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); 111 112 /* Buffer refilled, invalidate cache */ 113 dma_sync_single_for_cpu(jrdev, bd->addr, RN_BUF_SIZE, DMA_FROM_DEVICE); 114 115#ifdef DEBUG 116 print_hex_dump(KERN_ERR, "rng refreshed buf@: ", 117 DUMP_PREFIX_ADDRESS, 16, 4, bd->buf, RN_BUF_SIZE, 1); 118#endif 119} 120 121static inline int submit_job(struct caam_rng_ctx *ctx, int to_current) 122{ 123 struct buf_data *bd = &ctx->bufs[!(to_current ^ ctx->current_buf)]; 124 struct device *jrdev = ctx->jrdev; 125 u32 *desc = bd->hw_desc; 126 int err; 127 128 dev_dbg(jrdev, "submitting job %d\n", !(to_current ^ ctx->current_buf)); 129 init_completion(&bd->filled); 130 err = caam_jr_enqueue(jrdev, desc, rng_done, ctx); 131 if (err) 132 complete(&bd->filled); /* don't wait on failed job*/ 133 else 134 atomic_inc(&bd->empty); /* note if pending */ 135 136 return err; 137} 138 139static int caam_read(struct hwrng *rng, void *data, size_t max, bool wait) 140{ 141 struct caam_rng_ctx *ctx = rng_ctx; 142 struct buf_data *bd = &ctx->bufs[ctx->current_buf]; 143 int next_buf_idx, copied_idx; 144 int err; 145 146 if (atomic_read(&bd->empty)) { 147 /* try to submit job if there wasn't one */ 148 if (atomic_read(&bd->empty) == BUF_EMPTY) { 149 err = submit_job(ctx, 1); 150 /* if can't submit job, can't even wait */ 151 if (err) 152 return 0; 153 } 154 /* no immediate data, so exit if not waiting */ 155 if (!wait) 156 return 0; 157 158 /* waiting for pending job */ 159 if (atomic_read(&bd->empty)) 160 wait_for_completion(&bd->filled); 161 } 162 163 next_buf_idx = ctx->cur_buf_idx + max; 164 dev_dbg(ctx->jrdev, "%s: start reading at buffer %d, idx %d\n", 165 __func__, ctx->current_buf, ctx->cur_buf_idx); 166 167 /* if enough data in current buffer */ 168 if (next_buf_idx < RN_BUF_SIZE) { 169 memcpy(data, bd->buf + ctx->cur_buf_idx, max); 170 ctx->cur_buf_idx = next_buf_idx; 171 return max; 172 } 173 174 /* else, copy what's left... */ 175 copied_idx = RN_BUF_SIZE - ctx->cur_buf_idx; 176 memcpy(data, bd->buf + ctx->cur_buf_idx, copied_idx); 177 ctx->cur_buf_idx = 0; 178 atomic_set(&bd->empty, BUF_EMPTY); 179 180 /* ...refill... */ 181 submit_job(ctx, 1); 182 183 /* and use next buffer */ 184 ctx->current_buf = !ctx->current_buf; 185 dev_dbg(ctx->jrdev, "switched to buffer %d\n", ctx->current_buf); 186 187 /* since there already is some data read, don't wait */ 188 return copied_idx + caam_read(rng, data + copied_idx, 189 max - copied_idx, false); 190} 191 192static inline int rng_create_sh_desc(struct caam_rng_ctx *ctx) 193{ 194 struct device *jrdev = ctx->jrdev; 195 u32 *desc = ctx->sh_desc; 196 197 init_sh_desc(desc, HDR_SHARE_SERIAL); 198 199 /* Propagate errors from shared to job descriptor */ 200 append_cmd(desc, SET_OK_NO_PROP_ERRORS | CMD_LOAD); 201 202 /* Generate random bytes */ 203 append_operation(desc, OP_ALG_ALGSEL_RNG | OP_TYPE_CLASS1_ALG); 204 205 /* Store bytes */ 206 append_seq_fifo_store(desc, RN_BUF_SIZE, FIFOST_TYPE_RNGSTORE); 207 208 ctx->sh_desc_dma = dma_map_single(jrdev, desc, desc_bytes(desc), 209 DMA_TO_DEVICE); 210 if (dma_mapping_error(jrdev, ctx->sh_desc_dma)) { 211 dev_err(jrdev, "unable to map shared descriptor\n"); 212 return -ENOMEM; 213 } 214#ifdef DEBUG 215 print_hex_dump(KERN_ERR, "rng shdesc@: ", DUMP_PREFIX_ADDRESS, 16, 4, 216 desc, desc_bytes(desc), 1); 217#endif 218 return 0; 219} 220 221static inline int rng_create_job_desc(struct caam_rng_ctx *ctx, int buf_id) 222{ 223 struct device *jrdev = ctx->jrdev; 224 struct buf_data *bd = &ctx->bufs[buf_id]; 225 u32 *desc = bd->hw_desc; 226 int sh_len = desc_len(ctx->sh_desc); 227 228 init_job_desc_shared(desc, ctx->sh_desc_dma, sh_len, HDR_SHARE_DEFER | 229 HDR_REVERSE); 230 231 bd->addr = dma_map_single(jrdev, bd->buf, RN_BUF_SIZE, DMA_FROM_DEVICE); 232 if (dma_mapping_error(jrdev, bd->addr)) { 233 dev_err(jrdev, "unable to map dst\n"); 234 return -ENOMEM; 235 } 236 237 append_seq_out_ptr_intlen(desc, bd->addr, RN_BUF_SIZE, 0); 238#ifdef DEBUG 239 print_hex_dump(KERN_ERR, "rng job desc@: ", DUMP_PREFIX_ADDRESS, 16, 4, 240 desc, desc_bytes(desc), 1); 241#endif 242 return 0; 243} 244 245static void caam_cleanup(struct hwrng *rng) 246{ 247 int i; 248 struct buf_data *bd; 249 250 for (i = 0; i < 2; i++) { 251 bd = &rng_ctx->bufs[i]; 252 if (atomic_read(&bd->empty) == BUF_PENDING) 253 wait_for_completion(&bd->filled); 254 } 255 256 rng_unmap_ctx(rng_ctx); 257} 258 259static int caam_init_buf(struct caam_rng_ctx *ctx, int buf_id) 260{ 261 struct buf_data *bd = &ctx->bufs[buf_id]; 262 int err; 263 264 err = rng_create_job_desc(ctx, buf_id); 265 if (err) 266 return err; 267 268 atomic_set(&bd->empty, BUF_EMPTY); 269 submit_job(ctx, buf_id == ctx->current_buf); 270 wait_for_completion(&bd->filled); 271 272 return 0; 273} 274 275static int caam_init_rng(struct caam_rng_ctx *ctx, struct device *jrdev) 276{ 277 int err; 278 279 ctx->jrdev = jrdev; 280 281 err = rng_create_sh_desc(ctx); 282 if (err) 283 return err; 284 285 ctx->current_buf = 0; 286 ctx->cur_buf_idx = 0; 287 288 err = caam_init_buf(ctx, 0); 289 if (err) 290 return err; 291 292 err = caam_init_buf(ctx, 1); 293 if (err) 294 return err; 295 296 return 0; 297} 298 299static struct hwrng caam_rng = { 300 .name = "rng-caam", 301 .cleanup = caam_cleanup, 302 .read = caam_read, 303}; 304 305static void __exit caam_rng_exit(void) 306{ 307 caam_jr_free(rng_ctx->jrdev); 308 hwrng_unregister(&caam_rng); 309 kfree(rng_ctx); 310} 311 312static int __init caam_rng_init(void) 313{ 314 struct device *dev; 315 struct device_node *dev_node; 316 struct platform_device *pdev; 317 struct device *ctrldev; 318 struct caam_drv_private *priv; 319 int err; 320 321 dev_node = of_find_compatible_node(NULL, NULL, "fsl,sec-v4.0"); 322 if (!dev_node) { 323 dev_node = of_find_compatible_node(NULL, NULL, "fsl,sec4.0"); 324 if (!dev_node) 325 return -ENODEV; 326 } 327 328 pdev = of_find_device_by_node(dev_node); 329 if (!pdev) { 330 of_node_put(dev_node); 331 return -ENODEV; 332 } 333 334 ctrldev = &pdev->dev; 335 priv = dev_get_drvdata(ctrldev); 336 of_node_put(dev_node); 337 338 /* 339 * If priv is NULL, it's probably because the caam driver wasn't 340 * properly initialized (e.g. RNG4 init failed). Thus, bail out here. 341 */ 342 if (!priv) 343 return -ENODEV; 344 345 /* Check for an instantiated RNG before registration */ 346 if (!(rd_reg32(&priv->ctrl->perfmon.cha_num_ls) & CHA_ID_LS_RNG_MASK)) 347 return -ENODEV; 348 349 dev = caam_jr_alloc(); 350 if (IS_ERR(dev)) { 351 pr_err("Job Ring Device allocation for transform failed\n"); 352 return PTR_ERR(dev); 353 } 354 rng_ctx = kmalloc(sizeof(*rng_ctx), GFP_DMA); 355 if (!rng_ctx) { 356 err = -ENOMEM; 357 goto free_caam_alloc; 358 } 359 err = caam_init_rng(rng_ctx, dev); 360 if (err) 361 goto free_rng_ctx; 362 363 dev_info(dev, "registering rng-caam\n"); 364 return hwrng_register(&caam_rng); 365 366free_rng_ctx: 367 kfree(rng_ctx); 368free_caam_alloc: 369 caam_jr_free(dev); 370 return err; 371} 372 373module_init(caam_rng_init); 374module_exit(caam_rng_exit); 375 376MODULE_LICENSE("GPL"); 377MODULE_DESCRIPTION("FSL CAAM support for hw_random API"); 378MODULE_AUTHOR("Freescale Semiconductor - NMG"); 379