Merge pull request #4228 from NarcolepticK/lle-mapped-buffer

LLE Mapped Buffer: Add unmapping of write buffer, zero-size, and multiple page handling

src/core/hle/kernel/ipc.cpp

@@ -14,6 +14,68 @@
 
 namespace Kernel {
 
+void ScanForAndUnmapBuffer(std::array<u32, IPC::COMMAND_BUFFER_LENGTH>& dst_cmd_buf,
+                           const std::size_t dst_command_size, std::size_t& target_index,
+                           SharedPtr<Process> src_process, SharedPtr<Process> dst_process,
+                           const VAddr source_address, const VAddr page_start, const u32 num_pages,
+                           const u32 size, const IPC::MappedBufferPermissions permissions) {
+    while (target_index < dst_command_size) {
+        u32 desc = dst_cmd_buf[target_index++];
+
+        if (IPC::GetDescriptorType(desc) == IPC::DescriptorType::CopyHandle ||
+            IPC::GetDescriptorType(desc) == IPC::DescriptorType::MoveHandle) {
+            u32 num_handles = IPC::HandleNumberFromDesc(desc);
+            for (u32 j = 0; j < num_handles; ++j) {
+                target_index += 1;
+            }
+            continue;
+        }
+
+        if (IPC::GetDescriptorType(desc) == IPC::DescriptorType::CallingPid ||
+            IPC::GetDescriptorType(desc) == IPC::DescriptorType::StaticBuffer) {
+            target_index += 1;
+            continue;
+        }
+
+        if (IPC::GetDescriptorType(desc) == IPC::DescriptorType::MappedBuffer) {
+            VAddr dest_address = dst_cmd_buf[target_index];
+            IPC::MappedBufferDescInfo dest_descInfo{desc};
+            u32 dest_size = static_cast<u32>(dest_descInfo.size);
+            IPC::MappedBufferPermissions dest_permissions = dest_descInfo.perms;
+
+            if (dest_size == 0) {
+                target_index += 1;
+                continue;
+            }
+
+            ASSERT(permissions == dest_permissions && size == dest_size);
+            // Readonly buffers do not need to be copied over to the target
+            // process again because they were (presumably) not modified. This
+            // behavior is consistent with the real kernel.
+            if (permissions != IPC::MappedBufferPermissions::R) {
+                // Copy the modified buffer back into the target process
+                Memory::CopyBlock(*src_process, *dst_process, source_address, dest_address, size);
+            }
+
+            VAddr prev_reserve = page_start - Memory::PAGE_SIZE;
+            VAddr next_reserve = page_start + num_pages * Memory::PAGE_SIZE;
+
+            auto& prev_vma = src_process->vm_manager.FindVMA(prev_reserve)->second;
+            auto& next_vma = src_process->vm_manager.FindVMA(next_reserve)->second;
+            ASSERT(prev_vma.meminfo_state == MemoryState::Reserved &&
+                   next_vma.meminfo_state == MemoryState::Reserved);
+
+            // Unmap the buffer and guard pages from the source process
+            ResultCode result = src_process->vm_manager.UnmapRange(
+                page_start - Memory::PAGE_SIZE, (num_pages + 2) * Memory::PAGE_SIZE);
+            ASSERT(result == RESULT_SUCCESS);
+
+            target_index += 1;
+            break;
+        }
+    }
+}
+
 ResultCode TranslateCommandBuffer(SharedPtr<Thread> src_thread, SharedPtr<Thread> dst_thread,
                                   VAddr src_address, VAddr dst_address, bool reply) {
 
@@ -33,6 +95,18 @@ ResultCode TranslateCommandBuffer(SharedPtr<Thread> src_thread, SharedPtr<Thread
     std::array<u32, IPC::COMMAND_BUFFER_LENGTH> cmd_buf;
     Memory::ReadBlock(*src_process, src_address, cmd_buf.data(), command_size * sizeof(u32));
 
+    // Create a copy of the target's command buffer
+    IPC::Header dst_header;
+    Memory::ReadBlock(*dst_process, dst_address, &dst_header.raw, sizeof(dst_header.raw));
+
+    std::size_t dst_untranslated_size = 1u + dst_header.normal_params_size;
+    std::size_t dst_command_size = dst_untranslated_size + dst_header.translate_params_size;
+    std::size_t target_index = dst_untranslated_size;
+
+    std::array<u32, IPC::COMMAND_BUFFER_LENGTH> dst_cmd_buf;
+    Memory::ReadBlock(*dst_process, dst_address, dst_cmd_buf.data(),
+                      dst_command_size * sizeof(u32));
+
     std::size_t i = untranslated_size;
     while (i < command_size) {
         u32 descriptor = cmd_buf[i];
@@ -128,76 +202,50 @@ ResultCode TranslateCommandBuffer(SharedPtr<Thread> src_thread, SharedPtr<Thread
             u32 num_pages =
                 Common::AlignUp(page_offset + size, Memory::PAGE_SIZE) >> Memory::PAGE_BITS;
 
+            // Skip when the size is zero and num_pages == 0
+            if (size == 0) {
+                cmd_buf[i++] = 0;
+                break;
+            }
             ASSERT(num_pages >= 1);
 
             if (reply) {
-                // TODO(Subv): Scan the target's command buffer to make sure that there was a
+                // Scan the target's command buffer for the matching mapped buffer.
-                // MappedBuffer descriptor in the original request. The real kernel panics if you
+                // The real kernel panics if you try to reply with an unsolicited MappedBuffer.
-                // try to reply with an unsolicited MappedBuffer.
+                ScanForAndUnmapBuffer(dst_cmd_buf, dst_command_size, target_index, src_process,
+                                      dst_process, source_address, page_start, num_pages, size,
+                                      permissions);
 
-                // Unmap the buffers. Readonly buffers do not need to be copied over to the target
-                // process again because they were (presumably) not modified. This behavior is
-                // consistent with the real kernel.
-                if (permissions == IPC::MappedBufferPermissions::R) {
-                    ResultCode result = src_process->vm_manager.UnmapRange(
-                        page_start, num_pages * Memory::PAGE_SIZE);
-                    ASSERT(result == RESULT_SUCCESS);
-                }
-
-                ASSERT_MSG(permissions == IPC::MappedBufferPermissions::R,
-                           "Unmapping Write MappedBuffers is unimplemented");
                 i += 1;
                 break;
             }
 
             VAddr target_address = 0;
 
-            auto IsPageAligned = [](VAddr address) -> bool {
-                return (address & Memory::PAGE_MASK) == 0;
-            };
-
-            // TODO(Subv): Support more than 1 page and aligned page mappings
-            ASSERT_MSG(
-                num_pages == 1 &&
-                    (!IsPageAligned(source_address) || !IsPageAligned(source_address + size)),
-                "MappedBuffers of more than one page or aligned transfers are not implemented");
-
             // TODO(Subv): Perform permission checks.
 
-            // TODO(Subv): Leave a page of Reserved memory before the first page and after the last
+            // Reserve a page of memory before the mapped buffer
-            // page.
+            auto reserve_buffer = std::make_shared<std::vector<u8>>(Memory::PAGE_SIZE);
+            dst_process->vm_manager.MapMemoryBlockToBase(
+                Memory::IPC_MAPPING_VADDR, Memory::IPC_MAPPING_SIZE, reserve_buffer, 0,
+                static_cast<u32>(reserve_buffer->size()), Kernel::MemoryState::Reserved);
 
-            if (!IsPageAligned(source_address) ||
+            auto buffer = std::make_shared<std::vector<u8>>(num_pages * Memory::PAGE_SIZE);
-                (num_pages == 1 && !IsPageAligned(source_address + size))) {
+            Memory::ReadBlock(*src_process, source_address, buffer->data() + page_offset, size);
-                // If the address of the source buffer is not page-aligned or if the buffer doesn't
-                // fill an entire page, then we have to allocate a page of memory in the target
-                // process and copy over the data from the input buffer. This allocated buffer will
-                // be copied back to the source process and deallocated when the server replies to
-                // the request via ReplyAndReceive.
 
-                auto buffer = std::make_shared<std::vector<u8>>(Memory::PAGE_SIZE);
+            // Map the page(s) into the target process' address space.
-
+            target_address = dst_process->vm_manager
-                // Number of bytes until the next page.
+                                 .MapMemoryBlockToBase(
-                std::size_t difference_to_page =
+                                     Memory::IPC_MAPPING_VADDR, Memory::IPC_MAPPING_SIZE, buffer, 0,
-                    Common::AlignUp(source_address, Memory::PAGE_SIZE) - source_address;
+                                     static_cast<u32>(buffer->size()), Kernel::MemoryState::Shared)
-                // If the data fits in one page we can just copy the required size instead of the
-                // entire page.
-                std::size_t read_size =
-                    num_pages == 1 ? static_cast<std::size_t>(size) : difference_to_page;
-
-                Memory::ReadBlock(*src_process, source_address, buffer->data() + page_offset,
-                                  read_size);
-
-                // Map the page into the target process' address space.
-                target_address =
-                    dst_process->vm_manager
-                        .MapMemoryBlockToBase(Memory::IPC_MAPPING_VADDR, Memory::IPC_MAPPING_SIZE,
-                                              buffer, 0, static_cast<u32>(buffer->size()),
-                                              Kernel::MemoryState::Shared)
                                  .Unwrap();
-            }
 
             cmd_buf[i++] = target_address + page_offset;
+
+            // Reserve a page of memory after the mapped buffer
+            dst_process->vm_manager.MapMemoryBlockToBase(
+                Memory::IPC_MAPPING_VADDR, Memory::IPC_MAPPING_SIZE, reserve_buffer, 0,
+                static_cast<u32>(reserve_buffer->size()), Kernel::MemoryState::Reserved);
             break;
         }
         default:

src/core/memory.cpp

@@ -705,6 +705,58 @@ void CopyBlock(VAddr dest_addr, VAddr src_addr, const std::size_t size) {
     CopyBlock(*Core::System::GetInstance().Kernel().GetCurrentProcess(), dest_addr, src_addr, size);
 }
 
+void CopyBlock(const Kernel::Process& src_process, const Kernel::Process& dest_process,
+               VAddr src_addr, VAddr dest_addr, std::size_t size) {
+    auto& page_table = src_process.vm_manager.page_table;
+    std::size_t remaining_size = size;
+    std::size_t page_index = src_addr >> PAGE_BITS;
+    std::size_t page_offset = src_addr & PAGE_MASK;
+
+    while (remaining_size > 0) {
+        const std::size_t copy_amount = std::min(PAGE_SIZE - page_offset, remaining_size);
+        const VAddr current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
+
+        switch (page_table.attributes[page_index]) {
+        case PageType::Unmapped: {
+            LOG_ERROR(HW_Memory,
+                      "unmapped CopyBlock @ 0x{:08X} (start address = 0x{:08X}, size = {})",
+                      current_vaddr, src_addr, size);
+            ZeroBlock(dest_process, dest_addr, copy_amount);
+            break;
+        }
+        case PageType::Memory: {
+            DEBUG_ASSERT(page_table.pointers[page_index]);
+            const u8* src_ptr = page_table.pointers[page_index] + page_offset;
+            WriteBlock(dest_process, dest_addr, src_ptr, copy_amount);
+            break;
+        }
+        case PageType::Special: {
+            MMIORegionPointer handler = GetMMIOHandler(page_table, current_vaddr);
+            DEBUG_ASSERT(handler);
+            std::vector<u8> buffer(copy_amount);
+            handler->ReadBlock(current_vaddr, buffer.data(), buffer.size());
+            WriteBlock(dest_process, dest_addr, buffer.data(), buffer.size());
+            break;
+        }
+        case PageType::RasterizerCachedMemory: {
+            RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
+                                         FlushMode::Flush);
+            WriteBlock(dest_process, dest_addr, GetPointerFromVMA(src_process, current_vaddr),
+                       copy_amount);
+            break;
+        }
+        default:
+            UNREACHABLE();
+        }
+
+        page_index++;
+        page_offset = 0;
+        dest_addr += static_cast<VAddr>(copy_amount);
+        src_addr += static_cast<VAddr>(copy_amount);
+        remaining_size -= copy_amount;
+    }
+}
+
 template <>
 u8 ReadMMIO<u8>(MMIORegionPointer mmio_handler, VAddr addr) {
     return mmio_handler->Read8(addr);

src/core/memory.h

@@ -205,6 +205,8 @@ void ZeroBlock(const Kernel::Process& process, VAddr dest_addr, const std::size_
 void ZeroBlock(VAddr dest_addr, const std::size_t size);
 void CopyBlock(const Kernel::Process& process, VAddr dest_addr, VAddr src_addr, std::size_t size);
 void CopyBlock(VAddr dest_addr, VAddr src_addr, std::size_t size);
+void CopyBlock(const Kernel::Process& src_process, const Kernel::Process& dest_process,
+               VAddr src_addr, VAddr dest_addr, std::size_t size);
 
 u8* GetPointer(VAddr vaddr);