https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87482
Bug ID: 87482
Summary: Clarify behaviour of resolvers with parameters in for
__attribute__((ifunc))
Product: gcc
Version: 8.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: peter.smith at linaro dot org
Target Milestone: ---
In the documentation for
https://gcc.gnu.org/onlinedocs/gcc-8.2.0/gcc/Common-Function-Attributes.html#Common-Function-Attributes
it says: "The resolver should be declared to be a function taking no arguments
and returning a pointer to a function of the same type as the implementation."
I would like to check if the documentation is correct here as the
implementation accepts a parameter and on some architectures, such as Arm and
AArch64, the hwcaps is passed in to the resolver as a parameter by the dynamic
loader? To me it looks like the documentation isn't quite right, but I could be
missing something important.
For example a quick modification of the example on the page:
#include <stdlib.h>
void *my_memcpy (void *dst, const void *src, size_t len)
{
return dst;
}
void *my_memcpy2 (void *dst, const void *src, size_t len)
{
return dst;
}
static void * (*resolve_memcpy (int hwcap))(void *, const void *, size_t)
{
if (hwcap == 1)
return my_memcpy;
else
return my_memcpy2;
}
void *memcpy(void *, const void*, size_t) __attribute__((ifunc
("resolve_memcpy")));
Compiles without a problem.
From what I can see glibc doesn't use __attribute__((ifunc("resolver")))
instead it uses __asm__ (".type " resolver, %gnu_indirect_function).
For context clang will give an error message if the ifunc resolver has a
parameter.
ifunc2.c:20:58: error: ifunc resolver function must have no parameters
void *memcpy(void *, const void*, size_t) __attribute__((ifunc ("resolve...
There is a thread on cfe-dev asking if the GCC documentation is correct and to
see if clang should be following the documentation or the implementation:
http://lists.llvm.org/pipermail/cfe-dev/2018-September/059548.html
There was a recent change for 8.0 that made "The resolver should be declared to
be a function taking no arguments" more explicit.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81882
