include/ust/processor.h

   1 /* Copyright (C) 2009  Pierre-Marc Fournier
   2  *
   3  * This library is free software; you can redistribute it and/or
   4  * modify it under the terms of the GNU Lesser General Public
   5  * License as published by the Free Software Foundation; either
   6  * version 2.1 of the License, or (at your option) any later version.
   7  *
   8  * This library is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * Lesser General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU Lesser General Public
  14  * License along with this library; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301 USA
  16  */
  17
  18 #ifndef UST_PROCESSOR_H
  19 #define UST_PROCESSOR_H
  20
  21 #include <stddef.h>
  22 #include <string.h>
  23 #include <urcu-bp.h>
  24
  25 extern __thread long ust_reg_stack[500];
  26 extern volatile __thread long *ust_reg_stack_ptr;
  27
  28 #define ____cacheline_aligned __attribute__((aligned(CAA_CACHE_LINE_SIZE)))
  29
  30 #ifdef __i386
  31
  32 struct registers {
  33         short ss;
  34         short cs;
  35         long esi;
  36         long ebp;
  37         long edx;
  38         long edi;
  39         long ecx;
  40         long ebx;
  41         long eax;
  42         long eflags;
  43         long esp;
  44 };
  45
  46 static inline int fls(int x)
  47 {
  48         int r;
  49         asm("bsrl %1,%0\n\t"
  50             "cmovzl %2,%0"
  51             : "=&r" (r) : "rm" (x), "rm" (-1));
  52         return r + 1;
  53 }
  54
  55 #ifdef CONFIG_UST_GDB_INTEGRATION
  56
  57 /* save_registers - saves most of the processor's registers so
  58  * they are available to the probe. gdb uses this to give the
  59  * value of local variables.
  60  *
  61  * Saving all registers without losing any of their values is
  62  * tricky.
  63  *
  64  * We cannot pass to the asm stub the address of a registers structure
  65  * on the stack, because it will use a register and override its value.
  66  *
  67  * We don't want to use a stub to push the regs on the stack and then
  68  * another stub to copy them to a structure because changing %sp in asm
  69  * and then returning to C (even briefly) can have unexpected results.
  70  * Also, gcc might modify %sp between the stubs in reaction to the
  71  * register needs of the second stub that needs to know where to copy
  72  * the register values.
  73  *
  74  * So the chosen approach is to use another stack, declared in thread-
  75  * local storage, to push the registers. They are subsequently copied
  76  * to the stack, by C code.
  77  */
  78
  79 #define save_registers(regsptr) \
  80         asm volatile ( \
  81              /* save original esp */ \
  82              "pushl %%esp\n\t" \
  83              /* push original eflags */ \
  84              "pushfl\n\t" \
  85               /* eax will hold the ptr to the private stack bottom */ \
  86              "pushl %%eax\n\t" \
  87              /* ebx is used for TLS access */ \
  88              "pushl %%ebx\n\t" \
  89              /* ecx will be used to temporarily hold the stack bottom addr */\
  90              "pushl %%ecx\n\t"                                               \
  91              /* rdi is the input to __tls_get_addr, and also a temp var */   \
  92              "pushl %%edi\n\t"                                               \
  93              /* For TLS access, we have to do function calls. However,       \
  94               * we must not lose the original value of:                      \
  95               *   esp, eflags, eax, ebx, ecx, edx, esi, edi, ebp, cs, ss     \
  96               *                                                              \
  97               * Some registers' original values have already been saved:     \
  98               *   esp, eflags, eax, ebx, ecx, edi                            \
  99               *                                                              \
 100               * In addition, the i386 ABI says the following registers belong\
 101               * to the caller function:                                      \
 102               *   esp, ebp, esi, edi, ebx                                    \
 103               *                                                              \
 104               * The following registers should not be changed by the callee: \
 105               *   cs, ss                                                     \
 106               *                                                              \
 107               * Therefore, the following registers must be explicitly        \
 108               * preserved:                                                   \
 109               *   edx                                                        \
 110               */ \
 111              "pushl %%edx\n\t" \
 112              /* Get GOT address */ \
 113              "call __i686.get_pc_thunk.bx\n\t" \
 114              "addl $_GLOBAL_OFFSET_TABLE_, %%ebx\n\t" \
 115              /* Start TLS access of private reg stack pointer */ \
 116              "leal ust_reg_stack_ptr@tlsgd(,%%ebx,1),%%eax\n\t" \
 117              "call ___tls_get_addr@plt\n\t" \
 118              /* --- End TLS access */ \
 119              /* check if ust_reg_stack_ptr has been initialized */ \
 120              "movl (%%eax),%%ecx\n\t" \
 121              "testl %%ecx,%%ecx\n\t" \
 122              "jne 1f\n\t" \
 123              "movl %%eax,%%ecx\n\t" \
 124              /* Save ecx because we are using it. */ \
 125              "pushl %%ecx\n\t" \
 126              /* Start TLS access of private reg stack */ \
 127              "leal ust_reg_stack@tlsgd(,%%ebx,1),%%eax\n\t" \
 128              "call ___tls_get_addr@plt\n\t" \
 129              /* --- End TLS access */ \
 130              "popl %%ecx\n\t" \
 131              "addl $500,%%eax\n\t" \
 132              "movl %%eax,(%%ecx)\n\t" \
 133              "movl %%ecx,%%eax\n\t" \
 134              /* now the pointer to the private stack is in eax. \
 135                 must add stack size so the ptr points to the stack bottom. */ \
 136         "1:\n\t" \
 137              /* edx was pushed for function calls */ \
 138              "popl %%edx\n\t" \
 139              /* Manually push esp to private stack */ \
 140              "addl $-4,(%%eax)\n\t" \
 141              "movl 20(%%esp), %%edi\n\t" \
 142              "movl (%%eax), %%ebx\n\t" \
 143              "movl %%edi, (%%ebx)\n\t" \
 144              /* Manually push eflags to private stack */ \
 145              "addl $-4,(%%eax)\n\t" \
 146              "movl 16(%%esp), %%edi\n\t" \
 147              "movl (%%eax), %%ebx\n\t" \
 148              "movl %%edi, (%%ebx)\n\t" \
 149              /* Manually push eax to private stack */ \
 150              "addl $-4,(%%eax)\n\t" \
 151              "movl 12(%%esp), %%edi\n\t" \
 152              "movl (%%eax), %%ebx\n\t" \
 153              "movl %%edi, (%%ebx)\n\t" \
 154              /* Manually push ebx to private stack */ \
 155              "addl $-4,(%%eax)\n\t" \
 156              "movl 8(%%esp), %%edi\n\t" \
 157              "movl (%%eax), %%ebx\n\t" \
 158              "movl %%edi, (%%ebx)\n\t" \
 159              /* Manually push ecx to private stack */ \
 160              "addl $-4,(%%eax)\n\t" \
 161              "movl 4(%%esp), %%edi\n\t" \
 162              "movl (%%eax), %%ebx\n\t" \
 163              "movl %%edi, (%%ebx)\n\t" \
 164              /* Manually push edi to private stack */ \
 165              "addl $-4,(%%eax)\n\t" \
 166              "movl 0(%%esp), %%edi\n\t" \
 167              "movl (%%eax), %%ebx\n\t" \
 168              "movl %%edi, (%%ebx)\n\t" \
 169              /* now push regs to tls */ \
 170              /* -- esp already pushed -- */ \
 171              /* -- eax already pushed -- */ \
 172              /* -- ebx already pushed -- */ \
 173              /* -- ecx already pushed -- */ \
 174              /* -- edi already pushed -- */ \
 175              "addl $-4,(%%eax)\n\t" \
 176              "movl (%%eax), %%ebx\n\t" \
 177              "movl %%edx,(%%ebx)\n\t" \
 178              "addl $-4,(%%eax)\n\t" \
 179              "movl (%%eax), %%ebx\n\t" \
 180              "movl %%ebp,(%%ebx)\n\t" \
 181              "addl $-4,(%%eax)\n\t" \
 182              "movl (%%eax), %%ebx\n\t" \
 183              "movl %%esi,(%%ebx)\n\t" \
 184              /* push cs */ \
 185              "addl $-2,(%%eax)\n\t" \
 186              "movl (%%eax), %%ebx\n\t" \
 187              "movw %%cs, (%%ebx)\n\t" \
 188              /* push ss */ \
 189              "addl $-2,(%%eax)\n\t" \
 190              "movl (%%eax), %%ebx\n\t" \
 191              "movw %%ss, (%%ebx)\n\t" \
 192              /* restore original values of regs that were used internally */ \
 193              "popl %%edi\n\t" \
 194              "popl %%ecx\n\t" \
 195              "popl %%ebx\n\t" \
 196              "popl %%eax\n\t" \
 197              /* cancel push of rsp */ \
 198              "addl $4,%%esp\n\t" \
 199              /* cancel push of eflags */ \
 200              "addl $4,%%esp\n\t" \
 201              ::: "memory"); \
 202         memcpy(regsptr, (void *)ust_reg_stack_ptr, sizeof(struct registers)); \
 203         ust_reg_stack_ptr = (void *)(((long)ust_reg_stack_ptr) + sizeof(struct registers));
 204
 205 #else /* CONFIG_UST_GDB_INTEGRATION */
 206
 207 #define save_registers(a)
 208
 209 #endif /* CONFIG_UST_GDB_INTEGRATION */
 210
 211 #define RELATIVE_ADDRESS(__rel_label__) __rel_label__
 212
 213 #define ARCH_COPY_ADDR(dst) "lea 2b," dst "\n\t"
 214
 215 #define _ASM_PTR ".long "
 216
 217 #elif defined(__x86_64)
 218
 219 struct registers {
 220         int padding; /* 4 bytes */
 221         short ss;
 222         short cs;
 223         unsigned long r15;
 224         unsigned long r14;
 225         unsigned long r13;
 226         unsigned long r12;
 227         unsigned long r11;
 228         unsigned long r10;
 229         unsigned long r9;
 230         unsigned long r8;
 231         unsigned long rsi;
 232         unsigned long rbp;
 233         unsigned long rdx;
 234         unsigned long rcx;
 235         unsigned long rdi;
 236         unsigned long rbx;
 237         unsigned long rax;
 238         unsigned long rflags;
 239         unsigned long rsp;
 240 };
 241
 242 static inline int fls(int x)
 243 {
 244         int r;
 245         asm("bsrl %1,%0\n\t"
 246             "cmovzl %2,%0"
 247             : "=&r" (r) : "rm" (x), "rm" (-1));
 248         return r + 1;
 249 }
 250
 251 #ifdef CONFIG_UST_GDB_INTEGRATION
 252
 253 #define save_registers(regsptr) \
 254         asm volatile ( \
 255              /* save original rsp */ \
 256              "pushq %%rsp\n\t" \
 257              /* push original rflags */ \
 258              "pushfq\n\t" \
 259               /* rax will hold the ptr to the private stack bottom */ \
 260              "pushq %%rax\n\t" \
 261              /* rbx will be used to temporarily hold the stack bottom addr */ \
 262              "pushq %%rbx\n\t" \
 263              /* rdi is the input to __tls_get_addr, and also a temp var */ \
 264              "pushq %%rdi\n\t" \
 265              /* For TLS access, we have to do function calls. However,       \
 266               * we must not lose the original value of:                      \
 267               *   rsp, rflags, rax, rbx, rcx, rdx, rsi, rdi, rbp, r8, r9     \
 268               *   r10, r11, r12, r13, r14, r15, cs, ss                       \
 269               *                                                              \
 270               * Some registers' original values have already been saved:     \
 271               *   rsp, rflags, rax, rbx, rdi                                 \
 272               *                                                              \
 273               * In addition, the x86-64 ABI says the following registers     \
 274               * belong to the caller function:                               \
 275               *   rbp, rbx, r12, r13, r14, r15                               \
 276               *                                                              \
 277               * The following registers should not be changed by the callee: \
 278               *   cs, ss                                                     \
 279               *                                                              \
 280               * Therefore, the following registers must be explicitly        \
 281               * preserved:                                                   \
 282               *   rcx, rdx, rsi, r8, r9, r10, r11                            \
 283               */ \
 284              "pushq %%rcx\n\t" \
 285              "pushq %%rdx\n\t" \
 286              "pushq %%rsi\n\t" \
 287              "pushq %%r8\n\t" \
 288              "pushq %%r9\n\t" \
 289              "pushq %%r10\n\t" \
 290              "pushq %%r11\n\t" \
 291              /* Start TLS access of private reg stack pointer */ \
 292              ".byte 0x66\n\t" \
 293              "leaq ust_reg_stack_ptr@tlsgd(%%rip), %%rdi\n\t" \
 294              ".hword 0x6666\n\t" \
 295              "rex64\n\t" \
 296              "call __tls_get_addr@plt\n\t" \
 297              /* --- End TLS access */ \
 298              /* check if ust_reg_stack_ptr has been initialized */ \
 299              "movq (%%rax),%%rbx\n\t" \
 300              "testq %%rbx,%%rbx\n\t" \
 301              "jne 1f\n\t" \
 302              "movq %%rax,%%rbx\n\t" \
 303              /* Start TLS access of private reg stack */ \
 304              ".byte 0x66\n\t" \
 305              "leaq ust_reg_stack@tlsgd(%%rip), %%rdi\n\t" \
 306              ".hword 0x6666\n\t" \
 307              "rex64\n\t" \
 308              "call __tls_get_addr@plt\n\t" \
 309              /* --- End TLS access */ \
 310              "addq $500,%%rax\n\t" \
 311              "movq %%rax,(%%rbx)\n\t" \
 312              "movq %%rbx,%%rax\n\t" \
 313              /* now the pointer to the private stack is in rax.
 314                 must add stack size so the ptr points to the stack bottom. */ \
 315         "1:\n\t" \
 316              /* Pop regs that were pushed for function calls */ \
 317              "popq %%r11\n\t" \
 318              "popq %%r10\n\t" \
 319              "popq %%r9\n\t" \
 320              "popq %%r8\n\t" \
 321              "popq %%rsi\n\t" \
 322              "popq %%rdx\n\t" \
 323              "popq %%rcx\n\t" \
 324              /* Manually push rsp to private stack */ \
 325              "addq $-8,(%%rax)\n\t" \
 326              "movq 32(%%rsp), %%rdi\n\t" \
 327              "movq (%%rax), %%rbx\n\t" \
 328              "movq %%rdi, (%%rbx)\n\t" \
 329              /* Manually push eflags to private stack */ \
 330              "addq $-8,(%%rax)\n\t" \
 331              "movq 24(%%rsp), %%rdi\n\t" \
 332              "movq (%%rax), %%rbx\n\t" \
 333              "movq %%rdi, (%%rbx)\n\t" \
 334              /* Manually push rax to private stack */ \
 335              "addq $-8,(%%rax)\n\t" \
 336              "movq 16(%%rsp), %%rdi\n\t" \
 337              "movq (%%rax), %%rbx\n\t" \
 338              "movq %%rdi, (%%rbx)\n\t" \
 339              /* Manually push rbx to private stack */ \
 340              "addq $-8,(%%rax)\n\t" \
 341              "movq 8(%%rsp), %%rdi\n\t" \
 342              "movq (%%rax), %%rbx\n\t" \
 343              "movq %%rdi, (%%rbx)\n\t" \
 344              /* Manually push rdi to private stack */ \
 345              "addq $-8,(%%rax)\n\t" \
 346              "movq 0(%%rsp), %%rdi\n\t" \
 347              "movq (%%rax), %%rbx\n\t" \
 348              "movq %%rdi, (%%rbx)\n\t" \
 349              /* now push regs to tls */ \
 350              /* -- rsp already pushed -- */ \
 351              /* -- rax already pushed -- */ \
 352              /* -- rbx already pushed -- */ \
 353              /* -- rdi already pushed -- */ \
 354              "addq $-8,(%%rax)\n\t" \
 355              "movq (%%rax), %%rbx\n\t" \
 356              "movq %%rcx,(%%rbx)\n\t" \
 357              "addq $-8,(%%rax)\n\t" \
 358              "movq (%%rax), %%rbx\n\t" \
 359              "movq %%rdx,(%%rbx)\n\t" \
 360              "addq $-8,(%%rax)\n\t" \
 361              "movq (%%rax), %%rbx\n\t" \
 362              "movq %%rbp,(%%rbx)\n\t" \
 363              "addq $-8,(%%rax)\n\t" \
 364              "movq (%%rax), %%rbx\n\t" \
 365              "movq %%rsi,(%%rbx)\n\t" \
 366              "addq $-8,(%%rax)\n\t" \
 367              "movq (%%rax), %%rbx\n\t" \
 368              "movq %%r8,(%%rbx)\n\t" \
 369              "addq $-8,(%%rax)\n\t" \
 370              "movq (%%rax), %%rbx\n\t" \
 371              "movq %%r9,(%%rbx)\n\t" \
 372              "addq $-8,(%%rax)\n\t" \
 373              "movq (%%rax), %%rbx\n\t" \
 374              "movq %%r10,(%%rbx)\n\t" \
 375              "addq $-8,(%%rax)\n\t" \
 376              "movq (%%rax), %%rbx\n\t" \
 377              "movq %%r11,(%%rbx)\n\t" \
 378              "addq $-8,(%%rax)\n\t" \
 379              "movq (%%rax), %%rbx\n\t" \
 380              "movq %%r12,(%%rbx)\n\t" \
 381              "addq $-8,(%%rax)\n\t" \
 382              "movq (%%rax), %%rbx\n\t" \
 383              "movq %%r13,(%%rbx)\n\t" \
 384              "addq $-8,(%%rax)\n\t" \
 385              "movq (%%rax), %%rbx\n\t" \
 386              "movq %%r14,(%%rbx)\n\t" \
 387              "addq $-8,(%%rax)\n\t" \
 388              "movq (%%rax), %%rbx\n\t" \
 389              "movq %%r15,(%%rbx)\n\t" \
 390              /* push cs */ \
 391              "addq $-2,(%%rax)\n\t" \
 392              "movq (%%rax), %%rbx\n\t" \
 393              "movw %%cs, (%%rbx)\n\t" \
 394              /* push ss */ \
 395              "addq $-2,(%%rax)\n\t" \
 396              "movq (%%rax), %%rbx\n\t" \
 397              "movw %%ss, (%%rbx)\n\t" \
 398              /* add padding for struct registers */ \
 399              "addq $-4,(%%rax)\n\t" \
 400              /* restore original values of regs that were used internally */ \
 401              "popq %%rdi\n\t" \
 402              "popq %%rbx\n\t" \
 403              "popq %%rax\n\t" \
 404              /* cancel push of rsp */ \
 405              "addq $8,%%rsp\n\t" \
 406              /* cancel push of rflags */ \
 407              "addq $8,%%rsp\n\t" \
 408              ::); \
 409         memcpy(regsptr, (void *)ust_reg_stack_ptr, sizeof(struct registers)); \
 410         ust_reg_stack_ptr = (void *)(((long)ust_reg_stack_ptr) + sizeof(struct registers));
 411
 412 #else /* CONFIG_UST_GDB_INTEGRATION */
 413
 414 #define save_registers(a)
 415
 416 #endif /* CONFIG_UST_GDB_INTEGRATION */
 417
 418 /* Macro to insert the address of a relative jump in an assembly stub,
 419  * in a relocatable way. On x86-64, this uses a special (%rip) notation. */
 420 #define RELATIVE_ADDRESS(__rel_label__) __rel_label__(%%rip)
 421
 422 #define ARCH_COPY_ADDR(dst) "lea 2b(%%rip)," dst "\n\t"
 423
 424 #define _ASM_PTR ".quad "
 425
 426 #elif defined(__PPC__)
 427
 428 struct registers {
 429 };
 430
 431 static __inline__ int fls(unsigned int x)
 432 {
 433         int lz;
 434
 435         asm ("cntlzw %0,%1" : "=r" (lz) : "r" (x));
 436         return 32 - lz;
 437 }
 438
 439 #define ARCH_COPY_ADDR(dst) \
 440                      "lis " dst ",2b@h\n\t" /* load high bytes */ \
 441                      "ori " dst "," dst ",2b@l\n\t" /* load low bytes */
 442
 443 #define _ASM_PTR ".long "
 444 #define save_registers(a)
 445
 446 #else /* arch-agnostic */
 447
 448 static __inline__ int fls(unsigned int x)
 449 {
 450         int r = 32;
 451
 452         if (!x)
 453                 return 0;
 454         if (!(x & 0xFFFF0000U)) {
 455                 x <<= 16;
 456                 r -= 16;
 457         }
 458         if (!(x & 0xFF000000U)) {
 459                 x <<= 8;
 460                 r -= 8;
 461         }
 462         if (!(x & 0xF0000000U)) {
 463                 x <<= 4;
 464                 r -= 4;
 465         }
 466         if (!(x & 0xC0000000U)) {
 467                 x <<= 2;
 468                 r -= 2;
 469         }
 470         if (!(x & 0x80000000U)) {
 471                 x <<= 1;
 472                 r -= 1;
 473         }
 474         return r;
 475 }
 476
 477 #endif
 478
 479 #ifdef __arm__
 480
 481 struct registers {
 482 };
 483
 484 #define ARCH_COPY_ADDR(dst) "ldr "dst", =2b\n\t" \
 485                 "b 55f\n\t" \
 486                 ".ltorg\n\t" \
 487                 "55:\n\t"
 488
 489 #define _ASM_PTR ".long "
 490 #define save_registers(a)
 491
 492 #endif /* __arm__ */
 493
 494 #endif /* UST_PROCESSOR_H */