tech/connect/ril.jd

page.title=RIL Refactoring
@jd:body

<!--
    Copyright 2016 The Android Open Source Project

    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at

        http://www.apache.org/licenses/LICENSE-2.0

    Unless required by applicable law or agreed to in writing, software
    distributed under the License is distributed on an "AS IS" BASIS,
    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    See the License for the specific language governing permissions and
    limitations under the License.
-->
<div id="qv-wrapper">
  <div id="qv">
    <h2>In this document</h2>
    <ol id="auto-toc">
    </ol>
  </div>
</div>

<h2 id="introduction">Introduction</h2>

<p>The Radio Interface Layer (RIL) refactoring feature
of the Android 7.0 release is a set of subfeatures
that improves RIL functionality. Implementing the features is optional but
encouraged. Partner code changes are required to implement these features. The
refactoring changes are backward compatible, so prior implementations of
the refactored features will still work.</p>

<p>The following subfeatures are included in the RIL refactoring feature. You
can implement any or all of the subfeatures:</p>

<ul>
<li>Enhanced RIL error codes: Code can return more specific error codes
than the existing <code>GENERIC_FAILURE</code> code. This enhances error
troubleshooting by providing more specific information about the cause
of errors.</li>

<li>Enhanced RIL versioning: The RIL versioning mechanism is enhanced to
provide more accurate and easier to configure version information.</li>

<li>Redesigned RIL communication using wakelocks: RIL communication using
wakelocks is enhanced to improve device battery performance.</li>
</ul>

<h2 id="examples">Examples and source</h2>

<p>Documentation for RIL versioning is also in code comments in <a
href="https://android.googlesource.com/platform/hardware/ril/+/master/include/telephony/ril.h"><code>https://android.googlesource.com/platform/hardware/ril/+/master/include/telephony/ril.h</code></a>.</p>

<h2 id="implementation">Implementation</h2>

<p>The following sections describe how to implement the subfeatures of the
RIL refactoring feature.</p>

<h3 id="errorcodes">Implementing enhanced RIL error codes</h3>

<h4 id="errorcodes-problem">Problem</h4>

<p>Almost all RIL request calls can return the <code>GENERIC_FAILURE</code>
error code in response to an error. This is an issue with all solicited
responses returned by the OEMs. It is difficult to debug an issue from
the bug report if the same <code>GENERIC_FAILURE</code> error code is
returned by RIL calls for different reasons. It can take considerable time
for vendors to even identify what part of the code could have returned a
<code>GENERIC_FAILURE</code> code.</p>

<h4 id="errorcodes-solution">Solution</h4>

<p>OEMs should return a distinct error code value associated
with each of the different errors that are currently categorized as
<code>GENERIC_FAILURE</code>.</p>

<p>If OEMs do not want to publicly reveal their custom error codes, they may
return errors as a distinct set of integers (for example, from 1 to x) that
are mapped as <code>OEM_ERROR_1</code> to <code>OEM_ERROR_X</code>. The
vendor should make sure each such masked error code returned maps to a unique
error reason in their code. The purpose of doing this is
to speed up debugging RIL issues whenever generic errors are returned
by the OEM. It can take too much time to identify what exactly caused
<code>GENERIC_FAILURE</code>, and sometimes it's impossible to figure out.<p>

<p>In <code>ril.h</code>, more error codes are
added for enums <code>RIL_LastCallFailCause</code> and
<code>RIL_DataCallFailCause</code> so that vendor code avoids returning
generic errors like <code>CALL_FAIL_ERROR_UNSPECIFIED</code> and
<code>PDP_FAIL_ERROR_UNSPECIFIED</code>.</p>

<h3 id="version">Implementing enhanced RIL versioning</h3>

<h4 id="version-problem">Problem</h4>

<p>RIL versioning is not accurate enough. The mechanism for vendors to
report their RIL version is not clear, causing vendors to report an incorrect
version. A workaround method of estimating the version is used, but it can
be inaccurate.</p>

<h4 id="version-solution">Solution</h4>

<p>There is a documented section in <code>ril.h</code> describing what a
particular RIL version value corresponds to. Each
RIL version is documented, including what changes correspond
to that version. Vendors must update their version in code when making
changes corresponding to that version, and return that version while doing
<code>RIL_REGISTER</code>.</p>

<h3 id="wakelocks">Implementing redesigned RIL communication using
wakelocks</h3>

<h4 id="wakelocks-prob-sum">Problem summary</h4>

<p>Timed wakelocks are used in RIL communication in an imprecise way,
which negatively affects battery performance. RIL requests can be either
solicited or unsolicited. Solicited requests should be classified as one of
the following:</p>

<ul>
<li>synchronous: Those that do not take considerable time to respond back. For
example, <code>RIL_REQUEST_GET_SIM_STATUS</code>.</li>

<li>asynchronous: Those that take considerable time to respond back. For
example, <code>RIL_REQUEST_QUERY_AVAILABLE_NETWORKS</code>.</li>
</ul>

<p>Follow these steps to implement redesigned wakelocks:</p>

<ol>
<li>
Classify solicited RIL commands as either synchronous or asynchronous
depending on how much time they take to respond.
<p>Here are some things to consider while making
that decision:</p>

<ul>
<li>As explained in the solution of asynchronous solicited RIL requests,
because the requests take considerable time, RIL Java releases the wakelock
after receiving ack from vendor code. This might cause the application
processor to go from idle to suspend state. When the response is available
from vendor code, RIL Java (the application processor) will re-acquire the
wakelock and process the response, and later go to idle state again. This
process of moving from idle to suspend state and back to idle can consume
a lot of power.</li>

<li>If the response time isn't long enough then holding the wakelock and
staying in idle state for the entire time it takes to respond can be more
power efficient than going in suspend state by releasing the wakelock and
then waking up when the response arrives. So vendors should use
platform-specific power measurement to find out the threshold value of time 't' when
power consumed by staying in idle state for the entire time 't' consumes
more power than moving from idle to suspend and back to idle in same time
't'. When that time 't' is discovered, RIL commands that take more than time
't' can be classified as asynchronous, and the rest of the RIL commands can
be classified as synchronous.</li>
</ul>
</li>

<li>Understand the RIL communications scenarios described in the <a
href="#ril-comm-scenarios">RIL communication scenarios</a> section.</li>

<li>Follow the solutions in the scenarios by modifying your code to handle
RIL solicited and unsolicited requests.</li>
</ol>

<h4 id="ril-comm-scenarios">RIL communication scenarios</h4>

<p>For implementation details of the functions used in the
following diagrams, see the source code of <code>ril.cpp</code>:
<code>acquireWakeLock()</code>, <code>decrementWakeLock()</code>,
<code>clearWakeLock(</code>)</p>

<h5>Scenario 1: RIL request from Java APIs and solicited asynchronous response
to that request</h5>

<p><img src="images/ril-refactor-scenario-1.png"></p>

<h6>Problem</h6>

<p>If the RIL solicited response is expected to take considerable time (for
example, <code>RIL_REQUEST_GET_AVAILABLE_NETWORKS</code>), then wakelock
is held for a long time on the Application processor side, which is a
problem. Also, modem problems result in a long wait.</p>

<h6>Solution part 1</h6>

<p>In this scenario, wakelock equivalent is held by Modem code (RIL request
and asynchronous response back).</p>

<p><img src="images/ril-refactor-scenario-1-solution-1.png"></p>

<p>As shown in the above sequence diagram:</p>

<ol>
<li>RIL request is sent, and the modem needs to acquire wakelock to process
the request.</li>

<li>The modem code sends acknowledgement that causes the Java side to decrement
the wakelock counter and release it if the wakelock counter value is 0.</li>

<li>After the modem processes the request, it sends an interrupt to the
vendor code that acquires wakelock and sends a response to ril.cpp. ril.cpp
then acquires wakelock and sends a response to the Java side.</li>

<li>When the response reaches the Java side, wakelock is acquired and response
is sent back to caller.</li>

<li>After that response is processed by all modules, acknowledgement is
sent back to <code>ril.cpp</code> over a socket. <code>ril.cpp</code> then
releases the wakelock that was acquired in step 3.</li>
</ol>

<p>Note that the wakelock timeout duration for the request-ack sequence
would be smaller than the currently used timeout duration because the ack
should be received back fairly quickly.</p>

<h6>Solution part 2</h6>

<p>In this scenario, wakelock is not held by modem and response is quick
(synchronous RIL request and response).</p>

<p><img src="images/ril-refactor-scenario-1-solution-2.png"></p>

<p>As shown in the above sequence diagram:</p>

<ol>
<li>RIL request is sent by calling <code>acquireWakeLock()</code> on the
Java side.</li>

<li>Vendor code doesn't need to acquire wakelock and can process the request
and respond quickly.</li>

<li>When the response is received by the Java side,
<code>decrementWakeLock()</code> is called, which decreases wakelock counter
and releases wakelock if the counter value is 0.</li>
</ol>

<p>Note that this synchronous vs. asynchronous behavior is hardcoded for a
particular RIL command and decided on a call-by-call basis.</p>

<h5>Scenario 2: RIL unsolicited response</h5>

<p><img src="images/ril-refactor-scenario-2.png"></p>

<p>As shown in the above diagram, RIL unsolicited responses have a wakelock
type flag in the response that indicates whether a wakelock needs to be
acquired or not for the particular response received from the vendor. If
the flag is set, then a timed wakelock is set and response is sent over a
socket to the Java side. When the timer expires, the wakelock is released.</p>

<h6>Problem</h6>

<p>The timed wakelock illustrated in Scenario 2 could be too long or too
short for different RIL unsolicited responses.</p>

<h6>Solution</h6>

<p><img src="images/ril-refactor-scenario-2-solution.png"></p>

<p>As shown, the problem can be solved by sending an acknowledgement from
the Java code to the native side (<code>ril.cpp</code>), instead of holding
a timed wakelock on the native side while sending an unsolicited response.</p>

<h2 id="validation">Validation</h2>

<p>The following sections describe how to validate the implementation of
the RIL refactoring feature's subfeatures.</p>

<h3 id="validate-error">Validating enhanced RIL error codes</h3>

<p>After adding new error codes to replace the <code>GENERIC_FAILURE</code>
code, verify that the new error codes are returned by the RIL call instead
of <code>GENERIC_FAILURE</code>.</p>

<h3 id="validate-version">Validating enhanced RIL versioning</h3>

<p>Verify that the RIL version corresponding to your RIL code is returned
during <code>RIL_REGISTER</code> rather than the <code>RIL_VERSION</code>
defined in <code>ril.h</code>.</p>

<h3 id="validate-wakelocks">Validating redesigned wakelocks</h3>

<p>Verify that RIL calls are identified as synchronous or asynchronous.</p>

<p>Because battery power consumption can be hardware/platform dependent,
vendors should do some internal testing to find out if using the new wakelock
semantics for asynchronous calls leads to battery power savings.</p>