diff --git a/sys/compat/linuxkpi/common/include/linux/compat.h b/sys/compat/linuxkpi/common/include/linux/compat.h
--- a/sys/compat/linuxkpi/common/include/linux/compat.h
+++ b/sys/compat/linuxkpi/common/include/linux/compat.h
@@ -41,17 +41,20 @@
 extern void linux_free_current(struct task_struct *);
 extern struct domainset *linux_get_vm_domain_set(int node);
 
+#define	__current_unallocated(td)	\
+	__predict_false((td)->td_lkpi_task == NULL)
+
 static inline void
 linux_set_current(struct thread *td)
 {
-	if (__predict_false(td->td_lkpi_task == NULL))
+	if (__current_unallocated(td))
 		lkpi_alloc_current(td, M_WAITOK);
 }
 
 static inline int
 linux_set_current_flags(struct thread *td, int flags)
 {
-	if (__predict_false(td->td_lkpi_task == NULL))
+	if (__current_unallocated(td))
 		return (lkpi_alloc_current(td, flags));
 	return (0);
 }
diff --git a/sys/compat/linuxkpi/common/include/linux/slab.h b/sys/compat/linuxkpi/common/include/linux/slab.h
--- a/sys/compat/linuxkpi/common/include/linux/slab.h
+++ b/sys/compat/linuxkpi/common/include/linux/slab.h
@@ -106,12 +106,8 @@
 	return (flags & GFP_NATIVE_MASK);
 }
 
-static inline void *
-kmalloc(size_t size, gfp_t flags)
-{
-	return (malloc(MAX(size, sizeof(struct llist_node)), M_KMALLOC,
-	    linux_check_m_flags(flags)));
-}
+extern void *lkpi_kmalloc(size_t size, gfp_t flags);
+#define	kmalloc(size, flags)	lkpi_kmalloc(size, flags)
 
 static inline void *
 kmalloc_node(size_t size, gfp_t flags, int node)
diff --git a/sys/compat/linuxkpi/common/src/linux_slab.c b/sys/compat/linuxkpi/common/src/linux_slab.c
--- a/sys/compat/linuxkpi/common/src/linux_slab.c
+++ b/sys/compat/linuxkpi/common/src/linux_slab.c
@@ -30,6 +30,9 @@
 #include <linux/kernel.h>
 #include <linux/irq_work.h>
 #include <linux/llist.h>
+#include <linux/sched.h>
+
+#include <asm/fpu/api.h>
 
 #include <sys/param.h>
 #include <sys/taskqueue.h>
@@ -206,6 +209,28 @@
 	free(c, M_KMALLOC);
 }
 
+void *
+lkpi_kmalloc(size_t size, gfp_t flags)
+{
+	void *rv;
+	unsigned int save_fpu_level;
+
+	save_fpu_level =
+	    __current_unallocated(curthread) ? 0 : current->fpu_ctx_level;
+	if (__predict_false(save_fpu_level != 0)) {
+		current->fpu_ctx_level = 1;
+		kernel_fpu_end();
+	}
+	rv = malloc(MAX(size, sizeof(struct llist_node)), M_KMALLOC,
+	    linux_check_m_flags(flags));
+	if (__predict_false(save_fpu_level != 0)) {
+		kernel_fpu_begin();
+		current->fpu_ctx_level = save_fpu_level;
+	}
+
+	return (rv);
+}
+
 static void
 linux_kfree_async_fn(void *context, int pending)
 {