Differences

This shows you the differences between two versions of the page.

--- etc:users:jcmvbkbc:linux-mm [2009/07/13 17:16] – jcmvbkbc
+++ etc:users:jcmvbkbc:linux-mm [2016/08/08 20:53] (current) – ↷ Page moved from users:jcmvbkbc:linux-mm to etc:users:jcmvbkbc:linux-mm kel
@@ Line 1: / Line 1: @@
 ====== Linux memory management summary ======
 Every part of process address space is some sort of mapping.
 ===== Memory accounting =====
@@ Line 25: / Line 27: @@
   * private_clean -- size of memory mapped only to this process and haven't been written to;
   * private_clean -- size of memory mapped only to this process and changed;
-  * referenced -- size of data that were accessed (either read or written);
+  * referenced -- size of data that were accessed. Each physical memory page has an attribute, which may be reset by software and asserted by CPU automatically when the page is accessed. This is used to monitor memory usage activity & to recycle least recently used pages first;
   * swap -- size of region data that is currently in swap file;
-  * Charging memory to processes
-  * total_vm
 ==== Information sources ====
+Quotes from linux/Documentation/filesystems/proc.txt
 Per-process data:
-  * /proc/<PID>/maps -- process memory map;
+  * /proc/<PID>/map -- process memory map;
+<code>
+The /proc/PID/map file containing the currently mapped memory regions and
+their access permissions.
+The format is:
+address           perms offset  dev   inode      pathname
+08048000-08049000 r-xp 00000000 03:00 8312       /opt/test
+08049000-0804a000 rw-p 00001000 03:00 8312       /opt/test
+a000-0806b000 rw-p 00000000 00:00 0          [heap]
+a7cb1000-a7cb2000 ---p 00000000 00:00 0
+a7cb2000-a7eb2000 rw-p 00000000 00:00 0
+a7eb2000-a7eb3000 ---p 00000000 00:00 0
+a7eb3000-a7ed5000 rw-p 00000000 00:00 0
+a7ed5000-a8008000 r-xp 00000000 03:00 4222       /lib/libc.so.6
+a8008000-a800a000 r--p 00133000 03:00 4222       /lib/libc.so.6
+a800a000-a800b000 rw-p 00135000 03:00 4222       /lib/libc.so.6
+a800b000-a800e000 rw-p 00000000 00:00 0
+a800e000-a8022000 r-xp 00000000 03:00 14462      /lib/libpthread.so.0
+a8022000-a8023000 r--p 00013000 03:00 14462      /lib/libpthread.so.0
+a8023000-a8024000 rw-p 00014000 03:00 14462      /lib/libpthread.so.0
+a8024000-a8027000 rw-p 00000000 00:00 0
+a8027000-a8043000 r-xp 00000000 03:00 8317       /lib/ld-linux.so.2
+a8043000-a8044000 r--p 0001b000 03:00 8317       /lib/ld-linux.so.2
+a8044000-a8045000 rw-p 0001c000 03:00 8317       /lib/ld-linux.so.2
+aff35000-aff4a000 rw-p 00000000 00:00 0          [stack]
+ffffe000-fffff000 r-xp 00000000 00:00 0          [vdso]
+where "address" is the address space in the process that it occupies, "perms"
+is a set of permissions:
+ r = read
+ w = write
+ x = execute
+ s = shared
+ p = private (copy on write)
+"offset" is the offset into the mapping, "dev" is the device (major:minor), and
+"inode" is the inode  on that device.  0 indicates that  no inode is associated
+with the memory region, as the case would be with BSS (uninitialized data).
+The "pathname" shows the name associated file for this mapping.  If the mapping
+is not associated with a file:
+ [heap]                   = the heap of the program
+ [stack]                  = the stack of the main process
+ [vdso]                   = the "virtual dynamic shared object",
+                            the kernel system call handler
+ or if empty, the mapping is anonymous.
+</code>
   * /proc/<PID>/smaps -- detailed process memory map;
+<code>
+08048000-080bc000 r-xp 00000000 03:02 13130      /bin/bash
+Size:               1084 kB
+Rss:                 892 kB
+Pss:                 374 kB
+Shared_Clean:        892 kB
+Shared_Dirty:          0 kB
+Private_Clean:         0 kB
+Private_Dirty:         0 kB
+Referenced:          892 kB
+Swap:                  0 kB
+KernelPageSize:        4 kB
+MMUPageSize:           4 kB
+</code>
   * /proc/<PID>/statm -- process memory usage summary;
+<code>
+ Field    Content
+ size     total program size (pages)            (same as VmSize in status)
+ resident size of memory portions (pages)       (same as VmRSS in status)
+ shared   number of pages that are shared       (i.e. backed by a file)
+ trs      number of pages that are 'code'       (not including libs; broken,
+                                                        includes data segment)
+ lrs      number of pages of library            (always 0 on 2.6)
+ drs      number of pages of data/stack         (including libs; broken,
+                                                        includes library text)
+ dt       number of dirty pages                 (always 0 on 2.6)
+</code>
   * /proc/<PID>/status -- process status;
+<code>
+  >cat /proc/self/status
+  Name:   cat
+  State:  R (running)
+  Tgid:   5452
+  Pid:    5452
+  PPid:   743
+  TracerPid:      0                                             (2.4)
+  Uid:    501     501     501     501
+  Gid:    100     100     100     100
+  FDSize: 256
+  Groups: 100 14 16
+  VmPeak:     5004 kB
+  VmSize:     5004 kB
+  VmLck:         0 kB
+  VmHWM:       476 kB
+  VmRSS:       476 kB
+  VmData:      156 kB
+  VmStk:        88 kB
+  VmExe:        68 kB
+  VmLib:      1412 kB
+  VmPTE:        20 kb
+  Threads:        1
+  SigQ:   0/28578
+  SigPnd: 0000000000000000
+  ShdPnd: 0000000000000000
+  SigBlk: 0000000000000000
+  SigIgn: 0000000000000000
+  SigCgt: 0000000000000000
+  CapInh: 00000000fffffeff
+  CapPrm: 0000000000000000
+  CapEff: 0000000000000000
+  CapBnd: ffffffffffffffff
+  voluntary_ctxt_switches:        0
+  nonvoluntary_ctxt_switches:     1
+</code>
   * /proc/<PID>/oom_adj
   * /proc/<PID>/oom_score
@@ Line 61: / Line 176: @@
     * Resident is what currently in physical memory;
     * If not resident, it may be swapped or not backed by whatever storage at all.
 ==== What to expect ====
   * of heap usage
+    * private_dirty or swap size of [heap] region grows;
+    * size of [heap] region grows;
   * of file mapping
   * of anonymous mapping
@@ Line 71: / Line 189: @@
   * of threads
     * tls?
 ===== OOM killing =====
 ==== When this happens ====
@@ Line 115: / Line 234: @@
     * if (oom_adj > 0) points <<= oom_adj; //(if points == 0 before shift, points = 1)//
     * if (oom_adj < 0) points >>= -oom_adj;
 ==== How to control OOM-killer ====
@@ Line 122: / Line 243: @@
   * vm.panic_on_oom -- panic in case of OOM, instead of trying to kill some processes;
   * vm.oom_kill_allocating_task -- try first to kill task that issued request for memory that caused OOM condition;
-  * vm.oom_dump_tasks
+  * vm.oom_dump_tasks -- dump memory summary, stack of the process caused oom and table of processes to log before oom-killing;
   * vm.would_have_oomkilled
 ===== Memleak detection =====
 ==== Direct memleak evidences ====
+$ cat /proc/<PID>/smaps
+And monitor [heap] swap+private_dirty
+<code>
+08143000-bfd30000 rw-p 08143000 00:00 0          [heap]
+Size:            3010484 kB
+Rss:              475660 kB
+Pss:              475660 kB
+Shared_Clean:          0 kB
+Shared_Dirty:          0 kB
+Private_Clean:         0 kB
+Private_Dirty:    475660 kB
+Referenced:            0 kB
+Swap:             727624 kB
+</code>
+<code>
+08143000-bfd30000 rw-p 08143000 00:00 0          [heap]
+Size:            3010484 kB
+Rss:                   0 kB
+Pss:                   0 kB
+Shared_Clean:          0 kB
+Shared_Dirty:          0 kB
+Private_Clean:         0 kB
+Private_Dirty:         0 kB
+Referenced:            0 kB
+Swap:            1203284 kB
+</code>