Re: [PATCH] Export mlock information via smaps

[Nikanth Karthikesan <knikanth@suse.de>]

On Wednesday 18 August 2010 11:22:53 Balbir Singh wrote:
> * Nikanth Karthikesan <knikanth@suse.de> [2010-08-18 10:23:41]:
> > On Tuesday 17 August 2010 21:55:36 Matt Mackall wrote:
> > > On Tue, 2010-08-17 at 10:39 +0530, Nikanth Karthikesan wrote:
> > > > Currently there is no way to find whether a process has locked its
> > > > pages in memory or not. And which of the memory regions are locked in
> > > > memory.
> > > >
> > > > Add a new field to perms field 'l' to export this information. The
> > > > information exported via maps file is not changed.
> > >
> > > I'm worried that your new 'l' flag will fatally surprise some naive
> > > parser of this file.
> >
> > So how to proceed? Create another "ssmaps" file or something? :) Or is
> > the following patch any better? Even the "Yes"/"No" could be changed to
> > "0 kB" or "x kB", if it would make it better.
> >
> > Thanks
> > Nikanth
> >
> > Currently there is no way to find whether a process has locked its pages
> > in memory or not. And which of the memory regions are locked in memory.
> >
> > Add a new field "Locked" to export this information via smaps file.
> >
> > Signed-off-by: Nikanth Karthikesan <knikanth@suse.de>
> >
> > ---
> >
> > diff --git a/Documentation/filesystems/proc.txt
> > b/Documentation/filesystems/proc.txt index a6aca87..6eafd26 100644
> > --- a/Documentation/filesystems/proc.txt
> > +++ b/Documentation/filesystems/proc.txt
> > @@ -373,6 +373,7 @@ Referenced:          892 kB
> >  Swap:                  0 kB
> >  KernelPageSize:        4 kB
> >  MMUPageSize:           4 kB
> > +Locked:         No
> >
> >  The first  of these lines shows  the same information  as is displayed
> > for the mapping in /proc/PID/maps.  The remaining lines show  the size of
> > the mapping, @@ -397,6 +398,8 @@ To clear the bits for the file mapped
> > pages associated with the process
> >
> >      > echo 3 > /proc/PID/clear_refs
> >
> >  Any other value written to /proc/PID/clear_refs will have no effect.
> >
> > +The "Locked" indicates whether the mapping is locked in memory or not.
> > +
> >
> >  1.2 Kernel data
> >  ---------------
> > diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
> > index aea1d3f..7bafbcf 100644
> > --- a/fs/proc/task_mmu.c
> > +++ b/fs/proc/task_mmu.c
> > @@ -405,7 +405,8 @@ static int show_smap(struct seq_file *m, void *v)
> >  		   "Referenced:     %8lu kB\n"
> >  		   "Swap:           %8lu kB\n"
> >  		   "KernelPageSize: %8lu kB\n"
> > -		   "MMUPageSize:    %8lu kB\n",
> > +		   "MMUPageSize:    %8lu kB\n"
> > +		   "Locked:         %s\n",
> >  		   (vma->vm_end - vma->vm_start) >> 10,
> >  		   mss.resident >> 10,
> >  		   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
> > @@ -416,7 +417,8 @@ static int show_smap(struct seq_file *m, void *v)
> >  		   mss.referenced >> 10,
> >  		   mss.swap >> 10,
> >  		   vma_kernel_pagesize(vma) >> 10,
> > -		   vma_mmu_pagesize(vma) >> 10);
> > +		   vma_mmu_pagesize(vma) >> 10,
> > +		   (vma->vm_flags & VM_LOCKED) ? "Yes" : "No");
> 
> Why not show the Locked as kB as well? I know that the entire VMA is
> locked, but ideally if we can show mss.pss as locked, one can write
> a simple script to accumulate locked memory for the process. NOTE:
> One could choose RSS or PSS, but I'd prefer PSS (even though the
> value is not stable across the system) since it is more accurate
> representation of the truly locked memory.
> 

Agreed. Attached patch does this, i.e, uses PSS as amount of locked memory.

Matt, will this be okay?

Thanks
Nikanth

Currently there is no way to find whether a process has locked its pages in
memory or not. And which of the memory regions are locked in memory.

Add a new field "Locked" to export this information via smaps file.

Signed-off-by: Nikanth Karthikesan <knikanth@suse.de>

---

diff --git a/Documentation/filesystems/proc.txt b/Documentation/filesystems/proc.txt
index a6aca87..17b0ae0 100644
--- a/Documentation/filesystems/proc.txt
+++ b/Documentation/filesystems/proc.txt
@@ -373,6 +373,7 @@ Referenced:          892 kB
 Swap:                  0 kB
 KernelPageSize:        4 kB
 MMUPageSize:           4 kB
+Locked:              374 kB
 
 The first  of these lines shows  the same information  as is displayed for the
 mapping in /proc/PID/maps.  The remaining lines show  the size of the mapping,
@@ -397,6 +398,8 @@ To clear the bits for the file mapped pages associated with the process
     > echo 3 > /proc/PID/clear_refs
 Any other value written to /proc/PID/clear_refs will have no effect.
 
+The "Locked" indicates whether the mapping is locked in memory or not.
+
 
 1.2 Kernel data
 ---------------
diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
index aea1d3f..58e586c 100644
--- a/fs/proc/task_mmu.c
+++ b/fs/proc/task_mmu.c
@@ -405,7 +405,8 @@ static int show_smap(struct seq_file *m, void *v)
 		   "Referenced:     %8lu kB\n"
 		   "Swap:           %8lu kB\n"
 		   "KernelPageSize: %8lu kB\n"
-		   "MMUPageSize:    %8lu kB\n",
+		   "MMUPageSize:    %8lu kB\n"
+		   "Locked:         %8lu kB\n",
 		   (vma->vm_end - vma->vm_start) >> 10,
 		   mss.resident >> 10,
 		   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
@@ -416,7 +417,9 @@ static int show_smap(struct seq_file *m, void *v)
 		   mss.referenced >> 10,
 		   mss.swap >> 10,
 		   vma_kernel_pagesize(vma) >> 10,
-		   vma_mmu_pagesize(vma) >> 10);
+		   vma_mmu_pagesize(vma) >> 10,
+		   (vma->vm_flags & VM_LOCKED) ?
+			(unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 
 	if (m->count < m->size)  /* vma is copied successfully */
 		m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;