Re: [PATCH] vfs: add simple direct-mapped dcache lookup front-end

[Randy Dunlap <rdunlap@infradead.org>]

Hi,
Just a small typo in Kconfig below..

On 06/11/16 16:51, George Spelvin wrote:
> 
>  fs/Kconfig  |  28 ++++++++
>  fs/dcache.c | 207 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++--
>  2 files changed, 231 insertions(+), 4 deletions(-)
> 
> diff --git a/fs/Kconfig b/fs/Kconfig
> index b8fcb416..3b111b77 100644
> --- a/fs/Kconfig
> +++ b/fs/Kconfig
> @@ -8,6 +8,34 @@ menu "File systems"
>  config DCACHE_WORD_ACCESS
>         bool
>  
> +config L1_DCACHE_BITS
> +	int "Dcache level-1 cache size (bits)"
> +	range 0 20
> +	default 0 if !EXPERT
> +	default 0 if NUMA
> +	default 10 if BASE_SMALL
> +	default 13
> +	help
> +	  The Linus kernel maintains a large cache of "dentries"

	      Linux

> +	  (directory entries) for the performance-critical task of
> +	  converting file names to inodes.  This option enables a smaller
> +	  direct-mapped "level-1 cache" in front of the main dcache.
> +
> +	  (This software "dcache" is quite different from the CPU's data
> +	  cache, or "D-cache".	Sorry for the confusingly similar names.)
> +
> +	  This option specifies the size of this cache, as a power of 2.
> +	  For example, 13 means 2^13 = 8192 entries in the L1 dcache.
> +	  Specify 0 to turn off the L1 dcache entirely.
> +
> +	  The cost of enabling this is one pointer per entry, plus a
> +	  small amount of code.
> +
> +	  This is an experimental feature which hopes to speed up
> +	  single-socket machines.  On larger systems, the extra updates
> +	  generated by the L1 dcache probably cause too much cache-line
> +	  bouncing to be worth it.
> +
>  if BLOCK
>  
>  source "fs/ext2/Kconfig"


-- 
~Randy