proc: partially revert "procfs: provide stack information for threads"
authorKOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Fri, 8 Jan 2010 22:42:56 +0000 (14:42 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Mon, 11 Jan 2010 17:34:06 +0000 (09:34 -0800)
Commit d899bf7b (procfs: provide stack information for threads) introduced
to show stack information in /proc/{pid}/status.  But it cause large
performance regression.  Unfortunately /proc/{pid}/status is used ps
command too and ps is one of most important component.  Because both to
take mmap_sem and page table walk are heavily operation.

If many process run, the ps performance is,

[before d899bf7b]

% perf stat ps >/dev/null

 Performance counter stats for 'ps':

     4090.435806  task-clock-msecs         #      0.032 CPUs
             229  context-switches         #      0.000 M/sec
               0  CPU-migrations           #      0.000 M/sec
             234  page-faults              #      0.000 M/sec
      8587565207  cycles                   #   2099.425 M/sec
      9866662403  instructions             #      1.149 IPC
      3789415411  cache-references         #    926.409 M/sec
        30419509  cache-misses             #      7.437 M/sec

   128.859521955  seconds time elapsed

[after d899bf7b]

% perf stat  ps  > /dev/null

 Performance counter stats for 'ps':

     4305.081146  task-clock-msecs         #      0.028 CPUs
             480  context-switches         #      0.000 M/sec
               2  CPU-migrations           #      0.000 M/sec
             237  page-faults              #      0.000 M/sec
      9021211334  cycles                   #   2095.480 M/sec
     10605887536  instructions             #      1.176 IPC
      3612650999  cache-references         #    839.160 M/sec
        23917502  cache-misses             #      5.556 M/sec

   152.277819582  seconds time elapsed

Thus, this patch revert it. Fortunately /proc/{pid}/task/{tid}/smaps
provide almost same information. we can use it.

Commit d899bf7b introduced two features:

 1) Add the annotattion of [thread stack: xxxx] mark to
    /proc/{pid}/task/{tid}/maps.
 2) Add StackUsage field to /proc/{pid}/status.

I only revert (2), because I haven't seen (1) cause regression.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Stefani Seibold <stefani@seibold.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Documentation/filesystems/proc.txt
fs/proc/array.c

index 220cc6376ef80e0c9bcfec162d45552e729cdf5a..0d07513a67a661440fc1599121bb1dc38560bc33 100644 (file)
@@ -177,7 +177,6 @@ read the file /proc/PID/status:
   CapBnd: ffffffffffffffff
   voluntary_ctxt_switches:        0
   nonvoluntary_ctxt_switches:     1
-  Stack usage:    12 kB
 
 This shows you nearly the same information you would get if you viewed it with
 the ps  command.  In  fact,  ps  uses  the  proc  file  system  to  obtain its
@@ -231,7 +230,6 @@ Table 1-2: Contents of the statm files (as of 2.6.30-rc7)
  Mems_allowed_list           Same as previous, but in "list format"
  voluntary_ctxt_switches     number of voluntary context switches
  nonvoluntary_ctxt_switches  number of non voluntary context switches
- Stack usage:                stack usage high water mark (round up to page size)
 ..............................................................................
 
 Table 1-3: Contents of the statm files (as of 2.6.8-rc3)
index f560325c444f0c10e6258ce7c5d9836c58d703b9..13b5d07081751e1cb42aca4ce491c9a52b70ab4f 100644 (file)
@@ -327,94 +327,6 @@ static inline void task_context_switch_counts(struct seq_file *m,
                        p->nivcsw);
 }
 
-#ifdef CONFIG_MMU
-
-struct stack_stats {
-       struct vm_area_struct *vma;
-       unsigned long   startpage;
-       unsigned long   usage;
-};
-
-static int stack_usage_pte_range(pmd_t *pmd, unsigned long addr,
-                               unsigned long end, struct mm_walk *walk)
-{
-       struct stack_stats *ss = walk->private;
-       struct vm_area_struct *vma = ss->vma;
-       pte_t *pte, ptent;
-       spinlock_t *ptl;
-       int ret = 0;
-
-       pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
-       for (; addr != end; pte++, addr += PAGE_SIZE) {
-               ptent = *pte;
-
-#ifdef CONFIG_STACK_GROWSUP
-               if (pte_present(ptent) || is_swap_pte(ptent))
-                       ss->usage = addr - ss->startpage + PAGE_SIZE;
-#else
-               if (pte_present(ptent) || is_swap_pte(ptent)) {
-                       ss->usage = ss->startpage - addr + PAGE_SIZE;
-                       pte++;
-                       ret = 1;
-                       break;
-               }
-#endif
-       }
-       pte_unmap_unlock(pte - 1, ptl);
-       cond_resched();
-       return ret;
-}
-
-static inline unsigned long get_stack_usage_in_bytes(struct vm_area_struct *vma,
-                               struct task_struct *task)
-{
-       struct stack_stats ss;
-       struct mm_walk stack_walk = {
-               .pmd_entry = stack_usage_pte_range,
-               .mm = vma->vm_mm,
-               .private = &ss,
-       };
-
-       if (!vma->vm_mm || is_vm_hugetlb_page(vma))
-               return 0;
-
-       ss.vma = vma;
-       ss.startpage = task->stack_start & PAGE_MASK;
-       ss.usage = 0;
-
-#ifdef CONFIG_STACK_GROWSUP
-       walk_page_range(KSTK_ESP(task) & PAGE_MASK, vma->vm_end,
-               &stack_walk);
-#else
-       walk_page_range(vma->vm_start, (KSTK_ESP(task) & PAGE_MASK) + PAGE_SIZE,
-               &stack_walk);
-#endif
-       return ss.usage;
-}
-
-static inline void task_show_stack_usage(struct seq_file *m,
-                                               struct task_struct *task)
-{
-       struct vm_area_struct   *vma;
-       struct mm_struct        *mm = get_task_mm(task);
-
-       if (mm) {
-               down_read(&mm->mmap_sem);
-               vma = find_vma(mm, task->stack_start);
-               if (vma)
-                       seq_printf(m, "Stack usage:\t%lu kB\n",
-                               get_stack_usage_in_bytes(vma, task) >> 10);
-
-               up_read(&mm->mmap_sem);
-               mmput(mm);
-       }
-}
-#else
-static void task_show_stack_usage(struct seq_file *m, struct task_struct *task)
-{
-}
-#endif         /* CONFIG_MMU */
-
 static void task_cpus_allowed(struct seq_file *m, struct task_struct *task)
 {
        seq_printf(m, "Cpus_allowed:\t");
@@ -445,7 +357,6 @@ int proc_pid_status(struct seq_file *m, struct pid_namespace *ns,
        task_show_regs(m, task);
 #endif
        task_context_switch_counts(m, task);
-       task_show_stack_usage(m, task);
        return 0;
 }