1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * Implement AES algorithm in Intel AES-NI instructions.
4 *
5 * The white paper of AES-NI instructions can be downloaded from:
6 *   http://softwarecommunity.intel.com/isn/downloads/intelavx/AES-Instructions-Set_WP.pdf
7 *
8 * Copyright (C) 2008, Intel Corp.
9 *    Author: Huang Ying <ying.huang@intel.com>
10 *            Vinodh Gopal <vinodh.gopal@intel.com>
11 *            Kahraman Akdemir
12 *
13 * Added RFC4106 AES-GCM support for 128-bit keys under the AEAD
14 * interface for 64-bit kernels.
15 *    Authors: Erdinc Ozturk (erdinc.ozturk@intel.com)
16 *             Aidan O'Mahony (aidan.o.mahony@intel.com)
17 *             Adrian Hoban <adrian.hoban@intel.com>
18 *             James Guilford (james.guilford@intel.com)
19 *             Gabriele Paoloni <gabriele.paoloni@intel.com>
20 *             Tadeusz Struk (tadeusz.struk@intel.com)
21 *             Wajdi Feghali (wajdi.k.feghali@intel.com)
22 *    Copyright (c) 2010, Intel Corporation.
23 *
24 * Ported x86_64 version to x86:
25 *    Author: Mathias Krause <minipli@googlemail.com>
26 */
27
28#include <linux/linkage.h>
29#include <asm/frame.h>
30#include <asm/nospec-branch.h>
31
32/*
33 * The following macros are used to move an (un)aligned 16 byte value to/from
34 * an XMM register.  This can done for either FP or integer values, for FP use
35 * movaps (move aligned packed single) or integer use movdqa (move double quad
36 * aligned).  It doesn't make a performance difference which instruction is used
37 * since Nehalem (original Core i7) was released.  However, the movaps is a byte
38 * shorter, so that is the one we'll use for now. (same for unaligned).
39 */
40#define MOVADQ	movaps
41#define MOVUDQ	movups
42
43#ifdef __x86_64__
44
45# constants in mergeable sections, linker can reorder and merge
46.section	.rodata.cst16.POLY, "aM", @progbits, 16
47.align 16
48POLY:   .octa 0xC2000000000000000000000000000001
49.section	.rodata.cst16.TWOONE, "aM", @progbits, 16
50.align 16
51TWOONE: .octa 0x00000001000000000000000000000001
52
53.section	.rodata.cst16.SHUF_MASK, "aM", @progbits, 16
54.align 16
55SHUF_MASK:  .octa 0x000102030405060708090A0B0C0D0E0F
56.section	.rodata.cst16.MASK1, "aM", @progbits, 16
57.align 16
58MASK1:      .octa 0x0000000000000000ffffffffffffffff
59.section	.rodata.cst16.MASK2, "aM", @progbits, 16
60.align 16
61MASK2:      .octa 0xffffffffffffffff0000000000000000
62.section	.rodata.cst16.ONE, "aM", @progbits, 16
63.align 16
64ONE:        .octa 0x00000000000000000000000000000001
65.section	.rodata.cst16.F_MIN_MASK, "aM", @progbits, 16
66.align 16
67F_MIN_MASK: .octa 0xf1f2f3f4f5f6f7f8f9fafbfcfdfeff0
68.section	.rodata.cst16.dec, "aM", @progbits, 16
69.align 16
70dec:        .octa 0x1
71.section	.rodata.cst16.enc, "aM", @progbits, 16
72.align 16
73enc:        .octa 0x2
74
75# order of these constants should not change.
76# more specifically, ALL_F should follow SHIFT_MASK,
77# and zero should follow ALL_F
78.section	.rodata, "a", @progbits
79.align 16
80SHIFT_MASK: .octa 0x0f0e0d0c0b0a09080706050403020100
81ALL_F:      .octa 0xffffffffffffffffffffffffffffffff
82            .octa 0x00000000000000000000000000000000
83
84.text
85
86
87#define	STACK_OFFSET    8*3
88
89#define AadHash 16*0
90#define AadLen 16*1
91#define InLen (16*1)+8
92#define PBlockEncKey 16*2
93#define OrigIV 16*3
94#define CurCount 16*4
95#define PBlockLen 16*5
96#define	HashKey		16*6	// store HashKey <<1 mod poly here
97#define	HashKey_2	16*7	// store HashKey^2 <<1 mod poly here
98#define	HashKey_3	16*8	// store HashKey^3 <<1 mod poly here
99#define	HashKey_4	16*9	// store HashKey^4 <<1 mod poly here
100#define	HashKey_k	16*10	// store XOR of High 64 bits and Low 64
101				// bits of  HashKey <<1 mod poly here
102				//(for Karatsuba purposes)
103#define	HashKey_2_k	16*11	// store XOR of High 64 bits and Low 64
104				// bits of  HashKey^2 <<1 mod poly here
105				// (for Karatsuba purposes)
106#define	HashKey_3_k	16*12	// store XOR of High 64 bits and Low 64
107				// bits of  HashKey^3 <<1 mod poly here
108				// (for Karatsuba purposes)
109#define	HashKey_4_k	16*13	// store XOR of High 64 bits and Low 64
110				// bits of  HashKey^4 <<1 mod poly here
111				// (for Karatsuba purposes)
112
113#define arg1 rdi
114#define arg2 rsi
115#define arg3 rdx
116#define arg4 rcx
117#define arg5 r8
118#define arg6 r9
119#define arg7 STACK_OFFSET+8(%rsp)
120#define arg8 STACK_OFFSET+16(%rsp)
121#define arg9 STACK_OFFSET+24(%rsp)
122#define arg10 STACK_OFFSET+32(%rsp)
123#define arg11 STACK_OFFSET+40(%rsp)
124#define keysize 2*15*16(%arg1)
125#endif
126
127
128#define STATE1	%xmm0
129#define STATE2	%xmm4
130#define STATE3	%xmm5
131#define STATE4	%xmm6
132#define STATE	STATE1
133#define IN1	%xmm1
134#define IN2	%xmm7
135#define IN3	%xmm8
136#define IN4	%xmm9
137#define IN	IN1
138#define KEY	%xmm2
139#define IV	%xmm3
140
141#define BSWAP_MASK %xmm10
142#define CTR	%xmm11
143#define INC	%xmm12
144
145#define GF128MUL_MASK %xmm7
146
147#ifdef __x86_64__
148#define AREG	%rax
149#define KEYP	%rdi
150#define OUTP	%rsi
151#define UKEYP	OUTP
152#define INP	%rdx
153#define LEN	%rcx
154#define IVP	%r8
155#define KLEN	%r9d
156#define T1	%r10
157#define TKEYP	T1
158#define T2	%r11
159#define TCTR_LOW T2
160#else
161#define AREG	%eax
162#define KEYP	%edi
163#define OUTP	AREG
164#define UKEYP	OUTP
165#define INP	%edx
166#define LEN	%esi
167#define IVP	%ebp
168#define KLEN	%ebx
169#define T1	%ecx
170#define TKEYP	T1
171#endif
172
173.macro FUNC_SAVE
174	push	%r12
175	push	%r13
176	push	%r14
177#
178# states of %xmm registers %xmm6:%xmm15 not saved
179# all %xmm registers are clobbered
180#
181.endm
182
183
184.macro FUNC_RESTORE
185	pop	%r14
186	pop	%r13
187	pop	%r12
188.endm
189
190# Precompute hashkeys.
191# Input: Hash subkey.
192# Output: HashKeys stored in gcm_context_data.  Only needs to be called
193# once per key.
194# clobbers r12, and tmp xmm registers.
195.macro PRECOMPUTE SUBKEY TMP1 TMP2 TMP3 TMP4 TMP5 TMP6 TMP7
196	mov	\SUBKEY, %r12
197	movdqu	(%r12), \TMP3
198	movdqa	SHUF_MASK(%rip), \TMP2
199	pshufb	\TMP2, \TMP3
200
201	# precompute HashKey<<1 mod poly from the HashKey (required for GHASH)
202
203	movdqa	\TMP3, \TMP2
204	psllq	$1, \TMP3
205	psrlq	$63, \TMP2
206	movdqa	\TMP2, \TMP1
207	pslldq	$8, \TMP2
208	psrldq	$8, \TMP1
209	por	\TMP2, \TMP3
210
211	# reduce HashKey<<1
212
213	pshufd	$0x24, \TMP1, \TMP2
214	pcmpeqd TWOONE(%rip), \TMP2
215	pand	POLY(%rip), \TMP2
216	pxor	\TMP2, \TMP3
217	movdqu	\TMP3, HashKey(%arg2)
218
219	movdqa	   \TMP3, \TMP5
220	pshufd	   $78, \TMP3, \TMP1
221	pxor	   \TMP3, \TMP1
222	movdqu	   \TMP1, HashKey_k(%arg2)
223
224	GHASH_MUL  \TMP5, \TMP3, \TMP1, \TMP2, \TMP4, \TMP6, \TMP7
225# TMP5 = HashKey^2<<1 (mod poly)
226	movdqu	   \TMP5, HashKey_2(%arg2)
227# HashKey_2 = HashKey^2<<1 (mod poly)
228	pshufd	   $78, \TMP5, \TMP1
229	pxor	   \TMP5, \TMP1
230	movdqu	   \TMP1, HashKey_2_k(%arg2)
231
232	GHASH_MUL  \TMP5, \TMP3, \TMP1, \TMP2, \TMP4, \TMP6, \TMP7
233# TMP5 = HashKey^3<<1 (mod poly)
234	movdqu	   \TMP5, HashKey_3(%arg2)
235	pshufd	   $78, \TMP5, \TMP1
236	pxor	   \TMP5, \TMP1
237	movdqu	   \TMP1, HashKey_3_k(%arg2)
238
239	GHASH_MUL  \TMP5, \TMP3, \TMP1, \TMP2, \TMP4, \TMP6, \TMP7
240# TMP5 = HashKey^3<<1 (mod poly)
241	movdqu	   \TMP5, HashKey_4(%arg2)
242	pshufd	   $78, \TMP5, \TMP1
243	pxor	   \TMP5, \TMP1
244	movdqu	   \TMP1, HashKey_4_k(%arg2)
245.endm
246
247# GCM_INIT initializes a gcm_context struct to prepare for encoding/decoding.
248# Clobbers rax, r10-r13 and xmm0-xmm6, %xmm13
249.macro GCM_INIT Iv SUBKEY AAD AADLEN
250	mov \AADLEN, %r11
251	mov %r11, AadLen(%arg2) # ctx_data.aad_length = aad_length
252	xor %r11d, %r11d
253	mov %r11, InLen(%arg2) # ctx_data.in_length = 0
254	mov %r11, PBlockLen(%arg2) # ctx_data.partial_block_length = 0
255	mov %r11, PBlockEncKey(%arg2) # ctx_data.partial_block_enc_key = 0
256	mov \Iv, %rax
257	movdqu (%rax), %xmm0
258	movdqu %xmm0, OrigIV(%arg2) # ctx_data.orig_IV = iv
259
260	movdqa  SHUF_MASK(%rip), %xmm2
261	pshufb %xmm2, %xmm0
262	movdqu %xmm0, CurCount(%arg2) # ctx_data.current_counter = iv
263
264	PRECOMPUTE \SUBKEY, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7
265	movdqu HashKey(%arg2), %xmm13
266
267	CALC_AAD_HASH %xmm13, \AAD, \AADLEN, %xmm0, %xmm1, %xmm2, %xmm3, \
268	%xmm4, %xmm5, %xmm6
269.endm
270
271# GCM_ENC_DEC Encodes/Decodes given data. Assumes that the passed gcm_context
272# struct has been initialized by GCM_INIT.
273# Requires the input data be at least 1 byte long because of READ_PARTIAL_BLOCK
274# Clobbers rax, r10-r13, and xmm0-xmm15
275.macro GCM_ENC_DEC operation
276	movdqu AadHash(%arg2), %xmm8
277	movdqu HashKey(%arg2), %xmm13
278	add %arg5, InLen(%arg2)
279
280	xor %r11d, %r11d # initialise the data pointer offset as zero
281	PARTIAL_BLOCK %arg3 %arg4 %arg5 %r11 %xmm8 \operation
282
283	sub %r11, %arg5		# sub partial block data used
284	mov %arg5, %r13		# save the number of bytes
285
286	and $-16, %r13		# %r13 = %r13 - (%r13 mod 16)
287	mov %r13, %r12
288	# Encrypt/Decrypt first few blocks
289
290	and	$(3<<4), %r12
291	jz	.L_initial_num_blocks_is_0_\@
292	cmp	$(2<<4), %r12
293	jb	.L_initial_num_blocks_is_1_\@
294	je	.L_initial_num_blocks_is_2_\@
295.L_initial_num_blocks_is_3_\@:
296	INITIAL_BLOCKS_ENC_DEC	%xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
297%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 5, 678, \operation
298	sub	$48, %r13
299	jmp	.L_initial_blocks_\@
300.L_initial_num_blocks_is_2_\@:
301	INITIAL_BLOCKS_ENC_DEC	%xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
302%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 6, 78, \operation
303	sub	$32, %r13
304	jmp	.L_initial_blocks_\@
305.L_initial_num_blocks_is_1_\@:
306	INITIAL_BLOCKS_ENC_DEC	%xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
307%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 7, 8, \operation
308	sub	$16, %r13
309	jmp	.L_initial_blocks_\@
310.L_initial_num_blocks_is_0_\@:
311	INITIAL_BLOCKS_ENC_DEC	%xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
312%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 8, 0, \operation
313.L_initial_blocks_\@:
314
315	# Main loop - Encrypt/Decrypt remaining blocks
316
317	test	%r13, %r13
318	je	.L_zero_cipher_left_\@
319	sub	$64, %r13
320	je	.L_four_cipher_left_\@
321.L_crypt_by_4_\@:
322	GHASH_4_ENCRYPT_4_PARALLEL_\operation	%xmm9, %xmm10, %xmm11, %xmm12, \
323	%xmm13, %xmm14, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, \
324	%xmm7, %xmm8, enc
325	add	$64, %r11
326	sub	$64, %r13
327	jne	.L_crypt_by_4_\@
328.L_four_cipher_left_\@:
329	GHASH_LAST_4	%xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14, \
330%xmm15, %xmm1, %xmm2, %xmm3, %xmm4, %xmm8
331.L_zero_cipher_left_\@:
332	movdqu %xmm8, AadHash(%arg2)
333	movdqu %xmm0, CurCount(%arg2)
334
335	mov	%arg5, %r13
336	and	$15, %r13			# %r13 = arg5 (mod 16)
337	je	.L_multiple_of_16_bytes_\@
338
339	mov %r13, PBlockLen(%arg2)
340
341	# Handle the last <16 Byte block separately
342	paddd ONE(%rip), %xmm0                # INCR CNT to get Yn
343	movdqu %xmm0, CurCount(%arg2)
344	movdqa SHUF_MASK(%rip), %xmm10
345	pshufb %xmm10, %xmm0
346
347	ENCRYPT_SINGLE_BLOCK	%xmm0, %xmm1        # Encrypt(K, Yn)
348	movdqu %xmm0, PBlockEncKey(%arg2)
349
350	cmp	$16, %arg5
351	jge	.L_large_enough_update_\@
352
353	lea (%arg4,%r11,1), %r10
354	mov %r13, %r12
355	READ_PARTIAL_BLOCK %r10 %r12 %xmm2 %xmm1
356	jmp	.L_data_read_\@
357
358.L_large_enough_update_\@:
359	sub	$16, %r11
360	add	%r13, %r11
361
362	# receive the last <16 Byte block
363	movdqu	(%arg4, %r11, 1), %xmm1
364
365	sub	%r13, %r11
366	add	$16, %r11
367
368	lea	SHIFT_MASK+16(%rip), %r12
369	# adjust the shuffle mask pointer to be able to shift 16-r13 bytes
370	# (r13 is the number of bytes in plaintext mod 16)
371	sub	%r13, %r12
372	# get the appropriate shuffle mask
373	movdqu	(%r12), %xmm2
374	# shift right 16-r13 bytes
375	pshufb  %xmm2, %xmm1
376
377.L_data_read_\@:
378	lea ALL_F+16(%rip), %r12
379	sub %r13, %r12
380
381.ifc \operation, dec
382	movdqa  %xmm1, %xmm2
383.endif
384	pxor	%xmm1, %xmm0            # XOR Encrypt(K, Yn)
385	movdqu	(%r12), %xmm1
386	# get the appropriate mask to mask out top 16-r13 bytes of xmm0
387	pand	%xmm1, %xmm0            # mask out top 16-r13 bytes of xmm0
388.ifc \operation, dec
389	pand    %xmm1, %xmm2
390	movdqa SHUF_MASK(%rip), %xmm10
391	pshufb %xmm10 ,%xmm2
392
393	pxor %xmm2, %xmm8
394.else
395	movdqa SHUF_MASK(%rip), %xmm10
396	pshufb %xmm10,%xmm0
397
398	pxor	%xmm0, %xmm8
399.endif
400
401	movdqu %xmm8, AadHash(%arg2)
402.ifc \operation, enc
403	# GHASH computation for the last <16 byte block
404	movdqa SHUF_MASK(%rip), %xmm10
405	# shuffle xmm0 back to output as ciphertext
406	pshufb %xmm10, %xmm0
407.endif
408
409	# Output %r13 bytes
410	movq %xmm0, %rax
411	cmp $8, %r13
412	jle .L_less_than_8_bytes_left_\@
413	mov %rax, (%arg3 , %r11, 1)
414	add $8, %r11
415	psrldq $8, %xmm0
416	movq %xmm0, %rax
417	sub $8, %r13
418.L_less_than_8_bytes_left_\@:
419	mov %al,  (%arg3, %r11, 1)
420	add $1, %r11
421	shr $8, %rax
422	sub $1, %r13
423	jne .L_less_than_8_bytes_left_\@
424.L_multiple_of_16_bytes_\@:
425.endm
426
427# GCM_COMPLETE Finishes update of tag of last partial block
428# Output: Authorization Tag (AUTH_TAG)
429# Clobbers rax, r10-r12, and xmm0, xmm1, xmm5-xmm15
430.macro GCM_COMPLETE AUTHTAG AUTHTAGLEN
431	movdqu AadHash(%arg2), %xmm8
432	movdqu HashKey(%arg2), %xmm13
433
434	mov PBlockLen(%arg2), %r12
435
436	test %r12, %r12
437	je .L_partial_done\@
438
439	GHASH_MUL %xmm8, %xmm13, %xmm9, %xmm10, %xmm11, %xmm5, %xmm6
440
441.L_partial_done\@:
442	mov AadLen(%arg2), %r12  # %r13 = aadLen (number of bytes)
443	shl	$3, %r12		  # convert into number of bits
444	movd	%r12d, %xmm15		  # len(A) in %xmm15
445	mov InLen(%arg2), %r12
446	shl     $3, %r12                  # len(C) in bits (*128)
447	movq    %r12, %xmm1
448
449	pslldq	$8, %xmm15		  # %xmm15 = len(A)||0x0000000000000000
450	pxor	%xmm1, %xmm15		  # %xmm15 = len(A)||len(C)
451	pxor	%xmm15, %xmm8
452	GHASH_MUL	%xmm8, %xmm13, %xmm9, %xmm10, %xmm11, %xmm5, %xmm6
453	# final GHASH computation
454	movdqa SHUF_MASK(%rip), %xmm10
455	pshufb %xmm10, %xmm8
456
457	movdqu OrigIV(%arg2), %xmm0       # %xmm0 = Y0
458	ENCRYPT_SINGLE_BLOCK	%xmm0,  %xmm1	  # E(K, Y0)
459	pxor	%xmm8, %xmm0
460.L_return_T_\@:
461	mov	\AUTHTAG, %r10                     # %r10 = authTag
462	mov	\AUTHTAGLEN, %r11                    # %r11 = auth_tag_len
463	cmp	$16, %r11
464	je	.L_T_16_\@
465	cmp	$8, %r11
466	jl	.L_T_4_\@
467.L_T_8_\@:
468	movq	%xmm0, %rax
469	mov	%rax, (%r10)
470	add	$8, %r10
471	sub	$8, %r11
472	psrldq	$8, %xmm0
473	test	%r11, %r11
474	je	.L_return_T_done_\@
475.L_T_4_\@:
476	movd	%xmm0, %eax
477	mov	%eax, (%r10)
478	add	$4, %r10
479	sub	$4, %r11
480	psrldq	$4, %xmm0
481	test	%r11, %r11
482	je	.L_return_T_done_\@
483.L_T_123_\@:
484	movd	%xmm0, %eax
485	cmp	$2, %r11
486	jl	.L_T_1_\@
487	mov	%ax, (%r10)
488	cmp	$2, %r11
489	je	.L_return_T_done_\@
490	add	$2, %r10
491	sar	$16, %eax
492.L_T_1_\@:
493	mov	%al, (%r10)
494	jmp	.L_return_T_done_\@
495.L_T_16_\@:
496	movdqu	%xmm0, (%r10)
497.L_return_T_done_\@:
498.endm
499
500#ifdef __x86_64__
501/* GHASH_MUL MACRO to implement: Data*HashKey mod (128,127,126,121,0)
502*
503*
504* Input: A and B (128-bits each, bit-reflected)
505* Output: C = A*B*x mod poly, (i.e. >>1 )
506* To compute GH = GH*HashKey mod poly, give HK = HashKey<<1 mod poly as input
507* GH = GH * HK * x mod poly which is equivalent to GH*HashKey mod poly.
508*
509*/
510.macro GHASH_MUL GH HK TMP1 TMP2 TMP3 TMP4 TMP5
511	movdqa	  \GH, \TMP1
512	pshufd	  $78, \GH, \TMP2
513	pshufd	  $78, \HK, \TMP3
514	pxor	  \GH, \TMP2            # TMP2 = a1+a0
515	pxor	  \HK, \TMP3            # TMP3 = b1+b0
516	pclmulqdq $0x11, \HK, \TMP1     # TMP1 = a1*b1
517	pclmulqdq $0x00, \HK, \GH       # GH = a0*b0
518	pclmulqdq $0x00, \TMP3, \TMP2   # TMP2 = (a0+a1)*(b1+b0)
519	pxor	  \GH, \TMP2
520	pxor	  \TMP1, \TMP2          # TMP2 = (a0*b0)+(a1*b0)
521	movdqa	  \TMP2, \TMP3
522	pslldq	  $8, \TMP3             # left shift TMP3 2 DWs
523	psrldq	  $8, \TMP2             # right shift TMP2 2 DWs
524	pxor	  \TMP3, \GH
525	pxor	  \TMP2, \TMP1          # TMP2:GH holds the result of GH*HK
526
527        # first phase of the reduction
528
529	movdqa    \GH, \TMP2
530	movdqa    \GH, \TMP3
531	movdqa    \GH, \TMP4            # copy GH into TMP2,TMP3 and TMP4
532					# in in order to perform
533					# independent shifts
534	pslld     $31, \TMP2            # packed right shift <<31
535	pslld     $30, \TMP3            # packed right shift <<30
536	pslld     $25, \TMP4            # packed right shift <<25
537	pxor      \TMP3, \TMP2          # xor the shifted versions
538	pxor      \TMP4, \TMP2
539	movdqa    \TMP2, \TMP5
540	psrldq    $4, \TMP5             # right shift TMP5 1 DW
541	pslldq    $12, \TMP2            # left shift TMP2 3 DWs
542	pxor      \TMP2, \GH
543
544        # second phase of the reduction
545
546	movdqa    \GH,\TMP2             # copy GH into TMP2,TMP3 and TMP4
547					# in in order to perform
548					# independent shifts
549	movdqa    \GH,\TMP3
550	movdqa    \GH,\TMP4
551	psrld     $1,\TMP2              # packed left shift >>1
552	psrld     $2,\TMP3              # packed left shift >>2
553	psrld     $7,\TMP4              # packed left shift >>7
554	pxor      \TMP3,\TMP2		# xor the shifted versions
555	pxor      \TMP4,\TMP2
556	pxor      \TMP5, \TMP2
557	pxor      \TMP2, \GH
558	pxor      \TMP1, \GH            # result is in TMP1
559.endm
560
561# Reads DLEN bytes starting at DPTR and stores in XMMDst
562# where 0 < DLEN < 16
563# Clobbers %rax, DLEN and XMM1
564.macro READ_PARTIAL_BLOCK DPTR DLEN XMM1 XMMDst
565        cmp $8, \DLEN
566        jl .L_read_lt8_\@
567        mov (\DPTR), %rax
568        movq %rax, \XMMDst
569        sub $8, \DLEN
570        jz .L_done_read_partial_block_\@
571	xor %eax, %eax
572.L_read_next_byte_\@:
573        shl $8, %rax
574        mov 7(\DPTR, \DLEN, 1), %al
575        dec \DLEN
576        jnz .L_read_next_byte_\@
577        movq %rax, \XMM1
578	pslldq $8, \XMM1
579        por \XMM1, \XMMDst
580	jmp .L_done_read_partial_block_\@
581.L_read_lt8_\@:
582	xor %eax, %eax
583.L_read_next_byte_lt8_\@:
584        shl $8, %rax
585        mov -1(\DPTR, \DLEN, 1), %al
586        dec \DLEN
587        jnz .L_read_next_byte_lt8_\@
588        movq %rax, \XMMDst
589.L_done_read_partial_block_\@:
590.endm
591
592# CALC_AAD_HASH: Calculates the hash of the data which will not be encrypted.
593# clobbers r10-11, xmm14
594.macro CALC_AAD_HASH HASHKEY AAD AADLEN TMP1 TMP2 TMP3 TMP4 TMP5 \
595	TMP6 TMP7
596	MOVADQ	   SHUF_MASK(%rip), %xmm14
597	mov	   \AAD, %r10		# %r10 = AAD
598	mov	   \AADLEN, %r11		# %r11 = aadLen
599	pxor	   \TMP7, \TMP7
600	pxor	   \TMP6, \TMP6
601
602	cmp	   $16, %r11
603	jl	   .L_get_AAD_rest\@
604.L_get_AAD_blocks\@:
605	movdqu	   (%r10), \TMP7
606	pshufb	   %xmm14, \TMP7 # byte-reflect the AAD data
607	pxor	   \TMP7, \TMP6
608	GHASH_MUL  \TMP6, \HASHKEY, \TMP1, \TMP2, \TMP3, \TMP4, \TMP5
609	add	   $16, %r10
610	sub	   $16, %r11
611	cmp	   $16, %r11
612	jge	   .L_get_AAD_blocks\@
613
614	movdqu	   \TMP6, \TMP7
615
616	/* read the last <16B of AAD */
617.L_get_AAD_rest\@:
618	test	   %r11, %r11
619	je	   .L_get_AAD_done\@
620
621	READ_PARTIAL_BLOCK %r10, %r11, \TMP1, \TMP7
622	pshufb	   %xmm14, \TMP7 # byte-reflect the AAD data
623	pxor	   \TMP6, \TMP7
624	GHASH_MUL  \TMP7, \HASHKEY, \TMP1, \TMP2, \TMP3, \TMP4, \TMP5
625	movdqu \TMP7, \TMP6
626
627.L_get_AAD_done\@:
628	movdqu \TMP6, AadHash(%arg2)
629.endm
630
631# PARTIAL_BLOCK: Handles encryption/decryption and the tag partial blocks
632# between update calls.
633# Requires the input data be at least 1 byte long due to READ_PARTIAL_BLOCK
634# Outputs encrypted bytes, and updates hash and partial info in gcm_data_context
635# Clobbers rax, r10, r12, r13, xmm0-6, xmm9-13
636.macro PARTIAL_BLOCK CYPH_PLAIN_OUT PLAIN_CYPH_IN PLAIN_CYPH_LEN DATA_OFFSET \
637	AAD_HASH operation
638	mov 	PBlockLen(%arg2), %r13
639	test	%r13, %r13
640	je	.L_partial_block_done_\@	# Leave Macro if no partial blocks
641	# Read in input data without over reading
642	cmp	$16, \PLAIN_CYPH_LEN
643	jl	.L_fewer_than_16_bytes_\@
644	movups	(\PLAIN_CYPH_IN), %xmm1	# If more than 16 bytes, just fill xmm
645	jmp	.L_data_read_\@
646
647.L_fewer_than_16_bytes_\@:
648	lea	(\PLAIN_CYPH_IN, \DATA_OFFSET, 1), %r10
649	mov	\PLAIN_CYPH_LEN, %r12
650	READ_PARTIAL_BLOCK %r10 %r12 %xmm0 %xmm1
651
652	mov PBlockLen(%arg2), %r13
653
654.L_data_read_\@:				# Finished reading in data
655
656	movdqu	PBlockEncKey(%arg2), %xmm9
657	movdqu	HashKey(%arg2), %xmm13
658
659	lea	SHIFT_MASK(%rip), %r12
660
661	# adjust the shuffle mask pointer to be able to shift r13 bytes
662	# r16-r13 is the number of bytes in plaintext mod 16)
663	add	%r13, %r12
664	movdqu	(%r12), %xmm2		# get the appropriate shuffle mask
665	pshufb	%xmm2, %xmm9		# shift right r13 bytes
666
667.ifc \operation, dec
668	movdqa	%xmm1, %xmm3
669	pxor	%xmm1, %xmm9		# Cyphertext XOR E(K, Yn)
670
671	mov	\PLAIN_CYPH_LEN, %r10
672	add	%r13, %r10
673	# Set r10 to be the amount of data left in CYPH_PLAIN_IN after filling
674	sub	$16, %r10
675	# Determine if if partial block is not being filled and
676	# shift mask accordingly
677	jge	.L_no_extra_mask_1_\@
678	sub	%r10, %r12
679.L_no_extra_mask_1_\@:
680
681	movdqu	ALL_F-SHIFT_MASK(%r12), %xmm1
682	# get the appropriate mask to mask out bottom r13 bytes of xmm9
683	pand	%xmm1, %xmm9		# mask out bottom r13 bytes of xmm9
684
685	pand	%xmm1, %xmm3
686	movdqa	SHUF_MASK(%rip), %xmm10
687	pshufb	%xmm10, %xmm3
688	pshufb	%xmm2, %xmm3
689	pxor	%xmm3, \AAD_HASH
690
691	test	%r10, %r10
692	jl	.L_partial_incomplete_1_\@
693
694	# GHASH computation for the last <16 Byte block
695	GHASH_MUL \AAD_HASH, %xmm13, %xmm0, %xmm10, %xmm11, %xmm5, %xmm6
696	xor	%eax, %eax
697
698	mov	%rax, PBlockLen(%arg2)
699	jmp	.L_dec_done_\@
700.L_partial_incomplete_1_\@:
701	add	\PLAIN_CYPH_LEN, PBlockLen(%arg2)
702.L_dec_done_\@:
703	movdqu	\AAD_HASH, AadHash(%arg2)
704.else
705	pxor	%xmm1, %xmm9			# Plaintext XOR E(K, Yn)
706
707	mov	\PLAIN_CYPH_LEN, %r10
708	add	%r13, %r10
709	# Set r10 to be the amount of data left in CYPH_PLAIN_IN after filling
710	sub	$16, %r10
711	# Determine if if partial block is not being filled and
712	# shift mask accordingly
713	jge	.L_no_extra_mask_2_\@
714	sub	%r10, %r12
715.L_no_extra_mask_2_\@:
716
717	movdqu	ALL_F-SHIFT_MASK(%r12), %xmm1
718	# get the appropriate mask to mask out bottom r13 bytes of xmm9
719	pand	%xmm1, %xmm9
720
721	movdqa	SHUF_MASK(%rip), %xmm1
722	pshufb	%xmm1, %xmm9
723	pshufb	%xmm2, %xmm9
724	pxor	%xmm9, \AAD_HASH
725
726	test	%r10, %r10
727	jl	.L_partial_incomplete_2_\@
728
729	# GHASH computation for the last <16 Byte block
730	GHASH_MUL \AAD_HASH, %xmm13, %xmm0, %xmm10, %xmm11, %xmm5, %xmm6
731	xor	%eax, %eax
732
733	mov	%rax, PBlockLen(%arg2)
734	jmp	.L_encode_done_\@
735.L_partial_incomplete_2_\@:
736	add	\PLAIN_CYPH_LEN, PBlockLen(%arg2)
737.L_encode_done_\@:
738	movdqu	\AAD_HASH, AadHash(%arg2)
739
740	movdqa	SHUF_MASK(%rip), %xmm10
741	# shuffle xmm9 back to output as ciphertext
742	pshufb	%xmm10, %xmm9
743	pshufb	%xmm2, %xmm9
744.endif
745	# output encrypted Bytes
746	test	%r10, %r10
747	jl	.L_partial_fill_\@
748	mov	%r13, %r12
749	mov	$16, %r13
750	# Set r13 to be the number of bytes to write out
751	sub	%r12, %r13
752	jmp	.L_count_set_\@
753.L_partial_fill_\@:
754	mov	\PLAIN_CYPH_LEN, %r13
755.L_count_set_\@:
756	movdqa	%xmm9, %xmm0
757	movq	%xmm0, %rax
758	cmp	$8, %r13
759	jle	.L_less_than_8_bytes_left_\@
760
761	mov	%rax, (\CYPH_PLAIN_OUT, \DATA_OFFSET, 1)
762	add	$8, \DATA_OFFSET
763	psrldq	$8, %xmm0
764	movq	%xmm0, %rax
765	sub	$8, %r13
766.L_less_than_8_bytes_left_\@:
767	movb	%al, (\CYPH_PLAIN_OUT, \DATA_OFFSET, 1)
768	add	$1, \DATA_OFFSET
769	shr	$8, %rax
770	sub	$1, %r13
771	jne	.L_less_than_8_bytes_left_\@
772.L_partial_block_done_\@:
773.endm # PARTIAL_BLOCK
774
775/*
776* if a = number of total plaintext bytes
777* b = floor(a/16)
778* num_initial_blocks = b mod 4
779* encrypt the initial num_initial_blocks blocks and apply ghash on
780* the ciphertext
781* %r10, %r11, %r12, %rax, %xmm5, %xmm6, %xmm7, %xmm8, %xmm9 registers
782* are clobbered
783* arg1, %arg2, %arg3 are used as a pointer only, not modified
784*/
785
786
787.macro INITIAL_BLOCKS_ENC_DEC TMP1 TMP2 TMP3 TMP4 TMP5 XMM0 XMM1 \
788	XMM2 XMM3 XMM4 XMMDst TMP6 TMP7 i i_seq operation
789	MOVADQ		SHUF_MASK(%rip), %xmm14
790
791	movdqu AadHash(%arg2), %xmm\i		    # XMM0 = Y0
792
793	# start AES for num_initial_blocks blocks
794
795	movdqu CurCount(%arg2), \XMM0                # XMM0 = Y0
796
797.if (\i == 5) || (\i == 6) || (\i == 7)
798
799	MOVADQ		ONE(%RIP),\TMP1
800	MOVADQ		0(%arg1),\TMP2
801.irpc index, \i_seq
802	paddd		\TMP1, \XMM0                 # INCR Y0
803.ifc \operation, dec
804        movdqa     \XMM0, %xmm\index
805.else
806	MOVADQ		\XMM0, %xmm\index
807.endif
808	pshufb	%xmm14, %xmm\index      # perform a 16 byte swap
809	pxor		\TMP2, %xmm\index
810.endr
811	lea	0x10(%arg1),%r10
812	mov	keysize,%eax
813	shr	$2,%eax				# 128->4, 192->6, 256->8
814	add	$5,%eax			      # 128->9, 192->11, 256->13
815
816.Laes_loop_initial_\@:
817	MOVADQ	(%r10),\TMP1
818.irpc	index, \i_seq
819	aesenc	\TMP1, %xmm\index
820.endr
821	add	$16,%r10
822	sub	$1,%eax
823	jnz	.Laes_loop_initial_\@
824
825	MOVADQ	(%r10), \TMP1
826.irpc index, \i_seq
827	aesenclast \TMP1, %xmm\index         # Last Round
828.endr
829.irpc index, \i_seq
830	movdqu	   (%arg4 , %r11, 1), \TMP1
831	pxor	   \TMP1, %xmm\index
832	movdqu	   %xmm\index, (%arg3 , %r11, 1)
833	# write back plaintext/ciphertext for num_initial_blocks
834	add	   $16, %r11
835
836.ifc \operation, dec
837	movdqa     \TMP1, %xmm\index
838.endif
839	pshufb	   %xmm14, %xmm\index
840
841		# prepare plaintext/ciphertext for GHASH computation
842.endr
843.endif
844
845        # apply GHASH on num_initial_blocks blocks
846
847.if \i == 5
848        pxor       %xmm5, %xmm6
849	GHASH_MUL  %xmm6, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
850        pxor       %xmm6, %xmm7
851	GHASH_MUL  %xmm7, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
852        pxor       %xmm7, %xmm8
853	GHASH_MUL  %xmm8, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
854.elseif \i == 6
855        pxor       %xmm6, %xmm7
856	GHASH_MUL  %xmm7, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
857        pxor       %xmm7, %xmm8
858	GHASH_MUL  %xmm8, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
859.elseif \i == 7
860        pxor       %xmm7, %xmm8
861	GHASH_MUL  %xmm8, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
862.endif
863	cmp	   $64, %r13
864	jl	.L_initial_blocks_done\@
865	# no need for precomputed values
866/*
867*
868* Precomputations for HashKey parallel with encryption of first 4 blocks.
869* Haskey_i_k holds XORed values of the low and high parts of the Haskey_i
870*/
871	MOVADQ	   ONE(%RIP),\TMP1
872	paddd	   \TMP1, \XMM0              # INCR Y0
873	MOVADQ	   \XMM0, \XMM1
874	pshufb  %xmm14, \XMM1        # perform a 16 byte swap
875
876	paddd	   \TMP1, \XMM0              # INCR Y0
877	MOVADQ	   \XMM0, \XMM2
878	pshufb  %xmm14, \XMM2        # perform a 16 byte swap
879
880	paddd	   \TMP1, \XMM0              # INCR Y0
881	MOVADQ	   \XMM0, \XMM3
882	pshufb %xmm14, \XMM3        # perform a 16 byte swap
883
884	paddd	   \TMP1, \XMM0              # INCR Y0
885	MOVADQ	   \XMM0, \XMM4
886	pshufb %xmm14, \XMM4        # perform a 16 byte swap
887
888	MOVADQ	   0(%arg1),\TMP1
889	pxor	   \TMP1, \XMM1
890	pxor	   \TMP1, \XMM2
891	pxor	   \TMP1, \XMM3
892	pxor	   \TMP1, \XMM4
893.irpc index, 1234 # do 4 rounds
894	movaps 0x10*\index(%arg1), \TMP1
895	aesenc	   \TMP1, \XMM1
896	aesenc	   \TMP1, \XMM2
897	aesenc	   \TMP1, \XMM3
898	aesenc	   \TMP1, \XMM4
899.endr
900.irpc index, 56789 # do next 5 rounds
901	movaps 0x10*\index(%arg1), \TMP1
902	aesenc	   \TMP1, \XMM1
903	aesenc	   \TMP1, \XMM2
904	aesenc	   \TMP1, \XMM3
905	aesenc	   \TMP1, \XMM4
906.endr
907	lea	   0xa0(%arg1),%r10
908	mov	   keysize,%eax
909	shr	   $2,%eax			# 128->4, 192->6, 256->8
910	sub	   $4,%eax			# 128->0, 192->2, 256->4
911	jz	   .Laes_loop_pre_done\@
912
913.Laes_loop_pre_\@:
914	MOVADQ	   (%r10),\TMP2
915.irpc	index, 1234
916	aesenc	   \TMP2, %xmm\index
917.endr
918	add	   $16,%r10
919	sub	   $1,%eax
920	jnz	   .Laes_loop_pre_\@
921
922.Laes_loop_pre_done\@:
923	MOVADQ	   (%r10), \TMP2
924	aesenclast \TMP2, \XMM1
925	aesenclast \TMP2, \XMM2
926	aesenclast \TMP2, \XMM3
927	aesenclast \TMP2, \XMM4
928	movdqu	   16*0(%arg4 , %r11 , 1), \TMP1
929	pxor	   \TMP1, \XMM1
930.ifc \operation, dec
931	movdqu     \XMM1, 16*0(%arg3 , %r11 , 1)
932	movdqa     \TMP1, \XMM1
933.endif
934	movdqu	   16*1(%arg4 , %r11 , 1), \TMP1
935	pxor	   \TMP1, \XMM2
936.ifc \operation, dec
937	movdqu     \XMM2, 16*1(%arg3 , %r11 , 1)
938	movdqa     \TMP1, \XMM2
939.endif
940	movdqu	   16*2(%arg4 , %r11 , 1), \TMP1
941	pxor	   \TMP1, \XMM3
942.ifc \operation, dec
943	movdqu     \XMM3, 16*2(%arg3 , %r11 , 1)
944	movdqa     \TMP1, \XMM3
945.endif
946	movdqu	   16*3(%arg4 , %r11 , 1), \TMP1
947	pxor	   \TMP1, \XMM4
948.ifc \operation, dec
949	movdqu     \XMM4, 16*3(%arg3 , %r11 , 1)
950	movdqa     \TMP1, \XMM4
951.else
952	movdqu     \XMM1, 16*0(%arg3 , %r11 , 1)
953	movdqu     \XMM2, 16*1(%arg3 , %r11 , 1)
954	movdqu     \XMM3, 16*2(%arg3 , %r11 , 1)
955	movdqu     \XMM4, 16*3(%arg3 , %r11 , 1)
956.endif
957
958	add	   $64, %r11
959	pshufb %xmm14, \XMM1 # perform a 16 byte swap
960	pxor	   \XMMDst, \XMM1
961# combine GHASHed value with the corresponding ciphertext
962	pshufb %xmm14, \XMM2 # perform a 16 byte swap
963	pshufb %xmm14, \XMM3 # perform a 16 byte swap
964	pshufb %xmm14, \XMM4 # perform a 16 byte swap
965
966.L_initial_blocks_done\@:
967
968.endm
969
970/*
971* encrypt 4 blocks at a time
972* ghash the 4 previously encrypted ciphertext blocks
973* arg1, %arg3, %arg4 are used as pointers only, not modified
974* %r11 is the data offset value
975*/
976.macro GHASH_4_ENCRYPT_4_PARALLEL_enc TMP1 TMP2 TMP3 TMP4 TMP5 \
977TMP6 XMM0 XMM1 XMM2 XMM3 XMM4 XMM5 XMM6 XMM7 XMM8 operation
978
979	movdqa	  \XMM1, \XMM5
980	movdqa	  \XMM2, \XMM6
981	movdqa	  \XMM3, \XMM7
982	movdqa	  \XMM4, \XMM8
983
984        movdqa    SHUF_MASK(%rip), %xmm15
985        # multiply TMP5 * HashKey using karatsuba
986
987	movdqa	  \XMM5, \TMP4
988	pshufd	  $78, \XMM5, \TMP6
989	pxor	  \XMM5, \TMP6
990	paddd     ONE(%rip), \XMM0		# INCR CNT
991	movdqu	  HashKey_4(%arg2), \TMP5
992	pclmulqdq $0x11, \TMP5, \TMP4           # TMP4 = a1*b1
993	movdqa    \XMM0, \XMM1
994	paddd     ONE(%rip), \XMM0		# INCR CNT
995	movdqa    \XMM0, \XMM2
996	paddd     ONE(%rip), \XMM0		# INCR CNT
997	movdqa    \XMM0, \XMM3
998	paddd     ONE(%rip), \XMM0		# INCR CNT
999	movdqa    \XMM0, \XMM4
1000	pshufb %xmm15, \XMM1	# perform a 16 byte swap
1001	pclmulqdq $0x00, \TMP5, \XMM5           # XMM5 = a0*b0
1002	pshufb %xmm15, \XMM2	# perform a 16 byte swap
1003	pshufb %xmm15, \XMM3	# perform a 16 byte swap
1004	pshufb %xmm15, \XMM4	# perform a 16 byte swap
1005
1006	pxor	  (%arg1), \XMM1
1007	pxor	  (%arg1), \XMM2
1008	pxor	  (%arg1), \XMM3
1009	pxor	  (%arg1), \XMM4
1010	movdqu	  HashKey_4_k(%arg2), \TMP5
1011	pclmulqdq $0x00, \TMP5, \TMP6       # TMP6 = (a1+a0)*(b1+b0)
1012	movaps 0x10(%arg1), \TMP1
1013	aesenc	  \TMP1, \XMM1              # Round 1
1014	aesenc	  \TMP1, \XMM2
1015	aesenc	  \TMP1, \XMM3
1016	aesenc	  \TMP1, \XMM4
1017	movaps 0x20(%arg1), \TMP1
1018	aesenc	  \TMP1, \XMM1              # Round 2
1019	aesenc	  \TMP1, \XMM2
1020	aesenc	  \TMP1, \XMM3
1021	aesenc	  \TMP1, \XMM4
1022	movdqa	  \XMM6, \TMP1
1023	pshufd	  $78, \XMM6, \TMP2
1024	pxor	  \XMM6, \TMP2
1025	movdqu	  HashKey_3(%arg2), \TMP5
1026	pclmulqdq $0x11, \TMP5, \TMP1       # TMP1 = a1 * b1
1027	movaps 0x30(%arg1), \TMP3
1028	aesenc    \TMP3, \XMM1              # Round 3
1029	aesenc    \TMP3, \XMM2
1030	aesenc    \TMP3, \XMM3
1031	aesenc    \TMP3, \XMM4
1032	pclmulqdq $0x00, \TMP5, \XMM6       # XMM6 = a0*b0
1033	movaps 0x40(%arg1), \TMP3
1034	aesenc	  \TMP3, \XMM1              # Round 4
1035	aesenc	  \TMP3, \XMM2
1036	aesenc	  \TMP3, \XMM3
1037	aesenc	  \TMP3, \XMM4
1038	movdqu	  HashKey_3_k(%arg2), \TMP5
1039	pclmulqdq $0x00, \TMP5, \TMP2       # TMP2 = (a1+a0)*(b1+b0)
1040	movaps 0x50(%arg1), \TMP3
1041	aesenc	  \TMP3, \XMM1              # Round 5
1042	aesenc	  \TMP3, \XMM2
1043	aesenc	  \TMP3, \XMM3
1044	aesenc	  \TMP3, \XMM4
1045	pxor	  \TMP1, \TMP4
1046# accumulate the results in TMP4:XMM5, TMP6 holds the middle part
1047	pxor	  \XMM6, \XMM5
1048	pxor	  \TMP2, \TMP6
1049	movdqa	  \XMM7, \TMP1
1050	pshufd	  $78, \XMM7, \TMP2
1051	pxor	  \XMM7, \TMP2
1052	movdqu	  HashKey_2(%arg2), \TMP5
1053
1054        # Multiply TMP5 * HashKey using karatsuba
1055
1056	pclmulqdq $0x11, \TMP5, \TMP1       # TMP1 = a1*b1
1057	movaps 0x60(%arg1), \TMP3
1058	aesenc	  \TMP3, \XMM1              # Round 6
1059	aesenc	  \TMP3, \XMM2
1060	aesenc	  \TMP3, \XMM3
1061	aesenc	  \TMP3, \XMM4
1062	pclmulqdq $0x00, \TMP5, \XMM7       # XMM7 = a0*b0
1063	movaps 0x70(%arg1), \TMP3
1064	aesenc	  \TMP3, \XMM1              # Round 7
1065	aesenc	  \TMP3, \XMM2
1066	aesenc	  \TMP3, \XMM3
1067	aesenc	  \TMP3, \XMM4
1068	movdqu	  HashKey_2_k(%arg2), \TMP5
1069	pclmulqdq $0x00, \TMP5, \TMP2       # TMP2 = (a1+a0)*(b1+b0)
1070	movaps 0x80(%arg1), \TMP3
1071	aesenc	  \TMP3, \XMM1              # Round 8
1072	aesenc	  \TMP3, \XMM2
1073	aesenc	  \TMP3, \XMM3
1074	aesenc	  \TMP3, \XMM4
1075	pxor	  \TMP1, \TMP4
1076# accumulate the results in TMP4:XMM5, TMP6 holds the middle part
1077	pxor	  \XMM7, \XMM5
1078	pxor	  \TMP2, \TMP6
1079
1080        # Multiply XMM8 * HashKey
1081        # XMM8 and TMP5 hold the values for the two operands
1082
1083	movdqa	  \XMM8, \TMP1
1084	pshufd	  $78, \XMM8, \TMP2
1085	pxor	  \XMM8, \TMP2
1086	movdqu	  HashKey(%arg2), \TMP5
1087	pclmulqdq $0x11, \TMP5, \TMP1      # TMP1 = a1*b1
1088	movaps 0x90(%arg1), \TMP3
1089	aesenc	  \TMP3, \XMM1             # Round 9
1090	aesenc	  \TMP3, \XMM2
1091	aesenc	  \TMP3, \XMM3
1092	aesenc	  \TMP3, \XMM4
1093	pclmulqdq $0x00, \TMP5, \XMM8      # XMM8 = a0*b0
1094	lea	  0xa0(%arg1),%r10
1095	mov	  keysize,%eax
1096	shr	  $2,%eax			# 128->4, 192->6, 256->8
1097	sub	  $4,%eax			# 128->0, 192->2, 256->4
1098	jz	  .Laes_loop_par_enc_done\@
1099
1100.Laes_loop_par_enc\@:
1101	MOVADQ	  (%r10),\TMP3
1102.irpc	index, 1234
1103	aesenc	  \TMP3, %xmm\index
1104.endr
1105	add	  $16,%r10
1106	sub	  $1,%eax
1107	jnz	  .Laes_loop_par_enc\@
1108
1109.Laes_loop_par_enc_done\@:
1110	MOVADQ	  (%r10), \TMP3
1111	aesenclast \TMP3, \XMM1           # Round 10
1112	aesenclast \TMP3, \XMM2
1113	aesenclast \TMP3, \XMM3
1114	aesenclast \TMP3, \XMM4
1115	movdqu    HashKey_k(%arg2), \TMP5
1116	pclmulqdq $0x00, \TMP5, \TMP2          # TMP2 = (a1+a0)*(b1+b0)
1117	movdqu	  (%arg4,%r11,1), \TMP3
1118	pxor	  \TMP3, \XMM1                 # Ciphertext/Plaintext XOR EK
1119	movdqu	  16(%arg4,%r11,1), \TMP3
1120	pxor	  \TMP3, \XMM2                 # Ciphertext/Plaintext XOR EK
1121	movdqu	  32(%arg4,%r11,1), \TMP3
1122	pxor	  \TMP3, \XMM3                 # Ciphertext/Plaintext XOR EK
1123	movdqu	  48(%arg4,%r11,1), \TMP3
1124	pxor	  \TMP3, \XMM4                 # Ciphertext/Plaintext XOR EK
1125        movdqu    \XMM1, (%arg3,%r11,1)        # Write to the ciphertext buffer
1126        movdqu    \XMM2, 16(%arg3,%r11,1)      # Write to the ciphertext buffer
1127        movdqu    \XMM3, 32(%arg3,%r11,1)      # Write to the ciphertext buffer
1128        movdqu    \XMM4, 48(%arg3,%r11,1)      # Write to the ciphertext buffer
1129	pshufb %xmm15, \XMM1        # perform a 16 byte swap
1130	pshufb %xmm15, \XMM2	# perform a 16 byte swap
1131	pshufb %xmm15, \XMM3	# perform a 16 byte swap
1132	pshufb %xmm15, \XMM4	# perform a 16 byte swap
1133
1134	pxor	  \TMP4, \TMP1
1135	pxor	  \XMM8, \XMM5
1136	pxor	  \TMP6, \TMP2
1137	pxor	  \TMP1, \TMP2
1138	pxor	  \XMM5, \TMP2
1139	movdqa	  \TMP2, \TMP3
1140	pslldq	  $8, \TMP3                    # left shift TMP3 2 DWs
1141	psrldq	  $8, \TMP2                    # right shift TMP2 2 DWs
1142	pxor	  \TMP3, \XMM5
1143	pxor	  \TMP2, \TMP1	  # accumulate the results in TMP1:XMM5
1144
1145        # first phase of reduction
1146
1147	movdqa    \XMM5, \TMP2
1148	movdqa    \XMM5, \TMP3
1149	movdqa    \XMM5, \TMP4
1150# move XMM5 into TMP2, TMP3, TMP4 in order to perform shifts independently
1151	pslld     $31, \TMP2                   # packed right shift << 31
1152	pslld     $30, \TMP3                   # packed right shift << 30
1153	pslld     $25, \TMP4                   # packed right shift << 25
1154	pxor      \TMP3, \TMP2	               # xor the shifted versions
1155	pxor      \TMP4, \TMP2
1156	movdqa    \TMP2, \TMP5
1157	psrldq    $4, \TMP5                    # right shift T5 1 DW
1158	pslldq    $12, \TMP2                   # left shift T2 3 DWs
1159	pxor      \TMP2, \XMM5
1160
1161        # second phase of reduction
1162
1163	movdqa    \XMM5,\TMP2 # make 3 copies of XMM5 into TMP2, TMP3, TMP4
1164	movdqa    \XMM5,\TMP3
1165	movdqa    \XMM5,\TMP4
1166	psrld     $1, \TMP2                    # packed left shift >>1
1167	psrld     $2, \TMP3                    # packed left shift >>2
1168	psrld     $7, \TMP4                    # packed left shift >>7
1169	pxor      \TMP3,\TMP2		       # xor the shifted versions
1170	pxor      \TMP4,\TMP2
1171	pxor      \TMP5, \TMP2
1172	pxor      \TMP2, \XMM5
1173	pxor      \TMP1, \XMM5                 # result is in TMP1
1174
1175	pxor	  \XMM5, \XMM1
1176.endm
1177
1178/*
1179* decrypt 4 blocks at a time
1180* ghash the 4 previously decrypted ciphertext blocks
1181* arg1, %arg3, %arg4 are used as pointers only, not modified
1182* %r11 is the data offset value
1183*/
1184.macro GHASH_4_ENCRYPT_4_PARALLEL_dec TMP1 TMP2 TMP3 TMP4 TMP5 \
1185TMP6 XMM0 XMM1 XMM2 XMM3 XMM4 XMM5 XMM6 XMM7 XMM8 operation
1186
1187	movdqa	  \XMM1, \XMM5
1188	movdqa	  \XMM2, \XMM6
1189	movdqa	  \XMM3, \XMM7
1190	movdqa	  \XMM4, \XMM8
1191
1192        movdqa    SHUF_MASK(%rip), %xmm15
1193        # multiply TMP5 * HashKey using karatsuba
1194
1195	movdqa	  \XMM5, \TMP4
1196	pshufd	  $78, \XMM5, \TMP6
1197	pxor	  \XMM5, \TMP6
1198	paddd     ONE(%rip), \XMM0		# INCR CNT
1199	movdqu	  HashKey_4(%arg2), \TMP5
1200	pclmulqdq $0x11, \TMP5, \TMP4           # TMP4 = a1*b1
1201	movdqa    \XMM0, \XMM1
1202	paddd     ONE(%rip), \XMM0		# INCR CNT
1203	movdqa    \XMM0, \XMM2
1204	paddd     ONE(%rip), \XMM0		# INCR CNT
1205	movdqa    \XMM0, \XMM3
1206	paddd     ONE(%rip), \XMM0		# INCR CNT
1207	movdqa    \XMM0, \XMM4
1208	pshufb %xmm15, \XMM1	# perform a 16 byte swap
1209	pclmulqdq $0x00, \TMP5, \XMM5           # XMM5 = a0*b0
1210	pshufb %xmm15, \XMM2	# perform a 16 byte swap
1211	pshufb %xmm15, \XMM3	# perform a 16 byte swap
1212	pshufb %xmm15, \XMM4	# perform a 16 byte swap
1213
1214	pxor	  (%arg1), \XMM1
1215	pxor	  (%arg1), \XMM2
1216	pxor	  (%arg1), \XMM3
1217	pxor	  (%arg1), \XMM4
1218	movdqu	  HashKey_4_k(%arg2), \TMP5
1219	pclmulqdq $0x00, \TMP5, \TMP6       # TMP6 = (a1+a0)*(b1+b0)
1220	movaps 0x10(%arg1), \TMP1
1221	aesenc	  \TMP1, \XMM1              # Round 1
1222	aesenc	  \TMP1, \XMM2
1223	aesenc	  \TMP1, \XMM3
1224	aesenc	  \TMP1, \XMM4
1225	movaps 0x20(%arg1), \TMP1
1226	aesenc	  \TMP1, \XMM1              # Round 2
1227	aesenc	  \TMP1, \XMM2
1228	aesenc	  \TMP1, \XMM3
1229	aesenc	  \TMP1, \XMM4
1230	movdqa	  \XMM6, \TMP1
1231	pshufd	  $78, \XMM6, \TMP2
1232	pxor	  \XMM6, \TMP2
1233	movdqu	  HashKey_3(%arg2), \TMP5
1234	pclmulqdq $0x11, \TMP5, \TMP1       # TMP1 = a1 * b1
1235	movaps 0x30(%arg1), \TMP3
1236	aesenc    \TMP3, \XMM1              # Round 3
1237	aesenc    \TMP3, \XMM2
1238	aesenc    \TMP3, \XMM3
1239	aesenc    \TMP3, \XMM4
1240	pclmulqdq $0x00, \TMP5, \XMM6       # XMM6 = a0*b0
1241	movaps 0x40(%arg1), \TMP3
1242	aesenc	  \TMP3, \XMM1              # Round 4
1243	aesenc	  \TMP3, \XMM2
1244	aesenc	  \TMP3, \XMM3
1245	aesenc	  \TMP3, \XMM4
1246	movdqu	  HashKey_3_k(%arg2), \TMP5
1247	pclmulqdq $0x00, \TMP5, \TMP2       # TMP2 = (a1+a0)*(b1+b0)
1248	movaps 0x50(%arg1), \TMP3
1249	aesenc	  \TMP3, \XMM1              # Round 5
1250	aesenc	  \TMP3, \XMM2
1251	aesenc	  \TMP3, \XMM3
1252	aesenc	  \TMP3, \XMM4
1253	pxor	  \TMP1, \TMP4
1254# accumulate the results in TMP4:XMM5, TMP6 holds the middle part
1255	pxor	  \XMM6, \XMM5
1256	pxor	  \TMP2, \TMP6
1257	movdqa	  \XMM7, \TMP1
1258	pshufd	  $78, \XMM7, \TMP2
1259	pxor	  \XMM7, \TMP2
1260	movdqu	  HashKey_2(%arg2), \TMP5
1261
1262        # Multiply TMP5 * HashKey using karatsuba
1263
1264	pclmulqdq $0x11, \TMP5, \TMP1       # TMP1 = a1*b1
1265	movaps 0x60(%arg1), \TMP3
1266	aesenc	  \TMP3, \XMM1              # Round 6
1267	aesenc	  \TMP3, \XMM2
1268	aesenc	  \TMP3, \XMM3
1269	aesenc	  \TMP3, \XMM4
1270	pclmulqdq $0x00, \TMP5, \XMM7       # XMM7 = a0*b0
1271	movaps 0x70(%arg1), \TMP3
1272	aesenc	  \TMP3, \XMM1              # Round 7
1273	aesenc	  \TMP3, \XMM2
1274	aesenc	  \TMP3, \XMM3
1275	aesenc	  \TMP3, \XMM4
1276	movdqu	  HashKey_2_k(%arg2), \TMP5
1277	pclmulqdq $0x00, \TMP5, \TMP2       # TMP2 = (a1+a0)*(b1+b0)
1278	movaps 0x80(%arg1), \TMP3
1279	aesenc	  \TMP3, \XMM1              # Round 8
1280	aesenc	  \TMP3, \XMM2
1281	aesenc	  \TMP3, \XMM3
1282	aesenc	  \TMP3, \XMM4
1283	pxor	  \TMP1, \TMP4
1284# accumulate the results in TMP4:XMM5, TMP6 holds the middle part
1285	pxor	  \XMM7, \XMM5
1286	pxor	  \TMP2, \TMP6
1287
1288        # Multiply XMM8 * HashKey
1289        # XMM8 and TMP5 hold the values for the two operands
1290
1291	movdqa	  \XMM8, \TMP1
1292	pshufd	  $78, \XMM8, \TMP2
1293	pxor	  \XMM8, \TMP2
1294	movdqu	  HashKey(%arg2), \TMP5
1295	pclmulqdq $0x11, \TMP5, \TMP1      # TMP1 = a1*b1
1296	movaps 0x90(%arg1), \TMP3
1297	aesenc	  \TMP3, \XMM1             # Round 9
1298	aesenc	  \TMP3, \XMM2
1299	aesenc	  \TMP3, \XMM3
1300	aesenc	  \TMP3, \XMM4
1301	pclmulqdq $0x00, \TMP5, \XMM8      # XMM8 = a0*b0
1302	lea	  0xa0(%arg1),%r10
1303	mov	  keysize,%eax
1304	shr	  $2,%eax		        # 128->4, 192->6, 256->8
1305	sub	  $4,%eax			# 128->0, 192->2, 256->4
1306	jz	  .Laes_loop_par_dec_done\@
1307
1308.Laes_loop_par_dec\@:
1309	MOVADQ	  (%r10),\TMP3
1310.irpc	index, 1234
1311	aesenc	  \TMP3, %xmm\index
1312.endr
1313	add	  $16,%r10
1314	sub	  $1,%eax
1315	jnz	  .Laes_loop_par_dec\@
1316
1317.Laes_loop_par_dec_done\@:
1318	MOVADQ	  (%r10), \TMP3
1319	aesenclast \TMP3, \XMM1           # last round
1320	aesenclast \TMP3, \XMM2
1321	aesenclast \TMP3, \XMM3
1322	aesenclast \TMP3, \XMM4
1323	movdqu    HashKey_k(%arg2), \TMP5
1324	pclmulqdq $0x00, \TMP5, \TMP2          # TMP2 = (a1+a0)*(b1+b0)
1325	movdqu	  (%arg4,%r11,1), \TMP3
1326	pxor	  \TMP3, \XMM1                 # Ciphertext/Plaintext XOR EK
1327	movdqu	  \XMM1, (%arg3,%r11,1)        # Write to plaintext buffer
1328	movdqa    \TMP3, \XMM1
1329	movdqu	  16(%arg4,%r11,1), \TMP3
1330	pxor	  \TMP3, \XMM2                 # Ciphertext/Plaintext XOR EK
1331	movdqu	  \XMM2, 16(%arg3,%r11,1)      # Write to plaintext buffer
1332	movdqa    \TMP3, \XMM2
1333	movdqu	  32(%arg4,%r11,1), \TMP3
1334	pxor	  \TMP3, \XMM3                 # Ciphertext/Plaintext XOR EK
1335	movdqu	  \XMM3, 32(%arg3,%r11,1)      # Write to plaintext buffer
1336	movdqa    \TMP3, \XMM3
1337	movdqu	  48(%arg4,%r11,1), \TMP3
1338	pxor	  \TMP3, \XMM4                 # Ciphertext/Plaintext XOR EK
1339	movdqu	  \XMM4, 48(%arg3,%r11,1)      # Write to plaintext buffer
1340	movdqa    \TMP3, \XMM4
1341	pshufb %xmm15, \XMM1        # perform a 16 byte swap
1342	pshufb %xmm15, \XMM2	# perform a 16 byte swap
1343	pshufb %xmm15, \XMM3	# perform a 16 byte swap
1344	pshufb %xmm15, \XMM4	# perform a 16 byte swap
1345
1346	pxor	  \TMP4, \TMP1
1347	pxor	  \XMM8, \XMM5
1348	pxor	  \TMP6, \TMP2
1349	pxor	  \TMP1, \TMP2
1350	pxor	  \XMM5, \TMP2
1351	movdqa	  \TMP2, \TMP3
1352	pslldq	  $8, \TMP3                    # left shift TMP3 2 DWs
1353	psrldq	  $8, \TMP2                    # right shift TMP2 2 DWs
1354	pxor	  \TMP3, \XMM5
1355	pxor	  \TMP2, \TMP1	  # accumulate the results in TMP1:XMM5
1356
1357        # first phase of reduction
1358
1359	movdqa    \XMM5, \TMP2
1360	movdqa    \XMM5, \TMP3
1361	movdqa    \XMM5, \TMP4
1362# move XMM5 into TMP2, TMP3, TMP4 in order to perform shifts independently
1363	pslld     $31, \TMP2                   # packed right shift << 31
1364	pslld     $30, \TMP3                   # packed right shift << 30
1365	pslld     $25, \TMP4                   # packed right shift << 25
1366	pxor      \TMP3, \TMP2	               # xor the shifted versions
1367	pxor      \TMP4, \TMP2
1368	movdqa    \TMP2, \TMP5
1369	psrldq    $4, \TMP5                    # right shift T5 1 DW
1370	pslldq    $12, \TMP2                   # left shift T2 3 DWs
1371	pxor      \TMP2, \XMM5
1372
1373        # second phase of reduction
1374
1375	movdqa    \XMM5,\TMP2 # make 3 copies of XMM5 into TMP2, TMP3, TMP4
1376	movdqa    \XMM5,\TMP3
1377	movdqa    \XMM5,\TMP4
1378	psrld     $1, \TMP2                    # packed left shift >>1
1379	psrld     $2, \TMP3                    # packed left shift >>2
1380	psrld     $7, \TMP4                    # packed left shift >>7
1381	pxor      \TMP3,\TMP2		       # xor the shifted versions
1382	pxor      \TMP4,\TMP2
1383	pxor      \TMP5, \TMP2
1384	pxor      \TMP2, \XMM5
1385	pxor      \TMP1, \XMM5                 # result is in TMP1
1386
1387	pxor	  \XMM5, \XMM1
1388.endm
1389
1390/* GHASH the last 4 ciphertext blocks. */
1391.macro	GHASH_LAST_4 TMP1 TMP2 TMP3 TMP4 TMP5 TMP6 \
1392TMP7 XMM1 XMM2 XMM3 XMM4 XMMDst
1393
1394        # Multiply TMP6 * HashKey (using Karatsuba)
1395
1396	movdqa	  \XMM1, \TMP6
1397	pshufd	  $78, \XMM1, \TMP2
1398	pxor	  \XMM1, \TMP2
1399	movdqu	  HashKey_4(%arg2), \TMP5
1400	pclmulqdq $0x11, \TMP5, \TMP6       # TMP6 = a1*b1
1401	pclmulqdq $0x00, \TMP5, \XMM1       # XMM1 = a0*b0
1402	movdqu	  HashKey_4_k(%arg2), \TMP4
1403	pclmulqdq $0x00, \TMP4, \TMP2       # TMP2 = (a1+a0)*(b1+b0)
1404	movdqa	  \XMM1, \XMMDst
1405	movdqa	  \TMP2, \XMM1              # result in TMP6, XMMDst, XMM1
1406
1407        # Multiply TMP1 * HashKey (using Karatsuba)
1408
1409	movdqa	  \XMM2, \TMP1
1410	pshufd	  $78, \XMM2, \TMP2
1411	pxor	  \XMM2, \TMP2
1412	movdqu	  HashKey_3(%arg2), \TMP5
1413	pclmulqdq $0x11, \TMP5, \TMP1       # TMP1 = a1*b1
1414	pclmulqdq $0x00, \TMP5, \XMM2       # XMM2 = a0*b0
1415	movdqu	  HashKey_3_k(%arg2), \TMP4
1416	pclmulqdq $0x00, \TMP4, \TMP2       # TMP2 = (a1+a0)*(b1+b0)
1417	pxor	  \TMP1, \TMP6
1418	pxor	  \XMM2, \XMMDst
1419	pxor	  \TMP2, \XMM1
1420# results accumulated in TMP6, XMMDst, XMM1
1421
1422        # Multiply TMP1 * HashKey (using Karatsuba)
1423
1424	movdqa	  \XMM3, \TMP1
1425	pshufd	  $78, \XMM3, \TMP2
1426	pxor	  \XMM3, \TMP2
1427	movdqu	  HashKey_2(%arg2), \TMP5
1428	pclmulqdq $0x11, \TMP5, \TMP1       # TMP1 = a1*b1
1429	pclmulqdq $0x00, \TMP5, \XMM3       # XMM3 = a0*b0
1430	movdqu	  HashKey_2_k(%arg2), \TMP4
1431	pclmulqdq $0x00, \TMP4, \TMP2       # TMP2 = (a1+a0)*(b1+b0)
1432	pxor	  \TMP1, \TMP6
1433	pxor	  \XMM3, \XMMDst
1434	pxor	  \TMP2, \XMM1   # results accumulated in TMP6, XMMDst, XMM1
1435
1436        # Multiply TMP1 * HashKey (using Karatsuba)
1437	movdqa	  \XMM4, \TMP1
1438	pshufd	  $78, \XMM4, \TMP2
1439	pxor	  \XMM4, \TMP2
1440	movdqu	  HashKey(%arg2), \TMP5
1441	pclmulqdq $0x11, \TMP5, \TMP1	    # TMP1 = a1*b1
1442	pclmulqdq $0x00, \TMP5, \XMM4       # XMM4 = a0*b0
1443	movdqu	  HashKey_k(%arg2), \TMP4
1444	pclmulqdq $0x00, \TMP4, \TMP2       # TMP2 = (a1+a0)*(b1+b0)
1445	pxor	  \TMP1, \TMP6
1446	pxor	  \XMM4, \XMMDst
1447	pxor	  \XMM1, \TMP2
1448	pxor	  \TMP6, \TMP2
1449	pxor	  \XMMDst, \TMP2
1450	# middle section of the temp results combined as in karatsuba algorithm
1451	movdqa	  \TMP2, \TMP4
1452	pslldq	  $8, \TMP4                 # left shift TMP4 2 DWs
1453	psrldq	  $8, \TMP2                 # right shift TMP2 2 DWs
1454	pxor	  \TMP4, \XMMDst
1455	pxor	  \TMP2, \TMP6
1456# TMP6:XMMDst holds the result of the accumulated carry-less multiplications
1457	# first phase of the reduction
1458	movdqa    \XMMDst, \TMP2
1459	movdqa    \XMMDst, \TMP3
1460	movdqa    \XMMDst, \TMP4
1461# move XMMDst into TMP2, TMP3, TMP4 in order to perform 3 shifts independently
1462	pslld     $31, \TMP2                # packed right shifting << 31
1463	pslld     $30, \TMP3                # packed right shifting << 30
1464	pslld     $25, \TMP4                # packed right shifting << 25
1465	pxor      \TMP3, \TMP2              # xor the shifted versions
1466	pxor      \TMP4, \TMP2
1467	movdqa    \TMP2, \TMP7
1468	psrldq    $4, \TMP7                 # right shift TMP7 1 DW
1469	pslldq    $12, \TMP2                # left shift TMP2 3 DWs
1470	pxor      \TMP2, \XMMDst
1471
1472        # second phase of the reduction
1473	movdqa    \XMMDst, \TMP2
1474	# make 3 copies of XMMDst for doing 3 shift operations
1475	movdqa    \XMMDst, \TMP3
1476	movdqa    \XMMDst, \TMP4
1477	psrld     $1, \TMP2                 # packed left shift >> 1
1478	psrld     $2, \TMP3                 # packed left shift >> 2
1479	psrld     $7, \TMP4                 # packed left shift >> 7
1480	pxor      \TMP3, \TMP2              # xor the shifted versions
1481	pxor      \TMP4, \TMP2
1482	pxor      \TMP7, \TMP2
1483	pxor      \TMP2, \XMMDst
1484	pxor      \TMP6, \XMMDst            # reduced result is in XMMDst
1485.endm
1486
1487
1488/* Encryption of a single block
1489* uses eax & r10
1490*/
1491
1492.macro ENCRYPT_SINGLE_BLOCK XMM0 TMP1
1493
1494	pxor		(%arg1), \XMM0
1495	mov		keysize,%eax
1496	shr		$2,%eax			# 128->4, 192->6, 256->8
1497	add		$5,%eax			# 128->9, 192->11, 256->13
1498	lea		16(%arg1), %r10	  # get first expanded key address
1499
1500_esb_loop_\@:
1501	MOVADQ		(%r10),\TMP1
1502	aesenc		\TMP1,\XMM0
1503	add		$16,%r10
1504	sub		$1,%eax
1505	jnz		_esb_loop_\@
1506
1507	MOVADQ		(%r10),\TMP1
1508	aesenclast	\TMP1,\XMM0
1509.endm
1510/*****************************************************************************
1511* void aesni_gcm_dec(void *aes_ctx,    // AES Key schedule. Starts on a 16 byte boundary.
1512*                   struct gcm_context_data *data
1513*                                      // Context data
1514*                   u8 *out,           // Plaintext output. Encrypt in-place is allowed.
1515*                   const u8 *in,      // Ciphertext input
1516*                   u64 plaintext_len, // Length of data in bytes for decryption.
1517*                   u8 *iv,            // Pre-counter block j0: 4 byte salt (from Security Association)
1518*                                      // concatenated with 8 byte Initialisation Vector (from IPSec ESP Payload)
1519*                                      // concatenated with 0x00000001. 16-byte aligned pointer.
1520*                   u8 *hash_subkey,   // H, the Hash sub key input. Data starts on a 16-byte boundary.
1521*                   const u8 *aad,     // Additional Authentication Data (AAD)
1522*                   u64 aad_len,       // Length of AAD in bytes. With RFC4106 this is going to be 8 or 12 bytes
1523*                   u8  *auth_tag,     // Authenticated Tag output. The driver will compare this to the
1524*                                      // given authentication tag and only return the plaintext if they match.
1525*                   u64 auth_tag_len); // Authenticated Tag Length in bytes. Valid values are 16
1526*                                      // (most likely), 12 or 8.
1527*
1528* Assumptions:
1529*
1530* keys:
1531*       keys are pre-expanded and aligned to 16 bytes. we are using the first
1532*       set of 11 keys in the data structure void *aes_ctx
1533*
1534* iv:
1535*       0                   1                   2                   3
1536*       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
1537*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1538*       |                             Salt  (From the SA)               |
1539*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1540*       |                     Initialization Vector                     |
1541*       |         (This is the sequence number from IPSec header)       |
1542*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1543*       |                              0x1                              |
1544*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1545*
1546*
1547*
1548* AAD:
1549*       AAD padded to 128 bits with 0
1550*       for example, assume AAD is a u32 vector
1551*
1552*       if AAD is 8 bytes:
1553*       AAD[3] = {A0, A1};
1554*       padded AAD in xmm register = {A1 A0 0 0}
1555*
1556*       0                   1                   2                   3
1557*       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
1558*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1559*       |                               SPI (A1)                        |
1560*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1561*       |                     32-bit Sequence Number (A0)               |
1562*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1563*       |                              0x0                              |
1564*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1565*
1566*                                       AAD Format with 32-bit Sequence Number
1567*
1568*       if AAD is 12 bytes:
1569*       AAD[3] = {A0, A1, A2};
1570*       padded AAD in xmm register = {A2 A1 A0 0}
1571*
1572*       0                   1                   2                   3
1573*       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
1574*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1575*       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
1576*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1577*       |                               SPI (A2)                        |
1578*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1579*       |                 64-bit Extended Sequence Number {A1,A0}       |
1580*       |                                                               |
1581*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1582*       |                              0x0                              |
1583*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1584*
1585*                        AAD Format with 64-bit Extended Sequence Number
1586*
1587* poly = x^128 + x^127 + x^126 + x^121 + 1
1588*
1589*****************************************************************************/
1590SYM_FUNC_START(aesni_gcm_dec)
1591	FUNC_SAVE
1592
1593	GCM_INIT %arg6, arg7, arg8, arg9
1594	GCM_ENC_DEC dec
1595	GCM_COMPLETE arg10, arg11
1596	FUNC_RESTORE
1597	RET
1598SYM_FUNC_END(aesni_gcm_dec)
1599
1600
1601/*****************************************************************************
1602* void aesni_gcm_enc(void *aes_ctx,      // AES Key schedule. Starts on a 16 byte boundary.
1603*                    struct gcm_context_data *data
1604*                                        // Context data
1605*                    u8 *out,            // Ciphertext output. Encrypt in-place is allowed.
1606*                    const u8 *in,       // Plaintext input
1607*                    u64 plaintext_len,  // Length of data in bytes for encryption.
1608*                    u8 *iv,             // Pre-counter block j0: 4 byte salt (from Security Association)
1609*                                        // concatenated with 8 byte Initialisation Vector (from IPSec ESP Payload)
1610*                                        // concatenated with 0x00000001. 16-byte aligned pointer.
1611*                    u8 *hash_subkey,    // H, the Hash sub key input. Data starts on a 16-byte boundary.
1612*                    const u8 *aad,      // Additional Authentication Data (AAD)
1613*                    u64 aad_len,        // Length of AAD in bytes. With RFC4106 this is going to be 8 or 12 bytes
1614*                    u8 *auth_tag,       // Authenticated Tag output.
1615*                    u64 auth_tag_len);  // Authenticated Tag Length in bytes. Valid values are 16 (most likely),
1616*                                        // 12 or 8.
1617*
1618* Assumptions:
1619*
1620* keys:
1621*       keys are pre-expanded and aligned to 16 bytes. we are using the
1622*       first set of 11 keys in the data structure void *aes_ctx
1623*
1624*
1625* iv:
1626*       0                   1                   2                   3
1627*       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
1628*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1629*       |                             Salt  (From the SA)               |
1630*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1631*       |                     Initialization Vector                     |
1632*       |         (This is the sequence number from IPSec header)       |
1633*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1634*       |                              0x1                              |
1635*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1636*
1637*
1638*
1639* AAD:
1640*       AAD padded to 128 bits with 0
1641*       for example, assume AAD is a u32 vector
1642*
1643*       if AAD is 8 bytes:
1644*       AAD[3] = {A0, A1};
1645*       padded AAD in xmm register = {A1 A0 0 0}
1646*
1647*       0                   1                   2                   3
1648*       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
1649*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1650*       |                               SPI (A1)                        |
1651*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1652*       |                     32-bit Sequence Number (A0)               |
1653*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1654*       |                              0x0                              |
1655*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1656*
1657*                                 AAD Format with 32-bit Sequence Number
1658*
1659*       if AAD is 12 bytes:
1660*       AAD[3] = {A0, A1, A2};
1661*       padded AAD in xmm register = {A2 A1 A0 0}
1662*
1663*       0                   1                   2                   3
1664*       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
1665*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1666*       |                               SPI (A2)                        |
1667*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1668*       |                 64-bit Extended Sequence Number {A1,A0}       |
1669*       |                                                               |
1670*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1671*       |                              0x0                              |
1672*       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1673*
1674*                         AAD Format with 64-bit Extended Sequence Number
1675*
1676* poly = x^128 + x^127 + x^126 + x^121 + 1
1677***************************************************************************/
1678SYM_FUNC_START(aesni_gcm_enc)
1679	FUNC_SAVE
1680
1681	GCM_INIT %arg6, arg7, arg8, arg9
1682	GCM_ENC_DEC enc
1683
1684	GCM_COMPLETE arg10, arg11
1685	FUNC_RESTORE
1686	RET
1687SYM_FUNC_END(aesni_gcm_enc)
1688
1689/*****************************************************************************
1690* void aesni_gcm_init(void *aes_ctx,      // AES Key schedule. Starts on a 16 byte boundary.
1691*                     struct gcm_context_data *data,
1692*                                         // context data
1693*                     u8 *iv,             // Pre-counter block j0: 4 byte salt (from Security Association)
1694*                                         // concatenated with 8 byte Initialisation Vector (from IPSec ESP Payload)
1695*                                         // concatenated with 0x00000001. 16-byte aligned pointer.
1696*                     u8 *hash_subkey,    // H, the Hash sub key input. Data starts on a 16-byte boundary.
1697*                     const u8 *aad,      // Additional Authentication Data (AAD)
1698*                     u64 aad_len)        // Length of AAD in bytes.
1699*/
1700SYM_FUNC_START(aesni_gcm_init)
1701	FUNC_SAVE
1702	GCM_INIT %arg3, %arg4,%arg5, %arg6
1703	FUNC_RESTORE
1704	RET
1705SYM_FUNC_END(aesni_gcm_init)
1706
1707/*****************************************************************************
1708* void aesni_gcm_enc_update(void *aes_ctx,      // AES Key schedule. Starts on a 16 byte boundary.
1709*                    struct gcm_context_data *data,
1710*                                        // context data
1711*                    u8 *out,            // Ciphertext output. Encrypt in-place is allowed.
1712*                    const u8 *in,       // Plaintext input
1713*                    u64 plaintext_len,  // Length of data in bytes for encryption.
1714*/
1715SYM_FUNC_START(aesni_gcm_enc_update)
1716	FUNC_SAVE
1717	GCM_ENC_DEC enc
1718	FUNC_RESTORE
1719	RET
1720SYM_FUNC_END(aesni_gcm_enc_update)
1721
1722/*****************************************************************************
1723* void aesni_gcm_dec_update(void *aes_ctx,      // AES Key schedule. Starts on a 16 byte boundary.
1724*                    struct gcm_context_data *data,
1725*                                        // context data
1726*                    u8 *out,            // Ciphertext output. Encrypt in-place is allowed.
1727*                    const u8 *in,       // Plaintext input
1728*                    u64 plaintext_len,  // Length of data in bytes for encryption.
1729*/
1730SYM_FUNC_START(aesni_gcm_dec_update)
1731	FUNC_SAVE
1732	GCM_ENC_DEC dec
1733	FUNC_RESTORE
1734	RET
1735SYM_FUNC_END(aesni_gcm_dec_update)
1736
1737/*****************************************************************************
1738* void aesni_gcm_finalize(void *aes_ctx,      // AES Key schedule. Starts on a 16 byte boundary.
1739*                    struct gcm_context_data *data,
1740*                                        // context data
1741*                    u8 *auth_tag,       // Authenticated Tag output.
1742*                    u64 auth_tag_len);  // Authenticated Tag Length in bytes. Valid values are 16 (most likely),
1743*                                        // 12 or 8.
1744*/
1745SYM_FUNC_START(aesni_gcm_finalize)
1746	FUNC_SAVE
1747	GCM_COMPLETE %arg3 %arg4
1748	FUNC_RESTORE
1749	RET
1750SYM_FUNC_END(aesni_gcm_finalize)
1751
1752#endif
1753
1754SYM_FUNC_START_LOCAL(_key_expansion_256a)
1755	pshufd $0b11111111, %xmm1, %xmm1
1756	shufps $0b00010000, %xmm0, %xmm4
1757	pxor %xmm4, %xmm0
1758	shufps $0b10001100, %xmm0, %xmm4
1759	pxor %xmm4, %xmm0
1760	pxor %xmm1, %xmm0
1761	movaps %xmm0, (TKEYP)
1762	add $0x10, TKEYP
1763	RET
1764SYM_FUNC_END(_key_expansion_256a)
1765SYM_FUNC_ALIAS_LOCAL(_key_expansion_128, _key_expansion_256a)
1766
1767SYM_FUNC_START_LOCAL(_key_expansion_192a)
1768	pshufd $0b01010101, %xmm1, %xmm1
1769	shufps $0b00010000, %xmm0, %xmm4
1770	pxor %xmm4, %xmm0
1771	shufps $0b10001100, %xmm0, %xmm4
1772	pxor %xmm4, %xmm0
1773	pxor %xmm1, %xmm0
1774
1775	movaps %xmm2, %xmm5
1776	movaps %xmm2, %xmm6
1777	pslldq $4, %xmm5
1778	pshufd $0b11111111, %xmm0, %xmm3
1779	pxor %xmm3, %xmm2
1780	pxor %xmm5, %xmm2
1781
1782	movaps %xmm0, %xmm1
1783	shufps $0b01000100, %xmm0, %xmm6
1784	movaps %xmm6, (TKEYP)
1785	shufps $0b01001110, %xmm2, %xmm1
1786	movaps %xmm1, 0x10(TKEYP)
1787	add $0x20, TKEYP
1788	RET
1789SYM_FUNC_END(_key_expansion_192a)
1790
1791SYM_FUNC_START_LOCAL(_key_expansion_192b)
1792	pshufd $0b01010101, %xmm1, %xmm1
1793	shufps $0b00010000, %xmm0, %xmm4
1794	pxor %xmm4, %xmm0
1795	shufps $0b10001100, %xmm0, %xmm4
1796	pxor %xmm4, %xmm0
1797	pxor %xmm1, %xmm0
1798
1799	movaps %xmm2, %xmm5
1800	pslldq $4, %xmm5
1801	pshufd $0b11111111, %xmm0, %xmm3
1802	pxor %xmm3, %xmm2
1803	pxor %xmm5, %xmm2
1804
1805	movaps %xmm0, (TKEYP)
1806	add $0x10, TKEYP
1807	RET
1808SYM_FUNC_END(_key_expansion_192b)
1809
1810SYM_FUNC_START_LOCAL(_key_expansion_256b)
1811	pshufd $0b10101010, %xmm1, %xmm1
1812	shufps $0b00010000, %xmm2, %xmm4
1813	pxor %xmm4, %xmm2
1814	shufps $0b10001100, %xmm2, %xmm4
1815	pxor %xmm4, %xmm2
1816	pxor %xmm1, %xmm2
1817	movaps %xmm2, (TKEYP)
1818	add $0x10, TKEYP
1819	RET
1820SYM_FUNC_END(_key_expansion_256b)
1821
1822/*
1823 * int aesni_set_key(struct crypto_aes_ctx *ctx, const u8 *in_key,
1824 *                   unsigned int key_len)
1825 */
1826SYM_FUNC_START(aesni_set_key)
1827	FRAME_BEGIN
1828#ifndef __x86_64__
1829	pushl KEYP
1830	movl (FRAME_OFFSET+8)(%esp), KEYP	# ctx
1831	movl (FRAME_OFFSET+12)(%esp), UKEYP	# in_key
1832	movl (FRAME_OFFSET+16)(%esp), %edx	# key_len
1833#endif
1834	movups (UKEYP), %xmm0		# user key (first 16 bytes)
1835	movaps %xmm0, (KEYP)
1836	lea 0x10(KEYP), TKEYP		# key addr
1837	movl %edx, 480(KEYP)
1838	pxor %xmm4, %xmm4		# xmm4 is assumed 0 in _key_expansion_x
1839	cmp $24, %dl
1840	jb .Lenc_key128
1841	je .Lenc_key192
1842	movups 0x10(UKEYP), %xmm2	# other user key
1843	movaps %xmm2, (TKEYP)
1844	add $0x10, TKEYP
1845	aeskeygenassist $0x1, %xmm2, %xmm1	# round 1
1846	call _key_expansion_256a
1847	aeskeygenassist $0x1, %xmm0, %xmm1
1848	call _key_expansion_256b
1849	aeskeygenassist $0x2, %xmm2, %xmm1	# round 2
1850	call _key_expansion_256a
1851	aeskeygenassist $0x2, %xmm0, %xmm1
1852	call _key_expansion_256b
1853	aeskeygenassist $0x4, %xmm2, %xmm1	# round 3
1854	call _key_expansion_256a
1855	aeskeygenassist $0x4, %xmm0, %xmm1
1856	call _key_expansion_256b
1857	aeskeygenassist $0x8, %xmm2, %xmm1	# round 4
1858	call _key_expansion_256a
1859	aeskeygenassist $0x8, %xmm0, %xmm1
1860	call _key_expansion_256b
1861	aeskeygenassist $0x10, %xmm2, %xmm1	# round 5
1862	call _key_expansion_256a
1863	aeskeygenassist $0x10, %xmm0, %xmm1
1864	call _key_expansion_256b
1865	aeskeygenassist $0x20, %xmm2, %xmm1	# round 6
1866	call _key_expansion_256a
1867	aeskeygenassist $0x20, %xmm0, %xmm1
1868	call _key_expansion_256b
1869	aeskeygenassist $0x40, %xmm2, %xmm1	# round 7
1870	call _key_expansion_256a
1871	jmp .Ldec_key
1872.Lenc_key192:
1873	movq 0x10(UKEYP), %xmm2		# other user key
1874	aeskeygenassist $0x1, %xmm2, %xmm1	# round 1
1875	call _key_expansion_192a
1876	aeskeygenassist $0x2, %xmm2, %xmm1	# round 2
1877	call _key_expansion_192b
1878	aeskeygenassist $0x4, %xmm2, %xmm1	# round 3
1879	call _key_expansion_192a
1880	aeskeygenassist $0x8, %xmm2, %xmm1	# round 4
1881	call _key_expansion_192b
1882	aeskeygenassist $0x10, %xmm2, %xmm1	# round 5
1883	call _key_expansion_192a
1884	aeskeygenassist $0x20, %xmm2, %xmm1	# round 6
1885	call _key_expansion_192b
1886	aeskeygenassist $0x40, %xmm2, %xmm1	# round 7
1887	call _key_expansion_192a
1888	aeskeygenassist $0x80, %xmm2, %xmm1	# round 8
1889	call _key_expansion_192b
1890	jmp .Ldec_key
1891.Lenc_key128:
1892	aeskeygenassist $0x1, %xmm0, %xmm1	# round 1
1893	call _key_expansion_128
1894	aeskeygenassist $0x2, %xmm0, %xmm1	# round 2
1895	call _key_expansion_128
1896	aeskeygenassist $0x4, %xmm0, %xmm1	# round 3
1897	call _key_expansion_128
1898	aeskeygenassist $0x8, %xmm0, %xmm1	# round 4
1899	call _key_expansion_128
1900	aeskeygenassist $0x10, %xmm0, %xmm1	# round 5
1901	call _key_expansion_128
1902	aeskeygenassist $0x20, %xmm0, %xmm1	# round 6
1903	call _key_expansion_128
1904	aeskeygenassist $0x40, %xmm0, %xmm1	# round 7
1905	call _key_expansion_128
1906	aeskeygenassist $0x80, %xmm0, %xmm1	# round 8
1907	call _key_expansion_128
1908	aeskeygenassist $0x1b, %xmm0, %xmm1	# round 9
1909	call _key_expansion_128
1910	aeskeygenassist $0x36, %xmm0, %xmm1	# round 10
1911	call _key_expansion_128
1912.Ldec_key:
1913	sub $0x10, TKEYP
1914	movaps (KEYP), %xmm0
1915	movaps (TKEYP), %xmm1
1916	movaps %xmm0, 240(TKEYP)
1917	movaps %xmm1, 240(KEYP)
1918	add $0x10, KEYP
1919	lea 240-16(TKEYP), UKEYP
1920.align 4
1921.Ldec_key_loop:
1922	movaps (KEYP), %xmm0
1923	aesimc %xmm0, %xmm1
1924	movaps %xmm1, (UKEYP)
1925	add $0x10, KEYP
1926	sub $0x10, UKEYP
1927	cmp TKEYP, KEYP
1928	jb .Ldec_key_loop
1929	xor AREG, AREG
1930#ifndef __x86_64__
1931	popl KEYP
1932#endif
1933	FRAME_END
1934	RET
1935SYM_FUNC_END(aesni_set_key)
1936
1937/*
1938 * void aesni_enc(const void *ctx, u8 *dst, const u8 *src)
1939 */
1940SYM_FUNC_START(aesni_enc)
1941	FRAME_BEGIN
1942#ifndef __x86_64__
1943	pushl KEYP
1944	pushl KLEN
1945	movl (FRAME_OFFSET+12)(%esp), KEYP	# ctx
1946	movl (FRAME_OFFSET+16)(%esp), OUTP	# dst
1947	movl (FRAME_OFFSET+20)(%esp), INP	# src
1948#endif
1949	movl 480(KEYP), KLEN		# key length
1950	movups (INP), STATE		# input
1951	call _aesni_enc1
1952	movups STATE, (OUTP)		# output
1953#ifndef __x86_64__
1954	popl KLEN
1955	popl KEYP
1956#endif
1957	FRAME_END
1958	RET
1959SYM_FUNC_END(aesni_enc)
1960
1961/*
1962 * _aesni_enc1:		internal ABI
1963 * input:
1964 *	KEYP:		key struct pointer
1965 *	KLEN:		round count
1966 *	STATE:		initial state (input)
1967 * output:
1968 *	STATE:		finial state (output)
1969 * changed:
1970 *	KEY
1971 *	TKEYP (T1)
1972 */
1973SYM_FUNC_START_LOCAL(_aesni_enc1)
1974	movaps (KEYP), KEY		# key
1975	mov KEYP, TKEYP
1976	pxor KEY, STATE		# round 0
1977	add $0x30, TKEYP
1978	cmp $24, KLEN
1979	jb .Lenc128
1980	lea 0x20(TKEYP), TKEYP
1981	je .Lenc192
1982	add $0x20, TKEYP
1983	movaps -0x60(TKEYP), KEY
1984	aesenc KEY, STATE
1985	movaps -0x50(TKEYP), KEY
1986	aesenc KEY, STATE
1987.align 4
1988.Lenc192:
1989	movaps -0x40(TKEYP), KEY
1990	aesenc KEY, STATE
1991	movaps -0x30(TKEYP), KEY
1992	aesenc KEY, STATE
1993.align 4
1994.Lenc128:
1995	movaps -0x20(TKEYP), KEY
1996	aesenc KEY, STATE
1997	movaps -0x10(TKEYP), KEY
1998	aesenc KEY, STATE
1999	movaps (TKEYP), KEY
2000	aesenc KEY, STATE
2001	movaps 0x10(TKEYP), KEY
2002	aesenc KEY, STATE
2003	movaps 0x20(TKEYP), KEY
2004	aesenc KEY, STATE
2005	movaps 0x30(TKEYP), KEY
2006	aesenc KEY, STATE
2007	movaps 0x40(TKEYP), KEY
2008	aesenc KEY, STATE
2009	movaps 0x50(TKEYP), KEY
2010	aesenc KEY, STATE
2011	movaps 0x60(TKEYP), KEY
2012	aesenc KEY, STATE
2013	movaps 0x70(TKEYP), KEY
2014	aesenclast KEY, STATE
2015	RET
2016SYM_FUNC_END(_aesni_enc1)
2017
2018/*
2019 * _aesni_enc4:	internal ABI
2020 * input:
2021 *	KEYP:		key struct pointer
2022 *	KLEN:		round count
2023 *	STATE1:		initial state (input)
2024 *	STATE2
2025 *	STATE3
2026 *	STATE4
2027 * output:
2028 *	STATE1:		finial state (output)
2029 *	STATE2
2030 *	STATE3
2031 *	STATE4
2032 * changed:
2033 *	KEY
2034 *	TKEYP (T1)
2035 */
2036SYM_FUNC_START_LOCAL(_aesni_enc4)
2037	movaps (KEYP), KEY		# key
2038	mov KEYP, TKEYP
2039	pxor KEY, STATE1		# round 0
2040	pxor KEY, STATE2
2041	pxor KEY, STATE3
2042	pxor KEY, STATE4
2043	add $0x30, TKEYP
2044	cmp $24, KLEN
2045	jb .L4enc128
2046	lea 0x20(TKEYP), TKEYP
2047	je .L4enc192
2048	add $0x20, TKEYP
2049	movaps -0x60(TKEYP), KEY
2050	aesenc KEY, STATE1
2051	aesenc KEY, STATE2
2052	aesenc KEY, STATE3
2053	aesenc KEY, STATE4
2054	movaps -0x50(TKEYP), KEY
2055	aesenc KEY, STATE1
2056	aesenc KEY, STATE2
2057	aesenc KEY, STATE3
2058	aesenc KEY, STATE4
2059#.align 4
2060.L4enc192:
2061	movaps -0x40(TKEYP), KEY
2062	aesenc KEY, STATE1
2063	aesenc KEY, STATE2
2064	aesenc KEY, STATE3
2065	aesenc KEY, STATE4
2066	movaps -0x30(TKEYP), KEY
2067	aesenc KEY, STATE1
2068	aesenc KEY, STATE2
2069	aesenc KEY, STATE3
2070	aesenc KEY, STATE4
2071#.align 4
2072.L4enc128:
2073	movaps -0x20(TKEYP), KEY
2074	aesenc KEY, STATE1
2075	aesenc KEY, STATE2
2076	aesenc KEY, STATE3
2077	aesenc KEY, STATE4
2078	movaps -0x10(TKEYP), KEY
2079	aesenc KEY, STATE1
2080	aesenc KEY, STATE2
2081	aesenc KEY, STATE3
2082	aesenc KEY, STATE4
2083	movaps (TKEYP), KEY
2084	aesenc KEY, STATE1
2085	aesenc KEY, STATE2
2086	aesenc KEY, STATE3
2087	aesenc KEY, STATE4
2088	movaps 0x10(TKEYP), KEY
2089	aesenc KEY, STATE1
2090	aesenc KEY, STATE2
2091	aesenc KEY, STATE3
2092	aesenc KEY, STATE4
2093	movaps 0x20(TKEYP), KEY
2094	aesenc KEY, STATE1
2095	aesenc KEY, STATE2
2096	aesenc KEY, STATE3
2097	aesenc KEY, STATE4
2098	movaps 0x30(TKEYP), KEY
2099	aesenc KEY, STATE1
2100	aesenc KEY, STATE2
2101	aesenc KEY, STATE3
2102	aesenc KEY, STATE4
2103	movaps 0x40(TKEYP), KEY
2104	aesenc KEY, STATE1
2105	aesenc KEY, STATE2
2106	aesenc KEY, STATE3
2107	aesenc KEY, STATE4
2108	movaps 0x50(TKEYP), KEY
2109	aesenc KEY, STATE1
2110	aesenc KEY, STATE2
2111	aesenc KEY, STATE3
2112	aesenc KEY, STATE4
2113	movaps 0x60(TKEYP), KEY
2114	aesenc KEY, STATE1
2115	aesenc KEY, STATE2
2116	aesenc KEY, STATE3
2117	aesenc KEY, STATE4
2118	movaps 0x70(TKEYP), KEY
2119	aesenclast KEY, STATE1		# last round
2120	aesenclast KEY, STATE2
2121	aesenclast KEY, STATE3
2122	aesenclast KEY, STATE4
2123	RET
2124SYM_FUNC_END(_aesni_enc4)
2125
2126/*
2127 * void aesni_dec (const void *ctx, u8 *dst, const u8 *src)
2128 */
2129SYM_FUNC_START(aesni_dec)
2130	FRAME_BEGIN
2131#ifndef __x86_64__
2132	pushl KEYP
2133	pushl KLEN
2134	movl (FRAME_OFFSET+12)(%esp), KEYP	# ctx
2135	movl (FRAME_OFFSET+16)(%esp), OUTP	# dst
2136	movl (FRAME_OFFSET+20)(%esp), INP	# src
2137#endif
2138	mov 480(KEYP), KLEN		# key length
2139	add $240, KEYP
2140	movups (INP), STATE		# input
2141	call _aesni_dec1
2142	movups STATE, (OUTP)		#output
2143#ifndef __x86_64__
2144	popl KLEN
2145	popl KEYP
2146#endif
2147	FRAME_END
2148	RET
2149SYM_FUNC_END(aesni_dec)
2150
2151/*
2152 * _aesni_dec1:		internal ABI
2153 * input:
2154 *	KEYP:		key struct pointer
2155 *	KLEN:		key length
2156 *	STATE:		initial state (input)
2157 * output:
2158 *	STATE:		finial state (output)
2159 * changed:
2160 *	KEY
2161 *	TKEYP (T1)
2162 */
2163SYM_FUNC_START_LOCAL(_aesni_dec1)
2164	movaps (KEYP), KEY		# key
2165	mov KEYP, TKEYP
2166	pxor KEY, STATE		# round 0
2167	add $0x30, TKEYP
2168	cmp $24, KLEN
2169	jb .Ldec128
2170	lea 0x20(TKEYP), TKEYP
2171	je .Ldec192
2172	add $0x20, TKEYP
2173	movaps -0x60(TKEYP), KEY
2174	aesdec KEY, STATE
2175	movaps -0x50(TKEYP), KEY
2176	aesdec KEY, STATE
2177.align 4
2178.Ldec192:
2179	movaps -0x40(TKEYP), KEY
2180	aesdec KEY, STATE
2181	movaps -0x30(TKEYP), KEY
2182	aesdec KEY, STATE
2183.align 4
2184.Ldec128:
2185	movaps -0x20(TKEYP), KEY
2186	aesdec KEY, STATE
2187	movaps -0x10(TKEYP), KEY
2188	aesdec KEY, STATE
2189	movaps (TKEYP), KEY
2190	aesdec KEY, STATE
2191	movaps 0x10(TKEYP), KEY
2192	aesdec KEY, STATE
2193	movaps 0x20(TKEYP), KEY
2194	aesdec KEY, STATE
2195	movaps 0x30(TKEYP), KEY
2196	aesdec KEY, STATE
2197	movaps 0x40(TKEYP), KEY
2198	aesdec KEY, STATE
2199	movaps 0x50(TKEYP), KEY
2200	aesdec KEY, STATE
2201	movaps 0x60(TKEYP), KEY
2202	aesdec KEY, STATE
2203	movaps 0x70(TKEYP), KEY
2204	aesdeclast KEY, STATE
2205	RET
2206SYM_FUNC_END(_aesni_dec1)
2207
2208/*
2209 * _aesni_dec4:	internal ABI
2210 * input:
2211 *	KEYP:		key struct pointer
2212 *	KLEN:		key length
2213 *	STATE1:		initial state (input)
2214 *	STATE2
2215 *	STATE3
2216 *	STATE4
2217 * output:
2218 *	STATE1:		finial state (output)
2219 *	STATE2
2220 *	STATE3
2221 *	STATE4
2222 * changed:
2223 *	KEY
2224 *	TKEYP (T1)
2225 */
2226SYM_FUNC_START_LOCAL(_aesni_dec4)
2227	movaps (KEYP), KEY		# key
2228	mov KEYP, TKEYP
2229	pxor KEY, STATE1		# round 0
2230	pxor KEY, STATE2
2231	pxor KEY, STATE3
2232	pxor KEY, STATE4
2233	add $0x30, TKEYP
2234	cmp $24, KLEN
2235	jb .L4dec128
2236	lea 0x20(TKEYP), TKEYP
2237	je .L4dec192
2238	add $0x20, TKEYP
2239	movaps -0x60(TKEYP), KEY
2240	aesdec KEY, STATE1
2241	aesdec KEY, STATE2
2242	aesdec KEY, STATE3
2243	aesdec KEY, STATE4
2244	movaps -0x50(TKEYP), KEY
2245	aesdec KEY, STATE1
2246	aesdec KEY, STATE2
2247	aesdec KEY, STATE3
2248	aesdec KEY, STATE4
2249.align 4
2250.L4dec192:
2251	movaps -0x40(TKEYP), KEY
2252	aesdec KEY, STATE1
2253	aesdec KEY, STATE2
2254	aesdec KEY, STATE3
2255	aesdec KEY, STATE4
2256	movaps -0x30(TKEYP), KEY
2257	aesdec KEY, STATE1
2258	aesdec KEY, STATE2
2259	aesdec KEY, STATE3
2260	aesdec KEY, STATE4
2261.align 4
2262.L4dec128:
2263	movaps -0x20(TKEYP), KEY
2264	aesdec KEY, STATE1
2265	aesdec KEY, STATE2
2266	aesdec KEY, STATE3
2267	aesdec KEY, STATE4
2268	movaps -0x10(TKEYP), KEY
2269	aesdec KEY, STATE1
2270	aesdec KEY, STATE2
2271	aesdec KEY, STATE3
2272	aesdec KEY, STATE4
2273	movaps (TKEYP), KEY
2274	aesdec KEY, STATE1
2275	aesdec KEY, STATE2
2276	aesdec KEY, STATE3
2277	aesdec KEY, STATE4
2278	movaps 0x10(TKEYP), KEY
2279	aesdec KEY, STATE1
2280	aesdec KEY, STATE2
2281	aesdec KEY, STATE3
2282	aesdec KEY, STATE4
2283	movaps 0x20(TKEYP), KEY
2284	aesdec KEY, STATE1
2285	aesdec KEY, STATE2
2286	aesdec KEY, STATE3
2287	aesdec KEY, STATE4
2288	movaps 0x30(TKEYP), KEY
2289	aesdec KEY, STATE1
2290	aesdec KEY, STATE2
2291	aesdec KEY, STATE3
2292	aesdec KEY, STATE4
2293	movaps 0x40(TKEYP), KEY
2294	aesdec KEY, STATE1
2295	aesdec KEY, STATE2
2296	aesdec KEY, STATE3
2297	aesdec KEY, STATE4
2298	movaps 0x50(TKEYP), KEY
2299	aesdec KEY, STATE1
2300	aesdec KEY, STATE2
2301	aesdec KEY, STATE3
2302	aesdec KEY, STATE4
2303	movaps 0x60(TKEYP), KEY
2304	aesdec KEY, STATE1
2305	aesdec KEY, STATE2
2306	aesdec KEY, STATE3
2307	aesdec KEY, STATE4
2308	movaps 0x70(TKEYP), KEY
2309	aesdeclast KEY, STATE1		# last round
2310	aesdeclast KEY, STATE2
2311	aesdeclast KEY, STATE3
2312	aesdeclast KEY, STATE4
2313	RET
2314SYM_FUNC_END(_aesni_dec4)
2315
2316/*
2317 * void aesni_ecb_enc(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
2318 *		      size_t len)
2319 */
2320SYM_FUNC_START(aesni_ecb_enc)
2321	FRAME_BEGIN
2322#ifndef __x86_64__
2323	pushl LEN
2324	pushl KEYP
2325	pushl KLEN
2326	movl (FRAME_OFFSET+16)(%esp), KEYP	# ctx
2327	movl (FRAME_OFFSET+20)(%esp), OUTP	# dst
2328	movl (FRAME_OFFSET+24)(%esp), INP	# src
2329	movl (FRAME_OFFSET+28)(%esp), LEN	# len
2330#endif
2331	test LEN, LEN		# check length
2332	jz .Lecb_enc_ret
2333	mov 480(KEYP), KLEN
2334	cmp $16, LEN
2335	jb .Lecb_enc_ret
2336	cmp $64, LEN
2337	jb .Lecb_enc_loop1
2338.align 4
2339.Lecb_enc_loop4:
2340	movups (INP), STATE1
2341	movups 0x10(INP), STATE2
2342	movups 0x20(INP), STATE3
2343	movups 0x30(INP), STATE4
2344	call _aesni_enc4
2345	movups STATE1, (OUTP)
2346	movups STATE2, 0x10(OUTP)
2347	movups STATE3, 0x20(OUTP)
2348	movups STATE4, 0x30(OUTP)
2349	sub $64, LEN
2350	add $64, INP
2351	add $64, OUTP
2352	cmp $64, LEN
2353	jge .Lecb_enc_loop4
2354	cmp $16, LEN
2355	jb .Lecb_enc_ret
2356.align 4
2357.Lecb_enc_loop1:
2358	movups (INP), STATE1
2359	call _aesni_enc1
2360	movups STATE1, (OUTP)
2361	sub $16, LEN
2362	add $16, INP
2363	add $16, OUTP
2364	cmp $16, LEN
2365	jge .Lecb_enc_loop1
2366.Lecb_enc_ret:
2367#ifndef __x86_64__
2368	popl KLEN
2369	popl KEYP
2370	popl LEN
2371#endif
2372	FRAME_END
2373	RET
2374SYM_FUNC_END(aesni_ecb_enc)
2375
2376/*
2377 * void aesni_ecb_dec(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
2378 *		      size_t len);
2379 */
2380SYM_FUNC_START(aesni_ecb_dec)
2381	FRAME_BEGIN
2382#ifndef __x86_64__
2383	pushl LEN
2384	pushl KEYP
2385	pushl KLEN
2386	movl (FRAME_OFFSET+16)(%esp), KEYP	# ctx
2387	movl (FRAME_OFFSET+20)(%esp), OUTP	# dst
2388	movl (FRAME_OFFSET+24)(%esp), INP	# src
2389	movl (FRAME_OFFSET+28)(%esp), LEN	# len
2390#endif
2391	test LEN, LEN
2392	jz .Lecb_dec_ret
2393	mov 480(KEYP), KLEN
2394	add $240, KEYP
2395	cmp $16, LEN
2396	jb .Lecb_dec_ret
2397	cmp $64, LEN
2398	jb .Lecb_dec_loop1
2399.align 4
2400.Lecb_dec_loop4:
2401	movups (INP), STATE1
2402	movups 0x10(INP), STATE2
2403	movups 0x20(INP), STATE3
2404	movups 0x30(INP), STATE4
2405	call _aesni_dec4
2406	movups STATE1, (OUTP)
2407	movups STATE2, 0x10(OUTP)
2408	movups STATE3, 0x20(OUTP)
2409	movups STATE4, 0x30(OUTP)
2410	sub $64, LEN
2411	add $64, INP
2412	add $64, OUTP
2413	cmp $64, LEN
2414	jge .Lecb_dec_loop4
2415	cmp $16, LEN
2416	jb .Lecb_dec_ret
2417.align 4
2418.Lecb_dec_loop1:
2419	movups (INP), STATE1
2420	call _aesni_dec1
2421	movups STATE1, (OUTP)
2422	sub $16, LEN
2423	add $16, INP
2424	add $16, OUTP
2425	cmp $16, LEN
2426	jge .Lecb_dec_loop1
2427.Lecb_dec_ret:
2428#ifndef __x86_64__
2429	popl KLEN
2430	popl KEYP
2431	popl LEN
2432#endif
2433	FRAME_END
2434	RET
2435SYM_FUNC_END(aesni_ecb_dec)
2436
2437/*
2438 * void aesni_cbc_enc(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
2439 *		      size_t len, u8 *iv)
2440 */
2441SYM_FUNC_START(aesni_cbc_enc)
2442	FRAME_BEGIN
2443#ifndef __x86_64__
2444	pushl IVP
2445	pushl LEN
2446	pushl KEYP
2447	pushl KLEN
2448	movl (FRAME_OFFSET+20)(%esp), KEYP	# ctx
2449	movl (FRAME_OFFSET+24)(%esp), OUTP	# dst
2450	movl (FRAME_OFFSET+28)(%esp), INP	# src
2451	movl (FRAME_OFFSET+32)(%esp), LEN	# len
2452	movl (FRAME_OFFSET+36)(%esp), IVP	# iv
2453#endif
2454	cmp $16, LEN
2455	jb .Lcbc_enc_ret
2456	mov 480(KEYP), KLEN
2457	movups (IVP), STATE	# load iv as initial state
2458.align 4
2459.Lcbc_enc_loop:
2460	movups (INP), IN	# load input
2461	pxor IN, STATE
2462	call _aesni_enc1
2463	movups STATE, (OUTP)	# store output
2464	sub $16, LEN
2465	add $16, INP
2466	add $16, OUTP
2467	cmp $16, LEN
2468	jge .Lcbc_enc_loop
2469	movups STATE, (IVP)
2470.Lcbc_enc_ret:
2471#ifndef __x86_64__
2472	popl KLEN
2473	popl KEYP
2474	popl LEN
2475	popl IVP
2476#endif
2477	FRAME_END
2478	RET
2479SYM_FUNC_END(aesni_cbc_enc)
2480
2481/*
2482 * void aesni_cbc_dec(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
2483 *		      size_t len, u8 *iv)
2484 */
2485SYM_FUNC_START(aesni_cbc_dec)
2486	FRAME_BEGIN
2487#ifndef __x86_64__
2488	pushl IVP
2489	pushl LEN
2490	pushl KEYP
2491	pushl KLEN
2492	movl (FRAME_OFFSET+20)(%esp), KEYP	# ctx
2493	movl (FRAME_OFFSET+24)(%esp), OUTP	# dst
2494	movl (FRAME_OFFSET+28)(%esp), INP	# src
2495	movl (FRAME_OFFSET+32)(%esp), LEN	# len
2496	movl (FRAME_OFFSET+36)(%esp), IVP	# iv
2497#endif
2498	cmp $16, LEN
2499	jb .Lcbc_dec_just_ret
2500	mov 480(KEYP), KLEN
2501	add $240, KEYP
2502	movups (IVP), IV
2503	cmp $64, LEN
2504	jb .Lcbc_dec_loop1
2505.align 4
2506.Lcbc_dec_loop4:
2507	movups (INP), IN1
2508	movaps IN1, STATE1
2509	movups 0x10(INP), IN2
2510	movaps IN2, STATE2
2511#ifdef __x86_64__
2512	movups 0x20(INP), IN3
2513	movaps IN3, STATE3
2514	movups 0x30(INP), IN4
2515	movaps IN4, STATE4
2516#else
2517	movups 0x20(INP), IN1
2518	movaps IN1, STATE3
2519	movups 0x30(INP), IN2
2520	movaps IN2, STATE4
2521#endif
2522	call _aesni_dec4
2523	pxor IV, STATE1
2524#ifdef __x86_64__
2525	pxor IN1, STATE2
2526	pxor IN2, STATE3
2527	pxor IN3, STATE4
2528	movaps IN4, IV
2529#else
2530	pxor IN1, STATE4
2531	movaps IN2, IV
2532	movups (INP), IN1
2533	pxor IN1, STATE2
2534	movups 0x10(INP), IN2
2535	pxor IN2, STATE3
2536#endif
2537	movups STATE1, (OUTP)
2538	movups STATE2, 0x10(OUTP)
2539	movups STATE3, 0x20(OUTP)
2540	movups STATE4, 0x30(OUTP)
2541	sub $64, LEN
2542	add $64, INP
2543	add $64, OUTP
2544	cmp $64, LEN
2545	jge .Lcbc_dec_loop4
2546	cmp $16, LEN
2547	jb .Lcbc_dec_ret
2548.align 4
2549.Lcbc_dec_loop1:
2550	movups (INP), IN
2551	movaps IN, STATE
2552	call _aesni_dec1
2553	pxor IV, STATE
2554	movups STATE, (OUTP)
2555	movaps IN, IV
2556	sub $16, LEN
2557	add $16, INP
2558	add $16, OUTP
2559	cmp $16, LEN
2560	jge .Lcbc_dec_loop1
2561.Lcbc_dec_ret:
2562	movups IV, (IVP)
2563.Lcbc_dec_just_ret:
2564#ifndef __x86_64__
2565	popl KLEN
2566	popl KEYP
2567	popl LEN
2568	popl IVP
2569#endif
2570	FRAME_END
2571	RET
2572SYM_FUNC_END(aesni_cbc_dec)
2573
2574/*
2575 * void aesni_cts_cbc_enc(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
2576 *			  size_t len, u8 *iv)
2577 */
2578SYM_FUNC_START(aesni_cts_cbc_enc)
2579	FRAME_BEGIN
2580#ifndef __x86_64__
2581	pushl IVP
2582	pushl LEN
2583	pushl KEYP
2584	pushl KLEN
2585	movl (FRAME_OFFSET+20)(%esp), KEYP	# ctx
2586	movl (FRAME_OFFSET+24)(%esp), OUTP	# dst
2587	movl (FRAME_OFFSET+28)(%esp), INP	# src
2588	movl (FRAME_OFFSET+32)(%esp), LEN	# len
2589	movl (FRAME_OFFSET+36)(%esp), IVP	# iv
2590	lea .Lcts_permute_table, T1
2591#else
2592	lea .Lcts_permute_table(%rip), T1
2593#endif
2594	mov 480(KEYP), KLEN
2595	movups (IVP), STATE
2596	sub $16, LEN
2597	mov T1, IVP
2598	add $32, IVP
2599	add LEN, T1
2600	sub LEN, IVP
2601	movups (T1), %xmm4
2602	movups (IVP), %xmm5
2603
2604	movups (INP), IN1
2605	add LEN, INP
2606	movups (INP), IN2
2607
2608	pxor IN1, STATE
2609	call _aesni_enc1
2610
2611	pshufb %xmm5, IN2
2612	pxor STATE, IN2
2613	pshufb %xmm4, STATE
2614	add OUTP, LEN
2615	movups STATE, (LEN)
2616
2617	movaps IN2, STATE
2618	call _aesni_enc1
2619	movups STATE, (OUTP)
2620
2621#ifndef __x86_64__
2622	popl KLEN
2623	popl KEYP
2624	popl LEN
2625	popl IVP
2626#endif
2627	FRAME_END
2628	RET
2629SYM_FUNC_END(aesni_cts_cbc_enc)
2630
2631/*
2632 * void aesni_cts_cbc_dec(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
2633 *			  size_t len, u8 *iv)
2634 */
2635SYM_FUNC_START(aesni_cts_cbc_dec)
2636	FRAME_BEGIN
2637#ifndef __x86_64__
2638	pushl IVP
2639	pushl LEN
2640	pushl KEYP
2641	pushl KLEN
2642	movl (FRAME_OFFSET+20)(%esp), KEYP	# ctx
2643	movl (FRAME_OFFSET+24)(%esp), OUTP	# dst
2644	movl (FRAME_OFFSET+28)(%esp), INP	# src
2645	movl (FRAME_OFFSET+32)(%esp), LEN	# len
2646	movl (FRAME_OFFSET+36)(%esp), IVP	# iv
2647	lea .Lcts_permute_table, T1
2648#else
2649	lea .Lcts_permute_table(%rip), T1
2650#endif
2651	mov 480(KEYP), KLEN
2652	add $240, KEYP
2653	movups (IVP), IV
2654	sub $16, LEN
2655	mov T1, IVP
2656	add $32, IVP
2657	add LEN, T1
2658	sub LEN, IVP
2659	movups (T1), %xmm4
2660
2661	movups (INP), STATE
2662	add LEN, INP
2663	movups (INP), IN1
2664
2665	call _aesni_dec1
2666	movaps STATE, IN2
2667	pshufb %xmm4, STATE
2668	pxor IN1, STATE
2669
2670	add OUTP, LEN
2671	movups STATE, (LEN)
2672
2673	movups (IVP), %xmm0
2674	pshufb %xmm0, IN1
2675	pblendvb IN2, IN1
2676	movaps IN1, STATE
2677	call _aesni_dec1
2678
2679	pxor IV, STATE
2680	movups STATE, (OUTP)
2681
2682#ifndef __x86_64__
2683	popl KLEN
2684	popl KEYP
2685	popl LEN
2686	popl IVP
2687#endif
2688	FRAME_END
2689	RET
2690SYM_FUNC_END(aesni_cts_cbc_dec)
2691
2692.pushsection .rodata
2693.align 16
2694.Lcts_permute_table:
2695	.byte		0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
2696	.byte		0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
2697	.byte		0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07
2698	.byte		0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f
2699	.byte		0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
2700	.byte		0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
2701#ifdef __x86_64__
2702.Lbswap_mask:
2703	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
2704#endif
2705.popsection
2706
2707#ifdef __x86_64__
2708/*
2709 * _aesni_inc_init:	internal ABI
2710 *	setup registers used by _aesni_inc
2711 * input:
2712 *	IV
2713 * output:
2714 *	CTR:	== IV, in little endian
2715 *	TCTR_LOW: == lower qword of CTR
2716 *	INC:	== 1, in little endian
2717 *	BSWAP_MASK == endian swapping mask
2718 */
2719SYM_FUNC_START_LOCAL(_aesni_inc_init)
2720	movaps .Lbswap_mask(%rip), BSWAP_MASK
2721	movaps IV, CTR
2722	pshufb BSWAP_MASK, CTR
2723	mov $1, TCTR_LOW
2724	movq TCTR_LOW, INC
2725	movq CTR, TCTR_LOW
2726	RET
2727SYM_FUNC_END(_aesni_inc_init)
2728
2729/*
2730 * _aesni_inc:		internal ABI
2731 *	Increase IV by 1, IV is in big endian
2732 * input:
2733 *	IV
2734 *	CTR:	== IV, in little endian
2735 *	TCTR_LOW: == lower qword of CTR
2736 *	INC:	== 1, in little endian
2737 *	BSWAP_MASK == endian swapping mask
2738 * output:
2739 *	IV:	Increase by 1
2740 * changed:
2741 *	CTR:	== output IV, in little endian
2742 *	TCTR_LOW: == lower qword of CTR
2743 */
2744SYM_FUNC_START_LOCAL(_aesni_inc)
2745	paddq INC, CTR
2746	add $1, TCTR_LOW
2747	jnc .Linc_low
2748	pslldq $8, INC
2749	paddq INC, CTR
2750	psrldq $8, INC
2751.Linc_low:
2752	movaps CTR, IV
2753	pshufb BSWAP_MASK, IV
2754	RET
2755SYM_FUNC_END(_aesni_inc)
2756
2757/*
2758 * void aesni_ctr_enc(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
2759 *		      size_t len, u8 *iv)
2760 */
2761SYM_FUNC_START(aesni_ctr_enc)
2762	FRAME_BEGIN
2763	cmp $16, LEN
2764	jb .Lctr_enc_just_ret
2765	mov 480(KEYP), KLEN
2766	movups (IVP), IV
2767	call _aesni_inc_init
2768	cmp $64, LEN
2769	jb .Lctr_enc_loop1
2770.align 4
2771.Lctr_enc_loop4:
2772	movaps IV, STATE1
2773	call _aesni_inc
2774	movups (INP), IN1
2775	movaps IV, STATE2
2776	call _aesni_inc
2777	movups 0x10(INP), IN2
2778	movaps IV, STATE3
2779	call _aesni_inc
2780	movups 0x20(INP), IN3
2781	movaps IV, STATE4
2782	call _aesni_inc
2783	movups 0x30(INP), IN4
2784	call _aesni_enc4
2785	pxor IN1, STATE1
2786	movups STATE1, (OUTP)
2787	pxor IN2, STATE2
2788	movups STATE2, 0x10(OUTP)
2789	pxor IN3, STATE3
2790	movups STATE3, 0x20(OUTP)
2791	pxor IN4, STATE4
2792	movups STATE4, 0x30(OUTP)
2793	sub $64, LEN
2794	add $64, INP
2795	add $64, OUTP
2796	cmp $64, LEN
2797	jge .Lctr_enc_loop4
2798	cmp $16, LEN
2799	jb .Lctr_enc_ret
2800.align 4
2801.Lctr_enc_loop1:
2802	movaps IV, STATE
2803	call _aesni_inc
2804	movups (INP), IN
2805	call _aesni_enc1
2806	pxor IN, STATE
2807	movups STATE, (OUTP)
2808	sub $16, LEN
2809	add $16, INP
2810	add $16, OUTP
2811	cmp $16, LEN
2812	jge .Lctr_enc_loop1
2813.Lctr_enc_ret:
2814	movups IV, (IVP)
2815.Lctr_enc_just_ret:
2816	FRAME_END
2817	RET
2818SYM_FUNC_END(aesni_ctr_enc)
2819
2820#endif
2821
2822.section	.rodata.cst16.gf128mul_x_ble_mask, "aM", @progbits, 16
2823.align 16
2824.Lgf128mul_x_ble_mask:
2825	.octa 0x00000000000000010000000000000087
2826.previous
2827
2828/*
2829 * _aesni_gf128mul_x_ble:		internal ABI
2830 *	Multiply in GF(2^128) for XTS IVs
2831 * input:
2832 *	IV:	current IV
2833 *	GF128MUL_MASK == mask with 0x87 and 0x01
2834 * output:
2835 *	IV:	next IV
2836 * changed:
2837 *	CTR:	== temporary value
2838 */
2839#define _aesni_gf128mul_x_ble() \
2840	pshufd $0x13, IV, KEY; \
2841	paddq IV, IV; \
2842	psrad $31, KEY; \
2843	pand GF128MUL_MASK, KEY; \
2844	pxor KEY, IV;
2845
2846/*
2847 * void aesni_xts_encrypt(const struct crypto_aes_ctx *ctx, u8 *dst,
2848 *			  const u8 *src, unsigned int len, le128 *iv)
2849 */
2850SYM_FUNC_START(aesni_xts_encrypt)
2851	FRAME_BEGIN
2852#ifndef __x86_64__
2853	pushl IVP
2854	pushl LEN
2855	pushl KEYP
2856	pushl KLEN
2857	movl (FRAME_OFFSET+20)(%esp), KEYP	# ctx
2858	movl (FRAME_OFFSET+24)(%esp), OUTP	# dst
2859	movl (FRAME_OFFSET+28)(%esp), INP	# src
2860	movl (FRAME_OFFSET+32)(%esp), LEN	# len
2861	movl (FRAME_OFFSET+36)(%esp), IVP	# iv
2862	movdqa .Lgf128mul_x_ble_mask, GF128MUL_MASK
2863#else
2864	movdqa .Lgf128mul_x_ble_mask(%rip), GF128MUL_MASK
2865#endif
2866	movups (IVP), IV
2867
2868	mov 480(KEYP), KLEN
2869
2870.Lxts_enc_loop4:
2871	sub $64, LEN
2872	jl .Lxts_enc_1x
2873
2874	movdqa IV, STATE1
2875	movdqu 0x00(INP), IN
2876	pxor IN, STATE1
2877	movdqu IV, 0x00(OUTP)
2878
2879	_aesni_gf128mul_x_ble()
2880	movdqa IV, STATE2
2881	movdqu 0x10(INP), IN
2882	pxor IN, STATE2
2883	movdqu IV, 0x10(OUTP)
2884
2885	_aesni_gf128mul_x_ble()
2886	movdqa IV, STATE3
2887	movdqu 0x20(INP), IN
2888	pxor IN, STATE3
2889	movdqu IV, 0x20(OUTP)
2890
2891	_aesni_gf128mul_x_ble()
2892	movdqa IV, STATE4
2893	movdqu 0x30(INP), IN
2894	pxor IN, STATE4
2895	movdqu IV, 0x30(OUTP)
2896
2897	call _aesni_enc4
2898
2899	movdqu 0x00(OUTP), IN
2900	pxor IN, STATE1
2901	movdqu STATE1, 0x00(OUTP)
2902
2903	movdqu 0x10(OUTP), IN
2904	pxor IN, STATE2
2905	movdqu STATE2, 0x10(OUTP)
2906
2907	movdqu 0x20(OUTP), IN
2908	pxor IN, STATE3
2909	movdqu STATE3, 0x20(OUTP)
2910
2911	movdqu 0x30(OUTP), IN
2912	pxor IN, STATE4
2913	movdqu STATE4, 0x30(OUTP)
2914
2915	_aesni_gf128mul_x_ble()
2916
2917	add $64, INP
2918	add $64, OUTP
2919	test LEN, LEN
2920	jnz .Lxts_enc_loop4
2921
2922.Lxts_enc_ret_iv:
2923	movups IV, (IVP)
2924
2925.Lxts_enc_ret:
2926#ifndef __x86_64__
2927	popl KLEN
2928	popl KEYP
2929	popl LEN
2930	popl IVP
2931#endif
2932	FRAME_END
2933	RET
2934
2935.Lxts_enc_1x:
2936	add $64, LEN
2937	jz .Lxts_enc_ret_iv
2938	sub $16, LEN
2939	jl .Lxts_enc_cts4
2940
2941.Lxts_enc_loop1:
2942	movdqu (INP), STATE
2943	pxor IV, STATE
2944	call _aesni_enc1
2945	pxor IV, STATE
2946	_aesni_gf128mul_x_ble()
2947
2948	test LEN, LEN
2949	jz .Lxts_enc_out
2950
2951	add $16, INP
2952	sub $16, LEN
2953	jl .Lxts_enc_cts1
2954
2955	movdqu STATE, (OUTP)
2956	add $16, OUTP
2957	jmp .Lxts_enc_loop1
2958
2959.Lxts_enc_out:
2960	movdqu STATE, (OUTP)
2961	jmp .Lxts_enc_ret_iv
2962
2963.Lxts_enc_cts4:
2964	movdqa STATE4, STATE
2965	sub $16, OUTP
2966
2967.Lxts_enc_cts1:
2968#ifndef __x86_64__
2969	lea .Lcts_permute_table, T1
2970#else
2971	lea .Lcts_permute_table(%rip), T1
2972#endif
2973	add LEN, INP		/* rewind input pointer */
2974	add $16, LEN		/* # bytes in final block */
2975	movups (INP), IN1
2976
2977	mov T1, IVP
2978	add $32, IVP
2979	add LEN, T1
2980	sub LEN, IVP
2981	add OUTP, LEN
2982
2983	movups (T1), %xmm4
2984	movaps STATE, IN2
2985	pshufb %xmm4, STATE
2986	movups STATE, (LEN)
2987
2988	movups (IVP), %xmm0
2989	pshufb %xmm0, IN1
2990	pblendvb IN2, IN1
2991	movaps IN1, STATE
2992
2993	pxor IV, STATE
2994	call _aesni_enc1
2995	pxor IV, STATE
2996
2997	movups STATE, (OUTP)
2998	jmp .Lxts_enc_ret
2999SYM_FUNC_END(aesni_xts_encrypt)
3000
3001/*
3002 * void aesni_xts_decrypt(const struct crypto_aes_ctx *ctx, u8 *dst,
3003 *			  const u8 *src, unsigned int len, le128 *iv)
3004 */
3005SYM_FUNC_START(aesni_xts_decrypt)
3006	FRAME_BEGIN
3007#ifndef __x86_64__
3008	pushl IVP
3009	pushl LEN
3010	pushl KEYP
3011	pushl KLEN
3012	movl (FRAME_OFFSET+20)(%esp), KEYP	# ctx
3013	movl (FRAME_OFFSET+24)(%esp), OUTP	# dst
3014	movl (FRAME_OFFSET+28)(%esp), INP	# src
3015	movl (FRAME_OFFSET+32)(%esp), LEN	# len
3016	movl (FRAME_OFFSET+36)(%esp), IVP	# iv
3017	movdqa .Lgf128mul_x_ble_mask, GF128MUL_MASK
3018#else
3019	movdqa .Lgf128mul_x_ble_mask(%rip), GF128MUL_MASK
3020#endif
3021	movups (IVP), IV
3022
3023	mov 480(KEYP), KLEN
3024	add $240, KEYP
3025
3026	test $15, LEN
3027	jz .Lxts_dec_loop4
3028	sub $16, LEN
3029
3030.Lxts_dec_loop4:
3031	sub $64, LEN
3032	jl .Lxts_dec_1x
3033
3034	movdqa IV, STATE1
3035	movdqu 0x00(INP), IN
3036	pxor IN, STATE1
3037	movdqu IV, 0x00(OUTP)
3038
3039	_aesni_gf128mul_x_ble()
3040	movdqa IV, STATE2
3041	movdqu 0x10(INP), IN
3042	pxor IN, STATE2
3043	movdqu IV, 0x10(OUTP)
3044
3045	_aesni_gf128mul_x_ble()
3046	movdqa IV, STATE3
3047	movdqu 0x20(INP), IN
3048	pxor IN, STATE3
3049	movdqu IV, 0x20(OUTP)
3050
3051	_aesni_gf128mul_x_ble()
3052	movdqa IV, STATE4
3053	movdqu 0x30(INP), IN
3054	pxor IN, STATE4
3055	movdqu IV, 0x30(OUTP)
3056
3057	call _aesni_dec4
3058
3059	movdqu 0x00(OUTP), IN
3060	pxor IN, STATE1
3061	movdqu STATE1, 0x00(OUTP)
3062
3063	movdqu 0x10(OUTP), IN
3064	pxor IN, STATE2
3065	movdqu STATE2, 0x10(OUTP)
3066
3067	movdqu 0x20(OUTP), IN
3068	pxor IN, STATE3
3069	movdqu STATE3, 0x20(OUTP)
3070
3071	movdqu 0x30(OUTP), IN
3072	pxor IN, STATE4
3073	movdqu STATE4, 0x30(OUTP)
3074
3075	_aesni_gf128mul_x_ble()
3076
3077	add $64, INP
3078	add $64, OUTP
3079	test LEN, LEN
3080	jnz .Lxts_dec_loop4
3081
3082.Lxts_dec_ret_iv:
3083	movups IV, (IVP)
3084
3085.Lxts_dec_ret:
3086#ifndef __x86_64__
3087	popl KLEN
3088	popl KEYP
3089	popl LEN
3090	popl IVP
3091#endif
3092	FRAME_END
3093	RET
3094
3095.Lxts_dec_1x:
3096	add $64, LEN
3097	jz .Lxts_dec_ret_iv
3098
3099.Lxts_dec_loop1:
3100	movdqu (INP), STATE
3101
3102	add $16, INP
3103	sub $16, LEN
3104	jl .Lxts_dec_cts1
3105
3106	pxor IV, STATE
3107	call _aesni_dec1
3108	pxor IV, STATE
3109	_aesni_gf128mul_x_ble()
3110
3111	test LEN, LEN
3112	jz .Lxts_dec_out
3113
3114	movdqu STATE, (OUTP)
3115	add $16, OUTP
3116	jmp .Lxts_dec_loop1
3117
3118.Lxts_dec_out:
3119	movdqu STATE, (OUTP)
3120	jmp .Lxts_dec_ret_iv
3121
3122.Lxts_dec_cts1:
3123	movdqa IV, STATE4
3124	_aesni_gf128mul_x_ble()
3125
3126	pxor IV, STATE
3127	call _aesni_dec1
3128	pxor IV, STATE
3129
3130#ifndef __x86_64__
3131	lea .Lcts_permute_table, T1
3132#else
3133	lea .Lcts_permute_table(%rip), T1
3134#endif
3135	add LEN, INP		/* rewind input pointer */
3136	add $16, LEN		/* # bytes in final block */
3137	movups (INP), IN1
3138
3139	mov T1, IVP
3140	add $32, IVP
3141	add LEN, T1
3142	sub LEN, IVP
3143	add OUTP, LEN
3144
3145	movups (T1), %xmm4
3146	movaps STATE, IN2
3147	pshufb %xmm4, STATE
3148	movups STATE, (LEN)
3149
3150	movups (IVP), %xmm0
3151	pshufb %xmm0, IN1
3152	pblendvb IN2, IN1
3153	movaps IN1, STATE
3154
3155	pxor STATE4, STATE
3156	call _aesni_dec1
3157	pxor STATE4, STATE
3158
3159	movups STATE, (OUTP)
3160	jmp .Lxts_dec_ret
3161SYM_FUNC_END(aesni_xts_decrypt)
3162