libcfs/libcfs/tracefile.h

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #ifndef __LIBCFS_TRACEFILE_H__
  38 #define __LIBCFS_TRACEFILE_H__
  39
  40 #include <libcfs/libcfs.h>
  41
  42 #if defined(__linux__)
  43 #include "linux/linux-tracefile.h"
  44 #elif defined(__WINNT__)
  45 #include "winnt/winnt-tracefile.h"
  46 #else
  47 #error Unsupported operating system.
  48 #endif
  49
  50 /* trace file lock routines */
  51
  52 #define TRACEFILE_NAME_SIZE 1024
  53 extern char      cfs_tracefile[TRACEFILE_NAME_SIZE];
  54 extern long long cfs_tracefile_size;
  55
  56 extern void libcfs_run_debug_log_upcall(char *file);
  57
  58 int  cfs_tracefile_init_arch(void);
  59 void cfs_tracefile_fini_arch(void);
  60
  61 void cfs_tracefile_read_lock(void);
  62 void cfs_tracefile_read_unlock(void);
  63 void cfs_tracefile_write_lock(void);
  64 void cfs_tracefile_write_unlock(void);
  65
  66 int cfs_tracefile_dump_all_pages(char *filename);
  67 void cfs_trace_debug_print(void);
  68 void cfs_trace_flush_pages(void);
  69 int cfs_trace_start_thread(void);
  70 void cfs_trace_stop_thread(void);
  71 int cfs_tracefile_init(int max_pages);
  72 void cfs_tracefile_exit(void);
  73
  74
  75
  76 int cfs_trace_copyin_string(char *knl_buffer, int knl_buffer_nob,
  77                             const char *usr_buffer, int usr_buffer_nob);
  78 int cfs_trace_copyout_string(char *usr_buffer, int usr_buffer_nob,
  79                              const char *knl_str, char *append);
  80 int cfs_trace_allocate_string_buffer(char **str, int nob);
  81 void cfs_trace_free_string_buffer(char *str, int nob);
  82 int cfs_trace_dump_debug_buffer_usrstr(void *usr_str, int usr_str_nob);
  83 int cfs_trace_daemon_command(char *str);
  84 int cfs_trace_daemon_command_usrstr(void *usr_str, int usr_str_nob);
  85 int cfs_trace_set_debug_mb(int mb);
  86 int cfs_trace_set_debug_mb_usrstr(void *usr_str, int usr_str_nob);
  87 int cfs_trace_get_debug_mb(void);
  88
  89 extern void libcfs_debug_dumplog_internal(void *arg);
  90 extern void libcfs_register_panic_notifier(void);
  91 extern void libcfs_unregister_panic_notifier(void);
  92 extern int  libcfs_panic_in_progress;
  93 extern int  cfs_trace_max_debug_mb(void);
  94
  95 #define TCD_MAX_PAGES (5 << (20 - CFS_PAGE_SHIFT))
  96 #define TCD_STOCK_PAGES (TCD_MAX_PAGES)
  97 #define CFS_TRACEFILE_SIZE (500 << 20)
  98
  99 #ifdef LUSTRE_TRACEFILE_PRIVATE
 100
 101 /*
 102  * Private declare for tracefile
 103  */
 104 #define TCD_MAX_PAGES (5 << (20 - CFS_PAGE_SHIFT))
 105 #define TCD_STOCK_PAGES (TCD_MAX_PAGES)
 106
 107 #define CFS_TRACEFILE_SIZE (500 << 20)
 108
 109 /* Size of a buffer for sprinting console messages if we can't get a page
 110  * from system */
 111 #define CFS_TRACE_CONSOLE_BUFFER_SIZE   1024
 112
 113 union cfs_trace_data_union {
 114         struct cfs_trace_cpu_data {
 115                 /*
 116                  * Even though this structure is meant to be per-CPU, locking
 117                  * is needed because in some places the data may be accessed
 118                  * from other CPUs. This lock is directly used in trace_get_tcd
 119                  * and trace_put_tcd, which are called in libcfs_debug_vmsg2 and
 120                  * tcd_for_each_type_lock
 121                  */
 122                 cfs_spinlock_t          tcd_lock;
 123                 unsigned long           tcd_lock_flags;
 124
 125                 /*
 126                  * pages with trace records not yet processed by tracefiled.
 127                  */
 128                 cfs_list_t              tcd_pages;
 129                 /* number of pages on ->tcd_pages */
 130                 unsigned long           tcd_cur_pages;
 131
 132                 /*
 133                  * pages with trace records already processed by
 134                  * tracefiled. These pages are kept in memory, so that some
 135                  * portion of log can be written in the event of LBUG. This
 136                  * list is maintained in LRU order.
 137                  *
 138                  * Pages are moved to ->tcd_daemon_pages by tracefiled()
 139                  * (put_pages_on_daemon_list()). LRU pages from this list are
 140                  * discarded when list grows too large.
 141                  */
 142                 cfs_list_t              tcd_daemon_pages;
 143                 /* number of pages on ->tcd_daemon_pages */
 144                 unsigned long           tcd_cur_daemon_pages;
 145
 146                 /*
 147                  * Maximal number of pages allowed on ->tcd_pages and
 148                  * ->tcd_daemon_pages each.
 149                  * Always TCD_MAX_PAGES * tcd_pages_factor / 100 in current
 150                  * implementation.
 151                  */
 152                 unsigned long           tcd_max_pages;
 153
 154                 /*
 155                  * preallocated pages to write trace records into. Pages from
 156                  * ->tcd_stock_pages are moved to ->tcd_pages by
 157                  * portals_debug_msg().
 158                  *
 159                  * This list is necessary, because on some platforms it's
 160                  * impossible to perform efficient atomic page allocation in a
 161                  * non-blockable context.
 162                  *
 163                  * Such platforms fill ->tcd_stock_pages "on occasion", when
 164                  * tracing code is entered in blockable context.
 165                  *
 166                  * trace_get_tage_try() tries to get a page from
 167                  * ->tcd_stock_pages first and resorts to atomic page
 168                  * allocation only if this queue is empty. ->tcd_stock_pages
 169                  * is replenished when tracing code is entered in blocking
 170                  * context (darwin-tracefile.c:trace_get_tcd()). We try to
 171                  * maintain TCD_STOCK_PAGES (40 by default) pages in this
 172                  * queue. Atomic allocation is only required if more than
 173                  * TCD_STOCK_PAGES pagesful are consumed by trace records all
 174                  * emitted in non-blocking contexts. Which is quite unlikely.
 175                  */
 176                 cfs_list_t              tcd_stock_pages;
 177                 /* number of pages on ->tcd_stock_pages */
 178                 unsigned long           tcd_cur_stock_pages;
 179
 180                 unsigned short          tcd_shutting_down;
 181                 unsigned short          tcd_cpu;
 182                 unsigned short          tcd_type;
 183                 /* The factors to share debug memory. */
 184                 unsigned short          tcd_pages_factor;
 185         } tcd;
 186         char __pad[CFS_L1_CACHE_ALIGN(sizeof(struct cfs_trace_cpu_data))];
 187 };
 188
 189 #define TCD_MAX_TYPES      8
 190 extern union cfs_trace_data_union (*cfs_trace_data[TCD_MAX_TYPES])[CFS_NR_CPUS];
 191
 192 #define cfs_tcd_for_each(tcd, i, j)                                       \
 193     for (i = 0; cfs_trace_data[i] != NULL; i++)                           \
 194         for (j = 0, ((tcd) = &(*cfs_trace_data[i])[j].tcd);               \
 195              j < cfs_num_possible_cpus();                                 \
 196              j++, (tcd) = &(*cfs_trace_data[i])[j].tcd)
 197
 198 #define cfs_tcd_for_each_type_lock(tcd, i, cpu)                           \
 199     for (i = 0; cfs_trace_data[i] &&                                      \
 200          (tcd = &(*cfs_trace_data[i])[cpu].tcd) &&                        \
 201          cfs_trace_lock_tcd(tcd); cfs_trace_unlock_tcd(tcd), i++)
 202
 203 /* XXX nikita: this declaration is internal to tracefile.c and should probably
 204  * be moved there */
 205 struct page_collection {
 206         cfs_list_t              pc_pages;
 207         /*
 208          * spin-lock protecting ->pc_pages. It is taken by smp_call_function()
 209          * call-back functions. XXX nikita: Which is horrible: all processors
 210          * receive NMI at the same time only to be serialized by this
 211          * lock. Probably ->pc_pages should be replaced with an array of
 212          * NR_CPUS elements accessed locklessly.
 213          */
 214         cfs_spinlock_t          pc_lock;
 215         /*
 216          * if this flag is set, collect_pages() will spill both
 217          * ->tcd_daemon_pages and ->tcd_pages to the ->pc_pages. Otherwise,
 218          * only ->tcd_pages are spilled.
 219          */
 220         int                     pc_want_daemon_pages;
 221 };
 222
 223 /* XXX nikita: this declaration is internal to tracefile.c and should probably
 224  * be moved there */
 225 struct tracefiled_ctl {
 226         cfs_completion_t       tctl_start;
 227         cfs_completion_t       tctl_stop;
 228         cfs_waitq_t            tctl_waitq;
 229         pid_t                  tctl_pid;
 230         cfs_atomic_t           tctl_shutdown;
 231 };
 232
 233 /*
 234  * small data-structure for each page owned by tracefiled.
 235  */
 236 /* XXX nikita: this declaration is internal to tracefile.c and should probably
 237  * be moved there */
 238 struct cfs_trace_page {
 239         /*
 240          * page itself
 241          */
 242         cfs_page_t          *page;
 243         /*
 244          * linkage into one of the lists in trace_data_union or
 245          * page_collection
 246          */
 247         cfs_list_t           linkage;
 248         /*
 249          * number of bytes used within this page
 250          */
 251         unsigned int         used;
 252         /*
 253          * cpu that owns this page
 254          */
 255         unsigned short       cpu;
 256         /*
 257          * type(context) of this page
 258          */
 259         unsigned short       type;
 260 };
 261
 262 extern void cfs_set_ptldebug_header(struct ptldebug_header *header,
 263                                     int subsys, int mask, const int line,
 264                                     unsigned long stack);
 265 extern void cfs_print_to_console(struct ptldebug_header *hdr, int mask,
 266                                  const char *buf, int len, const char *file,
 267                                  const char *fn);
 268
 269 extern int cfs_trace_lock_tcd(struct cfs_trace_cpu_data *tcd);
 270 extern void cfs_trace_unlock_tcd(struct cfs_trace_cpu_data *tcd);
 271
 272 /**
 273  * trace_buf_type_t, trace_buf_idx_get() and trace_console_buffers[][]
 274  * are not public libcfs API; they should be defined in
 275  * platform-specific tracefile include files
 276  * (see, for example, linux-tracefile.h).
 277  */
 278
 279 extern char *cfs_trace_console_buffers[CFS_NR_CPUS][CFS_TCD_TYPE_MAX];
 280 extern cfs_trace_buf_type_t cfs_trace_buf_idx_get(void);
 281
 282 static inline char *
 283 cfs_trace_get_console_buffer(void)
 284 {
 285         unsigned int i = cfs_get_cpu();
 286         unsigned int j = cfs_trace_buf_idx_get();
 287
 288         return cfs_trace_console_buffers[i][j];
 289 }
 290
 291 static inline void
 292 cfs_trace_put_console_buffer(char *buffer)
 293 {
 294         cfs_put_cpu();
 295 }
 296
 297 static inline struct cfs_trace_cpu_data *
 298 cfs_trace_get_tcd(void)
 299 {
 300         struct cfs_trace_cpu_data *tcd =
 301                 &(*cfs_trace_data[cfs_trace_buf_idx_get()])[cfs_get_cpu()].tcd;
 302
 303         cfs_trace_lock_tcd(tcd);
 304
 305         return tcd;
 306 }
 307
 308 static inline void
 309 cfs_trace_put_tcd (struct cfs_trace_cpu_data *tcd)
 310 {
 311         cfs_trace_unlock_tcd(tcd);
 312
 313         cfs_put_cpu();
 314 }
 315
 316 int cfs_trace_refill_stock(struct cfs_trace_cpu_data *tcd, int gfp,
 317                            cfs_list_t *stock);
 318
 319
 320 int cfs_tcd_owns_tage(struct cfs_trace_cpu_data *tcd,
 321                       struct cfs_trace_page *tage);
 322
 323 extern void cfs_trace_assertion_failed(const char *str, const char *fn,
 324                                        const char *file, int line);
 325
 326 /* ASSERTION that is safe to use within the debug system */
 327 #define __LASSERT(cond)                                                 \
 328     do {                                                                \
 329         if (unlikely(!(cond))) {                                        \
 330                 cfs_trace_assertion_failed("ASSERTION("#cond") failed", \
 331                                  __FUNCTION__, __FILE__, __LINE__);     \
 332         }                                                               \
 333     } while (0)
 334
 335 #define __LASSERT_TAGE_INVARIANT(tage)                                  \
 336     do {                                                                \
 337         __LASSERT(tage != NULL);                                        \
 338         __LASSERT(tage->page != NULL);                                  \
 339         __LASSERT(tage->used <= CFS_PAGE_SIZE);                         \
 340         __LASSERT(cfs_page_count(tage->page) > 0);                      \
 341     } while (0)
 342
 343 #endif  /* LUSTRE_TRACEFILE_PRIVATE */
 344
 345 #endif /* __LIBCFS_TRACEFILE_H__ */