DevTools can't unambiguously determine whether code is JS or wasm.
This CL adds a string to the tracing output that will be 'JS', 'wasm' or
'other'.
Bug: chromium:1168052
Change-Id: Iaacb5ea9a83327e22d60bf6114f607e6fa5532ad
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2637859
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/master@{#72193}
76217f5 fixed the profiler so it would only sample a thread if
it had the Isolate lock. Unfortunately, this fix missed a timing
window where a thread might have the Isolate lock but might not
have restored the thread-specific data such as thread_local_top_
for the locked thread yet, so the sampler might end up using data
from a different thread.
This doesn't cause any seg faults or the like because the thread
we *meant* to sample has the Isolate lock so the thread we're
accidentally sampling can't mess with any Isolate data but we can
still get incorrect sample data which can be especially obvious if
the accidentally sampled thread is inside code that would never
run on the thread we meant to sample.
Fortunately, we can tell when all thread-specific data has been
restored to the Isolate because thread_state_ in the
PerIsolateThreadData for a thread is set to a non-null value
until everything has been restored, at which point it gets set
to null. So the fix adds a check after the test for the Isolate
lock to check if thread_state_ is null for the thread we mean to
sample. If so, we know all the data in the Isolate is good to go
for sampling.
Bug: v8:11316
Change-Id: I02d6361d8cbd6ec809ad8fb7ef07f5e9c94c7d1e
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2628133
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#72112}
As a first step towards freeing CodeEntry objects that are neither still
referenced by JS or stored in a profile, enable freeing of refcounted
strings by CodeEntry instances. For now, this leaves behaviour unchanged
until we receive CodeEntry destruction events.
Bug: v8:11054
Change-Id: Iabd05aa730343cd1a879ff5b04326f23e68aa948
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2590604
Commit-Queue: Andrew Comminos <acomminos@fb.com>
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#71858}
This CL changes SharedFunctionInfo::GetBytecodeArray to a function
template, which is specialized for Isolate and LocalIsolate arguments.
This allows main thread only uses to avoid taking a lock.
Bug: v8:7790, chromium:1154603
Change-Id: I3462c4e36b66073e09393c01c765dd8a018a98f0
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2595307
Commit-Queue: Nico Hartmann <nicohartmann@chromium.org>
Reviewed-by: Michael Stanton <mvstanton@chromium.org>
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Cr-Commit-Position: refs/heads/master@{#71833}
There is a race in the way we handle deopts that made this test flaky.
The race is not hugely important to fix, and is difficult without
breaking something else.
The best thing to do here is update the test to reflect reality so we
can get the test coverage back.
This updates the test so that the deopt reason can be found either
on the first or second level function. The test assumed it would
always be available on the second level function in the profile,
but if we get a regular profile tick at the exact wrong time, we
could end up with the deopt info getting attached to the first level
function. So we accept either.
Bug: v8:5193
Change-Id: Ia43880ebafd1341a514b3143dc215514b5dccf15
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2594775
Reviewed-by: Simon Zünd <szuend@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#71797}
This CL moves the disabling to the test source code.
Bug: v8:11178
Change-Id: I9771c1eb2ea474143bb401ba40f4452f6e6572ec
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2549963
Commit-Queue: Maya Lekova <mslekova@chromium.org>
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Auto-Submit: Maya Lekova <mslekova@chromium.org>
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#71278}
This CL introduces a new fast_api_call_target field on the isolate,
which is set by Turbofan before making the fast call. It then uses
the field when creating a stack sample and stores it in the existing
external_callback_entry used for regular API callbacks. The CL also
adds a cctest with simple usage scenario and introduces a minor
refactoring in test-api.cc.
Design doc:
https://docs.google.com/document/d/1r32qlPzGz0P7nieisJ5h2qfSnWOs40Cigt0LXPipejE/edit
Bug: chromium:1052746
Change-Id: I2dab1bc395ccab0c14088f7c354fb52b08df8d32
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2488683
Commit-Queue: Maya Lekova <mslekova@chromium.org>
Reviewed-by: Georg Neis <neis@chromium.org>
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#71254}
- Created status enum with statuses kStarted, kAlreadyStarted and
kErrorTooManyProfilers, returning when StartProfiling is invoked
- Tests spin up one profiler, check kStarted returned; spin up
another with same name, check kAlreadyStarted returned; Spin up 99
more profilers (100 total), check each returning kStarted, and
one more, expecting 101st to return kErrorTooManyProfilers
R=acomminos@fb.com, petermarshall@chromium.org, ulan@chromium.org
Change-Id: I64e2e6396775f90f9f49f75331a075a47efa7fca
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2486240
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Commit-Queue: Ulan Degenbaev <ulan@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70808}
We kept the CodeMap filled with entries between profiles, even in
kLazyLogging mode which will re-fill the CodeMap when profiling starts
again. See the bug for more details.
This fix manually clears the CodeMap after the last profile is deleted.
We already call DisableLogging() when the last profile is stopped. At
this point we still need the CodeMap alive because the profile object
we expose via the API is backed by the CodeEntry objects in the CodeMap.
Once the last profile is deleted though, we can empty the CodeMap.
There is still another bug, which is that we never delete CodeEntry
objects for deleted code, as there are no CodeDeleteEvents from the GC.
We will work on that separately, but this fix should stop those leaks
accumulating between profiles as we wipe the CodeMap entirely between
profiles (at least for kLazyLogging mode). kEagerLogging mode still has
this problem and will only be fixed by introducing CodeDelete events or
similar.
Bug: v8:11051
Change-Id: Iab9570747d17c657e6e318d434f935af8047d05f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2491033
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Simon Zünd <szuend@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70792}
No major functional changes in this CL, mostly it moves code around
to make follow-up CLs less messy.
- Document Code layout.
- New concepts: 'body' and 'metadata' areas of Code objects. The
metadata area contains metadata tables, the body area includes
both instructions and metadata (this is currently the 'instructions'
area). Add accessors for these new areas.
- An interesting detail: embedded builtins will have non-adjacent
instruction and metadata areas, thus a concept of 'body' doesn't
make sense there.
- Also add raw_instruction_X_future accessors; these are used where
we are actually interested in the instructions range, not the entire
body. In a follow-up, current raw_instruction_X accessors will be
replaced by raw_body_X, and raw_instruction_X_future by
raw_instruction_X.
Bug: v8:11036
Change-Id: I1d85146b652e0c097c3602d4db1862d5d3898a7e
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2491023
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70701}
Always spend 1ms per iteration.
Previously if the profilerthread took a long time to start up then we
would skip through iterations and potentially not gather enough samples.
This forces each iteration to take 1ms.
Bug: v8:10996
Change-Id: I0dd7bb7e31636c9ebf5dd99110c8a976cbc8f045
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2461727
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Simon Zünd <szuend@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70466}
1. Set profiling interval to 100us to get 10x the samples
2. Guarantee we spend at least 1ms per iteration, instead of only
bailing out if we spend more than 1ms. This gives us enough samples on
release mode.
3. Increase the time spent profiling optimized code
Bug: v8:10996
Change-Id: I1348ebce48fe998e79b5847f3e3d037148302dcc
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2460823
Reviewed-by: Simon Zünd <szuend@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70403}
Rename it to Symbolizer because it does exactly that.
Change the SymbolizeTickSample method to return the symbolized state
rather than pass it on to the ProfilesCollection. This makes it easier
to test as now it only relies on the CodeMap provided to it.
Make EntryForVMState a free-floating function as it doesn't rely on
state and then we can avoid importing the StateTag definition in the
header.
Remove the UNREACHABLE from EntryForVMState as the compiler got smarter
and doesn't need it anymore.
Pass the CpuProfilesCollection to SamplingEventsProcessor instead,
as it is now responsible for putting the symbolized samples into the
collection to be sorted into the appropriate profiles.
Change-Id: I104290eff22b7d94a1bd34ba904036badccf4e13
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2440522
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Simon Zünd <szuend@chromium.org>
Cr-Commit-Position: refs/heads/master@{#70248}
This reverts commit 8b60d8fcbf.
Reason for revert: Flaky on windows: https://ci.chromium.org/p/v8/builders/ci/V8%20Win32%20-%20debug/27302
Original change's description:
> Reland "[cpu-profiler] Log OSR code when starting the profiler"
>
> This is a reland of f696528189
>
> Updated the test:
> 1. Set profiling interval to 100us to get 10x the samples
> 2. Guarantee we spend at least 1ms per iteration, instead of only
> bailing out if we spend more than 1ms. This gives us enough samples on
> release mode.
> 3. Increase the time spent profiling optimized code by 50% to make sure
> we have a big enough difference.
>
> With 1000 iterations I didn't see any flakes locally so this looks solid
> now.
>
> Original change's description:
> > [cpu-profiler] Log OSR code when starting the profiler
> >
> > OSR code doesn't hang off any JSFunction or SFI, so we missed it when
> > starting up the profiler. This meant we didn't properly attribute
> > ticks to SFI code. The ticks ended up going to the caller instead.
> >
> > There is a weak cache of OSR code per native context, so iterate that
> > on profiler startup and log all the code objects.
> >
> > Change-Id: I2e9738b86a488b37f36ac89803561607dc76f745
> > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2414216
> > Commit-Queue: Peter Marshall <petermarshall@chromium.org>
> > Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
> > Reviewed-by: Mythri Alle <mythria@chromium.org>
> > Cr-Commit-Position: refs/heads/master@{#69964}
>
> Change-Id: Ib506e88b546008e462967259763bbf985b74b462
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2418092
> Commit-Queue: Peter Marshall <petermarshall@chromium.org>
> Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
> Reviewed-by: Mythri Alle <mythria@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#69990}
TBR=mythria@chromium.org,petermarshall@chromium.org,dinfuehr@chromium.org
Change-Id: Ie3272c4fd297ca6f10a47c3fe8826e226a9f0545
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2418714
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69999}
This is a reland of f696528189
Updated the test:
1. Set profiling interval to 100us to get 10x the samples
2. Guarantee we spend at least 1ms per iteration, instead of only
bailing out if we spend more than 1ms. This gives us enough samples on
release mode.
3. Increase the time spent profiling optimized code by 50% to make sure
we have a big enough difference.
With 1000 iterations I didn't see any flakes locally so this looks solid
now.
Original change's description:
> [cpu-profiler] Log OSR code when starting the profiler
>
> OSR code doesn't hang off any JSFunction or SFI, so we missed it when
> starting up the profiler. This meant we didn't properly attribute
> ticks to SFI code. The ticks ended up going to the caller instead.
>
> There is a weak cache of OSR code per native context, so iterate that
> on profiler startup and log all the code objects.
>
> Change-Id: I2e9738b86a488b37f36ac89803561607dc76f745
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2414216
> Commit-Queue: Peter Marshall <petermarshall@chromium.org>
> Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
> Reviewed-by: Mythri Alle <mythria@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#69964}
Change-Id: Ib506e88b546008e462967259763bbf985b74b462
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2418092
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
Reviewed-by: Mythri Alle <mythria@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69990}
This reverts commit f696528189.
Reason for revert: Test is flaky: https://ci.chromium.org/p/v8/builders/ci/V8%20Linux64/39092
Original change's description:
> [cpu-profiler] Log OSR code when starting the profiler
>
> OSR code doesn't hang off any JSFunction or SFI, so we missed it when
> starting up the profiler. This meant we didn't properly attribute
> ticks to SFI code. The ticks ended up going to the caller instead.
>
> There is a weak cache of OSR code per native context, so iterate that
> on profiler startup and log all the code objects.
>
> Change-Id: I2e9738b86a488b37f36ac89803561607dc76f745
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2414216
> Commit-Queue: Peter Marshall <petermarshall@chromium.org>
> Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
> Reviewed-by: Mythri Alle <mythria@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#69964}
TBR=mythria@chromium.org,petermarshall@chromium.org,dinfuehr@chromium.org
Change-Id: I1e69f8af88d901bab6f257652d3536d24a4777f9
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2415994
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69969}
OSR code doesn't hang off any JSFunction or SFI, so we missed it when
starting up the profiler. This meant we didn't properly attribute
ticks to SFI code. The ticks ended up going to the caller instead.
There is a weak cache of OSR code per native context, so iterate that
on profiler startup and log all the code objects.
Change-Id: I2e9738b86a488b37f36ac89803561607dc76f745
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2414216
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Dominik Inführ <dinfuehr@chromium.org>
Reviewed-by: Mythri Alle <mythria@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69964}
While the sampler checked if the sampled thread had the Isolate locked
(if locks are being used) under Linux, the check was not done under
Windows (or Fuchsia) which meant that in a multi-threading application
under Windows, thread locking was not checked making it prone to seg
faults and the like as the profiler would be using isolate->js_entry_sp
to determine the stack to walk but isolate->js_entry_sp is the stack
pointer for the thread that currently has the Isolate lock so, if the
sampled thread does not have the lock, the sampler woud be iterating
over the wrong stack, one that might actually be actively changing on
another thread. The fix was to move the lock check into CpuSampler
and Ticker (--prof) so all OSes would do the correct check.
The basic concept is that on all operating systems a CpuProfiler, and
so its corresponding CpuCampler, the profiler is tied to a thread.
This is not based on first principles or anything, it's simply the
way it works in V8, though it is a useful conceit as it makes
visualization and interpretation of profile data much easier.
To collect a sample on a thread associated with a profiler the thread
must be stopped for obvious reasons -- walking the stack of a running
thread is a formula for disaster. The mechanism for stopping a thread
is OS-specific and is done in sample.cc. There are currently three
basic approaches, one for Linux/Unix variants, one for Windows and one
for Fuchsia. The approaches vary as to which thread actually collects
the sample -- under Linux the sample is actually collected on the
(interrupted) sampled thread whereas under Fuchsia/Windows it's on
a separate thread.
However, in a multi-threaded environment (where Locker is used), it's
not sufficient for the sampled thread to be stopped. Because the stack
walk involves looking in the Isolate heap, no other thread can be
messing with the heap while the sample is collected. The only ways to
ensure this would be to either stop all threads whenever collecting a
sample, or to ensure that the thread being sampled holds the Isolate
lock so prevents other threads from messing with the heap. While there
might be something to be said for the "stop all threads" approach, the
current approach in V8 is to only stop the sampled thread so, if in a
multi-threaded environment, the profiler must check if the thread being
sampled holds the Isolate lock.
Since this check must be done, independent of which thread the sample
is being collected on (since it varies from OS to OS), the approach is
to save the thread id of the thread to be profiled/sampled when the
CpuSampler is instantiated (on all OSes it is instantiated on the
sampled thread) and then check that thread id against the Isolate lock
holder thread id before collecting a sample. If it matches, we know
sample.cc has stop the sampled thread, one way or another, and we know
that no other thread can mess with the heap (since the stopped thread
holds the Isolate lock) so it's safe to walk the stack and collect data
from the heap so the sample can be taken. It it doesn't match, we can't
safely collect the sample so we don't.
Bug: v8:10850
Change-Id: Iba6cabcd3e11a19c261c004103e37e806934dc6f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2411343
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69952}
If code is deoptimized while CPU profiling, a deoptimization event
record is sent to the profiler. But if the profiler could not find
the associated CodeMap entry in CodeDeoptEventRecord::UpdateCodeMap
it would simply return without freeing the deopt_frames array.
This change frees the deopt_frames array no matter what in
CodeDeoptEventRecord::UpdateCodeMap, eliminating a storage leak.
Bug: v8:10861
Change-Id: I4e68566bb91dff13b38e255ddfed24b85b7a1d57
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2386332
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69901}
Since the web-exposed profiler will require COOP/COEP, it is no longer
necessary to perform isolation at the V8 level. Strip the unnecessary
complexity and unreliability of context filtering accordingly.
Bug: chromium:956688, v8:9881, v8:9860
Change-Id: I21a30d51f8daf7565ec95de8c265e9d3b9d10fad
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2386144
Commit-Queue: Andrew Comminos <acomminos@fb.com>
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69894}
Using the Template::Set method which const char name is more ergonomic
and it creates directly an internalized name instead of the normal
string that most users pass in.
Bug: v8:10884
Change-Id: I00c6d49fee9de16b8ebbfe75be4b383831f0d4dd
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2400980
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Commit-Queue: Camillo Bruni <cbruni@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69787}
This adds a global counter for the various reasons we might fail to
attribute a tick.
The counters are cleared and printed when Profile::Print() is called,
which we call in our tests, so flaky test output will now contain these
stats along with the printed profile tree.
Drive-by cleanup some print functions and make them const.
Change-Id: Ia3a27405f5b5346adfdbb32afc7e414857969cc5
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1550406
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69647}
This reverts commit dfb3f7daa5.
Reason for revert: Breaks LSAN & ASAN flakily: https://bugs.chromium.org/p/v8/issues/detail?id=10861
Original change's description:
> [cpu-profiler] Ensure sampled thread has Isolate lock under Windows
>
> While the sampler checked if the sampled thread had the Isolate locked
> (if locks are being used) under Linux, the check was not done under
> Windows (or Fuchsia) which meant that in a multi-threading application
> under Windows, thread locking was not checked making it prone to seg
> faults and the like as the profiler would be extracting info from a
> heap in motion. The fix was to move the lock check into CpuSampler
> and Ticker (--prof) so all OSes would do the correct check.
>
> The basic concept is that on all operating systems a CpuProfiler, and
> so its corresponding CpuCampler, the profiler is tied to a thread.
> This is not based on first principles or anything, it's simply the
> way it works in V8, though it is a useful conceit as it makes
> visualization and interpretation of profile data much easier.
>
> To collect a sample on a thread associated with a profiler the thread
> must be stopped for obvious reasons -- walking the stack of a running
> thread is a formula for disaster. The mechanism for stopping a thread
> is OS-specific and is done in sample.cc. There are currently three
> basic approaches, one for Linux/Unix variants, one for Windows and one
> for Fuchsia. The approaches vary as to which thread actually collects
> the sample -- under Linux the sample is actually collected on the
> (interrupted) sampled thread whereas under Fuchsia/Windows it's on
> a separate thread.
>
> However, in a multi-threaded environment (where Locker is used), it's
> not sufficient for the sampled thread to be stopped. Because the stack
> walk involves looking in the Isolate heap, no other thread can be
> messing with the heap while the sample is collected. The only ways to
> ensure this would be to either stop all threads whenever collecting a
> sample, or to ensure that the thread being sampled holds the Isolate
> lock so prevents other threads from messing with the heap. While there
> might be something to be said for the "stop all threads" approach, the
> current approach in V8 is to only stop the sampled thread so, if in a
> multi-threaded environment, the profiler must check if the thread being
> sampled holds the Isolate lock.
>
> Since this check must be done, independent of which thread the sample
> is being collected on (since it varies from OS to OS), the approach is
> to save the thread id of the thread to be profiled/sampled when the
> CpuSampler is instantiated (on all OSes it is instantiated on the
> sampled thread) and then check that thread id against the Isolate lock
> holder thread id before collecting a sample. If it matches, we know
> sample.cc has stop the sampled thread, one way or another, and we know
> that no other thread can mess with the heap (since the stopped thread
> holds the Isolate lock) so it's safe to walk the stack and collect data
> from the heap so the sample can be taken. It it doesn't match, we can't
> safely collect the sample so we don't.
>
> Bug: v8:10850
> Change-Id: Iab2493130b9328430d7e5f5d3cf90ad6d10b1892
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2377108
> Reviewed-by: Peter Marshall <petermarshall@chromium.org>
> Commit-Queue: Peter Marshall <petermarshall@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#69623}
TBR=akodat@rocketsoftware.com,petermarshall@chromium.org,petermarshall@google.com
Change-Id: Ib6b6dc4ce109d5aa4e504fa7c9769f5cd95ddd0c
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: v8:10850
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2387570
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69638}
While the sampler checked if the sampled thread had the Isolate locked
(if locks are being used) under Linux, the check was not done under
Windows (or Fuchsia) which meant that in a multi-threading application
under Windows, thread locking was not checked making it prone to seg
faults and the like as the profiler would be extracting info from a
heap in motion. The fix was to move the lock check into CpuSampler
and Ticker (--prof) so all OSes would do the correct check.
The basic concept is that on all operating systems a CpuProfiler, and
so its corresponding CpuCampler, the profiler is tied to a thread.
This is not based on first principles or anything, it's simply the
way it works in V8, though it is a useful conceit as it makes
visualization and interpretation of profile data much easier.
To collect a sample on a thread associated with a profiler the thread
must be stopped for obvious reasons -- walking the stack of a running
thread is a formula for disaster. The mechanism for stopping a thread
is OS-specific and is done in sample.cc. There are currently three
basic approaches, one for Linux/Unix variants, one for Windows and one
for Fuchsia. The approaches vary as to which thread actually collects
the sample -- under Linux the sample is actually collected on the
(interrupted) sampled thread whereas under Fuchsia/Windows it's on
a separate thread.
However, in a multi-threaded environment (where Locker is used), it's
not sufficient for the sampled thread to be stopped. Because the stack
walk involves looking in the Isolate heap, no other thread can be
messing with the heap while the sample is collected. The only ways to
ensure this would be to either stop all threads whenever collecting a
sample, or to ensure that the thread being sampled holds the Isolate
lock so prevents other threads from messing with the heap. While there
might be something to be said for the "stop all threads" approach, the
current approach in V8 is to only stop the sampled thread so, if in a
multi-threaded environment, the profiler must check if the thread being
sampled holds the Isolate lock.
Since this check must be done, independent of which thread the sample
is being collected on (since it varies from OS to OS), the approach is
to save the thread id of the thread to be profiled/sampled when the
CpuSampler is instantiated (on all OSes it is instantiated on the
sampled thread) and then check that thread id against the Isolate lock
holder thread id before collecting a sample. If it matches, we know
sample.cc has stop the sampled thread, one way or another, and we know
that no other thread can mess with the heap (since the stopped thread
holds the Isolate lock) so it's safe to walk the stack and collect data
from the heap so the sample can be taken. It it doesn't match, we can't
safely collect the sample so we don't.
Bug: v8:10850
Change-Id: Iab2493130b9328430d7e5f5d3cf90ad6d10b1892
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2377108
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69623}
Changes the isolate's string table into an off-heap structure. This
allows the string table to be resized without allocating on the V8 heap,
and potentially triggering a GC. This allows existing strings to be
inserted into the string table without requiring allocation.
This has two important benefits:
1) It allows the deserializer to insert strings directly into the
string table, rather than having to defer string insertion until
deserialization completes.
2) It simplifies the concurrent string table lookup to allow resizing
the table inside the write lock, therefore eliminating the race
where two concurrent lookups could both resize the table.
The off-heap string table has the following properties:
1) The general hashmap behaviour matches the HashTable, i.e. open
addressing, power-of-two sized, quadratic probing. This could, of
course, now be changed.
2) The empty and deleted sentinels are changed to Smi 0 and 1,
respectively, to make those comparisons a bit cheaper and not
require roots access.
3) When the HashTable is resized, the old elements array is kept
alive in a linked list of previous arrays, so that concurrent
lookups don't lose the data they're accessing. This linked list
is cleared by the GC, as then we know that all threads are in
a safepoint.
4) The GC treats the hash table entries as weak roots, and only walks
them for non-live reference clearing and for evacuation.
5) Since there is no longer a FixedArray to serialize for the startup
snapshot, there is now a custom serialization of the string table,
and the string table root is considered unserializable during weak
root iteration. As a bonus, the custom serialization is more
efficient, as it skips non-string entries.
As a drive-by, rename LookupStringExists_NoAllocate to
TryStringToIndexOrLookupExisting, to make it clearer that it returns
a non-string for the case when the string is an array index. As another
drive-by, extract StringSet into a separate header.
Bug: v8:10729
Change-Id: I9c990fb2d74d1fe222920408670974a70e969bca
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2339104
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69270}
The devtools frontend used this to set a lower interval (100us). Now the
frontend will stop setting this category and we will always profile
with 100us (which was already the default).
Bug: chromium:1082220
Change-Id: Iaa671d9750ad4c13b95c2c1ba1e433a1077c858a
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2198989
Reviewed-by: Sigurd Schneider <sigurds@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#67805}
This patch replaces V8's tracing implementation (i.e., the TRACE_EVENT
macros) with the track event base implementation from Perfetto. The
advantages of doing this are:
1) This allows us to remove most tracing-related backend code from V8.
2) V8 can start writing strongly typed trace event arguments, which
are more compact, easier to process and more extensible than legacy
JSON-based trace arguments.
For the time being, we still support the old trace macros when V8 is
embedded into Chrome and other embedders.
Design doc: https://docs.google.com/document/d/1f7tt4cb-JcA5bQFR1oXk60ncJPpkL02_Hi_Bc6MfTQk/edit#heading=h.398p6b4eaen2
Bug: chromium:1006766
Change-Id: Ie71474fbe065821772b13d851487ebbca680c4ae
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1947688
Commit-Queue: Sami Kyöstilä <skyostil@chromium.org>
Auto-Submit: Sami Kyöstilä <skyostil@chromium.org>
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Cr-Commit-Position: refs/heads/master@{#67217}
This improves runtime and guarantees a number of samples, reducing
the chance of flakes and simplfying the code as well.
Change-Id: Ife559475240670e40e194285ba73853d18672e97
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1981506
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#65575}
This test tried to get enough samples by varying the amount of work
done based on simulator vs. non-simulator builds. This didn't take
into account other things like release vs. debug, and some bots took
a very long time to run this test.
This CL uses the existing ProfilerHelper and sets a minimum number of
samples so that we can actually run the minimum amount of work for
any platform without manual tweaking.
This decreases runtime on x64.optdebug from 18.5 to 1.08s, and from
10+ min on x64.debug to 1.8s. arm.debug is ~4s now.
Bug: v8:10013
Change-Id: Ibdbdec00a300683554e3c67013e56d27dc7e6257
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1981504
Auto-Submit: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#65574}
Just a cleanup, should not change behavior, although we will allocate
more handles in some cases.
Also re-orders some of the implementations of the interface to try
and keep things consistent.
Included cleanup: Change CodeEventDispatcher so that it now implements
CodeEventListener, given that it had that exact interface already.
Also remove the macro dispatch to try and make things a bit easier to
read.
Bug: chromium:1033407
Change-Id: Id943b10c49f102d9783d8f4cf3a8c43e04364c77
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1976390
Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Reviewed-by: Tobias Tebbi <tebbi@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#65571}
This is a reland of 096d9c5663
Fixed DEPS files which were not caught by presubmit trybot
Original change's description:
> [tracing] Roll perfetto @ 28b633cd
>
> This catches up with [1] that make the proto include path
> relative to the project root rather than ./protos/
>
> [1] https://android-review.googlesource.com/c/platform/external/perfetto/+/1108421
>
> Cq-Include-Trybots: luci.v8.try:v8_linux64_perfetto_dbg_ng
> Bug: v8:8339
> Change-Id: I1f2dec93120142ea61cee864e4bf76a6947d958d
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1776088
> Reviewed-by: Peter Marshall <petermarshall@chromium.org>
> Reviewed-by: Tamer Tas <tmrts@chromium.org>
> Commit-Queue: Primiano Tucci <primiano@chromium.org>
> Auto-Submit: Primiano Tucci <primiano@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#63584}
Bug: v8:8339
Change-Id: I70e8b2d4520c620d02d5251d14bd61b90fb1d73f
Cq-Include-Trybots: luci.v8.try:v8_linux64_perfetto_dbg_ng
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1789143
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Tamer Tas <tmrts@chromium.org>
Cr-Commit-Position: refs/heads/master@{#63598}
This catches up with [1] that make the proto include path
relative to the project root rather than ./protos/
[1] https://android-review.googlesource.com/c/platform/external/perfetto/+/1108421
Cq-Include-Trybots: luci.v8.try:v8_linux64_perfetto_dbg_ng
Bug: v8:8339
Change-Id: I1f2dec93120142ea61cee864e4bf76a6947d958d
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1776088
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Tamer Tas <tmrts@chromium.org>
Commit-Queue: Primiano Tucci <primiano@chromium.org>
Auto-Submit: Primiano Tucci <primiano@chromium.org>
Cr-Commit-Position: refs/heads/master@{#63584}
We have internal::TickSample which inherits from this, but we never
use the public version in the API despite defining it there.
Change-Id: I6f0ce7ee663ef821be57cfbad540c1660484a525
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1745472
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Alexei Filippov <alph@chromium.org>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Cr-Commit-Position: refs/heads/master@{#63329}
Adds support to the CPU profiler for scraping the incumbent contexts of
V8 stack frames. While it is generally unsafe to access heap objects
during a profiling interrupt, the native context is uniquely usable due
to being guaranteed an alive root on the stack, as well as its slots
being immutable after context creation.
Change-Id: I2c3149c1302b74d2f13aa99d1fdd0cf006e0f9d1
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1580020
Commit-Queue: Andrew Comminos <acomminos@fb.com>
Reviewed-by: Yang Guo <yangguo@chromium.org>
Reviewed-by: Alexei Filippov <alph@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#63258}
Update the TracingCpuProfiler test to work properly with perfetto.
Roll perfetto to get fixes for bugs encountered with starting/stopping
tracing rapidly, which happens in the test for the tracing profiler.
Add a check that the DataSource::Register call was successful to flush
out any errors there (although they are fixed by the perfetto roll).
Emit a fake trace event when stopping tracing in order to avoid
losing the final trace event that the user provided. Remove the ad-hoc
fake final trace events that the cctests for perfetto added.
Add a test StartAndStopRepeated which flushes out the issues fixed by
the perfetto roll.
TBR=yangguo@chromium.org
Cq-Include-Trybots: luci.v8.try:v8_linux64_perfetto_dbg_ng
Bug: v8:8339
Change-Id: I042d7385486bf42c86f1631406974693868a477f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1731006
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#63083}
This is a reland of a0728e869b
Original change's description:
> [d8] Remove maximum workers limitation
>
> This CL refactors the lifetime management of the v8::Worker C++ object
> and in the process lifts the 100 maximum worker limitation. To do this,
> it uses a Managed<v8::Worker> heap object and attaches the managed to
> the API worker object.
>
> R=mstarzinger@chromium.org
> BUG=v8:9524
>
> Change-Id: I279b7aeb6645a87f9108ee6f572105739721cef4
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1715453
> Commit-Queue: Ben Titzer <titzer@chromium.org>
> Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
> Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#62932}
Bug: v8:9524
Change-Id: I7d903fb12ddb00909a9429455f46c55db2fd02de
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1722562
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Commit-Queue: Ben Titzer <titzer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#62974}
We can get repeated positions from optimized code objects in some cases
but for our purposes of looking up a line number from a PC, we can only
return one line number so just use the first one that is reported in
the source position table on the code object.
Change-Id: I4c0e866fb1948f65bf6c988d992ef55f520dd874
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1724375
Reviewed-by: Alexei Filippov <alph@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#62972}
Implements ProfilerCodeObserver, a class to track the generation and
movement of code on the heap for the lifetime of each CpuProfiler. When
sampling is inactive, logged code is committed directly to the CodeMap.
During profiling, ProfilerCodeObserver redirects these events onto the
profiling thread for later dispatch.
Bug: v8:9151
Change-Id: Ib5b152446d2a3838e1b00a80253fc4fbd2f6e8c3
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1604143
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Commit-Queue: Andrew Comminos <acomminos@fb.com>
Cr-Commit-Position: refs/heads/master@{#62943}
Add a bit on the isolate which indicates that the stack is currently
not iterable for the SafeStackFrameIterator.
This is needed during deoptimization, when we do a fast C call without
a return address on the stack, meaning we can't iterate the stack
frames.
Re-enable DeoptAtFirstLevelInlinedSource which is fixed by this CL.
Bug: v8:9057
Change-Id: I76379a2dd38023be7e6f5153edeb1f838e9ac4d6
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1688049
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Reviewed-by: Sigurd Schneider <sigurds@chromium.org>
Cr-Commit-Position: refs/heads/master@{#62634}
This replaces all typedefs that define types and not functions by the
equivalent "using" declaration.
This was done mostly automatically using this command:
ag -l '\btypedef\b' src test | xargs -L1 \
perl -i -p0e 's/typedef ([^*;{}]+) (\w+);/using \2 = \1;/sg'
Patchset 2 then adds some manual changes for typedefs for pointer types,
where the regular expression did not match.
R=mstarzinger@chromium.orgTBR=yangguo@chromium.org, jarin@chromium.org
Bug: v8:9183
Change-Id: I6f6ee28d1793b7ac34a58f980b94babc21874b78
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1631409
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#61849}
This is a reland of a03ed62679
Removed the added test which was failing on win32. The test was unrelated
to the CL; we can add it later.
Original change's description:
> [tracing] Add a configurable output stream for perfetto tracing
>
> Add the ability to provide perfetto with an output stream for the JSON
> consumer rather than hardcode it. D8 will use this interface exclusively
> once the old trace controller is removed.
>
> Also add a test for scope-managed trace events and their duration - this
> was leftover from a previous CL.
>
> Cq-Include-Trybots: luci.v8.try:v8_linux64_perfetto_dbg_ng
> Bug: v8:8339
> Change-Id: I1c45e17e528b549a4cfdaecabd33c7ac4ab4af77
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1611801
> Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Commit-Queue: Peter Marshall <petermarshall@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#61753}
TBR=jgruber@chromium.org, ulan@chromium.org
Bug: v8:8339
Change-Id: I3442a4d111e12947c107e7d0c226ae934acd06e4
Cq-Include-Trybots: luci.v8.try:v8_linux64_perfetto_dbg_ng
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1627334
Reviewed-by: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Cr-Commit-Position: refs/heads/master@{#61780}
This CL was generated by an automatic clang AST rewriter using this
matcher expression:
callExpr(
callee(
cxxMethodDecl(
hasName("operator->"),
ofClass(isSameOrDerivedFrom("v8::internal::Object"))
)
),
argumentCountIs(1)
)
The "->" at the expression location was then rewritten to ".".
R=jkummerow@chromium.orgTBR=mstarzinger@chromium.org,verwaest@chromium.org,yangguo@chromium.org
Bug: v8:9183, v8:3770
No-Try: true
No-Tree-Checks: true
Change-Id: I0a7ecabdeafe51d0cf427f5280af0c7cab96869e
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1624209
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Jakob Kummerow <jkummerow@chromium.org>
Reviewed-by: Toon Verwaest <verwaest@chromium.org>
Commit-Queue: Yang Guo <yangguo@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#61764}
This reverts commit a03ed62679.
Reason for revert: Fails on win32: https://ci.chromium.org/p/v8/builders/ci/V8%20Win32/21252
Original change's description:
> [tracing] Add a configurable output stream for perfetto tracing
>
> Add the ability to provide perfetto with an output stream for the JSON
> consumer rather than hardcode it. D8 will use this interface exclusively
> once the old trace controller is removed.
>
> Also add a test for scope-managed trace events and their duration - this
> was leftover from a previous CL.
>
> Cq-Include-Trybots: luci.v8.try:v8_linux64_perfetto_dbg_ng
> Bug: v8:8339
> Change-Id: I1c45e17e528b549a4cfdaecabd33c7ac4ab4af77
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1611801
> Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Commit-Queue: Peter Marshall <petermarshall@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#61753}
TBR=ulan@chromium.org,yangguo@chromium.org,jgruber@chromium.org,petermarshall@chromium.org
Change-Id: Ie7dbe4fc5f9e496fafc8e3ad2b05011b88c9e2c3
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: v8:8339
Cq-Include-Trybots: luci.v8.try:v8_linux64_perfetto_dbg_ng
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1625465
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#61754}
