add folio_headpage() macro
Subsystems (5)
cover add folio_headpage() macro
The standard idiom for getting head page of a given folio is '&folio->page'. It is efficient and safe even if the folio is NULL, because the offset of page field in folio is zero. However, it makes the code not that easy to understand at the first glance, especially the NULL safety. Also, sometimes people forget the idiom and use 'folio_page(folio, 0)' instead. To make it easier to read and remember, add a new macro function called 'folio_headpage()' with the NULL case explanation. Then, replace the 'folio_page(folio, 0)' calls with 'folio_headpage(folio)'.
SeongJae Park (3):
include/linux/page-flags: add folio_headpage()
mm: use folio_headpage() instead of folio_page() fs/ceph/addr: use folio_headpage() instead of folio_page()
fs/ceph/addr.c | 2 +- include/linux/page-flags.h | 8 ++++++++ mm/shmem.c | 4 ++-- mm/slab.c | 6 +++--- mm/slab_common.c | 4 ++-- mm/slub.c | 4 ++-- 6 files changed, 18 insertions(+), 10 deletions(-)
Signature · 1 line
2.25.1
PATCH 1/3 include/linux/page-flags: add folio_headpage()
The standard idiom for getting head page of a given folio is '&folio->page'. It is efficient and safe even if the folio is NULL, because the offset of page field in folio is zero. However, it makes the code not that easy to understand at the first glance, especially the NULL safety. Also, sometimes people forget the idiom and use 'folio_page(folio, 0)' instead. To make it easier to read and remember, add a new macro function called 'folio_headpage()' with the NULL case explanation.
--- include/linux/page-flags.h | 8 ++++++++ 1 file changed, 8 insertions(+)
include/linux/page-flags.h +8 -0
@@ -285,6 +285,14 @@ static inline unsigned long _compound_head(const struct page *page)*/ #define folio_page (folio, n) nth_page(&(folio)->page, n) +/** +* folio_headpage - Return the head page from a folio. +* @folio: The pointer to the folio. +* +* Return: The head page of the folio, or NULL if the folio is NULL. +*/ +#define folio_headpage (folio) (&(folio)->page) + static __always_inline int PageTail(struct page *page) {return READ_ONCE (page->compound_head) &1 ||page_is_fake_head (page);
Signature · 1 line
2.25.1
PATCH 2/3 mm: use folio_headpage() instead of folio_page()
In reply to SeongJae ParkSeveral code in mm is using 'folio_page(folio, 0)' for getting the head pages of folios. It's not the standard idiom and inefficient. Replace the calls to 'folio_headpage()'.
--- mm/shmem.c | 4 ++-- mm/slab.c | 6 +++--- mm/slab_common.c | 4 ++-- mm/slub.c | 4 ++-- 4 files changed, 9 insertions(+), 9 deletions(-)
mm/shmem.c +2 -2
@@ -3211,7 +3211,7 @@ static const char *shmem_get_link(struct dentry *dentry, folio =filemap_get_folio (inode->i_mapping,0 );if (!folio)return ERR_PTR (-ECHILD ); -if (PageHWPoison (folio_page (folio,0 )) || +if (PageHWPoison (folio_headpage (folio)) || !folio_test_uptodate (folio)) {folio_put (folio);return ERR_PTR (-ECHILD );
@@ -3222,7 +3222,7 @@ static const char *shmem_get_link(struct dentry *dentry,return ERR_PTR (error);if (!folio)return ERR_PTR (-ECHILD ); -if (PageHWPoison (folio_page (folio,0 ))) { +if (PageHWPoison (folio_headpage (folio))) {folio_unlock (folio);folio_put (folio);return ERR_PTR (-ECHILD );
mm/slab.c +3 -3
@@ -1373,7 +1373,7 @@ static struct slab *kmem_getpages(struct kmem_cache *cachep, gfp_t flags,/* Make the flag visible before any changes to folio->mapping */ smp_wmb ();/* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ -if (sk_memalloc_socks () &&page_is_pfmemalloc (folio_page (folio,0 ))) +if (sk_memalloc_socks () &&page_is_pfmemalloc (folio_headpage (folio)))slab_set_pfmemalloc (slab);return slab;
@@ -1389,7 +1389,7 @@ static void kmem_freepages(struct kmem_cache *cachep, struct slab *slab)BUG_ON (!folio_test_slab (folio));__slab_clear_pfmemalloc (slab); -page_mapcount_reset (folio_page (folio,0 )); +page_mapcount_reset (folio_headpage (folio)); folio->mapping =NULL ;/* Make the mapping reset visible before clearing the flag */ smp_wmb ();
@@ -1398,7 +1398,7 @@ static void kmem_freepages(struct kmem_cache *cachep, struct slab *slab)if (current->reclaim_state) current->reclaim_state->reclaimed_slab +=1 << order;unaccount_slab (slab, order, cachep); -__free_pages (folio_page (folio,0 ), order); +__free_pages (folio_headpage (folio), order); }static void kmem_rcu_free (struct rcu_head *head)
mm/slab_common.c +2 -2
@@ -939,9 +939,9 @@ void free_large_kmalloc(struct folio *folio, void *object)kasan_kfree_large (object);kmsan_kfree_large (object); -mod_lruvec_page_state (folio_page (folio,0 ),NR_SLAB_UNRECLAIMABLE_B , +mod_lruvec_page_state (folio_headpage (folio),NR_SLAB_UNRECLAIMABLE_B , -(PAGE_SIZE << order)); -__free_pages (folio_page (folio,0 ), order); +__free_pages (folio_headpage (folio), order); }static void *__kmalloc_large_node (size_t size,gfp_t flags,int node);
mm/slub.c +2 -2
@@ -1859,7 +1859,7 @@ static inline struct slab *alloc_slab_page(gfp_t flags, int node,__folio_set_slab (folio);/* Make the flag visible before any changes to folio->mapping */ smp_wmb (); -if (page_is_pfmemalloc (folio_page (folio,0 ))) +if (page_is_pfmemalloc (folio_headpage (folio)))slab_set_pfmemalloc (slab);return slab;
@@ -2066,7 +2066,7 @@ static void __free_slab(struct kmem_cache *s, struct slab *slab)if (current->reclaim_state) current->reclaim_state->reclaimed_slab += pages;unaccount_slab (slab, order, s); -__free_pages (folio_page (folio,0 ), order); +__free_pages (folio_headpage (folio), order); }static void rcu_free_slab (struct rcu_head *h)
Signature · 1 line
2.25.1
Re: [PATCH 2/3] mm: use folio_headpage() instead of folio_page()
mm/shmem.c static const char *shmem_get_link(struct dentry *dentry, · line 3214
Quoted 10 lines · SeongJae Park
mm/shmem.c +1
-1@@ -3211,7 +3211,7 @@ static const char *shmem_get_link(struct dentry *dentry, folio =filemap_get_folio (inode->i_mapping,0 );if (!folio)return ERR_PTR (-ECHILD ); - if (PageHWPoison(folio_page(folio, 0)) || + if (PageHWPoison(folio_headpage(folio)) ||
This is actually incorrect. We don't want the head page, we want the page at index 0. It's a subtle but important difference later on.
mm/shmem.c static const char *shmem_get_link(struct dentry *dentry, · line 3225
Quoted 6 lines
mm/shmem.c +1
-1@@ -3222,7 +3222,7 @@ static const char *shmem_get_link(struct dentry *dentry,return ERR_PTR (error);if (!folio)return ERR_PTR (-ECHILD ); -if (PageHWPoison (folio_page (folio,0 ))) { +if (PageHWPoison (folio_headpage (folio))) {
Same here.
mm/shmem.c static struct slab *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, · line 1376
Quoted 7 lines
mm/shmem.c +2
-1+++ b/mm/slab.c@@ -1373,7 +1373,7 @@ static struct slab *kmem_getpages(struct kmem_cache *cachep, gfp_t flags,/* Make the flag visible before any changes to folio->mapping */ smp_wmb ();/* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ -if (sk_memalloc_socks () &&page_is_pfmemalloc (folio_page (folio,0 ))) +if (sk_memalloc_socks () &&page_is_pfmemalloc (folio_headpage (folio)))
We should have a folio_is_pfmemalloc().
mm/shmem.c static void kmem_freepages(struct kmem_cache *cachep, struct slab *slab) · line 1392
Quoted 6 lines
mm/shmem.c +1
-1@@ -1389,7 +1389,7 @@ static void kmem_freepages(struct kmem_cache *cachep, struct slab *slab)BUG_ON (!folio_test_slab (folio));__slab_clear_pfmemalloc (slab); -page_mapcount_reset (folio_page (folio,0 )); +page_mapcount_reset (folio_headpage (folio));
This one should be &folio->page.
mm/shmem.c static void kmem_freepages(struct kmem_cache *cachep, struct slab *slab) · line 1401
Quoted 6 lines
mm/shmem.c +1
-1@@ -1398,7 +1398,7 @@ static void kmem_freepages(struct kmem_cache *cachep, struct slab *slab)if (current->reclaim_state) current->reclaim_state->reclaimed_slab +=1 << order;unaccount_slab (slab, order, cachep); -__free_pages (folio_page (folio,0 ), order); +__free_pages (folio_headpage (folio), order);
&folio->page.
mm/shmem.c void free_large_kmalloc(struct folio *folio, void *object) · line 943
Quoted 7 lines
mm/shmem.c +1
-1@@ -939,9 +939,9 @@ void free_large_kmalloc(struct folio *folio, void *object)kasan_kfree_large (object);kmsan_kfree_large (object); -mod_lruvec_page_state (folio_page (folio,0 ),NR_SLAB_UNRECLAIMABLE_B , +mod_lruvec_page_state (folio_headpage (folio),NR_SLAB_UNRECLAIMABLE_B , -(PAGE_SIZE << order));
lruvec_stat_mod_folio(folio, NR_SLAB_UNRECLAIMABLE_B, ...
mm/shmem.c void free_large_kmalloc(struct folio *folio, void *object) · line 944
Quoted 2 lines
mm/shmem.c +1
-1-__free_pages (folio_page (folio,0 ), order); +__free_pages (folio_headpage (folio), order);
&folio->page.
mm/shmem.c static inline struct slab *alloc_slab_page(gfp_t flags, int node, · line 1862
Quoted 7 lines
mm/shmem.c +2
-1+++ b/mm/slub.c@@ -1859,7 +1859,7 @@ static inline struct slab *alloc_slab_page(gfp_t flags, int node,__folio_set_slab (folio);/* Make the flag visible before any changes to folio->mapping */ smp_wmb (); -if (page_is_pfmemalloc (folio_page (folio,0 ))) +if (page_is_pfmemalloc (folio_headpage (folio)))
folio_is_pfmemalloc()
mm/shmem.c static void __free_slab(struct kmem_cache *s, struct slab *slab) · line 2069
Quoted 6 lines
mm/shmem.c +1
-1@@ -2066,7 +2066,7 @@ static void __free_slab(struct kmem_cache *s, struct slab *slab)if (current->reclaim_state) current->reclaim_state->reclaimed_slab += pages;unaccount_slab (slab, order, s); -__free_pages (folio_page (folio,0 ), order); +__free_pages (folio_headpage (folio), order);
&folio->page.
PATCH 3/3 fs/ceph/addr: use folio_headpage() instead of folio_page()
In reply to SeongJae ParkUsing 'folio_page(folio, 0)' for getting the head page of a folios is not the standard idiom and inefficient. Replace the call in fs/ceph/ to 'folio_headpage()'.
--- fs/ceph/addr.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
fs/ceph/addr.c +1 -1
@@ -1290,7 +1290,7 @@ static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsignedstruct ceph_inode_info *ci =ceph_inode (inode);struct ceph_snap_context *snapc; - snapc =ceph_find_incompatible (folio_page (*foliop,0 )); + snapc =ceph_find_incompatible (folio_headpage (*foliop));if (snapc) {int r;
Signature · 1 line
2.25.1
Re: [PATCH 3/3] fs/ceph/addr: use folio_headpage() instead of folio_page()
fs/ceph/addr.c static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned · line 1293
Quoted 2 lines · SeongJae Park
fs/ceph/addr.c +1
-1- snapc =ceph_find_incompatible (folio_page (*foliop,0 )); + snapc =ceph_find_incompatible (folio_headpage (*foliop));
ceph_find_incompatible() should take a folio.
Quoted 9 lines · SeongJae Park
The standard idiom for getting head page of a given folio is '
&folio->page'. It is efficient and safe even if the folio is NULL, because the offset of page field in folio is zero. However, it makes the code not that easy to understand at the first glance, especially the NULL safety. Also, sometimes people forget the idiom and use 'folio_page(folio, 0)' instead. To make it easier to read and remember, add a new macro function called 'folio_headpage()' with the NULL case explanation. Then, replace the 'folio_page(folio, 0)' calls with 'folio_headpage(folio)'.
No. Everywhere that uses &folio->page is a place that needs to be fixed. It shouldn't have a nice convenience macro. It should make you mildly uncomfortable.
Quoted 11 lines · On Fri, 6 Jan 2023 19:21:40 +0000 Matthew Wilcox <[email protected]> wrote:
Quoted 9 lines · SeongJae Park
The standard idiom for getting head page of a given folio is '
&folio->page'. It is efficient and safe even if the folio is NULL, because the offset of page field in folio is zero. However, it makes the code not that easy to understand at the first glance, especially the NULL safety. Also, sometimes people forget the idiom and use 'folio_page(folio, 0)' instead. To make it easier to read and remember, add a new macro function called 'folio_headpage()' with the NULL case explanation. Then, replace the 'folio_page(folio, 0)' calls with 'folio_headpage(folio)'.
Quoted 3 lines
No. Everywhere that uses
&folio->pageis a place that needs to be fixed. It shouldn't have a nice convenience macro. It should make you mildly uncomfortable.
It's true that it's just a mild uncomfortableness. I will respect your opinion here. Thanks for the input.
Thanks,
SJ
Sorry I didn't see the [1/3] and [2/3] patches in my inbox, it seems you didn't CCed the ceph-devel@ mail list.
Thanks
Quoted 24 lines · SeongJae Park
The standard idiom for getting head page of a given folio is '
&folio->page'. It is efficient and safe even if the folio is NULL, because the offset of page field in folio is zero. However, it makes the code not that easy to understand at the first glance, especially the NULL safety. Also, sometimes people forget the idiom and use 'folio_page(folio, 0)' instead. To make it easier to read and remember, add a new macro function called 'folio_headpage()' with the NULL case explanation. Then, replace the 'folio_page(folio, 0)' calls with 'folio_headpage(folio)'.SeongJae Park (3):
include/linux/page-flags: addfolio_headpage()
mm: usefolio_headpage()instead offolio_page()fs/ceph/addr: usefolio_headpage()instead offolio_page()fs/ceph/addr.c | 2 +- include/linux/page-flags.h | 8 ++++++++ mm/shmem.c | 4 ++-- mm/slab.c | 6 +++--- mm/slab_common.c | 4 ++-- mm/slub.c | 4 ++-- 6 files changed, 18 insertions(+), 10 deletions(-)
Signature · 6 lines
Best Regards, Xiubo Li (李秀波) Email: [email protected]/[email protected] Slack: @Xiubo Li