Fix ps command CPU usage on Apple Silicon M1 macs. #4142

[obaudys]

Description

The cpu user and system times returned by libproc are not in
nanoseconds; they are in mach ticks units. This is not documented very
well. The convert from mach ticks to ns, the kernel provides a timebase
info function and datatype:

(https://developer.apple.com/documentation/driverkit/3433733-mach_timebase_info)

The commit makes the PS command work for me.

Tests

Existing tests should cover this just fine. However it would be good to get confirmation from someone with an Intel mac to see if ps output makes sense.

[fdncred]

oh wow. it's so nice to have cpu utilization again. thanks so much for this. now, if we can only fix the virtual memory, why oh why is everything ~400-~500GB? it's not like that in activity monitor. I wonder if it's the same sort of thing?

[obaudys]

if we can only fix the virtual memory, why oh why is everything ~400-~500GB?

As far as I can see, the data for virtual memory size is populated by the C library function proc_pidinfo, which is a wrapper around a syscall __proc_info. So, I'll say it is accurate, and the native /bin/ps utility's "rss" field matches nushell's "mem":

Funnily enough my Google Chrome worker processes have 1.5TB mapped into their address space, apparently. Both nushell and /bin/ps agree on this.

Activity Monitor does not seem to report the the rss field unless you enable the "Real Mem" column, and I have no idea how it actually calculates the "Memory" column:

[fdncred]

@obaudys When I compared activity monitor to nushell's ps command earlier this morning. The virtual memory was way off, but maybe I was looking at the wrong column. All I saw was 95% (or greater) of my processes were using over 400GB of virtual memory. I'll take a closer look later this afternoon like you suggest. Thanks for the info.

[rgwood]

Thanks @obaudys! Really appreciate these kinds of platform-specific fixes :)