Also includes simple tests for it + disambiguates
including hash_function.h throughout the stack.

This will allow us to do deeper equality on things like
bluetooth addresses where the actual pointers are different
but the values of the bluetooth addresses are the same.

The callback in BTA was null, so it never was actually
used.

The core security manager stores whether we have link keys
for each peer. This other call up to the bta device manager
to see if it has link keys is a dead end and does nothing.

This script helps simplify pushing and running bluedroid unit tests.
You can run all of the known ones, or provide names of selected unit
test executables you wish to run.

Make this check automatic and print out a friendly message for reminder.

These macros should replace ALOG* and the various trace macros
used throughout bluedroid. This change eliminates all uses of the
ALOG* macros in favor of the new ones.

This is a great way to run controlled audio quality experiments for
the Handsfree profile.

The code enumerating the list of possible error codes is not
a C enum and the original authors skipped a value. As a result, the
table mapping error codes to strings is off by one after that skipped
value. I'm inserting a string for that error code even though it's not
defined so the strings line up with the actual error.

If the SDP database is mutated while a client is fetching SDP records,
we can end up in an infinite request/response loop. Specifically, if
an SDP record is deleted after a client has received the first fragment
of a multi-fragment response, the server will not be able to complete
the request. Instead, it will return the same continuation token back
to the client which will request the next fragment and the server,
again, will return the same continuation token. This process repeats
forever, resulting in a large amount of unncessary BT traffic and
power costs on both devices.

Unfortunately, this seems to be a design flaw in the current SDP
server implementation. This change simply detects the above condition
and aborts the transaction entirely by returning an error to the
client. Future changes will eliminate this class of problem entirely.

Arrays are useful for storing a collection of small objects where
linked list node allocations would be a lot of overhead. They are
also useful for random access through the container. Lastly, this
implementation assumes copy semantics which are intentionally
different from list's reference semantics (since objects are assumed
to be smaller).

The current interface is intentionally sparse.

Bug: 18324148

Allocation test harness included relatively local
header allocation_tracker.h.  Change to include
absolute header location from bluedroid root.

This change will result in a constant increase in memory utilization
per allocation but will help us catch memory errors earlier.

rcv_hold_q -> rcv_pending_q

soc_queue

- Alphabetize include files
- Add asserts
- Ensure static funtions are marked as such
- Remove unnecessary logging

Ensure we hand off the node iterator before
proceeding to remove from the list.

Also added PAN API

Hopefully this will squash the sem_t semaphore_t collision

It looks like we're somehow getting NULL back from buffer allocation.
Backing GKI buffers with malloc should fix this.

Bug: 18324148

ASSERTC isn't a real assert. It just prints a message and allows the
code to carry on. Switch to asserts so the exception condition is
evident in the crash log.

Make the profile queue use osi_malloc/free instead of gki buffers, so it
can't drain the buffer pool.

If the hci layer ever entered the ignore state,
synchronization death was immenent. This patch fixes that.
Also adds additional info print out when entering the ignored state.

Adds tests for the ignored state.

Doesn't solve the problem of why we were getting NULL back from
the buffer allocator, though. (The trigger for the ignored state.)

Bug: 18156298