aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/crypto
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2011-01-13 13:25:58 -0500
committerLinus Torvalds <torvalds@linux-foundation.org>2011-01-13 13:25:58 -0500
commit27d189c02ba25851973c8582e419c0bded9f7e5b (patch)
treebe142d664bc4e3cec7ab2878a243343f46e897ee /arch/x86/crypto
parenta1703154200c390ab03c10224c586e815d3e31e8 (diff)
parent55db8387a5e8d07407f0b7c6b2526417a2bc6243 (diff)
Merge git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
* git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6: (46 commits) hwrng: via_rng - Fix memory scribbling on some CPUs crypto: padlock - Move padlock.h into include/crypto hwrng: via_rng - Fix asm constraints crypto: n2 - use __devexit not __exit in n2_unregister_algs crypto: mark crypto workqueues CPU_INTENSIVE crypto: mv_cesa - dont return PTR_ERR() of wrong pointer crypto: ripemd - Set module author and update email address crypto: omap-sham - backlog handling fix crypto: gf128mul - Remove experimental tag crypto: af_alg - fix af_alg memory_allocated data type crypto: aesni-intel - Fixed build with binutils 2.16 crypto: af_alg - Make sure sk_security is initialized on accept()ed sockets net: Add missing lockdep class names for af_alg include: Install linux/if_alg.h for user-space crypto API crypto: omap-aes - checkpatch --file warning fixes crypto: omap-aes - initialize aes module once per request crypto: omap-aes - unnecessary code removed crypto: omap-aes - error handling implementation improved crypto: omap-aes - redundant locking is removed crypto: omap-aes - DMA initialization fixes for OMAP off mode ...
Diffstat (limited to 'arch/x86/crypto')
-rw-r--r--arch/x86/crypto/aesni-intel_asm.S1832
-rw-r--r--arch/x86/crypto/aesni-intel_glue.c540
2 files changed, 2335 insertions, 37 deletions
diff --git a/arch/x86/crypto/aesni-intel_asm.S b/arch/x86/crypto/aesni-intel_asm.S
index ff16756a51c..8fe2a4966b7 100644
--- a/arch/x86/crypto/aesni-intel_asm.S
+++ b/arch/x86/crypto/aesni-intel_asm.S
@@ -9,6 +9,20 @@
9 * Vinodh Gopal <vinodh.gopal@intel.com> 9 * Vinodh Gopal <vinodh.gopal@intel.com>
10 * Kahraman Akdemir 10 * Kahraman Akdemir
11 * 11 *
12 * Added RFC4106 AES-GCM support for 128-bit keys under the AEAD
13 * interface for 64-bit kernels.
14 * Authors: Erdinc Ozturk (erdinc.ozturk@intel.com)
15 * Aidan O'Mahony (aidan.o.mahony@intel.com)
16 * Adrian Hoban <adrian.hoban@intel.com>
17 * James Guilford (james.guilford@intel.com)
18 * Gabriele Paoloni <gabriele.paoloni@intel.com>
19 * Tadeusz Struk (tadeusz.struk@intel.com)
20 * Wajdi Feghali (wajdi.k.feghali@intel.com)
21 * Copyright (c) 2010, Intel Corporation.
22 *
23 * Ported x86_64 version to x86:
24 * Author: Mathias Krause <minipli@googlemail.com>
25 *
12 * This program is free software; you can redistribute it and/or modify 26 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License as published by 27 * it under the terms of the GNU General Public License as published by
14 * the Free Software Foundation; either version 2 of the License, or 28 * the Free Software Foundation; either version 2 of the License, or
@@ -18,8 +32,62 @@
18#include <linux/linkage.h> 32#include <linux/linkage.h>
19#include <asm/inst.h> 33#include <asm/inst.h>
20 34
35#ifdef __x86_64__
36.data
37POLY: .octa 0xC2000000000000000000000000000001
38TWOONE: .octa 0x00000001000000000000000000000001
39
40# order of these constants should not change.
41# more specifically, ALL_F should follow SHIFT_MASK,
42# and ZERO should follow ALL_F
43
44SHUF_MASK: .octa 0x000102030405060708090A0B0C0D0E0F
45MASK1: .octa 0x0000000000000000ffffffffffffffff
46MASK2: .octa 0xffffffffffffffff0000000000000000
47SHIFT_MASK: .octa 0x0f0e0d0c0b0a09080706050403020100
48ALL_F: .octa 0xffffffffffffffffffffffffffffffff
49ZERO: .octa 0x00000000000000000000000000000000
50ONE: .octa 0x00000000000000000000000000000001
51F_MIN_MASK: .octa 0xf1f2f3f4f5f6f7f8f9fafbfcfdfeff0
52dec: .octa 0x1
53enc: .octa 0x2
54
55
21.text 56.text
22 57
58
59#define STACK_OFFSET 8*3
60#define HashKey 16*0 // store HashKey <<1 mod poly here
61#define HashKey_2 16*1 // store HashKey^2 <<1 mod poly here
62#define HashKey_3 16*2 // store HashKey^3 <<1 mod poly here
63#define HashKey_4 16*3 // store HashKey^4 <<1 mod poly here
64#define HashKey_k 16*4 // store XOR of High 64 bits and Low 64
65 // bits of HashKey <<1 mod poly here
66 //(for Karatsuba purposes)
67#define HashKey_2_k 16*5 // store XOR of High 64 bits and Low 64
68 // bits of HashKey^2 <<1 mod poly here
69 // (for Karatsuba purposes)
70#define HashKey_3_k 16*6 // store XOR of High 64 bits and Low 64
71 // bits of HashKey^3 <<1 mod poly here
72 // (for Karatsuba purposes)
73#define HashKey_4_k 16*7 // store XOR of High 64 bits and Low 64
74 // bits of HashKey^4 <<1 mod poly here
75 // (for Karatsuba purposes)
76#define VARIABLE_OFFSET 16*8
77
78#define arg1 rdi
79#define arg2 rsi
80#define arg3 rdx
81#define arg4 rcx
82#define arg5 r8
83#define arg6 r9
84#define arg7 STACK_OFFSET+8(%r14)
85#define arg8 STACK_OFFSET+16(%r14)
86#define arg9 STACK_OFFSET+24(%r14)
87#define arg10 STACK_OFFSET+32(%r14)
88#endif
89
90
23#define STATE1 %xmm0 91#define STATE1 %xmm0
24#define STATE2 %xmm4 92#define STATE2 %xmm4
25#define STATE3 %xmm5 93#define STATE3 %xmm5
@@ -32,12 +100,16 @@
32#define IN IN1 100#define IN IN1
33#define KEY %xmm2 101#define KEY %xmm2
34#define IV %xmm3 102#define IV %xmm3
103
35#define BSWAP_MASK %xmm10 104#define BSWAP_MASK %xmm10
36#define CTR %xmm11 105#define CTR %xmm11
37#define INC %xmm12 106#define INC %xmm12
38 107
108#ifdef __x86_64__
109#define AREG %rax
39#define KEYP %rdi 110#define KEYP %rdi
40#define OUTP %rsi 111#define OUTP %rsi
112#define UKEYP OUTP
41#define INP %rdx 113#define INP %rdx
42#define LEN %rcx 114#define LEN %rcx
43#define IVP %r8 115#define IVP %r8
@@ -46,6 +118,1588 @@
46#define TKEYP T1 118#define TKEYP T1
47#define T2 %r11 119#define T2 %r11
48#define TCTR_LOW T2 120#define TCTR_LOW T2
121#else
122#define AREG %eax
123#define KEYP %edi
124#define OUTP AREG
125#define UKEYP OUTP
126#define INP %edx
127#define LEN %esi
128#define IVP %ebp
129#define KLEN %ebx
130#define T1 %ecx
131#define TKEYP T1
132#endif
133
134
135#ifdef __x86_64__
136/* GHASH_MUL MACRO to implement: Data*HashKey mod (128,127,126,121,0)
137*
138*
139* Input: A and B (128-bits each, bit-reflected)
140* Output: C = A*B*x mod poly, (i.e. >>1 )
141* To compute GH = GH*HashKey mod poly, give HK = HashKey<<1 mod poly as input
142* GH = GH * HK * x mod poly which is equivalent to GH*HashKey mod poly.
143*
144*/
145.macro GHASH_MUL GH HK TMP1 TMP2 TMP3 TMP4 TMP5
146 movdqa \GH, \TMP1
147 pshufd $78, \GH, \TMP2
148 pshufd $78, \HK, \TMP3
149 pxor \GH, \TMP2 # TMP2 = a1+a0
150 pxor \HK, \TMP3 # TMP3 = b1+b0
151 PCLMULQDQ 0x11, \HK, \TMP1 # TMP1 = a1*b1
152 PCLMULQDQ 0x00, \HK, \GH # GH = a0*b0
153 PCLMULQDQ 0x00, \TMP3, \TMP2 # TMP2 = (a0+a1)*(b1+b0)
154 pxor \GH, \TMP2
155 pxor \TMP1, \TMP2 # TMP2 = (a0*b0)+(a1*b0)
156 movdqa \TMP2, \TMP3
157 pslldq $8, \TMP3 # left shift TMP3 2 DWs
158 psrldq $8, \TMP2 # right shift TMP2 2 DWs
159 pxor \TMP3, \GH
160 pxor \TMP2, \TMP1 # TMP2:GH holds the result of GH*HK
161
162 # first phase of the reduction
163
164 movdqa \GH, \TMP2
165 movdqa \GH, \TMP3
166 movdqa \GH, \TMP4 # copy GH into TMP2,TMP3 and TMP4
167 # in in order to perform
168 # independent shifts
169 pslld $31, \TMP2 # packed right shift <<31
170 pslld $30, \TMP3 # packed right shift <<30
171 pslld $25, \TMP4 # packed right shift <<25
172 pxor \TMP3, \TMP2 # xor the shifted versions
173 pxor \TMP4, \TMP2
174 movdqa \TMP2, \TMP5
175 psrldq $4, \TMP5 # right shift TMP5 1 DW
176 pslldq $12, \TMP2 # left shift TMP2 3 DWs
177 pxor \TMP2, \GH
178
179 # second phase of the reduction
180
181 movdqa \GH,\TMP2 # copy GH into TMP2,TMP3 and TMP4
182 # in in order to perform
183 # independent shifts
184 movdqa \GH,\TMP3
185 movdqa \GH,\TMP4
186 psrld $1,\TMP2 # packed left shift >>1
187 psrld $2,\TMP3 # packed left shift >>2
188 psrld $7,\TMP4 # packed left shift >>7
189 pxor \TMP3,\TMP2 # xor the shifted versions
190 pxor \TMP4,\TMP2
191 pxor \TMP5, \TMP2
192 pxor \TMP2, \GH
193 pxor \TMP1, \GH # result is in TMP1
194.endm
195
196/*
197* if a = number of total plaintext bytes
198* b = floor(a/16)
199* num_initial_blocks = b mod 4
200* encrypt the initial num_initial_blocks blocks and apply ghash on
201* the ciphertext
202* %r10, %r11, %r12, %rax, %xmm5, %xmm6, %xmm7, %xmm8, %xmm9 registers
203* are clobbered
204* arg1, %arg2, %arg3, %r14 are used as a pointer only, not modified
205*/
206
207
208.macro INITIAL_BLOCKS_DEC num_initial_blocks TMP1 TMP2 TMP3 TMP4 TMP5 XMM0 XMM1 \
209XMM2 XMM3 XMM4 XMMDst TMP6 TMP7 i i_seq operation
210 mov arg7, %r10 # %r10 = AAD
211 mov arg8, %r12 # %r12 = aadLen
212 mov %r12, %r11
213 pxor %xmm\i, %xmm\i
214_get_AAD_loop\num_initial_blocks\operation:
215 movd (%r10), \TMP1
216 pslldq $12, \TMP1
217 psrldq $4, %xmm\i
218 pxor \TMP1, %xmm\i
219 add $4, %r10
220 sub $4, %r12
221 jne _get_AAD_loop\num_initial_blocks\operation
222 cmp $16, %r11
223 je _get_AAD_loop2_done\num_initial_blocks\operation
224 mov $16, %r12
225_get_AAD_loop2\num_initial_blocks\operation:
226 psrldq $4, %xmm\i
227 sub $4, %r12
228 cmp %r11, %r12
229 jne _get_AAD_loop2\num_initial_blocks\operation
230_get_AAD_loop2_done\num_initial_blocks\operation:
231 movdqa SHUF_MASK(%rip), %xmm14
232 PSHUFB_XMM %xmm14, %xmm\i # byte-reflect the AAD data
233
234 xor %r11, %r11 # initialise the data pointer offset as zero
235
236 # start AES for num_initial_blocks blocks
237
238 mov %arg5, %rax # %rax = *Y0
239 movdqu (%rax), \XMM0 # XMM0 = Y0
240 movdqa SHUF_MASK(%rip), %xmm14
241 PSHUFB_XMM %xmm14, \XMM0
242
243.if (\i == 5) || (\i == 6) || (\i == 7)
244.irpc index, \i_seq
245 paddd ONE(%rip), \XMM0 # INCR Y0
246 movdqa \XMM0, %xmm\index
247 movdqa SHUF_MASK(%rip), %xmm14
248 PSHUFB_XMM %xmm14, %xmm\index # perform a 16 byte swap
249
250.endr
251.irpc index, \i_seq
252 pxor 16*0(%arg1), %xmm\index
253.endr
254.irpc index, \i_seq
255 movaps 0x10(%rdi), \TMP1
256 AESENC \TMP1, %xmm\index # Round 1
257.endr
258.irpc index, \i_seq
259 movaps 0x20(%arg1), \TMP1
260 AESENC \TMP1, %xmm\index # Round 2
261.endr
262.irpc index, \i_seq
263 movaps 0x30(%arg1), \TMP1
264 AESENC \TMP1, %xmm\index # Round 2
265.endr
266.irpc index, \i_seq
267 movaps 0x40(%arg1), \TMP1
268 AESENC \TMP1, %xmm\index # Round 2
269.endr
270.irpc index, \i_seq
271 movaps 0x50(%arg1), \TMP1
272 AESENC \TMP1, %xmm\index # Round 2
273.endr
274.irpc index, \i_seq
275 movaps 0x60(%arg1), \TMP1
276 AESENC \TMP1, %xmm\index # Round 2
277.endr
278.irpc index, \i_seq
279 movaps 0x70(%arg1), \TMP1
280 AESENC \TMP1, %xmm\index # Round 2
281.endr
282.irpc index, \i_seq
283 movaps 0x80(%arg1), \TMP1
284 AESENC \TMP1, %xmm\index # Round 2
285.endr
286.irpc index, \i_seq
287 movaps 0x90(%arg1), \TMP1
288 AESENC \TMP1, %xmm\index # Round 2
289.endr
290.irpc index, \i_seq
291 movaps 0xa0(%arg1), \TMP1
292 AESENCLAST \TMP1, %xmm\index # Round 10
293.endr
294.irpc index, \i_seq
295 movdqu (%arg3 , %r11, 1), \TMP1
296 pxor \TMP1, %xmm\index
297 movdqu %xmm\index, (%arg2 , %r11, 1)
298 # write back plaintext/ciphertext for num_initial_blocks
299 add $16, %r11
300
301 movdqa \TMP1, %xmm\index
302 movdqa SHUF_MASK(%rip), %xmm14
303 PSHUFB_XMM %xmm14, %xmm\index
304
305 # prepare plaintext/ciphertext for GHASH computation
306.endr
307.endif
308 GHASH_MUL %xmm\i, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
309 # apply GHASH on num_initial_blocks blocks
310
311.if \i == 5
312 pxor %xmm5, %xmm6
313 GHASH_MUL %xmm6, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
314 pxor %xmm6, %xmm7
315 GHASH_MUL %xmm7, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
316 pxor %xmm7, %xmm8
317 GHASH_MUL %xmm8, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
318.elseif \i == 6
319 pxor %xmm6, %xmm7
320 GHASH_MUL %xmm7, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
321 pxor %xmm7, %xmm8
322 GHASH_MUL %xmm8, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
323.elseif \i == 7
324 pxor %xmm7, %xmm8
325 GHASH_MUL %xmm8, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
326.endif
327 cmp $64, %r13
328 jl _initial_blocks_done\num_initial_blocks\operation
329 # no need for precomputed values
330/*
331*
332* Precomputations for HashKey parallel with encryption of first 4 blocks.
333* Haskey_i_k holds XORed values of the low and high parts of the Haskey_i
334*/
335 paddd ONE(%rip), \XMM0 # INCR Y0
336 movdqa \XMM0, \XMM1
337 movdqa SHUF_MASK(%rip), %xmm14
338 PSHUFB_XMM %xmm14, \XMM1 # perform a 16 byte swap
339
340 paddd ONE(%rip), \XMM0 # INCR Y0
341 movdqa \XMM0, \XMM2
342 movdqa SHUF_MASK(%rip), %xmm14
343 PSHUFB_XMM %xmm14, \XMM2 # perform a 16 byte swap
344
345 paddd ONE(%rip), \XMM0 # INCR Y0
346 movdqa \XMM0, \XMM3
347 movdqa SHUF_MASK(%rip), %xmm14
348 PSHUFB_XMM %xmm14, \XMM3 # perform a 16 byte swap
349
350 paddd ONE(%rip), \XMM0 # INCR Y0
351 movdqa \XMM0, \XMM4
352 movdqa SHUF_MASK(%rip), %xmm14
353 PSHUFB_XMM %xmm14, \XMM4 # perform a 16 byte swap
354
355 pxor 16*0(%arg1), \XMM1
356 pxor 16*0(%arg1), \XMM2
357 pxor 16*0(%arg1), \XMM3
358 pxor 16*0(%arg1), \XMM4
359 movdqa \TMP3, \TMP5
360 pshufd $78, \TMP3, \TMP1
361 pxor \TMP3, \TMP1
362 movdqa \TMP1, HashKey_k(%rsp)
363 GHASH_MUL \TMP5, \TMP3, \TMP1, \TMP2, \TMP4, \TMP6, \TMP7
364# TMP5 = HashKey^2<<1 (mod poly)
365 movdqa \TMP5, HashKey_2(%rsp)
366# HashKey_2 = HashKey^2<<1 (mod poly)
367 pshufd $78, \TMP5, \TMP1
368 pxor \TMP5, \TMP1
369 movdqa \TMP1, HashKey_2_k(%rsp)
370.irpc index, 1234 # do 4 rounds
371 movaps 0x10*\index(%arg1), \TMP1
372 AESENC \TMP1, \XMM1
373 AESENC \TMP1, \XMM2
374 AESENC \TMP1, \XMM3
375 AESENC \TMP1, \XMM4
376.endr
377 GHASH_MUL \TMP5, \TMP3, \TMP1, \TMP2, \TMP4, \TMP6, \TMP7
378# TMP5 = HashKey^3<<1 (mod poly)
379 movdqa \TMP5, HashKey_3(%rsp)
380 pshufd $78, \TMP5, \TMP1
381 pxor \TMP5, \TMP1
382 movdqa \TMP1, HashKey_3_k(%rsp)
383.irpc index, 56789 # do next 5 rounds
384 movaps 0x10*\index(%arg1), \TMP1
385 AESENC \TMP1, \XMM1
386 AESENC \TMP1, \XMM2
387 AESENC \TMP1, \XMM3
388 AESENC \TMP1, \XMM4
389.endr
390 GHASH_MUL \TMP5, \TMP3, \TMP1, \TMP2, \TMP4, \TMP6, \TMP7
391# TMP5 = HashKey^3<<1 (mod poly)
392 movdqa \TMP5, HashKey_4(%rsp)
393 pshufd $78, \TMP5, \TMP1
394 pxor \TMP5, \TMP1
395 movdqa \TMP1, HashKey_4_k(%rsp)
396 movaps 0xa0(%arg1), \TMP2
397 AESENCLAST \TMP2, \XMM1
398 AESENCLAST \TMP2, \XMM2
399 AESENCLAST \TMP2, \XMM3
400 AESENCLAST \TMP2, \XMM4
401 movdqu 16*0(%arg3 , %r11 , 1), \TMP1
402 pxor \TMP1, \XMM1
403 movdqu \XMM1, 16*0(%arg2 , %r11 , 1)
404 movdqa \TMP1, \XMM1
405 movdqu 16*1(%arg3 , %r11 , 1), \TMP1
406 pxor \TMP1, \XMM2
407 movdqu \XMM2, 16*1(%arg2 , %r11 , 1)
408 movdqa \TMP1, \XMM2
409 movdqu 16*2(%arg3 , %r11 , 1), \TMP1
410 pxor \TMP1, \XMM3
411 movdqu \XMM3, 16*2(%arg2 , %r11 , 1)
412 movdqa \TMP1, \XMM3
413 movdqu 16*3(%arg3 , %r11 , 1), \TMP1
414 pxor \TMP1, \XMM4
415 movdqu \XMM4, 16*3(%arg2 , %r11 , 1)
416 movdqa \TMP1, \XMM4
417 add $64, %r11
418 movdqa SHUF_MASK(%rip), %xmm14
419 PSHUFB_XMM %xmm14, \XMM1 # perform a 16 byte swap
420 pxor \XMMDst, \XMM1
421# combine GHASHed value with the corresponding ciphertext
422 movdqa SHUF_MASK(%rip), %xmm14
423 PSHUFB_XMM %xmm14, \XMM2 # perform a 16 byte swap
424 movdqa SHUF_MASK(%rip), %xmm14
425 PSHUFB_XMM %xmm14, \XMM3 # perform a 16 byte swap
426 movdqa SHUF_MASK(%rip), %xmm14
427 PSHUFB_XMM %xmm14, \XMM4 # perform a 16 byte swap
428
429_initial_blocks_done\num_initial_blocks\operation:
430
431.endm
432
433
434/*
435* if a = number of total plaintext bytes
436* b = floor(a/16)
437* num_initial_blocks = b mod 4
438* encrypt the initial num_initial_blocks blocks and apply ghash on
439* the ciphertext
440* %r10, %r11, %r12, %rax, %xmm5, %xmm6, %xmm7, %xmm8, %xmm9 registers
441* are clobbered
442* arg1, %arg2, %arg3, %r14 are used as a pointer only, not modified
443*/
444
445
446.macro INITIAL_BLOCKS_ENC num_initial_blocks TMP1 TMP2 TMP3 TMP4 TMP5 XMM0 XMM1 \
447XMM2 XMM3 XMM4 XMMDst TMP6 TMP7 i i_seq operation
448 mov arg7, %r10 # %r10 = AAD
449 mov arg8, %r12 # %r12 = aadLen
450 mov %r12, %r11
451 pxor %xmm\i, %xmm\i
452_get_AAD_loop\num_initial_blocks\operation:
453 movd (%r10), \TMP1
454 pslldq $12, \TMP1
455 psrldq $4, %xmm\i
456 pxor \TMP1, %xmm\i
457 add $4, %r10
458 sub $4, %r12
459 jne _get_AAD_loop\num_initial_blocks\operation
460 cmp $16, %r11
461 je _get_AAD_loop2_done\num_initial_blocks\operation
462 mov $16, %r12
463_get_AAD_loop2\num_initial_blocks\operation:
464 psrldq $4, %xmm\i
465 sub $4, %r12
466 cmp %r11, %r12
467 jne _get_AAD_loop2\num_initial_blocks\operation
468_get_AAD_loop2_done\num_initial_blocks\operation:
469 movdqa SHUF_MASK(%rip), %xmm14
470 PSHUFB_XMM %xmm14, %xmm\i # byte-reflect the AAD data
471
472 xor %r11, %r11 # initialise the data pointer offset as zero
473
474 # start AES for num_initial_blocks blocks
475
476 mov %arg5, %rax # %rax = *Y0
477 movdqu (%rax), \XMM0 # XMM0 = Y0
478 movdqa SHUF_MASK(%rip), %xmm14
479 PSHUFB_XMM %xmm14, \XMM0
480
481.if (\i == 5) || (\i == 6) || (\i == 7)
482.irpc index, \i_seq
483 paddd ONE(%rip), \XMM0 # INCR Y0
484 movdqa \XMM0, %xmm\index
485 movdqa SHUF_MASK(%rip), %xmm14
486 PSHUFB_XMM %xmm14, %xmm\index # perform a 16 byte swap
487
488.endr
489.irpc index, \i_seq
490 pxor 16*0(%arg1), %xmm\index
491.endr
492.irpc index, \i_seq
493 movaps 0x10(%rdi), \TMP1
494 AESENC \TMP1, %xmm\index # Round 1
495.endr
496.irpc index, \i_seq
497 movaps 0x20(%arg1), \TMP1
498 AESENC \TMP1, %xmm\index # Round 2
499.endr
500.irpc index, \i_seq
501 movaps 0x30(%arg1), \TMP1
502 AESENC \TMP1, %xmm\index # Round 2
503.endr
504.irpc index, \i_seq
505 movaps 0x40(%arg1), \TMP1
506 AESENC \TMP1, %xmm\index # Round 2
507.endr
508.irpc index, \i_seq
509 movaps 0x50(%arg1), \TMP1
510 AESENC \TMP1, %xmm\index # Round 2
511.endr
512.irpc index, \i_seq
513 movaps 0x60(%arg1), \TMP1
514 AESENC \TMP1, %xmm\index # Round 2
515.endr
516.irpc index, \i_seq
517 movaps 0x70(%arg1), \TMP1
518 AESENC \TMP1, %xmm\index # Round 2
519.endr
520.irpc index, \i_seq
521 movaps 0x80(%arg1), \TMP1
522 AESENC \TMP1, %xmm\index # Round 2
523.endr
524.irpc index, \i_seq
525 movaps 0x90(%arg1), \TMP1
526 AESENC \TMP1, %xmm\index # Round 2
527.endr
528.irpc index, \i_seq
529 movaps 0xa0(%arg1), \TMP1
530 AESENCLAST \TMP1, %xmm\index # Round 10
531.endr
532.irpc index, \i_seq
533 movdqu (%arg3 , %r11, 1), \TMP1
534 pxor \TMP1, %xmm\index
535 movdqu %xmm\index, (%arg2 , %r11, 1)
536 # write back plaintext/ciphertext for num_initial_blocks
537 add $16, %r11
538
539 movdqa SHUF_MASK(%rip), %xmm14
540 PSHUFB_XMM %xmm14, %xmm\index
541
542 # prepare plaintext/ciphertext for GHASH computation
543.endr
544.endif
545 GHASH_MUL %xmm\i, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
546 # apply GHASH on num_initial_blocks blocks
547
548.if \i == 5
549 pxor %xmm5, %xmm6
550 GHASH_MUL %xmm6, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
551 pxor %xmm6, %xmm7
552 GHASH_MUL %xmm7, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
553 pxor %xmm7, %xmm8
554 GHASH_MUL %xmm8, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
555.elseif \i == 6
556 pxor %xmm6, %xmm7
557 GHASH_MUL %xmm7, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
558 pxor %xmm7, %xmm8
559 GHASH_MUL %xmm8, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
560.elseif \i == 7
561 pxor %xmm7, %xmm8
562 GHASH_MUL %xmm8, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
563.endif
564 cmp $64, %r13
565 jl _initial_blocks_done\num_initial_blocks\operation
566 # no need for precomputed values
567/*
568*
569* Precomputations for HashKey parallel with encryption of first 4 blocks.
570* Haskey_i_k holds XORed values of the low and high parts of the Haskey_i
571*/
572 paddd ONE(%rip), \XMM0 # INCR Y0
573 movdqa \XMM0, \XMM1
574 movdqa SHUF_MASK(%rip), %xmm14
575 PSHUFB_XMM %xmm14, \XMM1 # perform a 16 byte swap
576
577 paddd ONE(%rip), \XMM0 # INCR Y0
578 movdqa \XMM0, \XMM2
579 movdqa SHUF_MASK(%rip), %xmm14
580 PSHUFB_XMM %xmm14, \XMM2 # perform a 16 byte swap
581
582 paddd ONE(%rip), \XMM0 # INCR Y0
583 movdqa \XMM0, \XMM3
584 movdqa SHUF_MASK(%rip), %xmm14
585 PSHUFB_XMM %xmm14, \XMM3 # perform a 16 byte swap
586
587 paddd ONE(%rip), \XMM0 # INCR Y0
588 movdqa \XMM0, \XMM4
589 movdqa SHUF_MASK(%rip), %xmm14
590 PSHUFB_XMM %xmm14, \XMM4 # perform a 16 byte swap
591
592 pxor 16*0(%arg1), \XMM1
593 pxor 16*0(%arg1), \XMM2
594 pxor 16*0(%arg1), \XMM3
595 pxor 16*0(%arg1), \XMM4
596 movdqa \TMP3, \TMP5
597 pshufd $78, \TMP3, \TMP1
598 pxor \TMP3, \TMP1
599 movdqa \TMP1, HashKey_k(%rsp)
600 GHASH_MUL \TMP5, \TMP3, \TMP1, \TMP2, \TMP4, \TMP6, \TMP7
601# TMP5 = HashKey^2<<1 (mod poly)
602 movdqa \TMP5, HashKey_2(%rsp)
603# HashKey_2 = HashKey^2<<1 (mod poly)
604 pshufd $78, \TMP5, \TMP1
605 pxor \TMP5, \TMP1
606 movdqa \TMP1, HashKey_2_k(%rsp)
607.irpc index, 1234 # do 4 rounds
608 movaps 0x10*\index(%arg1), \TMP1
609 AESENC \TMP1, \XMM1
610 AESENC \TMP1, \XMM2
611 AESENC \TMP1, \XMM3
612 AESENC \TMP1, \XMM4
613.endr
614 GHASH_MUL \TMP5, \TMP3, \TMP1, \TMP2, \TMP4, \TMP6, \TMP7
615# TMP5 = HashKey^3<<1 (mod poly)
616 movdqa \TMP5, HashKey_3(%rsp)
617 pshufd $78, \TMP5, \TMP1
618 pxor \TMP5, \TMP1
619 movdqa \TMP1, HashKey_3_k(%rsp)
620.irpc index, 56789 # do next 5 rounds
621 movaps 0x10*\index(%arg1), \TMP1
622 AESENC \TMP1, \XMM1
623 AESENC \TMP1, \XMM2
624 AESENC \TMP1, \XMM3
625 AESENC \TMP1, \XMM4
626.endr
627 GHASH_MUL \TMP5, \TMP3, \TMP1, \TMP2, \TMP4, \TMP6, \TMP7
628# TMP5 = HashKey^3<<1 (mod poly)
629 movdqa \TMP5, HashKey_4(%rsp)
630 pshufd $78, \TMP5, \TMP1
631 pxor \TMP5, \TMP1
632 movdqa \TMP1, HashKey_4_k(%rsp)
633 movaps 0xa0(%arg1), \TMP2
634 AESENCLAST \TMP2, \XMM1
635 AESENCLAST \TMP2, \XMM2
636 AESENCLAST \TMP2, \XMM3
637 AESENCLAST \TMP2, \XMM4
638 movdqu 16*0(%arg3 , %r11 , 1), \TMP1
639 pxor \TMP1, \XMM1
640 movdqu 16*1(%arg3 , %r11 , 1), \TMP1
641 pxor \TMP1, \XMM2
642 movdqu 16*2(%arg3 , %r11 , 1), \TMP1
643 pxor \TMP1, \XMM3
644 movdqu 16*3(%arg3 , %r11 , 1), \TMP1
645 pxor \TMP1, \XMM4
646 movdqu \XMM1, 16*0(%arg2 , %r11 , 1)
647 movdqu \XMM2, 16*1(%arg2 , %r11 , 1)
648 movdqu \XMM3, 16*2(%arg2 , %r11 , 1)
649 movdqu \XMM4, 16*3(%arg2 , %r11 , 1)
650
651 add $64, %r11
652 movdqa SHUF_MASK(%rip), %xmm14
653 PSHUFB_XMM %xmm14, \XMM1 # perform a 16 byte swap
654 pxor \XMMDst, \XMM1
655# combine GHASHed value with the corresponding ciphertext
656 movdqa SHUF_MASK(%rip), %xmm14
657 PSHUFB_XMM %xmm14, \XMM2 # perform a 16 byte swap
658 movdqa SHUF_MASK(%rip), %xmm14
659 PSHUFB_XMM %xmm14, \XMM3 # perform a 16 byte swap
660 movdqa SHUF_MASK(%rip), %xmm14
661 PSHUFB_XMM %xmm14, \XMM4 # perform a 16 byte swap
662
663_initial_blocks_done\num_initial_blocks\operation:
664
665.endm
666
667/*
668* encrypt 4 blocks at a time
669* ghash the 4 previously encrypted ciphertext blocks
670* arg1, %arg2, %arg3 are used as pointers only, not modified
671* %r11 is the data offset value
672*/
673.macro GHASH_4_ENCRYPT_4_PARALLEL_ENC TMP1 TMP2 TMP3 TMP4 TMP5 \
674TMP6 XMM0 XMM1 XMM2 XMM3 XMM4 XMM5 XMM6 XMM7 XMM8 operation
675
676 movdqa \XMM1, \XMM5
677 movdqa \XMM2, \XMM6
678 movdqa \XMM3, \XMM7
679 movdqa \XMM4, \XMM8
680
681 movdqa SHUF_MASK(%rip), %xmm15
682 # multiply TMP5 * HashKey using karatsuba
683
684 movdqa \XMM5, \TMP4
685 pshufd $78, \XMM5, \TMP6
686 pxor \XMM5, \TMP6
687 paddd ONE(%rip), \XMM0 # INCR CNT
688 movdqa HashKey_4(%rsp), \TMP5
689 PCLMULQDQ 0x11, \TMP5, \TMP4 # TMP4 = a1*b1
690 movdqa \XMM0, \XMM1
691 paddd ONE(%rip), \XMM0 # INCR CNT
692 movdqa \XMM0, \XMM2
693 paddd ONE(%rip), \XMM0 # INCR CNT
694 movdqa \XMM0, \XMM3
695 paddd ONE(%rip), \XMM0 # INCR CNT
696 movdqa \XMM0, \XMM4
697 PSHUFB_XMM %xmm15, \XMM1 # perform a 16 byte swap
698 PCLMULQDQ 0x00, \TMP5, \XMM5 # XMM5 = a0*b0
699 PSHUFB_XMM %xmm15, \XMM2 # perform a 16 byte swap
700 PSHUFB_XMM %xmm15, \XMM3 # perform a 16 byte swap
701 PSHUFB_XMM %xmm15, \XMM4 # perform a 16 byte swap
702
703 pxor (%arg1), \XMM1
704 pxor (%arg1), \XMM2
705 pxor (%arg1), \XMM3
706 pxor (%arg1), \XMM4
707 movdqa HashKey_4_k(%rsp), \TMP5
708 PCLMULQDQ 0x00, \TMP5, \TMP6 # TMP6 = (a1+a0)*(b1+b0)
709 movaps 0x10(%arg1), \TMP1
710 AESENC \TMP1, \XMM1 # Round 1
711 AESENC \TMP1, \XMM2
712 AESENC \TMP1, \XMM3
713 AESENC \TMP1, \XMM4
714 movaps 0x20(%arg1), \TMP1
715 AESENC \TMP1, \XMM1 # Round 2
716 AESENC \TMP1, \XMM2
717 AESENC \TMP1, \XMM3
718 AESENC \TMP1, \XMM4
719 movdqa \XMM6, \TMP1
720 pshufd $78, \XMM6, \TMP2
721 pxor \XMM6, \TMP2
722 movdqa HashKey_3(%rsp), \TMP5
723 PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1 * b1
724 movaps 0x30(%arg1), \TMP3
725 AESENC \TMP3, \XMM1 # Round 3
726 AESENC \TMP3, \XMM2
727 AESENC \TMP3, \XMM3
728 AESENC \TMP3, \XMM4
729 PCLMULQDQ 0x00, \TMP5, \XMM6 # XMM6 = a0*b0
730 movaps 0x40(%arg1), \TMP3
731 AESENC \TMP3, \XMM1 # Round 4
732 AESENC \TMP3, \XMM2
733 AESENC \TMP3, \XMM3
734 AESENC \TMP3, \XMM4
735 movdqa HashKey_3_k(%rsp), \TMP5
736 PCLMULQDQ 0x00, \TMP5, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
737 movaps 0x50(%arg1), \TMP3
738 AESENC \TMP3, \XMM1 # Round 5
739 AESENC \TMP3, \XMM2
740 AESENC \TMP3, \XMM3
741 AESENC \TMP3, \XMM4
742 pxor \TMP1, \TMP4
743# accumulate the results in TMP4:XMM5, TMP6 holds the middle part
744 pxor \XMM6, \XMM5
745 pxor \TMP2, \TMP6
746 movdqa \XMM7, \TMP1
747 pshufd $78, \XMM7, \TMP2
748 pxor \XMM7, \TMP2
749 movdqa HashKey_2(%rsp ), \TMP5
750
751 # Multiply TMP5 * HashKey using karatsuba
752
753 PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1*b1
754 movaps 0x60(%arg1), \TMP3
755 AESENC \TMP3, \XMM1 # Round 6
756 AESENC \TMP3, \XMM2
757 AESENC \TMP3, \XMM3
758 AESENC \TMP3, \XMM4
759 PCLMULQDQ 0x00, \TMP5, \XMM7 # XMM7 = a0*b0
760 movaps 0x70(%arg1), \TMP3
761 AESENC \TMP3, \XMM1 # Round 7
762 AESENC \TMP3, \XMM2
763 AESENC \TMP3, \XMM3
764 AESENC \TMP3, \XMM4
765 movdqa HashKey_2_k(%rsp), \TMP5
766 PCLMULQDQ 0x00, \TMP5, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
767 movaps 0x80(%arg1), \TMP3
768 AESENC \TMP3, \XMM1 # Round 8
769 AESENC \TMP3, \XMM2
770 AESENC \TMP3, \XMM3
771 AESENC \TMP3, \XMM4
772 pxor \TMP1, \TMP4
773# accumulate the results in TMP4:XMM5, TMP6 holds the middle part
774 pxor \XMM7, \XMM5
775 pxor \TMP2, \TMP6
776
777 # Multiply XMM8 * HashKey
778 # XMM8 and TMP5 hold the values for the two operands
779
780 movdqa \XMM8, \TMP1
781 pshufd $78, \XMM8, \TMP2
782 pxor \XMM8, \TMP2
783 movdqa HashKey(%rsp), \TMP5
784 PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1*b1
785 movaps 0x90(%arg1), \TMP3
786 AESENC \TMP3, \XMM1 # Round 9
787 AESENC \TMP3, \XMM2
788 AESENC \TMP3, \XMM3
789 AESENC \TMP3, \XMM4
790 PCLMULQDQ 0x00, \TMP5, \XMM8 # XMM8 = a0*b0
791 movaps 0xa0(%arg1), \TMP3
792 AESENCLAST \TMP3, \XMM1 # Round 10
793 AESENCLAST \TMP3, \XMM2
794 AESENCLAST \TMP3, \XMM3
795 AESENCLAST \TMP3, \XMM4
796 movdqa HashKey_k(%rsp), \TMP5
797 PCLMULQDQ 0x00, \TMP5, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
798 movdqu (%arg3,%r11,1), \TMP3
799 pxor \TMP3, \XMM1 # Ciphertext/Plaintext XOR EK
800 movdqu 16(%arg3,%r11,1), \TMP3
801 pxor \TMP3, \XMM2 # Ciphertext/Plaintext XOR EK
802 movdqu 32(%arg3,%r11,1), \TMP3
803 pxor \TMP3, \XMM3 # Ciphertext/Plaintext XOR EK
804 movdqu 48(%arg3,%r11,1), \TMP3
805 pxor \TMP3, \XMM4 # Ciphertext/Plaintext XOR EK
806 movdqu \XMM1, (%arg2,%r11,1) # Write to the ciphertext buffer
807 movdqu \XMM2, 16(%arg2,%r11,1) # Write to the ciphertext buffer
808 movdqu \XMM3, 32(%arg2,%r11,1) # Write to the ciphertext buffer
809 movdqu \XMM4, 48(%arg2,%r11,1) # Write to the ciphertext buffer
810 PSHUFB_XMM %xmm15, \XMM1 # perform a 16 byte swap
811 PSHUFB_XMM %xmm15, \XMM2 # perform a 16 byte swap
812 PSHUFB_XMM %xmm15, \XMM3 # perform a 16 byte swap
813 PSHUFB_XMM %xmm15, \XMM4 # perform a 16 byte swap
814
815 pxor \TMP4, \TMP1
816 pxor \XMM8, \XMM5
817 pxor \TMP6, \TMP2
818 pxor \TMP1, \TMP2
819 pxor \XMM5, \TMP2
820 movdqa \TMP2, \TMP3
821 pslldq $8, \TMP3 # left shift TMP3 2 DWs
822 psrldq $8, \TMP2 # right shift TMP2 2 DWs
823 pxor \TMP3, \XMM5
824 pxor \TMP2, \TMP1 # accumulate the results in TMP1:XMM5
825
826 # first phase of reduction
827
828 movdqa \XMM5, \TMP2
829 movdqa \XMM5, \TMP3
830 movdqa \XMM5, \TMP4
831# move XMM5 into TMP2, TMP3, TMP4 in order to perform shifts independently
832 pslld $31, \TMP2 # packed right shift << 31
833 pslld $30, \TMP3 # packed right shift << 30
834 pslld $25, \TMP4 # packed right shift << 25
835 pxor \TMP3, \TMP2 # xor the shifted versions
836 pxor \TMP4, \TMP2
837 movdqa \TMP2, \TMP5
838 psrldq $4, \TMP5 # right shift T5 1 DW
839 pslldq $12, \TMP2 # left shift T2 3 DWs
840 pxor \TMP2, \XMM5
841
842 # second phase of reduction
843
844 movdqa \XMM5,\TMP2 # make 3 copies of XMM5 into TMP2, TMP3, TMP4
845 movdqa \XMM5,\TMP3
846 movdqa \XMM5,\TMP4
847 psrld $1, \TMP2 # packed left shift >>1
848 psrld $2, \TMP3 # packed left shift >>2
849 psrld $7, \TMP4 # packed left shift >>7
850 pxor \TMP3,\TMP2 # xor the shifted versions
851 pxor \TMP4,\TMP2
852 pxor \TMP5, \TMP2
853 pxor \TMP2, \XMM5
854 pxor \TMP1, \XMM5 # result is in TMP1
855
856 pxor \XMM5, \XMM1
857.endm
858
859/*
860* decrypt 4 blocks at a time
861* ghash the 4 previously decrypted ciphertext blocks
862* arg1, %arg2, %arg3 are used as pointers only, not modified
863* %r11 is the data offset value
864*/
865.macro GHASH_4_ENCRYPT_4_PARALLEL_DEC TMP1 TMP2 TMP3 TMP4 TMP5 \
866TMP6 XMM0 XMM1 XMM2 XMM3 XMM4 XMM5 XMM6 XMM7 XMM8 operation
867
868 movdqa \XMM1, \XMM5
869 movdqa \XMM2, \XMM6
870 movdqa \XMM3, \XMM7
871 movdqa \XMM4, \XMM8
872
873 movdqa SHUF_MASK(%rip), %xmm15
874 # multiply TMP5 * HashKey using karatsuba
875
876 movdqa \XMM5, \TMP4
877 pshufd $78, \XMM5, \TMP6
878 pxor \XMM5, \TMP6
879 paddd ONE(%rip), \XMM0 # INCR CNT
880 movdqa HashKey_4(%rsp), \TMP5
881 PCLMULQDQ 0x11, \TMP5, \TMP4 # TMP4 = a1*b1
882 movdqa \XMM0, \XMM1
883 paddd ONE(%rip), \XMM0 # INCR CNT
884 movdqa \XMM0, \XMM2
885 paddd ONE(%rip), \XMM0 # INCR CNT
886 movdqa \XMM0, \XMM3
887 paddd ONE(%rip), \XMM0 # INCR CNT
888 movdqa \XMM0, \XMM4
889 PSHUFB_XMM %xmm15, \XMM1 # perform a 16 byte swap
890 PCLMULQDQ 0x00, \TMP5, \XMM5 # XMM5 = a0*b0
891 PSHUFB_XMM %xmm15, \XMM2 # perform a 16 byte swap
892 PSHUFB_XMM %xmm15, \XMM3 # perform a 16 byte swap
893 PSHUFB_XMM %xmm15, \XMM4 # perform a 16 byte swap
894
895 pxor (%arg1), \XMM1
896 pxor (%arg1), \XMM2
897 pxor (%arg1), \XMM3
898 pxor (%arg1), \XMM4
899 movdqa HashKey_4_k(%rsp), \TMP5
900 PCLMULQDQ 0x00, \TMP5, \TMP6 # TMP6 = (a1+a0)*(b1+b0)
901 movaps 0x10(%arg1), \TMP1
902 AESENC \TMP1, \XMM1 # Round 1
903 AESENC \TMP1, \XMM2
904 AESENC \TMP1, \XMM3
905 AESENC \TMP1, \XMM4
906 movaps 0x20(%arg1), \TMP1
907 AESENC \TMP1, \XMM1 # Round 2
908 AESENC \TMP1, \XMM2
909 AESENC \TMP1, \XMM3
910 AESENC \TMP1, \XMM4
911 movdqa \XMM6, \TMP1
912 pshufd $78, \XMM6, \TMP2
913 pxor \XMM6, \TMP2
914 movdqa HashKey_3(%rsp), \TMP5
915 PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1 * b1
916 movaps 0x30(%arg1), \TMP3
917 AESENC \TMP3, \XMM1 # Round 3
918 AESENC \TMP3, \XMM2
919 AESENC \TMP3, \XMM3
920 AESENC \TMP3, \XMM4
921 PCLMULQDQ 0x00, \TMP5, \XMM6 # XMM6 = a0*b0
922 movaps 0x40(%arg1), \TMP3
923 AESENC \TMP3, \XMM1 # Round 4
924 AESENC \TMP3, \XMM2
925 AESENC \TMP3, \XMM3
926 AESENC \TMP3, \XMM4
927 movdqa HashKey_3_k(%rsp), \TMP5
928 PCLMULQDQ 0x00, \TMP5, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
929 movaps 0x50(%arg1), \TMP3
930 AESENC \TMP3, \XMM1 # Round 5
931 AESENC \TMP3, \XMM2
932 AESENC \TMP3, \XMM3
933 AESENC \TMP3, \XMM4
934 pxor \TMP1, \TMP4
935# accumulate the results in TMP4:XMM5, TMP6 holds the middle part
936 pxor \XMM6, \XMM5
937 pxor \TMP2, \TMP6
938 movdqa \XMM7, \TMP1
939 pshufd $78, \XMM7, \TMP2
940 pxor \XMM7, \TMP2
941 movdqa HashKey_2(%rsp ), \TMP5
942
943 # Multiply TMP5 * HashKey using karatsuba
944
945 PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1*b1
946 movaps 0x60(%arg1), \TMP3
947 AESENC \TMP3, \XMM1 # Round 6
948 AESENC \TMP3, \XMM2
949 AESENC \TMP3, \XMM3
950 AESENC \TMP3, \XMM4
951 PCLMULQDQ 0x00, \TMP5, \XMM7 # XMM7 = a0*b0
952 movaps 0x70(%arg1), \TMP3
953 AESENC \TMP3, \XMM1 # Round 7
954 AESENC \TMP3, \XMM2
955 AESENC \TMP3, \XMM3
956 AESENC \TMP3, \XMM4
957 movdqa HashKey_2_k(%rsp), \TMP5
958 PCLMULQDQ 0x00, \TMP5, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
959 movaps 0x80(%arg1), \TMP3
960 AESENC \TMP3, \XMM1 # Round 8
961 AESENC \TMP3, \XMM2
962 AESENC \TMP3, \XMM3
963 AESENC \TMP3, \XMM4
964 pxor \TMP1, \TMP4
965# accumulate the results in TMP4:XMM5, TMP6 holds the middle part
966 pxor \XMM7, \XMM5
967 pxor \TMP2, \TMP6
968
969 # Multiply XMM8 * HashKey
970 # XMM8 and TMP5 hold the values for the two operands
971
972 movdqa \XMM8, \TMP1
973 pshufd $78, \XMM8, \TMP2
974 pxor \XMM8, \TMP2
975 movdqa HashKey(%rsp), \TMP5
976 PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1*b1
977 movaps 0x90(%arg1), \TMP3
978 AESENC \TMP3, \XMM1 # Round 9
979 AESENC \TMP3, \XMM2
980 AESENC \TMP3, \XMM3
981 AESENC \TMP3, \XMM4
982 PCLMULQDQ 0x00, \TMP5, \XMM8 # XMM8 = a0*b0
983 movaps 0xa0(%arg1), \TMP3
984 AESENCLAST \TMP3, \XMM1 # Round 10
985 AESENCLAST \TMP3, \XMM2
986 AESENCLAST \TMP3, \XMM3
987 AESENCLAST \TMP3, \XMM4
988 movdqa HashKey_k(%rsp), \TMP5
989 PCLMULQDQ 0x00, \TMP5, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
990 movdqu (%arg3,%r11,1), \TMP3
991 pxor \TMP3, \XMM1 # Ciphertext/Plaintext XOR EK
992 movdqu \XMM1, (%arg2,%r11,1) # Write to plaintext buffer
993 movdqa \TMP3, \XMM1
994 movdqu 16(%arg3,%r11,1), \TMP3
995 pxor \TMP3, \XMM2 # Ciphertext/Plaintext XOR EK
996 movdqu \XMM2, 16(%arg2,%r11,1) # Write to plaintext buffer
997 movdqa \TMP3, \XMM2
998 movdqu 32(%arg3,%r11,1), \TMP3
999 pxor \TMP3, \XMM3 # Ciphertext/Plaintext XOR EK
1000 movdqu \XMM3, 32(%arg2,%r11,1) # Write to plaintext buffer
1001 movdqa \TMP3, \XMM3
1002 movdqu 48(%arg3,%r11,1), \TMP3
1003 pxor \TMP3, \XMM4 # Ciphertext/Plaintext XOR EK
1004 movdqu \XMM4, 48(%arg2,%r11,1) # Write to plaintext buffer
1005 movdqa \TMP3, \XMM4
1006 PSHUFB_XMM %xmm15, \XMM1 # perform a 16 byte swap
1007 PSHUFB_XMM %xmm15, \XMM2 # perform a 16 byte swap
1008 PSHUFB_XMM %xmm15, \XMM3 # perform a 16 byte swap
1009 PSHUFB_XMM %xmm15, \XMM4 # perform a 16 byte swap
1010
1011 pxor \TMP4, \TMP1
1012 pxor \XMM8, \XMM5
1013 pxor \TMP6, \TMP2
1014 pxor \TMP1, \TMP2
1015 pxor \XMM5, \TMP2
1016 movdqa \TMP2, \TMP3
1017 pslldq $8, \TMP3 # left shift TMP3 2 DWs
1018 psrldq $8, \TMP2 # right shift TMP2 2 DWs
1019 pxor \TMP3, \XMM5
1020 pxor \TMP2, \TMP1 # accumulate the results in TMP1:XMM5
1021
1022 # first phase of reduction
1023
1024 movdqa \XMM5, \TMP2
1025 movdqa \XMM5, \TMP3
1026 movdqa \XMM5, \TMP4
1027# move XMM5 into TMP2, TMP3, TMP4 in order to perform shifts independently
1028 pslld $31, \TMP2 # packed right shift << 31
1029 pslld $30, \TMP3 # packed right shift << 30
1030 pslld $25, \TMP4 # packed right shift << 25
1031 pxor \TMP3, \TMP2 # xor the shifted versions
1032 pxor \TMP4, \TMP2
1033 movdqa \TMP2, \TMP5
1034 psrldq $4, \TMP5 # right shift T5 1 DW
1035 pslldq $12, \TMP2 # left shift T2 3 DWs
1036 pxor \TMP2, \XMM5
1037
1038 # second phase of reduction
1039
1040 movdqa \XMM5,\TMP2 # make 3 copies of XMM5 into TMP2, TMP3, TMP4
1041 movdqa \XMM5,\TMP3
1042 movdqa \XMM5,\TMP4
1043 psrld $1, \TMP2 # packed left shift >>1
1044 psrld $2, \TMP3 # packed left shift >>2
1045 psrld $7, \TMP4 # packed left shift >>7
1046 pxor \TMP3,\TMP2 # xor the shifted versions
1047 pxor \TMP4,\TMP2
1048 pxor \TMP5, \TMP2
1049 pxor \TMP2, \XMM5
1050 pxor \TMP1, \XMM5 # result is in TMP1
1051
1052 pxor \XMM5, \XMM1
1053.endm
1054
1055/* GHASH the last 4 ciphertext blocks. */
1056.macro GHASH_LAST_4 TMP1 TMP2 TMP3 TMP4 TMP5 TMP6 \
1057TMP7 XMM1 XMM2 XMM3 XMM4 XMMDst
1058
1059 # Multiply TMP6 * HashKey (using Karatsuba)
1060
1061 movdqa \XMM1, \TMP6
1062 pshufd $78, \XMM1, \TMP2
1063 pxor \XMM1, \TMP2
1064 movdqa HashKey_4(%rsp), \TMP5
1065 PCLMULQDQ 0x11, \TMP5, \TMP6 # TMP6 = a1*b1
1066 PCLMULQDQ 0x00, \TMP5, \XMM1 # XMM1 = a0*b0
1067 movdqa HashKey_4_k(%rsp), \TMP4
1068 PCLMULQDQ 0x00, \TMP4, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
1069 movdqa \XMM1, \XMMDst
1070 movdqa \TMP2, \XMM1 # result in TMP6, XMMDst, XMM1
1071
1072 # Multiply TMP1 * HashKey (using Karatsuba)
1073
1074 movdqa \XMM2, \TMP1
1075 pshufd $78, \XMM2, \TMP2
1076 pxor \XMM2, \TMP2
1077 movdqa HashKey_3(%rsp), \TMP5
1078 PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1*b1
1079 PCLMULQDQ 0x00, \TMP5, \XMM2 # XMM2 = a0*b0
1080 movdqa HashKey_3_k(%rsp), \TMP4
1081 PCLMULQDQ 0x00, \TMP4, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
1082 pxor \TMP1, \TMP6
1083 pxor \XMM2, \XMMDst
1084 pxor \TMP2, \XMM1
1085# results accumulated in TMP6, XMMDst, XMM1
1086
1087 # Multiply TMP1 * HashKey (using Karatsuba)
1088
1089 movdqa \XMM3, \TMP1
1090 pshufd $78, \XMM3, \TMP2
1091 pxor \XMM3, \TMP2
1092 movdqa HashKey_2(%rsp), \TMP5
1093 PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1*b1
1094 PCLMULQDQ 0x00, \TMP5, \XMM3 # XMM3 = a0*b0
1095 movdqa HashKey_2_k(%rsp), \TMP4
1096 PCLMULQDQ 0x00, \TMP4, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
1097 pxor \TMP1, \TMP6
1098 pxor \XMM3, \XMMDst
1099 pxor \TMP2, \XMM1 # results accumulated in TMP6, XMMDst, XMM1
1100
1101 # Multiply TMP1 * HashKey (using Karatsuba)
1102 movdqa \XMM4, \TMP1
1103 pshufd $78, \XMM4, \TMP2
1104 pxor \XMM4, \TMP2
1105 movdqa HashKey(%rsp), \TMP5
1106 PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1*b1
1107 PCLMULQDQ 0x00, \TMP5, \XMM4 # XMM4 = a0*b0
1108 movdqa HashKey_k(%rsp), \TMP4
1109 PCLMULQDQ 0x00, \TMP4, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
1110 pxor \TMP1, \TMP6
1111 pxor \XMM4, \XMMDst
1112 pxor \XMM1, \TMP2
1113 pxor \TMP6, \TMP2
1114 pxor \XMMDst, \TMP2
1115 # middle section of the temp results combined as in karatsuba algorithm
1116 movdqa \TMP2, \TMP4
1117 pslldq $8, \TMP4 # left shift TMP4 2 DWs
1118 psrldq $8, \TMP2 # right shift TMP2 2 DWs
1119 pxor \TMP4, \XMMDst
1120 pxor \TMP2, \TMP6
1121# TMP6:XMMDst holds the result of the accumulated carry-less multiplications
1122 # first phase of the reduction
1123 movdqa \XMMDst, \TMP2
1124 movdqa \XMMDst, \TMP3
1125 movdqa \XMMDst, \TMP4
1126# move XMMDst into TMP2, TMP3, TMP4 in order to perform 3 shifts independently
1127 pslld $31, \TMP2 # packed right shifting << 31
1128 pslld $30, \TMP3 # packed right shifting << 30
1129 pslld $25, \TMP4 # packed right shifting << 25
1130 pxor \TMP3, \TMP2 # xor the shifted versions
1131 pxor \TMP4, \TMP2
1132 movdqa \TMP2, \TMP7
1133 psrldq $4, \TMP7 # right shift TMP7 1 DW
1134 pslldq $12, \TMP2 # left shift TMP2 3 DWs
1135 pxor \TMP2, \XMMDst
1136
1137 # second phase of the reduction
1138 movdqa \XMMDst, \TMP2
1139 # make 3 copies of XMMDst for doing 3 shift operations
1140 movdqa \XMMDst, \TMP3
1141 movdqa \XMMDst, \TMP4
1142 psrld $1, \TMP2 # packed left shift >> 1
1143 psrld $2, \TMP3 # packed left shift >> 2
1144 psrld $7, \TMP4 # packed left shift >> 7
1145 pxor \TMP3, \TMP2 # xor the shifted versions
1146 pxor \TMP4, \TMP2
1147 pxor \TMP7, \TMP2
1148 pxor \TMP2, \XMMDst
1149 pxor \TMP6, \XMMDst # reduced result is in XMMDst
1150.endm
1151
1152/* Encryption of a single block done*/
1153.macro ENCRYPT_SINGLE_BLOCK XMM0 TMP1
1154
1155 pxor (%arg1), \XMM0
1156 movaps 16(%arg1), \TMP1
1157 AESENC \TMP1, \XMM0
1158 movaps 32(%arg1), \TMP1
1159 AESENC \TMP1, \XMM0
1160 movaps 48(%arg1), \TMP1
1161 AESENC \TMP1, \XMM0
1162 movaps 64(%arg1), \TMP1
1163 AESENC \TMP1, \XMM0
1164 movaps 80(%arg1), \TMP1
1165 AESENC \TMP1, \XMM0
1166 movaps 96(%arg1), \TMP1
1167 AESENC \TMP1, \XMM0
1168 movaps 112(%arg1), \TMP1
1169 AESENC \TMP1, \XMM0
1170 movaps 128(%arg1), \TMP1
1171 AESENC \TMP1, \XMM0
1172 movaps 144(%arg1), \TMP1
1173 AESENC \TMP1, \XMM0
1174 movaps 160(%arg1), \TMP1
1175 AESENCLAST \TMP1, \XMM0
1176.endm
1177
1178
1179/*****************************************************************************
1180* void aesni_gcm_dec(void *aes_ctx, // AES Key schedule. Starts on a 16 byte boundary.
1181* u8 *out, // Plaintext output. Encrypt in-place is allowed.
1182* const u8 *in, // Ciphertext input
1183* u64 plaintext_len, // Length of data in bytes for decryption.
1184* u8 *iv, // Pre-counter block j0: 4 byte salt (from Security Association)
1185* // concatenated with 8 byte Initialisation Vector (from IPSec ESP Payload)
1186* // concatenated with 0x00000001. 16-byte aligned pointer.
1187* u8 *hash_subkey, // H, the Hash sub key input. Data starts on a 16-byte boundary.
1188* const u8 *aad, // Additional Authentication Data (AAD)
1189* u64 aad_len, // Length of AAD in bytes. With RFC4106 this is going to be 8 or 12 bytes
1190* u8 *auth_tag, // Authenticated Tag output. The driver will compare this to the
1191* // given authentication tag and only return the plaintext if they match.
1192* u64 auth_tag_len); // Authenticated Tag Length in bytes. Valid values are 16
1193* // (most likely), 12 or 8.
1194*
1195* Assumptions:
1196*
1197* keys:
1198* keys are pre-expanded and aligned to 16 bytes. we are using the first
1199* set of 11 keys in the data structure void *aes_ctx
1200*
1201* iv:
1202* 0 1 2 3
1203* 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
1204* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1205* | Salt (From the SA) |
1206* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1207* | Initialization Vector |
1208* | (This is the sequence number from IPSec header) |
1209* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1210* | 0x1 |
1211* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1212*
1213*
1214*
1215* AAD:
1216* AAD padded to 128 bits with 0
1217* for example, assume AAD is a u32 vector
1218*
1219* if AAD is 8 bytes:
1220* AAD[3] = {A0, A1};
1221* padded AAD in xmm register = {A1 A0 0 0}
1222*
1223* 0 1 2 3
1224* 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
1225* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1226* | SPI (A1) |
1227* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1228* | 32-bit Sequence Number (A0) |
1229* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1230* | 0x0 |
1231* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1232*
1233* AAD Format with 32-bit Sequence Number
1234*
1235* if AAD is 12 bytes:
1236* AAD[3] = {A0, A1, A2};
1237* padded AAD in xmm register = {A2 A1 A0 0}
1238*
1239* 0 1 2 3
1240* 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
1241* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1242* 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
1243* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1244* | SPI (A2) |
1245* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1246* | 64-bit Extended Sequence Number {A1,A0} |
1247* | |
1248* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1249* | 0x0 |
1250* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1251*
1252* AAD Format with 64-bit Extended Sequence Number
1253*
1254* aadLen:
1255* from the definition of the spec, aadLen can only be 8 or 12 bytes.
1256* The code supports 16 too but for other sizes, the code will fail.
1257*
1258* TLen:
1259* from the definition of the spec, TLen can only be 8, 12 or 16 bytes.
1260* For other sizes, the code will fail.
1261*
1262* poly = x^128 + x^127 + x^126 + x^121 + 1
1263*
1264*****************************************************************************/
1265
1266ENTRY(aesni_gcm_dec)
1267 push %r12
1268 push %r13
1269 push %r14
1270 mov %rsp, %r14
1271/*
1272* states of %xmm registers %xmm6:%xmm15 not saved
1273* all %xmm registers are clobbered
1274*/
1275 sub $VARIABLE_OFFSET, %rsp
1276 and $~63, %rsp # align rsp to 64 bytes
1277 mov %arg6, %r12
1278 movdqu (%r12), %xmm13 # %xmm13 = HashKey
1279 movdqa SHUF_MASK(%rip), %xmm2
1280 PSHUFB_XMM %xmm2, %xmm13
1281
1282
1283# Precompute HashKey<<1 (mod poly) from the hash key (required for GHASH)
1284
1285 movdqa %xmm13, %xmm2
1286 psllq $1, %xmm13
1287 psrlq $63, %xmm2
1288 movdqa %xmm2, %xmm1
1289 pslldq $8, %xmm2
1290 psrldq $8, %xmm1
1291 por %xmm2, %xmm13
1292
1293 # Reduction
1294
1295 pshufd $0x24, %xmm1, %xmm2
1296 pcmpeqd TWOONE(%rip), %xmm2
1297 pand POLY(%rip), %xmm2
1298 pxor %xmm2, %xmm13 # %xmm13 holds the HashKey<<1 (mod poly)
1299
1300
1301 # Decrypt first few blocks
1302
1303 movdqa %xmm13, HashKey(%rsp) # store HashKey<<1 (mod poly)
1304 mov %arg4, %r13 # save the number of bytes of plaintext/ciphertext
1305 and $-16, %r13 # %r13 = %r13 - (%r13 mod 16)
1306 mov %r13, %r12
1307 and $(3<<4), %r12
1308 jz _initial_num_blocks_is_0_decrypt
1309 cmp $(2<<4), %r12
1310 jb _initial_num_blocks_is_1_decrypt
1311 je _initial_num_blocks_is_2_decrypt
1312_initial_num_blocks_is_3_decrypt:
1313 INITIAL_BLOCKS_DEC 3, %xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
1314%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 5, 678, dec
1315 sub $48, %r13
1316 jmp _initial_blocks_decrypted
1317_initial_num_blocks_is_2_decrypt:
1318 INITIAL_BLOCKS_DEC 2, %xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
1319%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 6, 78, dec
1320 sub $32, %r13
1321 jmp _initial_blocks_decrypted
1322_initial_num_blocks_is_1_decrypt:
1323 INITIAL_BLOCKS_DEC 1, %xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
1324%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 7, 8, dec
1325 sub $16, %r13
1326 jmp _initial_blocks_decrypted
1327_initial_num_blocks_is_0_decrypt:
1328 INITIAL_BLOCKS_DEC 0, %xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
1329%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 8, 0, dec
1330_initial_blocks_decrypted:
1331 cmp $0, %r13
1332 je _zero_cipher_left_decrypt
1333 sub $64, %r13
1334 je _four_cipher_left_decrypt
1335_decrypt_by_4:
1336 GHASH_4_ENCRYPT_4_PARALLEL_DEC %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, \
1337%xmm14, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7, %xmm8, dec
1338 add $64, %r11
1339 sub $64, %r13
1340 jne _decrypt_by_4
1341_four_cipher_left_decrypt:
1342 GHASH_LAST_4 %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14, \
1343%xmm15, %xmm1, %xmm2, %xmm3, %xmm4, %xmm8
1344_zero_cipher_left_decrypt:
1345 mov %arg4, %r13
1346 and $15, %r13 # %r13 = arg4 (mod 16)
1347 je _multiple_of_16_bytes_decrypt
1348
1349 # Handle the last <16 byte block seperately
1350
1351 paddd ONE(%rip), %xmm0 # increment CNT to get Yn
1352 movdqa SHUF_MASK(%rip), %xmm10
1353 PSHUFB_XMM %xmm10, %xmm0
1354
1355 ENCRYPT_SINGLE_BLOCK %xmm0, %xmm1 # E(K, Yn)
1356 sub $16, %r11
1357 add %r13, %r11
1358 movdqu (%arg3,%r11,1), %xmm1 # recieve the last <16 byte block
1359 lea SHIFT_MASK+16(%rip), %r12
1360 sub %r13, %r12
1361# adjust the shuffle mask pointer to be able to shift 16-%r13 bytes
1362# (%r13 is the number of bytes in plaintext mod 16)
1363 movdqu (%r12), %xmm2 # get the appropriate shuffle mask
1364 PSHUFB_XMM %xmm2, %xmm1 # right shift 16-%r13 butes
1365
1366 movdqa %xmm1, %xmm2
1367 pxor %xmm1, %xmm0 # Ciphertext XOR E(K, Yn)
1368 movdqu ALL_F-SHIFT_MASK(%r12), %xmm1
1369 # get the appropriate mask to mask out top 16-%r13 bytes of %xmm0
1370 pand %xmm1, %xmm0 # mask out top 16-%r13 bytes of %xmm0
1371 pand %xmm1, %xmm2
1372 movdqa SHUF_MASK(%rip), %xmm10
1373 PSHUFB_XMM %xmm10 ,%xmm2
1374
1375 pxor %xmm2, %xmm8
1376 GHASH_MUL %xmm8, %xmm13, %xmm9, %xmm10, %xmm11, %xmm5, %xmm6
1377 # GHASH computation for the last <16 byte block
1378 sub %r13, %r11
1379 add $16, %r11
1380
1381 # output %r13 bytes
1382 MOVQ_R64_XMM %xmm0, %rax
1383 cmp $8, %r13
1384 jle _less_than_8_bytes_left_decrypt
1385 mov %rax, (%arg2 , %r11, 1)
1386 add $8, %r11
1387 psrldq $8, %xmm0
1388 MOVQ_R64_XMM %xmm0, %rax
1389 sub $8, %r13
1390_less_than_8_bytes_left_decrypt:
1391 mov %al, (%arg2, %r11, 1)
1392 add $1, %r11
1393 shr $8, %rax
1394 sub $1, %r13
1395 jne _less_than_8_bytes_left_decrypt
1396_multiple_of_16_bytes_decrypt:
1397 mov arg8, %r12 # %r13 = aadLen (number of bytes)
1398 shl $3, %r12 # convert into number of bits
1399 movd %r12d, %xmm15 # len(A) in %xmm15
1400 shl $3, %arg4 # len(C) in bits (*128)
1401 MOVQ_R64_XMM %arg4, %xmm1
1402 pslldq $8, %xmm15 # %xmm15 = len(A)||0x0000000000000000
1403 pxor %xmm1, %xmm15 # %xmm15 = len(A)||len(C)
1404 pxor %xmm15, %xmm8
1405 GHASH_MUL %xmm8, %xmm13, %xmm9, %xmm10, %xmm11, %xmm5, %xmm6
1406 # final GHASH computation
1407 movdqa SHUF_MASK(%rip), %xmm10
1408 PSHUFB_XMM %xmm10, %xmm8
1409
1410 mov %arg5, %rax # %rax = *Y0
1411 movdqu (%rax), %xmm0 # %xmm0 = Y0
1412 ENCRYPT_SINGLE_BLOCK %xmm0, %xmm1 # E(K, Y0)
1413 pxor %xmm8, %xmm0
1414_return_T_decrypt:
1415 mov arg9, %r10 # %r10 = authTag
1416 mov arg10, %r11 # %r11 = auth_tag_len
1417 cmp $16, %r11
1418 je _T_16_decrypt
1419 cmp $12, %r11
1420 je _T_12_decrypt
1421_T_8_decrypt:
1422 MOVQ_R64_XMM %xmm0, %rax
1423 mov %rax, (%r10)
1424 jmp _return_T_done_decrypt
1425_T_12_decrypt:
1426 MOVQ_R64_XMM %xmm0, %rax
1427 mov %rax, (%r10)
1428 psrldq $8, %xmm0
1429 movd %xmm0, %eax
1430 mov %eax, 8(%r10)
1431 jmp _return_T_done_decrypt
1432_T_16_decrypt:
1433 movdqu %xmm0, (%r10)
1434_return_T_done_decrypt:
1435 mov %r14, %rsp
1436 pop %r14
1437 pop %r13
1438 pop %r12
1439 ret
1440
1441
1442/*****************************************************************************
1443* void aesni_gcm_enc(void *aes_ctx, // AES Key schedule. Starts on a 16 byte boundary.
1444* u8 *out, // Ciphertext output. Encrypt in-place is allowed.
1445* const u8 *in, // Plaintext input
1446* u64 plaintext_len, // Length of data in bytes for encryption.
1447* u8 *iv, // Pre-counter block j0: 4 byte salt (from Security Association)
1448* // concatenated with 8 byte Initialisation Vector (from IPSec ESP Payload)
1449* // concatenated with 0x00000001. 16-byte aligned pointer.
1450* u8 *hash_subkey, // H, the Hash sub key input. Data starts on a 16-byte boundary.
1451* const u8 *aad, // Additional Authentication Data (AAD)
1452* u64 aad_len, // Length of AAD in bytes. With RFC4106 this is going to be 8 or 12 bytes
1453* u8 *auth_tag, // Authenticated Tag output.
1454* u64 auth_tag_len); // Authenticated Tag Length in bytes. Valid values are 16 (most likely),
1455* // 12 or 8.
1456*
1457* Assumptions:
1458*
1459* keys:
1460* keys are pre-expanded and aligned to 16 bytes. we are using the
1461* first set of 11 keys in the data structure void *aes_ctx
1462*
1463*
1464* iv:
1465* 0 1 2 3
1466* 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
1467* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1468* | Salt (From the SA) |
1469* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1470* | Initialization Vector |
1471* | (This is the sequence number from IPSec header) |
1472* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1473* | 0x1 |
1474* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1475*
1476*
1477*
1478* AAD:
1479* AAD padded to 128 bits with 0
1480* for example, assume AAD is a u32 vector
1481*
1482* if AAD is 8 bytes:
1483* AAD[3] = {A0, A1};
1484* padded AAD in xmm register = {A1 A0 0 0}
1485*
1486* 0 1 2 3
1487* 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
1488* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1489* | SPI (A1) |
1490* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1491* | 32-bit Sequence Number (A0) |
1492* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1493* | 0x0 |
1494* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1495*
1496* AAD Format with 32-bit Sequence Number
1497*
1498* if AAD is 12 bytes:
1499* AAD[3] = {A0, A1, A2};
1500* padded AAD in xmm register = {A2 A1 A0 0}
1501*
1502* 0 1 2 3
1503* 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
1504* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1505* | SPI (A2) |
1506* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1507* | 64-bit Extended Sequence Number {A1,A0} |
1508* | |
1509* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1510* | 0x0 |
1511* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1512*
1513* AAD Format with 64-bit Extended Sequence Number
1514*
1515* aadLen:
1516* from the definition of the spec, aadLen can only be 8 or 12 bytes.
1517* The code supports 16 too but for other sizes, the code will fail.
1518*
1519* TLen:
1520* from the definition of the spec, TLen can only be 8, 12 or 16 bytes.
1521* For other sizes, the code will fail.
1522*
1523* poly = x^128 + x^127 + x^126 + x^121 + 1
1524***************************************************************************/
1525ENTRY(aesni_gcm_enc)
1526 push %r12
1527 push %r13
1528 push %r14
1529 mov %rsp, %r14
1530#
1531# states of %xmm registers %xmm6:%xmm15 not saved
1532# all %xmm registers are clobbered
1533#
1534 sub $VARIABLE_OFFSET, %rsp
1535 and $~63, %rsp
1536 mov %arg6, %r12
1537 movdqu (%r12), %xmm13
1538 movdqa SHUF_MASK(%rip), %xmm2
1539 PSHUFB_XMM %xmm2, %xmm13
1540
1541
1542# precompute HashKey<<1 mod poly from the HashKey (required for GHASH)
1543
1544 movdqa %xmm13, %xmm2
1545 psllq $1, %xmm13
1546 psrlq $63, %xmm2
1547 movdqa %xmm2, %xmm1
1548 pslldq $8, %xmm2
1549 psrldq $8, %xmm1
1550 por %xmm2, %xmm13
1551
1552 # reduce HashKey<<1
1553
1554 pshufd $0x24, %xmm1, %xmm2
1555 pcmpeqd TWOONE(%rip), %xmm2
1556 pand POLY(%rip), %xmm2
1557 pxor %xmm2, %xmm13
1558 movdqa %xmm13, HashKey(%rsp)
1559 mov %arg4, %r13 # %xmm13 holds HashKey<<1 (mod poly)
1560 and $-16, %r13
1561 mov %r13, %r12
1562
1563 # Encrypt first few blocks
1564
1565 and $(3<<4), %r12
1566 jz _initial_num_blocks_is_0_encrypt
1567 cmp $(2<<4), %r12
1568 jb _initial_num_blocks_is_1_encrypt
1569 je _initial_num_blocks_is_2_encrypt
1570_initial_num_blocks_is_3_encrypt:
1571 INITIAL_BLOCKS_ENC 3, %xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
1572%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 5, 678, enc
1573 sub $48, %r13
1574 jmp _initial_blocks_encrypted
1575_initial_num_blocks_is_2_encrypt:
1576 INITIAL_BLOCKS_ENC 2, %xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
1577%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 6, 78, enc
1578 sub $32, %r13
1579 jmp _initial_blocks_encrypted
1580_initial_num_blocks_is_1_encrypt:
1581 INITIAL_BLOCKS_ENC 1, %xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
1582%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 7, 8, enc
1583 sub $16, %r13
1584 jmp _initial_blocks_encrypted
1585_initial_num_blocks_is_0_encrypt:
1586 INITIAL_BLOCKS_ENC 0, %xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
1587%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 8, 0, enc
1588_initial_blocks_encrypted:
1589
1590 # Main loop - Encrypt remaining blocks
1591
1592 cmp $0, %r13
1593 je _zero_cipher_left_encrypt
1594 sub $64, %r13
1595 je _four_cipher_left_encrypt
1596_encrypt_by_4_encrypt:
1597 GHASH_4_ENCRYPT_4_PARALLEL_ENC %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, \
1598%xmm14, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7, %xmm8, enc
1599 add $64, %r11
1600 sub $64, %r13
1601 jne _encrypt_by_4_encrypt
1602_four_cipher_left_encrypt:
1603 GHASH_LAST_4 %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14, \
1604%xmm15, %xmm1, %xmm2, %xmm3, %xmm4, %xmm8
1605_zero_cipher_left_encrypt:
1606 mov %arg4, %r13
1607 and $15, %r13 # %r13 = arg4 (mod 16)
1608 je _multiple_of_16_bytes_encrypt
1609
1610 # Handle the last <16 Byte block seperately
1611 paddd ONE(%rip), %xmm0 # INCR CNT to get Yn
1612 movdqa SHUF_MASK(%rip), %xmm10
1613 PSHUFB_XMM %xmm10, %xmm0
1614
1615 ENCRYPT_SINGLE_BLOCK %xmm0, %xmm1 # Encrypt(K, Yn)
1616 sub $16, %r11
1617 add %r13, %r11
1618 movdqu (%arg3,%r11,1), %xmm1 # receive the last <16 byte blocks
1619 lea SHIFT_MASK+16(%rip), %r12
1620 sub %r13, %r12
1621 # adjust the shuffle mask pointer to be able to shift 16-r13 bytes
1622 # (%r13 is the number of bytes in plaintext mod 16)
1623 movdqu (%r12), %xmm2 # get the appropriate shuffle mask
1624 PSHUFB_XMM %xmm2, %xmm1 # shift right 16-r13 byte
1625 pxor %xmm1, %xmm0 # Plaintext XOR Encrypt(K, Yn)
1626 movdqu ALL_F-SHIFT_MASK(%r12), %xmm1
1627 # get the appropriate mask to mask out top 16-r13 bytes of xmm0
1628 pand %xmm1, %xmm0 # mask out top 16-r13 bytes of xmm0
1629 movdqa SHUF_MASK(%rip), %xmm10
1630 PSHUFB_XMM %xmm10,%xmm0
1631
1632 pxor %xmm0, %xmm8
1633 GHASH_MUL %xmm8, %xmm13, %xmm9, %xmm10, %xmm11, %xmm5, %xmm6
1634 # GHASH computation for the last <16 byte block
1635 sub %r13, %r11
1636 add $16, %r11
1637 PSHUFB_XMM %xmm10, %xmm1
1638
1639 # shuffle xmm0 back to output as ciphertext
1640
1641 # Output %r13 bytes
1642 MOVQ_R64_XMM %xmm0, %rax
1643 cmp $8, %r13
1644 jle _less_than_8_bytes_left_encrypt
1645 mov %rax, (%arg2 , %r11, 1)
1646 add $8, %r11
1647 psrldq $8, %xmm0
1648 MOVQ_R64_XMM %xmm0, %rax
1649 sub $8, %r13
1650_less_than_8_bytes_left_encrypt:
1651 mov %al, (%arg2, %r11, 1)
1652 add $1, %r11
1653 shr $8, %rax
1654 sub $1, %r13
1655 jne _less_than_8_bytes_left_encrypt
1656_multiple_of_16_bytes_encrypt:
1657 mov arg8, %r12 # %r12 = addLen (number of bytes)
1658 shl $3, %r12
1659 movd %r12d, %xmm15 # len(A) in %xmm15
1660 shl $3, %arg4 # len(C) in bits (*128)
1661 MOVQ_R64_XMM %arg4, %xmm1
1662 pslldq $8, %xmm15 # %xmm15 = len(A)||0x0000000000000000
1663 pxor %xmm1, %xmm15 # %xmm15 = len(A)||len(C)
1664 pxor %xmm15, %xmm8
1665 GHASH_MUL %xmm8, %xmm13, %xmm9, %xmm10, %xmm11, %xmm5, %xmm6
1666 # final GHASH computation
1667 movdqa SHUF_MASK(%rip), %xmm10
1668 PSHUFB_XMM %xmm10, %xmm8 # perform a 16 byte swap
1669
1670 mov %arg5, %rax # %rax = *Y0
1671 movdqu (%rax), %xmm0 # %xmm0 = Y0
1672 ENCRYPT_SINGLE_BLOCK %xmm0, %xmm15 # Encrypt(K, Y0)
1673 pxor %xmm8, %xmm0
1674_return_T_encrypt:
1675 mov arg9, %r10 # %r10 = authTag
1676 mov arg10, %r11 # %r11 = auth_tag_len
1677 cmp $16, %r11
1678 je _T_16_encrypt
1679 cmp $12, %r11
1680 je _T_12_encrypt
1681_T_8_encrypt:
1682 MOVQ_R64_XMM %xmm0, %rax
1683 mov %rax, (%r10)
1684 jmp _return_T_done_encrypt
1685_T_12_encrypt:
1686 MOVQ_R64_XMM %xmm0, %rax
1687 mov %rax, (%r10)
1688 psrldq $8, %xmm0
1689 movd %xmm0, %eax
1690 mov %eax, 8(%r10)
1691 jmp _return_T_done_encrypt
1692_T_16_encrypt:
1693 movdqu %xmm0, (%r10)
1694_return_T_done_encrypt:
1695 mov %r14, %rsp
1696 pop %r14
1697 pop %r13
1698 pop %r12
1699 ret
1700
1701#endif
1702
49 1703
50_key_expansion_128: 1704_key_expansion_128:
51_key_expansion_256a: 1705_key_expansion_256a:
@@ -55,10 +1709,11 @@ _key_expansion_256a:
55 shufps $0b10001100, %xmm0, %xmm4 1709 shufps $0b10001100, %xmm0, %xmm4
56 pxor %xmm4, %xmm0 1710 pxor %xmm4, %xmm0
57 pxor %xmm1, %xmm0 1711 pxor %xmm1, %xmm0
58 movaps %xmm0, (%rcx) 1712 movaps %xmm0, (TKEYP)
59 add $0x10, %rcx 1713 add $0x10, TKEYP
60 ret 1714 ret
61 1715
1716.align 4
62_key_expansion_192a: 1717_key_expansion_192a:
63 pshufd $0b01010101, %xmm1, %xmm1 1718 pshufd $0b01010101, %xmm1, %xmm1
64 shufps $0b00010000, %xmm0, %xmm4 1719 shufps $0b00010000, %xmm0, %xmm4
@@ -76,12 +1731,13 @@ _key_expansion_192a:
76 1731
77 movaps %xmm0, %xmm1 1732 movaps %xmm0, %xmm1
78 shufps $0b01000100, %xmm0, %xmm6 1733 shufps $0b01000100, %xmm0, %xmm6
79 movaps %xmm6, (%rcx) 1734 movaps %xmm6, (TKEYP)
80 shufps $0b01001110, %xmm2, %xmm1 1735 shufps $0b01001110, %xmm2, %xmm1
81 movaps %xmm1, 16(%rcx) 1736 movaps %xmm1, 0x10(TKEYP)
82 add $0x20, %rcx 1737 add $0x20, TKEYP
83 ret 1738 ret
84 1739
1740.align 4
85_key_expansion_192b: 1741_key_expansion_192b:
86 pshufd $0b01010101, %xmm1, %xmm1 1742 pshufd $0b01010101, %xmm1, %xmm1
87 shufps $0b00010000, %xmm0, %xmm4 1743 shufps $0b00010000, %xmm0, %xmm4
@@ -96,10 +1752,11 @@ _key_expansion_192b:
96 pxor %xmm3, %xmm2 1752 pxor %xmm3, %xmm2
97 pxor %xmm5, %xmm2 1753 pxor %xmm5, %xmm2
98 1754
99 movaps %xmm0, (%rcx) 1755 movaps %xmm0, (TKEYP)
100 add $0x10, %rcx 1756 add $0x10, TKEYP
101 ret 1757 ret
102 1758
1759.align 4
103_key_expansion_256b: 1760_key_expansion_256b:
104 pshufd $0b10101010, %xmm1, %xmm1 1761 pshufd $0b10101010, %xmm1, %xmm1
105 shufps $0b00010000, %xmm2, %xmm4 1762 shufps $0b00010000, %xmm2, %xmm4
@@ -107,8 +1764,8 @@ _key_expansion_256b:
107 shufps $0b10001100, %xmm2, %xmm4 1764 shufps $0b10001100, %xmm2, %xmm4
108 pxor %xmm4, %xmm2 1765 pxor %xmm4, %xmm2
109 pxor %xmm1, %xmm2 1766 pxor %xmm1, %xmm2
110 movaps %xmm2, (%rcx) 1767 movaps %xmm2, (TKEYP)
111 add $0x10, %rcx 1768 add $0x10, TKEYP
112 ret 1769 ret
113 1770
114/* 1771/*
@@ -116,17 +1773,23 @@ _key_expansion_256b:
116 * unsigned int key_len) 1773 * unsigned int key_len)
117 */ 1774 */
118ENTRY(aesni_set_key) 1775ENTRY(aesni_set_key)
119 movups (%rsi), %xmm0 # user key (first 16 bytes) 1776#ifndef __x86_64__
120 movaps %xmm0, (%rdi) 1777 pushl KEYP
121 lea 0x10(%rdi), %rcx # key addr 1778 movl 8(%esp), KEYP # ctx
122 movl %edx, 480(%rdi) 1779 movl 12(%esp), UKEYP # in_key
1780 movl 16(%esp), %edx # key_len
1781#endif
1782 movups (UKEYP), %xmm0 # user key (first 16 bytes)
1783 movaps %xmm0, (KEYP)
1784 lea 0x10(KEYP), TKEYP # key addr
1785 movl %edx, 480(KEYP)
123 pxor %xmm4, %xmm4 # xmm4 is assumed 0 in _key_expansion_x 1786 pxor %xmm4, %xmm4 # xmm4 is assumed 0 in _key_expansion_x
124 cmp $24, %dl 1787 cmp $24, %dl
125 jb .Lenc_key128 1788 jb .Lenc_key128
126 je .Lenc_key192 1789 je .Lenc_key192
127 movups 0x10(%rsi), %xmm2 # other user key 1790 movups 0x10(UKEYP), %xmm2 # other user key
128 movaps %xmm2, (%rcx) 1791 movaps %xmm2, (TKEYP)
129 add $0x10, %rcx 1792 add $0x10, TKEYP
130 AESKEYGENASSIST 0x1 %xmm2 %xmm1 # round 1 1793 AESKEYGENASSIST 0x1 %xmm2 %xmm1 # round 1
131 call _key_expansion_256a 1794 call _key_expansion_256a
132 AESKEYGENASSIST 0x1 %xmm0 %xmm1 1795 AESKEYGENASSIST 0x1 %xmm0 %xmm1
@@ -155,7 +1818,7 @@ ENTRY(aesni_set_key)
155 call _key_expansion_256a 1818 call _key_expansion_256a
156 jmp .Ldec_key 1819 jmp .Ldec_key
157.Lenc_key192: 1820.Lenc_key192:
158 movq 0x10(%rsi), %xmm2 # other user key 1821 movq 0x10(UKEYP), %xmm2 # other user key
159 AESKEYGENASSIST 0x1 %xmm2 %xmm1 # round 1 1822 AESKEYGENASSIST 0x1 %xmm2 %xmm1 # round 1
160 call _key_expansion_192a 1823 call _key_expansion_192a
161 AESKEYGENASSIST 0x2 %xmm2 %xmm1 # round 2 1824 AESKEYGENASSIST 0x2 %xmm2 %xmm1 # round 2
@@ -195,33 +1858,47 @@ ENTRY(aesni_set_key)
195 AESKEYGENASSIST 0x36 %xmm0 %xmm1 # round 10 1858 AESKEYGENASSIST 0x36 %xmm0 %xmm1 # round 10
196 call _key_expansion_128 1859 call _key_expansion_128
197.Ldec_key: 1860.Ldec_key:
198 sub $0x10, %rcx 1861 sub $0x10, TKEYP
199 movaps (%rdi), %xmm0 1862 movaps (KEYP), %xmm0
200 movaps (%rcx), %xmm1 1863 movaps (TKEYP), %xmm1
201 movaps %xmm0, 240(%rcx) 1864 movaps %xmm0, 240(TKEYP)
202 movaps %xmm1, 240(%rdi) 1865 movaps %xmm1, 240(KEYP)
203 add $0x10, %rdi 1866 add $0x10, KEYP
204 lea 240-16(%rcx), %rsi 1867 lea 240-16(TKEYP), UKEYP
205.align 4 1868.align 4
206.Ldec_key_loop: 1869.Ldec_key_loop:
207 movaps (%rdi), %xmm0 1870 movaps (KEYP), %xmm0
208 AESIMC %xmm0 %xmm1 1871 AESIMC %xmm0 %xmm1
209 movaps %xmm1, (%rsi) 1872 movaps %xmm1, (UKEYP)
210 add $0x10, %rdi 1873 add $0x10, KEYP
211 sub $0x10, %rsi 1874 sub $0x10, UKEYP
212 cmp %rcx, %rdi 1875 cmp TKEYP, KEYP
213 jb .Ldec_key_loop 1876 jb .Ldec_key_loop
214 xor %rax, %rax 1877 xor AREG, AREG
1878#ifndef __x86_64__
1879 popl KEYP
1880#endif
215 ret 1881 ret
216 1882
217/* 1883/*
218 * void aesni_enc(struct crypto_aes_ctx *ctx, u8 *dst, const u8 *src) 1884 * void aesni_enc(struct crypto_aes_ctx *ctx, u8 *dst, const u8 *src)
219 */ 1885 */
220ENTRY(aesni_enc) 1886ENTRY(aesni_enc)
1887#ifndef __x86_64__
1888 pushl KEYP
1889 pushl KLEN
1890 movl 12(%esp), KEYP
1891 movl 16(%esp), OUTP
1892 movl 20(%esp), INP
1893#endif
221 movl 480(KEYP), KLEN # key length 1894 movl 480(KEYP), KLEN # key length
222 movups (INP), STATE # input 1895 movups (INP), STATE # input
223 call _aesni_enc1 1896 call _aesni_enc1
224 movups STATE, (OUTP) # output 1897 movups STATE, (OUTP) # output
1898#ifndef __x86_64__
1899 popl KLEN
1900 popl KEYP
1901#endif
225 ret 1902 ret
226 1903
227/* 1904/*
@@ -236,6 +1913,7 @@ ENTRY(aesni_enc)
236 * KEY 1913 * KEY
237 * TKEYP (T1) 1914 * TKEYP (T1)
238 */ 1915 */
1916.align 4
239_aesni_enc1: 1917_aesni_enc1:
240 movaps (KEYP), KEY # key 1918 movaps (KEYP), KEY # key
241 mov KEYP, TKEYP 1919 mov KEYP, TKEYP
@@ -298,6 +1976,7 @@ _aesni_enc1:
298 * KEY 1976 * KEY
299 * TKEYP (T1) 1977 * TKEYP (T1)
300 */ 1978 */
1979.align 4
301_aesni_enc4: 1980_aesni_enc4:
302 movaps (KEYP), KEY # key 1981 movaps (KEYP), KEY # key
303 mov KEYP, TKEYP 1982 mov KEYP, TKEYP
@@ -391,11 +2070,22 @@ _aesni_enc4:
391 * void aesni_dec (struct crypto_aes_ctx *ctx, u8 *dst, const u8 *src) 2070 * void aesni_dec (struct crypto_aes_ctx *ctx, u8 *dst, const u8 *src)
392 */ 2071 */
393ENTRY(aesni_dec) 2072ENTRY(aesni_dec)
2073#ifndef __x86_64__
2074 pushl KEYP
2075 pushl KLEN
2076 movl 12(%esp), KEYP
2077 movl 16(%esp), OUTP
2078 movl 20(%esp), INP
2079#endif
394 mov 480(KEYP), KLEN # key length 2080 mov 480(KEYP), KLEN # key length
395 add $240, KEYP 2081 add $240, KEYP
396 movups (INP), STATE # input 2082 movups (INP), STATE # input
397 call _aesni_dec1 2083 call _aesni_dec1
398 movups STATE, (OUTP) #output 2084 movups STATE, (OUTP) #output
2085#ifndef __x86_64__
2086 popl KLEN
2087 popl KEYP
2088#endif
399 ret 2089 ret
400 2090
401/* 2091/*
@@ -410,6 +2100,7 @@ ENTRY(aesni_dec)
410 * KEY 2100 * KEY
411 * TKEYP (T1) 2101 * TKEYP (T1)
412 */ 2102 */
2103.align 4
413_aesni_dec1: 2104_aesni_dec1:
414 movaps (KEYP), KEY # key 2105 movaps (KEYP), KEY # key
415 mov KEYP, TKEYP 2106 mov KEYP, TKEYP
@@ -472,6 +2163,7 @@ _aesni_dec1:
472 * KEY 2163 * KEY
473 * TKEYP (T1) 2164 * TKEYP (T1)
474 */ 2165 */
2166.align 4
475_aesni_dec4: 2167_aesni_dec4:
476 movaps (KEYP), KEY # key 2168 movaps (KEYP), KEY # key
477 mov KEYP, TKEYP 2169 mov KEYP, TKEYP
@@ -566,6 +2258,15 @@ _aesni_dec4:
566 * size_t len) 2258 * size_t len)
567 */ 2259 */
568ENTRY(aesni_ecb_enc) 2260ENTRY(aesni_ecb_enc)
2261#ifndef __x86_64__
2262 pushl LEN
2263 pushl KEYP
2264 pushl KLEN
2265 movl 16(%esp), KEYP
2266 movl 20(%esp), OUTP
2267 movl 24(%esp), INP
2268 movl 28(%esp), LEN
2269#endif
569 test LEN, LEN # check length 2270 test LEN, LEN # check length
570 jz .Lecb_enc_ret 2271 jz .Lecb_enc_ret
571 mov 480(KEYP), KLEN 2272 mov 480(KEYP), KLEN
@@ -602,6 +2303,11 @@ ENTRY(aesni_ecb_enc)
602 cmp $16, LEN 2303 cmp $16, LEN
603 jge .Lecb_enc_loop1 2304 jge .Lecb_enc_loop1
604.Lecb_enc_ret: 2305.Lecb_enc_ret:
2306#ifndef __x86_64__
2307 popl KLEN
2308 popl KEYP
2309 popl LEN
2310#endif
605 ret 2311 ret
606 2312
607/* 2313/*
@@ -609,6 +2315,15 @@ ENTRY(aesni_ecb_enc)
609 * size_t len); 2315 * size_t len);
610 */ 2316 */
611ENTRY(aesni_ecb_dec) 2317ENTRY(aesni_ecb_dec)
2318#ifndef __x86_64__
2319 pushl LEN
2320 pushl KEYP
2321 pushl KLEN
2322 movl 16(%esp), KEYP
2323 movl 20(%esp), OUTP
2324 movl 24(%esp), INP
2325 movl 28(%esp), LEN
2326#endif
612 test LEN, LEN 2327 test LEN, LEN
613 jz .Lecb_dec_ret 2328 jz .Lecb_dec_ret
614 mov 480(KEYP), KLEN 2329 mov 480(KEYP), KLEN
@@ -646,6 +2361,11 @@ ENTRY(aesni_ecb_dec)
646 cmp $16, LEN 2361 cmp $16, LEN
647 jge .Lecb_dec_loop1 2362 jge .Lecb_dec_loop1
648.Lecb_dec_ret: 2363.Lecb_dec_ret:
2364#ifndef __x86_64__
2365 popl KLEN
2366 popl KEYP
2367 popl LEN
2368#endif
649 ret 2369 ret
650 2370
651/* 2371/*
@@ -653,6 +2373,17 @@ ENTRY(aesni_ecb_dec)
653 * size_t len, u8 *iv) 2373 * size_t len, u8 *iv)
654 */ 2374 */
655ENTRY(aesni_cbc_enc) 2375ENTRY(aesni_cbc_enc)
2376#ifndef __x86_64__
2377 pushl IVP
2378 pushl LEN
2379 pushl KEYP
2380 pushl KLEN
2381 movl 20(%esp), KEYP
2382 movl 24(%esp), OUTP
2383 movl 28(%esp), INP
2384 movl 32(%esp), LEN
2385 movl 36(%esp), IVP
2386#endif
656 cmp $16, LEN 2387 cmp $16, LEN
657 jb .Lcbc_enc_ret 2388 jb .Lcbc_enc_ret
658 mov 480(KEYP), KLEN 2389 mov 480(KEYP), KLEN
@@ -670,6 +2401,12 @@ ENTRY(aesni_cbc_enc)
670 jge .Lcbc_enc_loop 2401 jge .Lcbc_enc_loop
671 movups STATE, (IVP) 2402 movups STATE, (IVP)
672.Lcbc_enc_ret: 2403.Lcbc_enc_ret:
2404#ifndef __x86_64__
2405 popl KLEN
2406 popl KEYP
2407 popl LEN
2408 popl IVP
2409#endif
673 ret 2410 ret
674 2411
675/* 2412/*
@@ -677,6 +2414,17 @@ ENTRY(aesni_cbc_enc)
677 * size_t len, u8 *iv) 2414 * size_t len, u8 *iv)
678 */ 2415 */
679ENTRY(aesni_cbc_dec) 2416ENTRY(aesni_cbc_dec)
2417#ifndef __x86_64__
2418 pushl IVP
2419 pushl LEN
2420 pushl KEYP
2421 pushl KLEN
2422 movl 20(%esp), KEYP
2423 movl 24(%esp), OUTP
2424 movl 28(%esp), INP
2425 movl 32(%esp), LEN
2426 movl 36(%esp), IVP
2427#endif
680 cmp $16, LEN 2428 cmp $16, LEN
681 jb .Lcbc_dec_just_ret 2429 jb .Lcbc_dec_just_ret
682 mov 480(KEYP), KLEN 2430 mov 480(KEYP), KLEN
@@ -690,16 +2438,30 @@ ENTRY(aesni_cbc_dec)
690 movaps IN1, STATE1 2438 movaps IN1, STATE1
691 movups 0x10(INP), IN2 2439 movups 0x10(INP), IN2
692 movaps IN2, STATE2 2440 movaps IN2, STATE2
2441#ifdef __x86_64__
693 movups 0x20(INP), IN3 2442 movups 0x20(INP), IN3
694 movaps IN3, STATE3 2443 movaps IN3, STATE3
695 movups 0x30(INP), IN4 2444 movups 0x30(INP), IN4
696 movaps IN4, STATE4 2445 movaps IN4, STATE4
2446#else
2447 movups 0x20(INP), IN1
2448 movaps IN1, STATE3
2449 movups 0x30(INP), IN2
2450 movaps IN2, STATE4
2451#endif
697 call _aesni_dec4 2452 call _aesni_dec4
698 pxor IV, STATE1 2453 pxor IV, STATE1
2454#ifdef __x86_64__
699 pxor IN1, STATE2 2455 pxor IN1, STATE2
700 pxor IN2, STATE3 2456 pxor IN2, STATE3
701 pxor IN3, STATE4 2457 pxor IN3, STATE4
702 movaps IN4, IV 2458 movaps IN4, IV
2459#else
2460 pxor (INP), STATE2
2461 pxor 0x10(INP), STATE3
2462 pxor IN1, STATE4
2463 movaps IN2, IV
2464#endif
703 movups STATE1, (OUTP) 2465 movups STATE1, (OUTP)
704 movups STATE2, 0x10(OUTP) 2466 movups STATE2, 0x10(OUTP)
705 movups STATE3, 0x20(OUTP) 2467 movups STATE3, 0x20(OUTP)
@@ -727,8 +2489,15 @@ ENTRY(aesni_cbc_dec)
727.Lcbc_dec_ret: 2489.Lcbc_dec_ret:
728 movups IV, (IVP) 2490 movups IV, (IVP)
729.Lcbc_dec_just_ret: 2491.Lcbc_dec_just_ret:
2492#ifndef __x86_64__
2493 popl KLEN
2494 popl KEYP
2495 popl LEN
2496 popl IVP
2497#endif
730 ret 2498 ret
731 2499
2500#ifdef __x86_64__
732.align 16 2501.align 16
733.Lbswap_mask: 2502.Lbswap_mask:
734 .byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0 2503 .byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
@@ -744,6 +2513,7 @@ ENTRY(aesni_cbc_dec)
744 * INC: == 1, in little endian 2513 * INC: == 1, in little endian
745 * BSWAP_MASK == endian swapping mask 2514 * BSWAP_MASK == endian swapping mask
746 */ 2515 */
2516.align 4
747_aesni_inc_init: 2517_aesni_inc_init:
748 movaps .Lbswap_mask, BSWAP_MASK 2518 movaps .Lbswap_mask, BSWAP_MASK
749 movaps IV, CTR 2519 movaps IV, CTR
@@ -768,6 +2538,7 @@ _aesni_inc_init:
768 * CTR: == output IV, in little endian 2538 * CTR: == output IV, in little endian
769 * TCTR_LOW: == lower qword of CTR 2539 * TCTR_LOW: == lower qword of CTR
770 */ 2540 */
2541.align 4
771_aesni_inc: 2542_aesni_inc:
772 paddq INC, CTR 2543 paddq INC, CTR
773 add $1, TCTR_LOW 2544 add $1, TCTR_LOW
@@ -839,3 +2610,4 @@ ENTRY(aesni_ctr_enc)
839 movups IV, (IVP) 2610 movups IV, (IVP)
840.Lctr_enc_just_ret: 2611.Lctr_enc_just_ret:
841 ret 2612 ret
2613#endif
diff --git a/arch/x86/crypto/aesni-intel_glue.c b/arch/x86/crypto/aesni-intel_glue.c
index 2cb3dcc4490..e1e60c7d581 100644
--- a/arch/x86/crypto/aesni-intel_glue.c
+++ b/arch/x86/crypto/aesni-intel_glue.c
@@ -5,6 +5,14 @@
5 * Copyright (C) 2008, Intel Corp. 5 * Copyright (C) 2008, Intel Corp.
6 * Author: Huang Ying <ying.huang@intel.com> 6 * Author: Huang Ying <ying.huang@intel.com>
7 * 7 *
8 * Added RFC4106 AES-GCM support for 128-bit keys under the AEAD
9 * interface for 64-bit kernels.
10 * Authors: Adrian Hoban <adrian.hoban@intel.com>
11 * Gabriele Paoloni <gabriele.paoloni@intel.com>
12 * Tadeusz Struk (tadeusz.struk@intel.com)
13 * Aidan O'Mahony (aidan.o.mahony@intel.com)
14 * Copyright (c) 2010, Intel Corporation.
15 *
8 * This program is free software; you can redistribute it and/or modify 16 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by 17 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or 18 * the Free Software Foundation; either version 2 of the License, or
@@ -21,6 +29,10 @@
21#include <crypto/ctr.h> 29#include <crypto/ctr.h>
22#include <asm/i387.h> 30#include <asm/i387.h>
23#include <asm/aes.h> 31#include <asm/aes.h>
32#include <crypto/scatterwalk.h>
33#include <crypto/internal/aead.h>
34#include <linux/workqueue.h>
35#include <linux/spinlock.h>
24 36
25#if defined(CONFIG_CRYPTO_CTR) || defined(CONFIG_CRYPTO_CTR_MODULE) 37#if defined(CONFIG_CRYPTO_CTR) || defined(CONFIG_CRYPTO_CTR_MODULE)
26#define HAS_CTR 38#define HAS_CTR
@@ -42,8 +54,31 @@ struct async_aes_ctx {
42 struct cryptd_ablkcipher *cryptd_tfm; 54 struct cryptd_ablkcipher *cryptd_tfm;
43}; 55};
44 56
45#define AESNI_ALIGN 16 57/* This data is stored at the end of the crypto_tfm struct.
58 * It's a type of per "session" data storage location.
59 * This needs to be 16 byte aligned.
60 */
61struct aesni_rfc4106_gcm_ctx {
62 u8 hash_subkey[16];
63 struct crypto_aes_ctx aes_key_expanded;
64 u8 nonce[4];
65 struct cryptd_aead *cryptd_tfm;
66};
67
68struct aesni_gcm_set_hash_subkey_result {
69 int err;
70 struct completion completion;
71};
72
73struct aesni_hash_subkey_req_data {
74 u8 iv[16];
75 struct aesni_gcm_set_hash_subkey_result result;
76 struct scatterlist sg;
77};
78
79#define AESNI_ALIGN (16)
46#define AES_BLOCK_MASK (~(AES_BLOCK_SIZE-1)) 80#define AES_BLOCK_MASK (~(AES_BLOCK_SIZE-1))
81#define RFC4106_HASH_SUBKEY_SIZE 16
47 82
48asmlinkage int aesni_set_key(struct crypto_aes_ctx *ctx, const u8 *in_key, 83asmlinkage int aesni_set_key(struct crypto_aes_ctx *ctx, const u8 *in_key,
49 unsigned int key_len); 84 unsigned int key_len);
@@ -59,9 +94,62 @@ asmlinkage void aesni_cbc_enc(struct crypto_aes_ctx *ctx, u8 *out,
59 const u8 *in, unsigned int len, u8 *iv); 94 const u8 *in, unsigned int len, u8 *iv);
60asmlinkage void aesni_cbc_dec(struct crypto_aes_ctx *ctx, u8 *out, 95asmlinkage void aesni_cbc_dec(struct crypto_aes_ctx *ctx, u8 *out,
61 const u8 *in, unsigned int len, u8 *iv); 96 const u8 *in, unsigned int len, u8 *iv);
97#ifdef CONFIG_X86_64
62asmlinkage void aesni_ctr_enc(struct crypto_aes_ctx *ctx, u8 *out, 98asmlinkage void aesni_ctr_enc(struct crypto_aes_ctx *ctx, u8 *out,
63 const u8 *in, unsigned int len, u8 *iv); 99 const u8 *in, unsigned int len, u8 *iv);
64 100
101/* asmlinkage void aesni_gcm_enc()
102 * void *ctx, AES Key schedule. Starts on a 16 byte boundary.
103 * u8 *out, Ciphertext output. Encrypt in-place is allowed.
104 * const u8 *in, Plaintext input
105 * unsigned long plaintext_len, Length of data in bytes for encryption.
106 * u8 *iv, Pre-counter block j0: 4 byte salt (from Security Association)
107 * concatenated with 8 byte Initialisation Vector (from IPSec ESP
108 * Payload) concatenated with 0x00000001. 16-byte aligned pointer.
109 * u8 *hash_subkey, the Hash sub key input. Data starts on a 16-byte boundary.
110 * const u8 *aad, Additional Authentication Data (AAD)
111 * unsigned long aad_len, Length of AAD in bytes. With RFC4106 this
112 * is going to be 8 or 12 bytes
113 * u8 *auth_tag, Authenticated Tag output.
114 * unsigned long auth_tag_len), Authenticated Tag Length in bytes.
115 * Valid values are 16 (most likely), 12 or 8.
116 */
117asmlinkage void aesni_gcm_enc(void *ctx, u8 *out,
118 const u8 *in, unsigned long plaintext_len, u8 *iv,
119 u8 *hash_subkey, const u8 *aad, unsigned long aad_len,
120 u8 *auth_tag, unsigned long auth_tag_len);
121
122/* asmlinkage void aesni_gcm_dec()
123 * void *ctx, AES Key schedule. Starts on a 16 byte boundary.
124 * u8 *out, Plaintext output. Decrypt in-place is allowed.
125 * const u8 *in, Ciphertext input
126 * unsigned long ciphertext_len, Length of data in bytes for decryption.
127 * u8 *iv, Pre-counter block j0: 4 byte salt (from Security Association)
128 * concatenated with 8 byte Initialisation Vector (from IPSec ESP
129 * Payload) concatenated with 0x00000001. 16-byte aligned pointer.
130 * u8 *hash_subkey, the Hash sub key input. Data starts on a 16-byte boundary.
131 * const u8 *aad, Additional Authentication Data (AAD)
132 * unsigned long aad_len, Length of AAD in bytes. With RFC4106 this is going
133 * to be 8 or 12 bytes
134 * u8 *auth_tag, Authenticated Tag output.
135 * unsigned long auth_tag_len) Authenticated Tag Length in bytes.
136 * Valid values are 16 (most likely), 12 or 8.
137 */
138asmlinkage void aesni_gcm_dec(void *ctx, u8 *out,
139 const u8 *in, unsigned long ciphertext_len, u8 *iv,
140 u8 *hash_subkey, const u8 *aad, unsigned long aad_len,
141 u8 *auth_tag, unsigned long auth_tag_len);
142
143static inline struct
144aesni_rfc4106_gcm_ctx *aesni_rfc4106_gcm_ctx_get(struct crypto_aead *tfm)
145{
146 return
147 (struct aesni_rfc4106_gcm_ctx *)
148 PTR_ALIGN((u8 *)
149 crypto_tfm_ctx(crypto_aead_tfm(tfm)), AESNI_ALIGN);
150}
151#endif
152
65static inline struct crypto_aes_ctx *aes_ctx(void *raw_ctx) 153static inline struct crypto_aes_ctx *aes_ctx(void *raw_ctx)
66{ 154{
67 unsigned long addr = (unsigned long)raw_ctx; 155 unsigned long addr = (unsigned long)raw_ctx;
@@ -324,6 +412,7 @@ static struct crypto_alg blk_cbc_alg = {
324 }, 412 },
325}; 413};
326 414
415#ifdef CONFIG_X86_64
327static void ctr_crypt_final(struct crypto_aes_ctx *ctx, 416static void ctr_crypt_final(struct crypto_aes_ctx *ctx,
328 struct blkcipher_walk *walk) 417 struct blkcipher_walk *walk)
329{ 418{
@@ -389,6 +478,7 @@ static struct crypto_alg blk_ctr_alg = {
389 }, 478 },
390 }, 479 },
391}; 480};
481#endif
392 482
393static int ablk_set_key(struct crypto_ablkcipher *tfm, const u8 *key, 483static int ablk_set_key(struct crypto_ablkcipher *tfm, const u8 *key,
394 unsigned int key_len) 484 unsigned int key_len)
@@ -536,6 +626,7 @@ static struct crypto_alg ablk_cbc_alg = {
536 }, 626 },
537}; 627};
538 628
629#ifdef CONFIG_X86_64
539static int ablk_ctr_init(struct crypto_tfm *tfm) 630static int ablk_ctr_init(struct crypto_tfm *tfm)
540{ 631{
541 struct cryptd_ablkcipher *cryptd_tfm; 632 struct cryptd_ablkcipher *cryptd_tfm;
@@ -612,6 +703,7 @@ static struct crypto_alg ablk_rfc3686_ctr_alg = {
612 }, 703 },
613}; 704};
614#endif 705#endif
706#endif
615 707
616#ifdef HAS_LRW 708#ifdef HAS_LRW
617static int ablk_lrw_init(struct crypto_tfm *tfm) 709static int ablk_lrw_init(struct crypto_tfm *tfm)
@@ -730,6 +822,424 @@ static struct crypto_alg ablk_xts_alg = {
730}; 822};
731#endif 823#endif
732 824
825#ifdef CONFIG_X86_64
826static int rfc4106_init(struct crypto_tfm *tfm)
827{
828 struct cryptd_aead *cryptd_tfm;
829 struct aesni_rfc4106_gcm_ctx *ctx = (struct aesni_rfc4106_gcm_ctx *)
830 PTR_ALIGN((u8 *)crypto_tfm_ctx(tfm), AESNI_ALIGN);
831 cryptd_tfm = cryptd_alloc_aead("__driver-gcm-aes-aesni", 0, 0);
832 if (IS_ERR(cryptd_tfm))
833 return PTR_ERR(cryptd_tfm);
834 ctx->cryptd_tfm = cryptd_tfm;
835 tfm->crt_aead.reqsize = sizeof(struct aead_request)
836 + crypto_aead_reqsize(&cryptd_tfm->base);
837 return 0;
838}
839
840static void rfc4106_exit(struct crypto_tfm *tfm)
841{
842 struct aesni_rfc4106_gcm_ctx *ctx =
843 (struct aesni_rfc4106_gcm_ctx *)
844 PTR_ALIGN((u8 *)crypto_tfm_ctx(tfm), AESNI_ALIGN);
845 if (!IS_ERR(ctx->cryptd_tfm))
846 cryptd_free_aead(ctx->cryptd_tfm);
847 return;
848}
849
850static void
851rfc4106_set_hash_subkey_done(struct crypto_async_request *req, int err)
852{
853 struct aesni_gcm_set_hash_subkey_result *result = req->data;
854
855 if (err == -EINPROGRESS)
856 return;
857 result->err = err;
858 complete(&result->completion);
859}
860
861static int
862rfc4106_set_hash_subkey(u8 *hash_subkey, const u8 *key, unsigned int key_len)
863{
864 struct crypto_ablkcipher *ctr_tfm;
865 struct ablkcipher_request *req;
866 int ret = -EINVAL;
867 struct aesni_hash_subkey_req_data *req_data;
868
869 ctr_tfm = crypto_alloc_ablkcipher("ctr(aes)", 0, 0);
870 if (IS_ERR(ctr_tfm))
871 return PTR_ERR(ctr_tfm);
872
873 crypto_ablkcipher_clear_flags(ctr_tfm, ~0);
874
875 ret = crypto_ablkcipher_setkey(ctr_tfm, key, key_len);
876 if (ret) {
877 crypto_free_ablkcipher(ctr_tfm);
878 return ret;
879 }
880
881 req = ablkcipher_request_alloc(ctr_tfm, GFP_KERNEL);
882 if (!req) {
883 crypto_free_ablkcipher(ctr_tfm);
884 return -EINVAL;
885 }
886
887 req_data = kmalloc(sizeof(*req_data), GFP_KERNEL);
888 if (!req_data) {
889 crypto_free_ablkcipher(ctr_tfm);
890 return -ENOMEM;
891 }
892 memset(req_data->iv, 0, sizeof(req_data->iv));
893
894 /* Clear the data in the hash sub key container to zero.*/
895 /* We want to cipher all zeros to create the hash sub key. */
896 memset(hash_subkey, 0, RFC4106_HASH_SUBKEY_SIZE);
897
898 init_completion(&req_data->result.completion);
899 sg_init_one(&req_data->sg, hash_subkey, RFC4106_HASH_SUBKEY_SIZE);
900 ablkcipher_request_set_tfm(req, ctr_tfm);
901 ablkcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_SLEEP |
902 CRYPTO_TFM_REQ_MAY_BACKLOG,
903 rfc4106_set_hash_subkey_done,
904 &req_data->result);
905
906 ablkcipher_request_set_crypt(req, &req_data->sg,
907 &req_data->sg, RFC4106_HASH_SUBKEY_SIZE, req_data->iv);
908
909 ret = crypto_ablkcipher_encrypt(req);
910 if (ret == -EINPROGRESS || ret == -EBUSY) {
911 ret = wait_for_completion_interruptible
912 (&req_data->result.completion);
913 if (!ret)
914 ret = req_data->result.err;
915 }
916 ablkcipher_request_free(req);
917 kfree(req_data);
918 crypto_free_ablkcipher(ctr_tfm);
919 return ret;
920}
921
922static int rfc4106_set_key(struct crypto_aead *parent, const u8 *key,
923 unsigned int key_len)
924{
925 int ret = 0;
926 struct crypto_tfm *tfm = crypto_aead_tfm(parent);
927 struct aesni_rfc4106_gcm_ctx *ctx = aesni_rfc4106_gcm_ctx_get(parent);
928 u8 *new_key_mem = NULL;
929
930 if (key_len < 4) {
931 crypto_tfm_set_flags(tfm, CRYPTO_TFM_RES_BAD_KEY_LEN);
932 return -EINVAL;
933 }
934 /*Account for 4 byte nonce at the end.*/
935 key_len -= 4;
936 if (key_len != AES_KEYSIZE_128) {
937 crypto_tfm_set_flags(tfm, CRYPTO_TFM_RES_BAD_KEY_LEN);
938 return -EINVAL;
939 }
940
941 memcpy(ctx->nonce, key + key_len, sizeof(ctx->nonce));
942 /*This must be on a 16 byte boundary!*/
943 if ((unsigned long)(&(ctx->aes_key_expanded.key_enc[0])) % AESNI_ALIGN)
944 return -EINVAL;
945
946 if ((unsigned long)key % AESNI_ALIGN) {
947 /*key is not aligned: use an auxuliar aligned pointer*/
948 new_key_mem = kmalloc(key_len+AESNI_ALIGN, GFP_KERNEL);
949 if (!new_key_mem)
950 return -ENOMEM;
951
952 new_key_mem = PTR_ALIGN(new_key_mem, AESNI_ALIGN);
953 memcpy(new_key_mem, key, key_len);
954 key = new_key_mem;
955 }
956
957 if (!irq_fpu_usable())
958 ret = crypto_aes_expand_key(&(ctx->aes_key_expanded),
959 key, key_len);
960 else {
961 kernel_fpu_begin();
962 ret = aesni_set_key(&(ctx->aes_key_expanded), key, key_len);
963 kernel_fpu_end();
964 }
965 /*This must be on a 16 byte boundary!*/
966 if ((unsigned long)(&(ctx->hash_subkey[0])) % AESNI_ALIGN) {
967 ret = -EINVAL;
968 goto exit;
969 }
970 ret = rfc4106_set_hash_subkey(ctx->hash_subkey, key, key_len);
971exit:
972 kfree(new_key_mem);
973 return ret;
974}
975
976/* This is the Integrity Check Value (aka the authentication tag length and can
977 * be 8, 12 or 16 bytes long. */
978static int rfc4106_set_authsize(struct crypto_aead *parent,
979 unsigned int authsize)
980{
981 struct aesni_rfc4106_gcm_ctx *ctx = aesni_rfc4106_gcm_ctx_get(parent);
982 struct crypto_aead *cryptd_child = cryptd_aead_child(ctx->cryptd_tfm);
983
984 switch (authsize) {
985 case 8:
986 case 12:
987 case 16:
988 break;
989 default:
990 return -EINVAL;
991 }
992 crypto_aead_crt(parent)->authsize = authsize;
993 crypto_aead_crt(cryptd_child)->authsize = authsize;
994 return 0;
995}
996
997static int rfc4106_encrypt(struct aead_request *req)
998{
999 int ret;
1000 struct crypto_aead *tfm = crypto_aead_reqtfm(req);
1001 struct aesni_rfc4106_gcm_ctx *ctx = aesni_rfc4106_gcm_ctx_get(tfm);
1002 struct crypto_aead *cryptd_child = cryptd_aead_child(ctx->cryptd_tfm);
1003
1004 if (!irq_fpu_usable()) {
1005 struct aead_request *cryptd_req =
1006 (struct aead_request *) aead_request_ctx(req);
1007 memcpy(cryptd_req, req, sizeof(*req));
1008 aead_request_set_tfm(cryptd_req, &ctx->cryptd_tfm->base);
1009 return crypto_aead_encrypt(cryptd_req);
1010 } else {
1011 kernel_fpu_begin();
1012 ret = cryptd_child->base.crt_aead.encrypt(req);
1013 kernel_fpu_end();
1014 return ret;
1015 }
1016}
1017
1018static int rfc4106_decrypt(struct aead_request *req)
1019{
1020 int ret;
1021 struct crypto_aead *tfm = crypto_aead_reqtfm(req);
1022 struct aesni_rfc4106_gcm_ctx *ctx = aesni_rfc4106_gcm_ctx_get(tfm);
1023 struct crypto_aead *cryptd_child = cryptd_aead_child(ctx->cryptd_tfm);
1024
1025 if (!irq_fpu_usable()) {
1026 struct aead_request *cryptd_req =
1027 (struct aead_request *) aead_request_ctx(req);
1028 memcpy(cryptd_req, req, sizeof(*req));
1029 aead_request_set_tfm(cryptd_req, &ctx->cryptd_tfm->base);
1030 return crypto_aead_decrypt(cryptd_req);
1031 } else {
1032 kernel_fpu_begin();
1033 ret = cryptd_child->base.crt_aead.decrypt(req);
1034 kernel_fpu_end();
1035 return ret;
1036 }
1037}
1038
1039static struct crypto_alg rfc4106_alg = {
1040 .cra_name = "rfc4106(gcm(aes))",
1041 .cra_driver_name = "rfc4106-gcm-aesni",
1042 .cra_priority = 400,
1043 .cra_flags = CRYPTO_ALG_TYPE_AEAD | CRYPTO_ALG_ASYNC,
1044 .cra_blocksize = 1,
1045 .cra_ctxsize = sizeof(struct aesni_rfc4106_gcm_ctx) + AESNI_ALIGN,
1046 .cra_alignmask = 0,
1047 .cra_type = &crypto_nivaead_type,
1048 .cra_module = THIS_MODULE,
1049 .cra_list = LIST_HEAD_INIT(rfc4106_alg.cra_list),
1050 .cra_init = rfc4106_init,
1051 .cra_exit = rfc4106_exit,
1052 .cra_u = {
1053 .aead = {
1054 .setkey = rfc4106_set_key,
1055 .setauthsize = rfc4106_set_authsize,
1056 .encrypt = rfc4106_encrypt,
1057 .decrypt = rfc4106_decrypt,
1058 .geniv = "seqiv",
1059 .ivsize = 8,
1060 .maxauthsize = 16,
1061 },
1062 },
1063};
1064
1065static int __driver_rfc4106_encrypt(struct aead_request *req)
1066{
1067 u8 one_entry_in_sg = 0;
1068 u8 *src, *dst, *assoc;
1069 __be32 counter = cpu_to_be32(1);
1070 struct crypto_aead *tfm = crypto_aead_reqtfm(req);
1071 struct aesni_rfc4106_gcm_ctx *ctx = aesni_rfc4106_gcm_ctx_get(tfm);
1072 void *aes_ctx = &(ctx->aes_key_expanded);
1073 unsigned long auth_tag_len = crypto_aead_authsize(tfm);
1074 u8 iv_tab[16+AESNI_ALIGN];
1075 u8* iv = (u8 *) PTR_ALIGN((u8 *)iv_tab, AESNI_ALIGN);
1076 struct scatter_walk src_sg_walk;
1077 struct scatter_walk assoc_sg_walk;
1078 struct scatter_walk dst_sg_walk;
1079 unsigned int i;
1080
1081 /* Assuming we are supporting rfc4106 64-bit extended */
1082 /* sequence numbers We need to have the AAD length equal */
1083 /* to 8 or 12 bytes */
1084 if (unlikely(req->assoclen != 8 && req->assoclen != 12))
1085 return -EINVAL;
1086 /* IV below built */
1087 for (i = 0; i < 4; i++)
1088 *(iv+i) = ctx->nonce[i];
1089 for (i = 0; i < 8; i++)
1090 *(iv+4+i) = req->iv[i];
1091 *((__be32 *)(iv+12)) = counter;
1092
1093 if ((sg_is_last(req->src)) && (sg_is_last(req->assoc))) {
1094 one_entry_in_sg = 1;
1095 scatterwalk_start(&src_sg_walk, req->src);
1096 scatterwalk_start(&assoc_sg_walk, req->assoc);
1097 src = scatterwalk_map(&src_sg_walk, 0);
1098 assoc = scatterwalk_map(&assoc_sg_walk, 0);
1099 dst = src;
1100 if (unlikely(req->src != req->dst)) {
1101 scatterwalk_start(&dst_sg_walk, req->dst);
1102 dst = scatterwalk_map(&dst_sg_walk, 0);
1103 }
1104
1105 } else {
1106 /* Allocate memory for src, dst, assoc */
1107 src = kmalloc(req->cryptlen + auth_tag_len + req->assoclen,
1108 GFP_ATOMIC);
1109 if (unlikely(!src))
1110 return -ENOMEM;
1111 assoc = (src + req->cryptlen + auth_tag_len);
1112 scatterwalk_map_and_copy(src, req->src, 0, req->cryptlen, 0);
1113 scatterwalk_map_and_copy(assoc, req->assoc, 0,
1114 req->assoclen, 0);
1115 dst = src;
1116 }
1117
1118 aesni_gcm_enc(aes_ctx, dst, src, (unsigned long)req->cryptlen, iv,
1119 ctx->hash_subkey, assoc, (unsigned long)req->assoclen, dst
1120 + ((unsigned long)req->cryptlen), auth_tag_len);
1121
1122 /* The authTag (aka the Integrity Check Value) needs to be written
1123 * back to the packet. */
1124 if (one_entry_in_sg) {
1125 if (unlikely(req->src != req->dst)) {
1126 scatterwalk_unmap(dst, 0);
1127 scatterwalk_done(&dst_sg_walk, 0, 0);
1128 }
1129 scatterwalk_unmap(src, 0);
1130 scatterwalk_unmap(assoc, 0);
1131 scatterwalk_done(&src_sg_walk, 0, 0);
1132 scatterwalk_done(&assoc_sg_walk, 0, 0);
1133 } else {
1134 scatterwalk_map_and_copy(dst, req->dst, 0,
1135 req->cryptlen + auth_tag_len, 1);
1136 kfree(src);
1137 }
1138 return 0;
1139}
1140
1141static int __driver_rfc4106_decrypt(struct aead_request *req)
1142{
1143 u8 one_entry_in_sg = 0;
1144 u8 *src, *dst, *assoc;
1145 unsigned long tempCipherLen = 0;
1146 __be32 counter = cpu_to_be32(1);
1147 int retval = 0;
1148 struct crypto_aead *tfm = crypto_aead_reqtfm(req);
1149 struct aesni_rfc4106_gcm_ctx *ctx = aesni_rfc4106_gcm_ctx_get(tfm);
1150 void *aes_ctx = &(ctx->aes_key_expanded);
1151 unsigned long auth_tag_len = crypto_aead_authsize(tfm);
1152 u8 iv_and_authTag[32+AESNI_ALIGN];
1153 u8 *iv = (u8 *) PTR_ALIGN((u8 *)iv_and_authTag, AESNI_ALIGN);
1154 u8 *authTag = iv + 16;
1155 struct scatter_walk src_sg_walk;
1156 struct scatter_walk assoc_sg_walk;
1157 struct scatter_walk dst_sg_walk;
1158 unsigned int i;
1159
1160 if (unlikely((req->cryptlen < auth_tag_len) ||
1161 (req->assoclen != 8 && req->assoclen != 12)))
1162 return -EINVAL;
1163 /* Assuming we are supporting rfc4106 64-bit extended */
1164 /* sequence numbers We need to have the AAD length */
1165 /* equal to 8 or 12 bytes */
1166
1167 tempCipherLen = (unsigned long)(req->cryptlen - auth_tag_len);
1168 /* IV below built */
1169 for (i = 0; i < 4; i++)
1170 *(iv+i) = ctx->nonce[i];
1171 for (i = 0; i < 8; i++)
1172 *(iv+4+i) = req->iv[i];
1173 *((__be32 *)(iv+12)) = counter;
1174
1175 if ((sg_is_last(req->src)) && (sg_is_last(req->assoc))) {
1176 one_entry_in_sg = 1;
1177 scatterwalk_start(&src_sg_walk, req->src);
1178 scatterwalk_start(&assoc_sg_walk, req->assoc);
1179 src = scatterwalk_map(&src_sg_walk, 0);
1180 assoc = scatterwalk_map(&assoc_sg_walk, 0);
1181 dst = src;
1182 if (unlikely(req->src != req->dst)) {
1183 scatterwalk_start(&dst_sg_walk, req->dst);
1184 dst = scatterwalk_map(&dst_sg_walk, 0);
1185 }
1186
1187 } else {
1188 /* Allocate memory for src, dst, assoc */
1189 src = kmalloc(req->cryptlen + req->assoclen, GFP_ATOMIC);
1190 if (!src)
1191 return -ENOMEM;
1192 assoc = (src + req->cryptlen + auth_tag_len);
1193 scatterwalk_map_and_copy(src, req->src, 0, req->cryptlen, 0);
1194 scatterwalk_map_and_copy(assoc, req->assoc, 0,
1195 req->assoclen, 0);
1196 dst = src;
1197 }
1198
1199 aesni_gcm_dec(aes_ctx, dst, src, tempCipherLen, iv,
1200 ctx->hash_subkey, assoc, (unsigned long)req->assoclen,
1201 authTag, auth_tag_len);
1202
1203 /* Compare generated tag with passed in tag. */
1204 retval = memcmp(src + tempCipherLen, authTag, auth_tag_len) ?
1205 -EBADMSG : 0;
1206
1207 if (one_entry_in_sg) {
1208 if (unlikely(req->src != req->dst)) {
1209 scatterwalk_unmap(dst, 0);
1210 scatterwalk_done(&dst_sg_walk, 0, 0);
1211 }
1212 scatterwalk_unmap(src, 0);
1213 scatterwalk_unmap(assoc, 0);
1214 scatterwalk_done(&src_sg_walk, 0, 0);
1215 scatterwalk_done(&assoc_sg_walk, 0, 0);
1216 } else {
1217 scatterwalk_map_and_copy(dst, req->dst, 0, req->cryptlen, 1);
1218 kfree(src);
1219 }
1220 return retval;
1221}
1222
1223static struct crypto_alg __rfc4106_alg = {
1224 .cra_name = "__gcm-aes-aesni",
1225 .cra_driver_name = "__driver-gcm-aes-aesni",
1226 .cra_priority = 0,
1227 .cra_flags = CRYPTO_ALG_TYPE_AEAD,
1228 .cra_blocksize = 1,
1229 .cra_ctxsize = sizeof(struct aesni_rfc4106_gcm_ctx) + AESNI_ALIGN,
1230 .cra_alignmask = 0,
1231 .cra_type = &crypto_aead_type,
1232 .cra_module = THIS_MODULE,
1233 .cra_list = LIST_HEAD_INIT(__rfc4106_alg.cra_list),
1234 .cra_u = {
1235 .aead = {
1236 .encrypt = __driver_rfc4106_encrypt,
1237 .decrypt = __driver_rfc4106_decrypt,
1238 },
1239 },
1240};
1241#endif
1242
733static int __init aesni_init(void) 1243static int __init aesni_init(void)
734{ 1244{
735 int err; 1245 int err;
@@ -738,6 +1248,7 @@ static int __init aesni_init(void)
738 printk(KERN_INFO "Intel AES-NI instructions are not detected.\n"); 1248 printk(KERN_INFO "Intel AES-NI instructions are not detected.\n");
739 return -ENODEV; 1249 return -ENODEV;
740 } 1250 }
1251
741 if ((err = crypto_register_alg(&aesni_alg))) 1252 if ((err = crypto_register_alg(&aesni_alg)))
742 goto aes_err; 1253 goto aes_err;
743 if ((err = crypto_register_alg(&__aesni_alg))) 1254 if ((err = crypto_register_alg(&__aesni_alg)))
@@ -746,18 +1257,24 @@ static int __init aesni_init(void)
746 goto blk_ecb_err; 1257 goto blk_ecb_err;
747 if ((err = crypto_register_alg(&blk_cbc_alg))) 1258 if ((err = crypto_register_alg(&blk_cbc_alg)))
748 goto blk_cbc_err; 1259 goto blk_cbc_err;
749 if ((err = crypto_register_alg(&blk_ctr_alg)))
750 goto blk_ctr_err;
751 if ((err = crypto_register_alg(&ablk_ecb_alg))) 1260 if ((err = crypto_register_alg(&ablk_ecb_alg)))
752 goto ablk_ecb_err; 1261 goto ablk_ecb_err;
753 if ((err = crypto_register_alg(&ablk_cbc_alg))) 1262 if ((err = crypto_register_alg(&ablk_cbc_alg)))
754 goto ablk_cbc_err; 1263 goto ablk_cbc_err;
1264#ifdef CONFIG_X86_64
1265 if ((err = crypto_register_alg(&blk_ctr_alg)))
1266 goto blk_ctr_err;
755 if ((err = crypto_register_alg(&ablk_ctr_alg))) 1267 if ((err = crypto_register_alg(&ablk_ctr_alg)))
756 goto ablk_ctr_err; 1268 goto ablk_ctr_err;
1269 if ((err = crypto_register_alg(&__rfc4106_alg)))
1270 goto __aead_gcm_err;
1271 if ((err = crypto_register_alg(&rfc4106_alg)))
1272 goto aead_gcm_err;
757#ifdef HAS_CTR 1273#ifdef HAS_CTR
758 if ((err = crypto_register_alg(&ablk_rfc3686_ctr_alg))) 1274 if ((err = crypto_register_alg(&ablk_rfc3686_ctr_alg)))
759 goto ablk_rfc3686_ctr_err; 1275 goto ablk_rfc3686_ctr_err;
760#endif 1276#endif
1277#endif
761#ifdef HAS_LRW 1278#ifdef HAS_LRW
762 if ((err = crypto_register_alg(&ablk_lrw_alg))) 1279 if ((err = crypto_register_alg(&ablk_lrw_alg)))
763 goto ablk_lrw_err; 1280 goto ablk_lrw_err;
@@ -770,7 +1287,6 @@ static int __init aesni_init(void)
770 if ((err = crypto_register_alg(&ablk_xts_alg))) 1287 if ((err = crypto_register_alg(&ablk_xts_alg)))
771 goto ablk_xts_err; 1288 goto ablk_xts_err;
772#endif 1289#endif
773
774 return err; 1290 return err;
775 1291
776#ifdef HAS_XTS 1292#ifdef HAS_XTS
@@ -784,18 +1300,24 @@ ablk_pcbc_err:
784 crypto_unregister_alg(&ablk_lrw_alg); 1300 crypto_unregister_alg(&ablk_lrw_alg);
785ablk_lrw_err: 1301ablk_lrw_err:
786#endif 1302#endif
1303#ifdef CONFIG_X86_64
787#ifdef HAS_CTR 1304#ifdef HAS_CTR
788 crypto_unregister_alg(&ablk_rfc3686_ctr_alg); 1305 crypto_unregister_alg(&ablk_rfc3686_ctr_alg);
789ablk_rfc3686_ctr_err: 1306ablk_rfc3686_ctr_err:
790#endif 1307#endif
1308 crypto_unregister_alg(&rfc4106_alg);
1309aead_gcm_err:
1310 crypto_unregister_alg(&__rfc4106_alg);
1311__aead_gcm_err:
791 crypto_unregister_alg(&ablk_ctr_alg); 1312 crypto_unregister_alg(&ablk_ctr_alg);
792ablk_ctr_err: 1313ablk_ctr_err:
1314 crypto_unregister_alg(&blk_ctr_alg);
1315blk_ctr_err:
1316#endif
793 crypto_unregister_alg(&ablk_cbc_alg); 1317 crypto_unregister_alg(&ablk_cbc_alg);
794ablk_cbc_err: 1318ablk_cbc_err:
795 crypto_unregister_alg(&ablk_ecb_alg); 1319 crypto_unregister_alg(&ablk_ecb_alg);
796ablk_ecb_err: 1320ablk_ecb_err:
797 crypto_unregister_alg(&blk_ctr_alg);
798blk_ctr_err:
799 crypto_unregister_alg(&blk_cbc_alg); 1321 crypto_unregister_alg(&blk_cbc_alg);
800blk_cbc_err: 1322blk_cbc_err:
801 crypto_unregister_alg(&blk_ecb_alg); 1323 crypto_unregister_alg(&blk_ecb_alg);
@@ -818,13 +1340,17 @@ static void __exit aesni_exit(void)
818#ifdef HAS_LRW 1340#ifdef HAS_LRW
819 crypto_unregister_alg(&ablk_lrw_alg); 1341 crypto_unregister_alg(&ablk_lrw_alg);
820#endif 1342#endif
1343#ifdef CONFIG_X86_64
821#ifdef HAS_CTR 1344#ifdef HAS_CTR
822 crypto_unregister_alg(&ablk_rfc3686_ctr_alg); 1345 crypto_unregister_alg(&ablk_rfc3686_ctr_alg);
823#endif 1346#endif
1347 crypto_unregister_alg(&rfc4106_alg);
1348 crypto_unregister_alg(&__rfc4106_alg);
824 crypto_unregister_alg(&ablk_ctr_alg); 1349 crypto_unregister_alg(&ablk_ctr_alg);
1350 crypto_unregister_alg(&blk_ctr_alg);
1351#endif
825 crypto_unregister_alg(&ablk_cbc_alg); 1352 crypto_unregister_alg(&ablk_cbc_alg);
826 crypto_unregister_alg(&ablk_ecb_alg); 1353 crypto_unregister_alg(&ablk_ecb_alg);
827 crypto_unregister_alg(&blk_ctr_alg);
828 crypto_unregister_alg(&blk_cbc_alg); 1354 crypto_unregister_alg(&blk_cbc_alg);
829 crypto_unregister_alg(&blk_ecb_alg); 1355 crypto_unregister_alg(&blk_ecb_alg);
830 crypto_unregister_alg(&__aesni_alg); 1356 crypto_unregister_alg(&__aesni_alg);
generated by cgit v1.2.2 (git 2.25.0) at 2024-10-22 07:29:17 -0400