linux-2.6-590-chopstix-intern.patch

   1 diff -Nurb linux-2.6.22-580/drivers/oprofile/cpu_buffer.c linux-2.6.22-590/drivers/oprofile/cpu_buffer.c
   2 --- linux-2.6.22-580/drivers/oprofile/cpu_buffer.c      2007-07-08 19:32:17.000000000 -0400
   3 +++ linux-2.6.22-590/drivers/oprofile/cpu_buffer.c      2008-02-27 13:53:47.000000000 -0500
   4 @@ -21,6 +21,7 @@
   5  #include <linux/oprofile.h>
   6  #include <linux/vmalloc.h>
   7  #include <linux/errno.h>
   8 +#include <linux/arrays.h>
   9
  10  #include "event_buffer.h"
  11  #include "cpu_buffer.h"
  12 @@ -143,6 +144,14 @@
  13                 b->head_pos = 0;
  14  }
  15
  16 +struct event_spec {
  17 +       unsigned int pc;
  18 +       unsigned long dcookie;
  19 +       unsigned count;
  20 +};
  21 +
  22 +extern void (*rec_event)(void *,unsigned int);
  23 +
  24  static inline void
  25  add_sample(struct oprofile_cpu_buffer * cpu_buf,
  26             unsigned long pc, unsigned long event)
  27 @@ -151,6 +160,7 @@
  28         entry->eip = pc;
  29         entry->event = event;
  30         increment_head(cpu_buf);
  31 +
  32  }
  33
  34  static inline void
  35 @@ -237,12 +247,66 @@
  36         oprofile_end_trace(cpu_buf);
  37  }
  38
  39 +static int proc_pid_cmdline(struct task_struct *task, char * buffer)
  40 +{
  41 +       int res = 0;
  42 +       unsigned int len;
  43 +       struct mm_struct *mm = get_task_mm(task);
  44 +       if (!mm)
  45 +               goto out;
  46 +       if (!mm->arg_end)
  47 +               goto out_mm;    /* Shh! No looking before we're done */
  48 +
  49 +       len = mm->arg_end - mm->arg_start;
  50 +
  51 +       if (len > PAGE_SIZE)
  52 +               len = PAGE_SIZE;
  53 +
  54 +       res = access_process_vm(task, mm->arg_start, buffer, len, 0);
  55 +
  56 +       // If the nul at the end of args has been overwritten, then
  57 +       // assume application is using setproctitle(3).
  58 +       if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
  59 +               len = strnlen(buffer, res);
  60 +               if (len < res) {
  61 +                   res = len;
  62 +               } else {
  63 +                       len = mm->env_end - mm->env_start;
  64 +                       if (len > PAGE_SIZE - res)
  65 +                               len = PAGE_SIZE - res;
  66 +                       res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
  67 +                       res = strnlen(buffer, res);
  68 +               }
  69 +       }
  70 +out_mm:
  71 +       mmput(mm);
  72 +out:
  73 +       return res;
  74 +}
  75 +
  76 +
  77 +
  78  void oprofile_add_sample(struct pt_regs * const regs, unsigned long event)
  79  {
  80         int is_kernel = !user_mode(regs);
  81         unsigned long pc = profile_pc(regs);
  82 +       int res=0;
  83
  84 +       if (rec_event) {
  85 +               struct event esig;
  86 +               struct event_spec espec;
  87 +               /*res = proc_pid_cmdline(current, espec->appname);*/
  88 +               esig.task = current;
  89 +               espec.pc=pc;
  90 +               espec.count=1;
  91 +               esig.event_data=&espec;
  92 +               esig.event_type=event; /* index in the event array currently set up */
  93 +                                       /* make sure the counters are loaded in the order we want them to show up*/
  94 +               (*rec_event)(&esig, 1);
  95 +       }
  96 +       else {
  97         oprofile_add_ext_sample(pc, regs, event, is_kernel);
  98 +       }
  99  }
 100
 101  void oprofile_add_pc(unsigned long pc, int is_kernel, unsigned long event)
 102 diff -Nurb linux-2.6.22-580/drivers/oprofile/oprof.c linux-2.6.22-590/drivers/oprofile/oprof.c
 103 --- linux-2.6.22-580/drivers/oprofile/oprof.c   2007-07-08 19:32:17.000000000 -0400
 104 +++ linux-2.6.22-590/drivers/oprofile/oprof.c   2008-02-27 13:48:29.000000000 -0500
 105 @@ -33,6 +33,8 @@
 106   */
 107  static int timer = 0;
 108
 109 +extern void *rec_event;
 110 +
 111  int oprofile_setup(void)
 112  {
 113         int err;
 114 @@ -53,7 +55,7 @@
 115          * us missing task deaths and eventually oopsing
 116          * when trying to process the event buffer.
 117          */
 118 -       if ((err = sync_start()))
 119 +       if (err = sync_start())
 120                 goto out3;
 121
 122         is_setup = 1;
 123 diff -Nurb linux-2.6.22-580/fs/exec.c linux-2.6.22-590/fs/exec.c
 124 --- linux-2.6.22-580/fs/exec.c  2008-02-27 13:46:38.000000000 -0500
 125 +++ linux-2.6.22-590/fs/exec.c  2008-02-27 13:49:58.000000000 -0500
 126 @@ -52,6 +52,7 @@
 127  #include <linux/audit.h>
 128  #include <linux/signalfd.h>
 129  #include <linux/vs_memory.h>
 130 +#include <linux/dcookies.h>
 131
 132  #include <asm/uaccess.h>
 133  #include <asm/mmu_context.h>
 134 @@ -488,6 +489,10 @@
 135
 136         if (!err) {
 137                 struct inode *inode = nd.dentry->d_inode;
 138 +               unsigned long cookie;
 139 +               if (!nd.dentry->d_cookie)
 140 +                       get_dcookie(nd.dentry, nd.mnt, &cookie);
 141 +
 142                 file = ERR_PTR(-EACCES);
 143                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 144                     S_ISREG(inode->i_mode)) {
 145 diff -Nurb linux-2.6.22-580/include/linux/arrays.h linux-2.6.22-590/include/linux/arrays.h
 146 --- linux-2.6.22-580/include/linux/arrays.h     1969-12-31 19:00:00.000000000 -0500
 147 +++ linux-2.6.22-590/include/linux/arrays.h     2008-02-27 13:48:29.000000000 -0500
 148 @@ -0,0 +1,35 @@
 149 +#ifndef __ARRAYS_H__
 150 +#define __ARRAYS_H__
 151 +#include <linux/list.h>
 152 +
 153 +#define SAMPLING_METHOD_DEFAULT 0
 154 +#define SAMPLING_METHOD_LOG 1
 155 +
 156 +/* Every probe has an array handler */
 157 +
 158 +/* XXX - Optimize this structure */
 159 +
 160 +struct array_handler {
 161 +       struct list_head link;
 162 +       unsigned int (*hash_func)(void *);
 163 +       unsigned int (*sampling_func)(void *,int,void *);
 164 +       unsigned short size;
 165 +       unsigned int threshold;
 166 +       unsigned char **expcount;
 167 +       unsigned int sampling_method;
 168 +       unsigned int **arrays;
 169 +       unsigned int arraysize;
 170 +       unsigned int num_samples[2];
 171 +       void **epoch_samples; /* size-sized lists of samples */
 172 +       unsigned int (*serialize)(void *, void *);
 173 +       unsigned char code[5];
 174 +};
 175 +
 176 +struct event {
 177 +       struct list_head link;
 178 +       void *event_data;
 179 +       unsigned int count;
 180 +       unsigned int event_type;
 181 +       struct task_struct *task;
 182 +};
 183 +#endif
 184 diff -Nurb linux-2.6.22-580/include/linux/sched.h linux-2.6.22-590/include/linux/sched.h
 185 --- linux-2.6.22-580/include/linux/sched.h      2008-02-27 13:46:40.000000000 -0500
 186 +++ linux-2.6.22-590/include/linux/sched.h      2008-02-27 13:48:29.000000000 -0500
 187 @@ -849,7 +849,7 @@
 188         unsigned int btrace_seq;
 189  #endif
 190         unsigned long sleep_avg;
 191 -       unsigned long long timestamp, last_ran;
 192 +       unsigned long long timestamp, last_ran, last_interrupted, last_ran_j;
 193         unsigned long long sched_time; /* sched_clock time spent running */
 194         enum sleep_type sleep_type;
 195
 196 diff -Nurb linux-2.6.22-580/kernel/fork.c linux-2.6.22-590/kernel/fork.c
 197 --- linux-2.6.22-580/kernel/fork.c      2008-02-27 13:46:40.000000000 -0500
 198 +++ linux-2.6.22-590/kernel/fork.c      2008-02-27 13:48:29.000000000 -0500
 199 @@ -197,6 +197,8 @@
 200         tsk->btrace_seq = 0;
 201  #endif
 202         tsk->splice_pipe = NULL;
 203 +       //tsk->cmdline[0]='\0';
 204 +       tsk->last_interrupted = 0;
 205         return tsk;
 206  }
 207
 208 diff -Nurb linux-2.6.22-580/kernel/sched.c linux-2.6.22-590/kernel/sched.c
 209 --- linux-2.6.22-580/kernel/sched.c     2008-02-27 13:46:40.000000000 -0500
 210 +++ linux-2.6.22-590/kernel/sched.c     2008-02-27 14:08:26.000000000 -0500
 211 @@ -10,7 +10,7 @@
 212   *  1998-11-19 Implemented schedule_timeout() and related stuff
 213   *             by Andrea Arcangeli
 214   *  2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar:
 215 - *             hybrid priority-list and round-robin design with
 216 + *             hybrid priority-list and round-robin deventn with
 217   *             an array-switch method of distributing timeslices
 218   *             and per-CPU runqueues.  Cleanups and useful suggestions
 219   *             by Davide Libenzi, preemptible kernel bits by Robert Love.
 220 @@ -56,6 +56,7 @@
 221
 222  #include <asm/tlb.h>
 223  #include <asm/unistd.h>
 224 +#include <linux/arrays.h>
 225  #include <linux/vs_sched.h>
 226  #include <linux/vs_cvirt.h>
 227
 228 @@ -3608,6 +3609,8 @@
 229
 230  #endif
 231
 232 +extern void (*rec_event)(void *,unsigned int);
 233 +
 234  static inline int interactive_sleep(enum sleep_type sleep_type)
 235  {
 236         return (sleep_type == SLEEP_INTERACTIVE ||
 237 @@ -3617,16 +3620,51 @@
 238  /*
 239   * schedule() is the main scheduler function.
 240   */
 241 +
 242 +struct event_spec {
 243 +       unsigned long pc;
 244 +       unsigned long dcookie;
 245 +       unsigned count;
 246 +       unsigned char reason;
 247 +};
 248 +
 249 +#define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 250 +#define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 251 +
 252 +static inline unsigned long my_get_wchan(struct task_struct *p)
 253 +{
 254 +        unsigned long ebp, esp, eip;
 255 +        unsigned long stack_page;
 256 +        int count = 0;
 257 +        stack_page = (unsigned long)task_stack_page(p);
 258 +        esp = p->thread.esp;
 259 +        if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
 260 +                return 0;
 261 +        /* include/asm-i386/system.h:switch_to() pushes ebp last. */
 262 +        ebp = *(unsigned long *) esp;
 263 +        do {
 264 +                if (ebp < stack_page || ebp > top_ebp+stack_page)
 265 +                        return 0;
 266 +                eip = *(unsigned long *) (ebp+4);
 267 +                if (!in_sched_functions(eip))
 268 +                        return eip;
 269 +                ebp = *(unsigned long *) ebp;
 270 +        } while (count++ < 16);
 271 +        return 0;
 272 +}
 273 +/* CHOPSTIX */
 274 +
 275  asmlinkage void __sched schedule(void)
 276  {
 277         struct task_struct *prev, *next;
 278         struct prio_array *array;
 279         struct list_head *queue;
 280         unsigned long long now;
 281 -       unsigned long run_time;
 282 +       unsigned long run_time, diff;
 283         int cpu, idx, new_prio;
 284         long *switch_count;
 285         struct rq *rq;
 286 +       int sampling_reason;
 287
 288         /*
 289          * Test if we are atomic.  Since do_exit() needs to call into
 290 @@ -3680,6 +3718,7 @@
 291         switch_count = &prev->nivcsw;
 292         if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
 293                 switch_count = &prev->nvcsw;
 294 +
 295                 if (unlikely((prev->state & TASK_INTERRUPTIBLE) &&
 296                                 unlikely(signal_pending(prev))))
 297                         prev->state = TASK_RUNNING;
 298 @@ -3689,6 +3728,9 @@
 299                                 vx_uninterruptible_inc(prev);
 300                         }
 301                         deactivate_task(prev, rq);
 302 +                       if (prev->state & TASK_INTERRUPTIBLE) {
 303 +                               prev->last_interrupted=jiffies;
 304 +                       }
 305                 }
 306         }
 307
 308 @@ -3763,8 +3805,44 @@
 309         prev->sleep_avg -= run_time;
 310         if ((long)prev->sleep_avg <= 0)
 311                 prev->sleep_avg = 0;
 312 +
 313         prev->timestamp = prev->last_ran = now;
 314
 315 +       /* CHOPSTIX */
 316 +
 317 +       prev->last_ran_j = jiffies;
 318 +       if (next->last_interrupted) {
 319 +               diff = (jiffies-next->last_interrupted);
 320 +               next->last_interrupted = 0;
 321 +               sampling_reason = 0;
 322 +       }
 323 +       else {
 324 +               diff = jiffies-next->last_ran_j;
 325 +               sampling_reason = 1;
 326 +       }
 327 +
 328 +       if (rec_event && (diff>HZ/5)) {
 329 +               struct event event;
 330 +               struct event_spec espec;
 331 +               unsigned long eip;
 332 +               unsigned int state = next->state;
 333 +
 334 +               espec.reason = sampling_reason;
 335 +
 336 +               next->state = 0;
 337 +               eip = next->thread.esp;
 338 +               next->state = state;
 339 +
 340 +               next->last_interrupted = 0;
 341 +               event.event_data=&espec;
 342 +               event.task=next;
 343 +               espec.pc=eip;
 344 +               event.event_type=2;
 345 +               /* index in the event array currently set up */
 346 +               /* make sure the counters are loaded in the order we want them to show up*/
 347 +               (*rec_event)(&event, diff);
 348 +       }
 349 +
 350         sched_info_switch(prev, next);
 351         if (likely(prev != next)) {
 352                 next->timestamp = next->last_ran = now;
 353 @@ -7275,3 +7353,7 @@
 354  }
 355
 356  #endif
 357 +
 358 +void (*rec_event)(void *,unsigned int);
 359 +EXPORT_SYMBOL(rec_event);
 360 +EXPORT_SYMBOL(in_sched_functions);
 361 diff -Nurb linux-2.6.22-580/kernel/sched.c.rej linux-2.6.22-590/kernel/sched.c.rej
 362 --- linux-2.6.22-580/kernel/sched.c.rej 1969-12-31 19:00:00.000000000 -0500
 363 +++ linux-2.6.22-590/kernel/sched.c.rej 2008-02-27 13:48:29.000000000 -0500
 364 @@ -0,0 +1,18 @@
 365 +***************
 366 +*** 56,61 ****
 367 +
 368 +  #include <asm/tlb.h>
 369 +  #include <asm/unistd.h>
 370 +
 371 +  /*
 372 +   * Scheduler clock - returns current time in nanosec units.
 373 +--- 56,64 ----
 374 +
 375 +  #include <asm/tlb.h>
 376 +  #include <asm/unistd.h>
 377 ++ #include <linux/arrays.h>
 378 ++
 379 ++
 380 +
 381 +  /*
 382 +   * Scheduler clock - returns current time in nanosec units.