Re: [Qemu-devel] [PULL 0/3] Linux user for 4.1 patches

[Peter Maydell]

On Thu, 18 Jul 2019 at 11:40, Laurent Vivier <address@hidden> wrote:
> It comes from linux-user/syscall.c:
>
>  6328         /* automatic consistency check if same arch */
>  6329 #if (defined(__i386__) && defined(TARGET_I386) && 
> defined(TARGET_ABI32)) || \
>  6330     (defined(__x86_64__) && defined(TARGET_X86_64))
>  6331         if (unlikely(ie->target_cmd != ie->host_cmd)) {
>  6332             fprintf(stderr, "ERROR: ioctl(%s): target=0x%x host=0x%x\n",
>  6333                     ie->name, ie->target_cmd, ie->host_cmd);
>  6334         }
>  6335 #endif
>
> because of:
>
> +  { TARGET_SIOCGSTAMP_OLD, SIOCGSTAMP, "IOCGSTAMP_OLD", IOC_R, \
> +    do_ioctl_SIOCGSTAMP },
> +  { TARGET_SIOCGSTAMPNS_OLD, SIOCGSTAMPNS, "IOCGSTAMPNS_OLD", IOC_R, \
> +    do_ioctl_SIOCGSTAMPNS },
> +  { TARGET_SIOCGSTAMP_NEW, SIOCGSTAMP, "IOCGSTAMP_NEW", IOC_R, \
> +    do_ioctl_SIOCGSTAMP },
> +  { TARGET_SIOCGSTAMPNS_NEW, SIOCGSTAMPNS, "IOCGSTAMPNS_NEW", IOC_R, \
> +    do_ioctl_SIOCGSTAMPNS },
>
> As the host_cmd is not used, the simplest way to fix that is
>
> +  { TARGET_SIOCGSTAMP_OLD, TARGET_SIOCGSTAMP_OLD, "IOCGSTAMP_OLD", IOC_R, \
> +    do_ioctl_SIOCGSTAMP },
> +  { TARGET_SIOCGSTAMPNS_OLD, TARGET_SIOCGSTAMPNS_OLD, "IOCGSTAMPNS_OLD", 
> IOC_R, \
> +    do_ioctl_SIOCGSTAMPNS },
> +  { TARGET_SIOCGSTAMP_NEW, TARGET_SIOCGSTAMP_NEW, "IOCGSTAMP_NEW", IOC_R, \
> +    do_ioctl_SIOCGSTAMP },
> +  { TARGET_SIOCGSTAMPNS_NEW, TARGET_SIOCGSTAMPNS_NEW, "IOCGSTAMPNS_NEW", 
> IOC_R, \
> +    do_ioctl_SIOCGSTAMPNS },
>
> As SIOCGSTAMP_OLD and SIOCGSTAMP_NEW can be undefined on the host (and not 
> needed
> because we always use SIOCGSTAMP and SIOCGSTAMPNS)

So we don't use the host_cmd because we have a custom do_ioctl_foo
function which doesn't look at that field?

Sounds OK, but please include a comment explaining why.

PS: why didn't you use IOCTL_SPECIAL() rather than hand-written
array entries? None of the other ioctls.h entries do that.
Of course now we're trying to sidestep the consistency check
we can't use the macro, but it wolud have been fine otherwise.
It also would get the names of the ioctls in the string form
right -- they are all missing the initial "S" here.

Perhaps for 4.2 it might be worth considering having a
macro for "IOCTL_SPECIAL but skip the consistency check"
to be a bit less hacky here.

thanks
-- PMM