]> git.infradead.org Git - users/hch/xfs.git/commit
workqueue: Remove cpus_read_lock() from apply_wqattrs_lock()
authorLai Jiangshan <jiangshan.ljs@antgroup.com>
Thu, 11 Jul 2024 08:35:43 +0000 (16:35 +0800)
committerTejun Heo <tj@kernel.org>
Thu, 11 Jul 2024 22:50:34 +0000 (12:50 -1000)
commit19af457573838785290d27dd09a59140f541d1d8
treec11a88c3866deb47af708f2b48437d4f98abc168
parentfbb3d4c15dc0fe8a439de267db927a9ab2f44e98
workqueue: Remove cpus_read_lock() from apply_wqattrs_lock()

1726a1713590 ("workqueue: Put PWQ allocation and WQ enlistment in the same
lock C.S.") led to the following possible deadlock:

  WARNING: possible recursive locking detected
  6.10.0-rc5-00004-g1d4c6111406c #1 Not tainted
   --------------------------------------------
   swapper/0/1 is trying to acquire lock:
   c27760f4 (cpu_hotplug_lock){++++}-{0:0}, at: alloc_workqueue (kernel/workqueue.c:5152 kernel/workqueue.c:5730)

   but task is already holding lock:
   c27760f4 (cpu_hotplug_lock){++++}-{0:0}, at: padata_alloc (kernel/padata.c:1007)
   ...
   stack backtrace:
   ...
   cpus_read_lock (include/linux/percpu-rwsem.h:53 kernel/cpu.c:488)
   alloc_workqueue (kernel/workqueue.c:5152 kernel/workqueue.c:5730)
   padata_alloc (kernel/padata.c:1007 (discriminator 1))
   pcrypt_init_padata (crypto/pcrypt.c:327 (discriminator 1))
   pcrypt_init (crypto/pcrypt.c:353)
   do_one_initcall (init/main.c:1267)
   do_initcalls (init/main.c:1328 (discriminator 1) init/main.c:1345 (discriminator 1))
   kernel_init_freeable (init/main.c:1364)
   kernel_init (init/main.c:1469)
   ret_from_fork (arch/x86/kernel/process.c:153)
   ret_from_fork_asm (arch/x86/entry/entry_32.S:737)
   entry_INT80_32 (arch/x86/entry/entry_32.S:944)

This is caused by pcrypt allocating a workqueue while holding
cpus_read_lock(), so workqueue code can't do it again as that can lead to
deadlocks if down_write starts after the first down_read.

The pwq creations and installations have been reworked based on
wq_online_cpumask rather than cpu_online_mask making cpus_read_lock() is
unneeded during wqattrs changes. Fix the deadlock by removing
cpus_read_lock() from apply_wqattrs_lock().

tj: Updated changelog.

Signed-off-by: Lai Jiangshan <jiangshan.ljs@antgroup.com>
Fixes: 1726a1713590 ("workqueue: Put PWQ allocation and WQ enlistment in the same lock C.S.")
Link: http://lkml.kernel.org/r/202407081521.83b627c1-lkp@intel.com
Signed-off-by: Tejun Heo <tj@kernel.org>
kernel/workqueue.c