v1->v2:
- reworked so it only affects x86, since its the only arch I can build and test.
- fixed the fact that do_page_mkwrite wasn't actually sending ALLOW_RETRY down
to ->page_mkwrite.
- fixed error handling in do_page_mkwrite/callers to explicitly catch
VM_FAULT_RETRY.
- fixed btrfs to set ->cached_page properly.
This time I've verified that the ->page_mkwrite retry path is actually getting
used (apparently I only verified the read side last time). xfstests is still
running but it passed the couple of mmap tests I ran directly. Again this is an
RFC, I'm still doing a bunch of testing, but I'd appreciate comments on the
overall strategy.
-- Original message --
Now that we have proper isolation in place with cgroups2 we have started going
through and fixing the various priority inversions. Most are all gone now, but
this one is sort of weird since it's not necessarily a priority inversion that
happens within the kernel, but rather because of something userspace does.
We have giant applications that we want to protect, and parts of these giant
applications do things like watch the system state to determine how healthy the
box is for load balancing and such. This involves running 'ps' or other such
utilities. These utilities will often walk /proc/<pid>/whatever, and these
files can sometimes need to down_read(&task->mmap_sem). Not usually a big deal,
but we noticed when we are stress testing that sometimes our protected
application has latency spikes trying to get the mmap_sem for tasks that are in
lower priority cgroups.
This is because any down_write() on a semaphore essentially turns it into a
mutex, so even if we currently have it held for reading, any new readers will
not be allowed on to keep from starving the writer. This is fine, except a
lower priority task could be stuck doing IO because it has been throttled to the
point that its IO is taking much longer than normal. But because a higher
priority group depends on this completing it is now stuck behind lower priority
work.
In order to avoid this particular priority inversion we want to use the existing
retry mechanism to stop from holding the mmap_sem at all if we are going to do
IO. This already exists in the read case sort of, but needed to be extended for
more than just grabbing the page lock. With io.latency we throttle at
submit_bio() time, so the readahead stuff can block and even page_cache_read can
block, so all these paths need to have the mmap_sem dropped.
The other big thing is ->page_mkwrite. btrfs is particularly shitty here
because we have to reserve space for the dirty page, which can be a very
expensive operation. We use the same retry method as the read path, and simply
cache the page and verify the page is still setup properly the next pass through
->page_mkwrite().
I've tested these patches with xfstests and there are no regressions. Let me
know what you think. Thanks,
Josef
->page_mkwrite is extremely expensive in btrfs. We have to reserve
space, which can take 6 lifetimes, and we could possibly have to wait on
writeback on the page, another several lifetimes. To avoid this simply
drop the mmap_sem if we didn't have the cached page and do all of our
work and return the appropriate retry error. If we have the cached page
we know we did all the right things to set this page up and we can just
carry on.
Signed-off-by: Josef Bacik <[email protected]>
---
fs/btrfs/inode.c | 41 +++++++++++++++++++++++++++++++++++++++--
include/linux/mm.h | 14 ++++++++++++++
mm/filemap.c | 3 ++-
3 files changed, 55 insertions(+), 3 deletions(-)
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 3ea5339603cf..6b723d29bc0c 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -8809,7 +8809,9 @@ static void btrfs_invalidatepage(struct page *page, unsigned int offset,
vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
{
struct page *page = vmf->page;
- struct inode *inode = file_inode(vmf->vma->vm_file);
+ struct file *file = vmf->vma->vm_file, *fpin;
+ struct mm_struct *mm = vmf->vma->vm_mm;
+ struct inode *inode = file_inode(file);
struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
struct btrfs_ordered_extent *ordered;
@@ -8828,6 +8830,29 @@ vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
reserved_space = PAGE_SIZE;
+ /*
+ * We have our cached page from a previous mkwrite, check it to make
+ * sure it's still dirty and our file size matches when we ran mkwrite
+ * the last time. If everything is OK then return VM_FAULT_LOCKED,
+ * otherwise do the mkwrite again.
+ */
+ if (vmf->flags & FAULT_FLAG_USED_CACHED) {
+ lock_page(page);
+ if (vmf->cached_size == i_size_read(inode) &&
+ PageDirty(page))
+ return VM_FAULT_LOCKED;
+ unlock_page(page);
+ }
+
+ /*
+ * mkwrite is extremely expensive, and we are holding the mmap_sem
+ * during this, which means we can starve out anybody trying to
+ * down_write(mmap_sem) for a long while, especially if we throw cgroups
+ * into the mix. So just drop the mmap_sem and do all of our work,
+ * we'll loop back through and verify everything is ok the next time and
+ * hopefully avoid doing the work twice.
+ */
+ fpin = maybe_unlock_mmap_for_io(vmf->vma, vmf->flags);
sb_start_pagefault(inode->i_sb);
page_start = page_offset(page);
page_end = page_start + PAGE_SIZE - 1;
@@ -8844,7 +8869,7 @@ vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
reserved_space);
if (!ret2) {
- ret2 = file_update_time(vmf->vma->vm_file);
+ ret2 = file_update_time(file);
reserved = 1;
}
if (ret2) {
@@ -8943,6 +8968,14 @@ vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true);
sb_end_pagefault(inode->i_sb);
extent_changeset_free(data_reserved);
+ if (fpin) {
+ unlock_page(page);
+ fput(fpin);
+ get_page(page);
+ vmf->cached_size = size;
+ vmf->cached_page = page;
+ return VM_FAULT_RETRY;
+ }
return VM_FAULT_LOCKED;
}
@@ -8955,6 +8988,10 @@ vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
out_noreserve:
sb_end_pagefault(inode->i_sb);
extent_changeset_free(data_reserved);
+ if (fpin) {
+ fput(fpin);
+ down_read(&mm->mmap_sem);
+ }
return ret;
}
diff --git a/include/linux/mm.h b/include/linux/mm.h
index a7305d193c71..9409845d0411 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -370,6 +370,13 @@ struct vm_fault {
* next time we loop through the fault
* handler for faster lookup.
*/
+ loff_t cached_size; /* ->page_mkwrite handlers may drop
+ * the mmap_sem to avoid starvation, in
+ * which case they need to save the
+ * i_size in order to verify the cached
+ * page we're using the next loop
+ * through hasn't changed under us.
+ */
/* These three entries are valid only while holding ptl lock */
pte_t *pte; /* Pointer to pte entry matching
* the 'address'. NULL if the page
@@ -1437,6 +1444,8 @@ extern vm_fault_t handle_mm_fault_cacheable(struct vm_fault *vmf);
extern int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
unsigned long address, unsigned int fault_flags,
bool *unlocked);
+extern struct file *maybe_unlock_mmap_for_io(struct vm_area_struct *vma,
+ int flags);
void unmap_mapping_pages(struct address_space *mapping,
pgoff_t start, pgoff_t nr, bool even_cows);
void unmap_mapping_range(struct address_space *mapping,
@@ -1463,6 +1472,11 @@ static inline int fixup_user_fault(struct task_struct *tsk,
BUG();
return -EFAULT;
}
+stiatc inline struct file *maybe_unlock_mmap_for_io(struct vm_area_struct *vma,
+ int flags)
+{
+ return NULL;
+}
static inline void unmap_mapping_pages(struct address_space *mapping,
pgoff_t start, pgoff_t nr, bool even_cows) { }
static inline void unmap_mapping_range(struct address_space *mapping,
diff --git a/mm/filemap.c b/mm/filemap.c
index e9cb44bd35aa..8027f082d74f 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2366,7 +2366,7 @@ generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
EXPORT_SYMBOL(generic_file_read_iter);
#ifdef CONFIG_MMU
-static struct file *maybe_unlock_mmap_for_io(struct vm_area_struct *vma, int flags)
+struct file *maybe_unlock_mmap_for_io(struct vm_area_struct *vma, int flags)
{
if ((flags & (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT)) == FAULT_FLAG_ALLOW_RETRY) {
struct file *file;
@@ -2377,6 +2377,7 @@ static struct file *maybe_unlock_mmap_for_io(struct vm_area_struct *vma, int fla
}
return NULL;
}
+EXPORT_SYMBOL_GPL(maybe_unlock_mmap_for_io);
/**
* page_cache_read - adds requested page to the page cache if not already there
--
2.14.3
Before we didn't set the retry flag on our vm_fault. We want to allow
file systems to drop the mmap_sem if they so choose, so set this flag
and deal with VM_FAULT_RETRY appropriately.
Signed-off-by: Josef Bacik <[email protected]>
---
mm/memory.c | 10 +++++++---
1 file changed, 7 insertions(+), 3 deletions(-)
diff --git a/mm/memory.c b/mm/memory.c
index 6f8abde84986..821435855177 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -2384,11 +2384,13 @@ static vm_fault_t do_page_mkwrite(struct vm_fault *vmf)
unsigned int old_flags = vmf->flags;
vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
+ vmf->flags |= old_flags & FAULT_FLAG_ALLOW_RETRY;
ret = vmf->vma->vm_ops->page_mkwrite(vmf);
/* Restore original flags so that caller is not surprised */
vmf->flags = old_flags;
- if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
+ if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE |
+ VM_FAULT_RETRY)))
return ret;
if (unlikely(!(ret & VM_FAULT_LOCKED))) {
lock_page(page);
@@ -2683,7 +2685,8 @@ static vm_fault_t wp_page_shared(struct vm_fault *vmf)
pte_unmap_unlock(vmf->pte, vmf->ptl);
tmp = do_page_mkwrite(vmf);
if (unlikely(!tmp || (tmp &
- (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
+ (VM_FAULT_ERROR | VM_FAULT_NOPAGE |
+ VM_FAULT_RETRY)))) {
put_page(vmf->page);
return tmp;
}
@@ -3716,7 +3719,8 @@ static vm_fault_t do_shared_fault(struct vm_fault *vmf)
unlock_page(vmf->page);
tmp = do_page_mkwrite(vmf);
if (unlikely(!tmp ||
- (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
+ (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE |
+ VM_FAULT_RETRY)))) {
put_page(vmf->page);
return tmp;
}
--
2.14.3
This is preparation for dropping the mmap_sem in page_mkwrite. We need
to know if we used our cached page so we can be sure it is the page we
already did the page_mkwrite stuff on so we don't have to redo all of
that work.
Signed-off-by: Josef Bacik <[email protected]>
---
include/linux/mm.h | 6 +++++-
mm/filemap.c | 5 ++++-
2 files changed, 9 insertions(+), 2 deletions(-)
diff --git a/include/linux/mm.h b/include/linux/mm.h
index 4a84ec976dfc..a7305d193c71 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -318,6 +318,9 @@ extern pgprot_t protection_map[16];
#define FAULT_FLAG_USER 0x40 /* The fault originated in userspace */
#define FAULT_FLAG_REMOTE 0x80 /* faulting for non current tsk/mm */
#define FAULT_FLAG_INSTRUCTION 0x100 /* The fault was during an instruction fetch */
+#define FAULT_FLAG_USED_CACHED 0x200 /* Our vmf->page was from a previous
+ * loop through the fault handler.
+ */
#define FAULT_FLAG_TRACE \
{ FAULT_FLAG_WRITE, "WRITE" }, \
@@ -328,7 +331,8 @@ extern pgprot_t protection_map[16];
{ FAULT_FLAG_TRIED, "TRIED" }, \
{ FAULT_FLAG_USER, "USER" }, \
{ FAULT_FLAG_REMOTE, "REMOTE" }, \
- { FAULT_FLAG_INSTRUCTION, "INSTRUCTION" }
+ { FAULT_FLAG_INSTRUCTION, "INSTRUCTION" }, \
+ { FAULT_FLAG_USED_CACHED, "USED_CACHED" }
/*
* vm_fault is filled by the the pagefault handler and passed to the vma's
diff --git a/mm/filemap.c b/mm/filemap.c
index 5212ab637832..e9cb44bd35aa 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2556,6 +2556,7 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
if (cached_page->mapping == mapping &&
cached_page->index == offset) {
page = cached_page;
+ vmf->flags |= FAULT_FLAG_USED_CACHED;
goto have_cached_page;
}
unlock_page(cached_page);
@@ -2619,8 +2620,10 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
* We have a locked page in the page cache, now we need to check
* that it's up-to-date. If not, it is going to be due to an error.
*/
- if (unlikely(!PageUptodate(page)))
+ if (unlikely(!PageUptodate(page))) {
+ vmf->flags &= ~(FAULT_FLAG_USED_CACHED);
goto page_not_uptodate;
+ }
/*
* Found the page and have a reference on it.
--
2.14.3
If we drop the mmap_sem we have to redo the vma lookup which requires
redoing the fault handler. Chances are we will just come back to the
same page, so save this page in our vmf->cached_page and reuse it in the
next loop through the fault handler.
Signed-off-by: Josef Bacik <[email protected]>
---
mm/filemap.c | 30 ++++++++++++++++++++++++++++--
1 file changed, 28 insertions(+), 2 deletions(-)
diff --git a/mm/filemap.c b/mm/filemap.c
index 65395ee132a0..5212ab637832 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2530,13 +2530,38 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
pgoff_t offset = vmf->pgoff;
int flags = vmf->flags;
pgoff_t max_off;
- struct page *page;
+ struct page *page = NULL;
+ struct page *cached_page = vmf->cached_page;
vm_fault_t ret = 0;
max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
if (unlikely(offset >= max_off))
return VM_FAULT_SIGBUS;
+ /*
+ * We may have read in the page already and have a page from an earlier
+ * loop. If so we need to see if this page is still valid, and if not
+ * do the whole dance over again.
+ */
+ if (cached_page) {
+ if (flags & FAULT_FLAG_KILLABLE) {
+ error = lock_page_killable(cached_page);
+ if (error) {
+ up_read(&mm->mmap_sem);
+ goto out_retry;
+ }
+ } else
+ lock_page(cached_page);
+ vmf->cached_page = NULL;
+ if (cached_page->mapping == mapping &&
+ cached_page->index == offset) {
+ page = cached_page;
+ goto have_cached_page;
+ }
+ unlock_page(cached_page);
+ put_page(cached_page);
+ }
+
/*
* Do we have something in the page cache already?
*/
@@ -2587,6 +2612,7 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
put_page(page);
goto retry_find;
}
+have_cached_page:
VM_BUG_ON_PAGE(page->index != offset, page);
/*
@@ -2677,7 +2703,7 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
if (fpin)
fput(fpin);
if (page)
- put_page(page);
+ vmf->cached_page = page;
return ret | VM_FAULT_RETRY;
}
EXPORT_SYMBOL(filemap_fault);
--
2.14.3
From: Johannes Weiner <[email protected]>
We don't need to hold the mmap_sem while we're doing the IO, simply drop
it and retry appropriately.
Signed-off-by: Johannes Weiner <[email protected]>
Signed-off-by: Josef Bacik <[email protected]>
---
mm/page_io.c | 14 ++++++++++++++
1 file changed, 14 insertions(+)
diff --git a/mm/page_io.c b/mm/page_io.c
index aafd19ec1db4..bf21b56a964e 100644
--- a/mm/page_io.c
+++ b/mm/page_io.c
@@ -365,6 +365,20 @@ int swap_readpage(struct page *page, bool synchronous)
goto out;
}
+ /*
+ * XXX:
+ *
+ * Propagate mm->mmap_sem into this function. Then:
+ *
+ * get_file(sis->swap_file)
+ * up_read(mm->mmap_sem)
+ * submit io request
+ * fput
+ *
+ * After mmap_sem is dropped, sis is no longer valid. Go
+ * through swap_file->blah->bdev.
+ */
+
if (sis->flags & SWP_FILE) {
struct file *swap_file = sis->swap_file;
struct address_space *mapping = swap_file->f_mapping;
--
2.14.3
From: Johannes Weiner <[email protected]>
__read_swap_cache_async() has a misleading name. All it does is look
up or create a page in swapcache; it doesn't initiate any IO.
The swapcache has many parallels to the page cache, and shares naming
schemes with it elsewhere. Analogous to the cache lookup and creation
API, rename __read_swap_cache_async() find_or_create_swap_cache() and
lookup_swap_cache() to find_swap_cache().
Signed-off-by: Johannes Weiner <[email protected]>
Signed-off-by: Josef Bacik <[email protected]>
---
include/linux/swap.h | 14 ++++++++------
mm/memory.c | 2 +-
mm/shmem.c | 2 +-
mm/swap_state.c | 43 ++++++++++++++++++++++---------------------
mm/zswap.c | 8 ++++----
5 files changed, 36 insertions(+), 33 deletions(-)
diff --git a/include/linux/swap.h b/include/linux/swap.h
index 8e2c11e692ba..293a84c34448 100644
--- a/include/linux/swap.h
+++ b/include/linux/swap.h
@@ -412,15 +412,17 @@ extern void __delete_from_swap_cache(struct page *);
extern void delete_from_swap_cache(struct page *);
extern void free_page_and_swap_cache(struct page *);
extern void free_pages_and_swap_cache(struct page **, int);
-extern struct page *lookup_swap_cache(swp_entry_t entry,
- struct vm_area_struct *vma,
- unsigned long addr);
+extern struct page *find_swap_cache(swp_entry_t entry,
+ struct vm_area_struct *vma,
+ unsigned long addr);
+extern struct page *find_or_create_swap_cache(swp_entry_t entry,
+ gfp_t gfp_mask,
+ struct vm_area_struct *vma,
+ unsigned long addr,
+ bool *created);
extern struct page *read_swap_cache_async(swp_entry_t, gfp_t,
struct vm_area_struct *vma, unsigned long addr,
bool do_poll);
-extern struct page *__read_swap_cache_async(swp_entry_t, gfp_t,
- struct vm_area_struct *vma, unsigned long addr,
- bool *new_page_allocated);
extern struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t flag,
struct vm_fault *vmf);
extern struct page *swapin_readahead(swp_entry_t entry, gfp_t flag,
diff --git a/mm/memory.c b/mm/memory.c
index 433075f722ea..6f8abde84986 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -2935,7 +2935,7 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
delayacct_set_flag(DELAYACCT_PF_SWAPIN);
- page = lookup_swap_cache(entry, vma, vmf->address);
+ page = find_swap_cache(entry, vma, vmf->address);
swapcache = page;
if (!page) {
diff --git a/mm/shmem.c b/mm/shmem.c
index 0376c124b043..9854903ae92f 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1679,7 +1679,7 @@ static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
if (swap.val) {
/* Look it up and read it in.. */
- page = lookup_swap_cache(swap, NULL, 0);
+ page = find_swap_cache(swap, NULL, 0);
if (!page) {
/* Or update major stats only when swapin succeeds?? */
if (fault_type) {
diff --git a/mm/swap_state.c b/mm/swap_state.c
index ecee9c6c4cc1..bae758e19f7a 100644
--- a/mm/swap_state.c
+++ b/mm/swap_state.c
@@ -330,8 +330,8 @@ static inline bool swap_use_vma_readahead(void)
* lock getting page table operations atomic even if we drop the page
* lock before returning.
*/
-struct page *lookup_swap_cache(swp_entry_t entry, struct vm_area_struct *vma,
- unsigned long addr)
+struct page *find_swap_cache(swp_entry_t entry, struct vm_area_struct *vma,
+ unsigned long addr)
{
struct page *page;
@@ -374,19 +374,20 @@ struct page *lookup_swap_cache(swp_entry_t entry, struct vm_area_struct *vma,
return page;
}
-struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
+struct page *find_or_create_swap_cache(swp_entry_t entry, gfp_t gfp_mask,
struct vm_area_struct *vma, unsigned long addr,
- bool *new_page_allocated)
+ bool *created)
{
struct page *found_page, *new_page = NULL;
struct address_space *swapper_space = swap_address_space(entry);
int err;
- *new_page_allocated = false;
+
+ *created = false;
do {
/*
* First check the swap cache. Since this is normally
- * called after lookup_swap_cache() failed, re-calling
+ * called after find_swap_cache() failed, re-calling
* that would confuse statistics.
*/
found_page = find_get_page(swapper_space, swp_offset(entry));
@@ -449,7 +450,7 @@ struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
* Initiate read into locked page and return.
*/
lru_cache_add_anon(new_page);
- *new_page_allocated = true;
+ *created = true;
return new_page;
}
radix_tree_preload_end();
@@ -475,14 +476,14 @@ struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
struct vm_area_struct *vma, unsigned long addr, bool do_poll)
{
- bool page_was_allocated;
- struct page *retpage = __read_swap_cache_async(entry, gfp_mask,
- vma, addr, &page_was_allocated);
+ struct page *page;
+ bool created;
- if (page_was_allocated)
- swap_readpage(retpage, do_poll);
+ page = find_or_create_swap_cache(entry, gfp_mask, vma, addr, &created);
+ if (created)
+ swap_readpage(page, do_poll);
- return retpage;
+ return page;
}
static unsigned int __swapin_nr_pages(unsigned long prev_offset,
@@ -573,7 +574,7 @@ struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
unsigned long mask;
struct swap_info_struct *si = swp_swap_info(entry);
struct blk_plug plug;
- bool do_poll = true, page_allocated;
+ bool do_poll = true, created;
struct vm_area_struct *vma = vmf->vma;
unsigned long addr = vmf->address;
@@ -593,12 +594,12 @@ struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
blk_start_plug(&plug);
for (offset = start_offset; offset <= end_offset ; offset++) {
/* Ok, do the async read-ahead now */
- page = __read_swap_cache_async(
+ page = find_or_create_swap_cache(
swp_entry(swp_type(entry), offset),
- gfp_mask, vma, addr, &page_allocated);
+ gfp_mask, vma, addr, &created);
if (!page)
continue;
- if (page_allocated) {
+ if (created) {
swap_readpage(page, false);
if (offset != entry_offset) {
SetPageReadahead(page);
@@ -738,7 +739,7 @@ static struct page *swap_vma_readahead(swp_entry_t fentry, gfp_t gfp_mask,
pte_t *pte, pentry;
swp_entry_t entry;
unsigned int i;
- bool page_allocated;
+ bool created;
struct vma_swap_readahead ra_info = {0,};
swap_ra_info(vmf, &ra_info);
@@ -756,11 +757,11 @@ static struct page *swap_vma_readahead(swp_entry_t fentry, gfp_t gfp_mask,
entry = pte_to_swp_entry(pentry);
if (unlikely(non_swap_entry(entry)))
continue;
- page = __read_swap_cache_async(entry, gfp_mask, vma,
- vmf->address, &page_allocated);
+ page = find_or_create_swap_cache(entry, gfp_mask, vma,
+ vmf->address, &created);
if (!page)
continue;
- if (page_allocated) {
+ if (created) {
swap_readpage(page, false);
if (i != ra_info.offset) {
SetPageReadahead(page);
diff --git a/mm/zswap.c b/mm/zswap.c
index cd91fd9d96b8..6f05faa75766 100644
--- a/mm/zswap.c
+++ b/mm/zswap.c
@@ -823,11 +823,11 @@ enum zswap_get_swap_ret {
static int zswap_get_swap_cache_page(swp_entry_t entry,
struct page **retpage)
{
- bool page_was_allocated;
+ bool created;
- *retpage = __read_swap_cache_async(entry, GFP_KERNEL,
- NULL, 0, &page_was_allocated);
- if (page_was_allocated)
+ *retpage = find_or_create_swap_cache(entry, GFP_KERNEL,
+ NULL, 0, &created);
+ if (created)
return ZSWAP_SWAPCACHE_NEW;
if (!*retpage)
return ZSWAP_SWAPCACHE_FAIL;
--
2.14.3
From: Johannes Weiner <[email protected]>
Reads can take a long time, and if anybody needs to take a write lock on
the mmap_sem it'll block any subsequent readers to the mmap_sem while
the read is outstanding, which could cause long delays. Instead drop
the mmap_sem if we do any reads at all.
Signed-off-by: Johannes Weiner <[email protected]>
Signed-off-by: Josef Bacik <[email protected]>
---
mm/filemap.c | 119 ++++++++++++++++++++++++++++++++++++++++++++---------------
1 file changed, 90 insertions(+), 29 deletions(-)
diff --git a/mm/filemap.c b/mm/filemap.c
index 52517f28e6f4..1ed35cd99b2c 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2366,6 +2366,18 @@ generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
EXPORT_SYMBOL(generic_file_read_iter);
#ifdef CONFIG_MMU
+static struct file *maybe_unlock_mmap_for_io(struct vm_area_struct *vma, int flags)
+{
+ if ((flags & (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT)) == FAULT_FLAG_ALLOW_RETRY) {
+ struct file *file;
+
+ file = get_file(vma->vm_file);
+ up_read(&vma->vm_mm->mmap_sem);
+ return file;
+ }
+ return NULL;
+}
+
/**
* page_cache_read - adds requested page to the page cache if not already there
* @file: file to read
@@ -2405,23 +2417,28 @@ static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
* Synchronous readahead happens when we don't even find
* a page in the page cache at all.
*/
-static void do_sync_mmap_readahead(struct vm_area_struct *vma,
- struct file_ra_state *ra,
- struct file *file,
- pgoff_t offset)
+static int do_sync_mmap_readahead(struct vm_area_struct *vma,
+ struct file_ra_state *ra,
+ struct file *file,
+ pgoff_t offset,
+ int flags)
{
struct address_space *mapping = file->f_mapping;
+ struct file *fpin;
/* If we don't want any read-ahead, don't bother */
if (vma->vm_flags & VM_RAND_READ)
- return;
+ return 0;
if (!ra->ra_pages)
- return;
+ return 0;
if (vma->vm_flags & VM_SEQ_READ) {
+ fpin = maybe_unlock_mmap_for_io(vma, flags);
page_cache_sync_readahead(mapping, ra, file, offset,
ra->ra_pages);
- return;
+ if (fpin)
+ fput(fpin);
+ return fpin ? -EAGAIN : 0;
}
/* Avoid banging the cache line if not needed */
@@ -2433,7 +2450,9 @@ static void do_sync_mmap_readahead(struct vm_area_struct *vma,
* stop bothering with read-ahead. It will only hurt.
*/
if (ra->mmap_miss > MMAP_LOTSAMISS)
- return;
+ return 0;
+
+ fpin = maybe_unlock_mmap_for_io(vma, flags);
/*
* mmap read-around
@@ -2442,28 +2461,40 @@ static void do_sync_mmap_readahead(struct vm_area_struct *vma,
ra->size = ra->ra_pages;
ra->async_size = ra->ra_pages / 4;
ra_submit(ra, mapping, file);
+
+ if (fpin)
+ fput(fpin);
+
+ return fpin ? -EAGAIN : 0;
}
/*
* Asynchronous readahead happens when we find the page and PG_readahead,
* so we want to possibly extend the readahead further..
*/
-static void do_async_mmap_readahead(struct vm_area_struct *vma,
- struct file_ra_state *ra,
- struct file *file,
- struct page *page,
- pgoff_t offset)
+static int do_async_mmap_readahead(struct vm_area_struct *vma,
+ struct file_ra_state *ra,
+ struct file *file,
+ struct page *page,
+ pgoff_t offset,
+ int flags)
{
struct address_space *mapping = file->f_mapping;
+ struct file *fpin;
/* If we don't want any read-ahead, don't bother */
if (vma->vm_flags & VM_RAND_READ)
- return;
+ return 0;
if (ra->mmap_miss > 0)
ra->mmap_miss--;
- if (PageReadahead(page))
- page_cache_async_readahead(mapping, ra, file,
- page, offset, ra->ra_pages);
+ if (!PageReadahead(page))
+ return 0;
+ fpin = maybe_unlock_mmap_for_io(vma, flags);
+ page_cache_async_readahead(mapping, ra, file,
+ page, offset, ra->ra_pages);
+ if (fpin)
+ fput(fpin);
+ return fpin ? -EAGAIN : 0;
}
/**
@@ -2479,10 +2510,8 @@ static void do_async_mmap_readahead(struct vm_area_struct *vma,
*
* vma->vm_mm->mmap_sem must be held on entry.
*
- * If our return value has VM_FAULT_RETRY set, it's because
- * lock_page_or_retry() returned 0.
- * The mmap_sem has usually been released in this case.
- * See __lock_page_or_retry() for the exception.
+ * If our return value has VM_FAULT_RETRY set, the mmap_sem has
+ * usually been released.
*
* If our return value does not have VM_FAULT_RETRY set, the mmap_sem
* has not been released.
@@ -2492,11 +2521,13 @@ static void do_async_mmap_readahead(struct vm_area_struct *vma,
vm_fault_t filemap_fault(struct vm_fault *vmf)
{
int error;
+ struct mm_struct *mm = vmf->vma->vm_mm;
struct file *file = vmf->vma->vm_file;
struct address_space *mapping = file->f_mapping;
struct file_ra_state *ra = &file->f_ra;
struct inode *inode = mapping->host;
pgoff_t offset = vmf->pgoff;
+ int flags = vmf->flags;
pgoff_t max_off;
struct page *page;
vm_fault_t ret = 0;
@@ -2509,27 +2540,44 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
* Do we have something in the page cache already?
*/
page = find_get_page(mapping, offset);
- if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
+ if (likely(page) && !(flags & FAULT_FLAG_TRIED)) {
/*
* We found the page, so try async readahead before
* waiting for the lock.
*/
- do_async_mmap_readahead(vmf->vma, ra, file, page, offset);
+ error = do_async_mmap_readahead(vmf->vma, ra, file, page, offset, vmf->flags);
+ if (error == -EAGAIN)
+ goto out_retry_wait;
} else if (!page) {
/* No page in the page cache at all */
- do_sync_mmap_readahead(vmf->vma, ra, file, offset);
- count_vm_event(PGMAJFAULT);
- count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
ret = VM_FAULT_MAJOR;
+ count_vm_event(PGMAJFAULT);
+ count_memcg_event_mm(mm, PGMAJFAULT);
+ error = do_sync_mmap_readahead(vmf->vma, ra, file, offset, vmf->flags);
+ if (error == -EAGAIN)
+ goto out_retry_wait;
retry_find:
page = find_get_page(mapping, offset);
if (!page)
goto no_cached_page;
}
- if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) {
- put_page(page);
- return ret | VM_FAULT_RETRY;
+ if (!trylock_page(page)) {
+ if (flags & FAULT_FLAG_ALLOW_RETRY) {
+ if (flags & FAULT_FLAG_RETRY_NOWAIT)
+ goto out_retry;
+ up_read(&mm->mmap_sem);
+ goto out_retry_wait;
+ }
+ if (flags & FAULT_FLAG_KILLABLE) {
+ int ret = __lock_page_killable(page);
+
+ if (ret) {
+ up_read(&mm->mmap_sem);
+ goto out_retry;
+ }
+ } else
+ __lock_page(page);
}
/* Did it get truncated? */
@@ -2607,6 +2655,19 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
/* Things didn't work out. Return zero to tell the mm layer so. */
shrink_readahead_size_eio(file, ra);
return VM_FAULT_SIGBUS;
+
+out_retry_wait:
+ if (page) {
+ if (flags & FAULT_FLAG_KILLABLE)
+ wait_on_page_locked_killable(page);
+ else
+ wait_on_page_locked(page);
+ }
+
+out_retry:
+ if (page)
+ put_page(page);
+ return ret | VM_FAULT_RETRY;
}
EXPORT_SYMBOL(filemap_fault);
--
2.14.3
We want to be able to cache the result of a previous loop of a page
fault in the case that we use VM_FAULT_RETRY, so introduce
handle_mm_fault_cacheable that will take a struct vm_fault directly, add
a ->cached_page field to vm_fault, and add helpers to init/cleanup the
struct vm_fault.
I've converted x86, other arch's can follow suit if they so wish, it's
relatively straightforward.
Signed-off-by: Josef Bacik <[email protected]>
---
arch/x86/mm/fault.c | 6 +++-
include/linux/mm.h | 31 +++++++++++++++++++++
mm/memory.c | 79 ++++++++++++++++++++++++++++++++---------------------
3 files changed, 84 insertions(+), 32 deletions(-)
diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c
index 47bebfe6efa7..ef6e538c4931 100644
--- a/arch/x86/mm/fault.c
+++ b/arch/x86/mm/fault.c
@@ -1211,6 +1211,7 @@ static noinline void
__do_page_fault(struct pt_regs *regs, unsigned long error_code,
unsigned long address)
{
+ struct vm_fault vmf = {};
struct vm_area_struct *vma;
struct task_struct *tsk;
struct mm_struct *mm;
@@ -1392,7 +1393,8 @@ __do_page_fault(struct pt_regs *regs, unsigned long error_code,
* fault, so we read the pkey beforehand.
*/
pkey = vma_pkey(vma);
- fault = handle_mm_fault(vma, address, flags);
+ vm_fault_init(&vmf, vma, address, flags);
+ fault = handle_mm_fault_cacheable(&vmf);
major |= fault & VM_FAULT_MAJOR;
/*
@@ -1408,6 +1410,7 @@ __do_page_fault(struct pt_regs *regs, unsigned long error_code,
if (!fatal_signal_pending(tsk))
goto retry;
}
+ vm_fault_cleanup(&vmf);
/* User mode? Just return to handle the fatal exception */
if (flags & FAULT_FLAG_USER)
@@ -1418,6 +1421,7 @@ __do_page_fault(struct pt_regs *regs, unsigned long error_code,
return;
}
+ vm_fault_cleanup(&vmf);
up_read(&mm->mmap_sem);
if (unlikely(fault & VM_FAULT_ERROR)) {
mm_fault_error(regs, error_code, address, &pkey, fault);
diff --git a/include/linux/mm.h b/include/linux/mm.h
index a61ebe8ad4ca..4a84ec976dfc 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -360,6 +360,12 @@ struct vm_fault {
* is set (which is also implied by
* VM_FAULT_ERROR).
*/
+ struct page *cached_page; /* ->fault handlers that return
+ * VM_FAULT_RETRY can store their
+ * previous page here to be reused the
+ * next time we loop through the fault
+ * handler for faster lookup.
+ */
/* These three entries are valid only while holding ptl lock */
pte_t *pte; /* Pointer to pte entry matching
* the 'address'. NULL if the page
@@ -378,6 +384,16 @@ struct vm_fault {
*/
};
+static inline void vm_fault_init(struct vm_fault *vmf,
+ struct vm_area_struct *vma,
+ unsigned long address,
+ unsigned int flags)
+{
+ vmf->vma = vma;
+ vmf->address = address;
+ vmf->flags = flags;
+}
+
/* page entry size for vm->huge_fault() */
enum page_entry_size {
PE_SIZE_PTE = 0,
@@ -943,6 +959,14 @@ static inline void put_page(struct page *page)
__put_page(page);
}
+static inline void vm_fault_cleanup(struct vm_fault *vmf)
+{
+ if (vmf->cached_page) {
+ put_page(vmf->cached_page);
+ vmf->cached_page = NULL;
+ }
+}
+
#if defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP)
#define SECTION_IN_PAGE_FLAGS
#endif
@@ -1405,6 +1429,7 @@ int invalidate_inode_page(struct page *page);
#ifdef CONFIG_MMU
extern vm_fault_t handle_mm_fault(struct vm_area_struct *vma,
unsigned long address, unsigned int flags);
+extern vm_fault_t handle_mm_fault_cacheable(struct vm_fault *vmf);
extern int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
unsigned long address, unsigned int fault_flags,
bool *unlocked);
@@ -1420,6 +1445,12 @@ static inline vm_fault_t handle_mm_fault(struct vm_area_struct *vma,
BUG();
return VM_FAULT_SIGBUS;
}
+static inline vm_fault_t handle_mm_fault_cacheable(struct vm_fault *vmf)
+{
+ /* should never happen if there's no MMU */
+ BUG();
+ return VM_FAULT_SIGBUS;
+}
static inline int fixup_user_fault(struct task_struct *tsk,
struct mm_struct *mm, unsigned long address,
unsigned int fault_flags, bool *unlocked)
diff --git a/mm/memory.c b/mm/memory.c
index c467102a5cbc..433075f722ea 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4024,36 +4024,34 @@ static vm_fault_t handle_pte_fault(struct vm_fault *vmf)
* The mmap_sem may have been released depending on flags and our
* return value. See filemap_fault() and __lock_page_or_retry().
*/
-static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma,
- unsigned long address, unsigned int flags)
+static vm_fault_t __handle_mm_fault(struct vm_fault *vmf)
{
- struct vm_fault vmf = {
- .vma = vma,
- .address = address & PAGE_MASK,
- .flags = flags,
- .pgoff = linear_page_index(vma, address),
- .gfp_mask = __get_fault_gfp_mask(vma),
- };
- unsigned int dirty = flags & FAULT_FLAG_WRITE;
+ struct vm_area_struct *vma = vmf->vma;
+ unsigned long address = vmf->address;
+ unsigned int dirty = vmf->flags & FAULT_FLAG_WRITE;
struct mm_struct *mm = vma->vm_mm;
pgd_t *pgd;
p4d_t *p4d;
vm_fault_t ret;
+ vmf->address = address & PAGE_MASK;
+ vmf->pgoff = linear_page_index(vma, address);
+ vmf->gfp_mask = __get_fault_gfp_mask(vma);
+
pgd = pgd_offset(mm, address);
p4d = p4d_alloc(mm, pgd, address);
if (!p4d)
return VM_FAULT_OOM;
- vmf.pud = pud_alloc(mm, p4d, address);
- if (!vmf.pud)
+ vmf->pud = pud_alloc(mm, p4d, address);
+ if (!vmf->pud)
return VM_FAULT_OOM;
- if (pud_none(*vmf.pud) && transparent_hugepage_enabled(vma)) {
- ret = create_huge_pud(&vmf);
+ if (pud_none(*vmf->pud) && transparent_hugepage_enabled(vma)) {
+ ret = create_huge_pud(vmf);
if (!(ret & VM_FAULT_FALLBACK))
return ret;
} else {
- pud_t orig_pud = *vmf.pud;
+ pud_t orig_pud = *vmf->pud;
barrier();
if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) {
@@ -4061,50 +4059,50 @@ static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma,
/* NUMA case for anonymous PUDs would go here */
if (dirty && !pud_write(orig_pud)) {
- ret = wp_huge_pud(&vmf, orig_pud);
+ ret = wp_huge_pud(vmf, orig_pud);
if (!(ret & VM_FAULT_FALLBACK))
return ret;
} else {
- huge_pud_set_accessed(&vmf, orig_pud);
+ huge_pud_set_accessed(vmf, orig_pud);
return 0;
}
}
}
- vmf.pmd = pmd_alloc(mm, vmf.pud, address);
- if (!vmf.pmd)
+ vmf->pmd = pmd_alloc(mm, vmf->pud, address);
+ if (!vmf->pmd)
return VM_FAULT_OOM;
- if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) {
- ret = create_huge_pmd(&vmf);
+ if (pmd_none(*vmf->pmd) && transparent_hugepage_enabled(vma)) {
+ ret = create_huge_pmd(vmf);
if (!(ret & VM_FAULT_FALLBACK))
return ret;
} else {
- pmd_t orig_pmd = *vmf.pmd;
+ pmd_t orig_pmd = *vmf->pmd;
barrier();
if (unlikely(is_swap_pmd(orig_pmd))) {
VM_BUG_ON(thp_migration_supported() &&
!is_pmd_migration_entry(orig_pmd));
if (is_pmd_migration_entry(orig_pmd))
- pmd_migration_entry_wait(mm, vmf.pmd);
+ pmd_migration_entry_wait(mm, vmf->pmd);
return 0;
}
if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) {
if (pmd_protnone(orig_pmd) && vma_is_accessible(vma))
- return do_huge_pmd_numa_page(&vmf, orig_pmd);
+ return do_huge_pmd_numa_page(vmf, orig_pmd);
if (dirty && !pmd_write(orig_pmd)) {
- ret = wp_huge_pmd(&vmf, orig_pmd);
+ ret = wp_huge_pmd(vmf, orig_pmd);
if (!(ret & VM_FAULT_FALLBACK))
return ret;
} else {
- huge_pmd_set_accessed(&vmf, orig_pmd);
+ huge_pmd_set_accessed(vmf, orig_pmd);
return 0;
}
}
}
- return handle_pte_fault(&vmf);
+ return handle_pte_fault(vmf);
}
/*
@@ -4113,9 +4111,10 @@ static vm_fault_t __handle_mm_fault(struct vm_area_struct *vma,
* The mmap_sem may have been released depending on flags and our
* return value. See filemap_fault() and __lock_page_or_retry().
*/
-vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
- unsigned int flags)
+static vm_fault_t do_handle_mm_fault(struct vm_fault *vmf)
{
+ struct vm_area_struct *vma = vmf->vma;
+ unsigned int flags = vmf->flags;
vm_fault_t ret;
__set_current_state(TASK_RUNNING);
@@ -4139,9 +4138,9 @@ vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
mem_cgroup_enter_user_fault();
if (unlikely(is_vm_hugetlb_page(vma)))
- ret = hugetlb_fault(vma->vm_mm, vma, address, flags);
+ ret = hugetlb_fault(vma->vm_mm, vma, vmf->address, flags);
else
- ret = __handle_mm_fault(vma, address, flags);
+ ret = __handle_mm_fault(vmf);
if (flags & FAULT_FLAG_USER) {
mem_cgroup_exit_user_fault();
@@ -4157,8 +4156,26 @@ vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
return ret;
}
+
+vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
+ unsigned int flags)
+{
+ struct vm_fault vmf = {};
+ vm_fault_t ret;
+
+ vm_fault_init(&vmf, vma, address, flags);
+ ret = do_handle_mm_fault(&vmf);
+ vm_fault_cleanup(&vmf);
+ return ret;
+}
EXPORT_SYMBOL_GPL(handle_mm_fault);
+vm_fault_t handle_mm_fault_cacheable(struct vm_fault *vmf)
+{
+ return do_handle_mm_fault(vmf);
+}
+EXPORT_SYMBOL_GPL(handle_mm_fault_cacheable);
+
#ifndef __PAGETABLE_P4D_FOLDED
/*
* Allocate p4d page table.
--
2.14.3
Johannes' patches didn't quite cover all of the IO cases that we need to
drop the mmap_sem for, this patch covers the rest of them.
Signed-off-by: Josef Bacik <[email protected]>
---
mm/filemap.c | 11 +++++++++++
1 file changed, 11 insertions(+)
diff --git a/mm/filemap.c b/mm/filemap.c
index 1ed35cd99b2c..65395ee132a0 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2523,6 +2523,7 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
int error;
struct mm_struct *mm = vmf->vma->vm_mm;
struct file *file = vmf->vma->vm_file;
+ struct file *fpin = NULL;
struct address_space *mapping = file->f_mapping;
struct file_ra_state *ra = &file->f_ra;
struct inode *inode = mapping->host;
@@ -2610,11 +2611,15 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
return ret | VM_FAULT_LOCKED;
no_cached_page:
+ fpin = maybe_unlock_mmap_for_io(vmf->vma, vmf->flags);
+
/*
* We're only likely to ever get here if MADV_RANDOM is in
* effect.
*/
error = page_cache_read(file, offset, vmf->gfp_mask);
+ if (fpin)
+ goto out_retry;
/*
* The page we want has now been added to the page cache.
@@ -2634,6 +2639,8 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
return VM_FAULT_SIGBUS;
page_not_uptodate:
+ fpin = maybe_unlock_mmap_for_io(vmf->vma, vmf->flags);
+
/*
* Umm, take care of errors if the page isn't up-to-date.
* Try to re-read it _once_. We do this synchronously,
@@ -2647,6 +2654,8 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
if (!PageUptodate(page))
error = -EIO;
}
+ if (fpin)
+ goto out_retry;
put_page(page);
if (!error || error == AOP_TRUNCATED_PAGE)
@@ -2665,6 +2674,8 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
}
out_retry:
+ if (fpin)
+ fput(fpin);
if (page)
put_page(page);
return ret | VM_FAULT_RETRY;
--
2.14.3
On Wed, Sep 26, 2018 at 05:08:48PM -0400, Josef Bacik wrote:
> We want to be able to cache the result of a previous loop of a page
> fault in the case that we use VM_FAULT_RETRY, so introduce
> handle_mm_fault_cacheable that will take a struct vm_fault directly, add
> a ->cached_page field to vm_fault, and add helpers to init/cleanup the
> struct vm_fault.
>
> I've converted x86, other arch's can follow suit if they so wish, it's
> relatively straightforward.
Here's what I did back in January ... feel free to steal any of it if you
like it better.
diff --git a/mm/memory.c b/mm/memory.c
index 5eb3d2524bdc..403934297a3d 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -3977,36 +3977,28 @@ static int handle_pte_fault(struct vm_fault *vmf)
* The mmap_sem may have been released depending on flags and our
* return value. See filemap_fault() and __lock_page_or_retry().
*/
-static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
- unsigned int flags)
+static int __handle_mm_fault(struct vm_fault *vmf)
{
- struct vm_fault vmf = {
- .vma = vma,
- .address = address & PAGE_MASK,
- .flags = flags,
- .pgoff = linear_page_index(vma, address),
- .gfp_mask = __get_fault_gfp_mask(vma),
- };
- unsigned int dirty = flags & FAULT_FLAG_WRITE;
- struct mm_struct *mm = vma->vm_mm;
+ unsigned int dirty = vmf->flags & FAULT_FLAG_WRITE;
+ struct mm_struct *mm = vmf->vma->vm_mm;
pgd_t *pgd;
p4d_t *p4d;
int ret;
- pgd = pgd_offset(mm, address);
- p4d = p4d_alloc(mm, pgd, address);
+ pgd = pgd_offset(mm, vmf->address);
+ p4d = p4d_alloc(mm, pgd, vmf->address);
if (!p4d)
return VM_FAULT_OOM;
- vmf.pud = pud_alloc(mm, p4d, address);
- if (!vmf.pud)
+ vmf->pud = pud_alloc(mm, p4d, vmf->address);
+ if (!vmf->pud)
return VM_FAULT_OOM;
- if (pud_none(*vmf.pud) && transparent_hugepage_enabled(vma)) {
- ret = create_huge_pud(&vmf);
+ if (pud_none(*vmf->pud) && transparent_hugepage_enabled(vmf->vma)) {
+ ret = create_huge_pud(vmf);
if (!(ret & VM_FAULT_FALLBACK))
return ret;
} else {
- pud_t orig_pud = *vmf.pud;
+ pud_t orig_pud = *vmf->pud;
barrier();
if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) {
@@ -4014,50 +4006,51 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
/* NUMA case for anonymous PUDs would go here */
if (dirty && !pud_access_permitted(orig_pud, WRITE)) {
- ret = wp_huge_pud(&vmf, orig_pud);
+ ret = wp_huge_pud(vmf, orig_pud);
if (!(ret & VM_FAULT_FALLBACK))
return ret;
} else {
- huge_pud_set_accessed(&vmf, orig_pud);
+ huge_pud_set_accessed(vmf, orig_pud);
return 0;
}
}
}
- vmf.pmd = pmd_alloc(mm, vmf.pud, address);
- if (!vmf.pmd)
+ vmf->pmd = pmd_alloc(mm, vmf->pud, vmf->address);
+ if (!vmf->pmd)
return VM_FAULT_OOM;
- if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) {
- ret = create_huge_pmd(&vmf);
+ if (pmd_none(*vmf->pmd) && transparent_hugepage_enabled(vmf->vma)) {
+ ret = create_huge_pmd(vmf);
if (!(ret & VM_FAULT_FALLBACK))
return ret;
} else {
- pmd_t orig_pmd = *vmf.pmd;
+ pmd_t orig_pmd = *vmf->pmd;
barrier();
if (unlikely(is_swap_pmd(orig_pmd))) {
VM_BUG_ON(thp_migration_supported() &&
!is_pmd_migration_entry(orig_pmd));
if (is_pmd_migration_entry(orig_pmd))
- pmd_migration_entry_wait(mm, vmf.pmd);
+ pmd_migration_entry_wait(mm, vmf->pmd);
return 0;
}
if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) {
- if (pmd_protnone(orig_pmd) && vma_is_accessible(vma))
- return do_huge_pmd_numa_page(&vmf, orig_pmd);
+ if (pmd_protnone(orig_pmd) &&
+ vma_is_accessible(vmf->vma))
+ return do_huge_pmd_numa_page(vmf, orig_pmd);
if (dirty && !pmd_access_permitted(orig_pmd, WRITE)) {
- ret = wp_huge_pmd(&vmf, orig_pmd);
+ ret = wp_huge_pmd(vmf, orig_pmd);
if (!(ret & VM_FAULT_FALLBACK))
return ret;
} else {
- huge_pmd_set_accessed(&vmf, orig_pmd);
+ huge_pmd_set_accessed(vmf, orig_pmd);
return 0;
}
}
}
- return handle_pte_fault(&vmf);
+ return handle_pte_fault(vmf);
}
/*
@@ -4066,9 +4059,10 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
* The mmap_sem may have been released depending on flags and our
* return value. See filemap_fault() and __lock_page_or_retry().
*/
-int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
- unsigned int flags)
+int vm_handle_fault(struct vm_fault *vmf)
{
+ unsigned int flags = vmf->flags;
+ struct vm_area_struct *vma = vmf->vma;
int ret;
__set_current_state(TASK_RUNNING);
@@ -4092,9 +4086,9 @@ int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
mem_cgroup_oom_enable();
if (unlikely(is_vm_hugetlb_page(vma)))
- ret = hugetlb_fault(vma->vm_mm, vma, address, flags);
+ ret = hugetlb_fault(vma->vm_mm, vma, vmf->address, flags);
else
- ret = __handle_mm_fault(vma, address, flags);
+ ret = __handle_mm_fault(vmf);
if (flags & FAULT_FLAG_USER) {
mem_cgroup_oom_disable();
@@ -4110,6 +4104,26 @@ int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
return ret;
}
+
+/*
+ * By the time we get here, we already hold the mm semaphore
+ *
+ * The mmap_sem may have been released depending on flags and our
+ * return value. See filemap_fault() and __lock_page_or_retry().
+ */
+int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
+ unsigned int flags)
+{
+ struct vm_fault vmf = {
+ .vma = vma,
+ .address = address & PAGE_MASK,
+ .flags = flags,
+ .pgoff = linear_page_index(vma, address),
+ .gfp_mask = __get_fault_gfp_mask(vma),
+ };
+
+ return vm_handle_fault(&vmf);
+}
EXPORT_SYMBOL_GPL(handle_mm_fault);
#ifndef __PAGETABLE_P4D_FOLDED