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
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 <[email protected]>
Signed-off-by: Davidlohr Bueso <[email protected]>
---
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
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 <[email protected]>
Cc: [email protected]
Signed-off-by: Davidlohr Bueso <[email protected]>
---
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
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 <[email protected]>
Cc: Robert Richter <[email protected]>
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Davidlohr Bueso <[email protected]>
---
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