1/* SPDX-License-Identifier: GPL-2.0 */ 2/* 3 * Copyright (C) 2014 Steven Rostedt, Red Hat Inc 4 */ 5 6#include <linux/cfi_types.h> 7#include <linux/linkage.h> 8#include <asm/asm-offsets.h> 9#include <asm/ptrace.h> 10#include <asm/ftrace.h> 11#include <asm/export.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 139SYM_TYPED_FUNC_START(ftrace_stub_graph) 140 CALL_DEPTH_ACCOUNT 141 RET 142SYM_FUNC_END(ftrace_stub_graph) 143 144#ifdef CONFIG_DYNAMIC_FTRACE 145 146SYM_FUNC_START(__fentry__) 147 CALL_DEPTH_ACCOUNT 148 RET 149SYM_FUNC_END(__fentry__) 150EXPORT_SYMBOL(__fentry__) 151 152SYM_FUNC_START(ftrace_caller) 153 /* save_mcount_regs fills in first two parameters */ 154 save_mcount_regs 155 156 CALL_DEPTH_ACCOUNT 157 158 /* Stack - skipping return address of ftrace_caller */ 159 leaq MCOUNT_REG_SIZE+8(%rsp), %rcx 160 movq %rcx, RSP(%rsp) 161 162SYM_INNER_LABEL(ftrace_caller_op_ptr, SYM_L_GLOBAL) 163 ANNOTATE_NOENDBR 164 /* Load the ftrace_ops into the 3rd parameter */ 165 movq function_trace_op(%rip), %rdx 166 167 /* regs go into 4th parameter */ 168 leaq (%rsp), %rcx 169 170 /* Only ops with REGS flag set should have CS register set */ 171 movq $0, CS(%rsp) 172 173 /* Account for the function call below */ 174 CALL_DEPTH_ACCOUNT 175 176SYM_INNER_LABEL(ftrace_call, SYM_L_GLOBAL) 177 ANNOTATE_NOENDBR 178 call ftrace_stub 179 180 /* Handlers can change the RIP */ 181 movq RIP(%rsp), %rax 182 movq %rax, MCOUNT_REG_SIZE(%rsp) 183 184 restore_mcount_regs 185 186 /* 187 * The code up to this label is copied into trampolines so 188 * think twice before adding any new code or changing the 189 * layout here. 190 */ 191SYM_INNER_LABEL(ftrace_caller_end, SYM_L_GLOBAL) 192 ANNOTATE_NOENDBR 193 RET 194SYM_FUNC_END(ftrace_caller); 195STACK_FRAME_NON_STANDARD_FP(ftrace_caller) 196 197SYM_FUNC_START(ftrace_regs_caller) 198 /* Save the current flags before any operations that can change them */ 199 pushfq 200 201 /* added 8 bytes to save flags */ 202 save_mcount_regs 8 203 /* save_mcount_regs fills in first two parameters */ 204 205 CALL_DEPTH_ACCOUNT 206 207SYM_INNER_LABEL(ftrace_regs_caller_op_ptr, SYM_L_GLOBAL) 208 ANNOTATE_NOENDBR 209 /* Load the ftrace_ops into the 3rd parameter */ 210 movq function_trace_op(%rip), %rdx 211 212 /* Save the rest of pt_regs */ 213 movq %r15, R15(%rsp) 214 movq %r14, R14(%rsp) 215 movq %r13, R13(%rsp) 216 movq %r12, R12(%rsp) 217 movq %r11, R11(%rsp) 218 movq %r10, R10(%rsp) 219 movq %rbx, RBX(%rsp) 220 /* Copy saved flags */ 221 movq MCOUNT_REG_SIZE(%rsp), %rcx 222 movq %rcx, EFLAGS(%rsp) 223 /* Kernel segments */ 224 movq $__KERNEL_DS, %rcx 225 movq %rcx, SS(%rsp) 226 movq $__KERNEL_CS, %rcx 227 movq %rcx, CS(%rsp) 228 /* Stack - skipping return address and flags */ 229 leaq MCOUNT_REG_SIZE+8*2(%rsp), %rcx 230 movq %rcx, RSP(%rsp) 231 232 ENCODE_FRAME_POINTER 233 234 /* regs go into 4th parameter */ 235 leaq (%rsp), %rcx 236 237 /* Account for the function call below */ 238 CALL_DEPTH_ACCOUNT 239 240SYM_INNER_LABEL(ftrace_regs_call, SYM_L_GLOBAL) 241 ANNOTATE_NOENDBR 242 call ftrace_stub 243 244 /* Copy flags back to SS, to restore them */ 245 movq EFLAGS(%rsp), %rax 246 movq %rax, MCOUNT_REG_SIZE(%rsp) 247 248 /* Handlers can change the RIP */ 249 movq RIP(%rsp), %rax 250 movq %rax, MCOUNT_REG_SIZE+8(%rsp) 251 252 /* restore the rest of pt_regs */ 253 movq R15(%rsp), %r15 254 movq R14(%rsp), %r14 255 movq R13(%rsp), %r13 256 movq R12(%rsp), %r12 257 movq R10(%rsp), %r10 258 movq RBX(%rsp), %rbx 259 260 movq ORIG_RAX(%rsp), %rax 261 movq %rax, MCOUNT_REG_SIZE-8(%rsp) 262 263 /* 264 * If ORIG_RAX is anything but zero, make this a call to that. 265 * See arch_ftrace_set_direct_caller(). 266 */ 267 testq %rax, %rax 268SYM_INNER_LABEL(ftrace_regs_caller_jmp, SYM_L_GLOBAL) 269 ANNOTATE_NOENDBR 270 jnz 1f 271 272 restore_mcount_regs 273 /* Restore flags */ 274 popfq 275 276 /* 277 * The trampoline will add the return. 278 */ 279SYM_INNER_LABEL(ftrace_regs_caller_end, SYM_L_GLOBAL) 280 ANNOTATE_NOENDBR 281 RET 282 283 /* Swap the flags with orig_rax */ 2841: movq MCOUNT_REG_SIZE(%rsp), %rdi 285 movq %rdi, MCOUNT_REG_SIZE-8(%rsp) 286 movq %rax, MCOUNT_REG_SIZE(%rsp) 287 288 restore_mcount_regs 8 289 /* Restore flags */ 290 popfq 291 UNWIND_HINT_FUNC 292 293 /* 294 * The above left an extra return value on the stack; effectively 295 * doing a tail-call without using a register. This PUSH;RET 296 * pattern unbalances the RSB, inject a pointless CALL to rebalance. 297 */ 298 ANNOTATE_INTRA_FUNCTION_CALL 299 CALL .Ldo_rebalance 300 int3 301.Ldo_rebalance: 302 add $8, %rsp 303 ALTERNATIVE __stringify(RET), \ 304 __stringify(ANNOTATE_UNRET_SAFE; ret; int3), \ 305 X86_FEATURE_CALL_DEPTH 306 307SYM_FUNC_END(ftrace_regs_caller) 308STACK_FRAME_NON_STANDARD_FP(ftrace_regs_caller) 309 310 311#else /* ! CONFIG_DYNAMIC_FTRACE */ 312 313SYM_FUNC_START(__fentry__) 314 CALL_DEPTH_ACCOUNT 315 316 cmpq $ftrace_stub, ftrace_trace_function 317 jnz trace 318 RET 319 320trace: 321 /* save_mcount_regs fills in first two parameters */ 322 save_mcount_regs 323 324 /* 325 * When DYNAMIC_FTRACE is not defined, ARCH_SUPPORTS_FTRACE_OPS is not 326 * set (see include/asm/ftrace.h and include/linux/ftrace.h). Only the 327 * ip and parent ip are used and the list function is called when 328 * function tracing is enabled. 329 */ 330 movq ftrace_trace_function, %r8 331 CALL_NOSPEC r8 332 restore_mcount_regs 333 334 jmp ftrace_stub 335SYM_FUNC_END(__fentry__) 336EXPORT_SYMBOL(__fentry__) 337STACK_FRAME_NON_STANDARD_FP(__fentry__) 338 339#endif /* CONFIG_DYNAMIC_FTRACE */ 340 341#ifdef CONFIG_FUNCTION_GRAPH_TRACER 342SYM_CODE_START(return_to_handler) 343 UNWIND_HINT_EMPTY 344 ANNOTATE_NOENDBR 345 subq $16, %rsp 346 347 /* Save the return values */ 348 movq %rax, (%rsp) 349 movq %rdx, 8(%rsp) 350 movq %rbp, %rdi 351 352 call ftrace_return_to_handler 353 354 movq %rax, %rdi 355 movq 8(%rsp), %rdx 356 movq (%rsp), %rax 357 358 addq $16, %rsp 359 /* 360 * Jump back to the old return address. This cannot be JMP_NOSPEC rdi 361 * since IBT would demand that contain ENDBR, which simply isn't so for 362 * return addresses. Use a retpoline here to keep the RSB balanced. 363 */ 364 ANNOTATE_INTRA_FUNCTION_CALL 365 call .Ldo_rop 366 int3 367.Ldo_rop: 368 mov %rdi, (%rsp) 369 ALTERNATIVE __stringify(RET), \ 370 __stringify(ANNOTATE_UNRET_SAFE; ret; int3), \ 371 X86_FEATURE_CALL_DEPTH 372SYM_CODE_END(return_to_handler) 373#endif 374