building for / running on Power Linux (ppc64)

[Bill Buros]

Had an issue with building/running wkhtmlpdf on Power Linux.

A segfault was reported to us...   with a workaround identified

> > I've reproduced at least one SIGSEGV:
> >
> > #0  0x0000000010167a3c in .WTF::OSAllocator::reserveUncommitted(unsigned long, WTF::OSAllocator::Usage, bool, bool) ()
> > #1  0x000000001160e734 in .WTF::PageAllocationAligned::allocate(unsigned long, unsigned long, WTF::OSAllocator::Usage, bool, bool) ()
> > #2  0x000000001144b4fc in .JSC::MarkedBlock::create(JSC::JSGlobalData*, unsigned long) ()
> > #3  0x0000000010da9680 in .JSC::MarkedSpace::allocateBlock(JSC::MarkedSpace::SizeClass&) ()
> > #4  0x0000000010da9ac0 in .JSC::MarkedSpace::allocateFromSizeClass(JSC::MarkedSpace::SizeClass&) ()
> > #5  0x0000000010de71e8 in .JSC::JSGlobalData::JSGlobalData(JSC::JSGlobalData::GlobalDataType, JSC::ThreadStackType) ()
> > #6  0x0000000010de8cec in .JSC::JSGlobalData::createLeaked(JSC::ThreadStackType) ()
> > #7  0x0000000010197aec in .WebCore::JSDOMWindowBase::commonJSGlobalData() ()
> > #8  0x00000000101a320c in .WebCore::ScriptController::getAllWorlds(WTF::Vector<WebCore::DOMWrapperWorld*, 0ul>&) ()
> > #9  0x000000001053f2b8 in .WebCore::FrameLoader::dispatchDidClearWindowObjectsInAllWorlds() ()
> > #10 0x0000000010540e24 in .WebCore::FrameLoader::receivedFirstData() ()
> > #11 0x0000000010537c40 in .WebCore::DocumentWriter::setEncoding(WTF::String const&, bool) ()
> > #12 0x000000001052fef4 in .WebCore::DocumentLoader::commitData(char const*, int)---Type <return> to continue, or q <return> to quit--- 
> >  ()
> > #13 0x0000000010117650 in .WebCore::FrameLoaderClientQt::committedLoad(WebCore::DocumentLoader*, char const*, int) ()
> > #14 0x0000000010530eb8 in .WebCore::DocumentLoader::commitLoad(char const*, int) ()
> > #15 0x00000000105645bc in .WebCore::MainResourceLoader::addData(char const*, int, bool) ()
> > #16 0x000000001057a9dc in .WebCore::ResourceLoader::didReceiveData(char const*, int, long long, bool) ()
> > #17 0x0000000010566ce4 in .WebCore::MainResourceLoader::didReceiveData(char const*, int, long long, bool) ()
> > #18 0x0000000010578364 in .WebCore::ResourceLoader::didReceiveData(WebCore::ResourceHandle*, char const*, int, int) ()
> > #19 0x000000001088e104 in .WebCore::QNetworkReplyHandler::forwardData() ()
> > #20 0x0000000010891210 in .WebCore::QNetworkReplyHandlerCallQueue::flush() [clone .part.47] ()
> > #21 0x0000000010892758 in .WebCore::QNetworkReplyWrapper::emitMetaDataChanged()
> >     ()
> > #22 0x0000000010892c9c in .WebCore::QNetworkReplyWrapper::receiveMetaData() ()
> > #23 0x0000000010893364 in .WebCore::QNetworkReplyWrapper::qt_static_metacall(QObject*, QMetaObject::Call, int, void**) ()
> > #24 0x00000000122b6e0c in .QMetaObject::activate(QObject*, QMetaObject const*, int, void**) ()
> > ---Type <return> to continue, or q <return> to quit---
> > #25 0x000000001230f5cc in .QIODevice::qt_static_metacall(QObject*, QMetaObject::Call, int, void**) ()
> > #26 0x00000000122b20d4 in .QMetaCallEvent::placeMetaCall(QObject*) ()
> > #27 0x00000000122bbbd0 in .QObject::event(QEvent*) ()
> > #28 0x00000000118c1e78 in .QApplicationPrivate::notify_helper(QObject*, QEvent*) ()
> > #29 0x00000000118cab4c in .QApplication::notify(QObject*, QEvent*) ()
> > #30 0x000000001229a180 in .QCoreApplication::notifyInternal(QObject*, QEvent*)
> >     ()
> > #31 0x000000001229e97c in .QCoreApplicationPrivate::sendPostedEvents(QObject*, int, QThreadData*) ()
> > #32 0x00000000122d4dfc in .QEventDispatcherUNIX::processEvents(QFlags<QEventLoop::ProcessEventsFlag>) ()
> > #33 0x000000001229eef8 in .QCoreApplication::processEvents(QFlags<QEventLoop::ProcessEventsFlag>) ()
> > #34 0x000000001006930c in .wkhtmltopdf::ConverterPrivate::convert() ()
> > #35 0x000000001004340c in .main ()
> 
> It is an Qt issue that is doing an unsigned operations that might overflow.  It 
> uses system pagesize to make some calculation regarding a future mmap call to 'fix'
> alignment. However, since pagesize on powerpc64 is usually 64K, the operations overflows
> and the negative alignment adjustments ends up creating an wrong value that will be passed
> to mmap.  More specifically, it is on:
> 
>  54     size_t alignmentDelta = alignment - pageSize();
>  55 
>  56     // Resererve with suffcient additional VM to correctly align.
>  57     size_t reservationSize = size + alignmentDelta;
>  58     void* reservationBase = OSAllocator::reserveUncommitted(reservationSize, usage, writable, executable);
> 
> You can easily fix by:
> 
>  53     size_t pagesize = pageSize();
>  54     size_t alignmentDelta = 0;
>  55     if (alignment > pagesize)
>  56       alignmentDelta = alignment - pagesize;
>  57 
>  58     // Resererve with suffcient additional VM to correctly align.
>  59     size_t reservationSize = size + alignmentDelta;
>  60     void* reservationBase = OSAllocator::reserveUncommitted(reservationSize, usage, writable, executable);

[Bill Buros]

A colleague (Nish) filed a bug the underlying issue.

https://bugs.webkit.org/show_bug.cgi?id=135702

Appreciate Adhemerval Zanella and Nishanth Aravamudan assistance.