mbox series

[v2,0/2] Fix thread synchronization issues

Message ID 20250424084429.3220757-1-akuchynski@chromium.org
Headers show
Series Fix thread synchronization issues | expand

Message

Andrei Kuchynski April 24, 2025, 8:44 a.m. UTC 
v2: ucsi_con_mutex_lock / ucsi_con_mutex_unlock functions are not longer
exported. Thanks to Heikki Krogerus for noticing it.

This patch series addresses deadlock issue and prevents a NULL pointer
access within the UCSI driver.

The first patch resolves a deadlock that occurs during partner device
removal. The deadlock scenario involves:
Thread 1: ucsi_handle_connector_change holding the connector mutex
(con->mutex) while waiting for dp_altmode_work() to complete
Thread 2: dp_altmode_work concurrently attempting to acquire the same mutex
Thread 3: ucsi_poll_worker is indirectly affected as it would be blocked
by the held mutex

The crash log:
 INFO: task kworker/u16:16:7254 blocked for more than 122 seconds.
       Not tainted 6.6.75-g5e7fdbab3246 #1
 task:kworker/u16:16  state:D stack:0 pid:7254  ppid:2 flags:0x00000208
 Workqueue: 16130000.spi:ec@0:typec ucsi_poll_worker [typec_ucsi]
 Call trace:
  __switch_to+0x130/0x158
  __schedule+0x950/0xa18
  schedule+0x68/0xc8
  schedule_preempt_disabled+0x28/0x40
  __mutex_lock+0x390/0xac0
  __mutex_lock_slowpath+0x18/0x28
  mutex_lock+0x28/0x60
  ucsi_poll_worker+0x24/0x138 [typec_ucsi (HASH:5d4a 1)]
  process_scheduled_works+0x258/0x4d8
  worker_thread+0x330/0x428
  kthread+0x108/0x1d0
  ret_from_fork+0x10/0x20
 INFO: task kworker/2:0:25045 blocked for more than 122 seconds.
       Not tainted 6.6.75-g5e7fdbab3246 #1
 task:kworker/2:0     state:D stack:0 pid:25045 ppid:2 flags:0x00000208
 Workqueue: events dp_altmode_work [typec_displayport]
 Call trace:
  __switch_to+0x130/0x158
  __schedule+0x950/0xa18
  schedule+0x68/0xc8
  schedule_preempt_disabled+0x28/0x40
  __mutex_lock+0x390/0xac0
  __mutex_lock_slowpath+0x18/0x28
  mutex_lock+0x28/0x60
  ucsi_displayport_enter+0x44/0x160 [typec_ucsi (HASH:5d4a 1)]
  typec_altmode_enter+0x110/0x120
  dp_altmode_work+0xe8/0x170 [typec_displayport (HASH:c690 2)]
  process_scheduled_works+0x258/0x4d8
  worker_thread+0x330/0x428
  kthread+0x108/0x1d0
  ret_from_fork+0x10/0x20
 INFO: task kworker/2:3:25952 blocked for more than 122 seconds.
       Not tainted 6.6.75-g5e7fdbab3246 #1
 task:kworker/2:3     state:D stack:0 pid:25952 ppid:2 flags:0x00000208
 Workqueue: events ucsi_handle_connector_change [typec_ucsi]
 Call trace:
  __switch_to+0x130/0x158
  __schedule+0x950/0xa18
  schedule+0x68/0xc8
  schedule_timeout+0x44/0x1b8
  wait_for_common+0xe4/0x1d8
  wait_for_completion+0x1c/0x30
  __flush_work+0x22c/0x2a8
  __cancel_work_timer+0x130/0x1d8
  cancel_work_sync+0x18/0x30
  dp_altmode_remove+0x20/0x70 [typec_displayport (HASH:c690 2)]
  typec_remove+0x60/0x120
  device_release_driver_internal+0x164/0x258
  device_release_driver+0x1c/0x30
  bus_remove_device+0x140/0x160
  device_del+0x1c0/0x330
  device_unregister+0x1c/0x40
  typec_unregister_altmode+0x40/0x58
  ucsi_unregister_partner+0x90/0x170 [typec_ucsi (HASH:5d4a 1)]
  ucsi_handle_connector_change+0x184/0x468 [typec_ucsi (HASH:5d4a 1)]
  process_scheduled_works+0x258/0x4d8
  worker_thread+0x330/0x428
  kthread+0x108/0x1d0
  ret_from_fork+0x10/0x20

The second patch prevents a NULL pointer access that occurs if
ucsi_displayport_work is still accessing partner-related data after
the partner has been unregistered.

The crash log:
 BUG: kernel NULL pointer dereference, address: 0000000000000300
 #PF: supervisor read access in kernel mode
 #PF: error_code(0x0000) - not-present page
 Oops: 0000 [#1] PREEMPT SMP NOPTI
 CPU: 3 PID: 3413 Comm: kworker/3:3 Tainted: G     U 6.6.76-07480
 Hardware name: Google Brox/Brox, BIOS Google_Brox.16080.59.0 12/18/2024
 Workqueue: events ucsi_displayport_work [typec_ucsi]
 RIP: 0010:typec_altmode_vdm+0x8f/0x4d0 [typec]
 RSP: 0018:ffffa929c825fe28 EFLAGS: 00010246
 RAX: 0000000000000014 RBX: 0000000000000001 RCX: 43032ac4505ac800
 RDX: ffff9d8def8e77c0 RSI: ffff9d8def8db5c8 RDI: ffff9d8def8db5c8
 RBP: ffffa929c825fe50 R08: 0000000000001fff R09: ffffffff9385e1e0
 R10: 0000000000005ffd R11: 00000000ffffdfff R12: ffff9d8ab4b67c08
 R13: 0000000000000000 R14: 0000000000000000 R15: 00000000ff01a144
 FS:  0000000000000000(0000) GS:ffff9d8def8c0000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 0000000000000300 CR3: 0000000148438000 CR4: 0000000000750ee0
 PKRU: 55555554
 Call Trace:
  ? __die_body+0xae/0xb0
  ? page_fault_oops+0x381/0x3e0
  ? prb_read_valid+0x29/0x50
  ? exc_page_fault+0x4f/0xa0
  ? asm_exc_page_fault+0x22/0x30
  ? typec_altmode_vdm+0x8f/0x4d0 [typec (HASH:c6ec 5)]
  ? typec_altmode_vdm+0x87/0x4d0 [typec (HASH:c6ec 5)]
  ucsi_displayport_work+0x34/0x70 [typec_ucsi (HASH:698a 6)]
  worker_thread+0x386/0x930

Andrei Kuchynski (2):
  usb: typec: ucsi: displayport: Fix deadlock
  usb: typec: ucsi: displayport: Fix NULL pointer access

 drivers/usb/typec/ucsi/displayport.c | 21 ++++++++++-------
 drivers/usb/typec/ucsi/ucsi.c        | 34 ++++++++++++++++++++++++++++
 drivers/usb/typec/ucsi/ucsi.h        |  2 ++
 3 files changed, 49 insertions(+), 8 deletions(-)