* [PATCH v2 0/2] target/s390x: Fix R[NOX]SBG with T=1
@ 2023-03-15 23:56 Ilya Leoshkevich
2023-03-15 23:56 ` [PATCH v2 1/2] " Ilya Leoshkevich
2023-03-15 23:56 ` [PATCH v2 2/2] tests/tcg/s390x: Add rxsbg.c Ilya Leoshkevich
0 siblings, 2 replies; 5+ messages in thread
From: Ilya Leoshkevich @ 2023-03-15 23:56 UTC (permalink / raw)
To: Richard Henderson, David Hildenbrand, Thomas Huth
Cc: qemu-s390x, qemu-devel, Ilya Leoshkevich
v1: https://lists.gnu.org/archive/html/qemu-devel/2023-03/msg04493.html
v1 -> v2: Work around a clang issue (Thomas).
Add cc=0 test, use more human-friendly constants.
Hi,
This series fixes ROTATE THEN <do something with> SELECTED BITS when
test-results control is on. The problem is the incorrect translation,
which confuses the register allocator.
Patch 1 is the fix, patch 2 adds a test.
Best regards,
Ilya
Ilya Leoshkevich (2):
target/s390x: Fix R[NOX]SBG with T=1
tests/tcg/s390x: Add rxsbg.c
target/s390x/tcg/translate.c | 3 +++
tests/tcg/s390x/Makefile.target | 3 +++
tests/tcg/s390x/rxsbg.c | 46 +++++++++++++++++++++++++++++++++
3 files changed, 52 insertions(+)
create mode 100644 tests/tcg/s390x/rxsbg.c
--
2.39.2
^ permalink raw reply [flat|nested] 5+ messages in thread* [PATCH v2 1/2] target/s390x: Fix R[NOX]SBG with T=1 2023-03-15 23:56 [PATCH v2 0/2] target/s390x: Fix R[NOX]SBG with T=1 Ilya Leoshkevich @ 2023-03-15 23:56 ` Ilya Leoshkevich 2023-03-16 8:41 ` Philippe Mathieu-Daudé 2023-03-15 23:56 ` [PATCH v2 2/2] tests/tcg/s390x: Add rxsbg.c Ilya Leoshkevich 1 sibling, 1 reply; 5+ messages in thread From: Ilya Leoshkevich @ 2023-03-15 23:56 UTC (permalink / raw) To: Richard Henderson, David Hildenbrand, Thomas Huth Cc: qemu-s390x, qemu-devel, Ilya Leoshkevich RXSBG usage in the "filetests" test from the wasmtime testsuite makes tcg_reg_alloc_op() attempt to temp_load() a TEMP_VAL_DEAD temporary, causing an assertion failure: 0x01000a70: ec14 b040 3057 rxsbg %r1, %r4, 0xb0, 0x40, 0x30 OP after optimization and liveness analysis: ---- 0000000001000a70 0000000000000004 0000000000000006 rotl_i64 tmp2,r4,$0x30 dead: 1 2 pref=0xffff and_i64 tmp2,tmp2,$0x800000000000ffff dead: 1 pref=0xffff [xor_i64 tmp3,tmp3,tmp2 dead: 1 2 pref=0xffff] and_i64 cc_dst,tmp3,$0x800000000000ffff sync: 0 dead: 0 1 2 pref=0xffff mov_i64 psw_addr,$0x1000a76 sync: 0 dead: 0 1 pref=0xffff mov_i32 cc_op,$0x6 sync: 0 dead: 0 1 pref=0xffff call lookup_tb_ptr,$0x6,$1,tmp8,env dead: 1 pref=none goto_ptr tmp8 dead: 0 set_label $L0 exit_tb $0x7fffe809d183 ../tcg/tcg.c:3865: tcg fatal error The reason is that tmp3 does not have an initial value, which confuses the register allocator. This also affects the correctness of the results. Fix by assigning R1 to it. Fixes: d6c6372e186e ("target-s390: Implement R[NOX]SBG") Reviewed-by: David Hildenbrand <david@redhat.com> Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com> --- target/s390x/tcg/translate.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/target/s390x/tcg/translate.c b/target/s390x/tcg/translate.c index 14c3896d529..6dd2f41ad08 100644 --- a/target/s390x/tcg/translate.c +++ b/target/s390x/tcg/translate.c @@ -3696,10 +3696,13 @@ static DisasJumpType op_rosbg(DisasContext *s, DisasOps *o) int i4 = get_field(s, i4); int i5 = get_field(s, i5); uint64_t mask; + TCGv_i64 tmp; /* If this is a test-only form, arrange to discard the result. */ if (i3 & 0x80) { + tmp = o->out; o->out = tcg_temp_new_i64(); + tcg_gen_mov_i64(o->out, tmp); } i3 &= 63; -- 2.39.2 ^ permalink raw reply related [flat|nested] 5+ messages in thread
* Re: [PATCH v2 1/2] target/s390x: Fix R[NOX]SBG with T=1 2023-03-15 23:56 ` [PATCH v2 1/2] " Ilya Leoshkevich @ 2023-03-16 8:41 ` Philippe Mathieu-Daudé 2023-03-16 17:15 ` Ilya Leoshkevich 0 siblings, 1 reply; 5+ messages in thread From: Philippe Mathieu-Daudé @ 2023-03-16 8:41 UTC (permalink / raw) To: Ilya Leoshkevich, Richard Henderson, David Hildenbrand, Thomas Huth Cc: qemu-s390x, qemu-devel On 16/3/23 00:56, Ilya Leoshkevich wrote: > RXSBG usage in the "filetests" test from the wasmtime testsuite makes > tcg_reg_alloc_op() attempt to temp_load() a TEMP_VAL_DEAD temporary, > causing an assertion failure: > > 0x01000a70: ec14 b040 3057 rxsbg %r1, %r4, 0xb0, 0x40, 0x30 > > OP after optimization and liveness analysis: > ---- 0000000001000a70 0000000000000004 0000000000000006 > rotl_i64 tmp2,r4,$0x30 dead: 1 2 pref=0xffff > and_i64 tmp2,tmp2,$0x800000000000ffff dead: 1 pref=0xffff > [xor_i64 tmp3,tmp3,tmp2 dead: 1 2 pref=0xffff] > and_i64 cc_dst,tmp3,$0x800000000000ffff sync: 0 dead: 0 1 2 pref=0xffff > mov_i64 psw_addr,$0x1000a76 sync: 0 dead: 0 1 pref=0xffff > mov_i32 cc_op,$0x6 sync: 0 dead: 0 1 pref=0xffff > call lookup_tb_ptr,$0x6,$1,tmp8,env dead: 1 pref=none > goto_ptr tmp8 dead: 0 > set_label $L0 > exit_tb $0x7fffe809d183 > > ../tcg/tcg.c:3865: tcg fatal error > > The reason is that tmp3 does not have an initial value, which confuses > the register allocator. This also affects the correctness of the > results. > > Fix by assigning R1 to it. > > Fixes: d6c6372e186e ("target-s390: Implement R[NOX]SBG") Exposed by 3ac6f91bca..dd161de75f? 3ac6f91bca target/s390x: Drop tcg_temp_free from translate.c dd161de75f target/s390x: Remove g_out, g_out2, g_in1, g_in2 > Reviewed-by: David Hildenbrand <david@redhat.com> > Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com> > --- > target/s390x/tcg/translate.c | 3 +++ > 1 file changed, 3 insertions(+) > > diff --git a/target/s390x/tcg/translate.c b/target/s390x/tcg/translate.c > index 14c3896d529..6dd2f41ad08 100644 > --- a/target/s390x/tcg/translate.c > +++ b/target/s390x/tcg/translate.c > @@ -3696,10 +3696,13 @@ static DisasJumpType op_rosbg(DisasContext *s, DisasOps *o) > int i4 = get_field(s, i4); > int i5 = get_field(s, i5); > uint64_t mask; > + TCGv_i64 tmp; > > /* If this is a test-only form, arrange to discard the result. */ > if (i3 & 0x80) { tcg_debug_assert(o->out != NULL); ? > + tmp = o->out; > o->out = tcg_temp_new_i64(); > + tcg_gen_mov_i64(o->out, tmp); Something bugs me with this pattern but I can't say why yet :( > } > > i3 &= 63; ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH v2 1/2] target/s390x: Fix R[NOX]SBG with T=1 2023-03-16 8:41 ` Philippe Mathieu-Daudé @ 2023-03-16 17:15 ` Ilya Leoshkevich 0 siblings, 0 replies; 5+ messages in thread From: Ilya Leoshkevich @ 2023-03-16 17:15 UTC (permalink / raw) To: Philippe Mathieu-Daudé, Richard Henderson, David Hildenbrand, Thomas Huth Cc: qemu-s390x, qemu-devel On Thu, 2023-03-16 at 09:41 +0100, Philippe Mathieu-Daudé wrote: > On 16/3/23 00:56, Ilya Leoshkevich wrote: > > RXSBG usage in the "filetests" test from the wasmtime testsuite > > makes > > tcg_reg_alloc_op() attempt to temp_load() a TEMP_VAL_DEAD > > temporary, > > causing an assertion failure: > > > > 0x01000a70: ec14 b040 3057 rxsbg %r1, %r4, 0xb0, 0x40, > > 0x30 > > > > OP after optimization and liveness analysis: > > ---- 0000000001000a70 0000000000000004 0000000000000006 > > rotl_i64 tmp2,r4,$0x30 dead: 1 2 > > pref=0xffff > > and_i64 tmp2,tmp2,$0x800000000000ffff dead: 1 pref=0xffff > > [xor_i64 tmp3,tmp3,tmp2 dead: 1 2 > > pref=0xffff] > > and_i64 cc_dst,tmp3,$0x800000000000ffff sync: 0 dead: 0 1 > > 2 pref=0xffff > > mov_i64 psw_addr,$0x1000a76 sync: 0 dead: 0 1 > > pref=0xffff > > mov_i32 cc_op,$0x6 sync: 0 dead: 0 1 > > pref=0xffff > > call lookup_tb_ptr,$0x6,$1,tmp8,env dead: 1 pref=none > > goto_ptr tmp8 dead: 0 > > set_label $L0 > > exit_tb $0x7fffe809d183 > > > > ../tcg/tcg.c:3865: tcg fatal error > > > > The reason is that tmp3 does not have an initial value, which > > confuses > > the register allocator. This also affects the correctness of the > > results. > > > > Fix by assigning R1 to it. > > > > Fixes: d6c6372e186e ("target-s390: Implement R[NOX]SBG") > > Exposed by 3ac6f91bca..dd161de75f? Bisect points to: commit e2e641fa3d5e730f128562d6901dcc729c9bf8a0 Author: Richard Henderson <richard.henderson@linaro.org> Date: Sun Jan 29 14:09:00 2023 -1000 tcg: Change default temp lifetime to TEMP_TB I will mention this. > 3ac6f91bca target/s390x: Drop tcg_temp_free from translate.c > dd161de75f target/s390x: Remove g_out, g_out2, g_in1, g_in2 > > > Reviewed-by: David Hildenbrand <david@redhat.com> > > Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com> > > --- > > target/s390x/tcg/translate.c | 3 +++ > > 1 file changed, 3 insertions(+) > > > > diff --git a/target/s390x/tcg/translate.c > > b/target/s390x/tcg/translate.c > > index 14c3896d529..6dd2f41ad08 100644 > > --- a/target/s390x/tcg/translate.c > > +++ b/target/s390x/tcg/translate.c > > @@ -3696,10 +3696,13 @@ static DisasJumpType op_rosbg(DisasContext > > *s, DisasOps *o) > > int i4 = get_field(s, i4); > > int i5 = get_field(s, i5); > > uint64_t mask; > > + TCGv_i64 tmp; > > > > /* If this is a test-only form, arrange to discard the > > result. */ > > if (i3 & 0x80) { > > tcg_debug_assert(o->out != NULL); ? Ok, I will add this. > > > + tmp = o->out; > > o->out = tcg_temp_new_i64(); > > + tcg_gen_mov_i64(o->out, tmp); > > Something bugs me with this pattern but I can't say why yet :( Please let me know once you come up with something. I will do s/tmp/orig_out/ send a v3 in the meantime. > > } > > > > i3 &= 63; ^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH v2 2/2] tests/tcg/s390x: Add rxsbg.c 2023-03-15 23:56 [PATCH v2 0/2] target/s390x: Fix R[NOX]SBG with T=1 Ilya Leoshkevich 2023-03-15 23:56 ` [PATCH v2 1/2] " Ilya Leoshkevich @ 2023-03-15 23:56 ` Ilya Leoshkevich 1 sibling, 0 replies; 5+ messages in thread From: Ilya Leoshkevich @ 2023-03-15 23:56 UTC (permalink / raw) To: Richard Henderson, David Hildenbrand, Thomas Huth Cc: qemu-s390x, qemu-devel, Ilya Leoshkevich Add a small test for RXSBG with T=1 to prevent regressions. Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com> --- tests/tcg/s390x/Makefile.target | 3 +++ tests/tcg/s390x/rxsbg.c | 46 +++++++++++++++++++++++++++++++++ 2 files changed, 49 insertions(+) create mode 100644 tests/tcg/s390x/rxsbg.c diff --git a/tests/tcg/s390x/Makefile.target b/tests/tcg/s390x/Makefile.target index cf93b966862..3c940ac952e 100644 --- a/tests/tcg/s390x/Makefile.target +++ b/tests/tcg/s390x/Makefile.target @@ -29,10 +29,13 @@ TESTS+=clst TESTS+=long-double TESTS+=cdsg TESTS+=chrl +TESTS+=rxsbg cdsg: CFLAGS+=-pthread cdsg: LDFLAGS+=-pthread +rxsbg: CFLAGS+=-O2 + Z13_TESTS=vistr $(Z13_TESTS): CFLAGS+=-march=z13 -O2 TESTS+=$(Z13_TESTS) diff --git a/tests/tcg/s390x/rxsbg.c b/tests/tcg/s390x/rxsbg.c new file mode 100644 index 00000000000..4b155db304e --- /dev/null +++ b/tests/tcg/s390x/rxsbg.c @@ -0,0 +1,46 @@ +/* + * Test the RXSBG instruction. + * + * SPDX-License-Identifier: GPL-2.0-or-later + */ +#include <assert.h> +#include <stdlib.h> + +static inline __attribute__((__always_inline__)) void +rxsbg(unsigned long *r1, unsigned long r2, int i3, int i4, int i5, int *cc) +{ + asm("rxsbg %[r1],%[r2],%[i3],%[i4],%[i5]\n" + "ipm %[cc]" + : [r1] "+r" (*r1), [cc] "=r" (*cc) + : [r2] "r" (r2) , [i3] "i" (i3) , [i4] "i" (i4) , [i5] "i" (i5) + : "cc"); + *cc = (*cc >> 28) & 3; +} + +void test_cc0(void) +{ + unsigned long r1 = 6; + int cc; + + rxsbg(&r1, 3, 61 | 0x80, 62, 1, &cc); + assert(r1 == 6); + assert(cc == 0); +} + +void test_cc1(void) +{ + unsigned long r1 = 2; + int cc; + + rxsbg(&r1, 3, 61 | 0x80, 62, 1, &cc); + assert(r1 == 2); + assert(cc == 1); +} + +int main(void) +{ + test_cc0(); + test_cc1(); + + return EXIT_SUCCESS; +} -- 2.39.2 ^ permalink raw reply related [flat|nested] 5+ messages in thread
end of thread, other threads:[~2023-03-16 17:16 UTC | newest] Thread overview: 5+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2023-03-15 23:56 [PATCH v2 0/2] target/s390x: Fix R[NOX]SBG with T=1 Ilya Leoshkevich 2023-03-15 23:56 ` [PATCH v2 1/2] " Ilya Leoshkevich 2023-03-16 8:41 ` Philippe Mathieu-Daudé 2023-03-16 17:15 ` Ilya Leoshkevich 2023-03-15 23:56 ` [PATCH v2 2/2] tests/tcg/s390x: Add rxsbg.c Ilya Leoshkevich
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).