Home | History | Annotate | Line # | Download | only in x86
aes_via.c revision 1.1
      1 /*	$NetBSD: aes_via.c,v 1.1 2020/06/29 23:39:30 riastradh Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2020 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  *
     16  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     17  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     18  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     19  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     20  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     21  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     22  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     23  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     24  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     25  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     26  * POSSIBILITY OF SUCH DAMAGE.
     27  */
     28 
     29 #include <sys/cdefs.h>
     30 __KERNEL_RCSID(1, "$NetBSD: aes_via.c,v 1.1 2020/06/29 23:39:30 riastradh Exp $");
     31 
     32 #include <sys/types.h>
     33 #include <sys/evcnt.h>
     34 #include <sys/systm.h>
     35 
     36 #include <crypto/aes/aes.h>
     37 #include <crypto/aes/aes_bear.h>
     38 
     39 #include <x86/cpufunc.h>
     40 #include <x86/cpuvar.h>
     41 #include <x86/fpu.h>
     42 #include <x86/specialreg.h>
     43 #include <x86/via_padlock.h>
     44 
     45 static void
     46 aesvia_reload_keys(void)
     47 {
     48 
     49 	asm volatile("pushf; popf");
     50 }
     51 
     52 static uint32_t
     53 aesvia_keylen_cw0(unsigned nrounds)
     54 {
     55 
     56 	/*
     57 	 * Determine the control word bits for the key size / number of
     58 	 * rounds.  For AES-128, the hardware can do key expansion on
     59 	 * the fly; for AES-192 and AES-256, software must do it.
     60 	 */
     61 	switch (nrounds) {
     62 	case AES_128_NROUNDS:
     63 		return C3_CRYPT_CWLO_KEY128;
     64 	case AES_192_NROUNDS:
     65 		return C3_CRYPT_CWLO_KEY192 | C3_CRYPT_CWLO_KEYGEN_SW;
     66 	case AES_256_NROUNDS:
     67 		return C3_CRYPT_CWLO_KEY256 | C3_CRYPT_CWLO_KEYGEN_SW;
     68 	default:
     69 		panic("invalid AES nrounds: %u", nrounds);
     70 	}
     71 }
     72 
     73 static void
     74 aesvia_setenckey(struct aesenc *enc, const uint8_t *key, uint32_t nrounds)
     75 {
     76 	size_t key_len;
     77 
     78 	switch (nrounds) {
     79 	case AES_128_NROUNDS:
     80 		enc->aese_aes.aes_rk[0] = le32dec(key + 4*0);
     81 		enc->aese_aes.aes_rk[1] = le32dec(key + 4*1);
     82 		enc->aese_aes.aes_rk[2] = le32dec(key + 4*2);
     83 		enc->aese_aes.aes_rk[3] = le32dec(key + 4*3);
     84 		return;
     85 	case AES_192_NROUNDS:
     86 		key_len = 24;
     87 		break;
     88 	case AES_256_NROUNDS:
     89 		key_len = 32;
     90 		break;
     91 	default:
     92 		panic("invalid AES nrounds: %u", nrounds);
     93 	}
     94 	br_aes_ct_keysched_stdenc(enc->aese_aes.aes_rk, key, key_len);
     95 }
     96 
     97 static void
     98 aesvia_setdeckey(struct aesdec *dec, const uint8_t *key, uint32_t nrounds)
     99 {
    100 	size_t key_len;
    101 
    102 	switch (nrounds) {
    103 	case AES_128_NROUNDS:
    104 		dec->aesd_aes.aes_rk[0] = le32dec(key + 4*0);
    105 		dec->aesd_aes.aes_rk[1] = le32dec(key + 4*1);
    106 		dec->aesd_aes.aes_rk[2] = le32dec(key + 4*2);
    107 		dec->aesd_aes.aes_rk[3] = le32dec(key + 4*3);
    108 		return;
    109 	case AES_192_NROUNDS:
    110 		key_len = 24;
    111 		break;
    112 	case AES_256_NROUNDS:
    113 		key_len = 32;
    114 		break;
    115 	default:
    116 		panic("invalid AES nrounds: %u", nrounds);
    117 	}
    118 	br_aes_ct_keysched_stddec(dec->aesd_aes.aes_rk, key, key_len);
    119 }
    120 
    121 static inline void
    122 aesvia_enc1(const struct aesenc *enc, const uint8_t in[static 16],
    123     uint8_t out[static 16], uint32_t cw0)
    124 {
    125 	const uint32_t cw[4] __aligned(16) = {
    126 		[0] = (cw0
    127 		    | C3_CRYPT_CWLO_ALG_AES
    128 		    | C3_CRYPT_CWLO_ENCRYPT
    129 		    | C3_CRYPT_CWLO_NORMAL),
    130 	};
    131 	size_t nblocks = 1;
    132 
    133 	KASSERT(((uintptr_t)enc & 0xf) == 0);
    134 	KASSERT(((uintptr_t)in & 0xf) == 0);
    135 	KASSERT(((uintptr_t)out & 0xf) == 0);
    136 
    137 	asm volatile("rep xcryptecb"
    138 	    : "+c"(nblocks), "+S"(in), "+D"(out)
    139 	    : "b"(enc), "d"(cw)
    140 	    : "memory", "cc");
    141 }
    142 
    143 static inline void
    144 aesvia_dec1(const struct aesdec *dec, const uint8_t in[static 16],
    145     uint8_t out[static 16], uint32_t cw0)
    146 {
    147 	const uint32_t cw[4] __aligned(16) = {
    148 		[0] = (cw0
    149 		    | C3_CRYPT_CWLO_ALG_AES
    150 		    | C3_CRYPT_CWLO_DECRYPT
    151 		    | C3_CRYPT_CWLO_NORMAL),
    152 	};
    153 	size_t nblocks = 1;
    154 
    155 	KASSERT(((uintptr_t)dec & 0xf) == 0);
    156 	KASSERT(((uintptr_t)in & 0xf) == 0);
    157 	KASSERT(((uintptr_t)out & 0xf) == 0);
    158 
    159 	asm volatile("rep xcryptecb"
    160 	    : "+c"(nblocks), "+S"(in), "+D"(out)
    161 	    : "b"(dec), "d"(cw)
    162 	    : "memory", "cc");
    163 }
    164 
    165 static struct evcnt enc_aligned_evcnt = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    166     NULL, "aesvia", "enc aligned");
    167 EVCNT_ATTACH_STATIC(enc_aligned_evcnt);
    168 static struct evcnt enc_unaligned_evcnt = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    169     NULL, "aesvia", "dec unaligned");
    170 EVCNT_ATTACH_STATIC(enc_unaligned_evcnt);
    171 
    172 static void
    173 aesvia_enc(const struct aesenc *enc, const uint8_t in[static 16],
    174     uint8_t out[static 16], uint32_t nrounds)
    175 {
    176 	const uint32_t cw0 = aesvia_keylen_cw0(nrounds);
    177 
    178 	fpu_kern_enter();
    179 	aesvia_reload_keys();
    180 	if ((((uintptr_t)in | (uintptr_t)out) & 0xf) == 0 &&
    181 	    ((uintptr_t)in & 0xff0) != 0xff0) {
    182 		enc_aligned_evcnt.ev_count++;
    183 		aesvia_enc1(enc, in, out, cw0);
    184 	} else {
    185 		enc_unaligned_evcnt.ev_count++;
    186 		/*
    187 		 * VIA requires 16-byte/128-bit alignment, and
    188 		 * xcrypt-ecb reads one block past the one we're
    189 		 * working on -- which may go past the end of the page
    190 		 * into unmapped territory.  Use a bounce buffer if
    191 		 * either constraint is violated.
    192 		 */
    193 		uint8_t inbuf[16] __aligned(16);
    194 		uint8_t outbuf[16] __aligned(16);
    195 
    196 		memcpy(inbuf, in, 16);
    197 		aesvia_enc1(enc, inbuf, outbuf, cw0);
    198 		memcpy(out, outbuf, 16);
    199 
    200 		explicit_memset(inbuf, 0, sizeof inbuf);
    201 		explicit_memset(outbuf, 0, sizeof outbuf);
    202 	}
    203 	fpu_kern_leave();
    204 }
    205 
    206 static struct evcnt dec_aligned_evcnt = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    207     NULL, "aesvia", "dec aligned");
    208 EVCNT_ATTACH_STATIC(dec_aligned_evcnt);
    209 static struct evcnt dec_unaligned_evcnt = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    210     NULL, "aesvia", "dec unaligned");
    211 EVCNT_ATTACH_STATIC(dec_unaligned_evcnt);
    212 
    213 static void
    214 aesvia_dec(const struct aesdec *dec, const uint8_t in[static 16],
    215     uint8_t out[static 16], uint32_t nrounds)
    216 {
    217 	const uint32_t cw0 = aesvia_keylen_cw0(nrounds);
    218 
    219 	fpu_kern_enter();
    220 	aesvia_reload_keys();
    221 	if ((((uintptr_t)in | (uintptr_t)out) & 0xf) == 0 &&
    222 	    ((uintptr_t)in & 0xff0) != 0xff0) {
    223 		dec_aligned_evcnt.ev_count++;
    224 		aesvia_dec1(dec, in, out, cw0);
    225 	} else {
    226 		dec_unaligned_evcnt.ev_count++;
    227 		/*
    228 		 * VIA requires 16-byte/128-bit alignment, and
    229 		 * xcrypt-ecb reads one block past the one we're
    230 		 * working on -- which may go past the end of the page
    231 		 * into unmapped territory.  Use a bounce buffer if
    232 		 * either constraint is violated.
    233 		 */
    234 		uint8_t inbuf[16] __aligned(16);
    235 		uint8_t outbuf[16] __aligned(16);
    236 
    237 		memcpy(inbuf, in, 16);
    238 		aesvia_dec1(dec, inbuf, outbuf, cw0);
    239 		memcpy(out, outbuf, 16);
    240 
    241 		explicit_memset(inbuf, 0, sizeof inbuf);
    242 		explicit_memset(outbuf, 0, sizeof outbuf);
    243 	}
    244 	fpu_kern_leave();
    245 }
    246 
    247 static inline void
    248 aesvia_cbc_enc1(const struct aesenc *enc, const uint8_t in[static 16],
    249     uint8_t out[static 16], size_t nblocks, uint8_t **ivp, uint32_t cw0)
    250 {
    251 	const uint32_t cw[4] __aligned(16) = {
    252 		[0] = (cw0
    253 		    | C3_CRYPT_CWLO_ALG_AES
    254 		    | C3_CRYPT_CWLO_ENCRYPT
    255 		    | C3_CRYPT_CWLO_NORMAL),
    256 	};
    257 
    258 	KASSERT(((uintptr_t)enc & 0xf) == 0);
    259 	KASSERT(((uintptr_t)in & 0xf) == 0);
    260 	KASSERT(((uintptr_t)out & 0xf) == 0);
    261 	KASSERT(((uintptr_t)*ivp & 0xf) == 0);
    262 
    263 	/*
    264 	 * Register effects:
    265 	 * - Counts nblocks down to zero.
    266 	 * - Advances in by nblocks (units of blocks).
    267 	 * - Advances out by nblocks (units of blocks).
    268 	 * - Updates *ivp to point at the last block of out.
    269 	 */
    270 	asm volatile("rep xcryptcbc"
    271 	    : "+c"(nblocks), "+S"(in), "+D"(out), "+a"(*ivp)
    272 	    : "b"(enc), "d"(cw)
    273 	    : "memory", "cc");
    274 }
    275 
    276 static inline void
    277 aesvia_cbc_dec1(const struct aesdec *dec, const uint8_t in[static 16],
    278     uint8_t out[static 16], size_t nblocks, uint8_t iv[static 16],
    279     uint32_t cw0)
    280 {
    281 	const uint32_t cw[4] __aligned(16) = {
    282 		[0] = (cw0
    283 		    | C3_CRYPT_CWLO_ALG_AES
    284 		    | C3_CRYPT_CWLO_DECRYPT
    285 		    | C3_CRYPT_CWLO_NORMAL),
    286 	};
    287 
    288 	KASSERT(((uintptr_t)dec & 0xf) == 0);
    289 	KASSERT(((uintptr_t)in & 0xf) == 0);
    290 	KASSERT(((uintptr_t)out & 0xf) == 0);
    291 	KASSERT(((uintptr_t)iv & 0xf) == 0);
    292 
    293 	/*
    294 	 * Register effects:
    295 	 * - Counts nblocks down to zero.
    296 	 * - Advances in by nblocks (units of blocks).
    297 	 * - Advances out by nblocks (units of blocks).
    298 	 * Memory side effects:
    299 	 * - Writes what was the last block of in at the address iv.
    300 	 */
    301 	asm volatile("rep xcryptcbc"
    302 	    : "+c"(nblocks), "+S"(in), "+D"(out)
    303 	    : "a"(iv), "b"(dec), "d"(cw)
    304 	    : "memory", "cc");
    305 }
    306 
    307 static inline void
    308 xor128(void *x, const void *a, const void *b)
    309 {
    310 	uint32_t *x32 = x;
    311 	const uint32_t *a32 = a;
    312 	const uint32_t *b32 = b;
    313 
    314 	x32[0] = a32[0] ^ b32[0];
    315 	x32[1] = a32[1] ^ b32[1];
    316 	x32[2] = a32[2] ^ b32[2];
    317 	x32[3] = a32[3] ^ b32[3];
    318 }
    319 
    320 static struct evcnt cbcenc_aligned_evcnt = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    321     NULL, "aesvia", "cbcenc aligned");
    322 EVCNT_ATTACH_STATIC(cbcenc_aligned_evcnt);
    323 static struct evcnt cbcenc_unaligned_evcnt = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    324     NULL, "aesvia", "cbcenc unaligned");
    325 EVCNT_ATTACH_STATIC(cbcenc_unaligned_evcnt);
    326 
    327 static void
    328 aesvia_cbc_enc(const struct aesenc *enc, const uint8_t in[static 16],
    329     uint8_t out[static 16], size_t nbytes, uint8_t iv[static 16],
    330     uint32_t nrounds)
    331 {
    332 	const uint32_t cw0 = aesvia_keylen_cw0(nrounds);
    333 
    334 	KASSERT(nbytes % 16 == 0);
    335 	if (nbytes == 0)
    336 		return;
    337 
    338 	fpu_kern_enter();
    339 	aesvia_reload_keys();
    340 	if ((((uintptr_t)in | (uintptr_t)out | (uintptr_t)iv) & 0xf) == 0) {
    341 		cbcenc_aligned_evcnt.ev_count++;
    342 		uint8_t *ivp = iv;
    343 		aesvia_cbc_enc1(enc, in, out, nbytes/16, &ivp, cw0);
    344 		memcpy(iv, ivp, 16);
    345 	} else {
    346 		cbcenc_unaligned_evcnt.ev_count++;
    347 		uint8_t cv[16] __aligned(16);
    348 		uint8_t tmp[16] __aligned(16);
    349 
    350 		memcpy(cv, iv, 16);
    351 		for (; nbytes; nbytes -= 16, in += 16, out += 16) {
    352 			memcpy(tmp, in, 16);
    353 			xor128(tmp, tmp, cv);
    354 			aesvia_enc1(enc, tmp, cv, cw0);
    355 			memcpy(out, cv, 16);
    356 		}
    357 		memcpy(iv, cv, 16);
    358 	}
    359 	fpu_kern_leave();
    360 }
    361 
    362 static struct evcnt cbcdec_aligned_evcnt = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    363     NULL, "aesvia", "cbcdec aligned");
    364 EVCNT_ATTACH_STATIC(cbcdec_aligned_evcnt);
    365 static struct evcnt cbcdec_unaligned_evcnt = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    366     NULL, "aesvia", "cbcdec unaligned");
    367 EVCNT_ATTACH_STATIC(cbcdec_unaligned_evcnt);
    368 
    369 static void
    370 aesvia_cbc_dec(const struct aesdec *dec, const uint8_t in[static 16],
    371     uint8_t out[static 16], size_t nbytes, uint8_t iv[static 16],
    372     uint32_t nrounds)
    373 {
    374 	const uint32_t cw0 = aesvia_keylen_cw0(nrounds);
    375 
    376 	KASSERT(nbytes % 16 == 0);
    377 	if (nbytes == 0)
    378 		return;
    379 
    380 	fpu_kern_enter();
    381 	aesvia_reload_keys();
    382 	if ((((uintptr_t)in | (uintptr_t)out | (uintptr_t)iv) & 0xf) == 0) {
    383 		cbcdec_aligned_evcnt.ev_count++;
    384 		aesvia_cbc_dec1(dec, in, out, nbytes/16, iv, cw0);
    385 	} else {
    386 		cbcdec_unaligned_evcnt.ev_count++;
    387 		uint8_t iv0[16] __aligned(16);
    388 		uint8_t cv[16] __aligned(16);
    389 		uint8_t tmp[16] __aligned(16);
    390 
    391 		memcpy(iv0, iv, 16);
    392 		memcpy(cv, in + nbytes - 16, 16);
    393 		memcpy(iv, cv, 16);
    394 
    395 		for (;;) {
    396 			aesvia_dec1(dec, cv, tmp, cw0);
    397 			if ((nbytes -= 16) == 0)
    398 				break;
    399 			memcpy(cv, in + nbytes - 16, 16);
    400 			xor128(tmp, tmp, cv);
    401 			memcpy(out + nbytes, tmp, 16);
    402 		}
    403 
    404 		xor128(tmp, tmp, iv0);
    405 		memcpy(out, tmp, 16);
    406 		explicit_memset(tmp, 0, sizeof tmp);
    407 	}
    408 	fpu_kern_leave();
    409 }
    410 
    411 static inline void
    412 aesvia_xts_update(uint32_t *t0, uint32_t *t1, uint32_t *t2, uint32_t *t3)
    413 {
    414 	uint32_t s0, s1, s2, s3;
    415 
    416 	s0 = *t0 >> 31;
    417 	s1 = *t1 >> 31;
    418 	s2 = *t2 >> 31;
    419 	s3 = *t3 >> 31;
    420 	*t0 = (*t0 << 1) ^ (-s3 & 0x87);
    421 	*t1 = (*t1 << 1) ^ s0;
    422 	*t2 = (*t2 << 1) ^ s1;
    423 	*t3 = (*t3 << 1) ^ s2;
    424 }
    425 
    426 static int
    427 aesvia_xts_update_selftest(void)
    428 {
    429 	static const struct {
    430 		uint32_t in[4], out[4];
    431 	} cases[] = {
    432 		{ {1}, {2} },
    433 		{ {0x80000000U,0,0,0}, {0,1,0,0} },
    434 		{ {0,0x80000000U,0,0}, {0,0,1,0} },
    435 		{ {0,0,0x80000000U,0}, {0,0,0,1} },
    436 		{ {0,0,0,0x80000000U}, {0x87,0,0,0} },
    437 		{ {0,0x80000000U,0,0x80000000U}, {0x87,0,1,0} },
    438 	};
    439 	unsigned i;
    440 	uint32_t t0, t1, t2, t3;
    441 
    442 	for (i = 0; i < sizeof(cases)/sizeof(cases[0]); i++) {
    443 		t0 = cases[i].in[0];
    444 		t1 = cases[i].in[1];
    445 		t2 = cases[i].in[2];
    446 		t3 = cases[i].in[3];
    447 		aesvia_xts_update(&t0, &t1, &t2, &t3);
    448 		if (t0 != cases[i].out[0] ||
    449 		    t1 != cases[i].out[1] ||
    450 		    t2 != cases[i].out[2] ||
    451 		    t3 != cases[i].out[3])
    452 			return -1;
    453 	}
    454 
    455 	/* Success!  */
    456 	return 0;
    457 }
    458 
    459 static struct evcnt xtsenc_aligned_evcnt = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    460     NULL, "aesvia", "xtsenc aligned");
    461 EVCNT_ATTACH_STATIC(xtsenc_aligned_evcnt);
    462 static struct evcnt xtsenc_unaligned_evcnt = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    463     NULL, "aesvia", "xtsenc unaligned");
    464 EVCNT_ATTACH_STATIC(xtsenc_unaligned_evcnt);
    465 
    466 static void
    467 aesvia_xts_enc(const struct aesenc *enc, const uint8_t in[static 16],
    468     uint8_t out[static 16], size_t nbytes, uint8_t tweak[static 16],
    469     uint32_t nrounds)
    470 {
    471 	const uint32_t cw0 = aesvia_keylen_cw0(nrounds);
    472 	uint32_t t[4];
    473 
    474 	KASSERT(nbytes % 16 == 0);
    475 
    476 	memcpy(t, tweak, 16);
    477 
    478 	fpu_kern_enter();
    479 	aesvia_reload_keys();
    480 	if ((((uintptr_t)in | (uintptr_t)out) & 0xf) == 0) {
    481 		xtsenc_aligned_evcnt.ev_count++;
    482 		unsigned lastblock = 0;
    483 
    484 		/*
    485 		 * Make sure the last block is not the last block of a
    486 		 * page.  (Note that we store the AES input in `out' as
    487 		 * a temporary buffer, rather than reading it directly
    488 		 * from `in', since we have to combine the tweak
    489 		 * first.)
    490 		 */
    491 		lastblock = 16*(((uintptr_t)(out + nbytes) & 0xfff) == 0);
    492 		nbytes -= lastblock;
    493 
    494 		for (; nbytes; nbytes -= 16, in += 16, out += 16) {
    495 			xor128(out, in, t);
    496 			aesvia_enc1(enc, out, out, cw0);
    497 			xor128(out, out, t);
    498 			aesvia_xts_update(&t[0], &t[1], &t[2], &t[3]);
    499 		}
    500 
    501 		/* Handle the last block of a page, if necessary.  */
    502 		if (lastblock) {
    503 			uint8_t buf[16] __aligned(16);
    504 			xor128(buf, in, t);
    505 			aesvia_enc1(enc, buf, out, cw0);
    506 			explicit_memset(buf, 0, sizeof buf);
    507 		}
    508 	} else {
    509 		xtsenc_unaligned_evcnt.ev_count++;
    510 		uint8_t buf[16] __aligned(16);
    511 
    512 		for (; nbytes; nbytes -= 16, in += 16, out += 16) {
    513 			memcpy(buf, in, 16);
    514 			xor128(buf, buf, t);
    515 			aesvia_enc1(enc, buf, buf, cw0);
    516 			xor128(buf, buf, t);
    517 			memcpy(out, buf, 16);
    518 			aesvia_xts_update(&t[0], &t[1], &t[2], &t[3]);
    519 		}
    520 
    521 		explicit_memset(buf, 0, sizeof buf);
    522 	}
    523 	fpu_kern_leave();
    524 
    525 	memcpy(tweak, t, 16);
    526 	explicit_memset(t, 0, sizeof t);
    527 }
    528 
    529 static struct evcnt xtsdec_aligned_evcnt = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    530     NULL, "aesvia", "xtsdec aligned");
    531 EVCNT_ATTACH_STATIC(xtsdec_aligned_evcnt);
    532 static struct evcnt xtsdec_unaligned_evcnt = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    533     NULL, "aesvia", "xtsdec unaligned");
    534 EVCNT_ATTACH_STATIC(xtsdec_unaligned_evcnt);
    535 
    536 static void
    537 aesvia_xts_dec(const struct aesdec *dec, const uint8_t in[static 16],
    538     uint8_t out[static 16], size_t nbytes, uint8_t tweak[static 16],
    539     uint32_t nrounds)
    540 {
    541 	const uint32_t cw0 = aesvia_keylen_cw0(nrounds);
    542 	uint32_t t[4];
    543 
    544 	KASSERT(nbytes % 16 == 0);
    545 
    546 	memcpy(t, tweak, 16);
    547 
    548 	fpu_kern_enter();
    549 	aesvia_reload_keys();
    550 	if ((((uintptr_t)in | (uintptr_t)out) & 0xf) == 0) {
    551 		xtsdec_aligned_evcnt.ev_count++;
    552 		unsigned lastblock = 0;
    553 
    554 		/*
    555 		 * Make sure the last block is not the last block of a
    556 		 * page.  (Note that we store the AES input in `out' as
    557 		 * a temporary buffer, rather than reading it directly
    558 		 * from `in', since we have to combine the tweak
    559 		 * first.)
    560 		 */
    561 		lastblock = 16*(((uintptr_t)(out + nbytes) & 0xfff) == 0);
    562 		nbytes -= lastblock;
    563 
    564 		for (; nbytes; nbytes -= 16, in += 16, out += 16) {
    565 			xor128(out, in, t);
    566 			aesvia_dec1(dec, out, out, cw0);
    567 			xor128(out, out, t);
    568 			aesvia_xts_update(&t[0], &t[1], &t[2], &t[3]);
    569 		}
    570 
    571 		/* Handle the last block of a page, if necessary.  */
    572 		if (lastblock) {
    573 			uint8_t buf[16] __aligned(16);
    574 			xor128(buf, in, t);
    575 			aesvia_dec1(dec, buf, out, cw0);
    576 			explicit_memset(buf, 0, sizeof buf);
    577 		}
    578 	} else {
    579 		xtsdec_unaligned_evcnt.ev_count++;
    580 		uint8_t buf[16] __aligned(16);
    581 
    582 		for (; nbytes; nbytes -= 16, in += 16, out += 16) {
    583 			memcpy(buf, in, 16);
    584 			xor128(buf, buf, t);
    585 			aesvia_dec1(dec, buf, buf, cw0);
    586 			xor128(buf, buf, t);
    587 			memcpy(out, buf, 16);
    588 			aesvia_xts_update(&t[0], &t[1], &t[2], &t[3]);
    589 		}
    590 
    591 		explicit_memset(buf, 0, sizeof buf);
    592 	}
    593 	fpu_kern_leave();
    594 
    595 	memcpy(tweak, t, 16);
    596 	explicit_memset(t, 0, sizeof t);
    597 }
    598 
    599 static int
    600 aesvia_probe(void)
    601 {
    602 
    603 	/* Verify that the CPU advertises VIA ACE support.  */
    604 	if ((cpu_feature[4] & CPUID_VIA_HAS_ACE) == 0)
    605 		return -1;
    606 
    607 	/* Verify that our XTS tweak update logic works.  */
    608 	if (aesvia_xts_update_selftest())
    609 		return -1;
    610 
    611 	/* Success!  */
    612 	return 0;
    613 }
    614 
    615 struct aes_impl aes_via_impl = {
    616 	.ai_name = "VIA ACE",
    617 	.ai_probe = aesvia_probe,
    618 	.ai_setenckey = aesvia_setenckey,
    619 	.ai_setdeckey = aesvia_setdeckey,
    620 	.ai_enc = aesvia_enc,
    621 	.ai_dec = aesvia_dec,
    622 	.ai_cbc_enc = aesvia_cbc_enc,
    623 	.ai_cbc_dec = aesvia_cbc_dec,
    624 	.ai_xts_enc = aesvia_xts_enc,
    625 	.ai_xts_dec = aesvia_xts_dec,
    626 };
    627