The patch titled
Subject: mm/page_alloc: replace local_lock with normal spinlock
has been added to the -mm mm-unstable branch. Its filename is
mm-page_alloc-replace-local_lock-with-normal-spinlock.patch
This patch will shortly appear at
https://git.kernel.org/pub/scm/linux/kernel/git/akpm/25-new.git/tree/patches/mm-page_alloc-replace-local_lock-with-normal-spinlock.patch
This patch will later appear in the mm-unstable branch at
git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm
Before you just go and hit "reply", please:
a) Consider who else should be cc'ed
b) Prefer to cc a suitable mailing list as well
c) Ideally: find the original patch on the mailing list and do a
reply-to-all to that, adding suitable additional cc's
*** Remember to use Documentation/process/submit-checklist.rst when testing your code ***
The -mm tree is included into linux-next via the mm-everything
branch at git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm
and is updated there every 2-3 working days
------------------------------------------------------
From: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx>
Subject: mm/page_alloc: replace local_lock with normal spinlock
Date: Mon, 13 Jun 2022 13:56:22 +0100
struct per_cpu_pages is no longer strictly local as PCP lists can be
drained remotely using a lock for protection. While the use of local_lock
works, it goes against the intent of local_lock which is for "pure CPU
local concurrency control mechanisms and not suited for inter-CPU
concurrency control" (Documentation/locking/locktypes.rst)
local_lock protects against migration between when the percpu pointer is
accessed and the pcp->lock acquired. The lock acquisition is a preemption
point so in the worst case, a task could migrate to another NUMA node and
accidentally allocate remote memory. The main requirement is to pin the
task to a CPU that is suitable for PREEMPT_RT and !PREEMPT_RT.
Replace local_lock with helpers that pin a task to a CPU, lookup the
per-cpu structure and acquire the embedded lock. It's similar to
local_lock without breaking the intent behind the API. It is not a
complete API as only the parts needed for PCP-alloc are implemented but in
theory, the generic helpers could be promoted to a general API if there
was demand for an embedded lock within a per-cpu struct with a guarantee
that the per-cpu structure locked matches the running CPU and cannot use
get_cpu_var due to RT concerns. PCP requires these semantics to avoid
accidentally allocating remote memory.
Link: https://lkml.kernel.org/r/20220613125622.18628-8-mgorman@xxxxxxxxxxxxxxxxxxx
Signed-off-by: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx>
Cc: Hugh Dickins <hughd@xxxxxxxxxx>
Cc: Marcelo Tosatti <mtosatti@xxxxxxxxxx>
Cc: Michal Hocko <mhocko@xxxxxxxxxx>
Cc: Minchan Kim <minchan@xxxxxxxxxx>
Cc: Nicolas Saenz Julienne <nsaenzju@xxxxxxxxxx>
Cc: Vlastimil Babka <vbabka@xxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
---
mm/page_alloc.c | 226 ++++++++++++++++++++++++----------------------
1 file changed, 121 insertions(+), 105 deletions(-)
--- a/mm/page_alloc.c~mm-page_alloc-replace-local_lock-with-normal-spinlock
+++ a/mm/page_alloc.c
@@ -126,13 +126,6 @@ typedef int __bitwise fpi_t;
static DEFINE_MUTEX(pcp_batch_high_lock);
#define MIN_PERCPU_PAGELIST_HIGH_FRACTION (8)
-struct pagesets {
- local_lock_t lock;
-};
-static DEFINE_PER_CPU(struct pagesets, pagesets) = {
- .lock = INIT_LOCAL_LOCK(lock),
-};
-
#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT)
/*
* On SMP, spin_trylock is sufficient protection.
@@ -147,6 +140,81 @@ static DEFINE_PER_CPU(struct pagesets, p
#define pcp_trylock_finish(flags) local_irq_restore(flags)
#endif
+/*
+ * Locking a pcp requires a PCP lookup followed by a spinlock. To avoid
+ * a migration causing the wrong PCP to be locked and remote memory being
+ * potentially allocated, pin the task to the CPU for the lookup+lock.
+ * preempt_disable is used on !RT because it is faster than migrate_disable.
+ * migrate_disable is used on RT because otherwise RT spinlock usage is
+ * interfered with and a high priority task cannot preempt the allocator.
+ */
+#ifndef CONFIG_PREEMPT_RT
+#define pcpu_task_pin() preempt_disable()
+#define pcpu_task_unpin() preempt_enable()
+#else
+#define pcpu_task_pin() migrate_disable()
+#define pcpu_task_unpin() migrate_enable()
+#endif
+
+/*
+ * Generic helper to lookup and a per-cpu variable with an embedded spinlock.
+ * Return value should be used with equivalent unlock helper.
+ */
+#define pcpu_spin_lock(type, member, ptr) \
+({ \
+ type *_ret; \
+ pcpu_task_pin(); \
+ _ret = this_cpu_ptr(ptr); \
+ spin_lock(&_ret->member); \
+ _ret; \
+})
+
+#define pcpu_spin_lock_irqsave(type, member, ptr, flags) \
+({ \
+ type *_ret; \
+ pcpu_task_pin(); \
+ _ret = this_cpu_ptr(ptr); \
+ spin_lock_irqsave(&_ret->member, flags); \
+ _ret; \
+})
+
+#define pcpu_spin_trylock_irqsave(type, member, ptr, flags) \
+({ \
+ type *_ret; \
+ pcpu_task_pin(); \
+ _ret = this_cpu_ptr(ptr); \
+ if (!spin_trylock_irqsave(&_ret->member, flags)) \
+ _ret = NULL; \
+ _ret; \
+})
+
+#define pcpu_spin_unlock(member, ptr) \
+({ \
+ spin_unlock(&ptr->member); \
+ pcpu_task_unpin(); \
+})
+
+#define pcpu_spin_unlock_irqrestore(member, ptr, flags) \
+({ \
+ spin_unlock_irqrestore(&ptr->member, flags); \
+ pcpu_task_unpin(); \
+})
+
+/* struct per_cpu_pages specific helpers. */
+#define pcp_spin_lock(ptr) \
+ pcpu_spin_lock(struct per_cpu_pages, lock, ptr)
+
+#define pcp_spin_lock_irqsave(ptr, flags) \
+ pcpu_spin_lock_irqsave(struct per_cpu_pages, lock, ptr, flags)
+
+#define pcp_spin_trylock_irqsave(ptr, flags) \
+ pcpu_spin_trylock_irqsave(struct per_cpu_pages, lock, ptr, flags)
+
+#define pcp_spin_unlock(ptr) \
+ pcpu_spin_unlock(lock, ptr)
+
+#define pcp_spin_unlock_irqrestore(ptr, flags) \
+ pcpu_spin_unlock_irqrestore(lock, ptr, flags)
#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
DEFINE_PER_CPU(int, numa_node);
EXPORT_PER_CPU_SYMBOL(numa_node);
@@ -1477,10 +1545,7 @@ static void free_pcppages_bulk(struct zo
/* Ensure requested pindex is drained first. */
pindex = pindex - 1;
- /*
- * local_lock_irq held so equivalent to spin_lock_irqsave for
- * both PREEMPT_RT and non-PREEMPT_RT configurations.
- */
+ /* Caller must hold IRQ-safe pcp->lock so IRQs are disabled. */
spin_lock(&zone->lock);
isolated_pageblocks = has_isolate_pageblock(zone);
@@ -3048,10 +3113,7 @@ static int rmqueue_bulk(struct zone *zon
{
int i, allocated = 0;
- /*
- * local_lock_irq held so equivalent to spin_lock_irqsave for
- * both PREEMPT_RT and non-PREEMPT_RT configurations.
- */
+ /* Caller must hold IRQ-safe pcp->lock so IRQs are disabled. */
spin_lock(&zone->lock);
for (i = 0; i < count; ++i) {
struct page *page = __rmqueue(zone, order, migratetype,
@@ -3363,30 +3425,17 @@ static int nr_pcp_high(struct per_cpu_pa
return min(READ_ONCE(pcp->batch) << 2, high);
}
-/* Returns true if the page was committed to the per-cpu list. */
-static bool free_unref_page_commit(struct page *page, int migratetype,
- unsigned int order, bool locked)
+static void free_unref_page_commit(struct per_cpu_pages *pcp, struct zone *zone,
+ struct page *page, int migratetype,
+ unsigned int order)
{
- struct zone *zone = page_zone(page);
- struct per_cpu_pages *pcp;
int high;
int pindex;
bool free_high;
- unsigned long __maybe_unused UP_flags;
__count_vm_event(PGFREE);
- pcp = this_cpu_ptr(zone->per_cpu_pageset);
pindex = order_to_pindex(migratetype, order);
- if (!locked) {
- /* Protect against a parallel drain. */
- pcp_trylock_prepare(UP_flags);
- if (!spin_trylock(&pcp->lock)) {
- pcp_trylock_finish(UP_flags);
- return false;
- }
- }
-
list_add(&page->pcp_list, &pcp->lists[pindex]);
pcp->count += 1 << order;
@@ -3404,13 +3453,6 @@ static bool free_unref_page_commit(struc
free_pcppages_bulk(zone, nr_pcp_free(pcp, high, batch, free_high), pcp, pindex);
}
-
- if (!locked) {
- spin_unlock(&pcp->lock);
- pcp_trylock_finish(UP_flags);
- }
-
- return true;
}
/*
@@ -3418,10 +3460,12 @@ static bool free_unref_page_commit(struc
*/
void free_unref_page(struct page *page, unsigned int order)
{
- unsigned long flags;
+ struct per_cpu_pages *pcp;
+ struct zone *zone;
unsigned long pfn = page_to_pfn(page);
int migratetype;
- bool freed_pcp = false;
+ unsigned long flags;
+ unsigned long __maybe_unused UP_flags;
if (!free_unref_page_prepare(page, pfn, order))
return;
@@ -3442,12 +3486,16 @@ void free_unref_page(struct page *page,
migratetype = MIGRATE_MOVABLE;
}
- local_lock_irqsave(&pagesets.lock, flags);
- freed_pcp = free_unref_page_commit(page, migratetype, order, false);
- local_unlock_irqrestore(&pagesets.lock, flags);
-
- if (unlikely(!freed_pcp))
+ zone = page_zone(page);
+ pcp_trylock_prepare(UP_flags);
+ pcp = pcpu_spin_trylock_irqsave(struct per_cpu_pages, lock, zone->per_cpu_pageset, flags);
+ if (pcp) {
+ free_unref_page_commit(pcp, zone, page, migratetype, order);
+ pcp_spin_unlock_irqrestore(pcp, flags);
+ } else {
free_one_page(page_zone(page), page, pfn, order, migratetype, FPI_NONE);
+ }
+ pcp_trylock_finish(UP_flags);
}
/*
@@ -3496,20 +3544,20 @@ void free_unref_page_list(struct list_he
if (list_empty(list))
return;
- local_lock_irqsave(&pagesets.lock, flags);
-
page = lru_to_page(list);
locked_zone = page_zone(page);
- pcp = this_cpu_ptr(locked_zone->per_cpu_pageset);
- spin_lock(&pcp->lock);
+ pcp = pcp_spin_lock_irqsave(locked_zone->per_cpu_pageset, flags);
list_for_each_entry_safe(page, next, list, lru) {
struct zone *zone = page_zone(page);
/* Different zone, different pcp lock. */
if (zone != locked_zone) {
+ /* Leave IRQs enabled as a new lock is acquired. */
spin_unlock(&pcp->lock);
locked_zone = zone;
+
+ /* Preemption disabled by pcp_spin_lock_irqsave. */
pcp = this_cpu_ptr(zone->per_cpu_pageset);
spin_lock(&pcp->lock);
}
@@ -3524,33 +3572,19 @@ void free_unref_page_list(struct list_he
trace_mm_page_free_batched(page);
- /*
- * If there is a parallel drain in progress, free to the buddy
- * allocator directly. This is expensive as the zone lock will
- * be acquired multiple times but if a drain is in progress
- * then an expensive operation is already taking place.
- *
- * TODO: Always false at the moment due to local_lock_irqsave
- * and is preparation for converting to local_lock.
- */
- if (unlikely(!free_unref_page_commit(page, migratetype, 0, true)))
- free_one_page(page_zone(page), page, page_to_pfn(page), 0, migratetype, FPI_NONE);
+ free_unref_page_commit(pcp, zone, page, migratetype, 0);
/*
* Guard against excessive IRQ disabled times when we get
* a large list of pages to free.
*/
if (++batch_count == SWAP_CLUSTER_MAX) {
- spin_unlock(&pcp->lock);
- local_unlock_irqrestore(&pagesets.lock, flags);
+ pcp_spin_unlock_irqrestore(pcp, flags);
batch_count = 0;
- local_lock_irqsave(&pagesets.lock, flags);
- pcp = this_cpu_ptr(locked_zone->per_cpu_pageset);
- spin_lock(&pcp->lock);
+ pcp = pcp_spin_lock_irqsave(locked_zone->per_cpu_pageset, flags);
}
}
- spin_unlock(&pcp->lock);
- local_unlock_irqrestore(&pagesets.lock, flags);
+ pcp_spin_unlock_irqrestore(pcp, flags);
}
/*
@@ -3718,28 +3752,9 @@ struct page *__rmqueue_pcplist(struct zo
int migratetype,
unsigned int alloc_flags,
struct per_cpu_pages *pcp,
- struct list_head *list,
- bool locked)
+ struct list_head *list)
{
struct page *page;
- unsigned long __maybe_unused UP_flags;
-
- /*
- * spin_trylock is not necessary right now due to due to
- * local_lock_irqsave and is a preparation step for
- * a conversion to local_lock using the trylock to prevent
- * IRQ re-entrancy. If pcp->lock cannot be acquired, the caller
- * uses rmqueue_buddy.
- *
- * TODO: Convert local_lock_irqsave to local_lock.
- */
- if (unlikely(!locked)) {
- pcp_trylock_prepare(UP_flags);
- if (!spin_trylock(&pcp->lock)) {
- pcp_trylock_finish(UP_flags);
- return NULL;
- }
- }
do {
if (list_empty(list)) {
@@ -3772,10 +3787,6 @@ struct page *__rmqueue_pcplist(struct zo
} while (check_new_pcp(page, order));
out:
- if (!locked) {
- spin_unlock(&pcp->lock);
- pcp_trylock_finish(UP_flags);
- }
return page;
}
@@ -3790,19 +3801,29 @@ static struct page *rmqueue_pcplist(stru
struct list_head *list;
struct page *page;
unsigned long flags;
+ unsigned long __maybe_unused UP_flags;
- local_lock_irqsave(&pagesets.lock, flags);
+ /*
+ * spin_trylock_irqsave is not necessary right now as it'll only be
+ * true when contending with a remote drain. It's in place as a
+ * preparation step before converting pcp locking to spin_trylock
+ * to protect against IRQ reentry.
+ */
+ pcp_trylock_prepare(UP_flags);
+ pcp = pcp_spin_trylock_irqsave(zone->per_cpu_pageset, flags);
+ if (!pcp)
+ return NULL;
/*
* On allocation, reduce the number of pages that are batch freed.
* See nr_pcp_free() where free_factor is increased for subsequent
* frees.
*/
- pcp = this_cpu_ptr(zone->per_cpu_pageset);
pcp->free_factor >>= 1;
list = &pcp->lists[order_to_pindex(migratetype, order)];
- page = __rmqueue_pcplist(zone, order, migratetype, alloc_flags, pcp, list, false);
- local_unlock_irqrestore(&pagesets.lock, flags);
+ page = __rmqueue_pcplist(zone, order, migratetype, alloc_flags, pcp, list);
+ pcp_spin_unlock_irqrestore(pcp, flags);
+ pcp_trylock_finish(UP_flags);
if (page) {
__count_zid_vm_events(PGALLOC, page_zonenum(page), 1);
zone_statistics(preferred_zone, zone, 1);
@@ -5395,10 +5416,8 @@ unsigned long __alloc_pages_bulk(gfp_t g
goto failed;
/* Attempt the batch allocation */
- local_lock_irqsave(&pagesets.lock, flags);
- pcp = this_cpu_ptr(zone->per_cpu_pageset);
+ pcp = pcp_spin_lock_irqsave(zone->per_cpu_pageset, flags);
pcp_list = &pcp->lists[order_to_pindex(ac.migratetype, 0)];
- spin_lock(&pcp->lock);
while (nr_populated < nr_pages) {
@@ -5409,13 +5428,11 @@ unsigned long __alloc_pages_bulk(gfp_t g
}
page = __rmqueue_pcplist(zone, 0, ac.migratetype, alloc_flags,
- pcp, pcp_list, true);
+ pcp, pcp_list);
if (unlikely(!page)) {
/* Try and allocate at least one page */
- if (!nr_account) {
- spin_unlock(&pcp->lock);
+ if (!nr_account)
goto failed_irq;
- }
break;
}
nr_account++;
@@ -5428,8 +5445,7 @@ unsigned long __alloc_pages_bulk(gfp_t g
nr_populated++;
}
- spin_unlock(&pcp->lock);
- local_unlock_irqrestore(&pagesets.lock, flags);
+ pcp_spin_unlock_irqrestore(pcp, flags);
__count_zid_vm_events(PGALLOC, zone_idx(zone), nr_account);
zone_statistics(ac.preferred_zoneref->zone, zone, nr_account);
@@ -5438,7 +5454,7 @@ out:
return nr_populated;
failed_irq:
- local_unlock_irqrestore(&pagesets.lock, flags);
+ pcp_spin_unlock_irqrestore(pcp, flags);
failed:
page = __alloc_pages(gfp, 0, preferred_nid, nodemask);
_
Patches currently in -mm which might be from mgorman@xxxxxxxxxxxxxxxxxxx are
mm-page_alloc-add-page-buddy_list-and-page-pcp_list.patch
mm-page_alloc-use-only-one-pcp-list-for-thp-sized-allocations.patch
mm-page_alloc-split-out-buddy-removal-code-from-rmqueue-into-separate-helper.patch
mm-page_alloc-remove-mistaken-page-==-null-check-in-rmqueue.patch
mm-page_alloc-protect-pcp-lists-with-a-spinlock.patch
mm-page_alloc-replace-local_lock-with-normal-spinlock.patch
