From: Heesub Shin
mm: add cma reuse feature
Reference: https://lkml.org/lkml/2020/11/2/646
-------------------------------------
cma: redirect page allocation to CMA
CMA pages are designed to be used as fallback for movable allocations
and cannot be used for non-movable allocations. If CMA pages are
utilized poorly, non-movable allocations may end up getting starved if
all regular movable pages are allocated and the only pages left are
CMA. Always using CMA pages first creates unacceptable performance
problems. As a midway alternative, use CMA pages for certain
userspace allocations. The userspace pages can be migrated or dropped
quickly which giving decent utilization.
Signed-off-by: Kyungmin Park
Signed-off-by: Heesub Shin
Signed-off-by: Vinayak Menon
[cgoldswo@codeaurora.org: Place in bugfixes]
Signed-off-by: Chris Goldsworthy
Reported-by: kernel test robot
---
include/linux/gfp.h | 15 ++++++++
include/linux/highmem.h | 4 +-
include/linux/mmzone.h | 4 ++
mm/page_alloc.c | 84 +++++++++++++++++++++++++----------------
4 files changed, 74 insertions(+), 33 deletions(-)
diff --git a/include/linux/gfp.h b/include/linux/gfp.h
index c603237e0..e80b7d2f5 100644
--- a/include/linux/gfp.h
+++ b/include/linux/gfp.h
@@ -39,11 +39,21 @@ struct vm_area_struct;
#define ___GFP_HARDWALL 0x100000u
#define ___GFP_THISNODE 0x200000u
#define ___GFP_ACCOUNT 0x400000u
+#ifdef CONFIG_CMA
+#define ___GFP_CMA 0x800000u
+#else
+#define ___GFP_CMA 0
+#endif
#ifdef CONFIG_LOCKDEP
+#ifdef CONFIG_CMA
+#define ___GFP_NOLOCKDEP 0x1000000u
+#else
#define ___GFP_NOLOCKDEP 0x800000u
+#endif
#else
#define ___GFP_NOLOCKDEP 0
#endif
+
/* If the above are modified, __GFP_BITS_SHIFT may need updating */
/*
@@ -57,6 +67,7 @@ struct vm_area_struct;
#define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM)
#define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32)
#define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */
+#define __GFP_CMA ((__force gfp_t)___GFP_CMA)
#define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
/**
@@ -224,7 +235,11 @@ struct vm_area_struct;
#define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
/* Room for N __GFP_FOO bits */
+#ifdef CONFIG_CMA
+#define __GFP_BITS_SHIFT (24 + IS_ENABLED(CONFIG_LOCKDEP))
+#else
#define __GFP_BITS_SHIFT (23 + IS_ENABLED(CONFIG_LOCKDEP))
+#endif
#define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
/**
diff --git a/include/linux/highmem.h b/include/linux/highmem.h
index 14e6202ce..97241457d 100644
--- a/include/linux/highmem.h
+++ b/include/linux/highmem.h
@@ -274,7 +274,9 @@ static inline struct page *
alloc_zeroed_user_highpage_movable(struct vm_area_struct *vma,
unsigned long vaddr)
{
- return __alloc_zeroed_user_highpage(__GFP_MOVABLE, vma, vaddr);
+ return __alloc_zeroed_user_highpage(
+ __GFP_MOVABLE | __GFP_CMA, vma,
+ vaddr);
}
static inline void clear_highpage(struct page *page)
diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index 63b550403..b6bcef45b 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -451,6 +451,10 @@ struct zone {
struct pglist_data *zone_pgdat;
struct per_cpu_pageset __percpu *pageset;
+#ifdef CONFIG_CMA
+ bool cma_alloc;
+#endif
+
#ifndef CONFIG_SPARSEMEM
/*
* Flags for a pageblock_nr_pages block. See pageblock-flags.h.
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 6d189b69a..7250a0b59 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -2844,36 +2844,34 @@ __rmqueue(struct zone *zone, unsigned int order, int migratetype,
{
struct page *page;
- if (IS_ENABLED(CONFIG_CMA)) {
- /*
- * Balance movable allocations between regular and CMA areas by
- * allocating from CMA when over half of the zone's free memory
- * is in the CMA area.
- */
- if (alloc_flags & ALLOC_CMA &&
- zone_page_state(zone, NR_FREE_CMA_PAGES) >
- zone_page_state(zone, NR_FREE_PAGES) / 2) {
- page = __rmqueue_cma_fallback(zone, order);
- if (page)
- goto out;
- }
- }
retry:
page = __rmqueue_smallest(zone, order, migratetype);
- if (unlikely(!page)) {
- if (alloc_flags & ALLOC_CMA)
- page = __rmqueue_cma_fallback(zone, order);
-
- if (!page && __rmqueue_fallback(zone, order, migratetype,
- alloc_flags))
- goto retry;
- }
+ if (unlikely(!page) && __rmqueue_fallback(zone, order, migratetype,
+ alloc_flags))
+ goto retry;
out:
if (page)
trace_mm_page_alloc_zone_locked(page, order, migratetype);
return page;
}
+static struct page *__rmqueue_cma(struct zone *zone, unsigned int order,
+ int migratetype,
+ unsigned int alloc_flags)
+{
+ struct page *page = 0;
+
+#ifdef CONFIG_CMA
+ if (migratetype == MIGRATE_MOVABLE && !zone->cma_alloc)
+ page = __rmqueue_cma_fallback(zone, order);
+ else
+#endif
+ page = __rmqueue_smallest(zone, order, migratetype);
+
+ trace_mm_page_alloc_zone_locked(page, order, MIGRATE_CMA);
+ return page;
+}
+
/*
* Obtain a specified number of elements from the buddy allocator, all under
* a single hold of the lock, for efficiency. Add them to the supplied list.
@@ -2881,14 +2879,20 @@ __rmqueue(struct zone *zone, unsigned int order, int migratetype,
*/
static int rmqueue_bulk(struct zone *zone, unsigned int order,
unsigned long count, struct list_head *list,
- int migratetype, unsigned int alloc_flags)
+ int migratetype, unsigned int alloc_flags, int cma)
{
int i, alloced = 0;
spin_lock(&zone->lock);
for (i = 0; i < count; ++i) {
- struct page *page = __rmqueue(zone, order, migratetype,
- alloc_flags);
+ struct page *page = NULL;
+
+ if (cma)
+ page = __rmqueue_cma(zone, order, migratetype,
+ alloc_flags);
+ else
+ page = __rmqueue(zone, order, migratetype, alloc_flags);
+
if (unlikely(page == NULL))
break;
@@ -3374,7 +3378,8 @@ static inline void zone_statistics(struct zone *preferred_zone, struct zone *z)
static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype,
unsigned int alloc_flags,
struct per_cpu_pages *pcp,
- struct list_head *list)
+ struct list_head *list,
+ gfp_t gfp_flags)
{
struct page *page;
@@ -3382,7 +3387,8 @@ static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype,
if (list_empty(list)) {
pcp->count += rmqueue_bulk(zone, 0,
pcp->batch, list,
- migratetype, alloc_flags);
+ migratetype, alloc_flags,
+ gfp_flags && __GFP_CMA);
if (unlikely(list_empty(list)))
return NULL;
}
@@ -3408,7 +3414,8 @@ static struct page *rmqueue_pcplist(struct zone *preferred_zone,
local_irq_save(flags);
pcp = &this_cpu_ptr(zone->pageset)->pcp;
list = &pcp->lists[migratetype];
- page = __rmqueue_pcplist(zone, migratetype, alloc_flags, pcp, list);
+ page = __rmqueue_pcplist(zone, migratetype, alloc_flags, pcp, list,
+ gfp_flags);
if (page) {
__count_zid_vm_events(PGALLOC, page_zonenum(page), 1);
zone_statistics(preferred_zone, zone);
@@ -3434,7 +3441,7 @@ struct page *rmqueue(struct zone *preferred_zone,
* MIGRATE_MOVABLE pcplist could have the pages on CMA area and
* we need to skip it when CMA area isn't allowed.
*/
- if (!IS_ENABLED(CONFIG_CMA) || alloc_flags & ALLOC_CMA ||
+ if (!IS_ENABLED(CONFIG_CMA) || gfp_flags & __GFP_CMA ||
migratetype != MIGRATE_MOVABLE) {
page = rmqueue_pcplist(preferred_zone, zone, gfp_flags,
migratetype, alloc_flags);
@@ -3462,8 +3469,14 @@ struct page *rmqueue(struct zone *preferred_zone,
if (page)
trace_mm_page_alloc_zone_locked(page, order, migratetype);
}
- if (!page)
- page = __rmqueue(zone, order, migratetype, alloc_flags);
+ if (!page) {
+ if (gfp_flags & __GFP_CMA)
+ page = __rmqueue_cma(zone, order, migratetype,
+ alloc_flags);
+ else
+ page = __rmqueue(zone, order, migratetype,
+ alloc_flags);
+ }
} while (page && check_new_pages(page, order));
spin_unlock(&zone->lock);
if (!page)
@@ -3776,7 +3789,8 @@ static inline unsigned int current_alloc_flags(gfp_t gfp_mask,
unsigned int pflags = current->flags;
if (!(pflags & PF_MEMALLOC_NOCMA) &&
- gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE)
+ gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE &&
+ gfp_mask & __GFP_CMA)
alloc_flags |= ALLOC_CMA;
#endif
@@ -8533,6 +8547,9 @@ int alloc_contig_range(unsigned long start, unsigned long end,
if (ret)
return ret;
+#ifdef CONFIG_CMA
+ cc.zone->cma_alloc = 1;
+#endif
/*
* In case of -EBUSY, we'd like to know which page causes problem.
* So, just fall through. test_pages_isolated() has a tracepoint
@@ -8614,6 +8631,9 @@ int alloc_contig_range(unsigned long start, unsigned long end,
done:
undo_isolate_page_range(pfn_max_align_down(start),
pfn_max_align_up(end), migratetype);
+#ifdef CONFIG_CMA
+ cc.zone->cma_alloc = 0;
+#endif
return ret;
}
EXPORT_SYMBOL(alloc_contig_range);
--
2.25.1
