On 31. 07. 25, 22:58, Nicolas Pitre wrote:
On Thu, 31 Jul 2025, Christophe Leroy wrote:
Le 31/07/2025 à 16:35, Christophe Leroy a écrit :
Hi Jiri,
Le 11/06/2025 à 12:02, Jiri Slaby (SUSE) a écrit :
_IO*() is the proper way of defining ioctl numbers. All these vt numbers
were synthetically built up the same way the _IO() macro does.
So instead of implicit hex numbers, use _IO() properly.
To not change the pre-existing numbers, use only _IO() (and not _IOR()
or _IOW()). The latter would change the numbers indeed.
On powerpc your assumption is wrong, because _IOC_NONE is not 0:
$ git grep _IOC_NONE arch/powerpc/
arch/powerpc/include/uapi/asm/ioctl.h:#define _IOC_NONE 1U
Therefore the value changes even with _IO(), leading to failure of Xorg as
reported by Christian.
And is likely an issue on the 4 following architectures:
$ git grep _IOC_NONE arch/ | grep 1U
arch/alpha/include/uapi/asm/ioctl.h:#define _IOC_NONE 1U
arch/mips/include/uapi/asm/ioctl.h:#define _IOC_NONE 1U
arch/powerpc/include/uapi/asm/ioctl.h:#define _IOC_NONE 1U
arch/sparc/include/uapi/asm/ioctl.h:#define _IOC_NONE 1U
IMHO this one patch could simply be reverted and the "old" code let be.
Oh, right -- it's easy to revert (no conflicts).
We could use _IOC(0, 'V', number, 0) directly, but I am not sure, that's
worth it.
thanks,
--
js
suse labs
