xref: /freebsd/sys/dev/glxsb/glxsb.c (revision 8881d206f4e68b564c2c5f50fc717086fc3e827a)
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 static devclass_t glxsb_devclass;
240 
241 DRIVER_MODULE(glxsb, pci, glxsb_driver, glxsb_devclass, 0, 0);
242 MODULE_VERSION(glxsb, 1);
243 MODULE_DEPEND(glxsb, crypto, 1, 1, 1);
244 
245 static int
246 glxsb_probe(device_t dev)
247 {
248 
249 	if (pci_get_vendor(dev) == PCI_VENDOR_AMD &&
250 	    pci_get_device(dev) == PCI_PRODUCT_AMD_GEODE_LX_CRYPTO) {
251 		device_set_desc(dev,
252 		    "AMD Geode LX Security Block (AES-128-CBC, RNG)");
253 		return (BUS_PROBE_DEFAULT);
254 	}
255 
256 	return (ENXIO);
257 }
258 
259 static int
260 glxsb_attach(device_t dev)
261 {
262 	struct glxsb_softc *sc = device_get_softc(dev);
263 	uint64_t msr;
264 
265 	sc->sc_dev = dev;
266 	msr = rdmsr(SB_GLD_MSR_CAP);
267 
268 	if ((msr & 0xFFFF00) != 0x130400) {
269 		device_printf(dev, "unknown ID 0x%x\n",
270 		    (int)((msr & 0xFFFF00) >> 16));
271 		return (ENXIO);
272 	}
273 
274 	pci_enable_busmaster(dev);
275 
276 	/* Map in the security block configuration/control registers */
277 	sc->sc_rid = PCIR_BAR(0);
278 	sc->sc_sr = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->sc_rid,
279 	    RF_ACTIVE);
280 	if (sc->sc_sr == NULL) {
281 		device_printf(dev, "cannot map register space\n");
282 		return (ENXIO);
283 	}
284 
285 	/*
286 	 * Configure the Security Block.
287 	 *
288 	 * We want to enable the noise generator (T_NE), and enable the
289 	 * linear feedback shift register and whitener post-processing
290 	 * (T_SEL = 3).  Also ensure that test mode (deterministic values)
291 	 * is disabled.
292 	 */
293 	msr = rdmsr(SB_GLD_MSR_CTRL);
294 	msr &= ~(SB_GMC_T_TM | SB_GMC_T_SEL_MASK);
295 	msr |= SB_GMC_T_NE | SB_GMC_T_SEL3;
296 #if 0
297 	msr |= SB_GMC_SBI | SB_GMC_SBY;		/* for AES, if necessary */
298 #endif
299 	wrmsr(SB_GLD_MSR_CTRL, msr);
300 
301 	/* Disable interrupts */
302 	bus_write_4(sc->sc_sr, SB_AES_INT, SB_AI_CLEAR_INTR);
303 
304 	/* Allocate a contiguous DMA-able buffer to work in */
305 	if (glxsb_dma_alloc(sc) != 0)
306 		goto fail0;
307 
308 	/* Initialize our task queue */
309 	sc->sc_tq = taskqueue_create("glxsb_taskq", M_NOWAIT | M_ZERO,
310 	    taskqueue_thread_enqueue, &sc->sc_tq);
311 	if (sc->sc_tq == NULL) {
312 		device_printf(dev, "cannot create task queue\n");
313 		goto fail0;
314 	}
315 	if (taskqueue_start_threads(&sc->sc_tq, 1, PI_NET, "%s taskq",
316 	    device_get_nameunit(dev)) != 0) {
317 		device_printf(dev, "cannot start task queue\n");
318 		goto fail1;
319 	}
320 	TASK_INIT(&sc->sc_cryptotask, 0, glxsb_crypto_task, sc);
321 
322 	/* Initialize crypto */
323 	if (glxsb_crypto_setup(sc) != 0)
324 		goto fail1;
325 
326 	/* Install a periodic collector for the "true" (AMD's word) RNG */
327 	if (hz > 100)
328 		sc->sc_rnghz = hz / 100;
329 	else
330 		sc->sc_rnghz = 1;
331 	callout_init(&sc->sc_rngco, 1);
332 	glxsb_rnd(sc);
333 
334 	return (0);
335 
336 fail1:
337 	taskqueue_free(sc->sc_tq);
338 fail0:
339 	bus_release_resource(dev, SYS_RES_MEMORY, sc->sc_rid, sc->sc_sr);
340 	return (ENXIO);
341 }
342 
343 static int
344 glxsb_detach(device_t dev)
345 {
346 	struct glxsb_softc *sc = device_get_softc(dev);
347 
348 	crypto_unregister_all(sc->sc_cid);
349 
350 	callout_drain(&sc->sc_rngco);
351 	taskqueue_drain(sc->sc_tq, &sc->sc_cryptotask);
352 	bus_generic_detach(dev);
353 	glxsb_dma_free(sc, &sc->sc_dma);
354 	bus_release_resource(dev, SYS_RES_MEMORY, sc->sc_rid, sc->sc_sr);
355 	taskqueue_free(sc->sc_tq);
356 	mtx_destroy(&sc->sc_task_mtx);
357 	return (0);
358 }
359 
360 /*
361  *	callback for bus_dmamap_load()
362  */
363 static void
364 glxsb_dmamap_cb(void *arg, bus_dma_segment_t *seg, int nseg, int error)
365 {
366 
367 	bus_addr_t *paddr = (bus_addr_t*) arg;
368 	*paddr = seg[0].ds_addr;
369 }
370 
371 static int
372 glxsb_dma_alloc(struct glxsb_softc *sc)
373 {
374 	struct glxsb_dma_map *dma = &sc->sc_dma;
375 	int rc;
376 
377 	dma->dma_nsegs = 1;
378 	dma->dma_size = GLXSB_MAX_AES_LEN * 2;
379 
380 	/* Setup DMA descriptor area */
381 	rc = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev),	/* parent */
382 				SB_AES_ALIGN, 0,	/* alignments, bounds */
383 				BUS_SPACE_MAXADDR_32BIT,/* lowaddr */
384 				BUS_SPACE_MAXADDR,	/* highaddr */
385 				NULL, NULL,		/* filter, filterarg */
386 				dma->dma_size,		/* maxsize */
387 				dma->dma_nsegs,		/* nsegments */
388 				dma->dma_size,		/* maxsegsize */
389 				BUS_DMA_ALLOCNOW,	/* flags */
390 				NULL, NULL,		/* lockfunc, lockarg */
391 				&sc->sc_dmat);
392 	if (rc != 0) {
393 		device_printf(sc->sc_dev,
394 		    "cannot allocate DMA tag (%d)\n", rc);
395 		return (rc);
396 	}
397 
398 	rc = bus_dmamem_alloc(sc->sc_dmat, (void **)&dma->dma_vaddr,
399 	    BUS_DMA_NOWAIT, &dma->dma_map);
400 	if (rc != 0) {
401 		device_printf(sc->sc_dev,
402 		    "cannot allocate DMA memory of %d bytes (%d)\n",
403 			dma->dma_size, rc);
404 		goto fail0;
405 	}
406 
407 	rc = bus_dmamap_load(sc->sc_dmat, dma->dma_map, dma->dma_vaddr,
408 	    dma->dma_size, glxsb_dmamap_cb, &dma->dma_paddr, BUS_DMA_NOWAIT);
409 	if (rc != 0) {
410 		device_printf(sc->sc_dev,
411 		    "cannot load DMA memory for %d bytes (%d)\n",
412 		   dma->dma_size, rc);
413 		goto fail1;
414 	}
415 
416 	return (0);
417 
418 fail1:
419 	bus_dmamem_free(sc->sc_dmat, dma->dma_vaddr, dma->dma_map);
420 fail0:
421 	bus_dma_tag_destroy(sc->sc_dmat);
422 	return (rc);
423 }
424 
425 static void
426 glxsb_dma_pre_op(struct glxsb_softc *sc, struct glxsb_dma_map *dma)
427 {
428 
429 	bus_dmamap_sync(sc->sc_dmat, dma->dma_map,
430 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
431 }
432 
433 static void
434 glxsb_dma_post_op(struct glxsb_softc *sc, struct glxsb_dma_map *dma)
435 {
436 
437 	bus_dmamap_sync(sc->sc_dmat, dma->dma_map,
438 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
439 }
440 
441 static void
442 glxsb_dma_free(struct glxsb_softc *sc, struct glxsb_dma_map *dma)
443 {
444 
445 	bus_dmamap_unload(sc->sc_dmat, dma->dma_map);
446 	bus_dmamem_free(sc->sc_dmat, dma->dma_vaddr, dma->dma_map);
447 	bus_dma_tag_destroy(sc->sc_dmat);
448 }
449 
450 static void
451 glxsb_rnd(void *v)
452 {
453 	struct glxsb_softc *sc = v;
454 	uint32_t status, value;
455 
456 	status = bus_read_4(sc->sc_sr, SB_RANDOM_NUM_STATUS);
457 	if (status & SB_RNS_TRNG_VALID) {
458 		value = bus_read_4(sc->sc_sr, SB_RANDOM_NUM);
459 		/* feed with one uint32 */
460 		/* MarkM: FIX!! Check that this does not swamp the harvester! */
461 		random_harvest_queue(&value, sizeof(value), RANDOM_PURE_GLXSB);
462 	}
463 
464 	callout_reset(&sc->sc_rngco, sc->sc_rnghz, glxsb_rnd, sc);
465 }
466 
467 static int
468 glxsb_crypto_setup(struct glxsb_softc *sc)
469 {
470 
471 	sc->sc_cid = crypto_get_driverid(sc->sc_dev,
472 	    sizeof(struct glxsb_session), CRYPTOCAP_F_HARDWARE);
473 
474 	if (sc->sc_cid < 0) {
475 		device_printf(sc->sc_dev, "cannot get crypto driver id\n");
476 		return (ENOMEM);
477 	}
478 
479 	mtx_init(&sc->sc_task_mtx, "glxsb_crypto_mtx", NULL, MTX_DEF);
480 
481 	return (0);
482 }
483 
484 static int
485 glxsb_crypto_probesession(device_t dev, const struct crypto_session_params *csp)
486 {
487 
488 	if (csp->csp_flags != 0)
489 		return (EINVAL);
490 
491 	/*
492 	 * We only support HMAC algorithms to be able to work with
493 	 * ipsec(4), so if we are asked only for authentication without
494 	 * encryption, don't pretend we can accelerate it.
495 	 */
496 	switch (csp->csp_mode) {
497 	case CSP_MODE_ETA:
498 		switch (csp->csp_auth_alg) {
499 		case CRYPTO_NULL_HMAC:
500 		case CRYPTO_SHA1_HMAC:
501 		case CRYPTO_RIPEMD160_HMAC:
502 		case CRYPTO_SHA2_256_HMAC:
503 		case CRYPTO_SHA2_384_HMAC:
504 		case CRYPTO_SHA2_512_HMAC:
505 			break;
506 		default:
507 			return (EINVAL);
508 		}
509 		/* FALLTHROUGH */
510 	case CSP_MODE_CIPHER:
511 		switch (csp->csp_cipher_alg) {
512 		case CRYPTO_AES_CBC:
513 			if (csp->csp_cipher_klen * 8 != 128)
514 				return (EINVAL);
515 			break;
516 		default:
517 			return (EINVAL);
518 		}
519 	default:
520 		return (EINVAL);
521 	}
522 	return (CRYPTODEV_PROBE_HARDWARE);
523 }
524 
525 static int
526 glxsb_crypto_newsession(device_t dev, crypto_session_t cses,
527     const struct crypto_session_params *csp)
528 {
529 	struct glxsb_softc *sc = device_get_softc(dev);
530 	struct glxsb_session *ses;
531 	int error;
532 
533 	ses = crypto_get_driver_session(cses);
534 
535 	/* Copy the key (Geode LX wants the primary key only) */
536 	if (csp->csp_cipher_key != NULL)
537 		bcopy(csp->csp_cipher_key, ses->ses_key, sizeof(ses->ses_key));
538 
539 	if (csp->csp_auth_alg != 0) {
540 		error = glxsb_hash_setup(ses, csp);
541 		if (error != 0) {
542 			glxsb_crypto_freesession(sc->sc_dev, cses);
543 			return (error);
544 		}
545 	}
546 
547 	return (0);
548 }
549 
550 static void
551 glxsb_crypto_freesession(device_t dev, crypto_session_t cses)
552 {
553 	struct glxsb_session *ses;
554 
555 	ses = crypto_get_driver_session(cses);
556 	glxsb_hash_free(ses);
557 }
558 
559 static int
560 glxsb_aes(struct glxsb_softc *sc, uint32_t control, uint32_t psrc,
561     uint32_t pdst, const void *key, int len, const void *iv)
562 {
563 	uint32_t status;
564 	int i;
565 
566 	if (len & 0xF) {
567 		device_printf(sc->sc_dev,
568 		    "len must be a multiple of 16 (not %d)\n", len);
569 		return (EINVAL);
570 	}
571 
572 	/* Set the source */
573 	bus_write_4(sc->sc_sr, SB_SOURCE_A, psrc);
574 
575 	/* Set the destination address */
576 	bus_write_4(sc->sc_sr, SB_DEST_A, pdst);
577 
578 	/* Set the data length */
579 	bus_write_4(sc->sc_sr, SB_LENGTH_A, len);
580 
581 	/* Set the IV */
582 	if (iv != NULL) {
583 		bus_write_region_4(sc->sc_sr, SB_CBC_IV, iv, 4);
584 		control |= SB_CTL_CBC;
585 	}
586 
587 	/* Set the key */
588 	bus_write_region_4(sc->sc_sr, SB_WKEY, key, 4);
589 
590 	/* Ask the security block to do it */
591 	bus_write_4(sc->sc_sr, SB_CTL_A,
592 	    control | SB_CTL_WK | SB_CTL_DC | SB_CTL_SC | SB_CTL_ST);
593 
594 	/*
595 	 * Now wait until it is done.
596 	 *
597 	 * We do a busy wait.  Obviously the number of iterations of
598 	 * the loop required to perform the AES operation depends upon
599 	 * the number of bytes to process.
600 	 *
601 	 * On a 500 MHz Geode LX we see
602 	 *
603 	 *	length (bytes)	typical max iterations
604 	 *	    16		   12
605 	 *	    64		   22
606 	 *	   256		   59
607 	 *	  1024		  212
608 	 *	  8192		1,537
609 	 *
610 	 * Since we have a maximum size of operation defined in
611 	 * GLXSB_MAX_AES_LEN, we use this constant to decide how long
612 	 * to wait.  Allow an order of magnitude longer than it should
613 	 * really take, just in case.
614 	 */
615 
616 	for (i = 0; i < GLXSB_MAX_AES_LEN * 10; i++) {
617 		status = bus_read_4(sc->sc_sr, SB_CTL_A);
618 		if ((status & SB_CTL_ST) == 0)		/* Done */
619 			return (0);
620 	}
621 
622 	device_printf(sc->sc_dev, "operation failed to complete\n");
623 	return (EIO);
624 }
625 
626 static int
627 glxsb_crypto_encdec(struct cryptop *crp, struct glxsb_session *ses,
628     struct glxsb_softc *sc)
629 {
630 	char *op_src, *op_dst;
631 	const void *key;
632 	uint32_t op_psrc, op_pdst;
633 	uint8_t op_iv[SB_AES_BLOCK_SIZE];
634 	int error;
635 	int len, tlen, xlen;
636 	int offset;
637 	uint32_t control;
638 
639 	if ((crp->crp_payload_length % SB_AES_BLOCK_SIZE) != 0)
640 		return (EINVAL);
641 
642 	/* How much of our buffer will we need to use? */
643 	xlen = crp->crp_payload_length > GLXSB_MAX_AES_LEN ?
644 	    GLXSB_MAX_AES_LEN : crp->crp_payload_length;
645 
646 	/*
647 	 * XXX Check if we can have input == output on Geode LX.
648 	 * XXX In the meantime, use two separate (adjacent) buffers.
649 	 */
650 	op_src = sc->sc_dma.dma_vaddr;
651 	op_dst = (char *)sc->sc_dma.dma_vaddr + xlen;
652 
653 	op_psrc = sc->sc_dma.dma_paddr;
654 	op_pdst = sc->sc_dma.dma_paddr + xlen;
655 
656 	if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
657 		control = SB_CTL_ENC;
658 	else
659 		control = SB_CTL_DEC;
660 
661 	crypto_read_iv(crp, op_iv);
662 
663 	offset = 0;
664 	tlen = crp->crp_payload_length;
665 
666 	if (crp->crp_cipher_key != NULL)
667 		key = crp->crp_cipher_key;
668 	else
669 		key = ses->ses_key;
670 
671 	/* Process the data in GLXSB_MAX_AES_LEN chunks */
672 	while (tlen > 0) {
673 		len = (tlen > GLXSB_MAX_AES_LEN) ? GLXSB_MAX_AES_LEN : tlen;
674 		crypto_copydata(crp, crp->crp_payload_start + offset, len,
675 		    op_src);
676 
677 		glxsb_dma_pre_op(sc, &sc->sc_dma);
678 
679 		error = glxsb_aes(sc, control, op_psrc, op_pdst, key, len,
680 		    op_iv);
681 
682 		glxsb_dma_post_op(sc, &sc->sc_dma);
683 		if (error != 0)
684 			return (error);
685 
686 		crypto_copyback(crp, crp->crp_payload_start + offset, len,
687 		    op_dst);
688 
689 		offset += len;
690 		tlen -= len;
691 
692 		/*
693 		 * Copy out last block for use as next iteration IV.
694 		 */
695 		if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
696 			bcopy(op_dst + len - sizeof(op_iv), op_iv,
697 			    sizeof(op_iv));
698 		else
699 			bcopy(op_src + len - sizeof(op_iv), op_iv,
700 			    sizeof(op_iv));
701 	} /* while */
702 
703 	/* All AES processing has now been done. */
704 	bzero(sc->sc_dma.dma_vaddr, xlen * 2);
705 
706 	return (0);
707 }
708 
709 static void
710 glxsb_crypto_task(void *arg, int pending)
711 {
712 	struct glxsb_softc *sc = arg;
713 	const struct crypto_session_params *csp;
714 	struct glxsb_session *ses;
715 	struct cryptop *crp;
716 	int error;
717 
718 	crp = sc->sc_to.to_crp;
719 	ses = sc->sc_to.to_ses;
720 	csp = crypto_get_params(crp->crp_session);
721 
722 	/* Perform data authentication if requested before encryption */
723 	if (csp->csp_mode == CSP_MODE_ETA &&
724 	    !CRYPTO_OP_IS_ENCRYPT(crp->crp_op)) {
725 		error = glxsb_hash_process(ses, csp, crp);
726 		if (error != 0)
727 			goto out;
728 	}
729 
730 	error = glxsb_crypto_encdec(crp, ses, sc);
731 	if (error != 0)
732 		goto out;
733 
734 	/* Perform data authentication if requested after encryption */
735 	if (csp->csp_mode == CSP_MODE_ETA &&
736 	    CRYPTO_OP_IS_ENCRYPT(crp->crp_op)) {
737 		error = glxsb_hash_process(ses, csp, crp);
738 		if (error != 0)
739 			goto out;
740 	}
741 out:
742 	mtx_lock(&sc->sc_task_mtx);
743 	sc->sc_task_count--;
744 	mtx_unlock(&sc->sc_task_mtx);
745 
746 	crp->crp_etype = error;
747 	crypto_unblock(sc->sc_cid, CRYPTO_SYMQ);
748 	crypto_done(crp);
749 }
750 
751 static int
752 glxsb_crypto_process(device_t dev, struct cryptop *crp, int hint)
753 {
754 	struct glxsb_softc *sc = device_get_softc(dev);
755 	struct glxsb_session *ses;
756 
757 	ses = crypto_get_driver_session(crp->crp_session);
758 
759 	mtx_lock(&sc->sc_task_mtx);
760 	if (sc->sc_task_count != 0) {
761 		mtx_unlock(&sc->sc_task_mtx);
762 		return (ERESTART);
763 	}
764 	sc->sc_task_count++;
765 
766 	sc->sc_to.to_crp = crp;
767 	sc->sc_to.to_ses = ses;
768 	mtx_unlock(&sc->sc_task_mtx);
769 
770 	taskqueue_enqueue(sc->sc_tq, &sc->sc_cryptotask);
771 	return(0);
772 }
773