From: "Christian König" <ckoenig.leichtzumerken@gmail.com>
To: Chris Wilson <chris@chris-wilson.co.uk>,
daniel.vetter@ffwll.ch, dri-devel@lists.freedesktop.org,
linaro-mm-sig@lists.linaro.org, linux-media@vger.kernel.org,
sumit.semwal@linaro.org
Subject: Re: [PATCH 08/10] dma-buf/resv: replace shared fence with new fences container
Date: Thu, 22 Aug 2019 11:16:26 +0200 [thread overview]
Message-ID: <25f13de1-c5e2-1ec8-cb79-4e7ee28bb392@gmail.com> (raw)
In-Reply-To: <5a2aea22-6857-d061-4a80-64f2f57da007@gmail.com>
Am 22.08.19 um 10:37 schrieb Christian König:
> Am 21.08.19 um 19:35 schrieb Chris Wilson:
>> Quoting Chris Wilson (2019-08-21 16:24:22)
>>> Quoting Christian König (2019-08-21 13:31:45)
>>>> @@ -117,17 +120,10 @@ i915_gem_busy_ioctl(struct drm_device *dev,
>>>> void *data,
>>>> busy_check_writer(rcu_dereference(obj->base.resv->fence_excl));
>>>> /* Translate shared fences to READ set of engines */
>>>> - list = rcu_dereference(obj->base.resv->fence);
>>>> - if (list) {
>>>> - unsigned int shared_count = list->shared_count, i;
>>>> -
>>>> - for (i = 0; i < shared_count; ++i) {
>>>> - struct dma_fence *fence =
>>>> - rcu_dereference(list->shared[i]);
>>>> -
>>>> - args->busy |= busy_check_reader(fence);
>>>> - }
>>>> - }
>>>> + readers = dma_resv_fences_get_rcu(&obj->base.resv->readers);
>>>> + dma_fence_array_for_each(fence, cursor, readers)
>>>> + args->busy |= busy_check_reader(fence);
>>>> + dma_fence_put(readers);
>>> That's underwhelming, the full-mb shows up in scaling tests (I'll test
>>> the impact of this series later). Something like,
>> To put some numbers to it, adding the full-mb adds 5ns to a single
>> thread on Kabylake and 20ns under contention.
>
> The question is if that's the use case we want to optimize for.
>
> Querying a buffer for business is something we do absolutely rarely on
> amdgpu, e.g. IIRC we even grab the full reservation lock for this.
>
> But adding new fences comes with every command submission.
>
> What could maybe work is the "do { } while (fence has changed); loop
> you suggested earlier in this mail thread, but I need to double check
> if that would really work or clash with recycling dma_fence_arrays().
Ok thinking about it some more that won't work either.
See the dma_fence_array is only guaranteed to not change when you hold a
reference to it. Otherwise we don't guarantee anything.
Christian.
>
> Christian.
>
>> -Chris
>
next prev parent reply other threads:[~2019-08-22 9:16 UTC|newest]
Thread overview: 33+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-08-21 12:31 [RFC] replacing dma_resv API Christian König
2019-08-21 12:31 ` [PATCH 01/10] dma-buf: make to_dma_fence_array NULL safe Christian König
2019-08-21 12:31 ` [PATCH 02/10] dma-buf: add dma_fence_array_alloc/free Christian König
2019-08-21 12:31 ` [PATCH 03/10] dma-buf: add dma_fence_array_recycle Christian König
2019-08-21 16:24 ` Chris Wilson
2019-08-22 8:38 ` Christian König
2019-08-21 12:31 ` [PATCH 04/10] dma-buf: add dma_fence_array_for_each Christian König
2019-08-21 12:31 ` [PATCH 05/10] dma-buf/resv: add dma_resv_prune_fences Christian König
2019-08-21 14:55 ` Chris Wilson
2019-08-21 14:56 ` Chris Wilson
2019-08-21 12:31 ` [PATCH 06/10] dma-buf/resv: stop pruning shared fences when exclusive is added Christian König
2019-08-21 12:31 ` [PATCH 07/10] dma-buf/resv: add new fences container implementation Christian König
2019-08-21 16:04 ` Daniel Vetter
2019-08-22 8:23 ` Christian König
2019-08-22 13:02 ` Daniel Vetter
2019-08-22 13:53 ` Koenig, Christian
2019-08-21 12:31 ` [PATCH 08/10] dma-buf/resv: replace shared fence with new fences container Christian König
2019-08-21 15:24 ` Chris Wilson
2019-08-21 17:35 ` Chris Wilson
2019-08-22 8:37 ` Christian König
2019-08-22 9:16 ` Christian König [this message]
2019-08-21 16:21 ` Chris Wilson
2019-08-24 13:22 ` Chris Wilson
2019-08-21 12:31 ` [PATCH 09/10] dma-buf/resv: replace exclusive " Christian König
2019-08-21 12:31 ` [PATCH 10/10] dma-buf/resv: add other operations Christian König
2019-08-22 12:28 ` Ville Syrjälä
2019-08-21 16:13 ` [RFC] replacing dma_resv API Daniel Vetter
2019-08-21 20:05 ` Daniel Vetter
2019-08-22 9:27 ` Christian König
2019-08-21 20:11 ` Chris Wilson
2019-08-21 20:22 ` Daniel Vetter
2019-08-22 9:14 ` Christian König
2019-08-22 10:00 ` Daniel Vetter
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=25f13de1-c5e2-1ec8-cb79-4e7ee28bb392@gmail.com \
--to=ckoenig.leichtzumerken@gmail.com \
--cc=chris@chris-wilson.co.uk \
--cc=christian.koenig@amd.com \
--cc=daniel.vetter@ffwll.ch \
--cc=dri-devel@lists.freedesktop.org \
--cc=linaro-mm-sig@lists.linaro.org \
--cc=linux-media@vger.kernel.org \
--cc=sumit.semwal@linaro.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).