1/* SPDX-License-Identifier: GPL-2.0 */ 2/* 3 * Copyright (C) 2014 Steven Rostedt, Red Hat Inc 4 */ 5 6#include <linux/linkage.h> 7#include <asm/ptrace.h> 8#include <asm/ftrace.h> 9#include <asm/export.h> 10#include <asm/nospec-branch.h> 11#include <asm/unwind_hints.h> 12#include <asm/frame.h> 13 14 .code64 15 .section .text, "ax" 16 17#ifdef CONFIG_FRAME_POINTER 18/* Save parent and function stack frames (rip and rbp) */ 19# define MCOUNT_FRAME_SIZE (8+16*2) 20#else 21/* No need to save a stack frame */ 22# define MCOUNT_FRAME_SIZE 0 23#endif /* CONFIG_FRAME_POINTER */ 24 25/* Size of stack used to save mcount regs in save_mcount_regs */ 26#define MCOUNT_REG_SIZE (FRAME_SIZE + MCOUNT_FRAME_SIZE) 27 28/* 29 * gcc -pg option adds a call to 'mcount' in most functions. 30 * When -mfentry is used, the call is to 'fentry' and not 'mcount' 31 * and is done before the function's stack frame is set up. 32 * They both require a set of regs to be saved before calling 33 * any C code and restored before returning back to the function. 34 * 35 * On boot up, all these calls are converted into nops. When tracing 36 * is enabled, the call can jump to either ftrace_caller or 37 * ftrace_regs_caller. Callbacks (tracing functions) that require 38 * ftrace_regs_caller (like kprobes) need to have pt_regs passed to 39 * it. For this reason, the size of the pt_regs structure will be 40 * allocated on the stack and the required mcount registers will 41 * be saved in the locations that pt_regs has them in. 42 */ 43 44/* 45 * @added: the amount of stack added before calling this 46 * 47 * After this is called, the following registers contain: 48 * 49 * %rdi - holds the address that called the trampoline 50 * %rsi - holds the parent function (traced function's return address) 51 * %rdx - holds the original %rbp 52 */ 53.macro save_mcount_regs added=0 54 55#ifdef CONFIG_FRAME_POINTER 56 /* Save the original rbp */ 57 pushq %rbp 58 59 /* 60 * Stack traces will stop at the ftrace trampoline if the frame pointer 61 * is not set up properly. If fentry is used, we need to save a frame 62 * pointer for the parent as well as the function traced, because the 63 * fentry is called before the stack frame is set up, where as mcount 64 * is called afterward. 65 */ 66 67 /* Save the parent pointer (skip orig rbp and our return address) */ 68 pushq \added+8*2(%rsp) 69 pushq %rbp 70 movq %rsp, %rbp 71 /* Save the return address (now skip orig rbp, rbp and parent) */ 72 pushq \added+8*3(%rsp) 73 pushq %rbp 74 movq %rsp, %rbp 75#endif /* CONFIG_FRAME_POINTER */ 76 77 /* 78 * We add enough stack to save all regs. 79 */ 80 subq $(FRAME_SIZE), %rsp 81 movq %rax, RAX(%rsp) 82 movq %rcx, RCX(%rsp) 83 movq %rdx, RDX(%rsp) 84 movq %rsi, RSI(%rsp) 85 movq %rdi, RDI(%rsp) 86 movq %r8, R8(%rsp) 87 movq %r9, R9(%rsp) 88 movq $0, ORIG_RAX(%rsp) 89 /* 90 * Save the original RBP. Even though the mcount ABI does not 91 * require this, it helps out callers. 92 */ 93#ifdef CONFIG_FRAME_POINTER 94 movq MCOUNT_REG_SIZE-8(%rsp), %rdx 95#else 96 movq %rbp, %rdx 97#endif 98 movq %rdx, RBP(%rsp) 99 100 /* Copy the parent address into %rsi (second parameter) */ 101 movq MCOUNT_REG_SIZE+8+\added(%rsp), %rsi 102 103 /* Move RIP to its proper location */ 104 movq MCOUNT_REG_SIZE+\added(%rsp), %rdi 105 movq %rdi, RIP(%rsp) 106 107 /* 108 * Now %rdi (the first parameter) has the return address of 109 * where ftrace_call returns. But the callbacks expect the 110 * address of the call itself. 111 */ 112 subq $MCOUNT_INSN_SIZE, %rdi 113 .endm 114 115.macro restore_mcount_regs save=0 116 117 /* ftrace_regs_caller or frame pointers require this */ 118 movq RBP(%rsp), %rbp 119 120 movq R9(%rsp), %r9 121 movq R8(%rsp), %r8 122 movq RDI(%rsp), %rdi 123 movq RSI(%rsp), %rsi 124 movq RDX(%rsp), %rdx 125 movq RCX(%rsp), %rcx 126 movq RAX(%rsp), %rax 127 128 addq $MCOUNT_REG_SIZE-\save, %rsp 129 130 .endm 131 132#ifdef CONFIG_DYNAMIC_FTRACE 133 134SYM_FUNC_START(__fentry__) 135 RET 136SYM_FUNC_END(__fentry__) 137EXPORT_SYMBOL(__fentry__) 138 139SYM_FUNC_START(ftrace_caller) 140 /* save_mcount_regs fills in first two parameters */ 141 save_mcount_regs 142 143 /* Stack - skipping return address of ftrace_caller */ 144 leaq MCOUNT_REG_SIZE+8(%rsp), %rcx 145 movq %rcx, RSP(%rsp) 146 147SYM_INNER_LABEL(ftrace_caller_op_ptr, SYM_L_GLOBAL) 148 ANNOTATE_NOENDBR 149 /* Load the ftrace_ops into the 3rd parameter */ 150 movq function_trace_op(%rip), %rdx 151 152 /* regs go into 4th parameter */ 153 leaq (%rsp), %rcx 154 155 /* Only ops with REGS flag set should have CS register set */ 156 movq $0, CS(%rsp) 157 158SYM_INNER_LABEL(ftrace_call, SYM_L_GLOBAL) 159 ANNOTATE_NOENDBR 160 call ftrace_stub 161 162 /* Handlers can change the RIP */ 163 movq RIP(%rsp), %rax 164 movq %rax, MCOUNT_REG_SIZE(%rsp) 165 166 restore_mcount_regs 167 168 /* 169 * The code up to this label is copied into trampolines so 170 * think twice before adding any new code or changing the 171 * layout here. 172 */ 173SYM_INNER_LABEL(ftrace_caller_end, SYM_L_GLOBAL) 174 ANNOTATE_NOENDBR 175 176 jmp ftrace_epilogue 177SYM_FUNC_END(ftrace_caller); 178 179SYM_FUNC_START(ftrace_epilogue) 180/* 181 * This is weak to keep gas from relaxing the jumps. 182 */ 183SYM_INNER_LABEL_ALIGN(ftrace_stub, SYM_L_WEAK) 184 UNWIND_HINT_FUNC 185 ENDBR 186 RET 187SYM_FUNC_END(ftrace_epilogue) 188 189SYM_FUNC_START(ftrace_regs_caller) 190 /* Save the current flags before any operations that can change them */ 191 pushfq 192 193 /* added 8 bytes to save flags */ 194 save_mcount_regs 8 195 /* save_mcount_regs fills in first two parameters */ 196 197SYM_INNER_LABEL(ftrace_regs_caller_op_ptr, SYM_L_GLOBAL) 198 ANNOTATE_NOENDBR 199 /* Load the ftrace_ops into the 3rd parameter */ 200 movq function_trace_op(%rip), %rdx 201 202 /* Save the rest of pt_regs */ 203 movq %r15, R15(%rsp) 204 movq %r14, R14(%rsp) 205 movq %r13, R13(%rsp) 206 movq %r12, R12(%rsp) 207 movq %r11, R11(%rsp) 208 movq %r10, R10(%rsp) 209 movq %rbx, RBX(%rsp) 210 /* Copy saved flags */ 211 movq MCOUNT_REG_SIZE(%rsp), %rcx 212 movq %rcx, EFLAGS(%rsp) 213 /* Kernel segments */ 214 movq $__KERNEL_DS, %rcx 215 movq %rcx, SS(%rsp) 216 movq $__KERNEL_CS, %rcx 217 movq %rcx, CS(%rsp) 218 /* Stack - skipping return address and flags */ 219 leaq MCOUNT_REG_SIZE+8*2(%rsp), %rcx 220 movq %rcx, RSP(%rsp) 221 222 ENCODE_FRAME_POINTER 223 224 /* regs go into 4th parameter */ 225 leaq (%rsp), %rcx 226 227SYM_INNER_LABEL(ftrace_regs_call, SYM_L_GLOBAL) 228 ANNOTATE_NOENDBR 229 call ftrace_stub 230 231 /* Copy flags back to SS, to restore them */ 232 movq EFLAGS(%rsp), %rax 233 movq %rax, MCOUNT_REG_SIZE(%rsp) 234 235 /* Handlers can change the RIP */ 236 movq RIP(%rsp), %rax 237 movq %rax, MCOUNT_REG_SIZE+8(%rsp) 238 239 /* restore the rest of pt_regs */ 240 movq R15(%rsp), %r15 241 movq R14(%rsp), %r14 242 movq R13(%rsp), %r13 243 movq R12(%rsp), %r12 244 movq R10(%rsp), %r10 245 movq RBX(%rsp), %rbx 246 247 movq ORIG_RAX(%rsp), %rax 248 movq %rax, MCOUNT_REG_SIZE-8(%rsp) 249 250 /* 251 * If ORIG_RAX is anything but zero, make this a call to that. 252 * See arch_ftrace_set_direct_caller(). 253 */ 254 testq %rax, %rax 255SYM_INNER_LABEL(ftrace_regs_caller_jmp, SYM_L_GLOBAL) 256 ANNOTATE_NOENDBR 257 jnz 1f 258 259 restore_mcount_regs 260 /* Restore flags */ 261 popfq 262 263 /* 264 * As this jmp to ftrace_epilogue can be a short jump 265 * it must not be copied into the trampoline. 266 * The trampoline will add the code to jump 267 * to the return. 268 */ 269SYM_INNER_LABEL(ftrace_regs_caller_end, SYM_L_GLOBAL) 270 ANNOTATE_NOENDBR 271 jmp ftrace_epilogue 272 273 /* Swap the flags with orig_rax */ 2741: movq MCOUNT_REG_SIZE(%rsp), %rdi 275 movq %rdi, MCOUNT_REG_SIZE-8(%rsp) 276 movq %rax, MCOUNT_REG_SIZE(%rsp) 277 278 restore_mcount_regs 8 279 /* Restore flags */ 280 popfq 281 UNWIND_HINT_FUNC 282 jmp ftrace_epilogue 283 284SYM_FUNC_END(ftrace_regs_caller) 285 286 287#else /* ! CONFIG_DYNAMIC_FTRACE */ 288 289SYM_FUNC_START(__fentry__) 290 cmpq $ftrace_stub, ftrace_trace_function 291 jnz trace 292 293SYM_INNER_LABEL(ftrace_stub, SYM_L_GLOBAL) 294 ENDBR 295 RET 296 297trace: 298 /* save_mcount_regs fills in first two parameters */ 299 save_mcount_regs 300 301 /* 302 * When DYNAMIC_FTRACE is not defined, ARCH_SUPPORTS_FTRACE_OPS is not 303 * set (see include/asm/ftrace.h and include/linux/ftrace.h). Only the 304 * ip and parent ip are used and the list function is called when 305 * function tracing is enabled. 306 */ 307 movq ftrace_trace_function, %r8 308 CALL_NOSPEC r8 309 restore_mcount_regs 310 311 jmp ftrace_stub 312SYM_FUNC_END(__fentry__) 313EXPORT_SYMBOL(__fentry__) 314#endif /* CONFIG_DYNAMIC_FTRACE */ 315 316#ifdef CONFIG_FUNCTION_GRAPH_TRACER 317SYM_FUNC_START(return_to_handler) 318 subq $16, %rsp 319 320 /* Save the return values */ 321 movq %rax, (%rsp) 322 movq %rdx, 8(%rsp) 323 movq %rbp, %rdi 324 325 call ftrace_return_to_handler 326 327 movq %rax, %rdi 328 movq 8(%rsp), %rdx 329 movq (%rsp), %rax 330 331 addq $16, %rsp 332 /* 333 * Jump back to the old return address. This cannot be JMP_NOSPEC rdi 334 * since IBT would demand that contain ENDBR, which simply isn't so for 335 * return addresses. Use a retpoline here to keep the RSB balanced. 336 */ 337 ANNOTATE_INTRA_FUNCTION_CALL 338 call .Ldo_rop 339 int3 340.Ldo_rop: 341 mov %rdi, (%rsp) 342 UNWIND_HINT_FUNC 343 RET 344SYM_FUNC_END(return_to_handler) 345#endif 346