arch/xtensa/kernel/process.c

   1 /*
   2  * arch/xtensa/kernel/process.c
   3  *
   4  * Xtensa Processor version.
   5  *
   6  * This file is subject to the terms and conditions of the GNU General Public
   7  * License.  See the file "COPYING" in the main directory of this archive
   8  * for more details.
   9  *
  10  * Copyright (C) 2001 - 2005 Tensilica Inc.
  11  *
  12  * Joe Taylor <joe@tensilica.com, joetylr@yahoo.com>
  13  * Chris Zankel <chris@zankel.net>
  14  * Marc Gauthier <marc@tensilica.com, marc@alumni.uwaterloo.ca>
  15  * Kevin Chea
  16  */
  17
  18 #include <linux/errno.h>
  19 #include <linux/sched.h>
  20 #include <linux/sched/debug.h>
  21 #include <linux/kernel.h>
  22 #include <linux/mm.h>
  23 #include <linux/smp.h>
  24 #include <linux/stddef.h>
  25 #include <linux/unistd.h>
  26 #include <linux/ptrace.h>
  27 #include <linux/elf.h>
  28 #include <linux/hw_breakpoint.h>
  29 #include <linux/init.h>
  30 #include <linux/prctl.h>
  31 #include <linux/init_task.h>
  32 #include <linux/module.h>
  33 #include <linux/mqueue.h>
  34 #include <linux/fs.h>
  35 #include <linux/slab.h>
  36 #include <linux/rcupdate.h>
  37
  38 #include <asm/pgtable.h>
  39 #include <linux/uaccess.h>
  40 #include <asm/io.h>
  41 #include <asm/processor.h>
  42 #include <asm/platform.h>
  43 #include <asm/mmu.h>
  44 #include <asm/irq.h>
  45 #include <linux/atomic.h>
  46 #include <asm/asm-offsets.h>
  47 #include <asm/regs.h>
  48 #include <asm/hw_breakpoint.h>
  49
  50 extern void ret_from_fork(void);
  51 extern void ret_from_kernel_thread(void);
  52
  53 struct task_struct *current_set[NR_CPUS] = {&init_task, };
  54
  55 void (*pm_power_off)(void) = NULL;
  56 EXPORT_SYMBOL(pm_power_off);
  57
  58
  59 #if XTENSA_HAVE_COPROCESSORS
  60
  61 void coprocessor_release_all(struct thread_info *ti)
  62 {
  63         unsigned long cpenable;
  64         int i;
  65
  66         /* Make sure we don't switch tasks during this operation. */
  67
  68         preempt_disable();
  69
  70         /* Walk through all cp owners and release it for the requested one. */
  71
  72         cpenable = ti->cpenable;
  73
  74         for (i = 0; i < XCHAL_CP_MAX; i++) {
  75                 if (coprocessor_owner[i] == ti) {
  76                         coprocessor_owner[i] = 0;
  77                         cpenable &= ~(1 << i);
  78                 }
  79         }
  80
  81         ti->cpenable = cpenable;
  82         coprocessor_clear_cpenable();
  83
  84         preempt_enable();
  85 }
  86
  87 void coprocessor_flush_all(struct thread_info *ti)
  88 {
  89         unsigned long cpenable;
  90         int i;
  91
  92         preempt_disable();
  93
  94         cpenable = ti->cpenable;
  95
  96         for (i = 0; i < XCHAL_CP_MAX; i++) {
  97                 if ((cpenable & 1) != 0 && coprocessor_owner[i] == ti)
  98                         coprocessor_flush(ti, i);
  99                 cpenable >>= 1;
 100         }
 101
 102         preempt_enable();
 103 }
 104
 105 #endif
 106
 107
 108 /*
 109  * Powermanagement idle function, if any is provided by the platform.
 110  */
 111 void arch_cpu_idle(void)
 112 {
 113         platform_idle();
 114 }
 115
 116 /*
 117  * This is called when the thread calls exit().
 118  */
 119 void exit_thread(struct task_struct *tsk)
 120 {
 121 #if XTENSA_HAVE_COPROCESSORS
 122         coprocessor_release_all(task_thread_info(tsk));
 123 #endif
 124 }
 125
 126 /*
 127  * Flush thread state. This is called when a thread does an execve()
 128  * Note that we flush coprocessor registers for the case execve fails.
 129  */
 130 void flush_thread(void)
 131 {
 132 #if XTENSA_HAVE_COPROCESSORS
 133         struct thread_info *ti = current_thread_info();
 134         coprocessor_flush_all(ti);
 135         coprocessor_release_all(ti);
 136 #endif
 137         flush_ptrace_hw_breakpoint(current);
 138 }
 139
 140 /*
 141  * this gets called so that we can store coprocessor state into memory and
 142  * copy the current task into the new thread.
 143  */
 144 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 145 {
 146 #if XTENSA_HAVE_COPROCESSORS
 147         coprocessor_flush_all(task_thread_info(src));
 148 #endif
 149         *dst = *src;
 150         return 0;
 151 }
 152
 153 /*
 154  * Copy thread.
 155  *
 156  * There are two modes in which this function is called:
 157  * 1) Userspace thread creation,
 158  *    regs != NULL, usp_thread_fn is userspace stack pointer.
 159  *    It is expected to copy parent regs (in case CLONE_VM is not set
 160  *    in the clone_flags) and set up passed usp in the childregs.
 161  * 2) Kernel thread creation,
 162  *    regs == NULL, usp_thread_fn is the function to run in the new thread
 163  *    and thread_fn_arg is its parameter.
 164  *    childregs are not used for the kernel threads.
 165  *
 166  * The stack layout for the new thread looks like this:
 167  *
 168  *      +------------------------+
 169  *      |       childregs        |
 170  *      +------------------------+ <- thread.sp = sp in dummy-frame
 171  *      |      dummy-frame       |    (saved in dummy-frame spill-area)
 172  *      +------------------------+
 173  *
 174  * We create a dummy frame to return to either ret_from_fork or
 175  *   ret_from_kernel_thread:
 176  *   a0 points to ret_from_fork/ret_from_kernel_thread (simulating a call4)
 177  *   sp points to itself (thread.sp)
 178  *   a2, a3 are unused for userspace threads,
 179  *   a2 points to thread_fn, a3 holds thread_fn arg for kernel threads.
 180  *
 181  * Note: This is a pristine frame, so we don't need any spill region on top of
 182  *       childregs.
 183  *
 184  * The fun part:  if we're keeping the same VM (i.e. cloning a thread,
 185  * not an entire process), we're normally given a new usp, and we CANNOT share
 186  * any live address register windows.  If we just copy those live frames over,
 187  * the two threads (parent and child) will overflow the same frames onto the
 188  * parent stack at different times, likely corrupting the parent stack (esp.
 189  * if the parent returns from functions that called clone() and calls new
 190  * ones, before the child overflows its now old copies of its parent windows).
 191  * One solution is to spill windows to the parent stack, but that's fairly
 192  * involved.  Much simpler to just not copy those live frames across.
 193  */
 194
 195 int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
 196                 unsigned long thread_fn_arg, struct task_struct *p)
 197 {
 198         struct pt_regs *childregs = task_pt_regs(p);
 199
 200 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
 201         struct thread_info *ti;
 202 #endif
 203
 204         /* Create a call4 dummy-frame: a0 = 0, a1 = childregs. */
 205         *((int*)childregs - 3) = (unsigned long)childregs;
 206         *((int*)childregs - 4) = 0;
 207
 208         p->thread.sp = (unsigned long)childregs;
 209
 210         if (!(p->flags & PF_KTHREAD)) {
 211                 struct pt_regs *regs = current_pt_regs();
 212                 unsigned long usp = usp_thread_fn ?
 213                         usp_thread_fn : regs->areg[1];
 214
 215                 p->thread.ra = MAKE_RA_FOR_CALL(
 216                                 (unsigned long)ret_from_fork, 0x1);
 217
 218                 /* This does not copy all the regs.
 219                  * In a bout of brilliance or madness,
 220                  * ARs beyond a0-a15 exist past the end of the struct.
 221                  */
 222                 *childregs = *regs;
 223                 childregs->areg[1] = usp;
 224                 childregs->areg[2] = 0;
 225
 226                 /* When sharing memory with the parent thread, the child
 227                    usually starts on a pristine stack, so we have to reset
 228                    windowbase, windowstart and wmask.
 229                    (Note that such a new thread is required to always create
 230                    an initial call4 frame)
 231                    The exception is vfork, where the new thread continues to
 232                    run on the parent's stack until it calls execve. This could
 233                    be a call8 or call12, which requires a legal stack frame
 234                    of the previous caller for the overflow handlers to work.
 235                    (Note that it's always legal to overflow live registers).
 236                    In this case, ensure to spill at least the stack pointer
 237                    of that frame. */
 238
 239                 if (clone_flags & CLONE_VM) {
 240                         /* check that caller window is live and same stack */
 241                         int len = childregs->wmask & ~0xf;
 242                         if (regs->areg[1] == usp && len != 0) {
 243                                 int callinc = (regs->areg[0] >> 30) & 3;
 244                                 int caller_ars = XCHAL_NUM_AREGS - callinc * 4;
 245                                 put_user(regs->areg[caller_ars+1],
 246                                          (unsigned __user*)(usp - 12));
 247                         }
 248                         childregs->wmask = 1;
 249                         childregs->windowstart = 1;
 250                         childregs->windowbase = 0;
 251                 } else {
 252                         int len = childregs->wmask & ~0xf;
 253                         memcpy(&childregs->areg[XCHAL_NUM_AREGS - len/4],
 254                                &regs->areg[XCHAL_NUM_AREGS - len/4], len);
 255                 }
 256
 257                 /* The thread pointer is passed in the '4th argument' (= a5) */
 258                 if (clone_flags & CLONE_SETTLS)
 259                         childregs->threadptr = childregs->areg[5];
 260         } else {
 261                 p->thread.ra = MAKE_RA_FOR_CALL(
 262                                 (unsigned long)ret_from_kernel_thread, 1);
 263
 264                 /* pass parameters to ret_from_kernel_thread:
 265                  * a2 = thread_fn, a3 = thread_fn arg
 266                  */
 267                 *((int *)childregs - 1) = thread_fn_arg;
 268                 *((int *)childregs - 2) = usp_thread_fn;
 269
 270                 /* Childregs are only used when we're going to userspace
 271                  * in which case start_thread will set them up.
 272                  */
 273         }
 274
 275 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
 276         ti = task_thread_info(p);
 277         ti->cpenable = 0;
 278 #endif
 279
 280         clear_ptrace_hw_breakpoint(p);
 281
 282         return 0;
 283 }
 284
 285
 286 /*
 287  * These bracket the sleeping functions..
 288  */
 289
 290 unsigned long get_wchan(struct task_struct *p)
 291 {
 292         unsigned long sp, pc;
 293         unsigned long stack_page = (unsigned long) task_stack_page(p);
 294         int count = 0;
 295
 296         if (!p || p == current || p->state == TASK_RUNNING)
 297                 return 0;
 298
 299         sp = p->thread.sp;
 300         pc = MAKE_PC_FROM_RA(p->thread.ra, p->thread.sp);
 301
 302         do {
 303                 if (sp < stack_page + sizeof(struct task_struct) ||
 304                     sp >= (stack_page + THREAD_SIZE) ||
 305                     pc == 0)
 306                         return 0;
 307                 if (!in_sched_functions(pc))
 308                         return pc;
 309
 310                 /* Stack layout: sp-4: ra, sp-3: sp' */
 311
 312                 pc = MAKE_PC_FROM_RA(*(unsigned long*)sp - 4, sp);
 313                 sp = *(unsigned long *)sp - 3;
 314         } while (count++ < 16);
 315         return 0;
 316 }
 317
 318 /*
 319  * xtensa_gregset_t and 'struct pt_regs' are vastly different formats
 320  * of processor registers.  Besides different ordering,
 321  * xtensa_gregset_t contains non-live register information that
 322  * 'struct pt_regs' does not.  Exception handling (primarily) uses
 323  * 'struct pt_regs'.  Core files and ptrace use xtensa_gregset_t.
 324  *
 325  */
 326
 327 void xtensa_elf_core_copy_regs (xtensa_gregset_t *elfregs, struct pt_regs *regs)
 328 {
 329         unsigned long wb, ws, wm;
 330         int live, last;
 331
 332         wb = regs->windowbase;
 333         ws = regs->windowstart;
 334         wm = regs->wmask;
 335         ws = ((ws >> wb) | (ws << (WSBITS - wb))) & ((1 << WSBITS) - 1);
 336
 337         /* Don't leak any random bits. */
 338
 339         memset(elfregs, 0, sizeof(*elfregs));
 340
 341         /* Note:  PS.EXCM is not set while user task is running; its
 342          * being set in regs->ps is for exception handling convenience.
 343          */
 344
 345         elfregs->pc             = regs->pc;
 346         elfregs->ps             = (regs->ps & ~(1 << PS_EXCM_BIT));
 347         elfregs->lbeg           = regs->lbeg;
 348         elfregs->lend           = regs->lend;
 349         elfregs->lcount         = regs->lcount;
 350         elfregs->sar            = regs->sar;
 351         elfregs->windowstart    = ws;
 352
 353         live = (wm & 2) ? 4 : (wm & 4) ? 8 : (wm & 8) ? 12 : 16;
 354         last = XCHAL_NUM_AREGS - (wm >> 4) * 4;
 355         memcpy(elfregs->a, regs->areg, live * 4);
 356         memcpy(elfregs->a + last, regs->areg + last, (wm >> 4) * 16);
 357 }
 358
 359 int dump_fpu(void)
 360 {
 361         return 0;
 362 }