I've recently summarized the cpuset/mempolicy issues in a LSF/MM proposal [1]
and the discussion itself [2]. I've been trying to rewrite the handling as
proposed, with the idea that changing semantics to make all mempolicies static
wrt cpuset updates (and discarding the relative and default modes) can be tried
on top, as there's a high risk of being rejected/reverted because somebody
might still care about the removed modes.
However I haven't yet figured out how to properly:
1) make mempolicies swappable instead of rebinding in place. I thought mbind()
already works that way and uses refcounting to avoid use-after-free of the old
policy by a parallel allocation, but turns out true refcounting is only done
for shared (shmem) mempolicies, and the actual protection for mbind() comes
from mmap_sem. Extending the refcounting means more overhead in allocator hot
path. Also swapping whole mempolicies means that we have to allocate the new
ones, which can fail, and reverting of the partially done work also means
allocating (note that mbind() doesn't care and will just leave part of the
range updated and part not updated when returning -ENOMEM...).
2) make cpuset's task->mems_allowed also swappable (after converting it from
nodemask to zonelist, which is the easy part) for mostly the same reasons.
The good news is that while trying to do the above, I've at least figured out
how to hopefully close the remaining premature OOM's, and do a buch of cleanups
on top, removing quite some of the code that was also supposed to prevent the
cpuset update races, but doesn't work anymore nowadays. This should fix the
most pressing concerns with this topic and give us a better baseline before
either proceeding with the original proposal, or pushing a change of semantics
that removes the problem 1) above. I'd be then fine with trying to change the
semantic first and rewrite later.
Patchset is based on next-20170411 and has been tested with the LTP cpuset01
stress test.
[1] https://lkml.kernel.org/r/[email protected]
[2] https://lwn.net/Articles/717797/
Vlastimil Babka (6):
mm, page_alloc: fix more premature OOM due to race with cpuset update
mm, mempolicy: stop adjusting current->il_next in
mpol_rebind_nodemask()
mm, page_alloc: pass preferred nid instead of zonelist to allocator
mm, mempolicy: simplify rebinding mempolicies when updating cpusets
mm, cpuset: always use seqlock when changing task's nodemask
mm, mempolicy: don't check cpuset seqlock where it doesn't matter
include/linux/gfp.h | 11 ++-
include/linux/mempolicy.h | 12 ++-
include/uapi/linux/mempolicy.h | 8 --
kernel/cgroup/cpuset.c | 33 ++-------
mm/hugetlb.c | 15 ++--
mm/memory_hotplug.c | 6 +-
mm/mempolicy.c | 165 +++++++++--------------------------------
mm/page_alloc.c | 61 ++++++++++-----
8 files changed, 109 insertions(+), 202 deletions(-)
--
2.12.2
The task->il_next variable remembers the last allocation node for task's
MPOL_INTERLEAVE policy. mpol_rebind_nodemask() updates interleave and
bind mempolicies due to changing cpuset mems. Currently it also tries to
make sure that current->il_next is valid within the updated nodemask. This is
bogus, because 1) we are updating potentially any task's mempolicy, not just
current, and 2) we might be updating per-vma mempolicy, not task one.
The interleave_nodes() function that uses il_next can cope fine with the value
not being within the currently allowed nodes, so this hasn't manifested as an
actual issue. Thus it also won't be an issue if we just remove this adjustment
completely.
Signed-off-by: Vlastimil Babka <[email protected]>
---
mm/mempolicy.c | 6 ------
1 file changed, 6 deletions(-)
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 37d0b334bfe9..efeec8d2bce5 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -349,12 +349,6 @@ static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes,
pol->v.nodes = tmp;
else
BUG();
-
- if (!node_isset(current->il_next, tmp)) {
- current->il_next = next_node_in(current->il_next, tmp);
- if (current->il_next >= MAX_NUMNODES)
- current->il_next = numa_node_id();
- }
}
static void mpol_rebind_preferred(struct mempolicy *pol,
--
2.12.2
Two wrappers of __alloc_pages_nodemask() are checking task->mems_allowed_seq
themselves to retry allocation that has raced with a cpuset update. This has
been shown to be ineffective in preventing premature OOM's which can happen in
__alloc_pages_slowpath() long before it returns back to the wrappers to detect
the race at that level. Previous patches have made __alloc_pages_slowpath()
more robust, so we can now simply remove the seqlock checking in the wrappers
to prevent further wrong impression that it can actually help.
Signed-off-by: Vlastimil Babka <[email protected]>
---
mm/mempolicy.c | 16 ----------------
1 file changed, 16 deletions(-)
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 72e5aeb1feeb..9a542b7a2189 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -1900,12 +1900,9 @@ alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
struct mempolicy *pol;
struct page *page;
int preferred_nid;
- unsigned int cpuset_mems_cookie;
nodemask_t *nmask;
-retry_cpuset:
pol = get_vma_policy(vma, addr);
- cpuset_mems_cookie = read_mems_allowed_begin();
if (pol->mode == MPOL_INTERLEAVE) {
unsigned nid;
@@ -1947,8 +1944,6 @@ alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
page = __alloc_pages_nodemask(gfp, order, preferred_nid, nmask);
mpol_cond_put(pol);
out:
- if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
- goto retry_cpuset;
return page;
}
@@ -1966,23 +1961,15 @@ alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
* Allocate a page from the kernel page pool. When not in
* interrupt context and apply the current process NUMA policy.
* Returns NULL when no page can be allocated.
- *
- * Don't call cpuset_update_task_memory_state() unless
- * 1) it's ok to take cpuset_sem (can WAIT), and
- * 2) allocating for current task (not interrupt).
*/
struct page *alloc_pages_current(gfp_t gfp, unsigned order)
{
struct mempolicy *pol = &default_policy;
struct page *page;
- unsigned int cpuset_mems_cookie;
if (!in_interrupt() && !(gfp & __GFP_THISNODE))
pol = get_task_policy(current);
-retry_cpuset:
- cpuset_mems_cookie = read_mems_allowed_begin();
-
/*
* No reference counting needed for current->mempolicy
* nor system default_policy
@@ -1994,9 +1981,6 @@ struct page *alloc_pages_current(gfp_t gfp, unsigned order)
policy_node(gfp, pol, numa_node_id()),
policy_nodemask(gfp, pol));
- if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
- goto retry_cpuset;
-
return page;
}
EXPORT_SYMBOL(alloc_pages_current);
--
2.12.2
When updating task's mems_allowed and rebinding its mempolicy due to cpuset's
mems being changed, we currently only take the seqlock for writing when either
the task has a mempolicy, or the new mems has no intersection with the old
mems. This should be enough to prevent a parallel allocation seeing no
available nodes, but the optimization is IMHO unnecessary (cpuset updates
should not be frequent), and we still potentially risk issues if the
intersection of new and old nodes has limited amount of free/reclaimable
memory. Let's just use the seqlock for all tasks.
Signed-off-by: Vlastimil Babka <[email protected]>
---
kernel/cgroup/cpuset.c | 29 +++++++----------------------
1 file changed, 7 insertions(+), 22 deletions(-)
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index b0159f8f8c89..e76d18daf085 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -1038,38 +1038,23 @@ static void cpuset_post_attach(void)
* @tsk: the task to change
* @newmems: new nodes that the task will be set
*
- * In order to avoid seeing no nodes if the old and new nodes are disjoint,
- * we structure updates as setting all new allowed nodes, then clearing newly
- * disallowed ones.
+ * We use the mems_allowed_seq seqlock to safely update both tsk->mems_allowed
+ * and rebind an eventual tasks' mempolicy. If the task is allocating in
+ * parallel, it might temporarily see an empty intersection, which results in
+ * a seqlock check and retry before OOM or allocation failure.
*/
static void cpuset_change_task_nodemask(struct task_struct *tsk,
nodemask_t *newmems)
{
- bool need_loop;
-
task_lock(tsk);
- /*
- * Determine if a loop is necessary if another thread is doing
- * read_mems_allowed_begin(). If at least one node remains unchanged and
- * tsk does not have a mempolicy, then an empty nodemask will not be
- * possible when mems_allowed is larger than a word.
- */
- need_loop = task_has_mempolicy(tsk) ||
- !nodes_intersects(*newmems, tsk->mems_allowed);
- if (need_loop) {
- local_irq_disable();
- write_seqcount_begin(&tsk->mems_allowed_seq);
- }
+ local_irq_disable();
+ write_seqcount_begin(&tsk->mems_allowed_seq);
- nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
mpol_rebind_task(tsk, newmems);
tsk->mems_allowed = *newmems;
- if (need_loop) {
- write_seqcount_end(&tsk->mems_allowed_seq);
- local_irq_enable();
- }
+ write_seqcount_end(&tsk->mems_allowed_seq);
task_unlock(tsk);
}
--
2.12.2
Commit c0ff7453bb5c ("cpuset,mm: fix no node to alloc memory when changing
cpuset's mems") has introduced a two-step protocol when rebinding task's
mempolicy due to cpuset update, in order to avoid a parallel allocation seeing
an empty effective nodemask and failing. Later, commit cc9a6c877661 ("cpuset:
mm: reduce large amounts of memory barrier related damage v3") introduced
a seqlock protection and removed the synchronization point between the two
update steps. At that point (or perhaps later), the two-step rebinding became
unnecessary. Currently it only makes sure that the update first adds new nodes
in step 1 and then removes nodes in step 2. Without memory barriers the effects
are questionable, and even then this cannot prevent a parallel zonelist
iteration checking the nodemask at each step to observe all nodes as unusable
for allocation. We now fully rely on the seqlock to prevent premature OOMs and
allocation failures.
We can thus remove the two-step update parts and simplify the code.
Signed-off-by: Vlastimil Babka <[email protected]>
---
include/linux/mempolicy.h | 6 +--
include/uapi/linux/mempolicy.h | 8 ----
kernel/cgroup/cpuset.c | 4 +-
mm/mempolicy.c | 102 ++++++++---------------------------------
4 files changed, 21 insertions(+), 99 deletions(-)
diff --git a/include/linux/mempolicy.h b/include/linux/mempolicy.h
index ecb6cbeede5a..3a58b4be1b0c 100644
--- a/include/linux/mempolicy.h
+++ b/include/linux/mempolicy.h
@@ -142,8 +142,7 @@ bool vma_policy_mof(struct vm_area_struct *vma);
extern void numa_default_policy(void);
extern void numa_policy_init(void);
-extern void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
- enum mpol_rebind_step step);
+extern void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new);
extern void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new);
extern int huge_node(struct vm_area_struct *vma,
@@ -260,8 +259,7 @@ static inline void numa_default_policy(void)
}
static inline void mpol_rebind_task(struct task_struct *tsk,
- const nodemask_t *new,
- enum mpol_rebind_step step)
+ const nodemask_t *new)
{
}
diff --git a/include/uapi/linux/mempolicy.h b/include/uapi/linux/mempolicy.h
index 9cd8b21dddbe..2a4d89508fec 100644
--- a/include/uapi/linux/mempolicy.h
+++ b/include/uapi/linux/mempolicy.h
@@ -24,13 +24,6 @@ enum {
MPOL_MAX, /* always last member of enum */
};
-enum mpol_rebind_step {
- MPOL_REBIND_ONCE, /* do rebind work at once(not by two step) */
- MPOL_REBIND_STEP1, /* first step(set all the newly nodes) */
- MPOL_REBIND_STEP2, /* second step(clean all the disallowed nodes)*/
- MPOL_REBIND_NSTEP,
-};
-
/* Flags for set_mempolicy */
#define MPOL_F_STATIC_NODES (1 << 15)
#define MPOL_F_RELATIVE_NODES (1 << 14)
@@ -65,7 +58,6 @@ enum mpol_rebind_step {
*/
#define MPOL_F_SHARED (1 << 0) /* identify shared policies */
#define MPOL_F_LOCAL (1 << 1) /* preferred local allocation */
-#define MPOL_F_REBINDING (1 << 2) /* identify policies in rebinding */
#define MPOL_F_MOF (1 << 3) /* this policy wants migrate on fault */
#define MPOL_F_MORON (1 << 4) /* Migrate On protnone Reference On Node */
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index f6501f4f6040..b0159f8f8c89 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -1063,9 +1063,7 @@ static void cpuset_change_task_nodemask(struct task_struct *tsk,
}
nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
- mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1);
-
- mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2);
+ mpol_rebind_task(tsk, newmems);
tsk->mems_allowed = *newmems;
if (need_loop) {
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 895d7a775f27..72e5aeb1feeb 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -146,22 +146,7 @@ struct mempolicy *get_task_policy(struct task_struct *p)
static const struct mempolicy_operations {
int (*create)(struct mempolicy *pol, const nodemask_t *nodes);
- /*
- * If read-side task has no lock to protect task->mempolicy, write-side
- * task will rebind the task->mempolicy by two step. The first step is
- * setting all the newly nodes, and the second step is cleaning all the
- * disallowed nodes. In this way, we can avoid finding no node to alloc
- * page.
- * If we have a lock to protect task->mempolicy in read-side, we do
- * rebind directly.
- *
- * step:
- * MPOL_REBIND_ONCE - do rebind work at once
- * MPOL_REBIND_STEP1 - set all the newly nodes
- * MPOL_REBIND_STEP2 - clean all the disallowed nodes
- */
- void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes,
- enum mpol_rebind_step step);
+ void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes);
} mpol_ops[MPOL_MAX];
static inline int mpol_store_user_nodemask(const struct mempolicy *pol)
@@ -304,19 +289,11 @@ void __mpol_put(struct mempolicy *p)
kmem_cache_free(policy_cache, p);
}
-static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes,
- enum mpol_rebind_step step)
+static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes)
{
}
-/*
- * step:
- * MPOL_REBIND_ONCE - do rebind work at once
- * MPOL_REBIND_STEP1 - set all the newly nodes
- * MPOL_REBIND_STEP2 - clean all the disallowed nodes
- */
-static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes,
- enum mpol_rebind_step step)
+static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes)
{
nodemask_t tmp;
@@ -325,35 +302,19 @@ static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes,
else if (pol->flags & MPOL_F_RELATIVE_NODES)
mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
else {
- /*
- * if step == 1, we use ->w.cpuset_mems_allowed to cache the
- * result
- */
- if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) {
- nodes_remap(tmp, pol->v.nodes,
- pol->w.cpuset_mems_allowed, *nodes);
- pol->w.cpuset_mems_allowed = step ? tmp : *nodes;
- } else if (step == MPOL_REBIND_STEP2) {
- tmp = pol->w.cpuset_mems_allowed;
- pol->w.cpuset_mems_allowed = *nodes;
- } else
- BUG();
+ nodes_remap(tmp, pol->v.nodes,pol->w.cpuset_mems_allowed,
+ *nodes);
+ pol->w.cpuset_mems_allowed = tmp;
}
if (nodes_empty(tmp))
tmp = *nodes;
- if (step == MPOL_REBIND_STEP1)
- nodes_or(pol->v.nodes, pol->v.nodes, tmp);
- else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2)
- pol->v.nodes = tmp;
- else
- BUG();
+ pol->v.nodes = tmp;
}
static void mpol_rebind_preferred(struct mempolicy *pol,
- const nodemask_t *nodes,
- enum mpol_rebind_step step)
+ const nodemask_t *nodes)
{
nodemask_t tmp;
@@ -379,42 +340,19 @@ static void mpol_rebind_preferred(struct mempolicy *pol,
/*
* mpol_rebind_policy - Migrate a policy to a different set of nodes
*
- * If read-side task has no lock to protect task->mempolicy, write-side
- * task will rebind the task->mempolicy by two step. The first step is
- * setting all the newly nodes, and the second step is cleaning all the
- * disallowed nodes. In this way, we can avoid finding no node to alloc
- * page.
- * If we have a lock to protect task->mempolicy in read-side, we do
- * rebind directly.
- *
- * step:
- * MPOL_REBIND_ONCE - do rebind work at once
- * MPOL_REBIND_STEP1 - set all the newly nodes
- * MPOL_REBIND_STEP2 - clean all the disallowed nodes
+ * Per-vma policies are protected by mmap_sem. Allocations using per-task
+ * policies are protected by task->mems_allowed_seq to prevent a premature
+ * OOM/allocation failure due to parallel nodemask modification.
*/
-static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask,
- enum mpol_rebind_step step)
+static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask)
{
if (!pol)
return;
- if (!mpol_store_user_nodemask(pol) && step == MPOL_REBIND_ONCE &&
+ if (!mpol_store_user_nodemask(pol) &&
nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
return;
- if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING))
- return;
-
- if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING))
- BUG();
-
- if (step == MPOL_REBIND_STEP1)
- pol->flags |= MPOL_F_REBINDING;
- else if (step == MPOL_REBIND_STEP2)
- pol->flags &= ~MPOL_F_REBINDING;
- else if (step >= MPOL_REBIND_NSTEP)
- BUG();
-
- mpol_ops[pol->mode].rebind(pol, newmask, step);
+ mpol_ops[pol->mode].rebind(pol, newmask);
}
/*
@@ -424,10 +362,9 @@ static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask,
* Called with task's alloc_lock held.
*/
-void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
- enum mpol_rebind_step step)
+void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new)
{
- mpol_rebind_policy(tsk->mempolicy, new, step);
+ mpol_rebind_policy(tsk->mempolicy, new);
}
/*
@@ -442,7 +379,7 @@ void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
down_write(&mm->mmap_sem);
for (vma = mm->mmap; vma; vma = vma->vm_next)
- mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE);
+ mpol_rebind_policy(vma->vm_policy, new);
up_write(&mm->mmap_sem);
}
@@ -2103,10 +2040,7 @@ struct mempolicy *__mpol_dup(struct mempolicy *old)
if (current_cpuset_is_being_rebound()) {
nodemask_t mems = cpuset_mems_allowed(current);
- if (new->flags & MPOL_F_REBINDING)
- mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2);
- else
- mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE);
+ mpol_rebind_policy(new, &mems);
}
atomic_set(&new->refcnt, 1);
return new;
--
2.12.2
The main allocator function __alloc_pages_nodemask() takes a zonelist pointer
as one of its parameters. All of its callers directly or indirectly obtain the
zonelist via node_zonelist() using a preferred node id and gfp_mask. We can
make the code a bit simpler by doing the zonelist lookup in
__alloc_pages_nodemask(), passing it a preferred node id instead (gfp_mask is
already another parameter).
There are some code size benefits thanks to removal of inlined node_zonelist():
bloat-o-meter add/remove: 2/2 grow/shrink: 4/36 up/down: 399/-1351 (-952)
This will also make things simpler if we proceed with converting cpusets to
zonelists.
Signed-off-by: Vlastimil Babka <[email protected]>
---
include/linux/gfp.h | 11 +++++------
include/linux/mempolicy.h | 6 +++---
mm/hugetlb.c | 15 +++++++++------
mm/memory_hotplug.c | 6 ++----
mm/mempolicy.c | 41 +++++++++++++++++++----------------------
mm/page_alloc.c | 10 +++++-----
6 files changed, 43 insertions(+), 46 deletions(-)
diff --git a/include/linux/gfp.h b/include/linux/gfp.h
index 2b1a44f5bdb6..666af3c39d00 100644
--- a/include/linux/gfp.h
+++ b/include/linux/gfp.h
@@ -432,14 +432,13 @@ static inline void arch_alloc_page(struct page *page, int order) { }
#endif
struct page *
-__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
- struct zonelist *zonelist, nodemask_t *nodemask);
+__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
+ nodemask_t *nodemask);
static inline struct page *
-__alloc_pages(gfp_t gfp_mask, unsigned int order,
- struct zonelist *zonelist)
+__alloc_pages(gfp_t gfp_mask, unsigned int order, int preferred_nid)
{
- return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
+ return __alloc_pages_nodemask(gfp_mask, order, preferred_nid, NULL);
}
/*
@@ -452,7 +451,7 @@ __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
VM_WARN_ON(!node_online(nid));
- return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
+ return __alloc_pages(gfp_mask, order, nid);
}
/*
diff --git a/include/linux/mempolicy.h b/include/linux/mempolicy.h
index 5f4d8281832b..ecb6cbeede5a 100644
--- a/include/linux/mempolicy.h
+++ b/include/linux/mempolicy.h
@@ -146,7 +146,7 @@ extern void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
enum mpol_rebind_step step);
extern void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new);
-extern struct zonelist *huge_zonelist(struct vm_area_struct *vma,
+extern int huge_node(struct vm_area_struct *vma,
unsigned long addr, gfp_t gfp_flags,
struct mempolicy **mpol, nodemask_t **nodemask);
extern bool init_nodemask_of_mempolicy(nodemask_t *mask);
@@ -269,13 +269,13 @@ static inline void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
{
}
-static inline struct zonelist *huge_zonelist(struct vm_area_struct *vma,
+static inline int huge_node(struct vm_area_struct *vma,
unsigned long addr, gfp_t gfp_flags,
struct mempolicy **mpol, nodemask_t **nodemask)
{
*mpol = NULL;
*nodemask = NULL;
- return node_zonelist(0, gfp_flags);
+ return 0;
}
static inline bool init_nodemask_of_mempolicy(nodemask_t *m)
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index e5828875f7bb..9f1f399bb913 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -904,6 +904,8 @@ static struct page *dequeue_huge_page_vma(struct hstate *h,
struct page *page = NULL;
struct mempolicy *mpol;
nodemask_t *nodemask;
+ gfp_t gfp_mask;
+ int nid;
struct zonelist *zonelist;
struct zone *zone;
struct zoneref *z;
@@ -924,12 +926,13 @@ static struct page *dequeue_huge_page_vma(struct hstate *h,
retry_cpuset:
cpuset_mems_cookie = read_mems_allowed_begin();
- zonelist = huge_zonelist(vma, address,
- htlb_alloc_mask(h), &mpol, &nodemask);
+ gfp_mask = htlb_alloc_mask(h);
+ nid = huge_node(vma, address, gfp_mask, &mpol, &nodemask);
+ zonelist = node_zonelist(nid, gfp_mask);
for_each_zone_zonelist_nodemask(zone, z, zonelist,
MAX_NR_ZONES - 1, nodemask) {
- if (cpuset_zone_allowed(zone, htlb_alloc_mask(h))) {
+ if (cpuset_zone_allowed(zone, gfp_mask)) {
page = dequeue_huge_page_node(h, zone_to_nid(zone));
if (page) {
if (avoid_reserve)
@@ -1545,13 +1548,13 @@ static struct page *__hugetlb_alloc_buddy_huge_page(struct hstate *h,
do {
struct page *page;
struct mempolicy *mpol;
- struct zonelist *zl;
+ int nid;
nodemask_t *nodemask;
cpuset_mems_cookie = read_mems_allowed_begin();
- zl = huge_zonelist(vma, addr, gfp, &mpol, &nodemask);
+ nid = huge_node(vma, addr, gfp, &mpol, &nodemask);
mpol_cond_put(mpol);
- page = __alloc_pages_nodemask(gfp, order, zl, nodemask);
+ page = __alloc_pages_nodemask(gfp, order, nid, nodemask);
if (page)
return page;
} while (read_mems_allowed_retry(cpuset_mems_cookie));
diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index 76d4745513ee..79787a55ebc1 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -1598,11 +1598,9 @@ static struct page *new_node_page(struct page *page, unsigned long private,
gfp_mask |= __GFP_HIGHMEM;
if (!nodes_empty(nmask))
- new_page = __alloc_pages_nodemask(gfp_mask, 0,
- node_zonelist(nid, gfp_mask), &nmask);
+ new_page = __alloc_pages_nodemask(gfp_mask, 0, nid, &nmask);
if (!new_page)
- new_page = __alloc_pages(gfp_mask, 0,
- node_zonelist(nid, gfp_mask));
+ new_page = __alloc_pages(gfp_mask, 0, nid);
return new_page;
}
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index efeec8d2bce5..895d7a775f27 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -1670,9 +1670,9 @@ static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
return NULL;
}
-/* Return a zonelist indicated by gfp for node representing a mempolicy */
-static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy,
- int nd)
+/* Return the node id preferred by the given mempolicy, or the given id */
+static int policy_node(gfp_t gfp, struct mempolicy *policy,
+ int nd)
{
if (policy->mode == MPOL_PREFERRED && !(policy->flags & MPOL_F_LOCAL))
nd = policy->v.preferred_node;
@@ -1685,7 +1685,7 @@ static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy,
WARN_ON_ONCE(policy->mode == MPOL_BIND && (gfp & __GFP_THISNODE));
}
- return node_zonelist(nd, gfp);
+ return nd;
}
/* Do dynamic interleaving for a process */
@@ -1793,38 +1793,37 @@ static inline unsigned interleave_nid(struct mempolicy *pol,
#ifdef CONFIG_HUGETLBFS
/*
- * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
+ * huge_node(@vma, @addr, @gfp_flags, @mpol)
* @vma: virtual memory area whose policy is sought
* @addr: address in @vma for shared policy lookup and interleave policy
* @gfp_flags: for requested zone
* @mpol: pointer to mempolicy pointer for reference counted mempolicy
* @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask
*
- * Returns a zonelist suitable for a huge page allocation and a pointer
+ * Returns a nid suitable for a huge page allocation and a pointer
* to the struct mempolicy for conditional unref after allocation.
* If the effective policy is 'BIND, returns a pointer to the mempolicy's
* @nodemask for filtering the zonelist.
*
* Must be protected by read_mems_allowed_begin()
*/
-struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr,
- gfp_t gfp_flags, struct mempolicy **mpol,
- nodemask_t **nodemask)
+int huge_node(struct vm_area_struct *vma, unsigned long addr, gfp_t gfp_flags,
+ struct mempolicy **mpol, nodemask_t **nodemask)
{
- struct zonelist *zl;
+ int nid;
*mpol = get_vma_policy(vma, addr);
*nodemask = NULL; /* assume !MPOL_BIND */
if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
- zl = node_zonelist(interleave_nid(*mpol, vma, addr,
- huge_page_shift(hstate_vma(vma))), gfp_flags);
+ nid = interleave_nid(*mpol, vma, addr,
+ huge_page_shift(hstate_vma(vma)));
} else {
- zl = policy_zonelist(gfp_flags, *mpol, numa_node_id());
+ nid = policy_node(gfp_flags, *mpol, numa_node_id());
if ((*mpol)->mode == MPOL_BIND)
*nodemask = &(*mpol)->v.nodes;
}
- return zl;
+ return nid;
}
/*
@@ -1926,12 +1925,10 @@ bool mempolicy_nodemask_intersects(struct task_struct *tsk,
static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
unsigned nid)
{
- struct zonelist *zl;
struct page *page;
- zl = node_zonelist(nid, gfp);
- page = __alloc_pages(gfp, order, zl);
- if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0]))
+ page = __alloc_pages(gfp, order, nid);
+ if (page && page_to_nid(page) == nid)
inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
return page;
}
@@ -1965,8 +1962,8 @@ alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
{
struct mempolicy *pol;
struct page *page;
+ int preferred_nid;
unsigned int cpuset_mems_cookie;
- struct zonelist *zl;
nodemask_t *nmask;
retry_cpuset:
@@ -2009,8 +2006,8 @@ alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
}
nmask = policy_nodemask(gfp, pol);
- zl = policy_zonelist(gfp, pol, node);
- page = __alloc_pages_nodemask(gfp, order, zl, nmask);
+ preferred_nid = policy_node(gfp, pol, node);
+ page = __alloc_pages_nodemask(gfp, order, preferred_nid, nmask);
mpol_cond_put(pol);
out:
if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
@@ -2057,7 +2054,7 @@ struct page *alloc_pages_current(gfp_t gfp, unsigned order)
page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
else
page = __alloc_pages_nodemask(gfp, order,
- policy_zonelist(gfp, pol, numa_node_id()),
+ policy_node(gfp, pol, numa_node_id()),
policy_nodemask(gfp, pol));
if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 502d82f0e004..028db51c953b 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -3969,12 +3969,12 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
}
static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
- struct zonelist *zonelist, nodemask_t *nodemask,
+ int preferred_nid, nodemask_t *nodemask,
struct alloc_context *ac, gfp_t *alloc_mask,
unsigned int *alloc_flags)
{
ac->high_zoneidx = gfp_zone(gfp_mask);
- ac->zonelist = zonelist;
+ ac->zonelist = node_zonelist(preferred_nid, gfp_mask);
ac->nodemask = nodemask;
ac->migratetype = gfpflags_to_migratetype(gfp_mask);
@@ -4019,8 +4019,8 @@ static inline void finalise_ac(gfp_t gfp_mask,
* This is the 'heart' of the zoned buddy allocator.
*/
struct page *
-__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
- struct zonelist *zonelist, nodemask_t *nodemask)
+__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
+ nodemask_t *nodemask)
{
struct page *page;
unsigned int alloc_flags = ALLOC_WMARK_LOW;
@@ -4028,7 +4028,7 @@ __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
struct alloc_context ac = { };
gfp_mask &= gfp_allowed_mask;
- if (!prepare_alloc_pages(gfp_mask, order, zonelist, nodemask, &ac, &alloc_mask, &alloc_flags))
+ if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags))
return NULL;
finalise_ac(gfp_mask, order, &ac);
--
2.12.2
Commit e47483bca2cc ("mm, page_alloc: fix premature OOM when racing with cpuset
mems update") has fixed known recent regressions found by LTP's cpuset01
testcase. I have however found that by modifying the testcase to use per-vma
mempolicies via bind(2) instead of per-task mempolicies via set_mempolicy(2),
the premature OOM still happens and the issue is much older.
The root of the problem is that the cpuset's mems_allowed and mempolicy's
nodemask can temporarily have no intersection, thus get_page_from_freelist()
cannot find any usable zone. The current semantic for empty intersection is to
ignore mempolicy's nodemask and honour cpuset restrictions. This is checked in
node_zonelist(), but the racy update can happen after we already passed the
check. Such races should be protected by the seqlock task->mems_allowed_seq,
but it doesn't work here, because 1) mpol_rebind_mm() does not happen under
seqlock for write, and doing so would lead to deadlock, as it takes mmap_sem
for write, while the allocation can have mmap_sem for read when it's taking the
seqlock for read. And 2) the seqlock cookie of callers of node_zonelist()
(alloc_pages_vma() and alloc_pages_current()) is different than the one of
__alloc_pages_slowpath(), so there's still a potential race window.
This patch fixes the issue by having __alloc_pages_slowpath() check for empty
intersection of cpuset and ac->nodemask before OOM or allocation failure. If
it's indeed empty, the nodemask is ignored and allocation retried, which mimics
node_zonelist(). This works fine, because almost all callers of
__alloc_pages_nodemask are obtaining the nodemask via node_zonelist(). The only
exception is new_node_page() from hotplug, where the potential violation of
nodemask isn't an issue, as there's already a fallback allocation attempt
without any nodemask. If there's a future caller that needs to have its specific
nodemask honoured over task's cpuset restrictions, we'll have to e.g. add a gfp
flag for that.
Signed-off-by: Vlastimil Babka <[email protected]>
---
mm/page_alloc.c | 51 ++++++++++++++++++++++++++++++++++++++-------------
1 file changed, 38 insertions(+), 13 deletions(-)
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 32b31d661c9c..502d82f0e004 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -3668,6 +3668,39 @@ should_reclaim_retry(gfp_t gfp_mask, unsigned order,
return false;
}
+static inline bool
+check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac)
+{
+ /*
+ * It's possible that cpuset's mems_allowed and the nodemask from
+ * mempolicy don't intersect. This should be normally dealt with by
+ * policy_nodemask(), but it's possible to race with cpuset update in
+ * such a way the check therein was true, and then it became false
+ * before we got our cpuset_mems_cookie here.
+ * This assumes that for all allocations, ac->nodemask can come only
+ * from MPOL_BIND mempolicy (whose documented semantics is to be ignored
+ * when it does not intersect with the cpuset restrictions) or the
+ * caller can deal with a violated nodemask.
+ */
+ if (cpusets_enabled() && ac->nodemask &&
+ !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) {
+ ac->nodemask = NULL;
+ return true;
+ }
+
+ /*
+ * When updating a task's mems_allowed or mempolicy nodemask, it is
+ * possible to race with parallel threads in such a way that our
+ * allocation can fail while the mask is being updated. If we are about
+ * to fail, check if the cpuset changed during allocation and if so,
+ * retry.
+ */
+ if (read_mems_allowed_retry(cpuset_mems_cookie))
+ return true;
+
+ return false;
+}
+
static inline struct page *
__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
struct alloc_context *ac)
@@ -3863,11 +3896,9 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
&compaction_retries))
goto retry;
- /*
- * It's possible we raced with cpuset update so the OOM would be
- * premature (see below the nopage: label for full explanation).
- */
- if (read_mems_allowed_retry(cpuset_mems_cookie))
+
+ /* Deal with possible cpuset update races before we start OOM killing */
+ if (check_retry_cpuset(cpuset_mems_cookie, ac))
goto retry_cpuset;
/* Reclaim has failed us, start killing things */
@@ -3886,14 +3917,8 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
}
nopage:
- /*
- * When updating a task's mems_allowed or mempolicy nodemask, it is
- * possible to race with parallel threads in such a way that our
- * allocation can fail while the mask is being updated. If we are about
- * to fail, check if the cpuset changed during allocation and if so,
- * retry.
- */
- if (read_mems_allowed_retry(cpuset_mems_cookie))
+ /* Deal with possible cpuset update races before we fail */
+ if (check_retry_cpuset(cpuset_mems_cookie, ac))
goto retry_cpuset;
/*
--
2.12.2
On Tue, 11 Apr 2017, Vlastimil Babka wrote:
> The root of the problem is that the cpuset's mems_allowed and mempolicy's
> nodemask can temporarily have no intersection, thus get_page_from_freelist()
> cannot find any usable zone. The current semantic for empty intersection is to
> ignore mempolicy's nodemask and honour cpuset restrictions. This is checked in
> node_zonelist(), but the racy update can happen after we already passed the
The fallback was only intended for a cpuset on which boundaries are not enforced
in critical conditions (softwall). A hardwall cpuset (CS_MEM_HARDWALL)
should fail the allocation.
> This patch fixes the issue by having __alloc_pages_slowpath() check for empty
> intersection of cpuset and ac->nodemask before OOM or allocation failure. If
> it's indeed empty, the nodemask is ignored and allocation retried, which mimics
> node_zonelist(). This works fine, because almost all callers of
Well that would need to be subject to the hardwall flag. Allocation needs
to fail for a hardwall cpuset.
On Tue, 11 Apr 2017, Vlastimil Babka wrote:
> The task->il_next variable remembers the last allocation node for task's
> MPOL_INTERLEAVE policy. mpol_rebind_nodemask() updates interleave and
> bind mempolicies due to changing cpuset mems. Currently it also tries to
> make sure that current->il_next is valid within the updated nodemask. This is
> bogus, because 1) we are updating potentially any task's mempolicy, not just
> current, and 2) we might be updating per-vma mempolicy, not task one.
>
> The interleave_nodes() function that uses il_next can cope fine with the value
> not being within the currently allowed nodes, so this hasn't manifested as an
> actual issue. Thus it also won't be an issue if we just remove this adjustment
> completely.
Well, interleave_nodes() will then potentially return a node outside of
the allowed memory policy when its called for the first time after
mpol_rebind_.. . But thenn it will find the next node within the
nodemask and work correctly for the next invocations.
But yea the race can probably be ignored. The idea was that the
application has a stable memory footprint during rebinding.
+CC linux-api
On 11.4.2017 19:24, Christoph Lameter wrote:
> On Tue, 11 Apr 2017, Vlastimil Babka wrote:
>
>> The root of the problem is that the cpuset's mems_allowed and mempolicy's
>> nodemask can temporarily have no intersection, thus get_page_from_freelist()
>> cannot find any usable zone. The current semantic for empty intersection is to
>> ignore mempolicy's nodemask and honour cpuset restrictions. This is checked in
>> node_zonelist(), but the racy update can happen after we already passed the
>
> The fallback was only intended for a cpuset on which boundaries are not enforced
> in critical conditions (softwall). A hardwall cpuset (CS_MEM_HARDWALL)
> should fail the allocation.
Hmm just to clarify - I'm talking about ignoring the *mempolicy's* nodemask on
the basis of cpuset having higher priority, while you seem to be talking about
ignoring a (softwall) cpuset nodemask, right? man set_mempolicy says "... if
required nodemask contains no nodes that are allowed by the process's current
cpuset context, the memory policy reverts to local allocation" which does come
down to ignoring mempolicy's nodemask.
>> This patch fixes the issue by having __alloc_pages_slowpath() check for empty
>> intersection of cpuset and ac->nodemask before OOM or allocation failure. If
>> it's indeed empty, the nodemask is ignored and allocation retried, which mimics
>> node_zonelist(). This works fine, because almost all callers of
>
> Well that would need to be subject to the hardwall flag. Allocation needs
> to fail for a hardwall cpuset.
They still do, if no hardwall cpuset node can satisfy the allocation with
mempolicy ignored.
On 11.4.2017 19:32, Christoph Lameter wrote:
> On Tue, 11 Apr 2017, Vlastimil Babka wrote:
>
>> The task->il_next variable remembers the last allocation node for task's
>> MPOL_INTERLEAVE policy. mpol_rebind_nodemask() updates interleave and
>> bind mempolicies due to changing cpuset mems. Currently it also tries to
>> make sure that current->il_next is valid within the updated nodemask. This is
>> bogus, because 1) we are updating potentially any task's mempolicy, not just
>> current, and 2) we might be updating per-vma mempolicy, not task one.
>>
>> The interleave_nodes() function that uses il_next can cope fine with the value
>> not being within the currently allowed nodes, so this hasn't manifested as an
>> actual issue. Thus it also won't be an issue if we just remove this adjustment
>> completely.
>
> Well, interleave_nodes() will then potentially return a node outside of
> the allowed memory policy when its called for the first time after
> mpol_rebind_.. . But thenn it will find the next node within the
> nodemask and work correctly for the next invocations.
Hmm, you're right. But that could be easily fixed if il_next became il_prev, so
we would return the result of next_node_in(il_prev) and also store it as the new
il_prev, right? I somehow assumed it already worked that way.
> But yea the race can probably be ignored. The idea was that the
> application has a stable memory footprint during rebinding.
On April 11, 2017 10:06 PM Vlastimil Babka wrote:
>
> static void cpuset_change_task_nodemask(struct task_struct *tsk,
> nodemask_t *newmems)
> {
> - bool need_loop;
> -
> task_lock(tsk);
> - /*
> - * Determine if a loop is necessary if another thread is doing
> - * read_mems_allowed_begin(). If at least one node remains unchanged and
> - * tsk does not have a mempolicy, then an empty nodemask will not be
> - * possible when mems_allowed is larger than a word.
> - */
> - need_loop = task_has_mempolicy(tsk) ||
> - !nodes_intersects(*newmems, tsk->mems_allowed);
>
> - if (need_loop) {
> - local_irq_disable();
> - write_seqcount_begin(&tsk->mems_allowed_seq);
> - }
> + local_irq_disable();
> + write_seqcount_begin(&tsk->mems_allowed_seq);
>
> - nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
> mpol_rebind_task(tsk, newmems);
> tsk->mems_allowed = *newmems;
>
> - if (need_loop) {
> - write_seqcount_end(&tsk->mems_allowed_seq);
> - local_irq_enable();
> - }
> + write_seqcount_end(&tsk->mems_allowed_seq);
>
Doubt if we'd listen irq again.
> task_unlock(tsk);
> }
> --
> 2.12.2
On 04/12/2017 10:10 AM, Hillf Danton wrote:
> On April 11, 2017 10:06 PM Vlastimil Babka wrote:
>>
>> static void cpuset_change_task_nodemask(struct task_struct *tsk,
>> nodemask_t *newmems)
>> {
>> - bool need_loop;
>> -
>> task_lock(tsk);
>> - /*
>> - * Determine if a loop is necessary if another thread is doing
>> - * read_mems_allowed_begin(). If at least one node remains unchanged and
>> - * tsk does not have a mempolicy, then an empty nodemask will not be
>> - * possible when mems_allowed is larger than a word.
>> - */
>> - need_loop = task_has_mempolicy(tsk) ||
>> - !nodes_intersects(*newmems, tsk->mems_allowed);
>>
>> - if (need_loop) {
>> - local_irq_disable();
>> - write_seqcount_begin(&tsk->mems_allowed_seq);
>> - }
>> + local_irq_disable();
>> + write_seqcount_begin(&tsk->mems_allowed_seq);
>>
>> - nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
>> mpol_rebind_task(tsk, newmems);
>> tsk->mems_allowed = *newmems;
>>
>> - if (need_loop) {
>> - write_seqcount_end(&tsk->mems_allowed_seq);
>> - local_irq_enable();
>> - }
>> + write_seqcount_end(&tsk->mems_allowed_seq);
>>
> Doubt if we'd listen irq again.
Ugh, thanks for catching this. Looks like my testing config didn't have
lockup detectors enabled.
>> task_unlock(tsk);
>> }
>> --
>> 2.12.2
>
On 04/11/2017 09:03 PM, Vlastimil Babka wrote:
> On 11.4.2017 19:32, Christoph Lameter wrote:
>> On Tue, 11 Apr 2017, Vlastimil Babka wrote:
>>
>>> The task->il_next variable remembers the last allocation node for task's
>>> MPOL_INTERLEAVE policy. mpol_rebind_nodemask() updates interleave and
>>> bind mempolicies due to changing cpuset mems. Currently it also tries to
>>> make sure that current->il_next is valid within the updated nodemask. This is
>>> bogus, because 1) we are updating potentially any task's mempolicy, not just
>>> current, and 2) we might be updating per-vma mempolicy, not task one.
>>>
>>> The interleave_nodes() function that uses il_next can cope fine with the value
>>> not being within the currently allowed nodes, so this hasn't manifested as an
>>> actual issue. Thus it also won't be an issue if we just remove this adjustment
>>> completely.
>>
>> Well, interleave_nodes() will then potentially return a node outside of
>> the allowed memory policy when its called for the first time after
>> mpol_rebind_.. . But thenn it will find the next node within the
>> nodemask and work correctly for the next invocations.
>
> Hmm, you're right. But that could be easily fixed if il_next became il_prev, so
> we would return the result of next_node_in(il_prev) and also store it as the new
> il_prev, right? I somehow assumed it already worked that way.
Like this?
----8<----
commit 0ec64a0b8e614ea655328d0fb539447c407ba7c1
Author: Vlastimil Babka <[email protected]>
Date: Mon Apr 3 13:11:32 2017 +0200
mm, mempolicy: stop adjusting current->il_next in mpol_rebind_nodemask()
The task->il_next variable stores the next allocation node id for task's
MPOL_INTERLEAVE policy. mpol_rebind_nodemask() updates interleave and
bind mempolicies due to changing cpuset mems. Currently it also tries to
make sure that current->il_next is valid within the updated nodemask. This is
bogus, because 1) we are updating potentially any task's mempolicy, not just
current, and 2) we might be updating a per-vma mempolicy, not task one.
The interleave_nodes() function that uses il_next can cope fine with the value
not being within the currently allowed nodes, so this hasn't manifested as an
actual issue.
We can remove the need for updating il_next completely by changing it to
il_prev and store the node id of the previous interleave allocation instead of
the next id. Then interleave_nodes() can calculate the next id using the
current nodemask and also store it as il_prev, except when querying the next
node via do_get_mempolicy().
Signed-off-by: Vlastimil Babka <[email protected]>
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 050d7113924a..9aca0db1e588 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -886,7 +886,7 @@ struct task_struct {
#ifdef CONFIG_NUMA
/* Protected by alloc_lock: */
struct mempolicy *mempolicy;
- short il_next;
+ short il_prev;
short pref_node_fork;
#endif
#ifdef CONFIG_NUMA_BALANCING
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 37d0b334bfe9..25f9bde58521 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -349,12 +349,6 @@ static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes,
pol->v.nodes = tmp;
else
BUG();
-
- if (!node_isset(current->il_next, tmp)) {
- current->il_next = next_node_in(current->il_next, tmp);
- if (current->il_next >= MAX_NUMNODES)
- current->il_next = numa_node_id();
- }
}
static void mpol_rebind_preferred(struct mempolicy *pol,
@@ -812,9 +806,8 @@ static long do_set_mempolicy(unsigned short mode, unsigned short flags,
}
old = current->mempolicy;
current->mempolicy = new;
- if (new && new->mode == MPOL_INTERLEAVE &&
- nodes_weight(new->v.nodes))
- current->il_next = first_node(new->v.nodes);
+ if (new && new->mode == MPOL_INTERLEAVE)
+ current->il_prev = MAX_NUMNODES-1;
task_unlock(current);
mpol_put(old);
ret = 0;
@@ -863,6 +856,18 @@ static int lookup_node(unsigned long addr)
return err;
}
+/* Do dynamic interleaving for a process */
+static unsigned interleave_nodes(struct mempolicy *policy, bool update_prev)
+{
+ unsigned next;
+ struct task_struct *me = current;
+
+ next = next_node_in(me->il_prev, policy->v.nodes);
+ if (next < MAX_NUMNODES && update_prev)
+ me->il_prev = next;
+ return next;
+}
+
/* Retrieve NUMA policy */
static long do_get_mempolicy(int *policy, nodemask_t *nmask,
unsigned long addr, unsigned long flags)
@@ -916,7 +921,7 @@ static long do_get_mempolicy(int *policy, nodemask_t *nmask,
*policy = err;
} else if (pol == current->mempolicy &&
pol->mode == MPOL_INTERLEAVE) {
- *policy = current->il_next;
+ *policy = interleave_nodes(current->mempolicy, false);
} else {
err = -EINVAL;
goto out;
@@ -1694,19 +1699,6 @@ static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy,
return node_zonelist(nd, gfp);
}
-/* Do dynamic interleaving for a process */
-static unsigned interleave_nodes(struct mempolicy *policy)
-{
- unsigned nid, next;
- struct task_struct *me = current;
-
- nid = me->il_next;
- next = next_node_in(nid, policy->v.nodes);
- if (next < MAX_NUMNODES)
- me->il_next = next;
- return nid;
-}
-
/*
* Depending on the memory policy provide a node from which to allocate the
* next slab entry.
@@ -1731,7 +1723,7 @@ unsigned int mempolicy_slab_node(void)
return policy->v.preferred_node;
case MPOL_INTERLEAVE:
- return interleave_nodes(policy);
+ return interleave_nodes(policy, true);
case MPOL_BIND: {
struct zoneref *z;
@@ -1794,7 +1786,7 @@ static inline unsigned interleave_nid(struct mempolicy *pol,
off += (addr - vma->vm_start) >> shift;
return offset_il_node(pol, vma, off);
} else
- return interleave_nodes(pol);
+ return interleave_nodes(pol, true);
}
#ifdef CONFIG_HUGETLBFS
@@ -2060,7 +2052,8 @@ struct page *alloc_pages_current(gfp_t gfp, unsigned order)
* nor system default_policy
*/
if (pol->mode == MPOL_INTERLEAVE)
- page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
+ page = alloc_page_interleave(gfp, order,
+ interleave_nodes(pol, true));
else
page = __alloc_pages_nodemask(gfp, order,
policy_zonelist(gfp, pol, numa_node_id()),
On Wed, 12 Apr 2017, Vlastimil Babka wrote:
> >> Well, interleave_nodes() will then potentially return a node outside of
> >> the allowed memory policy when its called for the first time after
> >> mpol_rebind_.. . But thenn it will find the next node within the
> >> nodemask and work correctly for the next invocations.
> >
> > Hmm, you're right. But that could be easily fixed if il_next became il_prev, so
> > we would return the result of next_node_in(il_prev) and also store it as the new
> > il_prev, right? I somehow assumed it already worked that way.
Yup that makes sense and I thought about that when I saw the problem too.
> @@ -863,6 +856,18 @@ static int lookup_node(unsigned long addr)
> return err;
> }
>
> +/* Do dynamic interleaving for a process */
> +static unsigned interleave_nodes(struct mempolicy *policy, bool update_prev)
Why do you need an additional flag? Would it not be better to always
update and switch the update_prev=false case to simply use
next_node_in()?
> +{
> + unsigned next;
> + struct task_struct *me = current;
> +
> + next = next_node_in(me->il_prev, policy->v.nodes);
> + if (next < MAX_NUMNODES && update_prev)
> + me->il_prev = next;
> + return next;
> +}
> +
> /* Retrieve NUMA policy */
> static long do_get_mempolicy(int *policy, nodemask_t *nmask,
> unsigned long addr, unsigned long flags)
> @@ -916,7 +921,7 @@ static long do_get_mempolicy(int *policy, nodemask_t *nmask,
> *policy = err;
> } else if (pol == current->mempolicy &&
> pol->mode == MPOL_INTERLEAVE) {
> - *policy = current->il_next;
> + *policy = interleave_nodes(current->mempolicy, false);
Here
On 12.4.2017 23:16, Christoph Lameter wrote:
> On Wed, 12 Apr 2017, Vlastimil Babka wrote:
>
>>>> Well, interleave_nodes() will then potentially return a node outside of
>>>> the allowed memory policy when its called for the first time after
>>>> mpol_rebind_.. . But thenn it will find the next node within the
>>>> nodemask and work correctly for the next invocations.
>>>
>>> Hmm, you're right. But that could be easily fixed if il_next became il_prev, so
>>> we would return the result of next_node_in(il_prev) and also store it as the new
>>> il_prev, right? I somehow assumed it already worked that way.
>
> Yup that makes sense and I thought about that when I saw the problem too.
>
>> @@ -863,6 +856,18 @@ static int lookup_node(unsigned long addr)
>> return err;
>> }
>>
>> +/* Do dynamic interleaving for a process */
>> +static unsigned interleave_nodes(struct mempolicy *policy, bool update_prev)
>
> Why do you need an additional flag? Would it not be better to always
> update and switch the update_prev=false case to simply use
> next_node_in()?
Looked to me as better wrapping, but probably overengineered, ok. Will change
for v2.
On Tue, 11 Apr 2017, Vlastimil Babka wrote:
> > The fallback was only intended for a cpuset on which boundaries are not enforced
> > in critical conditions (softwall). A hardwall cpuset (CS_MEM_HARDWALL)
> > should fail the allocation.
>
> Hmm just to clarify - I'm talking about ignoring the *mempolicy's* nodemask on
> the basis of cpuset having higher priority, while you seem to be talking about
> ignoring a (softwall) cpuset nodemask, right? man set_mempolicy says "... if
> required nodemask contains no nodes that are allowed by the process's current
> cpuset context, the memory policy reverts to local allocation" which does come
> down to ignoring mempolicy's nodemask.
I am talking of allocating outside of the current allowed nodes
(determined by mempolicy -- MPOL_BIND is the only concern as far as I can
tell -- as well as the current cpuset). One can violate the cpuset if its not
a hardwall but the MPOL_MBIND node restriction cannot be violated.
Those allocations are also not allowed if the allocation was for a user
space page even if this is a softwall cpuset.
> >> This patch fixes the issue by having __alloc_pages_slowpath() check for empty
> >> intersection of cpuset and ac->nodemask before OOM or allocation failure. If
> >> it's indeed empty, the nodemask is ignored and allocation retried, which mimics
> >> node_zonelist(). This works fine, because almost all callers of
> >
> > Well that would need to be subject to the hardwall flag. Allocation needs
> > to fail for a hardwall cpuset.
>
> They still do, if no hardwall cpuset node can satisfy the allocation with
> mempolicy ignored.
If the memory policy is MPOL_MBIND then allocations outside of the given
nodes should fail. They can violate the cpuset boundaries only if they are
kernel allocations and we are not in a hardwall cpuset.
That was at least my understand when working on this code years ago.
On 04/11/2017 07:36 PM, Vlastimil Babka wrote:
> Commit e47483bca2cc ("mm, page_alloc: fix premature OOM when racing with cpuset
> mems update") has fixed known recent regressions found by LTP's cpuset01
> testcase. I have however found that by modifying the testcase to use per-vma
> mempolicies via bind(2) instead of per-task mempolicies via set_mempolicy(2),
> the premature OOM still happens and the issue is much older.
Meanwhile while we are discussing this RFC, will it be better to WARN
out these situations where we dont have node in the intersection,
hence no usable zone during allocation. That might actually give
a hint to the user before a premature OOM/allocation failure comes.
>
> The root of the problem is that the cpuset's mems_allowed and mempolicy's
> nodemask can temporarily have no intersection, thus get_page_from_freelist()
> cannot find any usable zone. The current semantic for empty intersection is to
> ignore mempolicy's nodemask and honour cpuset restrictions. This is checked in
> node_zonelist(), but the racy update can happen after we already passed the
> check. Such races should be protected by the seqlock task->mems_allowed_seq,
> but it doesn't work here, because 1) mpol_rebind_mm() does not happen under
> seqlock for write, and doing so would lead to deadlock, as it takes mmap_sem
> for write, while the allocation can have mmap_sem for read when it's taking the
> seqlock for read. And 2) the seqlock cookie of callers of node_zonelist()
> (alloc_pages_vma() and alloc_pages_current()) is different than the one of
> __alloc_pages_slowpath(), so there's still a potential race window.
>
> This patch fixes the issue by having __alloc_pages_slowpath() check for empty
> intersection of cpuset and ac->nodemask before OOM or allocation failure. If
> it's indeed empty, the nodemask is ignored and allocation retried, which mimics
> node_zonelist(). This works fine, because almost all callers of
> __alloc_pages_nodemask are obtaining the nodemask via node_zonelist(). The only
> exception is new_node_page() from hotplug, where the potential violation of
> nodemask isn't an issue, as there's already a fallback allocation attempt
> without any nodemask. If there's a future caller that needs to have its specific
> nodemask honoured over task's cpuset restrictions, we'll have to e.g. add a gfp
> flag for that.
Did you really mean node_zonelist() in both the instances above. Because
that function just picks up either FALLBACK_ZONELIST or NOFALLBACK_ZONELIST
depending upon the passed GFP flags in the allocation request and does not
deal with ignoring the passed nodemask.
On 04/13/2017 07:42 AM, Anshuman Khandual wrote:
> On 04/11/2017 07:36 PM, Vlastimil Babka wrote:
>> Commit e47483bca2cc ("mm, page_alloc: fix premature OOM when racing with cpuset
>> mems update") has fixed known recent regressions found by LTP's cpuset01
>> testcase. I have however found that by modifying the testcase to use per-vma
>> mempolicies via bind(2) instead of per-task mempolicies via set_mempolicy(2),
>> the premature OOM still happens and the issue is much older.
>
> Meanwhile while we are discussing this RFC, will it be better to WARN
> out these situations where we dont have node in the intersection,
> hence no usable zone during allocation. That might actually give
> a hint to the user before a premature OOM/allocation failure comes.
Well, the bug is very old and nobody reported it so far, AFAIK. So it's
not that urgent.
>>
>> The root of the problem is that the cpuset's mems_allowed and mempolicy's
>> nodemask can temporarily have no intersection, thus get_page_from_freelist()
>> cannot find any usable zone. The current semantic for empty intersection is to
>> ignore mempolicy's nodemask and honour cpuset restrictions. This is checked in
>> node_zonelist(), but the racy update can happen after we already passed the
>> check. Such races should be protected by the seqlock task->mems_allowed_seq,
>> but it doesn't work here, because 1) mpol_rebind_mm() does not happen under
>> seqlock for write, and doing so would lead to deadlock, as it takes mmap_sem
>> for write, while the allocation can have mmap_sem for read when it's taking the
>> seqlock for read. And 2) the seqlock cookie of callers of node_zonelist()
>> (alloc_pages_vma() and alloc_pages_current()) is different than the one of
>> __alloc_pages_slowpath(), so there's still a potential race window.
>>
>> This patch fixes the issue by having __alloc_pages_slowpath() check for empty
>> intersection of cpuset and ac->nodemask before OOM or allocation failure. If
>> it's indeed empty, the nodemask is ignored and allocation retried, which mimics
>> node_zonelist(). This works fine, because almost all callers of
>> __alloc_pages_nodemask are obtaining the nodemask via node_zonelist(). The only
>> exception is new_node_page() from hotplug, where the potential violation of
>> nodemask isn't an issue, as there's already a fallback allocation attempt
>> without any nodemask. If there's a future caller that needs to have its specific
>> nodemask honoured over task's cpuset restrictions, we'll have to e.g. add a gfp
>> flag for that.
>
> Did you really mean node_zonelist() in both the instances above. Because
> that function just picks up either FALLBACK_ZONELIST or NOFALLBACK_ZONELIST
> depending upon the passed GFP flags in the allocation request and does not
> deal with ignoring the passed nodemask.
Oops, I meant policy_zonelist(), thanks for noticing.
On 04/13/2017 08:06 AM, Vlastimil Babka wrote:
>> Did you really mean node_zonelist() in both the instances above. Because
>> that function just picks up either FALLBACK_ZONELIST or NOFALLBACK_ZONELIST
>> depending upon the passed GFP flags in the allocation request and does not
>> deal with ignoring the passed nodemask.
>
> Oops, I meant policy_zonelist(), thanks for noticing.
Nah, policy_nodemask()... I need coffee.
On 04/12/2017 11:25 PM, Christoph Lameter wrote:
> On Tue, 11 Apr 2017, Vlastimil Babka wrote:
>
>>> The fallback was only intended for a cpuset on which boundaries are not enforced
>>> in critical conditions (softwall). A hardwall cpuset (CS_MEM_HARDWALL)
>>> should fail the allocation.
>>
>> Hmm just to clarify - I'm talking about ignoring the *mempolicy's* nodemask on
>> the basis of cpuset having higher priority, while you seem to be talking about
>> ignoring a (softwall) cpuset nodemask, right? man set_mempolicy says "... if
>> required nodemask contains no nodes that are allowed by the process's current
>> cpuset context, the memory policy reverts to local allocation" which does come
>> down to ignoring mempolicy's nodemask.
>
> I am talking of allocating outside of the current allowed nodes
> (determined by mempolicy -- MPOL_BIND is the only concern as far as I can
> tell -- as well as the current cpuset). One can violate the cpuset if its not
> a hardwall but the MPOL_MBIND node restriction cannot be violated.
>
> Those allocations are also not allowed if the allocation was for a user
> space page even if this is a softwall cpuset.
>
>>>> This patch fixes the issue by having __alloc_pages_slowpath() check for empty
>>>> intersection of cpuset and ac->nodemask before OOM or allocation failure. If
>>>> it's indeed empty, the nodemask is ignored and allocation retried, which mimics
>>>> node_zonelist(). This works fine, because almost all callers of
>>>
>>> Well that would need to be subject to the hardwall flag. Allocation needs
>>> to fail for a hardwall cpuset.
>>
>> They still do, if no hardwall cpuset node can satisfy the allocation with
>> mempolicy ignored.
>
> If the memory policy is MPOL_MBIND then allocations outside of the given
> nodes should fail. They can violate the cpuset boundaries only if they are
> kernel allocations and we are not in a hardwall cpuset.
>
> That was at least my understand when working on this code years ago.
Hmm, I see policy_nodemask() (I wrongly mentioned node_zonelist()
before) ignores BIND mempolicy nodemask when it doesn't overlap with
cpuset allowed nodes since initial git commit 1da177e4c3f4 (back then it
was zonelist_policy()). But AFAIU this couldn't actually happen (outside
of races), because 1) one is not allowed to create such effectively
empty BIND mempolicy in the first place and 2) an existing mempolicy is
rebound on cpuset changes to maintain the overlap.
The point 2) does not apply to MPOL_F_STATIC_NODES mempolicies
introduced in 2008 by DavidR, but it's documented in
Documentation/vm/numa_memory_policy.txt and manpages that when they
don't overlap with cpuset allowed nodes, the default mempolicy is used
instead.
I doubt we can change that now, because that can break existing
programs. It also makes some sense at least to me, because a task can
control its own mempolicy (for performance reasons), but cpuset changes
are admin decisions that the task cannot even anticipate. I think it's
better to continue working with suboptimal performance than start
failing allocations?
>
> --
> 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>
>
On Thu, 13 Apr 2017, Vlastimil Babka wrote:
>
> I doubt we can change that now, because that can break existing
> programs. It also makes some sense at least to me, because a task can
> control its own mempolicy (for performance reasons), but cpuset changes
> are admin decisions that the task cannot even anticipate. I think it's
> better to continue working with suboptimal performance than start
> failing allocations?
If the expected semantics (hardwall) are that allocations should fail then
lets be consistent and do so.
Adding more and more exceptions gets this convoluted mess into an even
worse shape. Adding the static binding of nodes was already a screwball
if used within a cpuset because now one has to anticipate how a user would
move the nodes of a cpuset and how the static bindings would work in such
a context.
The admin basically needs to know how the application has used memory
policies if one still wants to move the applications within a cpuset with
the fixed bindings.
Maybe the best way to handle this is to give up on cpuset migration of
live applications? After all this can be done with a script in the same
way as the kernel is doing:
1. Extend the cpuset to include the new nodes.
2. Loop over the processes and use the migrate_pages() to move the apps
one by one.
3. Remove the nodes no longer to be used.
Then forget about translating memory policies. If an application that is
supposed to run in a cpuset and supposed to be moveable has fixed bindings
then the application should be aware of that and be equipped with
some logic to rebind its memory on its own.
Such an application typically already has such logic and executes a
binding after discovering its numa node configuration on startup. It would
have to be modified to redo that action when it gets some sort of a signal
from the script telling it that the node config would be changed.
Having this logic in the application instead of the kernel avoids all the
kernel messes that we keep on trying to deal with and IMHO is much
cleaner.
On 04/14/2017 10:37 PM, Christoph Lameter wrote:
> On Thu, 13 Apr 2017, Vlastimil Babka wrote:
>
>>
>> I doubt we can change that now, because that can break existing
>> programs. It also makes some sense at least to me, because a task can
>> control its own mempolicy (for performance reasons), but cpuset changes
>> are admin decisions that the task cannot even anticipate. I think it's
>> better to continue working with suboptimal performance than start
>> failing allocations?
>
> If the expected semantics (hardwall) are that allocations should fail then
> lets be consistent and do so.
It's not "expected" right now. The documented semantics is that (static,
as the others are rebound) mempolicy is ignored when it's not compatible
with cpuset. I'm just reusing the same existing semantic for race
situations. We can discuss whether we can change the semantics now, but
I don't think it should block this fix.
> Adding more and more exceptions gets this convoluted mess into an even
> worse shape.
Again, it's not a new exception semantics-wise, but I agree that the
code of __alloc_pages_slowpath() is even more subtle. But I don't see
any other easy fix.
> Adding the static binding of nodes was already a screwball
> if used within a cpuset because now one has to anticipate how a user would
> move the nodes of a cpuset and how the static bindings would work in such
> a context.
On the other hand, static mempolicy is the only one that does not need
rebinding, and removing the other modes would allow much simpler
implementation. I thought the outcome of LSF/MM session was that we
should try to go that way.
> The admin basically needs to know how the application has used memory
> policies if one still wants to move the applications within a cpuset with
> the fixed bindings.
>
> Maybe the best way to handle this is to give up on cpuset migration of
> live applications? After all this can be done with a script in the same
> way as the kernel is doing:
>
> 1. Extend the cpuset to include the new nodes.
>
> 2. Loop over the processes and use the migrate_pages() to move the apps
> one by one.
>
> 3. Remove the nodes no longer to be used.
>
> Then forget about translating memory policies. If an application that is
> supposed to run in a cpuset and supposed to be moveable has fixed bindings
> then the application should be aware of that and be equipped with
> some logic to rebind its memory on its own.
>
> Such an application typically already has such logic and executes a
> binding after discovering its numa node configuration on startup. It would
> have to be modified to redo that action when it gets some sort of a signal
> from the script telling it that the node config would be changed.
>
> Having this logic in the application instead of the kernel avoids all the
> kernel messes that we keep on trying to deal with and IMHO is much
> cleaner.
That would be much simpler for us indeed. But we still IMHO can't
abruptly start denying page fault allocations for existing applications
that don't have the necessary awareness.
> --
> 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>
>
On Wed, 26 Apr 2017, Vlastimil Babka wrote:
> > Such an application typically already has such logic and executes a
> > binding after discovering its numa node configuration on startup. It would
> > have to be modified to redo that action when it gets some sort of a signal
> > from the script telling it that the node config would be changed.
> >
> > Having this logic in the application instead of the kernel avoids all the
> > kernel messes that we keep on trying to deal with and IMHO is much
> > cleaner.
>
> That would be much simpler for us indeed. But we still IMHO can't
> abruptly start denying page fault allocations for existing applications
> that don't have the necessary awareness.
We certainly can do that. The failure of the page faults are due to the
admin trying to move an application that is not aware of this and is using
mempols. That could be an error. Trying to move an application that
contains both absolute and relative node numbers is definitely something
that is potentiall so screwed up that the kernel should not muck around
with such an app.
Also user space can determine if the application is using memory policies
and can then take appropriate measures (message to the sysadmin to eval
tge situation f.e.) or mess aroud with the processes memory policies on
its own.
So this is certainly a way out of this mess.