On Saturday, March 13, 2021 at 2:04:29 AM UTC-5, grze...@gmail.com wrote:
> tcam...@mozilla.com schrieb am Freitag, 12. März 2021 um 20:14:54 UTC+1:
> > On Friday, March 12, 2021 at 6:20:41 AM UTC-5, grze...@gmail.com wrote:
> > > tcam...@mozilla.com schrieb am Freitag, 12. März 2021 um 02:12:11 UTC+1:
> > > > On Thursday, March 11, 2021 at 1:15:42 PM UTC-5, grze...@gmail.com
> > > > wrote:
> > > > > Carsten Grzemba schrieb am Donnerstag, 11. März 2021 um 14:25:41
> > > > > UTC+1:
> > > > > > Lars Hansen schrieb am Donnerstag, 11. März 2021 um 13:25:45 UTC+1:
> > > > > > > If you run jit-tests with the options -f --no-xdr it will print
> > > > > > > out the
> > > > > > > command lines you need to use to run the failed tests
> > > > > > > individually.
> > > > > > >
> > > > > > > --lrs
> > > > > > > On Thu, Mar 11, 2021 at 12:05 PM Carsten Grzemba
> > > > > > > <grze...@gmail.com> wrote:
> > > > > > >
> > > > > > > > I attempt to build Firefox for Openindiana (Illumos). Some of
> > > > > > > > the tests
> > > > > > > > are failing:
> > > > > > > > Result summary:
> > > > > > > > Passed: 30561
> > > > > > > > Failed: 480
> > > > > > > >
> > > > > > > > How can I check what the reason for a failed check is?
> > > > > > > >
> > > > > > > > In the log I see for example:
> > > > > > > > TEST-PASS |
> > > > > > > > js/src/jit-test/tests/wasm/timeout/debug-interrupt-2.js |
> > > > > > > > Success (code 6, args "--ion-eager --ion-offthread-compile=off
> > > > > > > > --more-compartments") [1.0 s]
> > > > > > > > #1 module successfully instantiated: PASS.
> > > > > > > > #2 A wast module that must return a particular value.: PASS.
> > > > > > > > #3 A wast module that must trap at runtime.: PASS.
> > > > > > > > #4 A wast module that must trap at runtime.: PASS.
> > > > > > > > #5 A wast module that must trap at runtime.: PASS.
> > > > > > > > #6 A wast module that must trap at runtime.: PASS.
> > > > > > > > Exit code: -11
> > > > > > > > FAIL - wasm/spec/memory_grow.wast.js
> > > > > > > > TEST-UNEXPECTED-FAIL |
> > > > > > > > js/src/jit-test/tests/wasm/spec/memory_grow.wast.js
> > > > > > > > | Unknown (code -11, args "--ion-eager
> > > > > > > > --ion-offthread-compile=off
> > > > > > > > --ion-check-range-analysis --ion-extra-checks --no-sse3
> > > > > > > > --no-threads")
> > > > > > > > [20.0 s]
> > > > > > > > INFO exit-status : -11
> > > > > > > > INFO timed-out : False
> > > > > > > > INFO stdout > #1 module successfully instantiated: PASS.
> > > > > > > > INFO stdout > #2 A wast module that must return a particular
> > > > > > > > value.: PASS.
> > > > > > > > INFO stdout > #3 A wast module that must trap at runtime.:
> > > > > > > > PASS.
> > > > > > > > INFO stdout > #4 A wast module that must trap at runtime.:
> > > > > > > > PASS.
> > > > > > > > INFO stdout > #5 A wast module that must trap at runtime.:
> > > > > > > > PASS.
> > > > > > > > INFO stdout > #6 A wast module that must trap at runtime.:
> > > > > > > > PASS.
> > > > > > > >
> > > > > > > > TEST-PASS | js/src/jit-test/tests/wasm/timeout/1.js | Success
> > > > > > > > (code 6,
> > > > > > > > args "--wasm-compiler=baseline") [1.2 s]
> > > > > > > >
> > > > > > > > How can I run a test standalone, perhapse with a little bit
> > > > > > > > more
> > > > > > > > informations?
> > > > > > > > _______________________________________________
> > > > > > > > dev-platform mailing list
> > > > > > > > dev-pl...@lists.mozilla.org
> > > > > > > > https://lists.mozilla.org/listinfo/dev-platform
> > > > > > > >
> > > > > > cool!
> > > > > > thats is what Im looking for:
> > > > > >
> > > > > > grzemba@oibuild:~/oi-userland/components/web/firefox/firefox-68.12.0/js/src/jit-test$
> > > > > >
> > > > > > /ws/grzemba/oi-userland/components/web/firefox/build_opt/amd64/dist/bin/js
> > > > > > -f
> > > > > > /ws/grzemba/oi-userland/components/web/firefox/firefox-68.12.0/js/src/jit-test/lib/prologue.js
> > > > > > -e 'const platform='"'"'sunos5'"'"'' -e 'const
> > > > > > libdir='"'"'/ws/grzemba/oi-userland/components/web/firefox/firefox-68.12.0/js/src/jit-test/lib/'"'"''
> > > > > > -e 'const
> > > > > > scriptdir='"'"'/ws/grzemba/oi-userland/components/web/firefox/firefox-68.12.0/js/src/jit-test/tests/asm.js/'"'"''
> > > > > > -e 'if ((!this.SharedArrayBuffer ||
> > > > > > !isAsmJSCompilationAvailable())) quit(59)' --module-load-path
> > > > > > /ws/grzemba/oi-userland/components/web/firefox/firefox-68.12.0/js/src/jit-test/modules/
> > > > > > -f
> > > > > > /ws/grzemba/oi-userland/components/web/firefox/firefox-68.12.0/js/src/jit-test/tests/asm.js/testBug1057248.js
> > > > > >
> > > > > > Segmentation Fault (core dumped)
> > > > > >
> > > > > > core 'core' of 4573:
> > > > > > /ws/grzemba/oi-userland/components/web/firefox/build_opt/amd64/dist/bin
> > > > > >
> > > > > > --------------------- thread# 1 / lwp# 1 ---------------------
> > > > > > 0000000000d51211
> > > > > > js::InternalBarrierMethods<JS::Value>::preBarrier(JS::Value const&)
> > > > > > () + 5b3
> > > > > > 0000000000b43671 js::SharedArrayBufferObject::dropRawBuffer() () +
> > > > > > 33
> > > > > > 0000000000b436eb js::SharedArrayBufferObject::Finalize(js::FreeOp*,
> > > > > > JSObject*) () + 51
> > > > > > 0000000000d3b700 bool FinalizeTypedArenas<JSObject>(js::FreeOp*,
> > > > > > js::gc::Arena**, js::gc::SortedArenaList&, js::gc::AllocKind,
> > > > > > js::SliceBudget&, js::gc::ArenaLists::KeepArenasEnum) () + 211
> > > > > > 0000000000d3bb98 FinalizeArenas(js::FreeOp*, js::gc::Arena**,
> > > > > > js::gc::SortedArenaList&, js::gc::AllocKind, js::SliceBudget&,
> > > > > > js::gc::ArenaLists::KeepArenasEnum) () + 187
> > > > > > 0000000000d3f124
> > > > > > js::gc::ArenaLists::backgroundFinalize(js::FreeOp*, js::gc::Arena*,
> > > > > > js::gc::Arena**) () + b0
> > > > > > 0000000000d3f39a
> > > > > > js::gc::GCRuntime::sweepBackgroundThings(js::gc::ZoneList&,
> > > > > > js::LifoAlloc&) () + ba
> > > > > > 0000000000d3f6b2
> > > > > > js::gc::GCRuntime::sweepFromBackgroundThread(js::AutoLockHelperThreadState&)
> > > > > > () + 124
> > > > > > 0000000000d3fa13 js::gc::BackgroundSweepTask::run() () + 47
> > > > > > 0000000000d3fa50
> > > > > > js::GCParallelTaskHelper<js::gc::BackgroundSweepTask>::runTaskTyped(js::GCParallelTask*)
> > > > > > () + 9
> > > > > > 00000000009f0ebf js::GCParallelTask::runTask() () + 7
> > > > > > 00000000009f3f90 js::GCParallelTask::runFromMainThread(JSRuntime*)
> > > > > > () + 20
> > > > > > 00000000009f403b
> > > > > > js::GCParallelTask::joinAndRunFromMainThread(JSRuntime*) () + 5b
> > > > > > 0000000000d2da40
> > > > > > js::gc::GCRuntime::queueZonesAndStartBackgroundSweep(js::gc::ZoneList&)
> > > > > > () + 9c
> > > > > > 0000000000d2de3a
> > > > > > js::gc::GCRuntime::endSweepingSweepGroup(js::FreeOp*,
> > > > > > js::SliceBudget&) () + 1e4
> > > > > > 0000000000d275c0 sweepaction::SweepActionCall<js::FreeOp*,
> > > > > > js::SliceBudget&>::run(js::gc::GCRuntime*, js::FreeOp*,
> > > > > > js::SliceBudget&) () + 2c
> > > > > > 0000000000d28fdf
> > > > > > sweepaction::SweepActionSequence<js::gc::GCRuntime*, js::FreeOp*,
> > > > > > js::SliceBudget&>::run(js::gc::GCRuntime*, js::FreeOp*,
> > > > > > js::SliceBudget&) () + 39
> > > > > > 0000000000d2e940
> > > > > > sweepaction::SweepActionRepeatFor<js::gc::SweepGroupsIter,
> > > > > > JSRuntime*, js::gc::GCRuntime*, js::FreeOp*,
> > > > > > js::SliceBudget&>::run(js::gc::GCRuntime*, js::FreeOp*,
> > > > > > js::SliceBudget&) () + 3e
> > > > > > 0000000000d5ea4f
> > > > > > js::gc::GCRuntime::performSweepActions(js::SliceBudget&) () + 67
> > > > > > 0000000000d5ee53
> > > > > > js::gc::GCRuntime::incrementalSlice(js::SliceBudget&, JS::GCReason,
> > > > > > js::gc::AutoGCSession&) () + 3ad
> > > > > > 0000000000d5f37e js::gc::GCRuntime::gcCycle(bool, js::SliceBudget,
> > > > > > JS::GCReason) () + 1fc
> > > > > > 0000000000d5f83c js::gc::GCRuntime::collect(bool, js::SliceBudget,
> > > > > > JS::GCReason) () + b2
> > > > > > 0000000000d5fa6d js::gc::GCRuntime::gc(JSGCInvocationKind,
> > > > > > JS::GCReason) () + 4f
> > > > > > 0000000000aaa4cd JSRuntime::destroyRuntime() () + 13d
> > > > > > 0000000000a1be11 js::DestroyContext(JSContext*) () + a9
> > > > > > 0000000000c14e5f JS_DestroyContext(JSContext*) () + 9
> > > > > > 00000000008560e6 main () + 21fb
> > > > > > 0000000000836427 _start_crt () + 87
> > > > > > 0000000000836388 _start () + 18
> > > > > For the failed tests I get core dumps initiated by
> > > > > MOZ_ASSERT(isDouble()) in setPrivate(void*) mostly, with stack traces
> > > > > similar like this:
> > > > > --------------------- thread# 1 / lwp# 1 ---------------------
> > > > > 0000000001c85879 JS::Value::setPrivate(void*) () + af
> > > > > 0000000001edda15 JS::PrivateValue(void*) () + 3d
> > > > > 0000000001ee9320
> > > > > js::ArrayBufferObject::setDataPointer(js::ArrayBufferObject::BufferContents)
> > > > > () + 32
> > > > >
> > > > > I assume the core is the wanted behaviour of the MOZ_ASSERT.
> > > > > How can I investigate was is going wrong here?
> > > > > I do the test with Firefox 68.1.20 esr. What I can found on
> > > > > https://bugzilla.mozilla.org is related to newer versions of FF and
> > > > > the failed test for bugs very old.
> > > > >
> > > > > Many Thanks
> > > > This sounds like you are hitting Bug 1568564 where the encoding of
> > > > `PrivateValue` type acted up on certain compiler configurations in
> > > > Firefox 68 ESR. We did backport a workaround to Firefox 68.2.0+ that
> > > > added an optional --enable-unaligned-private-values compilation flag.
> > > > If this flag is available in your version you should give it a try.
> > > >
> > > > The issue in that bug is that the code expects a certain alignment that
> > > > was stricter than some compiler / OS pairs guaranteed.
> > > The comment in setPrivate mentions: ptr must be a valid user-mode
> > > pointer, with the top 16 bits clear.
> > > But if I get the assertion, than the ptr is 0xfffffbffef120fb0
> > > Where do the pointer come from? In the sense of pointer the address is
> > > ok, it should be an mmap'ed area, but why expects setPrivate whith the
> > > top 16 bits clear?
> > The JS::Value representation uses a technique called "NaN-boxing" to
> > encoding pointers as special versions of floating-point NaN values. In this
> > approach, there is a limited number of bits available for storing pointer
> > values. To work around this we take advantage of the fact that 64-bit
> > processors generally do not use 64-bits of actual address space yet. See
> > this wikipedia diagram for more info:
> > https://en.wikipedia.org/wiki/X86-64#Virtual_address_space_details . In
> > that diagram there is an "upper" and "lower" half of memory. While it is
> > possible to sign-extend the 48-bit number to 64-bit to get the expected
> > value, we find that the main operating systems we support don't use the
> > "upper" region for user-space pointers so we avoid performance overhead of
> > this 48->64 sign-extension.
> >
> > If this failure mostly experienced in these WASM SharedArrayBuffer cases,
> > you may be able to modify where the mmap is called. I don't know what the
> > appropriate way on your OS to get a page that is in the "lower" half of
> > memory.
> > https://searchfox.org/mozilla-central/rev/491c8096b5dfdb328b2135895062e16e1e36d708/mfbt/TaggedAnonymousMemory.cpp#85
> >
> >
> > If you are investigating possible mmap solutions, it may be helpful to look
> > at the "scattershot" allocation that the GC uses:
> > https://searchfox.org/mozilla-central/rev/491c8096b5dfdb328b2135895062e16e1e36d708/js/src/gc/Memory.cpp#467-487
> I compiled without the option --enable-jemalloc.
> Can I expect a different behavior here if I compile with --enable-jemalloc ?
> Or it this mmap usage beside jemalloc.
These `mmap` calls are not related to jemalloc. The garbage collector (GC) and
the SharedArrayBuffer code use `mmap` directly. It is possible that you would
also need to modify jemalloc as well.
If you are changing code to be able to get the browser to just start, then it
might be easiest to change the definition of JS::Value::setPrivate /
JS::Value::toPrivate. The JITs do not have their own version of this code so
you do not need to worry about them. Something like:
void setPrivate(void* ptr) {
// Clear upper 16-bits
asBits_ = (uintptr_t(ptr) << 16) >> 16;
MOZ_ASSERT(isDouble());
}
void* toPrivate() const {
MOZ_ASSERT(isDouble());
// Sign-extend bit-47
intptr_t raw = (intptr_t(asBits_) << 16) >> 16;
return reinterpret_cast<void*>(raw);
}
I have not tested this code as I don't have any OS available that has this
behaviour.
_______________________________________________
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform
