1/*
2 * caam - Freescale FSL CAAM support for hw_random
3 *
4 * Copyright 2011 Freescale Semiconductor, Inc.
5 *
6 * Based on caamalg.c crypto API driver.
7 *
8 * relationship between job descriptors to shared descriptors:
9 *
10 * ---------------                     --------------
11 * | JobDesc #0  |-------------------->| ShareDesc  |
12 * | *(buffer 0) |      |------------->| (generate) |
13 * ---------------      |              | (move)     |
14 *                      |              | (store)    |
15 * ---------------      |              --------------
16 * | JobDesc #1  |------|
17 * | *(buffer 1) |
18 * ---------------
19 *
20 * A job desc looks like this:
21 *
22 * ---------------------
23 * | Header            |
24 * | ShareDesc Pointer |
25 * | SEQ_OUT_PTR       |
26 * | (output buffer)   |
27 * ---------------------
28 *
29 * The SharedDesc never changes, and each job descriptor points to one of two
30 * buffers for each device, from which the data will be copied into the
31 * requested destination
32 */
33
34#include <linux/hw_random.h>
35#include <linux/completion.h>
36#include <linux/atomic.h>
37
38#include "compat.h"
39
40#include "regs.h"
41#include "intern.h"
42#include "desc_constr.h"
43#include "jr.h"
44#include "error.h"
45
46/*
47 * Maximum buffer size: maximum number of random, cache-aligned bytes that
48 * will be generated and moved to seq out ptr (extlen not allowed)
49 */
50#define RN_BUF_SIZE			(0xffff / L1_CACHE_BYTES * \
51					 L1_CACHE_BYTES)
52
53/* length of descriptors */
54#define DESC_JOB_O_LEN			(CAAM_CMD_SZ * 2 + CAAM_PTR_SZ * 2)
55#define DESC_RNG_LEN			(4 * CAAM_CMD_SZ)
56
57/* Buffer, its dma address and lock */
58struct buf_data {
59	u8 buf[RN_BUF_SIZE] ____cacheline_aligned;
60	dma_addr_t addr;
61	struct completion filled;
62	u32 hw_desc[DESC_JOB_O_LEN];
63#define BUF_NOT_EMPTY 0
64#define BUF_EMPTY 1
65#define BUF_PENDING 2  /* Empty, but with job pending --don't submit another */
66	atomic_t empty;
67};
68
69/* rng per-device context */
70struct caam_rng_ctx {
71	struct device *jrdev;
72	dma_addr_t sh_desc_dma;
73	u32 sh_desc[DESC_RNG_LEN];
74	unsigned int cur_buf_idx;
75	int current_buf;
76	struct buf_data bufs[2];
77};
78
79static struct caam_rng_ctx *rng_ctx;
80
81static inline void rng_unmap_buf(struct device *jrdev, struct buf_data *bd)
82{
83	if (bd->addr)
84		dma_unmap_single(jrdev, bd->addr, RN_BUF_SIZE,
85				 DMA_FROM_DEVICE);
86}
87
88static inline void rng_unmap_ctx(struct caam_rng_ctx *ctx)
89{
90	struct device *jrdev = ctx->jrdev;
91
92	if (ctx->sh_desc_dma)
93		dma_unmap_single(jrdev, ctx->sh_desc_dma,
94				 desc_bytes(ctx->sh_desc), DMA_TO_DEVICE);
95	rng_unmap_buf(jrdev, &ctx->bufs[0]);
96	rng_unmap_buf(jrdev, &ctx->bufs[1]);
97}
98
99static void rng_done(struct device *jrdev, u32 *desc, u32 err, void *context)
100{
101	struct buf_data *bd;
102
103	bd = (struct buf_data *)((char *)desc -
104	      offsetof(struct buf_data, hw_desc));
105
106	if (err)
107		caam_jr_strstatus(jrdev, err);
108
109	atomic_set(&bd->empty, BUF_NOT_EMPTY);
110	complete(&bd->filled);
111#ifdef DEBUG
112	print_hex_dump(KERN_ERR, "rng refreshed buf@: ",
113		       DUMP_PREFIX_ADDRESS, 16, 4, bd->buf, RN_BUF_SIZE, 1);
114#endif
115}
116
117static inline int submit_job(struct caam_rng_ctx *ctx, int to_current)
118{
119	struct buf_data *bd = &ctx->bufs[!(to_current ^ ctx->current_buf)];
120	struct device *jrdev = ctx->jrdev;
121	u32 *desc = bd->hw_desc;
122	int err;
123
124	dev_dbg(jrdev, "submitting job %d\n", !(to_current ^ ctx->current_buf));
125	init_completion(&bd->filled);
126	err = caam_jr_enqueue(jrdev, desc, rng_done, ctx);
127	if (err)
128		complete(&bd->filled); /* don't wait on failed job*/
129	else
130		atomic_inc(&bd->empty); /* note if pending */
131
132	return err;
133}
134
135static int caam_read(struct hwrng *rng, void *data, size_t max, bool wait)
136{
137	struct caam_rng_ctx *ctx = rng_ctx;
138	struct buf_data *bd = &ctx->bufs[ctx->current_buf];
139	int next_buf_idx, copied_idx;
140	int err;
141
142	if (atomic_read(&bd->empty)) {
143		/* try to submit job if there wasn't one */
144		if (atomic_read(&bd->empty) == BUF_EMPTY) {
145			err = submit_job(ctx, 1);
146			/* if can't submit job, can't even wait */
147			if (err)
148				return 0;
149		}
150		/* no immediate data, so exit if not waiting */
151		if (!wait)
152			return 0;
153
154		/* waiting for pending job */
155		if (atomic_read(&bd->empty))
156			wait_for_completion(&bd->filled);
157	}
158
159	next_buf_idx = ctx->cur_buf_idx + max;
160	dev_dbg(ctx->jrdev, "%s: start reading at buffer %d, idx %d\n",
161		 __func__, ctx->current_buf, ctx->cur_buf_idx);
162
163	/* if enough data in current buffer */
164	if (next_buf_idx < RN_BUF_SIZE) {
165		memcpy(data, bd->buf + ctx->cur_buf_idx, max);
166		ctx->cur_buf_idx = next_buf_idx;
167		return max;
168	}
169
170	/* else, copy what's left... */
171	copied_idx = RN_BUF_SIZE - ctx->cur_buf_idx;
172	memcpy(data, bd->buf + ctx->cur_buf_idx, copied_idx);
173	ctx->cur_buf_idx = 0;
174	atomic_set(&bd->empty, BUF_EMPTY);
175
176	/* ...refill... */
177	submit_job(ctx, 1);
178
179	/* and use next buffer */
180	ctx->current_buf = !ctx->current_buf;
181	dev_dbg(ctx->jrdev, "switched to buffer %d\n", ctx->current_buf);
182
183	/* since there already is some data read, don't wait */
184	return copied_idx + caam_read(rng, data + copied_idx,
185				      max - copied_idx, false);
186}
187
188static inline int rng_create_sh_desc(struct caam_rng_ctx *ctx)
189{
190	struct device *jrdev = ctx->jrdev;
191	u32 *desc = ctx->sh_desc;
192
193	init_sh_desc(desc, HDR_SHARE_SERIAL);
194
195	/* Propagate errors from shared to job descriptor */
196	append_cmd(desc, SET_OK_NO_PROP_ERRORS | CMD_LOAD);
197
198	/* Generate random bytes */
199	append_operation(desc, OP_ALG_ALGSEL_RNG | OP_TYPE_CLASS1_ALG);
200
201	/* Store bytes */
202	append_seq_fifo_store(desc, RN_BUF_SIZE, FIFOST_TYPE_RNGSTORE);
203
204	ctx->sh_desc_dma = dma_map_single(jrdev, desc, desc_bytes(desc),
205					  DMA_TO_DEVICE);
206	if (dma_mapping_error(jrdev, ctx->sh_desc_dma)) {
207		dev_err(jrdev, "unable to map shared descriptor\n");
208		return -ENOMEM;
209	}
210#ifdef DEBUG
211	print_hex_dump(KERN_ERR, "rng shdesc@: ", DUMP_PREFIX_ADDRESS, 16, 4,
212		       desc, desc_bytes(desc), 1);
213#endif
214	return 0;
215}
216
217static inline int rng_create_job_desc(struct caam_rng_ctx *ctx, int buf_id)
218{
219	struct device *jrdev = ctx->jrdev;
220	struct buf_data *bd = &ctx->bufs[buf_id];
221	u32 *desc = bd->hw_desc;
222	int sh_len = desc_len(ctx->sh_desc);
223
224	init_job_desc_shared(desc, ctx->sh_desc_dma, sh_len, HDR_SHARE_DEFER |
225			     HDR_REVERSE);
226
227	bd->addr = dma_map_single(jrdev, bd->buf, RN_BUF_SIZE, DMA_FROM_DEVICE);
228	if (dma_mapping_error(jrdev, bd->addr)) {
229		dev_err(jrdev, "unable to map dst\n");
230		return -ENOMEM;
231	}
232
233	append_seq_out_ptr_intlen(desc, bd->addr, RN_BUF_SIZE, 0);
234#ifdef DEBUG
235	print_hex_dump(KERN_ERR, "rng job desc@: ", DUMP_PREFIX_ADDRESS, 16, 4,
236		       desc, desc_bytes(desc), 1);
237#endif
238	return 0;
239}
240
241static void caam_cleanup(struct hwrng *rng)
242{
243	int i;
244	struct buf_data *bd;
245
246	for (i = 0; i < 2; i++) {
247		bd = &rng_ctx->bufs[i];
248		if (atomic_read(&bd->empty) == BUF_PENDING)
249			wait_for_completion(&bd->filled);
250	}
251
252	rng_unmap_ctx(rng_ctx);
253}
254
255static int caam_init_buf(struct caam_rng_ctx *ctx, int buf_id)
256{
257	struct buf_data *bd = &ctx->bufs[buf_id];
258	int err;
259
260	err = rng_create_job_desc(ctx, buf_id);
261	if (err)
262		return err;
263
264	atomic_set(&bd->empty, BUF_EMPTY);
265	submit_job(ctx, buf_id == ctx->current_buf);
266	wait_for_completion(&bd->filled);
267
268	return 0;
269}
270
271static int caam_init_rng(struct caam_rng_ctx *ctx, struct device *jrdev)
272{
273	int err;
274
275	ctx->jrdev = jrdev;
276
277	err = rng_create_sh_desc(ctx);
278	if (err)
279		return err;
280
281	ctx->current_buf = 0;
282	ctx->cur_buf_idx = 0;
283
284	err = caam_init_buf(ctx, 0);
285	if (err)
286		return err;
287
288	err = caam_init_buf(ctx, 1);
289	if (err)
290		return err;
291
292	return 0;
293}
294
295static struct hwrng caam_rng = {
296	.name		= "rng-caam",
297	.cleanup	= caam_cleanup,
298	.read		= caam_read,
299};
300
301static void __exit caam_rng_exit(void)
302{
303	caam_jr_free(rng_ctx->jrdev);
304	hwrng_unregister(&caam_rng);
305	kfree(rng_ctx);
306}
307
308static int __init caam_rng_init(void)
309{
310	struct device *dev;
311	struct device_node *dev_node;
312	struct platform_device *pdev;
313	struct device *ctrldev;
314	void *priv;
315	int err;
316
317	dev_node = of_find_compatible_node(NULL, NULL, "fsl,sec-v4.0");
318	if (!dev_node) {
319		dev_node = of_find_compatible_node(NULL, NULL, "fsl,sec4.0");
320		if (!dev_node)
321			return -ENODEV;
322	}
323
324	pdev = of_find_device_by_node(dev_node);
325	if (!pdev) {
326		of_node_put(dev_node);
327		return -ENODEV;
328	}
329
330	ctrldev = &pdev->dev;
331	priv = dev_get_drvdata(ctrldev);
332	of_node_put(dev_node);
333
334	/*
335	 * If priv is NULL, it's probably because the caam driver wasn't
336	 * properly initialized (e.g. RNG4 init failed). Thus, bail out here.
337	 */
338	if (!priv)
339		return -ENODEV;
340
341	dev = caam_jr_alloc();
342	if (IS_ERR(dev)) {
343		pr_err("Job Ring Device allocation for transform failed\n");
344		return PTR_ERR(dev);
345	}
346	rng_ctx = kmalloc(sizeof(struct caam_rng_ctx), GFP_DMA);
347	if (!rng_ctx)
348		return -ENOMEM;
349	err = caam_init_rng(rng_ctx, dev);
350	if (err)
351		return err;
352
353	dev_info(dev, "registering rng-caam\n");
354	return hwrng_register(&caam_rng);
355}
356
357module_init(caam_rng_init);
358module_exit(caam_rng_exit);
359
360MODULE_LICENSE("GPL");
361MODULE_DESCRIPTION("FSL CAAM support for hw_random API");
362MODULE_AUTHOR("Freescale Semiconductor - NMG");
363