xref: /freebsd/sys/dev/safe/safe.c (revision 7cc42f6d25ef2e19059d088fa7d4853fe9afefb5)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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 static driver_t safe_driver = {
112 	"safe",
113 	safe_methods,
114 	sizeof (struct safe_softc)
115 };
116 static devclass_t safe_devclass;
117 
118 DRIVER_MODULE(safe, pci, safe_driver, safe_devclass, 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 		crypto_kregister(sc->sc_cid, CRK_MOD_EXP, 0);
389 		crypto_kregister(sc->sc_cid, CRK_MOD_EXP_CRT, 0);
390 #endif
391 	}
392 	if (sc->sc_devinfo & SAFE_DEVINFO_DES) {
393 		printf(" des/3des");
394 	}
395 	if (sc->sc_devinfo & SAFE_DEVINFO_AES) {
396 		printf(" aes");
397 	}
398 	if (sc->sc_devinfo & SAFE_DEVINFO_MD5) {
399 		printf(" md5");
400 	}
401 	if (sc->sc_devinfo & SAFE_DEVINFO_SHA1) {
402 		printf(" sha1");
403 	}
404 	/* XXX other supported algorithms */
405 	printf("\n");
406 
407 	safe_reset_board(sc);		/* reset h/w */
408 	safe_init_pciregs(dev);		/* init pci settings */
409 	safe_init_board(sc);		/* init h/w */
410 
411 #ifndef SAFE_NO_RNG
412 	if (sc->sc_flags & SAFE_FLAGS_RNG) {
413 #ifdef SAFE_RNDTEST
414 		sc->sc_rndtest = rndtest_attach(dev);
415 		if (sc->sc_rndtest)
416 			sc->sc_harvest = rndtest_harvest;
417 		else
418 			sc->sc_harvest = default_harvest;
419 #else
420 		sc->sc_harvest = default_harvest;
421 #endif
422 		safe_rng_init(sc);
423 
424 		callout_init(&sc->sc_rngto, 1);
425 		callout_reset(&sc->sc_rngto, hz*safe_rnginterval, safe_rng, sc);
426 	}
427 #endif /* SAFE_NO_RNG */
428 #ifdef SAFE_DEBUG
429 	safec = sc;			/* for use by hw.safe.dump */
430 #endif
431 	return (0);
432 bad4:
433 	crypto_unregister_all(sc->sc_cid);
434 bad3:
435 	bus_teardown_intr(dev, sc->sc_irq, sc->sc_ih);
436 bad2:
437 	bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq);
438 bad1:
439 	bus_release_resource(dev, SYS_RES_MEMORY, BS_BAR, sc->sc_sr);
440 bad:
441 	return (ENXIO);
442 }
443 
444 /*
445  * Detach a device that successfully probed.
446  */
447 static int
448 safe_detach(device_t dev)
449 {
450 	struct safe_softc *sc = device_get_softc(dev);
451 
452 	/* XXX wait/abort active ops */
453 
454 	WRITE_REG(sc, SAFE_HI_MASK, 0);		/* disable interrupts */
455 
456 	callout_stop(&sc->sc_rngto);
457 
458 	crypto_unregister_all(sc->sc_cid);
459 
460 #ifdef SAFE_RNDTEST
461 	if (sc->sc_rndtest)
462 		rndtest_detach(sc->sc_rndtest);
463 #endif
464 
465 	safe_cleanchip(sc);
466 	safe_dma_free(sc, &sc->sc_dpalloc);
467 	safe_dma_free(sc, &sc->sc_spalloc);
468 	mtx_destroy(&sc->sc_ringmtx);
469 	safe_dma_free(sc, &sc->sc_ringalloc);
470 
471 	bus_generic_detach(dev);
472 	bus_teardown_intr(dev, sc->sc_irq, sc->sc_ih);
473 	bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq);
474 
475 	bus_dma_tag_destroy(sc->sc_srcdmat);
476 	bus_dma_tag_destroy(sc->sc_dstdmat);
477 	bus_release_resource(dev, SYS_RES_MEMORY, BS_BAR, sc->sc_sr);
478 
479 	return (0);
480 }
481 
482 /*
483  * Stop all chip i/o so that the kernel's probe routines don't
484  * get confused by errant DMAs when rebooting.
485  */
486 static int
487 safe_shutdown(device_t dev)
488 {
489 #ifdef notyet
490 	safe_stop(device_get_softc(dev));
491 #endif
492 	return (0);
493 }
494 
495 /*
496  * Device suspend routine.
497  */
498 static int
499 safe_suspend(device_t dev)
500 {
501 	struct safe_softc *sc = device_get_softc(dev);
502 
503 #ifdef notyet
504 	/* XXX stop the device and save PCI settings */
505 #endif
506 	sc->sc_suspended = 1;
507 
508 	return (0);
509 }
510 
511 static int
512 safe_resume(device_t dev)
513 {
514 	struct safe_softc *sc = device_get_softc(dev);
515 
516 #ifdef notyet
517 	/* XXX retore PCI settings and start the device */
518 #endif
519 	sc->sc_suspended = 0;
520 	return (0);
521 }
522 
523 /*
524  * SafeXcel Interrupt routine
525  */
526 static void
527 safe_intr(void *arg)
528 {
529 	struct safe_softc *sc = arg;
530 	volatile u_int32_t stat;
531 
532 	stat = READ_REG(sc, SAFE_HM_STAT);
533 	if (stat == 0)			/* shared irq, not for us */
534 		return;
535 
536 	WRITE_REG(sc, SAFE_HI_CLR, stat);	/* IACK */
537 
538 	if ((stat & SAFE_INT_PE_DDONE)) {
539 		/*
540 		 * Descriptor(s) done; scan the ring and
541 		 * process completed operations.
542 		 */
543 		mtx_lock(&sc->sc_ringmtx);
544 		while (sc->sc_back != sc->sc_front) {
545 			struct safe_ringentry *re = sc->sc_back;
546 #ifdef SAFE_DEBUG
547 			if (safe_debug) {
548 				safe_dump_ringstate(sc, __func__);
549 				safe_dump_request(sc, __func__, re);
550 			}
551 #endif
552 			/*
553 			 * safe_process marks ring entries that were allocated
554 			 * but not used with a csr of zero.  This insures the
555 			 * ring front pointer never needs to be set backwards
556 			 * in the event that an entry is allocated but not used
557 			 * because of a setup error.
558 			 */
559 			if (re->re_desc.d_csr != 0) {
560 				if (!SAFE_PE_CSR_IS_DONE(re->re_desc.d_csr))
561 					break;
562 				if (!SAFE_PE_LEN_IS_DONE(re->re_desc.d_len))
563 					break;
564 				sc->sc_nqchip--;
565 				safe_callback(sc, re);
566 			}
567 			if (++(sc->sc_back) == sc->sc_ringtop)
568 				sc->sc_back = sc->sc_ring;
569 		}
570 		mtx_unlock(&sc->sc_ringmtx);
571 	}
572 
573 	/*
574 	 * Check to see if we got any DMA Error
575 	 */
576 	if (stat & SAFE_INT_PE_ERROR) {
577 		DPRINTF(("dmaerr dmastat %08x\n",
578 			READ_REG(sc, SAFE_PE_DMASTAT)));
579 		safestats.st_dmaerr++;
580 		safe_totalreset(sc);
581 #if 0
582 		safe_feed(sc);
583 #endif
584 	}
585 
586 	if (sc->sc_needwakeup) {		/* XXX check high watermark */
587 		int wakeup = sc->sc_needwakeup & (CRYPTO_SYMQ|CRYPTO_ASYMQ);
588 		DPRINTF(("%s: wakeup crypto %x\n", __func__,
589 			sc->sc_needwakeup));
590 		sc->sc_needwakeup &= ~wakeup;
591 		crypto_unblock(sc->sc_cid, wakeup);
592 	}
593 }
594 
595 /*
596  * safe_feed() - post a request to chip
597  */
598 static void
599 safe_feed(struct safe_softc *sc, struct safe_ringentry *re)
600 {
601 	bus_dmamap_sync(sc->sc_srcdmat, re->re_src_map, BUS_DMASYNC_PREWRITE);
602 	if (re->re_dst_map != NULL)
603 		bus_dmamap_sync(sc->sc_dstdmat, re->re_dst_map,
604 			BUS_DMASYNC_PREREAD);
605 	/* XXX have no smaller granularity */
606 	safe_dma_sync(&sc->sc_ringalloc,
607 		BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
608 	safe_dma_sync(&sc->sc_spalloc, BUS_DMASYNC_PREWRITE);
609 	safe_dma_sync(&sc->sc_dpalloc, BUS_DMASYNC_PREWRITE);
610 
611 #ifdef SAFE_DEBUG
612 	if (safe_debug) {
613 		safe_dump_ringstate(sc, __func__);
614 		safe_dump_request(sc, __func__, re);
615 	}
616 #endif
617 	sc->sc_nqchip++;
618 	if (sc->sc_nqchip > safestats.st_maxqchip)
619 		safestats.st_maxqchip = sc->sc_nqchip;
620 	/* poke h/w to check descriptor ring, any value can be written */
621 	WRITE_REG(sc, SAFE_HI_RD_DESCR, 0);
622 }
623 
624 #define	N(a)	(sizeof(a) / sizeof (a[0]))
625 static void
626 safe_setup_enckey(struct safe_session *ses, const void *key)
627 {
628 	int i;
629 
630 	bcopy(key, ses->ses_key, ses->ses_klen);
631 
632 	/* PE is little-endian, insure proper byte order */
633 	for (i = 0; i < N(ses->ses_key); i++)
634 		ses->ses_key[i] = htole32(ses->ses_key[i]);
635 }
636 
637 static void
638 safe_setup_mackey(struct safe_session *ses, int algo, const uint8_t *key,
639     int klen)
640 {
641 	SHA1_CTX sha1ctx;
642 	int i;
643 
644 	hmac_init_ipad(&auth_hash_hmac_sha1, key, klen, &sha1ctx);
645 	bcopy(sha1ctx.h.b32, ses->ses_hminner, sizeof(sha1ctx.h.b32));
646 
647 	hmac_init_opad(&auth_hash_hmac_sha1, key, klen, &sha1ctx);
648 	bcopy(sha1ctx.h.b32, ses->ses_hmouter, sizeof(sha1ctx.h.b32));
649 
650 	explicit_bzero(&sha1ctx, sizeof(sha1ctx));
651 
652 	/* PE is little-endian, insure proper byte order */
653 	for (i = 0; i < N(ses->ses_hminner); i++) {
654 		ses->ses_hminner[i] = htole32(ses->ses_hminner[i]);
655 		ses->ses_hmouter[i] = htole32(ses->ses_hmouter[i]);
656 	}
657 }
658 #undef N
659 
660 static bool
661 safe_auth_supported(struct safe_softc *sc,
662     const struct crypto_session_params *csp)
663 {
664 
665 	switch (csp->csp_auth_alg) {
666 	case CRYPTO_SHA1_HMAC:
667 		if ((sc->sc_devinfo & SAFE_DEVINFO_SHA1) == 0)
668 			return (false);
669 		break;
670 	default:
671 		return (false);
672 	}
673 	return (true);
674 }
675 
676 static bool
677 safe_cipher_supported(struct safe_softc *sc,
678     const struct crypto_session_params *csp)
679 {
680 
681 	switch (csp->csp_cipher_alg) {
682 	case CRYPTO_AES_CBC:
683 		if ((sc->sc_devinfo & SAFE_DEVINFO_AES) == 0)
684 			return (false);
685 		if (csp->csp_ivlen != 16)
686 			return (false);
687 		if (csp->csp_cipher_klen != 16 &&
688 		    csp->csp_cipher_klen != 24 &&
689 		    csp->csp_cipher_klen != 32)
690 			return (false);
691 		break;
692 	}
693 	return (true);
694 }
695 
696 static int
697 safe_probesession(device_t dev, const struct crypto_session_params *csp)
698 {
699 	struct safe_softc *sc = device_get_softc(dev);
700 
701 	if (csp->csp_flags != 0)
702 		return (EINVAL);
703 	switch (csp->csp_mode) {
704 	case CSP_MODE_DIGEST:
705 		if (!safe_auth_supported(sc, csp))
706 			return (EINVAL);
707 		break;
708 	case CSP_MODE_CIPHER:
709 		if (!safe_cipher_supported(sc, csp))
710 			return (EINVAL);
711 		break;
712 	case CSP_MODE_ETA:
713 		if (!safe_auth_supported(sc, csp) ||
714 		    !safe_cipher_supported(sc, csp))
715 			return (EINVAL);
716 		break;
717 	default:
718 		return (EINVAL);
719 	}
720 
721 	return (CRYPTODEV_PROBE_HARDWARE);
722 }
723 
724 /*
725  * Allocate a new 'session'.
726  */
727 static int
728 safe_newsession(device_t dev, crypto_session_t cses,
729     const struct crypto_session_params *csp)
730 {
731 	struct safe_session *ses;
732 
733 	ses = crypto_get_driver_session(cses);
734 	if (csp->csp_cipher_alg != 0) {
735 		ses->ses_klen = csp->csp_cipher_klen;
736 		if (csp->csp_cipher_key != NULL)
737 			safe_setup_enckey(ses, csp->csp_cipher_key);
738 	}
739 
740 	if (csp->csp_auth_alg != 0) {
741 		ses->ses_mlen = csp->csp_auth_mlen;
742 		if (ses->ses_mlen == 0) {
743 			ses->ses_mlen = SHA1_HASH_LEN;
744 		}
745 
746 		if (csp->csp_auth_key != NULL) {
747 			safe_setup_mackey(ses, csp->csp_auth_alg,
748 			    csp->csp_auth_key, csp->csp_auth_klen);
749 		}
750 	}
751 
752 	return (0);
753 }
754 
755 static void
756 safe_op_cb(void *arg, bus_dma_segment_t *seg, int nsegs, int error)
757 {
758 	struct safe_operand *op = arg;
759 
760 	DPRINTF(("%s: nsegs %d error %d\n", __func__,
761 		nsegs, error));
762 	if (error != 0)
763 		return;
764 	op->nsegs = nsegs;
765 	bcopy(seg, op->segs, nsegs * sizeof (seg[0]));
766 }
767 
768 static int
769 safe_process(device_t dev, struct cryptop *crp, int hint)
770 {
771 	struct safe_softc *sc = device_get_softc(dev);
772 	const struct crypto_session_params *csp;
773 	int err = 0, i, nicealign, uniform;
774 	int bypass, oplen;
775 	int16_t coffset;
776 	struct safe_session *ses;
777 	struct safe_ringentry *re;
778 	struct safe_sarec *sa;
779 	struct safe_pdesc *pd;
780 	u_int32_t cmd0, cmd1, staterec;
781 
782 	mtx_lock(&sc->sc_ringmtx);
783 	if (sc->sc_front == sc->sc_back && sc->sc_nqchip != 0) {
784 		safestats.st_ringfull++;
785 		sc->sc_needwakeup |= CRYPTO_SYMQ;
786 		mtx_unlock(&sc->sc_ringmtx);
787 		return (ERESTART);
788 	}
789 	re = sc->sc_front;
790 
791 	staterec = re->re_sa.sa_staterec;	/* save */
792 	/* NB: zero everything but the PE descriptor */
793 	bzero(&re->re_sa, sizeof(struct safe_ringentry) - sizeof(re->re_desc));
794 	re->re_sa.sa_staterec = staterec;	/* restore */
795 
796 	re->re_crp = crp;
797 
798 	sa = &re->re_sa;
799 	ses = crypto_get_driver_session(crp->crp_session);
800 	csp = crypto_get_params(crp->crp_session);
801 
802 	cmd0 = SAFE_SA_CMD0_BASIC;		/* basic group operation */
803 	cmd1 = 0;
804 	switch (csp->csp_mode) {
805 	case CSP_MODE_DIGEST:
806 		cmd0 |= SAFE_SA_CMD0_OP_HASH;
807 		break;
808 	case CSP_MODE_CIPHER:
809 		cmd0 |= SAFE_SA_CMD0_OP_CRYPT;
810 		break;
811 	case CSP_MODE_ETA:
812 		cmd0 |= SAFE_SA_CMD0_OP_BOTH;
813 		break;
814 	}
815 
816 	if (csp->csp_cipher_alg != 0) {
817 		if (crp->crp_cipher_key != NULL)
818 			safe_setup_enckey(ses, crp->crp_cipher_key);
819 
820 		switch (csp->csp_cipher_alg) {
821 		case CRYPTO_AES_CBC:
822 			cmd0 |= SAFE_SA_CMD0_AES;
823 			cmd1 |= SAFE_SA_CMD1_CBC;
824 			if (ses->ses_klen * 8 == 128)
825 			     cmd1 |=  SAFE_SA_CMD1_AES128;
826 			else if (ses->ses_klen * 8 == 192)
827 			     cmd1 |=  SAFE_SA_CMD1_AES192;
828 			else
829 			     cmd1 |=  SAFE_SA_CMD1_AES256;
830 		}
831 
832 		/*
833 		 * Setup encrypt/decrypt state.  When using basic ops
834 		 * we can't use an inline IV because hash/crypt offset
835 		 * must be from the end of the IV to the start of the
836 		 * crypt data and this leaves out the preceding header
837 		 * from the hash calculation.  Instead we place the IV
838 		 * in the state record and set the hash/crypt offset to
839 		 * copy both the header+IV.
840 		 */
841 		crypto_read_iv(crp, re->re_sastate.sa_saved_iv);
842 		cmd0 |= SAFE_SA_CMD0_IVLD_STATE;
843 
844 		if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op)) {
845 			cmd0 |= SAFE_SA_CMD0_OUTBOUND;
846 
847 			/*
848 			 * XXX: I suspect we don't need this since we
849 			 * don't save the returned IV.
850 			 */
851 			cmd0 |= SAFE_SA_CMD0_SAVEIV;
852 		} else {
853 			cmd0 |= SAFE_SA_CMD0_INBOUND;
854 		}
855 		/*
856 		 * For basic encryption use the zero pad algorithm.
857 		 * This pads results to an 8-byte boundary and
858 		 * suppresses padding verification for inbound (i.e.
859 		 * decrypt) operations.
860 		 *
861 		 * NB: Not sure if the 8-byte pad boundary is a problem.
862 		 */
863 		cmd0 |= SAFE_SA_CMD0_PAD_ZERO;
864 
865 		/* XXX assert key bufs have the same size */
866 		bcopy(ses->ses_key, sa->sa_key, sizeof(sa->sa_key));
867 	}
868 
869 	if (csp->csp_auth_alg != 0) {
870 		if (crp->crp_auth_key != NULL) {
871 			safe_setup_mackey(ses, csp->csp_auth_alg,
872 			    crp->crp_auth_key, csp->csp_auth_klen);
873 		}
874 
875 		switch (csp->csp_auth_alg) {
876 		case CRYPTO_SHA1_HMAC:
877 			cmd0 |= SAFE_SA_CMD0_SHA1;
878 			cmd1 |= SAFE_SA_CMD1_HMAC;	/* NB: enable HMAC */
879 			break;
880 		}
881 
882 		/*
883 		 * Digest data is loaded from the SA and the hash
884 		 * result is saved to the state block where we
885 		 * retrieve it for return to the caller.
886 		 */
887 		/* XXX assert digest bufs have the same size */
888 		bcopy(ses->ses_hminner, sa->sa_indigest,
889 			sizeof(sa->sa_indigest));
890 		bcopy(ses->ses_hmouter, sa->sa_outdigest,
891 			sizeof(sa->sa_outdigest));
892 
893 		cmd0 |= SAFE_SA_CMD0_HSLD_SA | SAFE_SA_CMD0_SAVEHASH;
894 		re->re_flags |= SAFE_QFLAGS_COPYOUTICV;
895 	}
896 
897 	if (csp->csp_mode == CSP_MODE_ETA) {
898 		/*
899 		 * The driver only supports ETA requests where there
900 		 * is no gap between the AAD and payload.
901 		 */
902 		if (crp->crp_aad_length != 0 &&
903 		    crp->crp_aad_start + crp->crp_aad_length !=
904 		    crp->crp_payload_start) {
905 			safestats.st_lenmismatch++;
906 			err = EINVAL;
907 			goto errout;
908 		}
909 		if (crp->crp_aad_length != 0)
910 			bypass = crp->crp_aad_start;
911 		else
912 			bypass = crp->crp_payload_start;
913 		coffset = crp->crp_aad_length;
914 		oplen = crp->crp_payload_start + crp->crp_payload_length;
915 #ifdef SAFE_DEBUG
916 		if (safe_debug) {
917 			printf("AAD: skip %d, len %d, digest %d\n",
918 			    crp->crp_aad_start, crp->crp_aad_length,
919 			    crp->crp_digest_start);
920 			printf("payload: skip %d, len %d, IV %d\n",
921 			    crp->crp_payload_start, crp->crp_payload_length,
922 			    crp->crp_iv_start);
923 			printf("bypass %d coffset %d oplen %d\n",
924 				bypass, coffset, oplen);
925 		}
926 #endif
927 		if (coffset & 3) {	/* offset must be 32-bit aligned */
928 			DPRINTF(("%s: coffset %u misaligned\n",
929 				__func__, coffset));
930 			safestats.st_coffmisaligned++;
931 			err = EINVAL;
932 			goto errout;
933 		}
934 		coffset >>= 2;
935 		if (coffset > 255) {	/* offset must be <256 dwords */
936 			DPRINTF(("%s: coffset %u too big\n",
937 				__func__, coffset));
938 			safestats.st_cofftoobig++;
939 			err = EINVAL;
940 			goto errout;
941 		}
942 		/*
943 		 * Tell the hardware to copy the header to the output.
944 		 * The header is defined as the data from the end of
945 		 * the bypass to the start of data to be encrypted.
946 		 * Typically this is the inline IV.  Note that you need
947 		 * to do this even if src+dst are the same; it appears
948 		 * that w/o this bit the crypted data is written
949 		 * immediately after the bypass data.
950 		 */
951 		cmd1 |= SAFE_SA_CMD1_HDRCOPY;
952 		/*
953 		 * Disable IP header mutable bit handling.  This is
954 		 * needed to get correct HMAC calculations.
955 		 */
956 		cmd1 |= SAFE_SA_CMD1_MUTABLE;
957 	} else {
958 		bypass = crp->crp_payload_start;
959 		oplen = bypass + crp->crp_payload_length;
960 		coffset = 0;
961 	}
962 	/* XXX verify multiple of 4 when using s/g */
963 	if (bypass > 96) {		/* bypass offset must be <= 96 bytes */
964 		DPRINTF(("%s: bypass %u too big\n", __func__, bypass));
965 		safestats.st_bypasstoobig++;
966 		err = EINVAL;
967 		goto errout;
968 	}
969 
970 	if (bus_dmamap_create(sc->sc_srcdmat, BUS_DMA_NOWAIT, &re->re_src_map)) {
971 		safestats.st_nomap++;
972 		err = ENOMEM;
973 		goto errout;
974 	}
975 	if (bus_dmamap_load_crp(sc->sc_srcdmat, re->re_src_map, crp, safe_op_cb,
976 	    &re->re_src, BUS_DMA_NOWAIT) != 0) {
977 		bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map);
978 		re->re_src_map = NULL;
979 		safestats.st_noload++;
980 		err = ENOMEM;
981 		goto errout;
982 	}
983 	re->re_src_mapsize = crypto_buffer_len(&crp->crp_buf);
984 	nicealign = safe_dmamap_aligned(&re->re_src);
985 	uniform = safe_dmamap_uniform(&re->re_src);
986 
987 	DPRINTF(("src nicealign %u uniform %u nsegs %u\n",
988 		nicealign, uniform, re->re_src.nsegs));
989 	if (re->re_src.nsegs > 1) {
990 		re->re_desc.d_src = sc->sc_spalloc.dma_paddr +
991 			((caddr_t) sc->sc_spfree - (caddr_t) sc->sc_spring);
992 		for (i = 0; i < re->re_src_nsegs; i++) {
993 			/* NB: no need to check if there's space */
994 			pd = sc->sc_spfree;
995 			if (++(sc->sc_spfree) == sc->sc_springtop)
996 				sc->sc_spfree = sc->sc_spring;
997 
998 			KASSERT((pd->pd_flags&3) == 0 ||
999 				(pd->pd_flags&3) == SAFE_PD_DONE,
1000 				("bogus source particle descriptor; flags %x",
1001 				pd->pd_flags));
1002 			pd->pd_addr = re->re_src_segs[i].ds_addr;
1003 			pd->pd_size = re->re_src_segs[i].ds_len;
1004 			pd->pd_flags = SAFE_PD_READY;
1005 		}
1006 		cmd0 |= SAFE_SA_CMD0_IGATHER;
1007 	} else {
1008 		/*
1009 		 * No need for gather, reference the operand directly.
1010 		 */
1011 		re->re_desc.d_src = re->re_src_segs[0].ds_addr;
1012 	}
1013 
1014 	if (csp->csp_mode == CSP_MODE_DIGEST) {
1015 		/*
1016 		 * Hash op; no destination needed.
1017 		 */
1018 	} else {
1019 		if (nicealign && uniform == 1) {
1020 			/*
1021 			 * Source layout is suitable for direct
1022 			 * sharing of the DMA map and segment list.
1023 			 */
1024 			re->re_dst = re->re_src;
1025 		} else if (nicealign && uniform == 2) {
1026 			/*
1027 			 * The source is properly aligned but requires a
1028 			 * different particle list to handle DMA of the
1029 			 * result.  Create a new map and do the load to
1030 			 * create the segment list.  The particle
1031 			 * descriptor setup code below will handle the
1032 			 * rest.
1033 			 */
1034 			if (bus_dmamap_create(sc->sc_dstdmat, BUS_DMA_NOWAIT,
1035 			    &re->re_dst_map)) {
1036 				safestats.st_nomap++;
1037 				err = ENOMEM;
1038 				goto errout;
1039 			}
1040 			if (bus_dmamap_load_crp(sc->sc_dstdmat, re->re_dst_map,
1041 			    crp, safe_op_cb, &re->re_dst, BUS_DMA_NOWAIT) !=
1042 			    0) {
1043 				bus_dmamap_destroy(sc->sc_dstdmat,
1044 				    re->re_dst_map);
1045 				re->re_dst_map = NULL;
1046 				safestats.st_noload++;
1047 				err = ENOMEM;
1048 				goto errout;
1049 			}
1050 		} else if (crp->crp_buf.cb_type == CRYPTO_BUF_MBUF) {
1051 			int totlen, len;
1052 			struct mbuf *m, *top, **mp;
1053 
1054 			/*
1055 			 * DMA constraints require that we allocate a
1056 			 * new mbuf chain for the destination.  We
1057 			 * allocate an entire new set of mbufs of
1058 			 * optimal/required size and then tell the
1059 			 * hardware to copy any bits that are not
1060 			 * created as a byproduct of the operation.
1061 			 */
1062 			if (!nicealign)
1063 				safestats.st_unaligned++;
1064 			if (!uniform)
1065 				safestats.st_notuniform++;
1066 			totlen = re->re_src_mapsize;
1067 			if (crp->crp_buf.cb_mbuf->m_flags & M_PKTHDR) {
1068 				len = MHLEN;
1069 				MGETHDR(m, M_NOWAIT, MT_DATA);
1070 				if (m && !m_dup_pkthdr(m, crp->crp_buf.cb_mbuf,
1071 				    M_NOWAIT)) {
1072 					m_free(m);
1073 					m = NULL;
1074 				}
1075 			} else {
1076 				len = MLEN;
1077 				MGET(m, M_NOWAIT, MT_DATA);
1078 			}
1079 			if (m == NULL) {
1080 				safestats.st_nombuf++;
1081 				err = sc->sc_nqchip ? ERESTART : ENOMEM;
1082 				goto errout;
1083 			}
1084 			if (totlen >= MINCLSIZE) {
1085 				if (!(MCLGET(m, M_NOWAIT))) {
1086 					m_free(m);
1087 					safestats.st_nomcl++;
1088 					err = sc->sc_nqchip ?
1089 					    ERESTART : ENOMEM;
1090 					goto errout;
1091 				}
1092 				len = MCLBYTES;
1093 			}
1094 			m->m_len = len;
1095 			top = NULL;
1096 			mp = &top;
1097 
1098 			while (totlen > 0) {
1099 				if (top) {
1100 					MGET(m, M_NOWAIT, MT_DATA);
1101 					if (m == NULL) {
1102 						m_freem(top);
1103 						safestats.st_nombuf++;
1104 						err = sc->sc_nqchip ?
1105 						    ERESTART : ENOMEM;
1106 						goto errout;
1107 					}
1108 					len = MLEN;
1109 				}
1110 				if (top && totlen >= MINCLSIZE) {
1111 					if (!(MCLGET(m, M_NOWAIT))) {
1112 						*mp = m;
1113 						m_freem(top);
1114 						safestats.st_nomcl++;
1115 						err = sc->sc_nqchip ?
1116 						    ERESTART : ENOMEM;
1117 						goto errout;
1118 					}
1119 					len = MCLBYTES;
1120 				}
1121 				m->m_len = len = min(totlen, len);
1122 				totlen -= len;
1123 				*mp = m;
1124 				mp = &m->m_next;
1125 			}
1126 			re->re_dst_m = top;
1127 			if (bus_dmamap_create(sc->sc_dstdmat,
1128 			    BUS_DMA_NOWAIT, &re->re_dst_map) != 0) {
1129 				safestats.st_nomap++;
1130 				err = ENOMEM;
1131 				goto errout;
1132 			}
1133 			if (bus_dmamap_load_mbuf_sg(sc->sc_dstdmat,
1134 			    re->re_dst_map, top, re->re_dst_segs,
1135 			    &re->re_dst_nsegs, 0) != 0) {
1136 				bus_dmamap_destroy(sc->sc_dstdmat,
1137 				    re->re_dst_map);
1138 				re->re_dst_map = NULL;
1139 				safestats.st_noload++;
1140 				err = ENOMEM;
1141 				goto errout;
1142 			}
1143 			re->re_dst_mapsize = re->re_src_mapsize;
1144 			if (re->re_src.mapsize > oplen) {
1145 				/*
1146 				 * There's data following what the
1147 				 * hardware will copy for us.  If this
1148 				 * isn't just the ICV (that's going to
1149 				 * be written on completion), copy it
1150 				 * to the new mbufs
1151 				 */
1152 				if (!(csp->csp_mode == CSP_MODE_ETA &&
1153 				    (re->re_src.mapsize-oplen) == ses->ses_mlen &&
1154 				    crp->crp_digest_start == oplen))
1155 					safe_mcopy(crp->crp_buf.cb_mbuf,
1156 					    re->re_dst_m, oplen);
1157 				else
1158 					safestats.st_noicvcopy++;
1159 			}
1160 		} else {
1161 			if (!nicealign) {
1162 				safestats.st_iovmisaligned++;
1163 				err = EINVAL;
1164 				goto errout;
1165 			} else {
1166 				/*
1167 				 * There's no way to handle the DMA
1168 				 * requirements with this uio.  We
1169 				 * could create a separate DMA area for
1170 				 * the result and then copy it back,
1171 				 * but for now we just bail and return
1172 				 * an error.  Note that uio requests
1173 				 * > SAFE_MAX_DSIZE are handled because
1174 				 * the DMA map and segment list for the
1175 				 * destination wil result in a
1176 				 * destination particle list that does
1177 				 * the necessary scatter DMA.
1178 				 */
1179 				safestats.st_iovnotuniform++;
1180 				err = EINVAL;
1181 				goto errout;
1182 			}
1183 		}
1184 
1185 		if (re->re_dst.nsegs > 1) {
1186 			re->re_desc.d_dst = sc->sc_dpalloc.dma_paddr +
1187 			    ((caddr_t) sc->sc_dpfree - (caddr_t) sc->sc_dpring);
1188 			for (i = 0; i < re->re_dst_nsegs; i++) {
1189 				pd = sc->sc_dpfree;
1190 				KASSERT((pd->pd_flags&3) == 0 ||
1191 					(pd->pd_flags&3) == SAFE_PD_DONE,
1192 					("bogus dest particle descriptor; flags %x",
1193 						pd->pd_flags));
1194 				if (++(sc->sc_dpfree) == sc->sc_dpringtop)
1195 					sc->sc_dpfree = sc->sc_dpring;
1196 				pd->pd_addr = re->re_dst_segs[i].ds_addr;
1197 				pd->pd_flags = SAFE_PD_READY;
1198 			}
1199 			cmd0 |= SAFE_SA_CMD0_OSCATTER;
1200 		} else {
1201 			/*
1202 			 * No need for scatter, reference the operand directly.
1203 			 */
1204 			re->re_desc.d_dst = re->re_dst_segs[0].ds_addr;
1205 		}
1206 	}
1207 
1208 	/*
1209 	 * All done with setup; fillin the SA command words
1210 	 * and the packet engine descriptor.  The operation
1211 	 * is now ready for submission to the hardware.
1212 	 */
1213 	sa->sa_cmd0 = cmd0 | SAFE_SA_CMD0_IPCI | SAFE_SA_CMD0_OPCI;
1214 	sa->sa_cmd1 = cmd1
1215 		    | (coffset << SAFE_SA_CMD1_OFFSET_S)
1216 		    | SAFE_SA_CMD1_SAREV1	/* Rev 1 SA data structure */
1217 		    | SAFE_SA_CMD1_SRPCI
1218 		    ;
1219 	/*
1220 	 * NB: the order of writes is important here.  In case the
1221 	 * chip is scanning the ring because of an outstanding request
1222 	 * it might nab this one too.  In that case we need to make
1223 	 * sure the setup is complete before we write the length
1224 	 * field of the descriptor as it signals the descriptor is
1225 	 * ready for processing.
1226 	 */
1227 	re->re_desc.d_csr = SAFE_PE_CSR_READY | SAFE_PE_CSR_SAPCI;
1228 	if (csp->csp_auth_alg != 0)
1229 		re->re_desc.d_csr |= SAFE_PE_CSR_LOADSA | SAFE_PE_CSR_HASHFINAL;
1230 	re->re_desc.d_len = oplen
1231 			  | SAFE_PE_LEN_READY
1232 			  | (bypass << SAFE_PE_LEN_BYPASS_S)
1233 			  ;
1234 
1235 	safestats.st_ipackets++;
1236 	safestats.st_ibytes += oplen;
1237 
1238 	if (++(sc->sc_front) == sc->sc_ringtop)
1239 		sc->sc_front = sc->sc_ring;
1240 
1241 	/* XXX honor batching */
1242 	safe_feed(sc, re);
1243 	mtx_unlock(&sc->sc_ringmtx);
1244 	return (0);
1245 
1246 errout:
1247 	if (re->re_dst_m != NULL)
1248 		m_freem(re->re_dst_m);
1249 
1250 	if (re->re_dst_map != NULL && re->re_dst_map != re->re_src_map) {
1251 		bus_dmamap_unload(sc->sc_dstdmat, re->re_dst_map);
1252 		bus_dmamap_destroy(sc->sc_dstdmat, re->re_dst_map);
1253 	}
1254 	if (re->re_src_map != NULL) {
1255 		bus_dmamap_unload(sc->sc_srcdmat, re->re_src_map);
1256 		bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map);
1257 	}
1258 	mtx_unlock(&sc->sc_ringmtx);
1259 	if (err != ERESTART) {
1260 		crp->crp_etype = err;
1261 		crypto_done(crp);
1262 		err = 0;
1263 	} else {
1264 		sc->sc_needwakeup |= CRYPTO_SYMQ;
1265 	}
1266 	return (err);
1267 }
1268 
1269 static void
1270 safe_callback(struct safe_softc *sc, struct safe_ringentry *re)
1271 {
1272 	const struct crypto_session_params *csp;
1273 	struct cryptop *crp = (struct cryptop *)re->re_crp;
1274 	struct safe_session *ses;
1275 	uint8_t hash[HASH_MAX_LEN];
1276 
1277 	ses = crypto_get_driver_session(crp->crp_session);
1278 	csp = crypto_get_params(crp->crp_session);
1279 
1280 	safestats.st_opackets++;
1281 	safestats.st_obytes += re->re_dst.mapsize;
1282 
1283 	safe_dma_sync(&sc->sc_ringalloc,
1284 		BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1285 	if (re->re_desc.d_csr & SAFE_PE_CSR_STATUS) {
1286 		device_printf(sc->sc_dev, "csr 0x%x cmd0 0x%x cmd1 0x%x\n",
1287 			re->re_desc.d_csr,
1288 			re->re_sa.sa_cmd0, re->re_sa.sa_cmd1);
1289 		safestats.st_peoperr++;
1290 		crp->crp_etype = EIO;		/* something more meaningful? */
1291 	}
1292 
1293 	/*
1294 	 * XXX: Should crp_buf.cb_mbuf be updated to re->re_dst_m if
1295 	 * it is non-NULL?
1296 	 */
1297 
1298 	if (re->re_dst_map != NULL && re->re_dst_map != re->re_src_map) {
1299 		bus_dmamap_sync(sc->sc_dstdmat, re->re_dst_map,
1300 		    BUS_DMASYNC_POSTREAD);
1301 		bus_dmamap_unload(sc->sc_dstdmat, re->re_dst_map);
1302 		bus_dmamap_destroy(sc->sc_dstdmat, re->re_dst_map);
1303 	}
1304 	bus_dmamap_sync(sc->sc_srcdmat, re->re_src_map, BUS_DMASYNC_POSTWRITE);
1305 	bus_dmamap_unload(sc->sc_srcdmat, re->re_src_map);
1306 	bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map);
1307 
1308 	if (re->re_flags & SAFE_QFLAGS_COPYOUTICV) {
1309 		if (csp->csp_auth_alg == CRYPTO_SHA1_HMAC) {
1310 			/*
1311 			 * SHA-1 ICV's are byte-swapped; fix 'em up
1312 			 * before copying them to their destination.
1313 			 */
1314 			re->re_sastate.sa_saved_indigest[0] =
1315 			    bswap32(re->re_sastate.sa_saved_indigest[0]);
1316 			re->re_sastate.sa_saved_indigest[1] =
1317 			    bswap32(re->re_sastate.sa_saved_indigest[1]);
1318 			re->re_sastate.sa_saved_indigest[2] =
1319 			    bswap32(re->re_sastate.sa_saved_indigest[2]);
1320 		}
1321 
1322 		if (crp->crp_op & CRYPTO_OP_VERIFY_DIGEST) {
1323 			crypto_copydata(crp, crp->crp_digest_start,
1324 			    ses->ses_mlen, hash);
1325 			if (timingsafe_bcmp(re->re_sastate.sa_saved_indigest,
1326 			    hash, ses->ses_mlen) != 0)
1327 				crp->crp_etype = EBADMSG;
1328 		} else
1329 			crypto_copyback(crp, crp->crp_digest_start,
1330 			    ses->ses_mlen, re->re_sastate.sa_saved_indigest);
1331 	}
1332 	crypto_done(crp);
1333 }
1334 
1335 /*
1336  * Copy all data past offset from srcm to dstm.
1337  */
1338 static void
1339 safe_mcopy(struct mbuf *srcm, struct mbuf *dstm, u_int offset)
1340 {
1341 	u_int j, dlen, slen;
1342 	caddr_t dptr, sptr;
1343 
1344 	/*
1345 	 * Advance src and dst to offset.
1346 	 */
1347 	j = offset;
1348 	while (j >= srcm->m_len) {
1349 		j -= srcm->m_len;
1350 		srcm = srcm->m_next;
1351 		if (srcm == NULL)
1352 			return;
1353 	}
1354 	sptr = mtod(srcm, caddr_t) + j;
1355 	slen = srcm->m_len - j;
1356 
1357 	j = offset;
1358 	while (j >= dstm->m_len) {
1359 		j -= dstm->m_len;
1360 		dstm = dstm->m_next;
1361 		if (dstm == NULL)
1362 			return;
1363 	}
1364 	dptr = mtod(dstm, caddr_t) + j;
1365 	dlen = dstm->m_len - j;
1366 
1367 	/*
1368 	 * Copy everything that remains.
1369 	 */
1370 	for (;;) {
1371 		j = min(slen, dlen);
1372 		bcopy(sptr, dptr, j);
1373 		if (slen == j) {
1374 			srcm = srcm->m_next;
1375 			if (srcm == NULL)
1376 				return;
1377 			sptr = srcm->m_data;
1378 			slen = srcm->m_len;
1379 		} else
1380 			sptr += j, slen -= j;
1381 		if (dlen == j) {
1382 			dstm = dstm->m_next;
1383 			if (dstm == NULL)
1384 				return;
1385 			dptr = dstm->m_data;
1386 			dlen = dstm->m_len;
1387 		} else
1388 			dptr += j, dlen -= j;
1389 	}
1390 }
1391 
1392 #ifndef SAFE_NO_RNG
1393 #define	SAFE_RNG_MAXWAIT	1000
1394 
1395 static void
1396 safe_rng_init(struct safe_softc *sc)
1397 {
1398 	u_int32_t w, v;
1399 	int i;
1400 
1401 	WRITE_REG(sc, SAFE_RNG_CTRL, 0);
1402 	/* use default value according to the manual */
1403 	WRITE_REG(sc, SAFE_RNG_CNFG, 0x834);	/* magic from SafeNet */
1404 	WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0);
1405 
1406 	/*
1407 	 * There is a bug in rev 1.0 of the 1140 that when the RNG
1408 	 * is brought out of reset the ready status flag does not
1409 	 * work until the RNG has finished its internal initialization.
1410 	 *
1411 	 * So in order to determine the device is through its
1412 	 * initialization we must read the data register, using the
1413 	 * status reg in the read in case it is initialized.  Then read
1414 	 * the data register until it changes from the first read.
1415 	 * Once it changes read the data register until it changes
1416 	 * again.  At this time the RNG is considered initialized.
1417 	 * This could take between 750ms - 1000ms in time.
1418 	 */
1419 	i = 0;
1420 	w = READ_REG(sc, SAFE_RNG_OUT);
1421 	do {
1422 		v = READ_REG(sc, SAFE_RNG_OUT);
1423 		if (v != w) {
1424 			w = v;
1425 			break;
1426 		}
1427 		DELAY(10);
1428 	} while (++i < SAFE_RNG_MAXWAIT);
1429 
1430 	/* Wait Until data changes again */
1431 	i = 0;
1432 	do {
1433 		v = READ_REG(sc, SAFE_RNG_OUT);
1434 		if (v != w)
1435 			break;
1436 		DELAY(10);
1437 	} while (++i < SAFE_RNG_MAXWAIT);
1438 }
1439 
1440 static __inline void
1441 safe_rng_disable_short_cycle(struct safe_softc *sc)
1442 {
1443 	WRITE_REG(sc, SAFE_RNG_CTRL,
1444 		READ_REG(sc, SAFE_RNG_CTRL) &~ SAFE_RNG_CTRL_SHORTEN);
1445 }
1446 
1447 static __inline void
1448 safe_rng_enable_short_cycle(struct safe_softc *sc)
1449 {
1450 	WRITE_REG(sc, SAFE_RNG_CTRL,
1451 		READ_REG(sc, SAFE_RNG_CTRL) | SAFE_RNG_CTRL_SHORTEN);
1452 }
1453 
1454 static __inline u_int32_t
1455 safe_rng_read(struct safe_softc *sc)
1456 {
1457 	int i;
1458 
1459 	i = 0;
1460 	while (READ_REG(sc, SAFE_RNG_STAT) != 0 && ++i < SAFE_RNG_MAXWAIT)
1461 		;
1462 	return READ_REG(sc, SAFE_RNG_OUT);
1463 }
1464 
1465 static void
1466 safe_rng(void *arg)
1467 {
1468 	struct safe_softc *sc = arg;
1469 	u_int32_t buf[SAFE_RNG_MAXBUFSIZ];	/* NB: maybe move to softc */
1470 	u_int maxwords;
1471 	int i;
1472 
1473 	safestats.st_rng++;
1474 	/*
1475 	 * Fetch the next block of data.
1476 	 */
1477 	maxwords = safe_rngbufsize;
1478 	if (maxwords > SAFE_RNG_MAXBUFSIZ)
1479 		maxwords = SAFE_RNG_MAXBUFSIZ;
1480 retry:
1481 	for (i = 0; i < maxwords; i++)
1482 		buf[i] = safe_rng_read(sc);
1483 	/*
1484 	 * Check the comparator alarm count and reset the h/w if
1485 	 * it exceeds our threshold.  This guards against the
1486 	 * hardware oscillators resonating with external signals.
1487 	 */
1488 	if (READ_REG(sc, SAFE_RNG_ALM_CNT) > safe_rngmaxalarm) {
1489 		u_int32_t freq_inc, w;
1490 
1491 		DPRINTF(("%s: alarm count %u exceeds threshold %u\n", __func__,
1492 			READ_REG(sc, SAFE_RNG_ALM_CNT), safe_rngmaxalarm));
1493 		safestats.st_rngalarm++;
1494 		safe_rng_enable_short_cycle(sc);
1495 		freq_inc = 18;
1496 		for (i = 0; i < 64; i++) {
1497 			w = READ_REG(sc, SAFE_RNG_CNFG);
1498 			freq_inc = ((w + freq_inc) & 0x3fL);
1499 			w = ((w & ~0x3fL) | freq_inc);
1500 			WRITE_REG(sc, SAFE_RNG_CNFG, w);
1501 
1502 			WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0);
1503 
1504 			(void) safe_rng_read(sc);
1505 			DELAY(25);
1506 
1507 			if (READ_REG(sc, SAFE_RNG_ALM_CNT) == 0) {
1508 				safe_rng_disable_short_cycle(sc);
1509 				goto retry;
1510 			}
1511 			freq_inc = 1;
1512 		}
1513 		safe_rng_disable_short_cycle(sc);
1514 	} else
1515 		WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0);
1516 
1517 	(*sc->sc_harvest)(sc->sc_rndtest, buf, maxwords*sizeof (u_int32_t));
1518 	callout_reset(&sc->sc_rngto,
1519 		hz * (safe_rnginterval ? safe_rnginterval : 1), safe_rng, sc);
1520 }
1521 #endif /* SAFE_NO_RNG */
1522 
1523 static void
1524 safe_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1525 {
1526 	bus_addr_t *paddr = (bus_addr_t*) arg;
1527 	*paddr = segs->ds_addr;
1528 }
1529 
1530 static int
1531 safe_dma_malloc(
1532 	struct safe_softc *sc,
1533 	bus_size_t size,
1534 	struct safe_dma_alloc *dma,
1535 	int mapflags
1536 )
1537 {
1538 	int r;
1539 
1540 	r = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev),	/* parent */
1541 			       sizeof(u_int32_t), 0,	/* alignment, bounds */
1542 			       BUS_SPACE_MAXADDR_32BIT,	/* lowaddr */
1543 			       BUS_SPACE_MAXADDR,	/* highaddr */
1544 			       NULL, NULL,		/* filter, filterarg */
1545 			       size,			/* maxsize */
1546 			       1,			/* nsegments */
1547 			       size,			/* maxsegsize */
1548 			       BUS_DMA_ALLOCNOW,	/* flags */
1549 			       NULL, NULL,		/* locking */
1550 			       &dma->dma_tag);
1551 	if (r != 0) {
1552 		device_printf(sc->sc_dev, "safe_dma_malloc: "
1553 			"bus_dma_tag_create failed; error %u\n", r);
1554 		goto fail_0;
1555 	}
1556 
1557 	r = bus_dmamem_alloc(dma->dma_tag, (void**) &dma->dma_vaddr,
1558 			     BUS_DMA_NOWAIT, &dma->dma_map);
1559 	if (r != 0) {
1560 		device_printf(sc->sc_dev, "safe_dma_malloc: "
1561 			"bus_dmammem_alloc failed; size %ju, error %u\n",
1562 			(uintmax_t)size, r);
1563 		goto fail_1;
1564 	}
1565 
1566 	r = bus_dmamap_load(dma->dma_tag, dma->dma_map, dma->dma_vaddr,
1567 		            size,
1568 			    safe_dmamap_cb,
1569 			    &dma->dma_paddr,
1570 			    mapflags | BUS_DMA_NOWAIT);
1571 	if (r != 0) {
1572 		device_printf(sc->sc_dev, "safe_dma_malloc: "
1573 			"bus_dmamap_load failed; error %u\n", r);
1574 		goto fail_2;
1575 	}
1576 
1577 	dma->dma_size = size;
1578 	return (0);
1579 
1580 	bus_dmamap_unload(dma->dma_tag, dma->dma_map);
1581 fail_2:
1582 	bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
1583 fail_1:
1584 	bus_dma_tag_destroy(dma->dma_tag);
1585 fail_0:
1586 	dma->dma_tag = NULL;
1587 	return (r);
1588 }
1589 
1590 static void
1591 safe_dma_free(struct safe_softc *sc, struct safe_dma_alloc *dma)
1592 {
1593 	bus_dmamap_unload(dma->dma_tag, dma->dma_map);
1594 	bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
1595 	bus_dma_tag_destroy(dma->dma_tag);
1596 }
1597 
1598 /*
1599  * Resets the board.  Values in the regesters are left as is
1600  * from the reset (i.e. initial values are assigned elsewhere).
1601  */
1602 static void
1603 safe_reset_board(struct safe_softc *sc)
1604 {
1605 	u_int32_t v;
1606 	/*
1607 	 * Reset the device.  The manual says no delay
1608 	 * is needed between marking and clearing reset.
1609 	 */
1610 	v = READ_REG(sc, SAFE_PE_DMACFG) &~
1611 		(SAFE_PE_DMACFG_PERESET | SAFE_PE_DMACFG_PDRRESET |
1612 		 SAFE_PE_DMACFG_SGRESET);
1613 	WRITE_REG(sc, SAFE_PE_DMACFG, v
1614 				    | SAFE_PE_DMACFG_PERESET
1615 				    | SAFE_PE_DMACFG_PDRRESET
1616 				    | SAFE_PE_DMACFG_SGRESET);
1617 	WRITE_REG(sc, SAFE_PE_DMACFG, v);
1618 }
1619 
1620 /*
1621  * Initialize registers we need to touch only once.
1622  */
1623 static void
1624 safe_init_board(struct safe_softc *sc)
1625 {
1626 	u_int32_t v, dwords;
1627 
1628 	v = READ_REG(sc, SAFE_PE_DMACFG);
1629 	v &=~ SAFE_PE_DMACFG_PEMODE;
1630 	v |= SAFE_PE_DMACFG_FSENA		/* failsafe enable */
1631 	  |  SAFE_PE_DMACFG_GPRPCI		/* gather ring on PCI */
1632 	  |  SAFE_PE_DMACFG_SPRPCI		/* scatter ring on PCI */
1633 	  |  SAFE_PE_DMACFG_ESDESC		/* endian-swap descriptors */
1634 	  |  SAFE_PE_DMACFG_ESSA		/* endian-swap SA's */
1635 	  |  SAFE_PE_DMACFG_ESPDESC		/* endian-swap part. desc's */
1636 	  ;
1637 	WRITE_REG(sc, SAFE_PE_DMACFG, v);
1638 #if 0
1639 	/* XXX select byte swap based on host byte order */
1640 	WRITE_REG(sc, SAFE_ENDIAN, 0x1b);
1641 #endif
1642 	if (sc->sc_chiprev == SAFE_REV(1,0)) {
1643 		/*
1644 		 * Avoid large PCI DMA transfers.  Rev 1.0 has a bug where
1645 		 * "target mode transfers" done while the chip is DMA'ing
1646 		 * >1020 bytes cause the hardware to lockup.  To avoid this
1647 		 * we reduce the max PCI transfer size and use small source
1648 		 * particle descriptors (<= 256 bytes).
1649 		 */
1650 		WRITE_REG(sc, SAFE_DMA_CFG, 256);
1651 		device_printf(sc->sc_dev,
1652 			"Reduce max DMA size to %u words for rev %u.%u WAR\n",
1653 			(READ_REG(sc, SAFE_DMA_CFG)>>2) & 0xff,
1654 			SAFE_REV_MAJ(sc->sc_chiprev),
1655 			SAFE_REV_MIN(sc->sc_chiprev));
1656 	}
1657 
1658 	/* NB: operands+results are overlaid */
1659 	WRITE_REG(sc, SAFE_PE_PDRBASE, sc->sc_ringalloc.dma_paddr);
1660 	WRITE_REG(sc, SAFE_PE_RDRBASE, sc->sc_ringalloc.dma_paddr);
1661 	/*
1662 	 * Configure ring entry size and number of items in the ring.
1663 	 */
1664 	KASSERT((sizeof(struct safe_ringentry) % sizeof(u_int32_t)) == 0,
1665 		("PE ring entry not 32-bit aligned!"));
1666 	dwords = sizeof(struct safe_ringentry) / sizeof(u_int32_t);
1667 	WRITE_REG(sc, SAFE_PE_RINGCFG,
1668 		(dwords << SAFE_PE_RINGCFG_OFFSET_S) | SAFE_MAX_NQUEUE);
1669 	WRITE_REG(sc, SAFE_PE_RINGPOLL, 0);	/* disable polling */
1670 
1671 	WRITE_REG(sc, SAFE_PE_GRNGBASE, sc->sc_spalloc.dma_paddr);
1672 	WRITE_REG(sc, SAFE_PE_SRNGBASE, sc->sc_dpalloc.dma_paddr);
1673 	WRITE_REG(sc, SAFE_PE_PARTSIZE,
1674 		(SAFE_TOTAL_DPART<<16) | SAFE_TOTAL_SPART);
1675 	/*
1676 	 * NB: destination particles are fixed size.  We use
1677 	 *     an mbuf cluster and require all results go to
1678 	 *     clusters or smaller.
1679 	 */
1680 	WRITE_REG(sc, SAFE_PE_PARTCFG, SAFE_MAX_DSIZE);
1681 
1682 	/* it's now safe to enable PE mode, do it */
1683 	WRITE_REG(sc, SAFE_PE_DMACFG, v | SAFE_PE_DMACFG_PEMODE);
1684 
1685 	/*
1686 	 * Configure hardware to use level-triggered interrupts and
1687 	 * to interrupt after each descriptor is processed.
1688 	 */
1689 	WRITE_REG(sc, SAFE_HI_CFG, SAFE_HI_CFG_LEVEL);
1690 	WRITE_REG(sc, SAFE_HI_DESC_CNT, 1);
1691 	WRITE_REG(sc, SAFE_HI_MASK, SAFE_INT_PE_DDONE | SAFE_INT_PE_ERROR);
1692 }
1693 
1694 /*
1695  * Init PCI registers
1696  */
1697 static void
1698 safe_init_pciregs(device_t dev)
1699 {
1700 }
1701 
1702 /*
1703  * Clean up after a chip crash.
1704  * It is assumed that the caller in splimp()
1705  */
1706 static void
1707 safe_cleanchip(struct safe_softc *sc)
1708 {
1709 
1710 	if (sc->sc_nqchip != 0) {
1711 		struct safe_ringentry *re = sc->sc_back;
1712 
1713 		while (re != sc->sc_front) {
1714 			if (re->re_desc.d_csr != 0)
1715 				safe_free_entry(sc, re);
1716 			if (++re == sc->sc_ringtop)
1717 				re = sc->sc_ring;
1718 		}
1719 		sc->sc_back = re;
1720 		sc->sc_nqchip = 0;
1721 	}
1722 }
1723 
1724 /*
1725  * free a safe_q
1726  * It is assumed that the caller is within splimp().
1727  */
1728 static int
1729 safe_free_entry(struct safe_softc *sc, struct safe_ringentry *re)
1730 {
1731 	struct cryptop *crp;
1732 
1733 	/*
1734 	 * Free header MCR
1735 	 */
1736 	if (re->re_dst_m != NULL)
1737 		m_freem(re->re_dst_m);
1738 
1739 	crp = (struct cryptop *)re->re_crp;
1740 
1741 	re->re_desc.d_csr = 0;
1742 
1743 	crp->crp_etype = EFAULT;
1744 	crypto_done(crp);
1745 	return(0);
1746 }
1747 
1748 /*
1749  * Routine to reset the chip and clean up.
1750  * It is assumed that the caller is in splimp()
1751  */
1752 static void
1753 safe_totalreset(struct safe_softc *sc)
1754 {
1755 	safe_reset_board(sc);
1756 	safe_init_board(sc);
1757 	safe_cleanchip(sc);
1758 }
1759 
1760 /*
1761  * Is the operand suitable aligned for direct DMA.  Each
1762  * segment must be aligned on a 32-bit boundary and all
1763  * but the last segment must be a multiple of 4 bytes.
1764  */
1765 static int
1766 safe_dmamap_aligned(const struct safe_operand *op)
1767 {
1768 	int i;
1769 
1770 	for (i = 0; i < op->nsegs; i++) {
1771 		if (op->segs[i].ds_addr & 3)
1772 			return (0);
1773 		if (i != (op->nsegs - 1) && (op->segs[i].ds_len & 3))
1774 			return (0);
1775 	}
1776 	return (1);
1777 }
1778 
1779 /*
1780  * Is the operand suitable for direct DMA as the destination
1781  * of an operation.  The hardware requires that each ``particle''
1782  * but the last in an operation result have the same size.  We
1783  * fix that size at SAFE_MAX_DSIZE bytes.  This routine returns
1784  * 0 if some segment is not a multiple of of this size, 1 if all
1785  * segments are exactly this size, or 2 if segments are at worst
1786  * a multple of this size.
1787  */
1788 static int
1789 safe_dmamap_uniform(const struct safe_operand *op)
1790 {
1791 	int result = 1;
1792 
1793 	if (op->nsegs > 0) {
1794 		int i;
1795 
1796 		for (i = 0; i < op->nsegs-1; i++) {
1797 			if (op->segs[i].ds_len % SAFE_MAX_DSIZE)
1798 				return (0);
1799 			if (op->segs[i].ds_len != SAFE_MAX_DSIZE)
1800 				result = 2;
1801 		}
1802 	}
1803 	return (result);
1804 }
1805 
1806 #ifdef SAFE_DEBUG
1807 static void
1808 safe_dump_dmastatus(struct safe_softc *sc, const char *tag)
1809 {
1810 	printf("%s: ENDIAN 0x%x SRC 0x%x DST 0x%x STAT 0x%x\n"
1811 		, tag
1812 		, READ_REG(sc, SAFE_DMA_ENDIAN)
1813 		, READ_REG(sc, SAFE_DMA_SRCADDR)
1814 		, READ_REG(sc, SAFE_DMA_DSTADDR)
1815 		, READ_REG(sc, SAFE_DMA_STAT)
1816 	);
1817 }
1818 
1819 static void
1820 safe_dump_intrstate(struct safe_softc *sc, const char *tag)
1821 {
1822 	printf("%s: HI_CFG 0x%x HI_MASK 0x%x HI_DESC_CNT 0x%x HU_STAT 0x%x HM_STAT 0x%x\n"
1823 		, tag
1824 		, READ_REG(sc, SAFE_HI_CFG)
1825 		, READ_REG(sc, SAFE_HI_MASK)
1826 		, READ_REG(sc, SAFE_HI_DESC_CNT)
1827 		, READ_REG(sc, SAFE_HU_STAT)
1828 		, READ_REG(sc, SAFE_HM_STAT)
1829 	);
1830 }
1831 
1832 static void
1833 safe_dump_ringstate(struct safe_softc *sc, const char *tag)
1834 {
1835 	u_int32_t estat = READ_REG(sc, SAFE_PE_ERNGSTAT);
1836 
1837 	/* NB: assume caller has lock on ring */
1838 	printf("%s: ERNGSTAT %x (next %u) back %lu front %lu\n",
1839 		tag,
1840 		estat, (estat >> SAFE_PE_ERNGSTAT_NEXT_S),
1841 		(unsigned long)(sc->sc_back - sc->sc_ring),
1842 		(unsigned long)(sc->sc_front - sc->sc_ring));
1843 }
1844 
1845 static void
1846 safe_dump_request(struct safe_softc *sc, const char* tag, struct safe_ringentry *re)
1847 {
1848 	int ix, nsegs;
1849 
1850 	ix = re - sc->sc_ring;
1851 	printf("%s: %p (%u): csr %x src %x dst %x sa %x len %x\n"
1852 		, tag
1853 		, re, ix
1854 		, re->re_desc.d_csr
1855 		, re->re_desc.d_src
1856 		, re->re_desc.d_dst
1857 		, re->re_desc.d_sa
1858 		, re->re_desc.d_len
1859 	);
1860 	if (re->re_src.nsegs > 1) {
1861 		ix = (re->re_desc.d_src - sc->sc_spalloc.dma_paddr) /
1862 			sizeof(struct safe_pdesc);
1863 		for (nsegs = re->re_src.nsegs; nsegs; nsegs--) {
1864 			printf(" spd[%u] %p: %p size %u flags %x"
1865 				, ix, &sc->sc_spring[ix]
1866 				, (caddr_t)(uintptr_t) sc->sc_spring[ix].pd_addr
1867 				, sc->sc_spring[ix].pd_size
1868 				, sc->sc_spring[ix].pd_flags
1869 			);
1870 			if (sc->sc_spring[ix].pd_size == 0)
1871 				printf(" (zero!)");
1872 			printf("\n");
1873 			if (++ix == SAFE_TOTAL_SPART)
1874 				ix = 0;
1875 		}
1876 	}
1877 	if (re->re_dst.nsegs > 1) {
1878 		ix = (re->re_desc.d_dst - sc->sc_dpalloc.dma_paddr) /
1879 			sizeof(struct safe_pdesc);
1880 		for (nsegs = re->re_dst.nsegs; nsegs; nsegs--) {
1881 			printf(" dpd[%u] %p: %p flags %x\n"
1882 				, ix, &sc->sc_dpring[ix]
1883 				, (caddr_t)(uintptr_t) sc->sc_dpring[ix].pd_addr
1884 				, sc->sc_dpring[ix].pd_flags
1885 			);
1886 			if (++ix == SAFE_TOTAL_DPART)
1887 				ix = 0;
1888 		}
1889 	}
1890 	printf("sa: cmd0 %08x cmd1 %08x staterec %x\n",
1891 		re->re_sa.sa_cmd0, re->re_sa.sa_cmd1, re->re_sa.sa_staterec);
1892 	printf("sa: key %x %x %x %x %x %x %x %x\n"
1893 		, re->re_sa.sa_key[0]
1894 		, re->re_sa.sa_key[1]
1895 		, re->re_sa.sa_key[2]
1896 		, re->re_sa.sa_key[3]
1897 		, re->re_sa.sa_key[4]
1898 		, re->re_sa.sa_key[5]
1899 		, re->re_sa.sa_key[6]
1900 		, re->re_sa.sa_key[7]
1901 	);
1902 	printf("sa: indigest %x %x %x %x %x\n"
1903 		, re->re_sa.sa_indigest[0]
1904 		, re->re_sa.sa_indigest[1]
1905 		, re->re_sa.sa_indigest[2]
1906 		, re->re_sa.sa_indigest[3]
1907 		, re->re_sa.sa_indigest[4]
1908 	);
1909 	printf("sa: outdigest %x %x %x %x %x\n"
1910 		, re->re_sa.sa_outdigest[0]
1911 		, re->re_sa.sa_outdigest[1]
1912 		, re->re_sa.sa_outdigest[2]
1913 		, re->re_sa.sa_outdigest[3]
1914 		, re->re_sa.sa_outdigest[4]
1915 	);
1916 	printf("sr: iv %x %x %x %x\n"
1917 		, re->re_sastate.sa_saved_iv[0]
1918 		, re->re_sastate.sa_saved_iv[1]
1919 		, re->re_sastate.sa_saved_iv[2]
1920 		, re->re_sastate.sa_saved_iv[3]
1921 	);
1922 	printf("sr: hashbc %u indigest %x %x %x %x %x\n"
1923 		, re->re_sastate.sa_saved_hashbc
1924 		, re->re_sastate.sa_saved_indigest[0]
1925 		, re->re_sastate.sa_saved_indigest[1]
1926 		, re->re_sastate.sa_saved_indigest[2]
1927 		, re->re_sastate.sa_saved_indigest[3]
1928 		, re->re_sastate.sa_saved_indigest[4]
1929 	);
1930 }
1931 
1932 static void
1933 safe_dump_ring(struct safe_softc *sc, const char *tag)
1934 {
1935 	mtx_lock(&sc->sc_ringmtx);
1936 	printf("\nSafeNet Ring State:\n");
1937 	safe_dump_intrstate(sc, tag);
1938 	safe_dump_dmastatus(sc, tag);
1939 	safe_dump_ringstate(sc, tag);
1940 	if (sc->sc_nqchip) {
1941 		struct safe_ringentry *re = sc->sc_back;
1942 		do {
1943 			safe_dump_request(sc, tag, re);
1944 			if (++re == sc->sc_ringtop)
1945 				re = sc->sc_ring;
1946 		} while (re != sc->sc_front);
1947 	}
1948 	mtx_unlock(&sc->sc_ringmtx);
1949 }
1950 
1951 static int
1952 sysctl_hw_safe_dump(SYSCTL_HANDLER_ARGS)
1953 {
1954 	char dmode[64];
1955 	int error;
1956 
1957 	strncpy(dmode, "", sizeof(dmode) - 1);
1958 	dmode[sizeof(dmode) - 1] = '\0';
1959 	error = sysctl_handle_string(oidp, &dmode[0], sizeof(dmode), req);
1960 
1961 	if (error == 0 && req->newptr != NULL) {
1962 		struct safe_softc *sc = safec;
1963 
1964 		if (!sc)
1965 			return EINVAL;
1966 		if (strncmp(dmode, "dma", 3) == 0)
1967 			safe_dump_dmastatus(sc, "safe0");
1968 		else if (strncmp(dmode, "int", 3) == 0)
1969 			safe_dump_intrstate(sc, "safe0");
1970 		else if (strncmp(dmode, "ring", 4) == 0)
1971 			safe_dump_ring(sc, "safe0");
1972 		else
1973 			return EINVAL;
1974 	}
1975 	return error;
1976 }
1977 SYSCTL_PROC(_hw_safe, OID_AUTO, dump,
1978     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, 0,
1979     sysctl_hw_safe_dump, "A",
1980     "Dump driver state");
1981 #endif /* SAFE_DEBUG */
1982