On Thu, Feb 07, 2013 at 06:19:35PM -0800, H. Peter Anvin wrote:
> On 02/07/2013 03:45 PM, kbuild test robot wrote:
> > tree: git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git x86/mm
> > head: 6fcddf474ae1c8e2fb5f14c850c8aa018e7a5034
> > commit: 6fcddf474ae1c8e2fb5f14c850c8aa018e7a5034 x86-32: Add support for
> > 64bit get_user()
> > date: 2 hours ago
> > config: make ARCH=i386 allmodconfig
> >
> > All warnings:
> >
> > sound/pci/asihpi/hpioctl.c: In function 'asihpi_hpi_ioctl':
> >>> sound/pci/asihpi/hpioctl.c:125:6: warning: cast to pointer from integer
> >>> of different size [-Wint-to-pointer-cast]
> >>> sound/pci/asihpi/hpioctl.c:126:6: warning: cast to pointer from integer
> >>> of different size [-Wint-to-pointer-cast]
> > --
> > net/atm/resources.c: In function 'atm_dev_ioctl':
> >>> net/atm/resources.c:223:8: warning: cast to pointer from integer of
> >>> different size [-Wint-to-pointer-cast]
> >>> net/atm/resources.c:274:7: warning: cast to pointer from integer of
> >>> different size [-Wint-to-pointer-cast]
> > --
>
> [... and so on ...]
>
> Now I remember this problem. I believe I discussed it with rmk in the
> context of ARM, too. Russell, did we ever find any way to implement
> 8-byte get_user() on 32 bits without massive warning spewage?
My last attempt for ARM that I can find (though it wasn't my last email)
was:
#define __get_user_x(__r2,__p,__e,__s,__i...) \
__asm__ __volatile__ ( \
"bl __get_user_" #__s \
: "=&r" (__e), "=r" (__r2) \
: "0" (__p) \
: __i, "cc")
#ifdef BIG_ENDIAN
#define __get_user_xb(__r2,__p,__e,__s,__i...) \
__get_user_x(__r2,(uintptr_t)__p + 4,__s,__i)
#else
#define __get_user_xb __get_user_x
#endif
#define get_user(x,p) \
({ \
register const typeof(*(p)) __user *__p asm("r0") = (p);\
register int __e asm("r0"); \
register typeof(x) __r2 asm("r2"); \
switch (sizeof(*(__p))) { \
case 1: \
__get_user_x(__r2, __p, __e, 1, "lr"); \
break; \
case 2: \
__get_user_x(__r2, __p, __e, 2, "r3", "lr"); \
break; \
case 4: \
__get_user_x(__r2, __p, __e, 4, "lr"); \
break; \
case 8: \
if (sizeof((x)) < 8) \
__get_user_xb(__r2, __p, __e, 4, "lr"); \
else \
__get_user_x(__r2, __p, __e, 8, "lr"); \
break; \
default: __e = __get_user_bad(); break; \
} \
x = (typeof(*(__p))) __r2; \
__e; \
})
which I claimed did work for ARM, but I doubted it'd work everywhere
because it relies upon there being no 8-bit or 16-bit "registers" being
available - iow, it replies upon a register for byte and half-word
variables being a full register where the value gets sign-extended to
the full register.
This is important because of the "typeof(x) __r2 asm("r2")" which may
end up being a char, short or int-sized, but __get_user_x() effectively
writing to the whole word.
Consider the case where 'x' and '*p' are two different sizes.
Whether that's safe for x86 or not, I don't know, but my suspicions are
that it's unsafe on x86 as it's possible to refer to the various bytes/
half-words of eax separately.
So, I came to the conclusion that if x86 remains a problem, there's
little point supporting it on ARM.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
