dri-devel Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: "Christian König" <ckoenig.leichtzumerken-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
To: "Daniel Vetter" <daniel-/w4YWyX8dFk@public.gmane.org>,
	"Christian König" <christian.koenig-5C7GfCeVMHo@public.gmane.org>
Cc: Alex Deucher <alexander.deucher-5C7GfCeVMHo@public.gmane.org>,
	Daniel Vetter <daniel.vetter-/w4YWyX8dFk@public.gmane.org>,
	Intel Graphics Development
	<intel-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org>,
	DRI Development
	<dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org>,
	amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
Subject: Re: [PATCH 13/17] drm/radeon: Remove custom dma_fence_ops->wait implementation
Date: Mon, 30 Apr 2018 20:26:14 +0200	[thread overview]
Message-ID: <ba53bdbd-3311-5836-4529-5374ccf6584a@gmail.com> (raw)
In-Reply-To: <20180430153854.GR12521-dv86pmgwkMBes7Z6vYuT8azUEOm+Xw19@public.gmane.org>

Am 30.04.2018 um 17:38 schrieb Daniel Vetter:
> On Sun, Apr 29, 2018 at 09:08:31AM +0200, Christian König wrote:
>> NAK, there is a subtitle but major difference:
>>
>>> -		if (rdev->needs_reset) {
>>> -			t = -EDEADLK;
>>> -			break;
>>> -		}
>> Without that the whole radeon GPU reset code breaks.
> Oops I've missed that. How does this work when you register a callback
> using ->enable_signaling and then block on it? Everything just dies?

The short answer is we simply avoid using enable_signaling() from inside 
driver IOCTLs.

> We have lots of users of that for buffer/fence sharing. A really ugly, but
> probably working fix for this would be a kthread worker that just looks
> for ->needs_reset and force-completes all fences with
> dma_fence_set_error(-EIO), which is kinda what's supposed to happen here
> anyway.

That actually won't help. Radeon does this dance to return an error from 
dma_fence_wait() when the GPU needs a reset.

This way all IOCTLs should return to userspace with -EAGAIN and when 
they are restarted we block for the running GPU reset to finish.

I was against this approach, but it works as long as radeon only has to 
deal with it's own fences.

Christian.

> -Daniel
>
>> Regards,
>> Christian.
>>
>>
>> Am 27.04.2018 um 08:17 schrieb Daniel Vetter:
>>> It's a copy of dma_fence_default_wait, written slightly differently.
>>>
>>> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>>> Cc: Alex Deucher <alexander.deucher@amd.com>
>>> Cc: "Christian König" <christian.koenig@amd.com>
>>> Cc: "David (ChunMing) Zhou" <David1.Zhou@amd.com>
>>> Cc: amd-gfx@lists.freedesktop.org
>>> ---
>>>    drivers/gpu/drm/radeon/radeon_fence.c | 63 ---------------------------
>>>    1 file changed, 63 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/radeon/radeon_fence.c b/drivers/gpu/drm/radeon/radeon_fence.c
>>> index e86f2bd38410..32690a525bfc 100644
>>> --- a/drivers/gpu/drm/radeon/radeon_fence.c
>>> +++ b/drivers/gpu/drm/radeon/radeon_fence.c
>>> @@ -1051,72 +1051,9 @@ static const char *radeon_fence_get_timeline_name(struct dma_fence *f)
>>>    	}
>>>    }
>>> -static inline bool radeon_test_signaled(struct radeon_fence *fence)
>>> -{
>>> -	return test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->base.flags);
>>> -}
>>> -
>>> -struct radeon_wait_cb {
>>> -	struct dma_fence_cb base;
>>> -	struct task_struct *task;
>>> -};
>>> -
>>> -static void
>>> -radeon_fence_wait_cb(struct dma_fence *fence, struct dma_fence_cb *cb)
>>> -{
>>> -	struct radeon_wait_cb *wait =
>>> -		container_of(cb, struct radeon_wait_cb, base);
>>> -
>>> -	wake_up_process(wait->task);
>>> -}
>>> -
>>> -static signed long radeon_fence_default_wait(struct dma_fence *f, bool intr,
>>> -					     signed long t)
>>> -{
>>> -	struct radeon_fence *fence = to_radeon_fence(f);
>>> -	struct radeon_device *rdev = fence->rdev;
>>> -	struct radeon_wait_cb cb;
>>> -
>>> -	cb.task = current;
>>> -
>>> -	if (dma_fence_add_callback(f, &cb.base, radeon_fence_wait_cb))
>>> -		return t;
>>> -
>>> -	while (t > 0) {
>>> -		if (intr)
>>> -			set_current_state(TASK_INTERRUPTIBLE);
>>> -		else
>>> -			set_current_state(TASK_UNINTERRUPTIBLE);
>>> -
>>> -		/*
>>> -		 * radeon_test_signaled must be called after
>>> -		 * set_current_state to prevent a race with wake_up_process
>>> -		 */
>>> -		if (radeon_test_signaled(fence))
>>> -			break;
>>> -
>>> -		if (rdev->needs_reset) {
>>> -			t = -EDEADLK;
>>> -			break;
>>> -		}
>>> -
>>> -		t = schedule_timeout(t);
>>> -
>>> -		if (t > 0 && intr && signal_pending(current))
>>> -			t = -ERESTARTSYS;
>>> -	}
>>> -
>>> -	__set_current_state(TASK_RUNNING);
>>> -	dma_fence_remove_callback(f, &cb.base);
>>> -
>>> -	return t;
>>> -}
>>> -
>>>    const struct dma_fence_ops radeon_fence_ops = {
>>>    	.get_driver_name = radeon_fence_get_driver_name,
>>>    	.get_timeline_name = radeon_fence_get_timeline_name,
>>>    	.enable_signaling = radeon_fence_enable_signaling,
>>>    	.signaled = radeon_fence_is_signaled,
>>> -	.wait = radeon_fence_default_wait,
>>> -	.release = NULL,
>>>    };

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

  parent reply	other threads:[~2018-04-30 18:26 UTC|newest]

Thread overview: 37+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-04-27  6:17 [PATCH 00/17] dma-fence doc polish and small cleanup Daniel Vetter
2018-04-27  6:17 ` [PATCH 01/17] dma-fence: Some kerneldoc polish for dma-fence.h Daniel Vetter
2018-04-30 17:49   ` Eric Anholt
2018-05-02  7:38     ` [Intel-gfx] " Daniel Vetter
2018-04-27  6:17 ` [PATCH 02/17] dma-fence: remove fill_driver_data callback Daniel Vetter
2018-04-30 17:49   ` Eric Anholt
2018-05-02  8:23   ` [PATCH] " Daniel Vetter
2018-05-02 16:57     ` Eric Anholt
2018-04-27  6:17 ` [PATCH 03/17] dma-fence: Make ->enable_signaling optional Daniel Vetter
2018-04-27  6:17 ` [PATCH 04/17] dma-fence: Allow wait_any_timeout for all fences Daniel Vetter
2018-04-29  7:11   ` Christian König
2018-04-30 15:35     ` Daniel Vetter
2018-04-27  6:17 ` [PATCH 05/17] dma-fence: Make ->wait callback optional Daniel Vetter
2018-04-27  6:17 ` [PATCH 06/17] drm/amdgpu: Remove unecessary dma_fence_ops Daniel Vetter
2018-04-29  7:12   ` Christian König
2018-04-27  6:17 ` [PATCH 07/17] drm: " Daniel Vetter
2018-04-30 17:51   ` Eric Anholt
2018-04-27  6:17 ` [PATCH 08/17] drm/etnaviv: " Daniel Vetter
2018-05-03 13:43   ` Lucas Stach
2018-04-27  6:17 ` [PATCH 09/17] drm/i915: " Daniel Vetter
     [not found] ` <20180427061724.28497-1-daniel.vetter-/w4YWyX8dFk@public.gmane.org>
2018-04-27  6:17   ` [PATCH 10/17] drm/msm: " Daniel Vetter
2018-04-27  6:17   ` [PATCH 11/17] drm/nouveau: " Daniel Vetter
2018-04-27  6:17 ` [PATCH 12/17] drm/qxl: " Daniel Vetter
2018-04-30 17:53   ` Eric Anholt
2018-05-02  8:23   ` [PATCH] " Daniel Vetter
2018-04-27  6:17 ` [PATCH 13/17] drm/radeon: Remove custom dma_fence_ops->wait implementation Daniel Vetter
2018-04-29  7:08   ` Christian König
2018-04-30 15:38     ` Daniel Vetter
     [not found]       ` <20180430153854.GR12521-dv86pmgwkMBes7Z6vYuT8azUEOm+Xw19@public.gmane.org>
2018-04-30 18:26         ` Christian König [this message]
     [not found]           ` <ba53bdbd-3311-5836-4529-5374ccf6584a-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2018-04-30 19:35             ` Daniel Vetter
2018-04-27  6:17 ` [PATCH 14/17] drm/vc4: Remove unecessary dma_fence_ops Daniel Vetter
2018-04-30 17:50   ` Eric Anholt
2018-04-27  6:17 ` [PATCH 15/17] drm/vgem: " Daniel Vetter
2018-04-27  6:17 ` [PATCH 16/17] drm/virtio: " Daniel Vetter
2018-04-30 17:54   ` Eric Anholt
2018-04-27  6:17 ` [PATCH 17/17] dma-fence: Polish kernel-doc for dma-fence.c Daniel Vetter
2018-04-29  7:15 ` [PATCH 00/17] dma-fence doc polish and small cleanup Christian König

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=ba53bdbd-3311-5836-4529-5374ccf6584a@gmail.com \
    --to=ckoenig.leichtzumerken-re5jqeeqqe8avxtiumwx3w@public.gmane.org \
    --cc=alexander.deucher-5C7GfCeVMHo@public.gmane.org \
    --cc=amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org \
    --cc=christian.koenig-5C7GfCeVMHo@public.gmane.org \
    --cc=daniel-/w4YWyX8dFk@public.gmane.org \
    --cc=daniel.vetter-/w4YWyX8dFk@public.gmane.org \
    --cc=dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org \
    --cc=intel-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.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