| Message ID | 20180820181900.GA27620@intel.com |
|---|---|
| State | New |
| Headers | show |
| Series | x86: Always update EH return address in word_mode | expand |
On Mon, Aug 20, 2018 at 8:19 PM, H.J. Lu <hongjiu.lu@intel.com> wrote: > On x86, return address is always popped in word_mode. eh_return needs > to put EH return address in word_mode on stack. > > Tested on x86-64 with x32. OK for trunk and release branches? OK. Perhaps the testcase should go into g++.dg/torture, since it is sensitive to optimization level? Thanks, Uros. > Thanks. > > H.J. > --- > gcc/ > > PR target/87014 > * config/i386/i386.md (eh_return): Always update EH return > address in word_mode. > > gcc/testsuite/ > > PR target/87014 > * g++.dg/pr87014.C: New file. > --- > gcc/config/i386/i386.md | 5 ++++- > gcc/testsuite/g++.dg/pr87014.C | 38 ++++++++++++++++++++++++++++++++++ > 2 files changed, 42 insertions(+), 1 deletion(-) > create mode 100644 gcc/testsuite/g++.dg/pr87014.C > > diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md > index 918241d953a..71faa218ffa 100644 > --- a/gcc/config/i386/i386.md > +++ b/gcc/config/i386/i386.md > @@ -13612,7 +13612,10 @@ > stack address we wish to restore. */ > tmp = gen_rtx_PLUS (Pmode, arg_pointer_rtx, sa); > tmp = plus_constant (Pmode, tmp, -UNITS_PER_WORD); > - tmp = gen_rtx_MEM (Pmode, tmp); > + /* Return address is always in word_mode. */ > + tmp = gen_rtx_MEM (word_mode, tmp); > + if (GET_MODE (ra) != word_mode) > + ra = convert_to_mode (word_mode, ra, 1); > emit_move_insn (tmp, ra); > > emit_jump_insn (gen_eh_return_internal ()); > diff --git a/gcc/testsuite/g++.dg/pr87014.C b/gcc/testsuite/g++.dg/pr87014.C > new file mode 100644 > index 00000000000..060ec948c90 > --- /dev/null > +++ b/gcc/testsuite/g++.dg/pr87014.C > @@ -0,0 +1,38 @@ > +// { dg-do run } > +// { dg-options "-O0" } > + > +void > +fillstack () > +{ > + long long foo[] = > + { > + -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, > + }; > +} > + > +void > +f (long long=-1,long long=-1,long long=-1,long long=-1, > + long long=-1,long long=-1,long long arg7_on_stack=-1) > +{ > + throw 0; > +} > + > +void > +g() > +{ > + try > + { > + f (); > + } > + catch (int) > + { > + } > +} > + > +int > +main() > +{ > + fillstack (); > + g (); > + return 0; > +} > -- > 2.17.1 >
On Mon, Aug 20, 2018 at 11:40 AM, Uros Bizjak <ubizjak@gmail.com> wrote: > On Mon, Aug 20, 2018 at 8:19 PM, H.J. Lu <hongjiu.lu@intel.com> wrote: >> On x86, return address is always popped in word_mode. eh_return needs >> to put EH return address in word_mode on stack. >> >> Tested on x86-64 with x32. OK for trunk and release branches? > > OK. Perhaps the testcase should go into g++.dg/torture, since it is > sensitive to optimization level? Sure. This is the patch I am checking in. Thanks.
diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md index 918241d953a..71faa218ffa 100644 --- a/gcc/config/i386/i386.md +++ b/gcc/config/i386/i386.md @@ -13612,7 +13612,10 @@ stack address we wish to restore. */ tmp = gen_rtx_PLUS (Pmode, arg_pointer_rtx, sa); tmp = plus_constant (Pmode, tmp, -UNITS_PER_WORD); - tmp = gen_rtx_MEM (Pmode, tmp); + /* Return address is always in word_mode. */ + tmp = gen_rtx_MEM (word_mode, tmp); + if (GET_MODE (ra) != word_mode) + ra = convert_to_mode (word_mode, ra, 1); emit_move_insn (tmp, ra); emit_jump_insn (gen_eh_return_internal ()); diff --git a/gcc/testsuite/g++.dg/pr87014.C b/gcc/testsuite/g++.dg/pr87014.C new file mode 100644 index 00000000000..060ec948c90 --- /dev/null +++ b/gcc/testsuite/g++.dg/pr87014.C @@ -0,0 +1,38 @@ +// { dg-do run } +// { dg-options "-O0" } + +void +fillstack () +{ + long long foo[] = + { + -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, + }; +} + +void +f (long long=-1,long long=-1,long long=-1,long long=-1, + long long=-1,long long=-1,long long arg7_on_stack=-1) +{ + throw 0; +} + +void +g() +{ + try + { + f (); + } + catch (int) + { + } +} + +int +main() +{ + fillstack (); + g (); + return 0; +}