Fw: [Bug 205579] New: rfcomm causes very high cpu usage during serial connectivity

Fw: [Bug 205579] New: rfcomm causes very high cpu usage during serial connectivity

[Stephen Hemminger]

Begin forwarded message:

Date: Tue, 19 Nov 2019 09:52:12 +0000
From: bugzilla-daemon@bugzilla.kernel.org
To: stephen@networkplumber.org
Subject: [Bug 205579] New: rfcomm causes very high cpu usage during serial connectivity


https://bugzilla.kernel.org/show_bug.cgi?id=205579

            Bug ID: 205579
           Summary: rfcomm causes very high cpu usage during serial
                    connectivity
           Product: Networking
           Version: 2.5
    Kernel Version: Everything
          Hardware: All
                OS: Linux
              Tree: Mainline
            Status: NEW
          Severity: normal
          Priority: P1
         Component: Other
          Assignee: stephen@networkplumber.org
          Reporter: epigramx@yahoo.com
        Regression: No

This appears to be a long standing issue that is mentioned only in old Debian
bug reports and current Raspbian discussions but it remains important for
embedded systems because they are not powerful systems.

Sources:

https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=565485
https://www.raspberrypi.org/forums/viewtopic.php?t=181796

Issue Description:

"I use 'rfcomm listen' in combination with 'gpspipe' to relay GPS data
from my eeePC to my Blackberry.  I noticed, however, that the CPU
overhead from 'rfcomm listen' is quite significant -- over 10% on my
eeePC 901.

Running 'strace' reveals that rfcomm is in a tight loop between
'waitpid' (with WNOHANG) and 'ppoll' (with a 200 nanosecond(!!)
timeout).  This is evidently the source of the extreme CPU usage.

If it's critical that rfcomm know the child has died ASAP, I would
expect it should be watching for the SIGCHLD signal rather than trying
to use waitpid().  If it's not all that critical, it probably doesn't
need to be watching for child death at a rate of five millions checks
per second. :)"

also

"I am experimenting with a serial connection to Pi Zero W using the on-board
Bluetooth module. I can successfully establish a connection and send data to
Pi. However as soon as the connection starts, the rfcomm process consumes
almost 50% of the processor (as top shows). This noticeably slows down other
processes."

The issue remains on current code.

-- 
You are receiving this mail because:
You are the assignee for the bug.