[RTEMS Project] #4892: MacOS tools build failures on Ventura and M series processors

[RTEMS trac]

#4892: MacOS tools build failures on Ventura and M series processors
-------------------------+--------------------------
 Reporter:  Chris Johns  |       Owner:  Chris Johns
     Type:  defect       |      Status:  assigned
 Priority:  normal       |   Milestone:  6.1
Component:  tool/rsb     |     Version:  6
 Severity:  normal       |  Resolution:
 Keywords:               |  Blocked By:
 Blocking:               |
-------------------------+--------------------------

Comment (by Chris Johns):

 Building on M2 Pro with Ventura 13.3 and Xcode Version 14.3 (14E222b)
 results in segmentation faults in `cc1` and `cc1plus`. Inspecting the
 MacOS crash reports showed invalid accesses to the stack guard area of
 memory. A typical crash report is:
 {{{
 Exception Type:        EXC_BAD_ACCESS (SIGSEGV)
 Exception Codes:       KERN_PROTECTION_FAILURE at 0x00000001699dbff0
 Exception Codes:       0x0000000000000002, 0x00000001699dbff0

 Termination Reason:    Namespace SIGNAL, Code 11 Segmentation fault: 11
 Terminating Process:   exc handler [30551]

 VM Region Info: 0x1699dbff0 is in 0x1699d8000-0x1699dc000;  bytes after
 start: 16368  bytes before end: 15
       REGION TYPE                    START - END         [ VSIZE] PRT/MAX
 SHRMOD  REGION DETAIL
       MALLOC_MEDIUM (reserved)    147800000-148000000    [ 8192K] rw-/rwx
 SM=NUL  ...(unallocated)
       GAP OF 0x219d8000 BYTES
 --->  STACK GUARD                 1699d8000-1699dc000    [   16K] ---/rwx
 SM=NUL  ... for thread 0
       Stack                       1699dc000-16d9d8000    [ 64.0M] rw-/rwx
 SM=PRV  thread 0

 Thread 0 Crashed::  Dispatch queue: com.apple.main-thread
 0   cc1                                        0x10335d630 __gmpn_lshift +
 208
 1   cc1                                        0x1033438c8 mpfr_mpn_exp +
 684 (mpn_exp.c:151)
 2   ???                                        0x2db39f030 ???


 Thread 0 crashed with ARM Thread State (64-bit):
     x0: 0x0000000000000000   x1: 0x00000001699dc010   x2:
 0x0000000000000000   x3: 0x0000000000000001
     x4: 0x0000000000000000   x5: 0x0000000000000000   x6:
 0x0000000000000000   x7: 0x0000000000000000
     x8: 0xffffffffffffffff   x9: 0x0000000000000007  x10:
 0x0000000000000000  x11: 0x0000000000000000
    x12: 0x0000000000000000  x13: 0x0000000000000000  x14:
 0x0000000000000001  x15: 0x00000000ffffffd0
    x16: 0x00000001699dc170  x17: 0x0000000082eff7fb  x18:
 0xfffffffffffffe00  x19: 0xffffffffffffff04
    x20: 0x0000000000000003  x21: 0x0000000000000003  x22:
 0x000000016d9cf818  x23: 0x0000000000000005
    x24: 0xffffffffffffff08  x25: 0x0000000000000004  x26:
 0x0000000000000012  x27: 0x000000016d9cf6c8
    x28: 0x000000016d9cf6e0   fp: 0x000000016d9cf7c0   lr:
 0x00000001033438c8
     sp: 0x000000016d9cf6b0   pc: 0x000000010335d630 cpsr: 0x20001000
    far: 0x00000001699dbff0  esr: 0x92000007 (Data Abort) byte read
 Translation fault

 Binary Images:
        0x102428000 -        0x10359bfff cc1 (*) <72d6a4cf-055f-369c-89cb-
 4cbf48bff6cf> /Users/USER/*/cc1
                0x0 - 0xffffffffffffffff ??? (*)
 <00000000-0000-0000-0000-000000000000> ???
 }}}
 The `FAR` is in the stack guard area. The crashes are always in the MPFR
 library and relate to calls to GMP. The GMP tests pass. The MPFR built by
 `gcc` tests fail and a hand built MPFR tests pass.

 The crash leaves the build so changing to the MPFR build directory lets
 you run the tests:
 {{{
 cd build/arm-rtems6-gcc-cf59d86-newlib-17ac400-arm64-apple-
 darwin22.4.0-1/build/mpfr
 make check
 }}}
 The results are:
 {{{
 ============================================================================
 Testsuite summary for MPFR 4.2.0
 ============================================================================
 # TOTAL: 197
 # PASS:  97
 # SKIP:  3
 # XFAIL: 0
 # FAIL:  97
 # XPASS: 0
 # ERROR: 0
 }}}
 The failures are all seg faults.

 I upgraded to MPFR 4.2.0 as it does not need the `fix-mac-arm64-mpfr-
 config.patch` patch. Building by hand and referencing the `gcc` build GMP
 has the following results:
 {{{
 ============================================================================
 Testsuite summary for MPFR 4.2.0
 ============================================================================
 # TOTAL: 197
 # PASS:  194
 # SKIP:  3
 # XFAIL: 0
 # FAIL:  0
 # XPASS: 0
 # ERROR: 0
 }}}
 I provided the same `configure` command line used by `gcc` when it builds
 the library internally. There are no sig faults.

 I ran the tests for GMP and there were no errors.

 Close inspection of the build and the `config.log` reveals a difference.
 The `gcc` build has:
 {{{
 -DHAVE___GMPN_INVERT_LIMB=1 -DHAVE___GMPN_RSBLSH1_N=1
 }}}
 and they are not present when building by hand. The `config.log` shows:
 {{{
 configure:19315: checking for __gmpn_invert_limb
  [..]
 configure:19315: result: yes
 }}}
 The hand built MPFR has:
 {{{
 configure:19315: checking for __gmpn_invert_limb
  [..]
 configure:19315: result: no
 }}}
 This is the only difference I could see.

--
Ticket URL: <http://devel.rtems.org/ticket/4892#comment:10>
RTEMS Project <http://www.rtems.org/>
RTEMS Project