mm: swap_cgroup: allocate swap_cgroup map using vcalloc()
PATCH 1/2 mm: swap_cgroup: allocate swap_cgroup map using vcalloc()
Currently swap_cgroup's map is constructed as a vmalloc()'s-based array of pointers to individual struct pages. This brings an unnecessary complexity into the code.
This patch turns the swap_cgroup's map into a single space allocated by vcalloc().
--- mm/swap_cgroup.c | 83 ++++++++---------------------------------------- 1 file changed, 13 insertions(+), 70 deletions(-)
mm/swap_cgroup.c +13 -70
@@ -6,17 +6,18 @@#include <linux/swapops.h> /* depends on mm.h include */ static DEFINE_MUTEX (swap_cgroup_mutex); + +struct swap_cgroup { +unsigned short id; +}; +struct swap_cgroup_ctrl { -struct page **map; -unsigned long length; +struct swap_cgroup *map;spinlock_t lock; };static struct swap_cgroup_ctrl swap_cgroup_ctrl[MAX_SWAPFILES ]; -struct swap_cgroup { -unsigned short id; -};#define SC_PER_PAGE (PAGE_SIZE/sizeof(struct swap_cgroup))/*
@@ -33,44 +34,10 @@ struct swap_cgroup {* TODO: we can push these buffers out to HIGHMEM. */ -/* -* allocate buffer for swap_cgroup. -*/ -static int swap_cgroup_prepare (int type) -{ -struct page *page; -struct swap_cgroup_ctrl *ctrl; -unsigned long idx, max; - - ctrl = &swap_cgroup_ctrl[type]; - -for (idx =0 ; idx < ctrl->length; idx++) { - page =alloc_page (GFP_KERNEL | __GFP_ZERO); -if (!page) - goto not_enough_page; - ctrl->map[idx] = page; - -if (!(idx %SWAP_CLUSTER_MAX )) -cond_resched (); - } -return 0 ; -not_enough_page: - max = idx; -for (idx =0 ; idx < max; idx++) -__free_page (ctrl->map[idx]); - -return -ENOMEM ; -} -static struct swap_cgroup *__lookup_swap_cgroup (struct swap_cgroup_ctrl *ctrl,pgoff_t offset) { -struct page *mappage; -struct swap_cgroup *sc; - - mappage = ctrl->map[offset /SC_PER_PAGE ]; - sc =page_address (mappage); -return sc + offset %SC_PER_PAGE ; +return &ctrl->map[offset]; } static struct swap_cgroup *lookup_swap_cgroup(swp_entry_t ent,
@@ -168,32 +135,20 @@ unsigned short lookup_swap_cgroup_id(swp_entry_t ent)int swap_cgroup_swapon (int type,unsigned long max_pages) { -void *array; -unsigned long length; +struct swap_cgroup *map;struct swap_cgroup_ctrl *ctrl;if (mem_cgroup_disabled ())return 0 ; - length =DIV_ROUND_UP (max_pages,SC_PER_PAGE ); - - array =vcalloc (length,sizeof (void *)); -if (!array) + map =vcalloc (max_pages,sizeof (struct swap_cgroup )); +if (!map) goto nomem; ctrl = &swap_cgroup_ctrl[type];mutex_lock (&swap_cgroup_mutex); - ctrl->length = length; - ctrl->map = array; + ctrl->map = map;spin_lock_init (&ctrl->lock); -if (swap_cgroup_prepare (type)) { -/* memory shortage */ - ctrl->map =NULL ; - ctrl->length =0 ; -mutex_unlock (&swap_cgroup_mutex); -vfree (array); - goto nomem; - }mutex_unlock (&swap_cgroup_mutex);return 0 ;
@@ -205,8 +160,7 @@ int swap_cgroup_swapon(int type, unsigned long max_pages)void swap_cgroup_swapoff (int type) { -struct page **map; -unsigned long i, length; +struct swap_cgroup *map;struct swap_cgroup_ctrl *ctrl;if (mem_cgroup_disabled ())
@@ -215,19 +169,8 @@ void swap_cgroup_swapoff(int type)mutex_lock (&swap_cgroup_mutex); ctrl = &swap_cgroup_ctrl[type]; map = ctrl->map; - length = ctrl->length; ctrl->map =NULL ; - ctrl->length =0 ;mutex_unlock (&swap_cgroup_mutex); -if (map) { -for (i =0 ; i < length; i++) { -struct page *page = map[i]; -if (page) -__free_page (page); -if (!(i %SWAP_CLUSTER_MAX )) -cond_resched (); - } -vfree (map); - } +kvfree (map); }
Signature · 1 line
2.47.0.338.g60cca15819-goog
PATCH 2/2 mm: swap_cgroup: get rid of __lookup_swap_cgroup()
Because swap_cgroup map is now virtually contiguous, swap_cgroup_record() can be simplified, which eliminates a need to use __lookup_swap_cgroup().
Now as __lookup_swap_cgroup() is really trivial and is used only once, it can be inlined.
--- mm/swap_cgroup.c | 18 ++---------------- 1 file changed, 2 insertions(+), 16 deletions(-)
mm/swap_cgroup.c +2 -16
@@ -33,13 +33,6 @@ static struct swap_cgroup_ctrl swap_cgroup_ctrl[MAX_SWAPFILES];* * TODO: we can push these buffers out to HIGHMEM. */ - -static struct swap_cgroup *__lookup_swap_cgroup (struct swap_cgroup_ctrl *ctrl, -pgoff_t offset) -{ -return &ctrl->map[offset]; -} -static struct swap_cgroup *lookup_swap_cgroup (swp_entry_t ent,struct swap_cgroup_ctrl **ctrlp) {
@@ -49,7 +42,7 @@ static struct swap_cgroup *lookup_swap_cgroup(swp_entry_t ent, ctrl = &swap_cgroup_ctrl[swp_type (ent)];if (ctrlp) *ctrlp = ctrl; -return __lookup_swap_cgroup (ctrl, offset); +return &ctrl->map[offset]; }/**
@@ -104,16 +97,9 @@ unsigned short swap_cgroup_record(swp_entry_t ent, unsigned short id,spin_lock_irqsave (&ctrl->lock, flags); old = sc->id; -for (;;) { +for (; offset < end; offset++, sc++) {VM_BUG_ON (sc->id != old); sc->id = id; - offset++; -if (offset == end) -break ; -if (offset %SC_PER_PAGE ) - sc++; -else - sc =__lookup_swap_cgroup (ctrl, offset); }spin_unlock_irqrestore (&ctrl->lock, flags);
Signature · 1 line
2.47.0.338.g60cca15819-goog
Re: [PATCH 2/2] mm: swap_cgroup: get rid of __lookup_swap_cgroup()
Quoted 8 lines · Roman Gushchin
Because
swap_cgroupmap is now virtually contiguous,swap_cgroup_record()can be simplified, which eliminates a need to use__lookup_swap_cgroup().Now as
__lookup_swap_cgroup()is really trivial and is used only once, it can be inlined.Signed-off-by Roman Gushchin <[email protected]>
Quoted 8 lines · Roman Gushchin
Currently swap_cgroup's map is constructed as a vmalloc()'s-based array of pointers to individual struct pages. This brings an unnecessary complexity into the code.
This patch turns the swap_cgroup's map into a single space allocated by
vcalloc().Signed-off-by Roman Gushchin <[email protected]>
[...]
Quoted 10 lines
mm/swap_cgroup.c +0
-3@@ -215,19 +169,8 @@ void swap_cgroup_swapoff(int type)mutex_lock (&swap_cgroup_mutex); ctrl = &swap_cgroup_ctrl[type]; map = ctrl->map; - length = ctrl->length; ctrl->map =NULL ; - ctrl->length =0 ;mutex_unlock (&swap_cgroup_mutex); -if (map) {
mm/swap_cgroup.c void swap_cgroup_swapoff(int type) · line 175
Quoted 10 lines
mm/swap_cgroup.c +1
-9-for (i =0 ; i < length; i++) { -struct page *page = map[i]; -if (page) -__free_page (page); -if (!(i %SWAP_CLUSTER_MAX )) -cond_resched (); - } -vfree (map); - } +kvfree (map);
Any reason to use kvfree() instead of just vfree()?
Quoted 4 lines
(quoted diff) +0
-0}Signature · 1 line
2.47.0.338.g60cca15819-goog
Quoted 11 lines · Shakeel Butt
Quoted 8 lines · Roman Gushchin
Currently swap_cgroup's map is constructed as a vmalloc()'s-based array of pointers to individual struct pages. This brings an unnecessary complexity into the code.
This patch turns the swap_cgroup's map into a single space allocated by
vcalloc().Signed-off-by Roman Gushchin <[email protected]>Acked-by Shakeel Butt <[email protected]>
Thank you for reviewing the patchset!
Quoted 11 lines
[...]
Quoted 9 lines
mm/swap_cgroup.c +0
-2@@ -215,19 +169,8 @@ void swap_cgroup_swapoff(int type)mutex_lock (&swap_cgroup_mutex); ctrl = &swap_cgroup_ctrl[type]; map = ctrl->map; - length = ctrl->length; ctrl->map =NULL ; - ctrl->length =0 ;mutex_unlock (&swap_cgroup_mutex);
Quoted 13 lines
mm/swap_cgroup.c void swap_cgroup_swapoff(int type) · line 175
Quoted 11 lines
mm/swap_cgroup.c +1
-10-if (map) { -for (i =0 ; i < length; i++) { -struct page *page = map[i]; -if (page) -__free_page (page); -if (!(i %SWAP_CLUSTER_MAX )) -cond_resched (); - } -vfree (map); - } +kvfree (map);Any reason to use
kvfree()instead of justvfree()?
No, you're right, vfree() is a better choice here.
Andrew, do you prefer a v2 or it's easier to master a fix-up?
Thank you!