Re: [RFC][PATCH] /proc/<pid>/pmaps - memory maps in granularity of pages

[Fengguang Wu <wfg@mail.ustc.edu.cn>]

On Tue, Aug 14, 2007 at 11:26:18AM -0500, Matt Mackall wrote:
> On Tue, Aug 14, 2007 at 04:52:04PM +0800, Fengguang Wu wrote:
> > Hello,
> > 
> > Matt Mackall brings us many memory-footprint-optimization
> > opportunities with his pagemap/kpagemap patches. However I wonder if
> > the binary interface can be improved.
> > 
> > This seq_file based implementation iterates in the unit of vmas.  So
> > super large vmas can be a problem. The next step is to to do address
> > based iteration. That should not be hard.
> > 
> > Andrew, please stop me early if it is at all a wrong interface :)
> > 
> > Thank you,
> > Fengguang
> > ===
> > 
> > Show a process's memory maps page-by-page in /proc/<pid>/pmaps.
> > It helps to analyze application's memory footprint in a comprehensive way.
> > 
> > Pages share the same states are grouped into a page range.
> > For each page range, the following fields are exported:
> > 	- first page index
> > 	- number of pages in the range
> > 	- well known page/pte flags
> > 	- number of mmap users
> > 
> > Only page flags not expected to disappear in the near future are exported:
> > 
> > 	Y:young R:referenced A:active U:uptodate P:ptedirty D:dirty W:writeback
> > 
> > Here is a sample output:
> > 
> > # cat /proc/$$/pmaps
> > 08048000-080c9000 r-xp 08048000 00:00 0
> > 32840   129     Y_A_P__ 1
> > 080c9000-080f6000 rwxp 080c9000 00:00 0                                  [heap]
> > 32969   45      Y_A_P__ 1
> > f7dba000-f7dc3000 r-xp 00000000 03:00 176633                             /lib/libnss_files-2.3.6.so
> > 0       1       Y_AU___ 1
> > 1       1       YR_U___ 1
> > 5       1       YR_U___ 1
> > 8       1       Y_AU___ 1
> > f7dc3000-f7dc5000 rwxp 00008000 03:00 176633                             /lib/libnss_files-2.3.6.so
> > 8       2       Y_A_P__ 1
> > f7dc5000-f7dcd000 r-xp 00000000 03:00 176635                             /lib/libnss_nis-2.3.6.so
> > 0       1       Y_AU___ 1
> > 1       1       YR_U___ 1
> > 4       1       YR_U___ 1
> > 7       1       Y_AU___ 1
> > f7dcd000-f7dcf000 rwxp 00007000 03:00 176635                             /lib/libnss_nis-2.3.6.so
> > 7       2       Y_A_P__ 1
> > f7dcf000-f7de1000 r-xp 00000000 03:00 176630                             /lib/libnsl-2.3.6.so
> > 0       1       Y_AU___ 1
> > 1       3       YR_U___ 1
> > 16      1       YR_U___ 1
> > f7de1000-f7de3000 rwxp 00011000 03:00 176630                             /lib/libnsl-2.3.6.so
> > 17      2       Y_A_P__ 1
> > [...]
> 
> That's a _lot_ of data to generate and parse. I doubt we can watch a

Yes, but won't be too bad because it works in a sparse way.

1) It will only generate output for resident pages, that normally is
much smaller than the mapped size. Take my shell for example, the
(size:rss) ratio is (7:1)!

wfg ~% cat /proc/$$/smaps |grep Size|sum
sum      50552.000 
avg        777.723 

wfg ~% cat /proc/$$/smaps |grep Rss|sum 
sum       7604.000 
avg        116.985 

2) The page range trick suppresses more output. Look at these lines:

> > 08048000-080c9000 r-xp 08048000 00:00 0
> > 32840   129     Y_A_P__ 1
> > 080c9000-080f6000 rwxp 080c9000 00:00 0 [heap]
> > 32969   45      Y_A_P__ 1

It's interesting to see that the seq_file interface demands some
more programming efforts, and provides this flexibility as well.

3) A 64bit number can be encoded in hex in 8 bytes, no more than its
binary presentation. And often the text string will be much shorter
because of the common small values.

4) String formatting is cheap. Much cheaper than pte/struct page
walkings, and context switches. Not to mention its user friendliness.

> larger app in realtime with this interface. And it doesn't give us
> physical page numbers either.

This could be a problem. Binary interface is discouraging, which is
good and bad. The good thing about it is that we can safely export
the raw PFN/page flag numbers without upsetting upset normal users. In
this way a possible useful kernel debugging interface can be shipped
everywhere.

If kpagemap is a reasonable kernel debug interface, I do not see the
reason why kpagemap and pmaps cannot coexist :)