>>>>> Mike Klar writes:
> Hi all,
> I've run across a problem with user apps that use the _syscall[0-7] macros
> in asm-mips/unistd.h. With the latest Linux-MIPS glibc release (the
> 2.0.6-5lm version), they're not getting errno set properly for error
> conditions. In particular, this breaks recent versions of modutils.
> The problem is this: In user space, errno is now a per-thread value. This
> seems like a good thing to me, but it means that the assignment of errno is
> no longer a simple assignment, but now involves a function call to retrieve
> a pointer to errno. This trips up the _syscall macros in unistd.h because
> they all do:
> errno = __res;
> where __res is declared to be the v0 register, which is set by the syscall
> itself. Meanwhile, the function that retrieves the pointer to errno returns
> it in... v0. So errno winds up getting set to the address of errno. This
> doesn't affect _syscall usage in kernel code, because errno is a simple
> assignment in that case, nor does it affect glibc itself, since glibc
> doesn't seem to use the _syscall macros.
> I worked around this problem by adding to each _syscall macro:
> unsigned long __res_save; \
> and changing:
> errno = __res; \
> __res_save = __res; \
> errno = __res_save; \
> I'm not sure that's the best/cleanest solution, but it does have the benefit
> that when errno is a simple assignment (as is the case for the kernel), the
> compiler optimizes out the additional assignment.
The question remains what we should do with glibc 2.2. Currently
<sys/syscalls.h> includes <asm/unistd.h> and this makes _syscall0 and
friends available to userspace.
I couldn't find any reference to <sys/syscalls.h> in the ABI and
consider dropping the include of <asm/unistd.h> since it's not needed
Any objections or better suggestions?
SuSE Labs email@example.com