[WIP RFC PATCH] fs: hide names_cachep behind runtime_const machinery

[WIP RFC PATCH] fs: hide names_cachep behind runtime_const machinery

[Mateusz Guzik]

All path lookups end up allocating and freeing a buffer. The namei cache
is created and at boot time and remains constant, meaning there is no
reason to spend a cacheline to load the pointer.

I verified this boots on x86-64.

The problem is that when building I get the following:
ld: warning: orphan section `runtime_ptr_names_cachep' from `vmlinux.o' being placed in section `runtime_ptr_names_cachep'

I don't know what's up with that yet, but I will sort it out. Before I
put any effort into it I need to know if the idea looks fine.

---
 fs/dcache.c        | 1 +
 include/linux/fs.h | 5 +++--
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/fs/dcache.c b/fs/dcache.c
index 035cccbc9276..786d09798313 100644
--- a/fs/dcache.c
+++ b/fs/dcache.c
@@ -3265,6 +3265,7 @@ void __init vfs_caches_init(void)
 {
 	names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
 			SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
+	runtime_const_init(ptr, names_cachep);
 
 	dcache_init();
 	inode_init();
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 68c4a59ec8fb..08ea27340309 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -2960,8 +2960,9 @@ extern void __init vfs_caches_init(void);
 
 extern struct kmem_cache *names_cachep;
 
-#define __getname()		kmem_cache_alloc(names_cachep, GFP_KERNEL)
-#define __putname(name)		kmem_cache_free(names_cachep, (void *)(name))
+#define __const_names_cachep	runtime_const_ptr(names_cachep)
+#define __getname()		kmem_cache_alloc(__const_names_cachep, GFP_KERNEL)
+#define __putname(name)		kmem_cache_free(__const_names_cachep, (void *)(name))
 
 extern struct super_block *blockdev_superblock;
 static inline bool sb_is_blkdev_sb(struct super_block *sb)
-- 
2.34.1

Re: [WIP RFC PATCH] fs: hide names_cachep behind runtime_const machinery

[Mateusz Guzik]

On Wed, Oct 29, 2025 at 2:55 PM Mateusz Guzik <mjguzik@gmail.com> wrote:
>
> All path lookups end up allocating and freeing a buffer. The namei cache
> is created and at boot time and remains constant, meaning there is no
> reason to spend a cacheline to load the pointer.
>
> I verified this boots on x86-64.
>
> The problem is that when building I get the following:
> ld: warning: orphan section `runtime_ptr_names_cachep' from `vmlinux.o' being placed in section `runtime_ptr_names_cachep'
>
> I don't know what's up with that yet, but I will sort it out. Before I
> put any effort into it I need to know if the idea looks fine.
>

The good news is that Pedro Falcato stepped in and found a spot I
failed to add the var to, this clears up the warning.

However, after further testing I found that kernel modules are not
being patched. I'll be sending a v2 with some more commentary.

> ---
>  fs/dcache.c        | 1 +
>  include/linux/fs.h | 5 +++--
>  2 files changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/fs/dcache.c b/fs/dcache.c
> index 035cccbc9276..786d09798313 100644
> --- a/fs/dcache.c
> +++ b/fs/dcache.c
> @@ -3265,6 +3265,7 @@ void __init vfs_caches_init(void)
>  {
>         names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
>                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
> +       runtime_const_init(ptr, names_cachep);
>
>         dcache_init();
>         inode_init();
> diff --git a/include/linux/fs.h b/include/linux/fs.h
> index 68c4a59ec8fb..08ea27340309 100644
> --- a/include/linux/fs.h
> +++ b/include/linux/fs.h
> @@ -2960,8 +2960,9 @@ extern void __init vfs_caches_init(void);
>
>  extern struct kmem_cache *names_cachep;
>
> -#define __getname()            kmem_cache_alloc(names_cachep, GFP_KERNEL)
> -#define __putname(name)                kmem_cache_free(names_cachep, (void *)(name))
> +#define __const_names_cachep   runtime_const_ptr(names_cachep)
> +#define __getname()            kmem_cache_alloc(__const_names_cachep, GFP_KERNEL)
> +#define __putname(name)                kmem_cache_free(__const_names_cachep, (void *)(name))
>
>  extern struct super_block *blockdev_superblock;
>  static inline bool sb_is_blkdev_sb(struct super_block *sb)
> --
> 2.34.1
>