ContextHubService: Allow cancelling load/unload
As a minimal risk improvement for Android N, we allow a new
load/unload transaction request to cancel a currently pending
one if enough time has passed since starting the current
transaction.
Bug: 30111539
Change-Id: Ide958c2454b88529305443fb789b8ae5af18c642
diff --git a/core/jni/android_hardware_location_ContextHubService.cpp b/core/jni/android_hardware_location_ContextHubService.cpp
index 8bee9c9..3644410 100644
--- a/core/jni/android_hardware_location_ContextHubService.cpp
+++ b/core/jni/android_hardware_location_ContextHubService.cpp
@@ -26,6 +26,10 @@
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
+
+// TOOD: On master, alphabetize these and move <mutex> into this
+// grouping.
+#include <chrono>
#include <unordered_map>
#include <queue>
@@ -51,6 +55,10 @@
static constexpr size_t HEADER_FIELD_LOAD_APP_ID_HI = MSG_HEADER_SIZE + 1;
static constexpr size_t MSG_HEADER_SIZE_LOAD_APP = MSG_HEADER_SIZE + 2;
+// Monotonically increasing clock we use to determine if we can cancel
+// a transaction.
+using std::chrono::steady_clock;
+
namespace android {
namespace {
@@ -107,6 +115,17 @@
struct hub_app_info appInfo; // returned from the HAL
};
+
+// If a transaction takes longer than this, we'll allow it to be
+// canceled by a new transaction. Note we do _not_ automatically
+// cancel a transaction after this much time. We can have a
+// legal transaction which takes longer than this amount of time,
+// as long as no other new transactions are attempted after this
+// time has expired.
+// TODO(b/31105001): Establish a clean timing approach for all
+// of our HAL interactions.
+constexpr auto kMinTransactionCancelTime = std::chrono::seconds(29);
+
/*
* TODO(ashutoshj): From original code review:
*
@@ -148,6 +167,7 @@
std::mutex m; // mutex for manager
hub_messages_e txnIdentifier; // What are we doing
void *txnData; // Details
+ steady_clock::time_point firstTimeTxnCanBeCanceled;
};
struct contextHubServiceDb_s {
@@ -177,25 +197,40 @@
std::lock_guard<std::mutex>lock(mgr->m);
mgr->txnPending = true;
+ mgr->firstTimeTxnCanBeCanceled = steady_clock::now() +
+ kMinTransactionCancelTime;
mgr->txnData = txnData;
mgr->txnIdentifier = txnIdentifier;
return 0;
}
-static int closeTxn() {
+// Only call this if you hold the db.txnManager.m lock.
+static void closeTxnUnlocked() {
txnManager_s *mgr = &db.txnManager;
- std::lock_guard<std::mutex>lock(mgr->m);
mgr->txnPending = false;
free(mgr->txnData);
mgr->txnData = nullptr;
+}
+static int closeTxn() {
+ std::lock_guard<std::mutex>lock(db.txnManager.m);
+ closeTxnUnlocked();
return 0;
}
+// If a transaction has been pending for longer than
+// kMinTransactionCancelTime, this call will "cancel" that
+// transaction and return that there are none pending.
static bool isTxnPending() {
txnManager_s *mgr = &db.txnManager;
std::lock_guard<std::mutex>lock(mgr->m);
+ if (mgr->txnPending) {
+ if (steady_clock::now() >= mgr->firstTimeTxnCanBeCanceled) {
+ ALOGW("Transaction canceled");
+ closeTxnUnlocked();
+ }
+ }
return mgr->txnPending;
}
