Linux MIPS Architecture development
 help / color / mirror / Atom feed
From: Ralf Baechle <ralf@linux-mips.org>
To: Philip Mucci <mucci@cs.utk.edu>
Cc: Linux MIPS <linux-mips@linux-mips.org>
Subject: Re: /dev/cpuid or /proc/cpuinfo
Date: Wed, 1 Feb 2006 11:30:07 +0000	[thread overview]
Message-ID: <20060201113006.GA3562@linux-mips.org> (raw)
In-Reply-To: <1138647269.4077.11.camel@localhost.localdomain>

On Mon, Jan 30, 2006 at 06:54:29PM +0000, Philip Mucci wrote:

> In reference to the performance counting thread we had going earlier,
> I've noticed a 'feature' I need out of MIPS/Linux that isn't currently
> available. This has also recently come up on the oprofile list with one
> of the oprofile/mips tools not being able to grab cpu Mhz
> from /proc/cpuinfo because it's not there.

Surprise - the kernel doesn't actually have that information.  The closest
available would be the counting speed of the c0 count register which - if
something like that is available at all - could be calibrated against an
external counter to figure out the clock speed.  To complicate matters
not all processors have have a c0 counter, it's sometimes incrementing
every cycle but it might just as well counting at half the clock rate and
oh yes, more recently there is clock scaling.

> I have need to execute the mfc0 instruction on the config register and
> grok the results to find out things like cache size etc. In addition, it
> might be nice to also actually be able to find out the clock rate.
> (Currently I grab BogoMIPS and punt.)

Considering above complexities a BogoMIPS-based approach doesn't sound too
bad - but it has other problem to implement since it requires knowing the
exact execution timing.  A think that could be done is meassuring two
loops containing a different number of nops and use that to compute the
execution time for the loop closing branch.  Now knowing that based on
knowing the # of cycles per loop we can compute the clock speed - all it
takes is a timer running at a known speed.

A funny complexity is that some multiprocessor MIPS systems have processors
running at different speeds.  So knowing the clock rate doesn't make life
too much easier ;-)

> On the intel and PPC systems, I believe you can execute similar
> instructions from user mode which makes things easy. However, of course
> an MFC0 is a privileged instruction...meaning that if the value or
> values aren't found in /proc/cpuinfo, I'm s.o.l.

We have a long tradition of emulating instructions, could use that to
permit access to a bunch of cp0 registers from userland ;)

> What does the list think about this? Making a mips /dev/cpuid is a bit
> gross but extending and grokking /proc/cpuinfo is perhaps grosser...and
> many tools do just this (like PAPI and oprofile's opreport...)
> 
> Comments? I'm certainly willing to implement this, but I'd rather 'do it
> right the first time' rather than get rotten vegetables thrown my way.

/proc/cpuinfo is meant to have per processor information so I don't have
a problem with adding cache configuration information.  I was considering
to do so to make such information available for bug reports.

  Ralf

      reply	other threads:[~2006-02-02 11:51 UTC|newest]

Thread overview: 8+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2006-01-16 15:36 Timer Interrupt Bharathi Subramanian
2006-01-17  7:30 ` Bharathi Subramanian
2006-01-17 14:13   ` Bharathi Subramanian
2006-01-17 14:52     ` Kevin D. Kissell
2006-01-17 15:28       ` Bharathi Subramanian
2006-01-27  8:32       ` Bharathi Subramanian
2006-01-30 18:54         ` /dev/cpuid or /proc/cpuinfo Philip Mucci
2006-02-01 11:30           ` Ralf Baechle [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20060201113006.GA3562@linux-mips.org \
    --to=ralf@linux-mips.org \
    --cc=linux-mips@linux-mips.org \
    --cc=mucci@cs.utk.edu \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox