Force a submit before any i2m engine writes

We need traps to be inplace so we dont end up overwriting a resource that's being actively used by the current context without setting it to dirty

app/src/main/cpp/skyline/soc/gm20b/channel.cpp

@@ -13,6 +13,6 @@ namespace skyline::soc::gm20b {
           fermi2D(std::make_unique<engine::fermi2d::Fermi2D>(state, *this, macroState, executor)),
           maxwellDma(state, *this, executor),
           keplerCompute(state, *this),
-          inline2Memory(asCtx),
+          inline2Memory(*this),
           gpfifo(state, *this, numEntries) {}
 }

app/src/main/cpp/skyline/soc/gm20b/engines/inline2memory.cpp

@@ -1,11 +1,11 @@
 // SPDX-License-Identifier: MPL-2.0
 // Copyright © 2022 Skyline Team and Contributors (https://github.com/skyline-emu/)
 
-#include <soc/gm20b/gmmu.h>
+#include <soc/gm20b/channel.h>
 #include "inline2memory.h"
 
 namespace skyline::soc::gm20b::engine {
-    Inline2MemoryBackend::Inline2MemoryBackend(std::shared_ptr<AddressSpaceContext> addressSpaceContext) : addressSpaceContext(std::move(addressSpaceContext)) {}
+    Inline2MemoryBackend::Inline2MemoryBackend(ChannelContext &channelCtx) : channelCtx(channelCtx) {}
 
     void Inline2MemoryBackend::LaunchDma(Inline2MemoryBackend::RegisterState &state) {
         writeOffset = 0;
@@ -17,10 +17,12 @@ namespace skyline::soc::gm20b::engine {
         if (state.launchDma.completion == RegisterState::DmaCompletionType::ReleaseSemaphore)
             throw exception("Semaphore release on I2M completion is not supported!");
 
+        channelCtx.executor.Submit();
+
         if (state.launchDma.layout == RegisterState::DmaDstMemoryLayout::Pitch && state.lineCount == 1) {
             // TODO: we can do this with the buffer manager to avoid some overhead in the future
             Logger::Debug("range: 0x{:X} -> 0x{:X}", u64{state.offsetOut}, u64{state.offsetOut} + buffer.size() * 0x4);
-            addressSpaceContext->gmmu.Write(state.offsetOut, span(buffer));
+            channelCtx.asCtx->gmmu.Write(state.offsetOut, span(buffer));
         } else {
             Logger::Warn("Non-linear I2M uploads are not supported!");
         }
@@ -47,7 +49,7 @@ namespace skyline::soc::gm20b::engine {
             CompleteDma(state);
     }
 
-    Inline2Memory::Inline2Memory(std::shared_ptr<AddressSpaceContext> addressSpaceContext) : backend(std::move(addressSpaceContext)) {}
+    Inline2Memory::Inline2Memory(ChannelContext &channelCtx) : backend(channelCtx) {}
 
     __attribute__((always_inline)) void Inline2Memory::CallMethod(u32 method, u32 argument) {
         Logger::Verbose("Called method in I2M: 0x{:X} args: 0x{:X}", method, argument);

app/src/main/cpp/skyline/soc/gm20b/engines/inline2memory.h

@@ -7,7 +7,7 @@
 #include "engine.h"
 
 namespace skyline::soc::gm20b {
-    struct AddressSpaceContext;
+    struct ChannelContext;
 }
 
 namespace skyline::soc::gm20b::engine {
@@ -18,7 +18,7 @@ namespace skyline::soc::gm20b::engine {
       private:
         std::vector<u32> buffer; //!< Temporary buffer to hold data being currently uploaded
         u32 writeOffset{}; //!< Current write offset in words into `buffer`
-        std::shared_ptr<AddressSpaceContext> addressSpaceContext;
+        ChannelContext &channelCtx;
 
       public:
         /**
@@ -124,7 +124,7 @@ namespace skyline::soc::gm20b::engine {
         void CompleteDma(RegisterState &state);
 
       public:
-        Inline2MemoryBackend(std::shared_ptr<AddressSpaceContext> addressSpaceContext);
+        Inline2MemoryBackend(ChannelContext &channelCtx);
 
         /**
          * @brief Should be called when launchDma in `state` is written to
@@ -164,7 +164,7 @@ namespace skyline::soc::gm20b::engine {
         } registers{};
 
       public:
-        Inline2Memory(std::shared_ptr<AddressSpaceContext> addressSpaceContext);
+        Inline2Memory(ChannelContext &channelCtx);
 
         void CallMethod(u32 method, u32 argument);
 

app/src/main/cpp/skyline/soc/gm20b/engines/kepler_compute.cpp

@@ -8,7 +8,7 @@
 
 namespace skyline::soc::gm20b::engine {
     KeplerCompute::KeplerCompute(const DeviceState &state, ChannelContext &channelCtx)
-        : syncpoints(state.soc->host1x.syncpoints), i2m(channelCtx.asCtx) {}
+        : syncpoints(state.soc->host1x.syncpoints), i2m(channelCtx) {}
 
     __attribute__((always_inline)) void KeplerCompute::CallMethod(u32 method, u32 argument) {
         Logger::Verbose("Called method in Kepler compute: 0x{:X} args: 0x{:X}", method, argument);

app/src/main/cpp/skyline/soc/gm20b/engines/maxwell_3d.cpp

@@ -11,7 +11,7 @@ namespace skyline::soc::gm20b::engine::maxwell3d {
         : MacroEngineBase(macroState),
           syncpoints(state.soc->host1x.syncpoints),
           context(*state.gpu, channelCtx, executor),
-          i2m(channelCtx.asCtx),
+          i2m(channelCtx),
           channelCtx(channelCtx) {
         executor.AddFlushCallback([this]() { FlushEngineState(); });
         InitializeRegisters();