Strange behaviour of sizeof(struct v4l2_queryctrl)

* Strange behaviour of sizeof(struct v4l2_queryctrl)
@ 2015-01-27 10:43 Francesco Marletta
  2015-01-27 10:44 ` Francesco Marletta
  0 siblings, 1 reply; 3+ messages in thread
From: Francesco Marletta @ 2015-01-27 10:43 UTC (permalink / raw)
  To: linux-media

Hello to anyone,
I'm working on a problem with V4L2 on Linux Kernel 2.6.37.

The problem arise when I try to query a video device to list the 
controls it provides.

When is call
     ioctl(fd, VIDIOC_QUERYCTRL, &queryctrl)

the function doesn't return 0 and errno is set to EINVAL

This happen for every control, even for the controls that the driver 
provides (checked in the code) like brightness.

After adding a lot of printk in the videodev.ko module I found that the 
problem is caused by a wrong value of VIDIOC_QUERYCTRL, that in the 
kernel module is 0xC0485624 while in userspace application is 0xC0445624.

I digged the kernel source to understand what's happening, and 
discovered the definition of VIDIOC_QUERYCTRL:
         #define VIDIOC_QUERYCTRL        _IOWR('V', 36, struct 
v4l2_queryctrl)

dove:
         #define _IOWR(type,nr,size)     \
_IOC(_IOC_READ|_IOC_WRITE,(type),(nr),(_IOC_TYPECHECK(size)))

with
         #define _IOC(dir,type,nr,size) \
                         (((dir)  << _IOC_DIRSHIFT)  | \
                         ((type)  << _IOC_TYPESHIFT) | \
                         ((nr)    << _IOC_NRSHIFT)   | \
                         ((size)  << _IOC_SIZESHIFT))

and
         #define _IOC_NRSHIFT    0
         #define _IOC_TYPESHIFT  (_IOC_NRSHIFT+_IOC_NRBITS)
         #define _IOC_SIZESHIFT  (_IOC_TYPESHIFT+_IOC_TYPEBITS)
         #define _IOC_DIRSHIFT   (_IOC_SIZESHIFT+_IOC_SIZEBITS)

         #define _IOC_NRBITS     8
         #define _IOC_TYPEBITS   8
         #define _IOC_SIZEBITS   14

         #define _IOC_NONE       0U
         #define _IOC_WRITE      1U
         #define _IOC_READ       2U

         #define _IOC_TYPECHECK(t) (sizeof(t))

thus, the _IOC definition means:
         _IOC(dir,type,nr,size)   <=>   [dir|size|type|nr]

that, for the aftermentioned values means:
         0xC0445624 <=> [3|044|56|24]  ==> size = 0x44 = 68
         0xC0485624 <=> [3|048|56|24]  ==> size = 0x48 = 72

To be sure that the number 68 and 72 are correct, I added a printk() in 
the videodev.ko module and a println() in the userspace application to 
print "sizeof(struct v4l2_queryctrl)", and the outputs are:

     Kernel module:
         printk("[DBG] %s() ~ sizeof(struct v4l2_queryctrl): %u\n", 
__func__, sizeof(struct v4l2_queryctrl));
             => [DBG] videodev_init() ~ sizeof(struct v4l2_queryctrl): 72

     Userspace application:
         printf("[dbg] sizeof(struct v4l2_queryctrl): %u\n", 
sizeof(struct v4l2_queryctrl));
             => [dbg] sizeof(struct v4l2_queryctrl): 68

In both cases (module and application), there is the inclusion of 
<linux/videodev2.h>.

When I compiled the application I istructed the compiler to pick the 
same kernel header of the module, with a proper CFLAGS += -I<PATH> in 
the Makefile.

Any idea about this strangeness?

Regards
Francesco

^ permalink raw reply	[flat|nested] 3+ messages in thread