Performance/MemShrink/DMD: Difference between revisions

From MozillaWiki
Jump to navigation Jump to search
No edit summary
 
(48 intermediate revisions by 4 users not shown)
Line 1: Line 1:
DMD (short for "dark matter detector") is a tool for that tracks which heap blocks have been reported by memory reporters.  It's designed to help us reduce the "heap-unclassified" value in Firefox's about:memory page.  It also detects if any heap blocks are reported twice.
The contents of this page have moved to [https://developer.mozilla.org/en-US/docs/Mozilla/Performance/DMD MDN].
 
== Build ==
 
=== Everything other than B2G-device builds ===
 
If you're building on Linux or Android, add this line to the mozconfig file of your choice:
 
  ac_add_options --enable-dmd
 
If you're building on Mac, add these lines:
 
  ac_add_options --enable-debug
  ac_add_options --enable-dmd
 
''Note: non-debug DMD builds do not currently work on Mac. see {{bug|995443}}.''
 
If you're building on Windows, add these lines:
 
  ac_add_options --enable-dmd
  ac_add_options --enable-profiling
 
Build with that mozconfig.  Optimized builds should work fine.  DMD enables jemalloc, and does not work in combination with <code>--enable-trace-malloc</code>.
 
If you don't want to do the build yourself, you can get try server to do it for you.  For a desktop build, modify the appropriate in-tree mozconfig file before pushing.  E.g. for an optimized Linux64 build, modify <code>browser/config/mozconfigs/linux64/common-opt</code> before pushing.  On Windows you'll also need to add this line to build/mozconfig.common:
 
  MOZ_CRASHREPORTER_UPLOAD_FULL_SYMBOLS=1
 
(and note that below the "Run" instructions for Windows try builds differ from the instructions for local Windows builds).
 
''Note: If you build with clang earlier than r171782 on Linux, you may not get any stack traces;  see {{bug|826962}}.  As of Jan 16, 2013, r171782 is not part of any Clang release, so you have to build LLVM/clang from scratch to get it.  GCC works fine.''
 
=== B2G device builds ===
 
'''First, update your B2G checkout''' with <tt>git pull</tt> or <tt>git fetch && git merge origin/master</tt>.  <tt>./repo sync</tt> is not sufficient!  You must git pull to get the latest version of the relevant tools.
 
For B2G device builds, we don't usually modify the mozconfig (although you can; it's hiding under gonk-misc/default-gecko-config).
 
Instead, modify your .userconfig and add
 
  export MOZ_DMD=1
 
(don't forget the <tt>export</tt>).
 
You probably need to clobber your objdir (rm -rf objdir-gecko).  Then build normally.
 
== Run ==
 
=== Desktop ===
 
To run DMD on a desktop build (including Firefox and B2G desktop), you need to precede your usual invocation of Firefox with three environment variable definitions.
 
On Linux, from a bourne-style shell, do this:
 
  LD_PRELOAD=$OBJDIR/dist/lib/libdmd.so \
  LD_LIBRARY_PATH=$OBJDIR/dist/lib/ \
  DMD=1 \
  <command>
 
If you are feeling masochistic enough to want to run DMD under gdb, one way is to do this:
 
  $ gdb --args <command>
  (gdb) set exec-wrapper env LD_PRELOAD=[path_to_lib]/libdmd.so LD_LIBRARY_PATH=[path_to_lib]/ DMD=1
  (gdb) run
 
On Mac OS X, do this:
 
  DYLD_INSERT_LIBRARIES=$OBJDIR/dist/lib/libdmd.dylib \
  LD_LIBRARY_PATH=$OBJDIR/dist/lib/ \
  DMD=1 \
  <command>
 
On a local Windows build, do this:
 
  set MOZ_REPLACE_MALLOC_LIB=path\to\dmd.dll
  set DMD=1
  <command>
 
On a Windows build done by the try server, follow [https://bugzilla.mozilla.org/show_bug.cgi?id=936784#c69 these instructions] instead.
 
On start-up, you'll see some commentary on stderr, such as:
 
  DMD[20523] $DMD = '1'
  DMD[20523] DMD is enabled
 
The number in brackets is the process ID.
 
The browser will run a little slower than usual.
 
=== Fennec ===
 
To run DMD on Fennec, run the following commands (be sure to replace "org.mozilla.fennec" with the app identifier as appropriate; this will usually be org.mozilla.fennec_$USERNAME for a local build).
 
  adb push $OBJDIR/dist/lib/libdmd.so /sdcard/
  adb shell am start -n org.mozilla.fennec/.App --es \
      env0 MOZ_REPLACE_MALLOC_LIB=/sdcard/libdmd.so \
      --es env1 DMD=1
 
The commentary on Fennec goes to logcat, and looks like this:
 
  I/DMD  (27314): $DMD = '1'
  I/DMD  (27314): DMD is enabled
 
The number in the parentheses is the process ID.
 
=== B2G for devices ===
 
If you built B2G with <tt>export MOZ_DMD=1</tt>, your build will automagically run with DMD enabled.  (The <tt>b2g.sh</tt> script figures out whether to enable DMD by checking for the presence of </tt>libdmd.so</tt> in <tt>/system/b2g</tt>)
 
If DMD is enabled, you'll see a message in logcat when a process starts up:
 
  I/DMD    (  305): $DMD = '1'
  I/DMD    (  305): DMD is enabled
 
The <tt>run-gdb.sh</tt> script also knows to start DMD builds with DMD enabled, so you don't need to do anything special.
 
== Analyze ==
 
DMD doesn't do anything notable until you ask it to.
 
=== Desktop ===
 
If you are on a sufficiently recent build, visit about:memory and click the "DMD" button. (The button won't be present in non-DMD builds, and will be grayed out in DMD builds if DMD wasn't enabled at start-up.)
 
If you are on an older build, create a bookmark with the following location:
 
  javascript:DMDReportAndDump("out.dmd")
 
and click that instead. Note that this silently fails on certain special pages such as about:memory or an empty tab.
 
Both actions invoke all the memory reporters and then DMD analyzes the reports, printing this commentary:
 
  DMD: running reporters...
  DMD[11420] Dump 1 {
  DMD[11420]  gathering stack trace records...
  DMD[11420]  creating and sorting twice-reported stack trace record array...
  DMD[11420]  creating and sorting unreported stack trace record array...
  DMD[11420]  printing unreported stack trace record array...
  DMD[11420]  creating and sorting once-reported stack trace record array...
  DMD[11420]  printing once-reported stack trace record array...
  DMD[11420] }
 
The output be written to a file called <code>out.dmd</code> in the current working directory.
 
On Linux the analysis is fast.  On Mac it can take 30+ seconds.
 
=== Fennec ===
 
'''Due to [https://bugzilla.mozilla.org/show_bug.cgi?id=823354 bug 823354], you may get empty stack traces on Fennec, rendering DMD's output largely unhelpful.  Sorry.'''
 
On Fennec you can use the existing memory-report dumping hook to get a DMD report as well, assuming you have a DMD-enabled build. Run the following command:
 
  adb shell am broadcast -a org.mozilla.gecko.MEMORY_DUMP
 
In logcat, you should see output similar to this:
 
  E/GeckoConsole (27314): nsIMemoryInfoDumper dumped reports to /data/data/org.mozilla.fennec_kats/app_tmp/memory-report-default-27314.json.gz
 
The path (should always be /data/data/$APPID/app_tmp/) is where the memory reports and DMD reports get dumped to. You can pull them like so:
 
  adb pull /data/data/org.mozilla.fennec_kats/app_tmp/memory-report-default-27314.json.gz
  adb pull /data/data/org.mozilla.fennec_kats/app_tmp/dmd-default-27314.txt.gz
 
=== B2G ===
 
''Did you remember to <tt>git pull</tt> or <tt>git fetch && git merge</tt> as instructed in the B2G building section?  If you skip this step, the tools below may be outdated.  <tt>./repo sync</tt> is not sufficient.''
 
Run <tt>tools/get_about_memory.py</tt>.  If DMD is enabled on your device, you should see output like the following:
 
    $ ./get_about_memory.py
    Got 3/3 files.
    Pulled files into about-memory-18.
    Got 3 DMD dump(s).
    [...]
    Done processing DMD files.  Have a look in about-memory-18.
 
<tt>get_about_memory.py</tt> invokes <tt>fix_b2g_stack.py</tt>, so you shouldn't need to run it yourself, but it's there in case you need it.  It works just like <tt>fix_linux_stack.pl</tt> on desktop.
 
See <tt>get_about_memory.py --help</tt> for more options, but you probably don't need anything other than the defaults.
 
== The output ==
 
=== Pre-processing ===
 
''Note: You can skip this step if you're on Windows or a B2G device build.''
 
DMD's output file contains a lot of stack traces.  As printed, many of the stack trace entries will look like this:
 
  ???[/home/njn/moz/mi2/dmdo64/dist/bin/libxul.so +0x1761BCD] 0x7f845186bbcd
 
To make them more useful, you need to run them through a "stack-fixing" script. 
* On Linux, use <tt>tools/rb/fix-linux-stack.pl</tt>.
* On Mac, use <tt>tools/rb/fix_macosx_stack.py</tt>.
Both scripts read from stdin and print to stdout.  After doing so, these lines should look something more like this:
 
  nsStringBuffer::Alloc(unsigned long) (/home/njn/moz/mi2/xpcom/string/src/nsSubstring.cpp:177) 0x7f845186bbcd
 
This shows the function, filename, line number, and PC.
 
Note that <tt>fix-linux-stack.pl</tt> is very slow, and can take 2+ minutes to process a DMD output file.  <tt>fix_macosx_stack.py</tt> is faster, but can still take 30+ seconds.
 
=== Output Sections ===
 
DMD's output is broken into multiple sections.
 
# "Invocation".  This tells you how DMD was invoked, i.e. what options were used.
# "Twice-reported stack trace records".  This tells you which heap blocks were reported twice or more.  The presence of any such records indicates bugs in one or more memory reporters.
# "Unreported stack trace records".  This tells you which heap blocks were not reported, which indicate where additional memory reporters would be most helpful.
# "Once-reported stack trace records": like the "Unreported stack trace records" section, but for blocks reported once.
# "Summary": gives measurements of the total heap, and the unreported/once-reported/twice-reported portions of it.
# "Execution measurements": gives some statistics about DMD's execution, which are mostly of interest to DMD's developers.
 
The "Twice-reported stack trace records" and "Unreported stack trace records" sections are the most important, because they indicate ways in which the memory reporters can be improved.
 
=== Stack trace records ===
 
The stack trace record sections are the most important ones.  Here's an example stack trace record from the "Unreported stack trace records" section.
 
Unreported: 3 blocks in stack trace record 209 of 1,891
  36,864 bytes (26,184 requested / 10,680 slop)
  0.03% of the heap (64.55% cumulative);  0.04% of unreported (86.78% cumulative)
  Allocated at
    malloc (/home/njn/moz/mi2/memory/build/replace_malloc.c:151) 0x417170
    PR_Malloc (/home/njn/moz/mi2/nsprpub/pr/src/malloc/prmem.c:435) 0x7f68650f423c
    PL_ArenaAllocate (/home/njn/moz/mi2/nsprpub/lib/ds/plarena.c:200) 0x7f68652463e1
    nsFixedSizeAllocator::Alloc(unsigned long) (/home/njn/moz/mi2/xpcom/ds/nsFixedSizeAllocator.cpp:95) 0x7f6860f528dc
    nsNodeInfo::Create(nsIAtom*, nsIAtom*, int, unsigned short, nsIAtom*, nsNodeInfoManager*) (/home/njn/moz/mi2/content/base/src/nsNodeInfo.cpp:64) 0x7f685f640933
    nsNodeInfoManager::GetNodeInfo(nsIAtom*, nsIAtom*, int, unsigned short, nsIAtom*) (/home/njn/moz/mi2/content/base/src/nsNodeInfoManager.cpp:225) 0x7f685f642d05
    mozilla::dom::Element::SetAttrAndNotify(int, nsIAtom*, nsIAtom*, nsAttrValue const&, nsAttrValue&, unsigned char, bool, bool, bool) (/home/njn/moz/mi2/content/base/src/Element.cpp:1862) 0x7f685f60ad87
    mozilla::dom::Element::SetAttr(int, nsIAtom*, nsIAtom*, nsAString_internal const&, bool) (/home/njn/moz/mi2/content/base/src/Element.cpp:1778) 0x7f685f60a9b3
    nsXMLContentSink::AddAttributes(unsigned short const**, nsIContent*) (/home/njn/moz/mi2/content/xml/document/src/nsXMLContentSink.cpp:1464) 0x7f685fa76c5c
    nsXBLContentSink::AddAttributes(unsigned short const**, nsIContent*) (/home/njn/moz/mi2/content/xbl/src/nsXBLContentSink.cpp:882) 0x7f685fb3ad42
    nsXMLContentSink::HandleStartElement(unsigned short const*, unsigned short const**, unsigned int, int, unsigned int, bool) (/home/njn/moz/mi2/content/xml/document/src/nsXMLContentSink.cpp:1018) 0x7f685fa73db5
    nsXMLContentSink::HandleStartElement(unsigned short const*, unsigned short const**, unsigned int, int, unsigned int) (/home/njn/moz/mi2/content/xml/document/src/nsXMLContentSink.cpp:947) 0x7f685fa7370a
    nsXBLContentSink::HandleStartElement(unsigned short const*, unsigned short const**, unsigned int, int, unsigned int) (/home/njn/moz/mi2/content/xbl/src/nsXBLContentSink.cpp:258) 0x7f685fb37cc0
 
It tells you that there were 3 heap blocks that were allocated from the program point indicated by the "Allocated at" stack trace, that these blocks took up 36,864 bytes, and that 10,680 of those bytes were "slop" (wasted space caused by the heap allocator rounding up request sizes).  It also indicates what percentage of the total heap size and the unreported portion of the heap these blocks represent.
 
Within each section, records are listed from largest to smallest.
 
Once-reported and twice-reported stack trace records also have stack traces for the report point(s).  For example:
 
Reported at
  mozilla::dmd::Report(void const*) (/home/njn/moz/mi2/memory/replace/dmd/DMD.cpp:1740) 0x7f68652581ca
  CycleCollectorMallocSizeOf(void const*) (/home/njn/moz/mi2/xpcom/base/nsCycleCollector.cpp:3008) 0x7f6860fdfe02
  nsPurpleBuffer::SizeOfExcludingThis(unsigned long (*)(void const*)) const (/home/njn/moz/mi2/xpcom/base/nsCycleCollector.cpp:933) 0x7f6860fdb7af
  nsCycleCollector::SizeOfIncludingThis(unsigned long (*)(void const*), unsigned long*, unsigned long*, unsigned long*, unsigned long*, unsigned long*) const (/home/njn/moz/mi2/xpcom/base/nsCycleCollector.cpp:3029) 0x7f6860fdb6b1
  CycleCollectorMultiReporter::CollectReports(nsIMemoryMultiReporterCallback*, nsISupports*) (/home/njn/moz/mi2/xpcom/base/nsCycleCollector.cpp:3075) 0x7f6860fde432
  nsMemoryInfoDumper::DumpMemoryReportsToFileImpl(nsAString_internal const&) (/home/njn/moz/mi2/xpcom/base/nsMemoryInfoDumper.cpp:626) 0x7f6860fece79
  nsMemoryInfoDumper::DumpMemoryReportsToFile(nsAString_internal const&, bool, bool) (/home/njn/moz/mi2/xpcom/base/nsMemoryInfoDumper.cpp:344) 0x7f6860febaf9
  mozilla::(anonymous namespace)::DumpMemoryReportsRunnable::Run() (/home/njn/moz/mi2/xpcom/base/nsMemoryInfoDumper.cpp:58) 0x7f6860fefe03
 
You can tell which memory reporter made the report by the name of the MallocSizeOf function near the top of the stack trace.  In this case it was the cycle collector's reporter.
 
By default, DMD measures heap blocks above a certain size precisely, but uses sampling to measure blocks below that size.  Any measurements that involve sampled blocks (even if combined with non-sampled measurements) are approximate, and this is indicated by a preceding '~'.  For example:
 
Unreported: ~273 blocks in block group 17 of 14,611
  ~1,125,590 bytes (~1,117,936 requested / ~7,654 slop)
  0.07% of the heap (2.58% cumulative);  0.43% of unreported (16.36% cumulative)
 
The sampling threshold can be adjusted with an option (see below).  This will affect the precision of the output and the speed at which Firefox+DMD runs.
 
== Options ==
 
Setting the <tt>DMD</tt> environment variable to <tt>1</tt> gives default options.  But you can also specify non-default options by setting <tt>DMD</tt> to a whitespace separated list of <tt>--option=val</tt> entries.
 
At the moment, you can provide two options to DMD: <tt>--sample-below=<1..n></tt> and <tt>--mode=<normal|test|stress></tt>.
 
=== --sample-below=<1..n> ===
 
By default, DMD ''samples'' blocks with a sample-below size of 4093.  I.e. it ignores some small allocations in order to run (much) faster.
 
If you want DMD to record all allocations precisely, pass <tt>--sample-below=1</tt>.  Otherwise, you should probably leave it unchanged.  If you do pick a different value, prime numbers work best.
 
=== --mode=<normal|test|stress> ===
 
<tt>--mode=<normal|test|stress></tt> can be used to invoke "test" or "stress" mode, which are useful if you're hacking on DMD.  The default is normal mode.
 
"test" and "stress" modes set their own <tt>--sample-below</tt> values, so you
should never have to specify both <tt>--sample-below</tt> and <tt>--mode</tt>.
 
To run the tests, specify <tt>--mode=test</tt> and start Firefox. It will print out some stuff and very quickly exit. Then run the following command from the top of your source directory.
 
  memory/replace/dmd/check_test_output.py . test.dmd
 
(If you invoked Firefox from a different directory to your source directory, you might need to specify the path to <tt>test.dmd</tt>.)
 
This script checks the output produced by the previous step, and will indicate if the test passed or failed. It should work on Linux and Mac, but is unreliable on Windows.
 
=== Setting options on B2G device builds ===
 
If you want to run B2G on a device with args other than DMD=1, you'll need to modify the <tt>gonk-misc/b2g.sh</tt> script and then push it to the device.
 
To push the modified script, do something like
 
  adb shell stop b2g
  adb remount
  adb push b2g.sh /system/bin
  adb shell chmod 0755 /system/bin/b2g.sh
  adb shell start b2g
 
If you want to run B2G on the device under GDB with args other than DMD=1, modify the run-gdb.sh script.  You don't need to push anything.
 
== Which heap blocks are reported? ==
 
At this stage you might wonder how DMD knows which allocations have been reported and which haven't.  DMD only knows about heap blocks that are measured via a function created with one of the following two macros:
 
  NS_MEMORY_REPORTER_MALLOC_SIZEOF_FUN
  NS_MEMORY_REPORTER_MALLOC_SIZEOF_ON_ALLOC_FUN
 
Fortunately, most of the existing memory reporters do this.  See [[Platform/Memory_Reporting]] for more details about how memory reporters are written.
 
== Troubleshooting DMD ==
 
Contact Nick Nethercote ("njn" on IRC) or Nathan Froyd ("froydnj" on IRC).

Latest revision as of 05:10, 21 October 2014

The contents of this page have moved to MDN.