1 | // SPDX-License-Identifier: GPL-2.0 |
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. |
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/resume_user_mode.h> |
27 | #include <asm/current.h> |
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> |
34 | #include <registers.h> |
35 | #include <linux/time-internal.h> |
36 | #include <linux/elfcore.h> |
37 | |
38 | /* |
39 | * This is a per-cpu array. A processor only modifies its entry and it only |
40 | * cares about its entry, so it's OK if another processor is modifying its |
41 | * entry. |
42 | */ |
43 | struct cpu_task cpu_tasks[NR_CPUS] = { [0 ... NR_CPUS - 1] = { -1, NULL } }; |
44 | |
45 | static inline int external_pid(void) |
46 | { |
47 | /* FIXME: Need to look up userspace_pid by cpu */ |
48 | return userspace_pid[0]; |
49 | } |
50 | |
51 | int pid_to_processor_id(int pid) |
52 | { |
53 | int i; |
54 | |
55 | for (i = 0; i < ncpus; i++) { |
56 | if (cpu_tasks[i].pid == pid) |
57 | return i; |
58 | } |
59 | return -1; |
60 | } |
61 | |
62 | void free_stack(unsigned long stack, int order) |
63 | { |
64 | free_pages(addr: stack, order); |
65 | } |
66 | |
67 | unsigned long alloc_stack(int order, int atomic) |
68 | { |
69 | unsigned long page; |
70 | gfp_t flags = GFP_KERNEL; |
71 | |
72 | if (atomic) |
73 | flags = GFP_ATOMIC; |
74 | page = __get_free_pages(gfp_mask: flags, order); |
75 | |
76 | return page; |
77 | } |
78 | |
79 | static inline void set_current(struct task_struct *task) |
80 | { |
81 | cpu_tasks[task_thread_info(task)->cpu] = ((struct cpu_task) |
82 | { external_pid(), task }); |
83 | } |
84 | |
85 | extern void arch_switch_to(struct task_struct *to); |
86 | |
87 | void *__switch_to(struct task_struct *from, struct task_struct *to) |
88 | { |
89 | to->thread.prev_sched = from; |
90 | set_current(to); |
91 | |
92 | switch_threads(&from->thread.switch_buf, &to->thread.switch_buf); |
93 | arch_switch_to(current); |
94 | |
95 | return current->thread.prev_sched; |
96 | } |
97 | |
98 | void interrupt_end(void) |
99 | { |
100 | struct pt_regs *regs = ¤t->thread.regs; |
101 | |
102 | if (need_resched()) |
103 | schedule(); |
104 | if (test_thread_flag(TIF_SIGPENDING) || |
105 | test_thread_flag(TIF_NOTIFY_SIGNAL)) |
106 | do_signal(regs); |
107 | if (test_thread_flag(TIF_NOTIFY_RESUME)) |
108 | resume_user_mode_work(regs); |
109 | } |
110 | |
111 | int get_current_pid(void) |
112 | { |
113 | return task_pid_nr(current); |
114 | } |
115 | |
116 | /* |
117 | * This is called magically, by its address being stuffed in a jmp_buf |
118 | * and being longjmp-d to. |
119 | */ |
120 | void new_thread_handler(void) |
121 | { |
122 | int (*fn)(void *), n; |
123 | void *arg; |
124 | |
125 | if (current->thread.prev_sched != NULL) |
126 | schedule_tail(current->thread.prev_sched); |
127 | current->thread.prev_sched = NULL; |
128 | |
129 | fn = current->thread.request.u.thread.proc; |
130 | arg = current->thread.request.u.thread.arg; |
131 | |
132 | /* |
133 | * callback returns only if the kernel thread execs a process |
134 | */ |
135 | n = fn(arg); |
136 | userspace(¤t->thread.regs.regs, current_thread_info()->aux_fp_regs); |
137 | } |
138 | |
139 | /* Called magically, see new_thread_handler above */ |
140 | void fork_handler(void) |
141 | { |
142 | force_flush_all(); |
143 | |
144 | schedule_tail(current->thread.prev_sched); |
145 | |
146 | /* |
147 | * XXX: if interrupt_end() calls schedule, this call to |
148 | * arch_switch_to isn't needed. We could want to apply this to |
149 | * improve performance. -bb |
150 | */ |
151 | arch_switch_to(current); |
152 | |
153 | current->thread.prev_sched = NULL; |
154 | |
155 | userspace(¤t->thread.regs.regs, current_thread_info()->aux_fp_regs); |
156 | } |
157 | |
158 | int copy_thread(struct task_struct * p, const struct kernel_clone_args *args) |
159 | { |
160 | unsigned long clone_flags = args->flags; |
161 | unsigned long sp = args->stack; |
162 | unsigned long tls = args->tls; |
163 | void (*handler)(void); |
164 | int ret = 0; |
165 | |
166 | p->thread = (struct thread_struct) INIT_THREAD; |
167 | |
168 | if (!args->fn) { |
169 | memcpy(&p->thread.regs.regs, current_pt_regs(), |
170 | sizeof(p->thread.regs.regs)); |
171 | PT_REGS_SET_SYSCALL_RETURN(&p->thread.regs, 0); |
172 | if (sp != 0) |
173 | REGS_SP(p->thread.regs.regs.gp) = sp; |
174 | |
175 | handler = fork_handler; |
176 | |
177 | arch_copy_thread(¤t->thread.arch, &p->thread.arch); |
178 | } else { |
179 | get_safe_registers(p->thread.regs.regs.gp, p->thread.regs.regs.fp); |
180 | p->thread.request.u.thread.proc = args->fn; |
181 | p->thread.request.u.thread.arg = args->fn_arg; |
182 | handler = new_thread_handler; |
183 | } |
184 | |
185 | new_thread(task_stack_page(task: p), &p->thread.switch_buf, handler); |
186 | |
187 | if (!args->fn) { |
188 | clear_flushed_tls(p); |
189 | |
190 | /* |
191 | * Set a new TLS for the child thread? |
192 | */ |
193 | if (clone_flags & CLONE_SETTLS) |
194 | ret = arch_set_tls(p, tls); |
195 | } |
196 | |
197 | return ret; |
198 | } |
199 | |
200 | void initial_thread_cb(void (*proc)(void *), void *arg) |
201 | { |
202 | int save_kmalloc_ok = kmalloc_ok; |
203 | |
204 | kmalloc_ok = 0; |
205 | initial_thread_cb_skas(proc, arg); |
206 | kmalloc_ok = save_kmalloc_ok; |
207 | } |
208 | |
209 | void um_idle_sleep(void) |
210 | { |
211 | if (time_travel_mode != TT_MODE_OFF) |
212 | time_travel_sleep(); |
213 | else |
214 | os_idle_sleep(); |
215 | } |
216 | |
217 | void arch_cpu_idle(void) |
218 | { |
219 | cpu_tasks[current_thread_info()->cpu].pid = os_getpid(); |
220 | um_idle_sleep(); |
221 | } |
222 | |
223 | int __uml_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(s: 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, n: size); |
256 | } |
257 | |
258 | int copy_from_user_proc(void *to, void __user *from, int size) |
259 | { |
260 | return copy_from_user(to, from, n: size); |
261 | } |
262 | |
263 | int clear_user_proc(void __user *buf, int size) |
264 | { |
265 | return clear_user(to: buf, n: size); |
266 | } |
267 | |
268 | static atomic_t using_sysemu = ATOMIC_INIT(0); |
269 | int sysemu_supported; |
270 | |
271 | void set_using_sysemu(int value) |
272 | { |
273 | if (value > sysemu_supported) |
274 | return; |
275 | atomic_set(v: &using_sysemu, i: value); |
276 | } |
277 | |
278 | int get_using_sysemu(void) |
279 | { |
280 | return atomic_read(v: &using_sysemu); |
281 | } |
282 | |
283 | static int sysemu_proc_show(struct seq_file *m, void *v) |
284 | { |
285 | seq_printf(m, fmt: "%d\n" , get_using_sysemu()); |
286 | return 0; |
287 | } |
288 | |
289 | static int sysemu_proc_open(struct inode *inode, struct file *file) |
290 | { |
291 | return single_open(file, sysemu_proc_show, NULL); |
292 | } |
293 | |
294 | static ssize_t sysemu_proc_write(struct file *file, const char __user *buf, |
295 | size_t count, loff_t *pos) |
296 | { |
297 | char tmp[2]; |
298 | |
299 | if (copy_from_user(to: tmp, from: buf, n: 1)) |
300 | return -EFAULT; |
301 | |
302 | if (tmp[0] >= '0' && tmp[0] <= '2') |
303 | set_using_sysemu(tmp[0] - '0'); |
304 | /* We use the first char, but pretend to write everything */ |
305 | return count; |
306 | } |
307 | |
308 | static const struct proc_ops sysemu_proc_ops = { |
309 | .proc_open = sysemu_proc_open, |
310 | .proc_read = seq_read, |
311 | .proc_lseek = seq_lseek, |
312 | .proc_release = single_release, |
313 | .proc_write = sysemu_proc_write, |
314 | }; |
315 | |
316 | int __init make_proc_sysemu(void) |
317 | { |
318 | struct proc_dir_entry *ent; |
319 | if (!sysemu_supported) |
320 | return 0; |
321 | |
322 | ent = proc_create(name: "sysemu" , mode: 0600, NULL, proc_ops: &sysemu_proc_ops); |
323 | |
324 | if (ent == NULL) |
325 | { |
326 | printk(KERN_WARNING "Failed to register /proc/sysemu\n" ); |
327 | return 0; |
328 | } |
329 | |
330 | return 0; |
331 | } |
332 | |
333 | late_initcall(make_proc_sysemu); |
334 | |
335 | int singlestepping(void) |
336 | { |
337 | return test_thread_flag(TIF_SINGLESTEP); |
338 | } |
339 | |
340 | /* |
341 | * Only x86 and x86_64 have an arch_align_stack(). |
342 | * All other arches have "#define arch_align_stack(x) (x)" |
343 | * in their asm/exec.h |
344 | * As this is included in UML from asm-um/system-generic.h, |
345 | * we can use it to behave as the subarch does. |
346 | */ |
347 | #ifndef arch_align_stack |
348 | unsigned long arch_align_stack(unsigned long sp) |
349 | { |
350 | if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space) |
351 | sp -= get_random_u32_below(ceil: 8192); |
352 | return sp & ~0xf; |
353 | } |
354 | #endif |
355 | |
356 | unsigned long __get_wchan(struct task_struct *p) |
357 | { |
358 | unsigned long stack_page, sp, ip; |
359 | bool seen_sched = 0; |
360 | |
361 | stack_page = (unsigned long) task_stack_page(task: p); |
362 | /* Bail if the process has no kernel stack for some reason */ |
363 | if (stack_page == 0) |
364 | return 0; |
365 | |
366 | sp = p->thread.switch_buf->JB_SP; |
367 | /* |
368 | * Bail if the stack pointer is below the bottom of the kernel |
369 | * stack for some reason |
370 | */ |
371 | if (sp < stack_page) |
372 | return 0; |
373 | |
374 | while (sp < stack_page + THREAD_SIZE) { |
375 | ip = *((unsigned long *) sp); |
376 | if (in_sched_functions(addr: ip)) |
377 | /* Ignore everything until we're above the scheduler */ |
378 | seen_sched = 1; |
379 | else if (kernel_text_address(addr: ip) && seen_sched) |
380 | return ip; |
381 | |
382 | sp += sizeof(unsigned long); |
383 | } |
384 | |
385 | return 0; |
386 | } |
387 | |
388 | int elf_core_copy_task_fpregs(struct task_struct *t, elf_fpregset_t *fpu) |
389 | { |
390 | int cpu = current_thread_info()->cpu; |
391 | |
392 | return save_i387_registers(userspace_pid[cpu], (unsigned long *) fpu); |
393 | } |
394 | |
395 | |