• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 #include <asm/xen/hypervisor.h>
53 
54 asmlinkage extern void ret_from_fork(void);
55 
56 __visible DEFINE_PER_CPU(unsigned long, old_rsp);
57 
58 /* Prints also some state that isn't saved in the pt_regs */
__show_regs(struct pt_regs * regs,int all)59 void __show_regs(struct pt_regs *regs, int all)
60 {
61 	unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
62 	unsigned long d0, d1, d2, d3, d6, d7;
63 	unsigned int fsindex, gsindex;
64 	unsigned int ds, cs, es;
65 
66 	printk(KERN_DEFAULT "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
67 	printk_address(regs->ip);
68 	printk(KERN_DEFAULT "RSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss,
69 			regs->sp, regs->flags);
70 	printk(KERN_DEFAULT "RAX: %016lx RBX: %016lx RCX: %016lx\n",
71 	       regs->ax, regs->bx, regs->cx);
72 	printk(KERN_DEFAULT "RDX: %016lx RSI: %016lx RDI: %016lx\n",
73 	       regs->dx, regs->si, regs->di);
74 	printk(KERN_DEFAULT "RBP: %016lx R08: %016lx R09: %016lx\n",
75 	       regs->bp, regs->r8, regs->r9);
76 	printk(KERN_DEFAULT "R10: %016lx R11: %016lx R12: %016lx\n",
77 	       regs->r10, regs->r11, regs->r12);
78 	printk(KERN_DEFAULT "R13: %016lx R14: %016lx R15: %016lx\n",
79 	       regs->r13, regs->r14, regs->r15);
80 
81 	asm("movl %%ds,%0" : "=r" (ds));
82 	asm("movl %%cs,%0" : "=r" (cs));
83 	asm("movl %%es,%0" : "=r" (es));
84 	asm("movl %%fs,%0" : "=r" (fsindex));
85 	asm("movl %%gs,%0" : "=r" (gsindex));
86 
87 	rdmsrl(MSR_FS_BASE, fs);
88 	rdmsrl(MSR_GS_BASE, gs);
89 	rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
90 
91 	if (!all)
92 		return;
93 
94 	cr0 = read_cr0();
95 	cr2 = read_cr2();
96 	cr3 = read_cr3();
97 	cr4 = __read_cr4();
98 
99 	printk(KERN_DEFAULT "FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
100 	       fs, fsindex, gs, gsindex, shadowgs);
101 	printk(KERN_DEFAULT "CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
102 			es, cr0);
103 	printk(KERN_DEFAULT "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
104 			cr4);
105 
106 	get_debugreg(d0, 0);
107 	get_debugreg(d1, 1);
108 	get_debugreg(d2, 2);
109 	get_debugreg(d3, 3);
110 	get_debugreg(d6, 6);
111 	get_debugreg(d7, 7);
112 
113 	/* Only print out debug registers if they are in their non-default state. */
114 	if ((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
115 	    (d6 == DR6_RESERVED) && (d7 == 0x400))
116 		return;
117 
118 	printk(KERN_DEFAULT "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
119 	printk(KERN_DEFAULT "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
120 
121 }
122 
release_thread(struct task_struct * dead_task)123 void release_thread(struct task_struct *dead_task)
124 {
125 	if (dead_task->mm) {
126 		if (dead_task->mm->context.ldt) {
127 			pr_warn("WARNING: dead process %s still has LDT? <%p/%d>\n",
128 				dead_task->comm,
129 				dead_task->mm->context.ldt,
130 				dead_task->mm->context.ldt->size);
131 			BUG();
132 		}
133 	}
134 }
135 
set_32bit_tls(struct task_struct * t,int tls,u32 addr)136 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
137 {
138 	struct user_desc ud = {
139 		.base_addr = addr,
140 		.limit = 0xfffff,
141 		.seg_32bit = 1,
142 		.limit_in_pages = 1,
143 		.useable = 1,
144 	};
145 	struct desc_struct *desc = t->thread.tls_array;
146 	desc += tls;
147 	fill_ldt(desc, &ud);
148 }
149 
read_32bit_tls(struct task_struct * t,int tls)150 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
151 {
152 	return get_desc_base(&t->thread.tls_array[tls]);
153 }
154 
copy_thread(unsigned long clone_flags,unsigned long sp,unsigned long arg,struct task_struct * p)155 int copy_thread(unsigned long clone_flags, unsigned long sp,
156 		unsigned long arg, struct task_struct *p)
157 {
158 	int err;
159 	struct pt_regs *childregs;
160 	struct task_struct *me = current;
161 
162 	p->thread.sp0 = (unsigned long)task_stack_page(p) + THREAD_SIZE;
163 	childregs = task_pt_regs(p);
164 	p->thread.sp = (unsigned long) childregs;
165 	p->thread.usersp = me->thread.usersp;
166 	set_tsk_thread_flag(p, TIF_FORK);
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 	if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
197 		p->thread.io_bitmap_ptr = kmemdup(me->thread.io_bitmap_ptr,
198 						  IO_BITMAP_BYTES, GFP_KERNEL);
199 		if (!p->thread.io_bitmap_ptr) {
200 			p->thread.io_bitmap_max = 0;
201 			return -ENOMEM;
202 		}
203 		set_tsk_thread_flag(p, TIF_IO_BITMAP);
204 	}
205 
206 	/*
207 	 * Set a new TLS for the child thread?
208 	 */
209 	if (clone_flags & CLONE_SETTLS) {
210 #ifdef CONFIG_IA32_EMULATION
211 		if (test_thread_flag(TIF_IA32))
212 			err = do_set_thread_area(p, -1,
213 				(struct user_desc __user *)childregs->si, 0);
214 		else
215 #endif
216 			err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
217 		if (err)
218 			goto out;
219 	}
220 	err = 0;
221 out:
222 	if (err && p->thread.io_bitmap_ptr) {
223 		kfree(p->thread.io_bitmap_ptr);
224 		p->thread.io_bitmap_max = 0;
225 	}
226 
227 	return err;
228 }
229 
230 static void
start_thread_common(struct pt_regs * regs,unsigned long new_ip,unsigned long new_sp,unsigned int _cs,unsigned int _ss,unsigned int _ds)231 start_thread_common(struct pt_regs *regs, unsigned long new_ip,
232 		    unsigned long new_sp,
233 		    unsigned int _cs, unsigned int _ss, unsigned int _ds)
234 {
235 	loadsegment(fs, 0);
236 	loadsegment(es, _ds);
237 	loadsegment(ds, _ds);
238 	load_gs_index(0);
239 	current->thread.usersp	= new_sp;
240 	regs->ip		= new_ip;
241 	regs->sp		= new_sp;
242 	this_cpu_write(old_rsp, new_sp);
243 	regs->cs		= _cs;
244 	regs->ss		= _ss;
245 	regs->flags		= X86_EFLAGS_IF;
246 }
247 
248 void
start_thread(struct pt_regs * regs,unsigned long new_ip,unsigned long new_sp)249 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
250 {
251 	start_thread_common(regs, new_ip, new_sp,
252 			    __USER_CS, __USER_DS, 0);
253 }
254 
255 #ifdef CONFIG_IA32_EMULATION
start_thread_ia32(struct pt_regs * regs,u32 new_ip,u32 new_sp)256 void start_thread_ia32(struct pt_regs *regs, u32 new_ip, u32 new_sp)
257 {
258 	start_thread_common(regs, new_ip, new_sp,
259 			    test_thread_flag(TIF_X32)
260 			    ? __USER_CS : __USER32_CS,
261 			    __USER_DS, __USER_DS);
262 }
263 #endif
264 
265 /*
266  *	switch_to(x,y) should switch tasks from x to y.
267  *
268  * This could still be optimized:
269  * - fold all the options into a flag word and test it with a single test.
270  * - could test fs/gs bitsliced
271  *
272  * Kprobes not supported here. Set the probe on schedule instead.
273  * Function graph tracer not supported too.
274  */
275 __visible __notrace_funcgraph struct task_struct *
__switch_to(struct task_struct * prev_p,struct task_struct * next_p)276 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
277 {
278 	struct thread_struct *prev = &prev_p->thread;
279 	struct thread_struct *next = &next_p->thread;
280 	int cpu = smp_processor_id();
281 	struct tss_struct *tss = &per_cpu(init_tss, cpu);
282 	unsigned fsindex, gsindex;
283 	fpu_switch_t fpu;
284 
285 	fpu = switch_fpu_prepare(prev_p, next_p, cpu);
286 
287 	/* Reload esp0 and ss1. */
288 	load_sp0(tss, next);
289 
290 	/* We must save %fs and %gs before load_TLS() because
291 	 * %fs and %gs may be cleared by load_TLS().
292 	 *
293 	 * (e.g. xen_load_tls())
294 	 */
295 	savesegment(fs, fsindex);
296 	savesegment(gs, gsindex);
297 
298 	/*
299 	 * Load TLS before restoring any segments so that segment loads
300 	 * reference the correct GDT entries.
301 	 */
302 	load_TLS(next, cpu);
303 
304 	/*
305 	 * Leave lazy mode, flushing any hypercalls made here.  This
306 	 * must be done after loading TLS entries in the GDT but before
307 	 * loading segments that might reference them, and and it must
308 	 * be done before math_state_restore, so the TS bit is up to
309 	 * date.
310 	 */
311 	arch_end_context_switch(next_p);
312 
313 	/* Switch DS and ES.
314 	 *
315 	 * Reading them only returns the selectors, but writing them (if
316 	 * nonzero) loads the full descriptor from the GDT or LDT.  The
317 	 * LDT for next is loaded in switch_mm, and the GDT is loaded
318 	 * above.
319 	 *
320 	 * We therefore need to write new values to the segment
321 	 * registers on every context switch unless both the new and old
322 	 * values are zero.
323 	 *
324 	 * Note that we don't need to do anything for CS and SS, as
325 	 * those are saved and restored as part of pt_regs.
326 	 */
327 	savesegment(es, prev->es);
328 	if (unlikely(next->es | prev->es))
329 		loadsegment(es, next->es);
330 
331 	savesegment(ds, prev->ds);
332 	if (unlikely(next->ds | prev->ds))
333 		loadsegment(ds, next->ds);
334 
335 	/*
336 	 * Switch FS and GS.
337 	 *
338 	 * These are even more complicated than FS and GS: they have
339 	 * 64-bit bases are that controlled by arch_prctl.  Those bases
340 	 * only differ from the values in the GDT or LDT if the selector
341 	 * is 0.
342 	 *
343 	 * Loading the segment register resets the hidden base part of
344 	 * the register to 0 or the value from the GDT / LDT.  If the
345 	 * next base address zero, writing 0 to the segment register is
346 	 * much faster than using wrmsr to explicitly zero the base.
347 	 *
348 	 * The thread_struct.fs and thread_struct.gs values are 0
349 	 * if the fs and gs bases respectively are not overridden
350 	 * from the values implied by fsindex and gsindex.  They
351 	 * are nonzero, and store the nonzero base addresses, if
352 	 * the bases are overridden.
353 	 *
354 	 * (fs != 0 && fsindex != 0) || (gs != 0 && gsindex != 0) should
355 	 * be impossible.
356 	 *
357 	 * Therefore we need to reload the segment registers if either
358 	 * the old or new selector is nonzero, and we need to override
359 	 * the base address if next thread expects it to be overridden.
360 	 *
361 	 * This code is unnecessarily slow in the case where the old and
362 	 * new indexes are zero and the new base is nonzero -- it will
363 	 * unnecessarily write 0 to the selector before writing the new
364 	 * base address.
365 	 *
366 	 * Note: This all depends on arch_prctl being the only way that
367 	 * user code can override the segment base.  Once wrfsbase and
368 	 * wrgsbase are enabled, most of this code will need to change.
369 	 */
370 	if (unlikely(fsindex | next->fsindex | prev->fs)) {
371 		loadsegment(fs, next->fsindex);
372 
373 		/*
374 		 * If user code wrote a nonzero value to FS, then it also
375 		 * cleared the overridden base address.
376 		 *
377 		 * XXX: if user code wrote 0 to FS and cleared the base
378 		 * address itself, we won't notice and we'll incorrectly
379 		 * restore the prior base address next time we reschdule
380 		 * the process.
381 		 */
382 		if (fsindex)
383 			prev->fs = 0;
384 	}
385 	if (next->fs)
386 		wrmsrl(MSR_FS_BASE, next->fs);
387 	prev->fsindex = fsindex;
388 
389 	if (unlikely(gsindex | next->gsindex | prev->gs)) {
390 		load_gs_index(next->gsindex);
391 
392 		/* This works (and fails) the same way as fsindex above. */
393 		if (gsindex)
394 			prev->gs = 0;
395 	}
396 	if (next->gs)
397 		wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
398 	prev->gsindex = gsindex;
399 
400 	switch_fpu_finish(next_p, fpu);
401 
402 	/*
403 	 * Switch the PDA and FPU contexts.
404 	 */
405 	prev->usersp = this_cpu_read(old_rsp);
406 	this_cpu_write(old_rsp, next->usersp);
407 	this_cpu_write(current_task, next_p);
408 
409 	/*
410 	 * If it were not for PREEMPT_ACTIVE we could guarantee that the
411 	 * preempt_count of all tasks was equal here and this would not be
412 	 * needed.
413 	 */
414 	task_thread_info(prev_p)->saved_preempt_count = this_cpu_read(__preempt_count);
415 	this_cpu_write(__preempt_count, task_thread_info(next_p)->saved_preempt_count);
416 
417 	this_cpu_write(kernel_stack,
418 		  (unsigned long)task_stack_page(next_p) +
419 		  THREAD_SIZE - KERNEL_STACK_OFFSET);
420 
421 	/*
422 	 * Now maybe reload the debug registers and handle I/O bitmaps
423 	 */
424 	if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
425 		     task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
426 		__switch_to_xtra(prev_p, next_p, tss);
427 
428 #ifdef CONFIG_XEN
429 	/*
430 	 * On Xen PV, IOPL bits in pt_regs->flags have no effect, and
431 	 * current_pt_regs()->flags may not match the current task's
432 	 * intended IOPL.  We need to switch it manually.
433 	 */
434 	if (unlikely(xen_pv_domain() &&
435 		     prev->iopl != next->iopl))
436 		xen_set_iopl_mask(next->iopl);
437 #endif
438 
439 	return prev_p;
440 }
441 
set_personality_64bit(void)442 void set_personality_64bit(void)
443 {
444 	/* inherit personality from parent */
445 
446 	/* Make sure to be in 64bit mode */
447 	clear_thread_flag(TIF_IA32);
448 	clear_thread_flag(TIF_ADDR32);
449 	clear_thread_flag(TIF_X32);
450 
451 	/* Ensure the corresponding mm is not marked. */
452 	if (current->mm)
453 		current->mm->context.ia32_compat = 0;
454 
455 	/* TBD: overwrites user setup. Should have two bits.
456 	   But 64bit processes have always behaved this way,
457 	   so it's not too bad. The main problem is just that
458 	   32bit childs are affected again. */
459 	current->personality &= ~READ_IMPLIES_EXEC;
460 }
461 
set_personality_ia32(bool x32)462 void set_personality_ia32(bool x32)
463 {
464 	/* inherit personality from parent */
465 
466 	/* Make sure to be in 32bit mode */
467 	set_thread_flag(TIF_ADDR32);
468 
469 	/* Mark the associated mm as containing 32-bit tasks. */
470 	if (x32) {
471 		clear_thread_flag(TIF_IA32);
472 		set_thread_flag(TIF_X32);
473 		if (current->mm)
474 			current->mm->context.ia32_compat = TIF_X32;
475 		current->personality &= ~READ_IMPLIES_EXEC;
476 		/* is_compat_task() uses the presence of the x32
477 		   syscall bit flag to determine compat status */
478 		current_thread_info()->status &= ~TS_COMPAT;
479 	} else {
480 		set_thread_flag(TIF_IA32);
481 		clear_thread_flag(TIF_X32);
482 		if (current->mm)
483 			current->mm->context.ia32_compat = TIF_IA32;
484 		current->personality |= force_personality32;
485 		/* Prepare the first "return" to user space */
486 		current_thread_info()->status |= TS_COMPAT;
487 	}
488 }
489 EXPORT_SYMBOL_GPL(set_personality_ia32);
490 
491 /*
492  * Called from fs/proc with a reference on @p to find the function
493  * which called into schedule(). This needs to be done carefully
494  * because the task might wake up and we might look at a stack
495  * changing under us.
496  */
get_wchan(struct task_struct * p)497 unsigned long get_wchan(struct task_struct *p)
498 {
499 	unsigned long start, bottom, top, sp, fp, ip;
500 	int count = 0;
501 
502 	if (!p || p == current || p->state == TASK_RUNNING)
503 		return 0;
504 
505 	start = (unsigned long)task_stack_page(p);
506 	if (!start)
507 		return 0;
508 
509 	/*
510 	 * Layout of the stack page:
511 	 *
512 	 * ----------- topmax = start + THREAD_SIZE - sizeof(unsigned long)
513 	 * PADDING
514 	 * ----------- top = topmax - TOP_OF_KERNEL_STACK_PADDING
515 	 * stack
516 	 * ----------- bottom = start + sizeof(thread_info)
517 	 * thread_info
518 	 * ----------- start
519 	 *
520 	 * The tasks stack pointer points at the location where the
521 	 * framepointer is stored. The data on the stack is:
522 	 * ... IP FP ... IP FP
523 	 *
524 	 * We need to read FP and IP, so we need to adjust the upper
525 	 * bound by another unsigned long.
526 	 */
527 	top = start + THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;
528 	top -= 2 * sizeof(unsigned long);
529 	bottom = start + sizeof(struct thread_info);
530 
531 	sp = READ_ONCE(p->thread.sp);
532 	if (sp < bottom || sp > top)
533 		return 0;
534 
535 	fp = READ_ONCE(*(unsigned long *)sp);
536 	do {
537 		if (fp < bottom || fp > top)
538 			return 0;
539 		ip = READ_ONCE(*(unsigned long *)(fp + sizeof(unsigned long)));
540 		if (!in_sched_functions(ip))
541 			return ip;
542 		fp = READ_ONCE(*(unsigned long *)fp);
543 	} while (count++ < 16 && p->state != TASK_RUNNING);
544 	return 0;
545 }
546 
do_arch_prctl(struct task_struct * task,int code,unsigned long addr)547 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
548 {
549 	int ret = 0;
550 	int doit = task == current;
551 	int cpu;
552 
553 	switch (code) {
554 	case ARCH_SET_GS:
555 		if (addr >= TASK_SIZE_OF(task))
556 			return -EPERM;
557 		cpu = get_cpu();
558 		/* handle small bases via the GDT because that's faster to
559 		   switch. */
560 		if (addr <= 0xffffffff) {
561 			set_32bit_tls(task, GS_TLS, addr);
562 			if (doit) {
563 				load_TLS(&task->thread, cpu);
564 				load_gs_index(GS_TLS_SEL);
565 			}
566 			task->thread.gsindex = GS_TLS_SEL;
567 			task->thread.gs = 0;
568 		} else {
569 			task->thread.gsindex = 0;
570 			task->thread.gs = addr;
571 			if (doit) {
572 				load_gs_index(0);
573 				ret = wrmsrl_safe(MSR_KERNEL_GS_BASE, addr);
574 			}
575 		}
576 		put_cpu();
577 		break;
578 	case ARCH_SET_FS:
579 		/* Not strictly needed for fs, but do it for symmetry
580 		   with gs */
581 		if (addr >= TASK_SIZE_OF(task))
582 			return -EPERM;
583 		cpu = get_cpu();
584 		/* handle small bases via the GDT because that's faster to
585 		   switch. */
586 		if (addr <= 0xffffffff) {
587 			set_32bit_tls(task, FS_TLS, addr);
588 			if (doit) {
589 				load_TLS(&task->thread, cpu);
590 				loadsegment(fs, FS_TLS_SEL);
591 			}
592 			task->thread.fsindex = FS_TLS_SEL;
593 			task->thread.fs = 0;
594 		} else {
595 			task->thread.fsindex = 0;
596 			task->thread.fs = addr;
597 			if (doit) {
598 				/* set the selector to 0 to not confuse
599 				   __switch_to */
600 				loadsegment(fs, 0);
601 				ret = wrmsrl_safe(MSR_FS_BASE, addr);
602 			}
603 		}
604 		put_cpu();
605 		break;
606 	case ARCH_GET_FS: {
607 		unsigned long base;
608 		if (task->thread.fsindex == FS_TLS_SEL)
609 			base = read_32bit_tls(task, FS_TLS);
610 		else if (doit)
611 			rdmsrl(MSR_FS_BASE, base);
612 		else
613 			base = task->thread.fs;
614 		ret = put_user(base, (unsigned long __user *)addr);
615 		break;
616 	}
617 	case ARCH_GET_GS: {
618 		unsigned long base;
619 		unsigned gsindex;
620 		if (task->thread.gsindex == GS_TLS_SEL)
621 			base = read_32bit_tls(task, GS_TLS);
622 		else if (doit) {
623 			savesegment(gs, gsindex);
624 			if (gsindex)
625 				rdmsrl(MSR_KERNEL_GS_BASE, base);
626 			else
627 				base = task->thread.gs;
628 		} else
629 			base = task->thread.gs;
630 		ret = put_user(base, (unsigned long __user *)addr);
631 		break;
632 	}
633 
634 	default:
635 		ret = -EINVAL;
636 		break;
637 	}
638 
639 	return ret;
640 }
641 
sys_arch_prctl(int code,unsigned long addr)642 long sys_arch_prctl(int code, unsigned long addr)
643 {
644 	return do_arch_prctl(current, code, addr);
645 }
646 
KSTK_ESP(struct task_struct * task)647 unsigned long KSTK_ESP(struct task_struct *task)
648 {
649 	return (test_tsk_thread_flag(task, TIF_IA32)) ?
650 			(task_pt_regs(task)->sp) : ((task)->thread.usersp);
651 }
652