atomic_sub_return() negates a long

https://github.com/riscv/opensbi/blob/0f20e8adcf42d31bc478d6700b625d00a68cb30e/lib/sbi/riscv_atomic.c#L46

Hello, I'm reviewing opensbi as part of Ubuntu's Main Inclusion Review process. It's just a quick look, not a full audit.

I noticed that atomic_sub_return() uses -value as part of its processing.

There's more negative long numbers than positive long numbers, on most platforms; on my amd64 Ubuntu 20.04 system:

LONG_MAX 9223372036854775807
LONG_MIN -9223372036854775808
LONG_MAX 7fffffffffffffff
LONG_MIN 8000000000000000

Trying to flip LONG_MIN with unary - leaves it unchanged:

$ cat negativelong.c
#include <stdio.h>
#include <limits.h>

int main(int argc, char* argv[]) {
	long i = LONG_MIN;
	printf("%ld negated: %ld\n", i, -i);
	return 0;
}

$ make ./negativelong
make: 'negativelong' is up to date.
$ ./negativelong 
-9223372036854775808 negated: -9223372036854775808

I honestly haven't got a clue if there is an actual problem here or not; it just seemed worth a few seconds to double-check.

Thanks

Start at lib/sbi/riscv_atomic.c line 46 and inspect how atomic_sub_return() handles the negated value, including the LONG_MIN case described in the issue. Reproduce or reason through the behavior on the supported platform and establish whether it causes an incorrect result; done means the concern is resolved with an appropriate code or test change if a defect is confirmed.