tree: 75ac631c4f418dfcef2dfd54ea56fff06a2bc23a [path history] [tgz]
  1. memdf/
  2. platform/
  3. .pylintrc
  4. block.py
  5. collect.py
  6. diffsyms.py
  7. gaps.py
  8. gh_db_load.py
  9. gh_db_query.py
  10. gh_report.py
  11. gh_sizes.py
  12. gh_sizes_environment.py
  13. README-GitHub-CI.md
  14. README.md
  15. report_summary.py
  16. report_tree.py
scripts/tools/memory/README.md

Memory usage scripts

Scripts to collect, aggregate, and report memory usage.

Common options

The following options are common to most of the scripts, where applicable:

optional arguments:

  • -h, --help Show this help message and exit.
  • --verbose, -v Show informational messages; repeat for debugging messages.
  • --config-file FILE Read configuration FILE. Typically this is a file from the platform/ subdirectory providing platform-specific option defaults. Command line options override the configuration file.

input options:

  • --collect-method METHOD, -f METHOD Method of input processing.

    This specifies how the input files are read. The available METHODs are:

    • elftools — use the python elftools library
    • readelf — use the external readelf program
    • bloaty — use the external bloaty program
    • csv — read a comma-separated table
    • tsv — read a tab-separated table
    • su — read .su stack usage reports

    Not all methods are usable for all scripts. Usually readelf is fastest.

  • --collect-prefix PATH, --prefix PATH, --strip-prefix PATH Strip PATH from the beginning of source file names.

output options:

  • --output-file FILENAME, --output FILENAME, -O FILENAME Output file. Defaults to standard output. For csv and tsv formats, this is actually an output file name prefix.
  • --output-format FORMAT, --to FORMAT, -t FORMAT Output format. One of:
    • text — Plain text tables, in a single file.
    • csv — Comma-separated tables (in several files, if not stdout).
    • tsv — Tab-separated tables (in several files, if not stdout).
    • json_split — JSON - see Pandas documentation for details.
    • json_records — JSON - see Pandas documentation for details.
    • json_index — JSON - see Pandas documentation for details.
    • json_columns — JSON - see Pandas documentation for details.
    • json_values — JSON - see Pandas documentation for details.
    • json_table — JSON - see Pandas documentation for details.
    • Any format provided by tabulate.
  • --report-limit BYTES, --limit BYTES Limit display to items above the given size. Suffixes (e.g. K) are accepted.
  • --report-by GROUP, --by GROUP Reporting group. One of:
    • region — Aggregate by region. A ‘region’ is a platform-defined memory area consisting of a number of segments. Commonly platforms define the regions FLASH and RAM.
    • section — Aggregate by section.
    • symbol — Aggregate by symbol name.

selection options:

  • --section-select NAME, --section NAME Section(s) to process; otherwise all not ignored. Note: platform configuration files typically define a default list of sections.
  • --section-select-all Select all sections.
  • --section-ignore NAME Section(s) to ignore.
  • --section-ignore-all Ignore all sections unless explicitly selected.
  • --symbol-select NAME, --symbol NAME Symbol(s) to process; otherwise all not ignored.
  • --symbol-select-all Select all symbols.
  • --symbol-ignore NAME Symbol(s) to ignore.
  • --symbol-ignore-all Ignore all symbols unless explicitly selected.
  • --region-select NAME, --region NAME Region(s) to process; otherwise all not ignored.
  • --region-select-all Select all regions.
  • --region-ignore NAME Region(s) to ignore.
  • --region-ignore-all Ignore all regions unless explicitly selected.

external tool options:

  • --tool-bloaty FILE File name of the bloaty executable.
  • --tool-nm FILE File name of the nm executable.
  • --tool-readelf FILE File name of the readelf executable.

Scripts

report_summary.py

Report the total size of each region or section (per --report-by).

Example:

$ report_summary.py --by=region --config-file=${PLATFORM}.cfg ${IMAGE}
   region   size
*unknown*    200
FLASH     524285
RAM       165501

report_tree.py

Present a tree-structured report of memory use.

For this script, --limit (or a per-section limit) is useful.

Example:

$ report_tree.py --demangle --by=region --region=RAM:8K \
    --prefix=${CHIP_TOOLS} --prefix=${PWD} \
    --config-file=${PLATFORM}.cfg \
    ${IMAGE}

REGION: RAM
100% 62540 *total*
├── 32% 19896 WS_vlatest
├── 28% 17671 out
│   └── 100% 17671 release
│       └── 100% 17671 ..
│           └── 100% 17671 ..
│               ├── 96% 16911 third_party
│               │   ├── 100% 16866 lwip
│               │   │   └── 100% 16866 repo
│               │   │       └── 100% 16866 lwip
│               │   │           └── 100% 16866 src
│               │   └──  0% 45 *other*
│               └──  4% 760 *other*
└── 40% 24973 *other*

gaps.py

Report parts of an image that are not defined as part of any symbol. Typically these are string constants or other anonymous data.

Note: currently this only works with the elftools reader and consequently only works on ELF files.

Example:


$ gaps.py --section=.text --limit=1K --config-file=${PLATFORM}.cfg ${IMAGE} 04065992 length 5482 in section .text of ${IMAGE} 04065990: 54 43 50 00 55 44 50 00 47 6F 74 20 6D 6F TCP.UDP.Got mo 040659A0: 72 65 20 41 43 4B 65 64 20 62 79 74 65 73 20 28 re ACKed bytes ( 040659B0: 25 64 29 20 74 68 61 6E 20 77 65 72 65 20 70 65 %d) than were pe 040659C0: 6E 64 69 6E 67 20 28 25 64 29 00 47 6F 74 20 41 nding (%d).Got A 040659D0: 43 4B 20 66 6F 72 20 25 64 20 62 79 74 65 73 20 CK for %d bytes ...

diffsyms.py

Reports differences in size (and/or presence) of individual symbols between two files. Generally this only makes sense between different versions of the same thing, e.g. between a build on a working branch vs master.

Example:

$ diffsyms.py --demangle ${IMAGE1} ${IMAGE2}
                                                                                         symbol   a   b
chip::Inet::InetLayer::NewUDPEndPoint(chip::Inet::UDPEndPoint**)                                196 194
chip::Transport::BLE::Init(chip::DevicePairingDelegate*, chip::RendezvousParameters const&)  80 100

block.py

Report symbol references found in a block list. Generally useful only on library (.a) or object files.

Example:


$ block.py --config-file=blocklist.cfg out/release/\${PLATFORM}/obj/src/transport/lib/libTransportLayer.a address kind symbol cu 0 U strcpy TransportMgrBase.cpp

collect.py

Read memory use and write it in another form.

Useful for example to capture symbols from an image and write them to file(s) in some useful format (e.g. csv, json) for further processing.