This patch enable MADV_FREE hint for madvise syscall, which have
been supported by other OSes. [PATCH 1] includes the details.
[1] support MADVISE_FREE for !THP page so if VM encounter
THP page in syscall context, it splits THP page.
[2-6] is to preparing to call madvise syscall without THP plitting
[7] enable THP page support for MADV_FREE.
* From v10
* Add Acked-by from arch stuff(x86, s390)
* Pagewalker based pagetable working - Kirill
* Fix try_to_unmap_one broken with hwpoison - Kirill
* Use VM_BUG_ON_PAGE in madvise_free_pmd - Kirill
* Fix pgtable-3level.h for arm - Steve
* From v9
* Add Acked-by - Rik
* Add THP page support - Kirill
* Rebased-on v3.16-rc3-mmotm-2014-07-02-15-07
* From v8
* Rebased-on v3.16-rc2-mmotm-2014-06-25-16-44
* From v7
* Rebased-on next-20140613
* From v6
* Remove page from swapcache in syscal time
* Move utility functions from memory.c to madvise.c - Johannes
* Rename untilify functtions - Johannes
* Remove unnecessary checks from vmscan.c - Johannes
* Rebased-on v3.15-rc5-mmotm-2014-05-16-16-56
* Drop Reviewe-by because there was some changes since then.
* From v5
* Fix PPC problem which don't flush TLB - Rik
* Remove unnecessary lazyfree_range stub function - Rik
* Rebased on v3.15-rc5
* From v4
* Add Reviewed-by: Zhang Yanfei
* Rebase on v3.15-rc1-mmotm-2014-04-15-16-14
* From v3
* Add "how to work part" in description - Zhang
* Add page_discardable utility function - Zhang
* Clean up
* From v2
* Remove forceful dirty marking of swap-readed page - Johannes
* Remove deactivation logic of lazyfreed page
* Rebased on 3.14
* Remove RFC tag
* From v1
* Use custom page table walker for madvise_free - Johannes
* Remove PG_lazypage flag - Johannes
* Do madvise_dontneed instead of madvise_freein swapless system
Minchan Kim (7):
[1] mm: support madvise(MADV_FREE)
[2] x86: add pmd_[dirty|mkclean] for THP
[3] sparc: add pmd_[dirty|mkclean] for THP
[4] powerpc: add pmd_[dirty|mkclean] for THP
[5] s390: add pmd_[dirty|mkclean] for THP
[6] ARM: add pmd_[dirty|mkclean] for THP
[7] mm: Don't split THP page when syscall is called
arch/arm/include/asm/pgtable-3level.h | 3 +
arch/arm64/include/asm/pgtable.h | 2 +
arch/powerpc/include/asm/pgtable-ppc64.h | 2 +
arch/s390/include/asm/pgtable.h | 12 +++
arch/sparc/include/asm/pgtable_64.h | 16 ++++
arch/x86/include/asm/pgtable.h | 10 ++
include/linux/rmap.h | 9 +-
include/linux/vm_event_item.h | 1 +
include/uapi/asm-generic/mman-common.h | 1 +
mm/madvise.c | 153 +++++++++++++++++++++++++++++++
mm/rmap.c | 41 ++++++++-
mm/vmscan.c | 64 +++++++++----
mm/vmstat.c | 1 +
13 files changed, 295 insertions(+), 20 deletions(-)
--
2.0.0
MADV_FREE needs pmd_dirty and pmd_mkclean for detecting recent
overwrite of the contents since MADV_FREE syscall is called for
THP page but for s390 pmds only referenced bit is available
because there is no free bit left in the pmd entry for the
software dirty bit so this patch adds dumb pmd_dirty which
returns always true by suggesting by Martin.
They finally find a solution in future.
http://marc.info/?l=linux-api&m=140440328820808&w=2
Cc: Martin Schwidefsky <[email protected]>
Cc: Heiko Carstens <[email protected]>
Cc: Dominik Dingel <[email protected]>
Cc: Christian Borntraeger <[email protected]>
Cc: [email protected]
Acked-by: Gerald Schaefer <[email protected]>
Signed-off-by: Minchan Kim <[email protected]>
---
arch/s390/include/asm/pgtable.h | 12 ++++++++++++
1 file changed, 12 insertions(+)
diff --git a/arch/s390/include/asm/pgtable.h b/arch/s390/include/asm/pgtable.h
index fcba5e03839f..9862fcb0592b 100644
--- a/arch/s390/include/asm/pgtable.h
+++ b/arch/s390/include/asm/pgtable.h
@@ -1586,6 +1586,18 @@ static inline pmd_t pmd_mkdirty(pmd_t pmd)
return pmd;
}
+static inline int pmd_dirty(pmd_t pmd)
+{
+ /* No dirty bit in the segment table entry */
+ return 1;
+}
+
+static inline pmd_t pmd_mkclean(pmd_t pmd)
+{
+ /* No dirty bit in the segment table entry */
+ return pmd;
+}
+
#define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
unsigned long address, pmd_t *pmdp)
--
2.0.0
MADV_FREE needs pmd_dirty and pmd_mkclean for detecting recent
overwrite of the contents since MADV_FREE syscall is called for
THP page.
This patch adds pmd_dirty and pmd_mkclean for THP page MADV_FREE
support.
Cc: "David S. Miller" <[email protected]>
Cc: [email protected]
Signed-off-by: Minchan Kim <[email protected]>
---
arch/sparc/include/asm/pgtable_64.h | 16 ++++++++++++++++
1 file changed, 16 insertions(+)
diff --git a/arch/sparc/include/asm/pgtable_64.h b/arch/sparc/include/asm/pgtable_64.h
index 3770bf5c6e1b..0a3e5fdfead2 100644
--- a/arch/sparc/include/asm/pgtable_64.h
+++ b/arch/sparc/include/asm/pgtable_64.h
@@ -666,6 +666,13 @@ static inline unsigned long pmd_young(pmd_t pmd)
return pte_young(pte);
}
+static inline int pmd_dirty(pmd_t pmd)
+{
+ pte_t pte = __pte(pmd_val(pmd));
+
+ return pte_dirty(pte);
+}
+
static inline unsigned long pmd_write(pmd_t pmd)
{
pte_t pte = __pte(pmd_val(pmd));
@@ -723,6 +730,15 @@ static inline pmd_t pmd_mkdirty(pmd_t pmd)
return __pmd(pte_val(pte));
}
+static inline pmd_mkclean(pmd_t pmd)
+{
+ pte_t pte = __pte(pmd_val(pmd));
+
+ pte = pte_mkclean(pte);
+
+ return __pmd(pte_val(pte));
+}
+
static inline pmd_t pmd_mkyoung(pmd_t pmd)
{
pte_t pte = __pte(pmd_val(pmd));
--
2.0.0
Linux doesn't have an ability to free pages lazy while other OS
already have been supported that named by madvise(MADV_FREE).
The gain is clear that kernel can discard freed pages rather than
swapping out or OOM if memory pressure happens.
Without memory pressure, freed pages would be reused by userspace
without another additional overhead(ex, page fault + allocation
+ zeroing).
How to work is following as.
When madvise syscall is called, VM clears dirty bit of ptes of
the range. If memory pressure happens, VM checks dirty bit of
page table and if it found still "clean", it means it's a
"lazyfree pages" so VM could discard the page instead of swapping out.
Once there was store operation for the page before VM peek a page
to reclaim, dirty bit is set so VM can swap out the page instead of
discarding.
Firstly, heavy users would be general allocators(ex, jemalloc,
tcmalloc and hope glibc supports it) and jemalloc/tcmalloc already
have supported the feature for other OS(ex, FreeBSD)
barrios@blaptop:~/benchmark/ebizzy$ lscpu
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 4
On-line CPU(s) list: 0-3
Thread(s) per core: 2
Core(s) per socket: 2
Socket(s): 1
NUMA node(s): 1
Vendor ID: GenuineIntel
CPU family: 6
Model: 42
Stepping: 7
CPU MHz: 2801.000
BogoMIPS: 5581.64
Virtualization: VT-x
L1d cache: 32K
L1i cache: 32K
L2 cache: 256K
L3 cache: 4096K
NUMA node0 CPU(s): 0-3
ebizzy benchmark(./ebizzy -S 10 -n 512)
vanilla-jemalloc MADV_free-jemalloc
1 thread
records: 10 records: 10
avg: 7682.10 avg: 15306.10
std: 62.35(0.81%) std: 347.99(2.27%)
max: 7770.00 max: 15622.00
min: 7598.00 min: 14772.00
2 thread
records: 10 records: 10
avg: 12747.50 avg: 24171.00
std: 792.06(6.21%) std: 895.18(3.70%)
max: 13337.00 max: 26023.00
min: 10535.00 min: 23152.00
4 thread
records: 10 records: 10
avg: 16474.60 avg: 33717.90
std: 1496.45(9.08%) std: 2008.97(5.96%)
max: 17877.00 max: 35958.00
min: 12224.00 min: 29565.00
8 thread
records: 10 records: 10
avg: 16778.50 avg: 33308.10
std: 825.53(4.92%) std: 1668.30(5.01%)
max: 17543.00 max: 36010.00
min: 14576.00 min: 29577.00
16 thread
records: 10 records: 10
avg: 20614.40 avg: 35516.30
std: 602.95(2.92%) std: 1283.65(3.61%)
max: 21753.00 max: 37178.00
min: 19605.00 min: 33217.00
32 thread
records: 10 records: 10
avg: 22771.70 avg: 36018.50
std: 598.94(2.63%) std: 1046.76(2.91%)
max: 24035.00 max: 37266.00
min: 22108.00 min: 34149.00
In summary, MADV_FREE is about 2 time faster than MADV_DONTNEED.
Cc: Michael Kerrisk <[email protected]>
Cc: Linux API <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Johannes Weiner <[email protected]>
Cc: KOSAKI Motohiro <[email protected]>
Cc: Mel Gorman <[email protected]>
Cc: Jason Evans <[email protected]>
Cc: Zhang Yanfei <[email protected]>
Acked-by: Rik van Riel <[email protected]>
Signed-off-by: Minchan Kim <[email protected]>
---
include/linux/rmap.h | 9 ++-
include/linux/vm_event_item.h | 1 +
include/uapi/asm-generic/mman-common.h | 1 +
mm/madvise.c | 135 +++++++++++++++++++++++++++++++++
mm/rmap.c | 42 +++++++++-
mm/vmscan.c | 40 ++++++++--
mm/vmstat.c | 1 +
7 files changed, 217 insertions(+), 12 deletions(-)
diff --git a/include/linux/rmap.h b/include/linux/rmap.h
index be574506e6a9..0ba377b97a38 100644
--- a/include/linux/rmap.h
+++ b/include/linux/rmap.h
@@ -75,6 +75,7 @@ enum ttu_flags {
TTU_UNMAP = 1, /* unmap mode */
TTU_MIGRATION = 2, /* migration mode */
TTU_MUNLOCK = 4, /* munlock mode */
+ TTU_FREE = 8, /* free mode */
TTU_IGNORE_MLOCK = (1 << 8), /* ignore mlock */
TTU_IGNORE_ACCESS = (1 << 9), /* don't age */
@@ -181,7 +182,8 @@ static inline void page_dup_rmap(struct page *page)
* Called from mm/vmscan.c to handle paging out
*/
int page_referenced(struct page *, int is_locked,
- struct mem_cgroup *memcg, unsigned long *vm_flags);
+ struct mem_cgroup *memcg, unsigned long *vm_flags,
+ int *is_dirty);
#define TTU_ACTION(x) ((x) & TTU_ACTION_MASK)
@@ -260,9 +262,12 @@ int rmap_walk(struct page *page, struct rmap_walk_control *rwc);
static inline int page_referenced(struct page *page, int is_locked,
struct mem_cgroup *memcg,
- unsigned long *vm_flags)
+ unsigned long *vm_flags,
+ int *is_pte_dirty)
{
*vm_flags = 0;
+ if (is_pte_dirty)
+ *is_pte_dirty = 0;
return 0;
}
diff --git a/include/linux/vm_event_item.h b/include/linux/vm_event_item.h
index ced92345c963..e2d3fb1e9814 100644
--- a/include/linux/vm_event_item.h
+++ b/include/linux/vm_event_item.h
@@ -25,6 +25,7 @@ enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
FOR_ALL_ZONES(PGALLOC),
PGFREE, PGACTIVATE, PGDEACTIVATE,
PGFAULT, PGMAJFAULT,
+ PGLAZYFREED,
FOR_ALL_ZONES(PGREFILL),
FOR_ALL_ZONES(PGSTEAL_KSWAPD),
FOR_ALL_ZONES(PGSTEAL_DIRECT),
diff --git a/include/uapi/asm-generic/mman-common.h b/include/uapi/asm-generic/mman-common.h
index ddc3b36f1046..7a94102b7a02 100644
--- a/include/uapi/asm-generic/mman-common.h
+++ b/include/uapi/asm-generic/mman-common.h
@@ -34,6 +34,7 @@
#define MADV_SEQUENTIAL 2 /* expect sequential page references */
#define MADV_WILLNEED 3 /* will need these pages */
#define MADV_DONTNEED 4 /* don't need these pages */
+#define MADV_FREE 5 /* free pages only if memory pressure */
/* common parameters: try to keep these consistent across architectures */
#define MADV_REMOVE 9 /* remove these pages & resources */
diff --git a/mm/madvise.c b/mm/madvise.c
index 0938b30da4ab..a6aa7d4c4e02 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -19,6 +19,14 @@
#include <linux/blkdev.h>
#include <linux/swap.h>
#include <linux/swapops.h>
+#include <linux/mmu_notifier.h>
+
+#include <asm/tlb.h>
+
+struct madvise_free_private {
+ struct vm_area_struct *vma;
+ struct mmu_gather *tlb;
+};
/*
* Any behaviour which results in changes to the vma->vm_flags needs to
@@ -31,6 +39,7 @@ static int madvise_need_mmap_write(int behavior)
case MADV_REMOVE:
case MADV_WILLNEED:
case MADV_DONTNEED:
+ case MADV_FREE:
return 0;
default:
/* be safe, default to 1. list exceptions explicitly */
@@ -251,6 +260,124 @@ static long madvise_willneed(struct vm_area_struct *vma,
return 0;
}
+static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
+ unsigned long end, struct mm_walk *walk)
+
+{
+ struct madvise_free_private *fp = walk->private;
+ struct mmu_gather *tlb = fp->tlb;
+ struct mm_struct *mm = tlb->mm;
+ struct vm_area_struct *vma = fp->vma;
+ spinlock_t *ptl;
+ pte_t *pte, ptent;
+ struct page *page;
+
+ split_huge_page_pmd(vma, addr, pmd);
+ if (pmd_trans_unstable(pmd))
+ return 0;
+
+ pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
+ arch_enter_lazy_mmu_mode();
+ for (; addr != end; pte++, addr += PAGE_SIZE) {
+ ptent = *pte;
+
+ if (pte_none(ptent))
+ continue;
+
+ if (!pte_present(ptent))
+ continue;
+
+ page = vm_normal_page(vma, addr, ptent);
+ if (page && PageSwapCache(page)) {
+ if (trylock_page(page)) {
+ if (try_to_free_swap(page))
+ ClearPageDirty(page);
+ unlock_page(page);
+ } else
+ continue;
+ }
+
+ /*
+ * Some of architecture(ex, PPC) don't update TLB
+ * with set_pte_at and tlb_remove_tlb_entry so for
+ * the portability, remap the pte with old|clean
+ * after pte clearing.
+ */
+ ptent = ptep_get_and_clear_full(mm, addr, pte,
+ tlb->fullmm);
+ ptent = pte_mkold(ptent);
+ ptent = pte_mkclean(ptent);
+ set_pte_at(mm, addr, pte, ptent);
+ tlb_remove_tlb_entry(tlb, pte, addr);
+ }
+ arch_leave_lazy_mmu_mode();
+ pte_unmap_unlock(pte - 1, ptl);
+ cond_resched();
+ return 0;
+}
+
+static void madvise_free_page_range(struct mmu_gather *tlb,
+ struct vm_area_struct *vma,
+ unsigned long addr, unsigned long end)
+{
+ struct madvise_free_private fp = {
+ .vma = vma,
+ .tlb = tlb,
+ };
+
+ struct mm_walk free_walk = {
+ .pmd_entry = madvise_free_pte_range,
+ .mm = vma->vm_mm,
+ .private = &fp,
+ };
+
+ BUG_ON(addr >= end);
+ tlb_start_vma(tlb, vma);
+ walk_page_range(addr, end, &free_walk);
+ tlb_end_vma(tlb, vma);
+}
+
+static int madvise_free_single_vma(struct vm_area_struct *vma,
+ unsigned long start_addr, unsigned long end_addr)
+{
+ unsigned long start, end;
+ struct mm_struct *mm = vma->vm_mm;
+ struct mmu_gather tlb;
+
+ if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP))
+ return -EINVAL;
+
+ /* MADV_FREE works for only anon vma at the moment */
+ if (vma->vm_file)
+ return -EINVAL;
+
+ start = max(vma->vm_start, start_addr);
+ if (start >= vma->vm_end)
+ return -EINVAL;
+ end = min(vma->vm_end, end_addr);
+ if (end <= vma->vm_start)
+ return -EINVAL;
+
+ lru_add_drain();
+ tlb_gather_mmu(&tlb, mm, start, end);
+ update_hiwater_rss(mm);
+
+ mmu_notifier_invalidate_range_start(mm, start, end);
+ madvise_free_page_range(&tlb, vma, start, end);
+ mmu_notifier_invalidate_range_end(mm, start, end);
+ tlb_finish_mmu(&tlb, start, end);
+
+ return 0;
+}
+
+static long madvise_free(struct vm_area_struct *vma,
+ struct vm_area_struct **prev,
+ unsigned long start, unsigned long end)
+{
+ *prev = vma;
+ return madvise_free_single_vma(vma, start, end);
+}
+
/*
* Application no longer needs these pages. If the pages are dirty,
* it's OK to just throw them away. The app will be more careful about
@@ -381,6 +508,13 @@ madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
return madvise_remove(vma, prev, start, end);
case MADV_WILLNEED:
return madvise_willneed(vma, prev, start, end);
+ case MADV_FREE:
+ /*
+ * XXX: In this implementation, MADV_FREE works like
+ * MADV_DONTNEED on swapless system or full swap.
+ */
+ if (get_nr_swap_pages() > 0)
+ return madvise_free(vma, prev, start, end);
case MADV_DONTNEED:
return madvise_dontneed(vma, prev, start, end);
default:
@@ -400,6 +534,7 @@ madvise_behavior_valid(int behavior)
case MADV_REMOVE:
case MADV_WILLNEED:
case MADV_DONTNEED:
+ case MADV_FREE:
#ifdef CONFIG_KSM
case MADV_MERGEABLE:
case MADV_UNMERGEABLE:
diff --git a/mm/rmap.c b/mm/rmap.c
index 7928ddd91b6e..a8e34596dc97 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -663,6 +663,7 @@ int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
}
struct page_referenced_arg {
+ int dirtied;
int mapcount;
int referenced;
unsigned long vm_flags;
@@ -677,6 +678,7 @@ static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
struct mm_struct *mm = vma->vm_mm;
spinlock_t *ptl;
int referenced = 0;
+ int dirty = 0;
struct page_referenced_arg *pra = arg;
if (unlikely(PageTransHuge(page))) {
@@ -700,6 +702,11 @@ static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
/* go ahead even if the pmd is pmd_trans_splitting() */
if (pmdp_clear_flush_young_notify(vma, address, pmd))
referenced++;
+
+ /*
+ * In this implmentation, MADV_FREE doesn't support THP free
+ */
+ dirty++;
spin_unlock(ptl);
} else {
pte_t *pte;
@@ -729,6 +736,10 @@ static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
if (likely(!(vma->vm_flags & VM_SEQ_READ)))
referenced++;
}
+
+ if (pte_dirty(*pte))
+ dirty++;
+
pte_unmap_unlock(pte, ptl);
}
@@ -737,6 +748,9 @@ static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
pra->vm_flags |= vma->vm_flags;
}
+ if (dirty)
+ pra->dirtied++;
+
pra->mapcount--;
if (!pra->mapcount)
return SWAP_SUCCESS; /* To break the loop */
@@ -761,6 +775,7 @@ static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg)
* @is_locked: caller holds lock on the page
* @memcg: target memory cgroup
* @vm_flags: collect encountered vma->vm_flags who actually referenced the page
+ * @is_pte_dirty: ptes which have marked dirty bit - used for lazyfree page
*
* Quick test_and_clear_referenced for all mappings to a page,
* returns the number of ptes which referenced the page.
@@ -768,7 +783,8 @@ static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg)
int page_referenced(struct page *page,
int is_locked,
struct mem_cgroup *memcg,
- unsigned long *vm_flags)
+ unsigned long *vm_flags,
+ int *is_pte_dirty)
{
int ret;
int we_locked = 0;
@@ -783,6 +799,9 @@ int page_referenced(struct page *page,
};
*vm_flags = 0;
+ if (is_pte_dirty)
+ *is_pte_dirty = 0;
+
if (!page_mapped(page))
return 0;
@@ -810,6 +829,9 @@ int page_referenced(struct page *page,
if (we_locked)
unlock_page(page);
+ if (is_pte_dirty)
+ *is_pte_dirty = pra.dirtied;
+
return pra.referenced;
}
@@ -1128,6 +1150,7 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
spinlock_t *ptl;
int ret = SWAP_AGAIN;
enum ttu_flags flags = (enum ttu_flags)arg;
+ int dirty = 0;
pte = page_check_address(page, mm, address, &ptl, 0);
if (!pte)
@@ -1157,7 +1180,8 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
pteval = ptep_clear_flush(vma, address, pte);
/* Move the dirty bit to the physical page now the pte is gone. */
- if (pte_dirty(pteval))
+ dirty = pte_dirty(pteval);
+ if (dirty)
set_page_dirty(page);
/* Update high watermark before we lower rss */
@@ -1186,6 +1210,19 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
swp_entry_t entry = { .val = page_private(page) };
pte_t swp_pte;
+ if (flags & TTU_FREE) {
+ VM_BUG_ON_PAGE(PageSwapCache(page), page);
+ if (dirty || PageDirty(page)) {
+ set_pte_at(mm, address, pte, pteval);
+ ret = SWAP_FAIL;
+ goto out_unmap;
+ } else {
+ /* It's a freeable page by MADV_FREE */
+ dec_mm_counter(mm, MM_ANONPAGES);
+ goto discard;
+ }
+ }
+
if (PageSwapCache(page)) {
/*
* Store the swap location in the pte.
@@ -1227,6 +1264,7 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
} else
dec_mm_counter(mm, MM_FILEPAGES);
+discard:
page_remove_rmap(page);
page_cache_release(page);
diff --git a/mm/vmscan.c b/mm/vmscan.c
index 6d24fd63b209..d88413ccadcc 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -707,13 +707,17 @@ enum page_references {
};
static enum page_references page_check_references(struct page *page,
- struct scan_control *sc)
+ struct scan_control *sc,
+ bool *freeable)
{
int referenced_ptes, referenced_page;
unsigned long vm_flags;
+ int pte_dirty;
+
+ VM_BUG_ON_PAGE(!PageLocked(page), page);
referenced_ptes = page_referenced(page, 1, sc->target_mem_cgroup,
- &vm_flags);
+ &vm_flags, &pte_dirty);
referenced_page = TestClearPageReferenced(page);
/*
@@ -754,6 +758,10 @@ static enum page_references page_check_references(struct page *page,
return PAGEREF_KEEP;
}
+ if (PageAnon(page) && !pte_dirty && !PageSwapCache(page) &&
+ !PageDirty(page))
+ *freeable = true;
+
/* Reclaim if clean, defer dirty pages to writeback */
if (referenced_page && !PageSwapBacked(page))
return PAGEREF_RECLAIM_CLEAN;
@@ -823,6 +831,7 @@ static unsigned long shrink_page_list(struct list_head *page_list,
int may_enter_fs;
enum page_references references = PAGEREF_RECLAIM_CLEAN;
bool dirty, writeback;
+ bool freeable = false;
cond_resched();
@@ -945,7 +954,8 @@ static unsigned long shrink_page_list(struct list_head *page_list,
}
if (!force_reclaim)
- references = page_check_references(page, sc);
+ references = page_check_references(page, sc,
+ &freeable);
switch (references) {
case PAGEREF_ACTIVATE:
@@ -961,7 +971,7 @@ static unsigned long shrink_page_list(struct list_head *page_list,
* Anonymous process memory has backing store?
* Try to allocate it some swap space here.
*/
- if (PageAnon(page) && !PageSwapCache(page)) {
+ if (PageAnon(page) && !PageSwapCache(page) && !freeable) {
if (!(sc->gfp_mask & __GFP_IO))
goto keep_locked;
if (!add_to_swap(page, page_list))
@@ -976,8 +986,9 @@ static unsigned long shrink_page_list(struct list_head *page_list,
* The page is mapped into the page tables of one or more
* processes. Try to unmap it here.
*/
- if (page_mapped(page) && mapping) {
- switch (try_to_unmap(page, ttu_flags)) {
+ if (page_mapped(page) && (mapping || freeable)) {
+ switch (try_to_unmap(page, ttu_flags |
+ (freeable ? TTU_FREE : 0))) {
case SWAP_FAIL:
goto activate_locked;
case SWAP_AGAIN:
@@ -985,7 +996,20 @@ static unsigned long shrink_page_list(struct list_head *page_list,
case SWAP_MLOCK:
goto cull_mlocked;
case SWAP_SUCCESS:
- ; /* try to free the page below */
+ /* try to free the page below */
+ if (!freeable)
+ break;
+ /*
+ * Freeable anon page doesn't have mapping
+ * due to skipping of swapcache so we free
+ * page in here rather than __remove_mapping.
+ */
+ VM_BUG_ON_PAGE(PageSwapCache(page), page);
+ if (!page_freeze_refs(page, 1))
+ goto keep_locked;
+ __clear_page_locked(page);
+ count_vm_event(PGLAZYFREED);
+ goto free_it;
}
}
@@ -1727,7 +1751,7 @@ static void shrink_active_list(unsigned long nr_to_scan,
}
if (page_referenced(page, 0, sc->target_mem_cgroup,
- &vm_flags)) {
+ &vm_flags, NULL)) {
nr_rotated += hpage_nr_pages(page);
/*
* Identify referenced, file-backed active pages and
diff --git a/mm/vmstat.c b/mm/vmstat.c
index eef6321c8470..da18337c6c66 100644
--- a/mm/vmstat.c
+++ b/mm/vmstat.c
@@ -794,6 +794,7 @@ const char * const vmstat_text[] = {
"pgfault",
"pgmajfault",
+ "pglazyfreed",
TEXTS_FOR_ZONES("pgrefill")
TEXTS_FOR_ZONES("pgsteal_kswapd")
--
2.0.0
We don't need to split THP page when MADV_FREE syscall is
called. It could be done when VM decide really frees it so
we could avoid unnecessary THP split.
Signed-off-by: Minchan Kim <[email protected]>
---
mm/madvise.c | 20 +++++++++++++++++++-
mm/rmap.c | 7 +++----
mm/vmscan.c | 28 ++++++++++++++++++----------
3 files changed, 40 insertions(+), 15 deletions(-)
diff --git a/mm/madvise.c b/mm/madvise.c
index a6aa7d4c4e02..77f13a99584c 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -272,7 +272,25 @@ static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
pte_t *pte, ptent;
struct page *page;
- split_huge_page_pmd(vma, addr, pmd);
+ if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
+ struct page *page;
+ pmd_t orig_pmd;
+
+ orig_pmd = pmdp_get_and_clear(mm, addr, pmd);
+
+ /* No hugepage in swapcache */
+ page = pmd_page(orig_pmd);
+ VM_BUG_ON_PAGE(PageSwapCache(page), page);
+
+ orig_pmd = pmd_mkold(orig_pmd);
+ orig_pmd = pmd_mkclean(orig_pmd);
+
+ set_pmd_at(mm, addr, pmd, orig_pmd);
+ tlb_remove_pmd_tlb_entry(tlb, pmd, addr);
+ spin_unlock(ptl);
+ return 0;
+ }
+
if (pmd_trans_unstable(pmd))
return 0;
diff --git a/mm/rmap.c b/mm/rmap.c
index a8e34596dc97..67e1c1859c1d 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -703,10 +703,9 @@ static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
if (pmdp_clear_flush_young_notify(vma, address, pmd))
referenced++;
- /*
- * In this implmentation, MADV_FREE doesn't support THP free
- */
- dirty++;
+ if (pmd_dirty(*pmd))
+ dirty++;
+
spin_unlock(ptl);
} else {
pte_t *pte;
diff --git a/mm/vmscan.c b/mm/vmscan.c
index d88413ccadcc..6557f0b36321 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -971,17 +971,25 @@ static unsigned long shrink_page_list(struct list_head *page_list,
* Anonymous process memory has backing store?
* Try to allocate it some swap space here.
*/
- if (PageAnon(page) && !PageSwapCache(page) && !freeable) {
- if (!(sc->gfp_mask & __GFP_IO))
- goto keep_locked;
- if (!add_to_swap(page, page_list))
- goto activate_locked;
- may_enter_fs = 1;
-
- /* Adding to swap updated mapping */
- mapping = page_mapping(page);
+ if (PageAnon(page) && !PageSwapCache(page)) {
+ if (!freeable) {
+ if (!(sc->gfp_mask & __GFP_IO))
+ goto keep_locked;
+ if (!add_to_swap(page, page_list))
+ goto activate_locked;
+ may_enter_fs = 1;
+ /* Adding to swap updated mapping */
+ mapping = page_mapping(page);
+ } else {
+ if (likely(!PageTransHuge(page)))
+ goto unmap;
+ /* try_to_unmap isn't aware of THP page */
+ if (unlikely(split_huge_page_to_list(page,
+ page_list)))
+ goto keep_locked;
+ }
}
-
+unmap:
/*
* The page is mapped into the page tables of one or more
* processes. Try to unmap it here.
--
2.0.0
MADV_FREE needs pmd_dirty and pmd_mkclean for detecting recent
overwrite of the contents since MADV_FREE syscall is called for
THP page.
This patch adds pmd_dirty and pmd_mkclean for THP page MADV_FREE
support.
Cc: Catalin Marinas <[email protected]>
Cc: Will Deacon <[email protected]>
Cc: Steve Capper <[email protected]>
Cc: Russell King <[email protected]>
Cc: [email protected]
Signed-off-by: Minchan Kim <[email protected]>
---
arch/arm/include/asm/pgtable-3level.h | 3 +++
arch/arm64/include/asm/pgtable.h | 2 ++
2 files changed, 5 insertions(+)
diff --git a/arch/arm/include/asm/pgtable-3level.h b/arch/arm/include/asm/pgtable-3level.h
index 85c60adc8b60..3a7bb8dc7d05 100644
--- a/arch/arm/include/asm/pgtable-3level.h
+++ b/arch/arm/include/asm/pgtable-3level.h
@@ -220,6 +220,8 @@ static inline pmd_t *pmd_offset(pud_t *pud, unsigned long addr)
#define pmd_trans_splitting(pmd) (pmd_val(pmd) & PMD_SECT_SPLITTING)
#endif
+#define pmd_dirty (pmd_val(pmd) & PMD_SECT_DIRTY)
+
#define PMD_BIT_FUNC(fn,op) \
static inline pmd_t pmd_##fn(pmd_t pmd) { pmd_val(pmd) op; return pmd; }
@@ -228,6 +230,7 @@ PMD_BIT_FUNC(mkold, &= ~PMD_SECT_AF);
PMD_BIT_FUNC(mksplitting, |= PMD_SECT_SPLITTING);
PMD_BIT_FUNC(mkwrite, &= ~PMD_SECT_RDONLY);
PMD_BIT_FUNC(mkdirty, |= PMD_SECT_DIRTY);
+PMD_BIT_FUNC(mkclean, &= ~PMD_SECT_DIRTY);
PMD_BIT_FUNC(mkyoung, |= PMD_SECT_AF);
#define pmd_mkhuge(pmd) (__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT))
diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h
index 579702086488..f3ec01cef04f 100644
--- a/arch/arm64/include/asm/pgtable.h
+++ b/arch/arm64/include/asm/pgtable.h
@@ -240,10 +240,12 @@ static inline pmd_t pte_pmd(pte_t pte)
#endif
#define pmd_young(pmd) pte_young(pmd_pte(pmd))
+#define pmd_dirty(pmd) pte_dirty(pmd_pte(pmd))
#define pmd_wrprotect(pmd) pte_pmd(pte_wrprotect(pmd_pte(pmd)))
#define pmd_mksplitting(pmd) pte_pmd(pte_mkspecial(pmd_pte(pmd)))
#define pmd_mkold(pmd) pte_pmd(pte_mkold(pmd_pte(pmd)))
#define pmd_mkwrite(pmd) pte_pmd(pte_mkwrite(pmd_pte(pmd)))
+#define pmd_mkclean(pmd) pte_pmd(pte_mkclean(pmd_pte(pmd)))
#define pmd_mkdirty(pmd) pte_pmd(pte_mkdirty(pmd_pte(pmd)))
#define pmd_mkyoung(pmd) pte_pmd(pte_mkyoung(pmd_pte(pmd)))
#define pmd_mknotpresent(pmd) (__pmd(pmd_val(pmd) & ~PMD_TYPE_MASK))
--
2.0.0
MADV_FREE needs pmd_dirty and pmd_mkclean for detecting recent
overwrite of the contents since MADV_FREE syscall is called for
THP page.
This patch adds pmd_dirty and pmd_mkclean for THP page MADV_FREE
support.
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: "Aneesh Kumar K.V" <[email protected]>
Cc: [email protected]
Signed-off-by: Minchan Kim <[email protected]>
---
arch/powerpc/include/asm/pgtable-ppc64.h | 2 ++
1 file changed, 2 insertions(+)
diff --git a/arch/powerpc/include/asm/pgtable-ppc64.h b/arch/powerpc/include/asm/pgtable-ppc64.h
index eb9261024f51..c9a4bbe8e179 100644
--- a/arch/powerpc/include/asm/pgtable-ppc64.h
+++ b/arch/powerpc/include/asm/pgtable-ppc64.h
@@ -468,9 +468,11 @@ static inline pte_t *pmdp_ptep(pmd_t *pmd)
#define pmd_pfn(pmd) pte_pfn(pmd_pte(pmd))
#define pmd_young(pmd) pte_young(pmd_pte(pmd))
+#define pmd_dirty(pmd) pte_dirty(pmd_pte(pmd))
#define pmd_mkold(pmd) pte_pmd(pte_mkold(pmd_pte(pmd)))
#define pmd_wrprotect(pmd) pte_pmd(pte_wrprotect(pmd_pte(pmd)))
#define pmd_mkdirty(pmd) pte_pmd(pte_mkdirty(pmd_pte(pmd)))
+#define pmd_mkclean(pmd) pte_pmd(pte_mkclean(pmd_pte(pmd)))
#define pmd_mkyoung(pmd) pte_pmd(pte_mkyoung(pmd_pte(pmd)))
#define pmd_mkwrite(pmd) pte_pmd(pte_mkwrite(pmd_pte(pmd)))
--
2.0.0
MADV_FREE needs pmd_dirty and pmd_mkclean for detecting recent
overwrite of the contents since MADV_FREE syscall is called for
THP page.
This patch adds pmd_dirty and pmd_mkclean for THP page MADV_FREE
support.
Cc: Thomas Gleixner <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
Cc: [email protected]
Acked-by: Kirill A. Shutemov <[email protected]>
Signed-off-by: Minchan Kim <[email protected]>
---
arch/x86/include/asm/pgtable.h | 10 ++++++++++
1 file changed, 10 insertions(+)
diff --git a/arch/x86/include/asm/pgtable.h b/arch/x86/include/asm/pgtable.h
index 0ec056012618..329865799653 100644
--- a/arch/x86/include/asm/pgtable.h
+++ b/arch/x86/include/asm/pgtable.h
@@ -104,6 +104,11 @@ static inline int pmd_young(pmd_t pmd)
return pmd_flags(pmd) & _PAGE_ACCESSED;
}
+static inline int pmd_dirty(pmd_t pmd)
+{
+ return pmd_flags(pmd) & _PAGE_DIRTY;
+}
+
static inline int pte_write(pte_t pte)
{
return pte_flags(pte) & _PAGE_RW;
@@ -267,6 +272,11 @@ static inline pmd_t pmd_mkold(pmd_t pmd)
return pmd_clear_flags(pmd, _PAGE_ACCESSED);
}
+static inline pmd_t pmd_mkclean(pmd_t pmd)
+{
+ return pmd_clear_flags(pmd, _PAGE_DIRTY);
+}
+
static inline pmd_t pmd_wrprotect(pmd_t pmd)
{
return pmd_clear_flags(pmd, _PAGE_RW);
--
2.0.0
On 07/08/2014 02:03 PM, Minchan Kim wrote:
> Linux doesn't have an ability to free pages lazy while other OS
> already have been supported that named by madvise(MADV_FREE).
>
> The gain is clear that kernel can discard freed pages rather than
> swapping out or OOM if memory pressure happens.
>
> Without memory pressure, freed pages would be reused by userspace
> without another additional overhead(ex, page fault + allocation
> + zeroing).
>
> How to work is following as.
>
> When madvise syscall is called, VM clears dirty bit of ptes of
> the range. If memory pressure happens, VM checks dirty bit of
> page table and if it found still "clean", it means it's a
> "lazyfree pages" so VM could discard the page instead of swapping out.
> Once there was store operation for the page before VM peek a page
> to reclaim, dirty bit is set so VM can swap out the page instead of
> discarding.
>
> Firstly, heavy users would be general allocators(ex, jemalloc,
> tcmalloc and hope glibc supports it) and jemalloc/tcmalloc already
> have supported the feature for other OS(ex, FreeBSD)
>
> barrios@blaptop:~/benchmark/ebizzy$ lscpu
> Architecture: x86_64
> CPU op-mode(s): 32-bit, 64-bit
> Byte Order: Little Endian
> CPU(s): 4
> On-line CPU(s) list: 0-3
> Thread(s) per core: 2
> Core(s) per socket: 2
> Socket(s): 1
> NUMA node(s): 1
> Vendor ID: GenuineIntel
> CPU family: 6
> Model: 42
> Stepping: 7
> CPU MHz: 2801.000
> BogoMIPS: 5581.64
> Virtualization: VT-x
> L1d cache: 32K
> L1i cache: 32K
> L2 cache: 256K
> L3 cache: 4096K
> NUMA node0 CPU(s): 0-3
>
> ebizzy benchmark(./ebizzy -S 10 -n 512)
>
> vanilla-jemalloc MADV_free-jemalloc
>
> 1 thread
> records: 10 records: 10
> avg: 7682.10 avg: 15306.10
> std: 62.35(0.81%) std: 347.99(2.27%)
> max: 7770.00 max: 15622.00
> min: 7598.00 min: 14772.00
>
> 2 thread
> records: 10 records: 10
> avg: 12747.50 avg: 24171.00
> std: 792.06(6.21%) std: 895.18(3.70%)
> max: 13337.00 max: 26023.00
> min: 10535.00 min: 23152.00
>
> 4 thread
> records: 10 records: 10
> avg: 16474.60 avg: 33717.90
> std: 1496.45(9.08%) std: 2008.97(5.96%)
> max: 17877.00 max: 35958.00
> min: 12224.00 min: 29565.00
>
> 8 thread
> records: 10 records: 10
> avg: 16778.50 avg: 33308.10
> std: 825.53(4.92%) std: 1668.30(5.01%)
> max: 17543.00 max: 36010.00
> min: 14576.00 min: 29577.00
>
> 16 thread
> records: 10 records: 10
> avg: 20614.40 avg: 35516.30
> std: 602.95(2.92%) std: 1283.65(3.61%)
> max: 21753.00 max: 37178.00
> min: 19605.00 min: 33217.00
>
> 32 thread
> records: 10 records: 10
> avg: 22771.70 avg: 36018.50
> std: 598.94(2.63%) std: 1046.76(2.91%)
> max: 24035.00 max: 37266.00
> min: 22108.00 min: 34149.00
>
> In summary, MADV_FREE is about 2 time faster than MADV_DONTNEED.
>
> Cc: Michael Kerrisk <[email protected]>
> Cc: Linux API <[email protected]>
> Cc: Hugh Dickins <[email protected]>
> Cc: Johannes Weiner <[email protected]>
> Cc: KOSAKI Motohiro <[email protected]>
> Cc: Mel Gorman <[email protected]>
> Cc: Jason Evans <[email protected]>
> Cc: Zhang Yanfei <[email protected]>
> Acked-by: Rik van Riel <[email protected]>
> Signed-off-by: Minchan Kim <[email protected]>
A quick respin, looks good to me now for this !THP part. And
looks neat with the Pagewalker.
Acked-by: Zhang Yanfei <[email protected]>
> ---
> include/linux/rmap.h | 9 ++-
> include/linux/vm_event_item.h | 1 +
> include/uapi/asm-generic/mman-common.h | 1 +
> mm/madvise.c | 135 +++++++++++++++++++++++++++++++++
> mm/rmap.c | 42 +++++++++-
> mm/vmscan.c | 40 ++++++++--
> mm/vmstat.c | 1 +
> 7 files changed, 217 insertions(+), 12 deletions(-)
>
> diff --git a/include/linux/rmap.h b/include/linux/rmap.h
> index be574506e6a9..0ba377b97a38 100644
> --- a/include/linux/rmap.h
> +++ b/include/linux/rmap.h
> @@ -75,6 +75,7 @@ enum ttu_flags {
> TTU_UNMAP = 1, /* unmap mode */
> TTU_MIGRATION = 2, /* migration mode */
> TTU_MUNLOCK = 4, /* munlock mode */
> + TTU_FREE = 8, /* free mode */
>
> TTU_IGNORE_MLOCK = (1 << 8), /* ignore mlock */
> TTU_IGNORE_ACCESS = (1 << 9), /* don't age */
> @@ -181,7 +182,8 @@ static inline void page_dup_rmap(struct page *page)
> * Called from mm/vmscan.c to handle paging out
> */
> int page_referenced(struct page *, int is_locked,
> - struct mem_cgroup *memcg, unsigned long *vm_flags);
> + struct mem_cgroup *memcg, unsigned long *vm_flags,
> + int *is_dirty);
>
> #define TTU_ACTION(x) ((x) & TTU_ACTION_MASK)
>
> @@ -260,9 +262,12 @@ int rmap_walk(struct page *page, struct rmap_walk_control *rwc);
>
> static inline int page_referenced(struct page *page, int is_locked,
> struct mem_cgroup *memcg,
> - unsigned long *vm_flags)
> + unsigned long *vm_flags,
> + int *is_pte_dirty)
> {
> *vm_flags = 0;
> + if (is_pte_dirty)
> + *is_pte_dirty = 0;
> return 0;
> }
>
> diff --git a/include/linux/vm_event_item.h b/include/linux/vm_event_item.h
> index ced92345c963..e2d3fb1e9814 100644
> --- a/include/linux/vm_event_item.h
> +++ b/include/linux/vm_event_item.h
> @@ -25,6 +25,7 @@ enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
> FOR_ALL_ZONES(PGALLOC),
> PGFREE, PGACTIVATE, PGDEACTIVATE,
> PGFAULT, PGMAJFAULT,
> + PGLAZYFREED,
> FOR_ALL_ZONES(PGREFILL),
> FOR_ALL_ZONES(PGSTEAL_KSWAPD),
> FOR_ALL_ZONES(PGSTEAL_DIRECT),
> diff --git a/include/uapi/asm-generic/mman-common.h b/include/uapi/asm-generic/mman-common.h
> index ddc3b36f1046..7a94102b7a02 100644
> --- a/include/uapi/asm-generic/mman-common.h
> +++ b/include/uapi/asm-generic/mman-common.h
> @@ -34,6 +34,7 @@
> #define MADV_SEQUENTIAL 2 /* expect sequential page references */
> #define MADV_WILLNEED 3 /* will need these pages */
> #define MADV_DONTNEED 4 /* don't need these pages */
> +#define MADV_FREE 5 /* free pages only if memory pressure */
>
> /* common parameters: try to keep these consistent across architectures */
> #define MADV_REMOVE 9 /* remove these pages & resources */
> diff --git a/mm/madvise.c b/mm/madvise.c
> index 0938b30da4ab..a6aa7d4c4e02 100644
> --- a/mm/madvise.c
> +++ b/mm/madvise.c
> @@ -19,6 +19,14 @@
> #include <linux/blkdev.h>
> #include <linux/swap.h>
> #include <linux/swapops.h>
> +#include <linux/mmu_notifier.h>
> +
> +#include <asm/tlb.h>
> +
> +struct madvise_free_private {
> + struct vm_area_struct *vma;
> + struct mmu_gather *tlb;
> +};
>
> /*
> * Any behaviour which results in changes to the vma->vm_flags needs to
> @@ -31,6 +39,7 @@ static int madvise_need_mmap_write(int behavior)
> case MADV_REMOVE:
> case MADV_WILLNEED:
> case MADV_DONTNEED:
> + case MADV_FREE:
> return 0;
> default:
> /* be safe, default to 1. list exceptions explicitly */
> @@ -251,6 +260,124 @@ static long madvise_willneed(struct vm_area_struct *vma,
> return 0;
> }
>
> +static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
> + unsigned long end, struct mm_walk *walk)
> +
> +{
> + struct madvise_free_private *fp = walk->private;
> + struct mmu_gather *tlb = fp->tlb;
> + struct mm_struct *mm = tlb->mm;
> + struct vm_area_struct *vma = fp->vma;
> + spinlock_t *ptl;
> + pte_t *pte, ptent;
> + struct page *page;
> +
> + split_huge_page_pmd(vma, addr, pmd);
> + if (pmd_trans_unstable(pmd))
> + return 0;
> +
> + pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
> + arch_enter_lazy_mmu_mode();
> + for (; addr != end; pte++, addr += PAGE_SIZE) {
> + ptent = *pte;
> +
> + if (pte_none(ptent))
> + continue;
> +
> + if (!pte_present(ptent))
> + continue;
> +
> + page = vm_normal_page(vma, addr, ptent);
> + if (page && PageSwapCache(page)) {
> + if (trylock_page(page)) {
> + if (try_to_free_swap(page))
> + ClearPageDirty(page);
> + unlock_page(page);
> + } else
> + continue;
> + }
> +
> + /*
> + * Some of architecture(ex, PPC) don't update TLB
> + * with set_pte_at and tlb_remove_tlb_entry so for
> + * the portability, remap the pte with old|clean
> + * after pte clearing.
> + */
> + ptent = ptep_get_and_clear_full(mm, addr, pte,
> + tlb->fullmm);
> + ptent = pte_mkold(ptent);
> + ptent = pte_mkclean(ptent);
> + set_pte_at(mm, addr, pte, ptent);
> + tlb_remove_tlb_entry(tlb, pte, addr);
> + }
> + arch_leave_lazy_mmu_mode();
> + pte_unmap_unlock(pte - 1, ptl);
> + cond_resched();
> + return 0;
> +}
> +
> +static void madvise_free_page_range(struct mmu_gather *tlb,
> + struct vm_area_struct *vma,
> + unsigned long addr, unsigned long end)
> +{
> + struct madvise_free_private fp = {
> + .vma = vma,
> + .tlb = tlb,
> + };
> +
> + struct mm_walk free_walk = {
> + .pmd_entry = madvise_free_pte_range,
> + .mm = vma->vm_mm,
> + .private = &fp,
> + };
> +
> + BUG_ON(addr >= end);
> + tlb_start_vma(tlb, vma);
> + walk_page_range(addr, end, &free_walk);
> + tlb_end_vma(tlb, vma);
> +}
> +
> +static int madvise_free_single_vma(struct vm_area_struct *vma,
> + unsigned long start_addr, unsigned long end_addr)
> +{
> + unsigned long start, end;
> + struct mm_struct *mm = vma->vm_mm;
> + struct mmu_gather tlb;
> +
> + if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP))
> + return -EINVAL;
> +
> + /* MADV_FREE works for only anon vma at the moment */
> + if (vma->vm_file)
> + return -EINVAL;
> +
> + start = max(vma->vm_start, start_addr);
> + if (start >= vma->vm_end)
> + return -EINVAL;
> + end = min(vma->vm_end, end_addr);
> + if (end <= vma->vm_start)
> + return -EINVAL;
> +
> + lru_add_drain();
> + tlb_gather_mmu(&tlb, mm, start, end);
> + update_hiwater_rss(mm);
> +
> + mmu_notifier_invalidate_range_start(mm, start, end);
> + madvise_free_page_range(&tlb, vma, start, end);
> + mmu_notifier_invalidate_range_end(mm, start, end);
> + tlb_finish_mmu(&tlb, start, end);
> +
> + return 0;
> +}
> +
> +static long madvise_free(struct vm_area_struct *vma,
> + struct vm_area_struct **prev,
> + unsigned long start, unsigned long end)
> +{
> + *prev = vma;
> + return madvise_free_single_vma(vma, start, end);
> +}
> +
> /*
> * Application no longer needs these pages. If the pages are dirty,
> * it's OK to just throw them away. The app will be more careful about
> @@ -381,6 +508,13 @@ madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
> return madvise_remove(vma, prev, start, end);
> case MADV_WILLNEED:
> return madvise_willneed(vma, prev, start, end);
> + case MADV_FREE:
> + /*
> + * XXX: In this implementation, MADV_FREE works like
> + * MADV_DONTNEED on swapless system or full swap.
> + */
> + if (get_nr_swap_pages() > 0)
> + return madvise_free(vma, prev, start, end);
> case MADV_DONTNEED:
> return madvise_dontneed(vma, prev, start, end);
> default:
> @@ -400,6 +534,7 @@ madvise_behavior_valid(int behavior)
> case MADV_REMOVE:
> case MADV_WILLNEED:
> case MADV_DONTNEED:
> + case MADV_FREE:
> #ifdef CONFIG_KSM
> case MADV_MERGEABLE:
> case MADV_UNMERGEABLE:
> diff --git a/mm/rmap.c b/mm/rmap.c
> index 7928ddd91b6e..a8e34596dc97 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -663,6 +663,7 @@ int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
> }
>
> struct page_referenced_arg {
> + int dirtied;
> int mapcount;
> int referenced;
> unsigned long vm_flags;
> @@ -677,6 +678,7 @@ static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
> struct mm_struct *mm = vma->vm_mm;
> spinlock_t *ptl;
> int referenced = 0;
> + int dirty = 0;
> struct page_referenced_arg *pra = arg;
>
> if (unlikely(PageTransHuge(page))) {
> @@ -700,6 +702,11 @@ static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
> /* go ahead even if the pmd is pmd_trans_splitting() */
> if (pmdp_clear_flush_young_notify(vma, address, pmd))
> referenced++;
> +
> + /*
> + * In this implmentation, MADV_FREE doesn't support THP free
> + */
> + dirty++;
> spin_unlock(ptl);
> } else {
> pte_t *pte;
> @@ -729,6 +736,10 @@ static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
> if (likely(!(vma->vm_flags & VM_SEQ_READ)))
> referenced++;
> }
> +
> + if (pte_dirty(*pte))
> + dirty++;
> +
> pte_unmap_unlock(pte, ptl);
> }
>
> @@ -737,6 +748,9 @@ static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
> pra->vm_flags |= vma->vm_flags;
> }
>
> + if (dirty)
> + pra->dirtied++;
> +
> pra->mapcount--;
> if (!pra->mapcount)
> return SWAP_SUCCESS; /* To break the loop */
> @@ -761,6 +775,7 @@ static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg)
> * @is_locked: caller holds lock on the page
> * @memcg: target memory cgroup
> * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
> + * @is_pte_dirty: ptes which have marked dirty bit - used for lazyfree page
> *
> * Quick test_and_clear_referenced for all mappings to a page,
> * returns the number of ptes which referenced the page.
> @@ -768,7 +783,8 @@ static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg)
> int page_referenced(struct page *page,
> int is_locked,
> struct mem_cgroup *memcg,
> - unsigned long *vm_flags)
> + unsigned long *vm_flags,
> + int *is_pte_dirty)
> {
> int ret;
> int we_locked = 0;
> @@ -783,6 +799,9 @@ int page_referenced(struct page *page,
> };
>
> *vm_flags = 0;
> + if (is_pte_dirty)
> + *is_pte_dirty = 0;
> +
> if (!page_mapped(page))
> return 0;
>
> @@ -810,6 +829,9 @@ int page_referenced(struct page *page,
> if (we_locked)
> unlock_page(page);
>
> + if (is_pte_dirty)
> + *is_pte_dirty = pra.dirtied;
> +
> return pra.referenced;
> }
>
> @@ -1128,6 +1150,7 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
> spinlock_t *ptl;
> int ret = SWAP_AGAIN;
> enum ttu_flags flags = (enum ttu_flags)arg;
> + int dirty = 0;
>
> pte = page_check_address(page, mm, address, &ptl, 0);
> if (!pte)
> @@ -1157,7 +1180,8 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
> pteval = ptep_clear_flush(vma, address, pte);
>
> /* Move the dirty bit to the physical page now the pte is gone. */
> - if (pte_dirty(pteval))
> + dirty = pte_dirty(pteval);
> + if (dirty)
> set_page_dirty(page);
>
> /* Update high watermark before we lower rss */
> @@ -1186,6 +1210,19 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
> swp_entry_t entry = { .val = page_private(page) };
> pte_t swp_pte;
>
> + if (flags & TTU_FREE) {
> + VM_BUG_ON_PAGE(PageSwapCache(page), page);
> + if (dirty || PageDirty(page)) {
> + set_pte_at(mm, address, pte, pteval);
> + ret = SWAP_FAIL;
> + goto out_unmap;
> + } else {
> + /* It's a freeable page by MADV_FREE */
> + dec_mm_counter(mm, MM_ANONPAGES);
> + goto discard;
> + }
> + }
> +
> if (PageSwapCache(page)) {
> /*
> * Store the swap location in the pte.
> @@ -1227,6 +1264,7 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
> } else
> dec_mm_counter(mm, MM_FILEPAGES);
>
> +discard:
> page_remove_rmap(page);
> page_cache_release(page);
>
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index 6d24fd63b209..d88413ccadcc 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -707,13 +707,17 @@ enum page_references {
> };
>
> static enum page_references page_check_references(struct page *page,
> - struct scan_control *sc)
> + struct scan_control *sc,
> + bool *freeable)
> {
> int referenced_ptes, referenced_page;
> unsigned long vm_flags;
> + int pte_dirty;
> +
> + VM_BUG_ON_PAGE(!PageLocked(page), page);
>
> referenced_ptes = page_referenced(page, 1, sc->target_mem_cgroup,
> - &vm_flags);
> + &vm_flags, &pte_dirty);
> referenced_page = TestClearPageReferenced(page);
>
> /*
> @@ -754,6 +758,10 @@ static enum page_references page_check_references(struct page *page,
> return PAGEREF_KEEP;
> }
>
> + if (PageAnon(page) && !pte_dirty && !PageSwapCache(page) &&
> + !PageDirty(page))
> + *freeable = true;
> +
> /* Reclaim if clean, defer dirty pages to writeback */
> if (referenced_page && !PageSwapBacked(page))
> return PAGEREF_RECLAIM_CLEAN;
> @@ -823,6 +831,7 @@ static unsigned long shrink_page_list(struct list_head *page_list,
> int may_enter_fs;
> enum page_references references = PAGEREF_RECLAIM_CLEAN;
> bool dirty, writeback;
> + bool freeable = false;
>
> cond_resched();
>
> @@ -945,7 +954,8 @@ static unsigned long shrink_page_list(struct list_head *page_list,
> }
>
> if (!force_reclaim)
> - references = page_check_references(page, sc);
> + references = page_check_references(page, sc,
> + &freeable);
>
> switch (references) {
> case PAGEREF_ACTIVATE:
> @@ -961,7 +971,7 @@ static unsigned long shrink_page_list(struct list_head *page_list,
> * Anonymous process memory has backing store?
> * Try to allocate it some swap space here.
> */
> - if (PageAnon(page) && !PageSwapCache(page)) {
> + if (PageAnon(page) && !PageSwapCache(page) && !freeable) {
> if (!(sc->gfp_mask & __GFP_IO))
> goto keep_locked;
> if (!add_to_swap(page, page_list))
> @@ -976,8 +986,9 @@ static unsigned long shrink_page_list(struct list_head *page_list,
> * The page is mapped into the page tables of one or more
> * processes. Try to unmap it here.
> */
> - if (page_mapped(page) && mapping) {
> - switch (try_to_unmap(page, ttu_flags)) {
> + if (page_mapped(page) && (mapping || freeable)) {
> + switch (try_to_unmap(page, ttu_flags |
> + (freeable ? TTU_FREE : 0))) {
> case SWAP_FAIL:
> goto activate_locked;
> case SWAP_AGAIN:
> @@ -985,7 +996,20 @@ static unsigned long shrink_page_list(struct list_head *page_list,
> case SWAP_MLOCK:
> goto cull_mlocked;
> case SWAP_SUCCESS:
> - ; /* try to free the page below */
> + /* try to free the page below */
> + if (!freeable)
> + break;
> + /*
> + * Freeable anon page doesn't have mapping
> + * due to skipping of swapcache so we free
> + * page in here rather than __remove_mapping.
> + */
> + VM_BUG_ON_PAGE(PageSwapCache(page), page);
> + if (!page_freeze_refs(page, 1))
> + goto keep_locked;
> + __clear_page_locked(page);
> + count_vm_event(PGLAZYFREED);
> + goto free_it;
> }
> }
>
> @@ -1727,7 +1751,7 @@ static void shrink_active_list(unsigned long nr_to_scan,
> }
>
> if (page_referenced(page, 0, sc->target_mem_cgroup,
> - &vm_flags)) {
> + &vm_flags, NULL)) {
> nr_rotated += hpage_nr_pages(page);
> /*
> * Identify referenced, file-backed active pages and
> diff --git a/mm/vmstat.c b/mm/vmstat.c
> index eef6321c8470..da18337c6c66 100644
> --- a/mm/vmstat.c
> +++ b/mm/vmstat.c
> @@ -794,6 +794,7 @@ const char * const vmstat_text[] = {
>
> "pgfault",
> "pgmajfault",
> + "pglazyfreed",
>
> TEXTS_FOR_ZONES("pgrefill")
> TEXTS_FOR_ZONES("pgsteal_kswapd")
>
--
Thanks.
Zhang Yanfei
On 07/08/2014 02:03 PM, Minchan Kim wrote:
> MADV_FREE needs pmd_dirty and pmd_mkclean for detecting recent
> overwrite of the contents since MADV_FREE syscall is called for
> THP page.
>
> This patch adds pmd_dirty and pmd_mkclean for THP page MADV_FREE
> support.
>
> Cc: Thomas Gleixner <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> Cc: "H. Peter Anvin" <[email protected]>
> Cc: [email protected]
> Acked-by: Kirill A. Shutemov <[email protected]>
> Signed-off-by: Minchan Kim <[email protected]>
Acked-by: Zhang Yanfei <[email protected]>
> ---
> arch/x86/include/asm/pgtable.h | 10 ++++++++++
> 1 file changed, 10 insertions(+)
>
> diff --git a/arch/x86/include/asm/pgtable.h b/arch/x86/include/asm/pgtable.h
> index 0ec056012618..329865799653 100644
> --- a/arch/x86/include/asm/pgtable.h
> +++ b/arch/x86/include/asm/pgtable.h
> @@ -104,6 +104,11 @@ static inline int pmd_young(pmd_t pmd)
> return pmd_flags(pmd) & _PAGE_ACCESSED;
> }
>
> +static inline int pmd_dirty(pmd_t pmd)
> +{
> + return pmd_flags(pmd) & _PAGE_DIRTY;
> +}
> +
> static inline int pte_write(pte_t pte)
> {
> return pte_flags(pte) & _PAGE_RW;
> @@ -267,6 +272,11 @@ static inline pmd_t pmd_mkold(pmd_t pmd)
> return pmd_clear_flags(pmd, _PAGE_ACCESSED);
> }
>
> +static inline pmd_t pmd_mkclean(pmd_t pmd)
> +{
> + return pmd_clear_flags(pmd, _PAGE_DIRTY);
> +}
> +
> static inline pmd_t pmd_wrprotect(pmd_t pmd)
> {
> return pmd_clear_flags(pmd, _PAGE_RW);
>
--
Thanks.
Zhang Yanfei
On Tue, Jul 08, 2014 at 03:03:38PM +0900, Minchan Kim wrote:
> Linux doesn't have an ability to free pages lazy while other OS
> already have been supported that named by madvise(MADV_FREE).
>
> The gain is clear that kernel can discard freed pages rather than
> swapping out or OOM if memory pressure happens.
>
> Without memory pressure, freed pages would be reused by userspace
> without another additional overhead(ex, page fault + allocation
> + zeroing).
>
> How to work is following as.
>
> When madvise syscall is called, VM clears dirty bit of ptes of
> the range. If memory pressure happens, VM checks dirty bit of
> page table and if it found still "clean", it means it's a
> "lazyfree pages" so VM could discard the page instead of swapping out.
> Once there was store operation for the page before VM peek a page
> to reclaim, dirty bit is set so VM can swap out the page instead of
> discarding.
>
> Firstly, heavy users would be general allocators(ex, jemalloc,
> tcmalloc and hope glibc supports it) and jemalloc/tcmalloc already
> have supported the feature for other OS(ex, FreeBSD)
>
> barrios@blaptop:~/benchmark/ebizzy$ lscpu
> Architecture: x86_64
> CPU op-mode(s): 32-bit, 64-bit
> Byte Order: Little Endian
> CPU(s): 4
> On-line CPU(s) list: 0-3
> Thread(s) per core: 2
> Core(s) per socket: 2
> Socket(s): 1
> NUMA node(s): 1
> Vendor ID: GenuineIntel
> CPU family: 6
> Model: 42
> Stepping: 7
> CPU MHz: 2801.000
> BogoMIPS: 5581.64
> Virtualization: VT-x
> L1d cache: 32K
> L1i cache: 32K
> L2 cache: 256K
> L3 cache: 4096K
> NUMA node0 CPU(s): 0-3
>
> ebizzy benchmark(./ebizzy -S 10 -n 512)
>
> vanilla-jemalloc MADV_free-jemalloc
>
> 1 thread
> records: 10 records: 10
> avg: 7682.10 avg: 15306.10
> std: 62.35(0.81%) std: 347.99(2.27%)
> max: 7770.00 max: 15622.00
> min: 7598.00 min: 14772.00
>
> 2 thread
> records: 10 records: 10
> avg: 12747.50 avg: 24171.00
> std: 792.06(6.21%) std: 895.18(3.70%)
> max: 13337.00 max: 26023.00
> min: 10535.00 min: 23152.00
>
> 4 thread
> records: 10 records: 10
> avg: 16474.60 avg: 33717.90
> std: 1496.45(9.08%) std: 2008.97(5.96%)
> max: 17877.00 max: 35958.00
> min: 12224.00 min: 29565.00
>
> 8 thread
> records: 10 records: 10
> avg: 16778.50 avg: 33308.10
> std: 825.53(4.92%) std: 1668.30(5.01%)
> max: 17543.00 max: 36010.00
> min: 14576.00 min: 29577.00
>
> 16 thread
> records: 10 records: 10
> avg: 20614.40 avg: 35516.30
> std: 602.95(2.92%) std: 1283.65(3.61%)
> max: 21753.00 max: 37178.00
> min: 19605.00 min: 33217.00
>
> 32 thread
> records: 10 records: 10
> avg: 22771.70 avg: 36018.50
> std: 598.94(2.63%) std: 1046.76(2.91%)
> max: 24035.00 max: 37266.00
> min: 22108.00 min: 34149.00
>
> In summary, MADV_FREE is about 2 time faster than MADV_DONTNEED.
>
> Cc: Michael Kerrisk <[email protected]>
> Cc: Linux API <[email protected]>
> Cc: Hugh Dickins <[email protected]>
> Cc: Johannes Weiner <[email protected]>
> Cc: KOSAKI Motohiro <[email protected]>
> Cc: Mel Gorman <[email protected]>
> Cc: Jason Evans <[email protected]>
> Cc: Zhang Yanfei <[email protected]>
> Acked-by: Rik van Riel <[email protected]>
> Signed-off-by: Minchan Kim <[email protected]>
> ---
> include/linux/rmap.h | 9 ++-
> include/linux/vm_event_item.h | 1 +
> include/uapi/asm-generic/mman-common.h | 1 +
> mm/madvise.c | 135 +++++++++++++++++++++++++++++++++
> mm/rmap.c | 42 +++++++++-
> mm/vmscan.c | 40 ++++++++--
> mm/vmstat.c | 1 +
> 7 files changed, 217 insertions(+), 12 deletions(-)
>
...
> @@ -251,6 +260,124 @@ static long madvise_willneed(struct vm_area_struct *vma,
> return 0;
> }
>
> +static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
> + unsigned long end, struct mm_walk *walk)
> +
> +{
> + struct madvise_free_private *fp = walk->private;
> + struct mmu_gather *tlb = fp->tlb;
> + struct mm_struct *mm = tlb->mm;
> + struct vm_area_struct *vma = fp->vma;
> + spinlock_t *ptl;
> + pte_t *pte, ptent;
> + struct page *page;
> +
> + split_huge_page_pmd(vma, addr, pmd);
> + if (pmd_trans_unstable(pmd))
> + return 0;
> +
> + pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
> + arch_enter_lazy_mmu_mode();
> + for (; addr != end; pte++, addr += PAGE_SIZE) {
> + ptent = *pte;
> +
> + if (pte_none(ptent))
> + continue;
The check is redundant: all pte_none() entries are also !pte_present().
> +
> + if (!pte_present(ptent))
> + continue;
> +
> + page = vm_normal_page(vma, addr, ptent);
> + if (page && PageSwapCache(page)) {
> + if (trylock_page(page)) {
> + if (try_to_free_swap(page))
> + ClearPageDirty(page);
> + unlock_page(page);
> + } else
> + continue;
> + }
Is it safe to touch non-vm_normal entries? I would suggest to put
if (!page)
continue;
instead.
> + /*
> + * Some of architecture(ex, PPC) don't update TLB
> + * with set_pte_at and tlb_remove_tlb_entry so for
> + * the portability, remap the pte with old|clean
> + * after pte clearing.
> + */
> + ptent = ptep_get_and_clear_full(mm, addr, pte,
> + tlb->fullmm);
> + ptent = pte_mkold(ptent);
> + ptent = pte_mkclean(ptent);
> + set_pte_at(mm, addr, pte, ptent);
> + tlb_remove_tlb_entry(tlb, pte, addr);
> + }
> + arch_leave_lazy_mmu_mode();
> + pte_unmap_unlock(pte - 1, ptl);
> + cond_resched();
> + return 0;
> +}
> +
> +static void madvise_free_page_range(struct mmu_gather *tlb,
> + struct vm_area_struct *vma,
> + unsigned long addr, unsigned long end)
> +{
> + struct madvise_free_private fp = {
> + .vma = vma,
> + .tlb = tlb,
> + };
> +
> + struct mm_walk free_walk = {
> + .pmd_entry = madvise_free_pte_range,
> + .mm = vma->vm_mm,
> + .private = &fp,
> + };
> +
> + BUG_ON(addr >= end);
> + tlb_start_vma(tlb, vma);
> + walk_page_range(addr, end, &free_walk);
> + tlb_end_vma(tlb, vma);
> +}
> +
> +static int madvise_free_single_vma(struct vm_area_struct *vma,
> + unsigned long start_addr, unsigned long end_addr)
> +{
> + unsigned long start, end;
> + struct mm_struct *mm = vma->vm_mm;
> + struct mmu_gather tlb;
> +
> + if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP))
> + return -EINVAL;
VM_MIXEDMAP? VM_IO? Should it be whitelist instead?
> +
> + /* MADV_FREE works for only anon vma at the moment */
> + if (vma->vm_file)
> + return -EINVAL;
> +
> + start = max(vma->vm_start, start_addr);
> + if (start >= vma->vm_end)
> + return -EINVAL;
> + end = min(vma->vm_end, end_addr);
> + if (end <= vma->vm_start)
> + return -EINVAL;
> +
> + lru_add_drain();
> + tlb_gather_mmu(&tlb, mm, start, end);
> + update_hiwater_rss(mm);
> +
> + mmu_notifier_invalidate_range_start(mm, start, end);
> + madvise_free_page_range(&tlb, vma, start, end);
> + mmu_notifier_invalidate_range_end(mm, start, end);
> + tlb_finish_mmu(&tlb, start, end);
> +
> + return 0;
> +}
> +
> +static long madvise_free(struct vm_area_struct *vma,
> + struct vm_area_struct **prev,
> + unsigned long start, unsigned long end)
> +{
> + *prev = vma;
> + return madvise_free_single_vma(vma, start, end);
> +}
> +
> /*
> * Application no longer needs these pages. If the pages are dirty,
> * it's OK to just throw them away. The app will be more careful about
> @@ -381,6 +508,13 @@ madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
> return madvise_remove(vma, prev, start, end);
> case MADV_WILLNEED:
> return madvise_willneed(vma, prev, start, end);
> + case MADV_FREE:
> + /*
> + * XXX: In this implementation, MADV_FREE works like
> + * MADV_DONTNEED on swapless system or full swap.
> + */
> + if (get_nr_swap_pages() > 0)
> + return madvise_free(vma, prev, start, end);
/* passthough */
> case MADV_DONTNEED:
> return madvise_dontneed(vma, prev, start, end);
> default:
...
> @@ -1186,6 +1210,19 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
> swp_entry_t entry = { .val = page_private(page) };
> pte_t swp_pte;
>
> + if (flags & TTU_FREE) {
> + VM_BUG_ON_PAGE(PageSwapCache(page), page);
> + if (dirty || PageDirty(page)) {
> + set_pte_at(mm, address, pte, pteval);
> + ret = SWAP_FAIL;
> + goto out_unmap;
Hm. Again: do we really want stop here if caller asks for
TTU_FREE|TTU_UNMAP or should proceed?
> + } else {
> + /* It's a freeable page by MADV_FREE */
> + dec_mm_counter(mm, MM_ANONPAGES);
> + goto discard;
> + }
> + }
> +
> if (PageSwapCache(page)) {
> /*
> * Store the swap location in the pte.
--
Kirill A. Shutemov
On Tue, Jul 08, 2014 at 03:03:43PM +0900, Minchan Kim wrote:
> MADV_FREE needs pmd_dirty and pmd_mkclean for detecting recent
> overwrite of the contents since MADV_FREE syscall is called for
> THP page.
>
> This patch adds pmd_dirty and pmd_mkclean for THP page MADV_FREE
> support.
>
> Cc: Catalin Marinas <[email protected]>
> Cc: Will Deacon <[email protected]>
> Cc: Steve Capper <[email protected]>
> Cc: Russell King <[email protected]>
> Cc: [email protected]
> Signed-off-by: Minchan Kim <[email protected]>
> ---
> arch/arm/include/asm/pgtable-3level.h | 3 +++
> arch/arm64/include/asm/pgtable.h | 2 ++
> 2 files changed, 5 insertions(+)
Hi Minchan,
arch/arm and arch/arm64 are handled separately.
Could you please split this patch into arm and arm64 versions?
>
> diff --git a/arch/arm/include/asm/pgtable-3level.h b/arch/arm/include/asm/pgtable-3level.h
> index 85c60adc8b60..3a7bb8dc7d05 100644
> --- a/arch/arm/include/asm/pgtable-3level.h
> +++ b/arch/arm/include/asm/pgtable-3level.h
> @@ -220,6 +220,8 @@ static inline pmd_t *pmd_offset(pud_t *pud, unsigned long addr)
> #define pmd_trans_splitting(pmd) (pmd_val(pmd) & PMD_SECT_SPLITTING)
> #endif
>
> +#define pmd_dirty (pmd_val(pmd) & PMD_SECT_DIRTY)
This macro is missing the parameter definition, and will generate a
compile errror when CONFIG_ARM_LPAE=y
For 32-bit ARM with CONFIG_ARM_LPAE=n, we don't have THP support and I noticed
some compiler errors in mm/madvise.c:
CC mm/madvise.o
mm/madvise.c: In function ‘madvise_free_pte_range’:
mm/madvise.c:279:3: error: implicit declaration of function ‘pmdp_get_and_clear’ [-Werror=implicit-function-declaration]
orig_pmd = pmdp_get_and_clear(mm, addr, pmd);
^
mm/madvise.c:285:3: error: implicit declaration of function ‘pmd_mkold’ [-Werror=implicit-function-declaration]
orig_pmd = pmd_mkold(orig_pmd);
^
mm/madvise.c:286:3: error: implicit declaration of function ‘pmd_mkclean’ [-Werror=implicit-function-declaration]
orig_pmd = pmd_mkclean(orig_pmd);
^
mm/madvise.c:288:3: error: implicit declaration of function ‘set_pmd_at’ [-Werror=implicit-function-declaration]
set_pmd_at(mm, addr, pmd, orig_pmd);
^
cc1: some warnings being treated as errors
make[1]: *** [mm/madvise.o] Error 1
Cheers,
--
Steve
On Tue, Jul 08, 2014 at 12:41:14PM +0300, Kirill A. Shutemov wrote:
> On Tue, Jul 08, 2014 at 03:03:38PM +0900, Minchan Kim wrote:
> > Linux doesn't have an ability to free pages lazy while other OS
> > already have been supported that named by madvise(MADV_FREE).
> >
> > The gain is clear that kernel can discard freed pages rather than
> > swapping out or OOM if memory pressure happens.
> >
> > Without memory pressure, freed pages would be reused by userspace
> > without another additional overhead(ex, page fault + allocation
> > + zeroing).
> >
> > How to work is following as.
> >
> > When madvise syscall is called, VM clears dirty bit of ptes of
> > the range. If memory pressure happens, VM checks dirty bit of
> > page table and if it found still "clean", it means it's a
> > "lazyfree pages" so VM could discard the page instead of swapping out.
> > Once there was store operation for the page before VM peek a page
> > to reclaim, dirty bit is set so VM can swap out the page instead of
> > discarding.
> >
> > Firstly, heavy users would be general allocators(ex, jemalloc,
> > tcmalloc and hope glibc supports it) and jemalloc/tcmalloc already
> > have supported the feature for other OS(ex, FreeBSD)
> >
> > barrios@blaptop:~/benchmark/ebizzy$ lscpu
> > Architecture: x86_64
> > CPU op-mode(s): 32-bit, 64-bit
> > Byte Order: Little Endian
> > CPU(s): 4
> > On-line CPU(s) list: 0-3
> > Thread(s) per core: 2
> > Core(s) per socket: 2
> > Socket(s): 1
> > NUMA node(s): 1
> > Vendor ID: GenuineIntel
> > CPU family: 6
> > Model: 42
> > Stepping: 7
> > CPU MHz: 2801.000
> > BogoMIPS: 5581.64
> > Virtualization: VT-x
> > L1d cache: 32K
> > L1i cache: 32K
> > L2 cache: 256K
> > L3 cache: 4096K
> > NUMA node0 CPU(s): 0-3
> >
> > ebizzy benchmark(./ebizzy -S 10 -n 512)
> >
> > vanilla-jemalloc MADV_free-jemalloc
> >
> > 1 thread
> > records: 10 records: 10
> > avg: 7682.10 avg: 15306.10
> > std: 62.35(0.81%) std: 347.99(2.27%)
> > max: 7770.00 max: 15622.00
> > min: 7598.00 min: 14772.00
> >
> > 2 thread
> > records: 10 records: 10
> > avg: 12747.50 avg: 24171.00
> > std: 792.06(6.21%) std: 895.18(3.70%)
> > max: 13337.00 max: 26023.00
> > min: 10535.00 min: 23152.00
> >
> > 4 thread
> > records: 10 records: 10
> > avg: 16474.60 avg: 33717.90
> > std: 1496.45(9.08%) std: 2008.97(5.96%)
> > max: 17877.00 max: 35958.00
> > min: 12224.00 min: 29565.00
> >
> > 8 thread
> > records: 10 records: 10
> > avg: 16778.50 avg: 33308.10
> > std: 825.53(4.92%) std: 1668.30(5.01%)
> > max: 17543.00 max: 36010.00
> > min: 14576.00 min: 29577.00
> >
> > 16 thread
> > records: 10 records: 10
> > avg: 20614.40 avg: 35516.30
> > std: 602.95(2.92%) std: 1283.65(3.61%)
> > max: 21753.00 max: 37178.00
> > min: 19605.00 min: 33217.00
> >
> > 32 thread
> > records: 10 records: 10
> > avg: 22771.70 avg: 36018.50
> > std: 598.94(2.63%) std: 1046.76(2.91%)
> > max: 24035.00 max: 37266.00
> > min: 22108.00 min: 34149.00
> >
> > In summary, MADV_FREE is about 2 time faster than MADV_DONTNEED.
> >
> > Cc: Michael Kerrisk <[email protected]>
> > Cc: Linux API <[email protected]>
> > Cc: Hugh Dickins <[email protected]>
> > Cc: Johannes Weiner <[email protected]>
> > Cc: KOSAKI Motohiro <[email protected]>
> > Cc: Mel Gorman <[email protected]>
> > Cc: Jason Evans <[email protected]>
> > Cc: Zhang Yanfei <[email protected]>
> > Acked-by: Rik van Riel <[email protected]>
> > Signed-off-by: Minchan Kim <[email protected]>
> > ---
> > include/linux/rmap.h | 9 ++-
> > include/linux/vm_event_item.h | 1 +
> > include/uapi/asm-generic/mman-common.h | 1 +
> > mm/madvise.c | 135 +++++++++++++++++++++++++++++++++
> > mm/rmap.c | 42 +++++++++-
> > mm/vmscan.c | 40 ++++++++--
> > mm/vmstat.c | 1 +
> > 7 files changed, 217 insertions(+), 12 deletions(-)
> >
>
> ...
>
> > @@ -251,6 +260,124 @@ static long madvise_willneed(struct vm_area_struct *vma,
> > return 0;
> > }
> >
> > +static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
> > + unsigned long end, struct mm_walk *walk)
> > +
> > +{
> > + struct madvise_free_private *fp = walk->private;
> > + struct mmu_gather *tlb = fp->tlb;
> > + struct mm_struct *mm = tlb->mm;
> > + struct vm_area_struct *vma = fp->vma;
> > + spinlock_t *ptl;
> > + pte_t *pte, ptent;
> > + struct page *page;
> > +
> > + split_huge_page_pmd(vma, addr, pmd);
> > + if (pmd_trans_unstable(pmd))
> > + return 0;
> > +
> > + pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
> > + arch_enter_lazy_mmu_mode();
> > + for (; addr != end; pte++, addr += PAGE_SIZE) {
> > + ptent = *pte;
> > +
> > + if (pte_none(ptent))
> > + continue;
>
> The check is redundant: all pte_none() entries are also !pte_present().
True.
>
> > +
> > + if (!pte_present(ptent))
> > + continue;
> > +
> > + page = vm_normal_page(vma, addr, ptent);
> > + if (page && PageSwapCache(page)) {
> > + if (trylock_page(page)) {
> > + if (try_to_free_swap(page))
> > + ClearPageDirty(page);
> > + unlock_page(page);
> > + } else
> > + continue;
> > + }
>
> Is it safe to touch non-vm_normal entries? I would suggest to put
> if (!page)
> continue;
> instead.
That's right!
>
> > + /*
> > + * Some of architecture(ex, PPC) don't update TLB
> > + * with set_pte_at and tlb_remove_tlb_entry so for
> > + * the portability, remap the pte with old|clean
> > + * after pte clearing.
> > + */
> > + ptent = ptep_get_and_clear_full(mm, addr, pte,
> > + tlb->fullmm);
> > + ptent = pte_mkold(ptent);
> > + ptent = pte_mkclean(ptent);
> > + set_pte_at(mm, addr, pte, ptent);
> > + tlb_remove_tlb_entry(tlb, pte, addr);
> > + }
> > + arch_leave_lazy_mmu_mode();
> > + pte_unmap_unlock(pte - 1, ptl);
> > + cond_resched();
> > + return 0;
> > +}
> > +
> > +static void madvise_free_page_range(struct mmu_gather *tlb,
> > + struct vm_area_struct *vma,
> > + unsigned long addr, unsigned long end)
> > +{
> > + struct madvise_free_private fp = {
> > + .vma = vma,
> > + .tlb = tlb,
> > + };
> > +
> > + struct mm_walk free_walk = {
> > + .pmd_entry = madvise_free_pte_range,
> > + .mm = vma->vm_mm,
> > + .private = &fp,
> > + };
> > +
> > + BUG_ON(addr >= end);
> > + tlb_start_vma(tlb, vma);
> > + walk_page_range(addr, end, &free_walk);
> > + tlb_end_vma(tlb, vma);
> > +}
> > +
> > +static int madvise_free_single_vma(struct vm_area_struct *vma,
> > + unsigned long start_addr, unsigned long end_addr)
> > +{
> > + unsigned long start, end;
> > + struct mm_struct *mm = vma->vm_mm;
> > + struct mmu_gather tlb;
> > +
> > + if (vma->vm_flags & (VM_LOCKED|VM_HUGETLB|VM_PFNMAP))
> > + return -EINVAL;
>
> VM_MIXEDMAP? VM_IO? Should it be whitelist instead?
Don't they work with vma->vm_file?
so, below check will filter it out.
>
> > +
> > + /* MADV_FREE works for only anon vma at the moment */
> > + if (vma->vm_file)
> > + return -EINVAL;
> > +
> > + start = max(vma->vm_start, start_addr);
> > + if (start >= vma->vm_end)
> > + return -EINVAL;
> > + end = min(vma->vm_end, end_addr);
> > + if (end <= vma->vm_start)
> > + return -EINVAL;
> > +
> > + lru_add_drain();
> > + tlb_gather_mmu(&tlb, mm, start, end);
> > + update_hiwater_rss(mm);
> > +
> > + mmu_notifier_invalidate_range_start(mm, start, end);
> > + madvise_free_page_range(&tlb, vma, start, end);
> > + mmu_notifier_invalidate_range_end(mm, start, end);
> > + tlb_finish_mmu(&tlb, start, end);
> > +
> > + return 0;
> > +}
> > +
> > +static long madvise_free(struct vm_area_struct *vma,
> > + struct vm_area_struct **prev,
> > + unsigned long start, unsigned long end)
> > +{
> > + *prev = vma;
> > + return madvise_free_single_vma(vma, start, end);
> > +}
> > +
> > /*
> > * Application no longer needs these pages. If the pages are dirty,
> > * it's OK to just throw them away. The app will be more careful about
> > @@ -381,6 +508,13 @@ madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
> > return madvise_remove(vma, prev, start, end);
> > case MADV_WILLNEED:
> > return madvise_willneed(vma, prev, start, end);
> > + case MADV_FREE:
> > + /*
> > + * XXX: In this implementation, MADV_FREE works like
> > + * MADV_DONTNEED on swapless system or full swap.
> > + */
> > + if (get_nr_swap_pages() > 0)
> > + return madvise_free(vma, prev, start, end);
>
> /* passthough */
>
> > case MADV_DONTNEED:
> > return madvise_dontneed(vma, prev, start, end);
> > default:
>
> ...
>
> > @@ -1186,6 +1210,19 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
> > swp_entry_t entry = { .val = page_private(page) };
> > pte_t swp_pte;
> >
> > + if (flags & TTU_FREE) {
> > + VM_BUG_ON_PAGE(PageSwapCache(page), page);
> > + if (dirty || PageDirty(page)) {
> > + set_pte_at(mm, address, pte, pteval);
> > + ret = SWAP_FAIL;
> > + goto out_unmap;
>
> Hm. Again: do we really want stop here if caller asks for
> TTU_FREE|TTU_UNMAP or should proceed?
I'd like to stop.
If it is dirty in here, it means the page have been accessed during window
between page_check_references and try_to_unmap in shrink_page_list so
the page should be cycled one more time in LRU list without swapping.
But it's not a good idea to pass TTU_FREE|TTU_UNMAP together for redability
because people can think try_to_unmap will try both.
so I will modify it.
Thanks for the review!
>
> > + } else {
> > + /* It's a freeable page by MADV_FREE */
> > + dec_mm_counter(mm, MM_ANONPAGES);
> > + goto discard;
> > + }
> > + }
> > +
> > if (PageSwapCache(page)) {
> > /*
> > * Store the swap location in the pte.
> --
> Kirill A. Shutemov
>
> --
> 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>
--
Kind regards,
Minchan Kim
Hello Steve,
On Tue, Jul 08, 2014 at 12:37:38PM +0100, Steve Capper wrote:
> On Tue, Jul 08, 2014 at 03:03:43PM +0900, Minchan Kim wrote:
> > MADV_FREE needs pmd_dirty and pmd_mkclean for detecting recent
> > overwrite of the contents since MADV_FREE syscall is called for
> > THP page.
> >
> > This patch adds pmd_dirty and pmd_mkclean for THP page MADV_FREE
> > support.
> >
> > Cc: Catalin Marinas <[email protected]>
> > Cc: Will Deacon <[email protected]>
> > Cc: Steve Capper <[email protected]>
> > Cc: Russell King <[email protected]>
> > Cc: [email protected]
> > Signed-off-by: Minchan Kim <[email protected]>
> > ---
> > arch/arm/include/asm/pgtable-3level.h | 3 +++
> > arch/arm64/include/asm/pgtable.h | 2 ++
> > 2 files changed, 5 insertions(+)
>
> Hi Minchan,
> arch/arm and arch/arm64 are handled separately.
> Could you please split this patch into arm and arm64 versions?
>
Sure.
>
> >
> > diff --git a/arch/arm/include/asm/pgtable-3level.h b/arch/arm/include/asm/pgtable-3level.h
> > index 85c60adc8b60..3a7bb8dc7d05 100644
> > --- a/arch/arm/include/asm/pgtable-3level.h
> > +++ b/arch/arm/include/asm/pgtable-3level.h
> > @@ -220,6 +220,8 @@ static inline pmd_t *pmd_offset(pud_t *pud, unsigned long addr)
> > #define pmd_trans_splitting(pmd) (pmd_val(pmd) & PMD_SECT_SPLITTING)
> > #endif
> >
> > +#define pmd_dirty (pmd_val(pmd) & PMD_SECT_DIRTY)
>
> This macro is missing the parameter definition, and will generate a
> compile errror when CONFIG_ARM_LPAE=y
>
> For 32-bit ARM with CONFIG_ARM_LPAE=n, we don't have THP support and I noticed
> some compiler errors in mm/madvise.c:
>
> CC mm/madvise.o
> mm/madvise.c: In function ‘madvise_free_pte_range’:
> mm/madvise.c:279:3: error: implicit declaration of function ‘pmdp_get_and_clear’ [-Werror=implicit-function-declaration]
> orig_pmd = pmdp_get_and_clear(mm, addr, pmd);
> ^
> mm/madvise.c:285:3: error: implicit declaration of function ‘pmd_mkold’ [-Werror=implicit-function-declaration]
> orig_pmd = pmd_mkold(orig_pmd);
> ^
> mm/madvise.c:286:3: error: implicit declaration of function ‘pmd_mkclean’ [-Werror=implicit-function-declaration]
> orig_pmd = pmd_mkclean(orig_pmd);
> ^
> mm/madvise.c:288:3: error: implicit declaration of function ‘set_pmd_at’ [-Werror=implicit-function-declaration]
> set_pmd_at(mm, addr, pmd, orig_pmd);
> ^
> cc1: some warnings being treated as errors
> make[1]: *** [mm/madvise.o] Error 1
>
Oops, will fix.
Thanks for the review!
>
> Cheers,
> --
> Steve
>
> --
> 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>
--
Kind regards,
Minchan Kim