selftests/powerpc: Fix TM resched DSCR test with some compilers
authorMichael Ellerman <mpe@ellerman.id.au>
Fri, 19 May 2017 01:29:04 +0000 (11:29 +1000)
committerMichael Ellerman <mpe@ellerman.id.au>
Fri, 19 May 2017 09:46:34 +0000 (19:46 +1000)
The tm-resched-dscr test has started failing sometimes, depending on
what compiler it's built with, eg:

  test: tm_resched_dscr
  Check DSCR TM context switch: tm-resched-dscr: tm-resched-dscr.c:76: test_body: Assertion `rv' failed.
  !! child died by signal 6

When it fails we see that the compiler doesn't initialise rv to 1 before
entering the inline asm block. Although that's counter intuitive, it
is allowed because we tell the compiler that the inline asm will write
to rv (using "=r"), meaning the original value is irrelevant.

Marking it as a read/write parameter would presumably work, but it seems
simpler to fix it by setting the initial value of rv in the inline asm.

Fixes: 96d016108640 ("powerpc: Correct DSCR during TM context switch")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Acked-by: Michael Neuling <mikey@neuling.org>
tools/testing/selftests/powerpc/tm/tm-resched-dscr.c

index d9c49f41515e704994ed1669dbe0a859a6dadd79..e79ccd6aada1a9acf10bb2860e76ddb444415d52 100644 (file)
@@ -42,12 +42,12 @@ int test_body(void)
        printf("Check DSCR TM context switch: ");
        fflush(stdout);
        for (;;) {
-               rv = 1;
                asm __volatile__ (
                        /* set a known value into the DSCR */
                        "ld      3, %[dscr1];"
                        "mtspr   %[sprn_dscr], 3;"
 
+                       "li      %[rv], 1;"
                        /* start and suspend a transaction */
                        "tbegin.;"
                        "beq     1f;"