arch/s390/kernel/process.c

   1 /*
   2  * This file handles the architecture dependent parts of process handling.
   3  *
   4  *    Copyright IBM Corp. 1999,2009
   5  *    Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>,
   6  *               Hartmut Penner <hp@de.ibm.com>,
   7  *               Denis Joseph Barrow,
   8  */
   9
  10 #include <linux/compiler.h>
  11 #include <linux/cpu.h>
  12 #include <linux/sched.h>
  13 #include <linux/kernel.h>
  14 #include <linux/mm.h>
  15 #include <linux/elfcore.h>
  16 #include <linux/smp.h>
  17 #include <linux/slab.h>
  18 #include <linux/interrupt.h>
  19 #include <linux/tick.h>
  20 #include <linux/personality.h>
  21 #include <linux/syscalls.h>
  22 #include <linux/compat.h>
  23 #include <linux/kprobes.h>
  24 #include <linux/random.h>
  25 #include <linux/module.h>
  26 #include <asm/system.h>
  27 #include <asm/io.h>
  28 #include <asm/processor.h>
  29 #include <asm/irq.h>
  30 #include <asm/timer.h>
  31 #include <asm/nmi.h>
  32 #include <asm/smp.h>
  33 #include "entry.h"
  34
  35 asmlinkage void ret_from_fork(void) asm ("ret_from_fork");
  36
  37 /*
  38  * Return saved PC of a blocked thread. used in kernel/sched.
  39  * resume in entry.S does not create a new stack frame, it
  40  * just stores the registers %r6-%r15 to the frame given by
  41  * schedule. We want to return the address of the caller of
  42  * schedule, so we have to walk the backchain one time to
  43  * find the frame schedule() store its return address.
  44  */
  45 unsigned long thread_saved_pc(struct task_struct *tsk)
  46 {
  47         struct stack_frame *sf, *low, *high;
  48
  49         if (!tsk || !task_stack_page(tsk))
  50                 return 0;
  51         low = task_stack_page(tsk);
  52         high = (struct stack_frame *) task_pt_regs(tsk);
  53         sf = (struct stack_frame *) (tsk->thread.ksp & PSW_ADDR_INSN);
  54         if (sf <= low || sf > high)
  55                 return 0;
  56         sf = (struct stack_frame *) (sf->back_chain & PSW_ADDR_INSN);
  57         if (sf <= low || sf > high)
  58                 return 0;
  59         return sf->gprs[8];
  60 }
  61
  62 /*
  63  * The idle loop on a S390...
  64  */
  65 static void default_idle(void)
  66 {
  67         if (cpu_is_offline(smp_processor_id()))
  68                 cpu_die();
  69         local_irq_disable();
  70         if (need_resched()) {
  71                 local_irq_enable();
  72                 return;
  73         }
  74         local_mcck_disable();
  75         if (test_thread_flag(TIF_MCCK_PENDING)) {
  76                 local_mcck_enable();
  77                 local_irq_enable();
  78                 return;
  79         }
  80         trace_hardirqs_on();
  81         /* Don't trace preempt off for idle. */
  82         stop_critical_timings();
  83         /* Stop virtual timer and halt the cpu. */
  84         vtime_stop_cpu();
  85         /* Reenable preemption tracer. */
  86         start_critical_timings();
  87 }
  88
  89 void cpu_idle(void)
  90 {
  91         for (;;) {
  92                 tick_nohz_idle_enter();
  93                 rcu_idle_enter();
  94                 while (!need_resched() && !test_thread_flag(TIF_MCCK_PENDING))
  95                         default_idle();
  96                 rcu_idle_exit();
  97                 tick_nohz_idle_exit();
  98                 if (test_thread_flag(TIF_MCCK_PENDING))
  99                         s390_handle_mcck();
 100                 preempt_enable_no_resched();
 101                 schedule();
 102                 preempt_disable();
 103         }
 104 }
 105
 106 extern void __kprobes kernel_thread_starter(void);
 107
 108 asm(
 109         ".section .kprobes.text, \"ax\"\n"
 110         ".global kernel_thread_starter\n"
 111         "kernel_thread_starter:\n"
 112         "    la    2,0(10)\n"
 113         "    basr  14,9\n"
 114         "    la    2,0\n"
 115         "    br    11\n"
 116         ".previous\n");
 117
 118 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 119 {
 120         struct pt_regs regs;
 121
 122         memset(&regs, 0, sizeof(regs));
 123         regs.psw.mask = psw_kernel_bits |
 124                 PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT | PSW_MASK_MCHECK;
 125         regs.psw.addr = (unsigned long) kernel_thread_starter | PSW_ADDR_AMODE;
 126         regs.gprs[9] = (unsigned long) fn;
 127         regs.gprs[10] = (unsigned long) arg;
 128         regs.gprs[11] = (unsigned long) do_exit;
 129         regs.orig_gpr2 = -1;
 130
 131         /* Ok, create the new process.. */
 132         return do_fork(flags | CLONE_VM | CLONE_UNTRACED,
 133                        0, &regs, 0, NULL, NULL);
 134 }
 135 EXPORT_SYMBOL(kernel_thread);
 136
 137 /*
 138  * Free current thread data structures etc..
 139  */
 140 void exit_thread(void)
 141 {
 142 }
 143
 144 void flush_thread(void)
 145 {
 146 }
 147
 148 void release_thread(struct task_struct *dead_task)
 149 {
 150 }
 151
 152 int copy_thread(unsigned long clone_flags, unsigned long new_stackp,
 153                 unsigned long unused,
 154                 struct task_struct *p, struct pt_regs *regs)
 155 {
 156         struct thread_info *ti;
 157         struct fake_frame
 158         {
 159                 struct stack_frame sf;
 160                 struct pt_regs childregs;
 161         } *frame;
 162
 163         frame = container_of(task_pt_regs(p), struct fake_frame, childregs);
 164         p->thread.ksp = (unsigned long) frame;
 165         /* Store access registers to kernel stack of new process. */
 166         frame->childregs = *regs;
 167         frame->childregs.gprs[2] = 0;   /* child returns 0 on fork. */
 168         frame->childregs.gprs[15] = new_stackp;
 169         frame->sf.back_chain = 0;
 170
 171         /* new return point is ret_from_fork */
 172         frame->sf.gprs[8] = (unsigned long) ret_from_fork;
 173
 174         /* fake return stack for resume(), don't go back to schedule */
 175         frame->sf.gprs[9] = (unsigned long) frame;
 176
 177         /* Save access registers to new thread structure. */
 178         save_access_regs(&p->thread.acrs[0]);
 179
 180 #ifndef CONFIG_64BIT
 181         /*
 182          * save fprs to current->thread.fp_regs to merge them with
 183          * the emulated registers and then copy the result to the child.
 184          */
 185         save_fp_regs(&current->thread.fp_regs);
 186         memcpy(&p->thread.fp_regs, &current->thread.fp_regs,
 187                sizeof(s390_fp_regs));
 188         /* Set a new TLS ?  */
 189         if (clone_flags & CLONE_SETTLS)
 190                 p->thread.acrs[0] = regs->gprs[6];
 191 #else /* CONFIG_64BIT */
 192         /* Save the fpu registers to new thread structure. */
 193         save_fp_regs(&p->thread.fp_regs);
 194         /* Set a new TLS ?  */
 195         if (clone_flags & CLONE_SETTLS) {
 196                 if (is_compat_task()) {
 197                         p->thread.acrs[0] = (unsigned int) regs->gprs[6];
 198                 } else {
 199                         p->thread.acrs[0] = (unsigned int)(regs->gprs[6] >> 32);
 200                         p->thread.acrs[1] = (unsigned int) regs->gprs[6];
 201                 }
 202         }
 203 #endif /* CONFIG_64BIT */
 204         /* start new process with ar4 pointing to the correct address space */
 205         p->thread.mm_segment = get_fs();
 206         /* Don't copy debug registers */
 207         memset(&p->thread.per_user, 0, sizeof(p->thread.per_user));
 208         memset(&p->thread.per_event, 0, sizeof(p->thread.per_event));
 209         clear_tsk_thread_flag(p, TIF_SINGLE_STEP);
 210         clear_tsk_thread_flag(p, TIF_PER_TRAP);
 211         /* Initialize per thread user and system timer values */
 212         ti = task_thread_info(p);
 213         ti->user_timer = 0;
 214         ti->system_timer = 0;
 215         return 0;
 216 }
 217
 218 SYSCALL_DEFINE0(fork)
 219 {
 220         struct pt_regs *regs = task_pt_regs(current);
 221         return do_fork(SIGCHLD, regs->gprs[15], regs, 0, NULL, NULL);
 222 }
 223
 224 SYSCALL_DEFINE4(clone, unsigned long, newsp, unsigned long, clone_flags,
 225                 int __user *, parent_tidptr, int __user *, child_tidptr)
 226 {
 227         struct pt_regs *regs = task_pt_regs(current);
 228
 229         if (!newsp)
 230                 newsp = regs->gprs[15];
 231         return do_fork(clone_flags, newsp, regs, 0,
 232                        parent_tidptr, child_tidptr);
 233 }
 234
 235 /*
 236  * This is trivial, and on the face of it looks like it
 237  * could equally well be done in user mode.
 238  *
 239  * Not so, for quite unobvious reasons - register pressure.
 240  * In user mode vfork() cannot have a stack frame, and if
 241  * done by calling the "clone()" system call directly, you
 242  * do not have enough call-clobbered registers to hold all
 243  * the information you need.
 244  */
 245 SYSCALL_DEFINE0(vfork)
 246 {
 247         struct pt_regs *regs = task_pt_regs(current);
 248         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD,
 249                        regs->gprs[15], regs, 0, NULL, NULL);
 250 }
 251
 252 asmlinkage void execve_tail(void)
 253 {
 254         current->thread.fp_regs.fpc = 0;
 255         if (MACHINE_HAS_IEEE)
 256                 asm volatile("sfpc %0,%0" : : "d" (0));
 257 }
 258
 259 /*
 260  * sys_execve() executes a new program.
 261  */
 262 SYSCALL_DEFINE3(execve, const char __user *, name,
 263                 const char __user *const __user *, argv,
 264                 const char __user *const __user *, envp)
 265 {
 266         struct pt_regs *regs = task_pt_regs(current);
 267         char *filename;
 268         long rc;
 269
 270         filename = getname(name);
 271         rc = PTR_ERR(filename);
 272         if (IS_ERR(filename))
 273                 return rc;
 274         rc = do_execve(filename, argv, envp, regs);
 275         if (rc)
 276                 goto out;
 277         execve_tail();
 278         rc = regs->gprs[2];
 279 out:
 280         putname(filename);
 281         return rc;
 282 }
 283
 284 /*
 285  * fill in the FPU structure for a core dump.
 286  */
 287 int dump_fpu (struct pt_regs * regs, s390_fp_regs *fpregs)
 288 {
 289 #ifndef CONFIG_64BIT
 290         /*
 291          * save fprs to current->thread.fp_regs to merge them with
 292          * the emulated registers and then copy the result to the dump.
 293          */
 294         save_fp_regs(&current->thread.fp_regs);
 295         memcpy(fpregs, &current->thread.fp_regs, sizeof(s390_fp_regs));
 296 #else /* CONFIG_64BIT */
 297         save_fp_regs(fpregs);
 298 #endif /* CONFIG_64BIT */
 299         return 1;
 300 }
 301 EXPORT_SYMBOL(dump_fpu);
 302
 303 unsigned long get_wchan(struct task_struct *p)
 304 {
 305         struct stack_frame *sf, *low, *high;
 306         unsigned long return_address;
 307         int count;
 308
 309         if (!p || p == current || p->state == TASK_RUNNING || !task_stack_page(p))
 310                 return 0;
 311         low = task_stack_page(p);
 312         high = (struct stack_frame *) task_pt_regs(p);
 313         sf = (struct stack_frame *) (p->thread.ksp & PSW_ADDR_INSN);
 314         if (sf <= low || sf > high)
 315                 return 0;
 316         for (count = 0; count < 16; count++) {
 317                 sf = (struct stack_frame *) (sf->back_chain & PSW_ADDR_INSN);
 318                 if (sf <= low || sf > high)
 319                         return 0;
 320                 return_address = sf->gprs[8] & PSW_ADDR_INSN;
 321                 if (!in_sched_functions(return_address))
 322                         return return_address;
 323         }
 324         return 0;
 325 }
 326
 327 unsigned long arch_align_stack(unsigned long sp)
 328 {
 329         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 330                 sp -= get_random_int() & ~PAGE_MASK;
 331         return sp & ~0xf;
 332 }
 333
 334 static inline unsigned long brk_rnd(void)
 335 {
 336         /* 8MB for 32bit, 1GB for 64bit */
 337         if (is_32bit_task())
 338                 return (get_random_int() & 0x7ffUL) << PAGE_SHIFT;
 339         else
 340                 return (get_random_int() & 0x3ffffUL) << PAGE_SHIFT;
 341 }
 342
 343 unsigned long arch_randomize_brk(struct mm_struct *mm)
 344 {
 345         unsigned long ret = PAGE_ALIGN(mm->brk + brk_rnd());
 346
 347         if (ret < mm->brk)
 348                 return mm->brk;
 349         return ret;
 350 }
 351
 352 unsigned long randomize_et_dyn(unsigned long base)
 353 {
 354         unsigned long ret = PAGE_ALIGN(base + brk_rnd());
 355
 356         if (!(current->flags & PF_RANDOMIZE))
 357                 return base;
 358         if (ret < base)
 359                 return base;
 360         return ret;
 361 }