[Bug target/103085] New: [12 Regression] -fPIC and -fstack-protector-strong broken AArch64

[Bug target/103085] New: [12 Regression] -fPIC and -fstack-protector-strong broken AArch64

[tnfchris at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103085

            Bug ID: 103085
           Summary: [12 Regression] -fPIC and -fstack-protector-strong
                    broken AArch64
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tnfchris at gcc dot gnu.org
                CC: wilco at gcc dot gnu.org
  Target Milestone: ---
            Target: aarch64-*

-fPIC and -fstack-protector-strong are broken since
g:a195c7270e4705a06a9aaade0d721d5f52ca2275

The following testcase

---

struct A {
  int operator*();
  void operator++();
  friend bool operator!=(A, A);
};
class B {
public:
  A begin();
  A end();
};
class C {
  void emitMaps();
  B Events;
};
void C::emitMaps() {
  for (auto a : Events)
    ;
}

----

ICEs with gcc -O2 -g -fstack-protector-strong -fPIC -c -std=c++11

foo.cpp: In member function 'void C::emitMaps()':
foo.cpp:18:1: error: unrecognizable insn:
   18 | }
      | ^
(insn 4 3 5 2 (set (reg:DI 96)
        (const:DI (unspec:DI [
                    (symbol_ref:DI ("__stack_chk_guard") [flags 0xc0] <var_decl
0xffffb7a96360 __stack_chk_guard>)
                    (const_int 0 [0])
                ] UNSPEC_SALT_ADDR))) "foo.cpp":15:20 -1
     (nil))
during RTL pass: vregs
foo.cpp:18:1: internal compiler error: in extract_insn, at recog.c:2769
0x750627 _fatal_insn(char const*, rtx_def const*, char const*, int, char
const*)
        ../../gcc-fsf/gcc/rtl-error.c:108
0x750653 _fatal_insn_not_found(rtx_def const*, char const*, int, char const*)
        ../../gcc-fsf/gcc/rtl-error.c:116
0x10e332b extract_insn(rtx_insn*)
        ../../gcc-fsf/gcc/recog.c:2769
0xdf6477 instantiate_virtual_regs_in_insn
        ../../gcc-fsf/gcc/function.c:1611
0xdf6477 instantiate_virtual_regs
        ../../gcc-fsf/gcc/function.c:1985
0xdf6477 execute
        ../../gcc-fsf/gcc/function.c:2034

because the salt address is no longer folded away.

[Bug target/103085] [12 Regression] -fPIC and -fstack-protector-strong broken AArch64

[wilco at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103085

Wilco <wilco at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2021-11-04
           Assignee|unassigned at gcc dot gnu.org      |wilco at gcc dot gnu.org
     Ever confirmed|0                           |1

--- Comment #1 from Wilco <wilco at gcc dot gnu.org> ---
Yes there is something hacky about the way the salt thingy works. It adds some
unspec around a memory operand and still expects the standard movdi pattern to
recognize it.

It looks like adding support for CONST in Usw constraint and stripping the salt
before checking the symbol in mov_operand_p fixes it. I'm testing a patch.

[Bug target/103085] [12 Regression] -fPIC and -fstack-protector-strong broken AArch64

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103085

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1
   Target Milestone|---                         |12.0

[Bug target/103085] [12 Regression] -fPIC and -fstack-protector-strong broken AArch64

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103085

--- Comment #2 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Wilco Dijkstra <wilco@gcc.gnu.org>:

https://gcc.gnu.org/g:b33b2678342c655ff8ddb9fab379b26f656adc03

commit r12-4942-gb33b2678342c655ff8ddb9fab379b26f656adc03
Author: Wilco Dijkstra <wdijkstr@arm.com>
Date:   Fri Nov 5 15:05:15 2021 +0000

    AArch64: Fix PR103085

    The stack protector implementation hides symbols in a const unspec, which
means
    movdi/movsi patterns must always support const on symbol operands and
    explicitly strip away the unspec. Do this for the recently added GOT
    alternatives. Add a test to ensure stack-protector tests GOT accesses as
well.

    2021-11-05  Wilco Dijkstra  <wdijkstr@arm.com>

            PR target/103085
            * config/aarch64/aarch64.c (aarch64_mov_operand_p): Strip the salt
            first.
            * config/aarch64/constraints.md: Support const in Usw.

    gcc/testsuite/
            PR target/103085
            * gcc.target/aarch64/pr103085.c: New test

[Bug target/103085] [12 Regression] -fPIC and -fstack-protector-strong broken AArch64

[wilco at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103085

Wilco <wilco at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
             Status|ASSIGNED                    |RESOLVED

--- Comment #3 from Wilco <wilco at gcc dot gnu.org> ---
Fixed