Always consider infcall breakpoints as non-permanent.

Always consider infcall breakpoints as non-permanent.

A recent change...

    commit 1a853c5224e2b8fedfac6d029365522b83080b40
    Date:   Wed Nov 12 10:10:49 2014 +0000
    Subject: make "permanent breakpoints" per location and disableable

... broke function calls on sparc-elf when running over QEMU. Any
function call should demonstrate the problem.

For instance, seen from the debugger:

    (gdb) call pn(1234)
    [Inferior 1 (Remote target) exited normally]
    The program being debugged exited while in a function called from GDB.
    Evaluation of the expression containing the function

And seen from QEMU:

    qemu: fatal: Trap 0x02 while interrupts disabled, Error state
    [register dump removed]

What happens in this case is that GDB sets the inferior function call
by not only creating the dummy frame, but also writing a breakpoint
instruction at the return address for our function call. See infcall.c:

        /* Write a legitimate instruction at the point where the infcall
           breakpoint is going to be inserted.  While this instruction
           is never going to be executed, a user investigating the
           memory from GDB would see this instruction instead of random
           uninitialized bytes.  We chose the breakpoint instruction
           as it may look as the most logical one to the user and also
           valgrind 3.7.0 needs it for proper vgdb inferior calls.

           If software breakpoints are unsupported for this target we
           leave the user visible memory content uninitialized.  */

        bp_addr_as_address = bp_addr;
        bp_bytes = gdbarch_breakpoint_from_pc (gdbarch, &bp_addr_as_address,
                                               &bp_size);
        if (bp_bytes != NULL)
          write_memory (bp_addr_as_address, bp_bytes, bp_size);

This instruction triggers a change introduced by the commit above,
where we consider bp locations as being permanent breakpoints
if there is already a breakpoint instruction at that address:

        +  if (bp_loc_is_permanent (loc))
        +    {
        +      loc->inserted = 1;
        +      loc->permanent = 1;
        +    }

As a result, when resuming the program's execution for the inferior
function call, GDB decides that it does not need to insert a breakpoint
at this address, expecting the target to just report a SIGTRAP when
trying to execute that instruction.

But unfortunately for us, at least some versions of QEMU for SPARC
just terminate the execution entirely instead of reporting a breakpoint,
thus producing the behavior reported here.

Although it appears like QEMU might be misbehaving and should therefore
be fixed (to be verified) from the user's point of view, the recent
change does introduce a regression. So this patch tries to mitigate
a bit the damage by handling such infcall breakpoints as special and
making sure that they are never considered permanent, thus restoring
the previous behavior specifically for those breakpoints.

The option of not writing the breakpoint instructions in the first
place was considered, and would probably work also. But the comment
associated to it seems to indicate that there is still reason to
keep it.

gdb/ChangeLog:

        * breakpoint.c (bp_loc_is_permanent): Return 0 if LOC corresponds
        to a bp_call_dummy breakpoint type.

Tested on x86_64-linux. Also testing on sparc-elf/QEMU using
AdaCore's testsuite.