aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kernel/process_64.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/kernel/process_64.c')
-rw-r--r--arch/x86/kernel/process_64.c170
1 files changed, 97 insertions, 73 deletions
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index e12e0e4dd25..2a8ccb9238b 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -37,11 +37,11 @@
37#include <linux/kdebug.h> 37#include <linux/kdebug.h>
38#include <linux/tick.h> 38#include <linux/tick.h>
39#include <linux/prctl.h> 39#include <linux/prctl.h>
40#include <linux/uaccess.h>
41#include <linux/io.h>
40 42
41#include <asm/uaccess.h>
42#include <asm/pgtable.h> 43#include <asm/pgtable.h>
43#include <asm/system.h> 44#include <asm/system.h>
44#include <asm/io.h>
45#include <asm/processor.h> 45#include <asm/processor.h>
46#include <asm/i387.h> 46#include <asm/i387.h>
47#include <asm/mmu_context.h> 47#include <asm/mmu_context.h>
@@ -51,6 +51,7 @@
51#include <asm/proto.h> 51#include <asm/proto.h>
52#include <asm/ia32.h> 52#include <asm/ia32.h>
53#include <asm/idle.h> 53#include <asm/idle.h>
54#include <asm/syscalls.h>
54 55
55asmlinkage extern void ret_from_fork(void); 56asmlinkage extern void ret_from_fork(void);
56 57
@@ -88,7 +89,7 @@ void exit_idle(void)
88#ifdef CONFIG_HOTPLUG_CPU 89#ifdef CONFIG_HOTPLUG_CPU
89DECLARE_PER_CPU(int, cpu_state); 90DECLARE_PER_CPU(int, cpu_state);
90 91
91#include <asm/nmi.h> 92#include <linux/nmi.h>
92/* We halt the CPU with physical CPU hotplug */ 93/* We halt the CPU with physical CPU hotplug */
93static inline void play_dead(void) 94static inline void play_dead(void)
94{ 95{
@@ -153,7 +154,7 @@ void cpu_idle(void)
153} 154}
154 155
155/* Prints also some state that isn't saved in the pt_regs */ 156/* Prints also some state that isn't saved in the pt_regs */
156void __show_regs(struct pt_regs * regs) 157void __show_regs(struct pt_regs *regs)
157{ 158{
158 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs; 159 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
159 unsigned long d0, d1, d2, d3, d6, d7; 160 unsigned long d0, d1, d2, d3, d6, d7;
@@ -162,59 +163,61 @@ void __show_regs(struct pt_regs * regs)
162 163
163 printk("\n"); 164 printk("\n");
164 print_modules(); 165 print_modules();
165 printk("Pid: %d, comm: %.20s %s %s %.*s\n", 166 printk(KERN_INFO "Pid: %d, comm: %.20s %s %s %.*s\n",
166 current->pid, current->comm, print_tainted(), 167 current->pid, current->comm, print_tainted(),
167 init_utsname()->release, 168 init_utsname()->release,
168 (int)strcspn(init_utsname()->version, " "), 169 (int)strcspn(init_utsname()->version, " "),
169 init_utsname()->version); 170 init_utsname()->version);
170 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip); 171 printk(KERN_INFO "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
171 printk_address(regs->ip, 1); 172 printk_address(regs->ip, 1);
172 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->sp, 173 printk(KERN_INFO "RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss,
173 regs->flags); 174 regs->sp, regs->flags);
174 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n", 175 printk(KERN_INFO "RAX: %016lx RBX: %016lx RCX: %016lx\n",
175 regs->ax, regs->bx, regs->cx); 176 regs->ax, regs->bx, regs->cx);
176 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n", 177 printk(KERN_INFO "RDX: %016lx RSI: %016lx RDI: %016lx\n",
177 regs->dx, regs->si, regs->di); 178 regs->dx, regs->si, regs->di);
178 printk("RBP: %016lx R08: %016lx R09: %016lx\n", 179 printk(KERN_INFO "RBP: %016lx R08: %016lx R09: %016lx\n",
179 regs->bp, regs->r8, regs->r9); 180 regs->bp, regs->r8, regs->r9);
180 printk("R10: %016lx R11: %016lx R12: %016lx\n", 181 printk(KERN_INFO "R10: %016lx R11: %016lx R12: %016lx\n",
181 regs->r10, regs->r11, regs->r12); 182 regs->r10, regs->r11, regs->r12);
182 printk("R13: %016lx R14: %016lx R15: %016lx\n", 183 printk(KERN_INFO "R13: %016lx R14: %016lx R15: %016lx\n",
183 regs->r13, regs->r14, regs->r15); 184 regs->r13, regs->r14, regs->r15);
184 185
185 asm("movl %%ds,%0" : "=r" (ds)); 186 asm("movl %%ds,%0" : "=r" (ds));
186 asm("movl %%cs,%0" : "=r" (cs)); 187 asm("movl %%cs,%0" : "=r" (cs));
187 asm("movl %%es,%0" : "=r" (es)); 188 asm("movl %%es,%0" : "=r" (es));
188 asm("movl %%fs,%0" : "=r" (fsindex)); 189 asm("movl %%fs,%0" : "=r" (fsindex));
189 asm("movl %%gs,%0" : "=r" (gsindex)); 190 asm("movl %%gs,%0" : "=r" (gsindex));
190 191
191 rdmsrl(MSR_FS_BASE, fs); 192 rdmsrl(MSR_FS_BASE, fs);
192 rdmsrl(MSR_GS_BASE, gs); 193 rdmsrl(MSR_GS_BASE, gs);
193 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs); 194 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
194 195
195 cr0 = read_cr0(); 196 cr0 = read_cr0();
196 cr2 = read_cr2(); 197 cr2 = read_cr2();
197 cr3 = read_cr3(); 198 cr3 = read_cr3();
198 cr4 = read_cr4(); 199 cr4 = read_cr4();
199 200
200 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n", 201 printk(KERN_INFO "FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
201 fs,fsindex,gs,gsindex,shadowgs); 202 fs, fsindex, gs, gsindex, shadowgs);
202 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0); 203 printk(KERN_INFO "CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
203 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4); 204 es, cr0);
205 printk(KERN_INFO "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
206 cr4);
204 207
205 get_debugreg(d0, 0); 208 get_debugreg(d0, 0);
206 get_debugreg(d1, 1); 209 get_debugreg(d1, 1);
207 get_debugreg(d2, 2); 210 get_debugreg(d2, 2);
208 printk("DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2); 211 printk(KERN_INFO "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
209 get_debugreg(d3, 3); 212 get_debugreg(d3, 3);
210 get_debugreg(d6, 6); 213 get_debugreg(d6, 6);
211 get_debugreg(d7, 7); 214 get_debugreg(d7, 7);
212 printk("DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7); 215 printk(KERN_INFO "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
213} 216}
214 217
215void show_regs(struct pt_regs *regs) 218void show_regs(struct pt_regs *regs)
216{ 219{
217 printk("CPU %d:", smp_processor_id()); 220 printk(KERN_INFO "CPU %d:", smp_processor_id());
218 __show_regs(regs); 221 __show_regs(regs);
219 show_trace(NULL, regs, (void *)(regs + 1), regs->bp); 222 show_trace(NULL, regs, (void *)(regs + 1), regs->bp);
220} 223}
@@ -240,6 +243,14 @@ void exit_thread(void)
240 t->io_bitmap_max = 0; 243 t->io_bitmap_max = 0;
241 put_cpu(); 244 put_cpu();
242 } 245 }
246#ifdef CONFIG_X86_DS
247 /* Free any DS contexts that have not been properly released. */
248 if (unlikely(t->ds_ctx)) {
249 /* we clear debugctl to make sure DS is not used. */
250 update_debugctlmsr(0);
251 ds_free(t->ds_ctx);
252 }
253#endif /* CONFIG_X86_DS */
243} 254}
244 255
245void flush_thread(void) 256void flush_thread(void)
@@ -315,10 +326,10 @@ void prepare_to_copy(struct task_struct *tsk)
315 326
316int copy_thread(int nr, unsigned long clone_flags, unsigned long sp, 327int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
317 unsigned long unused, 328 unsigned long unused,
318 struct task_struct * p, struct pt_regs * regs) 329 struct task_struct *p, struct pt_regs *regs)
319{ 330{
320 int err; 331 int err;
321 struct pt_regs * childregs; 332 struct pt_regs *childregs;
322 struct task_struct *me = current; 333 struct task_struct *me = current;
323 334
324 childregs = ((struct pt_regs *) 335 childregs = ((struct pt_regs *)
@@ -363,10 +374,10 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
363 if (test_thread_flag(TIF_IA32)) 374 if (test_thread_flag(TIF_IA32))
364 err = do_set_thread_area(p, -1, 375 err = do_set_thread_area(p, -1,
365 (struct user_desc __user *)childregs->si, 0); 376 (struct user_desc __user *)childregs->si, 0);
366 else 377 else
367#endif 378#endif
368 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8); 379 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
369 if (err) 380 if (err)
370 goto out; 381 goto out;
371 } 382 }
372 err = 0; 383 err = 0;
@@ -473,13 +484,27 @@ static inline void __switch_to_xtra(struct task_struct *prev_p,
473 next = &next_p->thread; 484 next = &next_p->thread;
474 485
475 debugctl = prev->debugctlmsr; 486 debugctl = prev->debugctlmsr;
476 if (next->ds_area_msr != prev->ds_area_msr) { 487
477 /* we clear debugctl to make sure DS 488#ifdef CONFIG_X86_DS
478 * is not in use when we change it */ 489 {
479 debugctl = 0; 490 unsigned long ds_prev = 0, ds_next = 0;
480 update_debugctlmsr(0); 491
481 wrmsrl(MSR_IA32_DS_AREA, next->ds_area_msr); 492 if (prev->ds_ctx)
493 ds_prev = (unsigned long)prev->ds_ctx->ds;
494 if (next->ds_ctx)
495 ds_next = (unsigned long)next->ds_ctx->ds;
496
497 if (ds_next != ds_prev) {
498 /*
499 * We clear debugctl to make sure DS
500 * is not in use when we change it:
501 */
502 debugctl = 0;
503 update_debugctlmsr(0);
504 wrmsrl(MSR_IA32_DS_AREA, ds_next);
505 }
482 } 506 }
507#endif /* CONFIG_X86_DS */
483 508
484 if (next->debugctlmsr != debugctl) 509 if (next->debugctlmsr != debugctl)
485 update_debugctlmsr(next->debugctlmsr); 510 update_debugctlmsr(next->debugctlmsr);
@@ -517,13 +542,13 @@ static inline void __switch_to_xtra(struct task_struct *prev_p,
517 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max); 542 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
518 } 543 }
519 544
520#ifdef X86_BTS 545#ifdef CONFIG_X86_PTRACE_BTS
521 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS)) 546 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
522 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS); 547 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
523 548
524 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS)) 549 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
525 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES); 550 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
526#endif 551#endif /* CONFIG_X86_PTRACE_BTS */
527} 552}
528 553
529/* 554/*
@@ -545,7 +570,7 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
545 unsigned fsindex, gsindex; 570 unsigned fsindex, gsindex;
546 571
547 /* we're going to use this soon, after a few expensive things */ 572 /* we're going to use this soon, after a few expensive things */
548 if (next_p->fpu_counter>5) 573 if (next_p->fpu_counter > 5)
549 prefetch(next->xstate); 574 prefetch(next->xstate);
550 575
551 /* 576 /*
@@ -553,13 +578,13 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
553 */ 578 */
554 load_sp0(tss, next); 579 load_sp0(tss, next);
555 580
556 /* 581 /*
557 * Switch DS and ES. 582 * Switch DS and ES.
558 * This won't pick up thread selector changes, but I guess that is ok. 583 * This won't pick up thread selector changes, but I guess that is ok.
559 */ 584 */
560 savesegment(es, prev->es); 585 savesegment(es, prev->es);
561 if (unlikely(next->es | prev->es)) 586 if (unlikely(next->es | prev->es))
562 loadsegment(es, next->es); 587 loadsegment(es, next->es);
563 588
564 savesegment(ds, prev->ds); 589 savesegment(ds, prev->ds);
565 if (unlikely(next->ds | prev->ds)) 590 if (unlikely(next->ds | prev->ds))
@@ -585,7 +610,7 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
585 */ 610 */
586 arch_leave_lazy_cpu_mode(); 611 arch_leave_lazy_cpu_mode();
587 612
588 /* 613 /*
589 * Switch FS and GS. 614 * Switch FS and GS.
590 * 615 *
591 * Segment register != 0 always requires a reload. Also 616 * Segment register != 0 always requires a reload. Also
@@ -594,13 +619,13 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
594 */ 619 */
595 if (unlikely(fsindex | next->fsindex | prev->fs)) { 620 if (unlikely(fsindex | next->fsindex | prev->fs)) {
596 loadsegment(fs, next->fsindex); 621 loadsegment(fs, next->fsindex);
597 /* 622 /*
598 * Check if the user used a selector != 0; if yes 623 * Check if the user used a selector != 0; if yes
599 * clear 64bit base, since overloaded base is always 624 * clear 64bit base, since overloaded base is always
600 * mapped to the Null selector 625 * mapped to the Null selector
601 */ 626 */
602 if (fsindex) 627 if (fsindex)
603 prev->fs = 0; 628 prev->fs = 0;
604 } 629 }
605 /* when next process has a 64bit base use it */ 630 /* when next process has a 64bit base use it */
606 if (next->fs) 631 if (next->fs)
@@ -610,7 +635,7 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
610 if (unlikely(gsindex | next->gsindex | prev->gs)) { 635 if (unlikely(gsindex | next->gsindex | prev->gs)) {
611 load_gs_index(next->gsindex); 636 load_gs_index(next->gsindex);
612 if (gsindex) 637 if (gsindex)
613 prev->gs = 0; 638 prev->gs = 0;
614 } 639 }
615 if (next->gs) 640 if (next->gs)
616 wrmsrl(MSR_KERNEL_GS_BASE, next->gs); 641 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
@@ -619,12 +644,12 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
619 /* Must be after DS reload */ 644 /* Must be after DS reload */
620 unlazy_fpu(prev_p); 645 unlazy_fpu(prev_p);
621 646
622 /* 647 /*
623 * Switch the PDA and FPU contexts. 648 * Switch the PDA and FPU contexts.
624 */ 649 */
625 prev->usersp = read_pda(oldrsp); 650 prev->usersp = read_pda(oldrsp);
626 write_pda(oldrsp, next->usersp); 651 write_pda(oldrsp, next->usersp);
627 write_pda(pcurrent, next_p); 652 write_pda(pcurrent, next_p);
628 653
629 write_pda(kernelstack, 654 write_pda(kernelstack,
630 (unsigned long)task_stack_page(next_p) + 655 (unsigned long)task_stack_page(next_p) +
@@ -665,7 +690,7 @@ long sys_execve(char __user *name, char __user * __user *argv,
665 char __user * __user *envp, struct pt_regs *regs) 690 char __user * __user *envp, struct pt_regs *regs)
666{ 691{
667 long error; 692 long error;
668 char * filename; 693 char *filename;
669 694
670 filename = getname(name); 695 filename = getname(name);
671 error = PTR_ERR(filename); 696 error = PTR_ERR(filename);
@@ -723,55 +748,55 @@ asmlinkage long sys_vfork(struct pt_regs *regs)
723unsigned long get_wchan(struct task_struct *p) 748unsigned long get_wchan(struct task_struct *p)
724{ 749{
725 unsigned long stack; 750 unsigned long stack;
726 u64 fp,ip; 751 u64 fp, ip;
727 int count = 0; 752 int count = 0;
728 753
729 if (!p || p == current || p->state==TASK_RUNNING) 754 if (!p || p == current || p->state == TASK_RUNNING)
730 return 0; 755 return 0;
731 stack = (unsigned long)task_stack_page(p); 756 stack = (unsigned long)task_stack_page(p);
732 if (p->thread.sp < stack || p->thread.sp > stack+THREAD_SIZE) 757 if (p->thread.sp < stack || p->thread.sp > stack+THREAD_SIZE)
733 return 0; 758 return 0;
734 fp = *(u64 *)(p->thread.sp); 759 fp = *(u64 *)(p->thread.sp);
735 do { 760 do {
736 if (fp < (unsigned long)stack || 761 if (fp < (unsigned long)stack ||
737 fp > (unsigned long)stack+THREAD_SIZE) 762 fp > (unsigned long)stack+THREAD_SIZE)
738 return 0; 763 return 0;
739 ip = *(u64 *)(fp+8); 764 ip = *(u64 *)(fp+8);
740 if (!in_sched_functions(ip)) 765 if (!in_sched_functions(ip))
741 return ip; 766 return ip;
742 fp = *(u64 *)fp; 767 fp = *(u64 *)fp;
743 } while (count++ < 16); 768 } while (count++ < 16);
744 return 0; 769 return 0;
745} 770}
746 771
747long do_arch_prctl(struct task_struct *task, int code, unsigned long addr) 772long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
748{ 773{
749 int ret = 0; 774 int ret = 0;
750 int doit = task == current; 775 int doit = task == current;
751 int cpu; 776 int cpu;
752 777
753 switch (code) { 778 switch (code) {
754 case ARCH_SET_GS: 779 case ARCH_SET_GS:
755 if (addr >= TASK_SIZE_OF(task)) 780 if (addr >= TASK_SIZE_OF(task))
756 return -EPERM; 781 return -EPERM;
757 cpu = get_cpu(); 782 cpu = get_cpu();
758 /* handle small bases via the GDT because that's faster to 783 /* handle small bases via the GDT because that's faster to
759 switch. */ 784 switch. */
760 if (addr <= 0xffffffff) { 785 if (addr <= 0xffffffff) {
761 set_32bit_tls(task, GS_TLS, addr); 786 set_32bit_tls(task, GS_TLS, addr);
762 if (doit) { 787 if (doit) {
763 load_TLS(&task->thread, cpu); 788 load_TLS(&task->thread, cpu);
764 load_gs_index(GS_TLS_SEL); 789 load_gs_index(GS_TLS_SEL);
765 } 790 }
766 task->thread.gsindex = GS_TLS_SEL; 791 task->thread.gsindex = GS_TLS_SEL;
767 task->thread.gs = 0; 792 task->thread.gs = 0;
768 } else { 793 } else {
769 task->thread.gsindex = 0; 794 task->thread.gsindex = 0;
770 task->thread.gs = addr; 795 task->thread.gs = addr;
771 if (doit) { 796 if (doit) {
772 load_gs_index(0); 797 load_gs_index(0);
773 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr); 798 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
774 } 799 }
775 } 800 }
776 put_cpu(); 801 put_cpu();
777 break; 802 break;
@@ -825,8 +850,7 @@ long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
825 rdmsrl(MSR_KERNEL_GS_BASE, base); 850 rdmsrl(MSR_KERNEL_GS_BASE, base);
826 else 851 else
827 base = task->thread.gs; 852 base = task->thread.gs;
828 } 853 } else
829 else
830 base = task->thread.gs; 854 base = task->thread.gs;
831 ret = put_user(base, (unsigned long __user *)addr); 855 ret = put_user(base, (unsigned long __user *)addr);
832 break; 856 break;