xref: /netbsd-src/sys/dev/pci/ubsec.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
1 /*	$NetBSD: ubsec.c,v 1.24 2009/11/26 15:17:10 njoly 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 1.24 2009/11/26 15:17:10 njoly 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 <uvm/uvm_extern.h>
63 
64 #include <opencrypto/cryptodev.h>
65 #include <opencrypto/xform.h>
66 #ifdef __OpenBSD__
67  #include <dev/rndvar.h>
68  #include <sys/md5k.h>
69 #else
70  #include <sys/rnd.h>
71  #include <sys/md5.h>
72 #endif
73 #include <sys/sha1.h>
74 
75 #include <dev/pci/pcireg.h>
76 #include <dev/pci/pcivar.h>
77 #include <dev/pci/pcidevs.h>
78 
79 #include <dev/pci/ubsecreg.h>
80 #include <dev/pci/ubsecvar.h>
81 
82 /*
83  * Prototypes and count for the pci_device structure
84  */
85 static	int ubsec_probe(device_t, cfdata_t, void *);
86 static	void ubsec_attach(device_t, device_t, void *);
87 static	void ubsec_reset_board(struct ubsec_softc *);
88 static	void ubsec_init_board(struct ubsec_softc *);
89 static	void ubsec_init_pciregs(struct pci_attach_args *pa);
90 static	void ubsec_cleanchip(struct ubsec_softc *);
91 static	void ubsec_totalreset(struct ubsec_softc *);
92 static	int  ubsec_free_q(struct ubsec_softc*, struct ubsec_q *);
93 
94 #ifdef __OpenBSD__
95 struct cfattach ubsec_ca = {
96 	sizeof(struct ubsec_softc), ubsec_probe, ubsec_attach,
97 };
98 
99 struct cfdriver ubsec_cd = {
100 	0, "ubsec", DV_DULL
101 };
102 #else
103 CFATTACH_DECL(ubsec, sizeof(struct ubsec_softc), ubsec_probe, ubsec_attach,
104 	      NULL, NULL);
105 extern struct cfdriver ubsec_cd;
106 #endif
107 
108 /* patchable */
109 #ifdef	UBSEC_DEBUG
110 extern int ubsec_debug;
111 int ubsec_debug=1;
112 #endif
113 
114 static	int	ubsec_intr(void *);
115 static	int	ubsec_newsession(void*, u_int32_t *, struct cryptoini *);
116 static	int	ubsec_freesession(void*, u_int64_t);
117 static	int	ubsec_process(void*, struct cryptop *, int hint);
118 static	void	ubsec_callback(struct ubsec_softc *, struct ubsec_q *);
119 static	void	ubsec_feed(struct ubsec_softc *);
120 static	void	ubsec_mcopy(struct mbuf *, struct mbuf *, int, int);
121 static	void	ubsec_callback2(struct ubsec_softc *, struct ubsec_q2 *);
122 static	void	ubsec_feed2(struct ubsec_softc *);
123 #ifndef UBSEC_NO_RNG
124 static	void	ubsec_rng(void *);
125 #endif /* UBSEC_NO_RNG */
126 static	int 	ubsec_dma_malloc(struct ubsec_softc *, bus_size_t,
127 				 struct ubsec_dma_alloc *, int);
128 static	void	ubsec_dma_free(struct ubsec_softc *, struct ubsec_dma_alloc *);
129 static	int	ubsec_dmamap_aligned(bus_dmamap_t);
130 
131 static	int	ubsec_kprocess(void*, struct cryptkop *, int);
132 static	int	ubsec_kprocess_modexp_sw(struct ubsec_softc *,
133 					 struct cryptkop *, int);
134 static	int	ubsec_kprocess_modexp_hw(struct ubsec_softc *,
135 					 struct cryptkop *, int);
136 static	int	ubsec_kprocess_rsapriv(struct ubsec_softc *,
137 				       struct cryptkop *, int);
138 static	void	ubsec_kfree(struct ubsec_softc *, struct ubsec_q2 *);
139 static	int	ubsec_ksigbits(struct crparam *);
140 static	void	ubsec_kshift_r(u_int, u_int8_t *, u_int, u_int8_t *, u_int);
141 static	void	ubsec_kshift_l(u_int, u_int8_t *, u_int, u_int8_t *, u_int);
142 
143 #ifdef UBSEC_DEBUG
144 static void	ubsec_dump_pb(volatile struct ubsec_pktbuf *);
145 static void	ubsec_dump_mcr(struct ubsec_mcr *);
146 static	void	ubsec_dump_ctx2(volatile struct ubsec_ctx_keyop *);
147 #endif
148 
149 #define	READ_REG(sc,r) \
150 	bus_space_read_4((sc)->sc_st, (sc)->sc_sh, (r))
151 
152 #define WRITE_REG(sc,reg,val) \
153 	bus_space_write_4((sc)->sc_st, (sc)->sc_sh, reg, val)
154 
155 #define	SWAP32(x) (x) = htole32(ntohl((x)))
156 #ifndef HTOLE32
157  #define	HTOLE32(x) (x) = htole32(x)
158 #endif
159 
160 struct ubsec_stats ubsecstats;
161 
162 /*
163  * ubsec_maxbatch controls the number of crypto ops to voluntarily
164  * collect into one submission to the hardware.  This batching happens
165  * when ops are dispatched from the crypto subsystem with a hint that
166  * more are to follow immediately.  These ops must also not be marked
167  * with a ``no delay'' flag.
168  */
169 static	int ubsec_maxbatch = 1;
170 #ifdef SYSCTL_INT
171 SYSCTL_INT(_kern, OID_AUTO, ubsec_maxbatch, CTLFLAG_RW, &ubsec_maxbatch,
172 	    0, "Broadcom driver: max ops to batch w/o interrupt");
173 #endif
174 
175 /*
176  * ubsec_maxaggr controls the number of crypto ops to submit to the
177  * hardware as a unit.  This aggregation reduces the number of interrupts
178  * to the host at the expense of increased latency (for all but the last
179  * operation).  For network traffic setting this to one yields the highest
180  * performance but at the expense of more interrupt processing.
181  */
182 static	int ubsec_maxaggr = 1;
183 #ifdef SYSCTL_INT
184 SYSCTL_INT(_kern, OID_AUTO, ubsec_maxaggr, CTLFLAG_RW, &ubsec_maxaggr,
185 	    0, "Broadcom driver: max ops to aggregate under one interrupt");
186 #endif
187 
188 static const struct ubsec_product {
189 	pci_vendor_id_t		ubsec_vendor;
190 	pci_product_id_t	ubsec_product;
191 	int			ubsec_flags;
192 	int			ubsec_statmask;
193 	const char		*ubsec_name;
194 } ubsec_products[] = {
195 	{ PCI_VENDOR_BLUESTEEL,	PCI_PRODUCT_BLUESTEEL_5501,
196 	  0,
197 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR,
198 	  "Bluesteel 5501"
199 	},
200 	{ PCI_VENDOR_BLUESTEEL,	PCI_PRODUCT_BLUESTEEL_5601,
201 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG,
202 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR,
203 	  "Bluesteel 5601"
204 	},
205 
206 	{ PCI_VENDOR_BROADCOM,	PCI_PRODUCT_BROADCOM_5801,
207 	  0,
208 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR,
209 	  "Broadcom BCM5801"
210 	},
211 
212 	{ PCI_VENDOR_BROADCOM,	PCI_PRODUCT_BROADCOM_5802,
213 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG,
214 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR,
215 	  "Broadcom BCM5802"
216 	},
217 
218 	{ PCI_VENDOR_BROADCOM,	PCI_PRODUCT_BROADCOM_5805,
219 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG,
220 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR,
221 	  "Broadcom BCM5805"
222 	},
223 
224 	{ PCI_VENDOR_BROADCOM,	PCI_PRODUCT_BROADCOM_5820,
225 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG | UBS_FLAGS_LONGCTX |
226 	      UBS_FLAGS_HWNORM | UBS_FLAGS_BIGKEY,
227 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR,
228 	  "Broadcom BCM5820"
229 	},
230 
231 	{ PCI_VENDOR_BROADCOM,	PCI_PRODUCT_BROADCOM_5821,
232 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG | UBS_FLAGS_LONGCTX |
233 	      UBS_FLAGS_HWNORM | UBS_FLAGS_BIGKEY,
234 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR |
235 	      BS_STAT_MCR1_ALLEMPTY | BS_STAT_MCR2_ALLEMPTY,
236 	  "Broadcom BCM5821"
237 	},
238 	{ PCI_VENDOR_SUN,	PCI_PRODUCT_SUN_SCA1K,
239 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG | UBS_FLAGS_LONGCTX |
240 	      UBS_FLAGS_HWNORM | UBS_FLAGS_BIGKEY,
241 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR |
242 	      BS_STAT_MCR1_ALLEMPTY | BS_STAT_MCR2_ALLEMPTY,
243 	  "Sun Crypto Accelerator 1000"
244 	},
245 	{ PCI_VENDOR_SUN,	PCI_PRODUCT_SUN_5821,
246 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG | UBS_FLAGS_LONGCTX |
247 	      UBS_FLAGS_HWNORM | UBS_FLAGS_BIGKEY,
248 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR |
249 	      BS_STAT_MCR1_ALLEMPTY | BS_STAT_MCR2_ALLEMPTY,
250 	  "Broadcom BCM5821 (Sun)"
251 	},
252 
253 	{ PCI_VENDOR_BROADCOM,	PCI_PRODUCT_BROADCOM_5822,
254 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG | UBS_FLAGS_LONGCTX |
255 	      UBS_FLAGS_HWNORM | UBS_FLAGS_BIGKEY,
256 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR |
257 	      BS_STAT_MCR1_ALLEMPTY | BS_STAT_MCR2_ALLEMPTY,
258 	  "Broadcom BCM5822"
259 	},
260 
261 	{ PCI_VENDOR_BROADCOM,	PCI_PRODUCT_BROADCOM_5823,
262 	  UBS_FLAGS_KEY | UBS_FLAGS_RNG | UBS_FLAGS_LONGCTX |
263 	      UBS_FLAGS_HWNORM | UBS_FLAGS_BIGKEY,
264 	  BS_STAT_MCR1_DONE | BS_STAT_DMAERR |
265 	      BS_STAT_MCR1_ALLEMPTY | BS_STAT_MCR2_ALLEMPTY,
266 	  "Broadcom BCM5823"
267 	},
268 
269 	{ 0,			0,
270 	  0,
271 	  0,
272 	  NULL
273 	}
274 };
275 
276 static const struct ubsec_product *
277 ubsec_lookup(const struct pci_attach_args *pa)
278 {
279 	const struct ubsec_product *up;
280 
281 	for (up = ubsec_products; up->ubsec_name != NULL; up++) {
282 		if (PCI_VENDOR(pa->pa_id) == up->ubsec_vendor &&
283 		    PCI_PRODUCT(pa->pa_id) == up->ubsec_product)
284 			return (up);
285 	}
286 	return (NULL);
287 }
288 
289 static int
290 ubsec_probe(device_t parent, cfdata_t match, void *aux)
291 {
292 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
293 
294 	if (ubsec_lookup(pa) != NULL)
295 		return (1);
296 
297 	return (0);
298 }
299 
300 static void
301 ubsec_attach(device_t parent, device_t self, void *aux)
302 {
303 	struct ubsec_softc *sc = device_private(self);
304 	struct pci_attach_args *pa = aux;
305 	const struct ubsec_product *up;
306 	pci_chipset_tag_t pc = pa->pa_pc;
307 	pci_intr_handle_t ih;
308 	const char *intrstr = NULL;
309 	struct ubsec_dma *dmap;
310 	u_int32_t cmd, i;
311 
312 	up = ubsec_lookup(pa);
313 	if (up == NULL) {
314 		printf("\n");
315 		panic("ubsec_attach: impossible");
316 	}
317 
318 	aprint_naive(": Crypto processor\n");
319 	aprint_normal(": %s, rev. %d\n", up->ubsec_name,
320 	    PCI_REVISION(pa->pa_class));
321 
322 	SIMPLEQ_INIT(&sc->sc_queue);
323 	SIMPLEQ_INIT(&sc->sc_qchip);
324 	SIMPLEQ_INIT(&sc->sc_queue2);
325 	SIMPLEQ_INIT(&sc->sc_qchip2);
326 	SIMPLEQ_INIT(&sc->sc_q2free);
327 
328 	sc->sc_flags = up->ubsec_flags;
329 	sc->sc_statmask = up->ubsec_statmask;
330 
331 	cmd = pci_conf_read(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
332 	cmd |= PCI_COMMAND_MASTER_ENABLE;
333 	pci_conf_write(pc, pa->pa_tag, PCI_COMMAND_STATUS_REG, cmd);
334 
335 	if (pci_mapreg_map(pa, BS_BAR, PCI_MAPREG_TYPE_MEM, 0,
336 	    &sc->sc_st, &sc->sc_sh, NULL, NULL)) {
337 		aprint_error_dev(&sc->sc_dv, "can't find mem space");
338 		return;
339 	}
340 
341 	sc->sc_dmat = pa->pa_dmat;
342 
343 	if (pci_intr_map(pa, &ih)) {
344 		aprint_error_dev(&sc->sc_dv, "couldn't map interrupt\n");
345 		return;
346 	}
347 	intrstr = pci_intr_string(pc, ih);
348 	sc->sc_ih = pci_intr_establish(pc, ih, IPL_NET, ubsec_intr, sc);
349 	if (sc->sc_ih == NULL) {
350 		aprint_error_dev(&sc->sc_dv, "couldn't establish interrupt");
351 		if (intrstr != NULL)
352 			aprint_error(" at %s", intrstr);
353 		aprint_error("\n");
354 		return;
355 	}
356 	aprint_normal_dev(&sc->sc_dv, "interrupting at %s\n", intrstr);
357 
358 	sc->sc_cid = crypto_get_driverid(0);
359 	if (sc->sc_cid < 0) {
360 		aprint_error_dev(&sc->sc_dv, "couldn't get crypto driver id\n");
361 		pci_intr_disestablish(pc, sc->sc_ih);
362 		return;
363 	}
364 
365 	SIMPLEQ_INIT(&sc->sc_freequeue);
366 	dmap = sc->sc_dmaa;
367 	for (i = 0; i < UBS_MAX_NQUEUE; i++, dmap++) {
368 		struct ubsec_q *q;
369 
370 		q = (struct ubsec_q *)malloc(sizeof(struct ubsec_q),
371 		    M_DEVBUF, M_NOWAIT);
372 		if (q == NULL) {
373 			aprint_error_dev(&sc->sc_dv, "can't allocate queue buffers\n");
374 			break;
375 		}
376 
377 		if (ubsec_dma_malloc(sc, sizeof(struct ubsec_dmachunk),
378 		    &dmap->d_alloc, 0)) {
379 			aprint_error_dev(&sc->sc_dv, "can't allocate dma buffers\n");
380 			free(q, M_DEVBUF);
381 			break;
382 		}
383 		dmap->d_dma = (struct ubsec_dmachunk *)dmap->d_alloc.dma_vaddr;
384 
385 		q->q_dma = dmap;
386 		sc->sc_queuea[i] = q;
387 
388 		SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
389 	}
390 
391 	crypto_register(sc->sc_cid, CRYPTO_3DES_CBC, 0, 0,
392 	    ubsec_newsession, ubsec_freesession, ubsec_process, sc);
393 	crypto_register(sc->sc_cid, CRYPTO_DES_CBC, 0, 0,
394 	    ubsec_newsession, ubsec_freesession, ubsec_process, sc);
395 	crypto_register(sc->sc_cid, CRYPTO_MD5_HMAC_96, 0, 0,
396 	    ubsec_newsession, ubsec_freesession, ubsec_process, sc);
397 	crypto_register(sc->sc_cid, CRYPTO_SHA1_HMAC_96, 0, 0,
398 	    ubsec_newsession, ubsec_freesession, ubsec_process, sc);
399 
400 	/*
401 	 * Reset Broadcom chip
402 	 */
403 	ubsec_reset_board(sc);
404 
405 	/*
406 	 * Init Broadcom specific PCI settings
407 	 */
408 	ubsec_init_pciregs(pa);
409 
410 	/*
411 	 * Init Broadcom chip
412 	 */
413 	ubsec_init_board(sc);
414 
415 #ifndef UBSEC_NO_RNG
416 	if (sc->sc_flags & UBS_FLAGS_RNG) {
417 		sc->sc_statmask |= BS_STAT_MCR2_DONE;
418 
419 		if (ubsec_dma_malloc(sc, sizeof(struct ubsec_mcr),
420 		    &sc->sc_rng.rng_q.q_mcr, 0))
421 			goto skip_rng;
422 
423 		if (ubsec_dma_malloc(sc, sizeof(struct ubsec_ctx_rngbypass),
424 		    &sc->sc_rng.rng_q.q_ctx, 0)) {
425 			ubsec_dma_free(sc, &sc->sc_rng.rng_q.q_mcr);
426 			goto skip_rng;
427 		}
428 
429 		if (ubsec_dma_malloc(sc, sizeof(u_int32_t) *
430 		    UBSEC_RNG_BUFSIZ, &sc->sc_rng.rng_buf, 0)) {
431 			ubsec_dma_free(sc, &sc->sc_rng.rng_q.q_ctx);
432 			ubsec_dma_free(sc, &sc->sc_rng.rng_q.q_mcr);
433 			goto skip_rng;
434 		}
435 
436 		if (hz >= 100)
437 			sc->sc_rnghz = hz / 100;
438 		else
439 			sc->sc_rnghz = 1;
440 #ifdef __OpenBSD__
441 		timeout_set(&sc->sc_rngto, ubsec_rng, sc);
442 		timeout_add(&sc->sc_rngto, sc->sc_rnghz);
443 #else
444 		callout_init(&sc->sc_rngto, 0);
445 		callout_reset(&sc->sc_rngto, sc->sc_rnghz, ubsec_rng, sc);
446 #endif
447  skip_rng:
448 		if (sc->sc_rnghz)
449 			aprint_normal_dev(&sc->sc_dv, "random number generator enabled\n");
450 		else
451 			aprint_error_dev(&sc->sc_dv, "WARNING: random number generator "
452 			    "disabled\n");
453 	}
454 #endif /* UBSEC_NO_RNG */
455 
456 	if (sc->sc_flags & UBS_FLAGS_KEY) {
457 		sc->sc_statmask |= BS_STAT_MCR2_DONE;
458 
459 		crypto_kregister(sc->sc_cid, CRK_MOD_EXP, 0,
460 				 ubsec_kprocess, sc);
461 #if 0
462 		crypto_kregister(sc->sc_cid, CRK_MOD_EXP_CRT, 0,
463 				 ubsec_kprocess, sc);
464 #endif
465 	}
466 }
467 
468 /*
469  * UBSEC Interrupt routine
470  */
471 static int
472 ubsec_intr(void *arg)
473 {
474 	struct ubsec_softc *sc = arg;
475 	volatile u_int32_t stat;
476 	struct ubsec_q *q;
477 	struct ubsec_dma *dmap;
478 	int npkts = 0, i;
479 
480 	stat = READ_REG(sc, BS_STAT);
481 	stat &= sc->sc_statmask;
482 	if (stat == 0) {
483 		return (0);
484 	}
485 
486 	WRITE_REG(sc, BS_STAT, stat);		/* IACK */
487 
488 	/*
489 	 * Check to see if we have any packets waiting for us
490 	 */
491 	if ((stat & BS_STAT_MCR1_DONE)) {
492 		while (!SIMPLEQ_EMPTY(&sc->sc_qchip)) {
493 			q = SIMPLEQ_FIRST(&sc->sc_qchip);
494 			dmap = q->q_dma;
495 
496 			if ((dmap->d_dma->d_mcr.mcr_flags & htole16(UBS_MCR_DONE)) == 0)
497 				break;
498 
499 			q = SIMPLEQ_FIRST(&sc->sc_qchip);
500 			SIMPLEQ_REMOVE_HEAD(&sc->sc_qchip, /*q,*/ q_next);
501 
502 			npkts = q->q_nstacked_mcrs;
503 			sc->sc_nqchip -= 1+npkts;
504 			/*
505 			 * search for further sc_qchip ubsec_q's that share
506 			 * the same MCR, and complete them too, they must be
507 			 * at the top.
508 			 */
509 			for (i = 0; i < npkts; i++) {
510 				if(q->q_stacked_mcr[i])
511 					ubsec_callback(sc, q->q_stacked_mcr[i]);
512 				else
513 					break;
514 			}
515 			ubsec_callback(sc, q);
516 		}
517 
518 		/*
519 		 * Don't send any more packet to chip if there has been
520 		 * a DMAERR.
521 		 */
522 		if (!(stat & BS_STAT_DMAERR))
523 			ubsec_feed(sc);
524 	}
525 
526 	/*
527 	 * Check to see if we have any key setups/rng's waiting for us
528 	 */
529 	if ((sc->sc_flags & (UBS_FLAGS_KEY|UBS_FLAGS_RNG)) &&
530 	    (stat & BS_STAT_MCR2_DONE)) {
531 		struct ubsec_q2 *q2;
532 		struct ubsec_mcr *mcr;
533 
534 		while (!SIMPLEQ_EMPTY(&sc->sc_qchip2)) {
535 			q2 = SIMPLEQ_FIRST(&sc->sc_qchip2);
536 
537 			bus_dmamap_sync(sc->sc_dmat, q2->q_mcr.dma_map,
538 			    0, q2->q_mcr.dma_map->dm_mapsize,
539 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
540 
541 			mcr = (struct ubsec_mcr *)q2->q_mcr.dma_vaddr;
542 			if ((mcr->mcr_flags & htole16(UBS_MCR_DONE)) == 0) {
543 				bus_dmamap_sync(sc->sc_dmat,
544 				    q2->q_mcr.dma_map, 0,
545 				    q2->q_mcr.dma_map->dm_mapsize,
546 				    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
547 				break;
548 			}
549 			q2 = SIMPLEQ_FIRST(&sc->sc_qchip2);
550 			SIMPLEQ_REMOVE_HEAD(&sc->sc_qchip2, /*q2,*/ q_next);
551 			ubsec_callback2(sc, q2);
552 			/*
553 			 * Don't send any more packet to chip if there has been
554 			 * a DMAERR.
555 			 */
556 			if (!(stat & BS_STAT_DMAERR))
557 				ubsec_feed2(sc);
558 		}
559 	}
560 
561 	/*
562 	 * Check to see if we got any DMA Error
563 	 */
564 	if (stat & BS_STAT_DMAERR) {
565 #ifdef UBSEC_DEBUG
566 		if (ubsec_debug) {
567 			volatile u_int32_t a = READ_REG(sc, BS_ERR);
568 
569 			printf("%s: dmaerr %s@%08x\n", device_xname(&sc->sc_dv),
570 			    (a & BS_ERR_READ) ? "read" : "write",
571 			       a & BS_ERR_ADDR);
572 		}
573 #endif /* UBSEC_DEBUG */
574 		ubsecstats.hst_dmaerr++;
575 		ubsec_totalreset(sc);
576 		ubsec_feed(sc);
577 	}
578 
579 	if (sc->sc_needwakeup) {		/* XXX check high watermark */
580 		int wkeup = sc->sc_needwakeup & (CRYPTO_SYMQ|CRYPTO_ASYMQ);
581 #ifdef UBSEC_DEBUG
582 		if (ubsec_debug)
583 			printf("%s: wakeup crypto (%x)\n", device_xname(&sc->sc_dv),
584 				sc->sc_needwakeup);
585 #endif /* UBSEC_DEBUG */
586 		sc->sc_needwakeup &= ~wkeup;
587 		crypto_unblock(sc->sc_cid, wkeup);
588 	}
589 	return (1);
590 }
591 
592 /*
593  * ubsec_feed() - aggregate and post requests to chip
594  * OpenBSD comments:
595  *		  It is assumed that the caller set splnet()
596  */
597 static void
598 ubsec_feed(struct ubsec_softc *sc)
599 {
600 	struct ubsec_q *q, *q2;
601 	int npkts, i;
602 	void *v;
603 	u_int32_t stat;
604 #ifdef UBSEC_DEBUG
605 	static int max;
606 #endif /* UBSEC_DEBUG */
607 
608 	npkts = sc->sc_nqueue;
609 	if (npkts > ubsecstats.hst_maxqueue)
610 		ubsecstats.hst_maxqueue = npkts;
611 	if (npkts < 2)
612 		goto feed1;
613 
614 	/*
615 	 * Decide how many ops to combine in a single MCR.  We cannot
616 	 * aggregate more than UBS_MAX_AGGR because this is the number
617 	 * of slots defined in the data structure.  Otherwise we clamp
618 	 * based on the tunable parameter ubsec_maxaggr.  Note that
619 	 * aggregation can happen in two ways: either by batching ops
620 	 * from above or because the h/w backs up and throttles us.
621 	 * Aggregating ops reduces the number of interrupts to the host
622 	 * but also (potentially) increases the latency for processing
623 	 * completed ops as we only get an interrupt when all aggregated
624 	 * ops have completed.
625 	 */
626 	if (npkts > UBS_MAX_AGGR)
627 		npkts = UBS_MAX_AGGR;
628 	if (npkts > ubsec_maxaggr)
629 		npkts = ubsec_maxaggr;
630 	if (npkts > ubsecstats.hst_maxbatch)
631 		ubsecstats.hst_maxbatch = npkts;
632 	if (npkts < 2)
633 		goto feed1;
634 	ubsecstats.hst_totbatch += npkts-1;
635 
636 	if ((stat = READ_REG(sc, BS_STAT)) & (BS_STAT_MCR1_FULL | BS_STAT_DMAERR)) {
637 		if (stat & BS_STAT_DMAERR) {
638 			ubsec_totalreset(sc);
639 			ubsecstats.hst_dmaerr++;
640 		} else {
641 			ubsecstats.hst_mcr1full++;
642 		}
643 		return;
644 	}
645 
646 #ifdef UBSEC_DEBUG
647 	if (ubsec_debug)
648 	    printf("merging %d records\n", npkts);
649 	/* XXX temporary aggregation statistics reporting code */
650 	if (max < npkts) {
651 		max = npkts;
652 		printf("%s: new max aggregate %d\n", device_xname(&sc->sc_dv), max);
653 	}
654 #endif /* UBSEC_DEBUG */
655 
656 	q = SIMPLEQ_FIRST(&sc->sc_queue);
657 	SIMPLEQ_REMOVE_HEAD(&sc->sc_queue, /*q,*/ q_next);
658 	--sc->sc_nqueue;
659 
660 	bus_dmamap_sync(sc->sc_dmat, q->q_src_map,
661 	    0, q->q_src_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
662 	if (q->q_dst_map != NULL)
663 		bus_dmamap_sync(sc->sc_dmat, q->q_dst_map,
664 		    0, q->q_dst_map->dm_mapsize, BUS_DMASYNC_PREREAD);
665 
666 	q->q_nstacked_mcrs = npkts - 1;		/* Number of packets stacked */
667 
668 	for (i = 0; i < q->q_nstacked_mcrs; i++) {
669 		q2 = SIMPLEQ_FIRST(&sc->sc_queue);
670 		bus_dmamap_sync(sc->sc_dmat, q2->q_src_map,
671 		    0, q2->q_src_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
672 		if (q2->q_dst_map != NULL)
673 			bus_dmamap_sync(sc->sc_dmat, q2->q_dst_map,
674 			    0, q2->q_dst_map->dm_mapsize, BUS_DMASYNC_PREREAD);
675 		q2= SIMPLEQ_FIRST(&sc->sc_queue);
676 		SIMPLEQ_REMOVE_HEAD(&sc->sc_queue, /*q2,*/ q_next);
677 		--sc->sc_nqueue;
678 
679 		v = ((void *)&q2->q_dma->d_dma->d_mcr);
680 		v = (char*)v + (sizeof(struct ubsec_mcr) -
681 				 sizeof(struct ubsec_mcr_add));
682 		memcpy(&q->q_dma->d_dma->d_mcradd[i], v, sizeof(struct ubsec_mcr_add));
683 		q->q_stacked_mcr[i] = q2;
684 	}
685 	q->q_dma->d_dma->d_mcr.mcr_pkts = htole16(npkts);
686 	SIMPLEQ_INSERT_TAIL(&sc->sc_qchip, q, q_next);
687 	sc->sc_nqchip += npkts;
688 	if (sc->sc_nqchip > ubsecstats.hst_maxqchip)
689 		ubsecstats.hst_maxqchip = sc->sc_nqchip;
690 	bus_dmamap_sync(sc->sc_dmat, q->q_dma->d_alloc.dma_map,
691 	    0, q->q_dma->d_alloc.dma_map->dm_mapsize,
692 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
693 	WRITE_REG(sc, BS_MCR1, q->q_dma->d_alloc.dma_paddr +
694 	    offsetof(struct ubsec_dmachunk, d_mcr));
695 	return;
696 
697 feed1:
698 	while (!SIMPLEQ_EMPTY(&sc->sc_queue)) {
699 		if ((stat = READ_REG(sc, BS_STAT)) & (BS_STAT_MCR1_FULL | BS_STAT_DMAERR)) {
700 			if (stat & BS_STAT_DMAERR) {
701 				ubsec_totalreset(sc);
702 				ubsecstats.hst_dmaerr++;
703 			} else {
704 				ubsecstats.hst_mcr1full++;
705 			}
706 			break;
707 		}
708 
709 		q = SIMPLEQ_FIRST(&sc->sc_queue);
710 
711 		bus_dmamap_sync(sc->sc_dmat, q->q_src_map,
712 		    0, q->q_src_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
713 		if (q->q_dst_map != NULL)
714 			bus_dmamap_sync(sc->sc_dmat, q->q_dst_map,
715 			    0, q->q_dst_map->dm_mapsize, BUS_DMASYNC_PREREAD);
716 		bus_dmamap_sync(sc->sc_dmat, q->q_dma->d_alloc.dma_map,
717 		    0, q->q_dma->d_alloc.dma_map->dm_mapsize,
718 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
719 
720 		WRITE_REG(sc, BS_MCR1, q->q_dma->d_alloc.dma_paddr +
721 		    offsetof(struct ubsec_dmachunk, d_mcr));
722 #ifdef UBSEC_DEBUG
723 		if (ubsec_debug)
724 			printf("feed: q->chip %p %08x stat %08x\n",
725  		    	       q, (u_int32_t)q->q_dma->d_alloc.dma_paddr,
726 			       stat);
727 #endif /* UBSEC_DEBUG */
728 		q = SIMPLEQ_FIRST(&sc->sc_queue);
729 		SIMPLEQ_REMOVE_HEAD(&sc->sc_queue, /*q,*/ q_next);
730 		--sc->sc_nqueue;
731 		SIMPLEQ_INSERT_TAIL(&sc->sc_qchip, q, q_next);
732 		sc->sc_nqchip++;
733 	}
734 	if (sc->sc_nqchip > ubsecstats.hst_maxqchip)
735 		ubsecstats.hst_maxqchip = sc->sc_nqchip;
736 }
737 
738 /*
739  * Allocate a new 'session' and return an encoded session id.  'sidp'
740  * contains our registration id, and should contain an encoded session
741  * id on successful allocation.
742  */
743 static int
744 ubsec_newsession(void *arg, u_int32_t *sidp, struct cryptoini *cri)
745 {
746 	struct cryptoini *c, *encini = NULL, *macini = NULL;
747 	struct ubsec_softc *sc;
748 	struct ubsec_session *ses = NULL;
749 	MD5_CTX md5ctx;
750 	SHA1_CTX sha1ctx;
751 	int i, sesn;
752 
753 	sc = arg;
754 	KASSERT(sc != NULL /*, ("ubsec_newsession: null softc")*/);
755 
756 	if (sidp == NULL || cri == NULL || sc == NULL)
757 		return (EINVAL);
758 
759 	for (c = cri; c != NULL; c = c->cri_next) {
760 		if (c->cri_alg == CRYPTO_MD5_HMAC_96 ||
761 		    c->cri_alg == CRYPTO_SHA1_HMAC_96) {
762 			if (macini)
763 				return (EINVAL);
764 			macini = c;
765 		} else if (c->cri_alg == CRYPTO_DES_CBC ||
766 		    c->cri_alg == CRYPTO_3DES_CBC) {
767 			if (encini)
768 				return (EINVAL);
769 			encini = c;
770 		} else
771 			return (EINVAL);
772 	}
773 	if (encini == NULL && macini == NULL)
774 		return (EINVAL);
775 
776 	if (sc->sc_sessions == NULL) {
777 		ses = sc->sc_sessions = (struct ubsec_session *)malloc(
778 		    sizeof(struct ubsec_session), M_DEVBUF, M_NOWAIT);
779 		if (ses == NULL)
780 			return (ENOMEM);
781 		sesn = 0;
782 		sc->sc_nsessions = 1;
783 	} else {
784 		for (sesn = 0; sesn < sc->sc_nsessions; sesn++) {
785 			if (sc->sc_sessions[sesn].ses_used == 0) {
786 				ses = &sc->sc_sessions[sesn];
787 				break;
788 			}
789 		}
790 
791 		if (ses == NULL) {
792 			sesn = sc->sc_nsessions;
793 			ses = (struct ubsec_session *)malloc((sesn + 1) *
794 			    sizeof(struct ubsec_session), M_DEVBUF, M_NOWAIT);
795 			if (ses == NULL)
796 				return (ENOMEM);
797 			memcpy(ses, sc->sc_sessions, sesn *
798 			    sizeof(struct ubsec_session));
799 			memset(sc->sc_sessions, 0, sesn *
800 			    sizeof(struct ubsec_session));
801 			free(sc->sc_sessions, M_DEVBUF);
802 			sc->sc_sessions = ses;
803 			ses = &sc->sc_sessions[sesn];
804 			sc->sc_nsessions++;
805 		}
806 	}
807 
808 	memset(ses, 0, sizeof(struct ubsec_session));
809 	ses->ses_used = 1;
810 	if (encini) {
811 		/* get an IV, network byte order */
812 #ifdef __NetBSD__
813 		rnd_extract_data(ses->ses_iv,
814 		    sizeof(ses->ses_iv), RND_EXTRACT_ANY);
815 #else
816 		get_random_bytes(ses->ses_iv, sizeof(ses->ses_iv));
817 #endif
818 
819 		/* Go ahead and compute key in ubsec's byte order */
820 		if (encini->cri_alg == CRYPTO_DES_CBC) {
821 			memcpy(&ses->ses_deskey[0], encini->cri_key, 8);
822 			memcpy(&ses->ses_deskey[2], encini->cri_key, 8);
823 			memcpy(&ses->ses_deskey[4], encini->cri_key, 8);
824 		} else
825 			memcpy(ses->ses_deskey, encini->cri_key, 24);
826 
827 		SWAP32(ses->ses_deskey[0]);
828 		SWAP32(ses->ses_deskey[1]);
829 		SWAP32(ses->ses_deskey[2]);
830 		SWAP32(ses->ses_deskey[3]);
831 		SWAP32(ses->ses_deskey[4]);
832 		SWAP32(ses->ses_deskey[5]);
833 	}
834 
835 	if (macini) {
836 		for (i = 0; i < macini->cri_klen / 8; i++)
837 			macini->cri_key[i] ^= HMAC_IPAD_VAL;
838 
839 		if (macini->cri_alg == CRYPTO_MD5_HMAC_96) {
840 			MD5Init(&md5ctx);
841 			MD5Update(&md5ctx, macini->cri_key,
842 			    macini->cri_klen / 8);
843 			MD5Update(&md5ctx, hmac_ipad_buffer,
844 			    HMAC_BLOCK_LEN - (macini->cri_klen / 8));
845 			memcpy(ses->ses_hminner, md5ctx.state,
846 			    sizeof(md5ctx.state));
847 		} else {
848 			SHA1Init(&sha1ctx);
849 			SHA1Update(&sha1ctx, macini->cri_key,
850 			    macini->cri_klen / 8);
851 			SHA1Update(&sha1ctx, hmac_ipad_buffer,
852 			    HMAC_BLOCK_LEN - (macini->cri_klen / 8));
853 			memcpy(ses->ses_hminner, sha1ctx.state,
854 			    sizeof(sha1ctx.state));
855 		}
856 
857 		for (i = 0; i < macini->cri_klen / 8; i++)
858 			macini->cri_key[i] ^= (HMAC_IPAD_VAL ^ HMAC_OPAD_VAL);
859 
860 		if (macini->cri_alg == CRYPTO_MD5_HMAC_96) {
861 			MD5Init(&md5ctx);
862 			MD5Update(&md5ctx, macini->cri_key,
863 			    macini->cri_klen / 8);
864 			MD5Update(&md5ctx, hmac_opad_buffer,
865 			    HMAC_BLOCK_LEN - (macini->cri_klen / 8));
866 			memcpy(ses->ses_hmouter, md5ctx.state,
867 			    sizeof(md5ctx.state));
868 		} else {
869 			SHA1Init(&sha1ctx);
870 			SHA1Update(&sha1ctx, macini->cri_key,
871 			    macini->cri_klen / 8);
872 			SHA1Update(&sha1ctx, hmac_opad_buffer,
873 			    HMAC_BLOCK_LEN - (macini->cri_klen / 8));
874 			memcpy(ses->ses_hmouter, sha1ctx.state,
875 			    sizeof(sha1ctx.state));
876 		}
877 
878 		for (i = 0; i < macini->cri_klen / 8; i++)
879 			macini->cri_key[i] ^= HMAC_OPAD_VAL;
880 	}
881 
882 	*sidp = UBSEC_SID(device_unit(&sc->sc_dv), sesn);
883 	return (0);
884 }
885 
886 /*
887  * Deallocate a session.
888  */
889 static int
890 ubsec_freesession(void *arg, u_int64_t tid)
891 {
892 	struct ubsec_softc *sc;
893 	int session;
894 	u_int32_t sid = ((u_int32_t) tid) & 0xffffffff;
895 
896 	sc = arg;
897 	KASSERT(sc != NULL /*, ("ubsec_freesession: null softc")*/);
898 
899 	session = UBSEC_SESSION(sid);
900 	if (session >= sc->sc_nsessions)
901 		return (EINVAL);
902 
903 	memset(&sc->sc_sessions[session], 0, sizeof(sc->sc_sessions[session]));
904 	return (0);
905 }
906 
907 #ifdef __FreeBSD__ /* Ugly gratuitous changes to bus_dma */
908 static void
909 ubsec_op_cb(void *arg, bus_dma_segment_t *seg, int nsegs, bus_size_t mapsize, int error)
910 {
911 	struct ubsec_operand *op = arg;
912 
913 	KASSERT(nsegs <= UBS_MAX_SCATTER
914 		/*, ("Too many DMA segments returned when mapping operand")*/);
915 #ifdef UBSEC_DEBUG
916 	if (ubsec_debug)
917 		printf("ubsec_op_cb: mapsize %u nsegs %d\n",
918 			(u_int) mapsize, nsegs);
919 #endif
920 	op->mapsize = mapsize;
921 	op->nsegs = nsegs;
922 	memcpy(op->segs, seg, nsegs * sizeof (seg[0]));
923 }
924 #endif
925 
926 static int
927 ubsec_process(void *arg, struct cryptop *crp, int hint)
928 {
929 	struct ubsec_q *q = NULL;
930 #ifdef	__OpenBSD__
931 	int card;
932 #endif
933 	int err = 0, i, j, s, nicealign;
934 	struct ubsec_softc *sc;
935 	struct cryptodesc *crd1, *crd2, *maccrd, *enccrd;
936 	int encoffset = 0, macoffset = 0, cpskip, cpoffset;
937 	int sskip, dskip, stheend, dtheend;
938 	int16_t coffset;
939 	struct ubsec_session *ses;
940 	struct ubsec_pktctx ctx;
941 	struct ubsec_dma *dmap = NULL;
942 
943 	sc = arg;
944 	KASSERT(sc != NULL /*, ("ubsec_process: null softc")*/);
945 
946 	if (crp == NULL || crp->crp_callback == NULL || sc == NULL) {
947 		ubsecstats.hst_invalid++;
948 		return (EINVAL);
949 	}
950 	if (UBSEC_SESSION(crp->crp_sid) >= sc->sc_nsessions) {
951 		ubsecstats.hst_badsession++;
952 		return (EINVAL);
953 	}
954 
955 	s = splnet();
956 
957 	if (SIMPLEQ_EMPTY(&sc->sc_freequeue)) {
958 		ubsecstats.hst_queuefull++;
959 		sc->sc_needwakeup |= CRYPTO_SYMQ;
960 		splx(s);
961 		return(ERESTART);
962 	}
963 
964 	q = SIMPLEQ_FIRST(&sc->sc_freequeue);
965 	SIMPLEQ_REMOVE_HEAD(&sc->sc_freequeue, /*q,*/ q_next);
966 	splx(s);
967 
968 	dmap = q->q_dma; /* Save dma pointer */
969 	memset(q, 0, sizeof(struct ubsec_q));
970 	memset(&ctx, 0, sizeof(ctx));
971 
972 	q->q_sesn = UBSEC_SESSION(crp->crp_sid);
973 	q->q_dma = dmap;
974 	ses = &sc->sc_sessions[q->q_sesn];
975 
976 	if (crp->crp_flags & CRYPTO_F_IMBUF) {
977 		q->q_src_m = (struct mbuf *)crp->crp_buf;
978 		q->q_dst_m = (struct mbuf *)crp->crp_buf;
979 	} else if (crp->crp_flags & CRYPTO_F_IOV) {
980 		q->q_src_io = (struct uio *)crp->crp_buf;
981 		q->q_dst_io = (struct uio *)crp->crp_buf;
982 	} else {
983 		ubsecstats.hst_badflags++;
984 		err = EINVAL;
985 		goto errout;	/* XXX we don't handle contiguous blocks! */
986 	}
987 
988 	memset(&dmap->d_dma->d_mcr, 0, sizeof(struct ubsec_mcr));
989 
990 	dmap->d_dma->d_mcr.mcr_pkts = htole16(1);
991 	dmap->d_dma->d_mcr.mcr_flags = 0;
992 	q->q_crp = crp;
993 
994 	crd1 = crp->crp_desc;
995 	if (crd1 == NULL) {
996 		ubsecstats.hst_nodesc++;
997 		err = EINVAL;
998 		goto errout;
999 	}
1000 	crd2 = crd1->crd_next;
1001 
1002 	if (crd2 == NULL) {
1003 		if (crd1->crd_alg == CRYPTO_MD5_HMAC_96 ||
1004 		    crd1->crd_alg == CRYPTO_SHA1_HMAC_96) {
1005 			maccrd = crd1;
1006 			enccrd = NULL;
1007 		} else if (crd1->crd_alg == CRYPTO_DES_CBC ||
1008 		    crd1->crd_alg == CRYPTO_3DES_CBC) {
1009 			maccrd = NULL;
1010 			enccrd = crd1;
1011 		} else {
1012 			ubsecstats.hst_badalg++;
1013 			err = EINVAL;
1014 			goto errout;
1015 		}
1016 	} else {
1017 		if ((crd1->crd_alg == CRYPTO_MD5_HMAC_96 ||
1018 		    crd1->crd_alg == CRYPTO_SHA1_HMAC_96) &&
1019 		    (crd2->crd_alg == CRYPTO_DES_CBC ||
1020 			crd2->crd_alg == CRYPTO_3DES_CBC) &&
1021 		    ((crd2->crd_flags & CRD_F_ENCRYPT) == 0)) {
1022 			maccrd = crd1;
1023 			enccrd = crd2;
1024 		} else if ((crd1->crd_alg == CRYPTO_DES_CBC ||
1025 		    crd1->crd_alg == CRYPTO_3DES_CBC) &&
1026 		    (crd2->crd_alg == CRYPTO_MD5_HMAC_96 ||
1027 			crd2->crd_alg == CRYPTO_SHA1_HMAC_96) &&
1028 		    (crd1->crd_flags & CRD_F_ENCRYPT)) {
1029 			enccrd = crd1;
1030 			maccrd = crd2;
1031 		} else {
1032 			/*
1033 			 * We cannot order the ubsec as requested
1034 			 */
1035 			ubsecstats.hst_badalg++;
1036 			err = EINVAL;
1037 			goto errout;
1038 		}
1039 	}
1040 
1041 	if (enccrd) {
1042 		encoffset = enccrd->crd_skip;
1043 		ctx.pc_flags |= htole16(UBS_PKTCTX_ENC_3DES);
1044 
1045 		if (enccrd->crd_flags & CRD_F_ENCRYPT) {
1046 			q->q_flags |= UBSEC_QFLAGS_COPYOUTIV;
1047 
1048 			if (enccrd->crd_flags & CRD_F_IV_EXPLICIT)
1049 				memcpy(ctx.pc_iv, enccrd->crd_iv, 8);
1050 			else {
1051 				ctx.pc_iv[0] = ses->ses_iv[0];
1052 				ctx.pc_iv[1] = ses->ses_iv[1];
1053 			}
1054 
1055 			if ((enccrd->crd_flags & CRD_F_IV_PRESENT) == 0) {
1056 				if (crp->crp_flags & CRYPTO_F_IMBUF)
1057 					m_copyback(q->q_src_m,
1058 					    enccrd->crd_inject,
1059 					    8, (void *)ctx.pc_iv);
1060 				else if (crp->crp_flags & CRYPTO_F_IOV)
1061 					cuio_copyback(q->q_src_io,
1062 					    enccrd->crd_inject,
1063 					    8, (void *)ctx.pc_iv);
1064 			}
1065 		} else {
1066 			ctx.pc_flags |= htole16(UBS_PKTCTX_INBOUND);
1067 
1068 			if (enccrd->crd_flags & CRD_F_IV_EXPLICIT)
1069 				memcpy(ctx.pc_iv, enccrd->crd_iv, 8);
1070 			else if (crp->crp_flags & CRYPTO_F_IMBUF)
1071 				m_copydata(q->q_src_m, enccrd->crd_inject,
1072 				    8, (void *)ctx.pc_iv);
1073 			else if (crp->crp_flags & CRYPTO_F_IOV)
1074 				cuio_copydata(q->q_src_io,
1075 				    enccrd->crd_inject, 8,
1076 				    (void *)ctx.pc_iv);
1077 		}
1078 
1079 		ctx.pc_deskey[0] = ses->ses_deskey[0];
1080 		ctx.pc_deskey[1] = ses->ses_deskey[1];
1081 		ctx.pc_deskey[2] = ses->ses_deskey[2];
1082 		ctx.pc_deskey[3] = ses->ses_deskey[3];
1083 		ctx.pc_deskey[4] = ses->ses_deskey[4];
1084 		ctx.pc_deskey[5] = ses->ses_deskey[5];
1085 		SWAP32(ctx.pc_iv[0]);
1086 		SWAP32(ctx.pc_iv[1]);
1087 	}
1088 
1089 	if (maccrd) {
1090 		macoffset = maccrd->crd_skip;
1091 
1092 		if (maccrd->crd_alg == CRYPTO_MD5_HMAC_96)
1093 			ctx.pc_flags |= htole16(UBS_PKTCTX_AUTH_MD5);
1094 		else
1095 			ctx.pc_flags |= htole16(UBS_PKTCTX_AUTH_SHA1);
1096 
1097 		for (i = 0; i < 5; i++) {
1098 			ctx.pc_hminner[i] = ses->ses_hminner[i];
1099 			ctx.pc_hmouter[i] = ses->ses_hmouter[i];
1100 
1101 			HTOLE32(ctx.pc_hminner[i]);
1102 			HTOLE32(ctx.pc_hmouter[i]);
1103 		}
1104 	}
1105 
1106 	if (enccrd && maccrd) {
1107 		/*
1108 		 * ubsec cannot handle packets where the end of encryption
1109 		 * and authentication are not the same, or where the
1110 		 * encrypted part begins before the authenticated part.
1111 		 */
1112 		if ((encoffset + enccrd->crd_len) !=
1113 		    (macoffset + maccrd->crd_len)) {
1114 			ubsecstats.hst_lenmismatch++;
1115 			err = EINVAL;
1116 			goto errout;
1117 		}
1118 		if (enccrd->crd_skip < maccrd->crd_skip) {
1119 			ubsecstats.hst_skipmismatch++;
1120 			err = EINVAL;
1121 			goto errout;
1122 		}
1123 		sskip = maccrd->crd_skip;
1124 		cpskip = dskip = enccrd->crd_skip;
1125 		stheend = maccrd->crd_len;
1126 		dtheend = enccrd->crd_len;
1127 		coffset = enccrd->crd_skip - maccrd->crd_skip;
1128 		cpoffset = cpskip + dtheend;
1129 #ifdef UBSEC_DEBUG
1130 		if (ubsec_debug) {
1131 			printf("mac: skip %d, len %d, inject %d\n",
1132 			       maccrd->crd_skip, maccrd->crd_len, maccrd->crd_inject);
1133 			printf("enc: skip %d, len %d, inject %d\n",
1134 			       enccrd->crd_skip, enccrd->crd_len, enccrd->crd_inject);
1135 			printf("src: skip %d, len %d\n", sskip, stheend);
1136 			printf("dst: skip %d, len %d\n", dskip, dtheend);
1137 			printf("ubs: coffset %d, pktlen %d, cpskip %d, cpoffset %d\n",
1138 			       coffset, stheend, cpskip, cpoffset);
1139 		}
1140 #endif
1141 	} else {
1142 		cpskip = dskip = sskip = macoffset + encoffset;
1143 		dtheend = stheend = (enccrd)?enccrd->crd_len:maccrd->crd_len;
1144 		cpoffset = cpskip + dtheend;
1145 		coffset = 0;
1146 	}
1147 	ctx.pc_offset = htole16(coffset >> 2);
1148 
1149 	/* XXX FIXME: jonathan asks, what the heck's that 0xfff0?  */
1150 	if (bus_dmamap_create(sc->sc_dmat, 0xfff0, UBS_MAX_SCATTER,
1151 		0xfff0, 0, BUS_DMA_NOWAIT, &q->q_src_map) != 0) {
1152 		err = ENOMEM;
1153 		goto errout;
1154 	}
1155 	if (crp->crp_flags & CRYPTO_F_IMBUF) {
1156 		if (bus_dmamap_load_mbuf(sc->sc_dmat, q->q_src_map,
1157 		    q->q_src_m, BUS_DMA_NOWAIT) != 0) {
1158 			bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
1159 			q->q_src_map = NULL;
1160 			ubsecstats.hst_noload++;
1161 			err = ENOMEM;
1162 			goto errout;
1163 		}
1164 	} else if (crp->crp_flags & CRYPTO_F_IOV) {
1165 		if (bus_dmamap_load_uio(sc->sc_dmat, q->q_src_map,
1166 		    q->q_src_io, BUS_DMA_NOWAIT) != 0) {
1167 			bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
1168 			q->q_src_map = NULL;
1169 			ubsecstats.hst_noload++;
1170 			err = ENOMEM;
1171 			goto errout;
1172 		}
1173 	}
1174 	nicealign = ubsec_dmamap_aligned(q->q_src_map);
1175 
1176 	dmap->d_dma->d_mcr.mcr_pktlen = htole16(stheend);
1177 
1178 #ifdef UBSEC_DEBUG
1179 	if (ubsec_debug)
1180 		printf("src skip: %d nicealign: %u\n", sskip, nicealign);
1181 #endif
1182 	for (i = j = 0; i < q->q_src_map->dm_nsegs; i++) {
1183 		struct ubsec_pktbuf *pb;
1184 		bus_size_t packl = q->q_src_map->dm_segs[i].ds_len;
1185 		bus_addr_t packp = q->q_src_map->dm_segs[i].ds_addr;
1186 
1187 		if (sskip >= packl) {
1188 			sskip -= packl;
1189 			continue;
1190 		}
1191 
1192 		packl -= sskip;
1193 		packp += sskip;
1194 		sskip = 0;
1195 
1196 		if (packl > 0xfffc) {
1197 			err = EIO;
1198 			goto errout;
1199 		}
1200 
1201 		if (j == 0)
1202 			pb = &dmap->d_dma->d_mcr.mcr_ipktbuf;
1203 		else
1204 			pb = &dmap->d_dma->d_sbuf[j - 1];
1205 
1206 		pb->pb_addr = htole32(packp);
1207 
1208 		if (stheend) {
1209 			if (packl > stheend) {
1210 				pb->pb_len = htole32(stheend);
1211 				stheend = 0;
1212 			} else {
1213 				pb->pb_len = htole32(packl);
1214 				stheend -= packl;
1215 			}
1216 		} else
1217 			pb->pb_len = htole32(packl);
1218 
1219 		if ((i + 1) == q->q_src_map->dm_nsegs)
1220 			pb->pb_next = 0;
1221 		else
1222 			pb->pb_next = htole32(dmap->d_alloc.dma_paddr +
1223 			    offsetof(struct ubsec_dmachunk, d_sbuf[j]));
1224 		j++;
1225 	}
1226 
1227 	if (enccrd == NULL && maccrd != NULL) {
1228 		dmap->d_dma->d_mcr.mcr_opktbuf.pb_addr = 0;
1229 		dmap->d_dma->d_mcr.mcr_opktbuf.pb_len = 0;
1230 		dmap->d_dma->d_mcr.mcr_opktbuf.pb_next = htole32(dmap->d_alloc.dma_paddr +
1231 		    offsetof(struct ubsec_dmachunk, d_macbuf[0]));
1232 #ifdef UBSEC_DEBUG
1233 		if (ubsec_debug)
1234 			printf("opkt: %x %x %x\n",
1235 	 		    dmap->d_dma->d_mcr.mcr_opktbuf.pb_addr,
1236 	 		    dmap->d_dma->d_mcr.mcr_opktbuf.pb_len,
1237 	 		    dmap->d_dma->d_mcr.mcr_opktbuf.pb_next);
1238 
1239 #endif
1240 	} else {
1241 		if (crp->crp_flags & CRYPTO_F_IOV) {
1242 			if (!nicealign) {
1243 				ubsecstats.hst_iovmisaligned++;
1244 				err = EINVAL;
1245 				goto errout;
1246 			}
1247 			/* XXX: ``what the heck's that'' 0xfff0? */
1248 			if (bus_dmamap_create(sc->sc_dmat, 0xfff0,
1249 			    UBS_MAX_SCATTER, 0xfff0, 0, BUS_DMA_NOWAIT,
1250 			    &q->q_dst_map) != 0) {
1251 				ubsecstats.hst_nomap++;
1252 				err = ENOMEM;
1253 				goto errout;
1254 			}
1255 			if (bus_dmamap_load_uio(sc->sc_dmat, q->q_dst_map,
1256 			    q->q_dst_io, BUS_DMA_NOWAIT) != 0) {
1257 				bus_dmamap_destroy(sc->sc_dmat, q->q_dst_map);
1258 				q->q_dst_map = NULL;
1259 				ubsecstats.hst_noload++;
1260 				err = ENOMEM;
1261 				goto errout;
1262 			}
1263 		} else if (crp->crp_flags & CRYPTO_F_IMBUF) {
1264 			if (nicealign) {
1265 				q->q_dst_m = q->q_src_m;
1266 				q->q_dst_map = q->q_src_map;
1267 			} else {
1268 				int totlen, len;
1269 				struct mbuf *m, *top, **mp;
1270 
1271 				ubsecstats.hst_unaligned++;
1272 				totlen = q->q_src_map->dm_mapsize;
1273 				if (q->q_src_m->m_flags & M_PKTHDR) {
1274 					len = MHLEN;
1275 					MGETHDR(m, M_DONTWAIT, MT_DATA);
1276 					/*XXX FIXME: m_dup_pkthdr */
1277 					if (m && 1 /*!m_dup_pkthdr(m, q->q_src_m, M_DONTWAIT)*/) {
1278 						m_free(m);
1279 						m = NULL;
1280 					}
1281 				} else {
1282 					len = MLEN;
1283 					MGET(m, M_DONTWAIT, MT_DATA);
1284 				}
1285 				if (m == NULL) {
1286 					ubsecstats.hst_nombuf++;
1287 					err = sc->sc_nqueue ? ERESTART : ENOMEM;
1288 					goto errout;
1289 				}
1290 				if (len == MHLEN)
1291 				  /*XXX was M_DUP_PKTHDR*/
1292 				  M_COPY_PKTHDR(m, q->q_src_m);
1293 				if (totlen >= MINCLSIZE) {
1294 					MCLGET(m, M_DONTWAIT);
1295 					if ((m->m_flags & M_EXT) == 0) {
1296 						m_free(m);
1297 						ubsecstats.hst_nomcl++;
1298 						err = sc->sc_nqueue ? ERESTART : ENOMEM;
1299 						goto errout;
1300 					}
1301 					len = MCLBYTES;
1302 				}
1303 				m->m_len = len;
1304 				top = NULL;
1305 				mp = &top;
1306 
1307 				while (totlen > 0) {
1308 					if (top) {
1309 						MGET(m, M_DONTWAIT, MT_DATA);
1310 						if (m == NULL) {
1311 							m_freem(top);
1312 							ubsecstats.hst_nombuf++;
1313 							err = sc->sc_nqueue ? ERESTART : ENOMEM;
1314 							goto errout;
1315 						}
1316 						len = MLEN;
1317 					}
1318 					if (top && totlen >= MINCLSIZE) {
1319 						MCLGET(m, M_DONTWAIT);
1320 						if ((m->m_flags & M_EXT) == 0) {
1321 							*mp = m;
1322 							m_freem(top);
1323 							ubsecstats.hst_nomcl++;
1324 							err = sc->sc_nqueue ? ERESTART : ENOMEM;
1325 							goto errout;
1326 						}
1327 						len = MCLBYTES;
1328 					}
1329 					m->m_len = len = min(totlen, len);
1330 					totlen -= len;
1331 					*mp = m;
1332 					mp = &m->m_next;
1333 				}
1334 				q->q_dst_m = top;
1335 				ubsec_mcopy(q->q_src_m, q->q_dst_m,
1336 				    cpskip, cpoffset);
1337 				/* XXX again, what the heck is that 0xfff0? */
1338 				if (bus_dmamap_create(sc->sc_dmat, 0xfff0,
1339 				    UBS_MAX_SCATTER, 0xfff0, 0, BUS_DMA_NOWAIT,
1340 				    &q->q_dst_map) != 0) {
1341 					ubsecstats.hst_nomap++;
1342 					err = ENOMEM;
1343 					goto errout;
1344 				}
1345 				if (bus_dmamap_load_mbuf(sc->sc_dmat,
1346 				    q->q_dst_map, q->q_dst_m,
1347 				    BUS_DMA_NOWAIT) != 0) {
1348 					bus_dmamap_destroy(sc->sc_dmat,
1349 					q->q_dst_map);
1350 					q->q_dst_map = NULL;
1351 					ubsecstats.hst_noload++;
1352 					err = ENOMEM;
1353 					goto errout;
1354 				}
1355 			}
1356 		} else {
1357 			ubsecstats.hst_badflags++;
1358 			err = EINVAL;
1359 			goto errout;
1360 		}
1361 
1362 #ifdef UBSEC_DEBUG
1363 		if (ubsec_debug)
1364 			printf("dst skip: %d\n", dskip);
1365 #endif
1366 		for (i = j = 0; i < q->q_dst_map->dm_nsegs; i++) {
1367 			struct ubsec_pktbuf *pb;
1368 			bus_size_t packl = q->q_dst_map->dm_segs[i].ds_len;
1369 			bus_addr_t packp = q->q_dst_map->dm_segs[i].ds_addr;
1370 
1371 			if (dskip >= packl) {
1372 				dskip -= packl;
1373 				continue;
1374 			}
1375 
1376 			packl -= dskip;
1377 			packp += dskip;
1378 			dskip = 0;
1379 
1380 			if (packl > 0xfffc) {
1381 				err = EIO;
1382 				goto errout;
1383 			}
1384 
1385 			if (j == 0)
1386 				pb = &dmap->d_dma->d_mcr.mcr_opktbuf;
1387 			else
1388 				pb = &dmap->d_dma->d_dbuf[j - 1];
1389 
1390 			pb->pb_addr = htole32(packp);
1391 
1392 			if (dtheend) {
1393 				if (packl > dtheend) {
1394 					pb->pb_len = htole32(dtheend);
1395 					dtheend = 0;
1396 				} else {
1397 					pb->pb_len = htole32(packl);
1398 					dtheend -= packl;
1399 				}
1400 			} else
1401 				pb->pb_len = htole32(packl);
1402 
1403 			if ((i + 1) == q->q_dst_map->dm_nsegs) {
1404 				if (maccrd)
1405 					pb->pb_next = htole32(dmap->d_alloc.dma_paddr +
1406 					    offsetof(struct ubsec_dmachunk, d_macbuf[0]));
1407 				else
1408 					pb->pb_next = 0;
1409 			} else
1410 				pb->pb_next = htole32(dmap->d_alloc.dma_paddr +
1411 				    offsetof(struct ubsec_dmachunk, d_dbuf[j]));
1412 			j++;
1413 		}
1414 	}
1415 
1416 	dmap->d_dma->d_mcr.mcr_cmdctxp = htole32(dmap->d_alloc.dma_paddr +
1417 	    offsetof(struct ubsec_dmachunk, d_ctx));
1418 
1419 	if (sc->sc_flags & UBS_FLAGS_LONGCTX) {
1420 		struct ubsec_pktctx_long *ctxl;
1421 
1422 		ctxl = (struct ubsec_pktctx_long *)((char *)dmap->d_alloc.dma_vaddr +
1423 		    offsetof(struct ubsec_dmachunk, d_ctx));
1424 
1425 		/* transform small context into long context */
1426 		ctxl->pc_len = htole16(sizeof(struct ubsec_pktctx_long));
1427 		ctxl->pc_type = htole16(UBS_PKTCTX_TYPE_IPSEC);
1428 		ctxl->pc_flags = ctx.pc_flags;
1429 		ctxl->pc_offset = ctx.pc_offset;
1430 		for (i = 0; i < 6; i++)
1431 			ctxl->pc_deskey[i] = ctx.pc_deskey[i];
1432 		for (i = 0; i < 5; i++)
1433 			ctxl->pc_hminner[i] = ctx.pc_hminner[i];
1434 		for (i = 0; i < 5; i++)
1435 			ctxl->pc_hmouter[i] = ctx.pc_hmouter[i];
1436 		ctxl->pc_iv[0] = ctx.pc_iv[0];
1437 		ctxl->pc_iv[1] = ctx.pc_iv[1];
1438 	} else
1439 		memcpy((char *)dmap->d_alloc.dma_vaddr +
1440 		    offsetof(struct ubsec_dmachunk, d_ctx), &ctx,
1441 		    sizeof(struct ubsec_pktctx));
1442 
1443 	s = splnet();
1444 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue, q, q_next);
1445 	sc->sc_nqueue++;
1446 	ubsecstats.hst_ipackets++;
1447 	ubsecstats.hst_ibytes += dmap->d_alloc.dma_map->dm_mapsize;
1448 	if ((hint & CRYPTO_HINT_MORE) == 0 || sc->sc_nqueue >= ubsec_maxbatch)
1449 		ubsec_feed(sc);
1450 	splx(s);
1451 	return (0);
1452 
1453 errout:
1454 	if (q != NULL) {
1455 		if ((q->q_dst_m != NULL) && (q->q_src_m != q->q_dst_m))
1456 			m_freem(q->q_dst_m);
1457 
1458 		if (q->q_dst_map != NULL && q->q_dst_map != q->q_src_map) {
1459 			bus_dmamap_unload(sc->sc_dmat, q->q_dst_map);
1460 			bus_dmamap_destroy(sc->sc_dmat, q->q_dst_map);
1461 		}
1462 		if (q->q_src_map != NULL) {
1463 			bus_dmamap_unload(sc->sc_dmat, q->q_src_map);
1464 			bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
1465 		}
1466 
1467 		s = splnet();
1468 		SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
1469 		splx(s);
1470 	}
1471 #if 0 /* jonathan says: this openbsd code seems to be subsumed elsewhere */
1472 	if (err == EINVAL)
1473 		ubsecstats.hst_invalid++;
1474 	else
1475 		ubsecstats.hst_nomem++;
1476 #endif
1477 	if (err != ERESTART) {
1478 		crp->crp_etype = err;
1479 		crypto_done(crp);
1480 	} else {
1481 		sc->sc_needwakeup |= CRYPTO_SYMQ;
1482 	}
1483 	return (err);
1484 }
1485 
1486 static void
1487 ubsec_callback(struct ubsec_softc *sc, struct ubsec_q *q)
1488 {
1489 	struct cryptop *crp = (struct cryptop *)q->q_crp;
1490 	struct cryptodesc *crd;
1491 	struct ubsec_dma *dmap = q->q_dma;
1492 
1493 	ubsecstats.hst_opackets++;
1494 	ubsecstats.hst_obytes += dmap->d_alloc.dma_size;
1495 
1496 	bus_dmamap_sync(sc->sc_dmat, dmap->d_alloc.dma_map, 0,
1497 	    dmap->d_alloc.dma_map->dm_mapsize,
1498 	    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1499 	if (q->q_dst_map != NULL && q->q_dst_map != q->q_src_map) {
1500 		bus_dmamap_sync(sc->sc_dmat, q->q_dst_map,
1501 		    0, q->q_dst_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1502 		bus_dmamap_unload(sc->sc_dmat, q->q_dst_map);
1503 		bus_dmamap_destroy(sc->sc_dmat, q->q_dst_map);
1504 	}
1505 	bus_dmamap_sync(sc->sc_dmat, q->q_src_map,
1506 	    0, q->q_src_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1507 	bus_dmamap_unload(sc->sc_dmat, q->q_src_map);
1508 	bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
1509 
1510 	if ((crp->crp_flags & CRYPTO_F_IMBUF) && (q->q_src_m != q->q_dst_m)) {
1511 		m_freem(q->q_src_m);
1512 		crp->crp_buf = (void *)q->q_dst_m;
1513 	}
1514 
1515 	/* copy out IV for future use */
1516 	if (q->q_flags & UBSEC_QFLAGS_COPYOUTIV) {
1517 		for (crd = crp->crp_desc; crd; crd = crd->crd_next) {
1518 			if (crd->crd_alg != CRYPTO_DES_CBC &&
1519 			    crd->crd_alg != CRYPTO_3DES_CBC)
1520 				continue;
1521 			if (crp->crp_flags & CRYPTO_F_IMBUF)
1522 				m_copydata((struct mbuf *)crp->crp_buf,
1523 				    crd->crd_skip + crd->crd_len - 8, 8,
1524 				    (void *)sc->sc_sessions[q->q_sesn].ses_iv);
1525 			else if (crp->crp_flags & CRYPTO_F_IOV) {
1526 				cuio_copydata((struct uio *)crp->crp_buf,
1527 				    crd->crd_skip + crd->crd_len - 8, 8,
1528 				    (void *)sc->sc_sessions[q->q_sesn].ses_iv);
1529 			}
1530 			break;
1531 		}
1532 	}
1533 
1534 	for (crd = crp->crp_desc; crd; crd = crd->crd_next) {
1535 		if (crd->crd_alg != CRYPTO_MD5_HMAC_96 &&
1536 		    crd->crd_alg != CRYPTO_SHA1_HMAC_96)
1537 			continue;
1538 		if (crp->crp_flags & CRYPTO_F_IMBUF)
1539 			m_copyback((struct mbuf *)crp->crp_buf,
1540 			    crd->crd_inject, 12,
1541 			    (void *)dmap->d_dma->d_macbuf);
1542 		else if (crp->crp_flags & CRYPTO_F_IOV && crp->crp_mac)
1543 			bcopy((void *)dmap->d_dma->d_macbuf,
1544 			    crp->crp_mac, 12);
1545 		break;
1546 	}
1547 	SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
1548 	crypto_done(crp);
1549 }
1550 
1551 static void
1552 ubsec_mcopy(struct mbuf *srcm, struct mbuf *dstm, int hoffset, int toffset)
1553 {
1554 	int i, j, dlen, slen;
1555 	char *dptr, *sptr;
1556 
1557 	j = 0;
1558 	sptr = srcm->m_data;
1559 	slen = srcm->m_len;
1560 	dptr = dstm->m_data;
1561 	dlen = dstm->m_len;
1562 
1563 	while (1) {
1564 		for (i = 0; i < min(slen, dlen); i++) {
1565 			if (j < hoffset || j >= toffset)
1566 				*dptr++ = *sptr++;
1567 			slen--;
1568 			dlen--;
1569 			j++;
1570 		}
1571 		if (slen == 0) {
1572 			srcm = srcm->m_next;
1573 			if (srcm == NULL)
1574 				return;
1575 			sptr = srcm->m_data;
1576 			slen = srcm->m_len;
1577 		}
1578 		if (dlen == 0) {
1579 			dstm = dstm->m_next;
1580 			if (dstm == NULL)
1581 				return;
1582 			dptr = dstm->m_data;
1583 			dlen = dstm->m_len;
1584 		}
1585 	}
1586 }
1587 
1588 /*
1589  * feed the key generator, must be called at splnet() or higher.
1590  */
1591 static void
1592 ubsec_feed2(struct ubsec_softc *sc)
1593 {
1594 	struct ubsec_q2 *q;
1595 
1596 	while (!SIMPLEQ_EMPTY(&sc->sc_queue2)) {
1597 		if (READ_REG(sc, BS_STAT) & BS_STAT_MCR2_FULL)
1598 			break;
1599 		q = SIMPLEQ_FIRST(&sc->sc_queue2);
1600 
1601 		bus_dmamap_sync(sc->sc_dmat, q->q_mcr.dma_map, 0,
1602 		    q->q_mcr.dma_map->dm_mapsize,
1603 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1604 		bus_dmamap_sync(sc->sc_dmat, q->q_ctx.dma_map, 0,
1605 		    q->q_ctx.dma_map->dm_mapsize,
1606 		    BUS_DMASYNC_PREWRITE);
1607 
1608 		WRITE_REG(sc, BS_MCR2, q->q_mcr.dma_paddr);
1609 		q = SIMPLEQ_FIRST(&sc->sc_queue2);
1610 		SIMPLEQ_REMOVE_HEAD(&sc->sc_queue2, /*q,*/ q_next);
1611 		--sc->sc_nqueue2;
1612 		SIMPLEQ_INSERT_TAIL(&sc->sc_qchip2, q, q_next);
1613 	}
1614 }
1615 
1616 /*
1617  * Callback for handling random numbers
1618  */
1619 static void
1620 ubsec_callback2(struct ubsec_softc *sc, struct ubsec_q2 *q)
1621 {
1622 	struct cryptkop *krp;
1623 	struct ubsec_ctx_keyop *ctx;
1624 
1625 	ctx = (struct ubsec_ctx_keyop *)q->q_ctx.dma_vaddr;
1626 	bus_dmamap_sync(sc->sc_dmat, q->q_ctx.dma_map, 0,
1627 	    q->q_ctx.dma_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1628 
1629 	switch (q->q_type) {
1630 #ifndef UBSEC_NO_RNG
1631 	case UBS_CTXOP_RNGSHA1:
1632 	case UBS_CTXOP_RNGBYPASS: {
1633 		struct ubsec_q2_rng *rng = (struct ubsec_q2_rng *)q;
1634 		u_int32_t *p;
1635 		int i;
1636 
1637 		bus_dmamap_sync(sc->sc_dmat, rng->rng_buf.dma_map, 0,
1638 		    rng->rng_buf.dma_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1639 		p = (u_int32_t *)rng->rng_buf.dma_vaddr;
1640 #ifndef __NetBSD__
1641 		for (i = 0; i < UBSEC_RNG_BUFSIZ; p++, i++)
1642 			add_true_randomness(letoh32(*p));
1643 		rng->rng_used = 0;
1644 #else
1645 		/* XXX NetBSD rnd subsystem too weak */
1646 		i = 0; (void)i;	/* shut off gcc warnings */
1647 #endif
1648 #ifdef __OpenBSD__
1649 		timeout_add(&sc->sc_rngto, sc->sc_rnghz);
1650 #else
1651 		callout_reset(&sc->sc_rngto, sc->sc_rnghz, ubsec_rng, sc);
1652 #endif
1653 		break;
1654 	}
1655 #endif
1656 	case UBS_CTXOP_MODEXP: {
1657 		struct ubsec_q2_modexp *me = (struct ubsec_q2_modexp *)q;
1658 		u_int rlen, clen;
1659 
1660 		krp = me->me_krp;
1661 		rlen = (me->me_modbits + 7) / 8;
1662 		clen = (krp->krp_param[krp->krp_iparams].crp_nbits + 7) / 8;
1663 
1664 		bus_dmamap_sync(sc->sc_dmat, me->me_M.dma_map,
1665 		    0, me->me_M.dma_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1666 		bus_dmamap_sync(sc->sc_dmat, me->me_E.dma_map,
1667 		    0, me->me_E.dma_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1668 		bus_dmamap_sync(sc->sc_dmat, me->me_C.dma_map,
1669 		    0, me->me_C.dma_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1670 		bus_dmamap_sync(sc->sc_dmat, me->me_epb.dma_map,
1671 		    0, me->me_epb.dma_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1672 
1673 		if (clen < rlen)
1674 			krp->krp_status = E2BIG;
1675 		else {
1676 			if (sc->sc_flags & UBS_FLAGS_HWNORM) {
1677 				memset(krp->krp_param[krp->krp_iparams].crp_p, 0,
1678 				    (krp->krp_param[krp->krp_iparams].crp_nbits
1679 					+ 7) / 8);
1680 				bcopy(me->me_C.dma_vaddr,
1681 				    krp->krp_param[krp->krp_iparams].crp_p,
1682 				    (me->me_modbits + 7) / 8);
1683 			} else
1684 				ubsec_kshift_l(me->me_shiftbits,
1685 				    me->me_C.dma_vaddr, me->me_normbits,
1686 				    krp->krp_param[krp->krp_iparams].crp_p,
1687 				    krp->krp_param[krp->krp_iparams].crp_nbits);
1688 		}
1689 
1690 		crypto_kdone(krp);
1691 
1692 		/* bzero all potentially sensitive data */
1693 		memset(me->me_E.dma_vaddr, 0, me->me_E.dma_size);
1694 		memset(me->me_M.dma_vaddr, 0, me->me_M.dma_size);
1695 		memset(me->me_C.dma_vaddr, 0, me->me_C.dma_size);
1696 		memset(me->me_q.q_ctx.dma_vaddr, 0, me->me_q.q_ctx.dma_size);
1697 
1698 		/* Can't free here, so put us on the free list. */
1699 		SIMPLEQ_INSERT_TAIL(&sc->sc_q2free, &me->me_q, q_next);
1700 		break;
1701 	}
1702 	case UBS_CTXOP_RSAPRIV: {
1703 		struct ubsec_q2_rsapriv *rp = (struct ubsec_q2_rsapriv *)q;
1704 		u_int len;
1705 
1706 		krp = rp->rpr_krp;
1707 		bus_dmamap_sync(sc->sc_dmat, rp->rpr_msgin.dma_map, 0,
1708 		    rp->rpr_msgin.dma_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1709 		bus_dmamap_sync(sc->sc_dmat, rp->rpr_msgout.dma_map, 0,
1710 		    rp->rpr_msgout.dma_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1711 
1712 		len = (krp->krp_param[UBS_RSAPRIV_PAR_MSGOUT].crp_nbits + 7) / 8;
1713 		bcopy(rp->rpr_msgout.dma_vaddr,
1714 		    krp->krp_param[UBS_RSAPRIV_PAR_MSGOUT].crp_p, len);
1715 
1716 		crypto_kdone(krp);
1717 
1718 		memset(rp->rpr_msgin.dma_vaddr, 0, rp->rpr_msgin.dma_size);
1719 		memset(rp->rpr_msgout.dma_vaddr, 0, rp->rpr_msgout.dma_size);
1720 		memset(rp->rpr_q.q_ctx.dma_vaddr, 0, rp->rpr_q.q_ctx.dma_size);
1721 
1722 		/* Can't free here, so put us on the free list. */
1723 		SIMPLEQ_INSERT_TAIL(&sc->sc_q2free, &rp->rpr_q, q_next);
1724 		break;
1725 	}
1726 	default:
1727 		printf("%s: unknown ctx op: %x\n", device_xname(&sc->sc_dv),
1728 		    letoh16(ctx->ctx_op));
1729 		break;
1730 	}
1731 }
1732 
1733 #ifndef UBSEC_NO_RNG
1734 static void
1735 ubsec_rng(void *vsc)
1736 {
1737 	struct ubsec_softc *sc = vsc;
1738 	struct ubsec_q2_rng *rng = &sc->sc_rng;
1739 	struct ubsec_mcr *mcr;
1740 	struct ubsec_ctx_rngbypass *ctx;
1741 	int s;
1742 
1743 	s = splnet();
1744 	if (rng->rng_used) {
1745 		splx(s);
1746 		return;
1747 	}
1748 	sc->sc_nqueue2++;
1749 	if (sc->sc_nqueue2 >= UBS_MAX_NQUEUE)
1750 		goto out;
1751 
1752 	mcr = (struct ubsec_mcr *)rng->rng_q.q_mcr.dma_vaddr;
1753 	ctx = (struct ubsec_ctx_rngbypass *)rng->rng_q.q_ctx.dma_vaddr;
1754 
1755 	mcr->mcr_pkts = htole16(1);
1756 	mcr->mcr_flags = 0;
1757 	mcr->mcr_cmdctxp = htole32(rng->rng_q.q_ctx.dma_paddr);
1758 	mcr->mcr_ipktbuf.pb_addr = mcr->mcr_ipktbuf.pb_next = 0;
1759 	mcr->mcr_ipktbuf.pb_len = 0;
1760 	mcr->mcr_reserved = mcr->mcr_pktlen = 0;
1761 	mcr->mcr_opktbuf.pb_addr = htole32(rng->rng_buf.dma_paddr);
1762 	mcr->mcr_opktbuf.pb_len = htole32(((sizeof(u_int32_t) * UBSEC_RNG_BUFSIZ)) &
1763 	    UBS_PKTBUF_LEN);
1764 	mcr->mcr_opktbuf.pb_next = 0;
1765 
1766 	ctx->rbp_len = htole16(sizeof(struct ubsec_ctx_rngbypass));
1767 	ctx->rbp_op = htole16(UBS_CTXOP_RNGSHA1);
1768 	rng->rng_q.q_type = UBS_CTXOP_RNGSHA1;
1769 
1770 	bus_dmamap_sync(sc->sc_dmat, rng->rng_buf.dma_map, 0,
1771 	    rng->rng_buf.dma_map->dm_mapsize, BUS_DMASYNC_PREREAD);
1772 
1773 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue2, &rng->rng_q, q_next);
1774 	rng->rng_used = 1;
1775 	ubsec_feed2(sc);
1776 	ubsecstats.hst_rng++;
1777 	splx(s);
1778 
1779 	return;
1780 
1781 out:
1782 	/*
1783 	 * Something weird happened, generate our own call back.
1784 	 */
1785 	sc->sc_nqueue2--;
1786 	splx(s);
1787 #ifdef __OpenBSD__
1788 	timeout_add(&sc->sc_rngto, sc->sc_rnghz);
1789 #else
1790 	callout_reset(&sc->sc_rngto, sc->sc_rnghz, ubsec_rng, sc);
1791 #endif
1792 }
1793 #endif /* UBSEC_NO_RNG */
1794 
1795 static int
1796 ubsec_dma_malloc(struct ubsec_softc *sc, bus_size_t size,
1797 		 struct ubsec_dma_alloc *dma,int mapflags)
1798 {
1799 	int r;
1800 
1801 	if ((r = bus_dmamem_alloc(sc->sc_dmat, size, PAGE_SIZE, 0,
1802 	    &dma->dma_seg, 1, &dma->dma_nseg, BUS_DMA_NOWAIT)) != 0)
1803 		goto fail_0;
1804 
1805 	if ((r = bus_dmamem_map(sc->sc_dmat, &dma->dma_seg, dma->dma_nseg,
1806 	    size, &dma->dma_vaddr, mapflags | BUS_DMA_NOWAIT)) != 0)
1807 		goto fail_1;
1808 
1809 	if ((r = bus_dmamap_create(sc->sc_dmat, size, 1, size, 0,
1810 	    BUS_DMA_NOWAIT, &dma->dma_map)) != 0)
1811 		goto fail_2;
1812 
1813 	if ((r = bus_dmamap_load(sc->sc_dmat, dma->dma_map, dma->dma_vaddr,
1814 	    size, NULL, BUS_DMA_NOWAIT)) != 0)
1815 		goto fail_3;
1816 
1817 	dma->dma_paddr = dma->dma_map->dm_segs[0].ds_addr;
1818 	dma->dma_size = size;
1819 	return (0);
1820 
1821 fail_3:
1822 	bus_dmamap_destroy(sc->sc_dmat, dma->dma_map);
1823 fail_2:
1824 	bus_dmamem_unmap(sc->sc_dmat, dma->dma_vaddr, size);
1825 fail_1:
1826 	bus_dmamem_free(sc->sc_dmat, &dma->dma_seg, dma->dma_nseg);
1827 fail_0:
1828 	dma->dma_map = NULL;
1829 	return (r);
1830 }
1831 
1832 static void
1833 ubsec_dma_free(struct ubsec_softc *sc, struct ubsec_dma_alloc *dma)
1834 {
1835 	bus_dmamap_unload(sc->sc_dmat, dma->dma_map);
1836 	bus_dmamem_unmap(sc->sc_dmat, dma->dma_vaddr, dma->dma_size);
1837 	bus_dmamem_free(sc->sc_dmat, &dma->dma_seg, dma->dma_nseg);
1838 	bus_dmamap_destroy(sc->sc_dmat, dma->dma_map);
1839 }
1840 
1841 /*
1842  * Resets the board.  Values in the regesters are left as is
1843  * from the reset (i.e. initial values are assigned elsewhere).
1844  */
1845 static void
1846 ubsec_reset_board(struct ubsec_softc *sc)
1847 {
1848     volatile u_int32_t ctrl;
1849 
1850     ctrl = READ_REG(sc, BS_CTRL);
1851     ctrl |= BS_CTRL_RESET;
1852     WRITE_REG(sc, BS_CTRL, ctrl);
1853 
1854     /*
1855      * Wait aprox. 30 PCI clocks = 900 ns = 0.9 us
1856      */
1857     DELAY(10);
1858 }
1859 
1860 /*
1861  * Init Broadcom registers
1862  */
1863 static void
1864 ubsec_init_board(struct ubsec_softc *sc)
1865 {
1866 	u_int32_t ctrl;
1867 
1868 	ctrl = READ_REG(sc, BS_CTRL);
1869 	ctrl &= ~(BS_CTRL_BE32 | BS_CTRL_BE64);
1870 	ctrl |= BS_CTRL_LITTLE_ENDIAN | BS_CTRL_MCR1INT;
1871 
1872 	/*
1873 	 * XXX: Sam Leffler's code has (UBS_FLAGS_KEY|UBS_FLAGS_RNG)).
1874 	 * anyone got hw docs?
1875 	 */
1876 	if (sc->sc_flags & UBS_FLAGS_KEY)
1877 		ctrl |= BS_CTRL_MCR2INT;
1878 	else
1879 		ctrl &= ~BS_CTRL_MCR2INT;
1880 
1881 	if (sc->sc_flags & UBS_FLAGS_HWNORM)
1882 		ctrl &= ~BS_CTRL_SWNORM;
1883 
1884 	WRITE_REG(sc, BS_CTRL, ctrl);
1885 }
1886 
1887 /*
1888  * Init Broadcom PCI registers
1889  */
1890 static void
1891 ubsec_init_pciregs(struct pci_attach_args *pa)
1892 {
1893 	pci_chipset_tag_t pc = pa->pa_pc;
1894 	u_int32_t misc;
1895 
1896 	/*
1897 	 * This will set the cache line size to 1, this will
1898 	 * force the BCM58xx chip just to do burst read/writes.
1899 	 * Cache line read/writes are to slow
1900 	 */
1901 	misc = pci_conf_read(pc, pa->pa_tag, PCI_BHLC_REG);
1902 	misc = (misc & ~(PCI_CACHELINE_MASK << PCI_CACHELINE_SHIFT))
1903 	    | ((UBS_DEF_CACHELINE & 0xff) << PCI_CACHELINE_SHIFT);
1904 	pci_conf_write(pc, pa->pa_tag, PCI_BHLC_REG, misc);
1905 }
1906 
1907 /*
1908  * Clean up after a chip crash.
1909  * It is assumed that the caller in splnet()
1910  */
1911 static void
1912 ubsec_cleanchip(struct ubsec_softc *sc)
1913 {
1914 	struct ubsec_q *q;
1915 
1916 	while (!SIMPLEQ_EMPTY(&sc->sc_qchip)) {
1917 		q = SIMPLEQ_FIRST(&sc->sc_qchip);
1918 		SIMPLEQ_REMOVE_HEAD(&sc->sc_qchip, /*q,*/ q_next);
1919 		ubsec_free_q(sc, q);
1920 	}
1921 	sc->sc_nqchip = 0;
1922 }
1923 
1924 /*
1925  * free a ubsec_q
1926  * It is assumed that the caller is within splnet()
1927  */
1928 static int
1929 ubsec_free_q(struct ubsec_softc *sc, struct ubsec_q *q)
1930 {
1931 	struct ubsec_q *q2;
1932 	struct cryptop *crp;
1933 	int npkts;
1934 	int i;
1935 
1936 	npkts = q->q_nstacked_mcrs;
1937 
1938 	for (i = 0; i < npkts; i++) {
1939 		if(q->q_stacked_mcr[i]) {
1940 			q2 = q->q_stacked_mcr[i];
1941 
1942 			if ((q2->q_dst_m != NULL) && (q2->q_src_m != q2->q_dst_m))
1943 				m_freem(q2->q_dst_m);
1944 
1945 			crp = (struct cryptop *)q2->q_crp;
1946 
1947 			SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q2, q_next);
1948 
1949 			crp->crp_etype = EFAULT;
1950 			crypto_done(crp);
1951 		} else {
1952 			break;
1953 		}
1954 	}
1955 
1956 	/*
1957 	 * Free header MCR
1958 	 */
1959 	if ((q->q_dst_m != NULL) && (q->q_src_m != q->q_dst_m))
1960 		m_freem(q->q_dst_m);
1961 
1962 	crp = (struct cryptop *)q->q_crp;
1963 
1964 	SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
1965 
1966 	crp->crp_etype = EFAULT;
1967 	crypto_done(crp);
1968 	return(0);
1969 }
1970 
1971 /*
1972  * Routine to reset the chip and clean up.
1973  * It is assumed that the caller is in splnet()
1974  */
1975 static void
1976 ubsec_totalreset(struct ubsec_softc *sc)
1977 {
1978 	ubsec_reset_board(sc);
1979 	ubsec_init_board(sc);
1980 	ubsec_cleanchip(sc);
1981 }
1982 
1983 static int
1984 ubsec_dmamap_aligned(bus_dmamap_t map)
1985 {
1986 	int i;
1987 
1988 	for (i = 0; i < map->dm_nsegs; i++) {
1989 		if (map->dm_segs[i].ds_addr & 3)
1990 			return (0);
1991 		if ((i != (map->dm_nsegs - 1)) &&
1992 		    (map->dm_segs[i].ds_len & 3))
1993 			return (0);
1994 	}
1995 	return (1);
1996 }
1997 
1998 #ifdef __OpenBSD__
1999 struct ubsec_softc *
2000 ubsec_kfind(struct cryptkop *krp)
2001 {
2002 	struct ubsec_softc *sc;
2003 	int i;
2004 
2005 	for (i = 0; i < ubsec_cd.cd_ndevs; i++) {
2006 		sc = ubsec_cd.cd_devs[i];
2007 		if (sc == NULL)
2008 			continue;
2009 		if (sc->sc_cid == krp->krp_hid)
2010 			return (sc);
2011 	}
2012 	return (NULL);
2013 }
2014 #endif
2015 
2016 static void
2017 ubsec_kfree(struct ubsec_softc *sc, struct ubsec_q2 *q)
2018 {
2019 	switch (q->q_type) {
2020 	case UBS_CTXOP_MODEXP: {
2021 		struct ubsec_q2_modexp *me = (struct ubsec_q2_modexp *)q;
2022 
2023 		ubsec_dma_free(sc, &me->me_q.q_mcr);
2024 		ubsec_dma_free(sc, &me->me_q.q_ctx);
2025 		ubsec_dma_free(sc, &me->me_M);
2026 		ubsec_dma_free(sc, &me->me_E);
2027 		ubsec_dma_free(sc, &me->me_C);
2028 		ubsec_dma_free(sc, &me->me_epb);
2029 		free(me, M_DEVBUF);
2030 		break;
2031 	}
2032 	case UBS_CTXOP_RSAPRIV: {
2033 		struct ubsec_q2_rsapriv *rp = (struct ubsec_q2_rsapriv *)q;
2034 
2035 		ubsec_dma_free(sc, &rp->rpr_q.q_mcr);
2036 		ubsec_dma_free(sc, &rp->rpr_q.q_ctx);
2037 		ubsec_dma_free(sc, &rp->rpr_msgin);
2038 		ubsec_dma_free(sc, &rp->rpr_msgout);
2039 		free(rp, M_DEVBUF);
2040 		break;
2041 	}
2042 	default:
2043 		printf("%s: invalid kfree 0x%x\n", device_xname(&sc->sc_dv),
2044 		    q->q_type);
2045 		break;
2046 	}
2047 }
2048 
2049 static int
2050 ubsec_kprocess(void *arg, struct cryptkop *krp, int hint)
2051 {
2052 	struct ubsec_softc *sc;
2053 	int r;
2054 
2055 	if (krp == NULL || krp->krp_callback == NULL)
2056 		return (EINVAL);
2057 #ifdef __OpenBSD__
2058 	if ((sc = ubsec_kfind(krp)) == NULL)
2059 		return (EINVAL);
2060 #else
2061 	sc = arg;
2062 	KASSERT(sc != NULL /*, ("ubsec_kprocess: null softc")*/);
2063 #endif
2064 
2065 	while (!SIMPLEQ_EMPTY(&sc->sc_q2free)) {
2066 		struct ubsec_q2 *q;
2067 
2068 		q = SIMPLEQ_FIRST(&sc->sc_q2free);
2069 		SIMPLEQ_REMOVE_HEAD(&sc->sc_q2free, /*q,*/ q_next);
2070 		ubsec_kfree(sc, q);
2071 	}
2072 
2073 	switch (krp->krp_op) {
2074 	case CRK_MOD_EXP:
2075 		if (sc->sc_flags & UBS_FLAGS_HWNORM)
2076 			r = ubsec_kprocess_modexp_hw(sc, krp, hint);
2077 		else
2078 			r = ubsec_kprocess_modexp_sw(sc, krp, hint);
2079 		break;
2080 	case CRK_MOD_EXP_CRT:
2081 		r = ubsec_kprocess_rsapriv(sc, krp, hint);
2082 		break;
2083 	default:
2084 		printf("%s: kprocess: invalid op 0x%x\n",
2085 		    device_xname(&sc->sc_dv), krp->krp_op);
2086 		krp->krp_status = EOPNOTSUPP;
2087 		crypto_kdone(krp);
2088 		r = 0;
2089 	}
2090 	return (r);
2091 }
2092 
2093 /*
2094  * Start computation of cr[C] = (cr[M] ^ cr[E]) mod cr[N] (sw normalization)
2095  */
2096 static int
2097 ubsec_kprocess_modexp_sw(struct ubsec_softc *sc, struct cryptkop *krp,
2098 			 int hint)
2099 {
2100 	struct ubsec_q2_modexp *me;
2101 	struct ubsec_mcr *mcr;
2102 	struct ubsec_ctx_modexp *ctx;
2103 	struct ubsec_pktbuf *epb;
2104 	int s, err = 0;
2105 	u_int nbits, normbits, mbits, shiftbits, ebits;
2106 
2107 	me = (struct ubsec_q2_modexp *)malloc(sizeof *me, M_DEVBUF, M_NOWAIT);
2108 	if (me == NULL) {
2109 		err = ENOMEM;
2110 		goto errout;
2111 	}
2112 	memset(me, 0, sizeof *me);
2113 	me->me_krp = krp;
2114 	me->me_q.q_type = UBS_CTXOP_MODEXP;
2115 
2116 	nbits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_N]);
2117 	if (nbits <= 512)
2118 		normbits = 512;
2119 	else if (nbits <= 768)
2120 		normbits = 768;
2121 	else if (nbits <= 1024)
2122 		normbits = 1024;
2123 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && nbits <= 1536)
2124 		normbits = 1536;
2125 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && nbits <= 2048)
2126 		normbits = 2048;
2127 	else {
2128 		err = E2BIG;
2129 		goto errout;
2130 	}
2131 
2132 	shiftbits = normbits - nbits;
2133 
2134 	me->me_modbits = nbits;
2135 	me->me_shiftbits = shiftbits;
2136 	me->me_normbits = normbits;
2137 
2138 	/* Sanity check: result bits must be >= true modulus bits. */
2139 	if (krp->krp_param[krp->krp_iparams].crp_nbits < nbits) {
2140 		err = ERANGE;
2141 		goto errout;
2142 	}
2143 
2144 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_mcr),
2145 	    &me->me_q.q_mcr, 0)) {
2146 		err = ENOMEM;
2147 		goto errout;
2148 	}
2149 	mcr = (struct ubsec_mcr *)me->me_q.q_mcr.dma_vaddr;
2150 
2151 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_ctx_modexp),
2152 	    &me->me_q.q_ctx, 0)) {
2153 		err = ENOMEM;
2154 		goto errout;
2155 	}
2156 
2157 	mbits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_M]);
2158 	if (mbits > nbits) {
2159 		err = E2BIG;
2160 		goto errout;
2161 	}
2162 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_M, 0)) {
2163 		err = ENOMEM;
2164 		goto errout;
2165 	}
2166 	ubsec_kshift_r(shiftbits,
2167 	    krp->krp_param[UBS_MODEXP_PAR_M].crp_p, mbits,
2168 	    me->me_M.dma_vaddr, normbits);
2169 
2170 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_C, 0)) {
2171 		err = ENOMEM;
2172 		goto errout;
2173 	}
2174 	memset(me->me_C.dma_vaddr, 0, me->me_C.dma_size);
2175 
2176 	ebits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_E]);
2177 	if (ebits > nbits) {
2178 		err = E2BIG;
2179 		goto errout;
2180 	}
2181 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_E, 0)) {
2182 		err = ENOMEM;
2183 		goto errout;
2184 	}
2185 	ubsec_kshift_r(shiftbits,
2186 	    krp->krp_param[UBS_MODEXP_PAR_E].crp_p, ebits,
2187 	    me->me_E.dma_vaddr, normbits);
2188 
2189 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_pktbuf),
2190 	    &me->me_epb, 0)) {
2191 		err = ENOMEM;
2192 		goto errout;
2193 	}
2194 	epb = (struct ubsec_pktbuf *)me->me_epb.dma_vaddr;
2195 	epb->pb_addr = htole32(me->me_E.dma_paddr);
2196 	epb->pb_next = 0;
2197 	epb->pb_len = htole32(normbits / 8);
2198 
2199 #ifdef UBSEC_DEBUG
2200 	if (ubsec_debug) {
2201 		printf("Epb ");
2202 		ubsec_dump_pb(epb);
2203 	}
2204 #endif
2205 
2206 	mcr->mcr_pkts = htole16(1);
2207 	mcr->mcr_flags = 0;
2208 	mcr->mcr_cmdctxp = htole32(me->me_q.q_ctx.dma_paddr);
2209 	mcr->mcr_reserved = 0;
2210 	mcr->mcr_pktlen = 0;
2211 
2212 	mcr->mcr_ipktbuf.pb_addr = htole32(me->me_M.dma_paddr);
2213 	mcr->mcr_ipktbuf.pb_len = htole32(normbits / 8);
2214 	mcr->mcr_ipktbuf.pb_next = htole32(me->me_epb.dma_paddr);
2215 
2216 	mcr->mcr_opktbuf.pb_addr = htole32(me->me_C.dma_paddr);
2217 	mcr->mcr_opktbuf.pb_next = 0;
2218 	mcr->mcr_opktbuf.pb_len = htole32(normbits / 8);
2219 
2220 #ifdef DIAGNOSTIC
2221 	/* Misaligned output buffer will hang the chip. */
2222 	if ((letoh32(mcr->mcr_opktbuf.pb_addr) & 3) != 0)
2223 		panic("%s: modexp invalid addr 0x%x",
2224 		    device_xname(&sc->sc_dv), letoh32(mcr->mcr_opktbuf.pb_addr));
2225 	if ((letoh32(mcr->mcr_opktbuf.pb_len) & 3) != 0)
2226 		panic("%s: modexp invalid len 0x%x",
2227 		    device_xname(&sc->sc_dv), letoh32(mcr->mcr_opktbuf.pb_len));
2228 #endif
2229 
2230 	ctx = (struct ubsec_ctx_modexp *)me->me_q.q_ctx.dma_vaddr;
2231 	memset(ctx, 0, sizeof(*ctx));
2232 	ubsec_kshift_r(shiftbits,
2233 	    krp->krp_param[UBS_MODEXP_PAR_N].crp_p, nbits,
2234 	    ctx->me_N, normbits);
2235 	ctx->me_len = htole16((normbits / 8) + (4 * sizeof(u_int16_t)));
2236 	ctx->me_op = htole16(UBS_CTXOP_MODEXP);
2237 	ctx->me_E_len = htole16(nbits);
2238 	ctx->me_N_len = htole16(nbits);
2239 
2240 #ifdef UBSEC_DEBUG
2241 	if (ubsec_debug) {
2242 		ubsec_dump_mcr(mcr);
2243 		ubsec_dump_ctx2((struct ubsec_ctx_keyop *)ctx);
2244 	}
2245 #endif
2246 
2247 	/*
2248 	 * ubsec_feed2 will sync mcr and ctx, we just need to sync
2249 	 * everything else.
2250 	 */
2251 	bus_dmamap_sync(sc->sc_dmat, me->me_M.dma_map,
2252 	    0, me->me_M.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2253 	bus_dmamap_sync(sc->sc_dmat, me->me_E.dma_map,
2254 	    0, me->me_E.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2255 	bus_dmamap_sync(sc->sc_dmat, me->me_C.dma_map,
2256 	    0, me->me_C.dma_map->dm_mapsize, BUS_DMASYNC_PREREAD);
2257 	bus_dmamap_sync(sc->sc_dmat, me->me_epb.dma_map,
2258 	    0, me->me_epb.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2259 
2260 	/* Enqueue and we're done... */
2261 	s = splnet();
2262 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue2, &me->me_q, q_next);
2263 	ubsec_feed2(sc);
2264 	ubsecstats.hst_modexp++;
2265 	splx(s);
2266 
2267 	return (0);
2268 
2269 errout:
2270 	if (me != NULL) {
2271 		if (me->me_q.q_mcr.dma_map != NULL)
2272 			ubsec_dma_free(sc, &me->me_q.q_mcr);
2273 		if (me->me_q.q_ctx.dma_map != NULL) {
2274 			memset(me->me_q.q_ctx.dma_vaddr, 0, me->me_q.q_ctx.dma_size);
2275 			ubsec_dma_free(sc, &me->me_q.q_ctx);
2276 		}
2277 		if (me->me_M.dma_map != NULL) {
2278 			memset(me->me_M.dma_vaddr, 0, me->me_M.dma_size);
2279 			ubsec_dma_free(sc, &me->me_M);
2280 		}
2281 		if (me->me_E.dma_map != NULL) {
2282 			memset(me->me_E.dma_vaddr, 0, me->me_E.dma_size);
2283 			ubsec_dma_free(sc, &me->me_E);
2284 		}
2285 		if (me->me_C.dma_map != NULL) {
2286 			memset(me->me_C.dma_vaddr, 0, me->me_C.dma_size);
2287 			ubsec_dma_free(sc, &me->me_C);
2288 		}
2289 		if (me->me_epb.dma_map != NULL)
2290 			ubsec_dma_free(sc, &me->me_epb);
2291 		free(me, M_DEVBUF);
2292 	}
2293 	krp->krp_status = err;
2294 	crypto_kdone(krp);
2295 	return (0);
2296 }
2297 
2298 /*
2299  * Start computation of cr[C] = (cr[M] ^ cr[E]) mod cr[N] (hw normalization)
2300  */
2301 static int
2302 ubsec_kprocess_modexp_hw(struct ubsec_softc *sc, struct cryptkop *krp,
2303 			 int hint)
2304 {
2305 	struct ubsec_q2_modexp *me;
2306 	struct ubsec_mcr *mcr;
2307 	struct ubsec_ctx_modexp *ctx;
2308 	struct ubsec_pktbuf *epb;
2309 	int s, err = 0;
2310 	u_int nbits, normbits, mbits, shiftbits, ebits;
2311 
2312 	me = (struct ubsec_q2_modexp *)malloc(sizeof *me, M_DEVBUF, M_NOWAIT);
2313 	if (me == NULL) {
2314 		err = ENOMEM;
2315 		goto errout;
2316 	}
2317 	memset(me, 0, sizeof *me);
2318 	me->me_krp = krp;
2319 	me->me_q.q_type = UBS_CTXOP_MODEXP;
2320 
2321 	nbits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_N]);
2322 	if (nbits <= 512)
2323 		normbits = 512;
2324 	else if (nbits <= 768)
2325 		normbits = 768;
2326 	else if (nbits <= 1024)
2327 		normbits = 1024;
2328 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && nbits <= 1536)
2329 		normbits = 1536;
2330 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && nbits <= 2048)
2331 		normbits = 2048;
2332 	else {
2333 		err = E2BIG;
2334 		goto errout;
2335 	}
2336 
2337 	shiftbits = normbits - nbits;
2338 
2339 	/* XXX ??? */
2340 	me->me_modbits = nbits;
2341 	me->me_shiftbits = shiftbits;
2342 	me->me_normbits = normbits;
2343 
2344 	/* Sanity check: result bits must be >= true modulus bits. */
2345 	if (krp->krp_param[krp->krp_iparams].crp_nbits < nbits) {
2346 		err = ERANGE;
2347 		goto errout;
2348 	}
2349 
2350 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_mcr),
2351 	    &me->me_q.q_mcr, 0)) {
2352 		err = ENOMEM;
2353 		goto errout;
2354 	}
2355 	mcr = (struct ubsec_mcr *)me->me_q.q_mcr.dma_vaddr;
2356 
2357 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_ctx_modexp),
2358 	    &me->me_q.q_ctx, 0)) {
2359 		err = ENOMEM;
2360 		goto errout;
2361 	}
2362 
2363 	mbits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_M]);
2364 	if (mbits > nbits) {
2365 		err = E2BIG;
2366 		goto errout;
2367 	}
2368 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_M, 0)) {
2369 		err = ENOMEM;
2370 		goto errout;
2371 	}
2372 	memset(me->me_M.dma_vaddr, 0, normbits / 8);
2373 	bcopy(krp->krp_param[UBS_MODEXP_PAR_M].crp_p,
2374 	    me->me_M.dma_vaddr, (mbits + 7) / 8);
2375 
2376 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_C, 0)) {
2377 		err = ENOMEM;
2378 		goto errout;
2379 	}
2380 	memset(me->me_C.dma_vaddr, 0, me->me_C.dma_size);
2381 
2382 	ebits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_E]);
2383 	if (ebits > nbits) {
2384 		err = E2BIG;
2385 		goto errout;
2386 	}
2387 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_E, 0)) {
2388 		err = ENOMEM;
2389 		goto errout;
2390 	}
2391 	memset(me->me_E.dma_vaddr, 0, normbits / 8);
2392 	bcopy(krp->krp_param[UBS_MODEXP_PAR_E].crp_p,
2393 	    me->me_E.dma_vaddr, (ebits + 7) / 8);
2394 
2395 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_pktbuf),
2396 	    &me->me_epb, 0)) {
2397 		err = ENOMEM;
2398 		goto errout;
2399 	}
2400 	epb = (struct ubsec_pktbuf *)me->me_epb.dma_vaddr;
2401 	epb->pb_addr = htole32(me->me_E.dma_paddr);
2402 	epb->pb_next = 0;
2403 	epb->pb_len = htole32((ebits + 7) / 8);
2404 
2405 #ifdef UBSEC_DEBUG
2406 	if (ubsec_debug) {
2407 		printf("Epb ");
2408 		ubsec_dump_pb(epb);
2409 	}
2410 #endif
2411 
2412 	mcr->mcr_pkts = htole16(1);
2413 	mcr->mcr_flags = 0;
2414 	mcr->mcr_cmdctxp = htole32(me->me_q.q_ctx.dma_paddr);
2415 	mcr->mcr_reserved = 0;
2416 	mcr->mcr_pktlen = 0;
2417 
2418 	mcr->mcr_ipktbuf.pb_addr = htole32(me->me_M.dma_paddr);
2419 	mcr->mcr_ipktbuf.pb_len = htole32(normbits / 8);
2420 	mcr->mcr_ipktbuf.pb_next = htole32(me->me_epb.dma_paddr);
2421 
2422 	mcr->mcr_opktbuf.pb_addr = htole32(me->me_C.dma_paddr);
2423 	mcr->mcr_opktbuf.pb_next = 0;
2424 	mcr->mcr_opktbuf.pb_len = htole32(normbits / 8);
2425 
2426 #ifdef DIAGNOSTIC
2427 	/* Misaligned output buffer will hang the chip. */
2428 	if ((letoh32(mcr->mcr_opktbuf.pb_addr) & 3) != 0)
2429 		panic("%s: modexp invalid addr 0x%x",
2430 		    device_xname(&sc->sc_dv), letoh32(mcr->mcr_opktbuf.pb_addr));
2431 	if ((letoh32(mcr->mcr_opktbuf.pb_len) & 3) != 0)
2432 		panic("%s: modexp invalid len 0x%x",
2433 		    device_xname(&sc->sc_dv), letoh32(mcr->mcr_opktbuf.pb_len));
2434 #endif
2435 
2436 	ctx = (struct ubsec_ctx_modexp *)me->me_q.q_ctx.dma_vaddr;
2437 	memset(ctx, 0, sizeof(*ctx));
2438 	memcpy(ctx->me_N, krp->krp_param[UBS_MODEXP_PAR_N].crp_p,
2439 	    (nbits + 7) / 8);
2440 	ctx->me_len = htole16((normbits / 8) + (4 * sizeof(u_int16_t)));
2441 	ctx->me_op = htole16(UBS_CTXOP_MODEXP);
2442 	ctx->me_E_len = htole16(ebits);
2443 	ctx->me_N_len = htole16(nbits);
2444 
2445 #ifdef UBSEC_DEBUG
2446 	if (ubsec_debug) {
2447 		ubsec_dump_mcr(mcr);
2448 		ubsec_dump_ctx2((struct ubsec_ctx_keyop *)ctx);
2449 	}
2450 #endif
2451 
2452 	/*
2453 	 * ubsec_feed2 will sync mcr and ctx, we just need to sync
2454 	 * everything else.
2455 	 */
2456 	bus_dmamap_sync(sc->sc_dmat, me->me_M.dma_map,
2457 	    0, me->me_M.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2458 	bus_dmamap_sync(sc->sc_dmat, me->me_E.dma_map,
2459 	    0, me->me_E.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2460 	bus_dmamap_sync(sc->sc_dmat, me->me_C.dma_map,
2461 	    0, me->me_C.dma_map->dm_mapsize, BUS_DMASYNC_PREREAD);
2462 	bus_dmamap_sync(sc->sc_dmat, me->me_epb.dma_map,
2463 	    0, me->me_epb.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2464 
2465 	/* Enqueue and we're done... */
2466 	s = splnet();
2467 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue2, &me->me_q, q_next);
2468 	ubsec_feed2(sc);
2469 	splx(s);
2470 
2471 	return (0);
2472 
2473 errout:
2474 	if (me != NULL) {
2475 		if (me->me_q.q_mcr.dma_map != NULL)
2476 			ubsec_dma_free(sc, &me->me_q.q_mcr);
2477 		if (me->me_q.q_ctx.dma_map != NULL) {
2478 			memset(me->me_q.q_ctx.dma_vaddr, 0, me->me_q.q_ctx.dma_size);
2479 			ubsec_dma_free(sc, &me->me_q.q_ctx);
2480 		}
2481 		if (me->me_M.dma_map != NULL) {
2482 			memset(me->me_M.dma_vaddr, 0, me->me_M.dma_size);
2483 			ubsec_dma_free(sc, &me->me_M);
2484 		}
2485 		if (me->me_E.dma_map != NULL) {
2486 			memset(me->me_E.dma_vaddr, 0, me->me_E.dma_size);
2487 			ubsec_dma_free(sc, &me->me_E);
2488 		}
2489 		if (me->me_C.dma_map != NULL) {
2490 			memset(me->me_C.dma_vaddr, 0, me->me_C.dma_size);
2491 			ubsec_dma_free(sc, &me->me_C);
2492 		}
2493 		if (me->me_epb.dma_map != NULL)
2494 			ubsec_dma_free(sc, &me->me_epb);
2495 		free(me, M_DEVBUF);
2496 	}
2497 	krp->krp_status = err;
2498 	crypto_kdone(krp);
2499 	return (0);
2500 }
2501 
2502 static int
2503 ubsec_kprocess_rsapriv(struct ubsec_softc *sc, struct cryptkop *krp,
2504 		       int hint)
2505 {
2506 	struct ubsec_q2_rsapriv *rp = NULL;
2507 	struct ubsec_mcr *mcr;
2508 	struct ubsec_ctx_rsapriv *ctx;
2509 	int s, err = 0;
2510 	u_int padlen, msglen;
2511 
2512 	msglen = ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_P]);
2513 	padlen = ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_Q]);
2514 	if (msglen > padlen)
2515 		padlen = msglen;
2516 
2517 	if (padlen <= 256)
2518 		padlen = 256;
2519 	else if (padlen <= 384)
2520 		padlen = 384;
2521 	else if (padlen <= 512)
2522 		padlen = 512;
2523 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && padlen <= 768)
2524 		padlen = 768;
2525 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && padlen <= 1024)
2526 		padlen = 1024;
2527 	else {
2528 		err = E2BIG;
2529 		goto errout;
2530 	}
2531 
2532 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_DP]) > padlen) {
2533 		err = E2BIG;
2534 		goto errout;
2535 	}
2536 
2537 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_DQ]) > padlen) {
2538 		err = E2BIG;
2539 		goto errout;
2540 	}
2541 
2542 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_PINV]) > padlen) {
2543 		err = E2BIG;
2544 		goto errout;
2545 	}
2546 
2547 	rp = malloc(sizeof *rp, M_DEVBUF, M_NOWAIT|M_ZERO);
2548 	if (rp == NULL)
2549 		return (ENOMEM);
2550 	rp->rpr_krp = krp;
2551 	rp->rpr_q.q_type = UBS_CTXOP_RSAPRIV;
2552 
2553 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_mcr),
2554 	    &rp->rpr_q.q_mcr, 0)) {
2555 		err = ENOMEM;
2556 		goto errout;
2557 	}
2558 	mcr = (struct ubsec_mcr *)rp->rpr_q.q_mcr.dma_vaddr;
2559 
2560 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_ctx_rsapriv),
2561 	    &rp->rpr_q.q_ctx, 0)) {
2562 		err = ENOMEM;
2563 		goto errout;
2564 	}
2565 	ctx = (struct ubsec_ctx_rsapriv *)rp->rpr_q.q_ctx.dma_vaddr;
2566 	memset(ctx, 0, sizeof *ctx);
2567 
2568 	/* Copy in p */
2569 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_P].crp_p,
2570 	    &ctx->rpr_buf[0 * (padlen / 8)],
2571 	    (krp->krp_param[UBS_RSAPRIV_PAR_P].crp_nbits + 7) / 8);
2572 
2573 	/* Copy in q */
2574 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_Q].crp_p,
2575 	    &ctx->rpr_buf[1 * (padlen / 8)],
2576 	    (krp->krp_param[UBS_RSAPRIV_PAR_Q].crp_nbits + 7) / 8);
2577 
2578 	/* Copy in dp */
2579 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_DP].crp_p,
2580 	    &ctx->rpr_buf[2 * (padlen / 8)],
2581 	    (krp->krp_param[UBS_RSAPRIV_PAR_DP].crp_nbits + 7) / 8);
2582 
2583 	/* Copy in dq */
2584 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_DQ].crp_p,
2585 	    &ctx->rpr_buf[3 * (padlen / 8)],
2586 	    (krp->krp_param[UBS_RSAPRIV_PAR_DQ].crp_nbits + 7) / 8);
2587 
2588 	/* Copy in pinv */
2589 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_PINV].crp_p,
2590 	    &ctx->rpr_buf[4 * (padlen / 8)],
2591 	    (krp->krp_param[UBS_RSAPRIV_PAR_PINV].crp_nbits + 7) / 8);
2592 
2593 	msglen = padlen * 2;
2594 
2595 	/* Copy in input message (aligned buffer/length). */
2596 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_MSGIN]) > msglen) {
2597 		/* Is this likely? */
2598 		err = E2BIG;
2599 		goto errout;
2600 	}
2601 	if (ubsec_dma_malloc(sc, (msglen + 7) / 8, &rp->rpr_msgin, 0)) {
2602 		err = ENOMEM;
2603 		goto errout;
2604 	}
2605 	memset(rp->rpr_msgin.dma_vaddr, 0, (msglen + 7) / 8);
2606 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_MSGIN].crp_p,
2607 	    rp->rpr_msgin.dma_vaddr,
2608 	    (krp->krp_param[UBS_RSAPRIV_PAR_MSGIN].crp_nbits + 7) / 8);
2609 
2610 	/* Prepare space for output message (aligned buffer/length). */
2611 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_MSGOUT]) < msglen) {
2612 		/* Is this likely? */
2613 		err = E2BIG;
2614 		goto errout;
2615 	}
2616 	if (ubsec_dma_malloc(sc, (msglen + 7) / 8, &rp->rpr_msgout, 0)) {
2617 		err = ENOMEM;
2618 		goto errout;
2619 	}
2620 	memset(rp->rpr_msgout.dma_vaddr, 0, (msglen + 7) / 8);
2621 
2622 	mcr->mcr_pkts = htole16(1);
2623 	mcr->mcr_flags = 0;
2624 	mcr->mcr_cmdctxp = htole32(rp->rpr_q.q_ctx.dma_paddr);
2625 	mcr->mcr_ipktbuf.pb_addr = htole32(rp->rpr_msgin.dma_paddr);
2626 	mcr->mcr_ipktbuf.pb_next = 0;
2627 	mcr->mcr_ipktbuf.pb_len = htole32(rp->rpr_msgin.dma_size);
2628 	mcr->mcr_reserved = 0;
2629 	mcr->mcr_pktlen = htole16(msglen);
2630 	mcr->mcr_opktbuf.pb_addr = htole32(rp->rpr_msgout.dma_paddr);
2631 	mcr->mcr_opktbuf.pb_next = 0;
2632 	mcr->mcr_opktbuf.pb_len = htole32(rp->rpr_msgout.dma_size);
2633 
2634 #ifdef DIAGNOSTIC
2635 	if (rp->rpr_msgin.dma_paddr & 3 || rp->rpr_msgin.dma_size & 3) {
2636 		panic("%s: rsapriv: invalid msgin 0x%lx(0x%lx)",
2637 		    device_xname(&sc->sc_dv), (u_long) rp->rpr_msgin.dma_paddr,
2638 		    (u_long) rp->rpr_msgin.dma_size);
2639 	}
2640 	if (rp->rpr_msgout.dma_paddr & 3 || rp->rpr_msgout.dma_size & 3) {
2641 		panic("%s: rsapriv: invalid msgout 0x%lx(0x%lx)",
2642 		    device_xname(&sc->sc_dv), (u_long) rp->rpr_msgout.dma_paddr,
2643 		    (u_long) rp->rpr_msgout.dma_size);
2644 	}
2645 #endif
2646 
2647 	ctx->rpr_len = (sizeof(u_int16_t) * 4) + (5 * (padlen / 8));
2648 	ctx->rpr_op = htole16(UBS_CTXOP_RSAPRIV);
2649 	ctx->rpr_q_len = htole16(padlen);
2650 	ctx->rpr_p_len = htole16(padlen);
2651 
2652 	/*
2653 	 * ubsec_feed2 will sync mcr and ctx, we just need to sync
2654 	 * everything else.
2655 	 */
2656 	bus_dmamap_sync(sc->sc_dmat, rp->rpr_msgin.dma_map,
2657 	    0, rp->rpr_msgin.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2658 	bus_dmamap_sync(sc->sc_dmat, rp->rpr_msgout.dma_map,
2659 	    0, rp->rpr_msgout.dma_map->dm_mapsize, BUS_DMASYNC_PREREAD);
2660 
2661 	/* Enqueue and we're done... */
2662 	s = splnet();
2663 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue2, &rp->rpr_q, q_next);
2664 	ubsec_feed2(sc);
2665 	ubsecstats.hst_modexpcrt++;
2666 	splx(s);
2667 	return (0);
2668 
2669 errout:
2670 	if (rp != NULL) {
2671 		if (rp->rpr_q.q_mcr.dma_map != NULL)
2672 			ubsec_dma_free(sc, &rp->rpr_q.q_mcr);
2673 		if (rp->rpr_msgin.dma_map != NULL) {
2674 			memset(rp->rpr_msgin.dma_vaddr, 0, rp->rpr_msgin.dma_size);
2675 			ubsec_dma_free(sc, &rp->rpr_msgin);
2676 		}
2677 		if (rp->rpr_msgout.dma_map != NULL) {
2678 			memset(rp->rpr_msgout.dma_vaddr, 0, rp->rpr_msgout.dma_size);
2679 			ubsec_dma_free(sc, &rp->rpr_msgout);
2680 		}
2681 		free(rp, M_DEVBUF);
2682 	}
2683 	krp->krp_status = err;
2684 	crypto_kdone(krp);
2685 	return (0);
2686 }
2687 
2688 #ifdef UBSEC_DEBUG
2689 static void
2690 ubsec_dump_pb(volatile struct ubsec_pktbuf *pb)
2691 {
2692 	printf("addr 0x%x (0x%x) next 0x%x\n",
2693 	    pb->pb_addr, pb->pb_len, pb->pb_next);
2694 }
2695 
2696 static void
2697 ubsec_dump_ctx2(volatile struct ubsec_ctx_keyop *c)
2698 {
2699 	printf("CTX (0x%x):\n", c->ctx_len);
2700 	switch (letoh16(c->ctx_op)) {
2701 	case UBS_CTXOP_RNGBYPASS:
2702 	case UBS_CTXOP_RNGSHA1:
2703 		break;
2704 	case UBS_CTXOP_MODEXP:
2705 	{
2706 		struct ubsec_ctx_modexp *cx = (void *)c;
2707 		int i, len;
2708 
2709 		printf(" Elen %u, Nlen %u\n",
2710 		    letoh16(cx->me_E_len), letoh16(cx->me_N_len));
2711 		len = (cx->me_N_len + 7)/8;
2712 		for (i = 0; i < len; i++)
2713 			printf("%s%02x", (i == 0) ? " N: " : ":", cx->me_N[i]);
2714 		printf("\n");
2715 		break;
2716 	}
2717 	default:
2718 		printf("unknown context: %x\n", c->ctx_op);
2719 	}
2720 	printf("END CTX\n");
2721 }
2722 
2723 static void
2724 ubsec_dump_mcr(struct ubsec_mcr *mcr)
2725 {
2726 	volatile struct ubsec_mcr_add *ma;
2727 	int i;
2728 
2729 	printf("MCR:\n");
2730 	printf(" pkts: %u, flags 0x%x\n",
2731 	    letoh16(mcr->mcr_pkts), letoh16(mcr->mcr_flags));
2732 	ma = (volatile struct ubsec_mcr_add *)&mcr->mcr_cmdctxp;
2733 	for (i = 0; i < letoh16(mcr->mcr_pkts); i++) {
2734 		printf(" %d: ctx 0x%x len 0x%x rsvd 0x%x\n", i,
2735 		    letoh32(ma->mcr_cmdctxp), letoh16(ma->mcr_pktlen),
2736 		    letoh16(ma->mcr_reserved));
2737 		printf(" %d: ipkt ", i);
2738 		ubsec_dump_pb(&ma->mcr_ipktbuf);
2739 		printf(" %d: opkt ", i);
2740 		ubsec_dump_pb(&ma->mcr_opktbuf);
2741 		ma++;
2742 	}
2743 	printf("END MCR\n");
2744 }
2745 #endif /* UBSEC_DEBUG */
2746 
2747 /*
2748  * Return the number of significant bits of a big number.
2749  */
2750 static int
2751 ubsec_ksigbits(struct crparam *cr)
2752 {
2753 	u_int plen = (cr->crp_nbits + 7) / 8;
2754 	int i, sig = plen * 8;
2755 	u_int8_t c, *p = cr->crp_p;
2756 
2757 	for (i = plen - 1; i >= 0; i--) {
2758 		c = p[i];
2759 		if (c != 0) {
2760 			while ((c & 0x80) == 0) {
2761 				sig--;
2762 				c <<= 1;
2763 			}
2764 			break;
2765 		}
2766 		sig -= 8;
2767 	}
2768 	return (sig);
2769 }
2770 
2771 static void
2772 ubsec_kshift_r(u_int shiftbits, u_int8_t *src, u_int srcbits,
2773     u_int8_t *dst, u_int dstbits)
2774 {
2775 	u_int slen, dlen;
2776 	int i, si, di, n;
2777 
2778 	slen = (srcbits + 7) / 8;
2779 	dlen = (dstbits + 7) / 8;
2780 
2781 	for (i = 0; i < slen; i++)
2782 		dst[i] = src[i];
2783 	for (i = 0; i < dlen - slen; i++)
2784 		dst[slen + i] = 0;
2785 
2786 	n = shiftbits / 8;
2787 	if (n != 0) {
2788 		si = dlen - n - 1;
2789 		di = dlen - 1;
2790 		while (si >= 0)
2791 			dst[di--] = dst[si--];
2792 		while (di >= 0)
2793 			dst[di--] = 0;
2794 	}
2795 
2796 	n = shiftbits % 8;
2797 	if (n != 0) {
2798 		for (i = dlen - 1; i > 0; i--)
2799 			dst[i] = (dst[i] << n) |
2800 			    (dst[i - 1] >> (8 - n));
2801 		dst[0] = dst[0] << n;
2802 	}
2803 }
2804 
2805 static void
2806 ubsec_kshift_l(u_int shiftbits, u_int8_t *src, u_int srcbits,
2807     u_int8_t *dst, u_int dstbits)
2808 {
2809 	int slen, dlen, i, n;
2810 
2811 	slen = (srcbits + 7) / 8;
2812 	dlen = (dstbits + 7) / 8;
2813 
2814 	n = shiftbits / 8;
2815 	for (i = 0; i < slen; i++)
2816 		dst[i] = src[i + n];
2817 	for (i = 0; i < dlen - slen; i++)
2818 		dst[slen + i] = 0;
2819 
2820 	n = shiftbits % 8;
2821 	if (n != 0) {
2822 		for (i = 0; i < (dlen - 1); i++)
2823 			dst[i] = (dst[i] >> n) | (dst[i + 1] << (8 - n));
2824 		dst[dlen - 1] = dst[dlen - 1] >> n;
2825 	}
2826 }
2827