Gitiles
Code Review Sign In
gerrit.openbmc.org / openbmc / hiomapd / 0d087f17f868587ca01259632ba0bd584b66aae4
commit0d087f17f868587ca01259632ba0bd584b66aae4[log] [tgz]
authorAndrew Jeffery <andrew@aj.id.au>Fri Mar 23 11:40:35 2018 +1030
committerAndrew Jeffery <andrew@aj.id.au>Wed Apr 04 07:56:23 2018 +0000
treec71ac35940f16ed5d2d52d0abe14abcb19ba3035
parent912c9bdf938893e41c675f6cd97b7b00bcc83940 [diff]
mboxd_windows: Reset evicted windows

After copying a portion of the backing store to a window,
create_map_window() "resizes" the window to the aligned-up size reported
by copy_flash(). This allows use of the window size as the content size
elsewhere in the codebase.

However, if we needed to evict a window to satisfy a request, the window
properties were not reset. This lead to inefficient use of the reserved
memory by limiting the effective window size to the minimum size of all
requests that were previously allocated the window in question.

Inefficient use of reserved memory isn't the only side effect; the host
takes an eye-watering hit to throughput that gets exponentionally
worse over time:

From the petitboot shell without the patch applied:

    / # time cat /dev/mtd0 > /dev/null
    real    0m 49.77s
    user    0m 0.00s
    sys     0m 49.76s
    / # time cat /dev/mtd0 > /dev/null
    real    1m 33.57s
    user    0m 0.00s
    sys     1m 33.55s
    / # time cat /dev/mtd0 > /dev/null
    real    4m 45.37s
    user    0m 0.00s
    sys     4m 45.35s
    / # time cat /dev/mtd0 > /dev/null
    real    9m 17.77s
    user    0m 0.00s
    sys     9m 17.76s
    / #

And with the patch applied:

    / # time cat /dev/mtd0 > /dev/null
    real    0m 43.00s
    user    0m 0.00s
    sys     0m 42.99s
    / # time cat /dev/mtd0 > /dev/null
    real    0m 42.40s
    user    0m 0.00s
    sys     0m 42.39s
    / # time cat /dev/mtd0 > /dev/null
    real    0m 42.41s
    user    0m 0.00s
    sys     0m 42.39s
    / #

Reset the properties to allow use of the entire reserved memory region
allocated to the window, improving memory efficiency, throughput, and
minimising throughput variance.

Change-Id: I7be78ec5e0a9ee0caf31133b0861e333844b8975
Signed-off-by: Andrew Jeffery <andrew@aj.id.au>
2 files changed
tree: c71ac35940f16ed5d2d52d0abe14abcb19ba3035
  1. Documentation/
  2. m4/
  3. test/
  4. xyz/
  5. .clang-format-c
  6. .clang-format-c++
  7. .gitignore
  8. bootstrap.sh
  9. common.c
  10. common.h
  11. configure.ac
  12. dbus.h
  13. format-code.sh
  14. LICENSE
  15. Makefile.am
  16. mbox.h
  17. mboxctl.c
  18. mboxd.c
  19. mboxd_dbus.c
  20. mboxd_dbus.h
  21. mboxd_flash.h
  22. mboxd_flash_physical.c
  23. mboxd_flash_virtual.cpp
  24. mboxd_lpc.c
  25. mboxd_lpc.h
  26. mboxd_lpc_physical.c
  27. mboxd_lpc_virtual.cpp
  28. mboxd_msg.c
  29. mboxd_msg.h
  30. mboxd_pnor_partition_table.cpp
  31. mboxd_pnor_partition_table.h
  32. mboxd_windows.c
  33. mboxd_windows.h
  34. mtd.c
  35. pnor_partition.cpp
  36. pnor_partition.hpp
  37. pnor_partition_defs.h
  38. pnor_partition_table.cpp
  39. pnor_partition_table.hpp
  40. README.md
README.md

Copyright 2017 IBM

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

MBOX

This repo contains the protocol definition for the host to BMC mailbox communication specification which can be found in Documentation/mbox_procotol.md.

There is also a reference implementation of a BMC mailbox daemon, the details of which can be found in Documentation/mboxd.md.

Finally there is also an implementation of a mailbox daemon control program, the details of which can be found in Documentation/mboxctl.md.

Style Guide

Preamble

This codebase is a mix of C (due to its heritage) and C++. This is an ugly split: message logging and error handling can be vastly different inside the same codebase. The aim is to remove the split one way or the other over time and have consistent approaches to solving problems.

phosphor-mboxd is developed as part of the OpenBMC project, which also leads to integration of frameworks such as phosphor-logging. Specifically on phosphor-logging, it's noted that without care we can achieve absurd duplication or irritating splits in where errors are reported, as the C code is not capable of making use of the interfaces provided.

Rules

  1. Message logging MUST be done to stdout or stderr, and MUST NOT be done directly via journal APIs or wrappers of the journal APIs.

    Rationale:

    We have two scenarios where we care about output, with the important restriction that the method must be consistent between C and C++:

    1. Running in-context on an OpenBMC-based system
    2. Running the test suite

    In the first case it is desirable that the messages appear in the system journal. To this end, systemd will by default capture stdout and stderr of the launched binary and redirect it to the journal.

    In the second case it is desirable that messages be captured by the test runner (make check) for test failure analysis, and it is undesirable for messages to appear in the system journal (as these are tests, not issues affecting the health of the system they are being executed on).

    Therefore direct calls to the journal MUST be avoided for the purpose of message logging.

    Note: This section specifically targets the use of phosphor-logging's log<T>(). It does not prevent the use of elog<T>().