David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2 | /* |
| 3 | * Copyright (C) 2015 Anton Ivanov (aivanov@{brocade.com,kot-begemot.co.uk}) |
| 4 | * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de) |
| 5 | * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com) |
| 6 | * Copyright 2003 PathScale, Inc. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 7 | */ |
| 8 | |
| 9 | #include <linux/stddef.h> |
| 10 | #include <linux/err.h> |
| 11 | #include <linux/hardirq.h> |
| 12 | #include <linux/mm.h> |
| 13 | #include <linux/module.h> |
| 14 | #include <linux/personality.h> |
| 15 | #include <linux/proc_fs.h> |
| 16 | #include <linux/ptrace.h> |
| 17 | #include <linux/random.h> |
| 18 | #include <linux/slab.h> |
| 19 | #include <linux/sched.h> |
| 20 | #include <linux/sched/debug.h> |
| 21 | #include <linux/sched/task.h> |
| 22 | #include <linux/sched/task_stack.h> |
| 23 | #include <linux/seq_file.h> |
| 24 | #include <linux/tick.h> |
| 25 | #include <linux/threads.h> |
| 26 | #include <linux/tracehook.h> |
| 27 | #include <asm/current.h> |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 28 | #include <asm/mmu_context.h> |
| 29 | #include <linux/uaccess.h> |
| 30 | #include <as-layout.h> |
| 31 | #include <kern_util.h> |
| 32 | #include <os.h> |
| 33 | #include <skas.h> |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame^] | 34 | #include <linux/time-internal.h> |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 35 | |
| 36 | /* |
| 37 | * This is a per-cpu array. A processor only modifies its entry and it only |
| 38 | * cares about its entry, so it's OK if another processor is modifying its |
| 39 | * entry. |
| 40 | */ |
| 41 | struct cpu_task cpu_tasks[NR_CPUS] = { [0 ... NR_CPUS - 1] = { -1, NULL } }; |
| 42 | |
| 43 | static inline int external_pid(void) |
| 44 | { |
| 45 | /* FIXME: Need to look up userspace_pid by cpu */ |
| 46 | return userspace_pid[0]; |
| 47 | } |
| 48 | |
| 49 | int pid_to_processor_id(int pid) |
| 50 | { |
| 51 | int i; |
| 52 | |
| 53 | for (i = 0; i < ncpus; i++) { |
| 54 | if (cpu_tasks[i].pid == pid) |
| 55 | return i; |
| 56 | } |
| 57 | return -1; |
| 58 | } |
| 59 | |
| 60 | void free_stack(unsigned long stack, int order) |
| 61 | { |
| 62 | free_pages(stack, order); |
| 63 | } |
| 64 | |
| 65 | unsigned long alloc_stack(int order, int atomic) |
| 66 | { |
| 67 | unsigned long page; |
| 68 | gfp_t flags = GFP_KERNEL; |
| 69 | |
| 70 | if (atomic) |
| 71 | flags = GFP_ATOMIC; |
| 72 | page = __get_free_pages(flags, order); |
| 73 | |
| 74 | return page; |
| 75 | } |
| 76 | |
| 77 | static inline void set_current(struct task_struct *task) |
| 78 | { |
| 79 | cpu_tasks[task_thread_info(task)->cpu] = ((struct cpu_task) |
| 80 | { external_pid(), task }); |
| 81 | } |
| 82 | |
| 83 | extern void arch_switch_to(struct task_struct *to); |
| 84 | |
| 85 | void *__switch_to(struct task_struct *from, struct task_struct *to) |
| 86 | { |
| 87 | to->thread.prev_sched = from; |
| 88 | set_current(to); |
| 89 | |
| 90 | switch_threads(&from->thread.switch_buf, &to->thread.switch_buf); |
| 91 | arch_switch_to(current); |
| 92 | |
| 93 | return current->thread.prev_sched; |
| 94 | } |
| 95 | |
| 96 | void interrupt_end(void) |
| 97 | { |
| 98 | struct pt_regs *regs = ¤t->thread.regs; |
| 99 | |
| 100 | if (need_resched()) |
| 101 | schedule(); |
| 102 | if (test_thread_flag(TIF_SIGPENDING)) |
| 103 | do_signal(regs); |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame^] | 104 | if (test_thread_flag(TIF_NOTIFY_RESUME)) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 105 | tracehook_notify_resume(regs); |
| 106 | } |
| 107 | |
| 108 | int get_current_pid(void) |
| 109 | { |
| 110 | return task_pid_nr(current); |
| 111 | } |
| 112 | |
| 113 | /* |
| 114 | * This is called magically, by its address being stuffed in a jmp_buf |
| 115 | * and being longjmp-d to. |
| 116 | */ |
| 117 | void new_thread_handler(void) |
| 118 | { |
| 119 | int (*fn)(void *), n; |
| 120 | void *arg; |
| 121 | |
| 122 | if (current->thread.prev_sched != NULL) |
| 123 | schedule_tail(current->thread.prev_sched); |
| 124 | current->thread.prev_sched = NULL; |
| 125 | |
| 126 | fn = current->thread.request.u.thread.proc; |
| 127 | arg = current->thread.request.u.thread.arg; |
| 128 | |
| 129 | /* |
| 130 | * callback returns only if the kernel thread execs a process |
| 131 | */ |
| 132 | n = fn(arg); |
| 133 | userspace(¤t->thread.regs.regs, current_thread_info()->aux_fp_regs); |
| 134 | } |
| 135 | |
| 136 | /* Called magically, see new_thread_handler above */ |
| 137 | void fork_handler(void) |
| 138 | { |
| 139 | force_flush_all(); |
| 140 | |
| 141 | schedule_tail(current->thread.prev_sched); |
| 142 | |
| 143 | /* |
| 144 | * XXX: if interrupt_end() calls schedule, this call to |
| 145 | * arch_switch_to isn't needed. We could want to apply this to |
| 146 | * improve performance. -bb |
| 147 | */ |
| 148 | arch_switch_to(current); |
| 149 | |
| 150 | current->thread.prev_sched = NULL; |
| 151 | |
| 152 | userspace(¤t->thread.regs.regs, current_thread_info()->aux_fp_regs); |
| 153 | } |
| 154 | |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame^] | 155 | int copy_thread(unsigned long clone_flags, unsigned long sp, |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 156 | unsigned long arg, struct task_struct * p, unsigned long tls) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 157 | { |
| 158 | void (*handler)(void); |
| 159 | int kthread = current->flags & PF_KTHREAD; |
| 160 | int ret = 0; |
| 161 | |
| 162 | p->thread = (struct thread_struct) INIT_THREAD; |
| 163 | |
| 164 | if (!kthread) { |
| 165 | memcpy(&p->thread.regs.regs, current_pt_regs(), |
| 166 | sizeof(p->thread.regs.regs)); |
| 167 | PT_REGS_SET_SYSCALL_RETURN(&p->thread.regs, 0); |
| 168 | if (sp != 0) |
| 169 | REGS_SP(p->thread.regs.regs.gp) = sp; |
| 170 | |
| 171 | handler = fork_handler; |
| 172 | |
| 173 | arch_copy_thread(¤t->thread.arch, &p->thread.arch); |
| 174 | } else { |
| 175 | get_safe_registers(p->thread.regs.regs.gp, p->thread.regs.regs.fp); |
| 176 | p->thread.request.u.thread.proc = (int (*)(void *))sp; |
| 177 | p->thread.request.u.thread.arg = (void *)arg; |
| 178 | handler = new_thread_handler; |
| 179 | } |
| 180 | |
| 181 | new_thread(task_stack_page(p), &p->thread.switch_buf, handler); |
| 182 | |
| 183 | if (!kthread) { |
| 184 | clear_flushed_tls(p); |
| 185 | |
| 186 | /* |
| 187 | * Set a new TLS for the child thread? |
| 188 | */ |
| 189 | if (clone_flags & CLONE_SETTLS) |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 190 | ret = arch_set_tls(p, tls); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 191 | } |
| 192 | |
| 193 | return ret; |
| 194 | } |
| 195 | |
| 196 | void initial_thread_cb(void (*proc)(void *), void *arg) |
| 197 | { |
| 198 | int save_kmalloc_ok = kmalloc_ok; |
| 199 | |
| 200 | kmalloc_ok = 0; |
| 201 | initial_thread_cb_skas(proc, arg); |
| 202 | kmalloc_ok = save_kmalloc_ok; |
| 203 | } |
| 204 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 205 | static void um_idle_sleep(void) |
| 206 | { |
| 207 | unsigned long long duration = UM_NSEC_PER_SEC; |
| 208 | |
| 209 | if (time_travel_mode != TT_MODE_OFF) { |
| 210 | time_travel_sleep(duration); |
| 211 | } else { |
| 212 | os_idle_sleep(duration); |
| 213 | } |
| 214 | } |
| 215 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 216 | void arch_cpu_idle(void) |
| 217 | { |
| 218 | cpu_tasks[current_thread_info()->cpu].pid = os_getpid(); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 219 | um_idle_sleep(); |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame^] | 220 | raw_local_irq_enable(); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 221 | } |
| 222 | |
| 223 | int __cant_sleep(void) { |
| 224 | return in_atomic() || irqs_disabled() || in_interrupt(); |
| 225 | /* Is in_interrupt() really needed? */ |
| 226 | } |
| 227 | |
| 228 | int user_context(unsigned long sp) |
| 229 | { |
| 230 | unsigned long stack; |
| 231 | |
| 232 | stack = sp & (PAGE_MASK << CONFIG_KERNEL_STACK_ORDER); |
| 233 | return stack != (unsigned long) current_thread_info(); |
| 234 | } |
| 235 | |
| 236 | extern exitcall_t __uml_exitcall_begin, __uml_exitcall_end; |
| 237 | |
| 238 | void do_uml_exitcalls(void) |
| 239 | { |
| 240 | exitcall_t *call; |
| 241 | |
| 242 | call = &__uml_exitcall_end; |
| 243 | while (--call >= &__uml_exitcall_begin) |
| 244 | (*call)(); |
| 245 | } |
| 246 | |
| 247 | char *uml_strdup(const char *string) |
| 248 | { |
| 249 | return kstrdup(string, GFP_KERNEL); |
| 250 | } |
| 251 | EXPORT_SYMBOL(uml_strdup); |
| 252 | |
| 253 | int copy_to_user_proc(void __user *to, void *from, int size) |
| 254 | { |
| 255 | return copy_to_user(to, from, size); |
| 256 | } |
| 257 | |
| 258 | int copy_from_user_proc(void *to, void __user *from, int size) |
| 259 | { |
| 260 | return copy_from_user(to, from, size); |
| 261 | } |
| 262 | |
| 263 | int clear_user_proc(void __user *buf, int size) |
| 264 | { |
| 265 | return clear_user(buf, size); |
| 266 | } |
| 267 | |
| 268 | int cpu(void) |
| 269 | { |
| 270 | return current_thread_info()->cpu; |
| 271 | } |
| 272 | |
| 273 | static atomic_t using_sysemu = ATOMIC_INIT(0); |
| 274 | int sysemu_supported; |
| 275 | |
| 276 | void set_using_sysemu(int value) |
| 277 | { |
| 278 | if (value > sysemu_supported) |
| 279 | return; |
| 280 | atomic_set(&using_sysemu, value); |
| 281 | } |
| 282 | |
| 283 | int get_using_sysemu(void) |
| 284 | { |
| 285 | return atomic_read(&using_sysemu); |
| 286 | } |
| 287 | |
| 288 | static int sysemu_proc_show(struct seq_file *m, void *v) |
| 289 | { |
| 290 | seq_printf(m, "%d\n", get_using_sysemu()); |
| 291 | return 0; |
| 292 | } |
| 293 | |
| 294 | static int sysemu_proc_open(struct inode *inode, struct file *file) |
| 295 | { |
| 296 | return single_open(file, sysemu_proc_show, NULL); |
| 297 | } |
| 298 | |
| 299 | static ssize_t sysemu_proc_write(struct file *file, const char __user *buf, |
| 300 | size_t count, loff_t *pos) |
| 301 | { |
| 302 | char tmp[2]; |
| 303 | |
| 304 | if (copy_from_user(tmp, buf, 1)) |
| 305 | return -EFAULT; |
| 306 | |
| 307 | if (tmp[0] >= '0' && tmp[0] <= '2') |
| 308 | set_using_sysemu(tmp[0] - '0'); |
| 309 | /* We use the first char, but pretend to write everything */ |
| 310 | return count; |
| 311 | } |
| 312 | |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame^] | 313 | static const struct proc_ops sysemu_proc_ops = { |
| 314 | .proc_open = sysemu_proc_open, |
| 315 | .proc_read = seq_read, |
| 316 | .proc_lseek = seq_lseek, |
| 317 | .proc_release = single_release, |
| 318 | .proc_write = sysemu_proc_write, |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 319 | }; |
| 320 | |
| 321 | int __init make_proc_sysemu(void) |
| 322 | { |
| 323 | struct proc_dir_entry *ent; |
| 324 | if (!sysemu_supported) |
| 325 | return 0; |
| 326 | |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame^] | 327 | ent = proc_create("sysemu", 0600, NULL, &sysemu_proc_ops); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 328 | |
| 329 | if (ent == NULL) |
| 330 | { |
| 331 | printk(KERN_WARNING "Failed to register /proc/sysemu\n"); |
| 332 | return 0; |
| 333 | } |
| 334 | |
| 335 | return 0; |
| 336 | } |
| 337 | |
| 338 | late_initcall(make_proc_sysemu); |
| 339 | |
| 340 | int singlestepping(void * t) |
| 341 | { |
| 342 | struct task_struct *task = t ? t : current; |
| 343 | |
| 344 | if (!(task->ptrace & PT_DTRACE)) |
| 345 | return 0; |
| 346 | |
| 347 | if (task->thread.singlestep_syscall) |
| 348 | return 1; |
| 349 | |
| 350 | return 2; |
| 351 | } |
| 352 | |
| 353 | /* |
| 354 | * Only x86 and x86_64 have an arch_align_stack(). |
| 355 | * All other arches have "#define arch_align_stack(x) (x)" |
| 356 | * in their asm/exec.h |
| 357 | * As this is included in UML from asm-um/system-generic.h, |
| 358 | * we can use it to behave as the subarch does. |
| 359 | */ |
| 360 | #ifndef arch_align_stack |
| 361 | unsigned long arch_align_stack(unsigned long sp) |
| 362 | { |
| 363 | if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space) |
| 364 | sp -= get_random_int() % 8192; |
| 365 | return sp & ~0xf; |
| 366 | } |
| 367 | #endif |
| 368 | |
| 369 | unsigned long get_wchan(struct task_struct *p) |
| 370 | { |
| 371 | unsigned long stack_page, sp, ip; |
| 372 | bool seen_sched = 0; |
| 373 | |
| 374 | if ((p == NULL) || (p == current) || (p->state == TASK_RUNNING)) |
| 375 | return 0; |
| 376 | |
| 377 | stack_page = (unsigned long) task_stack_page(p); |
| 378 | /* Bail if the process has no kernel stack for some reason */ |
| 379 | if (stack_page == 0) |
| 380 | return 0; |
| 381 | |
| 382 | sp = p->thread.switch_buf->JB_SP; |
| 383 | /* |
| 384 | * Bail if the stack pointer is below the bottom of the kernel |
| 385 | * stack for some reason |
| 386 | */ |
| 387 | if (sp < stack_page) |
| 388 | return 0; |
| 389 | |
| 390 | while (sp < stack_page + THREAD_SIZE) { |
| 391 | ip = *((unsigned long *) sp); |
| 392 | if (in_sched_functions(ip)) |
| 393 | /* Ignore everything until we're above the scheduler */ |
| 394 | seen_sched = 1; |
| 395 | else if (kernel_text_address(ip) && seen_sched) |
| 396 | return ip; |
| 397 | |
| 398 | sp += sizeof(unsigned long); |
| 399 | } |
| 400 | |
| 401 | return 0; |
| 402 | } |
| 403 | |
| 404 | int elf_core_copy_fpregs(struct task_struct *t, elf_fpregset_t *fpu) |
| 405 | { |
| 406 | int cpu = current_thread_info()->cpu; |
| 407 | |
| 408 | return save_i387_registers(userspace_pid[cpu], (unsigned long *) fpu); |
| 409 | } |
| 410 | |