Re: [PATCH] x86: split e820 reserved entries record to late v4 - fix

["Yinghai Lu" <yhlu.kernel@gmail.com>]

On Fri, Aug 29, 2008 at 9:05 AM, Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
>
> On Fri, 29 Aug 2008, Yinghai Lu wrote:
>>
>> try to insert_resource second time, by expand the resource...
>
> I would hold off on this unless it's shown to actually be needed. And _if_
> it is needed, I would just make a new function for doing this all:
> "insert_resource_expand_to_fit()"
>
> That said, I think the insert_resource()/__insert_resource() change is
> pretty ok. However, it doesn't follow the rules, and is racy. The rules
> for resources are:
>
>  - the "internal" version (with the "__" prepended) is static to
>   resource.c, because it must not be called from outside, which is in
>   turn because:
>
>  - it must be called with the lock taken by the caller, because otherwise
>   returning a "struct resource *" is racy - the resource is not protected
>   by anything!
>
> So the "insert_resource_expand_to_fit()" thing would look something like
> this:
>
>        void insert_resource_expand_to_fit(struct resource *root, struct resource *new)
>        {
>                write_lock(&resource_lock);
>                while (new->start && new->parent) {
>                        struct resource *conflict;
>
>                        conflict = __insert_resource(root, new);
>                        if (!conflict)
>                                break;
>                        if (conflict->start < new->start)
>                                new->start = conflict->start;
>                        if (conflict->end > new->end)
>                                new->end = conflict->end;
>                }
>                write_unlock(&resource_lock);
>        }
>
> but the above is obviously _totally_ untested.
>

good, will build one test stub to test it.

YH