1 /* 2 * Copyright (C) 2015-2016 Marvell International Ltd. 3 4 * This program is free software: you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License as 6 * published by the Free Software Foundation, either version 2 of the 7 * License, or any later version. 8 * 9 * This program is distributed in the hope that it will be useful, but 10 * WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 12 * General Public License for more details. 13 */ 14 15 #include <linux/clk.h> 16 #include <linux/dma-mapping.h> 17 #include <linux/interrupt.h> 18 #include <linux/io.h> 19 #include <linux/module.h> 20 #include <linux/msi.h> 21 #include <linux/of.h> 22 #include <linux/of_irq.h> 23 #include <linux/platform_device.h> 24 #include <linux/spinlock.h> 25 26 #include "dmaengine.h" 27 28 /* DMA Engine Registers */ 29 #define MV_XOR_V2_DMA_DESQ_BALR_OFF 0x000 30 #define MV_XOR_V2_DMA_DESQ_BAHR_OFF 0x004 31 #define MV_XOR_V2_DMA_DESQ_SIZE_OFF 0x008 32 #define MV_XOR_V2_DMA_DESQ_DONE_OFF 0x00C 33 #define MV_XOR_V2_DMA_DESQ_DONE_PENDING_MASK 0x7FFF 34 #define MV_XOR_V2_DMA_DESQ_DONE_PENDING_SHIFT 0 35 #define MV_XOR_V2_DMA_DESQ_DONE_READ_PTR_MASK 0x1FFF 36 #define MV_XOR_V2_DMA_DESQ_DONE_READ_PTR_SHIFT 16 37 #define MV_XOR_V2_DMA_DESQ_ARATTR_OFF 0x010 38 #define MV_XOR_V2_DMA_DESQ_ATTR_CACHE_MASK 0x3F3F 39 #define MV_XOR_V2_DMA_DESQ_ATTR_OUTER_SHAREABLE 0x202 40 #define MV_XOR_V2_DMA_DESQ_ATTR_CACHEABLE 0x3C3C 41 #define MV_XOR_V2_DMA_IMSG_CDAT_OFF 0x014 42 #define MV_XOR_V2_DMA_IMSG_THRD_OFF 0x018 43 #define MV_XOR_V2_DMA_IMSG_THRD_MASK 0x7FFF 44 #define MV_XOR_V2_DMA_IMSG_THRD_SHIFT 0x0 45 #define MV_XOR_V2_DMA_IMSG_TIMER_EN BIT(18) 46 #define MV_XOR_V2_DMA_DESQ_AWATTR_OFF 0x01C 47 /* Same flags as MV_XOR_V2_DMA_DESQ_ARATTR_OFF */ 48 #define MV_XOR_V2_DMA_DESQ_ALLOC_OFF 0x04C 49 #define MV_XOR_V2_DMA_DESQ_ALLOC_WRPTR_MASK 0xFFFF 50 #define MV_XOR_V2_DMA_DESQ_ALLOC_WRPTR_SHIFT 16 51 #define MV_XOR_V2_DMA_IMSG_BALR_OFF 0x050 52 #define MV_XOR_V2_DMA_IMSG_BAHR_OFF 0x054 53 #define MV_XOR_V2_DMA_DESQ_CTRL_OFF 0x100 54 #define MV_XOR_V2_DMA_DESQ_CTRL_32B 1 55 #define MV_XOR_V2_DMA_DESQ_CTRL_128B 7 56 #define MV_XOR_V2_DMA_DESQ_STOP_OFF 0x800 57 #define MV_XOR_V2_DMA_DESQ_DEALLOC_OFF 0x804 58 #define MV_XOR_V2_DMA_DESQ_ADD_OFF 0x808 59 #define MV_XOR_V2_DMA_IMSG_TMOT 0x810 60 #define MV_XOR_V2_DMA_IMSG_TIMER_THRD_MASK 0x1FFF 61 #define MV_XOR_V2_DMA_IMSG_TIMER_THRD_SHIFT 0 62 63 /* XOR Global registers */ 64 #define MV_XOR_V2_GLOB_BW_CTRL 0x4 65 #define MV_XOR_V2_GLOB_BW_CTRL_NUM_OSTD_RD_SHIFT 0 66 #define MV_XOR_V2_GLOB_BW_CTRL_NUM_OSTD_RD_VAL 64 67 #define MV_XOR_V2_GLOB_BW_CTRL_NUM_OSTD_WR_SHIFT 8 68 #define MV_XOR_V2_GLOB_BW_CTRL_NUM_OSTD_WR_VAL 8 69 #define MV_XOR_V2_GLOB_BW_CTRL_RD_BURST_LEN_SHIFT 12 70 #define MV_XOR_V2_GLOB_BW_CTRL_RD_BURST_LEN_VAL 4 71 #define MV_XOR_V2_GLOB_BW_CTRL_WR_BURST_LEN_SHIFT 16 72 #define MV_XOR_V2_GLOB_BW_CTRL_WR_BURST_LEN_VAL 4 73 #define MV_XOR_V2_GLOB_PAUSE 0x014 74 #define MV_XOR_V2_GLOB_PAUSE_AXI_TIME_DIS_VAL 0x8 75 #define MV_XOR_V2_GLOB_SYS_INT_CAUSE 0x200 76 #define MV_XOR_V2_GLOB_SYS_INT_MASK 0x204 77 #define MV_XOR_V2_GLOB_MEM_INT_CAUSE 0x220 78 #define MV_XOR_V2_GLOB_MEM_INT_MASK 0x224 79 80 #define MV_XOR_V2_MIN_DESC_SIZE 32 81 #define MV_XOR_V2_EXT_DESC_SIZE 128 82 83 #define MV_XOR_V2_DESC_RESERVED_SIZE 12 84 #define MV_XOR_V2_DESC_BUFF_D_ADDR_SIZE 12 85 86 #define MV_XOR_V2_CMD_LINE_NUM_MAX_D_BUF 8 87 88 /* 89 * Descriptors queue size. With 32 bytes descriptors, up to 2^14 90 * descriptors are allowed, with 128 bytes descriptors, up to 2^12 91 * descriptors are allowed. This driver uses 128 bytes descriptors, 92 * but experimentation has shown that a set of 1024 descriptors is 93 * sufficient to reach a good level of performance. 94 */ 95 #define MV_XOR_V2_DESC_NUM 1024 96 97 /* 98 * Threshold values for descriptors and timeout, determined by 99 * experimentation as giving a good level of performance. 100 */ 101 #define MV_XOR_V2_DONE_IMSG_THRD 0x14 102 #define MV_XOR_V2_TIMER_THRD 0xB0 103 104 /** 105 * struct mv_xor_v2_descriptor - DMA HW descriptor 106 * @desc_id: used by S/W and is not affected by H/W. 107 * @flags: error and status flags 108 * @crc32_result: CRC32 calculation result 109 * @desc_ctrl: operation mode and control flags 110 * @buff_size: amount of bytes to be processed 111 * @fill_pattern_src_addr: Fill-Pattern or Source-Address and 112 * AW-Attributes 113 * @data_buff_addr: Source (and might be RAID6 destination) 114 * addresses of data buffers in RAID5 and RAID6 115 * @reserved: reserved 116 */ 117 struct mv_xor_v2_descriptor { 118 u16 desc_id; 119 u16 flags; 120 u32 crc32_result; 121 u32 desc_ctrl; 122 123 /* Definitions for desc_ctrl */ 124 #define DESC_NUM_ACTIVE_D_BUF_SHIFT 22 125 #define DESC_OP_MODE_SHIFT 28 126 #define DESC_OP_MODE_NOP 0 /* Idle operation */ 127 #define DESC_OP_MODE_MEMCPY 1 /* Pure-DMA operation */ 128 #define DESC_OP_MODE_MEMSET 2 /* Mem-Fill operation */ 129 #define DESC_OP_MODE_MEMINIT 3 /* Mem-Init operation */ 130 #define DESC_OP_MODE_MEM_COMPARE 4 /* Mem-Compare operation */ 131 #define DESC_OP_MODE_CRC32 5 /* CRC32 calculation */ 132 #define DESC_OP_MODE_XOR 6 /* RAID5 (XOR) operation */ 133 #define DESC_OP_MODE_RAID6 7 /* RAID6 P&Q-generation */ 134 #define DESC_OP_MODE_RAID6_REC 8 /* RAID6 Recovery */ 135 #define DESC_Q_BUFFER_ENABLE BIT(16) 136 #define DESC_P_BUFFER_ENABLE BIT(17) 137 #define DESC_IOD BIT(27) 138 139 u32 buff_size; 140 u32 fill_pattern_src_addr[4]; 141 u32 data_buff_addr[MV_XOR_V2_DESC_BUFF_D_ADDR_SIZE]; 142 u32 reserved[MV_XOR_V2_DESC_RESERVED_SIZE]; 143 }; 144 145 /** 146 * struct mv_xor_v2_device - implements a xor device 147 * @lock: lock for the engine 148 * @dma_base: memory mapped DMA register base 149 * @glob_base: memory mapped global register base 150 * @irq_tasklet: 151 * @free_sw_desc: linked list of free SW descriptors 152 * @dmadev: dma device 153 * @dmachan: dma channel 154 * @hw_desq: HW descriptors queue 155 * @hw_desq_virt: virtual address of DESCQ 156 * @sw_desq: SW descriptors queue 157 * @desc_size: HW descriptor size 158 * @npendings: number of pending descriptors (for which tx_submit has 159 * been called, but not yet issue_pending) 160 */ 161 struct mv_xor_v2_device { 162 spinlock_t lock; 163 void __iomem *dma_base; 164 void __iomem *glob_base; 165 struct clk *clk; 166 struct clk *reg_clk; 167 struct tasklet_struct irq_tasklet; 168 struct list_head free_sw_desc; 169 struct dma_device dmadev; 170 struct dma_chan dmachan; 171 dma_addr_t hw_desq; 172 struct mv_xor_v2_descriptor *hw_desq_virt; 173 struct mv_xor_v2_sw_desc *sw_desq; 174 int desc_size; 175 unsigned int npendings; 176 unsigned int hw_queue_idx; 177 }; 178 179 /** 180 * struct mv_xor_v2_sw_desc - implements a xor SW descriptor 181 * @idx: descriptor index 182 * @async_tx: support for the async_tx api 183 * @hw_desc: assosiated HW descriptor 184 * @free_list: node of the free SW descriprots list 185 */ 186 struct mv_xor_v2_sw_desc { 187 int idx; 188 struct dma_async_tx_descriptor async_tx; 189 struct mv_xor_v2_descriptor hw_desc; 190 struct list_head free_list; 191 }; 192 193 /* 194 * Fill the data buffers to a HW descriptor 195 */ 196 static void mv_xor_v2_set_data_buffers(struct mv_xor_v2_device *xor_dev, 197 struct mv_xor_v2_descriptor *desc, 198 dma_addr_t src, int index) 199 { 200 int arr_index = ((index >> 1) * 3); 201 202 /* 203 * Fill the buffer's addresses to the descriptor. 204 * 205 * The format of the buffers address for 2 sequential buffers 206 * X and X + 1: 207 * 208 * First word: Buffer-DX-Address-Low[31:0] 209 * Second word: Buffer-DX+1-Address-Low[31:0] 210 * Third word: DX+1-Buffer-Address-High[47:32] [31:16] 211 * DX-Buffer-Address-High[47:32] [15:0] 212 */ 213 if ((index & 0x1) == 0) { 214 desc->data_buff_addr[arr_index] = lower_32_bits(src); 215 216 desc->data_buff_addr[arr_index + 2] &= ~0xFFFF; 217 desc->data_buff_addr[arr_index + 2] |= 218 upper_32_bits(src) & 0xFFFF; 219 } else { 220 desc->data_buff_addr[arr_index + 1] = 221 lower_32_bits(src); 222 223 desc->data_buff_addr[arr_index + 2] &= ~0xFFFF0000; 224 desc->data_buff_addr[arr_index + 2] |= 225 (upper_32_bits(src) & 0xFFFF) << 16; 226 } 227 } 228 229 /* 230 * notify the engine of new descriptors, and update the available index. 231 */ 232 static void mv_xor_v2_add_desc_to_desq(struct mv_xor_v2_device *xor_dev, 233 int num_of_desc) 234 { 235 /* write the number of new descriptors in the DESQ. */ 236 writel(num_of_desc, xor_dev->dma_base + MV_XOR_V2_DMA_DESQ_ADD_OFF); 237 } 238 239 /* 240 * free HW descriptors 241 */ 242 static void mv_xor_v2_free_desc_from_desq(struct mv_xor_v2_device *xor_dev, 243 int num_of_desc) 244 { 245 /* write the number of new descriptors in the DESQ. */ 246 writel(num_of_desc, xor_dev->dma_base + MV_XOR_V2_DMA_DESQ_DEALLOC_OFF); 247 } 248 249 /* 250 * Set descriptor size 251 * Return the HW descriptor size in bytes 252 */ 253 static int mv_xor_v2_set_desc_size(struct mv_xor_v2_device *xor_dev) 254 { 255 writel(MV_XOR_V2_DMA_DESQ_CTRL_128B, 256 xor_dev->dma_base + MV_XOR_V2_DMA_DESQ_CTRL_OFF); 257 258 return MV_XOR_V2_EXT_DESC_SIZE; 259 } 260 261 /* 262 * Set the IMSG threshold 263 */ 264 static inline 265 void mv_xor_v2_enable_imsg_thrd(struct mv_xor_v2_device *xor_dev) 266 { 267 u32 reg; 268 269 /* Configure threshold of number of descriptors, and enable timer */ 270 reg = readl(xor_dev->dma_base + MV_XOR_V2_DMA_IMSG_THRD_OFF); 271 reg &= (~MV_XOR_V2_DMA_IMSG_THRD_MASK << MV_XOR_V2_DMA_IMSG_THRD_SHIFT); 272 reg |= (MV_XOR_V2_DONE_IMSG_THRD << MV_XOR_V2_DMA_IMSG_THRD_SHIFT); 273 reg |= MV_XOR_V2_DMA_IMSG_TIMER_EN; 274 writel(reg, xor_dev->dma_base + MV_XOR_V2_DMA_IMSG_THRD_OFF); 275 276 /* Configure Timer Threshold */ 277 reg = readl(xor_dev->dma_base + MV_XOR_V2_DMA_IMSG_TMOT); 278 reg &= (~MV_XOR_V2_DMA_IMSG_TIMER_THRD_MASK << 279 MV_XOR_V2_DMA_IMSG_TIMER_THRD_SHIFT); 280 reg |= (MV_XOR_V2_TIMER_THRD << MV_XOR_V2_DMA_IMSG_TIMER_THRD_SHIFT); 281 writel(reg, xor_dev->dma_base + MV_XOR_V2_DMA_IMSG_TMOT); 282 } 283 284 static irqreturn_t mv_xor_v2_interrupt_handler(int irq, void *data) 285 { 286 struct mv_xor_v2_device *xor_dev = data; 287 unsigned int ndescs; 288 u32 reg; 289 290 reg = readl(xor_dev->dma_base + MV_XOR_V2_DMA_DESQ_DONE_OFF); 291 292 ndescs = ((reg >> MV_XOR_V2_DMA_DESQ_DONE_PENDING_SHIFT) & 293 MV_XOR_V2_DMA_DESQ_DONE_PENDING_MASK); 294 295 /* No descriptors to process */ 296 if (!ndescs) 297 return IRQ_NONE; 298 299 /* schedule a tasklet to handle descriptors callbacks */ 300 tasklet_schedule(&xor_dev->irq_tasklet); 301 302 return IRQ_HANDLED; 303 } 304 305 /* 306 * submit a descriptor to the DMA engine 307 */ 308 static dma_cookie_t 309 mv_xor_v2_tx_submit(struct dma_async_tx_descriptor *tx) 310 { 311 void *dest_hw_desc; 312 dma_cookie_t cookie; 313 struct mv_xor_v2_sw_desc *sw_desc = 314 container_of(tx, struct mv_xor_v2_sw_desc, async_tx); 315 struct mv_xor_v2_device *xor_dev = 316 container_of(tx->chan, struct mv_xor_v2_device, dmachan); 317 318 dev_dbg(xor_dev->dmadev.dev, 319 "%s sw_desc %p: async_tx %p\n", 320 __func__, sw_desc, &sw_desc->async_tx); 321 322 /* assign coookie */ 323 spin_lock_bh(&xor_dev->lock); 324 cookie = dma_cookie_assign(tx); 325 326 /* copy the HW descriptor from the SW descriptor to the DESQ */ 327 dest_hw_desc = xor_dev->hw_desq_virt + xor_dev->hw_queue_idx; 328 329 memcpy(dest_hw_desc, &sw_desc->hw_desc, xor_dev->desc_size); 330 331 xor_dev->npendings++; 332 xor_dev->hw_queue_idx++; 333 if (xor_dev->hw_queue_idx >= MV_XOR_V2_DESC_NUM) 334 xor_dev->hw_queue_idx = 0; 335 336 spin_unlock_bh(&xor_dev->lock); 337 338 return cookie; 339 } 340 341 /* 342 * Prepare a SW descriptor 343 */ 344 static struct mv_xor_v2_sw_desc * 345 mv_xor_v2_prep_sw_desc(struct mv_xor_v2_device *xor_dev) 346 { 347 struct mv_xor_v2_sw_desc *sw_desc; 348 bool found = false; 349 350 /* Lock the channel */ 351 spin_lock_bh(&xor_dev->lock); 352 353 if (list_empty(&xor_dev->free_sw_desc)) { 354 spin_unlock_bh(&xor_dev->lock); 355 /* schedule tasklet to free some descriptors */ 356 tasklet_schedule(&xor_dev->irq_tasklet); 357 return NULL; 358 } 359 360 list_for_each_entry(sw_desc, &xor_dev->free_sw_desc, free_list) { 361 if (async_tx_test_ack(&sw_desc->async_tx)) { 362 found = true; 363 break; 364 } 365 } 366 367 if (!found) { 368 spin_unlock_bh(&xor_dev->lock); 369 return NULL; 370 } 371 372 list_del(&sw_desc->free_list); 373 374 /* Release the channel */ 375 spin_unlock_bh(&xor_dev->lock); 376 377 return sw_desc; 378 } 379 380 /* 381 * Prepare a HW descriptor for a memcpy operation 382 */ 383 static struct dma_async_tx_descriptor * 384 mv_xor_v2_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, 385 dma_addr_t src, size_t len, unsigned long flags) 386 { 387 struct mv_xor_v2_sw_desc *sw_desc; 388 struct mv_xor_v2_descriptor *hw_descriptor; 389 struct mv_xor_v2_device *xor_dev; 390 391 xor_dev = container_of(chan, struct mv_xor_v2_device, dmachan); 392 393 dev_dbg(xor_dev->dmadev.dev, 394 "%s len: %zu src %pad dest %pad flags: %ld\n", 395 __func__, len, &src, &dest, flags); 396 397 sw_desc = mv_xor_v2_prep_sw_desc(xor_dev); 398 if (!sw_desc) 399 return NULL; 400 401 sw_desc->async_tx.flags = flags; 402 403 /* set the HW descriptor */ 404 hw_descriptor = &sw_desc->hw_desc; 405 406 /* save the SW descriptor ID to restore when operation is done */ 407 hw_descriptor->desc_id = sw_desc->idx; 408 409 /* Set the MEMCPY control word */ 410 hw_descriptor->desc_ctrl = 411 DESC_OP_MODE_MEMCPY << DESC_OP_MODE_SHIFT; 412 413 if (flags & DMA_PREP_INTERRUPT) 414 hw_descriptor->desc_ctrl |= DESC_IOD; 415 416 /* Set source address */ 417 hw_descriptor->fill_pattern_src_addr[0] = lower_32_bits(src); 418 hw_descriptor->fill_pattern_src_addr[1] = 419 upper_32_bits(src) & 0xFFFF; 420 421 /* Set Destination address */ 422 hw_descriptor->fill_pattern_src_addr[2] = lower_32_bits(dest); 423 hw_descriptor->fill_pattern_src_addr[3] = 424 upper_32_bits(dest) & 0xFFFF; 425 426 /* Set buffers size */ 427 hw_descriptor->buff_size = len; 428 429 /* return the async tx descriptor */ 430 return &sw_desc->async_tx; 431 } 432 433 /* 434 * Prepare a HW descriptor for a XOR operation 435 */ 436 static struct dma_async_tx_descriptor * 437 mv_xor_v2_prep_dma_xor(struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src, 438 unsigned int src_cnt, size_t len, unsigned long flags) 439 { 440 struct mv_xor_v2_sw_desc *sw_desc; 441 struct mv_xor_v2_descriptor *hw_descriptor; 442 struct mv_xor_v2_device *xor_dev = 443 container_of(chan, struct mv_xor_v2_device, dmachan); 444 int i; 445 446 if (src_cnt > MV_XOR_V2_CMD_LINE_NUM_MAX_D_BUF || src_cnt < 1) 447 return NULL; 448 449 dev_dbg(xor_dev->dmadev.dev, 450 "%s src_cnt: %d len: %zu dest %pad flags: %ld\n", 451 __func__, src_cnt, len, &dest, flags); 452 453 sw_desc = mv_xor_v2_prep_sw_desc(xor_dev); 454 if (!sw_desc) 455 return NULL; 456 457 sw_desc->async_tx.flags = flags; 458 459 /* set the HW descriptor */ 460 hw_descriptor = &sw_desc->hw_desc; 461 462 /* save the SW descriptor ID to restore when operation is done */ 463 hw_descriptor->desc_id = sw_desc->idx; 464 465 /* Set the XOR control word */ 466 hw_descriptor->desc_ctrl = 467 DESC_OP_MODE_XOR << DESC_OP_MODE_SHIFT; 468 hw_descriptor->desc_ctrl |= DESC_P_BUFFER_ENABLE; 469 470 if (flags & DMA_PREP_INTERRUPT) 471 hw_descriptor->desc_ctrl |= DESC_IOD; 472 473 /* Set the data buffers */ 474 for (i = 0; i < src_cnt; i++) 475 mv_xor_v2_set_data_buffers(xor_dev, hw_descriptor, src[i], i); 476 477 hw_descriptor->desc_ctrl |= 478 src_cnt << DESC_NUM_ACTIVE_D_BUF_SHIFT; 479 480 /* Set Destination address */ 481 hw_descriptor->fill_pattern_src_addr[2] = lower_32_bits(dest); 482 hw_descriptor->fill_pattern_src_addr[3] = 483 upper_32_bits(dest) & 0xFFFF; 484 485 /* Set buffers size */ 486 hw_descriptor->buff_size = len; 487 488 /* return the async tx descriptor */ 489 return &sw_desc->async_tx; 490 } 491 492 /* 493 * Prepare a HW descriptor for interrupt operation. 494 */ 495 static struct dma_async_tx_descriptor * 496 mv_xor_v2_prep_dma_interrupt(struct dma_chan *chan, unsigned long flags) 497 { 498 struct mv_xor_v2_sw_desc *sw_desc; 499 struct mv_xor_v2_descriptor *hw_descriptor; 500 struct mv_xor_v2_device *xor_dev = 501 container_of(chan, struct mv_xor_v2_device, dmachan); 502 503 sw_desc = mv_xor_v2_prep_sw_desc(xor_dev); 504 if (!sw_desc) 505 return NULL; 506 507 /* set the HW descriptor */ 508 hw_descriptor = &sw_desc->hw_desc; 509 510 /* save the SW descriptor ID to restore when operation is done */ 511 hw_descriptor->desc_id = sw_desc->idx; 512 513 /* Set the INTERRUPT control word */ 514 hw_descriptor->desc_ctrl = 515 DESC_OP_MODE_NOP << DESC_OP_MODE_SHIFT; 516 hw_descriptor->desc_ctrl |= DESC_IOD; 517 518 /* return the async tx descriptor */ 519 return &sw_desc->async_tx; 520 } 521 522 /* 523 * push pending transactions to hardware 524 */ 525 static void mv_xor_v2_issue_pending(struct dma_chan *chan) 526 { 527 struct mv_xor_v2_device *xor_dev = 528 container_of(chan, struct mv_xor_v2_device, dmachan); 529 530 spin_lock_bh(&xor_dev->lock); 531 532 /* 533 * update the engine with the number of descriptors to 534 * process 535 */ 536 mv_xor_v2_add_desc_to_desq(xor_dev, xor_dev->npendings); 537 xor_dev->npendings = 0; 538 539 spin_unlock_bh(&xor_dev->lock); 540 } 541 542 static inline 543 int mv_xor_v2_get_pending_params(struct mv_xor_v2_device *xor_dev, 544 int *pending_ptr) 545 { 546 u32 reg; 547 548 reg = readl(xor_dev->dma_base + MV_XOR_V2_DMA_DESQ_DONE_OFF); 549 550 /* get the next pending descriptor index */ 551 *pending_ptr = ((reg >> MV_XOR_V2_DMA_DESQ_DONE_READ_PTR_SHIFT) & 552 MV_XOR_V2_DMA_DESQ_DONE_READ_PTR_MASK); 553 554 /* get the number of descriptors pending handle */ 555 return ((reg >> MV_XOR_V2_DMA_DESQ_DONE_PENDING_SHIFT) & 556 MV_XOR_V2_DMA_DESQ_DONE_PENDING_MASK); 557 } 558 559 /* 560 * handle the descriptors after HW process 561 */ 562 static void mv_xor_v2_tasklet(unsigned long data) 563 { 564 struct mv_xor_v2_device *xor_dev = (struct mv_xor_v2_device *) data; 565 int pending_ptr, num_of_pending, i; 566 struct mv_xor_v2_sw_desc *next_pending_sw_desc = NULL; 567 568 dev_dbg(xor_dev->dmadev.dev, "%s %d\n", __func__, __LINE__); 569 570 /* get the pending descriptors parameters */ 571 num_of_pending = mv_xor_v2_get_pending_params(xor_dev, &pending_ptr); 572 573 /* loop over free descriptors */ 574 for (i = 0; i < num_of_pending; i++) { 575 struct mv_xor_v2_descriptor *next_pending_hw_desc = 576 xor_dev->hw_desq_virt + pending_ptr; 577 578 /* get the SW descriptor related to the HW descriptor */ 579 next_pending_sw_desc = 580 &xor_dev->sw_desq[next_pending_hw_desc->desc_id]; 581 582 /* call the callback */ 583 if (next_pending_sw_desc->async_tx.cookie > 0) { 584 /* 585 * update the channel's completed cookie - no 586 * lock is required the IMSG threshold provide 587 * the locking 588 */ 589 dma_cookie_complete(&next_pending_sw_desc->async_tx); 590 591 if (next_pending_sw_desc->async_tx.callback) 592 next_pending_sw_desc->async_tx.callback( 593 next_pending_sw_desc->async_tx.callback_param); 594 595 dma_descriptor_unmap(&next_pending_sw_desc->async_tx); 596 } 597 598 dma_run_dependencies(&next_pending_sw_desc->async_tx); 599 600 /* Lock the channel */ 601 spin_lock_bh(&xor_dev->lock); 602 603 /* add the SW descriptor to the free descriptors list */ 604 list_add(&next_pending_sw_desc->free_list, 605 &xor_dev->free_sw_desc); 606 607 /* Release the channel */ 608 spin_unlock_bh(&xor_dev->lock); 609 610 /* increment the next descriptor */ 611 pending_ptr++; 612 if (pending_ptr >= MV_XOR_V2_DESC_NUM) 613 pending_ptr = 0; 614 } 615 616 if (num_of_pending != 0) { 617 /* free the descriptores */ 618 mv_xor_v2_free_desc_from_desq(xor_dev, num_of_pending); 619 } 620 } 621 622 /* 623 * Set DMA Interrupt-message (IMSG) parameters 624 */ 625 static void mv_xor_v2_set_msi_msg(struct msi_desc *desc, struct msi_msg *msg) 626 { 627 struct mv_xor_v2_device *xor_dev = dev_get_drvdata(desc->dev); 628 629 writel(msg->address_lo, 630 xor_dev->dma_base + MV_XOR_V2_DMA_IMSG_BALR_OFF); 631 writel(msg->address_hi & 0xFFFF, 632 xor_dev->dma_base + MV_XOR_V2_DMA_IMSG_BAHR_OFF); 633 writel(msg->data, 634 xor_dev->dma_base + MV_XOR_V2_DMA_IMSG_CDAT_OFF); 635 } 636 637 static int mv_xor_v2_descq_init(struct mv_xor_v2_device *xor_dev) 638 { 639 u32 reg; 640 641 /* write the DESQ size to the DMA engine */ 642 writel(MV_XOR_V2_DESC_NUM, 643 xor_dev->dma_base + MV_XOR_V2_DMA_DESQ_SIZE_OFF); 644 645 /* write the DESQ address to the DMA enngine*/ 646 writel(xor_dev->hw_desq & 0xFFFFFFFF, 647 xor_dev->dma_base + MV_XOR_V2_DMA_DESQ_BALR_OFF); 648 writel((xor_dev->hw_desq & 0xFFFF00000000) >> 32, 649 xor_dev->dma_base + MV_XOR_V2_DMA_DESQ_BAHR_OFF); 650 651 /* 652 * This is a temporary solution, until we activate the 653 * SMMU. Set the attributes for reading & writing data buffers 654 * & descriptors to: 655 * 656 * - OuterShareable - Snoops will be performed on CPU caches 657 * - Enable cacheable - Bufferable, Modifiable, Other Allocate 658 * and Allocate 659 */ 660 reg = readl(xor_dev->dma_base + MV_XOR_V2_DMA_DESQ_ARATTR_OFF); 661 reg &= ~MV_XOR_V2_DMA_DESQ_ATTR_CACHE_MASK; 662 reg |= MV_XOR_V2_DMA_DESQ_ATTR_OUTER_SHAREABLE | 663 MV_XOR_V2_DMA_DESQ_ATTR_CACHEABLE; 664 writel(reg, xor_dev->dma_base + MV_XOR_V2_DMA_DESQ_ARATTR_OFF); 665 666 reg = readl(xor_dev->dma_base + MV_XOR_V2_DMA_DESQ_AWATTR_OFF); 667 reg &= ~MV_XOR_V2_DMA_DESQ_ATTR_CACHE_MASK; 668 reg |= MV_XOR_V2_DMA_DESQ_ATTR_OUTER_SHAREABLE | 669 MV_XOR_V2_DMA_DESQ_ATTR_CACHEABLE; 670 writel(reg, xor_dev->dma_base + MV_XOR_V2_DMA_DESQ_AWATTR_OFF); 671 672 /* BW CTRL - set values to optimize the XOR performance: 673 * 674 * - Set WrBurstLen & RdBurstLen - the unit will issue 675 * maximum of 256B write/read transactions. 676 * - Limit the number of outstanding write & read data 677 * (OBB/IBB) requests to the maximal value. 678 */ 679 reg = ((MV_XOR_V2_GLOB_BW_CTRL_NUM_OSTD_RD_VAL << 680 MV_XOR_V2_GLOB_BW_CTRL_NUM_OSTD_RD_SHIFT) | 681 (MV_XOR_V2_GLOB_BW_CTRL_NUM_OSTD_WR_VAL << 682 MV_XOR_V2_GLOB_BW_CTRL_NUM_OSTD_WR_SHIFT) | 683 (MV_XOR_V2_GLOB_BW_CTRL_RD_BURST_LEN_VAL << 684 MV_XOR_V2_GLOB_BW_CTRL_RD_BURST_LEN_SHIFT) | 685 (MV_XOR_V2_GLOB_BW_CTRL_WR_BURST_LEN_VAL << 686 MV_XOR_V2_GLOB_BW_CTRL_WR_BURST_LEN_SHIFT)); 687 writel(reg, xor_dev->glob_base + MV_XOR_V2_GLOB_BW_CTRL); 688 689 /* Disable the AXI timer feature */ 690 reg = readl(xor_dev->glob_base + MV_XOR_V2_GLOB_PAUSE); 691 reg |= MV_XOR_V2_GLOB_PAUSE_AXI_TIME_DIS_VAL; 692 writel(reg, xor_dev->glob_base + MV_XOR_V2_GLOB_PAUSE); 693 694 /* enable the DMA engine */ 695 writel(0, xor_dev->dma_base + MV_XOR_V2_DMA_DESQ_STOP_OFF); 696 697 return 0; 698 } 699 700 static int mv_xor_v2_suspend(struct platform_device *dev, pm_message_t state) 701 { 702 struct mv_xor_v2_device *xor_dev = platform_get_drvdata(dev); 703 704 /* Set this bit to disable to stop the XOR unit. */ 705 writel(0x1, xor_dev->dma_base + MV_XOR_V2_DMA_DESQ_STOP_OFF); 706 707 return 0; 708 } 709 710 static int mv_xor_v2_resume(struct platform_device *dev) 711 { 712 struct mv_xor_v2_device *xor_dev = platform_get_drvdata(dev); 713 714 mv_xor_v2_set_desc_size(xor_dev); 715 mv_xor_v2_enable_imsg_thrd(xor_dev); 716 mv_xor_v2_descq_init(xor_dev); 717 718 return 0; 719 } 720 721 static int mv_xor_v2_probe(struct platform_device *pdev) 722 { 723 struct mv_xor_v2_device *xor_dev; 724 struct resource *res; 725 int i, ret = 0; 726 struct dma_device *dma_dev; 727 struct mv_xor_v2_sw_desc *sw_desc; 728 struct msi_desc *msi_desc; 729 730 BUILD_BUG_ON(sizeof(struct mv_xor_v2_descriptor) != 731 MV_XOR_V2_EXT_DESC_SIZE); 732 733 xor_dev = devm_kzalloc(&pdev->dev, sizeof(*xor_dev), GFP_KERNEL); 734 if (!xor_dev) 735 return -ENOMEM; 736 737 res = platform_get_resource(pdev, IORESOURCE_MEM, 0); 738 xor_dev->dma_base = devm_ioremap_resource(&pdev->dev, res); 739 if (IS_ERR(xor_dev->dma_base)) 740 return PTR_ERR(xor_dev->dma_base); 741 742 res = platform_get_resource(pdev, IORESOURCE_MEM, 1); 743 xor_dev->glob_base = devm_ioremap_resource(&pdev->dev, res); 744 if (IS_ERR(xor_dev->glob_base)) 745 return PTR_ERR(xor_dev->glob_base); 746 747 platform_set_drvdata(pdev, xor_dev); 748 749 ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(40)); 750 if (ret) 751 return ret; 752 753 xor_dev->reg_clk = devm_clk_get(&pdev->dev, "reg"); 754 if (PTR_ERR(xor_dev->reg_clk) != -ENOENT) { 755 if (!IS_ERR(xor_dev->reg_clk)) { 756 ret = clk_prepare_enable(xor_dev->reg_clk); 757 if (ret) 758 return ret; 759 } else { 760 return PTR_ERR(xor_dev->reg_clk); 761 } 762 } 763 764 xor_dev->clk = devm_clk_get(&pdev->dev, NULL); 765 if (IS_ERR(xor_dev->clk) && PTR_ERR(xor_dev->clk) == -EPROBE_DEFER) { 766 ret = EPROBE_DEFER; 767 goto disable_reg_clk; 768 } 769 if (!IS_ERR(xor_dev->clk)) { 770 ret = clk_prepare_enable(xor_dev->clk); 771 if (ret) 772 goto disable_reg_clk; 773 } 774 775 ret = platform_msi_domain_alloc_irqs(&pdev->dev, 1, 776 mv_xor_v2_set_msi_msg); 777 if (ret) 778 goto disable_clk; 779 780 msi_desc = first_msi_entry(&pdev->dev); 781 if (!msi_desc) 782 goto free_msi_irqs; 783 784 ret = devm_request_irq(&pdev->dev, msi_desc->irq, 785 mv_xor_v2_interrupt_handler, 0, 786 dev_name(&pdev->dev), xor_dev); 787 if (ret) 788 goto free_msi_irqs; 789 790 tasklet_init(&xor_dev->irq_tasklet, mv_xor_v2_tasklet, 791 (unsigned long) xor_dev); 792 793 xor_dev->desc_size = mv_xor_v2_set_desc_size(xor_dev); 794 795 dma_cookie_init(&xor_dev->dmachan); 796 797 /* 798 * allocate coherent memory for hardware descriptors 799 * note: writecombine gives slightly better performance, but 800 * requires that we explicitly flush the writes 801 */ 802 xor_dev->hw_desq_virt = 803 dma_alloc_coherent(&pdev->dev, 804 xor_dev->desc_size * MV_XOR_V2_DESC_NUM, 805 &xor_dev->hw_desq, GFP_KERNEL); 806 if (!xor_dev->hw_desq_virt) { 807 ret = -ENOMEM; 808 goto free_msi_irqs; 809 } 810 811 /* alloc memory for the SW descriptors */ 812 xor_dev->sw_desq = devm_kzalloc(&pdev->dev, sizeof(*sw_desc) * 813 MV_XOR_V2_DESC_NUM, GFP_KERNEL); 814 if (!xor_dev->sw_desq) { 815 ret = -ENOMEM; 816 goto free_hw_desq; 817 } 818 819 spin_lock_init(&xor_dev->lock); 820 821 /* init the free SW descriptors list */ 822 INIT_LIST_HEAD(&xor_dev->free_sw_desc); 823 824 /* add all SW descriptors to the free list */ 825 for (i = 0; i < MV_XOR_V2_DESC_NUM; i++) { 826 struct mv_xor_v2_sw_desc *sw_desc = 827 xor_dev->sw_desq + i; 828 sw_desc->idx = i; 829 dma_async_tx_descriptor_init(&sw_desc->async_tx, 830 &xor_dev->dmachan); 831 sw_desc->async_tx.tx_submit = mv_xor_v2_tx_submit; 832 async_tx_ack(&sw_desc->async_tx); 833 834 list_add(&sw_desc->free_list, 835 &xor_dev->free_sw_desc); 836 } 837 838 dma_dev = &xor_dev->dmadev; 839 840 /* set DMA capabilities */ 841 dma_cap_zero(dma_dev->cap_mask); 842 dma_cap_set(DMA_MEMCPY, dma_dev->cap_mask); 843 dma_cap_set(DMA_XOR, dma_dev->cap_mask); 844 dma_cap_set(DMA_INTERRUPT, dma_dev->cap_mask); 845 846 /* init dma link list */ 847 INIT_LIST_HEAD(&dma_dev->channels); 848 849 /* set base routines */ 850 dma_dev->device_tx_status = dma_cookie_status; 851 dma_dev->device_issue_pending = mv_xor_v2_issue_pending; 852 dma_dev->dev = &pdev->dev; 853 854 dma_dev->device_prep_dma_memcpy = mv_xor_v2_prep_dma_memcpy; 855 dma_dev->device_prep_dma_interrupt = mv_xor_v2_prep_dma_interrupt; 856 dma_dev->max_xor = 8; 857 dma_dev->device_prep_dma_xor = mv_xor_v2_prep_dma_xor; 858 859 xor_dev->dmachan.device = dma_dev; 860 861 list_add_tail(&xor_dev->dmachan.device_node, 862 &dma_dev->channels); 863 864 mv_xor_v2_enable_imsg_thrd(xor_dev); 865 866 mv_xor_v2_descq_init(xor_dev); 867 868 ret = dma_async_device_register(dma_dev); 869 if (ret) 870 goto free_hw_desq; 871 872 dev_notice(&pdev->dev, "Marvell Version 2 XOR driver\n"); 873 874 return 0; 875 876 free_hw_desq: 877 dma_free_coherent(&pdev->dev, 878 xor_dev->desc_size * MV_XOR_V2_DESC_NUM, 879 xor_dev->hw_desq_virt, xor_dev->hw_desq); 880 free_msi_irqs: 881 platform_msi_domain_free_irqs(&pdev->dev); 882 disable_clk: 883 clk_disable_unprepare(xor_dev->clk); 884 disable_reg_clk: 885 clk_disable_unprepare(xor_dev->reg_clk); 886 return ret; 887 } 888 889 static int mv_xor_v2_remove(struct platform_device *pdev) 890 { 891 struct mv_xor_v2_device *xor_dev = platform_get_drvdata(pdev); 892 893 dma_async_device_unregister(&xor_dev->dmadev); 894 895 dma_free_coherent(&pdev->dev, 896 xor_dev->desc_size * MV_XOR_V2_DESC_NUM, 897 xor_dev->hw_desq_virt, xor_dev->hw_desq); 898 899 platform_msi_domain_free_irqs(&pdev->dev); 900 901 clk_disable_unprepare(xor_dev->clk); 902 903 return 0; 904 } 905 906 #ifdef CONFIG_OF 907 static const struct of_device_id mv_xor_v2_dt_ids[] = { 908 { .compatible = "marvell,xor-v2", }, 909 {}, 910 }; 911 MODULE_DEVICE_TABLE(of, mv_xor_v2_dt_ids); 912 #endif 913 914 static struct platform_driver mv_xor_v2_driver = { 915 .probe = mv_xor_v2_probe, 916 .suspend = mv_xor_v2_suspend, 917 .resume = mv_xor_v2_resume, 918 .remove = mv_xor_v2_remove, 919 .driver = { 920 .name = "mv_xor_v2", 921 .of_match_table = of_match_ptr(mv_xor_v2_dt_ids), 922 }, 923 }; 924 925 module_platform_driver(mv_xor_v2_driver); 926 927 MODULE_DESCRIPTION("DMA engine driver for Marvell's Version 2 of XOR engine"); 928 MODULE_LICENSE("GPL"); 929