target/riscv: zimop and zcmop extension for riscv
PATCH 1/3 target/riscv: zimop and zcmop extension for riscv
zimop stands for may be operations. zcmop stands for compressed may be operations. For some RISC-V CPU extension, once compiled into the binary are part of generated code which can't be gated behind a probe of whether an instruction set is supported or not. One such example is zicfiss [1] extension where shadow stack push and shadow stack pop and check will be part of every function body. Thus binaries compiled with such extensions need to run in following scenarios
- On machines where extension is present and enabled
- On machines where extension is present and disabled
- On machines where extension is not present/implemented.
zimop (for 32bit instructions) and zcmop (for compressed) were devised and defined [2] to support such future (like zicfiss) CPU extensions where zimops and zcmops provide a base non-faulting behavior for codepoints that may claimed by future ISA extensions. Minimally, any CPU implementation wanting to have binary compatibility with such binaries only has to implement zimop and zcmop. Furthermore, this allows per-task optin for software where user has the option to enable the feature on per-task basis.
zimop are defined to write zero to rd. zcmop are defined to *not* write to any register.
[1] - https://github.com/riscv/riscv-cfi/blob/main/src/cfi_backward.adoc
[2] - https://github.com/riscv/riscv-isa-manual/blob/main/src/zimop.adoc
--- target/riscv/cpu.c | 2 ++ target/riscv/cpu_cfg.h | 1 + 2 files changed, 3 insertions(+)
target/riscv/cpu.c +2 -0
@@ -113,6 +113,7 @@ const RISCVIsaExtData isa_edata_arr[] = {ISA_EXT_DATA_ENTRY (zihintntl,PRIV_VERSION_1_10_0 , ext_zihintntl),ISA_EXT_DATA_ENTRY (zihintpause,PRIV_VERSION_1_10_0 , ext_zihintpause),ISA_EXT_DATA_ENTRY (zihpm,PRIV_VERSION_1_12_0 , ext_zihpm), +ISA_EXT_DATA_ENTRY (zimops,PRIV_VERSION_1_12_0 , ext_zimops),ISA_EXT_DATA_ENTRY (zmmul,PRIV_VERSION_1_12_0 , ext_zmmul),ISA_EXT_DATA_ENTRY (za64rs,PRIV_VERSION_1_12_0 , has_priv_1_11), ISA_EXT_DATA_ENTRY(zaamo, PRIV_VERSION_1_12_0, ext_zaamo),
@@ -2273,6 +2274,7 @@ static Property riscv_cpu_properties[] = {* it with -x and default to 'false'. */ DEFINE_PROP_BOOL ("x-misa-w" ,RISCVCPU , cfg.misa_w, false), +DEFINE_PROP_BOOL ("zimops" ,RISCVCPU , cfg.ext_zimops, true), DEFINE_PROP_END_OF_LIST(), };
target/riscv/cpu_cfg.h +1 -0
@@ -124,6 +124,7 @@ struct RISCVCPUConfig {uint32_t mvendorid;uint64_t marchid;uint64_t mimpid; +bool ext_zimops;/* Named features */ bool ext_svade;
Signature · 20 lines
2.34.1 From 4d15b0e0037f444eb75e60b398e19dcf476f07d4 Mon Sep 17 00:00:00 2001 From: Deepak Gupta <[email protected]> Date: Fri, 28 Jun 2024 00:13:58 -0700 Subject: [PATCH 2/3] target/riscv: zimop instruction encoding and its implementation This patch adds assigned codepoints for decoder for 32bit instructions and provide implementation for instruction. If extension is present, then moves 0 to `rd`. Signed-off-by: Deepak Gupta <[email protected]> --- target/riscv/insn32.decode | 15 +++++++ target/riscv/insn_trans/trans_zimops.c.inc | 50 ++++++++++++++++++++++ target/riscv/translate.c | 3 ++ 3 files changed, 68 insertions(+) create mode 100644 target/riscv/insn_trans/trans_zimops.c.inc
target/riscv/insn32.decode +15 -0
@@ -167,6 +167,21 @@ csrrwi ............ ..... 101 ..... 1110011 @csr csrrsi ............ ..... 110 ..... 1110011 @csr csrrci ............ ..... 111 ..... 1110011 @csr +# zimops (unpriv integer may be operations) instructions with system opcode +# zimops_r and zimops_rr are two code points assigned to zimops +# Any new extension that claims zimops encoding should be placed above mop.r +# and mop.rr + +# mop.r +{ + zimops_r 1-00--0 111-- ----- 100 ..... 1110011 %rd +} + +# mop.rr +{ + zimops_rr 1-00--1 ----- ----- 100 ..... 1110011 %rd +} + # *** RV64I Base Instruction Set (in addition to RV32I) *** lwu ............ ..... 110 ..... 0000011 @i ld ............ ..... 011 ..... 0000011 @i
target/riscv/insn_trans/trans_zimops.c.inc new file +50 -0
50 lines not on this page · raw message
target/riscv/translate.c +3 -0
@@ -1115,6 +1115,9 @@ static uint32_t opcode_at(DisasContextBase *dcbase, target_ulong pc)/* Include decoders for factored-out extensions */ #include "decode-XVentanaCondOps.c.inc" +/* Include decoder for zimop */ +#include "insn_trans/trans_zimops.c.inc" +/* The specification allows for longer insns, but not supported by qemu. */ #define MAX_INSN_LEN 4
Signature · 21 lines
2.34.1 From dc5d2c62475ac7a9cf42b9ed1398961505b8b91d Mon Sep 17 00:00:00 2001 From: Deepak Gupta <[email protected]> Date: Tue, 2 Jan 2024 15:46:13 -0800 Subject: [PATCH 3/3] target/riscv: Introduce `compressed zimop` aka `zcmop` Analogous to zimop, there are 8 encodings carved out of illegal space encodings (c.lui xn, 0) in compressed instructions which are defined to be zcmops short for compressed may be operations. Unlike zimops (which write 0 to rd), zcmops don't actually write anything to any register. Their encodings allow future extensions to define them to read register x[n]. Signed-off-by: Deepak Gupta <[email protected]> --- target/riscv/insn16.decode | 6 ++++++ target/riscv/insn_trans/trans_zimops.c.inc | 11 +++++++++++ 2 files changed, 17 insertions(+)
target/riscv/insn16.decode +6 -0
@@ -32,6 +32,7 @@ %uimm_cl_w 5:1 10:3 6:1 !function=ex_shift_2 %imm_cb 12:s1 5:2 2:1 10:2 3:2 !function=ex_shift_1 %imm_cj 12:s1 8:1 9:2 6:1 7:1 2:1 11:1 3:3 !function=ex_shift_1 +%zcmop_n 8:3 %shlimm_6bit 12:1 2:5 !function=ex_rvc_shiftli %shrimm_6bit 12:1 2:5 !function=ex_rvc_shiftri
@@ -66,6 +67,8 @@ &cmpp urlist spimm &cmjt index +&cmop zcmop_n + # Formats 16: @cr .... ..... ..... .. &r rs2=%rs2_5 rs1=%rd %rd @ci ... . ..... ..... .. &i imm=%imm_ci rs1=%rd %rd
@@ -109,6 +112,8 @@ @cm_mv ... ... ... .. ... .. &r2_s rs2=%r2s rs1=%r1s @cm_jt ... ... ........ .. &cmjt %index +@c_mop ... . ..... ..... .. &cmop %zcmop_n + # *** RV32/64C Standard Extension (Quadrant 0) *** { # Opcode of all zeros is illegal; rd != 0, nzuimm == 0 is reserved.
@@ -140,6 +145,7 @@ sw 110 ... ... .. ... 00 @cs_w addi 000 . ..... ..... 01 @ci addi 010 . ..... ..... 01 @c_li { + zcmops 011 0 0...1 00000 01 @c_mop # zcmop carving out of illegal c.lui xn,0 space illegal 011 0 ----- 00000 01 # c.addi16sp and c.lui, RES nzimm=0 addi 011 . 00010 ..... 01 @c_addi16sp lui 011 . ..... ..... 01 @c_lui
target/riscv/insn_trans/trans_zimops.c.inc +11 -0
@@ -48,3 +48,14 @@ static bool trans_zimops_rr(DisasContext *ctx, arg_zimops_r * a) gen_set_gpr(ctx, a->rd, dest); return true; } + +static bool trans_zcmops(DisasContext *ctx, arg_zcmops * a) +{ + /* zimops not implemented, return false */ + if (!ctx->cfg_ptr->ext_zimops) { + gen_exception_illegal(ctx); + return false; + } + + return true; +}
Signature · 1 line
2.34.1
Please ignore this one.
Wrong command and all patches came out as one.
Sorry about that.
Quoted 267 lines · Deepak Gupta
zimopstands formay be operations.zcmopstands for compressedmay be operations. For some RISC-V CPU extension, once compiled into the binary are part of generated code which can't be gated behind a probe of whether an instruction set is supported or not. One such example iszicfiss[1] extension whereshadow stack pushandshadow stack pop and checkwill be part of every function body. Thus binaries compiled with such extensions need to run in following scenarios- On machines where extension is present and enabled
- On machines where extension is present and disabled
- On machines where extension is not present/implemented.
zimop(for 32bit instructions) andzcmop(for compressed) were devised and defined [2] to support such future (like zicfiss) CPU extensions where zimops and zcmops provide a base non-faulting behavior for codepoints that may claimed by future ISA extensions. Minimally, any CPU implementation wanting to have binary compatibility with such binaries only has to implementzimop and zcmop. Furthermore, this allows per-task optin for software where user has the option to enable the feature on per-task basis.
zimopare defined to write zero tord.zcmopare defined to *not* write to any register.[1] - https://github.com/riscv/riscv-cfi/blob/main/src/cfi_backward.adoc
[2] - https://github.com/riscv/riscv-isa-manual/blob/main/src/zimop.adocSigned-off-by Deepak Gupta <[email protected]>--- target/riscv/cpu.c | 2 ++ target/riscv/cpu_cfg.h | 1 + 2 files changed, 3 insertions(+)target/riscv/cpu.c +2
-0@@ -113,6 +113,7 @@ const RISCVIsaExtData isa_edata_arr[] = {ISA_EXT_DATA_ENTRY (zihintntl,PRIV_VERSION_1_10_0 , ext_zihintntl),ISA_EXT_DATA_ENTRY (zihintpause,PRIV_VERSION_1_10_0 , ext_zihintpause),ISA_EXT_DATA_ENTRY (zihpm,PRIV_VERSION_1_12_0 , ext_zihpm), +ISA_EXT_DATA_ENTRY (zimops,PRIV_VERSION_1_12_0 , ext_zimops),ISA_EXT_DATA_ENTRY (zmmul,PRIV_VERSION_1_12_0 , ext_zmmul),ISA_EXT_DATA_ENTRY (za64rs,PRIV_VERSION_1_12_0 , has_priv_1_11), ISA_EXT_DATA_ENTRY(zaamo, PRIV_VERSION_1_12_0, ext_zaamo),@@ -2273,6 +2274,7 @@ static Property riscv_cpu_properties[] = {* it with -x and default to 'false'. */ DEFINE_PROP_BOOL ("x-misa-w" ,RISCVCPU , cfg.misa_w, false), +DEFINE_PROP_BOOL ("zimops" ,RISCVCPU , cfg.ext_zimops, true), DEFINE_PROP_END_OF_LIST(), };target/riscv/cpu_cfg.h +1
-1@@ -124,6 +124,7 @@ struct RISCVCPUConfig {uint32_t mvendorid;uint64_t marchid;uint64_t mimpid; +bool ext_zimops;/* Named features */ bool ext_svade; --2.34.1
From
4d15b0e0037f444eb75e60b398e19dcf476f07d4Mon Sep 17 00:00:00 2001
From: Deepak Gupta <[email protected]>
Date: Fri, 28 Jun 2024 00:13:58 -0700
Subject: [PATCH 2/3] target/riscv: zimop instruction encoding and its
implementationThis patch adds assigned codepoints for decoder for 32bit instructions and provide implementation for instruction. If extension is present, then moves 0 to
rd.Signed-off-by Deepak Gupta <[email protected]>--- target/riscv/insn32.decode | 15 +++++++ target/riscv/insn_trans/trans_zimops.c.inc | 50 ++++++++++++++++++++++ target/riscv/translate.c | 3 ++ 3 files changed, 68 insertions(+) create mode 100644 target/riscv/insn_trans/trans_zimops.c.inctarget/riscv/insn32.decode +15
-0@@ -167,6 +167,21 @@ csrrwi ............ ..... 101 ..... 1110011 @csr csrrsi ............ ..... 110 ..... 1110011 @csr csrrci ............ ..... 111 ..... 1110011 @csr +# zimops (unpriv integer may be operations) instructions with system opcode +# zimops_r and zimops_rr are two code points assigned to zimops +# Any new extension that claims zimops encoding should be placed above mop.r +# and mop.rr + +# mop.r +{ + zimops_r 1-00--0 111-- ----- 100 ..... 1110011 %rd +} + +# mop.rr +{ + zimops_rr 1-00--1 ----- ----- 100 ..... 1110011 %rd +} + # *** RV64I Base Instruction Set (in addition to RV32I) *** lwu ............ ..... 110 ..... 0000011 @i ld ............ ..... 011 ..... 0000011 @itarget/riscv/insn_trans/trans_zimops.c.inc new file +50
-0@@ -0,0 +1,50 @@ +/* + * RISC-V translation routines for the Control-Flow Integrity Extension + * + * Copyright (c) 2024 Rivos Inc. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2 or later, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program. If not, see <http://www.gnu.org/licenses/>. + */ +static bool trans_zimops_r(DisasContext *ctx, arg_zimops_r * a) +{ + /* zimops not implemented, raise illegal instruction & return true */ + if (!ctx->cfg_ptr->ext_zimops) { + gen_exception_illegal(ctx); + return true; + } + /* + * zimops implemented, simply grab destination and mov zero. + * return true + */ + TCGv dest = dest_gpr(ctx, a->rd); + dest = tcg_constant_tl(0); + gen_set_gpr(ctx, a->rd, dest); + return true; +} + +static bool trans_zimops_rr(DisasContext *ctx, arg_zimops_r * a) +{ + /* zimops not implemented, raise illegal instruction & return true */ + if (!ctx->cfg_ptr->ext_zimops) { + gen_exception_illegal(ctx); + return true; + } + /* + * zimops implemented, simply grab destination and mov zero. + * return true + */ + TCGv dest = dest_gpr(ctx, a->rd); + dest = tcg_constant_tl(0); + gen_set_gpr(ctx, a->rd, dest); + return true; +}target/riscv/translate.c +3
-1@@ -1115,6 +1115,9 @@ static uint32_t opcode_at(DisasContextBase *dcbase, target_ulong pc)/* Include decoders for factored-out extensions */ #include "decode-XVentanaCondOps.c.inc" +/* Include decoder for zimop */ +#include "insn_trans/trans_zimops.c.inc" +/* The specification allows for longer insns, but not supported by qemu. */ #define MAX_INSN_LEN 4 --2.34.1
From
dc5d2c62475ac7a9cf42b9ed1398961505b8b91dMon Sep 17 00:00:00 2001
From: Deepak Gupta <[email protected]>
Date: Tue, 2 Jan 2024 15:46:13 -0800
Subject: [PATCH 3/3] target/riscv: Introducecompressed zimopakazcmopAnalogous to zimop, there are 8 encodings carved out of illegal space encodings (c.lui xn, 0) in compressed instructions which are defined to be zcmops short for compressed may be operations.
Unlike zimops (which write 0 to rd), zcmops don't actually write anything to any register. Their encodings allow future extensions to define them to read register x[n].
Signed-off-by Deepak Gupta <[email protected]>--- target/riscv/insn16.decode | 6 ++++++ target/riscv/insn_trans/trans_zimops.c.inc | 11 +++++++++++ 2 files changed, 17 insertions(+)target/riscv/insn16.decode +6
-0@@ -32,6 +32,7 @@ %uimm_cl_w 5:1 10:3 6:1 !function=ex_shift_2 %imm_cb 12:s1 5:2 2:1 10:2 3:2 !function=ex_shift_1 %imm_cj 12:s1 8:1 9:2 6:1 7:1 2:1 11:1 3:3 !function=ex_shift_1 +%zcmop_n 8:3 %shlimm_6bit 12:1 2:5 !function=ex_rvc_shiftli %shrimm_6bit 12:1 2:5 !function=ex_rvc_shiftri@@ -66,6 +67,8 @@ &cmpp urlist spimm &cmjt index +&cmop zcmop_n + # Formats 16: @cr .... ..... ..... .. &r rs2=%rs2_5 rs1=%rd %rd @ci ... . ..... ..... .. &i imm=%imm_ci rs1=%rd %rd@@ -109,6 +112,8 @@ @cm_mv ... ... ... .. ... .. &r2_s rs2=%r2s rs1=%r1s @cm_jt ... ... ........ .. &cmjt %index +@c_mop ... . ..... ..... .. &cmop %zcmop_n + # *** RV32/64C Standard Extension (Quadrant 0) *** { # Opcode of all zeros is illegal; rd != 0, nzuimm == 0 is reserved.@@ -140,6 +145,7 @@ sw 110 ... ... .. ... 00 @cs_w addi 000 . ..... ..... 01 @ci addi 010 . ..... ..... 01 @c_li { + zcmops 011 0 0...1 00000 01 @c_mop # zcmop carving out of illegal c.lui xn,0 space illegal 011 0 ----- 00000 01 # c.addi16sp and c.lui, RES nzimm=0 addi 011 . 00010 ..... 01 @c_addi16sp lui 011 . ..... ..... 01 @c_luitarget/riscv/insn_trans/trans_zimops.c.inc +11
-1@@ -48,3 +48,14 @@ static bool trans_zimops_rr(DisasContext *ctx, arg_zimops_r * a) gen_set_gpr(ctx, a->rd, dest); return true; } + +static bool trans_zcmops(DisasContext *ctx, arg_zcmops * a) +{ + /* zimops not implemented, return false */ + if (!ctx->cfg_ptr->ext_zimops) { + gen_exception_illegal(ctx); + return false; + } + + return true; +} --2.34.1