[PATCH] Allow a larger buffer for writes to cpuset files

[PATCH] Allow a larger buffer for writes to cpuset files

[Paul Menage]

When using fake NUMA setup, the number of memory nodes can greatly
exceed the number of CPUs. So the current limit in
cpuset_common_file_write() is insufficient. PAGE_SIZE is still a bit
of an arbitrary limit, but gives more breathing room.

Signed-off-by: Paul Menage <menage@google.com>

--- 2.6.19-rc2.orig/kernel/cpuset.c
+++ 2.6.19-rc2/kernel/cpuset.c
@@ -1292,7 +1292,7 @@ static ssize_t cpuset_common_file_write(
 	int retval = 0;

 	/* Crude upper limit on largest legitimate cpulist user might write. */
-	if (nbytes > 100 + 6 * NR_CPUS)
+	if (nbytes >= PAGE_SIZE)
 		return -E2BIG;

 	/* +1 for nul-terminator */

Re: [PATCH] Allow a larger buffer for writes to cpuset files

[Paul Jackson]

Paul M wrote:
> @@ -1292,7 +1292,7 @@ static ssize_t cpuset_common_file_write(
>  	int retval = 0;
> 
>  	/* Crude upper limit on largest legitimate cpulist user might write. */
> -	if (nbytes > 100 + 6 * NR_CPUS)
> +	if (nbytes >= PAGE_SIZE)

On a system with 4 Kbyte pages, this imposes a tighter limit
if NR_CPUS is > 666.

On a system with 16 Kbyte pages, this imposes a tighter limit
if NR_CPUS is > 2714.

Those CPU counts don't look that futuristic to me.

And it makes the limit more aribitrary ... like what the heck does
that check have to do with PAGE_SIZE ?

How about coding for the possibility that either NR_CPUS or MAX_NUMNODES
is larger, and removing 'cpu' from the comment:

	/* Crude upper limit on largest legitimate list user might write. */
	if (nbytes > 100 + 6 * max(NR_CPUS, MAX_NUMNODES))

-- 
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <pj@sgi.com> 1.925.600.0401

Re: [PATCH] Allow a larger buffer for writes to cpuset files

[Paul Menage]

On 10/30/06, Paul Jackson <pj@sgi.com> wrote:
>
> How about coding for the possibility that either NR_CPUS or MAX_NUMNODES
> is larger, and removing 'cpu' from the comment:
>
>         /* Crude upper limit on largest legitimate list user might write. */
>         if (nbytes > 100 + 6 * max(NR_CPUS, MAX_NUMNODES))
>

Yes, that does seem like a better idea.

Paul