xref: /freebsd/sys/dev/glxsb/glxsb.c (revision 7ef62cebc2f965b0f640263e179276928885e33d)
1 /* $OpenBSD: glxsb.c,v 1.7 2007/02/12 14:31:45 tom Exp $ */
2 
3 /*
4  * Copyright (c) 2006 Tom Cosgrove <tom@openbsd.org>
5  * Copyright (c) 2003, 2004 Theo de Raadt
6  * Copyright (c) 2003 Jason Wright
7  *
8  * Permission to use, copy, modify, and distribute this software for any
9  * purpose with or without fee is hereby granted, provided that the above
10  * copyright notice and this permission notice appear in all copies.
11  *
12  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19  */
20 
21 /*
22  * Driver for the security block on the AMD Geode LX processors
23  * http://www.amd.com/files/connectivitysolutions/geode/geode_lx/33234d_lx_ds.pdf
24  */
25 
26 #include <sys/cdefs.h>
27 __FBSDID("$FreeBSD$");
28 
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/bus.h>
32 #include <sys/errno.h>
33 #include <sys/kernel.h>
34 #include <sys/lock.h>
35 #include <sys/malloc.h>
36 #include <sys/mbuf.h>
37 #include <sys/module.h>
38 #include <sys/mutex.h>
39 #include <sys/proc.h>
40 #include <sys/random.h>
41 #include <sys/rman.h>
42 #include <sys/rwlock.h>
43 #include <sys/sysctl.h>
44 #include <sys/taskqueue.h>
45 
46 #include <machine/bus.h>
47 #include <machine/cpufunc.h>
48 #include <machine/resource.h>
49 
50 #include <dev/pci/pcivar.h>
51 #include <dev/pci/pcireg.h>
52 
53 #include <opencrypto/cryptodev.h>
54 #include <opencrypto/xform.h>
55 
56 #include "cryptodev_if.h"
57 #include "glxsb.h"
58 
59 #define PCI_VENDOR_AMD			0x1022	/* AMD */
60 #define PCI_PRODUCT_AMD_GEODE_LX_CRYPTO	0x2082	/* Geode LX Crypto */
61 
62 #define SB_GLD_MSR_CAP		0x58002000	/* RO - Capabilities */
63 #define SB_GLD_MSR_CONFIG	0x58002001	/* RW - Master Config */
64 #define SB_GLD_MSR_SMI		0x58002002	/* RW - SMI */
65 #define SB_GLD_MSR_ERROR	0x58002003	/* RW - Error */
66 #define SB_GLD_MSR_PM		0x58002004	/* RW - Power Mgmt */
67 #define SB_GLD_MSR_DIAG		0x58002005	/* RW - Diagnostic */
68 #define SB_GLD_MSR_CTRL		0x58002006	/* RW - Security Block Cntrl */
69 
70 						/* For GLD_MSR_CTRL: */
71 #define SB_GMC_DIV0		0x0000		/* AES update divisor values */
72 #define SB_GMC_DIV1		0x0001
73 #define SB_GMC_DIV2		0x0002
74 #define SB_GMC_DIV3		0x0003
75 #define SB_GMC_DIV_MASK		0x0003
76 #define SB_GMC_SBI		0x0004		/* AES swap bits */
77 #define SB_GMC_SBY		0x0008		/* AES swap bytes */
78 #define SB_GMC_TW		0x0010		/* Time write (EEPROM) */
79 #define SB_GMC_T_SEL0		0x0000		/* RNG post-proc: none */
80 #define SB_GMC_T_SEL1		0x0100		/* RNG post-proc: LFSR */
81 #define SB_GMC_T_SEL2		0x0200		/* RNG post-proc: whitener */
82 #define SB_GMC_T_SEL3		0x0300		/* RNG LFSR+whitener */
83 #define SB_GMC_T_SEL_MASK	0x0300
84 #define SB_GMC_T_NE		0x0400		/* Noise (generator) Enable */
85 #define SB_GMC_T_TM		0x0800		/* RNG test mode */
86 						/*     (deterministic) */
87 
88 /* Security Block configuration/control registers (offsets from base) */
89 #define SB_CTL_A		0x0000		/* RW - SB Control A */
90 #define SB_CTL_B		0x0004		/* RW - SB Control B */
91 #define SB_AES_INT		0x0008		/* RW - SB AES Interrupt */
92 #define SB_SOURCE_A		0x0010		/* RW - Source A */
93 #define SB_DEST_A		0x0014		/* RW - Destination A */
94 #define SB_LENGTH_A		0x0018		/* RW - Length A */
95 #define SB_SOURCE_B		0x0020		/* RW - Source B */
96 #define SB_DEST_B		0x0024		/* RW - Destination B */
97 #define SB_LENGTH_B		0x0028		/* RW - Length B */
98 #define SB_WKEY			0x0030		/* WO - Writable Key 0-3 */
99 #define SB_WKEY_0		0x0030		/* WO - Writable Key 0 */
100 #define SB_WKEY_1		0x0034		/* WO - Writable Key 1 */
101 #define SB_WKEY_2		0x0038		/* WO - Writable Key 2 */
102 #define SB_WKEY_3		0x003C		/* WO - Writable Key 3 */
103 #define SB_CBC_IV		0x0040		/* RW - CBC IV 0-3 */
104 #define SB_CBC_IV_0		0x0040		/* RW - CBC IV 0 */
105 #define SB_CBC_IV_1		0x0044		/* RW - CBC IV 1 */
106 #define SB_CBC_IV_2		0x0048		/* RW - CBC IV 2 */
107 #define SB_CBC_IV_3		0x004C		/* RW - CBC IV 3 */
108 #define SB_RANDOM_NUM		0x0050		/* RW - Random Number */
109 #define SB_RANDOM_NUM_STATUS	0x0054		/* RW - Random Number Status */
110 #define SB_EEPROM_COMM		0x0800		/* RW - EEPROM Command */
111 #define SB_EEPROM_ADDR		0x0804		/* RW - EEPROM Address */
112 #define SB_EEPROM_DATA		0x0808		/* RW - EEPROM Data */
113 #define SB_EEPROM_SEC_STATE	0x080C		/* RW - EEPROM Security State */
114 
115 						/* For SB_CTL_A and _B */
116 #define SB_CTL_ST		0x0001		/* Start operation (enc/dec) */
117 #define SB_CTL_ENC		0x0002		/* Encrypt (0 is decrypt) */
118 #define SB_CTL_DEC		0x0000		/* Decrypt */
119 #define SB_CTL_WK		0x0004		/* Use writable key (we set) */
120 #define SB_CTL_DC		0x0008		/* Destination coherent */
121 #define SB_CTL_SC		0x0010		/* Source coherent */
122 #define SB_CTL_CBC		0x0020		/* CBC (0 is ECB) */
123 
124 						/* For SB_AES_INT */
125 #define SB_AI_DISABLE_AES_A	0x0001		/* Disable AES A compl int */
126 #define SB_AI_ENABLE_AES_A	0x0000		/* Enable AES A compl int */
127 #define SB_AI_DISABLE_AES_B	0x0002		/* Disable AES B compl int */
128 #define SB_AI_ENABLE_AES_B	0x0000		/* Enable AES B compl int */
129 #define SB_AI_DISABLE_EEPROM	0x0004		/* Disable EEPROM op comp int */
130 #define SB_AI_ENABLE_EEPROM	0x0000		/* Enable EEPROM op compl int */
131 #define SB_AI_AES_A_COMPLETE   0x10000		/* AES A operation complete */
132 #define SB_AI_AES_B_COMPLETE   0x20000		/* AES B operation complete */
133 #define SB_AI_EEPROM_COMPLETE  0x40000		/* EEPROM operation complete */
134 
135 #define SB_AI_CLEAR_INTR \
136 	(SB_AI_DISABLE_AES_A | SB_AI_DISABLE_AES_B |\
137 	SB_AI_DISABLE_EEPROM | SB_AI_AES_A_COMPLETE |\
138 	SB_AI_AES_B_COMPLETE | SB_AI_EEPROM_COMPLETE)
139 
140 #define SB_RNS_TRNG_VALID	0x0001		/* in SB_RANDOM_NUM_STATUS */
141 
142 #define SB_MEM_SIZE		0x0810		/* Size of memory block */
143 
144 #define SB_AES_ALIGN		0x0010		/* Source and dest buffers */
145 						/* must be 16-byte aligned */
146 #define SB_AES_BLOCK_SIZE	0x0010
147 
148 /*
149  * The Geode LX security block AES acceleration doesn't perform scatter-
150  * gather: it just takes source and destination addresses.  Therefore the
151  * plain- and ciphertexts need to be contiguous.  To this end, we allocate
152  * a buffer for both, and accept the overhead of copying in and out.  If
153  * the number of bytes in one operation is bigger than allowed for by the
154  * buffer (buffer is twice the size of the max length, as it has both input
155  * and output) then we have to perform multiple encryptions/decryptions.
156  */
157 
158 #define GLXSB_MAX_AES_LEN	16384
159 
160 MALLOC_DEFINE(M_GLXSB, "glxsb_data", "Glxsb Data");
161 
162 struct glxsb_dma_map {
163 	bus_dmamap_t		dma_map;	/* DMA map */
164 	bus_dma_segment_t	dma_seg;	/* segments */
165 	int			dma_nsegs;	/* #segments */
166 	int			dma_size;	/* size */
167 	caddr_t			dma_vaddr;	/* virtual address */
168 	bus_addr_t		dma_paddr;	/* physical address */
169 };
170 
171 struct glxsb_taskop {
172 	struct glxsb_session	*to_ses;	/* crypto session */
173 	struct cryptop		*to_crp;	/* cryptop to perfom */
174 };
175 
176 struct glxsb_softc {
177 	device_t		sc_dev;		/* device backpointer */
178 	struct resource		*sc_sr;		/* resource */
179 	int			sc_rid;		/* resource rid */
180 	struct callout		sc_rngco;	/* RNG callout */
181 	int			sc_rnghz;	/* RNG callout ticks */
182 	bus_dma_tag_t		sc_dmat;	/* DMA tag */
183 	struct glxsb_dma_map	sc_dma;		/* DMA map */
184 	int32_t			sc_cid;		/* crypto tag */
185 	struct mtx		sc_task_mtx;	/* task mutex */
186 	struct taskqueue	*sc_tq;		/* task queue */
187 	struct task		sc_cryptotask;	/* task */
188 	struct glxsb_taskop	sc_to;		/* task's crypto operation */
189 	int			sc_task_count;	/* tasks count */
190 };
191 
192 static int glxsb_probe(device_t);
193 static int glxsb_attach(device_t);
194 static int glxsb_detach(device_t);
195 
196 static void glxsb_dmamap_cb(void *, bus_dma_segment_t *, int, int);
197 static int  glxsb_dma_alloc(struct glxsb_softc *);
198 static void glxsb_dma_pre_op(struct glxsb_softc *, struct glxsb_dma_map *);
199 static void glxsb_dma_post_op(struct glxsb_softc *, struct glxsb_dma_map *);
200 static void glxsb_dma_free(struct glxsb_softc *, struct glxsb_dma_map *);
201 
202 static void glxsb_rnd(void *);
203 static int  glxsb_crypto_setup(struct glxsb_softc *);
204 static int  glxsb_crypto_probesession(device_t,
205 	const struct crypto_session_params *);
206 static int  glxsb_crypto_newsession(device_t, crypto_session_t,
207 	const struct crypto_session_params *);
208 static void glxsb_crypto_freesession(device_t, crypto_session_t);
209 static int  glxsb_aes(struct glxsb_softc *, uint32_t, uint32_t,
210 	uint32_t, const void *, int, const void *);
211 
212 static int  glxsb_crypto_encdec(struct cryptop *, struct glxsb_session *,
213 	struct glxsb_softc *);
214 
215 static void glxsb_crypto_task(void *, int);
216 static int  glxsb_crypto_process(device_t, struct cryptop *, int);
217 
218 static device_method_t glxsb_methods[] = {
219 	/* device interface */
220 	DEVMETHOD(device_probe,		glxsb_probe),
221 	DEVMETHOD(device_attach,	glxsb_attach),
222 	DEVMETHOD(device_detach,	glxsb_detach),
223 
224 	/* crypto device methods */
225 	DEVMETHOD(cryptodev_probesession,	glxsb_crypto_probesession),
226 	DEVMETHOD(cryptodev_newsession,		glxsb_crypto_newsession),
227 	DEVMETHOD(cryptodev_freesession,	glxsb_crypto_freesession),
228 	DEVMETHOD(cryptodev_process,		glxsb_crypto_process),
229 
230 	{0,0}
231 };
232 
233 static driver_t glxsb_driver = {
234 	"glxsb",
235 	glxsb_methods,
236 	sizeof(struct glxsb_softc)
237 };
238 
239 DRIVER_MODULE(glxsb, pci, glxsb_driver, 0, 0);
240 MODULE_VERSION(glxsb, 1);
241 MODULE_DEPEND(glxsb, crypto, 1, 1, 1);
242 
243 static int
244 glxsb_probe(device_t dev)
245 {
246 
247 	if (pci_get_vendor(dev) == PCI_VENDOR_AMD &&
248 	    pci_get_device(dev) == PCI_PRODUCT_AMD_GEODE_LX_CRYPTO) {
249 		device_set_desc(dev,
250 		    "AMD Geode LX Security Block (AES-128-CBC, RNG)");
251 		return (BUS_PROBE_DEFAULT);
252 	}
253 
254 	return (ENXIO);
255 }
256 
257 static int
258 glxsb_attach(device_t dev)
259 {
260 	struct glxsb_softc *sc = device_get_softc(dev);
261 	uint64_t msr;
262 
263 	sc->sc_dev = dev;
264 	msr = rdmsr(SB_GLD_MSR_CAP);
265 
266 	if ((msr & 0xFFFF00) != 0x130400) {
267 		device_printf(dev, "unknown ID 0x%x\n",
268 		    (int)((msr & 0xFFFF00) >> 16));
269 		return (ENXIO);
270 	}
271 
272 	pci_enable_busmaster(dev);
273 
274 	/* Map in the security block configuration/control registers */
275 	sc->sc_rid = PCIR_BAR(0);
276 	sc->sc_sr = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->sc_rid,
277 	    RF_ACTIVE);
278 	if (sc->sc_sr == NULL) {
279 		device_printf(dev, "cannot map register space\n");
280 		return (ENXIO);
281 	}
282 
283 	/*
284 	 * Configure the Security Block.
285 	 *
286 	 * We want to enable the noise generator (T_NE), and enable the
287 	 * linear feedback shift register and whitener post-processing
288 	 * (T_SEL = 3).  Also ensure that test mode (deterministic values)
289 	 * is disabled.
290 	 */
291 	msr = rdmsr(SB_GLD_MSR_CTRL);
292 	msr &= ~(SB_GMC_T_TM | SB_GMC_T_SEL_MASK);
293 	msr |= SB_GMC_T_NE | SB_GMC_T_SEL3;
294 #if 0
295 	msr |= SB_GMC_SBI | SB_GMC_SBY;		/* for AES, if necessary */
296 #endif
297 	wrmsr(SB_GLD_MSR_CTRL, msr);
298 
299 	/* Disable interrupts */
300 	bus_write_4(sc->sc_sr, SB_AES_INT, SB_AI_CLEAR_INTR);
301 
302 	/* Allocate a contiguous DMA-able buffer to work in */
303 	if (glxsb_dma_alloc(sc) != 0)
304 		goto fail0;
305 
306 	/* Initialize our task queue */
307 	sc->sc_tq = taskqueue_create("glxsb_taskq", M_NOWAIT | M_ZERO,
308 	    taskqueue_thread_enqueue, &sc->sc_tq);
309 	if (sc->sc_tq == NULL) {
310 		device_printf(dev, "cannot create task queue\n");
311 		goto fail0;
312 	}
313 	if (taskqueue_start_threads(&sc->sc_tq, 1, PI_NET, "%s taskq",
314 	    device_get_nameunit(dev)) != 0) {
315 		device_printf(dev, "cannot start task queue\n");
316 		goto fail1;
317 	}
318 	TASK_INIT(&sc->sc_cryptotask, 0, glxsb_crypto_task, sc);
319 
320 	/* Initialize crypto */
321 	if (glxsb_crypto_setup(sc) != 0)
322 		goto fail1;
323 
324 	/* Install a periodic collector for the "true" (AMD's word) RNG */
325 	if (hz > 100)
326 		sc->sc_rnghz = hz / 100;
327 	else
328 		sc->sc_rnghz = 1;
329 	callout_init(&sc->sc_rngco, 1);
330 	glxsb_rnd(sc);
331 
332 	return (0);
333 
334 fail1:
335 	taskqueue_free(sc->sc_tq);
336 fail0:
337 	bus_release_resource(dev, SYS_RES_MEMORY, sc->sc_rid, sc->sc_sr);
338 	return (ENXIO);
339 }
340 
341 static int
342 glxsb_detach(device_t dev)
343 {
344 	struct glxsb_softc *sc = device_get_softc(dev);
345 
346 	crypto_unregister_all(sc->sc_cid);
347 
348 	callout_drain(&sc->sc_rngco);
349 	taskqueue_drain(sc->sc_tq, &sc->sc_cryptotask);
350 	bus_generic_detach(dev);
351 	glxsb_dma_free(sc, &sc->sc_dma);
352 	bus_release_resource(dev, SYS_RES_MEMORY, sc->sc_rid, sc->sc_sr);
353 	taskqueue_free(sc->sc_tq);
354 	mtx_destroy(&sc->sc_task_mtx);
355 	return (0);
356 }
357 
358 /*
359  *	callback for bus_dmamap_load()
360  */
361 static void
362 glxsb_dmamap_cb(void *arg, bus_dma_segment_t *seg, int nseg, int error)
363 {
364 
365 	bus_addr_t *paddr = (bus_addr_t*) arg;
366 	*paddr = seg[0].ds_addr;
367 }
368 
369 static int
370 glxsb_dma_alloc(struct glxsb_softc *sc)
371 {
372 	struct glxsb_dma_map *dma = &sc->sc_dma;
373 	int rc;
374 
375 	dma->dma_nsegs = 1;
376 	dma->dma_size = GLXSB_MAX_AES_LEN * 2;
377 
378 	/* Setup DMA descriptor area */
379 	rc = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev),	/* parent */
380 				SB_AES_ALIGN, 0,	/* alignments, bounds */
381 				BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
382 				BUS_SPACE_MAXADDR,	/* highaddr */
383 				NULL, NULL,		/* filter, filterarg */
384 				dma->dma_size,		/* maxsize */
385 				dma->dma_nsegs,		/* nsegments */
386 				dma->dma_size,		/* maxsegsize */
387 				BUS_DMA_ALLOCNOW,	/* flags */
388 				NULL, NULL,		/* lockfunc, lockarg */
389 				&sc->sc_dmat);
390 	if (rc != 0) {
391 		device_printf(sc->sc_dev,
392 		    "cannot allocate DMA tag (%d)\n", rc);
393 		return (rc);
394 	}
395 
396 	rc = bus_dmamem_alloc(sc->sc_dmat, (void **)&dma->dma_vaddr,
397 	    BUS_DMA_NOWAIT, &dma->dma_map);
398 	if (rc != 0) {
399 		device_printf(sc->sc_dev,
400 		    "cannot allocate DMA memory of %d bytes (%d)\n",
401 			dma->dma_size, rc);
402 		goto fail0;
403 	}
404 
405 	rc = bus_dmamap_load(sc->sc_dmat, dma->dma_map, dma->dma_vaddr,
406 	    dma->dma_size, glxsb_dmamap_cb, &dma->dma_paddr, BUS_DMA_NOWAIT);
407 	if (rc != 0) {
408 		device_printf(sc->sc_dev,
409 		    "cannot load DMA memory for %d bytes (%d)\n",
410 		   dma->dma_size, rc);
411 		goto fail1;
412 	}
413 
414 	return (0);
415 
416 fail1:
417 	bus_dmamem_free(sc->sc_dmat, dma->dma_vaddr, dma->dma_map);
418 fail0:
419 	bus_dma_tag_destroy(sc->sc_dmat);
420 	return (rc);
421 }
422 
423 static void
424 glxsb_dma_pre_op(struct glxsb_softc *sc, struct glxsb_dma_map *dma)
425 {
426 
427 	bus_dmamap_sync(sc->sc_dmat, dma->dma_map,
428 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
429 }
430 
431 static void
432 glxsb_dma_post_op(struct glxsb_softc *sc, struct glxsb_dma_map *dma)
433 {
434 
435 	bus_dmamap_sync(sc->sc_dmat, dma->dma_map,
436 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
437 }
438 
439 static void
440 glxsb_dma_free(struct glxsb_softc *sc, struct glxsb_dma_map *dma)
441 {
442 
443 	bus_dmamap_unload(sc->sc_dmat, dma->dma_map);
444 	bus_dmamem_free(sc->sc_dmat, dma->dma_vaddr, dma->dma_map);
445 	bus_dma_tag_destroy(sc->sc_dmat);
446 }
447 
448 static void
449 glxsb_rnd(void *v)
450 {
451 	struct glxsb_softc *sc = v;
452 	uint32_t status, value;
453 
454 	status = bus_read_4(sc->sc_sr, SB_RANDOM_NUM_STATUS);
455 	if (status & SB_RNS_TRNG_VALID) {
456 		value = bus_read_4(sc->sc_sr, SB_RANDOM_NUM);
457 		/* feed with one uint32 */
458 		/* MarkM: FIX!! Check that this does not swamp the harvester! */
459 		random_harvest_queue(&value, sizeof(value), RANDOM_PURE_GLXSB);
460 	}
461 
462 	callout_reset(&sc->sc_rngco, sc->sc_rnghz, glxsb_rnd, sc);
463 }
464 
465 static int
466 glxsb_crypto_setup(struct glxsb_softc *sc)
467 {
468 
469 	sc->sc_cid = crypto_get_driverid(sc->sc_dev,
470 	    sizeof(struct glxsb_session), CRYPTOCAP_F_HARDWARE);
471 
472 	if (sc->sc_cid < 0) {
473 		device_printf(sc->sc_dev, "cannot get crypto driver id\n");
474 		return (ENOMEM);
475 	}
476 
477 	mtx_init(&sc->sc_task_mtx, "glxsb_crypto_mtx", NULL, MTX_DEF);
478 
479 	return (0);
480 }
481 
482 static int
483 glxsb_crypto_probesession(device_t dev, const struct crypto_session_params *csp)
484 {
485 
486 	if (csp->csp_flags != 0)
487 		return (EINVAL);
488 
489 	/*
490 	 * We only support HMAC algorithms to be able to work with
491 	 * ipsec(4), so if we are asked only for authentication without
492 	 * encryption, don't pretend we can accelerate it.
493 	 */
494 	switch (csp->csp_mode) {
495 	case CSP_MODE_ETA:
496 		switch (csp->csp_auth_alg) {
497 		case CRYPTO_NULL_HMAC:
498 		case CRYPTO_SHA1_HMAC:
499 		case CRYPTO_RIPEMD160_HMAC:
500 		case CRYPTO_SHA2_256_HMAC:
501 		case CRYPTO_SHA2_384_HMAC:
502 		case CRYPTO_SHA2_512_HMAC:
503 			break;
504 		default:
505 			return (EINVAL);
506 		}
507 		/* FALLTHROUGH */
508 	case CSP_MODE_CIPHER:
509 		switch (csp->csp_cipher_alg) {
510 		case CRYPTO_AES_CBC:
511 			if (csp->csp_cipher_klen * 8 != 128)
512 				return (EINVAL);
513 			break;
514 		default:
515 			return (EINVAL);
516 		}
517 	default:
518 		return (EINVAL);
519 	}
520 	return (CRYPTODEV_PROBE_HARDWARE);
521 }
522 
523 static int
524 glxsb_crypto_newsession(device_t dev, crypto_session_t cses,
525     const struct crypto_session_params *csp)
526 {
527 	struct glxsb_softc *sc = device_get_softc(dev);
528 	struct glxsb_session *ses;
529 	int error;
530 
531 	ses = crypto_get_driver_session(cses);
532 
533 	/* Copy the key (Geode LX wants the primary key only) */
534 	if (csp->csp_cipher_key != NULL)
535 		bcopy(csp->csp_cipher_key, ses->ses_key, sizeof(ses->ses_key));
536 
537 	if (csp->csp_auth_alg != 0) {
538 		error = glxsb_hash_setup(ses, csp);
539 		if (error != 0) {
540 			glxsb_crypto_freesession(sc->sc_dev, cses);
541 			return (error);
542 		}
543 	}
544 
545 	return (0);
546 }
547 
548 static void
549 glxsb_crypto_freesession(device_t dev, crypto_session_t cses)
550 {
551 	struct glxsb_session *ses;
552 
553 	ses = crypto_get_driver_session(cses);
554 	glxsb_hash_free(ses);
555 }
556 
557 static int
558 glxsb_aes(struct glxsb_softc *sc, uint32_t control, uint32_t psrc,
559     uint32_t pdst, const void *key, int len, const void *iv)
560 {
561 	uint32_t status;
562 	int i;
563 
564 	if (len & 0xF) {
565 		device_printf(sc->sc_dev,
566 		    "len must be a multiple of 16 (not %d)\n", len);
567 		return (EINVAL);
568 	}
569 
570 	/* Set the source */
571 	bus_write_4(sc->sc_sr, SB_SOURCE_A, psrc);
572 
573 	/* Set the destination address */
574 	bus_write_4(sc->sc_sr, SB_DEST_A, pdst);
575 
576 	/* Set the data length */
577 	bus_write_4(sc->sc_sr, SB_LENGTH_A, len);
578 
579 	/* Set the IV */
580 	if (iv != NULL) {
581 		bus_write_region_4(sc->sc_sr, SB_CBC_IV, iv, 4);
582 		control |= SB_CTL_CBC;
583 	}
584 
585 	/* Set the key */
586 	bus_write_region_4(sc->sc_sr, SB_WKEY, key, 4);
587 
588 	/* Ask the security block to do it */
589 	bus_write_4(sc->sc_sr, SB_CTL_A,
590 	    control | SB_CTL_WK | SB_CTL_DC | SB_CTL_SC | SB_CTL_ST);
591 
592 	/*
593 	 * Now wait until it is done.
594 	 *
595 	 * We do a busy wait.  Obviously the number of iterations of
596 	 * the loop required to perform the AES operation depends upon
597 	 * the number of bytes to process.
598 	 *
599 	 * On a 500 MHz Geode LX we see
600 	 *
601 	 *	length (bytes)	typical max iterations
602 	 *	    16		   12
603 	 *	    64		   22
604 	 *	   256		   59
605 	 *	  1024		  212
606 	 *	  8192		1,537
607 	 *
608 	 * Since we have a maximum size of operation defined in
609 	 * GLXSB_MAX_AES_LEN, we use this constant to decide how long
610 	 * to wait.  Allow an order of magnitude longer than it should
611 	 * really take, just in case.
612 	 */
613 
614 	for (i = 0; i < GLXSB_MAX_AES_LEN * 10; i++) {
615 		status = bus_read_4(sc->sc_sr, SB_CTL_A);
616 		if ((status & SB_CTL_ST) == 0)		/* Done */
617 			return (0);
618 	}
619 
620 	device_printf(sc->sc_dev, "operation failed to complete\n");
621 	return (EIO);
622 }
623 
624 static int
625 glxsb_crypto_encdec(struct cryptop *crp, struct glxsb_session *ses,
626     struct glxsb_softc *sc)
627 {
628 	char *op_src, *op_dst;
629 	const void *key;
630 	uint32_t op_psrc, op_pdst;
631 	uint8_t op_iv[SB_AES_BLOCK_SIZE];
632 	int error;
633 	int len, tlen, xlen;
634 	int offset;
635 	uint32_t control;
636 
637 	if ((crp->crp_payload_length % SB_AES_BLOCK_SIZE) != 0)
638 		return (EINVAL);
639 
640 	/* How much of our buffer will we need to use? */
641 	xlen = crp->crp_payload_length > GLXSB_MAX_AES_LEN ?
642 	    GLXSB_MAX_AES_LEN : crp->crp_payload_length;
643 
644 	/*
645 	 * XXX Check if we can have input == output on Geode LX.
646 	 * XXX In the meantime, use two separate (adjacent) buffers.
647 	 */
648 	op_src = sc->sc_dma.dma_vaddr;
649 	op_dst = (char *)sc->sc_dma.dma_vaddr + xlen;
650 
651 	op_psrc = sc->sc_dma.dma_paddr;
652 	op_pdst = sc->sc_dma.dma_paddr + xlen;
653 
654 	if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
655 		control = SB_CTL_ENC;
656 	else
657 		control = SB_CTL_DEC;
658 
659 	crypto_read_iv(crp, op_iv);
660 
661 	offset = 0;
662 	tlen = crp->crp_payload_length;
663 
664 	if (crp->crp_cipher_key != NULL)
665 		key = crp->crp_cipher_key;
666 	else
667 		key = ses->ses_key;
668 
669 	/* Process the data in GLXSB_MAX_AES_LEN chunks */
670 	while (tlen > 0) {
671 		len = (tlen > GLXSB_MAX_AES_LEN) ? GLXSB_MAX_AES_LEN : tlen;
672 		crypto_copydata(crp, crp->crp_payload_start + offset, len,
673 		    op_src);
674 
675 		glxsb_dma_pre_op(sc, &sc->sc_dma);
676 
677 		error = glxsb_aes(sc, control, op_psrc, op_pdst, key, len,
678 		    op_iv);
679 
680 		glxsb_dma_post_op(sc, &sc->sc_dma);
681 		if (error != 0)
682 			return (error);
683 
684 		crypto_copyback(crp, crp->crp_payload_start + offset, len,
685 		    op_dst);
686 
687 		offset += len;
688 		tlen -= len;
689 
690 		/*
691 		 * Copy out last block for use as next iteration IV.
692 		 */
693 		if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
694 			bcopy(op_dst + len - sizeof(op_iv), op_iv,
695 			    sizeof(op_iv));
696 		else
697 			bcopy(op_src + len - sizeof(op_iv), op_iv,
698 			    sizeof(op_iv));
699 	} /* while */
700 
701 	/* All AES processing has now been done. */
702 	bzero(sc->sc_dma.dma_vaddr, xlen * 2);
703 
704 	return (0);
705 }
706 
707 static void
708 glxsb_crypto_task(void *arg, int pending)
709 {
710 	struct glxsb_softc *sc = arg;
711 	const struct crypto_session_params *csp;
712 	struct glxsb_session *ses;
713 	struct cryptop *crp;
714 	int error;
715 
716 	crp = sc->sc_to.to_crp;
717 	ses = sc->sc_to.to_ses;
718 	csp = crypto_get_params(crp->crp_session);
719 
720 	/* Perform data authentication if requested before encryption */
721 	if (csp->csp_mode == CSP_MODE_ETA &&
722 	    !CRYPTO_OP_IS_ENCRYPT(crp->crp_op)) {
723 		error = glxsb_hash_process(ses, csp, crp);
724 		if (error != 0)
725 			goto out;
726 	}
727 
728 	error = glxsb_crypto_encdec(crp, ses, sc);
729 	if (error != 0)
730 		goto out;
731 
732 	/* Perform data authentication if requested after encryption */
733 	if (csp->csp_mode == CSP_MODE_ETA &&
734 	    CRYPTO_OP_IS_ENCRYPT(crp->crp_op)) {
735 		error = glxsb_hash_process(ses, csp, crp);
736 		if (error != 0)
737 			goto out;
738 	}
739 out:
740 	mtx_lock(&sc->sc_task_mtx);
741 	sc->sc_task_count--;
742 	mtx_unlock(&sc->sc_task_mtx);
743 
744 	crp->crp_etype = error;
745 	crypto_unblock(sc->sc_cid, CRYPTO_SYMQ);
746 	crypto_done(crp);
747 }
748 
749 static int
750 glxsb_crypto_process(device_t dev, struct cryptop *crp, int hint)
751 {
752 	struct glxsb_softc *sc = device_get_softc(dev);
753 	struct glxsb_session *ses;
754 
755 	ses = crypto_get_driver_session(crp->crp_session);
756 
757 	mtx_lock(&sc->sc_task_mtx);
758 	if (sc->sc_task_count != 0) {
759 		mtx_unlock(&sc->sc_task_mtx);
760 		return (ERESTART);
761 	}
762 	sc->sc_task_count++;
763 
764 	sc->sc_to.to_crp = crp;
765 	sc->sc_to.to_ses = ses;
766 	mtx_unlock(&sc->sc_task_mtx);
767 
768 	taskqueue_enqueue(sc->sc_tq, &sc->sc_cryptotask);
769 	return(0);
770 }
771