Skip to content

Commit

Permalink
tools/vm/page-types.c: support KPF_SOFTDIRTY bit
Browse files Browse the repository at this point in the history 
Soft dirty bit allows us to track which pages are written since the last
clear_ref (by "echo 4 > /proc/pid/clear_refs".) This is useful for
userspace applications to know their memory footprints.

Note that the kernel exposes this flag via bit[55] of /proc/pid/pagemap,
and the semantics is not a default one (scheduled to be the default in the
near future.) However, it shifts to the new semantics at the first
clear_ref, and the users of soft dirty bit always do it before utilizing
the bit, so that's not a big deal.  Users must avoid relying on the bit in
page-types before the first clear_ref.

Signed-off-by: Naoya Horiguchi <[email protected]>
Cc: Wu Fengguang <[email protected]>
Cc: Pavel Emelyanov <[email protected]>
Cc: Cyrill Gorcunov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
  • Loading branch information
@torvalds
Naoya Horiguchi authored and torvalds committed Nov 13, 2013
1 parent ec8e41a commit 46c77e2
Show file tree
Hide file tree
Showing 2 changed files with 21 additions and 12 deletions.
1 change: 1 addition & 0 deletions include/linux/kernel-page-flags.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -15,5 +15,6 @@
#define KPF_OWNER_PRIVATE 37
#define KPF_ARCH 38
#define KPF_UNCACHED 39
#define KPF_SOFTDIRTY 40

#endif /* LINUX_KERNEL_PAGE_FLAGS_H */
32 changes: 20 additions & 12 deletions tools/vm/page-types.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -59,12 +59,14 @@
#define PM_PSHIFT_BITS 6
#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
#define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
#define __PM_PSHIFT(x) (((uint64_t) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)

#define __PM_SOFT_DIRTY (1LL)
#define PM_PRESENT PM_STATUS(4LL)
#define PM_SWAP PM_STATUS(2LL)
#define PM_SOFT_DIRTY __PM_PSHIFT(__PM_SOFT_DIRTY)


/*
Expand All @@ -83,6 +85,7 @@
#define KPF_OWNER_PRIVATE 37
#define KPF_ARCH 38
#define KPF_UNCACHED 39
#define KPF_SOFTDIRTY 40

/* [48-] take some arbitrary free slots for expanding overloaded flags
* not part of kernel API
Expand Down Expand Up @@ -132,6 +135,7 @@ static const char * const page_flag_names[] = {
[KPF_OWNER_PRIVATE] = "O:owner_private",
[KPF_ARCH] = "h:arch",
[KPF_UNCACHED] = "c:uncached",
[KPF_SOFTDIRTY] = "f:softdirty",

[KPF_READAHEAD] = "I:readahead",
[KPF_SLOB_FREE] = "P:slob_free",
Expand Down Expand Up @@ -417,7 +421,7 @@ static int bit_mask_ok(uint64_t flags)
return 1;
}

static uint64_t expand_overloaded_flags(uint64_t flags)
static uint64_t expand_overloaded_flags(uint64_t flags, uint64_t pme)
{
/* SLOB/SLUB overload several page flags */
if (flags & BIT(SLAB)) {
Expand All @@ -433,6 +437,9 @@ static uint64_t expand_overloaded_flags(uint64_t flags)
if ((flags & (BIT(RECLAIM) | BIT(WRITEBACK))) == BIT(RECLAIM))
flags ^= BIT(RECLAIM) | BIT(READAHEAD);

if (pme & PM_SOFT_DIRTY)
flags |= BIT(SOFTDIRTY);

return flags;
}

Expand All @@ -448,11 +455,11 @@ static uint64_t well_known_flags(uint64_t flags)
return flags;
}

static uint64_t kpageflags_flags(uint64_t flags)
static uint64_t kpageflags_flags(uint64_t flags, uint64_t pme)
{
flags = expand_overloaded_flags(flags);

if (!opt_raw)
if (opt_raw)
flags = expand_overloaded_flags(flags, pme);
else
flags = well_known_flags(flags);

return flags;
Expand Down Expand Up @@ -545,9 +552,9 @@ static size_t hash_slot(uint64_t flags)
}

static void add_page(unsigned long voffset,
unsigned long offset, uint64_t flags)
unsigned long offset, uint64_t flags, uint64_t pme)
{
flags = kpageflags_flags(flags);
flags = kpageflags_flags(flags, pme);

if (!bit_mask_ok(flags))
return;
Expand All @@ -569,7 +576,8 @@ static void add_page(unsigned long voffset,
#define KPAGEFLAGS_BATCH (64 << 10) /* 64k pages */
static void walk_pfn(unsigned long voffset,
unsigned long index,
unsigned long count)
unsigned long count,
uint64_t pme)
{
uint64_t buf[KPAGEFLAGS_BATCH];
unsigned long batch;
Expand All @@ -583,7 +591,7 @@ static void walk_pfn(unsigned long voffset,
break;

for (i = 0; i < pages; i++)
add_page(voffset + i, index + i, buf[i]);
add_page(voffset + i, index + i, buf[i], pme);

index += pages;
count -= pages;
Expand All @@ -608,7 +616,7 @@ static void walk_vma(unsigned long index, unsigned long count)
for (i = 0; i < pages; i++) {
pfn = pagemap_pfn(buf[i]);
if (pfn)
walk_pfn(index + i, pfn, 1);
walk_pfn(index + i, pfn, 1, buf[i]);
}

index += pages;
Expand Down Expand Up @@ -659,7 +667,7 @@ static void walk_addr_ranges(void)

for (i = 0; i < nr_addr_ranges; i++)
if (!opt_pid)
walk_pfn(0, opt_offset[i], opt_size[i]);
walk_pfn(0, opt_offset[i], opt_size[i], 0);
else
walk_task(opt_offset[i], opt_size[i]);

Expand Down

0 comments on commit 46c77e2

Please sign in to comment.