diff --git a/arch/x86_64/thread.c b/arch/x86_64/thread.c
index 70d696c..a1937fa 100644
--- a/arch/x86_64/thread.c
+++ b/arch/x86_64/thread.c
@@ -1,6 +1,14 @@
 #include <kernel/machine/cpu.h>
 #include <kernel/machine/thread.h>
 
+#define MAX_REG_ARGS 6
+#define REG_ARG_0    rdi
+#define REG_ARG_1    rsi
+#define REG_ARG_2    rdx
+#define REG_ARG_3    rcx
+#define REG_ARG_4    r8
+#define REG_ARG_5    r9
+
 /* this is the context information restored by ml_thread_switch.
  * since ml_thread_switch only jumps to kernel-mode, IRETQ isn't used,
  * and the extra register values needed by IRETQ aren't present. */
@@ -23,10 +31,12 @@ void ml_thread_prepare_kernel_context(uintptr_t ip, uintptr_t *sp)
 	ctx->rfl = 0x202;
 }
 
-extern void ml_thread_prepare_user_context(
+extern kern_status_t ml_thread_prepare_user_context(
 	virt_addr_t ip,
 	virt_addr_t user_sp,
-	virt_addr_t *kernel_sp)
+	virt_addr_t *kernel_sp,
+	const uintptr_t *args,
+	size_t nr_args)
 {
 	(*kernel_sp) -= sizeof(struct ml_cpu_context);
 
@@ -39,4 +49,31 @@ extern void ml_thread_prepare_user_context(
 	ctx->rflags = 0x202;
 	ctx->rdi = 0;   // arg 0
 	ctx->rsi = 0;   // arg 1
+
+	for (size_t i = 0; i < nr_args; i++) {
+		switch (i) {
+		case 0:
+			ctx->REG_ARG_0 = args[i];
+			break;
+		case 1:
+			ctx->REG_ARG_1 = args[i];
+			break;
+		case 2:
+			ctx->REG_ARG_2 = args[i];
+			break;
+		case 3:
+			ctx->REG_ARG_3 = args[i];
+			break;
+		case 4:
+			ctx->REG_ARG_4 = args[i];
+			break;
+		case 5:
+			ctx->REG_ARG_5 = args[i];
+			break;
+		default:
+			return KERN_INVALID_ARGUMENT;
+		}
+	}
+
+	return KERN_OK;
 }
diff --git a/kernel/bsp.c b/kernel/bsp.c
index df5808a..1db9a98 100644
--- a/kernel/bsp.c
+++ b/kernel/bsp.c
@@ -179,15 +179,52 @@ kern_status_t bsp_launch_async(struct bsp *bsp, struct task *task)
 		return status;
 	}
 
+	status = vm_region_map_object(
+		task->t_address_space,
+		VM_REGION_ANY_OFFSET,
+		bsp->bsp_vmo,
+		0,
+		bsp->bsp_trailer.bsp_exec_offset,
+		VM_PROT_READ | VM_PROT_USER,
+		&bsp_data_base);
+
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	status = map_executable(bsp, task, &entry);
+	if (status != KERN_OK) {
+		return status;
+	}
 #ifdef TRACE
-	vm_region_dump(task->t_address_space, 0);
+	vm_region_dump(task->t_address_space);
 #endif
 
 	sp = stack_buffer + BOOTSTRAP_STACK_SIZE;
 	tracek("bootstrap: entry=%llx, sp=%llx", entry, sp);
 
+	kern_handle_t self, self_address_space;
+	task_open_handle(task, &task->t_base, 0, &self);
+	task_open_handle(
+		task,
+		&task->t_address_space->vr_base,
+		0,
+		&self_address_space);
+
+	const uintptr_t args[] = {
+		0,                    // int argc
+		0,                    // const char ** argv
+		self,                 // kern_handle_t task
+		self_address_space,   // kern_handle_t address_space
+
+		/* this parameter is specific to the bsp bootstrap program, so
+	         * that it can access the rest of the bsp image. */
+		bsp_data_base,
+	};
+	const size_t nr_args = sizeof args / sizeof args[0];
+
 	struct thread *init_thread = task_create_thread(task);
-	thread_init_user(init_thread, entry, sp);
+	thread_init_user(init_thread, entry, sp, args, nr_args);
 	schedule_thread_on_cpu(init_thread);
 
 	return KERN_OK;
diff --git a/sched/thread.c b/sched/thread.c
index 36c73eb..d4d240e 100644
--- a/sched/thread.c
+++ b/sched/thread.c
@@ -54,7 +54,9 @@ kern_status_t thread_init_kernel(struct thread *thr, virt_addr_t ip)
 kern_status_t thread_init_user(
 	struct thread *thr,
 	virt_addr_t ip,
-	virt_addr_t sp)
+	virt_addr_t sp,
+	const uintptr_t *args,
+	size_t nr_args)
 {
 	thr->tr_id = thr->tr_parent->t_next_thread_id++;
 
@@ -84,7 +86,7 @@ kern_status_t thread_init_user(
 
 	/* this context will be used by ml_user_return to jump to userspace
 	 * with the specified instruction pointer and user stack */
-	ml_thread_prepare_user_context(ip, sp, &thr->tr_sp);
+	ml_thread_prepare_user_context(ip, sp, &thr->tr_sp, args, nr_args);
 	/* this context will be used by the scheduler and ml_thread_switch to
 	 * jump to ml_user_return in kernel mode with the thread's kernel stack.
 	 */