On Mon, 1 Apr 2002, Randy.Dunlap wrote:
> Of course the basic answer is something like
> Try cscope
I can't find this -- does it come with Red Hat??
> or
> cd /usr/src/linux
> grep -r "kstat.p" * | more
This worked.
>
> Using the latter one:
>
> | 1. Why are kstat.pgpgin and kstat.pgpgout shifted right / halved?
>
> I had wondered that also, so you made me look.
>
> pgpgin and pgpgout are maintained (counted) in units of 512-byte
> blocks but displayed in /proc/stat in 1 KB (KiB :) blocks by shifting
> right by 1.
Yes ... that I managed to figure out. Seems strange that pgpgin/out are
in units of kilobytes and pswpin/out are in units of pages, though. Just
another little hole that needs plugging in the documentation.
> | 2. Are the "page" and "swap" numbers mutually exclusive? That is, if a
> | page is brought in from swap and counted in kstat.pswpin, is it also
> | counted in kstat.pgpgin? I found the places in the code where the counts
> | are incremented, but I couldn't tell if the swapin routine calls the
> | block driver or not.
> No, "page" and "swap" counts are not mutually exclusive.
> Both paths call submit_bh().
>
> For swap:
>
> mm/page_io.c::rw_swap_page_base():
> /* reads or writes a swap page */
> kstat.pswpin++;
> kstat.pswpout++;
> calls brw_page();
> which calls submit_bh();
>
> For all low-level block I/O:
>
> drivers/block/ll_rw_blk.c::ll_rw_block():
> /* low-level access to block devices */
> calls submit_bh();
>
> drivers/block/ll_rw_blk.c::submit_bh():
> count := is number of 512-byte blocks;
> kstat.pgpgout += count;
> kstat.pgpgin += count;
> calls generic_make_request();
Ah ... so every page to/from swap is counted in pswpin/out (as a page)
and again in pgpgin/out as half-kilobytes :-). Incidentally, I also
followed the third fork in this road, the one which derives the "blocks"
read and written per device that show up in /proc/stat. Those "blocks"
turn out to be 512 bytes long.
All of this is quite useful, actually, because I'm trying to make sense
of some high-frequency (every 15 seconds) samples of I/O data. Thanks!!
-- M. Edward Borasky znmeb@borasky-research.netThe COUGAR Project http://www.borasky-research.com/Cougar.htm
How to Stop A Folksinger Cold # 4 "Tie me kangaroo down, sport..." Tie your own kangaroo down -- and stop calling me "sport"!
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sun Apr 07 2002 - 22:00:09 EST