| Message ID | 1459275349-11356-1-git-send-email-linux@roeck-us.net (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Tue, Mar 29, 2016 at 11:15:49AM -0700, Guenter Roeck wrote: > The sa1100 gpio driver was initialized from interrupt initialization code, > which is earlier than the gpio subsystem is initialized. Since commit > ff2b13592299 ("gpio: make the gpiochip a real device"), this is fatal > and causes the system to crash. > > The sa1100 gpio driver must be initialized prior to arch_initcall, since > its gpio pins are used in arch_initcall code, but after the gpio subsystem > is initialized (core_initcall). Initialize it with postcore_initcall. > > Fixes: ff2b13592299 ("gpio: make the gpiochip a real device") > Cc: Linus Walleij <linus.walleij@linaro.org> > Signed-off-by: Guenter Roeck <linux@roeck-us.net> Please ignore/drop this patch - a better (cleaner) fix is in the works for gpiolib. Thanks, Guenter > --- > > Tested by building collie_defconfig and: > > qemu-system-arm -M collie -kernel arch/arm/boot/zImage \ > -no-reboot -initrd busybox-armv4.cpio \ > --append "rdinit=/sbin/init console=ttySA1" -monitor null -nographic > > arch/arm/mach-sa1100/generic.c | 11 +++++++++-- > 1 file changed, 9 insertions(+), 2 deletions(-) > > diff --git a/arch/arm/mach-sa1100/generic.c b/arch/arm/mach-sa1100/generic.c > index 345e63f4eb71..850aa7f8f649 100644 > --- a/arch/arm/mach-sa1100/generic.c > +++ b/arch/arm/mach-sa1100/generic.c > @@ -321,6 +321,15 @@ static int __init sa1100_init(void) > > arch_initcall(sa1100_init); > > +static int __init sa1100_gpio_init(void) > +{ > + sa1100_init_gpio(); > + > + return 0; > +} > + > +postcore_initcall(sa1100_gpio_init); > + > void __init sa11x0_init_late(void) > { > sa11x0_pm_init(); > @@ -386,8 +395,6 @@ void __init sa1100_init_irq(void) > request_resource(&iomem_resource, &irq_resource); > > sa11x0_init_irq_nodt(IRQ_GPIO0_SC, irq_resource.start); > - > - sa1100_init_gpio(); > } > > /* > -- > 2.5.0 >
On Thu, Mar 31, 2016 at 6:21 PM, Guenter Roeck <linux@roeck-us.net> wrote: > On Tue, Mar 29, 2016 at 11:15:49AM -0700, Guenter Roeck wrote: >> The sa1100 gpio driver was initialized from interrupt initialization code, >> which is earlier than the gpio subsystem is initialized. Since commit >> ff2b13592299 ("gpio: make the gpiochip a real device"), this is fatal >> and causes the system to crash. >> >> The sa1100 gpio driver must be initialized prior to arch_initcall, since >> its gpio pins are used in arch_initcall code, but after the gpio subsystem >> is initialized (core_initcall). Initialize it with postcore_initcall. >> >> Fixes: ff2b13592299 ("gpio: make the gpiochip a real device") >> Cc: Linus Walleij <linus.walleij@linaro.org> >> Signed-off-by: Guenter Roeck <linux@roeck-us.net> > > Please ignore/drop this patch - a better (cleaner) fix is in the works > for gpiolib. This delivers though. I tested it on my also regressing iPAQ. Yours, Linus Walleij
On Fri, Apr 01, 2016 at 10:04:57AM +0200, Linus Walleij wrote: > On Thu, Mar 31, 2016 at 6:21 PM, Guenter Roeck <linux@roeck-us.net> wrote: > > On Tue, Mar 29, 2016 at 11:15:49AM -0700, Guenter Roeck wrote: > >> The sa1100 gpio driver was initialized from interrupt initialization code, > >> which is earlier than the gpio subsystem is initialized. Since commit > >> ff2b13592299 ("gpio: make the gpiochip a real device"), this is fatal > >> and causes the system to crash. > >> > >> The sa1100 gpio driver must be initialized prior to arch_initcall, since > >> its gpio pins are used in arch_initcall code, but after the gpio subsystem > >> is initialized (core_initcall). Initialize it with postcore_initcall. > >> > >> Fixes: ff2b13592299 ("gpio: make the gpiochip a real device") > >> Cc: Linus Walleij <linus.walleij@linaro.org> > >> Signed-off-by: Guenter Roeck <linux@roeck-us.net> > > > > Please ignore/drop this patch - a better (cleaner) fix is in the works > > for gpiolib. > > This delivers though. I tested it on my also regressing iPAQ. > I am a bit concerned that the gpio initialization was that early on purpose, and that by moving it we might miss some use cases. I did not find any, but that doesn't mean that there are none. Without knowing _why_ the initialization was that early, I would prefer not to touch the code if it can be avoided. Thanks, Guenter
On Fri, Apr 1, 2016 at 4:02 PM, Guenter Roeck <linux@roeck-us.net> wrote: > On Fri, Apr 01, 2016 at 10:04:57AM +0200, Linus Walleij wrote: >> > Please ignore/drop this patch - a better (cleaner) fix is in the works >> > for gpiolib. >> >> This delivers though. I tested it on my also regressing iPAQ. >> > I am a bit concerned that the gpio initialization was that early on purpose, > and that by moving it we might miss some use cases. I did not find any, but that > doesn't mean that there are none. Without knowing _why_ the initialization was > that early, I would prefer not to touch the code if it can be avoided. I think we'll go for the other patch to gpiolib. It seems much more clean and generic to support this kind of cases. Yours, Linus Walleij
diff --git a/arch/arm/mach-sa1100/generic.c b/arch/arm/mach-sa1100/generic.c index 345e63f4eb71..850aa7f8f649 100644 --- a/arch/arm/mach-sa1100/generic.c +++ b/arch/arm/mach-sa1100/generic.c @@ -321,6 +321,15 @@ static int __init sa1100_init(void) arch_initcall(sa1100_init); +static int __init sa1100_gpio_init(void) +{ + sa1100_init_gpio(); + + return 0; +} + +postcore_initcall(sa1100_gpio_init); + void __init sa11x0_init_late(void) { sa11x0_pm_init(); @@ -386,8 +395,6 @@ void __init sa1100_init_irq(void) request_resource(&iomem_resource, &irq_resource); sa11x0_init_irq_nodt(IRQ_GPIO0_SC, irq_resource.start); - - sa1100_init_gpio(); } /*
The sa1100 gpio driver was initialized from interrupt initialization code, which is earlier than the gpio subsystem is initialized. Since commit ff2b13592299 ("gpio: make the gpiochip a real device"), this is fatal and causes the system to crash. The sa1100 gpio driver must be initialized prior to arch_initcall, since its gpio pins are used in arch_initcall code, but after the gpio subsystem is initialized (core_initcall). Initialize it with postcore_initcall. Fixes: ff2b13592299 ("gpio: make the gpiochip a real device") Cc: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Guenter Roeck <linux@roeck-us.net> --- Tested by building collie_defconfig and: qemu-system-arm -M collie -kernel arch/arm/boot/zImage \ -no-reboot -initrd busybox-armv4.cpio \ --append "rdinit=/sbin/init console=ttySA1" -monitor null -nographic arch/arm/mach-sa1100/generic.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-)