1 /* 2 * AMD Cryptographic Coprocessor (CCP) driver 3 * 4 * Copyright (C) 2013,2016 Advanced Micro Devices, Inc. 5 * 6 * Author: Tom Lendacky <thomas.lendacky@amd.com> 7 * Author: Gary R Hook <gary.hook@amd.com> 8 * 9 * This program is free software; you can redistribute it and/or modify 10 * it under the terms of the GNU General Public License version 2 as 11 * published by the Free Software Foundation. 12 */ 13 14 #ifndef __CCP_DEV_H__ 15 #define __CCP_DEV_H__ 16 17 #include <linux/device.h> 18 #include <linux/pci.h> 19 #include <linux/spinlock.h> 20 #include <linux/mutex.h> 21 #include <linux/list.h> 22 #include <linux/wait.h> 23 #include <linux/dmapool.h> 24 #include <linux/hw_random.h> 25 #include <linux/bitops.h> 26 #include <linux/interrupt.h> 27 #include <linux/irqreturn.h> 28 #include <linux/dmaengine.h> 29 30 #define MAX_CCP_NAME_LEN 16 31 #define MAX_DMAPOOL_NAME_LEN 32 32 33 #define MAX_HW_QUEUES 5 34 #define MAX_CMD_QLEN 100 35 36 #define TRNG_RETRIES 10 37 38 #define CACHE_NONE 0x00 39 #define CACHE_WB_NO_ALLOC 0xb7 40 41 /****** Register Mappings ******/ 42 #define Q_MASK_REG 0x000 43 #define TRNG_OUT_REG 0x00c 44 #define IRQ_MASK_REG 0x040 45 #define IRQ_STATUS_REG 0x200 46 47 #define DEL_CMD_Q_JOB 0x124 48 #define DEL_Q_ACTIVE 0x00000200 49 #define DEL_Q_ID_SHIFT 6 50 51 #define CMD_REQ0 0x180 52 #define CMD_REQ_INCR 0x04 53 54 #define CMD_Q_STATUS_BASE 0x210 55 #define CMD_Q_INT_STATUS_BASE 0x214 56 #define CMD_Q_STATUS_INCR 0x20 57 58 #define CMD_Q_CACHE_BASE 0x228 59 #define CMD_Q_CACHE_INC 0x20 60 61 #define CMD_Q_ERROR(__qs) ((__qs) & 0x0000003f) 62 #define CMD_Q_DEPTH(__qs) (((__qs) >> 12) & 0x0000000f) 63 64 /* ------------------------ CCP Version 5 Specifics ------------------------ */ 65 #define CMD5_QUEUE_MASK_OFFSET 0x00 66 #define CMD5_QUEUE_PRIO_OFFSET 0x04 67 #define CMD5_REQID_CONFIG_OFFSET 0x08 68 #define CMD5_CMD_TIMEOUT_OFFSET 0x10 69 #define LSB_PUBLIC_MASK_LO_OFFSET 0x18 70 #define LSB_PUBLIC_MASK_HI_OFFSET 0x1C 71 #define LSB_PRIVATE_MASK_LO_OFFSET 0x20 72 #define LSB_PRIVATE_MASK_HI_OFFSET 0x24 73 74 #define CMD5_Q_CONTROL_BASE 0x0000 75 #define CMD5_Q_TAIL_LO_BASE 0x0004 76 #define CMD5_Q_HEAD_LO_BASE 0x0008 77 #define CMD5_Q_INT_ENABLE_BASE 0x000C 78 #define CMD5_Q_INTERRUPT_STATUS_BASE 0x0010 79 80 #define CMD5_Q_STATUS_BASE 0x0100 81 #define CMD5_Q_INT_STATUS_BASE 0x0104 82 #define CMD5_Q_DMA_STATUS_BASE 0x0108 83 #define CMD5_Q_DMA_READ_STATUS_BASE 0x010C 84 #define CMD5_Q_DMA_WRITE_STATUS_BASE 0x0110 85 #define CMD5_Q_ABORT_BASE 0x0114 86 #define CMD5_Q_AX_CACHE_BASE 0x0118 87 88 #define CMD5_CONFIG_0_OFFSET 0x6000 89 #define CMD5_TRNG_CTL_OFFSET 0x6008 90 #define CMD5_AES_MASK_OFFSET 0x6010 91 #define CMD5_CLK_GATE_CTL_OFFSET 0x603C 92 93 /* Address offset between two virtual queue registers */ 94 #define CMD5_Q_STATUS_INCR 0x1000 95 96 /* Bit masks */ 97 #define CMD5_Q_RUN 0x1 98 #define CMD5_Q_HALT 0x2 99 #define CMD5_Q_MEM_LOCATION 0x4 100 #define CMD5_Q_SIZE 0x1F 101 #define CMD5_Q_SHIFT 3 102 #define COMMANDS_PER_QUEUE 16 103 #define QUEUE_SIZE_VAL ((ffs(COMMANDS_PER_QUEUE) - 2) & \ 104 CMD5_Q_SIZE) 105 #define Q_PTR_MASK (2 << (QUEUE_SIZE_VAL + 5) - 1) 106 #define Q_DESC_SIZE sizeof(struct ccp5_desc) 107 #define Q_SIZE(n) (COMMANDS_PER_QUEUE*(n)) 108 109 #define INT_COMPLETION 0x1 110 #define INT_ERROR 0x2 111 #define INT_QUEUE_STOPPED 0x4 112 #define ALL_INTERRUPTS (INT_COMPLETION| \ 113 INT_ERROR| \ 114 INT_QUEUE_STOPPED) 115 116 #define LSB_REGION_WIDTH 5 117 #define MAX_LSB_CNT 8 118 119 #define LSB_SIZE 16 120 #define LSB_ITEM_SIZE 32 121 #define PLSB_MAP_SIZE (LSB_SIZE) 122 #define SLSB_MAP_SIZE (MAX_LSB_CNT * LSB_SIZE) 123 124 #define LSB_ENTRY_NUMBER(LSB_ADDR) (LSB_ADDR / LSB_ITEM_SIZE) 125 126 /* ------------------------ CCP Version 3 Specifics ------------------------ */ 127 #define REQ0_WAIT_FOR_WRITE 0x00000004 128 #define REQ0_INT_ON_COMPLETE 0x00000002 129 #define REQ0_STOP_ON_COMPLETE 0x00000001 130 131 #define REQ0_CMD_Q_SHIFT 9 132 #define REQ0_JOBID_SHIFT 3 133 134 /****** REQ1 Related Values ******/ 135 #define REQ1_PROTECT_SHIFT 27 136 #define REQ1_ENGINE_SHIFT 23 137 #define REQ1_KEY_KSB_SHIFT 2 138 139 #define REQ1_EOM 0x00000002 140 #define REQ1_INIT 0x00000001 141 142 /* AES Related Values */ 143 #define REQ1_AES_TYPE_SHIFT 21 144 #define REQ1_AES_MODE_SHIFT 18 145 #define REQ1_AES_ACTION_SHIFT 17 146 #define REQ1_AES_CFB_SIZE_SHIFT 10 147 148 /* XTS-AES Related Values */ 149 #define REQ1_XTS_AES_SIZE_SHIFT 10 150 151 /* SHA Related Values */ 152 #define REQ1_SHA_TYPE_SHIFT 21 153 154 /* RSA Related Values */ 155 #define REQ1_RSA_MOD_SIZE_SHIFT 10 156 157 /* Pass-Through Related Values */ 158 #define REQ1_PT_BW_SHIFT 12 159 #define REQ1_PT_BS_SHIFT 10 160 161 /* ECC Related Values */ 162 #define REQ1_ECC_AFFINE_CONVERT 0x00200000 163 #define REQ1_ECC_FUNCTION_SHIFT 18 164 165 /****** REQ4 Related Values ******/ 166 #define REQ4_KSB_SHIFT 18 167 #define REQ4_MEMTYPE_SHIFT 16 168 169 /****** REQ6 Related Values ******/ 170 #define REQ6_MEMTYPE_SHIFT 16 171 172 /****** Key Storage Block ******/ 173 #define KSB_START 77 174 #define KSB_END 127 175 #define KSB_COUNT (KSB_END - KSB_START + 1) 176 #define CCP_SB_BITS 256 177 178 #define CCP_JOBID_MASK 0x0000003f 179 180 /* ------------------------ General CCP Defines ------------------------ */ 181 182 #define CCP_DMAPOOL_MAX_SIZE 64 183 #define CCP_DMAPOOL_ALIGN BIT(5) 184 185 #define CCP_REVERSE_BUF_SIZE 64 186 187 #define CCP_AES_KEY_SB_COUNT 1 188 #define CCP_AES_CTX_SB_COUNT 1 189 190 #define CCP_XTS_AES_KEY_SB_COUNT 1 191 #define CCP_XTS_AES_CTX_SB_COUNT 1 192 193 #define CCP_SHA_SB_COUNT 1 194 195 #define CCP_RSA_MAX_WIDTH 4096 196 197 #define CCP_PASSTHRU_BLOCKSIZE 256 198 #define CCP_PASSTHRU_MASKSIZE 32 199 #define CCP_PASSTHRU_SB_COUNT 1 200 201 #define CCP_ECC_MODULUS_BYTES 48 /* 384-bits */ 202 #define CCP_ECC_MAX_OPERANDS 6 203 #define CCP_ECC_MAX_OUTPUTS 3 204 #define CCP_ECC_SRC_BUF_SIZE 448 205 #define CCP_ECC_DST_BUF_SIZE 192 206 #define CCP_ECC_OPERAND_SIZE 64 207 #define CCP_ECC_OUTPUT_SIZE 64 208 #define CCP_ECC_RESULT_OFFSET 60 209 #define CCP_ECC_RESULT_SUCCESS 0x0001 210 211 #define CCP_SB_BYTES 32 212 213 struct ccp_op; 214 struct ccp_device; 215 struct ccp_cmd; 216 struct ccp_fns; 217 218 struct ccp_dma_cmd { 219 struct list_head entry; 220 221 struct ccp_cmd ccp_cmd; 222 }; 223 224 struct ccp_dma_desc { 225 struct list_head entry; 226 227 struct ccp_device *ccp; 228 229 struct list_head pending; 230 struct list_head active; 231 232 enum dma_status status; 233 struct dma_async_tx_descriptor tx_desc; 234 size_t len; 235 }; 236 237 struct ccp_dma_chan { 238 struct ccp_device *ccp; 239 240 spinlock_t lock; 241 struct list_head pending; 242 struct list_head active; 243 struct list_head complete; 244 245 struct tasklet_struct cleanup_tasklet; 246 247 enum dma_status status; 248 struct dma_chan dma_chan; 249 }; 250 251 struct ccp_cmd_queue { 252 struct ccp_device *ccp; 253 254 /* Queue identifier */ 255 u32 id; 256 257 /* Queue dma pool */ 258 struct dma_pool *dma_pool; 259 260 /* Queue base address (not neccessarily aligned)*/ 261 struct ccp5_desc *qbase; 262 263 /* Aligned queue start address (per requirement) */ 264 struct mutex q_mutex ____cacheline_aligned; 265 unsigned int qidx; 266 267 /* Version 5 has different requirements for queue memory */ 268 unsigned int qsize; 269 dma_addr_t qbase_dma; 270 dma_addr_t qdma_tail; 271 272 /* Per-queue reserved storage block(s) */ 273 u32 sb_key; 274 u32 sb_ctx; 275 276 /* Bitmap of LSBs that can be accessed by this queue */ 277 DECLARE_BITMAP(lsbmask, MAX_LSB_CNT); 278 /* Private LSB that is assigned to this queue, or -1 if none. 279 * Bitmap for my private LSB, unused otherwise 280 */ 281 unsigned int lsb; 282 DECLARE_BITMAP(lsbmap, PLSB_MAP_SIZE); 283 284 /* Queue processing thread */ 285 struct task_struct *kthread; 286 unsigned int active; 287 unsigned int suspended; 288 289 /* Number of free command slots available */ 290 unsigned int free_slots; 291 292 /* Interrupt masks */ 293 u32 int_ok; 294 u32 int_err; 295 296 /* Register addresses for queue */ 297 void __iomem *reg_control; 298 void __iomem *reg_tail_lo; 299 void __iomem *reg_head_lo; 300 void __iomem *reg_int_enable; 301 void __iomem *reg_interrupt_status; 302 void __iomem *reg_status; 303 void __iomem *reg_int_status; 304 void __iomem *reg_dma_status; 305 void __iomem *reg_dma_read_status; 306 void __iomem *reg_dma_write_status; 307 u32 qcontrol; /* Cached control register */ 308 309 /* Status values from job */ 310 u32 int_status; 311 u32 q_status; 312 u32 q_int_status; 313 u32 cmd_error; 314 315 /* Interrupt wait queue */ 316 wait_queue_head_t int_queue; 317 unsigned int int_rcvd; 318 } ____cacheline_aligned; 319 320 struct ccp_device { 321 struct list_head entry; 322 323 struct ccp_vdata *vdata; 324 unsigned int ord; 325 char name[MAX_CCP_NAME_LEN]; 326 char rngname[MAX_CCP_NAME_LEN]; 327 328 struct device *dev; 329 330 /* Bus specific device information 331 */ 332 void *dev_specific; 333 int (*get_irq)(struct ccp_device *ccp); 334 void (*free_irq)(struct ccp_device *ccp); 335 unsigned int irq; 336 337 /* I/O area used for device communication. The register mapping 338 * starts at an offset into the mapped bar. 339 * The CMD_REQx registers and the Delete_Cmd_Queue_Job register 340 * need to be protected while a command queue thread is accessing 341 * them. 342 */ 343 struct mutex req_mutex ____cacheline_aligned; 344 void __iomem *io_map; 345 void __iomem *io_regs; 346 347 /* Master lists that all cmds are queued on. Because there can be 348 * more than one CCP command queue that can process a cmd a separate 349 * backlog list is neeeded so that the backlog completion call 350 * completes before the cmd is available for execution. 351 */ 352 spinlock_t cmd_lock ____cacheline_aligned; 353 unsigned int cmd_count; 354 struct list_head cmd; 355 struct list_head backlog; 356 357 /* The command queues. These represent the queues available on the 358 * CCP that are available for processing cmds 359 */ 360 struct ccp_cmd_queue cmd_q[MAX_HW_QUEUES]; 361 unsigned int cmd_q_count; 362 363 /* Support for the CCP True RNG 364 */ 365 struct hwrng hwrng; 366 unsigned int hwrng_retries; 367 368 /* Support for the CCP DMA capabilities 369 */ 370 struct dma_device dma_dev; 371 struct ccp_dma_chan *ccp_dma_chan; 372 struct kmem_cache *dma_cmd_cache; 373 struct kmem_cache *dma_desc_cache; 374 375 /* A counter used to generate job-ids for cmds submitted to the CCP 376 */ 377 atomic_t current_id ____cacheline_aligned; 378 379 /* The v3 CCP uses key storage blocks (SB) to maintain context for 380 * certain operations. To prevent multiple cmds from using the same 381 * SB range a command queue reserves an SB range for the duration of 382 * the cmd. Each queue, will however, reserve 2 SB blocks for 383 * operations that only require single SB entries (eg. AES context/iv 384 * and key) in order to avoid allocation contention. This will reserve 385 * at most 10 SB entries, leaving 40 SB entries available for dynamic 386 * allocation. 387 * 388 * The v5 CCP Local Storage Block (LSB) is broken up into 8 389 * memrory ranges, each of which can be enabled for access by one 390 * or more queues. Device initialization takes this into account, 391 * and attempts to assign one region for exclusive use by each 392 * available queue; the rest are then aggregated as "public" use. 393 * If there are fewer regions than queues, all regions are shared 394 * amongst all queues. 395 */ 396 struct mutex sb_mutex ____cacheline_aligned; 397 DECLARE_BITMAP(sb, KSB_COUNT); 398 wait_queue_head_t sb_queue; 399 unsigned int sb_avail; 400 unsigned int sb_count; 401 u32 sb_start; 402 403 /* Bitmap of shared LSBs, if any */ 404 DECLARE_BITMAP(lsbmap, SLSB_MAP_SIZE); 405 406 /* Suspend support */ 407 unsigned int suspending; 408 wait_queue_head_t suspend_queue; 409 410 /* DMA caching attribute support */ 411 unsigned int axcache; 412 }; 413 414 enum ccp_memtype { 415 CCP_MEMTYPE_SYSTEM = 0, 416 CCP_MEMTYPE_SB, 417 CCP_MEMTYPE_LOCAL, 418 CCP_MEMTYPE__LAST, 419 }; 420 #define CCP_MEMTYPE_LSB CCP_MEMTYPE_KSB 421 422 struct ccp_dma_info { 423 dma_addr_t address; 424 unsigned int offset; 425 unsigned int length; 426 enum dma_data_direction dir; 427 }; 428 429 struct ccp_dm_workarea { 430 struct device *dev; 431 struct dma_pool *dma_pool; 432 unsigned int length; 433 434 u8 *address; 435 struct ccp_dma_info dma; 436 }; 437 438 struct ccp_sg_workarea { 439 struct scatterlist *sg; 440 int nents; 441 442 struct scatterlist *dma_sg; 443 struct device *dma_dev; 444 unsigned int dma_count; 445 enum dma_data_direction dma_dir; 446 447 unsigned int sg_used; 448 449 u64 bytes_left; 450 }; 451 452 struct ccp_data { 453 struct ccp_sg_workarea sg_wa; 454 struct ccp_dm_workarea dm_wa; 455 }; 456 457 struct ccp_mem { 458 enum ccp_memtype type; 459 union { 460 struct ccp_dma_info dma; 461 u32 sb; 462 } u; 463 }; 464 465 struct ccp_aes_op { 466 enum ccp_aes_type type; 467 enum ccp_aes_mode mode; 468 enum ccp_aes_action action; 469 }; 470 471 struct ccp_xts_aes_op { 472 enum ccp_aes_action action; 473 enum ccp_xts_aes_unit_size unit_size; 474 }; 475 476 struct ccp_sha_op { 477 enum ccp_sha_type type; 478 u64 msg_bits; 479 }; 480 481 struct ccp_rsa_op { 482 u32 mod_size; 483 u32 input_len; 484 }; 485 486 struct ccp_passthru_op { 487 enum ccp_passthru_bitwise bit_mod; 488 enum ccp_passthru_byteswap byte_swap; 489 }; 490 491 struct ccp_ecc_op { 492 enum ccp_ecc_function function; 493 }; 494 495 struct ccp_op { 496 struct ccp_cmd_queue *cmd_q; 497 498 u32 jobid; 499 u32 ioc; 500 u32 soc; 501 u32 sb_key; 502 u32 sb_ctx; 503 u32 init; 504 u32 eom; 505 506 struct ccp_mem src; 507 struct ccp_mem dst; 508 struct ccp_mem exp; 509 510 union { 511 struct ccp_aes_op aes; 512 struct ccp_xts_aes_op xts; 513 struct ccp_sha_op sha; 514 struct ccp_rsa_op rsa; 515 struct ccp_passthru_op passthru; 516 struct ccp_ecc_op ecc; 517 } u; 518 struct ccp_mem key; 519 }; 520 521 static inline u32 ccp_addr_lo(struct ccp_dma_info *info) 522 { 523 return lower_32_bits(info->address + info->offset); 524 } 525 526 static inline u32 ccp_addr_hi(struct ccp_dma_info *info) 527 { 528 return upper_32_bits(info->address + info->offset) & 0x0000ffff; 529 } 530 531 /** 532 * descriptor for version 5 CPP commands 533 * 8 32-bit words: 534 * word 0: function; engine; control bits 535 * word 1: length of source data 536 * word 2: low 32 bits of source pointer 537 * word 3: upper 16 bits of source pointer; source memory type 538 * word 4: low 32 bits of destination pointer 539 * word 5: upper 16 bits of destination pointer; destination memory type 540 * word 6: low 32 bits of key pointer 541 * word 7: upper 16 bits of key pointer; key memory type 542 */ 543 struct dword0 { 544 __le32 soc:1; 545 __le32 ioc:1; 546 __le32 rsvd1:1; 547 __le32 init:1; 548 __le32 eom:1; /* AES/SHA only */ 549 __le32 function:15; 550 __le32 engine:4; 551 __le32 prot:1; 552 __le32 rsvd2:7; 553 }; 554 555 struct dword3 { 556 __le32 src_hi:16; 557 __le32 src_mem:2; 558 __le32 lsb_cxt_id:8; 559 __le32 rsvd1:5; 560 __le32 fixed:1; 561 }; 562 563 union dword4 { 564 __le32 dst_lo; /* NON-SHA */ 565 __le32 sha_len_lo; /* SHA */ 566 }; 567 568 union dword5 { 569 struct { 570 __le32 dst_hi:16; 571 __le32 dst_mem:2; 572 __le32 rsvd1:13; 573 __le32 fixed:1; 574 } fields; 575 __le32 sha_len_hi; 576 }; 577 578 struct dword7 { 579 __le32 key_hi:16; 580 __le32 key_mem:2; 581 __le32 rsvd1:14; 582 }; 583 584 struct ccp5_desc { 585 struct dword0 dw0; 586 __le32 length; 587 __le32 src_lo; 588 struct dword3 dw3; 589 union dword4 dw4; 590 union dword5 dw5; 591 __le32 key_lo; 592 struct dword7 dw7; 593 }; 594 595 int ccp_pci_init(void); 596 void ccp_pci_exit(void); 597 598 int ccp_platform_init(void); 599 void ccp_platform_exit(void); 600 601 void ccp_add_device(struct ccp_device *ccp); 602 void ccp_del_device(struct ccp_device *ccp); 603 604 extern void ccp_log_error(struct ccp_device *, int); 605 606 struct ccp_device *ccp_alloc_struct(struct device *dev); 607 bool ccp_queues_suspended(struct ccp_device *ccp); 608 int ccp_cmd_queue_thread(void *data); 609 int ccp_trng_read(struct hwrng *rng, void *data, size_t max, bool wait); 610 611 int ccp_run_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd); 612 613 int ccp_register_rng(struct ccp_device *ccp); 614 void ccp_unregister_rng(struct ccp_device *ccp); 615 int ccp_dmaengine_register(struct ccp_device *ccp); 616 void ccp_dmaengine_unregister(struct ccp_device *ccp); 617 618 /* Structure for computation functions that are device-specific */ 619 struct ccp_actions { 620 int (*aes)(struct ccp_op *); 621 int (*xts_aes)(struct ccp_op *); 622 int (*sha)(struct ccp_op *); 623 int (*rsa)(struct ccp_op *); 624 int (*passthru)(struct ccp_op *); 625 int (*ecc)(struct ccp_op *); 626 u32 (*sballoc)(struct ccp_cmd_queue *, unsigned int); 627 void (*sbfree)(struct ccp_cmd_queue *, unsigned int, 628 unsigned int); 629 unsigned int (*get_free_slots)(struct ccp_cmd_queue *); 630 int (*init)(struct ccp_device *); 631 void (*destroy)(struct ccp_device *); 632 irqreturn_t (*irqhandler)(int, void *); 633 }; 634 635 /* Structure to hold CCP version-specific values */ 636 struct ccp_vdata { 637 const unsigned int version; 638 void (*setup)(struct ccp_device *); 639 const struct ccp_actions *perform; 640 const unsigned int bar; 641 const unsigned int offset; 642 }; 643 644 extern const struct ccp_vdata ccpv3; 645 extern const struct ccp_vdata ccpv5a; 646 extern const struct ccp_vdata ccpv5b; 647 648 #endif 649