The standard placement new function is declared to not throw, which
means that, per spec, a null check on its result is required. There are
a number of places throughout xpcom/ where we know that we are passing
non-null pointers to placement new (and receiving them as a return
value), and we are therefore doing useless work performing these null
checks.
Therefore, we should be using an operator new overload that doesn't
require the null check. MFBT has just such an overload, so use that.
nsCycleCollector.cpp has three different structs named "Block", which makes it
hard to read. This patch renames them as EdgeBlock, NodeBlock, and PurpleBlock.
--HG--
extra : rebase_source : cae17ba925559b625f65a4741ae0bb4bee99d078
15% of our "small" OOM crashes are allocations of this struct. Halving its size
will hopefully help reduce that.
--HG--
extra : rebase_source : 5480a540a9584899def045c1401a5aa9a271d72c
The bulk of this commit was generated with a script, executed at the top
level of a typical source code checkout. The only non-machine-generated
part was modifying MFBT's moz.build to reflect the new naming.
CLOSED TREE makes big refactorings like this a piece of cake.
# The main substitution.
find . -name '*.cpp' -o -name '*.cc' -o -name '*.h' -o -name '*.mm' -o -name '*.idl'| \
xargs perl -p -i -e '
s/nsRefPtr\.h/RefPtr\.h/g; # handle includes
s/nsRefPtr ?</RefPtr</g; # handle declarations and variables
'
# Handle a special friend declaration in gfx/layers/AtomicRefCountedWithFinalize.h.
perl -p -i -e 's/::nsRefPtr;/::RefPtr;/' gfx/layers/AtomicRefCountedWithFinalize.h
# Handle nsRefPtr.h itself, a couple places that define constructors
# from nsRefPtr, and code generators specially. We do this here, rather
# than indiscriminantly s/nsRefPtr/RefPtr/, because that would rename
# things like nsRefPtrHashtable.
perl -p -i -e 's/nsRefPtr/RefPtr/g' \
mfbt/nsRefPtr.h \
xpcom/glue/nsCOMPtr.h \
xpcom/base/OwningNonNull.h \
ipc/ipdl/ipdl/lower.py \
ipc/ipdl/ipdl/builtin.py \
dom/bindings/Codegen.py \
python/lldbutils/lldbutils/utils.py
# In our indiscriminate substitution above, we renamed
# nsRefPtrGetterAddRefs, the class behind getter_AddRefs. Fix that up.
find . -name '*.cpp' -o -name '*.h' -o -name '*.idl' | \
xargs perl -p -i -e 's/nsRefPtrGetterAddRefs/RefPtrGetterAddRefs/g'
if [ -d .git ]; then
git mv mfbt/nsRefPtr.h mfbt/RefPtr.h
else
hg mv mfbt/nsRefPtr.h mfbt/RefPtr.h
fi
--HG--
rename : mfbt/nsRefPtr.h => mfbt/RefPtr.h
Jemalloc 4 purges dirty pages regularly during free() when the ratio of dirty
pages compared to active pages is higher than 1 << lg_dirty_mult. We set
lg_dirty_mult in jemalloc_config to limit RSS usage, but it also has an impact
on performance.
So instead of enforcing a high ratio to force more pages being purged, we keep
jemalloc's default ratio of 8, and force a regular purge of all dirty pages,
after cycle collection.
Keeping jemalloc's default ratio avoids cycle-collection-triggered purge to
have to go through really all dirty pages when there are a lot, in which case
the normal jemalloc purge during free() will already have kicked in. It also
takes care of everything that doesn't run the cycle collector still having
a level of purge, like plugins in the plugin-container.
At the same time, since jemalloc_purge_freed_pages does nothing with jemalloc 4,
repurpose the MEMORY_FREE_PURGED_PAGES_MS telemetry probe to track the time
spent in this cycle-collector-triggered purge.
This can cause leaks that are invisible to our XPCOM leak detection system.
To avoid this, classes should not addref or release in their Traverse methods.
--HG--
extra : rebase_source : acd0b070c63cbb4111c165d6b131b8e3b822773a
This makes it clearer that, unlike how SizeOf*() functions usually work, this
doesn't measure any children hanging off the array.
And do likewise for nsTObserverArray.
--HG--
extra : rebase_source : 6a8c8d8ffb53ad51b5773afea77126cdd767f149
This patch makes it so that while the cycle collector is running methods are called
on the concrete implementation nsCycleCollectorLogger, rather than the interface
nsICycleCollectorListener. This makes explicit the requirement that we have to be
very careful about what we call during the cycle collector, and should make it
possible for the GC rooting static analysis to understand what is happening.
The UUID of nsICycleCollectorHandler was changed to appease the UUID commit hook.
Various parts of the first half of BeginCollection() can start an incremental GC.
This is bad because running the GC and CC at the same time can cause the CC to end
up with pointers to dead JS objects.
To avoid this, we finish any incremental GC in progress in BeginCollection. This
is slow, but hopefully it is rare.
They are kept around for the sake of the standalone glue, which is used
for e.g. webapprt, which doesn't have direct access to jemalloc, and thus
still needs a wrapper to go through the xpcom function list and get to
jemalloc from there.
Crashing here is apparently fairly common. This restores the old behavior, so we at least
don't crash immediately, but instead enter a slow downward spiral of leaking.
This improves on the old behavior in that we only try and fail to grow the hash table once,
rather than on every add. khuey I think reported that the browser got very slow, because
you are going through the very slowest path of the allocator over and over.
It is a little cleaner to use this helper method if we only care about the CCJSRuntime pointer,
and it will let us move some of these methods out of this file more easily.
I kept all the existing PL_DHashTableAdd() calls fallible, in order to be
conservative, except for the ones in nsAtomTable.cpp which already were
followed immediately by an abort on failure.
--HG--
extra : rebase_source : 526d96ab65e4d7d71197b90d086d19fbdd79b7b5
I kept all the existing PL_DHashTableAdd() calls fallible, in order to be
conservative, except for the ones in nsAtomTable.cpp which already were
followed immediately by an abort on failure.
--HG--
extra : rebase_source : eeba14d732077ef2e412f4caca852de6b6b85f55
It feels safer to use a function with a new name, rather than just changing the
behaviour of the existing function.
For most of these cases the PL_DHashTableLookup() result was checked with
PL_DHASH_ENTRY_IS_{FREE,BUSY} so the conversion was easy. A few of them
preceded that check with a useless null check, but the intent of these was
still easy to determine.
I'll do the trickier ones in subsequent patches.
--HG--
extra : rebase_source : ab37a7a30be563861ded8631771181aacf054fd4
Currently the setting of PLDHashTable::ops is very haphazard.
- PLDHashTable has no constructor, so it's not auto-nulled, so lots of places
null it themselves.
- In the fallible PLDHashTable::Init() function, if the entry storage
allocation fails we'll be left with a table that has |ops| set -- indicating
it's been initialized -- but has null entry storage. I'm not certain this can
cause problems but it feels unsafe, and some (but not all) callers of Init()
null it on failure.
- PLDHashTable does not null |ops| in Finish(), so some (but not all) callers
do this themselves.
This patch makes things simpler.
- It adds a constructor that zeroes |ops|.
- It modifies Init() so that it only sets |ops| once success is ensured.
- It zeroes |ops| in Finish().
- Finally, it removes all the now-unnecessary |ops| nulling done by the users
of PLDHashTable.
--HG--
extra : rebase_source : bb34979c218d152562a2f9c7e5215256c111cc5b