1/* SPDX-License-Identifier: GPL-2.0 */ 2/* 3 * Copyright (C) 2014 Steven Rostedt, Red Hat Inc 4 */ 5 6#include <linux/export.h> 7#include <linux/cfi_types.h> 8#include <linux/linkage.h> 9#include <asm/asm-offsets.h> 10#include <asm/ptrace.h> 11#include <asm/ftrace.h> 12#include <asm/nospec-branch.h> 13#include <asm/unwind_hints.h> 14#include <asm/frame.h> 15 16 .code64 17 .section .text, "ax" 18 19#ifdef CONFIG_FRAME_POINTER 20/* Save parent and function stack frames (rip and rbp) */ 21# define MCOUNT_FRAME_SIZE (8+16*2) 22#else 23/* No need to save a stack frame */ 24# define MCOUNT_FRAME_SIZE 0 25#endif /* CONFIG_FRAME_POINTER */ 26 27/* Size of stack used to save mcount regs in save_mcount_regs */ 28#define MCOUNT_REG_SIZE (FRAME_SIZE + MCOUNT_FRAME_SIZE) 29 30/* 31 * gcc -pg option adds a call to 'mcount' in most functions. 32 * When -mfentry is used, the call is to 'fentry' and not 'mcount' 33 * and is done before the function's stack frame is set up. 34 * They both require a set of regs to be saved before calling 35 * any C code and restored before returning back to the function. 36 * 37 * On boot up, all these calls are converted into nops. When tracing 38 * is enabled, the call can jump to either ftrace_caller or 39 * ftrace_regs_caller. Callbacks (tracing functions) that require 40 * ftrace_regs_caller (like kprobes) need to have pt_regs passed to 41 * it. For this reason, the size of the pt_regs structure will be 42 * allocated on the stack and the required mcount registers will 43 * be saved in the locations that pt_regs has them in. 44 */ 45 46/* 47 * @added: the amount of stack added before calling this 48 * 49 * After this is called, the following registers contain: 50 * 51 * %rdi - holds the address that called the trampoline 52 * %rsi - holds the parent function (traced function's return address) 53 * %rdx - holds the original %rbp 54 */ 55.macro save_mcount_regs added=0 56 57#ifdef CONFIG_FRAME_POINTER 58 /* Save the original rbp */ 59 pushq %rbp 60 61 /* 62 * Stack traces will stop at the ftrace trampoline if the frame pointer 63 * is not set up properly. If fentry is used, we need to save a frame 64 * pointer for the parent as well as the function traced, because the 65 * fentry is called before the stack frame is set up, where as mcount 66 * is called afterward. 67 */ 68 69 /* Save the parent pointer (skip orig rbp and our return address) */ 70 pushq \added+8*2(%rsp) 71 pushq %rbp 72 movq %rsp, %rbp 73 /* Save the return address (now skip orig rbp, rbp and parent) */ 74 pushq \added+8*3(%rsp) 75 pushq %rbp 76 movq %rsp, %rbp 77#endif /* CONFIG_FRAME_POINTER */ 78 79 /* 80 * We add enough stack to save all regs. 81 */ 82 subq $(FRAME_SIZE), %rsp 83 movq %rax, RAX(%rsp) 84 movq %rcx, RCX(%rsp) 85 movq %rdx, RDX(%rsp) 86 movq %rsi, RSI(%rsp) 87 movq %rdi, RDI(%rsp) 88 movq %r8, R8(%rsp) 89 movq %r9, R9(%rsp) 90 movq $0, ORIG_RAX(%rsp) 91 /* 92 * Save the original RBP. Even though the mcount ABI does not 93 * require this, it helps out callers. 94 */ 95#ifdef CONFIG_FRAME_POINTER 96 movq MCOUNT_REG_SIZE-8(%rsp), %rdx 97#else 98 movq %rbp, %rdx 99#endif 100 movq %rdx, RBP(%rsp) 101 102 /* Copy the parent address into %rsi (second parameter) */ 103 movq MCOUNT_REG_SIZE+8+\added(%rsp), %rsi 104 105 /* Move RIP to its proper location */ 106 movq MCOUNT_REG_SIZE+\added(%rsp), %rdi 107 movq %rdi, RIP(%rsp) 108 109 /* 110 * Now %rdi (the first parameter) has the return address of 111 * where ftrace_call returns. But the callbacks expect the 112 * address of the call itself. 113 */ 114 subq $MCOUNT_INSN_SIZE, %rdi 115 .endm 116 117.macro restore_mcount_regs save=0 118 119 /* ftrace_regs_caller or frame pointers require this */ 120 movq RBP(%rsp), %rbp 121 122 movq R9(%rsp), %r9 123 movq R8(%rsp), %r8 124 movq RDI(%rsp), %rdi 125 movq RSI(%rsp), %rsi 126 movq RDX(%rsp), %rdx 127 movq RCX(%rsp), %rcx 128 movq RAX(%rsp), %rax 129 130 addq $MCOUNT_REG_SIZE-\save, %rsp 131 132 .endm 133 134SYM_TYPED_FUNC_START(ftrace_stub) 135 CALL_DEPTH_ACCOUNT 136 RET 137SYM_FUNC_END(ftrace_stub) 138 139#ifdef CONFIG_FUNCTION_GRAPH_TRACER 140SYM_TYPED_FUNC_START(ftrace_stub_graph) 141 CALL_DEPTH_ACCOUNT 142 RET 143SYM_FUNC_END(ftrace_stub_graph) 144#endif 145 146#ifdef CONFIG_DYNAMIC_FTRACE 147 148SYM_FUNC_START(__fentry__) 149 ANNOTATE_NOENDBR 150 CALL_DEPTH_ACCOUNT 151 RET 152SYM_FUNC_END(__fentry__) 153EXPORT_SYMBOL(__fentry__) 154 155SYM_FUNC_START(ftrace_caller) 156 ANNOTATE_NOENDBR 157 /* save_mcount_regs fills in first two parameters */ 158 save_mcount_regs 159 160 CALL_DEPTH_ACCOUNT 161 162 /* Stack - skipping return address of ftrace_caller */ 163 leaq MCOUNT_REG_SIZE+8(%rsp), %rcx 164 movq %rcx, RSP(%rsp) 165 166SYM_INNER_LABEL(ftrace_caller_op_ptr, SYM_L_GLOBAL) 167 ANNOTATE_NOENDBR 168 /* Load the ftrace_ops into the 3rd parameter */ 169 movq function_trace_op(%rip), %rdx 170 171 /* regs go into 4th parameter */ 172 leaq (%rsp), %rcx 173 174 /* Only ops with REGS flag set should have CS register set */ 175 movq $0, CS(%rsp) 176 177 /* Account for the function call below */ 178 CALL_DEPTH_ACCOUNT 179 180SYM_INNER_LABEL(ftrace_call, SYM_L_GLOBAL) 181 ANNOTATE_NOENDBR 182 call ftrace_stub 183 184 /* Handlers can change the RIP */ 185 movq RIP(%rsp), %rax 186 movq %rax, MCOUNT_REG_SIZE(%rsp) 187 188 restore_mcount_regs 189 190 /* 191 * The code up to this label is copied into trampolines so 192 * think twice before adding any new code or changing the 193 * layout here. 194 */ 195SYM_INNER_LABEL(ftrace_caller_end, SYM_L_GLOBAL) 196 ANNOTATE_NOENDBR 197 RET 198SYM_FUNC_END(ftrace_caller); 199STACK_FRAME_NON_STANDARD_FP(ftrace_caller) 200 201SYM_FUNC_START(ftrace_regs_caller) 202 ANNOTATE_NOENDBR 203 /* Save the current flags before any operations that can change them */ 204 pushfq 205 206 /* added 8 bytes to save flags */ 207 save_mcount_regs 8 208 /* save_mcount_regs fills in first two parameters */ 209 210 CALL_DEPTH_ACCOUNT 211 212SYM_INNER_LABEL(ftrace_regs_caller_op_ptr, SYM_L_GLOBAL) 213 ANNOTATE_NOENDBR 214 /* Load the ftrace_ops into the 3rd parameter */ 215 movq function_trace_op(%rip), %rdx 216 217 /* Save the rest of pt_regs */ 218 movq %r15, R15(%rsp) 219 movq %r14, R14(%rsp) 220 movq %r13, R13(%rsp) 221 movq %r12, R12(%rsp) 222 movq %r11, R11(%rsp) 223 movq %r10, R10(%rsp) 224 movq %rbx, RBX(%rsp) 225 /* Copy saved flags */ 226 movq MCOUNT_REG_SIZE(%rsp), %rcx 227 movq %rcx, EFLAGS(%rsp) 228 /* Kernel segments */ 229 movq $__KERNEL_DS, %rcx 230 movq %rcx, SS(%rsp) 231 movq $__KERNEL_CS, %rcx 232 movq %rcx, CS(%rsp) 233 /* Stack - skipping return address and flags */ 234 leaq MCOUNT_REG_SIZE+8*2(%rsp), %rcx 235 movq %rcx, RSP(%rsp) 236 237 ENCODE_FRAME_POINTER 238 239 /* regs go into 4th parameter */ 240 leaq (%rsp), %rcx 241 242 /* Account for the function call below */ 243 CALL_DEPTH_ACCOUNT 244 245SYM_INNER_LABEL(ftrace_regs_call, SYM_L_GLOBAL) 246 ANNOTATE_NOENDBR 247 call ftrace_stub 248 249 /* Copy flags back to SS, to restore them */ 250 movq EFLAGS(%rsp), %rax 251 movq %rax, MCOUNT_REG_SIZE(%rsp) 252 253 /* Handlers can change the RIP */ 254 movq RIP(%rsp), %rax 255 movq %rax, MCOUNT_REG_SIZE+8(%rsp) 256 257 /* restore the rest of pt_regs */ 258 movq R15(%rsp), %r15 259 movq R14(%rsp), %r14 260 movq R13(%rsp), %r13 261 movq R12(%rsp), %r12 262 movq R10(%rsp), %r10 263 movq RBX(%rsp), %rbx 264 265 movq ORIG_RAX(%rsp), %rax 266 movq %rax, MCOUNT_REG_SIZE-8(%rsp) 267 268 /* 269 * If ORIG_RAX is anything but zero, make this a call to that. 270 * See arch_ftrace_set_direct_caller(). 271 */ 272 testq %rax, %rax 273SYM_INNER_LABEL(ftrace_regs_caller_jmp, SYM_L_GLOBAL) 274 ANNOTATE_NOENDBR 275 jnz 1f 276 277 restore_mcount_regs 278 /* Restore flags */ 279 popfq 280 281 /* 282 * The trampoline will add the return. 283 */ 284SYM_INNER_LABEL(ftrace_regs_caller_end, SYM_L_GLOBAL) 285 ANNOTATE_NOENDBR 286 RET 287 288 /* Swap the flags with orig_rax */ 2891: movq MCOUNT_REG_SIZE(%rsp), %rdi 290 movq %rdi, MCOUNT_REG_SIZE-8(%rsp) 291 movq %rax, MCOUNT_REG_SIZE(%rsp) 292 293 restore_mcount_regs 8 294 /* Restore flags */ 295 popfq 296 UNWIND_HINT_FUNC 297 298 /* 299 * The above left an extra return value on the stack; effectively 300 * doing a tail-call without using a register. This PUSH;RET 301 * pattern unbalances the RSB, inject a pointless CALL to rebalance. 302 */ 303 ANNOTATE_INTRA_FUNCTION_CALL 304 CALL .Ldo_rebalance 305 int3 306.Ldo_rebalance: 307 add $8, %rsp 308 ALTERNATIVE __stringify(RET), \ 309 __stringify(ANNOTATE_UNRET_SAFE; ret; int3), \ 310 X86_FEATURE_CALL_DEPTH 311 312SYM_FUNC_END(ftrace_regs_caller) 313STACK_FRAME_NON_STANDARD_FP(ftrace_regs_caller) 314 315SYM_FUNC_START(ftrace_stub_direct_tramp) 316 ANNOTATE_NOENDBR 317 CALL_DEPTH_ACCOUNT 318 RET 319SYM_FUNC_END(ftrace_stub_direct_tramp) 320 321#else /* ! CONFIG_DYNAMIC_FTRACE */ 322 323SYM_FUNC_START(__fentry__) 324 ANNOTATE_NOENDBR 325 CALL_DEPTH_ACCOUNT 326 327 cmpq $ftrace_stub, ftrace_trace_function 328 jnz trace 329 RET 330 331trace: 332 /* save_mcount_regs fills in first two parameters */ 333 save_mcount_regs 334 335 /* 336 * When DYNAMIC_FTRACE is not defined, ARCH_SUPPORTS_FTRACE_OPS is not 337 * set (see include/asm/ftrace.h and include/linux/ftrace.h). Only the 338 * ip and parent ip are used and the list function is called when 339 * function tracing is enabled. 340 */ 341 movq ftrace_trace_function, %r8 342 CALL_NOSPEC r8 343 restore_mcount_regs 344 345 jmp ftrace_stub 346SYM_FUNC_END(__fentry__) 347EXPORT_SYMBOL(__fentry__) 348STACK_FRAME_NON_STANDARD_FP(__fentry__) 349 350#endif /* CONFIG_DYNAMIC_FTRACE */ 351 352#ifdef CONFIG_FUNCTION_GRAPH_TRACER 353SYM_CODE_START(return_to_handler) 354 UNWIND_HINT_UNDEFINED 355 ANNOTATE_NOENDBR 356 357 /* Save ftrace_regs for function exit context */ 358 subq $(FRAME_SIZE), %rsp 359 360 movq %rax, RAX(%rsp) 361 movq %rdx, RDX(%rsp) 362 movq %rbp, RBP(%rsp) 363 movq %rsp, %rdi 364 365 call ftrace_return_to_handler 366 367 movq %rax, %rdi 368 movq RDX(%rsp), %rdx 369 movq RAX(%rsp), %rax 370 371 addq $(FRAME_SIZE), %rsp 372 /* 373 * Jump back to the old return address. This cannot be JMP_NOSPEC rdi 374 * since IBT would demand that contain ENDBR, which simply isn't so for 375 * return addresses. Use a retpoline here to keep the RSB balanced. 376 */ 377 ANNOTATE_INTRA_FUNCTION_CALL 378 call .Ldo_rop 379 int3 380.Ldo_rop: 381 mov %rdi, (%rsp) 382 ALTERNATIVE __stringify(RET), \ 383 __stringify(ANNOTATE_UNRET_SAFE; ret; int3), \ 384 X86_FEATURE_CALL_DEPTH 385SYM_CODE_END(return_to_handler) 386#endif 387