1/*
2 * AMD Cryptographic Coprocessor (CCP) driver
3 *
4 * Copyright (C) 2016,2017 Advanced Micro Devices, Inc.
5 *
6 * Author: Gary R Hook <gary.hook@amd.com>
7 *
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.
11 */
12
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>
22
23#include "ccp-dev.h"
24
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
30 */
31static u32 ccp_lsb_alloc(struct ccp_cmd_queue *cmd_q, unsigned int count)
32{
33 struct ccp_device *ccp;
34 int start;
35
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,
39 LSB_SIZE,
40 0, count, 0);
41 if (start < LSB_SIZE) {
42 bitmap_set(cmd_q->lsbmap, start, count);
43 return start + cmd_q->lsb * LSB_SIZE;
44 }
45 }
46
47 /* No joy; try to get an entry from the shared blocks */
48 ccp = cmd_q->ccp;
49 for (;;) {
50 mutex_lock(&ccp->sb_mutex);
51
52 start = (u32)bitmap_find_next_zero_area(ccp->lsbmap,
53 MAX_LSB_CNT * LSB_SIZE,
54 0,
55 count, 0);
56 if (start <= MAX_LSB_CNT * LSB_SIZE) {
57 bitmap_set(ccp->lsbmap, start, count);
58
59 mutex_unlock(&ccp->sb_mutex);
60 return start;
61 }
62
63 ccp->sb_avail = 0;
64
65 mutex_unlock(&ccp->sb_mutex);
66
67 /* Wait for KSB entries to become available */
68 if (wait_event_interruptible(ccp->sb_queue, ccp->sb_avail))
69 return 0;
70 }
71}
72
73/* Free a number of LSB slots from the bitmap, starting at
74 * the indicated starting slot number.
75 */
76static void ccp_lsb_free(struct ccp_cmd_queue *cmd_q, unsigned int start,
77 unsigned int count)
78{
79 if (!start)
80 return;
81
82 if (cmd_q->lsb == start) {
83 /* An entry from the private LSB */
84 bitmap_clear(cmd_q->lsbmap, start, count);
85 } else {
86 /* From the shared LSBs */
87 struct ccp_device *ccp = cmd_q->ccp;
88
89 mutex_lock(&ccp->sb_mutex);
90 bitmap_clear(ccp->lsbmap, start, count);
91 ccp->sb_avail = 1;
92 mutex_unlock(&ccp->sb_mutex);
93 wake_up_interruptible_all(&ccp->sb_queue);
94 }
95}
96
97/* CCP version 5: Union to define the function field (cmd_reg1/dword0) */
98union ccp_function {
99 struct {
100 u16 size:7;
101 u16 encrypt:1;
102 u16 mode:5;
103 u16 type:2;
104 } aes;
105 struct {
106 u16 size:7;
107 u16 encrypt:1;
108 u16 rsvd:5;
109 u16 type:2;
110 } aes_xts;
111 struct {
112 u16 size:7;
113 u16 encrypt:1;
114 u16 mode:5;
115 u16 type:2;
116 } des3;
117 struct {
118 u16 rsvd1:10;
119 u16 type:4;
120 u16 rsvd2:1;
121 } sha;
122 struct {
123 u16 mode:3;
124 u16 size:12;
125 } rsa;
126 struct {
127 u16 byteswap:2;
128 u16 bitwise:3;
129 u16 reflect:2;
130 u16 rsvd:8;
131 } pt;
132 struct {
133 u16 rsvd:13;
134 } zlib;
135 struct {
136 u16 size:10;
137 u16 type:2;
138 u16 mode:3;
139 } ecc;
140 u16 raw;
141};
142
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)
160
161/* Word 0 */
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)
170
171/* Word 1 */
172#define CCP5_CMD_DW1(p) ((p)->length)
173#define CCP5_CMD_LEN(p) (CCP5_CMD_DW1(p))
174
175/* Word 2 */
176#define CCP5_CMD_DW2(p) ((p)->src_lo)
177#define CCP5_CMD_SRC_LO(p) (CCP5_CMD_DW2(p))
178
179/* Word 3 */
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)
185
186/* Words 4/5 */
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)
195
196/* Word 6/7 */
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)
202
203static inline u32 low_address(unsigned long addr)
204{
205 return (u64)addr & 0x0ffffffff;
206}
207
208static inline u32 high_address(unsigned long addr)
209{
210 return ((u64)addr >> 32) & 0x00000ffff;
211}
212
213static unsigned int ccp5_get_free_slots(struct ccp_cmd_queue *cmd_q)
214{
215 unsigned int head_idx, n;
216 u32 head_lo, queue_start;
217
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);
221
222 n = head_idx + COMMANDS_PER_QUEUE - cmd_q->qidx - 1;
223
224 return n % COMMANDS_PER_QUEUE; /* Always one unused spot */
225}
226
227static int ccp5_do_cmd(struct ccp5_desc *desc,
228 struct ccp_cmd_queue *cmd_q)
229{
230 u32 *mP;
231 __le32 *dP;
232 u32 tail;
233 int i;
234 int ret = 0;
235
236 cmd_q->total_ops++;
237
238 if (CCP5_CMD_SOC(desc)) {
239 CCP5_CMD_IOC(desc) = 1;
240 CCP5_CMD_SOC(desc) = 0;
241 }
242 mutex_lock(&cmd_q->q_mutex);
243
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 */
248
249 cmd_q->qidx = (cmd_q->qidx + 1) % COMMANDS_PER_QUEUE;
250
251 /* The data used by this command must be flushed to memory */
252 wmb();
253
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);
257
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);
261
262 if (CCP5_CMD_IOC(desc)) {
263 /* Wait for the job to complete */
264 ret = wait_event_interruptible(cmd_q->int_queue,
265 cmd_q->int_rcvd);
266 if (ret || cmd_q->cmd_error) {
267 /* Log the error and flush the queue by
268 * moving the head pointer
269 */
270 if (cmd_q->cmd_error)
271 ccp_log_error(cmd_q->ccp,
272 cmd_q->cmd_error);
273 iowrite32(tail, cmd_q->reg_head_lo);
274 if (!ret)
275 ret = -EIO;
276 }
277 cmd_q->int_rcvd = 0;
278 }
279
280 return ret;
281}
282
283static int ccp5_perform_aes(struct ccp_op *op)
284{
285 struct ccp5_desc desc;
286 union ccp_function function;
287 u32 key_addr = op->sb_key * LSB_ITEM_SIZE;
288
289 op->cmd_q->total_aes_ops++;
290
291 /* Zero out all the fields of the command desc */
292 memset(&desc, 0, Q_DESC_SIZE);
293
294 CCP5_CMD_ENGINE(&desc) = CCP_ENGINE_AES;
295
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;
301
302 function.raw = 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;
307
308 CCP5_CMD_FUNCTION(&desc) = function.raw;
309
310 CCP5_CMD_LEN(&desc) = op->src.u.dma.length;
311
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;
315
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;
319
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;
324
325 return ccp5_do_cmd(&desc, op->cmd_q);
326}
327
328static int ccp5_perform_xts_aes(struct ccp_op *op)
329{
330 struct ccp5_desc desc;
331 union ccp_function function;
332 u32 key_addr = op->sb_key * LSB_ITEM_SIZE;
333
334 op->cmd_q->total_xts_aes_ops++;
335
336 /* Zero out all the fields of the command desc */
337 memset(&desc, 0, Q_DESC_SIZE);
338
339 CCP5_CMD_ENGINE(&desc) = CCP_ENGINE_XTS_AES_128;
340
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;
346
347 function.raw = 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;
352
353 CCP5_CMD_LEN(&desc) = op->src.u.dma.length;
354
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;
358
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;
362
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;
367
368 return ccp5_do_cmd(&desc, op->cmd_q);
369}
370
371static int ccp5_perform_sha(struct ccp_op *op)
372{
373 struct ccp5_desc desc;
374 union ccp_function function;
375
376 op->cmd_q->total_sha_ops++;
377
378 /* Zero out all the fields of the command desc */
379 memset(&desc, 0, Q_DESC_SIZE);
380
381 CCP5_CMD_ENGINE(&desc) = CCP_ENGINE_SHA;
382
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;
388
389 function.raw = 0;
390 CCP_SHA_TYPE(&function) = op->u.sha.type;
391 CCP5_CMD_FUNCTION(&desc) = function.raw;
392
393 CCP5_CMD_LEN(&desc) = op->src.u.dma.length;
394
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;
398
399 CCP5_CMD_LSB_ID(&desc) = op->sb_ctx;
400
401 if (op->eom) {
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);
404 } else {
405 CCP5_CMD_SHA_LO(&desc) = 0;
406 CCP5_CMD_SHA_HI(&desc) = 0;
407 }
408
409 return ccp5_do_cmd(&desc, op->cmd_q);
410}
411
412static int ccp5_perform_des3(struct ccp_op *op)
413{
414 struct ccp5_desc desc;
415 union ccp_function function;
416 u32 key_addr = op->sb_key * LSB_ITEM_SIZE;
417
418 op->cmd_q->total_3des_ops++;
419
420 /* Zero out all the fields of the command desc */
421 memset(&desc, 0, sizeof(struct ccp5_desc));
422
423 CCP5_CMD_ENGINE(&desc) = CCP_ENGINE_DES3;
424
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;
430
431 function.raw = 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;
436
437 CCP5_CMD_LEN(&desc) = op->src.u.dma.length;
438
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;
442
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;
446
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;
451
452 return ccp5_do_cmd(&desc, op->cmd_q);
453}
454
455static int ccp5_perform_rsa(struct ccp_op *op)
456{
457 struct ccp5_desc desc;
458 union ccp_function function;
459
460 op->cmd_q->total_rsa_ops++;
461
462 /* Zero out all the fields of the command desc */
463 memset(&desc, 0, Q_DESC_SIZE);
464
465 CCP5_CMD_ENGINE(&desc) = CCP_ENGINE_RSA;
466
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;
472
473 function.raw = 0;
474 CCP_RSA_SIZE(&function) = (op->u.rsa.mod_size + 7) >> 3;
475 CCP5_CMD_FUNCTION(&desc) = function.raw;
476
477 CCP5_CMD_LEN(&desc) = op->u.rsa.input_len;
478
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;
483
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;
488
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;
493
494 return ccp5_do_cmd(&desc, op->cmd_q);
495}
496
497static int ccp5_perform_passthru(struct ccp_op *op)
498{
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;
503
504
505 op->cmd_q->total_pt_ops++;
506
507 memset(&desc, 0, Q_DESC_SIZE);
508
509 CCP5_CMD_ENGINE(&desc) = CCP_ENGINE_PASSTHRU;
510
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;
516
517 function.raw = 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;
521
522 /* Length of source data is always 256 bytes */
523 if (op->src.type == CCP_MEMTYPE_SYSTEM)
524 CCP5_CMD_LEN(&desc) = saddr->length;
525 else
526 CCP5_CMD_LEN(&desc) = daddr->length;
527
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;
532
533 if (op->u.passthru.bit_mod != CCP_PASSTHRU_BITWISE_NOOP)
534 CCP5_CMD_LSB_ID(&desc) = op->sb_key;
535 } else {
536 u32 key_addr = op->src.u.sb * CCP_SB_BYTES;
537
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;
541 }
542
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;
547 } else {
548 u32 key_addr = op->dst.u.sb * CCP_SB_BYTES;
549
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;
553 }
554
555 return ccp5_do_cmd(&desc, op->cmd_q);
556}
557
558static int ccp5_perform_ecc(struct ccp_op *op)
559{
560 struct ccp5_desc desc;
561 union ccp_function function;
562
563 op->cmd_q->total_ecc_ops++;
564
565 /* Zero out all the fields of the command desc */
566 memset(&desc, 0, Q_DESC_SIZE);
567
568 CCP5_CMD_ENGINE(&desc) = CCP_ENGINE_ECC;
569
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;
575
576 function.raw = 0;
577 function.ecc.mode = op->u.ecc.function;
578 CCP5_CMD_FUNCTION(&desc) = function.raw;
579
580 CCP5_CMD_LEN(&desc) = op->src.u.dma.length;
581
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;
585
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;
589
590 return ccp5_do_cmd(&desc, op->cmd_q);
591}
592
593static int ccp_find_lsb_regions(struct ccp_cmd_queue *cmd_q, u64 status)
594{
595 int q_mask = 1 << cmd_q->id;
596 int queues = 0;
597 int j;
598
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.
601 */
602 for (j = 1; j < MAX_LSB_CNT; j++) {
603 if (status & q_mask)
604 bitmap_set(cmd_q->lsbmask, j, 1);
605 status >>= LSB_REGION_WIDTH;
606 }
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",
609 cmd_q->id, queues);
610
611 return queues ? 0 : -EINVAL;
612}
613
614static int ccp_find_and_assign_lsb_to_q(struct ccp_device *ccp,
615 int lsb_cnt, int n_lsbs,
616 unsigned long *lsb_pub)
617{
618 DECLARE_BITMAP(qlsb, MAX_LSB_CNT);
619 int bitno;
620 int qlsb_wgt;
621 int i;
622
623 /* For each queue:
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.
632 */
633 for (i = 0; i < ccp->cmd_q_count; i++) {
634 struct ccp_cmd_queue *cmd_q = &ccp->cmd_q[i];
635
636 qlsb_wgt = bitmap_weight(cmd_q->lsbmask, MAX_LSB_CNT);
637
638 if (qlsb_wgt == lsb_cnt) {
639 bitmap_copy(qlsb, cmd_q->lsbmask, MAX_LSB_CNT);
640
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
646 */
647 cmd_q->lsb = bitno;
648 bitmap_clear(lsb_pub, bitno, 1);
649 dev_dbg(ccp->dev,
650 "Queue %d gets LSB %d\n",
651 i, bitno);
652 break;
653 }
654 bitmap_clear(qlsb, bitno, 1);
655 bitno = find_first_bit(qlsb, MAX_LSB_CNT);
656 }
657 if (bitno >= MAX_LSB_CNT)
658 return -EINVAL;
659 n_lsbs--;
660 }
661 }
662 return n_lsbs;
663}
664
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.
670 */
671static int ccp_assign_lsbs(struct ccp_device *ccp)
672{
673 DECLARE_BITMAP(lsb_pub, MAX_LSB_CNT);
674 DECLARE_BITMAP(qlsb, MAX_LSB_CNT);
675 int n_lsbs = 0;
676 int bitno;
677 int i, lsb_cnt;
678 int rc = 0;
679
680 bitmap_zero(lsb_pub, MAX_LSB_CNT);
681
682 /* Create an aggregate bitmap to get a total count of available LSBs */
683 for (i = 0; i < ccp->cmd_q_count; i++)
684 bitmap_or(lsb_pub,
685 lsb_pub, ccp->cmd_q[i].lsbmask,
686 MAX_LSB_CNT);
687
688 n_lsbs = bitmap_weight(lsb_pub, MAX_LSB_CNT);
689
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.
696 */
697 for (lsb_cnt = 1;
698 n_lsbs && (lsb_cnt <= MAX_LSB_CNT);
699 lsb_cnt++) {
700 rc = ccp_find_and_assign_lsb_to_q(ccp, lsb_cnt, n_lsbs,
701 lsb_pub);
702 if (rc < 0)
703 return -EINVAL;
704 n_lsbs = rc;
705 }
706 }
707
708 rc = 0;
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
712 * them as "in use".
713 */
714 bitmap_copy(qlsb, lsb_pub, MAX_LSB_CNT);
715
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);
721 }
722
723 return rc;
724}
725
726static void ccp5_disable_queue_interrupts(struct ccp_device *ccp)
727{
728 unsigned int i;
729
730 for (i = 0; i < ccp->cmd_q_count; i++)
731 iowrite32(0x0, ccp->cmd_q[i].reg_int_enable);
732}
733
734static void ccp5_enable_queue_interrupts(struct ccp_device *ccp)
735{
736 unsigned int i;
737
738 for (i = 0; i < ccp->cmd_q_count; i++)
739 iowrite32(SUPPORTED_INTERRUPTS, ccp->cmd_q[i].reg_int_enable);
740}
741
742static void ccp5_irq_bh(unsigned long data)
743{
744 struct ccp_device *ccp = (struct ccp_device *)data;
745 u32 status;
746 unsigned int i;
747
748 for (i = 0; i < ccp->cmd_q_count; i++) {
749 struct ccp_cmd_queue *cmd_q = &ccp->cmd_q[i];
750
751 status = ioread32(cmd_q->reg_interrupt_status);
752
753 if (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);
757
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);
761
762 cmd_q->int_rcvd = 1;
763
764 /* Acknowledge the interrupt and wake the kthread */
765 iowrite32(status, cmd_q->reg_interrupt_status);
766 wake_up_interruptible(&cmd_q->int_queue);
767 }
768 }
769 ccp5_enable_queue_interrupts(ccp);
770}
771
772static irqreturn_t ccp5_irq_handler(int irq, void *data)
773{
774 struct ccp_device *ccp = (struct ccp_device *)data;
775
776 ccp5_disable_queue_interrupts(ccp);
777 ccp->total_interrupts++;
778 if (ccp->use_tasklet)
779 tasklet_schedule(&ccp->irq_tasklet);
780 else
781 ccp5_irq_bh((unsigned long)ccp);
782 return IRQ_HANDLED;
783}
784
785static int ccp5_init(struct ccp_device *ccp)
786{
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];
791 unsigned int qmr, i;
792 u64 status;
793 u32 status_lo, status_hi;
794 int ret;
795
796 /* Find available queues */
797 qmr = ioread32(ccp->io_regs + Q_MASK_REG);
798 for (i = 0; i < MAX_HW_QUEUES; i++) {
799
800 if (!(qmr & (1 << i)))
801 continue;
802
803 /* Allocate a dma pool for this queue */
804 snprintf(dma_pool_name, sizeof(dma_pool_name), "%s_q%d",
805 ccp->name, i);
806 dma_pool = dma_pool_create(dma_pool_name, dev,
807 CCP_DMAPOOL_MAX_SIZE,
808 CCP_DMAPOOL_ALIGN, 0);
809 if (!dma_pool) {
810 dev_err(dev, "unable to allocate dma pool\n");
811 ret = -ENOMEM;
812 }
813
814 cmd_q = &ccp->cmd_q[ccp->cmd_q_count];
815 ccp->cmd_q_count++;
816
817 cmd_q->ccp = ccp;
818 cmd_q->id = i;
819 cmd_q->dma_pool = dma_pool;
820 mutex_init(&cmd_q->q_mutex);
821
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_alloc_coherent(dev, cmd_q->qsize,
826 &cmd_q->qbase_dma,
827 GFP_KERNEL);
828 if (!cmd_q->qbase) {
829 dev_err(dev, "unable to allocate command queue\n");
830 ret = -ENOMEM;
831 goto e_pool;
832 }
833
834 cmd_q->qidx = 0;
835 /* Preset some register values and masks that are queue
836 * number dependent
837 */
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;
855
856 init_waitqueue_head(&cmd_q->int_queue);
857
858 dev_dbg(dev, "queue #%u available\n", i);
859 }
860
861 if (ccp->cmd_q_count == 0) {
862 dev_notice(dev, "no command queues available\n");
863 ret = -EIO;
864 goto e_pool;
865 }
866
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];
871
872 cmd_q->qcontrol = 0; /* Start with nothing */
873 iowrite32(cmd_q->qcontrol, cmd_q->reg_control);
874
875 ioread32(cmd_q->reg_int_status);
876 ioread32(cmd_q->reg_status);
877
878 /* Clear the interrupt status */
879 iowrite32(SUPPORTED_INTERRUPTS, cmd_q->reg_interrupt_status);
880 }
881
882 dev_dbg(dev, "Requesting an IRQ...\n");
883 /* Request an irq */
884 ret = sp_request_ccp_irq(ccp->sp, ccp5_irq_handler, ccp->name, ccp);
885 if (ret) {
886 dev_err(dev, "unable to allocate an IRQ\n");
887 goto e_pool;
888 }
889 /* Initialize the ISR tasklet */
890 if (ccp->use_tasklet)
891 tasklet_init(&ccp->irq_tasklet, ccp5_irq_bh,
892 (unsigned long)ccp);
893
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;
901
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++) {
905 u32 dma_addr_lo;
906 u32 dma_addr_hi;
907
908 cmd_q = &ccp->cmd_q[i];
909
910 cmd_q->qcontrol &= ~(CMD5_Q_SIZE << CMD5_Q_SHIFT);
911 cmd_q->qcontrol |= QUEUE_SIZE_VAL << CMD5_Q_SHIFT;
912
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);
917
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);
921
922 /* Find the LSB regions accessible to the queue */
923 ccp_find_lsb_regions(cmd_q, status);
924 cmd_q->lsb = -1; /* Unassigned value */
925 }
926
927 dev_dbg(dev, "Assigning LSBs...\n");
928 ret = ccp_assign_lsbs(ccp);
929 if (ret) {
930 dev_err(dev, "Unable to assign LSBs (%d)\n", ret);
931 goto e_irq;
932 }
933
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);
938 }
939
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;
944
945 cmd_q = &ccp->cmd_q[i];
946
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",
951 PTR_ERR(kthread));
952 ret = PTR_ERR(kthread);
953 goto e_kthread;
954 }
955
956 cmd_q->kthread = kthread;
957 wake_up_process(kthread);
958 }
959
960 dev_dbg(dev, "Enabling interrupts...\n");
961 ccp5_enable_queue_interrupts(ccp);
962
963 dev_dbg(dev, "Registering device...\n");
964 /* Put this on the unit list to make it available */
965 ccp_add_device(ccp);
966
967 ret = ccp_register_rng(ccp);
968 if (ret)
969 goto e_kthread;
970
971 /* Register the DMA engine support */
972 ret = ccp_dmaengine_register(ccp);
973 if (ret)
974 goto e_hwrng;
975
976 /* Set up debugfs entries */
977 ccp5_debugfs_setup(ccp);
978
979 return 0;
980
981e_hwrng:
982 ccp_unregister_rng(ccp);
983
984e_kthread:
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);
988
989e_irq:
990 sp_free_ccp_irq(ccp->sp, ccp);
991
992e_pool:
993 for (i = 0; i < ccp->cmd_q_count; i++)
994 dma_pool_destroy(ccp->cmd_q[i].dma_pool);
995
996 return ret;
997}
998
999static void ccp5_destroy(struct ccp_device *ccp)
1000{
1001 struct device *dev = ccp->dev;
1002 struct ccp_cmd_queue *cmd_q;
1003 struct ccp_cmd *cmd;
1004 unsigned int i;
1005
1006 /* Unregister the DMA engine */
1007 ccp_dmaengine_unregister(ccp);
1008
1009 /* Unregister the RNG */
1010 ccp_unregister_rng(ccp);
1011
1012 /* Remove this device from the list of available units first */
1013 ccp_del_device(ccp);
1014
1015 /* We're in the process of tearing down the entire driver;
1016 * when all the devices are gone clean up debugfs
1017 */
1018 if (ccp_present())
1019 ccp5_debugfs_destroy();
1020
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];
1025
1026 /* Turn off the run bit */
1027 iowrite32(cmd_q->qcontrol & ~CMD5_Q_RUN, cmd_q->reg_control);
1028
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);
1033 }
1034
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);
1039
1040 sp_free_ccp_irq(ccp->sp, ccp);
1041
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,
1045 cmd_q->qbase_dma);
1046 }
1047
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);
1054 }
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);
1060 }
1061}
1062
1063static void ccp5_config(struct ccp_device *ccp)
1064{
1065 /* Public side */
1066 iowrite32(0x0, ccp->io_regs + CMD5_REQID_CONFIG_OFFSET);
1067}
1068
1069static void ccp5other_config(struct ccp_device *ccp)
1070{
1071 int i;
1072 u32 rnd;
1073
1074 /* We own all of the queues on the NTB CCP */
1075
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);
1081 }
1082
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);
1086
1087 iowrite32(0x3FFFFFFF, ccp->io_regs + LSB_PRIVATE_MASK_LO_OFFSET);
1088 iowrite32(0x000003FF, ccp->io_regs + LSB_PRIVATE_MASK_HI_OFFSET);
1089
1090 iowrite32(0x00108823, ccp->io_regs + CMD5_CLK_GATE_CTL_OFFSET);
1091
1092 ccp5_config(ccp);
1093}
1094
1095/* Version 5 adds some function, but is essentially the same as v5 */
1096static 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,
1106 .init = ccp5_init,
1107 .destroy = ccp5_destroy,
1108 .get_free_slots = ccp5_get_free_slots,
1109};
1110
1111const struct ccp_vdata ccpv5a = {
1112 .version = CCP_VERSION(5, 0),
1113 .setup = ccp5_config,
1114 .perform = &ccp5_actions,
1115 .offset = 0x0,
1116 .rsamax = CCP5_RSA_MAX_WIDTH,
1117};
1118
1119const struct ccp_vdata ccpv5b = {
1120 .version = CCP_VERSION(5, 0),
1121 .dma_chan_attr = DMA_PRIVATE,
1122 .setup = ccp5other_config,
1123 .perform = &ccp5_actions,
1124 .offset = 0x0,
1125 .rsamax = CCP5_RSA_MAX_WIDTH,
1126};
1127