GPU Memory: ensure bpf program is attached to the tracepoint am: a3dae23fbe

Original change: https://android-review.googlesource.com/c/platform/frameworks/native/+/1472016

Change-Id: I941874d954b9670c6fb0f05ce36bfdd976c23ce7

5 files changed, 25 insertions, 6 deletions

diff --git a/services/gpuservice/GpuService.cpp b/services/gpuservice/GpuService.cpp
index 0ca8e5d2e3..84ae608148 100644
--- a/services/gpuservice/GpuService.cpp
+++ b/services/gpuservice/GpuService.cpp
@@ -31,6 +31,8 @@
 #include <utils/Trace.h>
 #include <vkjson.h>
 
+#include <thread>
+
 namespace android {
 
 using base::StringAppendF;
@@ -47,7 +49,8 @@ const char* const GpuService::SERVICE_NAME = "gpu";
 
 GpuService::GpuService()
       : mGpuMem(std::make_unique<GpuMem>()), mGpuStats(std::make_unique<GpuStats>()) {
-    mGpuMem->initialize();
+    std::thread asyncInitThread([this]() { mGpuMem->initialize(); });
+    asyncInitThread.detach();
 };
 
 void GpuService::setGpuStats(const std::string& driverPackageName,
diff --git a/services/gpuservice/gpumem/GpuMem.cpp b/services/gpuservice/gpumem/GpuMem.cpp
index 1d4b524a7d..9b4053b7fd 100644
--- a/services/gpuservice/gpumem/GpuMem.cpp
+++ b/services/gpuservice/gpumem/GpuMem.cpp
@@ -24,6 +24,7 @@
 #include <libbpf.h>
 #include <libbpf_android.h>
 #include <log/log.h>
+#include <unistd.h>
 #include <utils/Trace.h>
 
 #include <unordered_map>
@@ -47,11 +48,17 @@ void GpuMem::initialize() {
         return;
     }
 
+    // TODO(http://b/159963505): Figure out a nicer way to wait until GPU driver loaded.
     // Attach the program to the tracepoint, and the tracepoint is automatically enabled here.
-    if (bpf_attach_tracepoint(fd, kGpuMemTraceGroup, kGpuMemTotalTracepoint) < 0) {
-        ALOGE("Failed to attach bpf program to %s/%s tracepoint", kGpuMemTraceGroup,
-              kGpuMemTotalTracepoint);
-        return;
+    int count = 0;
+    while (bpf_attach_tracepoint(fd, kGpuMemTraceGroup, kGpuMemTotalTracepoint) < 0) {
+        if (++count > kGpuWaitTimeout) {
+            ALOGE("Failed to attach bpf program to %s/%s tracepoint", kGpuMemTraceGroup,
+                  kGpuMemTotalTracepoint);
+            return;
+        }
+        // Retry until GPU driver loaded or timeout.
+        sleep(1);
     }
 
     // Use the read-only wrapper BpfMapRO to properly retrieve the read-only map.
@@ -61,6 +68,8 @@ void GpuMem::initialize() {
         return;
     }
     setGpuMemTotalMap(map);
+
+    mInitialized.store(true);
 }
 
 void GpuMem::setGpuMemTotalMap(bpf::BpfMap<uint64_t, uint64_t>& map) {
@@ -71,7 +80,7 @@ void GpuMem::setGpuMemTotalMap(bpf::BpfMap<uint64_t, uint64_t>& map) {
 void GpuMem::dump(const Vector<String16>& /* args */, std::string* result) {
     ATRACE_CALL();
 
-    if (!mGpuMemTotalMap.isValid()) {
+    if (!mInitialized.load() || !mGpuMemTotalMap.isValid()) {
         result->append("Failed to initialize GPU memory eBPF\n");
         return;
     }
diff --git a/services/gpuservice/gpumem/include/gpumem/GpuMem.h b/services/gpuservice/gpumem/include/gpumem/GpuMem.h
index 6d0322ad1f..ff8b4bc70a 100644
--- a/services/gpuservice/gpumem/include/gpumem/GpuMem.h
+++ b/services/gpuservice/gpumem/include/gpumem/GpuMem.h
@@ -39,6 +39,8 @@ private:
     // set gpu memory total map
     void setGpuMemTotalMap(bpf::BpfMap<uint64_t, uint64_t>& map);
 
+    // indicate whether ebpf has been initialized
+    std::atomic<bool> mInitialized = false;
     // bpf map for GPU memory total data
     android::bpf::BpfMap<uint64_t, uint64_t> mGpuMemTotalMap;
 
@@ -51,6 +53,8 @@ private:
             "/sys/fs/bpf/prog_gpu_mem_tracepoint_gpu_mem_gpu_mem_total";
     // pinned gpu memory total bpf map path in bpf sysfs
     static constexpr char kGpuMemTotalMapPath[] = "/sys/fs/bpf/map_gpu_mem_gpu_mem_total_map";
+    // 30 seconds timeout for trying to attach bpf program to tracepoint
+    static constexpr int kGpuWaitTimeout = 30;
 };
 
 } // namespace android
diff --git a/services/gpuservice/tests/unittests/GpuMemTest.cpp b/services/gpuservice/tests/unittests/GpuMemTest.cpp
index 6ba304c9a9..abaf30ab8a 100644
--- a/services/gpuservice/tests/unittests/GpuMemTest.cpp
+++ b/services/gpuservice/tests/unittests/GpuMemTest.cpp
@@ -62,6 +62,7 @@ public:
 
         mGpuMem = std::make_unique<GpuMem>();
         mTestableGpuMem = TestableGpuMem(mGpuMem.get());
+        mTestableGpuMem.setInitialized();
         errno = 0;
         mTestMap = bpf::BpfMap<uint64_t, uint64_t>(BPF_MAP_TYPE_HASH, TEST_MAP_SIZE,
                                                    BPF_F_NO_PREALLOC);
diff --git a/services/gpuservice/tests/unittests/TestableGpuMem.h b/services/gpuservice/tests/unittests/TestableGpuMem.h
index 0e4b01ca02..6c8becb075 100644
--- a/services/gpuservice/tests/unittests/TestableGpuMem.h
+++ b/services/gpuservice/tests/unittests/TestableGpuMem.h
@@ -26,6 +26,8 @@ public:
     TestableGpuMem() = default;
     explicit TestableGpuMem(GpuMem *gpuMem) : mGpuMem(gpuMem) {}
 
+    void setInitialized() { mGpuMem->mInitialized.store(true); }
+
     void setGpuMemTotalMap(bpf::BpfMap<uint64_t, uint64_t>& map) {
         mGpuMem->setGpuMemTotalMap(map);
     }