| Message ID | 20161220100941.B4FA5401BD5F6@oldenburg.str.redhat.com |
|---|---|
| State | New |
| Headers | show |
On 12/20/2016 11:09 AM, Florian Weimer wrote: > Some targets fail to apply dead store elimination to the > memset call in setup_ordinary_clear. Before this commit, > this causes the test case to fail. Instead, the test case > now logs lack of memset elimination as an informational > message. > > 2016-12-20 Florian Weimer <fweimer@redhat.com> > > Do not require memset elimination in explicit_bzero test. > * string/tst-xbzero-opt.c (prepare_test_buffer): Force inlining. > (enum test_expectation): Add NO_EXPECTATIONS. > (subtests): NO_EXPECTATIONS for ordinary clear. > (check_test_buffer): Handle NO_EXPECTATIONS. > * string/Makefile (CFLAGS-tst-xbzero-opt.c): Compile with -O3. Stefan, this test still fails for me on s390x: PASS: no clear/prepare: expected 32 got 32 PASS: no clear/test: expected some got 32 PASS: ordinary clear/prepare: expected 32 got 32 INFO: ordinary clear/test: found 0 patterns (memset not eliminated) PASS: explicit clear/prepare: expected 32 got 32 FAIL: explicit clear/test: expected 0 got 1 Do you have an idea what's going on there? Thanks, Florian
On 12/21/2016 07:04 PM, Florian Weimer wrote: > On 12/20/2016 11:09 AM, Florian Weimer wrote: >> Some targets fail to apply dead store elimination to the >> memset call in setup_ordinary_clear. Before this commit, >> this causes the test case to fail. Instead, the test case >> now logs lack of memset elimination as an informational >> message. >> >> 2016-12-20 Florian Weimer <fweimer@redhat.com> >> >> Do not require memset elimination in explicit_bzero test. >> * string/tst-xbzero-opt.c (prepare_test_buffer): Force inlining. >> (enum test_expectation): Add NO_EXPECTATIONS. >> (subtests): NO_EXPECTATIONS for ordinary clear. >> (check_test_buffer): Handle NO_EXPECTATIONS. >> * string/Makefile (CFLAGS-tst-xbzero-opt.c): Compile with -O3. > > Stefan, this test still fails for me on s390x: > > PASS: no clear/prepare: expected 32 got 32 > PASS: no clear/test: expected some got 32 > PASS: ordinary clear/prepare: expected 32 got 32 > INFO: ordinary clear/test: found 0 patterns (memset not eliminated) > PASS: explicit clear/prepare: expected 32 got 32 > FAIL: explicit clear/test: expected 0 got 1 > > Do you have an idea what's going on there? I filed bug 21006 and will add it as a release blocker. Thanks, Florian
diff --git a/string/Makefile b/string/Makefile index 9a8b46d..0816277 100644 --- a/string/Makefile +++ b/string/Makefile @@ -71,6 +71,7 @@ CFLAGS-tst-strlen.c = -fno-builtin CFLAGS-stratcliff.c = -fno-builtin CFLAGS-test-ffs.c = -fno-builtin CFLAGS-tst-inlcall.c = -fno-builtin +CFLAGS-tst-xbzero-opt.c = -O3 ifeq ($(run-built-tests),yes) $(objpfx)tst-svc-cmp.out: tst-svc.expect $(objpfx)tst-svc.out diff --git a/string/tst-xbzero-opt.c b/string/tst-xbzero-opt.c index 312857e..3fcaf28 100644 --- a/string/tst-xbzero-opt.c +++ b/string/tst-xbzero-opt.c @@ -88,19 +88,19 @@ static const unsigned char test_pattern[16] = between preparing it and letting it go out of scope, and we expect to find it. This confirms that the test buffer does get filled in and we can find it from the stack buffer. In the "ordinary_clear" - case, we clear it using memset, and we expect to find it. This - confirms that the compiler can optimize out block clears in this - context; if it can't, the real test might be succeeding for the - wrong reason. Finally, the "explicit_clear" case uses - explicit_bzero and expects _not_ to find the test buffer, which is - the real test. */ + case, we clear it using memset. Depending on the target, the + compiler may not be able to apply dead store elimination to the + memset call, so the test does not fail if the memset is not + eliminated. Finally, the "explicit_clear" case uses explicit_bzero + and expects _not_ to find the test buffer, which is the real + test. */ static ucontext_t uc_main, uc_co; /* Always check the test buffer immediately after filling it; this makes externally visible side effects depend on the buffer existing and having been filled in. */ -static void +static inline __attribute__ ((always_inline)) void prepare_test_buffer (unsigned char *buf) { for (unsigned int i = 0; i < PATTERN_REPS; i++) @@ -133,7 +133,10 @@ setup_explicit_clear (void) explicit_bzero (buf, TEST_BUFFER_SIZE); } -enum test_expectation { EXPECT_NONE, EXPECT_SOME, EXPECT_ALL }; +enum test_expectation + { + EXPECT_NONE, EXPECT_SOME, EXPECT_ALL, NO_EXPECTATIONS + }; struct subtest { void (*setup_subtest) (void); @@ -145,7 +148,9 @@ static const struct subtest *cur_subtest; static const struct subtest subtests[] = { { setup_no_clear, "no clear", EXPECT_SOME }, - { setup_ordinary_clear, "ordinary clear", EXPECT_SOME }, + /* The memset may happen or not, depending on compiler + optimizations. */ + { setup_ordinary_clear, "ordinary clear", NO_EXPECTATIONS }, { setup_explicit_clear, "explicit clear", EXPECT_NONE }, { 0, 0, -1 } }; @@ -225,6 +230,11 @@ check_test_buffer (enum test_expectation expected, } break; + case NO_EXPECTATIONS: + printf ("INFO: %s/%s: found %d patterns%s\n", label, stage, cnt, + cnt == 0 ? " (memset not eliminated)" : ""); + break; + default: printf ("ERROR: %s/%s: invalid value for 'expected' = %d\n", label, stage, (int)expected);