/* Copyright 2016-2023 melonDS team This file is part of melonDS. melonDS is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. melonDS is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with melonDS. If not, see http://www.gnu.org/licenses/. */ #include #include #include #ifdef __WIN32__ #include #else #include #include #include #ifdef __APPLE__ #include "sem_timedwait.h" #endif #endif #include #include #include "Config.h" #include "LocalMP.h" #include "Platform.h" using namespace melonDS; using namespace melonDS::Platform; using Platform::Log; using Platform::LogLevel; namespace LocalMP { u32 MPUniqueID; u8 PacketBuffer[2048]; struct MPQueueHeader { u16 NumInstances; u16 InstanceBitmask; // bitmask of all instances present u16 ConnectedBitmask; // bitmask of which instances are ready to send/receive packets u32 PacketWriteOffset; u32 ReplyWriteOffset; u16 MPHostInstanceID; // instance ID from which the last CMD frame was sent u16 MPReplyBitmask; // bitmask of which clients replied in time }; struct MPPacketHeader { u32 Magic; u32 SenderID; u32 Type; // 0=regular 1=CMD 2=reply 3=ack u32 Length; u64 Timestamp; }; struct MPSync { u32 Magic; u32 SenderID; u16 ClientMask; u16 Type; u64 Timestamp; }; QSharedMemory* MPQueue; int InstanceID; u32 PacketReadOffset; u32 ReplyReadOffset; const u32 kQueueSize = 0x20000; const u32 kMaxFrameSize = 0x800; const u32 kPacketStart = sizeof(MPQueueHeader); const u32 kReplyStart = kQueueSize / 2; const u32 kPacketEnd = kReplyStart; const u32 kReplyEnd = kQueueSize; int RecvTimeout; int LastHostID; // we need to come up with our own abstraction layer for named semaphores // because QSystemSemaphore doesn't support waiting with a timeout // and, as such, is unsuitable to our needs #ifdef __WIN32__ bool SemInited[32]; HANDLE SemPool[32]; void SemPoolInit() { for (int i = 0; i < 32; i++) { SemPool[i] = INVALID_HANDLE_VALUE; SemInited[i] = false; } } void SemDeinit(int num); void SemPoolDeinit() { for (int i = 0; i < 32; i++) SemDeinit(i); } bool SemInit(int num) { if (SemInited[num]) return true; char semname[64]; sprintf(semname, "Local\\melonNIFI_Sem%02d", num); HANDLE sem = CreateSemaphoreA(nullptr, 0, 64, semname); SemPool[num] = sem; SemInited[num] = true; return sem != INVALID_HANDLE_VALUE; } void SemDeinit(int num) { if (SemPool[num] != INVALID_HANDLE_VALUE) { CloseHandle(SemPool[num]); SemPool[num] = INVALID_HANDLE_VALUE; } SemInited[num] = false; } bool SemPost(int num) { SemInit(num); return ReleaseSemaphore(SemPool[num], 1, nullptr) != 0; } bool SemWait(int num, int timeout) { return WaitForSingleObject(SemPool[num], timeout) == WAIT_OBJECT_0; } void SemReset(int num) { while (WaitForSingleObject(SemPool[num], 0) == WAIT_OBJECT_0); } #else bool SemInited[32]; sem_t* SemPool[32]; void SemPoolInit() { for (int i = 0; i < 32; i++) { SemPool[i] = SEM_FAILED; SemInited[i] = false; } } void SemDeinit(int num); void SemPoolDeinit() { for (int i = 0; i < 32; i++) SemDeinit(i); } bool SemInit(int num) { if (SemInited[num]) return true; char semname[64]; sprintf(semname, "/melonNIFI_Sem%02d", num); sem_t* sem = sem_open(semname, O_CREAT, 0644, 0); SemPool[num] = sem; SemInited[num] = true; return sem != SEM_FAILED; } void SemDeinit(int num) { if (SemPool[num] != SEM_FAILED) { sem_close(SemPool[num]); SemPool[num] = SEM_FAILED; } SemInited[num] = false; } bool SemPost(int num) { SemInit(num); return sem_post(SemPool[num]) == 0; } bool SemWait(int num, int timeout) { if (!timeout) return sem_trywait(SemPool[num]) == 0; struct timespec ts; clock_gettime(CLOCK_REALTIME, &ts); ts.tv_nsec += timeout * 1000000; long sec = ts.tv_nsec / 1000000000; ts.tv_nsec -= sec * 1000000000; ts.tv_sec += sec; return sem_timedwait(SemPool[num], &ts) == 0; } void SemReset(int num) { while (sem_trywait(SemPool[num]) == 0); } #endif bool Init() { MPQueue = new QSharedMemory("melonNIFI"); if (!MPQueue->attach()) { Log(LogLevel::Info, "MP sharedmem doesn't exist. creating\n"); if (!MPQueue->create(kQueueSize)) { Log(LogLevel::Error, "MP sharedmem create failed :(\n"); return false; } MPQueue->lock(); memset(MPQueue->data(), 0, MPQueue->size()); MPQueueHeader* header = (MPQueueHeader*)MPQueue->data(); header->PacketWriteOffset = kPacketStart; header->ReplyWriteOffset = kReplyStart; MPQueue->unlock(); } MPQueue->lock(); MPQueueHeader* header = (MPQueueHeader*)MPQueue->data(); u16 mask = header->InstanceBitmask; for (int i = 0; i < 16; i++) { if (!(mask & (1<InstanceBitmask |= (1<ConnectedBitmask |= (1 << i); break; } } header->NumInstances++; PacketReadOffset = header->PacketWriteOffset; ReplyReadOffset = header->ReplyWriteOffset; MPQueue->unlock(); // prepare semaphores // semaphores 0-15: regular frames; semaphore I is posted when instance I needs to process a new frame // semaphores 16-31: MP replies; semaphore I is posted when instance I needs to process a new MP reply SemPoolInit(); SemInit(InstanceID); SemInit(16+InstanceID); LastHostID = -1; Log(LogLevel::Info, "MP comm init OK, instance ID %d\n", InstanceID); RecvTimeout = 25; return true; } void DeInit() { if (MPQueue) { MPQueue->lock(); if (MPQueue->data() != nullptr) { MPQueueHeader *header = (MPQueueHeader *) MPQueue->data(); header->ConnectedBitmask &= ~(1 << InstanceID); header->InstanceBitmask &= ~(1 << InstanceID); header->NumInstances--; } MPQueue->unlock(); SemPoolDeinit(); MPQueue->detach(); } delete MPQueue; MPQueue = nullptr; } void SetRecvTimeout(int timeout) { RecvTimeout = timeout; } void Begin() { MPQueue->lock(); MPQueueHeader* header = (MPQueueHeader*)MPQueue->data(); PacketReadOffset = header->PacketWriteOffset; ReplyReadOffset = header->ReplyWriteOffset; SemReset(InstanceID); SemReset(16+InstanceID); header->ConnectedBitmask |= (1 << InstanceID); MPQueue->unlock(); } void End() { MPQueue->lock(); MPQueueHeader* header = (MPQueueHeader*)MPQueue->data(); //SemReset(InstanceID); //SemReset(16+InstanceID); header->ConnectedBitmask &= ~(1 << InstanceID); MPQueue->unlock(); } void FIFORead(int fifo, void* buf, int len) { u8* data = (u8*)MPQueue->data(); u32 offset, start, end; if (fifo == 0) { offset = PacketReadOffset; start = kPacketStart; end = kPacketEnd; } else { offset = ReplyReadOffset; start = kReplyStart; end = kReplyEnd; } if ((offset + len) >= end) { u32 part1 = end - offset; memcpy(buf, &data[offset], part1); memcpy(&((u8*)buf)[part1], &data[start], len - part1); offset = start + len - part1; } else { memcpy(buf, &data[offset], len); offset += len; } if (fifo == 0) PacketReadOffset = offset; else ReplyReadOffset = offset; } void FIFOWrite(int fifo, void* buf, int len) { u8* data = (u8*)MPQueue->data(); MPQueueHeader* header = (MPQueueHeader*)&data[0]; u32 offset, start, end; if (fifo == 0) { offset = header->PacketWriteOffset; start = kPacketStart; end = kPacketEnd; } else { offset = header->ReplyWriteOffset; start = kReplyStart; end = kReplyEnd; } if ((offset + len) >= end) { u32 part1 = end - offset; memcpy(&data[offset], buf, part1); memcpy(&data[start], &((u8*)buf)[part1], len - part1); offset = start + len - part1; } else { memcpy(&data[offset], buf, len); offset += len; } if (fifo == 0) header->PacketWriteOffset = offset; else header->ReplyWriteOffset = offset; } int SendPacketGeneric(u32 type, u8* packet, int len, u64 timestamp) { MPQueue->lock(); u8* data = (u8*)MPQueue->data(); MPQueueHeader* header = (MPQueueHeader*)&data[0]; u16 mask = header->ConnectedBitmask; // TODO: check if the FIFO is full! MPPacketHeader pktheader; pktheader.Magic = 0x4946494E; pktheader.SenderID = InstanceID; pktheader.Type = type; pktheader.Length = len; pktheader.Timestamp = timestamp; type &= 0xFFFF; int nfifo = (type == 2) ? 1 : 0; FIFOWrite(nfifo, &pktheader, sizeof(pktheader)); if (len) FIFOWrite(nfifo, packet, len); if (type == 1) { // NOTE: this is not guarded against, say, multiple multiplay games happening on the same machine // we would need to pass the packet's SenderID through the wifi module for that header->MPHostInstanceID = InstanceID; header->MPReplyBitmask = 0; ReplyReadOffset = header->ReplyWriteOffset; SemReset(16 + InstanceID); } else if (type == 2) { header->MPReplyBitmask |= (1 << InstanceID); } MPQueue->unlock(); if (type == 2) { SemPost(16 + header->MPHostInstanceID); } else { for (int i = 0; i < 16; i++) { if (mask & (1<lock(); u8* data = (u8*)MPQueue->data(); MPQueueHeader* header = (MPQueueHeader*)&data[0]; MPPacketHeader pktheader; FIFORead(0, &pktheader, sizeof(pktheader)); if (pktheader.Magic != 0x4946494E) { Log(LogLevel::Warn, "PACKET FIFO OVERFLOW\n"); PacketReadOffset = header->PacketWriteOffset; SemReset(InstanceID); MPQueue->unlock(); return 0; } if (pktheader.SenderID == InstanceID) { // skip this packet PacketReadOffset += pktheader.Length; if (PacketReadOffset >= kPacketEnd) PacketReadOffset += kPacketStart - kPacketEnd; MPQueue->unlock(); continue; } if (pktheader.Length) { FIFORead(0, packet, pktheader.Length); if (pktheader.Type == 1) LastHostID = pktheader.SenderID; } if (timestamp) *timestamp = pktheader.Timestamp; MPQueue->unlock(); return pktheader.Length; } } int SendPacket(u8* packet, int len, u64 timestamp) { return SendPacketGeneric(0, packet, len, timestamp); } int RecvPacket(u8* packet, u64* timestamp) { return RecvPacketGeneric(packet, false, timestamp); } int SendCmd(u8* packet, int len, u64 timestamp) { return SendPacketGeneric(1, packet, len, timestamp); } int SendReply(u8* packet, int len, u64 timestamp, u16 aid) { return SendPacketGeneric(2 | (aid<<16), packet, len, timestamp); } int SendAck(u8* packet, int len, u64 timestamp) { return SendPacketGeneric(3, packet, len, timestamp); } int RecvHostPacket(u8* packet, u64* timestamp) { if (LastHostID != -1) { // check if the host is still connected MPQueue->lock(); u8* data = (u8*)MPQueue->data(); MPQueueHeader* header = (MPQueueHeader*)&data[0]; u16 curinstmask = header->ConnectedBitmask; MPQueue->unlock(); if (!(curinstmask & (1 << LastHostID))) return -1; } return RecvPacketGeneric(packet, true, timestamp); } u16 RecvReplies(u8* packets, u64 timestamp, u16 aidmask) { u16 ret = 0; u16 myinstmask = (1 << InstanceID); u16 curinstmask; { MPQueue->lock(); u8* data = (u8*)MPQueue->data(); MPQueueHeader* header = (MPQueueHeader*)&data[0]; curinstmask = header->ConnectedBitmask; MPQueue->unlock(); } // if all clients have left: return early if ((myinstmask & curinstmask) == curinstmask) return 0; for (;;) { if (!SemWait(16+InstanceID, RecvTimeout)) { // no more replies available return ret; } MPQueue->lock(); u8* data = (u8*)MPQueue->data(); MPQueueHeader* header = (MPQueueHeader*)&data[0]; MPPacketHeader pktheader; FIFORead(1, &pktheader, sizeof(pktheader)); if (pktheader.Magic != 0x4946494E) { Log(LogLevel::Warn, "REPLY FIFO OVERFLOW\n"); ReplyReadOffset = header->ReplyWriteOffset; SemReset(16+InstanceID); MPQueue->unlock(); return 0; } if ((pktheader.SenderID == InstanceID) || // packet we sent out (shouldn't happen, but hey) (pktheader.Timestamp < (timestamp - 32))) // stale packet { // skip this packet ReplyReadOffset += pktheader.Length; if (ReplyReadOffset >= kReplyEnd) ReplyReadOffset += kReplyStart - kReplyEnd; MPQueue->unlock(); continue; } if (pktheader.Length) { u32 aid = (pktheader.Type >> 16); FIFORead(1, &packets[(aid-1)*1024], pktheader.Length); ret |= (1 << aid); } myinstmask |= (1 << pktheader.SenderID); if (((myinstmask & curinstmask) == curinstmask) || ((ret & aidmask) == aidmask)) { // all the clients have sent their reply MPQueue->unlock(); return ret; } MPQueue->unlock(); } } }