Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
To:     "linux-perf-use." <linux-perf-users@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Arnaldo Carvalho de Melo <acme@kernel.org>
Cc:     Heiko Carstens <heiko.carstens@de.ibm.com>,
        Sumanth Korikkar <sumanthk@linux.ibm.com>,
        Vasily Gorbik <gor@linux.ibm.com>
From:   Thomas Richter <tmricht@linux.ibm.com>
Subject: kernel/perf: Sample data being lost
Organization: IBM
Date:   Tue, 21 Apr 2020 17:54:29 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.6.0
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Message-Id: <f09e62d0-af40-683a-648f-3c3b7137369b@linux.ibm.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Since a couple of days I see this warning popping up very often:

[root@m35lp76 perf]# ./perf record --call-graph dwarf -e rb0000 -- find /
[ perf record: Woken up 282 times to write data ]
Warning:
Processed 16999 events and lost 382 chunks!

Check IO/CPU overload!

[ perf record: Captured and wrote 125.730 MB perf.data (16219 samples) ]
[root@m35lp76 perf]#

The machine is idle, its my development system, so not much going on.
It also happens using a software event, for example cycles. It shows
up more often, the larger the sample size is. So for example:

[root@m35lp76 perf]# pwd
/root/linux/tools/perf
[root@m35lp76 perf]#  ./perf record  --call-graph dwarf -- find
[ perf record: Woken up 2 times to write data ]
Warning:
Processed 231 events and lost 7 chunks!

Check IO/CPU overload!

[ perf record: Captured and wrote 1.000 MB perf.data (130 samples) ]
[root@m35lp76 perf]#

I have very seldom observed this before, only in extremely rare cases with
a heavily loaded machine. I am wondering what has changed, I haven't
changed anything in the s390 PMU device drivers.
It could be
 - common kernel code when writing into the ringbuffer.
 - the perf tool too slow to read data from the mapped buffer.
   However I have not come across changes in this area.

Has anybody observed similar issue?

PS: I have added some printk messages into my PMU devices drivers.
I have seen messages that the 16384 pages for auxilary buffers are full
and that samples have been dropped.


Thanks a lot.                                 
-- 
Thomas Richter, Dept 3252, IBM s390 Linux Development, Boeblingen, Germany
--
Vorsitzender des Aufsichtsrats: Matthias Hartmann
Geschäftsführung: Dirk Wittkopp
Sitz der Gesellschaft: Böblingen / Registergericht: Amtsgericht Stuttgart, HRB 243294