[PATCH 4/5] Update linux/ioctlent.h.in file

[Gabriel Laskar]

On Fri, Jan 16, 2015 at 12:39 AM, Dmitry V. Levin <ldv at altlinux.org> wrote:
> On Thu, Jan 15, 2015 at 12:03:20PM -0800, enh wrote:
>> is there anything we can do about the SND* mess? the most common
>> conflict i see on Android is the same one i see on my desktop:
>>
>> ioctl(1, SNDCTL_TMR_TIMEBASE or SNDRV_TIMER_IOCTL_NEXT_DEVICE or
>> TCGETS, {B38400 opost isig icanon echo ...}) = 0
>>
>> Android does use _some_ of the SND* ioctls, so i don't want to just
>> hack them all out of our strace.
>
> strace uses only 16 bits (8-bit number and 8-bit type) of the 32-bit ioctl
> command, so the only real solution to this issue is to take into account
> remaining 16 bits (size and direction).
>
> For example,
> SNDCTL_TMR_TIMEBASE is _IOC(_IOC_READ|_IOC_WRITE, 'T', 1, sizeof(int)),
> SNDRV_TIMER_IOCTL_NEXT_DEVICE is _IOC(_IOC_READ|_IOC_WRITE, 'T', 1, sizeof(struct snd_timer_id)),
> TCGETS is architecture-specific, its generic value 0x5401 is _IOC(0, 'T', 1, 0).
>
> As you can see all these 32-bit values are different.
> The only question is how to evaluate these remaining bits reliably, especially
> 14 bits (or 13 on alpha, mips, powerpc, and sparc) of ioctl command size.

Or we could do that, yes. I don't know why I have occulted this
solution. I have missed the 14/13 bit snafu though.

Maybe we can create 2 ioctlent.h.in files, 1 for the 14 bits size, and
another one for the 13 bits one and compile the right one for each
arch.


-- 
Gabriel Laskar