Le 07/09/2016 16:45, PrasannaKumar Muralidharan a écrit :

On 7 September 2016 at 19:53, Romain Perier
[off-list ref] wrote:

quoted

Hello,


Le 06/09/2016 18:31, PrasannaKumar Muralidharan a écrit :

quoted

quoted

Use devm_hwrng_register instead of hwrng_register. It avoids the need
to handle unregistration explicitly from the remove function.

Signed-off-by: Romain Perier <redacted>
---
   drivers/char/hw_random/omap-rng.c | 4 +---
   1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/drivers/char/hw_random/omap-rng.c

b/drivers/char/hw_random/omap-rng.c
index d47b24d..171c3e8 100644

--- a/drivers/char/hw_random/omap-rng.c
+++ b/drivers/char/hw_random/omap-rng.c

@@ -381,7 +381,7 @@ static int omap_rng_probe(struct platform_device

*pdev)
          if (ret)
                  goto err_ioremap;

-       ret = hwrng_register(&omap_rng_ops);
+       ret = devm_hwrng_register(dev, &omap_rng_ops);
          if (ret)
                  goto err_register;

@@ -402,8 +402,6 @@ static int omap_rng_remove(struct platform_device

*pdev)
   {
          struct omap_rng_dev *priv = platform_get_drvdata(pdev);

-       hwrng_unregister(&omap_rng_ops);
-
          priv->pdata->cleanup(priv);

          pm_runtime_put_sync(&pdev->dev);
--

If devm_hwrng_register is used hwrng_unregister will be called after
pm_runtime_disable is called. If RNG device is in use calling
omap_rng_remove may not work properly.

The case where the remove function is called is if you unbind the driver by
hand or you call rmmod while the RNG device is used.
I don't think that the kernel will call platform->remove is the device is in
use (so /dev/hwrng). I mean the argument that the unregister function is
called after pm_runtime_disable is correct, but I don't think that the
remove function might be called while the device is in use. There is
necessarily a mutual exclusive case between "use the device" and "call the
remove function of the device". However, I am open to suggestions.

The way you explained is good :D. Good point too. But the device is
created by hw_random core (hwrng_modinit in core.c) so the device can
be in use when omap-rng module is removed. Please feel free to correct
me if I am wrong.

Cheers,
PrasannaKumar

Hi,

I was wondering something. hwrng_unregister does not check the kref 
reference counter of the object... so technically if the current 
rng_device is in use, with or without devm... calling platform->remove 
will break the driver anyway because hwrng_unregister will unbind the 
device from /dev/hwrng. What I mean is that I think that we had this 
issue even without devm_hwrng_register.

Herbert, could you confirm ?

Romain
-- 
Romain Perier, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com