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/fs/exec.c linux-2.6.22-590/fs/exec.c
 103 --- linux-2.6.22-580/fs/exec.c  2008-02-27 13:46:38.000000000 -0500
 104 +++ linux-2.6.22-590/fs/exec.c  2008-02-27 13:49:58.000000000 -0500
 105 @@ -52,6 +52,7 @@
 106  #include <linux/audit.h>
 107  #include <linux/signalfd.h>
 108  #include <linux/vs_memory.h>
 109 +#include <linux/dcookies.h>
 110
 111  #include <asm/uaccess.h>
 112  #include <asm/mmu_context.h>
 113 @@ -488,6 +489,10 @@
 114
 115         if (!err) {
 116                 struct inode *inode = nd.dentry->d_inode;
 117 +               unsigned long cookie;
 118 +               if (!nd.dentry->d_cookie)
 119 +                       get_dcookie(nd.dentry, nd.mnt, &cookie);
 120 +
 121                 file = ERR_PTR(-EACCES);
 122                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 123                     S_ISREG(inode->i_mode)) {
 124 diff -Nurb linux-2.6.22-580/include/linux/arrays.h linux-2.6.22-590/include/linux/arrays.h
 125 --- linux-2.6.22-580/include/linux/arrays.h     1969-12-31 19:00:00.000000000 -0500
 126 +++ linux-2.6.22-590/include/linux/arrays.h     2008-02-27 13:48:29.000000000 -0500
 127 @@ -0,0 +1,35 @@
 128 +#ifndef __ARRAYS_H__
 129 +#define __ARRAYS_H__
 130 +#include <linux/list.h>
 131 +
 132 +#define SAMPLING_METHOD_DEFAULT 0
 133 +#define SAMPLING_METHOD_LOG 1
 134 +
 135 +/* Every probe has an array handler */
 136 +
 137 +/* XXX - Optimize this structure */
 138 +
 139 +struct array_handler {
 140 +       struct list_head link;
 141 +       unsigned int (*hash_func)(void *);
 142 +       unsigned int (*sampling_func)(void *,int,void *);
 143 +       unsigned short size;
 144 +       unsigned int threshold;
 145 +       unsigned char **expcount;
 146 +       unsigned int sampling_method;
 147 +       unsigned int **arrays;
 148 +       unsigned int arraysize;
 149 +       unsigned int num_samples[2];
 150 +       void **epoch_samples; /* size-sized lists of samples */
 151 +       unsigned int (*serialize)(void *, void *);
 152 +       unsigned char code[5];
 153 +};
 154 +
 155 +struct event {
 156 +       struct list_head link;
 157 +       void *event_data;
 158 +       unsigned int count;
 159 +       unsigned int event_type;
 160 +       struct task_struct *task;
 161 +};
 162 +#endif
 163 diff -Nurb linux-2.6.22-580/include/linux/sched.h linux-2.6.22-590/include/linux/sched.h
 164 --- linux-2.6.22-580/include/linux/sched.h      2008-02-27 13:46:40.000000000 -0500
 165 +++ linux-2.6.22-590/include/linux/sched.h      2008-02-27 13:48:29.000000000 -0500
 166 @@ -849,7 +849,7 @@
 167         unsigned int btrace_seq;
 168  #endif
 169         unsigned long sleep_avg;
 170 -       unsigned long long timestamp, last_ran;
 171 +       unsigned long long timestamp, last_ran, last_interrupted, last_ran_j;
 172         unsigned long long sched_time; /* sched_clock time spent running */
 173         enum sleep_type sleep_type;
 174
 175 diff -Nurb linux-2.6.22-580/kernel/fork.c linux-2.6.22-590/kernel/fork.c
 176 --- linux-2.6.22-580/kernel/fork.c      2008-02-27 13:46:40.000000000 -0500
 177 +++ linux-2.6.22-590/kernel/fork.c      2008-02-27 13:48:29.000000000 -0500
 178 @@ -197,6 +197,8 @@
 179         tsk->btrace_seq = 0;
 180  #endif
 181         tsk->splice_pipe = NULL;
 182 +       //tsk->cmdline[0]='\0';
 183 +       tsk->last_interrupted = 0;
 184         return tsk;
 185  }
 186
 187 diff -Nurb linux-2.6.22-580/kernel/sched.c linux-2.6.22-590/kernel/sched.c
 188 --- linux-2.6.22-580/kernel/sched.c     2008-02-27 13:46:40.000000000 -0500
 189 +++ linux-2.6.22-590/kernel/sched.c     2008-02-27 14:08:26.000000000 -0500
 190 @@ -10,7 +10,7 @@
 191   *  1998-11-19 Implemented schedule_timeout() and related stuff
 192   *             by Andrea Arcangeli
 193   *  2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar:
 194 - *             hybrid priority-list and round-robin design with
 195 + *             hybrid priority-list and round-robin deventn with
 196   *             an array-switch method of distributing timeslices
 197   *             and per-CPU runqueues.  Cleanups and useful suggestions
 198   *             by Davide Libenzi, preemptible kernel bits by Robert Love.
 199 @@ -56,6 +56,7 @@
 200
 201  #include <asm/tlb.h>
 202  #include <asm/unistd.h>
 203 +#include <linux/arrays.h>
 204  #include <linux/vs_sched.h>
 205  #include <linux/vs_cvirt.h>
 206
 207 @@ -3608,6 +3609,8 @@
 208
 209  #endif
 210
 211 +extern void (*rec_event)(void *,unsigned int);
 212 +
 213  static inline int interactive_sleep(enum sleep_type sleep_type)
 214  {
 215         return (sleep_type == SLEEP_INTERACTIVE ||
 216 @@ -3617,16 +3620,51 @@
 217  /*
 218   * schedule() is the main scheduler function.
 219   */
 220 +
 221 +struct event_spec {
 222 +       unsigned long pc;
 223 +       unsigned long dcookie;
 224 +       unsigned count;
 225 +       unsigned char reason;
 226 +};
 227 +
 228 +#define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 229 +#define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 230 +
 231 +static inline unsigned long my_get_wchan(struct task_struct *p)
 232 +{
 233 +        unsigned long ebp, esp, eip;
 234 +        unsigned long stack_page;
 235 +        int count = 0;
 236 +        stack_page = (unsigned long)task_stack_page(p);
 237 +        esp = p->thread.esp;
 238 +        if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
 239 +                return 0;
 240 +        /* include/asm-i386/system.h:switch_to() pushes ebp last. */
 241 +        ebp = *(unsigned long *) esp;
 242 +        do {
 243 +                if (ebp < stack_page || ebp > top_ebp+stack_page)
 244 +                        return 0;
 245 +                eip = *(unsigned long *) (ebp+4);
 246 +                if (!in_sched_functions(eip))
 247 +                        return eip;
 248 +                ebp = *(unsigned long *) ebp;
 249 +        } while (count++ < 16);
 250 +        return 0;
 251 +}
 252 +/* CHOPSTIX */
 253 +
 254  asmlinkage void __sched schedule(void)
 255  {
 256         struct task_struct *prev, *next;
 257         struct prio_array *array;
 258         struct list_head *queue;
 259         unsigned long long now;
 260 -       unsigned long run_time;
 261 +       unsigned long run_time, diff;
 262         int cpu, idx, new_prio;
 263         long *switch_count;
 264         struct rq *rq;
 265 +       int sampling_reason;
 266
 267         /*
 268          * Test if we are atomic.  Since do_exit() needs to call into
 269 @@ -3680,6 +3718,7 @@
 270         switch_count = &prev->nivcsw;
 271         if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
 272                 switch_count = &prev->nvcsw;
 273 +
 274                 if (unlikely((prev->state & TASK_INTERRUPTIBLE) &&
 275                                 unlikely(signal_pending(prev))))
 276                         prev->state = TASK_RUNNING;
 277 @@ -3689,6 +3728,9 @@
 278                                 vx_uninterruptible_inc(prev);
 279                         }
 280                         deactivate_task(prev, rq);
 281 +                       if (prev->state & TASK_INTERRUPTIBLE) {
 282 +                               prev->last_interrupted=jiffies;
 283 +                       }
 284                 }
 285         }
 286
 287 @@ -3763,8 +3805,44 @@
 288         prev->sleep_avg -= run_time;
 289         if ((long)prev->sleep_avg <= 0)
 290                 prev->sleep_avg = 0;
 291 +
 292         prev->timestamp = prev->last_ran = now;
 293
 294 +       /* CHOPSTIX */
 295 +
 296 +       prev->last_ran_j = jiffies;
 297 +       if (next->last_interrupted) {
 298 +               diff = (jiffies-next->last_interrupted);
 299 +               next->last_interrupted = 0;
 300 +               sampling_reason = 0;
 301 +       }
 302 +       else {
 303 +               diff = jiffies-next->last_ran_j;
 304 +               sampling_reason = 1;
 305 +       }
 306 +
 307 +       if (rec_event && (diff>HZ/5)) {
 308 +               struct event event;
 309 +               struct event_spec espec;
 310 +               unsigned long eip;
 311 +               unsigned int state = next->state;
 312 +
 313 +               espec.reason = sampling_reason;
 314 +
 315 +               next->state = 0;
 316 +               eip = next->thread.esp;
 317 +               next->state = state;
 318 +
 319 +               next->last_interrupted = 0;
 320 +               event.event_data=&espec;
 321 +               event.task=next;
 322 +               espec.pc=eip;
 323 +               event.event_type=2;
 324 +               /* index in the event array currently set up */
 325 +               /* make sure the counters are loaded in the order we want them to show up*/
 326 +               (*rec_event)(&event, diff);
 327 +       }
 328 +
 329         sched_info_switch(prev, next);
 330         if (likely(prev != next)) {
 331                 next->timestamp = next->last_ran = now;
 332 @@ -7275,3 +7353,7 @@
 333  }
 334
 335  #endif
 336 +
 337 +void (*rec_event)(void *,unsigned int);
 338 +EXPORT_SYMBOL(rec_event);
 339 +EXPORT_SYMBOL(in_sched_functions);
 340 diff -Nurb linux-2.6.22-580/kernel/sched.c.rej linux-2.6.22-590/kernel/sched.c.rej
 341 --- linux-2.6.22-580/kernel/sched.c.rej 1969-12-31 19:00:00.000000000 -0500
 342 +++ linux-2.6.22-590/kernel/sched.c.rej 2008-02-27 13:48:29.000000000 -0500
 343 @@ -0,0 +1,18 @@
 344 +***************
 345 +*** 56,61 ****
 346 +
 347 +  #include <asm/tlb.h>
 348 +  #include <asm/unistd.h>
 349 +
 350 +  /*
 351 +   * Scheduler clock - returns current time in nanosec units.
 352 +--- 56,64 ----
 353 +
 354 +  #include <asm/tlb.h>
 355 +  #include <asm/unistd.h>
 356 ++ #include <linux/arrays.h>
 357 ++
 358 ++
 359 +
 360 +  /*
 361 +   * Scheduler clock - returns current time in nanosec units.