#
# Derived from source code of TrueCrypt 7.1a, which is
# Copyright (c) 2008-2012 TrueCrypt Developers Association and which is governed
# by the TrueCrypt License 3.0.
#
# Modifications and additions to the original source code (contained in this file)
# and all other portions of this file are Copyright (c) 2013-2017 IDRIX
# and are governed by the Apache License 2.0 the full text of which is
# contained in the file License.txt included in VeraCrypt binary and source
# code distribution packages.
#
#------ Command line arguments ------
# DEBUG: Disable optimizations and enable debugging checks
# DEBUGGER: Enable debugging information for use by debuggers
# NOASM: Exclude modules requiring assembler
# NOGUI: Disable graphical user interface (build console-only application)
# NOSTRIP: Do not strip release binary
# NOTEST: Do not test release binary
# RESOURCEDIR: Run-time resource directory
# VERBOSE: Enable verbose messages
# WXSTATIC: Use static wxWidgets library
# SSSE3: Enable SSSE3 support in compiler
# SSE41: Enable SSE4.1 support in compiler
# NOSSE2: Disable SEE2 support in compiler
# WITHGTK3: Build wxWidgets against GTK3
#------ Targets ------
# all
# clean
# wxbuild: Configure and build wxWidgets - source code must be located at $(WX_ROOT)
#------ Build configuration ------
export APPNAME := veracrypt
export BASE_DIR := $(CURDIR)
export BUILD_INC := $(BASE_DIR)/Build/Include
export AR ?= ar
export CC ?= gcc
export CXX ?= g++
export AS := yasm
export RANLIB ?= ranlib
export CFLAGS := -Wall
export CXXFLAGS := -Wall -Wno-unused-parameter
C_CXX_FLAGS := -MMD -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -D_LARGE_FILES -I$(BASE_DIR) -I$(BASE_DIR)/Crypto
export ASFLAGS := -D __GNUC__ -D __YASM__
export LFLAGS :=
export PKG_CONFIG_PATH ?= /usr/local/lib/pkgconfig
export WX_CONFIG ?= wx-config
export WX_CONFIG_ARGS := --unicode
WX_CONFIGURE_FLAGS :=
export WXCONFIG_CFLAGS :=
export WXCONFIG_CXXFLAGS :=
WX_ROOT ?= $(BASE_DIR)/wxWidgets
export TC_BUILD_CONFIG := Release
ifeq "$(origin DEBUG)" "command line"
ifneq "$(DEBUG)" "0"
TC_BUILD_CONFIG := Debug
endif
endif
ifeq "$(origin NOGUI)" "command line"
export TC_NO_GUI := 1
C_CXX_FLAGS += -DTC_NO_GUI -DwxUSE_GUI=0
WX_CONFIGURE_FLAGS += --disable-gui
endif
ifdef PKCS11_INC
C_CXX_FLAGS += -I$(PKCS11_INC)
else
C_CXX_FLAGS += -I$(CURDIR)/PKCS11
endif
ifeq "$(origin RESOURCEDIR)" "command line"
C_CXX_FLAGS += -DTC_RESOURCE_DIR="$(RESOURCEDIR)"
endif
ifneq "$(origin VERBOSE)" "command line"
MAKEFLAGS += -s
endif
ifeq "$(origin WXSTATIC)" "command line"
export VC_WX_STATIC := 1
WX_CONFIG = $(WX_BUILD_DIR)/wx-config
WX_CONFIG_ARGS += --static
ifneq "$(WXSTATIC)" "FULL"
export VC_WX_MINIMAL := 1
endif
endif
ifeq "$(origin INDICATOR)" "command line"
ifneq (,$(findstring gtk3,$(shell $(WX_CONFIG) --selected-config)))
INDICATOR_LIBRARY=ayatana-appindicator3-0.1
else
INDICATOR_LIBRARY=ayatana-appindicator-0.1
endif
export LIBS += $(shell pkg-config --libs $(INDICATOR_LIBRARY))
C_CXX_FLAGS += $(shell pkg-config --cflags $(INDICATOR_LIBRARY)) -DHAVE_INDICATORS
endif
#------ Release configuration ------
ifeq "$(TC_BUILD_CONFIG)" "Release"
C_CXX_FLAGS += -O2 -fno-strict-aliasing # Do not enable strict aliasing
export WX_BUILD_DIR ?= $(BASE_DIR)/wxrelease
WX_CONFIGURE_FLAGS += --disable-debug_flag --disable-debug_gdb --disable-debug_info
else
#------ Debug configuration ------
C_CXX_FLAGS += -DDEBUG
CXXFLAGS += -fno-default-inline -Wno-unused-function -Wno-unused-variable
export WX_BUILD_DIR ?= $(BASE_DIR)/wxdebug
WX_CONFIGURE_FLAGS += --enable-debug_flag --disable-debug_gdb --disable-debug_info
endif
#------ Debugger configuration ------
ifeq "$(origin DEBUGGER)" "command line"
C_CXX_FLAGS += -ggdb
WX_CONFIGURE_FLAGS += --enable-debug_gdb --enable-debug_info
endif
#------ Platform configuration ------
export PLATFORM := "Unknown"
export PLATFORM_ARCH := "Unknown"
export PLATFORM_UNSUPPORTED := 0
export CPU_ARCH ?= unknown
export SIMD_SUPPORTED := 0
export DISABLE_AESNI ?= 0
export GCC_GTEQ_440 := 0
export GCC_GTEQ_430 := 0
ARCH ?= $(shell uname -m)
ifneq (,$(filter i386 i486 i586 i686 x86,$(ARCH)))
CPU_ARCH = x86
ASFLAGS += -f elf32 -D __BITS__=32
else ifneq (,$(filter x86_64 x86-64 amd64 x64,$(ARCH)))
CPU_ARCH = x64
ASFLAGS += -f elf64 -D __BITS__=64
else ifneq (,$(filter armv7l,$(ARCH)))
PLATFORM_ARCH := armv7
CPU_ARCH = armv7
endif
ifeq "$(origin NOASM)" "command line"
CPU_ARCH = unknown
C_CXX_FLAGS += -DCRYPTOPP_DISABLE_X86ASM
endif
ifeq "$(CPU_ARCH)" "x86"
PLATFORM_ARCH := i386
SIMD_SUPPORTED := 1
C_CXX_FLAGS += -D TC_ARCH_X86
else ifeq "$(CPU_ARCH)" "x64"
PLATFORM_ARCH := amd64
SIMD_SUPPORTED := 1
C_CXX_FLAGS += -D TC_ARCH_X64
endif
ifeq "$(origin NOSSE2)" "command line"
SIMD_SUPPORTED := 0
endif
ifeq "$(origin NOAESNI)" "command line"
DISABLE_AESNI := 1
endif
#------ Linux configuration ------
ifeq "$(shell uname -s)" "Linux"
PLATFORM := Linux
C_CXX_FLAGS += -DTC_UNIX -DTC_LINUX
# GNU GCC version 11 and higher compile with -std=gnu++17 by default
# which breaks "byte" definitions in Crypto++ library. So set
# -std=gnu++14 instead.
GCC11PLUS := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 11)
ifeq "$(GCC11PLUS)" "1"
CXXFLAGS += -std=gnu++14
endif
ifeq "$(SIMD_SUPPORTED)" "1"
CFLAGS += -msse2
CXXFLAGS += -msse2
GCC_GTEQ_440 := $(shell expr `$(CC) -dumpversion | sed -e 's/\.\([0-9][0-9]\)/\1/g' -e 's/\.\([0-9]\)/0\1/g' -e 's/^[0-9]\{3,4\}$$/&00/' -e 's/^[0-9]\{1,2\}$$/&0000/'` \>= 40400)
GCC_GTEQ_430 := $(shell expr `$(CC) -dumpversion | sed -e 's/\.\([0-9][0-9]\)/\1/g' -e 's/\.\([0-9]\)/0\1/g' -e 's/^[0-9]\{3,4\}$$/&00/' -e 's/^[0-9]\{1,2\}$$/&0000/'` \>= 40300)
ifeq "$(DISABLE_AESNI)" "1"
CFLAGS += -mno-aes -DCRYPTOPP_DISABLE_AESNI
CXXFLAGS += -mno-aes -DCRYPTOPP_DISABLE_AESNI
else
ifeq "$(GCC_GTEQ_440)" "1"
CFLAGS += -maes
CXXFLAGS += -maes
endif
endif
ifeq "$(GCC_GTEQ_430)" "1"
ifeq "$(origin SSSE3)" "command line"
CFLAGS += -mssse3
CXXFLAGS += -mssse3
endif
ifeq "$(origin SSE41)" "command line"
CFLAGS += -mssse3 -msse4.1
CXXFLAGS += -mssse3 -msse4.1
endif
endif
endif
ifeq "$(TC_BUILD_CONFIG)" "Release"
C_CXX_FLAGS += -fdata-sections -ffunction-sections -fpie
LFLAGS += -Wl,--gc-sections -pie
ifneq "$(shell ld --help 2>&1 | grep sysv | wc -l)" "0"
LFLAGS += -Wl,--hash-style=sysv
endif
WXCONFIG_CFLAGS += -fdata-sections -ffunction-sections -fpie
WXCONFIG_CXXFLAGS += -fdata-sections -ffunction-sections -fpie
endif
ifneq "$(origin WXSTATIC)" "command line"
LFLAGS += -ldl
else
GCC5USED := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 5)
ifeq "$(GCC5USED)" "1"
CXXFLAGS += -D_GLIBCXX_USE_CXX11_ABI=0
WXCONFIG_CXXFLAGS += -D_GLIBCXX_USE_CXX11_ABI=0
endif
endif
ifeq "$(origin NOSSE2)" "command line"
CFLAGS += -mno-sse2
CXXFLAGS += -mno-sse2
WXCONFIG_CFLAGS += -mno-sse2
WXCONFIG_CXXFLAGS += -mno-sse2
endif
ifeq "$(origin WITHGTK3)" "command line"
WX_CONFIGURE_FLAGS += --with-gtk=3
endif
endif
#------ Mac OS X configuration ------
ifeq "$(shell uname -s)" "Darwin"
PLATFORM := MacOSX
APPNAME := VeraCrypt
export VC_OSX_TARGET ?= 10.7
export VC_OSX_SDK ?= $(VC_OSX_TARGET)
#check to see if XCode 3 path exists.Otherwise, use XCode 4 path
VC_OSX_SDK_PATH := /Developer/SDKs/MacOSX$(VC_OSX_SDK).sdk
ifeq ($(wildcard $(VC_OSX_SDK_PATH)/SDKSettings.plist),)
VC_OSX_SDK_PATH := /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX$(VC_OSX_SDK).sdk
endif
#----- Legacy build if OSX <= 10.8: we build both 32-bit and 64-bit ----
ifneq (,$(filter 10.6 10.7 10.8,$(VC_OSX_TARGET)))
export VC_LEGACY_BUILD := 1
endif
CC := gcc
CXX := g++
C_CXX_FLAGS += -DTC_UNIX -DTC_BSD -DTC_MACOSX -mmacosx-version-min=$(VC_OSX_TARGET) -isysroot $(VC_OSX_SDK_PATH)
LFLAGS += -mmacosx-version-min=$(VC_OSX_TARGET) -Wl,-syslibroot $(VC_OSX_SDK_PATH)
WX_CONFIGURE_FLAGS += --with-macosx-version-min=$(VC_OSX_TARGET) --with-macosx-sdk=$(VC_OSX_SDK_PATH)
ifeq "$(CPU_ARCH)" "x64"
CPU_ARCH = x86
endif
ifeq "$(CPU_ARCH)" "arm64"
CPU_ARCH = x86
endif
CFLAGS += -msse2
CXXFLAGS += -msse2
ifeq "$(origin SSSE3)" "command line"
CFLAGS += -mssse3
CXXFLAGS += -mssse3
endif
ifeq "$(origin SSE41)" "command line"
CFLAGS += -mssse3 -msse4.1
CXXFLAGS += -mssse3 -msse4.1
endif
AS := $(BASE_DIR)/Build/Tools/MacOSX/yasm
export ASFLAGS32 := -D __GNUC__ -D __YASM__ -D __BITS__=32 --prefix=_ -f macho32
export ASFLAGS64 := -D __GNUC__ -D __YASM__ -D __BITS__=64 --prefix=_ -f macho64
ifeq "$(TC_BUILD_CONFIG)" "Release"
export DISABLE_PRECOMPILED_HEADERS := 1
S := $(C_CXX_FLAGS)
C_CXX_FLAGS = $(subst -MMD,,$(S))
C_CXX_FLAGS += -gfull -arch x86_64
LFLAGS += -Wl,-dead_strip -arch x86_64
WX_CONFIGURE_FLAGS += --without-libpng --disable-gif --disable-pcx --disable-tga --disable-iff --disable-gif --disable-svg
#----- Legacy build: we build both 32-bit and 64-bit ----
ifdef VC_LEGACY_BUILD
C_CXX_FLAGS += -arch i386
LFLAGS += -arch i386
WX_CONFIGURE_FLAGS += --enable-universal_binary=i386,x86_64
else
CXXFLAGS += -std=c++11
C_CXX_FLAGS += -arch arm64
LFLAGS += -arch arm64
WX_CONFIGURE_FLAGS += --enable-universal_binary=arm64,x86_64
endif
WXCONFIG_CFLAGS += -gfull
WXCONFIG_CXXFLAGS += -gfull
else
WX_CONFIGURE_FLAGS += --disable-universal_binary
endif
endif
#------ FreeBSD configuration ------
ifeq "$(shell uname -s)" "FreeBSD"
PLATFORM := FreeBSD
PLATFORM_UNSUPPORTED := 1
C_CXX_FLAGS += -DTC_UNIX -DTC_BSD -DTC_FREEBSD
CC := cc
CXX := c++
ifeq "$(TC_BUILD_CONFIG)" "Release"
C_CXX_FLAGS += -fdata-sections -ffunction-sections -fpie
LFLAGS += -Wl,--gc-sections -pie
ifneq "$(shell ld --help 2>&1 | grep sysv | wc -l)" "0"
LFLAGS += -Wl,--hash-style=sysv
endif
WXCONFIG_CFLAGS += -fpie -fPIC
WXCONFIG_CXXFLAGS += -fpie -fPIC
endif
ifeq "$(SIMD_SUPPORTED)" "1"
CFLAGS += -msse2
CXXFLAGS += -msse2
ifeq "$(DISABLE_AESNI)" "1"
CFLAGS += -mno-aes -DCRYPTOPP_DISABLE_AESNI
CXXFLAGS += -mno-aes -DCRYPTOPP_DISABLE_AESNI
else
CFLAGS += -maes
CXXFLAGS += -maes
endif
ifeq "$(origin SSSE3)" "command line"
CFLAGS += -mssse3
CXXFLAGS += -mssse3
endif
ifeq "$(origin SSE41)" "command line"
CFLAGS += -mssse3 -msse4.1
CXXFLAGS += -mssse3 -msse4.1
endif
endif
ifeq "$(origin NOSSE2)" "command line"
CFLAGS += -mno-sse2
CXXFLAGS += -mno-sse2
WXCONFIG_CFLAGS += -mno-sse2
WXCONFIG_CXXFLAGS += -mno-sse2
endif
endif
#------ OpenBSD configuration ------
ifeq "$(shell uname -s)" "OpenBSD"
PLATFORM := OpenBSD
PLATFORM_UNSUPPORTED := 1
C_CXX_FLAGS += -DTC_UNIX -DTC_BSD -DTC_OPENBSD
CC := cc
CXX := c++
ifeq "$(TC_BUILD_CONFIG)" "Release"
C_CXX_FLAGS += -fdata-sections -ffunction-sections -fpie
LFLAGS += -Wl,--gc-sections -pie
WXCONFIG_CFLAGS += -fpie -fPIC
WXCONFIG_CXXFLAGS += -fpie -fPIC
endif
endif
#------ Solaris configuration ------
ifeq "$(shell uname -s)" "SunOS"
PLATFORM := Solaris
PLATFORM_UNSUPPORTED := 1
C_CXX_FLAGS += -DTC_UNIX -DTC_SOLARIS
WX_CONFIGURE_FLAGS += --with-gtk
endif
#------ Common configuration ------
CFLAGS := $(C_CXX_FLAGS) $(CFLAGS) $(TC_EXTRA_CFLAGS)
CXXFLAGS := $(C_CXX_FLAGS) $(CXXFLAGS) $(TC_EXTRA_CXXFLAGS)
LFLAGS := $(LFLAGS) $(TC_EXTRA_LFLAGS)
WX_CONFIGURE_FLAGS += --enable-unicode -disable-shared --disable-dependency-tracking --enable-exceptions --enable-std_string --enable-dataobj --enable-mimetype
ifdef VC_WX_MINIMAL
WX_CONFIGURE_FLAGS += --disable-protocol --disable-protocols --disable-url --disable-ipc --disable-sockets --disable-fs_inet --disable-ole --disable-docview --disable-clipboard \
--disable-help --disable-html --disable-mshtmlhelp --disable-htmlhelp --disable-mdi --disable-metafile --disable-webkit --disable-webview \
--disable-xrc --disable-aui --disable-postscript --disable-printarch \
--disable-arcstream --disable-fs_archive --disable-fs_zip --disable-tarstream --disable-zipstream \
--disable-animatectrl --disable-bmpcombobox --disable-calendar --disable-caret --disable-checklst --disable-collpane --disable-colourpicker --disable-comboctrl \
--disable-datepick --disable-display --disable-dirpicker --disable-filepicker --disable-fontpicker --disable-grid --disable-dataviewctrl \
--disable-listbook --disable-odcombobox --disable-sash --disable-searchctrl --disable-slider --disable-splitter --disable-togglebtn \
--disable-toolbar --disable-tbarnative --disable-treebook --disable-toolbook --disable-tipwindow --disable-popupwin \
--disable-commondlg --disable-aboutdlg --disable-coldlg --disable-finddlg --disable-fontdlg --disable-numberdlg --disable-splash \
--disable-tipdlg --disable-progressdlg --disable-wizarddlg --disable-miniframe --disable-splines --disable-palette \
--disable-richtext --disable-dialupman --disable-debugreport --disable-filesystem --disable-rearrangectrl --disable-treelist --disable-richmsgdlg \
--disable-richtooltip --disable-propgrid --disable-stc --without-libnotify \
--without-gtkprint --without-gnomevfs --disable-fsvolume --disable-fswatcher \
--disable-sound --disable-mediactrl --disable-joystick --disable-apple_ieee \
--disable-gif --disable-pcx --disable-tga --disable-iff --disable-gif --disable-pnm --disable-svg \
--without-expat --without-libtiff --without-libjpeg --without-libpng -without-regex --without-zlib
ifeq "$(PLATFORM)" "Linux"
WX_CONFIGURE_FLAGS += --disable-tooltips
ifneq "$(origin WITHGTK3)" "command line"
WX_CONFIGURE_FLAGS += --disable-graphics_ctx
endif
else
WX_CONFIGURE_FLAGS += --disable-graphics_ctx
endif
endif
#------ Project build ------
PROJ_DIRS := Platform Volume Driver/Fuse Core Main
.PHONY: all clean wxbuild
all clean:
@if pwd | grep -q ' '; then echo 'Error: source code is stored in a path containing spaces' >&2; exit 1; fi
@for DIR in $(PROJ_DIRS); do \
PROJ=$$(echo $$DIR | cut -d/ -f1); \
$(MAKE) -C $$DIR -f $$PROJ.make NAME=$$PROJ $(MAKECMDGOALS) || exit $?; \
export LIBS="$(BASE_DIR)/$$DIR/$$PROJ.a $$LIBS"; \
done
install:
$(MAKE) -C Main -f Main.make NAME=Main install
package:
$(MAKE) -C Main -f Main.make NAME=Main package
#------ wxWidgets build ------
ifeq "$(MAKECMDGOALS)" "wxbuild"
CFLAGS :=
CXXFLAGS :=
LFLAGS :=
endif
wxbuild:
ifneq "$(shell test -f $(WX_ROOT)/configure || test -f $(WX_BUILD_DIR)/../configure && echo 1)" "1"
@echo 'Error: WX_ROOT must point to wxWidgets source code directory' >&2
@exit 1
endif
rm -rf "$(WX_BUILD_DIR)"
mkdir -p "$(WX_BUILD_DIR)"
@echo Configuring wxWidgets library...
cd "$(WX_BUILD_DIR)" && "$(WX_ROOT)/configure" $(WX_CONFIGURE_FLAGS) >/dev/null
@echo Building wxWidgets library...
cd "$(WX_BUILD_DIR)" && $(MAKE) -j 4
/a>
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
; ---------------------------------------------------------------------------
; Copyright (c) 1998-2007, Brian Gladman, Worcester, UK. All rights reserved.
;
; LICENSE TERMS
;
; The free distribution and use of this software is allowed (with or without
; changes) provided that:
;
; 1. source code distributions include the above copyright notice, this
; list of conditions and the following disclaimer;
;
; 2. binary distributions include the above copyright notice, this list
; of conditions and the following disclaimer in their documentation;
;
; 3. the name of the copyright holder is not used to endorse products
; built using this software without specific written permission.
;
; DISCLAIMER
;
; This software is provided 'as is' with no explicit or implied warranties
; in respect of its properties, including, but not limited to, correctness
; and/or fitness for purpose.
; ---------------------------------------------------------------------------
; Issue 20/12/2007
;
; I am grateful to Dag Arne Osvik for many discussions of the techniques that
; can be used to optimise AES assembler code on AMD64/EM64T architectures.
; Some of the techniques used in this implementation are the result of
; suggestions made by him for which I am most grateful.
;
; Adapted for TrueCrypt:
; - Compatibility with NASM
;
; An AES implementation for AMD64 processors using the YASM assembler. This
; implemetation provides only encryption, decryption and hence requires key
; scheduling support in C. It uses 8k bytes of tables but its encryption and
; decryption performance is very close to that obtained using large tables.
; It can use either Windows or Gnu/Linux calling conventions, which are as
; follows:
; windows gnu/linux
;
; in_blk rcx rdi
; out_blk rdx rsi
; context (cx) r8 rdx
;
; preserved rsi - + rbx, rbp, rsp, r12, r13, r14 & r15
; registers rdi - on both
;
; destroyed - rsi + rax, rcx, rdx, r8, r9, r10 & r11
; registers - rdi on both
;
; The default convention is that for windows, the gnu/linux convention being
; used if __GNUC__ is defined.
;
; Define _SEH_ to include support for Win64 structured exception handling
; (this requires YASM version 0.6 or later).
;
; This code provides the standard AES block size (128 bits, 16 bytes) and the
; three standard AES key sizes (128, 192 and 256 bits). It has the same call
; interface as my C implementation. It uses the Microsoft C AMD64 calling
; conventions in which the three parameters are placed in rcx, rdx and r8
; respectively. The rbx, rsi, rdi, rbp and r12..r15 registers are preserved.
;
; AES_RETURN aes_encrypt(const unsigned char in_blk[],
; unsigned char out_blk[], const aes_encrypt_ctx cx[1]);
;
; AES_RETURN aes_decrypt(const unsigned char in_blk[],
; unsigned char out_blk[], const aes_decrypt_ctx cx[1]);
;
; AES_RETURN aes_encrypt_key<NNN>(const unsigned char key[],
; const aes_encrypt_ctx cx[1]);
;
; AES_RETURN aes_decrypt_key<NNN>(const unsigned char key[],
; const aes_decrypt_ctx cx[1]);
;
; AES_RETURN aes_encrypt_key(const unsigned char key[],
; unsigned int len, const aes_decrypt_ctx cx[1]);
;
; AES_RETURN aes_decrypt_key(const unsigned char key[],
; unsigned int len, const aes_decrypt_ctx cx[1]);
;
; where <NNN> is 128, 102 or 256. In the last two calls the length can be in
; either bits or bytes.
;
; Comment in/out the following lines to obtain the desired subroutines. These
; selections MUST match those in the C header file aes.h
; %define AES_128 ; define if AES with 128 bit keys is needed
; %define AES_192 ; define if AES with 192 bit keys is needed
%define AES_256 ; define if AES with 256 bit keys is needed
; %define AES_VAR ; define if a variable key size is needed
%define ENCRYPTION ; define if encryption is needed
%define DECRYPTION ; define if decryption is needed
%define AES_REV_DKS ; define if key decryption schedule is reversed
%define LAST_ROUND_TABLES ; define for the faster version using extra tables
; The encryption key schedule has the following in memory layout where N is the
; number of rounds (10, 12 or 14):
;
; lo: | input key (round 0) | ; each round is four 32-bit words
; | encryption round 1 |
; | encryption round 2 |
; ....
; | encryption round N-1 |
; hi: | encryption round N |
;
; The decryption key schedule is normally set up so that it has the same
; layout as above by actually reversing the order of the encryption key
; schedule in memory (this happens when AES_REV_DKS is set):
;
; lo: | decryption round 0 | = | encryption round N |
; | decryption round 1 | = INV_MIX_COL[ | encryption round N-1 | ]
; | decryption round 2 | = INV_MIX_COL[ | encryption round N-2 | ]
; .... ....
; | decryption round N-1 | = INV_MIX_COL[ | encryption round 1 | ]
; hi: | decryption round N | = | input key (round 0) |
;
; with rounds except the first and last modified using inv_mix_column()
; But if AES_REV_DKS is NOT set the order of keys is left as it is for
; encryption so that it has to be accessed in reverse when used for
; decryption (although the inverse mix column modifications are done)
;
; lo: | decryption round 0 | = | input key (round 0) |
; | decryption round 1 | = INV_MIX_COL[ | encryption round 1 | ]
; | decryption round 2 | = INV_MIX_COL[ | encryption round 2 | ]
; .... ....
; | decryption round N-1 | = INV_MIX_COL[ | encryption round N-1 | ]
; hi: | decryption round N | = | encryption round N |
;
; This layout is faster when the assembler key scheduling provided here
; is used.
;
; The DLL interface must use the _stdcall convention in which the number
; of bytes of parameter space is added after an @ to the sutine's name.
; We must also remove our parameters from the stack before return (see
; the do_exit macro). Define DLL_EXPORT for the Dynamic Link Library version.
;%define DLL_EXPORT
; End of user defines
%ifdef AES_VAR
%ifndef AES_128
%define AES_128
%endif
%ifndef AES_192
%define AES_192
%endif
%ifndef AES_256
%define AES_256
%endif
%endif
%ifdef AES_VAR
%define KS_LENGTH 60
%elifdef AES_256
%define KS_LENGTH 60
%elifdef AES_192
%define KS_LENGTH 52
%else
%define KS_LENGTH 44
%endif
%define r0 rax
%define r1 rdx
%define r2 rcx
%define r3 rbx
%define r4 rsi
%define r5 rdi
%define r6 rbp
%define r7 rsp
%define raxd eax
%define rdxd edx
%define rcxd ecx
%define rbxd ebx
%define rsid esi
%define rdid edi
%define rbpd ebp
%define rspd esp
%define raxb al
%define rdxb dl
%define rcxb cl
%define rbxb bl
%define rsib sil
%define rdib dil
%define rbpb bpl
%define rspb spl
%define r0h ah
%define r1h dh
%define r2h ch
%define r3h bh
%define r0d eax
%define r1d edx
%define r2d ecx
%define r3d ebx
; finite field multiplies by {02}, {04} and {08}
%define f2(x) ((x<<1)^(((x>>7)&1)*0x11b))
%define f4(x) ((x<<2)^(((x>>6)&1)*0x11b)^(((x>>6)&2)*0x11b))
%define f8(x) ((x<<3)^(((x>>5)&1)*0x11b)^(((x>>5)&2)*0x11b)^(((x>>5)&4)*0x11b))
; finite field multiplies required in table generation
%define f3(x) (f2(x) ^ x)
%define f9(x) (f8(x) ^ x)
%define fb(x) (f8(x) ^ f2(x) ^ x)
%define fd(x) (f8(x) ^ f4(x) ^ x)
%define fe(x) (f8(x) ^ f4(x) ^ f2(x))
; macro for expanding S-box data
%macro enc_vals 1
db %1(0x63),%1(0x7c),%1(0x77),%1(0x7b),%1(0xf2),%1(0x6b),%1(0x6f),%1(0xc5)
db %1(0x30),%1(0x01),%1(0x67),%1(0x2b),%1(0xfe),%1(0xd7),%1(0xab),%1(0x76)
db %1(0xca),%1(0x82),%1(0xc9),%1(0x7d),%1(0xfa),%1(0x59),%1(0x47),%1(0xf0)
db %1(0xad),%1(0xd4),%1(0xa2),%1(0xaf),%1(0x9c),%1(0xa4),%1(0x72),%1(0xc0)
db %1(0xb7),%1(0xfd),%1(0x93),%1(0x26),%1(0x36),%1(0x3f),%1(0xf7),%1(0xcc)
db %1(0x34),%1(0xa5),%1(0xe5),%1(0xf1),%1(0x71),%1(0xd8),%1(0x31),%1(0x15)
db %1(0x04),%1(0xc7),%1(0x23),%1(0xc3),%1(0x18),%1(0x96),%1(0x05),%1(0x9a)
db %1(0x07),%1(0x12),%1(0x80),%1(0xe2),%1(0xeb),%1(0x27),%1(0xb2),%1(0x75)
db %1(0x09),%1(0x83),%1(0x2c),%1(0x1a),%1(0x1b),%1(0x6e),%1(0x5a),%1(0xa0)
db %1(0x52),%1(0x3b),%1(0xd6),%1(0xb3),%1(0x29),%1(0xe3),%1(0x2f),%1(0x84)
db %1(0x53),%1(0xd1),%1(0x00),%1(0xed),%1(0x20),%1(0xfc),%1(0xb1),%1(0x5b)
db %1(0x6a),%1(0xcb),%1(0xbe),%1(0x39),%1(0x4a),%1(0x4c),%1(0x58),%1(0xcf)
db %1(0xd0),%1(0xef),%1(0xaa),%1(0xfb),%1(0x43),%1(0x4d),%1(0x33),%1(0x85)
db %1(0x45),%1(0xf9),%1(0x02),%1(0x7f),%1(0x50),%1(0x3c),%1(0x9f),%1(0xa8)
db %1(0x51),%1(0xa3),%1(0x40),%1(0x8f),%1(0x92),%1(0x9d),%1(0x38),%1(0xf5)
db %1(0xbc),%1(0xb6),%1(0xda),%1(0x21),%1(0x10),%1(0xff),%1(0xf3),%1(0xd2)
db %1(0xcd),%1(0x0c),%1(0x13),%1(0xec),%1(0x5f),%1(0x97),%1(0x44),%1(0x17)
db %1(0xc4),%1(0xa7),%1(0x7e),%1(0x3d),%1(0x64),%1(0x5d),%1(0x19),%1(0x73)
db %1(0x60),%1(0x81),%1(0x4f),%1(0xdc),%1(0x22),%1(0x2a),%1(0x90),%1(0x88)
db %1(0x46),%1(0xee),%1(0xb8),%1(0x14),%1(0xde),%1(0x5e),%1(0x0b),%1(0xdb)
db %1(0xe0),%1(0x32),%1(0x3a),%1(0x0a),%1(0x49),%1(0x06),%1(0x24),%1(0x5c)
db %1(0xc2),%1(0xd3),%1(0xac),%1(0x62),%1(0x91),%1(0x95),%1(0xe4),%1(0x79)
db %1(0xe7),%1(0xc8),%1(0x37),%1(0x6d),%1(0x8d),%1(0xd5),%1(0x4e),%1(0xa9)
db %1(0x6c),%1(0x56),%1(0xf4),%1(0xea),%1(0x65),%1(0x7a),%1(0xae),%1(0x08)
db %1(0xba),%1(0x78),%1(0x25),%1(0x2e),%1(0x1c),%1(0xa6),%1(0xb4),%1(0xc6)
db %1(0xe8),%1(0xdd),%1(0x74),%1(0x1f),%1(0x4b),%1(0xbd),%1(0x8b),%1(0x8a)
db %1(0x70),%1(0x3e),%1(0xb5),%1(0x66),%1(0x48),%1(0x03),%1(0xf6),%1(0x0e)
db %1(0x61),%1(0x35),%1(0x57),%1(0xb9),%1(0x86),%1(0xc1),%1(0x1d),%1(0x9e)
db %1(0xe1),%1(0xf8),%1(0x98),%1(0x11),%1(0x69),%1(0xd9),%1(0x8e),%1(0x94)
db %1(0x9b),%1(0x1e),%1(0x87),%1(0xe9),%1(0xce),%1(0x55),%1(0x28),%1(0xdf)
db %1(0x8c),%1(0xa1),%1(0x89),%1(0x0d),%1(0xbf),%1(0xe6),%1(0x42),%1(0x68)
db %1(0x41),%1(0x99),%1(0x2d),%1(0x0f),%1(0xb0),%1(0x54),%1(0xbb),%1(0x16)
%endmacro
%macro dec_vals 1
db %1(0x52),%1(0x09),%1(0x6a),%1(0xd5),%1(0x30),%1(0x36),%1(0xa5),%1(0x38)
db %1(0xbf),%1(0x40),%1(0xa3),%1(0x9e),%1(0x81),%1(0xf3),%1(0xd7),%1(0xfb)
db %1(0x7c),%1(0xe3),%1(0x39),%1(0x82),%1(0x9b),%1(0x2f),%1(0xff),%1(0x87)
db %1(0x34),%1(0x8e),%1(0x43),%1(0x44),%1(0xc4),%1(0xde),%1(0xe9),%1(0xcb)
db %1(0x54),%1(0x7b),%1(0x94),%1(0x32),%1(0xa6),%1(0xc2),%1(0x23),%1(0x3d)
db %1(0xee),%1(0x4c),%1(0x95),%1(0x0b),%1(0x42),%1(0xfa),%1(0xc3),%1(0x4e)
db %1(0x08),%1(0x2e),%1(0xa1),%1(0x66),%1(0x28),%1(0xd9),%1(0x24),%1(0xb2)
db %1(0x76),%1(0x5b),%1(0xa2),%1(0x49),%1(0x6d),%1(0x8b),%1(0xd1),%1(0x25)
db %1(0x72),%1(0xf8),%1(0xf6),%1(0x64),%1(0x86),%1(0x68),%1(0x98),%1(0x16)
db %1(0xd4),%1(0xa4),%1(0x5c),%1(0xcc),%1(0x5d),%1(0x65),%1(0xb6),%1(0x92)
db %1(0x6c),%1(0x70),%1(0x48),%1(0x50),%1(0xfd),%1(0xed),%1(0xb9),%1(0xda)
db %1(0x5e),%1(0x15),%1(0x46),%1(0x57),%1(0xa7),%1(0x8d),%1(0x9d),%1(0x84)
db %1(0x90),%1(0xd8),%1(0xab),%1(0x00),%1(0x8c),%1(0xbc),%1(0xd3),%1(0x0a)
db %1(0xf7),%1(0xe4),%1(0x58),%1(0x05),%1(0xb8),%1(0xb3),%1(0x45),%1(0x06)
db %1(0xd0),%1(0x2c),%1(0x1e),%1(0x8f),%1(0xca),%1(0x3f),%1(0x0f),%1(0x02)
db %1(0xc1),%1(0xaf),%1(0xbd),%1(0x03),%1(0x01),%1(0x13),%1(0x8a),%1(0x6b)
db %1(0x3a),%1(0x91),%1(0x11),%1(0x41),%1(0x4f),%1(0x67),%1(0xdc),%1(0xea)
db %1(0x97),%1(0xf2),%1(0xcf),%1(0xce),%1(0xf0),%1(0xb4),%1(0xe6),%1(0x73)
db %1(0x96),%1(0xac),%1(0x74),%1(0x22),%1(0xe7),%1(0xad),%1(0x35),%1(0x85)
db %1(0xe2),%1(0xf9),%1(0x37),%1(0xe8),%1(0x1c),%1(0x75),%1(0xdf),%1(0x6e)
db %1(0x47),%1(0xf1),%1(0x1a),%1(0x71),%1(0x1d),%1(0x29),%1(0xc5),%1(0x89)
db %1(0x6f),%1(0xb7),%1(0x62),%1(0x0e),%1(0xaa),%1(0x18),%1(0xbe),%1(0x1b)
db %1(0xfc),%1(0x56),%1(0x3e),%1(0x4b),%1(0xc6),%1(0xd2),%1(0x79),%1(0x20)
db %1(0x9a),%1(0xdb),%1(0xc0),%1(0xfe),%1(0x78),%1(0xcd),%1(0x5a),%1(0xf4)
db %1(0x1f),%1(0xdd),%1(0xa8),%1(0x33),%1(0x88),%1(0x07),%1(0xc7),%1(0x31)
db %1(0xb1),%1(0x12),%1(0x10),%1(0x59),%1(0x27),%1(0x80),%1(0xec),%1(0x5f)
db %1(0x60),%1(0x51),%1(0x7f),%1(0xa9),%1(0x19),%1(0xb5),%1(0x4a),%1(0x0d)
db %1(0x2d),%1(0xe5),%1(0x7a),%1(0x9f),%1(0x93),%1(0xc9),%1(0x9c),%1(0xef)
db %1(0xa0),%1(0xe0),%1(0x3b),%1(0x4d),%1(0xae),%1(0x2a),%1(0xf5),%1(0xb0)
db %1(0xc8),%1(0xeb),%1(0xbb),%1(0x3c),%1(0x83),%1(0x53),%1(0x99),%1(0x61)
db %1(0x17),%1(0x2b),%1(0x04),%1(0x7e),%1(0xba),%1(0x77),%1(0xd6),%1(0x26)
db %1(0xe1),%1(0x69),%1(0x14),%1(0x63),%1(0x55),%1(0x21),%1(0x0c),%1(0x7d)
%endmacro
%define u8(x) f2(x), x, x, f3(x), f2(x), x, x, f3(x)
%define v8(x) fe(x), f9(x), fd(x), fb(x), fe(x), f9(x), fd(x), x
%define w8(x) x, 0, 0, 0, x, 0, 0, 0
%define tptr rbp ; table pointer
%define kptr r8 ; key schedule pointer
%define fofs 128 ; adjust offset in key schedule to keep |disp| < 128
%define fk_ref(x,y) [kptr-16*x+fofs+4*y]
%ifdef AES_REV_DKS
%define rofs 128
%define ik_ref(x,y) [kptr-16*x+rofs+4*y]
%else
%define rofs -128
%define ik_ref(x,y) [kptr+16*x+rofs+4*y]
%endif
%define tab_0(x) [tptr+8*x]
%define tab_1(x) [tptr+8*x+3]
%define tab_2(x) [tptr+8*x+2]
%define tab_3(x) [tptr+8*x+1]
%define tab_f(x) byte [tptr+8*x+1]
%define tab_i(x) byte [tptr+8*x+7]
%define t_ref(x,r) tab_ %+ x(r)
%macro ff_rnd 5 ; normal forward round
mov %1d, fk_ref(%5,0)
mov %2d, fk_ref(%5,1)
mov %3d, fk_ref(%5,2)
mov %4d, fk_ref(%5,3)
movzx esi, al
movzx edi, ah
shr eax, 16
xor %1d, t_ref(0,rsi)
xor %4d, t_ref(1,rdi)
movzx esi, al
movzx edi, ah
xor %3d, t_ref(2,rsi)
xor %2d, t_ref(3,rdi)
movzx esi, bl
movzx edi, bh
shr ebx, 16
xor %2d, t_ref(0,rsi)
xor %1d, t_ref(1,rdi)
movzx esi, bl
movzx edi, bh
xor %4d, t_ref(2,rsi)
xor %3d, t_ref(3,rdi)
movzx esi, cl
movzx edi, ch
shr ecx, 16
xor %3d, t_ref(0,rsi)
xor %2d, t_ref(1,rdi)
movzx esi, cl
movzx edi, ch
xor %1d, t_ref(2,rsi)
xor %4d, t_ref(3,rdi)
movzx esi, dl
movzx edi, dh
shr edx, 16
xor %4d, t_ref(0,rsi)
xor %3d, t_ref(1,rdi)
movzx esi, dl
movzx edi, dh
xor %2d, t_ref(2,rsi)
xor %1d, t_ref(3,rdi)
mov eax,%1d
mov ebx,%2d
mov ecx,%3d
mov edx,%4d
%endmacro
%ifdef LAST_ROUND_TABLES
%macro fl_rnd 5 ; last forward round
add tptr, 2048
mov %1d, fk_ref(%5,0)
mov %2d, fk_ref(%5,1)
mov %3d, fk_ref(%5,2)
mov %4d, fk_ref(%5,3)
movzx esi, al
movzx edi, ah
shr eax, 16
xor %1d, t_ref(0,rsi)
xor %4d, t_ref(1,rdi)
movzx esi, al
movzx edi, ah
xor %3d, t_ref(2,rsi)
xor %2d, t_ref(3,rdi)
movzx esi, bl
movzx edi, bh
shr ebx, 16
xor %2d, t_ref(0,rsi)
xor %1d, t_ref(1,rdi)
movzx esi, bl
movzx edi, bh
xor %4d, t_ref(2,rsi)
xor %3d, t_ref(3,rdi)
movzx esi, cl
movzx edi, ch
shr ecx, 16
xor %3d, t_ref(0,rsi)
xor %2d, t_ref(1,rdi)
movzx esi, cl
movzx edi, ch
xor %1d, t_ref(2,rsi)
xor %4d, t_ref(3,rdi)
movzx esi, dl
movzx edi, dh
shr edx, 16
xor %4d, t_ref(0,rsi)
xor %3d, t_ref(1,rdi)
movzx esi, dl
movzx edi, dh
xor %2d, t_ref(2,rsi)
xor %1d, t_ref(3,rdi)
%endmacro
%else
%macro fl_rnd 5 ; last forward round
mov %1d, fk_ref(%5,0)
mov %2d, fk_ref(%5,1)
mov %3d, fk_ref(%5,2)
mov %4d, fk_ref(%5,3)
movzx esi, al
movzx edi, ah
shr eax, 16
movzx esi, t_ref(f,rsi)
movzx edi, t_ref(f,rdi)
xor %1d, esi
rol edi, 8
xor %4d, edi
movzx esi, al
movzx edi, ah
movzx esi, t_ref(f,rsi)
movzx edi, t_ref(f,rdi)
rol esi, 16
rol edi, 24
xor %3d, esi
xor %2d, edi
movzx esi, bl
movzx edi, bh
shr ebx, 16
movzx esi, t_ref(f,rsi)
movzx edi, t_ref(f,rdi)
xor %2d, esi
rol edi, 8
xor %1d, edi
movzx esi, bl
movzx edi, bh
movzx esi, t_ref(f,rsi)
movzx edi, t_ref(f,rdi)
rol esi, 16
rol edi, 24
xor %4d, esi
xor %3d, edi
movzx esi, cl
movzx edi, ch
movzx esi, t_ref(f,rsi)
movzx edi, t_ref(f,rdi)
shr ecx, 16
xor %3d, esi
rol edi, 8
xor %2d, edi
movzx esi, cl
movzx edi, ch
movzx esi, t_ref(f,rsi)
movzx edi, t_ref(f,rdi)
rol esi, 16
rol edi, 24
xor %1d, esi
xor %4d, edi
movzx esi, dl
movzx edi, dh
movzx esi, t_ref(f,rsi)
movzx edi, t_ref(f,rdi)
shr edx, 16
xor %4d, esi
rol edi, 8
xor %3d, edi
movzx esi, dl
movzx edi, dh
movzx esi, t_ref(f,rsi)
movzx edi, t_ref(f,rdi)
rol esi, 16
rol edi, 24
xor %2d, esi
xor %1d, edi
%endmacro
%endif
%macro ii_rnd 5 ; normal inverse round
mov %1d, ik_ref(%5,0)
mov %2d, ik_ref(%5,1)
mov %3d, ik_ref(%5,2)
mov %4d, ik_ref(%5,3)
movzx esi, al
movzx edi, ah
shr eax, 16
xor %1d, t_ref(0,rsi)
xor %2d, t_ref(1,rdi)
movzx esi, al
movzx edi, ah
xor %3d, t_ref(2,rsi)
xor %4d, t_ref(3,rdi)
movzx esi, bl
movzx edi, bh
shr ebx, 16
xor %2d, t_ref(0,rsi)
xor %3d, t_ref(1,rdi)
movzx esi, bl
movzx edi, bh
xor %4d, t_ref(2,rsi)
xor %1d, t_ref(3,rdi)
movzx esi, cl
movzx edi, ch
shr ecx, 16
xor %3d, t_ref(0,rsi)
xor %4d, t_ref(1,rdi)
movzx esi, cl
movzx edi, ch
xor %1d, t_ref(2,rsi)
xor %2d, t_ref(3,rdi)
movzx esi, dl
movzx edi, dh
shr edx, 16
xor %4d, t_ref(0,rsi)
xor %1d, t_ref(1,rdi)
movzx esi, dl
movzx edi, dh
xor %2d, t_ref(2,rsi)
xor %3d, t_ref(3,rdi)
mov eax,%1d
mov ebx,%2d
mov ecx,%3d
mov edx,%4d
%endmacro
%ifdef LAST_ROUND_TABLES
%macro il_rnd 5 ; last inverse round
add tptr, 2048
mov %1d, ik_ref(%5,0)
mov %2d, ik_ref(%5,1)
mov %3d, ik_ref(%5,2)
mov %4d, ik_ref(%5,3)
movzx esi, al
movzx edi, ah
shr eax, 16
xor %1d, t_ref(0,rsi)
xor %2d, t_ref(1,rdi)
movzx esi, al
movzx edi, ah
xor %3d, t_ref(2,rsi)
xor %4d, t_ref(3,rdi)
movzx esi, bl
movzx edi, bh
shr ebx, 16
xor %2d, t_ref(0,rsi)
xor %3d, t_ref(1,rdi)
movzx esi, bl
movzx edi, bh
xor %4d, t_ref(2,rsi)
xor %1d, t_ref(3,rdi)
movzx esi, cl
movzx edi, ch
shr ecx, 16
xor %3d, t_ref(0,rsi)
xor %4d, t_ref(1,rdi)
movzx esi, cl
movzx edi, ch
xor %1d, t_ref(2,rsi)
xor %2d, t_ref(3,rdi)
movzx esi, dl
movzx edi, dh
shr edx, 16
xor %4d, t_ref(0,rsi)
xor %1d, t_ref(1,rdi)
movzx esi, dl
movzx edi, dh
xor %2d, t_ref(2,rsi)
xor %3d, t_ref(3,rdi)
%endmacro
%else
%macro il_rnd 5 ; last inverse round
mov %1d, ik_ref(%5,0)
mov %2d, ik_ref(%5,1)
mov %3d, ik_ref(%5,2)
mov %4d, ik_ref(%5,3)
movzx esi, al
movzx edi, ah
movzx esi, t_ref(i,rsi)
movzx edi, t_ref(i,rdi)
shr eax, 16
xor %1d, esi
rol edi, 8
xor %2d, edi
movzx esi, al
movzx edi, ah
movzx esi, t_ref(i,rsi)
movzx edi, t_ref(i,rdi)
rol esi, 16
rol edi, 24
xor %3d, esi
xor %4d, edi
movzx esi, bl
movzx edi, bh
movzx esi, t_ref(i,rsi)
movzx edi, t_ref(i,rdi)
shr ebx, 16
xor %2d, esi
rol edi, 8
xor %3d, edi
movzx esi, bl
movzx edi, bh
movzx esi, t_ref(i,rsi)
movzx edi, t_ref(i,rdi)
rol esi, 16
rol edi, 24
xor %4d, esi
xor %1d, edi
movzx esi, cl
movzx edi, ch
movzx esi, t_ref(i,rsi)
movzx edi, t_ref(i,rdi)
shr ecx, 16
xor %3d, esi
rol edi, 8
xor %4d, edi
movzx esi, cl
movzx edi, ch
movzx esi, t_ref(i,rsi)
movzx edi, t_ref(i,rdi)
rol esi, 16
rol edi, 24
xor %1d, esi
xor %2d, edi
movzx esi, dl
movzx edi, dh
movzx esi, t_ref(i,rsi)
movzx edi, t_ref(i,rdi)
shr edx, 16
xor %4d, esi
rol edi, 8
xor %1d, edi
movzx esi, dl
movzx edi, dh
movzx esi, t_ref(i,rsi)
movzx edi, t_ref(i,rdi)
rol esi, 16
rol edi, 24
xor %2d, esi
xor %3d, edi
%endmacro
%endif
%ifdef ENCRYPTION
global aes_encrypt
%ifdef DLL_EXPORT
export aes_encrypt
%endif
section .data align=64
align 64
enc_tab:
enc_vals u8
%ifdef LAST_ROUND_TABLES
enc_vals w8
%endif
section .text align=16
align 16
%ifdef _SEH_
proc_frame aes_encrypt
alloc_stack 7*8 ; 7 to align stack to 16 bytes
save_reg rsi,4*8
save_reg rdi,5*8
save_reg rbx,1*8
save_reg rbp,2*8
save_reg r12,3*8
end_prologue
mov rdi, rcx ; input pointer
mov [rsp+0*8], rdx ; output pointer
%else
aes_encrypt:
%ifdef __GNUC__
sub rsp, 4*8 ; gnu/linux binary interface
mov [rsp+0*8], rsi ; output pointer
mov r8, rdx ; context
%else
sub rsp, 6*8 ; windows binary interface
mov [rsp+4*8], rsi
mov [rsp+5*8], rdi
mov rdi, rcx ; input pointer
mov [rsp+0*8], rdx ; output pointer
%endif
mov [rsp+1*8], rbx ; input pointer in rdi
mov [rsp+2*8], rbp ; output pointer in [rsp]
mov [rsp+3*8], r12 ; context in r8
%endif
movzx esi, byte [kptr+4*KS_LENGTH]
lea tptr, [rel enc_tab]
sub kptr, fofs
mov eax, [rdi+0*4]
mov ebx, [rdi+1*4]
mov ecx, [rdi+2*4]
mov edx, [rdi+3*4]
xor eax, [kptr+fofs]
xor ebx, [kptr+fofs+4]
xor ecx, [kptr+fofs+8]
xor edx, [kptr+fofs+12]
lea kptr,[kptr+rsi]
cmp esi, 10*16
je .3
cmp esi, 12*16
je .2
cmp esi, 14*16
je .1
mov rax, -1
jmp .4
.1: ff_rnd r9, r10, r11, r12, 13
ff_rnd r9, r10, r11, r12, 12
.2: ff_rnd r9, r10, r11, r12, 11
ff_rnd r9, r10, r11, r12, 10
.3: ff_rnd r9, r10, r11, r12, 9
ff_rnd r9, r10, r11, r12, 8
ff_rnd r9, r10, r11, r12, 7
ff_rnd r9, r10, r11, r12, 6
ff_rnd r9, r10, r11, r12, 5
ff_rnd r9, r10, r11, r12, 4
ff_rnd r9, r10, r11, r12, 3
ff_rnd r9, r10, r11, r12, 2
ff_rnd r9, r10, r11, r12, 1
fl_rnd r9, r10, r11, r12, 0
mov rbx, [rsp]
mov [rbx], r9d
mov [rbx+4], r10d
mov [rbx+8], r11d
mov [rbx+12], r12d
xor rax, rax
.4:
mov rbx, [rsp+1*8]
mov rbp, [rsp+2*8]
mov r12, [rsp+3*8]
%ifdef __GNUC__
add rsp, 4*8
ret
%else
mov rsi, [rsp+4*8]
mov rdi, [rsp+5*8]
%ifdef _SEH_
add rsp, 7*8
ret
endproc_frame
%else
add rsp, 6*8
ret
%endif
%endif
%endif
%ifdef DECRYPTION
global aes_decrypt
%ifdef DLL_EXPORT
export aes_decrypt
%endif
section .data
align 64
dec_tab:
dec_vals v8
%ifdef LAST_ROUND_TABLES
dec_vals w8
%endif
section .text
align 16
%ifdef _SEH_
proc_frame aes_decrypt
alloc_stack 7*8 ; 7 to align stack to 16 bytes
save_reg rsi,4*8
save_reg rdi,5*8
save_reg rbx,1*8
save_reg rbp,2*8
save_reg r12,3*8
end_prologue
mov rdi, rcx ; input pointer
mov [rsp+0*8], rdx ; output pointer
%else
aes_decrypt:
%ifdef __GNUC__
sub rsp, 4*8 ; gnu/linux binary interface
mov [rsp+0*8], rsi ; output pointer
mov r8, rdx ; context
%else
sub rsp, 6*8 ; windows binary interface
mov [rsp+4*8], rsi
mov [rsp+5*8], rdi
mov rdi, rcx ; input pointer
mov [rsp+0*8], rdx ; output pointer
%endif
mov [rsp+1*8], rbx ; input pointer in rdi
mov [rsp+2*8], rbp ; output pointer in [rsp]
mov [rsp+3*8], r12 ; context in r8
%endif
movzx esi,byte[kptr+4*KS_LENGTH]
lea tptr, [rel dec_tab]
sub kptr, rofs
mov eax, [rdi+0*4]
mov ebx, [rdi+1*4]
mov ecx, [rdi+2*4]
mov edx, [rdi+3*4]
%ifdef AES_REV_DKS
mov rdi, kptr
lea kptr,[kptr+rsi]
%else
lea rdi,[kptr+rsi]
%endif
xor eax, [rdi+rofs]
xor ebx, [rdi+rofs+4]
xor ecx, [rdi+rofs+8]
xor edx, [rdi+rofs+12]
cmp esi, 10*16
je .3
cmp esi, 12*16
je .2
cmp esi, 14*16
je .1
mov rax, -1
jmp .4
.1: ii_rnd r9, r10, r11, r12, 13
ii_rnd r9, r10, r11, r12, 12
.2: ii_rnd r9, r10, r11, r12, 11
ii_rnd r9, r10, r11, r12, 10
.3: ii_rnd r9, r10, r11, r12, 9
ii_rnd r9, r10, r11, r12, 8
ii_rnd r9, r10, r11, r12, 7
ii_rnd r9, r10, r11, r12, 6
ii_rnd r9, r10, r11, r12, 5
ii_rnd r9, r10, r11, r12, 4
ii_rnd r9, r10, r11, r12, 3
ii_rnd r9, r10, r11, r12, 2
ii_rnd r9, r10, r11, r12, 1
il_rnd r9, r10, r11, r12, 0
mov rbx, [rsp]
mov [rbx], r9d
mov [rbx+4], r10d
mov [rbx+8], r11d
mov [rbx+12], r12d
xor rax, rax
.4: mov rbx, [rsp+1*8]
mov rbp, [rsp+2*8]
mov r12, [rsp+3*8]
%ifdef __GNUC__
add rsp, 4*8
ret
%else
mov rsi, [rsp+4*8]
mov rdi, [rsp+5*8]
%ifdef _SEH_
add rsp, 7*8
ret
endproc_frame
%else
add rsp, 6*8
ret
%endif
%endif
%endif
%ifidn __OUTPUT_FORMAT__,elf
section .note.GNU-stack noalloc noexec nowrite progbits
%endif
%ifidn __OUTPUT_FORMAT__,elf32
section .note.GNU-stack noalloc noexec nowrite progbits
%endif
%ifidn __OUTPUT_FORMAT__,elf64
section .note.GNU-stack noalloc noexec nowrite progbits
%endif
|