diff --git a/base-hw/src/core/kernel/kernel.cc b/base-hw/src/core/kernel/kernel.cc
index bbb52eb4b..d6926d865 100644
--- a/base-hw/src/core/kernel/kernel.cc
+++ b/base-hw/src/core/kernel/kernel.cc
@@ -213,7 +213,7 @@ extern "C" void init_kernel_multiprocessor()
 
 	/* synchronize data view of all processors */
 	Processor::invalidate_data_caches();
-	Processor::invalidate_instruction_caches();
+	Processor::invalidate_instr_caches();
 	Processor::invalidate_control_flow_predictions();
 	Processor::data_synchronization_barrier();
 
diff --git a/base-hw/src/core/kernel/thread.cc b/base-hw/src/core/kernel/thread.cc
index e3935382d..6bd89fa75 100644
--- a/base-hw/src/core/kernel/thread.cc
+++ b/base-hw/src/core/kernel/thread.cc
@@ -543,6 +543,19 @@ void Thread::_call_update_pd()
 
 void Thread::_call_update_data_region()
 {
+	/*
+	 * FIXME: If the caller is not a core thread, the kernel operates in a
+	 *        different address space than the caller. Combined with the fact
+	 *        that at least ARMv7 doesn't provide cache operations by physical
+	 *        address, this prevents us from selectively maintaining caches.
+	 *        The future solution will be a kernel that is mapped to every
+	 *        address space so we can use virtual addresses of the caller. Up
+	 *        until then we apply operations to caches as a whole instead.
+	 */
+	if (!_core()) {
+		Processor::flush_data_caches();
+		return;
+	}
 	auto base = (addr_t)user_arg_1();
 	auto const size = (size_t)user_arg_2();
 	Processor::flush_data_caches_by_virt_region(base, size);
@@ -551,6 +564,20 @@ void Thread::_call_update_data_region()
 
 void Thread::_call_update_instr_region()
 {
+	/*
+	 * FIXME: If the caller is not a core thread, the kernel operates in a
+	 *        different address space than the caller. Combined with the fact
+	 *        that at least ARMv7 doesn't provide cache operations by physical
+	 *        address, this prevents us from selectively maintaining caches.
+	 *        The future solution will be a kernel that is mapped to every
+	 *        address space so we can use virtual addresses of the caller. Up
+	 *        until then we apply operations to caches as a whole instead.
+	 */
+	if (!_core()) {
+		Processor::flush_data_caches();
+		Processor::invalidate_instr_caches();
+		return;
+	}
 	auto base = (addr_t)user_arg_1();
 	auto const size = (size_t)user_arg_2();
 	Processor::flush_data_caches_by_virt_region(base, size);
diff --git a/base-hw/src/core/processor_driver/arm.h b/base-hw/src/core/processor_driver/arm.h
index e27664222..0d8bf092d 100644
--- a/base-hw/src/core/processor_driver/arm.h
+++ b/base-hw/src/core/processor_driver/arm.h
@@ -625,7 +625,7 @@ namespace Arm
 		 * Invalidate all entries of all instruction caches
 		 */
 		__attribute__((always_inline))
-		static void invalidate_instruction_caches()
+		static void invalidate_instr_caches()
 		{
 			asm volatile ("mcr p15, 0, %[rd], c7, c5, 0" :: [rd]"r"(0) : );
 		}
@@ -646,7 +646,7 @@ namespace Arm
 		static void flush_caches()
 		{
 			flush_data_caches();
-			invalidate_instruction_caches();
+			invalidate_instr_caches();
 		}
 
 		/**