Re: port-hppa/56849: Wacko kernel memory accounting in current/hppa

[Tom Lane <tgl%sss.pgh.pa.us@localhost>]

The following reply was made to PR port-hppa/56849; it has been noted by GNATS.

From: Tom Lane <tgl%sss.pgh.pa.us@localhost>
To: gnats-bugs%netbsd.org@localhost, Nick Hudson <nick.hudson%gmx.co.uk@localhost>
Cc: port-hppa-maintainer%netbsd.org@localhost, gnats-admin%netbsd.org@localhost,
        netbsd-bugs%netbsd.org@localhost
Subject: Re: port-hppa/56849: Wacko kernel memory accounting in current/hppa
Date: Tue, 24 May 2022 16:55:11 -0400

 Nick Hudson <nick.hudson%gmx.co.uk@localhost> writes:
 >  Candidate fix...
 
 I can confirm that with this patch installed, top and ps report a
 reasonably stable kernel size, ranging from ~80M just after boot
 to ~95M at peak load.  This is a bit more than I was expecting, but
 it's probably fine given that I kicked some things up after hitting
 ENFILE errors:
 
 $ cat /etc/sysctl.conf
 ...
 # tgl additions
 kern.ipc.semmni=3D100
 kern.ipc.semmns=3D1000
 kern.maxvnodes=3D50000
 kern.maxproc=3D1000
 kern.maxfiles=3D5000
 
 
 I poked around with some of the other monitoring tools mentioned in
 the NetBSD Guide, and I noticed that "sysstat bufcache" is a tad
 confused about metadata buffers:
 
                     /0   /1   /2   /3   /4   /5   /6   /7   /8   /9   /10
      Load Average   |||||||||||||||||||||||||||||||||||||||||||||||
 
      2144 metadata buffers using             98681168592896 kBytes of memo=
 ry (19777529200%).
     55451 pages for cached file data using   221804 kBytes of memory (44%)=
 .
     15502 pages for executables using         62008 kBytes of memory (12%)=
 .
     35216 pages for anon (non-file) data     140864 kBytes of memory (28%)=
 .
       495 free pages                           1980 kBytes of memory ( 0%)=
 .
 
 File System          Bufs used   %   kB in use   %  Bufsize kB   %  Util %
 /home                     1227  57       12852   0       12852   0     100
 /usr                       524  24        4460   0        4498   0      99
 /var                        36   1         226   0         226   0     100
 /                           19   0         186   0         186   0     100
 
 Total:                    1806  84       17724   0       17762   0     100
 
 
 error reading kmem for vaddr at 0xda030 (Bad address)
 
 
 That error message at the bottom of the screen might or might not
 be related to the silly stat.  I'm not sure to what extent these
 numbers should be expected to match what "top" shows, but they
 don't seem crazy on their own with the exception of that one.
 
 			regards, tom lane