[PATCH -part2 0/3] mm: improve handling of mm->exe_file

[PATCH -part2 0/3] mm: improve handling of mm->exe_file

[Davidlohr Bueso]

This set addresses exe_file use for users that require the mmap_sem
for other things (mostly looking up the related vma->vm_file). In
a lot of cases we end up with scenarios where we take the mmap_sem in
get_mm_exe_file(), just to then take it again soon after. This is only
temporary as we will remove the need for mmap_sem when dealing with
exe_file.

Applies on top of linux-next (-20150225). The arch bits are entirely
100% untested, so I apologize if there are any stupid build issues.

Thanks!

Davidlohr Bueso (3):
  tile/elf: reorganize notify_exec()
  oprofile: reduce mmap_sem hold for mm->exe_file
  powerpc/oprofile: reduce mmap_sem hold for exe_file

 arch/powerpc/oprofile/cell/spu_task_sync.c | 13 +++++----
 arch/tile/mm/elf.c                         | 47 ++++++++++++++++++------------
 drivers/oprofile/buffer_sync.c             | 30 ++++++++++---------
 3 files changed, 53 insertions(+), 37 deletions(-)

-- 
2.1.4

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[PATCH 1/3] tile/elf: reorganize notify_exec()

[Davidlohr Bueso]

In the future mm->exe_file will be done without mmap_sem
serialization, thus isolate and reorganize the tile elf
code to make the transition easier. Good users will, make
use of the more standard get_mm_exe_file(), requiring only
holding the mmap_sem to read the value, and relying on reference
counting to make sure that the exe file won't dissappear
underneath us.

The visible effects of this patch are:

   o We now take and drop the mmap_sem more often. Instead of
     just in arch_setup_additional_pages(), we also do it in:

     1) get_mm_exe_file()
     2) to get the mm->vm_file and notify the simulator.

    [Note that 1) will disappear once we change the locking
     rules for exe_file.]

   o We avoid getting a free page and doing d_path() while
     holding the mmap_sem. This requires reordering the checks.

Cc: Chris Metcalf <cmetcalf@ezchip.com>
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
---

completely untested.

 arch/tile/mm/elf.c | 47 +++++++++++++++++++++++++++++------------------
 1 file changed, 29 insertions(+), 18 deletions(-)

diff --git a/arch/tile/mm/elf.c b/arch/tile/mm/elf.c
index 23f044e..f7ddae3 100644
--- a/arch/tile/mm/elf.c
+++ b/arch/tile/mm/elf.c
@@ -17,6 +17,7 @@
 #include <linux/binfmts.h>
 #include <linux/compat.h>
 #include <linux/mman.h>
+#include <linux/file.h>
 #include <linux/elf.h>
 #include <asm/pgtable.h>
 #include <asm/pgalloc.h>
@@ -39,30 +40,34 @@ static void sim_notify_exec(const char *binary_name)
 
 static int notify_exec(struct mm_struct *mm)
 {
+	int ret = 0;
 	char *buf, *path;
 	struct vm_area_struct *vma;
+	struct file *exe_file;
 
 	if (!sim_is_simulator())
 		return 1;
 
-	if (mm->exe_file == NULL)
-		return 0;
-
-	for (vma = current->mm->mmap; ; vma = vma->vm_next) {
-		if (vma == NULL)
-			return 0;
-		if (vma->vm_file == mm->exe_file)
-			break;
-	}
-
 	buf = (char *) __get_free_page(GFP_KERNEL);
 	if (buf == NULL)
 		return 0;
 
-	path = d_path(&mm->exe_file->f_path, buf, PAGE_SIZE);
-	if (IS_ERR(path)) {
-		free_page((unsigned long)buf);
-		return 0;
+	exe_file = get_mm_exe_file(mm);
+	if (exe_file == NULL)
+		goto done_free;
+
+	path = d_path(&exe_file->f_path, buf, PAGE_SIZE);
+	if (IS_ERR(path))
+		goto done_put;
+
+	down_read(&mm->mmap_sem);
+	for (vma = current->mm->mmap; ; vma = vma->vm_next) {
+		if (vma == NULL) {
+			up_read(&mm->mmap_sem);
+			goto done_put;
+		}
+		if (vma->vm_file == exe_file)
+			break;
 	}
 
 	/*
@@ -80,14 +85,20 @@ static int notify_exec(struct mm_struct *mm)
 			__insn_mtspr(SPR_SIM_CONTROL,
 				     (SIM_CONTROL_DLOPEN
 				      | (c << _SIM_CONTROL_OPERATOR_BITS)));
-			if (c == '\0')
+			if (c == '\0') {
+				ret = 1; /* success */
 				break;
+			}
 		}
 	}
+	up_read(&mm->mmap_sem);
 
 	sim_notify_exec(path);
+done_put:
+	fput(exe_file);
+done_free:
 	free_page((unsigned long)buf);
-	return 1;
+	return ret;
 }
 
 /* Notify a running simulator, if any, that we loaded an interpreter. */
@@ -109,8 +120,6 @@ int arch_setup_additional_pages(struct linux_binprm *bprm,
 	struct mm_struct *mm = current->mm;
 	int retval = 0;
 
-	down_write(&mm->mmap_sem);
-
 	/*
 	 * Notify the simulator that an exec just occurred.
 	 * If we can't find the filename of the mapping, just use
@@ -119,6 +128,8 @@ int arch_setup_additional_pages(struct linux_binprm *bprm,
 	if (!notify_exec(mm))
 		sim_notify_exec(bprm->filename);
 
+	down_write(&mm->mmap_sem);
+
 	retval = setup_vdso_pages();
 
 #ifndef __tilegx__
-- 
2.1.4

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH 1/3] tile/elf: reorganize notify_exec()

[Chris Metcalf]

On 2/25/2015 4:58 PM, Davidlohr Bueso wrote:
> In the future mm->exe_file will be done without mmap_sem
> serialization, thus isolate and reorganize the tile elf
> code to make the transition easier. Good users will, make
> use of the more standard get_mm_exe_file(), requiring only
> holding the mmap_sem to read the value, and relying on reference
> counting to make sure that the exe file won't dissappear
> underneath us.
>
> The visible effects of this patch are:
>
>     o We now take and drop the mmap_sem more often. Instead of
>       just in arch_setup_additional_pages(), we also do it in:
>
>       1) get_mm_exe_file()
>       2) to get the mm->vm_file and notify the simulator.
>
>      [Note that 1) will disappear once we change the locking
>       rules for exe_file.]
>
>     o We avoid getting a free page and doing d_path() while
>       holding the mmap_sem. This requires reordering the checks.
>
> Cc: Chris Metcalf<cmetcalf@ezchip.com>
> Signed-off-by: Davidlohr Bueso<dbueso@suse.de>
> ---
>
> completely untested.
>
>   arch/tile/mm/elf.c | 47 +++++++++++++++++++++++++++++------------------
>   1 file changed, 29 insertions(+), 18 deletions(-)

This looks OK to me and passes basic testing.  So here is my

Acked-by: Chris Metcalf <cmetcalf@ezchip.com>

Or would you prefer I took this through the tile tree?

-- 
Chris Metcalf, EZChip Semiconductor
http://www.ezchip.com

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[PATCH 2/3] oprofile: reduce mmap_sem hold for mm->exe_file

[Davidlohr Bueso]

sync_buffer() needs the mmap_sem for two distinct operations,
both only occurring upon user context switch handling:

 1) Dealing with the exe_file.

 2) Adding the dcookie data as we need to lookup the vma that
   backs it. This is done via add_sample() and add_data().

This patch isolates 1), for it will no longer need the mmap_sem
for serialization. However, for now, make of the more standard
get_mm_exe_file(), requiring only holding the mmap_sem to read
the value, and relying on reference counting to make sure that
the exe file won't dissappear underneath us while doing the get
dcookie.

As a consequence, for 2) we move the mmap_sem locking into where
we really need it, in lookup_dcookie(). The benefits are twofold:
reduce mmap_sem hold times, and cleaner code.

Cc: Robert Richter <rric@kernel.org>
Cc: oprofile-list@lists.sf.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
---
 drivers/oprofile/buffer_sync.c | 30 ++++++++++++++++--------------
 1 file changed, 16 insertions(+), 14 deletions(-)

diff --git a/drivers/oprofile/buffer_sync.c b/drivers/oprofile/buffer_sync.c
index d93b2b6..82f7000 100644
--- a/drivers/oprofile/buffer_sync.c
+++ b/drivers/oprofile/buffer_sync.c
@@ -21,6 +21,7 @@
  * objects.
  */
 
+#include <linux/file.h>
 #include <linux/mm.h>
 #include <linux/workqueue.h>
 #include <linux/notifier.h>
@@ -224,10 +225,18 @@ static inline unsigned long fast_get_dcookie(struct path *path)
 static unsigned long get_exec_dcookie(struct mm_struct *mm)
 {
 	unsigned long cookie = NO_COOKIE;
+	struct file *exe_file;
 
-	if (mm && mm->exe_file)
-		cookie = fast_get_dcookie(&mm->exe_file->f_path);
+	if (!mm)
+		goto done;
+
+	exe_file = get_mm_exe_file(mm);
+	if (!exe_file)
+		goto done;
 
+	cookie = fast_get_dcookie(&exe_file->f_path);
+	fput(exe_file);
+done:
 	return cookie;
 }
 
@@ -236,6 +245,8 @@ static unsigned long get_exec_dcookie(struct mm_struct *mm)
  * pair that can then be added to the global event buffer. We make
  * sure to do this lookup before a mm->mmap modification happens so
  * we don't lose track.
+ *
+ * The caller must ensure the mm is not nil (ie: not a kernel thread).
  */
 static unsigned long
 lookup_dcookie(struct mm_struct *mm, unsigned long addr, off_t *offset)
@@ -243,6 +254,7 @@ lookup_dcookie(struct mm_struct *mm, unsigned long addr, off_t *offset)
 	unsigned long cookie = NO_COOKIE;
 	struct vm_area_struct *vma;
 
+	down_read(&mm->mmap_sem);
 	for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
 
 		if (addr < vma->vm_start || addr >= vma->vm_end)
@@ -262,6 +274,7 @@ lookup_dcookie(struct mm_struct *mm, unsigned long addr, off_t *offset)
 
 	if (!vma)
 		cookie = INVALID_COOKIE;
+	up_read(&mm->mmap_sem);
 
 	return cookie;
 }
@@ -402,20 +415,9 @@ static void release_mm(struct mm_struct *mm)
 {
 	if (!mm)
 		return;
-	up_read(&mm->mmap_sem);
 	mmput(mm);
 }
 
-
-static struct mm_struct *take_tasks_mm(struct task_struct *task)
-{
-	struct mm_struct *mm = get_task_mm(task);
-	if (mm)
-		down_read(&mm->mmap_sem);
-	return mm;
-}
-
-
 static inline int is_code(unsigned long val)
 {
 	return val == ESCAPE_CODE;
@@ -532,7 +534,7 @@ void sync_buffer(int cpu)
 				new = (struct task_struct *)val;
 				oldmm = mm;
 				release_mm(oldmm);
-				mm = take_tasks_mm(new);
+				mm = get_task_mm(new);
 				if (mm != oldmm)
 					cookie = get_exec_dcookie(mm);
 				add_user_ctx_switch(new, cookie);
-- 
2.1.4

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[PATCH 3/3] powerpc/oprofile: reduce mmap_sem hold for exe_file

[Davidlohr Bueso]

In the future mm->exe_file will be done without mmap_sem
serialization, thus isolate and reorganize the related
code to make the transition easier. Good users will, make
use of the more standard get_mm_exe_file(), requiring only
holding the mmap_sem to read the value, and relying on reference
counting to make sure that the exe file won't dissappear
underneath us while getting the dcookie.

Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Robert Richter <rric@kernel.org>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: cbe-oss-dev@lists.ozlabs.org
Cc: oprofile-list@lists.sourceforge.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
---

Completely untested.

 arch/powerpc/oprofile/cell/spu_task_sync.c | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/arch/powerpc/oprofile/cell/spu_task_sync.c b/arch/powerpc/oprofile/cell/spu_task_sync.c
index 1c27831..ed7b097 100644
--- a/arch/powerpc/oprofile/cell/spu_task_sync.c
+++ b/arch/powerpc/oprofile/cell/spu_task_sync.c
@@ -22,6 +22,7 @@
 #include <linux/kref.h>
 #include <linux/mm.h>
 #include <linux/fs.h>
+#include <linux/file.h>
 #include <linux/module.h>
 #include <linux/notifier.h>
 #include <linux/numa.h>
@@ -322,18 +323,20 @@ get_exec_dcookie_and_offset(struct spu *spu, unsigned int *offsetp,
 	unsigned long app_cookie = 0;
 	unsigned int my_offset = 0;
 	struct vm_area_struct *vma;
+	struct file *exe_file;
 	struct mm_struct *mm = spu->mm;
 
 	if (!mm)
 		goto out;
 
-	down_read(&mm->mmap_sem);
-
-	if (mm->exe_file) {
-		app_cookie = fast_get_dcookie(&mm->exe_file->f_path);
-		pr_debug("got dcookie for %pD\n", mm->exe_file);
+	exe_file = get_mm_exe_file(mm);
+	if (exe_file) {
+		app_cookie = fast_get_dcookie(&exe_file->f_path);
+		pr_debug("got dcookie for %pD\n", exe_file);
+		fput(exe_file);
 	}
 
+	down_read(&mm->mmap_sem);
 	for (vma = mm->mmap; vma; vma = vma->vm_next) {
 		if (vma->vm_start > spu_ref || vma->vm_end <= spu_ref)
 			continue;
-- 
2.1.4

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>