Kernel: Implement basic VirGL device

This commit flips VirtIOGPU back to using a Mutex for its operation
lock (instead of a spinlock). This is necessary for avoiding a few
system hangs when queuing actions on the driver from multiple
processes, which becomes much more of an issue when using VirGL from
multiple userspace process.

This does result in a few code paths where we inevitably have to grab
a mutex from inside a spinlock, the only way to fix both issues is to
move to issuing asynchronous virtio gpu commands.

Kernel/Graphics/VirtIOGPU/GPU3DDevice.h

@@ -0,0 +1,111 @@
+/*
+ * Copyright (c) 2021, Sahan Fernando <sahan.h.fernando@gmail.com>
+ *
+ * SPDX-License-Identifier: BSD-2-Clause
+ */
+
+#pragma once
+
+#include <AK/DistinctNumeric.h>
+#include <Kernel/Devices/CharacterDevice.h>
+#include <Kernel/Graphics/VirtIOGPU/FramebufferDevice.h>
+#include <Kernel/Graphics/VirtIOGPU/Protocol.h>
+
+namespace Kernel::Graphics::VirtIOGPU {
+
+enum class VirGLCommand : u32 {
+    NOP = 0,
+    CREATE_OBJECT = 1,
+    BIND_OBJECT,
+    DESTROY_OBJECT,
+    SET_VIEWPORT_STATE,
+    SET_FRAMEBUFFER_STATE,
+    SET_VERTEX_BUFFERS,
+    CLEAR,
+    DRAW_VBO,
+    RESOURCE_INLINE_WRITE,
+    SET_SAMPLER_VIEWS,
+    SET_INDEX_BUFFER,
+    SET_CONSTANT_BUFFER,
+    SET_STENCIL_REF,
+    SET_BLEND_COLOR,
+    SET_SCISSOR_STATE,
+    BLIT,
+    RESOURCE_COPY_REGION,
+    BIND_SAMPLER_STATES,
+    BEGIN_QUERY,
+    END_QUERY,
+    GET_QUERY_RESULT,
+    SET_POLYGON_STIPPLE,
+    SET_CLIP_STATE,
+    SET_SAMPLE_MASK,
+    SET_STREAMOUT_TARGETS,
+    SET_RENDER_CONDITION,
+    SET_UNIFORM_BUFFER,
+
+    SET_SUB_CTX,
+    CREATE_SUB_CTX,
+    DESTROY_SUB_CTX,
+    BIND_SHADER,
+    SET_TESS_STATE,
+    SET_MIN_SAMPLES,
+    SET_SHADER_BUFFERS,
+    SET_SHADER_IMAGES,
+    MEMORY_BARRIER,
+    LAUNCH_GRID,
+    SET_FRAMEBUFFER_STATE_NO_ATTACH,
+    TEXTURE_BARRIER,
+    SET_ATOMIC_BUFFERS,
+    SET_DBG_FLAGS,
+    GET_QUERY_RESULT_QBO,
+    TRANSFER3D,
+    END_TRANSFERS,
+    COPY_TRANSFER3D,
+    SET_TWEAKS,
+    CLEAR_TEXTURE,
+    PIPE_RESOURCE_CREATE,
+    PIPE_RESOURCE_SET_TYPE,
+    GET_MEMORY_INFO,
+    SEND_STRING_MARKER,
+    MAX_COMMANDS
+};
+
+union ClearType {
+    struct {
+        u32 depth : 1;
+        u32 stencil : 1;
+        u32 color0 : 1;
+        u32 color1 : 1;
+        u32 color2 : 1;
+        u32 color3 : 1;
+        u32 color4 : 1;
+        u32 color5 : 1;
+        u32 color6 : 1;
+        u32 color7 : 1;
+    } flags;
+    u32 value;
+};
+
+class GPU3DDevice : public CharacterDevice {
+public:
+    GPU3DDevice() = delete;
+    explicit GPU3DDevice(GraphicsAdapter& graphics_adapter);
+
+    virtual bool can_read(const OpenFileDescription&, u64) const override { return true; }
+    virtual bool can_write(const OpenFileDescription&, u64) const override { return true; }
+    virtual ErrorOr<size_t> read(OpenFileDescription&, u64, UserOrKernelBuffer&, size_t) override { return ENOTSUP; }
+    virtual ErrorOr<size_t> write(OpenFileDescription&, u64, const UserOrKernelBuffer&, size_t) override { return ENOTSUP; }
+    virtual StringView class_name() const override { return "virgl3d"; }
+
+    virtual ErrorOr<void> ioctl(OpenFileDescription&, unsigned request, Userspace<void*> arg) override;
+
+private:
+    Kernel::Graphics::VirtIOGPU::GraphicsAdapter& m_graphics_adapter;
+    // Context used for kernel operations (e.g. flushing resources to scanout)
+    ContextID m_kernel_context_id;
+    // Memory management for backing buffers
+    OwnPtr<Memory::Region> m_transfer_buffer_region;
+    constexpr static size_t NUM_TRANSFER_REGION_PAGES = 256;
+};
+
+}