blob: c6d6dc5f8bb2a2245bf0203c498abf38d6be0bc4 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1995 Linus Torvalds
3 *
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
6 */
7
8/*
9 * This file handles the architecture-dependent parts of process handling..
10 */
11
Zwane Mwaikambof3705132005-06-25 14:54:50 -070012#include <linux/cpu.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/errno.h>
14#include <linux/sched.h>
Ingo Molnar29930022017-02-08 18:51:36 +010015#include <linux/sched/task.h>
Ingo Molnar68db0cf2017-02-08 18:51:37 +010016#include <linux/sched/task_stack.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/fs.h>
18#include <linux/kernel.h>
19#include <linux/mm.h>
20#include <linux/elfcore.h>
21#include <linux/smp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/stddef.h>
23#include <linux/slab.h>
24#include <linux/vmalloc.h>
25#include <linux/user.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/interrupt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/delay.h>
28#include <linux/reboot.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/mc146818rtc.h>
Paul Gortmaker186f4362016-07-13 20:18:56 -040030#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/kallsyms.h>
32#include <linux/ptrace.h>
Andi Kleenc16b63e02006-09-26 10:52:28 +020033#include <linux/personality.h>
Jeremy Fitzhardinge7c3576d2007-05-02 19:27:16 +020034#include <linux/percpu.h>
Erik Bosman529e25f2008-04-14 00:24:18 +020035#include <linux/prctl.h>
Frederic Weisbecker8b96f012008-12-06 03:40:00 +010036#include <linux/ftrace.h>
Jaswinder Singh Rajputbefa9e72009-01-04 16:18:56 +053037#include <linux/uaccess.h>
38#include <linux/io.h>
39#include <linux/kdebug.h>
Kyle Huey79170fd2017-03-20 01:16:24 -070040#include <linux/syscalls.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <asm/pgtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <asm/ldt.h>
44#include <asm/processor.h>
Ingo Molnar78f7f1e2015-04-24 02:54:44 +020045#include <asm/fpu/internal.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <asm/desc.h>
47#ifdef CONFIG_MATH_EMULATION
48#include <asm/math_emu.h>
49#endif
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/err.h>
52
Zwane Mwaikambof3705132005-06-25 14:54:50 -070053#include <asm/tlbflush.h>
54#include <asm/cpu.h>
Jaswinder Singhbbc1f692008-07-21 21:34:13 +053055#include <asm/syscalls.h>
K.Prasad66cb5912009-06-01 23:44:55 +053056#include <asm/debugreg.h>
David Howellsf05e7982012-03-28 18:11:12 +010057#include <asm/switch_to.h>
Brian Gerstba3e1272015-07-29 01:41:21 -040058#include <asm/vm86.h>
Fenghua Yu4f341a52016-10-28 15:04:48 -070059#include <asm/intel_rdt.h>
Kyle Huey79170fd2017-03-20 01:16:24 -070060#include <asm/proto.h>
Zwane Mwaikambof3705132005-06-25 14:54:50 -070061
Pekka Enberge2ce07c2008-04-03 16:40:48 +030062void __show_regs(struct pt_regs *regs, int all)
Linus Torvalds1da177e2005-04-16 15:20:36 -070063{
64 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
Alan Sternbb1995d2007-07-21 17:10:42 +020065 unsigned long d0, d1, d2, d3, d6, d7;
H. Peter Anvin65ea5b02008-01-30 13:30:56 +010066 unsigned long sp;
Pavel Emelyanov9d975eb2007-10-19 20:35:03 +020067 unsigned short ss, gs;
68
Andy Lutomirskif39b6f02015-03-18 18:33:33 -070069 if (user_mode(regs)) {
H. Peter Anvin65ea5b02008-01-30 13:30:56 +010070 sp = regs->sp;
71 ss = regs->ss & 0xffff;
Tejun Heod9a89a22009-02-09 22:17:40 +090072 gs = get_user_gs(regs);
Pavel Emelyanov9d975eb2007-10-19 20:35:03 +020073 } else {
H. Peter Anvindef3c5d2009-10-12 14:09:07 -070074 sp = kernel_stack_pointer(regs);
Pavel Emelyanov9d975eb2007-10-19 20:35:03 +020075 savesegment(ss, ss);
76 savesegment(gs, gs);
77 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
Josh Poimboeufbb5e5ce2016-10-25 09:51:12 -050079 printk(KERN_DEFAULT "EIP: %pS\n", (void *)regs->ip);
80 printk(KERN_DEFAULT "EFLAGS: %08lx CPU: %d\n", regs->flags,
Borislav Petkov5d9070b2017-05-28 11:03:42 +020081 raw_smp_processor_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
Pekka Enbergd015a092009-12-28 10:26:59 +020083 printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
H. Peter Anvin65ea5b02008-01-30 13:30:56 +010084 regs->ax, regs->bx, regs->cx, regs->dx);
Pekka Enbergd015a092009-12-28 10:26:59 +020085 printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
H. Peter Anvin65ea5b02008-01-30 13:30:56 +010086 regs->si, regs->di, regs->bp, sp);
Pekka Enbergd015a092009-12-28 10:26:59 +020087 printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
Harvey Harrison92bc2052008-02-08 12:09:56 -080088 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
Pavel Emelyanov9d975eb2007-10-19 20:35:03 +020089
90 if (!all)
91 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Zachary Amsden4bb0d3e2005-09-03 15:56:36 -070093 cr0 = read_cr0();
94 cr2 = read_cr2();
Andy Lutomirski6c690ee2017-06-12 10:26:14 -070095 cr3 = __read_cr3();
Andy Lutomirski1ef55be12016-09-29 12:48:12 -070096 cr4 = __read_cr4();
Pekka Enbergd015a092009-12-28 10:26:59 +020097 printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
Pavel Emelyanov9d975eb2007-10-19 20:35:03 +020098 cr0, cr2, cr3, cr4);
Alan Sternbb1995d2007-07-21 17:10:42 +020099
100 get_debugreg(d0, 0);
101 get_debugreg(d1, 1);
102 get_debugreg(d2, 2);
103 get_debugreg(d3, 3);
Alan Sternbb1995d2007-07-21 17:10:42 +0200104 get_debugreg(d6, 6);
105 get_debugreg(d7, 7);
Dave Jones43387742013-06-18 12:09:11 -0400106
107 /* Only print out debug registers if they are in their non-default state. */
108 if ((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
109 (d6 == DR6_RESERVED) && (d7 == 0x400))
110 return;
111
112 printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
113 d0, d1, d2, d3);
Pekka Enbergd015a092009-12-28 10:26:59 +0200114 printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
Pavel Emelyanov9d975eb2007-10-19 20:35:03 +0200115 d6, d7);
116}
Alan Sternbb1995d2007-07-21 17:10:42 +0200117
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118void release_thread(struct task_struct *dead_task)
119{
Zachary Amsden26849272006-01-06 00:11:59 -0800120 BUG_ON(dead_task->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 release_vm86_irqs(dead_task);
122}
123
Josh Triplettc1bd55f2015-06-30 15:00:00 -0700124int copy_thread_tls(unsigned long clone_flags, unsigned long sp,
125 unsigned long arg, struct task_struct *p, unsigned long tls)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126{
Al Viro7076aad2012-09-10 16:44:54 -0400127 struct pt_regs *childregs = task_pt_regs(p);
Brian Gerst01003012016-08-13 12:38:19 -0400128 struct fork_frame *fork_frame = container_of(childregs, struct fork_frame, regs);
129 struct inactive_task_frame *frame = &fork_frame->frame;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 struct task_struct *tsk;
131 int err;
132
Brian Gerst01003012016-08-13 12:38:19 -0400133 frame->bp = 0;
Brian Gerst616d2482016-08-13 12:38:20 -0400134 frame->ret_addr = (unsigned long) ret_from_fork;
Brian Gerst01003012016-08-13 12:38:19 -0400135 p->thread.sp = (unsigned long) fork_frame;
H. Peter Anvinfaca6222008-01-30 13:31:02 +0100136 p->thread.sp0 = (unsigned long) (childregs+1);
Oleg Nesterov6f46b3a2014-09-02 19:57:33 +0200137 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
Al Viro1d4b4b22012-10-22 22:34:11 -0400139 if (unlikely(p->flags & PF_KTHREAD)) {
Al Viro7076aad2012-09-10 16:44:54 -0400140 /* kernel thread */
141 memset(childregs, 0, sizeof(struct pt_regs));
Brian Gerst616d2482016-08-13 12:38:20 -0400142 frame->bx = sp; /* function */
143 frame->di = arg;
Al Viro7076aad2012-09-10 16:44:54 -0400144 p->thread.io_bitmap_ptr = NULL;
Al Viro7076aad2012-09-10 16:44:54 -0400145 return 0;
146 }
Brian Gerst616d2482016-08-13 12:38:20 -0400147 frame->bx = 0;
Al Viro1d4b4b22012-10-22 22:34:11 -0400148 *childregs = *current_pt_regs();
Al Viro7076aad2012-09-10 16:44:54 -0400149 childregs->ax = 0;
Al Viro1d4b4b22012-10-22 22:34:11 -0400150 if (sp)
151 childregs->sp = sp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Al Viro1d4b4b22012-10-22 22:34:11 -0400153 task_user_gs(p) = get_user_gs(current_pt_regs());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
K.Prasad66cb5912009-06-01 23:44:55 +0530155 p->thread.io_bitmap_ptr = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 tsk = current;
K.Prasad66cb5912009-06-01 23:44:55 +0530157 err = -ENOMEM;
Frederic Weisbecker24f1e32c2009-09-09 19:22:48 +0200158
Stephane Eranianb3cf2572006-07-09 21:12:39 -0400159 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
Alexey Dobriyan52978be2006-09-30 23:27:21 -0700160 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
161 IO_BITMAP_BYTES, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 if (!p->thread.io_bitmap_ptr) {
163 p->thread.io_bitmap_max = 0;
164 return -ENOMEM;
165 }
Stephane Eranianb3cf2572006-07-09 21:12:39 -0400166 set_tsk_thread_flag(p, TIF_IO_BITMAP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 }
168
Roland McGrathefd1ca52008-01-30 13:30:46 +0100169 err = 0;
170
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 /*
172 * Set a new TLS for the child thread?
173 */
Roland McGrathefd1ca52008-01-30 13:30:46 +0100174 if (clone_flags & CLONE_SETTLS)
175 err = do_set_thread_area(p, -1,
Josh Triplettc1bd55f2015-06-30 15:00:00 -0700176 (struct user_desc __user *)tls, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 if (err && p->thread.io_bitmap_ptr) {
179 kfree(p->thread.io_bitmap_ptr);
180 p->thread.io_bitmap_max = 0;
181 }
182 return err;
183}
184
Ingo Molnar513ad842008-02-21 05:18:40 +0100185void
186start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
187{
Tejun Heod9a89a22009-02-09 22:17:40 +0900188 set_user_gs(regs, 0);
Ingo Molnar513ad842008-02-21 05:18:40 +0100189 regs->fs = 0;
Ingo Molnar513ad842008-02-21 05:18:40 +0100190 regs->ds = __USER_DS;
191 regs->es = __USER_DS;
192 regs->ss = __USER_DS;
193 regs->cs = __USER_CS;
194 regs->ip = new_ip;
195 regs->sp = new_sp;
Al Viro6783eaa2012-08-02 23:05:11 +0400196 regs->flags = X86_EFLAGS_IF;
Brian Gerst1daeaa32015-03-21 18:54:21 -0400197 force_iret();
Ingo Molnar513ad842008-02-21 05:18:40 +0100198}
199EXPORT_SYMBOL_GPL(start_thread);
200
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
202/*
Kamalesh Babulalea70ef32011-04-28 14:32:08 +0530203 * switch_to(x,y) should switch tasks from x to y.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 *
205 * We fsave/fwait so that an exception goes off at the right time
206 * (as a call from the fsave or fwait in effect) rather than to
207 * the wrong process. Lazy FP saving no longer makes any sense
208 * with modern CPU's, and this simplifies a lot of things (SMP
209 * and UP become the same).
210 *
211 * NOTE! We used to use the x86 hardware context switching. The
212 * reason for not using it any more becomes apparent when you
213 * try to recover gracefully from saved state that is no longer
214 * valid (stale segment register values in particular). With the
215 * hardware task-switch, there is no way to fix up bad state in
216 * a reasonable manner.
217 *
218 * The fact that Intel documents the hardware task-switching to
219 * be slow is a fairly red herring - this code is not noticeably
220 * faster. However, there _is_ some room for improvement here,
221 * so the performance issues may eventually be a valid point.
222 * More important, however, is the fact that this allows us much
223 * more flexibility.
224 *
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100225 * The return value (in %ax) will be the "prev" task after
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 * the task-switch, and shows up in ret_from_fork in entry.S,
227 * for example.
228 */
Andi Kleen35ea79032013-08-05 15:02:39 -0700229__visible __notrace_funcgraph struct task_struct *
Frederic Weisbecker8b96f012008-12-06 03:40:00 +0100230__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231{
232 struct thread_struct *prev = &prev_p->thread,
Ingo Molnar384a23f2015-04-23 17:43:27 +0200233 *next = &next_p->thread;
234 struct fpu *prev_fpu = &prev->fpu;
235 struct fpu *next_fpu = &next->fpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 int cpu = smp_processor_id();
Andy Lutomirski24933b82015-03-05 19:19:05 -0800237 struct tss_struct *tss = &per_cpu(cpu_tss, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
239 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
240
Rik van Rielc474e502016-10-14 08:15:31 -0400241 switch_fpu_prepare(prev_fpu, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
243 /*
Jeremy Fitzhardinge464d1a72007-02-13 13:26:20 +0100244 * Save away %gs. No need to save %fs, as it was saved on the
Jeremy Fitzhardingef95d47c2006-12-07 02:14:02 +0100245 * stack on entry. No need to save %es and %ds, as those are
246 * always kernel segments while inside the kernel. Doing this
247 * before setting the new TLS descriptors avoids the situation
248 * where we temporarily have non-reloadable segments in %fs
249 * and %gs. This could be an issue if the NMI handler ever
250 * used %fs or %gs (it does not today), or if the kernel is
251 * running inside of a hypervisor layer.
Zachary Amsdene7a2ff52005-09-03 15:56:39 -0700252 */
Tejun Heoccbeed32009-02-09 22:17:40 +0900253 lazy_save_gs(prev->gs);
Zachary Amsdene7a2ff52005-09-03 15:56:39 -0700254
255 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 * Load the per-thread Thread-Local Storage descriptor.
257 */
258 load_TLS(next, cpu);
259
260 /*
Zachary Amsden8b151142007-02-13 13:26:21 +0100261 * Restore IOPL if needed. In normal use, the flags restore
262 * in the switch assembly will handle this. But if the kernel
263 * is running virtualized at a non-zero CPL, the popf will
264 * not restore flags, so it must be done in a separate step.
265 */
266 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
267 set_iopl_mask(next->iopl);
268
269 /*
Stephane Eranianb3cf2572006-07-09 21:12:39 -0400270 * Now maybe handle debug registers and/or IO bitmaps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 */
Andrea Arcangelicf99aba2007-07-15 23:41:33 -0700272 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
273 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
274 __switch_to_xtra(prev_p, next_p, tss);
Andrea Arcangeliffaa8bd2005-06-27 14:36:36 -0700275
Zachary Amsden9226d122007-02-13 13:26:21 +0100276 /*
277 * Leave lazy mode, flushing any hypercalls made here.
278 * This must be done before restoring TLS segments so
279 * the GDT and LDT are properly updated, and must be
Ingo Molnar3a0aee42015-04-22 13:16:47 +0200280 * done before fpu__restore(), so the TS bit is up
Zachary Amsden9226d122007-02-13 13:26:21 +0100281 * to date.
282 */
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800283 arch_end_context_switch(next_p);
Zachary Amsden9226d122007-02-13 13:26:21 +0100284
Andy Lutomirskib27559a2015-03-06 17:50:18 -0800285 /*
Denys Vlasenkofed7c3f2015-04-24 17:31:34 +0200286 * Reload esp0 and cpu_current_top_of_stack. This changes
Andy Lutomirskia7fcf282015-03-06 17:50:19 -0800287 * current_thread_info().
Andy Lutomirskib27559a2015-03-06 17:50:18 -0800288 */
289 load_sp0(tss, next);
Andy Lutomirskia7fcf282015-03-06 17:50:19 -0800290 this_cpu_write(cpu_current_top_of_stack,
291 (unsigned long)task_stack_page(next_p) +
292 THREAD_SIZE);
Steven Rostedt198d2082014-02-06 09:41:31 -0500293
Zachary Amsden9226d122007-02-13 13:26:21 +0100294 /*
295 * Restore %gs if needed (which is common)
296 */
297 if (prev->gs | next->gs)
Tejun Heoccbeed32009-02-09 22:17:40 +0900298 lazy_load_gs(next->gs);
Zachary Amsden9226d122007-02-13 13:26:21 +0100299
Rik van Rielc474e502016-10-14 08:15:31 -0400300 switch_fpu_finish(next_fpu, cpu);
Linus Torvalds34ddc812012-02-18 12:56:35 -0800301
Alex Shic6ae41e2012-05-11 15:35:27 +0800302 this_cpu_write(current_task, next_p);
Zachary Amsden9226d122007-02-13 13:26:21 +0100303
Fenghua Yu4f341a52016-10-28 15:04:48 -0700304 /* Load the Intel cache allocation PQR MSR. */
305 intel_rdt_sched_in();
306
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 return prev_p;
308}
Kyle Huey79170fd2017-03-20 01:16:24 -0700309
310SYSCALL_DEFINE2(arch_prctl, int, option, unsigned long, arg2)
311{
312 return do_arch_prctl_common(current, option, arg2);
313}