Re: [PATCH v1 1/8] Deferred batching of dput()

[Mike Waychison <mikew@google.com>]

Evgeniy Polyakov wrote:
> Hi Mike.
> 
> On Fri, Jan 16, 2009 at 06:29:42PM -0800, Mike Waychison (mikew@google.com) wrote:
>> +static void postpone_dput(struct dentry *dentry)
>> +{
>> +	struct postponed_dentries *ppd, *new_ppd;
>> +
>> +again:
>> +	ppd = get_cpu_var(postponed_dentries);
>> +	if (!pending_dput_full(ppd)) {
>> +		add_pending_dput(ppd, dentry);
>> +		put_cpu_var(postponed_dentries);
>> +		return;
>> +	}
>> +
>> +	/* need to flush out existing pending dentries. */
>> +	put_cpu_var(postponed_dentries);
>> +	/* Allocate more space.. */
>> +	new_ppd = new_postponed_dentries();
>> +	if (!new_ppd) {
>> +		/* Take the slow path, memory is low */
>> +		struct postponed_dentries_onstack ppd_onstack;
>> +		struct postponed_dentries *ppd;
>> +
>> +		ppd = init_ppd_onstack(&ppd_onstack);
>> +		add_pending_dput(ppd, dentry);
>> +		process_postponed_dentries(ppd);
>> +		return;
>> +	}
> 
> Why don't you want just to put the dentry in the lowmem condition?
> 

You're right.  This path could just use the original dput path.  I'll 
fix it up in the next version.