xref: /dragonfly/sys/dev/crypto/ubsec/ubsec.c (revision b7367ef6)
1 /* $FreeBSD: src/sys/dev/ubsec/ubsec.c,v 1.6.2.12 2003/06/04 17:56:59 sam Exp $ */
2 /* $DragonFly: src/sys/dev/crypto/ubsec/ubsec.c,v 1.13 2006/12/22 23:26:15 swildner Exp $ */
3 /*	$OpenBSD: ubsec.c,v 1.115 2002/09/24 18:33:26 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  * All rights reserved.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by Jason L. Wright
23  * 4. The name of the author may not be used to endorse or promote products
24  *    derived from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
27  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
28  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
29  * DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT,
30  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
31  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
32  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
34  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
35  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  *
38  * Effort sponsored in part by the Defense Advanced Research Projects
39  * Agency (DARPA) and Air Force Research Laboratory, Air Force
40  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
41  *
42  */
43 
44 /*
45  * uBsec 5[56]01, 58xx hardware crypto accelerator
46  */
47 
48 #include "opt_ubsec.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/proc.h>
53 #include <sys/errno.h>
54 #include <sys/malloc.h>
55 #include <sys/kernel.h>
56 #include <sys/mbuf.h>
57 #include <sys/sysctl.h>
58 #include <sys/endian.h>
59 #include <sys/bus.h>
60 #include <sys/rman.h>
61 #include <sys/md5.h>
62 #include <sys/random.h>
63 #include <sys/thread2.h>
64 
65 #include <vm/vm.h>
66 #include <vm/pmap.h>
67 
68 #include <machine/clock.h>
69 
70 #include <crypto/sha1.h>
71 #include <opencrypto/cryptodev.h>
72 #include <opencrypto/cryptosoft.h>
73 
74 #include <bus/pci/pcivar.h>
75 #include <bus/pci/pcireg.h>
76 
77 /* grr, #defines for gratuitous incompatibility in queue.h */
78 #define	SIMPLEQ_HEAD		STAILQ_HEAD
79 #define	SIMPLEQ_ENTRY		STAILQ_ENTRY
80 #define	SIMPLEQ_INIT		STAILQ_INIT
81 #define	SIMPLEQ_INSERT_TAIL	STAILQ_INSERT_TAIL
82 #define	SIMPLEQ_EMPTY		STAILQ_EMPTY
83 #define	SIMPLEQ_FIRST		STAILQ_FIRST
84 #define	SIMPLEQ_REMOVE_HEAD	STAILQ_REMOVE_HEAD_UNTIL
85 #define	SIMPLEQ_FOREACH		STAILQ_FOREACH
86 /* ditto for endian.h */
87 #define	letoh16(x)		le16toh(x)
88 #define	letoh32(x)		le32toh(x)
89 
90 #ifdef UBSEC_RNDTEST
91 #include "../rndtest/rndtest.h"
92 #endif
93 #include "ubsecreg.h"
94 #include "ubsecvar.h"
95 
96 /*
97  * Prototypes and count for the pci_device structure
98  */
99 static	int ubsec_probe(device_t);
100 static	int ubsec_attach(device_t);
101 static	int ubsec_detach(device_t);
102 static	int ubsec_suspend(device_t);
103 static	int ubsec_resume(device_t);
104 static	void ubsec_shutdown(device_t);
105 
106 static device_method_t ubsec_methods[] = {
107 	/* Device interface */
108 	DEVMETHOD(device_probe,		ubsec_probe),
109 	DEVMETHOD(device_attach,	ubsec_attach),
110 	DEVMETHOD(device_detach,	ubsec_detach),
111 	DEVMETHOD(device_suspend,	ubsec_suspend),
112 	DEVMETHOD(device_resume,	ubsec_resume),
113 	DEVMETHOD(device_shutdown,	ubsec_shutdown),
114 
115 	/* bus interface */
116 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
117 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
118 
119 	{ 0, 0 }
120 };
121 static driver_t ubsec_driver = {
122 	"ubsec",
123 	ubsec_methods,
124 	sizeof (struct ubsec_softc)
125 };
126 static devclass_t ubsec_devclass;
127 
128 DECLARE_DUMMY_MODULE(ubsec);
129 DRIVER_MODULE(ubsec, pci, ubsec_driver, ubsec_devclass, 0, 0);
130 MODULE_DEPEND(ubsec, crypto, 1, 1, 1);
131 #ifdef UBSEC_RNDTEST
132 MODULE_DEPEND(ubsec, rndtest, 1, 1, 1);
133 #endif
134 
135 static	void ubsec_intr(void *);
136 static	int ubsec_newsession(void *, u_int32_t *, struct cryptoini *);
137 static	int ubsec_freesession(void *, u_int64_t);
138 static	int ubsec_process(void *, struct cryptop *, int);
139 static	void ubsec_callback(struct ubsec_softc *, struct ubsec_q *);
140 static	void ubsec_feed(struct ubsec_softc *);
141 static	void ubsec_mcopy(struct mbuf *, struct mbuf *, int, int);
142 static	void ubsec_callback2(struct ubsec_softc *, struct ubsec_q2 *);
143 static	int ubsec_feed2(struct ubsec_softc *);
144 static	void ubsec_rng(void *);
145 static	int ubsec_dma_malloc(struct ubsec_softc *, bus_size_t,
146 			     struct ubsec_dma_alloc *, int);
147 #define	ubsec_dma_sync(_dma, _flags) \
148 	bus_dmamap_sync((_dma)->dma_tag, (_dma)->dma_map, (_flags))
149 static	void ubsec_dma_free(struct ubsec_softc *, struct ubsec_dma_alloc *);
150 static	int ubsec_dmamap_aligned(struct ubsec_operand *op);
151 
152 static	void ubsec_reset_board(struct ubsec_softc *sc);
153 static	void ubsec_init_board(struct ubsec_softc *sc);
154 static	void ubsec_init_pciregs(device_t dev);
155 static	void ubsec_totalreset(struct ubsec_softc *sc);
156 
157 static	int ubsec_free_q(struct ubsec_softc *sc, struct ubsec_q *q);
158 
159 static	int ubsec_kprocess(void*, struct cryptkop *, int);
160 static	int ubsec_kprocess_modexp_hw(struct ubsec_softc *, struct cryptkop *, int);
161 static	int ubsec_kprocess_modexp_sw(struct ubsec_softc *, struct cryptkop *, int);
162 static	int ubsec_kprocess_rsapriv(struct ubsec_softc *, struct cryptkop *, int);
163 static	void ubsec_kfree(struct ubsec_softc *, struct ubsec_q2 *);
164 static	int ubsec_ksigbits(struct crparam *);
165 static	void ubsec_kshift_r(u_int, u_int8_t *, u_int, u_int8_t *, u_int);
166 static	void ubsec_kshift_l(u_int, u_int8_t *, u_int, u_int8_t *, u_int);
167 
168 SYSCTL_NODE(_hw, OID_AUTO, ubsec, CTLFLAG_RD, 0, "Broadcom driver parameters");
169 
170 #ifdef UBSEC_DEBUG
171 static	void ubsec_dump_pb(volatile struct ubsec_pktbuf *);
172 static	void ubsec_dump_mcr(struct ubsec_mcr *);
173 static	void ubsec_dump_ctx2(struct ubsec_ctx_keyop *);
174 
175 static	int ubsec_debug = 0;
176 SYSCTL_INT(_hw_ubsec, OID_AUTO, debug, CTLFLAG_RW, &ubsec_debug,
177 	    0, "control debugging msgs");
178 #endif
179 
180 #define	READ_REG(sc,r) \
181 	bus_space_read_4((sc)->sc_st, (sc)->sc_sh, (r))
182 
183 #define WRITE_REG(sc,reg,val) \
184 	bus_space_write_4((sc)->sc_st, (sc)->sc_sh, reg, val)
185 
186 #define	SWAP32(x) (x) = htole32(ntohl((x)))
187 #define	HTOLE32(x) (x) = htole32(x)
188 
189 
190 struct ubsec_stats ubsecstats;
191 SYSCTL_STRUCT(_hw_ubsec, OID_AUTO, stats, CTLFLAG_RD, &ubsecstats,
192 	    ubsec_stats, "driver statistics");
193 
194 static int
195 ubsec_probe(device_t dev)
196 {
197 	if (pci_get_vendor(dev) == PCI_VENDOR_SUN &&
198 	    (pci_get_device(dev) == PCI_PRODUCT_SUN_5821 ||
199 	     pci_get_device(dev) == PCI_PRODUCT_SUN_SCA1K))
200 		return (0);
201 	if (pci_get_vendor(dev) == PCI_VENDOR_BLUESTEEL &&
202 	    (pci_get_device(dev) == PCI_PRODUCT_BLUESTEEL_5501 ||
203 	     pci_get_device(dev) == PCI_PRODUCT_BLUESTEEL_5601))
204 		return (0);
205 	if (pci_get_vendor(dev) == PCI_VENDOR_BROADCOM &&
206 	    (pci_get_device(dev) == PCI_PRODUCT_BROADCOM_5801 ||
207 	     pci_get_device(dev) == PCI_PRODUCT_BROADCOM_5802 ||
208 	     pci_get_device(dev) == PCI_PRODUCT_BROADCOM_5805 ||
209 	     pci_get_device(dev) == PCI_PRODUCT_BROADCOM_5820 ||
210 	     pci_get_device(dev) == PCI_PRODUCT_BROADCOM_5821 ||
211 	     pci_get_device(dev) == PCI_PRODUCT_BROADCOM_5822 ||
212 	     pci_get_device(dev) == PCI_PRODUCT_BROADCOM_5823
213 	     ))
214 		return (0);
215 	return (ENXIO);
216 }
217 
218 static const char*
219 ubsec_partname(struct ubsec_softc *sc)
220 {
221 	/* XXX sprintf numbers when not decoded */
222 	switch (pci_get_vendor(sc->sc_dev)) {
223 	case PCI_VENDOR_BROADCOM:
224 		switch (pci_get_device(sc->sc_dev)) {
225 		case PCI_PRODUCT_BROADCOM_5801:	return "Broadcom 5801";
226 		case PCI_PRODUCT_BROADCOM_5802:	return "Broadcom 5802";
227 		case PCI_PRODUCT_BROADCOM_5805:	return "Broadcom 5805";
228 		case PCI_PRODUCT_BROADCOM_5820:	return "Broadcom 5820";
229 		case PCI_PRODUCT_BROADCOM_5821:	return "Broadcom 5821";
230 		case PCI_PRODUCT_BROADCOM_5822:	return "Broadcom 5822";
231 		case PCI_PRODUCT_BROADCOM_5823:	return "Broadcom 5823";
232 		}
233 		return "Broadcom unknown-part";
234 	case PCI_VENDOR_BLUESTEEL:
235 		switch (pci_get_device(sc->sc_dev)) {
236 		case PCI_PRODUCT_BLUESTEEL_5601: return "Bluesteel 5601";
237 		}
238 		return "Bluesteel unknown-part";
239 	case PCI_VENDOR_SUN:
240 		switch (pci_get_device(sc->sc_dev)) {
241 		case PCI_PRODUCT_SUN_5821: return "Sun Crypto 5821";
242 		case PCI_PRODUCT_SUN_SCA1K: return "Sun Crypto 1K";
243 		}
244 		return "Sun unknown-part";
245 	}
246 	return "Unknown-vendor unknown-part";
247 }
248 
249 static void
250 default_harvest(struct rndtest_state *rsp, void *buf, u_int count)
251 {
252 	u_int32_t *p = (u_int32_t *)buf;
253 	for (count /= sizeof (u_int32_t); count; count--)
254 		add_true_randomness(*p++);
255 }
256 
257 static int
258 ubsec_attach(device_t dev)
259 {
260 	struct ubsec_softc *sc = device_get_softc(dev);
261 	struct ubsec_dma *dmap;
262 	u_int32_t cmd, i;
263 	int rid;
264 
265 	KASSERT(sc != NULL, ("ubsec_attach: null software carrier!"));
266 	bzero(sc, sizeof (*sc));
267 	sc->sc_dev = dev;
268 
269 	SIMPLEQ_INIT(&sc->sc_queue);
270 	SIMPLEQ_INIT(&sc->sc_qchip);
271 	SIMPLEQ_INIT(&sc->sc_queue2);
272 	SIMPLEQ_INIT(&sc->sc_qchip2);
273 	SIMPLEQ_INIT(&sc->sc_q2free);
274 
275 	/* XXX handle power management */
276 
277 	sc->sc_statmask = BS_STAT_MCR1_DONE | BS_STAT_DMAERR;
278 
279 	if (pci_get_vendor(dev) == PCI_VENDOR_BLUESTEEL &&
280 	    pci_get_device(dev) == PCI_PRODUCT_BLUESTEEL_5601)
281 		sc->sc_flags |= UBS_FLAGS_KEY | UBS_FLAGS_RNG;
282 
283 	if (pci_get_vendor(dev) == PCI_VENDOR_BROADCOM &&
284 	    (pci_get_device(dev) == PCI_PRODUCT_BROADCOM_5802 ||
285 	     pci_get_device(dev) == PCI_PRODUCT_BROADCOM_5805))
286 		sc->sc_flags |= UBS_FLAGS_KEY | UBS_FLAGS_RNG;
287 
288 	if (pci_get_vendor(dev) == PCI_VENDOR_BROADCOM &&
289 	    pci_get_device(dev) == PCI_PRODUCT_BROADCOM_5820)
290 		sc->sc_flags |= UBS_FLAGS_KEY | UBS_FLAGS_RNG |
291 		    UBS_FLAGS_LONGCTX | UBS_FLAGS_HWNORM | UBS_FLAGS_BIGKEY;
292 
293 	if ((pci_get_vendor(dev) == PCI_VENDOR_BROADCOM &&
294 	     (pci_get_device(dev) == PCI_PRODUCT_BROADCOM_5821 ||
295 	      pci_get_device(dev) == PCI_PRODUCT_BROADCOM_5822 ||
296 	      pci_get_device(dev) == PCI_PRODUCT_BROADCOM_5823)) ||
297 	    (pci_get_vendor(dev) == PCI_VENDOR_SUN &&
298 	     (pci_get_device(dev) == PCI_PRODUCT_SUN_SCA1K ||
299 	      pci_get_device(dev) == PCI_PRODUCT_SUN_5821))) {
300 		/* NB: the 5821/5822 defines some additional status bits */
301 		sc->sc_statmask |= BS_STAT_MCR1_ALLEMPTY |
302 		    BS_STAT_MCR2_ALLEMPTY;
303 		sc->sc_flags |= UBS_FLAGS_KEY | UBS_FLAGS_RNG |
304 		    UBS_FLAGS_LONGCTX | UBS_FLAGS_HWNORM | UBS_FLAGS_BIGKEY;
305 	}
306 
307 	cmd = pci_read_config(dev, PCIR_COMMAND, 4);
308 	cmd |= PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN;
309 	pci_write_config(dev, PCIR_COMMAND, cmd, 4);
310 	cmd = pci_read_config(dev, PCIR_COMMAND, 4);
311 
312 	if (!(cmd & PCIM_CMD_MEMEN)) {
313 		device_printf(dev, "failed to enable memory mapping\n");
314 		goto bad;
315 	}
316 
317 	if (!(cmd & PCIM_CMD_BUSMASTEREN)) {
318 		device_printf(dev, "failed to enable bus mastering\n");
319 		goto bad;
320 	}
321 
322 	/*
323 	 * Setup memory-mapping of PCI registers.
324 	 */
325 	rid = BS_BAR;
326 	sc->sc_sr = bus_alloc_resource(dev, SYS_RES_MEMORY, &rid,
327 				       0, ~0, 1, RF_ACTIVE);
328 	if (sc->sc_sr == NULL) {
329 		device_printf(dev, "cannot map register space\n");
330 		goto bad;
331 	}
332 	sc->sc_st = rman_get_bustag(sc->sc_sr);
333 	sc->sc_sh = rman_get_bushandle(sc->sc_sr);
334 
335 	/*
336 	 * Arrange interrupt line.
337 	 */
338 	rid = 0;
339 	sc->sc_irq = bus_alloc_resource(dev, SYS_RES_IRQ, &rid,
340 					0, ~0, 1, RF_SHAREABLE|RF_ACTIVE);
341 	if (sc->sc_irq == NULL) {
342 		device_printf(dev, "could not map interrupt\n");
343 		goto bad1;
344 	}
345 	/*
346 	 * NB: Network code assumes we are blocked with splimp()
347 	 *     so make sure the IRQ is mapped appropriately.
348 	 */
349 	if (bus_setup_intr(dev, sc->sc_irq, 0,
350 			   ubsec_intr, sc,
351 			   &sc->sc_ih, NULL)) {
352 		device_printf(dev, "could not establish interrupt\n");
353 		goto bad2;
354 	}
355 
356 	sc->sc_cid = crypto_get_driverid(0);
357 	if (sc->sc_cid < 0) {
358 		device_printf(dev, "could not get crypto driver id\n");
359 		goto bad3;
360 	}
361 
362 	/*
363 	 * Setup DMA descriptor area.
364 	 */
365 	if (bus_dma_tag_create(NULL,			/* parent */
366 			       1, 0,			/* alignment, bounds */
367 			       BUS_SPACE_MAXADDR_32BIT,	/* lowaddr */
368 			       BUS_SPACE_MAXADDR,	/* highaddr */
369 			       NULL, NULL,		/* filter, filterarg */
370 			       0x3ffff,			/* maxsize */
371 			       UBS_MAX_SCATTER,		/* nsegments */
372 			       0xffff,			/* maxsegsize */
373 			       BUS_DMA_ALLOCNOW,	/* flags */
374 			       &sc->sc_dmat)) {
375 		device_printf(dev, "cannot allocate DMA tag\n");
376 		goto bad4;
377 	}
378 	SIMPLEQ_INIT(&sc->sc_freequeue);
379 	dmap = sc->sc_dmaa;
380 	for (i = 0; i < UBS_MAX_NQUEUE; i++, dmap++) {
381 		struct ubsec_q *q;
382 
383 		q = kmalloc(sizeof(struct ubsec_q), M_DEVBUF, M_WAITOK);
384 		if (ubsec_dma_malloc(sc, sizeof(struct ubsec_dmachunk),
385 		    &dmap->d_alloc, 0)) {
386 			device_printf(dev, "cannot allocate dma buffers\n");
387 			kfree(q, M_DEVBUF);
388 			break;
389 		}
390 		dmap->d_dma = (struct ubsec_dmachunk *)dmap->d_alloc.dma_vaddr;
391 
392 		q->q_dma = dmap;
393 		sc->sc_queuea[i] = q;
394 
395 		SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
396 	}
397 
398 	device_printf(sc->sc_dev, "%s\n", ubsec_partname(sc));
399 
400 	crypto_register(sc->sc_cid, CRYPTO_3DES_CBC, 0, 0,
401 	    ubsec_newsession, ubsec_freesession, ubsec_process, sc);
402 	crypto_register(sc->sc_cid, CRYPTO_DES_CBC, 0, 0,
403 	     ubsec_newsession, ubsec_freesession, ubsec_process, sc);
404 	crypto_register(sc->sc_cid, CRYPTO_MD5_HMAC, 0, 0,
405 	     ubsec_newsession, ubsec_freesession, ubsec_process, sc);
406 	crypto_register(sc->sc_cid, CRYPTO_SHA1_HMAC, 0, 0,
407 	     ubsec_newsession, ubsec_freesession, ubsec_process, sc);
408 
409 	/*
410 	 * Reset Broadcom chip
411 	 */
412 	ubsec_reset_board(sc);
413 
414 	/*
415 	 * Init Broadcom specific PCI settings
416 	 */
417 	ubsec_init_pciregs(dev);
418 
419 	/*
420 	 * Init Broadcom chip
421 	 */
422 	ubsec_init_board(sc);
423 
424 #ifndef UBSEC_NO_RNG
425 	if (sc->sc_flags & UBS_FLAGS_RNG) {
426 		sc->sc_statmask |= BS_STAT_MCR2_DONE;
427 #ifdef UBSEC_RNDTEST
428 		sc->sc_rndtest = rndtest_attach(dev);
429 		if (sc->sc_rndtest)
430 			sc->sc_harvest = rndtest_harvest;
431 		else
432 			sc->sc_harvest = default_harvest;
433 #else
434 		sc->sc_harvest = default_harvest;
435 #endif
436 
437 		if (ubsec_dma_malloc(sc, sizeof(struct ubsec_mcr),
438 		    &sc->sc_rng.rng_q.q_mcr, 0))
439 			goto skip_rng;
440 
441 		if (ubsec_dma_malloc(sc, sizeof(struct ubsec_ctx_rngbypass),
442 		    &sc->sc_rng.rng_q.q_ctx, 0)) {
443 			ubsec_dma_free(sc, &sc->sc_rng.rng_q.q_mcr);
444 			goto skip_rng;
445 		}
446 
447 		if (ubsec_dma_malloc(sc, sizeof(u_int32_t) *
448 		    UBSEC_RNG_BUFSIZ, &sc->sc_rng.rng_buf, 0)) {
449 			ubsec_dma_free(sc, &sc->sc_rng.rng_q.q_ctx);
450 			ubsec_dma_free(sc, &sc->sc_rng.rng_q.q_mcr);
451 			goto skip_rng;
452 		}
453 
454 		if (hz >= 100)
455 			sc->sc_rnghz = hz / 100;
456 		else
457 			sc->sc_rnghz = 1;
458 		callout_init(&sc->sc_rngto);
459 		callout_reset(&sc->sc_rngto, sc->sc_rnghz, ubsec_rng, sc);
460 skip_rng:
461 	;
462 	}
463 #endif /* UBSEC_NO_RNG */
464 
465 	if (sc->sc_flags & UBS_FLAGS_KEY) {
466 		sc->sc_statmask |= BS_STAT_MCR2_DONE;
467 
468 		crypto_kregister(sc->sc_cid, CRK_MOD_EXP, 0,
469 			ubsec_kprocess, sc);
470 #if 0
471 		crypto_kregister(sc->sc_cid, CRK_MOD_EXP_CRT, 0,
472 			ubsec_kprocess, sc);
473 #endif
474 	}
475 	return (0);
476 bad4:
477 	crypto_unregister_all(sc->sc_cid);
478 bad3:
479 	bus_teardown_intr(dev, sc->sc_irq, sc->sc_ih);
480 bad2:
481 	bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq);
482 bad1:
483 	bus_release_resource(dev, SYS_RES_MEMORY, BS_BAR, sc->sc_sr);
484 bad:
485 	return (ENXIO);
486 }
487 
488 /*
489  * Detach a device that successfully probed.
490  */
491 static int
492 ubsec_detach(device_t dev)
493 {
494 	struct ubsec_softc *sc = device_get_softc(dev);
495 
496 	KASSERT(sc != NULL, ("ubsec_detach: null software carrier"));
497 
498 	/* XXX wait/abort active ops */
499 
500 	crit_enter();
501 
502 	callout_stop(&sc->sc_rngto);
503 
504 	crypto_unregister_all(sc->sc_cid);
505 
506 #ifdef UBSEC_RNDTEST
507 	if (sc->sc_rndtest)
508 		rndtest_detach(sc->sc_rndtest);
509 #endif
510 
511 	while (!SIMPLEQ_EMPTY(&sc->sc_freequeue)) {
512 		struct ubsec_q *q;
513 
514 		q = SIMPLEQ_FIRST(&sc->sc_freequeue);
515 		SIMPLEQ_REMOVE_HEAD(&sc->sc_freequeue, q, q_next);
516 		ubsec_dma_free(sc, &q->q_dma->d_alloc);
517 		kfree(q, M_DEVBUF);
518 	}
519 #ifndef UBSEC_NO_RNG
520 	if (sc->sc_flags & UBS_FLAGS_RNG) {
521 		ubsec_dma_free(sc, &sc->sc_rng.rng_q.q_mcr);
522 		ubsec_dma_free(sc, &sc->sc_rng.rng_q.q_ctx);
523 		ubsec_dma_free(sc, &sc->sc_rng.rng_buf);
524 	}
525 #endif /* UBSEC_NO_RNG */
526 
527 	bus_generic_detach(dev);
528 	bus_teardown_intr(dev, sc->sc_irq, sc->sc_ih);
529 	bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq);
530 
531 	bus_dma_tag_destroy(sc->sc_dmat);
532 	bus_release_resource(dev, SYS_RES_MEMORY, BS_BAR, sc->sc_sr);
533 
534 	crit_exit();
535 
536 	return (0);
537 }
538 
539 /*
540  * Stop all chip i/o so that the kernel's probe routines don't
541  * get confused by errant DMAs when rebooting.
542  */
543 static void
544 ubsec_shutdown(device_t dev)
545 {
546 #ifdef notyet
547 	ubsec_stop(device_get_softc(dev));
548 #endif
549 }
550 
551 /*
552  * Device suspend routine.
553  */
554 static int
555 ubsec_suspend(device_t dev)
556 {
557 	struct ubsec_softc *sc = device_get_softc(dev);
558 
559 	KASSERT(sc != NULL, ("ubsec_suspend: null software carrier"));
560 #ifdef notyet
561 	/* XXX stop the device and save PCI settings */
562 #endif
563 	sc->sc_suspended = 1;
564 
565 	return (0);
566 }
567 
568 static int
569 ubsec_resume(device_t dev)
570 {
571 	struct ubsec_softc *sc = device_get_softc(dev);
572 
573 	KASSERT(sc != NULL, ("ubsec_resume: null software carrier"));
574 #ifdef notyet
575 	/* XXX retore PCI settings and start the device */
576 #endif
577 	sc->sc_suspended = 0;
578 	return (0);
579 }
580 
581 /*
582  * UBSEC Interrupt routine
583  */
584 static void
585 ubsec_intr(void *arg)
586 {
587 	struct ubsec_softc *sc = arg;
588 	volatile u_int32_t stat;
589 	struct ubsec_q *q;
590 	struct ubsec_dma *dmap;
591 	int npkts = 0, i;
592 
593 	stat = READ_REG(sc, BS_STAT);
594 	stat &= sc->sc_statmask;
595 	if (stat == 0) {
596 		return;
597 	}
598 
599 	WRITE_REG(sc, BS_STAT, stat);		/* IACK */
600 
601 	/*
602 	 * Check to see if we have any packets waiting for us
603 	 */
604 	if ((stat & BS_STAT_MCR1_DONE)) {
605 		while (!SIMPLEQ_EMPTY(&sc->sc_qchip)) {
606 			q = SIMPLEQ_FIRST(&sc->sc_qchip);
607 			dmap = q->q_dma;
608 
609 			if ((dmap->d_dma->d_mcr.mcr_flags & htole16(UBS_MCR_DONE)) == 0)
610 				break;
611 
612 			SIMPLEQ_REMOVE_HEAD(&sc->sc_qchip, q, q_next);
613 
614 			npkts = q->q_nstacked_mcrs;
615 			sc->sc_nqchip -= 1+npkts;
616 			/*
617 			 * search for further sc_qchip ubsec_q's that share
618 			 * the same MCR, and complete them too, they must be
619 			 * at the top.
620 			 */
621 			for (i = 0; i < npkts; i++) {
622 				if(q->q_stacked_mcr[i]) {
623 					ubsec_callback(sc, q->q_stacked_mcr[i]);
624 				} else {
625 					break;
626 				}
627 			}
628 			ubsec_callback(sc, q);
629 		}
630 
631 		/*
632 		 * Don't send any more packet to chip if there has been
633 		 * a DMAERR.
634 		 */
635 		if (!(stat & BS_STAT_DMAERR))
636 			ubsec_feed(sc);
637 	}
638 
639 	/*
640 	 * Check to see if we have any key setups/rng's waiting for us
641 	 */
642 	if ((sc->sc_flags & (UBS_FLAGS_KEY|UBS_FLAGS_RNG)) &&
643 	    (stat & BS_STAT_MCR2_DONE)) {
644 		struct ubsec_q2 *q2;
645 		struct ubsec_mcr *mcr;
646 
647 		while (!SIMPLEQ_EMPTY(&sc->sc_qchip2)) {
648 			q2 = SIMPLEQ_FIRST(&sc->sc_qchip2);
649 
650 			ubsec_dma_sync(&q2->q_mcr,
651 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
652 
653 			mcr = (struct ubsec_mcr *)q2->q_mcr.dma_vaddr;
654 			if ((mcr->mcr_flags & htole16(UBS_MCR_DONE)) == 0) {
655 				ubsec_dma_sync(&q2->q_mcr,
656 				    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
657 				break;
658 			}
659 			SIMPLEQ_REMOVE_HEAD(&sc->sc_qchip2, q2, q_next);
660 			ubsec_callback2(sc, q2);
661 			/*
662 			 * Don't send any more packet to chip if there has been
663 			 * a DMAERR.
664 			 */
665 			if (!(stat & BS_STAT_DMAERR))
666 				ubsec_feed2(sc);
667 		}
668 	}
669 
670 	/*
671 	 * Check to see if we got any DMA Error
672 	 */
673 	if (stat & BS_STAT_DMAERR) {
674 #ifdef UBSEC_DEBUG
675 		if (ubsec_debug) {
676 			volatile u_int32_t a = READ_REG(sc, BS_ERR);
677 
678 			kprintf("dmaerr %s@%08x\n",
679 			    (a & BS_ERR_READ) ? "read" : "write",
680 			    a & BS_ERR_ADDR);
681 		}
682 #endif /* UBSEC_DEBUG */
683 		ubsecstats.hst_dmaerr++;
684 		ubsec_totalreset(sc);
685 		ubsec_feed(sc);
686 	}
687 
688 	if (sc->sc_needwakeup) {		/* XXX check high watermark */
689 		int wakeup = sc->sc_needwakeup & (CRYPTO_SYMQ|CRYPTO_ASYMQ);
690 #ifdef UBSEC_DEBUG
691 		if (ubsec_debug)
692 			device_printf(sc->sc_dev, "wakeup crypto (%x)\n",
693 				sc->sc_needwakeup);
694 #endif /* UBSEC_DEBUG */
695 		sc->sc_needwakeup &= ~wakeup;
696 		crypto_unblock(sc->sc_cid, wakeup);
697 	}
698 }
699 
700 /*
701  * ubsec_feed() - aggregate and post requests to chip
702  */
703 static void
704 ubsec_feed(struct ubsec_softc *sc)
705 {
706 	struct ubsec_q *q, *q2;
707 	int npkts, i;
708 	void *v;
709 	u_int32_t stat;
710 
711 	/*
712 	 * Decide how many ops to combine in a single MCR.  We cannot
713 	 * aggregate more than UBS_MAX_AGGR because this is the number
714 	 * of slots defined in the data structure.  Note that
715 	 * aggregation only happens if ops are marked batch'able.
716 	 * Aggregating ops reduces the number of interrupts to the host
717 	 * but also (potentially) increases the latency for processing
718 	 * completed ops as we only get an interrupt when all aggregated
719 	 * ops have completed.
720 	 */
721 	if (sc->sc_nqueue == 0)
722 		return;
723 	if (sc->sc_nqueue > 1) {
724 		npkts = 0;
725 		SIMPLEQ_FOREACH(q, &sc->sc_queue, q_next) {
726 			npkts++;
727 			if ((q->q_crp->crp_flags & CRYPTO_F_BATCH) == 0)
728 				break;
729 		}
730 	} else
731 		npkts = 1;
732 	/*
733 	 * Check device status before going any further.
734 	 */
735 	if ((stat = READ_REG(sc, BS_STAT)) & (BS_STAT_MCR1_FULL | BS_STAT_DMAERR)) {
736 		if (stat & BS_STAT_DMAERR) {
737 			ubsec_totalreset(sc);
738 			ubsecstats.hst_dmaerr++;
739 		} else
740 			ubsecstats.hst_mcr1full++;
741 		return;
742 	}
743 	if (sc->sc_nqueue > ubsecstats.hst_maxqueue)
744 		ubsecstats.hst_maxqueue = sc->sc_nqueue;
745 	if (npkts > UBS_MAX_AGGR)
746 		npkts = UBS_MAX_AGGR;
747 	if (npkts < 2)				/* special case 1 op */
748 		goto feed1;
749 
750 	ubsecstats.hst_totbatch += npkts-1;
751 #ifdef UBSEC_DEBUG
752 	if (ubsec_debug)
753 		kprintf("merging %d records\n", npkts);
754 #endif /* UBSEC_DEBUG */
755 
756 	q = SIMPLEQ_FIRST(&sc->sc_queue);
757 	SIMPLEQ_REMOVE_HEAD(&sc->sc_queue, q, q_next);
758 	--sc->sc_nqueue;
759 
760 	bus_dmamap_sync(sc->sc_dmat, q->q_src_map, BUS_DMASYNC_PREWRITE);
761 	if (q->q_dst_map != NULL)
762 		bus_dmamap_sync(sc->sc_dmat, q->q_dst_map, BUS_DMASYNC_PREREAD);
763 
764 	q->q_nstacked_mcrs = npkts - 1;		/* Number of packets stacked */
765 
766 	for (i = 0; i < q->q_nstacked_mcrs; i++) {
767 		q2 = SIMPLEQ_FIRST(&sc->sc_queue);
768 		bus_dmamap_sync(sc->sc_dmat, q2->q_src_map,
769 		    BUS_DMASYNC_PREWRITE);
770 		if (q2->q_dst_map != NULL)
771 			bus_dmamap_sync(sc->sc_dmat, q2->q_dst_map,
772 			    BUS_DMASYNC_PREREAD);
773 		SIMPLEQ_REMOVE_HEAD(&sc->sc_queue, q2, q_next);
774 		--sc->sc_nqueue;
775 
776 		v = (void*)(((char *)&q2->q_dma->d_dma->d_mcr) + sizeof(struct ubsec_mcr) -
777 		    sizeof(struct ubsec_mcr_add));
778 		bcopy(v, &q->q_dma->d_dma->d_mcradd[i], sizeof(struct ubsec_mcr_add));
779 		q->q_stacked_mcr[i] = q2;
780 	}
781 	q->q_dma->d_dma->d_mcr.mcr_pkts = htole16(npkts);
782 	SIMPLEQ_INSERT_TAIL(&sc->sc_qchip, q, q_next);
783 	sc->sc_nqchip += npkts;
784 	if (sc->sc_nqchip > ubsecstats.hst_maxqchip)
785 		ubsecstats.hst_maxqchip = sc->sc_nqchip;
786 	ubsec_dma_sync(&q->q_dma->d_alloc,
787 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
788 	WRITE_REG(sc, BS_MCR1, q->q_dma->d_alloc.dma_paddr +
789 	    offsetof(struct ubsec_dmachunk, d_mcr));
790 	return;
791 
792 feed1:
793 	q = SIMPLEQ_FIRST(&sc->sc_queue);
794 
795 	bus_dmamap_sync(sc->sc_dmat, q->q_src_map, BUS_DMASYNC_PREWRITE);
796 	if (q->q_dst_map != NULL)
797 		bus_dmamap_sync(sc->sc_dmat, q->q_dst_map, BUS_DMASYNC_PREREAD);
798 	ubsec_dma_sync(&q->q_dma->d_alloc,
799 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
800 
801 	WRITE_REG(sc, BS_MCR1, q->q_dma->d_alloc.dma_paddr +
802 	    offsetof(struct ubsec_dmachunk, d_mcr));
803 #ifdef UBSEC_DEBUG
804 	if (ubsec_debug)
805 		kprintf("feed1: q->chip %p %08x stat %08x\n",
806 		      q, (u_int32_t)vtophys(&q->q_dma->d_dma->d_mcr),
807 		      stat);
808 #endif /* UBSEC_DEBUG */
809 	SIMPLEQ_REMOVE_HEAD(&sc->sc_queue, q, q_next);
810 	--sc->sc_nqueue;
811 	SIMPLEQ_INSERT_TAIL(&sc->sc_qchip, q, q_next);
812 	sc->sc_nqchip++;
813 	if (sc->sc_nqchip > ubsecstats.hst_maxqchip)
814 		ubsecstats.hst_maxqchip = sc->sc_nqchip;
815 	return;
816 }
817 
818 /*
819  * Allocate a new 'session' and return an encoded session id.  'sidp'
820  * contains our registration id, and should contain an encoded session
821  * id on successful allocation.
822  */
823 static int
824 ubsec_newsession(void *arg, u_int32_t *sidp, struct cryptoini *cri)
825 {
826 	struct cryptoini *c, *encini = NULL, *macini = NULL;
827 	struct ubsec_softc *sc = arg;
828 	struct ubsec_session *ses = NULL;
829 	MD5_CTX md5ctx;
830 	SHA1_CTX sha1ctx;
831 	int i, sesn;
832 
833 	KASSERT(sc != NULL, ("ubsec_newsession: null softc"));
834 	if (sidp == NULL || cri == NULL || sc == NULL)
835 		return (EINVAL);
836 
837 	for (c = cri; c != NULL; c = c->cri_next) {
838 		if (c->cri_alg == CRYPTO_MD5_HMAC ||
839 		    c->cri_alg == CRYPTO_SHA1_HMAC) {
840 			if (macini)
841 				return (EINVAL);
842 			macini = c;
843 		} else if (c->cri_alg == CRYPTO_DES_CBC ||
844 		    c->cri_alg == CRYPTO_3DES_CBC) {
845 			if (encini)
846 				return (EINVAL);
847 			encini = c;
848 		} else
849 			return (EINVAL);
850 	}
851 	if (encini == NULL && macini == NULL)
852 		return (EINVAL);
853 
854 	if (sc->sc_sessions == NULL) {
855 		ses = sc->sc_sessions = kmalloc(sizeof(struct ubsec_session),
856 						M_DEVBUF, M_INTWAIT);
857 		sesn = 0;
858 		sc->sc_nsessions = 1;
859 	} else {
860 		for (sesn = 0; sesn < sc->sc_nsessions; sesn++) {
861 			if (sc->sc_sessions[sesn].ses_used == 0) {
862 				ses = &sc->sc_sessions[sesn];
863 				break;
864 			}
865 		}
866 
867 		if (ses == NULL) {
868 			sesn = sc->sc_nsessions;
869 			ses = kmalloc((sesn + 1) * sizeof(struct ubsec_session),
870 					M_DEVBUF, M_INTWAIT);
871 			bcopy(sc->sc_sessions, ses, sesn *
872 			    sizeof(struct ubsec_session));
873 			bzero(sc->sc_sessions, sesn *
874 			    sizeof(struct ubsec_session));
875 			kfree(sc->sc_sessions, M_DEVBUF);
876 			sc->sc_sessions = ses;
877 			ses = &sc->sc_sessions[sesn];
878 			sc->sc_nsessions++;
879 		}
880 	}
881 
882 	bzero(ses, sizeof(struct ubsec_session));
883 	ses->ses_used = 1;
884 	if (encini) {
885 		/* get an IV, network byte order */
886 		/* XXX may read fewer than requested */
887 		read_random(ses->ses_iv, sizeof(ses->ses_iv));
888 
889 		/* Go ahead and compute key in ubsec's byte order */
890 		if (encini->cri_alg == CRYPTO_DES_CBC) {
891 			bcopy(encini->cri_key, &ses->ses_deskey[0], 8);
892 			bcopy(encini->cri_key, &ses->ses_deskey[2], 8);
893 			bcopy(encini->cri_key, &ses->ses_deskey[4], 8);
894 		} else
895 			bcopy(encini->cri_key, ses->ses_deskey, 24);
896 
897 		SWAP32(ses->ses_deskey[0]);
898 		SWAP32(ses->ses_deskey[1]);
899 		SWAP32(ses->ses_deskey[2]);
900 		SWAP32(ses->ses_deskey[3]);
901 		SWAP32(ses->ses_deskey[4]);
902 		SWAP32(ses->ses_deskey[5]);
903 	}
904 
905 	if (macini) {
906 		for (i = 0; i < macini->cri_klen / 8; i++)
907 			macini->cri_key[i] ^= HMAC_IPAD_VAL;
908 
909 		if (macini->cri_alg == CRYPTO_MD5_HMAC) {
910 			MD5Init(&md5ctx);
911 			MD5Update(&md5ctx, macini->cri_key,
912 			    macini->cri_klen / 8);
913 			MD5Update(&md5ctx, hmac_ipad_buffer,
914 			    HMAC_BLOCK_LEN - (macini->cri_klen / 8));
915 			bcopy(md5ctx.state, ses->ses_hminner,
916 			    sizeof(md5ctx.state));
917 		} else {
918 			SHA1Init(&sha1ctx);
919 			SHA1Update(&sha1ctx, macini->cri_key,
920 			    macini->cri_klen / 8);
921 			SHA1Update(&sha1ctx, hmac_ipad_buffer,
922 			    HMAC_BLOCK_LEN - (macini->cri_klen / 8));
923 			bcopy(sha1ctx.h.b32, ses->ses_hminner,
924 			    sizeof(sha1ctx.h.b32));
925 		}
926 
927 		for (i = 0; i < macini->cri_klen / 8; i++)
928 			macini->cri_key[i] ^= (HMAC_IPAD_VAL ^ HMAC_OPAD_VAL);
929 
930 		if (macini->cri_alg == CRYPTO_MD5_HMAC) {
931 			MD5Init(&md5ctx);
932 			MD5Update(&md5ctx, macini->cri_key,
933 			    macini->cri_klen / 8);
934 			MD5Update(&md5ctx, hmac_opad_buffer,
935 			    HMAC_BLOCK_LEN - (macini->cri_klen / 8));
936 			bcopy(md5ctx.state, ses->ses_hmouter,
937 			    sizeof(md5ctx.state));
938 		} else {
939 			SHA1Init(&sha1ctx);
940 			SHA1Update(&sha1ctx, macini->cri_key,
941 			    macini->cri_klen / 8);
942 			SHA1Update(&sha1ctx, hmac_opad_buffer,
943 			    HMAC_BLOCK_LEN - (macini->cri_klen / 8));
944 			bcopy(sha1ctx.h.b32, ses->ses_hmouter,
945 			    sizeof(sha1ctx.h.b32));
946 		}
947 
948 		for (i = 0; i < macini->cri_klen / 8; i++)
949 			macini->cri_key[i] ^= HMAC_OPAD_VAL;
950 	}
951 
952 	*sidp = UBSEC_SID(device_get_unit(sc->sc_dev), sesn);
953 	return (0);
954 }
955 
956 /*
957  * Deallocate a session.
958  */
959 static int
960 ubsec_freesession(void *arg, u_int64_t tid)
961 {
962 	struct ubsec_softc *sc = arg;
963 	int session;
964 	u_int32_t sid = ((u_int32_t) tid) & 0xffffffff;
965 
966 	KASSERT(sc != NULL, ("ubsec_freesession: null softc"));
967 	if (sc == NULL)
968 		return (EINVAL);
969 
970 	session = UBSEC_SESSION(sid);
971 	if (session >= sc->sc_nsessions)
972 		return (EINVAL);
973 
974 	bzero(&sc->sc_sessions[session], sizeof(sc->sc_sessions[session]));
975 	return (0);
976 }
977 
978 static void
979 ubsec_op_cb(void *arg, bus_dma_segment_t *seg, int nsegs, bus_size_t mapsize, int error)
980 {
981 	struct ubsec_operand *op = arg;
982 
983 	KASSERT(nsegs <= UBS_MAX_SCATTER,
984 		("Too many DMA segments returned when mapping operand"));
985 #ifdef UBSEC_DEBUG
986 	if (ubsec_debug)
987 		kprintf("ubsec_op_cb: mapsize %u nsegs %d\n",
988 			(u_int) mapsize, nsegs);
989 #endif
990 	op->mapsize = mapsize;
991 	op->nsegs = nsegs;
992 	bcopy(seg, op->segs, nsegs * sizeof (seg[0]));
993 }
994 
995 static int
996 ubsec_process(void *arg, struct cryptop *crp, int hint)
997 {
998 	struct ubsec_q *q = NULL;
999 	int err = 0, i, j, nicealign;
1000 	struct ubsec_softc *sc = arg;
1001 	struct cryptodesc *crd1, *crd2, *maccrd, *enccrd;
1002 	int encoffset = 0, macoffset = 0, cpskip, cpoffset;
1003 	int sskip, dskip, stheend, dtheend;
1004 	int16_t coffset;
1005 	struct ubsec_session *ses;
1006 	struct ubsec_pktctx ctx;
1007 	struct ubsec_dma *dmap = NULL;
1008 
1009 	if (crp == NULL || crp->crp_callback == NULL || sc == NULL) {
1010 		ubsecstats.hst_invalid++;
1011 		return (EINVAL);
1012 	}
1013 	if (UBSEC_SESSION(crp->crp_sid) >= sc->sc_nsessions) {
1014 		ubsecstats.hst_badsession++;
1015 		return (EINVAL);
1016 	}
1017 
1018 	crit_enter();
1019 
1020 	if (SIMPLEQ_EMPTY(&sc->sc_freequeue)) {
1021 		ubsecstats.hst_queuefull++;
1022 		sc->sc_needwakeup |= CRYPTO_SYMQ;
1023 		crit_exit();
1024 		return (ERESTART);
1025 	}
1026 	q = SIMPLEQ_FIRST(&sc->sc_freequeue);
1027 	SIMPLEQ_REMOVE_HEAD(&sc->sc_freequeue, q, q_next);
1028 	crit_exit();
1029 
1030 	dmap = q->q_dma; /* Save dma pointer */
1031 	bzero(q, sizeof(struct ubsec_q));
1032 	bzero(&ctx, sizeof(ctx));
1033 
1034 	q->q_sesn = UBSEC_SESSION(crp->crp_sid);
1035 	q->q_dma = dmap;
1036 	ses = &sc->sc_sessions[q->q_sesn];
1037 
1038 	if (crp->crp_flags & CRYPTO_F_IMBUF) {
1039 		q->q_src_m = (struct mbuf *)crp->crp_buf;
1040 		q->q_dst_m = (struct mbuf *)crp->crp_buf;
1041 	} else if (crp->crp_flags & CRYPTO_F_IOV) {
1042 		q->q_src_io = (struct uio *)crp->crp_buf;
1043 		q->q_dst_io = (struct uio *)crp->crp_buf;
1044 	} else {
1045 		ubsecstats.hst_badflags++;
1046 		err = EINVAL;
1047 		goto errout;	/* XXX we don't handle contiguous blocks! */
1048 	}
1049 
1050 	bzero(&dmap->d_dma->d_mcr, sizeof(struct ubsec_mcr));
1051 
1052 	dmap->d_dma->d_mcr.mcr_pkts = htole16(1);
1053 	dmap->d_dma->d_mcr.mcr_flags = 0;
1054 	q->q_crp = crp;
1055 
1056 	crd1 = crp->crp_desc;
1057 	if (crd1 == NULL) {
1058 		ubsecstats.hst_nodesc++;
1059 		err = EINVAL;
1060 		goto errout;
1061 	}
1062 	crd2 = crd1->crd_next;
1063 
1064 	if (crd2 == NULL) {
1065 		if (crd1->crd_alg == CRYPTO_MD5_HMAC ||
1066 		    crd1->crd_alg == CRYPTO_SHA1_HMAC) {
1067 			maccrd = crd1;
1068 			enccrd = NULL;
1069 		} else if (crd1->crd_alg == CRYPTO_DES_CBC ||
1070 		    crd1->crd_alg == CRYPTO_3DES_CBC) {
1071 			maccrd = NULL;
1072 			enccrd = crd1;
1073 		} else {
1074 			ubsecstats.hst_badalg++;
1075 			err = EINVAL;
1076 			goto errout;
1077 		}
1078 	} else {
1079 		if ((crd1->crd_alg == CRYPTO_MD5_HMAC ||
1080 		    crd1->crd_alg == CRYPTO_SHA1_HMAC) &&
1081 		    (crd2->crd_alg == CRYPTO_DES_CBC ||
1082 			crd2->crd_alg == CRYPTO_3DES_CBC) &&
1083 		    ((crd2->crd_flags & CRD_F_ENCRYPT) == 0)) {
1084 			maccrd = crd1;
1085 			enccrd = crd2;
1086 		} else if ((crd1->crd_alg == CRYPTO_DES_CBC ||
1087 		    crd1->crd_alg == CRYPTO_3DES_CBC) &&
1088 		    (crd2->crd_alg == CRYPTO_MD5_HMAC ||
1089 			crd2->crd_alg == CRYPTO_SHA1_HMAC) &&
1090 		    (crd1->crd_flags & CRD_F_ENCRYPT)) {
1091 			enccrd = crd1;
1092 			maccrd = crd2;
1093 		} else {
1094 			/*
1095 			 * We cannot order the ubsec as requested
1096 			 */
1097 			ubsecstats.hst_badalg++;
1098 			err = EINVAL;
1099 			goto errout;
1100 		}
1101 	}
1102 
1103 	if (enccrd) {
1104 		encoffset = enccrd->crd_skip;
1105 		ctx.pc_flags |= htole16(UBS_PKTCTX_ENC_3DES);
1106 
1107 		if (enccrd->crd_flags & CRD_F_ENCRYPT) {
1108 			q->q_flags |= UBSEC_QFLAGS_COPYOUTIV;
1109 
1110 			if (enccrd->crd_flags & CRD_F_IV_EXPLICIT)
1111 				bcopy(enccrd->crd_iv, ctx.pc_iv, 8);
1112 			else {
1113 				ctx.pc_iv[0] = ses->ses_iv[0];
1114 				ctx.pc_iv[1] = ses->ses_iv[1];
1115 			}
1116 
1117 			if ((enccrd->crd_flags & CRD_F_IV_PRESENT) == 0) {
1118 				if (crp->crp_flags & CRYPTO_F_IMBUF)
1119 					m_copyback(q->q_src_m,
1120 					    enccrd->crd_inject,
1121 					    8, (caddr_t)ctx.pc_iv);
1122 				else if (crp->crp_flags & CRYPTO_F_IOV)
1123 					cuio_copyback(q->q_src_io,
1124 					    enccrd->crd_inject,
1125 					    8, (caddr_t)ctx.pc_iv);
1126 			}
1127 		} else {
1128 			ctx.pc_flags |= htole16(UBS_PKTCTX_INBOUND);
1129 
1130 			if (enccrd->crd_flags & CRD_F_IV_EXPLICIT)
1131 				bcopy(enccrd->crd_iv, ctx.pc_iv, 8);
1132 			else if (crp->crp_flags & CRYPTO_F_IMBUF)
1133 				m_copydata(q->q_src_m, enccrd->crd_inject,
1134 				    8, (caddr_t)ctx.pc_iv);
1135 			else if (crp->crp_flags & CRYPTO_F_IOV)
1136 				cuio_copydata(q->q_src_io,
1137 				    enccrd->crd_inject, 8,
1138 				    (caddr_t)ctx.pc_iv);
1139 		}
1140 
1141 		ctx.pc_deskey[0] = ses->ses_deskey[0];
1142 		ctx.pc_deskey[1] = ses->ses_deskey[1];
1143 		ctx.pc_deskey[2] = ses->ses_deskey[2];
1144 		ctx.pc_deskey[3] = ses->ses_deskey[3];
1145 		ctx.pc_deskey[4] = ses->ses_deskey[4];
1146 		ctx.pc_deskey[5] = ses->ses_deskey[5];
1147 		SWAP32(ctx.pc_iv[0]);
1148 		SWAP32(ctx.pc_iv[1]);
1149 	}
1150 
1151 	if (maccrd) {
1152 		macoffset = maccrd->crd_skip;
1153 
1154 		if (maccrd->crd_alg == CRYPTO_MD5_HMAC)
1155 			ctx.pc_flags |= htole16(UBS_PKTCTX_AUTH_MD5);
1156 		else
1157 			ctx.pc_flags |= htole16(UBS_PKTCTX_AUTH_SHA1);
1158 
1159 		for (i = 0; i < 5; i++) {
1160 			ctx.pc_hminner[i] = ses->ses_hminner[i];
1161 			ctx.pc_hmouter[i] = ses->ses_hmouter[i];
1162 
1163 			HTOLE32(ctx.pc_hminner[i]);
1164 			HTOLE32(ctx.pc_hmouter[i]);
1165 		}
1166 	}
1167 
1168 	if (enccrd && maccrd) {
1169 		/*
1170 		 * ubsec cannot handle packets where the end of encryption
1171 		 * and authentication are not the same, or where the
1172 		 * encrypted part begins before the authenticated part.
1173 		 */
1174 		if ((encoffset + enccrd->crd_len) !=
1175 		    (macoffset + maccrd->crd_len)) {
1176 			ubsecstats.hst_lenmismatch++;
1177 			err = EINVAL;
1178 			goto errout;
1179 		}
1180 		if (enccrd->crd_skip < maccrd->crd_skip) {
1181 			ubsecstats.hst_skipmismatch++;
1182 			err = EINVAL;
1183 			goto errout;
1184 		}
1185 		sskip = maccrd->crd_skip;
1186 		cpskip = dskip = enccrd->crd_skip;
1187 		stheend = maccrd->crd_len;
1188 		dtheend = enccrd->crd_len;
1189 		coffset = enccrd->crd_skip - maccrd->crd_skip;
1190 		cpoffset = cpskip + dtheend;
1191 #ifdef UBSEC_DEBUG
1192 		if (ubsec_debug) {
1193 			kprintf("mac: skip %d, len %d, inject %d\n",
1194 			    maccrd->crd_skip, maccrd->crd_len, maccrd->crd_inject);
1195 			kprintf("enc: skip %d, len %d, inject %d\n",
1196 			    enccrd->crd_skip, enccrd->crd_len, enccrd->crd_inject);
1197 			kprintf("src: skip %d, len %d\n", sskip, stheend);
1198 			kprintf("dst: skip %d, len %d\n", dskip, dtheend);
1199 			kprintf("ubs: coffset %d, pktlen %d, cpskip %d, cpoffset %d\n",
1200 			    coffset, stheend, cpskip, cpoffset);
1201 		}
1202 #endif
1203 	} else {
1204 		cpskip = dskip = sskip = macoffset + encoffset;
1205 		dtheend = stheend = (enccrd)?enccrd->crd_len:maccrd->crd_len;
1206 		cpoffset = cpskip + dtheend;
1207 		coffset = 0;
1208 	}
1209 	ctx.pc_offset = htole16(coffset >> 2);
1210 
1211 	if (bus_dmamap_create(sc->sc_dmat, BUS_DMA_NOWAIT, &q->q_src_map)) {
1212 		ubsecstats.hst_nomap++;
1213 		err = ENOMEM;
1214 		goto errout;
1215 	}
1216 	if (crp->crp_flags & CRYPTO_F_IMBUF) {
1217 		if (bus_dmamap_load_mbuf(sc->sc_dmat, q->q_src_map,
1218 		    q->q_src_m, ubsec_op_cb, &q->q_src, BUS_DMA_NOWAIT) != 0) {
1219 			bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
1220 			q->q_src_map = NULL;
1221 			ubsecstats.hst_noload++;
1222 			err = ENOMEM;
1223 			goto errout;
1224 		}
1225 	} else if (crp->crp_flags & CRYPTO_F_IOV) {
1226 		if (bus_dmamap_load_uio(sc->sc_dmat, q->q_src_map,
1227 		    q->q_src_io, ubsec_op_cb, &q->q_src, BUS_DMA_NOWAIT) != 0) {
1228 			bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
1229 			q->q_src_map = NULL;
1230 			ubsecstats.hst_noload++;
1231 			err = ENOMEM;
1232 			goto errout;
1233 		}
1234 	}
1235 	nicealign = ubsec_dmamap_aligned(&q->q_src);
1236 
1237 	dmap->d_dma->d_mcr.mcr_pktlen = htole16(stheend);
1238 
1239 #ifdef UBSEC_DEBUG
1240 	if (ubsec_debug)
1241 		kprintf("src skip: %d nicealign: %u\n", sskip, nicealign);
1242 #endif
1243 	for (i = j = 0; i < q->q_src_nsegs; i++) {
1244 		struct ubsec_pktbuf *pb;
1245 		bus_size_t packl = q->q_src_segs[i].ds_len;
1246 		bus_addr_t packp = q->q_src_segs[i].ds_addr;
1247 
1248 		if (sskip >= packl) {
1249 			sskip -= packl;
1250 			continue;
1251 		}
1252 
1253 		packl -= sskip;
1254 		packp += sskip;
1255 		sskip = 0;
1256 
1257 		if (packl > 0xfffc) {
1258 			err = EIO;
1259 			goto errout;
1260 		}
1261 
1262 		if (j == 0)
1263 			pb = &dmap->d_dma->d_mcr.mcr_ipktbuf;
1264 		else
1265 			pb = &dmap->d_dma->d_sbuf[j - 1];
1266 
1267 		pb->pb_addr = htole32(packp);
1268 
1269 		if (stheend) {
1270 			if (packl > stheend) {
1271 				pb->pb_len = htole32(stheend);
1272 				stheend = 0;
1273 			} else {
1274 				pb->pb_len = htole32(packl);
1275 				stheend -= packl;
1276 			}
1277 		} else
1278 			pb->pb_len = htole32(packl);
1279 
1280 		if ((i + 1) == q->q_src_nsegs)
1281 			pb->pb_next = 0;
1282 		else
1283 			pb->pb_next = htole32(dmap->d_alloc.dma_paddr +
1284 			    offsetof(struct ubsec_dmachunk, d_sbuf[j]));
1285 		j++;
1286 	}
1287 
1288 	if (enccrd == NULL && maccrd != NULL) {
1289 		dmap->d_dma->d_mcr.mcr_opktbuf.pb_addr = 0;
1290 		dmap->d_dma->d_mcr.mcr_opktbuf.pb_len = 0;
1291 		dmap->d_dma->d_mcr.mcr_opktbuf.pb_next = htole32(dmap->d_alloc.dma_paddr +
1292 		    offsetof(struct ubsec_dmachunk, d_macbuf[0]));
1293 #ifdef UBSEC_DEBUG
1294 		if (ubsec_debug)
1295 			kprintf("opkt: %x %x %x\n",
1296 			    dmap->d_dma->d_mcr.mcr_opktbuf.pb_addr,
1297 			    dmap->d_dma->d_mcr.mcr_opktbuf.pb_len,
1298 			    dmap->d_dma->d_mcr.mcr_opktbuf.pb_next);
1299 #endif
1300 	} else {
1301 		if (crp->crp_flags & CRYPTO_F_IOV) {
1302 			if (!nicealign) {
1303 				ubsecstats.hst_iovmisaligned++;
1304 				err = EINVAL;
1305 				goto errout;
1306 			}
1307 			if (bus_dmamap_create(sc->sc_dmat, BUS_DMA_NOWAIT,
1308 			     &q->q_dst_map)) {
1309 				ubsecstats.hst_nomap++;
1310 				err = ENOMEM;
1311 				goto errout;
1312 			}
1313 			if (bus_dmamap_load_uio(sc->sc_dmat, q->q_dst_map,
1314 			    q->q_dst_io, ubsec_op_cb, &q->q_dst, BUS_DMA_NOWAIT) != 0) {
1315 				bus_dmamap_destroy(sc->sc_dmat, q->q_dst_map);
1316 				q->q_dst_map = NULL;
1317 				ubsecstats.hst_noload++;
1318 				err = ENOMEM;
1319 				goto errout;
1320 			}
1321 		} else if (crp->crp_flags & CRYPTO_F_IMBUF) {
1322 			if (nicealign) {
1323 				q->q_dst = q->q_src;
1324 			} else {
1325 				int totlen, len;
1326 				struct mbuf *m, *top, **mp;
1327 
1328 				ubsecstats.hst_unaligned++;
1329 				totlen = q->q_src_mapsize;
1330 				if (q->q_src_m->m_flags & M_PKTHDR) {
1331 					len = MHLEN;
1332 					MGETHDR(m, MB_DONTWAIT, MT_DATA);
1333 					if (m && !m_dup_pkthdr(m, q->q_src_m, MB_DONTWAIT)) {
1334 						m_free(m);
1335 						m = NULL;
1336 					}
1337 				} else {
1338 					len = MLEN;
1339 					MGET(m, MB_DONTWAIT, MT_DATA);
1340 				}
1341 				if (m == NULL) {
1342 					ubsecstats.hst_nombuf++;
1343 					err = sc->sc_nqueue ? ERESTART : ENOMEM;
1344 					goto errout;
1345 				}
1346 				if (totlen >= MINCLSIZE) {
1347 					MCLGET(m, MB_DONTWAIT);
1348 					if ((m->m_flags & M_EXT) == 0) {
1349 						m_free(m);
1350 						ubsecstats.hst_nomcl++;
1351 						err = sc->sc_nqueue ? ERESTART : ENOMEM;
1352 						goto errout;
1353 					}
1354 					len = MCLBYTES;
1355 				}
1356 				m->m_len = len;
1357 				top = NULL;
1358 				mp = &top;
1359 
1360 				while (totlen > 0) {
1361 					if (top) {
1362 						MGET(m, MB_DONTWAIT, MT_DATA);
1363 						if (m == NULL) {
1364 							m_freem(top);
1365 							ubsecstats.hst_nombuf++;
1366 							err = sc->sc_nqueue ? ERESTART : ENOMEM;
1367 							goto errout;
1368 						}
1369 						len = MLEN;
1370 					}
1371 					if (top && totlen >= MINCLSIZE) {
1372 						MCLGET(m, MB_DONTWAIT);
1373 						if ((m->m_flags & M_EXT) == 0) {
1374 							*mp = m;
1375 							m_freem(top);
1376 							ubsecstats.hst_nomcl++;
1377 							err = sc->sc_nqueue ? ERESTART : ENOMEM;
1378 							goto errout;
1379 						}
1380 						len = MCLBYTES;
1381 					}
1382 					m->m_len = len = min(totlen, len);
1383 					totlen -= len;
1384 					*mp = m;
1385 					mp = &m->m_next;
1386 				}
1387 				q->q_dst_m = top;
1388 				ubsec_mcopy(q->q_src_m, q->q_dst_m,
1389 				    cpskip, cpoffset);
1390 				if (bus_dmamap_create(sc->sc_dmat,
1391 				    BUS_DMA_NOWAIT, &q->q_dst_map) != 0) {
1392 					ubsecstats.hst_nomap++;
1393 					err = ENOMEM;
1394 					goto errout;
1395 				}
1396 				if (bus_dmamap_load_mbuf(sc->sc_dmat,
1397 				    q->q_dst_map, q->q_dst_m,
1398 				    ubsec_op_cb, &q->q_dst,
1399 				    BUS_DMA_NOWAIT) != 0) {
1400 					bus_dmamap_destroy(sc->sc_dmat,
1401 					q->q_dst_map);
1402 					q->q_dst_map = NULL;
1403 					ubsecstats.hst_noload++;
1404 					err = ENOMEM;
1405 					goto errout;
1406 				}
1407 			}
1408 		} else {
1409 			ubsecstats.hst_badflags++;
1410 			err = EINVAL;
1411 			goto errout;
1412 		}
1413 
1414 #ifdef UBSEC_DEBUG
1415 		if (ubsec_debug)
1416 			kprintf("dst skip: %d\n", dskip);
1417 #endif
1418 		for (i = j = 0; i < q->q_dst_nsegs; i++) {
1419 			struct ubsec_pktbuf *pb;
1420 			bus_size_t packl = q->q_dst_segs[i].ds_len;
1421 			bus_addr_t packp = q->q_dst_segs[i].ds_addr;
1422 
1423 			if (dskip >= packl) {
1424 				dskip -= packl;
1425 				continue;
1426 			}
1427 
1428 			packl -= dskip;
1429 			packp += dskip;
1430 			dskip = 0;
1431 
1432 			if (packl > 0xfffc) {
1433 				err = EIO;
1434 				goto errout;
1435 			}
1436 
1437 			if (j == 0)
1438 				pb = &dmap->d_dma->d_mcr.mcr_opktbuf;
1439 			else
1440 				pb = &dmap->d_dma->d_dbuf[j - 1];
1441 
1442 			pb->pb_addr = htole32(packp);
1443 
1444 			if (dtheend) {
1445 				if (packl > dtheend) {
1446 					pb->pb_len = htole32(dtheend);
1447 					dtheend = 0;
1448 				} else {
1449 					pb->pb_len = htole32(packl);
1450 					dtheend -= packl;
1451 				}
1452 			} else
1453 				pb->pb_len = htole32(packl);
1454 
1455 			if ((i + 1) == q->q_dst_nsegs) {
1456 				if (maccrd)
1457 					pb->pb_next = htole32(dmap->d_alloc.dma_paddr +
1458 					    offsetof(struct ubsec_dmachunk, d_macbuf[0]));
1459 				else
1460 					pb->pb_next = 0;
1461 			} else
1462 				pb->pb_next = htole32(dmap->d_alloc.dma_paddr +
1463 				    offsetof(struct ubsec_dmachunk, d_dbuf[j]));
1464 			j++;
1465 		}
1466 	}
1467 
1468 	dmap->d_dma->d_mcr.mcr_cmdctxp = htole32(dmap->d_alloc.dma_paddr +
1469 	    offsetof(struct ubsec_dmachunk, d_ctx));
1470 
1471 	if (sc->sc_flags & UBS_FLAGS_LONGCTX) {
1472 		struct ubsec_pktctx_long *ctxl;
1473 
1474 		ctxl = (struct ubsec_pktctx_long *)(dmap->d_alloc.dma_vaddr +
1475 		    offsetof(struct ubsec_dmachunk, d_ctx));
1476 
1477 		/* transform small context into long context */
1478 		ctxl->pc_len = htole16(sizeof(struct ubsec_pktctx_long));
1479 		ctxl->pc_type = htole16(UBS_PKTCTX_TYPE_IPSEC);
1480 		ctxl->pc_flags = ctx.pc_flags;
1481 		ctxl->pc_offset = ctx.pc_offset;
1482 		for (i = 0; i < 6; i++)
1483 			ctxl->pc_deskey[i] = ctx.pc_deskey[i];
1484 		for (i = 0; i < 5; i++)
1485 			ctxl->pc_hminner[i] = ctx.pc_hminner[i];
1486 		for (i = 0; i < 5; i++)
1487 			ctxl->pc_hmouter[i] = ctx.pc_hmouter[i];
1488 		ctxl->pc_iv[0] = ctx.pc_iv[0];
1489 		ctxl->pc_iv[1] = ctx.pc_iv[1];
1490 	} else
1491 		bcopy(&ctx, dmap->d_alloc.dma_vaddr +
1492 		    offsetof(struct ubsec_dmachunk, d_ctx),
1493 		    sizeof(struct ubsec_pktctx));
1494 
1495 	crit_enter();
1496 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue, q, q_next);
1497 	sc->sc_nqueue++;
1498 	ubsecstats.hst_ipackets++;
1499 	ubsecstats.hst_ibytes += dmap->d_alloc.dma_size;
1500 	if ((hint & CRYPTO_HINT_MORE) == 0 || sc->sc_nqueue >= UBS_MAX_AGGR)
1501 		ubsec_feed(sc);
1502 	crit_exit();
1503 	return (0);
1504 
1505 errout:
1506 	if (q != NULL) {
1507 		if ((q->q_dst_m != NULL) && (q->q_src_m != q->q_dst_m))
1508 			m_freem(q->q_dst_m);
1509 
1510 		if (q->q_dst_map != NULL && q->q_dst_map != q->q_src_map) {
1511 			bus_dmamap_unload(sc->sc_dmat, q->q_dst_map);
1512 			bus_dmamap_destroy(sc->sc_dmat, q->q_dst_map);
1513 		}
1514 		if (q->q_src_map != NULL) {
1515 			bus_dmamap_unload(sc->sc_dmat, q->q_src_map);
1516 			bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
1517 		}
1518 
1519 		crit_enter();
1520 		SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
1521 		crit_exit();
1522 	}
1523 	if (err != ERESTART) {
1524 		crp->crp_etype = err;
1525 		crypto_done(crp);
1526 	} else {
1527 		sc->sc_needwakeup |= CRYPTO_SYMQ;
1528 	}
1529 	return (err);
1530 }
1531 
1532 static void
1533 ubsec_callback(struct ubsec_softc *sc, struct ubsec_q *q)
1534 {
1535 	struct cryptop *crp = (struct cryptop *)q->q_crp;
1536 	struct cryptodesc *crd;
1537 	struct ubsec_dma *dmap = q->q_dma;
1538 
1539 	ubsecstats.hst_opackets++;
1540 	ubsecstats.hst_obytes += dmap->d_alloc.dma_size;
1541 
1542 	ubsec_dma_sync(&dmap->d_alloc,
1543 	    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1544 	if (q->q_dst_map != NULL && q->q_dst_map != q->q_src_map) {
1545 		bus_dmamap_sync(sc->sc_dmat, q->q_dst_map,
1546 		    BUS_DMASYNC_POSTREAD);
1547 		bus_dmamap_unload(sc->sc_dmat, q->q_dst_map);
1548 		bus_dmamap_destroy(sc->sc_dmat, q->q_dst_map);
1549 	}
1550 	bus_dmamap_sync(sc->sc_dmat, q->q_src_map, BUS_DMASYNC_POSTWRITE);
1551 	bus_dmamap_unload(sc->sc_dmat, q->q_src_map);
1552 	bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
1553 
1554 	if ((crp->crp_flags & CRYPTO_F_IMBUF) && (q->q_src_m != q->q_dst_m)) {
1555 		m_freem(q->q_src_m);
1556 		crp->crp_buf = (caddr_t)q->q_dst_m;
1557 	}
1558 	ubsecstats.hst_obytes += ((struct mbuf *)crp->crp_buf)->m_len;
1559 
1560 	/* copy out IV for future use */
1561 	if (q->q_flags & UBSEC_QFLAGS_COPYOUTIV) {
1562 		for (crd = crp->crp_desc; crd; crd = crd->crd_next) {
1563 			if (crd->crd_alg != CRYPTO_DES_CBC &&
1564 			    crd->crd_alg != CRYPTO_3DES_CBC)
1565 				continue;
1566 			if (crp->crp_flags & CRYPTO_F_IMBUF)
1567 				m_copydata((struct mbuf *)crp->crp_buf,
1568 				    crd->crd_skip + crd->crd_len - 8, 8,
1569 				    (caddr_t)sc->sc_sessions[q->q_sesn].ses_iv);
1570 			else if (crp->crp_flags & CRYPTO_F_IOV) {
1571 				cuio_copydata((struct uio *)crp->crp_buf,
1572 				    crd->crd_skip + crd->crd_len - 8, 8,
1573 				    (caddr_t)sc->sc_sessions[q->q_sesn].ses_iv);
1574 			}
1575 			break;
1576 		}
1577 	}
1578 
1579 	for (crd = crp->crp_desc; crd; crd = crd->crd_next) {
1580 		if (crd->crd_alg != CRYPTO_MD5_HMAC &&
1581 		    crd->crd_alg != CRYPTO_SHA1_HMAC)
1582 			continue;
1583 		if (crp->crp_flags & CRYPTO_F_IMBUF)
1584 			m_copyback((struct mbuf *)crp->crp_buf,
1585 			    crd->crd_inject, 12,
1586 			    (caddr_t)dmap->d_dma->d_macbuf);
1587 		else if (crp->crp_flags & CRYPTO_F_IOV && crp->crp_mac)
1588 			bcopy((caddr_t)dmap->d_dma->d_macbuf,
1589 			    crp->crp_mac, 12);
1590 		break;
1591 	}
1592 	SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
1593 	crypto_done(crp);
1594 }
1595 
1596 static void
1597 ubsec_mcopy(struct mbuf *srcm, struct mbuf *dstm, int hoffset, int toffset)
1598 {
1599 	int i, j, dlen, slen;
1600 	caddr_t dptr, sptr;
1601 
1602 	j = 0;
1603 	sptr = srcm->m_data;
1604 	slen = srcm->m_len;
1605 	dptr = dstm->m_data;
1606 	dlen = dstm->m_len;
1607 
1608 	while (1) {
1609 		for (i = 0; i < min(slen, dlen); i++) {
1610 			if (j < hoffset || j >= toffset)
1611 				*dptr++ = *sptr++;
1612 			slen--;
1613 			dlen--;
1614 			j++;
1615 		}
1616 		if (slen == 0) {
1617 			srcm = srcm->m_next;
1618 			if (srcm == NULL)
1619 				return;
1620 			sptr = srcm->m_data;
1621 			slen = srcm->m_len;
1622 		}
1623 		if (dlen == 0) {
1624 			dstm = dstm->m_next;
1625 			if (dstm == NULL)
1626 				return;
1627 			dptr = dstm->m_data;
1628 			dlen = dstm->m_len;
1629 		}
1630 	}
1631 }
1632 
1633 /*
1634  * feed the key generator, must be called at splimp() or higher.
1635  */
1636 static int
1637 ubsec_feed2(struct ubsec_softc *sc)
1638 {
1639 	struct ubsec_q2 *q;
1640 
1641 	while (!SIMPLEQ_EMPTY(&sc->sc_queue2)) {
1642 		if (READ_REG(sc, BS_STAT) & BS_STAT_MCR2_FULL)
1643 			break;
1644 		q = SIMPLEQ_FIRST(&sc->sc_queue2);
1645 
1646 		ubsec_dma_sync(&q->q_mcr,
1647 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1648 		ubsec_dma_sync(&q->q_ctx, BUS_DMASYNC_PREWRITE);
1649 
1650 		WRITE_REG(sc, BS_MCR2, q->q_mcr.dma_paddr);
1651 		SIMPLEQ_REMOVE_HEAD(&sc->sc_queue2, q, q_next);
1652 		--sc->sc_nqueue2;
1653 		SIMPLEQ_INSERT_TAIL(&sc->sc_qchip2, q, q_next);
1654 	}
1655 	return (0);
1656 }
1657 
1658 /*
1659  * Callback for handling random numbers
1660  */
1661 static void
1662 ubsec_callback2(struct ubsec_softc *sc, struct ubsec_q2 *q)
1663 {
1664 	struct cryptkop *krp;
1665 	struct ubsec_ctx_keyop *ctx;
1666 
1667 	ctx = (struct ubsec_ctx_keyop *)q->q_ctx.dma_vaddr;
1668 	ubsec_dma_sync(&q->q_ctx, BUS_DMASYNC_POSTWRITE);
1669 
1670 	switch (q->q_type) {
1671 #ifndef UBSEC_NO_RNG
1672 	case UBS_CTXOP_RNGBYPASS: {
1673 		struct ubsec_q2_rng *rng = (struct ubsec_q2_rng *)q;
1674 
1675 		ubsec_dma_sync(&rng->rng_buf, BUS_DMASYNC_POSTREAD);
1676 		(*sc->sc_harvest)(sc->sc_rndtest,
1677 			rng->rng_buf.dma_vaddr,
1678 			UBSEC_RNG_BUFSIZ*sizeof (u_int32_t));
1679 		rng->rng_used = 0;
1680 		callout_reset(&sc->sc_rngto, sc->sc_rnghz, ubsec_rng, sc);
1681 		break;
1682 	}
1683 #endif
1684 	case UBS_CTXOP_MODEXP: {
1685 		struct ubsec_q2_modexp *me = (struct ubsec_q2_modexp *)q;
1686 		u_int rlen, clen;
1687 
1688 		krp = me->me_krp;
1689 		rlen = (me->me_modbits + 7) / 8;
1690 		clen = (krp->krp_param[krp->krp_iparams].crp_nbits + 7) / 8;
1691 
1692 		ubsec_dma_sync(&me->me_M, BUS_DMASYNC_POSTWRITE);
1693 		ubsec_dma_sync(&me->me_E, BUS_DMASYNC_POSTWRITE);
1694 		ubsec_dma_sync(&me->me_C, BUS_DMASYNC_POSTREAD);
1695 		ubsec_dma_sync(&me->me_epb, BUS_DMASYNC_POSTWRITE);
1696 
1697 		if (clen < rlen)
1698 			krp->krp_status = E2BIG;
1699 		else {
1700 			if (sc->sc_flags & UBS_FLAGS_HWNORM) {
1701 				bzero(krp->krp_param[krp->krp_iparams].crp_p,
1702 				    (krp->krp_param[krp->krp_iparams].crp_nbits
1703 					+ 7) / 8);
1704 				bcopy(me->me_C.dma_vaddr,
1705 				    krp->krp_param[krp->krp_iparams].crp_p,
1706 				    (me->me_modbits + 7) / 8);
1707 			} else
1708 				ubsec_kshift_l(me->me_shiftbits,
1709 				    me->me_C.dma_vaddr, me->me_normbits,
1710 				    krp->krp_param[krp->krp_iparams].crp_p,
1711 				    krp->krp_param[krp->krp_iparams].crp_nbits);
1712 		}
1713 
1714 		crypto_kdone(krp);
1715 
1716 		/* bzero all potentially sensitive data */
1717 		bzero(me->me_E.dma_vaddr, me->me_E.dma_size);
1718 		bzero(me->me_M.dma_vaddr, me->me_M.dma_size);
1719 		bzero(me->me_C.dma_vaddr, me->me_C.dma_size);
1720 		bzero(me->me_q.q_ctx.dma_vaddr, me->me_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, &me->me_q, q_next);
1724 		break;
1725 	}
1726 	case UBS_CTXOP_RSAPRIV: {
1727 		struct ubsec_q2_rsapriv *rp = (struct ubsec_q2_rsapriv *)q;
1728 		u_int len;
1729 
1730 		krp = rp->rpr_krp;
1731 		ubsec_dma_sync(&rp->rpr_msgin, BUS_DMASYNC_POSTWRITE);
1732 		ubsec_dma_sync(&rp->rpr_msgout, BUS_DMASYNC_POSTREAD);
1733 
1734 		len = (krp->krp_param[UBS_RSAPRIV_PAR_MSGOUT].crp_nbits + 7) / 8;
1735 		bcopy(rp->rpr_msgout.dma_vaddr,
1736 		    krp->krp_param[UBS_RSAPRIV_PAR_MSGOUT].crp_p, len);
1737 
1738 		crypto_kdone(krp);
1739 
1740 		bzero(rp->rpr_msgin.dma_vaddr, rp->rpr_msgin.dma_size);
1741 		bzero(rp->rpr_msgout.dma_vaddr, rp->rpr_msgout.dma_size);
1742 		bzero(rp->rpr_q.q_ctx.dma_vaddr, rp->rpr_q.q_ctx.dma_size);
1743 
1744 		/* Can't free here, so put us on the free list. */
1745 		SIMPLEQ_INSERT_TAIL(&sc->sc_q2free, &rp->rpr_q, q_next);
1746 		break;
1747 	}
1748 	default:
1749 		device_printf(sc->sc_dev, "unknown ctx op: %x\n",
1750 		    letoh16(ctx->ctx_op));
1751 		break;
1752 	}
1753 }
1754 
1755 #ifndef UBSEC_NO_RNG
1756 static void
1757 ubsec_rng(void *vsc)
1758 {
1759 	struct ubsec_softc *sc = vsc;
1760 	struct ubsec_q2_rng *rng = &sc->sc_rng;
1761 	struct ubsec_mcr *mcr;
1762 	struct ubsec_ctx_rngbypass *ctx;
1763 
1764 	crit_enter();
1765 	if (rng->rng_used) {
1766 		crit_exit();
1767 		return;
1768 	}
1769 	sc->sc_nqueue2++;
1770 	if (sc->sc_nqueue2 >= UBS_MAX_NQUEUE)
1771 		goto out;
1772 
1773 	mcr = (struct ubsec_mcr *)rng->rng_q.q_mcr.dma_vaddr;
1774 	ctx = (struct ubsec_ctx_rngbypass *)rng->rng_q.q_ctx.dma_vaddr;
1775 
1776 	mcr->mcr_pkts = htole16(1);
1777 	mcr->mcr_flags = 0;
1778 	mcr->mcr_cmdctxp = htole32(rng->rng_q.q_ctx.dma_paddr);
1779 	mcr->mcr_ipktbuf.pb_addr = mcr->mcr_ipktbuf.pb_next = 0;
1780 	mcr->mcr_ipktbuf.pb_len = 0;
1781 	mcr->mcr_reserved = mcr->mcr_pktlen = 0;
1782 	mcr->mcr_opktbuf.pb_addr = htole32(rng->rng_buf.dma_paddr);
1783 	mcr->mcr_opktbuf.pb_len = htole32(((sizeof(u_int32_t) * UBSEC_RNG_BUFSIZ)) &
1784 	    UBS_PKTBUF_LEN);
1785 	mcr->mcr_opktbuf.pb_next = 0;
1786 
1787 	ctx->rbp_len = htole16(sizeof(struct ubsec_ctx_rngbypass));
1788 	ctx->rbp_op = htole16(UBS_CTXOP_RNGBYPASS);
1789 	rng->rng_q.q_type = UBS_CTXOP_RNGBYPASS;
1790 
1791 	ubsec_dma_sync(&rng->rng_buf, BUS_DMASYNC_PREREAD);
1792 
1793 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue2, &rng->rng_q, q_next);
1794 	rng->rng_used = 1;
1795 	ubsec_feed2(sc);
1796 	ubsecstats.hst_rng++;
1797 	crit_exit();
1798 
1799 	return;
1800 
1801 out:
1802 	/*
1803 	 * Something weird happened, generate our own call back.
1804 	 */
1805 	sc->sc_nqueue2--;
1806 	crit_exit();
1807 	callout_reset(&sc->sc_rngto, sc->sc_rnghz, ubsec_rng, sc);
1808 }
1809 #endif /* UBSEC_NO_RNG */
1810 
1811 static void
1812 ubsec_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1813 {
1814 	bus_addr_t *paddr = (bus_addr_t*) arg;
1815 	*paddr = segs->ds_addr;
1816 }
1817 
1818 static int
1819 ubsec_dma_malloc(
1820 	struct ubsec_softc *sc,
1821 	bus_size_t size,
1822 	struct ubsec_dma_alloc *dma,
1823 	int mapflags
1824 )
1825 {
1826 	int r;
1827 
1828 	/* XXX could specify sc_dmat as parent but that just adds overhead */
1829 	r = bus_dma_tag_create(NULL,			/* parent */
1830 			       1, 0,			/* alignment, bounds */
1831 			       BUS_SPACE_MAXADDR_32BIT,	/* lowaddr */
1832 			       BUS_SPACE_MAXADDR,	/* highaddr */
1833 			       NULL, NULL,		/* filter, filterarg */
1834 			       size,			/* maxsize */
1835 			       1,			/* nsegments */
1836 			       size,			/* maxsegsize */
1837 			       BUS_DMA_ALLOCNOW,	/* flags */
1838 			       &dma->dma_tag);
1839 	if (r != 0) {
1840 		device_printf(sc->sc_dev, "ubsec_dma_malloc: "
1841 			"bus_dma_tag_create failed; error %u\n", r);
1842 		goto fail_0;
1843 	}
1844 
1845 	r = bus_dmamap_create(dma->dma_tag, BUS_DMA_NOWAIT, &dma->dma_map);
1846 	if (r != 0) {
1847 		device_printf(sc->sc_dev, "ubsec_dma_malloc: "
1848 			"bus_dmamap_create failed; error %u\n", r);
1849 		goto fail_1;
1850 	}
1851 
1852 	r = bus_dmamem_alloc(dma->dma_tag, (void**) &dma->dma_vaddr,
1853 			     BUS_DMA_NOWAIT, &dma->dma_map);
1854 	if (r != 0) {
1855 		device_printf(sc->sc_dev, "ubsec_dma_malloc: "
1856 			"bus_dmammem_alloc failed; size %u, error %u\n",
1857 			size, r);
1858 		goto fail_2;
1859 	}
1860 
1861 	r = bus_dmamap_load(dma->dma_tag, dma->dma_map, dma->dma_vaddr,
1862 		            size,
1863 			    ubsec_dmamap_cb,
1864 			    &dma->dma_paddr,
1865 			    mapflags | BUS_DMA_NOWAIT);
1866 	if (r != 0) {
1867 		device_printf(sc->sc_dev, "ubsec_dma_malloc: "
1868 			"bus_dmamap_load failed; error %u\n", r);
1869 		goto fail_3;
1870 	}
1871 
1872 	dma->dma_size = size;
1873 	return (0);
1874 
1875 fail_3:
1876 	bus_dmamap_unload(dma->dma_tag, dma->dma_map);
1877 fail_2:
1878 	bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
1879 fail_1:
1880 	bus_dmamap_destroy(dma->dma_tag, dma->dma_map);
1881 	bus_dma_tag_destroy(dma->dma_tag);
1882 fail_0:
1883 	dma->dma_map = NULL;
1884 	dma->dma_tag = NULL;
1885 	return (r);
1886 }
1887 
1888 static void
1889 ubsec_dma_free(struct ubsec_softc *sc, struct ubsec_dma_alloc *dma)
1890 {
1891 	bus_dmamap_unload(dma->dma_tag, dma->dma_map);
1892 	bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
1893 	bus_dmamap_destroy(dma->dma_tag, dma->dma_map);
1894 	bus_dma_tag_destroy(dma->dma_tag);
1895 }
1896 
1897 /*
1898  * Resets the board.  Values in the regesters are left as is
1899  * from the reset (i.e. initial values are assigned elsewhere).
1900  */
1901 static void
1902 ubsec_reset_board(struct ubsec_softc *sc)
1903 {
1904     volatile u_int32_t ctrl;
1905 
1906     ctrl = READ_REG(sc, BS_CTRL);
1907     ctrl |= BS_CTRL_RESET;
1908     WRITE_REG(sc, BS_CTRL, ctrl);
1909 
1910     /*
1911      * Wait aprox. 30 PCI clocks = 900 ns = 0.9 us
1912      */
1913     DELAY(10);
1914 }
1915 
1916 /*
1917  * Init Broadcom registers
1918  */
1919 static void
1920 ubsec_init_board(struct ubsec_softc *sc)
1921 {
1922 	u_int32_t ctrl;
1923 
1924 	ctrl = READ_REG(sc, BS_CTRL);
1925 	ctrl &= ~(BS_CTRL_BE32 | BS_CTRL_BE64);
1926 	ctrl |= BS_CTRL_LITTLE_ENDIAN | BS_CTRL_MCR1INT;
1927 
1928 	if (sc->sc_flags & (UBS_FLAGS_KEY|UBS_FLAGS_RNG))
1929 		ctrl |= BS_CTRL_MCR2INT;
1930 	else
1931 		ctrl &= ~BS_CTRL_MCR2INT;
1932 
1933 	if (sc->sc_flags & UBS_FLAGS_HWNORM)
1934 		ctrl &= ~BS_CTRL_SWNORM;
1935 
1936 	WRITE_REG(sc, BS_CTRL, ctrl);
1937 }
1938 
1939 /*
1940  * Init Broadcom PCI registers
1941  */
1942 static void
1943 ubsec_init_pciregs(device_t dev)
1944 {
1945 #if 0
1946 	u_int32_t misc;
1947 
1948 	misc = pci_conf_read(pc, pa->pa_tag, BS_RTY_TOUT);
1949 	misc = (misc & ~(UBS_PCI_RTY_MASK << UBS_PCI_RTY_SHIFT))
1950 	    | ((UBS_DEF_RTY & 0xff) << UBS_PCI_RTY_SHIFT);
1951 	misc = (misc & ~(UBS_PCI_TOUT_MASK << UBS_PCI_TOUT_SHIFT))
1952 	    | ((UBS_DEF_TOUT & 0xff) << UBS_PCI_TOUT_SHIFT);
1953 	pci_conf_write(pc, pa->pa_tag, BS_RTY_TOUT, misc);
1954 #endif
1955 
1956 	/*
1957 	 * This will set the cache line size to 1, this will
1958 	 * force the BCM58xx chip just to do burst read/writes.
1959 	 * Cache line read/writes are to slow
1960 	 */
1961 	pci_write_config(dev, PCIR_CACHELNSZ, UBS_DEF_CACHELINE, 1);
1962 }
1963 
1964 /*
1965  * Clean up after a chip crash.
1966  * It is assumed that the caller in splimp()
1967  */
1968 static void
1969 ubsec_cleanchip(struct ubsec_softc *sc)
1970 {
1971 	struct ubsec_q *q;
1972 
1973 	while (!SIMPLEQ_EMPTY(&sc->sc_qchip)) {
1974 		q = SIMPLEQ_FIRST(&sc->sc_qchip);
1975 		SIMPLEQ_REMOVE_HEAD(&sc->sc_qchip, q, q_next);
1976 		ubsec_free_q(sc, q);
1977 	}
1978 	sc->sc_nqchip = 0;
1979 }
1980 
1981 /*
1982  * free a ubsec_q
1983  * It is assumed that the caller is within spimp()
1984  */
1985 static int
1986 ubsec_free_q(struct ubsec_softc *sc, struct ubsec_q *q)
1987 {
1988 	struct ubsec_q *q2;
1989 	struct cryptop *crp;
1990 	int npkts;
1991 	int i;
1992 
1993 	npkts = q->q_nstacked_mcrs;
1994 
1995 	for (i = 0; i < npkts; i++) {
1996 		if(q->q_stacked_mcr[i]) {
1997 			q2 = q->q_stacked_mcr[i];
1998 
1999 			if ((q2->q_dst_m != NULL) && (q2->q_src_m != q2->q_dst_m))
2000 				m_freem(q2->q_dst_m);
2001 
2002 			crp = (struct cryptop *)q2->q_crp;
2003 
2004 			SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q2, q_next);
2005 
2006 			crp->crp_etype = EFAULT;
2007 			crypto_done(crp);
2008 		} else {
2009 			break;
2010 		}
2011 	}
2012 
2013 	/*
2014 	 * Free header MCR
2015 	 */
2016 	if ((q->q_dst_m != NULL) && (q->q_src_m != q->q_dst_m))
2017 		m_freem(q->q_dst_m);
2018 
2019 	crp = (struct cryptop *)q->q_crp;
2020 
2021 	SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
2022 
2023 	crp->crp_etype = EFAULT;
2024 	crypto_done(crp);
2025 	return(0);
2026 }
2027 
2028 /*
2029  * Routine to reset the chip and clean up.
2030  * It is assumed that the caller is in splimp()
2031  */
2032 static void
2033 ubsec_totalreset(struct ubsec_softc *sc)
2034 {
2035 	ubsec_reset_board(sc);
2036 	ubsec_init_board(sc);
2037 	ubsec_cleanchip(sc);
2038 }
2039 
2040 static int
2041 ubsec_dmamap_aligned(struct ubsec_operand *op)
2042 {
2043 	int i;
2044 
2045 	for (i = 0; i < op->nsegs; i++) {
2046 		if (op->segs[i].ds_addr & 3)
2047 			return (0);
2048 		if ((i != (op->nsegs - 1)) &&
2049 		    (op->segs[i].ds_len & 3))
2050 			return (0);
2051 	}
2052 	return (1);
2053 }
2054 
2055 static void
2056 ubsec_kfree(struct ubsec_softc *sc, struct ubsec_q2 *q)
2057 {
2058 	switch (q->q_type) {
2059 	case UBS_CTXOP_MODEXP: {
2060 		struct ubsec_q2_modexp *me = (struct ubsec_q2_modexp *)q;
2061 
2062 		ubsec_dma_free(sc, &me->me_q.q_mcr);
2063 		ubsec_dma_free(sc, &me->me_q.q_ctx);
2064 		ubsec_dma_free(sc, &me->me_M);
2065 		ubsec_dma_free(sc, &me->me_E);
2066 		ubsec_dma_free(sc, &me->me_C);
2067 		ubsec_dma_free(sc, &me->me_epb);
2068 		kfree(me, M_DEVBUF);
2069 		break;
2070 	}
2071 	case UBS_CTXOP_RSAPRIV: {
2072 		struct ubsec_q2_rsapriv *rp = (struct ubsec_q2_rsapriv *)q;
2073 
2074 		ubsec_dma_free(sc, &rp->rpr_q.q_mcr);
2075 		ubsec_dma_free(sc, &rp->rpr_q.q_ctx);
2076 		ubsec_dma_free(sc, &rp->rpr_msgin);
2077 		ubsec_dma_free(sc, &rp->rpr_msgout);
2078 		kfree(rp, M_DEVBUF);
2079 		break;
2080 	}
2081 	default:
2082 		device_printf(sc->sc_dev, "invalid kfree 0x%x\n", q->q_type);
2083 		break;
2084 	}
2085 }
2086 
2087 static int
2088 ubsec_kprocess(void *arg, struct cryptkop *krp, int hint)
2089 {
2090 	struct ubsec_softc *sc = arg;
2091 	int r;
2092 
2093 	if (krp == NULL || krp->krp_callback == NULL)
2094 		return (EINVAL);
2095 
2096 	while (!SIMPLEQ_EMPTY(&sc->sc_q2free)) {
2097 		struct ubsec_q2 *q;
2098 
2099 		q = SIMPLEQ_FIRST(&sc->sc_q2free);
2100 		SIMPLEQ_REMOVE_HEAD(&sc->sc_q2free, q, q_next);
2101 		ubsec_kfree(sc, q);
2102 	}
2103 
2104 	switch (krp->krp_op) {
2105 	case CRK_MOD_EXP:
2106 		if (sc->sc_flags & UBS_FLAGS_HWNORM)
2107 			r = ubsec_kprocess_modexp_hw(sc, krp, hint);
2108 		else
2109 			r = ubsec_kprocess_modexp_sw(sc, krp, hint);
2110 		break;
2111 	case CRK_MOD_EXP_CRT:
2112 		return (ubsec_kprocess_rsapriv(sc, krp, hint));
2113 	default:
2114 		device_printf(sc->sc_dev, "kprocess: invalid op 0x%x\n",
2115 		    krp->krp_op);
2116 		krp->krp_status = EOPNOTSUPP;
2117 		crypto_kdone(krp);
2118 		return (0);
2119 	}
2120 	return (0);			/* silence compiler */
2121 }
2122 
2123 /*
2124  * Start computation of cr[C] = (cr[M] ^ cr[E]) mod cr[N] (sw normalization)
2125  */
2126 static int
2127 ubsec_kprocess_modexp_sw(struct ubsec_softc *sc, struct cryptkop *krp, int hint)
2128 {
2129 	struct ubsec_q2_modexp *me;
2130 	struct ubsec_mcr *mcr;
2131 	struct ubsec_ctx_modexp *ctx;
2132 	struct ubsec_pktbuf *epb;
2133 	int err = 0;
2134 	u_int nbits, normbits, mbits, shiftbits, ebits;
2135 
2136 	me = kmalloc(sizeof *me, M_DEVBUF, M_INTWAIT | M_ZERO);
2137 	me->me_krp = krp;
2138 	me->me_q.q_type = UBS_CTXOP_MODEXP;
2139 
2140 	nbits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_N]);
2141 	if (nbits <= 512)
2142 		normbits = 512;
2143 	else if (nbits <= 768)
2144 		normbits = 768;
2145 	else if (nbits <= 1024)
2146 		normbits = 1024;
2147 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && nbits <= 1536)
2148 		normbits = 1536;
2149 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && nbits <= 2048)
2150 		normbits = 2048;
2151 	else {
2152 		err = E2BIG;
2153 		goto errout;
2154 	}
2155 
2156 	shiftbits = normbits - nbits;
2157 
2158 	me->me_modbits = nbits;
2159 	me->me_shiftbits = shiftbits;
2160 	me->me_normbits = normbits;
2161 
2162 	/* Sanity check: result bits must be >= true modulus bits. */
2163 	if (krp->krp_param[krp->krp_iparams].crp_nbits < nbits) {
2164 		err = ERANGE;
2165 		goto errout;
2166 	}
2167 
2168 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_mcr),
2169 	    &me->me_q.q_mcr, 0)) {
2170 		err = ENOMEM;
2171 		goto errout;
2172 	}
2173 	mcr = (struct ubsec_mcr *)me->me_q.q_mcr.dma_vaddr;
2174 
2175 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_ctx_modexp),
2176 	    &me->me_q.q_ctx, 0)) {
2177 		err = ENOMEM;
2178 		goto errout;
2179 	}
2180 
2181 	mbits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_M]);
2182 	if (mbits > nbits) {
2183 		err = E2BIG;
2184 		goto errout;
2185 	}
2186 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_M, 0)) {
2187 		err = ENOMEM;
2188 		goto errout;
2189 	}
2190 	ubsec_kshift_r(shiftbits,
2191 	    krp->krp_param[UBS_MODEXP_PAR_M].crp_p, mbits,
2192 	    me->me_M.dma_vaddr, normbits);
2193 
2194 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_C, 0)) {
2195 		err = ENOMEM;
2196 		goto errout;
2197 	}
2198 	bzero(me->me_C.dma_vaddr, me->me_C.dma_size);
2199 
2200 	ebits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_E]);
2201 	if (ebits > nbits) {
2202 		err = E2BIG;
2203 		goto errout;
2204 	}
2205 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_E, 0)) {
2206 		err = ENOMEM;
2207 		goto errout;
2208 	}
2209 	ubsec_kshift_r(shiftbits,
2210 	    krp->krp_param[UBS_MODEXP_PAR_E].crp_p, ebits,
2211 	    me->me_E.dma_vaddr, normbits);
2212 
2213 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_pktbuf),
2214 	    &me->me_epb, 0)) {
2215 		err = ENOMEM;
2216 		goto errout;
2217 	}
2218 	epb = (struct ubsec_pktbuf *)me->me_epb.dma_vaddr;
2219 	epb->pb_addr = htole32(me->me_E.dma_paddr);
2220 	epb->pb_next = 0;
2221 	epb->pb_len = htole32(normbits / 8);
2222 
2223 #ifdef UBSEC_DEBUG
2224 	if (ubsec_debug) {
2225 		kprintf("Epb ");
2226 		ubsec_dump_pb(epb);
2227 	}
2228 #endif
2229 
2230 	mcr->mcr_pkts = htole16(1);
2231 	mcr->mcr_flags = 0;
2232 	mcr->mcr_cmdctxp = htole32(me->me_q.q_ctx.dma_paddr);
2233 	mcr->mcr_reserved = 0;
2234 	mcr->mcr_pktlen = 0;
2235 
2236 	mcr->mcr_ipktbuf.pb_addr = htole32(me->me_M.dma_paddr);
2237 	mcr->mcr_ipktbuf.pb_len = htole32(normbits / 8);
2238 	mcr->mcr_ipktbuf.pb_next = htole32(me->me_epb.dma_paddr);
2239 
2240 	mcr->mcr_opktbuf.pb_addr = htole32(me->me_C.dma_paddr);
2241 	mcr->mcr_opktbuf.pb_next = 0;
2242 	mcr->mcr_opktbuf.pb_len = htole32(normbits / 8);
2243 
2244 #ifdef DIAGNOSTIC
2245 	/* Misaligned output buffer will hang the chip. */
2246 	if ((letoh32(mcr->mcr_opktbuf.pb_addr) & 3) != 0)
2247 		panic("%s: modexp invalid addr 0x%x\n",
2248 		    device_get_nameunit(sc->sc_dev),
2249 		    letoh32(mcr->mcr_opktbuf.pb_addr));
2250 	if ((letoh32(mcr->mcr_opktbuf.pb_len) & 3) != 0)
2251 		panic("%s: modexp invalid len 0x%x\n",
2252 		    device_get_nameunit(sc->sc_dev),
2253 		    letoh32(mcr->mcr_opktbuf.pb_len));
2254 #endif
2255 
2256 	ctx = (struct ubsec_ctx_modexp *)me->me_q.q_ctx.dma_vaddr;
2257 	bzero(ctx, sizeof(*ctx));
2258 	ubsec_kshift_r(shiftbits,
2259 	    krp->krp_param[UBS_MODEXP_PAR_N].crp_p, nbits,
2260 	    ctx->me_N, normbits);
2261 	ctx->me_len = htole16((normbits / 8) + (4 * sizeof(u_int16_t)));
2262 	ctx->me_op = htole16(UBS_CTXOP_MODEXP);
2263 	ctx->me_E_len = htole16(nbits);
2264 	ctx->me_N_len = htole16(nbits);
2265 
2266 #ifdef UBSEC_DEBUG
2267 	if (ubsec_debug) {
2268 		ubsec_dump_mcr(mcr);
2269 		ubsec_dump_ctx2((struct ubsec_ctx_keyop *)ctx);
2270 	}
2271 #endif
2272 
2273 	/*
2274 	 * ubsec_feed2 will sync mcr and ctx, we just need to sync
2275 	 * everything else.
2276 	 */
2277 	ubsec_dma_sync(&me->me_M, BUS_DMASYNC_PREWRITE);
2278 	ubsec_dma_sync(&me->me_E, BUS_DMASYNC_PREWRITE);
2279 	ubsec_dma_sync(&me->me_C, BUS_DMASYNC_PREREAD);
2280 	ubsec_dma_sync(&me->me_epb, BUS_DMASYNC_PREWRITE);
2281 
2282 	/* Enqueue and we're done... */
2283 	crit_enter();
2284 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue2, &me->me_q, q_next);
2285 	ubsec_feed2(sc);
2286 	ubsecstats.hst_modexp++;
2287 	crit_exit();
2288 
2289 	return (0);
2290 
2291 errout:
2292 	if (me != NULL) {
2293 		if (me->me_q.q_mcr.dma_map != NULL)
2294 			ubsec_dma_free(sc, &me->me_q.q_mcr);
2295 		if (me->me_q.q_ctx.dma_map != NULL) {
2296 			bzero(me->me_q.q_ctx.dma_vaddr, me->me_q.q_ctx.dma_size);
2297 			ubsec_dma_free(sc, &me->me_q.q_ctx);
2298 		}
2299 		if (me->me_M.dma_map != NULL) {
2300 			bzero(me->me_M.dma_vaddr, me->me_M.dma_size);
2301 			ubsec_dma_free(sc, &me->me_M);
2302 		}
2303 		if (me->me_E.dma_map != NULL) {
2304 			bzero(me->me_E.dma_vaddr, me->me_E.dma_size);
2305 			ubsec_dma_free(sc, &me->me_E);
2306 		}
2307 		if (me->me_C.dma_map != NULL) {
2308 			bzero(me->me_C.dma_vaddr, me->me_C.dma_size);
2309 			ubsec_dma_free(sc, &me->me_C);
2310 		}
2311 		if (me->me_epb.dma_map != NULL)
2312 			ubsec_dma_free(sc, &me->me_epb);
2313 		kfree(me, M_DEVBUF);
2314 	}
2315 	krp->krp_status = err;
2316 	crypto_kdone(krp);
2317 	return (0);
2318 }
2319 
2320 /*
2321  * Start computation of cr[C] = (cr[M] ^ cr[E]) mod cr[N] (hw normalization)
2322  */
2323 static int
2324 ubsec_kprocess_modexp_hw(struct ubsec_softc *sc, struct cryptkop *krp, int hint)
2325 {
2326 	struct ubsec_q2_modexp *me;
2327 	struct ubsec_mcr *mcr;
2328 	struct ubsec_ctx_modexp *ctx;
2329 	struct ubsec_pktbuf *epb;
2330 	int err = 0;
2331 	u_int nbits, normbits, mbits, shiftbits, ebits;
2332 
2333 	me = kmalloc(sizeof *me, M_DEVBUF, M_INTWAIT | M_ZERO);
2334 	me->me_krp = krp;
2335 	me->me_q.q_type = UBS_CTXOP_MODEXP;
2336 
2337 	nbits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_N]);
2338 	if (nbits <= 512)
2339 		normbits = 512;
2340 	else if (nbits <= 768)
2341 		normbits = 768;
2342 	else if (nbits <= 1024)
2343 		normbits = 1024;
2344 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && nbits <= 1536)
2345 		normbits = 1536;
2346 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && nbits <= 2048)
2347 		normbits = 2048;
2348 	else {
2349 		err = E2BIG;
2350 		goto errout;
2351 	}
2352 
2353 	shiftbits = normbits - nbits;
2354 
2355 	/* XXX ??? */
2356 	me->me_modbits = nbits;
2357 	me->me_shiftbits = shiftbits;
2358 	me->me_normbits = normbits;
2359 
2360 	/* Sanity check: result bits must be >= true modulus bits. */
2361 	if (krp->krp_param[krp->krp_iparams].crp_nbits < nbits) {
2362 		err = ERANGE;
2363 		goto errout;
2364 	}
2365 
2366 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_mcr),
2367 	    &me->me_q.q_mcr, 0)) {
2368 		err = ENOMEM;
2369 		goto errout;
2370 	}
2371 	mcr = (struct ubsec_mcr *)me->me_q.q_mcr.dma_vaddr;
2372 
2373 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_ctx_modexp),
2374 	    &me->me_q.q_ctx, 0)) {
2375 		err = ENOMEM;
2376 		goto errout;
2377 	}
2378 
2379 	mbits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_M]);
2380 	if (mbits > nbits) {
2381 		err = E2BIG;
2382 		goto errout;
2383 	}
2384 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_M, 0)) {
2385 		err = ENOMEM;
2386 		goto errout;
2387 	}
2388 	bzero(me->me_M.dma_vaddr, normbits / 8);
2389 	bcopy(krp->krp_param[UBS_MODEXP_PAR_M].crp_p,
2390 	    me->me_M.dma_vaddr, (mbits + 7) / 8);
2391 
2392 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_C, 0)) {
2393 		err = ENOMEM;
2394 		goto errout;
2395 	}
2396 	bzero(me->me_C.dma_vaddr, me->me_C.dma_size);
2397 
2398 	ebits = ubsec_ksigbits(&krp->krp_param[UBS_MODEXP_PAR_E]);
2399 	if (ebits > nbits) {
2400 		err = E2BIG;
2401 		goto errout;
2402 	}
2403 	if (ubsec_dma_malloc(sc, normbits / 8, &me->me_E, 0)) {
2404 		err = ENOMEM;
2405 		goto errout;
2406 	}
2407 	bzero(me->me_E.dma_vaddr, normbits / 8);
2408 	bcopy(krp->krp_param[UBS_MODEXP_PAR_E].crp_p,
2409 	    me->me_E.dma_vaddr, (ebits + 7) / 8);
2410 
2411 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_pktbuf),
2412 	    &me->me_epb, 0)) {
2413 		err = ENOMEM;
2414 		goto errout;
2415 	}
2416 	epb = (struct ubsec_pktbuf *)me->me_epb.dma_vaddr;
2417 	epb->pb_addr = htole32(me->me_E.dma_paddr);
2418 	epb->pb_next = 0;
2419 	epb->pb_len = htole32((ebits + 7) / 8);
2420 
2421 #ifdef UBSEC_DEBUG
2422 	if (ubsec_debug) {
2423 		kprintf("Epb ");
2424 		ubsec_dump_pb(epb);
2425 	}
2426 #endif
2427 
2428 	mcr->mcr_pkts = htole16(1);
2429 	mcr->mcr_flags = 0;
2430 	mcr->mcr_cmdctxp = htole32(me->me_q.q_ctx.dma_paddr);
2431 	mcr->mcr_reserved = 0;
2432 	mcr->mcr_pktlen = 0;
2433 
2434 	mcr->mcr_ipktbuf.pb_addr = htole32(me->me_M.dma_paddr);
2435 	mcr->mcr_ipktbuf.pb_len = htole32(normbits / 8);
2436 	mcr->mcr_ipktbuf.pb_next = htole32(me->me_epb.dma_paddr);
2437 
2438 	mcr->mcr_opktbuf.pb_addr = htole32(me->me_C.dma_paddr);
2439 	mcr->mcr_opktbuf.pb_next = 0;
2440 	mcr->mcr_opktbuf.pb_len = htole32(normbits / 8);
2441 
2442 #ifdef DIAGNOSTIC
2443 	/* Misaligned output buffer will hang the chip. */
2444 	if ((letoh32(mcr->mcr_opktbuf.pb_addr) & 3) != 0)
2445 		panic("%s: modexp invalid addr 0x%x\n",
2446 		    device_get_nameunit(sc->sc_dev),
2447 		    letoh32(mcr->mcr_opktbuf.pb_addr));
2448 	if ((letoh32(mcr->mcr_opktbuf.pb_len) & 3) != 0)
2449 		panic("%s: modexp invalid len 0x%x\n",
2450 		    device_get_nameunit(sc->sc_dev),
2451 		    letoh32(mcr->mcr_opktbuf.pb_len));
2452 #endif
2453 
2454 	ctx = (struct ubsec_ctx_modexp *)me->me_q.q_ctx.dma_vaddr;
2455 	bzero(ctx, sizeof(*ctx));
2456 	bcopy(krp->krp_param[UBS_MODEXP_PAR_N].crp_p, ctx->me_N,
2457 	    (nbits + 7) / 8);
2458 	ctx->me_len = htole16((normbits / 8) + (4 * sizeof(u_int16_t)));
2459 	ctx->me_op = htole16(UBS_CTXOP_MODEXP);
2460 	ctx->me_E_len = htole16(ebits);
2461 	ctx->me_N_len = htole16(nbits);
2462 
2463 #ifdef UBSEC_DEBUG
2464 	if (ubsec_debug) {
2465 		ubsec_dump_mcr(mcr);
2466 		ubsec_dump_ctx2((struct ubsec_ctx_keyop *)ctx);
2467 	}
2468 #endif
2469 
2470 	/*
2471 	 * ubsec_feed2 will sync mcr and ctx, we just need to sync
2472 	 * everything else.
2473 	 */
2474 	ubsec_dma_sync(&me->me_M, BUS_DMASYNC_PREWRITE);
2475 	ubsec_dma_sync(&me->me_E, BUS_DMASYNC_PREWRITE);
2476 	ubsec_dma_sync(&me->me_C, BUS_DMASYNC_PREREAD);
2477 	ubsec_dma_sync(&me->me_epb, BUS_DMASYNC_PREWRITE);
2478 
2479 	/* Enqueue and we're done... */
2480 	crit_enter();
2481 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue2, &me->me_q, q_next);
2482 	ubsec_feed2(sc);
2483 	crit_exit();
2484 
2485 	return (0);
2486 
2487 errout:
2488 	if (me != NULL) {
2489 		if (me->me_q.q_mcr.dma_map != NULL)
2490 			ubsec_dma_free(sc, &me->me_q.q_mcr);
2491 		if (me->me_q.q_ctx.dma_map != NULL) {
2492 			bzero(me->me_q.q_ctx.dma_vaddr, me->me_q.q_ctx.dma_size);
2493 			ubsec_dma_free(sc, &me->me_q.q_ctx);
2494 		}
2495 		if (me->me_M.dma_map != NULL) {
2496 			bzero(me->me_M.dma_vaddr, me->me_M.dma_size);
2497 			ubsec_dma_free(sc, &me->me_M);
2498 		}
2499 		if (me->me_E.dma_map != NULL) {
2500 			bzero(me->me_E.dma_vaddr, me->me_E.dma_size);
2501 			ubsec_dma_free(sc, &me->me_E);
2502 		}
2503 		if (me->me_C.dma_map != NULL) {
2504 			bzero(me->me_C.dma_vaddr, me->me_C.dma_size);
2505 			ubsec_dma_free(sc, &me->me_C);
2506 		}
2507 		if (me->me_epb.dma_map != NULL)
2508 			ubsec_dma_free(sc, &me->me_epb);
2509 		kfree(me, M_DEVBUF);
2510 	}
2511 	krp->krp_status = err;
2512 	crypto_kdone(krp);
2513 	return (0);
2514 }
2515 
2516 static int
2517 ubsec_kprocess_rsapriv(struct ubsec_softc *sc, struct cryptkop *krp, int hint)
2518 {
2519 	struct ubsec_q2_rsapriv *rp = NULL;
2520 	struct ubsec_mcr *mcr;
2521 	struct ubsec_ctx_rsapriv *ctx;
2522 	int err = 0;
2523 	u_int padlen, msglen;
2524 
2525 	msglen = ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_P]);
2526 	padlen = ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_Q]);
2527 	if (msglen > padlen)
2528 		padlen = msglen;
2529 
2530 	if (padlen <= 256)
2531 		padlen = 256;
2532 	else if (padlen <= 384)
2533 		padlen = 384;
2534 	else if (padlen <= 512)
2535 		padlen = 512;
2536 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && padlen <= 768)
2537 		padlen = 768;
2538 	else if (sc->sc_flags & UBS_FLAGS_BIGKEY && padlen <= 1024)
2539 		padlen = 1024;
2540 	else {
2541 		err = E2BIG;
2542 		goto errout;
2543 	}
2544 
2545 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_DP]) > padlen) {
2546 		err = E2BIG;
2547 		goto errout;
2548 	}
2549 
2550 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_DQ]) > padlen) {
2551 		err = E2BIG;
2552 		goto errout;
2553 	}
2554 
2555 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_PINV]) > padlen) {
2556 		err = E2BIG;
2557 		goto errout;
2558 	}
2559 
2560 	rp = kmalloc(sizeof *rp, M_DEVBUF, M_INTWAIT | M_ZERO);
2561 	rp->rpr_krp = krp;
2562 	rp->rpr_q.q_type = UBS_CTXOP_RSAPRIV;
2563 
2564 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_mcr),
2565 	    &rp->rpr_q.q_mcr, 0)) {
2566 		err = ENOMEM;
2567 		goto errout;
2568 	}
2569 	mcr = (struct ubsec_mcr *)rp->rpr_q.q_mcr.dma_vaddr;
2570 
2571 	if (ubsec_dma_malloc(sc, sizeof(struct ubsec_ctx_rsapriv),
2572 	    &rp->rpr_q.q_ctx, 0)) {
2573 		err = ENOMEM;
2574 		goto errout;
2575 	}
2576 	ctx = (struct ubsec_ctx_rsapriv *)rp->rpr_q.q_ctx.dma_vaddr;
2577 	bzero(ctx, sizeof *ctx);
2578 
2579 	/* Copy in p */
2580 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_P].crp_p,
2581 	    &ctx->rpr_buf[0 * (padlen / 8)],
2582 	    (krp->krp_param[UBS_RSAPRIV_PAR_P].crp_nbits + 7) / 8);
2583 
2584 	/* Copy in q */
2585 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_Q].crp_p,
2586 	    &ctx->rpr_buf[1 * (padlen / 8)],
2587 	    (krp->krp_param[UBS_RSAPRIV_PAR_Q].crp_nbits + 7) / 8);
2588 
2589 	/* Copy in dp */
2590 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_DP].crp_p,
2591 	    &ctx->rpr_buf[2 * (padlen / 8)],
2592 	    (krp->krp_param[UBS_RSAPRIV_PAR_DP].crp_nbits + 7) / 8);
2593 
2594 	/* Copy in dq */
2595 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_DQ].crp_p,
2596 	    &ctx->rpr_buf[3 * (padlen / 8)],
2597 	    (krp->krp_param[UBS_RSAPRIV_PAR_DQ].crp_nbits + 7) / 8);
2598 
2599 	/* Copy in pinv */
2600 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_PINV].crp_p,
2601 	    &ctx->rpr_buf[4 * (padlen / 8)],
2602 	    (krp->krp_param[UBS_RSAPRIV_PAR_PINV].crp_nbits + 7) / 8);
2603 
2604 	msglen = padlen * 2;
2605 
2606 	/* Copy in input message (aligned buffer/length). */
2607 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_MSGIN]) > msglen) {
2608 		/* Is this likely? */
2609 		err = E2BIG;
2610 		goto errout;
2611 	}
2612 	if (ubsec_dma_malloc(sc, (msglen + 7) / 8, &rp->rpr_msgin, 0)) {
2613 		err = ENOMEM;
2614 		goto errout;
2615 	}
2616 	bzero(rp->rpr_msgin.dma_vaddr, (msglen + 7) / 8);
2617 	bcopy(krp->krp_param[UBS_RSAPRIV_PAR_MSGIN].crp_p,
2618 	    rp->rpr_msgin.dma_vaddr,
2619 	    (krp->krp_param[UBS_RSAPRIV_PAR_MSGIN].crp_nbits + 7) / 8);
2620 
2621 	/* Prepare space for output message (aligned buffer/length). */
2622 	if (ubsec_ksigbits(&krp->krp_param[UBS_RSAPRIV_PAR_MSGOUT]) < msglen) {
2623 		/* Is this likely? */
2624 		err = E2BIG;
2625 		goto errout;
2626 	}
2627 	if (ubsec_dma_malloc(sc, (msglen + 7) / 8, &rp->rpr_msgout, 0)) {
2628 		err = ENOMEM;
2629 		goto errout;
2630 	}
2631 	bzero(rp->rpr_msgout.dma_vaddr, (msglen + 7) / 8);
2632 
2633 	mcr->mcr_pkts = htole16(1);
2634 	mcr->mcr_flags = 0;
2635 	mcr->mcr_cmdctxp = htole32(rp->rpr_q.q_ctx.dma_paddr);
2636 	mcr->mcr_ipktbuf.pb_addr = htole32(rp->rpr_msgin.dma_paddr);
2637 	mcr->mcr_ipktbuf.pb_next = 0;
2638 	mcr->mcr_ipktbuf.pb_len = htole32(rp->rpr_msgin.dma_size);
2639 	mcr->mcr_reserved = 0;
2640 	mcr->mcr_pktlen = htole16(msglen);
2641 	mcr->mcr_opktbuf.pb_addr = htole32(rp->rpr_msgout.dma_paddr);
2642 	mcr->mcr_opktbuf.pb_next = 0;
2643 	mcr->mcr_opktbuf.pb_len = htole32(rp->rpr_msgout.dma_size);
2644 
2645 #ifdef DIAGNOSTIC
2646 	if (rp->rpr_msgin.dma_paddr & 3 || rp->rpr_msgin.dma_size & 3) {
2647 		panic("%s: rsapriv: invalid msgin %x(0x%x)",
2648 		    device_get_nameunit(sc->sc_dev),
2649 		    rp->rpr_msgin.dma_paddr, rp->rpr_msgin.dma_size);
2650 	}
2651 	if (rp->rpr_msgout.dma_paddr & 3 || rp->rpr_msgout.dma_size & 3) {
2652 		panic("%s: rsapriv: invalid msgout %x(0x%x)",
2653 		    device_get_nameunit(sc->sc_dev),
2654 		    rp->rpr_msgout.dma_paddr, rp->rpr_msgout.dma_size);
2655 	}
2656 #endif
2657 
2658 	ctx->rpr_len = (sizeof(u_int16_t) * 4) + (5 * (padlen / 8));
2659 	ctx->rpr_op = htole16(UBS_CTXOP_RSAPRIV);
2660 	ctx->rpr_q_len = htole16(padlen);
2661 	ctx->rpr_p_len = htole16(padlen);
2662 
2663 	/*
2664 	 * ubsec_feed2 will sync mcr and ctx, we just need to sync
2665 	 * everything else.
2666 	 */
2667 	ubsec_dma_sync(&rp->rpr_msgin, BUS_DMASYNC_PREWRITE);
2668 	ubsec_dma_sync(&rp->rpr_msgout, BUS_DMASYNC_PREREAD);
2669 
2670 	/* Enqueue and we're done... */
2671 	crit_enter();
2672 	SIMPLEQ_INSERT_TAIL(&sc->sc_queue2, &rp->rpr_q, q_next);
2673 	ubsec_feed2(sc);
2674 	ubsecstats.hst_modexpcrt++;
2675 	crit_exit();
2676 	return (0);
2677 
2678 errout:
2679 	if (rp != NULL) {
2680 		if (rp->rpr_q.q_mcr.dma_map != NULL)
2681 			ubsec_dma_free(sc, &rp->rpr_q.q_mcr);
2682 		if (rp->rpr_msgin.dma_map != NULL) {
2683 			bzero(rp->rpr_msgin.dma_vaddr, rp->rpr_msgin.dma_size);
2684 			ubsec_dma_free(sc, &rp->rpr_msgin);
2685 		}
2686 		if (rp->rpr_msgout.dma_map != NULL) {
2687 			bzero(rp->rpr_msgout.dma_vaddr, rp->rpr_msgout.dma_size);
2688 			ubsec_dma_free(sc, &rp->rpr_msgout);
2689 		}
2690 		kfree(rp, M_DEVBUF);
2691 	}
2692 	krp->krp_status = err;
2693 	crypto_kdone(krp);
2694 	return (0);
2695 }
2696 
2697 #ifdef UBSEC_DEBUG
2698 static void
2699 ubsec_dump_pb(volatile struct ubsec_pktbuf *pb)
2700 {
2701 	kprintf("addr 0x%x (0x%x) next 0x%x\n",
2702 	    pb->pb_addr, pb->pb_len, pb->pb_next);
2703 }
2704 
2705 static void
2706 ubsec_dump_ctx2(struct ubsec_ctx_keyop *c)
2707 {
2708 	kprintf("CTX (0x%x):\n", c->ctx_len);
2709 	switch (letoh16(c->ctx_op)) {
2710 	case UBS_CTXOP_RNGBYPASS:
2711 	case UBS_CTXOP_RNGSHA1:
2712 		break;
2713 	case UBS_CTXOP_MODEXP:
2714 	{
2715 		struct ubsec_ctx_modexp *cx = (void *)c;
2716 		int i, len;
2717 
2718 		kprintf(" Elen %u, Nlen %u\n",
2719 		    letoh16(cx->me_E_len), letoh16(cx->me_N_len));
2720 		len = (cx->me_N_len + 7)/8;
2721 		for (i = 0; i < len; i++)
2722 			kprintf("%s%02x", (i == 0) ? " N: " : ":", cx->me_N[i]);
2723 		kprintf("\n");
2724 		break;
2725 	}
2726 	default:
2727 		kprintf("unknown context: %x\n", c->ctx_op);
2728 	}
2729 	kprintf("END CTX\n");
2730 }
2731 
2732 static void
2733 ubsec_dump_mcr(struct ubsec_mcr *mcr)
2734 {
2735 	volatile struct ubsec_mcr_add *ma;
2736 	int i;
2737 
2738 	kprintf("MCR:\n");
2739 	kprintf(" pkts: %u, flags 0x%x\n",
2740 	    letoh16(mcr->mcr_pkts), letoh16(mcr->mcr_flags));
2741 	ma = (volatile struct ubsec_mcr_add *)&mcr->mcr_cmdctxp;
2742 	for (i = 0; i < letoh16(mcr->mcr_pkts); i++) {
2743 		kprintf(" %d: ctx 0x%x len 0x%x rsvd 0x%x\n", i,
2744 		    letoh32(ma->mcr_cmdctxp), letoh16(ma->mcr_pktlen),
2745 		    letoh16(ma->mcr_reserved));
2746 		kprintf(" %d: ipkt ", i);
2747 		ubsec_dump_pb(&ma->mcr_ipktbuf);
2748 		kprintf(" %d: opkt ", i);
2749 		ubsec_dump_pb(&ma->mcr_opktbuf);
2750 		ma++;
2751 	}
2752 	kprintf("END MCR\n");
2753 }
2754 #endif /* UBSEC_DEBUG */
2755 
2756 /*
2757  * Return the number of significant bits of a big number.
2758  */
2759 static int
2760 ubsec_ksigbits(struct crparam *cr)
2761 {
2762 	u_int plen = (cr->crp_nbits + 7) / 8;
2763 	int i, sig = plen * 8;
2764 	u_int8_t c, *p = cr->crp_p;
2765 
2766 	for (i = plen - 1; i >= 0; i--) {
2767 		c = p[i];
2768 		if (c != 0) {
2769 			while ((c & 0x80) == 0) {
2770 				sig--;
2771 				c <<= 1;
2772 			}
2773 			break;
2774 		}
2775 		sig -= 8;
2776 	}
2777 	return (sig);
2778 }
2779 
2780 static void
2781 ubsec_kshift_r(
2782 	u_int shiftbits,
2783 	u_int8_t *src, u_int srcbits,
2784 	u_int8_t *dst, u_int dstbits)
2785 {
2786 	u_int slen, dlen;
2787 	int i, si, di, n;
2788 
2789 	slen = (srcbits + 7) / 8;
2790 	dlen = (dstbits + 7) / 8;
2791 
2792 	for (i = 0; i < slen; i++)
2793 		dst[i] = src[i];
2794 	for (i = 0; i < dlen - slen; i++)
2795 		dst[slen + i] = 0;
2796 
2797 	n = shiftbits / 8;
2798 	if (n != 0) {
2799 		si = dlen - n - 1;
2800 		di = dlen - 1;
2801 		while (si >= 0)
2802 			dst[di--] = dst[si--];
2803 		while (di >= 0)
2804 			dst[di--] = 0;
2805 	}
2806 
2807 	n = shiftbits % 8;
2808 	if (n != 0) {
2809 		for (i = dlen - 1; i > 0; i--)
2810 			dst[i] = (dst[i] << n) |
2811 			    (dst[i - 1] >> (8 - n));
2812 		dst[0] = dst[0] << n;
2813 	}
2814 }
2815 
2816 static void
2817 ubsec_kshift_l(
2818 	u_int shiftbits,
2819 	u_int8_t *src, u_int srcbits,
2820 	u_int8_t *dst, u_int dstbits)
2821 {
2822 	int slen, dlen, i, n;
2823 
2824 	slen = (srcbits + 7) / 8;
2825 	dlen = (dstbits + 7) / 8;
2826 
2827 	n = shiftbits / 8;
2828 	for (i = 0; i < slen; i++)
2829 		dst[i] = src[i + n];
2830 	for (i = 0; i < dlen - slen; i++)
2831 		dst[slen + i] = 0;
2832 
2833 	n = shiftbits % 8;
2834 	if (n != 0) {
2835 		for (i = 0; i < (dlen - 1); i++)
2836 			dst[i] = (dst[i] >> n) | (dst[i + 1] << (8 - n));
2837 		dst[dlen - 1] = dst[dlen - 1] >> n;
2838 	}
2839 }
2840