content: Fix Context creation logic in ContextProviderFactoryImpl.

content/browser/renderer_host/context_provider_factory_impl_android.cc

Index: content/browser/renderer_host/context_provider_factory_impl_android.cc
diff --git a/content/browser/renderer_host/context_provider_factory_impl_android.cc b/content/browser/renderer_host/context_provider_factory_impl_android.cc
index 40cc97c1a4a271837d53117c4d92b6822012aae8..5661976f8324c9d746c45ff920c9ef3645170f96 100644
--- a/content/browser/renderer_host/context_provider_factory_impl_android.cc
+++ b/content/browser/renderer_host/context_provider_factory_impl_android.cc
@@ -39,6 +39,21 @@ command_buffer_metrics::ContextType ToCommandBufferContextType(
   return command_buffer_metrics::CONTEXT_TYPE_UNKNOWN;
 }
 
+bool IsGpuChannelHostFactoryAvailable() {
+  return BrowserGpuChannelHostFactory::instance() != nullptr;
+}
+
+gpu::GpuChannelHost* IsGpuChannelAvailable() {
+  BrowserGpuChannelHostFactory* factory =
+      BrowserGpuChannelHostFactory::instance();
+  DCHECK(factory);
+
+  if (factory->GetGpuChannel())
+    return factory->GetGpuChannel();

[no sievers, 2016/08/17 18:30:10]

nit: This is maybe a bit verbose vs. just returning factory->GetGpuChannel().
Also maybe call the function GetGpuChannel() (the current name sounds like it
returns a bool)?

[Khushal, 2016/08/17 21:24:55]

Yeah, that's better. Done.

+  else
+    return nullptr;
+}
+
 }  // namespace
 
 // static
@@ -145,10 +160,30 @@ void ContextProviderFactoryImpl::CreateContextProviderInternal(
   context_request.result_callback = result_callback;
 
   context_provider_requests_.push_back(context_request);
-  HandlePendingRequests();
+  CheckCanHandlePendingRequests();
+}
+
+void ContextProviderFactoryImpl::CheckCanHandlePendingRequests() {

[no sievers, 2016/08/17 18:30:10]

can this outer logic be merged into HandlePendingRequests()?

[Khushal, 2016/08/17 21:24:55]

We call HandlePendingRequests from places where we know what the result is, for
instance when the GpuChannelHostFactory gives us a null host in the callback,
and we pass the result back to the caller who made the context request. I added
one to the dtor too so we clear out all the requests when the factory dies.

In CheckCanHandlePendingRequests, if we don't have a gpu channel already, we
would make a request to the factory to establish the channel instead.

+  DCHECK(!context_provider_requests_.empty())
+      << "We don't have any pending requests?";
+  if (!IsGpuChannelHostFactoryAvailable()) {
+    HandlePendingRequests(nullptr,
+                          ContextCreationFailureReason::BROWSER_SHUTDOWN);
+    return;
+  }
+
+  scoped_refptr<gpu::GpuChannelHost> gpu_channel_host(IsGpuChannelAvailable());
+  if (gpu_channel_host) {
+    HandlePendingRequests(gpu_channel_host,
+                          ContextCreationFailureReason::FAILURE_NONE);
+  } else {
+    EstablishGpuChannel();
+  }
 }
 
-void ContextProviderFactoryImpl::HandlePendingRequests() {
+void ContextProviderFactoryImpl::HandlePendingRequests(
+    scoped_refptr<gpu::GpuChannelHost> gpu_channel_host,
+    ContextCreationFailureReason reason) {
   DCHECK(!context_provider_requests_.empty())
       << "We don't have any pending requests?";
 
@@ -161,15 +196,6 @@ void ContextProviderFactoryImpl::HandlePendingRequests() {
     base::AutoReset<bool> auto_reset_in_handle_requests(
         &in_handle_pending_requests_, true);
 
-    scoped_refptr<gpu::GpuChannelHost> gpu_channel_host(
-        EnsureGpuChannelEstablished());
-
-    // If we don't have a Gpu Channel Host, we will come back here when the Gpu
-    // channel is established, since OnGpuChannelEstablished triggers handling
-    // of the requests we couldn't process right now.
-    if (!gpu_channel_host)
-      return;
-
     std::list<ContextProvidersRequest> context_requests =
         context_provider_requests_;
     context_provider_requests_.clear();
@@ -181,33 +207,39 @@ void ContextProviderFactoryImpl::HandlePendingRequests() {
           context_request.surface_handle != gpu::kNullSurfaceHandle;
 
       // Is the request for an onscreen context? Make sure the surface is
-      // still valid in that case. DO NOT run the callback if we don't have a
-      // valid surface.
+      // still valid in that case.
       if (create_onscreen_context &&
           !GpuSurfaceTracker::GetInstance()->IsValidSurfaceHandle(
               context_request.surface_handle)) {
-        continue;
+        // GPU Surface handle loss trumps other context creation failure
+        // reasons.
+        context_request.result_callback.Run(
+            nullptr, ContextCreationFailureReason::GPU_SURFACE_HANDLE_LOST);
+      } else if (!gpu_channel_host) {
+        context_request.result_callback.Run(nullptr, reason);
+      } else {
+        DCHECK_EQ(ContextCreationFailureReason::FAILURE_NONE, reason);
+
+        context_provider = new ContextProviderCommandBuffer(
+            gpu_channel_host, gpu::GPU_STREAM_DEFAULT,
+            gpu::GpuStreamPriority::NORMAL, context_request.surface_handle,
+            GURL(std::string("chrome://gpu/ContextProviderFactoryImpl::") +
+                 std::string("CompositorContextProvider")),
+            context_request.automatic_flushes, context_request.support_locking,
+            context_request.shared_memory_limits, context_request.attributes,
+            static_cast<ContextProviderCommandBuffer*>(
+                context_request.shared_context_provider),
+            context_request.context_type);
+        context_request.result_callback.Run(context_provider, reason);
       }
-
-      context_provider = new ContextProviderCommandBuffer(
-          gpu_channel_host, gpu::GPU_STREAM_DEFAULT,
-          gpu::GpuStreamPriority::NORMAL, context_request.surface_handle,
-          GURL(std::string("chrome://gpu/ContextProviderFactoryImpl::") +
-               std::string("CompositorContextProvider")),
-          context_request.automatic_flushes, context_request.support_locking,
-          context_request.shared_memory_limits, context_request.attributes,
-          static_cast<ContextProviderCommandBuffer*>(
-              context_request.shared_context_provider),
-          context_request.context_type);
-      context_request.result_callback.Run(context_provider);
     }
   }
 
   if (!context_provider_requests_.empty())
-    HandlePendingRequests();
+    CheckCanHandlePendingRequests();
 }
 
-gpu::GpuChannelHost* ContextProviderFactoryImpl::EnsureGpuChannelEstablished() {
+void ContextProviderFactoryImpl::EstablishGpuChannel() {
 #if defined(ADDRESS_SANITIZER) || defined(THREAD_SANITIZER) || \
     defined(SYZYASAN) || defined(CYGPROFILE_INSTRUMENTATION)
   const int64_t kGpuChannelTimeoutInSeconds = 40;
@@ -217,9 +249,7 @@ gpu::GpuChannelHost* ContextProviderFactoryImpl::EnsureGpuChannelEstablished() {
 
   BrowserGpuChannelHostFactory* factory =
       BrowserGpuChannelHostFactory::instance();
-
-  if (factory->GetGpuChannel())
-    return factory->GetGpuChannel();
+  DCHECK(factory);
 
   factory->EstablishGpuChannel(
       base::Bind(&ContextProviderFactoryImpl::OnGpuChannelEstablished,
@@ -227,24 +257,29 @@ gpu::GpuChannelHost* ContextProviderFactoryImpl::EnsureGpuChannelEstablished() {
   establish_gpu_channel_timeout_.Start(
       FROM_HERE, base::TimeDelta::FromSeconds(kGpuChannelTimeoutInSeconds),
       this, &ContextProviderFactoryImpl::OnGpuChannelTimeout);
-
-  return nullptr;
 }
 
 void ContextProviderFactoryImpl::OnGpuChannelEstablished(
     scoped_refptr<gpu::GpuChannelHost> gpu_channel) {
   establish_gpu_channel_timeout_.Stop();
 
-  // This should happen only during shutdown. So early out instead of queuing
-  // more requests with the factory.
-  if (!gpu_channel)
-    return;
-
   // We can queue the Gpu Channel initialization requests multiple times as
   // we get context requests. So we might have already handled any pending
   // requests when this callback runs.
-  if (!context_provider_requests_.empty())
-    HandlePendingRequests();
+  if (context_provider_requests_.empty())
+    return;
+
+  if (gpu_channel) {
+    HandlePendingRequests(std::move(gpu_channel),
+                          ContextCreationFailureReason::FAILURE_NONE);
+  } else if (IsGpuChannelHostFactoryAvailable()) {
+    HandlePendingRequests(
+        nullptr,
+        ContextCreationFailureReason::GPU_PROCESS_INITIALIZATION_FAILURE);
+  } else {
+    HandlePendingRequests(nullptr,
+                          ContextCreationFailureReason::BROWSER_SHUTDOWN);
+  }
 }
 
 void ContextProviderFactoryImpl::OnGpuChannelTimeout() {