| Message ID | 1375981237-7167-1-git-send-email-mikedunn@newsguy.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Mike Dunn <mikedunn@newsguy.com> writes: > Commit 76abbdde2d95a3807d0dc6bf9f84d03d0dbd4f3d > > pwm: Add sysfs interface > > causes a kernel oops due to a null pointer dereference on pxa platforms. This > happens because the class added by the patch is registered in a subsys_initcall > (initcall4), but the pxa pwm driver is registered in arch_initcall (initcall3). > If the class is not registered before the driver probe function runs, the oops > occurs in device_add() when the uninitialized pointers in struct class are > dereferenced. I don't see a reason that the driver must be an arch_initcall, so > this patch makes it a regular module_platform_driver (initcall6), preventing the > oops. That makes sense. The only users I'm aware of for PWM on pxa is backlight, and that can certainly wait for regular driver initcall level. I'm not aware of any usage in early boot stages. Acked-by: Robert Jarzmik <robert.jarzmik@free.fr>
Dear Mike Dunn, > Commit 76abbdde2d95a3807d0dc6bf9f84d03d0dbd4f3d > > pwm: Add sysfs interface > > causes a kernel oops due to a null pointer dereference on pxa platforms. > This happens because the class added by the patch is registered in a > subsys_initcall (initcall4), but the pxa pwm driver is registered in > arch_initcall (initcall3). If the class is not registered before the > driver probe function runs, the oops occurs in device_add() when the > uninitialized pointers in struct class are dereferenced. I don't see a > reason that the driver must be an arch_initcall, so this patch makes it a > regular module_platform_driver (initcall6), preventing the oops. > > Signed-off-by: Mike Dunn <mikedunn@newsguy.com> > --- > drivers/pwm/pwm-pxa.c | 12 +----------- > 1 file changed, 1 insertion(+), 11 deletions(-) > > diff --git a/drivers/pwm/pwm-pxa.c b/drivers/pwm/pwm-pxa.c > index dc97175..a4d2164 100644 > --- a/drivers/pwm/pwm-pxa.c > +++ b/drivers/pwm/pwm-pxa.c > @@ -182,16 +182,6 @@ static struct platform_driver pwm_driver = { > .id_table = pwm_id_table, > }; > > -static int __init pwm_init(void) > -{ > - return platform_driver_register(&pwm_driver); > -} > -arch_initcall(pwm_init); > - > -static void __exit pwm_exit(void) > -{ > - platform_driver_unregister(&pwm_driver); > -} > -module_exit(pwm_exit); > +module_platform_driver(pwm_driver); > > MODULE_LICENSE("GPL v2"); Acked-by: Marek Vasut <marex@denx.de> Best regards, Marek Vasut
diff --git a/drivers/pwm/pwm-pxa.c b/drivers/pwm/pwm-pxa.c index dc97175..a4d2164 100644 --- a/drivers/pwm/pwm-pxa.c +++ b/drivers/pwm/pwm-pxa.c @@ -182,16 +182,6 @@ static struct platform_driver pwm_driver = { .id_table = pwm_id_table, }; -static int __init pwm_init(void) -{ - return platform_driver_register(&pwm_driver); -} -arch_initcall(pwm_init); - -static void __exit pwm_exit(void) -{ - platform_driver_unregister(&pwm_driver); -} -module_exit(pwm_exit); +module_platform_driver(pwm_driver); MODULE_LICENSE("GPL v2");
Commit 76abbdde2d95a3807d0dc6bf9f84d03d0dbd4f3d pwm: Add sysfs interface causes a kernel oops due to a null pointer dereference on pxa platforms. This happens because the class added by the patch is registered in a subsys_initcall (initcall4), but the pxa pwm driver is registered in arch_initcall (initcall3). If the class is not registered before the driver probe function runs, the oops occurs in device_add() when the uninitialized pointers in struct class are dereferenced. I don't see a reason that the driver must be an arch_initcall, so this patch makes it a regular module_platform_driver (initcall6), preventing the oops. Signed-off-by: Mike Dunn <mikedunn@newsguy.com> --- drivers/pwm/pwm-pxa.c | 12 +----------- 1 file changed, 1 insertion(+), 11 deletions(-)