Re: [Lse-tech] [RFC] dcache scalability patch (2.4.17)

[Christoph Hellwig]

> diff -urN linux-2.4.17-base/fs/dcache.c linux-2.4.17-dc8/fs/dcache.c
> --- linux-2.4.17-base/fs/dcache.c	Fri Dec 21 23:11:55 2001
> +++ linux-2.4.17-dc8/fs/dcache.c	Fri Jul 12 16:18:39 2002
> @@ -25,6 +25,7 @@
>  #include <linux/module.h>
>  
>  #include <asm/uaccess.h>
> +#include <linux/rcupdate.h>

Please try to include <linux/*.h> before <asm/*.h> headers.

> +static void d_callback(void *arg)
> +{
> +	struct dentry * dentry = (struct dentry *)arg;
> +
> +	if (dname_external(dentry)) 
> +		kfree((void *) dentry->d_name.name);
> +	kmem_cache_free(dentry_cache, dentry); 
> +}

why do you cast to void * before calling kfree?

> -	/* dput on a free dentry? */
> -	if (!list_empty(&dentry->d_lru))
> -		BUG();
> +	spin_lock(&dentry->d_lock);
> +        if (atomic_read(&dentry->d_count)) {
> +                spin_unlock(&dentry->d_lock);
> +                spin_unlock(&dcache_lock);
> +                return;
> +        }
> +

Please use tabs instead of eight spaces in kernel code.

Another implementation details is whether we shouldn't spin on a bit of
->d_vfs_flags instead of increasing struct dentry further.  Maybe the
spin_lock_bit interface that wli prototypes might be a godd choise.

Else the patch looks fine to me, although I'm wondering why you target 2.4.17