[Bug compile/34551] New: Summary: compile command: external calls from compiled modules crash on powerpc64le

https://sourceware.org/bugzilla/show_bug.cgi?id=34551

            Bug ID: 34551
           Summary: Summary:  compile command: external calls from
                    compiled modules crash on powerpc64le
           Product: gdb
           Version: HEAD
            Status: NEW
          Severity: normal
          Priority: P2
         Component: compile
          Assignee: unassigned at sourceware dot org
          Reporter: abhay at sourceware dot org
  Target Milestone: ---

On powerpc64le-linux, code injected by the "compile" command crashes when
it calls a function in another module.

The module is loaded into inferior memory without a linker, so it has no
PLT.  On ELFv2 the default code model still emits a direct bl for such a
call, which leaves r12 unset and the caller's TOC pointer unrestored, so
the callee derives a wrong r2 from a stale r12 and faults on its first
TOC-relative load.

gdb.compile/compile-setjmp.exp reproduces this:

  FAIL: gdb.compile/compile-setjmp.exp: compile file -r
  FAIL: gdb.compile/compile-setjmp.exp: p done

  Program received signal SIGSEGV, Segmentation fault.
  0x00007ffff7a47d64 in __sigsetjmp@@GLIBC_2.17 () from
    /lib64/glibc-hwcaps/power10/libc.so.6

At the fault, r12 still held the address of the injected _gdb_expr rather
than that of the callee, and r2 pointed past the end of libc.

This affects any external call from a compiled module; compile-setjmp is
simply the only test in gdb.compile that makes one.

Fix posted to gdb-patches: add -mlongcall -mno-pltseq to the PowerPC
compile options so GCC resolves the call against the module's own TOC.
<URL to the thread>

ELFv1 (powerpc64-linux) fails too, but differently — the symbol resolves
to a function descriptor and the branch lands in .opd.  There are also
other unrelated failures in gdb.compile there.  Not covered by this
report.

-- 
You are receiving this mail because:
You are on the CC list for the bug.