b.liu | e958203 | 2025-04-17 19:18:16 +0800 | [diff] [blame^] | 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * linux/arch/alpha/kernel/process.c |
| 4 | * |
| 5 | * Copyright (C) 1995 Linus Torvalds |
| 6 | */ |
| 7 | |
| 8 | /* |
| 9 | * This file handles the architecture-dependent parts of process handling. |
| 10 | */ |
| 11 | |
| 12 | #include <linux/errno.h> |
| 13 | #include <linux/module.h> |
| 14 | #include <linux/sched.h> |
| 15 | #include <linux/sched/debug.h> |
| 16 | #include <linux/sched/task.h> |
| 17 | #include <linux/sched/task_stack.h> |
| 18 | #include <linux/kernel.h> |
| 19 | #include <linux/mm.h> |
| 20 | #include <linux/smp.h> |
| 21 | #include <linux/stddef.h> |
| 22 | #include <linux/unistd.h> |
| 23 | #include <linux/ptrace.h> |
| 24 | #include <linux/user.h> |
| 25 | #include <linux/time.h> |
| 26 | #include <linux/major.h> |
| 27 | #include <linux/stat.h> |
| 28 | #include <linux/vt.h> |
| 29 | #include <linux/mman.h> |
| 30 | #include <linux/elfcore.h> |
| 31 | #include <linux/reboot.h> |
| 32 | #include <linux/tty.h> |
| 33 | #include <linux/console.h> |
| 34 | #include <linux/slab.h> |
| 35 | #include <linux/rcupdate.h> |
| 36 | |
| 37 | #include <asm/reg.h> |
| 38 | #include <linux/uaccess.h> |
| 39 | #include <asm/io.h> |
| 40 | #include <asm/pgtable.h> |
| 41 | #include <asm/hwrpb.h> |
| 42 | #include <asm/fpu.h> |
| 43 | |
| 44 | #include "proto.h" |
| 45 | #include "pci_impl.h" |
| 46 | |
| 47 | /* |
| 48 | * Power off function, if any |
| 49 | */ |
| 50 | void (*pm_power_off)(void) = machine_power_off; |
| 51 | EXPORT_SYMBOL(pm_power_off); |
| 52 | |
| 53 | #ifdef CONFIG_ALPHA_WTINT |
| 54 | /* |
| 55 | * Sleep the CPU. |
| 56 | * EV6, LCA45 and QEMU know how to power down, skipping N timer interrupts. |
| 57 | */ |
| 58 | void arch_cpu_idle(void) |
| 59 | { |
| 60 | wtint(0); |
| 61 | local_irq_enable(); |
| 62 | } |
| 63 | |
| 64 | void arch_cpu_idle_dead(void) |
| 65 | { |
| 66 | wtint(INT_MAX); |
| 67 | } |
| 68 | #endif /* ALPHA_WTINT */ |
| 69 | |
| 70 | struct halt_info { |
| 71 | int mode; |
| 72 | char *restart_cmd; |
| 73 | }; |
| 74 | |
| 75 | static void |
| 76 | common_shutdown_1(void *generic_ptr) |
| 77 | { |
| 78 | struct halt_info *how = (struct halt_info *)generic_ptr; |
| 79 | struct percpu_struct *cpup; |
| 80 | unsigned long *pflags, flags; |
| 81 | int cpuid = smp_processor_id(); |
| 82 | |
| 83 | /* No point in taking interrupts anymore. */ |
| 84 | local_irq_disable(); |
| 85 | |
| 86 | cpup = (struct percpu_struct *) |
| 87 | ((unsigned long)hwrpb + hwrpb->processor_offset |
| 88 | + hwrpb->processor_size * cpuid); |
| 89 | pflags = &cpup->flags; |
| 90 | flags = *pflags; |
| 91 | |
| 92 | /* Clear reason to "default"; clear "bootstrap in progress". */ |
| 93 | flags &= ~0x00ff0001UL; |
| 94 | |
| 95 | #ifdef CONFIG_SMP |
| 96 | /* Secondaries halt here. */ |
| 97 | if (cpuid != boot_cpuid) { |
| 98 | flags |= 0x00040000UL; /* "remain halted" */ |
| 99 | *pflags = flags; |
| 100 | set_cpu_present(cpuid, false); |
| 101 | set_cpu_possible(cpuid, false); |
| 102 | halt(); |
| 103 | } |
| 104 | #endif |
| 105 | |
| 106 | if (how->mode == LINUX_REBOOT_CMD_RESTART) { |
| 107 | if (!how->restart_cmd) { |
| 108 | flags |= 0x00020000UL; /* "cold bootstrap" */ |
| 109 | } else { |
| 110 | /* For SRM, we could probably set environment |
| 111 | variables to get this to work. We'd have to |
| 112 | delay this until after srm_paging_stop unless |
| 113 | we ever got srm_fixup working. |
| 114 | |
| 115 | At the moment, SRM will use the last boot device, |
| 116 | but the file and flags will be the defaults, when |
| 117 | doing a "warm" bootstrap. */ |
| 118 | flags |= 0x00030000UL; /* "warm bootstrap" */ |
| 119 | } |
| 120 | } else { |
| 121 | flags |= 0x00040000UL; /* "remain halted" */ |
| 122 | } |
| 123 | *pflags = flags; |
| 124 | |
| 125 | #ifdef CONFIG_SMP |
| 126 | /* Wait for the secondaries to halt. */ |
| 127 | set_cpu_present(boot_cpuid, false); |
| 128 | set_cpu_possible(boot_cpuid, false); |
| 129 | while (cpumask_weight(cpu_present_mask)) |
| 130 | barrier(); |
| 131 | #endif |
| 132 | |
| 133 | /* If booted from SRM, reset some of the original environment. */ |
| 134 | if (alpha_using_srm) { |
| 135 | #ifdef CONFIG_DUMMY_CONSOLE |
| 136 | /* If we've gotten here after SysRq-b, leave interrupt |
| 137 | context before taking over the console. */ |
| 138 | if (in_interrupt()) |
| 139 | irq_exit(); |
| 140 | /* This has the effect of resetting the VGA video origin. */ |
| 141 | console_lock(); |
| 142 | do_take_over_console(&dummy_con, 0, MAX_NR_CONSOLES-1, 1); |
| 143 | console_unlock(); |
| 144 | #endif |
| 145 | pci_restore_srm_config(); |
| 146 | set_hae(srm_hae); |
| 147 | } |
| 148 | |
| 149 | if (alpha_mv.kill_arch) |
| 150 | alpha_mv.kill_arch(how->mode); |
| 151 | |
| 152 | if (! alpha_using_srm && how->mode != LINUX_REBOOT_CMD_RESTART) { |
| 153 | /* Unfortunately, since MILO doesn't currently understand |
| 154 | the hwrpb bits above, we can't reliably halt the |
| 155 | processor and keep it halted. So just loop. */ |
| 156 | return; |
| 157 | } |
| 158 | |
| 159 | if (alpha_using_srm) |
| 160 | srm_paging_stop(); |
| 161 | |
| 162 | halt(); |
| 163 | } |
| 164 | |
| 165 | static void |
| 166 | common_shutdown(int mode, char *restart_cmd) |
| 167 | { |
| 168 | struct halt_info args; |
| 169 | args.mode = mode; |
| 170 | args.restart_cmd = restart_cmd; |
| 171 | on_each_cpu(common_shutdown_1, &args, 0); |
| 172 | } |
| 173 | |
| 174 | void |
| 175 | machine_restart(char *restart_cmd) |
| 176 | { |
| 177 | common_shutdown(LINUX_REBOOT_CMD_RESTART, restart_cmd); |
| 178 | } |
| 179 | |
| 180 | |
| 181 | void |
| 182 | machine_halt(void) |
| 183 | { |
| 184 | common_shutdown(LINUX_REBOOT_CMD_HALT, NULL); |
| 185 | } |
| 186 | |
| 187 | |
| 188 | void |
| 189 | machine_power_off(void) |
| 190 | { |
| 191 | common_shutdown(LINUX_REBOOT_CMD_POWER_OFF, NULL); |
| 192 | } |
| 193 | |
| 194 | |
| 195 | /* Used by sysrq-p, among others. I don't believe r9-r15 are ever |
| 196 | saved in the context it's used. */ |
| 197 | |
| 198 | void |
| 199 | show_regs(struct pt_regs *regs) |
| 200 | { |
| 201 | show_regs_print_info(KERN_DEFAULT); |
| 202 | dik_show_regs(regs, NULL); |
| 203 | } |
| 204 | |
| 205 | /* |
| 206 | * Re-start a thread when doing execve() |
| 207 | */ |
| 208 | void |
| 209 | start_thread(struct pt_regs * regs, unsigned long pc, unsigned long sp) |
| 210 | { |
| 211 | regs->pc = pc; |
| 212 | regs->ps = 8; |
| 213 | wrusp(sp); |
| 214 | } |
| 215 | EXPORT_SYMBOL(start_thread); |
| 216 | |
| 217 | void |
| 218 | flush_thread(void) |
| 219 | { |
| 220 | /* Arrange for each exec'ed process to start off with a clean slate |
| 221 | with respect to the FPU. This is all exceptions disabled. */ |
| 222 | current_thread_info()->ieee_state = 0; |
| 223 | wrfpcr(FPCR_DYN_NORMAL | ieee_swcr_to_fpcr(0)); |
| 224 | |
| 225 | /* Clean slate for TLS. */ |
| 226 | current_thread_info()->pcb.unique = 0; |
| 227 | } |
| 228 | |
| 229 | void |
| 230 | release_thread(struct task_struct *dead_task) |
| 231 | { |
| 232 | } |
| 233 | |
| 234 | /* |
| 235 | * Copy architecture-specific thread state |
| 236 | */ |
| 237 | int |
| 238 | copy_thread(unsigned long clone_flags, unsigned long usp, |
| 239 | unsigned long kthread_arg, |
| 240 | struct task_struct *p) |
| 241 | { |
| 242 | extern void ret_from_fork(void); |
| 243 | extern void ret_from_kernel_thread(void); |
| 244 | |
| 245 | struct thread_info *childti = task_thread_info(p); |
| 246 | struct pt_regs *childregs = task_pt_regs(p); |
| 247 | struct pt_regs *regs = current_pt_regs(); |
| 248 | struct switch_stack *childstack, *stack; |
| 249 | |
| 250 | childstack = ((struct switch_stack *) childregs) - 1; |
| 251 | childti->pcb.ksp = (unsigned long) childstack; |
| 252 | childti->pcb.flags = 1; /* set FEN, clear everything else */ |
| 253 | |
| 254 | if (unlikely(p->flags & PF_KTHREAD)) { |
| 255 | /* kernel thread */ |
| 256 | memset(childstack, 0, |
| 257 | sizeof(struct switch_stack) + sizeof(struct pt_regs)); |
| 258 | childstack->r26 = (unsigned long) ret_from_kernel_thread; |
| 259 | childstack->r9 = usp; /* function */ |
| 260 | childstack->r10 = kthread_arg; |
| 261 | childregs->hae = alpha_mv.hae_cache, |
| 262 | childti->pcb.usp = 0; |
| 263 | return 0; |
| 264 | } |
| 265 | /* Note: if CLONE_SETTLS is not set, then we must inherit the |
| 266 | value from the parent, which will have been set by the block |
| 267 | copy in dup_task_struct. This is non-intuitive, but is |
| 268 | required for proper operation in the case of a threaded |
| 269 | application calling fork. */ |
| 270 | if (clone_flags & CLONE_SETTLS) |
| 271 | childti->pcb.unique = regs->r20; |
| 272 | else |
| 273 | regs->r20 = 0; /* OSF/1 has some strange fork() semantics. */ |
| 274 | childti->pcb.usp = usp ?: rdusp(); |
| 275 | *childregs = *regs; |
| 276 | childregs->r0 = 0; |
| 277 | childregs->r19 = 0; |
| 278 | childregs->r20 = 1; /* OSF/1 has some strange fork() semantics. */ |
| 279 | stack = ((struct switch_stack *) regs) - 1; |
| 280 | *childstack = *stack; |
| 281 | childstack->r26 = (unsigned long) ret_from_fork; |
| 282 | return 0; |
| 283 | } |
| 284 | |
| 285 | /* |
| 286 | * Fill in the user structure for a ELF core dump. |
| 287 | */ |
| 288 | void |
| 289 | dump_elf_thread(elf_greg_t *dest, struct pt_regs *pt, struct thread_info *ti) |
| 290 | { |
| 291 | /* switch stack follows right below pt_regs: */ |
| 292 | struct switch_stack * sw = ((struct switch_stack *) pt) - 1; |
| 293 | |
| 294 | dest[ 0] = pt->r0; |
| 295 | dest[ 1] = pt->r1; |
| 296 | dest[ 2] = pt->r2; |
| 297 | dest[ 3] = pt->r3; |
| 298 | dest[ 4] = pt->r4; |
| 299 | dest[ 5] = pt->r5; |
| 300 | dest[ 6] = pt->r6; |
| 301 | dest[ 7] = pt->r7; |
| 302 | dest[ 8] = pt->r8; |
| 303 | dest[ 9] = sw->r9; |
| 304 | dest[10] = sw->r10; |
| 305 | dest[11] = sw->r11; |
| 306 | dest[12] = sw->r12; |
| 307 | dest[13] = sw->r13; |
| 308 | dest[14] = sw->r14; |
| 309 | dest[15] = sw->r15; |
| 310 | dest[16] = pt->r16; |
| 311 | dest[17] = pt->r17; |
| 312 | dest[18] = pt->r18; |
| 313 | dest[19] = pt->r19; |
| 314 | dest[20] = pt->r20; |
| 315 | dest[21] = pt->r21; |
| 316 | dest[22] = pt->r22; |
| 317 | dest[23] = pt->r23; |
| 318 | dest[24] = pt->r24; |
| 319 | dest[25] = pt->r25; |
| 320 | dest[26] = pt->r26; |
| 321 | dest[27] = pt->r27; |
| 322 | dest[28] = pt->r28; |
| 323 | dest[29] = pt->gp; |
| 324 | dest[30] = ti == current_thread_info() ? rdusp() : ti->pcb.usp; |
| 325 | dest[31] = pt->pc; |
| 326 | |
| 327 | /* Once upon a time this was the PS value. Which is stupid |
| 328 | since that is always 8 for usermode. Usurped for the more |
| 329 | useful value of the thread's UNIQUE field. */ |
| 330 | dest[32] = ti->pcb.unique; |
| 331 | } |
| 332 | EXPORT_SYMBOL(dump_elf_thread); |
| 333 | |
| 334 | int |
| 335 | dump_elf_task(elf_greg_t *dest, struct task_struct *task) |
| 336 | { |
| 337 | dump_elf_thread(dest, task_pt_regs(task), task_thread_info(task)); |
| 338 | return 1; |
| 339 | } |
| 340 | EXPORT_SYMBOL(dump_elf_task); |
| 341 | |
| 342 | int |
| 343 | dump_elf_task_fp(elf_fpreg_t *dest, struct task_struct *task) |
| 344 | { |
| 345 | struct switch_stack *sw = (struct switch_stack *)task_pt_regs(task) - 1; |
| 346 | memcpy(dest, sw->fp, 32 * 8); |
| 347 | return 1; |
| 348 | } |
| 349 | EXPORT_SYMBOL(dump_elf_task_fp); |
| 350 | |
| 351 | /* |
| 352 | * Return saved PC of a blocked thread. This assumes the frame |
| 353 | * pointer is the 6th saved long on the kernel stack and that the |
| 354 | * saved return address is the first long in the frame. This all |
| 355 | * holds provided the thread blocked through a call to schedule() ($15 |
| 356 | * is the frame pointer in schedule() and $15 is saved at offset 48 by |
| 357 | * entry.S:do_switch_stack). |
| 358 | * |
| 359 | * Under heavy swap load I've seen this lose in an ugly way. So do |
| 360 | * some extra sanity checking on the ranges we expect these pointers |
| 361 | * to be in so that we can fail gracefully. This is just for ps after |
| 362 | * all. -- r~ |
| 363 | */ |
| 364 | |
| 365 | static unsigned long |
| 366 | thread_saved_pc(struct task_struct *t) |
| 367 | { |
| 368 | unsigned long base = (unsigned long)task_stack_page(t); |
| 369 | unsigned long fp, sp = task_thread_info(t)->pcb.ksp; |
| 370 | |
| 371 | if (sp > base && sp+6*8 < base + 16*1024) { |
| 372 | fp = ((unsigned long*)sp)[6]; |
| 373 | if (fp > sp && fp < base + 16*1024) |
| 374 | return *(unsigned long *)fp; |
| 375 | } |
| 376 | |
| 377 | return 0; |
| 378 | } |
| 379 | |
| 380 | unsigned long |
| 381 | get_wchan(struct task_struct *p) |
| 382 | { |
| 383 | unsigned long schedule_frame; |
| 384 | unsigned long pc; |
| 385 | if (!p || p == current || p->state == TASK_RUNNING) |
| 386 | return 0; |
| 387 | /* |
| 388 | * This one depends on the frame size of schedule(). Do a |
| 389 | * "disass schedule" in gdb to find the frame size. Also, the |
| 390 | * code assumes that sleep_on() follows immediately after |
| 391 | * interruptible_sleep_on() and that add_timer() follows |
| 392 | * immediately after interruptible_sleep(). Ugly, isn't it? |
| 393 | * Maybe adding a wchan field to task_struct would be better, |
| 394 | * after all... |
| 395 | */ |
| 396 | |
| 397 | pc = thread_saved_pc(p); |
| 398 | if (in_sched_functions(pc)) { |
| 399 | schedule_frame = ((unsigned long *)task_thread_info(p)->pcb.ksp)[6]; |
| 400 | return ((unsigned long *)schedule_frame)[12]; |
| 401 | } |
| 402 | return pc; |
| 403 | } |