zstd: Disable intrinsics

Disable the aarch64 NEON SIMD intrinsics for kernel builds.  Safely
using them in the kernel context requires saving/restoring the FPU
registers which is not currently done.

This lets us remove the problematic workaround in the aarch64_compat.h
header which undefines the __aarch64__ macro.

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>

Author: behlendorf

module/Kbuild.in

@@ -293,10 +293,9 @@ ZSTD_UPSTREAM_OBJS := \
 
 zfs-objs += $(addprefix zstd/,$(ZSTD_OBJS) $(ZSTD_UPSTREAM_OBJS))
 
-# Disable aarch64 neon SIMD instructions for kernel mode
 $(addprefix $(obj)/zstd/,$(ZSTD_OBJS) $(ZSTD_UPSTREAM_OBJS)) : ccflags-y += -I$(zstd_include) $(ZFS_ZSTD_FLAGS)
 $(addprefix $(obj)/zstd/,$(ZSTD_OBJS) $(ZSTD_UPSTREAM_OBJS)) : asflags-y += -I$(zstd_include)
-$(addprefix $(obj)/zstd/,$(ZSTD_UPSTREAM_OBJS)) : ccflags-y += -include $(zstd_include)/aarch64_compat.h -include $(zstd_include)/zstd_compat_wrapper.h -Wp,-w
+$(addprefix $(obj)/zstd/,$(ZSTD_UPSTREAM_OBJS)) : ccflags-y += -include $(zstd_include)/zstd_compat_wrapper.h -Wp,-w
 $(obj)/zstd/zfs_zstd.o : ccflags-y += -include $(zstd_include)/zstd_compat_wrapper.h
 
 

module/Makefile.bsd

@@ -521,30 +521,6 @@ CFLAGS.zstd_ldm.c= -U__BMI__ -fno-tree-vectorize ${NO_WBITWISE_INSTEAD_OF_LOGICA
 CFLAGS.zstd_opt.c= -U__BMI__ -fno-tree-vectorize ${NO_WBITWISE_INSTEAD_OF_LOGICAL}
 
 .if ${MACHINE_ARCH} == "aarch64"
-__ZFS_ZSTD_AARCH64_FLAGS= -include ${SRCDIR}/zstd/include/aarch64_compat.h
-CFLAGS.zstd.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.entropy_common.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.error_private.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.fse_compress.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.fse_decompress.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.hist.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.huf_compress.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.huf_decompress.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.pool.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.xxhash.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.zstd_common.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.zstd_compress.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.zstd_compress_literals.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.zstd_compress_sequences.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.zstd_compress_superblock.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.zstd_ddict.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.zstd_decompress.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.zstd_decompress_block.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.zstd_double_fast.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.zstd_fast.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.zstd_lazy.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.zstd_ldm.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
-CFLAGS.zstd_opt.c+= ${__ZFS_ZSTD_AARCH64_FLAGS}
 
 sha256-armv8.o: sha256-armv8.S
 	${CC} -c ${CFLAGS:N-mgeneral-regs-only} ${WERROR} ${.IMPSRC} \

module/zstd/include/aarch64_compat.h

@@ -12,6 +12,15 @@
 #ifndef ZSTD_CCOMMON_H_MODULE
 #define ZSTD_CCOMMON_H_MODULE
 
+/*
+ * Disable the aarch64 NEON SIMD intrinsics for kernel builds.  Safely
+ * using them in the kernel context requires saving/restoring the FPU
+ * registers which is not currently done.
+ */
+#ifdef _KERNEL
+#define ZSTD_NO_INTRINSICS
+#endif
+
 /* this module contains definitions which must be identical
  * across compression, decompression and dictBuilder.
  * It also contains a few functions useful to at least 2 of them