[PATCH] Remove vfs_init_caches_early()

[PATCH] Remove vfs_init_caches_early()

[Rusty Russell]

vfs_init_caches_early() does nothing on IA-64 and x86-64 (unless you
turn off CONFIG_NUMA): hashdist is set by default on these platforms.

Maybe some obscure feature which requires VFS to be set up v. early
(hence is broken in this configuration), or something arch-specific
(which seems unlikely).  Most likely this is a relic.

Unless someone has inside knowledge, the easiest way to find out is to
rip it out and see if anyone screams.  Boots fine on my 32-bit x86 box here.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
---
 Documentation/kernel-parameters.txt |    2 +-
 fs/dcache.c                         |   34 ----------------------------------
 fs/inode.c                          |   31 -------------------------------
 include/linux/fs.h                  |    2 --
 init/main.c                         |    1 -
 5 files changed, 1 insertion(+), 69 deletions(-)

diff -r e4dd27867542 Documentation/kernel-parameters.txt
--- a/Documentation/kernel-parameters.txt	Fri Jan 18 09:59:22 2008 +1100
+++ b/Documentation/kernel-parameters.txt	Fri Jan 18 10:07:13 2008 +1100
@@ -662,7 +662,7 @@ and is between 256 and 4096 characters. 
 
 	hashdist=	[KNL,NUMA] Large hashes allocated during boot
 			are distributed across NUMA nodes.  Defaults on
-			for IA-64, off otherwise.
+			for IA-64 and x86-64, off otherwise.
 			Format: 0 | 1 (for off | on)
 
 	hcl=		[IA-64] SGI's Hardware Graph compatibility layer
diff -r e4dd27867542 fs/dcache.c
--- a/fs/dcache.c	Fri Jan 18 09:59:22 2008 +1100
+++ b/fs/dcache.c	Fri Jan 18 10:07:13 2008 +1100
@@ -2082,30 +2082,6 @@ static int __init set_dhash_entries(char
 }
 __setup("dhash_entries=", set_dhash_entries);
 
-static void __init dcache_init_early(void)
-{
-	int loop;
-
-	/* If hashes are distributed across NUMA nodes, defer
-	 * hash allocation until vmalloc space is available.
-	 */
-	if (hashdist)
-		return;
-
-	dentry_hashtable =
-		alloc_large_system_hash("Dentry cache",
-					sizeof(struct hlist_head),
-					dhash_entries,
-					13,
-					HASH_EARLY,
-					&d_hash_shift,
-					&d_hash_mask,
-					0);
-
-	for (loop = 0; loop < (1 << d_hash_shift); loop++)
-		INIT_HLIST_HEAD(&dentry_hashtable[loop]);
-}
-
 static void __init dcache_init(void)
 {
 	int loop;
@@ -2119,10 +2095,6 @@ static void __init dcache_init(void)
 		SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
 	
 	register_shrinker(&dcache_shrinker);
-
-	/* Hash may have been set up in dcache_init_early */
-	if (!hashdist)
-		return;
 
 	dentry_hashtable =
 		alloc_large_system_hash("Dentry cache",
@@ -2145,12 +2117,6 @@ struct kmem_cache *filp_cachep __read_mo
 struct kmem_cache *filp_cachep __read_mostly;
 
 EXPORT_SYMBOL(d_genocide);
-
-void __init vfs_caches_init_early(void)
-{
-	dcache_init_early();
-	inode_init_early();
-}
 
 void __init vfs_caches_init(unsigned long mempages)
 {
diff -r e4dd27867542 fs/inode.c
--- a/fs/inode.c	Fri Jan 18 09:59:22 2008 +1100
+++ b/fs/inode.c	Fri Jan 18 10:07:13 2008 +1100
@@ -1369,33 +1369,6 @@ static int __init set_ihash_entries(char
 }
 __setup("ihash_entries=", set_ihash_entries);
 
-/*
- * Initialize the waitqueues and inode hash table.
- */
-void __init inode_init_early(void)
-{
-	int loop;
-
-	/* If hashes are distributed across NUMA nodes, defer
-	 * hash allocation until vmalloc space is available.
-	 */
-	if (hashdist)
-		return;
-
-	inode_hashtable =
-		alloc_large_system_hash("Inode-cache",
-					sizeof(struct hlist_head),
-					ihash_entries,
-					14,
-					HASH_EARLY,
-					&i_hash_shift,
-					&i_hash_mask,
-					0);
-
-	for (loop = 0; loop < (1 << i_hash_shift); loop++)
-		INIT_HLIST_HEAD(&inode_hashtable[loop]);
-}
-
 void __init inode_init(void)
 {
 	int loop;
@@ -1408,10 +1381,6 @@ void __init inode_init(void)
 					 SLAB_MEM_SPREAD),
 					 init_once);
 	register_shrinker(&icache_shrinker);
-
-	/* Hash may have been set up in inode_init_early */
-	if (!hashdist)
-		return;
 
 	inode_hashtable =
 		alloc_large_system_hash("Inode-cache",
diff -r e4dd27867542 include/linux/fs.h
--- a/include/linux/fs.h	Fri Jan 18 09:59:22 2008 +1100
+++ b/include/linux/fs.h	Fri Jan 18 10:07:13 2008 +1100
@@ -302,7 +302,6 @@ struct vfsmount;
 struct vfsmount;
 
 extern void __init inode_init(void);
-extern void __init inode_init_early(void);
 extern void __init mnt_init(void);
 extern void __init files_init(unsigned long);
 
@@ -1546,7 +1545,6 @@ extern char * getname(const char __user 
 extern char * getname(const char __user *);
 
 /* fs/dcache.c */
-extern void __init vfs_caches_init_early(void);
 extern void __init vfs_caches_init(unsigned long);
 
 extern struct kmem_cache *names_cachep;
diff -r e4dd27867542 init/main.c
--- a/init/main.c	Fri Jan 18 09:59:22 2008 +1100
+++ b/init/main.c	Fri Jan 18 10:07:13 2008 +1100
@@ -604,7 +604,6 @@ asmlinkage void __init start_kernel(void
 		initrd_start = 0;
 	}
 #endif
-	vfs_caches_init_early();
 	cpuset_init_early();
 	mem_init();
 	kmem_cache_init();

Re: [PATCH] Remove vfs_init_caches_early()

[Rusty Russell]

On Friday 18 January 2008 10:41:21 Rusty Russell wrote:
> vfs_init_caches_early() does nothing on IA-64 and x86-64 (unless you
> turn off CONFIG_NUMA): hashdist is set by default on these platforms.
>
> Maybe some obscure feature which requires VFS to be set up v. early
> (hence is broken in this configuration), or something arch-specific
> (which seems unlikely).  Most likely this is a relic.
>
> Unless someone has inside knowledge, the easiest way to find out is to
> rip it out and see if anyone screams.  Boots fine on my 32-bit x86 box
> here.

Ok, Chris Wright pointed me to the justification: it was because on 64G 
powerpc machines, MAX_ORDER of 11 was too small to allocate sufficient 
hashes.

See 
http://git.kernel.org/?p=linux/kernel/git/torvalds/old-2.6-bkcvs.git;a=commitdiff;h=50c2cb5183e4f101f923212e39a58a66a530cf77

Since MAX_ORDER is now 13 on powerpc, I think we're good.

Thanks Cdub!
Rusty.

Re: [PATCH] Remove vfs_init_caches_early()

[Nick Piggin]

On Friday 18 January 2008 10:41, Rusty Russell wrote:
> vfs_init_caches_early() does nothing on IA-64 and x86-64 (unless you
> turn off CONFIG_NUMA): hashdist is set by default on these platforms.
>
> Maybe some obscure feature which requires VFS to be set up v. early
> (hence is broken in this configuration), or something arch-specific
> (which seems unlikely).

The use of hugepage mappings for your hashes rather than vmalloc
memory? Or did I misunderstand you? ... bootmem is needed for very
large physically contiguous allocations.