xref: /linux-6.15/arch/x86/kernel/process_64.c (revision bcefe12e)
1 /*
2  *  Copyright (C) 1995  Linus Torvalds
3  *
4  *  Pentium III FXSR, SSE support
5  *	Gareth Hughes <[email protected]>, May 2000
6  *
7  *  X86-64 port
8  *	Andi Kleen.
9  *
10  *	CPU hotplug support - [email protected]
11  */
12 
13 /*
14  * This file handles the architecture-dependent parts of process handling..
15  */
16 
17 #include <linux/stackprotector.h>
18 #include <linux/cpu.h>
19 #include <linux/errno.h>
20 #include <linux/sched.h>
21 #include <linux/fs.h>
22 #include <linux/kernel.h>
23 #include <linux/mm.h>
24 #include <linux/elfcore.h>
25 #include <linux/smp.h>
26 #include <linux/slab.h>
27 #include <linux/user.h>
28 #include <linux/interrupt.h>
29 #include <linux/utsname.h>
30 #include <linux/delay.h>
31 #include <linux/module.h>
32 #include <linux/ptrace.h>
33 #include <linux/notifier.h>
34 #include <linux/kprobes.h>
35 #include <linux/kdebug.h>
36 #include <linux/tick.h>
37 #include <linux/prctl.h>
38 #include <linux/uaccess.h>
39 #include <linux/io.h>
40 #include <linux/ftrace.h>
41 #include <linux/dmi.h>
42 
43 #include <asm/pgtable.h>
44 #include <asm/system.h>
45 #include <asm/processor.h>
46 #include <asm/i387.h>
47 #include <asm/mmu_context.h>
48 #include <asm/prctl.h>
49 #include <asm/desc.h>
50 #include <asm/proto.h>
51 #include <asm/ia32.h>
52 #include <asm/idle.h>
53 #include <asm/syscalls.h>
54 #include <asm/ds.h>
55 #include <asm/debugreg.h>
56 
57 asmlinkage extern void ret_from_fork(void);
58 
59 DEFINE_PER_CPU(unsigned long, old_rsp);
60 static DEFINE_PER_CPU(unsigned char, is_idle);
61 
62 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
63 
64 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
65 
66 void idle_notifier_register(struct notifier_block *n)
67 {
68 	atomic_notifier_chain_register(&idle_notifier, n);
69 }
70 EXPORT_SYMBOL_GPL(idle_notifier_register);
71 
72 void idle_notifier_unregister(struct notifier_block *n)
73 {
74 	atomic_notifier_chain_unregister(&idle_notifier, n);
75 }
76 EXPORT_SYMBOL_GPL(idle_notifier_unregister);
77 
78 void enter_idle(void)
79 {
80 	percpu_write(is_idle, 1);
81 	atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
82 }
83 
84 static void __exit_idle(void)
85 {
86 	if (x86_test_and_clear_bit_percpu(0, is_idle) == 0)
87 		return;
88 	atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
89 }
90 
91 /* Called from interrupts to signify idle end */
92 void exit_idle(void)
93 {
94 	/* idle loop has pid 0 */
95 	if (current->pid)
96 		return;
97 	__exit_idle();
98 }
99 
100 #ifndef CONFIG_SMP
101 static inline void play_dead(void)
102 {
103 	BUG();
104 }
105 #endif
106 
107 /*
108  * The idle thread. There's no useful work to be
109  * done, so just try to conserve power and have a
110  * low exit latency (ie sit in a loop waiting for
111  * somebody to say that they'd like to reschedule)
112  */
113 void cpu_idle(void)
114 {
115 	current_thread_info()->status |= TS_POLLING;
116 
117 	/*
118 	 * If we're the non-boot CPU, nothing set the stack canary up
119 	 * for us.  CPU0 already has it initialized but no harm in
120 	 * doing it again.  This is a good place for updating it, as
121 	 * we wont ever return from this function (so the invalid
122 	 * canaries already on the stack wont ever trigger).
123 	 */
124 	boot_init_stack_canary();
125 
126 	/* endless idle loop with no priority at all */
127 	while (1) {
128 		tick_nohz_stop_sched_tick(1);
129 		while (!need_resched()) {
130 
131 			rmb();
132 
133 			if (cpu_is_offline(smp_processor_id()))
134 				play_dead();
135 			/*
136 			 * Idle routines should keep interrupts disabled
137 			 * from here on, until they go to idle.
138 			 * Otherwise, idle callbacks can misfire.
139 			 */
140 			local_irq_disable();
141 			enter_idle();
142 			/* Don't trace irqs off for idle */
143 			stop_critical_timings();
144 			pm_idle();
145 			start_critical_timings();
146 			/* In many cases the interrupt that ended idle
147 			   has already called exit_idle. But some idle
148 			   loops can be woken up without interrupt. */
149 			__exit_idle();
150 		}
151 
152 		tick_nohz_restart_sched_tick();
153 		preempt_enable_no_resched();
154 		schedule();
155 		preempt_disable();
156 	}
157 }
158 
159 /* Prints also some state that isn't saved in the pt_regs */
160 void __show_regs(struct pt_regs *regs, int all)
161 {
162 	unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
163 	unsigned long d0, d1, d2, d3, d6, d7;
164 	unsigned int fsindex, gsindex;
165 	unsigned int ds, cs, es;
166 	const char *board;
167 
168 	printk("\n");
169 	print_modules();
170 	board = dmi_get_system_info(DMI_PRODUCT_NAME);
171 	if (!board)
172 		board = "";
173 	printk(KERN_INFO "Pid: %d, comm: %.20s %s %s %.*s %s\n",
174 		current->pid, current->comm, print_tainted(),
175 		init_utsname()->release,
176 		(int)strcspn(init_utsname()->version, " "),
177 		init_utsname()->version, board);
178 	printk(KERN_INFO "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
179 	printk_address(regs->ip, 1);
180 	printk(KERN_INFO "RSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss,
181 			regs->sp, regs->flags);
182 	printk(KERN_INFO "RAX: %016lx RBX: %016lx RCX: %016lx\n",
183 	       regs->ax, regs->bx, regs->cx);
184 	printk(KERN_INFO "RDX: %016lx RSI: %016lx RDI: %016lx\n",
185 	       regs->dx, regs->si, regs->di);
186 	printk(KERN_INFO "RBP: %016lx R08: %016lx R09: %016lx\n",
187 	       regs->bp, regs->r8, regs->r9);
188 	printk(KERN_INFO "R10: %016lx R11: %016lx R12: %016lx\n",
189 	       regs->r10, regs->r11, regs->r12);
190 	printk(KERN_INFO "R13: %016lx R14: %016lx R15: %016lx\n",
191 	       regs->r13, regs->r14, regs->r15);
192 
193 	asm("movl %%ds,%0" : "=r" (ds));
194 	asm("movl %%cs,%0" : "=r" (cs));
195 	asm("movl %%es,%0" : "=r" (es));
196 	asm("movl %%fs,%0" : "=r" (fsindex));
197 	asm("movl %%gs,%0" : "=r" (gsindex));
198 
199 	rdmsrl(MSR_FS_BASE, fs);
200 	rdmsrl(MSR_GS_BASE, gs);
201 	rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
202 
203 	if (!all)
204 		return;
205 
206 	cr0 = read_cr0();
207 	cr2 = read_cr2();
208 	cr3 = read_cr3();
209 	cr4 = read_cr4();
210 
211 	printk(KERN_INFO "FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
212 	       fs, fsindex, gs, gsindex, shadowgs);
213 	printk(KERN_INFO "CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
214 			es, cr0);
215 	printk(KERN_INFO "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
216 			cr4);
217 
218 	get_debugreg(d0, 0);
219 	get_debugreg(d1, 1);
220 	get_debugreg(d2, 2);
221 	printk(KERN_INFO "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
222 	get_debugreg(d3, 3);
223 	get_debugreg(d6, 6);
224 	get_debugreg(d7, 7);
225 	printk(KERN_INFO "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
226 }
227 
228 void show_regs(struct pt_regs *regs)
229 {
230 	printk(KERN_INFO "CPU %d:", smp_processor_id());
231 	__show_regs(regs, 1);
232 	show_trace(NULL, regs, (void *)(regs + 1), regs->bp);
233 }
234 
235 void release_thread(struct task_struct *dead_task)
236 {
237 	if (dead_task->mm) {
238 		if (dead_task->mm->context.size) {
239 			printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
240 					dead_task->comm,
241 					dead_task->mm->context.ldt,
242 					dead_task->mm->context.size);
243 			BUG();
244 		}
245 	}
246 }
247 
248 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
249 {
250 	struct user_desc ud = {
251 		.base_addr = addr,
252 		.limit = 0xfffff,
253 		.seg_32bit = 1,
254 		.limit_in_pages = 1,
255 		.useable = 1,
256 	};
257 	struct desc_struct *desc = t->thread.tls_array;
258 	desc += tls;
259 	fill_ldt(desc, &ud);
260 }
261 
262 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
263 {
264 	return get_desc_base(&t->thread.tls_array[tls]);
265 }
266 
267 /*
268  * This gets called before we allocate a new thread and copy
269  * the current task into it.
270  */
271 void prepare_to_copy(struct task_struct *tsk)
272 {
273 	unlazy_fpu(tsk);
274 }
275 
276 int copy_thread(unsigned long clone_flags, unsigned long sp,
277 		unsigned long unused,
278 	struct task_struct *p, struct pt_regs *regs)
279 {
280 	int err;
281 	struct pt_regs *childregs;
282 	struct task_struct *me = current;
283 
284 	childregs = ((struct pt_regs *)
285 			(THREAD_SIZE + task_stack_page(p))) - 1;
286 	*childregs = *regs;
287 
288 	childregs->ax = 0;
289 	childregs->sp = sp;
290 	if (sp == ~0UL)
291 		childregs->sp = (unsigned long)childregs;
292 
293 	p->thread.sp = (unsigned long) childregs;
294 	p->thread.sp0 = (unsigned long) (childregs+1);
295 	p->thread.usersp = me->thread.usersp;
296 
297 	set_tsk_thread_flag(p, TIF_FORK);
298 
299 	p->thread.fs = me->thread.fs;
300 	p->thread.gs = me->thread.gs;
301 	p->thread.io_bitmap_ptr = NULL;
302 
303 	savesegment(gs, p->thread.gsindex);
304 	savesegment(fs, p->thread.fsindex);
305 	savesegment(es, p->thread.es);
306 	savesegment(ds, p->thread.ds);
307 
308 	err = -ENOMEM;
309 	memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
310 
311 	if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
312 		p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
313 		if (!p->thread.io_bitmap_ptr) {
314 			p->thread.io_bitmap_max = 0;
315 			return -ENOMEM;
316 		}
317 		memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
318 				IO_BITMAP_BYTES);
319 		set_tsk_thread_flag(p, TIF_IO_BITMAP);
320 	}
321 
322 	/*
323 	 * Set a new TLS for the child thread?
324 	 */
325 	if (clone_flags & CLONE_SETTLS) {
326 #ifdef CONFIG_IA32_EMULATION
327 		if (test_thread_flag(TIF_IA32))
328 			err = do_set_thread_area(p, -1,
329 				(struct user_desc __user *)childregs->si, 0);
330 		else
331 #endif
332 			err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
333 		if (err)
334 			goto out;
335 	}
336 
337 	clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
338 	p->thread.ds_ctx = NULL;
339 
340 	clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
341 	p->thread.debugctlmsr = 0;
342 
343 	err = 0;
344 out:
345 	if (err && p->thread.io_bitmap_ptr) {
346 		kfree(p->thread.io_bitmap_ptr);
347 		p->thread.io_bitmap_max = 0;
348 	}
349 
350 	return err;
351 }
352 
353 void
354 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
355 {
356 	loadsegment(fs, 0);
357 	loadsegment(es, 0);
358 	loadsegment(ds, 0);
359 	load_gs_index(0);
360 	regs->ip		= new_ip;
361 	regs->sp		= new_sp;
362 	percpu_write(old_rsp, new_sp);
363 	regs->cs		= __USER_CS;
364 	regs->ss		= __USER_DS;
365 	regs->flags		= 0x200;
366 	set_fs(USER_DS);
367 	/*
368 	 * Free the old FP and other extended state
369 	 */
370 	free_thread_xstate(current);
371 }
372 EXPORT_SYMBOL_GPL(start_thread);
373 
374 /*
375  *	switch_to(x,y) should switch tasks from x to y.
376  *
377  * This could still be optimized:
378  * - fold all the options into a flag word and test it with a single test.
379  * - could test fs/gs bitsliced
380  *
381  * Kprobes not supported here. Set the probe on schedule instead.
382  * Function graph tracer not supported too.
383  */
384 __notrace_funcgraph struct task_struct *
385 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
386 {
387 	struct thread_struct *prev = &prev_p->thread;
388 	struct thread_struct *next = &next_p->thread;
389 	int cpu = smp_processor_id();
390 	struct tss_struct *tss = &per_cpu(init_tss, cpu);
391 	unsigned fsindex, gsindex;
392 	bool preload_fpu;
393 
394 	/*
395 	 * If the task has used fpu the last 5 timeslices, just do a full
396 	 * restore of the math state immediately to avoid the trap; the
397 	 * chances of needing FPU soon are obviously high now
398 	 */
399 	preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5;
400 
401 	/* we're going to use this soon, after a few expensive things */
402 	if (preload_fpu)
403 		prefetch(next->xstate);
404 
405 	/*
406 	 * Reload esp0, LDT and the page table pointer:
407 	 */
408 	load_sp0(tss, next);
409 
410 	/*
411 	 * Switch DS and ES.
412 	 * This won't pick up thread selector changes, but I guess that is ok.
413 	 */
414 	savesegment(es, prev->es);
415 	if (unlikely(next->es | prev->es))
416 		loadsegment(es, next->es);
417 
418 	savesegment(ds, prev->ds);
419 	if (unlikely(next->ds | prev->ds))
420 		loadsegment(ds, next->ds);
421 
422 
423 	/* We must save %fs and %gs before load_TLS() because
424 	 * %fs and %gs may be cleared by load_TLS().
425 	 *
426 	 * (e.g. xen_load_tls())
427 	 */
428 	savesegment(fs, fsindex);
429 	savesegment(gs, gsindex);
430 
431 	load_TLS(next, cpu);
432 
433 	/* Must be after DS reload */
434 	unlazy_fpu(prev_p);
435 
436 	/* Make sure cpu is ready for new context */
437 	if (preload_fpu)
438 		clts();
439 
440 	/*
441 	 * Leave lazy mode, flushing any hypercalls made here.
442 	 * This must be done before restoring TLS segments so
443 	 * the GDT and LDT are properly updated, and must be
444 	 * done before math_state_restore, so the TS bit is up
445 	 * to date.
446 	 */
447 	arch_end_context_switch(next_p);
448 
449 	/*
450 	 * Switch FS and GS.
451 	 *
452 	 * Segment register != 0 always requires a reload.  Also
453 	 * reload when it has changed.  When prev process used 64bit
454 	 * base always reload to avoid an information leak.
455 	 */
456 	if (unlikely(fsindex | next->fsindex | prev->fs)) {
457 		loadsegment(fs, next->fsindex);
458 		/*
459 		 * Check if the user used a selector != 0; if yes
460 		 *  clear 64bit base, since overloaded base is always
461 		 *  mapped to the Null selector
462 		 */
463 		if (fsindex)
464 			prev->fs = 0;
465 	}
466 	/* when next process has a 64bit base use it */
467 	if (next->fs)
468 		wrmsrl(MSR_FS_BASE, next->fs);
469 	prev->fsindex = fsindex;
470 
471 	if (unlikely(gsindex | next->gsindex | prev->gs)) {
472 		load_gs_index(next->gsindex);
473 		if (gsindex)
474 			prev->gs = 0;
475 	}
476 	if (next->gs)
477 		wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
478 	prev->gsindex = gsindex;
479 
480 	/*
481 	 * Switch the PDA and FPU contexts.
482 	 */
483 	prev->usersp = percpu_read(old_rsp);
484 	percpu_write(old_rsp, next->usersp);
485 	percpu_write(current_task, next_p);
486 
487 	percpu_write(kernel_stack,
488 		  (unsigned long)task_stack_page(next_p) +
489 		  THREAD_SIZE - KERNEL_STACK_OFFSET);
490 
491 	/*
492 	 * Now maybe reload the debug registers and handle I/O bitmaps
493 	 */
494 	if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
495 		     task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
496 		__switch_to_xtra(prev_p, next_p, tss);
497 
498 	/*
499 	 * Preload the FPU context, now that we've determined that the
500 	 * task is likely to be using it.
501 	 */
502 	if (preload_fpu)
503 		__math_state_restore();
504 
505 	return prev_p;
506 }
507 
508 /*
509  * sys_execve() executes a new program.
510  */
511 asmlinkage
512 long sys_execve(char __user *name, char __user * __user *argv,
513 		char __user * __user *envp, struct pt_regs *regs)
514 {
515 	long error;
516 	char *filename;
517 
518 	filename = getname(name);
519 	error = PTR_ERR(filename);
520 	if (IS_ERR(filename))
521 		return error;
522 	error = do_execve(filename, argv, envp, regs);
523 	putname(filename);
524 	return error;
525 }
526 
527 void set_personality_64bit(void)
528 {
529 	/* inherit personality from parent */
530 
531 	/* Make sure to be in 64bit mode */
532 	clear_thread_flag(TIF_IA32);
533 
534 	/* TBD: overwrites user setup. Should have two bits.
535 	   But 64bit processes have always behaved this way,
536 	   so it's not too bad. The main problem is just that
537 	   32bit childs are affected again. */
538 	current->personality &= ~READ_IMPLIES_EXEC;
539 }
540 
541 asmlinkage long
542 sys_clone(unsigned long clone_flags, unsigned long newsp,
543 	  void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
544 {
545 	if (!newsp)
546 		newsp = regs->sp;
547 	return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
548 }
549 
550 unsigned long get_wchan(struct task_struct *p)
551 {
552 	unsigned long stack;
553 	u64 fp, ip;
554 	int count = 0;
555 
556 	if (!p || p == current || p->state == TASK_RUNNING)
557 		return 0;
558 	stack = (unsigned long)task_stack_page(p);
559 	if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
560 		return 0;
561 	fp = *(u64 *)(p->thread.sp);
562 	do {
563 		if (fp < (unsigned long)stack ||
564 		    fp >= (unsigned long)stack+THREAD_SIZE)
565 			return 0;
566 		ip = *(u64 *)(fp+8);
567 		if (!in_sched_functions(ip))
568 			return ip;
569 		fp = *(u64 *)fp;
570 	} while (count++ < 16);
571 	return 0;
572 }
573 
574 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
575 {
576 	int ret = 0;
577 	int doit = task == current;
578 	int cpu;
579 
580 	switch (code) {
581 	case ARCH_SET_GS:
582 		if (addr >= TASK_SIZE_OF(task))
583 			return -EPERM;
584 		cpu = get_cpu();
585 		/* handle small bases via the GDT because that's faster to
586 		   switch. */
587 		if (addr <= 0xffffffff) {
588 			set_32bit_tls(task, GS_TLS, addr);
589 			if (doit) {
590 				load_TLS(&task->thread, cpu);
591 				load_gs_index(GS_TLS_SEL);
592 			}
593 			task->thread.gsindex = GS_TLS_SEL;
594 			task->thread.gs = 0;
595 		} else {
596 			task->thread.gsindex = 0;
597 			task->thread.gs = addr;
598 			if (doit) {
599 				load_gs_index(0);
600 				ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
601 			}
602 		}
603 		put_cpu();
604 		break;
605 	case ARCH_SET_FS:
606 		/* Not strictly needed for fs, but do it for symmetry
607 		   with gs */
608 		if (addr >= TASK_SIZE_OF(task))
609 			return -EPERM;
610 		cpu = get_cpu();
611 		/* handle small bases via the GDT because that's faster to
612 		   switch. */
613 		if (addr <= 0xffffffff) {
614 			set_32bit_tls(task, FS_TLS, addr);
615 			if (doit) {
616 				load_TLS(&task->thread, cpu);
617 				loadsegment(fs, FS_TLS_SEL);
618 			}
619 			task->thread.fsindex = FS_TLS_SEL;
620 			task->thread.fs = 0;
621 		} else {
622 			task->thread.fsindex = 0;
623 			task->thread.fs = addr;
624 			if (doit) {
625 				/* set the selector to 0 to not confuse
626 				   __switch_to */
627 				loadsegment(fs, 0);
628 				ret = checking_wrmsrl(MSR_FS_BASE, addr);
629 			}
630 		}
631 		put_cpu();
632 		break;
633 	case ARCH_GET_FS: {
634 		unsigned long base;
635 		if (task->thread.fsindex == FS_TLS_SEL)
636 			base = read_32bit_tls(task, FS_TLS);
637 		else if (doit)
638 			rdmsrl(MSR_FS_BASE, base);
639 		else
640 			base = task->thread.fs;
641 		ret = put_user(base, (unsigned long __user *)addr);
642 		break;
643 	}
644 	case ARCH_GET_GS: {
645 		unsigned long base;
646 		unsigned gsindex;
647 		if (task->thread.gsindex == GS_TLS_SEL)
648 			base = read_32bit_tls(task, GS_TLS);
649 		else if (doit) {
650 			savesegment(gs, gsindex);
651 			if (gsindex)
652 				rdmsrl(MSR_KERNEL_GS_BASE, base);
653 			else
654 				base = task->thread.gs;
655 		} else
656 			base = task->thread.gs;
657 		ret = put_user(base, (unsigned long __user *)addr);
658 		break;
659 	}
660 
661 	default:
662 		ret = -EINVAL;
663 		break;
664 	}
665 
666 	return ret;
667 }
668 
669 long sys_arch_prctl(int code, unsigned long addr)
670 {
671 	return do_arch_prctl(current, code, addr);
672 }
673 
674 unsigned long KSTK_ESP(struct task_struct *task)
675 {
676 	return (test_tsk_thread_flag(task, TIF_IA32)) ?
677 			(task_pt_regs(task)->sp) : ((task)->thread.usersp);
678 }
679