[PATCH 1/3] drm/ttm: fix ttm_bo_bulk_move_helper
Huang Rui
ray.huang at amd.com
Sat Sep 8 07:54:14 UTC 2018
On Fri, Sep 07, 2018 at 07:18:59PM +0800, Christian K�nig wrote:
> Am 07.09.2018 um 13:02 schrieb Huang, Ray:
>
> Yes, that was one problem. Another was that the cutting code was buggy
> and determined one of the positions to cut at the wrong time.
>
> I went through again about the list cutting behavior and wrote down with the attached picture.
> After do the second list_cut_position, the list2 should be point the end of "before" list. And list2 won't be used anymore after list cutting. May I know am I something missed?
>
>
> Let's take a look at the original code:
>
> list1 = is_swap ? &pos->last->swap : &pos->last->lru;
> list2 = is_swap ? pos->first->swap.prev : pos->first->lru.prev;
>
> list_cut_position(&entries, lru, list1);
> list_cut_position(&before, &entries, list2);
>
>
> As far as I can see the problem is that the first list_cur_position could
> modify the value of pos->first->lru.prev and so make the second
> list_cut_position work on the wrong position.
>
I think I understood. In this case (the first element of LRU ==
pos->first->lru), please see the picture. If we store first->lru.prev as
list2(LRU head), after do list cutting, the first->lru.prev will overwrite
as new head (entries), however, the orignal list2 will still point previous
head (that is the wrong position now). We actually expected to use the
latest first->lru.prev as the second cutting position. So we should adjust
code sequence like below:
list1 = is_swap ? &pos->last->swap : &pos->last->lru;
list_cut_position(&entries, lru, list1);
list2 = is_swap ? pos->first->swap.prev : pos->first->lru.prev;
list_cut_position(&before, &entries, list2);
Am I understanding right?
Thanks,
Ray
> Regards,
> Christian.
>
>
>
> Thanks,
> Ray
>
> From: amd-gfx <amd-gfx-bounces at lists.freedesktop.org> on behalf of Christian König <christian.koenig at amd.com>
> Sent: Thursday, September 6, 2018 6:06 PM
> To: Huang, Ray
> Cc: Michel Dänzer; amd-gfx at lists.freedesktop.org
> Subject: Re: [PATCH 1/3] drm/ttm: fix ttm_bo_bulk_move_helper
>
>
> Am 06.09.2018 um 12:02 schrieb Huang Rui:
>
> On Fri, Aug 31, 2018 at 05:17:33PM +0200, Christian König wrote:
>
> Am 31.08.2018 um 17:15 schrieb Michel Dänzer:
>
> On 2018-08-31 3:10 p.m., Christian König wrote:
>
> Staring at the function for six hours, just to essentially move one line
> of code.
>
> That sucks, but the commit log should describe what the problem was and
> how this patch solves it.
>
>
>
> Signed-off-by: Christian König <christian.koenig at amd.com>
> ---
> drivers/gpu/drm/ttm/ttm_bo.c | 13 ++++++++-----
> 1 file changed, 8 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index 35d53d81f486..138c98902033 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -250,15 +250,18 @@ EXPORT_SYMBOL(ttm_bo_move_to_lru_tail);
> static void ttm_bo_bulk_move_helper(struct ttm_lru_bulk_move_pos *pos,
> struct list_head *lru, bool is_swap)
> {
> + struct list_head *list;
> LIST_HEAD(entries);
> LIST_HEAD(before);
> - struct list_head *list1, *list2;
> - list1 = is_swap ? &pos->last->swap : &pos->last->lru;
> - list2 = is_swap ? pos->first->swap.prev : pos->first->lru.prev;
> + reservation_object_assert_held(pos->last->resv);
> + list = is_swap ? &pos->last->swap : &pos->last->lru;
> + list_cut_position(&entries, lru, list);
> +
> + reservation_object_assert_held(pos->first->resv);
> + list = is_swap ? pos->first->swap.prev : pos->first->lru.prev;
> + list_cut_position(&before, &entries, list);
>
> So the problem was that the first list_cut_position call could result in
> list2 pointing to la-la-land? Good catch!
>
> Yes, exactly. Thought that would be obvious, but going to add that
> to the commit log.
>
> Can I get a tested-by? You where much better at reproducing that than I'm.
>
>
> Michel, Christian, thanks so much to take care of this when I was on
> vacation. And sorry to let you take a long time for finding the cause.
>
> Is that because I didn't hold the resveration before cut the list from
> position "first" and "last"?
>
> Yes, that was one problem. Another was that the cutting code was buggy
> and determined one of the positions to cut at the wrong time.
>
>
> May I know in which cases, we must hold the
> bo's reservation firstly?
>
> BOs are reserved to prevent moving them. E.g. when the BO isn't reserved
> it can move around and so the LRU where you want to add/remove it could
> change.
>
> Christian.
>
>
> Thanks,
> Ray
>
>
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
>
>
> amd-gfx Info Page - freedesktop.org
> lists.freedesktop.org
> To see the collection of prior postings to the list, visit the amd-gfx Archives.. Using amd-gfx: To post a message to all the list members, send email to amd-gfx at lists.freedesktop.org. You can subscribe to the list, or change your existing subscription, in the sections below.
>
>
>
>
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Wrong_case.jpg
Type: image/jpeg
Size: 70757 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20180908/f892790d/attachment-0001.jpg>
More information about the amd-gfx
mailing list