1 /* 2 * Code for Kernel probes Jump optimization. 3 * 4 * Copyright 2017, Anju T, IBM Corp. 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License 8 * as published by the Free Software Foundation; either version 9 * 2 of the License, or (at your option) any later version. 10 */ 11 12 #include <linux/kprobes.h> 13 #include <linux/jump_label.h> 14 #include <linux/types.h> 15 #include <linux/slab.h> 16 #include <linux/list.h> 17 #include <asm/kprobes.h> 18 #include <asm/ptrace.h> 19 #include <asm/cacheflush.h> 20 #include <asm/code-patching.h> 21 #include <asm/sstep.h> 22 #include <asm/ppc-opcode.h> 23 24 #define TMPL_CALL_HDLR_IDX \ 25 (optprobe_template_call_handler - optprobe_template_entry) 26 #define TMPL_EMULATE_IDX \ 27 (optprobe_template_call_emulate - optprobe_template_entry) 28 #define TMPL_RET_IDX \ 29 (optprobe_template_ret - optprobe_template_entry) 30 #define TMPL_OP_IDX \ 31 (optprobe_template_op_address - optprobe_template_entry) 32 #define TMPL_INSN_IDX \ 33 (optprobe_template_insn - optprobe_template_entry) 34 #define TMPL_END_IDX \ 35 (optprobe_template_end - optprobe_template_entry) 36 37 DEFINE_INSN_CACHE_OPS(ppc_optinsn); 38 39 static bool insn_page_in_use; 40 41 static void *__ppc_alloc_insn_page(void) 42 { 43 if (insn_page_in_use) 44 return NULL; 45 insn_page_in_use = true; 46 return &optinsn_slot; 47 } 48 49 static void __ppc_free_insn_page(void *page __maybe_unused) 50 { 51 insn_page_in_use = false; 52 } 53 54 struct kprobe_insn_cache kprobe_ppc_optinsn_slots = { 55 .mutex = __MUTEX_INITIALIZER(kprobe_ppc_optinsn_slots.mutex), 56 .pages = LIST_HEAD_INIT(kprobe_ppc_optinsn_slots.pages), 57 /* insn_size initialized later */ 58 .alloc = __ppc_alloc_insn_page, 59 .free = __ppc_free_insn_page, 60 .nr_garbage = 0, 61 }; 62 63 /* 64 * Check if we can optimize this probe. Returns NIP post-emulation if this can 65 * be optimized and 0 otherwise. 66 */ 67 static unsigned long can_optimize(struct kprobe *p) 68 { 69 struct pt_regs regs; 70 struct instruction_op op; 71 unsigned long nip = 0; 72 73 /* 74 * kprobe placed for kretprobe during boot time 75 * has a 'nop' instruction, which can be emulated. 76 * So further checks can be skipped. 77 */ 78 if (p->addr == (kprobe_opcode_t *)&kretprobe_trampoline) 79 return (unsigned long)p->addr + sizeof(kprobe_opcode_t); 80 81 /* 82 * We only support optimizing kernel addresses, but not 83 * module addresses. 84 * 85 * FIXME: Optimize kprobes placed in module addresses. 86 */ 87 if (!is_kernel_addr((unsigned long)p->addr)) 88 return 0; 89 90 memset(®s, 0, sizeof(struct pt_regs)); 91 regs.nip = (unsigned long)p->addr; 92 regs.trap = 0x0; 93 regs.msr = MSR_KERNEL; 94 95 /* 96 * Kprobe placed in conditional branch instructions are 97 * not optimized, as we can't predict the nip prior with 98 * dummy pt_regs and can not ensure that the return branch 99 * from detour buffer falls in the range of address (i.e 32MB). 100 * A branch back from trampoline is set up in the detour buffer 101 * to the nip returned by the analyse_instr() here. 102 * 103 * Ensure that the instruction is not a conditional branch, 104 * and that can be emulated. 105 */ 106 if (!is_conditional_branch(*p->ainsn.insn) && 107 analyse_instr(&op, ®s, *p->ainsn.insn) == 1) { 108 emulate_update_regs(®s, &op); 109 nip = regs.nip; 110 } 111 112 return nip; 113 } 114 115 static void optimized_callback(struct optimized_kprobe *op, 116 struct pt_regs *regs) 117 { 118 /* This is possible if op is under delayed unoptimizing */ 119 if (kprobe_disabled(&op->kp)) 120 return; 121 122 preempt_disable(); 123 124 if (kprobe_running()) { 125 kprobes_inc_nmissed_count(&op->kp); 126 } else { 127 __this_cpu_write(current_kprobe, &op->kp); 128 regs->nip = (unsigned long)op->kp.addr; 129 get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE; 130 opt_pre_handler(&op->kp, regs); 131 __this_cpu_write(current_kprobe, NULL); 132 } 133 134 preempt_enable_no_resched(); 135 } 136 NOKPROBE_SYMBOL(optimized_callback); 137 138 void arch_remove_optimized_kprobe(struct optimized_kprobe *op) 139 { 140 if (op->optinsn.insn) { 141 free_ppc_optinsn_slot(op->optinsn.insn, 1); 142 op->optinsn.insn = NULL; 143 } 144 } 145 146 /* 147 * emulate_step() requires insn to be emulated as 148 * second parameter. Load register 'r4' with the 149 * instruction. 150 */ 151 void patch_imm32_load_insns(unsigned int val, kprobe_opcode_t *addr) 152 { 153 /* addis r4,0,(insn)@h */ 154 patch_instruction(addr, PPC_INST_ADDIS | ___PPC_RT(4) | 155 ((val >> 16) & 0xffff)); 156 addr++; 157 158 /* ori r4,r4,(insn)@l */ 159 patch_instruction(addr, PPC_INST_ORI | ___PPC_RA(4) | 160 ___PPC_RS(4) | (val & 0xffff)); 161 } 162 163 /* 164 * Generate instructions to load provided immediate 64-bit value 165 * to register 'r3' and patch these instructions at 'addr'. 166 */ 167 void patch_imm64_load_insns(unsigned long val, kprobe_opcode_t *addr) 168 { 169 /* lis r3,(op)@highest */ 170 patch_instruction(addr, PPC_INST_ADDIS | ___PPC_RT(3) | 171 ((val >> 48) & 0xffff)); 172 addr++; 173 174 /* ori r3,r3,(op)@higher */ 175 patch_instruction(addr, PPC_INST_ORI | ___PPC_RA(3) | 176 ___PPC_RS(3) | ((val >> 32) & 0xffff)); 177 addr++; 178 179 /* rldicr r3,r3,32,31 */ 180 patch_instruction(addr, PPC_INST_RLDICR | ___PPC_RA(3) | 181 ___PPC_RS(3) | __PPC_SH64(32) | __PPC_ME64(31)); 182 addr++; 183 184 /* oris r3,r3,(op)@h */ 185 patch_instruction(addr, PPC_INST_ORIS | ___PPC_RA(3) | 186 ___PPC_RS(3) | ((val >> 16) & 0xffff)); 187 addr++; 188 189 /* ori r3,r3,(op)@l */ 190 patch_instruction(addr, PPC_INST_ORI | ___PPC_RA(3) | 191 ___PPC_RS(3) | (val & 0xffff)); 192 } 193 194 int arch_prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p) 195 { 196 kprobe_opcode_t *buff, branch_op_callback, branch_emulate_step; 197 kprobe_opcode_t *op_callback_addr, *emulate_step_addr; 198 long b_offset; 199 unsigned long nip, size; 200 int rc, i; 201 202 kprobe_ppc_optinsn_slots.insn_size = MAX_OPTINSN_SIZE; 203 204 nip = can_optimize(p); 205 if (!nip) 206 return -EILSEQ; 207 208 /* Allocate instruction slot for detour buffer */ 209 buff = get_ppc_optinsn_slot(); 210 if (!buff) 211 return -ENOMEM; 212 213 /* 214 * OPTPROBE uses 'b' instruction to branch to optinsn.insn. 215 * 216 * The target address has to be relatively nearby, to permit use 217 * of branch instruction in powerpc, because the address is specified 218 * in an immediate field in the instruction opcode itself, ie 24 bits 219 * in the opcode specify the address. Therefore the address should 220 * be within 32MB on either side of the current instruction. 221 */ 222 b_offset = (unsigned long)buff - (unsigned long)p->addr; 223 if (!is_offset_in_branch_range(b_offset)) 224 goto error; 225 226 /* Check if the return address is also within 32MB range */ 227 b_offset = (unsigned long)(buff + TMPL_RET_IDX) - 228 (unsigned long)nip; 229 if (!is_offset_in_branch_range(b_offset)) 230 goto error; 231 232 /* Setup template */ 233 /* We can optimize this via patch_instruction_window later */ 234 size = (TMPL_END_IDX * sizeof(kprobe_opcode_t)) / sizeof(int); 235 pr_devel("Copying template to %p, size %lu\n", buff, size); 236 for (i = 0; i < size; i++) { 237 rc = patch_instruction(buff + i, *(optprobe_template_entry + i)); 238 if (rc < 0) 239 goto error; 240 } 241 242 /* 243 * Fixup the template with instructions to: 244 * 1. load the address of the actual probepoint 245 */ 246 patch_imm64_load_insns((unsigned long)op, buff + TMPL_OP_IDX); 247 248 /* 249 * 2. branch to optimized_callback() and emulate_step() 250 */ 251 op_callback_addr = (kprobe_opcode_t *)ppc_kallsyms_lookup_name("optimized_callback"); 252 emulate_step_addr = (kprobe_opcode_t *)ppc_kallsyms_lookup_name("emulate_step"); 253 if (!op_callback_addr || !emulate_step_addr) { 254 WARN(1, "Unable to lookup optimized_callback()/emulate_step()\n"); 255 goto error; 256 } 257 258 branch_op_callback = create_branch((unsigned int *)buff + TMPL_CALL_HDLR_IDX, 259 (unsigned long)op_callback_addr, 260 BRANCH_SET_LINK); 261 262 branch_emulate_step = create_branch((unsigned int *)buff + TMPL_EMULATE_IDX, 263 (unsigned long)emulate_step_addr, 264 BRANCH_SET_LINK); 265 266 if (!branch_op_callback || !branch_emulate_step) 267 goto error; 268 269 patch_instruction(buff + TMPL_CALL_HDLR_IDX, branch_op_callback); 270 patch_instruction(buff + TMPL_EMULATE_IDX, branch_emulate_step); 271 272 /* 273 * 3. load instruction to be emulated into relevant register, and 274 */ 275 patch_imm32_load_insns(*p->ainsn.insn, buff + TMPL_INSN_IDX); 276 277 /* 278 * 4. branch back from trampoline 279 */ 280 patch_branch(buff + TMPL_RET_IDX, (unsigned long)nip, 0); 281 282 flush_icache_range((unsigned long)buff, 283 (unsigned long)(&buff[TMPL_END_IDX])); 284 285 op->optinsn.insn = buff; 286 287 return 0; 288 289 error: 290 free_ppc_optinsn_slot(buff, 0); 291 return -ERANGE; 292 293 } 294 295 int arch_prepared_optinsn(struct arch_optimized_insn *optinsn) 296 { 297 return optinsn->insn != NULL; 298 } 299 300 /* 301 * On powerpc, Optprobes always replaces one instruction (4 bytes 302 * aligned and 4 bytes long). It is impossible to encounter another 303 * kprobe in this address range. So always return 0. 304 */ 305 int arch_check_optimized_kprobe(struct optimized_kprobe *op) 306 { 307 return 0; 308 } 309 310 void arch_optimize_kprobes(struct list_head *oplist) 311 { 312 struct optimized_kprobe *op; 313 struct optimized_kprobe *tmp; 314 315 list_for_each_entry_safe(op, tmp, oplist, list) { 316 /* 317 * Backup instructions which will be replaced 318 * by jump address 319 */ 320 memcpy(op->optinsn.copied_insn, op->kp.addr, 321 RELATIVEJUMP_SIZE); 322 patch_instruction(op->kp.addr, 323 create_branch((unsigned int *)op->kp.addr, 324 (unsigned long)op->optinsn.insn, 0)); 325 list_del_init(&op->list); 326 } 327 } 328 329 void arch_unoptimize_kprobe(struct optimized_kprobe *op) 330 { 331 arch_arm_kprobe(&op->kp); 332 } 333 334 void arch_unoptimize_kprobes(struct list_head *oplist, 335 struct list_head *done_list) 336 { 337 struct optimized_kprobe *op; 338 struct optimized_kprobe *tmp; 339 340 list_for_each_entry_safe(op, tmp, oplist, list) { 341 arch_unoptimize_kprobe(op); 342 list_move(&op->list, done_list); 343 } 344 } 345 346 int arch_within_optimized_kprobe(struct optimized_kprobe *op, 347 unsigned long addr) 348 { 349 return ((unsigned long)op->kp.addr <= addr && 350 (unsigned long)op->kp.addr + RELATIVEJUMP_SIZE > addr); 351 } 352