xref: /netbsd-src/sys/dev/pci/ubsec.c (revision a536ee5124e62c9a0051a252f7833dc8f50f44c9)
1 /*	$NetBSD: ubsec.c,v 1.28 2012/10/27 17:18:35 chs Exp $	*/
2 /* $FreeBSD: src/sys/dev/ubsec/ubsec.c,v 1.6.2.6 2003/01/23 21:06:43 sam Exp $ */
3 /*	$OpenBSD: ubsec.c,v 1.127 2003/06/04 14:04:58 jason Exp $	*/
4 
5 /*
6  * Copyright (c) 2000 Jason L. Wright (jason@thought.net)
7  * Copyright (c) 2000 Theo de Raadt (deraadt@openbsd.org)
8  * Copyright (c) 2001 Patrik Lindergren (patrik@ipunplugged.com)
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
21  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
22  * DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT,
23  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
24  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
25  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
27  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
28  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  *
31  * Effort sponsored in part by the Defense Advanced Research Projects
32  * Agency (DARPA) and Air Force Research Laboratory, Air Force
33  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
34  *
35  */
36 
37 #include <sys/cdefs.h>
38 __KERNEL_RCSID(0, "$NetBSD: ubsec.c,v 1.28 2012/10/27 17:18:35 chs Exp $");
39 
40 #undef UBSEC_DEBUG
41 
42 /*
43  * uBsec 5[56]01, bcm580xx, bcm582x hardware crypto accelerator
44  */
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/proc.h>
49 #include <sys/endian.h>
50 #ifdef __NetBSD__
51   #define letoh16 htole16
52   #define letoh32 htole32
53 #define UBSEC_NO_RNG		/* until statistically tested */
54 #endif
55 #include <sys/errno.h>
56 #include <sys/malloc.h>
57 #include <sys/kernel.h>
58 #include <sys/mbuf.h>
59 #include <sys/device.h>
60 #include <sys/queue.h>
61 
62 #include <opencrypto/cryptodev.h>
63 #include <opencrypto/xform.h>
64 #ifdef __OpenBSD__
65  #include <dev/rndvar.h>
66  #include <sys/md5k.h>
67 #else
68  #include <sys/cprng.h>
69  #include <sys/md5.h>
70 #endif
71 #include <sys/sha1.h>
72 
73 #include <dev/pci/pcireg.h>
74 #include <dev/pci/pcivar.h>
75 #include <dev/pci/pcidevs.h>
76 
77 #include <dev/pci/ubsecreg.h>
78 #include <dev/pci/ubsecvar.h>
79 
80 /*
81  * Prototypes and count for the pci_device structure
82  */
83 static	int ubsec_probe(device_t, cfdata_t, void *);
84 static	void ubsec_attach(device_t, device_t, void *);
85 static	void ubsec_reset_board(struct ubsec_softc *);
86 static	void ubsec_init_board(struct ubsec_softc *);
87 static	void ubsec_init_pciregs(struct pci_attach_args *pa);
88 static	void ubsec_cleanchip(struct ubsec_softc *);
89 static	void ubsec_totalreset(struct ubsec_softc *);
90 static	int  ubsec_free_q(struct ubsec_softc*, struct ubsec_q *);
91 
92 #ifdef __OpenBSD__
93 struct cfattach ubsec_ca = {
94 	sizeof(struct ubsec_softc), ubsec_probe, ubsec_attach,
95 };
96 
97 struct cfdriver ubsec_cd = {
98 	0, "ubsec", DV_DULL
99 };
100 #else
101 CFATTACH_DECL_NEW(ubsec, sizeof(struct ubsec_softc), ubsec_probe, ubsec_attach,
102 	      NULL, NULL);
103 extern struct cfdriver ubsec_cd;
104 #endif
105 
106 /* patchable */
107 #ifdef	UBSEC_DEBUG
108 extern int ubsec_debug;
109 int ubsec_debug=1;
110 #endif
111 
112 static	int	ubsec_intr(void *);
113 static	int	ubsec_newsession(void*, u_int32_t *, struct cryptoini *);
114 static	int	ubsec_freesession(void*, u_int64_t);
115 static	int	ubsec_process(void*, struct cryptop *, int hint);
116 static	void	ubsec_callback(struct ubsec_softc *, struct ubsec_q *);
117 static	void	ubsec_feed(struct ubsec_softc *);
118 static	void	ubsec_mcopy(struct mbuf *, struct mbuf *, int, int);
119 static	void	ubsec_callback2(struct ubsec_softc *, struct ubsec_q2 *);
120 static	void	ubsec_feed2(struct ubsec_softc *);
121 #ifndef UBSEC_NO_RNG
122 static	void	ubsec_rng(void *);
123 #endif /* UBSEC_NO_RNG */
124 static	int 	ubsec_dma_malloc(struct ubsec_softc *, bus_size_t,
125 				 struct ubsec_dma_alloc *, int);
126 static	void	ubsec_dma_free(struct ubsec_softc *, struct ubsec_dma_alloc *);
127 static	int	ubsec_dmamap_aligned(bus_dmamap_t);
128 
129 static	int	ubsec_kprocess(void*, struct cryptkop *, int);
130 static	int	ubsec_kprocess_modexp_sw(struct ubsec_softc *,
131 					 struct cryptkop *, int);
132 static	int	ubsec_kprocess_modexp_hw(struct ubsec_softc *,
133 					 struct cryptkop *, int);
134 static	int	ubsec_kprocess_rsapriv(struct ubsec_softc *,
135 				       struct cryptkop *, int);
136 static	void	ubsec_kfree(struct ubsec_softc *, struct ubsec_q2 *);
137 static	int	ubsec_ksigbits(struct crparam *);
138 static	void	ubsec_kshift_r(u_int, u_int8_t *, u_int, u_int8_t *, u_int);
139 static	void	ubsec_kshift_l(u_int, u_int8_t *, u_int, u_int8_t *, u_int);
140 
141 #ifdef UBSEC_DEBUG
142 static void	ubsec_dump_pb(volatile struct ubsec_pktbuf *);
143 static void	ubsec_dump_mcr(struct ubsec_mcr *);
144 static	void	ubsec_dump_ctx2(volatile struct ubsec_ctx_keyop *);
145 #endif
146 
147 #define	READ_REG(sc,r) \
148 	bus_space_read_4((sc)->sc_st, (sc)->sc_sh, (r))
149 
150 #define WRITE_REG(sc,reg,val) \
151 	bus_space_write_4((sc)->sc_st, (sc)->sc_sh, reg, val)
152 
153 #define	SWAP32(x) (x) = htole32(ntohl((x)))
154 #ifndef HTOLE32
155  #define	HTOLE32(x) (x) = htole32(x)
156 #endif
157 
158 struct ubsec_stats ubsecstats;
159 
160 /*
161  * ubsec_maxbatch controls the number of crypto ops to voluntarily
162  * collect into one submission to the hardware.  This batching happens
163  * when ops are dispatched from the crypto subsystem with a hint that
164  * more are to follow immediately.  These ops must also not be marked
165  * with a ``no delay'' flag.
166  */
167 static	int ubsec_maxbatch = 1;
168 #ifdef SYSCTL_INT
169 SYSCTL_INT(_kern, OID_AUTO, ubsec_maxbatch, CTLFLAG_RW, &ubsec_maxbatch,
170 	    0, "Broadcom driver: max ops to batch w/o interrupt");
171 #endif
172 
173 /*
174  * ubsec_maxaggr controls the number of crypto ops to submit to the
175  * hardware as a unit.  This aggregation reduces the number of interrupts
176  * to the host at the expense of increased latency (for all but the last
177  * operation).  For network traffic setting this to one yields the highest
178  * performance but at the expense of more interrupt processing.
179  */
180 static	int ubsec_maxaggr = 1;
181 #ifdef SYSCTL_INT
182 SYSCTL_INT(_kern, OID_AUTO, ubsec_maxaggr, CTLFLAG_RW, &ubsec_maxaggr,
183 	    0, "Broadcom driver: max ops to aggregate under one interrupt");
184 #endif
185 
186 static const struct ubsec_product {
187 	pci_vendor_id_t		ubsec_vendor;
188 	pci_product_id_t	ubsec_product;
189 	int			ubsec_flags;
190 	int			ubsec_statmask;
191 	const char		*ubsec_name;
192 } ubsec_products[] = {
193 	{ PCI_VENDOR_BLUESTEEL,	PCI_PRODUCT_BLUESTEEL_5501,
194 	  0,
195 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR,
196 	  "Bluesteel 5501"
197 	},
198 	{ PCI_VENDOR_BLUESTEEL,	PCI_PRODUCT_BLUESTEEL_5601,
199 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG,
200 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR,
201 	  "Bluesteel 5601"
202 	},
203 
204 	{ PCI_VENDOR_BROADCOM,	PCI_PRODUCT_BROADCOM_5801,
205 	  0,
206 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR,
207 	  "Broadcom BCM5801"
208 	},
209 
210 	{ PCI_VENDOR_BROADCOM,	PCI_PRODUCT_BROADCOM_5802,
211 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG,
212 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR,
213 	  "Broadcom BCM5802"
214 	},
215 
216 	{ PCI_VENDOR_BROADCOM,	PCI_PRODUCT_BROADCOM_5805,
217 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG,
218 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR,
219 	  "Broadcom BCM5805"
220 	},
221 
222 	{ PCI_VENDOR_BROADCOM,	PCI_PRODUCT_BROADCOM_5820,
223 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG | UBS_FLAGS_LONGCTX |
224 	      UBS_FLAGS_HWNORM | UBS_FLAGS_BIGKEY,
225 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR,
226 	  "Broadcom BCM5820"
227 	},
228 
229 	{ PCI_VENDOR_BROADCOM,	PCI_PRODUCT_BROADCOM_5821,
230 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG | UBS_FLAGS_LONGCTX |
231 	      UBS_FLAGS_HWNORM | UBS_FLAGS_BIGKEY,
232 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR |
233 	      BS_STAT_MCR1_ALLEMPTY | BS_STAT_MCR2_ALLEMPTY,
234 	  "Broadcom BCM5821"
235 	},
236 	{ PCI_VENDOR_SUN,	PCI_PRODUCT_SUN_SCA1K,
237 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG | UBS_FLAGS_LONGCTX |
238 	      UBS_FLAGS_HWNORM | UBS_FLAGS_BIGKEY,
239 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR |
240 	      BS_STAT_MCR1_ALLEMPTY | BS_STAT_MCR2_ALLEMPTY,
241 	  "Sun Crypto Accelerator 1000"
242 	},
243 	{ PCI_VENDOR_SUN,	PCI_PRODUCT_SUN_5821,
244 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG | UBS_FLAGS_LONGCTX |
245 	      UBS_FLAGS_HWNORM | UBS_FLAGS_BIGKEY,
246 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR |
247 	      BS_STAT_MCR1_ALLEMPTY | BS_STAT_MCR2_ALLEMPTY,
248 	  "Broadcom BCM5821 (Sun)"
249 	},
250 
251 	{ PCI_VENDOR_BROADCOM,	PCI_PRODUCT_BROADCOM_5822,
252 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG | UBS_FLAGS_LONGCTX |
253 	      UBS_FLAGS_HWNORM | UBS_FLAGS_BIGKEY,
254 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR |
255 	      BS_STAT_MCR1_ALLEMPTY | BS_STAT_MCR2_ALLEMPTY,
256 	  "Broadcom BCM5822"
257 	},
258 
259 	{ PCI_VENDOR_BROADCOM,	PCI_PRODUCT_BROADCOM_5823,
260 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG | UBS_FLAGS_LONGCTX |
261 	      UBS_FLAGS_HWNORM | UBS_FLAGS_BIGKEY,
262 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR |
263 	      BS_STAT_MCR1_ALLEMPTY | BS_STAT_MCR2_ALLEMPTY,
264 	  "Broadcom BCM5823"
265 	},
266 
267 	{ 0,			0,
268 	  0,
269 	  0,
270 	  NULL
271 	}
272 };
273 
274 static const struct ubsec_product *
275 ubsec_lookup(const struct pci_attach_args *pa)
276 {
277 	const struct ubsec_product *up;
278 
279 	for (up = ubsec_products; up->ubsec_name != NULL; up++) {
280 		if (PCI_VENDOR(pa->pa_id) == up->ubsec_vendor &&
281 		    PCI_PRODUCT(pa->pa_id) == up->ubsec_product)
282 			return (up);
283 	}
284 	return (NULL);
285 }
286 
287 static int
288 ubsec_probe(device_t parent, cfdata_t match, void *aux)
289 {
290 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
291 
292 	if (ubsec_lookup(pa) != NULL)
293 		return (1);
294 
295 	return (0);
296 }
297 
298 static void
299 ubsec_attach(device_t parent, device_t self, void *aux)
300 {
301 	struct ubsec_softc *sc = device_private(self);
302 	struct pci_attach_args *pa = aux;
303 	const struct ubsec_product *up;
304 	pci_chipset_tag_t pc = pa->pa_pc;
305 	pci_intr_handle_t ih;
306 	const char *intrstr = NULL;
307 	struct ubsec_dma *dmap;
308 	u_int32_t cmd, i;
309 
310 	sc->sc_dev = self;
311 	up = ubsec_lookup(pa);
312 	if (up == NULL) {
313 		printf("\n");
314 		panic("ubsec_attach: impossible");
315 	}
316 
317 	pci_aprint_devinfo_fancy(pa, "Crypto processor", up->ubsec_name, 1);
318 
319 	SIMPLEQ_INIT(&sc->sc_queue);
320 	SIMPLEQ_INIT(&sc->sc_qchip);
321 	SIMPLEQ_INIT(&sc->sc_queue2);
322 	SIMPLEQ_INIT(&sc->sc_qchip2);
323 	SIMPLEQ_INIT(&sc->sc_q2free);
324 
325 	sc->sc_flags = up->ubsec_flags;
326 	sc->sc_statmask = up->ubsec_statmask;
327 
328 	cmd = pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
329 	cmd |= PCI_COMMAND_MASTER_ENABLE;
330 	pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG, cmd);
331 
332 	if (pci_mapreg_map(pa, BS_BAR, PCI_MAPREG_TYPE_MEM, 0,
333 	    &sc->sc_st, &sc->sc_sh, NULL, NULL)) {
334 		aprint_error_dev(self, "can't find mem space");
335 		return;
336 	}
337 
338 	sc->sc_dmat = pa->pa_dmat;
339 
340 	if (pci_intr_map(pa, &ih)) {
341 		aprint_error_dev(self, "couldn't map interrupt\n");
342 		return;
343 	}
344 	intrstr = pci_intr_string(pc, ih);
345 	sc->sc_ih = pci_intr_establish(pc, ih, IPL_NET, ubsec_intr, sc);
346 	if (sc->sc_ih == NULL) {
347 		aprint_error_dev(self, "couldn't establish interrupt");
348 		if (intrstr != NULL)
349 			aprint_error(" at %s", intrstr);
350 		aprint_error("\n");
351 		return;
352 	}
353 	aprint_normal_dev(self, "interrupting at %s\n", intrstr);
354 
355 	sc->sc_cid = crypto_get_driverid(0);
356 	if (sc->sc_cid < 0) {
357 		aprint_error_dev(self, "couldn't get crypto driver id\n");
358 		pci_intr_disestablish(pc, sc->sc_ih);
359 		return;
360 	}
361 
362 	SIMPLEQ_INIT(&sc->sc_freequeue);
363 	dmap = sc->sc_dmaa;
364 	for (i = 0; i < UBS_MAX_NQUEUE; i++, dmap++) {
365 		struct ubsec_q *q;
366 
367 		q = (struct ubsec_q *)malloc(sizeof(struct ubsec_q),
368 		    M_DEVBUF, M_NOWAIT);
369 		if (q == NULL) {
370 			aprint_error_dev(self, "can't allocate queue buffers\n");
371 			break;
372 		}
373 
374 		if (ubsec_dma_malloc(sc, sizeof(struct ubsec_dmachunk),
375 		    &dmap->d_alloc, 0)) {
376 			aprint_error_dev(self, "can't allocate dma buffers\n");
377 			free(q, M_DEVBUF);
378 			break;
379 		}
380 		dmap->d_dma = (struct ubsec_dmachunk *)dmap->d_alloc.dma_vaddr;
381 
382 		q->q_dma = dmap;
383 		sc->sc_queuea[i] = q;
384 
385 		SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
386 	}
387 
388 	crypto_register(sc->sc_cid, CRYPTO_3DES_CBC, 0, 0,
389 	    ubsec_newsession, ubsec_freesession, ubsec_process, sc);
390 	crypto_register(sc->sc_cid, CRYPTO_DES_CBC, 0, 0,
391 	    ubsec_newsession, ubsec_freesession, ubsec_process, sc);
392 	crypto_register(sc->sc_cid, CRYPTO_MD5_HMAC_96, 0, 0,
393 	    ubsec_newsession, ubsec_freesession, ubsec_process, sc);
394 	crypto_register(sc->sc_cid, CRYPTO_SHA1_HMAC_96, 0, 0,
395 	    ubsec_newsession, ubsec_freesession, ubsec_process, sc);
396 
397 	/*
398 	 * Reset Broadcom chip
399 	 */
400 	ubsec_reset_board(sc);
401 
402 	/*
403 	 * Init Broadcom specific PCI settings
404 	 */
405 	ubsec_init_pciregs(pa);
406 
407 	/*
408 	 * Init Broadcom chip
409 	 */
410 	ubsec_init_board(sc);
411 
412 #ifndef UBSEC_NO_RNG
413 	if (sc->sc_flags & UBS_FLAGS_RNG) {
414 		sc->sc_statmask |= BS_STAT_MCR2_DONE;
415 
416 		if (ubsec_dma_malloc(sc, sizeof(struct ubsec_mcr),
417 		    &sc->sc_rng.rng_q.q_mcr, 0))
418 			goto skip_rng;
419 
420 		if (ubsec_dma_malloc(sc, sizeof(struct ubsec_ctx_rngbypass),
421 		    &sc->sc_rng.rng_q.q_ctx, 0)) {
422 			ubsec_dma_free(sc, &sc->sc_rng.rng_q.q_mcr);
423 			goto skip_rng;
424 		}
425 
426 		if (ubsec_dma_malloc(sc, sizeof(u_int32_t) *
427 		    UBSEC_RNG_BUFSIZ, &sc->sc_rng.rng_buf, 0)) {
428 			ubsec_dma_free(sc, &sc->sc_rng.rng_q.q_ctx);
429 			ubsec_dma_free(sc, &sc->sc_rng.rng_q.q_mcr);
430 			goto skip_rng;
431 		}
432 
433 		if (hz >= 100)
434 			sc->sc_rnghz = hz / 100;
435 		else
436 			sc->sc_rnghz = 1;
437 #ifdef __OpenBSD__
438 		timeout_set(&sc->sc_rngto, ubsec_rng, sc);
439 		timeout_add(&sc->sc_rngto, sc->sc_rnghz);
440 #else
441 		callout_init(&sc->sc_rngto, 0);
442 		callout_reset(&sc->sc_rngto, sc->sc_rnghz, ubsec_rng, sc);
443 #endif
444  skip_rng:
445 		if (sc->sc_rnghz)
446 			aprint_normal_dev(self, "random number generator enabled\n");
447 		else
448 			aprint_error_dev(self, "WARNING: random number generator "
449 			    "disabled\n");
450 	}
451 #endif /* UBSEC_NO_RNG */
452 
453 	if (sc->sc_flags & UBS_FLAGS_KEY) {
454 		sc->sc_statmask |= BS_STAT_MCR2_DONE;
455 
456 		crypto_kregister(sc->sc_cid, CRK_MOD_EXP, 0,
457 				 ubsec_kprocess, sc);
458 #if 0
459 		crypto_kregister(sc->sc_cid, CRK_MOD_EXP_CRT, 0,
460 				 ubsec_kprocess, sc);
461 #endif
462 	}
463 }
464 
465 /*
466  * UBSEC Interrupt routine
467  */
468 static int
469 ubsec_intr(void *arg)
470 {
471 	struct ubsec_softc *sc = arg;
472 	volatile u_int32_t stat;
473 	struct ubsec_q *q;
474 	struct ubsec_dma *dmap;
475 	int npkts = 0, i;
476 
477 	stat = READ_REG(sc, BS_STAT);
478 	stat &= sc->sc_statmask;
479 	if (stat == 0) {
480 		return (0);
481 	}
482 
483 	WRITE_REG(sc, BS_STAT, stat);		/* IACK */
484 
485 	/*
486 	 * Check to see if we have any packets waiting for us
487 	 */
488 	if ((stat & BS_STAT_MCR1_DONE)) {
489 		while (!SIMPLEQ_EMPTY(&sc->sc_qchip)) {
490 			q = SIMPLEQ_FIRST(&sc->sc_qchip);
491 			dmap = q->q_dma;
492 
493 			if ((dmap->d_dma->d_mcr.mcr_flags & htole16(UBS_MCR_DONE)) == 0)
494 				break;
495 
496 			q = SIMPLEQ_FIRST(&sc->sc_qchip);
497 			SIMPLEQ_REMOVE_HEAD(&sc->sc_qchip, /*q,*/ q_next);
498 
499 			npkts = q->q_nstacked_mcrs;
500 			sc->sc_nqchip -= 1+npkts;
501 			/*
502 			 * search for further sc_qchip ubsec_q's that share
503 			 * the same MCR, and complete them too, they must be
504 			 * at the top.
505 			 */
506 			for (i = 0; i < npkts; i++) {
507 				if(q->q_stacked_mcr[i])
508 					ubsec_callback(sc, q->q_stacked_mcr[i]);
509 				else
510 					break;
511 			}
512 			ubsec_callback(sc, q);
513 		}
514 
515 		/*
516 		 * Don't send any more packet to chip if there has been
517 		 * a DMAERR.
518 		 */
519 		if (!(stat & BS_STAT_DMAERR))
520 			ubsec_feed(sc);
521 	}
522 
523 	/*
524 	 * Check to see if we have any key setups/rng's waiting for us
525 	 */
526 	if ((sc->sc_flags & (UBS_FLAGS_KEY|UBS_FLAGS_RNG)) &&
527 	    (stat & BS_STAT_MCR2_DONE)) {
528 		struct ubsec_q2 *q2;
529 		struct ubsec_mcr *mcr;
530 
531 		while (!SIMPLEQ_EMPTY(&sc->sc_qchip2)) {
532 			q2 = SIMPLEQ_FIRST(&sc->sc_qchip2);
533 
534 			bus_dmamap_sync(sc->sc_dmat, q2->q_mcr.dma_map,
535 			    0, q2->q_mcr.dma_map->dm_mapsize,
536 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
537 
538 			mcr = (struct ubsec_mcr *)q2->q_mcr.dma_vaddr;
539 			if ((mcr->mcr_flags & htole16(UBS_MCR_DONE)) == 0) {
540 				bus_dmamap_sync(sc->sc_dmat,
541 				    q2->q_mcr.dma_map, 0,
542 				    q2->q_mcr.dma_map->dm_mapsize,
543 				    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
544 				break;
545 			}
546 			q2 = SIMPLEQ_FIRST(&sc->sc_qchip2);
547 			SIMPLEQ_REMOVE_HEAD(&sc->sc_qchip2, /*q2,*/ q_next);
548 			ubsec_callback2(sc, q2);
549 			/*
550 			 * Don't send any more packet to chip if there has been
551 			 * a DMAERR.
552 			 */
553 			if (!(stat & BS_STAT_DMAERR))
554 				ubsec_feed2(sc);
555 		}
556 	}
557 
558 	/*
559 	 * Check to see if we got any DMA Error
560 	 */
561 	if (stat & BS_STAT_DMAERR) {
562 #ifdef UBSEC_DEBUG
563 		if (ubsec_debug) {
564 			volatile u_int32_t a = READ_REG(sc, BS_ERR);
565 
566 			printf("%s: dmaerr %s@%08x\n", device_xname(sc->sc_dev),
567 			    (a & BS_ERR_READ) ? "read" : "write",
568 			       a & BS_ERR_ADDR);
569 		}
570 #endif /* UBSEC_DEBUG */
571 		ubsecstats.hst_dmaerr++;
572 		ubsec_totalreset(sc);
573 		ubsec_feed(sc);
574 	}
575 
576 	if (sc->sc_needwakeup) {		/* XXX check high watermark */
577 		int wkeup = sc->sc_needwakeup & (CRYPTO_SYMQ|CRYPTO_ASYMQ);
578 #ifdef UBSEC_DEBUG
579 		if (ubsec_debug)
580 			printf("%s: wakeup crypto (%x)\n", device_xname(sc->sc_dev),
581 				sc->sc_needwakeup);
582 #endif /* UBSEC_DEBUG */
583 		sc->sc_needwakeup &= ~wkeup;
584 		crypto_unblock(sc->sc_cid, wkeup);
585 	}
586 	return (1);
587 }
588 
589 /*
590  * ubsec_feed() - aggregate and post requests to chip
591  * OpenBSD comments:
592  *		  It is assumed that the caller set splnet()
593  */
594 static void
595 ubsec_feed(struct ubsec_softc *sc)
596 {
597 	struct ubsec_q *q, *q2;
598 	int npkts, i;
599 	void *v;
600 	u_int32_t stat;
601 #ifdef UBSEC_DEBUG
602 	static int max;
603 #endif /* UBSEC_DEBUG */
604 
605 	npkts = sc->sc_nqueue;
606 	if (npkts > ubsecstats.hst_maxqueue)
607 		ubsecstats.hst_maxqueue = npkts;
608 	if (npkts < 2)
609 		goto feed1;
610 
611 	/*
612 	 * Decide how many ops to combine in a single MCR.  We cannot
613 	 * aggregate more than UBS_MAX_AGGR because this is the number
614 	 * of slots defined in the data structure.  Otherwise we clamp
615 	 * based on the tunable parameter ubsec_maxaggr.  Note that
616 	 * aggregation can happen in two ways: either by batching ops
617 	 * from above or because the h/w backs up and throttles us.
618 	 * Aggregating ops reduces the number of interrupts to the host
619 	 * but also (potentially) increases the latency for processing
620 	 * completed ops as we only get an interrupt when all aggregated
621 	 * ops have completed.
622 	 */
623 	if (npkts > UBS_MAX_AGGR)
624 		npkts = UBS_MAX_AGGR;
625 	if (npkts > ubsec_maxaggr)
626 		npkts = ubsec_maxaggr;
627 	if (npkts > ubsecstats.hst_maxbatch)
628 		ubsecstats.hst_maxbatch = npkts;
629 	if (npkts < 2)
630 		goto feed1;
631 	ubsecstats.hst_totbatch += npkts-1;
632 
633 	if ((stat = READ_REG(sc, BS_STAT)) & (BS_STAT_MCR1_FULL | BS_STAT_DMAERR)) {
634 		if (stat & BS_STAT_DMAERR) {
635 			ubsec_totalreset(sc);
636 			ubsecstats.hst_dmaerr++;
637 		} else {
638 			ubsecstats.hst_mcr1full++;
639 		}
640 		return;
641 	}
642 
643 #ifdef UBSEC_DEBUG
644 	if (ubsec_debug)
645 	    printf("merging %d records\n", npkts);
646 	/* XXX temporary aggregation statistics reporting code */
647 	if (max < npkts) {
648 		max = npkts;
649 		printf("%s: new max aggregate %d\n", device_xname(sc->sc_dev), max);
650 	}
651 #endif /* UBSEC_DEBUG */
652 
653 	q = SIMPLEQ_FIRST(&sc->sc_queue);
654 	SIMPLEQ_REMOVE_HEAD(&sc->sc_queue, /*q,*/ q_next);
655 	--sc->sc_nqueue;
656 
657 	bus_dmamap_sync(sc->sc_dmat, q->q_src_map,
658 	    0, q->q_src_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
659 	if (q->q_dst_map != NULL)
660 		bus_dmamap_sync(sc->sc_dmat, q->q_dst_map,
661 		    0, q->q_dst_map->dm_mapsize, BUS_DMASYNC_PREREAD);
662 
663 	q->q_nstacked_mcrs = npkts - 1;		/* Number of packets stacked */
664 
665 	for (i = 0; i < q->q_nstacked_mcrs; i++) {
666 		q2 = SIMPLEQ_FIRST(&sc->sc_queue);
667 		bus_dmamap_sync(sc->sc_dmat, q2->q_src_map,
668 		    0, q2->q_src_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
669 		if (q2->q_dst_map != NULL)
670 			bus_dmamap_sync(sc->sc_dmat, q2->q_dst_map,
671 			    0, q2->q_dst_map->dm_mapsize, BUS_DMASYNC_PREREAD);
672 		q2= SIMPLEQ_FIRST(&sc->sc_queue);
673 		SIMPLEQ_REMOVE_HEAD(&sc->sc_queue, /*q2,*/ q_next);
674 		--sc->sc_nqueue;
675 
676 		v = ((void *)&q2->q_dma->d_dma->d_mcr);
677 		v = (char*)v + (sizeof(struct ubsec_mcr) -
678 				 sizeof(struct ubsec_mcr_add));
679 		memcpy(&q->q_dma->d_dma->d_mcradd[i], v, sizeof(struct ubsec_mcr_add));
680 		q->q_stacked_mcr[i] = q2;
681 	}
682 	q->q_dma->d_dma->d_mcr.mcr_pkts = htole16(npkts);
683 	SIMPLEQ_INSERT_TAIL(&sc->sc_qchip, q, q_next);
684 	sc->sc_nqchip += npkts;
685 	if (sc->sc_nqchip > ubsecstats.hst_maxqchip)
686 		ubsecstats.hst_maxqchip = sc->sc_nqchip;
687 	bus_dmamap_sync(sc->sc_dmat, q->q_dma->d_alloc.dma_map,
688 	    0, q->q_dma->d_alloc.dma_map->dm_mapsize,
689 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
690 	WRITE_REG(sc, BS_MCR1, q->q_dma->d_alloc.dma_paddr +
691 	    offsetof(struct ubsec_dmachunk, d_mcr));
692 	return;
693 
694 feed1:
695 	while (!SIMPLEQ_EMPTY(&sc->sc_queue)) {
696 		if ((stat = READ_REG(sc, BS_STAT)) & (BS_STAT_MCR1_FULL | BS_STAT_DMAERR)) {
697 			if (stat & BS_STAT_DMAERR) {
698 				ubsec_totalreset(sc);
699 				ubsecstats.hst_dmaerr++;
700 			} else {
701 				ubsecstats.hst_mcr1full++;
702 			}
703 			break;
704 		}
705 
706 		q = SIMPLEQ_FIRST(&sc->sc_queue);
707 
708 		bus_dmamap_sync(sc->sc_dmat, q->q_src_map,
709 		    0, q->q_src_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
710 		if (q->q_dst_map != NULL)
711 			bus_dmamap_sync(sc->sc_dmat, q->q_dst_map,
712 			    0, q->q_dst_map->dm_mapsize, BUS_DMASYNC_PREREAD);
713 		bus_dmamap_sync(sc->sc_dmat, q->q_dma->d_alloc.dma_map,
714 		    0, q->q_dma->d_alloc.dma_map->dm_mapsize,
715 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
716 
717 		WRITE_REG(sc, BS_MCR1, q->q_dma->d_alloc.dma_paddr +
718 		    offsetof(struct ubsec_dmachunk, d_mcr));
719 #ifdef UBSEC_DEBUG
720 		if (ubsec_debug)
721 			printf("feed: q->chip %p %08x stat %08x\n",
722  		    	       q, (u_int32_t)q->q_dma->d_alloc.dma_paddr,
723 			       stat);
724 #endif /* UBSEC_DEBUG */
725 		q = SIMPLEQ_FIRST(&sc->sc_queue);
726 		SIMPLEQ_REMOVE_HEAD(&sc->sc_queue, /*q,*/ q_next);
727 		--sc->sc_nqueue;
728 		SIMPLEQ_INSERT_TAIL(&sc->sc_qchip, q, q_next);
729 		sc->sc_nqchip++;
730 	}
731 	if (sc->sc_nqchip > ubsecstats.hst_maxqchip)
732 		ubsecstats.hst_maxqchip = sc->sc_nqchip;
733 }
734 
735 /*
736  * Allocate a new 'session' and return an encoded session id.  'sidp'
737  * contains our registration id, and should contain an encoded session
738  * id on successful allocation.
739  */
740 static int
741 ubsec_newsession(void *arg, u_int32_t *sidp, struct cryptoini *cri)
742 {
743 	struct cryptoini *c, *encini = NULL, *macini = NULL;
744 	struct ubsec_softc *sc;
745 	struct ubsec_session *ses = NULL;
746 	MD5_CTX md5ctx;
747 	SHA1_CTX sha1ctx;
748 	int i, sesn;
749 
750 	sc = arg;
751 	KASSERT(sc != NULL /*, ("ubsec_newsession: null softc")*/);
752 
753 	if (sidp == NULL || cri == NULL || sc == NULL)
754 		return (EINVAL);
755 
756 	for (c = cri; c != NULL; c = c->cri_next) {
757 		if (c->cri_alg == CRYPTO_MD5_HMAC_96 ||
758 		    c->cri_alg == CRYPTO_SHA1_HMAC_96) {
759 			if (macini)
760 				return (EINVAL);
761 			macini = c;
762 		} else if (c->cri_alg == CRYPTO_DES_CBC ||
763 		    c->cri_alg == CRYPTO_3DES_CBC) {
764 			if (encini)
765 				return (EINVAL);
766 			encini = c;
767 		} else
768 			return (EINVAL);
769 	}
770 	if (encini == NULL && macini == NULL)
771 		return (EINVAL);
772 
773 	if (sc->sc_sessions == NULL) {
774 		ses = sc->sc_sessions = (struct ubsec_session *)malloc(
775 		    sizeof(struct ubsec_session), M_DEVBUF, M_NOWAIT);
776 		if (ses == NULL)
777 			return (ENOMEM);
778 		sesn = 0;
779 		sc->sc_nsessions = 1;
780 	} else {
781 		for (sesn = 0; sesn < sc->sc_nsessions; sesn++) {
782 			if (sc->sc_sessions[sesn].ses_used == 0) {
783 				ses = &sc->sc_sessions[sesn];
784 				break;
785 			}
786 		}
787 
788 		if (ses == NULL) {
789 			sesn = sc->sc_nsessions;
790 			ses = (struct ubsec_session *)malloc((sesn + 1) *
791 			    sizeof(struct ubsec_session), M_DEVBUF, M_NOWAIT);
792 			if (ses == NULL)
793 				return (ENOMEM);
794 			memcpy(ses, sc->sc_sessions, sesn *
795 			    sizeof(struct ubsec_session));
796 			memset(sc->sc_sessions, 0, sesn *
797 			    sizeof(struct ubsec_session));
798 			free(sc->sc_sessions, M_DEVBUF);
799 			sc->sc_sessions = ses;
800 			ses = &sc->sc_sessions[sesn];
801 			sc->sc_nsessions++;
802 		}
803 	}
804 
805 	memset(ses, 0, sizeof(struct ubsec_session));
806 	ses->ses_used = 1;
807 	if (encini) {
808 		/* get an IV, network byte order */
809 #ifdef __NetBSD__
810 		cprng_fast(ses->ses_iv, sizeof(ses->ses_iv));
811 #else
812 		get_random_bytes(ses->ses_iv, sizeof(ses->ses_iv));
813 #endif
814 
815 		/* Go ahead and compute key in ubsec's byte order */
816 		if (encini->cri_alg == CRYPTO_DES_CBC) {
817 			memcpy(&ses->ses_deskey[0], encini->cri_key, 8);
818 			memcpy(&ses->ses_deskey[2], encini->cri_key, 8);
819 			memcpy(&ses->ses_deskey[4], encini->cri_key, 8);
820 		} else
821 			memcpy(ses->ses_deskey, encini->cri_key, 24);
822 
823 		SWAP32(ses->ses_deskey[0]);
824 		SWAP32(ses->ses_deskey[1]);
825 		SWAP32(ses->ses_deskey[2]);
826 		SWAP32(ses->ses_deskey[3]);
827 		SWAP32(ses->ses_deskey[4]);
828 		SWAP32(ses->ses_deskey[5]);
829 	}
830 
831 	if (macini) {
832 		for (i = 0; i < macini->cri_klen / 8; i++)
833 			macini->cri_key[i] ^= HMAC_IPAD_VAL;
834 
835 		if (macini->cri_alg == CRYPTO_MD5_HMAC_96) {
836 			MD5Init(&md5ctx);
837 			MD5Update(&md5ctx, macini->cri_key,
838 			    macini->cri_klen / 8);
839 			MD5Update(&md5ctx, hmac_ipad_buffer,
840 			    HMAC_BLOCK_LEN - (macini->cri_klen / 8));
841 			memcpy(ses->ses_hminner, md5ctx.state,
842 			    sizeof(md5ctx.state));
843 		} else {
844 			SHA1Init(&sha1ctx);
845 			SHA1Update(&sha1ctx, macini->cri_key,
846 			    macini->cri_klen / 8);
847 			SHA1Update(&sha1ctx, hmac_ipad_buffer,
848 			    HMAC_BLOCK_LEN - (macini->cri_klen / 8));
849 			memcpy(ses->ses_hminner, sha1ctx.state,
850 			    sizeof(sha1ctx.state));
851 		}
852 
853 		for (i = 0; i < macini->cri_klen / 8; i++)
854 			macini->cri_key[i] ^= (HMAC_IPAD_VAL ^ HMAC_OPAD_VAL);
855 
856 		if (macini->cri_alg == CRYPTO_MD5_HMAC_96) {
857 			MD5Init(&md5ctx);
858 			MD5Update(&md5ctx, macini->cri_key,
859 			    macini->cri_klen / 8);
860 			MD5Update(&md5ctx, hmac_opad_buffer,
861 			    HMAC_BLOCK_LEN - (macini->cri_klen / 8));
862 			memcpy(ses->ses_hmouter, md5ctx.state,
863 			    sizeof(md5ctx.state));
864 		} else {
865 			SHA1Init(&sha1ctx);
866 			SHA1Update(&sha1ctx, macini->cri_key,
867 			    macini->cri_klen / 8);
868 			SHA1Update(&sha1ctx, hmac_opad_buffer,
869 			    HMAC_BLOCK_LEN - (macini->cri_klen / 8));
870 			memcpy(ses->ses_hmouter, sha1ctx.state,
871 			    sizeof(sha1ctx.state));
872 		}
873 
874 		for (i = 0; i < macini->cri_klen / 8; i++)
875 			macini->cri_key[i] ^= HMAC_OPAD_VAL;
876 	}
877 
878 	*sidp = UBSEC_SID(device_unit(sc->sc_dev), sesn);
879 	return (0);
880 }
881 
882 /*
883  * Deallocate a session.
884  */
885 static int
886 ubsec_freesession(void *arg, u_int64_t tid)
887 {
888 	struct ubsec_softc *sc;
889 	int session;
890 	u_int32_t sid = ((u_int32_t) tid) & 0xffffffff;
891 
892 	sc = arg;
893 	KASSERT(sc != NULL /*, ("ubsec_freesession: null softc")*/);
894 
895 	session = UBSEC_SESSION(sid);
896 	if (session >= sc->sc_nsessions)
897 		return (EINVAL);
898 
899 	memset(&sc->sc_sessions[session], 0, sizeof(sc->sc_sessions[session]));
900 	return (0);
901 }
902 
903 #ifdef __FreeBSD__ /* Ugly gratuitous changes to bus_dma */
904 static void
905 ubsec_op_cb(void *arg, bus_dma_segment_t *seg, int nsegs, bus_size_t mapsize, int error)
906 {
907 	struct ubsec_operand *op = arg;
908 
909 	KASSERT(nsegs <= UBS_MAX_SCATTER
910 		/*, ("Too many DMA segments returned when mapping operand")*/);
911 #ifdef UBSEC_DEBUG
912 	if (ubsec_debug)
913 		printf("ubsec_op_cb: mapsize %u nsegs %d\n",
914 			(u_int) mapsize, nsegs);
915 #endif
916 	op->mapsize = mapsize;
917 	op->nsegs = nsegs;
918 	memcpy(op->segs, seg, nsegs * sizeof (seg[0]));
919 }
920 #endif
921 
922 static int
923 ubsec_process(void *arg, struct cryptop *crp, int hint)
924 {
925 	struct ubsec_q *q = NULL;
926 #ifdef	__OpenBSD__
927 	int card;
928 #endif
929 	int err = 0, i, j, s, nicealign;
930 	struct ubsec_softc *sc;
931 	struct cryptodesc *crd1, *crd2, *maccrd, *enccrd;
932 	int encoffset = 0, macoffset = 0, cpskip, cpoffset;
933 	int sskip, dskip, stheend, dtheend;
934 	int16_t coffset;
935 	struct ubsec_session *ses;
936 	struct ubsec_pktctx ctx;
937 	struct ubsec_dma *dmap = NULL;
938 
939 	sc = arg;
940 	KASSERT(sc != NULL /*, ("ubsec_process: null softc")*/);
941 
942 	if (crp == NULL || crp->crp_callback == NULL || sc == NULL) {
943 		ubsecstats.hst_invalid++;
944 		return (EINVAL);
945 	}
946 	if (UBSEC_SESSION(crp->crp_sid) >= sc->sc_nsessions) {
947 		ubsecstats.hst_badsession++;
948 		return (EINVAL);
949 	}
950 
951 	s = splnet();
952 
953 	if (SIMPLEQ_EMPTY(&sc->sc_freequeue)) {
954 		ubsecstats.hst_queuefull++;
955 		sc->sc_needwakeup |= CRYPTO_SYMQ;
956 		splx(s);
957 		return(ERESTART);
958 	}
959 
960 	q = SIMPLEQ_FIRST(&sc->sc_freequeue);
961 	SIMPLEQ_REMOVE_HEAD(&sc->sc_freequeue, /*q,*/ q_next);
962 	splx(s);
963 
964 	dmap = q->q_dma; /* Save dma pointer */
965 	memset(q, 0, sizeof(struct ubsec_q));
966 	memset(&ctx, 0, sizeof(ctx));
967 
968 	q->q_sesn = UBSEC_SESSION(crp->crp_sid);
969 	q->q_dma = dmap;
970 	ses = &sc->sc_sessions[q->q_sesn];
971 
972 	if (crp->crp_flags & CRYPTO_F_IMBUF) {
973 		q->q_src_m = (struct mbuf *)crp->crp_buf;
974 		q->q_dst_m = (struct mbuf *)crp->crp_buf;
975 	} else if (crp->crp_flags & CRYPTO_F_IOV) {
976 		q->q_src_io = (struct uio *)crp->crp_buf;
977 		q->q_dst_io = (struct uio *)crp->crp_buf;
978 	} else {
979 		ubsecstats.hst_badflags++;
980 		err = EINVAL;
981 		goto errout;	/* XXX we don't handle contiguous blocks! */
982 	}
983 
984 	memset(&dmap->d_dma->d_mcr, 0, sizeof(struct ubsec_mcr));
985 
986 	dmap->d_dma->d_mcr.mcr_pkts = htole16(1);
987 	dmap->d_dma->d_mcr.mcr_flags = 0;
988 	q->q_crp = crp;
989 
990 	crd1 = crp->crp_desc;
991 	if (crd1 == NULL) {
992 		ubsecstats.hst_nodesc++;
993 		err = EINVAL;
994 		goto errout;
995 	}
996 	crd2 = crd1->crd_next;
997 
998 	if (crd2 == NULL) {
999 		if (crd1->crd_alg == CRYPTO_MD5_HMAC_96 ||
1000 		    crd1->crd_alg == CRYPTO_SHA1_HMAC_96) {
1001 			maccrd = crd1;
1002 			enccrd = NULL;
1003 		} else if (crd1->crd_alg == CRYPTO_DES_CBC ||
1004 		    crd1->crd_alg == CRYPTO_3DES_CBC) {
1005 			maccrd = NULL;
1006 			enccrd = crd1;
1007 		} else {
1008 			ubsecstats.hst_badalg++;
1009 			err = EINVAL;
1010 			goto errout;
1011 		}
1012 	} else {
1013 		if ((crd1->crd_alg == CRYPTO_MD5_HMAC_96 ||
1014 		    crd1->crd_alg == CRYPTO_SHA1_HMAC_96) &&
1015 		    (crd2->crd_alg == CRYPTO_DES_CBC ||
1016 			crd2->crd_alg == CRYPTO_3DES_CBC) &&
1017 		    ((crd2->crd_flags & CRD_F_ENCRYPT) == 0)) {
1018 			maccrd = crd1;
1019 			enccrd = crd2;
1020 		} else if ((crd1->crd_alg == CRYPTO_DES_CBC ||
1021 		    crd1->crd_alg == CRYPTO_3DES_CBC) &&
1022 		    (crd2->crd_alg == CRYPTO_MD5_HMAC_96 ||
1023 			crd2->crd_alg == CRYPTO_SHA1_HMAC_96) &&
1024 		    (crd1->crd_flags & CRD_F_ENCRYPT)) {
1025 			enccrd = crd1;
1026 			maccrd = crd2;
1027 		} else {
1028 			/*
1029 			 * We cannot order the ubsec as requested
1030 			 */
1031 			ubsecstats.hst_badalg++;
1032 			err = EINVAL;
1033 			goto errout;
1034 		}
1035 	}
1036 
1037 	if (enccrd) {
1038 		encoffset = enccrd->crd_skip;
1039 		ctx.pc_flags |= htole16(UBS_PKTCTX_ENC_3DES);
1040 
1041 		if (enccrd->crd_flags & CRD_F_ENCRYPT) {
1042 			q->q_flags |= UBSEC_QFLAGS_COPYOUTIV;
1043 
1044 			if (enccrd->crd_flags & CRD_F_IV_EXPLICIT)
1045 				memcpy(ctx.pc_iv, enccrd->crd_iv, 8);
1046 			else {
1047 				ctx.pc_iv[0] = ses->ses_iv[0];
1048 				ctx.pc_iv[1] = ses->ses_iv[1];
1049 			}
1050 
1051 			if ((enccrd->crd_flags & CRD_F_IV_PRESENT) == 0) {
1052 				if (crp->crp_flags & CRYPTO_F_IMBUF)
1053 					m_copyback(q->q_src_m,
1054 					    enccrd->crd_inject,
1055 					    8, (void *)ctx.pc_iv);
1056 				else if (crp->crp_flags & CRYPTO_F_IOV)
1057 					cuio_copyback(q->q_src_io,
1058 					    enccrd->crd_inject,
1059 					    8, (void *)ctx.pc_iv);
1060 			}
1061 		} else {
1062 			ctx.pc_flags |= htole16(UBS_PKTCTX_INBOUND);
1063 
1064 			if (enccrd->crd_flags & CRD_F_IV_EXPLICIT)
1065 				memcpy(ctx.pc_iv, enccrd->crd_iv, 8);
1066 			else if (crp->crp_flags & CRYPTO_F_IMBUF)
1067 				m_copydata(q->q_src_m, enccrd->crd_inject,
1068 				    8, (void *)ctx.pc_iv);
1069 			else if (crp->crp_flags & CRYPTO_F_IOV)
1070 				cuio_copydata(q->q_src_io,
1071 				    enccrd->crd_inject, 8,
1072 				    (void *)ctx.pc_iv);
1073 		}
1074 
1075 		ctx.pc_deskey[0] = ses->ses_deskey[0];
1076 		ctx.pc_deskey[1] = ses->ses_deskey[1];
1077 		ctx.pc_deskey[2] = ses->ses_deskey[2];
1078 		ctx.pc_deskey[3] = ses->ses_deskey[3];
1079 		ctx.pc_deskey[4] = ses->ses_deskey[4];
1080 		ctx.pc_deskey[5] = ses->ses_deskey[5];
1081 		SWAP32(ctx.pc_iv[0]);
1082 		SWAP32(ctx.pc_iv[1]);
1083 	}
1084 
1085 	if (maccrd) {
1086 		macoffset = maccrd->crd_skip;
1087 
1088 		if (maccrd->crd_alg == CRYPTO_MD5_HMAC_96)
1089 			ctx.pc_flags |= htole16(UBS_PKTCTX_AUTH_MD5);
1090 		else
1091 			ctx.pc_flags |= htole16(UBS_PKTCTX_AUTH_SHA1);
1092 
1093 		for (i = 0; i < 5; i++) {
1094 			ctx.pc_hminner[i] = ses->ses_hminner[i];
1095 			ctx.pc_hmouter[i] = ses->ses_hmouter[i];
1096 
1097 			HTOLE32(ctx.pc_hminner[i]);
1098 			HTOLE32(ctx.pc_hmouter[i]);
1099 		}
1100 	}
1101 
1102 	if (enccrd && maccrd) {
1103 		/*
1104 		 * ubsec cannot handle packets where the end of encryption
1105 		 * and authentication are not the same, or where the
1106 		 * encrypted part begins before the authenticated part.
1107 		 */
1108 		if ((encoffset + enccrd->crd_len) !=
1109 		    (macoffset + maccrd->crd_len)) {
1110 			ubsecstats.hst_lenmismatch++;
1111 			err = EINVAL;
1112 			goto errout;
1113 		}
1114 		if (enccrd->crd_skip < maccrd->crd_skip) {
1115 			ubsecstats.hst_skipmismatch++;
1116 			err = EINVAL;
1117 			goto errout;
1118 		}
1119 		sskip = maccrd->crd_skip;
1120 		cpskip = dskip = enccrd->crd_skip;
1121 		stheend = maccrd->crd_len;
1122 		dtheend = enccrd->crd_len;
1123 		coffset = enccrd->crd_skip - maccrd->crd_skip;
1124 		cpoffset = cpskip + dtheend;
1125 #ifdef UBSEC_DEBUG
1126 		if (ubsec_debug) {
1127 			printf("mac: skip %d, len %d, inject %d\n",
1128 			       maccrd->crd_skip, maccrd->crd_len, maccrd->crd_inject);
1129 			printf("enc: skip %d, len %d, inject %d\n",
1130 			       enccrd->crd_skip, enccrd->crd_len, enccrd->crd_inject);
1131 			printf("src: skip %d, len %d\n", sskip, stheend);
1132 			printf("dst: skip %d, len %d\n", dskip, dtheend);
1133 			printf("ubs: coffset %d, pktlen %d, cpskip %d, cpoffset %d\n",
1134 			       coffset, stheend, cpskip, cpoffset);
1135 		}
1136 #endif
1137 	} else {
1138 		cpskip = dskip = sskip = macoffset + encoffset;
1139 		dtheend = stheend = (enccrd)?enccrd->crd_len:maccrd->crd_len;
1140 		cpoffset = cpskip + dtheend;
1141 		coffset = 0;
1142 	}
1143 	ctx.pc_offset = htole16(coffset >> 2);
1144 
1145 	/* XXX FIXME: jonathan asks, what the heck's that 0xfff0?  */
1146 	if (bus_dmamap_create(sc->sc_dmat, 0xfff0, UBS_MAX_SCATTER,
1147 		0xfff0, 0, BUS_DMA_NOWAIT, &q->q_src_map) != 0) {
1148 		err = ENOMEM;
1149 		goto errout;
1150 	}
1151 	if (crp->crp_flags & CRYPTO_F_IMBUF) {
1152 		if (bus_dmamap_load_mbuf(sc->sc_dmat, q->q_src_map,
1153 		    q->q_src_m, BUS_DMA_NOWAIT) != 0) {
1154 			bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
1155 			q->q_src_map = NULL;
1156 			ubsecstats.hst_noload++;
1157 			err = ENOMEM;
1158 			goto errout;
1159 		}
1160 	} else if (crp->crp_flags & CRYPTO_F_IOV) {
1161 		if (bus_dmamap_load_uio(sc->sc_dmat, q->q_src_map,
1162 		    q->q_src_io, BUS_DMA_NOWAIT) != 0) {
1163 			bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
1164 			q->q_src_map = NULL;
1165 			ubsecstats.hst_noload++;
1166 			err = ENOMEM;
1167 			goto errout;
1168 		}
1169 	}
1170 	nicealign = ubsec_dmamap_aligned(q->q_src_map);
1171 
1172 	dmap->d_dma->d_mcr.mcr_pktlen = htole16(stheend);
1173 
1174 #ifdef UBSEC_DEBUG
1175 	if (ubsec_debug)
1176 		printf("src skip: %d nicealign: %u\n", sskip, nicealign);
1177 #endif
1178 	for (i = j = 0; i < q->q_src_map->dm_nsegs; i++) {
1179 		struct ubsec_pktbuf *pb;
1180 		bus_size_t packl = q->q_src_map->dm_segs[i].ds_len;
1181 		bus_addr_t packp = q->q_src_map->dm_segs[i].ds_addr;
1182 
1183 		if (sskip >= packl) {
1184 			sskip -= packl;
1185 			continue;
1186 		}
1187 
1188 		packl -= sskip;
1189 		packp += sskip;
1190 		sskip = 0;
1191 
1192 		if (packl > 0xfffc) {
1193 			err = EIO;
1194 			goto errout;
1195 		}
1196 
1197 		if (j == 0)
1198 			pb = &dmap->d_dma->d_mcr.mcr_ipktbuf;
1199 		else
1200 			pb = &dmap->d_dma->d_sbuf[j - 1];
1201 
1202 		pb->pb_addr = htole32(packp);
1203 
1204 		if (stheend) {
1205 			if (packl > stheend) {
1206 				pb->pb_len = htole32(stheend);
1207 				stheend = 0;
1208 			} else {
1209 				pb->pb_len = htole32(packl);
1210 				stheend -= packl;
1211 			}
1212 		} else
1213 			pb->pb_len = htole32(packl);
1214 
1215 		if ((i + 1) == q->q_src_map->dm_nsegs)
1216 			pb->pb_next = 0;
1217 		else
1218 			pb->pb_next = htole32(dmap->d_alloc.dma_paddr +
1219 			    offsetof(struct ubsec_dmachunk, d_sbuf[j]));
1220 		j++;
1221 	}
1222 
1223 	if (enccrd == NULL && maccrd != NULL) {
1224 		dmap->d_dma->d_mcr.mcr_opktbuf.pb_addr = 0;
1225 		dmap->d_dma->d_mcr.mcr_opktbuf.pb_len = 0;
1226 		dmap->d_dma->d_mcr.mcr_opktbuf.pb_next = htole32(dmap->d_alloc.dma_paddr +
1227 		    offsetof(struct ubsec_dmachunk, d_macbuf[0]));
1228 #ifdef UBSEC_DEBUG
1229 		if (ubsec_debug)
1230 			printf("opkt: %x %x %x\n",
1231 	 		    dmap->d_dma->d_mcr.mcr_opktbuf.pb_addr,
1232 	 		    dmap->d_dma->d_mcr.mcr_opktbuf.pb_len,
1233 	 		    dmap->d_dma->d_mcr.mcr_opktbuf.pb_next);
1234 
1235 #endif
1236 	} else {
1237 		if (crp->crp_flags & CRYPTO_F_IOV) {
1238 			if (!nicealign) {
1239 				ubsecstats.hst_iovmisaligned++;
1240 				err = EINVAL;
1241 				goto errout;
1242 			}
1243 			/* XXX: ``what the heck's that'' 0xfff0? */
1244 			if (bus_dmamap_create(sc->sc_dmat, 0xfff0,
1245 			    UBS_MAX_SCATTER, 0xfff0, 0, BUS_DMA_NOWAIT,
1246 			    &q->q_dst_map) != 0) {
1247 				ubsecstats.hst_nomap++;
1248 				err = ENOMEM;
1249 				goto errout;
1250 			}
1251 			if (bus_dmamap_load_uio(sc->sc_dmat, q->q_dst_map,
1252 			    q->q_dst_io, BUS_DMA_NOWAIT) != 0) {
1253 				bus_dmamap_destroy(sc->sc_dmat, q->q_dst_map);
1254 				q->q_dst_map = NULL;
1255 				ubsecstats.hst_noload++;
1256 				err = ENOMEM;
1257 				goto errout;
1258 			}
1259 		} else if (crp->crp_flags & CRYPTO_F_IMBUF) {
1260 			if (nicealign) {
1261 				q->q_dst_m = q->q_src_m;
1262 				q->q_dst_map = q->q_src_map;
1263 			} else {
1264 				int totlen, len;
1265 				struct mbuf *m, *top, **mp;
1266 
1267 				ubsecstats.hst_unaligned++;
1268 				totlen = q->q_src_map->dm_mapsize;
1269 				if (q->q_src_m->m_flags & M_PKTHDR) {
1270 					len = MHLEN;
1271 					MGETHDR(m, M_DONTWAIT, MT_DATA);
1272 					/*XXX FIXME: m_dup_pkthdr */
1273 					if (m && 1 /*!m_dup_pkthdr(m, q->q_src_m, M_DONTWAIT)*/) {
1274 						m_free(m);
1275 						m = NULL;
1276 					}
1277 				} else {
1278 					len = MLEN;
1279 					MGET(m, M_DONTWAIT, MT_DATA);
1280 				}
1281 				if (m == NULL) {
1282 					ubsecstats.hst_nombuf++;
1283 					err = sc->sc_nqueue ? ERESTART : ENOMEM;
1284 					goto errout;
1285 				}
1286 				if (len == MHLEN)
1287 				  /*XXX was M_DUP_PKTHDR*/
1288 				  M_COPY_PKTHDR(m, q->q_src_m);
1289 				if (totlen >= MINCLSIZE) {
1290 					MCLGET(m, M_DONTWAIT);
1291 					if ((m->m_flags & M_EXT) == 0) {
1292 						m_free(m);
1293 						ubsecstats.hst_nomcl++;
1294 						err = sc->sc_nqueue ? ERESTART : ENOMEM;
1295 						goto errout;
1296 					}
1297 					len = MCLBYTES;
1298 				}
1299 				m->m_len = len;
1300 				top = NULL;
1301 				mp = &top;
1302 
1303 				while (totlen > 0) {
1304 					if (top) {
1305 						MGET(m, M_DONTWAIT, MT_DATA);
1306 						if (m == NULL) {
1307 							m_freem(top);
1308 							ubsecstats.hst_nombuf++;
1309 							err = sc->sc_nqueue ? ERESTART : ENOMEM;
1310 							goto errout;
1311 						}
1312 						len = MLEN;
1313 					}
1314 					if (top && totlen >= MINCLSIZE) {
1315 						MCLGET(m, M_DONTWAIT);
1316 						if ((m->m_flags & M_EXT) == 0) {
1317 							*mp = m;
1318 							m_freem(top);
1319 							ubsecstats.hst_nomcl++;
1320 							err = sc->sc_nqueue ? ERESTART : ENOMEM;
1321 							goto errout;
1322 						}
1323 						len = MCLBYTES;
1324 					}
1325 					m->m_len = len = min(totlen, len);
1326 					totlen -= len;
1327 					*mp = m;
1328 					mp = &m->m_next;
1329 				}
1330 				q->q_dst_m = top;
1331 				ubsec_mcopy(q->q_src_m, q->q_dst_m,
1332 				    cpskip, cpoffset);
1333 				/* XXX again, what the heck is that 0xfff0? */
1334 				if (bus_dmamap_create(sc->sc_dmat, 0xfff0,
1335 				    UBS_MAX_SCATTER, 0xfff0, 0, BUS_DMA_NOWAIT,
1336 				    &q->q_dst_map) != 0) {
1337 					ubsecstats.hst_nomap++;
1338 					err = ENOMEM;
1339 					goto errout;
1340 				}
1341 				if (bus_dmamap_load_mbuf(sc->sc_dmat,
1342 				    q->q_dst_map, q->q_dst_m,
1343 				    BUS_DMA_NOWAIT) != 0) {
1344 					bus_dmamap_destroy(sc->sc_dmat,
1345 					q->q_dst_map);
1346 					q->q_dst_map = NULL;
1347 					ubsecstats.hst_noload++;
1348 					err = ENOMEM;
1349 					goto errout;
1350 				}
1351 			}
1352 		} else {
1353 			ubsecstats.hst_badflags++;
1354 			err = EINVAL;
1355 			goto errout;
1356 		}
1357 
1358 #ifdef UBSEC_DEBUG
1359 		if (ubsec_debug)
1360 			printf("dst skip: %d\n", dskip);
1361 #endif
1362 		for (i = j = 0; i < q->q_dst_map->dm_nsegs; i++) {
1363 			struct ubsec_pktbuf *pb;
1364 			bus_size_t packl = q->q_dst_map->dm_segs[i].ds_len;
1365 			bus_addr_t packp = q->q_dst_map->dm_segs[i].ds_addr;
1366 
1367 			if (dskip >= packl) {
1368 				dskip -= packl;
1369 				continue;
1370 			}
1371 
1372 			packl -= dskip;
1373 			packp += dskip;
1374 			dskip = 0;
1375 
1376 			if (packl > 0xfffc) {
1377 				err = EIO;
1378 				goto errout;
1379 			}
1380 
1381 			if (j == 0)
1382 				pb = &dmap->d_dma->d_mcr.mcr_opktbuf;
1383 			else
1384 				pb = &dmap->d_dma->d_dbuf[j - 1];
1385 
1386 			pb->pb_addr = htole32(packp);
1387 
1388 			if (dtheend) {
1389 				if (packl > dtheend) {
1390 					pb->pb_len = htole32(dtheend);
1391 					dtheend = 0;
1392 				} else {
1393 					pb->pb_len = htole32(packl);
1394 					dtheend -= packl;
1395 				}
1396 			} else
1397 				pb->pb_len = htole32(packl);
1398 
1399 			if ((i + 1) == q->q_dst_map->dm_nsegs) {
1400 				if (maccrd)
1401 					pb->pb_next = htole32(dmap->d_alloc.dma_paddr +
1402 					    offsetof(struct ubsec_dmachunk, d_macbuf[0]));
1403 				else
1404 					pb->pb_next = 0;
1405 			} else
1406 				pb->pb_next = htole32(dmap->d_alloc.dma_paddr +
1407 				    offsetof(struct ubsec_dmachunk, d_dbuf[j]));
1408 			j++;
1409 		}
1410 	}
1411 
1412 	dmap->d_dma->d_mcr.mcr_cmdctxp = htole32(dmap->d_alloc.dma_paddr +
1413 	    offsetof(struct ubsec_dmachunk, d_ctx));
1414 
1415 	if (sc->sc_flags & UBS_FLAGS_LONGCTX) {
1416 		struct ubsec_pktctx_long *ctxl;
1417 
1418 		ctxl = (struct ubsec_pktctx_long *)((char *)dmap->d_alloc.dma_vaddr +
1419 		    offsetof(struct ubsec_dmachunk, d_ctx));
1420 
1421 		/* transform small context into long context */
1422 		ctxl->pc_len = htole16(sizeof(struct ubsec_pktctx_long));
1423 		ctxl->pc_type = htole16(UBS_PKTCTX_TYPE_IPSEC);
1424 		ctxl->pc_flags = ctx.pc_flags;
1425 		ctxl->pc_offset = ctx.pc_offset;
1426 		for (i = 0; i < 6; i++)
1427 			ctxl->pc_deskey[i] = ctx.pc_deskey[i];
1428 		for (i = 0; i < 5; i++)
1429 			ctxl->pc_hminner[i] = ctx.pc_hminner[i];
1430 		for (i = 0; i < 5; i++)
1431 			ctxl->pc_hmouter[i] = ctx.pc_hmouter[i];
1432 		ctxl->pc_iv[0] = ctx.pc_iv[0];
1433 		ctxl->pc_iv[1] = ctx.pc_iv[1];
1434 	} else
1435 		memcpy((char *)dmap->d_alloc.dma_vaddr +
1436 		    offsetof(struct ubsec_dmachunk, d_ctx), &ctx,
1437 		    sizeof(struct ubsec_pktctx));
1438 
1439 	s = splnet();
1440 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue, q, q_next);
1441 	sc->sc_nqueue++;
1442 	ubsecstats.hst_ipackets++;
1443 	ubsecstats.hst_ibytes += dmap->d_alloc.dma_map->dm_mapsize;
1444 	if ((hint & CRYPTO_HINT_MORE) == 0 || sc->sc_nqueue >= ubsec_maxbatch)
1445 		ubsec_feed(sc);
1446 	splx(s);
1447 	return (0);
1448 
1449 errout:
1450 	if (q != NULL) {
1451 		if ((q->q_dst_m != NULL) && (q->q_src_m != q->q_dst_m))
1452 			m_freem(q->q_dst_m);
1453 
1454 		if (q->q_dst_map != NULL && q->q_dst_map != q->q_src_map) {
1455 			bus_dmamap_unload(sc->sc_dmat, q->q_dst_map);
1456 			bus_dmamap_destroy(sc->sc_dmat, q->q_dst_map);
1457 		}
1458 		if (q->q_src_map != NULL) {
1459 			bus_dmamap_unload(sc->sc_dmat, q->q_src_map);
1460 			bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
1461 		}
1462 
1463 		s = splnet();
1464 		SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
1465 		splx(s);
1466 	}
1467 #if 0 /* jonathan says: this openbsd code seems to be subsumed elsewhere */
1468 	if (err == EINVAL)
1469 		ubsecstats.hst_invalid++;
1470 	else
1471 		ubsecstats.hst_nomem++;
1472 #endif
1473 	if (err != ERESTART) {
1474 		crp->crp_etype = err;
1475 		crypto_done(crp);
1476 	} else {
1477 		sc->sc_needwakeup |= CRYPTO_SYMQ;
1478 	}
1479 	return (err);
1480 }
1481 
1482 static void
1483 ubsec_callback(struct ubsec_softc *sc, struct ubsec_q *q)
1484 {
1485 	struct cryptop *crp = (struct cryptop *)q->q_crp;
1486 	struct cryptodesc *crd;
1487 	struct ubsec_dma *dmap = q->q_dma;
1488 
1489 	ubsecstats.hst_opackets++;
1490 	ubsecstats.hst_obytes += dmap->d_alloc.dma_size;
1491 
1492 	bus_dmamap_sync(sc->sc_dmat, dmap->d_alloc.dma_map, 0,
1493 	    dmap->d_alloc.dma_map->dm_mapsize,
1494 	    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1495 	if (q->q_dst_map != NULL && q->q_dst_map != q->q_src_map) {
1496 		bus_dmamap_sync(sc->sc_dmat, q->q_dst_map,
1497 		    0, q->q_dst_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1498 		bus_dmamap_unload(sc->sc_dmat, q->q_dst_map);
1499 		bus_dmamap_destroy(sc->sc_dmat, q->q_dst_map);
1500 	}
1501 	bus_dmamap_sync(sc->sc_dmat, q->q_src_map,
1502 	    0, q->q_src_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1503 	bus_dmamap_unload(sc->sc_dmat, q->q_src_map);
1504 	bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
1505 
1506 	if ((crp->crp_flags & CRYPTO_F_IMBUF) && (q->q_src_m != q->q_dst_m)) {
1507 		m_freem(q->q_src_m);
1508 		crp->crp_buf = (void *)q->q_dst_m;
1509 	}
1510 
1511 	/* copy out IV for future use */
1512 	if (q->q_flags & UBSEC_QFLAGS_COPYOUTIV) {
1513 		for (crd = crp->crp_desc; crd; crd = crd->crd_next) {
1514 			if (crd->crd_alg != CRYPTO_DES_CBC &&
1515 			    crd->crd_alg != CRYPTO_3DES_CBC)
1516 				continue;
1517 			if (crp->crp_flags & CRYPTO_F_IMBUF)
1518 				m_copydata((struct mbuf *)crp->crp_buf,
1519 				    crd->crd_skip + crd->crd_len - 8, 8,
1520 				    (void *)sc->sc_sessions[q->q_sesn].ses_iv);
1521 			else if (crp->crp_flags & CRYPTO_F_IOV) {
1522 				cuio_copydata((struct uio *)crp->crp_buf,
1523 				    crd->crd_skip + crd->crd_len - 8, 8,
1524 				    (void *)sc->sc_sessions[q->q_sesn].ses_iv);
1525 			}
1526 			break;
1527 		}
1528 	}
1529 
1530 	for (crd = crp->crp_desc; crd; crd = crd->crd_next) {
1531 		if (crd->crd_alg != CRYPTO_MD5_HMAC_96 &&
1532 		    crd->crd_alg != CRYPTO_SHA1_HMAC_96)
1533 			continue;
1534 		if (crp->crp_flags & CRYPTO_F_IMBUF)
1535 			m_copyback((struct mbuf *)crp->crp_buf,
1536 			    crd->crd_inject, 12,
1537 			    (void *)dmap->d_dma->d_macbuf);
1538 		else if (crp->crp_flags & CRYPTO_F_IOV && crp->crp_mac)
1539 			bcopy((void *)dmap->d_dma->d_macbuf,
1540 			    crp->crp_mac, 12);
1541 		break;
1542 	}
1543 	SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
1544 	crypto_done(crp);
1545 }
1546 
1547 static void
1548 ubsec_mcopy(struct mbuf *srcm, struct mbuf *dstm, int hoffset, int toffset)
1549 {
1550 	int i, j, dlen, slen;
1551 	char *dptr, *sptr;
1552 
1553 	j = 0;
1554 	sptr = srcm->m_data;
1555 	slen = srcm->m_len;
1556 	dptr = dstm->m_data;
1557 	dlen = dstm->m_len;
1558 
1559 	while (1) {
1560 		for (i = 0; i < min(slen, dlen); i++) {
1561 			if (j < hoffset || j >= toffset)
1562 				*dptr++ = *sptr++;
1563 			slen--;
1564 			dlen--;
1565 			j++;
1566 		}
1567 		if (slen == 0) {
1568 			srcm = srcm->m_next;
1569 			if (srcm == NULL)
1570 				return;
1571 			sptr = srcm->m_data;
1572 			slen = srcm->m_len;
1573 		}
1574 		if (dlen == 0) {
1575 			dstm = dstm->m_next;
1576 			if (dstm == NULL)
1577 				return;
1578 			dptr = dstm->m_data;
1579 			dlen = dstm->m_len;
1580 		}
1581 	}
1582 }
1583 
1584 /*
1585  * feed the key generator, must be called at splnet() or higher.
1586  */
1587 static void
1588 ubsec_feed2(struct ubsec_softc *sc)
1589 {
1590 	struct ubsec_q2 *q;
1591 
1592 	while (!SIMPLEQ_EMPTY(&sc->sc_queue2)) {
1593 		if (READ_REG(sc, BS_STAT) & BS_STAT_MCR2_FULL)
1594 			break;
1595 		q = SIMPLEQ_FIRST(&sc->sc_queue2);
1596 
1597 		bus_dmamap_sync(sc->sc_dmat, q->q_mcr.dma_map, 0,
1598 		    q->q_mcr.dma_map->dm_mapsize,
1599 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1600 		bus_dmamap_sync(sc->sc_dmat, q->q_ctx.dma_map, 0,
1601 		    q->q_ctx.dma_map->dm_mapsize,
1602 		    BUS_DMASYNC_PREWRITE);
1603 
1604 		WRITE_REG(sc, BS_MCR2, q->q_mcr.dma_paddr);
1605 		q = SIMPLEQ_FIRST(&sc->sc_queue2);
1606 		SIMPLEQ_REMOVE_HEAD(&sc->sc_queue2, /*q,*/ q_next);
1607 		--sc->sc_nqueue2;
1608 		SIMPLEQ_INSERT_TAIL(&sc->sc_qchip2, q, q_next);
1609 	}
1610 }
1611 
1612 /*
1613  * Callback for handling random numbers
1614  */
1615 static void
1616 ubsec_callback2(struct ubsec_softc *sc, struct ubsec_q2 *q)
1617 {
1618 	struct cryptkop *krp;
1619 	struct ubsec_ctx_keyop *ctx;
1620 
1621 	ctx = (struct ubsec_ctx_keyop *)q->q_ctx.dma_vaddr;
1622 	bus_dmamap_sync(sc->sc_dmat, q->q_ctx.dma_map, 0,
1623 	    q->q_ctx.dma_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1624 
1625 	switch (q->q_type) {
1626 #ifndef UBSEC_NO_RNG
1627 	case UBS_CTXOP_RNGSHA1:
1628 	case UBS_CTXOP_RNGBYPASS: {
1629 		struct ubsec_q2_rng *rng = (struct ubsec_q2_rng *)q;
1630 		u_int32_t *p;
1631 		int i;
1632 
1633 		bus_dmamap_sync(sc->sc_dmat, rng->rng_buf.dma_map, 0,
1634 		    rng->rng_buf.dma_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1635 		p = (u_int32_t *)rng->rng_buf.dma_vaddr;
1636 #ifndef __NetBSD__
1637 		for (i = 0; i < UBSEC_RNG_BUFSIZ; p++, i++)
1638 			add_true_randomness(letoh32(*p));
1639 		rng->rng_used = 0;
1640 #else
1641 		/* XXX NetBSD rnd subsystem too weak */
1642 		i = 0; (void)i;	/* shut off gcc warnings */
1643 #endif
1644 #ifdef __OpenBSD__
1645 		timeout_add(&sc->sc_rngto, sc->sc_rnghz);
1646 #else
1647 		callout_reset(&sc->sc_rngto, sc->sc_rnghz, ubsec_rng, sc);
1648 #endif
1649 		break;
1650 	}
1651 #endif
1652 	case UBS_CTXOP_MODEXP: {
1653 		struct ubsec_q2_modexp *me = (struct ubsec_q2_modexp *)q;
1654 		u_int rlen, clen;
1655 
1656 		krp = me->me_krp;
1657 		rlen = (me->me_modbits + 7) / 8;
1658 		clen = (krp->krp_param[krp->krp_iparams].crp_nbits + 7) / 8;
1659 
1660 		bus_dmamap_sync(sc->sc_dmat, me->me_M.dma_map,
1661 		    0, me->me_M.dma_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1662 		bus_dmamap_sync(sc->sc_dmat, me->me_E.dma_map,
1663 		    0, me->me_E.dma_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1664 		bus_dmamap_sync(sc->sc_dmat, me->me_C.dma_map,
1665 		    0, me->me_C.dma_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1666 		bus_dmamap_sync(sc->sc_dmat, me->me_epb.dma_map,
1667 		    0, me->me_epb.dma_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1668 
1669 		if (clen < rlen)
1670 			krp->krp_status = E2BIG;
1671 		else {
1672 			if (sc->sc_flags & UBS_FLAGS_HWNORM) {
1673 				memset(krp->krp_param[krp->krp_iparams].crp_p, 0,
1674 				    (krp->krp_param[krp->krp_iparams].crp_nbits
1675 					+ 7) / 8);
1676 				bcopy(me->me_C.dma_vaddr,
1677 				    krp->krp_param[krp->krp_iparams].crp_p,
1678 				    (me->me_modbits + 7) / 8);
1679 			} else
1680 				ubsec_kshift_l(me->me_shiftbits,
1681 				    me->me_C.dma_vaddr, me->me_normbits,
1682 				    krp->krp_param[krp->krp_iparams].crp_p,
1683 				    krp->krp_param[krp->krp_iparams].crp_nbits);
1684 		}
1685 
1686 		crypto_kdone(krp);
1687 
1688 		/* bzero all potentially sensitive data */
1689 		memset(me->me_E.dma_vaddr, 0, me->me_E.dma_size);
1690 		memset(me->me_M.dma_vaddr, 0, me->me_M.dma_size);
1691 		memset(me->me_C.dma_vaddr, 0, me->me_C.dma_size);
1692 		memset(me->me_q.q_ctx.dma_vaddr, 0, me->me_q.q_ctx.dma_size);
1693 
1694 		/* Can't free here, so put us on the free list. */
1695 		SIMPLEQ_INSERT_TAIL(&sc->sc_q2free, &me->me_q, q_next);
1696 		break;
1697 	}
1698 	case UBS_CTXOP_RSAPRIV: {
1699 		struct ubsec_q2_rsapriv *rp = (struct ubsec_q2_rsapriv *)q;
1700 		u_int len;
1701 
1702 		krp = rp->rpr_krp;
1703 		bus_dmamap_sync(sc->sc_dmat, rp->rpr_msgin.dma_map, 0,
1704 		    rp->rpr_msgin.dma_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1705 		bus_dmamap_sync(sc->sc_dmat, rp->rpr_msgout.dma_map, 0,
1706 		    rp->rpr_msgout.dma_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1707 
1708 		len = (krp->krp_param[UBS_RSAPRIV_PAR_MSGOUT].crp_nbits + 7) / 8;
1709 		bcopy(rp->rpr_msgout.dma_vaddr,
1710 		    krp->krp_param[UBS_RSAPRIV_PAR_MSGOUT].crp_p, len);
1711 
1712 		crypto_kdone(krp);
1713 
1714 		memset(rp->rpr_msgin.dma_vaddr, 0, rp->rpr_msgin.dma_size);
1715 		memset(rp->rpr_msgout.dma_vaddr, 0, rp->rpr_msgout.dma_size);
1716 		memset(rp->rpr_q.q_ctx.dma_vaddr, 0, rp->rpr_q.q_ctx.dma_size);
1717 
1718 		/* Can't free here, so put us on the free list. */
1719 		SIMPLEQ_INSERT_TAIL(&sc->sc_q2free, &rp->rpr_q, q_next);
1720 		break;
1721 	}
1722 	default:
1723 		printf("%s: unknown ctx op: %x\n", device_xname(sc->sc_dev),
1724 		    letoh16(ctx->ctx_op));
1725 		break;
1726 	}
1727 }
1728 
1729 #ifndef UBSEC_NO_RNG
1730 static void
1731 ubsec_rng(void *vsc)
1732 {
1733 	struct ubsec_softc *sc = vsc;
1734 	struct ubsec_q2_rng *rng = &sc->sc_rng;
1735 	struct ubsec_mcr *mcr;
1736 	struct ubsec_ctx_rngbypass *ctx;
1737 	int s;
1738 
1739 	s = splnet();
1740 	if (rng->rng_used) {
1741 		splx(s);
1742 		return;
1743 	}
1744 	sc->sc_nqueue2++;
1745 	if (sc->sc_nqueue2 >= UBS_MAX_NQUEUE)
1746 		goto out;
1747 
1748 	mcr = (struct ubsec_mcr *)rng->rng_q.q_mcr.dma_vaddr;
1749 	ctx = (struct ubsec_ctx_rngbypass *)rng->rng_q.q_ctx.dma_vaddr;
1750 
1751 	mcr->mcr_pkts = htole16(1);
1752 	mcr->mcr_flags = 0;
1753 	mcr->mcr_cmdctxp = htole32(rng->rng_q.q_ctx.dma_paddr);
1754 	mcr->mcr_ipktbuf.pb_addr = mcr->mcr_ipktbuf.pb_next = 0;
1755 	mcr->mcr_ipktbuf.pb_len = 0;
1756 	mcr->mcr_reserved = mcr->mcr_pktlen = 0;
1757 	mcr->mcr_opktbuf.pb_addr = htole32(rng->rng_buf.dma_paddr);
1758 	mcr->mcr_opktbuf.pb_len = htole32(((sizeof(u_int32_t) * UBSEC_RNG_BUFSIZ)) &
1759 	    UBS_PKTBUF_LEN);
1760 	mcr->mcr_opktbuf.pb_next = 0;
1761 
1762 	ctx->rbp_len = htole16(sizeof(struct ubsec_ctx_rngbypass));
1763 	ctx->rbp_op = htole16(UBS_CTXOP_RNGSHA1);
1764 	rng->rng_q.q_type = UBS_CTXOP_RNGSHA1;
1765 
1766 	bus_dmamap_sync(sc->sc_dmat, rng->rng_buf.dma_map, 0,
1767 	    rng->rng_buf.dma_map->dm_mapsize, BUS_DMASYNC_PREREAD);
1768 
1769 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue2, &rng->rng_q, q_next);
1770 	rng->rng_used = 1;
1771 	ubsec_feed2(sc);
1772 	ubsecstats.hst_rng++;
1773 	splx(s);
1774 
1775 	return;
1776 
1777 out:
1778 	/*
1779 	 * Something weird happened, generate our own call back.
1780 	 */
1781 	sc->sc_nqueue2--;
1782 	splx(s);
1783 #ifdef __OpenBSD__
1784 	timeout_add(&sc->sc_rngto, sc->sc_rnghz);
1785 #else
1786 	callout_reset(&sc->sc_rngto, sc->sc_rnghz, ubsec_rng, sc);
1787 #endif
1788 }
1789 #endif /* UBSEC_NO_RNG */
1790 
1791 static int
1792 ubsec_dma_malloc(struct ubsec_softc *sc, bus_size_t size,
1793 		 struct ubsec_dma_alloc *dma,int mapflags)
1794 {
1795 	int r;
1796 
1797 	if ((r = bus_dmamem_alloc(sc->sc_dmat, size, PAGE_SIZE, 0,
1798 	    &dma->dma_seg, 1, &dma->dma_nseg, BUS_DMA_NOWAIT)) != 0)
1799 		goto fail_0;
1800 
1801 	if ((r = bus_dmamem_map(sc->sc_dmat, &dma->dma_seg, dma->dma_nseg,
1802 	    size, &dma->dma_vaddr, mapflags | BUS_DMA_NOWAIT)) != 0)
1803 		goto fail_1;
1804 
1805 	if ((r = bus_dmamap_create(sc->sc_dmat, size, 1, size, 0,
1806 	    BUS_DMA_NOWAIT, &dma->dma_map)) != 0)
1807 		goto fail_2;
1808 
1809 	if ((r = bus_dmamap_load(sc->sc_dmat, dma->dma_map, dma->dma_vaddr,
1810 	    size, NULL, BUS_DMA_NOWAIT)) != 0)
1811 		goto fail_3;
1812 
1813 	dma->dma_paddr = dma->dma_map->dm_segs[0].ds_addr;
1814 	dma->dma_size = size;
1815 	return (0);
1816 
1817 fail_3:
1818 	bus_dmamap_destroy(sc->sc_dmat, dma->dma_map);
1819 fail_2:
1820 	bus_dmamem_unmap(sc->sc_dmat, dma->dma_vaddr, size);
1821 fail_1:
1822 	bus_dmamem_free(sc->sc_dmat, &dma->dma_seg, dma->dma_nseg);
1823 fail_0:
1824 	dma->dma_map = NULL;
1825 	return (r);
1826 }
1827 
1828 static void
1829 ubsec_dma_free(struct ubsec_softc *sc, struct ubsec_dma_alloc *dma)
1830 {
1831 	bus_dmamap_unload(sc->sc_dmat, dma->dma_map);
1832 	bus_dmamem_unmap(sc->sc_dmat, dma->dma_vaddr, dma->dma_size);
1833 	bus_dmamem_free(sc->sc_dmat, &dma->dma_seg, dma->dma_nseg);
1834 	bus_dmamap_destroy(sc->sc_dmat, dma->dma_map);
1835 }
1836 
1837 /*
1838  * Resets the board.  Values in the regesters are left as is
1839  * from the reset (i.e. initial values are assigned elsewhere).
1840  */
1841 static void
1842 ubsec_reset_board(struct ubsec_softc *sc)
1843 {
1844     volatile u_int32_t ctrl;
1845 
1846     ctrl = READ_REG(sc, BS_CTRL);
1847     ctrl |= BS_CTRL_RESET;
1848     WRITE_REG(sc, BS_CTRL, ctrl);
1849 
1850     /*
1851      * Wait aprox. 30 PCI clocks = 900 ns = 0.9 us
1852      */
1853     DELAY(10);
1854 }
1855 
1856 /*
1857  * Init Broadcom registers
1858  */
1859 static void
1860 ubsec_init_board(struct ubsec_softc *sc)
1861 {
1862 	u_int32_t ctrl;
1863 
1864 	ctrl = READ_REG(sc, BS_CTRL);
1865 	ctrl &= ~(BS_CTRL_BE32 | BS_CTRL_BE64);
1866 	ctrl |= BS_CTRL_LITTLE_ENDIAN | BS_CTRL_MCR1INT;
1867 
1868 	/*
1869 	 * XXX: Sam Leffler's code has (UBS_FLAGS_KEY|UBS_FLAGS_RNG)).
1870 	 * anyone got hw docs?
1871 	 */
1872 	if (sc->sc_flags & UBS_FLAGS_KEY)
1873 		ctrl |= BS_CTRL_MCR2INT;
1874 	else
1875 		ctrl &= ~BS_CTRL_MCR2INT;
1876 
1877 	if (sc->sc_flags & UBS_FLAGS_HWNORM)
1878 		ctrl &= ~BS_CTRL_SWNORM;
1879 
1880 	WRITE_REG(sc, BS_CTRL, ctrl);
1881 }
1882 
1883 /*
1884  * Init Broadcom PCI registers
1885  */
1886 static void
1887 ubsec_init_pciregs(struct pci_attach_args *pa)
1888 {
1889 	pci_chipset_tag_t pc = pa->pa_pc;
1890 	u_int32_t misc;
1891 
1892 	/*
1893 	 * This will set the cache line size to 1, this will
1894 	 * force the BCM58xx chip just to do burst read/writes.
1895 	 * Cache line read/writes are to slow
1896 	 */
1897 	misc = pci_conf_read(pc, pa->pa_tag, PCI_BHLC_REG);
1898 	misc = (misc & ~(PCI_CACHELINE_MASK << PCI_CACHELINE_SHIFT))
1899 	    | ((UBS_DEF_CACHELINE & 0xff) << PCI_CACHELINE_SHIFT);
1900 	pci_conf_write(pc, pa->pa_tag, PCI_BHLC_REG, misc);
1901 }
1902 
1903 /*
1904  * Clean up after a chip crash.
1905  * It is assumed that the caller in splnet()
1906  */
1907 static void
1908 ubsec_cleanchip(struct ubsec_softc *sc)
1909 {
1910 	struct ubsec_q *q;
1911 
1912 	while (!SIMPLEQ_EMPTY(&sc->sc_qchip)) {
1913 		q = SIMPLEQ_FIRST(&sc->sc_qchip);
1914 		SIMPLEQ_REMOVE_HEAD(&sc->sc_qchip, /*q,*/ q_next);
1915 		ubsec_free_q(sc, q);
1916 	}
1917 	sc->sc_nqchip = 0;
1918 }
1919 
1920 /*
1921  * free a ubsec_q
1922  * It is assumed that the caller is within splnet()
1923  */
1924 static int
1925 ubsec_free_q(struct ubsec_softc *sc, struct ubsec_q *q)
1926 {
1927 	struct ubsec_q *q2;
1928 	struct cryptop *crp;
1929 	int npkts;
1930 	int i;
1931 
1932 	npkts = q->q_nstacked_mcrs;
1933 
1934 	for (i = 0; i < npkts; i++) {
1935 		if(q->q_stacked_mcr[i]) {
1936 			q2 = q->q_stacked_mcr[i];
1937 
1938 			if ((q2->q_dst_m != NULL) && (q2->q_src_m != q2->q_dst_m))
1939 				m_freem(q2->q_dst_m);
1940 
1941 			crp = (struct cryptop *)q2->q_crp;
1942 
1943 			SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q2, q_next);
1944 
1945 			crp->crp_etype = EFAULT;
1946 			crypto_done(crp);
1947 		} else {
1948 			break;
1949 		}
1950 	}
1951 
1952 	/*
1953 	 * Free header MCR
1954 	 */
1955 	if ((q->q_dst_m != NULL) && (q->q_src_m != q->q_dst_m))
1956 		m_freem(q->q_dst_m);
1957 
1958 	crp = (struct cryptop *)q->q_crp;
1959 
1960 	SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
1961 
1962 	crp->crp_etype = EFAULT;
1963 	crypto_done(crp);
1964 	return(0);
1965 }
1966 
1967 /*
1968  * Routine to reset the chip and clean up.
1969  * It is assumed that the caller is in splnet()
1970  */
1971 static void
1972 ubsec_totalreset(struct ubsec_softc *sc)
1973 {
1974 	ubsec_reset_board(sc);
1975 	ubsec_init_board(sc);
1976 	ubsec_cleanchip(sc);
1977 }
1978 
1979 static int
1980 ubsec_dmamap_aligned(bus_dmamap_t map)
1981 {
1982 	int i;
1983 
1984 	for (i = 0; i < map->dm_nsegs; i++) {
1985 		if (map->dm_segs[i].ds_addr & 3)
1986 			return (0);
1987 		if ((i != (map->dm_nsegs - 1)) &&
1988 		    (map->dm_segs[i].ds_len & 3))
1989 			return (0);
1990 	}
1991 	return (1);
1992 }
1993 
1994 #ifdef __OpenBSD__
1995 struct ubsec_softc *
1996 ubsec_kfind(struct cryptkop *krp)
1997 {
1998 	struct ubsec_softc *sc;
1999 	int i;
2000 
2001 	for (i = 0; i < ubsec_cd.cd_ndevs; i++) {
2002 		sc = ubsec_cd.cd_devs[i];
2003 		if (sc == NULL)
2004 			continue;
2005 		if (sc->sc_cid == krp->krp_hid)
2006 			return (sc);
2007 	}
2008 	return (NULL);
2009 }
2010 #endif
2011 
2012 static void
2013 ubsec_kfree(struct ubsec_softc *sc, struct ubsec_q2 *q)
2014 {
2015 	switch (q->q_type) {
2016 	case UBS_CTXOP_MODEXP: {
2017 		struct ubsec_q2_modexp *me = (struct ubsec_q2_modexp *)q;
2018 
2019 		ubsec_dma_free(sc, &me->me_q.q_mcr);
2020 		ubsec_dma_free(sc, &me->me_q.q_ctx);
2021 		ubsec_dma_free(sc, &me->me_M);
2022 		ubsec_dma_free(sc, &me->me_E);
2023 		ubsec_dma_free(sc, &me->me_C);
2024 		ubsec_dma_free(sc, &me->me_epb);
2025 		free(me, M_DEVBUF);
2026 		break;
2027 	}
2028 	case UBS_CTXOP_RSAPRIV: {
2029 		struct ubsec_q2_rsapriv *rp = (struct ubsec_q2_rsapriv *)q;
2030 
2031 		ubsec_dma_free(sc, &rp->rpr_q.q_mcr);
2032 		ubsec_dma_free(sc, &rp->rpr_q.q_ctx);
2033 		ubsec_dma_free(sc, &rp->rpr_msgin);
2034 		ubsec_dma_free(sc, &rp->rpr_msgout);
2035 		free(rp, M_DEVBUF);
2036 		break;
2037 	}
2038 	default:
2039 		printf("%s: invalid kfree 0x%x\n", device_xname(sc->sc_dev),
2040 		    q->q_type);
2041 		break;
2042 	}
2043 }
2044 
2045 static int
2046 ubsec_kprocess(void *arg, struct cryptkop *krp, int hint)
2047 {
2048 	struct ubsec_softc *sc;
2049 	int r;
2050 
2051 	if (krp == NULL || krp->krp_callback == NULL)
2052 		return (EINVAL);
2053 #ifdef __OpenBSD__
2054 	if ((sc = ubsec_kfind(krp)) == NULL)
2055 		return (EINVAL);
2056 #else
2057 	sc = arg;
2058 	KASSERT(sc != NULL /*, ("ubsec_kprocess: null softc")*/);
2059 #endif
2060 
2061 	while (!SIMPLEQ_EMPTY(&sc->sc_q2free)) {
2062 		struct ubsec_q2 *q;
2063 
2064 		q = SIMPLEQ_FIRST(&sc->sc_q2free);
2065 		SIMPLEQ_REMOVE_HEAD(&sc->sc_q2free, /*q,*/ q_next);
2066 		ubsec_kfree(sc, q);
2067 	}
2068 
2069 	switch (krp->krp_op) {
2070 	case CRK_MOD_EXP:
2071 		if (sc->sc_flags & UBS_FLAGS_HWNORM)
2072 			r = ubsec_kprocess_modexp_hw(sc, krp, hint);
2073 		else
2074 			r = ubsec_kprocess_modexp_sw(sc, krp, hint);
2075 		break;
2076 	case CRK_MOD_EXP_CRT:
2077 		r = ubsec_kprocess_rsapriv(sc, krp, hint);
2078 		break;
2079 	default:
2080 		printf("%s: kprocess: invalid op 0x%x\n",
2081 		    device_xname(sc->sc_dev), krp->krp_op);
2082 		krp->krp_status = EOPNOTSUPP;
2083 		crypto_kdone(krp);
2084 		r = 0;
2085 	}
2086 	return (r);
2087 }
2088 
2089 /*
2090  * Start computation of cr[C] = (cr[M] ^ cr[E]) mod cr[N] (sw normalization)
2091  */
2092 static int
2093 ubsec_kprocess_modexp_sw(struct ubsec_softc *sc, struct cryptkop *krp,
2094 			 int hint)
2095 {
2096 	struct ubsec_q2_modexp *me;
2097 	struct ubsec_mcr *mcr;
2098 	struct ubsec_ctx_modexp *ctx;
2099 	struct ubsec_pktbuf *epb;
2100 	int s, err = 0;
2101 	u_int nbits, normbits, mbits, shiftbits, ebits;
2102 
2103 	me = (struct ubsec_q2_modexp *)malloc(sizeof *me, M_DEVBUF, M_NOWAIT);
2104 	if (me == NULL) {
2105 		err = ENOMEM;
2106 		goto errout;
2107 	}
2108 	memset(me, 0, sizeof *me);
2109 	me->me_krp = krp;
2110 	me->me_q.q_type = UBS_CTXOP_MODEXP;
2111 
2112 	nbits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_N]);
2113 	if (nbits <= 512)
2114 		normbits = 512;
2115 	else if (nbits <= 768)
2116 		normbits = 768;
2117 	else if (nbits <= 1024)
2118 		normbits = 1024;
2119 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && nbits <= 1536)
2120 		normbits = 1536;
2121 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && nbits <= 2048)
2122 		normbits = 2048;
2123 	else {
2124 		err = E2BIG;
2125 		goto errout;
2126 	}
2127 
2128 	shiftbits = normbits - nbits;
2129 
2130 	me->me_modbits = nbits;
2131 	me->me_shiftbits = shiftbits;
2132 	me->me_normbits = normbits;
2133 
2134 	/* Sanity check: result bits must be >= true modulus bits. */
2135 	if (krp->krp_param[krp->krp_iparams].crp_nbits < nbits) {
2136 		err = ERANGE;
2137 		goto errout;
2138 	}
2139 
2140 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_mcr),
2141 	    &me->me_q.q_mcr, 0)) {
2142 		err = ENOMEM;
2143 		goto errout;
2144 	}
2145 	mcr = (struct ubsec_mcr *)me->me_q.q_mcr.dma_vaddr;
2146 
2147 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_ctx_modexp),
2148 	    &me->me_q.q_ctx, 0)) {
2149 		err = ENOMEM;
2150 		goto errout;
2151 	}
2152 
2153 	mbits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_M]);
2154 	if (mbits > nbits) {
2155 		err = E2BIG;
2156 		goto errout;
2157 	}
2158 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_M, 0)) {
2159 		err = ENOMEM;
2160 		goto errout;
2161 	}
2162 	ubsec_kshift_r(shiftbits,
2163 	    krp->krp_param[UBS_MODEXP_PAR_M].crp_p, mbits,
2164 	    me->me_M.dma_vaddr, normbits);
2165 
2166 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_C, 0)) {
2167 		err = ENOMEM;
2168 		goto errout;
2169 	}
2170 	memset(me->me_C.dma_vaddr, 0, me->me_C.dma_size);
2171 
2172 	ebits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_E]);
2173 	if (ebits > nbits) {
2174 		err = E2BIG;
2175 		goto errout;
2176 	}
2177 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_E, 0)) {
2178 		err = ENOMEM;
2179 		goto errout;
2180 	}
2181 	ubsec_kshift_r(shiftbits,
2182 	    krp->krp_param[UBS_MODEXP_PAR_E].crp_p, ebits,
2183 	    me->me_E.dma_vaddr, normbits);
2184 
2185 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_pktbuf),
2186 	    &me->me_epb, 0)) {
2187 		err = ENOMEM;
2188 		goto errout;
2189 	}
2190 	epb = (struct ubsec_pktbuf *)me->me_epb.dma_vaddr;
2191 	epb->pb_addr = htole32(me->me_E.dma_paddr);
2192 	epb->pb_next = 0;
2193 	epb->pb_len = htole32(normbits / 8);
2194 
2195 #ifdef UBSEC_DEBUG
2196 	if (ubsec_debug) {
2197 		printf("Epb ");
2198 		ubsec_dump_pb(epb);
2199 	}
2200 #endif
2201 
2202 	mcr->mcr_pkts = htole16(1);
2203 	mcr->mcr_flags = 0;
2204 	mcr->mcr_cmdctxp = htole32(me->me_q.q_ctx.dma_paddr);
2205 	mcr->mcr_reserved = 0;
2206 	mcr->mcr_pktlen = 0;
2207 
2208 	mcr->mcr_ipktbuf.pb_addr = htole32(me->me_M.dma_paddr);
2209 	mcr->mcr_ipktbuf.pb_len = htole32(normbits / 8);
2210 	mcr->mcr_ipktbuf.pb_next = htole32(me->me_epb.dma_paddr);
2211 
2212 	mcr->mcr_opktbuf.pb_addr = htole32(me->me_C.dma_paddr);
2213 	mcr->mcr_opktbuf.pb_next = 0;
2214 	mcr->mcr_opktbuf.pb_len = htole32(normbits / 8);
2215 
2216 #ifdef DIAGNOSTIC
2217 	/* Misaligned output buffer will hang the chip. */
2218 	if ((letoh32(mcr->mcr_opktbuf.pb_addr) & 3) != 0)
2219 		panic("%s: modexp invalid addr 0x%x",
2220 		    device_xname(sc->sc_dev), letoh32(mcr->mcr_opktbuf.pb_addr));
2221 	if ((letoh32(mcr->mcr_opktbuf.pb_len) & 3) != 0)
2222 		panic("%s: modexp invalid len 0x%x",
2223 		    device_xname(sc->sc_dev), letoh32(mcr->mcr_opktbuf.pb_len));
2224 #endif
2225 
2226 	ctx = (struct ubsec_ctx_modexp *)me->me_q.q_ctx.dma_vaddr;
2227 	memset(ctx, 0, sizeof(*ctx));
2228 	ubsec_kshift_r(shiftbits,
2229 	    krp->krp_param[UBS_MODEXP_PAR_N].crp_p, nbits,
2230 	    ctx->me_N, normbits);
2231 	ctx->me_len = htole16((normbits / 8) + (4 * sizeof(u_int16_t)));
2232 	ctx->me_op = htole16(UBS_CTXOP_MODEXP);
2233 	ctx->me_E_len = htole16(nbits);
2234 	ctx->me_N_len = htole16(nbits);
2235 
2236 #ifdef UBSEC_DEBUG
2237 	if (ubsec_debug) {
2238 		ubsec_dump_mcr(mcr);
2239 		ubsec_dump_ctx2((struct ubsec_ctx_keyop *)ctx);
2240 	}
2241 #endif
2242 
2243 	/*
2244 	 * ubsec_feed2 will sync mcr and ctx, we just need to sync
2245 	 * everything else.
2246 	 */
2247 	bus_dmamap_sync(sc->sc_dmat, me->me_M.dma_map,
2248 	    0, me->me_M.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2249 	bus_dmamap_sync(sc->sc_dmat, me->me_E.dma_map,
2250 	    0, me->me_E.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2251 	bus_dmamap_sync(sc->sc_dmat, me->me_C.dma_map,
2252 	    0, me->me_C.dma_map->dm_mapsize, BUS_DMASYNC_PREREAD);
2253 	bus_dmamap_sync(sc->sc_dmat, me->me_epb.dma_map,
2254 	    0, me->me_epb.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2255 
2256 	/* Enqueue and we're done... */
2257 	s = splnet();
2258 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue2, &me->me_q, q_next);
2259 	ubsec_feed2(sc);
2260 	ubsecstats.hst_modexp++;
2261 	splx(s);
2262 
2263 	return (0);
2264 
2265 errout:
2266 	if (me != NULL) {
2267 		if (me->me_q.q_mcr.dma_map != NULL)
2268 			ubsec_dma_free(sc, &me->me_q.q_mcr);
2269 		if (me->me_q.q_ctx.dma_map != NULL) {
2270 			memset(me->me_q.q_ctx.dma_vaddr, 0, me->me_q.q_ctx.dma_size);
2271 			ubsec_dma_free(sc, &me->me_q.q_ctx);
2272 		}
2273 		if (me->me_M.dma_map != NULL) {
2274 			memset(me->me_M.dma_vaddr, 0, me->me_M.dma_size);
2275 			ubsec_dma_free(sc, &me->me_M);
2276 		}
2277 		if (me->me_E.dma_map != NULL) {
2278 			memset(me->me_E.dma_vaddr, 0, me->me_E.dma_size);
2279 			ubsec_dma_free(sc, &me->me_E);
2280 		}
2281 		if (me->me_C.dma_map != NULL) {
2282 			memset(me->me_C.dma_vaddr, 0, me->me_C.dma_size);
2283 			ubsec_dma_free(sc, &me->me_C);
2284 		}
2285 		if (me->me_epb.dma_map != NULL)
2286 			ubsec_dma_free(sc, &me->me_epb);
2287 		free(me, M_DEVBUF);
2288 	}
2289 	krp->krp_status = err;
2290 	crypto_kdone(krp);
2291 	return (0);
2292 }
2293 
2294 /*
2295  * Start computation of cr[C] = (cr[M] ^ cr[E]) mod cr[N] (hw normalization)
2296  */
2297 static int
2298 ubsec_kprocess_modexp_hw(struct ubsec_softc *sc, struct cryptkop *krp,
2299 			 int hint)
2300 {
2301 	struct ubsec_q2_modexp *me;
2302 	struct ubsec_mcr *mcr;
2303 	struct ubsec_ctx_modexp *ctx;
2304 	struct ubsec_pktbuf *epb;
2305 	int s, err = 0;
2306 	u_int nbits, normbits, mbits, shiftbits, ebits;
2307 
2308 	me = (struct ubsec_q2_modexp *)malloc(sizeof *me, M_DEVBUF, M_NOWAIT);
2309 	if (me == NULL) {
2310 		err = ENOMEM;
2311 		goto errout;
2312 	}
2313 	memset(me, 0, sizeof *me);
2314 	me->me_krp = krp;
2315 	me->me_q.q_type = UBS_CTXOP_MODEXP;
2316 
2317 	nbits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_N]);
2318 	if (nbits <= 512)
2319 		normbits = 512;
2320 	else if (nbits <= 768)
2321 		normbits = 768;
2322 	else if (nbits <= 1024)
2323 		normbits = 1024;
2324 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && nbits <= 1536)
2325 		normbits = 1536;
2326 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && nbits <= 2048)
2327 		normbits = 2048;
2328 	else {
2329 		err = E2BIG;
2330 		goto errout;
2331 	}
2332 
2333 	shiftbits = normbits - nbits;
2334 
2335 	/* XXX ??? */
2336 	me->me_modbits = nbits;
2337 	me->me_shiftbits = shiftbits;
2338 	me->me_normbits = normbits;
2339 
2340 	/* Sanity check: result bits must be >= true modulus bits. */
2341 	if (krp->krp_param[krp->krp_iparams].crp_nbits < nbits) {
2342 		err = ERANGE;
2343 		goto errout;
2344 	}
2345 
2346 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_mcr),
2347 	    &me->me_q.q_mcr, 0)) {
2348 		err = ENOMEM;
2349 		goto errout;
2350 	}
2351 	mcr = (struct ubsec_mcr *)me->me_q.q_mcr.dma_vaddr;
2352 
2353 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_ctx_modexp),
2354 	    &me->me_q.q_ctx, 0)) {
2355 		err = ENOMEM;
2356 		goto errout;
2357 	}
2358 
2359 	mbits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_M]);
2360 	if (mbits > nbits) {
2361 		err = E2BIG;
2362 		goto errout;
2363 	}
2364 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_M, 0)) {
2365 		err = ENOMEM;
2366 		goto errout;
2367 	}
2368 	memset(me->me_M.dma_vaddr, 0, normbits / 8);
2369 	bcopy(krp->krp_param[UBS_MODEXP_PAR_M].crp_p,
2370 	    me->me_M.dma_vaddr, (mbits + 7) / 8);
2371 
2372 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_C, 0)) {
2373 		err = ENOMEM;
2374 		goto errout;
2375 	}
2376 	memset(me->me_C.dma_vaddr, 0, me->me_C.dma_size);
2377 
2378 	ebits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_E]);
2379 	if (ebits > nbits) {
2380 		err = E2BIG;
2381 		goto errout;
2382 	}
2383 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_E, 0)) {
2384 		err = ENOMEM;
2385 		goto errout;
2386 	}
2387 	memset(me->me_E.dma_vaddr, 0, normbits / 8);
2388 	bcopy(krp->krp_param[UBS_MODEXP_PAR_E].crp_p,
2389 	    me->me_E.dma_vaddr, (ebits + 7) / 8);
2390 
2391 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_pktbuf),
2392 	    &me->me_epb, 0)) {
2393 		err = ENOMEM;
2394 		goto errout;
2395 	}
2396 	epb = (struct ubsec_pktbuf *)me->me_epb.dma_vaddr;
2397 	epb->pb_addr = htole32(me->me_E.dma_paddr);
2398 	epb->pb_next = 0;
2399 	epb->pb_len = htole32((ebits + 7) / 8);
2400 
2401 #ifdef UBSEC_DEBUG
2402 	if (ubsec_debug) {
2403 		printf("Epb ");
2404 		ubsec_dump_pb(epb);
2405 	}
2406 #endif
2407 
2408 	mcr->mcr_pkts = htole16(1);
2409 	mcr->mcr_flags = 0;
2410 	mcr->mcr_cmdctxp = htole32(me->me_q.q_ctx.dma_paddr);
2411 	mcr->mcr_reserved = 0;
2412 	mcr->mcr_pktlen = 0;
2413 
2414 	mcr->mcr_ipktbuf.pb_addr = htole32(me->me_M.dma_paddr);
2415 	mcr->mcr_ipktbuf.pb_len = htole32(normbits / 8);
2416 	mcr->mcr_ipktbuf.pb_next = htole32(me->me_epb.dma_paddr);
2417 
2418 	mcr->mcr_opktbuf.pb_addr = htole32(me->me_C.dma_paddr);
2419 	mcr->mcr_opktbuf.pb_next = 0;
2420 	mcr->mcr_opktbuf.pb_len = htole32(normbits / 8);
2421 
2422 #ifdef DIAGNOSTIC
2423 	/* Misaligned output buffer will hang the chip. */
2424 	if ((letoh32(mcr->mcr_opktbuf.pb_addr) & 3) != 0)
2425 		panic("%s: modexp invalid addr 0x%x",
2426 		    device_xname(sc->sc_dev), letoh32(mcr->mcr_opktbuf.pb_addr));
2427 	if ((letoh32(mcr->mcr_opktbuf.pb_len) & 3) != 0)
2428 		panic("%s: modexp invalid len 0x%x",
2429 		    device_xname(sc->sc_dev), letoh32(mcr->mcr_opktbuf.pb_len));
2430 #endif
2431 
2432 	ctx = (struct ubsec_ctx_modexp *)me->me_q.q_ctx.dma_vaddr;
2433 	memset(ctx, 0, sizeof(*ctx));
2434 	memcpy(ctx->me_N, krp->krp_param[UBS_MODEXP_PAR_N].crp_p,
2435 	    (nbits + 7) / 8);
2436 	ctx->me_len = htole16((normbits / 8) + (4 * sizeof(u_int16_t)));
2437 	ctx->me_op = htole16(UBS_CTXOP_MODEXP);
2438 	ctx->me_E_len = htole16(ebits);
2439 	ctx->me_N_len = htole16(nbits);
2440 
2441 #ifdef UBSEC_DEBUG
2442 	if (ubsec_debug) {
2443 		ubsec_dump_mcr(mcr);
2444 		ubsec_dump_ctx2((struct ubsec_ctx_keyop *)ctx);
2445 	}
2446 #endif
2447 
2448 	/*
2449 	 * ubsec_feed2 will sync mcr and ctx, we just need to sync
2450 	 * everything else.
2451 	 */
2452 	bus_dmamap_sync(sc->sc_dmat, me->me_M.dma_map,
2453 	    0, me->me_M.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2454 	bus_dmamap_sync(sc->sc_dmat, me->me_E.dma_map,
2455 	    0, me->me_E.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2456 	bus_dmamap_sync(sc->sc_dmat, me->me_C.dma_map,
2457 	    0, me->me_C.dma_map->dm_mapsize, BUS_DMASYNC_PREREAD);
2458 	bus_dmamap_sync(sc->sc_dmat, me->me_epb.dma_map,
2459 	    0, me->me_epb.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2460 
2461 	/* Enqueue and we're done... */
2462 	s = splnet();
2463 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue2, &me->me_q, q_next);
2464 	ubsec_feed2(sc);
2465 	splx(s);
2466 
2467 	return (0);
2468 
2469 errout:
2470 	if (me != NULL) {
2471 		if (me->me_q.q_mcr.dma_map != NULL)
2472 			ubsec_dma_free(sc, &me->me_q.q_mcr);
2473 		if (me->me_q.q_ctx.dma_map != NULL) {
2474 			memset(me->me_q.q_ctx.dma_vaddr, 0, me->me_q.q_ctx.dma_size);
2475 			ubsec_dma_free(sc, &me->me_q.q_ctx);
2476 		}
2477 		if (me->me_M.dma_map != NULL) {
2478 			memset(me->me_M.dma_vaddr, 0, me->me_M.dma_size);
2479 			ubsec_dma_free(sc, &me->me_M);
2480 		}
2481 		if (me->me_E.dma_map != NULL) {
2482 			memset(me->me_E.dma_vaddr, 0, me->me_E.dma_size);
2483 			ubsec_dma_free(sc, &me->me_E);
2484 		}
2485 		if (me->me_C.dma_map != NULL) {
2486 			memset(me->me_C.dma_vaddr, 0, me->me_C.dma_size);
2487 			ubsec_dma_free(sc, &me->me_C);
2488 		}
2489 		if (me->me_epb.dma_map != NULL)
2490 			ubsec_dma_free(sc, &me->me_epb);
2491 		free(me, M_DEVBUF);
2492 	}
2493 	krp->krp_status = err;
2494 	crypto_kdone(krp);
2495 	return (0);
2496 }
2497 
2498 static int
2499 ubsec_kprocess_rsapriv(struct ubsec_softc *sc, struct cryptkop *krp,
2500 		       int hint)
2501 {
2502 	struct ubsec_q2_rsapriv *rp = NULL;
2503 	struct ubsec_mcr *mcr;
2504 	struct ubsec_ctx_rsapriv *ctx;
2505 	int s, err = 0;
2506 	u_int padlen, msglen;
2507 
2508 	msglen = ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_P]);
2509 	padlen = ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_Q]);
2510 	if (msglen > padlen)
2511 		padlen = msglen;
2512 
2513 	if (padlen <= 256)
2514 		padlen = 256;
2515 	else if (padlen <= 384)
2516 		padlen = 384;
2517 	else if (padlen <= 512)
2518 		padlen = 512;
2519 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && padlen <= 768)
2520 		padlen = 768;
2521 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && padlen <= 1024)
2522 		padlen = 1024;
2523 	else {
2524 		err = E2BIG;
2525 		goto errout;
2526 	}
2527 
2528 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_DP]) > padlen) {
2529 		err = E2BIG;
2530 		goto errout;
2531 	}
2532 
2533 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_DQ]) > padlen) {
2534 		err = E2BIG;
2535 		goto errout;
2536 	}
2537 
2538 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_PINV]) > padlen) {
2539 		err = E2BIG;
2540 		goto errout;
2541 	}
2542 
2543 	rp = malloc(sizeof *rp, M_DEVBUF, M_NOWAIT|M_ZERO);
2544 	if (rp == NULL)
2545 		return (ENOMEM);
2546 	rp->rpr_krp = krp;
2547 	rp->rpr_q.q_type = UBS_CTXOP_RSAPRIV;
2548 
2549 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_mcr),
2550 	    &rp->rpr_q.q_mcr, 0)) {
2551 		err = ENOMEM;
2552 		goto errout;
2553 	}
2554 	mcr = (struct ubsec_mcr *)rp->rpr_q.q_mcr.dma_vaddr;
2555 
2556 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_ctx_rsapriv),
2557 	    &rp->rpr_q.q_ctx, 0)) {
2558 		err = ENOMEM;
2559 		goto errout;
2560 	}
2561 	ctx = (struct ubsec_ctx_rsapriv *)rp->rpr_q.q_ctx.dma_vaddr;
2562 	memset(ctx, 0, sizeof *ctx);
2563 
2564 	/* Copy in p */
2565 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_P].crp_p,
2566 	    &ctx->rpr_buf[0 * (padlen / 8)],
2567 	    (krp->krp_param[UBS_RSAPRIV_PAR_P].crp_nbits + 7) / 8);
2568 
2569 	/* Copy in q */
2570 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_Q].crp_p,
2571 	    &ctx->rpr_buf[1 * (padlen / 8)],
2572 	    (krp->krp_param[UBS_RSAPRIV_PAR_Q].crp_nbits + 7) / 8);
2573 
2574 	/* Copy in dp */
2575 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_DP].crp_p,
2576 	    &ctx->rpr_buf[2 * (padlen / 8)],
2577 	    (krp->krp_param[UBS_RSAPRIV_PAR_DP].crp_nbits + 7) / 8);
2578 
2579 	/* Copy in dq */
2580 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_DQ].crp_p,
2581 	    &ctx->rpr_buf[3 * (padlen / 8)],
2582 	    (krp->krp_param[UBS_RSAPRIV_PAR_DQ].crp_nbits + 7) / 8);
2583 
2584 	/* Copy in pinv */
2585 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_PINV].crp_p,
2586 	    &ctx->rpr_buf[4 * (padlen / 8)],
2587 	    (krp->krp_param[UBS_RSAPRIV_PAR_PINV].crp_nbits + 7) / 8);
2588 
2589 	msglen = padlen * 2;
2590 
2591 	/* Copy in input message (aligned buffer/length). */
2592 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_MSGIN]) > msglen) {
2593 		/* Is this likely? */
2594 		err = E2BIG;
2595 		goto errout;
2596 	}
2597 	if (ubsec_dma_malloc(sc, (msglen + 7) / 8, &rp->rpr_msgin, 0)) {
2598 		err = ENOMEM;
2599 		goto errout;
2600 	}
2601 	memset(rp->rpr_msgin.dma_vaddr, 0, (msglen + 7) / 8);
2602 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_MSGIN].crp_p,
2603 	    rp->rpr_msgin.dma_vaddr,
2604 	    (krp->krp_param[UBS_RSAPRIV_PAR_MSGIN].crp_nbits + 7) / 8);
2605 
2606 	/* Prepare space for output message (aligned buffer/length). */
2607 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_MSGOUT]) < msglen) {
2608 		/* Is this likely? */
2609 		err = E2BIG;
2610 		goto errout;
2611 	}
2612 	if (ubsec_dma_malloc(sc, (msglen + 7) / 8, &rp->rpr_msgout, 0)) {
2613 		err = ENOMEM;
2614 		goto errout;
2615 	}
2616 	memset(rp->rpr_msgout.dma_vaddr, 0, (msglen + 7) / 8);
2617 
2618 	mcr->mcr_pkts = htole16(1);
2619 	mcr->mcr_flags = 0;
2620 	mcr->mcr_cmdctxp = htole32(rp->rpr_q.q_ctx.dma_paddr);
2621 	mcr->mcr_ipktbuf.pb_addr = htole32(rp->rpr_msgin.dma_paddr);
2622 	mcr->mcr_ipktbuf.pb_next = 0;
2623 	mcr->mcr_ipktbuf.pb_len = htole32(rp->rpr_msgin.dma_size);
2624 	mcr->mcr_reserved = 0;
2625 	mcr->mcr_pktlen = htole16(msglen);
2626 	mcr->mcr_opktbuf.pb_addr = htole32(rp->rpr_msgout.dma_paddr);
2627 	mcr->mcr_opktbuf.pb_next = 0;
2628 	mcr->mcr_opktbuf.pb_len = htole32(rp->rpr_msgout.dma_size);
2629 
2630 #ifdef DIAGNOSTIC
2631 	if (rp->rpr_msgin.dma_paddr & 3 || rp->rpr_msgin.dma_size & 3) {
2632 		panic("%s: rsapriv: invalid msgin 0x%lx(0x%lx)",
2633 		    device_xname(sc->sc_dev), (u_long) rp->rpr_msgin.dma_paddr,
2634 		    (u_long) rp->rpr_msgin.dma_size);
2635 	}
2636 	if (rp->rpr_msgout.dma_paddr & 3 || rp->rpr_msgout.dma_size & 3) {
2637 		panic("%s: rsapriv: invalid msgout 0x%lx(0x%lx)",
2638 		    device_xname(sc->sc_dev), (u_long) rp->rpr_msgout.dma_paddr,
2639 		    (u_long) rp->rpr_msgout.dma_size);
2640 	}
2641 #endif
2642 
2643 	ctx->rpr_len = (sizeof(u_int16_t) * 4) + (5 * (padlen / 8));
2644 	ctx->rpr_op = htole16(UBS_CTXOP_RSAPRIV);
2645 	ctx->rpr_q_len = htole16(padlen);
2646 	ctx->rpr_p_len = htole16(padlen);
2647 
2648 	/*
2649 	 * ubsec_feed2 will sync mcr and ctx, we just need to sync
2650 	 * everything else.
2651 	 */
2652 	bus_dmamap_sync(sc->sc_dmat, rp->rpr_msgin.dma_map,
2653 	    0, rp->rpr_msgin.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2654 	bus_dmamap_sync(sc->sc_dmat, rp->rpr_msgout.dma_map,
2655 	    0, rp->rpr_msgout.dma_map->dm_mapsize, BUS_DMASYNC_PREREAD);
2656 
2657 	/* Enqueue and we're done... */
2658 	s = splnet();
2659 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue2, &rp->rpr_q, q_next);
2660 	ubsec_feed2(sc);
2661 	ubsecstats.hst_modexpcrt++;
2662 	splx(s);
2663 	return (0);
2664 
2665 errout:
2666 	if (rp != NULL) {
2667 		if (rp->rpr_q.q_mcr.dma_map != NULL)
2668 			ubsec_dma_free(sc, &rp->rpr_q.q_mcr);
2669 		if (rp->rpr_msgin.dma_map != NULL) {
2670 			memset(rp->rpr_msgin.dma_vaddr, 0, rp->rpr_msgin.dma_size);
2671 			ubsec_dma_free(sc, &rp->rpr_msgin);
2672 		}
2673 		if (rp->rpr_msgout.dma_map != NULL) {
2674 			memset(rp->rpr_msgout.dma_vaddr, 0, rp->rpr_msgout.dma_size);
2675 			ubsec_dma_free(sc, &rp->rpr_msgout);
2676 		}
2677 		free(rp, M_DEVBUF);
2678 	}
2679 	krp->krp_status = err;
2680 	crypto_kdone(krp);
2681 	return (0);
2682 }
2683 
2684 #ifdef UBSEC_DEBUG
2685 static void
2686 ubsec_dump_pb(volatile struct ubsec_pktbuf *pb)
2687 {
2688 	printf("addr 0x%x (0x%x) next 0x%x\n",
2689 	    pb->pb_addr, pb->pb_len, pb->pb_next);
2690 }
2691 
2692 static void
2693 ubsec_dump_ctx2(volatile struct ubsec_ctx_keyop *c)
2694 {
2695 	printf("CTX (0x%x):\n", c->ctx_len);
2696 	switch (letoh16(c->ctx_op)) {
2697 	case UBS_CTXOP_RNGBYPASS:
2698 	case UBS_CTXOP_RNGSHA1:
2699 		break;
2700 	case UBS_CTXOP_MODEXP:
2701 	{
2702 		struct ubsec_ctx_modexp *cx = (void *)c;
2703 		int i, len;
2704 
2705 		printf(" Elen %u, Nlen %u\n",
2706 		    letoh16(cx->me_E_len), letoh16(cx->me_N_len));
2707 		len = (cx->me_N_len + 7)/8;
2708 		for (i = 0; i < len; i++)
2709 			printf("%s%02x", (i == 0) ? " N: " : ":", cx->me_N[i]);
2710 		printf("\n");
2711 		break;
2712 	}
2713 	default:
2714 		printf("unknown context: %x\n", c->ctx_op);
2715 	}
2716 	printf("END CTX\n");
2717 }
2718 
2719 static void
2720 ubsec_dump_mcr(struct ubsec_mcr *mcr)
2721 {
2722 	volatile struct ubsec_mcr_add *ma;
2723 	int i;
2724 
2725 	printf("MCR:\n");
2726 	printf(" pkts: %u, flags 0x%x\n",
2727 	    letoh16(mcr->mcr_pkts), letoh16(mcr->mcr_flags));
2728 	ma = (volatile struct ubsec_mcr_add *)&mcr->mcr_cmdctxp;
2729 	for (i = 0; i < letoh16(mcr->mcr_pkts); i++) {
2730 		printf(" %d: ctx 0x%x len 0x%x rsvd 0x%x\n", i,
2731 		    letoh32(ma->mcr_cmdctxp), letoh16(ma->mcr_pktlen),
2732 		    letoh16(ma->mcr_reserved));
2733 		printf(" %d: ipkt ", i);
2734 		ubsec_dump_pb(&ma->mcr_ipktbuf);
2735 		printf(" %d: opkt ", i);
2736 		ubsec_dump_pb(&ma->mcr_opktbuf);
2737 		ma++;
2738 	}
2739 	printf("END MCR\n");
2740 }
2741 #endif /* UBSEC_DEBUG */
2742 
2743 /*
2744  * Return the number of significant bits of a big number.
2745  */
2746 static int
2747 ubsec_ksigbits(struct crparam *cr)
2748 {
2749 	u_int plen = (cr->crp_nbits + 7) / 8;
2750 	int i, sig = plen * 8;
2751 	u_int8_t c, *p = cr->crp_p;
2752 
2753 	for (i = plen - 1; i >= 0; i--) {
2754 		c = p[i];
2755 		if (c != 0) {
2756 			while ((c & 0x80) == 0) {
2757 				sig--;
2758 				c <<= 1;
2759 			}
2760 			break;
2761 		}
2762 		sig -= 8;
2763 	}
2764 	return (sig);
2765 }
2766 
2767 static void
2768 ubsec_kshift_r(u_int shiftbits, u_int8_t *src, u_int srcbits,
2769     u_int8_t *dst, u_int dstbits)
2770 {
2771 	u_int slen, dlen;
2772 	int i, si, di, n;
2773 
2774 	slen = (srcbits + 7) / 8;
2775 	dlen = (dstbits + 7) / 8;
2776 
2777 	for (i = 0; i < slen; i++)
2778 		dst[i] = src[i];
2779 	for (i = 0; i < dlen - slen; i++)
2780 		dst[slen + i] = 0;
2781 
2782 	n = shiftbits / 8;
2783 	if (n != 0) {
2784 		si = dlen - n - 1;
2785 		di = dlen - 1;
2786 		while (si >= 0)
2787 			dst[di--] = dst[si--];
2788 		while (di >= 0)
2789 			dst[di--] = 0;
2790 	}
2791 
2792 	n = shiftbits % 8;
2793 	if (n != 0) {
2794 		for (i = dlen - 1; i > 0; i--)
2795 			dst[i] = (dst[i] << n) |
2796 			    (dst[i - 1] >> (8 - n));
2797 		dst[0] = dst[0] << n;
2798 	}
2799 }
2800 
2801 static void
2802 ubsec_kshift_l(u_int shiftbits, u_int8_t *src, u_int srcbits,
2803     u_int8_t *dst, u_int dstbits)
2804 {
2805 	int slen, dlen, i, n;
2806 
2807 	slen = (srcbits + 7) / 8;
2808 	dlen = (dstbits + 7) / 8;
2809 
2810 	n = shiftbits / 8;
2811 	for (i = 0; i < slen; i++)
2812 		dst[i] = src[i + n];
2813 	for (i = 0; i < dlen - slen; i++)
2814 		dst[slen + i] = 0;
2815 
2816 	n = shiftbits % 8;
2817 	if (n != 0) {
2818 		for (i = 0; i < (dlen - 1); i++)
2819 			dst[i] = (dst[i] >> n) | (dst[i + 1] << (8 - n));
2820 		dst[dlen - 1] = dst[dlen - 1] >> n;
2821 	}
2822 }
2823