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/internal/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/internal/skcipher.h>
27 #include <crypto/scatterwalk.h>
28 #include <crypto/algapi.h>
30 #include <asm/hypervisor.h>
31 #include <asm/mdesc.h>
35 #define DRV_MODULE_NAME "n2_crypto"
36 #define DRV_MODULE_VERSION "0.2"
37 #define DRV_MODULE_RELDATE "July 28, 2011"
39 static const char version[] =
40 DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
42 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
43 MODULE_DESCRIPTION("Niagara2 Crypto driver");
44 MODULE_LICENSE("GPL");
45 MODULE_VERSION(DRV_MODULE_VERSION);
47 #define N2_CRA_PRIORITY 200
49 static DEFINE_MUTEX(spu_lock);
53 unsigned long qhandle;
60 struct list_head jobs;
67 struct list_head list;
71 struct spu_queue *queue;
75 static struct spu_queue **cpu_to_cwq;
76 static struct spu_queue **cpu_to_mau;
78 static unsigned long spu_next_offset(struct spu_queue *q, unsigned long off)
80 if (q->q_type == HV_NCS_QTYPE_MAU) {
81 off += MAU_ENTRY_SIZE;
82 if (off == (MAU_ENTRY_SIZE * MAU_NUM_ENTRIES))
85 off += CWQ_ENTRY_SIZE;
86 if (off == (CWQ_ENTRY_SIZE * CWQ_NUM_ENTRIES))
92 struct n2_request_common {
93 struct list_head entry;
96 #define OFFSET_NOT_RUNNING (~(unsigned int)0)
98 /* An async job request records the final tail value it used in
99 * n2_request_common->offset, test to see if that offset is in
100 * the range old_head, new_head, inclusive.
102 static inline bool job_finished(struct spu_queue *q, unsigned int offset,
103 unsigned long old_head, unsigned long new_head)
105 if (old_head <= new_head) {
106 if (offset > old_head && offset <= new_head)
109 if (offset > old_head || offset <= new_head)
115 /* When the HEAD marker is unequal to the actual HEAD, we get
116 * a virtual device INO interrupt. We should process the
117 * completed CWQ entries and adjust the HEAD marker to clear
120 static irqreturn_t cwq_intr(int irq, void *dev_id)
122 unsigned long off, new_head, hv_ret;
123 struct spu_queue *q = dev_id;
125 pr_err("CPU[%d]: Got CWQ interrupt for qhdl[%lx]\n",
126 smp_processor_id(), q->qhandle);
130 hv_ret = sun4v_ncs_gethead(q->qhandle, &new_head);
132 pr_err("CPU[%d]: CWQ gethead[%lx] hv_ret[%lu]\n",
133 smp_processor_id(), new_head, hv_ret);
135 for (off = q->head; off != new_head; off = spu_next_offset(q, off)) {
139 hv_ret = sun4v_ncs_sethead_marker(q->qhandle, new_head);
140 if (hv_ret == HV_EOK)
143 spin_unlock(&q->lock);
148 static irqreturn_t mau_intr(int irq, void *dev_id)
150 struct spu_queue *q = dev_id;
151 unsigned long head, hv_ret;
155 pr_err("CPU[%d]: Got MAU interrupt for qhdl[%lx]\n",
156 smp_processor_id(), q->qhandle);
158 hv_ret = sun4v_ncs_gethead(q->qhandle, &head);
160 pr_err("CPU[%d]: MAU gethead[%lx] hv_ret[%lu]\n",
161 smp_processor_id(), head, hv_ret);
163 sun4v_ncs_sethead_marker(q->qhandle, head);
165 spin_unlock(&q->lock);
170 static void *spu_queue_next(struct spu_queue *q, void *cur)
172 return q->q + spu_next_offset(q, cur - q->q);
175 static int spu_queue_num_free(struct spu_queue *q)
177 unsigned long head = q->head;
178 unsigned long tail = q->tail;
179 unsigned long end = (CWQ_ENTRY_SIZE * CWQ_NUM_ENTRIES);
185 diff = (end - tail) + head;
187 return (diff / CWQ_ENTRY_SIZE) - 1;
190 static void *spu_queue_alloc(struct spu_queue *q, int num_entries)
192 int avail = spu_queue_num_free(q);
194 if (avail >= num_entries)
195 return q->q + q->tail;
200 static unsigned long spu_queue_submit(struct spu_queue *q, void *last)
202 unsigned long hv_ret, new_tail;
204 new_tail = spu_next_offset(q, last - q->q);
206 hv_ret = sun4v_ncs_settail(q->qhandle, new_tail);
207 if (hv_ret == HV_EOK)
212 static u64 control_word_base(unsigned int len, unsigned int hmac_key_len,
213 int enc_type, int auth_type,
214 unsigned int hash_len,
215 bool sfas, bool sob, bool eob, bool encrypt,
218 u64 word = (len - 1) & CONTROL_LEN;
220 word |= ((u64) opcode << CONTROL_OPCODE_SHIFT);
221 word |= ((u64) enc_type << CONTROL_ENC_TYPE_SHIFT);
222 word |= ((u64) auth_type << CONTROL_AUTH_TYPE_SHIFT);
224 word |= CONTROL_STORE_FINAL_AUTH_STATE;
226 word |= CONTROL_START_OF_BLOCK;
228 word |= CONTROL_END_OF_BLOCK;
230 word |= CONTROL_ENCRYPT;
232 word |= ((u64) (hmac_key_len - 1)) << CONTROL_HMAC_KEY_LEN_SHIFT;
234 word |= ((u64) (hash_len - 1)) << CONTROL_HASH_LEN_SHIFT;
240 static inline bool n2_should_run_async(struct spu_queue *qp, int this_len)
242 if (this_len >= 64 ||
243 qp->head != qp->tail)
249 struct n2_ahash_alg {
250 struct list_head entry;
252 const u32 *hash_init;
257 struct ahash_alg alg;
260 static inline struct n2_ahash_alg *n2_ahash_alg(struct crypto_tfm *tfm)
262 struct crypto_alg *alg = tfm->__crt_alg;
263 struct ahash_alg *ahash_alg;
265 ahash_alg = container_of(alg, struct ahash_alg, halg.base);
267 return container_of(ahash_alg, struct n2_ahash_alg, alg);
271 const char *child_alg;
272 struct n2_ahash_alg derived;
275 static inline struct n2_hmac_alg *n2_hmac_alg(struct crypto_tfm *tfm)
277 struct crypto_alg *alg = tfm->__crt_alg;
278 struct ahash_alg *ahash_alg;
280 ahash_alg = container_of(alg, struct ahash_alg, halg.base);
282 return container_of(ahash_alg, struct n2_hmac_alg, derived.alg);
286 struct crypto_ahash *fallback_tfm;
289 #define N2_HASH_KEY_MAX 32 /* HW limit for all HMAC requests */
292 struct n2_hash_ctx base;
294 struct crypto_shash *child_shash;
297 unsigned char hash_key[N2_HASH_KEY_MAX];
300 struct n2_hash_req_ctx {
302 struct md5_state md5;
303 struct sha1_state sha1;
304 struct sha256_state sha256;
307 struct ahash_request fallback_req;
310 static int n2_hash_async_init(struct ahash_request *req)
312 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
313 struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
314 struct n2_hash_ctx *ctx = crypto_ahash_ctx(tfm);
316 ahash_request_set_tfm(&rctx->fallback_req, ctx->fallback_tfm);
317 rctx->fallback_req.base.flags = req->base.flags & CRYPTO_TFM_REQ_MAY_SLEEP;
319 return crypto_ahash_init(&rctx->fallback_req);
322 static int n2_hash_async_update(struct ahash_request *req)
324 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
325 struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
326 struct n2_hash_ctx *ctx = crypto_ahash_ctx(tfm);
328 ahash_request_set_tfm(&rctx->fallback_req, ctx->fallback_tfm);
329 rctx->fallback_req.base.flags = req->base.flags & CRYPTO_TFM_REQ_MAY_SLEEP;
330 rctx->fallback_req.nbytes = req->nbytes;
331 rctx->fallback_req.src = req->src;
333 return crypto_ahash_update(&rctx->fallback_req);
336 static int n2_hash_async_final(struct ahash_request *req)
338 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
339 struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
340 struct n2_hash_ctx *ctx = crypto_ahash_ctx(tfm);
342 ahash_request_set_tfm(&rctx->fallback_req, ctx->fallback_tfm);
343 rctx->fallback_req.base.flags = req->base.flags & CRYPTO_TFM_REQ_MAY_SLEEP;
344 rctx->fallback_req.result = req->result;
346 return crypto_ahash_final(&rctx->fallback_req);
349 static int n2_hash_async_finup(struct ahash_request *req)
351 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
352 struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
353 struct n2_hash_ctx *ctx = crypto_ahash_ctx(tfm);
355 ahash_request_set_tfm(&rctx->fallback_req, ctx->fallback_tfm);
356 rctx->fallback_req.base.flags = req->base.flags & CRYPTO_TFM_REQ_MAY_SLEEP;
357 rctx->fallback_req.nbytes = req->nbytes;
358 rctx->fallback_req.src = req->src;
359 rctx->fallback_req.result = req->result;
361 return crypto_ahash_finup(&rctx->fallback_req);
364 static int n2_hash_async_noimport(struct ahash_request *req, const void *in)
369 static int n2_hash_async_noexport(struct ahash_request *req, void *out)
374 static int n2_hash_cra_init(struct crypto_tfm *tfm)
376 const char *fallback_driver_name = crypto_tfm_alg_name(tfm);
377 struct crypto_ahash *ahash = __crypto_ahash_cast(tfm);
378 struct n2_hash_ctx *ctx = crypto_ahash_ctx(ahash);
379 struct crypto_ahash *fallback_tfm;
382 fallback_tfm = crypto_alloc_ahash(fallback_driver_name, 0,
383 CRYPTO_ALG_NEED_FALLBACK);
384 if (IS_ERR(fallback_tfm)) {
385 pr_warn("Fallback driver '%s' could not be loaded!\n",
386 fallback_driver_name);
387 err = PTR_ERR(fallback_tfm);
391 crypto_ahash_set_reqsize(ahash, (sizeof(struct n2_hash_req_ctx) +
392 crypto_ahash_reqsize(fallback_tfm)));
394 ctx->fallback_tfm = fallback_tfm;
401 static void n2_hash_cra_exit(struct crypto_tfm *tfm)
403 struct crypto_ahash *ahash = __crypto_ahash_cast(tfm);
404 struct n2_hash_ctx *ctx = crypto_ahash_ctx(ahash);
406 crypto_free_ahash(ctx->fallback_tfm);
409 static int n2_hmac_cra_init(struct crypto_tfm *tfm)
411 const char *fallback_driver_name = crypto_tfm_alg_name(tfm);
412 struct crypto_ahash *ahash = __crypto_ahash_cast(tfm);
413 struct n2_hmac_ctx *ctx = crypto_ahash_ctx(ahash);
414 struct n2_hmac_alg *n2alg = n2_hmac_alg(tfm);
415 struct crypto_ahash *fallback_tfm;
416 struct crypto_shash *child_shash;
419 fallback_tfm = crypto_alloc_ahash(fallback_driver_name, 0,
420 CRYPTO_ALG_NEED_FALLBACK);
421 if (IS_ERR(fallback_tfm)) {
422 pr_warn("Fallback driver '%s' could not be loaded!\n",
423 fallback_driver_name);
424 err = PTR_ERR(fallback_tfm);
428 child_shash = crypto_alloc_shash(n2alg->child_alg, 0, 0);
429 if (IS_ERR(child_shash)) {
430 pr_warn("Child shash '%s' could not be loaded!\n",
432 err = PTR_ERR(child_shash);
433 goto out_free_fallback;
436 crypto_ahash_set_reqsize(ahash, (sizeof(struct n2_hash_req_ctx) +
437 crypto_ahash_reqsize(fallback_tfm)));
439 ctx->child_shash = child_shash;
440 ctx->base.fallback_tfm = fallback_tfm;
444 crypto_free_ahash(fallback_tfm);
450 static void n2_hmac_cra_exit(struct crypto_tfm *tfm)
452 struct crypto_ahash *ahash = __crypto_ahash_cast(tfm);
453 struct n2_hmac_ctx *ctx = crypto_ahash_ctx(ahash);
455 crypto_free_ahash(ctx->base.fallback_tfm);
456 crypto_free_shash(ctx->child_shash);
459 static int n2_hmac_async_setkey(struct crypto_ahash *tfm, const u8 *key,
462 struct n2_hmac_ctx *ctx = crypto_ahash_ctx(tfm);
463 struct crypto_shash *child_shash = ctx->child_shash;
464 struct crypto_ahash *fallback_tfm;
467 fallback_tfm = ctx->base.fallback_tfm;
468 err = crypto_ahash_setkey(fallback_tfm, key, keylen);
472 bs = crypto_shash_blocksize(child_shash);
473 ds = crypto_shash_digestsize(child_shash);
474 BUG_ON(ds > N2_HASH_KEY_MAX);
476 err = crypto_shash_tfm_digest(child_shash, key, keylen,
481 } else if (keylen <= N2_HASH_KEY_MAX)
482 memcpy(ctx->hash_key, key, keylen);
484 ctx->hash_key_len = keylen;
489 static unsigned long wait_for_tail(struct spu_queue *qp)
491 unsigned long head, hv_ret;
494 hv_ret = sun4v_ncs_gethead(qp->qhandle, &head);
495 if (hv_ret != HV_EOK) {
496 pr_err("Hypervisor error on gethead\n");
499 if (head == qp->tail) {
507 static unsigned long submit_and_wait_for_tail(struct spu_queue *qp,
508 struct cwq_initial_entry *ent)
510 unsigned long hv_ret = spu_queue_submit(qp, ent);
512 if (hv_ret == HV_EOK)
513 hv_ret = wait_for_tail(qp);
518 static int n2_do_async_digest(struct ahash_request *req,
519 unsigned int auth_type, unsigned int digest_size,
520 unsigned int result_size, void *hash_loc,
521 unsigned long auth_key, unsigned int auth_key_len)
523 struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
524 struct cwq_initial_entry *ent;
525 struct crypto_hash_walk walk;
526 struct spu_queue *qp;
531 /* The total effective length of the operation may not
534 if (unlikely(req->nbytes > (1 << 16))) {
535 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
536 struct n2_hash_ctx *ctx = crypto_ahash_ctx(tfm);
538 ahash_request_set_tfm(&rctx->fallback_req, ctx->fallback_tfm);
539 rctx->fallback_req.base.flags =
540 req->base.flags & CRYPTO_TFM_REQ_MAY_SLEEP;
541 rctx->fallback_req.nbytes = req->nbytes;
542 rctx->fallback_req.src = req->src;
543 rctx->fallback_req.result = req->result;
545 return crypto_ahash_digest(&rctx->fallback_req);
548 nbytes = crypto_hash_walk_first(req, &walk);
551 qp = cpu_to_cwq[cpu];
555 spin_lock_irqsave(&qp->lock, flags);
557 /* XXX can do better, improve this later by doing a by-hand scatterlist
560 ent = qp->q + qp->tail;
562 ent->control = control_word_base(nbytes, auth_key_len, 0,
563 auth_type, digest_size,
564 false, true, false, false,
567 ent->src_addr = __pa(walk.data);
568 ent->auth_key_addr = auth_key;
569 ent->auth_iv_addr = __pa(hash_loc);
570 ent->final_auth_state_addr = 0UL;
571 ent->enc_key_addr = 0UL;
572 ent->enc_iv_addr = 0UL;
573 ent->dest_addr = __pa(hash_loc);
575 nbytes = crypto_hash_walk_done(&walk, 0);
577 ent = spu_queue_next(qp, ent);
579 ent->control = (nbytes - 1);
580 ent->src_addr = __pa(walk.data);
581 ent->auth_key_addr = 0UL;
582 ent->auth_iv_addr = 0UL;
583 ent->final_auth_state_addr = 0UL;
584 ent->enc_key_addr = 0UL;
585 ent->enc_iv_addr = 0UL;
586 ent->dest_addr = 0UL;
588 nbytes = crypto_hash_walk_done(&walk, 0);
590 ent->control |= CONTROL_END_OF_BLOCK;
592 if (submit_and_wait_for_tail(qp, ent) != HV_EOK)
597 spin_unlock_irqrestore(&qp->lock, flags);
600 memcpy(req->result, hash_loc, result_size);
607 static int n2_hash_async_digest(struct ahash_request *req)
609 struct n2_ahash_alg *n2alg = n2_ahash_alg(req->base.tfm);
610 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
613 ds = n2alg->digest_size;
614 if (unlikely(req->nbytes == 0)) {
615 memcpy(req->result, n2alg->hash_zero, ds);
618 memcpy(&rctx->u, n2alg->hash_init, n2alg->hw_op_hashsz);
620 return n2_do_async_digest(req, n2alg->auth_type,
621 n2alg->hw_op_hashsz, ds,
625 static int n2_hmac_async_digest(struct ahash_request *req)
627 struct n2_hmac_alg *n2alg = n2_hmac_alg(req->base.tfm);
628 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
629 struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
630 struct n2_hmac_ctx *ctx = crypto_ahash_ctx(tfm);
633 ds = n2alg->derived.digest_size;
634 if (unlikely(req->nbytes == 0) ||
635 unlikely(ctx->hash_key_len > N2_HASH_KEY_MAX)) {
636 struct n2_hash_req_ctx *rctx = ahash_request_ctx(req);
637 struct n2_hash_ctx *ctx = crypto_ahash_ctx(tfm);
639 ahash_request_set_tfm(&rctx->fallback_req, ctx->fallback_tfm);
640 rctx->fallback_req.base.flags =
641 req->base.flags & CRYPTO_TFM_REQ_MAY_SLEEP;
642 rctx->fallback_req.nbytes = req->nbytes;
643 rctx->fallback_req.src = req->src;
644 rctx->fallback_req.result = req->result;
646 return crypto_ahash_digest(&rctx->fallback_req);
648 memcpy(&rctx->u, n2alg->derived.hash_init,
649 n2alg->derived.hw_op_hashsz);
651 return n2_do_async_digest(req, n2alg->derived.hmac_type,
652 n2alg->derived.hw_op_hashsz, ds,
654 __pa(&ctx->hash_key),
658 struct n2_skcipher_context {
662 u8 aes[AES_MAX_KEY_SIZE];
663 u8 des[DES_KEY_SIZE];
664 u8 des3[3 * DES_KEY_SIZE];
665 u8 arc4[258]; /* S-box, X, Y */
669 #define N2_CHUNK_ARR_LEN 16
671 struct n2_crypto_chunk {
672 struct list_head entry;
673 unsigned long iv_paddr : 44;
674 unsigned long arr_len : 20;
675 unsigned long dest_paddr;
676 unsigned long dest_final;
678 unsigned long src_paddr : 44;
679 unsigned long src_len : 20;
680 } arr[N2_CHUNK_ARR_LEN];
683 struct n2_request_context {
684 struct skcipher_walk walk;
685 struct list_head chunk_list;
686 struct n2_crypto_chunk chunk;
690 /* The SPU allows some level of flexibility for partial cipher blocks
691 * being specified in a descriptor.
693 * It merely requires that every descriptor's length field is at least
694 * as large as the cipher block size. This means that a cipher block
695 * can span at most 2 descriptors. However, this does not allow a
696 * partial block to span into the final descriptor as that would
697 * violate the rule (since every descriptor's length must be at lest
698 * the block size). So, for example, assuming an 8 byte block size:
700 * 0xe --> 0xa --> 0x8
702 * is a valid length sequence, whereas:
704 * 0xe --> 0xb --> 0x7
706 * is not a valid sequence.
709 struct n2_skcipher_alg {
710 struct list_head entry;
712 struct skcipher_alg skcipher;
715 static inline struct n2_skcipher_alg *n2_skcipher_alg(struct crypto_skcipher *tfm)
717 struct skcipher_alg *alg = crypto_skcipher_alg(tfm);
719 return container_of(alg, struct n2_skcipher_alg, skcipher);
722 struct n2_skcipher_request_context {
723 struct skcipher_walk walk;
726 static int n2_aes_setkey(struct crypto_skcipher *skcipher, const u8 *key,
729 struct crypto_tfm *tfm = crypto_skcipher_tfm(skcipher);
730 struct n2_skcipher_context *ctx = crypto_tfm_ctx(tfm);
731 struct n2_skcipher_alg *n2alg = n2_skcipher_alg(skcipher);
733 ctx->enc_type = (n2alg->enc_type & ENC_TYPE_CHAINING_MASK);
736 case AES_KEYSIZE_128:
737 ctx->enc_type |= ENC_TYPE_ALG_AES128;
739 case AES_KEYSIZE_192:
740 ctx->enc_type |= ENC_TYPE_ALG_AES192;
742 case AES_KEYSIZE_256:
743 ctx->enc_type |= ENC_TYPE_ALG_AES256;
749 ctx->key_len = keylen;
750 memcpy(ctx->key.aes, key, keylen);
754 static int n2_des_setkey(struct crypto_skcipher *skcipher, const u8 *key,
757 struct crypto_tfm *tfm = crypto_skcipher_tfm(skcipher);
758 struct n2_skcipher_context *ctx = crypto_tfm_ctx(tfm);
759 struct n2_skcipher_alg *n2alg = n2_skcipher_alg(skcipher);
762 err = verify_skcipher_des_key(skcipher, key);
766 ctx->enc_type = n2alg->enc_type;
768 ctx->key_len = keylen;
769 memcpy(ctx->key.des, key, keylen);
773 static int n2_3des_setkey(struct crypto_skcipher *skcipher, const u8 *key,
776 struct crypto_tfm *tfm = crypto_skcipher_tfm(skcipher);
777 struct n2_skcipher_context *ctx = crypto_tfm_ctx(tfm);
778 struct n2_skcipher_alg *n2alg = n2_skcipher_alg(skcipher);
781 err = verify_skcipher_des3_key(skcipher, key);
785 ctx->enc_type = n2alg->enc_type;
787 ctx->key_len = keylen;
788 memcpy(ctx->key.des3, key, keylen);
792 static int n2_arc4_setkey(struct crypto_skcipher *skcipher, const u8 *key,
795 struct crypto_tfm *tfm = crypto_skcipher_tfm(skcipher);
796 struct n2_skcipher_context *ctx = crypto_tfm_ctx(tfm);
797 struct n2_skcipher_alg *n2alg = n2_skcipher_alg(skcipher);
798 u8 *s = ctx->key.arc4;
803 ctx->enc_type = n2alg->enc_type;
808 for (i = 0; i < 256; i++)
810 for (i = 0; i < 256; i++) {
812 j = (j + key[k] + a) & 0xff;
822 static inline int skcipher_descriptor_len(int nbytes, unsigned int block_size)
824 int this_len = nbytes;
826 this_len -= (nbytes & (block_size - 1));
827 return this_len > (1 << 16) ? (1 << 16) : this_len;
830 static int __n2_crypt_chunk(struct crypto_skcipher *skcipher,
831 struct n2_crypto_chunk *cp,
832 struct spu_queue *qp, bool encrypt)
834 struct n2_skcipher_context *ctx = crypto_skcipher_ctx(skcipher);
835 struct cwq_initial_entry *ent;
839 ent = spu_queue_alloc(qp, cp->arr_len);
841 pr_info("queue_alloc() of %d fails\n",
846 in_place = (cp->dest_paddr == cp->arr[0].src_paddr);
848 ent->control = control_word_base(cp->arr[0].src_len,
849 0, ctx->enc_type, 0, 0,
850 false, true, false, encrypt,
852 (in_place ? OPCODE_INPLACE_BIT : 0));
853 ent->src_addr = cp->arr[0].src_paddr;
854 ent->auth_key_addr = 0UL;
855 ent->auth_iv_addr = 0UL;
856 ent->final_auth_state_addr = 0UL;
857 ent->enc_key_addr = __pa(&ctx->key);
858 ent->enc_iv_addr = cp->iv_paddr;
859 ent->dest_addr = (in_place ? 0UL : cp->dest_paddr);
861 for (i = 1; i < cp->arr_len; i++) {
862 ent = spu_queue_next(qp, ent);
864 ent->control = cp->arr[i].src_len - 1;
865 ent->src_addr = cp->arr[i].src_paddr;
866 ent->auth_key_addr = 0UL;
867 ent->auth_iv_addr = 0UL;
868 ent->final_auth_state_addr = 0UL;
869 ent->enc_key_addr = 0UL;
870 ent->enc_iv_addr = 0UL;
871 ent->dest_addr = 0UL;
873 ent->control |= CONTROL_END_OF_BLOCK;
875 return (spu_queue_submit(qp, ent) != HV_EOK) ? -EINVAL : 0;
878 static int n2_compute_chunks(struct skcipher_request *req)
880 struct n2_request_context *rctx = skcipher_request_ctx(req);
881 struct skcipher_walk *walk = &rctx->walk;
882 struct n2_crypto_chunk *chunk;
883 unsigned long dest_prev;
884 unsigned int tot_len;
888 err = skcipher_walk_async(walk, req);
892 INIT_LIST_HEAD(&rctx->chunk_list);
894 chunk = &rctx->chunk;
895 INIT_LIST_HEAD(&chunk->entry);
897 chunk->iv_paddr = 0UL;
899 chunk->dest_paddr = 0UL;
901 prev_in_place = false;
905 while ((nbytes = walk->nbytes) != 0) {
906 unsigned long dest_paddr, src_paddr;
910 src_paddr = (page_to_phys(walk->src.phys.page) +
911 walk->src.phys.offset);
912 dest_paddr = (page_to_phys(walk->dst.phys.page) +
913 walk->dst.phys.offset);
914 in_place = (src_paddr == dest_paddr);
915 this_len = skcipher_descriptor_len(nbytes, walk->blocksize);
917 if (chunk->arr_len != 0) {
918 if (in_place != prev_in_place ||
920 dest_paddr != dest_prev) ||
921 chunk->arr_len == N2_CHUNK_ARR_LEN ||
922 tot_len + this_len > (1 << 16)) {
923 chunk->dest_final = dest_prev;
924 list_add_tail(&chunk->entry,
926 chunk = kzalloc(sizeof(*chunk), GFP_ATOMIC);
931 INIT_LIST_HEAD(&chunk->entry);
934 if (chunk->arr_len == 0) {
935 chunk->dest_paddr = dest_paddr;
938 chunk->arr[chunk->arr_len].src_paddr = src_paddr;
939 chunk->arr[chunk->arr_len].src_len = this_len;
942 dest_prev = dest_paddr + this_len;
943 prev_in_place = in_place;
946 err = skcipher_walk_done(walk, nbytes - this_len);
950 if (!err && chunk->arr_len != 0) {
951 chunk->dest_final = dest_prev;
952 list_add_tail(&chunk->entry, &rctx->chunk_list);
958 static void n2_chunk_complete(struct skcipher_request *req, void *final_iv)
960 struct n2_request_context *rctx = skcipher_request_ctx(req);
961 struct n2_crypto_chunk *c, *tmp;
964 memcpy(rctx->walk.iv, final_iv, rctx->walk.blocksize);
966 list_for_each_entry_safe(c, tmp, &rctx->chunk_list, entry) {
968 if (unlikely(c != &rctx->chunk))
974 static int n2_do_ecb(struct skcipher_request *req, bool encrypt)
976 struct n2_request_context *rctx = skcipher_request_ctx(req);
977 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
978 int err = n2_compute_chunks(req);
979 struct n2_crypto_chunk *c, *tmp;
980 unsigned long flags, hv_ret;
981 struct spu_queue *qp;
986 qp = cpu_to_cwq[get_cpu()];
991 spin_lock_irqsave(&qp->lock, flags);
993 list_for_each_entry_safe(c, tmp, &rctx->chunk_list, entry) {
994 err = __n2_crypt_chunk(tfm, c, qp, encrypt);
998 if (unlikely(c != &rctx->chunk))
1002 hv_ret = wait_for_tail(qp);
1003 if (hv_ret != HV_EOK)
1007 spin_unlock_irqrestore(&qp->lock, flags);
1012 n2_chunk_complete(req, NULL);
1016 static int n2_encrypt_ecb(struct skcipher_request *req)
1018 return n2_do_ecb(req, true);
1021 static int n2_decrypt_ecb(struct skcipher_request *req)
1023 return n2_do_ecb(req, false);
1026 static int n2_do_chaining(struct skcipher_request *req, bool encrypt)
1028 struct n2_request_context *rctx = skcipher_request_ctx(req);
1029 struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
1030 unsigned long flags, hv_ret, iv_paddr;
1031 int err = n2_compute_chunks(req);
1032 struct n2_crypto_chunk *c, *tmp;
1033 struct spu_queue *qp;
1034 void *final_iv_addr;
1036 final_iv_addr = NULL;
1041 qp = cpu_to_cwq[get_cpu()];
1046 spin_lock_irqsave(&qp->lock, flags);
1049 iv_paddr = __pa(rctx->walk.iv);
1050 list_for_each_entry_safe(c, tmp, &rctx->chunk_list,
1052 c->iv_paddr = iv_paddr;
1053 err = __n2_crypt_chunk(tfm, c, qp, true);
1056 iv_paddr = c->dest_final - rctx->walk.blocksize;
1057 list_del(&c->entry);
1058 if (unlikely(c != &rctx->chunk))
1061 final_iv_addr = __va(iv_paddr);
1063 list_for_each_entry_safe_reverse(c, tmp, &rctx->chunk_list,
1065 if (c == &rctx->chunk) {
1066 iv_paddr = __pa(rctx->walk.iv);
1068 iv_paddr = (tmp->arr[tmp->arr_len-1].src_paddr +
1069 tmp->arr[tmp->arr_len-1].src_len -
1070 rctx->walk.blocksize);
1072 if (!final_iv_addr) {
1075 pa = (c->arr[c->arr_len-1].src_paddr +
1076 c->arr[c->arr_len-1].src_len -
1077 rctx->walk.blocksize);
1078 final_iv_addr = rctx->temp_iv;
1079 memcpy(rctx->temp_iv, __va(pa),
1080 rctx->walk.blocksize);
1082 c->iv_paddr = iv_paddr;
1083 err = __n2_crypt_chunk(tfm, c, qp, false);
1086 list_del(&c->entry);
1087 if (unlikely(c != &rctx->chunk))
1092 hv_ret = wait_for_tail(qp);
1093 if (hv_ret != HV_EOK)
1097 spin_unlock_irqrestore(&qp->lock, flags);
1102 n2_chunk_complete(req, err ? NULL : final_iv_addr);
1106 static int n2_encrypt_chaining(struct skcipher_request *req)
1108 return n2_do_chaining(req, true);
1111 static int n2_decrypt_chaining(struct skcipher_request *req)
1113 return n2_do_chaining(req, false);
1116 struct n2_skcipher_tmpl {
1118 const char *drv_name;
1121 struct skcipher_alg skcipher;
1124 static const struct n2_skcipher_tmpl skcipher_tmpls[] = {
1125 /* ARC4: only ECB is supported (chaining bits ignored) */
1126 { .name = "ecb(arc4)",
1127 .drv_name = "ecb-arc4",
1129 .enc_type = (ENC_TYPE_ALG_RC4_STREAM |
1130 ENC_TYPE_CHAINING_ECB),
1134 .setkey = n2_arc4_setkey,
1135 .encrypt = n2_encrypt_ecb,
1136 .decrypt = n2_decrypt_ecb,
1140 /* DES: ECB CBC and CFB are supported */
1141 { .name = "ecb(des)",
1142 .drv_name = "ecb-des",
1143 .block_size = DES_BLOCK_SIZE,
1144 .enc_type = (ENC_TYPE_ALG_DES |
1145 ENC_TYPE_CHAINING_ECB),
1147 .min_keysize = DES_KEY_SIZE,
1148 .max_keysize = DES_KEY_SIZE,
1149 .setkey = n2_des_setkey,
1150 .encrypt = n2_encrypt_ecb,
1151 .decrypt = n2_decrypt_ecb,
1154 { .name = "cbc(des)",
1155 .drv_name = "cbc-des",
1156 .block_size = DES_BLOCK_SIZE,
1157 .enc_type = (ENC_TYPE_ALG_DES |
1158 ENC_TYPE_CHAINING_CBC),
1160 .ivsize = DES_BLOCK_SIZE,
1161 .min_keysize = DES_KEY_SIZE,
1162 .max_keysize = DES_KEY_SIZE,
1163 .setkey = n2_des_setkey,
1164 .encrypt = n2_encrypt_chaining,
1165 .decrypt = n2_decrypt_chaining,
1168 { .name = "cfb(des)",
1169 .drv_name = "cfb-des",
1170 .block_size = DES_BLOCK_SIZE,
1171 .enc_type = (ENC_TYPE_ALG_DES |
1172 ENC_TYPE_CHAINING_CFB),
1174 .min_keysize = DES_KEY_SIZE,
1175 .max_keysize = DES_KEY_SIZE,
1176 .setkey = n2_des_setkey,
1177 .encrypt = n2_encrypt_chaining,
1178 .decrypt = n2_decrypt_chaining,
1182 /* 3DES: ECB CBC and CFB are supported */
1183 { .name = "ecb(des3_ede)",
1184 .drv_name = "ecb-3des",
1185 .block_size = DES_BLOCK_SIZE,
1186 .enc_type = (ENC_TYPE_ALG_3DES |
1187 ENC_TYPE_CHAINING_ECB),
1189 .min_keysize = 3 * DES_KEY_SIZE,
1190 .max_keysize = 3 * DES_KEY_SIZE,
1191 .setkey = n2_3des_setkey,
1192 .encrypt = n2_encrypt_ecb,
1193 .decrypt = n2_decrypt_ecb,
1196 { .name = "cbc(des3_ede)",
1197 .drv_name = "cbc-3des",
1198 .block_size = DES_BLOCK_SIZE,
1199 .enc_type = (ENC_TYPE_ALG_3DES |
1200 ENC_TYPE_CHAINING_CBC),
1202 .ivsize = DES_BLOCK_SIZE,
1203 .min_keysize = 3 * DES_KEY_SIZE,
1204 .max_keysize = 3 * DES_KEY_SIZE,
1205 .setkey = n2_3des_setkey,
1206 .encrypt = n2_encrypt_chaining,
1207 .decrypt = n2_decrypt_chaining,
1210 { .name = "cfb(des3_ede)",
1211 .drv_name = "cfb-3des",
1212 .block_size = DES_BLOCK_SIZE,
1213 .enc_type = (ENC_TYPE_ALG_3DES |
1214 ENC_TYPE_CHAINING_CFB),
1216 .min_keysize = 3 * DES_KEY_SIZE,
1217 .max_keysize = 3 * DES_KEY_SIZE,
1218 .setkey = n2_3des_setkey,
1219 .encrypt = n2_encrypt_chaining,
1220 .decrypt = n2_decrypt_chaining,
1223 /* AES: ECB CBC and CTR are supported */
1224 { .name = "ecb(aes)",
1225 .drv_name = "ecb-aes",
1226 .block_size = AES_BLOCK_SIZE,
1227 .enc_type = (ENC_TYPE_ALG_AES128 |
1228 ENC_TYPE_CHAINING_ECB),
1230 .min_keysize = AES_MIN_KEY_SIZE,
1231 .max_keysize = AES_MAX_KEY_SIZE,
1232 .setkey = n2_aes_setkey,
1233 .encrypt = n2_encrypt_ecb,
1234 .decrypt = n2_decrypt_ecb,
1237 { .name = "cbc(aes)",
1238 .drv_name = "cbc-aes",
1239 .block_size = AES_BLOCK_SIZE,
1240 .enc_type = (ENC_TYPE_ALG_AES128 |
1241 ENC_TYPE_CHAINING_CBC),
1243 .ivsize = AES_BLOCK_SIZE,
1244 .min_keysize = AES_MIN_KEY_SIZE,
1245 .max_keysize = AES_MAX_KEY_SIZE,
1246 .setkey = n2_aes_setkey,
1247 .encrypt = n2_encrypt_chaining,
1248 .decrypt = n2_decrypt_chaining,
1251 { .name = "ctr(aes)",
1252 .drv_name = "ctr-aes",
1253 .block_size = AES_BLOCK_SIZE,
1254 .enc_type = (ENC_TYPE_ALG_AES128 |
1255 ENC_TYPE_CHAINING_COUNTER),
1257 .ivsize = AES_BLOCK_SIZE,
1258 .min_keysize = AES_MIN_KEY_SIZE,
1259 .max_keysize = AES_MAX_KEY_SIZE,
1260 .setkey = n2_aes_setkey,
1261 .encrypt = n2_encrypt_chaining,
1262 .decrypt = n2_encrypt_chaining,
1267 #define NUM_CIPHER_TMPLS ARRAY_SIZE(skcipher_tmpls)
1269 static LIST_HEAD(skcipher_algs);
1271 struct n2_hash_tmpl {
1273 const u8 *hash_zero;
1274 const u32 *hash_init;
1282 static const u32 n2_md5_init[MD5_HASH_WORDS] = {
1283 cpu_to_le32(MD5_H0),
1284 cpu_to_le32(MD5_H1),
1285 cpu_to_le32(MD5_H2),
1286 cpu_to_le32(MD5_H3),
1288 static const u32 n2_sha1_init[SHA1_DIGEST_SIZE / 4] = {
1289 SHA1_H0, SHA1_H1, SHA1_H2, SHA1_H3, SHA1_H4,
1291 static const u32 n2_sha256_init[SHA256_DIGEST_SIZE / 4] = {
1292 SHA256_H0, SHA256_H1, SHA256_H2, SHA256_H3,
1293 SHA256_H4, SHA256_H5, SHA256_H6, SHA256_H7,
1295 static const u32 n2_sha224_init[SHA256_DIGEST_SIZE / 4] = {
1296 SHA224_H0, SHA224_H1, SHA224_H2, SHA224_H3,
1297 SHA224_H4, SHA224_H5, SHA224_H6, SHA224_H7,
1300 static const struct n2_hash_tmpl hash_tmpls[] = {
1302 .hash_zero = md5_zero_message_hash,
1303 .hash_init = n2_md5_init,
1304 .auth_type = AUTH_TYPE_MD5,
1305 .hmac_type = AUTH_TYPE_HMAC_MD5,
1306 .hw_op_hashsz = MD5_DIGEST_SIZE,
1307 .digest_size = MD5_DIGEST_SIZE,
1308 .block_size = MD5_HMAC_BLOCK_SIZE },
1310 .hash_zero = sha1_zero_message_hash,
1311 .hash_init = n2_sha1_init,
1312 .auth_type = AUTH_TYPE_SHA1,
1313 .hmac_type = AUTH_TYPE_HMAC_SHA1,
1314 .hw_op_hashsz = SHA1_DIGEST_SIZE,
1315 .digest_size = SHA1_DIGEST_SIZE,
1316 .block_size = SHA1_BLOCK_SIZE },
1318 .hash_zero = sha256_zero_message_hash,
1319 .hash_init = n2_sha256_init,
1320 .auth_type = AUTH_TYPE_SHA256,
1321 .hmac_type = AUTH_TYPE_HMAC_SHA256,
1322 .hw_op_hashsz = SHA256_DIGEST_SIZE,
1323 .digest_size = SHA256_DIGEST_SIZE,
1324 .block_size = SHA256_BLOCK_SIZE },
1326 .hash_zero = sha224_zero_message_hash,
1327 .hash_init = n2_sha224_init,
1328 .auth_type = AUTH_TYPE_SHA256,
1329 .hmac_type = AUTH_TYPE_RESERVED,
1330 .hw_op_hashsz = SHA256_DIGEST_SIZE,
1331 .digest_size = SHA224_DIGEST_SIZE,
1332 .block_size = SHA224_BLOCK_SIZE },
1334 #define NUM_HASH_TMPLS ARRAY_SIZE(hash_tmpls)
1336 static LIST_HEAD(ahash_algs);
1337 static LIST_HEAD(hmac_algs);
1339 static int algs_registered;
1341 static void __n2_unregister_algs(void)
1343 struct n2_skcipher_alg *skcipher, *skcipher_tmp;
1344 struct n2_ahash_alg *alg, *alg_tmp;
1345 struct n2_hmac_alg *hmac, *hmac_tmp;
1347 list_for_each_entry_safe(skcipher, skcipher_tmp, &skcipher_algs, entry) {
1348 crypto_unregister_skcipher(&skcipher->skcipher);
1349 list_del(&skcipher->entry);
1352 list_for_each_entry_safe(hmac, hmac_tmp, &hmac_algs, derived.entry) {
1353 crypto_unregister_ahash(&hmac->derived.alg);
1354 list_del(&hmac->derived.entry);
1357 list_for_each_entry_safe(alg, alg_tmp, &ahash_algs, entry) {
1358 crypto_unregister_ahash(&alg->alg);
1359 list_del(&alg->entry);
1364 static int n2_skcipher_init_tfm(struct crypto_skcipher *tfm)
1366 crypto_skcipher_set_reqsize(tfm, sizeof(struct n2_request_context));
1370 static int __n2_register_one_skcipher(const struct n2_skcipher_tmpl *tmpl)
1372 struct n2_skcipher_alg *p = kzalloc(sizeof(*p), GFP_KERNEL);
1373 struct skcipher_alg *alg;
1380 *alg = tmpl->skcipher;
1382 snprintf(alg->base.cra_name, CRYPTO_MAX_ALG_NAME, "%s", tmpl->name);
1383 snprintf(alg->base.cra_driver_name, CRYPTO_MAX_ALG_NAME, "%s-n2", tmpl->drv_name);
1384 alg->base.cra_priority = N2_CRA_PRIORITY;
1385 alg->base.cra_flags = CRYPTO_ALG_KERN_DRIVER_ONLY | CRYPTO_ALG_ASYNC;
1386 alg->base.cra_blocksize = tmpl->block_size;
1387 p->enc_type = tmpl->enc_type;
1388 alg->base.cra_ctxsize = sizeof(struct n2_skcipher_context);
1389 alg->base.cra_module = THIS_MODULE;
1390 alg->init = n2_skcipher_init_tfm;
1392 list_add(&p->entry, &skcipher_algs);
1393 err = crypto_register_skcipher(alg);
1395 pr_err("%s alg registration failed\n", alg->base.cra_name);
1396 list_del(&p->entry);
1399 pr_info("%s alg registered\n", alg->base.cra_name);
1404 static int __n2_register_one_hmac(struct n2_ahash_alg *n2ahash)
1406 struct n2_hmac_alg *p = kzalloc(sizeof(*p), GFP_KERNEL);
1407 struct ahash_alg *ahash;
1408 struct crypto_alg *base;
1414 p->child_alg = n2ahash->alg.halg.base.cra_name;
1415 memcpy(&p->derived, n2ahash, sizeof(struct n2_ahash_alg));
1416 INIT_LIST_HEAD(&p->derived.entry);
1418 ahash = &p->derived.alg;
1419 ahash->digest = n2_hmac_async_digest;
1420 ahash->setkey = n2_hmac_async_setkey;
1422 base = &ahash->halg.base;
1423 snprintf(base->cra_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)", p->child_alg);
1424 snprintf(base->cra_driver_name, CRYPTO_MAX_ALG_NAME, "hmac-%s-n2", p->child_alg);
1426 base->cra_ctxsize = sizeof(struct n2_hmac_ctx);
1427 base->cra_init = n2_hmac_cra_init;
1428 base->cra_exit = n2_hmac_cra_exit;
1430 list_add(&p->derived.entry, &hmac_algs);
1431 err = crypto_register_ahash(ahash);
1433 pr_err("%s alg registration failed\n", base->cra_name);
1434 list_del(&p->derived.entry);
1437 pr_info("%s alg registered\n", base->cra_name);
1442 static int __n2_register_one_ahash(const struct n2_hash_tmpl *tmpl)
1444 struct n2_ahash_alg *p = kzalloc(sizeof(*p), GFP_KERNEL);
1445 struct hash_alg_common *halg;
1446 struct crypto_alg *base;
1447 struct ahash_alg *ahash;
1453 p->hash_zero = tmpl->hash_zero;
1454 p->hash_init = tmpl->hash_init;
1455 p->auth_type = tmpl->auth_type;
1456 p->hmac_type = tmpl->hmac_type;
1457 p->hw_op_hashsz = tmpl->hw_op_hashsz;
1458 p->digest_size = tmpl->digest_size;
1461 ahash->init = n2_hash_async_init;
1462 ahash->update = n2_hash_async_update;
1463 ahash->final = n2_hash_async_final;
1464 ahash->finup = n2_hash_async_finup;
1465 ahash->digest = n2_hash_async_digest;
1466 ahash->export = n2_hash_async_noexport;
1467 ahash->import = n2_hash_async_noimport;
1469 halg = &ahash->halg;
1470 halg->digestsize = tmpl->digest_size;
1473 snprintf(base->cra_name, CRYPTO_MAX_ALG_NAME, "%s", tmpl->name);
1474 snprintf(base->cra_driver_name, CRYPTO_MAX_ALG_NAME, "%s-n2", tmpl->name);
1475 base->cra_priority = N2_CRA_PRIORITY;
1476 base->cra_flags = CRYPTO_ALG_KERN_DRIVER_ONLY |
1477 CRYPTO_ALG_NEED_FALLBACK;
1478 base->cra_blocksize = tmpl->block_size;
1479 base->cra_ctxsize = sizeof(struct n2_hash_ctx);
1480 base->cra_module = THIS_MODULE;
1481 base->cra_init = n2_hash_cra_init;
1482 base->cra_exit = n2_hash_cra_exit;
1484 list_add(&p->entry, &ahash_algs);
1485 err = crypto_register_ahash(ahash);
1487 pr_err("%s alg registration failed\n", base->cra_name);
1488 list_del(&p->entry);
1491 pr_info("%s alg registered\n", base->cra_name);
1493 if (!err && p->hmac_type != AUTH_TYPE_RESERVED)
1494 err = __n2_register_one_hmac(p);
1498 static int n2_register_algs(void)
1502 mutex_lock(&spu_lock);
1503 if (algs_registered++)
1506 for (i = 0; i < NUM_HASH_TMPLS; i++) {
1507 err = __n2_register_one_ahash(&hash_tmpls[i]);
1509 __n2_unregister_algs();
1513 for (i = 0; i < NUM_CIPHER_TMPLS; i++) {
1514 err = __n2_register_one_skcipher(&skcipher_tmpls[i]);
1516 __n2_unregister_algs();
1522 mutex_unlock(&spu_lock);
1526 static void n2_unregister_algs(void)
1528 mutex_lock(&spu_lock);
1529 if (!--algs_registered)
1530 __n2_unregister_algs();
1531 mutex_unlock(&spu_lock);
1534 /* To map CWQ queues to interrupt sources, the hypervisor API provides
1535 * a devino. This isn't very useful to us because all of the
1536 * interrupts listed in the device_node have been translated to
1537 * Linux virtual IRQ cookie numbers.
1539 * So we have to back-translate, going through the 'intr' and 'ino'
1540 * property tables of the n2cp MDESC node, matching it with the OF
1541 * 'interrupts' property entries, in order to to figure out which
1542 * devino goes to which already-translated IRQ.
1544 static int find_devino_index(struct platform_device *dev, struct spu_mdesc_info *ip,
1545 unsigned long dev_ino)
1547 const unsigned int *dev_intrs;
1551 for (i = 0; i < ip->num_intrs; i++) {
1552 if (ip->ino_table[i].ino == dev_ino)
1555 if (i == ip->num_intrs)
1558 intr = ip->ino_table[i].intr;
1560 dev_intrs = of_get_property(dev->dev.of_node, "interrupts", NULL);
1564 for (i = 0; i < dev->archdata.num_irqs; i++) {
1565 if (dev_intrs[i] == intr)
1572 static int spu_map_ino(struct platform_device *dev, struct spu_mdesc_info *ip,
1573 const char *irq_name, struct spu_queue *p,
1574 irq_handler_t handler)
1579 herr = sun4v_ncs_qhandle_to_devino(p->qhandle, &p->devino);
1583 index = find_devino_index(dev, ip, p->devino);
1587 p->irq = dev->archdata.irqs[index];
1589 sprintf(p->irq_name, "%s-%d", irq_name, index);
1591 return request_irq(p->irq, handler, 0, p->irq_name, p);
1594 static struct kmem_cache *queue_cache[2];
1596 static void *new_queue(unsigned long q_type)
1598 return kmem_cache_zalloc(queue_cache[q_type - 1], GFP_KERNEL);
1601 static void free_queue(void *p, unsigned long q_type)
1603 kmem_cache_free(queue_cache[q_type - 1], p);
1606 static int queue_cache_init(void)
1608 if (!queue_cache[HV_NCS_QTYPE_MAU - 1])
1609 queue_cache[HV_NCS_QTYPE_MAU - 1] =
1610 kmem_cache_create("mau_queue",
1613 MAU_ENTRY_SIZE, 0, NULL);
1614 if (!queue_cache[HV_NCS_QTYPE_MAU - 1])
1617 if (!queue_cache[HV_NCS_QTYPE_CWQ - 1])
1618 queue_cache[HV_NCS_QTYPE_CWQ - 1] =
1619 kmem_cache_create("cwq_queue",
1622 CWQ_ENTRY_SIZE, 0, NULL);
1623 if (!queue_cache[HV_NCS_QTYPE_CWQ - 1]) {
1624 kmem_cache_destroy(queue_cache[HV_NCS_QTYPE_MAU - 1]);
1625 queue_cache[HV_NCS_QTYPE_MAU - 1] = NULL;
1631 static void queue_cache_destroy(void)
1633 kmem_cache_destroy(queue_cache[HV_NCS_QTYPE_MAU - 1]);
1634 kmem_cache_destroy(queue_cache[HV_NCS_QTYPE_CWQ - 1]);
1635 queue_cache[HV_NCS_QTYPE_MAU - 1] = NULL;
1636 queue_cache[HV_NCS_QTYPE_CWQ - 1] = NULL;
1639 static long spu_queue_register_workfn(void *arg)
1641 struct spu_qreg *qr = arg;
1642 struct spu_queue *p = qr->queue;
1643 unsigned long q_type = qr->type;
1644 unsigned long hv_ret;
1646 hv_ret = sun4v_ncs_qconf(q_type, __pa(p->q),
1647 CWQ_NUM_ENTRIES, &p->qhandle);
1649 sun4v_ncs_sethead_marker(p->qhandle, 0);
1651 return hv_ret ? -EINVAL : 0;
1654 static int spu_queue_register(struct spu_queue *p, unsigned long q_type)
1656 int cpu = cpumask_any_and(&p->sharing, cpu_online_mask);
1657 struct spu_qreg qr = { .queue = p, .type = q_type };
1659 return work_on_cpu_safe(cpu, spu_queue_register_workfn, &qr);
1662 static int spu_queue_setup(struct spu_queue *p)
1666 p->q = new_queue(p->q_type);
1670 err = spu_queue_register(p, p->q_type);
1672 free_queue(p->q, p->q_type);
1679 static void spu_queue_destroy(struct spu_queue *p)
1681 unsigned long hv_ret;
1686 hv_ret = sun4v_ncs_qconf(p->q_type, p->qhandle, 0, &p->qhandle);
1689 free_queue(p->q, p->q_type);
1692 static void spu_list_destroy(struct list_head *list)
1694 struct spu_queue *p, *n;
1696 list_for_each_entry_safe(p, n, list, list) {
1699 for (i = 0; i < NR_CPUS; i++) {
1700 if (cpu_to_cwq[i] == p)
1701 cpu_to_cwq[i] = NULL;
1705 free_irq(p->irq, p);
1708 spu_queue_destroy(p);
1714 /* Walk the backward arcs of a CWQ 'exec-unit' node,
1715 * gathering cpu membership information.
1717 static int spu_mdesc_walk_arcs(struct mdesc_handle *mdesc,
1718 struct platform_device *dev,
1719 u64 node, struct spu_queue *p,
1720 struct spu_queue **table)
1724 mdesc_for_each_arc(arc, mdesc, node, MDESC_ARC_TYPE_BACK) {
1725 u64 tgt = mdesc_arc_target(mdesc, arc);
1726 const char *name = mdesc_node_name(mdesc, tgt);
1729 if (strcmp(name, "cpu"))
1731 id = mdesc_get_property(mdesc, tgt, "id", NULL);
1732 if (table[*id] != NULL) {
1733 dev_err(&dev->dev, "%pOF: SPU cpu slot already set.\n",
1737 cpumask_set_cpu(*id, &p->sharing);
1743 /* Process an 'exec-unit' MDESC node of type 'cwq'. */
1744 static int handle_exec_unit(struct spu_mdesc_info *ip, struct list_head *list,
1745 struct platform_device *dev, struct mdesc_handle *mdesc,
1746 u64 node, const char *iname, unsigned long q_type,
1747 irq_handler_t handler, struct spu_queue **table)
1749 struct spu_queue *p;
1752 p = kzalloc(sizeof(struct spu_queue), GFP_KERNEL);
1754 dev_err(&dev->dev, "%pOF: Could not allocate SPU queue.\n",
1759 cpumask_clear(&p->sharing);
1760 spin_lock_init(&p->lock);
1762 INIT_LIST_HEAD(&p->jobs);
1763 list_add(&p->list, list);
1765 err = spu_mdesc_walk_arcs(mdesc, dev, node, p, table);
1769 err = spu_queue_setup(p);
1773 return spu_map_ino(dev, ip, iname, p, handler);
1776 static int spu_mdesc_scan(struct mdesc_handle *mdesc, struct platform_device *dev,
1777 struct spu_mdesc_info *ip, struct list_head *list,
1778 const char *exec_name, unsigned long q_type,
1779 irq_handler_t handler, struct spu_queue **table)
1784 mdesc_for_each_node_by_name(mdesc, node, "exec-unit") {
1787 type = mdesc_get_property(mdesc, node, "type", NULL);
1788 if (!type || strcmp(type, exec_name))
1791 err = handle_exec_unit(ip, list, dev, mdesc, node,
1792 exec_name, q_type, handler, table);
1794 spu_list_destroy(list);
1802 static int get_irq_props(struct mdesc_handle *mdesc, u64 node,
1803 struct spu_mdesc_info *ip)
1809 ino = mdesc_get_property(mdesc, node, "ino", &ino_len);
1811 printk("NO 'ino'\n");
1815 ip->num_intrs = ino_len / sizeof(u64);
1816 ip->ino_table = kzalloc((sizeof(struct ino_blob) *
1822 for (i = 0; i < ip->num_intrs; i++) {
1823 struct ino_blob *b = &ip->ino_table[i];
1831 static int grab_mdesc_irq_props(struct mdesc_handle *mdesc,
1832 struct platform_device *dev,
1833 struct spu_mdesc_info *ip,
1834 const char *node_name)
1836 const unsigned int *reg;
1839 reg = of_get_property(dev->dev.of_node, "reg", NULL);
1843 mdesc_for_each_node_by_name(mdesc, node, "virtual-device") {
1847 name = mdesc_get_property(mdesc, node, "name", NULL);
1848 if (!name || strcmp(name, node_name))
1850 chdl = mdesc_get_property(mdesc, node, "cfg-handle", NULL);
1851 if (!chdl || (*chdl != *reg))
1853 ip->cfg_handle = *chdl;
1854 return get_irq_props(mdesc, node, ip);
1860 static unsigned long n2_spu_hvapi_major;
1861 static unsigned long n2_spu_hvapi_minor;
1863 static int n2_spu_hvapi_register(void)
1867 n2_spu_hvapi_major = 2;
1868 n2_spu_hvapi_minor = 0;
1870 err = sun4v_hvapi_register(HV_GRP_NCS,
1872 &n2_spu_hvapi_minor);
1875 pr_info("Registered NCS HVAPI version %lu.%lu\n",
1877 n2_spu_hvapi_minor);
1882 static void n2_spu_hvapi_unregister(void)
1884 sun4v_hvapi_unregister(HV_GRP_NCS);
1887 static int global_ref;
1889 static int grab_global_resources(void)
1893 mutex_lock(&spu_lock);
1898 err = n2_spu_hvapi_register();
1902 err = queue_cache_init();
1904 goto out_hvapi_release;
1907 cpu_to_cwq = kcalloc(NR_CPUS, sizeof(struct spu_queue *),
1910 goto out_queue_cache_destroy;
1912 cpu_to_mau = kcalloc(NR_CPUS, sizeof(struct spu_queue *),
1915 goto out_free_cwq_table;
1922 mutex_unlock(&spu_lock);
1929 out_queue_cache_destroy:
1930 queue_cache_destroy();
1933 n2_spu_hvapi_unregister();
1937 static void release_global_resources(void)
1939 mutex_lock(&spu_lock);
1940 if (!--global_ref) {
1947 queue_cache_destroy();
1948 n2_spu_hvapi_unregister();
1950 mutex_unlock(&spu_lock);
1953 static struct n2_crypto *alloc_n2cp(void)
1955 struct n2_crypto *np = kzalloc(sizeof(struct n2_crypto), GFP_KERNEL);
1958 INIT_LIST_HEAD(&np->cwq_list);
1963 static void free_n2cp(struct n2_crypto *np)
1965 kfree(np->cwq_info.ino_table);
1966 np->cwq_info.ino_table = NULL;
1971 static void n2_spu_driver_version(void)
1973 static int n2_spu_version_printed;
1975 if (n2_spu_version_printed++ == 0)
1976 pr_info("%s", version);
1979 static int n2_crypto_probe(struct platform_device *dev)
1981 struct mdesc_handle *mdesc;
1982 struct n2_crypto *np;
1985 n2_spu_driver_version();
1987 pr_info("Found N2CP at %pOF\n", dev->dev.of_node);
1991 dev_err(&dev->dev, "%pOF: Unable to allocate n2cp.\n",
1996 err = grab_global_resources();
1998 dev_err(&dev->dev, "%pOF: Unable to grab global resources.\n",
2003 mdesc = mdesc_grab();
2006 dev_err(&dev->dev, "%pOF: Unable to grab MDESC.\n",
2009 goto out_free_global;
2011 err = grab_mdesc_irq_props(mdesc, dev, &np->cwq_info, "n2cp");
2013 dev_err(&dev->dev, "%pOF: Unable to grab IRQ props.\n",
2015 mdesc_release(mdesc);
2016 goto out_free_global;
2019 err = spu_mdesc_scan(mdesc, dev, &np->cwq_info, &np->cwq_list,
2020 "cwq", HV_NCS_QTYPE_CWQ, cwq_intr,
2022 mdesc_release(mdesc);
2025 dev_err(&dev->dev, "%pOF: CWQ MDESC scan failed.\n",
2027 goto out_free_global;
2030 err = n2_register_algs();
2032 dev_err(&dev->dev, "%pOF: Unable to register algorithms.\n",
2034 goto out_free_spu_list;
2037 dev_set_drvdata(&dev->dev, np);
2042 spu_list_destroy(&np->cwq_list);
2045 release_global_resources();
2053 static int n2_crypto_remove(struct platform_device *dev)
2055 struct n2_crypto *np = dev_get_drvdata(&dev->dev);
2057 n2_unregister_algs();
2059 spu_list_destroy(&np->cwq_list);
2061 release_global_resources();
2068 static struct n2_mau *alloc_ncp(void)
2070 struct n2_mau *mp = kzalloc(sizeof(struct n2_mau), GFP_KERNEL);
2073 INIT_LIST_HEAD(&mp->mau_list);
2078 static void free_ncp(struct n2_mau *mp)
2080 kfree(mp->mau_info.ino_table);
2081 mp->mau_info.ino_table = NULL;
2086 static int n2_mau_probe(struct platform_device *dev)
2088 struct mdesc_handle *mdesc;
2092 n2_spu_driver_version();
2094 pr_info("Found NCP at %pOF\n", dev->dev.of_node);
2098 dev_err(&dev->dev, "%pOF: Unable to allocate ncp.\n",
2103 err = grab_global_resources();
2105 dev_err(&dev->dev, "%pOF: Unable to grab global resources.\n",
2110 mdesc = mdesc_grab();
2113 dev_err(&dev->dev, "%pOF: Unable to grab MDESC.\n",
2116 goto out_free_global;
2119 err = grab_mdesc_irq_props(mdesc, dev, &mp->mau_info, "ncp");
2121 dev_err(&dev->dev, "%pOF: Unable to grab IRQ props.\n",
2123 mdesc_release(mdesc);
2124 goto out_free_global;
2127 err = spu_mdesc_scan(mdesc, dev, &mp->mau_info, &mp->mau_list,
2128 "mau", HV_NCS_QTYPE_MAU, mau_intr,
2130 mdesc_release(mdesc);
2133 dev_err(&dev->dev, "%pOF: MAU MDESC scan failed.\n",
2135 goto out_free_global;
2138 dev_set_drvdata(&dev->dev, mp);
2143 release_global_resources();
2151 static int n2_mau_remove(struct platform_device *dev)
2153 struct n2_mau *mp = dev_get_drvdata(&dev->dev);
2155 spu_list_destroy(&mp->mau_list);
2157 release_global_resources();
2164 static const struct of_device_id n2_crypto_match[] = {
2167 .compatible = "SUNW,n2-cwq",
2171 .compatible = "SUNW,vf-cwq",
2175 .compatible = "SUNW,kt-cwq",
2180 MODULE_DEVICE_TABLE(of, n2_crypto_match);
2182 static struct platform_driver n2_crypto_driver = {
2185 .of_match_table = n2_crypto_match,
2187 .probe = n2_crypto_probe,
2188 .remove = n2_crypto_remove,
2191 static const struct of_device_id n2_mau_match[] = {
2194 .compatible = "SUNW,n2-mau",
2198 .compatible = "SUNW,vf-mau",
2202 .compatible = "SUNW,kt-mau",
2207 MODULE_DEVICE_TABLE(of, n2_mau_match);
2209 static struct platform_driver n2_mau_driver = {
2212 .of_match_table = n2_mau_match,
2214 .probe = n2_mau_probe,
2215 .remove = n2_mau_remove,
2218 static struct platform_driver * const drivers[] = {
2223 static int __init n2_init(void)
2225 return platform_register_drivers(drivers, ARRAY_SIZE(drivers));
2228 static void __exit n2_exit(void)
2230 platform_unregister_drivers(drivers, ARRAY_SIZE(drivers));
2233 module_init(n2_init);
2234 module_exit(n2_exit);