Allocation of PIDs on Tru64 5.1

[Kevin Eady_1]

Does anybody know how Tru64 allocates the next PID number ?

The reason that I am asking, is that I have noticed that the PIDs are not always greater than the last one used, as on most other O/S and I think this is causing me grief when I store these in an internal table.

Is there some sort of re-allocating of PIDs going on ?

Is there a kernel parameter to control this ?

Is this a feature/problem with 5.1 only ?

Any info would be appreciated.

Kevin

[Greg Yates]

Hi Kevin,

I did a quick look and found a couple of things.

PIDs do get reused but they (pid_entry's) should get tacked to the end of the free pid list when the process exits.

This certainly could produce out-of-order PIDs upon new process creation. I'm not aware of a tuneable to change this behavior.

Here are a couple of data structures you can look at on your system to see more of what's going on:

(dbx) p pid_data
struct {
pd_npids = 0x4000
pd_ntabs = 0x1
pd_nents = 0x4000
pd_ebits = 0xe
pd_pidtab = 0xfffffc0000ce05c0
pd_ptbases = (nil)
pd_freelock = 0x0
pd_freelist = 0xfffffc0000d31f80
pd_taillink = 0xfffffc0000d327d0
pd_usecount = 0x5a
pd_expwait = 0x0
pd_expbusy = (nil)
}
(dbx) p *pid_data.pd_freelist
struct {
pe_pid = 0x6a934
pe_gen = 0x1a
pe_ibn = '^N'
pe_flag = '^@'
pe_proc = (nil)
pe_link = 0xfffffc0000d32940
pe_lock = 0x0
}


Hope this helps.

Greg

[Phillip Brown]

Yes, pid entries are reused. And there's an algorithm we use to ensure the same PIDs are not reused too quickly.

They're assigned distinct PIDs by adding 2pe_ibn for each new generation. (i.e. newpid = pid + pe_gen * 2 pe_ibn ).

We verify whether the entry in the pid table belongs to "our PID" by checking the pe_pid value (in pidtab array) once the entry is located.

The algorithm to locate your pidtab by PID is:

(kdbx) p PID & (pid_data.pd_nents -1)

That will give you the index into pidtab - p pidtab[index]. You should print it in hex so you can read the pe_ibn index. This will also give you the address to the proc structure where all things related may be found.

My pe_ibn is 0xc which would be 2^12, which equals 4096. So, we add that to our specific PID and compare that to the pe_pid to determine whether this slot belongs to this PID, or if the slot is being reused.

I am aware that certain code will scan the pidtab and if we find a slot that isn't in use (I believe we determine this by determining if the pd_proc field is null) then it's a candidate for re-use.

Regards,

Phil

[Kevin Eady_1]

Thanks for the info.
I'm going to change the way that I process my internal table.
Just for the record, a log file that I was checking seemed to indicate that I had a re-used PID 35 seconds after the original.