diff --git a/include/os/windows/spl/sys/simd.h b/include/os/windows/spl/sys/simd.h
index ed46cabd3770..cf13da1ceab0 100644
--- a/include/os/windows/spl/sys/simd.h
+++ b/include/os/windows/spl/sys/simd.h
@@ -116,6 +116,14 @@ extern uint32_t kfpu_state;
 	if (NT_SUCCESS(saveStatus)) \
 		KeRestoreExtendedProcessorState(&SaveState);
 
+#define	kfpu_begin_ctx(O) \
+	(O)->saveStatus = \
+	KeSaveExtendedProcessorState(kfpu_state, &(O)->SaveState);
+
+#define	kfpu_end_ctx(O) \
+	if (NT_SUCCESS(((O)->saveStatus))) \
+		KeRestoreExtendedProcessorState(&(O)->SaveState);
+
 /*
  * CPUID feature tests for user-space. Linux kernel provides an interface for
  * CPU feature testing.
diff --git a/include/sys/zio_checksum.h b/include/sys/zio_checksum.h
index 9fb79ab4a54b..b00deb290e46 100644
--- a/include/sys/zio_checksum.h
+++ b/include/sys/zio_checksum.h
@@ -69,6 +69,10 @@ typedef struct zio_abd_checksum_data {
 	fletcher_4_ctx_t	*acd_ctx;
 	zio_cksum_t 		*acd_zcp;
 	void 			*acd_private;
+#if defined(_WIN32) && defined(_KERNEL)
+	NTSTATUS saveStatus;
+	XSTATE_SAVE SaveState;
+#endif
 } zio_abd_checksum_data_t;
 
 typedef void zio_abd_checksum_init_t(zio_abd_checksum_data_t *);
diff --git a/module/zcommon/zfs_fletcher.c b/module/zcommon/zfs_fletcher.c
index fa65f4c584dc..3c23ddd99d88 100644
--- a/module/zcommon/zfs_fletcher.c
+++ b/module/zcommon/zfs_fletcher.c
@@ -457,12 +457,14 @@ fletcher_4_native_impl(const void *buf, uint64_t size, zio_cksum_t *zcp)
 
 	if (ops->uses_fpu == B_TRUE) {
 		kfpu_begin();
-	}
-	ops->init_native(&ctx);
-	ops->compute_native(&ctx, buf, size);
-	ops->fini_native(&ctx, zcp);
-	if (ops->uses_fpu == B_TRUE) {
+		ops->init_native(&ctx);
+		ops->compute_native(&ctx, buf, size);
+		ops->fini_native(&ctx, zcp);
 		kfpu_end();
+	} else {
+		ops->init_native(&ctx);
+		ops->compute_native(&ctx, buf, size);
+		ops->fini_native(&ctx, zcp);
 	}
 }
 
@@ -505,12 +507,14 @@ fletcher_4_byteswap_impl(const void *buf, uint64_t size, zio_cksum_t *zcp)
 
 	if (ops->uses_fpu == B_TRUE) {
 		kfpu_begin();
-	}
-	ops->init_byteswap(&ctx);
-	ops->compute_byteswap(&ctx, buf, size);
-	ops->fini_byteswap(&ctx, zcp);
-	if (ops->uses_fpu == B_TRUE) {
+		ops->init_byteswap(&ctx);
+		ops->compute_byteswap(&ctx, buf, size);
+		ops->fini_byteswap(&ctx, zcp);
 		kfpu_end();
+	} else {
+		ops->init_byteswap(&ctx);
+		ops->compute_byteswap(&ctx, buf, size);
+		ops->fini_byteswap(&ctx, zcp);
 	}
 }
 
@@ -827,7 +831,11 @@ abd_fletcher_4_init(zio_abd_checksum_data_t *cdp)
 	cdp->acd_private = (void *) ops;
 
 	if (ops->uses_fpu == B_TRUE) {
+#if defined(_WIN32) && defined(_KERNEL)
+		kfpu_begin_ctx(cdp);
+#else
 		kfpu_begin();
+#endif
 	}
 	if (cdp->acd_byteorder == ZIO_CHECKSUM_NATIVE)
 		ops->init_native(cdp->acd_ctx);
@@ -849,7 +857,11 @@ abd_fletcher_4_fini(zio_abd_checksum_data_t *cdp)
 		ops->fini_byteswap(cdp->acd_ctx, cdp->acd_zcp);
 
 	if (ops->uses_fpu == B_TRUE) {
+#if defined(_WIN32) && defined(_KERNEL)
+		kfpu_end_ctx(cdp);
+#else
 		kfpu_end();
+#endif
 	}
 }