diff --git a/firmware/buflib.c b/firmware/buflib.c
index f6a565715d..294d2926d3 100644
--- a/firmware/buflib.c
+++ b/firmware/buflib.c
@@ -898,6 +898,15 @@ const char* buflib_get_name(struct buflib_context *ctx, int handle)
 }
 
 #ifdef DEBUG
+
+void *buflib_get_data(struct buflib_context *ctx, int handle)
+{
+    if (handle <= 0)
+        buflib_panic(ctx, "invalid handle access: %d", handle);
+
+    return (void*)(ctx->handle_table[-handle].alloc);
+}
+
 void buflib_check_valid(struct buflib_context *ctx)
 {
     union buflib_data *crc_slot;
diff --git a/firmware/export/config/librockplay.h b/firmware/export/config/librockplay.h
new file mode 100644
index 0000000000..e69de29bb2
diff --git a/firmware/include/buflib.h b/firmware/include/buflib.h
index 171ab5bcd7..30484431f8 100644
--- a/firmware/include/buflib.h
+++ b/firmware/include/buflib.h
@@ -237,10 +237,14 @@ int buflib_alloc_maximum(struct buflib_context* ctx, const char* name,
  *
  * Returns: The start pointer of the allocation
  */
-static inline void* buflib_get_data(struct buflib_context *context, int handle)
+#ifdef DEBUG
+void* buflib_get_data(struct buflib_context *ctx, int handle);
+#else
+static inline void* buflib_get_data(struct buflib_context *ctx, int handle)
 {
-    return (void*)(context->handle_table[-handle].alloc);
+    return (void*)(ctx->handle_table[-handle].alloc);
 }
+#endif
 
 /**
  * Shrink the memory allocation associated with the given handle
diff --git a/firmware/kernel/pthread/corelock.c b/firmware/kernel/pthread/corelock.c
new file mode 100644
index 0000000000..10b4329639
--- /dev/null
+++ b/firmware/kernel/pthread/corelock.c
@@ -0,0 +1,18 @@
+#include <pthread.h>
+#include "kernel.h"
+
+void corelock_init(struct corelock *lk)
+{
+    lk->mutex = (pthread_mutex_t) PTHREAD_MUTEX_INITIALIZER;
+}
+
+void corelock_lock(struct corelock *lk)
+{
+    pthread_mutex_lock(&lk->mutex);
+}
+
+
+void corelock_unlock(struct corelock *lk)
+{
+    pthread_mutex_unlock(&lk->mutex);
+}
diff --git a/firmware/kernel/pthread/mutex.c b/firmware/kernel/pthread/mutex.c
new file mode 100644
index 0000000000..49503b5d82
--- /dev/null
+++ b/firmware/kernel/pthread/mutex.c
@@ -0,0 +1,21 @@
+#include <pthread.h>
+#include "kernel.h"
+
+void mutex_init(struct mutex *m)
+{
+    pthread_mutexattr_t attr;
+    pthread_mutexattr_init(&attr);
+    pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE);
+    pthread_mutex_init(&m->mutex, &attr);
+    pthread_mutexattr_destroy(&attr);
+}
+
+void mutex_lock(struct mutex *m)
+{
+    pthread_mutex_lock(&m->mutex);
+}
+
+void mutex_unlock(struct mutex *m)
+{
+    pthread_mutex_unlock(&m->mutex);
+}
diff --git a/firmware/kernel/pthread/thread.c b/firmware/kernel/pthread/thread.c
new file mode 100644
index 0000000000..a80ce876e8
--- /dev/null
+++ b/firmware/kernel/pthread/thread.c
@@ -0,0 +1,272 @@
+#include <stdlib.h>
+#include <stdbool.h>
+#include <errno.h>
+#include <pthread.h>
+#include "/usr/include/semaphore.h"
+#include "kernel.h"
+#include "thread.h"
+
+#define NSEC_PER_SEC 1000000000L
+static inline void timespec_add_ns(struct timespec *a, uint64_t ns)
+{
+    lldiv_t q = lldiv(a->tv_nsec + ns, NSEC_PER_SEC);
+    a->tv_sec += q.quot;
+    a->tv_nsec = q.rem;
+}
+
+static int threads_initialized;
+
+struct thread_init_data {
+    void (*function)(void);
+    bool start_frozen;
+    sem_t init_sem;
+    struct thread_entry *entry;
+};
+
+__thread struct thread_entry *_current;
+
+struct thread_entry* thread_self_entry(void)
+{
+    return _current;
+}
+
+unsigned int thread_self(void)
+{
+    return (unsigned) pthread_self();
+}
+
+static struct thread_entry_item {
+    unsigned thread_id;
+    struct thread_entry *entry;
+} entry_lookup[32];
+
+
+
+static struct thread_entry_item *__find_thread_entry(unsigned thread_id)
+{
+    int i;
+    
+    for (i = 0; i < 32; i++)
+    {
+        if (entry_lookup[i].thread_id == thread_id)
+            return &entry_lookup[i];
+    }
+    return NULL;
+}
+
+static struct thread_entry *find_thread_entry(unsigned thread_id)
+{
+    return __find_thread_entry(thread_id)->entry;
+}
+
+static void *trampoline(void *arg)
+{
+    struct thread_init_data *data = arg;
+
+    void (*thread_fn)(void) = data->function;
+
+    _current = data->entry;
+
+    if (data->start_frozen)
+    {
+        struct corelock thaw_lock;
+        struct thread_entry *queue = NULL;
+        corelock_init(&thaw_lock);
+        corelock_lock(&thaw_lock);
+
+        _current->lock = &thaw_lock;
+        _current->bqp = &queue;
+        sem_post(&data->init_sem);
+        block_thread_switch(_current, _current->lock);
+        _current->lock = NULL;
+
+        corelock_unlock(&thaw_lock);
+    }
+    else
+        sem_post(&data->init_sem);
+
+    free(data);
+    thread_fn();
+
+    return NULL;
+}
+
+void thread_thaw(unsigned int thread_id)
+{
+    struct thread_entry *e = find_thread_entry(thread_id);
+    if (e->lock)
+    {
+        corelock_lock(e->lock);
+        wakeup_thread(e->bqp);
+        corelock_unlock(e->lock);
+    }
+    /* else: no lock. must be running already */
+}
+
+void init_threads(void)
+{
+    struct thread_entry_item *item0 = &entry_lookup[0];
+    item0->entry = calloc(1, sizeof(struct thread_entry));
+    item0->thread_id = pthread_self();
+
+    _current = item0->entry;
+    pthread_cond_init(&item0->entry->cond, NULL);
+    threads_initialized = 1;
+}
+
+
+unsigned int create_thread(void (*function)(void),
+                           void* stack, size_t stack_size,
+                           unsigned flags, const char *name
+                           //~ IF_PRIO(, int priority)
+                           IF_COP(, unsigned int core))
+{
+    pthread_t retval;
+
+    struct thread_init_data *data = calloc(1, sizeof(struct thread_init_data));
+    struct thread_entry *entry = calloc(1, sizeof(struct thread_entry));
+    struct thread_entry_item *item;
+
+    if (!threads_initialized)
+        abort();
+
+    data->function = function;
+    data->start_frozen = flags & CREATE_THREAD_FROZEN;
+    data->entry = entry;
+    pthread_cond_init(&entry->cond, NULL);         
+    entry->runnable = true;
+    entry->l = (struct thread_list) { NULL, NULL };
+    sem_init(&data->init_sem, 0, 0);
+
+    if (pthread_create(&retval, NULL, trampoline, data) < 0)
+        return -1;
+
+    sem_wait(&data->init_sem);
+
+    item = __find_thread_entry(0);
+    item->thread_id = retval;
+    item->entry = entry;
+
+    pthread_setname_np(retval, name);
+    
+
+    return retval;
+}
+
+static void add_to_list_l(struct thread_entry **list,
+                          struct thread_entry *thread)
+{
+    if (*list == NULL)
+    {
+        /* Insert into unoccupied list */
+        thread->l.next = thread;
+        thread->l.prev = thread;
+        *list = thread;
+    }
+    else
+    {
+        /* Insert last */
+        thread->l.next = *list;
+        thread->l.prev = (*list)->l.prev;
+        thread->l.prev->l.next = thread;
+        (*list)->l.prev = thread;
+    }
+}
+
+static void remove_from_list_l(struct thread_entry **list,
+                               struct thread_entry *thread)
+{
+    if (thread == thread->l.next)
+    {
+        /* The only item */
+        *list = NULL;
+        return;
+    }
+
+    if (thread == *list)
+    {
+        /* List becomes next item */
+        *list = thread->l.next;
+    }
+
+    /* Fix links to jump over the removed entry. */
+    thread->l.prev->l.next = thread->l.next;
+    thread->l.next->l.prev = thread->l.prev;
+}
+
+unsigned int thread_queue_wake(struct thread_entry **list)
+{
+    unsigned int result = THREAD_NONE;
+
+    for (;;)
+    {
+        unsigned int rc = wakeup_thread(list);
+
+        if (rc == THREAD_NONE)
+            break;
+
+        result |= rc;        
+    }
+
+    return result;
+}
+
+/* for block_thread(), _w_tmp() and wakeup_thread() t->lock must point
+ * to a corelock instance, and this corelock must be held by the caller */
+void block_thread_switch(struct thread_entry *t, struct corelock *cl)
+{
+    t->runnable = false;
+    add_to_list_l(t->bqp, t);
+    while(!t->runnable)
+        pthread_cond_wait(&t->cond, &cl->mutex);
+}
+
+void block_thread_switch_w_tmo(struct thread_entry *t, int timeout, struct corelock *cl)
+{
+    int err = 0;
+    struct timespec ts;
+
+    clock_gettime(CLOCK_REALTIME, &ts);
+    timespec_add_ns(&ts, timeout * (NSEC_PER_SEC/HZ));
+
+    t->runnable = false;
+    add_to_list_l(t->bqp, t);
+    while(!t->runnable && !err)
+        err = pthread_cond_timedwait(&t->cond, &cl->mutex, &ts);
+
+    if (err == ETIMEDOUT)
+    {   /* the thread timed out and was not explicitely woken up.
+         * we need to do this now to mark it runnable again */
+        remove_from_list_l(t->bqp, t);
+        t->runnable = true;
+        if (t->wakeup_ext_cb)
+            t->wakeup_ext_cb(t);
+    }
+}
+
+unsigned int wakeup_thread(struct thread_entry **list)
+{
+    struct thread_entry *t = *list;
+    if (t)
+    {
+        remove_from_list_l(list, t);
+        t->runnable = true;
+        pthread_cond_signal(&t->cond);
+    }
+    return THREAD_NONE;
+}
+
+
+void yield(void) {}
+
+unsigned sleep(unsigned ticks)
+{
+    struct timespec ts;
+
+    ts.tv_sec = ticks/HZ;
+    ts.tv_nsec = (ticks % HZ) * (NSEC_PER_SEC/HZ);
+
+    nanosleep(&ts, NULL);
+
+    return 0;
+}