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 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 #ifdef DEBUG 117 print_hex_dump(KERN_ERR, "rng refreshed buf@: ", 118 DUMP_PREFIX_ADDRESS, 16, 4, bd->buf, RN_BUF_SIZE, 1); 119 #endif 120 } 121 122 static inline int submit_job(struct caam_rng_ctx *ctx, int to_current) 123 { 124 struct buf_data *bd = &ctx->bufs[!(to_current ^ ctx->current_buf)]; 125 struct device *jrdev = ctx->jrdev; 126 u32 *desc = bd->hw_desc; 127 int err; 128 129 dev_dbg(jrdev, "submitting job %d\n", !(to_current ^ ctx->current_buf)); 130 init_completion(&bd->filled); 131 err = caam_jr_enqueue(jrdev, desc, rng_done, ctx); 132 if (err) 133 complete(&bd->filled); /* don't wait on failed job*/ 134 else 135 atomic_inc(&bd->empty); /* note if pending */ 136 137 return err; 138 } 139 140 static int caam_read(struct hwrng *rng, void *data, size_t max, bool wait) 141 { 142 struct caam_rng_ctx *ctx = rng_ctx; 143 struct buf_data *bd = &ctx->bufs[ctx->current_buf]; 144 int next_buf_idx, copied_idx; 145 int err; 146 147 if (atomic_read(&bd->empty)) { 148 /* try to submit job if there wasn't one */ 149 if (atomic_read(&bd->empty) == BUF_EMPTY) { 150 err = submit_job(ctx, 1); 151 /* if can't submit job, can't even wait */ 152 if (err) 153 return 0; 154 } 155 /* no immediate data, so exit if not waiting */ 156 if (!wait) 157 return 0; 158 159 /* waiting for pending job */ 160 if (atomic_read(&bd->empty)) 161 wait_for_completion(&bd->filled); 162 } 163 164 next_buf_idx = ctx->cur_buf_idx + max; 165 dev_dbg(ctx->jrdev, "%s: start reading at buffer %d, idx %d\n", 166 __func__, ctx->current_buf, ctx->cur_buf_idx); 167 168 /* if enough data in current buffer */ 169 if (next_buf_idx < RN_BUF_SIZE) { 170 memcpy(data, bd->buf + ctx->cur_buf_idx, max); 171 ctx->cur_buf_idx = next_buf_idx; 172 return max; 173 } 174 175 /* else, copy what's left... */ 176 copied_idx = RN_BUF_SIZE - ctx->cur_buf_idx; 177 memcpy(data, bd->buf + ctx->cur_buf_idx, copied_idx); 178 ctx->cur_buf_idx = 0; 179 atomic_set(&bd->empty, BUF_EMPTY); 180 181 /* ...refill... */ 182 submit_job(ctx, 1); 183 184 /* and use next buffer */ 185 ctx->current_buf = !ctx->current_buf; 186 dev_dbg(ctx->jrdev, "switched to buffer %d\n", ctx->current_buf); 187 188 /* since there already is some data read, don't wait */ 189 return copied_idx + caam_read(rng, data + copied_idx, 190 max - copied_idx, false); 191 } 192 193 static inline int rng_create_sh_desc(struct caam_rng_ctx *ctx) 194 { 195 struct device *jrdev = ctx->jrdev; 196 u32 *desc = ctx->sh_desc; 197 198 init_sh_desc(desc, HDR_SHARE_SERIAL); 199 200 /* Generate random bytes */ 201 append_operation(desc, OP_ALG_ALGSEL_RNG | OP_TYPE_CLASS1_ALG); 202 203 /* Store bytes */ 204 append_seq_fifo_store(desc, RN_BUF_SIZE, FIFOST_TYPE_RNGSTORE); 205 206 ctx->sh_desc_dma = dma_map_single(jrdev, desc, desc_bytes(desc), 207 DMA_TO_DEVICE); 208 if (dma_mapping_error(jrdev, ctx->sh_desc_dma)) { 209 dev_err(jrdev, "unable to map shared descriptor\n"); 210 return -ENOMEM; 211 } 212 #ifdef DEBUG 213 print_hex_dump(KERN_ERR, "rng shdesc@: ", DUMP_PREFIX_ADDRESS, 16, 4, 214 desc, desc_bytes(desc), 1); 215 #endif 216 return 0; 217 } 218 219 static inline int rng_create_job_desc(struct caam_rng_ctx *ctx, int buf_id) 220 { 221 struct device *jrdev = ctx->jrdev; 222 struct buf_data *bd = &ctx->bufs[buf_id]; 223 u32 *desc = bd->hw_desc; 224 int sh_len = desc_len(ctx->sh_desc); 225 226 init_job_desc_shared(desc, ctx->sh_desc_dma, sh_len, HDR_SHARE_DEFER | 227 HDR_REVERSE); 228 229 bd->addr = dma_map_single(jrdev, bd->buf, RN_BUF_SIZE, DMA_FROM_DEVICE); 230 if (dma_mapping_error(jrdev, bd->addr)) { 231 dev_err(jrdev, "unable to map dst\n"); 232 return -ENOMEM; 233 } 234 235 append_seq_out_ptr_intlen(desc, bd->addr, RN_BUF_SIZE, 0); 236 #ifdef DEBUG 237 print_hex_dump(KERN_ERR, "rng job desc@: ", DUMP_PREFIX_ADDRESS, 16, 4, 238 desc, desc_bytes(desc), 1); 239 #endif 240 return 0; 241 } 242 243 static void caam_cleanup(struct hwrng *rng) 244 { 245 int i; 246 struct buf_data *bd; 247 248 for (i = 0; i < 2; i++) { 249 bd = &rng_ctx->bufs[i]; 250 if (atomic_read(&bd->empty) == BUF_PENDING) 251 wait_for_completion(&bd->filled); 252 } 253 254 rng_unmap_ctx(rng_ctx); 255 } 256 257 static int caam_init_buf(struct caam_rng_ctx *ctx, int buf_id) 258 { 259 struct buf_data *bd = &ctx->bufs[buf_id]; 260 int err; 261 262 err = rng_create_job_desc(ctx, buf_id); 263 if (err) 264 return err; 265 266 atomic_set(&bd->empty, BUF_EMPTY); 267 submit_job(ctx, buf_id == ctx->current_buf); 268 wait_for_completion(&bd->filled); 269 270 return 0; 271 } 272 273 static int caam_init_rng(struct caam_rng_ctx *ctx, struct device *jrdev) 274 { 275 int err; 276 277 ctx->jrdev = jrdev; 278 279 err = rng_create_sh_desc(ctx); 280 if (err) 281 return err; 282 283 ctx->current_buf = 0; 284 ctx->cur_buf_idx = 0; 285 286 err = caam_init_buf(ctx, 0); 287 if (err) 288 return err; 289 290 return caam_init_buf(ctx, 1); 291 } 292 293 static struct hwrng caam_rng = { 294 .name = "rng-caam", 295 .cleanup = caam_cleanup, 296 .read = caam_read, 297 }; 298 299 static void __exit caam_rng_exit(void) 300 { 301 caam_jr_free(rng_ctx->jrdev); 302 hwrng_unregister(&caam_rng); 303 kfree(rng_ctx); 304 } 305 306 static int __init caam_rng_init(void) 307 { 308 struct device *dev; 309 struct device_node *dev_node; 310 struct platform_device *pdev; 311 struct device *ctrldev; 312 struct caam_drv_private *priv; 313 u32 rng_inst; 314 int err; 315 316 dev_node = of_find_compatible_node(NULL, NULL, "fsl,sec-v4.0"); 317 if (!dev_node) { 318 dev_node = of_find_compatible_node(NULL, NULL, "fsl,sec4.0"); 319 if (!dev_node) 320 return -ENODEV; 321 } 322 323 pdev = of_find_device_by_node(dev_node); 324 if (!pdev) { 325 of_node_put(dev_node); 326 return -ENODEV; 327 } 328 329 ctrldev = &pdev->dev; 330 priv = dev_get_drvdata(ctrldev); 331 of_node_put(dev_node); 332 333 /* 334 * If priv is NULL, it's probably because the caam driver wasn't 335 * properly initialized (e.g. RNG4 init failed). Thus, bail out here. 336 */ 337 if (!priv) 338 return -ENODEV; 339 340 /* Check for an instantiated RNG before registration */ 341 if (priv->era < 10) 342 rng_inst = (rd_reg32(&priv->ctrl->perfmon.cha_num_ls) & 343 CHA_ID_LS_RNG_MASK) >> CHA_ID_LS_RNG_SHIFT; 344 else 345 rng_inst = rd_reg32(&priv->ctrl->vreg.rng) & CHA_VER_NUM_MASK; 346 347 if (!rng_inst) 348 return -ENODEV; 349 350 dev = caam_jr_alloc(); 351 if (IS_ERR(dev)) { 352 pr_err("Job Ring Device allocation for transform failed\n"); 353 return PTR_ERR(dev); 354 } 355 rng_ctx = kmalloc(sizeof(*rng_ctx), GFP_DMA | GFP_KERNEL); 356 if (!rng_ctx) { 357 err = -ENOMEM; 358 goto free_caam_alloc; 359 } 360 err = caam_init_rng(rng_ctx, dev); 361 if (err) 362 goto free_rng_ctx; 363 364 dev_info(dev, "registering rng-caam\n"); 365 return hwrng_register(&caam_rng); 366 367 free_rng_ctx: 368 kfree(rng_ctx); 369 free_caam_alloc: 370 caam_jr_free(dev); 371 return err; 372 } 373 374 module_init(caam_rng_init); 375 module_exit(caam_rng_exit); 376 377 MODULE_LICENSE("GPL"); 378 MODULE_DESCRIPTION("FSL CAAM support for hw_random API"); 379 MODULE_AUTHOR("Freescale Semiconductor - NMG"); 380