From 3f601ed8bc0833cdf23c63a3378c873bb97b7b4c Mon Sep 17 00:00:00 2001 From: ameerj <52414509+ameerj@users.noreply.github.com> Date: Thu, 24 Jun 2021 21:29:41 -0400 Subject: [PATCH 1/3] nvhost_nvdec_common: Fix {Slice/Write}Vectors return Plus some minor cleanup for consistency. --- .../nvdrv/devices/nvhost_nvdec_common.cpp | 75 ++++++++++--------- 1 file changed, 38 insertions(+), 37 deletions(-) diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp index 98e6296f1e..f045008eb3 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp +++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp @@ -19,26 +19,29 @@ namespace Service::Nvidia::Devices { namespace { -// Splice vectors will copy count amount of type T from the input vector into the dst vector. +// Copies count amount of type T from the input vector into the dst vector. +// Returns the number of bytes written into dst. template -std::size_t SpliceVectors(const std::vector& input, std::vector& dst, std::size_t count, - std::size_t offset) { - if (!dst.empty()) { - std::memcpy(dst.data(), input.data() + offset, count * sizeof(T)); +std::size_t SliceVectors(const std::vector& input, std::vector& dst, std::size_t count, + std::size_t offset) { + if (dst.empty()) { + return 0; } - return 0; + const size_t bytes_copied = count * sizeof(T); + std::memcpy(dst.data(), input.data() + offset, bytes_copied); + return bytes_copied; } -// Write vectors will write data to the output buffer +// Writes the data in src to an offset into the dst vector. The offset is specified in bytes +// Returns the number of bytes written into dst. template std::size_t WriteVectors(std::vector& dst, const std::vector& src, std::size_t offset) { if (src.empty()) { return 0; - } else { - std::memcpy(dst.data() + offset, src.data(), src.size() * sizeof(T)); - offset += src.size() * sizeof(T); - return offset; } + const size_t bytes_copied = src.size() * sizeof(T); + std::memcpy(dst.data() + offset, src.data(), bytes_copied); + return bytes_copied; } } // Anonymous namespace @@ -62,7 +65,6 @@ NvResult nvhost_nvdec_common::Submit(const std::vector& input, std::vector command_buffers(params.cmd_buffer_count); std::vector relocs(params.relocation_count); std::vector reloc_shifts(params.relocation_count); @@ -70,13 +72,14 @@ NvResult nvhost_nvdec_common::Submit(const std::vector& input, std::vector wait_checks(params.syncpoint_count); std::vector fences(params.fence_count); - // Splice input into their respective buffers - offset = SpliceVectors(input, command_buffers, params.cmd_buffer_count, offset); - offset = SpliceVectors(input, relocs, params.relocation_count, offset); - offset = SpliceVectors(input, reloc_shifts, params.relocation_count, offset); - offset = SpliceVectors(input, syncpt_increments, params.syncpoint_count, offset); - offset = SpliceVectors(input, wait_checks, params.syncpoint_count, offset); - offset = SpliceVectors(input, fences, params.fence_count, offset); + // Slice input into their respective buffers + std::size_t offset = sizeof(IoctlSubmit); + offset += SliceVectors(input, command_buffers, params.cmd_buffer_count, offset); + offset += SliceVectors(input, relocs, params.relocation_count, offset); + offset += SliceVectors(input, reloc_shifts, params.relocation_count, offset); + offset += SliceVectors(input, syncpt_increments, params.syncpoint_count, offset); + offset += SliceVectors(input, wait_checks, params.syncpoint_count, offset); + offset += SliceVectors(input, fences, params.fence_count, offset); auto& gpu = system.GPU(); if (gpu.UseNvdec()) { @@ -88,7 +91,7 @@ NvResult nvhost_nvdec_common::Submit(const std::vector& input, std::vectorGetObject(cmd_buffer.memory_id); + const auto object = nvmap_dev->GetObject(cmd_buffer.memory_id); ASSERT_OR_EXECUTE(object, return NvResult::InvalidState;); const auto map = FindBufferMap(object->dma_map_addr); if (!map) { @@ -102,21 +105,19 @@ NvResult nvhost_nvdec_common::Submit(const std::vector& input, std::vector& input, std::vecto std::memcpy(¶ms, input.data(), sizeof(IoctlMapBuffer)); std::vector cmd_buffer_handles(params.num_entries); - SpliceVectors(input, cmd_buffer_handles, params.num_entries, sizeof(IoctlMapBuffer)); + SliceVectors(input, cmd_buffer_handles, params.num_entries, sizeof(IoctlMapBuffer)); auto& gpu = system.GPU(); - for (auto& cmf_buff : cmd_buffer_handles) { - auto object{nvmap_dev->GetObject(cmf_buff.map_handle)}; + for (auto& cmd_buffer : cmd_buffer_handles) { + auto object{nvmap_dev->GetObject(cmd_buffer.map_handle)}; if (!object) { - LOG_ERROR(Service_NVDRV, "invalid cmd_buffer nvmap_handle={:X}", cmf_buff.map_handle); + LOG_ERROR(Service_NVDRV, "invalid cmd_buffer nvmap_handle={:X}", cmd_buffer.map_handle); std::memcpy(output.data(), ¶ms, output.size()); return NvResult::InvalidState; } @@ -170,7 +171,7 @@ NvResult nvhost_nvdec_common::MapBuffer(const std::vector& input, std::vecto if (!object->dma_map_addr) { LOG_ERROR(Service_NVDRV, "failed to map size={}", object->size); } else { - cmf_buff.map_address = object->dma_map_addr; + cmd_buffer.map_address = object->dma_map_addr; AddBufferMap(object->dma_map_addr, object->size, object->addr, object->status == nvmap::Object::Status::Allocated); } @@ -186,14 +187,14 @@ NvResult nvhost_nvdec_common::UnmapBuffer(const std::vector& input, std::vec IoctlMapBuffer params{}; std::memcpy(¶ms, input.data(), sizeof(IoctlMapBuffer)); std::vector cmd_buffer_handles(params.num_entries); - SpliceVectors(input, cmd_buffer_handles, params.num_entries, sizeof(IoctlMapBuffer)); + SliceVectors(input, cmd_buffer_handles, params.num_entries, sizeof(IoctlMapBuffer)); auto& gpu = system.GPU(); - for (auto& cmf_buff : cmd_buffer_handles) { - const auto object{nvmap_dev->GetObject(cmf_buff.map_handle)}; + for (auto& cmd_buffer : cmd_buffer_handles) { + const auto object{nvmap_dev->GetObject(cmd_buffer.map_handle)}; if (!object) { - LOG_ERROR(Service_NVDRV, "invalid cmd_buffer nvmap_handle={:X}", cmf_buff.map_handle); + LOG_ERROR(Service_NVDRV, "invalid cmd_buffer nvmap_handle={:X}", cmd_buffer.map_handle); std::memcpy(output.data(), ¶ms, output.size()); return NvResult::InvalidState; } From 8943f2158d0e8a9f708612f1b0ea7051a6b0ddff Mon Sep 17 00:00:00 2001 From: ameerj <52414509+ameerj@users.noreply.github.com> Date: Wed, 14 Jul 2021 23:56:24 -0400 Subject: [PATCH 2/3] nvhost_nvdec_common: Read Submit ioctl data from object addr Fixes Mario Golf intro video decoding. --- .../hle/service/nvdrv/devices/nvhost_nvdec_common.cpp | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-) diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp index f045008eb3..1403a39d0f 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp +++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp @@ -93,15 +93,9 @@ NvResult nvhost_nvdec_common::Submit(const std::vector& input, std::vectorGetObject(cmd_buffer.memory_id); ASSERT_OR_EXECUTE(object, return NvResult::InvalidState;); - const auto map = FindBufferMap(object->dma_map_addr); - if (!map) { - LOG_ERROR(Service_NVDRV, "Tried to submit an invalid offset 0x{:X} dma 0x{:X}", - object->addr, object->dma_map_addr); - return NvResult::Success; - } Tegra::ChCommandHeaderList cmdlist(cmd_buffer.word_count); - gpu.MemoryManager().ReadBlock(map->StartAddr() + cmd_buffer.offset, cmdlist.data(), - cmdlist.size() * sizeof(u32)); + system.Memory().ReadBlock(object->addr + cmd_buffer.offset, cmdlist.data(), + cmdlist.size() * sizeof(u32)); gpu.PushCommandBuffer(cmdlist); } if (gpu.UseNvdec()) { From b7fa264749ae4b8f01341e4a1392a76440a3611d Mon Sep 17 00:00:00 2001 From: ameerj <52414509+ameerj@users.noreply.github.com> Date: Thu, 15 Jul 2021 00:51:50 -0400 Subject: [PATCH 3/3] vic: Fix dimension compuation of YUV frames Fixes out of bound memory crashes in Mario Golf --- src/video_core/command_classes/vic.cpp | 21 ++++++++++----------- 1 file changed, 10 insertions(+), 11 deletions(-) diff --git a/src/video_core/command_classes/vic.cpp b/src/video_core/command_classes/vic.cpp index 0a8b82f2b8..43ac207289 100644 --- a/src/video_core/command_classes/vic.cpp +++ b/src/video_core/command_classes/vic.cpp @@ -119,28 +119,27 @@ void Vic::Execute() { const std::size_t surface_width = config.surface_width_minus1 + 1; const std::size_t surface_height = config.surface_height_minus1 + 1; - const std::size_t half_width = surface_width / 2; - const std::size_t half_height = config.surface_height_minus1 / 2; + const auto frame_width = std::min(surface_width, static_cast(frame->width)); + const auto frame_height = std::min(surface_height, static_cast(frame->height)); + const std::size_t half_width = frame_width / 2; + const std::size_t half_height = frame_height / 2; const std::size_t aligned_width = (surface_width + 0xff) & ~0xff; const auto* luma_ptr = frame->data[0]; const auto* chroma_b_ptr = frame->data[1]; const auto* chroma_r_ptr = frame->data[2]; - const auto stride = frame->linesize[0]; - const auto half_stride = frame->linesize[1]; + const auto stride = static_cast(frame->linesize[0]); + const auto half_stride = static_cast(frame->linesize[1]); luma_buffer.resize(aligned_width * surface_height); - chroma_buffer.resize(aligned_width * half_height); + chroma_buffer.resize(aligned_width * surface_height / 2); // Populate luma buffer - for (std::size_t y = 0; y < surface_height - 1; ++y) { + for (std::size_t y = 0; y < frame_height; ++y) { const std::size_t src = y * stride; const std::size_t dst = y * aligned_width; - - const std::size_t size = surface_width; - - for (std::size_t offset = 0; offset < size; ++offset) { - luma_buffer[dst + offset] = luma_ptr[src + offset]; + for (std::size_t x = 0; x < frame_width; ++x) { + luma_buffer[dst + x] = luma_ptr[src + x]; } } gpu.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(),