GitLab

This CL adds GN build files for libchrome, which is the ChromeOS/Brillo library
for many C++ tools, include libbase. I added a new GN build target for "base"
under secondary sources with a GN build file for libchrome which currently lives
in platform/external/libchrome in AOSP.

To build, clone libchrome into third_party/libchrome and remove
third_party/libchrome/base/BUILD.gn. Also clone external/modp_b64 into
third_party/modp_b64. Eventually I will have scripts set up to do
this automatically.

This means that we can start developing against libbase while running GN builds
on Goobuntu but this would break the tree until libchrome is buildable within
the Android build system (http://r.android.com/158392). So Brillo/Android builds
will depend on libchrome using the conventional Android.mk means.

Bug: 22175181
Change-Id: I9a737aa785488e5bc0ca1adc087d1d5dfcdf3bea

Update the LOG_* functions to take a tag argument which makes them more
consistent with the Android Log.*(TAG, s) common syntax and removes
some #define-dependency with osi/include/log.h.

Also update to never use Android log functions directly.

Also contains minor cleanup of some header includes.

Bug: 21569831
Change-Id: If07385cafbea062232ecdbc7c673f908d5ef8921

Eliminate cutils/sockets dependencies by pulling sources files from
core/libcutils into osi/.
   - Pulled source files from cutils/ into osi/ and modified GN files and
     Android.mk
   - Updated includes to use the headers in osi/ and removed unused/duplicated
     headers.
   - Renamed the functions of osi/sockets and updated wherever they are called
     to avoid usign symbols from cutils/sockets, since other cutils
     dependencies have not been eliminated yet.

Bug: 21667795
Change-Id: I3dd4371d585f120d97ac0433ab1f35edb820dbb9

This patch improves RFCOMM throughput and reduces CPU utilization.
Instead of using a counting semaphore to measure bytes and reading
one at a time from the eager reader's buffer, read in bulk based
on the incoming read request size.

Change-Id: I17046bfbc3ca49576a9c82b38911aeb84234881a

This CL introduces macros for safe pointer to integer conversion (and vice
versa). Also fixed a small style issue in log.h

Bug: 21570302
Change-Id: If76bf5e35970f9b33f9bef53fbd03a7effae08dc

Change-Id: I007b2211d9b52422b4442765bb97c0ba90613c50

Also, removed the OSI module's start_up and shut_down steps,
because now they are no-op.

Bug: 21558791
Change-Id: I24259b327f399af57c37937111158baa9704f644

Temporary disable the allocation tracker initialization
(even for BLUEDROID_DEBUG) when initializing Bluetooth.
This is a short-term workaround solution for several issues related to
the usage of the allocation tracker.

* Inconsistent usage of osi_malloc()/osi_calloc() and osi_free()
  - Within some of the Bluetooth-related unit tests we have two copies
    of the same libosi library: one copy statically linked against the
    unit test, and another dlopen() at runtime as part of bluetooth.default.so
    Each of those copy has its own static variables.
  - For the dlopen() copy we do call allocation_tracker_init(), while for the
    static copy within the unit test we don't call allocation_tracker_init()
  - Occasionally, there is a  memory allocation via osi_calloc()
    within the dlopen()-ed library, and then it is osi_free()-ed
    within the statically linked library.
 Such (mis)usage creates issues in two ways: (1) free()-ing incorrect
 pointer, and (2) the osi hash_map() in the dlopen()-ed library still considers
 the osi-tracked memory as allocated.
 NOTE: (1) could trigger random crashes, while (2) is the reason the unit
 test triggers an assert

* Avoid potential issues that could result from the fact that calling
  allocation_tracker_init() is not mandatory; i.e., the same
  issue described above could be triggered if osi_malloc()/osi_callod()
  was called before the call to allocation_tracker_init()

* There is still code that uses malloc(3)/calloc(3) and free(3) instead
  of osi_malloc()/osi_calloc() and osi_free()

Also, add missing pthread_mutex_lock()/pthread_mutex_unlock() guards,
and fix the allocation_tracker_uninit() implementation so it works
properly even if void allocation_tracker_init() wasn't called.

Bug: 21561735
Change-Id: Ic83d6cd40af1189c4ee9c1dbfd0ad8e4666e1502

Now the OSI module's shut_down processing is split into "shut_down"
and "clean_up".

Previously, there was an ordering issue manipulating some of the
internal state during graceful shutdown/cleanup. Some of the modules
had two steps: shut_down, followed by clean_up, while other had only
shut_down step. This triggered the following assert in file alarm.c

    alarm_cancel: assertion "alarms != NULL" failed

Bug: 21406940
Change-Id: Iab1f033a69cbff646a6b0f346760ae82f8b00b8f

Ensures all alarm resources are torn down before the
test ends, so the allocation tracker is happy and
doesn't assert-fail the tests.

Also guard alarm_shutdown to make sure it only runs
if lazy_initialize was run (to avoid undefined behavior)

Change-Id: Iaf78fe85edd2fc65b9c5ee11c74665a11a3bc9ba

Change-Id: Iecf1e2258da012bdac69a4f57d38b12a272e3edd

Includes support for BTSnoop logging in memory.

Bug: 18508263
Change-Id: I175da528cbcdc00d40622647d518a74210cfe6fd

Bug: 20884459
Change-Id: Id010cb29cf09b95800dc9f5fd5b660e71e4a47bc

Removed unnecessary library and include flags from build files. To make the
initial build outputs more similar to Android, this CL changes the usage of the
GN "source_set" idiom to "static_library". This comes at a performance loss
while keeping the builds similar until we get the whole stack to compile.

BUG=21339022

Change-Id: I46f70b3b64b2537634a1dbb517a5168566230efb

Added tinyxml2 as a secondary source. Moved //build/secondary/testing to
//build/secondary/third_party in the process. This means that all third_party
sources (gtest, tinyxml2) should be placed in a top-level third_party directory.

BUG=21339022

Change-Id: If324ef1eb0d5061e005f62a80d23fef25e7e78a4

Bug: 20427982
Change-Id: I003469d6dfce681a38baa71b5709236a1a4a2189

This CL adds support to build the net_test_osi target using GN + ninja.
The BUILD.gn file now pulls in gtest as a dependency. Missing libc
includes have been added to sources that prevented compilation on
Goobuntu (14.04).

The osi/include/log.h header has been modified to conditionally call the
Android log utilities vs plain-old printf on non-Android builds.

BUG=21339022

Change-Id: If3c356360c56c63b3bf551b41dc8d3269d8e7e09

BUG=21339022

Change-Id: I946fad7bf8116054f9081e064748793e79a6f85a

BUG=20491586

These patches will allow us to compile the whole AOSP with clang/llvm
before all the following problems are fixed.
* Suppress warnings on unused variables.
  * Remove unused static variables.
* Use only gcc to compile code that requires gcc atomic functions.
  They should be converted to standard atomic functions later.
* Suppress warnings on redefined typedefs.

BUG: 20765701
Change-Id: Ib1d6e2ef86c255ba87c1d51f9fdeabe5c8de2ada
(cherry picked from commit 23546918)

- This was a clang complaint, not a gcc complaint
- memset event to be portable to all compile environments
- error as follows:

external/bluetooth/bluedroid/osi/./src/reactor.c:90:34:
    error: missing field data initializer [-Werror,-Wmissing-field-initializers]
  struct epoll_event event = { 0 };
                                 ^
external/bluetooth/bluedroid/osi/./src/reactor.c:151:34:
    error: missing field data initializer [-Werror,-Wmissing-field-initializers]
  struct epoll_event event = { 0 };
                                 ^
external/bluetooth/bluedroid/osi/./src/reactor.c:174:34:
    error: missing field data initializer [-Werror,-Wmissing-field-initializers]
  struct epoll_event event = { 0 };
                                 ^
3 errors generated.
make: *** [out/target/product/hammerhead/obj/STATIC_LIBRARIES/libosi_intermediates/./src/reactor.o] Error 1

Change-Id: I14b884cf1665c68aff1e5afd3df65ac10af474d2

Change-Id: Ia856311187a0b2168cb859bd8b2300ff59df6569

Change-Id: I56cca343e07df83999b0ef0fb672dc8bcfae4bc4

glibc doesn't declare strlcpy(), strlcat() or gettid(). Add a header
of compatibility routines that only get included when compiling on this.

Change-Id: I889ac7f1d22cf7056757890fcff0403107488d62

'inline' is only a hint for the compiler, and if ignored this results
in extern functions being created. Add 'static' so that the resulting
functions are also only local to the compilation unit.

Change-Id: I947201b8170c8c73f15e7ccdd68cfc7f07cde1e9

Some standard library functions are used while relying on bionic headers
including the headers the functions were actually declared on. Add those
missing #includes so that bluetooth.default.so will compile on glibc.

Change-Id: Ied9f89ce5a05911fca63f6bfe1b8cc8196ab3b1c

This change uses generic format specifier for logging the
pointers which can be 64 bit or 32 bit depending upon the
system used for compilation.
Also address of operator is used before casting a int variable
to void pointer.

Change-Id: Ie9dec408f3a5f674079f77c047d8192f65a004ca

Change-Id: I0d3e93ea61dd03505fe9db902ed90bdd4141cab2

This change uses generic format specifier for logging the
pointers which can be 64 bit or 32 bit depending upon the
system used for compilation.
Also address of operator is used before casting a int variable
to void pointer.

Change-Id: Ie9dec408f3a5f674079f77c047d8192f65a004ca

If we're rescheduling a periodic timer in the context of an alarm
expiration we want to force rescheduling to happen.

The alarm at the front of the list after scheduling the next
instance may be a different alarm. The old code would assume it was
already scheduled correctly, meaning nothing would ever get scheduled
after that.

Change-Id: I8accf5d004e69dfd6477ab7ec529eea42b3a88f8

Turns out the posix timers we're dealing with aren't well behaved.

If the timer is TIMER_ABSTIME the following is supposed to be true:
"If the specified time has already passed, the function shall succeed
and the expiration notification shall be made."

But alas, this is not the case. If the expiration time happens to be
in the past (e.g. very short timer which gets hit with a context switch
before the timer can be set) the timer decides to disarm itself. This
means no more callbacks happen, and no more alarms are processed ever.

sadness.

But thankfully, we can use timer_gettime to check the state of the timer
after timer_settime. If timer_gettime tells us the timer is disarmed
right after we armed it, we want to perform the alarm callback
ourselves.

Put all timer callbacks on the same thread (as would be the case in the
underlying timer implementation already) and used a sempahore to signal
when an alarm expires in the normal posix timer callback and also in the
timer didn't set case.

Change-Id: I39854b241369b2da52afbaaba0c01d3167a47e32

This means we lose the ability to send the intended alarm
as context with the posix timer, but this can be handled
gracefully enough by making sure we actually have an
expired alarm at the front of the list when the timer
callback occurs.

Change-Id: I909195b5c3155075096328c529655eeb5b7d80d0

If we're rescheduling a periodic timer in the context of an alarm
expiration we want to force rescheduling to happen.

The alarm at the front of the list after scheduling the next
instance may be a different alarm. The old code would assume it was
already scheduled correctly, meaning nothing would ever get scheduled
after that.

Change-Id: I8accf5d004e69dfd6477ab7ec529eea42b3a88f8

Turns out the posix timers we're dealing with aren't well behaved.

If the timer is TIMER_ABSTIME the following is supposed to be true:
"If the specified time has already passed, the function shall succeed
and the expiration notification shall be made."

But alas, this is not the case. If the expiration time happens to be
in the past (e.g. very short timer which gets hit with a context switch
before the timer can be set) the timer decides to disarm itself. This
means no more callbacks happen, and no more alarms are processed ever.

sadness.

But thankfully, we can use timer_gettime to check the state of the timer
after timer_settime. If timer_gettime tells us the timer is disarmed
right after we armed it, we want to perform the alarm callback
ourselves.

Put all timer callbacks on the same thread (as would be the case in the
underlying timer implementation already) and used a sempahore to signal
when an alarm expires in the normal posix timer callback and also in the
timer didn't set case.

Change-Id: I39854b241369b2da52afbaaba0c01d3167a47e32

This means we lose the ability to send the intended alarm
as context with the posix timer, but this can be handled
gracefully enough by making sure we actually have an
expired alarm at the front of the list when the timer
callback occurs.

Change-Id: I909195b5c3155075096328c529655eeb5b7d80d0

Also improve names of some functions.

Change-Id: Ia5aeebc8226ce5c36abdd372ea4de5195314e273

There's a race condition on shutdown between Java and native code.
A stack shutdown is issued from Java and in the meantime the
calling thread calls System.exit(0). If that line is reached before
the native code has had a chance to shut down cleanly, the system
could be stuck in a bad state.

This change uses rename(2) to replace an existing config file
atomically instead of truncating, writing, and then flushing. In
case of a crash or the race condition above, the config file will
still be valid.

There's an implementation gotcha. The config saving code uses ".new"
as a suffix on the original filename to write to a temporary file.
If a file by that name already exists, it will be overwritten.

Adds alarm_set_periodic so the alarm code can have more
contextual information when rescheduling alarms.

Problem: A2DP would stream for a few seconds and then
stop working.

Cause: The Java garbage collector. Bluedroid reaches out
to javaland to acquire and release the wake lock. Alarm was
always reaching out to get the wake lock when it scheduled a
short timeout. If GC kicked in during that call out to make
sure we have the wake lock, it could take more than 100ms to
get back us. That would screw over the alarm implementation
particularly for small 20ms timers.

So now if the wake lock was already acquired, we don't try to
reacquire it.

Cool. But we still have thrashing. Why? Because the alarm code
doesn't know the alarm is actually being used in a periodic way.

Here's what used to happen:

alarm expires
alarm is removed
reschedule
alarm callback is called
alarm callback sets the alarm again
alarm is added
reschedule

The problem is the first reschedule will get rid of the wake lock
if the next alarm is too far out or doesn't exist, meaning the next
reschedule needs to get the wake lock again.

With the extra periodicity information we can eliminate the
unnecessary intermediate reschedule, meaning no thrashing on the
wake lock. yay!