Block: Check major number before allocate the buffer in register_blkdev()
PATCH 1/1 Block: Check major number before allocate the buffer in register_blkdev()
Check major number before allocate the buffer, if the major number is not exist, and the register_blkdev() called many times, kmalloc()/kfree() will be no need to invoked many times. So check the major number before use kmalloc() to allocate the buffer will be better.
--- block/genhd.c | 25 ++++++++++++------------- 1 files changed, 12 insertions(+), 13 deletions(-)
block/genhd.c +12 -13
@@ -300,30 +300,29 @@ int register_blkdev(unsigned int major, const char *name) ret = major; } - p =kmalloc (sizeof (struct blk_major_name ),GFP_KERNEL ); -if (p ==NULL ) { - ret = -ENOMEM ; - goto out; - } - - p->major = major; -strlcpy (p->name, name,sizeof (p->name)); - p->next =NULL ; index =major_to_index (major);for (n = &major_names[index]; *n; n = &(*n)->next) {if ((*n)->major == major)break ; } -if (!*n) +if (!*n) { + p =kmalloc (sizeof (struct blk_major_name ),GFP_KERNEL ); +if (p ==NULL ) { + ret = -ENOMEM ; + goto out; + } + + p->major = major; +strlcpy (p->name, name,sizeof (p->name)); + p->next =NULL ; *n = p; -else + } +else { ret = -EBUSY ; -if (ret <0 ) {printk ("register_blkdev: cannot get major %d for %s\n" , major, name); -kfree (p); } out:mutex_unlock (&block_class_lock);
Signature · 1 line
1.6.5.3
Quoted 4 lines · On Mon, May 10 2010, [email protected] wrote:
Check major number before allocate the buffer, if the major number is not exist, and the
register_blkdev()called many times, kmalloc()/kfree() will be no need to invoked many times. So check the major number before usekmalloc()to allocate the buffer will be better.
This would generally be sound advice for performance oriented code, but I can't see it making any difference here.
Signature · 1 line
Jens Axboe
the original code use kmalloc() to allocate struct blk_major_name buffer first, then find the major number in major_names array. if found it, it will kfree() the unused struct blk_major_name buffer, if register_blkdev() called many times like:register_blkdev(22, "aa");
....register_blkdev(22, "aa");
kmalloc()/kfree() will be no need to invoked many times, my point is that find the major number first, then allocate the buffer will be better. this patch can handle the special case.
Quoted 13 lines · Jens Axboe
Quoted 4 lines · On Mon, May 10 2010, [email protected] wrote:
Check major number before allocate the buffer, if the major number is not exist, and the
register_blkdev()called many times, kmalloc()/kfree() will be no need to invoked many times. So check the major number before usekmalloc()to allocate the buffer will be better.This would generally be sound advice for performance oriented code, but I can't see it making any difference here.
--
Jens Axboe
Please don't top post when replying to emails, I fixed this one up for you.
Quoted 10 lines · wzt wzt
Quoted 8 lines · Jens Axboe
Quoted 4 lines · On Mon, May 10 2010, [email protected] wrote:
Check major number before allocate the buffer, if the major number is not exist, and the
register_blkdev()called many times, kmalloc()/kfree() will be no need to invoked many times. So check the major number before usekmalloc()to allocate the buffer will be better.This would generally be sound advice for performance oriented code, but I can't see it making any difference here.
Quoted 10 lines
the original code use
kmalloc()to allocate structblk_major_namebuffer first, then find the major number inmajor_namesarray. if found it, it willkfree()the unused structblk_major_namebuffer, ifregister_blkdev()called many times like:register_blkdev(22, "aa");
....register_blkdev(22, "aa");
kmalloc()/kfree() will be no need to invoked many times, my point is that find the major number first, then allocate the buffer will be better. this patch can handle the special case.
Yes I realize how it works and what your patch does, my point is that it seems pointless to change code like that. Your 'test case' above isn't a valid one. If register_blkdev() was called tons of times per second and hence would be a hot code path, and it additionally most of the time ended up freeing the buffer, then there would be a case for changing it.
Signature · 1 line
Jens Axboe