Linux Audio

Check our new training course

Loading...
v3.1
 
  1/*
  2 *  Copyright (C) 1995  Linus Torvalds
  3 *
  4 *  Pentium III FXSR, SSE support
  5 *	Gareth Hughes <gareth@valinux.com>, May 2000
  6 *
  7 *  X86-64 port
  8 *	Andi Kleen.
  9 *
 10 *	CPU hotplug support - ashok.raj@intel.com
 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/delay.h>
 30#include <linux/module.h>
 31#include <linux/ptrace.h>
 32#include <linux/notifier.h>
 33#include <linux/kprobes.h>
 34#include <linux/kdebug.h>
 35#include <linux/tick.h>
 36#include <linux/prctl.h>
 37#include <linux/uaccess.h>
 38#include <linux/io.h>
 39#include <linux/ftrace.h>
 40#include <linux/cpuidle.h>
 41
 42#include <asm/pgtable.h>
 43#include <asm/system.h>
 44#include <asm/processor.h>
 45#include <asm/i387.h>
 
 46#include <asm/mmu_context.h>
 47#include <asm/prctl.h>
 48#include <asm/desc.h>
 49#include <asm/proto.h>
 50#include <asm/ia32.h>
 51#include <asm/idle.h>
 52#include <asm/syscalls.h>
 53#include <asm/debugreg.h>
 
 
 
 
 
 
 
 
 
 
 54
 55asmlinkage extern void ret_from_fork(void);
 56
 57DEFINE_PER_CPU(unsigned long, old_rsp);
 58static DEFINE_PER_CPU(unsigned char, is_idle);
 59
 60static ATOMIC_NOTIFIER_HEAD(idle_notifier);
 61
 62void idle_notifier_register(struct notifier_block *n)
 
 
 63{
 64	atomic_notifier_chain_register(&idle_notifier, n);
 65}
 66EXPORT_SYMBOL_GPL(idle_notifier_register);
 
 67
 68void idle_notifier_unregister(struct notifier_block *n)
 69{
 70	atomic_notifier_chain_unregister(&idle_notifier, n);
 71}
 72EXPORT_SYMBOL_GPL(idle_notifier_unregister);
 73
 74void enter_idle(void)
 75{
 76	percpu_write(is_idle, 1);
 77	atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
 78}
 79
 80static void __exit_idle(void)
 81{
 82	if (x86_test_and_clear_bit_percpu(0, is_idle) == 0)
 83		return;
 84	atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
 85}
 
 
 
 
 86
 87/* Called from interrupts to signify idle end */
 88void exit_idle(void)
 89{
 90	/* idle loop has pid 0 */
 91	if (current->pid)
 92		return;
 93	__exit_idle();
 94}
 95
 96#ifndef CONFIG_SMP
 97static inline void play_dead(void)
 98{
 99	BUG();
100}
101#endif
102
103/*
104 * The idle thread. There's no useful work to be
105 * done, so just try to conserve power and have a
106 * low exit latency (ie sit in a loop waiting for
107 * somebody to say that they'd like to reschedule)
108 */
109void cpu_idle(void)
110{
111	current_thread_info()->status |= TS_POLLING;
112
113	/*
114	 * If we're the non-boot CPU, nothing set the stack canary up
115	 * for us.  CPU0 already has it initialized but no harm in
116	 * doing it again.  This is a good place for updating it, as
117	 * we wont ever return from this function (so the invalid
118	 * canaries already on the stack wont ever trigger).
119	 */
120	boot_init_stack_canary();
121
122	/* endless idle loop with no priority at all */
123	while (1) {
124		tick_nohz_stop_sched_tick(1);
125		while (!need_resched()) {
126
127			rmb();
128
129			if (cpu_is_offline(smp_processor_id()))
130				play_dead();
131			/*
132			 * Idle routines should keep interrupts disabled
133			 * from here on, until they go to idle.
134			 * Otherwise, idle callbacks can misfire.
135			 */
136			local_irq_disable();
137			enter_idle();
138			/* Don't trace irqs off for idle */
139			stop_critical_timings();
140			if (cpuidle_idle_call())
141				pm_idle();
142			start_critical_timings();
143
144			/* In many cases the interrupt that ended idle
145			   has already called exit_idle. But some idle
146			   loops can be woken up without interrupt. */
147			__exit_idle();
148		}
149
150		tick_nohz_restart_sched_tick();
151		preempt_enable_no_resched();
152		schedule();
153		preempt_disable();
154	}
155}
156
157/* Prints also some state that isn't saved in the pt_regs */
158void __show_regs(struct pt_regs *regs, int all)
159{
160	unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
161	unsigned long d0, d1, d2, d3, d6, d7;
162	unsigned int fsindex, gsindex;
163	unsigned int ds, cs, es;
164
165	show_regs_common();
166	printk(KERN_DEFAULT "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
167	printk_address(regs->ip, 1);
168	printk(KERN_DEFAULT "RSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss,
169			regs->sp, regs->flags);
170	printk(KERN_DEFAULT "RAX: %016lx RBX: %016lx RCX: %016lx\n",
171	       regs->ax, regs->bx, regs->cx);
172	printk(KERN_DEFAULT "RDX: %016lx RSI: %016lx RDI: %016lx\n",
173	       regs->dx, regs->si, regs->di);
174	printk(KERN_DEFAULT "RBP: %016lx R08: %016lx R09: %016lx\n",
175	       regs->bp, regs->r8, regs->r9);
176	printk(KERN_DEFAULT "R10: %016lx R11: %016lx R12: %016lx\n",
177	       regs->r10, regs->r11, regs->r12);
178	printk(KERN_DEFAULT "R13: %016lx R14: %016lx R15: %016lx\n",
179	       regs->r13, regs->r14, regs->r15);
180
181	asm("movl %%ds,%0" : "=r" (ds));
182	asm("movl %%cs,%0" : "=r" (cs));
183	asm("movl %%es,%0" : "=r" (es));
184	asm("movl %%fs,%0" : "=r" (fsindex));
185	asm("movl %%gs,%0" : "=r" (gsindex));
186
187	rdmsrl(MSR_FS_BASE, fs);
188	rdmsrl(MSR_GS_BASE, gs);
189	rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
190
191	if (!all)
192		return;
193
194	cr0 = read_cr0();
195	cr2 = read_cr2();
196	cr3 = read_cr3();
197	cr4 = read_cr4();
198
199	printk(KERN_DEFAULT "FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
200	       fs, fsindex, gs, gsindex, shadowgs);
201	printk(KERN_DEFAULT "CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
202			es, cr0);
203	printk(KERN_DEFAULT "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
204			cr4);
205
206	get_debugreg(d0, 0);
207	get_debugreg(d1, 1);
208	get_debugreg(d2, 2);
209	printk(KERN_DEFAULT "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
210	get_debugreg(d3, 3);
211	get_debugreg(d6, 6);
212	get_debugreg(d7, 7);
213	printk(KERN_DEFAULT "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
 
 
 
 
 
 
 
 
 
 
 
214}
215
216void release_thread(struct task_struct *dead_task)
217{
218	if (dead_task->mm) {
219		if (dead_task->mm->context.size) {
220			printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
221					dead_task->comm,
222					dead_task->mm->context.ldt,
223					dead_task->mm->context.size);
224			BUG();
225		}
226	}
227}
228
229static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
 
 
 
 
 
 
 
 
 
 
 
 
 
230{
231	struct user_desc ud = {
232		.base_addr = addr,
233		.limit = 0xfffff,
234		.seg_32bit = 1,
235		.limit_in_pages = 1,
236		.useable = 1,
237	};
238	struct desc_struct *desc = t->thread.tls_array;
239	desc += tls;
240	fill_ldt(desc, &ud);
 
 
 
 
 
241}
242
243static inline u32 read_32bit_tls(struct task_struct *t, int tls)
 
 
 
 
 
 
 
 
244{
245	return get_desc_base(&t->thread.tls_array[tls]);
 
 
 
 
 
 
 
 
 
 
246}
247
248/*
249 * This gets called before we allocate a new thread and copy
250 * the current task into it.
 
 
251 */
252void prepare_to_copy(struct task_struct *tsk)
 
 
253{
254	unlazy_fpu(tsk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
255}
256
257int copy_thread(unsigned long clone_flags, unsigned long sp,
258		unsigned long unused,
259	struct task_struct *p, struct pt_regs *regs)
260{
261	int err;
262	struct pt_regs *childregs;
263	struct task_struct *me = current;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
264
265	childregs = ((struct pt_regs *)
266			(THREAD_SIZE + task_stack_page(p))) - 1;
267	*childregs = *regs;
 
 
 
 
 
268
269	childregs->ax = 0;
270	if (user_mode(regs))
271		childregs->sp = sp;
 
 
272	else
273		childregs->sp = (unsigned long)childregs;
 
274
275	p->thread.sp = (unsigned long) childregs;
276	p->thread.sp0 = (unsigned long) (childregs+1);
277	p->thread.usersp = me->thread.usersp;
278
279	set_tsk_thread_flag(p, TIF_FORK);
280
281	p->thread.io_bitmap_ptr = NULL;
282
283	savesegment(gs, p->thread.gsindex);
284	p->thread.gs = p->thread.gsindex ? 0 : me->thread.gs;
285	savesegment(fs, p->thread.fsindex);
286	p->thread.fs = p->thread.fsindex ? 0 : me->thread.fs;
287	savesegment(es, p->thread.es);
288	savesegment(ds, p->thread.ds);
289
290	err = -ENOMEM;
291	memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
292
293	if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
294		p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
295		if (!p->thread.io_bitmap_ptr) {
296			p->thread.io_bitmap_max = 0;
297			return -ENOMEM;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
298		}
299		memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
300				IO_BITMAP_BYTES);
301		set_tsk_thread_flag(p, TIF_IO_BITMAP);
 
 
 
302	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
303
304	/*
305	 * Set a new TLS for the child thread?
 
306	 */
307	if (clone_flags & CLONE_SETTLS) {
308#ifdef CONFIG_IA32_EMULATION
309		if (test_thread_flag(TIF_IA32))
310			err = do_set_thread_area(p, -1,
311				(struct user_desc __user *)childregs->si, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
312		else
 
 
 
 
313#endif
314			err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
315		if (err)
316			goto out;
317	}
318	err = 0;
319out:
320	if (err && p->thread.io_bitmap_ptr) {
321		kfree(p->thread.io_bitmap_ptr);
322		p->thread.io_bitmap_max = 0;
323	}
324
325	return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
326}
327
328static void
329start_thread_common(struct pt_regs *regs, unsigned long new_ip,
330		    unsigned long new_sp,
331		    unsigned int _cs, unsigned int _ss, unsigned int _ds)
332{
 
 
 
 
 
 
 
 
333	loadsegment(fs, 0);
334	loadsegment(es, _ds);
335	loadsegment(ds, _ds);
336	load_gs_index(0);
 
337	regs->ip		= new_ip;
338	regs->sp		= new_sp;
339	percpu_write(old_rsp, new_sp);
340	regs->cs		= _cs;
341	regs->ss		= _ss;
342	regs->flags		= X86_EFLAGS_IF;
343	/*
344	 * Free the old FP and other extended state
345	 */
346	free_thread_xstate(current);
347}
348
349void
350start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
351{
352	start_thread_common(regs, new_ip, new_sp,
353			    __USER_CS, __USER_DS, 0);
354}
 
355
356#ifdef CONFIG_IA32_EMULATION
357void start_thread_ia32(struct pt_regs *regs, u32 new_ip, u32 new_sp)
358{
359	start_thread_common(regs, new_ip, new_sp,
360			    __USER32_CS, __USER32_DS, __USER32_DS);
 
361}
362#endif
363
364/*
365 *	switch_to(x,y) should switch tasks from x to y.
366 *
367 * This could still be optimized:
368 * - fold all the options into a flag word and test it with a single test.
369 * - could test fs/gs bitsliced
370 *
371 * Kprobes not supported here. Set the probe on schedule instead.
372 * Function graph tracer not supported too.
373 */
374__notrace_funcgraph struct task_struct *
375__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
376{
377	struct thread_struct *prev = &prev_p->thread;
378	struct thread_struct *next = &next_p->thread;
 
 
379	int cpu = smp_processor_id();
380	struct tss_struct *tss = &per_cpu(init_tss, cpu);
381	unsigned fsindex, gsindex;
382	bool preload_fpu;
383
384	/*
385	 * If the task has used fpu the last 5 timeslices, just do a full
386	 * restore of the math state immediately to avoid the trap; the
387	 * chances of needing FPU soon are obviously high now
388	 */
389	preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5;
390
391	/* we're going to use this soon, after a few expensive things */
392	if (preload_fpu)
393		prefetch(next->fpu.state);
 
 
 
394
395	/*
396	 * Reload esp0, LDT and the page table pointer:
 
397	 */
398	load_sp0(tss, next);
399
400	/*
401	 * Switch DS and ES.
402	 * This won't pick up thread selector changes, but I guess that is ok.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
403	 */
404	savesegment(es, prev->es);
405	if (unlikely(next->es | prev->es))
406		loadsegment(es, next->es);
407
408	savesegment(ds, prev->ds);
409	if (unlikely(next->ds | prev->ds))
410		loadsegment(ds, next->ds);
411
 
412
413	/* We must save %fs and %gs before load_TLS() because
414	 * %fs and %gs may be cleared by load_TLS().
415	 *
416	 * (e.g. xen_load_tls())
417	 */
418	savesegment(fs, fsindex);
419	savesegment(gs, gsindex);
420
421	load_TLS(next, cpu);
422
423	/* Must be after DS reload */
424	__unlazy_fpu(prev_p);
425
426	/* Make sure cpu is ready for new context */
427	if (preload_fpu)
428		clts();
429
430	/*
431	 * Leave lazy mode, flushing any hypercalls made here.
432	 * This must be done before restoring TLS segments so
433	 * the GDT and LDT are properly updated, and must be
434	 * done before math_state_restore, so the TS bit is up
435	 * to date.
436	 */
437	arch_end_context_switch(next_p);
 
438
439	/*
440	 * Switch FS and GS.
441	 *
442	 * Segment register != 0 always requires a reload.  Also
443	 * reload when it has changed.  When prev process used 64bit
444	 * base always reload to avoid an information leak.
445	 */
446	if (unlikely(fsindex | next->fsindex | prev->fs)) {
447		loadsegment(fs, next->fsindex);
448		/*
449		 * Check if the user used a selector != 0; if yes
450		 *  clear 64bit base, since overloaded base is always
451		 *  mapped to the Null selector
452		 */
453		if (fsindex)
454			prev->fs = 0;
455	}
456	/* when next process has a 64bit base use it */
457	if (next->fs)
458		wrmsrl(MSR_FS_BASE, next->fs);
459	prev->fsindex = fsindex;
460
461	if (unlikely(gsindex | next->gsindex | prev->gs)) {
462		load_gs_index(next->gsindex);
463		if (gsindex)
464			prev->gs = 0;
465	}
466	if (next->gs)
467		wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
468	prev->gsindex = gsindex;
469
470	/*
471	 * Switch the PDA and FPU contexts.
472	 */
473	prev->usersp = percpu_read(old_rsp);
474	percpu_write(old_rsp, next->usersp);
475	percpu_write(current_task, next_p);
476
477	percpu_write(kernel_stack,
478		  (unsigned long)task_stack_page(next_p) +
479		  THREAD_SIZE - KERNEL_STACK_OFFSET);
480
481	/*
482	 * Now maybe reload the debug registers and handle I/O bitmaps
483	 */
484	if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
485		     task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
486		__switch_to_xtra(prev_p, next_p, tss);
487
488	/*
489	 * Preload the FPU context, now that we've determined that the
490	 * task is likely to be using it. 
491	 */
492	if (preload_fpu)
493		__math_state_restore();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
494
495	return prev_p;
496}
497
498void set_personality_64bit(void)
499{
500	/* inherit personality from parent */
501
502	/* Make sure to be in 64bit mode */
503	clear_thread_flag(TIF_IA32);
504
505	/* Ensure the corresponding mm is not marked. */
 
506	if (current->mm)
507		current->mm->context.ia32_compat = 0;
508
509	/* TBD: overwrites user setup. Should have two bits.
510	   But 64bit processes have always behaved this way,
511	   so it's not too bad. The main problem is just that
512	   32bit childs are affected again. */
513	current->personality &= ~READ_IMPLIES_EXEC;
514}
515
516void set_personality_ia32(void)
517{
518	/* inherit personality from parent */
 
 
519
520	/* Make sure to be in 32bit mode */
521	set_thread_flag(TIF_IA32);
522	current->personality |= force_personality32;
 
 
 
 
 
 
 
 
 
 
523
524	/* Mark the associated mm as containing 32-bit tasks. */
525	if (current->mm)
526		current->mm->context.ia32_compat = 1;
 
 
 
 
 
 
 
527
 
528	/* Prepare the first "return" to user space */
 
529	current_thread_info()->status |= TS_COMPAT;
 
530}
531
532unsigned long get_wchan(struct task_struct *p)
533{
534	unsigned long stack;
535	u64 fp, ip;
536	int count = 0;
537
538	if (!p || p == current || p->state == TASK_RUNNING)
539		return 0;
540	stack = (unsigned long)task_stack_page(p);
541	if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
542		return 0;
543	fp = *(u64 *)(p->thread.sp);
544	do {
545		if (fp < (unsigned long)stack ||
546		    fp >= (unsigned long)stack+THREAD_SIZE)
547			return 0;
548		ip = *(u64 *)(fp+8);
549		if (!in_sched_functions(ip))
550			return ip;
551		fp = *(u64 *)fp;
552	} while (count++ < 16);
553	return 0;
554}
 
555
556long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
557{
558	int ret = 0;
559	int doit = task == current;
560	int cpu;
561
562	switch (code) {
563	case ARCH_SET_GS:
564		if (addr >= TASK_SIZE_OF(task))
565			return -EPERM;
566		cpu = get_cpu();
567		/* handle small bases via the GDT because that's faster to
568		   switch. */
569		if (addr <= 0xffffffff) {
570			set_32bit_tls(task, GS_TLS, addr);
571			if (doit) {
572				load_TLS(&task->thread, cpu);
573				load_gs_index(GS_TLS_SEL);
574			}
575			task->thread.gsindex = GS_TLS_SEL;
576			task->thread.gs = 0;
 
 
 
 
 
 
 
577		} else {
578			task->thread.gsindex = 0;
579			task->thread.gs = addr;
580			if (doit) {
581				load_gs_index(0);
582				ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
583			}
584		}
585		put_cpu();
586		break;
587	case ARCH_SET_FS:
588		/* Not strictly needed for fs, but do it for symmetry
589		   with gs */
590		if (addr >= TASK_SIZE_OF(task))
 
 
 
591			return -EPERM;
592		cpu = get_cpu();
593		/* handle small bases via the GDT because that's faster to
594		   switch. */
595		if (addr <= 0xffffffff) {
596			set_32bit_tls(task, FS_TLS, addr);
597			if (doit) {
598				load_TLS(&task->thread, cpu);
599				loadsegment(fs, FS_TLS_SEL);
600			}
601			task->thread.fsindex = FS_TLS_SEL;
602			task->thread.fs = 0;
 
 
 
 
603		} else {
604			task->thread.fsindex = 0;
605			task->thread.fs = addr;
606			if (doit) {
607				/* set the selector to 0 to not confuse
608				   __switch_to */
609				loadsegment(fs, 0);
610				ret = checking_wrmsrl(MSR_FS_BASE, addr);
611			}
612		}
613		put_cpu();
614		break;
 
615	case ARCH_GET_FS: {
616		unsigned long base;
617		if (task->thread.fsindex == FS_TLS_SEL)
618			base = read_32bit_tls(task, FS_TLS);
619		else if (doit)
620			rdmsrl(MSR_FS_BASE, base);
621		else
622			base = task->thread.fs;
623		ret = put_user(base, (unsigned long __user *)addr);
624		break;
625	}
626	case ARCH_GET_GS: {
627		unsigned long base;
628		unsigned gsindex;
629		if (task->thread.gsindex == GS_TLS_SEL)
630			base = read_32bit_tls(task, GS_TLS);
631		else if (doit) {
632			savesegment(gs, gsindex);
633			if (gsindex)
634				rdmsrl(MSR_KERNEL_GS_BASE, base);
635			else
636				base = task->thread.gs;
637		} else
638			base = task->thread.gs;
639		ret = put_user(base, (unsigned long __user *)addr);
640		break;
641	}
642
 
 
 
 
 
 
 
 
 
 
 
 
 
643	default:
644		ret = -EINVAL;
645		break;
646	}
647
648	return ret;
649}
650
651long sys_arch_prctl(int code, unsigned long addr)
652{
653	return do_arch_prctl(current, code, addr);
 
 
 
 
 
 
 
 
 
 
 
 
654}
 
655
656unsigned long KSTK_ESP(struct task_struct *task)
657{
658	return (test_tsk_thread_flag(task, TIF_IA32)) ?
659			(task_pt_regs(task)->sp) : ((task)->thread.usersp);
660}
v5.14.15
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 *  Copyright (C) 1995  Linus Torvalds
  4 *
  5 *  Pentium III FXSR, SSE support
  6 *	Gareth Hughes <gareth@valinux.com>, May 2000
  7 *
  8 *  X86-64 port
  9 *	Andi Kleen.
 10 *
 11 *	CPU hotplug support - ashok.raj@intel.com
 12 */
 13
 14/*
 15 * This file handles the architecture-dependent parts of process handling..
 16 */
 17
 
 18#include <linux/cpu.h>
 19#include <linux/errno.h>
 20#include <linux/sched.h>
 21#include <linux/sched/task.h>
 22#include <linux/sched/task_stack.h>
 23#include <linux/fs.h>
 24#include <linux/kernel.h>
 25#include <linux/mm.h>
 26#include <linux/elfcore.h>
 27#include <linux/smp.h>
 28#include <linux/slab.h>
 29#include <linux/user.h>
 30#include <linux/interrupt.h>
 31#include <linux/delay.h>
 32#include <linux/export.h>
 33#include <linux/ptrace.h>
 34#include <linux/notifier.h>
 35#include <linux/kprobes.h>
 36#include <linux/kdebug.h>
 
 37#include <linux/prctl.h>
 38#include <linux/uaccess.h>
 39#include <linux/io.h>
 40#include <linux/ftrace.h>
 41#include <linux/syscalls.h>
 42
 
 
 43#include <asm/processor.h>
 44#include <asm/pkru.h>
 45#include <asm/fpu/internal.h>
 46#include <asm/mmu_context.h>
 47#include <asm/prctl.h>
 48#include <asm/desc.h>
 49#include <asm/proto.h>
 50#include <asm/ia32.h>
 
 
 51#include <asm/debugreg.h>
 52#include <asm/switch_to.h>
 53#include <asm/xen/hypervisor.h>
 54#include <asm/vdso.h>
 55#include <asm/resctrl.h>
 56#include <asm/unistd.h>
 57#include <asm/fsgsbase.h>
 58#ifdef CONFIG_IA32_EMULATION
 59/* Not included via unistd.h */
 60#include <asm/unistd_32_ia32.h>
 61#endif
 62
 63#include "process.h"
 
 
 
 
 
 64
 65/* Prints also some state that isn't saved in the pt_regs */
 66void __show_regs(struct pt_regs *regs, enum show_regs_mode mode,
 67		 const char *log_lvl)
 68{
 69	unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
 70	unsigned long d0, d1, d2, d3, d6, d7;
 71	unsigned int fsindex, gsindex;
 72	unsigned int ds, es;
 73
 74	show_iret_regs(regs, log_lvl);
 
 
 
 
 75
 76	if (regs->orig_ax != -1)
 77		pr_cont(" ORIG_RAX: %016lx\n", regs->orig_ax);
 78	else
 79		pr_cont("\n");
 
 80
 81	printk("%sRAX: %016lx RBX: %016lx RCX: %016lx\n",
 82	       log_lvl, regs->ax, regs->bx, regs->cx);
 83	printk("%sRDX: %016lx RSI: %016lx RDI: %016lx\n",
 84	       log_lvl, regs->dx, regs->si, regs->di);
 85	printk("%sRBP: %016lx R08: %016lx R09: %016lx\n",
 86	       log_lvl, regs->bp, regs->r8, regs->r9);
 87	printk("%sR10: %016lx R11: %016lx R12: %016lx\n",
 88	       log_lvl, regs->r10, regs->r11, regs->r12);
 89	printk("%sR13: %016lx R14: %016lx R15: %016lx\n",
 90	       log_lvl, regs->r13, regs->r14, regs->r15);
 91
 92	if (mode == SHOW_REGS_SHORT)
 
 
 
 
 93		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 94
 95	if (mode == SHOW_REGS_USER) {
 96		rdmsrl(MSR_FS_BASE, fs);
 97		rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
 98		printk("%sFS:  %016lx GS:  %016lx\n",
 99		       log_lvl, fs, shadowgs);
100		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
101	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
102
103	asm("movl %%ds,%0" : "=r" (ds));
 
104	asm("movl %%es,%0" : "=r" (es));
105	asm("movl %%fs,%0" : "=r" (fsindex));
106	asm("movl %%gs,%0" : "=r" (gsindex));
107
108	rdmsrl(MSR_FS_BASE, fs);
109	rdmsrl(MSR_GS_BASE, gs);
110	rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
111
 
 
 
112	cr0 = read_cr0();
113	cr2 = read_cr2();
114	cr3 = __read_cr3();
115	cr4 = __read_cr4();
116
117	printk("%sFS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
118	       log_lvl, fs, fsindex, gs, gsindex, shadowgs);
119	printk("%sCS:  %04lx DS: %04x ES: %04x CR0: %016lx\n",
120		log_lvl, regs->cs, ds, es, cr0);
121	printk("%sCR2: %016lx CR3: %016lx CR4: %016lx\n",
122		log_lvl, cr2, cr3, cr4);
123
124	get_debugreg(d0, 0);
125	get_debugreg(d1, 1);
126	get_debugreg(d2, 2);
 
127	get_debugreg(d3, 3);
128	get_debugreg(d6, 6);
129	get_debugreg(d7, 7);
130
131	/* Only print out debug registers if they are in their non-default state. */
132	if (!((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
133	    (d6 == DR6_RESERVED) && (d7 == 0x400))) {
134		printk("%sDR0: %016lx DR1: %016lx DR2: %016lx\n",
135		       log_lvl, d0, d1, d2);
136		printk("%sDR3: %016lx DR6: %016lx DR7: %016lx\n",
137		       log_lvl, d3, d6, d7);
138	}
139
140	if (cpu_feature_enabled(X86_FEATURE_OSPKE))
141		printk("%sPKRU: %08x\n", log_lvl, read_pkru());
142}
143
144void release_thread(struct task_struct *dead_task)
145{
146	WARN_ON(dead_task->mm);
 
 
 
 
 
 
 
 
147}
148
149enum which_selector {
150	FS,
151	GS
152};
153
154/*
155 * Out of line to be protected from kprobes and tracing. If this would be
156 * traced or probed than any access to a per CPU variable happens with
157 * the wrong GS.
158 *
159 * It is not used on Xen paravirt. When paravirt support is needed, it
160 * needs to be renamed with native_ prefix.
161 */
162static noinstr unsigned long __rdgsbase_inactive(void)
163{
164	unsigned long gsbase;
165
166	lockdep_assert_irqs_disabled();
167
168	if (!static_cpu_has(X86_FEATURE_XENPV)) {
169		native_swapgs();
170		gsbase = rdgsbase();
171		native_swapgs();
172	} else {
173		instrumentation_begin();
174		rdmsrl(MSR_KERNEL_GS_BASE, gsbase);
175		instrumentation_end();
176	}
177
178	return gsbase;
179}
180
181/*
182 * Out of line to be protected from kprobes and tracing. If this would be
183 * traced or probed than any access to a per CPU variable happens with
184 * the wrong GS.
185 *
186 * It is not used on Xen paravirt. When paravirt support is needed, it
187 * needs to be renamed with native_ prefix.
188 */
189static noinstr void __wrgsbase_inactive(unsigned long gsbase)
190{
191	lockdep_assert_irqs_disabled();
192
193	if (!static_cpu_has(X86_FEATURE_XENPV)) {
194		native_swapgs();
195		wrgsbase(gsbase);
196		native_swapgs();
197	} else {
198		instrumentation_begin();
199		wrmsrl(MSR_KERNEL_GS_BASE, gsbase);
200		instrumentation_end();
201	}
202}
203
204/*
205 * Saves the FS or GS base for an outgoing thread if FSGSBASE extensions are
206 * not available.  The goal is to be reasonably fast on non-FSGSBASE systems.
207 * It's forcibly inlined because it'll generate better code and this function
208 * is hot.
209 */
210static __always_inline void save_base_legacy(struct task_struct *prev_p,
211					     unsigned short selector,
212					     enum which_selector which)
213{
214	if (likely(selector == 0)) {
215		/*
216		 * On Intel (without X86_BUG_NULL_SEG), the segment base could
217		 * be the pre-existing saved base or it could be zero.  On AMD
218		 * (with X86_BUG_NULL_SEG), the segment base could be almost
219		 * anything.
220		 *
221		 * This branch is very hot (it's hit twice on almost every
222		 * context switch between 64-bit programs), and avoiding
223		 * the RDMSR helps a lot, so we just assume that whatever
224		 * value is already saved is correct.  This matches historical
225		 * Linux behavior, so it won't break existing applications.
226		 *
227		 * To avoid leaking state, on non-X86_BUG_NULL_SEG CPUs, if we
228		 * report that the base is zero, it needs to actually be zero:
229		 * see the corresponding logic in load_seg_legacy.
230		 */
231	} else {
232		/*
233		 * If the selector is 1, 2, or 3, then the base is zero on
234		 * !X86_BUG_NULL_SEG CPUs and could be anything on
235		 * X86_BUG_NULL_SEG CPUs.  In the latter case, Linux
236		 * has never attempted to preserve the base across context
237		 * switches.
238		 *
239		 * If selector > 3, then it refers to a real segment, and
240		 * saving the base isn't necessary.
241		 */
242		if (which == FS)
243			prev_p->thread.fsbase = 0;
244		else
245			prev_p->thread.gsbase = 0;
246	}
247}
248
249static __always_inline void save_fsgs(struct task_struct *task)
 
 
250{
251	savesegment(fs, task->thread.fsindex);
252	savesegment(gs, task->thread.gsindex);
253	if (static_cpu_has(X86_FEATURE_FSGSBASE)) {
254		/*
255		 * If FSGSBASE is enabled, we can't make any useful guesses
256		 * about the base, and user code expects us to save the current
257		 * value.  Fortunately, reading the base directly is efficient.
258		 */
259		task->thread.fsbase = rdfsbase();
260		task->thread.gsbase = __rdgsbase_inactive();
261	} else {
262		save_base_legacy(task, task->thread.fsindex, FS);
263		save_base_legacy(task, task->thread.gsindex, GS);
264	}
265}
266
267/*
268 * While a process is running,current->thread.fsbase and current->thread.gsbase
269 * may not match the corresponding CPU registers (see save_base_legacy()).
270 */
271void current_save_fsgs(void)
272{
273	unsigned long flags;
274
275	/* Interrupts need to be off for FSGSBASE */
276	local_irq_save(flags);
277	save_fsgs(current);
278	local_irq_restore(flags);
279}
280#if IS_ENABLED(CONFIG_KVM)
281EXPORT_SYMBOL_GPL(current_save_fsgs);
282#endif
283
284static __always_inline void loadseg(enum which_selector which,
285				    unsigned short sel)
286{
287	if (which == FS)
288		loadsegment(fs, sel);
289	else
290		load_gs_index(sel);
291}
292
293static __always_inline void load_seg_legacy(unsigned short prev_index,
294					    unsigned long prev_base,
295					    unsigned short next_index,
296					    unsigned long next_base,
297					    enum which_selector which)
298{
299	if (likely(next_index <= 3)) {
300		/*
301		 * The next task is using 64-bit TLS, is not using this
302		 * segment at all, or is having fun with arcane CPU features.
303		 */
304		if (next_base == 0) {
305			/*
306			 * Nasty case: on AMD CPUs, we need to forcibly zero
307			 * the base.
308			 */
309			if (static_cpu_has_bug(X86_BUG_NULL_SEG)) {
310				loadseg(which, __USER_DS);
311				loadseg(which, next_index);
312			} else {
313				/*
314				 * We could try to exhaustively detect cases
315				 * under which we can skip the segment load,
316				 * but there's really only one case that matters
317				 * for performance: if both the previous and
318				 * next states are fully zeroed, we can skip
319				 * the load.
320				 *
321				 * (This assumes that prev_base == 0 has no
322				 * false positives.  This is the case on
323				 * Intel-style CPUs.)
324				 */
325				if (likely(prev_index | next_index | prev_base))
326					loadseg(which, next_index);
327			}
328		} else {
329			if (prev_index != next_index)
330				loadseg(which, next_index);
331			wrmsrl(which == FS ? MSR_FS_BASE : MSR_KERNEL_GS_BASE,
332			       next_base);
333		}
334	} else {
335		/*
336		 * The next task is using a real segment.  Loading the selector
337		 * is sufficient.
338		 */
339		loadseg(which, next_index);
340	}
341}
342
343/*
344 * Store prev's PKRU value and load next's PKRU value if they differ. PKRU
345 * is not XSTATE managed on context switch because that would require a
346 * lookup in the task's FPU xsave buffer and require to keep that updated
347 * in various places.
348 */
349static __always_inline void x86_pkru_load(struct thread_struct *prev,
350					  struct thread_struct *next)
351{
352	if (!cpu_feature_enabled(X86_FEATURE_OSPKE))
353		return;
354
355	/* Stash the prev task's value: */
356	prev->pkru = rdpkru();
357
358	/*
359	 * PKRU writes are slightly expensive.  Avoid them when not
360	 * strictly necessary:
361	 */
362	if (prev->pkru != next->pkru)
363		wrpkru(next->pkru);
364}
365
366static __always_inline void x86_fsgsbase_load(struct thread_struct *prev,
367					      struct thread_struct *next)
368{
369	if (static_cpu_has(X86_FEATURE_FSGSBASE)) {
370		/* Update the FS and GS selectors if they could have changed. */
371		if (unlikely(prev->fsindex || next->fsindex))
372			loadseg(FS, next->fsindex);
373		if (unlikely(prev->gsindex || next->gsindex))
374			loadseg(GS, next->gsindex);
375
376		/* Update the bases. */
377		wrfsbase(next->fsbase);
378		__wrgsbase_inactive(next->gsbase);
379	} else {
380		load_seg_legacy(prev->fsindex, prev->fsbase,
381				next->fsindex, next->fsbase, FS);
382		load_seg_legacy(prev->gsindex, prev->gsbase,
383				next->gsindex, next->gsbase, GS);
384	}
385}
386
387unsigned long x86_fsgsbase_read_task(struct task_struct *task,
388				     unsigned short selector)
389{
390	unsigned short idx = selector >> 3;
391	unsigned long base;
392
393	if (likely((selector & SEGMENT_TI_MASK) == 0)) {
394		if (unlikely(idx >= GDT_ENTRIES))
395			return 0;
396
397		/*
398		 * There are no user segments in the GDT with nonzero bases
399		 * other than the TLS segments.
400		 */
401		if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
402			return 0;
403
404		idx -= GDT_ENTRY_TLS_MIN;
405		base = get_desc_base(&task->thread.tls_array[idx]);
406	} else {
407#ifdef CONFIG_MODIFY_LDT_SYSCALL
408		struct ldt_struct *ldt;
409
410		/*
411		 * If performance here mattered, we could protect the LDT
412		 * with RCU.  This is a slow path, though, so we can just
413		 * take the mutex.
414		 */
415		mutex_lock(&task->mm->context.lock);
416		ldt = task->mm->context.ldt;
417		if (unlikely(!ldt || idx >= ldt->nr_entries))
418			base = 0;
419		else
420			base = get_desc_base(ldt->entries + idx);
421		mutex_unlock(&task->mm->context.lock);
422#else
423		base = 0;
424#endif
 
 
 
 
 
 
 
 
 
425	}
426
427	return base;
428}
429
430unsigned long x86_gsbase_read_cpu_inactive(void)
431{
432	unsigned long gsbase;
433
434	if (boot_cpu_has(X86_FEATURE_FSGSBASE)) {
435		unsigned long flags;
436
437		local_irq_save(flags);
438		gsbase = __rdgsbase_inactive();
439		local_irq_restore(flags);
440	} else {
441		rdmsrl(MSR_KERNEL_GS_BASE, gsbase);
442	}
443
444	return gsbase;
445}
446
447void x86_gsbase_write_cpu_inactive(unsigned long gsbase)
448{
449	if (boot_cpu_has(X86_FEATURE_FSGSBASE)) {
450		unsigned long flags;
451
452		local_irq_save(flags);
453		__wrgsbase_inactive(gsbase);
454		local_irq_restore(flags);
455	} else {
456		wrmsrl(MSR_KERNEL_GS_BASE, gsbase);
457	}
458}
459
460unsigned long x86_fsbase_read_task(struct task_struct *task)
461{
462	unsigned long fsbase;
463
464	if (task == current)
465		fsbase = x86_fsbase_read_cpu();
466	else if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
467		 (task->thread.fsindex == 0))
468		fsbase = task->thread.fsbase;
469	else
470		fsbase = x86_fsgsbase_read_task(task, task->thread.fsindex);
471
472	return fsbase;
473}
474
475unsigned long x86_gsbase_read_task(struct task_struct *task)
476{
477	unsigned long gsbase;
478
479	if (task == current)
480		gsbase = x86_gsbase_read_cpu_inactive();
481	else if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
482		 (task->thread.gsindex == 0))
483		gsbase = task->thread.gsbase;
484	else
485		gsbase = x86_fsgsbase_read_task(task, task->thread.gsindex);
486
487	return gsbase;
488}
489
490void x86_fsbase_write_task(struct task_struct *task, unsigned long fsbase)
491{
492	WARN_ON_ONCE(task == current);
493
494	task->thread.fsbase = fsbase;
495}
496
497void x86_gsbase_write_task(struct task_struct *task, unsigned long gsbase)
498{
499	WARN_ON_ONCE(task == current);
500
501	task->thread.gsbase = gsbase;
502}
503
504static void
505start_thread_common(struct pt_regs *regs, unsigned long new_ip,
506		    unsigned long new_sp,
507		    unsigned int _cs, unsigned int _ss, unsigned int _ds)
508{
509	WARN_ON_ONCE(regs != current_pt_regs());
510
511	if (static_cpu_has(X86_BUG_NULL_SEG)) {
512		/* Loading zero below won't clear the base. */
513		loadsegment(fs, __USER_DS);
514		load_gs_index(__USER_DS);
515	}
516
517	loadsegment(fs, 0);
518	loadsegment(es, _ds);
519	loadsegment(ds, _ds);
520	load_gs_index(0);
521
522	regs->ip		= new_ip;
523	regs->sp		= new_sp;
 
524	regs->cs		= _cs;
525	regs->ss		= _ss;
526	regs->flags		= X86_EFLAGS_IF;
 
 
 
 
527}
528
529void
530start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
531{
532	start_thread_common(regs, new_ip, new_sp,
533			    __USER_CS, __USER_DS, 0);
534}
535EXPORT_SYMBOL_GPL(start_thread);
536
537#ifdef CONFIG_COMPAT
538void compat_start_thread(struct pt_regs *regs, u32 new_ip, u32 new_sp, bool x32)
539{
540	start_thread_common(regs, new_ip, new_sp,
541			    x32 ? __USER_CS : __USER32_CS,
542			    __USER_DS, __USER_DS);
543}
544#endif
545
546/*
547 *	switch_to(x,y) should switch tasks from x to y.
548 *
549 * This could still be optimized:
550 * - fold all the options into a flag word and test it with a single test.
551 * - could test fs/gs bitsliced
552 *
553 * Kprobes not supported here. Set the probe on schedule instead.
554 * Function graph tracer not supported too.
555 */
556__visible __notrace_funcgraph struct task_struct *
557__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
558{
559	struct thread_struct *prev = &prev_p->thread;
560	struct thread_struct *next = &next_p->thread;
561	struct fpu *prev_fpu = &prev->fpu;
562	struct fpu *next_fpu = &next->fpu;
563	int cpu = smp_processor_id();
 
 
 
564
565	WARN_ON_ONCE(IS_ENABLED(CONFIG_DEBUG_ENTRY) &&
566		     this_cpu_read(hardirq_stack_inuse));
567
568	if (!test_thread_flag(TIF_NEED_FPU_LOAD))
569		switch_fpu_prepare(prev_fpu, cpu);
 
570
571	/* We must save %fs and %gs before load_TLS() because
572	 * %fs and %gs may be cleared by load_TLS().
573	 *
574	 * (e.g. xen_load_tls())
575	 */
576	save_fsgs(prev_p);
577
578	/*
579	 * Load TLS before restoring any segments so that segment loads
580	 * reference the correct GDT entries.
581	 */
582	load_TLS(next, cpu);
583
584	/*
585	 * Leave lazy mode, flushing any hypercalls made here.  This
586	 * must be done after loading TLS entries in the GDT but before
587	 * loading segments that might reference them.
588	 */
589	arch_end_context_switch(next_p);
590
591	/* Switch DS and ES.
592	 *
593	 * Reading them only returns the selectors, but writing them (if
594	 * nonzero) loads the full descriptor from the GDT or LDT.  The
595	 * LDT for next is loaded in switch_mm, and the GDT is loaded
596	 * above.
597	 *
598	 * We therefore need to write new values to the segment
599	 * registers on every context switch unless both the new and old
600	 * values are zero.
601	 *
602	 * Note that we don't need to do anything for CS and SS, as
603	 * those are saved and restored as part of pt_regs.
604	 */
605	savesegment(es, prev->es);
606	if (unlikely(next->es | prev->es))
607		loadsegment(es, next->es);
608
609	savesegment(ds, prev->ds);
610	if (unlikely(next->ds | prev->ds))
611		loadsegment(ds, next->ds);
612
613	x86_fsgsbase_load(prev, next);
614
615	x86_pkru_load(prev, next);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
616
617	/*
618	 * Switch the PDA and FPU contexts.
 
 
 
 
619	 */
620	this_cpu_write(current_task, next_p);
621	this_cpu_write(cpu_current_top_of_stack, task_top_of_stack(next_p));
622
623	switch_fpu_finish(next_fpu);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
624
625	/* Reload sp0. */
626	update_task_stack(next_p);
 
 
 
 
 
 
 
 
627
628	switch_to_extra(prev_p, next_p);
 
 
 
 
 
629
630	if (static_cpu_has_bug(X86_BUG_SYSRET_SS_ATTRS)) {
631		/*
632		 * AMD CPUs have a misfeature: SYSRET sets the SS selector but
633		 * does not update the cached descriptor.  As a result, if we
634		 * do SYSRET while SS is NULL, we'll end up in user mode with
635		 * SS apparently equal to __USER_DS but actually unusable.
636		 *
637		 * The straightforward workaround would be to fix it up just
638		 * before SYSRET, but that would slow down the system call
639		 * fast paths.  Instead, we ensure that SS is never NULL in
640		 * system call context.  We do this by replacing NULL SS
641		 * selectors at every context switch.  SYSCALL sets up a valid
642		 * SS, so the only way to get NULL is to re-enter the kernel
643		 * from CPL 3 through an interrupt.  Since that can't happen
644		 * in the same task as a running syscall, we are guaranteed to
645		 * context switch between every interrupt vector entry and a
646		 * subsequent SYSRET.
647		 *
648		 * We read SS first because SS reads are much faster than
649		 * writes.  Out of caution, we force SS to __KERNEL_DS even if
650		 * it previously had a different non-NULL value.
651		 */
652		unsigned short ss_sel;
653		savesegment(ss, ss_sel);
654		if (ss_sel != __KERNEL_DS)
655			loadsegment(ss, __KERNEL_DS);
656	}
657
658	/* Load the Intel cache allocation PQR MSR. */
659	resctrl_sched_in();
660
661	return prev_p;
662}
663
664void set_personality_64bit(void)
665{
666	/* inherit personality from parent */
667
668	/* Make sure to be in 64bit mode */
669	clear_thread_flag(TIF_ADDR32);
670	/* Pretend that this comes from a 64bit execve */
671	task_pt_regs(current)->orig_ax = __NR_execve;
672	current_thread_info()->status &= ~TS_COMPAT;
673	if (current->mm)
674		current->mm->context.flags = MM_CONTEXT_HAS_VSYSCALL;
675
676	/* TBD: overwrites user setup. Should have two bits.
677	   But 64bit processes have always behaved this way,
678	   so it's not too bad. The main problem is just that
679	   32bit children are affected again. */
680	current->personality &= ~READ_IMPLIES_EXEC;
681}
682
683static void __set_personality_x32(void)
684{
685#ifdef CONFIG_X86_X32
686	if (current->mm)
687		current->mm->context.flags = 0;
688
689	current->personality &= ~READ_IMPLIES_EXEC;
690	/*
691	 * in_32bit_syscall() uses the presence of the x32 syscall bit
692	 * flag to determine compat status.  The x86 mmap() code relies on
693	 * the syscall bitness so set x32 syscall bit right here to make
694	 * in_32bit_syscall() work during exec().
695	 *
696	 * Pretend to come from a x32 execve.
697	 */
698	task_pt_regs(current)->orig_ax = __NR_x32_execve | __X32_SYSCALL_BIT;
699	current_thread_info()->status &= ~TS_COMPAT;
700#endif
701}
702
703static void __set_personality_ia32(void)
704{
705#ifdef CONFIG_IA32_EMULATION
706	if (current->mm) {
707		/*
708		 * uprobes applied to this MM need to know this and
709		 * cannot use user_64bit_mode() at that time.
710		 */
711		current->mm->context.flags = MM_CONTEXT_UPROBE_IA32;
712	}
713
714	current->personality |= force_personality32;
715	/* Prepare the first "return" to user space */
716	task_pt_regs(current)->orig_ax = __NR_ia32_execve;
717	current_thread_info()->status |= TS_COMPAT;
718#endif
719}
720
721void set_personality_ia32(bool x32)
722{
723	/* Make sure to be in 32bit mode */
724	set_thread_flag(TIF_ADDR32);
725
726	if (x32)
727		__set_personality_x32();
728	else
729		__set_personality_ia32();
730}
731EXPORT_SYMBOL_GPL(set_personality_ia32);
732
733#ifdef CONFIG_CHECKPOINT_RESTORE
734static long prctl_map_vdso(const struct vdso_image *image, unsigned long addr)
735{
736	int ret;
737
738	ret = map_vdso_once(image, addr);
739	if (ret)
740		return ret;
741
742	return (long)image->size;
743}
744#endif
745
746long do_arch_prctl_64(struct task_struct *task, int option, unsigned long arg2)
747{
748	int ret = 0;
 
 
749
750	switch (option) {
751	case ARCH_SET_GS: {
752		if (unlikely(arg2 >= TASK_SIZE_MAX))
753			return -EPERM;
754
755		preempt_disable();
756		/*
757		 * ARCH_SET_GS has always overwritten the index
758		 * and the base. Zero is the most sensible value
759		 * to put in the index, and is the only value that
760		 * makes any sense if FSGSBASE is unavailable.
761		 */
762		if (task == current) {
763			loadseg(GS, 0);
764			x86_gsbase_write_cpu_inactive(arg2);
765
766			/*
767			 * On non-FSGSBASE systems, save_base_legacy() expects
768			 * that we also fill in thread.gsbase.
769			 */
770			task->thread.gsbase = arg2;
771
772		} else {
773			task->thread.gsindex = 0;
774			x86_gsbase_write_task(task, arg2);
 
 
 
 
775		}
776		preempt_enable();
777		break;
778	}
779	case ARCH_SET_FS: {
780		/*
781		 * Not strictly needed for %fs, but do it for symmetry
782		 * with %gs
783		 */
784		if (unlikely(arg2 >= TASK_SIZE_MAX))
785			return -EPERM;
786
787		preempt_disable();
788		/*
789		 * Set the selector to 0 for the same reason
790		 * as %gs above.
791		 */
792		if (task == current) {
793			loadseg(FS, 0);
794			x86_fsbase_write_cpu(arg2);
795
796			/*
797			 * On non-FSGSBASE systems, save_base_legacy() expects
798			 * that we also fill in thread.fsbase.
799			 */
800			task->thread.fsbase = arg2;
801		} else {
802			task->thread.fsindex = 0;
803			x86_fsbase_write_task(task, arg2);
 
 
 
 
 
 
804		}
805		preempt_enable();
806		break;
807	}
808	case ARCH_GET_FS: {
809		unsigned long base = x86_fsbase_read_task(task);
810
811		ret = put_user(base, (unsigned long __user *)arg2);
 
 
 
 
 
812		break;
813	}
814	case ARCH_GET_GS: {
815		unsigned long base = x86_gsbase_read_task(task);
816
817		ret = put_user(base, (unsigned long __user *)arg2);
 
 
 
 
 
 
 
 
 
 
818		break;
819	}
820
821#ifdef CONFIG_CHECKPOINT_RESTORE
822# ifdef CONFIG_X86_X32_ABI
823	case ARCH_MAP_VDSO_X32:
824		return prctl_map_vdso(&vdso_image_x32, arg2);
825# endif
826# if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
827	case ARCH_MAP_VDSO_32:
828		return prctl_map_vdso(&vdso_image_32, arg2);
829# endif
830	case ARCH_MAP_VDSO_64:
831		return prctl_map_vdso(&vdso_image_64, arg2);
832#endif
833
834	default:
835		ret = -EINVAL;
836		break;
837	}
838
839	return ret;
840}
841
842SYSCALL_DEFINE2(arch_prctl, int, option, unsigned long, arg2)
843{
844	long ret;
845
846	ret = do_arch_prctl_64(current, option, arg2);
847	if (ret == -EINVAL)
848		ret = do_arch_prctl_common(current, option, arg2);
849
850	return ret;
851}
852
853#ifdef CONFIG_IA32_EMULATION
854COMPAT_SYSCALL_DEFINE2(arch_prctl, int, option, unsigned long, arg2)
855{
856	return do_arch_prctl_common(current, option, arg2);
857}
858#endif
859
860unsigned long KSTK_ESP(struct task_struct *task)
861{
862	return task_pt_regs(task)->sp;
 
863}