mips: pass code of conditional trap
PATCH 1/1 mips: pass code of conditional trap
Linux and We use the code of conditional trap instructions to emit signals other than simple SIGTRAP. Currently, code 6 (overflow), 7 (div by zero) are supported. It means that if code 7 is used with a conditional trap instruction, a SIGFPE instead of SIGTRAP will emit.
But when gen_trap we didn't pass the code as we use generate_exception, which has no info about the code. Let's introduce a new function generate_exception_code for it.
--- target/mips/tcg/translate.c | 8 +++++++- target/mips/tcg/translate.h | 1 + 2 files changed, 8 insertions(+), 1 deletion(-)
target/mips/tcg/translate.c +7 -1
@@ -1353,6 +1353,12 @@ void generate_exception(DisasContext *ctx, int excp)gen_helper_raise_exception (tcg_env,tcg_constant_i32 (excp)); } +void generate_exception_with_code (DisasContext *ctx,int excp,int code) +{ +gen_helper_raise_exception_err (tcg_env,tcg_constant_i32 (excp), +tcg_constant_i32 (code)); +} +void generate_exception_end (DisasContext *ctx,int excp) {generate_exception_err (ctx, excp,0 );
@@ -4553,7 +4559,7 @@ static void gen_trap(DisasContext *ctx, uint32_t opc,if (ctx->hflags != ctx->saved_hflags) {tcg_gen_movi_i32 (hflags, ctx->hflags); } -generate_exception (ctx,EXCP_TRAP ); +generate_exception_with_code (ctx,EXCP_TRAP , code);gen_set_label (l1); } }
target/mips/tcg/translate.h +1 -0
@@ -134,6 +134,7 @@ enum { }while (0 ) void generate_exception(DisasContext *ctx, int excp); +void generate_exception_with_code (DisasContext *ctx,int excp,int code);void generate_exception_err (DisasContext *ctx,int excp,int err);void generate_exception_end (DisasContext *ctx,int excp);void generate_exception_break (DisasContext *ctx,int code);
Signature · 1 line
2.39.3 (Apple Git-146)
Quoted 8 lines · YunQiang Su
Linux and We use the code of conditional trap instructions to emit signals other than simple SIGTRAP. Currently, code 6 (overflow), 7 (div by zero) are supported. It means that if code 7 is used with a conditional trap instruction, a SIGFPE instead of SIGTRAP will emit.
But when
gen_trapwe didn't pass the code as we usegenerate_exception, which has no info about the code. Let's introduce a new functiongenerate_exception_codefor it.
I haven't touched this stuff for ages, but AFAICT the code is already
passed where applicable via the environment for `do_tr_or_bp' to handle, so I can't understand why your change is needed.
What problem are you trying to solve?
Maciej
Quoted 8 lines · Maciej W. Rozycki <[email protected]> 于2024年6月21日周五 08:41写道:
Quoted 5 lines · YunQiang Su
Linux and We use the code of conditional trap instructions to emit signals other than simple SIGTRAP. Currently, code 6 (overflow), 7 (div by zero) are supported. It means that if code 7 is used with a conditional trap instruction, a SIGFPE instead of SIGTRAP will emit.
Quoted 8 lines
Quoted 3 lines
But when
gen_trapwe didn't pass the code as we usegenerate_exception, which has no info about the code. Let's introduce a new functiongenerate_exception_codefor it.I haven't touched this stuff for ages, but AFAICT the code is already
passed where applicable via the environment for `do_tr_or_bp' to handle, so I can't understand why your change is needed.
The error_code in env is always zero, as we need to set it here.
Quoted 2 lines
What problem are you trying to solve?
See the talk in GCC mailing list about testsuite/ubsan/overflow-div-3.c Qemu emits SIGTRAP instead of SIGFPE, due to it didn't initialize the code of conditional trap to env.
Quoted 1 line
Maciej
Quoted 6 lines · YunQiang Su
Quoted 4 lines
I haven't touched this stuff for ages, but AFAICT the code is already
passed where applicable via the environment for `do_tr_or_bp' to handle, so I can't understand why your change is needed.The
error_codein env is always zero, as we need to set it here.
There is code to set it there already, so when submitting a fix you need
to investigate why it doesn't work and describe in the change description.
Quoted 6 lines
Quoted 2 lines
What problem are you trying to solve?
See the talk in GCC mailing list about
testsuite/ubsan/overflow-div-3.cQemu emits SIGTRAP instead of SIGFPE, due to it didn't initialize the code of conditional trap to env.
Self-contained information about the reproducer needs to be included in
the change description. A general statement such as "this and that does not work" or referring to another mailing list is not sufficient.
Maciej
Quoted 16 lines · Maciej W. Rozycki <[email protected]> 于2024年6月21日周五 20:02写道:
Quoted 6 lines · YunQiang Su
Quoted 4 lines
I haven't touched this stuff for ages, but AFAICT the code is already
passed where applicable via the environment for `do_tr_or_bp' to handle, so I can't understand why your change is needed.The
error_codein env is always zero, as we need to set it here.There is code to set it there already, so when submitting a fix you need
to investigate why it doesn't work and describe in the change description.Quoted 3 lines
Quoted 2 lines
What problem are you trying to solve?
Quoted 8 lines
Quoted 3 lines
See the talk in GCC mailing list about
testsuite/ubsan/overflow-div-3.cQemu emits SIGTRAP instead of SIGFPE, due to it didn't initialize the code of conditional trap to env.Self-contained information about the reproducer needs to be included in
the change description. A general statement such as "this and that does not work" or referring to another mailing list is not sufficient.
I am trying to fix the problem like this
gcc/testsuite/gcc.c-torture/execute/20101011-1.c
void
sigfpe (int signum __attribute__ ((unused)))
{
exit (0);
}
int
main ()
{
#if DO_TEST
signal (SIGFPE, sigfpe);
k = i / j;
abort ();
#else
exit (0);
#endif
}
Quoted 1 line
Maciej
Quoted 19 lines · YunQiang Su
Quoted 9 lines
Quoted 5 lines
Quoted 1 line
What problem are you trying to solve?
See the talk in GCC mailing list about
testsuite/ubsan/overflow-div-3.cQemu emits SIGTRAP instead of SIGFPE, due to it didn't initialize the code of conditional trap to env.Self-contained information about the reproducer needs to be included in
the change description. A general statement such as "this and that does not work" or referring to another mailing list is not sufficient.I am trying to fix the problem like this
gcc/testsuite/gcc.c-torture/execute/20101011-1.cvoid
sigfpe (int signum__attribute__((unused)))
{
exit (0);
}
Quoted 7 lines
int
main ()
{
#ifDO_TEST
signal (SIGFPE, sigfpe);
k = i / j;
abort ();
I gather QEMU in the user emulation mode fails to interpret the embedded
break or trap code of a `teq $2,$0,7' or similar instruction produced by the compiler as a part of the integer division machine code sequence for the source code quoted above, and consequently issues the wrong signal to the program emulated. Is that a correct statement of the problem?
If so, then it has to be stated in the change description. Then Richard
has correctly identified the fix to make.
Maciej
target/mips/tcg/translate.c static void gen_trap(DisasContext *ctx, uint32_t opc, · line 4565
Quoted 9 lines · YunQiang Su
target/mips/tcg/translate.c +1
-1@@ -4553,7 +4559,7 @@ static void gen_trap(DisasContext *ctx, uint32_t opc,if (ctx->hflags != ctx->saved_hflags) {tcg_gen_movi_i32 (hflags, ctx->hflags); } -generate_exception (ctx,EXCP_TRAP ); +generate_exception_with_code (ctx,EXCP_TRAP , code);gen_set_label (l1); } }
There are two instances within gen_trap, one of which *does* store into error_code, but that gets overwritten by do_raise_exception_err.
Search for EXCP_TRAP.
r~
Quoted 15 lines · Richard Henderson <[email protected]> 于2024年6月21日周五 12:21写道:
target/mips/tcg/translate.c static void gen_trap(DisasContext *ctx, uint32_t opc, · line 4565
Quoted 9 lines · YunQiang Su
target/mips/tcg/translate.c +1
-1@@ -4553,7 +4559,7 @@ static void gen_trap(DisasContext *ctx, uint32_t opc,if (ctx->hflags != ctx->saved_hflags) {tcg_gen_movi_i32 (hflags, ctx->hflags); } -generate_exception (ctx,EXCP_TRAP ); +generate_exception_with_code (ctx,EXCP_TRAP , code);gen_set_label (l1); } }There are two instances within
gen_trap, one of which *does* store intoerror_code, but that gets overwritten bydo_raise_exception_err.
Ohh, yes. There is another generate_exception_end if cond == 0.
Quoted 4 lines
Search for
EXCP_TRAP.r~