Message ID | 4C46D1C5.90200@gmail.com |
---|---|
State | Not Applicable, archived |
Headers | show |
Wang Sheng-Hui wrote: > Sorry. regerated the patch, please check it. > I wrapped most code in single pair of spinlock ops for 2 reasons: > 1) get spinlock 2 times seems time consuming > 2) use single pair of spinlock ops can keep "count" > consistent for the shrink operation. 2 pairs may > get some new ces created by other processes. > Sorry, this patch appears to have whitespace cut & paste mangling. More comments below. > Signed-off-by: Wang Sheng-Hui <crosslonelyover@gmail.com> > --- > fs/mbcache.c | 24 ++++++++++++------------ > 1 files changed, 12 insertions(+), 12 deletions(-) > > diff --git a/fs/mbcache.c b/fs/mbcache.c > index ec88ff3..ee57aa3 100644 > --- a/fs/mbcache.c > +++ b/fs/mbcache.c > @@ -201,21 +201,15 @@ mb_cache_shrink_fn(int nr_to_scan, gfp_t gfp_mask) > { > LIST_HEAD(free_list); > struct list_head *l, *ltmp; > + struct mb_cache *cache; > int count = 0; > > - spin_lock(&mb_cache_spinlock); > - list_for_each(l, &mb_cache_list) { > - struct mb_cache *cache = > - list_entry(l, struct mb_cache, c_cache_list); > - mb_debug("cache %s (%d)", cache->c_name, > - atomic_read(&cache->c_entry_count)); > - count += atomic_read(&cache->c_entry_count); > - } > mb_debug("trying to free %d entries", nr_to_scan); > - if (nr_to_scan == 0) { > - spin_unlock(&mb_cache_spinlock); > + > + spin_lock(&mb_cache_spinlock); > + if (nr_to_scan == 0) > goto out; > - } > + > while (nr_to_scan-- && !list_empty(&mb_cache_lru_list)) { > struct mb_cache_entry *ce = > list_entry(mb_cache_lru_list.next, > @@ -223,12 +217,18 @@ mb_cache_shrink_fn(int nr_to_scan, gfp_t gfp_mask) > list_move_tail(&ce->e_lru_list, &free_list); > __mb_cache_entry_unhash(ce); > } > - spin_unlock(&mb_cache_spinlock); you can't do this because > list_for_each_safe(l, ltmp, &free_list) { > __mb_cache_entry_forget(list_entry(l, struct mb_cache_entry, this takes the spinlock too and you'll deadlock. Did you test this patch? -Eric > e_lru_list), gfp_mask); > } > out: > + list_for_each_entry(cache, &mb_cache_list, c_cache_list) { > + mb_debug("cache %s (%d)", cache->c_name, > + atomic_read(&cache->c_entry_count)); > + count += atomic_read(&cache->c_entry_count); > + } > + spin_unlock(&mb_cache_spinlock); > + > return (count / 100) * sysctl_vfs_cache_pressure; > } > -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Al, here is an mbcache cleanup and then a fixed version of Shenghui's minor shrinker function fix. The patches have survived functional testing here. This seems slightly too much for kernel-janitors, so could you please take the patches? Thanks, Andreas Andreas Gruenbacher (2): mbcache: Remove unused features mbcache: fix shrinker function return value fs/ext2/xattr.c | 12 ++-- fs/ext3/xattr.c | 12 ++-- fs/ext4/xattr.c | 12 ++-- fs/mbcache.c | 168 ++++++++++++++--------------------------------- include/linux/mbcache.h | 20 ++---- 5 files changed, 70 insertions(+), 154 deletions(-)
On Wed, Jul 21, 2010 at 07:57:20PM +0200, Andreas Gruenbacher wrote: > Al, > > here is an mbcache cleanup and then a fixed version of Shenghui's minor > shrinker function fix. The patches have survived functional testing > here. > > This seems slightly too much for kernel-janitors, so could you please > take the patches? Done. -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/fs/mbcache.c b/fs/mbcache.c index ec88ff3..ee57aa3 100644 --- a/fs/mbcache.c +++ b/fs/mbcache.c @@ -201,21 +201,15 @@ mb_cache_shrink_fn(int nr_to_scan, gfp_t gfp_mask) { LIST_HEAD(free_list); struct list_head *l, *ltmp; + struct mb_cache *cache; int count = 0; - spin_lock(&mb_cache_spinlock); - list_for_each(l, &mb_cache_list) { - struct mb_cache *cache = - list_entry(l, struct mb_cache, c_cache_list); - mb_debug("cache %s (%d)", cache->c_name, - atomic_read(&cache->c_entry_count)); - count += atomic_read(&cache->c_entry_count); - } mb_debug("trying to free %d entries", nr_to_scan); - if (nr_to_scan == 0) { - spin_unlock(&mb_cache_spinlock); + + spin_lock(&mb_cache_spinlock); + if (nr_to_scan == 0) goto out; - } + while (nr_to_scan-- && !list_empty(&mb_cache_lru_list)) { struct mb_cache_entry *ce = list_entry(mb_cache_lru_list.next, @@ -223,12 +217,18 @@ mb_cache_shrink_fn(int nr_to_scan, gfp_t gfp_mask) list_move_tail(&ce->e_lru_list, &free_list); __mb_cache_entry_unhash(ce); } - spin_unlock(&mb_cache_spinlock); list_for_each_safe(l, ltmp, &free_list) { __mb_cache_entry_forget(list_entry(l, struct mb_cache_entry, e_lru_list), gfp_mask); } out: + list_for_each_entry(cache, &mb_cache_list, c_cache_list) { + mb_debug("cache %s (%d)", cache->c_name, + atomic_read(&cache->c_entry_count)); + count += atomic_read(&cache->c_entry_count); + } + spin_unlock(&mb_cache_spinlock); + return (count / 100) * sysctl_vfs_cache_pressure; }
Sorry. regerated the patch, please check it. I wrapped most code in single pair of spinlock ops for 2 reasons: 1) get spinlock 2 times seems time consuming 2) use single pair of spinlock ops can keep "count" consistent for the shrink operation. 2 pairs may get some new ces created by other processes. Signed-off-by: Wang Sheng-Hui <crosslonelyover@gmail.com> --- fs/mbcache.c | 24 ++++++++++++------------ 1 files changed, 12 insertions(+), 12 deletions(-)