From: Stephane Eranian <eranian@frankl.hpl.hp.com>
To: linux-ia64@vger.kernel.org
Subject: [Linux-ia64] More information on perfmon in 2.4.10
Date: Wed, 26 Sep 2001 06:48:32 +0000 [thread overview]
Message-ID: <marc-linux-ia64-105590698805253@msgid-missing> (raw)
To all interested,
> ftp://ftp.kernel.org/pub/linux/kernel/ports/ia64/ in file:
>
> linux-2.4.10-ia64-010924.diff*
>
> The major changes this time include:
>
> - update perfmon to v0.3 (Stephane Eranian)
I would like to add some comments to the updates of perfmon.
What is in 2.4.10 now includes the following new features:
- optimized PMU context switch for per process monitoring.
This kernel now incoporates a lazy management scheme.
At this point it support only UP running systems.
This covers SMP kernels running in UP mode (nosmp).
- PFM_DESTROY_CONTEXT: a new command to destroy a perfmon context.
This allow a process to exec() another one that want to use
perfmon as well.
- Beginning of system-wide monitoring support. This kernel contains
a code that I would consider more as a 'proof of concept' rather
than final code. For now support is limited to UP running kernels
(this covers SMP kernels running UP). There can only be one such
session for the entire system and it is mutually exclusive with
per-process sessions. You can create a system-wide session
by using the PFM_FL_SYSTEM_WIDE when creating the context.
- user level notification via the SIGPROF signal is now sent in the
perfmon overflow handler and not when the task blocks (if blocking
mode is selected). The allows for better code overlapping on SMP systems.
Bug fixes:
This new kernel includes the important patches sent earlier to the
list that covered the following bugs:
- The counts reported by the API are incorrect if the
monitored process was context-switched during its execution.
More or less the counts would be double of what they should be.
- The content of PMD14-PMD17 was not saved and restored on
context-switch due to an error in the perfmon initialization code,
where the number of PMDs and PMCs reported by PAL got swapped.
Here are the new feature that did not make it in time for this release:
- support for address range restrictions (IBR/DBR)
- lazy PMU context switch in SMP mode
Important upcoming changes:
- perfmonctl() will loose the flag argument as it is not incorporated
into the argument of the command.
- Given that the use of siginfo to convey detailed information about the
overflows mandates the use of a fixed signal number, the option
of choosing the signal to receive overflow notification will be
abandoned. You indicate you want notification by simply providing
a valid process ID and you will receive a SIGPROF when necessary.
I will release a new version of pfmon (pfmon-0.07) once those changes make
it to the official kernel as this version of pfmon already includes the above
changes and therefore will not work with 2.4.10 as it is today.
If you have further questions about pfmon and/or perfmon, please contact me.
--
-Stephane
reply other threads:[~2001-09-26 6:48 UTC|newest]
Thread overview: [no followups] expand[flat|nested] mbox.gz Atom feed
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=marc-linux-ia64-105590698805253@msgid-missing \
--to=eranian@frankl.hpl.hp.com \
--cc=linux-ia64@vger.kernel.org \
/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