10ca87f05SMatt Evans /* bpf_jit.h: BPF JIT compiler for PPC64 20ca87f05SMatt Evans * 30ca87f05SMatt Evans * Copyright 2011 Matt Evans <matt@ozlabs.org>, IBM Corporation 40ca87f05SMatt Evans * 50ca87f05SMatt Evans * This program is free software; you can redistribute it and/or 60ca87f05SMatt Evans * modify it under the terms of the GNU General Public License 70ca87f05SMatt Evans * as published by the Free Software Foundation; version 2 80ca87f05SMatt Evans * of the License. 90ca87f05SMatt Evans */ 100ca87f05SMatt Evans #ifndef _BPF_JIT_H 110ca87f05SMatt Evans #define _BPF_JIT_H 120ca87f05SMatt Evans 130ca87f05SMatt Evans #define BPF_PPC_STACK_LOCALS 32 140ca87f05SMatt Evans #define BPF_PPC_STACK_BASIC (48+64) 150ca87f05SMatt Evans #define BPF_PPC_STACK_SAVE (18*8) 160ca87f05SMatt Evans #define BPF_PPC_STACKFRAME (BPF_PPC_STACK_BASIC+BPF_PPC_STACK_LOCALS+ \ 170ca87f05SMatt Evans BPF_PPC_STACK_SAVE) 180ca87f05SMatt Evans #define BPF_PPC_SLOWPATH_FRAME (48+64) 190ca87f05SMatt Evans 200ca87f05SMatt Evans /* 210ca87f05SMatt Evans * Generated code register usage: 220ca87f05SMatt Evans * 230ca87f05SMatt Evans * As normal PPC C ABI (e.g. r1=sp, r2=TOC), with: 240ca87f05SMatt Evans * 250ca87f05SMatt Evans * skb r3 (Entry parameter) 260ca87f05SMatt Evans * A register r4 270ca87f05SMatt Evans * X register r5 280ca87f05SMatt Evans * addr param r6 290ca87f05SMatt Evans * r7-r10 scratch 300ca87f05SMatt Evans * skb->data r14 310ca87f05SMatt Evans * skb headlen r15 (skb->len - skb->data_len) 320ca87f05SMatt Evans * m[0] r16 330ca87f05SMatt Evans * m[...] ... 340ca87f05SMatt Evans * m[15] r31 350ca87f05SMatt Evans */ 360ca87f05SMatt Evans #define r_skb 3 370ca87f05SMatt Evans #define r_ret 3 380ca87f05SMatt Evans #define r_A 4 390ca87f05SMatt Evans #define r_X 5 400ca87f05SMatt Evans #define r_addr 6 410ca87f05SMatt Evans #define r_scratch1 7 42b0c06d33SVladimir Murzin #define r_scratch2 8 430ca87f05SMatt Evans #define r_D 14 440ca87f05SMatt Evans #define r_HL 15 450ca87f05SMatt Evans #define r_M 16 460ca87f05SMatt Evans 470ca87f05SMatt Evans #ifndef __ASSEMBLY__ 480ca87f05SMatt Evans 490ca87f05SMatt Evans /* 500ca87f05SMatt Evans * Assembly helpers from arch/powerpc/net/bpf_jit.S: 510ca87f05SMatt Evans */ 5205be1824SJan Seiffert #define DECLARE_LOAD_FUNC(func) \ 5305be1824SJan Seiffert extern u8 func[], func##_negative_offset[], func##_positive_offset[] 5405be1824SJan Seiffert 5505be1824SJan Seiffert DECLARE_LOAD_FUNC(sk_load_word); 5605be1824SJan Seiffert DECLARE_LOAD_FUNC(sk_load_half); 5705be1824SJan Seiffert DECLARE_LOAD_FUNC(sk_load_byte); 5805be1824SJan Seiffert DECLARE_LOAD_FUNC(sk_load_byte_msh); 590ca87f05SMatt Evans 600ca87f05SMatt Evans #define FUNCTION_DESCR_SIZE 24 610ca87f05SMatt Evans 620ca87f05SMatt Evans /* 630ca87f05SMatt Evans * 16-bit immediate helper macros: HA() is for use with sign-extending instrs 640ca87f05SMatt Evans * (e.g. LD, ADDI). If the bottom 16 bits is "-ve", add another bit into the 650ca87f05SMatt Evans * top half to negate the effect (i.e. 0xffff + 1 = 0x(1)0000). 660ca87f05SMatt Evans */ 670ca87f05SMatt Evans #define IMM_H(i) ((uintptr_t)(i)>>16) 680ca87f05SMatt Evans #define IMM_HA(i) (((uintptr_t)(i)>>16) + \ 690ca87f05SMatt Evans (((uintptr_t)(i) & 0x8000) >> 15)) 700ca87f05SMatt Evans #define IMM_L(i) ((uintptr_t)(i) & 0xffff) 710ca87f05SMatt Evans 720ca87f05SMatt Evans #define PLANT_INSTR(d, idx, instr) \ 730ca87f05SMatt Evans do { if (d) { (d)[idx] = instr; } idx++; } while (0) 740ca87f05SMatt Evans #define EMIT(instr) PLANT_INSTR(image, ctx->idx, instr) 750ca87f05SMatt Evans 760ca87f05SMatt Evans #define PPC_NOP() EMIT(PPC_INST_NOP) 770ca87f05SMatt Evans #define PPC_BLR() EMIT(PPC_INST_BLR) 780ca87f05SMatt Evans #define PPC_BLRL() EMIT(PPC_INST_BLRL) 79cdaade71SMichael Neuling #define PPC_MTLR(r) EMIT(PPC_INST_MTLR | ___PPC_RT(r)) 80cdaade71SMichael Neuling #define PPC_ADDI(d, a, i) EMIT(PPC_INST_ADDI | ___PPC_RT(d) | \ 81cdaade71SMichael Neuling ___PPC_RA(a) | IMM_L(i)) 820ca87f05SMatt Evans #define PPC_MR(d, a) PPC_OR(d, a, a) 830ca87f05SMatt Evans #define PPC_LI(r, i) PPC_ADDI(r, 0, i) 840ca87f05SMatt Evans #define PPC_ADDIS(d, a, i) EMIT(PPC_INST_ADDIS | \ 85cdaade71SMichael Neuling ___PPC_RS(d) | ___PPC_RA(a) | IMM_L(i)) 860ca87f05SMatt Evans #define PPC_LIS(r, i) PPC_ADDIS(r, 0, i) 87cdaade71SMichael Neuling #define PPC_STD(r, base, i) EMIT(PPC_INST_STD | ___PPC_RS(r) | \ 88cdaade71SMichael Neuling ___PPC_RA(base) | ((i) & 0xfffc)) 890ca87f05SMatt Evans 904e235761SDenis Kirjanov 914e235761SDenis Kirjanov #define PPC_LBZ(r, base, i) EMIT(PPC_INST_LBZ | ___PPC_RT(r) | \ 924e235761SDenis Kirjanov ___PPC_RA(base) | IMM_L(i)) 93cdaade71SMichael Neuling #define PPC_LD(r, base, i) EMIT(PPC_INST_LD | ___PPC_RT(r) | \ 94cdaade71SMichael Neuling ___PPC_RA(base) | IMM_L(i)) 95cdaade71SMichael Neuling #define PPC_LWZ(r, base, i) EMIT(PPC_INST_LWZ | ___PPC_RT(r) | \ 96cdaade71SMichael Neuling ___PPC_RA(base) | IMM_L(i)) 97cdaade71SMichael Neuling #define PPC_LHZ(r, base, i) EMIT(PPC_INST_LHZ | ___PPC_RT(r) | \ 98cdaade71SMichael Neuling ___PPC_RA(base) | IMM_L(i)) 999c662cadSPhilippe Bergheaud #define PPC_LHBRX(r, base, b) EMIT(PPC_INST_LHBRX | ___PPC_RT(r) | \ 1009c662cadSPhilippe Bergheaud ___PPC_RA(base) | ___PPC_RB(b)) 1010ca87f05SMatt Evans /* Convenience helpers for the above with 'far' offsets: */ 1024e235761SDenis Kirjanov #define PPC_LBZ_OFFS(r, base, i) do { if ((i) < 32768) PPC_LBZ(r, base, i); \ 1034e235761SDenis Kirjanov else { PPC_ADDIS(r, base, IMM_HA(i)); \ 1044e235761SDenis Kirjanov PPC_LBZ(r, r, IMM_L(i)); } } while(0) 1054e235761SDenis Kirjanov 1060ca87f05SMatt Evans #define PPC_LD_OFFS(r, base, i) do { if ((i) < 32768) PPC_LD(r, base, i); \ 1070ca87f05SMatt Evans else { PPC_ADDIS(r, base, IMM_HA(i)); \ 1080ca87f05SMatt Evans PPC_LD(r, r, IMM_L(i)); } } while(0) 1090ca87f05SMatt Evans 1100ca87f05SMatt Evans #define PPC_LWZ_OFFS(r, base, i) do { if ((i) < 32768) PPC_LWZ(r, base, i); \ 1110ca87f05SMatt Evans else { PPC_ADDIS(r, base, IMM_HA(i)); \ 1120ca87f05SMatt Evans PPC_LWZ(r, r, IMM_L(i)); } } while(0) 1130ca87f05SMatt Evans 1140ca87f05SMatt Evans #define PPC_LHZ_OFFS(r, base, i) do { if ((i) < 32768) PPC_LHZ(r, base, i); \ 1150ca87f05SMatt Evans else { PPC_ADDIS(r, base, IMM_HA(i)); \ 1160ca87f05SMatt Evans PPC_LHZ(r, r, IMM_L(i)); } } while(0) 1170ca87f05SMatt Evans 118cdaade71SMichael Neuling #define PPC_CMPWI(a, i) EMIT(PPC_INST_CMPWI | ___PPC_RA(a) | IMM_L(i)) 119cdaade71SMichael Neuling #define PPC_CMPDI(a, i) EMIT(PPC_INST_CMPDI | ___PPC_RA(a) | IMM_L(i)) 120cdaade71SMichael Neuling #define PPC_CMPLWI(a, i) EMIT(PPC_INST_CMPLWI | ___PPC_RA(a) | IMM_L(i)) 121cdaade71SMichael Neuling #define PPC_CMPLW(a, b) EMIT(PPC_INST_CMPLW | ___PPC_RA(a) | ___PPC_RB(b)) 1220ca87f05SMatt Evans 123cdaade71SMichael Neuling #define PPC_SUB(d, a, b) EMIT(PPC_INST_SUB | ___PPC_RT(d) | \ 124cdaade71SMichael Neuling ___PPC_RB(a) | ___PPC_RA(b)) 125cdaade71SMichael Neuling #define PPC_ADD(d, a, b) EMIT(PPC_INST_ADD | ___PPC_RT(d) | \ 126cdaade71SMichael Neuling ___PPC_RA(a) | ___PPC_RB(b)) 127cdaade71SMichael Neuling #define PPC_MUL(d, a, b) EMIT(PPC_INST_MULLW | ___PPC_RT(d) | \ 128cdaade71SMichael Neuling ___PPC_RA(a) | ___PPC_RB(b)) 129cdaade71SMichael Neuling #define PPC_MULHWU(d, a, b) EMIT(PPC_INST_MULHWU | ___PPC_RT(d) | \ 130cdaade71SMichael Neuling ___PPC_RA(a) | ___PPC_RB(b)) 131cdaade71SMichael Neuling #define PPC_MULI(d, a, i) EMIT(PPC_INST_MULLI | ___PPC_RT(d) | \ 132cdaade71SMichael Neuling ___PPC_RA(a) | IMM_L(i)) 133cdaade71SMichael Neuling #define PPC_DIVWU(d, a, b) EMIT(PPC_INST_DIVWU | ___PPC_RT(d) | \ 134cdaade71SMichael Neuling ___PPC_RA(a) | ___PPC_RB(b)) 135cdaade71SMichael Neuling #define PPC_AND(d, a, b) EMIT(PPC_INST_AND | ___PPC_RA(d) | \ 136cdaade71SMichael Neuling ___PPC_RS(a) | ___PPC_RB(b)) 137cdaade71SMichael Neuling #define PPC_ANDI(d, a, i) EMIT(PPC_INST_ANDI | ___PPC_RA(d) | \ 138cdaade71SMichael Neuling ___PPC_RS(a) | IMM_L(i)) 139cdaade71SMichael Neuling #define PPC_AND_DOT(d, a, b) EMIT(PPC_INST_ANDDOT | ___PPC_RA(d) | \ 140cdaade71SMichael Neuling ___PPC_RS(a) | ___PPC_RB(b)) 141cdaade71SMichael Neuling #define PPC_OR(d, a, b) EMIT(PPC_INST_OR | ___PPC_RA(d) | \ 142cdaade71SMichael Neuling ___PPC_RS(a) | ___PPC_RB(b)) 143cdaade71SMichael Neuling #define PPC_ORI(d, a, i) EMIT(PPC_INST_ORI | ___PPC_RA(d) | \ 144cdaade71SMichael Neuling ___PPC_RS(a) | IMM_L(i)) 145cdaade71SMichael Neuling #define PPC_ORIS(d, a, i) EMIT(PPC_INST_ORIS | ___PPC_RA(d) | \ 146cdaade71SMichael Neuling ___PPC_RS(a) | IMM_L(i)) 14702871903SDaniel Borkmann #define PPC_XOR(d, a, b) EMIT(PPC_INST_XOR | ___PPC_RA(d) | \ 14802871903SDaniel Borkmann ___PPC_RS(a) | ___PPC_RB(b)) 14902871903SDaniel Borkmann #define PPC_XORI(d, a, i) EMIT(PPC_INST_XORI | ___PPC_RA(d) | \ 15002871903SDaniel Borkmann ___PPC_RS(a) | IMM_L(i)) 15102871903SDaniel Borkmann #define PPC_XORIS(d, a, i) EMIT(PPC_INST_XORIS | ___PPC_RA(d) | \ 15202871903SDaniel Borkmann ___PPC_RS(a) | IMM_L(i)) 153cdaade71SMichael Neuling #define PPC_SLW(d, a, s) EMIT(PPC_INST_SLW | ___PPC_RA(d) | \ 154cdaade71SMichael Neuling ___PPC_RS(a) | ___PPC_RB(s)) 155cdaade71SMichael Neuling #define PPC_SRW(d, a, s) EMIT(PPC_INST_SRW | ___PPC_RA(d) | \ 156cdaade71SMichael Neuling ___PPC_RS(a) | ___PPC_RB(s)) 1570ca87f05SMatt Evans /* slwi = rlwinm Rx, Ry, n, 0, 31-n */ 158cdaade71SMichael Neuling #define PPC_SLWI(d, a, i) EMIT(PPC_INST_RLWINM | ___PPC_RA(d) | \ 159cdaade71SMichael Neuling ___PPC_RS(a) | __PPC_SH(i) | \ 1600ca87f05SMatt Evans __PPC_MB(0) | __PPC_ME(31-(i))) 1610ca87f05SMatt Evans /* srwi = rlwinm Rx, Ry, 32-n, n, 31 */ 162cdaade71SMichael Neuling #define PPC_SRWI(d, a, i) EMIT(PPC_INST_RLWINM | ___PPC_RA(d) | \ 163cdaade71SMichael Neuling ___PPC_RS(a) | __PPC_SH(32-(i)) | \ 1640ca87f05SMatt Evans __PPC_MB(i) | __PPC_ME(31)) 1650ca87f05SMatt Evans /* sldi = rldicr Rx, Ry, n, 63-n */ 166cdaade71SMichael Neuling #define PPC_SLDI(d, a, i) EMIT(PPC_INST_RLDICR | ___PPC_RA(d) | \ 167cdaade71SMichael Neuling ___PPC_RS(a) | __PPC_SH(i) | \ 1680ca87f05SMatt Evans __PPC_MB(63-(i)) | (((i) & 0x20) >> 4)) 169cdaade71SMichael Neuling #define PPC_NEG(d, a) EMIT(PPC_INST_NEG | ___PPC_RT(d) | ___PPC_RA(a)) 1700ca87f05SMatt Evans 1710ca87f05SMatt Evans /* Long jump; (unconditional 'branch') */ 1720ca87f05SMatt Evans #define PPC_JMP(dest) EMIT(PPC_INST_BRANCH | \ 1730ca87f05SMatt Evans (((dest) - (ctx->idx * 4)) & 0x03fffffc)) 1740ca87f05SMatt Evans /* "cond" here covers BO:BI fields. */ 1750ca87f05SMatt Evans #define PPC_BCC_SHORT(cond, dest) EMIT(PPC_INST_BRANCH_COND | \ 1760ca87f05SMatt Evans (((cond) & 0x3ff) << 16) | \ 1770ca87f05SMatt Evans (((dest) - (ctx->idx * 4)) & \ 1780ca87f05SMatt Evans 0xfffc)) 1790ca87f05SMatt Evans #define PPC_LI32(d, i) do { PPC_LI(d, IMM_L(i)); \ 1800ca87f05SMatt Evans if ((u32)(uintptr_t)(i) >= 32768) { \ 1810ca87f05SMatt Evans PPC_ADDIS(d, d, IMM_HA(i)); \ 1820ca87f05SMatt Evans } } while(0) 1830ca87f05SMatt Evans #define PPC_LI64(d, i) do { \ 1840ca87f05SMatt Evans if (!((uintptr_t)(i) & 0xffffffff00000000ULL)) \ 1850ca87f05SMatt Evans PPC_LI32(d, i); \ 1860ca87f05SMatt Evans else { \ 1870ca87f05SMatt Evans PPC_LIS(d, ((uintptr_t)(i) >> 48)); \ 1880ca87f05SMatt Evans if ((uintptr_t)(i) & 0x0000ffff00000000ULL) \ 1890ca87f05SMatt Evans PPC_ORI(d, d, \ 1900ca87f05SMatt Evans ((uintptr_t)(i) >> 32) & 0xffff); \ 1910ca87f05SMatt Evans PPC_SLDI(d, d, 32); \ 1920ca87f05SMatt Evans if ((uintptr_t)(i) & 0x00000000ffff0000ULL) \ 1930ca87f05SMatt Evans PPC_ORIS(d, d, \ 1940ca87f05SMatt Evans ((uintptr_t)(i) >> 16) & 0xffff); \ 1950ca87f05SMatt Evans if ((uintptr_t)(i) & 0x000000000000ffffULL) \ 1960ca87f05SMatt Evans PPC_ORI(d, d, (uintptr_t)(i) & 0xffff); \ 1970ca87f05SMatt Evans } } while (0); 1980ca87f05SMatt Evans 1999c662cadSPhilippe Bergheaud #define PPC_LHBRX_OFFS(r, base, i) \ 2009c662cadSPhilippe Bergheaud do { PPC_LI32(r, i); PPC_LHBRX(r, r, base); } while(0) 2019c662cadSPhilippe Bergheaud #ifdef __LITTLE_ENDIAN__ 2029c662cadSPhilippe Bergheaud #define PPC_NTOHS_OFFS(r, base, i) PPC_LHBRX_OFFS(r, base, i) 2039c662cadSPhilippe Bergheaud #else 2049c662cadSPhilippe Bergheaud #define PPC_NTOHS_OFFS(r, base, i) PPC_LHZ_OFFS(r, base, i) 2059c662cadSPhilippe Bergheaud #endif 2069c662cadSPhilippe Bergheaud 2070ca87f05SMatt Evans static inline bool is_nearbranch(int offset) 2080ca87f05SMatt Evans { 2090ca87f05SMatt Evans return (offset < 32768) && (offset >= -32768); 2100ca87f05SMatt Evans } 2110ca87f05SMatt Evans 2120ca87f05SMatt Evans /* 2130ca87f05SMatt Evans * The fly in the ointment of code size changing from pass to pass is 2140ca87f05SMatt Evans * avoided by padding the short branch case with a NOP. If code size differs 2150ca87f05SMatt Evans * with different branch reaches we will have the issue of code moving from 2160ca87f05SMatt Evans * one pass to the next and will need a few passes to converge on a stable 2170ca87f05SMatt Evans * state. 2180ca87f05SMatt Evans */ 2190ca87f05SMatt Evans #define PPC_BCC(cond, dest) do { \ 2200ca87f05SMatt Evans if (is_nearbranch((dest) - (ctx->idx * 4))) { \ 2210ca87f05SMatt Evans PPC_BCC_SHORT(cond, dest); \ 2220ca87f05SMatt Evans PPC_NOP(); \ 2230ca87f05SMatt Evans } else { \ 2240ca87f05SMatt Evans /* Flip the 'T or F' bit to invert comparison */ \ 2250ca87f05SMatt Evans PPC_BCC_SHORT(cond ^ COND_CMP_TRUE, (ctx->idx+2)*4); \ 2260ca87f05SMatt Evans PPC_JMP(dest); \ 2270ca87f05SMatt Evans } } while(0) 2280ca87f05SMatt Evans 2290ca87f05SMatt Evans /* To create a branch condition, select a bit of cr0... */ 2300ca87f05SMatt Evans #define CR0_LT 0 2310ca87f05SMatt Evans #define CR0_GT 1 2320ca87f05SMatt Evans #define CR0_EQ 2 2330ca87f05SMatt Evans /* ...and modify BO[3] */ 2340ca87f05SMatt Evans #define COND_CMP_TRUE 0x100 2350ca87f05SMatt Evans #define COND_CMP_FALSE 0x000 2360ca87f05SMatt Evans /* Together, they make all required comparisons: */ 2370ca87f05SMatt Evans #define COND_GT (CR0_GT | COND_CMP_TRUE) 2380ca87f05SMatt Evans #define COND_GE (CR0_LT | COND_CMP_FALSE) 2390ca87f05SMatt Evans #define COND_EQ (CR0_EQ | COND_CMP_TRUE) 2400ca87f05SMatt Evans #define COND_NE (CR0_EQ | COND_CMP_FALSE) 2410ca87f05SMatt Evans #define COND_LT (CR0_LT | COND_CMP_TRUE) 2420ca87f05SMatt Evans 2430ca87f05SMatt Evans #define SEEN_DATAREF 0x10000 /* might call external helpers */ 2440ca87f05SMatt Evans #define SEEN_XREG 0x20000 /* X reg is used */ 2450ca87f05SMatt Evans #define SEEN_MEM 0x40000 /* SEEN_MEM+(1<<n) = use mem[n] for temporary 2460ca87f05SMatt Evans * storage */ 2470ca87f05SMatt Evans #define SEEN_MEM_MSK 0x0ffff 2480ca87f05SMatt Evans 2490ca87f05SMatt Evans struct codegen_context { 2500ca87f05SMatt Evans unsigned int seen; 2510ca87f05SMatt Evans unsigned int idx; 2520ca87f05SMatt Evans int pc_ret0; /* bpf index of first RET #0 instruction (if any) */ 2530ca87f05SMatt Evans }; 2540ca87f05SMatt Evans 2550ca87f05SMatt Evans #endif 2560ca87f05SMatt Evans 2570ca87f05SMatt Evans #endif 258