2006-09-22 09:36:52

by Cornelia Huck

[permalink] [raw]
Subject: [2/9] driver core fixes: device_register() retval check in platform.c

From: Cornelia Huck <[email protected]>

Check the return value of device_register() in platform_bus_init().

Signed-off-by: Cornelia Huck <[email protected]>

---
drivers/base/platform.c | 11 +++++++++--
1 file changed, 9 insertions(+), 2 deletions(-)

--- linux-2.6-CH.orig/drivers/base/platform.c
+++ linux-2.6-CH/drivers/base/platform.c
@@ -563,8 +563,15 @@ EXPORT_SYMBOL_GPL(platform_bus_type);

int __init platform_bus_init(void)
{
- device_register(&platform_bus);
- return bus_register(&platform_bus_type);
+ int error;
+
+ error = device_register(&platform_bus);
+ if (error)
+ return error;
+ error = bus_register(&platform_bus_type);
+ if (error)
+ device_unregister(&platform_bus);
+ return error;
}

#ifndef ARCH_HAS_DMA_GET_REQUIRED_MASK


2006-09-23 21:10:57

by Russell King

[permalink] [raw]
Subject: Re: [2/9] driver core fixes: device_register() retval check in platform.c

On Fri, Sep 22, 2006 at 11:36:55AM +0200, Cornelia Huck wrote:
> From: Cornelia Huck <[email protected]>
>
> Check the return value of device_register() in platform_bus_init().
>
> Signed-off-by: Cornelia Huck <[email protected]>
>
> ---
> drivers/base/platform.c | 11 +++++++++--
> 1 file changed, 9 insertions(+), 2 deletions(-)
>
> --- linux-2.6-CH.orig/drivers/base/platform.c
> +++ linux-2.6-CH/drivers/base/platform.c
> @@ -563,8 +563,15 @@ EXPORT_SYMBOL_GPL(platform_bus_type);
>
> int __init platform_bus_init(void)
> {
> - device_register(&platform_bus);
> - return bus_register(&platform_bus_type);
> + int error;
> +
> + error = device_register(&platform_bus);
> + if (error)
> + return error;
> + error = bus_register(&platform_bus_type);
> + if (error)
> + device_unregister(&platform_bus);
> + return error;

I don't think there's much value in patches such as this - if the
platform bus type didn't register, what happens when we then try
to register a platform device driver or a platform device? ISTR
doing that before the bus type is registered leads to an OOPS.

So, presumably to do this properly, if the platform_bus_type failed
to register, you need to force all platform device/platform device
driver registrations to also fail.

At that point, is the added complexity really worth it?

--
Russell King
Linux kernel 2.6 ARM Linux - http://www.arm.linux.org.uk/
maintainer of: 2.6 Serial core

2006-09-25 14:58:06

by Cornelia Huck

[permalink] [raw]
Subject: Re: [2/9] driver core fixes: device_register() retval check in platform.c

On Sat, 23 Sep 2006 22:10:32 +0100,
Russell King <[email protected]> wrote:

> I don't think there's much value in patches such as this - if the
> platform bus type didn't register, what happens when we then try
> to register a platform device driver or a platform device? ISTR
> doing that before the bus type is registered leads to an OOPS.

Yes, since the klists have not yet been initialized.

> So, presumably to do this properly, if the platform_bus_type failed
> to register, you need to force all platform device/platform device
> driver registrations to also fail.

We should fail registration (gracefully) of all devices/drivers which
specify a bus that is !NULL but has not been registered. Unfortunately,
I don't see an easy way to do this.

However, we can fail the registration of devices that specify a parent
that is !NULL but not yet added to the tree. This catches platform
devices registering before platform_bus_type (since platform_bus is not
registered then), similar for other bus types like iucv.
(Unfortunately, not drivers...)


From: Cornelia Huck <[email protected]>

Force parent devices to be registered before their children. Otherwise we'll
oops when creating the child's sysfs directory.

Signed-off-by: Cornelia Huck <[email protected]>

---
drivers/base/bus.c | 6 ++++--
drivers/base/core.c | 5 ++++-
2 files changed, 8 insertions(+), 3 deletions(-)

--- linux-2.6.18-mm1.orig/drivers/base/core.c
+++ linux-2.6.18-mm1/drivers/base/core.c
@@ -474,7 +474,10 @@ int device_add(struct device *dev)
}

parent = get_device(dev->parent);
-
+ if (parent && !device_is_registered(parent)) {
+ error = -EINVAL;
+ goto Error;
+ }
pr_debug("DEV: registering device: ID = '%s'\n", dev->bus_id);

/* first, register with generic layer. */
--- linux-2.6.18-mm1.orig/drivers/base/bus.c
+++ linux-2.6.18-mm1/drivers/base/bus.c
@@ -418,7 +418,8 @@ int bus_attach_device(struct device * de
ret = 0;
} else
dev->is_registered = 0;
- }
+ } else
+ dev->is_registered = 1;
return ret;
}

@@ -443,7 +444,8 @@ void bus_remove_device(struct device * d
pr_debug("bus %s: remove device %s\n", dev->bus->name, dev->bus_id);
device_release_driver(dev);
put_bus(dev->bus);
- }
+ } else
+ dev->is_registered = 0;
}

static int driver_add_attrs(struct bus_type * bus, struct device_driver * drv)