xref: /openbmc/linux/drivers/crypto/caam/caamrng.c (revision a20eefae)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * caam - Freescale FSL CAAM support for hw_random
4  *
5  * Copyright 2011 Freescale Semiconductor, Inc.
6  * Copyright 2018-2019 NXP
7  *
8  * Based on caamalg.c crypto API driver.
9  *
10  * relationship between job descriptors to shared descriptors:
11  *
12  * ---------------                     --------------
13  * | JobDesc #0  |-------------------->| ShareDesc  |
14  * | *(buffer 0) |      |------------->| (generate) |
15  * ---------------      |              | (move)     |
16  *                      |              | (store)    |
17  * ---------------      |              --------------
18  * | JobDesc #1  |------|
19  * | *(buffer 1) |
20  * ---------------
21  *
22  * A job desc looks like this:
23  *
24  * ---------------------
25  * | Header            |
26  * | ShareDesc Pointer |
27  * | SEQ_OUT_PTR       |
28  * | (output buffer)   |
29  * ---------------------
30  *
31  * The SharedDesc never changes, and each job descriptor points to one of two
32  * buffers for each device, from which the data will be copied into the
33  * requested destination
34  */
35 
36 #include <linux/hw_random.h>
37 #include <linux/completion.h>
38 #include <linux/atomic.h>
39 
40 #include "compat.h"
41 
42 #include "regs.h"
43 #include "intern.h"
44 #include "desc_constr.h"
45 #include "jr.h"
46 #include "error.h"
47 
48 /*
49  * Maximum buffer size: maximum number of random, cache-aligned bytes that
50  * will be generated and moved to seq out ptr (extlen not allowed)
51  */
52 #define RN_BUF_SIZE			(0xffff / L1_CACHE_BYTES * \
53 					 L1_CACHE_BYTES)
54 
55 /* length of descriptors */
56 #define DESC_JOB_O_LEN			(CAAM_CMD_SZ * 2 + CAAM_PTR_SZ * 2)
57 #define DESC_RNG_LEN			(3 * CAAM_CMD_SZ)
58 
59 /* Buffer, its dma address and lock */
60 struct buf_data {
61 	u8 buf[RN_BUF_SIZE] ____cacheline_aligned;
62 	dma_addr_t addr;
63 	struct completion filled;
64 	u32 hw_desc[DESC_JOB_O_LEN];
65 #define BUF_NOT_EMPTY 0
66 #define BUF_EMPTY 1
67 #define BUF_PENDING 2  /* Empty, but with job pending --don't submit another */
68 	atomic_t empty;
69 };
70 
71 /* rng per-device context */
72 struct caam_rng_ctx {
73 	struct device *jrdev;
74 	dma_addr_t sh_desc_dma;
75 	u32 sh_desc[DESC_RNG_LEN];
76 	unsigned int cur_buf_idx;
77 	int current_buf;
78 	struct buf_data bufs[2];
79 };
80 
81 static struct caam_rng_ctx *rng_ctx;
82 
83 static inline void rng_unmap_buf(struct device *jrdev, struct buf_data *bd)
84 {
85 	if (bd->addr)
86 		dma_unmap_single(jrdev, bd->addr, RN_BUF_SIZE,
87 				 DMA_FROM_DEVICE);
88 }
89 
90 static inline void rng_unmap_ctx(struct caam_rng_ctx *ctx)
91 {
92 	struct device *jrdev = ctx->jrdev;
93 
94 	if (ctx->sh_desc_dma)
95 		dma_unmap_single(jrdev, ctx->sh_desc_dma,
96 				 desc_bytes(ctx->sh_desc), DMA_TO_DEVICE);
97 	rng_unmap_buf(jrdev, &ctx->bufs[0]);
98 	rng_unmap_buf(jrdev, &ctx->bufs[1]);
99 }
100 
101 static void rng_done(struct device *jrdev, u32 *desc, u32 err, void *context)
102 {
103 	struct buf_data *bd;
104 
105 	bd = container_of(desc, struct buf_data, hw_desc[0]);
106 
107 	if (err)
108 		caam_jr_strstatus(jrdev, err);
109 
110 	atomic_set(&bd->empty, BUF_NOT_EMPTY);
111 	complete(&bd->filled);
112 
113 	/* Buffer refilled, invalidate cache */
114 	dma_sync_single_for_cpu(jrdev, bd->addr, RN_BUF_SIZE, DMA_FROM_DEVICE);
115 
116 	print_hex_dump_debug("rng refreshed buf@: ", DUMP_PREFIX_ADDRESS, 16, 4,
117 			     bd->buf, RN_BUF_SIZE, 1);
118 }
119 
120 static inline int submit_job(struct caam_rng_ctx *ctx, int to_current)
121 {
122 	struct buf_data *bd = &ctx->bufs[!(to_current ^ ctx->current_buf)];
123 	struct device *jrdev = ctx->jrdev;
124 	u32 *desc = bd->hw_desc;
125 	int err;
126 
127 	dev_dbg(jrdev, "submitting job %d\n", !(to_current ^ ctx->current_buf));
128 	init_completion(&bd->filled);
129 	err = caam_jr_enqueue(jrdev, desc, rng_done, ctx);
130 	if (err)
131 		complete(&bd->filled); /* don't wait on failed job*/
132 	else
133 		atomic_inc(&bd->empty); /* note if pending */
134 
135 	return err;
136 }
137 
138 static int caam_read(struct hwrng *rng, void *data, size_t max, bool wait)
139 {
140 	struct caam_rng_ctx *ctx = rng_ctx;
141 	struct buf_data *bd = &ctx->bufs[ctx->current_buf];
142 	int next_buf_idx, copied_idx;
143 	int err;
144 
145 	if (atomic_read(&bd->empty)) {
146 		/* try to submit job if there wasn't one */
147 		if (atomic_read(&bd->empty) == BUF_EMPTY) {
148 			err = submit_job(ctx, 1);
149 			/* if can't submit job, can't even wait */
150 			if (err)
151 				return 0;
152 		}
153 		/* no immediate data, so exit if not waiting */
154 		if (!wait)
155 			return 0;
156 
157 		/* waiting for pending job */
158 		if (atomic_read(&bd->empty))
159 			wait_for_completion(&bd->filled);
160 	}
161 
162 	next_buf_idx = ctx->cur_buf_idx + max;
163 	dev_dbg(ctx->jrdev, "%s: start reading at buffer %d, idx %d\n",
164 		 __func__, ctx->current_buf, ctx->cur_buf_idx);
165 
166 	/* if enough data in current buffer */
167 	if (next_buf_idx < RN_BUF_SIZE) {
168 		memcpy(data, bd->buf + ctx->cur_buf_idx, max);
169 		ctx->cur_buf_idx = next_buf_idx;
170 		return max;
171 	}
172 
173 	/* else, copy what's left... */
174 	copied_idx = RN_BUF_SIZE - ctx->cur_buf_idx;
175 	memcpy(data, bd->buf + ctx->cur_buf_idx, copied_idx);
176 	ctx->cur_buf_idx = 0;
177 	atomic_set(&bd->empty, BUF_EMPTY);
178 
179 	/* ...refill... */
180 	submit_job(ctx, 1);
181 
182 	/* and use next buffer */
183 	ctx->current_buf = !ctx->current_buf;
184 	dev_dbg(ctx->jrdev, "switched to buffer %d\n", ctx->current_buf);
185 
186 	/* since there already is some data read, don't wait */
187 	return copied_idx + caam_read(rng, data + copied_idx,
188 				      max - copied_idx, false);
189 }
190 
191 static inline int rng_create_sh_desc(struct caam_rng_ctx *ctx)
192 {
193 	struct device *jrdev = ctx->jrdev;
194 	u32 *desc = ctx->sh_desc;
195 
196 	init_sh_desc(desc, HDR_SHARE_SERIAL);
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 
211 	print_hex_dump_debug("rng shdesc@: ", DUMP_PREFIX_ADDRESS, 16, 4,
212 			     desc, desc_bytes(desc), 1);
213 
214 	return 0;
215 }
216 
217 static 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 
235 	print_hex_dump_debug("rng job desc@: ", DUMP_PREFIX_ADDRESS, 16, 4,
236 			     desc, desc_bytes(desc), 1);
237 
238 	return 0;
239 }
240 
241 static 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 
255 static 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 
271 static 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 	return caam_init_buf(ctx, 1);
289 }
290 
291 static struct hwrng caam_rng = {
292 	.name		= "rng-caam",
293 	.cleanup	= caam_cleanup,
294 	.read		= caam_read,
295 };
296 
297 void caam_rng_exit(void)
298 {
299 	caam_jr_free(rng_ctx->jrdev);
300 	hwrng_unregister(&caam_rng);
301 	kfree(rng_ctx);
302 }
303 
304 int caam_rng_init(struct device *ctrldev)
305 {
306 	struct device *dev;
307 	u32 rng_inst;
308 	struct caam_drv_private *priv = dev_get_drvdata(ctrldev);
309 	int err;
310 
311 	/* Check for an instantiated RNG before registration */
312 	if (priv->era < 10)
313 		rng_inst = (rd_reg32(&priv->ctrl->perfmon.cha_num_ls) &
314 			    CHA_ID_LS_RNG_MASK) >> CHA_ID_LS_RNG_SHIFT;
315 	else
316 		rng_inst = rd_reg32(&priv->ctrl->vreg.rng) & CHA_VER_NUM_MASK;
317 
318 	if (!rng_inst)
319 		return 0;
320 
321 	dev = caam_jr_alloc();
322 	if (IS_ERR(dev)) {
323 		pr_err("Job Ring Device allocation for transform failed\n");
324 		return PTR_ERR(dev);
325 	}
326 	rng_ctx = kmalloc(sizeof(*rng_ctx), GFP_DMA | GFP_KERNEL);
327 	if (!rng_ctx) {
328 		err = -ENOMEM;
329 		goto free_caam_alloc;
330 	}
331 	err = caam_init_rng(rng_ctx, dev);
332 	if (err)
333 		goto free_rng_ctx;
334 
335 	dev_info(dev, "registering rng-caam\n");
336 	return hwrng_register(&caam_rng);
337 
338 free_rng_ctx:
339 	kfree(rng_ctx);
340 free_caam_alloc:
341 	caam_jr_free(dev);
342 	return err;
343 }
344