Re: RTC_RD_TIME

[Alexander Monakov <amonakov@ispras.ru>]

On Tue, 29 Aug 2017, Rich Felker wrote:
> I don't know any good fix. The problem is that the glibc (and others')
> ioctl function has a signature mismatching the standard one, with an
> unsigned argument instead of signed, and the macro values are outside
> the range of signed int. Fortunately the warning is harmless but
> obviously it can impact builds with -Werror or policy about warnings..

Type confusion causes musl to wrongly pass sign-extended value to the
kernel (instead of zero-extended). That's a real issue, and it's a bit
disappointing you have not mentioned it.

(BSD flavours as well as Linux have 'unsigned long' there, so I don't
understand why it was ok for POSIX to document request type as 'int';
wouldn't it be appropriate for musl to just follow existing practice
here rather than the letter of the standard?)

Without changing the prototype, I think a possible fix is

#ifdef __GNUC__
#define ioctl(fd, req, ...) ioctl(fd, (int)(req), ##__VA_ARGS__)
#endif

(plus a fix in ioctl.c for the sign extension issue)

although I don't understand how the quoted warning appears; I can
produce another one with -Wsign-conversion, but not this.

Alexander