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_MAX * 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 /* 84 * Variable used to avoid double free of resources in case 85 * algorithm registration was unsuccessful 86 */ 87 static bool init_done; 88 89 static inline void rng_unmap_buf(struct device *jrdev, struct buf_data *bd) 90 { 91 if (bd->addr) 92 dma_unmap_single(jrdev, bd->addr, RN_BUF_SIZE, 93 DMA_FROM_DEVICE); 94 } 95 96 static inline void rng_unmap_ctx(struct caam_rng_ctx *ctx) 97 { 98 struct device *jrdev = ctx->jrdev; 99 100 if (ctx->sh_desc_dma) 101 dma_unmap_single(jrdev, ctx->sh_desc_dma, 102 desc_bytes(ctx->sh_desc), DMA_TO_DEVICE); 103 rng_unmap_buf(jrdev, &ctx->bufs[0]); 104 rng_unmap_buf(jrdev, &ctx->bufs[1]); 105 } 106 107 static void rng_done(struct device *jrdev, u32 *desc, u32 err, void *context) 108 { 109 struct buf_data *bd; 110 111 bd = container_of(desc, struct buf_data, hw_desc[0]); 112 113 if (err) 114 caam_jr_strstatus(jrdev, err); 115 116 atomic_set(&bd->empty, BUF_NOT_EMPTY); 117 complete(&bd->filled); 118 119 /* Buffer refilled, invalidate cache */ 120 dma_sync_single_for_cpu(jrdev, bd->addr, RN_BUF_SIZE, DMA_FROM_DEVICE); 121 122 print_hex_dump_debug("rng refreshed buf@: ", DUMP_PREFIX_ADDRESS, 16, 4, 123 bd->buf, RN_BUF_SIZE, 1); 124 } 125 126 static inline int submit_job(struct caam_rng_ctx *ctx, int to_current) 127 { 128 struct buf_data *bd = &ctx->bufs[!(to_current ^ ctx->current_buf)]; 129 struct device *jrdev = ctx->jrdev; 130 u32 *desc = bd->hw_desc; 131 int err; 132 133 dev_dbg(jrdev, "submitting job %d\n", !(to_current ^ ctx->current_buf)); 134 init_completion(&bd->filled); 135 err = caam_jr_enqueue(jrdev, desc, rng_done, ctx); 136 if (err) 137 complete(&bd->filled); /* don't wait on failed job*/ 138 else 139 atomic_inc(&bd->empty); /* note if pending */ 140 141 return err; 142 } 143 144 static int caam_read(struct hwrng *rng, void *data, size_t max, bool wait) 145 { 146 struct caam_rng_ctx *ctx = rng_ctx; 147 struct buf_data *bd = &ctx->bufs[ctx->current_buf]; 148 int next_buf_idx, copied_idx; 149 int err; 150 151 if (atomic_read(&bd->empty)) { 152 /* try to submit job if there wasn't one */ 153 if (atomic_read(&bd->empty) == BUF_EMPTY) { 154 err = submit_job(ctx, 1); 155 /* if can't submit job, can't even wait */ 156 if (err) 157 return 0; 158 } 159 /* no immediate data, so exit if not waiting */ 160 if (!wait) 161 return 0; 162 163 /* waiting for pending job */ 164 if (atomic_read(&bd->empty)) 165 wait_for_completion(&bd->filled); 166 } 167 168 next_buf_idx = ctx->cur_buf_idx + max; 169 dev_dbg(ctx->jrdev, "%s: start reading at buffer %d, idx %d\n", 170 __func__, ctx->current_buf, ctx->cur_buf_idx); 171 172 /* if enough data in current buffer */ 173 if (next_buf_idx < RN_BUF_SIZE) { 174 memcpy(data, bd->buf + ctx->cur_buf_idx, max); 175 ctx->cur_buf_idx = next_buf_idx; 176 return max; 177 } 178 179 /* else, copy what's left... */ 180 copied_idx = RN_BUF_SIZE - ctx->cur_buf_idx; 181 memcpy(data, bd->buf + ctx->cur_buf_idx, copied_idx); 182 ctx->cur_buf_idx = 0; 183 atomic_set(&bd->empty, BUF_EMPTY); 184 185 /* ...refill... */ 186 submit_job(ctx, 1); 187 188 /* and use next buffer */ 189 ctx->current_buf = !ctx->current_buf; 190 dev_dbg(ctx->jrdev, "switched to buffer %d\n", ctx->current_buf); 191 192 /* since there already is some data read, don't wait */ 193 return copied_idx + caam_read(rng, data + copied_idx, 194 max - copied_idx, false); 195 } 196 197 static inline int rng_create_sh_desc(struct caam_rng_ctx *ctx) 198 { 199 struct device *jrdev = ctx->jrdev; 200 u32 *desc = ctx->sh_desc; 201 202 init_sh_desc(desc, HDR_SHARE_SERIAL); 203 204 /* Generate random bytes */ 205 append_operation(desc, OP_ALG_ALGSEL_RNG | OP_TYPE_CLASS1_ALG); 206 207 /* Store bytes */ 208 append_seq_fifo_store(desc, RN_BUF_SIZE, FIFOST_TYPE_RNGSTORE); 209 210 ctx->sh_desc_dma = dma_map_single(jrdev, desc, desc_bytes(desc), 211 DMA_TO_DEVICE); 212 if (dma_mapping_error(jrdev, ctx->sh_desc_dma)) { 213 dev_err(jrdev, "unable to map shared descriptor\n"); 214 return -ENOMEM; 215 } 216 217 print_hex_dump_debug("rng shdesc@: ", DUMP_PREFIX_ADDRESS, 16, 4, 218 desc, desc_bytes(desc), 1); 219 220 return 0; 221 } 222 223 static inline int rng_create_job_desc(struct caam_rng_ctx *ctx, int buf_id) 224 { 225 struct device *jrdev = ctx->jrdev; 226 struct buf_data *bd = &ctx->bufs[buf_id]; 227 u32 *desc = bd->hw_desc; 228 int sh_len = desc_len(ctx->sh_desc); 229 230 init_job_desc_shared(desc, ctx->sh_desc_dma, sh_len, HDR_SHARE_DEFER | 231 HDR_REVERSE); 232 233 bd->addr = dma_map_single(jrdev, bd->buf, RN_BUF_SIZE, DMA_FROM_DEVICE); 234 if (dma_mapping_error(jrdev, bd->addr)) { 235 dev_err(jrdev, "unable to map dst\n"); 236 return -ENOMEM; 237 } 238 239 append_seq_out_ptr_intlen(desc, bd->addr, RN_BUF_SIZE, 0); 240 241 print_hex_dump_debug("rng job desc@: ", DUMP_PREFIX_ADDRESS, 16, 4, 242 desc, desc_bytes(desc), 1); 243 244 return 0; 245 } 246 247 static void caam_cleanup(struct hwrng *rng) 248 { 249 int i; 250 struct buf_data *bd; 251 252 for (i = 0; i < 2; i++) { 253 bd = &rng_ctx->bufs[i]; 254 if (atomic_read(&bd->empty) == BUF_PENDING) 255 wait_for_completion(&bd->filled); 256 } 257 258 rng_unmap_ctx(rng_ctx); 259 } 260 261 static int caam_init_buf(struct caam_rng_ctx *ctx, int buf_id) 262 { 263 struct buf_data *bd = &ctx->bufs[buf_id]; 264 int err; 265 266 err = rng_create_job_desc(ctx, buf_id); 267 if (err) 268 return err; 269 270 atomic_set(&bd->empty, BUF_EMPTY); 271 submit_job(ctx, buf_id == ctx->current_buf); 272 wait_for_completion(&bd->filled); 273 274 return 0; 275 } 276 277 static int caam_init_rng(struct caam_rng_ctx *ctx, struct device *jrdev) 278 { 279 int err; 280 281 ctx->jrdev = jrdev; 282 283 err = rng_create_sh_desc(ctx); 284 if (err) 285 return err; 286 287 ctx->current_buf = 0; 288 ctx->cur_buf_idx = 0; 289 290 err = caam_init_buf(ctx, 0); 291 if (err) 292 return err; 293 294 return caam_init_buf(ctx, 1); 295 } 296 297 static struct hwrng caam_rng = { 298 .name = "rng-caam", 299 .cleanup = caam_cleanup, 300 .read = caam_read, 301 }; 302 303 void caam_rng_exit(void) 304 { 305 if (!init_done) 306 return; 307 308 caam_jr_free(rng_ctx->jrdev); 309 hwrng_unregister(&caam_rng); 310 kfree(rng_ctx); 311 } 312 313 int caam_rng_init(struct device *ctrldev) 314 { 315 struct device *dev; 316 u32 rng_inst; 317 struct caam_drv_private *priv = dev_get_drvdata(ctrldev); 318 int err; 319 init_done = false; 320 321 /* Check for an instantiated RNG before registration */ 322 if (priv->era < 10) 323 rng_inst = (rd_reg32(&priv->ctrl->perfmon.cha_num_ls) & 324 CHA_ID_LS_RNG_MASK) >> CHA_ID_LS_RNG_SHIFT; 325 else 326 rng_inst = rd_reg32(&priv->ctrl->vreg.rng) & CHA_VER_NUM_MASK; 327 328 if (!rng_inst) 329 return 0; 330 331 dev = caam_jr_alloc(); 332 if (IS_ERR(dev)) { 333 pr_err("Job Ring Device allocation for transform failed\n"); 334 return PTR_ERR(dev); 335 } 336 rng_ctx = kmalloc(sizeof(*rng_ctx), GFP_DMA | GFP_KERNEL); 337 if (!rng_ctx) { 338 err = -ENOMEM; 339 goto free_caam_alloc; 340 } 341 err = caam_init_rng(rng_ctx, dev); 342 if (err) 343 goto free_rng_ctx; 344 345 dev_info(dev, "registering rng-caam\n"); 346 347 err = hwrng_register(&caam_rng); 348 if (!err) { 349 init_done = true; 350 return err; 351 } 352 353 free_rng_ctx: 354 kfree(rng_ctx); 355 free_caam_alloc: 356 caam_jr_free(dev); 357 return err; 358 } 359