Merge branch 'shm-multi'

library/DFError.h

@@ -221,6 +221,61 @@ namespace DFHack
                 return "The server process has disappeared";
             }
         };
+        class DFHACK_EXPORT SHMLockingError : public std::exception
+        {
+        public:
+            SHMLockingError(const char* _type) : type(_type) {}
+            virtual ~SHMLockingError() throw(){};
+            
+            std::string type;
+            
+            virtual const char* what() const throw()
+            {
+                std::stringstream s;
+                s << "SHM locking error: " << type;
+                return s.str().c_str();
+            }
+        };
+        class DFHACK_EXPORT SHMAccessDenied : public std::exception
+        {
+        public:
+            SHMAccessDenied() {}
+            virtual ~SHMAccessDenied() throw(){};
+            
+            std::string type;
+            
+            virtual const char* what() const throw()
+            {
+                return "SHM ACCESS DENIED";
+            }
+        };
+        class DFHACK_EXPORT SHMVersionMismatch : public std::exception
+        {
+        public:
+            SHMVersionMismatch() {}
+            virtual ~SHMVersionMismatch() throw(){};
+            
+            std::string type;
+            
+            virtual const char* what() const throw()
+            {
+                return "SHM VERSION MISMATCH";
+            }
+        };
+        class DFHACK_EXPORT SHMAttachFailure : public std::exception
+        {
+        public:
+            SHMAttachFailure() {}
+            virtual ~SHMAttachFailure() throw(){};
+            
+            std::string type;
+            
+            virtual const char* what() const throw()
+            {
+                return "SHM ATTACH FAILURE";
+            }
+        };
+        
     }
 }
 

library/DFHackAPI.cpp

@@ -190,7 +190,7 @@ API::~API()
     delete d;
 }
 
-#define SHMCMD ((shm_cmd *)d->shm_start)->pingpong
+#define SHMCMD(num) ((shm_cmd *)d->shm_start)[num]->pingpong
 #define SHMHDR ((shm_core_hdr *)d->shm_start)
 #define SHMMAPSHDR ((Maps::shm_maps_hdr *)d->shm_start)
 #define SHMDATA(type) ((type *)(d->shm_start + SHM_HEADER))
@@ -243,8 +243,7 @@ bool API::InitMap()
         off->z_count_offset = z_count_offset;
         full_barrier
         const uint32_t cmd = Maps::MAP_INIT + d->maps_module << 16;
-        SHMCMD = cmd;
-        g_pProcess->waitWhile(cmd);
+        g_pProcess->SetAndWait(cmd);
         //cerr << "Map acceleration enabled!" << endl;
     }
     
@@ -324,13 +323,8 @@ bool API::ReadBlock40d(uint32_t x, uint32_t y, uint32_t z, mapblock40d * buffer)
         SHMMAPSHDR->y = y;
         SHMMAPSHDR->z = z;
         volatile uint32_t cmd = Maps::MAP_READ_BLOCK_BY_COORDS + (d->maps_module << 16);
-        full_barrier
-        SHMCMD = cmd;
-        full_barrier
-        if(!g_pProcess->waitWhile(cmd))
-        {
+        if(!g_pProcess->SetAndWait(cmd))
             return false;
-        }
         memcpy(buffer,SHMDATA(mapblock40d),sizeof(mapblock40d));
         return true;
     }

library/DFProcess-linux-SHM.cpp

@@ -43,96 +43,77 @@ class SHMProcess::Private
     public:
     Private()
     {
-        my_descriptor = NULL;
-        my_pid = 0;
-        my_shm = 0;
-        my_shmid = -1;
-        my_window = NULL;
+        memdescriptor = NULL;
+        process_ID = 0;
+        shm_addr = 0;
+        //shm_addr_with_cl_idx = 0;
+        shm_ID = -1;
+        window = NULL;
         attached = false;
-        suspended = false;
         identified = false;
         useYield = false;
-        my_SVfileLock = -1;
-        my_CLfileLock = -1;
+        server_lock = -1;
+        client_lock = -1;
+        suspend_lock = -1;
+        attachmentIdx = 0;
+        locked = false;
     };
     ~Private(){};
-    memory_info * my_descriptor;
-    DFWindow * my_window;
-    pid_t my_pid;
-    char *my_shm;
-    int my_shmid;
+    memory_info * memdescriptor;
+    DFWindow * window;
+    pid_t process_ID;
+    char *shm_addr;
+    int shm_ID;
     Process* q;
-    int my_SVfileLock;
-    int my_CLfileLock;
+    int server_lock;
+    int client_lock;
+    int suspend_lock;
+    int attachmentIdx;
+    
+
     
     bool attached;
-    bool suspended;
+    bool locked;
     bool identified;
     bool useYield;
     
-    bool validate(char* exe_file, uint32_t pid, std::vector< memory_info* >& known_versions);
+    bool validate(std::vector< memory_info* >& known_versions);
     
     bool Aux_Core_Attach(bool & versionOK, pid_t & PID);
-    bool waitWhile (uint32_t state);
+    //bool waitWhile (uint32_t state);
+    bool SetAndWait (uint32_t state);
     bool GetLocks();
     bool AreLocksOk();
     void FreeLocks();
 };
 
-// some helpful macros to keep the code bloat in check
-#define SHMCMD ((shm_cmd *)my_shm)->pingpong
-#define D_SHMCMD ((shm_cmd *)d->my_shm)->pingpong
+#define SHMCMD ( (uint32_t *) shm_addr)[attachmentIdx]
+#define D_SHMCMD ( (uint32_t *) (d->shm_addr))[d->attachmentIdx]
 
-#define SHMHDR ((shm_core_hdr *)my_shm)
-#define D_SHMHDR ((shm_core_hdr *)d->my_shm)
+#define SHMHDR ((shm_core_hdr *)shm_addr)
+#define D_SHMHDR ((shm_core_hdr *)(d->shm_addr))
 
-#define SHMDATA(type) ((type *)(my_shm + SHM_HEADER))
-#define D_SHMDATA(type) ((type *)(d->my_shm + SHM_HEADER))
+#define SHMDATA(type) ((type *)(shm_addr + SHM_HEADER))
+#define D_SHMDATA(type) ((type *)(d->shm_addr + SHM_HEADER))
 
-/*
-Yeah. with no way to synchronize things (locks are slow, the OS doesn't give us enough control over scheduling)
-we end up with this silly thing
-*/
-bool SHMProcess::Private::waitWhile (uint32_t state)
+bool SHMProcess::Private::SetAndWait (uint32_t state)
 {
     uint32_t cnt = 0;
-    struct shmid_ds descriptor;
+    if(!attached) return false;
+    SHMCMD = state;
     while (SHMCMD == state)
     {
-        if(cnt == 10000)// check if the other process is still there
-        {
-            
-            /*
-            shmctl(my_shmid, IPC_STAT, &descriptor);
-            if(descriptor.shm_nattch == 1)// DF crashed or exited - no way to tell?
-            {
-                //detach the shared memory
-                shmdt(my_shm);
-                attached = suspended = false;
-                
-                // we aren't the current process anymore
-                g_pProcess = NULL;
-                
-                throw Error::SHMServerDisappeared();
-                return false;
-            }
-            else
+        if(cnt == 10000)// check if the other process is still there, don't hammer the kernel too much.
         {
-                cnt = 0;
-            }
-            */
             if(!AreLocksOk())
             {
                 //detach the shared memory
-                shmdt(my_shm);
-                attached = suspended = false;
-                
+                shmdt(shm_addr);
+                FreeLocks();
+                attached = locked = identified = false;
                 // we aren't the current process anymore
                 g_pProcess = NULL;
-                FreeLocks();
-                
                 throw Error::SHMServerDisappeared();
-                return false;
             }
             else
             {
@@ -145,12 +126,9 @@ bool SHMProcess::Private::waitWhile (uint32_t state)
         }
         cnt++;
     }
+    // server returned a generic error
     if(SHMCMD == CORE_ERROR)
     {
-        SHMCMD = CORE_RUNNING;
-        attached = suspended = false;
-        cerr << "shm server error!" << endl;
-        assert (false);
         return false;
     }
     return true;
@@ -160,9 +138,9 @@ bool SHMProcess::Private::waitWhile (uint32_t state)
 Yeah. with no way to synchronize things (locks are slow, the OS doesn't give us enough control over scheduling)
 we end up with this silly thing
 */
-bool SHMProcess::waitWhile (uint32_t state)
+bool SHMProcess::SetAndWait (uint32_t state)
 {
-    return d->waitWhile(state);
+    return d->SetAndWait(state);
 }
 
 uint32_t OS_getAffinity()
@@ -174,30 +152,13 @@ uint32_t OS_getAffinity()
     return affinity;
 }
 
-
-bool SHMProcess::Private::Aux_Core_Attach(bool & versionOK, pid_t & PID)
-{
-    SHMDATA(coreattach)->cl_affinity = OS_getAffinity();
-    gcc_barrier
-    SHMCMD = CORE_ATTACH;
-    if(!waitWhile(CORE_ATTACH))
-        return false;
-    gcc_barrier
-    versionOK =( SHMDATA(coreattach)->sv_version == CORE_VERSION );
-    PID = SHMDATA(coreattach)->sv_PID;
-    useYield = SHMDATA(coreattach)->sv_useYield;
-    #ifdef DEBUG
-        if(useYield) cerr << "Using Yield!" << endl;
-    #endif
-    return true;
-}
-
+// test if we have client and server locks and the server is present
 bool SHMProcess::Private::AreLocksOk()
 {
     // both locks are inited (we hold our lock)
-    if(my_CLfileLock != -1 && my_SVfileLock != -1) 
+    if(client_lock != -1 && server_lock != -1) 
     {
-        if(lockf(my_SVfileLock,F_TEST,0) == -1) // and server holds its lock
+        if(lockf(server_lock,F_TEST,0) == -1) // and server holds its lock
         {
             return true; // OK, locks are good
         }
@@ -208,16 +169,23 @@ bool SHMProcess::Private::AreLocksOk()
 
 void SHMProcess::Private::FreeLocks()
 {
-    if(my_CLfileLock != -1)
+    attachmentIdx = -1;
+    if(client_lock != -1)
     {
-        lockf(my_CLfileLock,F_ULOCK,0);
-        close(my_CLfileLock);
-        my_CLfileLock = -1;
+        lockf(client_lock,F_ULOCK,0);
+        close(client_lock);
+        client_lock = -1;
     }
-    if(my_SVfileLock != -1)
+    if(server_lock != -1)
     {
-        close(my_SVfileLock);
-        my_SVfileLock = -1;
+        close(server_lock);
+        server_lock = -1;
+    }
+    if(suspend_lock != -1)
+    {
+        close(suspend_lock);
+        locked = false;
+        suspend_lock = -1;
     }
 }
 
@@ -226,123 +194,102 @@ bool SHMProcess::Private::GetLocks()
     char name[256];
     // try to acquire locks
     // look at the server lock, if it's locked, the server is present
-    sprintf(name, "/tmp/DFHack/%d/SVlock",my_pid,name);
-    my_SVfileLock = open(name,O_WRONLY);
-    if(my_SVfileLock == -1)
+    sprintf(name, "/tmp/DFHack/%d/SVlock",process_ID);
+    server_lock = open(name,O_WRONLY);
+    if(server_lock == -1)
     {
+        // cerr << "can't open sv lock" << endl;
         return false;
     }
     
-    if(lockf( my_SVfileLock, F_TEST, 0 ) != -1)
+    if(lockf( server_lock, F_TEST, 0 ) != -1)
     {
-        close(my_SVfileLock);
+        cerr << "sv lock not locked" << endl;
+        close(server_lock);
+        server_lock = -1;
         return false;
     }
     
-    // open the client lock, try to lock it
-    sprintf(name, "/tmp/DFHack/%d/CLlock",my_pid,name);
-    my_CLfileLock = open(name,O_WRONLY);
-    if(my_CLfileLock == -1)
+    for(int i = 0; i < SHM_MAX_CLIENTS; i++)
     {
-        close(my_SVfileLock);
-        return false;
+        // open the client suspend locked
+        sprintf(name, "/tmp/DFHack/%d/CLSlock%d",process_ID,i);
+        suspend_lock = open(name,O_WRONLY);
+        if(suspend_lock == -1)
+        {
+            cerr << "can't open cl S-lock " << i << endl;
+            // couldn't open lock
+            continue;
         }
-    if(lockf(my_CLfileLock,F_TLOCK, 0) == -1)
+
+        // open the client lock, try to lock it
+        sprintf(name, "/tmp/DFHack/%d/CLlock%d",process_ID,i);
+        client_lock = open(name,O_WRONLY);
+        if(client_lock == -1)
+        {
+            cerr << "can't open cl lock " << i << endl;
+            close(suspend_lock);
+            locked = false;
+            suspend_lock = -1;
+            // couldn't open lock
+            continue;
+        }
+        if(lockf(client_lock,F_TLOCK, 0) == -1)
         {
             // couldn't acquire lock
-        close(my_SVfileLock);
-        close(my_CLfileLock);
-        return false;
+            cerr << "can't acquire cl lock " << i << endl;
+            close(suspend_lock);
+            locked = false;
+            suspend_lock = -1;
+            close(client_lock);
+            client_lock = -1;
+            continue;
         }
-    // ok, we have all the locks!
+        // ok, we have all the locks we need!
+        attachmentIdx = i;
         return true;
+    }
+    close(server_lock);
+    server_lock = -1;
+    cerr << "can't get any client locks" << endl;
+    return false;
 }
 
 SHMProcess::SHMProcess(uint32_t PID, vector< memory_info* >& known_versions)
 : d(new Private())
 {
-    char exe_link_name [256];
-    char target_name[1024];
-    int target_result;
-    
-    /*
-     * Locate the segment.
-     */
-    if ((d->my_shmid = shmget(SHM_KEY + PID, SHM_SIZE, 0666)) < 0)
-    {
-        return;
-    }
-    
-    /*
-     * Attach the segment
-     */
-    if ((d->my_shm = (char *) shmat(d->my_shmid, NULL, 0)) == (char *) -1)
+    d->process_ID = PID;
+    if(!attach())
     {
+        // couldn't attach to process
         return;
     }
-    
-    // set pid and gets lock for it
-    d->my_pid = PID;
-    if(!d->GetLocks())
-    {
-        fprintf(stderr,"Couldn't get locks for PID %d'\n", PID);
-        shmdt(d->my_shm);
-        return;
-    }
-   
     /*
      * Test bridge version, get PID, sync Yield
      */
     bool bridgeOK;
-    if(!d->Aux_Core_Attach(bridgeOK,d->my_pid))
+    if(!d->Aux_Core_Attach(bridgeOK,d->process_ID))
     {
-        fprintf(stderr,"DF terminated during reading\n");
-        shmdt(d->my_shm);
-        // free locks
-        d->FreeLocks();
-        return;
+        detach();
+        throw Error::SHMAttachFailure();
     }
     if(!bridgeOK)
     {
-        fprintf(stderr,"SHM bridge version mismatch\n");
-        shmdt(d->my_shm);
-        // free locks
-        d->FreeLocks();
-        return;
-    }
-    
-    // find the binary
-    sprintf(exe_link_name,"/proc/%d/exe", d->my_pid);
-    target_result = readlink(exe_link_name, target_name, sizeof(target_name)-1);
-    if (target_result == -1)
-    {
-        perror("readlink");
-        shmdt(d->my_shm);
-        // free locks
-        d->FreeLocks();
-        return;
+        detach();
+        throw Error::SHMVersionMismatch();
     }
     
-    // make sure we have a null terminated string...
-    // see http://www.opengroup.org/onlinepubs/000095399/functions/readlink.html
-    target_name[target_result] = 0;
-    
-    // try to identify the DF version
-    d->validate(target_name, d->my_pid, known_versions);
-    d->my_window = new DFWindow(this);
-
-    gcc_barrier
-    // at this point, DF is stopped and waiting for commands. make it run again
-    D_SHMCMD = CORE_RUNNING;
-    shmdt(d->my_shm); // detach so we don't attach twice when attach() is called
+    // try to identify the DF version (md5 the binary, compare with known versions)
+    d->validate(known_versions);
+    d->window = new DFWindow(this);
     
-    // free locks
-    d->FreeLocks();
+    // detach
+    detach();
 }
 
 bool SHMProcess::isSuspended()
 {
-    return d->suspended;
+    return d->locked;
 }
 bool SHMProcess::isAttached()
 {
@@ -354,13 +301,28 @@ bool SHMProcess::isIdentified()
     return d->identified;
 }
 
-bool SHMProcess::Private::validate(char * exe_file, uint32_t pid, vector <memory_info *> & known_versions)
+bool SHMProcess::Private::validate(vector <memory_info *> & known_versions)
 {
+    char exe_link_name [256];
+    char target_name[1024];
+    int target_result;
+    // find the binary
+    sprintf(exe_link_name,"/proc/%d/exe", process_ID);
+    target_result = readlink(exe_link_name, target_name, sizeof(target_name)-1);
+    if (target_result == -1)
+    {
+        perror("readlink");
+        return false;
+    }
+    // make sure we have a null terminated string...
+    // see http://www.opengroup.org/onlinepubs/000095399/functions/readlink.html
+    target_name[target_result] = 0;
+    
     md5wrapper md5;
     // get hash of the running DF process
-    string hash = md5.getHashFromFile(exe_file);
+    string hash = md5.getHashFromFile(target_name);
     vector<memory_info *>::iterator it;
-    cerr << exe_file << " " << hash <<  endl;
+    // cerr << exe_file << " " << hash <<  endl;
     // iterate over the list of memory locations
     for ( it=known_versions.begin() ; it < known_versions.end(); it++ )
     {
@@ -368,10 +330,9 @@ bool SHMProcess::Private::validate(char * exe_file, uint32_t pid, vector <memory
             if(hash == (*it)->getString("md5")) // are the md5 hashes the same?
             {
                 memory_info * m = *it;
-                my_descriptor = m;
-                my_pid = pid;
+                memdescriptor = m;
                 identified = true;
-                cerr << "identified " << m->getVersion() << endl;
+                // cerr << "identified " << m->getVersion() << endl;
                 return true;
             }
         }
@@ -390,31 +351,37 @@ SHMProcess::~SHMProcess()
         detach();
     }
     // destroy data model. this is assigned by processmanager
-    if(d->my_window)
+    if(d->window)
     {
-        delete d->my_window;
+        delete d->window;
     }
     delete d;
 }
 
 memory_info * SHMProcess::getDescriptor()
 {
-    return d->my_descriptor;
+    return d->memdescriptor;
 }
 
 DFWindow * SHMProcess::getWindow()
 {
-    return d->my_window;
+    return d->window;
 }
 
 int SHMProcess::getPID()
 {
-    return d->my_pid;
+    return d->process_ID;
 }
 
-//FIXME: implement
+// there is only one we care about.
 bool SHMProcess::getThreadIDs(vector<uint32_t> & threads )
 {
+    if(d->attached)
+    {
+        threads.clear();
+        threads.push_back(d->process_ID);
+        return true;
+    }
     return false;
 }
 
@@ -424,7 +391,7 @@ void SHMProcess::getMemRanges( vector<t_memrange> & ranges )
     char buffer[1024];
     char permissions[5]; // r/-, w/-, x/-, p/s, 0
     
-    sprintf(buffer, "/proc/%lu/maps", d->my_pid);
+    sprintf(buffer, "/proc/%lu/maps", d->process_ID);
     FILE *mapFile = ::fopen(buffer, "r");
     uint64_t offset, device1, device2, node;
     
@@ -451,37 +418,79 @@ bool SHMProcess::suspend()
     {
         return false;
     }
-    if(d->suspended)
+    if(d->locked)
     {
         return true;
     }
-    D_SHMCMD = CORE_SUSPEND;
-    if(!waitWhile(CORE_SUSPEND))
+    
+    // FIXME: this should be controlled on the server side
+    // FIXME: IF server got CORE_RUN in this frame, interpret CORE_SUSPEND as CORE_STEP
+    // did we just resume a moment ago?
+    if(D_SHMCMD == CORE_RUN)
     {
-        return false;
+        //fprintf(stderr,"%d invokes step\n",d->attachmentIdx);
+        // wait for the next window
+        if(!d->SetAndWait(CORE_STEP))
+        {
+            throw Error::SHMLockingError("if(!d->SetAndWait(CORE_STEP))");
+        }
     }
-    d->suspended = true;
+    else
+    {
+        //fprintf(stderr,"%d invokes suspend\n",d->attachmentIdx);
+        // lock now
+        if(!d->SetAndWait(CORE_SUSPEND))
+        {
+            throw Error::SHMLockingError("if(!d->SetAndWait(CORE_SUSPEND))");
+        }
+    }
+    //fprintf(stderr,"waiting for lock\n");
+    // we wait for the server to give up our suspend lock (held by default)
+    if(lockf(d->suspend_lock,F_LOCK,0) == 0)
+    {
+        d->locked = true;
         return true;
+    }
+    return false;
 }
 
+// FIXME: needs a good think-through
 bool SHMProcess::asyncSuspend()
 {
     if(!d->attached)
     {
         return false;
     }
-    if(d->suspended)
+    if(d->locked)
     {
         return true;
     }
-    if(D_SHMCMD == CORE_SUSPENDED)
+    uint32_t cmd = D_SHMCMD;
+    if(cmd == CORE_SUSPENDED)
     {
-        d->suspended = true;
+        // we have to hold the lock to be really suspended
+        if(lockf(d->suspend_lock,F_LOCK,0) == 0)
+        {
+            d->locked = true;
             return true;
         }
+        return false;
+    }
+    else
+    {
+        // did we just resume a moment ago?
+        if(cmd == CORE_STEP)
+        {
+            return false;
+        }
+        else if(cmd == CORE_RUN)
+        {
+            D_SHMCMD = CORE_STEP;
+        }
         else
         {
             D_SHMCMD = CORE_SUSPEND;
+        }
         return false;
     }
 }
@@ -491,21 +500,30 @@ bool SHMProcess::forceresume()
     return resume();
 }
 
+// FIXME: wait for the server to advance a step!
 bool SHMProcess::resume()
 {
     if(!d->attached)
         return false;
-    if(!d->suspended)
+    if(!d->locked)
         return true;
-    D_SHMCMD = CORE_RUNNING;
-    d->suspended = false;
+    // unlock the suspend lock
+    if(lockf(d->suspend_lock,F_ULOCK,0) == 0)
+    {
+        d->locked = false;
+        if(d->SetAndWait(CORE_RUN)) // we have to make sure the server responds!
+        {
             return true;
+        }
+        throw Error::SHMLockingError("if(d->SetAndWait(CORE_RUN))");
+    }
+    throw Error::SHMLockingError("if(lockf(d->suspend_lock,F_ULOCK,0) == 0)");
+    return false;
 }
 
 
 bool SHMProcess::attach()
 {
-    int status;
     if(g_pProcess != 0)
     {
         // FIXME: throw exception here - programmer error
@@ -514,31 +532,39 @@ bool SHMProcess::attach()
     }
     if(!d->GetLocks())
     {
-        cerr << "server is full or not really there!" << endl;
+        //cerr << "server is full or not really there!" << endl;
         return false;
     }
 
     /*
-    * Attach the segment
+    * Locate the segment.
     */
-    if ((d->my_shm = (char *) shmat(d->my_shmid, NULL, 0)) != (char *) -1)
+    if ((d->shm_ID = shmget(SHM_KEY + d->process_ID, SHM_SIZE, 0666)) < 0)
     {
-        d->attached = true;
-        if(suspend())
-        {
-            d->suspended = true;
-            g_pProcess = this;
-            return true;
+        d->FreeLocks();
+        cerr << "can't find segment" << endl; // FIXME: throw
+        return false;
     }
-        d->attached = false;
-        cerr << "unable to suspend" << endl;
-        shmdt(d->my_shm);
+    
+    /*
+    * Attach the segment
+    */
+    if ((d->shm_addr = (char *) shmat(d->shm_ID, NULL, 0)) == (char *) -1)
+    {
         d->FreeLocks();
+        cerr << "can't attach segment" << endl; // FIXME: throw
         return false;
     }
-    cerr << "unable to attach" << endl;
+    d->attached = true;
+    if(!suspend())
+    {
+        shmdt(d->shm_addr);
         d->FreeLocks();
+        cerr << "unable to suspend" << endl;
         return false;
+    }
+    g_pProcess = this;
+    return true;
 }
 
 bool SHMProcess::detach()
@@ -547,18 +573,18 @@ bool SHMProcess::detach()
     {
         return false;
     }
-    if(d->suspended)
+    if(d->locked)
     {
         resume();
     }
     // detach segment
-    if(shmdt(d->my_shm) != -1)
+    if(shmdt(d->shm_addr) != -1)
     {
+        d->FreeLocks();
+        d->locked = false;
         d->attached = false;
-        d->suspended = false;
-        d->my_shm = 0;
+        d->shm_addr = 0;
         g_pProcess = 0;
-        d->FreeLocks();
         return true;
     }
     // fail if we can't detach
@@ -569,15 +595,16 @@ bool SHMProcess::detach()
 
 void SHMProcess::read (uint32_t src_address, uint32_t size, uint8_t *target_buffer)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+    
     // normal read under 1MB
     if(size <= SHM_BODY)
     {
         D_SHMHDR->address = src_address;
         D_SHMHDR->length = size;
         gcc_barrier
-        D_SHMCMD = CORE_DFPP_READ;
-        waitWhile(CORE_DFPP_READ);
-        memcpy (target_buffer, d->my_shm + SHM_HEADER,size);
+        d->SetAndWait(CORE_READ);
+        memcpy (target_buffer, D_SHMDATA(void),size);
     }
     // a big read, we pull data over the shm in iterations
     else
@@ -590,9 +617,8 @@ void SHMProcess::read (uint32_t src_address, uint32_t size, uint8_t *target_buff
             D_SHMHDR->address = src_address;
             D_SHMHDR->length = to_read;
             gcc_barrier
-            D_SHMCMD = CORE_DFPP_READ;
-            waitWhile(CORE_DFPP_READ);
-            memcpy (target_buffer, d->my_shm + SHM_HEADER,size);
+            d->SetAndWait(CORE_READ);
+            memcpy (target_buffer, D_SHMDATA(void) ,size);
             // decrease size by bytes read
             size -= to_read;
             // move the cursors
@@ -606,54 +632,60 @@ void SHMProcess::read (uint32_t src_address, uint32_t size, uint8_t *target_buff
 
 uint8_t SHMProcess::readByte (const uint32_t offset)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+    
     D_SHMHDR->address = offset;
     gcc_barrier
-    D_SHMCMD = CORE_READ_BYTE;
-    waitWhile(CORE_READ_BYTE);
+    d->SetAndWait(CORE_READ_BYTE);
     return D_SHMHDR->value;
 }
 
 void SHMProcess::readByte (const uint32_t offset, uint8_t &val )
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+    
     D_SHMHDR->address = offset;
     gcc_barrier
-    D_SHMCMD = CORE_READ_BYTE;
-    waitWhile(CORE_READ_BYTE);
+    d->SetAndWait(CORE_READ_BYTE);
     val = D_SHMHDR->value;
 }
 
 uint16_t SHMProcess::readWord (const uint32_t offset)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+    
     D_SHMHDR->address = offset;
     gcc_barrier
-    D_SHMCMD = CORE_READ_WORD;
-    waitWhile(CORE_READ_WORD);
+    d->SetAndWait(CORE_READ_WORD);
     return D_SHMHDR->value;
 }
 
 void SHMProcess::readWord (const uint32_t offset, uint16_t &val)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+    
     D_SHMHDR->address = offset;
     gcc_barrier
-    D_SHMCMD = CORE_READ_WORD;
-    waitWhile(CORE_READ_WORD);
+    d->SetAndWait(CORE_READ_WORD);
     val = D_SHMHDR->value;
 }
 
 uint32_t SHMProcess::readDWord (const uint32_t offset)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+    
     D_SHMHDR->address = offset;
     gcc_barrier
-    D_SHMCMD = CORE_READ_DWORD;
-    waitWhile(CORE_READ_DWORD);
+    d->SetAndWait(CORE_READ_DWORD);
     return D_SHMHDR->value;
 }
 void SHMProcess::readDWord (const uint32_t offset, uint32_t &val)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+    
     D_SHMHDR->address = offset;
     gcc_barrier
-    D_SHMCMD = CORE_READ_DWORD;
-    waitWhile(CORE_READ_DWORD);
+    d->SetAndWait(CORE_READ_DWORD);
     val = D_SHMHDR->value;
 }
 
@@ -663,43 +695,47 @@ void SHMProcess::readDWord (const uint32_t offset, uint32_t &val)
 
 void SHMProcess::writeDWord (uint32_t offset, uint32_t data)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+    
     D_SHMHDR->address = offset;
     D_SHMHDR->value = data;
     gcc_barrier
-    D_SHMCMD = CORE_WRITE_DWORD;
-    waitWhile(CORE_WRITE_DWORD);
+    d->SetAndWait(CORE_WRITE_DWORD);
 }
 
 // using these is expensive.
 void SHMProcess::writeWord (uint32_t offset, uint16_t data)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+    
     D_SHMHDR->address = offset;
     D_SHMHDR->value = data;
     gcc_barrier
-    D_SHMCMD = CORE_WRITE_WORD;
-    waitWhile(CORE_WRITE_WORD);
+    d->SetAndWait(CORE_WRITE_WORD);
 }
 
 void SHMProcess::writeByte (uint32_t offset, uint8_t data)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+    
     D_SHMHDR->address = offset;
     D_SHMHDR->value = data;
     gcc_barrier
-    D_SHMCMD = CORE_WRITE_BYTE;
-    waitWhile(CORE_WRITE_BYTE);
+    d->SetAndWait(CORE_WRITE_BYTE);
 }
 
 void SHMProcess::write (uint32_t dst_address, uint32_t size, uint8_t *source_buffer)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+    
     // normal write under 1MB
     if(size <= SHM_BODY)
     {
         D_SHMHDR->address = dst_address;
         D_SHMHDR->length = size;
-        memcpy(d->my_shm+SHM_HEADER,source_buffer, size);
+        memcpy(D_SHMDATA(void),source_buffer, size);
         gcc_barrier
-        D_SHMCMD = CORE_WRITE;
-        waitWhile(CORE_WRITE);
+        d->SetAndWait(CORE_WRITE);
     }
     // a big write, we push this over the shm in iterations
     else
@@ -711,10 +747,9 @@ void SHMProcess::write (uint32_t dst_address, uint32_t size, uint8_t *source_buf
             // write to_write bytes to dst_cursor
             D_SHMHDR->address = dst_address;
             D_SHMHDR->length = to_write;
-            memcpy(d->my_shm+SHM_HEADER,source_buffer, to_write);
+            memcpy(D_SHMDATA(void),source_buffer, to_write);
             gcc_barrier
-            D_SHMCMD = CORE_WRITE;
-            waitWhile(CORE_WRITE);
+            d->SetAndWait(CORE_WRITE);
             // decrease size by bytes written
             size -= to_write;
             // move the cursors
@@ -729,6 +764,8 @@ void SHMProcess::write (uint32_t dst_address, uint32_t size, uint8_t *source_buf
 // FIXME: butt-fugly
 const std::string SHMProcess::readCString (uint32_t offset)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+        
     std::string temp;
     char temp_c[256];
     int counter = 0;
@@ -746,6 +783,8 @@ const std::string SHMProcess::readCString (uint32_t offset)
 
 DfVector SHMProcess::readVector (uint32_t offset, uint32_t item_size)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+    
     /*
         GNU libstdc++ vector is three pointers long
         ptr start
@@ -762,38 +801,42 @@ DfVector SHMProcess::readVector (uint32_t offset, uint32_t item_size)
 
 const std::string SHMProcess::readSTLString(uint32_t offset)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+        
     D_SHMHDR->address = offset;
     full_barrier
-    D_SHMCMD = CORE_READ_STL_STRING;
-    waitWhile(CORE_READ_STL_STRING);
-    //int length = ((shm_retval *)d->my_shm)->value;
-    return(string( (char *)d->my_shm+SHM_HEADER));
+    d->SetAndWait(CORE_READ_STL_STRING);
+    return(string( D_SHMDATA(char) ));
 }
 
 size_t SHMProcess::readSTLString (uint32_t offset, char * buffer, size_t bufcapacity)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+        
     D_SHMHDR->address = offset;
     full_barrier
-    D_SHMCMD = CORE_READ_STL_STRING;
-    waitWhile(CORE_READ_STL_STRING);
+    d->SetAndWait(CORE_READ_STL_STRING);
     size_t length = D_SHMHDR->value;
     size_t fit = min(bufcapacity - 1, length);
-    strncpy(buffer,(char *)d->my_shm+SHM_HEADER,fit);
+    strncpy(buffer,D_SHMDATA(char),fit);
     buffer[fit] = 0;
     return fit;
 }
 
 void SHMProcess::writeSTLString(const uint32_t address, const std::string writeString)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+        
     D_SHMHDR->address = address;
-    strncpy(d->my_shm+SHM_HEADER,writeString.c_str(),writeString.length()+1); // length + 1 for the null terminator
+    strncpy(D_SHMDATA(char),writeString.c_str(),writeString.length()+1); // length + 1 for the null terminator
     full_barrier
-    D_SHMCMD = CORE_WRITE_STL_STRING;
-    waitWhile(CORE_WRITE_STL_STRING);
+    d->SetAndWait(CORE_WRITE_STL_STRING);
 }
 
 string SHMProcess::readClassName (uint32_t vptr)
 {
+    if(!d->locked) throw Error::SHMAccessDenied();
+        
     int typeinfo = readDWord(vptr - 0x4);
     int typestring = readDWord(typeinfo + 0x4);
     string raw = readCString(typestring);
@@ -802,22 +845,18 @@ string SHMProcess::readClassName (uint32_t vptr)
     return raw.substr(start,end-start - 2); // trim the 'st' from the end
 }
 
-// FIXME: having this around could lead to bad things in the hands of unsuspecting fools
-// *!!DON'T BE AN UNSUSPECTING FOOL!!*
-// the whole SHM thing works only because copying DWORDS is an atomic operation on i386 and x86_64 archs
-// get module index by name and version. bool 1 = error
+// get module index by name and version. bool 0 = error
 bool SHMProcess::getModuleIndex (const char * name, const uint32_t version, uint32_t & OUTPUT)
 {
-    modulelookup * payload = (modulelookup *) (d->my_shm + SHM_HEADER);
+    if(!d->locked) throw Error::SHMAccessDenied();
+        
+    modulelookup * payload = D_SHMDATA(modulelookup);
     payload->version = version;
     
     strncpy(payload->name,name,255);
     payload->name[255] = 0;
     
-    full_barrier
-    
-    D_SHMCMD = CORE_ACQUIRE_MODULE;
-    if(!waitWhile(CORE_ACQUIRE_MODULE))
+    if(!SetAndWait(CORE_ACQUIRE_MODULE))
     {
         return false; // FIXME: throw a fatal exception instead
     }
@@ -832,5 +871,26 @@ bool SHMProcess::getModuleIndex (const char * name, const uint32_t version, uint
 
 char * SHMProcess::getSHMStart (void)
 {
-    return d->my_shm;
+    if(!d->locked) return 0; //THROW HERE!
+        
+    return /*d->shm_addr_with_cl_idx*/ d->shm_addr;
+}
+
+bool SHMProcess::Private::Aux_Core_Attach(bool & versionOK, pid_t & PID)
+{
+    if(!locked) throw Error::SHMAccessDenied();
+    
+    SHMDATA(coreattach)->cl_affinity = OS_getAffinity();
+    if(!SetAndWait(CORE_ATTACH)) return false;
+    /*
+    cerr <<"CORE_VERSION" << CORE_VERSION << endl;
+    cerr <<"server CORE_VERSION" << SHMDATA(coreattach)->sv_version << endl;
+    */
+    versionOK =( SHMDATA(coreattach)->sv_version == CORE_VERSION );
+    PID = SHMDATA(coreattach)->sv_PID;
+    useYield = SHMDATA(coreattach)->sv_useYield;
+    #ifdef DEBUG
+        if(useYield) cerr << "Using Yield!" << endl;
+    #endif
+    return true;
 }

library/DFProcess.h

@@ -123,8 +123,14 @@ namespace DFHack
             virtual bool getModuleIndex (const char * name, const uint32_t version, uint32_t & OUTPUT) = 0;
             // get the SHM start if available
             virtual char * getSHMStart (void) = 0;
-            // wait for a SHM state. returns 0 without the SHM
+            // set a SHM command and wait for a response, return 0 on error or throw exception
+            virtual bool SetAndWait (uint32_t state) = 0;
+            /*
+            // wait while SHM command == state. returns 0 without the SHM
             virtual bool waitWhile (uint32_t state) = 0;
+            // set SHM command.
+            virtual void setCmd (uint32_t newstate) = 0;
+            */
     };
 
     class DFHACK_EXPORT NormalProcess : virtual public Process
@@ -181,8 +187,14 @@ namespace DFHack
             bool getModuleIndex (const char * name, const uint32_t version, uint32_t & OUTPUT) {return false;};
             // get the SHM start if available
             char * getSHMStart (void){return 0;};
+            // set a SHM command and wait for a response
+            bool SetAndWait (uint32_t state){return false;};
+            /*
             // wait for a SHM state. returns 0 without the SHM
             bool waitWhile (uint32_t state){return false;};
+            // set SHM command.
+            void setCmd (uint32_t newstate){};
+            */
     };
     
     class DFHACK_EXPORT SHMProcess : virtual public Process
@@ -240,8 +252,13 @@ namespace DFHack
             bool getModuleIndex (const char * name, const uint32_t version, uint32_t & OUTPUT);
             // get the SHM start if available
             char * getSHMStart (void);
+            bool SetAndWait (uint32_t state);
+            /*
             // wait for a SHM state. returns 0 without the SHM
             bool waitWhile (uint32_t state);
+            // set SHM command.
+            void setCmd (uint32_t newstate);
+            */
     };
 
 #ifdef LINUX_BUILD
@@ -299,8 +316,13 @@ namespace DFHack
             bool getModuleIndex (const char * name, const uint32_t version, uint32_t & OUTPUT) {return false;};
             // get the SHM start if available
             char * getSHMStart (void){return 0;};
+            bool SetAndWait (uint32_t state){return false;};
+            /*
             // wait for a SHM state. returns 0 without the SHM
             bool waitWhile (uint32_t state){return false;};
+            // set SHM command.
+            void setCmd (uint32_t newstate){};
+            */
     };
 #endif
 }

shmserver/mod-core.cpp

@@ -41,19 +41,22 @@ distribution.
 
 std::vector <DFPP_module> module_registry;
 
-// various crud
+// shared by shms_OS
 extern int errorstate;
 extern char *shm;
 extern int shmid;
+
+// file-globals
 bool useYield = 0;
+int currentClient = -1;
 
 #define SHMHDR ((shm_core_hdr *)shm)
-#define SHMCMD ((shm_cmd *)shm)->pingpong
+#define SHMCMD ((uint32_t *)shm )[currentClient]
 #define SHMDATA(type) ((type *)(shm + SHM_HEADER))
 
 void ReadRaw (void * data)
 {
-    memcpy(shm + SHM_HEADER, (void *) SHMHDR->address,SHMHDR->length);
+    memcpy(SHMDATA(void), (void *) SHMHDR->address,SHMHDR->length);
 }
 
 void ReadDWord (void * data)
@@ -73,7 +76,7 @@ void ReadByte (void * data)
 
 void WriteRaw (void * data)
 {
-    memcpy((void *)SHMHDR->address, shm + SHM_HEADER,SHMHDR->length);
+    memcpy((void *)SHMHDR->address, SHMDATA(void),SHMHDR->length);
 }
 
 void WriteDWord (void * data)
@@ -97,14 +100,14 @@ void ReadSTLString (void * data)
     unsigned int l = myStringPtr->length();
     SHMHDR->value = l;
     // FIXME: there doesn't have to be a null terminator!
-    strncpy(shm+SHM_HEADER,myStringPtr->c_str(),l+1);
+    strncpy( SHMDATA(char),myStringPtr->c_str(),l+1);
 }
 
 void WriteSTLString (void * data)
 {
     std::string * myStringPtr = (std::string *) SHMHDR->address;
     // here we DO expect a 0 terminator
-    myStringPtr->assign((const char *) (shm + SHM_HEADER));
+    myStringPtr->assign( SHMDATA(const char) );
 }
 
 // MIT HAKMEM bitcount
@@ -133,7 +136,7 @@ void CoreAttach (void * data)
 void FindModule (void * data)
 {
     bool found = false;
-    modulelookup * payload = (modulelookup *) (shm + SHM_HEADER);
+    modulelookup * payload =  SHMDATA(modulelookup);
     std::string test = payload->name;
     uint32_t version = payload->version;
     for(unsigned int i = 0; i < module_registry.size();i++)
@@ -175,6 +178,17 @@ void FindCommand (void * data)
     SHMHDR->error = true;
 }
 
+void ReleaseSuspendLock( void * data )
+{
+    OS_releaseSuspendLock(currentClient);
+}
+
+void AcquireSuspendLock( void * data )
+{
+    OS_lockSuspendLock(currentClient);
+}
+
+
 DFPP_module InitCore(void)
 {
     DFPP_module core;
@@ -185,7 +199,10 @@ DFPP_module InitCore(void)
     core.reserve(NUM_CORE_CMDS);
     // basic states
     core.set_command(CORE_RUNNING, CANCELLATION, "Running");
-    core.set_command(CORE_SUSPEND, CLIENT_WAIT, "Suspend", 0 , CORE_SUSPENDED);
+    //core.set_command(CORE_RUN, FUNCTION, "Run!",AcquireSuspendLock,CORE_RUNNING);
+    core.set_command(CORE_RUN, CANCELLATION, "Run!",0,CORE_RUNNING);
+    core.set_command(CORE_STEP, CANCELLATION, "Suspend on next step",0,CORE_SUSPEND);// set command to CORE_SUSPEND, check next client
+    core.set_command(CORE_SUSPEND, FUNCTION, "Suspend", ReleaseSuspendLock , CORE_SUSPENDED);
     core.set_command(CORE_SUSPENDED, CLIENT_WAIT, "Suspended");
     core.set_command(CORE_ERROR, CANCELLATION, "Error");
     
@@ -195,7 +212,7 @@ DFPP_module InitCore(void)
     core.set_command(CORE_ACQUIRE_COMMAND, FUNCTION, "Command lookup", FindCommand, CORE_SUSPENDED);
     
     // raw reads
-    core.set_command(CORE_DFPP_READ, FUNCTION,"Raw read",ReadRaw, CORE_SUSPENDED);
+    core.set_command(CORE_READ, FUNCTION,"Raw read",ReadRaw, CORE_SUSPENDED);
     core.set_command(CORE_READ_DWORD, FUNCTION,"Read DWORD",ReadDWord, CORE_SUSPENDED);
     core.set_command(CORE_READ_WORD, FUNCTION,"Read WORD",ReadWord, CORE_SUSPENDED);
     core.set_command(CORE_READ_BYTE, FUNCTION,"Read BYTE",ReadByte, CORE_SUSPENDED);
@@ -233,16 +250,21 @@ void KillModules (void)
 
 void SHM_Act (void)
 {
+    volatile uint32_t atomic = 0;
     if(errorstate)
     {
         return;
     }
+    //static uint oldcl = 88;
+    for(currentClient = 0; currentClient < SHM_MAX_CLIENTS;currentClient++)
+    {
+        // set the offset for the shared memory used for the client
         uint32_t numwaits = 0;
         check_again: // goto target!!!
         if(numwaits == 10000)
         {
             // this tests if there's a process on the other side
-        if(isValidSHM())
+            if(isValidSHM(currentClient))
             {
                 numwaits = 0;
             }
@@ -253,32 +275,52 @@ void SHM_Act (void)
                 fprintf(stderr,"dfhack: Broke out of loop, other process disappeared.\n");
             }
         }
+        full_barrier // I don't want the compiler to reorder my code.
+
+
+        //fprintf(stderr,"%d: %x %x\n",currentClient, (uint) SHMHDR, (uint) &(SHMHDR->cmd[currentClient]));
         
         // this is very important! copying two words separately from the command variable leads to inconsistency.
         // Always copy the thing in one go.
         // Also, this whole SHM thing probably only works on intel processors
-    
-    volatile shm_cmd atomic = SHMHDR->cmd;
-    full_barrier
-    DFPP_module & mod = module_registry[atomic.parts.module];
-    DFPP_command & cmd = mod.commands[atomic.parts.command];
+        atomic  = SHMCMD;
         full_barrier
+        
+        DFPP_module & mod = module_registry[ ((shm_cmd)atomic).parts.module ];
+        DFPP_command & cmd = mod.commands[ ((shm_cmd)atomic).parts.command ];
         /*
-    fprintf(stderr, "Called %x\0", cmd._function);
-    fprintf(stderr, "Client invoked %d:%d = ",atomic.parts.module,atomic.parts.command);
-    fprintf(stderr, "%s\n",cmd.name.c_str());
-    */
+        if(atomic == CORE_RUNNING)
+        {
+            // we are running again for this process
+            // reaquire the suspend lock
+            OS_lockSuspendLock(currentClient);
+            continue;
+        }
         full_barrier
+        */
         if(cmd._function)
         {
             cmd._function(mod.modulestate);
         }
         full_barrier
+        
         if(cmd.nextState != -1)
         {
+            /*
+            char text [512];
+            char text2 [512];
+            sprintf (text,"Client %d invoked %d:%d = %x = %s\n",currentClient,((shm_cmd)atomic).parts.module,((shm_cmd)atomic).parts.command, cmd._function,cmd.name.c_str());
+            sprintf(text2, "Server set %d\n",cmd.nextState);
+            */
+            // FIXME: WHAT HAPPENS WHEN A 'NEXTSTATE' IS FROM A DIFFERENT MODULE THAN 'CORE'? Yeah. It doesn't work.
             SHMCMD = cmd.nextState;
+            //MessageBox(0,text,text2, MB_OK);
+            
+            //fflush(stderr); // make sure this finds its way to the terminal!
+            
         }
         full_barrier
+        
         if(cmd.type != CANCELLATION)
         {
             if(useYield)
@@ -288,5 +330,10 @@ void SHM_Act (void)
             numwaits ++; // watchdog timeout
             goto check_again;
         }
-}
+        full_barrier
         
+        // we are running again for this process
+        // reaquire the suspend lock
+        OS_lockSuspendLock(currentClient);
+    }
+}

shmserver/mod-core.h

@@ -26,11 +26,11 @@ distribution.
 #define SHMS_CORE_H
 
 // increment on every core change
-#define CORE_VERSION 7
+#define CORE_VERSION 8
 
 typedef struct
 {
-    shm_cmd cmd;
+    shm_cmd cmd[SHM_MAX_CLIENTS]; // MANDATORY!
     uint32_t address;
     uint32_t value;
     uint32_t length;
@@ -62,6 +62,8 @@ enum CORE_COMMAND
 {
     // basic states
     CORE_RUNNING = 0, // no command, normal server execution
+    CORE_RUN, // sent by the client to restart the server execution
+    CORE_STEP, // client suspend sets step
     CORE_SUSPEND, // client notifies server to wait for commands (server is stalled in busy wait)
     CORE_SUSPENDED, // response to WAIT, server is stalled in busy wait
     CORE_ERROR, // there was a server error
@@ -72,7 +74,7 @@ enum CORE_COMMAND
     CORE_ACQUIRE_COMMAND, // get module::command callsign by module name, command name and module version
 
     // raw reads
-    CORE_DFPP_READ, // cl -> sv, read some data
+    CORE_READ, // cl -> sv, read some data
     CORE_READ_DWORD, // cl -> sv, read a dword
     CORE_READ_WORD, // cl -> sv, read a word
     CORE_READ_BYTE, // cl -> sv, read a byte

shmserver/mod-maps.cpp

@@ -19,7 +19,7 @@ extern char *shm;
 
 #define SHMHDR ((shm_maps_hdr *)shm)
 #define SHMCMD ((shm_cmd *)shm)->pingpong
-#define SHMDATA ((char *)(shm + SHM_HEADER))
+#define SHMDATA(type) ((type *)(shm + SHM_HEADER))
 
 void NullCommand (void* data)
 {
@@ -28,7 +28,7 @@ void NullCommand (void* data)
 void InitOffsets (void* data)
 {
     maps_modulestate * state = (maps_modulestate *) data;
-    memcpy((void *) &(state->offsets), SHMDATA, sizeof(maps_offsets));
+    memcpy((void *) &(state->offsets), SHMDATA(void), sizeof(maps_offsets));
     ((maps_modulestate *) data)->inited = true;
 }
 
@@ -53,8 +53,6 @@ struct mblock
     uint32_t * ptr_to_dirty;
 };
 
-#define SHMBLOCK ((mapblock40d *)(shm + SHM_HEADER))
-
 inline void ReadBlockByAddress (void * data)
 {
     maps_modulestate * state = (maps_modulestate *) data;
@@ -62,13 +60,13 @@ inline void ReadBlockByAddress (void * data)
     mblock * block = (mblock *) SHMHDR->address;
     if(block)
     {
-        memcpy(&(SHMBLOCK->tiletypes), ((char *) block) + offsets.tile_type_offset, sizeof(SHMBLOCK->tiletypes));
-        memcpy(&(SHMBLOCK->designaton), ((char *) block) + offsets.designation_offset, sizeof(SHMBLOCK->designaton));
-        memcpy(&(SHMBLOCK->occupancy), ((char *) block) + offsets.occupancy_offset, sizeof(SHMBLOCK->occupancy));
-        memcpy(&(SHMBLOCK->biome_indices), ((char *) block) + offsets.biome_stuffs, sizeof(SHMBLOCK->biome_indices));
-        SHMBLOCK->dirty_dword = *block->ptr_to_dirty;
+        memcpy(&(SHMDATA(mapblock40d)->tiletypes), ((char *) block) + offsets.tile_type_offset, sizeof(SHMDATA(mapblock40d)->tiletypes));
+        memcpy(&(SHMDATA(mapblock40d)->designaton), ((char *) block) + offsets.designation_offset, sizeof(SHMDATA(mapblock40d)->designaton));
+        memcpy(&(SHMDATA(mapblock40d)->occupancy), ((char *) block) + offsets.occupancy_offset, sizeof(SHMDATA(mapblock40d)->occupancy));
+        memcpy(&(SHMDATA(mapblock40d)->biome_indices), ((char *) block) + offsets.biome_stuffs, sizeof(SHMDATA(mapblock40d)->biome_indices));
+        SHMDATA(mapblock40d)->dirty_dword = *block->ptr_to_dirty;
         
-        SHMBLOCK->origin = (uint32_t)block;
+        SHMDATA(mapblock40d)->origin = (uint32_t)block;
         SHMHDR->error = false;
     }
     else

shmserver/mod-maps.h

@@ -71,7 +71,7 @@ typedef struct
 
 typedef struct
 {
-    shm_cmd cmd;
+    shm_cmd cmd[SHM_MAX_CLIENTS]; // MANDATORY!
     uint32_t x;
     uint32_t y;
     uint32_t z;

shmserver/shms-linux.cpp

@@ -51,38 +51,27 @@ char *shm = 0;
 int shmid = 0;
 bool inited = 0;
 
-int fd_svlock = 0;
-int fd_cllock = 0;
-
+int fd_svlock = -1;
+int fd_cllock[SHM_MAX_CLIENTS];
+int fd_clSlock[SHM_MAX_CLIENTS];
+int held_clSlock[SHM_MAX_CLIENTS];
+int numheld = SHM_MAX_CLIENTS;
 
 /*******************************************************************************
 *                           SHM part starts here                               *
 *******************************************************************************/
-/*
-// FIXME: add error checking?
-bool isValidSHM()
-{
-    shmid_ds descriptor;
-    shmctl(shmid, IPC_STAT, &descriptor);
-    //fprintf(stderr,"ID %d, attached: %d\n",shmid, descriptor.shm_nattch);
-    return (descriptor.shm_nattch == 2);
-}
-*/
-
 // is the other side still there?
-bool isValidSHM()
+bool isValidSHM(int which)
 {
-    // try if CL lock file is free
-    int result = lockf(fd_cllock,F_TEST,0);
-    /*
-    F_TEST: Test the lock:
-    return 0 if the specified section is unlocked or locked by this process;
-    return -1, set errno to EAGAIN (EACCES on some other systems), if another process holds a lock.
-    */
-    
-    // if nobody holds the lock, the SHM isn't valid. file locks are unlocked when the owner closes or crashes.
-    if(result == 0) return false;
-    
+    // if we get the client lock here, the client process failed and we need to reclaim our suspend lock
+    if(lockf(fd_cllock[which],F_TLOCK,0) == 0)
+    {
+        // we get back our suspend lock from the cold, dead hands of the former client :P
+        OS_lockSuspendLock(which);
+        // free the client lock again
+        lockf(fd_cllock[which],F_ULOCK,0);
+        return false;
+    }
     return true;
 }
 
@@ -100,6 +89,45 @@ uint32_t OS_getAffinity()
     return affinity;
 }
 
+void OS_lockSuspendLock(int which)
+{
+    if(numheld == SHM_MAX_CLIENTS)
+        return;
+    // lock not held by server and can be picked up. OK.
+    if(held_clSlock[which] == 0 && lockf(fd_clSlock[which],F_LOCK,0) == 0)
+    {
+        held_clSlock[which] = 1;
+        numheld++;
+    }
+    // lock couldn't be picked up!
+    else if (held_clSlock[which] == 0)
+    {
+        fprintf(stderr,"lock %d failed to lock\n", which);
+    }
+}
+
+void OS_releaseSuspendLock(int which)
+{
+    /*
+    if(which >=0 && which < SHM_MAX_CLIENTS)
+        return;
+    */
+    if(numheld != SHM_MAX_CLIENTS)
+    {
+        fprintf(stderr,"TOTAL FAILURE OF LOCKING SYSTEM\n");
+        return;
+    }
+    // lock hel by server and can be released -> OK
+    if(held_clSlock[which] == 1 && lockf(fd_clSlock[which],F_ULOCK,0) == 0)
+    {
+        numheld--;
+        held_clSlock[which] = 0;
+    }
+    // locked and not can't be released? FAIL!
+    else if (held_clSlock[which] == 1)
+        fprintf(stderr,"lock %d failed to unlock\n", which);
+}
+
 void SHM_Init ( void )
 {
     // check that we do this only once per process
@@ -111,7 +139,6 @@ void SHM_Init ( void )
     inited = true;
     char name[256];
     char name2[256];
-    
     // make folder structure for our lock files
     sprintf(name, "/tmp/DFHack/%d",OS_getPID());
     mode_t createmode = S_IRWXU | S_IRWXG | S_IROTH | S_IXOTH;
@@ -123,9 +150,17 @@ void SHM_Init ( void )
     fd_svlock = open(name2,O_WRONLY | O_CREAT, createmode);
     lockf( fd_svlock, F_LOCK, 0 );
     
+    for(int i = 0; i < SHM_MAX_CLIENTS; i++)
+    {
         // create the client lock file
-    sprintf(name2, "%s/CLlock",name);
-    fd_cllock = open(name2,O_WRONLY | O_CREAT, createmode);
+        sprintf(name2, "%s/CLlock%d",name,i);
+        fd_cllock[i] = open(name2,O_WRONLY | O_CREAT, createmode);
+        // get and lock the suspend locks
+        sprintf(name2, "%s/CLSlock%d",name,i);
+        fd_clSlock[i] = open(name2,O_WRONLY | O_CREAT, createmode);
+        lockf(fd_clSlock[i],F_LOCK,0);
+        held_clSlock[i] = 1;
+    }
     
     // name for the segment, an accident waiting to happen
     key_t key = SHM_KEY + OS_getPID();
@@ -159,7 +194,10 @@ void SHM_Init ( void )
     }
     full_barrier
     // make sure we don't stall or do crazy stuff
-    ((shm_cmd *)shm)->pingpong = CORE_RUNNING;
+    for(int i = 0; i < SHM_MAX_CLIENTS;i++)
+    {
+        ((uint32_t *)shm)[i] = CORE_RUNNING;
+    }
     InitModules();
 }
 
@@ -177,21 +215,29 @@ void SHM_Destroy ( void )
         }
         shmctl(shmid,IPC_RMID,NULL);
         
-        // unlock and close server lock, close client lock
-        lockf(fd_svlock,F_ULOCK,0);
-        close(fd_svlock);
-        close(fd_cllock);
-        fd_svlock = 0;
-        fd_cllock = 0;
-
-        // destroy lock files
         char name[256];
         char name2[256];
         sprintf(name, "/tmp/DFHack/%d",OS_getPID());
-        sprintf(name2, "%s/SVlock",name);
+        
+        // unlock and close server lock, close client lock, destroy files
+        lockf(fd_svlock,F_ULOCK,0);
+        for(int i = 0; i < SHM_MAX_CLIENTS; i++)
+        {
+            close(fd_cllock[i]);
+            fd_cllock[i] = 0;
+            close(fd_clSlock[i]);
+            fd_clSlock[i] = 0;
+            held_clSlock[i] = 0;
+            sprintf(name2, "%s/CLlock%d",name,i);
             unlink(name2);
-        sprintf(name2, "%s/CLlock",name);
+            sprintf(name2, "%s/CLSlock%d",name,i);
             unlink(name2);
+        }
+        close(fd_svlock);
+        fd_svlock = 0;
+        sprintf(name2, "%s/SVlock",name);
+        unlink(name2);
+        // remove the PID folder
         rmdir(name);
         fprintf(stderr,"dfhack: destroyed shared segment.\n");
         inited = false;
@@ -213,7 +259,7 @@ DFhackCExport void SDL_GL_SwapBuffers(void)
 {
     if(_SDL_GL_SwapBuffers)
     {
-        if(!errorstate && ((shm_cmd *)shm)->pingpong != CORE_RUNNING)
+        if(!errorstate)
         {
             SHM_Act();
         }
@@ -227,7 +273,7 @@ DFhackCExport int SDL_Flip(void * some_ptr)
 {
     if(_SDL_Flip)
     {
-        if(!errorstate && ((shm_cmd *)shm)->pingpong != CORE_RUNNING)
+        if(!errorstate)
         {
             SHM_Act();
         }
@@ -285,7 +331,7 @@ DFhackCExport int refresh (void)
 {
     if(_refresh)
     {
-        if(!errorstate && ((shm_cmd *)shm)->pingpong != CORE_RUNNING)
+        if(!errorstate)
         {
             SHM_Act();
         }

shmserver/shms-windows.cpp

@@ -44,8 +44,65 @@ char *shm = 0;
 int shmid = 0;
 bool inited = 0;
 HANDLE shmHandle = 0;
+
 HANDLE DFSVMutex = 0;
-HANDLE DFCLMutex = 0;
+HANDLE DFCLMutex[SHM_MAX_CLIENTS];
+HANDLE DFCLSuspendMutex[SHM_MAX_CLIENTS];
+int held_DFCLSuspendMutex[SHM_MAX_CLIENTS];
+int numheld = SHM_MAX_CLIENTS;
+
+
+void OS_lockSuspendLock(int which)
+{
+    if(numheld == SHM_MAX_CLIENTS)
+        return;
+    // lock not held by server and can be picked up. OK.
+    if(held_DFCLSuspendMutex[which] == 0)
+    {
+        uint32_t state = WaitForSingleObject(DFCLSuspendMutex[which],INFINITE);
+        if(state == WAIT_ABANDONED || state == WAIT_OBJECT_0)
+        {
+            held_DFCLSuspendMutex[which] = 1;
+            numheld++;
+            return;
+        }
+        // lock couldn't be picked up!
+        errorstate = 1;
+        MessageBox(0,"Suspend lock locking failed. Further communication disabled!","Error", MB_OK);
+        return;
+    }
+    errorstate = 1;
+    MessageBox(0,"Server tried to lock already locked suspend lock?  Further communication disabled!","Error", MB_OK);
+    return;
+}
+
+void OS_releaseSuspendLock(int which)
+{
+    /*
+    if(which >=0 && which < SHM_MAX_CLIENTS)
+        return;
+    */
+    if(numheld != SHM_MAX_CLIENTS)
+    {
+        MessageBox(0,"Locking system failure. Further communication disabled!","Error", MB_OK);
+        errorstate = 1;
+        return;
+    }
+    // lock hel by server and can be released -> OK
+    if(held_DFCLSuspendMutex[which] == 1 && ReleaseMutex(DFCLSuspendMutex[which]))
+    {
+        numheld--;
+        held_DFCLSuspendMutex[which] = 0;
+    }
+    // locked and not can't be released? FAIL!
+    else if (held_DFCLSuspendMutex[which] == 1)
+    {
+        MessageBox(0,"Suspend lock failed to unlock.  Further communication disabled!","Error", MB_OK);
+        return;
+    }
+}
+
+
 void SHM_Init ( void )
 {
     // check that we do this only once per process
@@ -56,62 +113,44 @@ void SHM_Init ( void )
     }
     inited = true;
     
-    char svmutexname [256];
-    sprintf(svmutexname,"DFSVMutex-%d",OS_getPID());
+    
     char clmutexname [256];
-    sprintf(clmutexname,"DFCLMutex-%d",OS_getPID());
+    char clsmutexname [256];
     char shmname [256];
     sprintf(shmname,"DFShm-%d",OS_getPID());    
     
-    // create or open mutexes
+    // create a locked server mutex
+    char svmutexname [256];
+    sprintf(svmutexname,"DFSVMutex-%d",OS_getPID());
     DFSVMutex = CreateMutex( 0, 1, svmutexname);
     if(DFSVMutex == 0)
     {
-        DFSVMutex = OpenMutex(SYNCHRONIZE,false, svmutexname);
-        if(DFSVMutex == 0)
-        {
+        MessageBox(0,"Server mutex creation failed. Further communication disabled!","Error", MB_OK);
         errorstate = 1;
         return;
     }
-    }
-    DFCLMutex = CreateMutex( 0, 0, clmutexname);
-    if(DFCLMutex == 0)
-    {
-        DFCLMutex = OpenMutex(SYNCHRONIZE,false, clmutexname);
-        if(DFCLMutex == 0)
+    // the mutex already existed. we don't want to know.
+    if(GetLastError() == ERROR_ALREADY_EXISTS)
     {
-            CloseHandle(DFSVMutex);
+        MessageBox(0,"Server mutex already existed. Further communication disabled!","Error", MB_OK);
         errorstate = 1;
         return;
     }
-    }
     
-    // try locking server mutex
-    uint32_t result;
-    result = WaitForSingleObject(DFSVMutex,0);
-    switch (result)
-    {
-        case WAIT_ABANDONED:
+    // create client and suspend mutexes
+    for(int i = 0; i < SHM_MAX_CLIENTS; i++)
     {
-            // picked up after a crashed DF process
-            // do some sanity checks on client attached
-            // otherwise the same thing as WAIT_OBJECT_0
-            break;
-        }
-        case WAIT_OBJECT_0:
-        {
-            // all right, we have the mutex and are the one and only DF server
-            break;
-        }
-        case WAIT_TIMEOUT:
-        case WAIT_FAILED:
-        default:
+        sprintf(clmutexname,"DFCLMutex-%d-%d",OS_getPID(),i);
+        sprintf(clsmutexname,"DFCLSuspendMutex-%d-%d",OS_getPID(),i);
+        
+        DFCLMutex[i] = CreateMutex( 0, 0, clmutexname); // client mutex, not held
+        DFCLSuspendMutex[i] = CreateMutex( 0, 1, clsmutexname); // suspend mutexes held on start
+        held_DFCLSuspendMutex[i] = 1;
+        
+        if(DFCLMutex[i] == 0 || DFCLSuspendMutex[i] == 0 || GetLastError() == ERROR_ALREADY_EXISTS)
         {
-            // error, bail
+            MessageBox(0,"Client mutex creation failed. Close all tools before starting DF.","Error", MB_OK);
             errorstate = 1;
-            MessageBox(0,"Could not aquire mutex","Error", MB_OK);
-            CloseHandle(DFSVMutex);
-            CloseHandle(DFCLMutex);
             return;
         }
     }
@@ -123,35 +162,32 @@ void SHM_Init ( void )
     {
         MessageBox(0,"SHM bridge already in use","Error", MB_OK);
         errorstate = 1;
-        ReleaseMutex(DFSVMutex);
-        CloseHandle(DFSVMutex);
-        CloseHandle(DFCLMutex);
         return;
     }
     if(!shmHandle)
     {
         MessageBox(0,"Couldn't create SHM bridge","Error", MB_OK);
         errorstate = 1;
-        ReleaseMutex(DFSVMutex);
-        CloseHandle(DFSVMutex);
-        CloseHandle(DFCLMutex);
         return;
     }
     // attempt to attach the created mapping
     shm = (char *) MapViewOfFile(shmHandle,FILE_MAP_ALL_ACCESS, 0,0, SHM_SIZE);
     if(shm)
     {
-        ((shm_cmd *)shm)->pingpong = CORE_RUNNING;
+        // make sure we don't stall or do crazy stuff
+        for(int i = 0; i < SHM_MAX_CLIENTS;i++)
+        {
+            ((uint32_t *)shm)[i] = CORE_RUNNING;
+        }
+        // init modules :)
+        InitModules();
     }
     else
     {
         MessageBox(0,"Couldn't attach SHM bridge","Error", MB_OK);
         errorstate = 1;
-        ReleaseMutex(DFSVMutex);
-        CloseHandle(DFSVMutex);
-        CloseHandle(DFCLMutex);
+        return;
     }
-    InitModules();
 }
 
 void SHM_Destroy ( void )
@@ -159,9 +195,13 @@ void SHM_Destroy ( void )
     if(errorstate)
         return;
     KillModules();
-    ReleaseMutex(DFSVMutex);
+    // get rid of all the locks
     CloseHandle(DFSVMutex);
-    CloseHandle(DFCLMutex);
+    for(int i=0; i < SHM_MAX_CLIENTS; i++)
+    {
+        CloseHandle(DFCLSuspendMutex[i]);
+        CloseHandle(DFCLMutex[i]);
+    }
 }
 
 uint32_t OS_getPID()
@@ -181,17 +221,18 @@ uint32_t OS_getAffinity()
 
 
 // is the other side still there?
-bool isValidSHM()
+bool isValidSHM(int which)
 {
-    // try if CL mutex is free
-    uint32_t result = WaitForSingleObject(DFCLMutex,0);
+    // try if CL mutex is free (by locking client mutex)
+    uint32_t result = WaitForSingleObject(DFCLMutex[which],0);
     
     switch (result)
     {
         case WAIT_ABANDONED:
         case WAIT_OBJECT_0:
         {
-            ReleaseMutex(DFCLMutex);
+            OS_lockSuspendLock(which);
+            ReleaseMutex(DFCLMutex[which]);
             return false;
         }
         case WAIT_TIMEOUT:

shmserver/shms.h

@@ -2,10 +2,12 @@
 #define DFCONNECT_H
 
 #define SHM_KEY 123466
+#define SHM_MAX_CLIENTS 4
 #define SHM_HEADER 1024 // 1kB reserved for a header
-#define SHM_BODY 1024*1024 // 1MB reserved for bulk data transfer
+#define SHM_BODY 1024*1024 // 4MB reserved for bulk data transfer
 #define SHM_SIZE SHM_HEADER+SHM_BODY
-
+//#define SHM_ALL_CLIENTS SHM_MAX_CLIENTS*(SHM_SIZE)
+//#define SHM_CL(client_idx) client_idx*(SHM_SIZE)
 
 // FIXME: add YIELD for linux, add single-core and multi-core compile targets for optimal speed
 #ifdef LINUX_BUILD
@@ -79,7 +81,7 @@ struct DFPP_module
     void * modulestate;
 };
 
-typedef union
+union shm_cmd
 {
     struct
         {
@@ -87,18 +89,20 @@ typedef union
             volatile uint16_t module;
         } parts;
     volatile uint32_t pingpong;
-    inline void set(uint16_t module, uint16_t command)
+    shm_cmd(volatile uint32_t z)
     {
-        pingpong = module + command << 16;
+        pingpong = z;
     }
-} shm_cmd;
+};
 
 void SHM_Act (void);
 void InitModules (void);
 void KillModules (void);
-bool isValidSHM();
+bool isValidSHM(int current);
 uint32_t OS_getPID();
 DFPP_module InitMaps(void);
 uint32_t OS_getAffinity(); // limited to 32 processors. Silly, eh?
+void OS_releaseSuspendLock(int currentClient);
+void OS_lockSuspendLock(int currentClient);
 
 #endif