【Android】SurfaceFlinger Dumpsys信息分析

SurfaceFlinger Dumpsys信息分析

dumpsys Surfaceflinger 用来输出SurfaceFlinger服务的状态信息，利用这些信息可以分析Android 画面层次、Display配置等等信息。

• 本文基于Android14。

dumpsys的实现

dumpsys Surfaceflinger命令对应的源码实现如下

• 源文件：frameworks/native/cmds/dumpsys/dumpsys.cpp
• 函数：status_t Dumpsys::startDumpThread

status_t Dumpsys::startDumpThread(int dumpTypeFlags, const String16& serviceName,
                                  const Vector<String16>& args) {
    sp<IBinder> service = sm_->checkService(serviceName);
    if (service == nullptr) {
        std::cerr << "Can't find service: " << serviceName << std::endl;
        return NAME_NOT_FOUND;
    }

    int sfd[2];
    if (pipe(sfd) != 0) {
        std::cerr << "Failed to create pipe to dump service info for " << serviceName << ": "
             << strerror(errno) << std::endl;
        return -errno;
    }

    redirectFd_ = unique_fd(sfd[0]);
    unique_fd remote_end(sfd[1]);
    sfd[0] = sfd[1] = -1;

    // dump blocks until completion, so spawn a thread..
    activeThread_ = std::thread([=, remote_end{std::move(remote_end)}]() mutable {
        if (dumpTypeFlags & TYPE_PID) {
            status_t err = dumpPidToFd(service, remote_end, dumpTypeFlags == TYPE_PID);
            reportDumpError(serviceName, err, "dumping PID");
        }
        if (dumpTypeFlags & TYPE_STABILITY) {
            status_t err = dumpStabilityToFd(service, remote_end);
            reportDumpError(serviceName, err, "dumping stability");
        }
        if (dumpTypeFlags & TYPE_THREAD) {
            status_t err = dumpThreadsToFd(service, remote_end);
            reportDumpError(serviceName, err, "dumping thread info");
        }
        if (dumpTypeFlags & TYPE_CLIENTS) {
            status_t err = dumpClientsToFd(service, remote_end);
            reportDumpError(serviceName, err, "dumping clients info");
        }

        // other types always act as a header, this is usually longer
        if (dumpTypeFlags & TYPE_DUMP) {
        	// 走这里！！！！！
            status_t err = service->dump(remote_end.get(), args);
            reportDumpError(serviceName, err, "dumping");
        }
    });
    return OK;
}

其实，就是调用对应Service的dump函数。经过IPC，调用到SurfaceFlinger的doDump接口。

• 源文件：frameworks/native/services/surfaceflinger/SurfaceFlinger.cpp
• 函数：void SurfaceFlinger::setPowerMode
  这个函数中，SurfaceFlinger将服务的相关信息dump出来。

status_t SurfaceFlinger::doDump(int fd, const DumpArgs& args, bool asProto) {
    std::string result;

    IPCThreadState* ipc = IPCThreadState::self();
    const int pid = ipc->getCallingPid();
    const int uid = ipc->getCallingUid();

    if ((uid != AID_SHELL) &&
            !PermissionCache::checkPermission(sDump, pid, uid)) {
        StringAppendF(&result, "Permission Denial: can't dump SurfaceFlinger from pid=%d, uid=%d\n",
                      pid, uid);
    } else {
        static const std::unordered_map<std::string, Dumper> dumpers = {
                {"--comp-displays"s, dumper(&SurfaceFlinger::dumpCompositionDisplays)},
                {"--display-id"s, dumper(&SurfaceFlinger::dumpDisplayIdentificationData)},
                {"--displays"s, dumper(&SurfaceFlinger::dumpDisplays)},
                {"--edid"s, argsDumper(&SurfaceFlinger::dumpRawDisplayIdentificationData)},
                {"--events"s, dumper(&SurfaceFlinger::dumpEvents)},
                {"--frametimeline"s, argsDumper(&SurfaceFlinger::dumpFrameTimeline)},
                {"--hwclayers"s, dumper(&SurfaceFlinger::dumpHwcLayersMinidumpLocked)},
                {"--latency"s, argsDumper(&SurfaceFlinger::dumpStatsLocked)},
                {"--latency-clear"s, argsDumper(&SurfaceFlinger::clearStatsLocked)},
                {"--list"s, dumper(&SurfaceFlinger::listLayersLocked)},
                {"--planner"s, argsDumper(&SurfaceFlinger::dumpPlannerInfo)},
                {"--scheduler"s, dumper(&SurfaceFlinger::dumpScheduler)},
                {"--timestats"s, protoDumper(&SurfaceFlinger::dumpTimeStats)},
                {"--vsync"s, dumper(&SurfaceFlinger::dumpVsync)},
                {"--wide-color"s, dumper(&SurfaceFlinger::dumpWideColorInfo)},
        };

        const auto flag = args.empty() ? ""s : std::string(String8(args[0]));

        // Traversal of drawing state must happen on the main thread.
        // Otherwise, SortedVector may have shared ownership during concurrent
        // traversals, which can result in use-after-frees.
        std::string compositionLayers;
        mScheduler
                ->schedule([&] {
                    StringAppendF(&compositionLayers, "Composition layers\n");
                    mDrawingState.traverseInZOrder([&](Layer* layer) {
                        auto* compositionState = layer->getCompositionState();
                        if (!compositionState || !compositionState->isVisible) return;

                        android::base::StringAppendF(&compositionLayers, "* Layer %p (%s)\n", layer,
                                                     layer->getDebugName() ? layer->getDebugName()
                                                                           : "<unknown>");
                        compositionState->dump(compositionLayers);
                    });
                })
                .get();

        bool dumpLayers = true;
        {
            TimedLock lock(mStateLock, s2ns(1), __func__);
            if (!lock.locked()) {
                StringAppendF(&result, "Dumping without lock after timeout: %s (%d)\n",
                              strerror(-lock.status), lock.status);
            }

            if (const auto it = dumpers.find(flag); it != dumpers.end()) {
                (it->second)(args, asProto, result);
                dumpLayers = false;
            } else if (!asProto) {
                dumpAllLocked(args, compositionLayers, result);
            }
        }

        if (dumpLayers) {
            LayersTraceFileProto traceFileProto = mLayerTracing.createTraceFileProto();
            LayersTraceProto* layersTrace = traceFileProto.add_entry();
            LayersProto layersProto = dumpProtoFromMainThread();
            layersTrace->mutable_layers()->Swap(&layersProto);
            auto displayProtos = dumpDisplayProto();
            layersTrace->mutable_displays()->Swap(&displayProtos);

            if (asProto) {
                result.append(traceFileProto.SerializeAsString());
            } else {
                // Dump info that we need to access from the main thread
                const auto layerTree = LayerProtoParser::generateLayerTree(layersTrace->layers());
                result.append(LayerProtoParser::layerTreeToString(layerTree));
                result.append("\n");
                dumpOffscreenLayers(result);
            }
        }
    }
    write(fd, result.c_str(), result.size());
    return NO_ERROR;
}

SurfaceFlinger Dump部分信息分析

SurfaceFlinger编译配置

Build configuration: 
[sf PRESENT_TIME_OFFSET=0 
FORCE_HWC_FOR_RBG_TO_YUV=1 
MAX_VIRT_DISPLAY_DIM=4096 
RUNNING_WITHOUT_SYNC_FRAMEWORK=0 
NUM_FRAMEBUFFER_SURFACE_BUFFERS=3]

sf：表示SurfacFlinger

PRESENT_TIME_OFFSET：垂直同步偏移，用来兼容垂直同步信号误差的（主要指跨进程的通知耗时），nanoseconds级别。

FORCE_HWC_FOR_RBG_TO_YUV：使用HWC进行 RGB到YUV的转换。

MAX_VIRT_DISPLAY_DIM：虚拟Display的最大尺寸，创建虚拟Display时，其宽或高要在这个范围内超过的话会创建失败。

RUNNING_WITHOUT_SYNC_FRAMEWORK: Sync同步框架是否开启，默认是开启的。

NUM_FRAMEBUFFER_SURFACE_BUFFERS：SurfaceFlinger中BufferQueue，一次申请Buffer时可以申请的最大数量（NUM_FRAMEBUFFER_SURFACE_BUFFERS减去1）。如果是3，减去1就是2.就是双Buffer机制。默认最高支持63个。一般都使用2，如果出现Jank情况下，可以适当调大（但是会更耗时系统资源)

显示器信息

Display identification data:
Display 100 (HWC display 0): invalid EDID

此处表示显示器信息（硬件信息）。EDID全称Extended Display Identification Data，存储在显示器寄存器中的一组有关于显示器属性值的信息。比如显示器名称、端口号之类。如果显示器没有提供这些信息，解析时就是显示ivalid EDID（PS此处显示的HWC，也就是PhysicalDisplay）

Wide-Color information:
Device supports wide color: 1
Device uses color management: 1
DisplayColorSetting: Managed
Display 100 color modes:
    ColorMode::NATIVE (0)
    Current color mode: ColorMode::NATIVE (0)

此处是显示器的色域信息，支持广色域、支持色彩管理、设置管理功能已经开启、支持1中颜色模式（Native）。
广色域是一种色彩背光技术，其色彩覆盖率能达到NSTC（National Television System Committe)标准的92%及以上。比如量子点LED背光能达到110%（比如AOC的量子点显示器大概在3499元）。使用广色域，可以让显示器的显示效果更佳鲜艳，但并不是NSTC覆盖标准越高越高，超过人眼的可识别范围的色彩是人类是无法分辨的（人眼识别范围（380~780nm）的光波波长。

Sync configuration: [using: EGL_ANDROID_native_fence_sync EGL_KHR_wait_sync]

这条信息表示，支持OpenGLES同步（Sync）。OpenGLES Sync依赖于两个扩EGL_ANDROID_native_fence_sync、EGL_KHR_wait_sync

Layer信息

Visible layers (count = 200)
Composition layers

当前显示的图层数目是200个

Display状态

Displays (1 entries) 有1个Display
Display 100（Display ID是 100）
connectionType=Internal （内置类型）
ColorMode::NATIVE （颜色模式为Native）
deviceProductInfo=nullopt （产品信息为空）
name="Primary display" 默认的Display
powerMode=Off （电源状态是OFF）
activeMode=60.00 Hz (60.00 Hz) （刷新率是60帧）

• 如果通过scrpy投屏可以看到对应的Virtual Display

Virtual Display 12529715046768705014
    name="scrcpy"
    powerMode=On

其他信息省略。