xref: /freebsd/sys/dev/safe/safe.c (revision 5036d9652a5701d00e9e40ea942c278e9f77d33d)
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 /*
32  * SafeNet SafeXcel-1141 hardware crypto accelerator
33  */
34 #include "opt_safe.h"
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/proc.h>
39 #include <sys/errno.h>
40 #include <sys/malloc.h>
41 #include <sys/kernel.h>
42 #include <sys/mbuf.h>
43 #include <sys/module.h>
44 #include <sys/lock.h>
45 #include <sys/mutex.h>
46 #include <sys/sysctl.h>
47 #include <sys/endian.h>
48 #include <sys/uio.h>
49 
50 #include <vm/vm.h>
51 #include <vm/pmap.h>
52 
53 #include <machine/bus.h>
54 #include <machine/resource.h>
55 #include <sys/bus.h>
56 #include <sys/rman.h>
57 
58 #include <opencrypto/cryptodev.h>
59 #include <opencrypto/xform_auth.h>
60 #include <sys/random.h>
61 #include <sys/kobj.h>
62 
63 #include "cryptodev_if.h"
64 
65 #include <dev/pci/pcivar.h>
66 #include <dev/pci/pcireg.h>
67 
68 #ifdef SAFE_RNDTEST
69 #include <dev/rndtest/rndtest.h>
70 #endif
71 #include <dev/safe/safereg.h>
72 #include <dev/safe/safevar.h>
73 
74 #ifndef bswap32
75 #define	bswap32	NTOHL
76 #endif
77 
78 /*
79  * Prototypes and count for the pci_device structure
80  */
81 static	int safe_probe(device_t);
82 static	int safe_attach(device_t);
83 static	int safe_detach(device_t);
84 static	int safe_suspend(device_t);
85 static	int safe_resume(device_t);
86 static	int safe_shutdown(device_t);
87 
88 static	int safe_probesession(device_t, const struct crypto_session_params *);
89 static	int safe_newsession(device_t, crypto_session_t,
90 	    const struct crypto_session_params *);
91 static	int safe_process(device_t, struct cryptop *, int);
92 
93 static device_method_t safe_methods[] = {
94 	/* Device interface */
95 	DEVMETHOD(device_probe,		safe_probe),
96 	DEVMETHOD(device_attach,	safe_attach),
97 	DEVMETHOD(device_detach,	safe_detach),
98 	DEVMETHOD(device_suspend,	safe_suspend),
99 	DEVMETHOD(device_resume,	safe_resume),
100 	DEVMETHOD(device_shutdown,	safe_shutdown),
101 
102 	/* crypto device methods */
103 	DEVMETHOD(cryptodev_probesession, safe_probesession),
104 	DEVMETHOD(cryptodev_newsession,	safe_newsession),
105 	DEVMETHOD(cryptodev_process,	safe_process),
106 
107 	DEVMETHOD_END
108 };
109 
110 static driver_t safe_driver = {
111 	"safe",
112 	safe_methods,
113 	sizeof (struct safe_softc)
114 };
115 
116 DRIVER_MODULE(safe, pci, safe_driver, 0, 0);
117 MODULE_DEPEND(safe, crypto, 1, 1, 1);
118 #ifdef SAFE_RNDTEST
119 MODULE_DEPEND(safe, rndtest, 1, 1, 1);
120 #endif
121 
122 static	void safe_intr(void *);
123 static	void safe_callback(struct safe_softc *, struct safe_ringentry *);
124 static	void safe_feed(struct safe_softc *, struct safe_ringentry *);
125 static	void safe_mcopy(struct mbuf *, struct mbuf *, u_int);
126 #ifndef SAFE_NO_RNG
127 static	void safe_rng_init(struct safe_softc *);
128 static	void safe_rng(void *);
129 #endif /* SAFE_NO_RNG */
130 static	int safe_dma_malloc(struct safe_softc *, bus_size_t,
131 	        struct safe_dma_alloc *, int);
132 #define	safe_dma_sync(_dma, _flags) \
133 	bus_dmamap_sync((_dma)->dma_tag, (_dma)->dma_map, (_flags))
134 static	void safe_dma_free(struct safe_softc *, struct safe_dma_alloc *);
135 static	int safe_dmamap_aligned(const struct safe_operand *);
136 static	int safe_dmamap_uniform(const struct safe_operand *);
137 
138 static	void safe_reset_board(struct safe_softc *);
139 static	void safe_init_board(struct safe_softc *);
140 static	void safe_init_pciregs(device_t dev);
141 static	void safe_cleanchip(struct safe_softc *);
142 static	void safe_totalreset(struct safe_softc *);
143 
144 static	int safe_free_entry(struct safe_softc *, struct safe_ringentry *);
145 
146 static SYSCTL_NODE(_hw, OID_AUTO, safe, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
147     "SafeNet driver parameters");
148 
149 #ifdef SAFE_DEBUG
150 static	void safe_dump_dmastatus(struct safe_softc *, const char *);
151 static	void safe_dump_ringstate(struct safe_softc *, const char *);
152 static	void safe_dump_intrstate(struct safe_softc *, const char *);
153 static	void safe_dump_request(struct safe_softc *, const char *,
154 		struct safe_ringentry *);
155 
156 static	struct safe_softc *safec;		/* for use by hw.safe.dump */
157 
158 static	int safe_debug = 0;
159 SYSCTL_INT(_hw_safe, OID_AUTO, debug, CTLFLAG_RW, &safe_debug,
160 	    0, "control debugging msgs");
161 #define	DPRINTF(_x)	if (safe_debug) printf _x
162 #else
163 #define	DPRINTF(_x)
164 #endif
165 
166 #define	READ_REG(sc,r) \
167 	bus_space_read_4((sc)->sc_st, (sc)->sc_sh, (r))
168 
169 #define WRITE_REG(sc,reg,val) \
170 	bus_space_write_4((sc)->sc_st, (sc)->sc_sh, reg, val)
171 
172 struct safe_stats safestats;
173 SYSCTL_STRUCT(_hw_safe, OID_AUTO, stats, CTLFLAG_RD, &safestats,
174 	    safe_stats, "driver statistics");
175 #ifndef SAFE_NO_RNG
176 static	int safe_rnginterval = 1;		/* poll once a second */
177 SYSCTL_INT(_hw_safe, OID_AUTO, rnginterval, CTLFLAG_RW, &safe_rnginterval,
178 	    0, "RNG polling interval (secs)");
179 static	int safe_rngbufsize = 16;		/* 64 bytes each poll  */
180 SYSCTL_INT(_hw_safe, OID_AUTO, rngbufsize, CTLFLAG_RW, &safe_rngbufsize,
181 	    0, "RNG polling buffer size (32-bit words)");
182 static	int safe_rngmaxalarm = 8;		/* max alarms before reset */
183 SYSCTL_INT(_hw_safe, OID_AUTO, rngmaxalarm, CTLFLAG_RW, &safe_rngmaxalarm,
184 	    0, "RNG max alarms before reset");
185 #endif /* SAFE_NO_RNG */
186 
187 static int
188 safe_probe(device_t dev)
189 {
190 	if (pci_get_vendor(dev) == PCI_VENDOR_SAFENET &&
191 	    pci_get_device(dev) == PCI_PRODUCT_SAFEXCEL)
192 		return (BUS_PROBE_DEFAULT);
193 	return (ENXIO);
194 }
195 
196 static const char*
197 safe_partname(struct safe_softc *sc)
198 {
199 	/* XXX sprintf numbers when not decoded */
200 	switch (pci_get_vendor(sc->sc_dev)) {
201 	case PCI_VENDOR_SAFENET:
202 		switch (pci_get_device(sc->sc_dev)) {
203 		case PCI_PRODUCT_SAFEXCEL: return "SafeNet SafeXcel-1141";
204 		}
205 		return "SafeNet unknown-part";
206 	}
207 	return "Unknown-vendor unknown-part";
208 }
209 
210 #ifndef SAFE_NO_RNG
211 static void
212 default_harvest(struct rndtest_state *rsp, void *buf, u_int count)
213 {
214 	/* MarkM: FIX!! Check that this does not swamp the harvester! */
215 	random_harvest_queue(buf, count, RANDOM_PURE_SAFE);
216 }
217 #endif /* SAFE_NO_RNG */
218 
219 static int
220 safe_attach(device_t dev)
221 {
222 	struct safe_softc *sc = device_get_softc(dev);
223 	u_int32_t raddr;
224 	u_int32_t i;
225 	int rid;
226 
227 	bzero(sc, sizeof (*sc));
228 	sc->sc_dev = dev;
229 
230 	/* XXX handle power management */
231 
232 	pci_enable_busmaster(dev);
233 
234 	/*
235 	 * Setup memory-mapping of PCI registers.
236 	 */
237 	rid = BS_BAR;
238 	sc->sc_sr = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
239 					   RF_ACTIVE);
240 	if (sc->sc_sr == NULL) {
241 		device_printf(dev, "cannot map register space\n");
242 		goto bad;
243 	}
244 	sc->sc_st = rman_get_bustag(sc->sc_sr);
245 	sc->sc_sh = rman_get_bushandle(sc->sc_sr);
246 
247 	/*
248 	 * Arrange interrupt line.
249 	 */
250 	rid = 0;
251 	sc->sc_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
252 					    RF_SHAREABLE|RF_ACTIVE);
253 	if (sc->sc_irq == NULL) {
254 		device_printf(dev, "could not map interrupt\n");
255 		goto bad1;
256 	}
257 	/*
258 	 * NB: Network code assumes we are blocked with splimp()
259 	 *     so make sure the IRQ is mapped appropriately.
260 	 */
261 	if (bus_setup_intr(dev, sc->sc_irq, INTR_TYPE_NET | INTR_MPSAFE,
262 			   NULL, safe_intr, sc, &sc->sc_ih)) {
263 		device_printf(dev, "could not establish interrupt\n");
264 		goto bad2;
265 	}
266 
267 	sc->sc_cid = crypto_get_driverid(dev, sizeof(struct safe_session),
268 	    CRYPTOCAP_F_HARDWARE);
269 	if (sc->sc_cid < 0) {
270 		device_printf(dev, "could not get crypto driver id\n");
271 		goto bad3;
272 	}
273 
274 	sc->sc_chiprev = READ_REG(sc, SAFE_DEVINFO) &
275 		(SAFE_DEVINFO_REV_MAJ | SAFE_DEVINFO_REV_MIN);
276 
277 	/*
278 	 * Setup DMA descriptor area.
279 	 */
280 	if (bus_dma_tag_create(bus_get_dma_tag(dev),	/* parent */
281 			       1,			/* alignment */
282 			       SAFE_DMA_BOUNDARY,	/* boundary */
283 			       BUS_SPACE_MAXADDR_32BIT,	/* lowaddr */
284 			       BUS_SPACE_MAXADDR,	/* highaddr */
285 			       NULL, NULL,		/* filter, filterarg */
286 			       SAFE_MAX_DMA,		/* maxsize */
287 			       SAFE_MAX_PART,		/* nsegments */
288 			       SAFE_MAX_SSIZE,		/* maxsegsize */
289 			       BUS_DMA_ALLOCNOW,	/* flags */
290 			       NULL, NULL,		/* locking */
291 			       &sc->sc_srcdmat)) {
292 		device_printf(dev, "cannot allocate DMA tag\n");
293 		goto bad4;
294 	}
295 	if (bus_dma_tag_create(bus_get_dma_tag(dev),	/* parent */
296 			       1,			/* alignment */
297 			       SAFE_MAX_DSIZE,		/* boundary */
298 			       BUS_SPACE_MAXADDR_32BIT,	/* lowaddr */
299 			       BUS_SPACE_MAXADDR,	/* highaddr */
300 			       NULL, NULL,		/* filter, filterarg */
301 			       SAFE_MAX_DMA,		/* maxsize */
302 			       SAFE_MAX_PART,		/* nsegments */
303 			       SAFE_MAX_DSIZE,		/* maxsegsize */
304 			       BUS_DMA_ALLOCNOW,	/* flags */
305 			       NULL, NULL,		/* locking */
306 			       &sc->sc_dstdmat)) {
307 		device_printf(dev, "cannot allocate DMA tag\n");
308 		goto bad4;
309 	}
310 
311 	/*
312 	 * Allocate packet engine descriptors.
313 	 */
314 	if (safe_dma_malloc(sc,
315 	    SAFE_MAX_NQUEUE * sizeof (struct safe_ringentry),
316 	    &sc->sc_ringalloc, 0)) {
317 		device_printf(dev, "cannot allocate PE descriptor ring\n");
318 		bus_dma_tag_destroy(sc->sc_srcdmat);
319 		goto bad4;
320 	}
321 	/*
322 	 * Hookup the static portion of all our data structures.
323 	 */
324 	sc->sc_ring = (struct safe_ringentry *) sc->sc_ringalloc.dma_vaddr;
325 	sc->sc_ringtop = sc->sc_ring + SAFE_MAX_NQUEUE;
326 	sc->sc_front = sc->sc_ring;
327 	sc->sc_back = sc->sc_ring;
328 	raddr = sc->sc_ringalloc.dma_paddr;
329 	bzero(sc->sc_ring, SAFE_MAX_NQUEUE * sizeof(struct safe_ringentry));
330 	for (i = 0; i < SAFE_MAX_NQUEUE; i++) {
331 		struct safe_ringentry *re = &sc->sc_ring[i];
332 
333 		re->re_desc.d_sa = raddr +
334 			offsetof(struct safe_ringentry, re_sa);
335 		re->re_sa.sa_staterec = raddr +
336 			offsetof(struct safe_ringentry, re_sastate);
337 
338 		raddr += sizeof (struct safe_ringentry);
339 	}
340 	mtx_init(&sc->sc_ringmtx, device_get_nameunit(dev),
341 		"packet engine ring", MTX_DEF);
342 
343 	/*
344 	 * Allocate scatter and gather particle descriptors.
345 	 */
346 	if (safe_dma_malloc(sc, SAFE_TOTAL_SPART * sizeof (struct safe_pdesc),
347 	    &sc->sc_spalloc, 0)) {
348 		device_printf(dev, "cannot allocate source particle "
349 			"descriptor ring\n");
350 		mtx_destroy(&sc->sc_ringmtx);
351 		safe_dma_free(sc, &sc->sc_ringalloc);
352 		bus_dma_tag_destroy(sc->sc_srcdmat);
353 		goto bad4;
354 	}
355 	sc->sc_spring = (struct safe_pdesc *) sc->sc_spalloc.dma_vaddr;
356 	sc->sc_springtop = sc->sc_spring + SAFE_TOTAL_SPART;
357 	sc->sc_spfree = sc->sc_spring;
358 	bzero(sc->sc_spring, SAFE_TOTAL_SPART * sizeof(struct safe_pdesc));
359 
360 	if (safe_dma_malloc(sc, SAFE_TOTAL_DPART * sizeof (struct safe_pdesc),
361 	    &sc->sc_dpalloc, 0)) {
362 		device_printf(dev, "cannot allocate destination particle "
363 			"descriptor ring\n");
364 		mtx_destroy(&sc->sc_ringmtx);
365 		safe_dma_free(sc, &sc->sc_spalloc);
366 		safe_dma_free(sc, &sc->sc_ringalloc);
367 		bus_dma_tag_destroy(sc->sc_dstdmat);
368 		goto bad4;
369 	}
370 	sc->sc_dpring = (struct safe_pdesc *) sc->sc_dpalloc.dma_vaddr;
371 	sc->sc_dpringtop = sc->sc_dpring + SAFE_TOTAL_DPART;
372 	sc->sc_dpfree = sc->sc_dpring;
373 	bzero(sc->sc_dpring, SAFE_TOTAL_DPART * sizeof(struct safe_pdesc));
374 
375 	device_printf(sc->sc_dev, "%s", safe_partname(sc));
376 
377 	sc->sc_devinfo = READ_REG(sc, SAFE_DEVINFO);
378 	if (sc->sc_devinfo & SAFE_DEVINFO_RNG) {
379 		sc->sc_flags |= SAFE_FLAGS_RNG;
380 		printf(" rng");
381 	}
382 	if (sc->sc_devinfo & SAFE_DEVINFO_PKEY) {
383 #if 0
384 		printf(" key");
385 		sc->sc_flags |= SAFE_FLAGS_KEY;
386 #endif
387 	}
388 	if (sc->sc_devinfo & SAFE_DEVINFO_DES) {
389 		printf(" des/3des");
390 	}
391 	if (sc->sc_devinfo & SAFE_DEVINFO_AES) {
392 		printf(" aes");
393 	}
394 	if (sc->sc_devinfo & SAFE_DEVINFO_MD5) {
395 		printf(" md5");
396 	}
397 	if (sc->sc_devinfo & SAFE_DEVINFO_SHA1) {
398 		printf(" sha1");
399 	}
400 	/* XXX other supported algorithms */
401 	printf("\n");
402 
403 	safe_reset_board(sc);		/* reset h/w */
404 	safe_init_pciregs(dev);		/* init pci settings */
405 	safe_init_board(sc);		/* init h/w */
406 
407 #ifndef SAFE_NO_RNG
408 	if (sc->sc_flags & SAFE_FLAGS_RNG) {
409 #ifdef SAFE_RNDTEST
410 		sc->sc_rndtest = rndtest_attach(dev);
411 		if (sc->sc_rndtest)
412 			sc->sc_harvest = rndtest_harvest;
413 		else
414 			sc->sc_harvest = default_harvest;
415 #else
416 		sc->sc_harvest = default_harvest;
417 #endif
418 		safe_rng_init(sc);
419 
420 		callout_init(&sc->sc_rngto, 1);
421 		callout_reset(&sc->sc_rngto, hz*safe_rnginterval, safe_rng, sc);
422 	}
423 #endif /* SAFE_NO_RNG */
424 #ifdef SAFE_DEBUG
425 	safec = sc;			/* for use by hw.safe.dump */
426 #endif
427 	return (0);
428 bad4:
429 	crypto_unregister_all(sc->sc_cid);
430 bad3:
431 	bus_teardown_intr(dev, sc->sc_irq, sc->sc_ih);
432 bad2:
433 	bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq);
434 bad1:
435 	bus_release_resource(dev, SYS_RES_MEMORY, BS_BAR, sc->sc_sr);
436 bad:
437 	return (ENXIO);
438 }
439 
440 /*
441  * Detach a device that successfully probed.
442  */
443 static int
444 safe_detach(device_t dev)
445 {
446 	struct safe_softc *sc = device_get_softc(dev);
447 
448 	/* XXX wait/abort active ops */
449 
450 	WRITE_REG(sc, SAFE_HI_MASK, 0);		/* disable interrupts */
451 
452 	callout_stop(&sc->sc_rngto);
453 
454 	crypto_unregister_all(sc->sc_cid);
455 
456 #ifdef SAFE_RNDTEST
457 	if (sc->sc_rndtest)
458 		rndtest_detach(sc->sc_rndtest);
459 #endif
460 
461 	safe_cleanchip(sc);
462 	safe_dma_free(sc, &sc->sc_dpalloc);
463 	safe_dma_free(sc, &sc->sc_spalloc);
464 	mtx_destroy(&sc->sc_ringmtx);
465 	safe_dma_free(sc, &sc->sc_ringalloc);
466 
467 	bus_teardown_intr(dev, sc->sc_irq, sc->sc_ih);
468 	bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq);
469 
470 	bus_dma_tag_destroy(sc->sc_srcdmat);
471 	bus_dma_tag_destroy(sc->sc_dstdmat);
472 	bus_release_resource(dev, SYS_RES_MEMORY, BS_BAR, sc->sc_sr);
473 
474 	return (0);
475 }
476 
477 /*
478  * Stop all chip i/o so that the kernel's probe routines don't
479  * get confused by errant DMAs when rebooting.
480  */
481 static int
482 safe_shutdown(device_t dev)
483 {
484 #ifdef notyet
485 	safe_stop(device_get_softc(dev));
486 #endif
487 	return (0);
488 }
489 
490 /*
491  * Device suspend routine.
492  */
493 static int
494 safe_suspend(device_t dev)
495 {
496 	struct safe_softc *sc = device_get_softc(dev);
497 
498 #ifdef notyet
499 	/* XXX stop the device and save PCI settings */
500 #endif
501 	sc->sc_suspended = 1;
502 
503 	return (0);
504 }
505 
506 static int
507 safe_resume(device_t dev)
508 {
509 	struct safe_softc *sc = device_get_softc(dev);
510 
511 #ifdef notyet
512 	/* XXX retore PCI settings and start the device */
513 #endif
514 	sc->sc_suspended = 0;
515 	return (0);
516 }
517 
518 /*
519  * SafeXcel Interrupt routine
520  */
521 static void
522 safe_intr(void *arg)
523 {
524 	struct safe_softc *sc = arg;
525 	volatile u_int32_t stat;
526 
527 	stat = READ_REG(sc, SAFE_HM_STAT);
528 	if (stat == 0)			/* shared irq, not for us */
529 		return;
530 
531 	WRITE_REG(sc, SAFE_HI_CLR, stat);	/* IACK */
532 
533 	if ((stat & SAFE_INT_PE_DDONE)) {
534 		/*
535 		 * Descriptor(s) done; scan the ring and
536 		 * process completed operations.
537 		 */
538 		mtx_lock(&sc->sc_ringmtx);
539 		while (sc->sc_back != sc->sc_front) {
540 			struct safe_ringentry *re = sc->sc_back;
541 #ifdef SAFE_DEBUG
542 			if (safe_debug) {
543 				safe_dump_ringstate(sc, __func__);
544 				safe_dump_request(sc, __func__, re);
545 			}
546 #endif
547 			/*
548 			 * safe_process marks ring entries that were allocated
549 			 * but not used with a csr of zero.  This insures the
550 			 * ring front pointer never needs to be set backwards
551 			 * in the event that an entry is allocated but not used
552 			 * because of a setup error.
553 			 */
554 			if (re->re_desc.d_csr != 0) {
555 				if (!SAFE_PE_CSR_IS_DONE(re->re_desc.d_csr))
556 					break;
557 				if (!SAFE_PE_LEN_IS_DONE(re->re_desc.d_len))
558 					break;
559 				sc->sc_nqchip--;
560 				safe_callback(sc, re);
561 			}
562 			if (++(sc->sc_back) == sc->sc_ringtop)
563 				sc->sc_back = sc->sc_ring;
564 		}
565 		mtx_unlock(&sc->sc_ringmtx);
566 	}
567 
568 	/*
569 	 * Check to see if we got any DMA Error
570 	 */
571 	if (stat & SAFE_INT_PE_ERROR) {
572 		DPRINTF(("dmaerr dmastat %08x\n",
573 			READ_REG(sc, SAFE_PE_DMASTAT)));
574 		safestats.st_dmaerr++;
575 		safe_totalreset(sc);
576 #if 0
577 		safe_feed(sc);
578 #endif
579 	}
580 
581 	if (sc->sc_needwakeup) {		/* XXX check high watermark */
582 		int wakeup = sc->sc_needwakeup & CRYPTO_SYMQ;
583 		DPRINTF(("%s: wakeup crypto %x\n", __func__,
584 			sc->sc_needwakeup));
585 		sc->sc_needwakeup &= ~wakeup;
586 		crypto_unblock(sc->sc_cid, wakeup);
587 	}
588 }
589 
590 /*
591  * safe_feed() - post a request to chip
592  */
593 static void
594 safe_feed(struct safe_softc *sc, struct safe_ringentry *re)
595 {
596 	bus_dmamap_sync(sc->sc_srcdmat, re->re_src_map, BUS_DMASYNC_PREWRITE);
597 	if (re->re_dst_map != NULL)
598 		bus_dmamap_sync(sc->sc_dstdmat, re->re_dst_map,
599 			BUS_DMASYNC_PREREAD);
600 	/* XXX have no smaller granularity */
601 	safe_dma_sync(&sc->sc_ringalloc,
602 		BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
603 	safe_dma_sync(&sc->sc_spalloc, BUS_DMASYNC_PREWRITE);
604 	safe_dma_sync(&sc->sc_dpalloc, BUS_DMASYNC_PREWRITE);
605 
606 #ifdef SAFE_DEBUG
607 	if (safe_debug) {
608 		safe_dump_ringstate(sc, __func__);
609 		safe_dump_request(sc, __func__, re);
610 	}
611 #endif
612 	sc->sc_nqchip++;
613 	if (sc->sc_nqchip > safestats.st_maxqchip)
614 		safestats.st_maxqchip = sc->sc_nqchip;
615 	/* poke h/w to check descriptor ring, any value can be written */
616 	WRITE_REG(sc, SAFE_HI_RD_DESCR, 0);
617 }
618 
619 #define	N(a)	(sizeof(a) / sizeof (a[0]))
620 static void
621 safe_setup_enckey(struct safe_session *ses, const void *key)
622 {
623 	int i;
624 
625 	bcopy(key, ses->ses_key, ses->ses_klen);
626 
627 	/* PE is little-endian, insure proper byte order */
628 	for (i = 0; i < N(ses->ses_key); i++)
629 		ses->ses_key[i] = htole32(ses->ses_key[i]);
630 }
631 
632 static void
633 safe_setup_mackey(struct safe_session *ses, int algo, const uint8_t *key,
634     int klen)
635 {
636 	SHA1_CTX sha1ctx;
637 	int i;
638 
639 	hmac_init_ipad(&auth_hash_hmac_sha1, key, klen, &sha1ctx);
640 	bcopy(sha1ctx.h.b32, ses->ses_hminner, sizeof(sha1ctx.h.b32));
641 
642 	hmac_init_opad(&auth_hash_hmac_sha1, key, klen, &sha1ctx);
643 	bcopy(sha1ctx.h.b32, ses->ses_hmouter, sizeof(sha1ctx.h.b32));
644 
645 	explicit_bzero(&sha1ctx, sizeof(sha1ctx));
646 
647 	/* PE is little-endian, insure proper byte order */
648 	for (i = 0; i < N(ses->ses_hminner); i++) {
649 		ses->ses_hminner[i] = htole32(ses->ses_hminner[i]);
650 		ses->ses_hmouter[i] = htole32(ses->ses_hmouter[i]);
651 	}
652 }
653 #undef N
654 
655 static bool
656 safe_auth_supported(struct safe_softc *sc,
657     const struct crypto_session_params *csp)
658 {
659 
660 	switch (csp->csp_auth_alg) {
661 	case CRYPTO_SHA1_HMAC:
662 		if ((sc->sc_devinfo & SAFE_DEVINFO_SHA1) == 0)
663 			return (false);
664 		break;
665 	default:
666 		return (false);
667 	}
668 	return (true);
669 }
670 
671 static bool
672 safe_cipher_supported(struct safe_softc *sc,
673     const struct crypto_session_params *csp)
674 {
675 
676 	switch (csp->csp_cipher_alg) {
677 	case CRYPTO_AES_CBC:
678 		if ((sc->sc_devinfo & SAFE_DEVINFO_AES) == 0)
679 			return (false);
680 		if (csp->csp_ivlen != 16)
681 			return (false);
682 		if (csp->csp_cipher_klen != 16 &&
683 		    csp->csp_cipher_klen != 24 &&
684 		    csp->csp_cipher_klen != 32)
685 			return (false);
686 		break;
687 	}
688 	return (true);
689 }
690 
691 static int
692 safe_probesession(device_t dev, const struct crypto_session_params *csp)
693 {
694 	struct safe_softc *sc = device_get_softc(dev);
695 
696 	if (csp->csp_flags != 0)
697 		return (EINVAL);
698 	switch (csp->csp_mode) {
699 	case CSP_MODE_DIGEST:
700 		if (!safe_auth_supported(sc, csp))
701 			return (EINVAL);
702 		break;
703 	case CSP_MODE_CIPHER:
704 		if (!safe_cipher_supported(sc, csp))
705 			return (EINVAL);
706 		break;
707 	case CSP_MODE_ETA:
708 		if (!safe_auth_supported(sc, csp) ||
709 		    !safe_cipher_supported(sc, csp))
710 			return (EINVAL);
711 		break;
712 	default:
713 		return (EINVAL);
714 	}
715 
716 	return (CRYPTODEV_PROBE_HARDWARE);
717 }
718 
719 /*
720  * Allocate a new 'session'.
721  */
722 static int
723 safe_newsession(device_t dev, crypto_session_t cses,
724     const struct crypto_session_params *csp)
725 {
726 	struct safe_session *ses;
727 
728 	ses = crypto_get_driver_session(cses);
729 	if (csp->csp_cipher_alg != 0) {
730 		ses->ses_klen = csp->csp_cipher_klen;
731 		if (csp->csp_cipher_key != NULL)
732 			safe_setup_enckey(ses, csp->csp_cipher_key);
733 	}
734 
735 	if (csp->csp_auth_alg != 0) {
736 		ses->ses_mlen = csp->csp_auth_mlen;
737 		if (ses->ses_mlen == 0) {
738 			ses->ses_mlen = SHA1_HASH_LEN;
739 		}
740 
741 		if (csp->csp_auth_key != NULL) {
742 			safe_setup_mackey(ses, csp->csp_auth_alg,
743 			    csp->csp_auth_key, csp->csp_auth_klen);
744 		}
745 	}
746 
747 	return (0);
748 }
749 
750 static void
751 safe_op_cb(void *arg, bus_dma_segment_t *seg, int nsegs, int error)
752 {
753 	struct safe_operand *op = arg;
754 
755 	DPRINTF(("%s: nsegs %d error %d\n", __func__,
756 		nsegs, error));
757 	if (error != 0)
758 		return;
759 	op->nsegs = nsegs;
760 	bcopy(seg, op->segs, nsegs * sizeof (seg[0]));
761 }
762 
763 static int
764 safe_process(device_t dev, struct cryptop *crp, int hint)
765 {
766 	struct safe_softc *sc = device_get_softc(dev);
767 	const struct crypto_session_params *csp;
768 	int err = 0, i, nicealign, uniform;
769 	int bypass, oplen;
770 	int16_t coffset;
771 	struct safe_session *ses;
772 	struct safe_ringentry *re;
773 	struct safe_sarec *sa;
774 	struct safe_pdesc *pd;
775 	u_int32_t cmd0, cmd1, staterec;
776 
777 	mtx_lock(&sc->sc_ringmtx);
778 	if (sc->sc_front == sc->sc_back && sc->sc_nqchip != 0) {
779 		safestats.st_ringfull++;
780 		sc->sc_needwakeup |= CRYPTO_SYMQ;
781 		mtx_unlock(&sc->sc_ringmtx);
782 		return (ERESTART);
783 	}
784 	re = sc->sc_front;
785 
786 	staterec = re->re_sa.sa_staterec;	/* save */
787 	/* NB: zero everything but the PE descriptor */
788 	bzero(&re->re_sa, sizeof(struct safe_ringentry) - sizeof(re->re_desc));
789 	re->re_sa.sa_staterec = staterec;	/* restore */
790 
791 	re->re_crp = crp;
792 
793 	sa = &re->re_sa;
794 	ses = crypto_get_driver_session(crp->crp_session);
795 	csp = crypto_get_params(crp->crp_session);
796 
797 	cmd0 = SAFE_SA_CMD0_BASIC;		/* basic group operation */
798 	cmd1 = 0;
799 	switch (csp->csp_mode) {
800 	case CSP_MODE_DIGEST:
801 		cmd0 |= SAFE_SA_CMD0_OP_HASH;
802 		break;
803 	case CSP_MODE_CIPHER:
804 		cmd0 |= SAFE_SA_CMD0_OP_CRYPT;
805 		break;
806 	case CSP_MODE_ETA:
807 		cmd0 |= SAFE_SA_CMD0_OP_BOTH;
808 		break;
809 	}
810 
811 	if (csp->csp_cipher_alg != 0) {
812 		if (crp->crp_cipher_key != NULL)
813 			safe_setup_enckey(ses, crp->crp_cipher_key);
814 
815 		switch (csp->csp_cipher_alg) {
816 		case CRYPTO_AES_CBC:
817 			cmd0 |= SAFE_SA_CMD0_AES;
818 			cmd1 |= SAFE_SA_CMD1_CBC;
819 			if (ses->ses_klen * 8 == 128)
820 			     cmd1 |=  SAFE_SA_CMD1_AES128;
821 			else if (ses->ses_klen * 8 == 192)
822 			     cmd1 |=  SAFE_SA_CMD1_AES192;
823 			else
824 			     cmd1 |=  SAFE_SA_CMD1_AES256;
825 		}
826 
827 		/*
828 		 * Setup encrypt/decrypt state.  When using basic ops
829 		 * we can't use an inline IV because hash/crypt offset
830 		 * must be from the end of the IV to the start of the
831 		 * crypt data and this leaves out the preceding header
832 		 * from the hash calculation.  Instead we place the IV
833 		 * in the state record and set the hash/crypt offset to
834 		 * copy both the header+IV.
835 		 */
836 		crypto_read_iv(crp, re->re_sastate.sa_saved_iv);
837 		cmd0 |= SAFE_SA_CMD0_IVLD_STATE;
838 
839 		if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op)) {
840 			cmd0 |= SAFE_SA_CMD0_OUTBOUND;
841 
842 			/*
843 			 * XXX: I suspect we don't need this since we
844 			 * don't save the returned IV.
845 			 */
846 			cmd0 |= SAFE_SA_CMD0_SAVEIV;
847 		} else {
848 			cmd0 |= SAFE_SA_CMD0_INBOUND;
849 		}
850 		/*
851 		 * For basic encryption use the zero pad algorithm.
852 		 * This pads results to an 8-byte boundary and
853 		 * suppresses padding verification for inbound (i.e.
854 		 * decrypt) operations.
855 		 *
856 		 * NB: Not sure if the 8-byte pad boundary is a problem.
857 		 */
858 		cmd0 |= SAFE_SA_CMD0_PAD_ZERO;
859 
860 		/* XXX assert key bufs have the same size */
861 		bcopy(ses->ses_key, sa->sa_key, sizeof(sa->sa_key));
862 	}
863 
864 	if (csp->csp_auth_alg != 0) {
865 		if (crp->crp_auth_key != NULL) {
866 			safe_setup_mackey(ses, csp->csp_auth_alg,
867 			    crp->crp_auth_key, csp->csp_auth_klen);
868 		}
869 
870 		switch (csp->csp_auth_alg) {
871 		case CRYPTO_SHA1_HMAC:
872 			cmd0 |= SAFE_SA_CMD0_SHA1;
873 			cmd1 |= SAFE_SA_CMD1_HMAC;	/* NB: enable HMAC */
874 			break;
875 		}
876 
877 		/*
878 		 * Digest data is loaded from the SA and the hash
879 		 * result is saved to the state block where we
880 		 * retrieve it for return to the caller.
881 		 */
882 		/* XXX assert digest bufs have the same size */
883 		bcopy(ses->ses_hminner, sa->sa_indigest,
884 			sizeof(sa->sa_indigest));
885 		bcopy(ses->ses_hmouter, sa->sa_outdigest,
886 			sizeof(sa->sa_outdigest));
887 
888 		cmd0 |= SAFE_SA_CMD0_HSLD_SA | SAFE_SA_CMD0_SAVEHASH;
889 		re->re_flags |= SAFE_QFLAGS_COPYOUTICV;
890 	}
891 
892 	if (csp->csp_mode == CSP_MODE_ETA) {
893 		/*
894 		 * The driver only supports ETA requests where there
895 		 * is no gap between the AAD and payload.
896 		 */
897 		if (crp->crp_aad_length != 0 &&
898 		    crp->crp_aad_start + crp->crp_aad_length !=
899 		    crp->crp_payload_start) {
900 			safestats.st_lenmismatch++;
901 			err = EINVAL;
902 			goto errout;
903 		}
904 		if (crp->crp_aad_length != 0)
905 			bypass = crp->crp_aad_start;
906 		else
907 			bypass = crp->crp_payload_start;
908 		coffset = crp->crp_aad_length;
909 		oplen = crp->crp_payload_start + crp->crp_payload_length;
910 #ifdef SAFE_DEBUG
911 		if (safe_debug) {
912 			printf("AAD: skip %d, len %d, digest %d\n",
913 			    crp->crp_aad_start, crp->crp_aad_length,
914 			    crp->crp_digest_start);
915 			printf("payload: skip %d, len %d, IV %d\n",
916 			    crp->crp_payload_start, crp->crp_payload_length,
917 			    crp->crp_iv_start);
918 			printf("bypass %d coffset %d oplen %d\n",
919 				bypass, coffset, oplen);
920 		}
921 #endif
922 		if (coffset & 3) {	/* offset must be 32-bit aligned */
923 			DPRINTF(("%s: coffset %u misaligned\n",
924 				__func__, coffset));
925 			safestats.st_coffmisaligned++;
926 			err = EINVAL;
927 			goto errout;
928 		}
929 		coffset >>= 2;
930 		if (coffset > 255) {	/* offset must be <256 dwords */
931 			DPRINTF(("%s: coffset %u too big\n",
932 				__func__, coffset));
933 			safestats.st_cofftoobig++;
934 			err = EINVAL;
935 			goto errout;
936 		}
937 		/*
938 		 * Tell the hardware to copy the header to the output.
939 		 * The header is defined as the data from the end of
940 		 * the bypass to the start of data to be encrypted.
941 		 * Typically this is the inline IV.  Note that you need
942 		 * to do this even if src+dst are the same; it appears
943 		 * that w/o this bit the crypted data is written
944 		 * immediately after the bypass data.
945 		 */
946 		cmd1 |= SAFE_SA_CMD1_HDRCOPY;
947 		/*
948 		 * Disable IP header mutable bit handling.  This is
949 		 * needed to get correct HMAC calculations.
950 		 */
951 		cmd1 |= SAFE_SA_CMD1_MUTABLE;
952 	} else {
953 		bypass = crp->crp_payload_start;
954 		oplen = bypass + crp->crp_payload_length;
955 		coffset = 0;
956 	}
957 	/* XXX verify multiple of 4 when using s/g */
958 	if (bypass > 96) {		/* bypass offset must be <= 96 bytes */
959 		DPRINTF(("%s: bypass %u too big\n", __func__, bypass));
960 		safestats.st_bypasstoobig++;
961 		err = EINVAL;
962 		goto errout;
963 	}
964 
965 	if (bus_dmamap_create(sc->sc_srcdmat, BUS_DMA_NOWAIT, &re->re_src_map)) {
966 		safestats.st_nomap++;
967 		err = ENOMEM;
968 		goto errout;
969 	}
970 	if (bus_dmamap_load_crp(sc->sc_srcdmat, re->re_src_map, crp, safe_op_cb,
971 	    &re->re_src, BUS_DMA_NOWAIT) != 0) {
972 		bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map);
973 		re->re_src_map = NULL;
974 		safestats.st_noload++;
975 		err = ENOMEM;
976 		goto errout;
977 	}
978 	re->re_src_mapsize = crypto_buffer_len(&crp->crp_buf);
979 	nicealign = safe_dmamap_aligned(&re->re_src);
980 	uniform = safe_dmamap_uniform(&re->re_src);
981 
982 	DPRINTF(("src nicealign %u uniform %u nsegs %u\n",
983 		nicealign, uniform, re->re_src.nsegs));
984 	if (re->re_src.nsegs > 1) {
985 		re->re_desc.d_src = sc->sc_spalloc.dma_paddr +
986 			((caddr_t) sc->sc_spfree - (caddr_t) sc->sc_spring);
987 		for (i = 0; i < re->re_src_nsegs; i++) {
988 			/* NB: no need to check if there's space */
989 			pd = sc->sc_spfree;
990 			if (++(sc->sc_spfree) == sc->sc_springtop)
991 				sc->sc_spfree = sc->sc_spring;
992 
993 			KASSERT((pd->pd_flags&3) == 0 ||
994 				(pd->pd_flags&3) == SAFE_PD_DONE,
995 				("bogus source particle descriptor; flags %x",
996 				pd->pd_flags));
997 			pd->pd_addr = re->re_src_segs[i].ds_addr;
998 			pd->pd_size = re->re_src_segs[i].ds_len;
999 			pd->pd_flags = SAFE_PD_READY;
1000 		}
1001 		cmd0 |= SAFE_SA_CMD0_IGATHER;
1002 	} else {
1003 		/*
1004 		 * No need for gather, reference the operand directly.
1005 		 */
1006 		re->re_desc.d_src = re->re_src_segs[0].ds_addr;
1007 	}
1008 
1009 	if (csp->csp_mode == CSP_MODE_DIGEST) {
1010 		/*
1011 		 * Hash op; no destination needed.
1012 		 */
1013 	} else {
1014 		if (nicealign && uniform == 1) {
1015 			/*
1016 			 * Source layout is suitable for direct
1017 			 * sharing of the DMA map and segment list.
1018 			 */
1019 			re->re_dst = re->re_src;
1020 		} else if (nicealign && uniform == 2) {
1021 			/*
1022 			 * The source is properly aligned but requires a
1023 			 * different particle list to handle DMA of the
1024 			 * result.  Create a new map and do the load to
1025 			 * create the segment list.  The particle
1026 			 * descriptor setup code below will handle the
1027 			 * rest.
1028 			 */
1029 			if (bus_dmamap_create(sc->sc_dstdmat, BUS_DMA_NOWAIT,
1030 			    &re->re_dst_map)) {
1031 				safestats.st_nomap++;
1032 				err = ENOMEM;
1033 				goto errout;
1034 			}
1035 			if (bus_dmamap_load_crp(sc->sc_dstdmat, re->re_dst_map,
1036 			    crp, safe_op_cb, &re->re_dst, BUS_DMA_NOWAIT) !=
1037 			    0) {
1038 				bus_dmamap_destroy(sc->sc_dstdmat,
1039 				    re->re_dst_map);
1040 				re->re_dst_map = NULL;
1041 				safestats.st_noload++;
1042 				err = ENOMEM;
1043 				goto errout;
1044 			}
1045 		} else if (crp->crp_buf.cb_type == CRYPTO_BUF_MBUF) {
1046 			int totlen, len;
1047 			struct mbuf *m, *top, **mp;
1048 
1049 			/*
1050 			 * DMA constraints require that we allocate a
1051 			 * new mbuf chain for the destination.  We
1052 			 * allocate an entire new set of mbufs of
1053 			 * optimal/required size and then tell the
1054 			 * hardware to copy any bits that are not
1055 			 * created as a byproduct of the operation.
1056 			 */
1057 			if (!nicealign)
1058 				safestats.st_unaligned++;
1059 			if (!uniform)
1060 				safestats.st_notuniform++;
1061 			totlen = re->re_src_mapsize;
1062 			if (crp->crp_buf.cb_mbuf->m_flags & M_PKTHDR) {
1063 				len = MHLEN;
1064 				MGETHDR(m, M_NOWAIT, MT_DATA);
1065 				if (m && !m_dup_pkthdr(m, crp->crp_buf.cb_mbuf,
1066 				    M_NOWAIT)) {
1067 					m_free(m);
1068 					m = NULL;
1069 				}
1070 			} else {
1071 				len = MLEN;
1072 				MGET(m, M_NOWAIT, MT_DATA);
1073 			}
1074 			if (m == NULL) {
1075 				safestats.st_nombuf++;
1076 				err = sc->sc_nqchip ? ERESTART : ENOMEM;
1077 				goto errout;
1078 			}
1079 			if (totlen >= MINCLSIZE) {
1080 				if (!(MCLGET(m, M_NOWAIT))) {
1081 					m_free(m);
1082 					safestats.st_nomcl++;
1083 					err = sc->sc_nqchip ?
1084 					    ERESTART : ENOMEM;
1085 					goto errout;
1086 				}
1087 				len = MCLBYTES;
1088 			}
1089 			m->m_len = len;
1090 			top = NULL;
1091 			mp = &top;
1092 
1093 			while (totlen > 0) {
1094 				if (top) {
1095 					MGET(m, M_NOWAIT, MT_DATA);
1096 					if (m == NULL) {
1097 						m_freem(top);
1098 						safestats.st_nombuf++;
1099 						err = sc->sc_nqchip ?
1100 						    ERESTART : ENOMEM;
1101 						goto errout;
1102 					}
1103 					len = MLEN;
1104 				}
1105 				if (top && totlen >= MINCLSIZE) {
1106 					if (!(MCLGET(m, M_NOWAIT))) {
1107 						*mp = m;
1108 						m_freem(top);
1109 						safestats.st_nomcl++;
1110 						err = sc->sc_nqchip ?
1111 						    ERESTART : ENOMEM;
1112 						goto errout;
1113 					}
1114 					len = MCLBYTES;
1115 				}
1116 				m->m_len = len = min(totlen, len);
1117 				totlen -= len;
1118 				*mp = m;
1119 				mp = &m->m_next;
1120 			}
1121 			re->re_dst_m = top;
1122 			if (bus_dmamap_create(sc->sc_dstdmat,
1123 			    BUS_DMA_NOWAIT, &re->re_dst_map) != 0) {
1124 				safestats.st_nomap++;
1125 				err = ENOMEM;
1126 				goto errout;
1127 			}
1128 			if (bus_dmamap_load_mbuf_sg(sc->sc_dstdmat,
1129 			    re->re_dst_map, top, re->re_dst_segs,
1130 			    &re->re_dst_nsegs, 0) != 0) {
1131 				bus_dmamap_destroy(sc->sc_dstdmat,
1132 				    re->re_dst_map);
1133 				re->re_dst_map = NULL;
1134 				safestats.st_noload++;
1135 				err = ENOMEM;
1136 				goto errout;
1137 			}
1138 			re->re_dst_mapsize = re->re_src_mapsize;
1139 			if (re->re_src.mapsize > oplen) {
1140 				/*
1141 				 * There's data following what the
1142 				 * hardware will copy for us.  If this
1143 				 * isn't just the ICV (that's going to
1144 				 * be written on completion), copy it
1145 				 * to the new mbufs
1146 				 */
1147 				if (!(csp->csp_mode == CSP_MODE_ETA &&
1148 				    (re->re_src.mapsize-oplen) == ses->ses_mlen &&
1149 				    crp->crp_digest_start == oplen))
1150 					safe_mcopy(crp->crp_buf.cb_mbuf,
1151 					    re->re_dst_m, oplen);
1152 				else
1153 					safestats.st_noicvcopy++;
1154 			}
1155 		} else {
1156 			if (!nicealign) {
1157 				safestats.st_iovmisaligned++;
1158 				err = EINVAL;
1159 				goto errout;
1160 			} else {
1161 				/*
1162 				 * There's no way to handle the DMA
1163 				 * requirements with this uio.  We
1164 				 * could create a separate DMA area for
1165 				 * the result and then copy it back,
1166 				 * but for now we just bail and return
1167 				 * an error.  Note that uio requests
1168 				 * > SAFE_MAX_DSIZE are handled because
1169 				 * the DMA map and segment list for the
1170 				 * destination wil result in a
1171 				 * destination particle list that does
1172 				 * the necessary scatter DMA.
1173 				 */
1174 				safestats.st_iovnotuniform++;
1175 				err = EINVAL;
1176 				goto errout;
1177 			}
1178 		}
1179 
1180 		if (re->re_dst.nsegs > 1) {
1181 			re->re_desc.d_dst = sc->sc_dpalloc.dma_paddr +
1182 			    ((caddr_t) sc->sc_dpfree - (caddr_t) sc->sc_dpring);
1183 			for (i = 0; i < re->re_dst_nsegs; i++) {
1184 				pd = sc->sc_dpfree;
1185 				KASSERT((pd->pd_flags&3) == 0 ||
1186 					(pd->pd_flags&3) == SAFE_PD_DONE,
1187 					("bogus dest particle descriptor; flags %x",
1188 						pd->pd_flags));
1189 				if (++(sc->sc_dpfree) == sc->sc_dpringtop)
1190 					sc->sc_dpfree = sc->sc_dpring;
1191 				pd->pd_addr = re->re_dst_segs[i].ds_addr;
1192 				pd->pd_flags = SAFE_PD_READY;
1193 			}
1194 			cmd0 |= SAFE_SA_CMD0_OSCATTER;
1195 		} else {
1196 			/*
1197 			 * No need for scatter, reference the operand directly.
1198 			 */
1199 			re->re_desc.d_dst = re->re_dst_segs[0].ds_addr;
1200 		}
1201 	}
1202 
1203 	/*
1204 	 * All done with setup; fillin the SA command words
1205 	 * and the packet engine descriptor.  The operation
1206 	 * is now ready for submission to the hardware.
1207 	 */
1208 	sa->sa_cmd0 = cmd0 | SAFE_SA_CMD0_IPCI | SAFE_SA_CMD0_OPCI;
1209 	sa->sa_cmd1 = cmd1
1210 		    | (coffset << SAFE_SA_CMD1_OFFSET_S)
1211 		    | SAFE_SA_CMD1_SAREV1	/* Rev 1 SA data structure */
1212 		    | SAFE_SA_CMD1_SRPCI
1213 		    ;
1214 	/*
1215 	 * NB: the order of writes is important here.  In case the
1216 	 * chip is scanning the ring because of an outstanding request
1217 	 * it might nab this one too.  In that case we need to make
1218 	 * sure the setup is complete before we write the length
1219 	 * field of the descriptor as it signals the descriptor is
1220 	 * ready for processing.
1221 	 */
1222 	re->re_desc.d_csr = SAFE_PE_CSR_READY | SAFE_PE_CSR_SAPCI;
1223 	if (csp->csp_auth_alg != 0)
1224 		re->re_desc.d_csr |= SAFE_PE_CSR_LOADSA | SAFE_PE_CSR_HASHFINAL;
1225 	re->re_desc.d_len = oplen
1226 			  | SAFE_PE_LEN_READY
1227 			  | (bypass << SAFE_PE_LEN_BYPASS_S)
1228 			  ;
1229 
1230 	safestats.st_ipackets++;
1231 	safestats.st_ibytes += oplen;
1232 
1233 	if (++(sc->sc_front) == sc->sc_ringtop)
1234 		sc->sc_front = sc->sc_ring;
1235 
1236 	/* XXX honor batching */
1237 	safe_feed(sc, re);
1238 	mtx_unlock(&sc->sc_ringmtx);
1239 	return (0);
1240 
1241 errout:
1242 	if (re->re_dst_m != NULL)
1243 		m_freem(re->re_dst_m);
1244 
1245 	if (re->re_dst_map != NULL && re->re_dst_map != re->re_src_map) {
1246 		bus_dmamap_unload(sc->sc_dstdmat, re->re_dst_map);
1247 		bus_dmamap_destroy(sc->sc_dstdmat, re->re_dst_map);
1248 	}
1249 	if (re->re_src_map != NULL) {
1250 		bus_dmamap_unload(sc->sc_srcdmat, re->re_src_map);
1251 		bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map);
1252 	}
1253 	mtx_unlock(&sc->sc_ringmtx);
1254 	if (err != ERESTART) {
1255 		crp->crp_etype = err;
1256 		crypto_done(crp);
1257 		err = 0;
1258 	} else {
1259 		sc->sc_needwakeup |= CRYPTO_SYMQ;
1260 	}
1261 	return (err);
1262 }
1263 
1264 static void
1265 safe_callback(struct safe_softc *sc, struct safe_ringentry *re)
1266 {
1267 	const struct crypto_session_params *csp;
1268 	struct cryptop *crp = (struct cryptop *)re->re_crp;
1269 	struct safe_session *ses;
1270 	uint8_t hash[HASH_MAX_LEN];
1271 
1272 	ses = crypto_get_driver_session(crp->crp_session);
1273 	csp = crypto_get_params(crp->crp_session);
1274 
1275 	safestats.st_opackets++;
1276 	safestats.st_obytes += re->re_dst.mapsize;
1277 
1278 	safe_dma_sync(&sc->sc_ringalloc,
1279 		BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1280 	if (re->re_desc.d_csr & SAFE_PE_CSR_STATUS) {
1281 		device_printf(sc->sc_dev, "csr 0x%x cmd0 0x%x cmd1 0x%x\n",
1282 			re->re_desc.d_csr,
1283 			re->re_sa.sa_cmd0, re->re_sa.sa_cmd1);
1284 		safestats.st_peoperr++;
1285 		crp->crp_etype = EIO;		/* something more meaningful? */
1286 	}
1287 
1288 	/*
1289 	 * XXX: Should crp_buf.cb_mbuf be updated to re->re_dst_m if
1290 	 * it is non-NULL?
1291 	 */
1292 
1293 	if (re->re_dst_map != NULL && re->re_dst_map != re->re_src_map) {
1294 		bus_dmamap_sync(sc->sc_dstdmat, re->re_dst_map,
1295 		    BUS_DMASYNC_POSTREAD);
1296 		bus_dmamap_unload(sc->sc_dstdmat, re->re_dst_map);
1297 		bus_dmamap_destroy(sc->sc_dstdmat, re->re_dst_map);
1298 	}
1299 	bus_dmamap_sync(sc->sc_srcdmat, re->re_src_map, BUS_DMASYNC_POSTWRITE);
1300 	bus_dmamap_unload(sc->sc_srcdmat, re->re_src_map);
1301 	bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map);
1302 
1303 	if (re->re_flags & SAFE_QFLAGS_COPYOUTICV) {
1304 		if (csp->csp_auth_alg == CRYPTO_SHA1_HMAC) {
1305 			/*
1306 			 * SHA-1 ICV's are byte-swapped; fix 'em up
1307 			 * before copying them to their destination.
1308 			 */
1309 			re->re_sastate.sa_saved_indigest[0] =
1310 			    bswap32(re->re_sastate.sa_saved_indigest[0]);
1311 			re->re_sastate.sa_saved_indigest[1] =
1312 			    bswap32(re->re_sastate.sa_saved_indigest[1]);
1313 			re->re_sastate.sa_saved_indigest[2] =
1314 			    bswap32(re->re_sastate.sa_saved_indigest[2]);
1315 		}
1316 
1317 		if (crp->crp_op & CRYPTO_OP_VERIFY_DIGEST) {
1318 			crypto_copydata(crp, crp->crp_digest_start,
1319 			    ses->ses_mlen, hash);
1320 			if (timingsafe_bcmp(re->re_sastate.sa_saved_indigest,
1321 			    hash, ses->ses_mlen) != 0)
1322 				crp->crp_etype = EBADMSG;
1323 		} else
1324 			crypto_copyback(crp, crp->crp_digest_start,
1325 			    ses->ses_mlen, re->re_sastate.sa_saved_indigest);
1326 	}
1327 	crypto_done(crp);
1328 }
1329 
1330 /*
1331  * Copy all data past offset from srcm to dstm.
1332  */
1333 static void
1334 safe_mcopy(struct mbuf *srcm, struct mbuf *dstm, u_int offset)
1335 {
1336 	u_int j, dlen, slen;
1337 	caddr_t dptr, sptr;
1338 
1339 	/*
1340 	 * Advance src and dst to offset.
1341 	 */
1342 	j = offset;
1343 	while (j >= srcm->m_len) {
1344 		j -= srcm->m_len;
1345 		srcm = srcm->m_next;
1346 		if (srcm == NULL)
1347 			return;
1348 	}
1349 	sptr = mtod(srcm, caddr_t) + j;
1350 	slen = srcm->m_len - j;
1351 
1352 	j = offset;
1353 	while (j >= dstm->m_len) {
1354 		j -= dstm->m_len;
1355 		dstm = dstm->m_next;
1356 		if (dstm == NULL)
1357 			return;
1358 	}
1359 	dptr = mtod(dstm, caddr_t) + j;
1360 	dlen = dstm->m_len - j;
1361 
1362 	/*
1363 	 * Copy everything that remains.
1364 	 */
1365 	for (;;) {
1366 		j = min(slen, dlen);
1367 		bcopy(sptr, dptr, j);
1368 		if (slen == j) {
1369 			srcm = srcm->m_next;
1370 			if (srcm == NULL)
1371 				return;
1372 			sptr = srcm->m_data;
1373 			slen = srcm->m_len;
1374 		} else
1375 			sptr += j, slen -= j;
1376 		if (dlen == j) {
1377 			dstm = dstm->m_next;
1378 			if (dstm == NULL)
1379 				return;
1380 			dptr = dstm->m_data;
1381 			dlen = dstm->m_len;
1382 		} else
1383 			dptr += j, dlen -= j;
1384 	}
1385 }
1386 
1387 #ifndef SAFE_NO_RNG
1388 #define	SAFE_RNG_MAXWAIT	1000
1389 
1390 static void
1391 safe_rng_init(struct safe_softc *sc)
1392 {
1393 	u_int32_t w, v;
1394 	int i;
1395 
1396 	WRITE_REG(sc, SAFE_RNG_CTRL, 0);
1397 	/* use default value according to the manual */
1398 	WRITE_REG(sc, SAFE_RNG_CNFG, 0x834);	/* magic from SafeNet */
1399 	WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0);
1400 
1401 	/*
1402 	 * There is a bug in rev 1.0 of the 1140 that when the RNG
1403 	 * is brought out of reset the ready status flag does not
1404 	 * work until the RNG has finished its internal initialization.
1405 	 *
1406 	 * So in order to determine the device is through its
1407 	 * initialization we must read the data register, using the
1408 	 * status reg in the read in case it is initialized.  Then read
1409 	 * the data register until it changes from the first read.
1410 	 * Once it changes read the data register until it changes
1411 	 * again.  At this time the RNG is considered initialized.
1412 	 * This could take between 750ms - 1000ms in time.
1413 	 */
1414 	i = 0;
1415 	w = READ_REG(sc, SAFE_RNG_OUT);
1416 	do {
1417 		v = READ_REG(sc, SAFE_RNG_OUT);
1418 		if (v != w) {
1419 			w = v;
1420 			break;
1421 		}
1422 		DELAY(10);
1423 	} while (++i < SAFE_RNG_MAXWAIT);
1424 
1425 	/* Wait Until data changes again */
1426 	i = 0;
1427 	do {
1428 		v = READ_REG(sc, SAFE_RNG_OUT);
1429 		if (v != w)
1430 			break;
1431 		DELAY(10);
1432 	} while (++i < SAFE_RNG_MAXWAIT);
1433 }
1434 
1435 static __inline void
1436 safe_rng_disable_short_cycle(struct safe_softc *sc)
1437 {
1438 	WRITE_REG(sc, SAFE_RNG_CTRL,
1439 		READ_REG(sc, SAFE_RNG_CTRL) &~ SAFE_RNG_CTRL_SHORTEN);
1440 }
1441 
1442 static __inline void
1443 safe_rng_enable_short_cycle(struct safe_softc *sc)
1444 {
1445 	WRITE_REG(sc, SAFE_RNG_CTRL,
1446 		READ_REG(sc, SAFE_RNG_CTRL) | SAFE_RNG_CTRL_SHORTEN);
1447 }
1448 
1449 static __inline u_int32_t
1450 safe_rng_read(struct safe_softc *sc)
1451 {
1452 	int i;
1453 
1454 	i = 0;
1455 	while (READ_REG(sc, SAFE_RNG_STAT) != 0 && ++i < SAFE_RNG_MAXWAIT)
1456 		;
1457 	return READ_REG(sc, SAFE_RNG_OUT);
1458 }
1459 
1460 static void
1461 safe_rng(void *arg)
1462 {
1463 	struct safe_softc *sc = arg;
1464 	u_int32_t buf[SAFE_RNG_MAXBUFSIZ];	/* NB: maybe move to softc */
1465 	u_int maxwords;
1466 	int i;
1467 
1468 	safestats.st_rng++;
1469 	/*
1470 	 * Fetch the next block of data.
1471 	 */
1472 	maxwords = safe_rngbufsize;
1473 	if (maxwords > SAFE_RNG_MAXBUFSIZ)
1474 		maxwords = SAFE_RNG_MAXBUFSIZ;
1475 retry:
1476 	for (i = 0; i < maxwords; i++)
1477 		buf[i] = safe_rng_read(sc);
1478 	/*
1479 	 * Check the comparator alarm count and reset the h/w if
1480 	 * it exceeds our threshold.  This guards against the
1481 	 * hardware oscillators resonating with external signals.
1482 	 */
1483 	if (READ_REG(sc, SAFE_RNG_ALM_CNT) > safe_rngmaxalarm) {
1484 		u_int32_t freq_inc, w;
1485 
1486 		DPRINTF(("%s: alarm count %u exceeds threshold %u\n", __func__,
1487 			READ_REG(sc, SAFE_RNG_ALM_CNT), safe_rngmaxalarm));
1488 		safestats.st_rngalarm++;
1489 		safe_rng_enable_short_cycle(sc);
1490 		freq_inc = 18;
1491 		for (i = 0; i < 64; i++) {
1492 			w = READ_REG(sc, SAFE_RNG_CNFG);
1493 			freq_inc = ((w + freq_inc) & 0x3fL);
1494 			w = ((w & ~0x3fL) | freq_inc);
1495 			WRITE_REG(sc, SAFE_RNG_CNFG, w);
1496 
1497 			WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0);
1498 
1499 			(void) safe_rng_read(sc);
1500 			DELAY(25);
1501 
1502 			if (READ_REG(sc, SAFE_RNG_ALM_CNT) == 0) {
1503 				safe_rng_disable_short_cycle(sc);
1504 				goto retry;
1505 			}
1506 			freq_inc = 1;
1507 		}
1508 		safe_rng_disable_short_cycle(sc);
1509 	} else
1510 		WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0);
1511 
1512 	(*sc->sc_harvest)(sc->sc_rndtest, buf, maxwords*sizeof (u_int32_t));
1513 	callout_reset(&sc->sc_rngto,
1514 		hz * (safe_rnginterval ? safe_rnginterval : 1), safe_rng, sc);
1515 }
1516 #endif /* SAFE_NO_RNG */
1517 
1518 static void
1519 safe_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1520 {
1521 	bus_addr_t *paddr = (bus_addr_t*) arg;
1522 	*paddr = segs->ds_addr;
1523 }
1524 
1525 static int
1526 safe_dma_malloc(
1527 	struct safe_softc *sc,
1528 	bus_size_t size,
1529 	struct safe_dma_alloc *dma,
1530 	int mapflags
1531 )
1532 {
1533 	int r;
1534 
1535 	r = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev),	/* parent */
1536 			       sizeof(u_int32_t), 0,	/* alignment, bounds */
1537 			       BUS_SPACE_MAXADDR_32BIT,	/* lowaddr */
1538 			       BUS_SPACE_MAXADDR,	/* highaddr */
1539 			       NULL, NULL,		/* filter, filterarg */
1540 			       size,			/* maxsize */
1541 			       1,			/* nsegments */
1542 			       size,			/* maxsegsize */
1543 			       BUS_DMA_ALLOCNOW,	/* flags */
1544 			       NULL, NULL,		/* locking */
1545 			       &dma->dma_tag);
1546 	if (r != 0) {
1547 		device_printf(sc->sc_dev, "safe_dma_malloc: "
1548 			"bus_dma_tag_create failed; error %u\n", r);
1549 		goto fail_0;
1550 	}
1551 
1552 	r = bus_dmamem_alloc(dma->dma_tag, (void**) &dma->dma_vaddr,
1553 			     BUS_DMA_NOWAIT, &dma->dma_map);
1554 	if (r != 0) {
1555 		device_printf(sc->sc_dev, "safe_dma_malloc: "
1556 			"bus_dmammem_alloc failed; size %ju, error %u\n",
1557 			(uintmax_t)size, r);
1558 		goto fail_1;
1559 	}
1560 
1561 	r = bus_dmamap_load(dma->dma_tag, dma->dma_map, dma->dma_vaddr,
1562 		            size,
1563 			    safe_dmamap_cb,
1564 			    &dma->dma_paddr,
1565 			    mapflags | BUS_DMA_NOWAIT);
1566 	if (r != 0) {
1567 		device_printf(sc->sc_dev, "safe_dma_malloc: "
1568 			"bus_dmamap_load failed; error %u\n", r);
1569 		goto fail_2;
1570 	}
1571 
1572 	dma->dma_size = size;
1573 	return (0);
1574 
1575 	bus_dmamap_unload(dma->dma_tag, dma->dma_map);
1576 fail_2:
1577 	bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
1578 fail_1:
1579 	bus_dma_tag_destroy(dma->dma_tag);
1580 fail_0:
1581 	dma->dma_tag = NULL;
1582 	return (r);
1583 }
1584 
1585 static void
1586 safe_dma_free(struct safe_softc *sc, struct safe_dma_alloc *dma)
1587 {
1588 	bus_dmamap_unload(dma->dma_tag, dma->dma_map);
1589 	bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
1590 	bus_dma_tag_destroy(dma->dma_tag);
1591 }
1592 
1593 /*
1594  * Resets the board.  Values in the regesters are left as is
1595  * from the reset (i.e. initial values are assigned elsewhere).
1596  */
1597 static void
1598 safe_reset_board(struct safe_softc *sc)
1599 {
1600 	u_int32_t v;
1601 	/*
1602 	 * Reset the device.  The manual says no delay
1603 	 * is needed between marking and clearing reset.
1604 	 */
1605 	v = READ_REG(sc, SAFE_PE_DMACFG) &~
1606 		(SAFE_PE_DMACFG_PERESET | SAFE_PE_DMACFG_PDRRESET |
1607 		 SAFE_PE_DMACFG_SGRESET);
1608 	WRITE_REG(sc, SAFE_PE_DMACFG, v
1609 				    | SAFE_PE_DMACFG_PERESET
1610 				    | SAFE_PE_DMACFG_PDRRESET
1611 				    | SAFE_PE_DMACFG_SGRESET);
1612 	WRITE_REG(sc, SAFE_PE_DMACFG, v);
1613 }
1614 
1615 /*
1616  * Initialize registers we need to touch only once.
1617  */
1618 static void
1619 safe_init_board(struct safe_softc *sc)
1620 {
1621 	u_int32_t v, dwords;
1622 
1623 	v = READ_REG(sc, SAFE_PE_DMACFG);
1624 	v &=~ SAFE_PE_DMACFG_PEMODE;
1625 	v |= SAFE_PE_DMACFG_FSENA		/* failsafe enable */
1626 	  |  SAFE_PE_DMACFG_GPRPCI		/* gather ring on PCI */
1627 	  |  SAFE_PE_DMACFG_SPRPCI		/* scatter ring on PCI */
1628 	  |  SAFE_PE_DMACFG_ESDESC		/* endian-swap descriptors */
1629 	  |  SAFE_PE_DMACFG_ESSA		/* endian-swap SA's */
1630 	  |  SAFE_PE_DMACFG_ESPDESC		/* endian-swap part. desc's */
1631 	  ;
1632 	WRITE_REG(sc, SAFE_PE_DMACFG, v);
1633 #if 0
1634 	/* XXX select byte swap based on host byte order */
1635 	WRITE_REG(sc, SAFE_ENDIAN, 0x1b);
1636 #endif
1637 	if (sc->sc_chiprev == SAFE_REV(1,0)) {
1638 		/*
1639 		 * Avoid large PCI DMA transfers.  Rev 1.0 has a bug where
1640 		 * "target mode transfers" done while the chip is DMA'ing
1641 		 * >1020 bytes cause the hardware to lockup.  To avoid this
1642 		 * we reduce the max PCI transfer size and use small source
1643 		 * particle descriptors (<= 256 bytes).
1644 		 */
1645 		WRITE_REG(sc, SAFE_DMA_CFG, 256);
1646 		device_printf(sc->sc_dev,
1647 			"Reduce max DMA size to %u words for rev %u.%u WAR\n",
1648 			(READ_REG(sc, SAFE_DMA_CFG)>>2) & 0xff,
1649 			SAFE_REV_MAJ(sc->sc_chiprev),
1650 			SAFE_REV_MIN(sc->sc_chiprev));
1651 	}
1652 
1653 	/* NB: operands+results are overlaid */
1654 	WRITE_REG(sc, SAFE_PE_PDRBASE, sc->sc_ringalloc.dma_paddr);
1655 	WRITE_REG(sc, SAFE_PE_RDRBASE, sc->sc_ringalloc.dma_paddr);
1656 	/*
1657 	 * Configure ring entry size and number of items in the ring.
1658 	 */
1659 	KASSERT((sizeof(struct safe_ringentry) % sizeof(u_int32_t)) == 0,
1660 		("PE ring entry not 32-bit aligned!"));
1661 	dwords = sizeof(struct safe_ringentry) / sizeof(u_int32_t);
1662 	WRITE_REG(sc, SAFE_PE_RINGCFG,
1663 		(dwords << SAFE_PE_RINGCFG_OFFSET_S) | SAFE_MAX_NQUEUE);
1664 	WRITE_REG(sc, SAFE_PE_RINGPOLL, 0);	/* disable polling */
1665 
1666 	WRITE_REG(sc, SAFE_PE_GRNGBASE, sc->sc_spalloc.dma_paddr);
1667 	WRITE_REG(sc, SAFE_PE_SRNGBASE, sc->sc_dpalloc.dma_paddr);
1668 	WRITE_REG(sc, SAFE_PE_PARTSIZE,
1669 		(SAFE_TOTAL_DPART<<16) | SAFE_TOTAL_SPART);
1670 	/*
1671 	 * NB: destination particles are fixed size.  We use
1672 	 *     an mbuf cluster and require all results go to
1673 	 *     clusters or smaller.
1674 	 */
1675 	WRITE_REG(sc, SAFE_PE_PARTCFG, SAFE_MAX_DSIZE);
1676 
1677 	/* it's now safe to enable PE mode, do it */
1678 	WRITE_REG(sc, SAFE_PE_DMACFG, v | SAFE_PE_DMACFG_PEMODE);
1679 
1680 	/*
1681 	 * Configure hardware to use level-triggered interrupts and
1682 	 * to interrupt after each descriptor is processed.
1683 	 */
1684 	WRITE_REG(sc, SAFE_HI_CFG, SAFE_HI_CFG_LEVEL);
1685 	WRITE_REG(sc, SAFE_HI_DESC_CNT, 1);
1686 	WRITE_REG(sc, SAFE_HI_MASK, SAFE_INT_PE_DDONE | SAFE_INT_PE_ERROR);
1687 }
1688 
1689 /*
1690  * Init PCI registers
1691  */
1692 static void
1693 safe_init_pciregs(device_t dev)
1694 {
1695 }
1696 
1697 /*
1698  * Clean up after a chip crash.
1699  * It is assumed that the caller in splimp()
1700  */
1701 static void
1702 safe_cleanchip(struct safe_softc *sc)
1703 {
1704 
1705 	if (sc->sc_nqchip != 0) {
1706 		struct safe_ringentry *re = sc->sc_back;
1707 
1708 		while (re != sc->sc_front) {
1709 			if (re->re_desc.d_csr != 0)
1710 				safe_free_entry(sc, re);
1711 			if (++re == sc->sc_ringtop)
1712 				re = sc->sc_ring;
1713 		}
1714 		sc->sc_back = re;
1715 		sc->sc_nqchip = 0;
1716 	}
1717 }
1718 
1719 /*
1720  * free a safe_q
1721  * It is assumed that the caller is within splimp().
1722  */
1723 static int
1724 safe_free_entry(struct safe_softc *sc, struct safe_ringentry *re)
1725 {
1726 	struct cryptop *crp;
1727 
1728 	/*
1729 	 * Free header MCR
1730 	 */
1731 	if (re->re_dst_m != NULL)
1732 		m_freem(re->re_dst_m);
1733 
1734 	crp = (struct cryptop *)re->re_crp;
1735 
1736 	re->re_desc.d_csr = 0;
1737 
1738 	crp->crp_etype = EFAULT;
1739 	crypto_done(crp);
1740 	return(0);
1741 }
1742 
1743 /*
1744  * Routine to reset the chip and clean up.
1745  * It is assumed that the caller is in splimp()
1746  */
1747 static void
1748 safe_totalreset(struct safe_softc *sc)
1749 {
1750 	safe_reset_board(sc);
1751 	safe_init_board(sc);
1752 	safe_cleanchip(sc);
1753 }
1754 
1755 /*
1756  * Is the operand suitable aligned for direct DMA.  Each
1757  * segment must be aligned on a 32-bit boundary and all
1758  * but the last segment must be a multiple of 4 bytes.
1759  */
1760 static int
1761 safe_dmamap_aligned(const struct safe_operand *op)
1762 {
1763 	int i;
1764 
1765 	for (i = 0; i < op->nsegs; i++) {
1766 		if (op->segs[i].ds_addr & 3)
1767 			return (0);
1768 		if (i != (op->nsegs - 1) && (op->segs[i].ds_len & 3))
1769 			return (0);
1770 	}
1771 	return (1);
1772 }
1773 
1774 /*
1775  * Is the operand suitable for direct DMA as the destination
1776  * of an operation.  The hardware requires that each ``particle''
1777  * but the last in an operation result have the same size.  We
1778  * fix that size at SAFE_MAX_DSIZE bytes.  This routine returns
1779  * 0 if some segment is not a multiple of this size, 1 if all
1780  * segments are exactly this size, or 2 if segments are at worst
1781  * a multiple of this size.
1782  */
1783 static int
1784 safe_dmamap_uniform(const struct safe_operand *op)
1785 {
1786 	int result = 1;
1787 
1788 	if (op->nsegs > 0) {
1789 		int i;
1790 
1791 		for (i = 0; i < op->nsegs-1; i++) {
1792 			if (op->segs[i].ds_len % SAFE_MAX_DSIZE)
1793 				return (0);
1794 			if (op->segs[i].ds_len != SAFE_MAX_DSIZE)
1795 				result = 2;
1796 		}
1797 	}
1798 	return (result);
1799 }
1800 
1801 #ifdef SAFE_DEBUG
1802 static void
1803 safe_dump_dmastatus(struct safe_softc *sc, const char *tag)
1804 {
1805 	printf("%s: ENDIAN 0x%x SRC 0x%x DST 0x%x STAT 0x%x\n"
1806 		, tag
1807 		, READ_REG(sc, SAFE_DMA_ENDIAN)
1808 		, READ_REG(sc, SAFE_DMA_SRCADDR)
1809 		, READ_REG(sc, SAFE_DMA_DSTADDR)
1810 		, READ_REG(sc, SAFE_DMA_STAT)
1811 	);
1812 }
1813 
1814 static void
1815 safe_dump_intrstate(struct safe_softc *sc, const char *tag)
1816 {
1817 	printf("%s: HI_CFG 0x%x HI_MASK 0x%x HI_DESC_CNT 0x%x HU_STAT 0x%x HM_STAT 0x%x\n"
1818 		, tag
1819 		, READ_REG(sc, SAFE_HI_CFG)
1820 		, READ_REG(sc, SAFE_HI_MASK)
1821 		, READ_REG(sc, SAFE_HI_DESC_CNT)
1822 		, READ_REG(sc, SAFE_HU_STAT)
1823 		, READ_REG(sc, SAFE_HM_STAT)
1824 	);
1825 }
1826 
1827 static void
1828 safe_dump_ringstate(struct safe_softc *sc, const char *tag)
1829 {
1830 	u_int32_t estat = READ_REG(sc, SAFE_PE_ERNGSTAT);
1831 
1832 	/* NB: assume caller has lock on ring */
1833 	printf("%s: ERNGSTAT %x (next %u) back %lu front %lu\n",
1834 		tag,
1835 		estat, (estat >> SAFE_PE_ERNGSTAT_NEXT_S),
1836 		(unsigned long)(sc->sc_back - sc->sc_ring),
1837 		(unsigned long)(sc->sc_front - sc->sc_ring));
1838 }
1839 
1840 static void
1841 safe_dump_request(struct safe_softc *sc, const char* tag, struct safe_ringentry *re)
1842 {
1843 	int ix, nsegs;
1844 
1845 	ix = re - sc->sc_ring;
1846 	printf("%s: %p (%u): csr %x src %x dst %x sa %x len %x\n"
1847 		, tag
1848 		, re, ix
1849 		, re->re_desc.d_csr
1850 		, re->re_desc.d_src
1851 		, re->re_desc.d_dst
1852 		, re->re_desc.d_sa
1853 		, re->re_desc.d_len
1854 	);
1855 	if (re->re_src.nsegs > 1) {
1856 		ix = (re->re_desc.d_src - sc->sc_spalloc.dma_paddr) /
1857 			sizeof(struct safe_pdesc);
1858 		for (nsegs = re->re_src.nsegs; nsegs; nsegs--) {
1859 			printf(" spd[%u] %p: %p size %u flags %x"
1860 				, ix, &sc->sc_spring[ix]
1861 				, (caddr_t)(uintptr_t) sc->sc_spring[ix].pd_addr
1862 				, sc->sc_spring[ix].pd_size
1863 				, sc->sc_spring[ix].pd_flags
1864 			);
1865 			if (sc->sc_spring[ix].pd_size == 0)
1866 				printf(" (zero!)");
1867 			printf("\n");
1868 			if (++ix == SAFE_TOTAL_SPART)
1869 				ix = 0;
1870 		}
1871 	}
1872 	if (re->re_dst.nsegs > 1) {
1873 		ix = (re->re_desc.d_dst - sc->sc_dpalloc.dma_paddr) /
1874 			sizeof(struct safe_pdesc);
1875 		for (nsegs = re->re_dst.nsegs; nsegs; nsegs--) {
1876 			printf(" dpd[%u] %p: %p flags %x\n"
1877 				, ix, &sc->sc_dpring[ix]
1878 				, (caddr_t)(uintptr_t) sc->sc_dpring[ix].pd_addr
1879 				, sc->sc_dpring[ix].pd_flags
1880 			);
1881 			if (++ix == SAFE_TOTAL_DPART)
1882 				ix = 0;
1883 		}
1884 	}
1885 	printf("sa: cmd0 %08x cmd1 %08x staterec %x\n",
1886 		re->re_sa.sa_cmd0, re->re_sa.sa_cmd1, re->re_sa.sa_staterec);
1887 	printf("sa: key %x %x %x %x %x %x %x %x\n"
1888 		, re->re_sa.sa_key[0]
1889 		, re->re_sa.sa_key[1]
1890 		, re->re_sa.sa_key[2]
1891 		, re->re_sa.sa_key[3]
1892 		, re->re_sa.sa_key[4]
1893 		, re->re_sa.sa_key[5]
1894 		, re->re_sa.sa_key[6]
1895 		, re->re_sa.sa_key[7]
1896 	);
1897 	printf("sa: indigest %x %x %x %x %x\n"
1898 		, re->re_sa.sa_indigest[0]
1899 		, re->re_sa.sa_indigest[1]
1900 		, re->re_sa.sa_indigest[2]
1901 		, re->re_sa.sa_indigest[3]
1902 		, re->re_sa.sa_indigest[4]
1903 	);
1904 	printf("sa: outdigest %x %x %x %x %x\n"
1905 		, re->re_sa.sa_outdigest[0]
1906 		, re->re_sa.sa_outdigest[1]
1907 		, re->re_sa.sa_outdigest[2]
1908 		, re->re_sa.sa_outdigest[3]
1909 		, re->re_sa.sa_outdigest[4]
1910 	);
1911 	printf("sr: iv %x %x %x %x\n"
1912 		, re->re_sastate.sa_saved_iv[0]
1913 		, re->re_sastate.sa_saved_iv[1]
1914 		, re->re_sastate.sa_saved_iv[2]
1915 		, re->re_sastate.sa_saved_iv[3]
1916 	);
1917 	printf("sr: hashbc %u indigest %x %x %x %x %x\n"
1918 		, re->re_sastate.sa_saved_hashbc
1919 		, re->re_sastate.sa_saved_indigest[0]
1920 		, re->re_sastate.sa_saved_indigest[1]
1921 		, re->re_sastate.sa_saved_indigest[2]
1922 		, re->re_sastate.sa_saved_indigest[3]
1923 		, re->re_sastate.sa_saved_indigest[4]
1924 	);
1925 }
1926 
1927 static void
1928 safe_dump_ring(struct safe_softc *sc, const char *tag)
1929 {
1930 	mtx_lock(&sc->sc_ringmtx);
1931 	printf("\nSafeNet Ring State:\n");
1932 	safe_dump_intrstate(sc, tag);
1933 	safe_dump_dmastatus(sc, tag);
1934 	safe_dump_ringstate(sc, tag);
1935 	if (sc->sc_nqchip) {
1936 		struct safe_ringentry *re = sc->sc_back;
1937 		do {
1938 			safe_dump_request(sc, tag, re);
1939 			if (++re == sc->sc_ringtop)
1940 				re = sc->sc_ring;
1941 		} while (re != sc->sc_front);
1942 	}
1943 	mtx_unlock(&sc->sc_ringmtx);
1944 }
1945 
1946 static int
1947 sysctl_hw_safe_dump(SYSCTL_HANDLER_ARGS)
1948 {
1949 	char dmode[64];
1950 	int error;
1951 
1952 	strncpy(dmode, "", sizeof(dmode) - 1);
1953 	dmode[sizeof(dmode) - 1] = '\0';
1954 	error = sysctl_handle_string(oidp, &dmode[0], sizeof(dmode), req);
1955 
1956 	if (error == 0 && req->newptr != NULL) {
1957 		struct safe_softc *sc = safec;
1958 
1959 		if (!sc)
1960 			return EINVAL;
1961 		if (strncmp(dmode, "dma", 3) == 0)
1962 			safe_dump_dmastatus(sc, "safe0");
1963 		else if (strncmp(dmode, "int", 3) == 0)
1964 			safe_dump_intrstate(sc, "safe0");
1965 		else if (strncmp(dmode, "ring", 4) == 0)
1966 			safe_dump_ring(sc, "safe0");
1967 		else
1968 			return EINVAL;
1969 	}
1970 	return error;
1971 }
1972 SYSCTL_PROC(_hw_safe, OID_AUTO, dump,
1973     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, 0,
1974     sysctl_hw_safe_dump, "A",
1975     "Dump driver state");
1976 #endif /* SAFE_DEBUG */
1977