1 // SPDX-License-Identifier: GPL-2.0-only
2 /* n2_core.c: Niagara2 Stream Processing Unit (SPU) crypto support.
4 * Copyright (C) 2010, 2011 David S. Miller <davem@davemloft.net>
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/kernel.h>
10 #include <linux/module.h>
12 #include <linux/of_device.h>
13 #include <linux/cpumask.h>
14 #include <linux/slab.h>
15 #include <linux/interrupt.h>
16 #include <linux/crypto.h>
17 #include <crypto/md5.h>
18 #include <crypto/sha.h>
19 #include <crypto/aes.h>
20 #include <crypto/des.h>
21 #include <linux/mutex.h>
22 #include <linux/delay.h>
23 #include <linux/sched.h>
25 #include <crypto/internal/hash.h>
26 #include <crypto/scatterwalk.h>
27 #include <crypto/algapi.h>
29 #include <asm/hypervisor.h>
30 #include <asm/mdesc.h>
34 #define DRV_MODULE_NAME "n2_crypto"
35 #define DRV_MODULE_VERSION "0.2"
36 #define DRV_MODULE_RELDATE "July 28, 2011"
38 static const char version[] =
39 DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
41 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
42 MODULE_DESCRIPTION("Niagara2 Crypto driver");
43 MODULE_LICENSE("GPL");
44 MODULE_VERSION(DRV_MODULE_VERSION);
46 #define N2_CRA_PRIORITY 200
48 static DEFINE_MUTEX(spu_lock);
52 unsigned long qhandle;
59 struct list_head jobs;
66 struct list_head list;
70 struct spu_queue *queue;
74 static struct spu_queue **cpu_to_cwq;
75 static struct spu_queue **cpu_to_mau;
77 static unsigned long spu_next_offset(struct spu_queue *q, unsigned long off)
79 if (q->q_type == HV_NCS_QTYPE_MAU) {
80 off += MAU_ENTRY_SIZE;
81 if (off == (MAU_ENTRY_SIZE * MAU_NUM_ENTRIES))
84 off += CWQ_ENTRY_SIZE;
85 if (off == (CWQ_ENTRY_SIZE * CWQ_NUM_ENTRIES))
91 struct n2_request_common {
92 struct list_head entry;
95 #define OFFSET_NOT_RUNNING (~(unsigned int)0)
97 /* An async job request records the final tail value it used in
98 * n2_request_common->offset, test to see if that offset is in
99 * the range old_head, new_head, inclusive.
101 static inline bool job_finished(struct spu_queue *q, unsigned int offset,
102 unsigned long old_head, unsigned long new_head)
104 if (old_head <= new_head) {
105 if (offset > old_head && offset <= new_head)
108 if (offset > old_head || offset <= new_head)
114 /* When the HEAD marker is unequal to the actual HEAD, we get
115 * a virtual device INO interrupt. We should process the
116 * completed CWQ entries and adjust the HEAD marker to clear
119 static irqreturn_t cwq_intr(int irq, void *dev_id)
121 unsigned long off, new_head, hv_ret;
122 struct spu_queue *q = dev_id;
124 pr_err("CPU[%d]: Got CWQ interrupt for qhdl[%lx]\n",
125 smp_processor_id(), q->qhandle);
129 hv_ret = sun4v_ncs_gethead(q->qhandle, &new_head);
131 pr_err("CPU[%d]: CWQ gethead[%lx] hv_ret[%lu]\n",
132 smp_processor_id(), new_head, hv_ret);
134 for (off = q->head; off != new_head; off = spu_next_offset(q, off)) {
138 hv_ret = sun4v_ncs_sethead_marker(q->qhandle, new_head);
139 if (hv_ret == HV_EOK)
142 spin_unlock(&q->lock);
147 static irqreturn_t mau_intr(int irq, void *dev_id)
149 struct spu_queue *q = dev_id;
150 unsigned long head, hv_ret;
154 pr_err("CPU[%d]: Got MAU interrupt for qhdl[%lx]\n",
155 smp_processor_id(), q->qhandle);
157 hv_ret = sun4v_ncs_gethead(q->qhandle, &head);
159 pr_err("CPU[%d]: MAU gethead[%lx] hv_ret[%lu]\n",
160 smp_processor_id(), head, hv_ret);
162 sun4v_ncs_sethead_marker(q->qhandle, head);
164 spin_unlock(&q->lock);
169 static void *spu_queue_next(struct spu_queue *q, void *cur)
171 return q->q + spu_next_offset(q, cur - q->q);
174 static int spu_queue_num_free(struct spu_queue *q)
176 unsigned long head = q->head;
177 unsigned long tail = q->tail;
178 unsigned long end = (CWQ_ENTRY_SIZE * CWQ_NUM_ENTRIES);
184 diff = (end - tail) + head;
186 return (diff / CWQ_ENTRY_SIZE) - 1;
189 static void *spu_queue_alloc(struct spu_queue *q, int num_entries)
191 int avail = spu_queue_num_free(q);
193 if (avail >= num_entries)
194 return q->q + q->tail;
199 static unsigned long spu_queue_submit(struct spu_queue *q, void *last)
201 unsigned long hv_ret, new_tail;
203 new_tail = spu_next_offset(q, last - q->q);
205 hv_ret = sun4v_ncs_settail(q->qhandle, new_tail);
206 if (hv_ret == HV_EOK)
211 static u64 control_word_base(unsigned int len, unsigned int hmac_key_len,
212 int enc_type, int auth_type,
213 unsigned int hash_len,
214 bool sfas, bool sob, bool eob, bool encrypt,
217 u64 word = (len - 1) & CONTROL_LEN;
219 word |= ((u64) opcode << CONTROL_OPCODE_SHIFT);
220 word |= ((u64) enc_type << CONTROL_ENC_TYPE_SHIFT);
221 word |= ((u64) auth_type << CONTROL_AUTH_TYPE_SHIFT);
223 word |= CONTROL_STORE_FINAL_AUTH_STATE;
225 word |= CONTROL_START_OF_BLOCK;
227 word |= CONTROL_END_OF_BLOCK;
229 word |= CONTROL_ENCRYPT;
231 word |= ((u64) (hmac_key_len - 1)) << CONTROL_HMAC_KEY_LEN_SHIFT;
233 word |= ((u64) (hash_len - 1)) << CONTROL_HASH_LEN_SHIFT;
239 static inline bool n2_should_run_async(struct spu_queue *qp, int this_len)
241 if (this_len >= 64 ||
242 qp->head != qp->tail)
248 struct n2_ahash_alg {
249 struct list_head entry;
251 const u32 *hash_init;
256 struct ahash_alg alg;
259 static inline struct n2_ahash_alg *n2_ahash_alg(struct crypto_tfm *tfm)
261 struct crypto_alg *alg = tfm->__crt_alg;
262 struct ahash_alg *ahash_alg;
264 ahash_alg = container_of(alg, struct ahash_alg, halg.base);
266 return container_of(ahash_alg, struct n2_ahash_alg, alg);
270 const char *child_alg;
271 struct n2_ahash_alg derived;
274 static inline struct n2_hmac_alg *n2_hmac_alg(struct crypto_tfm *tfm)
276 struct crypto_alg *alg = tfm->__crt_alg;
277 struct ahash_alg *ahash_alg;
279 ahash_alg = container_of(alg, struct ahash_alg, halg.base);
281 return container_of(ahash_alg, struct n2_hmac_alg, derived.alg);
285 struct crypto_ahash *fallback_tfm;
288 #define N2_HASH_KEY_MAX 32 /* HW limit for all HMAC requests */
291 struct n2_hash_ctx base;
293 struct crypto_shash *child_shash;
296 unsigned char hash_key[N2_HASH_KEY_MAX];
299 struct n2_hash_req_ctx {
301 struct md5_state md5;
302 struct sha1_state sha1;
303 struct sha256_state sha256;
306 struct ahash_request fallback_req;
309 static int n2_hash_async_init(struct ahash_request *req)
311 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
312 struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
313 struct n2_hash_ctx *ctx = crypto_ahash_ctx(tfm);
315 ahash_request_set_tfm(&rctx->fallback_req, ctx->fallback_tfm);
316 rctx->fallback_req.base.flags = req->base.flags & CRYPTO_TFM_REQ_MAY_SLEEP;
318 return crypto_ahash_init(&rctx->fallback_req);
321 static int n2_hash_async_update(struct ahash_request *req)
323 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
324 struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
325 struct n2_hash_ctx *ctx = crypto_ahash_ctx(tfm);
327 ahash_request_set_tfm(&rctx->fallback_req, ctx->fallback_tfm);
328 rctx->fallback_req.base.flags = req->base.flags & CRYPTO_TFM_REQ_MAY_SLEEP;
329 rctx->fallback_req.nbytes = req->nbytes;
330 rctx->fallback_req.src = req->src;
332 return crypto_ahash_update(&rctx->fallback_req);
335 static int n2_hash_async_final(struct ahash_request *req)
337 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
338 struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
339 struct n2_hash_ctx *ctx = crypto_ahash_ctx(tfm);
341 ahash_request_set_tfm(&rctx->fallback_req, ctx->fallback_tfm);
342 rctx->fallback_req.base.flags = req->base.flags & CRYPTO_TFM_REQ_MAY_SLEEP;
343 rctx->fallback_req.result = req->result;
345 return crypto_ahash_final(&rctx->fallback_req);
348 static int n2_hash_async_finup(struct ahash_request *req)
350 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
351 struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
352 struct n2_hash_ctx *ctx = crypto_ahash_ctx(tfm);
354 ahash_request_set_tfm(&rctx->fallback_req, ctx->fallback_tfm);
355 rctx->fallback_req.base.flags = req->base.flags & CRYPTO_TFM_REQ_MAY_SLEEP;
356 rctx->fallback_req.nbytes = req->nbytes;
357 rctx->fallback_req.src = req->src;
358 rctx->fallback_req.result = req->result;
360 return crypto_ahash_finup(&rctx->fallback_req);
363 static int n2_hash_async_noimport(struct ahash_request *req, const void *in)
368 static int n2_hash_async_noexport(struct ahash_request *req, void *out)
373 static int n2_hash_cra_init(struct crypto_tfm *tfm)
375 const char *fallback_driver_name = crypto_tfm_alg_name(tfm);
376 struct crypto_ahash *ahash = __crypto_ahash_cast(tfm);
377 struct n2_hash_ctx *ctx = crypto_ahash_ctx(ahash);
378 struct crypto_ahash *fallback_tfm;
381 fallback_tfm = crypto_alloc_ahash(fallback_driver_name, 0,
382 CRYPTO_ALG_NEED_FALLBACK);
383 if (IS_ERR(fallback_tfm)) {
384 pr_warning("Fallback driver '%s' could not be loaded!\n",
385 fallback_driver_name);
386 err = PTR_ERR(fallback_tfm);
390 crypto_ahash_set_reqsize(ahash, (sizeof(struct n2_hash_req_ctx) +
391 crypto_ahash_reqsize(fallback_tfm)));
393 ctx->fallback_tfm = fallback_tfm;
400 static void n2_hash_cra_exit(struct crypto_tfm *tfm)
402 struct crypto_ahash *ahash = __crypto_ahash_cast(tfm);
403 struct n2_hash_ctx *ctx = crypto_ahash_ctx(ahash);
405 crypto_free_ahash(ctx->fallback_tfm);
408 static int n2_hmac_cra_init(struct crypto_tfm *tfm)
410 const char *fallback_driver_name = crypto_tfm_alg_name(tfm);
411 struct crypto_ahash *ahash = __crypto_ahash_cast(tfm);
412 struct n2_hmac_ctx *ctx = crypto_ahash_ctx(ahash);
413 struct n2_hmac_alg *n2alg = n2_hmac_alg(tfm);
414 struct crypto_ahash *fallback_tfm;
415 struct crypto_shash *child_shash;
418 fallback_tfm = crypto_alloc_ahash(fallback_driver_name, 0,
419 CRYPTO_ALG_NEED_FALLBACK);
420 if (IS_ERR(fallback_tfm)) {
421 pr_warning("Fallback driver '%s' could not be loaded!\n",
422 fallback_driver_name);
423 err = PTR_ERR(fallback_tfm);
427 child_shash = crypto_alloc_shash(n2alg->child_alg, 0, 0);
428 if (IS_ERR(child_shash)) {
429 pr_warning("Child shash '%s' could not be loaded!\n",
431 err = PTR_ERR(child_shash);
432 goto out_free_fallback;
435 crypto_ahash_set_reqsize(ahash, (sizeof(struct n2_hash_req_ctx) +
436 crypto_ahash_reqsize(fallback_tfm)));
438 ctx->child_shash = child_shash;
439 ctx->base.fallback_tfm = fallback_tfm;
443 crypto_free_ahash(fallback_tfm);
449 static void n2_hmac_cra_exit(struct crypto_tfm *tfm)
451 struct crypto_ahash *ahash = __crypto_ahash_cast(tfm);
452 struct n2_hmac_ctx *ctx = crypto_ahash_ctx(ahash);
454 crypto_free_ahash(ctx->base.fallback_tfm);
455 crypto_free_shash(ctx->child_shash);
458 static int n2_hmac_async_setkey(struct crypto_ahash *tfm, const u8 *key,
461 struct n2_hmac_ctx *ctx = crypto_ahash_ctx(tfm);
462 struct crypto_shash *child_shash = ctx->child_shash;
463 struct crypto_ahash *fallback_tfm;
464 SHASH_DESC_ON_STACK(shash, child_shash);
467 fallback_tfm = ctx->base.fallback_tfm;
468 err = crypto_ahash_setkey(fallback_tfm, key, keylen);
472 shash->tfm = child_shash;
474 bs = crypto_shash_blocksize(child_shash);
475 ds = crypto_shash_digestsize(child_shash);
476 BUG_ON(ds > N2_HASH_KEY_MAX);
478 err = crypto_shash_digest(shash, key, keylen,
483 } else if (keylen <= N2_HASH_KEY_MAX)
484 memcpy(ctx->hash_key, key, keylen);
486 ctx->hash_key_len = keylen;
491 static unsigned long wait_for_tail(struct spu_queue *qp)
493 unsigned long head, hv_ret;
496 hv_ret = sun4v_ncs_gethead(qp->qhandle, &head);
497 if (hv_ret != HV_EOK) {
498 pr_err("Hypervisor error on gethead\n");
501 if (head == qp->tail) {
509 static unsigned long submit_and_wait_for_tail(struct spu_queue *qp,
510 struct cwq_initial_entry *ent)
512 unsigned long hv_ret = spu_queue_submit(qp, ent);
514 if (hv_ret == HV_EOK)
515 hv_ret = wait_for_tail(qp);
520 static int n2_do_async_digest(struct ahash_request *req,
521 unsigned int auth_type, unsigned int digest_size,
522 unsigned int result_size, void *hash_loc,
523 unsigned long auth_key, unsigned int auth_key_len)
525 struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
526 struct cwq_initial_entry *ent;
527 struct crypto_hash_walk walk;
528 struct spu_queue *qp;
533 /* The total effective length of the operation may not
536 if (unlikely(req->nbytes > (1 << 16))) {
537 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
538 struct n2_hash_ctx *ctx = crypto_ahash_ctx(tfm);
540 ahash_request_set_tfm(&rctx->fallback_req, ctx->fallback_tfm);
541 rctx->fallback_req.base.flags =
542 req->base.flags & CRYPTO_TFM_REQ_MAY_SLEEP;
543 rctx->fallback_req.nbytes = req->nbytes;
544 rctx->fallback_req.src = req->src;
545 rctx->fallback_req.result = req->result;
547 return crypto_ahash_digest(&rctx->fallback_req);
550 nbytes = crypto_hash_walk_first(req, &walk);
553 qp = cpu_to_cwq[cpu];
557 spin_lock_irqsave(&qp->lock, flags);
559 /* XXX can do better, improve this later by doing a by-hand scatterlist
562 ent = qp->q + qp->tail;
564 ent->control = control_word_base(nbytes, auth_key_len, 0,
565 auth_type, digest_size,
566 false, true, false, false,
569 ent->src_addr = __pa(walk.data);
570 ent->auth_key_addr = auth_key;
571 ent->auth_iv_addr = __pa(hash_loc);
572 ent->final_auth_state_addr = 0UL;
573 ent->enc_key_addr = 0UL;
574 ent->enc_iv_addr = 0UL;
575 ent->dest_addr = __pa(hash_loc);
577 nbytes = crypto_hash_walk_done(&walk, 0);
579 ent = spu_queue_next(qp, ent);
581 ent->control = (nbytes - 1);
582 ent->src_addr = __pa(walk.data);
583 ent->auth_key_addr = 0UL;
584 ent->auth_iv_addr = 0UL;
585 ent->final_auth_state_addr = 0UL;
586 ent->enc_key_addr = 0UL;
587 ent->enc_iv_addr = 0UL;
588 ent->dest_addr = 0UL;
590 nbytes = crypto_hash_walk_done(&walk, 0);
592 ent->control |= CONTROL_END_OF_BLOCK;
594 if (submit_and_wait_for_tail(qp, ent) != HV_EOK)
599 spin_unlock_irqrestore(&qp->lock, flags);
602 memcpy(req->result, hash_loc, result_size);
609 static int n2_hash_async_digest(struct ahash_request *req)
611 struct n2_ahash_alg *n2alg = n2_ahash_alg(req->base.tfm);
612 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
615 ds = n2alg->digest_size;
616 if (unlikely(req->nbytes == 0)) {
617 memcpy(req->result, n2alg->hash_zero, ds);
620 memcpy(&rctx->u, n2alg->hash_init, n2alg->hw_op_hashsz);
622 return n2_do_async_digest(req, n2alg->auth_type,
623 n2alg->hw_op_hashsz, ds,
627 static int n2_hmac_async_digest(struct ahash_request *req)
629 struct n2_hmac_alg *n2alg = n2_hmac_alg(req->base.tfm);
630 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
631 struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
632 struct n2_hmac_ctx *ctx = crypto_ahash_ctx(tfm);
635 ds = n2alg->derived.digest_size;
636 if (unlikely(req->nbytes == 0) ||
637 unlikely(ctx->hash_key_len > N2_HASH_KEY_MAX)) {
638 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
639 struct n2_hash_ctx *ctx = crypto_ahash_ctx(tfm);
641 ahash_request_set_tfm(&rctx->fallback_req, ctx->fallback_tfm);
642 rctx->fallback_req.base.flags =
643 req->base.flags & CRYPTO_TFM_REQ_MAY_SLEEP;
644 rctx->fallback_req.nbytes = req->nbytes;
645 rctx->fallback_req.src = req->src;
646 rctx->fallback_req.result = req->result;
648 return crypto_ahash_digest(&rctx->fallback_req);
650 memcpy(&rctx->u, n2alg->derived.hash_init,
651 n2alg->derived.hw_op_hashsz);
653 return n2_do_async_digest(req, n2alg->derived.hmac_type,
654 n2alg->derived.hw_op_hashsz, ds,
656 __pa(&ctx->hash_key),
660 struct n2_cipher_context {
664 u8 aes[AES_MAX_KEY_SIZE];
665 u8 des[DES_KEY_SIZE];
666 u8 des3[3 * DES_KEY_SIZE];
667 u8 arc4[258]; /* S-box, X, Y */
671 #define N2_CHUNK_ARR_LEN 16
673 struct n2_crypto_chunk {
674 struct list_head entry;
675 unsigned long iv_paddr : 44;
676 unsigned long arr_len : 20;
677 unsigned long dest_paddr;
678 unsigned long dest_final;
680 unsigned long src_paddr : 44;
681 unsigned long src_len : 20;
682 } arr[N2_CHUNK_ARR_LEN];
685 struct n2_request_context {
686 struct ablkcipher_walk walk;
687 struct list_head chunk_list;
688 struct n2_crypto_chunk chunk;
692 /* The SPU allows some level of flexibility for partial cipher blocks
693 * being specified in a descriptor.
695 * It merely requires that every descriptor's length field is at least
696 * as large as the cipher block size. This means that a cipher block
697 * can span at most 2 descriptors. However, this does not allow a
698 * partial block to span into the final descriptor as that would
699 * violate the rule (since every descriptor's length must be at lest
700 * the block size). So, for example, assuming an 8 byte block size:
702 * 0xe --> 0xa --> 0x8
704 * is a valid length sequence, whereas:
706 * 0xe --> 0xb --> 0x7
708 * is not a valid sequence.
711 struct n2_cipher_alg {
712 struct list_head entry;
714 struct crypto_alg alg;
717 static inline struct n2_cipher_alg *n2_cipher_alg(struct crypto_tfm *tfm)
719 struct crypto_alg *alg = tfm->__crt_alg;
721 return container_of(alg, struct n2_cipher_alg, alg);
724 struct n2_cipher_request_context {
725 struct ablkcipher_walk walk;
728 static int n2_aes_setkey(struct crypto_ablkcipher *cipher, const u8 *key,
731 struct crypto_tfm *tfm = crypto_ablkcipher_tfm(cipher);
732 struct n2_cipher_context *ctx = crypto_tfm_ctx(tfm);
733 struct n2_cipher_alg *n2alg = n2_cipher_alg(tfm);
735 ctx->enc_type = (n2alg->enc_type & ENC_TYPE_CHAINING_MASK);
738 case AES_KEYSIZE_128:
739 ctx->enc_type |= ENC_TYPE_ALG_AES128;
741 case AES_KEYSIZE_192:
742 ctx->enc_type |= ENC_TYPE_ALG_AES192;
744 case AES_KEYSIZE_256:
745 ctx->enc_type |= ENC_TYPE_ALG_AES256;
748 crypto_ablkcipher_set_flags(cipher, CRYPTO_TFM_RES_BAD_KEY_LEN);
752 ctx->key_len = keylen;
753 memcpy(ctx->key.aes, key, keylen);
757 static int n2_des_setkey(struct crypto_ablkcipher *cipher, const u8 *key,
760 struct crypto_tfm *tfm = crypto_ablkcipher_tfm(cipher);
761 struct n2_cipher_context *ctx = crypto_tfm_ctx(tfm);
762 struct n2_cipher_alg *n2alg = n2_cipher_alg(tfm);
763 u32 tmp[DES_EXPKEY_WORDS];
766 ctx->enc_type = n2alg->enc_type;
768 if (keylen != DES_KEY_SIZE) {
769 crypto_ablkcipher_set_flags(cipher, CRYPTO_TFM_RES_BAD_KEY_LEN);
773 err = des_ekey(tmp, key);
774 if (err == 0 && (tfm->crt_flags & CRYPTO_TFM_REQ_FORBID_WEAK_KEYS)) {
775 tfm->crt_flags |= CRYPTO_TFM_RES_WEAK_KEY;
779 ctx->key_len = keylen;
780 memcpy(ctx->key.des, key, keylen);
784 static int n2_3des_setkey(struct crypto_ablkcipher *cipher, const u8 *key,
787 struct crypto_tfm *tfm = crypto_ablkcipher_tfm(cipher);
788 struct n2_cipher_context *ctx = crypto_tfm_ctx(tfm);
789 struct n2_cipher_alg *n2alg = n2_cipher_alg(tfm);
793 flags = crypto_ablkcipher_get_flags(cipher);
794 err = __des3_verify_key(&flags, key);
796 crypto_ablkcipher_set_flags(cipher, flags);
800 ctx->enc_type = n2alg->enc_type;
802 ctx->key_len = keylen;
803 memcpy(ctx->key.des3, key, keylen);
807 static int n2_arc4_setkey(struct crypto_ablkcipher *cipher, const u8 *key,
810 struct crypto_tfm *tfm = crypto_ablkcipher_tfm(cipher);
811 struct n2_cipher_context *ctx = crypto_tfm_ctx(tfm);
812 struct n2_cipher_alg *n2alg = n2_cipher_alg(tfm);
813 u8 *s = ctx->key.arc4;
818 ctx->enc_type = n2alg->enc_type;
823 for (i = 0; i < 256; i++)
825 for (i = 0; i < 256; i++) {
827 j = (j + key[k] + a) & 0xff;
837 static inline int cipher_descriptor_len(int nbytes, unsigned int block_size)
839 int this_len = nbytes;
841 this_len -= (nbytes & (block_size - 1));
842 return this_len > (1 << 16) ? (1 << 16) : this_len;
845 static int __n2_crypt_chunk(struct crypto_tfm *tfm, struct n2_crypto_chunk *cp,
846 struct spu_queue *qp, bool encrypt)
848 struct n2_cipher_context *ctx = crypto_tfm_ctx(tfm);
849 struct cwq_initial_entry *ent;
853 ent = spu_queue_alloc(qp, cp->arr_len);
855 pr_info("queue_alloc() of %d fails\n",
860 in_place = (cp->dest_paddr == cp->arr[0].src_paddr);
862 ent->control = control_word_base(cp->arr[0].src_len,
863 0, ctx->enc_type, 0, 0,
864 false, true, false, encrypt,
866 (in_place ? OPCODE_INPLACE_BIT : 0));
867 ent->src_addr = cp->arr[0].src_paddr;
868 ent->auth_key_addr = 0UL;
869 ent->auth_iv_addr = 0UL;
870 ent->final_auth_state_addr = 0UL;
871 ent->enc_key_addr = __pa(&ctx->key);
872 ent->enc_iv_addr = cp->iv_paddr;
873 ent->dest_addr = (in_place ? 0UL : cp->dest_paddr);
875 for (i = 1; i < cp->arr_len; i++) {
876 ent = spu_queue_next(qp, ent);
878 ent->control = cp->arr[i].src_len - 1;
879 ent->src_addr = cp->arr[i].src_paddr;
880 ent->auth_key_addr = 0UL;
881 ent->auth_iv_addr = 0UL;
882 ent->final_auth_state_addr = 0UL;
883 ent->enc_key_addr = 0UL;
884 ent->enc_iv_addr = 0UL;
885 ent->dest_addr = 0UL;
887 ent->control |= CONTROL_END_OF_BLOCK;
889 return (spu_queue_submit(qp, ent) != HV_EOK) ? -EINVAL : 0;
892 static int n2_compute_chunks(struct ablkcipher_request *req)
894 struct n2_request_context *rctx = ablkcipher_request_ctx(req);
895 struct ablkcipher_walk *walk = &rctx->walk;
896 struct n2_crypto_chunk *chunk;
897 unsigned long dest_prev;
898 unsigned int tot_len;
902 ablkcipher_walk_init(walk, req->dst, req->src, req->nbytes);
903 err = ablkcipher_walk_phys(req, walk);
907 INIT_LIST_HEAD(&rctx->chunk_list);
909 chunk = &rctx->chunk;
910 INIT_LIST_HEAD(&chunk->entry);
912 chunk->iv_paddr = 0UL;
914 chunk->dest_paddr = 0UL;
916 prev_in_place = false;
920 while ((nbytes = walk->nbytes) != 0) {
921 unsigned long dest_paddr, src_paddr;
925 src_paddr = (page_to_phys(walk->src.page) +
927 dest_paddr = (page_to_phys(walk->dst.page) +
929 in_place = (src_paddr == dest_paddr);
930 this_len = cipher_descriptor_len(nbytes, walk->blocksize);
932 if (chunk->arr_len != 0) {
933 if (in_place != prev_in_place ||
935 dest_paddr != dest_prev) ||
936 chunk->arr_len == N2_CHUNK_ARR_LEN ||
937 tot_len + this_len > (1 << 16)) {
938 chunk->dest_final = dest_prev;
939 list_add_tail(&chunk->entry,
941 chunk = kzalloc(sizeof(*chunk), GFP_ATOMIC);
946 INIT_LIST_HEAD(&chunk->entry);
949 if (chunk->arr_len == 0) {
950 chunk->dest_paddr = dest_paddr;
953 chunk->arr[chunk->arr_len].src_paddr = src_paddr;
954 chunk->arr[chunk->arr_len].src_len = this_len;
957 dest_prev = dest_paddr + this_len;
958 prev_in_place = in_place;
961 err = ablkcipher_walk_done(req, walk, nbytes - this_len);
965 if (!err && chunk->arr_len != 0) {
966 chunk->dest_final = dest_prev;
967 list_add_tail(&chunk->entry, &rctx->chunk_list);
973 static void n2_chunk_complete(struct ablkcipher_request *req, void *final_iv)
975 struct n2_request_context *rctx = ablkcipher_request_ctx(req);
976 struct n2_crypto_chunk *c, *tmp;
979 memcpy(rctx->walk.iv, final_iv, rctx->walk.blocksize);
981 ablkcipher_walk_complete(&rctx->walk);
982 list_for_each_entry_safe(c, tmp, &rctx->chunk_list, entry) {
984 if (unlikely(c != &rctx->chunk))
990 static int n2_do_ecb(struct ablkcipher_request *req, bool encrypt)
992 struct n2_request_context *rctx = ablkcipher_request_ctx(req);
993 struct crypto_tfm *tfm = req->base.tfm;
994 int err = n2_compute_chunks(req);
995 struct n2_crypto_chunk *c, *tmp;
996 unsigned long flags, hv_ret;
997 struct spu_queue *qp;
1002 qp = cpu_to_cwq[get_cpu()];
1007 spin_lock_irqsave(&qp->lock, flags);
1009 list_for_each_entry_safe(c, tmp, &rctx->chunk_list, entry) {
1010 err = __n2_crypt_chunk(tfm, c, qp, encrypt);
1013 list_del(&c->entry);
1014 if (unlikely(c != &rctx->chunk))
1018 hv_ret = wait_for_tail(qp);
1019 if (hv_ret != HV_EOK)
1023 spin_unlock_irqrestore(&qp->lock, flags);
1028 n2_chunk_complete(req, NULL);
1032 static int n2_encrypt_ecb(struct ablkcipher_request *req)
1034 return n2_do_ecb(req, true);
1037 static int n2_decrypt_ecb(struct ablkcipher_request *req)
1039 return n2_do_ecb(req, false);
1042 static int n2_do_chaining(struct ablkcipher_request *req, bool encrypt)
1044 struct n2_request_context *rctx = ablkcipher_request_ctx(req);
1045 struct crypto_tfm *tfm = req->base.tfm;
1046 unsigned long flags, hv_ret, iv_paddr;
1047 int err = n2_compute_chunks(req);
1048 struct n2_crypto_chunk *c, *tmp;
1049 struct spu_queue *qp;
1050 void *final_iv_addr;
1052 final_iv_addr = NULL;
1057 qp = cpu_to_cwq[get_cpu()];
1062 spin_lock_irqsave(&qp->lock, flags);
1065 iv_paddr = __pa(rctx->walk.iv);
1066 list_for_each_entry_safe(c, tmp, &rctx->chunk_list,
1068 c->iv_paddr = iv_paddr;
1069 err = __n2_crypt_chunk(tfm, c, qp, true);
1072 iv_paddr = c->dest_final - rctx->walk.blocksize;
1073 list_del(&c->entry);
1074 if (unlikely(c != &rctx->chunk))
1077 final_iv_addr = __va(iv_paddr);
1079 list_for_each_entry_safe_reverse(c, tmp, &rctx->chunk_list,
1081 if (c == &rctx->chunk) {
1082 iv_paddr = __pa(rctx->walk.iv);
1084 iv_paddr = (tmp->arr[tmp->arr_len-1].src_paddr +
1085 tmp->arr[tmp->arr_len-1].src_len -
1086 rctx->walk.blocksize);
1088 if (!final_iv_addr) {
1091 pa = (c->arr[c->arr_len-1].src_paddr +
1092 c->arr[c->arr_len-1].src_len -
1093 rctx->walk.blocksize);
1094 final_iv_addr = rctx->temp_iv;
1095 memcpy(rctx->temp_iv, __va(pa),
1096 rctx->walk.blocksize);
1098 c->iv_paddr = iv_paddr;
1099 err = __n2_crypt_chunk(tfm, c, qp, false);
1102 list_del(&c->entry);
1103 if (unlikely(c != &rctx->chunk))
1108 hv_ret = wait_for_tail(qp);
1109 if (hv_ret != HV_EOK)
1113 spin_unlock_irqrestore(&qp->lock, flags);
1118 n2_chunk_complete(req, err ? NULL : final_iv_addr);
1122 static int n2_encrypt_chaining(struct ablkcipher_request *req)
1124 return n2_do_chaining(req, true);
1127 static int n2_decrypt_chaining(struct ablkcipher_request *req)
1129 return n2_do_chaining(req, false);
1132 struct n2_cipher_tmpl {
1134 const char *drv_name;
1137 struct ablkcipher_alg ablkcipher;
1140 static const struct n2_cipher_tmpl cipher_tmpls[] = {
1141 /* ARC4: only ECB is supported (chaining bits ignored) */
1142 { .name = "ecb(arc4)",
1143 .drv_name = "ecb-arc4",
1145 .enc_type = (ENC_TYPE_ALG_RC4_STREAM |
1146 ENC_TYPE_CHAINING_ECB),
1150 .setkey = n2_arc4_setkey,
1151 .encrypt = n2_encrypt_ecb,
1152 .decrypt = n2_decrypt_ecb,
1156 /* DES: ECB CBC and CFB are supported */
1157 { .name = "ecb(des)",
1158 .drv_name = "ecb-des",
1159 .block_size = DES_BLOCK_SIZE,
1160 .enc_type = (ENC_TYPE_ALG_DES |
1161 ENC_TYPE_CHAINING_ECB),
1163 .min_keysize = DES_KEY_SIZE,
1164 .max_keysize = DES_KEY_SIZE,
1165 .setkey = n2_des_setkey,
1166 .encrypt = n2_encrypt_ecb,
1167 .decrypt = n2_decrypt_ecb,
1170 { .name = "cbc(des)",
1171 .drv_name = "cbc-des",
1172 .block_size = DES_BLOCK_SIZE,
1173 .enc_type = (ENC_TYPE_ALG_DES |
1174 ENC_TYPE_CHAINING_CBC),
1176 .ivsize = DES_BLOCK_SIZE,
1177 .min_keysize = DES_KEY_SIZE,
1178 .max_keysize = DES_KEY_SIZE,
1179 .setkey = n2_des_setkey,
1180 .encrypt = n2_encrypt_chaining,
1181 .decrypt = n2_decrypt_chaining,
1184 { .name = "cfb(des)",
1185 .drv_name = "cfb-des",
1186 .block_size = DES_BLOCK_SIZE,
1187 .enc_type = (ENC_TYPE_ALG_DES |
1188 ENC_TYPE_CHAINING_CFB),
1190 .min_keysize = DES_KEY_SIZE,
1191 .max_keysize = DES_KEY_SIZE,
1192 .setkey = n2_des_setkey,
1193 .encrypt = n2_encrypt_chaining,
1194 .decrypt = n2_decrypt_chaining,
1198 /* 3DES: ECB CBC and CFB are supported */
1199 { .name = "ecb(des3_ede)",
1200 .drv_name = "ecb-3des",
1201 .block_size = DES_BLOCK_SIZE,
1202 .enc_type = (ENC_TYPE_ALG_3DES |
1203 ENC_TYPE_CHAINING_ECB),
1205 .min_keysize = 3 * DES_KEY_SIZE,
1206 .max_keysize = 3 * DES_KEY_SIZE,
1207 .setkey = n2_3des_setkey,
1208 .encrypt = n2_encrypt_ecb,
1209 .decrypt = n2_decrypt_ecb,
1212 { .name = "cbc(des3_ede)",
1213 .drv_name = "cbc-3des",
1214 .block_size = DES_BLOCK_SIZE,
1215 .enc_type = (ENC_TYPE_ALG_3DES |
1216 ENC_TYPE_CHAINING_CBC),
1218 .ivsize = DES_BLOCK_SIZE,
1219 .min_keysize = 3 * DES_KEY_SIZE,
1220 .max_keysize = 3 * DES_KEY_SIZE,
1221 .setkey = n2_3des_setkey,
1222 .encrypt = n2_encrypt_chaining,
1223 .decrypt = n2_decrypt_chaining,
1226 { .name = "cfb(des3_ede)",
1227 .drv_name = "cfb-3des",
1228 .block_size = DES_BLOCK_SIZE,
1229 .enc_type = (ENC_TYPE_ALG_3DES |
1230 ENC_TYPE_CHAINING_CFB),
1232 .min_keysize = 3 * DES_KEY_SIZE,
1233 .max_keysize = 3 * DES_KEY_SIZE,
1234 .setkey = n2_3des_setkey,
1235 .encrypt = n2_encrypt_chaining,
1236 .decrypt = n2_decrypt_chaining,
1239 /* AES: ECB CBC and CTR are supported */
1240 { .name = "ecb(aes)",
1241 .drv_name = "ecb-aes",
1242 .block_size = AES_BLOCK_SIZE,
1243 .enc_type = (ENC_TYPE_ALG_AES128 |
1244 ENC_TYPE_CHAINING_ECB),
1246 .min_keysize = AES_MIN_KEY_SIZE,
1247 .max_keysize = AES_MAX_KEY_SIZE,
1248 .setkey = n2_aes_setkey,
1249 .encrypt = n2_encrypt_ecb,
1250 .decrypt = n2_decrypt_ecb,
1253 { .name = "cbc(aes)",
1254 .drv_name = "cbc-aes",
1255 .block_size = AES_BLOCK_SIZE,
1256 .enc_type = (ENC_TYPE_ALG_AES128 |
1257 ENC_TYPE_CHAINING_CBC),
1259 .ivsize = AES_BLOCK_SIZE,
1260 .min_keysize = AES_MIN_KEY_SIZE,
1261 .max_keysize = AES_MAX_KEY_SIZE,
1262 .setkey = n2_aes_setkey,
1263 .encrypt = n2_encrypt_chaining,
1264 .decrypt = n2_decrypt_chaining,
1267 { .name = "ctr(aes)",
1268 .drv_name = "ctr-aes",
1269 .block_size = AES_BLOCK_SIZE,
1270 .enc_type = (ENC_TYPE_ALG_AES128 |
1271 ENC_TYPE_CHAINING_COUNTER),
1273 .ivsize = AES_BLOCK_SIZE,
1274 .min_keysize = AES_MIN_KEY_SIZE,
1275 .max_keysize = AES_MAX_KEY_SIZE,
1276 .setkey = n2_aes_setkey,
1277 .encrypt = n2_encrypt_chaining,
1278 .decrypt = n2_encrypt_chaining,
1283 #define NUM_CIPHER_TMPLS ARRAY_SIZE(cipher_tmpls)
1285 static LIST_HEAD(cipher_algs);
1287 struct n2_hash_tmpl {
1289 const u8 *hash_zero;
1290 const u32 *hash_init;
1298 static const u32 md5_init[MD5_HASH_WORDS] = {
1299 cpu_to_le32(MD5_H0),
1300 cpu_to_le32(MD5_H1),
1301 cpu_to_le32(MD5_H2),
1302 cpu_to_le32(MD5_H3),
1304 static const u32 sha1_init[SHA1_DIGEST_SIZE / 4] = {
1305 SHA1_H0, SHA1_H1, SHA1_H2, SHA1_H3, SHA1_H4,
1307 static const u32 sha256_init[SHA256_DIGEST_SIZE / 4] = {
1308 SHA256_H0, SHA256_H1, SHA256_H2, SHA256_H3,
1309 SHA256_H4, SHA256_H5, SHA256_H6, SHA256_H7,
1311 static const u32 sha224_init[SHA256_DIGEST_SIZE / 4] = {
1312 SHA224_H0, SHA224_H1, SHA224_H2, SHA224_H3,
1313 SHA224_H4, SHA224_H5, SHA224_H6, SHA224_H7,
1316 static const struct n2_hash_tmpl hash_tmpls[] = {
1318 .hash_zero = md5_zero_message_hash,
1319 .hash_init = md5_init,
1320 .auth_type = AUTH_TYPE_MD5,
1321 .hmac_type = AUTH_TYPE_HMAC_MD5,
1322 .hw_op_hashsz = MD5_DIGEST_SIZE,
1323 .digest_size = MD5_DIGEST_SIZE,
1324 .block_size = MD5_HMAC_BLOCK_SIZE },
1326 .hash_zero = sha1_zero_message_hash,
1327 .hash_init = sha1_init,
1328 .auth_type = AUTH_TYPE_SHA1,
1329 .hmac_type = AUTH_TYPE_HMAC_SHA1,
1330 .hw_op_hashsz = SHA1_DIGEST_SIZE,
1331 .digest_size = SHA1_DIGEST_SIZE,
1332 .block_size = SHA1_BLOCK_SIZE },
1334 .hash_zero = sha256_zero_message_hash,
1335 .hash_init = sha256_init,
1336 .auth_type = AUTH_TYPE_SHA256,
1337 .hmac_type = AUTH_TYPE_HMAC_SHA256,
1338 .hw_op_hashsz = SHA256_DIGEST_SIZE,
1339 .digest_size = SHA256_DIGEST_SIZE,
1340 .block_size = SHA256_BLOCK_SIZE },
1342 .hash_zero = sha224_zero_message_hash,
1343 .hash_init = sha224_init,
1344 .auth_type = AUTH_TYPE_SHA256,
1345 .hmac_type = AUTH_TYPE_RESERVED,
1346 .hw_op_hashsz = SHA256_DIGEST_SIZE,
1347 .digest_size = SHA224_DIGEST_SIZE,
1348 .block_size = SHA224_BLOCK_SIZE },
1350 #define NUM_HASH_TMPLS ARRAY_SIZE(hash_tmpls)
1352 static LIST_HEAD(ahash_algs);
1353 static LIST_HEAD(hmac_algs);
1355 static int algs_registered;
1357 static void __n2_unregister_algs(void)
1359 struct n2_cipher_alg *cipher, *cipher_tmp;
1360 struct n2_ahash_alg *alg, *alg_tmp;
1361 struct n2_hmac_alg *hmac, *hmac_tmp;
1363 list_for_each_entry_safe(cipher, cipher_tmp, &cipher_algs, entry) {
1364 crypto_unregister_alg(&cipher->alg);
1365 list_del(&cipher->entry);
1368 list_for_each_entry_safe(hmac, hmac_tmp, &hmac_algs, derived.entry) {
1369 crypto_unregister_ahash(&hmac->derived.alg);
1370 list_del(&hmac->derived.entry);
1373 list_for_each_entry_safe(alg, alg_tmp, &ahash_algs, entry) {
1374 crypto_unregister_ahash(&alg->alg);
1375 list_del(&alg->entry);
1380 static int n2_cipher_cra_init(struct crypto_tfm *tfm)
1382 tfm->crt_ablkcipher.reqsize = sizeof(struct n2_request_context);
1386 static int __n2_register_one_cipher(const struct n2_cipher_tmpl *tmpl)
1388 struct n2_cipher_alg *p = kzalloc(sizeof(*p), GFP_KERNEL);
1389 struct crypto_alg *alg;
1397 snprintf(alg->cra_name, CRYPTO_MAX_ALG_NAME, "%s", tmpl->name);
1398 snprintf(alg->cra_driver_name, CRYPTO_MAX_ALG_NAME, "%s-n2", tmpl->drv_name);
1399 alg->cra_priority = N2_CRA_PRIORITY;
1400 alg->cra_flags = CRYPTO_ALG_TYPE_ABLKCIPHER |
1401 CRYPTO_ALG_KERN_DRIVER_ONLY | CRYPTO_ALG_ASYNC;
1402 alg->cra_blocksize = tmpl->block_size;
1403 p->enc_type = tmpl->enc_type;
1404 alg->cra_ctxsize = sizeof(struct n2_cipher_context);
1405 alg->cra_type = &crypto_ablkcipher_type;
1406 alg->cra_u.ablkcipher = tmpl->ablkcipher;
1407 alg->cra_init = n2_cipher_cra_init;
1408 alg->cra_module = THIS_MODULE;
1410 list_add(&p->entry, &cipher_algs);
1411 err = crypto_register_alg(alg);
1413 pr_err("%s alg registration failed\n", alg->cra_name);
1414 list_del(&p->entry);
1417 pr_info("%s alg registered\n", alg->cra_name);
1422 static int __n2_register_one_hmac(struct n2_ahash_alg *n2ahash)
1424 struct n2_hmac_alg *p = kzalloc(sizeof(*p), GFP_KERNEL);
1425 struct ahash_alg *ahash;
1426 struct crypto_alg *base;
1432 p->child_alg = n2ahash->alg.halg.base.cra_name;
1433 memcpy(&p->derived, n2ahash, sizeof(struct n2_ahash_alg));
1434 INIT_LIST_HEAD(&p->derived.entry);
1436 ahash = &p->derived.alg;
1437 ahash->digest = n2_hmac_async_digest;
1438 ahash->setkey = n2_hmac_async_setkey;
1440 base = &ahash->halg.base;
1441 snprintf(base->cra_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)", p->child_alg);
1442 snprintf(base->cra_driver_name, CRYPTO_MAX_ALG_NAME, "hmac-%s-n2", p->child_alg);
1444 base->cra_ctxsize = sizeof(struct n2_hmac_ctx);
1445 base->cra_init = n2_hmac_cra_init;
1446 base->cra_exit = n2_hmac_cra_exit;
1448 list_add(&p->derived.entry, &hmac_algs);
1449 err = crypto_register_ahash(ahash);
1451 pr_err("%s alg registration failed\n", base->cra_name);
1452 list_del(&p->derived.entry);
1455 pr_info("%s alg registered\n", base->cra_name);
1460 static int __n2_register_one_ahash(const struct n2_hash_tmpl *tmpl)
1462 struct n2_ahash_alg *p = kzalloc(sizeof(*p), GFP_KERNEL);
1463 struct hash_alg_common *halg;
1464 struct crypto_alg *base;
1465 struct ahash_alg *ahash;
1471 p->hash_zero = tmpl->hash_zero;
1472 p->hash_init = tmpl->hash_init;
1473 p->auth_type = tmpl->auth_type;
1474 p->hmac_type = tmpl->hmac_type;
1475 p->hw_op_hashsz = tmpl->hw_op_hashsz;
1476 p->digest_size = tmpl->digest_size;
1479 ahash->init = n2_hash_async_init;
1480 ahash->update = n2_hash_async_update;
1481 ahash->final = n2_hash_async_final;
1482 ahash->finup = n2_hash_async_finup;
1483 ahash->digest = n2_hash_async_digest;
1484 ahash->export = n2_hash_async_noexport;
1485 ahash->import = n2_hash_async_noimport;
1487 halg = &ahash->halg;
1488 halg->digestsize = tmpl->digest_size;
1491 snprintf(base->cra_name, CRYPTO_MAX_ALG_NAME, "%s", tmpl->name);
1492 snprintf(base->cra_driver_name, CRYPTO_MAX_ALG_NAME, "%s-n2", tmpl->name);
1493 base->cra_priority = N2_CRA_PRIORITY;
1494 base->cra_flags = CRYPTO_ALG_KERN_DRIVER_ONLY |
1495 CRYPTO_ALG_NEED_FALLBACK;
1496 base->cra_blocksize = tmpl->block_size;
1497 base->cra_ctxsize = sizeof(struct n2_hash_ctx);
1498 base->cra_module = THIS_MODULE;
1499 base->cra_init = n2_hash_cra_init;
1500 base->cra_exit = n2_hash_cra_exit;
1502 list_add(&p->entry, &ahash_algs);
1503 err = crypto_register_ahash(ahash);
1505 pr_err("%s alg registration failed\n", base->cra_name);
1506 list_del(&p->entry);
1509 pr_info("%s alg registered\n", base->cra_name);
1511 if (!err && p->hmac_type != AUTH_TYPE_RESERVED)
1512 err = __n2_register_one_hmac(p);
1516 static int n2_register_algs(void)
1520 mutex_lock(&spu_lock);
1521 if (algs_registered++)
1524 for (i = 0; i < NUM_HASH_TMPLS; i++) {
1525 err = __n2_register_one_ahash(&hash_tmpls[i]);
1527 __n2_unregister_algs();
1531 for (i = 0; i < NUM_CIPHER_TMPLS; i++) {
1532 err = __n2_register_one_cipher(&cipher_tmpls[i]);
1534 __n2_unregister_algs();
1540 mutex_unlock(&spu_lock);
1544 static void n2_unregister_algs(void)
1546 mutex_lock(&spu_lock);
1547 if (!--algs_registered)
1548 __n2_unregister_algs();
1549 mutex_unlock(&spu_lock);
1552 /* To map CWQ queues to interrupt sources, the hypervisor API provides
1553 * a devino. This isn't very useful to us because all of the
1554 * interrupts listed in the device_node have been translated to
1555 * Linux virtual IRQ cookie numbers.
1557 * So we have to back-translate, going through the 'intr' and 'ino'
1558 * property tables of the n2cp MDESC node, matching it with the OF
1559 * 'interrupts' property entries, in order to to figure out which
1560 * devino goes to which already-translated IRQ.
1562 static int find_devino_index(struct platform_device *dev, struct spu_mdesc_info *ip,
1563 unsigned long dev_ino)
1565 const unsigned int *dev_intrs;
1569 for (i = 0; i < ip->num_intrs; i++) {
1570 if (ip->ino_table[i].ino == dev_ino)
1573 if (i == ip->num_intrs)
1576 intr = ip->ino_table[i].intr;
1578 dev_intrs = of_get_property(dev->dev.of_node, "interrupts", NULL);
1582 for (i = 0; i < dev->archdata.num_irqs; i++) {
1583 if (dev_intrs[i] == intr)
1590 static int spu_map_ino(struct platform_device *dev, struct spu_mdesc_info *ip,
1591 const char *irq_name, struct spu_queue *p,
1592 irq_handler_t handler)
1597 herr = sun4v_ncs_qhandle_to_devino(p->qhandle, &p->devino);
1601 index = find_devino_index(dev, ip, p->devino);
1605 p->irq = dev->archdata.irqs[index];
1607 sprintf(p->irq_name, "%s-%d", irq_name, index);
1609 return request_irq(p->irq, handler, 0, p->irq_name, p);
1612 static struct kmem_cache *queue_cache[2];
1614 static void *new_queue(unsigned long q_type)
1616 return kmem_cache_zalloc(queue_cache[q_type - 1], GFP_KERNEL);
1619 static void free_queue(void *p, unsigned long q_type)
1621 kmem_cache_free(queue_cache[q_type - 1], p);
1624 static int queue_cache_init(void)
1626 if (!queue_cache[HV_NCS_QTYPE_MAU - 1])
1627 queue_cache[HV_NCS_QTYPE_MAU - 1] =
1628 kmem_cache_create("mau_queue",
1631 MAU_ENTRY_SIZE, 0, NULL);
1632 if (!queue_cache[HV_NCS_QTYPE_MAU - 1])
1635 if (!queue_cache[HV_NCS_QTYPE_CWQ - 1])
1636 queue_cache[HV_NCS_QTYPE_CWQ - 1] =
1637 kmem_cache_create("cwq_queue",
1640 CWQ_ENTRY_SIZE, 0, NULL);
1641 if (!queue_cache[HV_NCS_QTYPE_CWQ - 1]) {
1642 kmem_cache_destroy(queue_cache[HV_NCS_QTYPE_MAU - 1]);
1643 queue_cache[HV_NCS_QTYPE_MAU - 1] = NULL;
1649 static void queue_cache_destroy(void)
1651 kmem_cache_destroy(queue_cache[HV_NCS_QTYPE_MAU - 1]);
1652 kmem_cache_destroy(queue_cache[HV_NCS_QTYPE_CWQ - 1]);
1653 queue_cache[HV_NCS_QTYPE_MAU - 1] = NULL;
1654 queue_cache[HV_NCS_QTYPE_CWQ - 1] = NULL;
1657 static long spu_queue_register_workfn(void *arg)
1659 struct spu_qreg *qr = arg;
1660 struct spu_queue *p = qr->queue;
1661 unsigned long q_type = qr->type;
1662 unsigned long hv_ret;
1664 hv_ret = sun4v_ncs_qconf(q_type, __pa(p->q),
1665 CWQ_NUM_ENTRIES, &p->qhandle);
1667 sun4v_ncs_sethead_marker(p->qhandle, 0);
1669 return hv_ret ? -EINVAL : 0;
1672 static int spu_queue_register(struct spu_queue *p, unsigned long q_type)
1674 int cpu = cpumask_any_and(&p->sharing, cpu_online_mask);
1675 struct spu_qreg qr = { .queue = p, .type = q_type };
1677 return work_on_cpu_safe(cpu, spu_queue_register_workfn, &qr);
1680 static int spu_queue_setup(struct spu_queue *p)
1684 p->q = new_queue(p->q_type);
1688 err = spu_queue_register(p, p->q_type);
1690 free_queue(p->q, p->q_type);
1697 static void spu_queue_destroy(struct spu_queue *p)
1699 unsigned long hv_ret;
1704 hv_ret = sun4v_ncs_qconf(p->q_type, p->qhandle, 0, &p->qhandle);
1707 free_queue(p->q, p->q_type);
1710 static void spu_list_destroy(struct list_head *list)
1712 struct spu_queue *p, *n;
1714 list_for_each_entry_safe(p, n, list, list) {
1717 for (i = 0; i < NR_CPUS; i++) {
1718 if (cpu_to_cwq[i] == p)
1719 cpu_to_cwq[i] = NULL;
1723 free_irq(p->irq, p);
1726 spu_queue_destroy(p);
1732 /* Walk the backward arcs of a CWQ 'exec-unit' node,
1733 * gathering cpu membership information.
1735 static int spu_mdesc_walk_arcs(struct mdesc_handle *mdesc,
1736 struct platform_device *dev,
1737 u64 node, struct spu_queue *p,
1738 struct spu_queue **table)
1742 mdesc_for_each_arc(arc, mdesc, node, MDESC_ARC_TYPE_BACK) {
1743 u64 tgt = mdesc_arc_target(mdesc, arc);
1744 const char *name = mdesc_node_name(mdesc, tgt);
1747 if (strcmp(name, "cpu"))
1749 id = mdesc_get_property(mdesc, tgt, "id", NULL);
1750 if (table[*id] != NULL) {
1751 dev_err(&dev->dev, "%pOF: SPU cpu slot already set.\n",
1755 cpumask_set_cpu(*id, &p->sharing);
1761 /* Process an 'exec-unit' MDESC node of type 'cwq'. */
1762 static int handle_exec_unit(struct spu_mdesc_info *ip, struct list_head *list,
1763 struct platform_device *dev, struct mdesc_handle *mdesc,
1764 u64 node, const char *iname, unsigned long q_type,
1765 irq_handler_t handler, struct spu_queue **table)
1767 struct spu_queue *p;
1770 p = kzalloc(sizeof(struct spu_queue), GFP_KERNEL);
1772 dev_err(&dev->dev, "%pOF: Could not allocate SPU queue.\n",
1777 cpumask_clear(&p->sharing);
1778 spin_lock_init(&p->lock);
1780 INIT_LIST_HEAD(&p->jobs);
1781 list_add(&p->list, list);
1783 err = spu_mdesc_walk_arcs(mdesc, dev, node, p, table);
1787 err = spu_queue_setup(p);
1791 return spu_map_ino(dev, ip, iname, p, handler);
1794 static int spu_mdesc_scan(struct mdesc_handle *mdesc, struct platform_device *dev,
1795 struct spu_mdesc_info *ip, struct list_head *list,
1796 const char *exec_name, unsigned long q_type,
1797 irq_handler_t handler, struct spu_queue **table)
1802 mdesc_for_each_node_by_name(mdesc, node, "exec-unit") {
1805 type = mdesc_get_property(mdesc, node, "type", NULL);
1806 if (!type || strcmp(type, exec_name))
1809 err = handle_exec_unit(ip, list, dev, mdesc, node,
1810 exec_name, q_type, handler, table);
1812 spu_list_destroy(list);
1820 static int get_irq_props(struct mdesc_handle *mdesc, u64 node,
1821 struct spu_mdesc_info *ip)
1827 ino = mdesc_get_property(mdesc, node, "ino", &ino_len);
1829 printk("NO 'ino'\n");
1833 ip->num_intrs = ino_len / sizeof(u64);
1834 ip->ino_table = kzalloc((sizeof(struct ino_blob) *
1840 for (i = 0; i < ip->num_intrs; i++) {
1841 struct ino_blob *b = &ip->ino_table[i];
1849 static int grab_mdesc_irq_props(struct mdesc_handle *mdesc,
1850 struct platform_device *dev,
1851 struct spu_mdesc_info *ip,
1852 const char *node_name)
1854 const unsigned int *reg;
1857 reg = of_get_property(dev->dev.of_node, "reg", NULL);
1861 mdesc_for_each_node_by_name(mdesc, node, "virtual-device") {
1865 name = mdesc_get_property(mdesc, node, "name", NULL);
1866 if (!name || strcmp(name, node_name))
1868 chdl = mdesc_get_property(mdesc, node, "cfg-handle", NULL);
1869 if (!chdl || (*chdl != *reg))
1871 ip->cfg_handle = *chdl;
1872 return get_irq_props(mdesc, node, ip);
1878 static unsigned long n2_spu_hvapi_major;
1879 static unsigned long n2_spu_hvapi_minor;
1881 static int n2_spu_hvapi_register(void)
1885 n2_spu_hvapi_major = 2;
1886 n2_spu_hvapi_minor = 0;
1888 err = sun4v_hvapi_register(HV_GRP_NCS,
1890 &n2_spu_hvapi_minor);
1893 pr_info("Registered NCS HVAPI version %lu.%lu\n",
1895 n2_spu_hvapi_minor);
1900 static void n2_spu_hvapi_unregister(void)
1902 sun4v_hvapi_unregister(HV_GRP_NCS);
1905 static int global_ref;
1907 static int grab_global_resources(void)
1911 mutex_lock(&spu_lock);
1916 err = n2_spu_hvapi_register();
1920 err = queue_cache_init();
1922 goto out_hvapi_release;
1925 cpu_to_cwq = kcalloc(NR_CPUS, sizeof(struct spu_queue *),
1928 goto out_queue_cache_destroy;
1930 cpu_to_mau = kcalloc(NR_CPUS, sizeof(struct spu_queue *),
1933 goto out_free_cwq_table;
1940 mutex_unlock(&spu_lock);
1947 out_queue_cache_destroy:
1948 queue_cache_destroy();
1951 n2_spu_hvapi_unregister();
1955 static void release_global_resources(void)
1957 mutex_lock(&spu_lock);
1958 if (!--global_ref) {
1965 queue_cache_destroy();
1966 n2_spu_hvapi_unregister();
1968 mutex_unlock(&spu_lock);
1971 static struct n2_crypto *alloc_n2cp(void)
1973 struct n2_crypto *np = kzalloc(sizeof(struct n2_crypto), GFP_KERNEL);
1976 INIT_LIST_HEAD(&np->cwq_list);
1981 static void free_n2cp(struct n2_crypto *np)
1983 kfree(np->cwq_info.ino_table);
1984 np->cwq_info.ino_table = NULL;
1989 static void n2_spu_driver_version(void)
1991 static int n2_spu_version_printed;
1993 if (n2_spu_version_printed++ == 0)
1994 pr_info("%s", version);
1997 static int n2_crypto_probe(struct platform_device *dev)
1999 struct mdesc_handle *mdesc;
2000 struct n2_crypto *np;
2003 n2_spu_driver_version();
2005 pr_info("Found N2CP at %pOF\n", dev->dev.of_node);
2009 dev_err(&dev->dev, "%pOF: Unable to allocate n2cp.\n",
2014 err = grab_global_resources();
2016 dev_err(&dev->dev, "%pOF: Unable to grab global resources.\n",
2021 mdesc = mdesc_grab();
2024 dev_err(&dev->dev, "%pOF: Unable to grab MDESC.\n",
2027 goto out_free_global;
2029 err = grab_mdesc_irq_props(mdesc, dev, &np->cwq_info, "n2cp");
2031 dev_err(&dev->dev, "%pOF: Unable to grab IRQ props.\n",
2033 mdesc_release(mdesc);
2034 goto out_free_global;
2037 err = spu_mdesc_scan(mdesc, dev, &np->cwq_info, &np->cwq_list,
2038 "cwq", HV_NCS_QTYPE_CWQ, cwq_intr,
2040 mdesc_release(mdesc);
2043 dev_err(&dev->dev, "%pOF: CWQ MDESC scan failed.\n",
2045 goto out_free_global;
2048 err = n2_register_algs();
2050 dev_err(&dev->dev, "%pOF: Unable to register algorithms.\n",
2052 goto out_free_spu_list;
2055 dev_set_drvdata(&dev->dev, np);
2060 spu_list_destroy(&np->cwq_list);
2063 release_global_resources();
2071 static int n2_crypto_remove(struct platform_device *dev)
2073 struct n2_crypto *np = dev_get_drvdata(&dev->dev);
2075 n2_unregister_algs();
2077 spu_list_destroy(&np->cwq_list);
2079 release_global_resources();
2086 static struct n2_mau *alloc_ncp(void)
2088 struct n2_mau *mp = kzalloc(sizeof(struct n2_mau), GFP_KERNEL);
2091 INIT_LIST_HEAD(&mp->mau_list);
2096 static void free_ncp(struct n2_mau *mp)
2098 kfree(mp->mau_info.ino_table);
2099 mp->mau_info.ino_table = NULL;
2104 static int n2_mau_probe(struct platform_device *dev)
2106 struct mdesc_handle *mdesc;
2110 n2_spu_driver_version();
2112 pr_info("Found NCP at %pOF\n", dev->dev.of_node);
2116 dev_err(&dev->dev, "%pOF: Unable to allocate ncp.\n",
2121 err = grab_global_resources();
2123 dev_err(&dev->dev, "%pOF: Unable to grab global resources.\n",
2128 mdesc = mdesc_grab();
2131 dev_err(&dev->dev, "%pOF: Unable to grab MDESC.\n",
2134 goto out_free_global;
2137 err = grab_mdesc_irq_props(mdesc, dev, &mp->mau_info, "ncp");
2139 dev_err(&dev->dev, "%pOF: Unable to grab IRQ props.\n",
2141 mdesc_release(mdesc);
2142 goto out_free_global;
2145 err = spu_mdesc_scan(mdesc, dev, &mp->mau_info, &mp->mau_list,
2146 "mau", HV_NCS_QTYPE_MAU, mau_intr,
2148 mdesc_release(mdesc);
2151 dev_err(&dev->dev, "%pOF: MAU MDESC scan failed.\n",
2153 goto out_free_global;
2156 dev_set_drvdata(&dev->dev, mp);
2161 release_global_resources();
2169 static int n2_mau_remove(struct platform_device *dev)
2171 struct n2_mau *mp = dev_get_drvdata(&dev->dev);
2173 spu_list_destroy(&mp->mau_list);
2175 release_global_resources();
2182 static const struct of_device_id n2_crypto_match[] = {
2185 .compatible = "SUNW,n2-cwq",
2189 .compatible = "SUNW,vf-cwq",
2193 .compatible = "SUNW,kt-cwq",
2198 MODULE_DEVICE_TABLE(of, n2_crypto_match);
2200 static struct platform_driver n2_crypto_driver = {
2203 .of_match_table = n2_crypto_match,
2205 .probe = n2_crypto_probe,
2206 .remove = n2_crypto_remove,
2209 static const struct of_device_id n2_mau_match[] = {
2212 .compatible = "SUNW,n2-mau",
2216 .compatible = "SUNW,vf-mau",
2220 .compatible = "SUNW,kt-mau",
2225 MODULE_DEVICE_TABLE(of, n2_mau_match);
2227 static struct platform_driver n2_mau_driver = {
2230 .of_match_table = n2_mau_match,
2232 .probe = n2_mau_probe,
2233 .remove = n2_mau_remove,
2236 static struct platform_driver * const drivers[] = {
2241 static int __init n2_init(void)
2243 return platform_register_drivers(drivers, ARRAY_SIZE(drivers));
2246 static void __exit n2_exit(void)
2248 platform_unregister_drivers(drivers, ARRAY_SIZE(drivers));
2251 module_init(n2_init);
2252 module_exit(n2_exit);