On 12/27/2013 01:31 PM, Viresh Kumar wrote:
On 27 December 2013 08:36, Jane Li <jiel@xxxxxxxxxxx> wrote:
When gov_queue_work(), governor_enabled may be modified. Following patch
can fix it by adding cpufreq_governor_lock in gov_queue_work. But in this
way, cpufreq_governor_lock also protects __gov_queue_work(). Do you think
this is a good idea?
gov_queue_work() is called with timer_mutex on. Can you try placing this
lock around gov_cancel_work() and see if it resolves issues you have reported ?
--
viresh
There may be deadlock in this way.
The existing dependency chain (in reverse order) is:
-> #1 (&j_cdbs->timer_mutex){+.+.+.}:
[<c017ba0c>] lock_acquire+0x9c/0x150
[<c069d6d0>] mutex_lock_nested+0x50/0x3d8
[<c04849c0>] od_dbs_timer+0x3c/0x130
[<c0143b84>] process_one_work+0x1b8/0x518
[<c0144024>] worker_thread+0x140/0x3f0
[<c014a708>] kthread+0xa4/0xb0
[<c010e2c8>] ret_from_fork+0x14/0x2c
-> #0 ((&(&j_cdbs->work)->work)){+.+...}:
[<c017af18>] __lock_acquire+0x171c/0x1c64
[<c017ba0c>] lock_acquire+0x9c/0x150
[<c01432e8>] flush_work+0x3c/0x2a0
[<c0144b90>] __cancel_work_timer+0x90/0x138
[<c0485e90>] cpufreq_governor_dbs+0x528/0x6a4
[<c0481c40>] __cpufreq_governor+0x80/0x1b0
[<c0481dd8>] __cpufreq_remove_dev.isra.12+0x68/0x380
[<c0696eac>] cpufreq_cpu_callback+0x7c/0x90
[<c014fff8>] notifier_call_chain+0x44/0x84
[<c012b09c>] __cpu_notify+0x2c/0x48
[<c0691c98>] _cpu_down+0x80/0x258
[<c0691e98>] cpu_down+0x28/0x3c
[<c0692418>] store_online+0x30/0x74
[<c03a4400>] dev_attr_store+0x18/0x24
[<c0257498>] sysfs_write_file+0x100/0x180
[<c01ff154>] vfs_write+0xbc/0x184
[<c01ff4ec>] SyS_write+0x40/0x68
[<c010e200>] ret_fast_syscall+0x0/0x48
CPU0 CPU1
---- ----
lock(&j_cdbs->timer_mutex);
lock((&(&j_cdbs->work)->work));
lock(&j_cdbs->timer_mutex);
lock((&(&j_cdbs->work)->work));
diff --git a/drivers/cpufreq/cpufreq_governor.c b/drivers/cpufreq/cpufreq_governor.c
index e6be635..ff14647 100644
--- a/drivers/cpufreq/cpufreq_governor.c
+++ b/drivers/cpufreq/cpufreq_governor.c
@@ -351,7 +351,9 @@ int cpufreq_governor_dbs(struct cpufreq_policy *policy,
if (dbs_data->cdata->governor == GOV_CONSERVATIVE)
cs_dbs_info->enable = 0;
+ mutex_lock(&cpu_cdbs->timer_mutex);
gov_cancel_work(dbs_data, policy);
+ mutex_unlock(&cpu_cdbs->timer_mutex);
mutex_lock(&dbs_data->mutex);
mutex_destroy(&cpu_cdbs->timer_mutex);
--
To unsubscribe from this list: send the line "unsubscribe cpufreq" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
