Do not open the input device when opening the input stream but
only when the input stream starts. This avoids blocking the thread
executing openInputStream() on the pcm input driver mutex while this mutex
is constantly acquired and released by the high priority audio
input thread.
It also allows an app to open an input stream while another thread is recording
if it does not use it immediately (useful for voice search and VoIP).
Also did some clean up in output and input stream standby logic.
Change-Id: I69c82f5833f7a034d828001f87e02745d7683b65
Remember, the system and main logs are
- Shared resources
- Primarily for recording problems
- To be used only for large grained events during normal operation
Bug: 3104855
Change-Id: I1abde2ae74119ce8ffa69c8b6e8ff278c2c8029e
Merge commit '15a9e53fbeaf765bfa6ed795d270e5968b30acb8'
* commit '15a9e53fbeaf765bfa6ed795d270e5968b30acb8':
reject odd widths or heights for YV12 format
Merge commit 'ad20ef6be061091a95353fbf71712a3a97618304'
* commit 'ad20ef6be061091a95353fbf71712a3a97618304':
don't revert to ashmem if HW usage bits are set
Merge commit 'a91a220ae435832209702ec59040036d276b63ca'
* commit 'a91a220ae435832209702ec59040036d276b63ca':
only set the PRIV_FLAGS_NEEDS_FLUSH flag on PMEM buffers
this wouldn't make sense, since the h/w wouldn't be able
to use the buffer. in this case it it better to fail to
let the caller a change to try again with different usage bits.
Change-Id: Ibecaf069b6b58ee9c026affc95a45512660d5881
PRIV_FLAGS_NEEDS_FLUSH will trigger an ioctl call into the PMEM
driver in unlock(), so it makes not sense to set this flag for
non pmem buffers.
Change-Id: Ie1e077c10df86d3689b82fa6ce5d6c856fc95688
Merge commit '249a65a3a132ff136a7a529f1cf4a13afde9a272'
* commit '249a65a3a132ff136a7a529f1cf4a13afde9a272':
In the car dock, when playing media through the 3.5mm jack, ensure
Merge commit '972fb8879c46f774daec656b3b1c1f070f04ee3e'
* commit '972fb8879c46f774daec656b3b1c1f070f04ee3e':
Change the constant name identifying the NV21 Adreno format.
Merge commit 'a526b0683dba50dac8a37ed45e3a08536044e973'
* commit 'a526b0683dba50dac8a37ed45e3a08536044e973':
Patch from HTC: fix improper device ID selection for BT with noise
suppression off. This patch is accompanied by updated ACDB files
that define the matching acoustic configuration.
Change-Id: I8cf769c30993e36c4dc07b8df8d234e752d26ffb
This patch causes librpc to abort() when a read() fails on the ONCRPC
device. We're seeing this at bootup when the RIL initializes the WMS
(SMS) service, and we can't get back in sync with the protocol. By
calling abort(), rild will restart and the RIL will work correctly
on the second attempt.
We can remove this after we fix the underlying protocol error, but perhaps
we should keep it in since the original version of the code will
continue on a read() error and corrupt memory, thinking that it
received a new message when the input buffer contains only the
previous message received.
Change-Id: I0deded00f5c38a71aa56ce1859a8fe37d234e11b
- Change debug macros to use Android logcat functions.
- Compile with optimization (remove "-O0" from CFLAGS).
Change-Id: Ibf327513a36730e7667ede45966dba6fd3bc5a5d
The purpose of this change is to add support for allocating gralloc buffers
from either /dev/pmem or /dev/pmem_adsp depending on the usage flags. It does
this by factoring out and abstracting the interactions with the pmem device.
For /dev/pmem allocations, the kernel allocator is not used, so a single master
fd is opened, and all the allocations are sub-allocated from that by gralloc.
For /dev/pmem_adsp the kernel allocator is used, so it simply opens a new fd
for each allocation.
A very basic unit test that can be run on the host is included. It requires
gtest, so to run it on a host system gtest must (currently) be compiled with
BUILD_WITH_ASTL=true.
Change-Id: If2ae0151698fad8107e18e808a3fa012a846263f
