arch/x86/kernel/process_32.c

   1 /*
   2  *  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
  12 #include <linux/cpu.h>
  13 #include <linux/errno.h>
  14 #include <linux/sched.h>
  15 #include <linux/sched/task.h>
  16 #include <linux/sched/task_stack.h>
  17 #include <linux/fs.h>
  18 #include <linux/kernel.h>
  19 #include <linux/mm.h>
  20 #include <linux/elfcore.h>
  21 #include <linux/smp.h>
  22 #include <linux/stddef.h>
  23 #include <linux/slab.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/user.h>
  26 #include <linux/interrupt.h>
  27 #include <linux/delay.h>
  28 #include <linux/reboot.h>
  29 #include <linux/mc146818rtc.h>
  30 #include <linux/export.h>
  31 #include <linux/kallsyms.h>
  32 #include <linux/ptrace.h>
  33 #include <linux/personality.h>
  34 #include <linux/percpu.h>
  35 #include <linux/prctl.h>
  36 #include <linux/ftrace.h>
  37 #include <linux/uaccess.h>
  38 #include <linux/io.h>
  39 #include <linux/kdebug.h>
  40 #include <linux/syscalls.h>
  41
  42 #include <asm/pgtable.h>
  43 #include <asm/ldt.h>
  44 #include <asm/processor.h>
  45 #include <asm/fpu/internal.h>
  46 #include <asm/desc.h>
  47 #ifdef CONFIG_MATH_EMULATION
  48 #include <asm/math_emu.h>
  49 #endif
  50
  51 #include <linux/err.h>
  52
  53 #include <asm/tlbflush.h>
  54 #include <asm/cpu.h>
  55 #include <asm/syscalls.h>
  56 #include <asm/debugreg.h>
  57 #include <asm/switch_to.h>
  58 #include <asm/vm86.h>
  59 #include <asm/intel_rdt_sched.h>
  60 #include <asm/proto.h>
  61
  62 #include "process.h"
  63
  64 void __show_regs(struct pt_regs *regs, int all)
  65 {
  66         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
  67         unsigned long d0, d1, d2, d3, d6, d7;
  68         unsigned long sp;
  69         unsigned short ss, gs;
  70
  71         if (user_mode(regs)) {
  72                 sp = regs->sp;
  73                 ss = regs->ss;
  74                 gs = get_user_gs(regs);
  75         } else {
  76                 sp = kernel_stack_pointer(regs);
  77                 savesegment(ss, ss);
  78                 savesegment(gs, gs);
  79         }
  80
  81         printk(KERN_DEFAULT "EIP: %pS\n", (void *)regs->ip);
  82         printk(KERN_DEFAULT "EFLAGS: %08lx CPU: %d\n", regs->flags,
  83                 raw_smp_processor_id());
  84
  85         printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
  86                 regs->ax, regs->bx, regs->cx, regs->dx);
  87         printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
  88                 regs->si, regs->di, regs->bp, sp);
  89         printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
  90                (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
  91
  92         if (!all)
  93                 return;
  94
  95         cr0 = read_cr0();
  96         cr2 = read_cr2();
  97         cr3 = __read_cr3();
  98         cr4 = __read_cr4();
  99         printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
 100                         cr0, cr2, cr3, cr4);
 101
 102         get_debugreg(d0, 0);
 103         get_debugreg(d1, 1);
 104         get_debugreg(d2, 2);
 105         get_debugreg(d3, 3);
 106         get_debugreg(d6, 6);
 107         get_debugreg(d7, 7);
 108
 109         /* Only print out debug registers if they are in their non-default state. */
 110         if ((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
 111             (d6 == DR6_RESERVED) && (d7 == 0x400))
 112                 return;
 113
 114         printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
 115                         d0, d1, d2, d3);
 116         printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
 117                         d6, d7);
 118 }
 119
 120 void release_thread(struct task_struct *dead_task)
 121 {
 122         BUG_ON(dead_task->mm);
 123         release_vm86_irqs(dead_task);
 124 }
 125
 126 int copy_thread_tls(unsigned long clone_flags, unsigned long sp,
 127         unsigned long arg, struct task_struct *p, unsigned long tls)
 128 {
 129         struct pt_regs *childregs = task_pt_regs(p);
 130         struct fork_frame *fork_frame = container_of(childregs, struct fork_frame, regs);
 131         struct inactive_task_frame *frame = &fork_frame->frame;
 132         struct task_struct *tsk;
 133         int err;
 134
 135         /*
 136          * For a new task use the RESET flags value since there is no before.
 137          * All the status flags are zero; DF and all the system flags must also
 138          * be 0, specifically IF must be 0 because we context switch to the new
 139          * task with interrupts disabled.
 140          */
 141         frame->flags = X86_EFLAGS_FIXED;
 142         frame->bp = 0;
 143         frame->ret_addr = (unsigned long) ret_from_fork;
 144         p->thread.sp = (unsigned long) fork_frame;
 145         p->thread.sp0 = (unsigned long) (childregs+1);
 146         memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
 147
 148         if (unlikely(p->flags & PF_KTHREAD)) {
 149                 /* kernel thread */
 150                 memset(childregs, 0, sizeof(struct pt_regs));
 151                 frame->bx = sp;         /* function */
 152                 frame->di = arg;
 153                 p->thread.io_bitmap_ptr = NULL;
 154                 return 0;
 155         }
 156         frame->bx = 0;
 157         *childregs = *current_pt_regs();
 158         childregs->ax = 0;
 159         if (sp)
 160                 childregs->sp = sp;
 161
 162         task_user_gs(p) = get_user_gs(current_pt_regs());
 163
 164         p->thread.io_bitmap_ptr = NULL;
 165         tsk = current;
 166         err = -ENOMEM;
 167
 168         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 169                 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
 170                                                 IO_BITMAP_BYTES, GFP_KERNEL);
 171                 if (!p->thread.io_bitmap_ptr) {
 172                         p->thread.io_bitmap_max = 0;
 173                         return -ENOMEM;
 174                 }
 175                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 176         }
 177
 178         err = 0;
 179
 180         /*
 181          * Set a new TLS for the child thread?
 182          */
 183         if (clone_flags & CLONE_SETTLS)
 184                 err = do_set_thread_area(p, -1,
 185                         (struct user_desc __user *)tls, 0);
 186
 187         if (err && p->thread.io_bitmap_ptr) {
 188                 kfree(p->thread.io_bitmap_ptr);
 189                 p->thread.io_bitmap_max = 0;
 190         }
 191         return err;
 192 }
 193
 194 void
 195 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
 196 {
 197         set_user_gs(regs, 0);
 198         regs->fs                = 0;
 199         regs->ds                = __USER_DS;
 200         regs->es                = __USER_DS;
 201         regs->ss                = __USER_DS;
 202         regs->cs                = __USER_CS;
 203         regs->ip                = new_ip;
 204         regs->sp                = new_sp;
 205         regs->flags             = X86_EFLAGS_IF;
 206         force_iret();
 207 }
 208 EXPORT_SYMBOL_GPL(start_thread);
 209
 210
 211 /*
 212  *      switch_to(x,y) should switch tasks from x to y.
 213  *
 214  * We fsave/fwait so that an exception goes off at the right time
 215  * (as a call from the fsave or fwait in effect) rather than to
 216  * the wrong process. Lazy FP saving no longer makes any sense
 217  * with modern CPU's, and this simplifies a lot of things (SMP
 218  * and UP become the same).
 219  *
 220  * NOTE! We used to use the x86 hardware context switching. The
 221  * reason for not using it any more becomes apparent when you
 222  * try to recover gracefully from saved state that is no longer
 223  * valid (stale segment register values in particular). With the
 224  * hardware task-switch, there is no way to fix up bad state in
 225  * a reasonable manner.
 226  *
 227  * The fact that Intel documents the hardware task-switching to
 228  * be slow is a fairly red herring - this code is not noticeably
 229  * faster. However, there _is_ some room for improvement here,
 230  * so the performance issues may eventually be a valid point.
 231  * More important, however, is the fact that this allows us much
 232  * more flexibility.
 233  *
 234  * The return value (in %ax) will be the "prev" task after
 235  * the task-switch, and shows up in ret_from_fork in entry.S,
 236  * for example.
 237  */
 238 __visible __notrace_funcgraph struct task_struct *
 239 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 240 {
 241         struct thread_struct *prev = &prev_p->thread,
 242                              *next = &next_p->thread;
 243         struct fpu *prev_fpu = &prev->fpu;
 244         struct fpu *next_fpu = &next->fpu;
 245         int cpu = smp_processor_id();
 246
 247         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 248
 249         switch_fpu_prepare(prev_fpu, cpu);
 250
 251         /*
 252          * Save away %gs. No need to save %fs, as it was saved on the
 253          * stack on entry.  No need to save %es and %ds, as those are
 254          * always kernel segments while inside the kernel.  Doing this
 255          * before setting the new TLS descriptors avoids the situation
 256          * where we temporarily have non-reloadable segments in %fs
 257          * and %gs.  This could be an issue if the NMI handler ever
 258          * used %fs or %gs (it does not today), or if the kernel is
 259          * running inside of a hypervisor layer.
 260          */
 261         lazy_save_gs(prev->gs);
 262
 263         /*
 264          * Load the per-thread Thread-Local Storage descriptor.
 265          */
 266         load_TLS(next, cpu);
 267
 268         /*
 269          * Restore IOPL if needed.  In normal use, the flags restore
 270          * in the switch assembly will handle this.  But if the kernel
 271          * is running virtualized at a non-zero CPL, the popf will
 272          * not restore flags, so it must be done in a separate step.
 273          */
 274         if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
 275                 set_iopl_mask(next->iopl);
 276
 277         switch_to_extra(prev_p, next_p);
 278
 279         /*
 280          * Leave lazy mode, flushing any hypercalls made here.
 281          * This must be done before restoring TLS segments so
 282          * the GDT and LDT are properly updated, and must be
 283          * done before fpu__restore(), so the TS bit is up
 284          * to date.
 285          */
 286         arch_end_context_switch(next_p);
 287
 288         /*
 289          * Reload esp0 and cpu_current_top_of_stack.  This changes
 290          * current_thread_info().  Refresh the SYSENTER configuration in
 291          * case prev or next is vm86.
 292          */
 293         update_sp0(next_p);
 294         refresh_sysenter_cs(next);
 295         this_cpu_write(cpu_current_top_of_stack,
 296                        (unsigned long)task_stack_page(next_p) +
 297                        THREAD_SIZE);
 298
 299         /*
 300          * Restore %gs if needed (which is common)
 301          */
 302         if (prev->gs | next->gs)
 303                 lazy_load_gs(next->gs);
 304
 305         switch_fpu_finish(next_fpu, cpu);
 306
 307         this_cpu_write(current_task, next_p);
 308
 309         /* Load the Intel cache allocation PQR MSR. */
 310         intel_rdt_sched_in();
 311
 312         return prev_p;
 313 }
 314
 315 SYSCALL_DEFINE2(arch_prctl, int, option, unsigned long, arg2)
 316 {
 317         return do_arch_prctl_common(current, option, arg2);
 318 }