mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Ivan Lipski <ivan.lipski@amd.com>
To: Hugo Sales <hugo@hsal.es>,
	harry.wentland@amd.com, sunpeng.li@amd.com, siqueira@igalia.com,
	alexander.deucher@amd.com, amd-gfx@lists.freedesktop.org
Cc: alex.hung@amd.com, mario.kleiner.de@gmail.com,
	nicholas.kazlauskas@amd.com, ray.wu@amd.com,
	bhawanpreet.lakha@amd.com, chiahsuan.chung@amd.com,
	jerry.zuo@amd.com, Aric.Cyr@amd.com, christian.koenig@amd.com,
	airlied@gmail.com, simona@ffwll.ch,
	dri-devel@lists.freedesktop.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] drm/amd/display: Hold the VRR vblank reference only on active CRTCs
Date: Fri, 13 Nov 2026 17:27:36 -0500	[thread overview]
Message-ID: <65222154-22bf-4f31-a050-b7c0f0c99284@amd.com> (raw)
In-Reply-To: <20261031143351.3222033-1-hugo@hsal.es>



On 10/31/26 10:33, Hugo Sales wrote:
> amdgpu_dm_handle_vrr_transition() takes a vblank reference when a CRTC
> state turns VRR on and drops it when a later state turns it off. It only
> looks at freesync_config.state, and disabling a CRTC keeps that state,
> so a suspend or DPMS off with VRR active keeps the reference
> 
> On resume, drm_atomic_helper_resume() first calls drm_mode_config_reset(),
> which swaps the CRTC state for a zeroed one. The restored state then goes
> from VRR_STATE_UNSUPPORTED to active and takes a second reference, and
> the first is never dropped. The vblank irq stays on for good, so the CRTC
> never idles and PSR is never entered
> 
> A VRR change on a disabled CRTC can't take the reference at all, since
> vblank is off there and drm_crtc_vblank_get() fails with a WARN
> 
> On a Cezanne laptop with VRR always on, each suspend added one to the
> eDP CRTC's vblank refcount, and it stayed at 1 rather than 0 after VRR
> was turned off again
> 
> Treat VRR as active for the reference only while the CRTC is active too.
> Disabling the CRTC then drops it and enabling it takes it again, which
> balances suspend, resume and DPMS, and the get only happens once
> manage_dm_interrupts() has turned vblank on
> 
> Fixes: 66b0c973d7f7 ("drm/amd/display: Prevent vblank irq disable while VRR is active. (v3)")
> Signed-off-by: Hugo Sales <hugo@hsal.es>

Reviewed-by: Ivan Lipski <ivan.lipski@amd.com>

> ---
> 
> Tested on an ASUS ROG Flow X13 (GV301QH, Cezanne, DCN 2.1) with VRR
> set to always on in Hyprland. bpftrace on drm_handle_vblank showed the eDP
> CRTC's vblank refcount go up by one on every suspend before this patch.
> With it, the refcount reads the same before and after a systemctl
> suspend, and drops to 0 once VRR is turned off
> 
> dm_test_handle_vrr_transition_round_trip now sets base.active on both
> states, since an inactive CRTC no longer takes the reference. The
> three new cases cover the CRTC turning off and on with VRR left on
> 
> amdgpu_dm KUnit on amd-staging-drm-next (ba496bb83788) passes apart
> from dm_mst_test_dsc_link_bpp_between_limits and the two
> dm_test_crtc_enable_vblank_ips_restore cases, which fail or panic in
> drm_vblank_restore() without this patch as well
> 
>   .../display/amdgpu_dm/amdgpu_dm_freesync.c    | 12 ++-
>   .../amdgpu_dm/tests/amdgpu_dm_freesync_test.c | 98 +++++++++++++++++++
>   2 files changed, 108 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_freesync.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_freesync.c
> index 386ba88c57c6a..ee8a11f62a0d3 100644
> --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_freesync.c
> +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_freesync.c
> @@ -414,8 +414,16 @@ void amdgpu_dm_handle_vrr_transition(struct amdgpu_display_manager *dm,
>   				     struct dm_crtc_state *new_state)
>   {
>   	struct amdgpu_device *adev = dm->adev;
> -	bool old_vrr_active = amdgpu_dm_crtc_vrr_active(old_state);
> -	bool new_vrr_active = amdgpu_dm_crtc_vrr_active(new_state);
> +	/*
> +	 * A disabled CRTC keeps its freesync state, but drm_crtc_vblank_get()
> +	 * fails there because vblank is off. Hold the reference only while
> +	 * the CRTC is active as well, which also covers resume, where the
> +	 * reset state lost track of the reference taken before suspend
> +	 */
> +	bool old_vrr_active = old_state->base.active &&
> +			      amdgpu_dm_crtc_vrr_active(old_state);
> +	bool new_vrr_active = new_state->base.active &&
> +			      amdgpu_dm_crtc_vrr_active(new_state);
>   
>   	/* Only DCE gates vupdate on VRR, keep it enabled for DCN */
>   	bool vrr_gates_vupdate = amdgpu_ip_version(adev, DCE_HWIP, 0) == 0;
> diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/tests/amdgpu_dm_freesync_test.c b/drivers/gpu/drm/amd/display/amdgpu_dm/tests/amdgpu_dm_freesync_test.c
> index 710cdc7f40bf8..3268bdb1043ba 100644
> --- a/drivers/gpu/drm/amd/display/amdgpu_dm/tests/amdgpu_dm_freesync_test.c
> +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/tests/amdgpu_dm_freesync_test.c
> @@ -855,8 +855,10 @@ static void dm_test_handle_vrr_transition_round_trip(struct kunit *test)
>   	mutex_init(&adev->dm.dc_lock);
>   
>   	old_state.base.crtc = &acrtc->base;
> +	old_state.base.active = true;
>   	old_state.stream = stream;
>   	new_state.base.crtc = &acrtc->base;
> +	new_state.base.active = true;
>   	new_state.stream = stream;
>   	new_state.freesync_config.state = VRR_STATE_ACTIVE_VARIABLE;
>   
> @@ -867,6 +869,99 @@ static void dm_test_handle_vrr_transition_round_trip(struct kunit *test)
>   	KUNIT_EXPECT_EQ(test, atomic_read(&vblank->refcount), 0);
>   }
>   
> +/*
> + * Runs a transition on a CRTC whose vblank is on with two references held,
> + * so a get or put never has to enable or disable it, and returns the
> + * refcount afterwards
> + */
> +static int dm_test_vrr_transition_refcount(struct kunit *test,
> +					   bool old_active, enum mod_vrr_state old_vrr,
> +					   bool new_active, enum mod_vrr_state new_vrr)
> +{
> +	struct amdgpu_device *adev = dm_kunit_alloc_adev(test);
> +	struct amdgpu_display_manager *dm = dm_kunit_alloc_dm(test);
> +	struct dm_crtc_state *old_state, *new_state;
> +	struct drm_crtc *crtc;
> +	int refcount;
> +
> +	KUNIT_ASSERT_EQ(test, drm_vblank_init(&adev->ddev, 1), 0);
> +	adev->ip_versions[DCE_HWIP][0] = IP_VERSION(2, 1, 0);
> +	dm->adev = adev;
> +	mutex_init(&dm->dc_lock);
> +
> +	crtc = kunit_kzalloc(test, sizeof(*crtc), GFP_KERNEL);
> +	KUNIT_ASSERT_NOT_NULL(test, crtc);
> +	crtc->dev = &adev->ddev;
> +
> +	old_state = kunit_kzalloc(test, sizeof(*old_state), GFP_KERNEL);
> +	new_state = kunit_kzalloc(test, sizeof(*new_state), GFP_KERNEL);
> +	KUNIT_ASSERT_NOT_NULL(test, old_state);
> +	KUNIT_ASSERT_NOT_NULL(test, new_state);
> +	old_state->base.crtc = crtc;
> +	old_state->base.active = old_active;
> +	old_state->freesync_config.state = old_vrr;
> +	new_state->base.crtc = crtc;
> +	new_state->base.active = new_active;
> +	new_state->freesync_config.state = new_vrr;
> +
> +	adev->ddev.vblank[0].enabled = true;
> +	atomic_set(&adev->ddev.vblank[0].refcount, 2);
> +
> +	amdgpu_dm_handle_vrr_transition(dm, old_state, new_state);
> +	refcount = atomic_read(&adev->ddev.vblank[0].refcount);
> +
> +	/* drm_vblank_init_release() warns if vblank is still enabled */
> +	adev->ddev.vblank[0].enabled = false;
> +	atomic_set(&adev->ddev.vblank[0].refcount, 0);
> +
> +	return refcount;
> +}
> +
> +/**
> + * dm_test_vrr_transition_crtc_off_drops_ref - Test disabling the CRTC drops the ref
> + * @test: The KUnit test context
> + *
> + * Disabling the CRTC keeps the freesync state, as on suspend or DPMS off.
> + * The reference must still be dropped, since resume restores the state from
> + * a reset one and takes a new reference
> + */
> +static void dm_test_vrr_transition_crtc_off_drops_ref(struct kunit *test)
> +{
> +	KUNIT_EXPECT_EQ(test,
> +			dm_test_vrr_transition_refcount(test, true, VRR_STATE_ACTIVE_VARIABLE,
> +							false, VRR_STATE_ACTIVE_VARIABLE),
> +			1);
> +}
> +
> +/**
> + * dm_test_vrr_transition_crtc_on_takes_ref - Test enabling the CRTC takes the ref
> + * @test: The KUnit test context
> + *
> + * Covers resume, where the old state is a reset one, and DPMS on
> + */
> +static void dm_test_vrr_transition_crtc_on_takes_ref(struct kunit *test)
> +{
> +	KUNIT_EXPECT_EQ(test,
> +			dm_test_vrr_transition_refcount(test, false, VRR_STATE_UNSUPPORTED,
> +							true, VRR_STATE_ACTIVE_VARIABLE),
> +			3);
> +}
> +
> +/**
> + * dm_test_vrr_transition_crtc_off_keeps_ref - Test no ref change while off
> + * @test: The KUnit test context
> + *
> + * A disabled CRTC has vblank off, so a VRR change there must neither take
> + * nor drop a reference
> + */
> +static void dm_test_vrr_transition_crtc_off_keeps_ref(struct kunit *test)
> +{
> +	KUNIT_EXPECT_EQ(test,
> +			dm_test_vrr_transition_refcount(test, false, VRR_STATE_UNSUPPORTED,
> +							false, VRR_STATE_ACTIVE_VARIABLE),
> +			2);
> +}
> +
>   static struct kunit_case amdgpu_dm_freesync_tests[] = {
>   	/* amdgpu_dm_is_timing_unchanged_for_freesync */
>   	KUNIT_CASE(dm_test_timing_unchanged_null_args),
> @@ -910,6 +1005,9 @@ static struct kunit_case amdgpu_dm_freesync_tests[] = {
>   	/* amdgpu_dm_handle_vrr_transition */
>   	KUNIT_CASE(dm_test_handle_vrr_transition_inactive),
>   	KUNIT_CASE(dm_test_handle_vrr_transition_round_trip),
> +	KUNIT_CASE(dm_test_vrr_transition_crtc_off_drops_ref),
> +	KUNIT_CASE(dm_test_vrr_transition_crtc_on_takes_ref),
> +	KUNIT_CASE(dm_test_vrr_transition_crtc_off_keeps_ref),
>   	{}
>   };
>   

-- 
Thanks,
Ivan Lipski


      reply	other threads:[~2026-11-13 22:27 UTC|newest]

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-10-31 14:33 Hugo Sales
2026-11-13 22:27 ` Ivan Lipski [this message]

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=65222154-22bf-4f31-a050-b7c0f0c99284@amd.com \
    --to=ivan.lipski@amd.com \
    --cc=Aric.Cyr@amd.com \
    --cc=airlied@gmail.com \
    --cc=alex.hung@amd.com \
    --cc=alexander.deucher@amd.com \
    --cc=amd-gfx@lists.freedesktop.org \
    --cc=bhawanpreet.lakha@amd.com \
    --cc=chiahsuan.chung@amd.com \
    --cc=christian.koenig@amd.com \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=harry.wentland@amd.com \
    --cc=hugo@hsal.es \
    --cc=jerry.zuo@amd.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mario.kleiner.de@gmail.com \
    --cc=nicholas.kazlauskas@amd.com \
    --cc=ray.wu@amd.com \
    --cc=simona@ffwll.ch \
    --cc=siqueira@igalia.com \
    --cc=sunpeng.li@amd.com \
    /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

all inboxes | Powered by JetHome®