2 * AMD Cryptographic Coprocessor (CCP) driver
4 * Copyright (C) 2016,2017 Advanced Micro Devices, Inc.
6 * Author: Gary R Hook <gary.hook@amd.com>
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 as
10 * published by the Free Software Foundation.
13 #include <linux/module.h>
14 #include <linux/kernel.h>
15 #include <linux/pci.h>
16 #include <linux/kthread.h>
17 #include <linux/debugfs.h>
18 #include <linux/dma-mapping.h>
19 #include <linux/interrupt.h>
20 #include <linux/compiler.h>
21 #include <linux/ccp.h>
25 /* Allocate the requested number of contiguous LSB slots
26 * from the LSB bitmap. Look in the private range for this
27 * queue first; failing that, check the public area.
28 * If no space is available, wait around.
29 * Return: first slot number
31 static u32
ccp_lsb_alloc(struct ccp_cmd_queue
*cmd_q
, unsigned int count
)
33 struct ccp_device
*ccp
;
36 /* First look at the map for the queue */
37 if (cmd_q
->lsb
>= 0) {
38 start
= (u32
)bitmap_find_next_zero_area(cmd_q
->lsbmap
,
41 if (start
< LSB_SIZE
) {
42 bitmap_set(cmd_q
->lsbmap
, start
, count
);
43 return start
+ cmd_q
->lsb
* LSB_SIZE
;
47 /* No joy; try to get an entry from the shared blocks */
50 mutex_lock(&ccp
->sb_mutex
);
52 start
= (u32
)bitmap_find_next_zero_area(ccp
->lsbmap
,
53 MAX_LSB_CNT
* LSB_SIZE
,
56 if (start
<= MAX_LSB_CNT
* LSB_SIZE
) {
57 bitmap_set(ccp
->lsbmap
, start
, count
);
59 mutex_unlock(&ccp
->sb_mutex
);
65 mutex_unlock(&ccp
->sb_mutex
);
67 /* Wait for KSB entries to become available */
68 if (wait_event_interruptible(ccp
->sb_queue
, ccp
->sb_avail
))
73 /* Free a number of LSB slots from the bitmap, starting at
74 * the indicated starting slot number.
76 static void ccp_lsb_free(struct ccp_cmd_queue
*cmd_q
, unsigned int start
,
82 if (cmd_q
->lsb
== start
) {
83 /* An entry from the private LSB */
84 bitmap_clear(cmd_q
->lsbmap
, start
, count
);
86 /* From the shared LSBs */
87 struct ccp_device
*ccp
= cmd_q
->ccp
;
89 mutex_lock(&ccp
->sb_mutex
);
90 bitmap_clear(ccp
->lsbmap
, start
, count
);
92 mutex_unlock(&ccp
->sb_mutex
);
93 wake_up_interruptible_all(&ccp
->sb_queue
);
97 /* CCP version 5: Union to define the function field (cmd_reg1/dword0) */
143 #define CCP_AES_SIZE(p) ((p)->aes.size)
144 #define CCP_AES_ENCRYPT(p) ((p)->aes.encrypt)
145 #define CCP_AES_MODE(p) ((p)->aes.mode)
146 #define CCP_AES_TYPE(p) ((p)->aes.type)
147 #define CCP_XTS_SIZE(p) ((p)->aes_xts.size)
148 #define CCP_XTS_TYPE(p) ((p)->aes_xts.type)
149 #define CCP_XTS_ENCRYPT(p) ((p)->aes_xts.encrypt)
150 #define CCP_DES3_SIZE(p) ((p)->des3.size)
151 #define CCP_DES3_ENCRYPT(p) ((p)->des3.encrypt)
152 #define CCP_DES3_MODE(p) ((p)->des3.mode)
153 #define CCP_DES3_TYPE(p) ((p)->des3.type)
154 #define CCP_SHA_TYPE(p) ((p)->sha.type)
155 #define CCP_RSA_SIZE(p) ((p)->rsa.size)
156 #define CCP_PT_BYTESWAP(p) ((p)->pt.byteswap)
157 #define CCP_PT_BITWISE(p) ((p)->pt.bitwise)
158 #define CCP_ECC_MODE(p) ((p)->ecc.mode)
159 #define CCP_ECC_AFFINE(p) ((p)->ecc.one)
162 #define CCP5_CMD_DW0(p) ((p)->dw0)
163 #define CCP5_CMD_SOC(p) (CCP5_CMD_DW0(p).soc)
164 #define CCP5_CMD_IOC(p) (CCP5_CMD_DW0(p).ioc)
165 #define CCP5_CMD_INIT(p) (CCP5_CMD_DW0(p).init)
166 #define CCP5_CMD_EOM(p) (CCP5_CMD_DW0(p).eom)
167 #define CCP5_CMD_FUNCTION(p) (CCP5_CMD_DW0(p).function)
168 #define CCP5_CMD_ENGINE(p) (CCP5_CMD_DW0(p).engine)
169 #define CCP5_CMD_PROT(p) (CCP5_CMD_DW0(p).prot)
172 #define CCP5_CMD_DW1(p) ((p)->length)
173 #define CCP5_CMD_LEN(p) (CCP5_CMD_DW1(p))
176 #define CCP5_CMD_DW2(p) ((p)->src_lo)
177 #define CCP5_CMD_SRC_LO(p) (CCP5_CMD_DW2(p))
180 #define CCP5_CMD_DW3(p) ((p)->dw3)
181 #define CCP5_CMD_SRC_MEM(p) ((p)->dw3.src_mem)
182 #define CCP5_CMD_SRC_HI(p) ((p)->dw3.src_hi)
183 #define CCP5_CMD_LSB_ID(p) ((p)->dw3.lsb_cxt_id)
184 #define CCP5_CMD_FIX_SRC(p) ((p)->dw3.fixed)
187 #define CCP5_CMD_DW4(p) ((p)->dw4)
188 #define CCP5_CMD_DST_LO(p) (CCP5_CMD_DW4(p).dst_lo)
189 #define CCP5_CMD_DW5(p) ((p)->dw5.fields.dst_hi)
190 #define CCP5_CMD_DST_HI(p) (CCP5_CMD_DW5(p))
191 #define CCP5_CMD_DST_MEM(p) ((p)->dw5.fields.dst_mem)
192 #define CCP5_CMD_FIX_DST(p) ((p)->dw5.fields.fixed)
193 #define CCP5_CMD_SHA_LO(p) ((p)->dw4.sha_len_lo)
194 #define CCP5_CMD_SHA_HI(p) ((p)->dw5.sha_len_hi)
197 #define CCP5_CMD_DW6(p) ((p)->key_lo)
198 #define CCP5_CMD_KEY_LO(p) (CCP5_CMD_DW6(p))
199 #define CCP5_CMD_DW7(p) ((p)->dw7)
200 #define CCP5_CMD_KEY_HI(p) ((p)->dw7.key_hi)
201 #define CCP5_CMD_KEY_MEM(p) ((p)->dw7.key_mem)
203 static inline u32
low_address(unsigned long addr
)
205 return (u64
)addr
& 0x0ffffffff;
208 static inline u32
high_address(unsigned long addr
)
210 return ((u64
)addr
>> 32) & 0x00000ffff;
213 static unsigned int ccp5_get_free_slots(struct ccp_cmd_queue
*cmd_q
)
215 unsigned int head_idx
, n
;
216 u32 head_lo
, queue_start
;
218 queue_start
= low_address(cmd_q
->qdma_tail
);
219 head_lo
= ioread32(cmd_q
->reg_head_lo
);
220 head_idx
= (head_lo
- queue_start
) / sizeof(struct ccp5_desc
);
222 n
= head_idx
+ COMMANDS_PER_QUEUE
- cmd_q
->qidx
- 1;
224 return n
% COMMANDS_PER_QUEUE
; /* Always one unused spot */
227 static int ccp5_do_cmd(struct ccp5_desc
*desc
,
228 struct ccp_cmd_queue
*cmd_q
)
238 if (CCP5_CMD_SOC(desc
)) {
239 CCP5_CMD_IOC(desc
) = 1;
240 CCP5_CMD_SOC(desc
) = 0;
242 mutex_lock(&cmd_q
->q_mutex
);
244 mP
= (u32
*) &cmd_q
->qbase
[cmd_q
->qidx
];
245 dP
= (__le32
*) desc
;
246 for (i
= 0; i
< 8; i
++)
247 mP
[i
] = cpu_to_le32(dP
[i
]); /* handle endianness */
249 cmd_q
->qidx
= (cmd_q
->qidx
+ 1) % COMMANDS_PER_QUEUE
;
251 /* The data used by this command must be flushed to memory */
254 /* Write the new tail address back to the queue register */
255 tail
= low_address(cmd_q
->qdma_tail
+ cmd_q
->qidx
* Q_DESC_SIZE
);
256 iowrite32(tail
, cmd_q
->reg_tail_lo
);
258 /* Turn the queue back on using our cached control register */
259 iowrite32(cmd_q
->qcontrol
| CMD5_Q_RUN
, cmd_q
->reg_control
);
260 mutex_unlock(&cmd_q
->q_mutex
);
262 if (CCP5_CMD_IOC(desc
)) {
263 /* Wait for the job to complete */
264 ret
= wait_event_interruptible(cmd_q
->int_queue
,
266 if (ret
|| cmd_q
->cmd_error
) {
267 /* Log the error and flush the queue by
268 * moving the head pointer
270 if (cmd_q
->cmd_error
)
271 ccp_log_error(cmd_q
->ccp
,
273 iowrite32(tail
, cmd_q
->reg_head_lo
);
283 static int ccp5_perform_aes(struct ccp_op
*op
)
285 struct ccp5_desc desc
;
286 union ccp_function function
;
287 u32 key_addr
= op
->sb_key
* LSB_ITEM_SIZE
;
289 op
->cmd_q
->total_aes_ops
++;
291 /* Zero out all the fields of the command desc */
292 memset(&desc
, 0, Q_DESC_SIZE
);
294 CCP5_CMD_ENGINE(&desc
) = CCP_ENGINE_AES
;
296 CCP5_CMD_SOC(&desc
) = op
->soc
;
297 CCP5_CMD_IOC(&desc
) = 1;
298 CCP5_CMD_INIT(&desc
) = op
->init
;
299 CCP5_CMD_EOM(&desc
) = op
->eom
;
300 CCP5_CMD_PROT(&desc
) = 0;
303 CCP_AES_ENCRYPT(&function
) = op
->u
.aes
.action
;
304 CCP_AES_MODE(&function
) = op
->u
.aes
.mode
;
305 CCP_AES_TYPE(&function
) = op
->u
.aes
.type
;
306 CCP_AES_SIZE(&function
) = op
->u
.aes
.size
;
308 CCP5_CMD_FUNCTION(&desc
) = function
.raw
;
310 CCP5_CMD_LEN(&desc
) = op
->src
.u
.dma
.length
;
312 CCP5_CMD_SRC_LO(&desc
) = ccp_addr_lo(&op
->src
.u
.dma
);
313 CCP5_CMD_SRC_HI(&desc
) = ccp_addr_hi(&op
->src
.u
.dma
);
314 CCP5_CMD_SRC_MEM(&desc
) = CCP_MEMTYPE_SYSTEM
;
316 CCP5_CMD_DST_LO(&desc
) = ccp_addr_lo(&op
->dst
.u
.dma
);
317 CCP5_CMD_DST_HI(&desc
) = ccp_addr_hi(&op
->dst
.u
.dma
);
318 CCP5_CMD_DST_MEM(&desc
) = CCP_MEMTYPE_SYSTEM
;
320 CCP5_CMD_KEY_LO(&desc
) = lower_32_bits(key_addr
);
321 CCP5_CMD_KEY_HI(&desc
) = 0;
322 CCP5_CMD_KEY_MEM(&desc
) = CCP_MEMTYPE_SB
;
323 CCP5_CMD_LSB_ID(&desc
) = op
->sb_ctx
;
325 return ccp5_do_cmd(&desc
, op
->cmd_q
);
328 static int ccp5_perform_xts_aes(struct ccp_op
*op
)
330 struct ccp5_desc desc
;
331 union ccp_function function
;
332 u32 key_addr
= op
->sb_key
* LSB_ITEM_SIZE
;
334 op
->cmd_q
->total_xts_aes_ops
++;
336 /* Zero out all the fields of the command desc */
337 memset(&desc
, 0, Q_DESC_SIZE
);
339 CCP5_CMD_ENGINE(&desc
) = CCP_ENGINE_XTS_AES_128
;
341 CCP5_CMD_SOC(&desc
) = op
->soc
;
342 CCP5_CMD_IOC(&desc
) = 1;
343 CCP5_CMD_INIT(&desc
) = op
->init
;
344 CCP5_CMD_EOM(&desc
) = op
->eom
;
345 CCP5_CMD_PROT(&desc
) = 0;
348 CCP_XTS_TYPE(&function
) = op
->u
.xts
.type
;
349 CCP_XTS_ENCRYPT(&function
) = op
->u
.xts
.action
;
350 CCP_XTS_SIZE(&function
) = op
->u
.xts
.unit_size
;
351 CCP5_CMD_FUNCTION(&desc
) = function
.raw
;
353 CCP5_CMD_LEN(&desc
) = op
->src
.u
.dma
.length
;
355 CCP5_CMD_SRC_LO(&desc
) = ccp_addr_lo(&op
->src
.u
.dma
);
356 CCP5_CMD_SRC_HI(&desc
) = ccp_addr_hi(&op
->src
.u
.dma
);
357 CCP5_CMD_SRC_MEM(&desc
) = CCP_MEMTYPE_SYSTEM
;
359 CCP5_CMD_DST_LO(&desc
) = ccp_addr_lo(&op
->dst
.u
.dma
);
360 CCP5_CMD_DST_HI(&desc
) = ccp_addr_hi(&op
->dst
.u
.dma
);
361 CCP5_CMD_DST_MEM(&desc
) = CCP_MEMTYPE_SYSTEM
;
363 CCP5_CMD_KEY_LO(&desc
) = lower_32_bits(key_addr
);
364 CCP5_CMD_KEY_HI(&desc
) = 0;
365 CCP5_CMD_KEY_MEM(&desc
) = CCP_MEMTYPE_SB
;
366 CCP5_CMD_LSB_ID(&desc
) = op
->sb_ctx
;
368 return ccp5_do_cmd(&desc
, op
->cmd_q
);
371 static int ccp5_perform_sha(struct ccp_op
*op
)
373 struct ccp5_desc desc
;
374 union ccp_function function
;
376 op
->cmd_q
->total_sha_ops
++;
378 /* Zero out all the fields of the command desc */
379 memset(&desc
, 0, Q_DESC_SIZE
);
381 CCP5_CMD_ENGINE(&desc
) = CCP_ENGINE_SHA
;
383 CCP5_CMD_SOC(&desc
) = op
->soc
;
384 CCP5_CMD_IOC(&desc
) = 1;
385 CCP5_CMD_INIT(&desc
) = 1;
386 CCP5_CMD_EOM(&desc
) = op
->eom
;
387 CCP5_CMD_PROT(&desc
) = 0;
390 CCP_SHA_TYPE(&function
) = op
->u
.sha
.type
;
391 CCP5_CMD_FUNCTION(&desc
) = function
.raw
;
393 CCP5_CMD_LEN(&desc
) = op
->src
.u
.dma
.length
;
395 CCP5_CMD_SRC_LO(&desc
) = ccp_addr_lo(&op
->src
.u
.dma
);
396 CCP5_CMD_SRC_HI(&desc
) = ccp_addr_hi(&op
->src
.u
.dma
);
397 CCP5_CMD_SRC_MEM(&desc
) = CCP_MEMTYPE_SYSTEM
;
399 CCP5_CMD_LSB_ID(&desc
) = op
->sb_ctx
;
402 CCP5_CMD_SHA_LO(&desc
) = lower_32_bits(op
->u
.sha
.msg_bits
);
403 CCP5_CMD_SHA_HI(&desc
) = upper_32_bits(op
->u
.sha
.msg_bits
);
405 CCP5_CMD_SHA_LO(&desc
) = 0;
406 CCP5_CMD_SHA_HI(&desc
) = 0;
409 return ccp5_do_cmd(&desc
, op
->cmd_q
);
412 static int ccp5_perform_des3(struct ccp_op
*op
)
414 struct ccp5_desc desc
;
415 union ccp_function function
;
416 u32 key_addr
= op
->sb_key
* LSB_ITEM_SIZE
;
418 op
->cmd_q
->total_3des_ops
++;
420 /* Zero out all the fields of the command desc */
421 memset(&desc
, 0, sizeof(struct ccp5_desc
));
423 CCP5_CMD_ENGINE(&desc
) = CCP_ENGINE_DES3
;
425 CCP5_CMD_SOC(&desc
) = op
->soc
;
426 CCP5_CMD_IOC(&desc
) = 1;
427 CCP5_CMD_INIT(&desc
) = op
->init
;
428 CCP5_CMD_EOM(&desc
) = op
->eom
;
429 CCP5_CMD_PROT(&desc
) = 0;
432 CCP_DES3_ENCRYPT(&function
) = op
->u
.des3
.action
;
433 CCP_DES3_MODE(&function
) = op
->u
.des3
.mode
;
434 CCP_DES3_TYPE(&function
) = op
->u
.des3
.type
;
435 CCP5_CMD_FUNCTION(&desc
) = function
.raw
;
437 CCP5_CMD_LEN(&desc
) = op
->src
.u
.dma
.length
;
439 CCP5_CMD_SRC_LO(&desc
) = ccp_addr_lo(&op
->src
.u
.dma
);
440 CCP5_CMD_SRC_HI(&desc
) = ccp_addr_hi(&op
->src
.u
.dma
);
441 CCP5_CMD_SRC_MEM(&desc
) = CCP_MEMTYPE_SYSTEM
;
443 CCP5_CMD_DST_LO(&desc
) = ccp_addr_lo(&op
->dst
.u
.dma
);
444 CCP5_CMD_DST_HI(&desc
) = ccp_addr_hi(&op
->dst
.u
.dma
);
445 CCP5_CMD_DST_MEM(&desc
) = CCP_MEMTYPE_SYSTEM
;
447 CCP5_CMD_KEY_LO(&desc
) = lower_32_bits(key_addr
);
448 CCP5_CMD_KEY_HI(&desc
) = 0;
449 CCP5_CMD_KEY_MEM(&desc
) = CCP_MEMTYPE_SB
;
450 CCP5_CMD_LSB_ID(&desc
) = op
->sb_ctx
;
452 return ccp5_do_cmd(&desc
, op
->cmd_q
);
455 static int ccp5_perform_rsa(struct ccp_op
*op
)
457 struct ccp5_desc desc
;
458 union ccp_function function
;
460 op
->cmd_q
->total_rsa_ops
++;
462 /* Zero out all the fields of the command desc */
463 memset(&desc
, 0, Q_DESC_SIZE
);
465 CCP5_CMD_ENGINE(&desc
) = CCP_ENGINE_RSA
;
467 CCP5_CMD_SOC(&desc
) = op
->soc
;
468 CCP5_CMD_IOC(&desc
) = 1;
469 CCP5_CMD_INIT(&desc
) = 0;
470 CCP5_CMD_EOM(&desc
) = 1;
471 CCP5_CMD_PROT(&desc
) = 0;
474 CCP_RSA_SIZE(&function
) = (op
->u
.rsa
.mod_size
+ 7) >> 3;
475 CCP5_CMD_FUNCTION(&desc
) = function
.raw
;
477 CCP5_CMD_LEN(&desc
) = op
->u
.rsa
.input_len
;
479 /* Source is from external memory */
480 CCP5_CMD_SRC_LO(&desc
) = ccp_addr_lo(&op
->src
.u
.dma
);
481 CCP5_CMD_SRC_HI(&desc
) = ccp_addr_hi(&op
->src
.u
.dma
);
482 CCP5_CMD_SRC_MEM(&desc
) = CCP_MEMTYPE_SYSTEM
;
484 /* Destination is in external memory */
485 CCP5_CMD_DST_LO(&desc
) = ccp_addr_lo(&op
->dst
.u
.dma
);
486 CCP5_CMD_DST_HI(&desc
) = ccp_addr_hi(&op
->dst
.u
.dma
);
487 CCP5_CMD_DST_MEM(&desc
) = CCP_MEMTYPE_SYSTEM
;
489 /* Key (Exponent) is in external memory */
490 CCP5_CMD_KEY_LO(&desc
) = ccp_addr_lo(&op
->exp
.u
.dma
);
491 CCP5_CMD_KEY_HI(&desc
) = ccp_addr_hi(&op
->exp
.u
.dma
);
492 CCP5_CMD_KEY_MEM(&desc
) = CCP_MEMTYPE_SYSTEM
;
494 return ccp5_do_cmd(&desc
, op
->cmd_q
);
497 static int ccp5_perform_passthru(struct ccp_op
*op
)
499 struct ccp5_desc desc
;
500 union ccp_function function
;
501 struct ccp_dma_info
*saddr
= &op
->src
.u
.dma
;
502 struct ccp_dma_info
*daddr
= &op
->dst
.u
.dma
;
505 op
->cmd_q
->total_pt_ops
++;
507 memset(&desc
, 0, Q_DESC_SIZE
);
509 CCP5_CMD_ENGINE(&desc
) = CCP_ENGINE_PASSTHRU
;
511 CCP5_CMD_SOC(&desc
) = 0;
512 CCP5_CMD_IOC(&desc
) = 1;
513 CCP5_CMD_INIT(&desc
) = 0;
514 CCP5_CMD_EOM(&desc
) = op
->eom
;
515 CCP5_CMD_PROT(&desc
) = 0;
518 CCP_PT_BYTESWAP(&function
) = op
->u
.passthru
.byte_swap
;
519 CCP_PT_BITWISE(&function
) = op
->u
.passthru
.bit_mod
;
520 CCP5_CMD_FUNCTION(&desc
) = function
.raw
;
522 /* Length of source data is always 256 bytes */
523 if (op
->src
.type
== CCP_MEMTYPE_SYSTEM
)
524 CCP5_CMD_LEN(&desc
) = saddr
->length
;
526 CCP5_CMD_LEN(&desc
) = daddr
->length
;
528 if (op
->src
.type
== CCP_MEMTYPE_SYSTEM
) {
529 CCP5_CMD_SRC_LO(&desc
) = ccp_addr_lo(&op
->src
.u
.dma
);
530 CCP5_CMD_SRC_HI(&desc
) = ccp_addr_hi(&op
->src
.u
.dma
);
531 CCP5_CMD_SRC_MEM(&desc
) = CCP_MEMTYPE_SYSTEM
;
533 if (op
->u
.passthru
.bit_mod
!= CCP_PASSTHRU_BITWISE_NOOP
)
534 CCP5_CMD_LSB_ID(&desc
) = op
->sb_key
;
536 u32 key_addr
= op
->src
.u
.sb
* CCP_SB_BYTES
;
538 CCP5_CMD_SRC_LO(&desc
) = lower_32_bits(key_addr
);
539 CCP5_CMD_SRC_HI(&desc
) = 0;
540 CCP5_CMD_SRC_MEM(&desc
) = CCP_MEMTYPE_SB
;
543 if (op
->dst
.type
== CCP_MEMTYPE_SYSTEM
) {
544 CCP5_CMD_DST_LO(&desc
) = ccp_addr_lo(&op
->dst
.u
.dma
);
545 CCP5_CMD_DST_HI(&desc
) = ccp_addr_hi(&op
->dst
.u
.dma
);
546 CCP5_CMD_DST_MEM(&desc
) = CCP_MEMTYPE_SYSTEM
;
548 u32 key_addr
= op
->dst
.u
.sb
* CCP_SB_BYTES
;
550 CCP5_CMD_DST_LO(&desc
) = lower_32_bits(key_addr
);
551 CCP5_CMD_DST_HI(&desc
) = 0;
552 CCP5_CMD_DST_MEM(&desc
) = CCP_MEMTYPE_SB
;
555 return ccp5_do_cmd(&desc
, op
->cmd_q
);
558 static int ccp5_perform_ecc(struct ccp_op
*op
)
560 struct ccp5_desc desc
;
561 union ccp_function function
;
563 op
->cmd_q
->total_ecc_ops
++;
565 /* Zero out all the fields of the command desc */
566 memset(&desc
, 0, Q_DESC_SIZE
);
568 CCP5_CMD_ENGINE(&desc
) = CCP_ENGINE_ECC
;
570 CCP5_CMD_SOC(&desc
) = 0;
571 CCP5_CMD_IOC(&desc
) = 1;
572 CCP5_CMD_INIT(&desc
) = 0;
573 CCP5_CMD_EOM(&desc
) = 1;
574 CCP5_CMD_PROT(&desc
) = 0;
577 function
.ecc
.mode
= op
->u
.ecc
.function
;
578 CCP5_CMD_FUNCTION(&desc
) = function
.raw
;
580 CCP5_CMD_LEN(&desc
) = op
->src
.u
.dma
.length
;
582 CCP5_CMD_SRC_LO(&desc
) = ccp_addr_lo(&op
->src
.u
.dma
);
583 CCP5_CMD_SRC_HI(&desc
) = ccp_addr_hi(&op
->src
.u
.dma
);
584 CCP5_CMD_SRC_MEM(&desc
) = CCP_MEMTYPE_SYSTEM
;
586 CCP5_CMD_DST_LO(&desc
) = ccp_addr_lo(&op
->dst
.u
.dma
);
587 CCP5_CMD_DST_HI(&desc
) = ccp_addr_hi(&op
->dst
.u
.dma
);
588 CCP5_CMD_DST_MEM(&desc
) = CCP_MEMTYPE_SYSTEM
;
590 return ccp5_do_cmd(&desc
, op
->cmd_q
);
593 static int ccp_find_lsb_regions(struct ccp_cmd_queue
*cmd_q
, u64 status
)
595 int q_mask
= 1 << cmd_q
->id
;
599 /* Build a bit mask to know which LSBs this queue has access to.
600 * Don't bother with segment 0 as it has special privileges.
602 for (j
= 1; j
< MAX_LSB_CNT
; j
++) {
604 bitmap_set(cmd_q
->lsbmask
, j
, 1);
605 status
>>= LSB_REGION_WIDTH
;
607 queues
= bitmap_weight(cmd_q
->lsbmask
, MAX_LSB_CNT
);
608 dev_dbg(cmd_q
->ccp
->dev
, "Queue %d can access %d LSB regions\n",
611 return queues
? 0 : -EINVAL
;
614 static int ccp_find_and_assign_lsb_to_q(struct ccp_device
*ccp
,
615 int lsb_cnt
, int n_lsbs
,
616 unsigned long *lsb_pub
)
618 DECLARE_BITMAP(qlsb
, MAX_LSB_CNT
);
624 * If the count of potential LSBs available to a queue matches the
625 * ordinal given to us in lsb_cnt:
626 * Copy the mask of possible LSBs for this queue into "qlsb";
627 * For each bit in qlsb, see if the corresponding bit in the
628 * aggregation mask is set; if so, we have a match.
629 * If we have a match, clear the bit in the aggregation to
630 * mark it as no longer available.
631 * If there is no match, clear the bit in qlsb and keep looking.
633 for (i
= 0; i
< ccp
->cmd_q_count
; i
++) {
634 struct ccp_cmd_queue
*cmd_q
= &ccp
->cmd_q
[i
];
636 qlsb_wgt
= bitmap_weight(cmd_q
->lsbmask
, MAX_LSB_CNT
);
638 if (qlsb_wgt
== lsb_cnt
) {
639 bitmap_copy(qlsb
, cmd_q
->lsbmask
, MAX_LSB_CNT
);
641 bitno
= find_first_bit(qlsb
, MAX_LSB_CNT
);
642 while (bitno
< MAX_LSB_CNT
) {
643 if (test_bit(bitno
, lsb_pub
)) {
644 /* We found an available LSB
645 * that this queue can access
648 bitmap_clear(lsb_pub
, bitno
, 1);
650 "Queue %d gets LSB %d\n",
654 bitmap_clear(qlsb
, bitno
, 1);
655 bitno
= find_first_bit(qlsb
, MAX_LSB_CNT
);
657 if (bitno
>= MAX_LSB_CNT
)
665 /* For each queue, from the most- to least-constrained:
666 * find an LSB that can be assigned to the queue. If there are N queues that
667 * can only use M LSBs, where N > M, fail; otherwise, every queue will get a
668 * dedicated LSB. Remaining LSB regions become a shared resource.
669 * If we have fewer LSBs than queues, all LSB regions become shared resources.
671 static int ccp_assign_lsbs(struct ccp_device
*ccp
)
673 DECLARE_BITMAP(lsb_pub
, MAX_LSB_CNT
);
674 DECLARE_BITMAP(qlsb
, MAX_LSB_CNT
);
680 bitmap_zero(lsb_pub
, MAX_LSB_CNT
);
682 /* Create an aggregate bitmap to get a total count of available LSBs */
683 for (i
= 0; i
< ccp
->cmd_q_count
; i
++)
685 lsb_pub
, ccp
->cmd_q
[i
].lsbmask
,
688 n_lsbs
= bitmap_weight(lsb_pub
, MAX_LSB_CNT
);
690 if (n_lsbs
>= ccp
->cmd_q_count
) {
691 /* We have enough LSBS to give every queue a private LSB.
692 * Brute force search to start with the queues that are more
693 * constrained in LSB choice. When an LSB is privately
694 * assigned, it is removed from the public mask.
695 * This is an ugly N squared algorithm with some optimization.
698 n_lsbs
&& (lsb_cnt
<= MAX_LSB_CNT
);
700 rc
= ccp_find_and_assign_lsb_to_q(ccp
, lsb_cnt
, n_lsbs
,
709 /* What's left of the LSBs, according to the public mask, now become
710 * shared. Any zero bits in the lsb_pub mask represent an LSB region
711 * that can't be used as a shared resource, so mark the LSB slots for
714 bitmap_copy(qlsb
, lsb_pub
, MAX_LSB_CNT
);
716 bitno
= find_first_zero_bit(qlsb
, MAX_LSB_CNT
);
717 while (bitno
< MAX_LSB_CNT
) {
718 bitmap_set(ccp
->lsbmap
, bitno
* LSB_SIZE
, LSB_SIZE
);
719 bitmap_set(qlsb
, bitno
, 1);
720 bitno
= find_first_zero_bit(qlsb
, MAX_LSB_CNT
);
726 static void ccp5_disable_queue_interrupts(struct ccp_device
*ccp
)
730 for (i
= 0; i
< ccp
->cmd_q_count
; i
++)
731 iowrite32(0x0, ccp
->cmd_q
[i
].reg_int_enable
);
734 static void ccp5_enable_queue_interrupts(struct ccp_device
*ccp
)
738 for (i
= 0; i
< ccp
->cmd_q_count
; i
++)
739 iowrite32(SUPPORTED_INTERRUPTS
, ccp
->cmd_q
[i
].reg_int_enable
);
742 static void ccp5_irq_bh(unsigned long data
)
744 struct ccp_device
*ccp
= (struct ccp_device
*)data
;
748 for (i
= 0; i
< ccp
->cmd_q_count
; i
++) {
749 struct ccp_cmd_queue
*cmd_q
= &ccp
->cmd_q
[i
];
751 status
= ioread32(cmd_q
->reg_interrupt_status
);
754 cmd_q
->int_status
= status
;
755 cmd_q
->q_status
= ioread32(cmd_q
->reg_status
);
756 cmd_q
->q_int_status
= ioread32(cmd_q
->reg_int_status
);
758 /* On error, only save the first error value */
759 if ((status
& INT_ERROR
) && !cmd_q
->cmd_error
)
760 cmd_q
->cmd_error
= CMD_Q_ERROR(cmd_q
->q_status
);
764 /* Acknowledge the interrupt and wake the kthread */
765 iowrite32(status
, cmd_q
->reg_interrupt_status
);
766 wake_up_interruptible(&cmd_q
->int_queue
);
769 ccp5_enable_queue_interrupts(ccp
);
772 static irqreturn_t
ccp5_irq_handler(int irq
, void *data
)
774 struct ccp_device
*ccp
= (struct ccp_device
*)data
;
776 ccp5_disable_queue_interrupts(ccp
);
777 ccp
->total_interrupts
++;
778 if (ccp
->use_tasklet
)
779 tasklet_schedule(&ccp
->irq_tasklet
);
781 ccp5_irq_bh((unsigned long)ccp
);
785 static int ccp5_init(struct ccp_device
*ccp
)
787 struct device
*dev
= ccp
->dev
;
788 struct ccp_cmd_queue
*cmd_q
;
789 struct dma_pool
*dma_pool
;
790 char dma_pool_name
[MAX_DMAPOOL_NAME_LEN
];
793 u32 status_lo
, status_hi
;
796 /* Find available queues */
797 qmr
= ioread32(ccp
->io_regs
+ Q_MASK_REG
);
798 for (i
= 0; i
< MAX_HW_QUEUES
; i
++) {
800 if (!(qmr
& (1 << i
)))
803 /* Allocate a dma pool for this queue */
804 snprintf(dma_pool_name
, sizeof(dma_pool_name
), "%s_q%d",
806 dma_pool
= dma_pool_create(dma_pool_name
, dev
,
807 CCP_DMAPOOL_MAX_SIZE
,
808 CCP_DMAPOOL_ALIGN
, 0);
810 dev_err(dev
, "unable to allocate dma pool\n");
814 cmd_q
= &ccp
->cmd_q
[ccp
->cmd_q_count
];
819 cmd_q
->dma_pool
= dma_pool
;
820 mutex_init(&cmd_q
->q_mutex
);
822 /* Page alignment satisfies our needs for N <= 128 */
823 BUILD_BUG_ON(COMMANDS_PER_QUEUE
> 128);
824 cmd_q
->qsize
= Q_SIZE(Q_DESC_SIZE
);
825 cmd_q
->qbase
= dma_zalloc_coherent(dev
, cmd_q
->qsize
,
829 dev_err(dev
, "unable to allocate command queue\n");
835 /* Preset some register values and masks that are queue
838 cmd_q
->reg_control
= ccp
->io_regs
+
839 CMD5_Q_STATUS_INCR
* (i
+ 1);
840 cmd_q
->reg_tail_lo
= cmd_q
->reg_control
+ CMD5_Q_TAIL_LO_BASE
;
841 cmd_q
->reg_head_lo
= cmd_q
->reg_control
+ CMD5_Q_HEAD_LO_BASE
;
842 cmd_q
->reg_int_enable
= cmd_q
->reg_control
+
843 CMD5_Q_INT_ENABLE_BASE
;
844 cmd_q
->reg_interrupt_status
= cmd_q
->reg_control
+
845 CMD5_Q_INTERRUPT_STATUS_BASE
;
846 cmd_q
->reg_status
= cmd_q
->reg_control
+ CMD5_Q_STATUS_BASE
;
847 cmd_q
->reg_int_status
= cmd_q
->reg_control
+
848 CMD5_Q_INT_STATUS_BASE
;
849 cmd_q
->reg_dma_status
= cmd_q
->reg_control
+
850 CMD5_Q_DMA_STATUS_BASE
;
851 cmd_q
->reg_dma_read_status
= cmd_q
->reg_control
+
852 CMD5_Q_DMA_READ_STATUS_BASE
;
853 cmd_q
->reg_dma_write_status
= cmd_q
->reg_control
+
854 CMD5_Q_DMA_WRITE_STATUS_BASE
;
856 init_waitqueue_head(&cmd_q
->int_queue
);
858 dev_dbg(dev
, "queue #%u available\n", i
);
861 if (ccp
->cmd_q_count
== 0) {
862 dev_notice(dev
, "no command queues available\n");
867 /* Turn off the queues and disable interrupts until ready */
868 ccp5_disable_queue_interrupts(ccp
);
869 for (i
= 0; i
< ccp
->cmd_q_count
; i
++) {
870 cmd_q
= &ccp
->cmd_q
[i
];
872 cmd_q
->qcontrol
= 0; /* Start with nothing */
873 iowrite32(cmd_q
->qcontrol
, cmd_q
->reg_control
);
875 ioread32(cmd_q
->reg_int_status
);
876 ioread32(cmd_q
->reg_status
);
878 /* Clear the interrupt status */
879 iowrite32(SUPPORTED_INTERRUPTS
, cmd_q
->reg_interrupt_status
);
882 dev_dbg(dev
, "Requesting an IRQ...\n");
884 ret
= sp_request_ccp_irq(ccp
->sp
, ccp5_irq_handler
, ccp
->name
, ccp
);
886 dev_err(dev
, "unable to allocate an IRQ\n");
889 /* Initialize the ISR tasklet */
890 if (ccp
->use_tasklet
)
891 tasklet_init(&ccp
->irq_tasklet
, ccp5_irq_bh
,
894 dev_dbg(dev
, "Loading LSB map...\n");
895 /* Copy the private LSB mask to the public registers */
896 status_lo
= ioread32(ccp
->io_regs
+ LSB_PRIVATE_MASK_LO_OFFSET
);
897 status_hi
= ioread32(ccp
->io_regs
+ LSB_PRIVATE_MASK_HI_OFFSET
);
898 iowrite32(status_lo
, ccp
->io_regs
+ LSB_PUBLIC_MASK_LO_OFFSET
);
899 iowrite32(status_hi
, ccp
->io_regs
+ LSB_PUBLIC_MASK_HI_OFFSET
);
900 status
= ((u64
)status_hi
<<30) | (u64
)status_lo
;
902 dev_dbg(dev
, "Configuring virtual queues...\n");
903 /* Configure size of each virtual queue accessible to host */
904 for (i
= 0; i
< ccp
->cmd_q_count
; i
++) {
908 cmd_q
= &ccp
->cmd_q
[i
];
910 cmd_q
->qcontrol
&= ~(CMD5_Q_SIZE
<< CMD5_Q_SHIFT
);
911 cmd_q
->qcontrol
|= QUEUE_SIZE_VAL
<< CMD5_Q_SHIFT
;
913 cmd_q
->qdma_tail
= cmd_q
->qbase_dma
;
914 dma_addr_lo
= low_address(cmd_q
->qdma_tail
);
915 iowrite32((u32
)dma_addr_lo
, cmd_q
->reg_tail_lo
);
916 iowrite32((u32
)dma_addr_lo
, cmd_q
->reg_head_lo
);
918 dma_addr_hi
= high_address(cmd_q
->qdma_tail
);
919 cmd_q
->qcontrol
|= (dma_addr_hi
<< 16);
920 iowrite32(cmd_q
->qcontrol
, cmd_q
->reg_control
);
922 /* Find the LSB regions accessible to the queue */
923 ccp_find_lsb_regions(cmd_q
, status
);
924 cmd_q
->lsb
= -1; /* Unassigned value */
927 dev_dbg(dev
, "Assigning LSBs...\n");
928 ret
= ccp_assign_lsbs(ccp
);
930 dev_err(dev
, "Unable to assign LSBs (%d)\n", ret
);
934 /* Optimization: pre-allocate LSB slots for each queue */
935 for (i
= 0; i
< ccp
->cmd_q_count
; i
++) {
936 ccp
->cmd_q
[i
].sb_key
= ccp_lsb_alloc(&ccp
->cmd_q
[i
], 2);
937 ccp
->cmd_q
[i
].sb_ctx
= ccp_lsb_alloc(&ccp
->cmd_q
[i
], 2);
940 dev_dbg(dev
, "Starting threads...\n");
941 /* Create a kthread for each queue */
942 for (i
= 0; i
< ccp
->cmd_q_count
; i
++) {
943 struct task_struct
*kthread
;
945 cmd_q
= &ccp
->cmd_q
[i
];
947 kthread
= kthread_create(ccp_cmd_queue_thread
, cmd_q
,
948 "%s-q%u", ccp
->name
, cmd_q
->id
);
949 if (IS_ERR(kthread
)) {
950 dev_err(dev
, "error creating queue thread (%ld)\n",
952 ret
= PTR_ERR(kthread
);
956 cmd_q
->kthread
= kthread
;
957 wake_up_process(kthread
);
960 dev_dbg(dev
, "Enabling interrupts...\n");
961 ccp5_enable_queue_interrupts(ccp
);
963 dev_dbg(dev
, "Registering device...\n");
964 /* Put this on the unit list to make it available */
967 ret
= ccp_register_rng(ccp
);
971 /* Register the DMA engine support */
972 ret
= ccp_dmaengine_register(ccp
);
976 /* Set up debugfs entries */
977 ccp5_debugfs_setup(ccp
);
982 ccp_unregister_rng(ccp
);
985 for (i
= 0; i
< ccp
->cmd_q_count
; i
++)
986 if (ccp
->cmd_q
[i
].kthread
)
987 kthread_stop(ccp
->cmd_q
[i
].kthread
);
990 sp_free_ccp_irq(ccp
->sp
, ccp
);
993 for (i
= 0; i
< ccp
->cmd_q_count
; i
++)
994 dma_pool_destroy(ccp
->cmd_q
[i
].dma_pool
);
999 static void ccp5_destroy(struct ccp_device
*ccp
)
1001 struct device
*dev
= ccp
->dev
;
1002 struct ccp_cmd_queue
*cmd_q
;
1003 struct ccp_cmd
*cmd
;
1006 /* Unregister the DMA engine */
1007 ccp_dmaengine_unregister(ccp
);
1009 /* Unregister the RNG */
1010 ccp_unregister_rng(ccp
);
1012 /* Remove this device from the list of available units first */
1013 ccp_del_device(ccp
);
1015 /* We're in the process of tearing down the entire driver;
1016 * when all the devices are gone clean up debugfs
1019 ccp5_debugfs_destroy();
1021 /* Disable and clear interrupts */
1022 ccp5_disable_queue_interrupts(ccp
);
1023 for (i
= 0; i
< ccp
->cmd_q_count
; i
++) {
1024 cmd_q
= &ccp
->cmd_q
[i
];
1026 /* Turn off the run bit */
1027 iowrite32(cmd_q
->qcontrol
& ~CMD5_Q_RUN
, cmd_q
->reg_control
);
1029 /* Clear the interrupt status */
1030 iowrite32(SUPPORTED_INTERRUPTS
, cmd_q
->reg_interrupt_status
);
1031 ioread32(cmd_q
->reg_int_status
);
1032 ioread32(cmd_q
->reg_status
);
1035 /* Stop the queue kthreads */
1036 for (i
= 0; i
< ccp
->cmd_q_count
; i
++)
1037 if (ccp
->cmd_q
[i
].kthread
)
1038 kthread_stop(ccp
->cmd_q
[i
].kthread
);
1040 sp_free_ccp_irq(ccp
->sp
, ccp
);
1042 for (i
= 0; i
< ccp
->cmd_q_count
; i
++) {
1043 cmd_q
= &ccp
->cmd_q
[i
];
1044 dma_free_coherent(dev
, cmd_q
->qsize
, cmd_q
->qbase
,
1048 /* Flush the cmd and backlog queue */
1049 while (!list_empty(&ccp
->cmd
)) {
1050 /* Invoke the callback directly with an error code */
1051 cmd
= list_first_entry(&ccp
->cmd
, struct ccp_cmd
, entry
);
1052 list_del(&cmd
->entry
);
1053 cmd
->callback(cmd
->data
, -ENODEV
);
1055 while (!list_empty(&ccp
->backlog
)) {
1056 /* Invoke the callback directly with an error code */
1057 cmd
= list_first_entry(&ccp
->backlog
, struct ccp_cmd
, entry
);
1058 list_del(&cmd
->entry
);
1059 cmd
->callback(cmd
->data
, -ENODEV
);
1063 static void ccp5_config(struct ccp_device
*ccp
)
1066 iowrite32(0x0, ccp
->io_regs
+ CMD5_REQID_CONFIG_OFFSET
);
1069 static void ccp5other_config(struct ccp_device
*ccp
)
1074 /* We own all of the queues on the NTB CCP */
1076 iowrite32(0x00012D57, ccp
->io_regs
+ CMD5_TRNG_CTL_OFFSET
);
1077 iowrite32(0x00000003, ccp
->io_regs
+ CMD5_CONFIG_0_OFFSET
);
1078 for (i
= 0; i
< 12; i
++) {
1079 rnd
= ioread32(ccp
->io_regs
+ TRNG_OUT_REG
);
1080 iowrite32(rnd
, ccp
->io_regs
+ CMD5_AES_MASK_OFFSET
);
1083 iowrite32(0x0000001F, ccp
->io_regs
+ CMD5_QUEUE_MASK_OFFSET
);
1084 iowrite32(0x00005B6D, ccp
->io_regs
+ CMD5_QUEUE_PRIO_OFFSET
);
1085 iowrite32(0x00000000, ccp
->io_regs
+ CMD5_CMD_TIMEOUT_OFFSET
);
1087 iowrite32(0x3FFFFFFF, ccp
->io_regs
+ LSB_PRIVATE_MASK_LO_OFFSET
);
1088 iowrite32(0x000003FF, ccp
->io_regs
+ LSB_PRIVATE_MASK_HI_OFFSET
);
1090 iowrite32(0x00108823, ccp
->io_regs
+ CMD5_CLK_GATE_CTL_OFFSET
);
1095 /* Version 5 adds some function, but is essentially the same as v5 */
1096 static const struct ccp_actions ccp5_actions
= {
1097 .aes
= ccp5_perform_aes
,
1098 .xts_aes
= ccp5_perform_xts_aes
,
1099 .sha
= ccp5_perform_sha
,
1100 .des3
= ccp5_perform_des3
,
1101 .rsa
= ccp5_perform_rsa
,
1102 .passthru
= ccp5_perform_passthru
,
1103 .ecc
= ccp5_perform_ecc
,
1104 .sballoc
= ccp_lsb_alloc
,
1105 .sbfree
= ccp_lsb_free
,
1107 .destroy
= ccp5_destroy
,
1108 .get_free_slots
= ccp5_get_free_slots
,
1111 const struct ccp_vdata ccpv5a
= {
1112 .version
= CCP_VERSION(5, 0),
1113 .setup
= ccp5_config
,
1114 .perform
= &ccp5_actions
,
1116 .rsamax
= CCP5_RSA_MAX_WIDTH
,
1119 const struct ccp_vdata ccpv5b
= {
1120 .version
= CCP_VERSION(5, 0),
1121 .dma_chan_attr
= DMA_PRIVATE
,
1122 .setup
= ccp5other_config
,
1123 .perform
= &ccp5_actions
,
1125 .rsamax
= CCP5_RSA_MAX_WIDTH
,