| 1 | |
| 2 | |
| 3 | * Copyright (C) 2015 Anton Ivanov (aivanov@{brocade.com,kot-begemot.co.uk})
|
| 4 | * Copyright (C) 2015 Thomas Meyer ([email protected])
|
| 5 | * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
|
| 6 | * Copyright 2003 PathScale, Inc.
|
| 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/cpu.h>
|
| 19 | #include <linux/slab.h>
|
| 20 | #include <linux/sched.h>
|
| 21 | #include <linux/sched/debug.h>
|
| 22 | #include <linux/sched/task.h>
|
| 23 | #include <linux/sched/task_stack.h>
|
| 24 | #include <linux/seq_file.h>
|
| 25 | #include <linux/tick.h>
|
| 26 | #include <linux/threads.h>
|
| 27 | #include <linux/resume_user_mode.h>
|
| 28 | #include <asm/current.h>
|
| 29 | #include <asm/mmu_context.h>
|
| 30 | #include <asm/switch_to.h>
|
| 31 | #include <asm/exec.h>
|
| 32 | #include <linux/uaccess.h>
|
| 33 | #include <as-layout.h>
|
| 34 | #include <kern_util.h>
|
| 35 | #include <os.h>
|
| 36 | #include <skas.h>
|
| 37 | #include <registers.h>
|
| 38 | #include <linux/time-internal.h>
|
| 39 | #include <linux/elfcore.h>
|
| 40 |
|
| 41 | |
| 42 | * This is a per-cpu array. A processor only modifies its entry and it only
|
| 43 | * cares about its entry, so it's OK if another processor is modifying its
|
| 44 | * entry.
|
| 45 |
|
| 46 | struct task_struct *cpu_tasks[NR_CPUS] = {
|
| 47 | [0 ... NR_CPUS - 1] = &init_task,
|
| 48 | };
|
| 49 | EXPORT_SYMBOL(cpu_tasks);
|
| 50 |
|
| 51 | void free_stack(unsigned long stack, int order)
|
| 52 | {
|
| 53 | free_pages(stack, order);
|
| 54 | }
|
| 55 |
|
| 56 | unsigned long alloc_stack(int order, int atomic)
|
| 57 | {
|
| 58 | unsigned long page;
|
| 59 | gfp_t flags = GFP_KERNEL;
|
| 60 |
|
| 61 | if (atomic)
|
| 62 | flags = GFP_ATOMIC;
|
| 63 | page = __get_free_pages(flags, order);
|
| 64 |
|
| 65 | return page;
|
| 66 | }
|
| 67 |
|
| 68 | static inline void set_current(struct task_struct *task)
|
| 69 | {
|
| 70 | cpu_tasks[task_thread_info(task)->cpu] = task;
|
| 71 | }
|
| 72 |
|
| 73 | struct task_struct *__switch_to(struct task_struct *from, struct task_struct *to)
|
| 74 | {
|
| 75 | to->thread.prev_sched = from;
|
| 76 | set_current(to);
|
| 77 |
|
| 78 | switch_threads(&from->thread.switch_buf, &to->thread.switch_buf);
|
| 79 | arch_switch_to(current);
|
| 80 |
|
| 81 | return current->thread.prev_sched;
|
| 82 | }
|
| 83 |
|
| 84 | void interrupt_end(void)
|
| 85 | {
|
| 86 | struct pt_regs *regs = ¤t->thread.regs;
|
| 87 | unsigned long thread_flags;
|
| 88 |
|
| 89 | thread_flags = read_thread_flags();
|
| 90 | while (thread_flags & _TIF_WORK_MASK) {
|
| 91 | if (thread_flags & _TIF_NEED_RESCHED)
|
| 92 | schedule();
|
| 93 | if (thread_flags & (_TIF_SIGPENDING | _TIF_NOTIFY_SIGNAL))
|
| 94 | do_signal(regs);
|
| 95 | if (thread_flags & _TIF_NOTIFY_RESUME)
|
| 96 | resume_user_mode_work(regs);
|
| 97 | thread_flags = read_thread_flags();
|
| 98 | }
|
| 99 | }
|
| 100 |
|
| 101 | int get_current_pid(void)
|
| 102 | {
|
| 103 | return task_pid_nr(current);
|
| 104 | }
|
| 105 |
|
| 106 | |
| 107 | * This is called magically, by its address being stuffed in a jmp_buf
|
| 108 | * and being longjmp-d to.
|
| 109 |
|
| 110 | void new_thread_handler(void)
|
| 111 | {
|
| 112 | int (*fn)(void *);
|
| 113 | void *arg;
|
| 114 |
|
| 115 | if (current->thread.prev_sched != NULL)
|
| 116 | schedule_tail(current->thread.prev_sched);
|
| 117 | current->thread.prev_sched = NULL;
|
| 118 |
|
| 119 | fn = current->thread.request.thread.proc;
|
| 120 | arg = current->thread.request.thread.arg;
|
| 121 |
|
| 122 | |
| 123 | * callback returns only if the kernel thread execs a process
|
| 124 |
|
| 125 | fn(arg);
|
| 126 | userspace(¤t->thread.regs.regs);
|
| 127 | }
|
| 128 |
|
| 129 |
|
| 130 | static void fork_handler(void)
|
| 131 | {
|
| 132 | schedule_tail(current->thread.prev_sched);
|
| 133 |
|
| 134 | |
| 135 | * XXX: if interrupt_end() calls schedule, this call to
|
| 136 | * arch_switch_to isn't needed. We could want to apply this to
|
| 137 | * improve performance. -bb
|
| 138 |
|
| 139 | arch_switch_to(current);
|
| 140 |
|
| 141 | current->thread.prev_sched = NULL;
|
| 142 |
|
| 143 | userspace(¤t->thread.regs.regs);
|
| 144 | }
|
| 145 |
|
| 146 | int copy_thread(struct task_struct * p, const struct kernel_clone_args *args)
|
| 147 | {
|
| 148 | u64 clone_flags = args->flags;
|
| 149 | unsigned long sp = args->stack;
|
| 150 | unsigned long tls = args->tls;
|
| 151 | void (*handler)(void);
|
| 152 | int ret = 0;
|
| 153 |
|
| 154 | p->thread = (struct thread_struct) INIT_THREAD;
|
| 155 |
|
| 156 | if (!args->fn) {
|
| 157 | memcpy(&p->thread.regs.regs, current_pt_regs(),
|
| 158 | sizeof(p->thread.regs.regs));
|
| 159 | PT_REGS_SET_SYSCALL_RETURN(&p->thread.regs, 0);
|
| 160 | if (sp != 0)
|
| 161 | REGS_SP(p->thread.regs.regs.gp) = sp;
|
| 162 |
|
| 163 | handler = fork_handler;
|
| 164 |
|
| 165 | arch_copy_thread(¤t->thread.arch, &p->thread.arch);
|
| 166 | } else {
|
| 167 | get_safe_registers(p->thread.regs.regs.gp, p->thread.regs.regs.fp);
|
| 168 | p->thread.request.thread.proc = args->fn;
|
| 169 | p->thread.request.thread.arg = args->fn_arg;
|
| 170 | handler = new_thread_handler;
|
| 171 | }
|
| 172 |
|
| 173 | new_thread(task_stack_page(p), &p->thread.switch_buf, handler);
|
| 174 |
|
| 175 | if (!args->fn) {
|
| 176 | clear_flushed_tls(p);
|
| 177 |
|
| 178 | |
| 179 | * Set a new TLS for the child thread?
|
| 180 |
|
| 181 | if (clone_flags & CLONE_SETTLS)
|
| 182 | ret = arch_set_tls(p, tls);
|
| 183 | }
|
| 184 |
|
| 185 | return ret;
|
| 186 | }
|
| 187 |
|
| 188 | void initial_thread_cb(void (*proc)(void *), void *arg)
|
| 189 | {
|
| 190 | initial_thread_cb_skas(proc, arg);
|
| 191 | }
|
| 192 |
|
| 193 | int arch_dup_task_struct(struct task_struct *dst,
|
| 194 | struct task_struct *src)
|
| 195 | {
|
| 196 |
|
| 197 | if (unlikely(src == &init_task)) {
|
| 198 | memcpy(dst, src, sizeof(init_task));
|
| 199 | memset((void *)dst + sizeof(init_task), 0,
|
| 200 | arch_task_struct_size - sizeof(init_task));
|
| 201 | } else {
|
| 202 | memcpy(dst, src, arch_task_struct_size);
|
| 203 | }
|
| 204 |
|
| 205 | return 0;
|
| 206 | }
|
| 207 |
|
| 208 | void um_idle_sleep(void)
|
| 209 | {
|
| 210 | if (time_travel_mode != TT_MODE_OFF)
|
| 211 | time_travel_sleep();
|
| 212 | else
|
| 213 | os_idle_sleep();
|
| 214 | }
|
| 215 |
|
| 216 | void arch_cpu_idle(void)
|
| 217 | {
|
| 218 | um_idle_sleep();
|
| 219 | }
|
| 220 |
|
| 221 | void arch_cpu_idle_prepare(void)
|
| 222 | {
|
| 223 | os_idle_prepare();
|
| 224 | }
|
| 225 |
|
| 226 | int __uml_cant_sleep(void) {
|
| 227 | return in_atomic() || irqs_disabled() || in_interrupt();
|
| 228 |
|
| 229 | }
|
| 230 |
|
| 231 | int uml_need_resched(void)
|
| 232 | {
|
| 233 | return need_resched();
|
| 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_from_user_proc(void *to, void __user *from, int size)
|
| 254 | {
|
| 255 | return copy_from_user(to, from, size);
|
| 256 | }
|
| 257 |
|
| 258 | int singlestepping(void)
|
| 259 | {
|
| 260 | return test_thread_flag(TIF_SINGLESTEP);
|
| 261 | }
|
| 262 |
|
| 263 | |
| 264 | * Only x86 and x86_64 have an arch_align_stack().
|
| 265 | * All other arches have "#define arch_align_stack(x) (x)"
|
| 266 | * in their asm/exec.h
|
| 267 | * As this is included in UML from asm-um/system-generic.h,
|
| 268 | * we can use it to behave as the subarch does.
|
| 269 |
|
| 270 | #ifndef arch_align_stack
|
| 271 | unsigned long arch_align_stack(unsigned long sp)
|
| 272 | {
|
| 273 | if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
|
| 274 | sp -= get_random_u32_below(8192);
|
| 275 | return sp & ~0xf;
|
| 276 | }
|
| 277 | #endif
|
| 278 |
|
| 279 | unsigned long __get_wchan(struct task_struct *p)
|
| 280 | {
|
| 281 | unsigned long stack_page, sp, ip;
|
| 282 | bool seen_sched = 0;
|
| 283 |
|
| 284 | stack_page = (unsigned long) task_stack_page(p);
|
| 285 |
|
| 286 | if (stack_page == 0)
|
| 287 | return 0;
|
| 288 |
|
| 289 | sp = p->thread.switch_buf->JB_SP;
|
| 290 | |
| 291 | * Bail if the stack pointer is below the bottom of the kernel
|
| 292 | * stack for some reason
|
| 293 |
|
| 294 | if (sp < stack_page)
|
| 295 | return 0;
|
| 296 |
|
| 297 | while (sp < stack_page + THREAD_SIZE) {
|
| 298 | ip = *((unsigned long *) sp);
|
| 299 | if (in_sched_functions(ip))
|
| 300 |
|
| 301 | seen_sched = 1;
|
| 302 | else if (kernel_text_address(ip) && seen_sched)
|
| 303 | return ip;
|
| 304 |
|
| 305 | sp += sizeof(unsigned long);
|
| 306 | }
|
| 307 |
|
| 308 | return 0;
|
| 309 | }
|