Received: from eggs.gnu.org ([209.51.188.92]:34644) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1gz5gK-0007Qe-4W for qemu-devel@nongnu.org; Wed, 27 Feb 2019 15:19:57 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1gz5gB-0007NI-1j for qemu-devel@nongnu.org; Wed, 27 Feb 2019 15:19:51 -0500 Received: from mail-wm1-x342.google.com ([2a00:1450:4864:20::342]:36830) by eggs.gnu.org with esmtps (TLS1.0:RSA_AES_128_CBC_SHA1:16) (Exim 4.71) (envelope-from ) id 1gz5gA-0007AS-7r for qemu-devel@nongnu.org; Wed, 27 Feb 2019 15:19:46 -0500 Received: by mail-wm1-x342.google.com with SMTP id j125so6818975wmj.1 for ; Wed, 27 Feb 2019 12:19:26 -0800 (PST) References: <20190227111411.22890-1-david@redhat.com> <20190227111411.22890-3-david@redhat.com> From: Alex =?utf-8?Q?Benn=C3=A9e?= In-reply-to: Date: Wed, 27 Feb 2019 20:19:21 +0000 Message-ID: <87h8cpkmra.fsf@zen.linaroharston> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: quoted-printable Subject: Re: [Qemu-devel] [PATCH RFC 2/2] tests/tcg: target/s390: Add test for VECTOR LOAD GR FROM VR ELEMENT List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: David Hildenbrand Cc: qemu-devel@nongnu.org, qemu-s390x@nongnu.org, Thomas Huth , Cornelia Huck , Richard Henderson , Philippe =?utf-8?Q?Mathieu-Daud=C3=A9?= David Hildenbrand writes: > On 27.02.19 12:14, David Hildenbrand wrote: >> We want to make use of vectors, so use -march=3Dz13. To make it compile, >> use a reasonable optimization level (-O2), which seems to work just fine >> with all tests. >> >> Add some infrastructure for checking if SIGILL will be properly >> triggered. >> >> Signed-off-by: David Hildenbrand >> --- >> tests/tcg/s390x/Makefile.target | 3 ++- >> tests/tcg/s390x/helper.h | 28 +++++++++++++++++++++ >> tests/tcg/s390x/signal_helper.inc.c | 39 +++++++++++++++++++++++++++++ >> tests/tcg/s390x/vlgv.c | 37 +++++++++++++++++++++++++++ >> 4 files changed, 106 insertions(+), 1 deletion(-) >> create mode 100644 tests/tcg/s390x/helper.h >> create mode 100644 tests/tcg/s390x/signal_helper.inc.c >> create mode 100644 tests/tcg/s390x/vlgv.c >> >> diff --git a/tests/tcg/s390x/Makefile.target b/tests/tcg/s390x/Makefile.= target >> index 151dc075aa..d1ae755ab9 100644 >> --- a/tests/tcg/s390x/Makefile.target >> +++ b/tests/tcg/s390x/Makefile.target >> @@ -1,8 +1,9 @@ >> VPATH+=3D$(SRC_PATH)/tests/tcg/s390x >> -CFLAGS+=3D-march=3DzEC12 -m64 >> +CFLAGS+=3D-march=3Dz13 -m64 -O2 >> TESTS+=3Dhello-s390x >> TESTS+=3Dcsst >> TESTS+=3Dipm >> TESTS+=3Dexrl-trt >> TESTS+=3Dexrl-trtr >> TESTS+=3Dpack >> +TESTS+=3Dvlgv >> diff --git a/tests/tcg/s390x/helper.h b/tests/tcg/s390x/helper.h >> new file mode 100644 >> index 0000000000..845b8bb504 >> --- /dev/null >> +++ b/tests/tcg/s390x/helper.h >> @@ -0,0 +1,28 @@ >> +#ifndef TEST_TCG_S390x_VECTOR_H >> +#define TEST_TCG_S390x_VECTOR_H >> + >> +#include >> + >> +#define ES_8 0 >> +#define ES_16 1 >> +#define ES_32 2 >> +#define ES_64 3 >> +#define ES_128 4 >> + >> +typedef union S390Vector { >> + __uint128_t v; >> + uint64_t q[2]; >> + uint32_t d[4]; >> + uint16_t w[8]; >> + uint8_t h[16]; >> +} S390Vector; >> + >> +static inline void check(const char *s, bool cond) >> +{ >> + if (!cond) { >> + fprintf(stderr, "Check failed: %s\n", s); >> + exit(-1); >> + } >> +} >> + >> +#endif /* TEST_TCG_S390x_VECTOR_H */ >> diff --git a/tests/tcg/s390x/signal_helper.inc.c b/tests/tcg/s390x/signa= l_helper.inc.c >> new file mode 100644 >> index 0000000000..5bd69ca76a >> --- /dev/null >> +++ b/tests/tcg/s390x/signal_helper.inc.c >> @@ -0,0 +1,39 @@ >> +#include >> +#include >> +#include >> +#include >> +#include >> +#include "helper.h" >> + >> +jmp_buf jmp_env; >> + >> +static void sig_sigill(int sig) >> +{ >> + if (sig !=3D SIGILL) { >> + check("Wrong signal received", false); >> + } >> + longjmp(jmp_env, 1); >> +} >> + >> +#define CHECK_SIGILL(STATEMENT) \ >> +do { \ >> + struct sigaction act; \ >> + \ >> + act.sa_handler =3D sig_sigill; \ >> + act.sa_flags =3D 0; \ >> + if (sigaction(SIGILL, &act, NULL)) { \ >> + check("SIGILL handler not registered", false); \ >> + } \ >> + \ >> + if (setjmp(jmp_env) =3D=3D 0) { \ >> + STATEMENT; \ >> + check("SIGILL not triggered", false); \ >> + } \ >> + \ >> + act.sa_handler =3D SIG_DFL; \ >> + sigemptyset(&act.sa_mask); \ >> + act.sa_flags =3D 0; \ >> + if (sigaction(SIGILL, &act, NULL)) { \ >> + check("SIGILL handler not unregistered", false); \ >> + } \ >> +} while (0) > > Now this is some interesting hackery. > > I changed it to > > jmp_buf jmp_env; > > static void sig_sigill(int sig) > { > if (sig !=3D SIGILL) { > check("Wrong signal received", false); > } > longjmp(jmp_env, 1); > } > > #define CHECK_SIGILL(STATEMENT) \ > do { \ > if (signal(SIGILL, sig_sigill) =3D=3D SIG_ERR) { \ > check("SIGILL not registered", false); \ > } \ > if (setjmp(jmp_env) =3D=3D 0) { \ > STATEMENT; \ > check("SIGILL not triggered", false); \ > } \ > if (signal(SIGILL, SIG_DFL) =3D=3D SIG_ERR) { \ > check("SIGILL not registered", false); \ > } \ > } while (0) > > > However it only works once. During the second signal, QEMU decides to > set the default handler. > > 1. In a signal handler that signal is blocked. We leave that handler via > a longjump. So after the first signal, the signal is blocked. > > 2. In QEMU, we decide that synchronous signals cannot be blocked and > simply override the signal handler to default handler. So on the second > signal, we execute the default handler and not our defined handler. > > Multiple ways to fix: > > 1. We have to manually unblock the signal in that hackery above. > 2. In QEMU, don't block synchronous signals. > 3. In QEMU, don't set the signal handler to the default handler in case > a synchronous signal is blocked. This all seems a little over-engineered. This is a single-threaded test so what's wrong with a couple of flags: bool should_get_sigill; and the handler doing: if (!should_get_sigill) { unexpected_sigills++; } Tests don't have to be pretty, just reliable. > > > Trying to run the test on a real s390x linux indicates that 1. should be > the right approach. > > [linux1@redhat-kvm ~]$ ./vge > Illegal instruction (core dumped) -- Alex Benn=C3=A9e