VFS: make file->f_pos access atomic on 32bit arch
Subsystems (2)
Hi,
----- Messaggio originale -----
Quoted 3 lines
Da: Nick Piggin <[email protected]>
A: Michael Trimarchi <[email protected]>
Cc: Hisashi Hifumi <[email protected]>; [email protected]; [email protected]; [email protected]
Quoted 3 lines
Inviato: Martedì 16 settembre 2008, 16:11:21
Oggetto: Re: [PATCH] VFS: makefile->f_posaccess atomic on 32bit arch
...
Quoted 7 lines
Quoted 3 lines
i was thinking about introducing a
file_pos_update()not implemented using
file_pos_read()/file_pos_write() and taking theseq_lockor disabling preemption only one timeAFAIK, the only way to do an atomic 64 bit store on 32-bit x86 is "cmpxchg8b" (with lock prefix on SMP). That would be far slower I'm sure.
Maybe it is clear with an example:
+/*
+ * file_pos_read/write is atomic by using sequence lock on 32bit arch.
+ */
+static inline loff_t file_pos_read(struct file *file)
+{
+#if BITS_PER_LONG == 32 && defined(CONFIG_SMP)
+ loff_t pos;
+ unsigned int seq;
+
+ do {
+ seq = read_seqbegin(&file->f_pos_seqlock);
+ pos = file->f_pos;
+ } while (read_seqretry(&file->f_pos_seqlock, seq));
+ return pos;
+#elif BITS_PER_LONG == 32 && defined(CONFIG_PREEMPT)
+ loff_t pos;
+
+ preempt_disable();
+ pos = file->f_pos;
+ preempt_enable();
+ return pos;
+#else
+ return file->f_pos;
+#endif
+}
+
+static inline void file_pos_write(struct file *file, loff_t pos) +{
+#if BITS_PER_LONG == 32 && defined(CONFIG_SMP)
+ write_seqlock(&file->f_pos_seqlock);
+ file->f_pos = pos;
+ write_sequnlock(&file->f_pos_seqlock);
+#elif BITS_PER_LONG == 32 && defined(CONFIG_PREEMPT)
+ preempt_disable();
+ file->f_pos = pos;
+ preempt_enable();
+#else
+ file->f_pos = pos;
+#endif
+}
+
if (offset != file_pos_read(file))
file_pos_write(file, offset);
Compile with:
BITS_PER_LONG=32CONFIG_PREEMPT
The code does:
file_pos_read(...)preempt_disable
....preempt_enable
file_pos_write(...)preempt_disable
....preempt_enable
with the file_pos_update() the code does:
preempt_disable
...
...preempt_enable
Regards Michael
__________________________________________________
Do You Yahoo!?
Poco spazio e tanto spam? Yahoo! Mail ti protegge dallo spam e ti da tanto spazio gratuito per i tuoi file e i messaggi
http://mail.yahoo.it
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Quoted 20 lines · Michael Trimarchi
Hi,
----- Messaggio originale -----
Quoted 2 lines
Da: Nick Piggin <[email protected]>
A: Michael Trimarchi <[email protected]>Cc: Hisashi Hifumi <[email protected]>; [email protected]; [email protected]; [email protected]
Quoted 2 lines
Inviato: Martedì 16 settembre 2008, 16:11:21
Oggetto: Re: [PATCH] VFS: makefile->f_posaccess atomic on 32bit arch...
Quoted 4 lines
Quoted 3 lines
i was thinking about introducing a
file_pos_update()not implemented
using file_pos_read()/file_pos_write() and taking theseq_lockor disabling preemption only one time
Quoted 5 lines
Quoted 3 lines
AFAIK, the only way to do an atomic 64 bit store on 32-bit x86 is "cmpxchg8b" (with lock prefix on SMP). That would be far slower I'm sure.
Maybe it is clear with an example:
I understand and agree.
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Quoted 11 lines · Nick Piggin
Quoted 9 lines
Quoted 7 lines
Quoted 3 lines
i was thinking about introducing a
file_pos_update()not implemented
using file_pos_read()/file_pos_write() and taking theseq_lockor disabling preemption only one timeAFAIK, the only way to do an atomic 64 bit store on 32-bit x86 is "cmpxchg8b" (with lock prefix on SMP). That would be far slower I'm sure.
Maybe it is clear with an example:
I understand and agree.
I modified my patch by introducing file_pos_update().
Comments?
fs/block_dev.c +2 -4
@@ -225,13 +225,11 @@ static loff_t block_llseek(struct file * offset += size;break ;case 1 : - offset += file->f_pos; + offset +=file_pos_read (file); } retval = -EINVAL ;if (offset >=0 && offset <= size) { -if (offset != file->f_pos) { - file->f_pos = offset; - } +file_pos_update (file, offset); retval = offset; }mutex_unlock (&bd_inode->i_mutex);
fs/file_table.c +1 -0
@@ -121,6 +121,7 @@ struct file *get_empty_filp(void) tsk = current;INIT_LIST_HEAD (&f->f_u.fu_list);atomic_long_set (&f->f_count,1 ); +f_pos_ordered_init (f);rwlock_init (&f->f_owner.lock); f->f_uid = tsk->fsuid; f->f_gid = tsk->fsgid;
fs/locks.c +2 -2
@@ -317,7 +317,7 @@ static int flock_to_posix_lock(struct fi start =0 ;break ;case SEEK_CUR : - start = filp->f_pos; + start =file_pos_read (filp);break ;case SEEK_END : start =i_size_read (filp->f_path.dentry->d_inode);
@@ -367,7 +367,7 @@ static int flock64_to_posix_lock(struct start =0 ;break ;case SEEK_CUR : - start = filp->f_pos; + start =file_pos_read (filp);break ;case SEEK_END : start =i_size_read (filp->f_path.dentry->d_inode);
fs/read_write.c +4 -18
@@ -42,15 +42,13 @@ generic_file_llseek_unlocked(struct file offset += inode->i_size;break ;case SEEK_CUR : - offset += file->f_pos; + offset +=file_pos_read (file); } retval = -EINVAL ;if (offset>=0 && offset<=inode->i_sb->s_maxbytes) {/* Special lock needed here? */ -if (offset != file->f_pos) { - file->f_pos = offset; +if (file_pos_update (file, offset)) file->f_version =0 ; - } retval = offset; }return retval;
@@ -83,14 +81,12 @@ loff_t default_llseek(struct file *file, offset +=i_size_read (file->f_path.dentry->d_inode);break ;case SEEK_CUR : - offset += file->f_pos; + offset +=file_pos_read (file); } retval = -EINVAL ;if (offset >=0 ) { -if (offset != file->f_pos) { - file->f_pos = offset; +if (file_pos_update (file, offset)) file->f_version =0 ; - } retval = offset; }unlock_kernel ();
@@ -324,16 +320,6 @@ ssize_t vfs_write(struct file *file, conEXPORT_SYMBOL (vfs_write); -static inline loff_t file_pos_read (struct file *file) -{ -return file->f_pos; -} - -static inline void file_pos_write (struct file *file,loff_t pos) -{ - file->f_pos = pos; -} - asmlinkage ssize_t sys_read(unsigned int fd, char __user * buf, size_t count) {struct file *file;
include/linux/fs.h +80 -0
98 lines not on this page · raw message
Quoted 63 lines
Maybe it is clear with an example:
(quoted diff) +41
-0+/* + * file_pos_read/write is atomic by using sequence lock on 32bit arch. + */ +static inline loff_t file_pos_read(struct file *file) +{ +#if BITS_PER_LONG == 32 && defined(CONFIG_SMP) + loff_t pos; + unsigned int seq; + + do { + seq = read_seqbegin(&file->f_pos_seqlock); + pos = file->f_pos; + } while (read_seqretry(&file->f_pos_seqlock, seq)); + return pos; +#elif BITS_PER_LONG == 32 && defined(CONFIG_PREEMPT) + loff_t pos; + + preempt_disable(); + pos = file->f_pos; + preempt_enable(); + return pos; +#else + return file->f_pos; +#endif +} + +static inline void file_pos_write(struct file *file, loff_t pos) +{ +#if BITS_PER_LONG == 32 && defined(CONFIG_SMP) + write_seqlock(&file->f_pos_seqlock); + file->f_pos = pos; + write_sequnlock(&file->f_pos_seqlock); +#elif BITS_PER_LONG == 32 && defined(CONFIG_PREEMPT) + preempt_disable(); + file->f_pos = pos; + preempt_enable(); +#else + file->f_pos = pos; +#endif +} +if (offset !=
file_pos_read(file))file_pos_write(file, offset);Compile with:
BITS_PER_LONG=32CONFIG_PREEMPTThe code does:
file_pos_read(...)preempt_disable
....preempt_enable
file_pos_write(...)preempt_disable
....preempt_enable
Quoted 9 lines
with the
file_pos_update()the code does:
preempt_disable
...
...preempt_enableRegards Michael
I think file_pos_update() with BITS_PER_LONG=32 && CONFIG_PREEMPT is easy, this is like this.
static inline void file_pos_update(struct file *file, loff_t pos)
{
preempt_disable();
if (pos != file->f_pos)
file->f_pos = pos;
preempt_enable();
}But I think BITS_PER_LONG=32 && CONFIG_SMP version has a problem, this is like this.
static inline void file_pos_update(struct file *file, loff_t pos)
{
write_seqlock(&file->f_pos_seqlock);
if (pos != file->f_pos)
file->f_pos = pos;
write_sequnlock(&file->f_pos_seqlock);
}file_pos_update acquires seqlock only once, but write_seqlock is called whether file->f_pos is overwritten or not.