drivers/dma/ste_dma40.c

   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  */
  50 struct 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
 101 static 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 */
 111 static 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 */
 125 static 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  */
 146 enum 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
 162 enum 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  */
 174 static __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  */
 197 static 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
 218 static 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
 243 static __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  */
 266 struct d40_interrupt_lookup {
 267         u32 src;
 268         u32 clr;
 269         bool is_error;
 270         int offset;
 271 };
 272
 273
 274 static 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
 287 static 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  */
 308 struct d40_reg_val {
 309         unsigned int reg;
 310         unsigned int val;
 311 };
 312
 313 static __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 };
 331 static __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  */
 364 struct 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  */
 391 struct 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  */
 421 struct 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  */
 444 struct 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
 453 struct 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  */
 487 struct 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  */
 532 struct 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  */
 593 struct 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
 628 static struct device *chan2dev(struct d40_chan *d40c)
 629 {
 630         return &d40c->chan.dev->device;
 631 }
 632
 633 static bool chan_is_physical(struct d40_chan *chan)
 634 {
 635         return chan->log_num == D40_PHY_CHAN;
 636 }
 637
 638 static bool chan_is_logical(struct d40_chan *chan)
 639 {
 640         return !chan_is_physical(chan);
 641 }
 642
 643 static 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
 655 static int d40_set_runtime_config_write(struct dma_chan *chan,
 656                                   struct dma_slave_config *config,
 657                                   enum dma_transfer_direction direction);
 658
 659 static 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
 711 static 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
 726 static 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
 755 static 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
 786 static void d40_desc_remove(struct d40_desc *d40d)
 787 {
 788         list_del(&d40d->node);
 789 }
 790
 791 static 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
 818 static 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
 826 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
 827 {
 828         list_add_tail(&desc->node, &d40c->active);
 829 }
 830
 831 static 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
 848 static void d40_desc_done(struct d40_chan *d40c, struct d40_desc *desc)
 849 {
 850         list_add_tail(&desc->node, &d40c->done);
 851 }
 852
 853 static 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
 965 static 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
 974 static 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 */
 980 static 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
 987 static 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
 993 static 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
 998 static 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
1003 static 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  */
1022 static 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
1045 static 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
1063 static 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
1133 static 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
1180 static 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
1263 static 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
1279 static 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
1290 static 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
1339 static 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
1348 static 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
1373 static 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
1409 static 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
1425 static 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
1438 static 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
1463 static 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
1490 static 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
1507 static int d40_start(struct d40_chan *d40c)
1508 {
1509         return d40_channel_execute_command(d40c, D40_DMA_RUN);
1510 }
1511
1512 static 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 */
1546 static 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
1599 static 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
1664 static 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
1726 static 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
1769 static 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
1823 static 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
1856 static 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;
1916 found_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
1969 found_log:
1970         d40c->phy_chan = &phys[i];
1971         d40c->log_num = log_num;
1972 out:
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
1983 static 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
2013 static 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
2073 static 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
2119 static 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
2133 static int
2134 d40_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
2161 static int
2162 d40_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
2194 static struct d40_desc *
2195 d40_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
2232 static struct dma_async_tx_descriptor *
2233 d40_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
2296 static 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
2316 static 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
2349 static 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
2369 static 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 */
2412 static 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
2477 static 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
2498 static 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
2520 static struct dma_async_tx_descriptor *
2521 d40_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
2531 static struct dma_async_tx_descriptor *
2532 dma40_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
2561 static 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
2583 static 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
2604 static 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
2635 static int
2636 dma40_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
2668 static 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 */
2679 static 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
2805 static 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
2837 static 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
2869 static 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
2929 static 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
2943 static 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
2959 static 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
2974 static 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
3008 static 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
3022 static 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
3034 static 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
3043 static 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 */
3128 static 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
3135 static 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
3314 static 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
3368 static 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
3451 static 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
3502 static 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
3672  report_failure:
3673         d40_err(dev, "probe failed\n");
3674         return ret;
3675 }
3676
3677 static const struct of_device_id d40_match[] = {
3678         { .compatible = "stericsson,dma40", },
3679         {}
3680 };
3681
3682 static struct platform_driver d40_driver = {
3683         .driver = {
3684                 .name  = D40_NAME,
3685                 .pm = &dma40_pm_ops,
3686                 .of_match_table = d40_match,
3687         },
3688 };
3689
3690 static int __init stedma40_init(void)
3691 {
3692         return platform_driver_probe(&d40_driver, d40_probe);
3693 }
3694 subsys_initcall(stedma40_init);