Originally, we would restart the GPU process a fixed number of attempts
based on the layers.gpu-process.max_restarts pref. With this patch, we
now use this pref to control how many "unstable" restarts we allow. A
restart is "stable" if and only if the process uptime exceeds the pref
layers.gpu-process.stable.min-uptime-ts and if the process renders a
total number of frames exceeding the pref
layers.gpu-process.stable.frame-threshold. This allows users to keep the
GPU process for a lot longer if they are encountering infrequent
crashes. Should the user experience the GPU process crashing quickly
and/or without rendering many frames, we will disable it as before after
a few attempts and move into the parent process.
Differential Revision: https://phabricator.services.mozilla.com/D114531
On 32-bit Windows, we see crashes related to running out of virtual
memory address space in the GPU process. Prior to this patch, we did not
report any memory status information from the GPU process to the parent,
only from the parent to the GPU process. Now if we go below the
threshold we request memory to be cleared in the parent/content
processes. This should trickle down to the GPU process by freeing shared
memory resources such as images, allowing us to unmap them out of the
GPU process sooner.
We will see similar problems on any 32-bit platform. The only other
target of note with sufficient numbers is 32-bit Android. There is no
GPU process on Android, however we only monitor the virtual memory
address space on Windows.
Differential Revision: https://phabricator.services.mozilla.com/D106286
Upon starting the RDD and GPU process; we check their decoding capabilities and send it to the parent process. It will then broadcast this information to all content children so that RemodeDecoderModule::Supports can return if a codec is supported or not.
Differential Revision: https://phabricator.services.mozilla.com/D100306
Upon starting the RDD and GPU process; we check their decoding capabilities and send it to the parent process. It will then broadcast this information to all content children so that RemodeDecoderModule::Supports can return if a codec is supported or not.
Differential Revision: https://phabricator.services.mozilla.com/D100306
Out-of-process WebGL needs GfxInfo to exist in the composition
process (which is the GPU process if it exists and the parent process
otherwise). This patch enables the Linux version of that component in
the GPU process; the IPC currently used to give content processes copies
of the parent's GPU info is extended to also send it to the GPU process.
Differential Revision: https://phabricator.services.mozilla.com/D85443
Out-of-process WebGL needs GfxInfo to exist in the composition
process (which is the GPU process if it exists and the parent process
otherwise). This patch enables the Linux version of that component in
the GPU process; the IPC currently used to give content processes copies
of the parent's GPU info is extended to also send it to the GPU process.
Differential Revision: https://phabricator.services.mozilla.com/D85443
This patch uses IPDL's return feature to ensure that the memory
reporter manager won't wait for a report from a child process
that has already exited.
This fixes a memory reporter hang that can happen if a child process
exits during a memory report, when the parent half of the actor is
being held alive. (If the parent half of the actor is not being held
alive, then mMemoryReportRequest will be naturally cleared when it
goes away.)
This was happening frequently on Windows Fission AWSY because that test
does a minimize memory right before it attempts to get a memory report,
and the preallocated content process exits when it sees a message to
minimize memory.
Differential Revision: https://phabricator.services.mozilla.com/D85499
With these changes, on my Linux analysis with ClangBuildAnalyzer, the
top two expensive headers, DOMTypes.h and TabMessageUtils.h are no longer
among the 30 most expensive headers.
Differential Revision: https://phabricator.services.mozilla.com/D82935
Crash annotations in content processes are currently sent over IPC via
shared memory buffers. To pave the way for the Rust rewrite of the exception
handler we are removing this code and gathering all the crash annotations
within the content processes themselves. This patch causes annotations to be
stored in the global table of each content process. They are then streamed
out to the parent process by the exception handler together with the
exception-time annotations.
This has a number of benefits:
* we have one less channel to exchange data between content processes and
the parent process
* we save memory because we don't need to allocate the shared memory buffers
* annotations are faster because we don't stream them all out every time one
changes
* we won't truncate annotations anymore if we run out of space in the shared
segment.
* we don't need delayed annotations anymore, so we can get rid of the
associated machinery
As I refactored the code I tried to adjust all the obsolete comments,
consolidate shared code and remove the redundant steps that were sometimes
present. In many places we had two entire crash annotation tables we merged to
change just a couple; that comes from the fact that historically we loaded
them from disk. Now it doesn't matter anymore and we can just go ahead and
change the ones we care about.
Differential Revision: https://phabricator.services.mozilla.com/D62586
--HG--
extra : moz-landing-system : lando
Crash annotations in content processes are currently sent over IPC via
shared memory buffers. To pave the way for the Rust rewrite of the exception
handler we are removing this code and gathering all the crash annotations
within the content processes themselves. This patch causes annotations to be
stored in the global table of each content process. They are then streamed
out to the parent process by the exception handler together with the
exception-time annotations.
This has a number of benefits:
* we have one less channel to exchange data between content processes and
the parent process
* we save memory because we don't need to allocate the shared memory buffers
* annotations are faster because we don't stream them all out every time one
changes
* we won't truncate annotations anymore if we run out of space in the shared
segment.
* we don't need delayed annotations anymore, so we can get rid of the
associated machinery
As I refactored the code I tried to adjust all the obsolete comments,
consolidate shared code and remove the redundant steps that were sometimes
present. In many places we had two entire crash annotation tables we merged to
change just a couple; that comes from the fact that historically we loaded
them from disk. Now it doesn't matter anymore and we can just go ahead and
change the ones we care about.
Differential Revision: https://phabricator.services.mozilla.com/D62586
--HG--
extra : moz-landing-system : lando
Crash annotations in content processes are currently sent over IPC via
shared memory buffers. To pave the way for the Rust rewrite of the exception
handler we are removing this code and gathering all the crash annotations
within the content processes themselves. This patch causes annotations to be
stored in the global table of each content process. They are then streamed
out to the parent process by the exception handler together with the
exception-time annotations.
This has a number of benefits:
* we have one less channel to exchange data between content processes and
the parent process
* we save memory because we don't need to allocate the shared memory buffers
* annotations are faster because we don't stream them all out every time one
changes
* we won't truncate annotations anymore if we run out of space in the shared
segment.
* we don't need delayed annotations anymore, so we can get rid of the
associated machinery
As I refactored the code I tried to adjust all the obsolete comments,
consolidate shared code and remove the redundant steps that were sometimes
present. In many places we had two entire crash annotation tables we merged to
change just a couple; that comes from the fact that historically we loaded
them from disk. Now it doesn't matter anymore and we can just go ahead and
change the ones we care about.
Differential Revision: https://phabricator.services.mozilla.com/D62586
--HG--
extra : moz-landing-system : lando
Adds support for creating and using a PSandboxTesting actor in the GPU process.
Differential Revision: https://phabricator.services.mozilla.com/D42386
--HG--
extra : moz-landing-system : lando
This will allow to remove gfxPrefs later. On Windows in particular, the need to decide gfxPrefs vs StaticPrefs for the WMF decoders has caused several bugs in the past.
We will remove the confusion as a consequence.
Differential Revision: https://phabricator.services.mozilla.com/D30589
--HG--
extra : moz-landing-system : lando
This will allow to remove gfxPrefs later. On Windows in particular, the need to decide gfxPrefs vs StaticPrefs for the WMF decoders has caused several bugs in the past.
We will remove the confusion as a consequence.
Differential Revision: https://phabricator.services.mozilla.com/D30589
--HG--
extra : moz-landing-system : lando
This will allow to remove gfxPrefs later. On Windows in particular, the need to decide gfxPrefs vs StaticPrefs for the WMF decoders has caused several bugs in the past.
We will remove the confusion as a consequence.
Differential Revision: https://phabricator.services.mozilla.com/D30589
--HG--
extra : moz-landing-system : lando
The new struct is in LayersTypes.h, all the rest of the changes are just
replacing existing uint64_t instances with the new LayersId struct.
Note that there is one functional change, in
CompositorBridgeParent::DeallocPWebRenderBridgeParent, where we now
correctly convert the PipelineId to a LayersId before using it to index
into sIndirectLayerTrees, whereas before we were incorrectly just using
the mHandle part of the PipelineId.
MozReview-Commit-ID: GFHZSZiwMrP
--HG--
extra : rebase_source : d2b274f63aaee2ee9bba030297e0a37a19af0d6c
This just adds the boilerplate that goes with the new protocol, without
adding any of the actual messages. The protocol is managed by PGPU, and
there will be one instance per compositor. The parent side lives on the
main thread of the GPU process, and the child side lives on the main
thread of the UI process. The protocol is only instantiated if the GPU
process is active.
MozReview-Commit-ID: J4VzwmEfYTa
--HG--
extra : rebase_source : 397ddda8b0e76e5ed5f63783b1220ed7b4414d99
BHRTelemetryService only runs in the parent process (and we can only submit
pings from there), so we need to send the data which we collect in the GPU and
Content processes over IPC to the parent process.
MozReview-Commit-ID: 8B5uZKbjNbU
BHRTelemetryService only runs in the parent process (and we can only submit
pings from there), so we need to send the data which we collect in the GPU and
Content processes over IPC to the parent process.
MozReview-Commit-ID: 8B5uZKbjNbU
It is possible in extreme cases that the Telemetry IPC Accumulator might be
starved long enough that it cannot drain its stored accumulations for a while.
Once we hit 5x the high water mark, start discarding data.
Count each piece and type of discarded data and report it via a custom IPC
message.
MozReview-Commit-ID: JayRpa5QPec
--HG--
extra : rebase_source : eded449c932ced64bdb64e9da1a989407f4f2daf
Andrew Osmond
Jean-Yves Avenard
Mihai Alexandru Michis
Simon Giesecke
Jed Davis
Andrew McCreight
Gabriele Svelto
Brindusan Cristian
Bogdan Tara
Dan Glastonbury
Razvan Maries
Srujana Peddinti
Bas Schouten
Matt Woodrow
Gurzau Raul
Alex Gaynor
Daosheng Mu
Kartikaya Gupta
Nika Layzell
sotaro
FSeita
Michael Layzell
Sebastian Hengst
David Anderson
Carsten "Tomcat" Book
Chris H-C
Wes Kocher
Markus Stange