workqueue: Make the PWQ allocation and WQ enlistment atomic
Lai Jiangshan
reviewerv2 cover workqueue: Make the PWQ allocation and WQ enlistment atomic
From: Lai Jiangshan <[email protected]>
The PWQ allocation and WQ enlistment are not within the same lock-held critical section; therefore, their states can become out of sync when the user modifies the unbound mask or if CPU hotplug events occur in the interim since those operations only update the WQs that are already in the list.
Change from v1:
Init rescuer's affinities as the wq's effective cpumask
V1: https://lore.kernel.org/lkml/[email protected]/
Lai Jiangshan (5):
workqueue: Register sysfs after the whole creation of the new wq
workqueue: Make rescuer initialization as the last step of the
creation of a new wq
workqueue: Move kthread_flush_worker() out of alloc_and_link_pwqs()
workqueue: Put PWQ allocation and WQ enlistment in the same lock C.S.
workqueue: Init rescuer's affinities as the wq's effective cpumask
kernel/workqueue.c | 87 +++++++++++++++++++++++++--------------------- 1 file changed, 47 insertions(+), 40 deletions(-)
Signature · 1 line
2.19.1.6.gb485710b
Lai Jiangshan
reviewerPATCH v2 1/5 workqueue: Register sysfs after the whole creation of the new wq
From: Lai Jiangshan <[email protected]>
workqueue creation includes adding it to the workqueue list.
Prepare for moving the whole workqueue initializing procedure into wq_pool_mutex and cpu hotplug locks.
--- kernel/workqueue.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
kernel/workqueue.c +3 -3
@@ -5693,9 +5693,6 @@ struct workqueue_struct *alloc_workqueue(const char *fmt,if (wq_online &&init_rescuer (wq) <0 ) goto err_destroy; -if ((wq->flags &WQ_SYSFS ) &&workqueue_sysfs_register (wq)) - goto err_destroy; -/* * wq_pool_mutex protects global freeze state and workqueues list. * Grab it, adjust max_active and add the new @wq to workqueues
@@ -5711,6 +5708,9 @@ struct workqueue_struct *alloc_workqueue(const char *fmt,mutex_unlock (&wq_pool_mutex); +if ((wq->flags &WQ_SYSFS ) &&workqueue_sysfs_register (wq)) + goto err_destroy; +return wq; err_free_node_nr_active:
Signature · 1 line
2.19.1.6.gb485710b
Lai Jiangshan
reviewerPATCH v2 2/5 workqueue: Make rescuer initialization as the last step of the creation of a new wq
In reply to Lai JiangshanFrom: Lai Jiangshan <[email protected]>
For early wq allocation, rescuer initialization is the last step of the creation of a new wq. Make the behavior the same for all allocations.
Prepare for initializing rescuer's affinities with the default pwq's affinities.
Prepare for moving the whole workqueue initializing procedure into wq_pool_mutex and cpu hotplug locks.
--- kernel/workqueue.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
kernel/workqueue.c +3 -3
@@ -5690,9 +5690,6 @@ struct workqueue_struct *alloc_workqueue(const char *fmt,if (alloc_and_link_pwqs (wq) <0 ) goto err_free_node_nr_active; -if (wq_online &&init_rescuer (wq) <0 ) - goto err_destroy; -/* * wq_pool_mutex protects global freeze state and workqueues list. * Grab it, adjust max_active and add the new @wq to workqueues
@@ -5708,6 +5705,9 @@ struct workqueue_struct *alloc_workqueue(const char *fmt,mutex_unlock (&wq_pool_mutex); +if (wq_online &&init_rescuer (wq) <0 ) + goto err_destroy; +if ((wq->flags &WQ_SYSFS ) &&workqueue_sysfs_register (wq)) goto err_destroy;
Signature · 1 line
2.19.1.6.gb485710b
Lai Jiangshan
reviewerPATCH v2 3/5 workqueue: Move kthread_flush_worker() out of alloc_and_link_pwqs()
In reply to Lai JiangshanFrom: Lai Jiangshan <[email protected]>
kthread_flush_worker() can't be called with wq_pool_mutex held.
Prepare for moving wq_pool_mutex and cpu hotplug lock out of alloc_and_link_pwqs().
--- kernel/workqueue.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-)
kernel/workqueue.c +8 -7
@@ -5476,12 +5476,6 @@ static int alloc_and_link_pwqs(struct workqueue_struct *wq) }cpus_read_unlock (); -/* for unbound pwq, flush the pwq_release_worker ensures that the -* pwq_release_workfn() completes before calling kfree(wq). -*/ -if (ret) -kthread_flush_worker (pwq_release_worker); -return ret; enomem:
@@ -5714,8 +5708,15 @@ struct workqueue_struct *alloc_workqueue(const char *fmt,return wq; err_free_node_nr_active: -if (wq->flags &WQ_UNBOUND ) +/* +* Failed alloc_and_link_pwqs() may leave pending pwq->release_work, +* flushing the pwq_release_worker ensures that the pwq_release_workfn() +* completes before calling kfree(wq). +*/ +if (wq->flags &WQ_UNBOUND ) { +kthread_flush_worker (pwq_release_worker);free_node_nr_active (wq->node_nr_active); + } err_unreg_lockdep:wq_unregister_lockdep (wq);wq_free_lockdep (wq);
Signature · 1 line
2.19.1.6.gb485710b
Lai Jiangshan
reviewerPATCH v2 4/5 workqueue: Put PWQ allocation and WQ enlistment in the same lock C.S.
In reply to Lai JiangshanFrom: Lai Jiangshan <[email protected]>
The PWQ allocation and WQ enlistment are not within the same lock-held critical section; therefore, their states can become out of sync when the user modifies the unbound mask or if CPU hotplug events occur in the interim since those operations only update the WQs that are already in the list.
Make the PWQ allocation and WQ enlistment atomic.
--- kernel/workqueue.c | 54 ++++++++++++++++++++++++---------------------- 1 file changed, 28 insertions(+), 26 deletions(-)
kernel/workqueue.c +28 -26
@@ -5118,6 +5118,19 @@ static struct pool_workqueue *alloc_unbound_pwq(struct workqueue_struct *wq,return pwq; } +static void apply_wqattrs_lock (void ) +{ +/* CPUs should stay stable across pwq creations and installations */ +cpus_read_lock (); +mutex_lock (&wq_pool_mutex); +} + +static void apply_wqattrs_unlock (void ) +{ +mutex_unlock (&wq_pool_mutex); +cpus_read_unlock (); +} +/** * wq_calc_pod_cpumask - calculate a wq_attrs' cpumask for a pod * @attrs: the wq_attrs of the default pwq of the target workqueue
@@ -5429,6 +5442,9 @@ static int alloc_and_link_pwqs(struct workqueue_struct *wq)bool highpri = wq->flags &WQ_HIGHPRI ;int cpu, ret; +lockdep_assert_cpus_held (); +lockdep_assert_held (&wq_pool_mutex); + wq->cpu_pwq = alloc_percpu(struct pool_workqueue *);if (!wq->cpu_pwq) goto enomem;
@@ -5461,20 +5477,18 @@ static int alloc_and_link_pwqs(struct workqueue_struct *wq)return 0 ; } -cpus_read_lock ();if (wq->flags & __WQ_ORDERED) {struct pool_workqueue *dfl_pwq; - ret =apply_workqueue_attrs (wq, ordered_wq_attrs[highpri]); + ret =apply_workqueue_attrs_locked (wq, ordered_wq_attrs[highpri]);/* there should only be single pwq for ordering guarantee */ dfl_pwq =rcu_access_pointer (wq->dfl_pwq);WARN (!ret && (wq->pwqs.next != &dfl_pwq->pwqs_node || wq->pwqs.prev != &dfl_pwq->pwqs_node),"ordering guarantee broken for workqueue %s\n" , wq->name); }else { - ret =apply_workqueue_attrs (wq, unbound_std_wq_attrs[highpri]); + ret =apply_workqueue_attrs_locked (wq, unbound_std_wq_attrs[highpri]); } -cpus_read_unlock ();return ret;
@@ -5681,15 +5695,15 @@ struct workqueue_struct *alloc_workqueue(const char *fmt, goto err_unreg_lockdep; } -if (alloc_and_link_pwqs (wq) <0 ) - goto err_free_node_nr_active; -/* -* wq_pool_mutex protects global freeze state and workqueues list. -* Grab it, adjust max_active and add the new @wq to workqueues -* list. +* wq_pool_mutex protects the workqueues list, allocations of PWQs, +* and the global freeze state. alloc_and_link_pwqs() also requires +* cpus_read_lock() for PWQs' affinities. */ -mutex_lock (&wq_pool_mutex); +apply_wqattrs_lock (); + +if (alloc_and_link_pwqs (wq) <0 ) + goto err_unlock_free_node_nr_active;mutex_lock (&wq->mutex);wq_adjust_max_active (wq);
@@ -5697,7 +5711,7 @@ struct workqueue_struct *alloc_workqueue(const char *fmt,list_add_tail_rcu (&wq->list, &workqueues); -mutex_unlock (&wq_pool_mutex); +apply_wqattrs_unlock ();if (wq_online &&init_rescuer (wq) <0 ) goto err_destroy;
@@ -5707,7 +5721,8 @@ struct workqueue_struct *alloc_workqueue(const char *fmt,return wq; -err_free_node_nr_active: +err_unlock_free_node_nr_active: +apply_wqattrs_unlock ();/* * Failed alloc_and_link_pwqs() may leave pending pwq->release_work, * flushing the pwq_release_worker ensures that the pwq_release_workfn()
@@ -6996,19 +7011,6 @@ static struct attribute *wq_sysfs_attrs[] = { };ATTRIBUTE_GROUPS (wq_sysfs); -static void apply_wqattrs_lock (void ) -{ -/* CPUs should stay stable across pwq creations and installations */ -cpus_read_lock (); -mutex_lock (&wq_pool_mutex); -} - -static void apply_wqattrs_unlock (void ) -{ -mutex_unlock (&wq_pool_mutex); -cpus_read_unlock (); -} -static ssize_t wq_nice_show (struct device *dev,struct device_attribute *attr,char *buf) {
Signature · 1 line
2.19.1.6.gb485710b
kernel test robot
build botRe: [PATCH V2 4/5] workqueue: Put PWQ allocation and WQ enlistment in the same lock C.S.
kernel test robot noticed "WARNING:possible_recursive_locking_detected" on:
Hello,
kernel test robot noticed "WARNING:possible_recursive_locking_detected" on:
commit: 1d4c6111406c8306a9b87ba6c496996cd4539cfd ("[PATCH V2 4/5] workqueue: Put PWQ allocation and WQ enlistment in the same lock C.S.")
url: https://github.com/intel-lab-lkp/linux/commits/Lai-Jiangshan/workqueue-Register-sysfs-after-the-whole-creation-of-the-new-wq/20240705-043238
base: v6.10-rc5
patch subject: [PATCH V2 4/5] workqueue: Put PWQ allocation and WQ enlistment in the same lock C.S.
in testcase: rcutorture
version:
with following parameters:
runtime: 300s test: default torture_type: srcu
compiler: gcc-13
test machine: qemu-system-x86_64 -enable-kvm -cpu SandyBridge -smp 2 -m 16G
(please refer to attached dmesg/kmsg for entire log/backtrace)
If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-lkp/[email protected]
[ 2.233457][ T1] WARNING: possible recursive locking detected
[ 2.233457][ T1] 6.10.0-rc5-00004-g1d4c6111406c #1 Not tainted
[ 2.233457][ T1] --------------------------------------------
[ 2.233457][ T1] swapper/0/1 is trying to acquire lock:
[ 2.233457][ T1] c27760f4 (cpu_hotplug_lock){++++}-{0:0}, at: alloc_workqueue (kernel/workqueue.c:5152 kernel/workqueue.c:5730)
[ 2.233457][ T1]
[ 2.233457][ T1] but task is already holding lock:
[ 2.233457][ T1] c27760f4 (cpu_hotplug_lock){++++}-{0:0}, at: padata_alloc (kernel/padata.c:1007)
[ 2.233457][ T1]
[ 2.233457][ T1] other info that might help us debug this:
[ 2.245549][ T1] Possible unsafe locking scenario:
[ 2.245549][ T1]
[ 2.245549][ T1] CPU0
[ 2.245549][ T1] ----
[ 2.245549][ T1] lock(cpu_hotplug_lock);
[ 2.245549][ T1]
[ 2.245549][ T1] *** DEADLOCK ***
[ 2.245549][ T1]
[ 2.251678][ T1] May be due to missing lock nesting notation
[ 2.251678][ T1]
[ 2.253463][ T1] 1 lock held by swapper/0/1:
[ 2.253463][ T1] #0: c27760f4 (cpu_hotplug_lock){++++}-{0:0}, at: padata_alloc (kernel/padata.c:1007)
[ 2.253463][ T1]
[ 2.253463][ T1] stack backtrace:
[ 2.257461][ T1] CPU: 1 PID: 1 Comm: swapper/0 Not tainted 6.10.0-rc5-00004-g1d4c6111406c #1 c89023213b7b89ade58aa28d4c172b811b00908c
[ 2.257461][ T1] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.2-debian-1.16.2-1 04/01/2014
[ 2.261462][ T1] Call Trace:
[ 2.261462][ T1] dump_stack_lvl (lib/dump_stack.c:117)
[ 2.261462][ T1] dump_stack (lib/dump_stack.c:124)
[ 2.261462][ T1] print_deadlock_bug (kernel/locking/lockdep.c:3013)
[ 2.265464][ T1] validate_chain (kernel/locking/lockdep.c:3859)
[ 2.265464][ T1] __lock_acquire (kernel/locking/lockdep.c:5137 (discriminator 1))
[ 2.265464][ T1] ? __this_cpu_preempt_check (lib/smp_processor_id.c:67)
[ 2.265464][ T1] lock_acquire (kernel/locking/lockdep.c:467 kernel/locking/lockdep.c:5756)
[ 2.265464][ T1] ? alloc_workqueue (kernel/workqueue.c:5152 kernel/workqueue.c:5730)
[ 2.269463][ T1] ? __might_sleep (kernel/sched/core.c:10126)
[ 2.269463][ T1] ? is_dynamic_key (include/linux/rcupdate.h:779 kernel/locking/lockdep.c:1257)
[ 2.269463][ T1] cpus_read_lock (include/linux/percpu-rwsem.h:53 kernel/cpu.c:488)
[ 2.269463][ T1] ? alloc_workqueue (kernel/workqueue.c:5152 kernel/workqueue.c:5730)
[ 2.273463][ T1] alloc_workqueue (kernel/workqueue.c:5152 kernel/workqueue.c:5730)
[ 2.273463][ T1] padata_alloc (kernel/padata.c:1007 (discriminator 1))
[ 2.273463][ T1] ? nhpoly1305_mod_init (crypto/pcrypt.c:345)
[ 2.273463][ T1] pcrypt_init_padata (crypto/pcrypt.c:327 (discriminator 1))
[ 2.277463][ T1] ? nhpoly1305_mod_init (crypto/pcrypt.c:345)
[ 2.277463][ T1] pcrypt_init (crypto/pcrypt.c:353)
[ 2.277463][ T1] ? nhpoly1305_mod_init (crypto/pcrypt.c:345)
[ 2.277463][ T1] do_one_initcall (init/main.c:1267)
[ 2.281464][ T1] ? next_arg (lib/cmdline.c:274)
[ 2.281464][ T1] do_initcalls (init/main.c:1328 (discriminator 1) init/main.c:1345 (discriminator 1))
[ 2.281464][ T1] ? rdinit_setup (init/main.c:1313)
[ 2.281464][ T1] kernel_init_freeable (init/main.c:1364)
[ 2.285464][ T1] ? kernel_init (init/main.c:1469)
[ 2.285464][ T1] ? rest_init (init/main.c:1459)
[ 2.285464][ T1] kernel_init (init/main.c:1469)
[ 2.285464][ T1] ? schedule_tail (kernel/sched/core.c:5346)
[ 2.285464][ T1] ret_from_fork (arch/x86/kernel/process.c:153)
[ 2.289466][ T1] ? rest_init (init/main.c:1459)
[ 2.289466][ T1] ret_from_fork_asm (arch/x86/entry/entry_32.S:737)
[ 2.289466][ T1] entry_INT80_32 (arch/x86/entry/entry_32.S:944) The kernel config and materials to reproduce are available at:
https://download.01.org/0day-ci/archive/20240708/[email protected]
Signature · 2 lines
0-DAY CI Kernel Test Service https://github.com/intel/lkp-tests/wiki
Tejun Heo
maintainerRe: [PATCH V2 4/5] workqueue: Put PWQ allocation and WQ enlistment in the same lock C.S.
Hello,
On Mon, Jul 08, 2024 at 03:54:19PM +0800, kernel test robot wrote:
...
Quoted 4 lines
[ 2.269463][ T1] cpus_read_lock (include/linux/percpu-rwsem.h:53 kernel/cpu.c:488) [ 2.273463][ T1] alloc_workqueue (kernel/workqueue.c:5152 kernel/workqueue.c:5730) [ 2.273463][ T1] padata_alloc (kernel/padata.c:1007 (discriminator 1)) [ 2.273463][ T1] pcrypt_init_padata (crypto/pcrypt.c:327 (discriminator 1))
Quoted 8 lines
[ 2.277463][ T1] pcrypt_init (crypto/pcrypt.c:353) [ 2.277463][ T1] do_one_initcall (init/main.c:1267) [ 2.281464][ T1] do_initcalls (init/main.c:1328 (discriminator 1) init/main.c:1345 (discriminator 1)) [ 2.281464][ T1] kernel_init_freeable (init/main.c:1364) [ 2.285464][ T1] kernel_init (init/main.c:1469) [ 2.285464][ T1] ret_from_fork (arch/x86/kernel/process.c:153) [ 2.289466][ T1] ret_from_fork_asm (arch/x86/entry/entry_32.S:737) [ 2.289466][ T1] entry_INT80_32 (arch/x86/entry/entry_32.S:944)
Ah, this is unfortunate, so pcrypt is 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. Lai, it looks like we'd need to switch to workqueue specific locking so that we don't depend on cpus_read_lock from alloc path.
Thanks.
Signature · 1 line
tejun
Lai Jiangshan
reviewerRe: [PATCH V2 4/5] workqueue: Put PWQ allocation and WQ enlistment in the same lock C.S.
Hello,
Quoted 17 lines · Tejun Heo
Hello,
On Mon, Jul 08, 2024 at 03:54:19PM +0800, kernel test robot wrote:
...Quoted 12 lines
[ 2.269463][ T1] cpus_read_lock (include/linux/percpu-rwsem.h:53 kernel/cpu.c:488) [ 2.273463][ T1] alloc_workqueue (kernel/workqueue.c:5152 kernel/workqueue.c:5730) [ 2.273463][ T1] padata_alloc (kernel/padata.c:1007 (discriminator 1)) [ 2.273463][ T1] pcrypt_init_padata (crypto/pcrypt.c:327 (discriminator 1)) [ 2.277463][ T1] pcrypt_init (crypto/pcrypt.c:353) [ 2.277463][ T1] do_one_initcall (init/main.c:1267) [ 2.281464][ T1] do_initcalls (init/main.c:1328 (discriminator 1) init/main.c:1345 (discriminator 1)) [ 2.281464][ T1] kernel_init_freeable (init/main.c:1364) [ 2.285464][ T1] kernel_init (init/main.c:1469) [ 2.285464][ T1] ret_from_fork (arch/x86/kernel/process.c:153) [ 2.289466][ T1] ret_from_fork_asm (arch/x86/entry/entry_32.S:737) [ 2.289466][ T1] entry_INT80_32 (arch/x86/entry/entry_32.S:944)
Thanks for the report!
Quoted 7 lines
Ah, this is unfortunate, so pcrypt is allocating a workqueue while holding
cpus_read_lock(), so workqueue code can't do it again as that can lead to deadlocks ifdown_writestarts after the firstdown_read. Lai, it looks like we'd need to switch to workqueue specific locking so that we don't depend oncpus_read_lockfrom alloc path.
I'm working on it. A new wq_online_mask will be added. The new wq_online_mask mirrors the cpu_online_mask except during hotplugging; specifically, it differs between the hotplugging stages of workqueue_offline_cpu() and workqueue_online_cpu(), during which the transitioning CPU is not represented in the mask.
Thanks
Lai
Lai Jiangshan
reviewerPATCH v2 5/5 workqueue: Init rescuer's affinities as the wq's effective cpumask
In reply to Lai JiangshanFrom: Lai Jiangshan <[email protected]>
Make it consistent with apply_wqattrs_commit().
--- kernel/workqueue.c | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-)
kernel/workqueue.c +8 -4
@@ -5525,6 +5525,8 @@ static int init_rescuer(struct workqueue_struct *wq)struct worker *rescuer;int ret; +lockdep_assert_held (&wq_pool_mutex); +if (!(wq->flags &WQ_MEM_RECLAIM ))return 0 ;
@@ -5547,7 +5549,7 @@ static int init_rescuer(struct workqueue_struct *wq) wq->rescuer = rescuer;if (wq->flags &WQ_UNBOUND ) -kthread_bind_mask (rescuer->task, wq_unbound_cpumask); +kthread_bind_mask (rescuer->task,unbound_effective_cpumask (wq));else kthread_bind_mask (rescuer->task, cpu_possible_mask);wake_up_process (rescuer->task);
@@ -5711,10 +5713,10 @@ struct workqueue_struct *alloc_workqueue(const char *fmt,list_add_tail_rcu (&wq->list, &workqueues); -apply_wqattrs_unlock (); -if (wq_online &&init_rescuer (wq) <0 ) - goto err_destroy; + goto err_unlock_destroy; + +apply_wqattrs_unlock ();if ((wq->flags &WQ_SYSFS ) &&workqueue_sysfs_register (wq)) goto err_destroy;
@@ -5739,6 +5741,8 @@ struct workqueue_struct *alloc_workqueue(const char *fmt,free_workqueue_attrs (wq->unbound_attrs);kfree (wq);return NULL ; +err_unlock_destroy: +apply_wqattrs_unlock (); err_destroy:destroy_workqueue (wq);return NULL ;
Signature · 1 line
2.19.1.6.gb485710b
Quoted 7 lines · Lai Jiangshan
From: Lai Jiangshan <[email protected]>
The PWQ allocation and WQ enlistment are not within the same lock-held critical section; therefore, their states can become out of sync when the user modifies the unbound mask or if CPU hotplug events occur in the interim since those operations only update the WQs that are already in the list.
Applied to wq/for-6.11.
Thanks.
Signature · 1 line
tejun