xref: /freebsd/sys/dev/safe/safe.c (revision 8aac90f18aef7c9eea906c3ff9a001ca7b94f375)
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_generic_detach(dev);
468 	bus_teardown_intr(dev, sc->sc_irq, sc->sc_ih);
469 	bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq);
470 
471 	bus_dma_tag_destroy(sc->sc_srcdmat);
472 	bus_dma_tag_destroy(sc->sc_dstdmat);
473 	bus_release_resource(dev, SYS_RES_MEMORY, BS_BAR, sc->sc_sr);
474 
475 	return (0);
476 }
477 
478 /*
479  * Stop all chip i/o so that the kernel's probe routines don't
480  * get confused by errant DMAs when rebooting.
481  */
482 static int
483 safe_shutdown(device_t dev)
484 {
485 #ifdef notyet
486 	safe_stop(device_get_softc(dev));
487 #endif
488 	return (0);
489 }
490 
491 /*
492  * Device suspend routine.
493  */
494 static int
495 safe_suspend(device_t dev)
496 {
497 	struct safe_softc *sc = device_get_softc(dev);
498 
499 #ifdef notyet
500 	/* XXX stop the device and save PCI settings */
501 #endif
502 	sc->sc_suspended = 1;
503 
504 	return (0);
505 }
506 
507 static int
508 safe_resume(device_t dev)
509 {
510 	struct safe_softc *sc = device_get_softc(dev);
511 
512 #ifdef notyet
513 	/* XXX retore PCI settings and start the device */
514 #endif
515 	sc->sc_suspended = 0;
516 	return (0);
517 }
518 
519 /*
520  * SafeXcel Interrupt routine
521  */
522 static void
523 safe_intr(void *arg)
524 {
525 	struct safe_softc *sc = arg;
526 	volatile u_int32_t stat;
527 
528 	stat = READ_REG(sc, SAFE_HM_STAT);
529 	if (stat == 0)			/* shared irq, not for us */
530 		return;
531 
532 	WRITE_REG(sc, SAFE_HI_CLR, stat);	/* IACK */
533 
534 	if ((stat & SAFE_INT_PE_DDONE)) {
535 		/*
536 		 * Descriptor(s) done; scan the ring and
537 		 * process completed operations.
538 		 */
539 		mtx_lock(&sc->sc_ringmtx);
540 		while (sc->sc_back != sc->sc_front) {
541 			struct safe_ringentry *re = sc->sc_back;
542 #ifdef SAFE_DEBUG
543 			if (safe_debug) {
544 				safe_dump_ringstate(sc, __func__);
545 				safe_dump_request(sc, __func__, re);
546 			}
547 #endif
548 			/*
549 			 * safe_process marks ring entries that were allocated
550 			 * but not used with a csr of zero.  This insures the
551 			 * ring front pointer never needs to be set backwards
552 			 * in the event that an entry is allocated but not used
553 			 * because of a setup error.
554 			 */
555 			if (re->re_desc.d_csr != 0) {
556 				if (!SAFE_PE_CSR_IS_DONE(re->re_desc.d_csr))
557 					break;
558 				if (!SAFE_PE_LEN_IS_DONE(re->re_desc.d_len))
559 					break;
560 				sc->sc_nqchip--;
561 				safe_callback(sc, re);
562 			}
563 			if (++(sc->sc_back) == sc->sc_ringtop)
564 				sc->sc_back = sc->sc_ring;
565 		}
566 		mtx_unlock(&sc->sc_ringmtx);
567 	}
568 
569 	/*
570 	 * Check to see if we got any DMA Error
571 	 */
572 	if (stat & SAFE_INT_PE_ERROR) {
573 		DPRINTF(("dmaerr dmastat %08x\n",
574 			READ_REG(sc, SAFE_PE_DMASTAT)));
575 		safestats.st_dmaerr++;
576 		safe_totalreset(sc);
577 #if 0
578 		safe_feed(sc);
579 #endif
580 	}
581 
582 	if (sc->sc_needwakeup) {		/* XXX check high watermark */
583 		int wakeup = sc->sc_needwakeup & CRYPTO_SYMQ;
584 		DPRINTF(("%s: wakeup crypto %x\n", __func__,
585 			sc->sc_needwakeup));
586 		sc->sc_needwakeup &= ~wakeup;
587 		crypto_unblock(sc->sc_cid, wakeup);
588 	}
589 }
590 
591 /*
592  * safe_feed() - post a request to chip
593  */
594 static void
595 safe_feed(struct safe_softc *sc, struct safe_ringentry *re)
596 {
597 	bus_dmamap_sync(sc->sc_srcdmat, re->re_src_map, BUS_DMASYNC_PREWRITE);
598 	if (re->re_dst_map != NULL)
599 		bus_dmamap_sync(sc->sc_dstdmat, re->re_dst_map,
600 			BUS_DMASYNC_PREREAD);
601 	/* XXX have no smaller granularity */
602 	safe_dma_sync(&sc->sc_ringalloc,
603 		BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
604 	safe_dma_sync(&sc->sc_spalloc, BUS_DMASYNC_PREWRITE);
605 	safe_dma_sync(&sc->sc_dpalloc, BUS_DMASYNC_PREWRITE);
606 
607 #ifdef SAFE_DEBUG
608 	if (safe_debug) {
609 		safe_dump_ringstate(sc, __func__);
610 		safe_dump_request(sc, __func__, re);
611 	}
612 #endif
613 	sc->sc_nqchip++;
614 	if (sc->sc_nqchip > safestats.st_maxqchip)
615 		safestats.st_maxqchip = sc->sc_nqchip;
616 	/* poke h/w to check descriptor ring, any value can be written */
617 	WRITE_REG(sc, SAFE_HI_RD_DESCR, 0);
618 }
619 
620 #define	N(a)	(sizeof(a) / sizeof (a[0]))
621 static void
622 safe_setup_enckey(struct safe_session *ses, const void *key)
623 {
624 	int i;
625 
626 	bcopy(key, ses->ses_key, ses->ses_klen);
627 
628 	/* PE is little-endian, insure proper byte order */
629 	for (i = 0; i < N(ses->ses_key); i++)
630 		ses->ses_key[i] = htole32(ses->ses_key[i]);
631 }
632 
633 static void
634 safe_setup_mackey(struct safe_session *ses, int algo, const uint8_t *key,
635     int klen)
636 {
637 	SHA1_CTX sha1ctx;
638 	int i;
639 
640 	hmac_init_ipad(&auth_hash_hmac_sha1, key, klen, &sha1ctx);
641 	bcopy(sha1ctx.h.b32, ses->ses_hminner, sizeof(sha1ctx.h.b32));
642 
643 	hmac_init_opad(&auth_hash_hmac_sha1, key, klen, &sha1ctx);
644 	bcopy(sha1ctx.h.b32, ses->ses_hmouter, sizeof(sha1ctx.h.b32));
645 
646 	explicit_bzero(&sha1ctx, sizeof(sha1ctx));
647 
648 	/* PE is little-endian, insure proper byte order */
649 	for (i = 0; i < N(ses->ses_hminner); i++) {
650 		ses->ses_hminner[i] = htole32(ses->ses_hminner[i]);
651 		ses->ses_hmouter[i] = htole32(ses->ses_hmouter[i]);
652 	}
653 }
654 #undef N
655 
656 static bool
657 safe_auth_supported(struct safe_softc *sc,
658     const struct crypto_session_params *csp)
659 {
660 
661 	switch (csp->csp_auth_alg) {
662 	case CRYPTO_SHA1_HMAC:
663 		if ((sc->sc_devinfo & SAFE_DEVINFO_SHA1) == 0)
664 			return (false);
665 		break;
666 	default:
667 		return (false);
668 	}
669 	return (true);
670 }
671 
672 static bool
673 safe_cipher_supported(struct safe_softc *sc,
674     const struct crypto_session_params *csp)
675 {
676 
677 	switch (csp->csp_cipher_alg) {
678 	case CRYPTO_AES_CBC:
679 		if ((sc->sc_devinfo & SAFE_DEVINFO_AES) == 0)
680 			return (false);
681 		if (csp->csp_ivlen != 16)
682 			return (false);
683 		if (csp->csp_cipher_klen != 16 &&
684 		    csp->csp_cipher_klen != 24 &&
685 		    csp->csp_cipher_klen != 32)
686 			return (false);
687 		break;
688 	}
689 	return (true);
690 }
691 
692 static int
693 safe_probesession(device_t dev, const struct crypto_session_params *csp)
694 {
695 	struct safe_softc *sc = device_get_softc(dev);
696 
697 	if (csp->csp_flags != 0)
698 		return (EINVAL);
699 	switch (csp->csp_mode) {
700 	case CSP_MODE_DIGEST:
701 		if (!safe_auth_supported(sc, csp))
702 			return (EINVAL);
703 		break;
704 	case CSP_MODE_CIPHER:
705 		if (!safe_cipher_supported(sc, csp))
706 			return (EINVAL);
707 		break;
708 	case CSP_MODE_ETA:
709 		if (!safe_auth_supported(sc, csp) ||
710 		    !safe_cipher_supported(sc, csp))
711 			return (EINVAL);
712 		break;
713 	default:
714 		return (EINVAL);
715 	}
716 
717 	return (CRYPTODEV_PROBE_HARDWARE);
718 }
719 
720 /*
721  * Allocate a new 'session'.
722  */
723 static int
724 safe_newsession(device_t dev, crypto_session_t cses,
725     const struct crypto_session_params *csp)
726 {
727 	struct safe_session *ses;
728 
729 	ses = crypto_get_driver_session(cses);
730 	if (csp->csp_cipher_alg != 0) {
731 		ses->ses_klen = csp->csp_cipher_klen;
732 		if (csp->csp_cipher_key != NULL)
733 			safe_setup_enckey(ses, csp->csp_cipher_key);
734 	}
735 
736 	if (csp->csp_auth_alg != 0) {
737 		ses->ses_mlen = csp->csp_auth_mlen;
738 		if (ses->ses_mlen == 0) {
739 			ses->ses_mlen = SHA1_HASH_LEN;
740 		}
741 
742 		if (csp->csp_auth_key != NULL) {
743 			safe_setup_mackey(ses, csp->csp_auth_alg,
744 			    csp->csp_auth_key, csp->csp_auth_klen);
745 		}
746 	}
747 
748 	return (0);
749 }
750 
751 static void
752 safe_op_cb(void *arg, bus_dma_segment_t *seg, int nsegs, int error)
753 {
754 	struct safe_operand *op = arg;
755 
756 	DPRINTF(("%s: nsegs %d error %d\n", __func__,
757 		nsegs, error));
758 	if (error != 0)
759 		return;
760 	op->nsegs = nsegs;
761 	bcopy(seg, op->segs, nsegs * sizeof (seg[0]));
762 }
763 
764 static int
765 safe_process(device_t dev, struct cryptop *crp, int hint)
766 {
767 	struct safe_softc *sc = device_get_softc(dev);
768 	const struct crypto_session_params *csp;
769 	int err = 0, i, nicealign, uniform;
770 	int bypass, oplen;
771 	int16_t coffset;
772 	struct safe_session *ses;
773 	struct safe_ringentry *re;
774 	struct safe_sarec *sa;
775 	struct safe_pdesc *pd;
776 	u_int32_t cmd0, cmd1, staterec;
777 
778 	mtx_lock(&sc->sc_ringmtx);
779 	if (sc->sc_front == sc->sc_back && sc->sc_nqchip != 0) {
780 		safestats.st_ringfull++;
781 		sc->sc_needwakeup |= CRYPTO_SYMQ;
782 		mtx_unlock(&sc->sc_ringmtx);
783 		return (ERESTART);
784 	}
785 	re = sc->sc_front;
786 
787 	staterec = re->re_sa.sa_staterec;	/* save */
788 	/* NB: zero everything but the PE descriptor */
789 	bzero(&re->re_sa, sizeof(struct safe_ringentry) - sizeof(re->re_desc));
790 	re->re_sa.sa_staterec = staterec;	/* restore */
791 
792 	re->re_crp = crp;
793 
794 	sa = &re->re_sa;
795 	ses = crypto_get_driver_session(crp->crp_session);
796 	csp = crypto_get_params(crp->crp_session);
797 
798 	cmd0 = SAFE_SA_CMD0_BASIC;		/* basic group operation */
799 	cmd1 = 0;
800 	switch (csp->csp_mode) {
801 	case CSP_MODE_DIGEST:
802 		cmd0 |= SAFE_SA_CMD0_OP_HASH;
803 		break;
804 	case CSP_MODE_CIPHER:
805 		cmd0 |= SAFE_SA_CMD0_OP_CRYPT;
806 		break;
807 	case CSP_MODE_ETA:
808 		cmd0 |= SAFE_SA_CMD0_OP_BOTH;
809 		break;
810 	}
811 
812 	if (csp->csp_cipher_alg != 0) {
813 		if (crp->crp_cipher_key != NULL)
814 			safe_setup_enckey(ses, crp->crp_cipher_key);
815 
816 		switch (csp->csp_cipher_alg) {
817 		case CRYPTO_AES_CBC:
818 			cmd0 |= SAFE_SA_CMD0_AES;
819 			cmd1 |= SAFE_SA_CMD1_CBC;
820 			if (ses->ses_klen * 8 == 128)
821 			     cmd1 |=  SAFE_SA_CMD1_AES128;
822 			else if (ses->ses_klen * 8 == 192)
823 			     cmd1 |=  SAFE_SA_CMD1_AES192;
824 			else
825 			     cmd1 |=  SAFE_SA_CMD1_AES256;
826 		}
827 
828 		/*
829 		 * Setup encrypt/decrypt state.  When using basic ops
830 		 * we can't use an inline IV because hash/crypt offset
831 		 * must be from the end of the IV to the start of the
832 		 * crypt data and this leaves out the preceding header
833 		 * from the hash calculation.  Instead we place the IV
834 		 * in the state record and set the hash/crypt offset to
835 		 * copy both the header+IV.
836 		 */
837 		crypto_read_iv(crp, re->re_sastate.sa_saved_iv);
838 		cmd0 |= SAFE_SA_CMD0_IVLD_STATE;
839 
840 		if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op)) {
841 			cmd0 |= SAFE_SA_CMD0_OUTBOUND;
842 
843 			/*
844 			 * XXX: I suspect we don't need this since we
845 			 * don't save the returned IV.
846 			 */
847 			cmd0 |= SAFE_SA_CMD0_SAVEIV;
848 		} else {
849 			cmd0 |= SAFE_SA_CMD0_INBOUND;
850 		}
851 		/*
852 		 * For basic encryption use the zero pad algorithm.
853 		 * This pads results to an 8-byte boundary and
854 		 * suppresses padding verification for inbound (i.e.
855 		 * decrypt) operations.
856 		 *
857 		 * NB: Not sure if the 8-byte pad boundary is a problem.
858 		 */
859 		cmd0 |= SAFE_SA_CMD0_PAD_ZERO;
860 
861 		/* XXX assert key bufs have the same size */
862 		bcopy(ses->ses_key, sa->sa_key, sizeof(sa->sa_key));
863 	}
864 
865 	if (csp->csp_auth_alg != 0) {
866 		if (crp->crp_auth_key != NULL) {
867 			safe_setup_mackey(ses, csp->csp_auth_alg,
868 			    crp->crp_auth_key, csp->csp_auth_klen);
869 		}
870 
871 		switch (csp->csp_auth_alg) {
872 		case CRYPTO_SHA1_HMAC:
873 			cmd0 |= SAFE_SA_CMD0_SHA1;
874 			cmd1 |= SAFE_SA_CMD1_HMAC;	/* NB: enable HMAC */
875 			break;
876 		}
877 
878 		/*
879 		 * Digest data is loaded from the SA and the hash
880 		 * result is saved to the state block where we
881 		 * retrieve it for return to the caller.
882 		 */
883 		/* XXX assert digest bufs have the same size */
884 		bcopy(ses->ses_hminner, sa->sa_indigest,
885 			sizeof(sa->sa_indigest));
886 		bcopy(ses->ses_hmouter, sa->sa_outdigest,
887 			sizeof(sa->sa_outdigest));
888 
889 		cmd0 |= SAFE_SA_CMD0_HSLD_SA | SAFE_SA_CMD0_SAVEHASH;
890 		re->re_flags |= SAFE_QFLAGS_COPYOUTICV;
891 	}
892 
893 	if (csp->csp_mode == CSP_MODE_ETA) {
894 		/*
895 		 * The driver only supports ETA requests where there
896 		 * is no gap between the AAD and payload.
897 		 */
898 		if (crp->crp_aad_length != 0 &&
899 		    crp->crp_aad_start + crp->crp_aad_length !=
900 		    crp->crp_payload_start) {
901 			safestats.st_lenmismatch++;
902 			err = EINVAL;
903 			goto errout;
904 		}
905 		if (crp->crp_aad_length != 0)
906 			bypass = crp->crp_aad_start;
907 		else
908 			bypass = crp->crp_payload_start;
909 		coffset = crp->crp_aad_length;
910 		oplen = crp->crp_payload_start + crp->crp_payload_length;
911 #ifdef SAFE_DEBUG
912 		if (safe_debug) {
913 			printf("AAD: skip %d, len %d, digest %d\n",
914 			    crp->crp_aad_start, crp->crp_aad_length,
915 			    crp->crp_digest_start);
916 			printf("payload: skip %d, len %d, IV %d\n",
917 			    crp->crp_payload_start, crp->crp_payload_length,
918 			    crp->crp_iv_start);
919 			printf("bypass %d coffset %d oplen %d\n",
920 				bypass, coffset, oplen);
921 		}
922 #endif
923 		if (coffset & 3) {	/* offset must be 32-bit aligned */
924 			DPRINTF(("%s: coffset %u misaligned\n",
925 				__func__, coffset));
926 			safestats.st_coffmisaligned++;
927 			err = EINVAL;
928 			goto errout;
929 		}
930 		coffset >>= 2;
931 		if (coffset > 255) {	/* offset must be <256 dwords */
932 			DPRINTF(("%s: coffset %u too big\n",
933 				__func__, coffset));
934 			safestats.st_cofftoobig++;
935 			err = EINVAL;
936 			goto errout;
937 		}
938 		/*
939 		 * Tell the hardware to copy the header to the output.
940 		 * The header is defined as the data from the end of
941 		 * the bypass to the start of data to be encrypted.
942 		 * Typically this is the inline IV.  Note that you need
943 		 * to do this even if src+dst are the same; it appears
944 		 * that w/o this bit the crypted data is written
945 		 * immediately after the bypass data.
946 		 */
947 		cmd1 |= SAFE_SA_CMD1_HDRCOPY;
948 		/*
949 		 * Disable IP header mutable bit handling.  This is
950 		 * needed to get correct HMAC calculations.
951 		 */
952 		cmd1 |= SAFE_SA_CMD1_MUTABLE;
953 	} else {
954 		bypass = crp->crp_payload_start;
955 		oplen = bypass + crp->crp_payload_length;
956 		coffset = 0;
957 	}
958 	/* XXX verify multiple of 4 when using s/g */
959 	if (bypass > 96) {		/* bypass offset must be <= 96 bytes */
960 		DPRINTF(("%s: bypass %u too big\n", __func__, bypass));
961 		safestats.st_bypasstoobig++;
962 		err = EINVAL;
963 		goto errout;
964 	}
965 
966 	if (bus_dmamap_create(sc->sc_srcdmat, BUS_DMA_NOWAIT, &re->re_src_map)) {
967 		safestats.st_nomap++;
968 		err = ENOMEM;
969 		goto errout;
970 	}
971 	if (bus_dmamap_load_crp(sc->sc_srcdmat, re->re_src_map, crp, safe_op_cb,
972 	    &re->re_src, BUS_DMA_NOWAIT) != 0) {
973 		bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map);
974 		re->re_src_map = NULL;
975 		safestats.st_noload++;
976 		err = ENOMEM;
977 		goto errout;
978 	}
979 	re->re_src_mapsize = crypto_buffer_len(&crp->crp_buf);
980 	nicealign = safe_dmamap_aligned(&re->re_src);
981 	uniform = safe_dmamap_uniform(&re->re_src);
982 
983 	DPRINTF(("src nicealign %u uniform %u nsegs %u\n",
984 		nicealign, uniform, re->re_src.nsegs));
985 	if (re->re_src.nsegs > 1) {
986 		re->re_desc.d_src = sc->sc_spalloc.dma_paddr +
987 			((caddr_t) sc->sc_spfree - (caddr_t) sc->sc_spring);
988 		for (i = 0; i < re->re_src_nsegs; i++) {
989 			/* NB: no need to check if there's space */
990 			pd = sc->sc_spfree;
991 			if (++(sc->sc_spfree) == sc->sc_springtop)
992 				sc->sc_spfree = sc->sc_spring;
993 
994 			KASSERT((pd->pd_flags&3) == 0 ||
995 				(pd->pd_flags&3) == SAFE_PD_DONE,
996 				("bogus source particle descriptor; flags %x",
997 				pd->pd_flags));
998 			pd->pd_addr = re->re_src_segs[i].ds_addr;
999 			pd->pd_size = re->re_src_segs[i].ds_len;
1000 			pd->pd_flags = SAFE_PD_READY;
1001 		}
1002 		cmd0 |= SAFE_SA_CMD0_IGATHER;
1003 	} else {
1004 		/*
1005 		 * No need for gather, reference the operand directly.
1006 		 */
1007 		re->re_desc.d_src = re->re_src_segs[0].ds_addr;
1008 	}
1009 
1010 	if (csp->csp_mode == CSP_MODE_DIGEST) {
1011 		/*
1012 		 * Hash op; no destination needed.
1013 		 */
1014 	} else {
1015 		if (nicealign && uniform == 1) {
1016 			/*
1017 			 * Source layout is suitable for direct
1018 			 * sharing of the DMA map and segment list.
1019 			 */
1020 			re->re_dst = re->re_src;
1021 		} else if (nicealign && uniform == 2) {
1022 			/*
1023 			 * The source is properly aligned but requires a
1024 			 * different particle list to handle DMA of the
1025 			 * result.  Create a new map and do the load to
1026 			 * create the segment list.  The particle
1027 			 * descriptor setup code below will handle the
1028 			 * rest.
1029 			 */
1030 			if (bus_dmamap_create(sc->sc_dstdmat, BUS_DMA_NOWAIT,
1031 			    &re->re_dst_map)) {
1032 				safestats.st_nomap++;
1033 				err = ENOMEM;
1034 				goto errout;
1035 			}
1036 			if (bus_dmamap_load_crp(sc->sc_dstdmat, re->re_dst_map,
1037 			    crp, safe_op_cb, &re->re_dst, BUS_DMA_NOWAIT) !=
1038 			    0) {
1039 				bus_dmamap_destroy(sc->sc_dstdmat,
1040 				    re->re_dst_map);
1041 				re->re_dst_map = NULL;
1042 				safestats.st_noload++;
1043 				err = ENOMEM;
1044 				goto errout;
1045 			}
1046 		} else if (crp->crp_buf.cb_type == CRYPTO_BUF_MBUF) {
1047 			int totlen, len;
1048 			struct mbuf *m, *top, **mp;
1049 
1050 			/*
1051 			 * DMA constraints require that we allocate a
1052 			 * new mbuf chain for the destination.  We
1053 			 * allocate an entire new set of mbufs of
1054 			 * optimal/required size and then tell the
1055 			 * hardware to copy any bits that are not
1056 			 * created as a byproduct of the operation.
1057 			 */
1058 			if (!nicealign)
1059 				safestats.st_unaligned++;
1060 			if (!uniform)
1061 				safestats.st_notuniform++;
1062 			totlen = re->re_src_mapsize;
1063 			if (crp->crp_buf.cb_mbuf->m_flags & M_PKTHDR) {
1064 				len = MHLEN;
1065 				MGETHDR(m, M_NOWAIT, MT_DATA);
1066 				if (m && !m_dup_pkthdr(m, crp->crp_buf.cb_mbuf,
1067 				    M_NOWAIT)) {
1068 					m_free(m);
1069 					m = NULL;
1070 				}
1071 			} else {
1072 				len = MLEN;
1073 				MGET(m, M_NOWAIT, MT_DATA);
1074 			}
1075 			if (m == NULL) {
1076 				safestats.st_nombuf++;
1077 				err = sc->sc_nqchip ? ERESTART : ENOMEM;
1078 				goto errout;
1079 			}
1080 			if (totlen >= MINCLSIZE) {
1081 				if (!(MCLGET(m, M_NOWAIT))) {
1082 					m_free(m);
1083 					safestats.st_nomcl++;
1084 					err = sc->sc_nqchip ?
1085 					    ERESTART : ENOMEM;
1086 					goto errout;
1087 				}
1088 				len = MCLBYTES;
1089 			}
1090 			m->m_len = len;
1091 			top = NULL;
1092 			mp = &top;
1093 
1094 			while (totlen > 0) {
1095 				if (top) {
1096 					MGET(m, M_NOWAIT, MT_DATA);
1097 					if (m == NULL) {
1098 						m_freem(top);
1099 						safestats.st_nombuf++;
1100 						err = sc->sc_nqchip ?
1101 						    ERESTART : ENOMEM;
1102 						goto errout;
1103 					}
1104 					len = MLEN;
1105 				}
1106 				if (top && totlen >= MINCLSIZE) {
1107 					if (!(MCLGET(m, M_NOWAIT))) {
1108 						*mp = m;
1109 						m_freem(top);
1110 						safestats.st_nomcl++;
1111 						err = sc->sc_nqchip ?
1112 						    ERESTART : ENOMEM;
1113 						goto errout;
1114 					}
1115 					len = MCLBYTES;
1116 				}
1117 				m->m_len = len = min(totlen, len);
1118 				totlen -= len;
1119 				*mp = m;
1120 				mp = &m->m_next;
1121 			}
1122 			re->re_dst_m = top;
1123 			if (bus_dmamap_create(sc->sc_dstdmat,
1124 			    BUS_DMA_NOWAIT, &re->re_dst_map) != 0) {
1125 				safestats.st_nomap++;
1126 				err = ENOMEM;
1127 				goto errout;
1128 			}
1129 			if (bus_dmamap_load_mbuf_sg(sc->sc_dstdmat,
1130 			    re->re_dst_map, top, re->re_dst_segs,
1131 			    &re->re_dst_nsegs, 0) != 0) {
1132 				bus_dmamap_destroy(sc->sc_dstdmat,
1133 				    re->re_dst_map);
1134 				re->re_dst_map = NULL;
1135 				safestats.st_noload++;
1136 				err = ENOMEM;
1137 				goto errout;
1138 			}
1139 			re->re_dst_mapsize = re->re_src_mapsize;
1140 			if (re->re_src.mapsize > oplen) {
1141 				/*
1142 				 * There's data following what the
1143 				 * hardware will copy for us.  If this
1144 				 * isn't just the ICV (that's going to
1145 				 * be written on completion), copy it
1146 				 * to the new mbufs
1147 				 */
1148 				if (!(csp->csp_mode == CSP_MODE_ETA &&
1149 				    (re->re_src.mapsize-oplen) == ses->ses_mlen &&
1150 				    crp->crp_digest_start == oplen))
1151 					safe_mcopy(crp->crp_buf.cb_mbuf,
1152 					    re->re_dst_m, oplen);
1153 				else
1154 					safestats.st_noicvcopy++;
1155 			}
1156 		} else {
1157 			if (!nicealign) {
1158 				safestats.st_iovmisaligned++;
1159 				err = EINVAL;
1160 				goto errout;
1161 			} else {
1162 				/*
1163 				 * There's no way to handle the DMA
1164 				 * requirements with this uio.  We
1165 				 * could create a separate DMA area for
1166 				 * the result and then copy it back,
1167 				 * but for now we just bail and return
1168 				 * an error.  Note that uio requests
1169 				 * > SAFE_MAX_DSIZE are handled because
1170 				 * the DMA map and segment list for the
1171 				 * destination wil result in a
1172 				 * destination particle list that does
1173 				 * the necessary scatter DMA.
1174 				 */
1175 				safestats.st_iovnotuniform++;
1176 				err = EINVAL;
1177 				goto errout;
1178 			}
1179 		}
1180 
1181 		if (re->re_dst.nsegs > 1) {
1182 			re->re_desc.d_dst = sc->sc_dpalloc.dma_paddr +
1183 			    ((caddr_t) sc->sc_dpfree - (caddr_t) sc->sc_dpring);
1184 			for (i = 0; i < re->re_dst_nsegs; i++) {
1185 				pd = sc->sc_dpfree;
1186 				KASSERT((pd->pd_flags&3) == 0 ||
1187 					(pd->pd_flags&3) == SAFE_PD_DONE,
1188 					("bogus dest particle descriptor; flags %x",
1189 						pd->pd_flags));
1190 				if (++(sc->sc_dpfree) == sc->sc_dpringtop)
1191 					sc->sc_dpfree = sc->sc_dpring;
1192 				pd->pd_addr = re->re_dst_segs[i].ds_addr;
1193 				pd->pd_flags = SAFE_PD_READY;
1194 			}
1195 			cmd0 |= SAFE_SA_CMD0_OSCATTER;
1196 		} else {
1197 			/*
1198 			 * No need for scatter, reference the operand directly.
1199 			 */
1200 			re->re_desc.d_dst = re->re_dst_segs[0].ds_addr;
1201 		}
1202 	}
1203 
1204 	/*
1205 	 * All done with setup; fillin the SA command words
1206 	 * and the packet engine descriptor.  The operation
1207 	 * is now ready for submission to the hardware.
1208 	 */
1209 	sa->sa_cmd0 = cmd0 | SAFE_SA_CMD0_IPCI | SAFE_SA_CMD0_OPCI;
1210 	sa->sa_cmd1 = cmd1
1211 		    | (coffset << SAFE_SA_CMD1_OFFSET_S)
1212 		    | SAFE_SA_CMD1_SAREV1	/* Rev 1 SA data structure */
1213 		    | SAFE_SA_CMD1_SRPCI
1214 		    ;
1215 	/*
1216 	 * NB: the order of writes is important here.  In case the
1217 	 * chip is scanning the ring because of an outstanding request
1218 	 * it might nab this one too.  In that case we need to make
1219 	 * sure the setup is complete before we write the length
1220 	 * field of the descriptor as it signals the descriptor is
1221 	 * ready for processing.
1222 	 */
1223 	re->re_desc.d_csr = SAFE_PE_CSR_READY | SAFE_PE_CSR_SAPCI;
1224 	if (csp->csp_auth_alg != 0)
1225 		re->re_desc.d_csr |= SAFE_PE_CSR_LOADSA | SAFE_PE_CSR_HASHFINAL;
1226 	re->re_desc.d_len = oplen
1227 			  | SAFE_PE_LEN_READY
1228 			  | (bypass << SAFE_PE_LEN_BYPASS_S)
1229 			  ;
1230 
1231 	safestats.st_ipackets++;
1232 	safestats.st_ibytes += oplen;
1233 
1234 	if (++(sc->sc_front) == sc->sc_ringtop)
1235 		sc->sc_front = sc->sc_ring;
1236 
1237 	/* XXX honor batching */
1238 	safe_feed(sc, re);
1239 	mtx_unlock(&sc->sc_ringmtx);
1240 	return (0);
1241 
1242 errout:
1243 	if (re->re_dst_m != NULL)
1244 		m_freem(re->re_dst_m);
1245 
1246 	if (re->re_dst_map != NULL && re->re_dst_map != re->re_src_map) {
1247 		bus_dmamap_unload(sc->sc_dstdmat, re->re_dst_map);
1248 		bus_dmamap_destroy(sc->sc_dstdmat, re->re_dst_map);
1249 	}
1250 	if (re->re_src_map != NULL) {
1251 		bus_dmamap_unload(sc->sc_srcdmat, re->re_src_map);
1252 		bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map);
1253 	}
1254 	mtx_unlock(&sc->sc_ringmtx);
1255 	if (err != ERESTART) {
1256 		crp->crp_etype = err;
1257 		crypto_done(crp);
1258 		err = 0;
1259 	} else {
1260 		sc->sc_needwakeup |= CRYPTO_SYMQ;
1261 	}
1262 	return (err);
1263 }
1264 
1265 static void
1266 safe_callback(struct safe_softc *sc, struct safe_ringentry *re)
1267 {
1268 	const struct crypto_session_params *csp;
1269 	struct cryptop *crp = (struct cryptop *)re->re_crp;
1270 	struct safe_session *ses;
1271 	uint8_t hash[HASH_MAX_LEN];
1272 
1273 	ses = crypto_get_driver_session(crp->crp_session);
1274 	csp = crypto_get_params(crp->crp_session);
1275 
1276 	safestats.st_opackets++;
1277 	safestats.st_obytes += re->re_dst.mapsize;
1278 
1279 	safe_dma_sync(&sc->sc_ringalloc,
1280 		BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1281 	if (re->re_desc.d_csr & SAFE_PE_CSR_STATUS) {
1282 		device_printf(sc->sc_dev, "csr 0x%x cmd0 0x%x cmd1 0x%x\n",
1283 			re->re_desc.d_csr,
1284 			re->re_sa.sa_cmd0, re->re_sa.sa_cmd1);
1285 		safestats.st_peoperr++;
1286 		crp->crp_etype = EIO;		/* something more meaningful? */
1287 	}
1288 
1289 	/*
1290 	 * XXX: Should crp_buf.cb_mbuf be updated to re->re_dst_m if
1291 	 * it is non-NULL?
1292 	 */
1293 
1294 	if (re->re_dst_map != NULL && re->re_dst_map != re->re_src_map) {
1295 		bus_dmamap_sync(sc->sc_dstdmat, re->re_dst_map,
1296 		    BUS_DMASYNC_POSTREAD);
1297 		bus_dmamap_unload(sc->sc_dstdmat, re->re_dst_map);
1298 		bus_dmamap_destroy(sc->sc_dstdmat, re->re_dst_map);
1299 	}
1300 	bus_dmamap_sync(sc->sc_srcdmat, re->re_src_map, BUS_DMASYNC_POSTWRITE);
1301 	bus_dmamap_unload(sc->sc_srcdmat, re->re_src_map);
1302 	bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map);
1303 
1304 	if (re->re_flags & SAFE_QFLAGS_COPYOUTICV) {
1305 		if (csp->csp_auth_alg == CRYPTO_SHA1_HMAC) {
1306 			/*
1307 			 * SHA-1 ICV's are byte-swapped; fix 'em up
1308 			 * before copying them to their destination.
1309 			 */
1310 			re->re_sastate.sa_saved_indigest[0] =
1311 			    bswap32(re->re_sastate.sa_saved_indigest[0]);
1312 			re->re_sastate.sa_saved_indigest[1] =
1313 			    bswap32(re->re_sastate.sa_saved_indigest[1]);
1314 			re->re_sastate.sa_saved_indigest[2] =
1315 			    bswap32(re->re_sastate.sa_saved_indigest[2]);
1316 		}
1317 
1318 		if (crp->crp_op & CRYPTO_OP_VERIFY_DIGEST) {
1319 			crypto_copydata(crp, crp->crp_digest_start,
1320 			    ses->ses_mlen, hash);
1321 			if (timingsafe_bcmp(re->re_sastate.sa_saved_indigest,
1322 			    hash, ses->ses_mlen) != 0)
1323 				crp->crp_etype = EBADMSG;
1324 		} else
1325 			crypto_copyback(crp, crp->crp_digest_start,
1326 			    ses->ses_mlen, re->re_sastate.sa_saved_indigest);
1327 	}
1328 	crypto_done(crp);
1329 }
1330 
1331 /*
1332  * Copy all data past offset from srcm to dstm.
1333  */
1334 static void
1335 safe_mcopy(struct mbuf *srcm, struct mbuf *dstm, u_int offset)
1336 {
1337 	u_int j, dlen, slen;
1338 	caddr_t dptr, sptr;
1339 
1340 	/*
1341 	 * Advance src and dst to offset.
1342 	 */
1343 	j = offset;
1344 	while (j >= srcm->m_len) {
1345 		j -= srcm->m_len;
1346 		srcm = srcm->m_next;
1347 		if (srcm == NULL)
1348 			return;
1349 	}
1350 	sptr = mtod(srcm, caddr_t) + j;
1351 	slen = srcm->m_len - j;
1352 
1353 	j = offset;
1354 	while (j >= dstm->m_len) {
1355 		j -= dstm->m_len;
1356 		dstm = dstm->m_next;
1357 		if (dstm == NULL)
1358 			return;
1359 	}
1360 	dptr = mtod(dstm, caddr_t) + j;
1361 	dlen = dstm->m_len - j;
1362 
1363 	/*
1364 	 * Copy everything that remains.
1365 	 */
1366 	for (;;) {
1367 		j = min(slen, dlen);
1368 		bcopy(sptr, dptr, j);
1369 		if (slen == j) {
1370 			srcm = srcm->m_next;
1371 			if (srcm == NULL)
1372 				return;
1373 			sptr = srcm->m_data;
1374 			slen = srcm->m_len;
1375 		} else
1376 			sptr += j, slen -= j;
1377 		if (dlen == j) {
1378 			dstm = dstm->m_next;
1379 			if (dstm == NULL)
1380 				return;
1381 			dptr = dstm->m_data;
1382 			dlen = dstm->m_len;
1383 		} else
1384 			dptr += j, dlen -= j;
1385 	}
1386 }
1387 
1388 #ifndef SAFE_NO_RNG
1389 #define	SAFE_RNG_MAXWAIT	1000
1390 
1391 static void
1392 safe_rng_init(struct safe_softc *sc)
1393 {
1394 	u_int32_t w, v;
1395 	int i;
1396 
1397 	WRITE_REG(sc, SAFE_RNG_CTRL, 0);
1398 	/* use default value according to the manual */
1399 	WRITE_REG(sc, SAFE_RNG_CNFG, 0x834);	/* magic from SafeNet */
1400 	WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0);
1401 
1402 	/*
1403 	 * There is a bug in rev 1.0 of the 1140 that when the RNG
1404 	 * is brought out of reset the ready status flag does not
1405 	 * work until the RNG has finished its internal initialization.
1406 	 *
1407 	 * So in order to determine the device is through its
1408 	 * initialization we must read the data register, using the
1409 	 * status reg in the read in case it is initialized.  Then read
1410 	 * the data register until it changes from the first read.
1411 	 * Once it changes read the data register until it changes
1412 	 * again.  At this time the RNG is considered initialized.
1413 	 * This could take between 750ms - 1000ms in time.
1414 	 */
1415 	i = 0;
1416 	w = READ_REG(sc, SAFE_RNG_OUT);
1417 	do {
1418 		v = READ_REG(sc, SAFE_RNG_OUT);
1419 		if (v != w) {
1420 			w = v;
1421 			break;
1422 		}
1423 		DELAY(10);
1424 	} while (++i < SAFE_RNG_MAXWAIT);
1425 
1426 	/* Wait Until data changes again */
1427 	i = 0;
1428 	do {
1429 		v = READ_REG(sc, SAFE_RNG_OUT);
1430 		if (v != w)
1431 			break;
1432 		DELAY(10);
1433 	} while (++i < SAFE_RNG_MAXWAIT);
1434 }
1435 
1436 static __inline void
1437 safe_rng_disable_short_cycle(struct safe_softc *sc)
1438 {
1439 	WRITE_REG(sc, SAFE_RNG_CTRL,
1440 		READ_REG(sc, SAFE_RNG_CTRL) &~ SAFE_RNG_CTRL_SHORTEN);
1441 }
1442 
1443 static __inline void
1444 safe_rng_enable_short_cycle(struct safe_softc *sc)
1445 {
1446 	WRITE_REG(sc, SAFE_RNG_CTRL,
1447 		READ_REG(sc, SAFE_RNG_CTRL) | SAFE_RNG_CTRL_SHORTEN);
1448 }
1449 
1450 static __inline u_int32_t
1451 safe_rng_read(struct safe_softc *sc)
1452 {
1453 	int i;
1454 
1455 	i = 0;
1456 	while (READ_REG(sc, SAFE_RNG_STAT) != 0 && ++i < SAFE_RNG_MAXWAIT)
1457 		;
1458 	return READ_REG(sc, SAFE_RNG_OUT);
1459 }
1460 
1461 static void
1462 safe_rng(void *arg)
1463 {
1464 	struct safe_softc *sc = arg;
1465 	u_int32_t buf[SAFE_RNG_MAXBUFSIZ];	/* NB: maybe move to softc */
1466 	u_int maxwords;
1467 	int i;
1468 
1469 	safestats.st_rng++;
1470 	/*
1471 	 * Fetch the next block of data.
1472 	 */
1473 	maxwords = safe_rngbufsize;
1474 	if (maxwords > SAFE_RNG_MAXBUFSIZ)
1475 		maxwords = SAFE_RNG_MAXBUFSIZ;
1476 retry:
1477 	for (i = 0; i < maxwords; i++)
1478 		buf[i] = safe_rng_read(sc);
1479 	/*
1480 	 * Check the comparator alarm count and reset the h/w if
1481 	 * it exceeds our threshold.  This guards against the
1482 	 * hardware oscillators resonating with external signals.
1483 	 */
1484 	if (READ_REG(sc, SAFE_RNG_ALM_CNT) > safe_rngmaxalarm) {
1485 		u_int32_t freq_inc, w;
1486 
1487 		DPRINTF(("%s: alarm count %u exceeds threshold %u\n", __func__,
1488 			READ_REG(sc, SAFE_RNG_ALM_CNT), safe_rngmaxalarm));
1489 		safestats.st_rngalarm++;
1490 		safe_rng_enable_short_cycle(sc);
1491 		freq_inc = 18;
1492 		for (i = 0; i < 64; i++) {
1493 			w = READ_REG(sc, SAFE_RNG_CNFG);
1494 			freq_inc = ((w + freq_inc) & 0x3fL);
1495 			w = ((w & ~0x3fL) | freq_inc);
1496 			WRITE_REG(sc, SAFE_RNG_CNFG, w);
1497 
1498 			WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0);
1499 
1500 			(void) safe_rng_read(sc);
1501 			DELAY(25);
1502 
1503 			if (READ_REG(sc, SAFE_RNG_ALM_CNT) == 0) {
1504 				safe_rng_disable_short_cycle(sc);
1505 				goto retry;
1506 			}
1507 			freq_inc = 1;
1508 		}
1509 		safe_rng_disable_short_cycle(sc);
1510 	} else
1511 		WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0);
1512 
1513 	(*sc->sc_harvest)(sc->sc_rndtest, buf, maxwords*sizeof (u_int32_t));
1514 	callout_reset(&sc->sc_rngto,
1515 		hz * (safe_rnginterval ? safe_rnginterval : 1), safe_rng, sc);
1516 }
1517 #endif /* SAFE_NO_RNG */
1518 
1519 static void
1520 safe_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1521 {
1522 	bus_addr_t *paddr = (bus_addr_t*) arg;
1523 	*paddr = segs->ds_addr;
1524 }
1525 
1526 static int
1527 safe_dma_malloc(
1528 	struct safe_softc *sc,
1529 	bus_size_t size,
1530 	struct safe_dma_alloc *dma,
1531 	int mapflags
1532 )
1533 {
1534 	int r;
1535 
1536 	r = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev),	/* parent */
1537 			       sizeof(u_int32_t), 0,	/* alignment, bounds */
1538 			       BUS_SPACE_MAXADDR_32BIT,	/* lowaddr */
1539 			       BUS_SPACE_MAXADDR,	/* highaddr */
1540 			       NULL, NULL,		/* filter, filterarg */
1541 			       size,			/* maxsize */
1542 			       1,			/* nsegments */
1543 			       size,			/* maxsegsize */
1544 			       BUS_DMA_ALLOCNOW,	/* flags */
1545 			       NULL, NULL,		/* locking */
1546 			       &dma->dma_tag);
1547 	if (r != 0) {
1548 		device_printf(sc->sc_dev, "safe_dma_malloc: "
1549 			"bus_dma_tag_create failed; error %u\n", r);
1550 		goto fail_0;
1551 	}
1552 
1553 	r = bus_dmamem_alloc(dma->dma_tag, (void**) &dma->dma_vaddr,
1554 			     BUS_DMA_NOWAIT, &dma->dma_map);
1555 	if (r != 0) {
1556 		device_printf(sc->sc_dev, "safe_dma_malloc: "
1557 			"bus_dmammem_alloc failed; size %ju, error %u\n",
1558 			(uintmax_t)size, r);
1559 		goto fail_1;
1560 	}
1561 
1562 	r = bus_dmamap_load(dma->dma_tag, dma->dma_map, dma->dma_vaddr,
1563 		            size,
1564 			    safe_dmamap_cb,
1565 			    &dma->dma_paddr,
1566 			    mapflags | BUS_DMA_NOWAIT);
1567 	if (r != 0) {
1568 		device_printf(sc->sc_dev, "safe_dma_malloc: "
1569 			"bus_dmamap_load failed; error %u\n", r);
1570 		goto fail_2;
1571 	}
1572 
1573 	dma->dma_size = size;
1574 	return (0);
1575 
1576 	bus_dmamap_unload(dma->dma_tag, dma->dma_map);
1577 fail_2:
1578 	bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
1579 fail_1:
1580 	bus_dma_tag_destroy(dma->dma_tag);
1581 fail_0:
1582 	dma->dma_tag = NULL;
1583 	return (r);
1584 }
1585 
1586 static void
1587 safe_dma_free(struct safe_softc *sc, struct safe_dma_alloc *dma)
1588 {
1589 	bus_dmamap_unload(dma->dma_tag, dma->dma_map);
1590 	bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
1591 	bus_dma_tag_destroy(dma->dma_tag);
1592 }
1593 
1594 /*
1595  * Resets the board.  Values in the regesters are left as is
1596  * from the reset (i.e. initial values are assigned elsewhere).
1597  */
1598 static void
1599 safe_reset_board(struct safe_softc *sc)
1600 {
1601 	u_int32_t v;
1602 	/*
1603 	 * Reset the device.  The manual says no delay
1604 	 * is needed between marking and clearing reset.
1605 	 */
1606 	v = READ_REG(sc, SAFE_PE_DMACFG) &~
1607 		(SAFE_PE_DMACFG_PERESET | SAFE_PE_DMACFG_PDRRESET |
1608 		 SAFE_PE_DMACFG_SGRESET);
1609 	WRITE_REG(sc, SAFE_PE_DMACFG, v
1610 				    | SAFE_PE_DMACFG_PERESET
1611 				    | SAFE_PE_DMACFG_PDRRESET
1612 				    | SAFE_PE_DMACFG_SGRESET);
1613 	WRITE_REG(sc, SAFE_PE_DMACFG, v);
1614 }
1615 
1616 /*
1617  * Initialize registers we need to touch only once.
1618  */
1619 static void
1620 safe_init_board(struct safe_softc *sc)
1621 {
1622 	u_int32_t v, dwords;
1623 
1624 	v = READ_REG(sc, SAFE_PE_DMACFG);
1625 	v &=~ SAFE_PE_DMACFG_PEMODE;
1626 	v |= SAFE_PE_DMACFG_FSENA		/* failsafe enable */
1627 	  |  SAFE_PE_DMACFG_GPRPCI		/* gather ring on PCI */
1628 	  |  SAFE_PE_DMACFG_SPRPCI		/* scatter ring on PCI */
1629 	  |  SAFE_PE_DMACFG_ESDESC		/* endian-swap descriptors */
1630 	  |  SAFE_PE_DMACFG_ESSA		/* endian-swap SA's */
1631 	  |  SAFE_PE_DMACFG_ESPDESC		/* endian-swap part. desc's */
1632 	  ;
1633 	WRITE_REG(sc, SAFE_PE_DMACFG, v);
1634 #if 0
1635 	/* XXX select byte swap based on host byte order */
1636 	WRITE_REG(sc, SAFE_ENDIAN, 0x1b);
1637 #endif
1638 	if (sc->sc_chiprev == SAFE_REV(1,0)) {
1639 		/*
1640 		 * Avoid large PCI DMA transfers.  Rev 1.0 has a bug where
1641 		 * "target mode transfers" done while the chip is DMA'ing
1642 		 * >1020 bytes cause the hardware to lockup.  To avoid this
1643 		 * we reduce the max PCI transfer size and use small source
1644 		 * particle descriptors (<= 256 bytes).
1645 		 */
1646 		WRITE_REG(sc, SAFE_DMA_CFG, 256);
1647 		device_printf(sc->sc_dev,
1648 			"Reduce max DMA size to %u words for rev %u.%u WAR\n",
1649 			(READ_REG(sc, SAFE_DMA_CFG)>>2) & 0xff,
1650 			SAFE_REV_MAJ(sc->sc_chiprev),
1651 			SAFE_REV_MIN(sc->sc_chiprev));
1652 	}
1653 
1654 	/* NB: operands+results are overlaid */
1655 	WRITE_REG(sc, SAFE_PE_PDRBASE, sc->sc_ringalloc.dma_paddr);
1656 	WRITE_REG(sc, SAFE_PE_RDRBASE, sc->sc_ringalloc.dma_paddr);
1657 	/*
1658 	 * Configure ring entry size and number of items in the ring.
1659 	 */
1660 	KASSERT((sizeof(struct safe_ringentry) % sizeof(u_int32_t)) == 0,
1661 		("PE ring entry not 32-bit aligned!"));
1662 	dwords = sizeof(struct safe_ringentry) / sizeof(u_int32_t);
1663 	WRITE_REG(sc, SAFE_PE_RINGCFG,
1664 		(dwords << SAFE_PE_RINGCFG_OFFSET_S) | SAFE_MAX_NQUEUE);
1665 	WRITE_REG(sc, SAFE_PE_RINGPOLL, 0);	/* disable polling */
1666 
1667 	WRITE_REG(sc, SAFE_PE_GRNGBASE, sc->sc_spalloc.dma_paddr);
1668 	WRITE_REG(sc, SAFE_PE_SRNGBASE, sc->sc_dpalloc.dma_paddr);
1669 	WRITE_REG(sc, SAFE_PE_PARTSIZE,
1670 		(SAFE_TOTAL_DPART<<16) | SAFE_TOTAL_SPART);
1671 	/*
1672 	 * NB: destination particles are fixed size.  We use
1673 	 *     an mbuf cluster and require all results go to
1674 	 *     clusters or smaller.
1675 	 */
1676 	WRITE_REG(sc, SAFE_PE_PARTCFG, SAFE_MAX_DSIZE);
1677 
1678 	/* it's now safe to enable PE mode, do it */
1679 	WRITE_REG(sc, SAFE_PE_DMACFG, v | SAFE_PE_DMACFG_PEMODE);
1680 
1681 	/*
1682 	 * Configure hardware to use level-triggered interrupts and
1683 	 * to interrupt after each descriptor is processed.
1684 	 */
1685 	WRITE_REG(sc, SAFE_HI_CFG, SAFE_HI_CFG_LEVEL);
1686 	WRITE_REG(sc, SAFE_HI_DESC_CNT, 1);
1687 	WRITE_REG(sc, SAFE_HI_MASK, SAFE_INT_PE_DDONE | SAFE_INT_PE_ERROR);
1688 }
1689 
1690 /*
1691  * Init PCI registers
1692  */
1693 static void
1694 safe_init_pciregs(device_t dev)
1695 {
1696 }
1697 
1698 /*
1699  * Clean up after a chip crash.
1700  * It is assumed that the caller in splimp()
1701  */
1702 static void
1703 safe_cleanchip(struct safe_softc *sc)
1704 {
1705 
1706 	if (sc->sc_nqchip != 0) {
1707 		struct safe_ringentry *re = sc->sc_back;
1708 
1709 		while (re != sc->sc_front) {
1710 			if (re->re_desc.d_csr != 0)
1711 				safe_free_entry(sc, re);
1712 			if (++re == sc->sc_ringtop)
1713 				re = sc->sc_ring;
1714 		}
1715 		sc->sc_back = re;
1716 		sc->sc_nqchip = 0;
1717 	}
1718 }
1719 
1720 /*
1721  * free a safe_q
1722  * It is assumed that the caller is within splimp().
1723  */
1724 static int
1725 safe_free_entry(struct safe_softc *sc, struct safe_ringentry *re)
1726 {
1727 	struct cryptop *crp;
1728 
1729 	/*
1730 	 * Free header MCR
1731 	 */
1732 	if (re->re_dst_m != NULL)
1733 		m_freem(re->re_dst_m);
1734 
1735 	crp = (struct cryptop *)re->re_crp;
1736 
1737 	re->re_desc.d_csr = 0;
1738 
1739 	crp->crp_etype = EFAULT;
1740 	crypto_done(crp);
1741 	return(0);
1742 }
1743 
1744 /*
1745  * Routine to reset the chip and clean up.
1746  * It is assumed that the caller is in splimp()
1747  */
1748 static void
1749 safe_totalreset(struct safe_softc *sc)
1750 {
1751 	safe_reset_board(sc);
1752 	safe_init_board(sc);
1753 	safe_cleanchip(sc);
1754 }
1755 
1756 /*
1757  * Is the operand suitable aligned for direct DMA.  Each
1758  * segment must be aligned on a 32-bit boundary and all
1759  * but the last segment must be a multiple of 4 bytes.
1760  */
1761 static int
1762 safe_dmamap_aligned(const struct safe_operand *op)
1763 {
1764 	int i;
1765 
1766 	for (i = 0; i < op->nsegs; i++) {
1767 		if (op->segs[i].ds_addr & 3)
1768 			return (0);
1769 		if (i != (op->nsegs - 1) && (op->segs[i].ds_len & 3))
1770 			return (0);
1771 	}
1772 	return (1);
1773 }
1774 
1775 /*
1776  * Is the operand suitable for direct DMA as the destination
1777  * of an operation.  The hardware requires that each ``particle''
1778  * but the last in an operation result have the same size.  We
1779  * fix that size at SAFE_MAX_DSIZE bytes.  This routine returns
1780  * 0 if some segment is not a multiple of this size, 1 if all
1781  * segments are exactly this size, or 2 if segments are at worst
1782  * a multiple of this size.
1783  */
1784 static int
1785 safe_dmamap_uniform(const struct safe_operand *op)
1786 {
1787 	int result = 1;
1788 
1789 	if (op->nsegs > 0) {
1790 		int i;
1791 
1792 		for (i = 0; i < op->nsegs-1; i++) {
1793 			if (op->segs[i].ds_len % SAFE_MAX_DSIZE)
1794 				return (0);
1795 			if (op->segs[i].ds_len != SAFE_MAX_DSIZE)
1796 				result = 2;
1797 		}
1798 	}
1799 	return (result);
1800 }
1801 
1802 #ifdef SAFE_DEBUG
1803 static void
1804 safe_dump_dmastatus(struct safe_softc *sc, const char *tag)
1805 {
1806 	printf("%s: ENDIAN 0x%x SRC 0x%x DST 0x%x STAT 0x%x\n"
1807 		, tag
1808 		, READ_REG(sc, SAFE_DMA_ENDIAN)
1809 		, READ_REG(sc, SAFE_DMA_SRCADDR)
1810 		, READ_REG(sc, SAFE_DMA_DSTADDR)
1811 		, READ_REG(sc, SAFE_DMA_STAT)
1812 	);
1813 }
1814 
1815 static void
1816 safe_dump_intrstate(struct safe_softc *sc, const char *tag)
1817 {
1818 	printf("%s: HI_CFG 0x%x HI_MASK 0x%x HI_DESC_CNT 0x%x HU_STAT 0x%x HM_STAT 0x%x\n"
1819 		, tag
1820 		, READ_REG(sc, SAFE_HI_CFG)
1821 		, READ_REG(sc, SAFE_HI_MASK)
1822 		, READ_REG(sc, SAFE_HI_DESC_CNT)
1823 		, READ_REG(sc, SAFE_HU_STAT)
1824 		, READ_REG(sc, SAFE_HM_STAT)
1825 	);
1826 }
1827 
1828 static void
1829 safe_dump_ringstate(struct safe_softc *sc, const char *tag)
1830 {
1831 	u_int32_t estat = READ_REG(sc, SAFE_PE_ERNGSTAT);
1832 
1833 	/* NB: assume caller has lock on ring */
1834 	printf("%s: ERNGSTAT %x (next %u) back %lu front %lu\n",
1835 		tag,
1836 		estat, (estat >> SAFE_PE_ERNGSTAT_NEXT_S),
1837 		(unsigned long)(sc->sc_back - sc->sc_ring),
1838 		(unsigned long)(sc->sc_front - sc->sc_ring));
1839 }
1840 
1841 static void
1842 safe_dump_request(struct safe_softc *sc, const char* tag, struct safe_ringentry *re)
1843 {
1844 	int ix, nsegs;
1845 
1846 	ix = re - sc->sc_ring;
1847 	printf("%s: %p (%u): csr %x src %x dst %x sa %x len %x\n"
1848 		, tag
1849 		, re, ix
1850 		, re->re_desc.d_csr
1851 		, re->re_desc.d_src
1852 		, re->re_desc.d_dst
1853 		, re->re_desc.d_sa
1854 		, re->re_desc.d_len
1855 	);
1856 	if (re->re_src.nsegs > 1) {
1857 		ix = (re->re_desc.d_src - sc->sc_spalloc.dma_paddr) /
1858 			sizeof(struct safe_pdesc);
1859 		for (nsegs = re->re_src.nsegs; nsegs; nsegs--) {
1860 			printf(" spd[%u] %p: %p size %u flags %x"
1861 				, ix, &sc->sc_spring[ix]
1862 				, (caddr_t)(uintptr_t) sc->sc_spring[ix].pd_addr
1863 				, sc->sc_spring[ix].pd_size
1864 				, sc->sc_spring[ix].pd_flags
1865 			);
1866 			if (sc->sc_spring[ix].pd_size == 0)
1867 				printf(" (zero!)");
1868 			printf("\n");
1869 			if (++ix == SAFE_TOTAL_SPART)
1870 				ix = 0;
1871 		}
1872 	}
1873 	if (re->re_dst.nsegs > 1) {
1874 		ix = (re->re_desc.d_dst - sc->sc_dpalloc.dma_paddr) /
1875 			sizeof(struct safe_pdesc);
1876 		for (nsegs = re->re_dst.nsegs; nsegs; nsegs--) {
1877 			printf(" dpd[%u] %p: %p flags %x\n"
1878 				, ix, &sc->sc_dpring[ix]
1879 				, (caddr_t)(uintptr_t) sc->sc_dpring[ix].pd_addr
1880 				, sc->sc_dpring[ix].pd_flags
1881 			);
1882 			if (++ix == SAFE_TOTAL_DPART)
1883 				ix = 0;
1884 		}
1885 	}
1886 	printf("sa: cmd0 %08x cmd1 %08x staterec %x\n",
1887 		re->re_sa.sa_cmd0, re->re_sa.sa_cmd1, re->re_sa.sa_staterec);
1888 	printf("sa: key %x %x %x %x %x %x %x %x\n"
1889 		, re->re_sa.sa_key[0]
1890 		, re->re_sa.sa_key[1]
1891 		, re->re_sa.sa_key[2]
1892 		, re->re_sa.sa_key[3]
1893 		, re->re_sa.sa_key[4]
1894 		, re->re_sa.sa_key[5]
1895 		, re->re_sa.sa_key[6]
1896 		, re->re_sa.sa_key[7]
1897 	);
1898 	printf("sa: indigest %x %x %x %x %x\n"
1899 		, re->re_sa.sa_indigest[0]
1900 		, re->re_sa.sa_indigest[1]
1901 		, re->re_sa.sa_indigest[2]
1902 		, re->re_sa.sa_indigest[3]
1903 		, re->re_sa.sa_indigest[4]
1904 	);
1905 	printf("sa: outdigest %x %x %x %x %x\n"
1906 		, re->re_sa.sa_outdigest[0]
1907 		, re->re_sa.sa_outdigest[1]
1908 		, re->re_sa.sa_outdigest[2]
1909 		, re->re_sa.sa_outdigest[3]
1910 		, re->re_sa.sa_outdigest[4]
1911 	);
1912 	printf("sr: iv %x %x %x %x\n"
1913 		, re->re_sastate.sa_saved_iv[0]
1914 		, re->re_sastate.sa_saved_iv[1]
1915 		, re->re_sastate.sa_saved_iv[2]
1916 		, re->re_sastate.sa_saved_iv[3]
1917 	);
1918 	printf("sr: hashbc %u indigest %x %x %x %x %x\n"
1919 		, re->re_sastate.sa_saved_hashbc
1920 		, re->re_sastate.sa_saved_indigest[0]
1921 		, re->re_sastate.sa_saved_indigest[1]
1922 		, re->re_sastate.sa_saved_indigest[2]
1923 		, re->re_sastate.sa_saved_indigest[3]
1924 		, re->re_sastate.sa_saved_indigest[4]
1925 	);
1926 }
1927 
1928 static void
1929 safe_dump_ring(struct safe_softc *sc, const char *tag)
1930 {
1931 	mtx_lock(&sc->sc_ringmtx);
1932 	printf("\nSafeNet Ring State:\n");
1933 	safe_dump_intrstate(sc, tag);
1934 	safe_dump_dmastatus(sc, tag);
1935 	safe_dump_ringstate(sc, tag);
1936 	if (sc->sc_nqchip) {
1937 		struct safe_ringentry *re = sc->sc_back;
1938 		do {
1939 			safe_dump_request(sc, tag, re);
1940 			if (++re == sc->sc_ringtop)
1941 				re = sc->sc_ring;
1942 		} while (re != sc->sc_front);
1943 	}
1944 	mtx_unlock(&sc->sc_ringmtx);
1945 }
1946 
1947 static int
1948 sysctl_hw_safe_dump(SYSCTL_HANDLER_ARGS)
1949 {
1950 	char dmode[64];
1951 	int error;
1952 
1953 	strncpy(dmode, "", sizeof(dmode) - 1);
1954 	dmode[sizeof(dmode) - 1] = '\0';
1955 	error = sysctl_handle_string(oidp, &dmode[0], sizeof(dmode), req);
1956 
1957 	if (error == 0 && req->newptr != NULL) {
1958 		struct safe_softc *sc = safec;
1959 
1960 		if (!sc)
1961 			return EINVAL;
1962 		if (strncmp(dmode, "dma", 3) == 0)
1963 			safe_dump_dmastatus(sc, "safe0");
1964 		else if (strncmp(dmode, "int", 3) == 0)
1965 			safe_dump_intrstate(sc, "safe0");
1966 		else if (strncmp(dmode, "ring", 4) == 0)
1967 			safe_dump_ring(sc, "safe0");
1968 		else
1969 			return EINVAL;
1970 	}
1971 	return error;
1972 }
1973 SYSCTL_PROC(_hw_safe, OID_AUTO, dump,
1974     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, 0,
1975     sysctl_hw_safe_dump, "A",
1976     "Dump driver state");
1977 #endif /* SAFE_DEBUG */
1978