diff --git a/kernel-src/arch/x86-64/smp.c b/kernel-src/arch/x86-64/smp.c
index 4f7c82e4..6e1acdad 100644
--- a/kernel-src/arch/x86-64/smp.c
+++ b/kernel-src/arch/x86-64/smp.c
@@ -11,7 +11,7 @@ static volatile struct limine_smp_request smprequest = {
 };
 
 // set to 1 because of the bsp
-size_t cpusawake = 1;
+size_t arch_smp_cpusawake = 1;
 
 static void cpuwakeup(struct limine_smp_info *info) {
 	cpu_set((cpu_t *)info->extra_argument);
@@ -20,13 +20,14 @@ static void cpuwakeup(struct limine_smp_info *info) {
 	arch_idt_reload();
 	dpc_init();
 	arch_mmu_apswitch();
+	vmm_apinit();
 
 	cpu_initstate();
 	arch_apic_timerinit();
 
-	__atomic_add_fetch(&cpusawake, 1, __ATOMIC_SEQ_CST);
-	// TODO let other processors run threads and applications
-	for (;;) CPU_HALT();
+	__atomic_add_fetch(&arch_smp_cpusawake, 1, __ATOMIC_SEQ_CST);
+
+	sched_apentry();
 }
 
 // if the nosmp argument is given to the kernel, this is what the APs will jump to
@@ -34,6 +35,10 @@ static void cpuwakeuphalt(struct limine_smp_info *info) {
 	for (;;) CPU_HALT();
 }
 
+void arch_smp_sendipi(cpu_t *targcpu, isr_t *isr, int target) {
+	arch_apic_sendipi(targcpu ? targcpu->id : 0, INTERRUPT_IDTOVECTOR(isr->id), target, 0, 1);
+}
+
 void arch_smp_wakeup() {
 	struct limine_smp_response *response = smprequest.response;
 	if (response == NULL) {
@@ -64,7 +69,7 @@ void arch_smp_wakeup() {
 
 	// wait for other cpus to boot up
 	if (wakeupfn == cpuwakeup)
-		while (__atomic_load_n(&cpusawake, __ATOMIC_SEQ_CST) != response->cpu_count) asm("pause");
+		while (__atomic_load_n(&arch_smp_cpusawake, __ATOMIC_SEQ_CST) != response->cpu_count) asm("pause");
 
 	printf("smp: awoke other processors\n");
 }
diff --git a/kernel-src/include/kernel/pgcache.h b/kernel-src/include/kernel/pgcache.h
new file mode 100644
index 00000000..00fc7c76
--- /dev/null
+++ b/kernel-src/include/kernel/pgcache.h
@@ -0,0 +1,8 @@
+#ifndef _PGCACHE_H
+#define _PGCACHE_H
+
+void pgcache_init();
+int pgcache_getpage(vnode_t *node, uintmax_t offset, void **ret);
+int pgcache_setdirty(void *addr);
+
+#endif
diff --git a/kernel-src/include/kernel/scheduler.h b/kernel-src/include/kernel/scheduler.h
index 088f6ce6..d5c12a0e 100644
--- a/kernel-src/include/kernel/scheduler.h
+++ b/kernel-src/include/kernel/scheduler.h
@@ -11,6 +11,7 @@
 #define SCHED_THREAD_FLAGS_RUNNING 2
 #define SCHED_THREAD_FLAGS_SLEEP 4
 #define SCHED_THREAD_FLAGS_INTERRUPTIBLE 8
+#define SCHED_THREAD_FLAGS_PREEMPTED 16
 
 #define SCHED_PROC_STATE_NORMAL 0
 #define SCHED_PROC_STATE_ZOMBIE 1
@@ -91,5 +92,6 @@ void sched_destroyproc(proc_t *);
 void sched_destroythread(thread_t *);
 void sched_targetcpu(struct cpu_t *cpu);
 void sched_sleepus(size_t us);
+void sched_apentry();
 
 #endif
diff --git a/kernel-src/include/x86-64/arch/smp.h b/kernel-src/include/x86-64/arch/smp.h
index 8b75b9c5..904ba9cd 100644
--- a/kernel-src/include/x86-64/arch/smp.h
+++ b/kernel-src/include/x86-64/arch/smp.h
@@ -1,6 +1,14 @@
 #ifndef _SMP_H
 #define _SMP_H
 
+#define ARCH_SMP_IPI_TARGET 0
+#define ARCH_SMP_IPI_SELF 1
+#define ARCH_SMP_IPI_ALL 2
+#define ARCH_SMP_IPI_OTHERCPUS 3
+
+extern size_t arch_smp_cpusawake;
+
 void arch_smp_wakeup();
+void arch_smp_sendipi(cpu_t *targcpu, isr_t *isr, int target);
 
 #endif
diff --git a/kernel-src/lib/panic.c b/kernel-src/lib/panic.c
index 9ab03226..9072df9e 100644
--- a/kernel-src/lib/panic.c
+++ b/kernel-src/lib/panic.c
@@ -4,7 +4,7 @@
 #include <kernel/interrupt.h>
 
 __attribute__((noreturn)) void _panic(char *msg, context_t *ctx) {
-	printf("Oops.\n");
+	printf("cpu%lu: Oops.\n", _cpu()->id);
 
 	if (msg)
 		printf("%s\n", msg);
diff --git a/kernel-src/mm/vmm.c b/kernel-src/mm/vmm.c
index 188bfc27..2e3f9560 100644
--- a/kernel-src/mm/vmm.c
+++ b/kernel-src/mm/vmm.c
@@ -635,3 +635,7 @@ void vmm_init() {
 
 	printspace(&kernelspace);
 }
+
+void vmm_apinit() {
+	vmm_switchcontext(&vmm_kernelctx);
+}
diff --git a/kernel-src/sys/scheduler.c b/kernel-src/sys/scheduler.c
index f1a02744..30dfd9df 100644
--- a/kernel-src/sys/scheduler.c
+++ b/kernel-src/sys/scheduler.c
@@ -12,7 +12,7 @@
 #include <kernel/devfs.h>
 
 #define QUANTUM_US 100000
-#define SCHEDULER_STACK_SIZE 4096
+#define SCHEDULER_STACK_SIZE PAGE_SIZE
 
 static scache_t *threadcache;
 static scache_t *processcache;
@@ -151,7 +151,6 @@ static thread_t *runqueuenext(int minprio) {
 			runqueuebitmap &= ~((uint64_t)1 << i);
 	}
 
-
 	if (thread)
 		thread->flags &= ~SCHED_THREAD_FLAGS_QUEUED;
 
@@ -173,6 +172,14 @@ static __attribute__((noreturn)) void switchthread(thread_t *thread) {
 	thread->cpu = _cpu();
 	if (current)
 		current->flags &= ~SCHED_THREAD_FLAGS_RUNNING;
+
+	__assert((thread->flags & SCHED_THREAD_FLAGS_RUNNING) == 0);
+
+	if (current && current->flags & SCHED_THREAD_FLAGS_SLEEP)
+		spinlock_release(&current->sleeplock);
+
+	thread->flags |= SCHED_THREAD_FLAGS_RUNNING;
+
 	ARCH_CONTEXT_SWITCHTHREAD(thread);
 	__builtin_unreachable();
 }
@@ -258,9 +265,7 @@ static void yield(context_t *context) {
 	if (next || sleeping) {
 		ARCH_CONTEXT_THREADSAVE(thread, context);
 
-		if (sleeping)
-			spinlock_release(&thread->sleeplock);
-		else
+		if (sleeping == false)
 			runqueueinsert(thread);
 
 		if (next == NULL)
@@ -348,32 +353,44 @@ void sched_setroot(vnode_t *new) {
 	setnodeslock(&proc->root, new);
 }
 
-static void timerhook(context_t *context, dpcarg_t arg) {
-	thread_t* current = _cpu()->thread;
-
-	interrupt_set(false);
+// once a scheduler dpc gets run, the return context is set to this function using the scheduler stack
+static void dopreempt() {
+	// interrupts are disabled, the thread context is already saved
 	spinlock_acquire(&runqueuelock);
 
+	thread_t *current = _cpu()->thread;
 	thread_t *next = runqueuenext(current->priority);
 
+	current->flags &= ~SCHED_THREAD_FLAGS_PREEMPTED;
 	if (next) {
 		current->flags &= ~SCHED_THREAD_FLAGS_RUNNING;
-		ARCH_CONTEXT_THREADSAVE(current, context);
-
 		runqueueinsert(current);
-
-		ARCH_CONTEXT_THREADLOAD(next, context);
-		_cpu()->thread = next;
-		next->flags |= SCHED_THREAD_FLAGS_RUNNING;
-		next->cpu = _cpu();
-		if (next->vmmctx != current->vmmctx)
-			vmm_switchcontext(next->vmmctx);
+	} else {
+		next = current;
 	}
 
 	spinlock_release(&runqueuelock);
+	switchthread(next);
+}
+
+static void timerhook(context_t *context, dpcarg_t arg) {
+	thread_t* current = _cpu()->thread;
+	interrupt_set(false);
+
+	// no need to preempt it again
+	if (current->flags & SCHED_THREAD_FLAGS_PREEMPTED)
+		return;
+
+	current->flags |= SCHED_THREAD_FLAGS_PREEMPTED;
+	ARCH_CONTEXT_THREADSAVE(current, context);
+
+	CTX_INIT(context, false);
+	CTX_SP(context) = (uintptr_t)_cpu()->schedulerstack;
+	CTX_IP(context) = (uintptr_t)dopreempt;
 }
 
 static void cpuidlethread() {
+	sched_targetcpu(_cpu());
 	interrupt_set(true);
 	while (1) {
 		CPU_HALT();
@@ -400,13 +417,26 @@ void sched_sleepus(size_t us) {
 	sched_yield();
 }
 
+void sched_apentry() {
+	_cpu()->schedulerstack = vmm_map(NULL, SCHEDULER_STACK_SIZE, VMM_FLAGS_ALLOCATE, ARCH_MMU_FLAGS_READ | ARCH_MMU_FLAGS_WRITE | ARCH_MMU_FLAGS_NOEXEC, NULL);
+	__assert(_cpu()->schedulerstack);
+	_cpu()->schedulerstack = (void *)((uintptr_t)_cpu()->schedulerstack + SCHEDULER_STACK_SIZE);
+
+	_cpu()->idlethread = sched_newthread(cpuidlethread, PAGE_SIZE * 4, 3, NULL, NULL);
+	__assert(_cpu()->idlethread);
+
+	timer_insert(_cpu()->timer, &_cpu()->schedtimerentry, timerhook, NULL, QUANTUM_US, true);
+	timer_resume(_cpu()->timer);
+	sched_stopcurrentthread();
+}
+
 void sched_init() {
 	threadcache = slab_newcache(sizeof(thread_t), 0, NULL, NULL);
 	__assert(threadcache);
 	processcache = slab_newcache(sizeof(proc_t), 0, NULL, NULL);
 	__assert(processcache);
 
-	_cpu()->schedulerstack = alloc(SCHEDULER_STACK_SIZE);
+	_cpu()->schedulerstack = vmm_map(NULL, SCHEDULER_STACK_SIZE, VMM_FLAGS_ALLOCATE, ARCH_MMU_FLAGS_READ | ARCH_MMU_FLAGS_WRITE | ARCH_MMU_FLAGS_NOEXEC, NULL);
 	__assert(_cpu()->schedulerstack);
 	_cpu()->schedulerstack = (void *)((uintptr_t)_cpu()->schedulerstack + SCHEDULER_STACK_SIZE);