return ResultSessionClosed when a session is closed

- [x] I have read and followed the [Contribution Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/CONTRIBUTING.md#code-contributions).
- [x] I have read and followed the [AI Policy](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/AI.md)
- [x] I have read and followed the [Coding Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/Coding.md) to the best of my ability.

-------------------
This one deserves a long story, but imma try to resume:

During investigating Absolum 1.2 black screen of death upon loading intro video, i've discovered guest was aborting for failing to allocate room for the video.

By logging everything prior to crash and decoding guest side instructions managed to confirm its media allocator was reading data faster than it was updating free available bucket list.

Since the IStorage::Read was happening 247 times before the crash, i've decided to add a very small sleep there, and boom, not only Absolum but some other titles got the same issue fixed.

But i was unsatisfied with the sleep and kept tracking guest instructions upstream in order to find a sync point for the read worker and the memory allocation update. But unfortunately the media allocator helpers live in guest, accessing memory directly via MMU, so any sync signaling would need to come from some dynarmic hack.

It's been 6 days now, so i've decided to polish the sleep: Moved it upstream to where i could have access for proper predicate, and added machinery to service and nvhost to support that. Now the sleep is restricted only for nvdec istorage reads. Any other reads will flow normally.

TL;DR: currently our code is so blazing async that guest is capable to request reads before its very self refresh it have freed room to do so. The sleep accepted as broadly stable was 600 us (MICROseconds), and it affects ONLY nvdec chunk reading.
Reports confirm that now videos are smoother now.

Code was polished at my knowledge limits.
Mostly machinery to track when a request comes from a process with nvdec active, and is istorage read.
I can provide more details if it comes to be needed.

Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4316
Reviewed-by: MaranBr <maranbr@eden-emu.dev>
Reviewed-by: Samuel <lizzie@eden-emu.dev>

src/common/CMakeLists.txt

@@ -63,6 +63,8 @@ add_library(
     fs/path_util.cpp
     fs/path_util.h
     hash.h
+    heap_tracker.cpp
+    heap_tracker.h
     hex_util.cpp
     hex_util.h
     host_memory.cpp

src/common/heap_tracker.cpp

@@ -0,0 +1,282 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <fstream>
+#include "common/heap_tracker.h"
+#include "common/logging.h"
+#include "common/assert.h"
+
+namespace Common {
+
+namespace {
+
+s64 GetMaxPermissibleResidentMapCount() {
+    // Default value.
+    s64 value = 65530;
+
+    // Try to read how many mappings we can make.
+    std::ifstream s("/proc/sys/vm/max_map_count");
+    s >> value;
+
+    // Print, for debug.
+    LOG_INFO(HW_Memory, "Current maximum map count: {}", value);
+
+    // Allow 20000 maps for other code and to account for split inaccuracy.
+    return std::max<s64>(value - 20000, 0);
+}
+
+} // namespace
+
+HeapTracker::HeapTracker(Common::HostMemory& buffer)
+    : m_buffer(buffer), m_max_resident_map_count(GetMaxPermissibleResidentMapCount()) {}
+HeapTracker::~HeapTracker() = default;
+
+void HeapTracker::Map(size_t virtual_offset, size_t host_offset, size_t length,
+                      MemoryPermission perm, bool is_separate_heap) {
+    // When mapping other memory, map pages immediately.
+    if (!is_separate_heap) {
+        m_buffer.Map(virtual_offset, host_offset, length, perm, false);
+        return;
+    }
+
+    {
+        // We are mapping part of a separate heap.
+        std::scoped_lock lk{m_lock};
+
+        auto* const map = new SeparateHeapMap{
+            .vaddr = virtual_offset,
+            .paddr = host_offset,
+            .size = length,
+            .tick = m_tick++,
+            .perm = perm,
+            .is_resident = false,
+        };
+
+        // Insert into mappings.
+        m_map_count++;
+        m_mappings.insert(*map);
+    }
+
+    // Finally, map.
+    this->DeferredMapSeparateHeap(virtual_offset);
+}
+
+void HeapTracker::Unmap(size_t virtual_offset, size_t size, bool is_separate_heap) {
+    // If this is a separate heap...
+    if (is_separate_heap) {
+        std::scoped_lock lk{m_lock};
+
+        const SeparateHeapMap key{
+            .vaddr = virtual_offset,
+        };
+
+        // Split at the boundaries of the region we are removing.
+        this->SplitHeapMapLocked(virtual_offset);
+        this->SplitHeapMapLocked(virtual_offset + size);
+
+        // Erase all mappings in range.
+        auto it = m_mappings.find(key);
+        while (it != m_mappings.end() && it->vaddr < virtual_offset + size) {
+            // Get underlying item.
+            auto* const item = std::addressof(*it);
+
+            // If resident, erase from resident map.
+            if (item->is_resident) {
+                ASSERT(--m_resident_map_count >= 0);
+                m_resident_mappings.erase(m_resident_mappings.iterator_to(*item));
+            }
+
+            // Erase from map.
+            ASSERT(--m_map_count >= 0);
+            it = m_mappings.erase(it);
+
+            // Free the item.
+            delete item;
+        }
+    }
+
+    // Unmap pages.
+    m_buffer.Unmap(virtual_offset, size, false);
+}
+
+void HeapTracker::Protect(size_t virtual_offset, size_t size, MemoryPermission perm) {
+    // Ensure no rebuild occurs while reprotecting.
+    std::shared_lock lk{m_rebuild_lock};
+
+    // Split at the boundaries of the region we are reprotecting.
+    this->SplitHeapMap(virtual_offset, size);
+
+    // Declare tracking variables.
+    const VAddr end = virtual_offset + size;
+    VAddr cur = virtual_offset;
+
+    while (cur < end) {
+        VAddr next = cur;
+        bool should_protect = false;
+
+        {
+            std::scoped_lock lk2{m_lock};
+
+            const SeparateHeapMap key{
+                .vaddr = next,
+            };
+
+            // Try to get the next mapping corresponding to this address.
+            const auto it = m_mappings.nfind(key);
+
+            if (it == m_mappings.end()) {
+                // There are no separate heap mappings remaining.
+                next = end;
+                should_protect = true;
+            } else if (it->vaddr == cur) {
+                // We are in range.
+                // Update permission bits.
+                it->perm = perm;
+
+                // Determine next address and whether we should protect.
+                next = cur + it->size;
+                should_protect = it->is_resident;
+            } else /* if (it->vaddr > cur) */ {
+                // We weren't in range, but there is a block coming up that will be.
+                next = it->vaddr;
+                should_protect = true;
+            }
+        }
+
+        // Clamp to end.
+        next = (std::min)(next, end);
+        // Reprotect, if we need to.
+        if (should_protect) {
+            m_buffer.Protect(cur, next - cur, perm);
+        }
+
+        // Advance.
+        cur = next;
+    }
+}
+
+bool HeapTracker::DeferredMapSeparateHeap(u8* fault_address) {
+    if (m_buffer.IsInVirtualRange(fault_address)) {
+        return this->DeferredMapSeparateHeap(fault_address - m_buffer.VirtualBasePointer());
+    }
+
+    return false;
+}
+
+bool HeapTracker::DeferredMapSeparateHeap(size_t virtual_offset) {
+    bool rebuild_required = false;
+
+    {
+        std::scoped_lock lk{m_lock};
+
+        // Check to ensure this was a non-resident separate heap mapping.
+        const auto it = this->GetNearestHeapMapLocked(virtual_offset);
+        if (it == m_mappings.end() || it->is_resident) {
+            return false;
+        }
+
+        // Update tick before possible rebuild.
+        it->tick = m_tick++;
+
+        // Check if we need to rebuild.
+        if (m_resident_map_count > m_max_resident_map_count) {
+            rebuild_required = true;
+        }
+
+        // Map the area.
+        m_buffer.Map(it->vaddr, it->paddr, it->size, it->perm, false);
+
+        // This map is now resident.
+        it->is_resident = true;
+        m_resident_map_count++;
+        m_resident_mappings.insert(*it);
+    }
+
+    if (rebuild_required) {
+        // A rebuild was required, so perform it now.
+        this->RebuildSeparateHeapAddressSpace();
+    }
+
+    return true;
+}
+
+void HeapTracker::RebuildSeparateHeapAddressSpace() {
+    std::scoped_lock lk{m_rebuild_lock, m_lock};
+
+    ASSERT(!m_resident_mappings.empty());
+
+    // Dump half of the mappings.
+    //
+    // Despite being worse in theory, this has proven to be better in practice than more
+    // regularly dumping a smaller amount, because it significantly reduces average case
+    // lock contention.
+    std::size_t const desired_count = (std::min)(m_resident_map_count, m_max_resident_map_count) / 2;
+    std::size_t const evict_count = m_resident_map_count - desired_count;
+    auto it = m_resident_mappings.begin();
+
+    for (size_t i = 0; i < evict_count && it != m_resident_mappings.end(); i++) {
+        // Unmark and unmap.
+        it->is_resident = false;
+        m_buffer.Unmap(it->vaddr, it->size, false);
+
+        // Advance.
+        ASSERT(--m_resident_map_count >= 0);
+        it = m_resident_mappings.erase(it);
+    }
+}
+
+void HeapTracker::SplitHeapMap(VAddr offset, size_t size) {
+    std::scoped_lock lk{m_lock};
+
+    this->SplitHeapMapLocked(offset);
+    this->SplitHeapMapLocked(offset + size);
+}
+
+void HeapTracker::SplitHeapMapLocked(VAddr offset) {
+    const auto it = this->GetNearestHeapMapLocked(offset);
+    if (it == m_mappings.end() || it->vaddr == offset) {
+        // Not contained or no split required.
+        return;
+    }
+
+    // Cache the original values.
+    auto* const left = std::addressof(*it);
+    const size_t orig_size = left->size;
+
+    // Adjust the left map.
+    const size_t left_size = offset - left->vaddr;
+    left->size = left_size;
+
+    // Create the new right map.
+    auto* const right = new SeparateHeapMap{
+        .vaddr = left->vaddr + left_size,
+        .paddr = left->paddr + left_size,
+        .size = orig_size - left_size,
+        .tick = left->tick,
+        .perm = left->perm,
+        .is_resident = left->is_resident,
+    };
+
+    // Insert the new right map.
+    m_map_count++;
+    m_mappings.insert(*right);
+
+    // If resident, also insert into resident map.
+    if (right->is_resident) {
+        m_resident_map_count++;
+        m_resident_mappings.insert(*right);
+    }
+}
+
+HeapTracker::AddrTree::iterator HeapTracker::GetNearestHeapMapLocked(VAddr offset) {
+    const SeparateHeapMap key{
+        .vaddr = offset,
+    };
+
+    return m_mappings.find(key);
+}
+
+} // namespace Common

src/common/heap_tracker.h

@@ -0,0 +1,98 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include <atomic>
+#include <mutex>
+#include <set>
+#include <shared_mutex>
+
+#include "common/host_memory.h"
+#include "common/intrusive_red_black_tree.h"
+
+namespace Common {
+
+struct SeparateHeapMap {
+    Common::IntrusiveRedBlackTreeNode addr_node{};
+    Common::IntrusiveRedBlackTreeNode tick_node{};
+    VAddr vaddr{};
+    PAddr paddr{};
+    size_t size{};
+    size_t tick{};
+    MemoryPermission perm{};
+    bool is_resident{};
+};
+
+struct SeparateHeapMapAddrComparator {
+    static constexpr int Compare(const SeparateHeapMap& lhs, const SeparateHeapMap& rhs) {
+        if (lhs.vaddr < rhs.vaddr) {
+            return -1;
+        } else if (lhs.vaddr <= (rhs.vaddr + rhs.size - 1)) {
+            return 0;
+        } else {
+            return 1;
+        }
+    }
+};
+
+struct SeparateHeapMapTickComparator {
+    static constexpr int Compare(const SeparateHeapMap& lhs, const SeparateHeapMap& rhs) {
+        if (lhs.tick < rhs.tick) {
+            return -1;
+        } else if (lhs.tick > rhs.tick) {
+            return 1;
+        } else {
+            return SeparateHeapMapAddrComparator::Compare(lhs, rhs);
+        }
+    }
+};
+
+class HeapTracker {
+public:
+    explicit HeapTracker(Common::HostMemory& buffer);
+    ~HeapTracker();
+
+    void Map(size_t virtual_offset, size_t host_offset, size_t length, MemoryPermission perm,
+             bool is_separate_heap);
+    void Unmap(size_t virtual_offset, size_t size, bool is_separate_heap);
+    void Protect(size_t virtual_offset, size_t length, MemoryPermission perm);
+    u8* VirtualBasePointer() {
+        return m_buffer.VirtualBasePointer();
+    }
+
+    bool DeferredMapSeparateHeap(u8* fault_address);
+    bool DeferredMapSeparateHeap(size_t virtual_offset);
+
+private:
+    using AddrTreeTraits =
+        Common::IntrusiveRedBlackTreeMemberTraitsDeferredAssert<&SeparateHeapMap::addr_node>;
+    using AddrTree = AddrTreeTraits::TreeType<SeparateHeapMapAddrComparator>;
+
+    using TickTreeTraits =
+        Common::IntrusiveRedBlackTreeMemberTraitsDeferredAssert<&SeparateHeapMap::tick_node>;
+    using TickTree = TickTreeTraits::TreeType<SeparateHeapMapTickComparator>;
+
+    AddrTree m_mappings{};
+    TickTree m_resident_mappings{};
+
+private:
+    void SplitHeapMap(VAddr offset, size_t size);
+    void SplitHeapMapLocked(VAddr offset);
+
+    AddrTree::iterator GetNearestHeapMapLocked(VAddr offset);
+
+    void RebuildSeparateHeapAddressSpace();
+
+private:
+    Common::HostMemory& m_buffer;
+    const s64 m_max_resident_map_count;
+
+    std::shared_mutex m_rebuild_lock{};
+    std::mutex m_lock{};
+    s64 m_map_count{};
+    s64 m_resident_map_count{};
+    size_t m_tick{};
+};
+
+} // namespace Common

src/core/core.cpp

@@ -4,6 +4,7 @@
 #include <array>
 #include <atomic>
 #include <memory>
+#include <unordered_map>
 #include <utility>
 
 #include "game_settings.h"
@@ -248,12 +249,30 @@ struct System::Impl {
         }
     }
 
-    void SetNVDECActive(bool is_nvdec_active) {
-        nvdec_active = is_nvdec_active;
+    void NotifyNVDECChannelOpen(u64 process_id) {
+        std::scoped_lock lock{nvdec_active_mutex};
+        ++nvdec_active_channels[process_id];
+    }
+
+    void NotifyNVDECChannelClose(u64 process_id) {
+        std::scoped_lock lock{nvdec_active_mutex};
+        const auto it = nvdec_active_channels.find(process_id);
+        if (it == nvdec_active_channels.end()) {
+            return;
+        }
+        if (--it->second == 0) {
+            nvdec_active_channels.erase(it);
+        }
     }
 
     bool GetNVDECActive() {
-        return nvdec_active;
+        std::scoped_lock lock{nvdec_active_mutex};
+        return !nvdec_active_channels.empty();
+    }
+
+    bool IsNVDECActiveForProcess(u64 process_id) {
+        std::scoped_lock lock{nvdec_active_mutex};
+        return nvdec_active_channels.contains(process_id);
     }
 
     void InitializeDebugger(System& system, u16 port) {
@@ -505,6 +524,8 @@ struct System::Impl {
 
     mutable std::mutex suspend_guard;
     std::mutex general_channel_mutex;
+    std::mutex nvdec_active_mutex;
+    std::unordered_map<u64, u32> nvdec_active_channels;
     std::atomic_bool is_paused{};
     std::atomic_bool is_shutting_down{};
     std::atomic_bool is_powered_on{};
@@ -512,7 +533,6 @@ struct System::Impl {
     bool extended_memory_layout : 1 = false;
     bool exit_locked : 1 = false;
     bool exit_requested : 1 = false;
-    bool nvdec_active : 1 = false;
 
     void EnsureGeneralChannelInitialized(System& system) {
         if (!general_channel_event) {
@@ -576,14 +596,22 @@ void System::UnstallApplication() {
     impl->UnstallApplication();
 }
 
-void System::SetNVDECActive(bool is_nvdec_active) {
-    impl->SetNVDECActive(is_nvdec_active);
+void System::NotifyNVDECChannelOpen(u64 process_id) {
+    impl->NotifyNVDECChannelOpen(process_id);
+}
+
+void System::NotifyNVDECChannelClose(u64 process_id) {
+    impl->NotifyNVDECChannelClose(process_id);
 }
 
 bool System::GetNVDECActive() {
     return impl->GetNVDECActive();
 }
 
+bool System::IsNVDECActiveForProcess(u64 process_id) {
+    return impl->IsNVDECActiveForProcess(process_id);
+}
+
 void System::InitializeDebugger() {
     impl->InitializeDebugger(*this, Settings::values.gdbstub_port.GetValue());
 }

src/core/core.h

@@ -191,8 +191,10 @@ public:
     std::unique_lock<std::mutex> StallApplication();
     void UnstallApplication();
 
-    void SetNVDECActive(bool is_nvdec_active);
+    void NotifyNVDECChannelOpen(u64 process_id);
+    void NotifyNVDECChannelClose(u64 process_id);
     [[nodiscard]] bool GetNVDECActive();
+    [[nodiscard]] bool IsNVDECActiveForProcess(u64 process_id);
 
     /**
      * Initialize the debugger.

src/core/hle/kernel/k_server_session.cpp

@@ -1216,7 +1216,7 @@ Result KServerSession::ReceiveRequest(KernelCore& kernel, uintptr_t server_messa
 }
 
 Result KServerSession::SendReply(KernelCore& kernel, uintptr_t server_message, uintptr_t server_buffer_size,
-                                 KPhysicalAddress server_message_paddr, bool is_hle) {
+                                 KPhysicalAddress server_message_paddr, bool is_hle, bool session_closed) {
     // Lock the session.
     KScopedLightLock lk{m_lock};
 
@@ -1248,7 +1248,7 @@ Result KServerSession::SendReply(KernelCore& kernel, uintptr_t server_message, u
     KEvent* event = request->GetEvent();
 
     // Check whether we're closed.
-    const bool closed = (client_thread == nullptr || m_parent->IsClientClosed());
+    const bool closed = (client_thread == nullptr || m_parent->IsClientClosed() || session_closed);
 
     Result result = ResultSuccess;
     if (!closed) {

src/core/hle/kernel/k_server_session.h

@@ -54,14 +54,14 @@ public:
 
     Result OnRequest(KernelCore& kernel, KSessionRequest* request);
     Result SendReply(KernelCore& kernel, uintptr_t server_message, uintptr_t server_buffer_size,
-                     KPhysicalAddress server_message_paddr, bool is_hle = false);
+                     KPhysicalAddress server_message_paddr, bool is_hle = false, bool session_closed = false);
     Result ReceiveRequest(KernelCore& kernel, uintptr_t server_message, uintptr_t server_buffer_size,
                           KPhysicalAddress server_message_paddr,
                           std::shared_ptr<Service::HLERequestContext>* out_context = nullptr,
                           std::weak_ptr<Service::SessionRequestManager> manager = {});
 
-    Result SendReplyHLE(KernelCore& kernel) {
-        R_RETURN(this->SendReply(kernel, 0, 0, 0, true));
+    Result SendReplyHLE(KernelCore& kernel, bool session_closed = false) {
+        R_RETURN(this->SendReply(kernel, 0, 0, 0, true, session_closed));
     }
 
     Result ReceiveRequestHLE(KernelCore& kernel, std::shared_ptr<Service::HLERequestContext>* out_context,

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

@@ -8,6 +8,7 @@
 #include "common/assert.h"
 #include "common/logging.h"
 #include "core/core.h"
+#include "core/hle/kernel/k_process.h"
 #include "core/hle/service/nvdrv/core/container.h"
 #include "core/hle/service/nvdrv/devices/ioctl_serialization.h"
 #include "core/hle/service/nvdrv/devices/nvhost_nvdec.h"
@@ -71,17 +72,23 @@ NvResult nvhost_nvdec::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> in
 
 void nvhost_nvdec::OnOpen(NvCore::SessionId session_id, DeviceFD fd) {
     LOG_INFO(Service_NVDRV, "NVDEC video stream started");
-    system.SetNVDECActive(true);
     sessions[fd] = session_id;
+    if (const auto* session = core.GetSession(session_id);
+        session != nullptr && session->process != nullptr) {
+        system.NotifyNVDECChannelOpen(session->process->GetId());
+    }
     host1x.StartDevice(fd, Tegra::Host1x::ChannelType::NvDec, channel_syncpoint);
 }
 
 void nvhost_nvdec::OnClose(DeviceFD fd) {
     LOG_INFO(Service_NVDRV, "NVDEC video stream ended");
     host1x.StopDevice(fd, Tegra::Host1x::ChannelType::NvDec);
-    system.SetNVDECActive(false);
     auto it = sessions.find(fd);
     if (it != sessions.end()) {
+        if (const auto* session = core.GetSession(it->second);
+            session != nullptr && session->process != nullptr) {
+            system.NotifyNVDECChannelClose(session->process->GetId());
+        }
         sessions.erase(it);
     }
 }

src/core/hle/service/server_manager.cpp

@@ -393,7 +393,7 @@ Result ServerManager::CompleteSyncRequest(Session* session) {
     }
 
     // Send the reply.
-    res = server_session->SendReplyHLE(m_system.Kernel());
+    res = server_session->SendReplyHLE(m_system.Kernel(), service_res == IPC::ResultSessionClosed);
 
     // If the session has been closed, we're done.
     if (res == Kernel::ResultSessionClosed || service_res == IPC::ResultSessionClosed) {

src/core/hle/service/service.cpp

@@ -4,12 +4,16 @@
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <chrono>
 #include <fmt/ranges.h>
+#include <string_view>
+#include <thread>
 #include "common/assert.h"
 #include "common/logging.h"
 #include "common/settings.h"
 #include "core/core.h"
 #include "core/hle/ipc.h"
+#include "core/hle/kernel/k_process.h"
 #include "core/hle/kernel/kernel.h"
 #include "core/hle/service/ipc_helpers.h"
 #include "core/hle/service/service.h"
@@ -33,6 +37,7 @@ ServiceFrameworkBase::ServiceFrameworkBase(Core::System& system_, const char* se
     : SessionRequestHandler(system_.Kernel(), service_name_)
     , system{system_}
     , service_name{service_name_}
+    , is_i_storage{std::string_view{service_name_} == "IStorage"}
     , handler_invoker{handler_invoker_}
     , max_sessions{max_sessions_}
 {}
@@ -77,13 +82,22 @@ void ServiceFrameworkBase::ReportUnimplementedFunction(HLERequestContext& ctx,
 }
 
 void ServiceFrameworkBase::InvokeRequest(HLERequestContext& ctx) {
-    auto it = handlers.find(ctx.GetCommand());
+    const auto command = ctx.GetCommand();
+    auto it = handlers.find(command);
+    const bool is_cmd_read = command == 0;
     FunctionInfoBase const* info = it == handlers.end() ? nullptr : &it->second;
     if (info == nullptr || info->handler_callback == nullptr)
         return ReportUnimplementedFunction(ctx, info);
 
     LOG_TRACE(Service, "{}", MakeFunctionString(info->name, GetServiceName(), ctx.CommandBuffer()));
     handler_invoker(this, info->handler_callback, ctx);
+
+    if (is_i_storage && is_cmd_read) {
+        const auto* const process = ctx.GetThread().GetOwnerProcess();
+        if (process != nullptr && system.IsNVDECActiveForProcess(process->GetId())) {
+            std::this_thread::sleep_for(std::chrono::microseconds{600});
+        }
+    }
 }
 
 void ServiceFrameworkBase::InvokeRequestTipc(HLERequestContext& ctx) {

src/core/hle/service/service.h

@@ -107,6 +107,8 @@ protected:
     Core::System& system;
     /// Identifier string used to connect to the service.
     const char* service_name;
+    /// Whether this is the IStorage service.
+    const bool is_i_storage;
     /// Function used to safely up-cast pointers to the derived class before invoking a handler.
     InvokerFn* handler_invoker;
     /// Maximum number of concurrent sessions that this service can handle.

src/core/memory.cpp

@@ -16,6 +16,7 @@
 #include "common/assert.h"
 #include "common/atomic_ops.h"
 #include "common/common_types.h"
+#include "common/heap_tracker.h"
 #include "common/logging.h"
 #include "common/page_table.h"
 #include "common/scope_exit.h"
@@ -54,24 +55,36 @@ struct Memory::Impl {
         } else {
             current_page_table->fastmem_arena = nullptr;
         }
+
+#ifdef __ANDROID__
+        heap_tracker.emplace(system.DeviceMemory().buffer);
+        host_buffer = std::addressof(*heap_tracker);
+#else
         host_buffer = std::addressof(system.DeviceMemory().buffer);
+#endif
     }
 
-    void MapMemoryRegion(Common::PageTable& page_table, Common::ProcessAddress base, u64 size, Common::PhysicalAddress target, Common::MemoryPermission perms, bool separate_heap) {
+    void MapMemoryRegion(Common::PageTable& page_table, Common::ProcessAddress base, u64 size,
+                         Common::PhysicalAddress target, Common::MemoryPermission perms,
+                         bool separate_heap) {
         ASSERT_MSG((size & YUZU_PAGEMASK) == 0, "non-page aligned size: {:016X}", size);
         ASSERT_MSG((base & YUZU_PAGEMASK) == 0, "non-page aligned base: {:016X}", GetInteger(base));
-        ASSERT_MSG(target >= DramMemoryMap::Base, "Out of bounds target: {:016X}", GetInteger(target));
-        MapPages(page_table, base / YUZU_PAGESIZE, size / YUZU_PAGESIZE, target, Common::PageType::Memory);
+        ASSERT_MSG(target >= DramMemoryMap::Base, "Out of bounds target: {:016X}",
+                   GetInteger(target));
+        MapPages(page_table, base / YUZU_PAGESIZE, size / YUZU_PAGESIZE, target,
+                 Common::PageType::Memory);
 
         if (current_page_table->fastmem_arena) {
             host_buffer->Map(GetInteger(base), GetInteger(target) - DramMemoryMap::Base, size, perms, separate_heap);
         }
     }
 
-    void UnmapRegion(Common::PageTable& page_table, Common::ProcessAddress base, u64 size, bool separate_heap) {
+    void UnmapRegion(Common::PageTable& page_table, Common::ProcessAddress base, u64 size,
+                     bool separate_heap) {
         ASSERT_MSG((size & YUZU_PAGEMASK) == 0, "non-page aligned size: {:016X}", size);
         ASSERT_MSG((base & YUZU_PAGEMASK) == 0, "non-page aligned base: {:016X}", GetInteger(base));
-        MapPages(page_table, base / YUZU_PAGESIZE, size / YUZU_PAGESIZE, 0, Common::PageType::Unmapped);
+        MapPages(page_table, base / YUZU_PAGESIZE, size / YUZU_PAGESIZE, 0,
+                 Common::PageType::Unmapped);
 
         if (current_page_table->fastmem_arena) {
             host_buffer->Unmap(GetInteger(base), size, separate_heap);
@@ -754,7 +767,12 @@ struct Memory::Impl {
     std::array<Common::ScratchBuffer<u32>, Core::Hardware::NUM_CPU_CORES> scratch_buffers{};
     std::span<Core::GPUDirtyMemoryManager> gpu_dirty_managers;
     std::mutex sys_core_guard;
+#ifdef __ANDROID__
+    std::optional<Common::HeapTracker> heap_tracker;
+    Common::HeapTracker* host_buffer{};
+#else
     Common::HostMemory* host_buffer{};
+#endif
 };
 
 Memory::Memory(Core::System& system_) : system{system_} {
@@ -947,14 +965,30 @@ bool Memory::InvalidateNCE(Common::ProcessAddress vaddr, size_t size) {
     u8* const ptr = impl->GetPointerImpl(
         GetInteger(vaddr),
         [&] {
-            LOG_ERROR(HW_Memory, "Unmapped InvalidateNCE for {} bytes @ {:#x}", size, GetInteger(vaddr));
+            LOG_ERROR(HW_Memory, "Unmapped InvalidateNCE for {} bytes @ {:#x}", size,
+                      GetInteger(vaddr));
             mapped = false;
         },
         [&] { rasterizer = true; });
     if (rasterizer) {
         impl->InvalidateGPUMemory(ptr, size);
     }
+
+#ifdef __ANDROID__
+    if (!rasterizer && mapped) {
+        impl->host_buffer->DeferredMapSeparateHeap(GetInteger(vaddr));
+    }
+#endif
+
     return mapped && ptr != nullptr;
 }
 
+bool Memory::InvalidateSeparateHeap(void* fault_address) {
+#ifdef __ANDROID__
+    return impl->host_buffer->DeferredMapSeparateHeap(static_cast<u8*>(fault_address));
+#else
+    return false;
+#endif
+}
+
 } // namespace Core::Memory

src/core/memory.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: 2014 Citra Emulator Project
@@ -490,8 +490,13 @@ public:
      *              marked as debug or non-debug.
      */
     void MarkRegionDebug(Common::ProcessAddress vaddr, u64 size, bool debug);
+
     void SetGPUDirtyManagers(std::span<Core::GPUDirtyMemoryManager> managers);
+
     bool InvalidateNCE(Common::ProcessAddress vaddr, size_t size);
+
+    bool InvalidateSeparateHeap(void* fault_address);
+
 private:
     Core::System& system;