Skip to contentSkip to conversation

shrink_caches inconsistancy

PATCH Craig Christophel 2 participants 3 messages Dec 5, 2001 – Dec 6, 2001 VFS

lore
Share & tools
lore
#0 lore raw

This patch makes the comments match for icache,dcache,dqcache shrink
functions. Initially the comment stated that a priority of 0 could be used, but after looking into mm/vmscan.c::shrink_caches this cannot be true. So the comment now states that 1 is the highest priority. This appears __really true as at priority 1 all of the cache possible is removed.

Also shrink_dqcache_memory now uses the count variable like everyone else.

Possibly incorrect __GFP_FS check added to the dqcache function. but again
consistancy is my goal.

Another dqcache issue in that the dqcache was being shrunk at priority+1
rather than at priority, this looked suspect, and with no comment around the code, it to has been remanded to consistancy.

Well, here goes. Any comments and/or b#$chslaps welcome. This is my first
attempted patch after studying the VFS layer for a while, so please be (fairly) kind.

Craig.

fs/dcache.c Wed Dec 5 16:13:38 2001 renamed from fs/dcache.c Sat Dec 1 17:21:03 2001 +1 -1
diff -urN linux/fs/dcache.c linux.mt/fs/dcache.c
@@ -543,7 +543,7 @@
  * too much.
  *
  * Priority:
- *   0 - very urgent: shrink everything
+ *   1 - very urgent: shrink everything
  *  ...
  *   6 - base-level: try to shrink a bit.
  */
fs/dquot.c Wed Dec 5 17:07:23 2001 renamed from fs/dquot.c Thu Nov 22 13:38:31 2001 +20 -1
diff -urN linux/fs/dquot.c linux.mt/fs/dquot.c
@@ -407,11 +407,30 @@
                head = free_dquots.prev;
        }
 }
+/*
+ * This is called from kswapd when we think we need some
+ * more memory, but aren't really sure how much. So we
+ * carefully try to free a _bit_ of our dqcache, but not
+ * too much.
+ *
+ * Priority:
+ *   1 - very urgent: shrink everything
+ *   ...
+ *   6 - base-level: try to shrink a bit.
+ */
 
 int shrink_dqcache_memory(int priority, unsigned int gfp_mask)
 {
+       int count = 0;
+
        lock_kernel();
-       prune_dqcache(nr_free_dquots / (priority + 1));
+
+       if (!(gfp_mask & __GFP_FS))
+               return 0;
+
+       count = nr_free_dquots / priority;
+
+       prune_dqcache(count);
        unlock_kernel();
        kmem_cache_shrink(dquot_cachep);
        return 0;
fs/inode.c Wed Dec 5 16:37:25 2001 renamed from fs/inode.c Sat Dec 1 17:21:03 2001 +11 -1
diff -urN linux/fs/inode.c linux.mt/fs/inode.c
@@ -707,7 +707,17 @@
        if (goal)
                schedule_task(&unused_inodes_flush_task);
 }
-
+/*
+ * This is called from kswapd when we think we need some
+ * more memory, but aren't really sure how much. So we
+ * carefully try to free a _bit_ of our icache, but not
+ * too much.
+ *
+ * Priority:
+ *   1 - very urgent: shrink everything
+ *  ...
+ *   6 - base-level: try to shrink a bit.
+ */
 int shrink_icache_memory(int priority, int gfp_mask)
 {
        int count = 0;
reviewer
#1 lore raw

Hello,

Quoted 10 lines

This patch makes the comments match for icache,dcache,dqcache shrink
functions. Initially the comment stated that a priority of 0 could be used, but after looking into mm/vmscan.c::shrink_caches this cannot be true. So the comment now states that 1 is the highest priority. This appears __really true as at priority 1 all of the cache possible is removed.

Also shrink_dqcache_memory now uses the count variable like everyone else.

Possibly incorrect __GFP_FS check added to the dqcache function. but again
consistancy is my goal.

This check really isn't needed for shrink_dqcache() function. This function can
never recurse into fs so there's no need to have __GFP_FS set.

Quoted 3 lines

Another dqcache issue in that the dqcache was being shrunk at priority+1
rather than at priority, this looked suspect, and with no comment around the code, it to has been remanded to consistancy.

OK :). If I remeber well I saw 'priority' could be 0 somewhere in the comment
and so I added +1 to avoid division by zero. But you're right that code in vmscan.c actually never calls the functions with priority == 0.

Honza

#2 lore raw
In reply to Craig Christophel

Well here is the fixed patch with __GFP_FS check removed from dqcache. Thanks for the note.

Craig

Quoted 5 lines
Quoted 2 lines

Possibly incorrect __GFP_FS check added to the dqcache function. but
again consistancy is my goal.

This check really isn't needed for shrink_dqcache() function. This
function can never recurse into fs so there's no need to have __GFP_FS set.

fs/dcache.c Wed Dec 5 18:22:26 2001 renamed from fs/dcache.c Wed Dec 5 18:22:42 2001 +1 -1
diff -urN linux/fs/dcache.c linux.mt/fs/dcache.c
@@ -543,7 +543,7 @@
  * too much.
  *
  * Priority:
- *   0 - very urgent: shrink everything
+ *   1 - very urgent: shrink everything
  *  ...
  *   6 - base-level: try to shrink a bit.
  */
fs/dquot.c Thu Dec 6 10:09:24 2001 renamed from fs/dquot.c Wed Dec 5 18:22:42 2001 +18 -1
diff -urN linux/fs/dquot.c linux.mt/fs/dquot.c
@@ -407,11 +407,28 @@
                head = free_dquots.prev;
        }
 }
+/*
+ * This is called from kswapd when we think we need some
+ * more memory, but aren't really sure how much. So we
+ * carefully try to free a _bit_ of our dqcache, but not
+ * too much.
+ *
+ * Priority:
+ *   1 - very urgent: shrink everything
+ *   ...
+ *   6 - base-level: try to shrink a bit.
+ */
 
 int shrink_dqcache_memory(int priority, unsigned int gfp_mask)
 {
+       int count = 0;
+
+
        lock_kernel();
-       prune_dqcache(nr_free_dquots / (priority + 1));
+
+       count = nr_free_dquots / priority;
+
+       prune_dqcache(count);
        unlock_kernel();
        kmem_cache_shrink(dquot_cachep);
        return 0;
fs/inode.c Wed Dec 5 18:22:26 2001 renamed from fs/inode.c Wed Dec 5 18:22:42 2001 +11 -1
diff -urN linux/fs/inode.c linux.mt/fs/inode.c
@@ -707,7 +707,17 @@
        if (goal)
                schedule_task(&unused_inodes_flush_task);
 }
-
+/*
+ * This is called from kswapd when we think we need some
+ * more memory, but aren't really sure how much. So we
+ * carefully try to free a _bit_ of our icache, but not
+ * too much.
+ *
+ * Priority:
+ *   1 - very urgent: shrink everything
+ *  ...
+ *   6 - base-level: try to shrink a bit.
+ */
 int shrink_icache_memory(int priority, int gfp_mask)
 {
        int count = 0;