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 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk(); 119 unsigned long flags; 120 121 /* This is possible if op is under delayed unoptimizing */ 122 if (kprobe_disabled(&op->kp)) 123 return; 124 125 local_irq_save(flags); 126 hard_irq_disable(); 127 128 if (kprobe_running()) { 129 kprobes_inc_nmissed_count(&op->kp); 130 } else { 131 __this_cpu_write(current_kprobe, &op->kp); 132 regs->nip = (unsigned long)op->kp.addr; 133 kcb->kprobe_status = KPROBE_HIT_ACTIVE; 134 opt_pre_handler(&op->kp, regs); 135 __this_cpu_write(current_kprobe, NULL); 136 } 137 138 /* 139 * No need for an explicit __hard_irq_enable() here. 140 * local_irq_restore() will re-enable interrupts, 141 * if they were hard disabled. 142 */ 143 local_irq_restore(flags); 144 } 145 NOKPROBE_SYMBOL(optimized_callback); 146 147 void arch_remove_optimized_kprobe(struct optimized_kprobe *op) 148 { 149 if (op->optinsn.insn) { 150 free_ppc_optinsn_slot(op->optinsn.insn, 1); 151 op->optinsn.insn = NULL; 152 } 153 } 154 155 /* 156 * emulate_step() requires insn to be emulated as 157 * second parameter. Load register 'r4' with the 158 * instruction. 159 */ 160 void patch_imm32_load_insns(unsigned int val, kprobe_opcode_t *addr) 161 { 162 /* addis r4,0,(insn)@h */ 163 patch_instruction(addr, PPC_INST_ADDIS | ___PPC_RT(4) | 164 ((val >> 16) & 0xffff)); 165 addr++; 166 167 /* ori r4,r4,(insn)@l */ 168 patch_instruction(addr, PPC_INST_ORI | ___PPC_RA(4) | 169 ___PPC_RS(4) | (val & 0xffff)); 170 } 171 172 /* 173 * Generate instructions to load provided immediate 64-bit value 174 * to register 'r3' and patch these instructions at 'addr'. 175 */ 176 void patch_imm64_load_insns(unsigned long val, kprobe_opcode_t *addr) 177 { 178 /* lis r3,(op)@highest */ 179 patch_instruction(addr, PPC_INST_ADDIS | ___PPC_RT(3) | 180 ((val >> 48) & 0xffff)); 181 addr++; 182 183 /* ori r3,r3,(op)@higher */ 184 patch_instruction(addr, PPC_INST_ORI | ___PPC_RA(3) | 185 ___PPC_RS(3) | ((val >> 32) & 0xffff)); 186 addr++; 187 188 /* rldicr r3,r3,32,31 */ 189 patch_instruction(addr, PPC_INST_RLDICR | ___PPC_RA(3) | 190 ___PPC_RS(3) | __PPC_SH64(32) | __PPC_ME64(31)); 191 addr++; 192 193 /* oris r3,r3,(op)@h */ 194 patch_instruction(addr, PPC_INST_ORIS | ___PPC_RA(3) | 195 ___PPC_RS(3) | ((val >> 16) & 0xffff)); 196 addr++; 197 198 /* ori r3,r3,(op)@l */ 199 patch_instruction(addr, PPC_INST_ORI | ___PPC_RA(3) | 200 ___PPC_RS(3) | (val & 0xffff)); 201 } 202 203 int arch_prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p) 204 { 205 kprobe_opcode_t *buff, branch_op_callback, branch_emulate_step; 206 kprobe_opcode_t *op_callback_addr, *emulate_step_addr; 207 long b_offset; 208 unsigned long nip, size; 209 int rc, i; 210 211 kprobe_ppc_optinsn_slots.insn_size = MAX_OPTINSN_SIZE; 212 213 nip = can_optimize(p); 214 if (!nip) 215 return -EILSEQ; 216 217 /* Allocate instruction slot for detour buffer */ 218 buff = get_ppc_optinsn_slot(); 219 if (!buff) 220 return -ENOMEM; 221 222 /* 223 * OPTPROBE uses 'b' instruction to branch to optinsn.insn. 224 * 225 * The target address has to be relatively nearby, to permit use 226 * of branch instruction in powerpc, because the address is specified 227 * in an immediate field in the instruction opcode itself, ie 24 bits 228 * in the opcode specify the address. Therefore the address should 229 * be within 32MB on either side of the current instruction. 230 */ 231 b_offset = (unsigned long)buff - (unsigned long)p->addr; 232 if (!is_offset_in_branch_range(b_offset)) 233 goto error; 234 235 /* Check if the return address is also within 32MB range */ 236 b_offset = (unsigned long)(buff + TMPL_RET_IDX) - 237 (unsigned long)nip; 238 if (!is_offset_in_branch_range(b_offset)) 239 goto error; 240 241 /* Setup template */ 242 /* We can optimize this via patch_instruction_window later */ 243 size = (TMPL_END_IDX * sizeof(kprobe_opcode_t)) / sizeof(int); 244 pr_devel("Copying template to %p, size %lu\n", buff, size); 245 for (i = 0; i < size; i++) { 246 rc = patch_instruction(buff + i, *(optprobe_template_entry + i)); 247 if (rc < 0) 248 goto error; 249 } 250 251 /* 252 * Fixup the template with instructions to: 253 * 1. load the address of the actual probepoint 254 */ 255 patch_imm64_load_insns((unsigned long)op, buff + TMPL_OP_IDX); 256 257 /* 258 * 2. branch to optimized_callback() and emulate_step() 259 */ 260 op_callback_addr = (kprobe_opcode_t *)ppc_kallsyms_lookup_name("optimized_callback"); 261 emulate_step_addr = (kprobe_opcode_t *)ppc_kallsyms_lookup_name("emulate_step"); 262 if (!op_callback_addr || !emulate_step_addr) { 263 WARN(1, "Unable to lookup optimized_callback()/emulate_step()\n"); 264 goto error; 265 } 266 267 branch_op_callback = create_branch((unsigned int *)buff + TMPL_CALL_HDLR_IDX, 268 (unsigned long)op_callback_addr, 269 BRANCH_SET_LINK); 270 271 branch_emulate_step = create_branch((unsigned int *)buff + TMPL_EMULATE_IDX, 272 (unsigned long)emulate_step_addr, 273 BRANCH_SET_LINK); 274 275 if (!branch_op_callback || !branch_emulate_step) 276 goto error; 277 278 patch_instruction(buff + TMPL_CALL_HDLR_IDX, branch_op_callback); 279 patch_instruction(buff + TMPL_EMULATE_IDX, branch_emulate_step); 280 281 /* 282 * 3. load instruction to be emulated into relevant register, and 283 */ 284 patch_imm32_load_insns(*p->ainsn.insn, buff + TMPL_INSN_IDX); 285 286 /* 287 * 4. branch back from trampoline 288 */ 289 patch_branch(buff + TMPL_RET_IDX, (unsigned long)nip, 0); 290 291 flush_icache_range((unsigned long)buff, 292 (unsigned long)(&buff[TMPL_END_IDX])); 293 294 op->optinsn.insn = buff; 295 296 return 0; 297 298 error: 299 free_ppc_optinsn_slot(buff, 0); 300 return -ERANGE; 301 302 } 303 304 int arch_prepared_optinsn(struct arch_optimized_insn *optinsn) 305 { 306 return optinsn->insn != NULL; 307 } 308 309 /* 310 * On powerpc, Optprobes always replaces one instruction (4 bytes 311 * aligned and 4 bytes long). It is impossible to encounter another 312 * kprobe in this address range. So always return 0. 313 */ 314 int arch_check_optimized_kprobe(struct optimized_kprobe *op) 315 { 316 return 0; 317 } 318 319 void arch_optimize_kprobes(struct list_head *oplist) 320 { 321 struct optimized_kprobe *op; 322 struct optimized_kprobe *tmp; 323 324 list_for_each_entry_safe(op, tmp, oplist, list) { 325 /* 326 * Backup instructions which will be replaced 327 * by jump address 328 */ 329 memcpy(op->optinsn.copied_insn, op->kp.addr, 330 RELATIVEJUMP_SIZE); 331 patch_instruction(op->kp.addr, 332 create_branch((unsigned int *)op->kp.addr, 333 (unsigned long)op->optinsn.insn, 0)); 334 list_del_init(&op->list); 335 } 336 } 337 338 void arch_unoptimize_kprobe(struct optimized_kprobe *op) 339 { 340 arch_arm_kprobe(&op->kp); 341 } 342 343 void arch_unoptimize_kprobes(struct list_head *oplist, 344 struct list_head *done_list) 345 { 346 struct optimized_kprobe *op; 347 struct optimized_kprobe *tmp; 348 349 list_for_each_entry_safe(op, tmp, oplist, list) { 350 arch_unoptimize_kprobe(op); 351 list_move(&op->list, done_list); 352 } 353 } 354 355 int arch_within_optimized_kprobe(struct optimized_kprobe *op, 356 unsigned long addr) 357 { 358 return ((unsigned long)op->kp.addr <= addr && 359 (unsigned long)op->kp.addr + RELATIVEJUMP_SIZE > addr); 360 } 361