Linux Audio

Check our new training course

Loading...
v3.15
  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/cpu.h>
 18#include <linux/errno.h>
 19#include <linux/sched.h>
 20#include <linux/fs.h>
 21#include <linux/kernel.h>
 22#include <linux/mm.h>
 23#include <linux/elfcore.h>
 24#include <linux/smp.h>
 25#include <linux/slab.h>
 26#include <linux/user.h>
 27#include <linux/interrupt.h>
 28#include <linux/delay.h>
 29#include <linux/module.h>
 30#include <linux/ptrace.h>
 31#include <linux/notifier.h>
 32#include <linux/kprobes.h>
 33#include <linux/kdebug.h>
 34#include <linux/prctl.h>
 35#include <linux/uaccess.h>
 36#include <linux/io.h>
 37#include <linux/ftrace.h>
 38
 39#include <asm/pgtable.h>
 40#include <asm/processor.h>
 41#include <asm/i387.h>
 42#include <asm/fpu-internal.h>
 43#include <asm/mmu_context.h>
 44#include <asm/prctl.h>
 45#include <asm/desc.h>
 46#include <asm/proto.h>
 47#include <asm/ia32.h>
 48#include <asm/idle.h>
 49#include <asm/syscalls.h>
 50#include <asm/debugreg.h>
 51#include <asm/switch_to.h>
 
 
 
 52
 53asmlinkage extern void ret_from_fork(void);
 54
 55__visible DEFINE_PER_CPU(unsigned long, old_rsp);
 56
 57/* Prints also some state that isn't saved in the pt_regs */
 58void __show_regs(struct pt_regs *regs, int all)
 59{
 60	unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
 61	unsigned long d0, d1, d2, d3, d6, d7;
 62	unsigned int fsindex, gsindex;
 63	unsigned int ds, cs, es;
 64
 65	printk(KERN_DEFAULT "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
 66	printk_address(regs->ip);
 67	printk(KERN_DEFAULT "RSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss,
 68			regs->sp, regs->flags);
 
 
 
 
 
 69	printk(KERN_DEFAULT "RAX: %016lx RBX: %016lx RCX: %016lx\n",
 70	       regs->ax, regs->bx, regs->cx);
 71	printk(KERN_DEFAULT "RDX: %016lx RSI: %016lx RDI: %016lx\n",
 72	       regs->dx, regs->si, regs->di);
 73	printk(KERN_DEFAULT "RBP: %016lx R08: %016lx R09: %016lx\n",
 74	       regs->bp, regs->r8, regs->r9);
 75	printk(KERN_DEFAULT "R10: %016lx R11: %016lx R12: %016lx\n",
 76	       regs->r10, regs->r11, regs->r12);
 77	printk(KERN_DEFAULT "R13: %016lx R14: %016lx R15: %016lx\n",
 78	       regs->r13, regs->r14, regs->r15);
 79
 80	asm("movl %%ds,%0" : "=r" (ds));
 81	asm("movl %%cs,%0" : "=r" (cs));
 82	asm("movl %%es,%0" : "=r" (es));
 83	asm("movl %%fs,%0" : "=r" (fsindex));
 84	asm("movl %%gs,%0" : "=r" (gsindex));
 85
 86	rdmsrl(MSR_FS_BASE, fs);
 87	rdmsrl(MSR_GS_BASE, gs);
 88	rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
 89
 90	if (!all)
 91		return;
 92
 93	cr0 = read_cr0();
 94	cr2 = read_cr2();
 95	cr3 = read_cr3();
 96	cr4 = read_cr4();
 97
 98	printk(KERN_DEFAULT "FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
 99	       fs, fsindex, gs, gsindex, shadowgs);
100	printk(KERN_DEFAULT "CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
101			es, cr0);
102	printk(KERN_DEFAULT "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
103			cr4);
104
105	get_debugreg(d0, 0);
106	get_debugreg(d1, 1);
107	get_debugreg(d2, 2);
108	get_debugreg(d3, 3);
109	get_debugreg(d6, 6);
110	get_debugreg(d7, 7);
111
112	/* Only print out debug registers if they are in their non-default state. */
113	if ((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
114	    (d6 == DR6_RESERVED) && (d7 == 0x400))
115		return;
116
117	printk(KERN_DEFAULT "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
118	printk(KERN_DEFAULT "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
 
119
 
 
120}
121
122void release_thread(struct task_struct *dead_task)
123{
124	if (dead_task->mm) {
125		if (dead_task->mm->context.size) {
 
126			pr_warn("WARNING: dead process %s still has LDT? <%p/%d>\n",
127				dead_task->comm,
128				dead_task->mm->context.ldt,
129				dead_task->mm->context.size);
130			BUG();
131		}
 
132	}
133}
134
135static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
136{
137	struct user_desc ud = {
138		.base_addr = addr,
139		.limit = 0xfffff,
140		.seg_32bit = 1,
141		.limit_in_pages = 1,
142		.useable = 1,
143	};
144	struct desc_struct *desc = t->thread.tls_array;
145	desc += tls;
146	fill_ldt(desc, &ud);
147}
148
149static inline u32 read_32bit_tls(struct task_struct *t, int tls)
150{
151	return get_desc_base(&t->thread.tls_array[tls]);
152}
153
154int copy_thread(unsigned long clone_flags, unsigned long sp,
155		unsigned long arg, struct task_struct *p)
156{
157	int err;
158	struct pt_regs *childregs;
 
 
159	struct task_struct *me = current;
160
161	p->thread.sp0 = (unsigned long)task_stack_page(p) + THREAD_SIZE;
162	childregs = task_pt_regs(p);
163	p->thread.sp = (unsigned long) childregs;
164	p->thread.usersp = me->thread.usersp;
165	set_tsk_thread_flag(p, TIF_FORK);
166	p->thread.fpu_counter = 0;
 
167	p->thread.io_bitmap_ptr = NULL;
168
169	savesegment(gs, p->thread.gsindex);
170	p->thread.gs = p->thread.gsindex ? 0 : me->thread.gs;
171	savesegment(fs, p->thread.fsindex);
172	p->thread.fs = p->thread.fsindex ? 0 : me->thread.fs;
173	savesegment(es, p->thread.es);
174	savesegment(ds, p->thread.ds);
175	memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
176
177	if (unlikely(p->flags & PF_KTHREAD)) {
178		/* kernel thread */
179		memset(childregs, 0, sizeof(struct pt_regs));
180		childregs->sp = (unsigned long)childregs;
181		childregs->ss = __KERNEL_DS;
182		childregs->bx = sp; /* function */
183		childregs->bp = arg;
184		childregs->orig_ax = -1;
185		childregs->cs = __KERNEL_CS | get_kernel_rpl();
186		childregs->flags = X86_EFLAGS_IF | X86_EFLAGS_FIXED;
187		return 0;
188	}
 
189	*childregs = *current_pt_regs();
190
191	childregs->ax = 0;
192	if (sp)
193		childregs->sp = sp;
194
195	err = -ENOMEM;
196	memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
197
198	if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
199		p->thread.io_bitmap_ptr = kmemdup(me->thread.io_bitmap_ptr,
200						  IO_BITMAP_BYTES, GFP_KERNEL);
201		if (!p->thread.io_bitmap_ptr) {
202			p->thread.io_bitmap_max = 0;
203			return -ENOMEM;
204		}
205		set_tsk_thread_flag(p, TIF_IO_BITMAP);
206	}
207
208	/*
209	 * Set a new TLS for the child thread?
210	 */
211	if (clone_flags & CLONE_SETTLS) {
212#ifdef CONFIG_IA32_EMULATION
213		if (test_thread_flag(TIF_IA32))
214			err = do_set_thread_area(p, -1,
215				(struct user_desc __user *)childregs->si, 0);
216		else
217#endif
218			err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
219		if (err)
220			goto out;
221	}
222	err = 0;
223out:
224	if (err && p->thread.io_bitmap_ptr) {
225		kfree(p->thread.io_bitmap_ptr);
226		p->thread.io_bitmap_max = 0;
227	}
228
229	return err;
230}
231
232static void
233start_thread_common(struct pt_regs *regs, unsigned long new_ip,
234		    unsigned long new_sp,
235		    unsigned int _cs, unsigned int _ss, unsigned int _ds)
236{
237	loadsegment(fs, 0);
238	loadsegment(es, _ds);
239	loadsegment(ds, _ds);
240	load_gs_index(0);
241	current->thread.usersp	= new_sp;
242	regs->ip		= new_ip;
243	regs->sp		= new_sp;
244	this_cpu_write(old_rsp, new_sp);
245	regs->cs		= _cs;
246	regs->ss		= _ss;
247	regs->flags		= X86_EFLAGS_IF;
 
248}
249
250void
251start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
252{
253	start_thread_common(regs, new_ip, new_sp,
254			    __USER_CS, __USER_DS, 0);
255}
256
257#ifdef CONFIG_IA32_EMULATION
258void start_thread_ia32(struct pt_regs *regs, u32 new_ip, u32 new_sp)
259{
260	start_thread_common(regs, new_ip, new_sp,
261			    test_thread_flag(TIF_X32)
262			    ? __USER_CS : __USER32_CS,
263			    __USER_DS, __USER_DS);
264}
265#endif
266
267/*
268 *	switch_to(x,y) should switch tasks from x to y.
269 *
270 * This could still be optimized:
271 * - fold all the options into a flag word and test it with a single test.
272 * - could test fs/gs bitsliced
273 *
274 * Kprobes not supported here. Set the probe on schedule instead.
275 * Function graph tracer not supported too.
276 */
277__visible __notrace_funcgraph struct task_struct *
278__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
279{
280	struct thread_struct *prev = &prev_p->thread;
281	struct thread_struct *next = &next_p->thread;
 
 
282	int cpu = smp_processor_id();
283	struct tss_struct *tss = &per_cpu(init_tss, cpu);
284	unsigned fsindex, gsindex;
285	fpu_switch_t fpu;
286
287	fpu = switch_fpu_prepare(prev_p, next_p, cpu);
 
 
 
 
 
 
 
 
288
289	/*
290	 * Reload esp0, LDT and the page table pointer:
 
291	 */
292	load_sp0(tss, next);
293
294	/*
295	 * Switch DS and ES.
296	 * This won't pick up thread selector changes, but I guess that is ok.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
297	 */
298	savesegment(es, prev->es);
299	if (unlikely(next->es | prev->es))
300		loadsegment(es, next->es);
301
302	savesegment(ds, prev->ds);
303	if (unlikely(next->ds | prev->ds))
304		loadsegment(ds, next->ds);
305
306
307	/* We must save %fs and %gs before load_TLS() because
308	 * %fs and %gs may be cleared by load_TLS().
309	 *
310	 * (e.g. xen_load_tls())
311	 */
312	savesegment(fs, fsindex);
313	savesegment(gs, gsindex);
314
315	load_TLS(next, cpu);
316
317	/*
318	 * Leave lazy mode, flushing any hypercalls made here.
319	 * This must be done before restoring TLS segments so
320	 * the GDT and LDT are properly updated, and must be
321	 * done before math_state_restore, so the TS bit is up
322	 * to date.
323	 */
324	arch_end_context_switch(next_p);
325
326	/*
327	 * Switch FS and GS.
328	 *
329	 * Segment register != 0 always requires a reload.  Also
330	 * reload when it has changed.  When prev process used 64bit
331	 * base always reload to avoid an information leak.
 
 
 
 
 
 
 
 
 
 
 
 
332	 */
333	if (unlikely(fsindex | next->fsindex | prev->fs)) {
 
334		loadsegment(fs, next->fsindex);
335		/*
336		 * Check if the user used a selector != 0; if yes
337		 *  clear 64bit base, since overloaded base is always
338		 *  mapped to the Null selector
339		 */
340		if (fsindex)
341			prev->fs = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
342	}
343	/* when next process has a 64bit base use it */
344	if (next->fs)
345		wrmsrl(MSR_FS_BASE, next->fs);
346	prev->fsindex = fsindex;
 
 
 
 
 
 
347
348	if (unlikely(gsindex | next->gsindex | prev->gs)) {
 
349		load_gs_index(next->gsindex);
350		if (gsindex)
351			prev->gs = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
352	}
353	if (next->gs)
354		wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
355	prev->gsindex = gsindex;
 
 
 
 
 
 
 
356
357	switch_fpu_finish(next_p, fpu);
358
359	/*
360	 * Switch the PDA and FPU contexts.
361	 */
362	prev->usersp = this_cpu_read(old_rsp);
363	this_cpu_write(old_rsp, next->usersp);
364	this_cpu_write(current_task, next_p);
365
366	/*
367	 * If it were not for PREEMPT_ACTIVE we could guarantee that the
368	 * preempt_count of all tasks was equal here and this would not be
369	 * needed.
370	 */
371	task_thread_info(prev_p)->saved_preempt_count = this_cpu_read(__preempt_count);
372	this_cpu_write(__preempt_count, task_thread_info(next_p)->saved_preempt_count);
373
374	this_cpu_write(kernel_stack,
375		  (unsigned long)task_stack_page(next_p) +
376		  THREAD_SIZE - KERNEL_STACK_OFFSET);
377
378	/*
379	 * Now maybe reload the debug registers and handle I/O bitmaps
380	 */
381	if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
382		     task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
383		__switch_to_xtra(prev_p, next_p, tss);
384
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
385	return prev_p;
386}
387
388void set_personality_64bit(void)
389{
390	/* inherit personality from parent */
391
392	/* Make sure to be in 64bit mode */
393	clear_thread_flag(TIF_IA32);
394	clear_thread_flag(TIF_ADDR32);
395	clear_thread_flag(TIF_X32);
396
397	/* Ensure the corresponding mm is not marked. */
398	if (current->mm)
399		current->mm->context.ia32_compat = 0;
400
401	/* TBD: overwrites user setup. Should have two bits.
402	   But 64bit processes have always behaved this way,
403	   so it's not too bad. The main problem is just that
404	   32bit childs are affected again. */
405	current->personality &= ~READ_IMPLIES_EXEC;
406}
407
408void set_personality_ia32(bool x32)
409{
410	/* inherit personality from parent */
411
412	/* Make sure to be in 32bit mode */
413	set_thread_flag(TIF_ADDR32);
414
415	/* Mark the associated mm as containing 32-bit tasks. */
416	if (current->mm)
417		current->mm->context.ia32_compat = 1;
418
419	if (x32) {
420		clear_thread_flag(TIF_IA32);
421		set_thread_flag(TIF_X32);
 
 
422		current->personality &= ~READ_IMPLIES_EXEC;
423		/* is_compat_task() uses the presence of the x32
424		   syscall bit flag to determine compat status */
425		current_thread_info()->status &= ~TS_COMPAT;
426	} else {
427		set_thread_flag(TIF_IA32);
428		clear_thread_flag(TIF_X32);
 
 
429		current->personality |= force_personality32;
430		/* Prepare the first "return" to user space */
431		current_thread_info()->status |= TS_COMPAT;
432	}
433}
434EXPORT_SYMBOL_GPL(set_personality_ia32);
435
436unsigned long get_wchan(struct task_struct *p)
 
437{
438	unsigned long stack;
439	u64 fp, ip;
440	int count = 0;
441
442	if (!p || p == current || p->state == TASK_RUNNING)
443		return 0;
444	stack = (unsigned long)task_stack_page(p);
445	if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
446		return 0;
447	fp = *(u64 *)(p->thread.sp);
448	do {
449		if (fp < (unsigned long)stack ||
450		    fp >= (unsigned long)stack+THREAD_SIZE)
451			return 0;
452		ip = *(u64 *)(fp+8);
453		if (!in_sched_functions(ip))
454			return ip;
455		fp = *(u64 *)fp;
456	} while (count++ < 16);
457	return 0;
458}
 
459
460long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
461{
462	int ret = 0;
463	int doit = task == current;
464	int cpu;
465
466	switch (code) {
467	case ARCH_SET_GS:
468		if (addr >= TASK_SIZE_OF(task))
469			return -EPERM;
470		cpu = get_cpu();
471		/* handle small bases via the GDT because that's faster to
472		   switch. */
473		if (addr <= 0xffffffff) {
474			set_32bit_tls(task, GS_TLS, addr);
475			if (doit) {
476				load_TLS(&task->thread, cpu);
477				load_gs_index(GS_TLS_SEL);
478			}
479			task->thread.gsindex = GS_TLS_SEL;
480			task->thread.gs = 0;
481		} else {
482			task->thread.gsindex = 0;
483			task->thread.gs = addr;
484			if (doit) {
485				load_gs_index(0);
486				ret = wrmsrl_safe(MSR_KERNEL_GS_BASE, addr);
487			}
488		}
489		put_cpu();
490		break;
491	case ARCH_SET_FS:
492		/* Not strictly needed for fs, but do it for symmetry
493		   with gs */
494		if (addr >= TASK_SIZE_OF(task))
495			return -EPERM;
496		cpu = get_cpu();
497		/* handle small bases via the GDT because that's faster to
498		   switch. */
499		if (addr <= 0xffffffff) {
500			set_32bit_tls(task, FS_TLS, addr);
501			if (doit) {
502				load_TLS(&task->thread, cpu);
503				loadsegment(fs, FS_TLS_SEL);
504			}
505			task->thread.fsindex = FS_TLS_SEL;
506			task->thread.fs = 0;
507		} else {
508			task->thread.fsindex = 0;
509			task->thread.fs = addr;
510			if (doit) {
511				/* set the selector to 0 to not confuse
512				   __switch_to */
513				loadsegment(fs, 0);
514				ret = wrmsrl_safe(MSR_FS_BASE, addr);
515			}
516		}
517		put_cpu();
518		break;
519	case ARCH_GET_FS: {
520		unsigned long base;
521		if (task->thread.fsindex == FS_TLS_SEL)
522			base = read_32bit_tls(task, FS_TLS);
523		else if (doit)
524			rdmsrl(MSR_FS_BASE, base);
525		else
526			base = task->thread.fs;
527		ret = put_user(base, (unsigned long __user *)addr);
528		break;
529	}
530	case ARCH_GET_GS: {
531		unsigned long base;
532		unsigned gsindex;
533		if (task->thread.gsindex == GS_TLS_SEL)
534			base = read_32bit_tls(task, GS_TLS);
535		else if (doit) {
536			savesegment(gs, gsindex);
537			if (gsindex)
538				rdmsrl(MSR_KERNEL_GS_BASE, base);
539			else
540				base = task->thread.gs;
541		} else
542			base = task->thread.gs;
543		ret = put_user(base, (unsigned long __user *)addr);
544		break;
545	}
546
 
 
 
 
 
 
 
 
 
 
 
 
 
547	default:
548		ret = -EINVAL;
549		break;
550	}
551
552	return ret;
553}
554
555long sys_arch_prctl(int code, unsigned long addr)
556{
557	return do_arch_prctl(current, code, addr);
558}
559
560unsigned long KSTK_ESP(struct task_struct *task)
561{
562	return (test_tsk_thread_flag(task, TIF_IA32)) ?
563			(task_pt_regs(task)->sp) : ((task)->thread.usersp);
564}
v4.10.11
  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/cpu.h>
 18#include <linux/errno.h>
 19#include <linux/sched.h>
 20#include <linux/fs.h>
 21#include <linux/kernel.h>
 22#include <linux/mm.h>
 23#include <linux/elfcore.h>
 24#include <linux/smp.h>
 25#include <linux/slab.h>
 26#include <linux/user.h>
 27#include <linux/interrupt.h>
 28#include <linux/delay.h>
 29#include <linux/export.h>
 30#include <linux/ptrace.h>
 31#include <linux/notifier.h>
 32#include <linux/kprobes.h>
 33#include <linux/kdebug.h>
 34#include <linux/prctl.h>
 35#include <linux/uaccess.h>
 36#include <linux/io.h>
 37#include <linux/ftrace.h>
 38
 39#include <asm/pgtable.h>
 40#include <asm/processor.h>
 41#include <asm/fpu/internal.h>
 
 42#include <asm/mmu_context.h>
 43#include <asm/prctl.h>
 44#include <asm/desc.h>
 45#include <asm/proto.h>
 46#include <asm/ia32.h>
 
 47#include <asm/syscalls.h>
 48#include <asm/debugreg.h>
 49#include <asm/switch_to.h>
 50#include <asm/xen/hypervisor.h>
 51#include <asm/vdso.h>
 52#include <asm/intel_rdt.h>
 53
 54__visible DEFINE_PER_CPU(unsigned long, rsp_scratch);
 
 
 55
 56/* Prints also some state that isn't saved in the pt_regs */
 57void __show_regs(struct pt_regs *regs, int all)
 58{
 59	unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
 60	unsigned long d0, d1, d2, d3, d6, d7;
 61	unsigned int fsindex, gsindex;
 62	unsigned int ds, cs, es;
 63
 64	printk(KERN_DEFAULT "RIP: %04lx:%pS\n", regs->cs & 0xffff,
 65		(void *)regs->ip);
 66	printk(KERN_DEFAULT "RSP: %04lx:%016lx EFLAGS: %08lx", regs->ss,
 67		regs->sp, regs->flags);
 68	if (regs->orig_ax != -1)
 69		pr_cont(" ORIG_RAX: %016lx\n", regs->orig_ax);
 70	else
 71		pr_cont("\n");
 72
 73	printk(KERN_DEFAULT "RAX: %016lx RBX: %016lx RCX: %016lx\n",
 74	       regs->ax, regs->bx, regs->cx);
 75	printk(KERN_DEFAULT "RDX: %016lx RSI: %016lx RDI: %016lx\n",
 76	       regs->dx, regs->si, regs->di);
 77	printk(KERN_DEFAULT "RBP: %016lx R08: %016lx R09: %016lx\n",
 78	       regs->bp, regs->r8, regs->r9);
 79	printk(KERN_DEFAULT "R10: %016lx R11: %016lx R12: %016lx\n",
 80	       regs->r10, regs->r11, regs->r12);
 81	printk(KERN_DEFAULT "R13: %016lx R14: %016lx R15: %016lx\n",
 82	       regs->r13, regs->r14, regs->r15);
 83
 84	asm("movl %%ds,%0" : "=r" (ds));
 85	asm("movl %%cs,%0" : "=r" (cs));
 86	asm("movl %%es,%0" : "=r" (es));
 87	asm("movl %%fs,%0" : "=r" (fsindex));
 88	asm("movl %%gs,%0" : "=r" (gsindex));
 89
 90	rdmsrl(MSR_FS_BASE, fs);
 91	rdmsrl(MSR_GS_BASE, gs);
 92	rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
 93
 94	if (!all)
 95		return;
 96
 97	cr0 = read_cr0();
 98	cr2 = read_cr2();
 99	cr3 = read_cr3();
100	cr4 = __read_cr4();
101
102	printk(KERN_DEFAULT "FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
103	       fs, fsindex, gs, gsindex, shadowgs);
104	printk(KERN_DEFAULT "CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
105			es, cr0);
106	printk(KERN_DEFAULT "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
107			cr4);
108
109	get_debugreg(d0, 0);
110	get_debugreg(d1, 1);
111	get_debugreg(d2, 2);
112	get_debugreg(d3, 3);
113	get_debugreg(d6, 6);
114	get_debugreg(d7, 7);
115
116	/* Only print out debug registers if they are in their non-default state. */
117	if (!((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
118	    (d6 == DR6_RESERVED) && (d7 == 0x400))) {
119		printk(KERN_DEFAULT "DR0: %016lx DR1: %016lx DR2: %016lx\n",
120		       d0, d1, d2);
121		printk(KERN_DEFAULT "DR3: %016lx DR6: %016lx DR7: %016lx\n",
122		       d3, d6, d7);
123	}
124
125	if (boot_cpu_has(X86_FEATURE_OSPKE))
126		printk(KERN_DEFAULT "PKRU: %08x\n", read_pkru());
127}
128
129void release_thread(struct task_struct *dead_task)
130{
131	if (dead_task->mm) {
132#ifdef CONFIG_MODIFY_LDT_SYSCALL
133		if (dead_task->mm->context.ldt) {
134			pr_warn("WARNING: dead process %s still has LDT? <%p/%d>\n",
135				dead_task->comm,
136				dead_task->mm->context.ldt->entries,
137				dead_task->mm->context.ldt->size);
138			BUG();
139		}
140#endif
141	}
142}
143
144int copy_thread_tls(unsigned long clone_flags, unsigned long sp,
145		unsigned long arg, struct task_struct *p, unsigned long tls)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
146{
147	int err;
148	struct pt_regs *childregs;
149	struct fork_frame *fork_frame;
150	struct inactive_task_frame *frame;
151	struct task_struct *me = current;
152
153	p->thread.sp0 = (unsigned long)task_stack_page(p) + THREAD_SIZE;
154	childregs = task_pt_regs(p);
155	fork_frame = container_of(childregs, struct fork_frame, regs);
156	frame = &fork_frame->frame;
157	frame->bp = 0;
158	frame->ret_addr = (unsigned long) ret_from_fork;
159	p->thread.sp = (unsigned long) fork_frame;
160	p->thread.io_bitmap_ptr = NULL;
161
162	savesegment(gs, p->thread.gsindex);
163	p->thread.gsbase = p->thread.gsindex ? 0 : me->thread.gsbase;
164	savesegment(fs, p->thread.fsindex);
165	p->thread.fsbase = p->thread.fsindex ? 0 : me->thread.fsbase;
166	savesegment(es, p->thread.es);
167	savesegment(ds, p->thread.ds);
168	memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
169
170	if (unlikely(p->flags & PF_KTHREAD)) {
171		/* kernel thread */
172		memset(childregs, 0, sizeof(struct pt_regs));
173		frame->bx = sp;		/* function */
174		frame->r12 = arg;
 
 
 
 
 
175		return 0;
176	}
177	frame->bx = 0;
178	*childregs = *current_pt_regs();
179
180	childregs->ax = 0;
181	if (sp)
182		childregs->sp = sp;
183
184	err = -ENOMEM;
 
 
185	if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
186		p->thread.io_bitmap_ptr = kmemdup(me->thread.io_bitmap_ptr,
187						  IO_BITMAP_BYTES, GFP_KERNEL);
188		if (!p->thread.io_bitmap_ptr) {
189			p->thread.io_bitmap_max = 0;
190			return -ENOMEM;
191		}
192		set_tsk_thread_flag(p, TIF_IO_BITMAP);
193	}
194
195	/*
196	 * Set a new TLS for the child thread?
197	 */
198	if (clone_flags & CLONE_SETTLS) {
199#ifdef CONFIG_IA32_EMULATION
200		if (in_ia32_syscall())
201			err = do_set_thread_area(p, -1,
202				(struct user_desc __user *)tls, 0);
203		else
204#endif
205			err = do_arch_prctl(p, ARCH_SET_FS, tls);
206		if (err)
207			goto out;
208	}
209	err = 0;
210out:
211	if (err && p->thread.io_bitmap_ptr) {
212		kfree(p->thread.io_bitmap_ptr);
213		p->thread.io_bitmap_max = 0;
214	}
215
216	return err;
217}
218
219static void
220start_thread_common(struct pt_regs *regs, unsigned long new_ip,
221		    unsigned long new_sp,
222		    unsigned int _cs, unsigned int _ss, unsigned int _ds)
223{
224	loadsegment(fs, 0);
225	loadsegment(es, _ds);
226	loadsegment(ds, _ds);
227	load_gs_index(0);
 
228	regs->ip		= new_ip;
229	regs->sp		= new_sp;
 
230	regs->cs		= _cs;
231	regs->ss		= _ss;
232	regs->flags		= X86_EFLAGS_IF;
233	force_iret();
234}
235
236void
237start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
238{
239	start_thread_common(regs, new_ip, new_sp,
240			    __USER_CS, __USER_DS, 0);
241}
242
243#ifdef CONFIG_COMPAT
244void compat_start_thread(struct pt_regs *regs, u32 new_ip, u32 new_sp)
245{
246	start_thread_common(regs, new_ip, new_sp,
247			    test_thread_flag(TIF_X32)
248			    ? __USER_CS : __USER32_CS,
249			    __USER_DS, __USER_DS);
250}
251#endif
252
253/*
254 *	switch_to(x,y) should switch tasks from x to y.
255 *
256 * This could still be optimized:
257 * - fold all the options into a flag word and test it with a single test.
258 * - could test fs/gs bitsliced
259 *
260 * Kprobes not supported here. Set the probe on schedule instead.
261 * Function graph tracer not supported too.
262 */
263__visible __notrace_funcgraph struct task_struct *
264__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
265{
266	struct thread_struct *prev = &prev_p->thread;
267	struct thread_struct *next = &next_p->thread;
268	struct fpu *prev_fpu = &prev->fpu;
269	struct fpu *next_fpu = &next->fpu;
270	int cpu = smp_processor_id();
271	struct tss_struct *tss = &per_cpu(cpu_tss, cpu);
272	unsigned prev_fsindex, prev_gsindex;
 
273
274	switch_fpu_prepare(prev_fpu, cpu);
275
276	/* We must save %fs and %gs before load_TLS() because
277	 * %fs and %gs may be cleared by load_TLS().
278	 *
279	 * (e.g. xen_load_tls())
280	 */
281	savesegment(fs, prev_fsindex);
282	savesegment(gs, prev_gsindex);
283
284	/*
285	 * Load TLS before restoring any segments so that segment loads
286	 * reference the correct GDT entries.
287	 */
288	load_TLS(next, cpu);
289
290	/*
291	 * Leave lazy mode, flushing any hypercalls made here.  This
292	 * must be done after loading TLS entries in the GDT but before
293	 * loading segments that might reference them, and and it must
294	 * be done before fpu__restore(), so the TS bit is up to
295	 * date.
296	 */
297	arch_end_context_switch(next_p);
298
299	/* Switch DS and ES.
300	 *
301	 * Reading them only returns the selectors, but writing them (if
302	 * nonzero) loads the full descriptor from the GDT or LDT.  The
303	 * LDT for next is loaded in switch_mm, and the GDT is loaded
304	 * above.
305	 *
306	 * We therefore need to write new values to the segment
307	 * registers on every context switch unless both the new and old
308	 * values are zero.
309	 *
310	 * Note that we don't need to do anything for CS and SS, as
311	 * those are saved and restored as part of pt_regs.
312	 */
313	savesegment(es, prev->es);
314	if (unlikely(next->es | prev->es))
315		loadsegment(es, next->es);
316
317	savesegment(ds, prev->ds);
318	if (unlikely(next->ds | prev->ds))
319		loadsegment(ds, next->ds);
320
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
321	/*
322	 * Switch FS and GS.
323	 *
324	 * These are even more complicated than DS and ES: they have
325	 * 64-bit bases are that controlled by arch_prctl.  The bases
326	 * don't necessarily match the selectors, as user code can do
327	 * any number of things to cause them to be inconsistent.
328	 *
329	 * We don't promise to preserve the bases if the selectors are
330	 * nonzero.  We also don't promise to preserve the base if the
331	 * selector is zero and the base doesn't match whatever was
332	 * most recently passed to ARCH_SET_FS/GS.  (If/when the
333	 * FSGSBASE instructions are enabled, we'll need to offer
334	 * stronger guarantees.)
335	 *
336	 * As an invariant,
337	 * (fsbase != 0 && fsindex != 0) || (gsbase != 0 && gsindex != 0) is
338	 * impossible.
339	 */
340	if (next->fsindex) {
341		/* Loading a nonzero value into FS sets the index and base. */
342		loadsegment(fs, next->fsindex);
343	} else {
344		if (next->fsbase) {
345			/* Next index is zero but next base is nonzero. */
346			if (prev_fsindex)
347				loadsegment(fs, 0);
348			wrmsrl(MSR_FS_BASE, next->fsbase);
349		} else {
350			/* Next base and index are both zero. */
351			if (static_cpu_has_bug(X86_BUG_NULL_SEG)) {
352				/*
353				 * We don't know the previous base and can't
354				 * find out without RDMSR.  Forcibly clear it.
355				 */
356				loadsegment(fs, __USER_DS);
357				loadsegment(fs, 0);
358			} else {
359				/*
360				 * If the previous index is zero and ARCH_SET_FS
361				 * didn't change the base, then the base is
362				 * also zero and we don't need to do anything.
363				 */
364				if (prev->fsbase || prev_fsindex)
365					loadsegment(fs, 0);
366			}
367		}
368	}
369	/*
370	 * Save the old state and preserve the invariant.
371	 * NB: if prev_fsindex == 0, then we can't reliably learn the base
372	 * without RDMSR because Intel user code can zero it without telling
373	 * us and AMD user code can program any 32-bit value without telling
374	 * us.
375	 */
376	if (prev_fsindex)
377		prev->fsbase = 0;
378	prev->fsindex = prev_fsindex;
379
380	if (next->gsindex) {
381		/* Loading a nonzero value into GS sets the index and base. */
382		load_gs_index(next->gsindex);
383	} else {
384		if (next->gsbase) {
385			/* Next index is zero but next base is nonzero. */
386			if (prev_gsindex)
387				load_gs_index(0);
388			wrmsrl(MSR_KERNEL_GS_BASE, next->gsbase);
389		} else {
390			/* Next base and index are both zero. */
391			if (static_cpu_has_bug(X86_BUG_NULL_SEG)) {
392				/*
393				 * We don't know the previous base and can't
394				 * find out without RDMSR.  Forcibly clear it.
395				 *
396				 * This contains a pointless SWAPGS pair.
397				 * Fixing it would involve an explicit check
398				 * for Xen or a new pvop.
399				 */
400				load_gs_index(__USER_DS);
401				load_gs_index(0);
402			} else {
403				/*
404				 * If the previous index is zero and ARCH_SET_GS
405				 * didn't change the base, then the base is
406				 * also zero and we don't need to do anything.
407				 */
408				if (prev->gsbase || prev_gsindex)
409					load_gs_index(0);
410			}
411		}
412	}
413	/*
414	 * Save the old state and preserve the invariant.
415	 * NB: if prev_gsindex == 0, then we can't reliably learn the base
416	 * without RDMSR because Intel user code can zero it without telling
417	 * us and AMD user code can program any 32-bit value without telling
418	 * us.
419	 */
420	if (prev_gsindex)
421		prev->gsbase = 0;
422	prev->gsindex = prev_gsindex;
423
424	switch_fpu_finish(next_fpu, cpu);
425
426	/*
427	 * Switch the PDA and FPU contexts.
428	 */
 
 
429	this_cpu_write(current_task, next_p);
430
431	/* Reload esp0 and ss1.  This changes current_thread_info(). */
432	load_sp0(tss, next);
 
 
 
 
 
 
 
 
 
433
434	/*
435	 * Now maybe reload the debug registers and handle I/O bitmaps
436	 */
437	if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
438		     task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
439		__switch_to_xtra(prev_p, next_p, tss);
440
441#ifdef CONFIG_XEN
442	/*
443	 * On Xen PV, IOPL bits in pt_regs->flags have no effect, and
444	 * current_pt_regs()->flags may not match the current task's
445	 * intended IOPL.  We need to switch it manually.
446	 */
447	if (unlikely(static_cpu_has(X86_FEATURE_XENPV) &&
448		     prev->iopl != next->iopl))
449		xen_set_iopl_mask(next->iopl);
450#endif
451
452	if (static_cpu_has_bug(X86_BUG_SYSRET_SS_ATTRS)) {
453		/*
454		 * AMD CPUs have a misfeature: SYSRET sets the SS selector but
455		 * does not update the cached descriptor.  As a result, if we
456		 * do SYSRET while SS is NULL, we'll end up in user mode with
457		 * SS apparently equal to __USER_DS but actually unusable.
458		 *
459		 * The straightforward workaround would be to fix it up just
460		 * before SYSRET, but that would slow down the system call
461		 * fast paths.  Instead, we ensure that SS is never NULL in
462		 * system call context.  We do this by replacing NULL SS
463		 * selectors at every context switch.  SYSCALL sets up a valid
464		 * SS, so the only way to get NULL is to re-enter the kernel
465		 * from CPL 3 through an interrupt.  Since that can't happen
466		 * in the same task as a running syscall, we are guaranteed to
467		 * context switch between every interrupt vector entry and a
468		 * subsequent SYSRET.
469		 *
470		 * We read SS first because SS reads are much faster than
471		 * writes.  Out of caution, we force SS to __KERNEL_DS even if
472		 * it previously had a different non-NULL value.
473		 */
474		unsigned short ss_sel;
475		savesegment(ss, ss_sel);
476		if (ss_sel != __KERNEL_DS)
477			loadsegment(ss, __KERNEL_DS);
478	}
479
480	/* Load the Intel cache allocation PQR MSR. */
481	intel_rdt_sched_in();
482
483	return prev_p;
484}
485
486void set_personality_64bit(void)
487{
488	/* inherit personality from parent */
489
490	/* Make sure to be in 64bit mode */
491	clear_thread_flag(TIF_IA32);
492	clear_thread_flag(TIF_ADDR32);
493	clear_thread_flag(TIF_X32);
494
495	/* Ensure the corresponding mm is not marked. */
496	if (current->mm)
497		current->mm->context.ia32_compat = 0;
498
499	/* TBD: overwrites user setup. Should have two bits.
500	   But 64bit processes have always behaved this way,
501	   so it's not too bad. The main problem is just that
502	   32bit childs are affected again. */
503	current->personality &= ~READ_IMPLIES_EXEC;
504}
505
506void set_personality_ia32(bool x32)
507{
508	/* inherit personality from parent */
509
510	/* Make sure to be in 32bit mode */
511	set_thread_flag(TIF_ADDR32);
512
513	/* Mark the associated mm as containing 32-bit tasks. */
 
 
 
514	if (x32) {
515		clear_thread_flag(TIF_IA32);
516		set_thread_flag(TIF_X32);
517		if (current->mm)
518			current->mm->context.ia32_compat = TIF_X32;
519		current->personality &= ~READ_IMPLIES_EXEC;
520		/* in_compat_syscall() uses the presence of the x32
521		   syscall bit flag to determine compat status */
522		current->thread.status &= ~TS_COMPAT;
523	} else {
524		set_thread_flag(TIF_IA32);
525		clear_thread_flag(TIF_X32);
526		if (current->mm)
527			current->mm->context.ia32_compat = TIF_IA32;
528		current->personality |= force_personality32;
529		/* Prepare the first "return" to user space */
530		current->thread.status |= TS_COMPAT;
531	}
532}
533EXPORT_SYMBOL_GPL(set_personality_ia32);
534
535#ifdef CONFIG_CHECKPOINT_RESTORE
536static long prctl_map_vdso(const struct vdso_image *image, unsigned long addr)
537{
538	int ret;
 
 
539
540	ret = map_vdso_once(image, addr);
541	if (ret)
542		return ret;
543
544	return (long)image->size;
 
 
 
 
 
 
 
 
 
 
 
545}
546#endif
547
548long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
549{
550	int ret = 0;
551	int doit = task == current;
552	int cpu;
553
554	switch (code) {
555	case ARCH_SET_GS:
556		if (addr >= TASK_SIZE_MAX)
557			return -EPERM;
558		cpu = get_cpu();
559		task->thread.gsindex = 0;
560		task->thread.gsbase = addr;
561		if (doit) {
562			load_gs_index(0);
563			ret = wrmsrl_safe(MSR_KERNEL_GS_BASE, addr);
 
 
 
 
 
 
 
 
 
 
 
 
564		}
565		put_cpu();
566		break;
567	case ARCH_SET_FS:
568		/* Not strictly needed for fs, but do it for symmetry
569		   with gs */
570		if (addr >= TASK_SIZE_MAX)
571			return -EPERM;
572		cpu = get_cpu();
573		task->thread.fsindex = 0;
574		task->thread.fsbase = addr;
575		if (doit) {
576			/* set the selector to 0 to not confuse __switch_to */
577			loadsegment(fs, 0);
578			ret = wrmsrl_safe(MSR_FS_BASE, addr);
 
 
 
 
 
 
 
 
 
 
 
 
 
579		}
580		put_cpu();
581		break;
582	case ARCH_GET_FS: {
583		unsigned long base;
584		if (doit)
 
 
585			rdmsrl(MSR_FS_BASE, base);
586		else
587			base = task->thread.fsbase;
588		ret = put_user(base, (unsigned long __user *)addr);
589		break;
590	}
591	case ARCH_GET_GS: {
592		unsigned long base;
593		if (doit)
594			rdmsrl(MSR_KERNEL_GS_BASE, base);
595		else
596			base = task->thread.gsbase;
 
 
 
 
 
 
 
597		ret = put_user(base, (unsigned long __user *)addr);
598		break;
599	}
600
601#ifdef CONFIG_CHECKPOINT_RESTORE
602# ifdef CONFIG_X86_X32_ABI
603	case ARCH_MAP_VDSO_X32:
604		return prctl_map_vdso(&vdso_image_x32, addr);
605# endif
606# if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
607	case ARCH_MAP_VDSO_32:
608		return prctl_map_vdso(&vdso_image_32, addr);
609# endif
610	case ARCH_MAP_VDSO_64:
611		return prctl_map_vdso(&vdso_image_64, addr);
612#endif
613
614	default:
615		ret = -EINVAL;
616		break;
617	}
618
619	return ret;
620}
621
622long sys_arch_prctl(int code, unsigned long addr)
623{
624	return do_arch_prctl(current, code, addr);
625}
626
627unsigned long KSTK_ESP(struct task_struct *task)
628{
629	return task_pt_regs(task)->sp;
 
630}