[gpu/nvdrv] Remove redundant whitespace

This rewrite attempts to implement a fully correct ZBC (Zero Bandwith Clear) mechanism.
The zbc_mutex attempts to mitigate contention by assuring that only threads which hold the mutex can modify the table.
Notify drivers about the index size, I believe some drivers even need the notification.
Only add new entries if a entry was not previously available.

Reviewed-on: #2501
Reviewed-by: Shinmegumi <shinmegumi@eden-emu.dev>
Reviewed-by: MaranBr <maranbr@eden-emu.dev>
Reviewed-by: crueter <crueter@eden-emu.dev>
Co-authored-by: SDK-Chan <sdkchan@eden-emu.dev>
Co-committed-by: SDK-Chan <sdkchan@eden-emu.dev>

src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp

@@ -256,61 +256,103 @@ NvResult nvhost_ctrl_gpu::ZCullGetInfo(IoctlNvgpuGpuZcullGetInfoArgs& params) {
 }
 
 NvResult nvhost_ctrl_gpu::ZBCSetTable(IoctlZbcSetTable& params) {
-    LOG_DEBUG(Service_NVDRV, "called");
-    ZbcEntry entry = {};
-    std::memset(&entry, 0, sizeof(entry));
-    // TODO(ogniK): What does this even actually do?
-    // TODO(myself): This thing I guess
-    if (params.type == 1) {
-        for (auto i = 0; i < 4; ++i) {
-            entry.color_ds[i] = params.color_ds[i];
-            entry.color_l2[i] = params.color_l2[i];
-        }
-        ASSERT(this->max_color_entries < 16);
-        this->color_entries[this->max_color_entries] = entry;
-        ++this->max_color_entries;
-    } else if (params.type == 2) {
-        entry.depth = params.depth;
-        ASSERT(this->max_depth_entries < 16);
-        this->depth_entries[this->max_depth_entries] = entry;
-        ++this->max_depth_entries;
+    if (params.type > supported_types) {
+        LOG_ERROR(Service_NVDRV, "ZBCSetTable: invalid type {:#X}", params.type);
+        return NvResult::BadParameter;
     }
+
+    std::scoped_lock lk(zbc_mutex);
+
+    switch (static_cast<ZBCTypes>(params.type)) {
+        case ZBCTypes::color: {
+            ZbcColorEntry color_entry{};
+            std::copy_n(std::begin(params.color_ds), color_entry.color_ds.size(), color_entry.color_ds.begin());
+            std::copy_n(std::begin(params.color_l2), color_entry.color_l2.size(), color_entry.color_l2.begin());
+            color_entry.format = params.format;
+            color_entry.ref_cnt = 1u;
+
+            auto color_it = std::ranges::find_if(zbc_colors,
+                                                 [&](const ZbcColorEntry& color_in_question) {
+                                                     return color_entry.format == color_in_question.format &&
+                                                            color_entry.color_ds == color_in_question.color_ds &&
+                                                            color_entry.color_l2 == color_in_question.color_l2;
+                                                 });
+
+            if (color_it != zbc_colors.end()) {
+                ++color_it->ref_cnt;
+                LOG_DEBUG(Service_NVDRV, "ZBCSetTable: reused color entry fmt={:#X}, ref_cnt={:#X}",
+                          params.format, color_it->ref_cnt);
+            } else {
+                zbc_colors.push_back(color_entry);
+                LOG_DEBUG(Service_NVDRV, "ZBCSetTable: added color entry fmt={:#X}, index={:#X}",
+                          params.format, zbc_colors.size() - 1);
+            }
+            break;
+        }
+        case ZBCTypes::depth: {
+            ZbcDepthEntry depth_entry{params.depth, params.format, 1u};
+
+            auto depth_it = std::ranges::find_if(zbc_depths,
+                                                 [&](const ZbcDepthEntry& depth_entry_in_question) {
+                                                     return depth_entry.format == depth_entry_in_question.format &&
+                                                            depth_entry.depth == depth_entry_in_question.depth;
+                                                 });
+
+            if (depth_it != zbc_depths.end()) {
+                ++depth_it->ref_cnt;
+                LOG_DEBUG(Service_NVDRV, "ZBCSetTable: reused depth entry fmt={:#X}, ref_cnt={:#X}",
+                          depth_entry.format, depth_it->ref_cnt);
+            } else {
+                zbc_depths.push_back(depth_entry);
+                LOG_DEBUG(Service_NVDRV, "ZBCSetTable: added depth entry fmt={:#X}, index={:#X}",
+                          depth_entry.format, zbc_depths.size() - 1);
+            }
+        }
+    }
+
     return NvResult::Success;
 }
 
 NvResult nvhost_ctrl_gpu::ZBCQueryTable(IoctlZbcQueryTable& params) {
-    LOG_DEBUG(Service_NVDRV, "called");
-    struct ZbcQueryParams {
-        u32_le color_ds[4];
-        u32_le color_l2[4];
-        u32_le depth;
-        u32_le ref_cnt;
-        u32_le format;
-        u32_le type;
-        u32_le index_size;
-    } entry = {};
-    std::memset(&entry, 0, sizeof(entry));
-    auto const index = params.index_size;
-    if (params.type == 0) { //no
-        entry.index_size = 15;
-    } else if (params.type == 1) { //color
-        ASSERT(index < 16);
-        for (auto i = 0; i < 4; ++i) {
-            params.color_ds[i] = this->color_entries[index].color_ds[i];
-            params.color_l2[i] = this->color_entries[index].color_l2[i];
-        }
-        // TODO: Only if no error thrown (otherwise dont modify)
-        params.format = this->color_entries[index].format;
-        //params.ref_cnt = this->color_entries[index].ref_cnt;
-    } else if (params.type == 2) { //depth
-        ASSERT(index < 16);
-        params.depth = this->depth_entries[index].depth;
-        // TODO: Only if no error thrown (otherwise dont modify)
-        params.format = this->depth_entries[index].format;
-        //params.ref_cnt = this->depth_entries[index].ref_cnt;
-    } else {
-        UNREACHABLE();
+    if (params.type > supported_types) {
+        LOG_ERROR(Service_NVDRV, "ZBCQueryTable: invalid type {:#X}", params.type);
+        return NvResult::BadParameter;
     }
+
+    std::scoped_lock lk(zbc_mutex);
+
+    switch (static_cast<ZBCTypes>(params.type)) {
+        case ZBCTypes::color: {
+            if (params.index_size >= zbc_colors.size()) {
+                LOG_ERROR(Service_NVDRV, "ZBCQueryTable: invalid color index {:#X}", params.index_size);
+                return NvResult::BadParameter;
+            }
+
+            const auto& colors = zbc_colors[params.index_size];
+            std::copy_n(colors.color_ds.begin(), colors.color_ds.size(), std::begin(params.color_ds));
+            std::copy_n(colors.color_l2.begin(), colors.color_l2.size(), std::begin(params.color_l2));
+            params.depth = 0;
+            params.ref_cnt = colors.ref_cnt;
+            params.format = colors.format;
+            params.index_size = static_cast<u32>(zbc_colors.size());
+            break;
+        }
+        case ZBCTypes::depth: {
+            if (params.index_size >= zbc_depths.size()) {
+                LOG_ERROR(Service_NVDRV, "ZBCQueryTable: invalid depth index {:#X}", params.index_size);
+                return NvResult::BadParameter;
+            }
+
+            const auto& depth_entry = zbc_depths[params.index_size];
+            std::fill(std::begin(params.color_ds), std::end(params.color_ds), 0);
+            std::fill(std::begin(params.color_l2), std::end(params.color_l2), 0);
+            params.depth = depth_entry.depth;
+            params.ref_cnt = depth_entry.ref_cnt;
+            params.format = depth_entry.format;
+            params.index_size = static_cast<u32>(zbc_depths.size());
+        }
+    }
+
     return NvResult::Success;
 }
 

src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -34,6 +37,11 @@ public:
     Kernel::KEvent* QueryEvent(u32 event_id) override;
 
 private:
+    enum class ZBCTypes {
+        color = 1,
+        depth = 2,
+    };
+
     struct IoctlGpuCharacteristics {
         u32_le arch;                       // 0x120 (NVGPU_GPU_ARCH_GM200)
         u32_le impl;                       // 0xB (NVGPU_GPU_IMPL_GM20B)
@@ -139,6 +147,21 @@ private:
     };
     static_assert(sizeof(IoctlZbcQueryTable) == 52, "IoctlZbcQueryTable is incorrect size");
 
+    struct ZbcColorEntry {
+        std::array<u32, 4> color_ds{};
+        std::array<u32, 4> color_l2{};
+        u32 format{};
+        u32 ref_cnt{};
+    };
+    static_assert(sizeof(ZbcColorEntry) == 40, "ZbcColorEntry is incorrect size");
+
+    struct ZbcDepthEntry {
+        u32 depth{};
+        u32 format{};
+        u32 ref_cnt{};
+    };
+    static_assert(sizeof(ZbcDepthEntry) == 12, "ZbcDepthEntry is incorrect size");
+
     struct IoctlFlushL2 {
         u32_le flush; // l2_flush | l2_invalidate << 1 | fb_flush << 2
         u32_le reserved;
@@ -182,17 +205,11 @@ private:
     Kernel::KEvent* error_notifier_event;
     Kernel::KEvent* unknown_event;
 
-    struct ZbcEntry {
-        u32_le color_ds[4];
-        u32_le color_l2[4];
-        u32_le depth;
-        u32_le type;
-        u32_le format;
-    };
-    std::array<ZbcEntry, 16> color_entries;
-    std::array<ZbcEntry, 16> depth_entries;
-    u8 max_color_entries;
-    u8 max_depth_entries;
+    // ZBC Tables
+    std::mutex zbc_mutex{};
+    std::vector<ZbcColorEntry> zbc_colors{};
+    std::vector<ZbcDepthEntry> zbc_depths{};
+    const u32 supported_types = 2u;
 };
 
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp

@@ -8,6 +8,7 @@
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "core/core.h"
+#include "core/hle/kernel/k_event.h"
 #include "core/hle/kernel/k_process.h"
 #include "core/hle/service/nvdrv/core/container.h"
 #include "core/hle/service/nvdrv/core/nvmap.h"
@@ -161,11 +162,114 @@ NvResult nvhost_gpu::ZCullBind(IoctlZCullBind& params) {
 }
 
 NvResult nvhost_gpu::SetErrorNotifier(IoctlSetErrorNotifier& params) {
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called, offset={:X}, size={:X}, mem={:X}", params.offset,
-                params.size, params.mem);
+    LOG_DEBUG(Service_NVDRV, "called, offset={:#X}, size={:#X}, mem={:#X}",
+              params.offset, params.size, params.mem);
+
+    if (!params.mem || !params.size) {
+        std::scoped_lock lk(channel_mutex);
+        if (!channel_state->initialized) {
+            LOG_CRITICAL(Service_NVDRV, "No address space bound for setting up error notifier!");
+            return NvResult::NotInitialized;
+        }
+
+        error_notifier_params = {};
+        LOG_DEBUG(Service_NVDRV, "Error notifier disabled!");
+        return NvResult::Success;
+    }
+
+    constexpr u64 error_notification_size = sizeof(IoctlGetErrorNotification);
+    if (params.size < error_notification_size) {
+        LOG_ERROR(Service_NVDRV, "Error notification size: {:#X} is too small. Need at least {:#X}", params.size,
+                  error_notification_size);
+        return NvResult::InvalidSize;
+    }
+
+    auto handle = nvmap.GetHandle(static_cast<NvCore::NvMap::Handle::Id>(params.mem));
+    if (!handle) {
+        LOG_ERROR(Service_NVDRV, "Unknown nvmap handle id {:#X}", params.mem);
+        return NvResult::BadParameter;
+    }
+
+    if (params.offset > handle->size || params.size > (handle->size - params.offset)) {
+        LOG_ERROR(Service_NVDRV, "Error notifier out of bounds: offset={:#X} size={:#X} handle size={:#X}", params.offset,
+                  params.size, handle->size);
+        return NvResult::InvalidSize;
+    }
+
+    u64 write_address, write_offset, handle_id;
+    {
+        std::scoped_lock lk(channel_mutex);
+        if (!channel_state->initialized) {
+            LOG_CRITICAL(Service_NVDRV, "No address space bound for setting up error notifier!");
+            return NvResult::NotInitialized;
+        }
+
+        error_notifier_params = params;
+        write_address = handle->address;
+        write_offset = params.offset;
+        handle_id = handle->id;
+    }
+
+    if (write_address) {
+        IoctlGetErrorNotification error_notification{};
+        error_notification.status = static_cast<u16>(NotifierStatus::NoError);
+        system.ApplicationMemory().WriteBlock(write_address + write_offset, &error_notification, sizeof(error_notification));
+    } else {
+        LOG_WARNING(Service_NVDRV,
+                    "nvmap handle id {:#X} has no virtual address assigned; "
+                    "skipping initialization write for error notification!",
+                    handle_id);
+    }
+
     return NvResult::Success;
 }
 
+void nvhost_gpu::PostErrorNotification(u32 info32, u16 info16, NotifierStatus status) {
+    IoctlSetErrorNotifier error_notifier_params_snapshot{};
+    Kernel::KEvent *error_notifier_event_snapshot{};
+    {
+        std::scoped_lock lk(channel_mutex);
+        error_notifier_params_snapshot = error_notifier_params;
+        error_notifier_event_snapshot = error_notifier_event;
+    }
+
+    if (!error_notifier_params_snapshot.mem || error_notifier_params_snapshot.size < sizeof(IoctlGetErrorNotification)) {
+        LOG_DEBUG(Service_NVDRV, "PostErrorNotification: notifier not configured or too small!");
+        return;
+    }
+
+    auto handle = nvmap.GetHandle(static_cast<NvCore::NvMap::Handle::Id>(error_notifier_params_snapshot.mem));
+    if (!handle || !handle->address) {
+        LOG_ERROR(Service_NVDRV, "PostErrorNotification: invalid handle or virtual address!");
+        return;
+    }
+
+    IoctlGetErrorNotification error_init{};
+    error_init.info32 = info32;
+    error_init.info16 = info16;
+    error_init.status = static_cast<u16>(status);
+    const u64 write_size = std::min<u64>(sizeof(IoctlGetErrorNotification),
+                                         error_notifier_params_snapshot.size);
+    if (error_notifier_params_snapshot.offset >= handle->size ||
+        write_size > (handle->size - error_notifier_params_snapshot.offset)) {
+        LOG_ERROR(Service_NVDRV, "PostErrorNotification: bounds check failed!");
+        return;
+    }
+
+    const u64 virtual_address = handle->address + error_notifier_params_snapshot.offset;
+    if (virtual_address < handle->address) {
+        LOG_ERROR(Service_NVDRV, "PostErrorNotification: virtual address overflow!");
+        return;
+    }
+
+    auto &application_memory = system.ApplicationMemory();
+    application_memory.WriteBlock(virtual_address, &error_init, write_size);
+
+    if (error_notifier_event_snapshot) {
+        error_notifier_event_snapshot->Signal();
+    }
+}
+
 NvResult nvhost_gpu::SetChannelPriority(IoctlChannelSetPriority& params) {
     channel_priority = params.priority;
     LOG_DEBUG(Service_NVDRV, "(STUBBED) called, priority={:X}", channel_priority);
@@ -251,7 +355,7 @@ NvResult nvhost_gpu::AllocateObjectContext(IoctlAllocObjCtx& params) {
         params.flags = allowed_mask;
     }
 
-    s32_le ctx_class_number_index = 
+    s32_le ctx_class_number_index =
         GetObjectContextClassNumberIndex(static_cast<CtxClasses>(params.class_num));
     if (ctx_class_number_index < 0) {
         LOG_ERROR(Service_NVDRV, "Invalid class number for object context: {:#X}",
@@ -324,6 +428,7 @@ NvResult nvhost_gpu::SubmitGPFIFOImpl(IoctlSubmitGpfifo& params, Tegra::CommandL
 
     if (flags.fence_wait.Value()) {
         if (flags.increment_value.Value()) {
+            PostErrorNotification(flags.raw, 0, NotifierStatus::GenericError);
             return NvResult::BadParameter;
         }
 
@@ -357,7 +462,11 @@ NvResult nvhost_gpu::SubmitGPFIFOImpl(IoctlSubmitGpfifo& params, Tegra::CommandL
 NvResult nvhost_gpu::SubmitGPFIFOBase1(IoctlSubmitGpfifo& params,
                                        std::span<Tegra::CommandListHeader> commands, bool kickoff) {
     if (params.num_entries > commands.size()) {
-        UNIMPLEMENTED();
+        LOG_ERROR(Service_NVDRV,
+                  "SubmitGPFIFO: num_entries={:#X} > provided commands={:#X}",
+                  params.num_entries, commands.size());
+
+        PostErrorNotification(params.num_entries, 0, NotifierStatus::BadGpfifo);
         return NvResult::InvalidSize;
     }
 
@@ -376,7 +485,7 @@ NvResult nvhost_gpu::SubmitGPFIFOBase1(IoctlSubmitGpfifo& params,
 NvResult nvhost_gpu::SubmitGPFIFOBase2(IoctlSubmitGpfifo& params,
                                        std::span<const Tegra::CommandListHeader> commands) {
     if (params.num_entries > commands.size()) {
-        UNIMPLEMENTED();
+        PostErrorNotification(params.num_entries, 0, NotifierStatus::BadGpfifo);
         return NvResult::InvalidSize;
     }
 

src/core/hle/service/nvdrv/devices/nvhost_gpu.h

@@ -66,6 +66,16 @@ private:
         CtxChannelGPFIFO = 0xB06F,
     };
 
+    enum class NotifierStatus : u16_le {
+        NoError      = 0xFFFF,
+        GenericError = 0x0001,
+        MmuFault     = 0x0002,
+        IllegalMethod= 0x0003,
+        InvalidObject= 0x0004,
+        BadGpfifo    = 0x0005,
+        TimeoutHang  = 0x0006,
+    };
+
     struct IoctlSetNvmapFD {
         s32_le nvmap_fd{};
     };
@@ -172,6 +182,8 @@ private:
     s32_le nvmap_fd{};
     u64_le user_data{};
     IoctlZCullBind zcull_params{};
+    IoctlSetErrorNotifier error_notifier_params{};
+    void PostErrorNotification(u32 info32, u16 info16, NotifierStatus status);
     std::array<std::optional<IoctlAllocObjCtx>, 6> ctxObjs{};
     u32_le channel_priority{};
     u32_le channel_timeslice{};