Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 1995 Linus Torvalds |
3 | * | |
4 | * Pentium III FXSR, SSE support | |
5 | * Gareth Hughes <gareth@valinux.com>, May 2000 | |
6612538c | 6 | * |
1da177e4 LT |
7 | * X86-64 port |
8 | * Andi Kleen. | |
76e4f660 AR |
9 | * |
10 | * CPU hotplug support - ashok.raj@intel.com | |
1da177e4 LT |
11 | */ |
12 | ||
13 | /* | |
14 | * This file handles the architecture-dependent parts of process handling.. | |
15 | */ | |
16 | ||
17 | #include <stdarg.h> | |
18 | ||
76e4f660 | 19 | #include <linux/cpu.h> |
1da177e4 LT |
20 | #include <linux/errno.h> |
21 | #include <linux/sched.h> | |
6612538c | 22 | #include <linux/fs.h> |
1da177e4 LT |
23 | #include <linux/kernel.h> |
24 | #include <linux/mm.h> | |
25 | #include <linux/elfcore.h> | |
26 | #include <linux/smp.h> | |
27 | #include <linux/slab.h> | |
28 | #include <linux/user.h> | |
1da177e4 | 29 | #include <linux/interrupt.h> |
6612538c | 30 | #include <linux/utsname.h> |
1da177e4 | 31 | #include <linux/delay.h> |
6612538c | 32 | #include <linux/module.h> |
1da177e4 | 33 | #include <linux/ptrace.h> |
1da177e4 | 34 | #include <linux/random.h> |
95833c83 | 35 | #include <linux/notifier.h> |
c6fd91f0 | 36 | #include <linux/kprobes.h> |
1eeb66a1 | 37 | #include <linux/kdebug.h> |
02290683 | 38 | #include <linux/tick.h> |
529e25f6 | 39 | #include <linux/prctl.h> |
7de08b4e GP |
40 | #include <linux/uaccess.h> |
41 | #include <linux/io.h> | |
1da177e4 | 42 | |
1da177e4 LT |
43 | #include <asm/pgtable.h> |
44 | #include <asm/system.h> | |
1da177e4 LT |
45 | #include <asm/processor.h> |
46 | #include <asm/i387.h> | |
47 | #include <asm/mmu_context.h> | |
48 | #include <asm/pda.h> | |
49 | #include <asm/prctl.h> | |
1da177e4 LT |
50 | #include <asm/desc.h> |
51 | #include <asm/proto.h> | |
52 | #include <asm/ia32.h> | |
95833c83 | 53 | #include <asm/idle.h> |
bbc1f698 | 54 | #include <asm/syscalls.h> |
1da177e4 LT |
55 | |
56 | asmlinkage extern void ret_from_fork(void); | |
57 | ||
58 | unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED; | |
59 | ||
e041c683 | 60 | static ATOMIC_NOTIFIER_HEAD(idle_notifier); |
95833c83 AK |
61 | |
62 | void idle_notifier_register(struct notifier_block *n) | |
63 | { | |
e041c683 | 64 | atomic_notifier_chain_register(&idle_notifier, n); |
95833c83 | 65 | } |
95833c83 | 66 | |
95833c83 AK |
67 | void enter_idle(void) |
68 | { | |
a15da49d | 69 | write_pda(isidle, 1); |
e041c683 | 70 | atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL); |
95833c83 AK |
71 | } |
72 | ||
73 | static void __exit_idle(void) | |
74 | { | |
9446868b | 75 | if (test_and_clear_bit_pda(0, isidle) == 0) |
a15da49d | 76 | return; |
e041c683 | 77 | atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL); |
95833c83 AK |
78 | } |
79 | ||
80 | /* Called from interrupts to signify idle end */ | |
81 | void exit_idle(void) | |
82 | { | |
a15da49d AK |
83 | /* idle loop has pid 0 */ |
84 | if (current->pid) | |
95833c83 AK |
85 | return; |
86 | __exit_idle(); | |
87 | } | |
88 | ||
913da64b AN |
89 | #ifndef CONFIG_SMP |
90 | static inline void play_dead(void) | |
91 | { | |
92 | BUG(); | |
93 | } | |
94 | #endif | |
95 | ||
1da177e4 LT |
96 | /* |
97 | * The idle thread. There's no useful work to be | |
98 | * done, so just try to conserve power and have a | |
99 | * low exit latency (ie sit in a loop waiting for | |
100 | * somebody to say that they'd like to reschedule) | |
101 | */ | |
b10db7f0 | 102 | void cpu_idle(void) |
1da177e4 | 103 | { |
495ab9c0 | 104 | current_thread_info()->status |= TS_POLLING; |
1da177e4 LT |
105 | /* endless idle loop with no priority at all */ |
106 | while (1) { | |
b8f8c3cf | 107 | tick_nohz_stop_sched_tick(1); |
1da177e4 | 108 | while (!need_resched()) { |
1da177e4 | 109 | |
1da177e4 | 110 | rmb(); |
6ddd2a27 | 111 | |
76e4f660 AR |
112 | if (cpu_is_offline(smp_processor_id())) |
113 | play_dead(); | |
d331e739 VP |
114 | /* |
115 | * Idle routines should keep interrupts disabled | |
116 | * from here on, until they go to idle. | |
117 | * Otherwise, idle callbacks can misfire. | |
118 | */ | |
119 | local_irq_disable(); | |
95833c83 | 120 | enter_idle(); |
81d68a96 SR |
121 | /* Don't trace irqs off for idle */ |
122 | stop_critical_timings(); | |
6ddd2a27 | 123 | pm_idle(); |
81d68a96 | 124 | start_critical_timings(); |
a15da49d AK |
125 | /* In many cases the interrupt that ended idle |
126 | has already called exit_idle. But some idle | |
127 | loops can be woken up without interrupt. */ | |
95833c83 | 128 | __exit_idle(); |
1da177e4 LT |
129 | } |
130 | ||
02290683 | 131 | tick_nohz_restart_sched_tick(); |
5bfb5d69 | 132 | preempt_enable_no_resched(); |
1da177e4 | 133 | schedule(); |
5bfb5d69 | 134 | preempt_disable(); |
1da177e4 LT |
135 | } |
136 | } | |
137 | ||
6612538c | 138 | /* Prints also some state that isn't saved in the pt_regs */ |
e2ce07c8 | 139 | void __show_regs(struct pt_regs *regs, int all) |
1da177e4 LT |
140 | { |
141 | unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs; | |
bb1995d5 | 142 | unsigned long d0, d1, d2, d3, d6, d7; |
6612538c HS |
143 | unsigned int fsindex, gsindex; |
144 | unsigned int ds, cs, es; | |
1da177e4 LT |
145 | |
146 | printk("\n"); | |
147 | print_modules(); | |
8092c654 | 148 | printk(KERN_INFO "Pid: %d, comm: %.20s %s %s %.*s\n", |
9acf23c4 | 149 | current->pid, current->comm, print_tainted(), |
96b644bd SH |
150 | init_utsname()->release, |
151 | (int)strcspn(init_utsname()->version, " "), | |
152 | init_utsname()->version); | |
8092c654 | 153 | printk(KERN_INFO "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip); |
aafbd7eb | 154 | printk_address(regs->ip, 1); |
8092c654 GP |
155 | printk(KERN_INFO "RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, |
156 | regs->sp, regs->flags); | |
157 | printk(KERN_INFO "RAX: %016lx RBX: %016lx RCX: %016lx\n", | |
65ea5b03 | 158 | regs->ax, regs->bx, regs->cx); |
8092c654 | 159 | printk(KERN_INFO "RDX: %016lx RSI: %016lx RDI: %016lx\n", |
65ea5b03 | 160 | regs->dx, regs->si, regs->di); |
8092c654 | 161 | printk(KERN_INFO "RBP: %016lx R08: %016lx R09: %016lx\n", |
65ea5b03 | 162 | regs->bp, regs->r8, regs->r9); |
8092c654 | 163 | printk(KERN_INFO "R10: %016lx R11: %016lx R12: %016lx\n", |
7de08b4e | 164 | regs->r10, regs->r11, regs->r12); |
8092c654 | 165 | printk(KERN_INFO "R13: %016lx R14: %016lx R15: %016lx\n", |
7de08b4e | 166 | regs->r13, regs->r14, regs->r15); |
1da177e4 | 167 | |
7de08b4e GP |
168 | asm("movl %%ds,%0" : "=r" (ds)); |
169 | asm("movl %%cs,%0" : "=r" (cs)); | |
170 | asm("movl %%es,%0" : "=r" (es)); | |
1da177e4 LT |
171 | asm("movl %%fs,%0" : "=r" (fsindex)); |
172 | asm("movl %%gs,%0" : "=r" (gsindex)); | |
173 | ||
174 | rdmsrl(MSR_FS_BASE, fs); | |
7de08b4e GP |
175 | rdmsrl(MSR_GS_BASE, gs); |
176 | rdmsrl(MSR_KERNEL_GS_BASE, shadowgs); | |
1da177e4 | 177 | |
e2ce07c8 PE |
178 | if (!all) |
179 | return; | |
180 | ||
f51c9452 GOC |
181 | cr0 = read_cr0(); |
182 | cr2 = read_cr2(); | |
183 | cr3 = read_cr3(); | |
184 | cr4 = read_cr4(); | |
1da177e4 | 185 | |
8092c654 | 186 | printk(KERN_INFO "FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n", |
7de08b4e | 187 | fs, fsindex, gs, gsindex, shadowgs); |
8092c654 GP |
188 | printk(KERN_INFO "CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, |
189 | es, cr0); | |
190 | printk(KERN_INFO "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, | |
191 | cr4); | |
bb1995d5 AS |
192 | |
193 | get_debugreg(d0, 0); | |
194 | get_debugreg(d1, 1); | |
195 | get_debugreg(d2, 2); | |
8092c654 | 196 | printk(KERN_INFO "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2); |
bb1995d5 AS |
197 | get_debugreg(d3, 3); |
198 | get_debugreg(d6, 6); | |
199 | get_debugreg(d7, 7); | |
8092c654 | 200 | printk(KERN_INFO "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7); |
1da177e4 LT |
201 | } |
202 | ||
203 | void show_regs(struct pt_regs *regs) | |
204 | { | |
8092c654 | 205 | printk(KERN_INFO "CPU %d:", smp_processor_id()); |
e2ce07c8 | 206 | __show_regs(regs, 1); |
bc850d6b | 207 | show_trace(NULL, regs, (void *)(regs + 1), regs->bp); |
1da177e4 LT |
208 | } |
209 | ||
210 | /* | |
211 | * Free current thread data structures etc.. | |
212 | */ | |
213 | void exit_thread(void) | |
214 | { | |
215 | struct task_struct *me = current; | |
216 | struct thread_struct *t = &me->thread; | |
73649dab | 217 | |
6612538c | 218 | if (me->thread.io_bitmap_ptr) { |
1da177e4 LT |
219 | struct tss_struct *tss = &per_cpu(init_tss, get_cpu()); |
220 | ||
221 | kfree(t->io_bitmap_ptr); | |
222 | t->io_bitmap_ptr = NULL; | |
d3a4f48d | 223 | clear_thread_flag(TIF_IO_BITMAP); |
1da177e4 LT |
224 | /* |
225 | * Careful, clear this in the TSS too: | |
226 | */ | |
227 | memset(tss->io_bitmap, 0xff, t->io_bitmap_max); | |
228 | t->io_bitmap_max = 0; | |
229 | put_cpu(); | |
230 | } | |
93fa7636 MM |
231 | #ifdef CONFIG_X86_DS |
232 | /* Free any DS contexts that have not been properly released. */ | |
233 | if (unlikely(t->ds_ctx)) { | |
234 | /* we clear debugctl to make sure DS is not used. */ | |
235 | update_debugctlmsr(0); | |
236 | ds_free(t->ds_ctx); | |
237 | } | |
238 | #endif /* CONFIG_X86_DS */ | |
1da177e4 LT |
239 | } |
240 | ||
241 | void flush_thread(void) | |
242 | { | |
243 | struct task_struct *tsk = current; | |
1da177e4 | 244 | |
303cd153 MD |
245 | if (test_tsk_thread_flag(tsk, TIF_ABI_PENDING)) { |
246 | clear_tsk_thread_flag(tsk, TIF_ABI_PENDING); | |
247 | if (test_tsk_thread_flag(tsk, TIF_IA32)) { | |
248 | clear_tsk_thread_flag(tsk, TIF_IA32); | |
249 | } else { | |
250 | set_tsk_thread_flag(tsk, TIF_IA32); | |
4d9bc79c | 251 | current_thread_info()->status |= TS_COMPAT; |
303cd153 | 252 | } |
4d9bc79c | 253 | } |
303cd153 | 254 | clear_tsk_thread_flag(tsk, TIF_DEBUG); |
1da177e4 LT |
255 | |
256 | tsk->thread.debugreg0 = 0; | |
257 | tsk->thread.debugreg1 = 0; | |
258 | tsk->thread.debugreg2 = 0; | |
259 | tsk->thread.debugreg3 = 0; | |
260 | tsk->thread.debugreg6 = 0; | |
261 | tsk->thread.debugreg7 = 0; | |
6612538c | 262 | memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array)); |
1da177e4 LT |
263 | /* |
264 | * Forget coprocessor state.. | |
265 | */ | |
75118a82 | 266 | tsk->fpu_counter = 0; |
1da177e4 LT |
267 | clear_fpu(tsk); |
268 | clear_used_math(); | |
269 | } | |
270 | ||
271 | void release_thread(struct task_struct *dead_task) | |
272 | { | |
273 | if (dead_task->mm) { | |
274 | if (dead_task->mm->context.size) { | |
275 | printk("WARNING: dead process %8s still has LDT? <%p/%d>\n", | |
276 | dead_task->comm, | |
277 | dead_task->mm->context.ldt, | |
278 | dead_task->mm->context.size); | |
279 | BUG(); | |
280 | } | |
281 | } | |
282 | } | |
283 | ||
284 | static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr) | |
285 | { | |
6612538c | 286 | struct user_desc ud = { |
1da177e4 LT |
287 | .base_addr = addr, |
288 | .limit = 0xfffff, | |
289 | .seg_32bit = 1, | |
290 | .limit_in_pages = 1, | |
291 | .useable = 1, | |
292 | }; | |
ade1af77 | 293 | struct desc_struct *desc = t->thread.tls_array; |
1da177e4 | 294 | desc += tls; |
80fbb69a | 295 | fill_ldt(desc, &ud); |
1da177e4 LT |
296 | } |
297 | ||
298 | static inline u32 read_32bit_tls(struct task_struct *t, int tls) | |
299 | { | |
91394eb0 | 300 | return get_desc_base(&t->thread.tls_array[tls]); |
1da177e4 LT |
301 | } |
302 | ||
303 | /* | |
304 | * This gets called before we allocate a new thread and copy | |
305 | * the current task into it. | |
306 | */ | |
307 | void prepare_to_copy(struct task_struct *tsk) | |
308 | { | |
309 | unlazy_fpu(tsk); | |
310 | } | |
311 | ||
65ea5b03 | 312 | int copy_thread(int nr, unsigned long clone_flags, unsigned long sp, |
1da177e4 | 313 | unsigned long unused, |
7de08b4e | 314 | struct task_struct *p, struct pt_regs *regs) |
1da177e4 LT |
315 | { |
316 | int err; | |
7de08b4e | 317 | struct pt_regs *childregs; |
1da177e4 LT |
318 | struct task_struct *me = current; |
319 | ||
a88cde13 | 320 | childregs = ((struct pt_regs *) |
57eafdc2 | 321 | (THREAD_SIZE + task_stack_page(p))) - 1; |
1da177e4 LT |
322 | *childregs = *regs; |
323 | ||
65ea5b03 PA |
324 | childregs->ax = 0; |
325 | childregs->sp = sp; | |
326 | if (sp == ~0UL) | |
327 | childregs->sp = (unsigned long)childregs; | |
1da177e4 | 328 | |
faca6227 PA |
329 | p->thread.sp = (unsigned long) childregs; |
330 | p->thread.sp0 = (unsigned long) (childregs+1); | |
331 | p->thread.usersp = me->thread.usersp; | |
1da177e4 | 332 | |
e4f17c43 | 333 | set_tsk_thread_flag(p, TIF_FORK); |
1da177e4 LT |
334 | |
335 | p->thread.fs = me->thread.fs; | |
336 | p->thread.gs = me->thread.gs; | |
337 | ||
ada85708 JF |
338 | savesegment(gs, p->thread.gsindex); |
339 | savesegment(fs, p->thread.fsindex); | |
340 | savesegment(es, p->thread.es); | |
341 | savesegment(ds, p->thread.ds); | |
1da177e4 | 342 | |
d3a4f48d | 343 | if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) { |
1da177e4 LT |
344 | p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL); |
345 | if (!p->thread.io_bitmap_ptr) { | |
346 | p->thread.io_bitmap_max = 0; | |
347 | return -ENOMEM; | |
348 | } | |
a88cde13 AK |
349 | memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr, |
350 | IO_BITMAP_BYTES); | |
d3a4f48d | 351 | set_tsk_thread_flag(p, TIF_IO_BITMAP); |
6612538c | 352 | } |
1da177e4 LT |
353 | |
354 | /* | |
355 | * Set a new TLS for the child thread? | |
356 | */ | |
357 | if (clone_flags & CLONE_SETTLS) { | |
358 | #ifdef CONFIG_IA32_EMULATION | |
359 | if (test_thread_flag(TIF_IA32)) | |
efd1ca52 | 360 | err = do_set_thread_area(p, -1, |
65ea5b03 | 361 | (struct user_desc __user *)childregs->si, 0); |
7de08b4e GP |
362 | else |
363 | #endif | |
364 | err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8); | |
365 | if (err) | |
1da177e4 LT |
366 | goto out; |
367 | } | |
368 | err = 0; | |
369 | out: | |
370 | if (err && p->thread.io_bitmap_ptr) { | |
371 | kfree(p->thread.io_bitmap_ptr); | |
372 | p->thread.io_bitmap_max = 0; | |
373 | } | |
374 | return err; | |
375 | } | |
376 | ||
513ad84b IM |
377 | void |
378 | start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp) | |
379 | { | |
ada85708 JF |
380 | loadsegment(fs, 0); |
381 | loadsegment(es, 0); | |
382 | loadsegment(ds, 0); | |
513ad84b IM |
383 | load_gs_index(0); |
384 | regs->ip = new_ip; | |
385 | regs->sp = new_sp; | |
386 | write_pda(oldrsp, new_sp); | |
387 | regs->cs = __USER_CS; | |
388 | regs->ss = __USER_DS; | |
389 | regs->flags = 0x200; | |
390 | set_fs(USER_DS); | |
aa283f49 SS |
391 | /* |
392 | * Free the old FP and other extended state | |
393 | */ | |
394 | free_thread_xstate(current); | |
513ad84b IM |
395 | } |
396 | EXPORT_SYMBOL_GPL(start_thread); | |
397 | ||
529e25f6 EB |
398 | static void hard_disable_TSC(void) |
399 | { | |
400 | write_cr4(read_cr4() | X86_CR4_TSD); | |
401 | } | |
402 | ||
403 | void disable_TSC(void) | |
404 | { | |
405 | preempt_disable(); | |
406 | if (!test_and_set_thread_flag(TIF_NOTSC)) | |
407 | /* | |
408 | * Must flip the CPU state synchronously with | |
409 | * TIF_NOTSC in the current running context. | |
410 | */ | |
411 | hard_disable_TSC(); | |
412 | preempt_enable(); | |
413 | } | |
414 | ||
415 | static void hard_enable_TSC(void) | |
416 | { | |
417 | write_cr4(read_cr4() & ~X86_CR4_TSD); | |
418 | } | |
419 | ||
a4928cff | 420 | static void enable_TSC(void) |
529e25f6 EB |
421 | { |
422 | preempt_disable(); | |
423 | if (test_and_clear_thread_flag(TIF_NOTSC)) | |
424 | /* | |
425 | * Must flip the CPU state synchronously with | |
426 | * TIF_NOTSC in the current running context. | |
427 | */ | |
428 | hard_enable_TSC(); | |
429 | preempt_enable(); | |
430 | } | |
431 | ||
432 | int get_tsc_mode(unsigned long adr) | |
433 | { | |
434 | unsigned int val; | |
435 | ||
436 | if (test_thread_flag(TIF_NOTSC)) | |
437 | val = PR_TSC_SIGSEGV; | |
438 | else | |
439 | val = PR_TSC_ENABLE; | |
440 | ||
441 | return put_user(val, (unsigned int __user *)adr); | |
442 | } | |
443 | ||
444 | int set_tsc_mode(unsigned int val) | |
445 | { | |
446 | if (val == PR_TSC_SIGSEGV) | |
447 | disable_TSC(); | |
448 | else if (val == PR_TSC_ENABLE) | |
449 | enable_TSC(); | |
450 | else | |
451 | return -EINVAL; | |
452 | ||
453 | return 0; | |
454 | } | |
455 | ||
1da177e4 LT |
456 | /* |
457 | * This special macro can be used to load a debugging register | |
458 | */ | |
6612538c HS |
459 | #define loaddebug(thread, r) set_debugreg(thread->debugreg ## r, r) |
460 | ||
d3a4f48d | 461 | static inline void __switch_to_xtra(struct task_struct *prev_p, |
6612538c HS |
462 | struct task_struct *next_p, |
463 | struct tss_struct *tss) | |
d3a4f48d SE |
464 | { |
465 | struct thread_struct *prev, *next; | |
eee3af4a | 466 | unsigned long debugctl; |
d3a4f48d SE |
467 | |
468 | prev = &prev_p->thread, | |
469 | next = &next_p->thread; | |
470 | ||
eee3af4a | 471 | debugctl = prev->debugctlmsr; |
93fa7636 MM |
472 | |
473 | #ifdef CONFIG_X86_DS | |
34b2cd5b IM |
474 | { |
475 | unsigned long ds_prev = 0, ds_next = 0; | |
476 | ||
477 | if (prev->ds_ctx) | |
478 | ds_prev = (unsigned long)prev->ds_ctx->ds; | |
479 | if (next->ds_ctx) | |
480 | ds_next = (unsigned long)next->ds_ctx->ds; | |
481 | ||
482 | if (ds_next != ds_prev) { | |
483 | /* | |
484 | * We clear debugctl to make sure DS | |
485 | * is not in use when we change it: | |
486 | */ | |
487 | debugctl = 0; | |
488 | update_debugctlmsr(0); | |
489 | wrmsrl(MSR_IA32_DS_AREA, ds_next); | |
490 | } | |
eee3af4a | 491 | } |
93fa7636 | 492 | #endif /* CONFIG_X86_DS */ |
eee3af4a MM |
493 | |
494 | if (next->debugctlmsr != debugctl) | |
5b0e5084 | 495 | update_debugctlmsr(next->debugctlmsr); |
7e991604 | 496 | |
d3a4f48d SE |
497 | if (test_tsk_thread_flag(next_p, TIF_DEBUG)) { |
498 | loaddebug(next, 0); | |
499 | loaddebug(next, 1); | |
500 | loaddebug(next, 2); | |
501 | loaddebug(next, 3); | |
502 | /* no 4 and 5 */ | |
503 | loaddebug(next, 6); | |
504 | loaddebug(next, 7); | |
505 | } | |
506 | ||
529e25f6 EB |
507 | if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^ |
508 | test_tsk_thread_flag(next_p, TIF_NOTSC)) { | |
509 | /* prev and next are different */ | |
510 | if (test_tsk_thread_flag(next_p, TIF_NOTSC)) | |
511 | hard_disable_TSC(); | |
512 | else | |
513 | hard_enable_TSC(); | |
514 | } | |
515 | ||
d3a4f48d SE |
516 | if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) { |
517 | /* | |
518 | * Copy the relevant range of the IO bitmap. | |
519 | * Normally this is 128 bytes or less: | |
520 | */ | |
521 | memcpy(tss->io_bitmap, next->io_bitmap_ptr, | |
522 | max(prev->io_bitmap_max, next->io_bitmap_max)); | |
523 | } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) { | |
524 | /* | |
525 | * Clear any possible leftover bits: | |
526 | */ | |
527 | memset(tss->io_bitmap, 0xff, prev->io_bitmap_max); | |
528 | } | |
eee3af4a | 529 | |
93fa7636 | 530 | #ifdef CONFIG_X86_PTRACE_BTS |
eee3af4a MM |
531 | if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS)) |
532 | ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS); | |
533 | ||
534 | if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS)) | |
535 | ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES); | |
93fa7636 | 536 | #endif /* CONFIG_X86_PTRACE_BTS */ |
d3a4f48d SE |
537 | } |
538 | ||
1da177e4 LT |
539 | /* |
540 | * switch_to(x,y) should switch tasks from x to y. | |
541 | * | |
6612538c | 542 | * This could still be optimized: |
1da177e4 LT |
543 | * - fold all the options into a flag word and test it with a single test. |
544 | * - could test fs/gs bitsliced | |
099f318b AK |
545 | * |
546 | * Kprobes not supported here. Set the probe on schedule instead. | |
1da177e4 | 547 | */ |
f438d914 | 548 | struct task_struct * |
a88cde13 | 549 | __switch_to(struct task_struct *prev_p, struct task_struct *next_p) |
1da177e4 | 550 | { |
87b935a0 JF |
551 | struct thread_struct *prev = &prev_p->thread; |
552 | struct thread_struct *next = &next_p->thread; | |
6612538c | 553 | int cpu = smp_processor_id(); |
1da177e4 | 554 | struct tss_struct *tss = &per_cpu(init_tss, cpu); |
478de5a9 | 555 | unsigned fsindex, gsindex; |
1da177e4 | 556 | |
e07e23e1 | 557 | /* we're going to use this soon, after a few expensive things */ |
7de08b4e | 558 | if (next_p->fpu_counter > 5) |
61c4628b | 559 | prefetch(next->xstate); |
e07e23e1 | 560 | |
1da177e4 LT |
561 | /* |
562 | * Reload esp0, LDT and the page table pointer: | |
563 | */ | |
7818a1e0 | 564 | load_sp0(tss, next); |
1da177e4 | 565 | |
7de08b4e | 566 | /* |
1da177e4 LT |
567 | * Switch DS and ES. |
568 | * This won't pick up thread selector changes, but I guess that is ok. | |
569 | */ | |
ada85708 | 570 | savesegment(es, prev->es); |
1da177e4 | 571 | if (unlikely(next->es | prev->es)) |
7de08b4e | 572 | loadsegment(es, next->es); |
ada85708 JF |
573 | |
574 | savesegment(ds, prev->ds); | |
1da177e4 LT |
575 | if (unlikely(next->ds | prev->ds)) |
576 | loadsegment(ds, next->ds); | |
577 | ||
478de5a9 JF |
578 | |
579 | /* We must save %fs and %gs before load_TLS() because | |
580 | * %fs and %gs may be cleared by load_TLS(). | |
581 | * | |
582 | * (e.g. xen_load_tls()) | |
583 | */ | |
584 | savesegment(fs, fsindex); | |
585 | savesegment(gs, gsindex); | |
586 | ||
1da177e4 LT |
587 | load_TLS(next, cpu); |
588 | ||
3fe0a63e JF |
589 | /* |
590 | * Leave lazy mode, flushing any hypercalls made here. | |
591 | * This must be done before restoring TLS segments so | |
592 | * the GDT and LDT are properly updated, and must be | |
593 | * done before math_state_restore, so the TS bit is up | |
594 | * to date. | |
595 | */ | |
596 | arch_leave_lazy_cpu_mode(); | |
597 | ||
7de08b4e | 598 | /* |
1da177e4 | 599 | * Switch FS and GS. |
87b935a0 JF |
600 | * |
601 | * Segment register != 0 always requires a reload. Also | |
602 | * reload when it has changed. When prev process used 64bit | |
603 | * base always reload to avoid an information leak. | |
1da177e4 | 604 | */ |
87b935a0 JF |
605 | if (unlikely(fsindex | next->fsindex | prev->fs)) { |
606 | loadsegment(fs, next->fsindex); | |
7de08b4e | 607 | /* |
87b935a0 JF |
608 | * Check if the user used a selector != 0; if yes |
609 | * clear 64bit base, since overloaded base is always | |
610 | * mapped to the Null selector | |
611 | */ | |
612 | if (fsindex) | |
7de08b4e | 613 | prev->fs = 0; |
87b935a0 JF |
614 | } |
615 | /* when next process has a 64bit base use it */ | |
616 | if (next->fs) | |
617 | wrmsrl(MSR_FS_BASE, next->fs); | |
618 | prev->fsindex = fsindex; | |
619 | ||
620 | if (unlikely(gsindex | next->gsindex | prev->gs)) { | |
621 | load_gs_index(next->gsindex); | |
622 | if (gsindex) | |
7de08b4e | 623 | prev->gs = 0; |
1da177e4 | 624 | } |
87b935a0 JF |
625 | if (next->gs) |
626 | wrmsrl(MSR_KERNEL_GS_BASE, next->gs); | |
627 | prev->gsindex = gsindex; | |
1da177e4 | 628 | |
0a5ace2a AK |
629 | /* Must be after DS reload */ |
630 | unlazy_fpu(prev_p); | |
631 | ||
7de08b4e | 632 | /* |
45948d77 | 633 | * Switch the PDA and FPU contexts. |
1da177e4 | 634 | */ |
faca6227 PA |
635 | prev->usersp = read_pda(oldrsp); |
636 | write_pda(oldrsp, next->usersp); | |
7de08b4e | 637 | write_pda(pcurrent, next_p); |
18bd057b | 638 | |
a88cde13 | 639 | write_pda(kernelstack, |
87b935a0 JF |
640 | (unsigned long)task_stack_page(next_p) + |
641 | THREAD_SIZE - PDA_STACKOFFSET); | |
0a425405 AV |
642 | #ifdef CONFIG_CC_STACKPROTECTOR |
643 | write_pda(stack_canary, next_p->stack_canary); | |
644 | /* | |
645 | * Build time only check to make sure the stack_canary is at | |
646 | * offset 40 in the pda; this is a gcc ABI requirement | |
647 | */ | |
648 | BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40); | |
649 | #endif | |
1da177e4 LT |
650 | |
651 | /* | |
d3a4f48d | 652 | * Now maybe reload the debug registers and handle I/O bitmaps |
1da177e4 | 653 | */ |
eee3af4a MM |
654 | if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT || |
655 | task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV)) | |
d3a4f48d | 656 | __switch_to_xtra(prev_p, next_p, tss); |
1da177e4 | 657 | |
e07e23e1 AV |
658 | /* If the task has used fpu the last 5 timeslices, just do a full |
659 | * restore of the math state immediately to avoid the trap; the | |
660 | * chances of needing FPU soon are obviously high now | |
870568b3 SS |
661 | * |
662 | * tsk_used_math() checks prevent calling math_state_restore(), | |
663 | * which can sleep in the case of !tsk_used_math() | |
e07e23e1 | 664 | */ |
870568b3 | 665 | if (tsk_used_math(next_p) && next_p->fpu_counter > 5) |
e07e23e1 | 666 | math_state_restore(); |
1da177e4 LT |
667 | return prev_p; |
668 | } | |
669 | ||
670 | /* | |
671 | * sys_execve() executes a new program. | |
672 | */ | |
6612538c | 673 | asmlinkage |
1da177e4 | 674 | long sys_execve(char __user *name, char __user * __user *argv, |
5d119b2c | 675 | char __user * __user *envp, struct pt_regs *regs) |
1da177e4 LT |
676 | { |
677 | long error; | |
7de08b4e | 678 | char *filename; |
1da177e4 LT |
679 | |
680 | filename = getname(name); | |
681 | error = PTR_ERR(filename); | |
5d119b2c | 682 | if (IS_ERR(filename)) |
1da177e4 | 683 | return error; |
5d119b2c | 684 | error = do_execve(filename, argv, envp, regs); |
1da177e4 LT |
685 | putname(filename); |
686 | return error; | |
687 | } | |
688 | ||
689 | void set_personality_64bit(void) | |
690 | { | |
691 | /* inherit personality from parent */ | |
692 | ||
693 | /* Make sure to be in 64bit mode */ | |
6612538c | 694 | clear_thread_flag(TIF_IA32); |
1da177e4 LT |
695 | |
696 | /* TBD: overwrites user setup. Should have two bits. | |
697 | But 64bit processes have always behaved this way, | |
698 | so it's not too bad. The main problem is just that | |
6612538c | 699 | 32bit childs are affected again. */ |
1da177e4 LT |
700 | current->personality &= ~READ_IMPLIES_EXEC; |
701 | } | |
702 | ||
703 | asmlinkage long sys_fork(struct pt_regs *regs) | |
704 | { | |
65ea5b03 | 705 | return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL); |
1da177e4 LT |
706 | } |
707 | ||
a88cde13 AK |
708 | asmlinkage long |
709 | sys_clone(unsigned long clone_flags, unsigned long newsp, | |
710 | void __user *parent_tid, void __user *child_tid, struct pt_regs *regs) | |
1da177e4 LT |
711 | { |
712 | if (!newsp) | |
65ea5b03 | 713 | newsp = regs->sp; |
1da177e4 LT |
714 | return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid); |
715 | } | |
716 | ||
717 | /* | |
718 | * This is trivial, and on the face of it looks like it | |
719 | * could equally well be done in user mode. | |
720 | * | |
721 | * Not so, for quite unobvious reasons - register pressure. | |
722 | * In user mode vfork() cannot have a stack frame, and if | |
723 | * done by calling the "clone()" system call directly, you | |
724 | * do not have enough call-clobbered registers to hold all | |
725 | * the information you need. | |
726 | */ | |
727 | asmlinkage long sys_vfork(struct pt_regs *regs) | |
728 | { | |
65ea5b03 | 729 | return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0, |
1da177e4 LT |
730 | NULL, NULL); |
731 | } | |
732 | ||
733 | unsigned long get_wchan(struct task_struct *p) | |
734 | { | |
735 | unsigned long stack; | |
7de08b4e | 736 | u64 fp, ip; |
1da177e4 LT |
737 | int count = 0; |
738 | ||
7de08b4e GP |
739 | if (!p || p == current || p->state == TASK_RUNNING) |
740 | return 0; | |
57eafdc2 | 741 | stack = (unsigned long)task_stack_page(p); |
e1e23bb0 | 742 | if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE) |
1da177e4 | 743 | return 0; |
faca6227 | 744 | fp = *(u64 *)(p->thread.sp); |
7de08b4e | 745 | do { |
a88cde13 | 746 | if (fp < (unsigned long)stack || |
e1e23bb0 | 747 | fp >= (unsigned long)stack+THREAD_SIZE) |
7de08b4e | 748 | return 0; |
65ea5b03 PA |
749 | ip = *(u64 *)(fp+8); |
750 | if (!in_sched_functions(ip)) | |
751 | return ip; | |
7de08b4e GP |
752 | fp = *(u64 *)fp; |
753 | } while (count++ < 16); | |
1da177e4 LT |
754 | return 0; |
755 | } | |
756 | ||
757 | long do_arch_prctl(struct task_struct *task, int code, unsigned long addr) | |
7de08b4e GP |
758 | { |
759 | int ret = 0; | |
1da177e4 LT |
760 | int doit = task == current; |
761 | int cpu; | |
762 | ||
7de08b4e | 763 | switch (code) { |
1da177e4 | 764 | case ARCH_SET_GS: |
84929801 | 765 | if (addr >= TASK_SIZE_OF(task)) |
7de08b4e | 766 | return -EPERM; |
1da177e4 | 767 | cpu = get_cpu(); |
7de08b4e | 768 | /* handle small bases via the GDT because that's faster to |
1da177e4 | 769 | switch. */ |
7de08b4e GP |
770 | if (addr <= 0xffffffff) { |
771 | set_32bit_tls(task, GS_TLS, addr); | |
772 | if (doit) { | |
1da177e4 | 773 | load_TLS(&task->thread, cpu); |
7de08b4e | 774 | load_gs_index(GS_TLS_SEL); |
1da177e4 | 775 | } |
7de08b4e | 776 | task->thread.gsindex = GS_TLS_SEL; |
1da177e4 | 777 | task->thread.gs = 0; |
7de08b4e | 778 | } else { |
1da177e4 LT |
779 | task->thread.gsindex = 0; |
780 | task->thread.gs = addr; | |
781 | if (doit) { | |
a88cde13 AK |
782 | load_gs_index(0); |
783 | ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr); | |
7de08b4e | 784 | } |
1da177e4 LT |
785 | } |
786 | put_cpu(); | |
787 | break; | |
788 | case ARCH_SET_FS: | |
789 | /* Not strictly needed for fs, but do it for symmetry | |
790 | with gs */ | |
84929801 | 791 | if (addr >= TASK_SIZE_OF(task)) |
6612538c | 792 | return -EPERM; |
1da177e4 | 793 | cpu = get_cpu(); |
6612538c | 794 | /* handle small bases via the GDT because that's faster to |
1da177e4 | 795 | switch. */ |
6612538c | 796 | if (addr <= 0xffffffff) { |
1da177e4 | 797 | set_32bit_tls(task, FS_TLS, addr); |
6612538c HS |
798 | if (doit) { |
799 | load_TLS(&task->thread, cpu); | |
ada85708 | 800 | loadsegment(fs, FS_TLS_SEL); |
1da177e4 LT |
801 | } |
802 | task->thread.fsindex = FS_TLS_SEL; | |
803 | task->thread.fs = 0; | |
6612538c | 804 | } else { |
1da177e4 LT |
805 | task->thread.fsindex = 0; |
806 | task->thread.fs = addr; | |
807 | if (doit) { | |
808 | /* set the selector to 0 to not confuse | |
809 | __switch_to */ | |
ada85708 | 810 | loadsegment(fs, 0); |
a88cde13 | 811 | ret = checking_wrmsrl(MSR_FS_BASE, addr); |
1da177e4 LT |
812 | } |
813 | } | |
814 | put_cpu(); | |
815 | break; | |
6612538c HS |
816 | case ARCH_GET_FS: { |
817 | unsigned long base; | |
1da177e4 LT |
818 | if (task->thread.fsindex == FS_TLS_SEL) |
819 | base = read_32bit_tls(task, FS_TLS); | |
a88cde13 | 820 | else if (doit) |
1da177e4 | 821 | rdmsrl(MSR_FS_BASE, base); |
a88cde13 | 822 | else |
1da177e4 | 823 | base = task->thread.fs; |
6612538c HS |
824 | ret = put_user(base, (unsigned long __user *)addr); |
825 | break; | |
1da177e4 | 826 | } |
6612538c | 827 | case ARCH_GET_GS: { |
1da177e4 | 828 | unsigned long base; |
97c2803c | 829 | unsigned gsindex; |
1da177e4 LT |
830 | if (task->thread.gsindex == GS_TLS_SEL) |
831 | base = read_32bit_tls(task, GS_TLS); | |
97c2803c | 832 | else if (doit) { |
ada85708 | 833 | savesegment(gs, gsindex); |
97c2803c JB |
834 | if (gsindex) |
835 | rdmsrl(MSR_KERNEL_GS_BASE, base); | |
836 | else | |
837 | base = task->thread.gs; | |
7de08b4e | 838 | } else |
1da177e4 | 839 | base = task->thread.gs; |
6612538c | 840 | ret = put_user(base, (unsigned long __user *)addr); |
1da177e4 LT |
841 | break; |
842 | } | |
843 | ||
844 | default: | |
845 | ret = -EINVAL; | |
846 | break; | |
6612538c | 847 | } |
1da177e4 | 848 | |
6612538c HS |
849 | return ret; |
850 | } | |
1da177e4 LT |
851 | |
852 | long sys_arch_prctl(int code, unsigned long addr) | |
853 | { | |
854 | return do_arch_prctl(current, code, addr); | |
1da177e4 LT |
855 | } |
856 | ||
857 | unsigned long arch_align_stack(unsigned long sp) | |
858 | { | |
c16b63e0 | 859 | if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space) |
1da177e4 LT |
860 | sp -= get_random_int() % 8192; |
861 | return sp & ~0xf; | |
862 | } | |
c1d171a0 JK |
863 | |
864 | unsigned long arch_randomize_brk(struct mm_struct *mm) | |
865 | { | |
866 | unsigned long range_end = mm->brk + 0x02000000; | |
867 | return randomize_range(mm->brk, range_end, 0) ? : mm->brk; | |
868 | } |