to top

RIL Refactoring

In this document

Introduction

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.

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

  • Enhanced RIL error codes: Code can return more specific error codes than the existing GENERIC_FAILURE code. This enhances error troubleshooting by providing more specific information about the cause of errors.
  • Enhanced RIL versioning: The RIL versioning mechanism is enhanced to provide more accurate and easier to configure version information.
  • Redesigned RIL communication using wakelocks: RIL communication using wakelocks is enhanced to improve device battery performance.

Examples and source

Documentation for RIL versioning is also in code comments in https://android.googlesource.com/platform/hardware/ril/+/master/include/telephony/ril.h.

Implementation

The following sections describe how to implement the subfeatures of the RIL refactoring feature.

Implementing enhanced RIL error codes

Problem

Almost all RIL request calls can return the GENERIC_FAILURE 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 GENERIC_FAILURE 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 GENERIC_FAILURE code.

Solution

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

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 OEM_ERROR_1 to OEM_ERROR_X. 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 GENERIC_FAILURE, and sometimes it's impossible to figure out.

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

Implementing enhanced RIL versioning

Problem

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.

Solution

There is a documented section in ril.h 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 RIL_REGISTER.

Implementing redesigned RIL communication using wakelocks

Problem summary

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:

  • synchronous: Those that do not take considerable time to respond back. For example, RIL_REQUEST_GET_SIM_STATUS.
  • asynchronous: Those that take considerable time to respond back. For example, RIL_REQUEST_QUERY_AVAILABLE_NETWORKS.

Follow these steps to implement redesigned wakelocks:

  1. Classify solicited RIL commands as either synchronous or asynchronous depending on how much time they take to respond.

    Here are some things to consider while making that decision:

    • 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.
    • 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.
  2. Understand the RIL communications scenarios described in the RIL communication scenarios section.
  3. Follow the solutions in the scenarios by modifying your code to handle RIL solicited and unsolicited requests.

RIL communication scenarios

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

Scenario 1: RIL request from Java APIs and solicited asynchronous response to that request

Problem

If the RIL solicited response is expected to take considerable time (for example, RIL_REQUEST_GET_AVAILABLE_NETWORKS), 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.

Solution part 1

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

As shown in the above sequence diagram:

  1. RIL request is sent, and the modem needs to acquire wakelock to process the request.
  2. 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.
  3. 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.
  4. When the response reaches the Java side, wakelock is acquired and response is sent back to caller.
  5. After that response is processed by all modules, acknowledgement is sent back to ril.cpp over a socket. ril.cpp then releases the wakelock that was acquired in step 3.

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.

Solution part 2

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

As shown in the above sequence diagram:

  1. RIL request is sent by calling acquireWakeLock() on the Java side.
  2. Vendor code doesn't need to acquire wakelock and can process the request and respond quickly.
  3. When the response is received by the Java side, decrementWakeLock() is called, which decreases wakelock counter and releases wakelock if the counter value is 0.

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

Scenario 2: RIL unsolicited response

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.

Problem

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

Solution

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

Validation

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

Validating enhanced RIL error codes

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

Validating enhanced RIL versioning

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

Validating redesigned wakelocks

Verify that RIL calls are identified as synchronous or asynchronous.

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.