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/compat.h>
  33 #include <asm/smp.h>
  34 #include "entry.h"
  35
  36 asmlinkage void ret_from_fork(void) asm ("ret_from_fork");
  37
  38 /*
  39  * Return saved PC of a blocked thread. used in kernel/sched.
  40  * resume in entry.S does not create a new stack frame, it
  41  * just stores the registers %r6-%r15 to the frame given by
  42  * schedule. We want to return the address of the caller of
  43  * schedule, so we have to walk the backchain one time to
  44  * find the frame schedule() store its return address.
  45  */
  46 unsigned long thread_saved_pc(struct task_struct *tsk)
  47 {
  48         struct stack_frame *sf, *low, *high;
  49
  50         if (!tsk || !task_stack_page(tsk))
  51                 return 0;
  52         low = task_stack_page(tsk);
  53         high = (struct stack_frame *) task_pt_regs(tsk);
  54         sf = (struct stack_frame *) (tsk->thread.ksp & PSW_ADDR_INSN);
  55         if (sf <= low || sf > high)
  56                 return 0;
  57         sf = (struct stack_frame *) (sf->back_chain & PSW_ADDR_INSN);
  58         if (sf <= low || sf > high)
  59                 return 0;
  60         return sf->gprs[8];
  61 }
  62
  63 /*
  64  * The idle loop on a S390...
  65  */
  66 static void default_idle(void)
  67 {
  68         if (cpu_is_offline(smp_processor_id()))
  69                 cpu_die();
  70         local_irq_disable();
  71         if (need_resched()) {
  72                 local_irq_enable();
  73                 return;
  74         }
  75         local_mcck_disable();
  76         if (test_thread_flag(TIF_MCCK_PENDING)) {
  77                 local_mcck_enable();
  78                 local_irq_enable();
  79                 return;
  80         }
  81         trace_hardirqs_on();
  82         /* Don't trace preempt off for idle. */
  83         stop_critical_timings();
  84         /* Stop virtual timer and halt the cpu. */
  85         vtime_stop_cpu();
  86         /* Reenable preemption tracer. */
  87         start_critical_timings();
  88 }
  89
  90 void cpu_idle(void)
  91 {
  92         for (;;) {
  93                 tick_nohz_idle_enter();
  94                 rcu_idle_enter();
  95                 while (!need_resched() && !test_thread_flag(TIF_MCCK_PENDING))
  96                         default_idle();
  97                 rcu_idle_exit();
  98                 tick_nohz_idle_exit();
  99                 if (test_thread_flag(TIF_MCCK_PENDING))
 100                         s390_handle_mcck();
 101                 preempt_enable_no_resched();
 102                 schedule();
 103                 preempt_disable();
 104         }
 105 }
 106
 107 extern void __kprobes kernel_thread_starter(void);
 108
 109 asm(
 110         ".section .kprobes.text, \"ax\"\n"
 111         ".global kernel_thread_starter\n"
 112         "kernel_thread_starter:\n"
 113         "    la    2,0(10)\n"
 114         "    basr  14,9\n"
 115         "    la    2,0\n"
 116         "    br    11\n"
 117         ".previous\n");
 118
 119 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 120 {
 121         struct pt_regs regs;
 122
 123         memset(&regs, 0, sizeof(regs));
 124         regs.psw.mask = psw_kernel_bits |
 125                 PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT | PSW_MASK_MCHECK;
 126         regs.psw.addr = (unsigned long) kernel_thread_starter | PSW_ADDR_AMODE;
 127         regs.gprs[9] = (unsigned long) fn;
 128         regs.gprs[10] = (unsigned long) arg;
 129         regs.gprs[11] = (unsigned long) do_exit;
 130         regs.orig_gpr2 = -1;
 131
 132         /* Ok, create the new process.. */
 133         return do_fork(flags | CLONE_VM | CLONE_UNTRACED,
 134                        0, &regs, 0, NULL, NULL);
 135 }
 136 EXPORT_SYMBOL(kernel_thread);
 137
 138 /*
 139  * Free current thread data structures etc..
 140  */
 141 void exit_thread(void)
 142 {
 143 }
 144
 145 void flush_thread(void)
 146 {
 147 }
 148
 149 void release_thread(struct task_struct *dead_task)
 150 {
 151 }
 152
 153 int copy_thread(unsigned long clone_flags, unsigned long new_stackp,
 154                 unsigned long unused,
 155                 struct task_struct *p, struct pt_regs *regs)
 156 {
 157         struct thread_info *ti;
 158         struct fake_frame
 159         {
 160                 struct stack_frame sf;
 161                 struct pt_regs childregs;
 162         } *frame;
 163
 164         frame = container_of(task_pt_regs(p), struct fake_frame, childregs);
 165         p->thread.ksp = (unsigned long) frame;
 166         /* Store access registers to kernel stack of new process. */
 167         frame->childregs = *regs;
 168         frame->childregs.gprs[2] = 0;   /* child returns 0 on fork. */
 169         frame->childregs.gprs[15] = new_stackp;
 170         frame->sf.back_chain = 0;
 171
 172         /* new return point is ret_from_fork */
 173         frame->sf.gprs[8] = (unsigned long) ret_from_fork;
 174
 175         /* fake return stack for resume(), don't go back to schedule */
 176         frame->sf.gprs[9] = (unsigned long) frame;
 177
 178         /* Save access registers to new thread structure. */
 179         save_access_regs(&p->thread.acrs[0]);
 180
 181 #ifndef CONFIG_64BIT
 182         /*
 183          * save fprs to current->thread.fp_regs to merge them with
 184          * the emulated registers and then copy the result to the child.
 185          */
 186         save_fp_regs(&current->thread.fp_regs);
 187         memcpy(&p->thread.fp_regs, &current->thread.fp_regs,
 188                sizeof(s390_fp_regs));
 189         /* Set a new TLS ?  */
 190         if (clone_flags & CLONE_SETTLS)
 191                 p->thread.acrs[0] = regs->gprs[6];
 192 #else /* CONFIG_64BIT */
 193         /* Save the fpu registers to new thread structure. */
 194         save_fp_regs(&p->thread.fp_regs);
 195         /* Set a new TLS ?  */
 196         if (clone_flags & CLONE_SETTLS) {
 197                 if (is_compat_task()) {
 198                         p->thread.acrs[0] = (unsigned int) regs->gprs[6];
 199                 } else {
 200                         p->thread.acrs[0] = (unsigned int)(regs->gprs[6] >> 32);
 201                         p->thread.acrs[1] = (unsigned int) regs->gprs[6];
 202                 }
 203         }
 204 #endif /* CONFIG_64BIT */
 205         /* start new process with ar4 pointing to the correct address space */
 206         p->thread.mm_segment = get_fs();
 207         /* Don't copy debug registers */
 208         memset(&p->thread.per_user, 0, sizeof(p->thread.per_user));
 209         memset(&p->thread.per_event, 0, sizeof(p->thread.per_event));
 210         clear_tsk_thread_flag(p, TIF_SINGLE_STEP);
 211         clear_tsk_thread_flag(p, TIF_PER_TRAP);
 212         /* Initialize per thread user and system timer values */
 213         ti = task_thread_info(p);
 214         ti->user_timer = 0;
 215         ti->system_timer = 0;
 216         return 0;
 217 }
 218
 219 SYSCALL_DEFINE0(fork)
 220 {
 221         struct pt_regs *regs = task_pt_regs(current);
 222         return do_fork(SIGCHLD, regs->gprs[15], regs, 0, NULL, NULL);
 223 }
 224
 225 SYSCALL_DEFINE4(clone, unsigned long, newsp, unsigned long, clone_flags,
 226                 int __user *, parent_tidptr, int __user *, child_tidptr)
 227 {
 228         struct pt_regs *regs = task_pt_regs(current);
 229
 230         if (!newsp)
 231                 newsp = regs->gprs[15];
 232         return do_fork(clone_flags, newsp, regs, 0,
 233                        parent_tidptr, child_tidptr);
 234 }
 235
 236 /*
 237  * This is trivial, and on the face of it looks like it
 238  * could equally well be done in user mode.
 239  *
 240  * Not so, for quite unobvious reasons - register pressure.
 241  * In user mode vfork() cannot have a stack frame, and if
 242  * done by calling the "clone()" system call directly, you
 243  * do not have enough call-clobbered registers to hold all
 244  * the information you need.
 245  */
 246 SYSCALL_DEFINE0(vfork)
 247 {
 248         struct pt_regs *regs = task_pt_regs(current);
 249         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD,
 250                        regs->gprs[15], regs, 0, NULL, NULL);
 251 }
 252
 253 asmlinkage void execve_tail(void)
 254 {
 255         current->thread.fp_regs.fpc = 0;
 256         if (MACHINE_HAS_IEEE)
 257                 asm volatile("sfpc %0,%0" : : "d" (0));
 258 }
 259
 260 /*
 261  * sys_execve() executes a new program.
 262  */
 263 SYSCALL_DEFINE3(execve, const char __user *, name,
 264                 const char __user *const __user *, argv,
 265                 const char __user *const __user *, envp)
 266 {
 267         struct pt_regs *regs = task_pt_regs(current);
 268         char *filename;
 269         long rc;
 270
 271         filename = getname(name);
 272         rc = PTR_ERR(filename);
 273         if (IS_ERR(filename))
 274                 return rc;
 275         rc = do_execve(filename, argv, envp, regs);
 276         if (rc)
 277                 goto out;
 278         execve_tail();
 279         rc = regs->gprs[2];
 280 out:
 281         putname(filename);
 282         return rc;
 283 }
 284
 285 /*
 286  * fill in the FPU structure for a core dump.
 287  */
 288 int dump_fpu (struct pt_regs * regs, s390_fp_regs *fpregs)
 289 {
 290 #ifndef CONFIG_64BIT
 291         /*
 292          * save fprs to current->thread.fp_regs to merge them with
 293          * the emulated registers and then copy the result to the dump.
 294          */
 295         save_fp_regs(&current->thread.fp_regs);
 296         memcpy(fpregs, &current->thread.fp_regs, sizeof(s390_fp_regs));
 297 #else /* CONFIG_64BIT */
 298         save_fp_regs(fpregs);
 299 #endif /* CONFIG_64BIT */
 300         return 1;
 301 }
 302 EXPORT_SYMBOL(dump_fpu);
 303
 304 unsigned long get_wchan(struct task_struct *p)
 305 {
 306         struct stack_frame *sf, *low, *high;
 307         unsigned long return_address;
 308         int count;
 309
 310         if (!p || p == current || p->state == TASK_RUNNING || !task_stack_page(p))
 311                 return 0;
 312         low = task_stack_page(p);
 313         high = (struct stack_frame *) task_pt_regs(p);
 314         sf = (struct stack_frame *) (p->thread.ksp & PSW_ADDR_INSN);
 315         if (sf <= low || sf > high)
 316                 return 0;
 317         for (count = 0; count < 16; count++) {
 318                 sf = (struct stack_frame *) (sf->back_chain & PSW_ADDR_INSN);
 319                 if (sf <= low || sf > high)
 320                         return 0;
 321                 return_address = sf->gprs[8] & PSW_ADDR_INSN;
 322                 if (!in_sched_functions(return_address))
 323                         return return_address;
 324         }
 325         return 0;
 326 }
 327
 328 unsigned long arch_align_stack(unsigned long sp)
 329 {
 330         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 331                 sp -= get_random_int() & ~PAGE_MASK;
 332         return sp & ~0xf;
 333 }
 334
 335 static inline unsigned long brk_rnd(void)
 336 {
 337         /* 8MB for 32bit, 1GB for 64bit */
 338         if (is_32bit_task())
 339                 return (get_random_int() & 0x7ffUL) << PAGE_SHIFT;
 340         else
 341                 return (get_random_int() & 0x3ffffUL) << PAGE_SHIFT;
 342 }
 343
 344 unsigned long arch_randomize_brk(struct mm_struct *mm)
 345 {
 346         unsigned long ret = PAGE_ALIGN(mm->brk + brk_rnd());
 347
 348         if (ret < mm->brk)
 349                 return mm->brk;
 350         return ret;
 351 }
 352
 353 unsigned long randomize_et_dyn(unsigned long base)
 354 {
 355         unsigned long ret = PAGE_ALIGN(base + brk_rnd());
 356
 357         if (!(current->flags & PF_RANDOMIZE))
 358                 return base;
 359         if (ret < base)
 360                 return base;
 361         return ret;
 362 }