]>
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 | ||
76e4f660 | 17 | #include <linux/cpu.h> |
1da177e4 LT |
18 | #include <linux/errno.h> |
19 | #include <linux/sched.h> | |
29930025 | 20 | #include <linux/sched/task.h> |
68db0cf1 | 21 | #include <linux/sched/task_stack.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 LT |
29 | #include <linux/interrupt.h> |
30 | #include <linux/delay.h> | |
186f4360 | 31 | #include <linux/export.h> |
1da177e4 | 32 | #include <linux/ptrace.h> |
95833c83 | 33 | #include <linux/notifier.h> |
c6fd91f0 | 34 | #include <linux/kprobes.h> |
1eeb66a1 | 35 | #include <linux/kdebug.h> |
529e25f6 | 36 | #include <linux/prctl.h> |
7de08b4e GP |
37 | #include <linux/uaccess.h> |
38 | #include <linux/io.h> | |
8b96f011 | 39 | #include <linux/ftrace.h> |
ff3f097e | 40 | #include <linux/syscalls.h> |
1da177e4 | 41 | |
1da177e4 | 42 | #include <asm/pgtable.h> |
1da177e4 | 43 | #include <asm/processor.h> |
78f7f1e5 | 44 | #include <asm/fpu/internal.h> |
1da177e4 | 45 | #include <asm/mmu_context.h> |
1da177e4 | 46 | #include <asm/prctl.h> |
1da177e4 LT |
47 | #include <asm/desc.h> |
48 | #include <asm/proto.h> | |
49 | #include <asm/ia32.h> | |
bbc1f698 | 50 | #include <asm/syscalls.h> |
66cb5917 | 51 | #include <asm/debugreg.h> |
f05e798a | 52 | #include <asm/switch_to.h> |
b7a58459 | 53 | #include <asm/xen/hypervisor.h> |
2eefd878 | 54 | #include <asm/vdso.h> |
05830204 | 55 | #include <asm/intel_rdt_sched.h> |
ada26481 DS |
56 | #include <asm/unistd.h> |
57 | #ifdef CONFIG_IA32_EMULATION | |
58 | /* Not included via unistd.h */ | |
59 | #include <asm/unistd_32_ia32.h> | |
60 | #endif | |
1da177e4 | 61 | |
c38e5038 | 62 | __visible DEFINE_PER_CPU(unsigned long, rsp_scratch); |
1da177e4 | 63 | |
6612538c | 64 | /* Prints also some state that isn't saved in the pt_regs */ |
e2ce07c8 | 65 | void __show_regs(struct pt_regs *regs, int all) |
1da177e4 LT |
66 | { |
67 | unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs; | |
bb1995d5 | 68 | unsigned long d0, d1, d2, d3, d6, d7; |
6612538c HS |
69 | unsigned int fsindex, gsindex; |
70 | unsigned int ds, cs, es; | |
814e2c84 | 71 | |
b02fcf9b JP |
72 | show_iret_regs(regs); |
73 | ||
6fa81a12 JP |
74 | if (regs->orig_ax != -1) |
75 | pr_cont(" ORIG_RAX: %016lx\n", regs->orig_ax); | |
76 | else | |
77 | pr_cont("\n"); | |
78 | ||
d015a092 | 79 | printk(KERN_DEFAULT "RAX: %016lx RBX: %016lx RCX: %016lx\n", |
65ea5b03 | 80 | regs->ax, regs->bx, regs->cx); |
d015a092 | 81 | printk(KERN_DEFAULT "RDX: %016lx RSI: %016lx RDI: %016lx\n", |
65ea5b03 | 82 | regs->dx, regs->si, regs->di); |
d015a092 | 83 | printk(KERN_DEFAULT "RBP: %016lx R08: %016lx R09: %016lx\n", |
65ea5b03 | 84 | regs->bp, regs->r8, regs->r9); |
d015a092 | 85 | printk(KERN_DEFAULT "R10: %016lx R11: %016lx R12: %016lx\n", |
7de08b4e | 86 | regs->r10, regs->r11, regs->r12); |
d015a092 | 87 | printk(KERN_DEFAULT "R13: %016lx R14: %016lx R15: %016lx\n", |
7de08b4e | 88 | regs->r13, regs->r14, regs->r15); |
1da177e4 | 89 | |
b02fcf9b JP |
90 | if (!all) |
91 | return; | |
92 | ||
7de08b4e GP |
93 | asm("movl %%ds,%0" : "=r" (ds)); |
94 | asm("movl %%cs,%0" : "=r" (cs)); | |
95 | asm("movl %%es,%0" : "=r" (es)); | |
1da177e4 LT |
96 | asm("movl %%fs,%0" : "=r" (fsindex)); |
97 | asm("movl %%gs,%0" : "=r" (gsindex)); | |
98 | ||
99 | rdmsrl(MSR_FS_BASE, fs); | |
7de08b4e GP |
100 | rdmsrl(MSR_GS_BASE, gs); |
101 | rdmsrl(MSR_KERNEL_GS_BASE, shadowgs); | |
1da177e4 | 102 | |
f51c9452 GOC |
103 | cr0 = read_cr0(); |
104 | cr2 = read_cr2(); | |
6c690ee1 | 105 | cr3 = __read_cr3(); |
1e02ce4c | 106 | cr4 = __read_cr4(); |
1da177e4 | 107 | |
d015a092 | 108 | printk(KERN_DEFAULT "FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n", |
7de08b4e | 109 | fs, fsindex, gs, gsindex, shadowgs); |
d015a092 | 110 | printk(KERN_DEFAULT "CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, |
8092c654 | 111 | es, cr0); |
d015a092 | 112 | printk(KERN_DEFAULT "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, |
8092c654 | 113 | cr4); |
bb1995d5 AS |
114 | |
115 | get_debugreg(d0, 0); | |
116 | get_debugreg(d1, 1); | |
117 | get_debugreg(d2, 2); | |
bb1995d5 AS |
118 | get_debugreg(d3, 3); |
119 | get_debugreg(d6, 6); | |
120 | get_debugreg(d7, 7); | |
4338774c DJ |
121 | |
122 | /* Only print out debug registers if they are in their non-default state. */ | |
ba6d018e NI |
123 | if (!((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) && |
124 | (d6 == DR6_RESERVED) && (d7 == 0x400))) { | |
125 | printk(KERN_DEFAULT "DR0: %016lx DR1: %016lx DR2: %016lx\n", | |
126 | d0, d1, d2); | |
127 | printk(KERN_DEFAULT "DR3: %016lx DR6: %016lx DR7: %016lx\n", | |
128 | d3, d6, d7); | |
129 | } | |
4338774c | 130 | |
c0b17b5b DH |
131 | if (boot_cpu_has(X86_FEATURE_OSPKE)) |
132 | printk(KERN_DEFAULT "PKRU: %08x\n", read_pkru()); | |
1da177e4 LT |
133 | } |
134 | ||
1da177e4 LT |
135 | void release_thread(struct task_struct *dead_task) |
136 | { | |
137 | if (dead_task->mm) { | |
a5b9e5a2 | 138 | #ifdef CONFIG_MODIFY_LDT_SYSCALL |
37868fe1 | 139 | if (dead_task->mm->context.ldt) { |
349eab6e | 140 | pr_warn("WARNING: dead process %s still has LDT? <%p/%d>\n", |
c767a54b | 141 | dead_task->comm, |
0d430e3f | 142 | dead_task->mm->context.ldt->entries, |
bbf79d21 | 143 | dead_task->mm->context.ldt->nr_entries); |
1da177e4 LT |
144 | BUG(); |
145 | } | |
a5b9e5a2 | 146 | #endif |
1da177e4 LT |
147 | } |
148 | } | |
149 | ||
e137a4d8 AL |
150 | enum which_selector { |
151 | FS, | |
152 | GS | |
153 | }; | |
154 | ||
155 | /* | |
156 | * Saves the FS or GS base for an outgoing thread if FSGSBASE extensions are | |
157 | * not available. The goal is to be reasonably fast on non-FSGSBASE systems. | |
158 | * It's forcibly inlined because it'll generate better code and this function | |
159 | * is hot. | |
160 | */ | |
161 | static __always_inline void save_base_legacy(struct task_struct *prev_p, | |
162 | unsigned short selector, | |
163 | enum which_selector which) | |
164 | { | |
165 | if (likely(selector == 0)) { | |
166 | /* | |
167 | * On Intel (without X86_BUG_NULL_SEG), the segment base could | |
168 | * be the pre-existing saved base or it could be zero. On AMD | |
169 | * (with X86_BUG_NULL_SEG), the segment base could be almost | |
170 | * anything. | |
171 | * | |
172 | * This branch is very hot (it's hit twice on almost every | |
173 | * context switch between 64-bit programs), and avoiding | |
174 | * the RDMSR helps a lot, so we just assume that whatever | |
175 | * value is already saved is correct. This matches historical | |
176 | * Linux behavior, so it won't break existing applications. | |
177 | * | |
178 | * To avoid leaking state, on non-X86_BUG_NULL_SEG CPUs, if we | |
179 | * report that the base is zero, it needs to actually be zero: | |
180 | * see the corresponding logic in load_seg_legacy. | |
181 | */ | |
182 | } else { | |
183 | /* | |
184 | * If the selector is 1, 2, or 3, then the base is zero on | |
185 | * !X86_BUG_NULL_SEG CPUs and could be anything on | |
186 | * X86_BUG_NULL_SEG CPUs. In the latter case, Linux | |
187 | * has never attempted to preserve the base across context | |
188 | * switches. | |
189 | * | |
190 | * If selector > 3, then it refers to a real segment, and | |
191 | * saving the base isn't necessary. | |
192 | */ | |
193 | if (which == FS) | |
194 | prev_p->thread.fsbase = 0; | |
195 | else | |
196 | prev_p->thread.gsbase = 0; | |
197 | } | |
198 | } | |
199 | ||
200 | static __always_inline void save_fsgs(struct task_struct *task) | |
201 | { | |
202 | savesegment(fs, task->thread.fsindex); | |
203 | savesegment(gs, task->thread.gsindex); | |
204 | save_base_legacy(task, task->thread.fsindex, FS); | |
205 | save_base_legacy(task, task->thread.gsindex, GS); | |
206 | } | |
207 | ||
42b933b5 VK |
208 | #if IS_ENABLED(CONFIG_KVM) |
209 | /* | |
210 | * While a process is running,current->thread.fsbase and current->thread.gsbase | |
211 | * may not match the corresponding CPU registers (see save_base_legacy()). KVM | |
212 | * wants an efficient way to save and restore FSBASE and GSBASE. | |
213 | * When FSGSBASE extensions are enabled, this will have to use RD{FS,GS}BASE. | |
214 | */ | |
215 | void save_fsgs_for_kvm(void) | |
216 | { | |
217 | save_fsgs(current); | |
218 | } | |
219 | EXPORT_SYMBOL_GPL(save_fsgs_for_kvm); | |
220 | #endif | |
221 | ||
e137a4d8 AL |
222 | static __always_inline void loadseg(enum which_selector which, |
223 | unsigned short sel) | |
224 | { | |
225 | if (which == FS) | |
226 | loadsegment(fs, sel); | |
227 | else | |
228 | load_gs_index(sel); | |
229 | } | |
230 | ||
231 | static __always_inline void load_seg_legacy(unsigned short prev_index, | |
232 | unsigned long prev_base, | |
233 | unsigned short next_index, | |
234 | unsigned long next_base, | |
235 | enum which_selector which) | |
236 | { | |
237 | if (likely(next_index <= 3)) { | |
238 | /* | |
239 | * The next task is using 64-bit TLS, is not using this | |
240 | * segment at all, or is having fun with arcane CPU features. | |
241 | */ | |
242 | if (next_base == 0) { | |
243 | /* | |
244 | * Nasty case: on AMD CPUs, we need to forcibly zero | |
245 | * the base. | |
246 | */ | |
247 | if (static_cpu_has_bug(X86_BUG_NULL_SEG)) { | |
248 | loadseg(which, __USER_DS); | |
249 | loadseg(which, next_index); | |
250 | } else { | |
251 | /* | |
252 | * We could try to exhaustively detect cases | |
253 | * under which we can skip the segment load, | |
254 | * but there's really only one case that matters | |
255 | * for performance: if both the previous and | |
256 | * next states are fully zeroed, we can skip | |
257 | * the load. | |
258 | * | |
259 | * (This assumes that prev_base == 0 has no | |
260 | * false positives. This is the case on | |
261 | * Intel-style CPUs.) | |
262 | */ | |
263 | if (likely(prev_index | next_index | prev_base)) | |
264 | loadseg(which, next_index); | |
265 | } | |
266 | } else { | |
267 | if (prev_index != next_index) | |
268 | loadseg(which, next_index); | |
269 | wrmsrl(which == FS ? MSR_FS_BASE : MSR_KERNEL_GS_BASE, | |
270 | next_base); | |
271 | } | |
272 | } else { | |
273 | /* | |
274 | * The next task is using a real segment. Loading the selector | |
275 | * is sufficient. | |
276 | */ | |
277 | loadseg(which, next_index); | |
278 | } | |
279 | } | |
280 | ||
c1bd55f9 JT |
281 | int copy_thread_tls(unsigned long clone_flags, unsigned long sp, |
282 | unsigned long arg, struct task_struct *p, unsigned long tls) | |
1da177e4 LT |
283 | { |
284 | int err; | |
7de08b4e | 285 | struct pt_regs *childregs; |
0100301b BG |
286 | struct fork_frame *fork_frame; |
287 | struct inactive_task_frame *frame; | |
1da177e4 LT |
288 | struct task_struct *me = current; |
289 | ||
7076aada | 290 | childregs = task_pt_regs(p); |
0100301b BG |
291 | fork_frame = container_of(childregs, struct fork_frame, regs); |
292 | frame = &fork_frame->frame; | |
293 | frame->bp = 0; | |
294 | frame->ret_addr = (unsigned long) ret_from_fork; | |
295 | p->thread.sp = (unsigned long) fork_frame; | |
66cb5917 | 296 | p->thread.io_bitmap_ptr = NULL; |
1da177e4 | 297 | |
ada85708 | 298 | savesegment(gs, p->thread.gsindex); |
296f781a | 299 | p->thread.gsbase = p->thread.gsindex ? 0 : me->thread.gsbase; |
ada85708 | 300 | savesegment(fs, p->thread.fsindex); |
296f781a | 301 | p->thread.fsbase = p->thread.fsindex ? 0 : me->thread.fsbase; |
ada85708 JF |
302 | savesegment(es, p->thread.es); |
303 | savesegment(ds, p->thread.ds); | |
7076aada AV |
304 | memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps)); |
305 | ||
1d4b4b29 | 306 | if (unlikely(p->flags & PF_KTHREAD)) { |
7076aada AV |
307 | /* kernel thread */ |
308 | memset(childregs, 0, sizeof(struct pt_regs)); | |
616d2483 BG |
309 | frame->bx = sp; /* function */ |
310 | frame->r12 = arg; | |
7076aada AV |
311 | return 0; |
312 | } | |
616d2483 | 313 | frame->bx = 0; |
1d4b4b29 | 314 | *childregs = *current_pt_regs(); |
7076aada AV |
315 | |
316 | childregs->ax = 0; | |
1d4b4b29 AV |
317 | if (sp) |
318 | childregs->sp = sp; | |
1da177e4 | 319 | |
66cb5917 | 320 | err = -ENOMEM; |
d3a4f48d | 321 | if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) { |
cced4022 TM |
322 | p->thread.io_bitmap_ptr = kmemdup(me->thread.io_bitmap_ptr, |
323 | IO_BITMAP_BYTES, GFP_KERNEL); | |
1da177e4 LT |
324 | if (!p->thread.io_bitmap_ptr) { |
325 | p->thread.io_bitmap_max = 0; | |
326 | return -ENOMEM; | |
327 | } | |
d3a4f48d | 328 | set_tsk_thread_flag(p, TIF_IO_BITMAP); |
6612538c | 329 | } |
1da177e4 LT |
330 | |
331 | /* | |
332 | * Set a new TLS for the child thread? | |
333 | */ | |
334 | if (clone_flags & CLONE_SETTLS) { | |
335 | #ifdef CONFIG_IA32_EMULATION | |
abfb9498 | 336 | if (in_ia32_syscall()) |
efd1ca52 | 337 | err = do_set_thread_area(p, -1, |
c1bd55f9 | 338 | (struct user_desc __user *)tls, 0); |
7de08b4e GP |
339 | else |
340 | #endif | |
17a6e1b8 | 341 | err = do_arch_prctl_64(p, ARCH_SET_FS, tls); |
7de08b4e | 342 | if (err) |
1da177e4 LT |
343 | goto out; |
344 | } | |
345 | err = 0; | |
346 | out: | |
347 | if (err && p->thread.io_bitmap_ptr) { | |
348 | kfree(p->thread.io_bitmap_ptr); | |
349 | p->thread.io_bitmap_max = 0; | |
350 | } | |
66cb5917 | 351 | |
1da177e4 LT |
352 | return err; |
353 | } | |
354 | ||
e634d8fc PA |
355 | static void |
356 | start_thread_common(struct pt_regs *regs, unsigned long new_ip, | |
357 | unsigned long new_sp, | |
358 | unsigned int _cs, unsigned int _ss, unsigned int _ds) | |
513ad84b | 359 | { |
767d035d AL |
360 | WARN_ON_ONCE(regs != current_pt_regs()); |
361 | ||
362 | if (static_cpu_has(X86_BUG_NULL_SEG)) { | |
363 | /* Loading zero below won't clear the base. */ | |
364 | loadsegment(fs, __USER_DS); | |
365 | load_gs_index(__USER_DS); | |
366 | } | |
367 | ||
ada85708 | 368 | loadsegment(fs, 0); |
e634d8fc PA |
369 | loadsegment(es, _ds); |
370 | loadsegment(ds, _ds); | |
513ad84b | 371 | load_gs_index(0); |
767d035d | 372 | |
513ad84b IM |
373 | regs->ip = new_ip; |
374 | regs->sp = new_sp; | |
e634d8fc PA |
375 | regs->cs = _cs; |
376 | regs->ss = _ss; | |
a6f05a6a | 377 | regs->flags = X86_EFLAGS_IF; |
1daeaa31 | 378 | force_iret(); |
513ad84b | 379 | } |
e634d8fc PA |
380 | |
381 | void | |
382 | start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp) | |
383 | { | |
384 | start_thread_common(regs, new_ip, new_sp, | |
385 | __USER_CS, __USER_DS, 0); | |
386 | } | |
513ad84b | 387 | |
7da77078 BG |
388 | #ifdef CONFIG_COMPAT |
389 | void compat_start_thread(struct pt_regs *regs, u32 new_ip, u32 new_sp) | |
a6f05a6a | 390 | { |
e634d8fc | 391 | start_thread_common(regs, new_ip, new_sp, |
d1a797f3 PA |
392 | test_thread_flag(TIF_X32) |
393 | ? __USER_CS : __USER32_CS, | |
394 | __USER_DS, __USER_DS); | |
a6f05a6a PA |
395 | } |
396 | #endif | |
513ad84b | 397 | |
1da177e4 LT |
398 | /* |
399 | * switch_to(x,y) should switch tasks from x to y. | |
400 | * | |
6612538c | 401 | * This could still be optimized: |
1da177e4 LT |
402 | * - fold all the options into a flag word and test it with a single test. |
403 | * - could test fs/gs bitsliced | |
099f318b AK |
404 | * |
405 | * Kprobes not supported here. Set the probe on schedule instead. | |
8b96f011 | 406 | * Function graph tracer not supported too. |
1da177e4 | 407 | */ |
35ea7903 | 408 | __visible __notrace_funcgraph struct task_struct * |
a88cde13 | 409 | __switch_to(struct task_struct *prev_p, struct task_struct *next_p) |
1da177e4 | 410 | { |
87b935a0 JF |
411 | struct thread_struct *prev = &prev_p->thread; |
412 | struct thread_struct *next = &next_p->thread; | |
384a23f9 IM |
413 | struct fpu *prev_fpu = &prev->fpu; |
414 | struct fpu *next_fpu = &next->fpu; | |
6612538c | 415 | int cpu = smp_processor_id(); |
c482feef | 416 | struct tss_struct *tss = &per_cpu(cpu_tss_rw, cpu); |
e07e23e1 | 417 | |
1d3e53e8 AL |
418 | WARN_ON_ONCE(IS_ENABLED(CONFIG_DEBUG_ENTRY) && |
419 | this_cpu_read(irq_count) != -1); | |
420 | ||
c474e507 | 421 | switch_fpu_prepare(prev_fpu, cpu); |
4903062b | 422 | |
478de5a9 JF |
423 | /* We must save %fs and %gs before load_TLS() because |
424 | * %fs and %gs may be cleared by load_TLS(). | |
425 | * | |
426 | * (e.g. xen_load_tls()) | |
427 | */ | |
e137a4d8 | 428 | save_fsgs(prev_p); |
478de5a9 | 429 | |
f647d7c1 AL |
430 | /* |
431 | * Load TLS before restoring any segments so that segment loads | |
432 | * reference the correct GDT entries. | |
433 | */ | |
1da177e4 LT |
434 | load_TLS(next, cpu); |
435 | ||
3fe0a63e | 436 | /* |
f647d7c1 AL |
437 | * Leave lazy mode, flushing any hypercalls made here. This |
438 | * must be done after loading TLS entries in the GDT but before | |
439 | * loading segments that might reference them, and and it must | |
3a0aee48 | 440 | * be done before fpu__restore(), so the TS bit is up to |
f647d7c1 | 441 | * date. |
3fe0a63e | 442 | */ |
224101ed | 443 | arch_end_context_switch(next_p); |
3fe0a63e | 444 | |
f647d7c1 AL |
445 | /* Switch DS and ES. |
446 | * | |
447 | * Reading them only returns the selectors, but writing them (if | |
448 | * nonzero) loads the full descriptor from the GDT or LDT. The | |
449 | * LDT for next is loaded in switch_mm, and the GDT is loaded | |
450 | * above. | |
451 | * | |
452 | * We therefore need to write new values to the segment | |
453 | * registers on every context switch unless both the new and old | |
454 | * values are zero. | |
455 | * | |
456 | * Note that we don't need to do anything for CS and SS, as | |
457 | * those are saved and restored as part of pt_regs. | |
458 | */ | |
459 | savesegment(es, prev->es); | |
460 | if (unlikely(next->es | prev->es)) | |
461 | loadsegment(es, next->es); | |
462 | ||
463 | savesegment(ds, prev->ds); | |
464 | if (unlikely(next->ds | prev->ds)) | |
465 | loadsegment(ds, next->ds); | |
466 | ||
e137a4d8 AL |
467 | load_seg_legacy(prev->fsindex, prev->fsbase, |
468 | next->fsindex, next->fsbase, FS); | |
469 | load_seg_legacy(prev->gsindex, prev->gsbase, | |
470 | next->gsindex, next->gsbase, GS); | |
1da177e4 | 471 | |
c474e507 | 472 | switch_fpu_finish(next_fpu, cpu); |
34ddc81a | 473 | |
7de08b4e | 474 | /* |
45948d77 | 475 | * Switch the PDA and FPU contexts. |
1da177e4 | 476 | */ |
c6ae41e7 | 477 | this_cpu_write(current_task, next_p); |
9aaefe7b | 478 | this_cpu_write(cpu_current_top_of_stack, task_top_of_stack(next_p)); |
18bd057b | 479 | |
bd7dc5a6 | 480 | /* Reload sp0. */ |
252e1a05 | 481 | update_task_stack(next_p); |
b27559a4 | 482 | |
1da177e4 | 483 | /* |
d3a4f48d | 484 | * Now maybe reload the debug registers and handle I/O bitmaps |
1da177e4 | 485 | */ |
eee3af4a MM |
486 | if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT || |
487 | task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV)) | |
d3a4f48d | 488 | __switch_to_xtra(prev_p, next_p, tss); |
1da177e4 | 489 | |
5e57f1d6 | 490 | #ifdef CONFIG_XEN_PV |
b7a58459 AL |
491 | /* |
492 | * On Xen PV, IOPL bits in pt_regs->flags have no effect, and | |
493 | * current_pt_regs()->flags may not match the current task's | |
494 | * intended IOPL. We need to switch it manually. | |
495 | */ | |
496 | if (unlikely(static_cpu_has(X86_FEATURE_XENPV) && | |
497 | prev->iopl != next->iopl)) | |
498 | xen_set_iopl_mask(next->iopl); | |
499 | #endif | |
500 | ||
61f01dd9 AL |
501 | if (static_cpu_has_bug(X86_BUG_SYSRET_SS_ATTRS)) { |
502 | /* | |
503 | * AMD CPUs have a misfeature: SYSRET sets the SS selector but | |
504 | * does not update the cached descriptor. As a result, if we | |
505 | * do SYSRET while SS is NULL, we'll end up in user mode with | |
506 | * SS apparently equal to __USER_DS but actually unusable. | |
507 | * | |
508 | * The straightforward workaround would be to fix it up just | |
509 | * before SYSRET, but that would slow down the system call | |
510 | * fast paths. Instead, we ensure that SS is never NULL in | |
511 | * system call context. We do this by replacing NULL SS | |
512 | * selectors at every context switch. SYSCALL sets up a valid | |
513 | * SS, so the only way to get NULL is to re-enter the kernel | |
514 | * from CPL 3 through an interrupt. Since that can't happen | |
515 | * in the same task as a running syscall, we are guaranteed to | |
516 | * context switch between every interrupt vector entry and a | |
517 | * subsequent SYSRET. | |
518 | * | |
519 | * We read SS first because SS reads are much faster than | |
520 | * writes. Out of caution, we force SS to __KERNEL_DS even if | |
521 | * it previously had a different non-NULL value. | |
522 | */ | |
523 | unsigned short ss_sel; | |
524 | savesegment(ss, ss_sel); | |
525 | if (ss_sel != __KERNEL_DS) | |
526 | loadsegment(ss, __KERNEL_DS); | |
527 | } | |
528 | ||
4f341a5e FY |
529 | /* Load the Intel cache allocation PQR MSR. */ |
530 | intel_rdt_sched_in(); | |
531 | ||
1da177e4 LT |
532 | return prev_p; |
533 | } | |
534 | ||
1da177e4 LT |
535 | void set_personality_64bit(void) |
536 | { | |
537 | /* inherit personality from parent */ | |
538 | ||
539 | /* Make sure to be in 64bit mode */ | |
6612538c | 540 | clear_thread_flag(TIF_IA32); |
6bd33008 | 541 | clear_thread_flag(TIF_ADDR32); |
bb212724 | 542 | clear_thread_flag(TIF_X32); |
ada26481 DS |
543 | /* Pretend that this comes from a 64bit execve */ |
544 | task_pt_regs(current)->orig_ax = __NR_execve; | |
acf46020 | 545 | current_thread_info()->status &= ~TS_COMPAT; |
1da177e4 | 546 | |
375906f8 SW |
547 | /* Ensure the corresponding mm is not marked. */ |
548 | if (current->mm) | |
549 | current->mm->context.ia32_compat = 0; | |
550 | ||
1da177e4 LT |
551 | /* TBD: overwrites user setup. Should have two bits. |
552 | But 64bit processes have always behaved this way, | |
553 | so it's not too bad. The main problem is just that | |
6612538c | 554 | 32bit childs are affected again. */ |
1da177e4 LT |
555 | current->personality &= ~READ_IMPLIES_EXEC; |
556 | } | |
557 | ||
ada26481 | 558 | static void __set_personality_x32(void) |
05d43ed8 | 559 | { |
ada26481 DS |
560 | #ifdef CONFIG_X86_X32 |
561 | clear_thread_flag(TIF_IA32); | |
562 | set_thread_flag(TIF_X32); | |
563 | if (current->mm) | |
564 | current->mm->context.ia32_compat = TIF_X32; | |
565 | current->personality &= ~READ_IMPLIES_EXEC; | |
566 | /* | |
567 | * in_compat_syscall() uses the presence of the x32 syscall bit | |
568 | * flag to determine compat status. The x86 mmap() code relies on | |
569 | * the syscall bitness so set x32 syscall bit right here to make | |
570 | * in_compat_syscall() work during exec(). | |
571 | * | |
572 | * Pretend to come from a x32 execve. | |
573 | */ | |
574 | task_pt_regs(current)->orig_ax = __NR_x32_execve | __X32_SYSCALL_BIT; | |
37a8f7c3 | 575 | current_thread_info()->status &= ~TS_COMPAT; |
ada26481 DS |
576 | #endif |
577 | } | |
05d43ed8 | 578 | |
ada26481 DS |
579 | static void __set_personality_ia32(void) |
580 | { | |
581 | #ifdef CONFIG_IA32_EMULATION | |
582 | set_thread_flag(TIF_IA32); | |
583 | clear_thread_flag(TIF_X32); | |
584 | if (current->mm) | |
585 | current->mm->context.ia32_compat = TIF_IA32; | |
586 | current->personality |= force_personality32; | |
587 | /* Prepare the first "return" to user space */ | |
588 | task_pt_regs(current)->orig_ax = __NR_ia32_execve; | |
37a8f7c3 | 589 | current_thread_info()->status |= TS_COMPAT; |
ada26481 DS |
590 | #endif |
591 | } | |
592 | ||
593 | void set_personality_ia32(bool x32) | |
594 | { | |
05d43ed8 | 595 | /* Make sure to be in 32bit mode */ |
6bd33008 | 596 | set_thread_flag(TIF_ADDR32); |
05d43ed8 | 597 | |
ada26481 DS |
598 | if (x32) |
599 | __set_personality_x32(); | |
600 | else | |
601 | __set_personality_ia32(); | |
05d43ed8 | 602 | } |
febb72a6 | 603 | EXPORT_SYMBOL_GPL(set_personality_ia32); |
05d43ed8 | 604 | |
91b7bd39 | 605 | #ifdef CONFIG_CHECKPOINT_RESTORE |
2eefd878 DS |
606 | static long prctl_map_vdso(const struct vdso_image *image, unsigned long addr) |
607 | { | |
608 | int ret; | |
609 | ||
610 | ret = map_vdso_once(image, addr); | |
611 | if (ret) | |
612 | return ret; | |
613 | ||
614 | return (long)image->size; | |
615 | } | |
91b7bd39 | 616 | #endif |
2eefd878 | 617 | |
17a6e1b8 | 618 | long do_arch_prctl_64(struct task_struct *task, int option, unsigned long arg2) |
7de08b4e GP |
619 | { |
620 | int ret = 0; | |
1da177e4 LT |
621 | int doit = task == current; |
622 | int cpu; | |
623 | ||
dd93938a | 624 | switch (option) { |
1da177e4 | 625 | case ARCH_SET_GS: |
17a6e1b8 | 626 | if (arg2 >= TASK_SIZE_MAX) |
7de08b4e | 627 | return -EPERM; |
1da177e4 | 628 | cpu = get_cpu(); |
731e33e3 | 629 | task->thread.gsindex = 0; |
17a6e1b8 | 630 | task->thread.gsbase = arg2; |
731e33e3 AL |
631 | if (doit) { |
632 | load_gs_index(0); | |
17a6e1b8 | 633 | ret = wrmsrl_safe(MSR_KERNEL_GS_BASE, arg2); |
1da177e4 | 634 | } |
4afd0565 | 635 | put_cpu(); |
1da177e4 LT |
636 | break; |
637 | case ARCH_SET_FS: | |
638 | /* Not strictly needed for fs, but do it for symmetry | |
639 | with gs */ | |
17a6e1b8 | 640 | if (arg2 >= TASK_SIZE_MAX) |
6612538c | 641 | return -EPERM; |
1da177e4 | 642 | cpu = get_cpu(); |
731e33e3 | 643 | task->thread.fsindex = 0; |
17a6e1b8 | 644 | task->thread.fsbase = arg2; |
731e33e3 AL |
645 | if (doit) { |
646 | /* set the selector to 0 to not confuse __switch_to */ | |
647 | loadsegment(fs, 0); | |
17a6e1b8 | 648 | ret = wrmsrl_safe(MSR_FS_BASE, arg2); |
1da177e4 LT |
649 | } |
650 | put_cpu(); | |
651 | break; | |
6612538c HS |
652 | case ARCH_GET_FS: { |
653 | unsigned long base; | |
17a6e1b8 | 654 | |
d47b50e7 | 655 | if (doit) |
1da177e4 | 656 | rdmsrl(MSR_FS_BASE, base); |
a88cde13 | 657 | else |
296f781a | 658 | base = task->thread.fsbase; |
17a6e1b8 | 659 | ret = put_user(base, (unsigned long __user *)arg2); |
6612538c | 660 | break; |
1da177e4 | 661 | } |
6612538c | 662 | case ARCH_GET_GS: { |
1da177e4 | 663 | unsigned long base; |
17a6e1b8 | 664 | |
d47b50e7 AL |
665 | if (doit) |
666 | rdmsrl(MSR_KERNEL_GS_BASE, base); | |
d47b50e7 | 667 | else |
296f781a | 668 | base = task->thread.gsbase; |
17a6e1b8 | 669 | ret = put_user(base, (unsigned long __user *)arg2); |
1da177e4 LT |
670 | break; |
671 | } | |
672 | ||
2eefd878 | 673 | #ifdef CONFIG_CHECKPOINT_RESTORE |
6e68b087 | 674 | # ifdef CONFIG_X86_X32_ABI |
2eefd878 | 675 | case ARCH_MAP_VDSO_X32: |
17a6e1b8 | 676 | return prctl_map_vdso(&vdso_image_x32, arg2); |
91b7bd39 IM |
677 | # endif |
678 | # if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION | |
2eefd878 | 679 | case ARCH_MAP_VDSO_32: |
17a6e1b8 | 680 | return prctl_map_vdso(&vdso_image_32, arg2); |
91b7bd39 | 681 | # endif |
2eefd878 | 682 | case ARCH_MAP_VDSO_64: |
17a6e1b8 | 683 | return prctl_map_vdso(&vdso_image_64, arg2); |
2eefd878 DS |
684 | #endif |
685 | ||
1da177e4 LT |
686 | default: |
687 | ret = -EINVAL; | |
688 | break; | |
6612538c | 689 | } |
1da177e4 | 690 | |
6612538c HS |
691 | return ret; |
692 | } | |
1da177e4 | 693 | |
17a6e1b8 | 694 | SYSCALL_DEFINE2(arch_prctl, int, option, unsigned long, arg2) |
1da177e4 | 695 | { |
b0b9b014 KH |
696 | long ret; |
697 | ||
698 | ret = do_arch_prctl_64(current, option, arg2); | |
699 | if (ret == -EINVAL) | |
700 | ret = do_arch_prctl_common(current, option, arg2); | |
701 | ||
702 | return ret; | |
1da177e4 LT |
703 | } |
704 | ||
79170fda KH |
705 | #ifdef CONFIG_IA32_EMULATION |
706 | COMPAT_SYSCALL_DEFINE2(arch_prctl, int, option, unsigned long, arg2) | |
707 | { | |
708 | return do_arch_prctl_common(current, option, arg2); | |
709 | } | |
710 | #endif | |
711 | ||
89240ba0 SS |
712 | unsigned long KSTK_ESP(struct task_struct *task) |
713 | { | |
263042e4 | 714 | return task_pt_regs(task)->sp; |
89240ba0 | 715 | } |