1 /* bpf_jit.h: BPF JIT compiler for PPC64
3 * Copyright 2011 Matt Evans <matt@ozlabs.org>, IBM Corporation
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License
7 * as published by the Free Software Foundation; version 2
13 #define BPF_PPC_STACK_LOCALS 32
14 #define BPF_PPC_STACK_BASIC (48+64)
15 #define BPF_PPC_STACK_SAVE (18*8)
16 #define BPF_PPC_STACKFRAME (BPF_PPC_STACK_BASIC+BPF_PPC_STACK_LOCALS+ \
18 #define BPF_PPC_SLOWPATH_FRAME (48+64)
21 * Generated code register usage:
23 * As normal PPC C ABI (e.g. r1=sp, r2=TOC), with:
25 * skb r3 (Entry parameter)
31 * skb headlen r15 (skb->len - skb->data_len)
50 * Assembly helpers from arch/powerpc/net/bpf_jit.S:
52 #define DECLARE_LOAD_FUNC(func) \
53 extern u8 func[], func##_negative_offset[], func##_positive_offset[]
55 DECLARE_LOAD_FUNC(sk_load_word
);
56 DECLARE_LOAD_FUNC(sk_load_half
);
57 DECLARE_LOAD_FUNC(sk_load_byte
);
58 DECLARE_LOAD_FUNC(sk_load_byte_msh
);
60 #define FUNCTION_DESCR_SIZE 24
63 * 16-bit immediate helper macros: HA() is for use with sign-extending instrs
64 * (e.g. LD, ADDI). If the bottom 16 bits is "-ve", add another bit into the
65 * top half to negate the effect (i.e. 0xffff + 1 = 0x(1)0000).
67 #define IMM_H(i) ((uintptr_t)(i)>>16)
68 #define IMM_HA(i) (((uintptr_t)(i)>>16) + \
69 (((uintptr_t)(i) & 0x8000) >> 15))
70 #define IMM_L(i) ((uintptr_t)(i) & 0xffff)
72 #define PLANT_INSTR(d, idx, instr) \
73 do { if (d) { (d)[idx] = instr; } idx++; } while (0)
74 #define EMIT(instr) PLANT_INSTR(image, ctx->idx, instr)
76 #define PPC_NOP() EMIT(PPC_INST_NOP)
77 #define PPC_BLR() EMIT(PPC_INST_BLR)
78 #define PPC_BLRL() EMIT(PPC_INST_BLRL)
79 #define PPC_MTLR(r) EMIT(PPC_INST_MTLR | ___PPC_RT(r))
80 #define PPC_ADDI(d, a, i) EMIT(PPC_INST_ADDI | ___PPC_RT(d) | \
81 ___PPC_RA(a) | IMM_L(i))
82 #define PPC_MR(d, a) PPC_OR(d, a, a)
83 #define PPC_LI(r, i) PPC_ADDI(r, 0, i)
84 #define PPC_ADDIS(d, a, i) EMIT(PPC_INST_ADDIS | \
85 ___PPC_RS(d) | ___PPC_RA(a) | IMM_L(i))
86 #define PPC_LIS(r, i) PPC_ADDIS(r, 0, i)
87 #define PPC_STD(r, base, i) EMIT(PPC_INST_STD | ___PPC_RS(r) | \
88 ___PPC_RA(base) | ((i) & 0xfffc))
90 #define PPC_LD(r, base, i) EMIT(PPC_INST_LD | ___PPC_RT(r) | \
91 ___PPC_RA(base) | IMM_L(i))
92 #define PPC_LWZ(r, base, i) EMIT(PPC_INST_LWZ | ___PPC_RT(r) | \
93 ___PPC_RA(base) | IMM_L(i))
94 #define PPC_LHZ(r, base, i) EMIT(PPC_INST_LHZ | ___PPC_RT(r) | \
95 ___PPC_RA(base) | IMM_L(i))
96 #define PPC_LHBRX(r, base, b) EMIT(PPC_INST_LHBRX | ___PPC_RT(r) | \
97 ___PPC_RA(base) | ___PPC_RB(b))
98 /* Convenience helpers for the above with 'far' offsets: */
99 #define PPC_LD_OFFS(r, base, i) do { if ((i) < 32768) PPC_LD(r, base, i); \
100 else { PPC_ADDIS(r, base, IMM_HA(i)); \
101 PPC_LD(r, r, IMM_L(i)); } } while(0)
103 #define PPC_LWZ_OFFS(r, base, i) do { if ((i) < 32768) PPC_LWZ(r, base, i); \
104 else { PPC_ADDIS(r, base, IMM_HA(i)); \
105 PPC_LWZ(r, r, IMM_L(i)); } } while(0)
107 #define PPC_LHZ_OFFS(r, base, i) do { if ((i) < 32768) PPC_LHZ(r, base, i); \
108 else { PPC_ADDIS(r, base, IMM_HA(i)); \
109 PPC_LHZ(r, r, IMM_L(i)); } } while(0)
111 #define PPC_CMPWI(a, i) EMIT(PPC_INST_CMPWI | ___PPC_RA(a) | IMM_L(i))
112 #define PPC_CMPDI(a, i) EMIT(PPC_INST_CMPDI | ___PPC_RA(a) | IMM_L(i))
113 #define PPC_CMPLWI(a, i) EMIT(PPC_INST_CMPLWI | ___PPC_RA(a) | IMM_L(i))
114 #define PPC_CMPLW(a, b) EMIT(PPC_INST_CMPLW | ___PPC_RA(a) | ___PPC_RB(b))
116 #define PPC_SUB(d, a, b) EMIT(PPC_INST_SUB | ___PPC_RT(d) | \
117 ___PPC_RB(a) | ___PPC_RA(b))
118 #define PPC_ADD(d, a, b) EMIT(PPC_INST_ADD | ___PPC_RT(d) | \
119 ___PPC_RA(a) | ___PPC_RB(b))
120 #define PPC_MUL(d, a, b) EMIT(PPC_INST_MULLW | ___PPC_RT(d) | \
121 ___PPC_RA(a) | ___PPC_RB(b))
122 #define PPC_MULHWU(d, a, b) EMIT(PPC_INST_MULHWU | ___PPC_RT(d) | \
123 ___PPC_RA(a) | ___PPC_RB(b))
124 #define PPC_MULI(d, a, i) EMIT(PPC_INST_MULLI | ___PPC_RT(d) | \
125 ___PPC_RA(a) | IMM_L(i))
126 #define PPC_DIVWU(d, a, b) EMIT(PPC_INST_DIVWU | ___PPC_RT(d) | \
127 ___PPC_RA(a) | ___PPC_RB(b))
128 #define PPC_AND(d, a, b) EMIT(PPC_INST_AND | ___PPC_RA(d) | \
129 ___PPC_RS(a) | ___PPC_RB(b))
130 #define PPC_ANDI(d, a, i) EMIT(PPC_INST_ANDI | ___PPC_RA(d) | \
131 ___PPC_RS(a) | IMM_L(i))
132 #define PPC_AND_DOT(d, a, b) EMIT(PPC_INST_ANDDOT | ___PPC_RA(d) | \
133 ___PPC_RS(a) | ___PPC_RB(b))
134 #define PPC_OR(d, a, b) EMIT(PPC_INST_OR | ___PPC_RA(d) | \
135 ___PPC_RS(a) | ___PPC_RB(b))
136 #define PPC_ORI(d, a, i) EMIT(PPC_INST_ORI | ___PPC_RA(d) | \
137 ___PPC_RS(a) | IMM_L(i))
138 #define PPC_ORIS(d, a, i) EMIT(PPC_INST_ORIS | ___PPC_RA(d) | \
139 ___PPC_RS(a) | IMM_L(i))
140 #define PPC_XOR(d, a, b) EMIT(PPC_INST_XOR | ___PPC_RA(d) | \
141 ___PPC_RS(a) | ___PPC_RB(b))
142 #define PPC_XORI(d, a, i) EMIT(PPC_INST_XORI | ___PPC_RA(d) | \
143 ___PPC_RS(a) | IMM_L(i))
144 #define PPC_XORIS(d, a, i) EMIT(PPC_INST_XORIS | ___PPC_RA(d) | \
145 ___PPC_RS(a) | IMM_L(i))
146 #define PPC_SLW(d, a, s) EMIT(PPC_INST_SLW | ___PPC_RA(d) | \
147 ___PPC_RS(a) | ___PPC_RB(s))
148 #define PPC_SRW(d, a, s) EMIT(PPC_INST_SRW | ___PPC_RA(d) | \
149 ___PPC_RS(a) | ___PPC_RB(s))
150 /* slwi = rlwinm Rx, Ry, n, 0, 31-n */
151 #define PPC_SLWI(d, a, i) EMIT(PPC_INST_RLWINM | ___PPC_RA(d) | \
152 ___PPC_RS(a) | __PPC_SH(i) | \
153 __PPC_MB(0) | __PPC_ME(31-(i)))
154 /* srwi = rlwinm Rx, Ry, 32-n, n, 31 */
155 #define PPC_SRWI(d, a, i) EMIT(PPC_INST_RLWINM | ___PPC_RA(d) | \
156 ___PPC_RS(a) | __PPC_SH(32-(i)) | \
157 __PPC_MB(i) | __PPC_ME(31))
158 /* sldi = rldicr Rx, Ry, n, 63-n */
159 #define PPC_SLDI(d, a, i) EMIT(PPC_INST_RLDICR | ___PPC_RA(d) | \
160 ___PPC_RS(a) | __PPC_SH(i) | \
161 __PPC_MB(63-(i)) | (((i) & 0x20) >> 4))
162 #define PPC_NEG(d, a) EMIT(PPC_INST_NEG | ___PPC_RT(d) | ___PPC_RA(a))
164 /* Long jump; (unconditional 'branch') */
165 #define PPC_JMP(dest) EMIT(PPC_INST_BRANCH | \
166 (((dest) - (ctx->idx * 4)) & 0x03fffffc))
167 /* "cond" here covers BO:BI fields. */
168 #define PPC_BCC_SHORT(cond, dest) EMIT(PPC_INST_BRANCH_COND | \
169 (((cond) & 0x3ff) << 16) | \
170 (((dest) - (ctx->idx * 4)) & \
172 #define PPC_LI32(d, i) do { PPC_LI(d, IMM_L(i)); \
173 if ((u32)(uintptr_t)(i) >= 32768) { \
174 PPC_ADDIS(d, d, IMM_HA(i)); \
176 #define PPC_LI64(d, i) do { \
177 if (!((uintptr_t)(i) & 0xffffffff00000000ULL)) \
180 PPC_LIS(d, ((uintptr_t)(i) >> 48)); \
181 if ((uintptr_t)(i) & 0x0000ffff00000000ULL) \
183 ((uintptr_t)(i) >> 32) & 0xffff); \
184 PPC_SLDI(d, d, 32); \
185 if ((uintptr_t)(i) & 0x00000000ffff0000ULL) \
187 ((uintptr_t)(i) >> 16) & 0xffff); \
188 if ((uintptr_t)(i) & 0x000000000000ffffULL) \
189 PPC_ORI(d, d, (uintptr_t)(i) & 0xffff); \
192 #define PPC_LHBRX_OFFS(r, base, i) \
193 do { PPC_LI32(r, i); PPC_LHBRX(r, r, base); } while(0)
194 #ifdef __LITTLE_ENDIAN__
195 #define PPC_NTOHS_OFFS(r, base, i) PPC_LHBRX_OFFS(r, base, i)
197 #define PPC_NTOHS_OFFS(r, base, i) PPC_LHZ_OFFS(r, base, i)
200 static inline bool is_nearbranch(int offset
)
202 return (offset
< 32768) && (offset
>= -32768);
206 * The fly in the ointment of code size changing from pass to pass is
207 * avoided by padding the short branch case with a NOP. If code size differs
208 * with different branch reaches we will have the issue of code moving from
209 * one pass to the next and will need a few passes to converge on a stable
212 #define PPC_BCC(cond, dest) do { \
213 if (is_nearbranch((dest) - (ctx->idx * 4))) { \
214 PPC_BCC_SHORT(cond, dest); \
217 /* Flip the 'T or F' bit to invert comparison */ \
218 PPC_BCC_SHORT(cond ^ COND_CMP_TRUE, (ctx->idx+2)*4); \
222 /* To create a branch condition, select a bit of cr0... */
226 /* ...and modify BO[3] */
227 #define COND_CMP_TRUE 0x100
228 #define COND_CMP_FALSE 0x000
229 /* Together, they make all required comparisons: */
230 #define COND_GT (CR0_GT | COND_CMP_TRUE)
231 #define COND_GE (CR0_LT | COND_CMP_FALSE)
232 #define COND_EQ (CR0_EQ | COND_CMP_TRUE)
233 #define COND_NE (CR0_EQ | COND_CMP_FALSE)
234 #define COND_LT (CR0_LT | COND_CMP_TRUE)
236 #define SEEN_DATAREF 0x10000 /* might call external helpers */
237 #define SEEN_XREG 0x20000 /* X reg is used */
238 #define SEEN_MEM 0x40000 /* SEEN_MEM+(1<<n) = use mem[n] for temporary
240 #define SEEN_MEM_MSK 0x0ffff
242 struct codegen_context
{
245 int pc_ret0
; /* bpf index of first RET #0 instruction (if any) */