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 .entry.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 (SS+8 + 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 $(MCOUNT_REG_SIZE - MCOUNT_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 retq 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 143SYM_INNER_LABEL(ftrace_caller_op_ptr, SYM_L_GLOBAL) 144 /* Load the ftrace_ops into the 3rd parameter */ 145 movq function_trace_op(%rip), %rdx 146 147 /* regs go into 4th parameter (but make it NULL) */ 148 movq $0, %rcx 149 150SYM_INNER_LABEL(ftrace_call, SYM_L_GLOBAL) 151 call ftrace_stub 152 153 restore_mcount_regs 154 155 /* 156 * The code up to this label is copied into trampolines so 157 * think twice before adding any new code or changing the 158 * layout here. 159 */ 160SYM_INNER_LABEL(ftrace_epilogue, SYM_L_GLOBAL) 161 162#ifdef CONFIG_FUNCTION_GRAPH_TRACER 163SYM_INNER_LABEL(ftrace_graph_call, SYM_L_GLOBAL) 164 jmp ftrace_stub 165#endif 166 167/* 168 * This is weak to keep gas from relaxing the jumps. 169 * It is also used to copy the retq for trampolines. 170 */ 171SYM_INNER_LABEL_ALIGN(ftrace_stub, SYM_L_WEAK) 172 retq 173SYM_FUNC_END(ftrace_caller) 174 175SYM_FUNC_START(ftrace_regs_caller) 176 /* Save the current flags before any operations that can change them */ 177 pushfq 178 179 UNWIND_HINT_SAVE 180 181 /* added 8 bytes to save flags */ 182 save_mcount_regs 8 183 /* save_mcount_regs fills in first two parameters */ 184 185SYM_INNER_LABEL(ftrace_regs_caller_op_ptr, SYM_L_GLOBAL) 186 /* Load the ftrace_ops into the 3rd parameter */ 187 movq function_trace_op(%rip), %rdx 188 189 /* Save the rest of pt_regs */ 190 movq %r15, R15(%rsp) 191 movq %r14, R14(%rsp) 192 movq %r13, R13(%rsp) 193 movq %r12, R12(%rsp) 194 movq %r11, R11(%rsp) 195 movq %r10, R10(%rsp) 196 movq %rbx, RBX(%rsp) 197 /* Copy saved flags */ 198 movq MCOUNT_REG_SIZE(%rsp), %rcx 199 movq %rcx, EFLAGS(%rsp) 200 /* Kernel segments */ 201 movq $__KERNEL_DS, %rcx 202 movq %rcx, SS(%rsp) 203 movq $__KERNEL_CS, %rcx 204 movq %rcx, CS(%rsp) 205 /* Stack - skipping return address and flags */ 206 leaq MCOUNT_REG_SIZE+8*2(%rsp), %rcx 207 movq %rcx, RSP(%rsp) 208 209 ENCODE_FRAME_POINTER 210 211 /* regs go into 4th parameter */ 212 leaq (%rsp), %rcx 213 214SYM_INNER_LABEL(ftrace_regs_call, SYM_L_GLOBAL) 215 call ftrace_stub 216 217 /* Copy flags back to SS, to restore them */ 218 movq EFLAGS(%rsp), %rax 219 movq %rax, MCOUNT_REG_SIZE(%rsp) 220 221 /* Handlers can change the RIP */ 222 movq RIP(%rsp), %rax 223 movq %rax, MCOUNT_REG_SIZE+8(%rsp) 224 225 /* restore the rest of pt_regs */ 226 movq R15(%rsp), %r15 227 movq R14(%rsp), %r14 228 movq R13(%rsp), %r13 229 movq R12(%rsp), %r12 230 movq R10(%rsp), %r10 231 movq RBX(%rsp), %rbx 232 233 movq ORIG_RAX(%rsp), %rax 234 movq %rax, MCOUNT_REG_SIZE-8(%rsp) 235 236 /* If ORIG_RAX is anything but zero, make this a call to that */ 237 movq ORIG_RAX(%rsp), %rax 238 cmpq $0, %rax 239 je 1f 240 241 /* Swap the flags with orig_rax */ 242 movq MCOUNT_REG_SIZE(%rsp), %rdi 243 movq %rdi, MCOUNT_REG_SIZE-8(%rsp) 244 movq %rax, MCOUNT_REG_SIZE(%rsp) 245 246 restore_mcount_regs 8 247 248 jmp 2f 249 2501: restore_mcount_regs 251 252 2532: 254 /* 255 * The stack layout is nondetermistic here, depending on which path was 256 * taken. This confuses objtool and ORC, rightfully so. For now, 257 * pretend the stack always looks like the non-direct case. 258 */ 259 UNWIND_HINT_RESTORE 260 261 /* Restore flags */ 262 popfq 263 264 /* 265 * As this jmp to ftrace_epilogue can be a short jump 266 * it must not be copied into the trampoline. 267 * The trampoline will add the code to jump 268 * to the return. 269 */ 270SYM_INNER_LABEL(ftrace_regs_caller_end, SYM_L_GLOBAL) 271 272 jmp ftrace_epilogue 273 274SYM_FUNC_END(ftrace_regs_caller) 275 276 277#else /* ! CONFIG_DYNAMIC_FTRACE */ 278 279SYM_FUNC_START(__fentry__) 280 cmpq $ftrace_stub, ftrace_trace_function 281 jnz trace 282 283fgraph_trace: 284#ifdef CONFIG_FUNCTION_GRAPH_TRACER 285 cmpq $ftrace_stub, ftrace_graph_return 286 jnz ftrace_graph_caller 287 288 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry 289 jnz ftrace_graph_caller 290#endif 291 292SYM_INNER_LABEL(ftrace_stub, SYM_L_GLOBAL) 293 retq 294 295trace: 296 /* save_mcount_regs fills in first two parameters */ 297 save_mcount_regs 298 299 /* 300 * When DYNAMIC_FTRACE is not defined, ARCH_SUPPORTS_FTRACE_OPS is not 301 * set (see include/asm/ftrace.h and include/linux/ftrace.h). Only the 302 * ip and parent ip are used and the list function is called when 303 * function tracing is enabled. 304 */ 305 movq ftrace_trace_function, %r8 306 CALL_NOSPEC %r8 307 restore_mcount_regs 308 309 jmp fgraph_trace 310SYM_FUNC_END(__fentry__) 311EXPORT_SYMBOL(__fentry__) 312#endif /* CONFIG_DYNAMIC_FTRACE */ 313 314#ifdef CONFIG_FUNCTION_GRAPH_TRACER 315SYM_FUNC_START(ftrace_graph_caller) 316 /* Saves rbp into %rdx and fills first parameter */ 317 save_mcount_regs 318 319 leaq MCOUNT_REG_SIZE+8(%rsp), %rsi 320 movq $0, %rdx /* No framepointers needed */ 321 call prepare_ftrace_return 322 323 restore_mcount_regs 324 325 retq 326SYM_FUNC_END(ftrace_graph_caller) 327 328SYM_CODE_START(return_to_handler) 329 UNWIND_HINT_EMPTY 330 subq $24, %rsp 331 332 /* Save the return values */ 333 movq %rax, (%rsp) 334 movq %rdx, 8(%rsp) 335 movq %rbp, %rdi 336 337 call ftrace_return_to_handler 338 339 movq %rax, %rdi 340 movq 8(%rsp), %rdx 341 movq (%rsp), %rax 342 addq $24, %rsp 343 JMP_NOSPEC %rdi 344SYM_CODE_END(return_to_handler) 345#endif 346