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 */
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);
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)
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))
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 | \
84 __PPC_RS(d) | __PPC_RA(a) | IMM_L(i))
85 #define PPC_LIS(r, i) PPC_ADDIS(r, 0, i)
86 #define PPC_STD(r, base, i) EMIT(PPC_INST_STD | __PPC_RS(r) | \
87 __PPC_RA(base) | ((i) & 0xfffc))
88
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))
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
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))
112
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))
137 #define PPC_SLW(d, a, s) EMIT(PPC_INST_SLW | __PPC_RA(d) | \
138 __PPC_RS(a) | __PPC_RB(s))
139 #define PPC_SRW(d, a, s) EMIT(PPC_INST_SRW | __PPC_RA(d) | \
140 __PPC_RS(a) | __PPC_RB(s))
141 /* slwi = rlwinm Rx, Ry, n, 0, 31-n */
142 #define PPC_SLWI(d, a, i) EMIT(PPC_INST_RLWINM | __PPC_RA(d) | \
143 __PPC_RS(a) | __PPC_SH(i) | \
144 __PPC_MB(0) | __PPC_ME(31-(i)))
145 /* srwi = rlwinm Rx, Ry, 32-n, n, 31 */
146 #define PPC_SRWI(d, a, i) EMIT(PPC_INST_RLWINM | __PPC_RA(d) | \
147 __PPC_RS(a) | __PPC_SH(32-(i)) | \
148 __PPC_MB(i) | __PPC_ME(31))
149 /* sldi = rldicr Rx, Ry, n, 63-n */
150 #define PPC_SLDI(d, a, i) EMIT(PPC_INST_RLDICR | __PPC_RA(d) | \
151 __PPC_RS(a) | __PPC_SH(i) | \
152 __PPC_MB(63-(i)) | (((i) & 0x20) >> 4))
153 #define PPC_NEG(d, a) EMIT(PPC_INST_NEG | __PPC_RT(d) | __PPC_RA(a))
154
155 /* Long jump; (unconditional 'branch') */
156 #define PPC_JMP(dest) EMIT(PPC_INST_BRANCH | \
157 (((dest) - (ctx->idx * 4)) & 0x03fffffc))
158 /* "cond" here covers BO:BI fields. */
159 #define PPC_BCC_SHORT(cond, dest) EMIT(PPC_INST_BRANCH_COND | \
160 (((cond) & 0x3ff) << 16) | \
161 (((dest) - (ctx->idx * 4)) & \
162 0xfffc))
163 #define PPC_LI32(d, i) do { PPC_LI(d, IMM_L(i)); \
164 if ((u32)(uintptr_t)(i) >= 32768) { \
165 PPC_ADDIS(d, d, IMM_HA(i)); \
166 } } while(0)
167 #define PPC_LI64(d, i) do { \
168 if (!((uintptr_t)(i) & 0xffffffff00000000ULL)) \
169 PPC_LI32(d, i); \
170 else { \
171 PPC_LIS(d, ((uintptr_t)(i) >> 48)); \
172 if ((uintptr_t)(i) & 0x0000ffff00000000ULL) \
173 PPC_ORI(d, d, \
174 ((uintptr_t)(i) >> 32) & 0xffff); \
175 PPC_SLDI(d, d, 32); \
176 if ((uintptr_t)(i) & 0x00000000ffff0000ULL) \
177 PPC_ORIS(d, d, \
178 ((uintptr_t)(i) >> 16) & 0xffff); \
179 if ((uintptr_t)(i) & 0x000000000000ffffULL) \
180 PPC_ORI(d, d, (uintptr_t)(i) & 0xffff); \
181 } } while (0);
182
is_nearbranch(int offset)183 static inline bool is_nearbranch(int offset)
184 {
185 return (offset < 32768) && (offset >= -32768);
186 }
187
188 /*
189 * The fly in the ointment of code size changing from pass to pass is
190 * avoided by padding the short branch case with a NOP. If code size differs
191 * with different branch reaches we will have the issue of code moving from
192 * one pass to the next and will need a few passes to converge on a stable
193 * state.
194 */
195 #define PPC_BCC(cond, dest) do { \
196 if (is_nearbranch((dest) - (ctx->idx * 4))) { \
197 PPC_BCC_SHORT(cond, dest); \
198 PPC_NOP(); \
199 } else { \
200 /* Flip the 'T or F' bit to invert comparison */ \
201 PPC_BCC_SHORT(cond ^ COND_CMP_TRUE, (ctx->idx+2)*4); \
202 PPC_JMP(dest); \
203 } } while(0)
204
205 /* To create a branch condition, select a bit of cr0... */
206 #define CR0_LT 0
207 #define CR0_GT 1
208 #define CR0_EQ 2
209 /* ...and modify BO[3] */
210 #define COND_CMP_TRUE 0x100
211 #define COND_CMP_FALSE 0x000
212 /* Together, they make all required comparisons: */
213 #define COND_GT (CR0_GT | COND_CMP_TRUE)
214 #define COND_GE (CR0_LT | COND_CMP_FALSE)
215 #define COND_EQ (CR0_EQ | COND_CMP_TRUE)
216 #define COND_NE (CR0_EQ | COND_CMP_FALSE)
217 #define COND_LT (CR0_LT | COND_CMP_TRUE)
218
219 #define SEEN_DATAREF 0x10000 /* might call external helpers */
220 #define SEEN_XREG 0x20000 /* X reg is used */
221 #define SEEN_MEM 0x40000 /* SEEN_MEM+(1<<n) = use mem[n] for temporary
222 * storage */
223 #define SEEN_MEM_MSK 0x0ffff
224
225 struct codegen_context {
226 unsigned int seen;
227 unsigned int idx;
228 int pc_ret0; /* bpf index of first RET #0 instruction (if any) */
229 };
230
231 #endif
232
233 #endif
234