Skip to contentSkip to conversation

cpufreq: fix locking order in store_local_boost to prevent deadlock

PATCH Superseded Seyediman Seyedarab 2 participants 3 messages Apr 28, 2025 – Apr 29, 2025 CPU frequency scaling framework

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 cpufreq: fix locking order in store_local_boost to prevent deadlock

Lockdep reports a possible circular locking dependency[1] when writing to /sys/devices/system/cpu/cpufreq/policyN/boost, triggered by power-profiles-daemon at boot.

store_local_boost() acquires cpu_hotplug_lock *AFTER* policy->rwsem has already been taken by the store() handler. However, the expected locking hierarchy is to acquire cpu_hotplug_lock before policy->rwsem. This inverted lock order creates a *theoretical* deadlock possibility.

Acquire cpu_hotplug_lock in the store() handler *only* for the local_boost, before taking down_write(&policy->rwsem), and remove the cpus_read_lock/unlock() calls inside the store_local_boost().

[1]
======================================================
WARNING: possible circular locking dependency detected 6.15.0-rc3-debug #28 Not tainted
------------------------------------------------------
power-profiles-/596 is trying to acquire lock:
ffffffffb147e910 (cpu_hotplug_lock){++++}-{0:0}, at: store_local_boost+0x6a/0xd0

but task is already holding lock:
ffff9eaa48377b80 (&policy->rwsem){++++}-{4:4}, at: store+0x37/0x90

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #2 (&policy->rwsem){++++}-{4:4}:

        down_write+0x29/0xb0
        cpufreq_online+0x841/0xa00
        cpufreq_add_dev+0x71/0x80
        subsys_interface_register+0x14b/0x170
        cpufreq_register_driver+0x154/0x250
        amd_pstate_register_driver+0x36/0x70
        amd_pstate_init+0x1e7/0x270
        do_one_initcall+0x67/0x2c0
        kernel_init_freeable+0x230/0x270
        kernel_init+0x15/0x130
        ret_from_fork+0x2c/0x50
        ret_from_fork_asm+0x11/0x20

-> #1 (subsys mutex#3){+.+.}-{4:4}:

        __mutex_lock+0xc2/0x930
        subsys_interface_register+0x83/0x170
        cpufreq_register_driver+0x154/0x250
        amd_pstate_register_driver+0x36/0x70
        amd_pstate_init+0x1e7/0x270
        do_one_initcall+0x67/0x2c0
        kernel_init_freeable+0x230/0x270
        kernel_init+0x15/0x130
        ret_from_fork+0x2c/0x50
        ret_from_fork_asm+0x11/0x20

-> #0 (cpu_hotplug_lock){++++}-{0:0}:

        __lock_acquire+0x1087/0x17e0
        lock_acquire.part.0+0x66/0x1b0
        cpus_read_lock+0x2a/0xc0
        store_local_boost+0x6a/0xd0
        store+0x50/0x90
        kernfs_fop_write_iter+0x135/0x200
        vfs_write+0x2ab/0x540
        ksys_write+0x6c/0xe0
        do_syscall_64+0xbb/0x1d0
        entry_SYSCALL_64_after_hwframe+0x56/0x5e

---
Changes in v2:
- Restrict cpu_hotplug_lock acquisition to only
the local_boost attribute in store() handler.

Regards,
Seyediman

 drivers/cpufreq/cpufreq.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
drivers/cpufreq/cpufreq.c +6 -3
index acf19b004..5464c8487 100644
@@ -653,10 +653,7 @@ static ssize_t store_local_boost(struct cpufreq_policy *policy,
 
 	policy->boost_enabled = enable;
 
-	cpus_read_lock();
 	ret = cpufreq_driver->set_boost(policy, enable);
-	cpus_read_unlock();
-
 	if (ret) {
 		policy->boost_enabled = !policy->boost_enabled;
 		return ret;
@@ -1045,11 +1042,17 @@ static ssize_t store(struct kobject *kobj, struct attribute *attr,
 	if (!fattr->store)
 		return -EIO;
 
+	if (fattr == &local_boost)
+		cpus_read_lock();
+
 	down_write(&policy->rwsem);
 	if (likely(!policy_is_inactive(policy)))
 		ret = fattr->store(policy, buf, count);
 	up_write(&policy->rwsem);
 
+	if (fattr == &local_boost)
+		cpus_read_unlock();
+
 	return ret;
 }
 
Signature · 1 line
2.49.0
maintainer
#1 lore raw
Quoted 16 lines · Seyediman Seyedarab
drivers/cpufreq/cpufreq.c +0 -3
index acf19b004..5464c8487 100644
@@ -653,10 +653,7 @@ static ssize_t store_local_boost(struct cpufreq_policy *policy,
 
 	policy->boost_enabled = enable;
 
-	cpus_read_lock();
 	ret = cpufreq_driver->set_boost(policy, enable);
-	cpus_read_unlock();
-
 	if (ret) {
 		policy->boost_enabled = !policy->boost_enabled;
 		return ret;

drivers/cpufreq/cpufreq.c static ssize_t store(struct kobject *kobj, struct attribute *attr, · line 1047

Quoted 6 lines
drivers/cpufreq/cpufreq.c +3 -0
@@ -1045,11 +1042,17 @@ static ssize_t store(struct kobject *kobj, struct attribute *attr,
 	if (!fattr->store)
 		return -EIO;
 
+	if (fattr == &local_boost)
+		cpus_read_lock();
+

Please add a comment as well to explain why this is required here.

drivers/cpufreq/cpufreq.c static ssize_t store(struct kobject *kobj, struct attribute *attr, · line 1048

Quoted 1 line
drivers/cpufreq/cpufreq.c +0 -0
 	down_write(&policy->rwsem);

And please rebase over PM tree's linux-next branch, your patch won't apply anymore due to recent changes.

Quoted 9 lines
drivers/cpufreq/cpufreq.c +3 -0
 	if (likely(!policy_is_inactive(policy)))
 		ret = fattr->store(policy, buf, count);
 	up_write(&policy->rwsem);
 
+	if (fattr == &local_boost)
+		cpus_read_unlock();
+
 	return ret;
 }
Signature · 1 line
viresh
#2 lore raw
Quoted 9 lines · Viresh Kumar
Quoted 8 lines · Seyediman Seyedarab
drivers/cpufreq/cpufreq.c +0 -0
index acf19b004..5464c8487 100644
@@ -1045,11 +1042,17 @@ static ssize_t store(struct kobject *kobj, struct attribute *attr,
 	if (!fattr->store)
 		return -EIO;
 
Quoted 7 lines

drivers/cpufreq/cpufreq.c static ssize_t store(struct kobject *kobj, struct attribute *attr, · line 1048

Quoted 4 lines
drivers/cpufreq/cpufreq.c +3 -0
+	if (fattr == &local_boost)
+		cpus_read_lock();
+
 	down_write(&policy->rwsem);

And please rebase over PM tree's linux-next branch, your patch won't apply anymore due to recent changes.

Since the new code uses the guard() macro to lock the policy, I see two possible ways to address this:
1. Switch to scoped_guard() and manually wrap the local boost case in an if block, calling cpus_read_lock() and cpus_read_unlock() around the scoped_guard(cpufreq_policy_write, policy) block.

2. Use DEFINE_GUARD() to create a cpu_hotplug_read guard, but this requires introducing an empty struct to satisfy the macro. Feels semantically awkward, since there's no actual object being guarded.

Option 1 looks more reasonable to me, because it's more direct and avoids misusing the guard pattern for something that's not resource-based.

Let me know if you'd prefer the second approach for consistency.

Regards,
Seyediman