[PATCH] i2c_dw: deadlock happening when system is trying to suspend

[PATCH] i2c_dw: deadlock happening when system is trying to suspend

[Liu, Chuansheng]

From: liu chuansheng <chuansheng.liu@intel.com>
Subject: [PATCH] i2c_dw: deadlock happening when system is trying to suspend

In i2c_dw code, there is a race condition that causes pm suspend
thread blocking there always. The scenerio is as below:

PM thread:
suspend -->
pm_suspend -->
enter_state -->
dpm_suspend_start(will call i2c_dw_pci_suspend(),
and the dw_i2c_dev->lock is hold)
...
suspend_enter -->
dpm_suspend_noirq -->
suspend_device_irqs -->
synchronize_irq()

synchronize_irq will wait for any pending irq is handled, and
the correpsonding irq thread is finished.

In this case, there is a i2c device interrupt is pending, the irq
thread do the below things:
IRQ thread:
i2c_smbus_read_byte_data -->
i2c_smbus_xfer -->
i2c_transfer -->
i2c_dw_xfer -->
down()

The irq thread blocked at down dw_i2c_dev->lock, because in PM thread,
it has been hold after calling i2c_dw_pci_suspend(), but PM thread is
waiting for IRQ thread, then deadlock happened.

The solution is moving the down() action after pm_runtime_get_sync().

Signed-off-by: liu chuansheng <chuansheng.liu@intel.com>
---
 drivers/i2c/busses/i2c-designware-core.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/i2c/busses/i2c-designware-core.c b/drivers/i2c/busses/i2c-designware-core.c
index 1e48bec..748ecb1 100644
--- a/drivers/i2c/busses/i2c-designware-core.c
+++ b/drivers/i2c/busses/i2c-designware-core.c
@@ -512,8 +512,8 @@ i2c_dw_xfer(struct i2c_adapter *adap, struct i2c_msg msgs[], int num)
 
        dev_dbg(dev->dev, "%s: msgs: %d\n", __func__, num);
 
-       mutex_lock(&dev->lock);
        pm_runtime_get_sync(dev->dev);
+       mutex_lock(&dev->lock);
 
        INIT_COMPLETION(dev->cmd_complete);
        dev->msgs = msgs;
-- 
1.7.0.4

Re: [PATCH] i2c_dw: deadlock happening when system is trying to suspend

[Wolfram Sang]

[-- Attachment #1: Type: text/plain, Size: 2297 bytes --]

On Tue, Jun 26, 2012 at 02:29:13PM +0000, Liu, Chuansheng wrote:
> From: liu chuansheng <chuansheng.liu-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> Subject: [PATCH] i2c_dw: deadlock happening when system is trying to suspend
> 
> In i2c_dw code, there is a race condition that causes pm suspend
> thread blocking there always. The scenerio is as below:
> 
> PM thread:
> suspend -->
> pm_suspend -->
> enter_state -->
> dpm_suspend_start(will call i2c_dw_pci_suspend(),
> and the dw_i2c_dev->lock is hold)
> ...
> suspend_enter -->
> dpm_suspend_noirq -->
> suspend_device_irqs -->
> synchronize_irq()
> 
> synchronize_irq will wait for any pending irq is handled, and
> the correpsonding irq thread is finished.
> 
> In this case, there is a i2c device interrupt is pending, the irq
> thread do the below things:
> IRQ thread:
> i2c_smbus_read_byte_data -->
> i2c_smbus_xfer -->
> i2c_transfer -->
> i2c_dw_xfer -->
> down()
> 
> The irq thread blocked at down dw_i2c_dev->lock, because in PM thread,
> it has been hold after calling i2c_dw_pci_suspend(), but PM thread is
> waiting for IRQ thread, then deadlock happened.
> 
> The solution is moving the down() action after pm_runtime_get_sync().
> 
> Signed-off-by: liu chuansheng <chuansheng.liu-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> ---
>  drivers/i2c/busses/i2c-designware-core.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
> 
> diff --git a/drivers/i2c/busses/i2c-designware-core.c b/drivers/i2c/busses/i2c-designware-core.c
> index 1e48bec..748ecb1 100644
> --- a/drivers/i2c/busses/i2c-designware-core.c
> +++ b/drivers/i2c/busses/i2c-designware-core.c
> @@ -512,8 +512,8 @@ i2c_dw_xfer(struct i2c_adapter *adap, struct i2c_msg msgs[], int num)
>  
>         dev_dbg(dev->dev, "%s: msgs: %d\n", __func__, num);
>  
> -       mutex_lock(&dev->lock);
>         pm_runtime_get_sync(dev->dev);
> +       mutex_lock(&dev->lock);
>  
>         INIT_COMPLETION(dev->cmd_complete);
>         dev->msgs = msgs;

Don't you need to place the mutex_unlock() before pm_runtime_put then?

Thanks,

   Wolfram

-- 
Pengutronix e.K.                           | Wolfram Sang                |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

RE: [PATCH] i2c_dw: deadlock happening when system is trying to suspend

[Liu, Chuansheng]

Sorry, I found the latest kernel code has some difference with my analysis.
I need more thoughts, then update this issue.

> -----Original Message-----
> From: Wolfram Sang [mailto:w.sang-bIcnvbaLZ9MEGnE8C9+IrQ@public.gmane.org]
> Sent: Monday, July 09, 2012 9:32 PM
> To: Liu, Chuansheng
> Cc: linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; linux-i2c-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; khali-PUYAD+kWke1g9hUCZPvPmw@public.gmane.org;
> ben-linux-elnMNo+KYs3YtjvyW6yDsg@public.gmane.org; Yanmin Zhang <yanmin_zhang-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>
> (yanmin_zhang-VuQAYsv1563Yd54FQh9/CA@public.gmane.org); Srivatsa S. Bhat; Tu, Xiaobing
> Subject: Re: [PATCH] i2c_dw: deadlock happening when system is trying to
> suspend
> 
> On Tue, Jun 26, 2012 at 02:29:13PM +0000, Liu, Chuansheng wrote:
> > From: liu chuansheng <chuansheng.liu-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> > Subject: [PATCH] i2c_dw: deadlock happening when system is trying to
> > suspend
> >
> > In i2c_dw code, there is a race condition that causes pm suspend
> > thread blocking there always. The scenerio is as below:
> >
> > PM thread:
> > suspend -->
> > pm_suspend -->
> > enter_state -->
> > dpm_suspend_start(will call i2c_dw_pci_suspend(), and the
> > dw_i2c_dev->lock is hold) ...
> > suspend_enter -->
> > dpm_suspend_noirq -->
> > suspend_device_irqs -->
> > synchronize_irq()
> >
> > synchronize_irq will wait for any pending irq is handled, and the
> > correpsonding irq thread is finished.
> >
> > In this case, there is a i2c device interrupt is pending, the irq
> > thread do the below things:
> > IRQ thread:
> > i2c_smbus_read_byte_data -->
> > i2c_smbus_xfer -->
> > i2c_transfer -->
> > i2c_dw_xfer -->
> > down()
> >
> > The irq thread blocked at down dw_i2c_dev->lock, because in PM thread,
> > it has been hold after calling i2c_dw_pci_suspend(), but PM thread is
> > waiting for IRQ thread, then deadlock happened.
> >
> > The solution is moving the down() action after pm_runtime_get_sync().
> >
> > Signed-off-by: liu chuansheng <chuansheng.liu-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> > ---
> >  drivers/i2c/busses/i2c-designware-core.c |    2 +-
> >  1 files changed, 1 insertions(+), 1 deletions(-)
> >
> > diff --git a/drivers/i2c/busses/i2c-designware-core.c
> > b/drivers/i2c/busses/i2c-designware-core.c
> > index 1e48bec..748ecb1 100644
> > --- a/drivers/i2c/busses/i2c-designware-core.c
> > +++ b/drivers/i2c/busses/i2c-designware-core.c
> > @@ -512,8 +512,8 @@ i2c_dw_xfer(struct i2c_adapter *adap, struct
> > i2c_msg msgs[], int num)
> >
> >         dev_dbg(dev->dev, "%s: msgs: %d\n", __func__, num);
> >
> > -       mutex_lock(&dev->lock);
> >         pm_runtime_get_sync(dev->dev);
> > +       mutex_lock(&dev->lock);
> >
> >         INIT_COMPLETION(dev->cmd_complete);
> >         dev->msgs = msgs;
> 
> Don't you need to place the mutex_unlock() before pm_runtime_put then?
> 
> Thanks,
> 
>    Wolfram
> 
> --
> Pengutronix e.K.                           | Wolfram Sang
> |
> Industrial Linux Solutions                 | http://www.pengutronix.de/
> |