| Message ID | 1440592108-3740-17-git-send-email-holler@ahsoftware.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi, On Wednesday 26 August 2015 05:58 PM, Alexander Holler wrote: > The phy-core has to be initialized before other dependent usb-drivers, > otherwise a crash might occur. > > Currently phy_core_init() is called in the initcall-level device, which is > the same level where most usb-drivers will end up. By luck this seemed to > have been called most of the time before other usb-drivers without having > been explicitly enforced. But if phy_core_init() is not called before a > dependent driver, a null-pointer exception might occur (e.g. because the > phy device class isn't registered). Did you actually face a problem? IIUC the modules get loaded based on the drivers/Makefile order (unless the other modules are in a different initcall table). IMHO the fix should be in the module that caused the crash. Change it to use module_init? Thanks Kishon > > To fix this, phy_core_init() is moved to the initcall-level fs (right > before the standard initcall level device). > > Signed-off-by: Alexander Holler <holler@ahsoftware.de> > --- > drivers/phy/phy-core.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/drivers/phy/phy-core.c b/drivers/phy/phy-core.c > index fc48fac..4945029 100644 > --- a/drivers/phy/phy-core.c > +++ b/drivers/phy/phy-core.c > @@ -930,7 +930,7 @@ static int __init phy_core_init(void) > > return 0; > } > -module_init(phy_core_init); > +fs_initcall_sync(phy_core_init); > > static void __exit phy_core_exit(void) > { >
Am 18.09.2015 um 08:16 schrieb Kishon Vijay Abraham I: > Hi, > > On Wednesday 26 August 2015 05:58 PM, Alexander Holler wrote: >> The phy-core has to be initialized before other dependent usb-drivers, >> otherwise a crash might occur. >> >> Currently phy_core_init() is called in the initcall-level device, which is >> the same level where most usb-drivers will end up. By luck this seemed to >> have been called most of the time before other usb-drivers without having >> been explicitly enforced. But if phy_core_init() is not called before a >> dependent driver, a null-pointer exception might occur (e.g. because the >> phy device class isn't registered). > > Did you actually face a problem? IIUC the modules get loaded based on > the drivers/Makefile order (unless the other modules are in a different > initcall table). I had a problem while playing with a modified init-system (based on dependencies). So not an actual problem. > IMHO the fix should be in the module that caused the crash. Change it to > use module_init? The problem arises if the init-system ignores the link order and assumes all drivers in the same initcall level can be called without any special ordering. The problem might also appear if a driver changes its name, directory or position in file system. E.g. how to you make sure that a driver in staging will be linked after the phy-core? Actually this happens, but I would assume its by luck. I assume if staging would be renamed to 'beta-quality' a lot of stuff would actually fail because of the problem with the implicit link order. Anyway, nothing which really has to be fixed. It's just a notice that maybe another initcall level of 'subsys' or something else before 'device' might be a better place for phy-core. I've chosen fs_sync instead of subsys because otherwise I would have had to look up if phy-core depends on another subsystem and therefore has to be initialized after subsys. Regards, Alexander Holler > > Thanks > Kishon > >> >> To fix this, phy_core_init() is moved to the initcall-level fs (right >> before the standard initcall level device). >> >> Signed-off-by: Alexander Holler <holler@ahsoftware.de> >> --- >> drivers/phy/phy-core.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/drivers/phy/phy-core.c b/drivers/phy/phy-core.c >> index fc48fac..4945029 100644 >> --- a/drivers/phy/phy-core.c >> +++ b/drivers/phy/phy-core.c >> @@ -930,7 +930,7 @@ static int __init phy_core_init(void) >> >> return 0; >> } >> -module_init(phy_core_init); >> +fs_initcall_sync(phy_core_init); >> >> static void __exit phy_core_exit(void) >> { >> > -- > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html > Please read the FAQ at http://www.tux.org/lkml/ >
diff --git a/drivers/phy/phy-core.c b/drivers/phy/phy-core.c index fc48fac..4945029 100644 --- a/drivers/phy/phy-core.c +++ b/drivers/phy/phy-core.c @@ -930,7 +930,7 @@ static int __init phy_core_init(void) return 0; } -module_init(phy_core_init); +fs_initcall_sync(phy_core_init); static void __exit phy_core_exit(void) {
The phy-core has to be initialized before other dependent usb-drivers, otherwise a crash might occur. Currently phy_core_init() is called in the initcall-level device, which is the same level where most usb-drivers will end up. By luck this seemed to have been called most of the time before other usb-drivers without having been explicitly enforced. But if phy_core_init() is not called before a dependent driver, a null-pointer exception might occur (e.g. because the phy device class isn't registered). To fix this, phy_core_init() is moved to the initcall-level fs (right before the standard initcall level device). Signed-off-by: Alexander Holler <holler@ahsoftware.de> --- drivers/phy/phy-core.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)