Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Tue, 29 Jan 2019 21:30:45 +0100
From:   Daniel Vetter <daniel@ffwll.ch>
To:     Lyude Paul <lyude@redhat.com>
Cc:     intel-gfx@lists.freedesktop.org, dri-devel@lists.freedesktop.org,
        Daniel Vetter <daniel@ffwll.ch>,
        Maarten Lankhorst <maarten.lankhorst@linux.intel.com>,
        Maxime Ripard <maxime.ripard@bootlin.com>,
        Sean Paul <sean@poorly.run>, David Airlie <airlied@linux.ie>,
        Jani Nikula <jani.nikula@linux.intel.com>,
        Joonas Lahtinen <joonas.lahtinen@linux.intel.com>,
        Rodrigo Vivi <rodrigo.vivi@intel.com>,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH 2/3] drm/atomic: Add drm_atomic_state->suspend_or_resume
Message-ID: <20190129203045.GZ3271@phenom.ffwll.local>
Mail-Followup-To: Lyude Paul <lyude@redhat.com>,
        intel-gfx@lists.freedesktop.org, dri-devel@lists.freedesktop.org,
        Maarten Lankhorst <maarten.lankhorst@linux.intel.com>,
        Maxime Ripard <maxime.ripard@bootlin.com>,
        Sean Paul <sean@poorly.run>, David Airlie <airlied@linux.ie>,
        Jani Nikula <jani.nikula@linux.intel.com>,
        Joonas Lahtinen <joonas.lahtinen@linux.intel.com>,
        Rodrigo Vivi <rodrigo.vivi@intel.com>, linux-kernel@vger.kernel.org
References: <20190129183928.26779-1-lyude@redhat.com>
 <20190129183928.26779-3-lyude@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190129183928.26779-3-lyude@redhat.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Jan 29, 2019 at 01:39:26PM -0500, Lyude Paul wrote:
> Since
> 
> commit 39b50c603878 ("drm/atomic_helper: Stop modesets on unregistered
> connectors harder")
> 
> We've been failing atomic checks if they try to enable new displays on
> unregistered connectors. This is fine except for the one situation that
> breaks atomic assumptions: suspend/resume. If a connector is
> unregistered before we attempt to restore the atomic state, something we
> end up failing the atomic check that happens when trying to restore the
> state during resume.
> 
> Normally this would be OK: we try our best to make sure that the atomic
> state pre-suspend can be restored post-suspend, but failures at that
> point usually don't cause problems. That is of course, until we
> introduced the new atomic MST VCPI helpers:
> 
> [drm:drm_atomic_helper_check_modeset [drm_kms_helper]] [CRTC:65:pipe B] active changed
> [drm:drm_atomic_helper_check_modeset [drm_kms_helper]] Updating routing for [CONNECTOR:123:DP-5]
> [drm:drm_atomic_helper_check_modeset [drm_kms_helper]] Disabling [CONNECTOR:123:DP-5]
> [drm:drm_atomic_get_private_obj_state [drm]] Added new private object 0000000025844636 state 000000009fd2899a to 000000003a13d7b8
> WARNING: CPU: 6 PID: 1070 at drivers/gpu/drm/drm_dp_mst_topology.c:3153 drm_dp_atomic_release_vcpi_slots+0xb9/0x200 [drm_kms_helper]
> Modules linked in: fuse vfat fat snd_hda_codec_hdmi snd_hda_codec_realtek snd_hda_codec_generic joydev iTCO_wdt i915(O) wmi_bmof intel_rapl btusb btrtl x86_pkg_temp_thermal btbcm btintel coretemp i2c_algo_bit drm_kms_helper(O) crc32_pclmul snd_hda_intel syscopyarea sysfillrect snd_hda_codec sysimgblt snd_hda_core bluetooth fb_sys_fops snd_pcm pcspkr drm(O) psmouse snd_timer mei_me ecdh_generic i2c_i801 mei i2c_core ucsi_acpi typec_ucsi typec wmi thinkpad_acpi ledtrig_audio snd soundcore tpm_tis rfkill tpm_tis_core video tpm acpi_pad pcc_cpufreq uas usb_storage crc32c_intel nvme serio_raw xhci_pci nvme_core xhci_hcd
> CPU: 6 PID: 1070 Comm: gnome-shell Tainted: G        W  O      5.0.0-rc2Lyude-Test+ #1
> Hardware name: LENOVO 20L8S2N800/20L8S2N800, BIOS N22ET35W (1.12 ) 04/09/2018
> RIP: 0010:drm_dp_atomic_release_vcpi_slots+0xb9/0x200 [drm_kms_helper]
> Code: 00 4c 39 6d f0 74 49 48 8d 7b 10 48 89 f9 48 c1 e9 03 42 80 3c 21 00 0f 85 d2 00 00 00 48 8b 6b 10 48 8d 5d f0 49 39 ee 75 c5 <0f> 0b 48 c7 c7 c0 78 b3 a0 48 89 c2 4c 89 ee e8 03 6c aa ff b8 ea
> RSP: 0018:ffff88841235f268 EFLAGS: 00010246
> RAX: ffff88841bf12ab0 RBX: ffff88841bf12aa8 RCX: 1ffff110837e2557
> RDX: dffffc0000000000 RSI: 0000000000000000 RDI: ffffed108246bde0
> RBP: ffff88841bf12ab8 R08: ffffed1083db3c93 R09: ffffed1083db3c92
> R10: ffffed1083db3c92 R11: ffff88841ed9e497 R12: ffff888419555d80
> R13: ffff8883bc499100 R14: ffff88841bf12ab8 R15: 0000000000000000
> FS:  00007f16fbd4cd00(0000) GS:ffff88841ed80000(0000) knlGS:0000000000000000
> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 00007f1687c9f000 CR3: 00000003ba3cc003 CR4: 00000000003606e0
> DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> Call Trace:
>  drm_atomic_helper_check_modeset+0xf21/0x2f50 [drm_kms_helper]
>  ? drm_atomic_helper_commit_modeset_enables+0xa90/0xa90 [drm_kms_helper]
>  ? __printk_safe_exit+0x10/0x10
>  ? save_stack+0x8c/0xb0
>  ? vprintk_func+0x96/0x1bf
>  ? __printk_safe_exit+0x10/0x10
>  intel_atomic_check+0x234/0x4750 [i915]
>  ? printk+0x9f/0xc5
>  ? kmsg_dump_rewind_nolock+0xd9/0xd9
>  ? _raw_spin_lock_irqsave+0xa4/0x140
>  ? drm_atomic_check_only+0xb1/0x28b0 [drm]
>  ? drm_dbg+0x186/0x1b0 [drm]
>  ? drm_dev_dbg+0x200/0x200 [drm]
>  ? intel_link_compute_m_n+0xb0/0xb0 [i915]
>  ? drm_mode_put_tile_group+0x20/0x20 [drm]
>  ? skl_plane_format_mod_supported+0x17f/0x1b0 [i915]
>  ? drm_plane_check_pixel_format+0x14a/0x310 [drm]
>  drm_atomic_check_only+0x13c4/0x28b0 [drm]
>  ? drm_state_info+0x220/0x220 [drm]
>  ? drm_atomic_helper_disable_plane+0x1d0/0x1d0 [drm_kms_helper]
>  ? pick_single_encoder_for_connector+0xe0/0xe0 [drm_kms_helper]
>  ? kasan_unpoison_shadow+0x35/0x40
>  drm_atomic_commit+0x3b/0x100 [drm]
>  drm_atomic_helper_set_config+0xd5/0x100 [drm_kms_helper]
>  drm_mode_setcrtc+0x636/0x1660 [drm]
>  ? vprintk_func+0x96/0x1bf
>  ? drm_dev_dbg+0x200/0x200 [drm]
>  ? drm_mode_getcrtc+0x790/0x790 [drm]
>  ? printk+0x9f/0xc5
>  ? mutex_unlock+0x1d/0x40
>  ? drm_mode_addfb2+0x2e9/0x3a0 [drm]
>  ? rcu_sync_dtor+0x2e0/0x2e0
>  ? drm_dbg+0x186/0x1b0 [drm]
>  ? set_page_dirty+0x271/0x4d0
>  drm_ioctl_kernel+0x203/0x290 [drm]
>  ? drm_mode_getcrtc+0x790/0x790 [drm]
>  ? drm_setversion+0x7f0/0x7f0 [drm]
>  ? __switch_to_asm+0x34/0x70
>  ? __switch_to_asm+0x34/0x70
>  drm_ioctl+0x445/0x950 [drm]
>  ? drm_mode_getcrtc+0x790/0x790 [drm]
>  ? drm_getunique+0x220/0x220 [drm]
>  ? expand_files.part.10+0x920/0x920
>  do_vfs_ioctl+0x1a1/0x13d0
>  ? ioctl_preallocate+0x2b0/0x2b0
>  ? __fget_light+0x2d6/0x390
>  ? schedule+0xd7/0x2e0
>  ? fget_raw+0x10/0x10
>  ? apic_timer_interrupt+0xa/0x20
>  ? apic_timer_interrupt+0xa/0x20
>  ? rcu_cleanup_dead_rnp+0x2c0/0x2c0
>  ksys_ioctl+0x60/0x90
>  __x64_sys_ioctl+0x6f/0xb0
>  do_syscall_64+0x136/0x440
>  ? syscall_return_slowpath+0x2d0/0x2d0
>  ? do_page_fault+0x89/0x330
>  ? __do_page_fault+0x9c0/0x9c0
>  ? prepare_exit_to_usermode+0x188/0x200
>  ? perf_trace_sys_enter+0x1090/0x1090
>  ? __x64_sys_sigaltstack+0x280/0x280
>  ? __put_user_4+0x1c/0x30
>  entry_SYSCALL_64_after_hwframe+0x44/0xa9
> RIP: 0033:0x7f16ff89a09b
> Code: 0f 1e fa 48 8b 05 ed bd 0c 00 64 c7 00 26 00 00 00 48 c7 c0 ff ff ff ff c3 66 0f 1f 44 00 00 f3 0f 1e fa b8 10 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d bd bd 0c 00 f7 d8 64 89 01 48
> RSP: 002b:00007fff001232b8 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
> RAX: ffffffffffffffda RBX: 00007fff001232f0 RCX: 00007f16ff89a09b
> RDX: 00007fff001232f0 RSI: 00000000c06864a2 RDI: 000000000000000b
> RBP: 00007fff001232f0 R08: 0000000000000000 R09: 000055a79d484460
> R10: 000055a79d44e770 R11: 0000000000000246 R12: 00000000c06864a2
> R13: 000000000000000b R14: 0000000000000000 R15: 000055a79d44e770
> WARNING: CPU: 6 PID: 1070 at drivers/gpu/drm/drm_dp_mst_topology.c:3153 drm_dp_atomic_release_vcpi_slots+0xb9/0x200 [drm_kms_helper]
> ---[ end trace d536c05c13c83be2 ]---
> [drm:drm_dp_atomic_release_vcpi_slots [drm_kms_helper]] *ERROR* no VCPI for [MST PORT:00000000f9e2b143] found in mst state 000000009fd2899a
> 
> This appears to be happening because we destroy the VCPI allocations
> when disabling all connected displays while suspending, and those VCPI
> allocations don't get restored on resume due to failing to restore the
> atomic state.
> 
> So, fix this by introducing the suspending option to
> drm_atomic_helper_duplicate_state() and use that to indicate in the
> atomic state that it's being used for suspending or resuming the system,
> and thus needs to be fixed up by the driver. We can then use the new
> state->suspend_or_resume hook to fixup the atomic VCPI helpers so they
> merely save and restore the VCPI allocations for such states instead of
> performing error checking on them. This fixes the warnings during
> suspend/resume when a topology is removed from the system, and allows
> us to once again restore the atomic state on resume without any issues.
> 
> Signed-off-by: Lyude Paul <lyude@redhat.com>
> Fixes: eceae1472467 ("drm/dp_mst: Start tracking per-port VCPI allocations")
> Cc: Daniel Vetter <daniel@ffwll.ch>
> ---
>  drivers/gpu/drm/drm_atomic_helper.c   | 16 +++++++++---
>  drivers/gpu/drm/drm_dp_mst_topology.c | 36 ++++++++++++++++++++++++---
>  drivers/gpu/drm/i915/intel_display.c  |  4 +--
>  include/drm/drm_atomic.h              | 11 ++++++++
>  include/drm/drm_atomic_helper.h       |  3 ++-
>  5 files changed, 61 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_atomic_helper.c b/drivers/gpu/drm/drm_atomic_helper.c
> index 6fe2303fccd9..7d23dafdffbc 100644
> --- a/drivers/gpu/drm/drm_atomic_helper.c
> +++ b/drivers/gpu/drm/drm_atomic_helper.c
> @@ -332,8 +332,15 @@ update_connector_routing(struct drm_atomic_state *state,
>  	 * about is ensuring that userspace can't do anything but shut off the
>  	 * display on a connector that was destroyed after its been notified,
>  	 * not before.
> +	 *
> +	 * Additionally, we also want to ignore connector registration when
> +	 * we're trying to restore an atomic state during system resume since
> +	 * there's a chance the connector may have been destroyed during the
> +	 * process, but it's better to ignore that then cause
> +	 * drm_atomic_helper_resume() to fail.
>  	 */
> -	if (drm_connector_is_unregistered(connector) && crtc_state->active) {
> +	if (!state->suspend_or_resume &&
> +	    drm_connector_is_unregistered(connector) && crtc_state->active) {
>  		DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] is not registered\n",
>  				 connector->base.id, connector->name);
>  		return -EINVAL;
> @@ -3144,6 +3151,7 @@ EXPORT_SYMBOL(drm_atomic_helper_shutdown);
>   * drm_atomic_helper_duplicate_state - duplicate an atomic state object
>   * @dev: DRM device
>   * @ctx: lock acquisition context
> + * @suspending: If this state is being duplicated as part of system suspend
>   *
>   * Makes a copy of the current atomic state by looping over all objects and
>   * duplicating their respective states. This is used for example by suspend/
> @@ -3166,7 +3174,8 @@ EXPORT_SYMBOL(drm_atomic_helper_shutdown);
>   */
>  struct drm_atomic_state *
>  drm_atomic_helper_duplicate_state(struct drm_device *dev,
> -				  struct drm_modeset_acquire_ctx *ctx)
> +				  struct drm_modeset_acquire_ctx *ctx,
> +				  bool suspending)

The duplicated state in the gpu reset code also wants suspending = true.
If you race a gpu hang with a hotunplug, we don't want blow up either. A
gpu reset which requires the display state reset too essentially looks
like a suspend/resume operation (the hw is completely off after the
reset).

For the watermark optimizer: That's run at boot-up, so right now 0 mst
connectors, and the setting doesn't matter. But if we ever get around to
fast-booting mst, then again I think blowing up when we race the unplug
with the watermark cleanup is probably not good.

tldr; I don't think we need this parameter. state->suspend_or_resume
becomes a bit misleading then, I'd call it state->duplicated. I think
that's also more accurately describing why we need special code: The issue
isn't resuming, it's trying to re-apply an atomic state that doesn't quite
fit to reality anymore.

Aside from the naming/api bikeshed looks good to me.
-Daniel

>  {
>  	struct drm_atomic_state *state;
>  	struct drm_connector *conn;
> @@ -3180,6 +3189,7 @@ drm_atomic_helper_duplicate_state(struct drm_device *dev,
>  		return ERR_PTR(-ENOMEM);
>  
>  	state->acquire_ctx = ctx;
> +	state->suspend_or_resume = suspending;
>  
>  	drm_for_each_crtc(crtc, dev) {
>  		struct drm_crtc_state *crtc_state;
> @@ -3263,7 +3273,7 @@ struct drm_atomic_state *drm_atomic_helper_suspend(struct drm_device *dev)
>  
>  	DRM_MODESET_LOCK_ALL_BEGIN(dev, ctx, 0, err);
>  
> -	state = drm_atomic_helper_duplicate_state(dev, &ctx);
> +	state = drm_atomic_helper_duplicate_state(dev, &ctx, true);
>  	if (IS_ERR(state))
>  		goto unlock;
>  
> diff --git a/drivers/gpu/drm/drm_dp_mst_topology.c b/drivers/gpu/drm/drm_dp_mst_topology.c
> index 8ba0e5b368da..371010f8d42e 100644
> --- a/drivers/gpu/drm/drm_dp_mst_topology.c
> +++ b/drivers/gpu/drm/drm_dp_mst_topology.c
> @@ -3032,6 +3032,10 @@ static int drm_dp_init_vcpi(struct drm_dp_mst_topology_mgr *mgr,
>   * @port as needed. It is not OK however, to call this function and
>   * drm_dp_atomic_release_vcpi_slots() in the same atomic check phase.
>   *
> + * When &drm_atomic_state.suspend_or_resume is set to %true%, this function
> + * will not perform any error checking and will instead simply retrieve the
> + * previously recorded VCPI allocation that was present before system suspend.
> + *
>   * See also:
>   * drm_dp_atomic_release_vcpi_slots()
>   * drm_dp_mst_atomic_check()
> @@ -3052,9 +3056,11 @@ int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
>  	if (IS_ERR(topology_state))
>  		return PTR_ERR(topology_state);
>  
> -	port = drm_dp_mst_topology_get_port_validated(mgr, port);
> -	if (port == NULL)
> -		return -EINVAL;
> +	if (!state->suspend_or_resume) {
> +		port = drm_dp_mst_topology_get_port_validated(mgr, port);
> +		if (!port)
> +			return -EINVAL;
> +	}
>  
>  	/* Find the current allocation for this port, if any */
>  	list_for_each_entry(pos, &topology_state->vcpis, next) {
> @@ -3062,6 +3068,19 @@ int drm_dp_atomic_find_vcpi_slots(struct drm_atomic_state *state,
>  			vcpi = pos;
>  			prev_slots = vcpi->vcpi;
>  
> +			/*
> +			 * When resuming, we just want to restore the previous
> +			 * VCPI without doing error checking
> +			 */
> +			if (state->suspend_or_resume) {
> +				DRM_DEBUG_ATOMIC("[CONNECTOR:%d:%s] [MST PORT:%p] restoring VCPI of %d\n",
> +						 port->connector->base.id,
> +						 port->connector->name,
> +						 port, prev_slots);
> +
> +				return prev_slots;
> +			}
> +
>  			/*
>  			 * This should never happen, unless the driver tries
>  			 * releasing and allocating the same VCPI allocation,
> @@ -3124,6 +3143,11 @@ EXPORT_SYMBOL(drm_dp_atomic_find_vcpi_slots);
>   * drm_dp_atomic_find_vcpi_slots() on the same @port in a single atomic check
>   * phase.
>   *
> + * When &drm_atomic_state.suspend_or_resume is set, this function will not
> + * modify the VCPI allocations in &drm_dp_mst_topology_state.vcpis, so that
> + * those VCPI allocations may be restored as-is during system resume. In this
> + * scenario, this function will always return 0.
> + *
>   * See also:
>   * drm_dp_atomic_find_vcpi_slots()
>   * drm_dp_mst_atomic_check()
> @@ -3140,6 +3164,12 @@ int drm_dp_atomic_release_vcpi_slots(struct drm_atomic_state *state,
>  	struct drm_dp_vcpi_allocation *pos;
>  	bool found = false;
>  
> +	/* During suspend, just keep our VCPI allocations as-is in the atomic
> +	 * state so they can be restored as-is at resume
> +	 */
> +	if (state->suspend_or_resume)
> +		return 0;
> +
>  	topology_state = drm_atomic_get_mst_topology_state(state, mgr);
>  	if (IS_ERR(topology_state))
>  		return PTR_ERR(topology_state);
> diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
> index 4a282532af81..4c70f1531119 100644
> --- a/drivers/gpu/drm/i915/intel_display.c
> +++ b/drivers/gpu/drm/i915/intel_display.c
> @@ -3824,7 +3824,7 @@ void intel_prepare_reset(struct drm_i915_private *dev_priv)
>  	 * Disabling the crtcs gracefully seems nicer. Also the
>  	 * g33 docs say we should at least disable all the planes.
>  	 */
> -	state = drm_atomic_helper_duplicate_state(dev, ctx);
> +	state = drm_atomic_helper_duplicate_state(dev, ctx, false);
>  	if (IS_ERR(state)) {
>  		ret = PTR_ERR(state);
>  		DRM_ERROR("Duplicating state failed with %i\n", ret);
> @@ -15043,7 +15043,7 @@ static void sanitize_watermarks(struct drm_device *dev)
>  		goto fail;
>  	}
>  
> -	state = drm_atomic_helper_duplicate_state(dev, &ctx);
> +	state = drm_atomic_helper_duplicate_state(dev, &ctx, false);
>  	if (WARN_ON(IS_ERR(state)))
>  		goto fail;
>  
> diff --git a/include/drm/drm_atomic.h b/include/drm/drm_atomic.h
> index 811b4a92568f..0897f51dcd62 100644
> --- a/include/drm/drm_atomic.h
> +++ b/include/drm/drm_atomic.h
> @@ -329,6 +329,17 @@ struct drm_atomic_state {
>  	bool allow_modeset : 1;
>  	bool legacy_cursor_update : 1;
>  	bool async_update : 1;
> +	/**
> +	 * @suspend_or_resume:
> +	 *
> +	 * Indicates whether or not this atomic state is being saved or
> +	 * committed as part of suspending or resuming the system
> +	 * respectively. Drivers and atomic helpers hould use this to fixup
> +	 * normal state inconsistencies that might arise between system
> +	 * suspend and resume, so as to ensure that restoring the atomic state
> +	 * at resume never fails.
> +	 */
> +	bool suspend_or_resume : 1;
>  	struct __drm_planes_state *planes;
>  	struct __drm_crtcs_state *crtcs;
>  	int num_connector;
> diff --git a/include/drm/drm_atomic_helper.h b/include/drm/drm_atomic_helper.h
> index 58214be3bf3d..359f557b6898 100644
> --- a/include/drm/drm_atomic_helper.h
> +++ b/include/drm/drm_atomic_helper.h
> @@ -129,7 +129,8 @@ int drm_atomic_helper_disable_all(struct drm_device *dev,
>  void drm_atomic_helper_shutdown(struct drm_device *dev);
>  struct drm_atomic_state *
>  drm_atomic_helper_duplicate_state(struct drm_device *dev,
> -				  struct drm_modeset_acquire_ctx *ctx);
> +				  struct drm_modeset_acquire_ctx *ctx,
> +				  bool suspending);
>  struct drm_atomic_state *drm_atomic_helper_suspend(struct drm_device *dev);
>  int drm_atomic_helper_commit_duplicated_state(struct drm_atomic_state *state,
>  					      struct drm_modeset_acquire_ctx *ctx);
> -- 
> 2.20.1
> 

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch