Device Drivers Which Include mkdir("/dev")

[Gedare Bloom]

On Thu, Feb 18, 2021 at 2:18 PM Joel Sherrill <joel at rtems.org> wrote:
>
>
>
> On Thu, Feb 18, 2021 at 2:08 PM Gedare Bloom <gedare at rtems.org> wrote:
>>
>> On Thu, Feb 18, 2021 at 12:52 PM Joel Sherrill <joel at rtems.org> wrote:
>> >
>> >
>> >
>> > On Thu, Feb 18, 2021 at 11:52 AM Gedare Bloom <gedare at rtems.org> wrote:
>> >>
>> >> On Thu, Feb 18, 2021 at 10:20 AM Joel Sherrill <joel at rtems.org> wrote:
>> >> >
>> >> >
>> >> >
>> >> > On Thu, Feb 18, 2021 at 11:06 AM Gedare Bloom <gedare at rtems.org> wrote:
>> >> >>
>> >> >> On Thu, Feb 18, 2021 at 8:56 AM Joel Sherrill <joel at rtems.org> wrote:
>> >> >> >
>> >> >> > Hi
>> >> >> >
>> >> >> > There are a lot of Coverity issues related to device drivers which call mkdir("/dev") and ignore the return value. My first thought was that they should have (void) added since /dev could have been created by an earlier driver.
>> >> >> >
>> >> >> > Then it occurred to me that libcsupport/src/base_fs.c includes mkdir("/dev") and a fatal error if it cannot create it.
>> >> >> >
>> >> >> > Doesn't this mean that every call to mkdir("/dev") in a BSP or device driver is redundant? They should be removed since the base FS contents are always in place before any device drivers are called.
>> >> >> >
>> >> >> > Thoughts?
>> >> >> >
>> >> >>
>> >> >> Seems reasonable. You could probably add an assert that /dev exists.
>> >> >
>> >> >
>> >> > The creation of the device node will fail if /dev does not exist so I think that
>> >> > is covered.
>> >> >
>> >> > I think the call is pointless and changing it to stat() to make sure it exists
>> >> > just adds bulk to the driver/BSP dependencies.
>> >> >
>> >> > I think just deleting it seems prudent. There is an error path if it doesn't
>> >> > exist.
>> >> >
>> >>
>> >> ok. I was just suggesting conversion to an assert since that is some
>> >> good practice when you have a belief/assumption but aren't totally
>> >> convinced. It's fine with me either way. assert(mkdir("/dev") == -1)
>> >> or just delete it.
>> >
>> >
>> > Grrr.. I've looked again at the code and it is all Gaisler code doing something
>> > like mkdir("/dev/leonXXX"). It really could fail. This should be a fatal error
>> > and would seem to indicate that we need a grlib category of fatal BSP/driver
>> > errors.
>>
>> Given the complexity of grlib, that would make some sense. Although
>> maybe we can make it a little more generic... FATAL_THIRD_PARTY_DRIVER
>> errors might be nicer on them, and can be reused
>
>
> Good suggestion. But I think this particular error is potentially even more generic.
> Any device driver wanting to make a subdirectory /dev/XXX could generate this.
> This seems like the right place to add this but I'm open for better names:
>
> diff --git a/bsps/include/bsp/fatal.h b/bsps/include/bsp/fatal.h
> index ec5902755e..0d8507e398 100644
> --- a/bsps/include/bsp/fatal.h
> +++ b/bsps/include/bsp/fatal.h
> @@ -41,6 +41,7 @@ typedef enum {
>    BSP_FATAL_CONSOLE_INSTALL_0,
>    BSP_FATAL_CONSOLE_INSTALL_1,
>    BSP_FATAL_CONSOLE_REGISTER_DEV_2,
> +  BSP_FATAL_MAKE_SUBDIR_OF_DEV,
>
Please try to come up with a name that is consistent with the other
patterns. If it can be any driver, I guess
BSP_FATAL_DEVFS_INVALID_NAME
or something?

>    /* ARM fatal codes */
>    BSP_ARM_A9MPCORE_FATAL_CLOCK_IRQ_INSTALL = BSP_FATAL_CODE_BLOCK(1),
>
>>
>>
>> >>
>> >>
>> >> > --joel
>> >> >>
>> >> >>
>> >> >> > --joel
>> >> >> > _______________________________________________
>> >> >> > devel mailing list
>> >> >> > devel at rtems.org
>> >> >> > http://lists.rtems.org/mailman/listinfo/devel