Merge pull request #12869 from FernandoS27/smmu-fixes

SMMU: A set of different fixes.

3 files changed, 45 insertions, 40 deletions

diff --git a/src/core/device_memory_manager.h b/src/core/device_memory_manager.h
index ffeed46cc..0568a821b 100644
--- a/src/core/device_memory_manager.h
+++ b/src/core/device_memory_manager.h
@@ -5,11 +5,13 @@
 
 #include <array>
 #include <atomic>
+#include <bit>
 #include <deque>
 #include <memory>
 #include <mutex>
 
 #include "common/common_types.h"
+#include "common/range_mutex.h"
 #include "common/scratch_buffer.h"
 #include "common/virtual_buffer.h"
 
@@ -180,31 +182,35 @@ private:
     }
 
     Common::VirtualBuffer<VAddr> cpu_backing_address;
-    static constexpr size_t subentries = 8 / sizeof(u8);
+    using CounterType = u8;
+    using CounterAtomicType = std::atomic_uint8_t;
+    static constexpr size_t subentries = 8 / sizeof(CounterType);
     static constexpr size_t subentries_mask = subentries - 1;
+    static constexpr size_t subentries_shift =
+        std::countr_zero(sizeof(u64)) - std::countr_zero(sizeof(CounterType));
     class CounterEntry final {
     public:
         CounterEntry() = default;
 
-        std::atomic_uint8_t& Count(std::size_t page) {
+        CounterAtomicType& Count(std::size_t page) {
             return values[page & subentries_mask];
         }
 
-        const std::atomic_uint8_t& Count(std::size_t page) const {
+        const CounterAtomicType& Count(std::size_t page) const {
             return values[page & subentries_mask];
         }
 
     private:
-        std::array<std::atomic_uint8_t, subentries> values{};
+        std::array<CounterAtomicType, subentries> values{};
     };
-    static_assert(sizeof(CounterEntry) == subentries * sizeof(u8),
+    static_assert(sizeof(CounterEntry) == subentries * sizeof(CounterType),
                   "CounterEntry should be 8 bytes!");
 
     static constexpr size_t num_counter_entries =
         (1ULL << (device_virtual_bits - page_bits)) / subentries;
     using CachedPages = std::array<CounterEntry, num_counter_entries>;
     std::unique_ptr<CachedPages> cached_pages;
-    std::mutex counter_guard;
+    Common::RangeMutex counter_guard;
     std::mutex mapping_guard;
 };
 
diff --git a/src/core/device_memory_manager.inc b/src/core/device_memory_manager.inc
index eab8a2731..b026f4220 100644
--- a/src/core/device_memory_manager.inc
+++ b/src/core/device_memory_manager.inc
@@ -213,8 +213,8 @@ void DeviceMemoryManager<Traits>::Free(DAddr start, size_t size) {
 }
 
 template <typename Traits>
-void DeviceMemoryManager<Traits>::Map(DAddr address, VAddr virtual_address, size_t size,
-                                      Asid asid, bool track) {
+void DeviceMemoryManager<Traits>::Map(DAddr address, VAddr virtual_address, size_t size, Asid asid,
+                                      bool track) {
     Core::Memory::Memory* process_memory = registered_processes[asid.id];
     size_t start_page_d = address >> Memory::YUZU_PAGEBITS;
     size_t num_pages = Common::AlignUp(size, Memory::YUZU_PAGESIZE) >> Memory::YUZU_PAGEBITS;
@@ -508,12 +508,7 @@ void DeviceMemoryManager<Traits>::UnregisterProcess(Asid asid) {
 
 template <typename Traits>
 void DeviceMemoryManager<Traits>::UpdatePagesCachedCount(DAddr addr, size_t size, s32 delta) {
-    std::unique_lock<std::mutex> lk(counter_guard, std::defer_lock);
-    const auto Lock = [&] {
-        if (!lk) {
-            lk.lock();
-        }
-    };
+    Common::ScopedRangeLock lk(counter_guard, addr, size);
     u64 uncache_begin = 0;
     u64 cache_begin = 0;
     u64 uncache_bytes = 0;
@@ -524,22 +519,36 @@ void DeviceMemoryManager<Traits>::UpdatePagesCachedCount(DAddr addr, size_t size
     const size_t page_end = Common::DivCeil(addr + size, Memory::YUZU_PAGESIZE);
     size_t page = addr >> Memory::YUZU_PAGEBITS;
     auto [asid, base_vaddress] = ExtractCPUBacking(page);
-    size_t vpage = base_vaddress >> Memory::YUZU_PAGEBITS;
     auto* memory_device_inter = registered_processes[asid.id];
+    const auto release_pending = [&] {
+        if (uncache_bytes > 0) {
+            MarkRegionCaching(memory_device_inter, uncache_begin << Memory::YUZU_PAGEBITS,
+                              uncache_bytes, false);
+            uncache_bytes = 0;
+        }
+        if (cache_bytes > 0) {
+            MarkRegionCaching(memory_device_inter, cache_begin << Memory::YUZU_PAGEBITS,
+                              cache_bytes, true);
+            cache_bytes = 0;
+        }
+    };
     for (; page != page_end; ++page) {
-        std::atomic_uint8_t& count = cached_pages->at(page >> 3).Count(page);
+        CounterAtomicType& count = cached_pages->at(page >> subentries_shift).Count(page);
+        auto [asid_2, vpage] = ExtractCPUBacking(page);
+        vpage >>= Memory::YUZU_PAGEBITS;
 
-        if (delta > 0) {
-            ASSERT_MSG(count.load(std::memory_order::relaxed) < std::numeric_limits<u8>::max(),
-                       "Count may overflow!");
-        } else if (delta < 0) {
-            ASSERT_MSG(count.load(std::memory_order::relaxed) > 0, "Count may underflow!");
-        } else {
-            ASSERT_MSG(false, "Delta must be non-zero!");
+        if (vpage == 0) [[unlikely]] {
+            release_pending();
+            continue;
+        }
+
+        if (asid.id != asid_2.id) [[unlikely]] {
+            release_pending();
+            memory_device_inter = registered_processes[asid_2.id];
         }
 
         // Adds or subtracts 1, as count is a unsigned 8-bit value
-        count.fetch_add(static_cast<u8>(delta), std::memory_order_release);
+        count.fetch_add(static_cast<CounterType>(delta), std::memory_order_release);
 
         // Assume delta is either -1 or 1
         if (count.load(std::memory_order::relaxed) == 0) {
@@ -548,7 +557,6 @@ void DeviceMemoryManager<Traits>::UpdatePagesCachedCount(DAddr addr, size_t size
             }
             uncache_bytes += Memory::YUZU_PAGESIZE;
         } else if (uncache_bytes > 0) {
-            Lock();
             MarkRegionCaching(memory_device_inter, uncache_begin << Memory::YUZU_PAGEBITS,
                               uncache_bytes, false);
             uncache_bytes = 0;
@@ -559,23 +567,12 @@ void DeviceMemoryManager<Traits>::UpdatePagesCachedCount(DAddr addr, size_t size
             }
             cache_bytes += Memory::YUZU_PAGESIZE;
         } else if (cache_bytes > 0) {
-            Lock();
-            MarkRegionCaching(memory_device_inter, cache_begin << Memory::YUZU_PAGEBITS, cache_bytes,
-                              true);
+            MarkRegionCaching(memory_device_inter, cache_begin << Memory::YUZU_PAGEBITS,
+                              cache_bytes, true);
             cache_bytes = 0;
         }
-        vpage++;
-    }
-    if (uncache_bytes > 0) {
-        Lock();
-        MarkRegionCaching(memory_device_inter, uncache_begin << Memory::YUZU_PAGEBITS, uncache_bytes,
-                          false);
-    }
-    if (cache_bytes > 0) {
-        Lock();
-        MarkRegionCaching(memory_device_inter, cache_begin << Memory::YUZU_PAGEBITS, cache_bytes,
-                          true);
     }
+    release_pending();
 }
 
 } // namespace Core
diff --git a/src/core/hle/service/nvdrv/core/container.cpp b/src/core/hle/service/nvdrv/core/container.cpp
index dc1b4d5be..e89cca6f2 100644
--- a/src/core/hle/service/nvdrv/core/container.cpp
+++ b/src/core/hle/service/nvdrv/core/container.cpp
@@ -83,7 +83,9 @@ SessionId Container::OpenSession(Kernel::KProcess* process) {
 
             // Check if this memory block is heap.
             if (svc_mem_info.state == Kernel::Svc::MemoryState::Normal) {
-                if (svc_mem_info.size > region_size) {
+                if (region_start + region_size == svc_mem_info.base_address) {
+                    region_size += svc_mem_info.size;
+                } else if (svc_mem_info.size > region_size) {
                     region_size = svc_mem_info.size;
                     region_start = svc_mem_info.base_address;
                 }