1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 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 __FBSDID("$FreeBSD$"); 32 33 /* 34 * SafeNet SafeXcel-1141 hardware crypto accelerator 35 */ 36 #include "opt_safe.h" 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/proc.h> 41 #include <sys/errno.h> 42 #include <sys/malloc.h> 43 #include <sys/kernel.h> 44 #include <sys/mbuf.h> 45 #include <sys/module.h> 46 #include <sys/lock.h> 47 #include <sys/mutex.h> 48 #include <sys/sysctl.h> 49 #include <sys/endian.h> 50 51 #include <vm/vm.h> 52 #include <vm/pmap.h> 53 54 #include <machine/bus.h> 55 #include <machine/resource.h> 56 #include <sys/bus.h> 57 #include <sys/rman.h> 58 59 #include <crypto/sha1.h> 60 #include <opencrypto/cryptodev.h> 61 #include <opencrypto/cryptosoft.h> 62 #include <sys/md5.h> 63 #include <sys/random.h> 64 #include <sys/kobj.h> 65 66 #include "cryptodev_if.h" 67 68 #include <dev/pci/pcivar.h> 69 #include <dev/pci/pcireg.h> 70 71 #ifdef SAFE_RNDTEST 72 #include <dev/rndtest/rndtest.h> 73 #endif 74 #include <dev/safe/safereg.h> 75 #include <dev/safe/safevar.h> 76 77 #ifndef bswap32 78 #define bswap32 NTOHL 79 #endif 80 81 /* 82 * Prototypes and count for the pci_device structure 83 */ 84 static int safe_probe(device_t); 85 static int safe_attach(device_t); 86 static int safe_detach(device_t); 87 static int safe_suspend(device_t); 88 static int safe_resume(device_t); 89 static int safe_shutdown(device_t); 90 91 static int safe_newsession(device_t, crypto_session_t, struct cryptoini *); 92 static int safe_process(device_t, struct cryptop *, int); 93 94 static device_method_t safe_methods[] = { 95 /* Device interface */ 96 DEVMETHOD(device_probe, safe_probe), 97 DEVMETHOD(device_attach, safe_attach), 98 DEVMETHOD(device_detach, safe_detach), 99 DEVMETHOD(device_suspend, safe_suspend), 100 DEVMETHOD(device_resume, safe_resume), 101 DEVMETHOD(device_shutdown, safe_shutdown), 102 103 /* crypto device methods */ 104 DEVMETHOD(cryptodev_newsession, safe_newsession), 105 DEVMETHOD(cryptodev_process, safe_process), 106 107 DEVMETHOD_END 108 }; 109 static driver_t safe_driver = { 110 "safe", 111 safe_methods, 112 sizeof (struct safe_softc) 113 }; 114 static devclass_t safe_devclass; 115 116 DRIVER_MODULE(safe, pci, safe_driver, safe_devclass, 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, devinfo; 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 devinfo = READ_REG(sc, SAFE_DEVINFO); 378 if (devinfo & SAFE_DEVINFO_RNG) { 379 sc->sc_flags |= SAFE_FLAGS_RNG; 380 printf(" rng"); 381 } 382 if (devinfo & SAFE_DEVINFO_PKEY) { 383 #if 0 384 printf(" key"); 385 sc->sc_flags |= SAFE_FLAGS_KEY; 386 crypto_kregister(sc->sc_cid, CRK_MOD_EXP, 0); 387 crypto_kregister(sc->sc_cid, CRK_MOD_EXP_CRT, 0); 388 #endif 389 } 390 if (devinfo & SAFE_DEVINFO_DES) { 391 printf(" des/3des"); 392 crypto_register(sc->sc_cid, CRYPTO_3DES_CBC, 0, 0); 393 crypto_register(sc->sc_cid, CRYPTO_DES_CBC, 0, 0); 394 } 395 if (devinfo & SAFE_DEVINFO_AES) { 396 printf(" aes"); 397 crypto_register(sc->sc_cid, CRYPTO_AES_CBC, 0, 0); 398 } 399 if (devinfo & SAFE_DEVINFO_MD5) { 400 printf(" md5"); 401 crypto_register(sc->sc_cid, CRYPTO_MD5_HMAC, 0, 0); 402 } 403 if (devinfo & SAFE_DEVINFO_SHA1) { 404 printf(" sha1"); 405 crypto_register(sc->sc_cid, CRYPTO_SHA1_HMAC, 0, 0); 406 } 407 printf(" null"); 408 crypto_register(sc->sc_cid, CRYPTO_NULL_CBC, 0, 0); 409 crypto_register(sc->sc_cid, CRYPTO_NULL_HMAC, 0, 0); 410 /* XXX other supported algorithms */ 411 printf("\n"); 412 413 safe_reset_board(sc); /* reset h/w */ 414 safe_init_pciregs(dev); /* init pci settings */ 415 safe_init_board(sc); /* init h/w */ 416 417 #ifndef SAFE_NO_RNG 418 if (sc->sc_flags & SAFE_FLAGS_RNG) { 419 #ifdef SAFE_RNDTEST 420 sc->sc_rndtest = rndtest_attach(dev); 421 if (sc->sc_rndtest) 422 sc->sc_harvest = rndtest_harvest; 423 else 424 sc->sc_harvest = default_harvest; 425 #else 426 sc->sc_harvest = default_harvest; 427 #endif 428 safe_rng_init(sc); 429 430 callout_init(&sc->sc_rngto, 1); 431 callout_reset(&sc->sc_rngto, hz*safe_rnginterval, safe_rng, sc); 432 } 433 #endif /* SAFE_NO_RNG */ 434 #ifdef SAFE_DEBUG 435 safec = sc; /* for use by hw.safe.dump */ 436 #endif 437 return (0); 438 bad4: 439 crypto_unregister_all(sc->sc_cid); 440 bad3: 441 bus_teardown_intr(dev, sc->sc_irq, sc->sc_ih); 442 bad2: 443 bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq); 444 bad1: 445 bus_release_resource(dev, SYS_RES_MEMORY, BS_BAR, sc->sc_sr); 446 bad: 447 return (ENXIO); 448 } 449 450 /* 451 * Detach a device that successfully probed. 452 */ 453 static int 454 safe_detach(device_t dev) 455 { 456 struct safe_softc *sc = device_get_softc(dev); 457 458 /* XXX wait/abort active ops */ 459 460 WRITE_REG(sc, SAFE_HI_MASK, 0); /* disable interrupts */ 461 462 callout_stop(&sc->sc_rngto); 463 464 crypto_unregister_all(sc->sc_cid); 465 466 #ifdef SAFE_RNDTEST 467 if (sc->sc_rndtest) 468 rndtest_detach(sc->sc_rndtest); 469 #endif 470 471 safe_cleanchip(sc); 472 safe_dma_free(sc, &sc->sc_dpalloc); 473 safe_dma_free(sc, &sc->sc_spalloc); 474 mtx_destroy(&sc->sc_ringmtx); 475 safe_dma_free(sc, &sc->sc_ringalloc); 476 477 bus_generic_detach(dev); 478 bus_teardown_intr(dev, sc->sc_irq, sc->sc_ih); 479 bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq); 480 481 bus_dma_tag_destroy(sc->sc_srcdmat); 482 bus_dma_tag_destroy(sc->sc_dstdmat); 483 bus_release_resource(dev, SYS_RES_MEMORY, BS_BAR, sc->sc_sr); 484 485 return (0); 486 } 487 488 /* 489 * Stop all chip i/o so that the kernel's probe routines don't 490 * get confused by errant DMAs when rebooting. 491 */ 492 static int 493 safe_shutdown(device_t dev) 494 { 495 #ifdef notyet 496 safe_stop(device_get_softc(dev)); 497 #endif 498 return (0); 499 } 500 501 /* 502 * Device suspend routine. 503 */ 504 static int 505 safe_suspend(device_t dev) 506 { 507 struct safe_softc *sc = device_get_softc(dev); 508 509 #ifdef notyet 510 /* XXX stop the device and save PCI settings */ 511 #endif 512 sc->sc_suspended = 1; 513 514 return (0); 515 } 516 517 static int 518 safe_resume(device_t dev) 519 { 520 struct safe_softc *sc = device_get_softc(dev); 521 522 #ifdef notyet 523 /* XXX retore PCI settings and start the device */ 524 #endif 525 sc->sc_suspended = 0; 526 return (0); 527 } 528 529 /* 530 * SafeXcel Interrupt routine 531 */ 532 static void 533 safe_intr(void *arg) 534 { 535 struct safe_softc *sc = arg; 536 volatile u_int32_t stat; 537 538 stat = READ_REG(sc, SAFE_HM_STAT); 539 if (stat == 0) /* shared irq, not for us */ 540 return; 541 542 WRITE_REG(sc, SAFE_HI_CLR, stat); /* IACK */ 543 544 if ((stat & SAFE_INT_PE_DDONE)) { 545 /* 546 * Descriptor(s) done; scan the ring and 547 * process completed operations. 548 */ 549 mtx_lock(&sc->sc_ringmtx); 550 while (sc->sc_back != sc->sc_front) { 551 struct safe_ringentry *re = sc->sc_back; 552 #ifdef SAFE_DEBUG 553 if (safe_debug) { 554 safe_dump_ringstate(sc, __func__); 555 safe_dump_request(sc, __func__, re); 556 } 557 #endif 558 /* 559 * safe_process marks ring entries that were allocated 560 * but not used with a csr of zero. This insures the 561 * ring front pointer never needs to be set backwards 562 * in the event that an entry is allocated but not used 563 * because of a setup error. 564 */ 565 if (re->re_desc.d_csr != 0) { 566 if (!SAFE_PE_CSR_IS_DONE(re->re_desc.d_csr)) 567 break; 568 if (!SAFE_PE_LEN_IS_DONE(re->re_desc.d_len)) 569 break; 570 sc->sc_nqchip--; 571 safe_callback(sc, re); 572 } 573 if (++(sc->sc_back) == sc->sc_ringtop) 574 sc->sc_back = sc->sc_ring; 575 } 576 mtx_unlock(&sc->sc_ringmtx); 577 } 578 579 /* 580 * Check to see if we got any DMA Error 581 */ 582 if (stat & SAFE_INT_PE_ERROR) { 583 DPRINTF(("dmaerr dmastat %08x\n", 584 READ_REG(sc, SAFE_PE_DMASTAT))); 585 safestats.st_dmaerr++; 586 safe_totalreset(sc); 587 #if 0 588 safe_feed(sc); 589 #endif 590 } 591 592 if (sc->sc_needwakeup) { /* XXX check high watermark */ 593 int wakeup = sc->sc_needwakeup & (CRYPTO_SYMQ|CRYPTO_ASYMQ); 594 DPRINTF(("%s: wakeup crypto %x\n", __func__, 595 sc->sc_needwakeup)); 596 sc->sc_needwakeup &= ~wakeup; 597 crypto_unblock(sc->sc_cid, wakeup); 598 } 599 } 600 601 /* 602 * safe_feed() - post a request to chip 603 */ 604 static void 605 safe_feed(struct safe_softc *sc, struct safe_ringentry *re) 606 { 607 bus_dmamap_sync(sc->sc_srcdmat, re->re_src_map, BUS_DMASYNC_PREWRITE); 608 if (re->re_dst_map != NULL) 609 bus_dmamap_sync(sc->sc_dstdmat, re->re_dst_map, 610 BUS_DMASYNC_PREREAD); 611 /* XXX have no smaller granularity */ 612 safe_dma_sync(&sc->sc_ringalloc, 613 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 614 safe_dma_sync(&sc->sc_spalloc, BUS_DMASYNC_PREWRITE); 615 safe_dma_sync(&sc->sc_dpalloc, BUS_DMASYNC_PREWRITE); 616 617 #ifdef SAFE_DEBUG 618 if (safe_debug) { 619 safe_dump_ringstate(sc, __func__); 620 safe_dump_request(sc, __func__, re); 621 } 622 #endif 623 sc->sc_nqchip++; 624 if (sc->sc_nqchip > safestats.st_maxqchip) 625 safestats.st_maxqchip = sc->sc_nqchip; 626 /* poke h/w to check descriptor ring, any value can be written */ 627 WRITE_REG(sc, SAFE_HI_RD_DESCR, 0); 628 } 629 630 #define N(a) (sizeof(a) / sizeof (a[0])) 631 static void 632 safe_setup_enckey(struct safe_session *ses, caddr_t key) 633 { 634 int i; 635 636 bcopy(key, ses->ses_key, ses->ses_klen / 8); 637 638 /* PE is little-endian, insure proper byte order */ 639 for (i = 0; i < N(ses->ses_key); i++) 640 ses->ses_key[i] = htole32(ses->ses_key[i]); 641 } 642 643 static void 644 safe_setup_mackey(struct safe_session *ses, int algo, caddr_t key, int klen) 645 { 646 MD5_CTX md5ctx; 647 SHA1_CTX sha1ctx; 648 int i; 649 650 651 for (i = 0; i < klen; i++) 652 key[i] ^= HMAC_IPAD_VAL; 653 654 if (algo == CRYPTO_MD5_HMAC) { 655 MD5Init(&md5ctx); 656 MD5Update(&md5ctx, key, klen); 657 MD5Update(&md5ctx, hmac_ipad_buffer, MD5_BLOCK_LEN - klen); 658 bcopy(md5ctx.state, ses->ses_hminner, sizeof(md5ctx.state)); 659 } else { 660 SHA1Init(&sha1ctx); 661 SHA1Update(&sha1ctx, key, klen); 662 SHA1Update(&sha1ctx, hmac_ipad_buffer, 663 SHA1_BLOCK_LEN - klen); 664 bcopy(sha1ctx.h.b32, ses->ses_hminner, sizeof(sha1ctx.h.b32)); 665 } 666 667 for (i = 0; i < klen; i++) 668 key[i] ^= (HMAC_IPAD_VAL ^ HMAC_OPAD_VAL); 669 670 if (algo == CRYPTO_MD5_HMAC) { 671 MD5Init(&md5ctx); 672 MD5Update(&md5ctx, key, klen); 673 MD5Update(&md5ctx, hmac_opad_buffer, MD5_BLOCK_LEN - klen); 674 bcopy(md5ctx.state, ses->ses_hmouter, sizeof(md5ctx.state)); 675 } else { 676 SHA1Init(&sha1ctx); 677 SHA1Update(&sha1ctx, key, klen); 678 SHA1Update(&sha1ctx, hmac_opad_buffer, 679 SHA1_BLOCK_LEN - klen); 680 bcopy(sha1ctx.h.b32, ses->ses_hmouter, sizeof(sha1ctx.h.b32)); 681 } 682 683 for (i = 0; i < klen; i++) 684 key[i] ^= HMAC_OPAD_VAL; 685 686 /* PE is little-endian, insure proper byte order */ 687 for (i = 0; i < N(ses->ses_hminner); i++) { 688 ses->ses_hminner[i] = htole32(ses->ses_hminner[i]); 689 ses->ses_hmouter[i] = htole32(ses->ses_hmouter[i]); 690 } 691 } 692 #undef N 693 694 /* 695 * Allocate a new 'session' and return an encoded session id. 'sidp' 696 * contains our registration id, and should contain an encoded session 697 * id on successful allocation. 698 */ 699 static int 700 safe_newsession(device_t dev, crypto_session_t cses, struct cryptoini *cri) 701 { 702 struct safe_softc *sc = device_get_softc(dev); 703 struct cryptoini *c, *encini = NULL, *macini = NULL; 704 struct safe_session *ses = NULL; 705 706 if (cri == NULL || sc == NULL) 707 return (EINVAL); 708 709 for (c = cri; c != NULL; c = c->cri_next) { 710 if (c->cri_alg == CRYPTO_MD5_HMAC || 711 c->cri_alg == CRYPTO_SHA1_HMAC || 712 c->cri_alg == CRYPTO_NULL_HMAC) { 713 if (macini) 714 return (EINVAL); 715 macini = c; 716 } else if (c->cri_alg == CRYPTO_DES_CBC || 717 c->cri_alg == CRYPTO_3DES_CBC || 718 c->cri_alg == CRYPTO_AES_CBC || 719 c->cri_alg == CRYPTO_NULL_CBC) { 720 if (encini) 721 return (EINVAL); 722 encini = c; 723 } else 724 return (EINVAL); 725 } 726 if (encini == NULL && macini == NULL) 727 return (EINVAL); 728 if (encini) { /* validate key length */ 729 switch (encini->cri_alg) { 730 case CRYPTO_DES_CBC: 731 if (encini->cri_klen != 64) 732 return (EINVAL); 733 break; 734 case CRYPTO_3DES_CBC: 735 if (encini->cri_klen != 192) 736 return (EINVAL); 737 break; 738 case CRYPTO_AES_CBC: 739 if (encini->cri_klen != 128 && 740 encini->cri_klen != 192 && 741 encini->cri_klen != 256) 742 return (EINVAL); 743 break; 744 } 745 } 746 747 ses = crypto_get_driver_session(cses); 748 if (encini) { 749 /* get an IV */ 750 /* XXX may read fewer than requested */ 751 read_random(ses->ses_iv, sizeof(ses->ses_iv)); 752 753 ses->ses_klen = encini->cri_klen; 754 if (encini->cri_key != NULL) 755 safe_setup_enckey(ses, encini->cri_key); 756 } 757 758 if (macini) { 759 ses->ses_mlen = macini->cri_mlen; 760 if (ses->ses_mlen == 0) { 761 if (macini->cri_alg == CRYPTO_MD5_HMAC) 762 ses->ses_mlen = MD5_HASH_LEN; 763 else 764 ses->ses_mlen = SHA1_HASH_LEN; 765 } 766 767 if (macini->cri_key != NULL) { 768 safe_setup_mackey(ses, macini->cri_alg, macini->cri_key, 769 macini->cri_klen / 8); 770 } 771 } 772 773 return (0); 774 } 775 776 static void 777 safe_op_cb(void *arg, bus_dma_segment_t *seg, int nsegs, bus_size_t mapsize, int error) 778 { 779 struct safe_operand *op = arg; 780 781 DPRINTF(("%s: mapsize %u nsegs %d error %d\n", __func__, 782 (u_int) mapsize, nsegs, error)); 783 if (error != 0) 784 return; 785 op->mapsize = mapsize; 786 op->nsegs = nsegs; 787 bcopy(seg, op->segs, nsegs * sizeof (seg[0])); 788 } 789 790 static int 791 safe_process(device_t dev, struct cryptop *crp, int hint) 792 { 793 struct safe_softc *sc = device_get_softc(dev); 794 int err = 0, i, nicealign, uniform; 795 struct cryptodesc *crd1, *crd2, *maccrd, *enccrd; 796 int bypass, oplen, ivsize; 797 caddr_t iv; 798 int16_t coffset; 799 struct safe_session *ses; 800 struct safe_ringentry *re; 801 struct safe_sarec *sa; 802 struct safe_pdesc *pd; 803 u_int32_t cmd0, cmd1, staterec; 804 805 if (crp == NULL || crp->crp_callback == NULL || sc == NULL) { 806 safestats.st_invalid++; 807 return (EINVAL); 808 } 809 810 mtx_lock(&sc->sc_ringmtx); 811 if (sc->sc_front == sc->sc_back && sc->sc_nqchip != 0) { 812 safestats.st_ringfull++; 813 sc->sc_needwakeup |= CRYPTO_SYMQ; 814 mtx_unlock(&sc->sc_ringmtx); 815 return (ERESTART); 816 } 817 re = sc->sc_front; 818 819 staterec = re->re_sa.sa_staterec; /* save */ 820 /* NB: zero everything but the PE descriptor */ 821 bzero(&re->re_sa, sizeof(struct safe_ringentry) - sizeof(re->re_desc)); 822 re->re_sa.sa_staterec = staterec; /* restore */ 823 824 re->re_crp = crp; 825 826 if (crp->crp_flags & CRYPTO_F_IMBUF) { 827 re->re_src_m = (struct mbuf *)crp->crp_buf; 828 re->re_dst_m = (struct mbuf *)crp->crp_buf; 829 } else if (crp->crp_flags & CRYPTO_F_IOV) { 830 re->re_src_io = (struct uio *)crp->crp_buf; 831 re->re_dst_io = (struct uio *)crp->crp_buf; 832 } else { 833 safestats.st_badflags++; 834 err = EINVAL; 835 goto errout; /* XXX we don't handle contiguous blocks! */ 836 } 837 838 sa = &re->re_sa; 839 ses = crypto_get_driver_session(crp->crp_session); 840 841 crd1 = crp->crp_desc; 842 if (crd1 == NULL) { 843 safestats.st_nodesc++; 844 err = EINVAL; 845 goto errout; 846 } 847 crd2 = crd1->crd_next; 848 849 cmd0 = SAFE_SA_CMD0_BASIC; /* basic group operation */ 850 cmd1 = 0; 851 if (crd2 == NULL) { 852 if (crd1->crd_alg == CRYPTO_MD5_HMAC || 853 crd1->crd_alg == CRYPTO_SHA1_HMAC || 854 crd1->crd_alg == CRYPTO_NULL_HMAC) { 855 maccrd = crd1; 856 enccrd = NULL; 857 cmd0 |= SAFE_SA_CMD0_OP_HASH; 858 } else if (crd1->crd_alg == CRYPTO_DES_CBC || 859 crd1->crd_alg == CRYPTO_3DES_CBC || 860 crd1->crd_alg == CRYPTO_AES_CBC || 861 crd1->crd_alg == CRYPTO_NULL_CBC) { 862 maccrd = NULL; 863 enccrd = crd1; 864 cmd0 |= SAFE_SA_CMD0_OP_CRYPT; 865 } else { 866 safestats.st_badalg++; 867 err = EINVAL; 868 goto errout; 869 } 870 } else { 871 if ((crd1->crd_alg == CRYPTO_MD5_HMAC || 872 crd1->crd_alg == CRYPTO_SHA1_HMAC || 873 crd1->crd_alg == CRYPTO_NULL_HMAC) && 874 (crd2->crd_alg == CRYPTO_DES_CBC || 875 crd2->crd_alg == CRYPTO_3DES_CBC || 876 crd2->crd_alg == CRYPTO_AES_CBC || 877 crd2->crd_alg == CRYPTO_NULL_CBC) && 878 ((crd2->crd_flags & CRD_F_ENCRYPT) == 0)) { 879 maccrd = crd1; 880 enccrd = crd2; 881 } else if ((crd1->crd_alg == CRYPTO_DES_CBC || 882 crd1->crd_alg == CRYPTO_3DES_CBC || 883 crd1->crd_alg == CRYPTO_AES_CBC || 884 crd1->crd_alg == CRYPTO_NULL_CBC) && 885 (crd2->crd_alg == CRYPTO_MD5_HMAC || 886 crd2->crd_alg == CRYPTO_SHA1_HMAC || 887 crd2->crd_alg == CRYPTO_NULL_HMAC) && 888 (crd1->crd_flags & CRD_F_ENCRYPT)) { 889 enccrd = crd1; 890 maccrd = crd2; 891 } else { 892 safestats.st_badalg++; 893 err = EINVAL; 894 goto errout; 895 } 896 cmd0 |= SAFE_SA_CMD0_OP_BOTH; 897 } 898 899 if (enccrd) { 900 if (enccrd->crd_flags & CRD_F_KEY_EXPLICIT) 901 safe_setup_enckey(ses, enccrd->crd_key); 902 903 if (enccrd->crd_alg == CRYPTO_DES_CBC) { 904 cmd0 |= SAFE_SA_CMD0_DES; 905 cmd1 |= SAFE_SA_CMD1_CBC; 906 ivsize = 2*sizeof(u_int32_t); 907 } else if (enccrd->crd_alg == CRYPTO_3DES_CBC) { 908 cmd0 |= SAFE_SA_CMD0_3DES; 909 cmd1 |= SAFE_SA_CMD1_CBC; 910 ivsize = 2*sizeof(u_int32_t); 911 } else if (enccrd->crd_alg == CRYPTO_AES_CBC) { 912 cmd0 |= SAFE_SA_CMD0_AES; 913 cmd1 |= SAFE_SA_CMD1_CBC; 914 if (ses->ses_klen == 128) 915 cmd1 |= SAFE_SA_CMD1_AES128; 916 else if (ses->ses_klen == 192) 917 cmd1 |= SAFE_SA_CMD1_AES192; 918 else 919 cmd1 |= SAFE_SA_CMD1_AES256; 920 ivsize = 4*sizeof(u_int32_t); 921 } else { 922 cmd0 |= SAFE_SA_CMD0_CRYPT_NULL; 923 ivsize = 0; 924 } 925 926 /* 927 * Setup encrypt/decrypt state. When using basic ops 928 * we can't use an inline IV because hash/crypt offset 929 * must be from the end of the IV to the start of the 930 * crypt data and this leaves out the preceding header 931 * from the hash calculation. Instead we place the IV 932 * in the state record and set the hash/crypt offset to 933 * copy both the header+IV. 934 */ 935 if (enccrd->crd_flags & CRD_F_ENCRYPT) { 936 cmd0 |= SAFE_SA_CMD0_OUTBOUND; 937 938 if (enccrd->crd_flags & CRD_F_IV_EXPLICIT) 939 iv = enccrd->crd_iv; 940 else 941 iv = (caddr_t) ses->ses_iv; 942 if ((enccrd->crd_flags & CRD_F_IV_PRESENT) == 0) { 943 crypto_copyback(crp->crp_flags, crp->crp_buf, 944 enccrd->crd_inject, ivsize, iv); 945 } 946 bcopy(iv, re->re_sastate.sa_saved_iv, ivsize); 947 cmd0 |= SAFE_SA_CMD0_IVLD_STATE | SAFE_SA_CMD0_SAVEIV; 948 re->re_flags |= SAFE_QFLAGS_COPYOUTIV; 949 } else { 950 cmd0 |= SAFE_SA_CMD0_INBOUND; 951 952 if (enccrd->crd_flags & CRD_F_IV_EXPLICIT) { 953 bcopy(enccrd->crd_iv, 954 re->re_sastate.sa_saved_iv, ivsize); 955 } else { 956 crypto_copydata(crp->crp_flags, crp->crp_buf, 957 enccrd->crd_inject, ivsize, 958 (caddr_t)re->re_sastate.sa_saved_iv); 959 } 960 cmd0 |= SAFE_SA_CMD0_IVLD_STATE; 961 } 962 /* 963 * For basic encryption use the zero pad algorithm. 964 * This pads results to an 8-byte boundary and 965 * suppresses padding verification for inbound (i.e. 966 * decrypt) operations. 967 * 968 * NB: Not sure if the 8-byte pad boundary is a problem. 969 */ 970 cmd0 |= SAFE_SA_CMD0_PAD_ZERO; 971 972 /* XXX assert key bufs have the same size */ 973 bcopy(ses->ses_key, sa->sa_key, sizeof(sa->sa_key)); 974 } 975 976 if (maccrd) { 977 if (maccrd->crd_flags & CRD_F_KEY_EXPLICIT) { 978 safe_setup_mackey(ses, maccrd->crd_alg, 979 maccrd->crd_key, maccrd->crd_klen / 8); 980 } 981 982 if (maccrd->crd_alg == CRYPTO_MD5_HMAC) { 983 cmd0 |= SAFE_SA_CMD0_MD5; 984 cmd1 |= SAFE_SA_CMD1_HMAC; /* NB: enable HMAC */ 985 } else if (maccrd->crd_alg == CRYPTO_SHA1_HMAC) { 986 cmd0 |= SAFE_SA_CMD0_SHA1; 987 cmd1 |= SAFE_SA_CMD1_HMAC; /* NB: enable HMAC */ 988 } else { 989 cmd0 |= SAFE_SA_CMD0_HASH_NULL; 990 } 991 /* 992 * Digest data is loaded from the SA and the hash 993 * result is saved to the state block where we 994 * retrieve it for return to the caller. 995 */ 996 /* XXX assert digest bufs have the same size */ 997 bcopy(ses->ses_hminner, sa->sa_indigest, 998 sizeof(sa->sa_indigest)); 999 bcopy(ses->ses_hmouter, sa->sa_outdigest, 1000 sizeof(sa->sa_outdigest)); 1001 1002 cmd0 |= SAFE_SA_CMD0_HSLD_SA | SAFE_SA_CMD0_SAVEHASH; 1003 re->re_flags |= SAFE_QFLAGS_COPYOUTICV; 1004 } 1005 1006 if (enccrd && maccrd) { 1007 /* 1008 * The offset from hash data to the start of 1009 * crypt data is the difference in the skips. 1010 */ 1011 bypass = maccrd->crd_skip; 1012 coffset = enccrd->crd_skip - maccrd->crd_skip; 1013 if (coffset < 0) { 1014 DPRINTF(("%s: hash does not precede crypt; " 1015 "mac skip %u enc skip %u\n", 1016 __func__, maccrd->crd_skip, enccrd->crd_skip)); 1017 safestats.st_skipmismatch++; 1018 err = EINVAL; 1019 goto errout; 1020 } 1021 oplen = enccrd->crd_skip + enccrd->crd_len; 1022 if (maccrd->crd_skip + maccrd->crd_len != oplen) { 1023 DPRINTF(("%s: hash amount %u != crypt amount %u\n", 1024 __func__, maccrd->crd_skip + maccrd->crd_len, 1025 oplen)); 1026 safestats.st_lenmismatch++; 1027 err = EINVAL; 1028 goto errout; 1029 } 1030 #ifdef SAFE_DEBUG 1031 if (safe_debug) { 1032 printf("mac: skip %d, len %d, inject %d\n", 1033 maccrd->crd_skip, maccrd->crd_len, 1034 maccrd->crd_inject); 1035 printf("enc: skip %d, len %d, inject %d\n", 1036 enccrd->crd_skip, enccrd->crd_len, 1037 enccrd->crd_inject); 1038 printf("bypass %d coffset %d oplen %d\n", 1039 bypass, coffset, oplen); 1040 } 1041 #endif 1042 if (coffset & 3) { /* offset must be 32-bit aligned */ 1043 DPRINTF(("%s: coffset %u misaligned\n", 1044 __func__, coffset)); 1045 safestats.st_coffmisaligned++; 1046 err = EINVAL; 1047 goto errout; 1048 } 1049 coffset >>= 2; 1050 if (coffset > 255) { /* offset must be <256 dwords */ 1051 DPRINTF(("%s: coffset %u too big\n", 1052 __func__, coffset)); 1053 safestats.st_cofftoobig++; 1054 err = EINVAL; 1055 goto errout; 1056 } 1057 /* 1058 * Tell the hardware to copy the header to the output. 1059 * The header is defined as the data from the end of 1060 * the bypass to the start of data to be encrypted. 1061 * Typically this is the inline IV. Note that you need 1062 * to do this even if src+dst are the same; it appears 1063 * that w/o this bit the crypted data is written 1064 * immediately after the bypass data. 1065 */ 1066 cmd1 |= SAFE_SA_CMD1_HDRCOPY; 1067 /* 1068 * Disable IP header mutable bit handling. This is 1069 * needed to get correct HMAC calculations. 1070 */ 1071 cmd1 |= SAFE_SA_CMD1_MUTABLE; 1072 } else { 1073 if (enccrd) { 1074 bypass = enccrd->crd_skip; 1075 oplen = bypass + enccrd->crd_len; 1076 } else { 1077 bypass = maccrd->crd_skip; 1078 oplen = bypass + maccrd->crd_len; 1079 } 1080 coffset = 0; 1081 } 1082 /* XXX verify multiple of 4 when using s/g */ 1083 if (bypass > 96) { /* bypass offset must be <= 96 bytes */ 1084 DPRINTF(("%s: bypass %u too big\n", __func__, bypass)); 1085 safestats.st_bypasstoobig++; 1086 err = EINVAL; 1087 goto errout; 1088 } 1089 1090 if (bus_dmamap_create(sc->sc_srcdmat, BUS_DMA_NOWAIT, &re->re_src_map)) { 1091 safestats.st_nomap++; 1092 err = ENOMEM; 1093 goto errout; 1094 } 1095 if (crp->crp_flags & CRYPTO_F_IMBUF) { 1096 if (bus_dmamap_load_mbuf(sc->sc_srcdmat, re->re_src_map, 1097 re->re_src_m, safe_op_cb, 1098 &re->re_src, BUS_DMA_NOWAIT) != 0) { 1099 bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map); 1100 re->re_src_map = NULL; 1101 safestats.st_noload++; 1102 err = ENOMEM; 1103 goto errout; 1104 } 1105 } else if (crp->crp_flags & CRYPTO_F_IOV) { 1106 if (bus_dmamap_load_uio(sc->sc_srcdmat, re->re_src_map, 1107 re->re_src_io, safe_op_cb, 1108 &re->re_src, BUS_DMA_NOWAIT) != 0) { 1109 bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map); 1110 re->re_src_map = NULL; 1111 safestats.st_noload++; 1112 err = ENOMEM; 1113 goto errout; 1114 } 1115 } 1116 nicealign = safe_dmamap_aligned(&re->re_src); 1117 uniform = safe_dmamap_uniform(&re->re_src); 1118 1119 DPRINTF(("src nicealign %u uniform %u nsegs %u\n", 1120 nicealign, uniform, re->re_src.nsegs)); 1121 if (re->re_src.nsegs > 1) { 1122 re->re_desc.d_src = sc->sc_spalloc.dma_paddr + 1123 ((caddr_t) sc->sc_spfree - (caddr_t) sc->sc_spring); 1124 for (i = 0; i < re->re_src_nsegs; i++) { 1125 /* NB: no need to check if there's space */ 1126 pd = sc->sc_spfree; 1127 if (++(sc->sc_spfree) == sc->sc_springtop) 1128 sc->sc_spfree = sc->sc_spring; 1129 1130 KASSERT((pd->pd_flags&3) == 0 || 1131 (pd->pd_flags&3) == SAFE_PD_DONE, 1132 ("bogus source particle descriptor; flags %x", 1133 pd->pd_flags)); 1134 pd->pd_addr = re->re_src_segs[i].ds_addr; 1135 pd->pd_size = re->re_src_segs[i].ds_len; 1136 pd->pd_flags = SAFE_PD_READY; 1137 } 1138 cmd0 |= SAFE_SA_CMD0_IGATHER; 1139 } else { 1140 /* 1141 * No need for gather, reference the operand directly. 1142 */ 1143 re->re_desc.d_src = re->re_src_segs[0].ds_addr; 1144 } 1145 1146 if (enccrd == NULL && maccrd != NULL) { 1147 /* 1148 * Hash op; no destination needed. 1149 */ 1150 } else { 1151 if (crp->crp_flags & CRYPTO_F_IOV) { 1152 if (!nicealign) { 1153 safestats.st_iovmisaligned++; 1154 err = EINVAL; 1155 goto errout; 1156 } 1157 if (uniform != 1) { 1158 /* 1159 * Source is not suitable for direct use as 1160 * the destination. Create a new scatter/gather 1161 * list based on the destination requirements 1162 * and check if that's ok. 1163 */ 1164 if (bus_dmamap_create(sc->sc_dstdmat, 1165 BUS_DMA_NOWAIT, &re->re_dst_map)) { 1166 safestats.st_nomap++; 1167 err = ENOMEM; 1168 goto errout; 1169 } 1170 if (bus_dmamap_load_uio(sc->sc_dstdmat, 1171 re->re_dst_map, re->re_dst_io, 1172 safe_op_cb, &re->re_dst, 1173 BUS_DMA_NOWAIT) != 0) { 1174 bus_dmamap_destroy(sc->sc_dstdmat, 1175 re->re_dst_map); 1176 re->re_dst_map = NULL; 1177 safestats.st_noload++; 1178 err = ENOMEM; 1179 goto errout; 1180 } 1181 uniform = safe_dmamap_uniform(&re->re_dst); 1182 if (!uniform) { 1183 /* 1184 * There's no way to handle the DMA 1185 * requirements with this uio. We 1186 * could create a separate DMA area for 1187 * the result and then copy it back, 1188 * but for now we just bail and return 1189 * an error. Note that uio requests 1190 * > SAFE_MAX_DSIZE are handled because 1191 * the DMA map and segment list for the 1192 * destination wil result in a 1193 * destination particle list that does 1194 * the necessary scatter DMA. 1195 */ 1196 safestats.st_iovnotuniform++; 1197 err = EINVAL; 1198 goto errout; 1199 } 1200 } else 1201 re->re_dst = re->re_src; 1202 } else if (crp->crp_flags & CRYPTO_F_IMBUF) { 1203 if (nicealign && uniform == 1) { 1204 /* 1205 * Source layout is suitable for direct 1206 * sharing of the DMA map and segment list. 1207 */ 1208 re->re_dst = re->re_src; 1209 } else if (nicealign && uniform == 2) { 1210 /* 1211 * The source is properly aligned but requires a 1212 * different particle list to handle DMA of the 1213 * result. Create a new map and do the load to 1214 * create the segment list. The particle 1215 * descriptor setup code below will handle the 1216 * rest. 1217 */ 1218 if (bus_dmamap_create(sc->sc_dstdmat, 1219 BUS_DMA_NOWAIT, &re->re_dst_map)) { 1220 safestats.st_nomap++; 1221 err = ENOMEM; 1222 goto errout; 1223 } 1224 if (bus_dmamap_load_mbuf(sc->sc_dstdmat, 1225 re->re_dst_map, re->re_dst_m, 1226 safe_op_cb, &re->re_dst, 1227 BUS_DMA_NOWAIT) != 0) { 1228 bus_dmamap_destroy(sc->sc_dstdmat, 1229 re->re_dst_map); 1230 re->re_dst_map = NULL; 1231 safestats.st_noload++; 1232 err = ENOMEM; 1233 goto errout; 1234 } 1235 } else { /* !(aligned and/or uniform) */ 1236 int totlen, len; 1237 struct mbuf *m, *top, **mp; 1238 1239 /* 1240 * DMA constraints require that we allocate a 1241 * new mbuf chain for the destination. We 1242 * allocate an entire new set of mbufs of 1243 * optimal/required size and then tell the 1244 * hardware to copy any bits that are not 1245 * created as a byproduct of the operation. 1246 */ 1247 if (!nicealign) 1248 safestats.st_unaligned++; 1249 if (!uniform) 1250 safestats.st_notuniform++; 1251 totlen = re->re_src_mapsize; 1252 if (re->re_src_m->m_flags & M_PKTHDR) { 1253 len = MHLEN; 1254 MGETHDR(m, M_NOWAIT, MT_DATA); 1255 if (m && !m_dup_pkthdr(m, re->re_src_m, 1256 M_NOWAIT)) { 1257 m_free(m); 1258 m = NULL; 1259 } 1260 } else { 1261 len = MLEN; 1262 MGET(m, M_NOWAIT, MT_DATA); 1263 } 1264 if (m == NULL) { 1265 safestats.st_nombuf++; 1266 err = sc->sc_nqchip ? ERESTART : ENOMEM; 1267 goto errout; 1268 } 1269 if (totlen >= MINCLSIZE) { 1270 if (!(MCLGET(m, M_NOWAIT))) { 1271 m_free(m); 1272 safestats.st_nomcl++; 1273 err = sc->sc_nqchip ? 1274 ERESTART : ENOMEM; 1275 goto errout; 1276 } 1277 len = MCLBYTES; 1278 } 1279 m->m_len = len; 1280 top = NULL; 1281 mp = ⊤ 1282 1283 while (totlen > 0) { 1284 if (top) { 1285 MGET(m, M_NOWAIT, MT_DATA); 1286 if (m == NULL) { 1287 m_freem(top); 1288 safestats.st_nombuf++; 1289 err = sc->sc_nqchip ? 1290 ERESTART : ENOMEM; 1291 goto errout; 1292 } 1293 len = MLEN; 1294 } 1295 if (top && totlen >= MINCLSIZE) { 1296 if (!(MCLGET(m, M_NOWAIT))) { 1297 *mp = m; 1298 m_freem(top); 1299 safestats.st_nomcl++; 1300 err = sc->sc_nqchip ? 1301 ERESTART : ENOMEM; 1302 goto errout; 1303 } 1304 len = MCLBYTES; 1305 } 1306 m->m_len = len = min(totlen, len); 1307 totlen -= len; 1308 *mp = m; 1309 mp = &m->m_next; 1310 } 1311 re->re_dst_m = top; 1312 if (bus_dmamap_create(sc->sc_dstdmat, 1313 BUS_DMA_NOWAIT, &re->re_dst_map) != 0) { 1314 safestats.st_nomap++; 1315 err = ENOMEM; 1316 goto errout; 1317 } 1318 if (bus_dmamap_load_mbuf(sc->sc_dstdmat, 1319 re->re_dst_map, re->re_dst_m, 1320 safe_op_cb, &re->re_dst, 1321 BUS_DMA_NOWAIT) != 0) { 1322 bus_dmamap_destroy(sc->sc_dstdmat, 1323 re->re_dst_map); 1324 re->re_dst_map = NULL; 1325 safestats.st_noload++; 1326 err = ENOMEM; 1327 goto errout; 1328 } 1329 if (re->re_src.mapsize > oplen) { 1330 /* 1331 * There's data following what the 1332 * hardware will copy for us. If this 1333 * isn't just the ICV (that's going to 1334 * be written on completion), copy it 1335 * to the new mbufs 1336 */ 1337 if (!(maccrd && 1338 (re->re_src.mapsize-oplen) == 12 && 1339 maccrd->crd_inject == oplen)) 1340 safe_mcopy(re->re_src_m, 1341 re->re_dst_m, 1342 oplen); 1343 else 1344 safestats.st_noicvcopy++; 1345 } 1346 } 1347 } else { 1348 safestats.st_badflags++; 1349 err = EINVAL; 1350 goto errout; 1351 } 1352 1353 if (re->re_dst.nsegs > 1) { 1354 re->re_desc.d_dst = sc->sc_dpalloc.dma_paddr + 1355 ((caddr_t) sc->sc_dpfree - (caddr_t) sc->sc_dpring); 1356 for (i = 0; i < re->re_dst_nsegs; i++) { 1357 pd = sc->sc_dpfree; 1358 KASSERT((pd->pd_flags&3) == 0 || 1359 (pd->pd_flags&3) == SAFE_PD_DONE, 1360 ("bogus dest particle descriptor; flags %x", 1361 pd->pd_flags)); 1362 if (++(sc->sc_dpfree) == sc->sc_dpringtop) 1363 sc->sc_dpfree = sc->sc_dpring; 1364 pd->pd_addr = re->re_dst_segs[i].ds_addr; 1365 pd->pd_flags = SAFE_PD_READY; 1366 } 1367 cmd0 |= SAFE_SA_CMD0_OSCATTER; 1368 } else { 1369 /* 1370 * No need for scatter, reference the operand directly. 1371 */ 1372 re->re_desc.d_dst = re->re_dst_segs[0].ds_addr; 1373 } 1374 } 1375 1376 /* 1377 * All done with setup; fillin the SA command words 1378 * and the packet engine descriptor. The operation 1379 * is now ready for submission to the hardware. 1380 */ 1381 sa->sa_cmd0 = cmd0 | SAFE_SA_CMD0_IPCI | SAFE_SA_CMD0_OPCI; 1382 sa->sa_cmd1 = cmd1 1383 | (coffset << SAFE_SA_CMD1_OFFSET_S) 1384 | SAFE_SA_CMD1_SAREV1 /* Rev 1 SA data structure */ 1385 | SAFE_SA_CMD1_SRPCI 1386 ; 1387 /* 1388 * NB: the order of writes is important here. In case the 1389 * chip is scanning the ring because of an outstanding request 1390 * it might nab this one too. In that case we need to make 1391 * sure the setup is complete before we write the length 1392 * field of the descriptor as it signals the descriptor is 1393 * ready for processing. 1394 */ 1395 re->re_desc.d_csr = SAFE_PE_CSR_READY | SAFE_PE_CSR_SAPCI; 1396 if (maccrd) 1397 re->re_desc.d_csr |= SAFE_PE_CSR_LOADSA | SAFE_PE_CSR_HASHFINAL; 1398 re->re_desc.d_len = oplen 1399 | SAFE_PE_LEN_READY 1400 | (bypass << SAFE_PE_LEN_BYPASS_S) 1401 ; 1402 1403 safestats.st_ipackets++; 1404 safestats.st_ibytes += oplen; 1405 1406 if (++(sc->sc_front) == sc->sc_ringtop) 1407 sc->sc_front = sc->sc_ring; 1408 1409 /* XXX honor batching */ 1410 safe_feed(sc, re); 1411 mtx_unlock(&sc->sc_ringmtx); 1412 return (0); 1413 1414 errout: 1415 if ((re->re_dst_m != NULL) && (re->re_src_m != re->re_dst_m)) 1416 m_freem(re->re_dst_m); 1417 1418 if (re->re_dst_map != NULL && re->re_dst_map != re->re_src_map) { 1419 bus_dmamap_unload(sc->sc_dstdmat, re->re_dst_map); 1420 bus_dmamap_destroy(sc->sc_dstdmat, re->re_dst_map); 1421 } 1422 if (re->re_src_map != NULL) { 1423 bus_dmamap_unload(sc->sc_srcdmat, re->re_src_map); 1424 bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map); 1425 } 1426 mtx_unlock(&sc->sc_ringmtx); 1427 if (err != ERESTART) { 1428 crp->crp_etype = err; 1429 crypto_done(crp); 1430 } else { 1431 sc->sc_needwakeup |= CRYPTO_SYMQ; 1432 } 1433 return (err); 1434 } 1435 1436 static void 1437 safe_callback(struct safe_softc *sc, struct safe_ringentry *re) 1438 { 1439 struct cryptop *crp = (struct cryptop *)re->re_crp; 1440 struct safe_session *ses; 1441 struct cryptodesc *crd; 1442 1443 ses = crypto_get_driver_session(crp->crp_session); 1444 1445 safestats.st_opackets++; 1446 safestats.st_obytes += re->re_dst.mapsize; 1447 1448 safe_dma_sync(&sc->sc_ringalloc, 1449 BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE); 1450 if (re->re_desc.d_csr & SAFE_PE_CSR_STATUS) { 1451 device_printf(sc->sc_dev, "csr 0x%x cmd0 0x%x cmd1 0x%x\n", 1452 re->re_desc.d_csr, 1453 re->re_sa.sa_cmd0, re->re_sa.sa_cmd1); 1454 safestats.st_peoperr++; 1455 crp->crp_etype = EIO; /* something more meaningful? */ 1456 } 1457 if (re->re_dst_map != NULL && re->re_dst_map != re->re_src_map) { 1458 bus_dmamap_sync(sc->sc_dstdmat, re->re_dst_map, 1459 BUS_DMASYNC_POSTREAD); 1460 bus_dmamap_unload(sc->sc_dstdmat, re->re_dst_map); 1461 bus_dmamap_destroy(sc->sc_dstdmat, re->re_dst_map); 1462 } 1463 bus_dmamap_sync(sc->sc_srcdmat, re->re_src_map, BUS_DMASYNC_POSTWRITE); 1464 bus_dmamap_unload(sc->sc_srcdmat, re->re_src_map); 1465 bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map); 1466 1467 /* 1468 * If result was written to a differet mbuf chain, swap 1469 * it in as the return value and reclaim the original. 1470 */ 1471 if ((crp->crp_flags & CRYPTO_F_IMBUF) && re->re_src_m != re->re_dst_m) { 1472 m_freem(re->re_src_m); 1473 crp->crp_buf = (caddr_t)re->re_dst_m; 1474 } 1475 1476 if (re->re_flags & SAFE_QFLAGS_COPYOUTIV) { 1477 /* copy out IV for future use */ 1478 for (crd = crp->crp_desc; crd; crd = crd->crd_next) { 1479 int ivsize; 1480 1481 if (crd->crd_alg == CRYPTO_DES_CBC || 1482 crd->crd_alg == CRYPTO_3DES_CBC) { 1483 ivsize = 2*sizeof(u_int32_t); 1484 } else if (crd->crd_alg == CRYPTO_AES_CBC) { 1485 ivsize = 4*sizeof(u_int32_t); 1486 } else 1487 continue; 1488 crypto_copydata(crp->crp_flags, crp->crp_buf, 1489 crd->crd_skip + crd->crd_len - ivsize, ivsize, 1490 (caddr_t)ses->ses_iv); 1491 break; 1492 } 1493 } 1494 1495 if (re->re_flags & SAFE_QFLAGS_COPYOUTICV) { 1496 /* copy out ICV result */ 1497 for (crd = crp->crp_desc; crd; crd = crd->crd_next) { 1498 if (!(crd->crd_alg == CRYPTO_MD5_HMAC || 1499 crd->crd_alg == CRYPTO_SHA1_HMAC || 1500 crd->crd_alg == CRYPTO_NULL_HMAC)) 1501 continue; 1502 if (crd->crd_alg == CRYPTO_SHA1_HMAC) { 1503 /* 1504 * SHA-1 ICV's are byte-swapped; fix 'em up 1505 * before copy them to their destination. 1506 */ 1507 re->re_sastate.sa_saved_indigest[0] = 1508 bswap32(re->re_sastate.sa_saved_indigest[0]); 1509 re->re_sastate.sa_saved_indigest[1] = 1510 bswap32(re->re_sastate.sa_saved_indigest[1]); 1511 re->re_sastate.sa_saved_indigest[2] = 1512 bswap32(re->re_sastate.sa_saved_indigest[2]); 1513 } 1514 crypto_copyback(crp->crp_flags, crp->crp_buf, 1515 crd->crd_inject, ses->ses_mlen, 1516 (caddr_t)re->re_sastate.sa_saved_indigest); 1517 break; 1518 } 1519 } 1520 crypto_done(crp); 1521 } 1522 1523 /* 1524 * Copy all data past offset from srcm to dstm. 1525 */ 1526 static void 1527 safe_mcopy(struct mbuf *srcm, struct mbuf *dstm, u_int offset) 1528 { 1529 u_int j, dlen, slen; 1530 caddr_t dptr, sptr; 1531 1532 /* 1533 * Advance src and dst to offset. 1534 */ 1535 j = offset; 1536 while (j >= srcm->m_len) { 1537 j -= srcm->m_len; 1538 srcm = srcm->m_next; 1539 if (srcm == NULL) 1540 return; 1541 } 1542 sptr = mtod(srcm, caddr_t) + j; 1543 slen = srcm->m_len - j; 1544 1545 j = offset; 1546 while (j >= dstm->m_len) { 1547 j -= dstm->m_len; 1548 dstm = dstm->m_next; 1549 if (dstm == NULL) 1550 return; 1551 } 1552 dptr = mtod(dstm, caddr_t) + j; 1553 dlen = dstm->m_len - j; 1554 1555 /* 1556 * Copy everything that remains. 1557 */ 1558 for (;;) { 1559 j = min(slen, dlen); 1560 bcopy(sptr, dptr, j); 1561 if (slen == j) { 1562 srcm = srcm->m_next; 1563 if (srcm == NULL) 1564 return; 1565 sptr = srcm->m_data; 1566 slen = srcm->m_len; 1567 } else 1568 sptr += j, slen -= j; 1569 if (dlen == j) { 1570 dstm = dstm->m_next; 1571 if (dstm == NULL) 1572 return; 1573 dptr = dstm->m_data; 1574 dlen = dstm->m_len; 1575 } else 1576 dptr += j, dlen -= j; 1577 } 1578 } 1579 1580 #ifndef SAFE_NO_RNG 1581 #define SAFE_RNG_MAXWAIT 1000 1582 1583 static void 1584 safe_rng_init(struct safe_softc *sc) 1585 { 1586 u_int32_t w, v; 1587 int i; 1588 1589 WRITE_REG(sc, SAFE_RNG_CTRL, 0); 1590 /* use default value according to the manual */ 1591 WRITE_REG(sc, SAFE_RNG_CNFG, 0x834); /* magic from SafeNet */ 1592 WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0); 1593 1594 /* 1595 * There is a bug in rev 1.0 of the 1140 that when the RNG 1596 * is brought out of reset the ready status flag does not 1597 * work until the RNG has finished its internal initialization. 1598 * 1599 * So in order to determine the device is through its 1600 * initialization we must read the data register, using the 1601 * status reg in the read in case it is initialized. Then read 1602 * the data register until it changes from the first read. 1603 * Once it changes read the data register until it changes 1604 * again. At this time the RNG is considered initialized. 1605 * This could take between 750ms - 1000ms in time. 1606 */ 1607 i = 0; 1608 w = READ_REG(sc, SAFE_RNG_OUT); 1609 do { 1610 v = READ_REG(sc, SAFE_RNG_OUT); 1611 if (v != w) { 1612 w = v; 1613 break; 1614 } 1615 DELAY(10); 1616 } while (++i < SAFE_RNG_MAXWAIT); 1617 1618 /* Wait Until data changes again */ 1619 i = 0; 1620 do { 1621 v = READ_REG(sc, SAFE_RNG_OUT); 1622 if (v != w) 1623 break; 1624 DELAY(10); 1625 } while (++i < SAFE_RNG_MAXWAIT); 1626 } 1627 1628 static __inline void 1629 safe_rng_disable_short_cycle(struct safe_softc *sc) 1630 { 1631 WRITE_REG(sc, SAFE_RNG_CTRL, 1632 READ_REG(sc, SAFE_RNG_CTRL) &~ SAFE_RNG_CTRL_SHORTEN); 1633 } 1634 1635 static __inline void 1636 safe_rng_enable_short_cycle(struct safe_softc *sc) 1637 { 1638 WRITE_REG(sc, SAFE_RNG_CTRL, 1639 READ_REG(sc, SAFE_RNG_CTRL) | SAFE_RNG_CTRL_SHORTEN); 1640 } 1641 1642 static __inline u_int32_t 1643 safe_rng_read(struct safe_softc *sc) 1644 { 1645 int i; 1646 1647 i = 0; 1648 while (READ_REG(sc, SAFE_RNG_STAT) != 0 && ++i < SAFE_RNG_MAXWAIT) 1649 ; 1650 return READ_REG(sc, SAFE_RNG_OUT); 1651 } 1652 1653 static void 1654 safe_rng(void *arg) 1655 { 1656 struct safe_softc *sc = arg; 1657 u_int32_t buf[SAFE_RNG_MAXBUFSIZ]; /* NB: maybe move to softc */ 1658 u_int maxwords; 1659 int i; 1660 1661 safestats.st_rng++; 1662 /* 1663 * Fetch the next block of data. 1664 */ 1665 maxwords = safe_rngbufsize; 1666 if (maxwords > SAFE_RNG_MAXBUFSIZ) 1667 maxwords = SAFE_RNG_MAXBUFSIZ; 1668 retry: 1669 for (i = 0; i < maxwords; i++) 1670 buf[i] = safe_rng_read(sc); 1671 /* 1672 * Check the comparator alarm count and reset the h/w if 1673 * it exceeds our threshold. This guards against the 1674 * hardware oscillators resonating with external signals. 1675 */ 1676 if (READ_REG(sc, SAFE_RNG_ALM_CNT) > safe_rngmaxalarm) { 1677 u_int32_t freq_inc, w; 1678 1679 DPRINTF(("%s: alarm count %u exceeds threshold %u\n", __func__, 1680 READ_REG(sc, SAFE_RNG_ALM_CNT), safe_rngmaxalarm)); 1681 safestats.st_rngalarm++; 1682 safe_rng_enable_short_cycle(sc); 1683 freq_inc = 18; 1684 for (i = 0; i < 64; i++) { 1685 w = READ_REG(sc, SAFE_RNG_CNFG); 1686 freq_inc = ((w + freq_inc) & 0x3fL); 1687 w = ((w & ~0x3fL) | freq_inc); 1688 WRITE_REG(sc, SAFE_RNG_CNFG, w); 1689 1690 WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0); 1691 1692 (void) safe_rng_read(sc); 1693 DELAY(25); 1694 1695 if (READ_REG(sc, SAFE_RNG_ALM_CNT) == 0) { 1696 safe_rng_disable_short_cycle(sc); 1697 goto retry; 1698 } 1699 freq_inc = 1; 1700 } 1701 safe_rng_disable_short_cycle(sc); 1702 } else 1703 WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0); 1704 1705 (*sc->sc_harvest)(sc->sc_rndtest, buf, maxwords*sizeof (u_int32_t)); 1706 callout_reset(&sc->sc_rngto, 1707 hz * (safe_rnginterval ? safe_rnginterval : 1), safe_rng, sc); 1708 } 1709 #endif /* SAFE_NO_RNG */ 1710 1711 static void 1712 safe_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error) 1713 { 1714 bus_addr_t *paddr = (bus_addr_t*) arg; 1715 *paddr = segs->ds_addr; 1716 } 1717 1718 static int 1719 safe_dma_malloc( 1720 struct safe_softc *sc, 1721 bus_size_t size, 1722 struct safe_dma_alloc *dma, 1723 int mapflags 1724 ) 1725 { 1726 int r; 1727 1728 r = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev), /* parent */ 1729 sizeof(u_int32_t), 0, /* alignment, bounds */ 1730 BUS_SPACE_MAXADDR_32BIT, /* lowaddr */ 1731 BUS_SPACE_MAXADDR, /* highaddr */ 1732 NULL, NULL, /* filter, filterarg */ 1733 size, /* maxsize */ 1734 1, /* nsegments */ 1735 size, /* maxsegsize */ 1736 BUS_DMA_ALLOCNOW, /* flags */ 1737 NULL, NULL, /* locking */ 1738 &dma->dma_tag); 1739 if (r != 0) { 1740 device_printf(sc->sc_dev, "safe_dma_malloc: " 1741 "bus_dma_tag_create failed; error %u\n", r); 1742 goto fail_0; 1743 } 1744 1745 r = bus_dmamem_alloc(dma->dma_tag, (void**) &dma->dma_vaddr, 1746 BUS_DMA_NOWAIT, &dma->dma_map); 1747 if (r != 0) { 1748 device_printf(sc->sc_dev, "safe_dma_malloc: " 1749 "bus_dmammem_alloc failed; size %ju, error %u\n", 1750 (uintmax_t)size, r); 1751 goto fail_1; 1752 } 1753 1754 r = bus_dmamap_load(dma->dma_tag, dma->dma_map, dma->dma_vaddr, 1755 size, 1756 safe_dmamap_cb, 1757 &dma->dma_paddr, 1758 mapflags | BUS_DMA_NOWAIT); 1759 if (r != 0) { 1760 device_printf(sc->sc_dev, "safe_dma_malloc: " 1761 "bus_dmamap_load failed; error %u\n", r); 1762 goto fail_2; 1763 } 1764 1765 dma->dma_size = size; 1766 return (0); 1767 1768 bus_dmamap_unload(dma->dma_tag, dma->dma_map); 1769 fail_2: 1770 bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map); 1771 fail_1: 1772 bus_dma_tag_destroy(dma->dma_tag); 1773 fail_0: 1774 dma->dma_tag = NULL; 1775 return (r); 1776 } 1777 1778 static void 1779 safe_dma_free(struct safe_softc *sc, struct safe_dma_alloc *dma) 1780 { 1781 bus_dmamap_unload(dma->dma_tag, dma->dma_map); 1782 bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map); 1783 bus_dma_tag_destroy(dma->dma_tag); 1784 } 1785 1786 /* 1787 * Resets the board. Values in the regesters are left as is 1788 * from the reset (i.e. initial values are assigned elsewhere). 1789 */ 1790 static void 1791 safe_reset_board(struct safe_softc *sc) 1792 { 1793 u_int32_t v; 1794 /* 1795 * Reset the device. The manual says no delay 1796 * is needed between marking and clearing reset. 1797 */ 1798 v = READ_REG(sc, SAFE_PE_DMACFG) &~ 1799 (SAFE_PE_DMACFG_PERESET | SAFE_PE_DMACFG_PDRRESET | 1800 SAFE_PE_DMACFG_SGRESET); 1801 WRITE_REG(sc, SAFE_PE_DMACFG, v 1802 | SAFE_PE_DMACFG_PERESET 1803 | SAFE_PE_DMACFG_PDRRESET 1804 | SAFE_PE_DMACFG_SGRESET); 1805 WRITE_REG(sc, SAFE_PE_DMACFG, v); 1806 } 1807 1808 /* 1809 * Initialize registers we need to touch only once. 1810 */ 1811 static void 1812 safe_init_board(struct safe_softc *sc) 1813 { 1814 u_int32_t v, dwords; 1815 1816 v = READ_REG(sc, SAFE_PE_DMACFG); 1817 v &=~ SAFE_PE_DMACFG_PEMODE; 1818 v |= SAFE_PE_DMACFG_FSENA /* failsafe enable */ 1819 | SAFE_PE_DMACFG_GPRPCI /* gather ring on PCI */ 1820 | SAFE_PE_DMACFG_SPRPCI /* scatter ring on PCI */ 1821 | SAFE_PE_DMACFG_ESDESC /* endian-swap descriptors */ 1822 | SAFE_PE_DMACFG_ESSA /* endian-swap SA's */ 1823 | SAFE_PE_DMACFG_ESPDESC /* endian-swap part. desc's */ 1824 ; 1825 WRITE_REG(sc, SAFE_PE_DMACFG, v); 1826 #if 0 1827 /* XXX select byte swap based on host byte order */ 1828 WRITE_REG(sc, SAFE_ENDIAN, 0x1b); 1829 #endif 1830 if (sc->sc_chiprev == SAFE_REV(1,0)) { 1831 /* 1832 * Avoid large PCI DMA transfers. Rev 1.0 has a bug where 1833 * "target mode transfers" done while the chip is DMA'ing 1834 * >1020 bytes cause the hardware to lockup. To avoid this 1835 * we reduce the max PCI transfer size and use small source 1836 * particle descriptors (<= 256 bytes). 1837 */ 1838 WRITE_REG(sc, SAFE_DMA_CFG, 256); 1839 device_printf(sc->sc_dev, 1840 "Reduce max DMA size to %u words for rev %u.%u WAR\n", 1841 (READ_REG(sc, SAFE_DMA_CFG)>>2) & 0xff, 1842 SAFE_REV_MAJ(sc->sc_chiprev), 1843 SAFE_REV_MIN(sc->sc_chiprev)); 1844 } 1845 1846 /* NB: operands+results are overlaid */ 1847 WRITE_REG(sc, SAFE_PE_PDRBASE, sc->sc_ringalloc.dma_paddr); 1848 WRITE_REG(sc, SAFE_PE_RDRBASE, sc->sc_ringalloc.dma_paddr); 1849 /* 1850 * Configure ring entry size and number of items in the ring. 1851 */ 1852 KASSERT((sizeof(struct safe_ringentry) % sizeof(u_int32_t)) == 0, 1853 ("PE ring entry not 32-bit aligned!")); 1854 dwords = sizeof(struct safe_ringentry) / sizeof(u_int32_t); 1855 WRITE_REG(sc, SAFE_PE_RINGCFG, 1856 (dwords << SAFE_PE_RINGCFG_OFFSET_S) | SAFE_MAX_NQUEUE); 1857 WRITE_REG(sc, SAFE_PE_RINGPOLL, 0); /* disable polling */ 1858 1859 WRITE_REG(sc, SAFE_PE_GRNGBASE, sc->sc_spalloc.dma_paddr); 1860 WRITE_REG(sc, SAFE_PE_SRNGBASE, sc->sc_dpalloc.dma_paddr); 1861 WRITE_REG(sc, SAFE_PE_PARTSIZE, 1862 (SAFE_TOTAL_DPART<<16) | SAFE_TOTAL_SPART); 1863 /* 1864 * NB: destination particles are fixed size. We use 1865 * an mbuf cluster and require all results go to 1866 * clusters or smaller. 1867 */ 1868 WRITE_REG(sc, SAFE_PE_PARTCFG, SAFE_MAX_DSIZE); 1869 1870 /* it's now safe to enable PE mode, do it */ 1871 WRITE_REG(sc, SAFE_PE_DMACFG, v | SAFE_PE_DMACFG_PEMODE); 1872 1873 /* 1874 * Configure hardware to use level-triggered interrupts and 1875 * to interrupt after each descriptor is processed. 1876 */ 1877 WRITE_REG(sc, SAFE_HI_CFG, SAFE_HI_CFG_LEVEL); 1878 WRITE_REG(sc, SAFE_HI_DESC_CNT, 1); 1879 WRITE_REG(sc, SAFE_HI_MASK, SAFE_INT_PE_DDONE | SAFE_INT_PE_ERROR); 1880 } 1881 1882 /* 1883 * Init PCI registers 1884 */ 1885 static void 1886 safe_init_pciregs(device_t dev) 1887 { 1888 } 1889 1890 /* 1891 * Clean up after a chip crash. 1892 * It is assumed that the caller in splimp() 1893 */ 1894 static void 1895 safe_cleanchip(struct safe_softc *sc) 1896 { 1897 1898 if (sc->sc_nqchip != 0) { 1899 struct safe_ringentry *re = sc->sc_back; 1900 1901 while (re != sc->sc_front) { 1902 if (re->re_desc.d_csr != 0) 1903 safe_free_entry(sc, re); 1904 if (++re == sc->sc_ringtop) 1905 re = sc->sc_ring; 1906 } 1907 sc->sc_back = re; 1908 sc->sc_nqchip = 0; 1909 } 1910 } 1911 1912 /* 1913 * free a safe_q 1914 * It is assumed that the caller is within splimp(). 1915 */ 1916 static int 1917 safe_free_entry(struct safe_softc *sc, struct safe_ringentry *re) 1918 { 1919 struct cryptop *crp; 1920 1921 /* 1922 * Free header MCR 1923 */ 1924 if ((re->re_dst_m != NULL) && (re->re_src_m != re->re_dst_m)) 1925 m_freem(re->re_dst_m); 1926 1927 crp = (struct cryptop *)re->re_crp; 1928 1929 re->re_desc.d_csr = 0; 1930 1931 crp->crp_etype = EFAULT; 1932 crypto_done(crp); 1933 return(0); 1934 } 1935 1936 /* 1937 * Routine to reset the chip and clean up. 1938 * It is assumed that the caller is in splimp() 1939 */ 1940 static void 1941 safe_totalreset(struct safe_softc *sc) 1942 { 1943 safe_reset_board(sc); 1944 safe_init_board(sc); 1945 safe_cleanchip(sc); 1946 } 1947 1948 /* 1949 * Is the operand suitable aligned for direct DMA. Each 1950 * segment must be aligned on a 32-bit boundary and all 1951 * but the last segment must be a multiple of 4 bytes. 1952 */ 1953 static int 1954 safe_dmamap_aligned(const struct safe_operand *op) 1955 { 1956 int i; 1957 1958 for (i = 0; i < op->nsegs; i++) { 1959 if (op->segs[i].ds_addr & 3) 1960 return (0); 1961 if (i != (op->nsegs - 1) && (op->segs[i].ds_len & 3)) 1962 return (0); 1963 } 1964 return (1); 1965 } 1966 1967 /* 1968 * Is the operand suitable for direct DMA as the destination 1969 * of an operation. The hardware requires that each ``particle'' 1970 * but the last in an operation result have the same size. We 1971 * fix that size at SAFE_MAX_DSIZE bytes. This routine returns 1972 * 0 if some segment is not a multiple of of this size, 1 if all 1973 * segments are exactly this size, or 2 if segments are at worst 1974 * a multple of this size. 1975 */ 1976 static int 1977 safe_dmamap_uniform(const struct safe_operand *op) 1978 { 1979 int result = 1; 1980 1981 if (op->nsegs > 0) { 1982 int i; 1983 1984 for (i = 0; i < op->nsegs-1; i++) { 1985 if (op->segs[i].ds_len % SAFE_MAX_DSIZE) 1986 return (0); 1987 if (op->segs[i].ds_len != SAFE_MAX_DSIZE) 1988 result = 2; 1989 } 1990 } 1991 return (result); 1992 } 1993 1994 #ifdef SAFE_DEBUG 1995 static void 1996 safe_dump_dmastatus(struct safe_softc *sc, const char *tag) 1997 { 1998 printf("%s: ENDIAN 0x%x SRC 0x%x DST 0x%x STAT 0x%x\n" 1999 , tag 2000 , READ_REG(sc, SAFE_DMA_ENDIAN) 2001 , READ_REG(sc, SAFE_DMA_SRCADDR) 2002 , READ_REG(sc, SAFE_DMA_DSTADDR) 2003 , READ_REG(sc, SAFE_DMA_STAT) 2004 ); 2005 } 2006 2007 static void 2008 safe_dump_intrstate(struct safe_softc *sc, const char *tag) 2009 { 2010 printf("%s: HI_CFG 0x%x HI_MASK 0x%x HI_DESC_CNT 0x%x HU_STAT 0x%x HM_STAT 0x%x\n" 2011 , tag 2012 , READ_REG(sc, SAFE_HI_CFG) 2013 , READ_REG(sc, SAFE_HI_MASK) 2014 , READ_REG(sc, SAFE_HI_DESC_CNT) 2015 , READ_REG(sc, SAFE_HU_STAT) 2016 , READ_REG(sc, SAFE_HM_STAT) 2017 ); 2018 } 2019 2020 static void 2021 safe_dump_ringstate(struct safe_softc *sc, const char *tag) 2022 { 2023 u_int32_t estat = READ_REG(sc, SAFE_PE_ERNGSTAT); 2024 2025 /* NB: assume caller has lock on ring */ 2026 printf("%s: ERNGSTAT %x (next %u) back %lu front %lu\n", 2027 tag, 2028 estat, (estat >> SAFE_PE_ERNGSTAT_NEXT_S), 2029 (unsigned long)(sc->sc_back - sc->sc_ring), 2030 (unsigned long)(sc->sc_front - sc->sc_ring)); 2031 } 2032 2033 static void 2034 safe_dump_request(struct safe_softc *sc, const char* tag, struct safe_ringentry *re) 2035 { 2036 int ix, nsegs; 2037 2038 ix = re - sc->sc_ring; 2039 printf("%s: %p (%u): csr %x src %x dst %x sa %x len %x\n" 2040 , tag 2041 , re, ix 2042 , re->re_desc.d_csr 2043 , re->re_desc.d_src 2044 , re->re_desc.d_dst 2045 , re->re_desc.d_sa 2046 , re->re_desc.d_len 2047 ); 2048 if (re->re_src.nsegs > 1) { 2049 ix = (re->re_desc.d_src - sc->sc_spalloc.dma_paddr) / 2050 sizeof(struct safe_pdesc); 2051 for (nsegs = re->re_src.nsegs; nsegs; nsegs--) { 2052 printf(" spd[%u] %p: %p size %u flags %x" 2053 , ix, &sc->sc_spring[ix] 2054 , (caddr_t)(uintptr_t) sc->sc_spring[ix].pd_addr 2055 , sc->sc_spring[ix].pd_size 2056 , sc->sc_spring[ix].pd_flags 2057 ); 2058 if (sc->sc_spring[ix].pd_size == 0) 2059 printf(" (zero!)"); 2060 printf("\n"); 2061 if (++ix == SAFE_TOTAL_SPART) 2062 ix = 0; 2063 } 2064 } 2065 if (re->re_dst.nsegs > 1) { 2066 ix = (re->re_desc.d_dst - sc->sc_dpalloc.dma_paddr) / 2067 sizeof(struct safe_pdesc); 2068 for (nsegs = re->re_dst.nsegs; nsegs; nsegs--) { 2069 printf(" dpd[%u] %p: %p flags %x\n" 2070 , ix, &sc->sc_dpring[ix] 2071 , (caddr_t)(uintptr_t) sc->sc_dpring[ix].pd_addr 2072 , sc->sc_dpring[ix].pd_flags 2073 ); 2074 if (++ix == SAFE_TOTAL_DPART) 2075 ix = 0; 2076 } 2077 } 2078 printf("sa: cmd0 %08x cmd1 %08x staterec %x\n", 2079 re->re_sa.sa_cmd0, re->re_sa.sa_cmd1, re->re_sa.sa_staterec); 2080 printf("sa: key %x %x %x %x %x %x %x %x\n" 2081 , re->re_sa.sa_key[0] 2082 , re->re_sa.sa_key[1] 2083 , re->re_sa.sa_key[2] 2084 , re->re_sa.sa_key[3] 2085 , re->re_sa.sa_key[4] 2086 , re->re_sa.sa_key[5] 2087 , re->re_sa.sa_key[6] 2088 , re->re_sa.sa_key[7] 2089 ); 2090 printf("sa: indigest %x %x %x %x %x\n" 2091 , re->re_sa.sa_indigest[0] 2092 , re->re_sa.sa_indigest[1] 2093 , re->re_sa.sa_indigest[2] 2094 , re->re_sa.sa_indigest[3] 2095 , re->re_sa.sa_indigest[4] 2096 ); 2097 printf("sa: outdigest %x %x %x %x %x\n" 2098 , re->re_sa.sa_outdigest[0] 2099 , re->re_sa.sa_outdigest[1] 2100 , re->re_sa.sa_outdigest[2] 2101 , re->re_sa.sa_outdigest[3] 2102 , re->re_sa.sa_outdigest[4] 2103 ); 2104 printf("sr: iv %x %x %x %x\n" 2105 , re->re_sastate.sa_saved_iv[0] 2106 , re->re_sastate.sa_saved_iv[1] 2107 , re->re_sastate.sa_saved_iv[2] 2108 , re->re_sastate.sa_saved_iv[3] 2109 ); 2110 printf("sr: hashbc %u indigest %x %x %x %x %x\n" 2111 , re->re_sastate.sa_saved_hashbc 2112 , re->re_sastate.sa_saved_indigest[0] 2113 , re->re_sastate.sa_saved_indigest[1] 2114 , re->re_sastate.sa_saved_indigest[2] 2115 , re->re_sastate.sa_saved_indigest[3] 2116 , re->re_sastate.sa_saved_indigest[4] 2117 ); 2118 } 2119 2120 static void 2121 safe_dump_ring(struct safe_softc *sc, const char *tag) 2122 { 2123 mtx_lock(&sc->sc_ringmtx); 2124 printf("\nSafeNet Ring State:\n"); 2125 safe_dump_intrstate(sc, tag); 2126 safe_dump_dmastatus(sc, tag); 2127 safe_dump_ringstate(sc, tag); 2128 if (sc->sc_nqchip) { 2129 struct safe_ringentry *re = sc->sc_back; 2130 do { 2131 safe_dump_request(sc, tag, re); 2132 if (++re == sc->sc_ringtop) 2133 re = sc->sc_ring; 2134 } while (re != sc->sc_front); 2135 } 2136 mtx_unlock(&sc->sc_ringmtx); 2137 } 2138 2139 static int 2140 sysctl_hw_safe_dump(SYSCTL_HANDLER_ARGS) 2141 { 2142 char dmode[64]; 2143 int error; 2144 2145 strncpy(dmode, "", sizeof(dmode) - 1); 2146 dmode[sizeof(dmode) - 1] = '\0'; 2147 error = sysctl_handle_string(oidp, &dmode[0], sizeof(dmode), req); 2148 2149 if (error == 0 && req->newptr != NULL) { 2150 struct safe_softc *sc = safec; 2151 2152 if (!sc) 2153 return EINVAL; 2154 if (strncmp(dmode, "dma", 3) == 0) 2155 safe_dump_dmastatus(sc, "safe0"); 2156 else if (strncmp(dmode, "int", 3) == 0) 2157 safe_dump_intrstate(sc, "safe0"); 2158 else if (strncmp(dmode, "ring", 4) == 0) 2159 safe_dump_ring(sc, "safe0"); 2160 else 2161 return EINVAL; 2162 } 2163 return error; 2164 } 2165 SYSCTL_PROC(_hw_safe, OID_AUTO, dump, 2166 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, 0, 2167 sysctl_hw_safe_dump, "A", 2168 "Dump driver state"); 2169 #endif /* SAFE_DEBUG */ 2170