From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by sourceware.org (Postfix) with ESMTP id 58561388981F for ; Thu, 31 Mar 2022 13:10:32 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 58561388981F Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.121.207.14]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id E916E23A; Thu, 31 Mar 2022 06:10:31 -0700 (PDT) Received: from localhost (e121540-lin.manchester.arm.com [10.32.98.88]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPSA id 5D33E3F66F; Thu, 31 Mar 2022 06:10:31 -0700 (PDT) From: Richard Sandiford To: Qing Zhao Mail-Followup-To: Qing Zhao , gcc-patches Paul A Clarke via , Xi Ruoyao , richard.sandiford@arm.com Cc: gcc-patches Paul A Clarke via , Xi Ruoyao Subject: Re: [patch]update the documentation for TARGET_ZERO_CALL_USED_REGS hook and add an assertion References: <01D26A82-EED3-48AA-BF56-F6300AC2C7CC@oracle.com> Date: Thu, 31 Mar 2022 14:10:30 +0100 In-Reply-To: <01D26A82-EED3-48AA-BF56-F6300AC2C7CC@oracle.com> (Qing Zhao's message of "Mon, 21 Mar 2022 21:27:54 +0000") Message-ID: User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/26.3 (gnu/linux) MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: quoted-printable X-Spam-Status: No, score=-12.3 required=5.0 tests=BAYES_00, GIT_PATCH_0, KAM_DMARC_STATUS, KAM_SHORT, SPF_HELO_NONE, SPF_PASS, TXREP, T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: gcc-patches@gcc.gnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Gcc-patches mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 31 Mar 2022 13:10:34 -0000 Qing Zhao writes: > Hi,=20 > > Per our discussion on: https://gcc.gnu.org/pipermail/gcc-patches/2022-Mar= ch/592002.html > > I come up with the following patch to: > > 1. Update the documentation for TARGET_ZERO_CALL_USED_REGS hook; > 2. Add an assertion in function.cc to make sure the actually zeroed_regs = is a subset of all call used regs; > (The reason I didn=E2=80=99t add a new parameter to TARGET_ZERO_CALL_U= SED_REGS is, I think adding the=20 > assertion in the common place function.cc is simpler to be implemente= d). Yeah, that's fair. I guess in theory, passing the parameter would allow targets to choose between two versions of zeroing the register, one with a temporary and one without. But that's a purely hypothetical situation and we could always add a parameter later if that turns out to be useful. Perhaps more realistically, there might be other uses of the hook in future that want to zero registers for different reasons, with their own rules about which registers can be zeroed. In other words, the hook is providing a general facility that happens to be useful for -fzero-call-used-regs. But again, we can deal with that if it ever happens. So I agree this is the right call, especially for stage 4. > 3. This new assertion identified a bug in i386 implementation. Fix this b= ug in i386. > > This patch is bootstrapped on both x86 and aarch64, no regression. > > Okay for commit? OK for the non-x86 bits. Thanks, Richard > thanks. > > Qing > > =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D > From 2e5bc1b25a707c6a17afbf03da2a8bec5b03454d Mon Sep 17 00:00:00 2001 > From: Qing Zhao > Date: Fri, 18 Mar 2022 20:49:56 +0000 > Subject: [PATCH] Add an assertion: the zeroed_hardregs set is a subset of= all > call used regs. > > We should make sure that the hard register set that is actually cleared by > the target hook zero_call_used_regs should be a subset of all call used > registers. > > At the same time, update documentation for the target hook > TARGET_ZERO_CALL_USED_REGS. > > This new assertion identified a bug in the i386 implemenation, which > incorrectly set the zeroed_hardregs for stack registers. Fixed this bug > in i386 implementation. > > gcc/ChangeLog: > > 2022-03-21 Qing Zhao > > * config/i386/i386.cc (zero_all_st_registers): Return the value of > num_of_st. > (ix86_zero_call_used_regs): Update zeroed_hardregs set according to > the return value of zero_all_st_registers. > * doc/tm.texi: Update the documentation of TARGET_ZERO_CALL_USED_REGS. > * function.cc (gen_call_used_regs_seq): Add an assertion. > * target.def: Update the documentation of TARGET_ZERO_CALL_USED_REGS. > --- > gcc/config/i386/i386.cc | 27 ++++++++++++++++++--------- > gcc/doc/tm.texi | 7 +++++++ > gcc/function.cc | 22 ++++++++++++++++++---- > gcc/target.def | 7 +++++++ > 4 files changed, 50 insertions(+), 13 deletions(-) > > diff --git a/gcc/config/i386/i386.cc b/gcc/config/i386/i386.cc > index 5a561966eb44..d84047a4bc1b 100644 > --- a/gcc/config/i386/i386.cc > +++ b/gcc/config/i386/i386.cc > @@ -3753,16 +3753,17 @@ zero_all_vector_registers (HARD_REG_SET need_zero= ed_hardregs) > needs to be cleared, the whole stack should be cleared. However, > x87 stack registers that hold the return value should be excluded. > x87 returns in the top (two for complex values) register, so > - num_of_st should be 7/6 when x87 returns, otherwise it will be 8. */ > + num_of_st should be 7/6 when x87 returns, otherwise it will be 8. > + return the value of num_of_st. */ >=20=20 >=20=20 > -static bool > +static int > zero_all_st_registers (HARD_REG_SET need_zeroed_hardregs) > { >=20=20 > /* If the FPU is disabled, no need to zero all st registers. */ > if (! (TARGET_80387 || TARGET_FLOAT_RETURNS_IN_80387)) > - return false; > + return 0; >=20=20 > unsigned int num_of_st =3D 0; > for (unsigned int regno =3D 0; regno < FIRST_PSEUDO_REGISTER; regno++) > @@ -3774,7 +3775,7 @@ zero_all_st_registers (HARD_REG_SET need_zeroed_har= dregs) > } >=20=20 > if (num_of_st =3D=3D 0) > - return false; > + return 0; >=20=20 > bool return_with_x87 =3D false; > return_with_x87 =3D (crtl->return_rtx > @@ -3802,7 +3803,7 @@ zero_all_st_registers (HARD_REG_SET need_zeroed_har= dregs) > insn =3D emit_insn (gen_rtx_SET (st_reg, st_reg)); > add_reg_note (insn, REG_DEAD, st_reg); > } > - return true; > + return num_of_st; > } >=20=20 >=20=20 > @@ -3851,7 +3852,7 @@ ix86_zero_call_used_regs (HARD_REG_SET need_zeroed_= hardregs) > { > HARD_REG_SET zeroed_hardregs; > bool all_sse_zeroed =3D false; > - bool all_st_zeroed =3D false; > + int all_st_zeroed_num =3D 0; > bool all_mm_zeroed =3D false; >=20=20 > CLEAR_HARD_REG_SET (zeroed_hardregs); > @@ -3881,9 +3882,17 @@ ix86_zero_call_used_regs (HARD_REG_SET need_zeroed= _hardregs) > if (!exit_with_mmx_mode) > /* x87 exit mode, we should zero all st registers together. */ > { > - all_st_zeroed =3D zero_all_st_registers (need_zeroed_hardregs); > - if (all_st_zeroed) > - SET_HARD_REG_BIT (zeroed_hardregs, FIRST_STACK_REG); > + all_st_zeroed_num =3D zero_all_st_registers (need_zeroed_hardregs); > + > + if (all_st_zeroed_num > 0) > + for (unsigned int regno =3D FIRST_STACK_REG; regno <=3D LAST_STACK_REG;= regno++) > + /* x87 stack registers that hold the return value should be excluded. > + x87 returns in the top (two for complex values) register. */ > + if (all_st_zeroed_num =3D=3D 8 > + || !((all_st_zeroed_num >=3D 6 && regno =3D=3D REGNO (crtl->retur= n_rtx)) > + || (all_st_zeroed_num =3D=3D 6 > + && (regno =3D=3D (REGNO (crtl->return_rtx) + 1))))) > + SET_HARD_REG_BIT (zeroed_hardregs, regno); > } > else > /* MMX exit mode, check whether we can zero all mm registers. */ > diff --git a/gcc/doc/tm.texi b/gcc/doc/tm.texi > index 2f92d37da8c0..c5006afc00d2 100644 > --- a/gcc/doc/tm.texi > +++ b/gcc/doc/tm.texi > @@ -12330,6 +12330,13 @@ This target hook emits instructions to zero the = subset of @var{selected_regs} > that could conceivably contain values that are useful to an attacker. > Return the set of registers that were actually cleared. >=20=20 > +For most targets, the returned set of registers is a subset of > +@var{selected_regs}, however, for some of the targets (for example MIPS), > +clearing some registers that are in the @var{selected_regs} requires > +clearing other call used registers that are not in the @var{selected_reg= s}, > +under such situation, the returned set of registers must be a subset of = all > +call used registers. > + > The default implementation uses normal move instructions to zero > all the registers in @var{selected_regs}. Define this hook if the > target has more efficient ways of zeroing certain registers, > diff --git a/gcc/function.cc b/gcc/function.cc > index d5ed51a6a663..ad0096a43eff 100644 > --- a/gcc/function.cc > +++ b/gcc/function.cc > @@ -5892,7 +5892,9 @@ gen_call_used_regs_seq (rtx_insn *ret, unsigned int= zero_regs_type) > df_simulate_one_insn_backwards (bb, ret, live_out); >=20=20 > HARD_REG_SET selected_hardregs; > + HARD_REG_SET all_call_used_regs; > CLEAR_HARD_REG_SET (selected_hardregs); > + CLEAR_HARD_REG_SET (all_call_used_regs); > for (unsigned int regno =3D 0; regno < FIRST_PSEUDO_REGISTER; regno++) > { > if (!crtl->abi->clobbers_full_reg_p (regno)) > @@ -5901,6 +5903,13 @@ gen_call_used_regs_seq (rtx_insn *ret, unsigned in= t zero_regs_type) > continue; > if (REGNO_REG_SET_P (live_out, regno)) > continue; > +#ifdef LEAF_REG_REMAP > + if (crtl->uses_only_leaf_regs && LEAF_REG_REMAP (regno) < 0) > + continue; > +#endif > + /* This is a call used register that is dead at return. */ > + SET_HARD_REG_BIT (all_call_used_regs, regno); > + > if (only_gpr > && !TEST_HARD_REG_BIT (reg_class_contents[GENERAL_REGS], regno)) > continue; > @@ -5908,10 +5917,6 @@ gen_call_used_regs_seq (rtx_insn *ret, unsigned in= t zero_regs_type) > continue; > if (only_arg && !FUNCTION_ARG_REGNO_P (regno)) > continue; > -#ifdef LEAF_REG_REMAP > - if (crtl->uses_only_leaf_regs && LEAF_REG_REMAP (regno) < 0) > - continue; > -#endif >=20=20 > /* Now this is a register that we might want to zero. */ > SET_HARD_REG_BIT (selected_hardregs, regno); > @@ -5925,6 +5930,15 @@ gen_call_used_regs_seq (rtx_insn *ret, unsigned in= t zero_regs_type) > HARD_REG_SET zeroed_hardregs; > start_sequence (); > zeroed_hardregs =3D targetm.calls.zero_call_used_regs (selected_hardre= gs); > + > + /* For most targets, the returned set of registers is a subset of > + selected_hardregs, however, for some of the targets (for example MI= PS), > + clearing some registers that are in selected_hardregs requires clea= ring > + other call used registers that are not in the selected_hardregs, un= der > + such situation, the returned set of registers must be a subset of > + all call used registers. */ > + gcc_assert (hard_reg_set_subset_p (zeroed_hardregs, all_call_used_regs= )); > + > rtx_insn *seq =3D get_insns (); > end_sequence (); > if (seq) > diff --git a/gcc/target.def b/gcc/target.def > index 72c2e1ef756c..d85adf36a391 100644 > --- a/gcc/target.def > +++ b/gcc/target.def > @@ -5122,6 +5122,13 @@ DEFHOOK > that could conceivably contain values that are useful to an attacker.\n\ > Return the set of registers that were actually cleared.\n\ > \n\ > +For most targets, the returned set of registers is a subset of\n\ > +@var{selected_regs}, however, for some of the targets (for example MIPS)= ,\n\ > +clearing some registers that are in the @var{selected_regs} requires\n\ > +clearing other call used registers that are not in the @var{selected_reg= s},\n\ > +under such situation, the returned set of registers must be a subset of = all\n\ > +call used registers.\n\ > +\n\ > The default implementation uses normal move instructions to zero\n\ > all the registers in @var{selected_regs}. Define this hook if the\n\ > target has more efficient ways of zeroing certain registers,\n\