Commit | Line | Data |
---|---|---|
0ca87f05 ME |
1 | /* bpf_jit.h: BPF JIT compiler for PPC64 |
2 | * | |
3 | * Copyright 2011 Matt Evans <matt@ozlabs.org>, IBM Corporation | |
4 | * | |
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 | |
8 | * of the License. | |
9 | */ | |
10 | #ifndef _BPF_JIT_H | |
11 | #define _BPF_JIT_H | |
12 | ||
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+ \ | |
17 | BPF_PPC_STACK_SAVE) | |
18 | #define BPF_PPC_SLOWPATH_FRAME (48+64) | |
19 | ||
20 | /* | |
21 | * Generated code register usage: | |
22 | * | |
23 | * As normal PPC C ABI (e.g. r1=sp, r2=TOC), with: | |
24 | * | |
25 | * skb r3 (Entry parameter) | |
26 | * A register r4 | |
27 | * X register r5 | |
28 | * addr param r6 | |
29 | * r7-r10 scratch | |
30 | * skb->data r14 | |
31 | * skb headlen r15 (skb->len - skb->data_len) | |
32 | * m[0] r16 | |
33 | * m[...] ... | |
34 | * m[15] r31 | |
35 | */ | |
36 | #define r_skb 3 | |
37 | #define r_ret 3 | |
38 | #define r_A 4 | |
39 | #define r_X 5 | |
40 | #define r_addr 6 | |
41 | #define r_scratch1 7 | |
42 | #define r_D 14 | |
43 | #define r_HL 15 | |
44 | #define r_M 16 | |
45 | ||
46 | #ifndef __ASSEMBLY__ | |
47 | ||
48 | /* | |
49 | * Assembly helpers from arch/powerpc/net/bpf_jit.S: | |
50 | */ | |
05be1824 JS |
51 | #define DECLARE_LOAD_FUNC(func) \ |
52 | extern u8 func[], func##_negative_offset[], func##_positive_offset[] | |
53 | ||
54 | DECLARE_LOAD_FUNC(sk_load_word); | |
55 | DECLARE_LOAD_FUNC(sk_load_half); | |
56 | DECLARE_LOAD_FUNC(sk_load_byte); | |
57 | DECLARE_LOAD_FUNC(sk_load_byte_msh); | |
0ca87f05 ME |
58 | |
59 | #define FUNCTION_DESCR_SIZE 24 | |
60 | ||
61 | /* | |
62 | * 16-bit immediate helper macros: HA() is for use with sign-extending instrs | |
63 | * (e.g. LD, ADDI). If the bottom 16 bits is "-ve", add another bit into the | |
64 | * top half to negate the effect (i.e. 0xffff + 1 = 0x(1)0000). | |
65 | */ | |
66 | #define IMM_H(i) ((uintptr_t)(i)>>16) | |
67 | #define IMM_HA(i) (((uintptr_t)(i)>>16) + \ | |
68 | (((uintptr_t)(i) & 0x8000) >> 15)) | |
69 | #define IMM_L(i) ((uintptr_t)(i) & 0xffff) | |
70 | ||
71 | #define PLANT_INSTR(d, idx, instr) \ | |
72 | do { if (d) { (d)[idx] = instr; } idx++; } while (0) | |
73 | #define EMIT(instr) PLANT_INSTR(image, ctx->idx, instr) | |
74 | ||
75 | #define PPC_NOP() EMIT(PPC_INST_NOP) | |
76 | #define PPC_BLR() EMIT(PPC_INST_BLR) | |
77 | #define PPC_BLRL() EMIT(PPC_INST_BLRL) | |
cdaade71 MN |
78 | #define PPC_MTLR(r) EMIT(PPC_INST_MTLR | ___PPC_RT(r)) |
79 | #define PPC_ADDI(d, a, i) EMIT(PPC_INST_ADDI | ___PPC_RT(d) | \ | |
80 | ___PPC_RA(a) | IMM_L(i)) | |
0ca87f05 ME |
81 | #define PPC_MR(d, a) PPC_OR(d, a, a) |
82 | #define PPC_LI(r, i) PPC_ADDI(r, 0, i) | |
83 | #define PPC_ADDIS(d, a, i) EMIT(PPC_INST_ADDIS | \ | |
cdaade71 | 84 | ___PPC_RS(d) | ___PPC_RA(a) | IMM_L(i)) |
0ca87f05 | 85 | #define PPC_LIS(r, i) PPC_ADDIS(r, 0, i) |
cdaade71 MN |
86 | #define PPC_STD(r, base, i) EMIT(PPC_INST_STD | ___PPC_RS(r) | \ |
87 | ___PPC_RA(base) | ((i) & 0xfffc)) | |
0ca87f05 | 88 | |
cdaade71 MN |
89 | #define PPC_LD(r, base, i) EMIT(PPC_INST_LD | ___PPC_RT(r) | \ |
90 | ___PPC_RA(base) | IMM_L(i)) | |
91 | #define PPC_LWZ(r, base, i) EMIT(PPC_INST_LWZ | ___PPC_RT(r) | \ | |
92 | ___PPC_RA(base) | IMM_L(i)) | |
93 | #define PPC_LHZ(r, base, i) EMIT(PPC_INST_LHZ | ___PPC_RT(r) | \ | |
94 | ___PPC_RA(base) | IMM_L(i)) | |
0ca87f05 ME |
95 | /* Convenience helpers for the above with 'far' offsets: */ |
96 | #define PPC_LD_OFFS(r, base, i) do { if ((i) < 32768) PPC_LD(r, base, i); \ | |
97 | else { PPC_ADDIS(r, base, IMM_HA(i)); \ | |
98 | PPC_LD(r, r, IMM_L(i)); } } while(0) | |
99 | ||
100 | #define PPC_LWZ_OFFS(r, base, i) do { if ((i) < 32768) PPC_LWZ(r, base, i); \ | |
101 | else { PPC_ADDIS(r, base, IMM_HA(i)); \ | |
102 | PPC_LWZ(r, r, IMM_L(i)); } } while(0) | |
103 | ||
104 | #define PPC_LHZ_OFFS(r, base, i) do { if ((i) < 32768) PPC_LHZ(r, base, i); \ | |
105 | else { PPC_ADDIS(r, base, IMM_HA(i)); \ | |
106 | PPC_LHZ(r, r, IMM_L(i)); } } while(0) | |
107 | ||
cdaade71 MN |
108 | #define PPC_CMPWI(a, i) EMIT(PPC_INST_CMPWI | ___PPC_RA(a) | IMM_L(i)) |
109 | #define PPC_CMPDI(a, i) EMIT(PPC_INST_CMPDI | ___PPC_RA(a) | IMM_L(i)) | |
110 | #define PPC_CMPLWI(a, i) EMIT(PPC_INST_CMPLWI | ___PPC_RA(a) | IMM_L(i)) | |
111 | #define PPC_CMPLW(a, b) EMIT(PPC_INST_CMPLW | ___PPC_RA(a) | ___PPC_RB(b)) | |
0ca87f05 | 112 | |
cdaade71 MN |
113 | #define PPC_SUB(d, a, b) EMIT(PPC_INST_SUB | ___PPC_RT(d) | \ |
114 | ___PPC_RB(a) | ___PPC_RA(b)) | |
115 | #define PPC_ADD(d, a, b) EMIT(PPC_INST_ADD | ___PPC_RT(d) | \ | |
116 | ___PPC_RA(a) | ___PPC_RB(b)) | |
117 | #define PPC_MUL(d, a, b) EMIT(PPC_INST_MULLW | ___PPC_RT(d) | \ | |
118 | ___PPC_RA(a) | ___PPC_RB(b)) | |
119 | #define PPC_MULHWU(d, a, b) EMIT(PPC_INST_MULHWU | ___PPC_RT(d) | \ | |
120 | ___PPC_RA(a) | ___PPC_RB(b)) | |
121 | #define PPC_MULI(d, a, i) EMIT(PPC_INST_MULLI | ___PPC_RT(d) | \ | |
122 | ___PPC_RA(a) | IMM_L(i)) | |
123 | #define PPC_DIVWU(d, a, b) EMIT(PPC_INST_DIVWU | ___PPC_RT(d) | \ | |
124 | ___PPC_RA(a) | ___PPC_RB(b)) | |
125 | #define PPC_AND(d, a, b) EMIT(PPC_INST_AND | ___PPC_RA(d) | \ | |
126 | ___PPC_RS(a) | ___PPC_RB(b)) | |
127 | #define PPC_ANDI(d, a, i) EMIT(PPC_INST_ANDI | ___PPC_RA(d) | \ | |
128 | ___PPC_RS(a) | IMM_L(i)) | |
129 | #define PPC_AND_DOT(d, a, b) EMIT(PPC_INST_ANDDOT | ___PPC_RA(d) | \ | |
130 | ___PPC_RS(a) | ___PPC_RB(b)) | |
131 | #define PPC_OR(d, a, b) EMIT(PPC_INST_OR | ___PPC_RA(d) | \ | |
132 | ___PPC_RS(a) | ___PPC_RB(b)) | |
133 | #define PPC_ORI(d, a, i) EMIT(PPC_INST_ORI | ___PPC_RA(d) | \ | |
134 | ___PPC_RS(a) | IMM_L(i)) | |
135 | #define PPC_ORIS(d, a, i) EMIT(PPC_INST_ORIS | ___PPC_RA(d) | \ | |
136 | ___PPC_RS(a) | IMM_L(i)) | |
02871903 DB |
137 | #define PPC_XOR(d, a, b) EMIT(PPC_INST_XOR | ___PPC_RA(d) | \ |
138 | ___PPC_RS(a) | ___PPC_RB(b)) | |
139 | #define PPC_XORI(d, a, i) EMIT(PPC_INST_XORI | ___PPC_RA(d) | \ | |
140 | ___PPC_RS(a) | IMM_L(i)) | |
141 | #define PPC_XORIS(d, a, i) EMIT(PPC_INST_XORIS | ___PPC_RA(d) | \ | |
142 | ___PPC_RS(a) | IMM_L(i)) | |
cdaade71 MN |
143 | #define PPC_SLW(d, a, s) EMIT(PPC_INST_SLW | ___PPC_RA(d) | \ |
144 | ___PPC_RS(a) | ___PPC_RB(s)) | |
145 | #define PPC_SRW(d, a, s) EMIT(PPC_INST_SRW | ___PPC_RA(d) | \ | |
146 | ___PPC_RS(a) | ___PPC_RB(s)) | |
0ca87f05 | 147 | /* slwi = rlwinm Rx, Ry, n, 0, 31-n */ |
cdaade71 MN |
148 | #define PPC_SLWI(d, a, i) EMIT(PPC_INST_RLWINM | ___PPC_RA(d) | \ |
149 | ___PPC_RS(a) | __PPC_SH(i) | \ | |
0ca87f05 ME |
150 | __PPC_MB(0) | __PPC_ME(31-(i))) |
151 | /* srwi = rlwinm Rx, Ry, 32-n, n, 31 */ | |
cdaade71 MN |
152 | #define PPC_SRWI(d, a, i) EMIT(PPC_INST_RLWINM | ___PPC_RA(d) | \ |
153 | ___PPC_RS(a) | __PPC_SH(32-(i)) | \ | |
0ca87f05 ME |
154 | __PPC_MB(i) | __PPC_ME(31)) |
155 | /* sldi = rldicr Rx, Ry, n, 63-n */ | |
cdaade71 MN |
156 | #define PPC_SLDI(d, a, i) EMIT(PPC_INST_RLDICR | ___PPC_RA(d) | \ |
157 | ___PPC_RS(a) | __PPC_SH(i) | \ | |
0ca87f05 | 158 | __PPC_MB(63-(i)) | (((i) & 0x20) >> 4)) |
cdaade71 | 159 | #define PPC_NEG(d, a) EMIT(PPC_INST_NEG | ___PPC_RT(d) | ___PPC_RA(a)) |
0ca87f05 ME |
160 | |
161 | /* Long jump; (unconditional 'branch') */ | |
162 | #define PPC_JMP(dest) EMIT(PPC_INST_BRANCH | \ | |
163 | (((dest) - (ctx->idx * 4)) & 0x03fffffc)) | |
164 | /* "cond" here covers BO:BI fields. */ | |
165 | #define PPC_BCC_SHORT(cond, dest) EMIT(PPC_INST_BRANCH_COND | \ | |
166 | (((cond) & 0x3ff) << 16) | \ | |
167 | (((dest) - (ctx->idx * 4)) & \ | |
168 | 0xfffc)) | |
169 | #define PPC_LI32(d, i) do { PPC_LI(d, IMM_L(i)); \ | |
170 | if ((u32)(uintptr_t)(i) >= 32768) { \ | |
171 | PPC_ADDIS(d, d, IMM_HA(i)); \ | |
172 | } } while(0) | |
173 | #define PPC_LI64(d, i) do { \ | |
174 | if (!((uintptr_t)(i) & 0xffffffff00000000ULL)) \ | |
175 | PPC_LI32(d, i); \ | |
176 | else { \ | |
177 | PPC_LIS(d, ((uintptr_t)(i) >> 48)); \ | |
178 | if ((uintptr_t)(i) & 0x0000ffff00000000ULL) \ | |
179 | PPC_ORI(d, d, \ | |
180 | ((uintptr_t)(i) >> 32) & 0xffff); \ | |
181 | PPC_SLDI(d, d, 32); \ | |
182 | if ((uintptr_t)(i) & 0x00000000ffff0000ULL) \ | |
183 | PPC_ORIS(d, d, \ | |
184 | ((uintptr_t)(i) >> 16) & 0xffff); \ | |
185 | if ((uintptr_t)(i) & 0x000000000000ffffULL) \ | |
186 | PPC_ORI(d, d, (uintptr_t)(i) & 0xffff); \ | |
187 | } } while (0); | |
188 | ||
189 | static inline bool is_nearbranch(int offset) | |
190 | { | |
191 | return (offset < 32768) && (offset >= -32768); | |
192 | } | |
193 | ||
194 | /* | |
195 | * The fly in the ointment of code size changing from pass to pass is | |
196 | * avoided by padding the short branch case with a NOP. If code size differs | |
197 | * with different branch reaches we will have the issue of code moving from | |
198 | * one pass to the next and will need a few passes to converge on a stable | |
199 | * state. | |
200 | */ | |
201 | #define PPC_BCC(cond, dest) do { \ | |
202 | if (is_nearbranch((dest) - (ctx->idx * 4))) { \ | |
203 | PPC_BCC_SHORT(cond, dest); \ | |
204 | PPC_NOP(); \ | |
205 | } else { \ | |
206 | /* Flip the 'T or F' bit to invert comparison */ \ | |
207 | PPC_BCC_SHORT(cond ^ COND_CMP_TRUE, (ctx->idx+2)*4); \ | |
208 | PPC_JMP(dest); \ | |
209 | } } while(0) | |
210 | ||
211 | /* To create a branch condition, select a bit of cr0... */ | |
212 | #define CR0_LT 0 | |
213 | #define CR0_GT 1 | |
214 | #define CR0_EQ 2 | |
215 | /* ...and modify BO[3] */ | |
216 | #define COND_CMP_TRUE 0x100 | |
217 | #define COND_CMP_FALSE 0x000 | |
218 | /* Together, they make all required comparisons: */ | |
219 | #define COND_GT (CR0_GT | COND_CMP_TRUE) | |
220 | #define COND_GE (CR0_LT | COND_CMP_FALSE) | |
221 | #define COND_EQ (CR0_EQ | COND_CMP_TRUE) | |
222 | #define COND_NE (CR0_EQ | COND_CMP_FALSE) | |
223 | #define COND_LT (CR0_LT | COND_CMP_TRUE) | |
224 | ||
225 | #define SEEN_DATAREF 0x10000 /* might call external helpers */ | |
226 | #define SEEN_XREG 0x20000 /* X reg is used */ | |
227 | #define SEEN_MEM 0x40000 /* SEEN_MEM+(1<<n) = use mem[n] for temporary | |
228 | * storage */ | |
229 | #define SEEN_MEM_MSK 0x0ffff | |
230 | ||
231 | struct codegen_context { | |
232 | unsigned int seen; | |
233 | unsigned int idx; | |
234 | int pc_ret0; /* bpf index of first RET #0 instruction (if any) */ | |
235 | }; | |
236 | ||
237 | #endif | |
238 | ||
239 | #endif |