T5435_gcc_v fails on ARM

The rts/T5435_gcc_v testcase sometimes fails on ARM. As far as I can tell there are two principle failure modes: hanging and segmentation faulting.

changed milestone to %8.0.1

changed weight to 5

added RTS Tbug Trac import labels

A segmentation fault look like this,

$ gdb --args ./T5435_v_gcc v ./T5435_load_v_gcc.o
...
(gdb) run
Starting program: /mnt/ext/exp/ghc/testsuite/tests/rts/T5435_v_gcc v ./T5435_load_v_gcc.o
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/arm-linux-gnueabihf/libthread_db.so.1".
initializer run

Program received signal SIGSEGV, Segmentation fault.
__GI__IO_fflush (fp=0xe3403000) at iofflush.c:40
40      iofflush.c: No such file or directory.
(gdb) bt
#0  __GI__IO_fflush (fp=0xe3403000) at iofflush.c:40
#1  0xb6ff6150 in ?? ()
Backtrace stopped: previous frame identical to this frame (corrupt stack?)

Sometimes it also segfaults inside of,

#1  0xb6e15c3c in _IO_new_file_write (f=0xb6ff7000, data=0xb6ff616c, n=6) at fileops.c:1253

changed the description

Here's another outcome with a useful backtrace due to compiling with -debug,

(gdb) run
Starting program: /mnt/ext/exp/ghc/testsuite/tests/rts/T5435_v_gcc v ./T5435_load_v_gcc.o
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/arm-linux-gnueabihf/libthread_db.so.1".
post-initLinker
post-if

Program received signal SIGILL, Illegal instruction.
0xb6ff7000 in ?? ()
(gdb) bt
#0  0xb6ff7000 in ?? ()
#1  0x002be160 in ocRunInit_ELF (oc=0x0) at rts/Linker.c:5685
#2  0x002be160 in ocRunInit_ELF (oc=0x38e9d8) at rts/Linker.c:5685
#3  0x002b9ea4 in resolveObjs_ () at rts/Linker.c:2447
#4  0x002b9f30 in resolveObjs () at rts/Linker.c:2475
#5  0x0000ad3c in r3ze_info$def ()
Backtrace stopped: previous frame identical to this frame (corrupt stack?)

I believe the issue here may be incomplete instruction cache flushing. With this patch,

diff --git a/rts/Linker.c b/rts/Linker.c
index aadd8d4..1101dea 100644
--- a/rts/Linker.c
+++ b/rts/Linker.c
@@ -2748,8 +2748,10 @@ static void
 ocFlushInstructionCache( ObjectCode *oc )
 {
     // Object code
+    printf("Flushed %p - %p\n", oc->image, oc->image + oc->fileSize);
     __clear_cache(oc->image, oc->image + oc->fileSize);
     // Jump islands
+    printf("Flushed %p - %p\n", oc->symbol_extras, &oc->symbol_extras[oc->n_symbol_extras]);
     __clear_cache(oc->symbol_extras, &oc->symbol_extras[oc->n_symbol_extras]);
 }
 
@@ -5706,6 +5708,7 @@ static int ocRunInit_ELF( ObjectCode *oc )
          init_start = (init_t*)init_startC;
          init_end = (init_t*)(init_startC + shdr[i].sh_size);
          for (init = init_start; init < init_end; init++) {
+            printf("init %p\n", init);
             (*init)(argc, argv, envv);
          }
       }

I see the following output on failure,

relocations for section 6 using symtab 12 and strtab 13
Rel entry   0 is raw( (nil)  0xa02)`initializer' resolves to 0xb6ff6138
Reloc: P = 0xb6ff6190   S = 0xb6ff6138   A = (nil)
Flushed 0xb6ff7000 - 0xb6ff753c
Flushed 0xb6ff6008 - 0xb6ff6138
init 0xb6ff6190

Program received signal SIGILL, Illegal instruction.
0xb6ff7000 in ?? ()

Indeed it looks like the location with the initializer (0xb6ff6190) was never flushed.

Trac metadata

Trac field	Value
Differential revisions	- → D1708

The test survives 500 runs without failure with D1708.

It also seems that this fixed the linker_unload test! Yay!

mentioned in commit 0054bcd4

closed

Trac metadata

Trac field	Value
Resolution	Unresolved → ResolvedFixed

mentioned in issue #11340 (closed)

added runtime crash label

added Pnormal label

mentioned in issue #17559

marked this issue as related to #17559

mentioned in issue #22970

marked this issue as related to #22970