1 /*- 2 * Copyright (c) 2003 Sam Leffler, Errno Consulting 3 * Copyright (c) 2003 Global Technology Associates, Inc. 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28 #include <sys/cdefs.h> 29 __FBSDID("$FreeBSD$"); 30 31 /* 32 * SafeNet SafeXcel-1141 hardware crypto accelerator 33 */ 34 #include "opt_safe.h" 35 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/proc.h> 39 #include <sys/errno.h> 40 #include <sys/malloc.h> 41 #include <sys/kernel.h> 42 #include <sys/mbuf.h> 43 #include <sys/module.h> 44 #include <sys/lock.h> 45 #include <sys/mutex.h> 46 #include <sys/sysctl.h> 47 #include <sys/endian.h> 48 49 #include <vm/vm.h> 50 #include <vm/pmap.h> 51 52 #include <machine/bus.h> 53 #include <machine/resource.h> 54 #include <sys/bus.h> 55 #include <sys/rman.h> 56 57 #include <crypto/sha1.h> 58 #include <opencrypto/cryptodev.h> 59 #include <opencrypto/cryptosoft.h> 60 #include <sys/md5.h> 61 #include <sys/random.h> 62 #include <sys/kobj.h> 63 64 #include "cryptodev_if.h" 65 66 #include <dev/pci/pcivar.h> 67 #include <dev/pci/pcireg.h> 68 69 #ifdef SAFE_RNDTEST 70 #include <dev/rndtest/rndtest.h> 71 #endif 72 #include <dev/safe/safereg.h> 73 #include <dev/safe/safevar.h> 74 75 #ifndef bswap32 76 #define bswap32 NTOHL 77 #endif 78 79 /* 80 * Prototypes and count for the pci_device structure 81 */ 82 static int safe_probe(device_t); 83 static int safe_attach(device_t); 84 static int safe_detach(device_t); 85 static int safe_suspend(device_t); 86 static int safe_resume(device_t); 87 static int safe_shutdown(device_t); 88 89 static int safe_newsession(device_t, u_int32_t *, struct cryptoini *); 90 static int safe_freesession(device_t, u_int64_t); 91 static int safe_process(device_t, struct cryptop *, int); 92 93 static device_method_t safe_methods[] = { 94 /* Device interface */ 95 DEVMETHOD(device_probe, safe_probe), 96 DEVMETHOD(device_attach, safe_attach), 97 DEVMETHOD(device_detach, safe_detach), 98 DEVMETHOD(device_suspend, safe_suspend), 99 DEVMETHOD(device_resume, safe_resume), 100 DEVMETHOD(device_shutdown, safe_shutdown), 101 102 /* crypto device methods */ 103 DEVMETHOD(cryptodev_newsession, safe_newsession), 104 DEVMETHOD(cryptodev_freesession,safe_freesession), 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, 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 random_harvest(buf, count, count*NBBY/2, RANDOM_PURE_SAFE); 215 } 216 #endif /* SAFE_NO_RNG */ 217 218 static int 219 safe_attach(device_t dev) 220 { 221 struct safe_softc *sc = device_get_softc(dev); 222 u_int32_t raddr; 223 u_int32_t i, devinfo; 224 int rid; 225 226 bzero(sc, sizeof (*sc)); 227 sc->sc_dev = dev; 228 229 /* XXX handle power management */ 230 231 pci_enable_busmaster(dev); 232 233 /* 234 * Setup memory-mapping of PCI registers. 235 */ 236 rid = BS_BAR; 237 sc->sc_sr = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, 238 RF_ACTIVE); 239 if (sc->sc_sr == NULL) { 240 device_printf(dev, "cannot map register space\n"); 241 goto bad; 242 } 243 sc->sc_st = rman_get_bustag(sc->sc_sr); 244 sc->sc_sh = rman_get_bushandle(sc->sc_sr); 245 246 /* 247 * Arrange interrupt line. 248 */ 249 rid = 0; 250 sc->sc_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid, 251 RF_SHAREABLE|RF_ACTIVE); 252 if (sc->sc_irq == NULL) { 253 device_printf(dev, "could not map interrupt\n"); 254 goto bad1; 255 } 256 /* 257 * NB: Network code assumes we are blocked with splimp() 258 * so make sure the IRQ is mapped appropriately. 259 */ 260 if (bus_setup_intr(dev, sc->sc_irq, INTR_TYPE_NET | INTR_MPSAFE, 261 NULL, safe_intr, sc, &sc->sc_ih)) { 262 device_printf(dev, "could not establish interrupt\n"); 263 goto bad2; 264 } 265 266 sc->sc_cid = crypto_get_driverid(dev, CRYPTOCAP_F_HARDWARE); 267 if (sc->sc_cid < 0) { 268 device_printf(dev, "could not get crypto driver id\n"); 269 goto bad3; 270 } 271 272 sc->sc_chiprev = READ_REG(sc, SAFE_DEVINFO) & 273 (SAFE_DEVINFO_REV_MAJ | SAFE_DEVINFO_REV_MIN); 274 275 /* 276 * Setup DMA descriptor area. 277 */ 278 if (bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */ 279 1, /* alignment */ 280 SAFE_DMA_BOUNDARY, /* boundary */ 281 BUS_SPACE_MAXADDR_32BIT, /* lowaddr */ 282 BUS_SPACE_MAXADDR, /* highaddr */ 283 NULL, NULL, /* filter, filterarg */ 284 SAFE_MAX_DMA, /* maxsize */ 285 SAFE_MAX_PART, /* nsegments */ 286 SAFE_MAX_SSIZE, /* maxsegsize */ 287 BUS_DMA_ALLOCNOW, /* flags */ 288 NULL, NULL, /* locking */ 289 &sc->sc_srcdmat)) { 290 device_printf(dev, "cannot allocate DMA tag\n"); 291 goto bad4; 292 } 293 if (bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */ 294 1, /* alignment */ 295 SAFE_MAX_DSIZE, /* boundary */ 296 BUS_SPACE_MAXADDR_32BIT, /* lowaddr */ 297 BUS_SPACE_MAXADDR, /* highaddr */ 298 NULL, NULL, /* filter, filterarg */ 299 SAFE_MAX_DMA, /* maxsize */ 300 SAFE_MAX_PART, /* nsegments */ 301 SAFE_MAX_DSIZE, /* maxsegsize */ 302 BUS_DMA_ALLOCNOW, /* flags */ 303 NULL, NULL, /* locking */ 304 &sc->sc_dstdmat)) { 305 device_printf(dev, "cannot allocate DMA tag\n"); 306 goto bad4; 307 } 308 309 /* 310 * Allocate packet engine descriptors. 311 */ 312 if (safe_dma_malloc(sc, 313 SAFE_MAX_NQUEUE * sizeof (struct safe_ringentry), 314 &sc->sc_ringalloc, 0)) { 315 device_printf(dev, "cannot allocate PE descriptor ring\n"); 316 bus_dma_tag_destroy(sc->sc_srcdmat); 317 goto bad4; 318 } 319 /* 320 * Hookup the static portion of all our data structures. 321 */ 322 sc->sc_ring = (struct safe_ringentry *) sc->sc_ringalloc.dma_vaddr; 323 sc->sc_ringtop = sc->sc_ring + SAFE_MAX_NQUEUE; 324 sc->sc_front = sc->sc_ring; 325 sc->sc_back = sc->sc_ring; 326 raddr = sc->sc_ringalloc.dma_paddr; 327 bzero(sc->sc_ring, SAFE_MAX_NQUEUE * sizeof(struct safe_ringentry)); 328 for (i = 0; i < SAFE_MAX_NQUEUE; i++) { 329 struct safe_ringentry *re = &sc->sc_ring[i]; 330 331 re->re_desc.d_sa = raddr + 332 offsetof(struct safe_ringentry, re_sa); 333 re->re_sa.sa_staterec = raddr + 334 offsetof(struct safe_ringentry, re_sastate); 335 336 raddr += sizeof (struct safe_ringentry); 337 } 338 mtx_init(&sc->sc_ringmtx, device_get_nameunit(dev), 339 "packet engine ring", MTX_DEF); 340 341 /* 342 * Allocate scatter and gather particle descriptors. 343 */ 344 if (safe_dma_malloc(sc, SAFE_TOTAL_SPART * sizeof (struct safe_pdesc), 345 &sc->sc_spalloc, 0)) { 346 device_printf(dev, "cannot allocate source particle " 347 "descriptor ring\n"); 348 mtx_destroy(&sc->sc_ringmtx); 349 safe_dma_free(sc, &sc->sc_ringalloc); 350 bus_dma_tag_destroy(sc->sc_srcdmat); 351 goto bad4; 352 } 353 sc->sc_spring = (struct safe_pdesc *) sc->sc_spalloc.dma_vaddr; 354 sc->sc_springtop = sc->sc_spring + SAFE_TOTAL_SPART; 355 sc->sc_spfree = sc->sc_spring; 356 bzero(sc->sc_spring, SAFE_TOTAL_SPART * sizeof(struct safe_pdesc)); 357 358 if (safe_dma_malloc(sc, SAFE_TOTAL_DPART * sizeof (struct safe_pdesc), 359 &sc->sc_dpalloc, 0)) { 360 device_printf(dev, "cannot allocate destination particle " 361 "descriptor ring\n"); 362 mtx_destroy(&sc->sc_ringmtx); 363 safe_dma_free(sc, &sc->sc_spalloc); 364 safe_dma_free(sc, &sc->sc_ringalloc); 365 bus_dma_tag_destroy(sc->sc_dstdmat); 366 goto bad4; 367 } 368 sc->sc_dpring = (struct safe_pdesc *) sc->sc_dpalloc.dma_vaddr; 369 sc->sc_dpringtop = sc->sc_dpring + SAFE_TOTAL_DPART; 370 sc->sc_dpfree = sc->sc_dpring; 371 bzero(sc->sc_dpring, SAFE_TOTAL_DPART * sizeof(struct safe_pdesc)); 372 373 device_printf(sc->sc_dev, "%s", safe_partname(sc)); 374 375 devinfo = READ_REG(sc, SAFE_DEVINFO); 376 if (devinfo & SAFE_DEVINFO_RNG) { 377 sc->sc_flags |= SAFE_FLAGS_RNG; 378 printf(" rng"); 379 } 380 if (devinfo & SAFE_DEVINFO_PKEY) { 381 #if 0 382 printf(" key"); 383 sc->sc_flags |= SAFE_FLAGS_KEY; 384 crypto_kregister(sc->sc_cid, CRK_MOD_EXP, 0); 385 crypto_kregister(sc->sc_cid, CRK_MOD_EXP_CRT, 0); 386 #endif 387 } 388 if (devinfo & SAFE_DEVINFO_DES) { 389 printf(" des/3des"); 390 crypto_register(sc->sc_cid, CRYPTO_3DES_CBC, 0, 0); 391 crypto_register(sc->sc_cid, CRYPTO_DES_CBC, 0, 0); 392 } 393 if (devinfo & SAFE_DEVINFO_AES) { 394 printf(" aes"); 395 crypto_register(sc->sc_cid, CRYPTO_AES_CBC, 0, 0); 396 } 397 if (devinfo & SAFE_DEVINFO_MD5) { 398 printf(" md5"); 399 crypto_register(sc->sc_cid, CRYPTO_MD5_HMAC, 0, 0); 400 } 401 if (devinfo & SAFE_DEVINFO_SHA1) { 402 printf(" sha1"); 403 crypto_register(sc->sc_cid, CRYPTO_SHA1_HMAC, 0, 0); 404 } 405 printf(" null"); 406 crypto_register(sc->sc_cid, CRYPTO_NULL_CBC, 0, 0); 407 crypto_register(sc->sc_cid, CRYPTO_NULL_HMAC, 0, 0); 408 /* XXX other supported algorithms */ 409 printf("\n"); 410 411 safe_reset_board(sc); /* reset h/w */ 412 safe_init_pciregs(dev); /* init pci settings */ 413 safe_init_board(sc); /* init h/w */ 414 415 #ifndef SAFE_NO_RNG 416 if (sc->sc_flags & SAFE_FLAGS_RNG) { 417 #ifdef SAFE_RNDTEST 418 sc->sc_rndtest = rndtest_attach(dev); 419 if (sc->sc_rndtest) 420 sc->sc_harvest = rndtest_harvest; 421 else 422 sc->sc_harvest = default_harvest; 423 #else 424 sc->sc_harvest = default_harvest; 425 #endif 426 safe_rng_init(sc); 427 428 callout_init(&sc->sc_rngto, 1); 429 callout_reset(&sc->sc_rngto, hz*safe_rnginterval, safe_rng, sc); 430 } 431 #endif /* SAFE_NO_RNG */ 432 #ifdef SAFE_DEBUG 433 safec = sc; /* for use by hw.safe.dump */ 434 #endif 435 return (0); 436 bad4: 437 crypto_unregister_all(sc->sc_cid); 438 bad3: 439 bus_teardown_intr(dev, sc->sc_irq, sc->sc_ih); 440 bad2: 441 bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq); 442 bad1: 443 bus_release_resource(dev, SYS_RES_MEMORY, BS_BAR, sc->sc_sr); 444 bad: 445 return (ENXIO); 446 } 447 448 /* 449 * Detach a device that successfully probed. 450 */ 451 static int 452 safe_detach(device_t dev) 453 { 454 struct safe_softc *sc = device_get_softc(dev); 455 456 /* XXX wait/abort active ops */ 457 458 WRITE_REG(sc, SAFE_HI_MASK, 0); /* disable interrupts */ 459 460 callout_stop(&sc->sc_rngto); 461 462 crypto_unregister_all(sc->sc_cid); 463 464 #ifdef SAFE_RNDTEST 465 if (sc->sc_rndtest) 466 rndtest_detach(sc->sc_rndtest); 467 #endif 468 469 safe_cleanchip(sc); 470 safe_dma_free(sc, &sc->sc_dpalloc); 471 safe_dma_free(sc, &sc->sc_spalloc); 472 mtx_destroy(&sc->sc_ringmtx); 473 safe_dma_free(sc, &sc->sc_ringalloc); 474 475 bus_generic_detach(dev); 476 bus_teardown_intr(dev, sc->sc_irq, sc->sc_ih); 477 bus_release_resource(dev, SYS_RES_IRQ, 0, sc->sc_irq); 478 479 bus_dma_tag_destroy(sc->sc_srcdmat); 480 bus_dma_tag_destroy(sc->sc_dstdmat); 481 bus_release_resource(dev, SYS_RES_MEMORY, BS_BAR, sc->sc_sr); 482 483 return (0); 484 } 485 486 /* 487 * Stop all chip i/o so that the kernel's probe routines don't 488 * get confused by errant DMAs when rebooting. 489 */ 490 static int 491 safe_shutdown(device_t dev) 492 { 493 #ifdef notyet 494 safe_stop(device_get_softc(dev)); 495 #endif 496 return (0); 497 } 498 499 /* 500 * Device suspend routine. 501 */ 502 static int 503 safe_suspend(device_t dev) 504 { 505 struct safe_softc *sc = device_get_softc(dev); 506 507 #ifdef notyet 508 /* XXX stop the device and save PCI settings */ 509 #endif 510 sc->sc_suspended = 1; 511 512 return (0); 513 } 514 515 static int 516 safe_resume(device_t dev) 517 { 518 struct safe_softc *sc = device_get_softc(dev); 519 520 #ifdef notyet 521 /* XXX retore PCI settings and start the device */ 522 #endif 523 sc->sc_suspended = 0; 524 return (0); 525 } 526 527 /* 528 * SafeXcel Interrupt routine 529 */ 530 static void 531 safe_intr(void *arg) 532 { 533 struct safe_softc *sc = arg; 534 volatile u_int32_t stat; 535 536 stat = READ_REG(sc, SAFE_HM_STAT); 537 if (stat == 0) /* shared irq, not for us */ 538 return; 539 540 WRITE_REG(sc, SAFE_HI_CLR, stat); /* IACK */ 541 542 if ((stat & SAFE_INT_PE_DDONE)) { 543 /* 544 * Descriptor(s) done; scan the ring and 545 * process completed operations. 546 */ 547 mtx_lock(&sc->sc_ringmtx); 548 while (sc->sc_back != sc->sc_front) { 549 struct safe_ringentry *re = sc->sc_back; 550 #ifdef SAFE_DEBUG 551 if (safe_debug) { 552 safe_dump_ringstate(sc, __func__); 553 safe_dump_request(sc, __func__, re); 554 } 555 #endif 556 /* 557 * safe_process marks ring entries that were allocated 558 * but not used with a csr of zero. This insures the 559 * ring front pointer never needs to be set backwards 560 * in the event that an entry is allocated but not used 561 * because of a setup error. 562 */ 563 if (re->re_desc.d_csr != 0) { 564 if (!SAFE_PE_CSR_IS_DONE(re->re_desc.d_csr)) 565 break; 566 if (!SAFE_PE_LEN_IS_DONE(re->re_desc.d_len)) 567 break; 568 sc->sc_nqchip--; 569 safe_callback(sc, re); 570 } 571 if (++(sc->sc_back) == sc->sc_ringtop) 572 sc->sc_back = sc->sc_ring; 573 } 574 mtx_unlock(&sc->sc_ringmtx); 575 } 576 577 /* 578 * Check to see if we got any DMA Error 579 */ 580 if (stat & SAFE_INT_PE_ERROR) { 581 DPRINTF(("dmaerr dmastat %08x\n", 582 READ_REG(sc, SAFE_PE_DMASTAT))); 583 safestats.st_dmaerr++; 584 safe_totalreset(sc); 585 #if 0 586 safe_feed(sc); 587 #endif 588 } 589 590 if (sc->sc_needwakeup) { /* XXX check high watermark */ 591 int wakeup = sc->sc_needwakeup & (CRYPTO_SYMQ|CRYPTO_ASYMQ); 592 DPRINTF(("%s: wakeup crypto %x\n", __func__, 593 sc->sc_needwakeup)); 594 sc->sc_needwakeup &= ~wakeup; 595 crypto_unblock(sc->sc_cid, wakeup); 596 } 597 } 598 599 /* 600 * safe_feed() - post a request to chip 601 */ 602 static void 603 safe_feed(struct safe_softc *sc, struct safe_ringentry *re) 604 { 605 bus_dmamap_sync(sc->sc_srcdmat, re->re_src_map, BUS_DMASYNC_PREWRITE); 606 if (re->re_dst_map != NULL) 607 bus_dmamap_sync(sc->sc_dstdmat, re->re_dst_map, 608 BUS_DMASYNC_PREREAD); 609 /* XXX have no smaller granularity */ 610 safe_dma_sync(&sc->sc_ringalloc, 611 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 612 safe_dma_sync(&sc->sc_spalloc, BUS_DMASYNC_PREWRITE); 613 safe_dma_sync(&sc->sc_dpalloc, BUS_DMASYNC_PREWRITE); 614 615 #ifdef SAFE_DEBUG 616 if (safe_debug) { 617 safe_dump_ringstate(sc, __func__); 618 safe_dump_request(sc, __func__, re); 619 } 620 #endif 621 sc->sc_nqchip++; 622 if (sc->sc_nqchip > safestats.st_maxqchip) 623 safestats.st_maxqchip = sc->sc_nqchip; 624 /* poke h/w to check descriptor ring, any value can be written */ 625 WRITE_REG(sc, SAFE_HI_RD_DESCR, 0); 626 } 627 628 #define N(a) (sizeof(a) / sizeof (a[0])) 629 static void 630 safe_setup_enckey(struct safe_session *ses, caddr_t key) 631 { 632 int i; 633 634 bcopy(key, ses->ses_key, ses->ses_klen / 8); 635 636 /* PE is little-endian, insure proper byte order */ 637 for (i = 0; i < N(ses->ses_key); i++) 638 ses->ses_key[i] = htole32(ses->ses_key[i]); 639 } 640 641 static void 642 safe_setup_mackey(struct safe_session *ses, int algo, caddr_t key, int klen) 643 { 644 MD5_CTX md5ctx; 645 SHA1_CTX sha1ctx; 646 int i; 647 648 649 for (i = 0; i < klen; i++) 650 key[i] ^= HMAC_IPAD_VAL; 651 652 if (algo == CRYPTO_MD5_HMAC) { 653 MD5Init(&md5ctx); 654 MD5Update(&md5ctx, key, klen); 655 MD5Update(&md5ctx, hmac_ipad_buffer, MD5_HMAC_BLOCK_LEN - klen); 656 bcopy(md5ctx.state, ses->ses_hminner, sizeof(md5ctx.state)); 657 } else { 658 SHA1Init(&sha1ctx); 659 SHA1Update(&sha1ctx, key, klen); 660 SHA1Update(&sha1ctx, hmac_ipad_buffer, 661 SHA1_HMAC_BLOCK_LEN - klen); 662 bcopy(sha1ctx.h.b32, ses->ses_hminner, sizeof(sha1ctx.h.b32)); 663 } 664 665 for (i = 0; i < klen; i++) 666 key[i] ^= (HMAC_IPAD_VAL ^ HMAC_OPAD_VAL); 667 668 if (algo == CRYPTO_MD5_HMAC) { 669 MD5Init(&md5ctx); 670 MD5Update(&md5ctx, key, klen); 671 MD5Update(&md5ctx, hmac_opad_buffer, MD5_HMAC_BLOCK_LEN - klen); 672 bcopy(md5ctx.state, ses->ses_hmouter, sizeof(md5ctx.state)); 673 } else { 674 SHA1Init(&sha1ctx); 675 SHA1Update(&sha1ctx, key, klen); 676 SHA1Update(&sha1ctx, hmac_opad_buffer, 677 SHA1_HMAC_BLOCK_LEN - klen); 678 bcopy(sha1ctx.h.b32, ses->ses_hmouter, sizeof(sha1ctx.h.b32)); 679 } 680 681 for (i = 0; i < klen; i++) 682 key[i] ^= HMAC_OPAD_VAL; 683 684 /* PE is little-endian, insure proper byte order */ 685 for (i = 0; i < N(ses->ses_hminner); i++) { 686 ses->ses_hminner[i] = htole32(ses->ses_hminner[i]); 687 ses->ses_hmouter[i] = htole32(ses->ses_hmouter[i]); 688 } 689 } 690 #undef N 691 692 /* 693 * Allocate a new 'session' and return an encoded session id. 'sidp' 694 * contains our registration id, and should contain an encoded session 695 * id on successful allocation. 696 */ 697 static int 698 safe_newsession(device_t dev, u_int32_t *sidp, struct cryptoini *cri) 699 { 700 struct safe_softc *sc = device_get_softc(dev); 701 struct cryptoini *c, *encini = NULL, *macini = NULL; 702 struct safe_session *ses = NULL; 703 int sesn; 704 705 if (sidp == NULL || cri == NULL || sc == NULL) 706 return (EINVAL); 707 708 for (c = cri; c != NULL; c = c->cri_next) { 709 if (c->cri_alg == CRYPTO_MD5_HMAC || 710 c->cri_alg == CRYPTO_SHA1_HMAC || 711 c->cri_alg == CRYPTO_NULL_HMAC) { 712 if (macini) 713 return (EINVAL); 714 macini = c; 715 } else if (c->cri_alg == CRYPTO_DES_CBC || 716 c->cri_alg == CRYPTO_3DES_CBC || 717 c->cri_alg == CRYPTO_AES_CBC || 718 c->cri_alg == CRYPTO_NULL_CBC) { 719 if (encini) 720 return (EINVAL); 721 encini = c; 722 } else 723 return (EINVAL); 724 } 725 if (encini == NULL && macini == NULL) 726 return (EINVAL); 727 if (encini) { /* validate key length */ 728 switch (encini->cri_alg) { 729 case CRYPTO_DES_CBC: 730 if (encini->cri_klen != 64) 731 return (EINVAL); 732 break; 733 case CRYPTO_3DES_CBC: 734 if (encini->cri_klen != 192) 735 return (EINVAL); 736 break; 737 case CRYPTO_AES_CBC: 738 if (encini->cri_klen != 128 && 739 encini->cri_klen != 192 && 740 encini->cri_klen != 256) 741 return (EINVAL); 742 break; 743 } 744 } 745 746 if (sc->sc_sessions == NULL) { 747 ses = sc->sc_sessions = (struct safe_session *)malloc( 748 sizeof(struct safe_session), M_DEVBUF, M_NOWAIT); 749 if (ses == NULL) 750 return (ENOMEM); 751 sesn = 0; 752 sc->sc_nsessions = 1; 753 } else { 754 for (sesn = 0; sesn < sc->sc_nsessions; sesn++) { 755 if (sc->sc_sessions[sesn].ses_used == 0) { 756 ses = &sc->sc_sessions[sesn]; 757 break; 758 } 759 } 760 761 if (ses == NULL) { 762 sesn = sc->sc_nsessions; 763 ses = (struct safe_session *)malloc((sesn + 1) * 764 sizeof(struct safe_session), M_DEVBUF, M_NOWAIT); 765 if (ses == NULL) 766 return (ENOMEM); 767 bcopy(sc->sc_sessions, ses, sesn * 768 sizeof(struct safe_session)); 769 bzero(sc->sc_sessions, sesn * 770 sizeof(struct safe_session)); 771 free(sc->sc_sessions, M_DEVBUF); 772 sc->sc_sessions = ses; 773 ses = &sc->sc_sessions[sesn]; 774 sc->sc_nsessions++; 775 } 776 } 777 778 bzero(ses, sizeof(struct safe_session)); 779 ses->ses_used = 1; 780 781 if (encini) { 782 /* get an IV */ 783 /* XXX may read fewer than requested */ 784 read_random(ses->ses_iv, sizeof(ses->ses_iv)); 785 786 ses->ses_klen = encini->cri_klen; 787 if (encini->cri_key != NULL) 788 safe_setup_enckey(ses, encini->cri_key); 789 } 790 791 if (macini) { 792 ses->ses_mlen = macini->cri_mlen; 793 if (ses->ses_mlen == 0) { 794 if (macini->cri_alg == CRYPTO_MD5_HMAC) 795 ses->ses_mlen = MD5_HASH_LEN; 796 else 797 ses->ses_mlen = SHA1_HASH_LEN; 798 } 799 800 if (macini->cri_key != NULL) { 801 safe_setup_mackey(ses, macini->cri_alg, macini->cri_key, 802 macini->cri_klen / 8); 803 } 804 } 805 806 *sidp = SAFE_SID(device_get_unit(sc->sc_dev), sesn); 807 return (0); 808 } 809 810 /* 811 * Deallocate a session. 812 */ 813 static int 814 safe_freesession(device_t dev, u_int64_t tid) 815 { 816 struct safe_softc *sc = device_get_softc(dev); 817 int session, ret; 818 u_int32_t sid = ((u_int32_t) tid) & 0xffffffff; 819 820 if (sc == NULL) 821 return (EINVAL); 822 823 session = SAFE_SESSION(sid); 824 if (session < sc->sc_nsessions) { 825 bzero(&sc->sc_sessions[session], sizeof(sc->sc_sessions[session])); 826 ret = 0; 827 } else 828 ret = EINVAL; 829 return (ret); 830 } 831 832 static void 833 safe_op_cb(void *arg, bus_dma_segment_t *seg, int nsegs, bus_size_t mapsize, int error) 834 { 835 struct safe_operand *op = arg; 836 837 DPRINTF(("%s: mapsize %u nsegs %d error %d\n", __func__, 838 (u_int) mapsize, nsegs, error)); 839 if (error != 0) 840 return; 841 op->mapsize = mapsize; 842 op->nsegs = nsegs; 843 bcopy(seg, op->segs, nsegs * sizeof (seg[0])); 844 } 845 846 static int 847 safe_process(device_t dev, struct cryptop *crp, int hint) 848 { 849 struct safe_softc *sc = device_get_softc(dev); 850 int err = 0, i, nicealign, uniform; 851 struct cryptodesc *crd1, *crd2, *maccrd, *enccrd; 852 int bypass, oplen, ivsize; 853 caddr_t iv; 854 int16_t coffset; 855 struct safe_session *ses; 856 struct safe_ringentry *re; 857 struct safe_sarec *sa; 858 struct safe_pdesc *pd; 859 u_int32_t cmd0, cmd1, staterec; 860 861 if (crp == NULL || crp->crp_callback == NULL || sc == NULL) { 862 safestats.st_invalid++; 863 return (EINVAL); 864 } 865 if (SAFE_SESSION(crp->crp_sid) >= sc->sc_nsessions) { 866 safestats.st_badsession++; 867 return (EINVAL); 868 } 869 870 mtx_lock(&sc->sc_ringmtx); 871 if (sc->sc_front == sc->sc_back && sc->sc_nqchip != 0) { 872 safestats.st_ringfull++; 873 sc->sc_needwakeup |= CRYPTO_SYMQ; 874 mtx_unlock(&sc->sc_ringmtx); 875 return (ERESTART); 876 } 877 re = sc->sc_front; 878 879 staterec = re->re_sa.sa_staterec; /* save */ 880 /* NB: zero everything but the PE descriptor */ 881 bzero(&re->re_sa, sizeof(struct safe_ringentry) - sizeof(re->re_desc)); 882 re->re_sa.sa_staterec = staterec; /* restore */ 883 884 re->re_crp = crp; 885 re->re_sesn = SAFE_SESSION(crp->crp_sid); 886 887 if (crp->crp_flags & CRYPTO_F_IMBUF) { 888 re->re_src_m = (struct mbuf *)crp->crp_buf; 889 re->re_dst_m = (struct mbuf *)crp->crp_buf; 890 } else if (crp->crp_flags & CRYPTO_F_IOV) { 891 re->re_src_io = (struct uio *)crp->crp_buf; 892 re->re_dst_io = (struct uio *)crp->crp_buf; 893 } else { 894 safestats.st_badflags++; 895 err = EINVAL; 896 goto errout; /* XXX we don't handle contiguous blocks! */ 897 } 898 899 sa = &re->re_sa; 900 ses = &sc->sc_sessions[re->re_sesn]; 901 902 crd1 = crp->crp_desc; 903 if (crd1 == NULL) { 904 safestats.st_nodesc++; 905 err = EINVAL; 906 goto errout; 907 } 908 crd2 = crd1->crd_next; 909 910 cmd0 = SAFE_SA_CMD0_BASIC; /* basic group operation */ 911 cmd1 = 0; 912 if (crd2 == NULL) { 913 if (crd1->crd_alg == CRYPTO_MD5_HMAC || 914 crd1->crd_alg == CRYPTO_SHA1_HMAC || 915 crd1->crd_alg == CRYPTO_NULL_HMAC) { 916 maccrd = crd1; 917 enccrd = NULL; 918 cmd0 |= SAFE_SA_CMD0_OP_HASH; 919 } else if (crd1->crd_alg == CRYPTO_DES_CBC || 920 crd1->crd_alg == CRYPTO_3DES_CBC || 921 crd1->crd_alg == CRYPTO_AES_CBC || 922 crd1->crd_alg == CRYPTO_NULL_CBC) { 923 maccrd = NULL; 924 enccrd = crd1; 925 cmd0 |= SAFE_SA_CMD0_OP_CRYPT; 926 } else { 927 safestats.st_badalg++; 928 err = EINVAL; 929 goto errout; 930 } 931 } else { 932 if ((crd1->crd_alg == CRYPTO_MD5_HMAC || 933 crd1->crd_alg == CRYPTO_SHA1_HMAC || 934 crd1->crd_alg == CRYPTO_NULL_HMAC) && 935 (crd2->crd_alg == CRYPTO_DES_CBC || 936 crd2->crd_alg == CRYPTO_3DES_CBC || 937 crd2->crd_alg == CRYPTO_AES_CBC || 938 crd2->crd_alg == CRYPTO_NULL_CBC) && 939 ((crd2->crd_flags & CRD_F_ENCRYPT) == 0)) { 940 maccrd = crd1; 941 enccrd = crd2; 942 } else if ((crd1->crd_alg == CRYPTO_DES_CBC || 943 crd1->crd_alg == CRYPTO_3DES_CBC || 944 crd1->crd_alg == CRYPTO_AES_CBC || 945 crd1->crd_alg == CRYPTO_NULL_CBC) && 946 (crd2->crd_alg == CRYPTO_MD5_HMAC || 947 crd2->crd_alg == CRYPTO_SHA1_HMAC || 948 crd2->crd_alg == CRYPTO_NULL_HMAC) && 949 (crd1->crd_flags & CRD_F_ENCRYPT)) { 950 enccrd = crd1; 951 maccrd = crd2; 952 } else { 953 safestats.st_badalg++; 954 err = EINVAL; 955 goto errout; 956 } 957 cmd0 |= SAFE_SA_CMD0_OP_BOTH; 958 } 959 960 if (enccrd) { 961 if (enccrd->crd_flags & CRD_F_KEY_EXPLICIT) 962 safe_setup_enckey(ses, enccrd->crd_key); 963 964 if (enccrd->crd_alg == CRYPTO_DES_CBC) { 965 cmd0 |= SAFE_SA_CMD0_DES; 966 cmd1 |= SAFE_SA_CMD1_CBC; 967 ivsize = 2*sizeof(u_int32_t); 968 } else if (enccrd->crd_alg == CRYPTO_3DES_CBC) { 969 cmd0 |= SAFE_SA_CMD0_3DES; 970 cmd1 |= SAFE_SA_CMD1_CBC; 971 ivsize = 2*sizeof(u_int32_t); 972 } else if (enccrd->crd_alg == CRYPTO_AES_CBC) { 973 cmd0 |= SAFE_SA_CMD0_AES; 974 cmd1 |= SAFE_SA_CMD1_CBC; 975 if (ses->ses_klen == 128) 976 cmd1 |= SAFE_SA_CMD1_AES128; 977 else if (ses->ses_klen == 192) 978 cmd1 |= SAFE_SA_CMD1_AES192; 979 else 980 cmd1 |= SAFE_SA_CMD1_AES256; 981 ivsize = 4*sizeof(u_int32_t); 982 } else { 983 cmd0 |= SAFE_SA_CMD0_CRYPT_NULL; 984 ivsize = 0; 985 } 986 987 /* 988 * Setup encrypt/decrypt state. When using basic ops 989 * we can't use an inline IV because hash/crypt offset 990 * must be from the end of the IV to the start of the 991 * crypt data and this leaves out the preceding header 992 * from the hash calculation. Instead we place the IV 993 * in the state record and set the hash/crypt offset to 994 * copy both the header+IV. 995 */ 996 if (enccrd->crd_flags & CRD_F_ENCRYPT) { 997 cmd0 |= SAFE_SA_CMD0_OUTBOUND; 998 999 if (enccrd->crd_flags & CRD_F_IV_EXPLICIT) 1000 iv = enccrd->crd_iv; 1001 else 1002 iv = (caddr_t) ses->ses_iv; 1003 if ((enccrd->crd_flags & CRD_F_IV_PRESENT) == 0) { 1004 crypto_copyback(crp->crp_flags, crp->crp_buf, 1005 enccrd->crd_inject, ivsize, iv); 1006 } 1007 bcopy(iv, re->re_sastate.sa_saved_iv, ivsize); 1008 cmd0 |= SAFE_SA_CMD0_IVLD_STATE | SAFE_SA_CMD0_SAVEIV; 1009 re->re_flags |= SAFE_QFLAGS_COPYOUTIV; 1010 } else { 1011 cmd0 |= SAFE_SA_CMD0_INBOUND; 1012 1013 if (enccrd->crd_flags & CRD_F_IV_EXPLICIT) { 1014 bcopy(enccrd->crd_iv, 1015 re->re_sastate.sa_saved_iv, ivsize); 1016 } else { 1017 crypto_copydata(crp->crp_flags, crp->crp_buf, 1018 enccrd->crd_inject, ivsize, 1019 (caddr_t)re->re_sastate.sa_saved_iv); 1020 } 1021 cmd0 |= SAFE_SA_CMD0_IVLD_STATE; 1022 } 1023 /* 1024 * For basic encryption use the zero pad algorithm. 1025 * This pads results to an 8-byte boundary and 1026 * suppresses padding verification for inbound (i.e. 1027 * decrypt) operations. 1028 * 1029 * NB: Not sure if the 8-byte pad boundary is a problem. 1030 */ 1031 cmd0 |= SAFE_SA_CMD0_PAD_ZERO; 1032 1033 /* XXX assert key bufs have the same size */ 1034 bcopy(ses->ses_key, sa->sa_key, sizeof(sa->sa_key)); 1035 } 1036 1037 if (maccrd) { 1038 if (maccrd->crd_flags & CRD_F_KEY_EXPLICIT) { 1039 safe_setup_mackey(ses, maccrd->crd_alg, 1040 maccrd->crd_key, maccrd->crd_klen / 8); 1041 } 1042 1043 if (maccrd->crd_alg == CRYPTO_MD5_HMAC) { 1044 cmd0 |= SAFE_SA_CMD0_MD5; 1045 cmd1 |= SAFE_SA_CMD1_HMAC; /* NB: enable HMAC */ 1046 } else if (maccrd->crd_alg == CRYPTO_SHA1_HMAC) { 1047 cmd0 |= SAFE_SA_CMD0_SHA1; 1048 cmd1 |= SAFE_SA_CMD1_HMAC; /* NB: enable HMAC */ 1049 } else { 1050 cmd0 |= SAFE_SA_CMD0_HASH_NULL; 1051 } 1052 /* 1053 * Digest data is loaded from the SA and the hash 1054 * result is saved to the state block where we 1055 * retrieve it for return to the caller. 1056 */ 1057 /* XXX assert digest bufs have the same size */ 1058 bcopy(ses->ses_hminner, sa->sa_indigest, 1059 sizeof(sa->sa_indigest)); 1060 bcopy(ses->ses_hmouter, sa->sa_outdigest, 1061 sizeof(sa->sa_outdigest)); 1062 1063 cmd0 |= SAFE_SA_CMD0_HSLD_SA | SAFE_SA_CMD0_SAVEHASH; 1064 re->re_flags |= SAFE_QFLAGS_COPYOUTICV; 1065 } 1066 1067 if (enccrd && maccrd) { 1068 /* 1069 * The offset from hash data to the start of 1070 * crypt data is the difference in the skips. 1071 */ 1072 bypass = maccrd->crd_skip; 1073 coffset = enccrd->crd_skip - maccrd->crd_skip; 1074 if (coffset < 0) { 1075 DPRINTF(("%s: hash does not precede crypt; " 1076 "mac skip %u enc skip %u\n", 1077 __func__, maccrd->crd_skip, enccrd->crd_skip)); 1078 safestats.st_skipmismatch++; 1079 err = EINVAL; 1080 goto errout; 1081 } 1082 oplen = enccrd->crd_skip + enccrd->crd_len; 1083 if (maccrd->crd_skip + maccrd->crd_len != oplen) { 1084 DPRINTF(("%s: hash amount %u != crypt amount %u\n", 1085 __func__, maccrd->crd_skip + maccrd->crd_len, 1086 oplen)); 1087 safestats.st_lenmismatch++; 1088 err = EINVAL; 1089 goto errout; 1090 } 1091 #ifdef SAFE_DEBUG 1092 if (safe_debug) { 1093 printf("mac: skip %d, len %d, inject %d\n", 1094 maccrd->crd_skip, maccrd->crd_len, 1095 maccrd->crd_inject); 1096 printf("enc: skip %d, len %d, inject %d\n", 1097 enccrd->crd_skip, enccrd->crd_len, 1098 enccrd->crd_inject); 1099 printf("bypass %d coffset %d oplen %d\n", 1100 bypass, coffset, oplen); 1101 } 1102 #endif 1103 if (coffset & 3) { /* offset must be 32-bit aligned */ 1104 DPRINTF(("%s: coffset %u misaligned\n", 1105 __func__, coffset)); 1106 safestats.st_coffmisaligned++; 1107 err = EINVAL; 1108 goto errout; 1109 } 1110 coffset >>= 2; 1111 if (coffset > 255) { /* offset must be <256 dwords */ 1112 DPRINTF(("%s: coffset %u too big\n", 1113 __func__, coffset)); 1114 safestats.st_cofftoobig++; 1115 err = EINVAL; 1116 goto errout; 1117 } 1118 /* 1119 * Tell the hardware to copy the header to the output. 1120 * The header is defined as the data from the end of 1121 * the bypass to the start of data to be encrypted. 1122 * Typically this is the inline IV. Note that you need 1123 * to do this even if src+dst are the same; it appears 1124 * that w/o this bit the crypted data is written 1125 * immediately after the bypass data. 1126 */ 1127 cmd1 |= SAFE_SA_CMD1_HDRCOPY; 1128 /* 1129 * Disable IP header mutable bit handling. This is 1130 * needed to get correct HMAC calculations. 1131 */ 1132 cmd1 |= SAFE_SA_CMD1_MUTABLE; 1133 } else { 1134 if (enccrd) { 1135 bypass = enccrd->crd_skip; 1136 oplen = bypass + enccrd->crd_len; 1137 } else { 1138 bypass = maccrd->crd_skip; 1139 oplen = bypass + maccrd->crd_len; 1140 } 1141 coffset = 0; 1142 } 1143 /* XXX verify multiple of 4 when using s/g */ 1144 if (bypass > 96) { /* bypass offset must be <= 96 bytes */ 1145 DPRINTF(("%s: bypass %u too big\n", __func__, bypass)); 1146 safestats.st_bypasstoobig++; 1147 err = EINVAL; 1148 goto errout; 1149 } 1150 1151 if (bus_dmamap_create(sc->sc_srcdmat, BUS_DMA_NOWAIT, &re->re_src_map)) { 1152 safestats.st_nomap++; 1153 err = ENOMEM; 1154 goto errout; 1155 } 1156 if (crp->crp_flags & CRYPTO_F_IMBUF) { 1157 if (bus_dmamap_load_mbuf(sc->sc_srcdmat, re->re_src_map, 1158 re->re_src_m, safe_op_cb, 1159 &re->re_src, BUS_DMA_NOWAIT) != 0) { 1160 bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map); 1161 re->re_src_map = NULL; 1162 safestats.st_noload++; 1163 err = ENOMEM; 1164 goto errout; 1165 } 1166 } else if (crp->crp_flags & CRYPTO_F_IOV) { 1167 if (bus_dmamap_load_uio(sc->sc_srcdmat, re->re_src_map, 1168 re->re_src_io, safe_op_cb, 1169 &re->re_src, BUS_DMA_NOWAIT) != 0) { 1170 bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map); 1171 re->re_src_map = NULL; 1172 safestats.st_noload++; 1173 err = ENOMEM; 1174 goto errout; 1175 } 1176 } 1177 nicealign = safe_dmamap_aligned(&re->re_src); 1178 uniform = safe_dmamap_uniform(&re->re_src); 1179 1180 DPRINTF(("src nicealign %u uniform %u nsegs %u\n", 1181 nicealign, uniform, re->re_src.nsegs)); 1182 if (re->re_src.nsegs > 1) { 1183 re->re_desc.d_src = sc->sc_spalloc.dma_paddr + 1184 ((caddr_t) sc->sc_spfree - (caddr_t) sc->sc_spring); 1185 for (i = 0; i < re->re_src_nsegs; i++) { 1186 /* NB: no need to check if there's space */ 1187 pd = sc->sc_spfree; 1188 if (++(sc->sc_spfree) == sc->sc_springtop) 1189 sc->sc_spfree = sc->sc_spring; 1190 1191 KASSERT((pd->pd_flags&3) == 0 || 1192 (pd->pd_flags&3) == SAFE_PD_DONE, 1193 ("bogus source particle descriptor; flags %x", 1194 pd->pd_flags)); 1195 pd->pd_addr = re->re_src_segs[i].ds_addr; 1196 pd->pd_size = re->re_src_segs[i].ds_len; 1197 pd->pd_flags = SAFE_PD_READY; 1198 } 1199 cmd0 |= SAFE_SA_CMD0_IGATHER; 1200 } else { 1201 /* 1202 * No need for gather, reference the operand directly. 1203 */ 1204 re->re_desc.d_src = re->re_src_segs[0].ds_addr; 1205 } 1206 1207 if (enccrd == NULL && maccrd != NULL) { 1208 /* 1209 * Hash op; no destination needed. 1210 */ 1211 } else { 1212 if (crp->crp_flags & CRYPTO_F_IOV) { 1213 if (!nicealign) { 1214 safestats.st_iovmisaligned++; 1215 err = EINVAL; 1216 goto errout; 1217 } 1218 if (uniform != 1) { 1219 /* 1220 * Source is not suitable for direct use as 1221 * the destination. Create a new scatter/gather 1222 * list based on the destination requirements 1223 * and check if that's ok. 1224 */ 1225 if (bus_dmamap_create(sc->sc_dstdmat, 1226 BUS_DMA_NOWAIT, &re->re_dst_map)) { 1227 safestats.st_nomap++; 1228 err = ENOMEM; 1229 goto errout; 1230 } 1231 if (bus_dmamap_load_uio(sc->sc_dstdmat, 1232 re->re_dst_map, re->re_dst_io, 1233 safe_op_cb, &re->re_dst, 1234 BUS_DMA_NOWAIT) != 0) { 1235 bus_dmamap_destroy(sc->sc_dstdmat, 1236 re->re_dst_map); 1237 re->re_dst_map = NULL; 1238 safestats.st_noload++; 1239 err = ENOMEM; 1240 goto errout; 1241 } 1242 uniform = safe_dmamap_uniform(&re->re_dst); 1243 if (!uniform) { 1244 /* 1245 * There's no way to handle the DMA 1246 * requirements with this uio. We 1247 * could create a separate DMA area for 1248 * the result and then copy it back, 1249 * but for now we just bail and return 1250 * an error. Note that uio requests 1251 * > SAFE_MAX_DSIZE are handled because 1252 * the DMA map and segment list for the 1253 * destination wil result in a 1254 * destination particle list that does 1255 * the necessary scatter DMA. 1256 */ 1257 safestats.st_iovnotuniform++; 1258 err = EINVAL; 1259 goto errout; 1260 } 1261 } else 1262 re->re_dst = re->re_src; 1263 } else if (crp->crp_flags & CRYPTO_F_IMBUF) { 1264 if (nicealign && uniform == 1) { 1265 /* 1266 * Source layout is suitable for direct 1267 * sharing of the DMA map and segment list. 1268 */ 1269 re->re_dst = re->re_src; 1270 } else if (nicealign && uniform == 2) { 1271 /* 1272 * The source is properly aligned but requires a 1273 * different particle list to handle DMA of the 1274 * result. Create a new map and do the load to 1275 * create the segment list. The particle 1276 * descriptor setup code below will handle the 1277 * rest. 1278 */ 1279 if (bus_dmamap_create(sc->sc_dstdmat, 1280 BUS_DMA_NOWAIT, &re->re_dst_map)) { 1281 safestats.st_nomap++; 1282 err = ENOMEM; 1283 goto errout; 1284 } 1285 if (bus_dmamap_load_mbuf(sc->sc_dstdmat, 1286 re->re_dst_map, re->re_dst_m, 1287 safe_op_cb, &re->re_dst, 1288 BUS_DMA_NOWAIT) != 0) { 1289 bus_dmamap_destroy(sc->sc_dstdmat, 1290 re->re_dst_map); 1291 re->re_dst_map = NULL; 1292 safestats.st_noload++; 1293 err = ENOMEM; 1294 goto errout; 1295 } 1296 } else { /* !(aligned and/or uniform) */ 1297 int totlen, len; 1298 struct mbuf *m, *top, **mp; 1299 1300 /* 1301 * DMA constraints require that we allocate a 1302 * new mbuf chain for the destination. We 1303 * allocate an entire new set of mbufs of 1304 * optimal/required size and then tell the 1305 * hardware to copy any bits that are not 1306 * created as a byproduct of the operation. 1307 */ 1308 if (!nicealign) 1309 safestats.st_unaligned++; 1310 if (!uniform) 1311 safestats.st_notuniform++; 1312 totlen = re->re_src_mapsize; 1313 if (re->re_src_m->m_flags & M_PKTHDR) { 1314 len = MHLEN; 1315 MGETHDR(m, M_NOWAIT, MT_DATA); 1316 if (m && !m_dup_pkthdr(m, re->re_src_m, 1317 M_NOWAIT)) { 1318 m_free(m); 1319 m = NULL; 1320 } 1321 } else { 1322 len = MLEN; 1323 MGET(m, M_NOWAIT, MT_DATA); 1324 } 1325 if (m == NULL) { 1326 safestats.st_nombuf++; 1327 err = sc->sc_nqchip ? ERESTART : ENOMEM; 1328 goto errout; 1329 } 1330 if (totlen >= MINCLSIZE) { 1331 if (!(MCLGET(m, M_NOWAIT))) { 1332 m_free(m); 1333 safestats.st_nomcl++; 1334 err = sc->sc_nqchip ? 1335 ERESTART : ENOMEM; 1336 goto errout; 1337 } 1338 len = MCLBYTES; 1339 } 1340 m->m_len = len; 1341 top = NULL; 1342 mp = ⊤ 1343 1344 while (totlen > 0) { 1345 if (top) { 1346 MGET(m, M_NOWAIT, MT_DATA); 1347 if (m == NULL) { 1348 m_freem(top); 1349 safestats.st_nombuf++; 1350 err = sc->sc_nqchip ? 1351 ERESTART : ENOMEM; 1352 goto errout; 1353 } 1354 len = MLEN; 1355 } 1356 if (top && totlen >= MINCLSIZE) { 1357 if (!(MCLGET(m, M_NOWAIT))) { 1358 *mp = m; 1359 m_freem(top); 1360 safestats.st_nomcl++; 1361 err = sc->sc_nqchip ? 1362 ERESTART : ENOMEM; 1363 goto errout; 1364 } 1365 len = MCLBYTES; 1366 } 1367 m->m_len = len = min(totlen, len); 1368 totlen -= len; 1369 *mp = m; 1370 mp = &m->m_next; 1371 } 1372 re->re_dst_m = top; 1373 if (bus_dmamap_create(sc->sc_dstdmat, 1374 BUS_DMA_NOWAIT, &re->re_dst_map) != 0) { 1375 safestats.st_nomap++; 1376 err = ENOMEM; 1377 goto errout; 1378 } 1379 if (bus_dmamap_load_mbuf(sc->sc_dstdmat, 1380 re->re_dst_map, re->re_dst_m, 1381 safe_op_cb, &re->re_dst, 1382 BUS_DMA_NOWAIT) != 0) { 1383 bus_dmamap_destroy(sc->sc_dstdmat, 1384 re->re_dst_map); 1385 re->re_dst_map = NULL; 1386 safestats.st_noload++; 1387 err = ENOMEM; 1388 goto errout; 1389 } 1390 if (re->re_src.mapsize > oplen) { 1391 /* 1392 * There's data following what the 1393 * hardware will copy for us. If this 1394 * isn't just the ICV (that's going to 1395 * be written on completion), copy it 1396 * to the new mbufs 1397 */ 1398 if (!(maccrd && 1399 (re->re_src.mapsize-oplen) == 12 && 1400 maccrd->crd_inject == oplen)) 1401 safe_mcopy(re->re_src_m, 1402 re->re_dst_m, 1403 oplen); 1404 else 1405 safestats.st_noicvcopy++; 1406 } 1407 } 1408 } else { 1409 safestats.st_badflags++; 1410 err = EINVAL; 1411 goto errout; 1412 } 1413 1414 if (re->re_dst.nsegs > 1) { 1415 re->re_desc.d_dst = sc->sc_dpalloc.dma_paddr + 1416 ((caddr_t) sc->sc_dpfree - (caddr_t) sc->sc_dpring); 1417 for (i = 0; i < re->re_dst_nsegs; i++) { 1418 pd = sc->sc_dpfree; 1419 KASSERT((pd->pd_flags&3) == 0 || 1420 (pd->pd_flags&3) == SAFE_PD_DONE, 1421 ("bogus dest particle descriptor; flags %x", 1422 pd->pd_flags)); 1423 if (++(sc->sc_dpfree) == sc->sc_dpringtop) 1424 sc->sc_dpfree = sc->sc_dpring; 1425 pd->pd_addr = re->re_dst_segs[i].ds_addr; 1426 pd->pd_flags = SAFE_PD_READY; 1427 } 1428 cmd0 |= SAFE_SA_CMD0_OSCATTER; 1429 } else { 1430 /* 1431 * No need for scatter, reference the operand directly. 1432 */ 1433 re->re_desc.d_dst = re->re_dst_segs[0].ds_addr; 1434 } 1435 } 1436 1437 /* 1438 * All done with setup; fillin the SA command words 1439 * and the packet engine descriptor. The operation 1440 * is now ready for submission to the hardware. 1441 */ 1442 sa->sa_cmd0 = cmd0 | SAFE_SA_CMD0_IPCI | SAFE_SA_CMD0_OPCI; 1443 sa->sa_cmd1 = cmd1 1444 | (coffset << SAFE_SA_CMD1_OFFSET_S) 1445 | SAFE_SA_CMD1_SAREV1 /* Rev 1 SA data structure */ 1446 | SAFE_SA_CMD1_SRPCI 1447 ; 1448 /* 1449 * NB: the order of writes is important here. In case the 1450 * chip is scanning the ring because of an outstanding request 1451 * it might nab this one too. In that case we need to make 1452 * sure the setup is complete before we write the length 1453 * field of the descriptor as it signals the descriptor is 1454 * ready for processing. 1455 */ 1456 re->re_desc.d_csr = SAFE_PE_CSR_READY | SAFE_PE_CSR_SAPCI; 1457 if (maccrd) 1458 re->re_desc.d_csr |= SAFE_PE_CSR_LOADSA | SAFE_PE_CSR_HASHFINAL; 1459 re->re_desc.d_len = oplen 1460 | SAFE_PE_LEN_READY 1461 | (bypass << SAFE_PE_LEN_BYPASS_S) 1462 ; 1463 1464 safestats.st_ipackets++; 1465 safestats.st_ibytes += oplen; 1466 1467 if (++(sc->sc_front) == sc->sc_ringtop) 1468 sc->sc_front = sc->sc_ring; 1469 1470 /* XXX honor batching */ 1471 safe_feed(sc, re); 1472 mtx_unlock(&sc->sc_ringmtx); 1473 return (0); 1474 1475 errout: 1476 if ((re->re_dst_m != NULL) && (re->re_src_m != re->re_dst_m)) 1477 m_freem(re->re_dst_m); 1478 1479 if (re->re_dst_map != NULL && re->re_dst_map != re->re_src_map) { 1480 bus_dmamap_unload(sc->sc_dstdmat, re->re_dst_map); 1481 bus_dmamap_destroy(sc->sc_dstdmat, re->re_dst_map); 1482 } 1483 if (re->re_src_map != NULL) { 1484 bus_dmamap_unload(sc->sc_srcdmat, re->re_src_map); 1485 bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map); 1486 } 1487 mtx_unlock(&sc->sc_ringmtx); 1488 if (err != ERESTART) { 1489 crp->crp_etype = err; 1490 crypto_done(crp); 1491 } else { 1492 sc->sc_needwakeup |= CRYPTO_SYMQ; 1493 } 1494 return (err); 1495 } 1496 1497 static void 1498 safe_callback(struct safe_softc *sc, struct safe_ringentry *re) 1499 { 1500 struct cryptop *crp = (struct cryptop *)re->re_crp; 1501 struct cryptodesc *crd; 1502 1503 safestats.st_opackets++; 1504 safestats.st_obytes += re->re_dst.mapsize; 1505 1506 safe_dma_sync(&sc->sc_ringalloc, 1507 BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE); 1508 if (re->re_desc.d_csr & SAFE_PE_CSR_STATUS) { 1509 device_printf(sc->sc_dev, "csr 0x%x cmd0 0x%x cmd1 0x%x\n", 1510 re->re_desc.d_csr, 1511 re->re_sa.sa_cmd0, re->re_sa.sa_cmd1); 1512 safestats.st_peoperr++; 1513 crp->crp_etype = EIO; /* something more meaningful? */ 1514 } 1515 if (re->re_dst_map != NULL && re->re_dst_map != re->re_src_map) { 1516 bus_dmamap_sync(sc->sc_dstdmat, re->re_dst_map, 1517 BUS_DMASYNC_POSTREAD); 1518 bus_dmamap_unload(sc->sc_dstdmat, re->re_dst_map); 1519 bus_dmamap_destroy(sc->sc_dstdmat, re->re_dst_map); 1520 } 1521 bus_dmamap_sync(sc->sc_srcdmat, re->re_src_map, BUS_DMASYNC_POSTWRITE); 1522 bus_dmamap_unload(sc->sc_srcdmat, re->re_src_map); 1523 bus_dmamap_destroy(sc->sc_srcdmat, re->re_src_map); 1524 1525 /* 1526 * If result was written to a differet mbuf chain, swap 1527 * it in as the return value and reclaim the original. 1528 */ 1529 if ((crp->crp_flags & CRYPTO_F_IMBUF) && re->re_src_m != re->re_dst_m) { 1530 m_freem(re->re_src_m); 1531 crp->crp_buf = (caddr_t)re->re_dst_m; 1532 } 1533 1534 if (re->re_flags & SAFE_QFLAGS_COPYOUTIV) { 1535 /* copy out IV for future use */ 1536 for (crd = crp->crp_desc; crd; crd = crd->crd_next) { 1537 int ivsize; 1538 1539 if (crd->crd_alg == CRYPTO_DES_CBC || 1540 crd->crd_alg == CRYPTO_3DES_CBC) { 1541 ivsize = 2*sizeof(u_int32_t); 1542 } else if (crd->crd_alg == CRYPTO_AES_CBC) { 1543 ivsize = 4*sizeof(u_int32_t); 1544 } else 1545 continue; 1546 crypto_copydata(crp->crp_flags, crp->crp_buf, 1547 crd->crd_skip + crd->crd_len - ivsize, ivsize, 1548 (caddr_t)sc->sc_sessions[re->re_sesn].ses_iv); 1549 break; 1550 } 1551 } 1552 1553 if (re->re_flags & SAFE_QFLAGS_COPYOUTICV) { 1554 /* copy out ICV result */ 1555 for (crd = crp->crp_desc; crd; crd = crd->crd_next) { 1556 if (!(crd->crd_alg == CRYPTO_MD5_HMAC || 1557 crd->crd_alg == CRYPTO_SHA1_HMAC || 1558 crd->crd_alg == CRYPTO_NULL_HMAC)) 1559 continue; 1560 if (crd->crd_alg == CRYPTO_SHA1_HMAC) { 1561 /* 1562 * SHA-1 ICV's are byte-swapped; fix 'em up 1563 * before copy them to their destination. 1564 */ 1565 re->re_sastate.sa_saved_indigest[0] = 1566 bswap32(re->re_sastate.sa_saved_indigest[0]); 1567 re->re_sastate.sa_saved_indigest[1] = 1568 bswap32(re->re_sastate.sa_saved_indigest[1]); 1569 re->re_sastate.sa_saved_indigest[2] = 1570 bswap32(re->re_sastate.sa_saved_indigest[2]); 1571 } 1572 crypto_copyback(crp->crp_flags, crp->crp_buf, 1573 crd->crd_inject, 1574 sc->sc_sessions[re->re_sesn].ses_mlen, 1575 (caddr_t)re->re_sastate.sa_saved_indigest); 1576 break; 1577 } 1578 } 1579 crypto_done(crp); 1580 } 1581 1582 /* 1583 * Copy all data past offset from srcm to dstm. 1584 */ 1585 static void 1586 safe_mcopy(struct mbuf *srcm, struct mbuf *dstm, u_int offset) 1587 { 1588 u_int j, dlen, slen; 1589 caddr_t dptr, sptr; 1590 1591 /* 1592 * Advance src and dst to offset. 1593 */ 1594 j = offset; 1595 while (j >= 0) { 1596 if (srcm->m_len > j) 1597 break; 1598 j -= srcm->m_len; 1599 srcm = srcm->m_next; 1600 if (srcm == NULL) 1601 return; 1602 } 1603 sptr = mtod(srcm, caddr_t) + j; 1604 slen = srcm->m_len - j; 1605 1606 j = offset; 1607 while (j >= 0) { 1608 if (dstm->m_len > j) 1609 break; 1610 j -= dstm->m_len; 1611 dstm = dstm->m_next; 1612 if (dstm == NULL) 1613 return; 1614 } 1615 dptr = mtod(dstm, caddr_t) + j; 1616 dlen = dstm->m_len - j; 1617 1618 /* 1619 * Copy everything that remains. 1620 */ 1621 for (;;) { 1622 j = min(slen, dlen); 1623 bcopy(sptr, dptr, j); 1624 if (slen == j) { 1625 srcm = srcm->m_next; 1626 if (srcm == NULL) 1627 return; 1628 sptr = srcm->m_data; 1629 slen = srcm->m_len; 1630 } else 1631 sptr += j, slen -= j; 1632 if (dlen == j) { 1633 dstm = dstm->m_next; 1634 if (dstm == NULL) 1635 return; 1636 dptr = dstm->m_data; 1637 dlen = dstm->m_len; 1638 } else 1639 dptr += j, dlen -= j; 1640 } 1641 } 1642 1643 #ifndef SAFE_NO_RNG 1644 #define SAFE_RNG_MAXWAIT 1000 1645 1646 static void 1647 safe_rng_init(struct safe_softc *sc) 1648 { 1649 u_int32_t w, v; 1650 int i; 1651 1652 WRITE_REG(sc, SAFE_RNG_CTRL, 0); 1653 /* use default value according to the manual */ 1654 WRITE_REG(sc, SAFE_RNG_CNFG, 0x834); /* magic from SafeNet */ 1655 WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0); 1656 1657 /* 1658 * There is a bug in rev 1.0 of the 1140 that when the RNG 1659 * is brought out of reset the ready status flag does not 1660 * work until the RNG has finished its internal initialization. 1661 * 1662 * So in order to determine the device is through its 1663 * initialization we must read the data register, using the 1664 * status reg in the read in case it is initialized. Then read 1665 * the data register until it changes from the first read. 1666 * Once it changes read the data register until it changes 1667 * again. At this time the RNG is considered initialized. 1668 * This could take between 750ms - 1000ms in time. 1669 */ 1670 i = 0; 1671 w = READ_REG(sc, SAFE_RNG_OUT); 1672 do { 1673 v = READ_REG(sc, SAFE_RNG_OUT); 1674 if (v != w) { 1675 w = v; 1676 break; 1677 } 1678 DELAY(10); 1679 } while (++i < SAFE_RNG_MAXWAIT); 1680 1681 /* Wait Until data changes again */ 1682 i = 0; 1683 do { 1684 v = READ_REG(sc, SAFE_RNG_OUT); 1685 if (v != w) 1686 break; 1687 DELAY(10); 1688 } while (++i < SAFE_RNG_MAXWAIT); 1689 } 1690 1691 static __inline void 1692 safe_rng_disable_short_cycle(struct safe_softc *sc) 1693 { 1694 WRITE_REG(sc, SAFE_RNG_CTRL, 1695 READ_REG(sc, SAFE_RNG_CTRL) &~ SAFE_RNG_CTRL_SHORTEN); 1696 } 1697 1698 static __inline void 1699 safe_rng_enable_short_cycle(struct safe_softc *sc) 1700 { 1701 WRITE_REG(sc, SAFE_RNG_CTRL, 1702 READ_REG(sc, SAFE_RNG_CTRL) | SAFE_RNG_CTRL_SHORTEN); 1703 } 1704 1705 static __inline u_int32_t 1706 safe_rng_read(struct safe_softc *sc) 1707 { 1708 int i; 1709 1710 i = 0; 1711 while (READ_REG(sc, SAFE_RNG_STAT) != 0 && ++i < SAFE_RNG_MAXWAIT) 1712 ; 1713 return READ_REG(sc, SAFE_RNG_OUT); 1714 } 1715 1716 static void 1717 safe_rng(void *arg) 1718 { 1719 struct safe_softc *sc = arg; 1720 u_int32_t buf[SAFE_RNG_MAXBUFSIZ]; /* NB: maybe move to softc */ 1721 u_int maxwords; 1722 int i; 1723 1724 safestats.st_rng++; 1725 /* 1726 * Fetch the next block of data. 1727 */ 1728 maxwords = safe_rngbufsize; 1729 if (maxwords > SAFE_RNG_MAXBUFSIZ) 1730 maxwords = SAFE_RNG_MAXBUFSIZ; 1731 retry: 1732 for (i = 0; i < maxwords; i++) 1733 buf[i] = safe_rng_read(sc); 1734 /* 1735 * Check the comparator alarm count and reset the h/w if 1736 * it exceeds our threshold. This guards against the 1737 * hardware oscillators resonating with external signals. 1738 */ 1739 if (READ_REG(sc, SAFE_RNG_ALM_CNT) > safe_rngmaxalarm) { 1740 u_int32_t freq_inc, w; 1741 1742 DPRINTF(("%s: alarm count %u exceeds threshold %u\n", __func__, 1743 READ_REG(sc, SAFE_RNG_ALM_CNT), safe_rngmaxalarm)); 1744 safestats.st_rngalarm++; 1745 safe_rng_enable_short_cycle(sc); 1746 freq_inc = 18; 1747 for (i = 0; i < 64; i++) { 1748 w = READ_REG(sc, SAFE_RNG_CNFG); 1749 freq_inc = ((w + freq_inc) & 0x3fL); 1750 w = ((w & ~0x3fL) | freq_inc); 1751 WRITE_REG(sc, SAFE_RNG_CNFG, w); 1752 1753 WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0); 1754 1755 (void) safe_rng_read(sc); 1756 DELAY(25); 1757 1758 if (READ_REG(sc, SAFE_RNG_ALM_CNT) == 0) { 1759 safe_rng_disable_short_cycle(sc); 1760 goto retry; 1761 } 1762 freq_inc = 1; 1763 } 1764 safe_rng_disable_short_cycle(sc); 1765 } else 1766 WRITE_REG(sc, SAFE_RNG_ALM_CNT, 0); 1767 1768 (*sc->sc_harvest)(sc->sc_rndtest, buf, maxwords*sizeof (u_int32_t)); 1769 callout_reset(&sc->sc_rngto, 1770 hz * (safe_rnginterval ? safe_rnginterval : 1), safe_rng, sc); 1771 } 1772 #endif /* SAFE_NO_RNG */ 1773 1774 static void 1775 safe_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error) 1776 { 1777 bus_addr_t *paddr = (bus_addr_t*) arg; 1778 *paddr = segs->ds_addr; 1779 } 1780 1781 static int 1782 safe_dma_malloc( 1783 struct safe_softc *sc, 1784 bus_size_t size, 1785 struct safe_dma_alloc *dma, 1786 int mapflags 1787 ) 1788 { 1789 int r; 1790 1791 r = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev), /* parent */ 1792 sizeof(u_int32_t), 0, /* alignment, bounds */ 1793 BUS_SPACE_MAXADDR_32BIT, /* lowaddr */ 1794 BUS_SPACE_MAXADDR, /* highaddr */ 1795 NULL, NULL, /* filter, filterarg */ 1796 size, /* maxsize */ 1797 1, /* nsegments */ 1798 size, /* maxsegsize */ 1799 BUS_DMA_ALLOCNOW, /* flags */ 1800 NULL, NULL, /* locking */ 1801 &dma->dma_tag); 1802 if (r != 0) { 1803 device_printf(sc->sc_dev, "safe_dma_malloc: " 1804 "bus_dma_tag_create failed; error %u\n", r); 1805 goto fail_0; 1806 } 1807 1808 r = bus_dmamem_alloc(dma->dma_tag, (void**) &dma->dma_vaddr, 1809 BUS_DMA_NOWAIT, &dma->dma_map); 1810 if (r != 0) { 1811 device_printf(sc->sc_dev, "safe_dma_malloc: " 1812 "bus_dmammem_alloc failed; size %ju, error %u\n", 1813 (uintmax_t)size, r); 1814 goto fail_1; 1815 } 1816 1817 r = bus_dmamap_load(dma->dma_tag, dma->dma_map, dma->dma_vaddr, 1818 size, 1819 safe_dmamap_cb, 1820 &dma->dma_paddr, 1821 mapflags | BUS_DMA_NOWAIT); 1822 if (r != 0) { 1823 device_printf(sc->sc_dev, "safe_dma_malloc: " 1824 "bus_dmamap_load failed; error %u\n", r); 1825 goto fail_2; 1826 } 1827 1828 dma->dma_size = size; 1829 return (0); 1830 1831 bus_dmamap_unload(dma->dma_tag, dma->dma_map); 1832 fail_2: 1833 bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map); 1834 fail_1: 1835 bus_dma_tag_destroy(dma->dma_tag); 1836 fail_0: 1837 dma->dma_tag = NULL; 1838 return (r); 1839 } 1840 1841 static void 1842 safe_dma_free(struct safe_softc *sc, struct safe_dma_alloc *dma) 1843 { 1844 bus_dmamap_unload(dma->dma_tag, dma->dma_map); 1845 bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map); 1846 bus_dma_tag_destroy(dma->dma_tag); 1847 } 1848 1849 /* 1850 * Resets the board. Values in the regesters are left as is 1851 * from the reset (i.e. initial values are assigned elsewhere). 1852 */ 1853 static void 1854 safe_reset_board(struct safe_softc *sc) 1855 { 1856 u_int32_t v; 1857 /* 1858 * Reset the device. The manual says no delay 1859 * is needed between marking and clearing reset. 1860 */ 1861 v = READ_REG(sc, SAFE_PE_DMACFG) &~ 1862 (SAFE_PE_DMACFG_PERESET | SAFE_PE_DMACFG_PDRRESET | 1863 SAFE_PE_DMACFG_SGRESET); 1864 WRITE_REG(sc, SAFE_PE_DMACFG, v 1865 | SAFE_PE_DMACFG_PERESET 1866 | SAFE_PE_DMACFG_PDRRESET 1867 | SAFE_PE_DMACFG_SGRESET); 1868 WRITE_REG(sc, SAFE_PE_DMACFG, v); 1869 } 1870 1871 /* 1872 * Initialize registers we need to touch only once. 1873 */ 1874 static void 1875 safe_init_board(struct safe_softc *sc) 1876 { 1877 u_int32_t v, dwords; 1878 1879 v = READ_REG(sc, SAFE_PE_DMACFG); 1880 v &=~ SAFE_PE_DMACFG_PEMODE; 1881 v |= SAFE_PE_DMACFG_FSENA /* failsafe enable */ 1882 | SAFE_PE_DMACFG_GPRPCI /* gather ring on PCI */ 1883 | SAFE_PE_DMACFG_SPRPCI /* scatter ring on PCI */ 1884 | SAFE_PE_DMACFG_ESDESC /* endian-swap descriptors */ 1885 | SAFE_PE_DMACFG_ESSA /* endian-swap SA's */ 1886 | SAFE_PE_DMACFG_ESPDESC /* endian-swap part. desc's */ 1887 ; 1888 WRITE_REG(sc, SAFE_PE_DMACFG, v); 1889 #if 0 1890 /* XXX select byte swap based on host byte order */ 1891 WRITE_REG(sc, SAFE_ENDIAN, 0x1b); 1892 #endif 1893 if (sc->sc_chiprev == SAFE_REV(1,0)) { 1894 /* 1895 * Avoid large PCI DMA transfers. Rev 1.0 has a bug where 1896 * "target mode transfers" done while the chip is DMA'ing 1897 * >1020 bytes cause the hardware to lockup. To avoid this 1898 * we reduce the max PCI transfer size and use small source 1899 * particle descriptors (<= 256 bytes). 1900 */ 1901 WRITE_REG(sc, SAFE_DMA_CFG, 256); 1902 device_printf(sc->sc_dev, 1903 "Reduce max DMA size to %u words for rev %u.%u WAR\n", 1904 (READ_REG(sc, SAFE_DMA_CFG)>>2) & 0xff, 1905 SAFE_REV_MAJ(sc->sc_chiprev), 1906 SAFE_REV_MIN(sc->sc_chiprev)); 1907 } 1908 1909 /* NB: operands+results are overlaid */ 1910 WRITE_REG(sc, SAFE_PE_PDRBASE, sc->sc_ringalloc.dma_paddr); 1911 WRITE_REG(sc, SAFE_PE_RDRBASE, sc->sc_ringalloc.dma_paddr); 1912 /* 1913 * Configure ring entry size and number of items in the ring. 1914 */ 1915 KASSERT((sizeof(struct safe_ringentry) % sizeof(u_int32_t)) == 0, 1916 ("PE ring entry not 32-bit aligned!")); 1917 dwords = sizeof(struct safe_ringentry) / sizeof(u_int32_t); 1918 WRITE_REG(sc, SAFE_PE_RINGCFG, 1919 (dwords << SAFE_PE_RINGCFG_OFFSET_S) | SAFE_MAX_NQUEUE); 1920 WRITE_REG(sc, SAFE_PE_RINGPOLL, 0); /* disable polling */ 1921 1922 WRITE_REG(sc, SAFE_PE_GRNGBASE, sc->sc_spalloc.dma_paddr); 1923 WRITE_REG(sc, SAFE_PE_SRNGBASE, sc->sc_dpalloc.dma_paddr); 1924 WRITE_REG(sc, SAFE_PE_PARTSIZE, 1925 (SAFE_TOTAL_DPART<<16) | SAFE_TOTAL_SPART); 1926 /* 1927 * NB: destination particles are fixed size. We use 1928 * an mbuf cluster and require all results go to 1929 * clusters or smaller. 1930 */ 1931 WRITE_REG(sc, SAFE_PE_PARTCFG, SAFE_MAX_DSIZE); 1932 1933 /* it's now safe to enable PE mode, do it */ 1934 WRITE_REG(sc, SAFE_PE_DMACFG, v | SAFE_PE_DMACFG_PEMODE); 1935 1936 /* 1937 * Configure hardware to use level-triggered interrupts and 1938 * to interrupt after each descriptor is processed. 1939 */ 1940 WRITE_REG(sc, SAFE_HI_CFG, SAFE_HI_CFG_LEVEL); 1941 WRITE_REG(sc, SAFE_HI_DESC_CNT, 1); 1942 WRITE_REG(sc, SAFE_HI_MASK, SAFE_INT_PE_DDONE | SAFE_INT_PE_ERROR); 1943 } 1944 1945 /* 1946 * Init PCI registers 1947 */ 1948 static void 1949 safe_init_pciregs(device_t dev) 1950 { 1951 } 1952 1953 /* 1954 * Clean up after a chip crash. 1955 * It is assumed that the caller in splimp() 1956 */ 1957 static void 1958 safe_cleanchip(struct safe_softc *sc) 1959 { 1960 1961 if (sc->sc_nqchip != 0) { 1962 struct safe_ringentry *re = sc->sc_back; 1963 1964 while (re != sc->sc_front) { 1965 if (re->re_desc.d_csr != 0) 1966 safe_free_entry(sc, re); 1967 if (++re == sc->sc_ringtop) 1968 re = sc->sc_ring; 1969 } 1970 sc->sc_back = re; 1971 sc->sc_nqchip = 0; 1972 } 1973 } 1974 1975 /* 1976 * free a safe_q 1977 * It is assumed that the caller is within splimp(). 1978 */ 1979 static int 1980 safe_free_entry(struct safe_softc *sc, struct safe_ringentry *re) 1981 { 1982 struct cryptop *crp; 1983 1984 /* 1985 * Free header MCR 1986 */ 1987 if ((re->re_dst_m != NULL) && (re->re_src_m != re->re_dst_m)) 1988 m_freem(re->re_dst_m); 1989 1990 crp = (struct cryptop *)re->re_crp; 1991 1992 re->re_desc.d_csr = 0; 1993 1994 crp->crp_etype = EFAULT; 1995 crypto_done(crp); 1996 return(0); 1997 } 1998 1999 /* 2000 * Routine to reset the chip and clean up. 2001 * It is assumed that the caller is in splimp() 2002 */ 2003 static void 2004 safe_totalreset(struct safe_softc *sc) 2005 { 2006 safe_reset_board(sc); 2007 safe_init_board(sc); 2008 safe_cleanchip(sc); 2009 } 2010 2011 /* 2012 * Is the operand suitable aligned for direct DMA. Each 2013 * segment must be aligned on a 32-bit boundary and all 2014 * but the last segment must be a multiple of 4 bytes. 2015 */ 2016 static int 2017 safe_dmamap_aligned(const struct safe_operand *op) 2018 { 2019 int i; 2020 2021 for (i = 0; i < op->nsegs; i++) { 2022 if (op->segs[i].ds_addr & 3) 2023 return (0); 2024 if (i != (op->nsegs - 1) && (op->segs[i].ds_len & 3)) 2025 return (0); 2026 } 2027 return (1); 2028 } 2029 2030 /* 2031 * Is the operand suitable for direct DMA as the destination 2032 * of an operation. The hardware requires that each ``particle'' 2033 * but the last in an operation result have the same size. We 2034 * fix that size at SAFE_MAX_DSIZE bytes. This routine returns 2035 * 0 if some segment is not a multiple of of this size, 1 if all 2036 * segments are exactly this size, or 2 if segments are at worst 2037 * a multple of this size. 2038 */ 2039 static int 2040 safe_dmamap_uniform(const struct safe_operand *op) 2041 { 2042 int result = 1; 2043 2044 if (op->nsegs > 0) { 2045 int i; 2046 2047 for (i = 0; i < op->nsegs-1; i++) { 2048 if (op->segs[i].ds_len % SAFE_MAX_DSIZE) 2049 return (0); 2050 if (op->segs[i].ds_len != SAFE_MAX_DSIZE) 2051 result = 2; 2052 } 2053 } 2054 return (result); 2055 } 2056 2057 #ifdef SAFE_DEBUG 2058 static void 2059 safe_dump_dmastatus(struct safe_softc *sc, const char *tag) 2060 { 2061 printf("%s: ENDIAN 0x%x SRC 0x%x DST 0x%x STAT 0x%x\n" 2062 , tag 2063 , READ_REG(sc, SAFE_DMA_ENDIAN) 2064 , READ_REG(sc, SAFE_DMA_SRCADDR) 2065 , READ_REG(sc, SAFE_DMA_DSTADDR) 2066 , READ_REG(sc, SAFE_DMA_STAT) 2067 ); 2068 } 2069 2070 static void 2071 safe_dump_intrstate(struct safe_softc *sc, const char *tag) 2072 { 2073 printf("%s: HI_CFG 0x%x HI_MASK 0x%x HI_DESC_CNT 0x%x HU_STAT 0x%x HM_STAT 0x%x\n" 2074 , tag 2075 , READ_REG(sc, SAFE_HI_CFG) 2076 , READ_REG(sc, SAFE_HI_MASK) 2077 , READ_REG(sc, SAFE_HI_DESC_CNT) 2078 , READ_REG(sc, SAFE_HU_STAT) 2079 , READ_REG(sc, SAFE_HM_STAT) 2080 ); 2081 } 2082 2083 static void 2084 safe_dump_ringstate(struct safe_softc *sc, const char *tag) 2085 { 2086 u_int32_t estat = READ_REG(sc, SAFE_PE_ERNGSTAT); 2087 2088 /* NB: assume caller has lock on ring */ 2089 printf("%s: ERNGSTAT %x (next %u) back %lu front %lu\n", 2090 tag, 2091 estat, (estat >> SAFE_PE_ERNGSTAT_NEXT_S), 2092 (unsigned long)(sc->sc_back - sc->sc_ring), 2093 (unsigned long)(sc->sc_front - sc->sc_ring)); 2094 } 2095 2096 static void 2097 safe_dump_request(struct safe_softc *sc, const char* tag, struct safe_ringentry *re) 2098 { 2099 int ix, nsegs; 2100 2101 ix = re - sc->sc_ring; 2102 printf("%s: %p (%u): csr %x src %x dst %x sa %x len %x\n" 2103 , tag 2104 , re, ix 2105 , re->re_desc.d_csr 2106 , re->re_desc.d_src 2107 , re->re_desc.d_dst 2108 , re->re_desc.d_sa 2109 , re->re_desc.d_len 2110 ); 2111 if (re->re_src.nsegs > 1) { 2112 ix = (re->re_desc.d_src - sc->sc_spalloc.dma_paddr) / 2113 sizeof(struct safe_pdesc); 2114 for (nsegs = re->re_src.nsegs; nsegs; nsegs--) { 2115 printf(" spd[%u] %p: %p size %u flags %x" 2116 , ix, &sc->sc_spring[ix] 2117 , (caddr_t)(uintptr_t) sc->sc_spring[ix].pd_addr 2118 , sc->sc_spring[ix].pd_size 2119 , sc->sc_spring[ix].pd_flags 2120 ); 2121 if (sc->sc_spring[ix].pd_size == 0) 2122 printf(" (zero!)"); 2123 printf("\n"); 2124 if (++ix == SAFE_TOTAL_SPART) 2125 ix = 0; 2126 } 2127 } 2128 if (re->re_dst.nsegs > 1) { 2129 ix = (re->re_desc.d_dst - sc->sc_dpalloc.dma_paddr) / 2130 sizeof(struct safe_pdesc); 2131 for (nsegs = re->re_dst.nsegs; nsegs; nsegs--) { 2132 printf(" dpd[%u] %p: %p flags %x\n" 2133 , ix, &sc->sc_dpring[ix] 2134 , (caddr_t)(uintptr_t) sc->sc_dpring[ix].pd_addr 2135 , sc->sc_dpring[ix].pd_flags 2136 ); 2137 if (++ix == SAFE_TOTAL_DPART) 2138 ix = 0; 2139 } 2140 } 2141 printf("sa: cmd0 %08x cmd1 %08x staterec %x\n", 2142 re->re_sa.sa_cmd0, re->re_sa.sa_cmd1, re->re_sa.sa_staterec); 2143 printf("sa: key %x %x %x %x %x %x %x %x\n" 2144 , re->re_sa.sa_key[0] 2145 , re->re_sa.sa_key[1] 2146 , re->re_sa.sa_key[2] 2147 , re->re_sa.sa_key[3] 2148 , re->re_sa.sa_key[4] 2149 , re->re_sa.sa_key[5] 2150 , re->re_sa.sa_key[6] 2151 , re->re_sa.sa_key[7] 2152 ); 2153 printf("sa: indigest %x %x %x %x %x\n" 2154 , re->re_sa.sa_indigest[0] 2155 , re->re_sa.sa_indigest[1] 2156 , re->re_sa.sa_indigest[2] 2157 , re->re_sa.sa_indigest[3] 2158 , re->re_sa.sa_indigest[4] 2159 ); 2160 printf("sa: outdigest %x %x %x %x %x\n" 2161 , re->re_sa.sa_outdigest[0] 2162 , re->re_sa.sa_outdigest[1] 2163 , re->re_sa.sa_outdigest[2] 2164 , re->re_sa.sa_outdigest[3] 2165 , re->re_sa.sa_outdigest[4] 2166 ); 2167 printf("sr: iv %x %x %x %x\n" 2168 , re->re_sastate.sa_saved_iv[0] 2169 , re->re_sastate.sa_saved_iv[1] 2170 , re->re_sastate.sa_saved_iv[2] 2171 , re->re_sastate.sa_saved_iv[3] 2172 ); 2173 printf("sr: hashbc %u indigest %x %x %x %x %x\n" 2174 , re->re_sastate.sa_saved_hashbc 2175 , re->re_sastate.sa_saved_indigest[0] 2176 , re->re_sastate.sa_saved_indigest[1] 2177 , re->re_sastate.sa_saved_indigest[2] 2178 , re->re_sastate.sa_saved_indigest[3] 2179 , re->re_sastate.sa_saved_indigest[4] 2180 ); 2181 } 2182 2183 static void 2184 safe_dump_ring(struct safe_softc *sc, const char *tag) 2185 { 2186 mtx_lock(&sc->sc_ringmtx); 2187 printf("\nSafeNet Ring State:\n"); 2188 safe_dump_intrstate(sc, tag); 2189 safe_dump_dmastatus(sc, tag); 2190 safe_dump_ringstate(sc, tag); 2191 if (sc->sc_nqchip) { 2192 struct safe_ringentry *re = sc->sc_back; 2193 do { 2194 safe_dump_request(sc, tag, re); 2195 if (++re == sc->sc_ringtop) 2196 re = sc->sc_ring; 2197 } while (re != sc->sc_front); 2198 } 2199 mtx_unlock(&sc->sc_ringmtx); 2200 } 2201 2202 static int 2203 sysctl_hw_safe_dump(SYSCTL_HANDLER_ARGS) 2204 { 2205 char dmode[64]; 2206 int error; 2207 2208 strncpy(dmode, "", sizeof(dmode) - 1); 2209 dmode[sizeof(dmode) - 1] = '\0'; 2210 error = sysctl_handle_string(oidp, &dmode[0], sizeof(dmode), req); 2211 2212 if (error == 0 && req->newptr != NULL) { 2213 struct safe_softc *sc = safec; 2214 2215 if (!sc) 2216 return EINVAL; 2217 if (strncmp(dmode, "dma", 3) == 0) 2218 safe_dump_dmastatus(sc, "safe0"); 2219 else if (strncmp(dmode, "int", 3) == 0) 2220 safe_dump_intrstate(sc, "safe0"); 2221 else if (strncmp(dmode, "ring", 4) == 0) 2222 safe_dump_ring(sc, "safe0"); 2223 else 2224 return EINVAL; 2225 } 2226 return error; 2227 } 2228 SYSCTL_PROC(_hw_safe, OID_AUTO, dump, CTLTYPE_STRING | CTLFLAG_RW, 2229 0, 0, sysctl_hw_safe_dump, "A", "Dump driver state"); 2230 #endif /* SAFE_DEBUG */ 2231