Wickman's [PATCH_3_of_3]_cpusets__update_cpumask_revision.patch I had had "iff", meaning "if and only if" in a comment. However, except for ancient mathematicians, the abbreviation "iff" was a tad too cryptic. Cliff changed it to "if", presumably figuring that the "iff" was a typo. However, it was the "only if" half of the conjunction that was most interesting. Reword to emphasis the "only if" aspect. The locking comment for remove_tasks_in_empty_cpuset() was wrong; it said callback_mutex had to be held on entry. The opposite is true. Several mentions of attach_task() in comments needed to be changed to cgroup_attach_task(). A comment about notify_on_release was no longer relevant, as the line of code it had commented, namely: set_bit(CS_RELEASED_RESOURCE, &parent->flags); is no longer present in that place in the cpuset.c code. Similarly a comment about notify_on_release before the scan_for_empty_cpusets() routine was no longer relevant. Removed extra parentheses and unnecessary return statement. Renamed attach_task() to cpuset_attach() in various comments. Removed comment about not needing memory migration, as it seems the migration is done anyway, via the cpuset_attach() callback from cgroup_attach_task(). Signed-off-by:
-
Paul Menage authored
Some of the comments in kernel/cpuset.c were stale following the transition to control groups; this patch updates them to more closely match reality. Signed-off-by:
Paul Menage <menage@google.com> Acked-by:
-
Cliff Wickman authored
Use the new function cgroup_scan_tasks() to step through all tasks in a cpuset. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by:
-
Cliff Wickman authored
This patch corrects a situation that occurs when one disables all the cpus in a cpuset. Currently, the disabled (cpu-less) cpuset inherits the cpus of its parent, which is incorrect because it may then overlap its cpu-exclusive sibling. Tasks of an empty cpuset should be moved to the cpuset which is the parent of their current cpuset. Or if the parent cpuset has no cpus, to its parent, etc. And the empty cpuset should be released (if it is flagged notify_on_release). Depends on the cgroup_scan_tasks() function (proposed by David Rientjes) to iterate through all tasks in the cpu-less cpuset. We are deliberately avoiding a walk of the tasklist. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by:
-
Cliff Wickman authored
Provide cgroup_scan_tasks(), which iterates through every task in a cgroup, calling a test function and a process function for each. And call the process function without holding the css_set_lock lock. The idea is David Rientjes', predicting that such a function will make it much easier in the future to extend things that require access to each task in a cgroup without holding the lock, [akpm@linux-foundation.org: cleanup] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by:
David Rientjes <rientjes@google.com> Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Documentation updates for memory controller. Signed-off-by:
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Reviewed-by:
Balbir Singh <balbir@linux.vnet.ibm.com> Signed-off-by:
-
Balbir Singh authored
Based on the discussion at http://lkml.org/lkml/2007/12/20/383, it was felt that control_type might not be a good thing to implement right away. We can add this flexibility at a later point when required. Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Now, lru is per-zone. Then, lru_lock can be (should be) per-zone, too. This patch implementes per-zone lru lock. lru_lock is placed into mem_cgroup_per_zone struct. lock can be accessed by mz = mem_cgroup_zoneinfo(mem_cgroup, node, zone); &mz->lru_lock or mz = page_cgroup_zoneinfo(page_cgroup); &mz->lru_lock Signed-off-by:
KAMEZAWA hiroyuki <kmaezawa.hiroyu@jp.fujitsu.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: David Rientjes <rientjes@google.com> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: Kirill Korotaev <dev@sw.ru> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Paul Menage <menage@google.com> Cc: Pavel Emelianov <xemul@openvz.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by:
-
KAMEZAWA Hiroyuki authored
This patch implements per-zone lru for memory cgroup. This patch makes use of mem_cgroup_per_zone struct for per zone lru. LRU can be accessed by mz = mem_cgroup_zoneinfo(mem_cgroup, node, zone); &mz->active_list &mz->inactive_list or mz = page_cgroup_zoneinfo(page_cgroup); &mz->active_list &mz->inactive_list Signed-off-by:
-
KAMEZAWA Hiroyuki authored
per-zone and reclaim enhancements for memory controller: modifies vmscan.c for isolate globa/cgroup lru activity When using memory controller, there are 2 levels of memory reclaim. 1. zone memory reclaim because of system/zone memory shortage. 2. memory cgroup memory reclaim because of hitting limit. These two can be distinguished by sc->mem_cgroup parameter. (scan_global_lru() macro) This patch tries to make memory cgroup reclaim routine avoid affecting system/zone memory reclaim. This patch inserts if (scan_global_lru()) and hook to memory_cgroup reclaim support functions. This patch can be a help for isolating system lru activity and group lru activity and shows what additional functions are necessary. * mem_cgroup_calc_mapped_ratio() ... calculate mapped ratio for cgroup. * mem_cgroup_reclaim_imbalance() ... calculate active/inactive balance in cgroup. * mem_cgroup_calc_reclaim_active() ... calculate the number of active pages to be scanned in this priority in mem_cgroup. * mem_cgroup_calc_reclaim_inactive() ... calculate the number of inactive pages to be scanned in this priority in mem_cgroup. * mem_cgroup_all_unreclaimable() .. checks cgroup's page is all unreclaimable or not. * mem_cgroup_get_reclaim_priority() ... * mem_cgroup_note_reclaim_priority() ... record reclaim priority (temporal) * mem_cgroup_remember_reclaim_priority() .... record reclaim priority as zone->prev_priority. This value is used for calc reclaim_mapped. [akpm@linux-foundation.org: fix unused var warning] Signed-off-by: -
KAMEZAWA Hiroyuki authored
per-zone and reclaim enhancements for memory controller: calculate the number of pages to be scanned per cgroup Define function for calculating the number of scan target on each Zone/LRU. Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Functions to remember reclaim priority per cgroup (as zone->prev_priority) [akpm@linux-foundation.org: build fixes] [akpm@linux-foundation.org: more build fixes] Signed-off-by:
-
KAMEZAWA Hiroyuki authored
per-zone and reclaim enhancements for memory controller: calculate active/inactive imbalance per cgroup Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Define function for calculating mapped_ratio in memory cgroup. Signed-off-by:
-
KAMEZAWA Hiroyuki authored
This patch adds per-zone status in memory cgroup. These values are often read (as per-zone value) by page reclaiming. In current design, per-zone stat is just a unsigned long value and not an atomic value because they are modified only under lru_lock. (So, atomic_ops is not necessary.) This patch adds ACTIVE and INACTIVE per-zone status values. For handling per-zone status, this patch adds struct mem_cgroup_per_zone { ... } and some helper functions. This will be useful to add per-zone objects in mem_cgroup. This patch turns memory controller's early_init to be 0 for calling kmalloc() in initialization. Acked-by: -
KAMEZAWA Hiroyuki authored
Add macro to get node_id and zone_id of page_cgroup. Will be used in per-zone-xxx patches and others. Signed-off-by:
-
KAMEZAWA Hiroyuki authored
This is used to detect which scan_control scans global lru or mem_cgroup lru. And compiled to be static value (1) when memory controller is not configured. This may make the meaning obvious. Acked-by:
-
KAMEZAWA Hiroyuki authored
Add pre_destroy handler for mem_cgroup and try to make mem_cgroup empty at rmdir(). Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Add a handler "pre_destroy" to cgroup_subsys. It is called before cgroup_rmdir() checks all subsys's refcnt. I think this is useful for subsys which have some extra refs even if there are no tasks in cgroup. By adding pre_destroy(), the kernel keeps the rule "destroy() against subsystem is called only when refcnt=0." and allows css ref to be used by other objects than tasks. Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Show accounted information of memory cgroup by memory.stat file [akpm@linux-foundation.org: coding-style fixes] [akpm@linux-foundation.org: fix printk warning] Signed-off-by:
YAMAMOTO Takashi <yamamoto@valinux.co.jp> Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Add statistics account infrastructure for memory controller. All account information is stored per-cpu and caller will not have to take lock or use atomic ops. This will be used by memory.stat file later. CACHE includes swapcache now. I'd like to divide it to PAGECACHE and SWAPCACHE later. This patch adds 3 functions for accounting. * __mem_cgroup_stat_add() ... for usual routine. * __mem_cgroup_stat_add_safe ... for calling under irq_disabled section. * mem_cgroup_read_stat() ... for reading stat value. * renamed PAGECACHE to CACHE (because it may include swapcache *now*) [akpm@linux-foundation.org: coding-style fixes] [akpm@linux-foundation.org: fix smp_processor_id-in-preemptible] [akpm@linux-foundation.org: uninline things] [akpm@linux-foundation.org: remove dead code] Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Remember page_cgroup is on active_list or not in page_cgroup->flags. Signed-off-by:
-
Hugh Dickins authored
The memcgroup regime relies upon a cgroup reclaiming pages from itself within add_to_page_cache: which may involve some waiting. Whereas shmem and tmpfs rely upon using add_to_page_cache while holding a spinlock: when it cannot wait. The consequence is that when a cgroup reaches its limit, shmem_getpage just hangs - unless there is outside memory pressure too, neither kswapd nor radix_tree_preload get it out of the retry loop. In most cases we can mem_cgroup_cache_charge the page waitably first, to attach the page_cgroup in advance, so add_to_page_cache will do no more than increment a count; then mem_cgroup_uncharge_page after (in both success and failure cases) to balance the books again. And where there used to be a congestion_wait for kswapd (recently made redundant by radix_tree_preload), use mem_cgroup_cache_charge with NULL page to go through a cycle of allocation and freeing, without accounting to any particular page, and without updating the statistics vector. This brings the cgroup below its limit so the next try usually succeeds. Signed-off-by:
Hugh Dickins <hugh@veritas.com> Cc: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Mel Gorman <mel@csn.ul.ie> Signed-off-by:
-
Hugh Dickins authored
Tidy up mem_cgroup_charge_common before extending it. Adjust some comments, but mainly clean up its loop: I've an aversion to loops full of continues, then a break or a goto at the bottom. And the is_atomic test should be on the __GFP_WAIT bit, not GFP_ATOMIC bits. Signed-off-by:
-
Balbir Singh authored
Hugh Dickins noticed that we were using rcu_dereference() without rcu_read_lock() in the cache charging routine. The patch below fixes this problem Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Add a flag to page_cgroup to remember "this page is charged as cache." cache here includes page caches and swap cache. This is useful for implementing precise accounting in memory cgroup. TODO: distinguish page-cache and swap-cache Signed-off-by:
-
KAMEZAWA Hiroyuki authored
This patch adds an interface "memory.force_empty". Any write to this file will drop all charges in this cgroup if there is no task under. %echo 1 > /....../memory.force_empty will drop all charges of memory cgroup if cgroup's tasks is empty. This is useful to invoke rmdir() against memory cgroup successfully. Tested and worked well on x86_64/fake-NUMA system. Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Because NODE_DATA(node)->node_zonelists[] is guaranteed to contain all necessary zones, it is not necessary to use for_each_online_node. And this for_each_online_node() makes reclaim routine start always from node 0. This is not good. This patch makes reclaim start from caller's node and just use usual (default) zonelist order. [akpm@linux-foundation.org: fix warning] Signed-off-by:
-
Hugh Dickins authored
If we're charging rss and we're charging cache, it seems obvious that we should be charging swapcache - as has been done. But in practice that doesn't work out so well: both swapin readahead and swapoff leave the majority of pages charged to the wrong cgroup (the cgroup that happened to read them in, rather than the cgroup to which they belong). (Which is why unuse_pte's GFP_KERNEL while holding pte lock never showed up as a problem: no allocation was ever done there, every page read being already charged to the cgroup which initiated the swapoff.) It all works rather better if we leave the charging to do_swap_page and unuse_pte, and do nothing for swapcache itself: revert mm/swap_state.c to what it was before the memory-controller patches. This also speeds up significantly a contained process working at its limit: because it no longer needs to keep waiting for swap writeback to complete. Is it unfair that swap pages become uncharged once they're unmapped, even though they're still clearly private to particular cgroups? For a short while, yes; but PageReclaim arranges for those pages to go to the end of the inactive list and be reclaimed soon if necessary. shmem/tmpfs pages are a distinct case: their charging also benefits from this change, but their second life on the lists as swapcache pages may prove more unfair - that I need to check next. Signed-off-by:
-
Hugh Dickins authored
mem_cgroup_charge_common shows a tendency to OOM without good reason, when a memhog goes well beyond its rss limit but with plenty of swap available. Seen on x86 but not on PowerPC; seen when the next patch omits swapcache from memcgroup, but we presume it can happen without. mem_cgroup_isolate_pages is not quite satisfying reclaim's criteria for OOM avoidance. Already it has to scan beyond the nr_to_scan limit when it finds a !LRU page or an active page when handling inactive or an inactive page when handling active. It needs to do exactly the same when it finds a page from the wrong zone (the x86 tests had two zones, the PowerPC tests had only one). Don't increment scan and then decrement it in these cases, just move the incrementation down. Fix recent off-by-one when checking against nr_to_scan. Cut out "Check if the meta page went away from under us", presumably left over from early debugging: no amount of such checks could save us if this list really were being updated without locking. This change does make the unlimited scan while holding two spinlocks even worse - bad for latency and bad for containment; but that's a separate issue which is better left to be fixed a little later. Signed-off-by:
-
KAMEZAWA Hiroyuki authored
This patch makes mem_cgroup_isolate_pages() to be - ignore !PageLRU pages. - fixes the bug that isolation makes no progress if page_zone(page) != zone page once find. (just increment scan in this case.) kswapd and memory migration removes a page from list when it handles a page for reclaiming/migration. Because __isolate_lru_page() doesn't moves page !PageLRU pages, it will be safe to avoid touching !PageLRU() page and its page_cgroup. Signed-off-by: -
KAMEZAWA Hiroyuki authored
While using memory control cgroup, page-migration under it works as following. == 1. uncharge all refs at try to unmap. 2. charge regs again remove_migration_ptes() == This is simple but has following problems. == The page is uncharged and charged back again if *mapped*. - This means that cgroup before migration can be different from one after migration - If page is not mapped but charged as page cache, charge is just ignored (because not mapped, it will not be uncharged before migration) This is memory leak. == This patch tries to keep memory cgroup at page migration by increasing one refcnt during it. 3 functions are added. mem_cgroup_prepare_migration() --- increase refcnt of page->page_cgroup mem_cgroup_end_migration() --- decrease refcnt of page->page_cgroup mem_cgroup_page_migration() --- copy page->page_cgroup from old page to new page. During migration - old page is under PG_locked. - new page is under PG_locked, too. - both old page and new page is not on LRU. These 3 facts guarantee that page_cgroup() migration has no race. Tested and worked well in x86_64/fake-NUMA box. Signed-off-by: -
KAMEZAWA Hiroyuki authored
This patch adds following functions. - clear_page_cgroup(page, pc) - page_cgroup_assign_new_page_group(page, pc) Mainly for cleanup. A manner "check page->cgroup again after lock_page_cgroup()" is implemented in straight way. A comment in mem_cgroup_uncharge() will be removed by force-empty patch Signed-off-by:
-
Rik van Riel authored
The current kswapd (and try_to_free_pages) code has an oddity where the code will wait on IO, even if there is no IO in flight. This problem is notable especially when the system scans through many unfreeable pages, causing unnecessary stalls in the VM. Additionally, tasks without __GFP_FS or __GFP_IO in the direct reclaim path will sleep if a significant number of pages are encountered that should be written out. This gives kswapd a chance to write out those pages, while the direct reclaim task sleeps. Signed-off-by:
Rik van Riel <riel@redhat.com> Signed-off-by:
-
David Rientjes authored
Adds a new sysctl, 'oom_dump_tasks', that enables the kernel to produce a dump of all system tasks (excluding kernel threads) when performing an OOM-killing. Information includes pid, uid, tgid, vm size, rss, cpu, oom_adj score, and name. This is helpful for determining why there was an OOM condition and which rogue task caused it. It is configurable so that large systems, such as those with several thousand tasks, do not incur a performance penalty associated with dumping data they may not desire. If an OOM was triggered as a result of a memory controller, the tasklist shall be filtered to exclude tasks that are not a member of the same cgroup. Cc: Andrea Arcangeli <andrea@suse.de> Cc: Christoph Lameter <clameter@sgi.com> Cc: Balbir Singh <balbir@linux.vnet.ibm.com> Signed-off-by:
-
David Rientjes authored
Creates a helper function to return non-zero if a task is a member of a memory controller: int task_in_mem_cgroup(const struct task_struct *task, const struct mem_cgroup *mem); When the OOM killer is constrained by the memory controller, the exclusion of tasks that are not a member of that controller was previously misplaced and appeared in the badness scoring function. It should be excluded during the tasklist scan in select_bad_process() instead. [akpm@linux-foundation.org: build fix] Cc: Christoph Lameter <clameter@sgi.com> Cc: Balbir Singh <balbir@linux.vnet.ibm.com> Signed-off-by:
-
Badari Pulavarty authored
Need to strip __GFP_HIGHMEM flag while passing to mem_container_cache_charge(). Signed-off-by:
Badari Pulavarty <pbadari@us.ibm.com> Signed-off-by:
-
Balbir Singh authored
Move mem_controller_cache_charge() above radix_tree_preload(). radix_tree_preload() disables preemption, even though the gfp_mask passed contains __GFP_WAIT, we cannot really do __GFP_WAIT allocations, thus we hit a BUG_ON() in kmem_cache_alloc(). This patch moves mem_controller_cache_charge() to above radix_tree_preload() for cache charging. Signed-off-by:
-
Hugh Dickins authored
This patch reinstates the "swapoff: scan ptes preemptibly" mod we started with: in due course it should be rendered down into the earlier patches, leaving us with a more straightforward mem_cgroup_charge mod to unuse_pte, allocating with GFP_KERNEL while holding no spinlock and no atomic kmap. Signed-off-by:
-
