1// SPDX-License-Identifier: GPL-2.0-only 2/* 3 * Copyright (C) Ericsson AB 2007-2008 4 * Copyright (C) ST-Ericsson SA 2008-2010 5 * Author: Per Forlin <per.forlin@stericsson.com> for ST-Ericsson 6 * Author: Jonas Aaberg <jonas.aberg@stericsson.com> for ST-Ericsson 7 */ 8 9#include <linux/dma-mapping.h> 10#include <linux/kernel.h> 11#include <linux/slab.h> 12#include <linux/export.h> 13#include <linux/dmaengine.h> 14#include <linux/platform_device.h> 15#include <linux/clk.h> 16#include <linux/delay.h> 17#include <linux/log2.h> 18#include <linux/pm.h> 19#include <linux/pm_runtime.h> 20#include <linux/err.h> 21#include <linux/of.h> 22#include <linux/of_address.h> 23#include <linux/of_dma.h> 24#include <linux/amba/bus.h> 25#include <linux/regulator/consumer.h> 26 27#include "dmaengine.h" 28#include "ste_dma40.h" 29#include "ste_dma40_ll.h" 30 31/** 32 * struct stedma40_platform_data - Configuration struct for the dma device. 33 * 34 * @dev_tx: mapping between destination event line and io address 35 * @dev_rx: mapping between source event line and io address 36 * @disabled_channels: A vector, ending with -1, that marks physical channels 37 * that are for different reasons not available for the driver. 38 * @soft_lli_chans: A vector, that marks physical channels will use LLI by SW 39 * which avoids HW bug that exists in some versions of the controller. 40 * SoftLLI introduces relink overhead that could impact performace for 41 * certain use cases. 42 * @num_of_soft_lli_chans: The number of channels that needs to be configured 43 * to use SoftLLI. 44 * @use_esram_lcla: flag for mapping the lcla into esram region 45 * @num_of_memcpy_chans: The number of channels reserved for memcpy. 46 * @num_of_phy_chans: The number of physical channels implemented in HW. 47 * 0 means reading the number of channels from DMA HW but this is only valid 48 * for 'multiple of 4' channels, like 8. 49 */ 50struct stedma40_platform_data { 51 int disabled_channels[STEDMA40_MAX_PHYS]; 52 int *soft_lli_chans; 53 int num_of_soft_lli_chans; 54 bool use_esram_lcla; 55 int num_of_memcpy_chans; 56 int num_of_phy_chans; 57}; 58 59#define D40_NAME "dma40" 60 61#define D40_PHY_CHAN -1 62 63/* For masking out/in 2 bit channel positions */ 64#define D40_CHAN_POS(chan) (2 * (chan / 2)) 65#define D40_CHAN_POS_MASK(chan) (0x3 << D40_CHAN_POS(chan)) 66 67/* Maximum iterations taken before giving up suspending a channel */ 68#define D40_SUSPEND_MAX_IT 500 69 70/* Milliseconds */ 71#define DMA40_AUTOSUSPEND_DELAY 100 72 73/* Hardware requirement on LCLA alignment */ 74#define LCLA_ALIGNMENT 0x40000 75 76/* Max number of links per event group */ 77#define D40_LCLA_LINK_PER_EVENT_GRP 128 78#define D40_LCLA_END D40_LCLA_LINK_PER_EVENT_GRP 79 80/* Max number of logical channels per physical channel */ 81#define D40_MAX_LOG_CHAN_PER_PHY 32 82 83/* Attempts before giving up to trying to get pages that are aligned */ 84#define MAX_LCLA_ALLOC_ATTEMPTS 256 85 86/* Bit markings for allocation map */ 87#define D40_ALLOC_FREE BIT(31) 88#define D40_ALLOC_PHY BIT(30) 89#define D40_ALLOC_LOG_FREE 0 90 91#define D40_MEMCPY_MAX_CHANS 8 92 93/* Reserved event lines for memcpy only. */ 94#define DB8500_DMA_MEMCPY_EV_0 51 95#define DB8500_DMA_MEMCPY_EV_1 56 96#define DB8500_DMA_MEMCPY_EV_2 57 97#define DB8500_DMA_MEMCPY_EV_3 58 98#define DB8500_DMA_MEMCPY_EV_4 59 99#define DB8500_DMA_MEMCPY_EV_5 60 100 101static int dma40_memcpy_channels[] = { 102 DB8500_DMA_MEMCPY_EV_0, 103 DB8500_DMA_MEMCPY_EV_1, 104 DB8500_DMA_MEMCPY_EV_2, 105 DB8500_DMA_MEMCPY_EV_3, 106 DB8500_DMA_MEMCPY_EV_4, 107 DB8500_DMA_MEMCPY_EV_5, 108}; 109 110/* Default configuration for physical memcpy */ 111static const struct stedma40_chan_cfg dma40_memcpy_conf_phy = { 112 .mode = STEDMA40_MODE_PHYSICAL, 113 .dir = DMA_MEM_TO_MEM, 114 115 .src_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE, 116 .src_info.psize = STEDMA40_PSIZE_PHY_1, 117 .src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL, 118 119 .dst_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE, 120 .dst_info.psize = STEDMA40_PSIZE_PHY_1, 121 .dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL, 122}; 123 124/* Default configuration for logical memcpy */ 125static const struct stedma40_chan_cfg dma40_memcpy_conf_log = { 126 .mode = STEDMA40_MODE_LOGICAL, 127 .dir = DMA_MEM_TO_MEM, 128 129 .src_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE, 130 .src_info.psize = STEDMA40_PSIZE_LOG_1, 131 .src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL, 132 133 .dst_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE, 134 .dst_info.psize = STEDMA40_PSIZE_LOG_1, 135 .dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL, 136}; 137 138/** 139 * enum d40_command - The different commands and/or statuses. 140 * 141 * @D40_DMA_STOP: DMA channel command STOP or status STOPPED, 142 * @D40_DMA_RUN: The DMA channel is RUNNING of the command RUN. 143 * @D40_DMA_SUSPEND_REQ: Request the DMA to SUSPEND as soon as possible. 144 * @D40_DMA_SUSPENDED: The DMA channel is SUSPENDED. 145 */ 146enum d40_command { 147 D40_DMA_STOP = 0, 148 D40_DMA_RUN = 1, 149 D40_DMA_SUSPEND_REQ = 2, 150 D40_DMA_SUSPENDED = 3 151}; 152 153/* 154 * enum d40_events - The different Event Enables for the event lines. 155 * 156 * @D40_DEACTIVATE_EVENTLINE: De-activate Event line, stopping the logical chan. 157 * @D40_ACTIVATE_EVENTLINE: Activate the Event line, to start a logical chan. 158 * @D40_SUSPEND_REQ_EVENTLINE: Requesting for suspending a event line. 159 * @D40_ROUND_EVENTLINE: Status check for event line. 160 */ 161 162enum d40_events { 163 D40_DEACTIVATE_EVENTLINE = 0, 164 D40_ACTIVATE_EVENTLINE = 1, 165 D40_SUSPEND_REQ_EVENTLINE = 2, 166 D40_ROUND_EVENTLINE = 3 167}; 168 169/* 170 * These are the registers that has to be saved and later restored 171 * when the DMA hw is powered off. 172 * TODO: Add save/restore of D40_DREG_GCC on dma40 v3 or later, if that works. 173 */ 174static __maybe_unused u32 d40_backup_regs[] = { 175 D40_DREG_LCPA, 176 D40_DREG_LCLA, 177 D40_DREG_PRMSE, 178 D40_DREG_PRMSO, 179 D40_DREG_PRMOE, 180 D40_DREG_PRMOO, 181}; 182 183#define BACKUP_REGS_SZ ARRAY_SIZE(d40_backup_regs) 184 185/* 186 * since 9540 and 8540 has the same HW revision 187 * use v4a for 9540 or ealier 188 * use v4b for 8540 or later 189 * HW revision: 190 * DB8500ed has revision 0 191 * DB8500v1 has revision 2 192 * DB8500v2 has revision 3 193 * AP9540v1 has revision 4 194 * DB8540v1 has revision 4 195 * TODO: Check if all these registers have to be saved/restored on dma40 v4a 196 */ 197static u32 d40_backup_regs_v4a[] = { 198 D40_DREG_PSEG1, 199 D40_DREG_PSEG2, 200 D40_DREG_PSEG3, 201 D40_DREG_PSEG4, 202 D40_DREG_PCEG1, 203 D40_DREG_PCEG2, 204 D40_DREG_PCEG3, 205 D40_DREG_PCEG4, 206 D40_DREG_RSEG1, 207 D40_DREG_RSEG2, 208 D40_DREG_RSEG3, 209 D40_DREG_RSEG4, 210 D40_DREG_RCEG1, 211 D40_DREG_RCEG2, 212 D40_DREG_RCEG3, 213 D40_DREG_RCEG4, 214}; 215 216#define BACKUP_REGS_SZ_V4A ARRAY_SIZE(d40_backup_regs_v4a) 217 218static u32 d40_backup_regs_v4b[] = { 219 D40_DREG_CPSEG1, 220 D40_DREG_CPSEG2, 221 D40_DREG_CPSEG3, 222 D40_DREG_CPSEG4, 223 D40_DREG_CPSEG5, 224 D40_DREG_CPCEG1, 225 D40_DREG_CPCEG2, 226 D40_DREG_CPCEG3, 227 D40_DREG_CPCEG4, 228 D40_DREG_CPCEG5, 229 D40_DREG_CRSEG1, 230 D40_DREG_CRSEG2, 231 D40_DREG_CRSEG3, 232 D40_DREG_CRSEG4, 233 D40_DREG_CRSEG5, 234 D40_DREG_CRCEG1, 235 D40_DREG_CRCEG2, 236 D40_DREG_CRCEG3, 237 D40_DREG_CRCEG4, 238 D40_DREG_CRCEG5, 239}; 240 241#define BACKUP_REGS_SZ_V4B ARRAY_SIZE(d40_backup_regs_v4b) 242 243static __maybe_unused u32 d40_backup_regs_chan[] = { 244 D40_CHAN_REG_SSCFG, 245 D40_CHAN_REG_SSELT, 246 D40_CHAN_REG_SSPTR, 247 D40_CHAN_REG_SSLNK, 248 D40_CHAN_REG_SDCFG, 249 D40_CHAN_REG_SDELT, 250 D40_CHAN_REG_SDPTR, 251 D40_CHAN_REG_SDLNK, 252}; 253 254#define BACKUP_REGS_SZ_MAX ((BACKUP_REGS_SZ_V4A > BACKUP_REGS_SZ_V4B) ? \ 255 BACKUP_REGS_SZ_V4A : BACKUP_REGS_SZ_V4B) 256 257/** 258 * struct d40_interrupt_lookup - lookup table for interrupt handler 259 * 260 * @src: Interrupt mask register. 261 * @clr: Interrupt clear register. 262 * @is_error: true if this is an error interrupt. 263 * @offset: start delta in the lookup_log_chans in d40_base. If equals to 264 * D40_PHY_CHAN, the lookup_phy_chans shall be used instead. 265 */ 266struct d40_interrupt_lookup { 267 u32 src; 268 u32 clr; 269 bool is_error; 270 int offset; 271}; 272 273 274static struct d40_interrupt_lookup il_v4a[] = { 275 {D40_DREG_LCTIS0, D40_DREG_LCICR0, false, 0}, 276 {D40_DREG_LCTIS1, D40_DREG_LCICR1, false, 32}, 277 {D40_DREG_LCTIS2, D40_DREG_LCICR2, false, 64}, 278 {D40_DREG_LCTIS3, D40_DREG_LCICR3, false, 96}, 279 {D40_DREG_LCEIS0, D40_DREG_LCICR0, true, 0}, 280 {D40_DREG_LCEIS1, D40_DREG_LCICR1, true, 32}, 281 {D40_DREG_LCEIS2, D40_DREG_LCICR2, true, 64}, 282 {D40_DREG_LCEIS3, D40_DREG_LCICR3, true, 96}, 283 {D40_DREG_PCTIS, D40_DREG_PCICR, false, D40_PHY_CHAN}, 284 {D40_DREG_PCEIS, D40_DREG_PCICR, true, D40_PHY_CHAN}, 285}; 286 287static struct d40_interrupt_lookup il_v4b[] = { 288 {D40_DREG_CLCTIS1, D40_DREG_CLCICR1, false, 0}, 289 {D40_DREG_CLCTIS2, D40_DREG_CLCICR2, false, 32}, 290 {D40_DREG_CLCTIS3, D40_DREG_CLCICR3, false, 64}, 291 {D40_DREG_CLCTIS4, D40_DREG_CLCICR4, false, 96}, 292 {D40_DREG_CLCTIS5, D40_DREG_CLCICR5, false, 128}, 293 {D40_DREG_CLCEIS1, D40_DREG_CLCICR1, true, 0}, 294 {D40_DREG_CLCEIS2, D40_DREG_CLCICR2, true, 32}, 295 {D40_DREG_CLCEIS3, D40_DREG_CLCICR3, true, 64}, 296 {D40_DREG_CLCEIS4, D40_DREG_CLCICR4, true, 96}, 297 {D40_DREG_CLCEIS5, D40_DREG_CLCICR5, true, 128}, 298 {D40_DREG_CPCTIS, D40_DREG_CPCICR, false, D40_PHY_CHAN}, 299 {D40_DREG_CPCEIS, D40_DREG_CPCICR, true, D40_PHY_CHAN}, 300}; 301 302/** 303 * struct d40_reg_val - simple lookup struct 304 * 305 * @reg: The register. 306 * @val: The value that belongs to the register in reg. 307 */ 308struct d40_reg_val { 309 unsigned int reg; 310 unsigned int val; 311}; 312 313static __initdata struct d40_reg_val dma_init_reg_v4a[] = { 314 /* Clock every part of the DMA block from start */ 315 { .reg = D40_DREG_GCC, .val = D40_DREG_GCC_ENABLE_ALL}, 316 317 /* Interrupts on all logical channels */ 318 { .reg = D40_DREG_LCMIS0, .val = 0xFFFFFFFF}, 319 { .reg = D40_DREG_LCMIS1, .val = 0xFFFFFFFF}, 320 { .reg = D40_DREG_LCMIS2, .val = 0xFFFFFFFF}, 321 { .reg = D40_DREG_LCMIS3, .val = 0xFFFFFFFF}, 322 { .reg = D40_DREG_LCICR0, .val = 0xFFFFFFFF}, 323 { .reg = D40_DREG_LCICR1, .val = 0xFFFFFFFF}, 324 { .reg = D40_DREG_LCICR2, .val = 0xFFFFFFFF}, 325 { .reg = D40_DREG_LCICR3, .val = 0xFFFFFFFF}, 326 { .reg = D40_DREG_LCTIS0, .val = 0xFFFFFFFF}, 327 { .reg = D40_DREG_LCTIS1, .val = 0xFFFFFFFF}, 328 { .reg = D40_DREG_LCTIS2, .val = 0xFFFFFFFF}, 329 { .reg = D40_DREG_LCTIS3, .val = 0xFFFFFFFF} 330}; 331static __initdata struct d40_reg_val dma_init_reg_v4b[] = { 332 /* Clock every part of the DMA block from start */ 333 { .reg = D40_DREG_GCC, .val = D40_DREG_GCC_ENABLE_ALL}, 334 335 /* Interrupts on all logical channels */ 336 { .reg = D40_DREG_CLCMIS1, .val = 0xFFFFFFFF}, 337 { .reg = D40_DREG_CLCMIS2, .val = 0xFFFFFFFF}, 338 { .reg = D40_DREG_CLCMIS3, .val = 0xFFFFFFFF}, 339 { .reg = D40_DREG_CLCMIS4, .val = 0xFFFFFFFF}, 340 { .reg = D40_DREG_CLCMIS5, .val = 0xFFFFFFFF}, 341 { .reg = D40_DREG_CLCICR1, .val = 0xFFFFFFFF}, 342 { .reg = D40_DREG_CLCICR2, .val = 0xFFFFFFFF}, 343 { .reg = D40_DREG_CLCICR3, .val = 0xFFFFFFFF}, 344 { .reg = D40_DREG_CLCICR4, .val = 0xFFFFFFFF}, 345 { .reg = D40_DREG_CLCICR5, .val = 0xFFFFFFFF}, 346 { .reg = D40_DREG_CLCTIS1, .val = 0xFFFFFFFF}, 347 { .reg = D40_DREG_CLCTIS2, .val = 0xFFFFFFFF}, 348 { .reg = D40_DREG_CLCTIS3, .val = 0xFFFFFFFF}, 349 { .reg = D40_DREG_CLCTIS4, .val = 0xFFFFFFFF}, 350 { .reg = D40_DREG_CLCTIS5, .val = 0xFFFFFFFF} 351}; 352 353/** 354 * struct d40_lli_pool - Structure for keeping LLIs in memory 355 * 356 * @base: Pointer to memory area when the pre_alloc_lli's are not large 357 * enough, IE bigger than the most common case, 1 dst and 1 src. NULL if 358 * pre_alloc_lli is used. 359 * @dma_addr: DMA address, if mapped 360 * @size: The size in bytes of the memory at base or the size of pre_alloc_lli. 361 * @pre_alloc_lli: Pre allocated area for the most common case of transfers, 362 * one buffer to one buffer. 363 */ 364struct d40_lli_pool { 365 void *base; 366 int size; 367 dma_addr_t dma_addr; 368 /* Space for dst and src, plus an extra for padding */ 369 u8 pre_alloc_lli[3 * sizeof(struct d40_phy_lli)]; 370}; 371 372/** 373 * struct d40_desc - A descriptor is one DMA job. 374 * 375 * @lli_phy: LLI settings for physical channel. Both src and dst= 376 * points into the lli_pool, to base if lli_len > 1 or to pre_alloc_lli if 377 * lli_len equals one. 378 * @lli_log: Same as above but for logical channels. 379 * @lli_pool: The pool with two entries pre-allocated. 380 * @lli_len: Number of llis of current descriptor. 381 * @lli_current: Number of transferred llis. 382 * @lcla_alloc: Number of LCLA entries allocated. 383 * @txd: DMA engine struct. Used for among other things for communication 384 * during a transfer. 385 * @node: List entry. 386 * @is_in_client_list: true if the client owns this descriptor. 387 * @cyclic: true if this is a cyclic job 388 * 389 * This descriptor is used for both logical and physical transfers. 390 */ 391struct d40_desc { 392 /* LLI physical */ 393 struct d40_phy_lli_bidir lli_phy; 394 /* LLI logical */ 395 struct d40_log_lli_bidir lli_log; 396 397 struct d40_lli_pool lli_pool; 398 int lli_len; 399 int lli_current; 400 int lcla_alloc; 401 402 struct dma_async_tx_descriptor txd; 403 struct list_head node; 404 405 bool is_in_client_list; 406 bool cyclic; 407}; 408 409/** 410 * struct d40_lcla_pool - LCLA pool settings and data. 411 * 412 * @base: The virtual address of LCLA. 18 bit aligned. 413 * @dma_addr: DMA address, if mapped 414 * @base_unaligned: The orignal kmalloc pointer, if kmalloc is used. 415 * This pointer is only there for clean-up on error. 416 * @pages: The number of pages needed for all physical channels. 417 * Only used later for clean-up on error 418 * @lock: Lock to protect the content in this struct. 419 * @alloc_map: big map over which LCLA entry is own by which job. 420 */ 421struct d40_lcla_pool { 422 void *base; 423 dma_addr_t dma_addr; 424 void *base_unaligned; 425 int pages; 426 spinlock_t lock; 427 struct d40_desc **alloc_map; 428}; 429 430/** 431 * struct d40_phy_res - struct for handling eventlines mapped to physical 432 * channels. 433 * 434 * @lock: A lock protection this entity. 435 * @reserved: True if used by secure world or otherwise. 436 * @num: The physical channel number of this entity. 437 * @allocated_src: Bit mapped to show which src event line's are mapped to 438 * this physical channel. Can also be free or physically allocated. 439 * @allocated_dst: Same as for src but is dst. 440 * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as 441 * event line number. 442 * @use_soft_lli: To mark if the linked lists of channel are managed by SW. 443 */ 444struct d40_phy_res { 445 spinlock_t lock; 446 bool reserved; 447 int num; 448 u32 allocated_src; 449 u32 allocated_dst; 450 bool use_soft_lli; 451}; 452 453struct d40_base; 454 455/** 456 * struct d40_chan - Struct that describes a channel. 457 * 458 * @lock: A spinlock to protect this struct. 459 * @log_num: The logical number, if any of this channel. 460 * @pending_tx: The number of pending transfers. Used between interrupt handler 461 * and tasklet. 462 * @busy: Set to true when transfer is ongoing on this channel. 463 * @phy_chan: Pointer to physical channel which this instance runs on. If this 464 * point is NULL, then the channel is not allocated. 465 * @chan: DMA engine handle. 466 * @tasklet: Tasklet that gets scheduled from interrupt context to complete a 467 * transfer and call client callback. 468 * @client: Cliented owned descriptor list. 469 * @pending_queue: Submitted jobs, to be issued by issue_pending() 470 * @active: Active descriptor. 471 * @done: Completed jobs 472 * @queue: Queued jobs. 473 * @prepare_queue: Prepared jobs. 474 * @dma_cfg: The client configuration of this dma channel. 475 * @slave_config: DMA slave configuration. 476 * @configured: whether the dma_cfg configuration is valid 477 * @base: Pointer to the device instance struct. 478 * @src_def_cfg: Default cfg register setting for src. 479 * @dst_def_cfg: Default cfg register setting for dst. 480 * @log_def: Default logical channel settings. 481 * @lcpa: Pointer to dst and src lcpa settings. 482 * @runtime_addr: runtime configured address. 483 * @runtime_direction: runtime configured direction. 484 * 485 * This struct can either "be" a logical or a physical channel. 486 */ 487struct d40_chan { 488 spinlock_t lock; 489 int log_num; 490 int pending_tx; 491 bool busy; 492 struct d40_phy_res *phy_chan; 493 struct dma_chan chan; 494 struct tasklet_struct tasklet; 495 struct list_head client; 496 struct list_head pending_queue; 497 struct list_head active; 498 struct list_head done; 499 struct list_head queue; 500 struct list_head prepare_queue; 501 struct stedma40_chan_cfg dma_cfg; 502 struct dma_slave_config slave_config; 503 bool configured; 504 struct d40_base *base; 505 /* Default register configurations */ 506 u32 src_def_cfg; 507 u32 dst_def_cfg; 508 struct d40_def_lcsp log_def; 509 struct d40_log_lli_full *lcpa; 510 /* Runtime reconfiguration */ 511 dma_addr_t runtime_addr; 512 enum dma_transfer_direction runtime_direction; 513}; 514 515/** 516 * struct d40_gen_dmac - generic values to represent u8500/u8540 DMA 517 * controller 518 * 519 * @backup: the pointer to the registers address array for backup 520 * @backup_size: the size of the registers address array for backup 521 * @realtime_en: the realtime enable register 522 * @realtime_clear: the realtime clear register 523 * @high_prio_en: the high priority enable register 524 * @high_prio_clear: the high priority clear register 525 * @interrupt_en: the interrupt enable register 526 * @interrupt_clear: the interrupt clear register 527 * @il: the pointer to struct d40_interrupt_lookup 528 * @il_size: the size of d40_interrupt_lookup array 529 * @init_reg: the pointer to the struct d40_reg_val 530 * @init_reg_size: the size of d40_reg_val array 531 */ 532struct d40_gen_dmac { 533 u32 *backup; 534 u32 backup_size; 535 u32 realtime_en; 536 u32 realtime_clear; 537 u32 high_prio_en; 538 u32 high_prio_clear; 539 u32 interrupt_en; 540 u32 interrupt_clear; 541 struct d40_interrupt_lookup *il; 542 u32 il_size; 543 struct d40_reg_val *init_reg; 544 u32 init_reg_size; 545}; 546 547/** 548 * struct d40_base - The big global struct, one for each probe'd instance. 549 * 550 * @interrupt_lock: Lock used to make sure one interrupt is handle a time. 551 * @execmd_lock: Lock for execute command usage since several channels share 552 * the same physical register. 553 * @dev: The device structure. 554 * @virtbase: The virtual base address of the DMA's register. 555 * @rev: silicon revision detected. 556 * @clk: Pointer to the DMA clock structure. 557 * @irq: The IRQ number. 558 * @num_memcpy_chans: The number of channels used for memcpy (mem-to-mem 559 * transfers). 560 * @num_phy_chans: The number of physical channels. Read from HW. This 561 * is the number of available channels for this driver, not counting "Secure 562 * mode" allocated physical channels. 563 * @num_log_chans: The number of logical channels. Calculated from 564 * num_phy_chans. 565 * @dma_both: dma_device channels that can do both memcpy and slave transfers. 566 * @dma_slave: dma_device channels that can do only do slave transfers. 567 * @dma_memcpy: dma_device channels that can do only do memcpy transfers. 568 * @phy_chans: Room for all possible physical channels in system. 569 * @log_chans: Room for all possible logical channels in system. 570 * @lookup_log_chans: Used to map interrupt number to logical channel. Points 571 * to log_chans entries. 572 * @lookup_phy_chans: Used to map interrupt number to physical channel. Points 573 * to phy_chans entries. 574 * @plat_data: Pointer to provided platform_data which is the driver 575 * configuration. 576 * @lcpa_regulator: Pointer to hold the regulator for the esram bank for lcla. 577 * @phy_res: Vector containing all physical channels. 578 * @lcla_pool: lcla pool settings and data. 579 * @lcpa_base: The virtual mapped address of LCPA. 580 * @phy_lcpa: The physical address of the LCPA. 581 * @lcpa_size: The size of the LCPA area. 582 * @desc_slab: cache for descriptors. 583 * @reg_val_backup: Here the values of some hardware registers are stored 584 * before the DMA is powered off. They are restored when the power is back on. 585 * @reg_val_backup_v4: Backup of registers that only exits on dma40 v3 and 586 * later 587 * @reg_val_backup_chan: Backup data for standard channel parameter registers. 588 * @regs_interrupt: Scratch space for registers during interrupt. 589 * @gcc_pwr_off_mask: Mask to maintain the channels that can be turned off. 590 * @gen_dmac: the struct for generic registers values to represent u8500/8540 591 * DMA controller 592 */ 593struct d40_base { 594 spinlock_t interrupt_lock; 595 spinlock_t execmd_lock; 596 struct device *dev; 597 void __iomem *virtbase; 598 u8 rev:4; 599 struct clk *clk; 600 int irq; 601 int num_memcpy_chans; 602 int num_phy_chans; 603 int num_log_chans; 604 struct dma_device dma_both; 605 struct dma_device dma_slave; 606 struct dma_device dma_memcpy; 607 struct d40_chan *phy_chans; 608 struct d40_chan *log_chans; 609 struct d40_chan **lookup_log_chans; 610 struct d40_chan **lookup_phy_chans; 611 struct stedma40_platform_data *plat_data; 612 struct regulator *lcpa_regulator; 613 /* Physical half channels */ 614 struct d40_phy_res *phy_res; 615 struct d40_lcla_pool lcla_pool; 616 void *lcpa_base; 617 dma_addr_t phy_lcpa; 618 resource_size_t lcpa_size; 619 struct kmem_cache *desc_slab; 620 u32 reg_val_backup[BACKUP_REGS_SZ]; 621 u32 reg_val_backup_v4[BACKUP_REGS_SZ_MAX]; 622 u32 *reg_val_backup_chan; 623 u32 *regs_interrupt; 624 u16 gcc_pwr_off_mask; 625 struct d40_gen_dmac gen_dmac; 626}; 627 628static struct device *chan2dev(struct d40_chan *d40c) 629{ 630 return &d40c->chan.dev->device; 631} 632 633static bool chan_is_physical(struct d40_chan *chan) 634{ 635 return chan->log_num == D40_PHY_CHAN; 636} 637 638static bool chan_is_logical(struct d40_chan *chan) 639{ 640 return !chan_is_physical(chan); 641} 642 643static void __iomem *chan_base(struct d40_chan *chan) 644{ 645 return chan->base->virtbase + D40_DREG_PCBASE + 646 chan->phy_chan->num * D40_DREG_PCDELTA; 647} 648 649#define d40_err(dev, format, arg...) \ 650 dev_err(dev, "[%s] " format, __func__, ## arg) 651 652#define chan_err(d40c, format, arg...) \ 653 d40_err(chan2dev(d40c), format, ## arg) 654 655static int d40_set_runtime_config_write(struct dma_chan *chan, 656 struct dma_slave_config *config, 657 enum dma_transfer_direction direction); 658 659static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d, 660 int lli_len) 661{ 662 bool is_log = chan_is_logical(d40c); 663 u32 align; 664 void *base; 665 666 if (is_log) 667 align = sizeof(struct d40_log_lli); 668 else 669 align = sizeof(struct d40_phy_lli); 670 671 if (lli_len == 1) { 672 base = d40d->lli_pool.pre_alloc_lli; 673 d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli); 674 d40d->lli_pool.base = NULL; 675 } else { 676 d40d->lli_pool.size = lli_len * 2 * align; 677 678 base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT); 679 d40d->lli_pool.base = base; 680 681 if (d40d->lli_pool.base == NULL) 682 return -ENOMEM; 683 } 684 685 if (is_log) { 686 d40d->lli_log.src = PTR_ALIGN(base, align); 687 d40d->lli_log.dst = d40d->lli_log.src + lli_len; 688 689 d40d->lli_pool.dma_addr = 0; 690 } else { 691 d40d->lli_phy.src = PTR_ALIGN(base, align); 692 d40d->lli_phy.dst = d40d->lli_phy.src + lli_len; 693 694 d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev, 695 d40d->lli_phy.src, 696 d40d->lli_pool.size, 697 DMA_TO_DEVICE); 698 699 if (dma_mapping_error(d40c->base->dev, 700 d40d->lli_pool.dma_addr)) { 701 kfree(d40d->lli_pool.base); 702 d40d->lli_pool.base = NULL; 703 d40d->lli_pool.dma_addr = 0; 704 return -ENOMEM; 705 } 706 } 707 708 return 0; 709} 710 711static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d) 712{ 713 if (d40d->lli_pool.dma_addr) 714 dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr, 715 d40d->lli_pool.size, DMA_TO_DEVICE); 716 717 kfree(d40d->lli_pool.base); 718 d40d->lli_pool.base = NULL; 719 d40d->lli_pool.size = 0; 720 d40d->lli_log.src = NULL; 721 d40d->lli_log.dst = NULL; 722 d40d->lli_phy.src = NULL; 723 d40d->lli_phy.dst = NULL; 724} 725 726static int d40_lcla_alloc_one(struct d40_chan *d40c, 727 struct d40_desc *d40d) 728{ 729 unsigned long flags; 730 int i; 731 int ret = -EINVAL; 732 733 spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags); 734 735 /* 736 * Allocate both src and dst at the same time, therefore the half 737 * start on 1 since 0 can't be used since zero is used as end marker. 738 */ 739 for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) { 740 int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i; 741 742 if (!d40c->base->lcla_pool.alloc_map[idx]) { 743 d40c->base->lcla_pool.alloc_map[idx] = d40d; 744 d40d->lcla_alloc++; 745 ret = i; 746 break; 747 } 748 } 749 750 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags); 751 752 return ret; 753} 754 755static int d40_lcla_free_all(struct d40_chan *d40c, 756 struct d40_desc *d40d) 757{ 758 unsigned long flags; 759 int i; 760 int ret = -EINVAL; 761 762 if (chan_is_physical(d40c)) 763 return 0; 764 765 spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags); 766 767 for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) { 768 int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i; 769 770 if (d40c->base->lcla_pool.alloc_map[idx] == d40d) { 771 d40c->base->lcla_pool.alloc_map[idx] = NULL; 772 d40d->lcla_alloc--; 773 if (d40d->lcla_alloc == 0) { 774 ret = 0; 775 break; 776 } 777 } 778 } 779 780 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags); 781 782 return ret; 783 784} 785 786static void d40_desc_remove(struct d40_desc *d40d) 787{ 788 list_del(&d40d->node); 789} 790 791static struct d40_desc *d40_desc_get(struct d40_chan *d40c) 792{ 793 struct d40_desc *desc = NULL; 794 795 if (!list_empty(&d40c->client)) { 796 struct d40_desc *d; 797 struct d40_desc *_d; 798 799 list_for_each_entry_safe(d, _d, &d40c->client, node) { 800 if (async_tx_test_ack(&d->txd)) { 801 d40_desc_remove(d); 802 desc = d; 803 memset(desc, 0, sizeof(*desc)); 804 break; 805 } 806 } 807 } 808 809 if (!desc) 810 desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT); 811 812 if (desc) 813 INIT_LIST_HEAD(&desc->node); 814 815 return desc; 816} 817 818static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d) 819{ 820 821 d40_pool_lli_free(d40c, d40d); 822 d40_lcla_free_all(d40c, d40d); 823 kmem_cache_free(d40c->base->desc_slab, d40d); 824} 825 826static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc) 827{ 828 list_add_tail(&desc->node, &d40c->active); 829} 830 831static void d40_phy_lli_load(struct d40_chan *chan, struct d40_desc *desc) 832{ 833 struct d40_phy_lli *lli_dst = desc->lli_phy.dst; 834 struct d40_phy_lli *lli_src = desc->lli_phy.src; 835 void __iomem *base = chan_base(chan); 836 837 writel(lli_src->reg_cfg, base + D40_CHAN_REG_SSCFG); 838 writel(lli_src->reg_elt, base + D40_CHAN_REG_SSELT); 839 writel(lli_src->reg_ptr, base + D40_CHAN_REG_SSPTR); 840 writel(lli_src->reg_lnk, base + D40_CHAN_REG_SSLNK); 841 842 writel(lli_dst->reg_cfg, base + D40_CHAN_REG_SDCFG); 843 writel(lli_dst->reg_elt, base + D40_CHAN_REG_SDELT); 844 writel(lli_dst->reg_ptr, base + D40_CHAN_REG_SDPTR); 845 writel(lli_dst->reg_lnk, base + D40_CHAN_REG_SDLNK); 846} 847 848static void d40_desc_done(struct d40_chan *d40c, struct d40_desc *desc) 849{ 850 list_add_tail(&desc->node, &d40c->done); 851} 852 853static void d40_log_lli_to_lcxa(struct d40_chan *chan, struct d40_desc *desc) 854{ 855 struct d40_lcla_pool *pool = &chan->base->lcla_pool; 856 struct d40_log_lli_bidir *lli = &desc->lli_log; 857 int lli_current = desc->lli_current; 858 int lli_len = desc->lli_len; 859 bool cyclic = desc->cyclic; 860 int curr_lcla = -EINVAL; 861 int first_lcla = 0; 862 bool use_esram_lcla = chan->base->plat_data->use_esram_lcla; 863 bool linkback; 864 865 /* 866 * We may have partially running cyclic transfers, in case we did't get 867 * enough LCLA entries. 868 */ 869 linkback = cyclic && lli_current == 0; 870 871 /* 872 * For linkback, we need one LCLA even with only one link, because we 873 * can't link back to the one in LCPA space 874 */ 875 if (linkback || (lli_len - lli_current > 1)) { 876 /* 877 * If the channel is expected to use only soft_lli don't 878 * allocate a lcla. This is to avoid a HW issue that exists 879 * in some controller during a peripheral to memory transfer 880 * that uses linked lists. 881 */ 882 if (!(chan->phy_chan->use_soft_lli && 883 chan->dma_cfg.dir == DMA_DEV_TO_MEM)) 884 curr_lcla = d40_lcla_alloc_one(chan, desc); 885 886 first_lcla = curr_lcla; 887 } 888 889 /* 890 * For linkback, we normally load the LCPA in the loop since we need to 891 * link it to the second LCLA and not the first. However, if we 892 * couldn't even get a first LCLA, then we have to run in LCPA and 893 * reload manually. 894 */ 895 if (!linkback || curr_lcla == -EINVAL) { 896 unsigned int flags = 0; 897 898 if (curr_lcla == -EINVAL) 899 flags |= LLI_TERM_INT; 900 901 d40_log_lli_lcpa_write(chan->lcpa, 902 &lli->dst[lli_current], 903 &lli->src[lli_current], 904 curr_lcla, 905 flags); 906 lli_current++; 907 } 908 909 if (curr_lcla < 0) 910 goto set_current; 911 912 for (; lli_current < lli_len; lli_current++) { 913 unsigned int lcla_offset = chan->phy_chan->num * 1024 + 914 8 * curr_lcla * 2; 915 struct d40_log_lli *lcla = pool->base + lcla_offset; 916 unsigned int flags = 0; 917 int next_lcla; 918 919 if (lli_current + 1 < lli_len) 920 next_lcla = d40_lcla_alloc_one(chan, desc); 921 else 922 next_lcla = linkback ? first_lcla : -EINVAL; 923 924 if (cyclic || next_lcla == -EINVAL) 925 flags |= LLI_TERM_INT; 926 927 if (linkback && curr_lcla == first_lcla) { 928 /* First link goes in both LCPA and LCLA */ 929 d40_log_lli_lcpa_write(chan->lcpa, 930 &lli->dst[lli_current], 931 &lli->src[lli_current], 932 next_lcla, flags); 933 } 934 935 /* 936 * One unused LCLA in the cyclic case if the very first 937 * next_lcla fails... 938 */ 939 d40_log_lli_lcla_write(lcla, 940 &lli->dst[lli_current], 941 &lli->src[lli_current], 942 next_lcla, flags); 943 944 /* 945 * Cache maintenance is not needed if lcla is 946 * mapped in esram 947 */ 948 if (!use_esram_lcla) { 949 dma_sync_single_range_for_device(chan->base->dev, 950 pool->dma_addr, lcla_offset, 951 2 * sizeof(struct d40_log_lli), 952 DMA_TO_DEVICE); 953 } 954 curr_lcla = next_lcla; 955 956 if (curr_lcla == -EINVAL || curr_lcla == first_lcla) { 957 lli_current++; 958 break; 959 } 960 } 961 set_current: 962 desc->lli_current = lli_current; 963} 964 965static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d) 966{ 967 if (chan_is_physical(d40c)) { 968 d40_phy_lli_load(d40c, d40d); 969 d40d->lli_current = d40d->lli_len; 970 } else 971 d40_log_lli_to_lcxa(d40c, d40d); 972} 973 974static struct d40_desc *d40_first_active_get(struct d40_chan *d40c) 975{ 976 return list_first_entry_or_null(&d40c->active, struct d40_desc, node); 977} 978 979/* remove desc from current queue and add it to the pending_queue */ 980static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc) 981{ 982 d40_desc_remove(desc); 983 desc->is_in_client_list = false; 984 list_add_tail(&desc->node, &d40c->pending_queue); 985} 986 987static struct d40_desc *d40_first_pending(struct d40_chan *d40c) 988{ 989 return list_first_entry_or_null(&d40c->pending_queue, struct d40_desc, 990 node); 991} 992 993static struct d40_desc *d40_first_queued(struct d40_chan *d40c) 994{ 995 return list_first_entry_or_null(&d40c->queue, struct d40_desc, node); 996} 997 998static struct d40_desc *d40_first_done(struct d40_chan *d40c) 999{ 1000 return list_first_entry_or_null(&d40c->done, struct d40_desc, node); 1001} 1002 1003static int d40_psize_2_burst_size(bool is_log, int psize) 1004{ 1005 if (is_log) { 1006 if (psize == STEDMA40_PSIZE_LOG_1) 1007 return 1; 1008 } else { 1009 if (psize == STEDMA40_PSIZE_PHY_1) 1010 return 1; 1011 } 1012 1013 return 2 << psize; 1014} 1015 1016/* 1017 * The dma only supports transmitting packages up to 1018 * STEDMA40_MAX_SEG_SIZE * data_width, where data_width is stored in Bytes. 1019 * 1020 * Calculate the total number of dma elements required to send the entire sg list. 1021 */ 1022static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2) 1023{ 1024 int dmalen; 1025 u32 max_w = max(data_width1, data_width2); 1026 u32 min_w = min(data_width1, data_width2); 1027 u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE * min_w, max_w); 1028 1029 if (seg_max > STEDMA40_MAX_SEG_SIZE) 1030 seg_max -= max_w; 1031 1032 if (!IS_ALIGNED(size, max_w)) 1033 return -EINVAL; 1034 1035 if (size <= seg_max) 1036 dmalen = 1; 1037 else { 1038 dmalen = size / seg_max; 1039 if (dmalen * seg_max < size) 1040 dmalen++; 1041 } 1042 return dmalen; 1043} 1044 1045static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len, 1046 u32 data_width1, u32 data_width2) 1047{ 1048 struct scatterlist *sg; 1049 int i; 1050 int len = 0; 1051 int ret; 1052 1053 for_each_sg(sgl, sg, sg_len, i) { 1054 ret = d40_size_2_dmalen(sg_dma_len(sg), 1055 data_width1, data_width2); 1056 if (ret < 0) 1057 return ret; 1058 len += ret; 1059 } 1060 return len; 1061} 1062 1063static int __d40_execute_command_phy(struct d40_chan *d40c, 1064 enum d40_command command) 1065{ 1066 u32 status; 1067 int i; 1068 void __iomem *active_reg; 1069 int ret = 0; 1070 unsigned long flags; 1071 u32 wmask; 1072 1073 if (command == D40_DMA_STOP) { 1074 ret = __d40_execute_command_phy(d40c, D40_DMA_SUSPEND_REQ); 1075 if (ret) 1076 return ret; 1077 } 1078 1079 spin_lock_irqsave(&d40c->base->execmd_lock, flags); 1080 1081 if (d40c->phy_chan->num % 2 == 0) 1082 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE; 1083 else 1084 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO; 1085 1086 if (command == D40_DMA_SUSPEND_REQ) { 1087 status = (readl(active_reg) & 1088 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >> 1089 D40_CHAN_POS(d40c->phy_chan->num); 1090 1091 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP) 1092 goto unlock; 1093 } 1094 1095 wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num)); 1096 writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)), 1097 active_reg); 1098 1099 if (command == D40_DMA_SUSPEND_REQ) { 1100 1101 for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) { 1102 status = (readl(active_reg) & 1103 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >> 1104 D40_CHAN_POS(d40c->phy_chan->num); 1105 1106 cpu_relax(); 1107 /* 1108 * Reduce the number of bus accesses while 1109 * waiting for the DMA to suspend. 1110 */ 1111 udelay(3); 1112 1113 if (status == D40_DMA_STOP || 1114 status == D40_DMA_SUSPENDED) 1115 break; 1116 } 1117 1118 if (i == D40_SUSPEND_MAX_IT) { 1119 chan_err(d40c, 1120 "unable to suspend the chl %d (log: %d) status %x\n", 1121 d40c->phy_chan->num, d40c->log_num, 1122 status); 1123 dump_stack(); 1124 ret = -EBUSY; 1125 } 1126 1127 } 1128 unlock: 1129 spin_unlock_irqrestore(&d40c->base->execmd_lock, flags); 1130 return ret; 1131} 1132 1133static void d40_term_all(struct d40_chan *d40c) 1134{ 1135 struct d40_desc *d40d; 1136 struct d40_desc *_d; 1137 1138 /* Release completed descriptors */ 1139 while ((d40d = d40_first_done(d40c))) { 1140 d40_desc_remove(d40d); 1141 d40_desc_free(d40c, d40d); 1142 } 1143 1144 /* Release active descriptors */ 1145 while ((d40d = d40_first_active_get(d40c))) { 1146 d40_desc_remove(d40d); 1147 d40_desc_free(d40c, d40d); 1148 } 1149 1150 /* Release queued descriptors waiting for transfer */ 1151 while ((d40d = d40_first_queued(d40c))) { 1152 d40_desc_remove(d40d); 1153 d40_desc_free(d40c, d40d); 1154 } 1155 1156 /* Release pending descriptors */ 1157 while ((d40d = d40_first_pending(d40c))) { 1158 d40_desc_remove(d40d); 1159 d40_desc_free(d40c, d40d); 1160 } 1161 1162 /* Release client owned descriptors */ 1163 if (!list_empty(&d40c->client)) 1164 list_for_each_entry_safe(d40d, _d, &d40c->client, node) { 1165 d40_desc_remove(d40d); 1166 d40_desc_free(d40c, d40d); 1167 } 1168 1169 /* Release descriptors in prepare queue */ 1170 if (!list_empty(&d40c->prepare_queue)) 1171 list_for_each_entry_safe(d40d, _d, 1172 &d40c->prepare_queue, node) { 1173 d40_desc_remove(d40d); 1174 d40_desc_free(d40c, d40d); 1175 } 1176 1177 d40c->pending_tx = 0; 1178} 1179 1180static void __d40_config_set_event(struct d40_chan *d40c, 1181 enum d40_events event_type, u32 event, 1182 int reg) 1183{ 1184 void __iomem *addr = chan_base(d40c) + reg; 1185 int tries; 1186 u32 status; 1187 1188 switch (event_type) { 1189 1190 case D40_DEACTIVATE_EVENTLINE: 1191 1192 writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event)) 1193 | ~D40_EVENTLINE_MASK(event), addr); 1194 break; 1195 1196 case D40_SUSPEND_REQ_EVENTLINE: 1197 status = (readl(addr) & D40_EVENTLINE_MASK(event)) >> 1198 D40_EVENTLINE_POS(event); 1199 1200 if (status == D40_DEACTIVATE_EVENTLINE || 1201 status == D40_SUSPEND_REQ_EVENTLINE) 1202 break; 1203 1204 writel((D40_SUSPEND_REQ_EVENTLINE << D40_EVENTLINE_POS(event)) 1205 | ~D40_EVENTLINE_MASK(event), addr); 1206 1207 for (tries = 0 ; tries < D40_SUSPEND_MAX_IT; tries++) { 1208 1209 status = (readl(addr) & D40_EVENTLINE_MASK(event)) >> 1210 D40_EVENTLINE_POS(event); 1211 1212 cpu_relax(); 1213 /* 1214 * Reduce the number of bus accesses while 1215 * waiting for the DMA to suspend. 1216 */ 1217 udelay(3); 1218 1219 if (status == D40_DEACTIVATE_EVENTLINE) 1220 break; 1221 } 1222 1223 if (tries == D40_SUSPEND_MAX_IT) { 1224 chan_err(d40c, 1225 "unable to stop the event_line chl %d (log: %d)" 1226 "status %x\n", d40c->phy_chan->num, 1227 d40c->log_num, status); 1228 } 1229 break; 1230 1231 case D40_ACTIVATE_EVENTLINE: 1232 /* 1233 * The hardware sometimes doesn't register the enable when src and dst 1234 * event lines are active on the same logical channel. Retry to ensure 1235 * it does. Usually only one retry is sufficient. 1236 */ 1237 tries = 100; 1238 while (--tries) { 1239 writel((D40_ACTIVATE_EVENTLINE << 1240 D40_EVENTLINE_POS(event)) | 1241 ~D40_EVENTLINE_MASK(event), addr); 1242 1243 if (readl(addr) & D40_EVENTLINE_MASK(event)) 1244 break; 1245 } 1246 1247 if (tries != 99) 1248 dev_dbg(chan2dev(d40c), 1249 "[%s] workaround enable S%cLNK (%d tries)\n", 1250 __func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D', 1251 100 - tries); 1252 1253 WARN_ON(!tries); 1254 break; 1255 1256 case D40_ROUND_EVENTLINE: 1257 BUG(); 1258 break; 1259 1260 } 1261} 1262 1263static void d40_config_set_event(struct d40_chan *d40c, 1264 enum d40_events event_type) 1265{ 1266 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type); 1267 1268 /* Enable event line connected to device (or memcpy) */ 1269 if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) || 1270 (d40c->dma_cfg.dir == DMA_DEV_TO_DEV)) 1271 __d40_config_set_event(d40c, event_type, event, 1272 D40_CHAN_REG_SSLNK); 1273 1274 if (d40c->dma_cfg.dir != DMA_DEV_TO_MEM) 1275 __d40_config_set_event(d40c, event_type, event, 1276 D40_CHAN_REG_SDLNK); 1277} 1278 1279static u32 d40_chan_has_events(struct d40_chan *d40c) 1280{ 1281 void __iomem *chanbase = chan_base(d40c); 1282 u32 val; 1283 1284 val = readl(chanbase + D40_CHAN_REG_SSLNK); 1285 val |= readl(chanbase + D40_CHAN_REG_SDLNK); 1286 1287 return val; 1288} 1289 1290static int 1291__d40_execute_command_log(struct d40_chan *d40c, enum d40_command command) 1292{ 1293 unsigned long flags; 1294 int ret = 0; 1295 u32 active_status; 1296 void __iomem *active_reg; 1297 1298 if (d40c->phy_chan->num % 2 == 0) 1299 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE; 1300 else 1301 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO; 1302 1303 1304 spin_lock_irqsave(&d40c->phy_chan->lock, flags); 1305 1306 switch (command) { 1307 case D40_DMA_STOP: 1308 case D40_DMA_SUSPEND_REQ: 1309 1310 active_status = (readl(active_reg) & 1311 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >> 1312 D40_CHAN_POS(d40c->phy_chan->num); 1313 1314 if (active_status == D40_DMA_RUN) 1315 d40_config_set_event(d40c, D40_SUSPEND_REQ_EVENTLINE); 1316 else 1317 d40_config_set_event(d40c, D40_DEACTIVATE_EVENTLINE); 1318 1319 if (!d40_chan_has_events(d40c) && (command == D40_DMA_STOP)) 1320 ret = __d40_execute_command_phy(d40c, command); 1321 1322 break; 1323 1324 case D40_DMA_RUN: 1325 1326 d40_config_set_event(d40c, D40_ACTIVATE_EVENTLINE); 1327 ret = __d40_execute_command_phy(d40c, command); 1328 break; 1329 1330 case D40_DMA_SUSPENDED: 1331 BUG(); 1332 break; 1333 } 1334 1335 spin_unlock_irqrestore(&d40c->phy_chan->lock, flags); 1336 return ret; 1337} 1338 1339static int d40_channel_execute_command(struct d40_chan *d40c, 1340 enum d40_command command) 1341{ 1342 if (chan_is_logical(d40c)) 1343 return __d40_execute_command_log(d40c, command); 1344 else 1345 return __d40_execute_command_phy(d40c, command); 1346} 1347 1348static u32 d40_get_prmo(struct d40_chan *d40c) 1349{ 1350 static const unsigned int phy_map[] = { 1351 [STEDMA40_PCHAN_BASIC_MODE] 1352 = D40_DREG_PRMO_PCHAN_BASIC, 1353 [STEDMA40_PCHAN_MODULO_MODE] 1354 = D40_DREG_PRMO_PCHAN_MODULO, 1355 [STEDMA40_PCHAN_DOUBLE_DST_MODE] 1356 = D40_DREG_PRMO_PCHAN_DOUBLE_DST, 1357 }; 1358 static const unsigned int log_map[] = { 1359 [STEDMA40_LCHAN_SRC_PHY_DST_LOG] 1360 = D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG, 1361 [STEDMA40_LCHAN_SRC_LOG_DST_PHY] 1362 = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY, 1363 [STEDMA40_LCHAN_SRC_LOG_DST_LOG] 1364 = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG, 1365 }; 1366 1367 if (chan_is_physical(d40c)) 1368 return phy_map[d40c->dma_cfg.mode_opt]; 1369 else 1370 return log_map[d40c->dma_cfg.mode_opt]; 1371} 1372 1373static void d40_config_write(struct d40_chan *d40c) 1374{ 1375 u32 addr_base; 1376 u32 var; 1377 1378 /* Odd addresses are even addresses + 4 */ 1379 addr_base = (d40c->phy_chan->num % 2) * 4; 1380 /* Setup channel mode to logical or physical */ 1381 var = ((u32)(chan_is_logical(d40c)) + 1) << 1382 D40_CHAN_POS(d40c->phy_chan->num); 1383 writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base); 1384 1385 /* Setup operational mode option register */ 1386 var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num); 1387 1388 writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base); 1389 1390 if (chan_is_logical(d40c)) { 1391 int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS) 1392 & D40_SREG_ELEM_LOG_LIDX_MASK; 1393 void __iomem *chanbase = chan_base(d40c); 1394 1395 /* Set default config for CFG reg */ 1396 writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG); 1397 writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG); 1398 1399 /* Set LIDX for lcla */ 1400 writel(lidx, chanbase + D40_CHAN_REG_SSELT); 1401 writel(lidx, chanbase + D40_CHAN_REG_SDELT); 1402 1403 /* Clear LNK which will be used by d40_chan_has_events() */ 1404 writel(0, chanbase + D40_CHAN_REG_SSLNK); 1405 writel(0, chanbase + D40_CHAN_REG_SDLNK); 1406 } 1407} 1408 1409static u32 d40_residue(struct d40_chan *d40c) 1410{ 1411 u32 num_elt; 1412 1413 if (chan_is_logical(d40c)) 1414 num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK) 1415 >> D40_MEM_LCSP2_ECNT_POS; 1416 else { 1417 u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT); 1418 num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK) 1419 >> D40_SREG_ELEM_PHY_ECNT_POS; 1420 } 1421 1422 return num_elt * d40c->dma_cfg.dst_info.data_width; 1423} 1424 1425static bool d40_tx_is_linked(struct d40_chan *d40c) 1426{ 1427 bool is_link; 1428 1429 if (chan_is_logical(d40c)) 1430 is_link = readl(&d40c->lcpa->lcsp3) & D40_MEM_LCSP3_DLOS_MASK; 1431 else 1432 is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK) 1433 & D40_SREG_LNK_PHYS_LNK_MASK; 1434 1435 return is_link; 1436} 1437 1438static int d40_pause(struct dma_chan *chan) 1439{ 1440 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan); 1441 int res = 0; 1442 unsigned long flags; 1443 1444 if (d40c->phy_chan == NULL) { 1445 chan_err(d40c, "Channel is not allocated!\n"); 1446 return -EINVAL; 1447 } 1448 1449 if (!d40c->busy) 1450 return 0; 1451 1452 spin_lock_irqsave(&d40c->lock, flags); 1453 pm_runtime_get_sync(d40c->base->dev); 1454 1455 res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ); 1456 1457 pm_runtime_mark_last_busy(d40c->base->dev); 1458 pm_runtime_put_autosuspend(d40c->base->dev); 1459 spin_unlock_irqrestore(&d40c->lock, flags); 1460 return res; 1461} 1462 1463static int d40_resume(struct dma_chan *chan) 1464{ 1465 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan); 1466 int res = 0; 1467 unsigned long flags; 1468 1469 if (d40c->phy_chan == NULL) { 1470 chan_err(d40c, "Channel is not allocated!\n"); 1471 return -EINVAL; 1472 } 1473 1474 if (!d40c->busy) 1475 return 0; 1476 1477 spin_lock_irqsave(&d40c->lock, flags); 1478 pm_runtime_get_sync(d40c->base->dev); 1479 1480 /* If bytes left to transfer or linked tx resume job */ 1481 if (d40_residue(d40c) || d40_tx_is_linked(d40c)) 1482 res = d40_channel_execute_command(d40c, D40_DMA_RUN); 1483 1484 pm_runtime_mark_last_busy(d40c->base->dev); 1485 pm_runtime_put_autosuspend(d40c->base->dev); 1486 spin_unlock_irqrestore(&d40c->lock, flags); 1487 return res; 1488} 1489 1490static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx) 1491{ 1492 struct d40_chan *d40c = container_of(tx->chan, 1493 struct d40_chan, 1494 chan); 1495 struct d40_desc *d40d = container_of(tx, struct d40_desc, txd); 1496 unsigned long flags; 1497 dma_cookie_t cookie; 1498 1499 spin_lock_irqsave(&d40c->lock, flags); 1500 cookie = dma_cookie_assign(tx); 1501 d40_desc_queue(d40c, d40d); 1502 spin_unlock_irqrestore(&d40c->lock, flags); 1503 1504 return cookie; 1505} 1506 1507static int d40_start(struct d40_chan *d40c) 1508{ 1509 return d40_channel_execute_command(d40c, D40_DMA_RUN); 1510} 1511 1512static struct d40_desc *d40_queue_start(struct d40_chan *d40c) 1513{ 1514 struct d40_desc *d40d; 1515 int err; 1516 1517 /* Start queued jobs, if any */ 1518 d40d = d40_first_queued(d40c); 1519 1520 if (d40d != NULL) { 1521 if (!d40c->busy) { 1522 d40c->busy = true; 1523 pm_runtime_get_sync(d40c->base->dev); 1524 } 1525 1526 /* Remove from queue */ 1527 d40_desc_remove(d40d); 1528 1529 /* Add to active queue */ 1530 d40_desc_submit(d40c, d40d); 1531 1532 /* Initiate DMA job */ 1533 d40_desc_load(d40c, d40d); 1534 1535 /* Start dma job */ 1536 err = d40_start(d40c); 1537 1538 if (err) 1539 return NULL; 1540 } 1541 1542 return d40d; 1543} 1544 1545/* called from interrupt context */ 1546static void dma_tc_handle(struct d40_chan *d40c) 1547{ 1548 struct d40_desc *d40d; 1549 1550 /* Get first active entry from list */ 1551 d40d = d40_first_active_get(d40c); 1552 1553 if (d40d == NULL) 1554 return; 1555 1556 if (d40d->cyclic) { 1557 /* 1558 * If this was a paritially loaded list, we need to reloaded 1559 * it, and only when the list is completed. We need to check 1560 * for done because the interrupt will hit for every link, and 1561 * not just the last one. 1562 */ 1563 if (d40d->lli_current < d40d->lli_len 1564 && !d40_tx_is_linked(d40c) 1565 && !d40_residue(d40c)) { 1566 d40_lcla_free_all(d40c, d40d); 1567 d40_desc_load(d40c, d40d); 1568 (void) d40_start(d40c); 1569 1570 if (d40d->lli_current == d40d->lli_len) 1571 d40d->lli_current = 0; 1572 } 1573 } else { 1574 d40_lcla_free_all(d40c, d40d); 1575 1576 if (d40d->lli_current < d40d->lli_len) { 1577 d40_desc_load(d40c, d40d); 1578 /* Start dma job */ 1579 (void) d40_start(d40c); 1580 return; 1581 } 1582 1583 if (d40_queue_start(d40c) == NULL) { 1584 d40c->busy = false; 1585 1586 pm_runtime_mark_last_busy(d40c->base->dev); 1587 pm_runtime_put_autosuspend(d40c->base->dev); 1588 } 1589 1590 d40_desc_remove(d40d); 1591 d40_desc_done(d40c, d40d); 1592 } 1593 1594 d40c->pending_tx++; 1595 tasklet_schedule(&d40c->tasklet); 1596 1597} 1598 1599static void dma_tasklet(struct tasklet_struct *t) 1600{ 1601 struct d40_chan *d40c = from_tasklet(d40c, t, tasklet); 1602 struct d40_desc *d40d; 1603 unsigned long flags; 1604 bool callback_active; 1605 struct dmaengine_desc_callback cb; 1606 1607 spin_lock_irqsave(&d40c->lock, flags); 1608 1609 /* Get first entry from the done list */ 1610 d40d = d40_first_done(d40c); 1611 if (d40d == NULL) { 1612 /* Check if we have reached here for cyclic job */ 1613 d40d = d40_first_active_get(d40c); 1614 if (d40d == NULL || !d40d->cyclic) 1615 goto check_pending_tx; 1616 } 1617 1618 if (!d40d->cyclic) 1619 dma_cookie_complete(&d40d->txd); 1620 1621 /* 1622 * If terminating a channel pending_tx is set to zero. 1623 * This prevents any finished active jobs to return to the client. 1624 */ 1625 if (d40c->pending_tx == 0) { 1626 spin_unlock_irqrestore(&d40c->lock, flags); 1627 return; 1628 } 1629 1630 /* Callback to client */ 1631 callback_active = !!(d40d->txd.flags & DMA_PREP_INTERRUPT); 1632 dmaengine_desc_get_callback(&d40d->txd, &cb); 1633 1634 if (!d40d->cyclic) { 1635 if (async_tx_test_ack(&d40d->txd)) { 1636 d40_desc_remove(d40d); 1637 d40_desc_free(d40c, d40d); 1638 } else if (!d40d->is_in_client_list) { 1639 d40_desc_remove(d40d); 1640 d40_lcla_free_all(d40c, d40d); 1641 list_add_tail(&d40d->node, &d40c->client); 1642 d40d->is_in_client_list = true; 1643 } 1644 } 1645 1646 d40c->pending_tx--; 1647 1648 if (d40c->pending_tx) 1649 tasklet_schedule(&d40c->tasklet); 1650 1651 spin_unlock_irqrestore(&d40c->lock, flags); 1652 1653 if (callback_active) 1654 dmaengine_desc_callback_invoke(&cb, NULL); 1655 1656 return; 1657 check_pending_tx: 1658 /* Rescue manouver if receiving double interrupts */ 1659 if (d40c->pending_tx > 0) 1660 d40c->pending_tx--; 1661 spin_unlock_irqrestore(&d40c->lock, flags); 1662} 1663 1664static irqreturn_t d40_handle_interrupt(int irq, void *data) 1665{ 1666 int i; 1667 u32 idx; 1668 u32 row; 1669 long chan = -1; 1670 struct d40_chan *d40c; 1671 struct d40_base *base = data; 1672 u32 *regs = base->regs_interrupt; 1673 struct d40_interrupt_lookup *il = base->gen_dmac.il; 1674 u32 il_size = base->gen_dmac.il_size; 1675 1676 spin_lock(&base->interrupt_lock); 1677 1678 /* Read interrupt status of both logical and physical channels */ 1679 for (i = 0; i < il_size; i++) 1680 regs[i] = readl(base->virtbase + il[i].src); 1681 1682 for (;;) { 1683 1684 chan = find_next_bit((unsigned long *)regs, 1685 BITS_PER_LONG * il_size, chan + 1); 1686 1687 /* No more set bits found? */ 1688 if (chan == BITS_PER_LONG * il_size) 1689 break; 1690 1691 row = chan / BITS_PER_LONG; 1692 idx = chan & (BITS_PER_LONG - 1); 1693 1694 if (il[row].offset == D40_PHY_CHAN) 1695 d40c = base->lookup_phy_chans[idx]; 1696 else 1697 d40c = base->lookup_log_chans[il[row].offset + idx]; 1698 1699 if (!d40c) { 1700 /* 1701 * No error because this can happen if something else 1702 * in the system is using the channel. 1703 */ 1704 continue; 1705 } 1706 1707 /* ACK interrupt */ 1708 writel(BIT(idx), base->virtbase + il[row].clr); 1709 1710 spin_lock(&d40c->lock); 1711 1712 if (!il[row].is_error) 1713 dma_tc_handle(d40c); 1714 else 1715 d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n", 1716 chan, il[row].offset, idx); 1717 1718 spin_unlock(&d40c->lock); 1719 } 1720 1721 spin_unlock(&base->interrupt_lock); 1722 1723 return IRQ_HANDLED; 1724} 1725 1726static int d40_validate_conf(struct d40_chan *d40c, 1727 struct stedma40_chan_cfg *conf) 1728{ 1729 int res = 0; 1730 bool is_log = conf->mode == STEDMA40_MODE_LOGICAL; 1731 1732 if (!conf->dir) { 1733 chan_err(d40c, "Invalid direction.\n"); 1734 res = -EINVAL; 1735 } 1736 1737 if ((is_log && conf->dev_type > d40c->base->num_log_chans) || 1738 (!is_log && conf->dev_type > d40c->base->num_phy_chans) || 1739 (conf->dev_type < 0)) { 1740 chan_err(d40c, "Invalid device type (%d)\n", conf->dev_type); 1741 res = -EINVAL; 1742 } 1743 1744 if (conf->dir == DMA_DEV_TO_DEV) { 1745 /* 1746 * DMAC HW supports it. Will be added to this driver, 1747 * in case any dma client requires it. 1748 */ 1749 chan_err(d40c, "periph to periph not supported\n"); 1750 res = -EINVAL; 1751 } 1752 1753 if (d40_psize_2_burst_size(is_log, conf->src_info.psize) * 1754 conf->src_info.data_width != 1755 d40_psize_2_burst_size(is_log, conf->dst_info.psize) * 1756 conf->dst_info.data_width) { 1757 /* 1758 * The DMAC hardware only supports 1759 * src (burst x width) == dst (burst x width) 1760 */ 1761 1762 chan_err(d40c, "src (burst x width) != dst (burst x width)\n"); 1763 res = -EINVAL; 1764 } 1765 1766 return res; 1767} 1768 1769static bool d40_alloc_mask_set(struct d40_phy_res *phy, 1770 bool is_src, int log_event_line, bool is_log, 1771 bool *first_user) 1772{ 1773 unsigned long flags; 1774 spin_lock_irqsave(&phy->lock, flags); 1775 1776 *first_user = ((phy->allocated_src | phy->allocated_dst) 1777 == D40_ALLOC_FREE); 1778 1779 if (!is_log) { 1780 /* Physical interrupts are masked per physical full channel */ 1781 if (phy->allocated_src == D40_ALLOC_FREE && 1782 phy->allocated_dst == D40_ALLOC_FREE) { 1783 phy->allocated_dst = D40_ALLOC_PHY; 1784 phy->allocated_src = D40_ALLOC_PHY; 1785 goto found_unlock; 1786 } else 1787 goto not_found_unlock; 1788 } 1789 1790 /* Logical channel */ 1791 if (is_src) { 1792 if (phy->allocated_src == D40_ALLOC_PHY) 1793 goto not_found_unlock; 1794 1795 if (phy->allocated_src == D40_ALLOC_FREE) 1796 phy->allocated_src = D40_ALLOC_LOG_FREE; 1797 1798 if (!(phy->allocated_src & BIT(log_event_line))) { 1799 phy->allocated_src |= BIT(log_event_line); 1800 goto found_unlock; 1801 } else 1802 goto not_found_unlock; 1803 } else { 1804 if (phy->allocated_dst == D40_ALLOC_PHY) 1805 goto not_found_unlock; 1806 1807 if (phy->allocated_dst == D40_ALLOC_FREE) 1808 phy->allocated_dst = D40_ALLOC_LOG_FREE; 1809 1810 if (!(phy->allocated_dst & BIT(log_event_line))) { 1811 phy->allocated_dst |= BIT(log_event_line); 1812 goto found_unlock; 1813 } 1814 } 1815 not_found_unlock: 1816 spin_unlock_irqrestore(&phy->lock, flags); 1817 return false; 1818 found_unlock: 1819 spin_unlock_irqrestore(&phy->lock, flags); 1820 return true; 1821} 1822 1823static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src, 1824 int log_event_line) 1825{ 1826 unsigned long flags; 1827 bool is_free = false; 1828 1829 spin_lock_irqsave(&phy->lock, flags); 1830 if (!log_event_line) { 1831 phy->allocated_dst = D40_ALLOC_FREE; 1832 phy->allocated_src = D40_ALLOC_FREE; 1833 is_free = true; 1834 goto unlock; 1835 } 1836 1837 /* Logical channel */ 1838 if (is_src) { 1839 phy->allocated_src &= ~BIT(log_event_line); 1840 if (phy->allocated_src == D40_ALLOC_LOG_FREE) 1841 phy->allocated_src = D40_ALLOC_FREE; 1842 } else { 1843 phy->allocated_dst &= ~BIT(log_event_line); 1844 if (phy->allocated_dst == D40_ALLOC_LOG_FREE) 1845 phy->allocated_dst = D40_ALLOC_FREE; 1846 } 1847 1848 is_free = ((phy->allocated_src | phy->allocated_dst) == 1849 D40_ALLOC_FREE); 1850 unlock: 1851 spin_unlock_irqrestore(&phy->lock, flags); 1852 1853 return is_free; 1854} 1855 1856static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user) 1857{ 1858 int dev_type = d40c->dma_cfg.dev_type; 1859 int event_group; 1860 int event_line; 1861 struct d40_phy_res *phys; 1862 int i; 1863 int j; 1864 int log_num; 1865 int num_phy_chans; 1866 bool is_src; 1867 bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL; 1868 1869 phys = d40c->base->phy_res; 1870 num_phy_chans = d40c->base->num_phy_chans; 1871 1872 if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) { 1873 log_num = 2 * dev_type; 1874 is_src = true; 1875 } else if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV || 1876 d40c->dma_cfg.dir == DMA_MEM_TO_MEM) { 1877 /* dst event lines are used for logical memcpy */ 1878 log_num = 2 * dev_type + 1; 1879 is_src = false; 1880 } else 1881 return -EINVAL; 1882 1883 event_group = D40_TYPE_TO_GROUP(dev_type); 1884 event_line = D40_TYPE_TO_EVENT(dev_type); 1885 1886 if (!is_log) { 1887 if (d40c->dma_cfg.dir == DMA_MEM_TO_MEM) { 1888 /* Find physical half channel */ 1889 if (d40c->dma_cfg.use_fixed_channel) { 1890 i = d40c->dma_cfg.phy_channel; 1891 if (d40_alloc_mask_set(&phys[i], is_src, 1892 0, is_log, 1893 first_phy_user)) 1894 goto found_phy; 1895 } else { 1896 for (i = 0; i < num_phy_chans; i++) { 1897 if (d40_alloc_mask_set(&phys[i], is_src, 1898 0, is_log, 1899 first_phy_user)) 1900 goto found_phy; 1901 } 1902 } 1903 } else 1904 for (j = 0; j < d40c->base->num_phy_chans; j += 8) { 1905 int phy_num = j + event_group * 2; 1906 for (i = phy_num; i < phy_num + 2; i++) { 1907 if (d40_alloc_mask_set(&phys[i], 1908 is_src, 1909 0, 1910 is_log, 1911 first_phy_user)) 1912 goto found_phy; 1913 } 1914 } 1915 return -EINVAL; 1916found_phy: 1917 d40c->phy_chan = &phys[i]; 1918 d40c->log_num = D40_PHY_CHAN; 1919 goto out; 1920 } 1921 if (dev_type == -1) 1922 return -EINVAL; 1923 1924 /* Find logical channel */ 1925 for (j = 0; j < d40c->base->num_phy_chans; j += 8) { 1926 int phy_num = j + event_group * 2; 1927 1928 if (d40c->dma_cfg.use_fixed_channel) { 1929 i = d40c->dma_cfg.phy_channel; 1930 1931 if ((i != phy_num) && (i != phy_num + 1)) { 1932 dev_err(chan2dev(d40c), 1933 "invalid fixed phy channel %d\n", i); 1934 return -EINVAL; 1935 } 1936 1937 if (d40_alloc_mask_set(&phys[i], is_src, event_line, 1938 is_log, first_phy_user)) 1939 goto found_log; 1940 1941 dev_err(chan2dev(d40c), 1942 "could not allocate fixed phy channel %d\n", i); 1943 return -EINVAL; 1944 } 1945 1946 /* 1947 * Spread logical channels across all available physical rather 1948 * than pack every logical channel at the first available phy 1949 * channels. 1950 */ 1951 if (is_src) { 1952 for (i = phy_num; i < phy_num + 2; i++) { 1953 if (d40_alloc_mask_set(&phys[i], is_src, 1954 event_line, is_log, 1955 first_phy_user)) 1956 goto found_log; 1957 } 1958 } else { 1959 for (i = phy_num + 1; i >= phy_num; i--) { 1960 if (d40_alloc_mask_set(&phys[i], is_src, 1961 event_line, is_log, 1962 first_phy_user)) 1963 goto found_log; 1964 } 1965 } 1966 } 1967 return -EINVAL; 1968 1969found_log: 1970 d40c->phy_chan = &phys[i]; 1971 d40c->log_num = log_num; 1972out: 1973 1974 if (is_log) 1975 d40c->base->lookup_log_chans[d40c->log_num] = d40c; 1976 else 1977 d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c; 1978 1979 return 0; 1980 1981} 1982 1983static int d40_config_memcpy(struct d40_chan *d40c) 1984{ 1985 dma_cap_mask_t cap = d40c->chan.device->cap_mask; 1986 1987 if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) { 1988 d40c->dma_cfg = dma40_memcpy_conf_log; 1989 d40c->dma_cfg.dev_type = dma40_memcpy_channels[d40c->chan.chan_id]; 1990 1991 d40_log_cfg(&d40c->dma_cfg, 1992 &d40c->log_def.lcsp1, &d40c->log_def.lcsp3); 1993 1994 } else if (dma_has_cap(DMA_MEMCPY, cap) && 1995 dma_has_cap(DMA_SLAVE, cap)) { 1996 d40c->dma_cfg = dma40_memcpy_conf_phy; 1997 1998 /* Generate interrupt at end of transfer or relink. */ 1999 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_TIM_POS); 2000 2001 /* Generate interrupt on error. */ 2002 d40c->src_def_cfg |= BIT(D40_SREG_CFG_EIM_POS); 2003 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_EIM_POS); 2004 2005 } else { 2006 chan_err(d40c, "No memcpy\n"); 2007 return -EINVAL; 2008 } 2009 2010 return 0; 2011} 2012 2013static int d40_free_dma(struct d40_chan *d40c) 2014{ 2015 2016 int res = 0; 2017 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type); 2018 struct d40_phy_res *phy = d40c->phy_chan; 2019 bool is_src; 2020 2021 /* Terminate all queued and active transfers */ 2022 d40_term_all(d40c); 2023 2024 if (phy == NULL) { 2025 chan_err(d40c, "phy == null\n"); 2026 return -EINVAL; 2027 } 2028 2029 if (phy->allocated_src == D40_ALLOC_FREE && 2030 phy->allocated_dst == D40_ALLOC_FREE) { 2031 chan_err(d40c, "channel already free\n"); 2032 return -EINVAL; 2033 } 2034 2035 if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV || 2036 d40c->dma_cfg.dir == DMA_MEM_TO_MEM) 2037 is_src = false; 2038 else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) 2039 is_src = true; 2040 else { 2041 chan_err(d40c, "Unknown direction\n"); 2042 return -EINVAL; 2043 } 2044 2045 pm_runtime_get_sync(d40c->base->dev); 2046 res = d40_channel_execute_command(d40c, D40_DMA_STOP); 2047 if (res) { 2048 chan_err(d40c, "stop failed\n"); 2049 goto mark_last_busy; 2050 } 2051 2052 d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0); 2053 2054 if (chan_is_logical(d40c)) 2055 d40c->base->lookup_log_chans[d40c->log_num] = NULL; 2056 else 2057 d40c->base->lookup_phy_chans[phy->num] = NULL; 2058 2059 if (d40c->busy) { 2060 pm_runtime_mark_last_busy(d40c->base->dev); 2061 pm_runtime_put_autosuspend(d40c->base->dev); 2062 } 2063 2064 d40c->busy = false; 2065 d40c->phy_chan = NULL; 2066 d40c->configured = false; 2067 mark_last_busy: 2068 pm_runtime_mark_last_busy(d40c->base->dev); 2069 pm_runtime_put_autosuspend(d40c->base->dev); 2070 return res; 2071} 2072 2073static bool d40_is_paused(struct d40_chan *d40c) 2074{ 2075 void __iomem *chanbase = chan_base(d40c); 2076 bool is_paused = false; 2077 unsigned long flags; 2078 void __iomem *active_reg; 2079 u32 status; 2080 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type); 2081 2082 spin_lock_irqsave(&d40c->lock, flags); 2083 2084 if (chan_is_physical(d40c)) { 2085 if (d40c->phy_chan->num % 2 == 0) 2086 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE; 2087 else 2088 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO; 2089 2090 status = (readl(active_reg) & 2091 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >> 2092 D40_CHAN_POS(d40c->phy_chan->num); 2093 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP) 2094 is_paused = true; 2095 goto unlock; 2096 } 2097 2098 if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV || 2099 d40c->dma_cfg.dir == DMA_MEM_TO_MEM) { 2100 status = readl(chanbase + D40_CHAN_REG_SDLNK); 2101 } else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) { 2102 status = readl(chanbase + D40_CHAN_REG_SSLNK); 2103 } else { 2104 chan_err(d40c, "Unknown direction\n"); 2105 goto unlock; 2106 } 2107 2108 status = (status & D40_EVENTLINE_MASK(event)) >> 2109 D40_EVENTLINE_POS(event); 2110 2111 if (status != D40_DMA_RUN) 2112 is_paused = true; 2113 unlock: 2114 spin_unlock_irqrestore(&d40c->lock, flags); 2115 return is_paused; 2116 2117} 2118 2119static u32 stedma40_residue(struct dma_chan *chan) 2120{ 2121 struct d40_chan *d40c = 2122 container_of(chan, struct d40_chan, chan); 2123 u32 bytes_left; 2124 unsigned long flags; 2125 2126 spin_lock_irqsave(&d40c->lock, flags); 2127 bytes_left = d40_residue(d40c); 2128 spin_unlock_irqrestore(&d40c->lock, flags); 2129 2130 return bytes_left; 2131} 2132 2133static int 2134d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc, 2135 struct scatterlist *sg_src, struct scatterlist *sg_dst, 2136 unsigned int sg_len, dma_addr_t src_dev_addr, 2137 dma_addr_t dst_dev_addr) 2138{ 2139 struct stedma40_chan_cfg *cfg = &chan->dma_cfg; 2140 struct stedma40_half_channel_info *src_info = &cfg->src_info; 2141 struct stedma40_half_channel_info *dst_info = &cfg->dst_info; 2142 int ret; 2143 2144 ret = d40_log_sg_to_lli(sg_src, sg_len, 2145 src_dev_addr, 2146 desc->lli_log.src, 2147 chan->log_def.lcsp1, 2148 src_info->data_width, 2149 dst_info->data_width); 2150 2151 ret = d40_log_sg_to_lli(sg_dst, sg_len, 2152 dst_dev_addr, 2153 desc->lli_log.dst, 2154 chan->log_def.lcsp3, 2155 dst_info->data_width, 2156 src_info->data_width); 2157 2158 return ret < 0 ? ret : 0; 2159} 2160 2161static int 2162d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc, 2163 struct scatterlist *sg_src, struct scatterlist *sg_dst, 2164 unsigned int sg_len, dma_addr_t src_dev_addr, 2165 dma_addr_t dst_dev_addr) 2166{ 2167 struct stedma40_chan_cfg *cfg = &chan->dma_cfg; 2168 struct stedma40_half_channel_info *src_info = &cfg->src_info; 2169 struct stedma40_half_channel_info *dst_info = &cfg->dst_info; 2170 unsigned long flags = 0; 2171 int ret; 2172 2173 if (desc->cyclic) 2174 flags |= LLI_CYCLIC | LLI_TERM_INT; 2175 2176 ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr, 2177 desc->lli_phy.src, 2178 virt_to_phys(desc->lli_phy.src), 2179 chan->src_def_cfg, 2180 src_info, dst_info, flags); 2181 2182 ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr, 2183 desc->lli_phy.dst, 2184 virt_to_phys(desc->lli_phy.dst), 2185 chan->dst_def_cfg, 2186 dst_info, src_info, flags); 2187 2188 dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr, 2189 desc->lli_pool.size, DMA_TO_DEVICE); 2190 2191 return ret < 0 ? ret : 0; 2192} 2193 2194static struct d40_desc * 2195d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg, 2196 unsigned int sg_len, unsigned long dma_flags) 2197{ 2198 struct stedma40_chan_cfg *cfg; 2199 struct d40_desc *desc; 2200 int ret; 2201 2202 desc = d40_desc_get(chan); 2203 if (!desc) 2204 return NULL; 2205 2206 cfg = &chan->dma_cfg; 2207 desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width, 2208 cfg->dst_info.data_width); 2209 if (desc->lli_len < 0) { 2210 chan_err(chan, "Unaligned size\n"); 2211 goto free_desc; 2212 } 2213 2214 ret = d40_pool_lli_alloc(chan, desc, desc->lli_len); 2215 if (ret < 0) { 2216 chan_err(chan, "Could not allocate lli\n"); 2217 goto free_desc; 2218 } 2219 2220 desc->lli_current = 0; 2221 desc->txd.flags = dma_flags; 2222 desc->txd.tx_submit = d40_tx_submit; 2223 2224 dma_async_tx_descriptor_init(&desc->txd, &chan->chan); 2225 2226 return desc; 2227 free_desc: 2228 d40_desc_free(chan, desc); 2229 return NULL; 2230} 2231 2232static struct dma_async_tx_descriptor * 2233d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src, 2234 struct scatterlist *sg_dst, unsigned int sg_len, 2235 enum dma_transfer_direction direction, unsigned long dma_flags) 2236{ 2237 struct d40_chan *chan = container_of(dchan, struct d40_chan, chan); 2238 dma_addr_t src_dev_addr; 2239 dma_addr_t dst_dev_addr; 2240 struct d40_desc *desc; 2241 unsigned long flags; 2242 int ret; 2243 2244 if (!chan->phy_chan) { 2245 chan_err(chan, "Cannot prepare unallocated channel\n"); 2246 return NULL; 2247 } 2248 2249 d40_set_runtime_config_write(dchan, &chan->slave_config, direction); 2250 2251 spin_lock_irqsave(&chan->lock, flags); 2252 2253 desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags); 2254 if (desc == NULL) 2255 goto unlock; 2256 2257 if (sg_next(&sg_src[sg_len - 1]) == sg_src) 2258 desc->cyclic = true; 2259 2260 src_dev_addr = 0; 2261 dst_dev_addr = 0; 2262 if (direction == DMA_DEV_TO_MEM) 2263 src_dev_addr = chan->runtime_addr; 2264 else if (direction == DMA_MEM_TO_DEV) 2265 dst_dev_addr = chan->runtime_addr; 2266 2267 if (chan_is_logical(chan)) 2268 ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst, 2269 sg_len, src_dev_addr, dst_dev_addr); 2270 else 2271 ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst, 2272 sg_len, src_dev_addr, dst_dev_addr); 2273 2274 if (ret) { 2275 chan_err(chan, "Failed to prepare %s sg job: %d\n", 2276 chan_is_logical(chan) ? "log" : "phy", ret); 2277 goto free_desc; 2278 } 2279 2280 /* 2281 * add descriptor to the prepare queue in order to be able 2282 * to free them later in terminate_all 2283 */ 2284 list_add_tail(&desc->node, &chan->prepare_queue); 2285 2286 spin_unlock_irqrestore(&chan->lock, flags); 2287 2288 return &desc->txd; 2289 free_desc: 2290 d40_desc_free(chan, desc); 2291 unlock: 2292 spin_unlock_irqrestore(&chan->lock, flags); 2293 return NULL; 2294} 2295 2296static bool stedma40_filter(struct dma_chan *chan, void *data) 2297{ 2298 struct stedma40_chan_cfg *info = data; 2299 struct d40_chan *d40c = 2300 container_of(chan, struct d40_chan, chan); 2301 int err; 2302 2303 if (data) { 2304 err = d40_validate_conf(d40c, info); 2305 if (!err) 2306 d40c->dma_cfg = *info; 2307 } else 2308 err = d40_config_memcpy(d40c); 2309 2310 if (!err) 2311 d40c->configured = true; 2312 2313 return err == 0; 2314} 2315 2316static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src) 2317{ 2318 bool realtime = d40c->dma_cfg.realtime; 2319 bool highprio = d40c->dma_cfg.high_priority; 2320 u32 rtreg; 2321 u32 event = D40_TYPE_TO_EVENT(dev_type); 2322 u32 group = D40_TYPE_TO_GROUP(dev_type); 2323 u32 bit = BIT(event); 2324 u32 prioreg; 2325 struct d40_gen_dmac *dmac = &d40c->base->gen_dmac; 2326 2327 rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear; 2328 /* 2329 * Due to a hardware bug, in some cases a logical channel triggered by 2330 * a high priority destination event line can generate extra packet 2331 * transactions. 2332 * 2333 * The workaround is to not set the high priority level for the 2334 * destination event lines that trigger logical channels. 2335 */ 2336 if (!src && chan_is_logical(d40c)) 2337 highprio = false; 2338 2339 prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear; 2340 2341 /* Destination event lines are stored in the upper halfword */ 2342 if (!src) 2343 bit <<= 16; 2344 2345 writel(bit, d40c->base->virtbase + prioreg + group * 4); 2346 writel(bit, d40c->base->virtbase + rtreg + group * 4); 2347} 2348 2349static void d40_set_prio_realtime(struct d40_chan *d40c) 2350{ 2351 if (d40c->base->rev < 3) 2352 return; 2353 2354 if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) || 2355 (d40c->dma_cfg.dir == DMA_DEV_TO_DEV)) 2356 __d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, true); 2357 2358 if ((d40c->dma_cfg.dir == DMA_MEM_TO_DEV) || 2359 (d40c->dma_cfg.dir == DMA_DEV_TO_DEV)) 2360 __d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, false); 2361} 2362 2363#define D40_DT_FLAGS_MODE(flags) ((flags >> 0) & 0x1) 2364#define D40_DT_FLAGS_DIR(flags) ((flags >> 1) & 0x1) 2365#define D40_DT_FLAGS_BIG_ENDIAN(flags) ((flags >> 2) & 0x1) 2366#define D40_DT_FLAGS_FIXED_CHAN(flags) ((flags >> 3) & 0x1) 2367#define D40_DT_FLAGS_HIGH_PRIO(flags) ((flags >> 4) & 0x1) 2368 2369static struct dma_chan *d40_xlate(struct of_phandle_args *dma_spec, 2370 struct of_dma *ofdma) 2371{ 2372 struct stedma40_chan_cfg cfg; 2373 dma_cap_mask_t cap; 2374 u32 flags; 2375 2376 memset(&cfg, 0, sizeof(struct stedma40_chan_cfg)); 2377 2378 dma_cap_zero(cap); 2379 dma_cap_set(DMA_SLAVE, cap); 2380 2381 cfg.dev_type = dma_spec->args[0]; 2382 flags = dma_spec->args[2]; 2383 2384 switch (D40_DT_FLAGS_MODE(flags)) { 2385 case 0: cfg.mode = STEDMA40_MODE_LOGICAL; break; 2386 case 1: cfg.mode = STEDMA40_MODE_PHYSICAL; break; 2387 } 2388 2389 switch (D40_DT_FLAGS_DIR(flags)) { 2390 case 0: 2391 cfg.dir = DMA_MEM_TO_DEV; 2392 cfg.dst_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags); 2393 break; 2394 case 1: 2395 cfg.dir = DMA_DEV_TO_MEM; 2396 cfg.src_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags); 2397 break; 2398 } 2399 2400 if (D40_DT_FLAGS_FIXED_CHAN(flags)) { 2401 cfg.phy_channel = dma_spec->args[1]; 2402 cfg.use_fixed_channel = true; 2403 } 2404 2405 if (D40_DT_FLAGS_HIGH_PRIO(flags)) 2406 cfg.high_priority = true; 2407 2408 return dma_request_channel(cap, stedma40_filter, &cfg); 2409} 2410 2411/* DMA ENGINE functions */ 2412static int d40_alloc_chan_resources(struct dma_chan *chan) 2413{ 2414 int err; 2415 unsigned long flags; 2416 struct d40_chan *d40c = 2417 container_of(chan, struct d40_chan, chan); 2418 bool is_free_phy; 2419 spin_lock_irqsave(&d40c->lock, flags); 2420 2421 dma_cookie_init(chan); 2422 2423 /* If no dma configuration is set use default configuration (memcpy) */ 2424 if (!d40c->configured) { 2425 err = d40_config_memcpy(d40c); 2426 if (err) { 2427 chan_err(d40c, "Failed to configure memcpy channel\n"); 2428 goto mark_last_busy; 2429 } 2430 } 2431 2432 err = d40_allocate_channel(d40c, &is_free_phy); 2433 if (err) { 2434 chan_err(d40c, "Failed to allocate channel\n"); 2435 d40c->configured = false; 2436 goto mark_last_busy; 2437 } 2438 2439 pm_runtime_get_sync(d40c->base->dev); 2440 2441 d40_set_prio_realtime(d40c); 2442 2443 if (chan_is_logical(d40c)) { 2444 if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) 2445 d40c->lcpa = d40c->base->lcpa_base + 2446 d40c->dma_cfg.dev_type * D40_LCPA_CHAN_SIZE; 2447 else 2448 d40c->lcpa = d40c->base->lcpa_base + 2449 d40c->dma_cfg.dev_type * 2450 D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA; 2451 2452 /* Unmask the Global Interrupt Mask. */ 2453 d40c->src_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS); 2454 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS); 2455 } 2456 2457 dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n", 2458 chan_is_logical(d40c) ? "logical" : "physical", 2459 d40c->phy_chan->num, 2460 d40c->dma_cfg.use_fixed_channel ? ", fixed" : ""); 2461 2462 2463 /* 2464 * Only write channel configuration to the DMA if the physical 2465 * resource is free. In case of multiple logical channels 2466 * on the same physical resource, only the first write is necessary. 2467 */ 2468 if (is_free_phy) 2469 d40_config_write(d40c); 2470 mark_last_busy: 2471 pm_runtime_mark_last_busy(d40c->base->dev); 2472 pm_runtime_put_autosuspend(d40c->base->dev); 2473 spin_unlock_irqrestore(&d40c->lock, flags); 2474 return err; 2475} 2476 2477static void d40_free_chan_resources(struct dma_chan *chan) 2478{ 2479 struct d40_chan *d40c = 2480 container_of(chan, struct d40_chan, chan); 2481 int err; 2482 unsigned long flags; 2483 2484 if (d40c->phy_chan == NULL) { 2485 chan_err(d40c, "Cannot free unallocated channel\n"); 2486 return; 2487 } 2488 2489 spin_lock_irqsave(&d40c->lock, flags); 2490 2491 err = d40_free_dma(d40c); 2492 2493 if (err) 2494 chan_err(d40c, "Failed to free channel\n"); 2495 spin_unlock_irqrestore(&d40c->lock, flags); 2496} 2497 2498static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan, 2499 dma_addr_t dst, 2500 dma_addr_t src, 2501 size_t size, 2502 unsigned long dma_flags) 2503{ 2504 struct scatterlist dst_sg; 2505 struct scatterlist src_sg; 2506 2507 sg_init_table(&dst_sg, 1); 2508 sg_init_table(&src_sg, 1); 2509 2510 sg_dma_address(&dst_sg) = dst; 2511 sg_dma_address(&src_sg) = src; 2512 2513 sg_dma_len(&dst_sg) = size; 2514 sg_dma_len(&src_sg) = size; 2515 2516 return d40_prep_sg(chan, &src_sg, &dst_sg, 1, 2517 DMA_MEM_TO_MEM, dma_flags); 2518} 2519 2520static struct dma_async_tx_descriptor * 2521d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl, 2522 unsigned int sg_len, enum dma_transfer_direction direction, 2523 unsigned long dma_flags, void *context) 2524{ 2525 if (!is_slave_direction(direction)) 2526 return NULL; 2527 2528 return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags); 2529} 2530 2531static struct dma_async_tx_descriptor * 2532dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr, 2533 size_t buf_len, size_t period_len, 2534 enum dma_transfer_direction direction, unsigned long flags) 2535{ 2536 unsigned int periods = buf_len / period_len; 2537 struct dma_async_tx_descriptor *txd; 2538 struct scatterlist *sg; 2539 int i; 2540 2541 sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT); 2542 if (!sg) 2543 return NULL; 2544 2545 for (i = 0; i < periods; i++) { 2546 sg_dma_address(&sg[i]) = dma_addr; 2547 sg_dma_len(&sg[i]) = period_len; 2548 dma_addr += period_len; 2549 } 2550 2551 sg_chain(sg, periods + 1, sg); 2552 2553 txd = d40_prep_sg(chan, sg, sg, periods, direction, 2554 DMA_PREP_INTERRUPT); 2555 2556 kfree(sg); 2557 2558 return txd; 2559} 2560 2561static enum dma_status d40_tx_status(struct dma_chan *chan, 2562 dma_cookie_t cookie, 2563 struct dma_tx_state *txstate) 2564{ 2565 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan); 2566 enum dma_status ret; 2567 2568 if (d40c->phy_chan == NULL) { 2569 chan_err(d40c, "Cannot read status of unallocated channel\n"); 2570 return -EINVAL; 2571 } 2572 2573 ret = dma_cookie_status(chan, cookie, txstate); 2574 if (ret != DMA_COMPLETE && txstate) 2575 dma_set_residue(txstate, stedma40_residue(chan)); 2576 2577 if (d40_is_paused(d40c)) 2578 ret = DMA_PAUSED; 2579 2580 return ret; 2581} 2582 2583static void d40_issue_pending(struct dma_chan *chan) 2584{ 2585 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan); 2586 unsigned long flags; 2587 2588 if (d40c->phy_chan == NULL) { 2589 chan_err(d40c, "Channel is not allocated!\n"); 2590 return; 2591 } 2592 2593 spin_lock_irqsave(&d40c->lock, flags); 2594 2595 list_splice_tail_init(&d40c->pending_queue, &d40c->queue); 2596 2597 /* Busy means that queued jobs are already being processed */ 2598 if (!d40c->busy) 2599 (void) d40_queue_start(d40c); 2600 2601 spin_unlock_irqrestore(&d40c->lock, flags); 2602} 2603 2604static int d40_terminate_all(struct dma_chan *chan) 2605{ 2606 unsigned long flags; 2607 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan); 2608 int ret; 2609 2610 if (d40c->phy_chan == NULL) { 2611 chan_err(d40c, "Channel is not allocated!\n"); 2612 return -EINVAL; 2613 } 2614 2615 spin_lock_irqsave(&d40c->lock, flags); 2616 2617 pm_runtime_get_sync(d40c->base->dev); 2618 ret = d40_channel_execute_command(d40c, D40_DMA_STOP); 2619 if (ret) 2620 chan_err(d40c, "Failed to stop channel\n"); 2621 2622 d40_term_all(d40c); 2623 pm_runtime_mark_last_busy(d40c->base->dev); 2624 pm_runtime_put_autosuspend(d40c->base->dev); 2625 if (d40c->busy) { 2626 pm_runtime_mark_last_busy(d40c->base->dev); 2627 pm_runtime_put_autosuspend(d40c->base->dev); 2628 } 2629 d40c->busy = false; 2630 2631 spin_unlock_irqrestore(&d40c->lock, flags); 2632 return 0; 2633} 2634 2635static int 2636dma40_config_to_halfchannel(struct d40_chan *d40c, 2637 struct stedma40_half_channel_info *info, 2638 u32 maxburst) 2639{ 2640 int psize; 2641 2642 if (chan_is_logical(d40c)) { 2643 if (maxburst >= 16) 2644 psize = STEDMA40_PSIZE_LOG_16; 2645 else if (maxburst >= 8) 2646 psize = STEDMA40_PSIZE_LOG_8; 2647 else if (maxburst >= 4) 2648 psize = STEDMA40_PSIZE_LOG_4; 2649 else 2650 psize = STEDMA40_PSIZE_LOG_1; 2651 } else { 2652 if (maxburst >= 16) 2653 psize = STEDMA40_PSIZE_PHY_16; 2654 else if (maxburst >= 8) 2655 psize = STEDMA40_PSIZE_PHY_8; 2656 else if (maxburst >= 4) 2657 psize = STEDMA40_PSIZE_PHY_4; 2658 else 2659 psize = STEDMA40_PSIZE_PHY_1; 2660 } 2661 2662 info->psize = psize; 2663 info->flow_ctrl = STEDMA40_NO_FLOW_CTRL; 2664 2665 return 0; 2666} 2667 2668static int d40_set_runtime_config(struct dma_chan *chan, 2669 struct dma_slave_config *config) 2670{ 2671 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan); 2672 2673 memcpy(&d40c->slave_config, config, sizeof(*config)); 2674 2675 return 0; 2676} 2677 2678/* Runtime reconfiguration extension */ 2679static int d40_set_runtime_config_write(struct dma_chan *chan, 2680 struct dma_slave_config *config, 2681 enum dma_transfer_direction direction) 2682{ 2683 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan); 2684 struct stedma40_chan_cfg *cfg = &d40c->dma_cfg; 2685 enum dma_slave_buswidth src_addr_width, dst_addr_width; 2686 dma_addr_t config_addr; 2687 u32 src_maxburst, dst_maxburst; 2688 int ret; 2689 2690 if (d40c->phy_chan == NULL) { 2691 chan_err(d40c, "Channel is not allocated!\n"); 2692 return -EINVAL; 2693 } 2694 2695 src_addr_width = config->src_addr_width; 2696 src_maxburst = config->src_maxburst; 2697 dst_addr_width = config->dst_addr_width; 2698 dst_maxburst = config->dst_maxburst; 2699 2700 if (direction == DMA_DEV_TO_MEM) { 2701 config_addr = config->src_addr; 2702 2703 if (cfg->dir != DMA_DEV_TO_MEM) 2704 dev_dbg(d40c->base->dev, 2705 "channel was not configured for peripheral " 2706 "to memory transfer (%d) overriding\n", 2707 cfg->dir); 2708 cfg->dir = DMA_DEV_TO_MEM; 2709 2710 /* Configure the memory side */ 2711 if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED) 2712 dst_addr_width = src_addr_width; 2713 if (dst_maxburst == 0) 2714 dst_maxburst = src_maxburst; 2715 2716 } else if (direction == DMA_MEM_TO_DEV) { 2717 config_addr = config->dst_addr; 2718 2719 if (cfg->dir != DMA_MEM_TO_DEV) 2720 dev_dbg(d40c->base->dev, 2721 "channel was not configured for memory " 2722 "to peripheral transfer (%d) overriding\n", 2723 cfg->dir); 2724 cfg->dir = DMA_MEM_TO_DEV; 2725 2726 /* Configure the memory side */ 2727 if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED) 2728 src_addr_width = dst_addr_width; 2729 if (src_maxburst == 0) 2730 src_maxburst = dst_maxburst; 2731 } else { 2732 dev_err(d40c->base->dev, 2733 "unrecognized channel direction %d\n", 2734 direction); 2735 return -EINVAL; 2736 } 2737 2738 if (config_addr <= 0) { 2739 dev_err(d40c->base->dev, "no address supplied\n"); 2740 return -EINVAL; 2741 } 2742 2743 if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) { 2744 dev_err(d40c->base->dev, 2745 "src/dst width/maxburst mismatch: %d*%d != %d*%d\n", 2746 src_maxburst, 2747 src_addr_width, 2748 dst_maxburst, 2749 dst_addr_width); 2750 return -EINVAL; 2751 } 2752 2753 if (src_maxburst > 16) { 2754 src_maxburst = 16; 2755 dst_maxburst = src_maxburst * src_addr_width / dst_addr_width; 2756 } else if (dst_maxburst > 16) { 2757 dst_maxburst = 16; 2758 src_maxburst = dst_maxburst * dst_addr_width / src_addr_width; 2759 } 2760 2761 /* Only valid widths are; 1, 2, 4 and 8. */ 2762 if (src_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED || 2763 src_addr_width > DMA_SLAVE_BUSWIDTH_8_BYTES || 2764 dst_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED || 2765 dst_addr_width > DMA_SLAVE_BUSWIDTH_8_BYTES || 2766 !is_power_of_2(src_addr_width) || 2767 !is_power_of_2(dst_addr_width)) 2768 return -EINVAL; 2769 2770 cfg->src_info.data_width = src_addr_width; 2771 cfg->dst_info.data_width = dst_addr_width; 2772 2773 ret = dma40_config_to_halfchannel(d40c, &cfg->src_info, 2774 src_maxburst); 2775 if (ret) 2776 return ret; 2777 2778 ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info, 2779 dst_maxburst); 2780 if (ret) 2781 return ret; 2782 2783 /* Fill in register values */ 2784 if (chan_is_logical(d40c)) 2785 d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3); 2786 else 2787 d40_phy_cfg(cfg, &d40c->src_def_cfg, &d40c->dst_def_cfg); 2788 2789 /* These settings will take precedence later */ 2790 d40c->runtime_addr = config_addr; 2791 d40c->runtime_direction = direction; 2792 dev_dbg(d40c->base->dev, 2793 "configured channel %s for %s, data width %d/%d, " 2794 "maxburst %d/%d elements, LE, no flow control\n", 2795 dma_chan_name(chan), 2796 (direction == DMA_DEV_TO_MEM) ? "RX" : "TX", 2797 src_addr_width, dst_addr_width, 2798 src_maxburst, dst_maxburst); 2799 2800 return 0; 2801} 2802 2803/* Initialization functions */ 2804 2805static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma, 2806 struct d40_chan *chans, int offset, 2807 int num_chans) 2808{ 2809 int i = 0; 2810 struct d40_chan *d40c; 2811 2812 INIT_LIST_HEAD(&dma->channels); 2813 2814 for (i = offset; i < offset + num_chans; i++) { 2815 d40c = &chans[i]; 2816 d40c->base = base; 2817 d40c->chan.device = dma; 2818 2819 spin_lock_init(&d40c->lock); 2820 2821 d40c->log_num = D40_PHY_CHAN; 2822 2823 INIT_LIST_HEAD(&d40c->done); 2824 INIT_LIST_HEAD(&d40c->active); 2825 INIT_LIST_HEAD(&d40c->queue); 2826 INIT_LIST_HEAD(&d40c->pending_queue); 2827 INIT_LIST_HEAD(&d40c->client); 2828 INIT_LIST_HEAD(&d40c->prepare_queue); 2829 2830 tasklet_setup(&d40c->tasklet, dma_tasklet); 2831 2832 list_add_tail(&d40c->chan.device_node, 2833 &dma->channels); 2834 } 2835} 2836 2837static void d40_ops_init(struct d40_base *base, struct dma_device *dev) 2838{ 2839 if (dma_has_cap(DMA_SLAVE, dev->cap_mask)) { 2840 dev->device_prep_slave_sg = d40_prep_slave_sg; 2841 dev->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV); 2842 } 2843 2844 if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) { 2845 dev->device_prep_dma_memcpy = d40_prep_memcpy; 2846 dev->directions = BIT(DMA_MEM_TO_MEM); 2847 /* 2848 * This controller can only access address at even 2849 * 32bit boundaries, i.e. 2^2 2850 */ 2851 dev->copy_align = DMAENGINE_ALIGN_4_BYTES; 2852 } 2853 2854 if (dma_has_cap(DMA_CYCLIC, dev->cap_mask)) 2855 dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic; 2856 2857 dev->device_alloc_chan_resources = d40_alloc_chan_resources; 2858 dev->device_free_chan_resources = d40_free_chan_resources; 2859 dev->device_issue_pending = d40_issue_pending; 2860 dev->device_tx_status = d40_tx_status; 2861 dev->device_config = d40_set_runtime_config; 2862 dev->device_pause = d40_pause; 2863 dev->device_resume = d40_resume; 2864 dev->device_terminate_all = d40_terminate_all; 2865 dev->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST; 2866 dev->dev = base->dev; 2867} 2868 2869static int __init d40_dmaengine_init(struct d40_base *base, 2870 int num_reserved_chans) 2871{ 2872 int err ; 2873 2874 d40_chan_init(base, &base->dma_slave, base->log_chans, 2875 0, base->num_log_chans); 2876 2877 dma_cap_zero(base->dma_slave.cap_mask); 2878 dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask); 2879 dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask); 2880 2881 d40_ops_init(base, &base->dma_slave); 2882 2883 err = dmaenginem_async_device_register(&base->dma_slave); 2884 2885 if (err) { 2886 d40_err(base->dev, "Failed to register slave channels\n"); 2887 goto exit; 2888 } 2889 2890 d40_chan_init(base, &base->dma_memcpy, base->log_chans, 2891 base->num_log_chans, base->num_memcpy_chans); 2892 2893 dma_cap_zero(base->dma_memcpy.cap_mask); 2894 dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask); 2895 2896 d40_ops_init(base, &base->dma_memcpy); 2897 2898 err = dmaenginem_async_device_register(&base->dma_memcpy); 2899 2900 if (err) { 2901 d40_err(base->dev, 2902 "Failed to register memcpy only channels\n"); 2903 goto exit; 2904 } 2905 2906 d40_chan_init(base, &base->dma_both, base->phy_chans, 2907 0, num_reserved_chans); 2908 2909 dma_cap_zero(base->dma_both.cap_mask); 2910 dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask); 2911 dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask); 2912 dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask); 2913 2914 d40_ops_init(base, &base->dma_both); 2915 err = dmaenginem_async_device_register(&base->dma_both); 2916 2917 if (err) { 2918 d40_err(base->dev, 2919 "Failed to register logical and physical capable channels\n"); 2920 goto exit; 2921 } 2922 return 0; 2923 exit: 2924 return err; 2925} 2926 2927/* Suspend resume functionality */ 2928#ifdef CONFIG_PM_SLEEP 2929static int dma40_suspend(struct device *dev) 2930{ 2931 struct d40_base *base = dev_get_drvdata(dev); 2932 int ret; 2933 2934 ret = pm_runtime_force_suspend(dev); 2935 if (ret) 2936 return ret; 2937 2938 if (base->lcpa_regulator) 2939 ret = regulator_disable(base->lcpa_regulator); 2940 return ret; 2941} 2942 2943static int dma40_resume(struct device *dev) 2944{ 2945 struct d40_base *base = dev_get_drvdata(dev); 2946 int ret = 0; 2947 2948 if (base->lcpa_regulator) { 2949 ret = regulator_enable(base->lcpa_regulator); 2950 if (ret) 2951 return ret; 2952 } 2953 2954 return pm_runtime_force_resume(dev); 2955} 2956#endif 2957 2958#ifdef CONFIG_PM 2959static void dma40_backup(void __iomem *baseaddr, u32 *backup, 2960 u32 *regaddr, int num, bool save) 2961{ 2962 int i; 2963 2964 for (i = 0; i < num; i++) { 2965 void __iomem *addr = baseaddr + regaddr[i]; 2966 2967 if (save) 2968 backup[i] = readl_relaxed(addr); 2969 else 2970 writel_relaxed(backup[i], addr); 2971 } 2972} 2973 2974static void d40_save_restore_registers(struct d40_base *base, bool save) 2975{ 2976 int i; 2977 2978 /* Save/Restore channel specific registers */ 2979 for (i = 0; i < base->num_phy_chans; i++) { 2980 void __iomem *addr; 2981 int idx; 2982 2983 if (base->phy_res[i].reserved) 2984 continue; 2985 2986 addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA; 2987 idx = i * ARRAY_SIZE(d40_backup_regs_chan); 2988 2989 dma40_backup(addr, &base->reg_val_backup_chan[idx], 2990 d40_backup_regs_chan, 2991 ARRAY_SIZE(d40_backup_regs_chan), 2992 save); 2993 } 2994 2995 /* Save/Restore global registers */ 2996 dma40_backup(base->virtbase, base->reg_val_backup, 2997 d40_backup_regs, ARRAY_SIZE(d40_backup_regs), 2998 save); 2999 3000 /* Save/Restore registers only existing on dma40 v3 and later */ 3001 if (base->gen_dmac.backup) 3002 dma40_backup(base->virtbase, base->reg_val_backup_v4, 3003 base->gen_dmac.backup, 3004 base->gen_dmac.backup_size, 3005 save); 3006} 3007 3008static int dma40_runtime_suspend(struct device *dev) 3009{ 3010 struct d40_base *base = dev_get_drvdata(dev); 3011 3012 d40_save_restore_registers(base, true); 3013 3014 /* Don't disable/enable clocks for v1 due to HW bugs */ 3015 if (base->rev != 1) 3016 writel_relaxed(base->gcc_pwr_off_mask, 3017 base->virtbase + D40_DREG_GCC); 3018 3019 return 0; 3020} 3021 3022static int dma40_runtime_resume(struct device *dev) 3023{ 3024 struct d40_base *base = dev_get_drvdata(dev); 3025 3026 d40_save_restore_registers(base, false); 3027 3028 writel_relaxed(D40_DREG_GCC_ENABLE_ALL, 3029 base->virtbase + D40_DREG_GCC); 3030 return 0; 3031} 3032#endif 3033 3034static const struct dev_pm_ops dma40_pm_ops = { 3035 SET_LATE_SYSTEM_SLEEP_PM_OPS(dma40_suspend, dma40_resume) 3036 SET_RUNTIME_PM_OPS(dma40_runtime_suspend, 3037 dma40_runtime_resume, 3038 NULL) 3039}; 3040 3041/* Initialization functions. */ 3042 3043static int __init d40_phy_res_init(struct d40_base *base) 3044{ 3045 int i; 3046 int num_phy_chans_avail = 0; 3047 u32 val[2]; 3048 int odd_even_bit = -2; 3049 int gcc = D40_DREG_GCC_ENA; 3050 3051 val[0] = readl(base->virtbase + D40_DREG_PRSME); 3052 val[1] = readl(base->virtbase + D40_DREG_PRSMO); 3053 3054 for (i = 0; i < base->num_phy_chans; i++) { 3055 base->phy_res[i].num = i; 3056 odd_even_bit += 2 * ((i % 2) == 0); 3057 if (((val[i % 2] >> odd_even_bit) & 3) == 1) { 3058 /* Mark security only channels as occupied */ 3059 base->phy_res[i].allocated_src = D40_ALLOC_PHY; 3060 base->phy_res[i].allocated_dst = D40_ALLOC_PHY; 3061 base->phy_res[i].reserved = true; 3062 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i), 3063 D40_DREG_GCC_SRC); 3064 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i), 3065 D40_DREG_GCC_DST); 3066 3067 3068 } else { 3069 base->phy_res[i].allocated_src = D40_ALLOC_FREE; 3070 base->phy_res[i].allocated_dst = D40_ALLOC_FREE; 3071 base->phy_res[i].reserved = false; 3072 num_phy_chans_avail++; 3073 } 3074 spin_lock_init(&base->phy_res[i].lock); 3075 } 3076 3077 /* Mark disabled channels as occupied */ 3078 for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) { 3079 int chan = base->plat_data->disabled_channels[i]; 3080 3081 base->phy_res[chan].allocated_src = D40_ALLOC_PHY; 3082 base->phy_res[chan].allocated_dst = D40_ALLOC_PHY; 3083 base->phy_res[chan].reserved = true; 3084 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan), 3085 D40_DREG_GCC_SRC); 3086 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan), 3087 D40_DREG_GCC_DST); 3088 num_phy_chans_avail--; 3089 } 3090 3091 /* Mark soft_lli channels */ 3092 for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) { 3093 int chan = base->plat_data->soft_lli_chans[i]; 3094 3095 base->phy_res[chan].use_soft_lli = true; 3096 } 3097 3098 dev_info(base->dev, "%d of %d physical DMA channels available\n", 3099 num_phy_chans_avail, base->num_phy_chans); 3100 3101 /* Verify settings extended vs standard */ 3102 val[0] = readl(base->virtbase + D40_DREG_PRTYP); 3103 3104 for (i = 0; i < base->num_phy_chans; i++) { 3105 3106 if (base->phy_res[i].allocated_src == D40_ALLOC_FREE && 3107 (val[0] & 0x3) != 1) 3108 dev_info(base->dev, 3109 "[%s] INFO: channel %d is misconfigured (%d)\n", 3110 __func__, i, val[0] & 0x3); 3111 3112 val[0] = val[0] >> 2; 3113 } 3114 3115 /* 3116 * To keep things simple, Enable all clocks initially. 3117 * The clocks will get managed later post channel allocation. 3118 * The clocks for the event lines on which reserved channels exists 3119 * are not managed here. 3120 */ 3121 writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC); 3122 base->gcc_pwr_off_mask = gcc; 3123 3124 return num_phy_chans_avail; 3125} 3126 3127/* Called from the registered devm action */ 3128static void d40_drop_kmem_cache_action(void *d) 3129{ 3130 struct kmem_cache *desc_slab = d; 3131 3132 kmem_cache_destroy(desc_slab); 3133} 3134 3135static int __init d40_hw_detect_init(struct platform_device *pdev, 3136 struct d40_base **retbase) 3137{ 3138 struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev); 3139 struct device *dev = &pdev->dev; 3140 struct clk *clk; 3141 void __iomem *virtbase; 3142 struct d40_base *base; 3143 int num_log_chans; 3144 int num_phy_chans; 3145 int num_memcpy_chans; 3146 int i; 3147 u32 pid; 3148 u32 cid; 3149 u8 rev; 3150 int ret; 3151 3152 clk = devm_clk_get_enabled(dev, NULL); 3153 if (IS_ERR(clk)) 3154 return PTR_ERR(clk); 3155 3156 /* Get IO for DMAC base address */ 3157 virtbase = devm_platform_ioremap_resource_byname(pdev, "base"); 3158 if (IS_ERR(virtbase)) 3159 return PTR_ERR(virtbase); 3160 3161 /* This is just a regular AMBA PrimeCell ID actually */ 3162 for (pid = 0, i = 0; i < 4; i++) 3163 pid |= (readl(virtbase + SZ_4K - 0x20 + 4 * i) 3164 & 255) << (i * 8); 3165 for (cid = 0, i = 0; i < 4; i++) 3166 cid |= (readl(virtbase + SZ_4K - 0x10 + 4 * i) 3167 & 255) << (i * 8); 3168 3169 if (cid != AMBA_CID) { 3170 d40_err(dev, "Unknown hardware! No PrimeCell ID\n"); 3171 return -EINVAL; 3172 } 3173 if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) { 3174 d40_err(dev, "Unknown designer! Got %x wanted %x\n", 3175 AMBA_MANF_BITS(pid), 3176 AMBA_VENDOR_ST); 3177 return -EINVAL; 3178 } 3179 /* 3180 * HW revision: 3181 * DB8500ed has revision 0 3182 * ? has revision 1 3183 * DB8500v1 has revision 2 3184 * DB8500v2 has revision 3 3185 * AP9540v1 has revision 4 3186 * DB8540v1 has revision 4 3187 */ 3188 rev = AMBA_REV_BITS(pid); 3189 if (rev < 2) { 3190 d40_err(dev, "hardware revision: %d is not supported", rev); 3191 return -EINVAL; 3192 } 3193 3194 /* The number of physical channels on this HW */ 3195 if (plat_data->num_of_phy_chans) 3196 num_phy_chans = plat_data->num_of_phy_chans; 3197 else 3198 num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4; 3199 3200 /* The number of channels used for memcpy */ 3201 if (plat_data->num_of_memcpy_chans) 3202 num_memcpy_chans = plat_data->num_of_memcpy_chans; 3203 else 3204 num_memcpy_chans = ARRAY_SIZE(dma40_memcpy_channels); 3205 3206 num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY; 3207 3208 dev_info(dev, 3209 "hardware rev: %d with %d physical and %d logical channels\n", 3210 rev, num_phy_chans, num_log_chans); 3211 3212 base = devm_kzalloc(dev, 3213 ALIGN(sizeof(struct d40_base), 4) + 3214 (num_phy_chans + num_log_chans + num_memcpy_chans) * 3215 sizeof(struct d40_chan), GFP_KERNEL); 3216 3217 if (!base) 3218 return -ENOMEM; 3219 3220 base->rev = rev; 3221 base->clk = clk; 3222 base->num_memcpy_chans = num_memcpy_chans; 3223 base->num_phy_chans = num_phy_chans; 3224 base->num_log_chans = num_log_chans; 3225 base->virtbase = virtbase; 3226 base->plat_data = plat_data; 3227 base->dev = dev; 3228 base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4); 3229 base->log_chans = &base->phy_chans[num_phy_chans]; 3230 3231 if (base->plat_data->num_of_phy_chans == 14) { 3232 base->gen_dmac.backup = d40_backup_regs_v4b; 3233 base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B; 3234 base->gen_dmac.interrupt_en = D40_DREG_CPCMIS; 3235 base->gen_dmac.interrupt_clear = D40_DREG_CPCICR; 3236 base->gen_dmac.realtime_en = D40_DREG_CRSEG1; 3237 base->gen_dmac.realtime_clear = D40_DREG_CRCEG1; 3238 base->gen_dmac.high_prio_en = D40_DREG_CPSEG1; 3239 base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1; 3240 base->gen_dmac.il = il_v4b; 3241 base->gen_dmac.il_size = ARRAY_SIZE(il_v4b); 3242 base->gen_dmac.init_reg = dma_init_reg_v4b; 3243 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b); 3244 } else { 3245 if (base->rev >= 3) { 3246 base->gen_dmac.backup = d40_backup_regs_v4a; 3247 base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A; 3248 } 3249 base->gen_dmac.interrupt_en = D40_DREG_PCMIS; 3250 base->gen_dmac.interrupt_clear = D40_DREG_PCICR; 3251 base->gen_dmac.realtime_en = D40_DREG_RSEG1; 3252 base->gen_dmac.realtime_clear = D40_DREG_RCEG1; 3253 base->gen_dmac.high_prio_en = D40_DREG_PSEG1; 3254 base->gen_dmac.high_prio_clear = D40_DREG_PCEG1; 3255 base->gen_dmac.il = il_v4a; 3256 base->gen_dmac.il_size = ARRAY_SIZE(il_v4a); 3257 base->gen_dmac.init_reg = dma_init_reg_v4a; 3258 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a); 3259 } 3260 3261 base->phy_res = devm_kcalloc(dev, num_phy_chans, 3262 sizeof(*base->phy_res), 3263 GFP_KERNEL); 3264 if (!base->phy_res) 3265 return -ENOMEM; 3266 3267 base->lookup_phy_chans = devm_kcalloc(dev, num_phy_chans, 3268 sizeof(*base->lookup_phy_chans), 3269 GFP_KERNEL); 3270 if (!base->lookup_phy_chans) 3271 return -ENOMEM; 3272 3273 base->lookup_log_chans = devm_kcalloc(dev, num_log_chans, 3274 sizeof(*base->lookup_log_chans), 3275 GFP_KERNEL); 3276 if (!base->lookup_log_chans) 3277 return -ENOMEM; 3278 3279 base->reg_val_backup_chan = devm_kmalloc_array(dev, base->num_phy_chans, 3280 sizeof(d40_backup_regs_chan), 3281 GFP_KERNEL); 3282 if (!base->reg_val_backup_chan) 3283 return -ENOMEM; 3284 3285 base->lcla_pool.alloc_map = devm_kcalloc(dev, num_phy_chans 3286 * D40_LCLA_LINK_PER_EVENT_GRP, 3287 sizeof(*base->lcla_pool.alloc_map), 3288 GFP_KERNEL); 3289 if (!base->lcla_pool.alloc_map) 3290 return -ENOMEM; 3291 3292 base->regs_interrupt = devm_kmalloc_array(dev, base->gen_dmac.il_size, 3293 sizeof(*base->regs_interrupt), 3294 GFP_KERNEL); 3295 if (!base->regs_interrupt) 3296 return -ENOMEM; 3297 3298 base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc), 3299 0, SLAB_HWCACHE_ALIGN, 3300 NULL); 3301 if (!base->desc_slab) 3302 return -ENOMEM; 3303 3304 ret = devm_add_action_or_reset(dev, d40_drop_kmem_cache_action, 3305 base->desc_slab); 3306 if (ret) 3307 return ret; 3308 3309 *retbase = base; 3310 3311 return 0; 3312} 3313 3314static void __init d40_hw_init(struct d40_base *base) 3315{ 3316 3317 int i; 3318 u32 prmseo[2] = {0, 0}; 3319 u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF}; 3320 u32 pcmis = 0; 3321 u32 pcicr = 0; 3322 struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg; 3323 u32 reg_size = base->gen_dmac.init_reg_size; 3324 3325 for (i = 0; i < reg_size; i++) 3326 writel(dma_init_reg[i].val, 3327 base->virtbase + dma_init_reg[i].reg); 3328 3329 /* Configure all our dma channels to default settings */ 3330 for (i = 0; i < base->num_phy_chans; i++) { 3331 3332 activeo[i % 2] = activeo[i % 2] << 2; 3333 3334 if (base->phy_res[base->num_phy_chans - i - 1].allocated_src 3335 == D40_ALLOC_PHY) { 3336 activeo[i % 2] |= 3; 3337 continue; 3338 } 3339 3340 /* Enable interrupt # */ 3341 pcmis = (pcmis << 1) | 1; 3342 3343 /* Clear interrupt # */ 3344 pcicr = (pcicr << 1) | 1; 3345 3346 /* Set channel to physical mode */ 3347 prmseo[i % 2] = prmseo[i % 2] << 2; 3348 prmseo[i % 2] |= 1; 3349 3350 } 3351 3352 writel(prmseo[1], base->virtbase + D40_DREG_PRMSE); 3353 writel(prmseo[0], base->virtbase + D40_DREG_PRMSO); 3354 writel(activeo[1], base->virtbase + D40_DREG_ACTIVE); 3355 writel(activeo[0], base->virtbase + D40_DREG_ACTIVO); 3356 3357 /* Write which interrupt to enable */ 3358 writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en); 3359 3360 /* Write which interrupt to clear */ 3361 writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear); 3362 3363 /* These are __initdata and cannot be accessed after init */ 3364 base->gen_dmac.init_reg = NULL; 3365 base->gen_dmac.init_reg_size = 0; 3366} 3367 3368static int __init d40_lcla_allocate(struct d40_base *base) 3369{ 3370 struct d40_lcla_pool *pool = &base->lcla_pool; 3371 unsigned long *page_list; 3372 int i, j; 3373 int ret; 3374 3375 /* 3376 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned, 3377 * To full fill this hardware requirement without wasting 256 kb 3378 * we allocate pages until we get an aligned one. 3379 */ 3380 page_list = kmalloc_array(MAX_LCLA_ALLOC_ATTEMPTS, 3381 sizeof(*page_list), 3382 GFP_KERNEL); 3383 if (!page_list) 3384 return -ENOMEM; 3385 3386 /* Calculating how many pages that are required */ 3387 base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE; 3388 3389 for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) { 3390 page_list[i] = __get_free_pages(GFP_KERNEL, 3391 base->lcla_pool.pages); 3392 if (!page_list[i]) { 3393 3394 d40_err(base->dev, "Failed to allocate %d pages.\n", 3395 base->lcla_pool.pages); 3396 ret = -ENOMEM; 3397 3398 for (j = 0; j < i; j++) 3399 free_pages(page_list[j], base->lcla_pool.pages); 3400 goto free_page_list; 3401 } 3402 3403 if ((virt_to_phys((void *)page_list[i]) & 3404 (LCLA_ALIGNMENT - 1)) == 0) 3405 break; 3406 } 3407 3408 for (j = 0; j < i; j++) 3409 free_pages(page_list[j], base->lcla_pool.pages); 3410 3411 if (i < MAX_LCLA_ALLOC_ATTEMPTS) { 3412 base->lcla_pool.base = (void *)page_list[i]; 3413 } else { 3414 /* 3415 * After many attempts and no succees with finding the correct 3416 * alignment, try with allocating a big buffer. 3417 */ 3418 dev_warn(base->dev, 3419 "[%s] Failed to get %d pages @ 18 bit align.\n", 3420 __func__, base->lcla_pool.pages); 3421 base->lcla_pool.base_unaligned = kmalloc(SZ_1K * 3422 base->num_phy_chans + 3423 LCLA_ALIGNMENT, 3424 GFP_KERNEL); 3425 if (!base->lcla_pool.base_unaligned) { 3426 ret = -ENOMEM; 3427 goto free_page_list; 3428 } 3429 3430 base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned, 3431 LCLA_ALIGNMENT); 3432 } 3433 3434 pool->dma_addr = dma_map_single(base->dev, pool->base, 3435 SZ_1K * base->num_phy_chans, 3436 DMA_TO_DEVICE); 3437 if (dma_mapping_error(base->dev, pool->dma_addr)) { 3438 pool->dma_addr = 0; 3439 ret = -ENOMEM; 3440 goto free_page_list; 3441 } 3442 3443 writel(virt_to_phys(base->lcla_pool.base), 3444 base->virtbase + D40_DREG_LCLA); 3445 ret = 0; 3446 free_page_list: 3447 kfree(page_list); 3448 return ret; 3449} 3450 3451static int __init d40_of_probe(struct device *dev, 3452 struct device_node *np) 3453{ 3454 struct stedma40_platform_data *pdata; 3455 int num_phy = 0, num_memcpy = 0, num_disabled = 0; 3456 const __be32 *list; 3457 3458 pdata = devm_kzalloc(dev, sizeof(*pdata), GFP_KERNEL); 3459 if (!pdata) 3460 return -ENOMEM; 3461 3462 /* If absent this value will be obtained from h/w. */ 3463 of_property_read_u32(np, "dma-channels", &num_phy); 3464 if (num_phy > 0) 3465 pdata->num_of_phy_chans = num_phy; 3466 3467 list = of_get_property(np, "memcpy-channels", &num_memcpy); 3468 num_memcpy /= sizeof(*list); 3469 3470 if (num_memcpy > D40_MEMCPY_MAX_CHANS || num_memcpy <= 0) { 3471 d40_err(dev, 3472 "Invalid number of memcpy channels specified (%d)\n", 3473 num_memcpy); 3474 return -EINVAL; 3475 } 3476 pdata->num_of_memcpy_chans = num_memcpy; 3477 3478 of_property_read_u32_array(np, "memcpy-channels", 3479 dma40_memcpy_channels, 3480 num_memcpy); 3481 3482 list = of_get_property(np, "disabled-channels", &num_disabled); 3483 num_disabled /= sizeof(*list); 3484 3485 if (num_disabled >= STEDMA40_MAX_PHYS || num_disabled < 0) { 3486 d40_err(dev, 3487 "Invalid number of disabled channels specified (%d)\n", 3488 num_disabled); 3489 return -EINVAL; 3490 } 3491 3492 of_property_read_u32_array(np, "disabled-channels", 3493 pdata->disabled_channels, 3494 num_disabled); 3495 pdata->disabled_channels[num_disabled] = -1; 3496 3497 dev->platform_data = pdata; 3498 3499 return 0; 3500} 3501 3502static int __init d40_probe(struct platform_device *pdev) 3503{ 3504 struct device *dev = &pdev->dev; 3505 struct device_node *np = pdev->dev.of_node; 3506 struct device_node *np_lcpa; 3507 struct d40_base *base; 3508 struct resource *res; 3509 struct resource res_lcpa; 3510 int num_reserved_chans; 3511 u32 val; 3512 int ret; 3513 3514 if (d40_of_probe(dev, np)) { 3515 ret = -ENOMEM; 3516 goto report_failure; 3517 } 3518 3519 ret = d40_hw_detect_init(pdev, &base); 3520 if (ret) 3521 goto report_failure; 3522 3523 num_reserved_chans = d40_phy_res_init(base); 3524 3525 platform_set_drvdata(pdev, base); 3526 3527 spin_lock_init(&base->interrupt_lock); 3528 spin_lock_init(&base->execmd_lock); 3529 3530 /* Get IO for logical channel parameter address (LCPA) */ 3531 np_lcpa = of_parse_phandle(np, "sram", 0); 3532 if (!np_lcpa) { 3533 dev_err(dev, "no LCPA SRAM node\n"); 3534 ret = -EINVAL; 3535 goto report_failure; 3536 } 3537 /* This is no device so read the address directly from the node */ 3538 ret = of_address_to_resource(np_lcpa, 0, &res_lcpa); 3539 if (ret) { 3540 dev_err(dev, "no LCPA SRAM resource\n"); 3541 goto report_failure; 3542 } 3543 base->lcpa_size = resource_size(&res_lcpa); 3544 base->phy_lcpa = res_lcpa.start; 3545 dev_info(dev, "found LCPA SRAM at %pad, size %pa\n", 3546 &base->phy_lcpa, &base->lcpa_size); 3547 3548 /* We make use of ESRAM memory for this. */ 3549 val = readl(base->virtbase + D40_DREG_LCPA); 3550 if (base->phy_lcpa != val && val != 0) { 3551 dev_warn(dev, 3552 "[%s] Mismatch LCPA dma 0x%x, def %08x\n", 3553 __func__, val, (u32)base->phy_lcpa); 3554 } else 3555 writel(base->phy_lcpa, base->virtbase + D40_DREG_LCPA); 3556 3557 base->lcpa_base = devm_ioremap(dev, base->phy_lcpa, base->lcpa_size); 3558 if (!base->lcpa_base) { 3559 ret = -ENOMEM; 3560 d40_err(dev, "Failed to ioremap LCPA region\n"); 3561 goto report_failure; 3562 } 3563 /* If lcla has to be located in ESRAM we don't need to allocate */ 3564 if (base->plat_data->use_esram_lcla) { 3565 res = platform_get_resource_byname(pdev, IORESOURCE_MEM, 3566 "lcla_esram"); 3567 if (!res) { 3568 ret = -ENOENT; 3569 d40_err(dev, 3570 "No \"lcla_esram\" memory resource\n"); 3571 goto report_failure; 3572 } 3573 base->lcla_pool.base = devm_ioremap(dev, res->start, 3574 resource_size(res)); 3575 if (!base->lcla_pool.base) { 3576 ret = -ENOMEM; 3577 d40_err(dev, "Failed to ioremap LCLA region\n"); 3578 goto report_failure; 3579 } 3580 writel(res->start, base->virtbase + D40_DREG_LCLA); 3581 3582 } else { 3583 ret = d40_lcla_allocate(base); 3584 if (ret) { 3585 d40_err(dev, "Failed to allocate LCLA area\n"); 3586 goto destroy_cache; 3587 } 3588 } 3589 3590 spin_lock_init(&base->lcla_pool.lock); 3591 3592 base->irq = platform_get_irq(pdev, 0); 3593 if (base->irq < 0) { 3594 ret = base->irq; 3595 goto destroy_cache; 3596 } 3597 3598 ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base); 3599 if (ret) { 3600 d40_err(dev, "No IRQ defined\n"); 3601 goto destroy_cache; 3602 } 3603 3604 if (base->plat_data->use_esram_lcla) { 3605 3606 base->lcpa_regulator = regulator_get(base->dev, "lcla_esram"); 3607 if (IS_ERR(base->lcpa_regulator)) { 3608 d40_err(dev, "Failed to get lcpa_regulator\n"); 3609 ret = PTR_ERR(base->lcpa_regulator); 3610 base->lcpa_regulator = NULL; 3611 goto destroy_cache; 3612 } 3613 3614 ret = regulator_enable(base->lcpa_regulator); 3615 if (ret) { 3616 d40_err(dev, 3617 "Failed to enable lcpa_regulator\n"); 3618 regulator_put(base->lcpa_regulator); 3619 base->lcpa_regulator = NULL; 3620 goto destroy_cache; 3621 } 3622 } 3623 3624 writel_relaxed(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC); 3625 3626 pm_runtime_irq_safe(base->dev); 3627 pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY); 3628 pm_runtime_use_autosuspend(base->dev); 3629 pm_runtime_mark_last_busy(base->dev); 3630 pm_runtime_set_active(base->dev); 3631 pm_runtime_enable(base->dev); 3632 3633 ret = d40_dmaengine_init(base, num_reserved_chans); 3634 if (ret) 3635 goto destroy_cache; 3636 3637 ret = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE); 3638 if (ret) { 3639 d40_err(dev, "Failed to set dma max seg size\n"); 3640 goto destroy_cache; 3641 } 3642 3643 d40_hw_init(base); 3644 3645 ret = of_dma_controller_register(np, d40_xlate, NULL); 3646 if (ret) { 3647 dev_err(dev, 3648 "could not register of_dma_controller\n"); 3649 goto destroy_cache; 3650 } 3651 3652 dev_info(base->dev, "initialized\n"); 3653 return 0; 3654 3655 destroy_cache: 3656 if (base->lcla_pool.dma_addr) 3657 dma_unmap_single(base->dev, base->lcla_pool.dma_addr, 3658 SZ_1K * base->num_phy_chans, 3659 DMA_TO_DEVICE); 3660 3661 if (!base->lcla_pool.base_unaligned && base->lcla_pool.base) 3662 free_pages((unsigned long)base->lcla_pool.base, 3663 base->lcla_pool.pages); 3664 3665 kfree(base->lcla_pool.base_unaligned); 3666 3667 if (base->lcpa_regulator) { 3668 regulator_disable(base->lcpa_regulator); 3669 regulator_put(base->lcpa_regulator); 3670 } 3671 pm_runtime_disable(base->dev); 3672 3673 report_failure: 3674 d40_err(dev, "probe failed\n"); 3675 return ret; 3676} 3677 3678static const struct of_device_id d40_match[] = { 3679 { .compatible = "stericsson,dma40", }, 3680 {} 3681}; 3682 3683static struct platform_driver d40_driver = { 3684 .driver = { 3685 .name = D40_NAME, 3686 .pm = &dma40_pm_ops, 3687 .of_match_table = d40_match, 3688 }, 3689}; 3690 3691static int __init stedma40_init(void) 3692{ 3693 return platform_driver_probe(&d40_driver, d40_probe); 3694} 3695subsys_initcall(stedma40_init); 3696