xref: /openbsd/sys/arch/i386/pci/glxsb.c (revision a6445c1d)
1 /*	$OpenBSD: glxsb.c,v 1.28 2014/07/12 18:44:42 tedu Exp $	*/
2 
3 /*
4  * Copyright (c) 2006 Tom Cosgrove <tom@openbsd.org>
5  * Copyright (c) 2003, 2004 Theo de Raadt
6  * Copyright (c) 2003 Jason Wright
7  *
8  * Permission to use, copy, modify, and distribute this software for any
9  * purpose with or without fee is hereby granted, provided that the above
10  * copyright notice and this permission notice appear in all copies.
11  *
12  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19  */
20 
21 /*
22  * Driver for the security block on the AMD Geode LX processors
23  * http://www.amd.com/files/connectivitysolutions/geode/geode_lx/33234d_lx_ds.pdf
24  */
25 
26 #include <sys/param.h>
27 #include <sys/systm.h>
28 #include <sys/device.h>
29 #include <sys/malloc.h>
30 #include <sys/mbuf.h>
31 #include <sys/types.h>
32 #include <sys/timeout.h>
33 
34 #include <machine/bus.h>
35 #include <machine/cpufunc.h>
36 
37 #include <dev/rndvar.h>
38 #include <dev/pci/pcivar.h>
39 #include <dev/pci/pcidevs.h>
40 
41 #ifdef CRYPTO
42 #include <crypto/cryptodev.h>
43 #include <crypto/rijndael.h>
44 #include <crypto/xform.h>
45 #include <crypto/cryptosoft.h>
46 #endif
47 
48 #define SB_GLD_MSR_CAP		0x58002000	/* RO - Capabilities */
49 #define SB_GLD_MSR_CONFIG	0x58002001	/* RW - Master Config */
50 #define SB_GLD_MSR_SMI		0x58002002	/* RW - SMI */
51 #define SB_GLD_MSR_ERROR	0x58002003	/* RW - Error */
52 #define SB_GLD_MSR_PM		0x58002004	/* RW - Power Mgmt */
53 #define SB_GLD_MSR_DIAG		0x58002005	/* RW - Diagnostic */
54 #define SB_GLD_MSR_CTRL		0x58002006	/* RW - Security Block Cntrl */
55 
56 						/* For GLD_MSR_CTRL: */
57 #define SB_GMC_DIV0		0x0000		/* AES update divisor values */
58 #define SB_GMC_DIV1		0x0001
59 #define SB_GMC_DIV2		0x0002
60 #define SB_GMC_DIV3		0x0003
61 #define SB_GMC_DIV_MASK		0x0003
62 #define SB_GMC_SBI		0x0004		/* AES swap bits */
63 #define SB_GMC_SBY		0x0008		/* AES swap bytes */
64 #define SB_GMC_TW		0x0010		/* Time write (EEPROM) */
65 #define SB_GMC_T_SEL0		0x0000		/* RNG post-proc: none */
66 #define SB_GMC_T_SEL1		0x0100		/* RNG post-proc: LFSR */
67 #define SB_GMC_T_SEL2		0x0200		/* RNG post-proc: whitener */
68 #define SB_GMC_T_SEL3		0x0300		/* RNG LFSR+whitener */
69 #define SB_GMC_T_SEL_MASK	0x0300
70 #define SB_GMC_T_NE		0x0400		/* Noise (generator) Enable */
71 #define SB_GMC_T_TM		0x0800		/* RNG test mode */
72 						/*     (deterministic) */
73 
74 /* Security Block configuration/control registers (offsets from base) */
75 
76 #define SB_CTL_A		0x0000		/* RW - SB Control A */
77 #define SB_CTL_B		0x0004		/* RW - SB Control B */
78 #define SB_AES_INT		0x0008		/* RW - SB AES Interrupt */
79 #define SB_SOURCE_A		0x0010		/* RW - Source A */
80 #define SB_DEST_A		0x0014		/* RW - Destination A */
81 #define SB_LENGTH_A		0x0018		/* RW - Length A */
82 #define SB_SOURCE_B		0x0020		/* RW - Source B */
83 #define SB_DEST_B		0x0024		/* RW - Destination B */
84 #define SB_LENGTH_B		0x0028		/* RW - Length B */
85 #define SB_WKEY			0x0030		/* WO - Writable Key 0-3 */
86 #define SB_WKEY_0		0x0030		/* WO - Writable Key 0 */
87 #define SB_WKEY_1		0x0034		/* WO - Writable Key 1 */
88 #define SB_WKEY_2		0x0038		/* WO - Writable Key 2 */
89 #define SB_WKEY_3		0x003C		/* WO - Writable Key 3 */
90 #define SB_CBC_IV		0x0040		/* RW - CBC IV 0-3 */
91 #define SB_CBC_IV_0		0x0040		/* RW - CBC IV 0 */
92 #define SB_CBC_IV_1		0x0044		/* RW - CBC IV 1 */
93 #define SB_CBC_IV_2		0x0048		/* RW - CBC IV 2 */
94 #define SB_CBC_IV_3		0x004C		/* RW - CBC IV 3 */
95 #define SB_RANDOM_NUM		0x0050		/* RW - Random Number */
96 #define SB_RANDOM_NUM_STATUS	0x0054		/* RW - Random Number Status */
97 #define SB_EEPROM_COMM		0x0800		/* RW - EEPROM Command */
98 #define SB_EEPROM_ADDR		0x0804		/* RW - EEPROM Address */
99 #define SB_EEPROM_DATA		0x0808		/* RW - EEPROM Data */
100 #define SB_EEPROM_SEC_STATE	0x080C		/* RW - EEPROM Security State */
101 
102 						/* For SB_CTL_A and _B */
103 #define SB_CTL_ST		0x0001		/* Start operation (enc/dec) */
104 #define SB_CTL_ENC		0x0002		/* Encrypt (0 is decrypt) */
105 #define SB_CTL_DEC		0x0000		/* Decrypt */
106 #define SB_CTL_WK		0x0004		/* Use writable key (we set) */
107 #define SB_CTL_DC		0x0008		/* Destination coherent */
108 #define SB_CTL_SC		0x0010		/* Source coherent */
109 #define SB_CTL_CBC		0x0020		/* CBC (0 is ECB) */
110 
111 						/* For SB_AES_INT */
112 #define SB_AI_DISABLE_AES_A	0x00001		/* Disable AES A compl int */
113 #define SB_AI_ENABLE_AES_A	0x00000		/* Enable AES A compl int */
114 #define SB_AI_DISABLE_AES_B	0x00002		/* Disable AES B compl int */
115 #define SB_AI_ENABLE_AES_B	0x00000		/* Enable AES B compl int */
116 #define SB_AI_DISABLE_EEPROM	0x00004		/* Disable EEPROM op comp int */
117 #define SB_AI_ENABLE_EEPROM	0x00000		/* Enable EEPROM op compl int */
118 #define SB_AI_AES_A_COMPLETE	0x10000		/* AES A operation complete */
119 #define SB_AI_AES_B_COMPLETE	0x20000		/* AES B operation complete */
120 #define SB_AI_EEPROM_COMPLETE	0x40000		/* EEPROM operation complete */
121 
122 #define SB_RNS_TRNG_VALID	0x0001		/* in SB_RANDOM_NUM_STATUS */
123 
124 #define SB_MEM_SIZE		0x0810		/* Size of memory block */
125 
126 #define SB_AES_ALIGN		0x0010		/* Source and dest buffers */
127 						/* must be 16-byte aligned */
128 #define SB_AES_BLOCK_SIZE	0x0010
129 
130 /*
131  * The Geode LX security block AES acceleration doesn't perform scatter-
132  * gather: it just takes source and destination addresses.  Therefore the
133  * plain- and ciphertexts need to be contiguous.  To this end, we allocate
134  * a buffer for both, and accept the overhead of copying in and out.  If
135  * the number of bytes in one operation is bigger than allowed for by the
136  * buffer (buffer is twice the size of the max length, as it has both input
137  * and output) then we have to perform multiple encryptions/decryptions.
138  */
139 #define GLXSB_MAX_AES_LEN	16384
140 
141 #ifdef CRYPTO
142 struct glxsb_dma_map {
143 	bus_dmamap_t		dma_map;
144 	bus_dma_segment_t	dma_seg;
145 	int			dma_nsegs;
146 	int			dma_size;
147 	caddr_t			dma_vaddr;
148 	uint32_t		dma_paddr;
149 };
150 struct glxsb_session {
151 	uint32_t	ses_key[4];
152 	int		ses_klen;
153 	int		ses_used;
154 	struct swcr_data *ses_swd_auth;
155 	struct swcr_data *ses_swd_enc;
156 };
157 #endif /* CRYPTO */
158 
159 struct glxsb_softc {
160 	struct device		sc_dev;
161 	bus_space_tag_t		sc_iot;
162 	bus_space_handle_t	sc_ioh;
163 	struct timeout		sc_to;
164 
165 #ifdef CRYPTO
166 	bus_dma_tag_t		sc_dmat;
167 	struct glxsb_dma_map	sc_dma;
168 	int32_t			sc_cid;
169 	int			sc_nsessions;
170 	struct glxsb_session	*sc_sessions;
171 #endif /* CRYPTO */
172 
173 	uint64_t		save_gld_msr;
174 };
175 
176 int	glxsb_match(struct device *, void *, void *);
177 void	glxsb_attach(struct device *, struct device *, void *);
178 int	glxsb_activate(struct device *, int);
179 void	glxsb_rnd(void *);
180 
181 struct cfattach glxsb_ca = {
182 	sizeof(struct glxsb_softc), glxsb_match, glxsb_attach, NULL,
183 	glxsb_activate
184 };
185 
186 struct cfdriver glxsb_cd = {
187 	NULL, "glxsb", DV_DULL
188 };
189 
190 
191 #ifdef CRYPTO
192 
193 #define GLXSB_SESSION(sid)		((sid) & 0x0fffffff)
194 #define	GLXSB_SID(crd,ses)		(((crd) << 28) | ((ses) & 0x0fffffff))
195 
196 static struct glxsb_softc *glxsb_sc;
197 
198 int glxsb_crypto_setup(struct glxsb_softc *);
199 int glxsb_crypto_newsession(uint32_t *, struct cryptoini *);
200 int glxsb_crypto_process(struct cryptop *);
201 int glxsb_crypto_freesession(uint64_t);
202 static __inline void glxsb_aes(struct glxsb_softc *, uint32_t, uint32_t,
203     uint32_t, void *, int, void *);
204 
205 int glxsb_dma_alloc(struct glxsb_softc *, int, struct glxsb_dma_map *);
206 void glxsb_dma_pre_op(struct glxsb_softc *, struct glxsb_dma_map *);
207 void glxsb_dma_post_op(struct glxsb_softc *, struct glxsb_dma_map *);
208 void glxsb_dma_free(struct glxsb_softc *, struct glxsb_dma_map *);
209 
210 #endif /* CRYPTO */
211 
212 
213 int
214 glxsb_match(struct device *parent, void *match, void *aux)
215 {
216 	struct pci_attach_args *pa = aux;
217 
218 	if (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_AMD &&
219 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_AMD_GEODE_LX_CRYPTO)
220 		return (1);
221 
222 	return (0);
223 }
224 
225 void
226 glxsb_attach(struct device *parent, struct device *self, void *aux)
227 {
228 	struct glxsb_softc *sc = (void *) self;
229 	struct pci_attach_args *pa = aux;
230 	bus_addr_t membase;
231 	bus_size_t memsize;
232 	uint64_t msr;
233 #ifdef CRYPTO
234 	uint32_t intr;
235 #endif
236 
237 	msr = rdmsr(SB_GLD_MSR_CAP);
238 	if ((msr & 0xFFFF00) != 0x130400) {
239 		printf(": unknown ID 0x%x\n", (int) ((msr & 0xFFFF00) >> 16));
240 		return;
241 	}
242 
243 	/* printf(": revision %d", (int) (msr & 0xFF)); */
244 
245 	/* Map in the security block configuration/control registers */
246 	if (pci_mapreg_map(pa, PCI_MAPREG_START,
247 	    PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT, 0, &sc->sc_iot,
248 	    &sc->sc_ioh, &membase, &memsize, SB_MEM_SIZE)) {
249 		printf(": can't find mem space\n");
250 		return;
251 	}
252 
253 	/*
254 	 * Configure the Security Block.
255 	 *
256 	 * We want to enable the noise generator (T_NE), and enable the
257 	 * linear feedback shift register and whitener post-processing
258 	 * (T_SEL = 3).  Also ensure that test mode (deterministic values)
259 	 * is disabled.
260 	 */
261 	msr = rdmsr(SB_GLD_MSR_CTRL);
262 	msr &= ~(SB_GMC_T_TM | SB_GMC_T_SEL_MASK);
263 	msr |= SB_GMC_T_NE | SB_GMC_T_SEL3;
264 #if 0
265 	msr |= SB_GMC_SBI | SB_GMC_SBY;		/* for AES, if necessary */
266 #endif
267 	wrmsr(SB_GLD_MSR_CTRL, msr);
268 
269 	/* Install a periodic collector for the "true" (AMD's word) RNG */
270 	timeout_set(&sc->sc_to, glxsb_rnd, sc);
271 	glxsb_rnd(sc);
272 	printf(": RNG");
273 
274 #ifdef CRYPTO
275 	/* We don't have an interrupt handler, so disable completion INTs */
276 	intr = SB_AI_DISABLE_AES_A | SB_AI_DISABLE_AES_B |
277 	    SB_AI_DISABLE_EEPROM | SB_AI_AES_A_COMPLETE |
278 	    SB_AI_AES_B_COMPLETE | SB_AI_EEPROM_COMPLETE;
279 	bus_space_write_4(sc->sc_iot, sc->sc_ioh, SB_AES_INT, intr);
280 
281 	sc->sc_dmat = pa->pa_dmat;
282 
283 	if (glxsb_crypto_setup(sc))
284 		printf(" AES");
285 #endif
286 
287 	printf("\n");
288 }
289 
290 int
291 glxsb_activate(struct device *self, int act)
292 {
293 	struct glxsb_softc *sc = (struct glxsb_softc *)self;
294 
295 	switch (act) {
296 	case DVACT_QUIESCE:
297 		/* XXX should wait for current crypto op to finish */
298 		break;
299 	case DVACT_SUSPEND:
300 		sc->save_gld_msr = rdmsr(SB_GLD_MSR_CTRL);
301 		break;
302 	case DVACT_RESUME:
303 		wrmsr(SB_GLD_MSR_CTRL, sc->save_gld_msr);
304 		break;
305 	}
306 	return (0);
307 }
308 
309 void
310 glxsb_rnd(void *v)
311 {
312 	struct glxsb_softc *sc = v;
313 	uint32_t status, value;
314 
315 	status = bus_space_read_4(sc->sc_iot, sc->sc_ioh, SB_RANDOM_NUM_STATUS);
316 	if (status & SB_RNS_TRNG_VALID) {
317 		value = bus_space_read_4(sc->sc_iot, sc->sc_ioh, SB_RANDOM_NUM);
318 		add_true_randomness(value);
319 	}
320 
321 	timeout_add_msec(&sc->sc_to, 10);
322 }
323 
324 #ifdef CRYPTO
325 int
326 glxsb_crypto_setup(struct glxsb_softc *sc)
327 {
328 	int algs[CRYPTO_ALGORITHM_MAX + 1];
329 
330 	/* Allocate a contiguous DMA-able buffer to work in */
331 	if (glxsb_dma_alloc(sc, GLXSB_MAX_AES_LEN * 2, &sc->sc_dma) != 0)
332 		return 0;
333 
334 	bzero(algs, sizeof(algs));
335 	algs[CRYPTO_AES_CBC] = CRYPTO_ALG_FLAG_SUPPORTED;
336 	algs[CRYPTO_MD5_HMAC] = CRYPTO_ALG_FLAG_SUPPORTED;
337 	algs[CRYPTO_SHA1_HMAC] = CRYPTO_ALG_FLAG_SUPPORTED;
338 	algs[CRYPTO_RIPEMD160_HMAC] = CRYPTO_ALG_FLAG_SUPPORTED;
339 	algs[CRYPTO_SHA2_256_HMAC] = CRYPTO_ALG_FLAG_SUPPORTED;
340 	algs[CRYPTO_SHA2_384_HMAC] = CRYPTO_ALG_FLAG_SUPPORTED;
341 	algs[CRYPTO_SHA2_512_HMAC] = CRYPTO_ALG_FLAG_SUPPORTED;
342 
343 	sc->sc_cid = crypto_get_driverid(0);
344 	if (sc->sc_cid < 0)
345 		return 0;
346 
347 	crypto_register(sc->sc_cid, algs, glxsb_crypto_newsession,
348 	    glxsb_crypto_freesession, glxsb_crypto_process);
349 
350 	sc->sc_nsessions = 0;
351 
352 	glxsb_sc = sc;
353 
354 	return 1;
355 }
356 
357 int
358 glxsb_crypto_newsession(uint32_t *sidp, struct cryptoini *cri)
359 {
360 	struct glxsb_softc *sc = glxsb_sc;
361 	struct glxsb_session *ses = NULL;
362 	struct auth_hash	*axf;
363 	struct enc_xform	*txf;
364 	struct cryptoini	*c;
365 	struct swcr_data	*swd;
366 	int sesn, i;
367 
368 	if (sc == NULL || sidp == NULL || cri == NULL)
369 		return (EINVAL);
370 
371 	for (sesn = 0; sesn < sc->sc_nsessions; sesn++) {
372 		if (sc->sc_sessions[sesn].ses_used == 0) {
373 			ses = &sc->sc_sessions[sesn];
374 			break;
375 		}
376 	}
377 
378 	if (ses == NULL) {
379 		sesn = sc->sc_nsessions;
380 		ses = malloc((sesn + 1) * sizeof(*ses), M_DEVBUF, M_NOWAIT);
381 		if (ses == NULL)
382 			return (ENOMEM);
383 		if (sesn != 0) {
384 			bcopy(sc->sc_sessions, ses, sesn * sizeof(*ses));
385 			explicit_bzero(sc->sc_sessions, sesn * sizeof(*ses));
386 			free(sc->sc_sessions, M_DEVBUF, 0);
387 		}
388 		sc->sc_sessions = ses;
389 		ses = &sc->sc_sessions[sesn];
390 		sc->sc_nsessions++;
391 	}
392 
393 	bzero(ses, sizeof(*ses));
394 	ses->ses_used = 1;
395 
396 	for (c = cri; c != NULL; c = c->cri_next) {
397 		switch (c->cri_alg) {
398 		case CRYPTO_AES_CBC:
399 
400 			if (c->cri_klen != 128) {
401 				swd = malloc(sizeof(struct swcr_data),
402 				    M_CRYPTO_DATA, M_NOWAIT|M_ZERO);
403 				if (swd == NULL) {
404 					glxsb_crypto_freesession(sesn);
405 					return (ENOMEM);
406 				}
407 				ses->ses_swd_enc = swd;
408 				txf = &enc_xform_rijndael128;
409 				if (txf->ctxsize > 0) {
410 					swd->sw_kschedule =
411 					    malloc(txf->ctxsize,
412 						M_CRYPTO_DATA,
413 						M_NOWAIT|M_ZERO);
414 					if (swd->sw_kschedule == NULL) {
415 						glxsb_crypto_freesession(sesn);
416 						return (EINVAL);
417 					}
418 				}
419 				if (txf->setkey(swd->sw_kschedule, c->cri_key,
420 				    c->cri_klen / 8) < 0) {
421 					glxsb_crypto_freesession(sesn);
422 					return (EINVAL);
423 				}
424 				swd->sw_exf = txf;
425 				break;
426 			}
427 
428 			ses->ses_klen = c->cri_klen;
429 
430 			/* Copy the key (Geode LX wants the primary key only) */
431 			bcopy(c->cri_key, ses->ses_key, sizeof(ses->ses_key));
432 			break;
433 
434 		case CRYPTO_MD5_HMAC:
435 			axf = &auth_hash_hmac_md5_96;
436 			goto authcommon;
437 		case CRYPTO_SHA1_HMAC:
438 			axf = &auth_hash_hmac_sha1_96;
439 			goto authcommon;
440 		case CRYPTO_RIPEMD160_HMAC:
441 			axf = &auth_hash_hmac_ripemd_160_96;
442 			goto authcommon;
443 		case CRYPTO_SHA2_256_HMAC:
444 			axf = &auth_hash_hmac_sha2_256_128;
445 			goto authcommon;
446 		case CRYPTO_SHA2_384_HMAC:
447 			axf = &auth_hash_hmac_sha2_384_192;
448 			goto authcommon;
449 		case CRYPTO_SHA2_512_HMAC:
450 			axf = &auth_hash_hmac_sha2_512_256;
451 		authcommon:
452 			swd = malloc(sizeof(struct swcr_data), M_CRYPTO_DATA,
453 			    M_NOWAIT|M_ZERO);
454 			if (swd == NULL) {
455 				glxsb_crypto_freesession(sesn);
456 				return (ENOMEM);
457 			}
458 			ses->ses_swd_auth = swd;
459 
460 			swd->sw_ictx = malloc(axf->ctxsize, M_CRYPTO_DATA,
461 			    M_NOWAIT);
462 			if (swd->sw_ictx == NULL) {
463 				glxsb_crypto_freesession(sesn);
464 				return (ENOMEM);
465 			}
466 
467 			swd->sw_octx = malloc(axf->ctxsize, M_CRYPTO_DATA,
468 			    M_NOWAIT);
469 			if (swd->sw_octx == NULL) {
470 				glxsb_crypto_freesession(sesn);
471 				return (ENOMEM);
472 			}
473 
474 			for (i = 0; i < c->cri_klen / 8; i++)
475 				c->cri_key[i] ^= HMAC_IPAD_VAL;
476 
477 			axf->Init(swd->sw_ictx);
478 			axf->Update(swd->sw_ictx, c->cri_key, c->cri_klen / 8);
479 			axf->Update(swd->sw_ictx, hmac_ipad_buffer,
480 			    axf->blocksize - (c->cri_klen / 8));
481 
482 			for (i = 0; i < c->cri_klen / 8; i++)
483 				c->cri_key[i] ^= (HMAC_IPAD_VAL ^
484 				    HMAC_OPAD_VAL);
485 
486 			axf->Init(swd->sw_octx);
487 			axf->Update(swd->sw_octx, c->cri_key, c->cri_klen / 8);
488 			axf->Update(swd->sw_octx, hmac_opad_buffer,
489 			    axf->blocksize - (c->cri_klen / 8));
490 
491 			for (i = 0; i < c->cri_klen / 8; i++)
492 				c->cri_key[i] ^= HMAC_OPAD_VAL;
493 
494 			swd->sw_axf = axf;
495 			swd->sw_alg = c->cri_alg;
496 
497 			break;
498 		default:
499 			glxsb_crypto_freesession(sesn);
500 			return (EINVAL);
501 		}
502 	}
503 
504 	*sidp = GLXSB_SID(0, sesn);
505 	return (0);
506 }
507 
508 int
509 glxsb_crypto_freesession(uint64_t tid)
510 {
511 	struct glxsb_softc *sc = glxsb_sc;
512 	struct swcr_data *swd;
513 	struct auth_hash *axf;
514 	struct enc_xform *txf;
515 	int sesn;
516 	uint32_t sid = ((uint32_t)tid) & 0xffffffff;
517 
518 	if (sc == NULL)
519 		return (EINVAL);
520 	sesn = GLXSB_SESSION(sid);
521 	if (sesn >= sc->sc_nsessions)
522 		return (EINVAL);
523 	if ((swd = sc->sc_sessions[sesn].ses_swd_enc)) {
524 		txf = swd->sw_exf;
525 
526 		if (swd->sw_kschedule) {
527 			explicit_bzero(swd->sw_kschedule, txf->ctxsize);
528 			free(swd->sw_kschedule, M_CRYPTO_DATA, 0);
529 		}
530 		free(swd, M_CRYPTO_DATA, 0);
531 	}
532 	if ((swd = sc->sc_sessions[sesn].ses_swd_auth)) {
533 		axf = swd->sw_axf;
534 
535 		if (swd->sw_ictx) {
536 			explicit_bzero(swd->sw_ictx, axf->ctxsize);
537 			free(swd->sw_ictx, M_CRYPTO_DATA, 0);
538 		}
539 		if (swd->sw_octx) {
540 			explicit_bzero(swd->sw_octx, axf->ctxsize);
541 			free(swd->sw_octx, M_CRYPTO_DATA, 0);
542 		}
543 		free(swd, M_CRYPTO_DATA, 0);
544 	}
545 	explicit_bzero(&sc->sc_sessions[sesn], sizeof(sc->sc_sessions[sesn]));
546 	return (0);
547 }
548 
549 /*
550  * Must be called at splnet() or higher
551  */
552 static __inline void
553 glxsb_aes(struct glxsb_softc *sc, uint32_t control, uint32_t psrc,
554     uint32_t pdst, void *key, int len, void *iv)
555 {
556 	uint32_t status;
557 	int i;
558 
559 	if (len & 0xF) {
560 		printf("%s: len must be a multiple of 16 (not %d)\n",
561 		    sc->sc_dev.dv_xname, len);
562 		return;
563 	}
564 
565 	/* Set the source */
566 	bus_space_write_4(sc->sc_iot, sc->sc_ioh, SB_SOURCE_A, psrc);
567 
568 	/* Set the destination address */
569 	bus_space_write_4(sc->sc_iot, sc->sc_ioh, SB_DEST_A, pdst);
570 
571 	/* Set the data length */
572 	bus_space_write_4(sc->sc_iot, sc->sc_ioh, SB_LENGTH_A, len);
573 
574 	/* Set the IV */
575 	if (iv != NULL) {
576 		bus_space_write_region_4(sc->sc_iot, sc->sc_ioh,
577 		    SB_CBC_IV, iv, 4);
578 		control |= SB_CTL_CBC;
579 	}
580 
581 	/* Set the key */
582 	bus_space_write_region_4(sc->sc_iot, sc->sc_ioh, SB_WKEY, key, 4);
583 
584 	/* Ask the security block to do it */
585 	bus_space_write_4(sc->sc_iot, sc->sc_ioh, SB_CTL_A,
586 	    control | SB_CTL_WK | SB_CTL_DC | SB_CTL_SC | SB_CTL_ST);
587 
588 	/*
589 	 * Now wait until it is done.
590 	 *
591 	 * We do a busy wait.  Obviously the number of iterations of
592 	 * the loop required to perform the AES operation depends upon
593 	 * the number of bytes to process.
594 	 *
595 	 * On a 500 MHz Geode LX we see
596 	 *
597 	 *	length (bytes)	typical max iterations
598 	 *	    16		   12
599 	 *	    64		   22
600 	 *	   256		   59
601 	 *	  1024		  212
602 	 *	  8192		1,537
603 	 *
604 	 * Since we have a maximum size of operation defined in
605 	 * GLXSB_MAX_AES_LEN, we use this constant to decide how long
606 	 * to wait.  Allow an order of magnitude longer than it should
607 	 * really take, just in case.
608 	 */
609 	for (i = 0; i < GLXSB_MAX_AES_LEN * 10; i++) {
610 		status = bus_space_read_4(sc->sc_iot, sc->sc_ioh, SB_CTL_A);
611 
612 		if ((status & SB_CTL_ST) == 0)		/* Done */
613 			return;
614 	}
615 
616 	printf("%s: operation failed to complete\n", sc->sc_dev.dv_xname);
617 }
618 
619 static int
620 glxsb_crypto_swauth(struct cryptop *crp, struct cryptodesc *crd,
621     struct swcr_data *sw, caddr_t buf)
622 {
623 	int	type;
624 
625 	if (crp->crp_flags & CRYPTO_F_IMBUF)
626 		type = CRYPTO_BUF_MBUF;
627 	else
628 		type = CRYPTO_BUF_IOV;
629 
630 	return (swcr_authcompute(crp, crd, sw, buf, type));
631 }
632 
633 static int
634 glxsb_crypto_swenc(struct cryptop *crp, struct cryptodesc *crd,
635     struct swcr_data *sw, caddr_t buf)
636 {
637 	int	type;
638 
639 	if (crp->crp_flags & CRYPTO_F_IMBUF)
640 		type = CRYPTO_BUF_MBUF;
641 	else
642 		type = CRYPTO_BUF_IOV;
643 
644 	return (swcr_encdec(crd, sw, buf, type));
645 }
646 
647 static int
648 glxsb_crypto_encdec(struct cryptop *crp, struct cryptodesc *crd,
649     struct glxsb_session *ses, struct glxsb_softc *sc, caddr_t buf)
650 {
651 	char *op_src, *op_dst;
652 	uint32_t op_psrc, op_pdst;
653 	uint8_t op_iv[SB_AES_BLOCK_SIZE];
654 	int err = 0;
655 	int len, tlen, xlen;
656 	int offset;
657 	uint32_t control;
658 
659 	if (crd == NULL || (crd->crd_len % SB_AES_BLOCK_SIZE) != 0) {
660 		err = EINVAL;
661 		goto out;
662 	}
663 
664 	/* How much of our buffer will we need to use? */
665 	xlen = crd->crd_len > GLXSB_MAX_AES_LEN ?
666 	    GLXSB_MAX_AES_LEN : crd->crd_len;
667 
668 	/*
669 	 * XXX Check if we can have input == output on Geode LX.
670 	 * XXX In the meantime, use two separate (adjacent) buffers.
671 	 */
672 	op_src = sc->sc_dma.dma_vaddr;
673 	op_dst = sc->sc_dma.dma_vaddr + xlen;
674 
675 	op_psrc = sc->sc_dma.dma_paddr;
676 	op_pdst = sc->sc_dma.dma_paddr + xlen;
677 
678 	if (crd->crd_flags & CRD_F_ENCRYPT) {
679 		control = SB_CTL_ENC;
680 		if (crd->crd_flags & CRD_F_IV_EXPLICIT)
681 			bcopy(crd->crd_iv, op_iv, sizeof(op_iv));
682 		else
683 			arc4random_buf(op_iv, sizeof(op_iv));
684 
685 		if ((crd->crd_flags & CRD_F_IV_PRESENT) == 0) {
686 			if (crp->crp_flags & CRYPTO_F_IMBUF)
687 				err = m_copyback((struct mbuf *)crp->crp_buf,
688 				    crd->crd_inject, sizeof(op_iv), op_iv,
689 				    M_NOWAIT);
690 			else if (crp->crp_flags & CRYPTO_F_IOV)
691 				cuio_copyback((struct uio *)crp->crp_buf,
692 				    crd->crd_inject, sizeof(op_iv), op_iv);
693 			else
694 				bcopy(op_iv,
695 				    crp->crp_buf + crd->crd_inject, sizeof(op_iv));
696 			if (err)
697 				goto out;
698 		}
699 	} else {
700 		control = SB_CTL_DEC;
701 		if (crd->crd_flags & CRD_F_IV_EXPLICIT)
702 			bcopy(crd->crd_iv, op_iv, sizeof(op_iv));
703 		else {
704 			if (crp->crp_flags & CRYPTO_F_IMBUF)
705 				m_copydata((struct mbuf *)crp->crp_buf,
706 				    crd->crd_inject, sizeof(op_iv), op_iv);
707 			else if (crp->crp_flags & CRYPTO_F_IOV)
708 				cuio_copydata((struct uio *)crp->crp_buf,
709 				    crd->crd_inject, sizeof(op_iv), op_iv);
710 			else
711 				bcopy(crp->crp_buf + crd->crd_inject,
712 				    op_iv, sizeof(op_iv));
713 		}
714 	}
715 
716 	offset = 0;
717 	tlen = crd->crd_len;
718 
719 	/* Process the data in GLXSB_MAX_AES_LEN chunks */
720 	while (tlen > 0) {
721 		len = (tlen > GLXSB_MAX_AES_LEN) ? GLXSB_MAX_AES_LEN : tlen;
722 
723 		if (crp->crp_flags & CRYPTO_F_IMBUF)
724 			m_copydata((struct mbuf *)crp->crp_buf,
725 			    crd->crd_skip + offset, len, op_src);
726 		else if (crp->crp_flags & CRYPTO_F_IOV)
727 			cuio_copydata((struct uio *)crp->crp_buf,
728 			    crd->crd_skip + offset, len, op_src);
729 		else
730 			bcopy(crp->crp_buf + crd->crd_skip + offset, op_src,
731 			    len);
732 
733 		glxsb_dma_pre_op(sc, &sc->sc_dma);
734 
735 		glxsb_aes(sc, control, op_psrc, op_pdst, ses->ses_key,
736 		    len, op_iv);
737 
738 		glxsb_dma_post_op(sc, &sc->sc_dma);
739 
740 		if (crp->crp_flags & CRYPTO_F_IMBUF)
741 			err = m_copyback((struct mbuf *)crp->crp_buf,
742 			    crd->crd_skip + offset, len, op_dst, M_NOWAIT);
743 		else if (crp->crp_flags & CRYPTO_F_IOV)
744 			cuio_copyback((struct uio *)crp->crp_buf,
745 			    crd->crd_skip + offset, len, op_dst);
746 		else
747 			bcopy(op_dst, crp->crp_buf + crd->crd_skip + offset,
748 			    len);
749 		if (err)
750 			break;
751 
752 		offset += len;
753 		tlen -= len;
754 
755 		if (tlen > 0) {
756 			/* Copy out last block for use as next iteration */
757 			if (crd->crd_flags & CRD_F_ENCRYPT)
758 				bcopy(op_dst + len - sizeof(op_iv), op_iv,
759 				    sizeof(op_iv));
760 			else
761 				bcopy(op_src + len - sizeof(op_iv), op_iv,
762 				    sizeof(op_iv));
763 		}
764 	}
765 
766 	/* All AES processing has now been done. */
767 	explicit_bzero(sc->sc_dma.dma_vaddr, xlen * 2);
768 
769 out:
770 	return (err);
771 }
772 
773 int
774 glxsb_crypto_process(struct cryptop *crp)
775 {
776 	struct glxsb_softc *sc = glxsb_sc;
777 	struct glxsb_session *ses;
778 	struct cryptodesc *crd;
779 	int sesn,err = 0;
780 	int s;
781 
782 	s = splnet();
783 
784 	if (crp == NULL || crp->crp_callback == NULL) {
785 		err = EINVAL;
786 		goto out;
787 	}
788 	crd = crp->crp_desc;
789 	if (crd == NULL) {
790 		err = EINVAL;
791 		goto out;
792 	}
793 
794 	sesn = GLXSB_SESSION(crp->crp_sid);
795 	if (sesn >= sc->sc_nsessions) {
796 		err = EINVAL;
797 		goto out;
798 	}
799 	ses = &sc->sc_sessions[sesn];
800 	if (ses->ses_used == 0) {
801 		err = EINVAL;
802 		goto out;
803 	}
804 
805 	for (crd = crp->crp_desc; crd; crd = crd->crd_next) {
806 		switch (crd->crd_alg) {
807 		case CRYPTO_AES_CBC:
808 			if (ses->ses_swd_enc) {
809 				if ((err = glxsb_crypto_swenc(crp, crd, ses->ses_swd_enc,
810 				    crp->crp_buf)) != 0)
811 					goto out;
812 			} else if ((err = glxsb_crypto_encdec(crp, crd, ses, sc,
813 			    crp->crp_buf)) != 0)
814 				goto out;
815 			break;
816 
817 		case CRYPTO_MD5_HMAC:
818 		case CRYPTO_SHA1_HMAC:
819 		case CRYPTO_RIPEMD160_HMAC:
820 		case CRYPTO_SHA2_256_HMAC:
821 		case CRYPTO_SHA2_384_HMAC:
822 		case CRYPTO_SHA2_512_HMAC:
823 			if ((err = glxsb_crypto_swauth(crp, crd, ses->ses_swd_auth,
824 			    crp->crp_buf)) != 0)
825 				goto out;
826 			break;
827 
828 		default:
829 			err = EINVAL;
830 			goto out;
831 		}
832 	}
833 
834 out:
835 	crp->crp_etype = err;
836 	crypto_done(crp);
837 	splx(s);
838 	return (err);
839 }
840 
841 int
842 glxsb_dma_alloc(struct glxsb_softc *sc, int size, struct glxsb_dma_map *dma)
843 {
844 	int rc;
845 
846 	dma->dma_nsegs = 1;
847 	dma->dma_size = size;
848 
849 	rc = bus_dmamap_create(sc->sc_dmat, size, dma->dma_nsegs, size,
850 	    0, BUS_DMA_NOWAIT, &dma->dma_map);
851 	if (rc != 0) {
852 		printf("%s: couldn't create DMA map for %d bytes (%d)\n",
853 		    sc->sc_dev.dv_xname, size, rc);
854 
855 		goto fail0;
856 	}
857 
858 	rc = bus_dmamem_alloc(sc->sc_dmat, size, SB_AES_ALIGN, 0,
859 	    &dma->dma_seg, dma->dma_nsegs, &dma->dma_nsegs, BUS_DMA_NOWAIT);
860 	if (rc != 0) {
861 		printf("%s: couldn't allocate DMA memory of %d bytes (%d)\n",
862 		    sc->sc_dev.dv_xname, size, rc);
863 
864 		goto fail1;
865 	}
866 
867 	rc = bus_dmamem_map(sc->sc_dmat, &dma->dma_seg, 1, size,
868 	    &dma->dma_vaddr, BUS_DMA_NOWAIT);
869 	if (rc != 0) {
870 		printf("%s: couldn't map DMA memory for %d bytes (%d)\n",
871 		    sc->sc_dev.dv_xname, size, rc);
872 
873 		goto fail2;
874 	}
875 
876 	rc = bus_dmamap_load(sc->sc_dmat, dma->dma_map, dma->dma_vaddr,
877 	    size, NULL, BUS_DMA_NOWAIT);
878 	if (rc != 0) {
879 		printf("%s: couldn't load DMA memory for %d bytes (%d)\n",
880 		    sc->sc_dev.dv_xname, size, rc);
881 
882 		goto fail3;
883 	}
884 
885 	dma->dma_paddr = dma->dma_map->dm_segs[0].ds_addr;
886 
887 	return 0;
888 
889 fail3:
890 	bus_dmamem_unmap(sc->sc_dmat, dma->dma_vaddr, size);
891 fail2:
892 	bus_dmamem_free(sc->sc_dmat, &dma->dma_seg, dma->dma_nsegs);
893 fail1:
894 	bus_dmamap_destroy(sc->sc_dmat, dma->dma_map);
895 fail0:
896 	return rc;
897 }
898 
899 void
900 glxsb_dma_pre_op(struct glxsb_softc *sc, struct glxsb_dma_map *dma)
901 {
902 	bus_dmamap_sync(sc->sc_dmat, dma->dma_map, 0, dma->dma_size,
903 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
904 }
905 
906 void
907 glxsb_dma_post_op(struct glxsb_softc *sc, struct glxsb_dma_map *dma)
908 {
909 	bus_dmamap_sync(sc->sc_dmat, dma->dma_map, 0, dma->dma_size,
910 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
911 }
912 
913 void
914 glxsb_dma_free(struct glxsb_softc *sc, struct glxsb_dma_map *dma)
915 {
916 	bus_dmamap_unload(sc->sc_dmat, dma->dma_map);
917 	bus_dmamem_unmap(sc->sc_dmat, dma->dma_vaddr, dma->dma_size);
918 	bus_dmamem_free(sc->sc_dmat, &dma->dma_seg, dma->dma_nsegs);
919 	bus_dmamap_destroy(sc->sc_dmat, dma->dma_map);
920 }
921 
922 #endif /* CRYPTO */
923