Re: [PATCH] dentry/inode accounting for vm_enough_mem()

[Andrew Morton <akpm@digeo.com>]

Dave Hansen <haveblue@us.ibm.com> wrote:
>
>   struct dentry_stat_t {
>   	int nr_dentry;
>   	int nr_unused;
>  +	atomic_t nr_alloced;
>   	int age_limit;          /* age in seconds */
>   	int want_pages;         /* pages requested by system */
>   	int dummy[2];

We're not at liberty to do this because /proc/sys/fs/dentry-state and
inode-state are implemented assuming that these structs are an array of
integers.  It'll screw up if the architecture's "int" and "atomic_t"
representations are different.

Probably you can just make this an integer and add a spinlock for it, or
not place it in dentry_stat.

Seems otherwise OK though.  Thanks.