#702 mach (darwin) support for CPU utilization

[Bill Fenner]

tanders asked me to take a look at reporting CPU stats on Darwin. I've come up with the attached. It collects stats for all CPUs and reports them, and adds them all up to report in the CPU #-1. It doesn't support the following values:
- kern_ticks, intrpt_ticks, wait_ticks, sirq_ticks, sys2_ticks
since I couldn't find mach interfaces to report these. (I used sys_ticks instead of sys2_ticks as other modules use because that way the CPUSYSTEM percentage gets reported properly).

The paging stats are reported in swap*, so pageIn and pageOut are unused. It should be the other way around, but the other modules work this way so I followed them. I also couldn't find mach interfaces for interrupts or context switches (Well, I found a per-process context switch value, but I didn't want to try to fetch them all and add them up) so nInterrupts and nCtxSwitches are also unused.

[Bill Fenner]

[Bill Fenner]

hardware/cpu.h diff

[Dave Shield]

Logged In: YES
user_id=88893
Originator: NO

Question:
Why do you believe that other implementations use
'swap{In,Out}' for reporting paging statistics?

For example, from cpu_sysctl.c:
#define NS_VM_PAGEIN pageins
#define NS_VM_PAGEOUT pdpageouts
#define NS_VM_SWAPIN swapins
#define NS_VM_SWAPOUT swapouts

cpu->swapIn       = \(unsigned long\)mem\_stats.NS\_VM\_SWAPIN;
cpu->swapOut      = \(unsigned long\)mem\_stats.NS\_VM\_SWAPOUT;
cpu->pageIn       = \(unsigned long\)mem\_stats.NS\_VM\_PAGEIN;
cpu->pageOut      = \(unsigned long\)mem\_stats.NS\_VM\_PAGEOUT;

[Bill Fenner]

Logged In: YES
user_id=109593
Originator: YES

Now that I revisit it, I don't know where I got that idea; perhaps from cpu_nlist's

cpu->swapIn  = \(unsigned long\)mem\_stats.v\_swappgsin+mem\_stats.v\_vnodepgsin;
cpu->swapOut = \(unsigned long\)mem\_stats.v\_swappgsout+mem\_stats.v\_vnodepgsout

since I was pretty sure that at least the vnode stats are paging.

I'd be happy to see cpu_mach doing whatever is most appropriate. If you'd like me to submit a new version, that's no problem, or of course it's easy enough to fix my source before adding it to the net-snmp repository.

[Bill Fenner]

Logged In: YES
user_id=109593
Originator: YES

I was also confused because I wasn't expecting the paging stats to be reported under ssIORawSent and ssIORawReceived, and thought that maybe the reasoning for reporting using swapping was because at least those stats were exposed in the MIB.

I've uploaded a new cpu_mach.c that uses the paging values.

[Bill Fenner]

[Bill Fenner]

Logged In: YES
user_id=109593
Originator: YES

I see that this has been labelled too late for 5.4. I submitted it a few hours after tanders asked me to work on it. That's quite an obnoxious release engineering process you have.

[Thomas Anders]

Logged In: YES
user_id=848638
Originator: NO

I'm sorry if you feel discouraged, but the fact it didn't make it into 5.4 doesn't say anything about the usefulness of your patch. It will likely make it into 5.4.1 and 5.5 and will be easily available as a patch on top of 5.4 for any interested OSX user. Again: thanks for working on it.

[Thomas Anders]

Logged In: YES
user_id=848638
Originator: NO

Thanks for the patch! It has been applied to the 5.4.x code branches and the main development tree, and will appear in future releases of the Net-SNMP package.

[Troy]

Logged In: YES
user_id=1652863
Originator: NO

This patch breaks the make on

Darwin Kernel Version 8.8.0: Fri Sep 8 17:18:57 PDT 2006; root:xnu-792.12.6.obj~1/RELEASE_PPC Power Macintosh powerpc

Using the "configure" from the README.osx

./configure --with-persistent-directory=/var/db/net-snmp --with-defaults --without-rpm --with-mib-modules="host ucd-snmp/diskio" --with-out-mib-modules="mibII/icmp host/hr_swrun" --disable-static --without-kmem-usage

make fails with:
gcc -I../include -I. -I../agent -I../agent/mibgroup -I../snmplib -g -O2 -Udarwin8 -Ddarwin8=darwin8 -c mibgroup/host/hr_network.c -fno-common -DPIC -o mibgroup/host/.libs/hr_network.o
mibgroup/host/hr_network.c: In function 'Save_HR_Network_Info':
mibgroup/host/hr_network.c:250: error: invalid use of undefined type 'struct ifnet'
mibgroup/host/hr_network.c:251: error: invalid use of undefined type 'struct ifnet'
mibgroup/host/hr_network.c:251: error: invalid use of undefined type 'struct ifnet'
make[1]: *** [mibgroup/host/hr_network.lo] Error 1
make: *** [subdirs] Error 1

[Thomas Anders]

Logged In: YES
user_id=848638
Originator: NO

Please post a separate bug for this with all the details (including what patches you added on top of 5.4).