v1->v2: Fix compile error. v2->v3: Wrap core functions with dpool_ prefix to avoid call them directly.
This part include:
1. Compatible with memory reliable feature[1-3]. 2. Compatible with memory migration[4] 3. Fix an incorrect set of subpool[5]. 4. Make the lock in dynamic pool irq safe[6]. 5. Wrap core functions with dpool_ prefix to avoid call them directly[7].
Liu Shixin (7): mm/dynamic_pool: Stop alloc reliable page from dynamic pool mm/mem_reliable: Treat page from dhugetlb pool as unreliable page mm/mem_reliable: Fallback to dpool if reliable memory is not enough mm/dynamic_pool: skip unexpected migration mm/dynamic_pool: don't set subpool for page from dynamic pool mm/dynamic_pool: disable irq for dynamic_pool lock mm/dynamic_pool: Wrap some core functions with dpool prefix
include/linux/dynamic_pool.h | 6 ++ include/linux/mem_reliable.h | 3 + mm/compaction.c | 4 ++ mm/dynamic_pool.c | 128 +++++++++++++++++++---------------- mm/hugetlb.c | 3 +- mm/internal.h | 12 ++-- mm/migrate.c | 7 ++ mm/page_alloc.c | 53 +++++++++++++-- mm/page_isolation.c | 9 ++- 9 files changed, 155 insertions(+), 70 deletions(-)
hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9CDZZ
--------------------------------
With memory reliable enabled, user expect to alloc memory from mirrored regions (also known as the non-movable zone). However dynamic pool does not contain such reliable information. So bypass it.
Signed-off-by: Liu Shixin liushixin2@huawei.com --- mm/dynamic_pool.c | 5 +++++ 1 file changed, 5 insertions(+)
diff --git a/mm/dynamic_pool.c b/mm/dynamic_pool.c index 82cbaa27f3d2..ce1225b85a2e 100644 --- a/mm/dynamic_pool.c +++ b/mm/dynamic_pool.c @@ -700,6 +700,11 @@ static bool dpool_should_alloc(gfp_t gfp_mask, unsigned int order) if ((gfp | __GFP_IO | __GFP_FS) != GFP_HIGHUSER_MOVABLE) return false;
+#ifdef CONFIG_MEMORY_RELIABLE + if (mem_reliable_is_enabled() && (gfp_mask & GFP_RELIABLE)) + return false; +#endif + return true; }
hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9CDZZ
--------------------------------
Memory from dhugetlb pool is lacking reliable attributes. Treat them as unreliable pages.
Signed-off-by: Liu Shixin liushixin2@huawei.com --- include/linux/mem_reliable.h | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/include/linux/mem_reliable.h b/include/linux/mem_reliable.h index 15f69349a2a8..d8cabf94f4a3 100644 --- a/include/linux/mem_reliable.h +++ b/include/linux/mem_reliable.h @@ -51,6 +51,9 @@ static inline bool page_reliable(struct page *page) if (!page) return false;
+ if (PagePool(page)) + return false; + return page_zonenum(page) < ZONE_MOVABLE; }
hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9CDZZ
--------------------------------
Fallback to dpool if reliable memory is not enough.
Kswapd and driect reclaim will not be trigger. Since the later normal memory allocation can trigger this, there is no problem here.
Signed-off-by: Liu Shixin liushixin2@huawei.com --- include/linux/dynamic_pool.h | 6 ++++++ mm/dynamic_pool.c | 4 ++-- mm/page_alloc.c | 27 +++++++++++++++++++++++++++ 3 files changed, 35 insertions(+), 2 deletions(-)
diff --git a/include/linux/dynamic_pool.h b/include/linux/dynamic_pool.h index d6ef9214c586..1d41c6a853c3 100644 --- a/include/linux/dynamic_pool.h +++ b/include/linux/dynamic_pool.h @@ -111,6 +111,7 @@ void dynamic_pool_bind_file(struct hugetlbfs_inode_info *p, struct hstate *h); void dynamic_pool_unbind_file(struct hugetlbfs_inode_info *p); int dynamic_pool_hugetlb_acct_memory(struct hstate *h, long delta, struct hugetlbfs_inode_info *p); +bool dynamic_pool_should_alloc(gfp_t gfp_mask, unsigned int order); struct folio *dynamic_pool_alloc_hugepage(struct hugetlbfs_inode_info *p, struct hstate *h, bool reserved); void dynamic_pool_free_hugepage(struct folio *folio, bool restore_reserve); @@ -186,6 +187,11 @@ static inline int dynamic_pool_hugetlb_acct_memory(struct hstate *h, long delta, return -ENOMEM; }
+static inline bool dynamic_pool_should_alloc(gfp_t gfp_mask, unsigned int order) +{ + return false; +} + static inline struct folio *dynamic_pool_alloc_hugepage(struct hugetlbfs_inode_info *p, struct hstate *h, bool reserved) { diff --git a/mm/dynamic_pool.c b/mm/dynamic_pool.c index ce1225b85a2e..751a2dcc6816 100644 --- a/mm/dynamic_pool.c +++ b/mm/dynamic_pool.c @@ -682,7 +682,7 @@ int dynamic_pool_can_attach(struct task_struct *tsk, struct mem_cgroup *memcg) return ret; }
-static bool dpool_should_alloc(gfp_t gfp_mask, unsigned int order) +bool dynamic_pool_should_alloc(gfp_t gfp_mask, unsigned int order) { gfp_t gfp = gfp_mask & GFP_HIGHUSER_MOVABLE;
@@ -719,7 +719,7 @@ struct page *dynamic_pool_alloc_page(gfp_t gfp, unsigned int order, if (!dpool_enabled) return NULL;
- if (!dpool_should_alloc(gfp, order)) + if (!dynamic_pool_should_alloc(gfp, order)) return NULL;
dpool = dpool_get_from_task(current); diff --git a/mm/page_alloc.c b/mm/page_alloc.c index b063dd0d456d..815b0c0212fd 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -4098,6 +4098,17 @@ static inline void mem_reliable_fallback_slowpath(gfp_t gfp_mask, return; } } + +static inline bool mem_reliable_fallback_dpool(gfp_t gfp_mask, unsigned int order) +{ + if (!reliable_allow_fb_enabled()) + return false; + + if (!(gfp_mask & GFP_RELIABLE)) + return false; + + return dynamic_pool_should_alloc(gfp_mask & ~GFP_RELIABLE, order); +} #else static inline struct zone *mem_reliable_fallback_zone(gfp_t gfp_mask, struct alloc_context *ac) @@ -4106,6 +4117,10 @@ static inline struct zone *mem_reliable_fallback_zone(gfp_t gfp_mask, } static inline void mem_reliable_fallback_slowpath(gfp_t gfp_mask, struct alloc_context *ac) {} +static inline bool mem_reliable_fallback_dpool(gfp_t gfp_mask, unsigned int order) +{ + return false; +} #endif
static inline struct page * @@ -4765,6 +4780,18 @@ struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid, if (likely(page)) goto out;
+ /* + * Fallback to dpool if mirrored momory is not enough. + * + * Kswapd and driect reclaim will not be trigger, since the later + * normal memory allocation can trigger this, there is no problem + * here. + */ + if (mem_reliable_fallback_dpool(gfp, order)) { + gfp &= ~GFP_RELIABLE; + goto retry; + } + alloc_gfp = gfp; ac.spread_dirty_pages = false;
hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9CDZZ
--------------------------------
With dynamic pool feature, some memory is isolated in the dynamic pool. When try to compact memory, the kcompactd thread will scan all memory, althougt some memory is belonging to dynamic pool, kcompactd still try to migrate them. After migration, these memory will free to dynamic pool rather than buddy system, which results the free pages in buddy system decreased.
Since it is unnecessary to compact the memory in the dynamic pool, skip migrate them to fix the problem.
The same problem also existed in alloc_contig_range(), offline_pages() and numa balancing. Skip it again in these three scenarios.
In addition to this, we have to consider the migration of hugepage, if a hugepage is from dynamic pool, we should not allow to migrate it.
Signed-off-by: Liu Shixin liushixin2@huawei.com --- mm/compaction.c | 4 ++++ mm/migrate.c | 7 +++++++ mm/page_isolation.c | 9 +++++++-- 3 files changed, 18 insertions(+), 2 deletions(-)
diff --git a/mm/compaction.c b/mm/compaction.c index 771e9629b95c..e9fe6777c8a3 100644 --- a/mm/compaction.c +++ b/mm/compaction.c @@ -23,6 +23,7 @@ #include <linux/freezer.h> #include <linux/page_owner.h> #include <linux/psi.h> +#include <linux/dynamic_pool.h> #include "internal.h"
#ifdef CONFIG_COMPACTION @@ -2024,6 +2025,9 @@ static isolate_migrate_t isolate_migratepages(struct compact_control *cc) continue; }
+ if (page_in_dynamic_pool(page)) + continue; + /* * If isolation recently failed, do not retry. Only check the * pageblock once. COMPACT_CLUSTER_MAX causes a pageblock diff --git a/mm/migrate.c b/mm/migrate.c index 1e88f81d2369..f13c0fcb989c 100644 --- a/mm/migrate.c +++ b/mm/migrate.c @@ -50,6 +50,7 @@ #include <linux/random.h> #include <linux/sched/sysctl.h> #include <linux/memory-tiers.h> +#include <linux/dynamic_pool.h>
#include <asm/tlbflush.h>
@@ -2009,6 +2010,9 @@ struct folio *alloc_migration_target(struct folio *src, unsigned long private) if (folio_test_hugetlb(src)) { struct hstate *h = folio_hstate(src);
+ if (page_in_dynamic_pool(folio_page(src, 0))) + return NULL; + gfp_mask = htlb_modify_alloc_mask(h, gfp_mask); return alloc_hugetlb_folio_nodemask(h, nid, mtc->nmask, gfp_mask); @@ -2525,6 +2529,9 @@ static int numamigrate_isolate_folio(pg_data_t *pgdat, struct folio *folio) return 0; }
+ if (page_in_dynamic_pool(folio_page(folio, 0))) + return 0; + if (!folio_isolate_lru(folio)) return 0;
diff --git a/mm/page_isolation.c b/mm/page_isolation.c index bcf99ba747a0..fefc8a926944 100644 --- a/mm/page_isolation.c +++ b/mm/page_isolation.c @@ -10,6 +10,7 @@ #include <linux/hugetlb.h> #include <linux/page_owner.h> #include <linux/migrate.h> +#include <linux/dynamic_pool.h> #include "internal.h"
#define CREATE_TRACE_POINTS @@ -329,6 +330,9 @@ static int isolate_single_pageblock(unsigned long boundary_pfn, int flags, start_pfn = max(ALIGN_DOWN(isolate_pageblock, MAX_ORDER_NR_PAGES), zone->zone_start_pfn);
+ if (page_in_dynamic_pool(pfn_to_page(isolate_pageblock))) + return -EBUSY; + if (skip_isolation) { int mt __maybe_unused = get_pageblock_migratetype(pfn_to_page(isolate_pageblock));
@@ -558,8 +562,9 @@ int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn, pfn < isolate_end - pageblock_nr_pages; pfn += pageblock_nr_pages) { page = __first_valid_page(pfn, pageblock_nr_pages); - if (page && set_migratetype_isolate(page, migratetype, flags, - start_pfn, end_pfn)) { + if (page && (page_in_dynamic_pool(page) || + set_migratetype_isolate(page, migratetype, flags, + start_pfn, end_pfn))) { undo_isolate_page_range(isolate_start, pfn, migratetype); unset_migratetype_isolate( pfn_to_page(isolate_end - pageblock_nr_pages),
hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9CDZZ
--------------------------------
Dynamic pool has individual resv_huge_pages value, don't account the page into subpool. The subpool record in page is only used when free the page, so there is no other effects.
Signed-off-by: Liu Shixin liushixin2@huawei.com --- mm/hugetlb.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 0d3648572898..b45e35e8af2c 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -3258,7 +3258,8 @@ struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
spin_unlock_irq(&hugetlb_lock);
- hugetlb_set_folio_subpool(folio, spool); + if (!page_from_dynamic_pool(folio_page(folio, 0))) + hugetlb_set_folio_subpool(folio, spool);
map_commit = vma_commit_reservation(h, vma, addr); if (unlikely(map_chg > map_commit)) {
hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9CDZZ
--------------------------------
The alloc/free path can be called from irq context. So make the lock irq safe.
Signed-off-by: Liu Shixin liushixin2@huawei.com --- mm/dynamic_pool.c | 103 +++++++++++++++++++++++++--------------------- 1 file changed, 55 insertions(+), 48 deletions(-)
diff --git a/mm/dynamic_pool.c b/mm/dynamic_pool.c index 751a2dcc6816..15a491a76a58 100644 --- a/mm/dynamic_pool.c +++ b/mm/dynamic_pool.c @@ -420,7 +420,7 @@ static int dpool_promote_pool(struct dynamic_pool *dpool, int type) src_pool = &dpool->pool[type + 1]; dst_pool = &dpool->pool[type];
- spin_lock(&dpool->lock); + spin_lock_irq(&dpool->lock);
if (!dst_pool->split_pages) goto unlock; @@ -438,13 +438,13 @@ static int dpool_promote_pool(struct dynamic_pool *dpool, int type) * there is no way to free spage_next, so * it is safe to unlock here. */ - spin_unlock(&dpool->lock); + spin_unlock_irq(&dpool->lock); cond_resched(); lru_add_drain_all(); dpool_disable_pcp_pool(dpool, true); do_migrate_range(spage->start_pfn, spage->start_pfn + nr_pages); - spin_lock(&dpool->lock); + spin_lock_irq(&dpool->lock); dpool_enable_pcp_pool(dpool); ret = dpool_promote_huge_page(src_pool, dst_pool, spage); break; @@ -463,7 +463,7 @@ static int dpool_promote_pool(struct dynamic_pool *dpool, int type) }
unlock: - spin_unlock(&dpool->lock); + spin_unlock_irq(&dpool->lock); if (!ret) kfree(spage); trace_dpool_promote(dpool, type, page, ret); @@ -479,11 +479,12 @@ static void dpool_refill_pcp_pool(struct dynamic_pool *dpool, { struct pages_pool *pool = &dpool->pool[PAGES_POOL_4K]; struct page *page, *next; + unsigned long flags; int i = 0;
lockdep_assert_held(&pcp_pool->lock);
- spin_lock(&dpool->lock); + spin_lock_irqsave(&dpool->lock, flags);
if (!pool->free_pages && dpool_demote_pool_locked(dpool, PAGES_POOL_2M)) goto unlock; @@ -498,7 +499,7 @@ static void dpool_refill_pcp_pool(struct dynamic_pool *dpool, }
unlock: - spin_unlock(&dpool->lock); + spin_unlock_irqrestore(&dpool->lock, flags); }
static void dpool_drain_pcp_pool(struct dynamic_pool *dpool, @@ -507,11 +508,12 @@ static void dpool_drain_pcp_pool(struct dynamic_pool *dpool, { struct pages_pool *pool = &dpool->pool[PAGES_POOL_4K]; struct page *page, *next; + unsigned long flags; int i = 0;
lockdep_assert_held(&pcp_pool->lock);
- spin_lock(&dpool->lock); + spin_lock_irqsave(&dpool->lock, flags); list_for_each_entry_safe(page, next, &pcp_pool->freelist, lru) { list_move_tail(&page->lru, &pool->freelist); __SetPageDpool(page); @@ -523,7 +525,7 @@ static void dpool_drain_pcp_pool(struct dynamic_pool *dpool,
pool->used_pages += pcp_pool->used_pages; pcp_pool->used_pages = 0; - spin_unlock(&dpool->lock); + spin_unlock_irqrestore(&dpool->lock, flags); }
static void dpool_drain_all_pcp_pool(struct dynamic_pool *dpool) @@ -890,16 +892,16 @@ struct folio *dynamic_pool_alloc_hugepage(struct hugetlbfs_inode_info *p, if (!dpool) return NULL;
- spin_lock_irqsave(&dpool->lock, flags); - if (!dpool->online) - goto unlock; - if (hstate_is_gigantic(h)) type = PAGES_POOL_1G; else type = PAGES_POOL_2M; pool = &dpool->pool[type];
+ spin_lock_irqsave(&dpool->lock, flags); + if (!dpool->online) + goto unlock; + list_for_each_entry(folio, &pool->freelist, lru) { if (folio_test_hwpoison(folio)) continue; @@ -943,13 +945,14 @@ void dynamic_pool_free_hugepage(struct folio *folio, bool restore_reserve) return; }
- spin_lock_irqsave(&dpool->lock, flags); if (hstate_is_gigantic(h)) type = PAGES_POOL_1G; else type = PAGES_POOL_2M; pool = &dpool->pool[type];
+ spin_lock_irqsave(&dpool->lock, flags); + if (folio_test_hwpoison(folio)) goto unlock;
@@ -1209,10 +1212,10 @@ static int dpool_fill_from_hugetlb(struct dynamic_pool *dpool, void *arg) if (!h) return -EINVAL;
- spin_lock(&hugetlb_lock); + spin_lock_irq(&hugetlb_lock); if ((h->free_huge_pages_node[nid] < nr_pages) || (h->free_huge_pages - h->resv_huge_pages < nr_pages)) { - spin_unlock(&hugetlb_lock); + spin_unlock_irq(&hugetlb_lock); return -ENOMEM; }
@@ -1233,24 +1236,24 @@ static int dpool_fill_from_hugetlb(struct dynamic_pool *dpool, void *arg) list_move(&page->lru, &page_list); count++; } - spin_unlock(&hugetlb_lock); + spin_unlock_irq(&hugetlb_lock);
list_for_each_entry_safe(page, next, &page_list, lru) { if (hugetlb_vmemmap_restore(h, page)) { - spin_lock(&hugetlb_lock); + spin_lock_irq(&hugetlb_lock); enqueue_hugetlb_folio(h, folio); - spin_unlock(&hugetlb_lock); + spin_unlock_irq(&hugetlb_lock); pr_err("restore hugetlb_vmemmap failed page 0x%px\n", page); continue; }
__SetPageDpool(page); - spin_lock(&dpool->lock); + spin_lock_irq(&dpool->lock); list_move(&page->lru, &pool->freelist); pool->free_pages++; dpool->total_pages++; - spin_unlock(&dpool->lock); + spin_unlock_irq(&dpool->lock); }
return 0; @@ -1267,7 +1270,7 @@ static int dpool_drain_to_hugetlb(struct dynamic_pool *dpool) if (!h) return -EINVAL;
- spin_lock(&dpool->lock); + spin_lock_irq(&dpool->lock); list_for_each_entry_safe(page, next, &pool->freelist, lru) { WARN_ON(PageHWPoison(page)); idx = hugepage_index(page_to_pfn(page)); @@ -1278,13 +1281,13 @@ static int dpool_drain_to_hugetlb(struct dynamic_pool *dpool) pool->free_pages--; dpool->total_pages--; } - spin_unlock(&dpool->lock); + spin_unlock_irq(&dpool->lock);
list_for_each_entry_safe(page, next, &page_list, lru) { hugetlb_vmemmap_optimize(h, page); - spin_lock(&hugetlb_lock); + spin_lock_irq(&hugetlb_lock); enqueue_hugetlb_folio(h, page_folio(page)); - spin_unlock(&hugetlb_lock); + spin_unlock_irq(&hugetlb_lock); }
return dpool->total_pages ? -ENOMEM : 0; @@ -1308,20 +1311,20 @@ static int dpool_merge_all(struct dynamic_pool *dpool) } }
- spin_lock(&dpool->lock); + spin_lock_irq(&dpool->lock); if (pool->split_pages || pool->used_huge_pages || pool->resv_huge_pages) { ret = -ENOMEM; pr_err("some 2M pages are still in use or mmap, delete failed: "); pr_cont_cgroup_name(dpool->memcg->css.cgroup); pr_cont("\n"); - spin_unlock(&dpool->lock); + spin_unlock_irq(&dpool->lock); goto out; }
pool->free_pages += pool->nr_huge_pages; pool->nr_huge_pages = 0; pool->free_huge_pages = 0; - spin_unlock(&dpool->lock); + spin_unlock_irq(&dpool->lock);
pool = &dpool->pool[PAGES_POOL_1G]; while (pool->split_pages) { @@ -1336,20 +1339,20 @@ static int dpool_merge_all(struct dynamic_pool *dpool) } }
- spin_lock(&dpool->lock); + spin_lock_irq(&dpool->lock); if (pool->split_pages || pool->used_huge_pages || pool->resv_huge_pages) { ret = -ENOMEM; pr_err("some 1G pages are still in use or mmap, delete failed: "); pr_cont_cgroup_name(dpool->memcg->css.cgroup); pr_cont("\n"); - spin_unlock(&dpool->lock); + spin_unlock_irq(&dpool->lock); goto out; }
pool->free_pages += pool->nr_huge_pages; pool->nr_huge_pages = 0; pool->free_huge_pages = 0; - spin_unlock(&dpool->lock); + spin_unlock_irq(&dpool->lock); ret = 0;
out: @@ -1441,7 +1444,7 @@ void dynamic_pool_show(struct mem_cgroup *memcg, struct seq_file *m) }
dpool_disable_pcp_pool(dpool, false); - spin_lock(&dpool->lock); + spin_lock_irq(&dpool->lock);
/* * no others can modify the count because pcp pool is disabled and @@ -1476,7 +1479,7 @@ void dynamic_pool_show(struct mem_cgroup *memcg, struct seq_file *m) seq_printf(m, "4K_free_pages %lu\n", free_pages); seq_printf(m, "4K_used_pages %ld\n", used_pages);
- spin_unlock(&dpool->lock); + spin_unlock_irq(&dpool->lock); dpool_enable_pcp_pool(dpool); dpool_put(dpool); } @@ -1499,22 +1502,25 @@ int dynamic_pool_reserve_hugepage(struct mem_cgroup *memcg, goto unlock;
pool = &dpool->pool[type]; - spin_lock(&dpool->lock); + spin_lock_irq(&dpool->lock); if (nr_pages > pool->nr_huge_pages) { delta = nr_pages - pool->nr_huge_pages; while (delta > pool->free_pages && - !dpool_demote_pool_locked(dpool, type - 1)) - cond_resched_lock(&dpool->lock); + !dpool_demote_pool_locked(dpool, type - 1)) { + spin_unlock_irq(&dpool->lock); + cond_resched(); + spin_lock_irq(&dpool->lock); + } /* Only try merge pages for 2M pages */ if (type == PAGES_POOL_2M) { while (delta > pool->free_pages) { - spin_unlock(&dpool->lock); + spin_unlock_irq(&dpool->lock); cond_resched(); if (dpool_promote_pool(dpool, type)) { - spin_lock(&dpool->lock); + spin_lock_irq(&dpool->lock); break; } - spin_lock(&dpool->lock); + spin_lock_irq(&dpool->lock); } } delta = min(delta, pool->free_pages); @@ -1528,7 +1534,7 @@ int dynamic_pool_reserve_hugepage(struct mem_cgroup *memcg, pool->free_huge_pages -= delta; pool->free_pages += delta; } - spin_unlock(&dpool->lock); + spin_unlock_irq(&dpool->lock); dpool_put(dpool); ret = 0;
@@ -1564,7 +1570,7 @@ static int dpool_fill_from_pagelist(struct dynamic_pool *dpool, void *arg) memcpy(dpool->pfn_ranges, info->pfn_ranges, sizeof(struct range) * dpool->range_cnt);
- spin_lock(&dpool->lock); + spin_lock_irq(&dpool->lock);
for (i = 0; i < dpool->range_cnt; i++) { struct range *range = &dpool->pfn_ranges[i]; @@ -1591,7 +1597,7 @@ static int dpool_fill_from_pagelist(struct dynamic_pool *dpool, void *arg) ret = 0;
unlock: - spin_unlock(&dpool->lock); + spin_unlock_irq(&dpool->lock);
return ret; } @@ -1609,7 +1615,7 @@ static int dpool_migrate_used_pages(struct dynamic_pool *dpool) int range_cnt = dpool->range_cnt; int i;
- spin_lock(&dpool->lock); + spin_lock_irq(&dpool->lock);
dpool->nr_poisoned_pages = 0; for (i = 0; i < range_cnt; i++) { @@ -1620,11 +1626,11 @@ static int dpool_migrate_used_pages(struct dynamic_pool *dpool) struct page *page = pfn_to_page(pfn);
/* Unlock and try migration. */ - spin_unlock(&dpool->lock); + spin_unlock_irq(&dpool->lock); cond_resched();
if (PageDpool(page)) { - spin_lock(&dpool->lock); + spin_lock_irq(&dpool->lock); continue; }
@@ -1633,11 +1639,11 @@ static int dpool_migrate_used_pages(struct dynamic_pool *dpool)
lru_add_drain_all(); do_migrate_range(pfn, pfn + 1); - spin_lock(&dpool->lock); + spin_lock_irq(&dpool->lock); } }
- spin_unlock(&dpool->lock); + spin_unlock_irq(&dpool->lock);
return 0; } @@ -1701,6 +1707,7 @@ void dynamic_pool_show_meminfo(struct seq_file *m) struct pages_pool *pool; unsigned long free_pages = 0; long used_pages = 0; + unsigned long flags;
if (!dpool_enabled || !enable_dpagelist) return; @@ -1711,11 +1718,11 @@ void dynamic_pool_show_meminfo(struct seq_file *m)
pool = &dpool->pool[PAGES_POOL_4K]; dpool_disable_pcp_pool(dpool, false); - spin_lock(&dpool->lock); + spin_lock_irqsave(&dpool->lock, flags); dpool_sum_pcp_pool(dpool, &free_pages, &used_pages); free_pages += pool->free_pages; used_pages += pool->used_pages; - spin_unlock(&dpool->lock); + spin_unlock_irqrestore(&dpool->lock, flags); dpool_enable_pcp_pool(dpool);
out:
hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9CDZZ
--------------------------------
There are some core functions called by function from dynamic_pool.c. This may impact performance in some case. Wrap these core functions with dpool_ prefix and use the wrapper instead.
Signed-off-by: Liu Shixin liushixin2@huawei.com --- mm/dynamic_pool.c | 16 ++++++++-------- mm/internal.h | 11 ++++++----- mm/page_alloc.c | 29 +++++++++++++++++++++++++---- 3 files changed, 39 insertions(+), 17 deletions(-)
diff --git a/mm/dynamic_pool.c b/mm/dynamic_pool.c index 15a491a76a58..b1590362c2c9 100644 --- a/mm/dynamic_pool.c +++ b/mm/dynamic_pool.c @@ -272,7 +272,7 @@ static int dpool_demote_huge_page(struct pages_pool *src_pool, clear_compound_page(page_folio(page), PMD_ORDER); for (i = 0; i < nr_pages; i++) { subpage = folio_page(folio, i); - free_pages_prepare(subpage, 0, 0); + dpool_free_page_prepare(subpage); __SetPageDpool(subpage); list_add_tail(&subpage->lru, &dst_pool->freelist); dst_pool->free_pages++; @@ -395,7 +395,7 @@ static int dpool_promote_huge_page(struct pages_pool *src_pool, }
page = pfn_to_page(spage->start_pfn); - prep_new_page(page, PMD_ORDER, __GFP_COMP, 0); + dpool_prep_new_page(page, PMD_ORDER, __GFP_COMP, 0); set_page_count(page, 0); folio_change_private(page_folio(page), NULL); __SetPageDpool(page); @@ -616,7 +616,7 @@ static struct page *dpool_alloc_pcp_page(struct dynamic_pool *dpool) pcp_pool->free_pages--; pcp_pool->used_pages++;
- if (check_new_page(page)) { + if (dpool_check_new_page(page)) { SetPagePool(page); goto retry; } @@ -643,7 +643,7 @@ static int dpool_free_pcp_page(struct dynamic_pool *dpool, struct page *page) }
ClearPagePool(page); - if (!free_pages_prepare(page, 0, 0)) { + if (!dpool_free_page_prepare(page)) { SetPagePool(page); goto unlock; } @@ -756,7 +756,7 @@ struct page *dynamic_pool_alloc_page(gfp_t gfp, unsigned int order, pool->free_pages--; pool->used_pages++;
- if (check_new_page(page)) { + if (dpool_check_new_page(page)) { /* This is a bad page, treat it as a used pages */ SetPagePool(page); goto retry; @@ -769,7 +769,7 @@ struct page *dynamic_pool_alloc_page(gfp_t gfp, unsigned int order, put: dpool_put(dpool); if (page) - prep_new_page(page, order, gfp, alloc_flags); + dpool_prep_new_page(page, order, gfp, alloc_flags);
return page; } @@ -796,7 +796,7 @@ void dynamic_pool_free_page(struct page *page) spin_lock_irqsave(&dpool->lock, flags);
ClearPagePool(page); - if (!free_pages_prepare(page, 0, 0)) { + if (!dpool_free_page_prepare(page)) { SetPagePool(page); goto unlock; } @@ -1582,7 +1582,7 @@ static int dpool_fill_from_pagelist(struct dynamic_pool *dpool, void *arg) set_page_count(page, 0); page_mapcount_reset(page);
- if (!free_pages_prepare(page, 0, 0)) { + if (!dpool_free_page_prepare(page)) { pr_err("fill pool failed, check pages failed\n"); goto unlock; } diff --git a/mm/internal.h b/mm/internal.h index eab2199d16c8..a7372dfd84c9 100644 --- a/mm/internal.h +++ b/mm/internal.h @@ -439,11 +439,12 @@ extern void prep_compound_page(struct page *page, unsigned int order);
extern void post_alloc_hook(struct page *page, unsigned int order, gfp_t gfp_flags); -extern bool free_pages_prepare(struct page *page, unsigned int order, - fpi_t fpi_flags); -extern int check_new_page(struct page *page); -extern void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, - unsigned int alloc_flags); +#ifdef CONFIG_DYNAMIC_POOL +extern bool dpool_free_page_prepare(struct page *page); +extern int dpool_check_new_page(struct page *page); +extern void dpool_prep_new_page(struct page *page, unsigned int order, + gfp_t gfp_flags, unsigned int alloc_flags); +#endif extern int user_min_free_kbytes;
extern void free_unref_page(struct page *page, unsigned int order); diff --git a/mm/page_alloc.c b/mm/page_alloc.c index 815b0c0212fd..10facd4d65ec 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -1080,7 +1080,7 @@ static void kernel_init_pages(struct page *page, int numpages) kasan_enable_current(); }
-__always_inline bool free_pages_prepare(struct page *page, +static __always_inline bool free_pages_prepare(struct page *page, unsigned int order, fpi_t fpi_flags) { int bad = 0; @@ -1433,7 +1433,7 @@ static void check_new_page_bad(struct page *page) /* * This page is about to be returned from the page allocator */ -int check_new_page(struct page *page) +static int check_new_page(struct page *page) { if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) @@ -1545,8 +1545,8 @@ inline void post_alloc_hook(struct page *page, unsigned int order, page_table_check_alloc(page, order); }
-void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, - unsigned int alloc_flags) +static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, + unsigned int alloc_flags) { post_alloc_hook(page, order, gfp_flags);
@@ -1565,6 +1565,27 @@ void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, clear_page_pfmemalloc(page); }
+#ifdef CONFIG_DYNAMIC_POOL +/* + * Wrap the core functions with dpool_ prefix to avoid to call them directly. + */ +bool dpool_free_page_prepare(struct page *page) +{ + return free_pages_prepare(page, 0, 0); +} + +int dpool_check_new_page(struct page *page) +{ + return check_new_page(page); +} + +void dpool_prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, + unsigned int alloc_flags) +{ + prep_new_page(page, order, gfp_flags, alloc_flags); +} +#endif + /* * Go through the free lists for the given migratetype and remove * the smallest available page from the freelists
反馈： 您发送到kernel@openeuler.org的补丁/补丁集，已成功转换为PR！ PR链接地址： https://gitee.com/openeuler/kernel/pulls/5661 邮件列表地址：https://mailweb.openeuler.org/hyperkitty/list/kernel@openeuler.org/message/S...
FeedBack: The patch(es) which you have sent to kernel@openeuler.org mailing list has been converted to a pull request successfully! Pull request link: https://gitee.com/openeuler/kernel/pulls/5661 Mailing list address: https://mailweb.openeuler.org/hyperkitty/list/kernel@openeuler.org/message/S...
-
Liu Shixin -
patchwork bot