Loading...
1/*
2 * Copyright (C) 1995 Linus Torvalds
3 *
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
6 *
7 * X86-64 port
8 * Andi Kleen.
9 *
10 * CPU hotplug support - ashok.raj@intel.com
11 */
12
13/*
14 * This file handles the architecture-dependent parts of process handling..
15 */
16
17#include <linux/cpu.h>
18#include <linux/errno.h>
19#include <linux/sched.h>
20#include <linux/fs.h>
21#include <linux/kernel.h>
22#include <linux/mm.h>
23#include <linux/elfcore.h>
24#include <linux/smp.h>
25#include <linux/slab.h>
26#include <linux/user.h>
27#include <linux/interrupt.h>
28#include <linux/delay.h>
29#include <linux/module.h>
30#include <linux/ptrace.h>
31#include <linux/notifier.h>
32#include <linux/kprobes.h>
33#include <linux/kdebug.h>
34#include <linux/prctl.h>
35#include <linux/uaccess.h>
36#include <linux/io.h>
37#include <linux/ftrace.h>
38
39#include <asm/pgtable.h>
40#include <asm/processor.h>
41#include <asm/i387.h>
42#include <asm/fpu-internal.h>
43#include <asm/mmu_context.h>
44#include <asm/prctl.h>
45#include <asm/desc.h>
46#include <asm/proto.h>
47#include <asm/ia32.h>
48#include <asm/idle.h>
49#include <asm/syscalls.h>
50#include <asm/debugreg.h>
51#include <asm/switch_to.h>
52
53asmlinkage extern void ret_from_fork(void);
54
55__visible DEFINE_PER_CPU(unsigned long, old_rsp);
56
57/* Prints also some state that isn't saved in the pt_regs */
58void __show_regs(struct pt_regs *regs, int all)
59{
60 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
61 unsigned long d0, d1, d2, d3, d6, d7;
62 unsigned int fsindex, gsindex;
63 unsigned int ds, cs, es;
64
65 printk(KERN_DEFAULT "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
66 printk_address(regs->ip);
67 printk(KERN_DEFAULT "RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss,
68 regs->sp, regs->flags);
69 printk(KERN_DEFAULT "RAX: %016lx RBX: %016lx RCX: %016lx\n",
70 regs->ax, regs->bx, regs->cx);
71 printk(KERN_DEFAULT "RDX: %016lx RSI: %016lx RDI: %016lx\n",
72 regs->dx, regs->si, regs->di);
73 printk(KERN_DEFAULT "RBP: %016lx R08: %016lx R09: %016lx\n",
74 regs->bp, regs->r8, regs->r9);
75 printk(KERN_DEFAULT "R10: %016lx R11: %016lx R12: %016lx\n",
76 regs->r10, regs->r11, regs->r12);
77 printk(KERN_DEFAULT "R13: %016lx R14: %016lx R15: %016lx\n",
78 regs->r13, regs->r14, regs->r15);
79
80 asm("movl %%ds,%0" : "=r" (ds));
81 asm("movl %%cs,%0" : "=r" (cs));
82 asm("movl %%es,%0" : "=r" (es));
83 asm("movl %%fs,%0" : "=r" (fsindex));
84 asm("movl %%gs,%0" : "=r" (gsindex));
85
86 rdmsrl(MSR_FS_BASE, fs);
87 rdmsrl(MSR_GS_BASE, gs);
88 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
89
90 if (!all)
91 return;
92
93 cr0 = read_cr0();
94 cr2 = read_cr2();
95 cr3 = read_cr3();
96 cr4 = read_cr4();
97
98 printk(KERN_DEFAULT "FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
99 fs, fsindex, gs, gsindex, shadowgs);
100 printk(KERN_DEFAULT "CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
101 es, cr0);
102 printk(KERN_DEFAULT "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
103 cr4);
104
105 get_debugreg(d0, 0);
106 get_debugreg(d1, 1);
107 get_debugreg(d2, 2);
108 get_debugreg(d3, 3);
109 get_debugreg(d6, 6);
110 get_debugreg(d7, 7);
111
112 /* Only print out debug registers if they are in their non-default state. */
113 if ((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
114 (d6 == DR6_RESERVED) && (d7 == 0x400))
115 return;
116
117 printk(KERN_DEFAULT "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
118 printk(KERN_DEFAULT "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
119
120}
121
122void release_thread(struct task_struct *dead_task)
123{
124 if (dead_task->mm) {
125 if (dead_task->mm->context.size) {
126 pr_warn("WARNING: dead process %s still has LDT? <%p/%d>\n",
127 dead_task->comm,
128 dead_task->mm->context.ldt,
129 dead_task->mm->context.size);
130 BUG();
131 }
132 }
133}
134
135static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
136{
137 struct user_desc ud = {
138 .base_addr = addr,
139 .limit = 0xfffff,
140 .seg_32bit = 1,
141 .limit_in_pages = 1,
142 .useable = 1,
143 };
144 struct desc_struct *desc = t->thread.tls_array;
145 desc += tls;
146 fill_ldt(desc, &ud);
147}
148
149static inline u32 read_32bit_tls(struct task_struct *t, int tls)
150{
151 return get_desc_base(&t->thread.tls_array[tls]);
152}
153
154int copy_thread(unsigned long clone_flags, unsigned long sp,
155 unsigned long arg, struct task_struct *p)
156{
157 int err;
158 struct pt_regs *childregs;
159 struct task_struct *me = current;
160
161 p->thread.sp0 = (unsigned long)task_stack_page(p) + THREAD_SIZE;
162 childregs = task_pt_regs(p);
163 p->thread.sp = (unsigned long) childregs;
164 p->thread.usersp = me->thread.usersp;
165 set_tsk_thread_flag(p, TIF_FORK);
166 p->thread.fpu_counter = 0;
167 p->thread.io_bitmap_ptr = NULL;
168
169 savesegment(gs, p->thread.gsindex);
170 p->thread.gs = p->thread.gsindex ? 0 : me->thread.gs;
171 savesegment(fs, p->thread.fsindex);
172 p->thread.fs = p->thread.fsindex ? 0 : me->thread.fs;
173 savesegment(es, p->thread.es);
174 savesegment(ds, p->thread.ds);
175 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
176
177 if (unlikely(p->flags & PF_KTHREAD)) {
178 /* kernel thread */
179 memset(childregs, 0, sizeof(struct pt_regs));
180 childregs->sp = (unsigned long)childregs;
181 childregs->ss = __KERNEL_DS;
182 childregs->bx = sp; /* function */
183 childregs->bp = arg;
184 childregs->orig_ax = -1;
185 childregs->cs = __KERNEL_CS | get_kernel_rpl();
186 childregs->flags = X86_EFLAGS_IF | X86_EFLAGS_FIXED;
187 return 0;
188 }
189 *childregs = *current_pt_regs();
190
191 childregs->ax = 0;
192 if (sp)
193 childregs->sp = sp;
194
195 err = -ENOMEM;
196 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
197
198 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
199 p->thread.io_bitmap_ptr = kmemdup(me->thread.io_bitmap_ptr,
200 IO_BITMAP_BYTES, GFP_KERNEL);
201 if (!p->thread.io_bitmap_ptr) {
202 p->thread.io_bitmap_max = 0;
203 return -ENOMEM;
204 }
205 set_tsk_thread_flag(p, TIF_IO_BITMAP);
206 }
207
208 /*
209 * Set a new TLS for the child thread?
210 */
211 if (clone_flags & CLONE_SETTLS) {
212#ifdef CONFIG_IA32_EMULATION
213 if (test_thread_flag(TIF_IA32))
214 err = do_set_thread_area(p, -1,
215 (struct user_desc __user *)childregs->si, 0);
216 else
217#endif
218 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
219 if (err)
220 goto out;
221 }
222 err = 0;
223out:
224 if (err && p->thread.io_bitmap_ptr) {
225 kfree(p->thread.io_bitmap_ptr);
226 p->thread.io_bitmap_max = 0;
227 }
228
229 return err;
230}
231
232static void
233start_thread_common(struct pt_regs *regs, unsigned long new_ip,
234 unsigned long new_sp,
235 unsigned int _cs, unsigned int _ss, unsigned int _ds)
236{
237 loadsegment(fs, 0);
238 loadsegment(es, _ds);
239 loadsegment(ds, _ds);
240 load_gs_index(0);
241 current->thread.usersp = new_sp;
242 regs->ip = new_ip;
243 regs->sp = new_sp;
244 this_cpu_write(old_rsp, new_sp);
245 regs->cs = _cs;
246 regs->ss = _ss;
247 regs->flags = X86_EFLAGS_IF;
248}
249
250void
251start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
252{
253 start_thread_common(regs, new_ip, new_sp,
254 __USER_CS, __USER_DS, 0);
255}
256
257#ifdef CONFIG_IA32_EMULATION
258void start_thread_ia32(struct pt_regs *regs, u32 new_ip, u32 new_sp)
259{
260 start_thread_common(regs, new_ip, new_sp,
261 test_thread_flag(TIF_X32)
262 ? __USER_CS : __USER32_CS,
263 __USER_DS, __USER_DS);
264}
265#endif
266
267/*
268 * switch_to(x,y) should switch tasks from x to y.
269 *
270 * This could still be optimized:
271 * - fold all the options into a flag word and test it with a single test.
272 * - could test fs/gs bitsliced
273 *
274 * Kprobes not supported here. Set the probe on schedule instead.
275 * Function graph tracer not supported too.
276 */
277__visible __notrace_funcgraph struct task_struct *
278__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
279{
280 struct thread_struct *prev = &prev_p->thread;
281 struct thread_struct *next = &next_p->thread;
282 int cpu = smp_processor_id();
283 struct tss_struct *tss = &per_cpu(init_tss, cpu);
284 unsigned fsindex, gsindex;
285 fpu_switch_t fpu;
286
287 fpu = switch_fpu_prepare(prev_p, next_p, cpu);
288
289 /*
290 * Reload esp0, LDT and the page table pointer:
291 */
292 load_sp0(tss, next);
293
294 /*
295 * Switch DS and ES.
296 * This won't pick up thread selector changes, but I guess that is ok.
297 */
298 savesegment(es, prev->es);
299 if (unlikely(next->es | prev->es))
300 loadsegment(es, next->es);
301
302 savesegment(ds, prev->ds);
303 if (unlikely(next->ds | prev->ds))
304 loadsegment(ds, next->ds);
305
306
307 /* We must save %fs and %gs before load_TLS() because
308 * %fs and %gs may be cleared by load_TLS().
309 *
310 * (e.g. xen_load_tls())
311 */
312 savesegment(fs, fsindex);
313 savesegment(gs, gsindex);
314
315 load_TLS(next, cpu);
316
317 /*
318 * Leave lazy mode, flushing any hypercalls made here.
319 * This must be done before restoring TLS segments so
320 * the GDT and LDT are properly updated, and must be
321 * done before math_state_restore, so the TS bit is up
322 * to date.
323 */
324 arch_end_context_switch(next_p);
325
326 /*
327 * Switch FS and GS.
328 *
329 * Segment register != 0 always requires a reload. Also
330 * reload when it has changed. When prev process used 64bit
331 * base always reload to avoid an information leak.
332 */
333 if (unlikely(fsindex | next->fsindex | prev->fs)) {
334 loadsegment(fs, next->fsindex);
335 /*
336 * Check if the user used a selector != 0; if yes
337 * clear 64bit base, since overloaded base is always
338 * mapped to the Null selector
339 */
340 if (fsindex)
341 prev->fs = 0;
342 }
343 /* when next process has a 64bit base use it */
344 if (next->fs)
345 wrmsrl(MSR_FS_BASE, next->fs);
346 prev->fsindex = fsindex;
347
348 if (unlikely(gsindex | next->gsindex | prev->gs)) {
349 load_gs_index(next->gsindex);
350 if (gsindex)
351 prev->gs = 0;
352 }
353 if (next->gs)
354 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
355 prev->gsindex = gsindex;
356
357 switch_fpu_finish(next_p, fpu);
358
359 /*
360 * Switch the PDA and FPU contexts.
361 */
362 prev->usersp = this_cpu_read(old_rsp);
363 this_cpu_write(old_rsp, next->usersp);
364 this_cpu_write(current_task, next_p);
365
366 /*
367 * If it were not for PREEMPT_ACTIVE we could guarantee that the
368 * preempt_count of all tasks was equal here and this would not be
369 * needed.
370 */
371 task_thread_info(prev_p)->saved_preempt_count = this_cpu_read(__preempt_count);
372 this_cpu_write(__preempt_count, task_thread_info(next_p)->saved_preempt_count);
373
374 this_cpu_write(kernel_stack,
375 (unsigned long)task_stack_page(next_p) +
376 THREAD_SIZE - KERNEL_STACK_OFFSET);
377
378 /*
379 * Now maybe reload the debug registers and handle I/O bitmaps
380 */
381 if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
382 task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
383 __switch_to_xtra(prev_p, next_p, tss);
384
385 return prev_p;
386}
387
388void set_personality_64bit(void)
389{
390 /* inherit personality from parent */
391
392 /* Make sure to be in 64bit mode */
393 clear_thread_flag(TIF_IA32);
394 clear_thread_flag(TIF_ADDR32);
395 clear_thread_flag(TIF_X32);
396
397 /* Ensure the corresponding mm is not marked. */
398 if (current->mm)
399 current->mm->context.ia32_compat = 0;
400
401 /* TBD: overwrites user setup. Should have two bits.
402 But 64bit processes have always behaved this way,
403 so it's not too bad. The main problem is just that
404 32bit childs are affected again. */
405 current->personality &= ~READ_IMPLIES_EXEC;
406}
407
408void set_personality_ia32(bool x32)
409{
410 /* inherit personality from parent */
411
412 /* Make sure to be in 32bit mode */
413 set_thread_flag(TIF_ADDR32);
414
415 /* Mark the associated mm as containing 32-bit tasks. */
416 if (current->mm)
417 current->mm->context.ia32_compat = 1;
418
419 if (x32) {
420 clear_thread_flag(TIF_IA32);
421 set_thread_flag(TIF_X32);
422 current->personality &= ~READ_IMPLIES_EXEC;
423 /* is_compat_task() uses the presence of the x32
424 syscall bit flag to determine compat status */
425 current_thread_info()->status &= ~TS_COMPAT;
426 } else {
427 set_thread_flag(TIF_IA32);
428 clear_thread_flag(TIF_X32);
429 current->personality |= force_personality32;
430 /* Prepare the first "return" to user space */
431 current_thread_info()->status |= TS_COMPAT;
432 }
433}
434EXPORT_SYMBOL_GPL(set_personality_ia32);
435
436unsigned long get_wchan(struct task_struct *p)
437{
438 unsigned long stack;
439 u64 fp, ip;
440 int count = 0;
441
442 if (!p || p == current || p->state == TASK_RUNNING)
443 return 0;
444 stack = (unsigned long)task_stack_page(p);
445 if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
446 return 0;
447 fp = *(u64 *)(p->thread.sp);
448 do {
449 if (fp < (unsigned long)stack ||
450 fp >= (unsigned long)stack+THREAD_SIZE)
451 return 0;
452 ip = *(u64 *)(fp+8);
453 if (!in_sched_functions(ip))
454 return ip;
455 fp = *(u64 *)fp;
456 } while (count++ < 16);
457 return 0;
458}
459
460long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
461{
462 int ret = 0;
463 int doit = task == current;
464 int cpu;
465
466 switch (code) {
467 case ARCH_SET_GS:
468 if (addr >= TASK_SIZE_OF(task))
469 return -EPERM;
470 cpu = get_cpu();
471 /* handle small bases via the GDT because that's faster to
472 switch. */
473 if (addr <= 0xffffffff) {
474 set_32bit_tls(task, GS_TLS, addr);
475 if (doit) {
476 load_TLS(&task->thread, cpu);
477 load_gs_index(GS_TLS_SEL);
478 }
479 task->thread.gsindex = GS_TLS_SEL;
480 task->thread.gs = 0;
481 } else {
482 task->thread.gsindex = 0;
483 task->thread.gs = addr;
484 if (doit) {
485 load_gs_index(0);
486 ret = wrmsrl_safe(MSR_KERNEL_GS_BASE, addr);
487 }
488 }
489 put_cpu();
490 break;
491 case ARCH_SET_FS:
492 /* Not strictly needed for fs, but do it for symmetry
493 with gs */
494 if (addr >= TASK_SIZE_OF(task))
495 return -EPERM;
496 cpu = get_cpu();
497 /* handle small bases via the GDT because that's faster to
498 switch. */
499 if (addr <= 0xffffffff) {
500 set_32bit_tls(task, FS_TLS, addr);
501 if (doit) {
502 load_TLS(&task->thread, cpu);
503 loadsegment(fs, FS_TLS_SEL);
504 }
505 task->thread.fsindex = FS_TLS_SEL;
506 task->thread.fs = 0;
507 } else {
508 task->thread.fsindex = 0;
509 task->thread.fs = addr;
510 if (doit) {
511 /* set the selector to 0 to not confuse
512 __switch_to */
513 loadsegment(fs, 0);
514 ret = wrmsrl_safe(MSR_FS_BASE, addr);
515 }
516 }
517 put_cpu();
518 break;
519 case ARCH_GET_FS: {
520 unsigned long base;
521 if (task->thread.fsindex == FS_TLS_SEL)
522 base = read_32bit_tls(task, FS_TLS);
523 else if (doit)
524 rdmsrl(MSR_FS_BASE, base);
525 else
526 base = task->thread.fs;
527 ret = put_user(base, (unsigned long __user *)addr);
528 break;
529 }
530 case ARCH_GET_GS: {
531 unsigned long base;
532 unsigned gsindex;
533 if (task->thread.gsindex == GS_TLS_SEL)
534 base = read_32bit_tls(task, GS_TLS);
535 else if (doit) {
536 savesegment(gs, gsindex);
537 if (gsindex)
538 rdmsrl(MSR_KERNEL_GS_BASE, base);
539 else
540 base = task->thread.gs;
541 } else
542 base = task->thread.gs;
543 ret = put_user(base, (unsigned long __user *)addr);
544 break;
545 }
546
547 default:
548 ret = -EINVAL;
549 break;
550 }
551
552 return ret;
553}
554
555long sys_arch_prctl(int code, unsigned long addr)
556{
557 return do_arch_prctl(current, code, addr);
558}
559
560unsigned long KSTK_ESP(struct task_struct *task)
561{
562 return (test_tsk_thread_flag(task, TIF_IA32)) ?
563 (task_pt_regs(task)->sp) : ((task)->thread.usersp);
564}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 1995 Linus Torvalds
4 *
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
7 *
8 * X86-64 port
9 * Andi Kleen.
10 *
11 * CPU hotplug support - ashok.raj@intel.com
12 */
13
14/*
15 * This file handles the architecture-dependent parts of process handling..
16 */
17
18#include <linux/cpu.h>
19#include <linux/errno.h>
20#include <linux/sched.h>
21#include <linux/sched/task.h>
22#include <linux/sched/task_stack.h>
23#include <linux/fs.h>
24#include <linux/kernel.h>
25#include <linux/mm.h>
26#include <linux/elfcore.h>
27#include <linux/smp.h>
28#include <linux/slab.h>
29#include <linux/user.h>
30#include <linux/interrupt.h>
31#include <linux/delay.h>
32#include <linux/export.h>
33#include <linux/ptrace.h>
34#include <linux/notifier.h>
35#include <linux/kprobes.h>
36#include <linux/kdebug.h>
37#include <linux/prctl.h>
38#include <linux/uaccess.h>
39#include <linux/io.h>
40#include <linux/ftrace.h>
41#include <linux/syscalls.h>
42
43#include <asm/processor.h>
44#include <asm/fpu/internal.h>
45#include <asm/mmu_context.h>
46#include <asm/prctl.h>
47#include <asm/desc.h>
48#include <asm/proto.h>
49#include <asm/ia32.h>
50#include <asm/debugreg.h>
51#include <asm/switch_to.h>
52#include <asm/xen/hypervisor.h>
53#include <asm/vdso.h>
54#include <asm/resctrl.h>
55#include <asm/unistd.h>
56#include <asm/fsgsbase.h>
57#ifdef CONFIG_IA32_EMULATION
58/* Not included via unistd.h */
59#include <asm/unistd_32_ia32.h>
60#endif
61
62#include "process.h"
63
64/* Prints also some state that isn't saved in the pt_regs */
65void __show_regs(struct pt_regs *regs, enum show_regs_mode mode,
66 const char *log_lvl)
67{
68 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
69 unsigned long d0, d1, d2, d3, d6, d7;
70 unsigned int fsindex, gsindex;
71 unsigned int ds, es;
72
73 show_iret_regs(regs, log_lvl);
74
75 if (regs->orig_ax != -1)
76 pr_cont(" ORIG_RAX: %016lx\n", regs->orig_ax);
77 else
78 pr_cont("\n");
79
80 printk("%sRAX: %016lx RBX: %016lx RCX: %016lx\n",
81 log_lvl, regs->ax, regs->bx, regs->cx);
82 printk("%sRDX: %016lx RSI: %016lx RDI: %016lx\n",
83 log_lvl, regs->dx, regs->si, regs->di);
84 printk("%sRBP: %016lx R08: %016lx R09: %016lx\n",
85 log_lvl, regs->bp, regs->r8, regs->r9);
86 printk("%sR10: %016lx R11: %016lx R12: %016lx\n",
87 log_lvl, regs->r10, regs->r11, regs->r12);
88 printk("%sR13: %016lx R14: %016lx R15: %016lx\n",
89 log_lvl, regs->r13, regs->r14, regs->r15);
90
91 if (mode == SHOW_REGS_SHORT)
92 return;
93
94 if (mode == SHOW_REGS_USER) {
95 rdmsrl(MSR_FS_BASE, fs);
96 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
97 printk("%sFS: %016lx GS: %016lx\n",
98 log_lvl, fs, shadowgs);
99 return;
100 }
101
102 asm("movl %%ds,%0" : "=r" (ds));
103 asm("movl %%es,%0" : "=r" (es));
104 asm("movl %%fs,%0" : "=r" (fsindex));
105 asm("movl %%gs,%0" : "=r" (gsindex));
106
107 rdmsrl(MSR_FS_BASE, fs);
108 rdmsrl(MSR_GS_BASE, gs);
109 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
110
111 cr0 = read_cr0();
112 cr2 = read_cr2();
113 cr3 = __read_cr3();
114 cr4 = __read_cr4();
115
116 printk("%sFS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
117 log_lvl, fs, fsindex, gs, gsindex, shadowgs);
118 printk("%sCS: %04lx DS: %04x ES: %04x CR0: %016lx\n",
119 log_lvl, regs->cs, ds, es, cr0);
120 printk("%sCR2: %016lx CR3: %016lx CR4: %016lx\n",
121 log_lvl, cr2, cr3, cr4);
122
123 get_debugreg(d0, 0);
124 get_debugreg(d1, 1);
125 get_debugreg(d2, 2);
126 get_debugreg(d3, 3);
127 get_debugreg(d6, 6);
128 get_debugreg(d7, 7);
129
130 /* Only print out debug registers if they are in their non-default state. */
131 if (!((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
132 (d6 == DR6_RESERVED) && (d7 == 0x400))) {
133 printk("%sDR0: %016lx DR1: %016lx DR2: %016lx\n",
134 log_lvl, d0, d1, d2);
135 printk("%sDR3: %016lx DR6: %016lx DR7: %016lx\n",
136 log_lvl, d3, d6, d7);
137 }
138
139 if (boot_cpu_has(X86_FEATURE_OSPKE))
140 printk("%sPKRU: %08x\n", log_lvl, read_pkru());
141}
142
143void release_thread(struct task_struct *dead_task)
144{
145 WARN_ON(dead_task->mm);
146}
147
148enum which_selector {
149 FS,
150 GS
151};
152
153/*
154 * Out of line to be protected from kprobes and tracing. If this would be
155 * traced or probed than any access to a per CPU variable happens with
156 * the wrong GS.
157 *
158 * It is not used on Xen paravirt. When paravirt support is needed, it
159 * needs to be renamed with native_ prefix.
160 */
161static noinstr unsigned long __rdgsbase_inactive(void)
162{
163 unsigned long gsbase;
164
165 lockdep_assert_irqs_disabled();
166
167 if (!static_cpu_has(X86_FEATURE_XENPV)) {
168 native_swapgs();
169 gsbase = rdgsbase();
170 native_swapgs();
171 } else {
172 instrumentation_begin();
173 rdmsrl(MSR_KERNEL_GS_BASE, gsbase);
174 instrumentation_end();
175 }
176
177 return gsbase;
178}
179
180/*
181 * Out of line to be protected from kprobes and tracing. If this would be
182 * traced or probed than any access to a per CPU variable happens with
183 * the wrong GS.
184 *
185 * It is not used on Xen paravirt. When paravirt support is needed, it
186 * needs to be renamed with native_ prefix.
187 */
188static noinstr void __wrgsbase_inactive(unsigned long gsbase)
189{
190 lockdep_assert_irqs_disabled();
191
192 if (!static_cpu_has(X86_FEATURE_XENPV)) {
193 native_swapgs();
194 wrgsbase(gsbase);
195 native_swapgs();
196 } else {
197 instrumentation_begin();
198 wrmsrl(MSR_KERNEL_GS_BASE, gsbase);
199 instrumentation_end();
200 }
201}
202
203/*
204 * Saves the FS or GS base for an outgoing thread if FSGSBASE extensions are
205 * not available. The goal is to be reasonably fast on non-FSGSBASE systems.
206 * It's forcibly inlined because it'll generate better code and this function
207 * is hot.
208 */
209static __always_inline void save_base_legacy(struct task_struct *prev_p,
210 unsigned short selector,
211 enum which_selector which)
212{
213 if (likely(selector == 0)) {
214 /*
215 * On Intel (without X86_BUG_NULL_SEG), the segment base could
216 * be the pre-existing saved base or it could be zero. On AMD
217 * (with X86_BUG_NULL_SEG), the segment base could be almost
218 * anything.
219 *
220 * This branch is very hot (it's hit twice on almost every
221 * context switch between 64-bit programs), and avoiding
222 * the RDMSR helps a lot, so we just assume that whatever
223 * value is already saved is correct. This matches historical
224 * Linux behavior, so it won't break existing applications.
225 *
226 * To avoid leaking state, on non-X86_BUG_NULL_SEG CPUs, if we
227 * report that the base is zero, it needs to actually be zero:
228 * see the corresponding logic in load_seg_legacy.
229 */
230 } else {
231 /*
232 * If the selector is 1, 2, or 3, then the base is zero on
233 * !X86_BUG_NULL_SEG CPUs and could be anything on
234 * X86_BUG_NULL_SEG CPUs. In the latter case, Linux
235 * has never attempted to preserve the base across context
236 * switches.
237 *
238 * If selector > 3, then it refers to a real segment, and
239 * saving the base isn't necessary.
240 */
241 if (which == FS)
242 prev_p->thread.fsbase = 0;
243 else
244 prev_p->thread.gsbase = 0;
245 }
246}
247
248static __always_inline void save_fsgs(struct task_struct *task)
249{
250 savesegment(fs, task->thread.fsindex);
251 savesegment(gs, task->thread.gsindex);
252 if (static_cpu_has(X86_FEATURE_FSGSBASE)) {
253 /*
254 * If FSGSBASE is enabled, we can't make any useful guesses
255 * about the base, and user code expects us to save the current
256 * value. Fortunately, reading the base directly is efficient.
257 */
258 task->thread.fsbase = rdfsbase();
259 task->thread.gsbase = __rdgsbase_inactive();
260 } else {
261 save_base_legacy(task, task->thread.fsindex, FS);
262 save_base_legacy(task, task->thread.gsindex, GS);
263 }
264}
265
266/*
267 * While a process is running,current->thread.fsbase and current->thread.gsbase
268 * may not match the corresponding CPU registers (see save_base_legacy()).
269 */
270void current_save_fsgs(void)
271{
272 unsigned long flags;
273
274 /* Interrupts need to be off for FSGSBASE */
275 local_irq_save(flags);
276 save_fsgs(current);
277 local_irq_restore(flags);
278}
279#if IS_ENABLED(CONFIG_KVM)
280EXPORT_SYMBOL_GPL(current_save_fsgs);
281#endif
282
283static __always_inline void loadseg(enum which_selector which,
284 unsigned short sel)
285{
286 if (which == FS)
287 loadsegment(fs, sel);
288 else
289 load_gs_index(sel);
290}
291
292static __always_inline void load_seg_legacy(unsigned short prev_index,
293 unsigned long prev_base,
294 unsigned short next_index,
295 unsigned long next_base,
296 enum which_selector which)
297{
298 if (likely(next_index <= 3)) {
299 /*
300 * The next task is using 64-bit TLS, is not using this
301 * segment at all, or is having fun with arcane CPU features.
302 */
303 if (next_base == 0) {
304 /*
305 * Nasty case: on AMD CPUs, we need to forcibly zero
306 * the base.
307 */
308 if (static_cpu_has_bug(X86_BUG_NULL_SEG)) {
309 loadseg(which, __USER_DS);
310 loadseg(which, next_index);
311 } else {
312 /*
313 * We could try to exhaustively detect cases
314 * under which we can skip the segment load,
315 * but there's really only one case that matters
316 * for performance: if both the previous and
317 * next states are fully zeroed, we can skip
318 * the load.
319 *
320 * (This assumes that prev_base == 0 has no
321 * false positives. This is the case on
322 * Intel-style CPUs.)
323 */
324 if (likely(prev_index | next_index | prev_base))
325 loadseg(which, next_index);
326 }
327 } else {
328 if (prev_index != next_index)
329 loadseg(which, next_index);
330 wrmsrl(which == FS ? MSR_FS_BASE : MSR_KERNEL_GS_BASE,
331 next_base);
332 }
333 } else {
334 /*
335 * The next task is using a real segment. Loading the selector
336 * is sufficient.
337 */
338 loadseg(which, next_index);
339 }
340}
341
342static __always_inline void x86_fsgsbase_load(struct thread_struct *prev,
343 struct thread_struct *next)
344{
345 if (static_cpu_has(X86_FEATURE_FSGSBASE)) {
346 /* Update the FS and GS selectors if they could have changed. */
347 if (unlikely(prev->fsindex || next->fsindex))
348 loadseg(FS, next->fsindex);
349 if (unlikely(prev->gsindex || next->gsindex))
350 loadseg(GS, next->gsindex);
351
352 /* Update the bases. */
353 wrfsbase(next->fsbase);
354 __wrgsbase_inactive(next->gsbase);
355 } else {
356 load_seg_legacy(prev->fsindex, prev->fsbase,
357 next->fsindex, next->fsbase, FS);
358 load_seg_legacy(prev->gsindex, prev->gsbase,
359 next->gsindex, next->gsbase, GS);
360 }
361}
362
363unsigned long x86_fsgsbase_read_task(struct task_struct *task,
364 unsigned short selector)
365{
366 unsigned short idx = selector >> 3;
367 unsigned long base;
368
369 if (likely((selector & SEGMENT_TI_MASK) == 0)) {
370 if (unlikely(idx >= GDT_ENTRIES))
371 return 0;
372
373 /*
374 * There are no user segments in the GDT with nonzero bases
375 * other than the TLS segments.
376 */
377 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
378 return 0;
379
380 idx -= GDT_ENTRY_TLS_MIN;
381 base = get_desc_base(&task->thread.tls_array[idx]);
382 } else {
383#ifdef CONFIG_MODIFY_LDT_SYSCALL
384 struct ldt_struct *ldt;
385
386 /*
387 * If performance here mattered, we could protect the LDT
388 * with RCU. This is a slow path, though, so we can just
389 * take the mutex.
390 */
391 mutex_lock(&task->mm->context.lock);
392 ldt = task->mm->context.ldt;
393 if (unlikely(!ldt || idx >= ldt->nr_entries))
394 base = 0;
395 else
396 base = get_desc_base(ldt->entries + idx);
397 mutex_unlock(&task->mm->context.lock);
398#else
399 base = 0;
400#endif
401 }
402
403 return base;
404}
405
406unsigned long x86_gsbase_read_cpu_inactive(void)
407{
408 unsigned long gsbase;
409
410 if (static_cpu_has(X86_FEATURE_FSGSBASE)) {
411 unsigned long flags;
412
413 local_irq_save(flags);
414 gsbase = __rdgsbase_inactive();
415 local_irq_restore(flags);
416 } else {
417 rdmsrl(MSR_KERNEL_GS_BASE, gsbase);
418 }
419
420 return gsbase;
421}
422
423void x86_gsbase_write_cpu_inactive(unsigned long gsbase)
424{
425 if (static_cpu_has(X86_FEATURE_FSGSBASE)) {
426 unsigned long flags;
427
428 local_irq_save(flags);
429 __wrgsbase_inactive(gsbase);
430 local_irq_restore(flags);
431 } else {
432 wrmsrl(MSR_KERNEL_GS_BASE, gsbase);
433 }
434}
435
436unsigned long x86_fsbase_read_task(struct task_struct *task)
437{
438 unsigned long fsbase;
439
440 if (task == current)
441 fsbase = x86_fsbase_read_cpu();
442 else if (static_cpu_has(X86_FEATURE_FSGSBASE) ||
443 (task->thread.fsindex == 0))
444 fsbase = task->thread.fsbase;
445 else
446 fsbase = x86_fsgsbase_read_task(task, task->thread.fsindex);
447
448 return fsbase;
449}
450
451unsigned long x86_gsbase_read_task(struct task_struct *task)
452{
453 unsigned long gsbase;
454
455 if (task == current)
456 gsbase = x86_gsbase_read_cpu_inactive();
457 else if (static_cpu_has(X86_FEATURE_FSGSBASE) ||
458 (task->thread.gsindex == 0))
459 gsbase = task->thread.gsbase;
460 else
461 gsbase = x86_fsgsbase_read_task(task, task->thread.gsindex);
462
463 return gsbase;
464}
465
466void x86_fsbase_write_task(struct task_struct *task, unsigned long fsbase)
467{
468 WARN_ON_ONCE(task == current);
469
470 task->thread.fsbase = fsbase;
471}
472
473void x86_gsbase_write_task(struct task_struct *task, unsigned long gsbase)
474{
475 WARN_ON_ONCE(task == current);
476
477 task->thread.gsbase = gsbase;
478}
479
480static void
481start_thread_common(struct pt_regs *regs, unsigned long new_ip,
482 unsigned long new_sp,
483 unsigned int _cs, unsigned int _ss, unsigned int _ds)
484{
485 WARN_ON_ONCE(regs != current_pt_regs());
486
487 if (static_cpu_has(X86_BUG_NULL_SEG)) {
488 /* Loading zero below won't clear the base. */
489 loadsegment(fs, __USER_DS);
490 load_gs_index(__USER_DS);
491 }
492
493 loadsegment(fs, 0);
494 loadsegment(es, _ds);
495 loadsegment(ds, _ds);
496 load_gs_index(0);
497
498 regs->ip = new_ip;
499 regs->sp = new_sp;
500 regs->cs = _cs;
501 regs->ss = _ss;
502 regs->flags = X86_EFLAGS_IF;
503}
504
505void
506start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
507{
508 start_thread_common(regs, new_ip, new_sp,
509 __USER_CS, __USER_DS, 0);
510}
511EXPORT_SYMBOL_GPL(start_thread);
512
513#ifdef CONFIG_COMPAT
514void compat_start_thread(struct pt_regs *regs, u32 new_ip, u32 new_sp)
515{
516 start_thread_common(regs, new_ip, new_sp,
517 test_thread_flag(TIF_X32)
518 ? __USER_CS : __USER32_CS,
519 __USER_DS, __USER_DS);
520}
521#endif
522
523/*
524 * switch_to(x,y) should switch tasks from x to y.
525 *
526 * This could still be optimized:
527 * - fold all the options into a flag word and test it with a single test.
528 * - could test fs/gs bitsliced
529 *
530 * Kprobes not supported here. Set the probe on schedule instead.
531 * Function graph tracer not supported too.
532 */
533__visible __notrace_funcgraph struct task_struct *
534__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
535{
536 struct thread_struct *prev = &prev_p->thread;
537 struct thread_struct *next = &next_p->thread;
538 struct fpu *prev_fpu = &prev->fpu;
539 struct fpu *next_fpu = &next->fpu;
540 int cpu = smp_processor_id();
541
542 WARN_ON_ONCE(IS_ENABLED(CONFIG_DEBUG_ENTRY) &&
543 this_cpu_read(irq_count) != -1);
544
545 if (!test_thread_flag(TIF_NEED_FPU_LOAD))
546 switch_fpu_prepare(prev_fpu, cpu);
547
548 /* We must save %fs and %gs before load_TLS() because
549 * %fs and %gs may be cleared by load_TLS().
550 *
551 * (e.g. xen_load_tls())
552 */
553 save_fsgs(prev_p);
554
555 /*
556 * Load TLS before restoring any segments so that segment loads
557 * reference the correct GDT entries.
558 */
559 load_TLS(next, cpu);
560
561 /*
562 * Leave lazy mode, flushing any hypercalls made here. This
563 * must be done after loading TLS entries in the GDT but before
564 * loading segments that might reference them.
565 */
566 arch_end_context_switch(next_p);
567
568 /* Switch DS and ES.
569 *
570 * Reading them only returns the selectors, but writing them (if
571 * nonzero) loads the full descriptor from the GDT or LDT. The
572 * LDT for next is loaded in switch_mm, and the GDT is loaded
573 * above.
574 *
575 * We therefore need to write new values to the segment
576 * registers on every context switch unless both the new and old
577 * values are zero.
578 *
579 * Note that we don't need to do anything for CS and SS, as
580 * those are saved and restored as part of pt_regs.
581 */
582 savesegment(es, prev->es);
583 if (unlikely(next->es | prev->es))
584 loadsegment(es, next->es);
585
586 savesegment(ds, prev->ds);
587 if (unlikely(next->ds | prev->ds))
588 loadsegment(ds, next->ds);
589
590 x86_fsgsbase_load(prev, next);
591
592 /*
593 * Switch the PDA and FPU contexts.
594 */
595 this_cpu_write(current_task, next_p);
596 this_cpu_write(cpu_current_top_of_stack, task_top_of_stack(next_p));
597
598 switch_fpu_finish(next_fpu);
599
600 /* Reload sp0. */
601 update_task_stack(next_p);
602
603 switch_to_extra(prev_p, next_p);
604
605 if (static_cpu_has_bug(X86_BUG_SYSRET_SS_ATTRS)) {
606 /*
607 * AMD CPUs have a misfeature: SYSRET sets the SS selector but
608 * does not update the cached descriptor. As a result, if we
609 * do SYSRET while SS is NULL, we'll end up in user mode with
610 * SS apparently equal to __USER_DS but actually unusable.
611 *
612 * The straightforward workaround would be to fix it up just
613 * before SYSRET, but that would slow down the system call
614 * fast paths. Instead, we ensure that SS is never NULL in
615 * system call context. We do this by replacing NULL SS
616 * selectors at every context switch. SYSCALL sets up a valid
617 * SS, so the only way to get NULL is to re-enter the kernel
618 * from CPL 3 through an interrupt. Since that can't happen
619 * in the same task as a running syscall, we are guaranteed to
620 * context switch between every interrupt vector entry and a
621 * subsequent SYSRET.
622 *
623 * We read SS first because SS reads are much faster than
624 * writes. Out of caution, we force SS to __KERNEL_DS even if
625 * it previously had a different non-NULL value.
626 */
627 unsigned short ss_sel;
628 savesegment(ss, ss_sel);
629 if (ss_sel != __KERNEL_DS)
630 loadsegment(ss, __KERNEL_DS);
631 }
632
633 /* Load the Intel cache allocation PQR MSR. */
634 resctrl_sched_in();
635
636 return prev_p;
637}
638
639void set_personality_64bit(void)
640{
641 /* inherit personality from parent */
642
643 /* Make sure to be in 64bit mode */
644 clear_thread_flag(TIF_IA32);
645 clear_thread_flag(TIF_ADDR32);
646 clear_thread_flag(TIF_X32);
647 /* Pretend that this comes from a 64bit execve */
648 task_pt_regs(current)->orig_ax = __NR_execve;
649 current_thread_info()->status &= ~TS_COMPAT;
650
651 /* Ensure the corresponding mm is not marked. */
652 if (current->mm)
653 current->mm->context.ia32_compat = 0;
654
655 /* TBD: overwrites user setup. Should have two bits.
656 But 64bit processes have always behaved this way,
657 so it's not too bad. The main problem is just that
658 32bit children are affected again. */
659 current->personality &= ~READ_IMPLIES_EXEC;
660}
661
662static void __set_personality_x32(void)
663{
664#ifdef CONFIG_X86_X32
665 clear_thread_flag(TIF_IA32);
666 set_thread_flag(TIF_X32);
667 if (current->mm)
668 current->mm->context.ia32_compat = TIF_X32;
669 current->personality &= ~READ_IMPLIES_EXEC;
670 /*
671 * in_32bit_syscall() uses the presence of the x32 syscall bit
672 * flag to determine compat status. The x86 mmap() code relies on
673 * the syscall bitness so set x32 syscall bit right here to make
674 * in_32bit_syscall() work during exec().
675 *
676 * Pretend to come from a x32 execve.
677 */
678 task_pt_regs(current)->orig_ax = __NR_x32_execve | __X32_SYSCALL_BIT;
679 current_thread_info()->status &= ~TS_COMPAT;
680#endif
681}
682
683static void __set_personality_ia32(void)
684{
685#ifdef CONFIG_IA32_EMULATION
686 set_thread_flag(TIF_IA32);
687 clear_thread_flag(TIF_X32);
688 if (current->mm)
689 current->mm->context.ia32_compat = TIF_IA32;
690 current->personality |= force_personality32;
691 /* Prepare the first "return" to user space */
692 task_pt_regs(current)->orig_ax = __NR_ia32_execve;
693 current_thread_info()->status |= TS_COMPAT;
694#endif
695}
696
697void set_personality_ia32(bool x32)
698{
699 /* Make sure to be in 32bit mode */
700 set_thread_flag(TIF_ADDR32);
701
702 if (x32)
703 __set_personality_x32();
704 else
705 __set_personality_ia32();
706}
707EXPORT_SYMBOL_GPL(set_personality_ia32);
708
709#ifdef CONFIG_CHECKPOINT_RESTORE
710static long prctl_map_vdso(const struct vdso_image *image, unsigned long addr)
711{
712 int ret;
713
714 ret = map_vdso_once(image, addr);
715 if (ret)
716 return ret;
717
718 return (long)image->size;
719}
720#endif
721
722long do_arch_prctl_64(struct task_struct *task, int option, unsigned long arg2)
723{
724 int ret = 0;
725
726 switch (option) {
727 case ARCH_SET_GS: {
728 if (unlikely(arg2 >= TASK_SIZE_MAX))
729 return -EPERM;
730
731 preempt_disable();
732 /*
733 * ARCH_SET_GS has always overwritten the index
734 * and the base. Zero is the most sensible value
735 * to put in the index, and is the only value that
736 * makes any sense if FSGSBASE is unavailable.
737 */
738 if (task == current) {
739 loadseg(GS, 0);
740 x86_gsbase_write_cpu_inactive(arg2);
741
742 /*
743 * On non-FSGSBASE systems, save_base_legacy() expects
744 * that we also fill in thread.gsbase.
745 */
746 task->thread.gsbase = arg2;
747
748 } else {
749 task->thread.gsindex = 0;
750 x86_gsbase_write_task(task, arg2);
751 }
752 preempt_enable();
753 break;
754 }
755 case ARCH_SET_FS: {
756 /*
757 * Not strictly needed for %fs, but do it for symmetry
758 * with %gs
759 */
760 if (unlikely(arg2 >= TASK_SIZE_MAX))
761 return -EPERM;
762
763 preempt_disable();
764 /*
765 * Set the selector to 0 for the same reason
766 * as %gs above.
767 */
768 if (task == current) {
769 loadseg(FS, 0);
770 x86_fsbase_write_cpu(arg2);
771
772 /*
773 * On non-FSGSBASE systems, save_base_legacy() expects
774 * that we also fill in thread.fsbase.
775 */
776 task->thread.fsbase = arg2;
777 } else {
778 task->thread.fsindex = 0;
779 x86_fsbase_write_task(task, arg2);
780 }
781 preempt_enable();
782 break;
783 }
784 case ARCH_GET_FS: {
785 unsigned long base = x86_fsbase_read_task(task);
786
787 ret = put_user(base, (unsigned long __user *)arg2);
788 break;
789 }
790 case ARCH_GET_GS: {
791 unsigned long base = x86_gsbase_read_task(task);
792
793 ret = put_user(base, (unsigned long __user *)arg2);
794 break;
795 }
796
797#ifdef CONFIG_CHECKPOINT_RESTORE
798# ifdef CONFIG_X86_X32_ABI
799 case ARCH_MAP_VDSO_X32:
800 return prctl_map_vdso(&vdso_image_x32, arg2);
801# endif
802# if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
803 case ARCH_MAP_VDSO_32:
804 return prctl_map_vdso(&vdso_image_32, arg2);
805# endif
806 case ARCH_MAP_VDSO_64:
807 return prctl_map_vdso(&vdso_image_64, arg2);
808#endif
809
810 default:
811 ret = -EINVAL;
812 break;
813 }
814
815 return ret;
816}
817
818SYSCALL_DEFINE2(arch_prctl, int, option, unsigned long, arg2)
819{
820 long ret;
821
822 ret = do_arch_prctl_64(current, option, arg2);
823 if (ret == -EINVAL)
824 ret = do_arch_prctl_common(current, option, arg2);
825
826 return ret;
827}
828
829#ifdef CONFIG_IA32_EMULATION
830COMPAT_SYSCALL_DEFINE2(arch_prctl, int, option, unsigned long, arg2)
831{
832 return do_arch_prctl_common(current, option, arg2);
833}
834#endif
835
836unsigned long KSTK_ESP(struct task_struct *task)
837{
838 return task_pt_regs(task)->sp;
839}