drm/nouveau: Fix deadlock in nv50_mstm_register_connector()
authorLyude Paul <lyude@redhat.com>
Wed, 2 May 2018 23:38:48 +0000 (19:38 -0400)
committerBen Skeggs <bskeggs@redhat.com>
Thu, 10 May 2018 03:11:52 +0000 (13:11 +1000)
commit352672db857290ab5b0e2b6a99c414f92bee024c
treeda77a102c3da20082f4c45e4ce297d0486b5a264
parent0d5a03c3d9254813ca76d7886ff9ed76a0aea545
drm/nouveau: Fix deadlock in nv50_mstm_register_connector()

Currently; we're grabbing all of the modesetting locks before adding MST
connectors to fbdev. This isn't actually necessary, and causes a
deadlock as well:

======================================================
WARNING: possible circular locking dependency detected
4.17.0-rc3Lyude-Test+ #1 Tainted: G           O
------------------------------------------------------
kworker/1:0/18 is trying to acquire lock:
00000000c832f62d (&helper->lock){+.+.}, at: drm_fb_helper_add_one_connector+0x2a/0x60 [drm_kms_helper]

but task is already holding lock:
00000000942e28e2 (crtc_ww_class_mutex){+.+.}, at: drm_modeset_backoff+0x8e/0x1c0 [drm]

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #3 (crtc_ww_class_mutex){+.+.}:
       ww_mutex_lock+0x43/0x80
       drm_modeset_lock+0x71/0x130 [drm]
       drm_helper_probe_single_connector_modes+0x7d/0x6b0 [drm_kms_helper]
       drm_setup_crtcs+0x15e/0xc90 [drm_kms_helper]
       __drm_fb_helper_initial_config_and_unlock+0x29/0x480 [drm_kms_helper]
       nouveau_fbcon_init+0x138/0x1a0 [nouveau]
       nouveau_drm_load+0x173/0x7e0 [nouveau]
       drm_dev_register+0x134/0x1c0 [drm]
       drm_get_pci_dev+0x8e/0x160 [drm]
       nouveau_drm_probe+0x1a9/0x230 [nouveau]
       pci_device_probe+0xcd/0x150
       driver_probe_device+0x30b/0x480
       __driver_attach+0xbc/0xe0
       bus_for_each_dev+0x67/0x90
       bus_add_driver+0x164/0x260
       driver_register+0x57/0xc0
       do_one_initcall+0x4d/0x323
       do_init_module+0x5b/0x1f8
       load_module+0x20e5/0x2ac0
       __do_sys_finit_module+0xb7/0xd0
       do_syscall_64+0x60/0x1b0
       entry_SYSCALL_64_after_hwframe+0x49/0xbe

-> #2 (crtc_ww_class_acquire){+.+.}:
       drm_helper_probe_single_connector_modes+0x58/0x6b0 [drm_kms_helper]
       drm_setup_crtcs+0x15e/0xc90 [drm_kms_helper]
       __drm_fb_helper_initial_config_and_unlock+0x29/0x480 [drm_kms_helper]
       nouveau_fbcon_init+0x138/0x1a0 [nouveau]
       nouveau_drm_load+0x173/0x7e0 [nouveau]
       drm_dev_register+0x134/0x1c0 [drm]
       drm_get_pci_dev+0x8e/0x160 [drm]
       nouveau_drm_probe+0x1a9/0x230 [nouveau]
       pci_device_probe+0xcd/0x150
       driver_probe_device+0x30b/0x480
       __driver_attach+0xbc/0xe0
       bus_for_each_dev+0x67/0x90
       bus_add_driver+0x164/0x260
       driver_register+0x57/0xc0
       do_one_initcall+0x4d/0x323
       do_init_module+0x5b/0x1f8
       load_module+0x20e5/0x2ac0
       __do_sys_finit_module+0xb7/0xd0
       do_syscall_64+0x60/0x1b0
       entry_SYSCALL_64_after_hwframe+0x49/0xbe

-> #1 (&dev->mode_config.mutex){+.+.}:
       drm_setup_crtcs+0x10c/0xc90 [drm_kms_helper]
       __drm_fb_helper_initial_config_and_unlock+0x29/0x480 [drm_kms_helper]
       nouveau_fbcon_init+0x138/0x1a0 [nouveau]
       nouveau_drm_load+0x173/0x7e0 [nouveau]
       drm_dev_register+0x134/0x1c0 [drm]
       drm_get_pci_dev+0x8e/0x160 [drm]
       nouveau_drm_probe+0x1a9/0x230 [nouveau]
       pci_device_probe+0xcd/0x150
       driver_probe_device+0x30b/0x480
       __driver_attach+0xbc/0xe0
       bus_for_each_dev+0x67/0x90
       bus_add_driver+0x164/0x260
       driver_register+0x57/0xc0
       do_one_initcall+0x4d/0x323
       do_init_module+0x5b/0x1f8
       load_module+0x20e5/0x2ac0
       __do_sys_finit_module+0xb7/0xd0
       do_syscall_64+0x60/0x1b0
       entry_SYSCALL_64_after_hwframe+0x49/0xbe

-> #0 (&helper->lock){+.+.}:
       __mutex_lock+0x70/0x9d0
       drm_fb_helper_add_one_connector+0x2a/0x60 [drm_kms_helper]
       nv50_mstm_register_connector+0x2c/0x50 [nouveau]
       drm_dp_add_port+0x2f5/0x420 [drm_kms_helper]
       drm_dp_send_link_address+0x155/0x1e0 [drm_kms_helper]
       drm_dp_add_port+0x33f/0x420 [drm_kms_helper]
       drm_dp_send_link_address+0x155/0x1e0 [drm_kms_helper]
       drm_dp_check_and_send_link_address+0x87/0xd0 [drm_kms_helper]
       drm_dp_mst_link_probe_work+0x4d/0x80 [drm_kms_helper]
       process_one_work+0x20d/0x650
       worker_thread+0x3a/0x390
       kthread+0x11e/0x140
       ret_from_fork+0x3a/0x50

other info that might help us debug this:
Chain exists of:
  &helper->lock --> crtc_ww_class_acquire --> crtc_ww_class_mutex
 Possible unsafe locking scenario:
       CPU0                    CPU1
       ----                    ----
  lock(crtc_ww_class_mutex);
                               lock(crtc_ww_class_acquire);
                               lock(crtc_ww_class_mutex);
  lock(&helper->lock);

 *** DEADLOCK ***
5 locks held by kworker/1:0/18:
 #0: 000000004a05cd50 ((wq_completion)"events_long"){+.+.}, at: process_one_work+0x187/0x650
 #1: 00000000601c11d1 ((work_completion)(&mgr->work)){+.+.}, at: process_one_work+0x187/0x650
 #2: 00000000586ca0df (&dev->mode_config.mutex){+.+.}, at: drm_modeset_lock_all+0x3a/0x1b0 [drm]
 #3: 00000000d3ca0ffa (crtc_ww_class_acquire){+.+.}, at: drm_modeset_lock_all+0x44/0x1b0 [drm]
 #4: 00000000942e28e2 (crtc_ww_class_mutex){+.+.}, at: drm_modeset_backoff+0x8e/0x1c0 [drm]

stack backtrace:
CPU: 1 PID: 18 Comm: kworker/1:0 Tainted: G           O      4.17.0-rc3Lyude-Test+ #1
Hardware name: Gateway FX6840/FX6840, BIOS P01-A3         05/17/2010
Workqueue: events_long drm_dp_mst_link_probe_work [drm_kms_helper]
Call Trace:
 dump_stack+0x85/0xcb
 print_circular_bug.isra.38+0x1ce/0x1db
 __lock_acquire+0x128f/0x1350
 ? lock_acquire+0x9f/0x200
 ? lock_acquire+0x9f/0x200
 ? __ww_mutex_lock.constprop.13+0x8f/0x1000
 lock_acquire+0x9f/0x200
 ? drm_fb_helper_add_one_connector+0x2a/0x60 [drm_kms_helper]
 ? drm_fb_helper_add_one_connector+0x2a/0x60 [drm_kms_helper]
 __mutex_lock+0x70/0x9d0
 ? drm_fb_helper_add_one_connector+0x2a/0x60 [drm_kms_helper]
 ? ww_mutex_lock+0x43/0x80
 ? _cond_resched+0x15/0x30
 ? ww_mutex_lock+0x43/0x80
 ? drm_modeset_lock+0xb2/0x130 [drm]
 ? drm_fb_helper_add_one_connector+0x2a/0x60 [drm_kms_helper]
 drm_fb_helper_add_one_connector+0x2a/0x60 [drm_kms_helper]
 nv50_mstm_register_connector+0x2c/0x50 [nouveau]
 drm_dp_add_port+0x2f5/0x420 [drm_kms_helper]
 ? mark_held_locks+0x50/0x80
 ? kfree+0xcf/0x2a0
 ? drm_dp_check_mstb_guid+0xd6/0x120 [drm_kms_helper]
 ? trace_hardirqs_on_caller+0xed/0x180
 ? drm_dp_check_mstb_guid+0xd6/0x120 [drm_kms_helper]
 drm_dp_send_link_address+0x155/0x1e0 [drm_kms_helper]
 drm_dp_add_port+0x33f/0x420 [drm_kms_helper]
 ? nouveau_connector_aux_xfer+0x7c/0xb0 [nouveau]
 ? find_held_lock+0x2d/0x90
 ? drm_dp_dpcd_access+0xd9/0xf0 [drm_kms_helper]
 ? __mutex_unlock_slowpath+0x3b/0x280
 ? drm_dp_dpcd_access+0xd9/0xf0 [drm_kms_helper]
 drm_dp_send_link_address+0x155/0x1e0 [drm_kms_helper]
 drm_dp_check_and_send_link_address+0x87/0xd0 [drm_kms_helper]
 drm_dp_mst_link_probe_work+0x4d/0x80 [drm_kms_helper]
 process_one_work+0x20d/0x650
 worker_thread+0x3a/0x390
 ? process_one_work+0x650/0x650
 kthread+0x11e/0x140
 ? kthread_create_worker_on_cpu+0x50/0x50
 ret_from_fork+0x3a/0x50

Taking example from i915, the only time we need to hold any modesetting
locks is when changing the port on the mstc, and in that case we only
need to hold the connection mutex.

Signed-off-by: Lyude Paul <lyude@redhat.com>
Cc: Karol Herbst <kherbst@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
drivers/gpu/drm/nouveau/nv50_display.c