tests/tcg: Vector instruction tests for target/s390x
cover tests/tcg: Vector instruction tests for target/s390x
As I currently work on vector instruction support for s390x/tcg, for now I wrote my tests for kvm-unit-tests, but tests/tcg seems to be a better fit. The only tricky part is testing interrupt handling, but that also seems to be possible using some signal hackery.
This is only one test to discuss if the approach make sense. These patches only work when applied on top of:
https://github.com/davidhildenbrand/qemu/tree/vx
David Hildenbrand (2):
tests/tcg: Allow targets to set the optimization level tests/tcg: target/s390: Add test for VECTOR LOAD GR FROM VR ELEMENT
tests/tcg/Makefile | 4 +-- 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 +++++++++++++++++++++++++++ 5 files changed, 108 insertions(+), 3 deletions(-) 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
Signature · 1 line
2.17.2
RFC PATCH 1/2 tests/tcg: Allow targets to set the optimization level
We sometimes want basic optimizations, e.g. for constant propagation into inlined functions.
Especially for inline asm with immediates;
static inline void some_instr(uint8_t imm)
{
asm volatile("...", :: "i" (imm));
}static void test()
{
some_instr(1);
}Which is impossible with -O0. So make -O0 the default but let targets override it.
--- tests/tcg/Makefile | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
tests/tcg/Makefile +2 -2
@@ -47,7 +47,7 @@ skip-test = @printf " SKIPPED %s on $(TARGET_NAME) because %s\n" $1 $2 TESTS= # Start with a blank slate, the build targets get to add stuff first -CFLAGS= +CFLAGS=-O0 QEMU_CFLAGS= LDFLAGS=
@@ -70,7 +70,7 @@ ifneq ($(TARGET_BASE_ARCH),$(TARGET_NAME)) endif # Add the common build options -CFLAGS+=-Wall -O0 -g -fno-strict-aliasing +CFLAGS+=-Wall -g -fno-strict-aliasing ifeq ($(BUILD_STATIC),y) LDFLAGS+=-static endif
Signature · 1 line
2.17.2
Re: [Qemu-devel] [PATCH RFC 1/2] tests/tcg: Allow targets to set the optimization level
Quoted 4 lines · David Hildenbrand
We sometimes want basic optimizations, e.g. for constant propagation into inlined functions.
Especially for inline asm with immediates;
I agree you need this, but...
<snip>
tests/tcg/Makefile ifneq ($(TARGET_BASE_ARCH),$(TARGET_NAME)) · line 73
Quoted 12 lines
# Start with a blank slate, the build targets get to add stuff first
tests/tcg/Makefile +2
-2-CFLAGS= +CFLAGS=-O0 QEMU_CFLAGS= LDFLAGS=@@ -70,7 +70,7 @@ ifneq ($(TARGET_BASE_ARCH),$(TARGET_NAME)) endif # Add the common build options -CFLAGS+=-Wall -O0 -g -fno-strict-aliasing +CFLAGS+=-Wall -g -fno-strict-aliasing
You don't need to do this - we already have EXTRA_CFLAGS which will always be at the end of the compile invocation:
# enable vectors and optimisation to vectorise for this test vglv: CFLAGS+=-march=z13 -m64 vglv: EXTRA_CFLAGS+=-O2
tests/tcg/Makefile ifneq ($(TARGET_BASE_ARCH),$(TARGET_NAME)) · line 76
Quoted 3 lines
tests/tcg/Makefile +0
-0ifeq ($(BUILD_STATIC),y) LDFLAGS+=-static endif
--
Alex Bennée
Re: [Qemu-devel] [PATCH RFC 1/2] tests/tcg: Allow targets to set the optimization level
Quoted 28 lines · Alex Bennée
Quoted 4 lines · David Hildenbrand
We sometimes want basic optimizations, e.g. for constant propagation into inlined functions.
Especially for inline asm with immediates;
I agree you need this, but...
<snip>
tests/tcg/Makefile ifneq ($(TARGET_BASE_ARCH),$(TARGET_NAME)) · line 73
Quoted 12 lines
# Start with a blank slate, the build targets get to add stuff first
tests/tcg/Makefile +2
-2-CFLAGS= +CFLAGS=-O0 QEMU_CFLAGS= LDFLAGS=@@ -70,7 +70,7 @@ ifneq ($(TARGET_BASE_ARCH),$(TARGET_NAME)) endif # Add the common build options -CFLAGS+=-Wall -O0 -g -fno-strict-aliasing +CFLAGS+=-Wall -g -fno-strict-aliasingYou don't need to do this - we already have
EXTRA_CFLAGSwhich will always be at the end of the compile invocation:
Quoted 3 lines
# enable vectors and optimisation to vectorise for this test vglv: CFLAGS+=-march=z13 -m64 vglv: EXTRA_CFLAGS+=-O2
For whatever reason I can't get this to work. EXTRA_CFLAGS+=-O2 is always silently dropped:
make SHELL='sh -x' run-tcg-tests-s390x-linux-user
+ /home/dhildenb/git/qemu/tests/docker/docker.py cc --user 100813 --cc s390x-linux-gnu-gcc -i qemu:debian-s390x-cross -s /home/dhildenb/git/qemu -- -march=zEC12 -m64 -Wall -O0 -g -fno-strict-aliasing -march=z13 /home/dhildenb/git/qemu/tests/tcg/s390x/vlgv.c -o vlgv -static /home/dhildenb/git/qemu/tests/tcg/s390x/vlgv.c: In function 'vlgv':/home/dhildenb/git/qemu/tests/tcg/s390x/vlgv.c:8:5: warning: asm operand 3 probably doesn't match constraints
asm volatile("vlgv %[r1], %[v3], 0(%[a2]), %[m4]\n"
^~~/home/dhildenb/git/qemu/tests/tcg/s390x/vlgv.c:8:5: error: impossible constraint in 'asm'
Signature · 3 lines
Thanks, David / dhildenb
Re: [Qemu-devel] [PATCH RFC 1/2] tests/tcg: Allow targets to set the optimization level
Quoted 46 lines · David Hildenbrand
Quoted 28 lines · Alex Bennée
Quoted 4 lines · David Hildenbrand
We sometimes want basic optimizations, e.g. for constant propagation into inlined functions.
Especially for inline asm with immediates;
I agree you need this, but...
<snip>
tests/tcg/Makefile ifneq ($(TARGET_BASE_ARCH),$(TARGET_NAME)) · line 73
Quoted 12 lines
# Start with a blank slate, the build targets get to add stuff first
tests/tcg/Makefile +2
-2-CFLAGS= +CFLAGS=-O0 QEMU_CFLAGS= LDFLAGS=@@ -70,7 +70,7 @@ ifneq ($(TARGET_BASE_ARCH),$(TARGET_NAME)) endif # Add the common build options -CFLAGS+=-Wall -O0 -g -fno-strict-aliasing +CFLAGS+=-Wall -g -fno-strict-aliasingYou don't need to do this - we already have
EXTRA_CFLAGSwhich will always be at the end of the compile invocation:Quoted 3 lines
# enable vectors and optimisation to vectorise for this test vglv: CFLAGS+=-march=z13 -m64 vglv: EXTRA_CFLAGS+=-O2For whatever reason I can't get this to work. EXTRA_CFLAGS+=-O2 is always silently dropped:
make SHELL='sh -x' run-tcg-tests-s390x-linux-user
+
/home/dhildenb/git/qemu/tests/docker/docker.pycc --user 100813 --cc s390x-linux-gnu-gcc -i qemu:debian-s390x-cross -s/home/dhildenb/git/qemu-- -march=zEC12 -m64 -Wall -O0 -g -fno-strict-aliasing -march=z13/home/dhildenb/git/qemu/tests/tcg/s390x/vlgv.c-o vlgv -static/home/dhildenb/git/qemu/tests/tcg/s390x/vlgv.c: In function 'vlgv':/home/dhildenb/git/qemu/tests/tcg/s390x/vlgv.c:8:5: warning: asm operand 3 probably doesn't match constraints
Quoted 5 lines
asm volatile("vlgv %[r1], %[v3], 0(%[a2]), %[m4]\n" ^~~
/home/dhildenb/git/qemu/tests/tcg/s390x/vlgv.c:8:5: error: impossible constraint in 'asm'
vlgv: CFLAGS+=-march=z13 -O2
However works. I am very confused now. I guess this is about evaluation order or something like that.
Signature · 3 lines
Thanks, David / dhildenb
Re: [Qemu-devel] [PATCH RFC 1/2] tests/tcg: Allow targets to set the optimization level
In reply to Alex BennéeQuoted 28 lines · Alex Bennée
Quoted 4 lines · David Hildenbrand
We sometimes want basic optimizations, e.g. for constant propagation into inlined functions.
Especially for inline asm with immediates;
I agree you need this, but...
<snip>
tests/tcg/Makefile ifneq ($(TARGET_BASE_ARCH),$(TARGET_NAME)) · line 73
Quoted 12 lines
# Start with a blank slate, the build targets get to add stuff first
tests/tcg/Makefile +2
-2-CFLAGS= +CFLAGS=-O0 QEMU_CFLAGS= LDFLAGS=@@ -70,7 +70,7 @@ ifneq ($(TARGET_BASE_ARCH),$(TARGET_NAME)) endif # Add the common build options -CFLAGS+=-Wall -O0 -g -fno-strict-aliasing +CFLAGS+=-Wall -g -fno-strict-aliasingYou don't need to do this - we already have
EXTRA_CFLAGSwhich will always be at the end of the compile invocation:
Quoted 4 lines
# enable vectors and optimisation to vectorise for this test vglv: CFLAGS+=-march=z13 -m64 vglv: EXTRA_CFLAGS+=-O2
Cool, missed that - thanks!
Signature · 3 lines
Thanks, David / dhildenb
Re: [Qemu-devel] [PATCH RFC 1/2] tests/tcg: Allow targets to set the optimization level
In reply to David HildenbrandQuoted 11 lines · David Hildenbrand
Especially for inline asm with immediates;
static inline void
some_instr(uint8_t imm){ asm volatile("...", :: "i" (imm)); }static void
test(){ some_instr(1); }
FWIW, __attribute__((always_inline)) should work even with -O0.
r~
RFC PATCH 2/2 tests/tcg: target/s390: Add test for VECTOR LOAD GR FROM VR ELEMENT
In reply to David HildenbrandWe want to make use of vectors, so use -march=z13. To make it compile, use a reasonable optimization level (-O2), which seems to work just fine with all tests.
Add some infrastructure for checking if SIGILL will be properly triggered.
--- tests/tcg/s390x/Makefile.target | 3 ++- tests/tcg/s390x/helper.h | 28 +++++++++++++++++++++ tests/tcg/s390x/signal_helper.inc.c | 39 +++++++++++++++++++++++++++++ tests/tcg/s390x/vlgv.c | 37 +++++++++++++++++++++++++++ 4 files changed, 106 insertions(+), 1 deletion(-) create mode 100644 tests/tcg/s390x/helper.h create mode 100644 tests/tcg/s390x/signal_helper.inc.c create mode 100644 tests/tcg/s390x/vlgv.c
tests/tcg/s390x/Makefile.target +2 -1
@@ -1,8 +1,9 @@ VPATH+=$(SRC_PATH)/tests/tcg/s390x -CFLAGS+=-march=zEC12 -m64 +CFLAGS+=-march=z13 -m64 -O2 TESTS+=hello-s390x TESTS+=csst TESTS+=ipm TESTS+=exrl-trt TESTS+=exrl-trtr TESTS+=pack +TESTS+=vlgv
tests/tcg/s390x/helper.h new file +28 -0
@@ -0,0 +1,28 @@ +#ifndef TEST_TCG_S390x_VECTOR_H +#define TEST_TCG_S390x_VECTOR_H + +#include <stdint.h> + +#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 */
tests/tcg/s390x/signal_helper.inc.c new file +39 -0
@@ -0,0 +1,39 @@ +#include <stdlib.h> +#include <stdio.h> +#include <stdbool.h> +#include <signal.h> +#include <setjmp.h> +#include "helper.h" + +jmp_buf jmp_env; + +static void sig_sigill (int sig) +{ +if (sig !=SIGILL ) { +check ("Wrong signal received" , false); + } +longjmp (jmp_env,1 ); +} + +#define CHECK_SIGILL (STATEMENT ) \ +do { \ + struct sigaction act; \ + \ + act.sa_handler = sig_sigill; \ + act.sa_flags = 0; \ + if (sigaction(SIGILL, &act, NULL)) { \ + check("SIGILL handler not registered", false); \ + } \ + \ + if (setjmp(jmp_env) == 0) { \ + STATEMENT; \ + check("SIGILL not triggered", false); \ + } \ + \ + act.sa_handler = SIG_DFL; \ + sigemptyset(&act.sa_mask); \ + act.sa_flags = 0; \ + if (sigaction(SIGILL, &act, NULL)) { \ + check("SIGILL handler not unregistered", false); \ + } \ +} while (0)
tests/tcg/s390x/vlgv.c new file +37 -0
@@ -0,0 +1,37 @@ +#include <stdint.h> +#include <unistd.h> +#include "signal_helper.inc.c" + +static inline void vlgv (uint64_t *r1,S390Vector *v3,const void *a2, +uint8_t m4) +{ + asmvolatile ("vlgv %[r1], %[v3], 0(%[a2]), %[m4]\n" + : [r1]"+d" (*r1), + [v3]"+v" (v3->v) + : [a2]"d" (a2), + [m4]"i" (m4)); +} + +int main (void ) +{ +S390Vector v3 = { + .q[0 ] =0x0011223344556677ull , + .q[1 ] =0x8899aabbccddeeffull , + }; +uint64_t r1 =0 ; + +/* Directly set all ignored bits to */ +vlgv (&r1, &v3, (void *)(7 | ~0xf ),ES_8 ); +check ("8 bit" , r1 ==0x77 ); +vlgv (&r1, &v3, (void *)(4 | ~0x7 ),ES_16 ); +check ("16 bit" , r1 ==0x8899 ); +vlgv (&r1, &v3, (void *)(3 | ~0x3 ),ES_32 ); +check ("32 bit" , r1 ==0xccddeeff ); +vlgv (&r1, &v3, (void *)(1 | ~0x1 ),ES_64 ); +check ("64 bit" , r1 ==0x8899aabbccddeeffull ); +check ("v3 not modified" , v3.q[0 ] ==0x0011223344556677ull && + v3.q[1 ] ==0x8899aabbccddeeffull ); + +CHECK_SIGILL (vlgv (&r1, &v3,NULL ,ES_128 )); +return 0 ; +}
Signature · 1 line
2.17.2
Re: [Qemu-devel] [PATCH RFC 2/2] tests/tcg: target/s390: Add test for VECTOR LOAD GR FROM VR ELEMENT
Quoted 3 lines · David Hildenbrand
We want to make use of vectors, so use -march=z13. To make it compile, use a reasonable optimization level (-O2), which seems to work just fine with all tests.
Just double-checked, at least exrl-trt will need a fixup to work with -O2.
Quoted 153 lines
Add some infrastructure for checking if SIGILL will be properly triggered.
Signed-off-by David Hildenbrand <[email protected]>--- 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.ctests/tcg/s390x/Makefile.target +2
-1@@ -1,8 +1,9 @@ VPATH+=$(SRC_PATH)/tests/tcg/s390x -CFLAGS+=-march=zEC12 -m64 +CFLAGS+=-march=z13 -m64 -O2 TESTS+=hello-s390x TESTS+=csst TESTS+=ipm TESTS+=exrl-trt TESTS+=exrl-trtr TESTS+=pack +TESTS+=vlgvtests/tcg/s390x/helper.h new file +28
-0@@ -0,0 +1,28 @@ +#ifndef TEST_TCG_S390x_VECTOR_H +#define TEST_TCG_S390x_VECTOR_H + +#include <stdint.h> + +#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 */ tests/tcg/s390x/signal_helper.inc.c new file +39
-0@@ -0,0 +1,39 @@ +#include <stdlib.h> +#include <stdio.h> +#include <stdbool.h> +#include <signal.h> +#include <setjmp.h> +#include "helper.h" + +jmp_buf jmp_env; + +static void sig_sigill (int sig) +{ +if (sig !=SIGILL ) { +check ("Wrong signal received" , false); + } +longjmp (jmp_env,1 ); +} + +#define CHECK_SIGILL (STATEMENT ) \ +do { \ + struct sigaction act; \ + \ + act.sa_handler = sig_sigill; \ + act.sa_flags = 0; \ + if (sigaction(SIGILL, &act, NULL)) { \ + check("SIGILL handler not registered", false); \ + } \ + \ + if (setjmp(jmp_env) == 0) { \ + STATEMENT; \ + check("SIGILL not triggered", false); \ + } \ + \ + act.sa_handler = SIG_DFL; \ + sigemptyset(&act.sa_mask); \ + act.sa_flags = 0; \ + if (sigaction(SIGILL, &act, NULL)) { \ + check("SIGILL handler not unregistered", false); \ + } \ +} while (0)tests/tcg/s390x/vlgv.c new file +37
-0@@ -0,0 +1,37 @@ +#include <stdint.h> +#include <unistd.h> +#include "signal_helper.inc.c" + +static inline void vlgv (uint64_t *r1,S390Vector *v3,const void *a2, +uint8_t m4) +{ + asmvolatile ("vlgv %[r1], %[v3], 0(%[a2]), %[m4]\n" + : [r1]"+d" (*r1), + [v3]"+v" (v3->v) + : [a2]"d" (a2), + [m4]"i" (m4)); +} + +int main (void ) +{ +S390Vector v3 = { + .q[0 ] =0x0011223344556677ull , + .q[1 ] =0x8899aabbccddeeffull , + }; +uint64_t r1 =0 ; + +/* Directly set all ignored bits to */ +vlgv (&r1, &v3, (void *)(7 | ~0xf ),ES_8 ); +check ("8 bit" , r1 ==0x77 ); +vlgv (&r1, &v3, (void *)(4 | ~0x7 ),ES_16 ); +check ("16 bit" , r1 ==0x8899 ); +vlgv (&r1, &v3, (void *)(3 | ~0x3 ),ES_32 ); +check ("32 bit" , r1 ==0xccddeeff ); +vlgv (&r1, &v3, (void *)(1 | ~0x1 ),ES_64 ); +check ("64 bit" , r1 ==0x8899aabbccddeeffull ); +check ("v3 not modified" , v3.q[0 ] ==0x0011223344556677ull && + v3.q[1 ] ==0x8899aabbccddeeffull ); + +CHECK_SIGILL (vlgv (&r1, &v3,NULL ,ES_128 )); +return 0 ; +}
Signature · 3 lines
Thanks, David / dhildenb
Re: [Qemu-devel] [PATCH RFC 2/2] tests/tcg: target/s390: Add test for VECTOR LOAD GR FROM VR ELEMENT
In reply to David HildenbrandQuoted 102 lines · David Hildenbrand
We want to make use of vectors, so use -march=z13. 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 <[email protected]>--- 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.ctests/tcg/s390x/Makefile.target +2
-1@@ -1,8 +1,9 @@ VPATH+=$(SRC_PATH)/tests/tcg/s390x -CFLAGS+=-march=zEC12 -m64 +CFLAGS+=-march=z13 -m64 -O2 TESTS+=hello-s390x TESTS+=csst TESTS+=ipm TESTS+=exrl-trt TESTS+=exrl-trtr TESTS+=pack +TESTS+=vlgvtests/tcg/s390x/helper.h new file +28
-0@@ -0,0 +1,28 @@ +#ifndef TEST_TCG_S390x_VECTOR_H +#define TEST_TCG_S390x_VECTOR_H + +#include <stdint.h> + +#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 */ tests/tcg/s390x/signal_helper.inc.c new file +29
-0@@ -0,0 +1,39 @@ +#include <stdlib.h> +#include <stdio.h> +#include <stdbool.h> +#include <signal.h> +#include <setjmp.h> +#include "helper.h" + +jmp_buf jmp_env; + +static void sig_sigill (int sig) +{ +if (sig !=SIGILL ) { +check ("Wrong signal received" , false); + } +longjmp (jmp_env,1 ); +} + +#define CHECK_SIGILL (STATEMENT ) \ +do { \ + struct sigaction act; \ + \ + act.sa_handler = sig_sigill; \ + act.sa_flags = 0; \ + if (sigaction(SIGILL, &act, NULL)) { \ + check("SIGILL handler not registered", false); \ + } \ + \ + if (setjmp(jmp_env) == 0) { \ + STATEMENT; \
tests/tcg/s390x/signal_helper.inc.c line 39
Quoted 10 lines
tests/tcg/s390x/signal_helper.inc.c new file +10
-0+check ("SIGILL not triggered" , false); \ + } \ + \ + act.sa_handler =SIG_DFL ; \ +sigemptyset (&act.sa_mask); \ + act.sa_flags =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 != SIGILL) {
check("Wrong signal received", false);
}
longjmp(jmp_env, 1);
}#define CHECK_SIGILL(STATEMENT) \
do { \
if (signal(SIGILL, sig_sigill) == SIG_ERR) { \
check("SIGILL not registered", false); \
} \
if (setjmp(jmp_env) == 0) { \
STATEMENT; \
check("SIGILL not triggered", false); \
} \
if (signal(SIGILL, SIG_DFL) == 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.
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)
Signature · 3 lines
Thanks, David / dhildenb
Re: [Qemu-devel] [PATCH RFC 2/2] tests/tcg: target/s390: Add test for VECTOR LOAD GR FROM VR ELEMENT
Quoted 153 lines · David Hildenbrand
Quoted 102 lines · David Hildenbrand
We want to make use of vectors, so use -march=z13. 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 <[email protected]>--- 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.ctests/tcg/s390x/Makefile.target +2
-1@@ -1,8 +1,9 @@ VPATH+=$(SRC_PATH)/tests/tcg/s390x -CFLAGS+=-march=zEC12 -m64 +CFLAGS+=-march=z13 -m64 -O2 TESTS+=hello-s390x TESTS+=csst TESTS+=ipm TESTS+=exrl-trt TESTS+=exrl-trtr TESTS+=pack +TESTS+=vlgvtests/tcg/s390x/helper.h new file +28
-0@@ -0,0 +1,28 @@ +#ifndef TEST_TCG_S390x_VECTOR_H +#define TEST_TCG_S390x_VECTOR_H + +#include <stdint.h> + +#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 */ tests/tcg/s390x/signal_helper.inc.c new file +29
-0@@ -0,0 +1,39 @@ +#include <stdlib.h> +#include <stdio.h> +#include <stdbool.h> +#include <signal.h> +#include <setjmp.h> +#include "helper.h" + +jmp_buf jmp_env; + +static void sig_sigill (int sig) +{ +if (sig !=SIGILL ) { +check ("Wrong signal received" , false); + } +longjmp (jmp_env,1 ); +} + +#define CHECK_SIGILL (STATEMENT ) \ +do { \ + struct sigaction act; \ + \ + act.sa_handler = sig_sigill; \ + act.sa_flags = 0; \ + if (sigaction(SIGILL, &act, NULL)) { \ + check("SIGILL handler not registered", false); \ + } \ + \ + if (setjmp(jmp_env) == 0) { \ + STATEMENT; \tests/tcg/s390x/signal_helper.inc.c line 39
Quoted 10 lines
tests/tcg/s390x/signal_helper.inc.c new file +10
-0+check ("SIGILL not triggered" , false); \ + } \ + \ + act.sa_handler =SIG_DFL ; \ +sigemptyset (&act.sa_mask); \ + act.sa_flags =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_bufjmp_env;static void
sig_sigill(int sig){ if (sig != SIGILL) { check("Wrong signal received", false); } longjmp(jmp_env, 1); }#define
CHECK_SIGILL(STATEMENT)\do { \ if (signal(SIGILL, sig_sigill) == SIG_ERR) { \ check("SIGILL not registered", false); \ } \ if (setjmp(jmp_env) == 0) { \ STATEMENT; \ check("SIGILL not triggered", false); \ } \ if (signal(SIGILL, SIG_DFL) == 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.
Quoted 6 lines
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.
Quoted 7 lines
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 Bennée
Re: [Qemu-devel] [PATCH RFC 2/2] tests/tcg: target/s390: Add test for VECTOR LOAD GR FROM VR ELEMENT
Quoted 158 lines · Alex Bennée
Quoted 155 lines · David Hildenbrand
Quoted 102 lines · David Hildenbrand
We want to make use of vectors, so use -march=z13. 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 <[email protected]>--- 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.ctests/tcg/s390x/Makefile.target +2
-1@@ -1,8 +1,9 @@ VPATH+=$(SRC_PATH)/tests/tcg/s390x -CFLAGS+=-march=zEC12 -m64 +CFLAGS+=-march=z13 -m64 -O2 TESTS+=hello-s390x TESTS+=csst TESTS+=ipm TESTS+=exrl-trt TESTS+=exrl-trtr TESTS+=pack +TESTS+=vlgvtests/tcg/s390x/helper.h new file +28
-0@@ -0,0 +1,28 @@ +#ifndef TEST_TCG_S390x_VECTOR_H +#define TEST_TCG_S390x_VECTOR_H + +#include <stdint.h> + +#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 */ tests/tcg/s390x/signal_helper.inc.c new file +29
-0@@ -0,0 +1,39 @@ +#include <stdlib.h> +#include <stdio.h> +#include <stdbool.h> +#include <signal.h> +#include <setjmp.h> +#include "helper.h" + +jmp_buf jmp_env; + +static void sig_sigill (int sig) +{ +if (sig !=SIGILL ) { +check ("Wrong signal received" , false); + } +longjmp (jmp_env,1 ); +} + +#define CHECK_SIGILL (STATEMENT ) \ +do { \ + struct sigaction act; \ + \ + act.sa_handler = sig_sigill; \ + act.sa_flags = 0; \ + if (sigaction(SIGILL, &act, NULL)) { \ + check("SIGILL handler not registered", false); \ + } \ + \ + if (setjmp(jmp_env) == 0) { \ + STATEMENT; \tests/tcg/s390x/signal_helper.inc.c line 39
Quoted 10 lines
tests/tcg/s390x/signal_helper.inc.c new file +10
-0+check ("SIGILL not triggered" , false); \ + } \ + \ + act.sa_handler =SIG_DFL ; \ +sigemptyset (&act.sa_mask); \ + act.sa_flags =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_bufjmp_env;static void
sig_sigill(int sig){ if (sig != SIGILL) { check("Wrong signal received", false); } longjmp(jmp_env, 1); }#define
CHECK_SIGILL(STATEMENT)\do { \ if (signal(SIGILL, sig_sigill) == SIG_ERR) { \ check("SIGILL not registered", false); \ } \ if (setjmp(jmp_env) == 0) { \ STATEMENT; \ check("SIGILL not triggered", false); \ } \ if (signal(SIGILL, SIG_DFL) == 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:
Quoted 8 lines
Quoted 4 lines
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:
I have to jump over the actual broken instruction and want to avoid patching it. Otherwise I'll loop forever ...
Signature · 3 lines
Thanks, David / dhildenb
Re: [Qemu-devel] [PATCH RFC 2/2] tests/tcg: target/s390: Add test for VECTOR LOAD GR FROM VR ELEMENT
Quoted 165 lines · David Hildenbrand
Quoted 164 lines · Alex Bennée
Quoted 159 lines · David Hildenbrand
Quoted 102 lines · David Hildenbrand
We want to make use of vectors, so use -march=z13. 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 <[email protected]>--- 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.ctests/tcg/s390x/Makefile.target +2
-1@@ -1,8 +1,9 @@ VPATH+=$(SRC_PATH)/tests/tcg/s390x -CFLAGS+=-march=zEC12 -m64 +CFLAGS+=-march=z13 -m64 -O2 TESTS+=hello-s390x TESTS+=csst TESTS+=ipm TESTS+=exrl-trt TESTS+=exrl-trtr TESTS+=pack +TESTS+=vlgvtests/tcg/s390x/helper.h new file +28
-0@@ -0,0 +1,28 @@ +#ifndef TEST_TCG_S390x_VECTOR_H +#define TEST_TCG_S390x_VECTOR_H + +#include <stdint.h> + +#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 */ tests/tcg/s390x/signal_helper.inc.c new file +29
-0@@ -0,0 +1,39 @@ +#include <stdlib.h> +#include <stdio.h> +#include <stdbool.h> +#include <signal.h> +#include <setjmp.h> +#include "helper.h" + +jmp_buf jmp_env; + +static void sig_sigill (int sig) +{ +if (sig !=SIGILL ) { +check ("Wrong signal received" , false); + } +longjmp (jmp_env,1 ); +} + +#define CHECK_SIGILL (STATEMENT ) \ +do { \ + struct sigaction act; \ + \ + act.sa_handler = sig_sigill; \ + act.sa_flags = 0; \ + if (sigaction(SIGILL, &act, NULL)) { \ + check("SIGILL handler not registered", false); \ + } \ + \ + if (setjmp(jmp_env) == 0) { \ + STATEMENT; \tests/tcg/s390x/signal_helper.inc.c line 39
Quoted 10 lines
tests/tcg/s390x/signal_helper.inc.c new file +10
-0+check ("SIGILL not triggered" , false); \ + } \ + \ + act.sa_handler =SIG_DFL ; \ +sigemptyset (&act.sa_mask); \ + act.sa_flags =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_bufjmp_env;static void
sig_sigill(int sig){ if (sig != SIGILL) { check("Wrong signal received", false); } longjmp(jmp_env, 1); }#define
CHECK_SIGILL(STATEMENT)\do { \ if (signal(SIGILL, sig_sigill) == SIG_ERR) { \ check("SIGILL not registered", false); \ } \ if (setjmp(jmp_env) == 0) { \ STATEMENT; \ check("SIGILL not triggered", false); \ } \ if (signal(SIGILL, SIG_DFL) == 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.
Quoted 5 lines
Quoted 2 lines
This all seems a little over-engineered. This is a single-threaded test so what's wrong with a couple of flags:
I have to jump over the actual broken instruction and want to avoid patching it. Otherwise I'll loop forever ...
So from:
Subject: [PATCH v2 6/6] tests/tcg/aarch64: userspace system register test
Date: Tue, 5 Feb 2019 19:02:24 +0000
Message-Id: <[email protected]>
Where I had to do something similar:
bool should_fail;
int should_fail_count;
int should_not_fail_count;
uintptr_t failed_pc[10];
void sigill_handler(int signo, siginfo_t *si, void *data)
{
ucontext_t *uc = (ucontext_t *)data; if (should_fail) {
should_fail_count++;
} else {
uintptr_t pc = (uintptr_t) uc->uc_mcontext.pc;
failed_pc[should_not_fail_count++] = pc;
}
uc->uc_mcontext.pc += 4;}
But that does depend on arch making pc hackable:
/* Hook in a SIGILL handler */ memset(&sa, 0, sizeof(struct sigaction)); sa.sa_flags = SA_SIGINFO; sa.sa_sigaction = &sigill_handler; sigemptyset(&sa.sa_mask);
And crucially have nice regular sized instructions. Is that not an option?
--
Alex Bennée
Re: [Qemu-devel] [PATCH RFC 2/2] tests/tcg: target/s390: Add test for VECTOR LOAD GR FROM VR ELEMENT
Quoted 1 line · Alex Bennée
And crucially have nice regular sized instructions. Is that not an option?
s390 insns are {2,4,6} bytes. I don't think that there's an easy way to pick out the hw status codes that would give the ilen of the faulting insn.
r~
Re: [Qemu-devel] [PATCH RFC 2/2] tests/tcg: target/s390: Add test for VECTOR LOAD GR FROM VR ELEMENT
Quoted 6 lines · Richard Henderson
Quoted 1 line · Alex Bennée
And crucially have nice regular sized instructions. Is that not an option?
s390 insns are {2,4,6} bytes. I don't think that there's an easy way to pick out the hw status codes that would give the ilen of the faulting insn.
We would have to read the faulting instruction to determine based on the two leftmost bits the ilen. But I don't consider that approach much cleaner compared to signal + setjmp (signal + sigsetjmp after I fixed it ;) )
Thanks for the pointer though, Alex
Quoted 3 lines
r~
Signature · 3 lines
Thanks, David / dhildenb
Re: [Qemu-devel] [PATCH RFC 2/2] tests/tcg: target/s390: Add test for VECTOR LOAD GR FROM VR ELEMENT
In reply to David HildenbrandQuoted 15 lines · David Hildenbrand
#define
CHECK_SIGILL(STATEMENT)\do { \ if (signal(SIGILL, sig_sigill) == SIG_ERR) { \ check("SIGILL not registered", false); \ } \ if (setjmp(jmp_env) == 0) { \ STATEMENT; \ check("SIGILL not triggered", false); \ } \ if (signal(SIGILL, SIG_DFL) == SIG_ERR) { \ check("SIGILL not registered", false); \ } \ } while (0)
Quoted 5 lines
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.
And this is why we use sigaction not signal.
You can set action.sa_flags |= SA_RESETHAND to avoid needing the explicit reset (or leave it clear to avoid the reset action that you are seeing).
You can set action.sa_flags |= SA_NODEFER to avoid having the signal added to the signal mask of the thread, to avoid neeing to use sigsetjmp/siglongjmp. Or you can just use sigsetjmp/siglongjmp. ;-)
r~
Re: [Qemu-devel] [PATCH RFC 2/2] tests/tcg: target/s390: Add test for VECTOR LOAD GR FROM VR ELEMENT
Quoted 22 lines · Richard Henderson
Quoted 15 lines · David Hildenbrand
#define
CHECK_SIGILL(STATEMENT)\do { \ if (signal(SIGILL, sig_sigill) == SIG_ERR) { \ check("SIGILL not registered", false); \ } \ if (setjmp(jmp_env) == 0) { \ STATEMENT; \ check("SIGILL not triggered", false); \ } \ if (signal(SIGILL, SIG_DFL) == SIG_ERR) { \ check("SIGILL not registered", false); \ } \ } while (0)Quoted 5 lines
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.
Quoted 8 lines
And this is why we use sigaction not signal.
You can set action.sa_flags |=
SA_RESETHANDto avoid needing the explicit reset (or leave it clear to avoid the reset action that you are seeing).You can set action.sa_flags |=
SA_NODEFERto avoid having the signal added to the signal mask of the thread, to avoid neeing to use sigsetjmp/siglongjmp. Or you can just use sigsetjmp/siglongjmp. ;-)
You might tell at this point that it is my first time hacking on signals and I already learned a lot, thanks ;)
I guess signal + sigsetjmp/siglongjmp is the one with the smalles LOC? Will give it a try.
BTW: I got the idea of using setjml/longjmp from tests/tcg/multiarch/linux-test.c - we're also not caring about blocked signals there, but I guess it doesn't matter as we trigger SISEGV only once.
Signature · 3 lines
Thanks, David / dhildenb
Re: [Qemu-devel] [PATCH RFC 2/2] tests/tcg: target/s390: Add test for VECTOR LOAD GR FROM VR ELEMENT
Quoted 2 lines · David Hildenbrand
I guess signal + sigsetjmp/siglongjmp is the one with the smalles LOC? Will give it a try.
I suppose it depends on how you write it.
struct sigaction sa = {
.sa_mask = SA_RESETHAND | SA_NODEFER,
.sa_handler = sig_sigill,
};
check("SIGILL not registered", sigaction(SIGILL, &sa, NULL) == 0);is 5 lines to the 6 you currently use for registering and unregistering the signal handler. ;-)
r~