From: Reinette Chatre <reinette.chatre@intel.com>
To: "Moger, Babu" <bmoger@amd.com>, Babu Moger <babu.moger@amd.com>,
"Borislav Petkov" <bp@alien8.de>
Cc: <tony.luck@intel.com>, <x86@kernel.org>, <Dave.Martin@arm.com>,
<james.morse@arm.com>, <corbet@lwn.net>,
<skhan@linuxfoundation.org>, <tglx@kernel.org>,
<mingo@redhat.com>, <dave.hansen@linux.intel.com>,
<hpa@zytor.com>, <linux-kernel@vger.kernel.org>,
<linux-doc@vger.kernel.org>, <eranian@google.com>,
<peternewman@google.com>
Subject: Re: [PATCH 1/2] x86/resctrl, Documentation: Keep mbm_assign_mode "default" on boot
Date: Mon, 27 Jul 2026 17:29:53 -0700 [thread overview]
Message-ID: <77f77d02-fae7-401d-9bb5-c62b244d23cd@intel.com> (raw)
In-Reply-To: <386b441c-286a-41e5-b878-50a6debb7571@amd.com>
Hi Babu,
On 7/27/26 4:29 PM, Moger, Babu wrote:
> I still have to get back to few questions(test details etc..) on this thread. Let me get back to on that later tomorrow.
> thanks
Don't bother.
I just installed a distro version of pqos utility and modified resctrl to return "Unavailable"
once in a while on an Intel system.
Below is the output on an idle system that confirms the pqos team's analysis that
it perceives counter wraparound when there is none. This is when I run with something like
"pqos -m all:0-143" to show all counts. When I run with "pqos -m all:[0-143]" to provide an
average I see the number repeatedly jump to the large number and back over time.
In summary I see:
- pqos utility does not support all scenarios of memory bandwidth measurement
on AMD systems without ABMC active. pqos may intermittently report very large
and incorrect bandwidth numbers.
- pqos utility reports 0MB/s when there are unassigned counters on AMD systems
that support ABMC (and it is enabled).
As I mused earlier (https://lore.kernel.org/lkml/119188c4-1127-4155-9a05-dff47e25be85@intel.com/),
I do not know (and can thus only speculate) why the issue with "Unavailable" handling has not
been reported before.
If you are expecting more pqos usage on AMD systems then users may encounter output like
below more and more until the fix is created and propagated to users. (sidenote: You may
want to participate in the fix since it may involve exposing the text state, including
"Unassigned", to users where currently it is 0.)
Considering this, could you please confirm that you consider default mode the best
choice? After all, "it will make things keep working as it used to", which sometimes
will look something like:
CORE IPC MISSES LLC[KB] MBL[MB/s] MBR[MB/s]
0 1.77 16k 0.0 17592186044410.4 0.0
1 2.74 9k 0.0 2.0 0.0
2 0.75 0k 0.0 0.0 0.0
3 0.71 0k 0.0 0.0 0.0
4 0.71 0k 0.0 0.0 0.1
5 0.71 0k 0.0 17592186044415.9 0.0
6 1.59 0k 0.0 0.0 0.0
7 0.72 0k 0.0 0.0 0.0
8 1.20 1k 0.0 0.0 0.0
9 2.48 887k 0.0 0.0 90.6
10 0.73 0k 0.0 17592186044415.4 0.0
11 0.68 0k 0.0 0.0 0.0
12 0.53 0k 0.0 0.0 0.0
13 0.71 0k 0.0 0.0 0.0
14 0.72 1k 0.0 0.0 0.1
15 0.70 0k 0.0 17592186044415.9 0.0
16 0.76 0k 0.0 0.0 0.0
17 0.72 0k 0.0 0.0 0.0
18 2.48 116k 3456.0 0.0 0.0
19 0.66 0k 0.0 0.0 0.1
20 0.64 0k 0.0 0.0 0.0
21 0.65 0k 0.0 0.1 0.0
22 0.41 0k 0.0 0.0 0.0
23 2.49 88k 6408.0 0.0 17592186044313.9
24 0.96 1k 0.0 0.0 0.1
25 0.03 0k 0.0 0.0 0.0
26 0.72 0k 0.0 0.4 0.0
27 0.71 0k 0.0 0.0 0.0
28 0.67 0k 72.0 0.0 0.0
29 0.79 0k 0.0 0.0 0.0
30 0.79 1k 0.0 17592186044415.9 0.0
31 0.71 0k 0.0 0.3 0.0
32 0.67 0k 0.0 0.0 0.0
33 0.72 0k 0.0 0.0 0.0
34 0.72 0k 0.0 0.0 0.0
35 0.71 0k 0.0 0.0 0.0
36 0.83 28k 0.0 2.0 0.0
37 0.86 0k 0.0 0.0 0.0
38 0.62 0k 0.0 0.0 0.0
39 0.67 0k 0.0 0.0 0.0
40 0.57 0k 0.0 0.0 0.0
41 0.87 0k 0.0 0.0 0.0
42 0.74 0k 0.0 0.0 0.0
43 0.79 0k 0.0 0.0 0.0
44 0.73 0k 0.0 0.0 0.0
45 0.86 0k 0.0 0.0 0.0
46 0.93 2k 0.0 0.0 0.0^C
Reinette
next prev parent reply other threads:[~2026-07-28 0:30 UTC|newest]
Thread overview: 23+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-07-17 21:13 [PATCH 1/2] x86/resctrl, Documentation: Keep mbm_assign_mode "default" on boot Babu Moger
2026-07-17 21:13 ` [PATCH 2/2] x86/resctrl: Fix ABMC counter programming for extended counter ranges Babu Moger
2026-07-17 22:56 ` [PATCH 1/2] x86/resctrl, Documentation: Keep mbm_assign_mode "default" on boot Reinette Chatre
2026-07-20 17:00 ` Babu Moger
2026-07-20 18:27 ` Reinette Chatre
2026-07-20 19:15 ` Babu Moger
2026-07-21 20:54 ` Reinette Chatre
2026-07-23 18:03 ` Babu Moger
2026-07-23 21:46 ` Reinette Chatre
2026-07-24 20:57 ` Moger, Babu
2026-07-24 22:28 ` Reinette Chatre
2026-07-24 23:12 ` Borislav Petkov
2026-07-24 23:53 ` Reinette Chatre
2026-07-27 14:05 ` Borislav Petkov
2026-07-27 15:25 ` Reinette Chatre
2026-07-27 17:24 ` Babu Moger
2026-07-27 18:12 ` Reinette Chatre
2026-07-27 20:17 ` Babu Moger
2026-07-27 21:42 ` Reinette Chatre
2026-07-27 22:44 ` Moger, Babu
2026-07-27 23:29 ` Moger, Babu
2026-07-28 0:29 ` Reinette Chatre [this message]
2026-07-20 20:12 ` Babu Moger
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=77f77d02-fae7-401d-9bb5-c62b244d23cd@intel.com \
--to=reinette.chatre@intel.com \
--cc=Dave.Martin@arm.com \
--cc=babu.moger@amd.com \
--cc=bmoger@amd.com \
--cc=bp@alien8.de \
--cc=corbet@lwn.net \
--cc=dave.hansen@linux.intel.com \
--cc=eranian@google.com \
--cc=hpa@zytor.com \
--cc=james.morse@arm.com \
--cc=linux-doc@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@redhat.com \
--cc=peternewman@google.com \
--cc=skhan@linuxfoundation.org \
--cc=tglx@kernel.org \
--cc=tony.luck@intel.com \
--cc=x86@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