Comment 6 for bug 1920784

[10] outlined to use PPC_FEATURE2_SCV but [4] does just that.
In addition [6] added power9 machine settings as only on this ISA it
is available - like:
+ .machine "push"
+ .machine "power9"
        scv 0
+ .machine "pop"

Maybe there is some generated "scv 0" left that needs the same [6] treatment?

OTOH In a normal test program I can run "scv 0" just fine.
But not other scv levels (expected).

# cat test.c
#include <stdio.h>

int main() {
   printf("Hello scv 0\n");
   __asm__(
   "scv 0\n\t"
   );
   printf("survived\n");
   __asm__(
   "scv 1\n\t"
   );
   printf("survived level 1\n");
   return 0;
}
# gcc -Wall -o test test.c
./test
Hello scv 0
survived
Illegal instruction (core dumped)

IIRC .machine is only a psedo-op for the assembler.
So it is correct that I can't see it in the live disassembly of gdb.

The failing "svc 0" from glibcs __GI___ioctl is
   0x00007ffff66c49a0 <+320>: 01 00 00 44 scv 0
And in my test program it is
   0x0000000100000848 <+44>: 01 00 00 44 scv 0

Hmm, this is the same opcode but fails in just one of the cases.
This might need someone being more an ppc64/glibc expert than me :-/

@Frank - could you modify this bug to become mirrored to IBM for their arch-expertise please?