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 void __show_regs(struct pt_regs *regs, int all)
  63 {
  64         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
  65         unsigned long d0, d1, d2, d3, d6, d7;
  66         unsigned long sp;
  67         unsigned short ss, gs;
  68
  69         if (user_mode(regs)) {
  70                 sp = regs->sp;
  71                 ss = regs->ss;
  72                 gs = get_user_gs(regs);
  73         } else {
  74                 sp = kernel_stack_pointer(regs);
  75                 savesegment(ss, ss);
  76                 savesegment(gs, gs);
  77         }
  78
  79         printk(KERN_DEFAULT "EIP: %pS\n", (void *)regs->ip);
  80         printk(KERN_DEFAULT "EFLAGS: %08lx CPU: %d\n", regs->flags,
  81                 raw_smp_processor_id());
  82
  83         printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
  84                 regs->ax, regs->bx, regs->cx, regs->dx);
  85         printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
  86                 regs->si, regs->di, regs->bp, sp);
  87         printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
  88                (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
  89
  90         if (!all)
  91                 return;
  92
  93         cr0 = read_cr0();
  94         cr2 = read_cr2();
  95         cr3 = __read_cr3();
  96         cr4 = __read_cr4();
  97         printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
  98                         cr0, cr2, cr3, cr4);
  99
 100         get_debugreg(d0, 0);
 101         get_debugreg(d1, 1);
 102         get_debugreg(d2, 2);
 103         get_debugreg(d3, 3);
 104         get_debugreg(d6, 6);
 105         get_debugreg(d7, 7);
 106
 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);
 114         printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
 115                         d6, d7);
 116 }
 117
 118 void release_thread(struct task_struct *dead_task)
 119 {
 120         BUG_ON(dead_task->mm);
 121         release_vm86_irqs(dead_task);
 122 }
 123
 124 int copy_thread_tls(unsigned long clone_flags, unsigned long sp,
 125         unsigned long arg, struct task_struct *p, unsigned long tls)
 126 {
 127         struct pt_regs *childregs = task_pt_regs(p);
 128         struct fork_frame *fork_frame = container_of(childregs, struct fork_frame, regs);
 129         struct inactive_task_frame *frame = &fork_frame->frame;
 130         struct task_struct *tsk;
 131         int err;
 132
 133         frame->bp = 0;
 134         frame->ret_addr = (unsigned long) ret_from_fork;
 135         p->thread.sp = (unsigned long) fork_frame;
 136         p->thread.sp0 = (unsigned long) (childregs+1);
 137         memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
 138
 139         if (unlikely(p->flags & PF_KTHREAD)) {
 140                 /* kernel thread */
 141                 memset(childregs, 0, sizeof(struct pt_regs));
 142                 frame->bx = sp;         /* function */
 143                 frame->di = arg;
 144                 p->thread.io_bitmap_ptr = NULL;
 145                 return 0;
 146         }
 147         frame->bx = 0;
 148         *childregs = *current_pt_regs();
 149         childregs->ax = 0;
 150         if (sp)
 151                 childregs->sp = sp;
 152
 153         task_user_gs(p) = get_user_gs(current_pt_regs());
 154
 155         p->thread.io_bitmap_ptr = NULL;
 156         tsk = current;
 157         err = -ENOMEM;
 158
 159         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 160                 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
 161                                                 IO_BITMAP_BYTES, GFP_KERNEL);
 162                 if (!p->thread.io_bitmap_ptr) {
 163                         p->thread.io_bitmap_max = 0;
 164                         return -ENOMEM;
 165                 }
 166                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 167         }
 168
 169         err = 0;
 170
 171         /*
 172          * Set a new TLS for the child thread?
 173          */
 174         if (clone_flags & CLONE_SETTLS)
 175                 err = do_set_thread_area(p, -1,
 176                         (struct user_desc __user *)tls, 0);
 177
 178         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
 185 void
 186 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
 187 {
 188         set_user_gs(regs, 0);
 189         regs->fs                = 0;
 190         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;
 196         regs->flags             = X86_EFLAGS_IF;
 197         force_iret();
 198 }
 199 EXPORT_SYMBOL_GPL(start_thread);
 200
 201
 202 /*
 203  *      switch_to(x,y) should switch tasks from x to y.
 204  *
 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  *
 225  * The return value (in %ax) will be the "prev" task after
 226  * the task-switch, and shows up in ret_from_fork in entry.S,
 227  * for example.
 228  */
 229 __visible __notrace_funcgraph struct task_struct *
 230 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 231 {
 232         struct thread_struct *prev = &prev_p->thread,
 233                              *next = &next_p->thread;
 234         struct fpu *prev_fpu = &prev->fpu;
 235         struct fpu *next_fpu = &next->fpu;
 236         int cpu = smp_processor_id();
 237         struct tss_struct *tss = &per_cpu(cpu_tss_rw, cpu);
 238
 239         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 240
 241         switch_fpu_prepare(prev_fpu, cpu);
 242
 243         /*
 244          * Save away %gs. No need to save %fs, as it was saved on the
 245          * 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.
 252          */
 253         lazy_save_gs(prev->gs);
 254
 255         /*
 256          * Load the per-thread Thread-Local Storage descriptor.
 257          */
 258         load_TLS(next, cpu);
 259
 260         /*
 261          * 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         /*
 270          * Now maybe handle debug registers and/or IO bitmaps
 271          */
 272         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);
 275
 276         /*
 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
 280          * done before fpu__restore(), so the TS bit is up
 281          * to date.
 282          */
 283         arch_end_context_switch(next_p);
 284
 285         /*
 286          * Reload esp0 and cpu_current_top_of_stack.  This changes
 287          * current_thread_info().  Refresh the SYSENTER configuration in
 288          * case prev or next is vm86.
 289          */
 290         update_sp0(next_p);
 291         refresh_sysenter_cs(next);
 292         this_cpu_write(cpu_current_top_of_stack,
 293                        (unsigned long)task_stack_page(next_p) +
 294                        THREAD_SIZE);
 295
 296         /*
 297          * Restore %gs if needed (which is common)
 298          */
 299         if (prev->gs | next->gs)
 300                 lazy_load_gs(next->gs);
 301
 302         switch_fpu_finish(next_fpu, cpu);
 303
 304         this_cpu_write(current_task, next_p);
 305
 306         /* Load the Intel cache allocation PQR MSR. */
 307         intel_rdt_sched_in();
 308
 309         return prev_p;
 310 }
 311
 312 SYSCALL_DEFINE2(arch_prctl, int, option, unsigned long, arg2)
 313 {
 314         return do_arch_prctl_common(current, option, arg2);
 315 }