diff options
Diffstat (limited to 'src/core/hle/service/nvdrv/devices')
-rw-r--r-- | src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp | 33 | ||||
-rw-r--r-- | src/core/hle/service/nvdrv/devices/nvhost_ctrl.h | 4 | ||||
-rw-r--r-- | src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp | 135 | ||||
-rw-r--r-- | src/core/hle/service/nvdrv/devices/nvhost_gpu.h | 20 |
4 files changed, 140 insertions, 52 deletions
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp index 75d9191ff..8356a8139 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp +++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp @@ -15,8 +15,9 @@ namespace Service::Nvidia::Devices { -nvhost_ctrl::nvhost_ctrl(Core::System& system, EventInterface& events_interface) - : nvdevice(system), events_interface{events_interface} {} +nvhost_ctrl::nvhost_ctrl(Core::System& system, EventInterface& events_interface, + SyncpointManager& syncpoint_manager) + : nvdevice(system), events_interface{events_interface}, syncpoint_manager{syncpoint_manager} {} nvhost_ctrl::~nvhost_ctrl() = default; u32 nvhost_ctrl::ioctl(Ioctl command, const std::vector<u8>& input, const std::vector<u8>& input2, @@ -70,19 +71,33 @@ u32 nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>& return NvResult::BadParameter; } + if (syncpoint_manager.IsSyncpointExpired(params.syncpt_id, params.threshold)) { + params.value = syncpoint_manager.GetSyncpointMin(params.syncpt_id); + std::memcpy(output.data(), ¶ms, sizeof(params)); + return NvResult::Success; + } + + if (const auto new_value = syncpoint_manager.RefreshSyncpoint(params.syncpt_id); + syncpoint_manager.IsSyncpointExpired(params.syncpt_id, params.threshold)) { + params.value = new_value; + std::memcpy(output.data(), ¶ms, sizeof(params)); + return NvResult::Success; + } + auto event = events_interface.events[event_id]; auto& gpu = system.GPU(); + // This is mostly to take into account unimplemented features. As synced // gpu is always synced. if (!gpu.IsAsync()) { - event.writable->Signal(); + event.event.writable->Signal(); return NvResult::Success; } auto lock = gpu.LockSync(); - const u32 current_syncpoint_value = gpu.GetSyncpointValue(params.syncpt_id); + const u32 current_syncpoint_value = event.fence.value; const s32 diff = current_syncpoint_value - params.threshold; if (diff >= 0) { - event.writable->Signal(); + event.event.writable->Signal(); params.value = current_syncpoint_value; std::memcpy(output.data(), ¶ms, sizeof(params)); return NvResult::Success; @@ -109,7 +124,7 @@ u32 nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>& params.value = ((params.syncpt_id & 0xfff) << 16) | 0x10000000; } params.value |= event_id; - event.writable->Clear(); + event.event.writable->Clear(); gpu.RegisterSyncptInterrupt(params.syncpt_id, target_value); if (!is_async && ctrl.fresh_call) { ctrl.must_delay = true; @@ -157,15 +172,19 @@ u32 nvhost_ctrl::IocCtrlEventUnregister(const std::vector<u8>& input, std::vecto u32 nvhost_ctrl::IocCtrlClearEventWait(const std::vector<u8>& input, std::vector<u8>& output) { IocCtrlEventSignalParams params{}; std::memcpy(¶ms, input.data(), sizeof(params)); + u32 event_id = params.event_id & 0x00FF; LOG_WARNING(Service_NVDRV, "cleared event wait on, event_id: {:X}", event_id); + if (event_id >= MaxNvEvents) { return NvResult::BadParameter; } if (events_interface.status[event_id] == EventState::Waiting) { events_interface.LiberateEvent(event_id); - events_interface.events[event_id].writable->Signal(); } + + syncpoint_manager.RefreshSyncpoint(events_interface.events[event_id].fence.id); + return NvResult::Success; } diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h index f7b04d9f1..24ad96cb9 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h +++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h @@ -14,7 +14,8 @@ namespace Service::Nvidia::Devices { class nvhost_ctrl final : public nvdevice { public: - explicit nvhost_ctrl(Core::System& system, EventInterface& events_interface); + explicit nvhost_ctrl(Core::System& system, EventInterface& events_interface, + SyncpointManager& syncpoint_manager); ~nvhost_ctrl() override; u32 ioctl(Ioctl command, const std::vector<u8>& input, const std::vector<u8>& input2, @@ -145,6 +146,7 @@ private: u32 IocCtrlClearEventWait(const std::vector<u8>& input, std::vector<u8>& output); EventInterface& events_interface; + SyncpointManager& syncpoint_manager; }; } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp index f1966ac0e..152019548 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp +++ b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp @@ -7,14 +7,20 @@ #include "common/logging/log.h" #include "core/core.h" #include "core/hle/service/nvdrv/devices/nvhost_gpu.h" +#include "core/hle/service/nvdrv/syncpoint_manager.h" #include "core/memory.h" #include "video_core/gpu.h" #include "video_core/memory_manager.h" namespace Service::Nvidia::Devices { -nvhost_gpu::nvhost_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev) - : nvdevice(system), nvmap_dev(std::move(nvmap_dev)) {} +nvhost_gpu::nvhost_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev, + SyncpointManager& syncpoint_manager) + : nvdevice(system), nvmap_dev(std::move(nvmap_dev)), syncpoint_manager{syncpoint_manager} { + channel_fence.id = syncpoint_manager.AllocateSyncpoint(); + channel_fence.value = system.GPU().GetSyncpointValue(channel_fence.id); +} + nvhost_gpu::~nvhost_gpu() = default; u32 nvhost_gpu::ioctl(Ioctl command, const std::vector<u8>& input, const std::vector<u8>& input2, @@ -126,10 +132,10 @@ u32 nvhost_gpu::AllocGPFIFOEx2(const std::vector<u8>& input, std::vector<u8>& ou params.num_entries, params.flags, params.unk0, params.unk1, params.unk2, params.unk3); - auto& gpu = system.GPU(); - params.fence_out.id = assigned_syncpoints; - params.fence_out.value = gpu.GetSyncpointValue(assigned_syncpoints); - assigned_syncpoints++; + channel_fence.value = system.GPU().GetSyncpointValue(channel_fence.id); + + params.fence_out = channel_fence; + std::memcpy(output.data(), ¶ms, output.size()); return 0; } @@ -145,39 +151,100 @@ u32 nvhost_gpu::AllocateObjectContext(const std::vector<u8>& input, std::vector< return 0; } -u32 nvhost_gpu::SubmitGPFIFO(const std::vector<u8>& input, std::vector<u8>& output) { - if (input.size() < sizeof(IoctlSubmitGpfifo)) { - UNIMPLEMENTED(); +static std::vector<Tegra::CommandHeader> BuildWaitCommandList(Fence fence) { + return { + Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceValue, 1, + Tegra::SubmissionMode::Increasing), + {fence.value}, + Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceAction, 1, + Tegra::SubmissionMode::Increasing), + Tegra::GPU::FenceAction::Build(Tegra::GPU::FenceOperation::Acquire, fence.id), + }; +} + +static std::vector<Tegra::CommandHeader> BuildIncrementCommandList(Fence fence, u32 add_increment) { + std::vector<Tegra::CommandHeader> result{ + Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceValue, 1, + Tegra::SubmissionMode::Increasing), + {}}; + + for (u32 count = 0; count < add_increment; ++count) { + result.emplace_back(Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceAction, 1, + Tegra::SubmissionMode::Increasing)); + result.emplace_back( + Tegra::GPU::FenceAction::Build(Tegra::GPU::FenceOperation::Increment, fence.id)); } - IoctlSubmitGpfifo params{}; - std::memcpy(¶ms, input.data(), sizeof(IoctlSubmitGpfifo)); + + return result; +} + +static std::vector<Tegra::CommandHeader> BuildIncrementWithWfiCommandList(Fence fence, + u32 add_increment) { + std::vector<Tegra::CommandHeader> result{ + Tegra::BuildCommandHeader(Tegra::BufferMethods::WaitForInterrupt, 1, + Tegra::SubmissionMode::Increasing), + {}}; + const std::vector<Tegra::CommandHeader> increment{ + BuildIncrementCommandList(fence, add_increment)}; + + result.insert(result.end(), increment.begin(), increment.end()); + + return result; +} + +u32 nvhost_gpu::SubmitGPFIFOImpl(IoctlSubmitGpfifo& params, std::vector<u8>& output, + Tegra::CommandList&& entries) { LOG_TRACE(Service_NVDRV, "called, gpfifo={:X}, num_entries={:X}, flags={:X}", params.address, params.num_entries, params.flags.raw); - ASSERT_MSG(input.size() == sizeof(IoctlSubmitGpfifo) + - params.num_entries * sizeof(Tegra::CommandListHeader), - "Incorrect input size"); + auto& gpu = system.GPU(); - Tegra::CommandList entries(params.num_entries); - std::memcpy(entries.data(), &input[sizeof(IoctlSubmitGpfifo)], - params.num_entries * sizeof(Tegra::CommandListHeader)); + params.fence_out.id = channel_fence.id; - UNIMPLEMENTED_IF(params.flags.add_wait.Value() != 0); - UNIMPLEMENTED_IF(params.flags.add_increment.Value() != 0); + if (params.flags.add_wait.Value() && + !syncpoint_manager.IsSyncpointExpired(params.fence_out.id, params.fence_out.value)) { + gpu.PushGPUEntries(Tegra::CommandList{BuildWaitCommandList(params.fence_out)}); + } - auto& gpu = system.GPU(); - u32 current_syncpoint_value = gpu.GetSyncpointValue(params.fence_out.id); - if (params.flags.increment.Value()) { - params.fence_out.value += current_syncpoint_value; + if (params.flags.add_increment.Value() || params.flags.increment.Value()) { + const u32 increment_value = params.flags.increment.Value() ? params.fence_out.value : 0; + params.fence_out.value = syncpoint_manager.IncreaseSyncpoint( + params.fence_out.id, params.AddIncrementValue() + increment_value); } else { - params.fence_out.value = current_syncpoint_value; + params.fence_out.value = syncpoint_manager.GetSyncpointMax(params.fence_out.id); } + + entries.RefreshIntegrityChecks(gpu); gpu.PushGPUEntries(std::move(entries)); + if (params.flags.add_increment.Value()) { + if (params.flags.suppress_wfi) { + gpu.PushGPUEntries(Tegra::CommandList{ + BuildIncrementCommandList(params.fence_out, params.AddIncrementValue())}); + } else { + gpu.PushGPUEntries(Tegra::CommandList{ + BuildIncrementWithWfiCommandList(params.fence_out, params.AddIncrementValue())}); + } + } + std::memcpy(output.data(), ¶ms, sizeof(IoctlSubmitGpfifo)); return 0; } +u32 nvhost_gpu::SubmitGPFIFO(const std::vector<u8>& input, std::vector<u8>& output) { + if (input.size() < sizeof(IoctlSubmitGpfifo)) { + UNIMPLEMENTED(); + } + IoctlSubmitGpfifo params{}; + std::memcpy(¶ms, input.data(), sizeof(IoctlSubmitGpfifo)); + + Tegra::CommandList entries(params.num_entries); + std::memcpy(entries.command_lists.data(), &input[sizeof(IoctlSubmitGpfifo)], + params.num_entries * sizeof(Tegra::CommandListHeader)); + + return SubmitGPFIFOImpl(params, output, std::move(entries)); +} + u32 nvhost_gpu::KickoffPB(const std::vector<u8>& input, std::vector<u8>& output, const std::vector<u8>& input2, IoctlVersion version) { if (input.size() < sizeof(IoctlSubmitGpfifo)) { @@ -185,31 +252,17 @@ u32 nvhost_gpu::KickoffPB(const std::vector<u8>& input, std::vector<u8>& output, } IoctlSubmitGpfifo params{}; std::memcpy(¶ms, input.data(), sizeof(IoctlSubmitGpfifo)); - LOG_TRACE(Service_NVDRV, "called, gpfifo={:X}, num_entries={:X}, flags={:X}", params.address, - params.num_entries, params.flags.raw); Tegra::CommandList entries(params.num_entries); if (version == IoctlVersion::Version2) { - std::memcpy(entries.data(), input2.data(), + std::memcpy(entries.command_lists.data(), input2.data(), params.num_entries * sizeof(Tegra::CommandListHeader)); } else { - system.Memory().ReadBlock(params.address, entries.data(), + system.Memory().ReadBlock(params.address, entries.command_lists.data(), params.num_entries * sizeof(Tegra::CommandListHeader)); } - UNIMPLEMENTED_IF(params.flags.add_wait.Value() != 0); - UNIMPLEMENTED_IF(params.flags.add_increment.Value() != 0); - - auto& gpu = system.GPU(); - u32 current_syncpoint_value = gpu.GetSyncpointValue(params.fence_out.id); - if (params.flags.increment.Value()) { - params.fence_out.value += current_syncpoint_value; - } else { - params.fence_out.value = current_syncpoint_value; - } - gpu.PushGPUEntries(std::move(entries)); - std::memcpy(output.data(), ¶ms, output.size()); - return 0; + return SubmitGPFIFOImpl(params, output, std::move(entries)); } u32 nvhost_gpu::GetWaitbase(const std::vector<u8>& input, std::vector<u8>& output) { diff --git a/src/core/hle/service/nvdrv/devices/nvhost_gpu.h b/src/core/hle/service/nvdrv/devices/nvhost_gpu.h index 2ac74743f..a252fc06d 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_gpu.h +++ b/src/core/hle/service/nvdrv/devices/nvhost_gpu.h @@ -11,6 +11,11 @@ #include "common/swap.h" #include "core/hle/service/nvdrv/devices/nvdevice.h" #include "core/hle/service/nvdrv/nvdata.h" +#include "video_core/dma_pusher.h" + +namespace Service::Nvidia { +class SyncpointManager; +} namespace Service::Nvidia::Devices { @@ -21,7 +26,8 @@ constexpr u32 NVGPU_IOCTL_CHANNEL_KICKOFF_PB(0x1b); class nvhost_gpu final : public nvdevice { public: - explicit nvhost_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev); + explicit nvhost_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev, + SyncpointManager& syncpoint_manager); ~nvhost_gpu() override; u32 ioctl(Ioctl command, const std::vector<u8>& input, const std::vector<u8>& input2, @@ -162,10 +168,15 @@ private: u32_le raw; BitField<0, 1, u32_le> add_wait; // append a wait sync_point to the list BitField<1, 1, u32_le> add_increment; // append an increment to the list - BitField<2, 1, u32_le> new_hw_format; // Mostly ignored + BitField<2, 1, u32_le> new_hw_format; // mostly ignored + BitField<4, 1, u32_le> suppress_wfi; // suppress wait for interrupt BitField<8, 1, u32_le> increment; // increment the returned fence } flags; Fence fence_out; // returned new fence object for others to wait on + + u32 AddIncrementValue() const { + return flags.add_increment.Value() << 1; + } }; static_assert(sizeof(IoctlSubmitGpfifo) == 16 + sizeof(Fence), "IoctlSubmitGpfifo is incorrect size"); @@ -190,6 +201,8 @@ private: u32 SetChannelPriority(const std::vector<u8>& input, std::vector<u8>& output); u32 AllocGPFIFOEx2(const std::vector<u8>& input, std::vector<u8>& output); u32 AllocateObjectContext(const std::vector<u8>& input, std::vector<u8>& output); + u32 SubmitGPFIFOImpl(IoctlSubmitGpfifo& params, std::vector<u8>& output, + Tegra::CommandList&& entries); u32 SubmitGPFIFO(const std::vector<u8>& input, std::vector<u8>& output); u32 KickoffPB(const std::vector<u8>& input, std::vector<u8>& output, const std::vector<u8>& input2, IoctlVersion version); @@ -198,7 +211,8 @@ private: u32 ChannelSetTimeslice(const std::vector<u8>& input, std::vector<u8>& output); std::shared_ptr<nvmap> nvmap_dev; - u32 assigned_syncpoints{}; + SyncpointManager& syncpoint_manager; + Fence channel_fence; }; } // namespace Service::Nvidia::Devices |