HasSPS, ExtractExtraData and CompareExtraData have nothing to do with the handling of annex B format. They are raw H264 related methods.
It will also prevent in the following change to have cycling references between two headers.
MozReview-Commit-ID: FCs5aU4GcTU
--HG--
extra : rebase_source : a96fe0c70416d38690b0c2f1dee567b0b025e947
HasSPS, ExtractExtraData and CompareExtraData have nothing to do with the handling of annex B format. They are raw H264 related methods.
It will also prevent in the following change to have cycling references between two headers.
MozReview-Commit-ID: FCs5aU4GcTU
--HG--
extra : rebase_source : b204723cdbb599d4f0a227871ed28f5da39e9cff
On Linux some implementations of time(0) appear to be suffering from integer
overflow and giving us the wrong dates. This causes the time we expose to the
CDM to be wrong, and so licenses passed to the CDM are failing to authenticate,
and Netflix is thus broken on some Linux systems.
This is only happening in Firefox 54 and earlier, as in those versions we use
the WidevineDecryptor to talk to the CDM. In 55 (in beta) and later we use the
PChromiumCDM protocol to talk to the CDM. This doesn't use time(0) to get a
time for the CDM, so it's immune to the problem here.
So this patch makes the GetCurrentWallTime() implementation in
WidevineDecryptor match the code currently being used on Nightly and Beta in
the ChromiumCDMChild::GetCurrentWallTime() function.
Since we use the PChromiumCDM protocol to talk to the CDM on Nightly and Beta
by default, the WidevineDecryptor isn't actually being used on Nightly and
Beta. So this patch will only cause a behaviour change in Release, which still
uses the old backend. However it will make Release run the same code that we're
running in Nightly and Beta, so it should be safe to uplift to Release.
MozReview-Commit-ID: J58iDyinyQG
--HG--
extra : rebase_source : dcdf4a846f7b007526aa626db24598942f13f01d
Includes re-importing gyp files removed from upstream in v56, and then
updating them to match the BUILD.gn file changes.
--HG--
rename : media/webrtc/trunk/webrtc/call/audio_send_stream.cc => media/webrtc/trunk/webrtc/call/audio_send_stream_call.cc
The next version of the Widevine CDM (970) has a new H.264 decoder and it does
not appear to be outputing frames it decodes in presentation order, so we need
to reorder the frames output by the CDM.
MozReview-Commit-ID: HMsQVN3NCIU
--HG--
extra : rebase_source : 68ef406556087434fa12b72ae5ed5c2e1bce2b64
The sandbox blocks loading of GMPs when the GMP resides in a directory stored
in a path which contains a symlink or junction point. So resolve GMP paths
fully before instantiating the GMP process.
MozReview-Commit-ID: EvPCpNIDNwg
--HG--
extra : rebase_source : 7df8236c9988a674ae128faf63b949fd711ab4e0
This will enable us to pre-allocate the correct number of shared memory buffers
that we pre-allocate for sending video frames between the CDM and Gecko. That
means we won't need to take the slow path to recover from underestimating how
many shmems we need.
MozReview-Commit-ID: Q4mX2rYMz3
--HG--
extra : rebase_source : f9573cfbf4e65013803b46c0909be2c68566e512
This function is arguably nicer than calling NS_ProcessNextEvent
manually, is slightly more efficient, and will enable better auditing
for NS_ProcessNextEvent when we do Quantum DOM scheduling changes.
This allows more use of the implicit version of InvokeAsync() without specifying the storage types explicitly.
MozReview-Commit-ID: 40WisaVX8Jy
--HG--
extra : rebase_source : ba34515788f0bc8264fac9a6897e234966d8b762
extra : source : b651963fe562755c0b2998ae6a95ffad400060ad
The ChromiumCDMParent is informed of the shutdown of its plugin, so we can
use that to inform the CDMProxy that its connection to the CDM has been
severed. This means we shutdown cleanly if the browser closes while playing.
MozReview-Commit-ID: HphQ2exu1gj
--HG--
extra : rebase_source : ff9ee3699915e8b7527570e839eb3bb0a0ab46bc
We are pre-allocating shmems in the content process for use by the CDM in the
GMP process. We guess the size of shmem required. However if we guess wrong,
currently we always end up taking the non-shmem path for video frames to
return to the content process, which results in us sending another shmem
(of the wrong size) to the CDM, and this continues until we hit the limit
on the number of shmems that we tolerate the CDM asking for.
So in this patch, I change our behaviour to detect when we're allocating
shmems that are too small, whereupon we purge the existing shmems and switch
to allocating them at the size being requested by the CDM.
This means we recover from incorrectly guessing the size of shmems required
by the CDM. The overhead of an incorrect guess should be one video frame
transferred via the nsTArray path.
MozReview-Commit-ID: 8o1s7FI2UBd
--HG--
extra : rebase_source : 0612d199686278612e8c58dc97e96a9304ea3ee9
The CDM process can't allocate shmems itself because it's sandboxed, so we
pre-allocate shmems in the content process and send them to the GMP process for
the CDM to use. Some sites seem to have encoded their content in such a way as
to cause the CDM to allocate and hang onto more frames than we pre-allocate; so
we run out of shmems in the GMP process, and the CDM can't allocate further
buffers to return output, and we fail.
So change the ChromiumCDMChild to allocate non-shmem-backed buffers if it runs
out of shmems, and return the result to the content process as an nsTArray.
Upon receiving that, the parent will send an extra shmem to the child, to
hopefully avoid the slow path again.
Also increase media.eme.chromium-api.video-shmems to 4, so that we're less
likely to hit this slow path in the wild. I've seen that Lightbox.co.nz and
Microsoft's EME test-drive require 4 shmems.
MozReview-Commit-ID: ISQYDkTj5uY
--HG--
extra : rebase_source : 92870f1adc7ae68e58b15443e4223012bdf0e39a
The CDM process can't allocate shmems itself because it's sandboxed, so we
pre-allocate shmems in the content process and send them to the GMP process for
the CDM to use. Some sites seem to have encoded their content in such a way as
to cause the CDM to allocate and hang onto more frames than we pre-allocate; so
we run out of shmems in the GMP process, and the CDM can't allocate further
buffers to return output, and we fail.
So change the ChromiumCDMChild to allocate non-shmem-backed buffers if it runs
out of shmems, and return the result to the content process as an nsTArray.
Upon receiving that, the parent will send an extra shmem to the child, to
hopefully avoid the slow path again.
Also increase media.eme.chromium-api.video-shmems to 4, so that we're less
likely to hit this slow path in the wild. I've seen that Lightbox.co.nz and
Microsoft's EME test-drive require 4 shmems.
MozReview-Commit-ID: ISQYDkTj5uY
--HG--
extra : rebase_source : 4beba0212411a0f5867feb506bbf732f5a934fa9
We want to replace the use of int64_t for microseconds by TimeUnit
whenever possible since int64_t is ambiguous which could be microseconds
or milliseconds.
MozReview-Commit-ID: LRz9d4yKBYJ
--HG--
extra : rebase_source : 1f73f1f338142b3183491d04726821a881ccabbe
extra : intermediate-source : 88e167b7b06303d10d92cd5317502f405d1c553e
extra : source : 98deb30ec93d395f9951f5fc488170ae35e29675
mServiceChild is a UniquePtr, so nulling it will destroy the GMPServiceChild,
which will destroy the associated message channel. So we need to close the
channel first before it gets destroyed. (Just as it was correctly done in
Observe() above.)
MozReview-Commit-ID: INuHN2Is7bC
--HG--
extra : rebase_source : 2a927bb06dd8fb4f1114dc0b64025cbdddc7c133
Makes transfer of samples between the content and CDM processes use shmems.
The Chromium CDM API requires us to implement a synchronous interface to supply
buffers to the CDM for it to write decrypted samples into. We want our buffers
to be backed by shmems, in order to reduce the overhead of transferring decoded
frames. However due to sandboxing restrictions, the CDM process cannot allocate
shmems itself. We don't want to be doing synchronous IPC to request shmems
from the content process, nor do we want to have to do intr IPC or make async
IPC conform to the sync allocation interface. So instead we have the content
process pre-allocate a set of shmems and give them to the CDM process in
advance of them being needed.
When the CDM needs to allocate a buffer for storing a decrypted sample, the CDM
host gives it one of these shmems' buffers. When this is sent back to the
content process, we copy the result out (uploading to a GPU surface for video
frames), and send the shmem back to the CDM process so it can reuse it.
We predict the size of buffer the CDM will allocate, and prepopulate the CDM's
list of shmems with shmems of at least that size, plus a bit of padding for
safety. We pad frames out to be the next multiple of 16, as we've seen some
decoders do that.
Normally the CDM won't allocate more than one buffer at once, but we've seen
cases where it allocates two buffers, returns one and holds onto the other. So
the minimum number of shmems we give to the CDM must be at least two, and the
default is three for safety.
MozReview-Commit-ID: 5FaWAst3aeh
--HG--
extra : rebase_source : a0cb126e72bfb2905bcdf02e864dc654e8340410
This means we can pass anything that converts implicitly to a Span to
PostResult, including an nsTArray<uint8_t>. We can also pass a Span
that contains the contents of a Shmem's buffer.
MozReview-Commit-ID: 8AAcRmVCEVy
--HG--
extra : rebase_source : 44dfbc465db14bb689a653e6c0b3cbc626c0a0d1
This ensures that the IPC connection from the content process to the main
process is shut down as soon as possible. Once all the IPC connections are
closed, the main process removes its async shutdown blocker, and Firefox
can shutdown.
MozReview-Commit-ID: 8rqa384ayd9
--HG--
extra : rebase_source : b9cbbb9f4c22016284a8d49cddaea0d96666acf9
This ensures that when we've started shutdown we don't try to start up new
GMPs. Doing so would create more connections from the content process to the
main process, and the main process can't shutdown until all such connections
are shut down.
MozReview-Commit-ID: KE8nCoLXjdd
--HG--
extra : rebase_source : 674f3c4ddcb5bb93dd775a861b425d25510871e9
This will allow us to broadcast a notification to the GMPServices running in
the content processes when they need to shutdown.
MozReview-Commit-ID: FviFDgNMnUV
--HG--
extra : rebase_source : f4ad3c6df0e14c88db1199fbe6281d67f98590ae
When we shutdown the browser while the GMPService is active we can end up
leaking a GMPParent, GeckoMediaPluginServiceParent, and a Runnable. I tracked
this down to the runnable dispatched to the GMP thread in
GMPParent::ChildTerminated(). The dispatch of this runnable is failing as we
are dispatching the runnable to a reference of the GMP thread which we have
previously acquired, but that thread is now shutdown. So the dispatch fails,
and if you look in nsThread::DispatchInternal() you'll see that we deliberately
leak the runnable if dispatch fails! The runnable leaking means that the
references it holds to the GMPParent and the GMP service parent leak.
The solution in this patch is to not cache a reference to the GMP thread on the
GMPParent; instead we re-request the GMP thread from the GMPService when we
want it. This means that in the case where the browser is shutting down,
GMPParent::GMPThread() will return null, and we'll not leak the runnable. We'll
then follow the (hacky) shutdown path added in bug 1163239.
We also need to change GMPParent::GMPThread() and GMPContentParent::GMPThread()
to return a reference to the GMP thread with a refcount held on it, in order
to ensure we don't race with the GMP service shutting down the GMP thread
while we're trying to dispatch to in on shutdown.
MozReview-Commit-ID: CXv9VZqTRzY
--HG--
extra : rebase_source : e507e48ee633cad8911287fb7296bbb1679a7bcb
This is required for the browser clearing persistence tests to pass.
MozReview-Commit-ID: Ai9qc6Ds1IG
--HG--
extra : rebase_source : 80c2133e26742410fda983e3c18c35736fc013d0
This severs the ChromiumCDMVideoDecoder's connection with the CDM. The CDM process
will shutdown when the MediaKeys also severs its connection.
MozReview-Commit-ID: Aqc4y5Nxjvc
--HG--
extra : rebase_source : 5a2f77ffe84f9b99b4668520c838b29a428578d3
At this stage, I store video frames in memory in nsTArrays rather than in
shmems just so we can get this working. Once this is working, I'll follow up
with patches to switch to storing all large buffer traffic between the CDM and
other processes in shmems.
I'm not planning on preffing this new CDM path on until that's in place.
MozReview-Commit-ID: LSTb42msWQS
--HG--
extra : rebase_source : b7f162515a1a32b2c344c11d0fa5c7004cec2e15
