[Intel-gfx] [PATCH 3/4] drm/i915: Add support for stealing purgable stolen pages

[Tvrtko Ursulin]

On 09/23/2015 10:28 AM, Daniel Vetter wrote:
> On Wed, Sep 16, 2015 at 02:31:38PM +0530, Ankitprasad Sharma wrote:
>> On Tue, 2015-09-15 at 16:14 +0100, Tvrtko Ursulin wrote:
>>> On 09/15/2015 09:33 AM, ankitprasad.r.sharma at intel.com wrote:
>>> In what circumstances can drm_mm_scan_remove_block fail?
>> It works some thing like this:
>> If there are 10 purgable nodes in the unwind list and 4 of them are
>> positioned in a way to reap enough contiguous space for the new object
>> (not necessarily purging all nodes will give us the amount of space we
>> need), then for the remaining 6 nodes drm_mm_scan_remove_block will
>> fail, while the rest will be removed to make space for the new object.
>
> Quoting the nice kerneldoc we have:
>
>   * Returns:
>   * True if this block should be evicted, false otherwise. Will always
>   * return false when no hole has been found.
>
> Was that not clear enough or did you simply not bother to read the docs?
> If it's not clear (together with the obligatory DOC: overview section) we
> need to improve them ...

Not sure to whom you are addressing this? I did not read the docs, after 
years of no docs the notion that there aren't any is kind of hard 
embedded in me at least. :)

Ankit's explanation is also more detailed, answers the interesting 
question, again for me at least.

Regards,

Tvrtko