diff --git a/base-hw/src/core/cpu/arm.h b/base-hw/src/core/cpu/arm.h
index ff032073c..00f34b93f 100644
--- a/base-hw/src/core/cpu/arm.h
+++ b/base-hw/src/core/cpu/arm.h
@@ -469,16 +469,6 @@ namespace Arm
 			uint32_t cidr;          /* context ID register backup */
 			uint32_t section_table; /* base address of applied section table */
 
-			/***************
-			 ** Accessors **
-			 ***************/
-
-			void tlb(addr_t const st) { section_table = st; }
-
-			addr_t tlb() const { return section_table; }
-
-			void protection_domain(unsigned const id) { cidr = id; }
-
 			/**
 			 * Copy CPU state data to 'c'
 			 */
@@ -524,6 +514,21 @@ namespace Arm
 				lr  = s->lr;
 				ip  = s->ip;
 			}
+
+			/**
+			 * Get base of assigned translation lookaside buffer
+			 */
+			addr_t tlb() const { return section_table; }
+
+			/**
+			 * Assign translation lookaside buffer
+			 */
+			void tlb(addr_t const st) { section_table = st; }
+
+			/**
+			 * Assign protection domain
+			 */
+			void protection_domain(unsigned const id) { cidr = id; }
 		};
 
 		/**
@@ -557,6 +562,62 @@ namespace Arm
 			unsigned user_arg_6() const { return r6; }
 			unsigned user_arg_7() const { return r7; }
 
+			/**
+			 * Part of context init that is common for all types of threads
+			 */
+			void init_thread_common(void * const   instr_p,
+			                        addr_t const   tlb,
+			                        unsigned const pd_id)
+			{
+				ip = (addr_t)instr_p;
+				cidr = pd_id;
+				section_table = tlb;
+			}
+
+			/**
+			 * Init context of the first thread of core
+			 */
+			void init_core_main_thread(void * const   instr_p,
+			                           void * const   stack_p,
+			                           addr_t const   tlb,
+			                           unsigned const pd_id)
+			{
+				sp = (addr_t)stack_p;
+				init_thread_common(instr_p, tlb, pd_id);
+			}
+
+			/**
+			 * Init context of a thread that isn't first thread of a program
+			 */
+			void init_thread(void * const   instr_p,
+			                 void * const   stack_p,
+			                 addr_t const   tlb,
+			                 unsigned const pd_id)
+			{
+				sp = (addr_t)stack_p;
+				init_thread_common(instr_p, tlb, pd_id);
+			}
+
+			/**
+			 * Init context of the first thread of a program other than core
+			 */
+			void init_main_thread(void * const   instr_p,
+			                      void * const   utcb_virt,
+			                      addr_t const   tlb,
+			                      unsigned const pd_id)
+			{
+				/*
+				 * Normally threads receive their UTCB pointer through their
+				 * 'Thread_base' but the first thread of a program doesn't
+				 * have such object. Thus the kernel hands out the UTCB pointer
+				 * through the main threads initial CPU context. 'crt0.s' then
+				 * can save the received pointer to local mem before polluting
+				 * the CPU context.
+				 */
+				sp = (addr_t)utcb_virt;
+				init_thread_common(instr_p, tlb, pd_id);
+			}
+
 			/**
 			 * Check if a pagefault has occured due to a translation miss
 			 *
diff --git a/base-hw/src/core/include/platform_thread.h b/base-hw/src/core/include/platform_thread.h
index bc99b0cd5..13fcd96a0 100644
--- a/base-hw/src/core/include/platform_thread.h
+++ b/base-hw/src/core/include/platform_thread.h
@@ -178,7 +178,7 @@ namespace Genode {
 
 			Thread_base * thread_base()
 			{
-				if (!_thread_base) assert(_main_thread);
+				if (!_thread_base) assert(main_thread());
 				return _thread_base;
 			}
 
@@ -188,6 +188,8 @@ namespace Genode {
 
 			Ram_dataspace_capability utcb() const { return _utcb; }
 
+			bool main_thread() const { return _main_thread; }
+
 			Tlb * tlb() const { return _tlb; }
 	};
 }
diff --git a/base-hw/src/core/kernel.cc b/base-hw/src/core/kernel.cc
index fa55abcc9..af8f3b243 100644
--- a/base-hw/src/core/kernel.cc
+++ b/base-hw/src/core/kernel.cc
@@ -766,7 +766,15 @@ namespace Kernel
 			/* initialize idle thread */
 			void * sp;
 			sp = (void *)&idle_stack[sizeof(idle_stack)/sizeof(idle_stack[0])];
-			idle.init_context((void *)&idle_main, sp, core_id());
+
+			/*
+			 * Idle doesn't use its UTCB pointer, thus
+			 * utcb_phys = utcb_virt = 0 is save.
+			 * Base-hw doesn't support multiple cores, thus
+			 * cpu_no = 0 is ok. We don't use 'start' to avoid
+			 * recursive call of'cpu_scheduler'.
+			 */
+			idle.prepare_to_start((void *)&idle_main, sp, 0, core_id(), 0, 0);
 			initial = 0;
 		}
 		/* create scheduler with a permanent idle thread */
@@ -929,8 +937,7 @@ namespace Kernel
 		assert(t);
 
 		/* start thread */
-		assert(!t->start(ip, sp, cpu, pt->pd_id(),
-		                 pt->phys_utcb(), pt->virt_utcb()))
+		t->start(ip, sp, cpu, pt->pd_id(), pt->phys_utcb(), pt->virt_utcb());
 
 		/* return software TLB that the thread is assigned to */
 		Pd::Pool * const pp = Pd::pool();
@@ -1529,42 +1536,49 @@ int Kernel::Thread::resume()
 }
 
 
-int Thread::start(void *ip, void *sp, unsigned cpu_no,
-                         unsigned const pd_id,
-                         Native_utcb * const phys_utcb,
-                         Native_utcb * const virt_utcb)
+void Thread::prepare_to_start(void * const        ip,
+                              void * const        sp,
+                              unsigned const      cpu_id,
+                              unsigned const      pd_id,
+                              Native_utcb * const utcb_phys,
+                              Native_utcb * const utcb_virt)
 {
 	/* check state and arguments */
 	assert(_state == AWAIT_START)
-	assert(!cpu_no);
+	assert(!cpu_id);
 
-	/* apply thread configuration */
-	init_context(ip, sp, pd_id);
-	_phys_utcb = phys_utcb;
-	_virt_utcb = virt_utcb;
+	/* store thread parameters */
+	_phys_utcb = utcb_phys;
+	_virt_utcb = utcb_virt;
+	_pd_id     = pd_id;
 
-	/* offer thread-entry arguments */
-	user_arg_0((unsigned)_virt_utcb);
+	/* join a protection domain */
+	Pd * const pd = Pd::pool()->object(_pd_id);
+	assert(pd)
+	addr_t const tlb = pd->tlb()->base();
 
-	/* start thread */
-	_schedule();
-	return 0;
+	/* initialize CPU context */
+	if (!_platform_thread)
+		/* this is the main thread of core */
+		User_context::init_core_main_thread(ip, sp, tlb, pd_id);
+	else if (!_platform_thread->main_thread())
+		/* this is not a main thread */
+		User_context::init_thread(ip, sp, tlb, pd_id);
+	else
+		/* this is the main thread of a program other than core */
+		User_context::init_main_thread(ip, _virt_utcb, tlb, pd_id);
 }
 
 
-void Thread::init_context(void * const instr_p, void * const stack_p,
-                                  unsigned const pd_id)
+void Thread::start(void * const        ip,
+                   void * const        sp,
+                   unsigned const      cpu_id,
+                   unsigned const      pd_id,
+                   Native_utcb * const utcb_phys,
+                   Native_utcb * const utcb_virt)
 {
-	/* basic thread state */
-	sp = (addr_t)stack_p;
-	ip = (addr_t)instr_p;
-
-	/* join a pd */
-	_pd_id = pd_id;
-	Pd * const pd = Pd::pool()->object(_pd_id);
-	assert(pd)
-	protection_domain(pd_id);
-	tlb(pd->tlb()->base());
+	prepare_to_start(ip, sp, cpu_id, pd_id, utcb_phys, utcb_virt);
+	_schedule();
 }
 
 
diff --git a/base-hw/src/core/kernel/thread.h b/base-hw/src/core/kernel/thread.h
index 3ca1a6d01..838c6f5bb 100644
--- a/base-hw/src/core/kernel/thread.h
+++ b/base-hw/src/core/kernel/thread.h
@@ -847,19 +847,39 @@ namespace Kernel
 				_phys_utcb(0), _virt_utcb(0), _signal_receiver(0)
 			{ }
 
+			/**
+			 * Prepare thread to get scheduled the first time
+			 *
+			 * \param ip         initial instruction pointer
+			 * \param sp         initial stack pointer
+			 * \param cpu_id     target cpu
+			 * \param pd_id      target protection-domain
+			 * \param utcb_phys  physical UTCB pointer
+			 * \param utcb_virt  virtual UTCB pointer
+			 */
+			void prepare_to_start(void * const        ip,
+			                      void * const        sp,
+			                      unsigned const      cpu_id,
+			                      unsigned const      pd_id,
+			                      Native_utcb * const utcb_phys,
+			                      Native_utcb * const utcb_virt);
+
 			/**
 			 * Start this thread
 			 *
-			 * \param ip      instruction pointer to start at
-			 * \param sp      stack pointer to use
-			 * \param cpu_no  target cpu
-			 *
-			 * \retval  0  successful
-			 * \retval -1  thread could not be started
+			 * \param ip         initial instruction pointer
+			 * \param sp         initial stack pointer
+			 * \param cpu_id     target cpu
+			 * \param pd_id      target protection-domain
+			 * \param utcb_phys  physical UTCB pointer
+			 * \param utcb_virt  virtual UTCB pointer
 			 */
-			int start(void *ip, void *sp, unsigned cpu_no,
-			          unsigned const pd_id, Native_utcb * const phys_utcb,
-			          Native_utcb * const virt_utcb);
+			void start(void * const        ip,
+			           void * const        sp,
+			           unsigned const      cpu_id,
+			           unsigned const      pd_id,
+			           Native_utcb * const utcb_phys,
+			           Native_utcb * const utcb_virt);
 
 			/**
 			 * Pause this thread
@@ -891,16 +911,6 @@ namespace Kernel
 			 */
 			void reply(size_t const size, bool const await_request);
 
-			/**
-			 * Initialize our execution context
-			 *
-			 * \param ip     instruction pointer
-			 * \param sp     stack pointer
-			 * \param pd_id  identifies protection domain we're assigned to
-			 */
-			void init_context(void * const ip, void * const sp,
-			                  unsigned const pd_id);
-
 			/**
 			 * Handle a pagefault that originates from this thread
 			 *
diff --git a/base-hw/src/core/platform_thread.cc b/base-hw/src/core/platform_thread.cc
index c6b44dacd..3c12c34e1 100644
--- a/base-hw/src/core/platform_thread.cc
+++ b/base-hw/src/core/platform_thread.cc
@@ -28,7 +28,7 @@ bool Platform_thread::_attaches_utcb_by_itself()
 	 * virtual context area by itself, as it is done for other threads
 	 * through a sub RM-session.
 	 */
-	return _pd_id == Kernel::core_id() || !_main_thread;
+	return _pd_id == Kernel::core_id() || !main_thread();
 }
 
 
diff --git a/base-hw/src/platform/arm/crt0.s b/base-hw/src/platform/arm/crt0.s
index 093f9b0d6..dba4a8adc 100644
--- a/base-hw/src/platform/arm/crt0.s
+++ b/base-hw/src/platform/arm/crt0.s
@@ -19,8 +19,8 @@
 	_start:
 
 		/* fetch thread-entry arguments to their destinations in BSS */
-		ldr r1, =_main_utcb
-		str r0, [r1]
+		ldr r0, =_main_utcb
+		str sp, [r0]
 
 		/* call _main routine */
 		ldr sp, =_stack_high