]>
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 | |
b84511cd TG |
62 | #include "process.h" |
63 | ||
c38e5038 | 64 | __visible DEFINE_PER_CPU(unsigned long, rsp_scratch); |
1da177e4 | 65 | |
6612538c | 66 | /* Prints also some state that isn't saved in the pt_regs */ |
e2ce07c8 | 67 | void __show_regs(struct pt_regs *regs, int all) |
1da177e4 LT |
68 | { |
69 | unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs; | |
bb1995d5 | 70 | unsigned long d0, d1, d2, d3, d6, d7; |
6612538c HS |
71 | unsigned int fsindex, gsindex; |
72 | unsigned int ds, cs, es; | |
814e2c84 | 73 | |
b02fcf9b JP |
74 | show_iret_regs(regs); |
75 | ||
6fa81a12 JP |
76 | if (regs->orig_ax != -1) |
77 | pr_cont(" ORIG_RAX: %016lx\n", regs->orig_ax); | |
78 | else | |
79 | pr_cont("\n"); | |
80 | ||
d015a092 | 81 | printk(KERN_DEFAULT "RAX: %016lx RBX: %016lx RCX: %016lx\n", |
65ea5b03 | 82 | regs->ax, regs->bx, regs->cx); |
d015a092 | 83 | printk(KERN_DEFAULT "RDX: %016lx RSI: %016lx RDI: %016lx\n", |
65ea5b03 | 84 | regs->dx, regs->si, regs->di); |
d015a092 | 85 | printk(KERN_DEFAULT "RBP: %016lx R08: %016lx R09: %016lx\n", |
65ea5b03 | 86 | regs->bp, regs->r8, regs->r9); |
d015a092 | 87 | printk(KERN_DEFAULT "R10: %016lx R11: %016lx R12: %016lx\n", |
7de08b4e | 88 | regs->r10, regs->r11, regs->r12); |
d015a092 | 89 | printk(KERN_DEFAULT "R13: %016lx R14: %016lx R15: %016lx\n", |
7de08b4e | 90 | regs->r13, regs->r14, regs->r15); |
1da177e4 | 91 | |
b02fcf9b JP |
92 | if (!all) |
93 | return; | |
94 | ||
7de08b4e GP |
95 | asm("movl %%ds,%0" : "=r" (ds)); |
96 | asm("movl %%cs,%0" : "=r" (cs)); | |
97 | asm("movl %%es,%0" : "=r" (es)); | |
1da177e4 LT |
98 | asm("movl %%fs,%0" : "=r" (fsindex)); |
99 | asm("movl %%gs,%0" : "=r" (gsindex)); | |
100 | ||
101 | rdmsrl(MSR_FS_BASE, fs); | |
7de08b4e GP |
102 | rdmsrl(MSR_GS_BASE, gs); |
103 | rdmsrl(MSR_KERNEL_GS_BASE, shadowgs); | |
1da177e4 | 104 | |
f51c9452 GOC |
105 | cr0 = read_cr0(); |
106 | cr2 = read_cr2(); | |
6c690ee1 | 107 | cr3 = __read_cr3(); |
1e02ce4c | 108 | cr4 = __read_cr4(); |
1da177e4 | 109 | |
d015a092 | 110 | printk(KERN_DEFAULT "FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n", |
7de08b4e | 111 | fs, fsindex, gs, gsindex, shadowgs); |
d015a092 | 112 | printk(KERN_DEFAULT "CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, |
8092c654 | 113 | es, cr0); |
d015a092 | 114 | printk(KERN_DEFAULT "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, |
8092c654 | 115 | cr4); |
bb1995d5 AS |
116 | |
117 | get_debugreg(d0, 0); | |
118 | get_debugreg(d1, 1); | |
119 | get_debugreg(d2, 2); | |
bb1995d5 AS |
120 | get_debugreg(d3, 3); |
121 | get_debugreg(d6, 6); | |
122 | get_debugreg(d7, 7); | |
4338774c DJ |
123 | |
124 | /* Only print out debug registers if they are in their non-default state. */ | |
ba6d018e NI |
125 | if (!((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) && |
126 | (d6 == DR6_RESERVED) && (d7 == 0x400))) { | |
127 | printk(KERN_DEFAULT "DR0: %016lx DR1: %016lx DR2: %016lx\n", | |
128 | d0, d1, d2); | |
129 | printk(KERN_DEFAULT "DR3: %016lx DR6: %016lx DR7: %016lx\n", | |
130 | d3, d6, d7); | |
131 | } | |
4338774c | 132 | |
c0b17b5b DH |
133 | if (boot_cpu_has(X86_FEATURE_OSPKE)) |
134 | printk(KERN_DEFAULT "PKRU: %08x\n", read_pkru()); | |
1da177e4 LT |
135 | } |
136 | ||
1da177e4 LT |
137 | void release_thread(struct task_struct *dead_task) |
138 | { | |
139 | if (dead_task->mm) { | |
a5b9e5a2 | 140 | #ifdef CONFIG_MODIFY_LDT_SYSCALL |
37868fe1 | 141 | if (dead_task->mm->context.ldt) { |
349eab6e | 142 | pr_warn("WARNING: dead process %s still has LDT? <%p/%d>\n", |
c767a54b | 143 | dead_task->comm, |
0d430e3f | 144 | dead_task->mm->context.ldt->entries, |
bbf79d21 | 145 | dead_task->mm->context.ldt->nr_entries); |
1da177e4 LT |
146 | BUG(); |
147 | } | |
a5b9e5a2 | 148 | #endif |
1da177e4 LT |
149 | } |
150 | } | |
151 | ||
e137a4d8 AL |
152 | enum which_selector { |
153 | FS, | |
154 | GS | |
155 | }; | |
156 | ||
157 | /* | |
158 | * Saves the FS or GS base for an outgoing thread if FSGSBASE extensions are | |
159 | * not available. The goal is to be reasonably fast on non-FSGSBASE systems. | |
160 | * It's forcibly inlined because it'll generate better code and this function | |
161 | * is hot. | |
162 | */ | |
163 | static __always_inline void save_base_legacy(struct task_struct *prev_p, | |
164 | unsigned short selector, | |
165 | enum which_selector which) | |
166 | { | |
167 | if (likely(selector == 0)) { | |
168 | /* | |
169 | * On Intel (without X86_BUG_NULL_SEG), the segment base could | |
170 | * be the pre-existing saved base or it could be zero. On AMD | |
171 | * (with X86_BUG_NULL_SEG), the segment base could be almost | |
172 | * anything. | |
173 | * | |
174 | * This branch is very hot (it's hit twice on almost every | |
175 | * context switch between 64-bit programs), and avoiding | |
176 | * the RDMSR helps a lot, so we just assume that whatever | |
177 | * value is already saved is correct. This matches historical | |
178 | * Linux behavior, so it won't break existing applications. | |
179 | * | |
180 | * To avoid leaking state, on non-X86_BUG_NULL_SEG CPUs, if we | |
181 | * report that the base is zero, it needs to actually be zero: | |
182 | * see the corresponding logic in load_seg_legacy. | |
183 | */ | |
184 | } else { | |
185 | /* | |
186 | * If the selector is 1, 2, or 3, then the base is zero on | |
187 | * !X86_BUG_NULL_SEG CPUs and could be anything on | |
188 | * X86_BUG_NULL_SEG CPUs. In the latter case, Linux | |
189 | * has never attempted to preserve the base across context | |
190 | * switches. | |
191 | * | |
192 | * If selector > 3, then it refers to a real segment, and | |
193 | * saving the base isn't necessary. | |
194 | */ | |
195 | if (which == FS) | |
196 | prev_p->thread.fsbase = 0; | |
197 | else | |
198 | prev_p->thread.gsbase = 0; | |
199 | } | |
200 | } | |
201 | ||
202 | static __always_inline void save_fsgs(struct task_struct *task) | |
203 | { | |
204 | savesegment(fs, task->thread.fsindex); | |
205 | savesegment(gs, task->thread.gsindex); | |
206 | save_base_legacy(task, task->thread.fsindex, FS); | |
207 | save_base_legacy(task, task->thread.gsindex, GS); | |
208 | } | |
209 | ||
210 | static __always_inline void loadseg(enum which_selector which, | |
211 | unsigned short sel) | |
212 | { | |
213 | if (which == FS) | |
214 | loadsegment(fs, sel); | |
215 | else | |
216 | load_gs_index(sel); | |
217 | } | |
218 | ||
219 | static __always_inline void load_seg_legacy(unsigned short prev_index, | |
220 | unsigned long prev_base, | |
221 | unsigned short next_index, | |
222 | unsigned long next_base, | |
223 | enum which_selector which) | |
224 | { | |
225 | if (likely(next_index <= 3)) { | |
226 | /* | |
227 | * The next task is using 64-bit TLS, is not using this | |
228 | * segment at all, or is having fun with arcane CPU features. | |
229 | */ | |
230 | if (next_base == 0) { | |
231 | /* | |
232 | * Nasty case: on AMD CPUs, we need to forcibly zero | |
233 | * the base. | |
234 | */ | |
235 | if (static_cpu_has_bug(X86_BUG_NULL_SEG)) { | |
236 | loadseg(which, __USER_DS); | |
237 | loadseg(which, next_index); | |
238 | } else { | |
239 | /* | |
240 | * We could try to exhaustively detect cases | |
241 | * under which we can skip the segment load, | |
242 | * but there's really only one case that matters | |
243 | * for performance: if both the previous and | |
244 | * next states are fully zeroed, we can skip | |
245 | * the load. | |
246 | * | |
247 | * (This assumes that prev_base == 0 has no | |
248 | * false positives. This is the case on | |
249 | * Intel-style CPUs.) | |
250 | */ | |
251 | if (likely(prev_index | next_index | prev_base)) | |
252 | loadseg(which, next_index); | |
253 | } | |
254 | } else { | |
255 | if (prev_index != next_index) | |
256 | loadseg(which, next_index); | |
257 | wrmsrl(which == FS ? MSR_FS_BASE : MSR_KERNEL_GS_BASE, | |
258 | next_base); | |
259 | } | |
260 | } else { | |
261 | /* | |
262 | * The next task is using a real segment. Loading the selector | |
263 | * is sufficient. | |
264 | */ | |
265 | loadseg(which, next_index); | |
266 | } | |
267 | } | |
268 | ||
c1bd55f9 JT |
269 | int copy_thread_tls(unsigned long clone_flags, unsigned long sp, |
270 | unsigned long arg, struct task_struct *p, unsigned long tls) | |
1da177e4 LT |
271 | { |
272 | int err; | |
7de08b4e | 273 | struct pt_regs *childregs; |
0100301b BG |
274 | struct fork_frame *fork_frame; |
275 | struct inactive_task_frame *frame; | |
1da177e4 LT |
276 | struct task_struct *me = current; |
277 | ||
7076aada | 278 | childregs = task_pt_regs(p); |
0100301b BG |
279 | fork_frame = container_of(childregs, struct fork_frame, regs); |
280 | frame = &fork_frame->frame; | |
281 | frame->bp = 0; | |
282 | frame->ret_addr = (unsigned long) ret_from_fork; | |
283 | p->thread.sp = (unsigned long) fork_frame; | |
66cb5917 | 284 | p->thread.io_bitmap_ptr = NULL; |
1da177e4 | 285 | |
ada85708 | 286 | savesegment(gs, p->thread.gsindex); |
296f781a | 287 | p->thread.gsbase = p->thread.gsindex ? 0 : me->thread.gsbase; |
ada85708 | 288 | savesegment(fs, p->thread.fsindex); |
296f781a | 289 | p->thread.fsbase = p->thread.fsindex ? 0 : me->thread.fsbase; |
ada85708 JF |
290 | savesegment(es, p->thread.es); |
291 | savesegment(ds, p->thread.ds); | |
7076aada AV |
292 | memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps)); |
293 | ||
1d4b4b29 | 294 | if (unlikely(p->flags & PF_KTHREAD)) { |
7076aada AV |
295 | /* kernel thread */ |
296 | memset(childregs, 0, sizeof(struct pt_regs)); | |
616d2483 BG |
297 | frame->bx = sp; /* function */ |
298 | frame->r12 = arg; | |
7076aada AV |
299 | return 0; |
300 | } | |
616d2483 | 301 | frame->bx = 0; |
1d4b4b29 | 302 | *childregs = *current_pt_regs(); |
7076aada AV |
303 | |
304 | childregs->ax = 0; | |
1d4b4b29 AV |
305 | if (sp) |
306 | childregs->sp = sp; | |
1da177e4 | 307 | |
66cb5917 | 308 | err = -ENOMEM; |
d3a4f48d | 309 | if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) { |
cced4022 TM |
310 | p->thread.io_bitmap_ptr = kmemdup(me->thread.io_bitmap_ptr, |
311 | IO_BITMAP_BYTES, GFP_KERNEL); | |
1da177e4 LT |
312 | if (!p->thread.io_bitmap_ptr) { |
313 | p->thread.io_bitmap_max = 0; | |
314 | return -ENOMEM; | |
315 | } | |
d3a4f48d | 316 | set_tsk_thread_flag(p, TIF_IO_BITMAP); |
6612538c | 317 | } |
1da177e4 LT |
318 | |
319 | /* | |
320 | * Set a new TLS for the child thread? | |
321 | */ | |
322 | if (clone_flags & CLONE_SETTLS) { | |
323 | #ifdef CONFIG_IA32_EMULATION | |
abfb9498 | 324 | if (in_ia32_syscall()) |
efd1ca52 | 325 | err = do_set_thread_area(p, -1, |
c1bd55f9 | 326 | (struct user_desc __user *)tls, 0); |
7de08b4e GP |
327 | else |
328 | #endif | |
17a6e1b8 | 329 | err = do_arch_prctl_64(p, ARCH_SET_FS, tls); |
7de08b4e | 330 | if (err) |
1da177e4 LT |
331 | goto out; |
332 | } | |
333 | err = 0; | |
334 | out: | |
335 | if (err && p->thread.io_bitmap_ptr) { | |
336 | kfree(p->thread.io_bitmap_ptr); | |
337 | p->thread.io_bitmap_max = 0; | |
338 | } | |
66cb5917 | 339 | |
1da177e4 LT |
340 | return err; |
341 | } | |
342 | ||
e634d8fc PA |
343 | static void |
344 | start_thread_common(struct pt_regs *regs, unsigned long new_ip, | |
345 | unsigned long new_sp, | |
346 | unsigned int _cs, unsigned int _ss, unsigned int _ds) | |
513ad84b | 347 | { |
767d035d AL |
348 | WARN_ON_ONCE(regs != current_pt_regs()); |
349 | ||
350 | if (static_cpu_has(X86_BUG_NULL_SEG)) { | |
351 | /* Loading zero below won't clear the base. */ | |
352 | loadsegment(fs, __USER_DS); | |
353 | load_gs_index(__USER_DS); | |
354 | } | |
355 | ||
ada85708 | 356 | loadsegment(fs, 0); |
e634d8fc PA |
357 | loadsegment(es, _ds); |
358 | loadsegment(ds, _ds); | |
513ad84b | 359 | load_gs_index(0); |
767d035d | 360 | |
513ad84b IM |
361 | regs->ip = new_ip; |
362 | regs->sp = new_sp; | |
e634d8fc PA |
363 | regs->cs = _cs; |
364 | regs->ss = _ss; | |
a6f05a6a | 365 | regs->flags = X86_EFLAGS_IF; |
1daeaa31 | 366 | force_iret(); |
513ad84b | 367 | } |
e634d8fc PA |
368 | |
369 | void | |
370 | start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp) | |
371 | { | |
372 | start_thread_common(regs, new_ip, new_sp, | |
373 | __USER_CS, __USER_DS, 0); | |
374 | } | |
513ad84b | 375 | |
7da77078 BG |
376 | #ifdef CONFIG_COMPAT |
377 | void compat_start_thread(struct pt_regs *regs, u32 new_ip, u32 new_sp) | |
a6f05a6a | 378 | { |
e634d8fc | 379 | start_thread_common(regs, new_ip, new_sp, |
d1a797f3 PA |
380 | test_thread_flag(TIF_X32) |
381 | ? __USER_CS : __USER32_CS, | |
382 | __USER_DS, __USER_DS); | |
a6f05a6a PA |
383 | } |
384 | #endif | |
513ad84b | 385 | |
1da177e4 LT |
386 | /* |
387 | * switch_to(x,y) should switch tasks from x to y. | |
388 | * | |
6612538c | 389 | * This could still be optimized: |
1da177e4 LT |
390 | * - fold all the options into a flag word and test it with a single test. |
391 | * - could test fs/gs bitsliced | |
099f318b AK |
392 | * |
393 | * Kprobes not supported here. Set the probe on schedule instead. | |
8b96f011 | 394 | * Function graph tracer not supported too. |
1da177e4 | 395 | */ |
35ea7903 | 396 | __visible __notrace_funcgraph struct task_struct * |
a88cde13 | 397 | __switch_to(struct task_struct *prev_p, struct task_struct *next_p) |
1da177e4 | 398 | { |
87b935a0 JF |
399 | struct thread_struct *prev = &prev_p->thread; |
400 | struct thread_struct *next = &next_p->thread; | |
384a23f9 IM |
401 | struct fpu *prev_fpu = &prev->fpu; |
402 | struct fpu *next_fpu = &next->fpu; | |
6612538c | 403 | int cpu = smp_processor_id(); |
e07e23e1 | 404 | |
1d3e53e8 AL |
405 | WARN_ON_ONCE(IS_ENABLED(CONFIG_DEBUG_ENTRY) && |
406 | this_cpu_read(irq_count) != -1); | |
407 | ||
c474e507 | 408 | switch_fpu_prepare(prev_fpu, cpu); |
4903062b | 409 | |
478de5a9 JF |
410 | /* We must save %fs and %gs before load_TLS() because |
411 | * %fs and %gs may be cleared by load_TLS(). | |
412 | * | |
413 | * (e.g. xen_load_tls()) | |
414 | */ | |
e137a4d8 | 415 | save_fsgs(prev_p); |
478de5a9 | 416 | |
f647d7c1 AL |
417 | /* |
418 | * Load TLS before restoring any segments so that segment loads | |
419 | * reference the correct GDT entries. | |
420 | */ | |
1da177e4 LT |
421 | load_TLS(next, cpu); |
422 | ||
3fe0a63e | 423 | /* |
f647d7c1 AL |
424 | * Leave lazy mode, flushing any hypercalls made here. This |
425 | * must be done after loading TLS entries in the GDT but before | |
426 | * loading segments that might reference them, and and it must | |
3a0aee48 | 427 | * be done before fpu__restore(), so the TS bit is up to |
f647d7c1 | 428 | * date. |
3fe0a63e | 429 | */ |
224101ed | 430 | arch_end_context_switch(next_p); |
3fe0a63e | 431 | |
f647d7c1 AL |
432 | /* Switch DS and ES. |
433 | * | |
434 | * Reading them only returns the selectors, but writing them (if | |
435 | * nonzero) loads the full descriptor from the GDT or LDT. The | |
436 | * LDT for next is loaded in switch_mm, and the GDT is loaded | |
437 | * above. | |
438 | * | |
439 | * We therefore need to write new values to the segment | |
440 | * registers on every context switch unless both the new and old | |
441 | * values are zero. | |
442 | * | |
443 | * Note that we don't need to do anything for CS and SS, as | |
444 | * those are saved and restored as part of pt_regs. | |
445 | */ | |
446 | savesegment(es, prev->es); | |
447 | if (unlikely(next->es | prev->es)) | |
448 | loadsegment(es, next->es); | |
449 | ||
450 | savesegment(ds, prev->ds); | |
451 | if (unlikely(next->ds | prev->ds)) | |
452 | loadsegment(ds, next->ds); | |
453 | ||
e137a4d8 AL |
454 | load_seg_legacy(prev->fsindex, prev->fsbase, |
455 | next->fsindex, next->fsbase, FS); | |
456 | load_seg_legacy(prev->gsindex, prev->gsbase, | |
457 | next->gsindex, next->gsbase, GS); | |
1da177e4 | 458 | |
c474e507 | 459 | switch_fpu_finish(next_fpu, cpu); |
34ddc81a | 460 | |
7de08b4e | 461 | /* |
45948d77 | 462 | * Switch the PDA and FPU contexts. |
1da177e4 | 463 | */ |
c6ae41e7 | 464 | this_cpu_write(current_task, next_p); |
9aaefe7b | 465 | this_cpu_write(cpu_current_top_of_stack, task_top_of_stack(next_p)); |
18bd057b | 466 | |
bd7dc5a6 | 467 | /* Reload sp0. */ |
954aab97 | 468 | update_task_stack(next_p); |
b27559a4 | 469 | |
b84511cd | 470 | switch_to_extra(prev_p, next_p); |
1da177e4 | 471 | |
5e57f1d6 | 472 | #ifdef CONFIG_XEN_PV |
b7a58459 AL |
473 | /* |
474 | * On Xen PV, IOPL bits in pt_regs->flags have no effect, and | |
475 | * current_pt_regs()->flags may not match the current task's | |
476 | * intended IOPL. We need to switch it manually. | |
477 | */ | |
478 | if (unlikely(static_cpu_has(X86_FEATURE_XENPV) && | |
479 | prev->iopl != next->iopl)) | |
480 | xen_set_iopl_mask(next->iopl); | |
481 | #endif | |
482 | ||
61f01dd9 AL |
483 | if (static_cpu_has_bug(X86_BUG_SYSRET_SS_ATTRS)) { |
484 | /* | |
485 | * AMD CPUs have a misfeature: SYSRET sets the SS selector but | |
486 | * does not update the cached descriptor. As a result, if we | |
487 | * do SYSRET while SS is NULL, we'll end up in user mode with | |
488 | * SS apparently equal to __USER_DS but actually unusable. | |
489 | * | |
490 | * The straightforward workaround would be to fix it up just | |
491 | * before SYSRET, but that would slow down the system call | |
492 | * fast paths. Instead, we ensure that SS is never NULL in | |
493 | * system call context. We do this by replacing NULL SS | |
494 | * selectors at every context switch. SYSCALL sets up a valid | |
495 | * SS, so the only way to get NULL is to re-enter the kernel | |
496 | * from CPL 3 through an interrupt. Since that can't happen | |
497 | * in the same task as a running syscall, we are guaranteed to | |
498 | * context switch between every interrupt vector entry and a | |
499 | * subsequent SYSRET. | |
500 | * | |
501 | * We read SS first because SS reads are much faster than | |
502 | * writes. Out of caution, we force SS to __KERNEL_DS even if | |
503 | * it previously had a different non-NULL value. | |
504 | */ | |
505 | unsigned short ss_sel; | |
506 | savesegment(ss, ss_sel); | |
507 | if (ss_sel != __KERNEL_DS) | |
508 | loadsegment(ss, __KERNEL_DS); | |
509 | } | |
510 | ||
4f341a5e FY |
511 | /* Load the Intel cache allocation PQR MSR. */ |
512 | intel_rdt_sched_in(); | |
513 | ||
1da177e4 LT |
514 | return prev_p; |
515 | } | |
516 | ||
1da177e4 LT |
517 | void set_personality_64bit(void) |
518 | { | |
519 | /* inherit personality from parent */ | |
520 | ||
521 | /* Make sure to be in 64bit mode */ | |
6612538c | 522 | clear_thread_flag(TIF_IA32); |
6bd33008 | 523 | clear_thread_flag(TIF_ADDR32); |
bb212724 | 524 | clear_thread_flag(TIF_X32); |
ada26481 DS |
525 | /* Pretend that this comes from a 64bit execve */ |
526 | task_pt_regs(current)->orig_ax = __NR_execve; | |
a0add795 | 527 | current_thread_info()->status &= ~TS_COMPAT; |
1da177e4 | 528 | |
375906f8 SW |
529 | /* Ensure the corresponding mm is not marked. */ |
530 | if (current->mm) | |
531 | current->mm->context.ia32_compat = 0; | |
532 | ||
1da177e4 LT |
533 | /* TBD: overwrites user setup. Should have two bits. |
534 | But 64bit processes have always behaved this way, | |
535 | so it's not too bad. The main problem is just that | |
6612538c | 536 | 32bit childs are affected again. */ |
1da177e4 LT |
537 | current->personality &= ~READ_IMPLIES_EXEC; |
538 | } | |
539 | ||
ada26481 | 540 | static void __set_personality_x32(void) |
05d43ed8 | 541 | { |
ada26481 DS |
542 | #ifdef CONFIG_X86_X32 |
543 | clear_thread_flag(TIF_IA32); | |
544 | set_thread_flag(TIF_X32); | |
545 | if (current->mm) | |
546 | current->mm->context.ia32_compat = TIF_X32; | |
547 | current->personality &= ~READ_IMPLIES_EXEC; | |
548 | /* | |
549 | * in_compat_syscall() uses the presence of the x32 syscall bit | |
550 | * flag to determine compat status. The x86 mmap() code relies on | |
551 | * the syscall bitness so set x32 syscall bit right here to make | |
552 | * in_compat_syscall() work during exec(). | |
553 | * | |
554 | * Pretend to come from a x32 execve. | |
555 | */ | |
556 | task_pt_regs(current)->orig_ax = __NR_x32_execve | __X32_SYSCALL_BIT; | |
6c46cb88 | 557 | current_thread_info()->status &= ~TS_COMPAT; |
ada26481 DS |
558 | #endif |
559 | } | |
05d43ed8 | 560 | |
ada26481 DS |
561 | static void __set_personality_ia32(void) |
562 | { | |
563 | #ifdef CONFIG_IA32_EMULATION | |
564 | set_thread_flag(TIF_IA32); | |
565 | clear_thread_flag(TIF_X32); | |
566 | if (current->mm) | |
567 | current->mm->context.ia32_compat = TIF_IA32; | |
568 | current->personality |= force_personality32; | |
569 | /* Prepare the first "return" to user space */ | |
570 | task_pt_regs(current)->orig_ax = __NR_ia32_execve; | |
6c46cb88 | 571 | current_thread_info()->status |= TS_COMPAT; |
ada26481 DS |
572 | #endif |
573 | } | |
574 | ||
575 | void set_personality_ia32(bool x32) | |
576 | { | |
05d43ed8 | 577 | /* Make sure to be in 32bit mode */ |
6bd33008 | 578 | set_thread_flag(TIF_ADDR32); |
05d43ed8 | 579 | |
ada26481 DS |
580 | if (x32) |
581 | __set_personality_x32(); | |
582 | else | |
583 | __set_personality_ia32(); | |
05d43ed8 | 584 | } |
febb72a6 | 585 | EXPORT_SYMBOL_GPL(set_personality_ia32); |
05d43ed8 | 586 | |
91b7bd39 | 587 | #ifdef CONFIG_CHECKPOINT_RESTORE |
2eefd878 DS |
588 | static long prctl_map_vdso(const struct vdso_image *image, unsigned long addr) |
589 | { | |
590 | int ret; | |
591 | ||
592 | ret = map_vdso_once(image, addr); | |
593 | if (ret) | |
594 | return ret; | |
595 | ||
596 | return (long)image->size; | |
597 | } | |
91b7bd39 | 598 | #endif |
2eefd878 | 599 | |
17a6e1b8 | 600 | long do_arch_prctl_64(struct task_struct *task, int option, unsigned long arg2) |
7de08b4e GP |
601 | { |
602 | int ret = 0; | |
1da177e4 LT |
603 | int doit = task == current; |
604 | int cpu; | |
605 | ||
dd93938a | 606 | switch (option) { |
1da177e4 | 607 | case ARCH_SET_GS: |
17a6e1b8 | 608 | if (arg2 >= TASK_SIZE_MAX) |
7de08b4e | 609 | return -EPERM; |
1da177e4 | 610 | cpu = get_cpu(); |
731e33e3 | 611 | task->thread.gsindex = 0; |
17a6e1b8 | 612 | task->thread.gsbase = arg2; |
731e33e3 AL |
613 | if (doit) { |
614 | load_gs_index(0); | |
17a6e1b8 | 615 | ret = wrmsrl_safe(MSR_KERNEL_GS_BASE, arg2); |
1da177e4 | 616 | } |
4afd0565 | 617 | put_cpu(); |
1da177e4 LT |
618 | break; |
619 | case ARCH_SET_FS: | |
620 | /* Not strictly needed for fs, but do it for symmetry | |
621 | with gs */ | |
17a6e1b8 | 622 | if (arg2 >= TASK_SIZE_MAX) |
6612538c | 623 | return -EPERM; |
1da177e4 | 624 | cpu = get_cpu(); |
731e33e3 | 625 | task->thread.fsindex = 0; |
17a6e1b8 | 626 | task->thread.fsbase = arg2; |
731e33e3 AL |
627 | if (doit) { |
628 | /* set the selector to 0 to not confuse __switch_to */ | |
629 | loadsegment(fs, 0); | |
17a6e1b8 | 630 | ret = wrmsrl_safe(MSR_FS_BASE, arg2); |
1da177e4 LT |
631 | } |
632 | put_cpu(); | |
633 | break; | |
6612538c HS |
634 | case ARCH_GET_FS: { |
635 | unsigned long base; | |
17a6e1b8 | 636 | |
d47b50e7 | 637 | if (doit) |
1da177e4 | 638 | rdmsrl(MSR_FS_BASE, base); |
a88cde13 | 639 | else |
296f781a | 640 | base = task->thread.fsbase; |
17a6e1b8 | 641 | ret = put_user(base, (unsigned long __user *)arg2); |
6612538c | 642 | break; |
1da177e4 | 643 | } |
6612538c | 644 | case ARCH_GET_GS: { |
1da177e4 | 645 | unsigned long base; |
17a6e1b8 | 646 | |
d47b50e7 AL |
647 | if (doit) |
648 | rdmsrl(MSR_KERNEL_GS_BASE, base); | |
d47b50e7 | 649 | else |
296f781a | 650 | base = task->thread.gsbase; |
17a6e1b8 | 651 | ret = put_user(base, (unsigned long __user *)arg2); |
1da177e4 LT |
652 | break; |
653 | } | |
654 | ||
2eefd878 | 655 | #ifdef CONFIG_CHECKPOINT_RESTORE |
6e68b087 | 656 | # ifdef CONFIG_X86_X32_ABI |
2eefd878 | 657 | case ARCH_MAP_VDSO_X32: |
17a6e1b8 | 658 | return prctl_map_vdso(&vdso_image_x32, arg2); |
91b7bd39 IM |
659 | # endif |
660 | # if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION | |
2eefd878 | 661 | case ARCH_MAP_VDSO_32: |
17a6e1b8 | 662 | return prctl_map_vdso(&vdso_image_32, arg2); |
91b7bd39 | 663 | # endif |
2eefd878 | 664 | case ARCH_MAP_VDSO_64: |
17a6e1b8 | 665 | return prctl_map_vdso(&vdso_image_64, arg2); |
2eefd878 DS |
666 | #endif |
667 | ||
1da177e4 LT |
668 | default: |
669 | ret = -EINVAL; | |
670 | break; | |
6612538c | 671 | } |
1da177e4 | 672 | |
6612538c HS |
673 | return ret; |
674 | } | |
1da177e4 | 675 | |
17a6e1b8 | 676 | SYSCALL_DEFINE2(arch_prctl, int, option, unsigned long, arg2) |
1da177e4 | 677 | { |
b0b9b014 KH |
678 | long ret; |
679 | ||
680 | ret = do_arch_prctl_64(current, option, arg2); | |
681 | if (ret == -EINVAL) | |
682 | ret = do_arch_prctl_common(current, option, arg2); | |
683 | ||
684 | return ret; | |
1da177e4 LT |
685 | } |
686 | ||
79170fda KH |
687 | #ifdef CONFIG_IA32_EMULATION |
688 | COMPAT_SYSCALL_DEFINE2(arch_prctl, int, option, unsigned long, arg2) | |
689 | { | |
690 | return do_arch_prctl_common(current, option, arg2); | |
691 | } | |
692 | #endif | |
693 | ||
89240ba0 SS |
694 | unsigned long KSTK_ESP(struct task_struct *task) |
695 | { | |
263042e4 | 696 | return task_pt_regs(task)->sp; |
89240ba0 | 697 | } |