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_dma.h>
  23 #include <linux/amba/bus.h>
  24 #include <linux/regulator/consumer.h>
  25 #include <linux/platform_data/dma-ste-dma40.h>
  26
  27 #include "dmaengine.h"
  28 #include "ste_dma40_ll.h"
  29
  30 #define D40_NAME "dma40"
  31
  32 #define D40_PHY_CHAN -1
  33
  34 /* For masking out/in 2 bit channel positions */
  35 #define D40_CHAN_POS(chan)  (2 * (chan / 2))
  36 #define D40_CHAN_POS_MASK(chan) (0x3 << D40_CHAN_POS(chan))
  37
  38 /* Maximum iterations taken before giving up suspending a channel */
  39 #define D40_SUSPEND_MAX_IT 500
  40
  41 /* Milliseconds */
  42 #define DMA40_AUTOSUSPEND_DELAY 100
  43
  44 /* Hardware requirement on LCLA alignment */
  45 #define LCLA_ALIGNMENT 0x40000
  46
  47 /* Max number of links per event group */
  48 #define D40_LCLA_LINK_PER_EVENT_GRP 128
  49 #define D40_LCLA_END D40_LCLA_LINK_PER_EVENT_GRP
  50
  51 /* Max number of logical channels per physical channel */
  52 #define D40_MAX_LOG_CHAN_PER_PHY 32
  53
  54 /* Attempts before giving up to trying to get pages that are aligned */
  55 #define MAX_LCLA_ALLOC_ATTEMPTS 256
  56
  57 /* Bit markings for allocation map */
  58 #define D40_ALLOC_FREE          BIT(31)
  59 #define D40_ALLOC_PHY           BIT(30)
  60 #define D40_ALLOC_LOG_FREE      0
  61
  62 #define D40_MEMCPY_MAX_CHANS    8
  63
  64 /* Reserved event lines for memcpy only. */
  65 #define DB8500_DMA_MEMCPY_EV_0  51
  66 #define DB8500_DMA_MEMCPY_EV_1  56
  67 #define DB8500_DMA_MEMCPY_EV_2  57
  68 #define DB8500_DMA_MEMCPY_EV_3  58
  69 #define DB8500_DMA_MEMCPY_EV_4  59
  70 #define DB8500_DMA_MEMCPY_EV_5  60
  71
  72 static int dma40_memcpy_channels[] = {
  73         DB8500_DMA_MEMCPY_EV_0,
  74         DB8500_DMA_MEMCPY_EV_1,
  75         DB8500_DMA_MEMCPY_EV_2,
  76         DB8500_DMA_MEMCPY_EV_3,
  77         DB8500_DMA_MEMCPY_EV_4,
  78         DB8500_DMA_MEMCPY_EV_5,
  79 };
  80
  81 /* Default configuration for physical memcpy */
  82 static const struct stedma40_chan_cfg dma40_memcpy_conf_phy = {
  83         .mode = STEDMA40_MODE_PHYSICAL,
  84         .dir = DMA_MEM_TO_MEM,
  85
  86         .src_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
  87         .src_info.psize = STEDMA40_PSIZE_PHY_1,
  88         .src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
  89
  90         .dst_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
  91         .dst_info.psize = STEDMA40_PSIZE_PHY_1,
  92         .dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
  93 };
  94
  95 /* Default configuration for logical memcpy */
  96 static const struct stedma40_chan_cfg dma40_memcpy_conf_log = {
  97         .mode = STEDMA40_MODE_LOGICAL,
  98         .dir = DMA_MEM_TO_MEM,
  99
 100         .src_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
 101         .src_info.psize = STEDMA40_PSIZE_LOG_1,
 102         .src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
 103
 104         .dst_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
 105         .dst_info.psize = STEDMA40_PSIZE_LOG_1,
 106         .dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
 107 };
 108
 109 /**
 110  * enum 40_command - The different commands and/or statuses.
 111  *
 112  * @D40_DMA_STOP: DMA channel command STOP or status STOPPED,
 113  * @D40_DMA_RUN: The DMA channel is RUNNING of the command RUN.
 114  * @D40_DMA_SUSPEND_REQ: Request the DMA to SUSPEND as soon as possible.
 115  * @D40_DMA_SUSPENDED: The DMA channel is SUSPENDED.
 116  */
 117 enum d40_command {
 118         D40_DMA_STOP            = 0,
 119         D40_DMA_RUN             = 1,
 120         D40_DMA_SUSPEND_REQ     = 2,
 121         D40_DMA_SUSPENDED       = 3
 122 };
 123
 124 /*
 125  * enum d40_events - The different Event Enables for the event lines.
 126  *
 127  * @D40_DEACTIVATE_EVENTLINE: De-activate Event line, stopping the logical chan.
 128  * @D40_ACTIVATE_EVENTLINE: Activate the Event line, to start a logical chan.
 129  * @D40_SUSPEND_REQ_EVENTLINE: Requesting for suspending a event line.
 130  * @D40_ROUND_EVENTLINE: Status check for event line.
 131  */
 132
 133 enum d40_events {
 134         D40_DEACTIVATE_EVENTLINE        = 0,
 135         D40_ACTIVATE_EVENTLINE          = 1,
 136         D40_SUSPEND_REQ_EVENTLINE       = 2,
 137         D40_ROUND_EVENTLINE             = 3
 138 };
 139
 140 /*
 141  * These are the registers that has to be saved and later restored
 142  * when the DMA hw is powered off.
 143  * TODO: Add save/restore of D40_DREG_GCC on dma40 v3 or later, if that works.
 144  */
 145 static __maybe_unused u32 d40_backup_regs[] = {
 146         D40_DREG_LCPA,
 147         D40_DREG_LCLA,
 148         D40_DREG_PRMSE,
 149         D40_DREG_PRMSO,
 150         D40_DREG_PRMOE,
 151         D40_DREG_PRMOO,
 152 };
 153
 154 #define BACKUP_REGS_SZ ARRAY_SIZE(d40_backup_regs)
 155
 156 /*
 157  * since 9540 and 8540 has the same HW revision
 158  * use v4a for 9540 or ealier
 159  * use v4b for 8540 or later
 160  * HW revision:
 161  * DB8500ed has revision 0
 162  * DB8500v1 has revision 2
 163  * DB8500v2 has revision 3
 164  * AP9540v1 has revision 4
 165  * DB8540v1 has revision 4
 166  * TODO: Check if all these registers have to be saved/restored on dma40 v4a
 167  */
 168 static u32 d40_backup_regs_v4a[] = {
 169         D40_DREG_PSEG1,
 170         D40_DREG_PSEG2,
 171         D40_DREG_PSEG3,
 172         D40_DREG_PSEG4,
 173         D40_DREG_PCEG1,
 174         D40_DREG_PCEG2,
 175         D40_DREG_PCEG3,
 176         D40_DREG_PCEG4,
 177         D40_DREG_RSEG1,
 178         D40_DREG_RSEG2,
 179         D40_DREG_RSEG3,
 180         D40_DREG_RSEG4,
 181         D40_DREG_RCEG1,
 182         D40_DREG_RCEG2,
 183         D40_DREG_RCEG3,
 184         D40_DREG_RCEG4,
 185 };
 186
 187 #define BACKUP_REGS_SZ_V4A ARRAY_SIZE(d40_backup_regs_v4a)
 188
 189 static u32 d40_backup_regs_v4b[] = {
 190         D40_DREG_CPSEG1,
 191         D40_DREG_CPSEG2,
 192         D40_DREG_CPSEG3,
 193         D40_DREG_CPSEG4,
 194         D40_DREG_CPSEG5,
 195         D40_DREG_CPCEG1,
 196         D40_DREG_CPCEG2,
 197         D40_DREG_CPCEG3,
 198         D40_DREG_CPCEG4,
 199         D40_DREG_CPCEG5,
 200         D40_DREG_CRSEG1,
 201         D40_DREG_CRSEG2,
 202         D40_DREG_CRSEG3,
 203         D40_DREG_CRSEG4,
 204         D40_DREG_CRSEG5,
 205         D40_DREG_CRCEG1,
 206         D40_DREG_CRCEG2,
 207         D40_DREG_CRCEG3,
 208         D40_DREG_CRCEG4,
 209         D40_DREG_CRCEG5,
 210 };
 211
 212 #define BACKUP_REGS_SZ_V4B ARRAY_SIZE(d40_backup_regs_v4b)
 213
 214 static __maybe_unused u32 d40_backup_regs_chan[] = {
 215         D40_CHAN_REG_SSCFG,
 216         D40_CHAN_REG_SSELT,
 217         D40_CHAN_REG_SSPTR,
 218         D40_CHAN_REG_SSLNK,
 219         D40_CHAN_REG_SDCFG,
 220         D40_CHAN_REG_SDELT,
 221         D40_CHAN_REG_SDPTR,
 222         D40_CHAN_REG_SDLNK,
 223 };
 224
 225 #define BACKUP_REGS_SZ_MAX ((BACKUP_REGS_SZ_V4A > BACKUP_REGS_SZ_V4B) ? \
 226                              BACKUP_REGS_SZ_V4A : BACKUP_REGS_SZ_V4B)
 227
 228 /**
 229  * struct d40_interrupt_lookup - lookup table for interrupt handler
 230  *
 231  * @src: Interrupt mask register.
 232  * @clr: Interrupt clear register.
 233  * @is_error: true if this is an error interrupt.
 234  * @offset: start delta in the lookup_log_chans in d40_base. If equals to
 235  * D40_PHY_CHAN, the lookup_phy_chans shall be used instead.
 236  */
 237 struct d40_interrupt_lookup {
 238         u32 src;
 239         u32 clr;
 240         bool is_error;
 241         int offset;
 242 };
 243
 244
 245 static struct d40_interrupt_lookup il_v4a[] = {
 246         {D40_DREG_LCTIS0, D40_DREG_LCICR0, false,  0},
 247         {D40_DREG_LCTIS1, D40_DREG_LCICR1, false, 32},
 248         {D40_DREG_LCTIS2, D40_DREG_LCICR2, false, 64},
 249         {D40_DREG_LCTIS3, D40_DREG_LCICR3, false, 96},
 250         {D40_DREG_LCEIS0, D40_DREG_LCICR0, true,   0},
 251         {D40_DREG_LCEIS1, D40_DREG_LCICR1, true,  32},
 252         {D40_DREG_LCEIS2, D40_DREG_LCICR2, true,  64},
 253         {D40_DREG_LCEIS3, D40_DREG_LCICR3, true,  96},
 254         {D40_DREG_PCTIS,  D40_DREG_PCICR,  false, D40_PHY_CHAN},
 255         {D40_DREG_PCEIS,  D40_DREG_PCICR,  true,  D40_PHY_CHAN},
 256 };
 257
 258 static struct d40_interrupt_lookup il_v4b[] = {
 259         {D40_DREG_CLCTIS1, D40_DREG_CLCICR1, false,  0},
 260         {D40_DREG_CLCTIS2, D40_DREG_CLCICR2, false, 32},
 261         {D40_DREG_CLCTIS3, D40_DREG_CLCICR3, false, 64},
 262         {D40_DREG_CLCTIS4, D40_DREG_CLCICR4, false, 96},
 263         {D40_DREG_CLCTIS5, D40_DREG_CLCICR5, false, 128},
 264         {D40_DREG_CLCEIS1, D40_DREG_CLCICR1, true,   0},
 265         {D40_DREG_CLCEIS2, D40_DREG_CLCICR2, true,  32},
 266         {D40_DREG_CLCEIS3, D40_DREG_CLCICR3, true,  64},
 267         {D40_DREG_CLCEIS4, D40_DREG_CLCICR4, true,  96},
 268         {D40_DREG_CLCEIS5, D40_DREG_CLCICR5, true,  128},
 269         {D40_DREG_CPCTIS,  D40_DREG_CPCICR,  false, D40_PHY_CHAN},
 270         {D40_DREG_CPCEIS,  D40_DREG_CPCICR,  true,  D40_PHY_CHAN},
 271 };
 272
 273 /**
 274  * struct d40_reg_val - simple lookup struct
 275  *
 276  * @reg: The register.
 277  * @val: The value that belongs to the register in reg.
 278  */
 279 struct d40_reg_val {
 280         unsigned int reg;
 281         unsigned int val;
 282 };
 283
 284 static __initdata struct d40_reg_val dma_init_reg_v4a[] = {
 285         /* Clock every part of the DMA block from start */
 286         { .reg = D40_DREG_GCC,    .val = D40_DREG_GCC_ENABLE_ALL},
 287
 288         /* Interrupts on all logical channels */
 289         { .reg = D40_DREG_LCMIS0, .val = 0xFFFFFFFF},
 290         { .reg = D40_DREG_LCMIS1, .val = 0xFFFFFFFF},
 291         { .reg = D40_DREG_LCMIS2, .val = 0xFFFFFFFF},
 292         { .reg = D40_DREG_LCMIS3, .val = 0xFFFFFFFF},
 293         { .reg = D40_DREG_LCICR0, .val = 0xFFFFFFFF},
 294         { .reg = D40_DREG_LCICR1, .val = 0xFFFFFFFF},
 295         { .reg = D40_DREG_LCICR2, .val = 0xFFFFFFFF},
 296         { .reg = D40_DREG_LCICR3, .val = 0xFFFFFFFF},
 297         { .reg = D40_DREG_LCTIS0, .val = 0xFFFFFFFF},
 298         { .reg = D40_DREG_LCTIS1, .val = 0xFFFFFFFF},
 299         { .reg = D40_DREG_LCTIS2, .val = 0xFFFFFFFF},
 300         { .reg = D40_DREG_LCTIS3, .val = 0xFFFFFFFF}
 301 };
 302 static __initdata struct d40_reg_val dma_init_reg_v4b[] = {
 303         /* Clock every part of the DMA block from start */
 304         { .reg = D40_DREG_GCC,    .val = D40_DREG_GCC_ENABLE_ALL},
 305
 306         /* Interrupts on all logical channels */
 307         { .reg = D40_DREG_CLCMIS1, .val = 0xFFFFFFFF},
 308         { .reg = D40_DREG_CLCMIS2, .val = 0xFFFFFFFF},
 309         { .reg = D40_DREG_CLCMIS3, .val = 0xFFFFFFFF},
 310         { .reg = D40_DREG_CLCMIS4, .val = 0xFFFFFFFF},
 311         { .reg = D40_DREG_CLCMIS5, .val = 0xFFFFFFFF},
 312         { .reg = D40_DREG_CLCICR1, .val = 0xFFFFFFFF},
 313         { .reg = D40_DREG_CLCICR2, .val = 0xFFFFFFFF},
 314         { .reg = D40_DREG_CLCICR3, .val = 0xFFFFFFFF},
 315         { .reg = D40_DREG_CLCICR4, .val = 0xFFFFFFFF},
 316         { .reg = D40_DREG_CLCICR5, .val = 0xFFFFFFFF},
 317         { .reg = D40_DREG_CLCTIS1, .val = 0xFFFFFFFF},
 318         { .reg = D40_DREG_CLCTIS2, .val = 0xFFFFFFFF},
 319         { .reg = D40_DREG_CLCTIS3, .val = 0xFFFFFFFF},
 320         { .reg = D40_DREG_CLCTIS4, .val = 0xFFFFFFFF},
 321         { .reg = D40_DREG_CLCTIS5, .val = 0xFFFFFFFF}
 322 };
 323
 324 /**
 325  * struct d40_lli_pool - Structure for keeping LLIs in memory
 326  *
 327  * @base: Pointer to memory area when the pre_alloc_lli's are not large
 328  * enough, IE bigger than the most common case, 1 dst and 1 src. NULL if
 329  * pre_alloc_lli is used.
 330  * @dma_addr: DMA address, if mapped
 331  * @size: The size in bytes of the memory at base or the size of pre_alloc_lli.
 332  * @pre_alloc_lli: Pre allocated area for the most common case of transfers,
 333  * one buffer to one buffer.
 334  */
 335 struct d40_lli_pool {
 336         void    *base;
 337         int      size;
 338         dma_addr_t      dma_addr;
 339         /* Space for dst and src, plus an extra for padding */
 340         u8       pre_alloc_lli[3 * sizeof(struct d40_phy_lli)];
 341 };
 342
 343 /**
 344  * struct d40_desc - A descriptor is one DMA job.
 345  *
 346  * @lli_phy: LLI settings for physical channel. Both src and dst=
 347  * points into the lli_pool, to base if lli_len > 1 or to pre_alloc_lli if
 348  * lli_len equals one.
 349  * @lli_log: Same as above but for logical channels.
 350  * @lli_pool: The pool with two entries pre-allocated.
 351  * @lli_len: Number of llis of current descriptor.
 352  * @lli_current: Number of transferred llis.
 353  * @lcla_alloc: Number of LCLA entries allocated.
 354  * @txd: DMA engine struct. Used for among other things for communication
 355  * during a transfer.
 356  * @node: List entry.
 357  * @is_in_client_list: true if the client owns this descriptor.
 358  * @cyclic: true if this is a cyclic job
 359  *
 360  * This descriptor is used for both logical and physical transfers.
 361  */
 362 struct d40_desc {
 363         /* LLI physical */
 364         struct d40_phy_lli_bidir         lli_phy;
 365         /* LLI logical */
 366         struct d40_log_lli_bidir         lli_log;
 367
 368         struct d40_lli_pool              lli_pool;
 369         int                              lli_len;
 370         int                              lli_current;
 371         int                              lcla_alloc;
 372
 373         struct dma_async_tx_descriptor   txd;
 374         struct list_head                 node;
 375
 376         bool                             is_in_client_list;
 377         bool                             cyclic;
 378 };
 379
 380 /**
 381  * struct d40_lcla_pool - LCLA pool settings and data.
 382  *
 383  * @base: The virtual address of LCLA. 18 bit aligned.
 384  * @dma_addr: DMA address, if mapped
 385  * @base_unaligned: The orignal kmalloc pointer, if kmalloc is used.
 386  * This pointer is only there for clean-up on error.
 387  * @pages: The number of pages needed for all physical channels.
 388  * Only used later for clean-up on error
 389  * @lock: Lock to protect the content in this struct.
 390  * @alloc_map: big map over which LCLA entry is own by which job.
 391  */
 392 struct d40_lcla_pool {
 393         void            *base;
 394         dma_addr_t      dma_addr;
 395         void            *base_unaligned;
 396         int              pages;
 397         spinlock_t       lock;
 398         struct d40_desc **alloc_map;
 399 };
 400
 401 /**
 402  * struct d40_phy_res - struct for handling eventlines mapped to physical
 403  * channels.
 404  *
 405  * @lock: A lock protection this entity.
 406  * @reserved: True if used by secure world or otherwise.
 407  * @num: The physical channel number of this entity.
 408  * @allocated_src: Bit mapped to show which src event line's are mapped to
 409  * this physical channel. Can also be free or physically allocated.
 410  * @allocated_dst: Same as for src but is dst.
 411  * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as
 412  * event line number.
 413  * @use_soft_lli: To mark if the linked lists of channel are managed by SW.
 414  */
 415 struct d40_phy_res {
 416         spinlock_t lock;
 417         bool       reserved;
 418         int        num;
 419         u32        allocated_src;
 420         u32        allocated_dst;
 421         bool       use_soft_lli;
 422 };
 423
 424 struct d40_base;
 425
 426 /**
 427  * struct d40_chan - Struct that describes a channel.
 428  *
 429  * @lock: A spinlock to protect this struct.
 430  * @log_num: The logical number, if any of this channel.
 431  * @pending_tx: The number of pending transfers. Used between interrupt handler
 432  * and tasklet.
 433  * @busy: Set to true when transfer is ongoing on this channel.
 434  * @phy_chan: Pointer to physical channel which this instance runs on. If this
 435  * point is NULL, then the channel is not allocated.
 436  * @chan: DMA engine handle.
 437  * @tasklet: Tasklet that gets scheduled from interrupt context to complete a
 438  * transfer and call client callback.
 439  * @client: Cliented owned descriptor list.
 440  * @pending_queue: Submitted jobs, to be issued by issue_pending()
 441  * @active: Active descriptor.
 442  * @done: Completed jobs
 443  * @queue: Queued jobs.
 444  * @prepare_queue: Prepared jobs.
 445  * @dma_cfg: The client configuration of this dma channel.
 446  * @slave_config: DMA slave configuration.
 447  * @configured: whether the dma_cfg configuration is valid
 448  * @base: Pointer to the device instance struct.
 449  * @src_def_cfg: Default cfg register setting for src.
 450  * @dst_def_cfg: Default cfg register setting for dst.
 451  * @log_def: Default logical channel settings.
 452  * @lcpa: Pointer to dst and src lcpa settings.
 453  * @runtime_addr: runtime configured address.
 454  * @runtime_direction: runtime configured direction.
 455  *
 456  * This struct can either "be" a logical or a physical channel.
 457  */
 458 struct d40_chan {
 459         spinlock_t                       lock;
 460         int                              log_num;
 461         int                              pending_tx;
 462         bool                             busy;
 463         struct d40_phy_res              *phy_chan;
 464         struct dma_chan                  chan;
 465         struct tasklet_struct            tasklet;
 466         struct list_head                 client;
 467         struct list_head                 pending_queue;
 468         struct list_head                 active;
 469         struct list_head                 done;
 470         struct list_head                 queue;
 471         struct list_head                 prepare_queue;
 472         struct stedma40_chan_cfg         dma_cfg;
 473         struct dma_slave_config          slave_config;
 474         bool                             configured;
 475         struct d40_base                 *base;
 476         /* Default register configurations */
 477         u32                              src_def_cfg;
 478         u32                              dst_def_cfg;
 479         struct d40_def_lcsp              log_def;
 480         struct d40_log_lli_full         *lcpa;
 481         /* Runtime reconfiguration */
 482         dma_addr_t                      runtime_addr;
 483         enum dma_transfer_direction     runtime_direction;
 484 };
 485
 486 /**
 487  * struct d40_gen_dmac - generic values to represent u8500/u8540 DMA
 488  * controller
 489  *
 490  * @backup: the pointer to the registers address array for backup
 491  * @backup_size: the size of the registers address array for backup
 492  * @realtime_en: the realtime enable register
 493  * @realtime_clear: the realtime clear register
 494  * @high_prio_en: the high priority enable register
 495  * @high_prio_clear: the high priority clear register
 496  * @interrupt_en: the interrupt enable register
 497  * @interrupt_clear: the interrupt clear register
 498  * @il: the pointer to struct d40_interrupt_lookup
 499  * @il_size: the size of d40_interrupt_lookup array
 500  * @init_reg: the pointer to the struct d40_reg_val
 501  * @init_reg_size: the size of d40_reg_val array
 502  */
 503 struct d40_gen_dmac {
 504         u32                             *backup;
 505         u32                              backup_size;
 506         u32                              realtime_en;
 507         u32                              realtime_clear;
 508         u32                              high_prio_en;
 509         u32                              high_prio_clear;
 510         u32                              interrupt_en;
 511         u32                              interrupt_clear;
 512         struct d40_interrupt_lookup     *il;
 513         u32                              il_size;
 514         struct d40_reg_val              *init_reg;
 515         u32                              init_reg_size;
 516 };
 517
 518 /**
 519  * struct d40_base - The big global struct, one for each probe'd instance.
 520  *
 521  * @interrupt_lock: Lock used to make sure one interrupt is handle a time.
 522  * @execmd_lock: Lock for execute command usage since several channels share
 523  * the same physical register.
 524  * @dev: The device structure.
 525  * @virtbase: The virtual base address of the DMA's register.
 526  * @rev: silicon revision detected.
 527  * @clk: Pointer to the DMA clock structure.
 528  * @phy_start: Physical memory start of the DMA registers.
 529  * @phy_size: Size of the DMA register map.
 530  * @irq: The IRQ number.
 531  * @num_memcpy_chans: The number of channels used for memcpy (mem-to-mem
 532  * transfers).
 533  * @num_phy_chans: The number of physical channels. Read from HW. This
 534  * is the number of available channels for this driver, not counting "Secure
 535  * mode" allocated physical channels.
 536  * @num_log_chans: The number of logical channels. Calculated from
 537  * num_phy_chans.
 538  * @dma_both: dma_device channels that can do both memcpy and slave transfers.
 539  * @dma_slave: dma_device channels that can do only do slave transfers.
 540  * @dma_memcpy: dma_device channels that can do only do memcpy transfers.
 541  * @phy_chans: Room for all possible physical channels in system.
 542  * @log_chans: Room for all possible logical channels in system.
 543  * @lookup_log_chans: Used to map interrupt number to logical channel. Points
 544  * to log_chans entries.
 545  * @lookup_phy_chans: Used to map interrupt number to physical channel. Points
 546  * to phy_chans entries.
 547  * @plat_data: Pointer to provided platform_data which is the driver
 548  * configuration.
 549  * @lcpa_regulator: Pointer to hold the regulator for the esram bank for lcla.
 550  * @phy_res: Vector containing all physical channels.
 551  * @lcla_pool: lcla pool settings and data.
 552  * @lcpa_base: The virtual mapped address of LCPA.
 553  * @phy_lcpa: The physical address of the LCPA.
 554  * @lcpa_size: The size of the LCPA area.
 555  * @desc_slab: cache for descriptors.
 556  * @reg_val_backup: Here the values of some hardware registers are stored
 557  * before the DMA is powered off. They are restored when the power is back on.
 558  * @reg_val_backup_v4: Backup of registers that only exits on dma40 v3 and
 559  * later
 560  * @reg_val_backup_chan: Backup data for standard channel parameter registers.
 561  * @regs_interrupt: Scratch space for registers during interrupt.
 562  * @gcc_pwr_off_mask: Mask to maintain the channels that can be turned off.
 563  * @gen_dmac: the struct for generic registers values to represent u8500/8540
 564  * DMA controller
 565  */
 566 struct d40_base {
 567         spinlock_t                       interrupt_lock;
 568         spinlock_t                       execmd_lock;
 569         struct device                    *dev;
 570         void __iomem                     *virtbase;
 571         u8                                rev:4;
 572         struct clk                       *clk;
 573         phys_addr_t                       phy_start;
 574         resource_size_t                   phy_size;
 575         int                               irq;
 576         int                               num_memcpy_chans;
 577         int                               num_phy_chans;
 578         int                               num_log_chans;
 579         struct dma_device                 dma_both;
 580         struct dma_device                 dma_slave;
 581         struct dma_device                 dma_memcpy;
 582         struct d40_chan                  *phy_chans;
 583         struct d40_chan                  *log_chans;
 584         struct d40_chan                 **lookup_log_chans;
 585         struct d40_chan                 **lookup_phy_chans;
 586         struct stedma40_platform_data    *plat_data;
 587         struct regulator                 *lcpa_regulator;
 588         /* Physical half channels */
 589         struct d40_phy_res               *phy_res;
 590         struct d40_lcla_pool              lcla_pool;
 591         void                             *lcpa_base;
 592         dma_addr_t                        phy_lcpa;
 593         resource_size_t                   lcpa_size;
 594         struct kmem_cache                *desc_slab;
 595         u32                               reg_val_backup[BACKUP_REGS_SZ];
 596         u32                               reg_val_backup_v4[BACKUP_REGS_SZ_MAX];
 597         u32                              *reg_val_backup_chan;
 598         u32                              *regs_interrupt;
 599         u16                               gcc_pwr_off_mask;
 600         struct d40_gen_dmac               gen_dmac;
 601 };
 602
 603 static struct device *chan2dev(struct d40_chan *d40c)
 604 {
 605         return &d40c->chan.dev->device;
 606 }
 607
 608 static bool chan_is_physical(struct d40_chan *chan)
 609 {
 610         return chan->log_num == D40_PHY_CHAN;
 611 }
 612
 613 static bool chan_is_logical(struct d40_chan *chan)
 614 {
 615         return !chan_is_physical(chan);
 616 }
 617
 618 static void __iomem *chan_base(struct d40_chan *chan)
 619 {
 620         return chan->base->virtbase + D40_DREG_PCBASE +
 621                chan->phy_chan->num * D40_DREG_PCDELTA;
 622 }
 623
 624 #define d40_err(dev, format, arg...)            \
 625         dev_err(dev, "[%s] " format, __func__, ## arg)
 626
 627 #define chan_err(d40c, format, arg...)          \
 628         d40_err(chan2dev(d40c), format, ## arg)
 629
 630 static int d40_set_runtime_config_write(struct dma_chan *chan,
 631                                   struct dma_slave_config *config,
 632                                   enum dma_transfer_direction direction);
 633
 634 static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d,
 635                               int lli_len)
 636 {
 637         bool is_log = chan_is_logical(d40c);
 638         u32 align;
 639         void *base;
 640
 641         if (is_log)
 642                 align = sizeof(struct d40_log_lli);
 643         else
 644                 align = sizeof(struct d40_phy_lli);
 645
 646         if (lli_len == 1) {
 647                 base = d40d->lli_pool.pre_alloc_lli;
 648                 d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
 649                 d40d->lli_pool.base = NULL;
 650         } else {
 651                 d40d->lli_pool.size = lli_len * 2 * align;
 652
 653                 base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
 654                 d40d->lli_pool.base = base;
 655
 656                 if (d40d->lli_pool.base == NULL)
 657                         return -ENOMEM;
 658         }
 659
 660         if (is_log) {
 661                 d40d->lli_log.src = PTR_ALIGN(base, align);
 662                 d40d->lli_log.dst = d40d->lli_log.src + lli_len;
 663
 664                 d40d->lli_pool.dma_addr = 0;
 665         } else {
 666                 d40d->lli_phy.src = PTR_ALIGN(base, align);
 667                 d40d->lli_phy.dst = d40d->lli_phy.src + lli_len;
 668
 669                 d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev,
 670                                                          d40d->lli_phy.src,
 671                                                          d40d->lli_pool.size,
 672                                                          DMA_TO_DEVICE);
 673
 674                 if (dma_mapping_error(d40c->base->dev,
 675                                       d40d->lli_pool.dma_addr)) {
 676                         kfree(d40d->lli_pool.base);
 677                         d40d->lli_pool.base = NULL;
 678                         d40d->lli_pool.dma_addr = 0;
 679                         return -ENOMEM;
 680                 }
 681         }
 682
 683         return 0;
 684 }
 685
 686 static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d)
 687 {
 688         if (d40d->lli_pool.dma_addr)
 689                 dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr,
 690                                  d40d->lli_pool.size, DMA_TO_DEVICE);
 691
 692         kfree(d40d->lli_pool.base);
 693         d40d->lli_pool.base = NULL;
 694         d40d->lli_pool.size = 0;
 695         d40d->lli_log.src = NULL;
 696         d40d->lli_log.dst = NULL;
 697         d40d->lli_phy.src = NULL;
 698         d40d->lli_phy.dst = NULL;
 699 }
 700
 701 static int d40_lcla_alloc_one(struct d40_chan *d40c,
 702                               struct d40_desc *d40d)
 703 {
 704         unsigned long flags;
 705         int i;
 706         int ret = -EINVAL;
 707
 708         spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
 709
 710         /*
 711          * Allocate both src and dst at the same time, therefore the half
 712          * start on 1 since 0 can't be used since zero is used as end marker.
 713          */
 714         for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
 715                 int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
 716
 717                 if (!d40c->base->lcla_pool.alloc_map[idx]) {
 718                         d40c->base->lcla_pool.alloc_map[idx] = d40d;
 719                         d40d->lcla_alloc++;
 720                         ret = i;
 721                         break;
 722                 }
 723         }
 724
 725         spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
 726
 727         return ret;
 728 }
 729
 730 static int d40_lcla_free_all(struct d40_chan *d40c,
 731                              struct d40_desc *d40d)
 732 {
 733         unsigned long flags;
 734         int i;
 735         int ret = -EINVAL;
 736
 737         if (chan_is_physical(d40c))
 738                 return 0;
 739
 740         spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
 741
 742         for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
 743                 int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
 744
 745                 if (d40c->base->lcla_pool.alloc_map[idx] == d40d) {
 746                         d40c->base->lcla_pool.alloc_map[idx] = NULL;
 747                         d40d->lcla_alloc--;
 748                         if (d40d->lcla_alloc == 0) {
 749                                 ret = 0;
 750                                 break;
 751                         }
 752                 }
 753         }
 754
 755         spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
 756
 757         return ret;
 758
 759 }
 760
 761 static void d40_desc_remove(struct d40_desc *d40d)
 762 {
 763         list_del(&d40d->node);
 764 }
 765
 766 static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
 767 {
 768         struct d40_desc *desc = NULL;
 769
 770         if (!list_empty(&d40c->client)) {
 771                 struct d40_desc *d;
 772                 struct d40_desc *_d;
 773
 774                 list_for_each_entry_safe(d, _d, &d40c->client, node) {
 775                         if (async_tx_test_ack(&d->txd)) {
 776                                 d40_desc_remove(d);
 777                                 desc = d;
 778                                 memset(desc, 0, sizeof(*desc));
 779                                 break;
 780                         }
 781                 }
 782         }
 783
 784         if (!desc)
 785                 desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT);
 786
 787         if (desc)
 788                 INIT_LIST_HEAD(&desc->node);
 789
 790         return desc;
 791 }
 792
 793 static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
 794 {
 795
 796         d40_pool_lli_free(d40c, d40d);
 797         d40_lcla_free_all(d40c, d40d);
 798         kmem_cache_free(d40c->base->desc_slab, d40d);
 799 }
 800
 801 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
 802 {
 803         list_add_tail(&desc->node, &d40c->active);
 804 }
 805
 806 static void d40_phy_lli_load(struct d40_chan *chan, struct d40_desc *desc)
 807 {
 808         struct d40_phy_lli *lli_dst = desc->lli_phy.dst;
 809         struct d40_phy_lli *lli_src = desc->lli_phy.src;
 810         void __iomem *base = chan_base(chan);
 811
 812         writel(lli_src->reg_cfg, base + D40_CHAN_REG_SSCFG);
 813         writel(lli_src->reg_elt, base + D40_CHAN_REG_SSELT);
 814         writel(lli_src->reg_ptr, base + D40_CHAN_REG_SSPTR);
 815         writel(lli_src->reg_lnk, base + D40_CHAN_REG_SSLNK);
 816
 817         writel(lli_dst->reg_cfg, base + D40_CHAN_REG_SDCFG);
 818         writel(lli_dst->reg_elt, base + D40_CHAN_REG_SDELT);
 819         writel(lli_dst->reg_ptr, base + D40_CHAN_REG_SDPTR);
 820         writel(lli_dst->reg_lnk, base + D40_CHAN_REG_SDLNK);
 821 }
 822
 823 static void d40_desc_done(struct d40_chan *d40c, struct d40_desc *desc)
 824 {
 825         list_add_tail(&desc->node, &d40c->done);
 826 }
 827
 828 static void d40_log_lli_to_lcxa(struct d40_chan *chan, struct d40_desc *desc)
 829 {
 830         struct d40_lcla_pool *pool = &chan->base->lcla_pool;
 831         struct d40_log_lli_bidir *lli = &desc->lli_log;
 832         int lli_current = desc->lli_current;
 833         int lli_len = desc->lli_len;
 834         bool cyclic = desc->cyclic;
 835         int curr_lcla = -EINVAL;
 836         int first_lcla = 0;
 837         bool use_esram_lcla = chan->base->plat_data->use_esram_lcla;
 838         bool linkback;
 839
 840         /*
 841          * We may have partially running cyclic transfers, in case we did't get
 842          * enough LCLA entries.
 843          */
 844         linkback = cyclic && lli_current == 0;
 845
 846         /*
 847          * For linkback, we need one LCLA even with only one link, because we
 848          * can't link back to the one in LCPA space
 849          */
 850         if (linkback || (lli_len - lli_current > 1)) {
 851                 /*
 852                  * If the channel is expected to use only soft_lli don't
 853                  * allocate a lcla. This is to avoid a HW issue that exists
 854                  * in some controller during a peripheral to memory transfer
 855                  * that uses linked lists.
 856                  */
 857                 if (!(chan->phy_chan->use_soft_lli &&
 858                         chan->dma_cfg.dir == DMA_DEV_TO_MEM))
 859                         curr_lcla = d40_lcla_alloc_one(chan, desc);
 860
 861                 first_lcla = curr_lcla;
 862         }
 863
 864         /*
 865          * For linkback, we normally load the LCPA in the loop since we need to
 866          * link it to the second LCLA and not the first.  However, if we
 867          * couldn't even get a first LCLA, then we have to run in LCPA and
 868          * reload manually.
 869          */
 870         if (!linkback || curr_lcla == -EINVAL) {
 871                 unsigned int flags = 0;
 872
 873                 if (curr_lcla == -EINVAL)
 874                         flags |= LLI_TERM_INT;
 875
 876                 d40_log_lli_lcpa_write(chan->lcpa,
 877                                        &lli->dst[lli_current],
 878                                        &lli->src[lli_current],
 879                                        curr_lcla,
 880                                        flags);
 881                 lli_current++;
 882         }
 883
 884         if (curr_lcla < 0)
 885                 goto set_current;
 886
 887         for (; lli_current < lli_len; lli_current++) {
 888                 unsigned int lcla_offset = chan->phy_chan->num * 1024 +
 889                                            8 * curr_lcla * 2;
 890                 struct d40_log_lli *lcla = pool->base + lcla_offset;
 891                 unsigned int flags = 0;
 892                 int next_lcla;
 893
 894                 if (lli_current + 1 < lli_len)
 895                         next_lcla = d40_lcla_alloc_one(chan, desc);
 896                 else
 897                         next_lcla = linkback ? first_lcla : -EINVAL;
 898
 899                 if (cyclic || next_lcla == -EINVAL)
 900                         flags |= LLI_TERM_INT;
 901
 902                 if (linkback && curr_lcla == first_lcla) {
 903                         /* First link goes in both LCPA and LCLA */
 904                         d40_log_lli_lcpa_write(chan->lcpa,
 905                                                &lli->dst[lli_current],
 906                                                &lli->src[lli_current],
 907                                                next_lcla, flags);
 908                 }
 909
 910                 /*
 911                  * One unused LCLA in the cyclic case if the very first
 912                  * next_lcla fails...
 913                  */
 914                 d40_log_lli_lcla_write(lcla,
 915                                        &lli->dst[lli_current],
 916                                        &lli->src[lli_current],
 917                                        next_lcla, flags);
 918
 919                 /*
 920                  * Cache maintenance is not needed if lcla is
 921                  * mapped in esram
 922                  */
 923                 if (!use_esram_lcla) {
 924                         dma_sync_single_range_for_device(chan->base->dev,
 925                                                 pool->dma_addr, lcla_offset,
 926                                                 2 * sizeof(struct d40_log_lli),
 927                                                 DMA_TO_DEVICE);
 928                 }
 929                 curr_lcla = next_lcla;
 930
 931                 if (curr_lcla == -EINVAL || curr_lcla == first_lcla) {
 932                         lli_current++;
 933                         break;
 934                 }
 935         }
 936  set_current:
 937         desc->lli_current = lli_current;
 938 }
 939
 940 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
 941 {
 942         if (chan_is_physical(d40c)) {
 943                 d40_phy_lli_load(d40c, d40d);
 944                 d40d->lli_current = d40d->lli_len;
 945         } else
 946                 d40_log_lli_to_lcxa(d40c, d40d);
 947 }
 948
 949 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
 950 {
 951         return list_first_entry_or_null(&d40c->active, struct d40_desc, node);
 952 }
 953
 954 /* remove desc from current queue and add it to the pending_queue */
 955 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
 956 {
 957         d40_desc_remove(desc);
 958         desc->is_in_client_list = false;
 959         list_add_tail(&desc->node, &d40c->pending_queue);
 960 }
 961
 962 static struct d40_desc *d40_first_pending(struct d40_chan *d40c)
 963 {
 964         return list_first_entry_or_null(&d40c->pending_queue, struct d40_desc,
 965                                         node);
 966 }
 967
 968 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
 969 {
 970         return list_first_entry_or_null(&d40c->queue, struct d40_desc, node);
 971 }
 972
 973 static struct d40_desc *d40_first_done(struct d40_chan *d40c)
 974 {
 975         return list_first_entry_or_null(&d40c->done, struct d40_desc, node);
 976 }
 977
 978 static int d40_psize_2_burst_size(bool is_log, int psize)
 979 {
 980         if (is_log) {
 981                 if (psize == STEDMA40_PSIZE_LOG_1)
 982                         return 1;
 983         } else {
 984                 if (psize == STEDMA40_PSIZE_PHY_1)
 985                         return 1;
 986         }
 987
 988         return 2 << psize;
 989 }
 990
 991 /*
 992  * The dma only supports transmitting packages up to
 993  * STEDMA40_MAX_SEG_SIZE * data_width, where data_width is stored in Bytes.
 994  *
 995  * Calculate the total number of dma elements required to send the entire sg list.
 996  */
 997 static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
 998 {
 999         int dmalen;
1000         u32 max_w = max(data_width1, data_width2);
1001         u32 min_w = min(data_width1, data_width2);
1002         u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE * min_w, max_w);
1003
1004         if (seg_max > STEDMA40_MAX_SEG_SIZE)
1005                 seg_max -= max_w;
1006
1007         if (!IS_ALIGNED(size, max_w))
1008                 return -EINVAL;
1009
1010         if (size <= seg_max)
1011                 dmalen = 1;
1012         else {
1013                 dmalen = size / seg_max;
1014                 if (dmalen * seg_max < size)
1015                         dmalen++;
1016         }
1017         return dmalen;
1018 }
1019
1020 static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
1021                            u32 data_width1, u32 data_width2)
1022 {
1023         struct scatterlist *sg;
1024         int i;
1025         int len = 0;
1026         int ret;
1027
1028         for_each_sg(sgl, sg, sg_len, i) {
1029                 ret = d40_size_2_dmalen(sg_dma_len(sg),
1030                                         data_width1, data_width2);
1031                 if (ret < 0)
1032                         return ret;
1033                 len += ret;
1034         }
1035         return len;
1036 }
1037
1038 static int __d40_execute_command_phy(struct d40_chan *d40c,
1039                                      enum d40_command command)
1040 {
1041         u32 status;
1042         int i;
1043         void __iomem *active_reg;
1044         int ret = 0;
1045         unsigned long flags;
1046         u32 wmask;
1047
1048         if (command == D40_DMA_STOP) {
1049                 ret = __d40_execute_command_phy(d40c, D40_DMA_SUSPEND_REQ);
1050                 if (ret)
1051                         return ret;
1052         }
1053
1054         spin_lock_irqsave(&d40c->base->execmd_lock, flags);
1055
1056         if (d40c->phy_chan->num % 2 == 0)
1057                 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1058         else
1059                 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1060
1061         if (command == D40_DMA_SUSPEND_REQ) {
1062                 status = (readl(active_reg) &
1063                           D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1064                         D40_CHAN_POS(d40c->phy_chan->num);
1065
1066                 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1067                         goto unlock;
1068         }
1069
1070         wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
1071         writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
1072                active_reg);
1073
1074         if (command == D40_DMA_SUSPEND_REQ) {
1075
1076                 for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
1077                         status = (readl(active_reg) &
1078                                   D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1079                                 D40_CHAN_POS(d40c->phy_chan->num);
1080
1081                         cpu_relax();
1082                         /*
1083                          * Reduce the number of bus accesses while
1084                          * waiting for the DMA to suspend.
1085                          */
1086                         udelay(3);
1087
1088                         if (status == D40_DMA_STOP ||
1089                             status == D40_DMA_SUSPENDED)
1090                                 break;
1091                 }
1092
1093                 if (i == D40_SUSPEND_MAX_IT) {
1094                         chan_err(d40c,
1095                                 "unable to suspend the chl %d (log: %d) status %x\n",
1096                                 d40c->phy_chan->num, d40c->log_num,
1097                                 status);
1098                         dump_stack();
1099                         ret = -EBUSY;
1100                 }
1101
1102         }
1103  unlock:
1104         spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
1105         return ret;
1106 }
1107
1108 static void d40_term_all(struct d40_chan *d40c)
1109 {
1110         struct d40_desc *d40d;
1111         struct d40_desc *_d;
1112
1113         /* Release completed descriptors */
1114         while ((d40d = d40_first_done(d40c))) {
1115                 d40_desc_remove(d40d);
1116                 d40_desc_free(d40c, d40d);
1117         }
1118
1119         /* Release active descriptors */
1120         while ((d40d = d40_first_active_get(d40c))) {
1121                 d40_desc_remove(d40d);
1122                 d40_desc_free(d40c, d40d);
1123         }
1124
1125         /* Release queued descriptors waiting for transfer */
1126         while ((d40d = d40_first_queued(d40c))) {
1127                 d40_desc_remove(d40d);
1128                 d40_desc_free(d40c, d40d);
1129         }
1130
1131         /* Release pending descriptors */
1132         while ((d40d = d40_first_pending(d40c))) {
1133                 d40_desc_remove(d40d);
1134                 d40_desc_free(d40c, d40d);
1135         }
1136
1137         /* Release client owned descriptors */
1138         if (!list_empty(&d40c->client))
1139                 list_for_each_entry_safe(d40d, _d, &d40c->client, node) {
1140                         d40_desc_remove(d40d);
1141                         d40_desc_free(d40c, d40d);
1142                 }
1143
1144         /* Release descriptors in prepare queue */
1145         if (!list_empty(&d40c->prepare_queue))
1146                 list_for_each_entry_safe(d40d, _d,
1147                                          &d40c->prepare_queue, node) {
1148                         d40_desc_remove(d40d);
1149                         d40_desc_free(d40c, d40d);
1150                 }
1151
1152         d40c->pending_tx = 0;
1153 }
1154
1155 static void __d40_config_set_event(struct d40_chan *d40c,
1156                                    enum d40_events event_type, u32 event,
1157                                    int reg)
1158 {
1159         void __iomem *addr = chan_base(d40c) + reg;
1160         int tries;
1161         u32 status;
1162
1163         switch (event_type) {
1164
1165         case D40_DEACTIVATE_EVENTLINE:
1166
1167                 writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
1168                        | ~D40_EVENTLINE_MASK(event), addr);
1169                 break;
1170
1171         case D40_SUSPEND_REQ_EVENTLINE:
1172                 status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1173                           D40_EVENTLINE_POS(event);
1174
1175                 if (status == D40_DEACTIVATE_EVENTLINE ||
1176                     status == D40_SUSPEND_REQ_EVENTLINE)
1177                         break;
1178
1179                 writel((D40_SUSPEND_REQ_EVENTLINE << D40_EVENTLINE_POS(event))
1180                        | ~D40_EVENTLINE_MASK(event), addr);
1181
1182                 for (tries = 0 ; tries < D40_SUSPEND_MAX_IT; tries++) {
1183
1184                         status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1185                                   D40_EVENTLINE_POS(event);
1186
1187                         cpu_relax();
1188                         /*
1189                          * Reduce the number of bus accesses while
1190                          * waiting for the DMA to suspend.
1191                          */
1192                         udelay(3);
1193
1194                         if (status == D40_DEACTIVATE_EVENTLINE)
1195                                 break;
1196                 }
1197
1198                 if (tries == D40_SUSPEND_MAX_IT) {
1199                         chan_err(d40c,
1200                                 "unable to stop the event_line chl %d (log: %d)"
1201                                 "status %x\n", d40c->phy_chan->num,
1202                                  d40c->log_num, status);
1203                 }
1204                 break;
1205
1206         case D40_ACTIVATE_EVENTLINE:
1207         /*
1208          * The hardware sometimes doesn't register the enable when src and dst
1209          * event lines are active on the same logical channel.  Retry to ensure
1210          * it does.  Usually only one retry is sufficient.
1211          */
1212                 tries = 100;
1213                 while (--tries) {
1214                         writel((D40_ACTIVATE_EVENTLINE <<
1215                                 D40_EVENTLINE_POS(event)) |
1216                                 ~D40_EVENTLINE_MASK(event), addr);
1217
1218                         if (readl(addr) & D40_EVENTLINE_MASK(event))
1219                                 break;
1220                 }
1221
1222                 if (tries != 99)
1223                         dev_dbg(chan2dev(d40c),
1224                                 "[%s] workaround enable S%cLNK (%d tries)\n",
1225                                 __func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
1226                                 100 - tries);
1227
1228                 WARN_ON(!tries);
1229                 break;
1230
1231         case D40_ROUND_EVENTLINE:
1232                 BUG();
1233                 break;
1234
1235         }
1236 }
1237
1238 static void d40_config_set_event(struct d40_chan *d40c,
1239                                  enum d40_events event_type)
1240 {
1241         u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1242
1243         /* Enable event line connected to device (or memcpy) */
1244         if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) ||
1245             (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
1246                 __d40_config_set_event(d40c, event_type, event,
1247                                        D40_CHAN_REG_SSLNK);
1248
1249         if (d40c->dma_cfg.dir !=  DMA_DEV_TO_MEM)
1250                 __d40_config_set_event(d40c, event_type, event,
1251                                        D40_CHAN_REG_SDLNK);
1252 }
1253
1254 static u32 d40_chan_has_events(struct d40_chan *d40c)
1255 {
1256         void __iomem *chanbase = chan_base(d40c);
1257         u32 val;
1258
1259         val = readl(chanbase + D40_CHAN_REG_SSLNK);
1260         val |= readl(chanbase + D40_CHAN_REG_SDLNK);
1261
1262         return val;
1263 }
1264
1265 static int
1266 __d40_execute_command_log(struct d40_chan *d40c, enum d40_command command)
1267 {
1268         unsigned long flags;
1269         int ret = 0;
1270         u32 active_status;
1271         void __iomem *active_reg;
1272
1273         if (d40c->phy_chan->num % 2 == 0)
1274                 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1275         else
1276                 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1277
1278
1279         spin_lock_irqsave(&d40c->phy_chan->lock, flags);
1280
1281         switch (command) {
1282         case D40_DMA_STOP:
1283         case D40_DMA_SUSPEND_REQ:
1284
1285                 active_status = (readl(active_reg) &
1286                                  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1287                                  D40_CHAN_POS(d40c->phy_chan->num);
1288
1289                 if (active_status == D40_DMA_RUN)
1290                         d40_config_set_event(d40c, D40_SUSPEND_REQ_EVENTLINE);
1291                 else
1292                         d40_config_set_event(d40c, D40_DEACTIVATE_EVENTLINE);
1293
1294                 if (!d40_chan_has_events(d40c) && (command == D40_DMA_STOP))
1295                         ret = __d40_execute_command_phy(d40c, command);
1296
1297                 break;
1298
1299         case D40_DMA_RUN:
1300
1301                 d40_config_set_event(d40c, D40_ACTIVATE_EVENTLINE);
1302                 ret = __d40_execute_command_phy(d40c, command);
1303                 break;
1304
1305         case D40_DMA_SUSPENDED:
1306                 BUG();
1307                 break;
1308         }
1309
1310         spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
1311         return ret;
1312 }
1313
1314 static int d40_channel_execute_command(struct d40_chan *d40c,
1315                                        enum d40_command command)
1316 {
1317         if (chan_is_logical(d40c))
1318                 return __d40_execute_command_log(d40c, command);
1319         else
1320                 return __d40_execute_command_phy(d40c, command);
1321 }
1322
1323 static u32 d40_get_prmo(struct d40_chan *d40c)
1324 {
1325         static const unsigned int phy_map[] = {
1326                 [STEDMA40_PCHAN_BASIC_MODE]
1327                         = D40_DREG_PRMO_PCHAN_BASIC,
1328                 [STEDMA40_PCHAN_MODULO_MODE]
1329                         = D40_DREG_PRMO_PCHAN_MODULO,
1330                 [STEDMA40_PCHAN_DOUBLE_DST_MODE]
1331                         = D40_DREG_PRMO_PCHAN_DOUBLE_DST,
1332         };
1333         static const unsigned int log_map[] = {
1334                 [STEDMA40_LCHAN_SRC_PHY_DST_LOG]
1335                         = D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG,
1336                 [STEDMA40_LCHAN_SRC_LOG_DST_PHY]
1337                         = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY,
1338                 [STEDMA40_LCHAN_SRC_LOG_DST_LOG]
1339                         = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG,
1340         };
1341
1342         if (chan_is_physical(d40c))
1343                 return phy_map[d40c->dma_cfg.mode_opt];
1344         else
1345                 return log_map[d40c->dma_cfg.mode_opt];
1346 }
1347
1348 static void d40_config_write(struct d40_chan *d40c)
1349 {
1350         u32 addr_base;
1351         u32 var;
1352
1353         /* Odd addresses are even addresses + 4 */
1354         addr_base = (d40c->phy_chan->num % 2) * 4;
1355         /* Setup channel mode to logical or physical */
1356         var = ((u32)(chan_is_logical(d40c)) + 1) <<
1357                 D40_CHAN_POS(d40c->phy_chan->num);
1358         writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
1359
1360         /* Setup operational mode option register */
1361         var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num);
1362
1363         writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
1364
1365         if (chan_is_logical(d40c)) {
1366                 int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS)
1367                            & D40_SREG_ELEM_LOG_LIDX_MASK;
1368                 void __iomem *chanbase = chan_base(d40c);
1369
1370                 /* Set default config for CFG reg */
1371                 writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG);
1372                 writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG);
1373
1374                 /* Set LIDX for lcla */
1375                 writel(lidx, chanbase + D40_CHAN_REG_SSELT);
1376                 writel(lidx, chanbase + D40_CHAN_REG_SDELT);
1377
1378                 /* Clear LNK which will be used by d40_chan_has_events() */
1379                 writel(0, chanbase + D40_CHAN_REG_SSLNK);
1380                 writel(0, chanbase + D40_CHAN_REG_SDLNK);
1381         }
1382 }
1383
1384 static u32 d40_residue(struct d40_chan *d40c)
1385 {
1386         u32 num_elt;
1387
1388         if (chan_is_logical(d40c))
1389                 num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
1390                         >> D40_MEM_LCSP2_ECNT_POS;
1391         else {
1392                 u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT);
1393                 num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK)
1394                           >> D40_SREG_ELEM_PHY_ECNT_POS;
1395         }
1396
1397         return num_elt * d40c->dma_cfg.dst_info.data_width;
1398 }
1399
1400 static bool d40_tx_is_linked(struct d40_chan *d40c)
1401 {
1402         bool is_link;
1403
1404         if (chan_is_logical(d40c))
1405                 is_link = readl(&d40c->lcpa->lcsp3) &  D40_MEM_LCSP3_DLOS_MASK;
1406         else
1407                 is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK)
1408                           & D40_SREG_LNK_PHYS_LNK_MASK;
1409
1410         return is_link;
1411 }
1412
1413 static int d40_pause(struct dma_chan *chan)
1414 {
1415         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
1416         int res = 0;
1417         unsigned long flags;
1418
1419         if (d40c->phy_chan == NULL) {
1420                 chan_err(d40c, "Channel is not allocated!\n");
1421                 return -EINVAL;
1422         }
1423
1424         if (!d40c->busy)
1425                 return 0;
1426
1427         spin_lock_irqsave(&d40c->lock, flags);
1428         pm_runtime_get_sync(d40c->base->dev);
1429
1430         res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1431
1432         pm_runtime_mark_last_busy(d40c->base->dev);
1433         pm_runtime_put_autosuspend(d40c->base->dev);
1434         spin_unlock_irqrestore(&d40c->lock, flags);
1435         return res;
1436 }
1437
1438 static int d40_resume(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         /* If bytes left to transfer or linked tx resume job */
1456         if (d40_residue(d40c) || d40_tx_is_linked(d40c))
1457                 res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1458
1459         pm_runtime_mark_last_busy(d40c->base->dev);
1460         pm_runtime_put_autosuspend(d40c->base->dev);
1461         spin_unlock_irqrestore(&d40c->lock, flags);
1462         return res;
1463 }
1464
1465 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1466 {
1467         struct d40_chan *d40c = container_of(tx->chan,
1468                                              struct d40_chan,
1469                                              chan);
1470         struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1471         unsigned long flags;
1472         dma_cookie_t cookie;
1473
1474         spin_lock_irqsave(&d40c->lock, flags);
1475         cookie = dma_cookie_assign(tx);
1476         d40_desc_queue(d40c, d40d);
1477         spin_unlock_irqrestore(&d40c->lock, flags);
1478
1479         return cookie;
1480 }
1481
1482 static int d40_start(struct d40_chan *d40c)
1483 {
1484         return d40_channel_execute_command(d40c, D40_DMA_RUN);
1485 }
1486
1487 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1488 {
1489         struct d40_desc *d40d;
1490         int err;
1491
1492         /* Start queued jobs, if any */
1493         d40d = d40_first_queued(d40c);
1494
1495         if (d40d != NULL) {
1496                 if (!d40c->busy) {
1497                         d40c->busy = true;
1498                         pm_runtime_get_sync(d40c->base->dev);
1499                 }
1500
1501                 /* Remove from queue */
1502                 d40_desc_remove(d40d);
1503
1504                 /* Add to active queue */
1505                 d40_desc_submit(d40c, d40d);
1506
1507                 /* Initiate DMA job */
1508                 d40_desc_load(d40c, d40d);
1509
1510                 /* Start dma job */
1511                 err = d40_start(d40c);
1512
1513                 if (err)
1514                         return NULL;
1515         }
1516
1517         return d40d;
1518 }
1519
1520 /* called from interrupt context */
1521 static void dma_tc_handle(struct d40_chan *d40c)
1522 {
1523         struct d40_desc *d40d;
1524
1525         /* Get first active entry from list */
1526         d40d = d40_first_active_get(d40c);
1527
1528         if (d40d == NULL)
1529                 return;
1530
1531         if (d40d->cyclic) {
1532                 /*
1533                  * If this was a paritially loaded list, we need to reloaded
1534                  * it, and only when the list is completed.  We need to check
1535                  * for done because the interrupt will hit for every link, and
1536                  * not just the last one.
1537                  */
1538                 if (d40d->lli_current < d40d->lli_len
1539                     && !d40_tx_is_linked(d40c)
1540                     && !d40_residue(d40c)) {
1541                         d40_lcla_free_all(d40c, d40d);
1542                         d40_desc_load(d40c, d40d);
1543                         (void) d40_start(d40c);
1544
1545                         if (d40d->lli_current == d40d->lli_len)
1546                                 d40d->lli_current = 0;
1547                 }
1548         } else {
1549                 d40_lcla_free_all(d40c, d40d);
1550
1551                 if (d40d->lli_current < d40d->lli_len) {
1552                         d40_desc_load(d40c, d40d);
1553                         /* Start dma job */
1554                         (void) d40_start(d40c);
1555                         return;
1556                 }
1557
1558                 if (d40_queue_start(d40c) == NULL) {
1559                         d40c->busy = false;
1560
1561                         pm_runtime_mark_last_busy(d40c->base->dev);
1562                         pm_runtime_put_autosuspend(d40c->base->dev);
1563                 }
1564
1565                 d40_desc_remove(d40d);
1566                 d40_desc_done(d40c, d40d);
1567         }
1568
1569         d40c->pending_tx++;
1570         tasklet_schedule(&d40c->tasklet);
1571
1572 }
1573
1574 static void dma_tasklet(struct tasklet_struct *t)
1575 {
1576         struct d40_chan *d40c = from_tasklet(d40c, t, tasklet);
1577         struct d40_desc *d40d;
1578         unsigned long flags;
1579         bool callback_active;
1580         struct dmaengine_desc_callback cb;
1581
1582         spin_lock_irqsave(&d40c->lock, flags);
1583
1584         /* Get first entry from the done list */
1585         d40d = d40_first_done(d40c);
1586         if (d40d == NULL) {
1587                 /* Check if we have reached here for cyclic job */
1588                 d40d = d40_first_active_get(d40c);
1589                 if (d40d == NULL || !d40d->cyclic)
1590                         goto check_pending_tx;
1591         }
1592
1593         if (!d40d->cyclic)
1594                 dma_cookie_complete(&d40d->txd);
1595
1596         /*
1597          * If terminating a channel pending_tx is set to zero.
1598          * This prevents any finished active jobs to return to the client.
1599          */
1600         if (d40c->pending_tx == 0) {
1601                 spin_unlock_irqrestore(&d40c->lock, flags);
1602                 return;
1603         }
1604
1605         /* Callback to client */
1606         callback_active = !!(d40d->txd.flags & DMA_PREP_INTERRUPT);
1607         dmaengine_desc_get_callback(&d40d->txd, &cb);
1608
1609         if (!d40d->cyclic) {
1610                 if (async_tx_test_ack(&d40d->txd)) {
1611                         d40_desc_remove(d40d);
1612                         d40_desc_free(d40c, d40d);
1613                 } else if (!d40d->is_in_client_list) {
1614                         d40_desc_remove(d40d);
1615                         d40_lcla_free_all(d40c, d40d);
1616                         list_add_tail(&d40d->node, &d40c->client);
1617                         d40d->is_in_client_list = true;
1618                 }
1619         }
1620
1621         d40c->pending_tx--;
1622
1623         if (d40c->pending_tx)
1624                 tasklet_schedule(&d40c->tasklet);
1625
1626         spin_unlock_irqrestore(&d40c->lock, flags);
1627
1628         if (callback_active)
1629                 dmaengine_desc_callback_invoke(&cb, NULL);
1630
1631         return;
1632  check_pending_tx:
1633         /* Rescue manouver if receiving double interrupts */
1634         if (d40c->pending_tx > 0)
1635                 d40c->pending_tx--;
1636         spin_unlock_irqrestore(&d40c->lock, flags);
1637 }
1638
1639 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1640 {
1641         int i;
1642         u32 idx;
1643         u32 row;
1644         long chan = -1;
1645         struct d40_chan *d40c;
1646         struct d40_base *base = data;
1647         u32 *regs = base->regs_interrupt;
1648         struct d40_interrupt_lookup *il = base->gen_dmac.il;
1649         u32 il_size = base->gen_dmac.il_size;
1650
1651         spin_lock(&base->interrupt_lock);
1652
1653         /* Read interrupt status of both logical and physical channels */
1654         for (i = 0; i < il_size; i++)
1655                 regs[i] = readl(base->virtbase + il[i].src);
1656
1657         for (;;) {
1658
1659                 chan = find_next_bit((unsigned long *)regs,
1660                                      BITS_PER_LONG * il_size, chan + 1);
1661
1662                 /* No more set bits found? */
1663                 if (chan == BITS_PER_LONG * il_size)
1664                         break;
1665
1666                 row = chan / BITS_PER_LONG;
1667                 idx = chan & (BITS_PER_LONG - 1);
1668
1669                 if (il[row].offset == D40_PHY_CHAN)
1670                         d40c = base->lookup_phy_chans[idx];
1671                 else
1672                         d40c = base->lookup_log_chans[il[row].offset + idx];
1673
1674                 if (!d40c) {
1675                         /*
1676                          * No error because this can happen if something else
1677                          * in the system is using the channel.
1678                          */
1679                         continue;
1680                 }
1681
1682                 /* ACK interrupt */
1683                 writel(BIT(idx), base->virtbase + il[row].clr);
1684
1685                 spin_lock(&d40c->lock);
1686
1687                 if (!il[row].is_error)
1688                         dma_tc_handle(d40c);
1689                 else
1690                         d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1691                                 chan, il[row].offset, idx);
1692
1693                 spin_unlock(&d40c->lock);
1694         }
1695
1696         spin_unlock(&base->interrupt_lock);
1697
1698         return IRQ_HANDLED;
1699 }
1700
1701 static int d40_validate_conf(struct d40_chan *d40c,
1702                              struct stedma40_chan_cfg *conf)
1703 {
1704         int res = 0;
1705         bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1706
1707         if (!conf->dir) {
1708                 chan_err(d40c, "Invalid direction.\n");
1709                 res = -EINVAL;
1710         }
1711
1712         if ((is_log && conf->dev_type > d40c->base->num_log_chans)  ||
1713             (!is_log && conf->dev_type > d40c->base->num_phy_chans) ||
1714             (conf->dev_type < 0)) {
1715                 chan_err(d40c, "Invalid device type (%d)\n", conf->dev_type);
1716                 res = -EINVAL;
1717         }
1718
1719         if (conf->dir == DMA_DEV_TO_DEV) {
1720                 /*
1721                  * DMAC HW supports it. Will be added to this driver,
1722                  * in case any dma client requires it.
1723                  */
1724                 chan_err(d40c, "periph to periph not supported\n");
1725                 res = -EINVAL;
1726         }
1727
1728         if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1729             conf->src_info.data_width !=
1730             d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1731             conf->dst_info.data_width) {
1732                 /*
1733                  * The DMAC hardware only supports
1734                  * src (burst x width) == dst (burst x width)
1735                  */
1736
1737                 chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1738                 res = -EINVAL;
1739         }
1740
1741         return res;
1742 }
1743
1744 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1745                                bool is_src, int log_event_line, bool is_log,
1746                                bool *first_user)
1747 {
1748         unsigned long flags;
1749         spin_lock_irqsave(&phy->lock, flags);
1750
1751         *first_user = ((phy->allocated_src | phy->allocated_dst)
1752                         == D40_ALLOC_FREE);
1753
1754         if (!is_log) {
1755                 /* Physical interrupts are masked per physical full channel */
1756                 if (phy->allocated_src == D40_ALLOC_FREE &&
1757                     phy->allocated_dst == D40_ALLOC_FREE) {
1758                         phy->allocated_dst = D40_ALLOC_PHY;
1759                         phy->allocated_src = D40_ALLOC_PHY;
1760                         goto found_unlock;
1761                 } else
1762                         goto not_found_unlock;
1763         }
1764
1765         /* Logical channel */
1766         if (is_src) {
1767                 if (phy->allocated_src == D40_ALLOC_PHY)
1768                         goto not_found_unlock;
1769
1770                 if (phy->allocated_src == D40_ALLOC_FREE)
1771                         phy->allocated_src = D40_ALLOC_LOG_FREE;
1772
1773                 if (!(phy->allocated_src & BIT(log_event_line))) {
1774                         phy->allocated_src |= BIT(log_event_line);
1775                         goto found_unlock;
1776                 } else
1777                         goto not_found_unlock;
1778         } else {
1779                 if (phy->allocated_dst == D40_ALLOC_PHY)
1780                         goto not_found_unlock;
1781
1782                 if (phy->allocated_dst == D40_ALLOC_FREE)
1783                         phy->allocated_dst = D40_ALLOC_LOG_FREE;
1784
1785                 if (!(phy->allocated_dst & BIT(log_event_line))) {
1786                         phy->allocated_dst |= BIT(log_event_line);
1787                         goto found_unlock;
1788                 }
1789         }
1790  not_found_unlock:
1791         spin_unlock_irqrestore(&phy->lock, flags);
1792         return false;
1793  found_unlock:
1794         spin_unlock_irqrestore(&phy->lock, flags);
1795         return true;
1796 }
1797
1798 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1799                                int log_event_line)
1800 {
1801         unsigned long flags;
1802         bool is_free = false;
1803
1804         spin_lock_irqsave(&phy->lock, flags);
1805         if (!log_event_line) {
1806                 phy->allocated_dst = D40_ALLOC_FREE;
1807                 phy->allocated_src = D40_ALLOC_FREE;
1808                 is_free = true;
1809                 goto unlock;
1810         }
1811
1812         /* Logical channel */
1813         if (is_src) {
1814                 phy->allocated_src &= ~BIT(log_event_line);
1815                 if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1816                         phy->allocated_src = D40_ALLOC_FREE;
1817         } else {
1818                 phy->allocated_dst &= ~BIT(log_event_line);
1819                 if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1820                         phy->allocated_dst = D40_ALLOC_FREE;
1821         }
1822
1823         is_free = ((phy->allocated_src | phy->allocated_dst) ==
1824                    D40_ALLOC_FREE);
1825  unlock:
1826         spin_unlock_irqrestore(&phy->lock, flags);
1827
1828         return is_free;
1829 }
1830
1831 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1832 {
1833         int dev_type = d40c->dma_cfg.dev_type;
1834         int event_group;
1835         int event_line;
1836         struct d40_phy_res *phys;
1837         int i;
1838         int j;
1839         int log_num;
1840         int num_phy_chans;
1841         bool is_src;
1842         bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1843
1844         phys = d40c->base->phy_res;
1845         num_phy_chans = d40c->base->num_phy_chans;
1846
1847         if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
1848                 log_num = 2 * dev_type;
1849                 is_src = true;
1850         } else if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
1851                    d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1852                 /* dst event lines are used for logical memcpy */
1853                 log_num = 2 * dev_type + 1;
1854                 is_src = false;
1855         } else
1856                 return -EINVAL;
1857
1858         event_group = D40_TYPE_TO_GROUP(dev_type);
1859         event_line = D40_TYPE_TO_EVENT(dev_type);
1860
1861         if (!is_log) {
1862                 if (d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1863                         /* Find physical half channel */
1864                         if (d40c->dma_cfg.use_fixed_channel) {
1865                                 i = d40c->dma_cfg.phy_channel;
1866                                 if (d40_alloc_mask_set(&phys[i], is_src,
1867                                                        0, is_log,
1868                                                        first_phy_user))
1869                                         goto found_phy;
1870                         } else {
1871                                 for (i = 0; i < num_phy_chans; i++) {
1872                                         if (d40_alloc_mask_set(&phys[i], is_src,
1873                                                        0, is_log,
1874                                                        first_phy_user))
1875                                                 goto found_phy;
1876                                 }
1877                         }
1878                 } else
1879                         for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1880                                 int phy_num = j  + event_group * 2;
1881                                 for (i = phy_num; i < phy_num + 2; i++) {
1882                                         if (d40_alloc_mask_set(&phys[i],
1883                                                                is_src,
1884                                                                0,
1885                                                                is_log,
1886                                                                first_phy_user))
1887                                                 goto found_phy;
1888                                 }
1889                         }
1890                 return -EINVAL;
1891 found_phy:
1892                 d40c->phy_chan = &phys[i];
1893                 d40c->log_num = D40_PHY_CHAN;
1894                 goto out;
1895         }
1896         if (dev_type == -1)
1897                 return -EINVAL;
1898
1899         /* Find logical channel */
1900         for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1901                 int phy_num = j + event_group * 2;
1902
1903                 if (d40c->dma_cfg.use_fixed_channel) {
1904                         i = d40c->dma_cfg.phy_channel;
1905
1906                         if ((i != phy_num) && (i != phy_num + 1)) {
1907                                 dev_err(chan2dev(d40c),
1908                                         "invalid fixed phy channel %d\n", i);
1909                                 return -EINVAL;
1910                         }
1911
1912                         if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1913                                                is_log, first_phy_user))
1914                                 goto found_log;
1915
1916                         dev_err(chan2dev(d40c),
1917                                 "could not allocate fixed phy channel %d\n", i);
1918                         return -EINVAL;
1919                 }
1920
1921                 /*
1922                  * Spread logical channels across all available physical rather
1923                  * than pack every logical channel at the first available phy
1924                  * channels.
1925                  */
1926                 if (is_src) {
1927                         for (i = phy_num; i < phy_num + 2; i++) {
1928                                 if (d40_alloc_mask_set(&phys[i], is_src,
1929                                                        event_line, is_log,
1930                                                        first_phy_user))
1931                                         goto found_log;
1932                         }
1933                 } else {
1934                         for (i = phy_num + 1; i >= phy_num; i--) {
1935                                 if (d40_alloc_mask_set(&phys[i], is_src,
1936                                                        event_line, is_log,
1937                                                        first_phy_user))
1938                                         goto found_log;
1939                         }
1940                 }
1941         }
1942         return -EINVAL;
1943
1944 found_log:
1945         d40c->phy_chan = &phys[i];
1946         d40c->log_num = log_num;
1947 out:
1948
1949         if (is_log)
1950                 d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1951         else
1952                 d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1953
1954         return 0;
1955
1956 }
1957
1958 static int d40_config_memcpy(struct d40_chan *d40c)
1959 {
1960         dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1961
1962         if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1963                 d40c->dma_cfg = dma40_memcpy_conf_log;
1964                 d40c->dma_cfg.dev_type = dma40_memcpy_channels[d40c->chan.chan_id];
1965
1966                 d40_log_cfg(&d40c->dma_cfg,
1967                             &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
1968
1969         } else if (dma_has_cap(DMA_MEMCPY, cap) &&
1970                    dma_has_cap(DMA_SLAVE, cap)) {
1971                 d40c->dma_cfg = dma40_memcpy_conf_phy;
1972
1973                 /* Generate interrrupt at end of transfer or relink. */
1974                 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_TIM_POS);
1975
1976                 /* Generate interrupt on error. */
1977                 d40c->src_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
1978                 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
1979
1980         } else {
1981                 chan_err(d40c, "No memcpy\n");
1982                 return -EINVAL;
1983         }
1984
1985         return 0;
1986 }
1987
1988 static int d40_free_dma(struct d40_chan *d40c)
1989 {
1990
1991         int res = 0;
1992         u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1993         struct d40_phy_res *phy = d40c->phy_chan;
1994         bool is_src;
1995
1996         /* Terminate all queued and active transfers */
1997         d40_term_all(d40c);
1998
1999         if (phy == NULL) {
2000                 chan_err(d40c, "phy == null\n");
2001                 return -EINVAL;
2002         }
2003
2004         if (phy->allocated_src == D40_ALLOC_FREE &&
2005             phy->allocated_dst == D40_ALLOC_FREE) {
2006                 chan_err(d40c, "channel already free\n");
2007                 return -EINVAL;
2008         }
2009
2010         if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2011             d40c->dma_cfg.dir == DMA_MEM_TO_MEM)
2012                 is_src = false;
2013         else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2014                 is_src = true;
2015         else {
2016                 chan_err(d40c, "Unknown direction\n");
2017                 return -EINVAL;
2018         }
2019
2020         pm_runtime_get_sync(d40c->base->dev);
2021         res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2022         if (res) {
2023                 chan_err(d40c, "stop failed\n");
2024                 goto mark_last_busy;
2025         }
2026
2027         d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2028
2029         if (chan_is_logical(d40c))
2030                 d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2031         else
2032                 d40c->base->lookup_phy_chans[phy->num] = NULL;
2033
2034         if (d40c->busy) {
2035                 pm_runtime_mark_last_busy(d40c->base->dev);
2036                 pm_runtime_put_autosuspend(d40c->base->dev);
2037         }
2038
2039         d40c->busy = false;
2040         d40c->phy_chan = NULL;
2041         d40c->configured = false;
2042  mark_last_busy:
2043         pm_runtime_mark_last_busy(d40c->base->dev);
2044         pm_runtime_put_autosuspend(d40c->base->dev);
2045         return res;
2046 }
2047
2048 static bool d40_is_paused(struct d40_chan *d40c)
2049 {
2050         void __iomem *chanbase = chan_base(d40c);
2051         bool is_paused = false;
2052         unsigned long flags;
2053         void __iomem *active_reg;
2054         u32 status;
2055         u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2056
2057         spin_lock_irqsave(&d40c->lock, flags);
2058
2059         if (chan_is_physical(d40c)) {
2060                 if (d40c->phy_chan->num % 2 == 0)
2061                         active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2062                 else
2063                         active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2064
2065                 status = (readl(active_reg) &
2066                           D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2067                         D40_CHAN_POS(d40c->phy_chan->num);
2068                 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2069                         is_paused = true;
2070                 goto unlock;
2071         }
2072
2073         if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2074             d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
2075                 status = readl(chanbase + D40_CHAN_REG_SDLNK);
2076         } else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
2077                 status = readl(chanbase + D40_CHAN_REG_SSLNK);
2078         } else {
2079                 chan_err(d40c, "Unknown direction\n");
2080                 goto unlock;
2081         }
2082
2083         status = (status & D40_EVENTLINE_MASK(event)) >>
2084                 D40_EVENTLINE_POS(event);
2085
2086         if (status != D40_DMA_RUN)
2087                 is_paused = true;
2088  unlock:
2089         spin_unlock_irqrestore(&d40c->lock, flags);
2090         return is_paused;
2091
2092 }
2093
2094 static u32 stedma40_residue(struct dma_chan *chan)
2095 {
2096         struct d40_chan *d40c =
2097                 container_of(chan, struct d40_chan, chan);
2098         u32 bytes_left;
2099         unsigned long flags;
2100
2101         spin_lock_irqsave(&d40c->lock, flags);
2102         bytes_left = d40_residue(d40c);
2103         spin_unlock_irqrestore(&d40c->lock, flags);
2104
2105         return bytes_left;
2106 }
2107
2108 static int
2109 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2110                 struct scatterlist *sg_src, struct scatterlist *sg_dst,
2111                 unsigned int sg_len, dma_addr_t src_dev_addr,
2112                 dma_addr_t dst_dev_addr)
2113 {
2114         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2115         struct stedma40_half_channel_info *src_info = &cfg->src_info;
2116         struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2117         int ret;
2118
2119         ret = d40_log_sg_to_lli(sg_src, sg_len,
2120                                 src_dev_addr,
2121                                 desc->lli_log.src,
2122                                 chan->log_def.lcsp1,
2123                                 src_info->data_width,
2124                                 dst_info->data_width);
2125
2126         ret = d40_log_sg_to_lli(sg_dst, sg_len,
2127                                 dst_dev_addr,
2128                                 desc->lli_log.dst,
2129                                 chan->log_def.lcsp3,
2130                                 dst_info->data_width,
2131                                 src_info->data_width);
2132
2133         return ret < 0 ? ret : 0;
2134 }
2135
2136 static int
2137 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2138                 struct scatterlist *sg_src, struct scatterlist *sg_dst,
2139                 unsigned int sg_len, dma_addr_t src_dev_addr,
2140                 dma_addr_t dst_dev_addr)
2141 {
2142         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2143         struct stedma40_half_channel_info *src_info = &cfg->src_info;
2144         struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2145         unsigned long flags = 0;
2146         int ret;
2147
2148         if (desc->cyclic)
2149                 flags |= LLI_CYCLIC | LLI_TERM_INT;
2150
2151         ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2152                                 desc->lli_phy.src,
2153                                 virt_to_phys(desc->lli_phy.src),
2154                                 chan->src_def_cfg,
2155                                 src_info, dst_info, flags);
2156
2157         ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2158                                 desc->lli_phy.dst,
2159                                 virt_to_phys(desc->lli_phy.dst),
2160                                 chan->dst_def_cfg,
2161                                 dst_info, src_info, flags);
2162
2163         dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2164                                    desc->lli_pool.size, DMA_TO_DEVICE);
2165
2166         return ret < 0 ? ret : 0;
2167 }
2168
2169 static struct d40_desc *
2170 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2171               unsigned int sg_len, unsigned long dma_flags)
2172 {
2173         struct stedma40_chan_cfg *cfg;
2174         struct d40_desc *desc;
2175         int ret;
2176
2177         desc = d40_desc_get(chan);
2178         if (!desc)
2179                 return NULL;
2180
2181         cfg = &chan->dma_cfg;
2182         desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2183                                         cfg->dst_info.data_width);
2184         if (desc->lli_len < 0) {
2185                 chan_err(chan, "Unaligned size\n");
2186                 goto free_desc;
2187         }
2188
2189         ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2190         if (ret < 0) {
2191                 chan_err(chan, "Could not allocate lli\n");
2192                 goto free_desc;
2193         }
2194
2195         desc->lli_current = 0;
2196         desc->txd.flags = dma_flags;
2197         desc->txd.tx_submit = d40_tx_submit;
2198
2199         dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2200
2201         return desc;
2202  free_desc:
2203         d40_desc_free(chan, desc);
2204         return NULL;
2205 }
2206
2207 static struct dma_async_tx_descriptor *
2208 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2209             struct scatterlist *sg_dst, unsigned int sg_len,
2210             enum dma_transfer_direction direction, unsigned long dma_flags)
2211 {
2212         struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2213         dma_addr_t src_dev_addr;
2214         dma_addr_t dst_dev_addr;
2215         struct d40_desc *desc;
2216         unsigned long flags;
2217         int ret;
2218
2219         if (!chan->phy_chan) {
2220                 chan_err(chan, "Cannot prepare unallocated channel\n");
2221                 return NULL;
2222         }
2223
2224         d40_set_runtime_config_write(dchan, &chan->slave_config, direction);
2225
2226         spin_lock_irqsave(&chan->lock, flags);
2227
2228         desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2229         if (desc == NULL)
2230                 goto unlock;
2231
2232         if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2233                 desc->cyclic = true;
2234
2235         src_dev_addr = 0;
2236         dst_dev_addr = 0;
2237         if (direction == DMA_DEV_TO_MEM)
2238                 src_dev_addr = chan->runtime_addr;
2239         else if (direction == DMA_MEM_TO_DEV)
2240                 dst_dev_addr = chan->runtime_addr;
2241
2242         if (chan_is_logical(chan))
2243                 ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2244                                       sg_len, src_dev_addr, dst_dev_addr);
2245         else
2246                 ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2247                                       sg_len, src_dev_addr, dst_dev_addr);
2248
2249         if (ret) {
2250                 chan_err(chan, "Failed to prepare %s sg job: %d\n",
2251                          chan_is_logical(chan) ? "log" : "phy", ret);
2252                 goto free_desc;
2253         }
2254
2255         /*
2256          * add descriptor to the prepare queue in order to be able
2257          * to free them later in terminate_all
2258          */
2259         list_add_tail(&desc->node, &chan->prepare_queue);
2260
2261         spin_unlock_irqrestore(&chan->lock, flags);
2262
2263         return &desc->txd;
2264  free_desc:
2265         d40_desc_free(chan, desc);
2266  unlock:
2267         spin_unlock_irqrestore(&chan->lock, flags);
2268         return NULL;
2269 }
2270
2271 bool stedma40_filter(struct dma_chan *chan, void *data)
2272 {
2273         struct stedma40_chan_cfg *info = data;
2274         struct d40_chan *d40c =
2275                 container_of(chan, struct d40_chan, chan);
2276         int err;
2277
2278         if (data) {
2279                 err = d40_validate_conf(d40c, info);
2280                 if (!err)
2281                         d40c->dma_cfg = *info;
2282         } else
2283                 err = d40_config_memcpy(d40c);
2284
2285         if (!err)
2286                 d40c->configured = true;
2287
2288         return err == 0;
2289 }
2290 EXPORT_SYMBOL(stedma40_filter);
2291
2292 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2293 {
2294         bool realtime = d40c->dma_cfg.realtime;
2295         bool highprio = d40c->dma_cfg.high_priority;
2296         u32 rtreg;
2297         u32 event = D40_TYPE_TO_EVENT(dev_type);
2298         u32 group = D40_TYPE_TO_GROUP(dev_type);
2299         u32 bit = BIT(event);
2300         u32 prioreg;
2301         struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2302
2303         rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2304         /*
2305          * Due to a hardware bug, in some cases a logical channel triggered by
2306          * a high priority destination event line can generate extra packet
2307          * transactions.
2308          *
2309          * The workaround is to not set the high priority level for the
2310          * destination event lines that trigger logical channels.
2311          */
2312         if (!src && chan_is_logical(d40c))
2313                 highprio = false;
2314
2315         prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2316
2317         /* Destination event lines are stored in the upper halfword */
2318         if (!src)
2319                 bit <<= 16;
2320
2321         writel(bit, d40c->base->virtbase + prioreg + group * 4);
2322         writel(bit, d40c->base->virtbase + rtreg + group * 4);
2323 }
2324
2325 static void d40_set_prio_realtime(struct d40_chan *d40c)
2326 {
2327         if (d40c->base->rev < 3)
2328                 return;
2329
2330         if ((d40c->dma_cfg.dir ==  DMA_DEV_TO_MEM) ||
2331             (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2332                 __d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, true);
2333
2334         if ((d40c->dma_cfg.dir ==  DMA_MEM_TO_DEV) ||
2335             (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2336                 __d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, false);
2337 }
2338
2339 #define D40_DT_FLAGS_MODE(flags)       ((flags >> 0) & 0x1)
2340 #define D40_DT_FLAGS_DIR(flags)        ((flags >> 1) & 0x1)
2341 #define D40_DT_FLAGS_BIG_ENDIAN(flags) ((flags >> 2) & 0x1)
2342 #define D40_DT_FLAGS_FIXED_CHAN(flags) ((flags >> 3) & 0x1)
2343 #define D40_DT_FLAGS_HIGH_PRIO(flags)  ((flags >> 4) & 0x1)
2344
2345 static struct dma_chan *d40_xlate(struct of_phandle_args *dma_spec,
2346                                   struct of_dma *ofdma)
2347 {
2348         struct stedma40_chan_cfg cfg;
2349         dma_cap_mask_t cap;
2350         u32 flags;
2351
2352         memset(&cfg, 0, sizeof(struct stedma40_chan_cfg));
2353
2354         dma_cap_zero(cap);
2355         dma_cap_set(DMA_SLAVE, cap);
2356
2357         cfg.dev_type = dma_spec->args[0];
2358         flags = dma_spec->args[2];
2359
2360         switch (D40_DT_FLAGS_MODE(flags)) {
2361         case 0: cfg.mode = STEDMA40_MODE_LOGICAL; break;
2362         case 1: cfg.mode = STEDMA40_MODE_PHYSICAL; break;
2363         }
2364
2365         switch (D40_DT_FLAGS_DIR(flags)) {
2366         case 0:
2367                 cfg.dir = DMA_MEM_TO_DEV;
2368                 cfg.dst_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2369                 break;
2370         case 1:
2371                 cfg.dir = DMA_DEV_TO_MEM;
2372                 cfg.src_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2373                 break;
2374         }
2375
2376         if (D40_DT_FLAGS_FIXED_CHAN(flags)) {
2377                 cfg.phy_channel = dma_spec->args[1];
2378                 cfg.use_fixed_channel = true;
2379         }
2380
2381         if (D40_DT_FLAGS_HIGH_PRIO(flags))
2382                 cfg.high_priority = true;
2383
2384         return dma_request_channel(cap, stedma40_filter, &cfg);
2385 }
2386
2387 /* DMA ENGINE functions */
2388 static int d40_alloc_chan_resources(struct dma_chan *chan)
2389 {
2390         int err;
2391         unsigned long flags;
2392         struct d40_chan *d40c =
2393                 container_of(chan, struct d40_chan, chan);
2394         bool is_free_phy;
2395         spin_lock_irqsave(&d40c->lock, flags);
2396
2397         dma_cookie_init(chan);
2398
2399         /* If no dma configuration is set use default configuration (memcpy) */
2400         if (!d40c->configured) {
2401                 err = d40_config_memcpy(d40c);
2402                 if (err) {
2403                         chan_err(d40c, "Failed to configure memcpy channel\n");
2404                         goto mark_last_busy;
2405                 }
2406         }
2407
2408         err = d40_allocate_channel(d40c, &is_free_phy);
2409         if (err) {
2410                 chan_err(d40c, "Failed to allocate channel\n");
2411                 d40c->configured = false;
2412                 goto mark_last_busy;
2413         }
2414
2415         pm_runtime_get_sync(d40c->base->dev);
2416
2417         d40_set_prio_realtime(d40c);
2418
2419         if (chan_is_logical(d40c)) {
2420                 if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2421                         d40c->lcpa = d40c->base->lcpa_base +
2422                                 d40c->dma_cfg.dev_type * D40_LCPA_CHAN_SIZE;
2423                 else
2424                         d40c->lcpa = d40c->base->lcpa_base +
2425                                 d40c->dma_cfg.dev_type *
2426                                 D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2427
2428                 /* Unmask the Global Interrupt Mask. */
2429                 d40c->src_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2430                 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2431         }
2432
2433         dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2434                  chan_is_logical(d40c) ? "logical" : "physical",
2435                  d40c->phy_chan->num,
2436                  d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2437
2438
2439         /*
2440          * Only write channel configuration to the DMA if the physical
2441          * resource is free. In case of multiple logical channels
2442          * on the same physical resource, only the first write is necessary.
2443          */
2444         if (is_free_phy)
2445                 d40_config_write(d40c);
2446  mark_last_busy:
2447         pm_runtime_mark_last_busy(d40c->base->dev);
2448         pm_runtime_put_autosuspend(d40c->base->dev);
2449         spin_unlock_irqrestore(&d40c->lock, flags);
2450         return err;
2451 }
2452
2453 static void d40_free_chan_resources(struct dma_chan *chan)
2454 {
2455         struct d40_chan *d40c =
2456                 container_of(chan, struct d40_chan, chan);
2457         int err;
2458         unsigned long flags;
2459
2460         if (d40c->phy_chan == NULL) {
2461                 chan_err(d40c, "Cannot free unallocated channel\n");
2462                 return;
2463         }
2464
2465         spin_lock_irqsave(&d40c->lock, flags);
2466
2467         err = d40_free_dma(d40c);
2468
2469         if (err)
2470                 chan_err(d40c, "Failed to free channel\n");
2471         spin_unlock_irqrestore(&d40c->lock, flags);
2472 }
2473
2474 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2475                                                        dma_addr_t dst,
2476                                                        dma_addr_t src,
2477                                                        size_t size,
2478                                                        unsigned long dma_flags)
2479 {
2480         struct scatterlist dst_sg;
2481         struct scatterlist src_sg;
2482
2483         sg_init_table(&dst_sg, 1);
2484         sg_init_table(&src_sg, 1);
2485
2486         sg_dma_address(&dst_sg) = dst;
2487         sg_dma_address(&src_sg) = src;
2488
2489         sg_dma_len(&dst_sg) = size;
2490         sg_dma_len(&src_sg) = size;
2491
2492         return d40_prep_sg(chan, &src_sg, &dst_sg, 1,
2493                            DMA_MEM_TO_MEM, dma_flags);
2494 }
2495
2496 static struct dma_async_tx_descriptor *
2497 d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2498                   unsigned int sg_len, enum dma_transfer_direction direction,
2499                   unsigned long dma_flags, void *context)
2500 {
2501         if (!is_slave_direction(direction))
2502                 return NULL;
2503
2504         return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2505 }
2506
2507 static struct dma_async_tx_descriptor *
2508 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2509                      size_t buf_len, size_t period_len,
2510                      enum dma_transfer_direction direction, unsigned long flags)
2511 {
2512         unsigned int periods = buf_len / period_len;
2513         struct dma_async_tx_descriptor *txd;
2514         struct scatterlist *sg;
2515         int i;
2516
2517         sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2518         if (!sg)
2519                 return NULL;
2520
2521         for (i = 0; i < periods; i++) {
2522                 sg_dma_address(&sg[i]) = dma_addr;
2523                 sg_dma_len(&sg[i]) = period_len;
2524                 dma_addr += period_len;
2525         }
2526
2527         sg_chain(sg, periods + 1, sg);
2528
2529         txd = d40_prep_sg(chan, sg, sg, periods, direction,
2530                           DMA_PREP_INTERRUPT);
2531
2532         kfree(sg);
2533
2534         return txd;
2535 }
2536
2537 static enum dma_status d40_tx_status(struct dma_chan *chan,
2538                                      dma_cookie_t cookie,
2539                                      struct dma_tx_state *txstate)
2540 {
2541         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2542         enum dma_status ret;
2543
2544         if (d40c->phy_chan == NULL) {
2545                 chan_err(d40c, "Cannot read status of unallocated channel\n");
2546                 return -EINVAL;
2547         }
2548
2549         ret = dma_cookie_status(chan, cookie, txstate);
2550         if (ret != DMA_COMPLETE && txstate)
2551                 dma_set_residue(txstate, stedma40_residue(chan));
2552
2553         if (d40_is_paused(d40c))
2554                 ret = DMA_PAUSED;
2555
2556         return ret;
2557 }
2558
2559 static void d40_issue_pending(struct dma_chan *chan)
2560 {
2561         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2562         unsigned long flags;
2563
2564         if (d40c->phy_chan == NULL) {
2565                 chan_err(d40c, "Channel is not allocated!\n");
2566                 return;
2567         }
2568
2569         spin_lock_irqsave(&d40c->lock, flags);
2570
2571         list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2572
2573         /* Busy means that queued jobs are already being processed */
2574         if (!d40c->busy)
2575                 (void) d40_queue_start(d40c);
2576
2577         spin_unlock_irqrestore(&d40c->lock, flags);
2578 }
2579
2580 static int d40_terminate_all(struct dma_chan *chan)
2581 {
2582         unsigned long flags;
2583         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2584         int ret;
2585
2586         if (d40c->phy_chan == NULL) {
2587                 chan_err(d40c, "Channel is not allocated!\n");
2588                 return -EINVAL;
2589         }
2590
2591         spin_lock_irqsave(&d40c->lock, flags);
2592
2593         pm_runtime_get_sync(d40c->base->dev);
2594         ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2595         if (ret)
2596                 chan_err(d40c, "Failed to stop channel\n");
2597
2598         d40_term_all(d40c);
2599         pm_runtime_mark_last_busy(d40c->base->dev);
2600         pm_runtime_put_autosuspend(d40c->base->dev);
2601         if (d40c->busy) {
2602                 pm_runtime_mark_last_busy(d40c->base->dev);
2603                 pm_runtime_put_autosuspend(d40c->base->dev);
2604         }
2605         d40c->busy = false;
2606
2607         spin_unlock_irqrestore(&d40c->lock, flags);
2608         return 0;
2609 }
2610
2611 static int
2612 dma40_config_to_halfchannel(struct d40_chan *d40c,
2613                             struct stedma40_half_channel_info *info,
2614                             u32 maxburst)
2615 {
2616         int psize;
2617
2618         if (chan_is_logical(d40c)) {
2619                 if (maxburst >= 16)
2620                         psize = STEDMA40_PSIZE_LOG_16;
2621                 else if (maxburst >= 8)
2622                         psize = STEDMA40_PSIZE_LOG_8;
2623                 else if (maxburst >= 4)
2624                         psize = STEDMA40_PSIZE_LOG_4;
2625                 else
2626                         psize = STEDMA40_PSIZE_LOG_1;
2627         } else {
2628                 if (maxburst >= 16)
2629                         psize = STEDMA40_PSIZE_PHY_16;
2630                 else if (maxburst >= 8)
2631                         psize = STEDMA40_PSIZE_PHY_8;
2632                 else if (maxburst >= 4)
2633                         psize = STEDMA40_PSIZE_PHY_4;
2634                 else
2635                         psize = STEDMA40_PSIZE_PHY_1;
2636         }
2637
2638         info->psize = psize;
2639         info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2640
2641         return 0;
2642 }
2643
2644 static int d40_set_runtime_config(struct dma_chan *chan,
2645                                   struct dma_slave_config *config)
2646 {
2647         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2648
2649         memcpy(&d40c->slave_config, config, sizeof(*config));
2650
2651         return 0;
2652 }
2653
2654 /* Runtime reconfiguration extension */
2655 static int d40_set_runtime_config_write(struct dma_chan *chan,
2656                                   struct dma_slave_config *config,
2657                                   enum dma_transfer_direction direction)
2658 {
2659         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2660         struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2661         enum dma_slave_buswidth src_addr_width, dst_addr_width;
2662         dma_addr_t config_addr;
2663         u32 src_maxburst, dst_maxburst;
2664         int ret;
2665
2666         if (d40c->phy_chan == NULL) {
2667                 chan_err(d40c, "Channel is not allocated!\n");
2668                 return -EINVAL;
2669         }
2670
2671         src_addr_width = config->src_addr_width;
2672         src_maxburst = config->src_maxburst;
2673         dst_addr_width = config->dst_addr_width;
2674         dst_maxburst = config->dst_maxburst;
2675
2676         if (direction == DMA_DEV_TO_MEM) {
2677                 config_addr = config->src_addr;
2678
2679                 if (cfg->dir != DMA_DEV_TO_MEM)
2680                         dev_dbg(d40c->base->dev,
2681                                 "channel was not configured for peripheral "
2682                                 "to memory transfer (%d) overriding\n",
2683                                 cfg->dir);
2684                 cfg->dir = DMA_DEV_TO_MEM;
2685
2686                 /* Configure the memory side */
2687                 if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2688                         dst_addr_width = src_addr_width;
2689                 if (dst_maxburst == 0)
2690                         dst_maxburst = src_maxburst;
2691
2692         } else if (direction == DMA_MEM_TO_DEV) {
2693                 config_addr = config->dst_addr;
2694
2695                 if (cfg->dir != DMA_MEM_TO_DEV)
2696                         dev_dbg(d40c->base->dev,
2697                                 "channel was not configured for memory "
2698                                 "to peripheral transfer (%d) overriding\n",
2699                                 cfg->dir);
2700                 cfg->dir = DMA_MEM_TO_DEV;
2701
2702                 /* Configure the memory side */
2703                 if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2704                         src_addr_width = dst_addr_width;
2705                 if (src_maxburst == 0)
2706                         src_maxburst = dst_maxburst;
2707         } else {
2708                 dev_err(d40c->base->dev,
2709                         "unrecognized channel direction %d\n",
2710                         direction);
2711                 return -EINVAL;
2712         }
2713
2714         if (config_addr <= 0) {
2715                 dev_err(d40c->base->dev, "no address supplied\n");
2716                 return -EINVAL;
2717         }
2718
2719         if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2720                 dev_err(d40c->base->dev,
2721                         "src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2722                         src_maxburst,
2723                         src_addr_width,
2724                         dst_maxburst,
2725                         dst_addr_width);
2726                 return -EINVAL;
2727         }
2728
2729         if (src_maxburst > 16) {
2730                 src_maxburst = 16;
2731                 dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2732         } else if (dst_maxburst > 16) {
2733                 dst_maxburst = 16;
2734                 src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2735         }
2736
2737         /* Only valid widths are; 1, 2, 4 and 8. */
2738         if (src_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2739             src_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2740             dst_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2741             dst_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2742             !is_power_of_2(src_addr_width) ||
2743             !is_power_of_2(dst_addr_width))
2744                 return -EINVAL;
2745
2746         cfg->src_info.data_width = src_addr_width;
2747         cfg->dst_info.data_width = dst_addr_width;
2748
2749         ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2750                                           src_maxburst);
2751         if (ret)
2752                 return ret;
2753
2754         ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2755                                           dst_maxburst);
2756         if (ret)
2757                 return ret;
2758
2759         /* Fill in register values */
2760         if (chan_is_logical(d40c))
2761                 d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2762         else
2763                 d40_phy_cfg(cfg, &d40c->src_def_cfg, &d40c->dst_def_cfg);
2764
2765         /* These settings will take precedence later */
2766         d40c->runtime_addr = config_addr;
2767         d40c->runtime_direction = direction;
2768         dev_dbg(d40c->base->dev,
2769                 "configured channel %s for %s, data width %d/%d, "
2770                 "maxburst %d/%d elements, LE, no flow control\n",
2771                 dma_chan_name(chan),
2772                 (direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2773                 src_addr_width, dst_addr_width,
2774                 src_maxburst, dst_maxburst);
2775
2776         return 0;
2777 }
2778
2779 /* Initialization functions */
2780
2781 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2782                                  struct d40_chan *chans, int offset,
2783                                  int num_chans)
2784 {
2785         int i = 0;
2786         struct d40_chan *d40c;
2787
2788         INIT_LIST_HEAD(&dma->channels);
2789
2790         for (i = offset; i < offset + num_chans; i++) {
2791                 d40c = &chans[i];
2792                 d40c->base = base;
2793                 d40c->chan.device = dma;
2794
2795                 spin_lock_init(&d40c->lock);
2796
2797                 d40c->log_num = D40_PHY_CHAN;
2798
2799                 INIT_LIST_HEAD(&d40c->done);
2800                 INIT_LIST_HEAD(&d40c->active);
2801                 INIT_LIST_HEAD(&d40c->queue);
2802                 INIT_LIST_HEAD(&d40c->pending_queue);
2803                 INIT_LIST_HEAD(&d40c->client);
2804                 INIT_LIST_HEAD(&d40c->prepare_queue);
2805
2806                 tasklet_setup(&d40c->tasklet, dma_tasklet);
2807
2808                 list_add_tail(&d40c->chan.device_node,
2809                               &dma->channels);
2810         }
2811 }
2812
2813 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2814 {
2815         if (dma_has_cap(DMA_SLAVE, dev->cap_mask)) {
2816                 dev->device_prep_slave_sg = d40_prep_slave_sg;
2817                 dev->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
2818         }
2819
2820         if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2821                 dev->device_prep_dma_memcpy = d40_prep_memcpy;
2822                 dev->directions = BIT(DMA_MEM_TO_MEM);
2823                 /*
2824                  * This controller can only access address at even
2825                  * 32bit boundaries, i.e. 2^2
2826                  */
2827                 dev->copy_align = DMAENGINE_ALIGN_4_BYTES;
2828         }
2829
2830         if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2831                 dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2832
2833         dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2834         dev->device_free_chan_resources = d40_free_chan_resources;
2835         dev->device_issue_pending = d40_issue_pending;
2836         dev->device_tx_status = d40_tx_status;
2837         dev->device_config = d40_set_runtime_config;
2838         dev->device_pause = d40_pause;
2839         dev->device_resume = d40_resume;
2840         dev->device_terminate_all = d40_terminate_all;
2841         dev->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
2842         dev->dev = base->dev;
2843 }
2844
2845 static int __init d40_dmaengine_init(struct d40_base *base,
2846                                      int num_reserved_chans)
2847 {
2848         int err ;
2849
2850         d40_chan_init(base, &base->dma_slave, base->log_chans,
2851                       0, base->num_log_chans);
2852
2853         dma_cap_zero(base->dma_slave.cap_mask);
2854         dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2855         dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2856
2857         d40_ops_init(base, &base->dma_slave);
2858
2859         err = dmaenginem_async_device_register(&base->dma_slave);
2860
2861         if (err) {
2862                 d40_err(base->dev, "Failed to register slave channels\n");
2863                 goto exit;
2864         }
2865
2866         d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2867                       base->num_log_chans, base->num_memcpy_chans);
2868
2869         dma_cap_zero(base->dma_memcpy.cap_mask);
2870         dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2871
2872         d40_ops_init(base, &base->dma_memcpy);
2873
2874         err = dmaenginem_async_device_register(&base->dma_memcpy);
2875
2876         if (err) {
2877                 d40_err(base->dev,
2878                         "Failed to register memcpy only channels\n");
2879                 goto exit;
2880         }
2881
2882         d40_chan_init(base, &base->dma_both, base->phy_chans,
2883                       0, num_reserved_chans);
2884
2885         dma_cap_zero(base->dma_both.cap_mask);
2886         dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2887         dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2888         dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2889
2890         d40_ops_init(base, &base->dma_both);
2891         err = dmaenginem_async_device_register(&base->dma_both);
2892
2893         if (err) {
2894                 d40_err(base->dev,
2895                         "Failed to register logical and physical capable channels\n");
2896                 goto exit;
2897         }
2898         return 0;
2899  exit:
2900         return err;
2901 }
2902
2903 /* Suspend resume functionality */
2904 #ifdef CONFIG_PM_SLEEP
2905 static int dma40_suspend(struct device *dev)
2906 {
2907         struct d40_base *base = dev_get_drvdata(dev);
2908         int ret;
2909
2910         ret = pm_runtime_force_suspend(dev);
2911         if (ret)
2912                 return ret;
2913
2914         if (base->lcpa_regulator)
2915                 ret = regulator_disable(base->lcpa_regulator);
2916         return ret;
2917 }
2918
2919 static int dma40_resume(struct device *dev)
2920 {
2921         struct d40_base *base = dev_get_drvdata(dev);
2922         int ret = 0;
2923
2924         if (base->lcpa_regulator) {
2925                 ret = regulator_enable(base->lcpa_regulator);
2926                 if (ret)
2927                         return ret;
2928         }
2929
2930         return pm_runtime_force_resume(dev);
2931 }
2932 #endif
2933
2934 #ifdef CONFIG_PM
2935 static void dma40_backup(void __iomem *baseaddr, u32 *backup,
2936                          u32 *regaddr, int num, bool save)
2937 {
2938         int i;
2939
2940         for (i = 0; i < num; i++) {
2941                 void __iomem *addr = baseaddr + regaddr[i];
2942
2943                 if (save)
2944                         backup[i] = readl_relaxed(addr);
2945                 else
2946                         writel_relaxed(backup[i], addr);
2947         }
2948 }
2949
2950 static void d40_save_restore_registers(struct d40_base *base, bool save)
2951 {
2952         int i;
2953
2954         /* Save/Restore channel specific registers */
2955         for (i = 0; i < base->num_phy_chans; i++) {
2956                 void __iomem *addr;
2957                 int idx;
2958
2959                 if (base->phy_res[i].reserved)
2960                         continue;
2961
2962                 addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA;
2963                 idx = i * ARRAY_SIZE(d40_backup_regs_chan);
2964
2965                 dma40_backup(addr, &base->reg_val_backup_chan[idx],
2966                              d40_backup_regs_chan,
2967                              ARRAY_SIZE(d40_backup_regs_chan),
2968                              save);
2969         }
2970
2971         /* Save/Restore global registers */
2972         dma40_backup(base->virtbase, base->reg_val_backup,
2973                      d40_backup_regs, ARRAY_SIZE(d40_backup_regs),
2974                      save);
2975
2976         /* Save/Restore registers only existing on dma40 v3 and later */
2977         if (base->gen_dmac.backup)
2978                 dma40_backup(base->virtbase, base->reg_val_backup_v4,
2979                              base->gen_dmac.backup,
2980                         base->gen_dmac.backup_size,
2981                         save);
2982 }
2983
2984 static int dma40_runtime_suspend(struct device *dev)
2985 {
2986         struct d40_base *base = dev_get_drvdata(dev);
2987
2988         d40_save_restore_registers(base, true);
2989
2990         /* Don't disable/enable clocks for v1 due to HW bugs */
2991         if (base->rev != 1)
2992                 writel_relaxed(base->gcc_pwr_off_mask,
2993                                base->virtbase + D40_DREG_GCC);
2994
2995         return 0;
2996 }
2997
2998 static int dma40_runtime_resume(struct device *dev)
2999 {
3000         struct d40_base *base = dev_get_drvdata(dev);
3001
3002         d40_save_restore_registers(base, false);
3003
3004         writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
3005                        base->virtbase + D40_DREG_GCC);
3006         return 0;
3007 }
3008 #endif
3009
3010 static const struct dev_pm_ops dma40_pm_ops = {
3011         SET_LATE_SYSTEM_SLEEP_PM_OPS(dma40_suspend, dma40_resume)
3012         SET_RUNTIME_PM_OPS(dma40_runtime_suspend,
3013                                 dma40_runtime_resume,
3014                                 NULL)
3015 };
3016
3017 /* Initialization functions. */
3018
3019 static int __init d40_phy_res_init(struct d40_base *base)
3020 {
3021         int i;
3022         int num_phy_chans_avail = 0;
3023         u32 val[2];
3024         int odd_even_bit = -2;
3025         int gcc = D40_DREG_GCC_ENA;
3026
3027         val[0] = readl(base->virtbase + D40_DREG_PRSME);
3028         val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3029
3030         for (i = 0; i < base->num_phy_chans; i++) {
3031                 base->phy_res[i].num = i;
3032                 odd_even_bit += 2 * ((i % 2) == 0);
3033                 if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3034                         /* Mark security only channels as occupied */
3035                         base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3036                         base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3037                         base->phy_res[i].reserved = true;
3038                         gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3039                                                        D40_DREG_GCC_SRC);
3040                         gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3041                                                        D40_DREG_GCC_DST);
3042
3043
3044                 } else {
3045                         base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3046                         base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3047                         base->phy_res[i].reserved = false;
3048                         num_phy_chans_avail++;
3049                 }
3050                 spin_lock_init(&base->phy_res[i].lock);
3051         }
3052
3053         /* Mark disabled channels as occupied */
3054         for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3055                 int chan = base->plat_data->disabled_channels[i];
3056
3057                 base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3058                 base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3059                 base->phy_res[chan].reserved = true;
3060                 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3061                                                D40_DREG_GCC_SRC);
3062                 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3063                                                D40_DREG_GCC_DST);
3064                 num_phy_chans_avail--;
3065         }
3066
3067         /* Mark soft_lli channels */
3068         for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3069                 int chan = base->plat_data->soft_lli_chans[i];
3070
3071                 base->phy_res[chan].use_soft_lli = true;
3072         }
3073
3074         dev_info(base->dev, "%d of %d physical DMA channels available\n",
3075                  num_phy_chans_avail, base->num_phy_chans);
3076
3077         /* Verify settings extended vs standard */
3078         val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3079
3080         for (i = 0; i < base->num_phy_chans; i++) {
3081
3082                 if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3083                     (val[0] & 0x3) != 1)
3084                         dev_info(base->dev,
3085                                  "[%s] INFO: channel %d is misconfigured (%d)\n",
3086                                  __func__, i, val[0] & 0x3);
3087
3088                 val[0] = val[0] >> 2;
3089         }
3090
3091         /*
3092          * To keep things simple, Enable all clocks initially.
3093          * The clocks will get managed later post channel allocation.
3094          * The clocks for the event lines on which reserved channels exists
3095          * are not managed here.
3096          */
3097         writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3098         base->gcc_pwr_off_mask = gcc;
3099
3100         return num_phy_chans_avail;
3101 }
3102
3103 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
3104 {
3105         struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3106         struct clk *clk;
3107         void __iomem *virtbase;
3108         struct resource *res;
3109         struct d40_base *base;
3110         int num_log_chans;
3111         int num_phy_chans;
3112         int num_memcpy_chans;
3113         int clk_ret = -EINVAL;
3114         int i;
3115         u32 pid;
3116         u32 cid;
3117         u8 rev;
3118
3119         clk = clk_get(&pdev->dev, NULL);
3120         if (IS_ERR(clk)) {
3121                 d40_err(&pdev->dev, "No matching clock found\n");
3122                 goto check_prepare_enabled;
3123         }
3124
3125         clk_ret = clk_prepare_enable(clk);
3126         if (clk_ret) {
3127                 d40_err(&pdev->dev, "Failed to prepare/enable clock\n");
3128                 goto disable_unprepare;
3129         }
3130
3131         /* Get IO for DMAC base address */
3132         res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
3133         if (!res)
3134                 goto disable_unprepare;
3135
3136         if (request_mem_region(res->start, resource_size(res),
3137                                D40_NAME " I/O base") == NULL)
3138                 goto release_region;
3139
3140         virtbase = ioremap(res->start, resource_size(res));
3141         if (!virtbase)
3142                 goto release_region;
3143
3144         /* This is just a regular AMBA PrimeCell ID actually */
3145         for (pid = 0, i = 0; i < 4; i++)
3146                 pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
3147                         & 255) << (i * 8);
3148         for (cid = 0, i = 0; i < 4; i++)
3149                 cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
3150                         & 255) << (i * 8);
3151
3152         if (cid != AMBA_CID) {
3153                 d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
3154                 goto unmap_io;
3155         }
3156         if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3157                 d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
3158                         AMBA_MANF_BITS(pid),
3159                         AMBA_VENDOR_ST);
3160                 goto unmap_io;
3161         }
3162         /*
3163          * HW revision:
3164          * DB8500ed has revision 0
3165          * ? has revision 1
3166          * DB8500v1 has revision 2
3167          * DB8500v2 has revision 3
3168          * AP9540v1 has revision 4
3169          * DB8540v1 has revision 4
3170          */
3171         rev = AMBA_REV_BITS(pid);
3172         if (rev < 2) {
3173                 d40_err(&pdev->dev, "hardware revision: %d is not supported", rev);
3174                 goto unmap_io;
3175         }
3176
3177         /* The number of physical channels on this HW */
3178         if (plat_data->num_of_phy_chans)
3179                 num_phy_chans = plat_data->num_of_phy_chans;
3180         else
3181                 num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3182
3183         /* The number of channels used for memcpy */
3184         if (plat_data->num_of_memcpy_chans)
3185                 num_memcpy_chans = plat_data->num_of_memcpy_chans;
3186         else
3187                 num_memcpy_chans = ARRAY_SIZE(dma40_memcpy_channels);
3188
3189         num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY;
3190
3191         dev_info(&pdev->dev,
3192                  "hardware rev: %d @ %pa with %d physical and %d logical channels\n",
3193                  rev, &res->start, num_phy_chans, num_log_chans);
3194
3195         base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
3196                        (num_phy_chans + num_log_chans + num_memcpy_chans) *
3197                        sizeof(struct d40_chan), GFP_KERNEL);
3198
3199         if (base == NULL)
3200                 goto unmap_io;
3201
3202         base->rev = rev;
3203         base->clk = clk;
3204         base->num_memcpy_chans = num_memcpy_chans;
3205         base->num_phy_chans = num_phy_chans;
3206         base->num_log_chans = num_log_chans;
3207         base->phy_start = res->start;
3208         base->phy_size = resource_size(res);
3209         base->virtbase = virtbase;
3210         base->plat_data = plat_data;
3211         base->dev = &pdev->dev;
3212         base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3213         base->log_chans = &base->phy_chans[num_phy_chans];
3214
3215         if (base->plat_data->num_of_phy_chans == 14) {
3216                 base->gen_dmac.backup = d40_backup_regs_v4b;
3217                 base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3218                 base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3219                 base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3220                 base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3221                 base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3222                 base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3223                 base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3224                 base->gen_dmac.il = il_v4b;
3225                 base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3226                 base->gen_dmac.init_reg = dma_init_reg_v4b;
3227                 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3228         } else {
3229                 if (base->rev >= 3) {
3230                         base->gen_dmac.backup = d40_backup_regs_v4a;
3231                         base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3232                 }
3233                 base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3234                 base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3235                 base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3236                 base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3237                 base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3238                 base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3239                 base->gen_dmac.il = il_v4a;
3240                 base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3241                 base->gen_dmac.init_reg = dma_init_reg_v4a;
3242                 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3243         }
3244
3245         base->phy_res = kcalloc(num_phy_chans,
3246                                 sizeof(*base->phy_res),
3247                                 GFP_KERNEL);
3248         if (!base->phy_res)
3249                 goto free_base;
3250
3251         base->lookup_phy_chans = kcalloc(num_phy_chans,
3252                                          sizeof(*base->lookup_phy_chans),
3253                                          GFP_KERNEL);
3254         if (!base->lookup_phy_chans)
3255                 goto free_phy_res;
3256
3257         base->lookup_log_chans = kcalloc(num_log_chans,
3258                                          sizeof(*base->lookup_log_chans),
3259                                          GFP_KERNEL);
3260         if (!base->lookup_log_chans)
3261                 goto free_phy_chans;
3262
3263         base->reg_val_backup_chan = kmalloc_array(base->num_phy_chans,
3264                                                   sizeof(d40_backup_regs_chan),
3265                                                   GFP_KERNEL);
3266         if (!base->reg_val_backup_chan)
3267                 goto free_log_chans;
3268
3269         base->lcla_pool.alloc_map = kcalloc(num_phy_chans
3270                                             * D40_LCLA_LINK_PER_EVENT_GRP,
3271                                             sizeof(*base->lcla_pool.alloc_map),
3272                                             GFP_KERNEL);
3273         if (!base->lcla_pool.alloc_map)
3274                 goto free_backup_chan;
3275
3276         base->regs_interrupt = kmalloc_array(base->gen_dmac.il_size,
3277                                              sizeof(*base->regs_interrupt),
3278                                              GFP_KERNEL);
3279         if (!base->regs_interrupt)
3280                 goto free_map;
3281
3282         base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3283                                             0, SLAB_HWCACHE_ALIGN,
3284                                             NULL);
3285         if (base->desc_slab == NULL)
3286                 goto free_regs;
3287
3288
3289         return base;
3290  free_regs:
3291         kfree(base->regs_interrupt);
3292  free_map:
3293         kfree(base->lcla_pool.alloc_map);
3294  free_backup_chan:
3295         kfree(base->reg_val_backup_chan);
3296  free_log_chans:
3297         kfree(base->lookup_log_chans);
3298  free_phy_chans:
3299         kfree(base->lookup_phy_chans);
3300  free_phy_res:
3301         kfree(base->phy_res);
3302  free_base:
3303         kfree(base);
3304  unmap_io:
3305         iounmap(virtbase);
3306  release_region:
3307         release_mem_region(res->start, resource_size(res));
3308  check_prepare_enabled:
3309         if (!clk_ret)
3310  disable_unprepare:
3311                 clk_disable_unprepare(clk);
3312         if (!IS_ERR(clk))
3313                 clk_put(clk);
3314         return NULL;
3315 }
3316
3317 static void __init d40_hw_init(struct d40_base *base)
3318 {
3319
3320         int i;
3321         u32 prmseo[2] = {0, 0};
3322         u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3323         u32 pcmis = 0;
3324         u32 pcicr = 0;
3325         struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3326         u32 reg_size = base->gen_dmac.init_reg_size;
3327
3328         for (i = 0; i < reg_size; i++)
3329                 writel(dma_init_reg[i].val,
3330                        base->virtbase + dma_init_reg[i].reg);
3331
3332         /* Configure all our dma channels to default settings */
3333         for (i = 0; i < base->num_phy_chans; i++) {
3334
3335                 activeo[i % 2] = activeo[i % 2] << 2;
3336
3337                 if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3338                     == D40_ALLOC_PHY) {
3339                         activeo[i % 2] |= 3;
3340                         continue;
3341                 }
3342
3343                 /* Enable interrupt # */
3344                 pcmis = (pcmis << 1) | 1;
3345
3346                 /* Clear interrupt # */
3347                 pcicr = (pcicr << 1) | 1;
3348
3349                 /* Set channel to physical mode */
3350                 prmseo[i % 2] = prmseo[i % 2] << 2;
3351                 prmseo[i % 2] |= 1;
3352
3353         }
3354
3355         writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3356         writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3357         writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3358         writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3359
3360         /* Write which interrupt to enable */
3361         writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3362
3363         /* Write which interrupt to clear */
3364         writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3365
3366         /* These are __initdata and cannot be accessed after init */
3367         base->gen_dmac.init_reg = NULL;
3368         base->gen_dmac.init_reg_size = 0;
3369 }
3370
3371 static int __init d40_lcla_allocate(struct d40_base *base)
3372 {
3373         struct d40_lcla_pool *pool = &base->lcla_pool;
3374         unsigned long *page_list;
3375         int i, j;
3376         int ret;
3377
3378         /*
3379          * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3380          * To full fill this hardware requirement without wasting 256 kb
3381          * we allocate pages until we get an aligned one.
3382          */
3383         page_list = kmalloc_array(MAX_LCLA_ALLOC_ATTEMPTS,
3384                                   sizeof(*page_list),
3385                                   GFP_KERNEL);
3386         if (!page_list)
3387                 return -ENOMEM;
3388
3389         /* Calculating how many pages that are required */
3390         base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3391
3392         for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3393                 page_list[i] = __get_free_pages(GFP_KERNEL,
3394                                                 base->lcla_pool.pages);
3395                 if (!page_list[i]) {
3396
3397                         d40_err(base->dev, "Failed to allocate %d pages.\n",
3398                                 base->lcla_pool.pages);
3399                         ret = -ENOMEM;
3400
3401                         for (j = 0; j < i; j++)
3402                                 free_pages(page_list[j], base->lcla_pool.pages);
3403                         goto free_page_list;
3404                 }
3405
3406                 if ((virt_to_phys((void *)page_list[i]) &
3407                      (LCLA_ALIGNMENT - 1)) == 0)
3408                         break;
3409         }
3410
3411         for (j = 0; j < i; j++)
3412                 free_pages(page_list[j], base->lcla_pool.pages);
3413
3414         if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3415                 base->lcla_pool.base = (void *)page_list[i];
3416         } else {
3417                 /*
3418                  * After many attempts and no succees with finding the correct
3419                  * alignment, try with allocating a big buffer.
3420                  */
3421                 dev_warn(base->dev,
3422                          "[%s] Failed to get %d pages @ 18 bit align.\n",
3423                          __func__, base->lcla_pool.pages);
3424                 base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3425                                                          base->num_phy_chans +
3426                                                          LCLA_ALIGNMENT,
3427                                                          GFP_KERNEL);
3428                 if (!base->lcla_pool.base_unaligned) {
3429                         ret = -ENOMEM;
3430                         goto free_page_list;
3431                 }
3432
3433                 base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3434                                                  LCLA_ALIGNMENT);
3435         }
3436
3437         pool->dma_addr = dma_map_single(base->dev, pool->base,
3438                                         SZ_1K * base->num_phy_chans,
3439                                         DMA_TO_DEVICE);
3440         if (dma_mapping_error(base->dev, pool->dma_addr)) {
3441                 pool->dma_addr = 0;
3442                 ret = -ENOMEM;
3443                 goto free_page_list;
3444         }
3445
3446         writel(virt_to_phys(base->lcla_pool.base),
3447                base->virtbase + D40_DREG_LCLA);
3448         ret = 0;
3449  free_page_list:
3450         kfree(page_list);
3451         return ret;
3452 }
3453
3454 static int __init d40_of_probe(struct platform_device *pdev,
3455                                struct device_node *np)
3456 {
3457         struct stedma40_platform_data *pdata;
3458         int num_phy = 0, num_memcpy = 0, num_disabled = 0;
3459         const __be32 *list;
3460
3461         pdata = devm_kzalloc(&pdev->dev, sizeof(*pdata), GFP_KERNEL);
3462         if (!pdata)
3463                 return -ENOMEM;
3464
3465         /* If absent this value will be obtained from h/w. */
3466         of_property_read_u32(np, "dma-channels", &num_phy);
3467         if (num_phy > 0)
3468                 pdata->num_of_phy_chans = num_phy;
3469
3470         list = of_get_property(np, "memcpy-channels", &num_memcpy);
3471         num_memcpy /= sizeof(*list);
3472
3473         if (num_memcpy > D40_MEMCPY_MAX_CHANS || num_memcpy <= 0) {
3474                 d40_err(&pdev->dev,
3475                         "Invalid number of memcpy channels specified (%d)\n",
3476                         num_memcpy);
3477                 return -EINVAL;
3478         }
3479         pdata->num_of_memcpy_chans = num_memcpy;
3480
3481         of_property_read_u32_array(np, "memcpy-channels",
3482                                    dma40_memcpy_channels,
3483                                    num_memcpy);
3484
3485         list = of_get_property(np, "disabled-channels", &num_disabled);
3486         num_disabled /= sizeof(*list);
3487
3488         if (num_disabled >= STEDMA40_MAX_PHYS || num_disabled < 0) {
3489                 d40_err(&pdev->dev,
3490                         "Invalid number of disabled channels specified (%d)\n",
3491                         num_disabled);
3492                 return -EINVAL;
3493         }
3494
3495         of_property_read_u32_array(np, "disabled-channels",
3496                                    pdata->disabled_channels,
3497                                    num_disabled);
3498         pdata->disabled_channels[num_disabled] = -1;
3499
3500         pdev->dev.platform_data = pdata;
3501
3502         return 0;
3503 }
3504
3505 static int __init d40_probe(struct platform_device *pdev)
3506 {
3507         struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3508         struct device_node *np = pdev->dev.of_node;
3509         int ret = -ENOENT;
3510         struct d40_base *base;
3511         struct resource *res;
3512         int num_reserved_chans;
3513         u32 val;
3514
3515         if (!plat_data) {
3516                 if (np) {
3517                         if (d40_of_probe(pdev, np)) {
3518                                 ret = -ENOMEM;
3519                                 goto report_failure;
3520                         }
3521                 } else {
3522                         d40_err(&pdev->dev, "No pdata or Device Tree provided\n");
3523                         goto report_failure;
3524                 }
3525         }
3526
3527         base = d40_hw_detect_init(pdev);
3528         if (!base)
3529                 goto report_failure;
3530
3531         num_reserved_chans = d40_phy_res_init(base);
3532
3533         platform_set_drvdata(pdev, base);
3534
3535         spin_lock_init(&base->interrupt_lock);
3536         spin_lock_init(&base->execmd_lock);
3537
3538         /* Get IO for logical channel parameter address */
3539         res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
3540         if (!res) {
3541                 ret = -ENOENT;
3542                 d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
3543                 goto destroy_cache;
3544         }
3545         base->lcpa_size = resource_size(res);
3546         base->phy_lcpa = res->start;
3547
3548         if (request_mem_region(res->start, resource_size(res),
3549                                D40_NAME " I/O lcpa") == NULL) {
3550                 ret = -EBUSY;
3551                 d40_err(&pdev->dev, "Failed to request LCPA region %pR\n", res);
3552                 goto destroy_cache;
3553         }
3554
3555         /* We make use of ESRAM memory for this. */
3556         val = readl(base->virtbase + D40_DREG_LCPA);
3557         if (res->start != val && val != 0) {
3558                 dev_warn(&pdev->dev,
3559                          "[%s] Mismatch LCPA dma 0x%x, def %pa\n",
3560                          __func__, val, &res->start);
3561         } else
3562                 writel(res->start, base->virtbase + D40_DREG_LCPA);
3563
3564         base->lcpa_base = ioremap(res->start, resource_size(res));
3565         if (!base->lcpa_base) {
3566                 ret = -ENOMEM;
3567                 d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
3568                 goto destroy_cache;
3569         }
3570         /* If lcla has to be located in ESRAM we don't need to allocate */
3571         if (base->plat_data->use_esram_lcla) {
3572                 res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3573                                                         "lcla_esram");
3574                 if (!res) {
3575                         ret = -ENOENT;
3576                         d40_err(&pdev->dev,
3577                                 "No \"lcla_esram\" memory resource\n");
3578                         goto destroy_cache;
3579                 }
3580                 base->lcla_pool.base = ioremap(res->start,
3581                                                 resource_size(res));
3582                 if (!base->lcla_pool.base) {
3583                         ret = -ENOMEM;
3584                         d40_err(&pdev->dev, "Failed to ioremap LCLA region\n");
3585                         goto destroy_cache;
3586                 }
3587                 writel(res->start, base->virtbase + D40_DREG_LCLA);
3588
3589         } else {
3590                 ret = d40_lcla_allocate(base);
3591                 if (ret) {
3592                         d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
3593                         goto destroy_cache;
3594                 }
3595         }
3596
3597         spin_lock_init(&base->lcla_pool.lock);
3598
3599         base->irq = platform_get_irq(pdev, 0);
3600
3601         ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3602         if (ret) {
3603                 d40_err(&pdev->dev, "No IRQ defined\n");
3604                 goto destroy_cache;
3605         }
3606
3607         if (base->plat_data->use_esram_lcla) {
3608
3609                 base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3610                 if (IS_ERR(base->lcpa_regulator)) {
3611                         d40_err(&pdev->dev, "Failed to get lcpa_regulator\n");
3612                         ret = PTR_ERR(base->lcpa_regulator);
3613                         base->lcpa_regulator = NULL;
3614                         goto destroy_cache;
3615                 }
3616
3617                 ret = regulator_enable(base->lcpa_regulator);
3618                 if (ret) {
3619                         d40_err(&pdev->dev,
3620                                 "Failed to enable lcpa_regulator\n");
3621                         regulator_put(base->lcpa_regulator);
3622                         base->lcpa_regulator = NULL;
3623                         goto destroy_cache;
3624                 }
3625         }
3626
3627         writel_relaxed(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3628
3629         pm_runtime_irq_safe(base->dev);
3630         pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3631         pm_runtime_use_autosuspend(base->dev);
3632         pm_runtime_mark_last_busy(base->dev);
3633         pm_runtime_set_active(base->dev);
3634         pm_runtime_enable(base->dev);
3635
3636         ret = d40_dmaengine_init(base, num_reserved_chans);
3637         if (ret)
3638                 goto destroy_cache;
3639
3640         ret = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3641         if (ret) {
3642                 d40_err(&pdev->dev, "Failed to set dma max seg size\n");
3643                 goto destroy_cache;
3644         }
3645
3646         d40_hw_init(base);
3647
3648         if (np) {
3649                 ret = of_dma_controller_register(np, d40_xlate, NULL);
3650                 if (ret)
3651                         dev_err(&pdev->dev,
3652                                 "could not register of_dma_controller\n");
3653         }
3654
3655         dev_info(base->dev, "initialized\n");
3656         return 0;
3657  destroy_cache:
3658         kmem_cache_destroy(base->desc_slab);
3659         if (base->virtbase)
3660                 iounmap(base->virtbase);
3661
3662         if (base->lcla_pool.base && base->plat_data->use_esram_lcla) {
3663                 iounmap(base->lcla_pool.base);
3664                 base->lcla_pool.base = NULL;
3665         }
3666
3667         if (base->lcla_pool.dma_addr)
3668                 dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3669                                  SZ_1K * base->num_phy_chans,
3670                                  DMA_TO_DEVICE);
3671
3672         if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3673                 free_pages((unsigned long)base->lcla_pool.base,
3674                            base->lcla_pool.pages);
3675
3676         kfree(base->lcla_pool.base_unaligned);
3677
3678         if (base->lcpa_base)
3679                 iounmap(base->lcpa_base);
3680
3681         if (base->phy_lcpa)
3682                 release_mem_region(base->phy_lcpa,
3683                                    base->lcpa_size);
3684         if (base->phy_start)
3685                 release_mem_region(base->phy_start,
3686                                    base->phy_size);
3687         if (base->clk) {
3688                 clk_disable_unprepare(base->clk);
3689                 clk_put(base->clk);
3690         }
3691
3692         if (base->lcpa_regulator) {
3693                 regulator_disable(base->lcpa_regulator);
3694                 regulator_put(base->lcpa_regulator);
3695         }
3696
3697         kfree(base->lcla_pool.alloc_map);
3698         kfree(base->lookup_log_chans);
3699         kfree(base->lookup_phy_chans);
3700         kfree(base->phy_res);
3701         kfree(base);
3702  report_failure:
3703         d40_err(&pdev->dev, "probe failed\n");
3704         return ret;
3705 }
3706
3707 static const struct of_device_id d40_match[] = {
3708         { .compatible = "stericsson,dma40", },
3709         {}
3710 };
3711
3712 static struct platform_driver d40_driver = {
3713         .driver = {
3714                 .name  = D40_NAME,
3715                 .pm = &dma40_pm_ops,
3716                 .of_match_table = d40_match,
3717         },
3718 };
3719
3720 static int __init stedma40_init(void)
3721 {
3722         return platform_driver_probe(&d40_driver, d40_probe);
3723 }
3724 subsys_initcall(stedma40_init);