https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96180
Bug ID: 96180
Summary: gcc.dg/torture/pr96133.c FAILs
Product: gcc
Version: 11.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: ro at gcc dot gnu.org
CC: ebotcazou at gcc dot gnu.org, rguenth at gcc dot gnu.org
Target Milestone: ---
Target: sparc-sun-solaris2.11
The new gcc.dg/torture/pr96133.c test FAILs on Solaris/SPARC (both 32 and
64-bit)
at -O0 only:
+FAIL: gcc.dg/torture/pr96133.c -O0 execution test
The failure mode is a bit weird in so far as the test reliably fails with make
check/runtest, but works when run manually. With runtest -v, gcc.log shows
spawning command ./pr96133.exe
spawn [open ...]
exp10 file9
close result is child killed: bus error
FAIL: gcc.dg/torture/pr96133.c -O0 execution test
and running the whole runtest under truss agrees:
5090: Incurred fault #5, FLTACCESS %pc = 0x00010CB0
5090: siginfo: SIGBUS BUS_ADRALN addr=0x00010A34
5090: Received signal #10, SIGBUS [default]
5090: siginfo: SIGBUS BUS_ADRALN addr=0x00010A34
If I let a core dump be created, I see
0x10c9c: save %sp, -112, %sp
0x10ca0: sethi %hi(0x10800), %g1
0x10ca4: or %g1, 0x234, %g1 ! 0x10a34 <_start+84>
0x10ca8: st %g1, [ %fp + -4 ]
0x10cac: ld [ %fp + -4 ], %g1
0x10cb0: ldtw [ %g1 ], %g2
(which is the beginning of main), and %g1 at that point is only 4-byte aligned
while ldtw requires double-word aligment.
