xref: /netbsd/sys/arch/i386/pci/glxsb.c (revision fdf161e4)
1 /*	$NetBSD: glxsb.c,v 1.19 2022/05/22 11:39:26 riastradh Exp $	*/
2 /* $OpenBSD: glxsb.c,v 1.7 2007/02/12 14:31:45 tom Exp $ */
3 
4 /*
5  * Copyright (c) 2006 Tom Cosgrove <tom@openbsd.org>
6  * Copyright (c) 2003, 2004 Theo de Raadt
7  * Copyright (c) 2003 Jason Wright
8  *
9  * Permission to use, copy, modify, and distribute this software for any
10  * purpose with or without fee is hereby granted, provided that the above
11  * copyright notice and this permission notice appear in all copies.
12  *
13  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
14  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
15  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
16  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
17  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
18  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
19  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
20  */
21 
22 /*
23  * Driver for the security block on the AMD Geode LX processors
24  * http://www.amd.com/files/connectivitysolutions/geode/geode_lx/33234d_lx_ds.pdf
25  */
26 
27 #include <sys/cdefs.h>
28 __KERNEL_RCSID(0, "$NetBSD: glxsb.c,v 1.19 2022/05/22 11:39:26 riastradh Exp $");
29 
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/device.h>
33 #include <sys/malloc.h>
34 #include <sys/mbuf.h>
35 #include <sys/types.h>
36 #include <sys/callout.h>
37 #include <sys/bus.h>
38 #include <sys/cprng.h>
39 #include <sys/rndsource.h>
40 
41 #include <machine/cpufunc.h>
42 
43 #include <dev/pci/pcivar.h>
44 #include <dev/pci/pcidevs.h>
45 
46 #include <opencrypto/cryptodev.h>
47 
48 #define SB_GLD_MSR_CAP		0x58002000	/* RO - Capabilities */
49 #define SB_GLD_MSR_CONFIG	0x58002001	/* RW - Master Config */
50 #define SB_GLD_MSR_SMI		0x58002002	/* RW - SMI */
51 #define SB_GLD_MSR_ERROR	0x58002003	/* RW - Error */
52 #define SB_GLD_MSR_PM		0x58002004	/* RW - Power Mgmt */
53 #define SB_GLD_MSR_DIAG		0x58002005	/* RW - Diagnostic */
54 #define SB_GLD_MSR_CTRL		0x58002006	/* RW - Security Block Cntrl */
55 
56 						/* For GLD_MSR_CTRL: */
57 #define SB_GMC_DIV0		0x0000		/* AES update divisor values */
58 #define SB_GMC_DIV1		0x0001
59 #define SB_GMC_DIV2		0x0002
60 #define SB_GMC_DIV3		0x0003
61 #define SB_GMC_DIV_MASK		0x0003
62 #define SB_GMC_SBI		0x0004		/* AES swap bits */
63 #define SB_GMC_SBY		0x0008		/* AES swap bytes */
64 #define SB_GMC_TW		0x0010		/* Time write (EEPROM) */
65 #define SB_GMC_T_SEL0		0x0000		/* RNG post-proc: none */
66 #define SB_GMC_T_SEL1		0x0100		/* RNG post-proc: LFSR */
67 #define SB_GMC_T_SEL2		0x0200		/* RNG post-proc: whitener */
68 #define SB_GMC_T_SEL3		0x0300		/* RNG LFSR+whitener */
69 #define SB_GMC_T_SEL_MASK	0x0300
70 #define SB_GMC_T_NE		0x0400		/* Noise (generator) Enable */
71 #define SB_GMC_T_TM		0x0800		/* RNG test mode */
72 						/*     (deterministic) */
73 
74 /* Security Block configuration/control registers (offsets from base) */
75 
76 #define SB_CTL_A		0x0000		/* RW - SB Control A */
77 #define SB_CTL_B		0x0004		/* RW - SB Control B */
78 #define SB_AES_INT		0x0008		/* RW - SB AES Interrupt */
79 #define SB_SOURCE_A		0x0010		/* RW - Source A */
80 #define SB_DEST_A		0x0014		/* RW - Destination A */
81 #define SB_LENGTH_A		0x0018		/* RW - Length A */
82 #define SB_SOURCE_B		0x0020		/* RW - Source B */
83 #define SB_DEST_B		0x0024		/* RW - Destination B */
84 #define SB_LENGTH_B		0x0028		/* RW - Length B */
85 #define SB_WKEY			0x0030		/* WO - Writable Key 0-3 */
86 #define SB_WKEY_0		0x0030		/* WO - Writable Key 0 */
87 #define SB_WKEY_1		0x0034		/* WO - Writable Key 1 */
88 #define SB_WKEY_2		0x0038		/* WO - Writable Key 2 */
89 #define SB_WKEY_3		0x003C		/* WO - Writable Key 3 */
90 #define SB_CBC_IV		0x0040		/* RW - CBC IV 0-3 */
91 #define SB_CBC_IV_0		0x0040		/* RW - CBC IV 0 */
92 #define SB_CBC_IV_1		0x0044		/* RW - CBC IV 1 */
93 #define SB_CBC_IV_2		0x0048		/* RW - CBC IV 2 */
94 #define SB_CBC_IV_3		0x004C		/* RW - CBC IV 3 */
95 #define SB_RANDOM_NUM		0x0050		/* RW - Random Number */
96 #define SB_RANDOM_NUM_STATUS	0x0054		/* RW - Random Number Status */
97 #define SB_EEPROM_COMM		0x0800		/* RW - EEPROM Command */
98 #define SB_EEPROM_ADDR		0x0804		/* RW - EEPROM Address */
99 #define SB_EEPROM_DATA		0x0808		/* RW - EEPROM Data */
100 #define SB_EEPROM_SEC_STATE	0x080C		/* RW - EEPROM Security State */
101 
102 						/* For SB_CTL_A and _B */
103 #define SB_CTL_ST		0x0001		/* Start operation (enc/dec) */
104 #define SB_CTL_ENC		0x0002		/* Encrypt (0 is decrypt) */
105 #define SB_CTL_DEC		0x0000		/* Decrypt */
106 #define SB_CTL_WK		0x0004		/* Use writable key (we set) */
107 #define SB_CTL_DC		0x0008		/* Destination coherent */
108 #define SB_CTL_SC		0x0010		/* Source coherent */
109 #define SB_CTL_CBC		0x0020		/* CBC (0 is ECB) */
110 
111 						/* For SB_AES_INT */
112 #define SB_AI_DISABLE_AES_A	0x0001		/* Disable AES A compl int */
113 #define SB_AI_ENABLE_AES_A	0x0000		/* Enable AES A compl int */
114 #define SB_AI_DISABLE_AES_B	0x0002		/* Disable AES B compl int */
115 #define SB_AI_ENABLE_AES_B	0x0000		/* Enable AES B compl int */
116 #define SB_AI_DISABLE_EEPROM	0x0004		/* Disable EEPROM op comp int */
117 #define SB_AI_ENABLE_EEPROM	0x0000		/* Enable EEPROM op compl int */
118 #define SB_AI_AES_A_COMPLETE	0x0100		/* AES A operation complete */
119 #define SB_AI_AES_B_COMPLETE	0x0200		/* AES B operation complete */
120 #define SB_AI_EEPROM_COMPLETE	0x0400		/* EEPROM operation complete */
121 
122 #define SB_RNS_TRNG_VALID	0x0001		/* in SB_RANDOM_NUM_STATUS */
123 
124 #define SB_MEM_SIZE		0x0810		/* Size of memory block */
125 
126 #define SB_AES_ALIGN		0x0010		/* Source and dest buffers */
127 						/* must be 16-byte aligned */
128 #define SB_AES_BLOCK_SIZE	0x0010
129 
130 /*
131  * The Geode LX security block AES acceleration doesn't perform scatter-
132  * gather: it just takes source and destination addresses.  Therefore the
133  * plain- and ciphertexts need to be contiguous.  To this end, we allocate
134  * a buffer for both, and accept the overhead of copying in and out.  If
135  * the number of bytes in one operation is bigger than allowed for by the
136  * buffer (buffer is twice the size of the max length, as it has both input
137  * and output) then we have to perform multiple encryptions/decryptions.
138  */
139 #define GLXSB_MAX_AES_LEN	16384
140 
141 struct glxsb_dma_map {
142 	bus_dmamap_t		dma_map;
143 	bus_dma_segment_t	dma_seg;
144 	int			dma_nsegs;
145 	int			dma_size;
146 	void *			dma_vaddr;
147 	uint32_t		dma_paddr;
148 };
149 struct glxsb_session {
150 	uint32_t	ses_key[4];
151 	int		ses_klen;
152 	int		ses_used;
153 };
154 
155 struct glxsb_softc {
156 	device_t		sc_dev;
157 	bus_space_tag_t		sc_iot;
158 	bus_space_handle_t	sc_ioh;
159 	struct callout		sc_co;
160 
161 	bus_dma_tag_t		sc_dmat;
162 	struct glxsb_dma_map	sc_dma;
163 	int32_t			sc_cid;
164 	int			sc_nsessions;
165 	struct glxsb_session	*sc_sessions;
166 
167 	krndsource_t	sc_rnd_source;
168 };
169 
170 int	glxsb_match(device_t, cfdata_t, void *);
171 void	glxsb_attach(device_t, device_t, void *);
172 void	glxsb_rnd(void *);
173 
174 CFATTACH_DECL_NEW(glxsb, sizeof(struct glxsb_softc),
175     glxsb_match, glxsb_attach, NULL, NULL);
176 
177 #define GLXSB_SESSION(sid)		((sid) & 0x0fffffff)
178 #define	GLXSB_SID(crd,ses)		(((crd) << 28) | ((ses) & 0x0fffffff))
179 
180 int glxsb_crypto_setup(struct glxsb_softc *);
181 int glxsb_crypto_newsession(void *, uint32_t *, struct cryptoini *);
182 int glxsb_crypto_process(void *, struct cryptop *, int);
183 void glxsb_crypto_freesession(void *, uint64_t);
184 static __inline void glxsb_aes(struct glxsb_softc *, uint32_t, uint32_t,
185     uint32_t, void *, int, void *);
186 
187 int glxsb_dma_alloc(struct glxsb_softc *, int, struct glxsb_dma_map *);
188 void glxsb_dma_pre_op(struct glxsb_softc *, struct glxsb_dma_map *);
189 void glxsb_dma_post_op(struct glxsb_softc *, struct glxsb_dma_map *);
190 void glxsb_dma_free(struct glxsb_softc *, struct glxsb_dma_map *);
191 
192 int
glxsb_match(device_t parent,cfdata_t match,void * aux)193 glxsb_match(device_t parent, cfdata_t match, void *aux)
194 {
195 	struct pci_attach_args *pa = aux;
196 
197 	if (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_AMD &&
198 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_AMD_GEODELX_AES)
199 		return (1);
200 
201 	return (0);
202 }
203 
204 void
glxsb_attach(device_t parent,device_t self,void * aux)205 glxsb_attach(device_t parent, device_t self, void *aux)
206 {
207 	struct glxsb_softc *sc = device_private(self);
208 	struct pci_attach_args *pa = aux;
209 	bus_addr_t membase;
210 	bus_size_t memsize;
211 	uint64_t msr;
212 	uint32_t intr;
213 
214 	msr = rdmsr(SB_GLD_MSR_CAP);
215 	if ((msr & 0xFFFF00) != 0x130400) {
216 		aprint_error(": unknown ID 0x%x\n",
217 		    (int)((msr & 0xFFFF00) >> 16));
218 		return;
219 	}
220 
221 	/* printf(": revision %d", (int) (msr & 0xFF)); */
222 
223 	/* Map in the security block configuration/control registers */
224 	if (pci_mapreg_map(pa, PCI_MAPREG_START,
225 	    PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT, 0,
226 	    &sc->sc_iot, &sc->sc_ioh, &membase, &memsize)) {
227 		aprint_error(": can't find mem space\n");
228 		return;
229 	}
230 
231 	sc->sc_dev = self;
232 
233 	/*
234 	 * Configure the Security Block.
235 	 *
236 	 * We want to enable the noise generator (T_NE), and enable the
237 	 * linear feedback shift register and whitener post-processing
238 	 * (T_SEL = 3).  Also ensure that test mode (deterministic values)
239 	 * is disabled.
240 	 */
241 	msr = rdmsr(SB_GLD_MSR_CTRL);
242 	msr &= ~(SB_GMC_T_TM | SB_GMC_T_SEL_MASK);
243 	msr |= SB_GMC_T_NE | SB_GMC_T_SEL3;
244 #if 0
245 	msr |= SB_GMC_SBI | SB_GMC_SBY;		/* for AES, if necessary */
246 #endif
247 	wrmsr(SB_GLD_MSR_CTRL, msr);
248 
249 	rnd_attach_source(&sc->sc_rnd_source, device_xname(self),
250 			  RND_TYPE_RNG, RND_FLAG_COLLECT_VALUE);
251 
252 	/* Install a periodic collector for the "true" (AMD's word) RNG */
253 	callout_init(&sc->sc_co, 0);
254 	callout_setfunc(&sc->sc_co, glxsb_rnd, sc);
255 	glxsb_rnd(sc);
256 	aprint_normal(": RNG");
257 
258 	/* We don't have an interrupt handler, so disable completion INTs */
259 	intr = SB_AI_DISABLE_AES_A | SB_AI_DISABLE_AES_B |
260 	    SB_AI_DISABLE_EEPROM | SB_AI_AES_A_COMPLETE |
261 	    SB_AI_AES_B_COMPLETE | SB_AI_EEPROM_COMPLETE;
262 	bus_space_write_4(sc->sc_iot, sc->sc_ioh, SB_AES_INT, intr);
263 
264 	sc->sc_dmat = pa->pa_dmat;
265 
266 	if (glxsb_crypto_setup(sc))
267 		aprint_normal(" AES");
268 
269 	aprint_normal("\n");
270 }
271 
272 void
glxsb_rnd(void * v)273 glxsb_rnd(void *v)
274 {
275 	struct glxsb_softc *sc = v;
276 	uint32_t status, value;
277 	extern int hz;
278 
279 	status = bus_space_read_4(sc->sc_iot, sc->sc_ioh, SB_RANDOM_NUM_STATUS);
280 	if (status & SB_RNS_TRNG_VALID) {
281 		value = bus_space_read_4(sc->sc_iot, sc->sc_ioh, SB_RANDOM_NUM);
282 		rnd_add_data(&sc->sc_rnd_source, &value, sizeof(value),
283 			     sizeof(value) * NBBY);
284 	}
285 
286 	callout_schedule(&sc->sc_co, (hz > 100) ? (hz / 100) : 1);
287 }
288 
289 int
glxsb_crypto_setup(struct glxsb_softc * sc)290 glxsb_crypto_setup(struct glxsb_softc *sc)
291 {
292 
293 	/* Allocate a contiguous DMA-able buffer to work in */
294 	if (glxsb_dma_alloc(sc, GLXSB_MAX_AES_LEN * 2, &sc->sc_dma) != 0)
295 		return 0;
296 
297 	sc->sc_cid = crypto_get_driverid(0);
298 	if (sc->sc_cid < 0)
299 		return 0;
300 
301 	crypto_register(sc->sc_cid, CRYPTO_AES_CBC, 0, 0,
302 	    glxsb_crypto_newsession, glxsb_crypto_freesession,
303 	    glxsb_crypto_process, sc);
304 
305 	sc->sc_nsessions = 0;
306 
307 	return 1;
308 }
309 
310 int
glxsb_crypto_newsession(void * aux,uint32_t * sidp,struct cryptoini * cri)311 glxsb_crypto_newsession(void *aux, uint32_t *sidp, struct cryptoini *cri)
312 {
313 	struct glxsb_softc *sc = aux;
314 	struct glxsb_session *ses = NULL;
315 	int sesn;
316 
317 	if (cri->cri_next != NULL || cri->cri_alg != CRYPTO_AES_CBC ||
318 	    cri->cri_klen != 128)
319 		return (EINVAL);
320 
321 	for (sesn = 0; sesn < sc->sc_nsessions; sesn++) {
322 		if (sc->sc_sessions[sesn].ses_used == 0) {
323 			ses = &sc->sc_sessions[sesn];
324 			break;
325 		}
326 	}
327 
328 	if (ses == NULL) {
329 		sesn = sc->sc_nsessions;
330 		ses = malloc((sesn + 1) * sizeof(*ses), M_DEVBUF, M_NOWAIT);
331 		if (ses == NULL)
332 			return (ENOMEM);
333 		if (sesn != 0) {
334 			memcpy(ses, sc->sc_sessions, sesn * sizeof(*ses));
335 			memset(sc->sc_sessions, 0, sesn * sizeof(*ses));
336 			free(sc->sc_sessions, M_DEVBUF);
337 		}
338 		sc->sc_sessions = ses;
339 		ses = &sc->sc_sessions[sesn];
340 		sc->sc_nsessions++;
341 	}
342 
343 	memset(ses, 0, sizeof(*ses));
344 	ses->ses_used = 1;
345 
346 	ses->ses_klen = cri->cri_klen;
347 
348 	/* Copy the key (Geode LX wants the primary key only) */
349 	memcpy(ses->ses_key, cri->cri_key, sizeof(ses->ses_key));
350 
351 	*sidp = GLXSB_SID(0, sesn);
352 	return (0);
353 }
354 
355 void
glxsb_crypto_freesession(void * aux,uint64_t tid)356 glxsb_crypto_freesession(void *aux, uint64_t tid)
357 {
358 	struct glxsb_softc *sc = aux;
359 	int sesn;
360 	uint32_t sid = ((uint32_t)tid) & 0xffffffff;
361 
362 	sesn = GLXSB_SESSION(sid);
363 	KASSERTMSG(sesn < sc->sc_nsessions, "sesn=%d nsessions=%d",
364 	    sesn, sc->sc_nsessions);
365 
366 	memset(&sc->sc_sessions[sesn], 0, sizeof(sc->sc_sessions[sesn]));
367 }
368 
369 /*
370  * Must be called at splnet() or higher
371  */
372 static __inline void
glxsb_aes(struct glxsb_softc * sc,uint32_t control,uint32_t psrc,uint32_t pdst,void * key,int len,void * iv)373 glxsb_aes(struct glxsb_softc *sc, uint32_t control, uint32_t psrc,
374     uint32_t pdst, void *key, int len, void *iv)
375 {
376 	uint32_t status;
377 	int i;
378 
379 	if (len & 0xF) {
380 		printf("%s: len must be a multiple of 16 (not %d)\n",
381 		    device_xname(sc->sc_dev), len);
382 		return;
383 	}
384 
385 	/* Set the source */
386 	bus_space_write_4(sc->sc_iot, sc->sc_ioh, SB_SOURCE_A, psrc);
387 
388 	/* Set the destination address */
389 	bus_space_write_4(sc->sc_iot, sc->sc_ioh, SB_DEST_A, pdst);
390 
391 	/* Set the data length */
392 	bus_space_write_4(sc->sc_iot, sc->sc_ioh, SB_LENGTH_A, len);
393 
394 	/* Set the IV */
395 	if (iv != NULL) {
396 		bus_space_write_region_4(sc->sc_iot, sc->sc_ioh,
397 		    SB_CBC_IV, iv, 4);
398 		control |= SB_CTL_CBC;
399 	}
400 
401 	/* Set the key */
402 	bus_space_write_region_4(sc->sc_iot, sc->sc_ioh, SB_WKEY, key, 4);
403 
404 	/* Ask the security block to do it */
405 	bus_space_write_4(sc->sc_iot, sc->sc_ioh, SB_CTL_A,
406 	    control | SB_CTL_WK | SB_CTL_DC | SB_CTL_SC | SB_CTL_ST);
407 
408 	/*
409 	 * Now wait until it is done.
410 	 *
411 	 * We do a busy wait.  Obviously the number of iterations of
412 	 * the loop required to perform the AES operation depends upon
413 	 * the number of bytes to process.
414 	 *
415 	 * On a 500 MHz Geode LX we see
416 	 *
417 	 *	length (bytes)	typical max iterations
418 	 *	    16		   12
419 	 *	    64		   22
420 	 *	   256		   59
421 	 *	  1024		  212
422 	 *	  8192		1,537
423 	 *
424 	 * Since we have a maximum size of operation defined in
425 	 * GLXSB_MAX_AES_LEN, we use this constant to decide how long
426 	 * to wait.  Allow an order of magnitude longer than it should
427 	 * really take, just in case.
428 	 */
429 	for (i = 0; i < GLXSB_MAX_AES_LEN * 10; i++) {
430 		status = bus_space_read_4(sc->sc_iot, sc->sc_ioh, SB_CTL_A);
431 
432 		if ((status & SB_CTL_ST) == 0)		/* Done */
433 			return;
434 	}
435 
436 	aprint_error_dev(sc->sc_dev, "operation failed to complete\n");
437 }
438 
439 int
glxsb_crypto_process(void * aux,struct cryptop * crp,int hint)440 glxsb_crypto_process(void *aux, struct cryptop *crp, int hint)
441 {
442 	struct glxsb_softc *sc = aux;
443 	struct glxsb_session *ses;
444 	struct cryptodesc *crd;
445 	char *op_src, *op_dst;
446 	uint32_t op_psrc, op_pdst;
447 	uint8_t op_iv[SB_AES_BLOCK_SIZE];
448 	int sesn, err = 0;
449 	int len, tlen, xlen;
450 	int offset;
451 	uint32_t control;
452 	int s;
453 
454 	s = splnet();
455 
456 	if (crp == NULL || crp->crp_callback == NULL) {
457 		err = EINVAL;
458 		goto out;
459 	}
460 	crd = crp->crp_desc;
461 	if (crd == NULL || crd->crd_next != NULL ||
462 	    crd->crd_alg != CRYPTO_AES_CBC ||
463 	    (crd->crd_len % SB_AES_BLOCK_SIZE) != 0) {
464 		err = EINVAL;
465 		goto out;
466 	}
467 
468 	sesn = GLXSB_SESSION(crp->crp_sid);
469 	if (sesn >= sc->sc_nsessions) {
470 		err = EINVAL;
471 		goto out;
472 	}
473 	ses = &sc->sc_sessions[sesn];
474 
475 	/* How much of our buffer will we need to use? */
476 	xlen = crd->crd_len > GLXSB_MAX_AES_LEN ?
477 	    GLXSB_MAX_AES_LEN : crd->crd_len;
478 
479 	/*
480 	 * XXX Check if we can have input == output on Geode LX.
481 	 * XXX In the meantime, use two separate (adjacent) buffers.
482 	 */
483 	op_src = sc->sc_dma.dma_vaddr;
484 	op_dst = (char *)sc->sc_dma.dma_vaddr + xlen;
485 
486 	op_psrc = sc->sc_dma.dma_paddr;
487 	op_pdst = sc->sc_dma.dma_paddr + xlen;
488 
489 	if (crd->crd_flags & CRD_F_ENCRYPT) {
490 		control = SB_CTL_ENC;
491 		if (crd->crd_flags & CRD_F_IV_EXPLICIT)
492 			memcpy(op_iv, crd->crd_iv, sizeof(op_iv));
493 		else
494 			cprng_fast(op_iv, sizeof(op_iv));
495 
496 		if ((crd->crd_flags & CRD_F_IV_PRESENT) == 0) {
497 			if (crp->crp_flags & CRYPTO_F_IMBUF)
498 				m_copyback((struct mbuf *)crp->crp_buf,
499 				    crd->crd_inject, sizeof(op_iv), op_iv);
500 			else if (crp->crp_flags & CRYPTO_F_IOV)
501 				cuio_copyback((struct uio *)crp->crp_buf,
502 				    crd->crd_inject, sizeof(op_iv), op_iv);
503 			else
504 				bcopy(op_iv,
505 				    (char *)crp->crp_buf + crd->crd_inject,
506 				    sizeof(op_iv));
507 		}
508 	} else {
509 		control = SB_CTL_DEC;
510 		if (crd->crd_flags & CRD_F_IV_EXPLICIT)
511 			memcpy(op_iv, crd->crd_iv, sizeof(op_iv));
512 		else {
513 			if (crp->crp_flags & CRYPTO_F_IMBUF)
514 				m_copydata((struct mbuf *)crp->crp_buf,
515 				    crd->crd_inject, sizeof(op_iv), op_iv);
516 			else if (crp->crp_flags & CRYPTO_F_IOV)
517 				cuio_copydata((struct uio *)crp->crp_buf,
518 				    crd->crd_inject, sizeof(op_iv), op_iv);
519 			else
520 				bcopy((char *)crp->crp_buf + crd->crd_inject,
521 				    op_iv, sizeof(op_iv));
522 		}
523 	}
524 
525 	offset = 0;
526 	tlen = crd->crd_len;
527 
528 	/* Process the data in GLXSB_MAX_AES_LEN chunks */
529 	while (tlen > 0) {
530 		len = (tlen > GLXSB_MAX_AES_LEN) ? GLXSB_MAX_AES_LEN : tlen;
531 
532 		if (crp->crp_flags & CRYPTO_F_IMBUF)
533 			m_copydata((struct mbuf *)crp->crp_buf,
534 			    crd->crd_skip + offset, len, op_src);
535 		else if (crp->crp_flags & CRYPTO_F_IOV)
536 			cuio_copydata((struct uio *)crp->crp_buf,
537 			    crd->crd_skip + offset, len, op_src);
538 		else
539 			bcopy((char *)crp->crp_buf + crd->crd_skip + offset,
540 			    op_src, len);
541 
542 		glxsb_dma_pre_op(sc, &sc->sc_dma);
543 
544 		glxsb_aes(sc, control, op_psrc, op_pdst, ses->ses_key,
545 		    len, op_iv);
546 
547 		glxsb_dma_post_op(sc, &sc->sc_dma);
548 
549 		if (crp->crp_flags & CRYPTO_F_IMBUF)
550 			m_copyback((struct mbuf *)crp->crp_buf,
551 			    crd->crd_skip + offset, len, op_dst);
552 		else if (crp->crp_flags & CRYPTO_F_IOV)
553 			cuio_copyback((struct uio *)crp->crp_buf,
554 			    crd->crd_skip + offset, len, op_dst);
555 		else
556 			memcpy((char *)crp->crp_buf + crd->crd_skip + offset,
557 			    op_dst, len);
558 
559 		offset += len;
560 		tlen -= len;
561 
562 		if (crd->crd_flags & CRD_F_ENCRYPT) {
563 			memcpy(op_iv, op_dst + len - sizeof(op_iv),
564 			    sizeof(op_iv));
565 		} else {
566 			/* Decryption, only need this if another iteration */
567 			if (tlen > 0) {
568 				memcpy(op_iv, op_src + len - sizeof(op_iv),
569 				    sizeof(op_iv));
570 			}
571 		}
572 	}
573 
574 	/* All AES processing has now been done. */
575 
576 	memset(sc->sc_dma.dma_vaddr, 0, xlen * 2);
577 out:
578 	crp->crp_etype = err;
579 	crypto_done(crp);
580 	splx(s);
581 	return 0;
582 }
583 
584 int
glxsb_dma_alloc(struct glxsb_softc * sc,int size,struct glxsb_dma_map * dma)585 glxsb_dma_alloc(struct glxsb_softc *sc, int size, struct glxsb_dma_map *dma)
586 {
587 	int rc;
588 
589 	dma->dma_nsegs = 1;
590 	dma->dma_size = size;
591 
592 	rc = bus_dmamap_create(sc->sc_dmat, size, dma->dma_nsegs, size,
593 	    0, BUS_DMA_NOWAIT, &dma->dma_map);
594 	if (rc != 0) {
595 		aprint_error_dev(sc->sc_dev,
596 		    "couldn't create DMA map for %d bytes (%d)\n", size, rc);
597 
598 		goto fail0;
599 	}
600 
601 	rc = bus_dmamem_alloc(sc->sc_dmat, size, SB_AES_ALIGN, 0,
602 	    &dma->dma_seg, dma->dma_nsegs, &dma->dma_nsegs, BUS_DMA_NOWAIT);
603 	if (rc != 0) {
604 		aprint_error_dev(sc->sc_dev,
605 		    "couldn't allocate DMA memory of %d bytes (%d)\n",
606 		    size, rc);
607 
608 		goto fail1;
609 	}
610 
611 	rc = bus_dmamem_map(sc->sc_dmat, &dma->dma_seg, 1, size,
612 	    &dma->dma_vaddr, BUS_DMA_NOWAIT);
613 	if (rc != 0) {
614 		aprint_error_dev(sc->sc_dev,
615 		    "couldn't map DMA memory for %d bytes (%d)\n", size, rc);
616 
617 		goto fail2;
618 	}
619 
620 	rc = bus_dmamap_load(sc->sc_dmat, dma->dma_map, dma->dma_vaddr,
621 	    size, NULL, BUS_DMA_NOWAIT);
622 	if (rc != 0) {
623 		aprint_error_dev(sc->sc_dev,
624 		    "couldn't load DMA memory for %d bytes (%d)\n", size, rc);
625 
626 		goto fail3;
627 	}
628 
629 	dma->dma_paddr = dma->dma_map->dm_segs[0].ds_addr;
630 
631 	return 0;
632 
633 fail3:
634 	bus_dmamem_unmap(sc->sc_dmat, dma->dma_vaddr, size);
635 fail2:
636 	bus_dmamem_free(sc->sc_dmat, &dma->dma_seg, dma->dma_nsegs);
637 fail1:
638 	bus_dmamap_destroy(sc->sc_dmat, dma->dma_map);
639 fail0:
640 	return rc;
641 }
642 
643 void
glxsb_dma_pre_op(struct glxsb_softc * sc,struct glxsb_dma_map * dma)644 glxsb_dma_pre_op(struct glxsb_softc *sc, struct glxsb_dma_map *dma)
645 {
646 	bus_dmamap_sync(sc->sc_dmat, dma->dma_map, 0, dma->dma_size,
647 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
648 }
649 
650 void
glxsb_dma_post_op(struct glxsb_softc * sc,struct glxsb_dma_map * dma)651 glxsb_dma_post_op(struct glxsb_softc *sc, struct glxsb_dma_map *dma)
652 {
653 	bus_dmamap_sync(sc->sc_dmat, dma->dma_map, 0, dma->dma_size,
654 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
655 }
656 
657 void
glxsb_dma_free(struct glxsb_softc * sc,struct glxsb_dma_map * dma)658 glxsb_dma_free(struct glxsb_softc *sc, struct glxsb_dma_map *dma)
659 {
660 	bus_dmamap_unload(sc->sc_dmat, dma->dma_map);
661 	bus_dmamem_unmap(sc->sc_dmat, dma->dma_vaddr, dma->dma_size);
662 	bus_dmamem_free(sc->sc_dmat, &dma->dma_seg, dma->dma_nsegs);
663 	bus_dmamap_destroy(sc->sc_dmat, dma->dma_map);
664 }
665