Page Space Problems

[Jeff Peek]

In article <1991Mar17.185553.9181 at menudo.uh.edu> jet at karazm.math.uh.edu ("J. Eric Townsend") writes:
>Again, I repeat my plea for some IBM employee to come out of the woodwork
>and explain this:
>
>oiler> ps aux
>    USER   PID %CPU %MEM    SZ   RSS     TT STAT  TIME CMD
>    root     0 0.0%  78%  7592  6376      -    S  1:31 swapper
>    root     1 0.1%   1%   164   120      -    S  2:41 /etc/init 
>    root   514 0.0%   0%    12     8      -    R 3994:08 kproc
>    root   771 0.0%   0%    20    20      -    S  0:05 kproc
>    root  1028 0.1%   0%    16    16      -    S  3:45 kproc
>    root  1285 0.0%   0%    16     8      -    S  0:00 kproc
>    root  1685 0.0%   0%    84    28      -    S  0:18 /etc/syncd 60 
>     jet  1867 2.9%   5%   168   404  pts/0    S  0:01 -csh 
>....
[my previous posting about what SZ meant for process 0 was incorrect.]
OK. You wanted it you got it. I did some mucking around in the code and 
actual experimentation on the machine and here is what I found. 
When you run ps -lk (the way I used to get the sizes for all the kprocs)
you get:
    F S  UID   PID  PPID   C PRI NI ADDR    SZ    WCHAN    TTY  TIME CMD
  b03 S    0     0     0 120  16 -- 1008 11008               -  4:50 swapper
  303 R    0   514     0 103 127 -- 1209    28               - 9332:43 kproc
  303 S    0   771     0   0  37 --  a05    36    20cd0      -  4:00 kproc
  303 S    0  1028     0   1  39 -- 160b    36    20cd8      - 10:20 kproc
  303 S    0  1285     0   0  36 -- 180c    32    20ce0      -  0:00 kproc
40201 S    0  1645     1   0  60 20 3058    32               -  0:00 kproc
40a01 S    0  2641     1   0  60 20  e27    32               -  0:00 kproc
40b03 S    0  4016     1   0  39 --  8a4    32  1bbd844      -  3:08 kproc
  a01 S    7  7856  4527   0  60 20  fc7    32               -  0:00 kproc
Here is a description of what is in the SZ field:
  SZ (-l, l)    The size in 1024 byte units of the core image of the
    		process.
So, SZ is in 1K chunks. Then it appears my swapper is close to 11meg.

The nitty gritty:
The SZ is based on the number of paging space blocks (4K) associated with
the TEXT and DATA segments for a process. Note, this is NOT real memory, just
the number of page spaces blocks used by the process. The DATA segment includes
user data, user stack, kernel stack, and the ublock.

Now, the segments register values used for the calculation are taken from the
u_adspace structure from the process's ublock. These srvals amount to the user's
address space in USER MODE! 

The swapper is a special kproc, handcrafted at System Initialization. The u_adspace
structure for the swapper is never initialized, i.e., it is all zeroes. On the 
RISC System/6000 an invalid segment register value is 7fffff. '0' is a valid
srval and in fact is the srval for the kernel's text segment. This segment is 
only addressable to processes in kernel mode. So, ps is counting the kernel's TEXT
segment TWICE for process 0. The reason this is not a problem for other kernel processes
is that their u_adspace structure is inherited from their parent when they are
created, thus initialized properly. 

Only process 0 has no parent (or is it it's own mother and father?).

The strange but true facts about the swapper.

>--
>J. Eric Townsend - jet at uh.edu - bitnet: jet at UHOU - vox: (713) 749-2120
>Skate UNIX or bleed, boyo...
>(UNIX is a trademark of Bell Laboratories).


Jeff Peek
AIX Operating System Architecture -- IBM Personal Systems Programiming        
ransom at perelandra.austin.ibm.com 	VNET PEEK at AUSVMQ T/L 793-3935
Austin, TX