camera.device@3.2: Clarify documentation
- Clarify handling of input/output buffer handles, and
that they must be null in CaptureResults.
- Clarify handling of fence handles and when they may/must be null.
Test: Compiles; comment-only changes
Bug: 32991603
Change-Id: I70367aa490fb0027791e6e35bb34c54601a0d144
diff --git a/camera/device/3.2/types.hal b/camera/device/3.2/types.hal
index ed6ef7d..a7ab4ab 100644
--- a/camera/device/3.2/types.hal
+++ b/camera/device/3.2/types.hal
@@ -408,7 +408,16 @@
int32_t streamId;
/**
- * The graphics buffer handle to the buffer
+ * The graphics buffer handle to the buffer.
+ *
+ * For StreamBuffers sent to the HAL in a CaptureRequest, this must be a
+ * valid handle to a graphics buffer, with dimensions and format matching
+ * that of the stream.
+ *
+ * For StreamBuffers returned in a CaptureResult, this must be null, since
+ * the handle to the buffer is already known to the client (since the client
+ * sent it in the matching CaptureRequest), and the handle can be identified
+ * by the combination of frame number and stream ID.
*/
handle buffer;
@@ -424,44 +433,34 @@
* The acquire sync fence for this buffer. The HAL must wait on this fence
* fd before attempting to read from or write to this buffer.
*
- * The framework may be set to -1 to indicate that no waiting is necessary
- * for this buffer.
+ * In a buffer included in a CaptureRequest, the client may set this to null
+ * to indicate that no waiting is necessary for this buffer.
*
- * When the HAL returns an output buffer to the framework with
- * processCaptureResult(), the acquireFence must be set to -1. If the HAL
- * never waits on the acquireFence due to an error in filling a buffer,
- * when calling processCaptureResult() the HAL must set the releaseFence
- * of the buffer to be the acquireFence passed to it by the framework. This
- * allows the framework to wait on the fence before reusing the buffer.
- *
- * For input buffers, the HAL must not change the acquireFence field during
- * the processCaptureRequest() call.
- *
- * When the HAL returns an input buffer to the framework with
- * processCaptureResult(), the acquireFence must be set to -1. If the HAL
- * never waits on input buffer acquire fence due to an error, the sync
- * fences must be handled similarly to the way they are handled for output
- * buffers.
+ * When the HAL returns an input or output buffer to the framework with
+ * processCaptureResult(), the acquireFence must be set to null. If the HAL
+ * never waits on the acquireFence due to an error in filling or reading a
+ * buffer, when calling processCaptureResult() the HAL must set the
+ * releaseFence of the buffer to be the acquireFence passed to it by the
+ * client. This allows the client to wait on the fence before reusing the
+ * buffer.
*/
handle acquireFence;
/**
- * The release sync fence for this buffer. The HAL must set this fence when
- * returning buffers to the framework, or write -1 to indicate that no
- * waiting is required for this buffer.
+ * The release sync fence for this buffer. The HAL must set this to a valid
+ * fence fd when returning the input buffer or output buffers to the client
+ * in a CaptureResult, or set it to null to indicate that no waiting is
+ * required for this buffer.
*
- * For the output buffers, the fences must be set in the outputBuffers
- * array passed to processCaptureResult().
- *
- * For the input buffer, the fences must be set in the inputBuffer
- * passed to processCaptureResult().
+ * The client must set this to be null for all buffers included in a
+ * processCaptureRequest call.
*
* After signaling the releaseFence for this buffer, the HAL
* must not make any further attempts to access this buffer as the
- * ownership has been fully transferred back to the framework.
+ * ownership has been fully transferred back to the client.
*
- * If a fence of -1 was specified then the ownership of this buffer
- * is transferred back immediately upon the call of processCaptureResult.
+ * If this is null, then the ownership of this buffer is transferred back
+ * immediately upon the call of processCaptureResult.
*/
handle releaseFence;
@@ -722,13 +721,13 @@
/**
* The input stream buffer to use for this request, if any.
*
+ * An invalid inputBuffer is signified by a null inputBuffer::buffer, in
+ * which case the value of all other members of inputBuffer must be ignored.
+ *
* If inputBuffer is invalid, then the request is for a new capture from the
* imager. If inputBuffer is valid, the request is for reprocessing the
- * image contained in inputBuffer.
- *
- * In the latter case, the HAL must set the releaseFence of the
- * inputBuffer to a valid sync fence, or to -1 if the HAL does not support
- * sync, before processCaptureRequest() returns.
+ * image contained in inputBuffer, and the HAL must release the inputBuffer
+ * back to the client in a subsequent processCaptureResult call.
*
* The HAL is required to wait on the acquire sync fence of the input buffer
* before accessing it.
@@ -741,9 +740,8 @@
* data from this capture/reprocess. The HAL must wait on the acquire fences
* of each stream buffer before writing to them.
*
- * The HAL takes ownership of the actual buffer_handle_t entries in
- * outputBuffers; the framework must not access them until they are
- * returned in a CaptureResult.
+ * The HAL takes ownership of the handles in outputBuffers; the client
+ * must not access them until they are returned in a CaptureResult.
*
* Any or all of the buffers included here may be brand new in this
* request (having never before seen by the HAL).
@@ -823,10 +821,15 @@
CameraMetadata result;
/**
- * The handles for the output stream buffers for this capture. They may not
- * yet be filled at the time the HAL calls processCaptureResult(); the
- * framework must wait on the release sync fences provided by the HAL before
- * reading the buffers.
+ * The completed output stream buffers for this capture.
+ *
+ * They may not yet be filled at the time the HAL calls
+ * processCaptureResult(); the framework must wait on the release sync
+ * fences provided by the HAL before reading the buffers.
+ *
+ * The StreamBuffer::buffer handle must be null for all returned buffers;
+ * the client must cache the handle and look it up via the combination of
+ * frame number and stream ID.
*
* The number of output buffers returned must be less than or equal to the
* matching capture request's count. If this is less than the buffer count
@@ -836,7 +839,7 @@
* valid result metadata or an input buffer is returned in this result.
*
* The HAL must set the stream buffer's release sync fence to a valid sync
- * fd, or to -1 if the buffer has already been filled.
+ * fd, or to null if the buffer has already been filled.
*
* If the HAL encounters an error while processing the buffer, and the
* buffer is not filled, the buffer's status field must be set to ERROR. If
@@ -844,7 +847,7 @@
* the acquire fence must be copied into the release fence, to allow the
* framework to wait on the fence before reusing the buffer.
*
- * The acquire fence must be set to -1 for all output buffers.
+ * The acquire fence must be set to null for all output buffers.
*
* This vector may be empty; if so, at least one other processCaptureResult
* call must be made (or have been made) by the HAL to provide the filled
@@ -854,8 +857,8 @@
* all previous frames' buffers for that corresponding stream must have been
* already delivered (the fences need not have yet been signaled).
*
- * Gralloc buffers for a frame may be sent to framework before the
- * corresponding SHUTTER-notify.
+ * Buffers for a frame may be sent to framework before the corresponding
+ * SHUTTER-notify call is made by the HAL.
*
* Performance considerations:
*
@@ -867,10 +870,11 @@
vec<StreamBuffer> outputBuffers;
/**
- * The handle for the input stream buffer for this capture. It may not
- * yet be consumed at the time the HAL calls processCaptureResult(); the
- * framework must wait on the release sync fence provided by the HAL before
- * reusing the buffer.
+ * The handle for the input stream buffer for this capture, if any.
+ *
+ * It may not yet be consumed at the time the HAL calls
+ * processCaptureResult(); the framework must wait on the release sync fence
+ * provided by the HAL before reusing the buffer.
*
* The HAL must handle the sync fences the same way they are done for
* outputBuffers.