diff --git a/kernel.c b/kernel.c
index 017bc81..1bbaaae 100644
--- a/kernel.c
+++ b/kernel.c
@@ -54,7 +54,10 @@ void read_binary_table()
 
         // create a new process for each binary found
         // it should have around 4kb stack
-        create_new_process(binary_table+i, 1<<12);
+        optional_pcbptr res = create_new_process(binary_table+i, 1<<12);
+        if (has_error(res)) {
+            dbgln("Error creating initial process!", 31);
+        }
     }
 }
 
diff --git a/kernel.h b/kernel.h
index 3c4a2cd..0fcb60e 100644
--- a/kernel.h
+++ b/kernel.h
@@ -4,12 +4,11 @@
 #define true 1
 #define false 0
 
-#define XLEN          32            // 32 bit system
-#define PROCESS_COUNT 32            // number of concurrent processes
-#define NUM_BINARIES  16            // number of binaries loaded simultaneously 
+#define PROCESS_COUNT 8            // number of concurrent processes
+#define NUM_BINARIES  4            // number of binaries loaded simultaneously 
 
 // scheduler settings
-#define TIME_SLICE_LEN 100          // number of cpu time ticks per slice
+#define TIME_SLICE_LEN 10          // number of cpu time ticks per slice
 
 // init function
 extern __attribute__((__noreturn__)) void init();
diff --git a/kinclude/boot.S b/kinclude/boot.S
index 5ed7767..4f2be0d 100644
--- a/kinclude/boot.S
+++ b/kinclude/boot.S
@@ -45,7 +45,8 @@ _start:
             jal     init
             
             // halt machine after returning from init
-            csrwi   CSR_HALT, 1
+            li      t0, -1
+            csrw    CSR_HALT, t0
 1:
             j       1b
 
@@ -162,4 +163,13 @@ memset:
             addi    a1, a1, 32
             blt     a1, a2, 1b
             ret
-#endif
\ No newline at end of file
+#endif
+
+// this is where instantiated threads return to once they are finished
+.section    .thread_fini
+.global     thread_finalizer
+thread_finalizer:
+// just a simple exit syscall
+            nop
+            li      a7, 5
+            ecall
\ No newline at end of file
diff --git a/kinclude/ecall.c b/kinclude/ecall.c
index d01ec4f..4c1b91d 100644
--- a/kinclude/ecall.c
+++ b/kinclude/ecall.c
@@ -16,11 +16,16 @@ ecall_handler ecall_table[ECALL_TABLE_LEN] = { 0 };
 
 int ecall_handle_spawn_thread(int* args_ptr, ProcessControlBlock* pcb)
 {
-    void* entry = (void*) args_ptr[0];
-    void* args  = (void*) args_ptr[1];
+    void* entry = (void*) args_ptr[0];  // a0
+    void* args  = (void*) args_ptr[1];  // a1
+    //int   flags = args_ptr[2];          // a2
 
+    optional_pcbptr pcb_or_err = create_new_thread(pcb, entry, args, 1<<14);
+
+    if (has_error(pcb_or_err))
+        return pcb_or_err.error;
+    
     return EINVAL;
-// void* entry, void* args
 }
 
 int ecall_handle_sleep(int* args, ProcessControlBlock* pcb)
@@ -38,13 +43,32 @@ int ecall_handle_sleep(int* args, ProcessControlBlock* pcb)
 
 int ecall_handle_join(int* args, ProcessControlBlock* pcb)
 {
-    return EINVAL;
+    int pid = args[0];  // a0
+
+    ProcessControlBlock* target = process_from_pid(pid);
+
+    if (target == NULL)
+        return ESRCH;
+    
+    if (target->status == PROC_DEAD)
+        return target->exit_code;
+    
+    pcb->status = PROC_WAIT_PROC;
+    pcb->waiting_for_process = target;
+
+    int timeout = args[1];
+    if (timeout <= 0)
+        return 0;
+
+    unsigned int len = (unsigned int) timeout;
+    pcb->asleep_until = read_time() + len;
+    
+    return 0;
 }
 
 int ecall_handle_kill(int* args, ProcessControlBlock* pcb)
 {
     return EINVAL;
-
 }
 
 int ecall_handle_exit(int* args, ProcessControlBlock* pcb)
@@ -52,6 +76,8 @@ int ecall_handle_exit(int* args, ProcessControlBlock* pcb)
     pcb->status = PROC_DEAD;
     pcb->exit_code = *args;
 
+    dbgln("exit", 4);
+
     char msg[34] = "process    exited with code   ";
 
     itoa(pcb->pid, &msg[8], 10);
@@ -59,28 +85,36 @@ int ecall_handle_exit(int* args, ProcessControlBlock* pcb)
 
     dbgln(msg, 34);
 
+    // recursively kill all child processes
+    kill_child_processes(pcb);
+
     return 0;
 }
 
 #pragma GCC diagnostic pop
 
-void trap_handle_ecall() {
-    {
-        
-        mark_ecall_entry();
-    };
+static void print_num(int num) {
+    char buff[16];
+    char* end = itoa(num, buff, 10);
+    dbgln(buff, (int)(end - buff));
+}
+
+void trap_handle_ecall() { 
+    mark_ecall_entry();
     ProcessControlBlock* pcb = get_current_process();
     int *regs = pcb->regs;
-    int code = regs[16];    // code is inside a7
+    int code = regs[REG_A0 + 7];    // code is inside a7
+
+    dbgln("ecall:", 6);
+    print_num(code);
 
     // check if the code is too large/small or if the handler is zero
     if (code < 0 || code > ECALL_TABLE_LEN || ecall_table[code] == 0) {
-        regs[9] = ENOCODE;
-        __asm__("ebreak");
+        regs[REG_A0] = ENOCODE;
     } else {
         // get the corresponding ecall handler
         ecall_handler handler = ecall_table[code];
-        regs[9] = handler(&regs[9], pcb);
+        regs[REG_A0] = handler(&regs[REG_A0], pcb);
     }
 
     // increment pc of this process
@@ -99,8 +133,8 @@ void trap_handle(int interrupt_bit, int code, int mtval)
             case 5:
             case 6:
             case 7:
-              scheduler_run_next();
-              break;
+                scheduler_run_next();
+                break;
             default:
               // impossible
               HALT(12);
@@ -131,7 +165,7 @@ void trap_handle(int interrupt_bit, int code, int mtval)
                 HALT(13);
         }
     }
-    HALT(1);
+    HALT(14);
     __builtin_unreachable();
 }
 
@@ -146,6 +180,6 @@ void init_ecall_table()
 
 void handle_exception(int ecode, int mtval) 
 {
-    dbgln("Trap encountered!", 17);
+    dbgln("exception encountered!", 17);
     __asm__("ebreak");
 }
diff --git a/kinclude/ecall.h b/kinclude/ecall.h
index 51c1aeb..e5e03a2 100644
--- a/kinclude/ecall.h
+++ b/kinclude/ecall.h
@@ -15,7 +15,7 @@ enum ecall_codes {
     ECALL_EXIT  = 5,
 };
 
-#define ECALL_TABLE_LEN 16
+#define ECALL_TABLE_LEN 8
 
 // initializer for ecall lookup table
 void init_ecall_table();
diff --git a/kinclude/io.c b/kinclude/io.c
index 32f2fd2..021d1eb 100644
--- a/kinclude/io.c
+++ b/kinclude/io.c
@@ -44,13 +44,24 @@ char* itoa (int value, char* str, int base)
         value *= -1;
     }
 
-    int num;
+    int digits = 0;
+    int num = 0;
+    // reverse number
+    do {
+        num = num * base;
+        num += value % base;
+        value = value / base;
+        digits++;
+    } while (value > 0);
+
+    value = num;
     do {
         num = value % base;
         value = value / base;
         *str++ = alpha[num];
+        digits--;
     }
-    while (value > 0);
+    while (digits > 0);
 
     return str;
 }
diff --git a/kinclude/ktypes.h b/kinclude/ktypes.h
index 54302f3..0a065e5 100644
--- a/kinclude/ktypes.h
+++ b/kinclude/ktypes.h
@@ -1,15 +1,18 @@
 #ifndef H_ktypes
 #define H_ktypes
 
+#define NULL ((void*) 0)
+
 /*
  *  Error codes
  */
 
 enum error_code {
-    ENOCODE = 1,   // invalid syscall code
-    EINVAL  = 2,   // invalid argument value
-    ENOMEM  = 3,   // not enough memory
-    ENOBUFS = 4,   // no space left in buffer
+    ENOCODE = 1,    // invalid syscall code
+    EINVAL  = 2,    // invalid argument value
+    ENOMEM  = 3,    // not enough memory
+    ENOBUFS = 4,    // no space left in buffer
+    ESRCH   = 5,    // no such process
 };
 
 /*
@@ -37,6 +40,24 @@ struct ProcessControlBlock {
     ProcessControlBlock* waiting_for_process;
     struct loaded_binary* binary;
     unsigned long long int asleep_until;
+    // parent
+    ProcessControlBlock* parent;
+};
+
+enum pcb_struct_registers {
+    REG_RA = 0,
+    REG_SP = 1,
+    REG_GP = 2,
+    REG_TP = 3,
+    REG_T0 = 4,
+    REG_T1 = 5,
+    REG_t2 = 6,
+    REG_FP = 7,
+    REG_S0 = 7,
+    REG_S1 = 8,
+    REG_A0 = 9,
+    REG_S2 = 17,
+    REG_T3 = 27
 };
 
 
diff --git a/kinclude/sched.c b/kinclude/sched.c
index 9dde2f8..bf22623 100644
--- a/kinclude/sched.c
+++ b/kinclude/sched.c
@@ -4,10 +4,37 @@
 #include "io.h"
 #include "malloc.h"
 
+// use memset provided in boot.S
+extern void memset(int, void*, void*);
+// this is the address where threads return to
+extern int thread_finalizer;
+
+static void print_num(int num) {
+    char buff[16];
+    char* end = itoa(num, buff, 10);
+    dbgln(buff, (int)(end - buff));
+}
+
+static void print_processes() {
+    char line[12] = "============";
+    char alive[5] = "alive";
+    char dead[4] = "dead";
+    for (int i = 0; i < PROCESS_COUNT; i++) {
+        ProcessControlBlock* pcb = processes + i;
+        print_num(pcb->pid);
+        if (pcb->status == PROC_DEAD) {
+            dbgln(dead, 4);
+        } else {
+            dbgln(alive, 5);
+        }
+        print_num(i);
+        dbgln(line, 12);
+    }
+}
 
 // scheduling data:
 ProcessControlBlock processes[PROCESS_COUNT];
-ProcessControlBlock* current_process;
+ProcessControlBlock* current_process = NULL;
 unsigned long long int scheduling_interrupted_start;
 unsigned long long int next_interrupt_scheduled_for;
 int next_process_id = 1;
@@ -15,10 +42,6 @@ int next_process_id = 1;
 void scheduler_run_next ()
 {
     current_process = scheduler_select_free();
-    char msg[30] = "scheduling ";
-    char* end = itoa(current_process->pid, &msg[11], 10);
-    dbgln(msg, ((int) end) - ((int) msg));
-
     // set up timer interrupt
     set_next_interrupt();
     scheduler_switch_to(current_process);
@@ -29,6 +52,7 @@ void scheduler_try_return_to(ProcessControlBlock* pcb)
     if (pcb->status != PROC_RDY) {
         scheduler_run_next();
     } else {
+        dbgln("returning to process...", 23);
         current_process = pcb;
         // add time spent in ecall handler to the processes time slice
         next_interrupt_scheduled_for = next_interrupt_scheduled_for + (read_time() - scheduling_interrupted_start);
@@ -43,12 +67,18 @@ ProcessControlBlock* scheduler_select_free()
     int i;
     int timeout_available = false; // note if a timeout is available
 
+    if (current_process == NULL)
+        current_process = processes + PROCESS_COUNT - 1;
+
     while (true) {
         mtime = read_time();
+        int i = 1;
+        ProcessControlBlock* pcb = current_process + 1;
+        if (pcb > processes + PROCESS_COUNT)
+            pcb = processes;
 
-        for (i=0; i < PROCESS_COUNT; i++) {
-            ProcessControlBlock* pcb = processes + i;
-            if (pcb->status == PROC_RDY && pcb != current_process)
+        while (pcb != current_process) {
+            if (pcb->status == PROC_RDY)
                 return pcb;
             
             if (pcb->status == PROC_WAIT_SLEEP) {
@@ -68,7 +98,11 @@ ProcessControlBlock* scheduler_select_free()
                     timeout_available = true;
                 }
             }
+            pcb++;
+            if (pcb > processes + PROCESS_COUNT)
+                pcb = processes;
         }
+
         if (current_process->status == PROC_RDY) {
             return current_process;
         }
@@ -127,13 +161,13 @@ void scheduler_switch_to(ProcessControlBlock* pcb)
     __builtin_unreachable();
 }
 
-int  scheduler_index_from_pid(int pid)
+ProcessControlBlock* process_from_pid(int pid)
 {
     for (int i = 0; i < PROCESS_COUNT; i++) {
         if (processes[i].pid == pid)
-            return i;
+            return processes + i;
     }
-    return -1;
+    return NULL;
 }
 
 int* get_current_process_registers()
@@ -168,17 +202,19 @@ optional_pcbptr find_available_pcb_slot() {
             return (optional_pcbptr) { .error = ENOBUFS };
         pcb = processes + index;
     }
+    index++;
+
     return (optional_pcbptr) { .value = pcb };
 }
 
-int create_new_process(loaded_binary* bin, int stack_size)
+optional_pcbptr create_new_process(loaded_binary* bin, int stack_size)
 {
     // try to get a position in the processes list
     optional_pcbptr slot_or_err = find_available_pcb_slot();
     // if that failed, we cannot creat a new process
     if (has_error(slot_or_err)) {
         dbgln("No more process structs!", 24);
-        return slot_or_err.error;
+        return slot_or_err;
     }
 
     // allocate stack for the new process
@@ -186,7 +222,7 @@ int create_new_process(loaded_binary* bin, int stack_size)
     // if that failed, we also can't create a new process
     if (has_error(stack_top_or_err)) {
         dbgln("Error while allocating stack for process", 40);
-        return stack_top_or_err.error;
+        return (optional_pcbptr) { .error = stack_top_or_err.error };
     }
 
     ProcessControlBlock* pcb = slot_or_err.value;
@@ -199,11 +235,75 @@ int create_new_process(loaded_binary* bin, int stack_size)
     pcb->pid = pid;
     pcb->pc = bin->entrypoint;
     pcb->binary = bin;
+    pcb->parent = NULL;
+    pcb->asleep_until = 0;
+    // zero out registers
+    memset(0, pcb->regs, pcb->regs + 31);
     // load stack top into stack pointer register
-    pcb->regs[1] = (int) stack_top_or_err.value;
+    pcb->regs[REG_SP] = (int) stack_top_or_err.value;
     // load pid into a0 register
-    pcb->regs[9] = pid;
+    pcb->regs[REG_A0] = pid;
 
     dbgln("Created new process!", 20);
 
+    return (optional_pcbptr) { .value = pcb };
+}
+
+optional_pcbptr create_new_thread(ProcessControlBlock* parent, void* entrypoint, void* args, int stack_size)
+{
+    // try to get a position in the processes list
+    optional_pcbptr slot_or_err = find_available_pcb_slot();
+    // if that failed, we cannot creat a new process
+    if (has_error(slot_or_err)) {
+        dbgln("No more process structs!", 24);
+        return slot_or_err;
+    }
+
+    // allocate stack for the new process
+    optional_voidptr stack_top_or_err = malloc_stack(stack_size); // allocate 4Kib stack
+    // if that failed, we also can't create a new process
+    if (has_error(stack_top_or_err)) {
+        dbgln("Error while allocating stack for thread", 39);
+        return (optional_pcbptr) { .error = stack_top_or_err.error };
+    }
+
+    ProcessControlBlock* pcb = slot_or_err.value;
+
+    // determine next pid
+    int pid = next_process_id++;
+
+    // mark process as ready
+    pcb->status = PROC_RDY;
+    pcb->pid = pid;
+    pcb->pc = (int) entrypoint;
+    pcb->binary = parent->binary;
+    pcb->parent = parent;
+    pcb->asleep_until = 0;
+    // zero out registers
+    memset(0, pcb->regs, pcb->regs + 31);
+    // set return address to global thread finalizer
+    pcb->regs[REG_RA] = (int) &thread_finalizer;
+    // load stack top into stack pointer register
+    pcb->regs[REG_SP] = (int) stack_top_or_err.value;
+    // copy global pointer from parent
+    pcb->regs[REG_GP] = parent->regs[REG_GP];
+    // load args pointer into a0 register
+    pcb->regs[REG_A0] = (int) args;
+
+    dbgln("Created new thread!", 19);
+
+    return (optional_pcbptr) { .value = pcb };
+}
+
+void kill_child_processes(ProcessControlBlock* pcb)
+{
+    for (int i = 0; i < PROCESS_COUNT; i++) {
+        ProcessControlBlock* proc = processes + i;
+        if (proc->parent != pcb) 
+            continue;
+
+        proc->status = PROC_DEAD;
+        proc->exit_code = -9;   // set arbitrary exit code
+        kill_child_processes(proc);
+    }
 }
diff --git a/kinclude/sched.h b/kinclude/sched.h
index a4937fe..50366ab 100644
--- a/kinclude/sched.h
+++ b/kinclude/sched.h
@@ -9,16 +9,16 @@ extern ProcessControlBlock processes[PROCESS_COUNT];
 
 // scheduler methods
 ProcessControlBlock* scheduler_select_free();
-int  scheduler_create_process(int binid);
 void set_next_interrupt();
 void __attribute__((noreturn)) scheduler_run_next();
 void __attribute__((noreturn)) scheduler_try_return_to(ProcessControlBlock*);
 void __attribute__((noreturn)) scheduler_switch_to(ProcessControlBlock*);
-int  scheduler_index_from_pid(int pid);
+ProcessControlBlock* process_from_pid(int pid);
 int* get_current_process_registers();
 ProcessControlBlock* get_current_process();
 void mark_ecall_entry();
 
-int create_new_process(loaded_binary* bin, int stack_size);
-
+optional_pcbptr create_new_process(loaded_binary*, int);
+optional_pcbptr create_new_thread(ProcessControlBlock*, void*, void*, int);
+void kill_child_processes(ProcessControlBlock* pcb);
 #endif
\ No newline at end of file
diff --git a/linker.ld b/linker.ld
index acd54b5..cfb9805 100644
--- a/linker.ld
+++ b/linker.ld
@@ -237,4 +237,9 @@ SECTIONS
   /* End of uninitialized data segment (used by syscalls.c for heap) */
   PROVIDE( end = . );
   _end = ALIGN(8);
+
+  .thread_fini :
+  {
+    *(.thread_fini)
+  }
 }
diff --git a/package.py b/package.py
index 7b57bc5..331e709 100644
--- a/package.py
+++ b/package.py
@@ -14,7 +14,7 @@ INCLUDE_THESE_SECTIONS = set((
     '.text', '.stack', '.bss', '.sdata', '.rdata', '.rodata'
     '.sbss', '.data', '.stack', '.init', 
     '.fini', '.preinit_array', '.init_array', 
-    '.fini_array', '.rodata'
+    '.fini_array', '.rodata', '.thread_fini'
 ))
 # these sections are empty, so we don't want to read the elf here
 EMPTY_SECTIONS = set((