linux-2.6-591-chopstix-intern.patch

   1 diff -Nurb --exclude='*.cmd' --exclude='*.orig' --exclude='*.swp' --exclude=tags --exclude='*.patch' --exclude='*.diff' --exclude='*.svn*' linux-2.6.22-590/arch/i386/Kconfig linux-2.6.22-591/arch/i386/Kconfig
   2 --- linux-2.6.22-590/arch/i386/Kconfig  2009-03-16 20:49:42.000000000 -0400
   3 +++ linux-2.6.22-591/arch/i386/Kconfig  2009-03-16 20:58:59.000000000 -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 --- linux-2.6.22-590/arch/i386/kernel/asm-offsets.c     2007-07-08 19:32:17.000000000 -0400
  20 +++ linux-2.6.22-591/arch/i386/kernel/asm-offsets.c     2009-03-16 20:58:59.000000000 -0400
  21 @@ -9,6 +9,7 @@
  22  #include <linux/signal.h>
  23  #include <linux/personality.h>
  24  #include <linux/suspend.h>
  25 +#include <linux/arrays.h>
  26  #include <asm/ucontext.h>
  27  #include "sigframe.h"
  28  #include <asm/pgtable.h>
  29 @@ -25,9 +26,19 @@
  30  #define OFFSET(sym, str, mem) \
  31         DEFINE(sym, offsetof(struct str, mem));
  32
  33 +#define STACKOFFSET(sym, str, mem) \
  34 +       DEFINE(sym, offsetof(struct str, mem)-sizeof(struct str));
  35 +
  36  /* workaround for a warning with -Wmissing-prototypes */
  37  void foo(void);
  38
  39 +struct event_spec {
  40 +       unsigned long pc;
  41 +       unsigned long dcookie;
  42 +       unsigned count;
  43 +       unsigned int number;
  44 +};
  45 +
  46  void foo(void)
  47  {
  48         OFFSET(SIGCONTEXT_eax, sigcontext, eax);
  49 @@ -51,7 +62,16 @@
  50         OFFSET(CPUINFO_x86_vendor_id, cpuinfo_x86, x86_vendor_id);
  51         BLANK();
  52
  53 -       OFFSET(TI_task, thread_info, task);
  54 +    STACKOFFSET(TASK_thread, task_struct, thread);
  55 +    STACKOFFSET(THREAD_esp, thread_struct, esp);
  56 +    STACKOFFSET(EVENT_event_data, event, event_data);
  57 +    STACKOFFSET(EVENT_task, event, task);
  58 +    STACKOFFSET(EVENT_event_type, event, event_type);
  59 +    STACKOFFSET(SPEC_number, event_spec, number);
  60 +    DEFINE(EVENT_SIZE, sizeof(struct event));
  61 +    DEFINE(SPEC_SIZE, sizeof(struct event_spec));
  62 +    DEFINE(SPEC_EVENT_SIZE, sizeof(struct event_spec)+sizeof(struct event));
  63 +
  64         OFFSET(TI_exec_domain, thread_info, exec_domain);
  65         OFFSET(TI_flags, thread_info, flags);
  66         OFFSET(TI_status, thread_info, status);
  67 --- linux-2.6.22-590/arch/i386/kernel/entry.S   2009-03-16 20:49:07.000000000 -0400
  68 +++ linux-2.6.22-591/arch/i386/kernel/entry.S   2009-03-16 20:58:59.000000000 -0400
  69 @@ -374,6 +374,33 @@
  70         cmpl $(nr_syscalls), %eax
  71         jae syscall_badsys
  72  syscall_call:
  73 +    /* Move Chopstix syscall probe here */
  74 +    /* Save and clobber: eax, ecx, ebp  */
  75 +    pushl   %eax
  76 +    pushl   %ecx
  77 +    pushl   %ebp
  78 +    movl    %esp, %ebp
  79 +    subl    $SPEC_EVENT_SIZE, %esp
  80 +    movl    rec_event, %ecx
  81 +    testl   %ecx, %ecx
  82 +    jz  carry_on
  83 +    # struct event is first, just below %ebp
  84 +    movl    %eax, (SPEC_number-EVENT_SIZE)(%ebp)
  85 +    leal    -SPEC_EVENT_SIZE(%ebp), %eax
  86 +    movl    %eax, EVENT_event_data(%ebp)
  87 +    movl    $6, EVENT_event_type(%ebp)
  88 +    movl    rec_event, %edx
  89 +    movl    $1, 4(%esp)
  90 +    leal    -EVENT_SIZE(%ebp), %eax
  91 +    movl    %eax, (%esp)
  92 +    call    rec_event_asm
  93 +carry_on:
  94 +    addl $SPEC_EVENT_SIZE, %esp
  95 +    popl %ebp
  96 +    popl %ecx
  97 +    popl %eax
  98 +     /* End chopstix */
  99 +
 100         call *sys_call_table(,%eax,4)
 101         movl %eax,PT_EAX(%esp)          # store the return value
 102  syscall_exit:
 103 --- linux-2.6.22-590/arch/i386/mm/fault.c       2009-03-16 20:49:42.000000000 -0400
 104 +++ linux-2.6.22-591/arch/i386/mm/fault.c       2009-03-16 20:58:59.000000000 -0400
 105 @@ -60,6 +60,15 @@
 106                                           DIE_PAGE_FAULT, &args);
 107  }
 108
 109 +
 110 +extern void (*rec_event)(void *,unsigned int);
 111 +struct event_spec {
 112 +       unsigned long pc;
 113 +       unsigned long dcookie;
 114 +       unsigned count;
 115 +       unsigned char reason;
 116 +};
 117 +
 118  /*
 119   * Return EIP plus the CS segment base.  The segment limit is also
 120   * adjusted, clamped to the kernel/user address space (whichever is
 121 @@ -296,6 +305,8 @@
 122   *     bit 3 == 1 means use of reserved bit detected
 123   *     bit 4 == 1 means fault was an instruction fetch
 124   */
 125 +
 126 +
 127  fastcall void __kprobes do_page_fault(struct pt_regs *regs,
 128                                       unsigned long error_code)
 129  {
 130 --- linux-2.6.22-590/block/ll_rw_blk.c  2009-03-16 20:49:07.000000000 -0400
 131 +++ linux-2.6.22-591/block/ll_rw_blk.c  2009-03-16 20:58:59.000000000 -0400
 132 @@ -30,6 +30,7 @@
 133  #include <linux/cpu.h>
 134  #include <linux/blktrace_api.h>
 135  #include <linux/fault-inject.h>
 136 +#include <linux/arrays.h>
 137
 138  /*
 139   * for max sense size
 140 @@ -3102,6 +3103,13 @@
 141
 142  #endif /* CONFIG_FAIL_MAKE_REQUEST */
 143
 144 +extern void (*rec_event)(void *,unsigned int);
 145 +struct event_spec {
 146 +       unsigned long pc;
 147 +       unsigned long dcookie;
 148 +       unsigned count;
 149 +       unsigned char reason;
 150 +};
 151  /**
 152   * generic_make_request: hand a buffer to its device driver for I/O
 153   * @bio:  The bio describing the location in memory and on the device.
 154 @@ -3220,7 +3228,23 @@
 155                                 goto end_io;
 156                         }
 157                 }
 158 -
 159 +#ifdef CONFIG_CHOPSTIX
 160 +               if (rec_event) {
 161 +                       struct event event;
 162 +                       struct event_spec espec;
 163 +                       unsigned long eip;
 164 +
 165 +                       espec.reason = 0;/*request */
 166 +
 167 +                       eip = bio->bi_end_io;
 168 +                       event.event_data=&espec;
 169 +                       espec.pc=eip;
 170 +                       event.event_type=3;
 171 +                       /* index in the event array currently set up */
 172 +                       /* make sure the counters are loaded in the order we want them to show up*/
 173 +                       (*rec_event)(&event, bio->bi_size);
 174 +               }
 175 +#endif
 176                 ret = q->make_request_fn(q, bio);
 177         } while (ret);
 178  }
 179 --- linux-2.6.22-590/drivers/oprofile/cpu_buffer.c      2007-07-08 19:32:17.000000000 -0400
 180 +++ linux-2.6.22-591/drivers/oprofile/cpu_buffer.c      2009-03-16 20:58:59.000000000 -0400
 181 @@ -21,6 +21,7 @@
 182  #include <linux/oprofile.h>
 183  #include <linux/vmalloc.h>
 184  #include <linux/errno.h>
 185 +#include <linux/arrays.h>
 186
 187  #include "event_buffer.h"
 188  #include "cpu_buffer.h"
 189 @@ -143,6 +144,17 @@
 190                 b->head_pos = 0;
 191  }
 192
 193 +#ifdef CONFIG_CHOPSTIX
 194 +
 195 +struct event_spec {
 196 +       unsigned int pc;
 197 +       unsigned long dcookie;
 198 +       unsigned count;
 199 +};
 200 +
 201 +extern void (*rec_event)(void *,unsigned int);
 202 +#endif
 203 +
 204  static inline void
 205  add_sample(struct oprofile_cpu_buffer * cpu_buf,
 206             unsigned long pc, unsigned long event)
 207 @@ -151,6 +163,7 @@
 208         entry->eip = pc;
 209         entry->event = event;
 210         increment_head(cpu_buf);
 211 +
 212  }
 213
 214  static inline void
 215 @@ -241,8 +254,28 @@
 216  {
 217         int is_kernel = !user_mode(regs);
 218         unsigned long pc = profile_pc(regs);
 219 +       int res=0;
 220
 221 +#ifdef CONFIG_CHOPSTIX
 222 +       if (rec_event) {
 223 +               struct event esig;
 224 +               struct event_spec espec;
 225 +               esig.task = current;
 226 +               espec.pc=pc;
 227 +               espec.count=1;
 228 +               esig.event_data=&espec;
 229 +               esig.event_type=event; /* index in the event array currently set up */
 230 +                                       /* make sure the counters are loaded in the order we want them to show up*/
 231 +               (*rec_event)(&esig, 1);
 232 +       }
 233 +       else {
 234         oprofile_add_ext_sample(pc, regs, event, is_kernel);
 235 +       }
 236 +#else
 237 +       oprofile_add_ext_sample(pc, regs, event, is_kernel);
 238 +#endif
 239 +
 240 +
 241  }
 242
 243  void oprofile_add_pc(unsigned long pc, int is_kernel, unsigned long event)
 244 --- linux-2.6.22-590/fs/bio.c   2007-07-08 19:32:17.000000000 -0400
 245 +++ linux-2.6.22-591/fs/bio.c   2009-03-16 20:58:59.000000000 -0400
 246 @@ -27,6 +27,7 @@
 247  #include <linux/workqueue.h>
 248  #include <linux/blktrace_api.h>
 249  #include <scsi/sg.h>           /* for struct sg_iovec */
 250 +#include <linux/arrays.h>
 251
 252  #define BIO_POOL_SIZE 2
 253
 254 @@ -47,6 +48,7 @@
 255         struct kmem_cache *slab;
 256  };
 257
 258 +
 259  /*
 260   * if you change this list, also change bvec_alloc or things will
 261   * break badly! cannot be bigger than what you can fit into an
 262 @@ -999,6 +1001,14 @@
 263         }
 264  }
 265
 266 +struct event_spec {
 267 +       unsigned long pc;
 268 +       unsigned long dcookie;
 269 +       unsigned count;
 270 +       unsigned char reason;
 271 +};
 272 +
 273 +extern void (*rec_event)(void *,unsigned int);
 274  /**
 275   * bio_endio - end I/O on a bio
 276   * @bio:       bio
 277 @@ -1028,6 +1038,24 @@
 278         bio->bi_size -= bytes_done;
 279         bio->bi_sector += (bytes_done >> 9);
 280
 281 +#ifdef CONFIG_CHOPSTIX
 282 +               if (rec_event) {
 283 +                       struct event event;
 284 +                       struct event_spec espec;
 285 +                       unsigned long eip;
 286 +
 287 +                       espec.reason = 1;/*response */
 288 +
 289 +                       eip = bio->bi_end_io;
 290 +                       event.event_data=&espec;
 291 +                       espec.pc=eip;
 292 +                       event.event_type=3;
 293 +                       /* index in the event array currently set up */
 294 +                       /* make sure the counters are loaded in the order we want them to show up*/
 295 +                       (*rec_event)(&event, bytes_done);
 296 +               }
 297 +#endif
 298 +
 299         if (bio->bi_end_io)
 300                 bio->bi_end_io(bio, bytes_done, error);
 301  }
 302 --- linux-2.6.22-590/fs/exec.c  2009-03-16 20:49:42.000000000 -0400
 303 +++ linux-2.6.22-591/fs/exec.c  2009-03-16 20:58:59.000000000 -0400
 304 @@ -27,6 +27,7 @@
 305  #include <linux/mman.h>
 306  #include <linux/a.out.h>
 307  #include <linux/stat.h>
 308 +#include <linux/dcookies.h>
 309  #include <linux/fcntl.h>
 310  #include <linux/smp_lock.h>
 311  #include <linux/init.h>
 312 @@ -38,7 +39,7 @@
 313  #include <linux/binfmts.h>
 314  #include <linux/swap.h>
 315  #include <linux/utsname.h>
 316 -#include <linux/pid_namespace.h>
 317 +/*#include <linux/pid_namespace.h>*/
 318  #include <linux/module.h>
 319  #include <linux/namei.h>
 320  #include <linux/proc_fs.h>
 321 @@ -488,6 +489,12 @@
 322
 323         if (!err) {
 324                 struct inode *inode = nd.dentry->d_inode;
 325 +#ifdef CONFIG_CHOPSTIX
 326 +               unsigned long cookie;
 327 +               if (!nd.dentry->d_cookie)
 328 +                       get_dcookie(nd.dentry, nd.mnt, &cookie);
 329 +#endif
 330 +
 331                 file = ERR_PTR(-EACCES);
 332                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 333                     S_ISREG(inode->i_mode)) {
 334 @@ -627,8 +634,10 @@
 335          * Reparenting needs write_lock on tasklist_lock,
 336          * so it is safe to do it under read_lock.
 337          */
 338 +       /*
 339         if (unlikely(tsk->group_leader == child_reaper(tsk)))
 340                 tsk->nsproxy->pid_ns->child_reaper = tsk;
 341 +               */
 342
 343         zap_other_threads(tsk);
 344         read_unlock(&tasklist_lock);
 345 --- linux-2.6.22-590/include/linux/arrays.h     1969-12-31 19:00:00.000000000 -0500
 346 +++ linux-2.6.22-591/include/linux/arrays.h     2009-03-16 20:58:59.000000000 -0400
 347 @@ -0,0 +1,36 @@
 348 +#ifndef __ARRAYS_H__
 349 +#define __ARRAYS_H__
 350 +#include <linux/list.h>
 351 +
 352 +#define SAMPLING_METHOD_DEFAULT 0
 353 +#define SAMPLING_METHOD_LOG 1
 354 +
 355 +/* Every probe has an array handler */
 356 +
 357 +/* XXX - Optimize this structure */
 358 +
 359 +extern void (*rec_event)(void *,unsigned int);
 360 +struct array_handler {
 361 +       struct list_head link;
 362 +       unsigned int (*hash_func)(void *);
 363 +       unsigned int (*sampling_func)(void *,int,void *);
 364 +       unsigned short size;
 365 +       unsigned int threshold;
 366 +       unsigned char **expcount;
 367 +       unsigned int sampling_method;
 368 +       unsigned int **arrays;
 369 +       unsigned int arraysize;
 370 +       unsigned int num_samples[2];
 371 +       void **epoch_samples; /* size-sized lists of samples */
 372 +       unsigned int (*serialize)(void *, void *);
 373 +       unsigned char code[5];
 374 +};
 375 +
 376 +struct event {
 377 +       struct list_head link;
 378 +       void *event_data;
 379 +       unsigned int count;
 380 +       unsigned int event_type;
 381 +       struct task_struct *task;
 382 +};
 383 +#endif
 384 --- linux-2.6.22-590/include/linux/mutex.h      2007-07-08 19:32:17.000000000 -0400
 385 +++ linux-2.6.22-591/include/linux/mutex.h      2009-03-16 20:58:59.000000000 -0400
 386 @@ -53,6 +53,10 @@
 387         struct thread_info      *owner;
 388         const char              *name;
 389         void                    *magic;
 390 +#else
 391 +#ifdef CONFIG_CHOPSTIX
 392 +       struct thread_info      *owner;
 393 +#endif
 394  #endif
 395  #ifdef CONFIG_DEBUG_LOCK_ALLOC
 396         struct lockdep_map      dep_map;
 397 --- linux-2.6.22-590/include/linux/sched.h      2009-03-16 20:49:42.000000000 -0400
 398 +++ linux-2.6.22-591/include/linux/sched.h      2009-03-16 20:58:59.000000000 -0400
 399 @@ -850,6 +850,10 @@
 400  #endif
 401         unsigned long sleep_avg;
 402         unsigned long long timestamp, last_ran;
 403 +#ifdef CONFIG_CHOPSTIX
 404 +       unsigned long last_interrupted, last_ran_j;
 405 +#endif
 406 +
 407         unsigned long long sched_time; /* sched_clock time spent running */
 408         enum sleep_type sleep_type;
 409
 410 --- linux-2.6.22-590/kernel/mutex.c     2007-07-08 19:32:17.000000000 -0400
 411 +++ linux-2.6.22-591/kernel/mutex.c     2009-03-16 20:58:59.000000000 -0400
 412 @@ -18,6 +18,17 @@
 413  #include <linux/spinlock.h>
 414  #include <linux/interrupt.h>
 415  #include <linux/debug_locks.h>
 416 +#include <linux/arrays.h>
 417 +
 418 +#undef CONFIG_CHOPSTIX
 419 +#ifdef CONFIG_CHOPSTIX
 420 +struct event_spec {
 421 +       unsigned long pc;
 422 +       unsigned long dcookie;
 423 +       unsigned count;
 424 +       unsigned char reason;
 425 +};
 426 +#endif
 427
 428  /*
 429   * In the DEBUG case we are using the "NULL fastpath" for mutexes,
 430 @@ -43,6 +54,9 @@
 431  __mutex_init(struct mutex *lock, const char *name, struct lock_class_key *key)
 432  {
 433         atomic_set(&lock->count, 1);
 434 +#ifdef CONFIG_CHOPSTIX
 435 +       lock->owner=NULL;
 436 +#endif
 437         spin_lock_init(&lock->wait_lock);
 438         INIT_LIST_HEAD(&lock->wait_list);
 439
 440 @@ -88,6 +102,7 @@
 441          * The locking fastpath is the 1->0 transition from
 442          * 'unlocked' into 'locked' state.
 443          */
 444 +
 445         __mutex_fastpath_lock(&lock->count, __mutex_lock_slowpath);
 446  }
 447
 448 @@ -168,6 +183,27 @@
 449                 }
 450                 __set_task_state(task, state);
 451
 452 +#ifdef CONFIG_CHOPSTIX
 453 +               if (rec_event) {
 454 +                       if (lock->owner) {
 455 +                               struct event event;
 456 +                               struct event_spec espec;
 457 +                               struct task_struct *p = lock->owner->task;
 458 +                               /*spin_lock(&p->alloc_lock);*/
 459 +                               espec.reason = 0; /* lock */
 460 +                               event.event_data=&espec;
 461 +                               event.task = p;
 462 +                               espec.pc=lock;
 463 +                               event.event_type=5;
 464 +                               (*rec_event)(&event, 1);
 465 +                               /*spin_unlock(&p->alloc_lock);*/
 466 +
 467 +                       }
 468 +                       else
 469 +                               BUG();
 470 +               }
 471 +#endif
 472 +
 473                 /* didnt get the lock, go to sleep: */
 474                 spin_unlock_mutex(&lock->wait_lock, flags);
 475                 schedule();
 476 @@ -177,6 +213,9 @@
 477         /* got the lock - rejoice! */
 478         mutex_remove_waiter(lock, &waiter, task_thread_info(task));
 479         debug_mutex_set_owner(lock, task_thread_info(task));
 480 +#ifdef CONFIG_CHOPSTIX
 481 +       lock->owner = task_thread_info(task);
 482 +#endif
 483
 484         /* set it to 0 if there are no waiters left: */
 485         if (likely(list_empty(&lock->wait_list)))
 486 @@ -202,6 +241,7 @@
 487  mutex_lock_nested(struct mutex *lock, unsigned int subclass)
 488  {
 489         might_sleep();
 490 +
 491         __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, subclass);
 492  }
 493
 494 @@ -211,6 +251,7 @@
 495  mutex_lock_interruptible_nested(struct mutex *lock, unsigned int subclass)
 496  {
 497         might_sleep();
 498 +
 499         return __mutex_lock_common(lock, TASK_INTERRUPTIBLE, subclass);
 500  }
 501
 502 @@ -246,6 +287,23 @@
 503
 504                 debug_mutex_wake_waiter(lock, waiter);
 505
 506 +#ifdef CONFIG_CHOPSTIX
 507 +               if (rec_event) {
 508 +                       if (lock->owner) {
 509 +                               struct event event;
 510 +                               struct event_spec espec;
 511 +
 512 +                               espec.reason = 1; /* unlock */
 513 +                               event.event_data=&espec;
 514 +                               event.task = lock->owner->task;
 515 +                               espec.pc=lock;
 516 +                               event.event_type=5;
 517 +                               (*rec_event)(&event, 1);
 518 +                       }
 519 +                       else
 520 +                               BUG();
 521 +               }
 522 +#endif
 523                 wake_up_process(waiter->task);
 524         }
 525
 526 --- linux-2.6.22-590/kernel/sched.c     2009-03-16 20:49:42.000000000 -0400
 527 +++ linux-2.6.22-591/kernel/sched.c     2009-03-16 20:58:59.000000000 -0400
 528 @@ -10,7 +10,7 @@
 529   *  1998-11-19 Implemented schedule_timeout() and related stuff
 530   *             by Andrea Arcangeli
 531   *  2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar:
 532 - *             hybrid priority-list and round-robin design with
 533 + *             hybrid priority-list and round-robin deventn with
 534   *             an array-switch method of distributing timeslices
 535   *             and per-CPU runqueues.  Cleanups and useful suggestions
 536   *             by Davide Libenzi, preemptible kernel bits by Robert Love.
 537 @@ -23,6 +23,7 @@
 538  #include <linux/nmi.h>
 539  #include <linux/init.h>
 540  #include <asm/uaccess.h>
 541 +#include <linux/arrays.h>
 542  #include <linux/highmem.h>
 543  #include <linux/smp_lock.h>
 544  #include <asm/mmu_context.h>
 545 @@ -59,6 +60,9 @@
 546  #include <linux/vs_sched.h>
 547  #include <linux/vs_cvirt.h>
 548
 549 +#define INTERRUPTIBLE   -1
 550 +#define RUNNING         0
 551 +
 552  /*
 553   * Scheduler clock - returns current time in nanosec units.
 554   * This is default implementation.
 555 @@ -431,6 +435,7 @@
 556
 557  repeat_lock_task:
 558         rq = task_rq(p);
 559 +
 560         spin_lock(&rq->lock);
 561         if (unlikely(rq != task_rq(p))) {
 562                 spin_unlock(&rq->lock);
 563 @@ -1741,6 +1746,21 @@
 564          * event cannot wake it up and insert it on the runqueue either.
 565          */
 566         p->state = TASK_RUNNING;
 567 +#ifdef CONFIG_CHOPSTIX
 568 +    /* The jiffy of last interruption */
 569 +    if (p->state & TASK_UNINTERRUPTIBLE) {
 570 +                               p->last_interrupted=jiffies;
 571 +       }
 572 +    else
 573 +    if (p->state & TASK_INTERRUPTIBLE) {
 574 +                               p->last_interrupted=INTERRUPTIBLE;
 575 +       }
 576 +    else
 577 +           p->last_interrupted=RUNNING;
 578 +
 579 +    /* The jiffy of last execution */
 580 +       p->last_ran_j=jiffies;
 581 +#endif
 582
 583         /*
 584          * Make sure we do not leak PI boosting priority to the child:
 585 @@ -3608,6 +3628,7 @@
 586
 587  #endif
 588
 589 +
 590  static inline int interactive_sleep(enum sleep_type sleep_type)
 591  {
 592         return (sleep_type == SLEEP_INTERACTIVE ||
 593 @@ -3617,16 +3638,28 @@
 594  /*
 595   * schedule() is the main scheduler function.
 596   */
 597 +
 598 +#ifdef CONFIG_CHOPSTIX
 599 +extern void (*rec_event)(void *,unsigned int);
 600 +struct event_spec {
 601 +       unsigned long pc;
 602 +       unsigned long dcookie;
 603 +       unsigned int count;
 604 +       unsigned int reason;
 605 +};
 606 +#endif
 607 +
 608  asmlinkage void __sched schedule(void)
 609  {
 610         struct task_struct *prev, *next;
 611         struct prio_array *array;
 612         struct list_head *queue;
 613         unsigned long long now;
 614 -       unsigned long run_time;
 615 +       unsigned long run_time, diff;
 616         int cpu, idx, new_prio;
 617         long *switch_count;
 618         struct rq *rq;
 619 +       int sampling_reason;
 620
 621         /*
 622          * Test if we are atomic.  Since do_exit() needs to call into
 623 @@ -3680,6 +3713,7 @@
 624         switch_count = &prev->nivcsw;
 625         if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
 626                 switch_count = &prev->nvcsw;
 627 +
 628                 if (unlikely((prev->state & TASK_INTERRUPTIBLE) &&
 629                                 unlikely(signal_pending(prev))))
 630                         prev->state = TASK_RUNNING;
 631 @@ -3689,6 +3723,17 @@
 632                                 vx_uninterruptible_inc(prev);
 633                         }
 634                         deactivate_task(prev, rq);
 635 +#ifdef CONFIG_CHOPSTIX
 636 +            /* An uninterruptible process just yielded. Record the current jiffie */
 637 +                       if (prev->state & TASK_UNINTERRUPTIBLE) {
 638 +                               prev->last_interrupted=jiffies;
 639 +                       }
 640 +            /* An interruptible process just yielded, or it got preempted.
 641 +             * Mark it as interruptible */
 642 +                       else if (prev->state & TASK_INTERRUPTIBLE) {
 643 +                               prev->last_interrupted=INTERRUPTIBLE;
 644 +                       }
 645 +#endif
 646                 }
 647         }
 648
 649 @@ -3765,6 +3810,40 @@
 650                 prev->sleep_avg = 0;
 651         prev->timestamp = prev->last_ran = now;
 652
 653 +#ifdef CONFIG_CHOPSTIX
 654 +       /* Run only if the Chopstix module so decrees it */
 655 +       if (rec_event) {
 656 +               prev->last_ran_j = jiffies;
 657 +               if (next->last_interrupted!=INTERRUPTIBLE) {
 658 +                       if (next->last_interrupted!=RUNNING) {
 659 +                               diff = (jiffies-next->last_interrupted);
 660 +                               sampling_reason = 0;/* BLOCKING */
 661 +                       }
 662 +                       else {
 663 +                               diff = jiffies-next->last_ran_j;
 664 +                               sampling_reason = 1;/* PREEMPTION */
 665 +                       }
 666 +
 667 +                       if (diff >= HZ/10) {
 668 +                               struct event event;
 669 +                               struct event_spec espec;
 670 +                struct pt_regs *regs;
 671 +                regs = task_pt_regs(current);
 672 +
 673 +                               espec.reason = sampling_reason;
 674 +                               event.event_data=&espec;
 675 +                               event.task=next;
 676 +                               espec.pc=regs->eip;
 677 +                               event.event_type=2;
 678 +                               /* index in the event array currently set up */
 679 +                               /* make sure the counters are loaded in the order we want them to show up*/
 680 +                               (*rec_event)(&event, diff);
 681 +                       }
 682 +               }
 683 +        /* next has been elected to run */
 684 +               next->last_interrupted=0;
 685 +       }
 686 +#endif
 687         sched_info_switch(prev, next);
 688         if (likely(prev != next)) {
 689                 next->timestamp = next->last_ran = now;
 690 @@ -4664,6 +4743,7 @@
 691         get_task_struct(p);
 692         read_unlock(&tasklist_lock);
 693
 694 +
 695         retval = -EPERM;
 696         if ((current->euid != p->euid) && (current->euid != p->uid) &&
 697                         !capable(CAP_SYS_NICE))
 698 @@ -5032,6 +5112,7 @@
 699         jiffies_to_timespec(p->policy == SCHED_FIFO ?
 700                                 0 : task_timeslice(p), &t);
 701         read_unlock(&tasklist_lock);
 702 +
 703         retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
 704  out_nounlock:
 705         return retval;
 706 @@ -7275,3 +7356,20 @@
 707  }
 708
 709  #endif
 710 +
 711 +#ifdef CONFIG_CHOPSTIX
 712 +void (*rec_event)(void *,unsigned int) = NULL;
 713 +
 714 +/* To support safe calling from asm */
 715 +asmlinkage void rec_event_asm (struct event *event_signature_in, unsigned int count) {
 716 +    struct pt_regs *regs;
 717 +    struct event_spec *es = event_signature_in->event_data;
 718 +    regs = task_pt_regs(current);
 719 +       event_signature_in->task=current;
 720 +       es->pc=regs->eip;
 721 +    event_signature_in->count=1;
 722 +    (*rec_event)(event_signature_in, count);
 723 +}
 724 +EXPORT_SYMBOL(rec_event);
 725 +EXPORT_SYMBOL(in_sched_functions);
 726 +#endif
 727 --- linux-2.6.22-590/mm/memory.c        2009-03-16 20:49:42.000000000 -0400
 728 +++ linux-2.6.22-591/mm/memory.c        2009-03-16 20:58:59.000000000 -0400
 729 @@ -59,6 +59,7 @@
 730
 731  #include <linux/swapops.h>
 732  #include <linux/elf.h>
 733 +#include <linux/arrays.h>
 734
 735  #ifndef CONFIG_NEED_MULTIPLE_NODES
 736  /* use the per-pgdat data instead for discontigmem - mbligh */
 737 @@ -2601,6 +2602,15 @@
 738         return ret;
 739  }
 740
 741 +extern void (*rec_event)(void *,unsigned int);
 742 +struct event_spec {
 743 +       unsigned long pc;
 744 +       unsigned long dcookie;
 745 +       unsigned count;
 746 +       unsigned char reason;
 747 +};
 748 +
 749 +
 750  /*
 751   * By the time we get here, we already hold the mm semaphore
 752   */
 753 @@ -2630,6 +2640,24 @@
 754         if (!pte)
 755                 return VM_FAULT_OOM;
 756
 757 +#ifdef CONFIG_CHOPSTIX
 758 +       if (rec_event) {
 759 +               struct event event;
 760 +               struct event_spec espec;
 761 +        struct pt_regs *regs;
 762 +        unsigned int pc;
 763 +        regs = task_pt_regs(current);
 764 +        pc = regs->eip & (unsigned int) ~4095;
 765 +
 766 +               espec.reason = 0; /* alloc */
 767 +               event.event_data=&espec;
 768 +               event.task = current;
 769 +               espec.pc=pc;
 770 +               event.event_type=5;
 771 +               (*rec_event)(&event, 1);
 772 +       }
 773 +#endif
 774 +
 775         return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
 776  }
 777
 778 --- linux-2.6.22-590/mm/slab.c  2009-03-16 20:49:42.000000000 -0400
 779 +++ linux-2.6.22-591/mm/slab.c  2009-03-16 21:00:27.000000000 -0400
 780 @@ -110,11 +110,13 @@
 781  #include       <linux/fault-inject.h>
 782  #include       <linux/rtmutex.h>
 783  #include       <linux/reciprocal_div.h>
 784 +#include <linux/arrays.h>
 785
 786  #include       <asm/cacheflush.h>
 787  #include       <asm/tlbflush.h>
 788  #include       <asm/page.h>
 789
 790 +
 791  /*
 792   * DEBUG       - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON.
 793   *               0 for faster, smaller code (especially in the critical paths).
 794 @@ -249,6 +251,14 @@
 795         void *addr;
 796  };
 797
 798 +extern void (*rec_event)(void *,unsigned int);
 799 +struct event_spec {
 800 +       unsigned long pc;
 801 +       unsigned long dcookie;
 802 +       unsigned count;
 803 +       unsigned char reason;
 804 +};
 805 +
 806  /*
 807   * struct array_cache
 808   *
 809 @@ -3443,6 +3453,19 @@
 810         local_irq_restore(save_flags);
 811         objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller);
 812         prefetchw(objp);
 813 +#ifdef CONFIG_CHOPSTIX
 814 +       if (rec_event && objp) {
 815 +               struct event event;
 816 +               struct event_spec espec;
 817 +
 818 +               espec.reason = 0; /* alloc */
 819 +               event.event_data=&espec;
 820 +               event.task = current;
 821 +               espec.pc=caller;
 822 +               event.event_type=5;
 823 +               (*rec_event)(&event, cachep->buffer_size);
 824 +       }
 825 +#endif
 826
 827         return objp;
 828  }
 829 @@ -3549,12 +3572,26 @@
 830   * Release an obj back to its cache. If the obj has a constructed state, it must
 831   * be in this state _before_ it is released.  Called with disabled ints.
 832   */
 833 -static inline void __cache_free(struct kmem_cache *cachep, void *objp)
 834 +static inline void __cache_free(struct kmem_cache *cachep, void *objp, void *caller)
 835  {
 836         struct array_cache *ac = cpu_cache_get(cachep);
 837
 838         check_irq_off();
 839 -       objp = cache_free_debugcheck(cachep, objp, __builtin_return_address(0));
 840 +       objp = cache_free_debugcheck(cachep, objp, caller);
 841 + #ifdef CONFIG_CHOPSTIX
 842 +       if (rec_event && objp) {
 843 +               struct event event;
 844 +               struct event_spec espec;
 845 +
 846 +               espec.reason = 1; /* free */
 847 +               event.event_data=&espec;
 848 +               event.task = current;
 849 +               espec.pc=caller;
 850 +               event.event_type=4;
 851 +               (*rec_event)(&event, cachep->buffer_size);
 852 +       }
 853 + #endif
 854 +
 855         vx_slab_free(cachep);
 856
 857         if (cache_free_alien(cachep, objp))
 858 @@ -3651,16 +3688,19 @@
 859                         __builtin_return_address(0));
 860  }
 861  EXPORT_SYMBOL(kmem_cache_alloc_node);
 862 -
 863  static __always_inline void *
 864  __do_kmalloc_node(size_t size, gfp_t flags, int node, void *caller)
 865  {
 866         struct kmem_cache *cachep;
 867 +       void *ret;
 868 +
 869
 870         cachep = kmem_find_general_cachep(size, flags);
 871         if (unlikely(cachep == NULL))
 872                 return NULL;
 873 -       return kmem_cache_alloc_node(cachep, flags, node);
 874 +       ret = kmem_cache_alloc_node(cachep, flags, node);
 875 +
 876 +       return ret;
 877  }
 878
 879  #ifdef CONFIG_DEBUG_SLAB
 880 @@ -3696,6 +3736,7 @@
 881                                           void *caller)
 882  {
 883         struct kmem_cache *cachep;
 884 +       void *ret;
 885
 886         /* If you want to save a few bytes .text space: replace
 887          * __ with kmem_.
 888 @@ -3705,9 +3746,10 @@
 889         cachep = __find_general_cachep(size, flags);
 890         if (unlikely(cachep == NULL))
 891                 return NULL;
 892 -       return __cache_alloc(cachep, flags, caller);
 893 -}
 894 +       ret = __cache_alloc(cachep, flags, caller);
 895
 896 +       return ret;
 897 +}
 898
 899  #ifdef CONFIG_DEBUG_SLAB
 900  void *__kmalloc(size_t size, gfp_t flags)
 901 @@ -3723,10 +3765,17 @@
 902  EXPORT_SYMBOL(__kmalloc_track_caller);
 903
 904  #else
 905 +#ifdef CONFIG_CHOPSTIX
 906 +void *__kmalloc(size_t size, gfp_t flags)
 907 +{
 908 +       return __do_kmalloc(size, flags, __builtin_return_address(0));
 909 +}
 910 +#else
 911  void *__kmalloc(size_t size, gfp_t flags)
 912  {
 913         return __do_kmalloc(size, flags, NULL);
 914  }
 915 +#endif
 916  EXPORT_SYMBOL(__kmalloc);
 917  #endif
 918
 919 @@ -3792,7 +3841,7 @@
 920
 921         local_irq_save(flags);
 922         debug_check_no_locks_freed(objp, obj_size(cachep));
 923 -       __cache_free(cachep, objp);
 924 +       __cache_free(cachep, objp,__builtin_return_address(0));
 925         local_irq_restore(flags);
 926  }
 927  EXPORT_SYMBOL(kmem_cache_free);
 928 @@ -3817,7 +3866,7 @@
 929         kfree_debugcheck(objp);
 930         c = virt_to_cache(objp);
 931         debug_check_no_locks_freed(objp, obj_size(c));
 932 -       __cache_free(c, (void *)objp);
 933 +       __cache_free(c, (void *)objp,__builtin_return_address(0));
 934         local_irq_restore(flags);
 935  }
 936  EXPORT_SYMBOL(kfree);
 937