Re: [PATCH] 2/5: Device-mapper: kcopyd

[Andrew Morton <akpm@osdl.org>]

Alasdair G Kergon <agk@redhat.com> wrote:
>
> kcopyd

I dunno about the rest of the people around here but I for one do not know
what kcopyd does, nor why it is being added to the kernel.  The changelog
was your chance to enlighten us - it is a shame nobody took the five
minutes to prepare a description.


> +static void drop_pages(struct page_list *pl)
> +{
> +	struct page_list *next;
> +
> +	while (pl) {
> +		next = pl->next;
> +		free_pl(pl);
> +		pl = next;
> +	}
> +}

What is the page pool for?  It is unfortunate that there is no suitable
library code for managing this - it is a fairly common pattern.  Maybe we
could massage mempools in some manner.

Why are the pooled pages locked?

> +static LIST_HEAD(_complete_jobs);
> +static LIST_HEAD(_io_jobs);
> +static LIST_HEAD(_pages_jobs);
> +
> +static int __init jobs_init(void)
> +{
> +	INIT_LIST_HEAD(&_complete_jobs);
> +	INIT_LIST_HEAD(&_io_jobs);
> +	INIT_LIST_HEAD(&_pages_jobs);

Do these lists need to be initialised a second time?

> +		memcpy(sub_job, job, sizeof(*job));
> +	memcpy(&job->source, from, sizeof(*from));

Structure assignments are nice.  If the struct is small the compiler will
do an element-by-element copy.  At some compiler-determined breakpoint it
will do a memcpy.  And it's typesafe.