]> www.infradead.org Git - users/hch/misc.git/commit
usb: typec: tcpm: move tcpm_queue_vdm_unlocked to asynchronous work
authorRD Babiera <rdbabiera@google.com>
Tue, 6 May 2025 23:28:53 +0000 (23:28 +0000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 21 May 2025 11:05:41 +0000 (13:05 +0200)
commit324d45e53f1a36c88bc649dc39e0c8300a41be0a
tree14ebc2a90ad85e247f691f941fe415402d955674
parent0736299d090f5c6a1032678705c4bc0a9511a3db
usb: typec: tcpm: move tcpm_queue_vdm_unlocked to asynchronous work

A state check was previously added to tcpm_queue_vdm_unlocked to
prevent a deadlock where the DisplayPort Alt Mode driver would be
executing work and attempting to grab the tcpm_lock while the TCPM
was holding the lock and attempting to unregister the altmode, blocking
on the altmode driver's cancel_work_sync call.

Because the state check isn't protected, there is a small window
where the Alt Mode driver could determine that the TCPM is
in a ready state and attempt to grab the lock while the
TCPM grabs the lock and changes the TCPM state to one that
causes the deadlock. The callstack is provided below:

[110121.667392][    C7] Call trace:
[110121.667396][    C7]  __switch_to+0x174/0x338
[110121.667406][    C7]  __schedule+0x608/0x9f0
[110121.667414][    C7]  schedule+0x7c/0xe8
[110121.667423][    C7]  kernfs_drain+0xb0/0x114
[110121.667431][    C7]  __kernfs_remove+0x16c/0x20c
[110121.667436][    C7]  kernfs_remove_by_name_ns+0x74/0xe8
[110121.667442][    C7]  sysfs_remove_group+0x84/0xe8
[110121.667450][    C7]  sysfs_remove_groups+0x34/0x58
[110121.667458][    C7]  device_remove_groups+0x10/0x20
[110121.667464][    C7]  device_release_driver_internal+0x164/0x2e4
[110121.667475][    C7]  device_release_driver+0x18/0x28
[110121.667484][    C7]  bus_remove_device+0xec/0x118
[110121.667491][    C7]  device_del+0x1e8/0x4ac
[110121.667498][    C7]  device_unregister+0x18/0x38
[110121.667504][    C7]  typec_unregister_altmode+0x30/0x44
[110121.667515][    C7]  tcpm_reset_port+0xac/0x370
[110121.667523][    C7]  tcpm_snk_detach+0x84/0xb8
[110121.667529][    C7]  run_state_machine+0x4c0/0x1b68
[110121.667536][    C7]  tcpm_state_machine_work+0x94/0xe4
[110121.667544][    C7]  kthread_worker_fn+0x10c/0x244
[110121.667552][    C7]  kthread+0x104/0x1d4
[110121.667557][    C7]  ret_from_fork+0x10/0x20

[110121.667689][    C7] Workqueue: events dp_altmode_work
[110121.667697][    C7] Call trace:
[110121.667701][    C7]  __switch_to+0x174/0x338
[110121.667710][    C7]  __schedule+0x608/0x9f0
[110121.667717][    C7]  schedule+0x7c/0xe8
[110121.667725][    C7]  schedule_preempt_disabled+0x24/0x40
[110121.667733][    C7]  __mutex_lock+0x408/0xdac
[110121.667741][    C7]  __mutex_lock_slowpath+0x14/0x24
[110121.667748][    C7]  mutex_lock+0x40/0xec
[110121.667757][    C7]  tcpm_altmode_enter+0x78/0xb4
[110121.667764][    C7]  typec_altmode_enter+0xdc/0x10c
[110121.667769][    C7]  dp_altmode_work+0x68/0x164
[110121.667775][    C7]  process_one_work+0x1e4/0x43c
[110121.667783][    C7]  worker_thread+0x25c/0x430
[110121.667789][    C7]  kthread+0x104/0x1d4
[110121.667794][    C7]  ret_from_fork+0x10/0x20

Change tcpm_queue_vdm_unlocked to queue for tcpm_queue_vdm_work,
which can perform the state check while holding the TCPM lock
while the Alt Mode lock is no longer held. This requires a new
struct to hold the vdm data, altmode_vdm_event.

Fixes: cdc9946ea637 ("usb: typec: tcpm: enforce ready state when queueing alt mode vdm")
Cc: stable <stable@kernel.org>
Signed-off-by: RD Babiera <rdbabiera@google.com>
Reviewed-by: Heikki Krogerus <heikki.krogerus@linux.intel.com>
Reviewed-by: Badhri Jagan Sridharan <badhri@google.com>
Link: https://lore.kernel.org/r/20250506232853.1968304-2-rdbabiera@google.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/usb/typec/tcpm/tcpm.c