xref: /freebsd/sys/dev/safe/safe.c (revision ba3c1f5972d7b90feb6e6da47905ff2757e0fe57)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2003 Sam Leffler, Errno Consulting
5  * Copyright (c) 2003 Global Technology Associates, Inc.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 /*
34  * SafeNet SafeXcel-1141 hardware crypto accelerator
35  */
36 #include "opt_safe.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/proc.h>
41 #include <sys/errno.h>
42 #include <sys/malloc.h>
43 #include <sys/kernel.h>
44 #include <sys/mbuf.h>
45 #include <sys/module.h>
46 #include <sys/lock.h>
47 #include <sys/mutex.h>
48 #include <sys/sysctl.h>
49 #include <sys/endian.h>
50 #include <sys/uio.h>
51 
52 #include <vm/vm.h>
53 #include <vm/pmap.h>
54 
55 #include <machine/bus.h>
56 #include <machine/resource.h>
57 #include <sys/bus.h>
58 #include <sys/rman.h>
59 
60 #include <opencrypto/cryptodev.h>
61 #include <opencrypto/xform_auth.h>
62 #include <sys/random.h>
63 #include <sys/kobj.h>
64 
65 #include "cryptodev_if.h"
66 
67 #include <dev/pci/pcivar.h>
68 #include <dev/pci/pcireg.h>
69 
70 #ifdef SAFE_RNDTEST
71 #include <dev/rndtest/rndtest.h>
72 #endif
73 #include <dev/safe/safereg.h>
74 #include <dev/safe/safevar.h>
75 
76 #ifndef bswap32
77 #define	bswap32	NTOHL
78 #endif
79 
80 /*
81  * Prototypes and count for the pci_device structure
82  */
83 static	int safe_probe(device_t);
84 static	int safe_attach(device_t);
85 static	int safe_detach(device_t);
86 static	int safe_suspend(device_t);
87 static	int safe_resume(device_t);
88 static	int safe_shutdown(device_t);
89 
90 static	int safe_probesession(device_t, const struct crypto_session_params *);
91 static	int safe_newsession(device_t, crypto_session_t,
92 	    const struct crypto_session_params *);
93 static	int safe_process(device_t, struct cryptop *, int);
94 
95 static device_method_t safe_methods[] = {
96 	/* Device interface */
97 	DEVMETHOD(device_probe,		safe_probe),
98 	DEVMETHOD(device_attach,	safe_attach),
99 	DEVMETHOD(device_detach,	safe_detach),
100 	DEVMETHOD(device_suspend,	safe_suspend),
101 	DEVMETHOD(device_resume,	safe_resume),
102 	DEVMETHOD(device_shutdown,	safe_shutdown),
103 
104 	/* crypto device methods */
105 	DEVMETHOD(cryptodev_probesession, safe_probesession),
106 	DEVMETHOD(cryptodev_newsession,	safe_newsession),
107 	DEVMETHOD(cryptodev_process,	safe_process),
108 
109 	DEVMETHOD_END
110 };
111 
112 static driver_t safe_driver = {
113 	"safe",
114 	safe_methods,
115 	sizeof (struct safe_softc)
116 };
117 
118 DRIVER_MODULE(safe, pci, safe_driver, 0, 0);
119 MODULE_DEPEND(safe, crypto, 1, 1, 1);
120 #ifdef SAFE_RNDTEST
121 MODULE_DEPEND(safe, rndtest, 1, 1, 1);
122 #endif
123 
124 static	void safe_intr(void *);
125 static	void safe_callback(struct safe_softc *, struct safe_ringentry *);
126 static	void safe_feed(struct safe_softc *, struct safe_ringentry *);
127 static	void safe_mcopy(struct mbuf *, struct mbuf *, u_int);
128 #ifndef SAFE_NO_RNG
129 static	void safe_rng_init(struct safe_softc *);
130 static	void safe_rng(void *);
131 #endif /* SAFE_NO_RNG */
132 static	int safe_dma_malloc(struct safe_softc *, bus_size_t,
133 	        struct safe_dma_alloc *, int);
134 #define	safe_dma_sync(_dma, _flags) \
135 	bus_dmamap_sync((_dma)->dma_tag, (_dma)->dma_map, (_flags))
136 static	void safe_dma_free(struct safe_softc *, struct safe_dma_alloc *);
137 static	int safe_dmamap_aligned(const struct safe_operand *);
138 static	int safe_dmamap_uniform(const struct safe_operand *);
139 
140 static	void safe_reset_board(struct safe_softc *);
141 static	void safe_init_board(struct safe_softc *);
142 static	void safe_init_pciregs(device_t dev);
143 static	void safe_cleanchip(struct safe_softc *);
144 static	void safe_totalreset(struct safe_softc *);
145 
146 static	int safe_free_entry(struct safe_softc *, struct safe_ringentry *);
147 
148 static SYSCTL_NODE(_hw, OID_AUTO, safe, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
149     "SafeNet driver parameters");
150 
151 #ifdef SAFE_DEBUG
152 static	void safe_dump_dmastatus(struct safe_softc *, const char *);
153 static	void safe_dump_ringstate(struct safe_softc *, const char *);
154 static	void safe_dump_intrstate(struct safe_softc *, const char *);
155 static	void safe_dump_request(struct safe_softc *, const char *,
156 		struct safe_ringentry *);
157 
158 static	struct safe_softc *safec;		/* for use by hw.safe.dump */
159 
160 static	int safe_debug = 0;
161 SYSCTL_INT(_hw_safe, OID_AUTO, debug, CTLFLAG_RW, &safe_debug,
162 	    0, "control debugging msgs");
163 #define	DPRINTF(_x)	if (safe_debug) printf _x
164 #else
165 #define	DPRINTF(_x)
166 #endif
167 
168 #define	READ_REG(sc,r) \
169 	bus_space_read_4((sc)->sc_st, (sc)->sc_sh, (r))
170 
171 #define WRITE_REG(sc,reg,val) \
172 	bus_space_write_4((sc)->sc_st, (sc)->sc_sh, reg, val)
173 
174 struct safe_stats safestats;
175 SYSCTL_STRUCT(_hw_safe, OID_AUTO, stats, CTLFLAG_RD, &safestats,
176 	    safe_stats, "driver statistics");
177 #ifndef SAFE_NO_RNG
178 static	int safe_rnginterval = 1;		/* poll once a second */
179 SYSCTL_INT(_hw_safe, OID_AUTO, rnginterval, CTLFLAG_RW, &safe_rnginterval,
180 	    0, "RNG polling interval (secs)");
181 static	int safe_rngbufsize = 16;		/* 64 bytes each poll  */
182 SYSCTL_INT(_hw_safe, OID_AUTO, rngbufsize, CTLFLAG_RW, &safe_rngbufsize,
183 	    0, "RNG polling buffer size (32-bit words)");
184 static	int safe_rngmaxalarm = 8;		/* max alarms before reset */
185 SYSCTL_INT(_hw_safe, OID_AUTO, rngmaxalarm, CTLFLAG_RW, &safe_rngmaxalarm,
186 	    0, "RNG max alarms before reset");
187 #endif /* SAFE_NO_RNG */
188 
189 static int
190 safe_probe(device_t dev)
191 {
192 	if (pci_get_vendor(dev) == PCI_VENDOR_SAFENET &&
193 	    pci_get_device(dev) == PCI_PRODUCT_SAFEXCEL)
194 		return (BUS_PROBE_DEFAULT);
195 	return (ENXIO);
196 }
197 
198 static const char*
199 safe_partname(struct safe_softc *sc)
200 {
201 	/* XXX sprintf numbers when not decoded */
202 	switch (pci_get_vendor(sc->sc_dev)) {
203 	case PCI_VENDOR_SAFENET:
204 		switch (pci_get_device(sc->sc_dev)) {
205 		case PCI_PRODUCT_SAFEXCEL: return "SafeNet SafeXcel-1141";
206 		}
207 		return "SafeNet unknown-part";
208 	}
209 	return "Unknown-vendor unknown-part";
210 }
211 
212 #ifndef SAFE_NO_RNG
213 static void
214 default_harvest(struct rndtest_state *rsp, void *buf, u_int count)
215 {
216 	/* MarkM: FIX!! Check that this does not swamp the harvester! */
217 	random_harvest_queue(buf, count, RANDOM_PURE_SAFE);
218 }
219 #endif /* SAFE_NO_RNG */
220 
221 static int
222 safe_attach(device_t dev)
223 {
224 	struct safe_softc *sc = device_get_softc(dev);
225 	u_int32_t raddr;
226 	u_int32_t i;
227 	int rid;
228 
229 	bzero(sc, sizeof (*sc));
230 	sc->sc_dev = dev;
231 
232 	/* XXX handle power management */
233 
234 	pci_enable_busmaster(dev);
235 
236 	/*
237 	 * Setup memory-mapping of PCI registers.
238 	 */
239 	rid = BS_BAR;
240 	sc->sc_sr = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
241 					   RF_ACTIVE);
242 	if (sc->sc_sr == NULL) {
243 		device_printf(dev, "cannot map register space\n");
244 		goto bad;
245 	}
246 	sc->sc_st = rman_get_bustag(sc->sc_sr);
247 	sc->sc_sh = rman_get_bushandle(sc->sc_sr);
248 
249 	/*
250 	 * Arrange interrupt line.
251 	 */
252 	rid = 0;
253 	sc->sc_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
254 					    RF_SHAREABLE|RF_ACTIVE);
255 	if (sc->sc_irq == NULL) {
256 		device_printf(dev, "could not map interrupt\n");
257 		goto bad1;
258 	}
259 	/*
260 	 * NB: Network code assumes we are blocked with splimp()
261 	 *     so make sure the IRQ is mapped appropriately.
262 	 */
263 	if (bus_setup_intr(dev, sc->sc_irq, INTR_TYPE_NET | INTR_MPSAFE,
264 			   NULL, safe_intr, sc, &sc->sc_ih)) {
265 		device_printf(dev, "could not establish interrupt\n");
266 		goto bad2;
267 	}
268 
269 	sc->sc_cid = crypto_get_driverid(dev, sizeof(struct safe_session),
270 	    CRYPTOCAP_F_HARDWARE);
271 	if (sc->sc_cid < 0) {
272 		device_printf(dev, "could not get crypto driver id\n");
273 		goto bad3;
274 	}
275 
276 	sc->sc_chiprev = READ_REG(sc, SAFE_DEVINFO) &
277 		(SAFE_DEVINFO_REV_MAJ | SAFE_DEVINFO_REV_MIN);
278 
279 	/*
280 	 * Setup DMA descriptor area.
281 	 */
282 	if (bus_dma_tag_create(bus_get_dma_tag(dev),	/* parent */
283 			       1,			/* alignment */
284 			       SAFE_DMA_BOUNDARY,	/* boundary */
285 			       BUS_SPACE_MAXADDR_32BIT,	/* lowaddr */
286 			       BUS_SPACE_MAXADDR,	/* highaddr */
287 			       NULL, NULL,		/* filter, filterarg */
288 			       SAFE_MAX_DMA,		/* maxsize */
289 			       SAFE_MAX_PART,		/* nsegments */
290 			       SAFE_MAX_SSIZE,		/* maxsegsize */
291 			       BUS_DMA_ALLOCNOW,	/* flags */
292 			       NULL, NULL,		/* locking */
293 			       &sc->sc_srcdmat)) {
294 		device_printf(dev, "cannot allocate DMA tag\n");
295 		goto bad4;
296 	}
297 	if (bus_dma_tag_create(bus_get_dma_tag(dev),	/* parent */
298 			       1,			/* alignment */
299 			       SAFE_MAX_DSIZE,		/* boundary */
300 			       BUS_SPACE_MAXADDR_32BIT,	/* lowaddr */
301 			       BUS_SPACE_MAXADDR,	/* highaddr */
302 			       NULL, NULL,		/* filter, filterarg */
303 			       SAFE_MAX_DMA,		/* maxsize */
304 			       SAFE_MAX_PART,		/* nsegments */
305 			       SAFE_MAX_DSIZE,		/* maxsegsize */
306 			       BUS_DMA_ALLOCNOW,	/* flags */
307 			       NULL, NULL,		/* locking */
308 			       &sc->sc_dstdmat)) {
309 		device_printf(dev, "cannot allocate DMA tag\n");
310 		goto bad4;
311 	}
312 
313 	/*
314 	 * Allocate packet engine descriptors.
315 	 */
316 	if (safe_dma_malloc(sc,
317 	    SAFE_MAX_NQUEUE * sizeof (struct safe_ringentry),
318 	    &sc->sc_ringalloc, 0)) {
319 		device_printf(dev, "cannot allocate PE descriptor ring\n");
320 		bus_dma_tag_destroy(sc->sc_srcdmat);
321 		goto bad4;
322 	}
323 	/*
324 	 * Hookup the static portion of all our data structures.
325 	 */
326 	sc->sc_ring = (struct safe_ringentry *) sc->sc_ringalloc.dma_vaddr;
327 	sc->sc_ringtop = sc->sc_ring + SAFE_MAX_NQUEUE;
328 	sc->sc_front = sc->sc_ring;
329 	sc->sc_back = sc->sc_ring;
330 	raddr = sc->sc_ringalloc.dma_paddr;
331 	bzero(sc->sc_ring, SAFE_MAX_NQUEUE * sizeof(struct safe_ringentry));
332 	for (i = 0; i < SAFE_MAX_NQUEUE; i++) {
333 		struct safe_ringentry *re = &sc->sc_ring[i];
334 
335 		re->re_desc.d_sa = raddr +
336 			offsetof(struct safe_ringentry, re_sa);
337 		re->re_sa.sa_staterec = raddr +
338 			offsetof(struct safe_ringentry, re_sastate);
339 
340 		raddr += sizeof (struct safe_ringentry);
341 	}
342 	mtx_init(&sc->sc_ringmtx, device_get_nameunit(dev),
343 		"packet engine ring", MTX_DEF);
344 
345 	/*
346 	 * Allocate scatter and gather particle descriptors.
347 	 */
348 	if (safe_dma_malloc(sc, SAFE_TOTAL_SPART * sizeof (struct safe_pdesc),
349 	    &sc->sc_spalloc, 0)) {
350 		device_printf(dev, "cannot allocate source particle "
351 			"descriptor ring\n");
352 		mtx_destroy(&sc->sc_ringmtx);
353 		safe_dma_free(sc, &sc->sc_ringalloc);
354 		bus_dma_tag_destroy(sc->sc_srcdmat);
355 		goto bad4;
356 	}
357 	sc->sc_spring = (struct safe_pdesc *) sc->sc_spalloc.dma_vaddr;
358 	sc->sc_springtop = sc->sc_spring + SAFE_TOTAL_SPART;
359 	sc->sc_spfree = sc->sc_spring;
360 	bzero(sc->sc_spring, SAFE_TOTAL_SPART * sizeof(struct safe_pdesc));
361 
362 	if (safe_dma_malloc(sc, SAFE_TOTAL_DPART * sizeof (struct safe_pdesc),
363 	    &sc->sc_dpalloc, 0)) {
364 		device_printf(dev, "cannot allocate destination particle "
365 			"descriptor ring\n");
366 		mtx_destroy(&sc->sc_ringmtx);
367 		safe_dma_free(sc, &sc->sc_spalloc);
368 		safe_dma_free(sc, &sc->sc_ringalloc);
369 		bus_dma_tag_destroy(sc->sc_dstdmat);
370 		goto bad4;
371 	}
372 	sc->sc_dpring = (struct safe_pdesc *) sc->sc_dpalloc.dma_vaddr;
373 	sc->sc_dpringtop = sc->sc_dpring + SAFE_TOTAL_DPART;
374 	sc->sc_dpfree = sc->sc_dpring;
375 	bzero(sc->sc_dpring, SAFE_TOTAL_DPART * sizeof(struct safe_pdesc));
376 
377 	device_printf(sc->sc_dev, "%s", safe_partname(sc));
378 
379 	sc->sc_devinfo = READ_REG(sc, SAFE_DEVINFO);
380 	if (sc->sc_devinfo & SAFE_DEVINFO_RNG) {
381 		sc->sc_flags |= SAFE_FLAGS_RNG;
382 		printf(" rng");
383 	}
384 	if (sc->sc_devinfo & SAFE_DEVINFO_PKEY) {
385 #if 0
386 		printf(" key");
387 		sc->sc_flags |= SAFE_FLAGS_KEY;
388 #endif
389 	}
390 	if (sc->sc_devinfo & SAFE_DEVINFO_DES) {
391 		printf(" des/3des");
392 	}
393 	if (sc->sc_devinfo & SAFE_DEVINFO_AES) {
394 		printf(" aes");
395 	}
396 	if (sc->sc_devinfo & SAFE_DEVINFO_MD5) {
397 		printf(" md5");
398 	}
399 	if (sc->sc_devinfo & SAFE_DEVINFO_SHA1) {
400 		printf(" sha1");
401 	}
402 	/* XXX other supported algorithms */
403 	printf("\n");
404 
405 	safe_reset_board(sc);		/* reset h/w */
406 	safe_init_pciregs(dev);		/* init pci settings */
407 	safe_init_board(sc);		/* init h/w */
408 
409 #ifndef SAFE_NO_RNG
410 	if (sc->sc_flags & SAFE_FLAGS_RNG) {
411 #ifdef SAFE_RNDTEST
412 		sc->sc_rndtest = rndtest_attach(dev);
413 		if (sc->sc_rndtest)
414 			sc->sc_harvest = rndtest_harvest;
415 		else
416 			sc->sc_harvest = default_harvest;
417 #else
418 		sc->sc_harvest = default_harvest;
419 #endif
420 		safe_rng_init(sc);
421 
422 		callout_init(&sc->sc_rngto, 1);
423 		callout_reset(&sc->sc_rngto, hz*safe_rnginterval, safe_rng, sc);
424 	}
425 #endif /* SAFE_NO_RNG */
426 #ifdef SAFE_DEBUG
427 	safec = sc;			/* for use by hw.safe.dump */
428 #endif
429 	return (0);
430 bad4:
431 	crypto_unregister_all(sc->sc_cid);
432 bad3:
433 	bus_teardown_intr(dev, sc->sc_irq, sc->sc_ih);
434 bad2:
435 	bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq);
436 bad1:
437 	bus_release_resource(dev, SYS_RES_MEMORY, BS_BAR, sc->sc_sr);
438 bad:
439 	return (ENXIO);
440 }
441 
442 /*
443  * Detach a device that successfully probed.
444  */
445 static int
446 safe_detach(device_t dev)
447 {
448 	struct safe_softc *sc = device_get_softc(dev);
449 
450 	/* XXX wait/abort active ops */
451 
452 	WRITE_REG(sc, SAFE_HI_MASK, 0);		/* disable interrupts */
453 
454 	callout_stop(&sc->sc_rngto);
455 
456 	crypto_unregister_all(sc->sc_cid);
457 
458 #ifdef SAFE_RNDTEST
459 	if (sc->sc_rndtest)
460 		rndtest_detach(sc->sc_rndtest);
461 #endif
462 
463 	safe_cleanchip(sc);
464 	safe_dma_free(sc, &sc->sc_dpalloc);
465 	safe_dma_free(sc, &sc->sc_spalloc);
466 	mtx_destroy(&sc->sc_ringmtx);
467 	safe_dma_free(sc, &sc->sc_ringalloc);
468 
469 	bus_generic_detach(dev);
470 	bus_teardown_intr(dev, sc->sc_irq, sc->sc_ih);
471 	bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq);
472 
473 	bus_dma_tag_destroy(sc->sc_srcdmat);
474 	bus_dma_tag_destroy(sc->sc_dstdmat);
475 	bus_release_resource(dev, SYS_RES_MEMORY, BS_BAR, sc->sc_sr);
476 
477 	return (0);
478 }
479 
480 /*
481  * Stop all chip i/o so that the kernel's probe routines don't
482  * get confused by errant DMAs when rebooting.
483  */
484 static int
485 safe_shutdown(device_t dev)
486 {
487 #ifdef notyet
488 	safe_stop(device_get_softc(dev));
489 #endif
490 	return (0);
491 }
492 
493 /*
494  * Device suspend routine.
495  */
496 static int
497 safe_suspend(device_t dev)
498 {
499 	struct safe_softc *sc = device_get_softc(dev);
500 
501 #ifdef notyet
502 	/* XXX stop the device and save PCI settings */
503 #endif
504 	sc->sc_suspended = 1;
505 
506 	return (0);
507 }
508 
509 static int
510 safe_resume(device_t dev)
511 {
512 	struct safe_softc *sc = device_get_softc(dev);
513 
514 #ifdef notyet
515 	/* XXX retore PCI settings and start the device */
516 #endif
517 	sc->sc_suspended = 0;
518 	return (0);
519 }
520 
521 /*
522  * SafeXcel Interrupt routine
523  */
524 static void
525 safe_intr(void *arg)
526 {
527 	struct safe_softc *sc = arg;
528 	volatile u_int32_t stat;
529 
530 	stat = READ_REG(sc, SAFE_HM_STAT);
531 	if (stat == 0)			/* shared irq, not for us */
532 		return;
533 
534 	WRITE_REG(sc, SAFE_HI_CLR, stat);	/* IACK */
535 
536 	if ((stat & SAFE_INT_PE_DDONE)) {
537 		/*
538 		 * Descriptor(s) done; scan the ring and
539 		 * process completed operations.
540 		 */
541 		mtx_lock(&sc->sc_ringmtx);
542 		while (sc->sc_back != sc->sc_front) {
543 			struct safe_ringentry *re = sc->sc_back;
544 #ifdef SAFE_DEBUG
545 			if (safe_debug) {
546 				safe_dump_ringstate(sc, __func__);
547 				safe_dump_request(sc, __func__, re);
548 			}
549 #endif
550 			/*
551 			 * safe_process marks ring entries that were allocated
552 			 * but not used with a csr of zero.  This insures the
553 			 * ring front pointer never needs to be set backwards
554 			 * in the event that an entry is allocated but not used
555 			 * because of a setup error.
556 			 */
557 			if (re->re_desc.d_csr != 0) {
558 				if (!SAFE_PE_CSR_IS_DONE(re->re_desc.d_csr))
559 					break;
560 				if (!SAFE_PE_LEN_IS_DONE(re->re_desc.d_len))
561 					break;
562 				sc->sc_nqchip--;
563 				safe_callback(sc, re);
564 			}
565 			if (++(sc->sc_back) == sc->sc_ringtop)
566 				sc->sc_back = sc->sc_ring;
567 		}
568 		mtx_unlock(&sc->sc_ringmtx);
569 	}
570 
571 	/*
572 	 * Check to see if we got any DMA Error
573 	 */
574 	if (stat & SAFE_INT_PE_ERROR) {
575 		DPRINTF(("dmaerr dmastat %08x\n",
576 			READ_REG(sc, SAFE_PE_DMASTAT)));
577 		safestats.st_dmaerr++;
578 		safe_totalreset(sc);
579 #if 0
580 		safe_feed(sc);
581 #endif
582 	}
583 
584 	if (sc->sc_needwakeup) {		/* XXX check high watermark */
585 		int wakeup = sc->sc_needwakeup & CRYPTO_SYMQ;
586 		DPRINTF(("%s: wakeup crypto %x\n", __func__,
587 			sc->sc_needwakeup));
588 		sc->sc_needwakeup &= ~wakeup;
589 		crypto_unblock(sc->sc_cid, wakeup);
590 	}
591 }
592 
593 /*
594  * safe_feed() - post a request to chip
595  */
596 static void
597 safe_feed(struct safe_softc *sc, struct safe_ringentry *re)
598 {
599 	bus_dmamap_sync(sc->sc_srcdmat, re->re_src_map, BUS_DMASYNC_PREWRITE);
600 	if (re->re_dst_map != NULL)
601 		bus_dmamap_sync(sc->sc_dstdmat, re->re_dst_map,
602 			BUS_DMASYNC_PREREAD);
603 	/* XXX have no smaller granularity */
604 	safe_dma_sync(&sc->sc_ringalloc,
605 		BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
606 	safe_dma_sync(&sc->sc_spalloc, BUS_DMASYNC_PREWRITE);
607 	safe_dma_sync(&sc->sc_dpalloc, BUS_DMASYNC_PREWRITE);
608 
609 #ifdef SAFE_DEBUG
610 	if (safe_debug) {
611 		safe_dump_ringstate(sc, __func__);
612 		safe_dump_request(sc, __func__, re);
613 	}
614 #endif
615 	sc->sc_nqchip++;
616 	if (sc->sc_nqchip > safestats.st_maxqchip)
617 		safestats.st_maxqchip = sc->sc_nqchip;
618 	/* poke h/w to check descriptor ring, any value can be written */
619 	WRITE_REG(sc, SAFE_HI_RD_DESCR, 0);
620 }
621 
622 #define	N(a)	(sizeof(a) / sizeof (a[0]))
623 static void
624 safe_setup_enckey(struct safe_session *ses, const void *key)
625 {
626 	int i;
627 
628 	bcopy(key, ses->ses_key, ses->ses_klen);
629 
630 	/* PE is little-endian, insure proper byte order */
631 	for (i = 0; i < N(ses->ses_key); i++)
632 		ses->ses_key[i] = htole32(ses->ses_key[i]);
633 }
634 
635 static void
636 safe_setup_mackey(struct safe_session *ses, int algo, const uint8_t *key,
637     int klen)
638 {
639 	SHA1_CTX sha1ctx;
640 	int i;
641 
642 	hmac_init_ipad(&auth_hash_hmac_sha1, key, klen, &sha1ctx);
643 	bcopy(sha1ctx.h.b32, ses->ses_hminner, sizeof(sha1ctx.h.b32));
644 
645 	hmac_init_opad(&auth_hash_hmac_sha1, key, klen, &sha1ctx);
646 	bcopy(sha1ctx.h.b32, ses->ses_hmouter, sizeof(sha1ctx.h.b32));
647 
648 	explicit_bzero(&sha1ctx, sizeof(sha1ctx));
649 
650 	/* PE is little-endian, insure proper byte order */
651 	for (i = 0; i < N(ses->ses_hminner); i++) {
652 		ses->ses_hminner[i] = htole32(ses->ses_hminner[i]);
653 		ses->ses_hmouter[i] = htole32(ses->ses_hmouter[i]);
654 	}
655 }
656 #undef N
657 
658 static bool
659 safe_auth_supported(struct safe_softc *sc,
660     const struct crypto_session_params *csp)
661 {
662 
663 	switch (csp->csp_auth_alg) {
664 	case CRYPTO_SHA1_HMAC:
665 		if ((sc->sc_devinfo & SAFE_DEVINFO_SHA1) == 0)
666 			return (false);
667 		break;
668 	default:
669 		return (false);
670 	}
671 	return (true);
672 }
673 
674 static bool
675 safe_cipher_supported(struct safe_softc *sc,
676     const struct crypto_session_params *csp)
677 {
678 
679 	switch (csp->csp_cipher_alg) {
680 	case CRYPTO_AES_CBC:
681 		if ((sc->sc_devinfo & SAFE_DEVINFO_AES) == 0)
682 			return (false);
683 		if (csp->csp_ivlen != 16)
684 			return (false);
685 		if (csp->csp_cipher_klen != 16 &&
686 		    csp->csp_cipher_klen != 24 &&
687 		    csp->csp_cipher_klen != 32)
688 			return (false);
689 		break;
690 	}
691 	return (true);
692 }
693 
694 static int
695 safe_probesession(device_t dev, const struct crypto_session_params *csp)
696 {
697 	struct safe_softc *sc = device_get_softc(dev);
698 
699 	if (csp->csp_flags != 0)
700 		return (EINVAL);
701 	switch (csp->csp_mode) {
702 	case CSP_MODE_DIGEST:
703 		if (!safe_auth_supported(sc, csp))
704 			return (EINVAL);
705 		break;
706 	case CSP_MODE_CIPHER:
707 		if (!safe_cipher_supported(sc, csp))
708 			return (EINVAL);
709 		break;
710 	case CSP_MODE_ETA:
711 		if (!safe_auth_supported(sc, csp) ||
712 		    !safe_cipher_supported(sc, csp))
713 			return (EINVAL);
714 		break;
715 	default:
716 		return (EINVAL);
717 	}
718 
719 	return (CRYPTODEV_PROBE_HARDWARE);
720 }
721 
722 /*
723  * Allocate a new 'session'.
724  */
725 static int
726 safe_newsession(device_t dev, crypto_session_t cses,
727     const struct crypto_session_params *csp)
728 {
729 	struct safe_session *ses;
730 
731 	ses = crypto_get_driver_session(cses);
732 	if (csp->csp_cipher_alg != 0) {
733 		ses->ses_klen = csp->csp_cipher_klen;
734 		if (csp->csp_cipher_key != NULL)
735 			safe_setup_enckey(ses, csp->csp_cipher_key);
736 	}
737 
738 	if (csp->csp_auth_alg != 0) {
739 		ses->ses_mlen = csp->csp_auth_mlen;
740 		if (ses->ses_mlen == 0) {
741 			ses->ses_mlen = SHA1_HASH_LEN;
742 		}
743 
744 		if (csp->csp_auth_key != NULL) {
745 			safe_setup_mackey(ses, csp->csp_auth_alg,
746 			    csp->csp_auth_key, csp->csp_auth_klen);
747 		}
748 	}
749 
750 	return (0);
751 }
752 
753 static void
754 safe_op_cb(void *arg, bus_dma_segment_t *seg, int nsegs, int error)
755 {
756 	struct safe_operand *op = arg;
757 
758 	DPRINTF(("%s: nsegs %d error %d\n", __func__,
759 		nsegs, error));
760 	if (error != 0)
761 		return;
762 	op->nsegs = nsegs;
763 	bcopy(seg, op->segs, nsegs * sizeof (seg[0]));
764 }
765 
766 static int
767 safe_process(device_t dev, struct cryptop *crp, int hint)
768 {
769 	struct safe_softc *sc = device_get_softc(dev);
770 	const struct crypto_session_params *csp;
771 	int err = 0, i, nicealign, uniform;
772 	int bypass, oplen;
773 	int16_t coffset;
774 	struct safe_session *ses;
775 	struct safe_ringentry *re;
776 	struct safe_sarec *sa;
777 	struct safe_pdesc *pd;
778 	u_int32_t cmd0, cmd1, staterec;
779 
780 	mtx_lock(&sc->sc_ringmtx);
781 	if (sc->sc_front == sc->sc_back && sc->sc_nqchip != 0) {
782 		safestats.st_ringfull++;
783 		sc->sc_needwakeup |= CRYPTO_SYMQ;
784 		mtx_unlock(&sc->sc_ringmtx);
785 		return (ERESTART);
786 	}
787 	re = sc->sc_front;
788 
789 	staterec = re->re_sa.sa_staterec;	/* save */
790 	/* NB: zero everything but the PE descriptor */
791 	bzero(&re->re_sa, sizeof(struct safe_ringentry) - sizeof(re->re_desc));
792 	re->re_sa.sa_staterec = staterec;	/* restore */
793 
794 	re->re_crp = crp;
795 
796 	sa = &re->re_sa;
797 	ses = crypto_get_driver_session(crp->crp_session);
798 	csp = crypto_get_params(crp->crp_session);
799 
800 	cmd0 = SAFE_SA_CMD0_BASIC;		/* basic group operation */
801 	cmd1 = 0;
802 	switch (csp->csp_mode) {
803 	case CSP_MODE_DIGEST:
804 		cmd0 |= SAFE_SA_CMD0_OP_HASH;
805 		break;
806 	case CSP_MODE_CIPHER:
807 		cmd0 |= SAFE_SA_CMD0_OP_CRYPT;
808 		break;
809 	case CSP_MODE_ETA:
810 		cmd0 |= SAFE_SA_CMD0_OP_BOTH;
811 		break;
812 	}
813 
814 	if (csp->csp_cipher_alg != 0) {
815 		if (crp->crp_cipher_key != NULL)
816 			safe_setup_enckey(ses, crp->crp_cipher_key);
817 
818 		switch (csp->csp_cipher_alg) {
819 		case CRYPTO_AES_CBC:
820 			cmd0 |= SAFE_SA_CMD0_AES;
821 			cmd1 |= SAFE_SA_CMD1_CBC;
822 			if (ses->ses_klen * 8 == 128)
823 			     cmd1 |=  SAFE_SA_CMD1_AES128;
824 			else if (ses->ses_klen * 8 == 192)
825 			     cmd1 |=  SAFE_SA_CMD1_AES192;
826 			else
827 			     cmd1 |=  SAFE_SA_CMD1_AES256;
828 		}
829 
830 		/*
831 		 * Setup encrypt/decrypt state.  When using basic ops
832 		 * we can't use an inline IV because hash/crypt offset
833 		 * must be from the end of the IV to the start of the
834 		 * crypt data and this leaves out the preceding header
835 		 * from the hash calculation.  Instead we place the IV
836 		 * in the state record and set the hash/crypt offset to
837 		 * copy both the header+IV.
838 		 */
839 		crypto_read_iv(crp, re->re_sastate.sa_saved_iv);
840 		cmd0 |= SAFE_SA_CMD0_IVLD_STATE;
841 
842 		if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op)) {
843 			cmd0 |= SAFE_SA_CMD0_OUTBOUND;
844 
845 			/*
846 			 * XXX: I suspect we don't need this since we
847 			 * don't save the returned IV.
848 			 */
849 			cmd0 |= SAFE_SA_CMD0_SAVEIV;
850 		} else {
851 			cmd0 |= SAFE_SA_CMD0_INBOUND;
852 		}
853 		/*
854 		 * For basic encryption use the zero pad algorithm.
855 		 * This pads results to an 8-byte boundary and
856 		 * suppresses padding verification for inbound (i.e.
857 		 * decrypt) operations.
858 		 *
859 		 * NB: Not sure if the 8-byte pad boundary is a problem.
860 		 */
861 		cmd0 |= SAFE_SA_CMD0_PAD_ZERO;
862 
863 		/* XXX assert key bufs have the same size */
864 		bcopy(ses->ses_key, sa->sa_key, sizeof(sa->sa_key));
865 	}
866 
867 	if (csp->csp_auth_alg != 0) {
868 		if (crp->crp_auth_key != NULL) {
869 			safe_setup_mackey(ses, csp->csp_auth_alg,
870 			    crp->crp_auth_key, csp->csp_auth_klen);
871 		}
872 
873 		switch (csp->csp_auth_alg) {
874 		case CRYPTO_SHA1_HMAC:
875 			cmd0 |= SAFE_SA_CMD0_SHA1;
876 			cmd1 |= SAFE_SA_CMD1_HMAC;	/* NB: enable HMAC */
877 			break;
878 		}
879 
880 		/*
881 		 * Digest data is loaded from the SA and the hash
882 		 * result is saved to the state block where we
883 		 * retrieve it for return to the caller.
884 		 */
885 		/* XXX assert digest bufs have the same size */
886 		bcopy(ses->ses_hminner, sa->sa_indigest,
887 			sizeof(sa->sa_indigest));
888 		bcopy(ses->ses_hmouter, sa->sa_outdigest,
889 			sizeof(sa->sa_outdigest));
890 
891 		cmd0 |= SAFE_SA_CMD0_HSLD_SA | SAFE_SA_CMD0_SAVEHASH;
892 		re->re_flags |= SAFE_QFLAGS_COPYOUTICV;
893 	}
894 
895 	if (csp->csp_mode == CSP_MODE_ETA) {
896 		/*
897 		 * The driver only supports ETA requests where there
898 		 * is no gap between the AAD and payload.
899 		 */
900 		if (crp->crp_aad_length != 0 &&
901 		    crp->crp_aad_start + crp->crp_aad_length !=
902 		    crp->crp_payload_start) {
903 			safestats.st_lenmismatch++;
904 			err = EINVAL;
905 			goto errout;
906 		}
907 		if (crp->crp_aad_length != 0)
908 			bypass = crp->crp_aad_start;
909 		else
910 			bypass = crp->crp_payload_start;
911 		coffset = crp->crp_aad_length;
912 		oplen = crp->crp_payload_start + crp->crp_payload_length;
913 #ifdef SAFE_DEBUG
914 		if (safe_debug) {
915 			printf("AAD: skip %d, len %d, digest %d\n",
916 			    crp->crp_aad_start, crp->crp_aad_length,
917 			    crp->crp_digest_start);
918 			printf("payload: skip %d, len %d, IV %d\n",
919 			    crp->crp_payload_start, crp->crp_payload_length,
920 			    crp->crp_iv_start);
921 			printf("bypass %d coffset %d oplen %d\n",
922 				bypass, coffset, oplen);
923 		}
924 #endif
925 		if (coffset & 3) {	/* offset must be 32-bit aligned */
926 			DPRINTF(("%s: coffset %u misaligned\n",
927 				__func__, coffset));
928 			safestats.st_coffmisaligned++;
929 			err = EINVAL;
930 			goto errout;
931 		}
932 		coffset >>= 2;
933 		if (coffset > 255) {	/* offset must be <256 dwords */
934 			DPRINTF(("%s: coffset %u too big\n",
935 				__func__, coffset));
936 			safestats.st_cofftoobig++;
937 			err = EINVAL;
938 			goto errout;
939 		}
940 		/*
941 		 * Tell the hardware to copy the header to the output.
942 		 * The header is defined as the data from the end of
943 		 * the bypass to the start of data to be encrypted.
944 		 * Typically this is the inline IV.  Note that you need
945 		 * to do this even if src+dst are the same; it appears
946 		 * that w/o this bit the crypted data is written
947 		 * immediately after the bypass data.
948 		 */
949 		cmd1 |= SAFE_SA_CMD1_HDRCOPY;
950 		/*
951 		 * Disable IP header mutable bit handling.  This is
952 		 * needed to get correct HMAC calculations.
953 		 */
954 		cmd1 |= SAFE_SA_CMD1_MUTABLE;
955 	} else {
956 		bypass = crp->crp_payload_start;
957 		oplen = bypass + crp->crp_payload_length;
958 		coffset = 0;
959 	}
960 	/* XXX verify multiple of 4 when using s/g */
961 	if (bypass > 96) {		/* bypass offset must be <= 96 bytes */
962 		DPRINTF(("%s: bypass %u too big\n", __func__, bypass));
963 		safestats.st_bypasstoobig++;
964 		err = EINVAL;
965 		goto errout;
966 	}
967 
968 	if (bus_dmamap_create(sc->sc_srcdmat, BUS_DMA_NOWAIT, &re->re_src_map)) {
969 		safestats.st_nomap++;
970 		err = ENOMEM;
971 		goto errout;
972 	}
973 	if (bus_dmamap_load_crp(sc->sc_srcdmat, re->re_src_map, crp, safe_op_cb,
974 	    &re->re_src, BUS_DMA_NOWAIT) != 0) {
975 		bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map);
976 		re->re_src_map = NULL;
977 		safestats.st_noload++;
978 		err = ENOMEM;
979 		goto errout;
980 	}
981 	re->re_src_mapsize = crypto_buffer_len(&crp->crp_buf);
982 	nicealign = safe_dmamap_aligned(&re->re_src);
983 	uniform = safe_dmamap_uniform(&re->re_src);
984 
985 	DPRINTF(("src nicealign %u uniform %u nsegs %u\n",
986 		nicealign, uniform, re->re_src.nsegs));
987 	if (re->re_src.nsegs > 1) {
988 		re->re_desc.d_src = sc->sc_spalloc.dma_paddr +
989 			((caddr_t) sc->sc_spfree - (caddr_t) sc->sc_spring);
990 		for (i = 0; i < re->re_src_nsegs; i++) {
991 			/* NB: no need to check if there's space */
992 			pd = sc->sc_spfree;
993 			if (++(sc->sc_spfree) == sc->sc_springtop)
994 				sc->sc_spfree = sc->sc_spring;
995 
996 			KASSERT((pd->pd_flags&3) == 0 ||
997 				(pd->pd_flags&3) == SAFE_PD_DONE,
998 				("bogus source particle descriptor; flags %x",
999 				pd->pd_flags));
1000 			pd->pd_addr = re->re_src_segs[i].ds_addr;
1001 			pd->pd_size = re->re_src_segs[i].ds_len;
1002 			pd->pd_flags = SAFE_PD_READY;
1003 		}
1004 		cmd0 |= SAFE_SA_CMD0_IGATHER;
1005 	} else {
1006 		/*
1007 		 * No need for gather, reference the operand directly.
1008 		 */
1009 		re->re_desc.d_src = re->re_src_segs[0].ds_addr;
1010 	}
1011 
1012 	if (csp->csp_mode == CSP_MODE_DIGEST) {
1013 		/*
1014 		 * Hash op; no destination needed.
1015 		 */
1016 	} else {
1017 		if (nicealign && uniform == 1) {
1018 			/*
1019 			 * Source layout is suitable for direct
1020 			 * sharing of the DMA map and segment list.
1021 			 */
1022 			re->re_dst = re->re_src;
1023 		} else if (nicealign && uniform == 2) {
1024 			/*
1025 			 * The source is properly aligned but requires a
1026 			 * different particle list to handle DMA of the
1027 			 * result.  Create a new map and do the load to
1028 			 * create the segment list.  The particle
1029 			 * descriptor setup code below will handle the
1030 			 * rest.
1031 			 */
1032 			if (bus_dmamap_create(sc->sc_dstdmat, BUS_DMA_NOWAIT,
1033 			    &re->re_dst_map)) {
1034 				safestats.st_nomap++;
1035 				err = ENOMEM;
1036 				goto errout;
1037 			}
1038 			if (bus_dmamap_load_crp(sc->sc_dstdmat, re->re_dst_map,
1039 			    crp, safe_op_cb, &re->re_dst, BUS_DMA_NOWAIT) !=
1040 			    0) {
1041 				bus_dmamap_destroy(sc->sc_dstdmat,
1042 				    re->re_dst_map);
1043 				re->re_dst_map = NULL;
1044 				safestats.st_noload++;
1045 				err = ENOMEM;
1046 				goto errout;
1047 			}
1048 		} else if (crp->crp_buf.cb_type == CRYPTO_BUF_MBUF) {
1049 			int totlen, len;
1050 			struct mbuf *m, *top, **mp;
1051 
1052 			/*
1053 			 * DMA constraints require that we allocate a
1054 			 * new mbuf chain for the destination.  We
1055 			 * allocate an entire new set of mbufs of
1056 			 * optimal/required size and then tell the
1057 			 * hardware to copy any bits that are not
1058 			 * created as a byproduct of the operation.
1059 			 */
1060 			if (!nicealign)
1061 				safestats.st_unaligned++;
1062 			if (!uniform)
1063 				safestats.st_notuniform++;
1064 			totlen = re->re_src_mapsize;
1065 			if (crp->crp_buf.cb_mbuf->m_flags & M_PKTHDR) {
1066 				len = MHLEN;
1067 				MGETHDR(m, M_NOWAIT, MT_DATA);
1068 				if (m && !m_dup_pkthdr(m, crp->crp_buf.cb_mbuf,
1069 				    M_NOWAIT)) {
1070 					m_free(m);
1071 					m = NULL;
1072 				}
1073 			} else {
1074 				len = MLEN;
1075 				MGET(m, M_NOWAIT, MT_DATA);
1076 			}
1077 			if (m == NULL) {
1078 				safestats.st_nombuf++;
1079 				err = sc->sc_nqchip ? ERESTART : ENOMEM;
1080 				goto errout;
1081 			}
1082 			if (totlen >= MINCLSIZE) {
1083 				if (!(MCLGET(m, M_NOWAIT))) {
1084 					m_free(m);
1085 					safestats.st_nomcl++;
1086 					err = sc->sc_nqchip ?
1087 					    ERESTART : ENOMEM;
1088 					goto errout;
1089 				}
1090 				len = MCLBYTES;
1091 			}
1092 			m->m_len = len;
1093 			top = NULL;
1094 			mp = &top;
1095 
1096 			while (totlen > 0) {
1097 				if (top) {
1098 					MGET(m, M_NOWAIT, MT_DATA);
1099 					if (m == NULL) {
1100 						m_freem(top);
1101 						safestats.st_nombuf++;
1102 						err = sc->sc_nqchip ?
1103 						    ERESTART : ENOMEM;
1104 						goto errout;
1105 					}
1106 					len = MLEN;
1107 				}
1108 				if (top && totlen >= MINCLSIZE) {
1109 					if (!(MCLGET(m, M_NOWAIT))) {
1110 						*mp = m;
1111 						m_freem(top);
1112 						safestats.st_nomcl++;
1113 						err = sc->sc_nqchip ?
1114 						    ERESTART : ENOMEM;
1115 						goto errout;
1116 					}
1117 					len = MCLBYTES;
1118 				}
1119 				m->m_len = len = min(totlen, len);
1120 				totlen -= len;
1121 				*mp = m;
1122 				mp = &m->m_next;
1123 			}
1124 			re->re_dst_m = top;
1125 			if (bus_dmamap_create(sc->sc_dstdmat,
1126 			    BUS_DMA_NOWAIT, &re->re_dst_map) != 0) {
1127 				safestats.st_nomap++;
1128 				err = ENOMEM;
1129 				goto errout;
1130 			}
1131 			if (bus_dmamap_load_mbuf_sg(sc->sc_dstdmat,
1132 			    re->re_dst_map, top, re->re_dst_segs,
1133 			    &re->re_dst_nsegs, 0) != 0) {
1134 				bus_dmamap_destroy(sc->sc_dstdmat,
1135 				    re->re_dst_map);
1136 				re->re_dst_map = NULL;
1137 				safestats.st_noload++;
1138 				err = ENOMEM;
1139 				goto errout;
1140 			}
1141 			re->re_dst_mapsize = re->re_src_mapsize;
1142 			if (re->re_src.mapsize > oplen) {
1143 				/*
1144 				 * There's data following what the
1145 				 * hardware will copy for us.  If this
1146 				 * isn't just the ICV (that's going to
1147 				 * be written on completion), copy it
1148 				 * to the new mbufs
1149 				 */
1150 				if (!(csp->csp_mode == CSP_MODE_ETA &&
1151 				    (re->re_src.mapsize-oplen) == ses->ses_mlen &&
1152 				    crp->crp_digest_start == oplen))
1153 					safe_mcopy(crp->crp_buf.cb_mbuf,
1154 					    re->re_dst_m, oplen);
1155 				else
1156 					safestats.st_noicvcopy++;
1157 			}
1158 		} else {
1159 			if (!nicealign) {
1160 				safestats.st_iovmisaligned++;
1161 				err = EINVAL;
1162 				goto errout;
1163 			} else {
1164 				/*
1165 				 * There's no way to handle the DMA
1166 				 * requirements with this uio.  We
1167 				 * could create a separate DMA area for
1168 				 * the result and then copy it back,
1169 				 * but for now we just bail and return
1170 				 * an error.  Note that uio requests
1171 				 * > SAFE_MAX_DSIZE are handled because
1172 				 * the DMA map and segment list for the
1173 				 * destination wil result in a
1174 				 * destination particle list that does
1175 				 * the necessary scatter DMA.
1176 				 */
1177 				safestats.st_iovnotuniform++;
1178 				err = EINVAL;
1179 				goto errout;
1180 			}
1181 		}
1182 
1183 		if (re->re_dst.nsegs > 1) {
1184 			re->re_desc.d_dst = sc->sc_dpalloc.dma_paddr +
1185 			    ((caddr_t) sc->sc_dpfree - (caddr_t) sc->sc_dpring);
1186 			for (i = 0; i < re->re_dst_nsegs; i++) {
1187 				pd = sc->sc_dpfree;
1188 				KASSERT((pd->pd_flags&3) == 0 ||
1189 					(pd->pd_flags&3) == SAFE_PD_DONE,
1190 					("bogus dest particle descriptor; flags %x",
1191 						pd->pd_flags));
1192 				if (++(sc->sc_dpfree) == sc->sc_dpringtop)
1193 					sc->sc_dpfree = sc->sc_dpring;
1194 				pd->pd_addr = re->re_dst_segs[i].ds_addr;
1195 				pd->pd_flags = SAFE_PD_READY;
1196 			}
1197 			cmd0 |= SAFE_SA_CMD0_OSCATTER;
1198 		} else {
1199 			/*
1200 			 * No need for scatter, reference the operand directly.
1201 			 */
1202 			re->re_desc.d_dst = re->re_dst_segs[0].ds_addr;
1203 		}
1204 	}
1205 
1206 	/*
1207 	 * All done with setup; fillin the SA command words
1208 	 * and the packet engine descriptor.  The operation
1209 	 * is now ready for submission to the hardware.
1210 	 */
1211 	sa->sa_cmd0 = cmd0 | SAFE_SA_CMD0_IPCI | SAFE_SA_CMD0_OPCI;
1212 	sa->sa_cmd1 = cmd1
1213 		    | (coffset << SAFE_SA_CMD1_OFFSET_S)
1214 		    | SAFE_SA_CMD1_SAREV1	/* Rev 1 SA data structure */
1215 		    | SAFE_SA_CMD1_SRPCI
1216 		    ;
1217 	/*
1218 	 * NB: the order of writes is important here.  In case the
1219 	 * chip is scanning the ring because of an outstanding request
1220 	 * it might nab this one too.  In that case we need to make
1221 	 * sure the setup is complete before we write the length
1222 	 * field of the descriptor as it signals the descriptor is
1223 	 * ready for processing.
1224 	 */
1225 	re->re_desc.d_csr = SAFE_PE_CSR_READY | SAFE_PE_CSR_SAPCI;
1226 	if (csp->csp_auth_alg != 0)
1227 		re->re_desc.d_csr |= SAFE_PE_CSR_LOADSA | SAFE_PE_CSR_HASHFINAL;
1228 	re->re_desc.d_len = oplen
1229 			  | SAFE_PE_LEN_READY
1230 			  | (bypass << SAFE_PE_LEN_BYPASS_S)
1231 			  ;
1232 
1233 	safestats.st_ipackets++;
1234 	safestats.st_ibytes += oplen;
1235 
1236 	if (++(sc->sc_front) == sc->sc_ringtop)
1237 		sc->sc_front = sc->sc_ring;
1238 
1239 	/* XXX honor batching */
1240 	safe_feed(sc, re);
1241 	mtx_unlock(&sc->sc_ringmtx);
1242 	return (0);
1243 
1244 errout:
1245 	if (re->re_dst_m != NULL)
1246 		m_freem(re->re_dst_m);
1247 
1248 	if (re->re_dst_map != NULL && re->re_dst_map != re->re_src_map) {
1249 		bus_dmamap_unload(sc->sc_dstdmat, re->re_dst_map);
1250 		bus_dmamap_destroy(sc->sc_dstdmat, re->re_dst_map);
1251 	}
1252 	if (re->re_src_map != NULL) {
1253 		bus_dmamap_unload(sc->sc_srcdmat, re->re_src_map);
1254 		bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map);
1255 	}
1256 	mtx_unlock(&sc->sc_ringmtx);
1257 	if (err != ERESTART) {
1258 		crp->crp_etype = err;
1259 		crypto_done(crp);
1260 		err = 0;
1261 	} else {
1262 		sc->sc_needwakeup |= CRYPTO_SYMQ;
1263 	}
1264 	return (err);
1265 }
1266 
1267 static void
1268 safe_callback(struct safe_softc *sc, struct safe_ringentry *re)
1269 {
1270 	const struct crypto_session_params *csp;
1271 	struct cryptop *crp = (struct cryptop *)re->re_crp;
1272 	struct safe_session *ses;
1273 	uint8_t hash[HASH_MAX_LEN];
1274 
1275 	ses = crypto_get_driver_session(crp->crp_session);
1276 	csp = crypto_get_params(crp->crp_session);
1277 
1278 	safestats.st_opackets++;
1279 	safestats.st_obytes += re->re_dst.mapsize;
1280 
1281 	safe_dma_sync(&sc->sc_ringalloc,
1282 		BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1283 	if (re->re_desc.d_csr & SAFE_PE_CSR_STATUS) {
1284 		device_printf(sc->sc_dev, "csr 0x%x cmd0 0x%x cmd1 0x%x\n",
1285 			re->re_desc.d_csr,
1286 			re->re_sa.sa_cmd0, re->re_sa.sa_cmd1);
1287 		safestats.st_peoperr++;
1288 		crp->crp_etype = EIO;		/* something more meaningful? */
1289 	}
1290 
1291 	/*
1292 	 * XXX: Should crp_buf.cb_mbuf be updated to re->re_dst_m if
1293 	 * it is non-NULL?
1294 	 */
1295 
1296 	if (re->re_dst_map != NULL && re->re_dst_map != re->re_src_map) {
1297 		bus_dmamap_sync(sc->sc_dstdmat, re->re_dst_map,
1298 		    BUS_DMASYNC_POSTREAD);
1299 		bus_dmamap_unload(sc->sc_dstdmat, re->re_dst_map);
1300 		bus_dmamap_destroy(sc->sc_dstdmat, re->re_dst_map);
1301 	}
1302 	bus_dmamap_sync(sc->sc_srcdmat, re->re_src_map, BUS_DMASYNC_POSTWRITE);
1303 	bus_dmamap_unload(sc->sc_srcdmat, re->re_src_map);
1304 	bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map);
1305 
1306 	if (re->re_flags & SAFE_QFLAGS_COPYOUTICV) {
1307 		if (csp->csp_auth_alg == CRYPTO_SHA1_HMAC) {
1308 			/*
1309 			 * SHA-1 ICV's are byte-swapped; fix 'em up
1310 			 * before copying them to their destination.
1311 			 */
1312 			re->re_sastate.sa_saved_indigest[0] =
1313 			    bswap32(re->re_sastate.sa_saved_indigest[0]);
1314 			re->re_sastate.sa_saved_indigest[1] =
1315 			    bswap32(re->re_sastate.sa_saved_indigest[1]);
1316 			re->re_sastate.sa_saved_indigest[2] =
1317 			    bswap32(re->re_sastate.sa_saved_indigest[2]);
1318 		}
1319 
1320 		if (crp->crp_op & CRYPTO_OP_VERIFY_DIGEST) {
1321 			crypto_copydata(crp, crp->crp_digest_start,
1322 			    ses->ses_mlen, hash);
1323 			if (timingsafe_bcmp(re->re_sastate.sa_saved_indigest,
1324 			    hash, ses->ses_mlen) != 0)
1325 				crp->crp_etype = EBADMSG;
1326 		} else
1327 			crypto_copyback(crp, crp->crp_digest_start,
1328 			    ses->ses_mlen, re->re_sastate.sa_saved_indigest);
1329 	}
1330 	crypto_done(crp);
1331 }
1332 
1333 /*
1334  * Copy all data past offset from srcm to dstm.
1335  */
1336 static void
1337 safe_mcopy(struct mbuf *srcm, struct mbuf *dstm, u_int offset)
1338 {
1339 	u_int j, dlen, slen;
1340 	caddr_t dptr, sptr;
1341 
1342 	/*
1343 	 * Advance src and dst to offset.
1344 	 */
1345 	j = offset;
1346 	while (j >= srcm->m_len) {
1347 		j -= srcm->m_len;
1348 		srcm = srcm->m_next;
1349 		if (srcm == NULL)
1350 			return;
1351 	}
1352 	sptr = mtod(srcm, caddr_t) + j;
1353 	slen = srcm->m_len - j;
1354 
1355 	j = offset;
1356 	while (j >= dstm->m_len) {
1357 		j -= dstm->m_len;
1358 		dstm = dstm->m_next;
1359 		if (dstm == NULL)
1360 			return;
1361 	}
1362 	dptr = mtod(dstm, caddr_t) + j;
1363 	dlen = dstm->m_len - j;
1364 
1365 	/*
1366 	 * Copy everything that remains.
1367 	 */
1368 	for (;;) {
1369 		j = min(slen, dlen);
1370 		bcopy(sptr, dptr, j);
1371 		if (slen == j) {
1372 			srcm = srcm->m_next;
1373 			if (srcm == NULL)
1374 				return;
1375 			sptr = srcm->m_data;
1376 			slen = srcm->m_len;
1377 		} else
1378 			sptr += j, slen -= j;
1379 		if (dlen == j) {
1380 			dstm = dstm->m_next;
1381 			if (dstm == NULL)
1382 				return;
1383 			dptr = dstm->m_data;
1384 			dlen = dstm->m_len;
1385 		} else
1386 			dptr += j, dlen -= j;
1387 	}
1388 }
1389 
1390 #ifndef SAFE_NO_RNG
1391 #define	SAFE_RNG_MAXWAIT	1000
1392 
1393 static void
1394 safe_rng_init(struct safe_softc *sc)
1395 {
1396 	u_int32_t w, v;
1397 	int i;
1398 
1399 	WRITE_REG(sc, SAFE_RNG_CTRL, 0);
1400 	/* use default value according to the manual */
1401 	WRITE_REG(sc, SAFE_RNG_CNFG, 0x834);	/* magic from SafeNet */
1402 	WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0);
1403 
1404 	/*
1405 	 * There is a bug in rev 1.0 of the 1140 that when the RNG
1406 	 * is brought out of reset the ready status flag does not
1407 	 * work until the RNG has finished its internal initialization.
1408 	 *
1409 	 * So in order to determine the device is through its
1410 	 * initialization we must read the data register, using the
1411 	 * status reg in the read in case it is initialized.  Then read
1412 	 * the data register until it changes from the first read.
1413 	 * Once it changes read the data register until it changes
1414 	 * again.  At this time the RNG is considered initialized.
1415 	 * This could take between 750ms - 1000ms in time.
1416 	 */
1417 	i = 0;
1418 	w = READ_REG(sc, SAFE_RNG_OUT);
1419 	do {
1420 		v = READ_REG(sc, SAFE_RNG_OUT);
1421 		if (v != w) {
1422 			w = v;
1423 			break;
1424 		}
1425 		DELAY(10);
1426 	} while (++i < SAFE_RNG_MAXWAIT);
1427 
1428 	/* Wait Until data changes again */
1429 	i = 0;
1430 	do {
1431 		v = READ_REG(sc, SAFE_RNG_OUT);
1432 		if (v != w)
1433 			break;
1434 		DELAY(10);
1435 	} while (++i < SAFE_RNG_MAXWAIT);
1436 }
1437 
1438 static __inline void
1439 safe_rng_disable_short_cycle(struct safe_softc *sc)
1440 {
1441 	WRITE_REG(sc, SAFE_RNG_CTRL,
1442 		READ_REG(sc, SAFE_RNG_CTRL) &~ SAFE_RNG_CTRL_SHORTEN);
1443 }
1444 
1445 static __inline void
1446 safe_rng_enable_short_cycle(struct safe_softc *sc)
1447 {
1448 	WRITE_REG(sc, SAFE_RNG_CTRL,
1449 		READ_REG(sc, SAFE_RNG_CTRL) | SAFE_RNG_CTRL_SHORTEN);
1450 }
1451 
1452 static __inline u_int32_t
1453 safe_rng_read(struct safe_softc *sc)
1454 {
1455 	int i;
1456 
1457 	i = 0;
1458 	while (READ_REG(sc, SAFE_RNG_STAT) != 0 && ++i < SAFE_RNG_MAXWAIT)
1459 		;
1460 	return READ_REG(sc, SAFE_RNG_OUT);
1461 }
1462 
1463 static void
1464 safe_rng(void *arg)
1465 {
1466 	struct safe_softc *sc = arg;
1467 	u_int32_t buf[SAFE_RNG_MAXBUFSIZ];	/* NB: maybe move to softc */
1468 	u_int maxwords;
1469 	int i;
1470 
1471 	safestats.st_rng++;
1472 	/*
1473 	 * Fetch the next block of data.
1474 	 */
1475 	maxwords = safe_rngbufsize;
1476 	if (maxwords > SAFE_RNG_MAXBUFSIZ)
1477 		maxwords = SAFE_RNG_MAXBUFSIZ;
1478 retry:
1479 	for (i = 0; i < maxwords; i++)
1480 		buf[i] = safe_rng_read(sc);
1481 	/*
1482 	 * Check the comparator alarm count and reset the h/w if
1483 	 * it exceeds our threshold.  This guards against the
1484 	 * hardware oscillators resonating with external signals.
1485 	 */
1486 	if (READ_REG(sc, SAFE_RNG_ALM_CNT) > safe_rngmaxalarm) {
1487 		u_int32_t freq_inc, w;
1488 
1489 		DPRINTF(("%s: alarm count %u exceeds threshold %u\n", __func__,
1490 			READ_REG(sc, SAFE_RNG_ALM_CNT), safe_rngmaxalarm));
1491 		safestats.st_rngalarm++;
1492 		safe_rng_enable_short_cycle(sc);
1493 		freq_inc = 18;
1494 		for (i = 0; i < 64; i++) {
1495 			w = READ_REG(sc, SAFE_RNG_CNFG);
1496 			freq_inc = ((w + freq_inc) & 0x3fL);
1497 			w = ((w & ~0x3fL) | freq_inc);
1498 			WRITE_REG(sc, SAFE_RNG_CNFG, w);
1499 
1500 			WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0);
1501 
1502 			(void) safe_rng_read(sc);
1503 			DELAY(25);
1504 
1505 			if (READ_REG(sc, SAFE_RNG_ALM_CNT) == 0) {
1506 				safe_rng_disable_short_cycle(sc);
1507 				goto retry;
1508 			}
1509 			freq_inc = 1;
1510 		}
1511 		safe_rng_disable_short_cycle(sc);
1512 	} else
1513 		WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0);
1514 
1515 	(*sc->sc_harvest)(sc->sc_rndtest, buf, maxwords*sizeof (u_int32_t));
1516 	callout_reset(&sc->sc_rngto,
1517 		hz * (safe_rnginterval ? safe_rnginterval : 1), safe_rng, sc);
1518 }
1519 #endif /* SAFE_NO_RNG */
1520 
1521 static void
1522 safe_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1523 {
1524 	bus_addr_t *paddr = (bus_addr_t*) arg;
1525 	*paddr = segs->ds_addr;
1526 }
1527 
1528 static int
1529 safe_dma_malloc(
1530 	struct safe_softc *sc,
1531 	bus_size_t size,
1532 	struct safe_dma_alloc *dma,
1533 	int mapflags
1534 )
1535 {
1536 	int r;
1537 
1538 	r = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev),	/* parent */
1539 			       sizeof(u_int32_t), 0,	/* alignment, bounds */
1540 			       BUS_SPACE_MAXADDR_32BIT,	/* lowaddr */
1541 			       BUS_SPACE_MAXADDR,	/* highaddr */
1542 			       NULL, NULL,		/* filter, filterarg */
1543 			       size,			/* maxsize */
1544 			       1,			/* nsegments */
1545 			       size,			/* maxsegsize */
1546 			       BUS_DMA_ALLOCNOW,	/* flags */
1547 			       NULL, NULL,		/* locking */
1548 			       &dma->dma_tag);
1549 	if (r != 0) {
1550 		device_printf(sc->sc_dev, "safe_dma_malloc: "
1551 			"bus_dma_tag_create failed; error %u\n", r);
1552 		goto fail_0;
1553 	}
1554 
1555 	r = bus_dmamem_alloc(dma->dma_tag, (void**) &dma->dma_vaddr,
1556 			     BUS_DMA_NOWAIT, &dma->dma_map);
1557 	if (r != 0) {
1558 		device_printf(sc->sc_dev, "safe_dma_malloc: "
1559 			"bus_dmammem_alloc failed; size %ju, error %u\n",
1560 			(uintmax_t)size, r);
1561 		goto fail_1;
1562 	}
1563 
1564 	r = bus_dmamap_load(dma->dma_tag, dma->dma_map, dma->dma_vaddr,
1565 		            size,
1566 			    safe_dmamap_cb,
1567 			    &dma->dma_paddr,
1568 			    mapflags | BUS_DMA_NOWAIT);
1569 	if (r != 0) {
1570 		device_printf(sc->sc_dev, "safe_dma_malloc: "
1571 			"bus_dmamap_load failed; error %u\n", r);
1572 		goto fail_2;
1573 	}
1574 
1575 	dma->dma_size = size;
1576 	return (0);
1577 
1578 	bus_dmamap_unload(dma->dma_tag, dma->dma_map);
1579 fail_2:
1580 	bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
1581 fail_1:
1582 	bus_dma_tag_destroy(dma->dma_tag);
1583 fail_0:
1584 	dma->dma_tag = NULL;
1585 	return (r);
1586 }
1587 
1588 static void
1589 safe_dma_free(struct safe_softc *sc, struct safe_dma_alloc *dma)
1590 {
1591 	bus_dmamap_unload(dma->dma_tag, dma->dma_map);
1592 	bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
1593 	bus_dma_tag_destroy(dma->dma_tag);
1594 }
1595 
1596 /*
1597  * Resets the board.  Values in the regesters are left as is
1598  * from the reset (i.e. initial values are assigned elsewhere).
1599  */
1600 static void
1601 safe_reset_board(struct safe_softc *sc)
1602 {
1603 	u_int32_t v;
1604 	/*
1605 	 * Reset the device.  The manual says no delay
1606 	 * is needed between marking and clearing reset.
1607 	 */
1608 	v = READ_REG(sc, SAFE_PE_DMACFG) &~
1609 		(SAFE_PE_DMACFG_PERESET | SAFE_PE_DMACFG_PDRRESET |
1610 		 SAFE_PE_DMACFG_SGRESET);
1611 	WRITE_REG(sc, SAFE_PE_DMACFG, v
1612 				    | SAFE_PE_DMACFG_PERESET
1613 				    | SAFE_PE_DMACFG_PDRRESET
1614 				    | SAFE_PE_DMACFG_SGRESET);
1615 	WRITE_REG(sc, SAFE_PE_DMACFG, v);
1616 }
1617 
1618 /*
1619  * Initialize registers we need to touch only once.
1620  */
1621 static void
1622 safe_init_board(struct safe_softc *sc)
1623 {
1624 	u_int32_t v, dwords;
1625 
1626 	v = READ_REG(sc, SAFE_PE_DMACFG);
1627 	v &=~ SAFE_PE_DMACFG_PEMODE;
1628 	v |= SAFE_PE_DMACFG_FSENA		/* failsafe enable */
1629 	  |  SAFE_PE_DMACFG_GPRPCI		/* gather ring on PCI */
1630 	  |  SAFE_PE_DMACFG_SPRPCI		/* scatter ring on PCI */
1631 	  |  SAFE_PE_DMACFG_ESDESC		/* endian-swap descriptors */
1632 	  |  SAFE_PE_DMACFG_ESSA		/* endian-swap SA's */
1633 	  |  SAFE_PE_DMACFG_ESPDESC		/* endian-swap part. desc's */
1634 	  ;
1635 	WRITE_REG(sc, SAFE_PE_DMACFG, v);
1636 #if 0
1637 	/* XXX select byte swap based on host byte order */
1638 	WRITE_REG(sc, SAFE_ENDIAN, 0x1b);
1639 #endif
1640 	if (sc->sc_chiprev == SAFE_REV(1,0)) {
1641 		/*
1642 		 * Avoid large PCI DMA transfers.  Rev 1.0 has a bug where
1643 		 * "target mode transfers" done while the chip is DMA'ing
1644 		 * >1020 bytes cause the hardware to lockup.  To avoid this
1645 		 * we reduce the max PCI transfer size and use small source
1646 		 * particle descriptors (<= 256 bytes).
1647 		 */
1648 		WRITE_REG(sc, SAFE_DMA_CFG, 256);
1649 		device_printf(sc->sc_dev,
1650 			"Reduce max DMA size to %u words for rev %u.%u WAR\n",
1651 			(READ_REG(sc, SAFE_DMA_CFG)>>2) & 0xff,
1652 			SAFE_REV_MAJ(sc->sc_chiprev),
1653 			SAFE_REV_MIN(sc->sc_chiprev));
1654 	}
1655 
1656 	/* NB: operands+results are overlaid */
1657 	WRITE_REG(sc, SAFE_PE_PDRBASE, sc->sc_ringalloc.dma_paddr);
1658 	WRITE_REG(sc, SAFE_PE_RDRBASE, sc->sc_ringalloc.dma_paddr);
1659 	/*
1660 	 * Configure ring entry size and number of items in the ring.
1661 	 */
1662 	KASSERT((sizeof(struct safe_ringentry) % sizeof(u_int32_t)) == 0,
1663 		("PE ring entry not 32-bit aligned!"));
1664 	dwords = sizeof(struct safe_ringentry) / sizeof(u_int32_t);
1665 	WRITE_REG(sc, SAFE_PE_RINGCFG,
1666 		(dwords << SAFE_PE_RINGCFG_OFFSET_S) | SAFE_MAX_NQUEUE);
1667 	WRITE_REG(sc, SAFE_PE_RINGPOLL, 0);	/* disable polling */
1668 
1669 	WRITE_REG(sc, SAFE_PE_GRNGBASE, sc->sc_spalloc.dma_paddr);
1670 	WRITE_REG(sc, SAFE_PE_SRNGBASE, sc->sc_dpalloc.dma_paddr);
1671 	WRITE_REG(sc, SAFE_PE_PARTSIZE,
1672 		(SAFE_TOTAL_DPART<<16) | SAFE_TOTAL_SPART);
1673 	/*
1674 	 * NB: destination particles are fixed size.  We use
1675 	 *     an mbuf cluster and require all results go to
1676 	 *     clusters or smaller.
1677 	 */
1678 	WRITE_REG(sc, SAFE_PE_PARTCFG, SAFE_MAX_DSIZE);
1679 
1680 	/* it's now safe to enable PE mode, do it */
1681 	WRITE_REG(sc, SAFE_PE_DMACFG, v | SAFE_PE_DMACFG_PEMODE);
1682 
1683 	/*
1684 	 * Configure hardware to use level-triggered interrupts and
1685 	 * to interrupt after each descriptor is processed.
1686 	 */
1687 	WRITE_REG(sc, SAFE_HI_CFG, SAFE_HI_CFG_LEVEL);
1688 	WRITE_REG(sc, SAFE_HI_DESC_CNT, 1);
1689 	WRITE_REG(sc, SAFE_HI_MASK, SAFE_INT_PE_DDONE | SAFE_INT_PE_ERROR);
1690 }
1691 
1692 /*
1693  * Init PCI registers
1694  */
1695 static void
1696 safe_init_pciregs(device_t dev)
1697 {
1698 }
1699 
1700 /*
1701  * Clean up after a chip crash.
1702  * It is assumed that the caller in splimp()
1703  */
1704 static void
1705 safe_cleanchip(struct safe_softc *sc)
1706 {
1707 
1708 	if (sc->sc_nqchip != 0) {
1709 		struct safe_ringentry *re = sc->sc_back;
1710 
1711 		while (re != sc->sc_front) {
1712 			if (re->re_desc.d_csr != 0)
1713 				safe_free_entry(sc, re);
1714 			if (++re == sc->sc_ringtop)
1715 				re = sc->sc_ring;
1716 		}
1717 		sc->sc_back = re;
1718 		sc->sc_nqchip = 0;
1719 	}
1720 }
1721 
1722 /*
1723  * free a safe_q
1724  * It is assumed that the caller is within splimp().
1725  */
1726 static int
1727 safe_free_entry(struct safe_softc *sc, struct safe_ringentry *re)
1728 {
1729 	struct cryptop *crp;
1730 
1731 	/*
1732 	 * Free header MCR
1733 	 */
1734 	if (re->re_dst_m != NULL)
1735 		m_freem(re->re_dst_m);
1736 
1737 	crp = (struct cryptop *)re->re_crp;
1738 
1739 	re->re_desc.d_csr = 0;
1740 
1741 	crp->crp_etype = EFAULT;
1742 	crypto_done(crp);
1743 	return(0);
1744 }
1745 
1746 /*
1747  * Routine to reset the chip and clean up.
1748  * It is assumed that the caller is in splimp()
1749  */
1750 static void
1751 safe_totalreset(struct safe_softc *sc)
1752 {
1753 	safe_reset_board(sc);
1754 	safe_init_board(sc);
1755 	safe_cleanchip(sc);
1756 }
1757 
1758 /*
1759  * Is the operand suitable aligned for direct DMA.  Each
1760  * segment must be aligned on a 32-bit boundary and all
1761  * but the last segment must be a multiple of 4 bytes.
1762  */
1763 static int
1764 safe_dmamap_aligned(const struct safe_operand *op)
1765 {
1766 	int i;
1767 
1768 	for (i = 0; i < op->nsegs; i++) {
1769 		if (op->segs[i].ds_addr & 3)
1770 			return (0);
1771 		if (i != (op->nsegs - 1) && (op->segs[i].ds_len & 3))
1772 			return (0);
1773 	}
1774 	return (1);
1775 }
1776 
1777 /*
1778  * Is the operand suitable for direct DMA as the destination
1779  * of an operation.  The hardware requires that each ``particle''
1780  * but the last in an operation result have the same size.  We
1781  * fix that size at SAFE_MAX_DSIZE bytes.  This routine returns
1782  * 0 if some segment is not a multiple of of this size, 1 if all
1783  * segments are exactly this size, or 2 if segments are at worst
1784  * a multiple of this size.
1785  */
1786 static int
1787 safe_dmamap_uniform(const struct safe_operand *op)
1788 {
1789 	int result = 1;
1790 
1791 	if (op->nsegs > 0) {
1792 		int i;
1793 
1794 		for (i = 0; i < op->nsegs-1; i++) {
1795 			if (op->segs[i].ds_len % SAFE_MAX_DSIZE)
1796 				return (0);
1797 			if (op->segs[i].ds_len != SAFE_MAX_DSIZE)
1798 				result = 2;
1799 		}
1800 	}
1801 	return (result);
1802 }
1803 
1804 #ifdef SAFE_DEBUG
1805 static void
1806 safe_dump_dmastatus(struct safe_softc *sc, const char *tag)
1807 {
1808 	printf("%s: ENDIAN 0x%x SRC 0x%x DST 0x%x STAT 0x%x\n"
1809 		, tag
1810 		, READ_REG(sc, SAFE_DMA_ENDIAN)
1811 		, READ_REG(sc, SAFE_DMA_SRCADDR)
1812 		, READ_REG(sc, SAFE_DMA_DSTADDR)
1813 		, READ_REG(sc, SAFE_DMA_STAT)
1814 	);
1815 }
1816 
1817 static void
1818 safe_dump_intrstate(struct safe_softc *sc, const char *tag)
1819 {
1820 	printf("%s: HI_CFG 0x%x HI_MASK 0x%x HI_DESC_CNT 0x%x HU_STAT 0x%x HM_STAT 0x%x\n"
1821 		, tag
1822 		, READ_REG(sc, SAFE_HI_CFG)
1823 		, READ_REG(sc, SAFE_HI_MASK)
1824 		, READ_REG(sc, SAFE_HI_DESC_CNT)
1825 		, READ_REG(sc, SAFE_HU_STAT)
1826 		, READ_REG(sc, SAFE_HM_STAT)
1827 	);
1828 }
1829 
1830 static void
1831 safe_dump_ringstate(struct safe_softc *sc, const char *tag)
1832 {
1833 	u_int32_t estat = READ_REG(sc, SAFE_PE_ERNGSTAT);
1834 
1835 	/* NB: assume caller has lock on ring */
1836 	printf("%s: ERNGSTAT %x (next %u) back %lu front %lu\n",
1837 		tag,
1838 		estat, (estat >> SAFE_PE_ERNGSTAT_NEXT_S),
1839 		(unsigned long)(sc->sc_back - sc->sc_ring),
1840 		(unsigned long)(sc->sc_front - sc->sc_ring));
1841 }
1842 
1843 static void
1844 safe_dump_request(struct safe_softc *sc, const char* tag, struct safe_ringentry *re)
1845 {
1846 	int ix, nsegs;
1847 
1848 	ix = re - sc->sc_ring;
1849 	printf("%s: %p (%u): csr %x src %x dst %x sa %x len %x\n"
1850 		, tag
1851 		, re, ix
1852 		, re->re_desc.d_csr
1853 		, re->re_desc.d_src
1854 		, re->re_desc.d_dst
1855 		, re->re_desc.d_sa
1856 		, re->re_desc.d_len
1857 	);
1858 	if (re->re_src.nsegs > 1) {
1859 		ix = (re->re_desc.d_src - sc->sc_spalloc.dma_paddr) /
1860 			sizeof(struct safe_pdesc);
1861 		for (nsegs = re->re_src.nsegs; nsegs; nsegs--) {
1862 			printf(" spd[%u] %p: %p size %u flags %x"
1863 				, ix, &sc->sc_spring[ix]
1864 				, (caddr_t)(uintptr_t) sc->sc_spring[ix].pd_addr
1865 				, sc->sc_spring[ix].pd_size
1866 				, sc->sc_spring[ix].pd_flags
1867 			);
1868 			if (sc->sc_spring[ix].pd_size == 0)
1869 				printf(" (zero!)");
1870 			printf("\n");
1871 			if (++ix == SAFE_TOTAL_SPART)
1872 				ix = 0;
1873 		}
1874 	}
1875 	if (re->re_dst.nsegs > 1) {
1876 		ix = (re->re_desc.d_dst - sc->sc_dpalloc.dma_paddr) /
1877 			sizeof(struct safe_pdesc);
1878 		for (nsegs = re->re_dst.nsegs; nsegs; nsegs--) {
1879 			printf(" dpd[%u] %p: %p flags %x\n"
1880 				, ix, &sc->sc_dpring[ix]
1881 				, (caddr_t)(uintptr_t) sc->sc_dpring[ix].pd_addr
1882 				, sc->sc_dpring[ix].pd_flags
1883 			);
1884 			if (++ix == SAFE_TOTAL_DPART)
1885 				ix = 0;
1886 		}
1887 	}
1888 	printf("sa: cmd0 %08x cmd1 %08x staterec %x\n",
1889 		re->re_sa.sa_cmd0, re->re_sa.sa_cmd1, re->re_sa.sa_staterec);
1890 	printf("sa: key %x %x %x %x %x %x %x %x\n"
1891 		, re->re_sa.sa_key[0]
1892 		, re->re_sa.sa_key[1]
1893 		, re->re_sa.sa_key[2]
1894 		, re->re_sa.sa_key[3]
1895 		, re->re_sa.sa_key[4]
1896 		, re->re_sa.sa_key[5]
1897 		, re->re_sa.sa_key[6]
1898 		, re->re_sa.sa_key[7]
1899 	);
1900 	printf("sa: indigest %x %x %x %x %x\n"
1901 		, re->re_sa.sa_indigest[0]
1902 		, re->re_sa.sa_indigest[1]
1903 		, re->re_sa.sa_indigest[2]
1904 		, re->re_sa.sa_indigest[3]
1905 		, re->re_sa.sa_indigest[4]
1906 	);
1907 	printf("sa: outdigest %x %x %x %x %x\n"
1908 		, re->re_sa.sa_outdigest[0]
1909 		, re->re_sa.sa_outdigest[1]
1910 		, re->re_sa.sa_outdigest[2]
1911 		, re->re_sa.sa_outdigest[3]
1912 		, re->re_sa.sa_outdigest[4]
1913 	);
1914 	printf("sr: iv %x %x %x %x\n"
1915 		, re->re_sastate.sa_saved_iv[0]
1916 		, re->re_sastate.sa_saved_iv[1]
1917 		, re->re_sastate.sa_saved_iv[2]
1918 		, re->re_sastate.sa_saved_iv[3]
1919 	);
1920 	printf("sr: hashbc %u indigest %x %x %x %x %x\n"
1921 		, re->re_sastate.sa_saved_hashbc
1922 		, re->re_sastate.sa_saved_indigest[0]
1923 		, re->re_sastate.sa_saved_indigest[1]
1924 		, re->re_sastate.sa_saved_indigest[2]
1925 		, re->re_sastate.sa_saved_indigest[3]
1926 		, re->re_sastate.sa_saved_indigest[4]
1927 	);
1928 }
1929 
1930 static void
1931 safe_dump_ring(struct safe_softc *sc, const char *tag)
1932 {
1933 	mtx_lock(&sc->sc_ringmtx);
1934 	printf("\nSafeNet Ring State:\n");
1935 	safe_dump_intrstate(sc, tag);
1936 	safe_dump_dmastatus(sc, tag);
1937 	safe_dump_ringstate(sc, tag);
1938 	if (sc->sc_nqchip) {
1939 		struct safe_ringentry *re = sc->sc_back;
1940 		do {
1941 			safe_dump_request(sc, tag, re);
1942 			if (++re == sc->sc_ringtop)
1943 				re = sc->sc_ring;
1944 		} while (re != sc->sc_front);
1945 	}
1946 	mtx_unlock(&sc->sc_ringmtx);
1947 }
1948 
1949 static int
1950 sysctl_hw_safe_dump(SYSCTL_HANDLER_ARGS)
1951 {
1952 	char dmode[64];
1953 	int error;
1954 
1955 	strncpy(dmode, "", sizeof(dmode) - 1);
1956 	dmode[sizeof(dmode) - 1] = '\0';
1957 	error = sysctl_handle_string(oidp, &dmode[0], sizeof(dmode), req);
1958 
1959 	if (error == 0 && req->newptr != NULL) {
1960 		struct safe_softc *sc = safec;
1961 
1962 		if (!sc)
1963 			return EINVAL;
1964 		if (strncmp(dmode, "dma", 3) == 0)
1965 			safe_dump_dmastatus(sc, "safe0");
1966 		else if (strncmp(dmode, "int", 3) == 0)
1967 			safe_dump_intrstate(sc, "safe0");
1968 		else if (strncmp(dmode, "ring", 4) == 0)
1969 			safe_dump_ring(sc, "safe0");
1970 		else
1971 			return EINVAL;
1972 	}
1973 	return error;
1974 }
1975 SYSCTL_PROC(_hw_safe, OID_AUTO, dump,
1976     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, 0,
1977     sysctl_hw_safe_dump, "A",
1978     "Dump driver state");
1979 #endif /* SAFE_DEBUG */
1980