linux-2.6-590-chopstix-intern.patch

   1 diff -Nurb linux-2.6.22-580/arch/i386/Kconfig linux-2.6.22-590/arch/i386/Kconfig
   2 --- linux-2.6.22-580/arch/i386/Kconfig  2009-04-08 16:36:16.000000000 -0400
   3 +++ linux-2.6.22-590/arch/i386/Kconfig  2009-04-08 16:37:10.000000000 -0400
   4 @@ -1217,6 +1217,14 @@
   5
   6  source "arch/i386/oprofile/Kconfig"
   7
   8 +config CHOPSTIX
   9 +       bool "Chopstix (PlanetLab)"
  10 +       depends on MODULES && OPROFILE
  11 +       help
  12 +         Chopstix allows you to monitor various events by summarizing them
  13 +         in lossy data structures and transferring these data structures
  14 +         into user space. If in doubt, say "N".
  15 +
  16  config KPROBES
  17         bool "Kprobes (EXPERIMENTAL)"
  18         depends on KALLSYMS && EXPERIMENTAL && MODULES
  19 diff -Nurb linux-2.6.22-580/arch/i386/kernel/asm-offsets.c linux-2.6.22-590/arch/i386/kernel/asm-offsets.c
  20 --- linux-2.6.22-580/arch/i386/kernel/asm-offsets.c     2007-07-08 19:32:17.000000000 -0400
  21 +++ linux-2.6.22-590/arch/i386/kernel/asm-offsets.c     2009-04-08 16:37:10.000000000 -0400
  22 @@ -9,6 +9,7 @@
  23  #include <linux/signal.h>
  24  #include <linux/personality.h>
  25  #include <linux/suspend.h>
  26 +#include <linux/arrays.h>
  27  #include <asm/ucontext.h>
  28  #include "sigframe.h"
  29  #include <asm/pgtable.h>
  30 @@ -25,9 +26,19 @@
  31  #define OFFSET(sym, str, mem) \
  32         DEFINE(sym, offsetof(struct str, mem));
  33
  34 +#define STACKOFFSET(sym, str, mem) \
  35 +       DEFINE(sym, offsetof(struct str, mem)-sizeof(struct str));
  36 +
  37  /* workaround for a warning with -Wmissing-prototypes */
  38  void foo(void);
  39
  40 +struct event_spec {
  41 +       unsigned long pc;
  42 +       unsigned long dcookie;
  43 +       unsigned count;
  44 +       unsigned int number;
  45 +};
  46 +
  47  void foo(void)
  48  {
  49         OFFSET(SIGCONTEXT_eax, sigcontext, eax);
  50 @@ -51,7 +62,16 @@
  51         OFFSET(CPUINFO_x86_vendor_id, cpuinfo_x86, x86_vendor_id);
  52         BLANK();
  53
  54 -       OFFSET(TI_task, thread_info, task);
  55 +    STACKOFFSET(TASK_thread, task_struct, thread);
  56 +    STACKOFFSET(THREAD_esp, thread_struct, esp);
  57 +    STACKOFFSET(EVENT_event_data, event, event_data);
  58 +    STACKOFFSET(EVENT_task, event, task);
  59 +    STACKOFFSET(EVENT_event_type, event, event_type);
  60 +    STACKOFFSET(SPEC_number, event_spec, number);
  61 +    DEFINE(EVENT_SIZE, sizeof(struct event));
  62 +    DEFINE(SPEC_SIZE, sizeof(struct event_spec));
  63 +    DEFINE(SPEC_EVENT_SIZE, sizeof(struct event_spec)+sizeof(struct event));
  64 +
  65         OFFSET(TI_exec_domain, thread_info, exec_domain);
  66         OFFSET(TI_flags, thread_info, flags);
  67         OFFSET(TI_status, thread_info, status);
  68 diff -Nurb linux-2.6.22-580/arch/i386/mm/fault.c linux-2.6.22-590/arch/i386/mm/fault.c
  69 --- linux-2.6.22-580/arch/i386/mm/fault.c       2009-04-08 16:36:16.000000000 -0400
  70 +++ linux-2.6.22-590/arch/i386/mm/fault.c       2009-04-08 16:37:10.000000000 -0400
  71 @@ -60,6 +60,15 @@
  72                                           DIE_PAGE_FAULT, &args);
  73  }
  74
  75 +
  76 +extern void (*rec_event)(void *,unsigned int);
  77 +struct event_spec {
  78 +       unsigned long pc;
  79 +       unsigned long dcookie;
  80 +       unsigned count;
  81 +       unsigned char reason;
  82 +};
  83 +
  84  /*
  85   * Return EIP plus the CS segment base.  The segment limit is also
  86   * adjusted, clamped to the kernel/user address space (whichever is
  87 @@ -296,6 +305,8 @@
  88   *     bit 3 == 1 means use of reserved bit detected
  89   *     bit 4 == 1 means fault was an instruction fetch
  90   */
  91 +
  92 +
  93  fastcall void __kprobes do_page_fault(struct pt_regs *regs,
  94                                       unsigned long error_code)
  95  {
  96 diff -Nurb linux-2.6.22-580/block/ll_rw_blk.c linux-2.6.22-590/block/ll_rw_blk.c
  97 --- linux-2.6.22-580/block/ll_rw_blk.c  2009-04-08 16:36:09.000000000 -0400
  98 +++ linux-2.6.22-590/block/ll_rw_blk.c  2009-04-08 16:37:10.000000000 -0400
  99 @@ -30,6 +30,7 @@
 100  #include <linux/cpu.h>
 101  #include <linux/blktrace_api.h>
 102  #include <linux/fault-inject.h>
 103 +#include <linux/arrays.h>
 104
 105  /*
 106   * for max sense size
 107 @@ -3102,6 +3103,13 @@
 108
 109  #endif /* CONFIG_FAIL_MAKE_REQUEST */
 110
 111 +extern void (*rec_event)(void *,unsigned int);
 112 +struct event_spec {
 113 +       unsigned long pc;
 114 +       unsigned long dcookie;
 115 +       unsigned count;
 116 +       unsigned char reason;
 117 +};
 118  /**
 119   * generic_make_request: hand a buffer to its device driver for I/O
 120   * @bio:  The bio describing the location in memory and on the device.
 121 @@ -3220,7 +3228,23 @@
 122                                 goto end_io;
 123                         }
 124                 }
 125 -
 126 +#ifdef CONFIG_CHOPSTIX
 127 +               if (rec_event) {
 128 +                       struct event event;
 129 +                       struct event_spec espec;
 130 +                       unsigned long eip;
 131 +
 132 +                       espec.reason = 0;/*request */
 133 +
 134 +                       eip = bio->bi_end_io;
 135 +                       event.event_data=&espec;
 136 +                       espec.pc=eip;
 137 +                       event.event_type=3;
 138 +                       /* index in the event array currently set up */
 139 +                       /* make sure the counters are loaded in the order we want them to show up*/
 140 +                       (*rec_event)(&event, bio->bi_size);
 141 +               }
 142 +#endif
 143                 ret = q->make_request_fn(q, bio);
 144         } while (ret);
 145  }
 146 diff -Nurb linux-2.6.22-580/drivers/oprofile/cpu_buffer.c linux-2.6.22-590/drivers/oprofile/cpu_buffer.c
 147 --- linux-2.6.22-580/drivers/oprofile/cpu_buffer.c      2007-07-08 19:32:17.000000000 -0400
 148 +++ linux-2.6.22-590/drivers/oprofile/cpu_buffer.c      2009-04-08 16:37:10.000000000 -0400
 149 @@ -21,6 +21,7 @@
 150  #include <linux/oprofile.h>
 151  #include <linux/vmalloc.h>
 152  #include <linux/errno.h>
 153 +#include <linux/arrays.h>
 154
 155  #include "event_buffer.h"
 156  #include "cpu_buffer.h"
 157 @@ -143,6 +144,17 @@
 158                 b->head_pos = 0;
 159  }
 160
 161 +#ifdef CONFIG_CHOPSTIX
 162 +
 163 +struct event_spec {
 164 +       unsigned int pc;
 165 +       unsigned long dcookie;
 166 +       unsigned count;
 167 +};
 168 +
 169 +extern void (*rec_event)(void *,unsigned int);
 170 +#endif
 171 +
 172  static inline void
 173  add_sample(struct oprofile_cpu_buffer * cpu_buf,
 174             unsigned long pc, unsigned long event)
 175 @@ -151,6 +163,7 @@
 176         entry->eip = pc;
 177         entry->event = event;
 178         increment_head(cpu_buf);
 179 +
 180  }
 181
 182  static inline void
 183 @@ -241,8 +254,28 @@
 184  {
 185         int is_kernel = !user_mode(regs);
 186         unsigned long pc = profile_pc(regs);
 187 +       int res=0;
 188
 189 +#ifdef CONFIG_CHOPSTIX
 190 +       if (rec_event) {
 191 +               struct event esig;
 192 +               struct event_spec espec;
 193 +               esig.task = current;
 194 +               espec.pc=pc;
 195 +               espec.count=1;
 196 +               esig.event_data=&espec;
 197 +               esig.event_type=event; /* index in the event array currently set up */
 198 +                                       /* make sure the counters are loaded in the order we want them to show up*/
 199 +               (*rec_event)(&esig, 1);
 200 +       }
 201 +       else {
 202         oprofile_add_ext_sample(pc, regs, event, is_kernel);
 203 +       }
 204 +#else
 205 +       oprofile_add_ext_sample(pc, regs, event, is_kernel);
 206 +#endif
 207 +
 208 +
 209  }
 210
 211  void oprofile_add_pc(unsigned long pc, int is_kernel, unsigned long event)
 212 diff -Nurb linux-2.6.22-580/fs/bio.c linux-2.6.22-590/fs/bio.c
 213 --- linux-2.6.22-580/fs/bio.c   2007-07-08 19:32:17.000000000 -0400
 214 +++ linux-2.6.22-590/fs/bio.c   2009-04-08 16:37:10.000000000 -0400
 215 @@ -27,6 +27,7 @@
 216  #include <linux/workqueue.h>
 217  #include <linux/blktrace_api.h>
 218  #include <scsi/sg.h>           /* for struct sg_iovec */
 219 +#include <linux/arrays.h>
 220
 221  #define BIO_POOL_SIZE 2
 222
 223 @@ -47,6 +48,7 @@
 224         struct kmem_cache *slab;
 225  };
 226
 227 +
 228  /*
 229   * if you change this list, also change bvec_alloc or things will
 230   * break badly! cannot be bigger than what you can fit into an
 231 @@ -999,6 +1001,14 @@
 232         }
 233  }
 234
 235 +struct event_spec {
 236 +       unsigned long pc;
 237 +       unsigned long dcookie;
 238 +       unsigned count;
 239 +       unsigned char reason;
 240 +};
 241 +
 242 +extern void (*rec_event)(void *,unsigned int);
 243  /**
 244   * bio_endio - end I/O on a bio
 245   * @bio:       bio
 246 @@ -1028,6 +1038,24 @@
 247         bio->bi_size -= bytes_done;
 248         bio->bi_sector += (bytes_done >> 9);
 249
 250 +#ifdef CONFIG_CHOPSTIX
 251 +               if (rec_event) {
 252 +                       struct event event;
 253 +                       struct event_spec espec;
 254 +                       unsigned long eip;
 255 +
 256 +                       espec.reason = 1;/*response */
 257 +
 258 +                       eip = bio->bi_end_io;
 259 +                       event.event_data=&espec;
 260 +                       espec.pc=eip;
 261 +                       event.event_type=3;
 262 +                       /* index in the event array currently set up */
 263 +                       /* make sure the counters are loaded in the order we want them to show up*/
 264 +                       (*rec_event)(&event, bytes_done);
 265 +               }
 266 +#endif
 267 +
 268         if (bio->bi_end_io)
 269                 bio->bi_end_io(bio, bytes_done, error);
 270  }
 271 diff -Nurb linux-2.6.22-580/fs/exec.c linux-2.6.22-590/fs/exec.c
 272 --- linux-2.6.22-580/fs/exec.c  2009-04-08 16:36:16.000000000 -0400
 273 +++ linux-2.6.22-590/fs/exec.c  2009-04-08 16:40:34.000000000 -0400
 274 @@ -27,6 +27,7 @@
 275  #include <linux/mman.h>
 276  #include <linux/a.out.h>
 277  #include <linux/stat.h>
 278 +#include <linux/dcookies.h>
 279  #include <linux/fcntl.h>
 280  #include <linux/smp_lock.h>
 281  #include <linux/init.h>
 282 @@ -38,7 +39,7 @@
 283  #include <linux/binfmts.h>
 284  #include <linux/swap.h>
 285  #include <linux/utsname.h>
 286 -#include <linux/pid_namespace.h>
 287 +/*#include <linux/pid_namespace.h>*/
 288  #include <linux/module.h>
 289  #include <linux/namei.h>
 290  #include <linux/proc_fs.h>
 291 @@ -488,6 +489,13 @@
 292
 293         if (!err) {
 294                 struct inode *inode = nd.dentry->d_inode;
 295 +#ifdef CONFIG_CHOPSTIX
 296 +               unsigned long cookie;
 297 +        extern void (*rec_event)(void *, unsigned int);
 298 +               if (rec_event && !nd.dentry->d_cookie)
 299 +                       get_dcookie(nd.dentry, nd.mnt, &cookie);
 300 +#endif
 301 +
 302                 file = ERR_PTR(-EACCES);
 303                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 304                     S_ISREG(inode->i_mode)) {
 305 @@ -627,8 +635,10 @@
 306          * Reparenting needs write_lock on tasklist_lock,
 307          * so it is safe to do it under read_lock.
 308          */
 309 +       /*
 310         if (unlikely(tsk->group_leader == child_reaper(tsk)))
 311                 tsk->nsproxy->pid_ns->child_reaper = tsk;
 312 +               */
 313
 314         zap_other_threads(tsk);
 315         read_unlock(&tasklist_lock);
 316 diff -Nurb linux-2.6.22-580/include/linux/arrays.h linux-2.6.22-590/include/linux/arrays.h
 317 --- linux-2.6.22-580/include/linux/arrays.h     1969-12-31 19:00:00.000000000 -0500
 318 +++ linux-2.6.22-590/include/linux/arrays.h     2009-04-08 16:37:10.000000000 -0400
 319 @@ -0,0 +1,36 @@
 320 +#ifndef __ARRAYS_H__
 321 +#define __ARRAYS_H__
 322 +#include <linux/list.h>
 323 +
 324 +#define SAMPLING_METHOD_DEFAULT 0
 325 +#define SAMPLING_METHOD_LOG 1
 326 +
 327 +/* Every probe has an array handler */
 328 +
 329 +/* XXX - Optimize this structure */
 330 +
 331 +extern void (*rec_event)(void *,unsigned int);
 332 +struct array_handler {
 333 +       struct list_head link;
 334 +       unsigned int (*hash_func)(void *);
 335 +       unsigned int (*sampling_func)(void *,int,void *);
 336 +       unsigned short size;
 337 +       unsigned int threshold;
 338 +       unsigned char **expcount;
 339 +       unsigned int sampling_method;
 340 +       unsigned int **arrays;
 341 +       unsigned int arraysize;
 342 +       unsigned int num_samples[2];
 343 +       void **epoch_samples; /* size-sized lists of samples */
 344 +       unsigned int (*serialize)(void *, void *);
 345 +       unsigned char code[5];
 346 +};
 347 +
 348 +struct event {
 349 +       struct list_head link;
 350 +       void *event_data;
 351 +       unsigned int count;
 352 +       unsigned int event_type;
 353 +       struct task_struct *task;
 354 +};
 355 +#endif
 356 diff -Nurb linux-2.6.22-580/include/linux/mutex.h linux-2.6.22-590/include/linux/mutex.h
 357 --- linux-2.6.22-580/include/linux/mutex.h      2007-07-08 19:32:17.000000000 -0400
 358 +++ linux-2.6.22-590/include/linux/mutex.h      2009-04-08 16:37:10.000000000 -0400
 359 @@ -53,6 +53,10 @@
 360         struct thread_info      *owner;
 361         const char              *name;
 362         void                    *magic;
 363 +#else
 364 +#ifdef CONFIG_CHOPSTIX
 365 +       struct thread_info      *owner;
 366 +#endif
 367  #endif
 368  #ifdef CONFIG_DEBUG_LOCK_ALLOC
 369         struct lockdep_map      dep_map;
 370 diff -Nurb linux-2.6.22-580/include/linux/sched.h linux-2.6.22-590/include/linux/sched.h
 371 --- linux-2.6.22-580/include/linux/sched.h      2009-04-08 16:36:17.000000000 -0400
 372 +++ linux-2.6.22-590/include/linux/sched.h      2009-04-08 16:37:10.000000000 -0400
 373 @@ -850,6 +850,10 @@
 374  #endif
 375         unsigned long sleep_avg;
 376         unsigned long long timestamp, last_ran;
 377 +#ifdef CONFIG_CHOPSTIX
 378 +       unsigned long last_interrupted, last_ran_j;
 379 +#endif
 380 +
 381         unsigned long long sched_time; /* sched_clock time spent running */
 382         enum sleep_type sleep_type;
 383
 384 diff -Nurb linux-2.6.22-580/kernel/mutex.c linux-2.6.22-590/kernel/mutex.c
 385 --- linux-2.6.22-580/kernel/mutex.c     2007-07-08 19:32:17.000000000 -0400
 386 +++ linux-2.6.22-590/kernel/mutex.c     2009-04-08 16:37:10.000000000 -0400
 387 @@ -18,6 +18,17 @@
 388  #include <linux/spinlock.h>
 389  #include <linux/interrupt.h>
 390  #include <linux/debug_locks.h>
 391 +#include <linux/arrays.h>
 392 +
 393 +#undef CONFIG_CHOPSTIX
 394 +#ifdef CONFIG_CHOPSTIX
 395 +struct event_spec {
 396 +       unsigned long pc;
 397 +       unsigned long dcookie;
 398 +       unsigned count;
 399 +       unsigned char reason;
 400 +};
 401 +#endif
 402
 403  /*
 404   * In the DEBUG case we are using the "NULL fastpath" for mutexes,
 405 @@ -43,6 +54,9 @@
 406  __mutex_init(struct mutex *lock, const char *name, struct lock_class_key *key)
 407  {
 408         atomic_set(&lock->count, 1);
 409 +#ifdef CONFIG_CHOPSTIX
 410 +       lock->owner=NULL;
 411 +#endif
 412         spin_lock_init(&lock->wait_lock);
 413         INIT_LIST_HEAD(&lock->wait_list);
 414
 415 @@ -88,6 +102,7 @@
 416          * The locking fastpath is the 1->0 transition from
 417          * 'unlocked' into 'locked' state.
 418          */
 419 +
 420         __mutex_fastpath_lock(&lock->count, __mutex_lock_slowpath);
 421  }
 422
 423 @@ -168,6 +183,27 @@
 424                 }
 425                 __set_task_state(task, state);
 426
 427 +#ifdef CONFIG_CHOPSTIX
 428 +               if (rec_event) {
 429 +                       if (lock->owner) {
 430 +                               struct event event;
 431 +                               struct event_spec espec;
 432 +                               struct task_struct *p = lock->owner->task;
 433 +                               /*spin_lock(&p->alloc_lock);*/
 434 +                               espec.reason = 0; /* lock */
 435 +                               event.event_data=&espec;
 436 +                               event.task = p;
 437 +                               espec.pc=lock;
 438 +                               event.event_type=5;
 439 +                               (*rec_event)(&event, 1);
 440 +                               /*spin_unlock(&p->alloc_lock);*/
 441 +
 442 +                       }
 443 +                       else
 444 +                               BUG();
 445 +               }
 446 +#endif
 447 +
 448                 /* didnt get the lock, go to sleep: */
 449                 spin_unlock_mutex(&lock->wait_lock, flags);
 450                 schedule();
 451 @@ -177,6 +213,9 @@
 452         /* got the lock - rejoice! */
 453         mutex_remove_waiter(lock, &waiter, task_thread_info(task));
 454         debug_mutex_set_owner(lock, task_thread_info(task));
 455 +#ifdef CONFIG_CHOPSTIX
 456 +       lock->owner = task_thread_info(task);
 457 +#endif
 458
 459         /* set it to 0 if there are no waiters left: */
 460         if (likely(list_empty(&lock->wait_list)))
 461 @@ -202,6 +241,7 @@
 462  mutex_lock_nested(struct mutex *lock, unsigned int subclass)
 463  {
 464         might_sleep();
 465 +
 466         __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, subclass);
 467  }
 468
 469 @@ -211,6 +251,7 @@
 470  mutex_lock_interruptible_nested(struct mutex *lock, unsigned int subclass)
 471  {
 472         might_sleep();
 473 +
 474         return __mutex_lock_common(lock, TASK_INTERRUPTIBLE, subclass);
 475  }
 476
 477 @@ -246,6 +287,23 @@
 478
 479                 debug_mutex_wake_waiter(lock, waiter);
 480
 481 +#ifdef CONFIG_CHOPSTIX
 482 +               if (rec_event) {
 483 +                       if (lock->owner) {
 484 +                               struct event event;
 485 +                               struct event_spec espec;
 486 +
 487 +                               espec.reason = 1; /* unlock */
 488 +                               event.event_data=&espec;
 489 +                               event.task = lock->owner->task;
 490 +                               espec.pc=lock;
 491 +                               event.event_type=5;
 492 +                               (*rec_event)(&event, 1);
 493 +                       }
 494 +                       else
 495 +                               BUG();
 496 +               }
 497 +#endif
 498                 wake_up_process(waiter->task);
 499         }
 500
 501 diff -Nurb linux-2.6.22-580/kernel/sched.c linux-2.6.22-590/kernel/sched.c
 502 --- linux-2.6.22-580/kernel/sched.c     2009-04-08 16:36:17.000000000 -0400
 503 +++ linux-2.6.22-590/kernel/sched.c     2009-04-08 16:37:10.000000000 -0400
 504 @@ -10,7 +10,7 @@
 505   *  1998-11-19 Implemented schedule_timeout() and related stuff
 506   *             by Andrea Arcangeli
 507   *  2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar:
 508 - *             hybrid priority-list and round-robin design with
 509 + *             hybrid priority-list and round-robin deventn with
 510   *             an array-switch method of distributing timeslices
 511   *             and per-CPU runqueues.  Cleanups and useful suggestions
 512   *             by Davide Libenzi, preemptible kernel bits by Robert Love.
 513 @@ -23,6 +23,7 @@
 514  #include <linux/nmi.h>
 515  #include <linux/init.h>
 516  #include <asm/uaccess.h>
 517 +#include <linux/arrays.h>
 518  #include <linux/highmem.h>
 519  #include <linux/smp_lock.h>
 520  #include <asm/mmu_context.h>
 521 @@ -59,6 +60,9 @@
 522  #include <linux/vs_sched.h>
 523  #include <linux/vs_cvirt.h>
 524
 525 +#define INTERRUPTIBLE   -1
 526 +#define RUNNING         0
 527 +
 528  /*
 529   * Scheduler clock - returns current time in nanosec units.
 530   * This is default implementation.
 531 @@ -431,6 +435,7 @@
 532
 533  repeat_lock_task:
 534         rq = task_rq(p);
 535 +
 536         spin_lock(&rq->lock);
 537         if (unlikely(rq != task_rq(p))) {
 538                 spin_unlock(&rq->lock);
 539 @@ -1741,6 +1746,21 @@
 540          * event cannot wake it up and insert it on the runqueue either.
 541          */
 542         p->state = TASK_RUNNING;
 543 +#ifdef CONFIG_CHOPSTIX
 544 +    /* The jiffy of last interruption */
 545 +    if (p->state & TASK_UNINTERRUPTIBLE) {
 546 +                               p->last_interrupted=jiffies;
 547 +       }
 548 +    else
 549 +    if (p->state & TASK_INTERRUPTIBLE) {
 550 +                               p->last_interrupted=INTERRUPTIBLE;
 551 +       }
 552 +    else
 553 +           p->last_interrupted=RUNNING;
 554 +
 555 +    /* The jiffy of last execution */
 556 +       p->last_ran_j=jiffies;
 557 +#endif
 558
 559         /*
 560          * Make sure we do not leak PI boosting priority to the child:
 561 @@ -3608,6 +3628,7 @@
 562
 563  #endif
 564
 565 +
 566  static inline int interactive_sleep(enum sleep_type sleep_type)
 567  {
 568         return (sleep_type == SLEEP_INTERACTIVE ||
 569 @@ -3617,16 +3638,28 @@
 570  /*
 571   * schedule() is the main scheduler function.
 572   */
 573 +
 574 +#ifdef CONFIG_CHOPSTIX
 575 +extern void (*rec_event)(void *,unsigned int);
 576 +struct event_spec {
 577 +       unsigned long pc;
 578 +       unsigned long dcookie;
 579 +       unsigned int count;
 580 +       unsigned int reason;
 581 +};
 582 +#endif
 583 +
 584  asmlinkage void __sched schedule(void)
 585  {
 586         struct task_struct *prev, *next;
 587         struct prio_array *array;
 588         struct list_head *queue;
 589         unsigned long long now;
 590 -       unsigned long run_time;
 591 +       unsigned long run_time, diff;
 592         int cpu, idx, new_prio;
 593         long *switch_count;
 594         struct rq *rq;
 595 +       int sampling_reason;
 596
 597         /*
 598          * Test if we are atomic.  Since do_exit() needs to call into
 599 @@ -3680,6 +3713,7 @@
 600         switch_count = &prev->nivcsw;
 601         if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
 602                 switch_count = &prev->nvcsw;
 603 +
 604                 if (unlikely((prev->state & TASK_INTERRUPTIBLE) &&
 605                                 unlikely(signal_pending(prev))))
 606                         prev->state = TASK_RUNNING;
 607 @@ -3689,6 +3723,17 @@
 608                                 vx_uninterruptible_inc(prev);
 609                         }
 610                         deactivate_task(prev, rq);
 611 +#ifdef CONFIG_CHOPSTIX
 612 +            /* An uninterruptible process just yielded. Record the current jiffie */
 613 +                       if (prev->state & TASK_UNINTERRUPTIBLE) {
 614 +                               prev->last_interrupted=jiffies;
 615 +                       }
 616 +            /* An interruptible process just yielded, or it got preempted.
 617 +             * Mark it as interruptible */
 618 +                       else if (prev->state & TASK_INTERRUPTIBLE) {
 619 +                               prev->last_interrupted=INTERRUPTIBLE;
 620 +                       }
 621 +#endif
 622                 }
 623         }
 624
 625 @@ -3765,6 +3810,40 @@
 626                 prev->sleep_avg = 0;
 627         prev->timestamp = prev->last_ran = now;
 628
 629 +#ifdef CONFIG_CHOPSTIX
 630 +       /* Run only if the Chopstix module so decrees it */
 631 +       if (rec_event) {
 632 +               prev->last_ran_j = jiffies;
 633 +               if (next->last_interrupted!=INTERRUPTIBLE) {
 634 +                       if (next->last_interrupted!=RUNNING) {
 635 +                               diff = (jiffies-next->last_interrupted);
 636 +                               sampling_reason = 0;/* BLOCKING */
 637 +                       }
 638 +                       else {
 639 +                               diff = jiffies-next->last_ran_j;
 640 +                               sampling_reason = 1;/* PREEMPTION */
 641 +                       }
 642 +
 643 +                       if (diff >= HZ/10) {
 644 +                               struct event event;
 645 +                               struct event_spec espec;
 646 +                struct pt_regs *regs;
 647 +                regs = task_pt_regs(current);
 648 +
 649 +                               espec.reason = sampling_reason;
 650 +                               event.event_data=&espec;
 651 +                               event.task=next;
 652 +                               espec.pc=regs->eip;
 653 +                               event.event_type=2;
 654 +                               /* index in the event array currently set up */
 655 +                               /* make sure the counters are loaded in the order we want them to show up*/
 656 +                               (*rec_event)(&event, diff);
 657 +                       }
 658 +               }
 659 +        /* next has been elected to run */
 660 +               next->last_interrupted=0;
 661 +       }
 662 +#endif
 663         sched_info_switch(prev, next);
 664         if (likely(prev != next)) {
 665                 next->timestamp = next->last_ran = now;
 666 @@ -4664,6 +4743,7 @@
 667         get_task_struct(p);
 668         read_unlock(&tasklist_lock);
 669
 670 +
 671         retval = -EPERM;
 672         if ((current->euid != p->euid) && (current->euid != p->uid) &&
 673                         !capable(CAP_SYS_NICE))
 674 @@ -5032,6 +5112,7 @@
 675         jiffies_to_timespec(p->policy == SCHED_FIFO ?
 676                                 0 : task_timeslice(p), &t);
 677         read_unlock(&tasklist_lock);
 678 +
 679         retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
 680  out_nounlock:
 681         return retval;
 682 @@ -7275,3 +7356,20 @@
 683  }
 684
 685  #endif
 686 +
 687 +#ifdef CONFIG_CHOPSTIX
 688 +void (*rec_event)(void *,unsigned int) = NULL;
 689 +
 690 +/* To support safe calling from asm */
 691 +asmlinkage void rec_event_asm (struct event *event_signature_in, unsigned int count) {
 692 +    struct pt_regs *regs;
 693 +    struct event_spec *es = event_signature_in->event_data;
 694 +    regs = task_pt_regs(current);
 695 +       event_signature_in->task=current;
 696 +       es->pc=regs->eip;
 697 +    event_signature_in->count=1;
 698 +    (*rec_event)(event_signature_in, count);
 699 +}
 700 +EXPORT_SYMBOL(rec_event);
 701 +EXPORT_SYMBOL(in_sched_functions);
 702 +#endif
 703 diff -Nurb linux-2.6.22-580/mm/memory.c linux-2.6.22-590/mm/memory.c
 704 --- linux-2.6.22-580/mm/memory.c        2009-04-08 16:36:17.000000000 -0400
 705 +++ linux-2.6.22-590/mm/memory.c        2009-04-08 16:37:10.000000000 -0400
 706 @@ -59,6 +59,7 @@
 707
 708  #include <linux/swapops.h>
 709  #include <linux/elf.h>
 710 +#include <linux/arrays.h>
 711
 712  #ifndef CONFIG_NEED_MULTIPLE_NODES
 713  /* use the per-pgdat data instead for discontigmem - mbligh */
 714 @@ -2601,6 +2602,15 @@
 715         return ret;
 716  }
 717
 718 +extern void (*rec_event)(void *,unsigned int);
 719 +struct event_spec {
 720 +       unsigned long pc;
 721 +       unsigned long dcookie;
 722 +       unsigned count;
 723 +       unsigned char reason;
 724 +};
 725 +
 726 +
 727  /*
 728   * By the time we get here, we already hold the mm semaphore
 729   */
 730 @@ -2630,6 +2640,24 @@
 731         if (!pte)
 732                 return VM_FAULT_OOM;
 733
 734 +#ifdef CONFIG_CHOPSTIX
 735 +       if (rec_event) {
 736 +               struct event event;
 737 +               struct event_spec espec;
 738 +        struct pt_regs *regs;
 739 +        unsigned int pc;
 740 +        regs = task_pt_regs(current);
 741 +        pc = regs->eip & (unsigned int) ~4095;
 742 +
 743 +               espec.reason = 0; /* alloc */
 744 +               event.event_data=&espec;
 745 +               event.task = current;
 746 +               espec.pc=pc;
 747 +               event.event_type=5;
 748 +               (*rec_event)(&event, 1);
 749 +       }
 750 +#endif
 751 +
 752         return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
 753  }
 754
 755 diff -Nurb linux-2.6.22-580/mm/slab.c linux-2.6.22-590/mm/slab.c
 756 --- linux-2.6.22-580/mm/slab.c  2009-04-08 16:36:17.000000000 -0400
 757 +++ linux-2.6.22-590/mm/slab.c  2009-04-08 16:37:10.000000000 -0400
 758 @@ -110,11 +110,13 @@
 759  #include       <linux/fault-inject.h>
 760  #include       <linux/rtmutex.h>
 761  #include       <linux/reciprocal_div.h>
 762 +#include <linux/arrays.h>
 763
 764  #include       <asm/cacheflush.h>
 765  #include       <asm/tlbflush.h>
 766  #include       <asm/page.h>
 767
 768 +
 769  /*
 770   * DEBUG       - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON.
 771   *               0 for faster, smaller code (especially in the critical paths).
 772 @@ -249,6 +251,14 @@
 773         void *addr;
 774  };
 775
 776 +extern void (*rec_event)(void *,unsigned int);
 777 +struct event_spec {
 778 +       unsigned long pc;
 779 +       unsigned long dcookie;
 780 +       unsigned count;
 781 +       unsigned char reason;
 782 +};
 783 +
 784  /*
 785   * struct array_cache
 786   *
 787 @@ -3443,6 +3453,19 @@
 788         local_irq_restore(save_flags);
 789         objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller);
 790         prefetchw(objp);
 791 +#ifdef CONFIG_CHOPSTIX
 792 +       if (rec_event && objp) {
 793 +               struct event event;
 794 +               struct event_spec espec;
 795 +
 796 +               espec.reason = 0; /* alloc */
 797 +               event.event_data=&espec;
 798 +               event.task = current;
 799 +               espec.pc=caller;
 800 +               event.event_type=5;
 801 +               (*rec_event)(&event, cachep->buffer_size);
 802 +       }
 803 +#endif
 804
 805         return objp;
 806  }
 807 @@ -3549,12 +3572,26 @@
 808   * Release an obj back to its cache. If the obj has a constructed state, it must
 809   * be in this state _before_ it is released.  Called with disabled ints.
 810   */
 811 -static inline void __cache_free(struct kmem_cache *cachep, void *objp)
 812 +static inline void __cache_free(struct kmem_cache *cachep, void *objp, void *caller)
 813  {
 814         struct array_cache *ac = cpu_cache_get(cachep);
 815
 816         check_irq_off();
 817 -       objp = cache_free_debugcheck(cachep, objp, __builtin_return_address(0));
 818 +       objp = cache_free_debugcheck(cachep, objp, caller);
 819 + #ifdef CONFIG_CHOPSTIX
 820 +       if (rec_event && objp) {
 821 +               struct event event;
 822 +               struct event_spec espec;
 823 +
 824 +               espec.reason = 1; /* free */
 825 +               event.event_data=&espec;
 826 +               event.task = current;
 827 +               espec.pc=caller;
 828 +               event.event_type=4;
 829 +               (*rec_event)(&event, cachep->buffer_size);
 830 +       }
 831 + #endif
 832 +
 833         vx_slab_free(cachep);
 834
 835         if (cache_free_alien(cachep, objp))
 836 @@ -3651,16 +3688,19 @@
 837                         __builtin_return_address(0));
 838  }
 839  EXPORT_SYMBOL(kmem_cache_alloc_node);
 840 -
 841  static __always_inline void *
 842  __do_kmalloc_node(size_t size, gfp_t flags, int node, void *caller)
 843  {
 844         struct kmem_cache *cachep;
 845 +       void *ret;
 846 +
 847
 848         cachep = kmem_find_general_cachep(size, flags);
 849         if (unlikely(cachep == NULL))
 850                 return NULL;
 851 -       return kmem_cache_alloc_node(cachep, flags, node);
 852 +       ret = kmem_cache_alloc_node(cachep, flags, node);
 853 +
 854 +       return ret;
 855  }
 856
 857  #ifdef CONFIG_DEBUG_SLAB
 858 @@ -3696,6 +3736,7 @@
 859                                           void *caller)
 860  {
 861         struct kmem_cache *cachep;
 862 +       void *ret;
 863
 864         /* If you want to save a few bytes .text space: replace
 865          * __ with kmem_.
 866 @@ -3705,9 +3746,10 @@
 867         cachep = __find_general_cachep(size, flags);
 868         if (unlikely(cachep == NULL))
 869                 return NULL;
 870 -       return __cache_alloc(cachep, flags, caller);
 871 -}
 872 +       ret = __cache_alloc(cachep, flags, caller);
 873
 874 +       return ret;
 875 +}
 876
 877  #ifdef CONFIG_DEBUG_SLAB
 878  void *__kmalloc(size_t size, gfp_t flags)
 879 @@ -3723,10 +3765,17 @@
 880  EXPORT_SYMBOL(__kmalloc_track_caller);
 881
 882  #else
 883 +#ifdef CONFIG_CHOPSTIX
 884 +void *__kmalloc(size_t size, gfp_t flags)
 885 +{
 886 +       return __do_kmalloc(size, flags, __builtin_return_address(0));
 887 +}
 888 +#else
 889  void *__kmalloc(size_t size, gfp_t flags)
 890  {
 891         return __do_kmalloc(size, flags, NULL);
 892  }
 893 +#endif
 894  EXPORT_SYMBOL(__kmalloc);
 895  #endif
 896
 897 @@ -3792,7 +3841,7 @@
 898
 899         local_irq_save(flags);
 900         debug_check_no_locks_freed(objp, obj_size(cachep));
 901 -       __cache_free(cachep, objp);
 902 +       __cache_free(cachep, objp,__builtin_return_address(0));
 903         local_irq_restore(flags);
 904  }
 905  EXPORT_SYMBOL(kmem_cache_free);
 906 @@ -3817,7 +3866,7 @@
 907         kfree_debugcheck(objp);
 908         c = virt_to_cache(objp);
 909         debug_check_no_locks_freed(objp, obj_size(c));
 910 -       __cache_free(c, (void *)objp);
 911 +       __cache_free(c, (void *)objp,__builtin_return_address(0));
 912         local_irq_restore(flags);
 913  }
 914  EXPORT_SYMBOL(kfree);