[loader] add ZBIC (zstd variant) NSO decompression support for Switch 22.0+

src/common/CMakeLists.txt

@@ -137,6 +137,11 @@ add_library(
     uuid.cpp
     uuid.h
     vector_math.h
+    zbic_compression.cpp
+    zbic_compression.h
+    zstd.c
+    zstd.h
+    zstd_errors.h
     zstd_compression.cpp
     zstd_compression.h
     fs/ryujinx_compat.h fs/ryujinx_compat.cpp
@@ -241,6 +246,7 @@ endif()
 
 target_link_libraries(common PUBLIC fmt::fmt stb::headers Threads::Threads)
 target_link_libraries(common PRIVATE lz4::lz4 zstd::zstd)
+target_compile_definitions(common PRIVATE ZSTD_ZBIC_SUPPORT=1)
 
 # Please refer to src/common/demangle.cpp
 if (WIN32)

src/common/sparse_large_vector.cpp

@@ -8,14 +8,7 @@
 #ifdef _WIN32
 #include <windows.h>
 #include <mutex>
-#include <algorithm>
-#include <vector>
-#endif
-
-#include <cerrno>
-#include <cstring>
-
-#ifndef _WIN32
+#else
 #include <sys/mman.h>
 #endif
 
@@ -26,60 +19,54 @@
 namespace Common {
 
 #ifdef _WIN32
+static std::vector<std::pair<u64, u64>> vector_regions {};
 
-struct VectorRegion {
-    u64 start_page;
-    u64 end_page;
-};
-
-static std::mutex& GetVectorRegionsMutex() {
-    static std::mutex* m = new std::mutex();
-    return *m;
-}
-
-static std::vector<VectorRegion>& GetVectorRegions() {
-    static std::vector<VectorRegion>* v = new std::vector<VectorRegion>();
-    return *v;
-}
-
+// Workaround for handling non-commited memory accessed by Dynarmic; usually result of an error
 static LONG WINAPI FakePageFaultHandler(PEXCEPTION_POINTERS info) {
-    if (info->ExceptionRecord->ExceptionCode != EXCEPTION_ACCESS_VIOLATION) {
+    DWORD code = info->ExceptionRecord->ExceptionCode;
+    u64 exception_addr = reinterpret_cast<u64>(info->ExceptionRecord->ExceptionAddress);
+
+    if (code != EXCEPTION_ACCESS_VIOLATION) {
+        // Not our problem
         return EXCEPTION_CONTINUE_SEARCH;
     }
 
-    const u64 fault_addr = info->ExceptionRecord->ExceptionInformation[1];
-    const u64 access_type = info->ExceptionRecord->ExceptionInformation[0];
-    const bool is_write = (access_type == 1);
-    const u64 fault_page = fault_addr >> HostPageBits;
+    u64 addr = 0, addr2 = 0;
 
-    u64 addr = 0;
-    u64 addr2 = 0;
+    for (auto region: vector_regions) {
+        auto addr_shifted = exception_addr >> HostPageBits;
+        if (region.first <= addr_shifted && addr_shifted <= region.second) {
+            addr = addr_shifted;
+        }
 
-    {
-        std::lock_guard lock(GetVectorRegionsMutex());
-        for (const auto& region : GetVectorRegions()) {
-            if (fault_page >= region.start_page && fault_page < region.end_page) {
-                addr = fault_page;
-            }
-            const u64 page2 = (fault_addr + 0x3F) >> HostPageBits;
-            if (page2 != fault_page && page2 >= region.start_page && page2 < region.end_page) {
-                addr2 = page2;
-            }
-            if (addr != 0 || addr2 != 0) break;
+        // Page-boundary accesses
+        if (auto addr_ = (exception_addr + 0x40) >> HostPageBits; addr_ != addr_shifted && region.first <= addr_ && addr_ <= region.second) {
+            addr2 = addr_;
+        }
+
+        if (addr != 0 || addr2 != 0) {
+            break;
         }
     }
 
     if (addr == 0 && addr2 == 0) {
+        // Not our problem
         return EXCEPTION_CONTINUE_SEARCH;
     }
 
-    LOG_ERROR(HW_Memory, "Accessing an unallocated region of a SparseLargeVector at {:#x}; this shouldn't happen and is likely a Dynarmic error!", fault_addr);
+    LOG_ERROR(HW_Memory, "Accessing an unallocated region of a SparseLargeVector at {:#x}; this shouldn't happen and is likely a Dynarmic error!", exception_addr);
 
-    if (addr != 0 && !CommitVectorPage(addr << HostPageBits, is_write)) {
-        return EXCEPTION_CONTINUE_SEARCH;
+    // Commit this region
+    if (addr != 0) {
+        if (!CommitVectorPage(addr << HostPageBits, false)) {
+            return EXCEPTION_CONTINUE_SEARCH;
+        }
     }
-    if (addr2 != 0 && !CommitVectorPage(addr2 << HostPageBits, is_write)) {
-        return EXCEPTION_CONTINUE_SEARCH;
+    // Commit next region if needed
+    if (addr2 != 0) {
+        if (!CommitVectorPage(addr2 << HostPageBits, false)) {
+            return EXCEPTION_CONTINUE_SEARCH;
+        }
     }
 
     return EXCEPTION_CONTINUE_EXECUTION;
@@ -87,31 +74,23 @@ static LONG WINAPI FakePageFaultHandler(PEXCEPTION_POINTERS info) {
 
 bool CommitVectorPage(uintptr_t addr, bool write) noexcept {
     MEMORY_BASIC_INFORMATION info {};
-    const auto res = VirtualQuery(reinterpret_cast<void*>(addr), &info, sizeof(info));
-    const DWORD perm = write ? PAGE_READWRITE : PAGE_READONLY;
-
+    auto res = VirtualQuery(reinterpret_cast<void*>(addr), &info, sizeof(info));
     if (res == 0) {
         LOG_CRITICAL(HW_Memory, "Failed to query large buffer region at {:#x} with error {}, will try committing anyway", addr, GetLastError());
-    } else if (info.State == MEM_COMMIT) {
-        DWORD old_protect {};
-        if (!VirtualProtect(reinterpret_cast<void*>(addr), HostPageSize, perm, &old_protect)) {
-            LOG_ERROR(HW_Memory, "VirtualProtect failed at {:#x}, error {}", addr, GetLastError());
-            return false;
-        }
-        return true;
     } else if (info.State != MEM_RESERVE) {
-        LOG_ERROR(HW_Memory, "Tried to commit an unreserved large buffer region at {:#x} (state {:#x})", addr, info.State);
+        LOG_ERROR(HW_Memory, "Tried to commit an unreserved large buffer region at {:#x} that is not mapped or is already committed (state {:#x})", addr, info.State);
         return false;
     }
 
-    if (VirtualAlloc(reinterpret_cast<LPVOID>(addr), HostPageSize, MEM_COMMIT, perm) == nullptr) {
+    auto perm = write ? PAGE_READWRITE : PAGE_READONLY;
+    void* res2 = VirtualAlloc(reinterpret_cast<LPVOID>(addr), HostPageSize, MEM_COMMIT, perm);
+    if (res2 == nullptr) {
         LOG_ERROR(HW_Memory, "Failed to commit large buffer region at {:#x}, error {}", addr, GetLastError());
         return false;
     }
 
     return true;
 }
-
 #endif
 
 #ifndef MAP_NOCORE
@@ -123,103 +102,57 @@ bool CommitVectorPage(uintptr_t addr, bool write) noexcept {
 
 void DecommitVectorPage(uintptr_t base) noexcept {
 #if defined(_WIN32)
-    if (!VirtualFree(reinterpret_cast<LPVOID>(base), HostPageSize, MEM_DECOMMIT)) {
-        LOG_WARNING(HW_Memory, "VirtualFree(MEM_DECOMMIT) failed at {:#x}, error {}", base, GetLastError());
-    }
+    VirtualFree(reinterpret_cast<LPVOID>(base), HostPageSize, MEM_DECOMMIT);
 #elif defined(__linux__)
-    if (madvise(reinterpret_cast<void*>(base), HostPageSize, MADV_DONTNEED) != 0) {
-        LOG_WARNING(HW_Memory, "madvise(MADV_DONTNEED) failed at {:#x}: {}", base, std::strerror(errno));
-    }
+    // Linux's MADV_DONTNEED zeros out pages for us
+    madvise(reinterpret_cast<void*>(base), HostPageSize, MADV_DONTNEED);
 #else
-    if (madvise(reinterpret_cast<void*>(base), HostPageSize, MADV_FREE) != 0) {
-        LOG_WARNING(HW_Memory, "madvise(MADV_FREE) failed at {:#x}: {}", base, std::strerror(errno));
-    }
+    madvise(reinterpret_cast<void*>(base), HostPageSize, MADV_FREE);
     std::memset(reinterpret_cast<void*>(base), 0, HostPageSize);
 #endif
 }
 
 void* AllocateMemoryPages(std::size_t size) noexcept {
-    if (size == 0) {
-        return nullptr;
-    }
-
-    const auto page = HostPageSize;
-    if (size % page != 0) {
+    if (auto page = HostPageSize; size % page != 0) {
         LOG_WARNING(HW_Memory, "Allocating unaligned large vector with size {:#x}; aligning to {} page size", size, page);
-        if (size > SIZE_MAX - (page - 1)) {
-            LOG_CRITICAL(HW_Memory, "Size {:#x} would overflow page alignment", size);
-            return nullptr;
-        }
         size = AlignUp(size, page);
     }
 
 #ifdef _WIN32
+    // We will never use this memory entirely so instead of committing it up front let's just reserve it and commit each page individually
     void* base = VirtualAlloc(nullptr, size, MEM_RESERVE, PAGE_READWRITE);
 
     if (base != nullptr) {
-        {
-            std::lock_guard lock(GetVectorRegionsMutex());
-            GetVectorRegions().push_back({
-                reinterpret_cast<u64>(base) >> HostPageBits,
-                (reinterpret_cast<u64>(base) + size) >> HostPageBits,
-            });
-        }
+        vector_regions.emplace_back(reinterpret_cast<u64>(base), reinterpret_cast<u64>(base) + size);
 
         static std::once_flag flag;
         std::call_once(flag, []() { AddVectoredExceptionHandler(1, FakePageFaultHandler); });
     } else {
+        // Try committing everything instead??
         LOG_WARNING(HW_Memory, "Failed to reserve large vector region with error {}, trying to commit instead..", GetLastError());
         base = VirtualAlloc(nullptr, size, MEM_COMMIT, PAGE_READWRITE);
     }
     ASSERT_MSG(base, "Failed to reserve {:#x} sized region with error {}", size, GetLastError());
 #else
-    int flags = MAP_ANON | MAP_PRIVATE;
-#ifdef MAP_NORESERVE
-    flags |= MAP_NORESERVE;
-#endif
-#if defined(MAP_NOCORE)
-    flags |= MAP_NOCORE;
-#endif
-    void* base = mmap(nullptr, size, PROT_READ, flags, -1, 0);
-    if (base == MAP_FAILED) {
+    void* base = mmap(nullptr, size, PROT_READ, MAP_ANON | MAP_PRIVATE | MAP_NOCORE, -1, 0);
+    if (base == MAP_FAILED)
         base = nullptr;
-    }
-#ifdef MADV_HUGEPAGE
-    if (base != nullptr) {
-        madvise(base, size, MADV_HUGEPAGE);
-    }
+    ASSERT_MSG(base, "Failed to allocate {:#x} sized region with error {}", size, strerror(errno));
 #endif
-    ASSERT_MSG(base, "Failed to allocate {:#x} sized region with error {}", size, std::strerror(errno));
-#endif
-
     return base;
 }
 
 void FreeMemoryPages(void* base, [[maybe_unused]] std::size_t size) noexcept {
-    if (base == nullptr) {
-        return;
-    }
-
-    if (const auto page = HostPageSize; size % page != 0) {
+    if (auto page = HostPageSize; size % page != 0) {
         size = AlignUp(size, page);
     }
-
+    if (!base)
+        return;
 #ifdef _WIN32
-    {
-        std::lock_guard lock(GetVectorRegionsMutex());
-        auto& regions = GetVectorRegions();
-        const u64 base_page = reinterpret_cast<u64>(base) >> HostPageBits;
-        regions.erase(std::remove_if(regions.begin(), regions.end(),
-            [base_page](const VectorRegion& r) { return r.start_page == base_page; }), regions.end());
-    }
-    if (!VirtualFree(base, 0, MEM_RELEASE)) {
-        LOG_ERROR(HW_Memory, "VirtualFree failed, error {}", GetLastError());
-    }
+    ASSERT(VirtualFree(base, 0, MEM_RELEASE));
 #else
-    if (munmap(base, size) != 0) {
-        LOG_ERROR(HW_Memory, "munmap failed: {}", std::strerror(errno));
-    }
+    ASSERT(munmap(base, size) == 0);
 #endif
 }
 
-} // namespace Common
+} // namespace Common

src/common/sparse_large_vector.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* virtual_buffer.h */
@@ -7,14 +7,10 @@
 
 #pragma once
 
-#include <array>
 #include <atomic>
 #include <bit>
-#include <cerrno>
-#include <cstdlib>
-#include <cstring>
-#include <memory>
-#include <type_traits>
+#include <utility>
+#include <vector>
 
 #ifndef _WIN32
 #include <unistd.h>
@@ -32,9 +28,9 @@ constexpr u64 HostPageBits = 12;
 constexpr u64 HostPageMask = ~(HostPageSize - 1);
 bool CommitVectorPage(uintptr_t addr, bool write) noexcept;
 #else
-inline const u64 HostPageSize = static_cast<u64>(sysconf(_SC_PAGESIZE));
-inline const u64 HostPageBits = std::countr_zero(HostPageSize);
-inline const u64 HostPageMask = ~(HostPageSize - 1);
+const u64 HostPageSize = sysconf(_SC_PAGESIZE);
+const u64 HostPageBits = std::countr_zero(HostPageSize);
+const u64 HostPageMask = ~(HostPageSize - 1);
 #endif
 
 void* AllocateMemoryPages(std::size_t size) noexcept;
@@ -47,18 +43,20 @@ template <typename T>
     // requires std::is_trivially_copyable_v<T>
 class SparseLargeVector final {
 public:
-    SparseLargeVector() = default;
+    constexpr SparseLargeVector() = default;
 
-    explicit SparseLargeVector(std::size_t count) noexcept {
-        if (count > SIZE_MAX / sizeof(T)) {
-            LOG_CRITICAL(Common_Memory, "SparseLargeVector size overflow: {} elements", count);
-            return;
-        }
-        Allocate(count * sizeof(T));
+    explicit SparseLargeVector(std::size_t count) noexcept
+        : alloc_size{count * sizeof(T)}
+    {
+        base_ptr = static_cast<T*>(AllocateMemoryPages(alloc_size));
+
+        // each item in vector holds information for 64 pages
+        auto denom = HostPageSize * 64;
+        committed_pages = std::vector<std::atomic<u64>>((alloc_size + denom - 1) / denom);
     }
 
     ~SparseLargeVector() noexcept {
-        Release();
+        FreeMemoryPages(base_ptr, alloc_size);
     }
 
     SparseLargeVector(const SparseLargeVector&) = delete;
@@ -67,181 +65,145 @@ public:
     SparseLargeVector& operator=(SparseLargeVector&& other) = delete;
 
     void ResizeAndClear(std::size_t count) noexcept {
-        if (count > SIZE_MAX / sizeof(T)) {
-            LOG_CRITICAL(Common_Memory, "SparseLargeVector resize overflow: {} elements", count);
-            return;
+        if (auto const new_size = count * sizeof(T); new_size != alloc_size) {
+            FreeMemoryPages(base_ptr, alloc_size);
+            alloc_size = new_size;
+            base_ptr = static_cast<T*>(AllocateMemoryPages(alloc_size));
+
+            auto denom = HostPageSize * 64;
+            committed_pages = std::vector<std::atomic<u64>>((alloc_size + denom - 1) / denom);
         }
-        const std::size_t new_size = count * sizeof(T);
-        if (new_size == alloc_size) {
-            ZeroRegion(0, alloc_size / sizeof(T));
-            return;
-        }
-        Release();
-        Allocate(new_size);
     }
 
+    /// Returns a reference to the value of the requested index and allocates memory if needed.
     T& GetAndFault(std::size_t index) noexcept {
-        if (base_ptr == nullptr || index >= size()) [[unlikely]] {
-            LOG_CRITICAL(Common_Memory, "SparseLargeVector RW access out of bounds @ {} (size {})", index, size());
-            std::abort();
+        if (index > alloc_size / sizeof(T)) {
+            UNREACHABLE_MSG("Out of bounds RW access on SparseLargeVector @ {}", index);
         }
-        const u64 byte_offset = static_cast<u64>(index) * sizeof(T);
-        if (!CommitPage(byte_offset)) [[unlikely]] {
-            LOG_CRITICAL(Common_Memory, "SparseLargeVector commit failed @ {} (offset {:#x})", index, byte_offset);
-            std::abort();
+
+        if (!IsCommittedPage(index)) {
+            CommitPage(index);
         }
         return base_ptr[index];
     }
 
-    const T& GetOrDefault(std::size_t index) const noexcept {
-        if (base_ptr == nullptr || index >= size()) [[unlikely]] {
-            LOG_CRITICAL(Common_Memory, "SparseLargeVector RO access out of bounds @ {}", index);
-            return DefaultValue();
-        }
+    /// Returns a reference to the value of the requested index if initialized, or will otherwise return a zero-initialized object.
+    const T& GetOrDefault(std::size_t index) const {
 #ifdef _WIN32
-        if (!IsPageCommitted(static_cast<u64>(index) * sizeof(T))) {
-            return DefaultValue();
+        if (!IsCommittedPage(index)) {
+            return *reinterpret_cast<const T*>(&default_val);
         }
 #endif
+        // On non-Windows, OS page table should optimize this by pointing to a zero page if unallocated.
         return base_ptr[index];
     }
 
     void Set(std::size_t index, const T& value) noexcept {
-        if (base_ptr == nullptr || index >= size()) [[unlikely]] {
-            LOG_CRITICAL(Common_Memory, "SparseLargeVector write out of bounds @ {}", index);
-            return;
-        }
-        const u64 byte_offset = static_cast<u64>(index) * sizeof(T);
-        if (!CommitPage(byte_offset)) [[unlikely]] {
-            LOG_CRITICAL(Common_Memory, "SparseLargeVector commit failed for write @ {}", index);
+        if (index > alloc_size / sizeof(T)) {
+            LOG_CRITICAL(Common_Memory, "Out of bounds write on SparseLargeVector @ {}", index);
             return;
         }
+        if (!IsCommittedPage(index))
+            CommitPage(index);
         base_ptr[index] = value;
     }
 
     void ZeroRegion(std::size_t start, std::size_t end_) noexcept {
-        if (base_ptr == nullptr || start >= end_) return;
+        u64 base = reinterpret_cast<u64>(&base_ptr[start]);
+        const u64 end = reinterpret_cast<u64>(&base_ptr[end_]);
 
-        const u64 start_off = static_cast<u64>(start) * sizeof(T);
-        const u64 end_off = static_cast<u64>(end_) * sizeof(T);
-        const u64 first_page_end = (start_off + HostPageSize - 1) & HostPageMask;
+        const u64 end_page = AlignUp(base, HostPageSize);
+        const u64 first_size = (std::min)(end_page, end) - base;
 
-        if (start_off < first_page_end) {
-            const u64 chunk_end = (std::min)(first_page_end, end_off);
-            const u64 chunk_size = chunk_end - start_off;
-            if (chunk_size != 0 && IsPageCommitted(start_off)) {
-                std::memset(reinterpret_cast<void*>(reinterpret_cast<uintptr_t>(base_ptr) + start_off), 0, chunk_size);
-            }
-            if (end_off <= first_page_end) return;
+        if (IsCommittedPage(start)) {
+            std::memset(reinterpret_cast<void*>(base), 0, first_size);
         }
 
-        for (u64 off = first_page_end; off < end_off; off += HostPageSize) {
-            if (!IsPageCommitted(off)) continue;
-            const u64 remaining = end_off - off;
-            if (remaining >= HostPageSize) {
-                DecommitPage(off);
+        if (end <= end_page)
+            return;
+
+        base = end_page;
+
+        for (u64 page = base; page < end; page += HostPageSize) {
+            auto index = (page - reinterpret_cast<u64>(base_ptr)) / sizeof(T);
+            if (!IsCommittedPage(index)) {
+                continue;
+            }
+
+            if (end - page >= HostPageSize) {
+                DecommitPage(index);
             } else {
-                std::memset(reinterpret_cast<void*>(reinterpret_cast<uintptr_t>(base_ptr) + off), 0, remaining);
+                std::memset(reinterpret_cast<void*>(page), 0, end - page);
             }
         }
     }
 
-    void CommitRegion(std::size_t index, std::size_t end_) noexcept {
-        if (base_ptr == nullptr || index >= end_) return;
-        const u64 start_off = static_cast<u64>(index) * sizeof(T);
-        const u64 end_off = static_cast<u64>(end_) * sizeof(T);
-        const u64 start_page = start_off & HostPageMask;
-        for (u64 off = start_page; off < end_off; off += HostPageSize) {
-            if (!IsPageCommitted(off)) {
-                (void)CommitPage(off);
+    constexpr void CommitRegion(size_t index, size_t end_) {
+        const u64 base = static_cast<u64>(index) * sizeof(T);
+        const u64 end = static_cast<u64>(end_) * sizeof(T);
+
+        for (u64 page = AlignDown(base, HostPageSize); page < end; page += HostPageSize) {
+            if (!IsCommittedPage(page / sizeof(T))) {
+                CommitPage(page / sizeof(T));
             }
         }
     }
 
-    T& GetUnchecked(std::size_t index) noexcept { return base_ptr[index]; }
+    constexpr T& GetUnchecked(size_t index) {
+        return base_ptr[index];
+    }
 
-    [[nodiscard]] const T& operator[](std::size_t index) const noexcept { return GetOrDefault(index); }
-    [[nodiscard]] const T* data() const noexcept { return base_ptr; }
-    [[nodiscard]] std::size_t size() const noexcept { return alloc_size / sizeof(T); }
+    [[nodiscard]] constexpr const T& operator[](std::size_t index) const noexcept {
+        return GetOrDefault(index);
+    }
+
+    [[nodiscard]] constexpr const T* data() const noexcept {
+        return base_ptr;
+    }
+
+    [[nodiscard]] constexpr std::size_t size() const noexcept {
+        return alloc_size / sizeof(T);
+    }
 
 private:
-    void Allocate(std::size_t new_size) noexcept {
-        alloc_size = new_size;
-        if (alloc_size == 0) {
-            base_ptr = nullptr;
-            committed_pages.reset();
-            return;
-        }
-        base_ptr = static_cast<T*>(AllocateMemoryPages(alloc_size));
-        const std::size_t num_pages = NumPages();
-        const std::size_t num_words = (num_pages + 63) / 64;
-        committed_pages = std::make_unique<std::atomic<u64>[]>(num_words);
-    }
-
-    void Release() noexcept {
-        if (base_ptr != nullptr) {
-            FreeMemoryPages(base_ptr, alloc_size);
-            base_ptr = nullptr;
-        }
-        committed_pages.reset();
-        alloc_size = 0;
-    }
-
-    [[nodiscard]] u64 NumPages() const noexcept {
-        return (alloc_size + HostPageSize - 1) >> HostPageBits;
-    }
-
-    [[nodiscard]] bool IsPageCommitted(u64 byte_offset) const noexcept {
-        const u64 page_index = byte_offset >> HostPageBits;
-        if (committed_pages == nullptr || page_index >= NumPages()) return false;
-        const auto val = committed_pages[page_index >> 6].load(std::memory_order_acquire);
-        return (val >> (page_index & 63)) & 1;
-    }
-
-    void SetPageBit(u64 page_index, bool value) noexcept {
-        if (committed_pages == nullptr) return;
-        const u64 bit = 1ULL << (page_index & 63);
-        auto& atom = committed_pages[page_index >> 6];
-        if (value) {
-            atom.fetch_or(bit, std::memory_order_release);
-        } else {
-            atom.fetch_and(~bit, std::memory_order_release);
-        }
-    }
-
-    bool CommitPage(u64 byte_offset) noexcept {
-        const u64 page_index = byte_offset >> HostPageBits;
-        const uintptr_t page_addr = (reinterpret_cast<uintptr_t>(base_ptr) + byte_offset) & HostPageMask;
-
-        if (IsPageCommitted(byte_offset)) return true;
-
-#if defined(_WIN32)
-        if (!CommitVectorPage(page_addr, true)) return false;
-#else
-        if (mprotect(reinterpret_cast<void*>(page_addr), HostPageSize, PROT_READ | PROT_WRITE) != 0) {
-            LOG_ERROR(Common_Memory, "mprotect failed at {:#x}: {}", page_addr, std::strerror(errno));
+    [[nodiscard]] constexpr bool IsCommittedPage(std::size_t index) const noexcept {
+        if (index > alloc_size / sizeof(T)) {
+            LOG_CRITICAL(Common_Memory, "Out of bounds access on large vector @ {}", index);
             return false;
         }
+
+        auto page = (index * sizeof(T)) >> HostPageBits;
+        auto val = committed_pages[page >> 6].load(std::memory_order_acquire);
+        return (val >> (page & 63)) & 1;
+    }
+
+    constexpr void CommitPage(std::size_t index) noexcept {
+        auto page_index = (index * sizeof(T)) >> HostPageBits;
+        auto page = reinterpret_cast<uintptr_t>(base_ptr + index) & HostPageMask;
+#if defined(_WIN32)
+        CommitVectorPage(page, true);
+#else
+        mprotect(reinterpret_cast<void*>(page), HostPageSize, PROT_READ | PROT_WRITE);
 #endif
-        SetPageBit(page_index, true);
-        return true;
+
+        committed_pages[page_index >> 6].fetch_or(1ULL << (page_index & 63), std::memory_order_release);
     }
 
-    void DecommitPage(u64 byte_offset) noexcept {
-        const u64 page_index = byte_offset >> HostPageBits;
-        const uintptr_t page_addr = (reinterpret_cast<uintptr_t>(base_ptr) + byte_offset) & HostPageMask;
-        DecommitVectorPage(page_addr);
-        SetPageBit(page_index, false);
-    }
+    constexpr void DecommitPage(std::size_t index) noexcept {
+        auto page_index = (index * sizeof(T)) >> HostPageBits;
+        auto page = reinterpret_cast<uintptr_t>(base_ptr + index) & HostPageMask;
 
-    [[nodiscard]] const T& DefaultValue() const noexcept {
-        return *reinterpret_cast<const T*>(&default_val);
+        committed_pages[page_index >> 6].fetch_and(~(1ULL << (page_index & 63)), std::memory_order_release);
+        DecommitVectorPage(page);
     }
 
     std::size_t alloc_size{};
     T* base_ptr{};
-    std::unique_ptr<std::atomic<u64>[]> committed_pages{};
-    alignas(T) const std::array<u8, sizeof(T)> default_val{};
+
+    std::vector<std::atomic<u64>> committed_pages{};
+#ifdef _WIN32
+    const std::array<u8, sizeof(T)> default_val{};
+#endif
 };
 
-} // namespace Common
+} // namespace Common

src/common/zbic_compression.cpp

@@ -0,0 +1,43 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#include <cstring>
+
+#include "common/zbic_compression.h"
+#include "common/zstd.h"
+
+namespace Common::Compression {
+
+bool IsZBIC(const void* src, size_t src_size) {
+    if (!src || src_size < 4) {
+        return false;
+    }
+    u32 magic = 0;
+    std::memcpy(&magic, src, sizeof(u32));
+    return magic == ZSTD_MAGICNUMBER; // 0x4349425A ("ZBIC")
+}
+
+int DecompressDataZBIC(void* dst, size_t dst_size, const void* src, size_t src_size) {
+    if (!dst || !src || dst_size == 0 || src_size == 0) {
+        return -1;
+    }
+    const size_t res = ZSTD_decompress(dst, dst_size, src, src_size);
+    if (ZSTD_isError(res)) {
+        return -1;
+    }
+    return static_cast<int>(res);
+}
+
+std::vector<u8> DecompressDataZBIC(std::span<const u8> compressed, std::size_t uncompressed_size) {
+    std::vector<u8> uncompressed(uncompressed_size);
+    const int r = DecompressDataZBIC(uncompressed.data(), uncompressed_size, compressed.data(), compressed.size());
+    if (r <= 0) {
+        return {};
+    }
+    if (static_cast<size_t>(r) < uncompressed_size) {
+        uncompressed.resize(static_cast<size_t>(r));
+    }
+    return uncompressed;
+}
+
+} // namespace Common::Compression

src/common/zbic_compression.h

@@ -0,0 +1,18 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include <span>
+#include <vector>
+#include "common/common_types.h"
+
+namespace Common::Compression {
+
+[[nodiscard]] bool IsZBIC(const void* src, size_t src_size);
+
+[[nodiscard]] std::vector<u8> DecompressDataZBIC(std::span<const u8> compressed, std::size_t uncompressed_size);
+
+[[nodiscard]] int DecompressDataZBIC(void* dst, size_t dst_size, const void* src, size_t src_size);
+
+} // namespace Common::Compression

src/common/zstd_errors.h

@@ -0,0 +1,107 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_ERRORS_H_398273423
+#define ZSTD_ERRORS_H_398273423
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* =====   ZSTDERRORLIB_API : control library symbols visibility   ===== */
+#ifndef ZSTDERRORLIB_VISIBLE
+   /* Backwards compatibility with old macro name */
+#  ifdef ZSTDERRORLIB_VISIBILITY
+#    define ZSTDERRORLIB_VISIBLE ZSTDERRORLIB_VISIBILITY
+#  elif defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZSTDERRORLIB_VISIBLE __attribute__ ((visibility ("default")))
+#  else
+#    define ZSTDERRORLIB_VISIBLE
+#  endif
+#endif
+
+#ifndef ZSTDERRORLIB_HIDDEN
+#  if defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZSTDERRORLIB_HIDDEN __attribute__ ((visibility ("hidden")))
+#  else
+#    define ZSTDERRORLIB_HIDDEN
+#  endif
+#endif
+
+#if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
+#  define ZSTDERRORLIB_API __declspec(dllexport) ZSTDERRORLIB_VISIBLE
+#elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
+#  define ZSTDERRORLIB_API __declspec(dllimport) ZSTDERRORLIB_VISIBLE /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define ZSTDERRORLIB_API ZSTDERRORLIB_VISIBLE
+#endif
+
+/*-*********************************************
+ *  Error codes list
+ *-*********************************************
+ *  Error codes _values_ are pinned down since v1.3.1 only.
+ *  Therefore, don't rely on values if you may link to any version < v1.3.1.
+ *
+ *  Only values < 100 are considered stable.
+ *
+ *  note 1 : this API shall be used with static linking only.
+ *           dynamic linking is not yet officially supported.
+ *  note 2 : Prefer relying on the enum than on its value whenever possible
+ *           This is the only supported way to use the error list < v1.3.1
+ *  note 3 : ZSTD_isError() is always correct, whatever the library version.
+ **********************************************/
+typedef enum {
+  ZSTD_error_no_error = 0,
+  ZSTD_error_GENERIC  = 1,
+  ZSTD_error_prefix_unknown                = 10,
+  ZSTD_error_version_unsupported           = 12,
+  ZSTD_error_frameParameter_unsupported    = 14,
+  ZSTD_error_frameParameter_windowTooLarge = 16,
+  ZSTD_error_corruption_detected = 20,
+  ZSTD_error_checksum_wrong      = 22,
+  ZSTD_error_literals_headerWrong = 24,
+  ZSTD_error_dictionary_corrupted      = 30,
+  ZSTD_error_dictionary_wrong          = 32,
+  ZSTD_error_dictionaryCreation_failed = 34,
+  ZSTD_error_parameter_unsupported   = 40,
+  ZSTD_error_parameter_combination_unsupported = 41,
+  ZSTD_error_parameter_outOfBound    = 42,
+  ZSTD_error_tableLog_tooLarge       = 44,
+  ZSTD_error_maxSymbolValue_tooLarge = 46,
+  ZSTD_error_maxSymbolValue_tooSmall = 48,
+  ZSTD_error_cannotProduce_uncompressedBlock = 49,
+  ZSTD_error_stabilityCondition_notRespected = 50,
+  ZSTD_error_stage_wrong       = 60,
+  ZSTD_error_init_missing      = 62,
+  ZSTD_error_memory_allocation = 64,
+  ZSTD_error_workSpace_tooSmall= 66,
+  ZSTD_error_dstSize_tooSmall = 70,
+  ZSTD_error_srcSize_wrong    = 72,
+  ZSTD_error_dstBuffer_null   = 74,
+  ZSTD_error_noForwardProgress_destFull = 80,
+  ZSTD_error_noForwardProgress_inputEmpty = 82,
+  /* following error codes are __NOT STABLE__, they can be removed or changed in future versions */
+  ZSTD_error_frameIndex_tooLarge = 100,
+  ZSTD_error_seekableIO          = 102,
+  ZSTD_error_dstBuffer_wrong     = 104,
+  ZSTD_error_srcBuffer_wrong     = 105,
+  ZSTD_error_sequenceProducer_failed = 106,
+  ZSTD_error_externalSequences_invalid = 107,
+  ZSTD_error_maxCode = 120  /* never EVER use this value directly, it can change in future versions! Use ZSTD_isError() instead */
+} ZSTD_ErrorCode;
+
+ZSTDERRORLIB_API const char* ZSTD_getErrorString(ZSTD_ErrorCode code);   /**< Same as ZSTD_getErrorName, but using a `ZSTD_ErrorCode` enum argument */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_ERRORS_H_398273423 */

src/core/loader/nso.cpp

@@ -13,6 +13,7 @@
 #include "common/hex_util.h"
 #include "common/logging.h"
 #include "common/lz4_compression.h"
+#include "common/zbic_compression.h"
 #include "common/settings.h"
 #include "common/swap.h"
 #include "core/core.h"
@@ -104,11 +105,38 @@ std::optional<VAddr> AppLoader_NSO::LoadModule(Kernel::KProcess& process, Core::
         for (std::size_t i = 0; i < nso_header.segments.size(); ++i) {
             nso_file.Read(compressed_data.data(), nso_header.segments_compressed_size[i], nso_header.segments[i].offset);
             if (nso_header.IsSegmentCompressed(i)) {
-                int r = Common::Compression::DecompressDataLZ4(decompressed_size.data(), nso_header.segments[i].size, compressed_data.data(), nso_header.segments_compressed_size[i]);
-                ASSERT(r == int(nso_header.segments[i].size));
-                std::memcpy(codeset.memory.data() + module_start + nso_header.segments[i].location, decompressed_size.data(), nso_header.segments[i].size);
+                if (nso_header.IsZBICCompressed()) {
+                    // ZBIC compression
+                    const int r = Common::Compression::DecompressDataZBIC(
+                        decompressed_size.data(),
+                        nso_header.segments[i].size,
+                        compressed_data.data(),
+                        nso_header.segments_compressed_size[i]
+                    );
+                    ASSERT(r > 0);
+                } else {
+                    // LZ4 compression
+                    int r = Common::Compression::DecompressDataLZ4(
+                        decompressed_size.data(),
+                        nso_header.segments[i].size,
+                        compressed_data.data(),
+                        nso_header.segments_compressed_size[i]
+                    );
+                    ASSERT(r == int(nso_header.segments[i].size));
+                }
+
+                std::memcpy(
+                    codeset.memory.data() + module_start + nso_header.segments[i].location,
+                    decompressed_size.data(),
+                    nso_header.segments[i].size
+                );
             } else {
-                std::memcpy(codeset.memory.data() + module_start + nso_header.segments[i].location, compressed_data.data(), nso_header.segments[i].size);
+                // Not compressed
+                std::memcpy(
+                    codeset.memory.data() + module_start + nso_header.segments[i].location,
+                    compressed_data.data(),
+                    nso_header.segments[i].size
+                );
             }
             codeset.segments[i].addr = module_start + nso_header.segments[i].location;
             codeset.segments[i].offset = module_start + nso_header.segments[i].location;

src/core/loader/nso.h

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 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
 
@@ -58,6 +61,9 @@ struct NSOHeader {
     std::array<SHA256Hash, 3> segment_hashes;
 
     bool IsSegmentCompressed(size_t segment_num) const;
+    bool IsZBICCompressed() const {
+        return ((flags >> 7) & 1) != 0;
+    }
 };
 static_assert(sizeof(NSOHeader) == 0x100, "NSOHeader has incorrect size.");
 static_assert(std::is_trivially_copyable_v<NSOHeader>, "NSOHeader must be trivially copyable.");