]> git.ipfire.org Git - thirdparty/kernel/stable-queue.git/blame - releases/4.18.11/drm-nouveau-fix-deadlocks-in-nouveau_connector_detect.patch
projects / thirdparty / kernel / stable-queue.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Drop watchdog patch
[thirdparty/kernel/stable-queue.git] / releases / 4.18.11 / drm-nouveau-fix-deadlocks-in-nouveau_connector_detect.patch
CommitLineData
9d09275e
GKH		 1From 3e1a12754d4df5804bfca5dedf09d2ba291bdc2a Mon Sep 17 00:00:00 2001
2From: Lyude Paul <lyude@redhat.com>
3Date: Wed, 15 Aug 2018 15:00:15 -0400
4Subject: drm/nouveau: Fix deadlocks in nouveau_connector_detect()
5
6From: Lyude Paul <lyude@redhat.com>
7
8commit 3e1a12754d4df5804bfca5dedf09d2ba291bdc2a upstream.
9
10When we disable hotplugging on the GPU, we need to be able to
11synchronize with each connector's hotplug interrupt handler before the
12interrupt is finally disabled. This can be a problem however, since
13nouveau_connector_detect() currently grabs a runtime power reference
14when handling connector probing. This will deadlock the runtime suspend
15handler like so:
16
17[ 861.480896] INFO: task kworker/0:2:61 blocked for more than 120 seconds.
18[ 861.483290] Tainted: G O 4.18.0-rc6Lyude-Test+ #1
19[ 861.485158] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
20[ 861.486332] kworker/0:2 D 0 61 2 0x80000000
21[ 861.487044] Workqueue: events nouveau_display_hpd_work [nouveau]
22[ 861.487737] Call Trace:
23[ 861.488394] __schedule+0x322/0xaf0
24[ 861.489070] schedule+0x33/0x90
25[ 861.489744] rpm_resume+0x19c/0x850
26[ 861.490392] ? finish_wait+0x90/0x90
27[ 861.491068] __pm_runtime_resume+0x4e/0x90
28[ 861.491753] nouveau_display_hpd_work+0x22/0x60 [nouveau]
29[ 861.492416] process_one_work+0x231/0x620
30[ 861.493068] worker_thread+0x44/0x3a0
31[ 861.493722] kthread+0x12b/0x150
32[ 861.494342] ? wq_pool_ids_show+0x140/0x140
33[ 861.494991] ? kthread_create_worker_on_cpu+0x70/0x70
34[ 861.495648] ret_from_fork+0x3a/0x50
35[ 861.496304] INFO: task kworker/6:2:320 blocked for more than 120 seconds.
36[ 861.496968] Tainted: G O 4.18.0-rc6Lyude-Test+ #1
37[ 861.497654] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
38[ 861.498341] kworker/6:2 D 0 320 2 0x80000080
39[ 861.499045] Workqueue: pm pm_runtime_work
40[ 861.499739] Call Trace:
41[ 861.500428] __schedule+0x322/0xaf0
42[ 861.501134] ? wait_for_completion+0x104/0x190
43[ 861.501851] schedule+0x33/0x90
44[ 861.502564] schedule_timeout+0x3a5/0x590
45[ 861.503284] ? mark_held_locks+0x58/0x80
46[ 861.503988] ? _raw_spin_unlock_irq+0x2c/0x40
47[ 861.504710] ? wait_for_completion+0x104/0x190
48[ 861.505417] ? trace_hardirqs_on_caller+0xf4/0x190
49[ 861.506136] ? wait_for_completion+0x104/0x190
50[ 861.506845] wait_for_completion+0x12c/0x190
51[ 861.507555] ? wake_up_q+0x80/0x80
52[ 861.508268] flush_work+0x1c9/0x280
53[ 861.508990] ? flush_workqueue_prep_pwqs+0x1b0/0x1b0
54[ 861.509735] nvif_notify_put+0xb1/0xc0 [nouveau]
55[ 861.510482] nouveau_display_fini+0xbd/0x170 [nouveau]
56[ 861.511241] nouveau_display_suspend+0x67/0x120 [nouveau]
57[ 861.511969] nouveau_do_suspend+0x5e/0x2d0 [nouveau]
58[ 861.512715] nouveau_pmops_runtime_suspend+0x47/0xb0 [nouveau]
59[ 861.513435] pci_pm_runtime_suspend+0x6b/0x180
60[ 861.514165] ? pci_has_legacy_pm_support+0x70/0x70
61[ 861.514897] __rpm_callback+0x7a/0x1d0
62[ 861.515618] ? pci_has_legacy_pm_support+0x70/0x70
63[ 861.516313] rpm_callback+0x24/0x80
64[ 861.517027] ? pci_has_legacy_pm_support+0x70/0x70
65[ 861.517741] rpm_suspend+0x142/0x6b0
66[ 861.518449] pm_runtime_work+0x97/0xc0
67[ 861.519144] process_one_work+0x231/0x620
68[ 861.519831] worker_thread+0x44/0x3a0
69[ 861.520522] kthread+0x12b/0x150
70[ 861.521220] ? wq_pool_ids_show+0x140/0x140
71[ 861.521925] ? kthread_create_worker_on_cpu+0x70/0x70
72[ 861.522622] ret_from_fork+0x3a/0x50
73[ 861.523299] INFO: task kworker/6:0:1329 blocked for more than 120 seconds.
74[ 861.523977] Tainted: G O 4.18.0-rc6Lyude-Test+ #1
75[ 861.524644] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
76[ 861.525349] kworker/6:0 D 0 1329 2 0x80000000
77[ 861.526073] Workqueue: events nvif_notify_work [nouveau]
78[ 861.526751] Call Trace:
79[ 861.527411] __schedule+0x322/0xaf0
80[ 861.528089] schedule+0x33/0x90
81[ 861.528758] rpm_resume+0x19c/0x850
82[ 861.529399] ? finish_wait+0x90/0x90
83[ 861.530073] __pm_runtime_resume+0x4e/0x90
84[ 861.530798] nouveau_connector_detect+0x7e/0x510 [nouveau]
85[ 861.531459] ? ww_mutex_lock+0x47/0x80
86[ 861.532097] ? ww_mutex_lock+0x47/0x80
87[ 861.532819] ? drm_modeset_lock+0x88/0x130 [drm]
88[ 861.533481] drm_helper_probe_detect_ctx+0xa0/0x100 [drm_kms_helper]
89[ 861.534127] drm_helper_hpd_irq_event+0xa4/0x120 [drm_kms_helper]
90[ 861.534940] nouveau_connector_hotplug+0x98/0x120 [nouveau]
91[ 861.535556] nvif_notify_work+0x2d/0xb0 [nouveau]
92[ 861.536221] process_one_work+0x231/0x620
93[ 861.536994] worker_thread+0x44/0x3a0
94[ 861.537757] kthread+0x12b/0x150
95[ 861.538463] ? wq_pool_ids_show+0x140/0x140
96[ 861.539102] ? kthread_create_worker_on_cpu+0x70/0x70
97[ 861.539815] ret_from_fork+0x3a/0x50
98[ 861.540521]
99 Showing all locks held in the system:
100[ 861.541696] 2 locks held by kworker/0:2/61:
101[ 861.542406] #0: 000000002dbf8af5 ((wq_completion)"events"){+.+.}, at: process_one_work+0x1b3/0x620
102[ 861.543071] #1: 0000000076868126 ((work_completion)(&drm->hpd_work)){+.+.}, at: process_one_work+0x1b3/0x620
103[ 861.543814] 1 lock held by khungtaskd/64:
104[ 861.544535] #0: 0000000059db4b53 (rcu_read_lock){....}, at: debug_show_all_locks+0x23/0x185
105[ 861.545160] 3 locks held by kworker/6:2/320:
106[ 861.545896] #0: 00000000d9e1bc59 ((wq_completion)"pm"){+.+.}, at: process_one_work+0x1b3/0x620
107[ 861.546702] #1: 00000000c9f92d84 ((work_completion)(&dev->power.work)){+.+.}, at: process_one_work+0x1b3/0x620
108[ 861.547443] #2: 000000004afc5de1 (drm_connector_list_iter){.+.+}, at: nouveau_display_fini+0x96/0x170 [nouveau]
109[ 861.548146] 1 lock held by dmesg/983:
110[ 861.548889] 2 locks held by zsh/1250:
111[ 861.549605] #0: 00000000348e3cf6 (&tty->ldisc_sem){++++}, at: ldsem_down_read+0x37/0x40
112[ 861.550393] #1: 000000007009a7a8 (&ldata->atomic_read_lock){+.+.}, at: n_tty_read+0xc1/0x870
113[ 861.551122] 6 locks held by kworker/6:0/1329:
114[ 861.551957] #0: 000000002dbf8af5 ((wq_completion)"events"){+.+.}, at: process_one_work+0x1b3/0x620
115[ 861.552765] #1: 00000000ddb499ad ((work_completion)(&notify->work)#2){+.+.}, at: process_one_work+0x1b3/0x620
116[ 861.553582] #2: 000000006e013cbe (&dev->mode_config.mutex){+.+.}, at: drm_helper_hpd_irq_event+0x6c/0x120 [drm_kms_helper]
117[ 861.554357] #3: 000000004afc5de1 (drm_connector_list_iter){.+.+}, at: drm_helper_hpd_irq_event+0x78/0x120 [drm_kms_helper]
118[ 861.555227] #4: 0000000044f294d9 (crtc_ww_class_acquire){+.+.}, at: drm_helper_probe_detect_ctx+0x3d/0x100 [drm_kms_helper]
119[ 861.556133] #5: 00000000db193642 (crtc_ww_class_mutex){+.+.}, at: drm_modeset_lock+0x4b/0x130 [drm]
120
121[ 861.557864] =============================================
122
123[ 861.559507] NMI backtrace for cpu 2
124[ 861.560363] CPU: 2 PID: 64 Comm: khungtaskd Tainted: G O 4.18.0-rc6Lyude-Test+ #1
125[ 861.561197] Hardware name: LENOVO 20EQS64N0B/20EQS64N0B, BIOS N1EET78W (1.51 ) 05/18/2018
126[ 861.561948] Call Trace:
127[ 861.562757] dump_stack+0x8e/0xd3
128[ 861.563516] nmi_cpu_backtrace.cold.3+0x14/0x5a
129[ 861.564269] ? lapic_can_unplug_cpu.cold.27+0x42/0x42
130[ 861.565029] nmi_trigger_cpumask_backtrace+0xa1/0xae
131[ 861.565789] arch_trigger_cpumask_backtrace+0x19/0x20
132[ 861.566558] watchdog+0x316/0x580
133[ 861.567355] kthread+0x12b/0x150
134[ 861.568114] ? reset_hung_task_detector+0x20/0x20
135[ 861.568863] ? kthread_create_worker_on_cpu+0x70/0x70
136[ 861.569598] ret_from_fork+0x3a/0x50
137[ 861.570370] Sending NMI from CPU 2 to CPUs 0-1,3-7:
138[ 861.571426] NMI backtrace for cpu 6 skipped: idling at intel_idle+0x7f/0x120
139[ 861.571429] NMI backtrace for cpu 7 skipped: idling at intel_idle+0x7f/0x120
140[ 861.571432] NMI backtrace for cpu 3 skipped: idling at intel_idle+0x7f/0x120
141[ 861.571464] NMI backtrace for cpu 5 skipped: idling at intel_idle+0x7f/0x120
142[ 861.571467] NMI backtrace for cpu 0 skipped: idling at intel_idle+0x7f/0x120
143[ 861.571469] NMI backtrace for cpu 4 skipped: idling at intel_idle+0x7f/0x120
144[ 861.571472] NMI backtrace for cpu 1 skipped: idling at intel_idle+0x7f/0x120
145[ 861.572428] Kernel panic - not syncing: hung_task: blocked tasks
146
147So: fix this by making it so that normal hotplug handling /only/ happens
148so long as the GPU is currently awake without any pending runtime PM
149requests. In the event that a hotplug occurs while the device is
150suspending or resuming, we can simply defer our response until the GPU
151is fully runtime resumed again.
152
153Changes since v4:
154- Use a new trick I came up with using pm_runtime_get() instead of the
155 hackish junk we had before
156
157Signed-off-by: Lyude Paul <lyude@redhat.com>
158Reviewed-by: Karol Herbst <kherbst@redhat.com>
159Acked-by: Daniel Vetter <daniel@ffwll.ch>
160Cc: stable@vger.kernel.org
161Cc: Lukas Wunner <lukas@wunner.de>
162Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
163Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
164
165---
166 drivers/gpu/drm/nouveau/nouveau_connector.c | 22 ++++++++++++++++++++++
167 1 file changed, 22 insertions(+)
168
169--- a/drivers/gpu/drm/nouveau/nouveau_connector.c
170+++ b/drivers/gpu/drm/nouveau/nouveau_connector.c
171@@ -1120,6 +1120,26 @@ nouveau_connector_hotplug(struct nvif_no
172 const struct nvif_notify_conn_rep_v0 *rep = notify->data;
173 const char *name = connector->name;
174 struct nouveau_encoder *nv_encoder;
175+ int ret;
176+
177+ ret = pm_runtime_get(drm->dev->dev);
178+ if (ret == 0) {
179+ /* We can't block here if there's a pending PM request
180+ * running, as we'll deadlock nouveau_display_fini() when it
181+ * calls nvif_put() on our nvif_notify struct. So, simply
182+ * defer the hotplug event until the device finishes resuming
183+ */
184+ NV_DEBUG(drm, "Deferring HPD on %s until runtime resume\n",
185+ name);
186+ schedule_work(&drm->hpd_work);
187+
188+ pm_runtime_put_noidle(drm->dev->dev);
189+ return NVIF_NOTIFY_KEEP;
190+ } else if (ret != 1 && ret != -EACCES) {
191+ NV_WARN(drm, "HPD on %s dropped due to RPM failure: %d\n",
192+ name, ret);
193+ return NVIF_NOTIFY_DROP;
194+ }
195
196 if (rep->mask & NVIF_NOTIFY_CONN_V0_IRQ) {
197 NV_DEBUG(drm, "service %s\n", name);
198@@ -1137,6 +1157,8 @@ nouveau_connector_hotplug(struct nvif_no
199 drm_helper_hpd_irq_event(connector->dev);
200 }
201
202+ pm_runtime_mark_last_busy(drm->dev->dev);
203+ pm_runtime_put_autosuspend(drm->dev->dev);
204 return NVIF_NOTIFY_KEEP;
205 }
206
Mirror of https://git.kernel.org/pub/scm/linux/kernel/git/stable/stable-queue.git