[gpu/nvdrv] Unstub SetErrorNotifier, add PostErrorNotification function (#2500)

This commit should ensure that GPU error handling is handled correctly.
Replace some unimplemented stubs with PostErrorNotification.
I believe this should mitigate some hiccups encountered in unreal engine based games.

Co-authored-by: Maufeat <sahyno1996@gmail.com>
Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/2500
Reviewed-by: Shinmegumi <shinmegumi@eden-emu.dev>
Reviewed-by: Caio Oliveira <caiooliveirafarias0@gmail.com>
Reviewed-by: Maufeat <sahyno1996@gmail.com>
Reviewed-by: Lizzie <lizzie@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_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_INFO(Service_NVDRV, "called, priority={:X}", channel_priority);
@@ -278,7 +382,7 @@ 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}",
@@ -357,6 +461,7 @@ 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;
         }
 
@@ -390,7 +495,11 @@ 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;
     }
 
@@ -409,7 +518,7 @@ 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

@@ -72,6 +72,16 @@ private:
         High = 0x96,
     };
 
+    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{};
     };
@@ -178,6 +188,8 @@ 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{};