Kernel/IPC: Implement MappedBuffer translation for HLE services that use the HLERequestContext architecture.

src/core/hle/kernel/hle_ipc.cpp

@@ -107,6 +107,12 @@ ResultCode HLERequestContext::PopulateFromIncomingCommandBuffer(const u32_le* sr
             cmd_buf[i++] = source_address;
             break;
         }
+        case IPC::DescriptorType::MappedBuffer: {
+            u32 next_id = static_cast<u32>(request_mapped_buffers.size());
+            request_mapped_buffers.emplace_back(src_process, descriptor, src_cmdbuf[i], next_id);
+            cmd_buf[i++] = next_id;
+            break;
+        }
         default:
             UNIMPLEMENTED_MSG("Unsupported handle translation: 0x%08X", descriptor);
         }
@@ -166,6 +172,11 @@ ResultCode HLERequestContext::WriteToOutgoingCommandBuffer(u32_le* dst_cmdbuf, P
             dst_cmdbuf[i++] = target_address;
             break;
         }
+        case IPC::DescriptorType::MappedBuffer: {
+            VAddr addr = request_mapped_buffers[cmd_buf[i]].address;
+            dst_cmdbuf[i++] = addr;
+            break;
+        }
         default:
             UNIMPLEMENTED_MSG("Unsupported handle translation: 0x%08X", descriptor);
         }
@@ -174,4 +185,28 @@ ResultCode HLERequestContext::WriteToOutgoingCommandBuffer(u32_le* dst_cmdbuf, P
     return RESULT_SUCCESS;
 }
 
+MappedBuffer& HLERequestContext::GetMappedBuffer(u32 id_from_cmdbuf) {
+    ASSERT_MSG(id_from_cmdbuf < request_mapped_buffers.size(), "Mapped Buffer ID out of range!");
+    return request_mapped_buffers[id_from_cmdbuf];
+}
+
+MappedBuffer::MappedBuffer(const Process& process, u32 descriptor, VAddr address, u32 id)
+    : process(&process), address(address), id(id) {
+    IPC::MappedBufferDescInfo desc{descriptor};
+    size = desc.size;
+    perms = desc.perms;
+}
+
+void MappedBuffer::Read(void* dest_buffer, size_t offset, size_t size) {
+    ASSERT(perms & IPC::R);
+    ASSERT(offset + size <= this->size);
+    Memory::ReadBlock(*process, address + offset, dest_buffer, size);
+}
+
+void MappedBuffer::Write(const void* src_buffer, size_t offset, size_t size) {
+    ASSERT(perms & IPC::W);
+    ASSERT(offset + size <= this->size);
+    Memory::WriteBlock(*process, address + offset, src_buffer, size);
+}
+
 } // namespace Kernel

src/core/hle/kernel/hle_ipc.h

@@ -63,6 +63,35 @@ protected:
     std::vector<SharedPtr<ServerSession>> connected_sessions;
 };
 
+class MappedBuffer {
+public:
+    MappedBuffer(const Process& process, u32 descriptor, VAddr address, u32 id);
+
+    // interface for service
+    void Read(void* dest_buffer, size_t offset, size_t size);
+    void Write(const void* src_buffer, size_t offset, size_t size);
+    size_t GetSize() const {
+        return size;
+    }
+
+    // interface for ipc helper
+    u32 GenerateDescriptor() const {
+        return IPC::MappedBufferDesc(size, perms);
+    }
+
+    u32 GetId() const {
+        return id;
+    }
+
+private:
+    friend class HLERequestContext;
+    u32 id;
+    VAddr address;
+    const Process* process;
+    size_t size;
+    IPC::MappedBufferPermissions perms;
+};
+
 /**
  * Class containing information about an in-flight IPC request being handled by an HLE service
  * implementation. Services should avoid using old global APIs (e.g. Kernel::GetCommandBuffer()) and
@@ -81,6 +110,17 @@ protected:
  * The end result is similar to just giving services their own real handle tables, but since these
  * ids are local to a specific context, it avoids requiring services to manage handles for objects
  * across multiple calls and ensuring that unneeded handles are cleaned up.
+ *
+ * HLE mapped buffer protocol
+ * ==========================
+ *
+ * HLE services don't have their own virtual memory space, a tweaked protocol is used to simulate
+ * memory mapping. The kernel will wrap the incoming buffers into a memory interface on which HLE
+ * services can operate, and insert a id in the buffer where the vaddr would normally be. The
+ * service then calls GetMappedBuffer with that id to get the memory interface. On response, like
+ * real services pushing back the mapped buffer address to unmap it, HLE services push back the
+ * id of the memory interface and let kernel convert it back to client vaddr. No real unmapping is
+ * needed in this case, though.
  */
 class HLERequestContext {
 public:
@@ -131,6 +171,12 @@ public:
      */
     void AddStaticBuffer(u8 buffer_id, std::vector<u8> data);
 
+    /**
+     * Gets a memory interface by the id from the request command buffer. See the "HLE mapped buffer
+     * protocol" section in the class documentation for more details.
+     */
+    MappedBuffer& GetMappedBuffer(u32 id_from_cmdbuf);
+
     /// Populates this context with data from the requesting process/thread.
     ResultCode PopulateFromIncomingCommandBuffer(const u32_le* src_cmdbuf, Process& src_process,
                                                  HandleTable& src_table);
@@ -145,6 +191,8 @@ private:
     boost::container::small_vector<SharedPtr<Object>, 8> request_handles;
     // The static buffers will be created when the IPC request is translated.
     std::array<std::vector<u8>, IPC::MAX_STATIC_BUFFERS> static_buffers;
+    // The mapped buffers will be created when the IPC request is translated
+    boost::container::small_vector<MappedBuffer, 8> request_mapped_buffers;
 };
 
 } // namespace Kernel