drivers/dma/ste_dma40.c

   1 /*
   2  * Copyright (C) Ericsson AB 2007-2008
   3  * Copyright (C) ST-Ericsson SA 2008-2010
   4  * Author: Per Forlin <per.forlin@stericsson.com> for ST-Ericsson
   5  * Author: Jonas Aaberg <jonas.aberg@stericsson.com> for ST-Ericsson
   6  * License terms: GNU General Public License (GPL) version 2
   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/pm.h>
  18 #include <linux/pm_runtime.h>
  19 #include <linux/err.h>
  20 #include <linux/amba/bus.h>
  21
  22 #include <plat/ste_dma40.h>
  23
  24 #include "ste_dma40_ll.h"
  25
  26 #define D40_NAME "dma40"
  27
  28 #define D40_PHY_CHAN -1
  29
  30 /* For masking out/in 2 bit channel positions */
  31 #define D40_CHAN_POS(chan)  (2 * (chan / 2))
  32 #define D40_CHAN_POS_MASK(chan) (0x3 << D40_CHAN_POS(chan))
  33
  34 /* Maximum iterations taken before giving up suspending a channel */
  35 #define D40_SUSPEND_MAX_IT 500
  36
  37 /* Milliseconds */
  38 #define DMA40_AUTOSUSPEND_DELAY 100
  39
  40 /* Hardware requirement on LCLA alignment */
  41 #define LCLA_ALIGNMENT 0x40000
  42
  43 /* Max number of links per event group */
  44 #define D40_LCLA_LINK_PER_EVENT_GRP 128
  45 #define D40_LCLA_END D40_LCLA_LINK_PER_EVENT_GRP
  46
  47 /* Attempts before giving up to trying to get pages that are aligned */
  48 #define MAX_LCLA_ALLOC_ATTEMPTS 256
  49
  50 /* Bit markings for allocation map */
  51 #define D40_ALLOC_FREE          (1 << 31)
  52 #define D40_ALLOC_PHY           (1 << 30)
  53 #define D40_ALLOC_LOG_FREE      0
  54
  55 /**
  56  * enum 40_command - The different commands and/or statuses.
  57  *
  58  * @D40_DMA_STOP: DMA channel command STOP or status STOPPED,
  59  * @D40_DMA_RUN: The DMA channel is RUNNING of the command RUN.
  60  * @D40_DMA_SUSPEND_REQ: Request the DMA to SUSPEND as soon as possible.
  61  * @D40_DMA_SUSPENDED: The DMA channel is SUSPENDED.
  62  */
  63 enum d40_command {
  64         D40_DMA_STOP            = 0,
  65         D40_DMA_RUN             = 1,
  66         D40_DMA_SUSPEND_REQ     = 2,
  67         D40_DMA_SUSPENDED       = 3
  68 };
  69
  70 /*
  71  * These are the registers that has to be saved and later restored
  72  * when the DMA hw is powered off.
  73  * TODO: Add save/restore of D40_DREG_GCC on dma40 v3 or later, if that works.
  74  */
  75 static u32 d40_backup_regs[] = {
  76         D40_DREG_LCPA,
  77         D40_DREG_LCLA,
  78         D40_DREG_PRMSE,
  79         D40_DREG_PRMSO,
  80         D40_DREG_PRMOE,
  81         D40_DREG_PRMOO,
  82 };
  83
  84 #define BACKUP_REGS_SZ ARRAY_SIZE(d40_backup_regs)
  85
  86 /* TODO: Check if all these registers have to be saved/restored on dma40 v3 */
  87 static u32 d40_backup_regs_v3[] = {
  88         D40_DREG_PSEG1,
  89         D40_DREG_PSEG2,
  90         D40_DREG_PSEG3,
  91         D40_DREG_PSEG4,
  92         D40_DREG_PCEG1,
  93         D40_DREG_PCEG2,
  94         D40_DREG_PCEG3,
  95         D40_DREG_PCEG4,
  96         D40_DREG_RSEG1,
  97         D40_DREG_RSEG2,
  98         D40_DREG_RSEG3,
  99         D40_DREG_RSEG4,
 100         D40_DREG_RCEG1,
 101         D40_DREG_RCEG2,
 102         D40_DREG_RCEG3,
 103         D40_DREG_RCEG4,
 104 };
 105
 106 #define BACKUP_REGS_SZ_V3 ARRAY_SIZE(d40_backup_regs_v3)
 107
 108 static u32 d40_backup_regs_chan[] = {
 109         D40_CHAN_REG_SSCFG,
 110         D40_CHAN_REG_SSELT,
 111         D40_CHAN_REG_SSPTR,
 112         D40_CHAN_REG_SSLNK,
 113         D40_CHAN_REG_SDCFG,
 114         D40_CHAN_REG_SDELT,
 115         D40_CHAN_REG_SDPTR,
 116         D40_CHAN_REG_SDLNK,
 117 };
 118
 119 /**
 120  * struct d40_lli_pool - Structure for keeping LLIs in memory
 121  *
 122  * @base: Pointer to memory area when the pre_alloc_lli's are not large
 123  * enough, IE bigger than the most common case, 1 dst and 1 src. NULL if
 124  * pre_alloc_lli is used.
 125  * @dma_addr: DMA address, if mapped
 126  * @size: The size in bytes of the memory at base or the size of pre_alloc_lli.
 127  * @pre_alloc_lli: Pre allocated area for the most common case of transfers,
 128  * one buffer to one buffer.
 129  */
 130 struct d40_lli_pool {
 131         void    *base;
 132         int      size;
 133         dma_addr_t      dma_addr;
 134         /* Space for dst and src, plus an extra for padding */
 135         u8       pre_alloc_lli[3 * sizeof(struct d40_phy_lli)];
 136 };
 137
 138 /**
 139  * struct d40_desc - A descriptor is one DMA job.
 140  *
 141  * @lli_phy: LLI settings for physical channel. Both src and dst=
 142  * points into the lli_pool, to base if lli_len > 1 or to pre_alloc_lli if
 143  * lli_len equals one.
 144  * @lli_log: Same as above but for logical channels.
 145  * @lli_pool: The pool with two entries pre-allocated.
 146  * @lli_len: Number of llis of current descriptor.
 147  * @lli_current: Number of transferred llis.
 148  * @lcla_alloc: Number of LCLA entries allocated.
 149  * @txd: DMA engine struct. Used for among other things for communication
 150  * during a transfer.
 151  * @node: List entry.
 152  * @is_in_client_list: true if the client owns this descriptor.
 153  * @cyclic: true if this is a cyclic job
 154  *
 155  * This descriptor is used for both logical and physical transfers.
 156  */
 157 struct d40_desc {
 158         /* LLI physical */
 159         struct d40_phy_lli_bidir         lli_phy;
 160         /* LLI logical */
 161         struct d40_log_lli_bidir         lli_log;
 162
 163         struct d40_lli_pool              lli_pool;
 164         int                              lli_len;
 165         int                              lli_current;
 166         int                              lcla_alloc;
 167
 168         struct dma_async_tx_descriptor   txd;
 169         struct list_head                 node;
 170
 171         bool                             is_in_client_list;
 172         bool                             cyclic;
 173 };
 174
 175 /**
 176  * struct d40_lcla_pool - LCLA pool settings and data.
 177  *
 178  * @base: The virtual address of LCLA. 18 bit aligned.
 179  * @base_unaligned: The orignal kmalloc pointer, if kmalloc is used.
 180  * This pointer is only there for clean-up on error.
 181  * @pages: The number of pages needed for all physical channels.
 182  * Only used later for clean-up on error
 183  * @lock: Lock to protect the content in this struct.
 184  * @alloc_map: big map over which LCLA entry is own by which job.
 185  */
 186 struct d40_lcla_pool {
 187         void            *base;
 188         dma_addr_t      dma_addr;
 189         void            *base_unaligned;
 190         int              pages;
 191         spinlock_t       lock;
 192         struct d40_desc **alloc_map;
 193 };
 194
 195 /**
 196  * struct d40_phy_res - struct for handling eventlines mapped to physical
 197  * channels.
 198  *
 199  * @lock: A lock protection this entity.
 200  * @reserved: True if used by secure world or otherwise.
 201  * @num: The physical channel number of this entity.
 202  * @allocated_src: Bit mapped to show which src event line's are mapped to
 203  * this physical channel. Can also be free or physically allocated.
 204  * @allocated_dst: Same as for src but is dst.
 205  * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as
 206  * event line number.
 207  */
 208 struct d40_phy_res {
 209         spinlock_t lock;
 210         bool       reserved;
 211         int        num;
 212         u32        allocated_src;
 213         u32        allocated_dst;
 214 };
 215
 216 struct d40_base;
 217
 218 /**
 219  * struct d40_chan - Struct that describes a channel.
 220  *
 221  * @lock: A spinlock to protect this struct.
 222  * @log_num: The logical number, if any of this channel.
 223  * @completed: Starts with 1, after first interrupt it is set to dma engine's
 224  * current cookie.
 225  * @pending_tx: The number of pending transfers. Used between interrupt handler
 226  * and tasklet.
 227  * @busy: Set to true when transfer is ongoing on this channel.
 228  * @phy_chan: Pointer to physical channel which this instance runs on. If this
 229  * point is NULL, then the channel is not allocated.
 230  * @chan: DMA engine handle.
 231  * @tasklet: Tasklet that gets scheduled from interrupt context to complete a
 232  * transfer and call client callback.
 233  * @client: Cliented owned descriptor list.
 234  * @pending_queue: Submitted jobs, to be issued by issue_pending()
 235  * @active: Active descriptor.
 236  * @queue: Queued jobs.
 237  * @prepare_queue: Prepared jobs.
 238  * @dma_cfg: The client configuration of this dma channel.
 239  * @configured: whether the dma_cfg configuration is valid
 240  * @base: Pointer to the device instance struct.
 241  * @src_def_cfg: Default cfg register setting for src.
 242  * @dst_def_cfg: Default cfg register setting for dst.
 243  * @log_def: Default logical channel settings.
 244  * @lcpa: Pointer to dst and src lcpa settings.
 245  * @runtime_addr: runtime configured address.
 246  * @runtime_direction: runtime configured direction.
 247  *
 248  * This struct can either "be" a logical or a physical channel.
 249  */
 250 struct d40_chan {
 251         spinlock_t                       lock;
 252         int                              log_num;
 253         /* ID of the most recent completed transfer */
 254         int                              completed;
 255         int                              pending_tx;
 256         bool                             busy;
 257         struct d40_phy_res              *phy_chan;
 258         struct dma_chan                  chan;
 259         struct tasklet_struct            tasklet;
 260         struct list_head                 client;
 261         struct list_head                 pending_queue;
 262         struct list_head                 active;
 263         struct list_head                 queue;
 264         struct list_head                 prepare_queue;
 265         struct stedma40_chan_cfg         dma_cfg;
 266         bool                             configured;
 267         struct d40_base                 *base;
 268         /* Default register configurations */
 269         u32                              src_def_cfg;
 270         u32                              dst_def_cfg;
 271         struct d40_def_lcsp              log_def;
 272         struct d40_log_lli_full         *lcpa;
 273         /* Runtime reconfiguration */
 274         dma_addr_t                      runtime_addr;
 275         enum dma_transfer_direction     runtime_direction;
 276 };
 277
 278 /**
 279  * struct d40_base - The big global struct, one for each probe'd instance.
 280  *
 281  * @interrupt_lock: Lock used to make sure one interrupt is handle a time.
 282  * @execmd_lock: Lock for execute command usage since several channels share
 283  * the same physical register.
 284  * @dev: The device structure.
 285  * @virtbase: The virtual base address of the DMA's register.
 286  * @rev: silicon revision detected.
 287  * @clk: Pointer to the DMA clock structure.
 288  * @phy_start: Physical memory start of the DMA registers.
 289  * @phy_size: Size of the DMA register map.
 290  * @irq: The IRQ number.
 291  * @num_phy_chans: The number of physical channels. Read from HW. This
 292  * is the number of available channels for this driver, not counting "Secure
 293  * mode" allocated physical channels.
 294  * @num_log_chans: The number of logical channels. Calculated from
 295  * num_phy_chans.
 296  * @dma_both: dma_device channels that can do both memcpy and slave transfers.
 297  * @dma_slave: dma_device channels that can do only do slave transfers.
 298  * @dma_memcpy: dma_device channels that can do only do memcpy transfers.
 299  * @phy_chans: Room for all possible physical channels in system.
 300  * @log_chans: Room for all possible logical channels in system.
 301  * @lookup_log_chans: Used to map interrupt number to logical channel. Points
 302  * to log_chans entries.
 303  * @lookup_phy_chans: Used to map interrupt number to physical channel. Points
 304  * to phy_chans entries.
 305  * @plat_data: Pointer to provided platform_data which is the driver
 306  * configuration.
 307  * @lcpa_regulator: Pointer to hold the regulator for the esram bank for lcla.
 308  * @phy_res: Vector containing all physical channels.
 309  * @lcla_pool: lcla pool settings and data.
 310  * @lcpa_base: The virtual mapped address of LCPA.
 311  * @phy_lcpa: The physical address of the LCPA.
 312  * @lcpa_size: The size of the LCPA area.
 313  * @desc_slab: cache for descriptors.
 314  * @reg_val_backup: Here the values of some hardware registers are stored
 315  * before the DMA is powered off. They are restored when the power is back on.
 316  * @reg_val_backup_v3: Backup of registers that only exits on dma40 v3 and
 317  * later.
 318  * @reg_val_backup_chan: Backup data for standard channel parameter registers.
 319  * @gcc_pwr_off_mask: Mask to maintain the channels that can be turned off.
 320  * @initialized: true if the dma has been initialized
 321  */
 322 struct d40_base {
 323         spinlock_t                       interrupt_lock;
 324         spinlock_t                       execmd_lock;
 325         struct device                    *dev;
 326         void __iomem                     *virtbase;
 327         u8                                rev:4;
 328         struct clk                       *clk;
 329         phys_addr_t                       phy_start;
 330         resource_size_t                   phy_size;
 331         int                               irq;
 332         int                               num_phy_chans;
 333         int                               num_log_chans;
 334         struct dma_device                 dma_both;
 335         struct dma_device                 dma_slave;
 336         struct dma_device                 dma_memcpy;
 337         struct d40_chan                  *phy_chans;
 338         struct d40_chan                  *log_chans;
 339         struct d40_chan                 **lookup_log_chans;
 340         struct d40_chan                 **lookup_phy_chans;
 341         struct stedma40_platform_data    *plat_data;
 342         struct regulator                 *lcpa_regulator;
 343         /* Physical half channels */
 344         struct d40_phy_res               *phy_res;
 345         struct d40_lcla_pool              lcla_pool;
 346         void                             *lcpa_base;
 347         dma_addr_t                        phy_lcpa;
 348         resource_size_t                   lcpa_size;
 349         struct kmem_cache                *desc_slab;
 350         u32                               reg_val_backup[BACKUP_REGS_SZ];
 351         u32                               reg_val_backup_v3[BACKUP_REGS_SZ_V3];
 352         u32                              *reg_val_backup_chan;
 353         u16                               gcc_pwr_off_mask;
 354         bool                              initialized;
 355 };
 356
 357 /**
 358  * struct d40_interrupt_lookup - lookup table for interrupt handler
 359  *
 360  * @src: Interrupt mask register.
 361  * @clr: Interrupt clear register.
 362  * @is_error: true if this is an error interrupt.
 363  * @offset: start delta in the lookup_log_chans in d40_base. If equals to
 364  * D40_PHY_CHAN, the lookup_phy_chans shall be used instead.
 365  */
 366 struct d40_interrupt_lookup {
 367         u32 src;
 368         u32 clr;
 369         bool is_error;
 370         int offset;
 371 };
 372
 373 /**
 374  * struct d40_reg_val - simple lookup struct
 375  *
 376  * @reg: The register.
 377  * @val: The value that belongs to the register in reg.
 378  */
 379 struct d40_reg_val {
 380         unsigned int reg;
 381         unsigned int val;
 382 };
 383
 384 static struct device *chan2dev(struct d40_chan *d40c)
 385 {
 386         return &d40c->chan.dev->device;
 387 }
 388
 389 static bool chan_is_physical(struct d40_chan *chan)
 390 {
 391         return chan->log_num == D40_PHY_CHAN;
 392 }
 393
 394 static bool chan_is_logical(struct d40_chan *chan)
 395 {
 396         return !chan_is_physical(chan);
 397 }
 398
 399 static void __iomem *chan_base(struct d40_chan *chan)
 400 {
 401         return chan->base->virtbase + D40_DREG_PCBASE +
 402                chan->phy_chan->num * D40_DREG_PCDELTA;
 403 }
 404
 405 #define d40_err(dev, format, arg...)            \
 406         dev_err(dev, "[%s] " format, __func__, ## arg)
 407
 408 #define chan_err(d40c, format, arg...)          \
 409         d40_err(chan2dev(d40c), format, ## arg)
 410
 411 static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d,
 412                               int lli_len)
 413 {
 414         bool is_log = chan_is_logical(d40c);
 415         u32 align;
 416         void *base;
 417
 418         if (is_log)
 419                 align = sizeof(struct d40_log_lli);
 420         else
 421                 align = sizeof(struct d40_phy_lli);
 422
 423         if (lli_len == 1) {
 424                 base = d40d->lli_pool.pre_alloc_lli;
 425                 d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
 426                 d40d->lli_pool.base = NULL;
 427         } else {
 428                 d40d->lli_pool.size = lli_len * 2 * align;
 429
 430                 base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
 431                 d40d->lli_pool.base = base;
 432
 433                 if (d40d->lli_pool.base == NULL)
 434                         return -ENOMEM;
 435         }
 436
 437         if (is_log) {
 438                 d40d->lli_log.src = PTR_ALIGN(base, align);
 439                 d40d->lli_log.dst = d40d->lli_log.src + lli_len;
 440
 441                 d40d->lli_pool.dma_addr = 0;
 442         } else {
 443                 d40d->lli_phy.src = PTR_ALIGN(base, align);
 444                 d40d->lli_phy.dst = d40d->lli_phy.src + lli_len;
 445
 446                 d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev,
 447                                                          d40d->lli_phy.src,
 448                                                          d40d->lli_pool.size,
 449                                                          DMA_TO_DEVICE);
 450
 451                 if (dma_mapping_error(d40c->base->dev,
 452                                       d40d->lli_pool.dma_addr)) {
 453                         kfree(d40d->lli_pool.base);
 454                         d40d->lli_pool.base = NULL;
 455                         d40d->lli_pool.dma_addr = 0;
 456                         return -ENOMEM;
 457                 }
 458         }
 459
 460         return 0;
 461 }
 462
 463 static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d)
 464 {
 465         if (d40d->lli_pool.dma_addr)
 466                 dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr,
 467                                  d40d->lli_pool.size, DMA_TO_DEVICE);
 468
 469         kfree(d40d->lli_pool.base);
 470         d40d->lli_pool.base = NULL;
 471         d40d->lli_pool.size = 0;
 472         d40d->lli_log.src = NULL;
 473         d40d->lli_log.dst = NULL;
 474         d40d->lli_phy.src = NULL;
 475         d40d->lli_phy.dst = NULL;
 476 }
 477
 478 static int d40_lcla_alloc_one(struct d40_chan *d40c,
 479                               struct d40_desc *d40d)
 480 {
 481         unsigned long flags;
 482         int i;
 483         int ret = -EINVAL;
 484         int p;
 485
 486         spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
 487
 488         p = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP;
 489
 490         /*
 491          * Allocate both src and dst at the same time, therefore the half
 492          * start on 1 since 0 can't be used since zero is used as end marker.
 493          */
 494         for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
 495                 if (!d40c->base->lcla_pool.alloc_map[p + i]) {
 496                         d40c->base->lcla_pool.alloc_map[p + i] = d40d;
 497                         d40d->lcla_alloc++;
 498                         ret = i;
 499                         break;
 500                 }
 501         }
 502
 503         spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
 504
 505         return ret;
 506 }
 507
 508 static int d40_lcla_free_all(struct d40_chan *d40c,
 509                              struct d40_desc *d40d)
 510 {
 511         unsigned long flags;
 512         int i;
 513         int ret = -EINVAL;
 514
 515         if (chan_is_physical(d40c))
 516                 return 0;
 517
 518         spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
 519
 520         for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
 521                 if (d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num *
 522                                                     D40_LCLA_LINK_PER_EVENT_GRP + i] == d40d) {
 523                         d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num *
 524                                                         D40_LCLA_LINK_PER_EVENT_GRP + i] = NULL;
 525                         d40d->lcla_alloc--;
 526                         if (d40d->lcla_alloc == 0) {
 527                                 ret = 0;
 528                                 break;
 529                         }
 530                 }
 531         }
 532
 533         spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
 534
 535         return ret;
 536
 537 }
 538
 539 static void d40_desc_remove(struct d40_desc *d40d)
 540 {
 541         list_del(&d40d->node);
 542 }
 543
 544 static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
 545 {
 546         struct d40_desc *desc = NULL;
 547
 548         if (!list_empty(&d40c->client)) {
 549                 struct d40_desc *d;
 550                 struct d40_desc *_d;
 551
 552                 list_for_each_entry_safe(d, _d, &d40c->client, node) {
 553                         if (async_tx_test_ack(&d->txd)) {
 554                                 d40_desc_remove(d);
 555                                 desc = d;
 556                                 memset(desc, 0, sizeof(*desc));
 557                                 break;
 558                         }
 559                 }
 560         }
 561
 562         if (!desc)
 563                 desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT);
 564
 565         if (desc)
 566                 INIT_LIST_HEAD(&desc->node);
 567
 568         return desc;
 569 }
 570
 571 static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
 572 {
 573
 574         d40_pool_lli_free(d40c, d40d);
 575         d40_lcla_free_all(d40c, d40d);
 576         kmem_cache_free(d40c->base->desc_slab, d40d);
 577 }
 578
 579 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
 580 {
 581         list_add_tail(&desc->node, &d40c->active);
 582 }
 583
 584 static void d40_phy_lli_load(struct d40_chan *chan, struct d40_desc *desc)
 585 {
 586         struct d40_phy_lli *lli_dst = desc->lli_phy.dst;
 587         struct d40_phy_lli *lli_src = desc->lli_phy.src;
 588         void __iomem *base = chan_base(chan);
 589
 590         writel(lli_src->reg_cfg, base + D40_CHAN_REG_SSCFG);
 591         writel(lli_src->reg_elt, base + D40_CHAN_REG_SSELT);
 592         writel(lli_src->reg_ptr, base + D40_CHAN_REG_SSPTR);
 593         writel(lli_src->reg_lnk, base + D40_CHAN_REG_SSLNK);
 594
 595         writel(lli_dst->reg_cfg, base + D40_CHAN_REG_SDCFG);
 596         writel(lli_dst->reg_elt, base + D40_CHAN_REG_SDELT);
 597         writel(lli_dst->reg_ptr, base + D40_CHAN_REG_SDPTR);
 598         writel(lli_dst->reg_lnk, base + D40_CHAN_REG_SDLNK);
 599 }
 600
 601 static void d40_log_lli_to_lcxa(struct d40_chan *chan, struct d40_desc *desc)
 602 {
 603         struct d40_lcla_pool *pool = &chan->base->lcla_pool;
 604         struct d40_log_lli_bidir *lli = &desc->lli_log;
 605         int lli_current = desc->lli_current;
 606         int lli_len = desc->lli_len;
 607         bool cyclic = desc->cyclic;
 608         int curr_lcla = -EINVAL;
 609         int first_lcla = 0;
 610         bool use_esram_lcla = chan->base->plat_data->use_esram_lcla;
 611         bool linkback;
 612
 613         /*
 614          * We may have partially running cyclic transfers, in case we did't get
 615          * enough LCLA entries.
 616          */
 617         linkback = cyclic && lli_current == 0;
 618
 619         /*
 620          * For linkback, we need one LCLA even with only one link, because we
 621          * can't link back to the one in LCPA space
 622          */
 623         if (linkback || (lli_len - lli_current > 1)) {
 624                 curr_lcla = d40_lcla_alloc_one(chan, desc);
 625                 first_lcla = curr_lcla;
 626         }
 627
 628         /*
 629          * For linkback, we normally load the LCPA in the loop since we need to
 630          * link it to the second LCLA and not the first.  However, if we
 631          * couldn't even get a first LCLA, then we have to run in LCPA and
 632          * reload manually.
 633          */
 634         if (!linkback || curr_lcla == -EINVAL) {
 635                 unsigned int flags = 0;
 636
 637                 if (curr_lcla == -EINVAL)
 638                         flags |= LLI_TERM_INT;
 639
 640                 d40_log_lli_lcpa_write(chan->lcpa,
 641                                        &lli->dst[lli_current],
 642                                        &lli->src[lli_current],
 643                                        curr_lcla,
 644                                        flags);
 645                 lli_current++;
 646         }
 647
 648         if (curr_lcla < 0)
 649                 goto out;
 650
 651         for (; lli_current < lli_len; lli_current++) {
 652                 unsigned int lcla_offset = chan->phy_chan->num * 1024 +
 653                                            8 * curr_lcla * 2;
 654                 struct d40_log_lli *lcla = pool->base + lcla_offset;
 655                 unsigned int flags = 0;
 656                 int next_lcla;
 657
 658                 if (lli_current + 1 < lli_len)
 659                         next_lcla = d40_lcla_alloc_one(chan, desc);
 660                 else
 661                         next_lcla = linkback ? first_lcla : -EINVAL;
 662
 663                 if (cyclic || next_lcla == -EINVAL)
 664                         flags |= LLI_TERM_INT;
 665
 666                 if (linkback && curr_lcla == first_lcla) {
 667                         /* First link goes in both LCPA and LCLA */
 668                         d40_log_lli_lcpa_write(chan->lcpa,
 669                                                &lli->dst[lli_current],
 670                                                &lli->src[lli_current],
 671                                                next_lcla, flags);
 672                 }
 673
 674                 /*
 675                  * One unused LCLA in the cyclic case if the very first
 676                  * next_lcla fails...
 677                  */
 678                 d40_log_lli_lcla_write(lcla,
 679                                        &lli->dst[lli_current],
 680                                        &lli->src[lli_current],
 681                                        next_lcla, flags);
 682
 683                 /*
 684                  * Cache maintenance is not needed if lcla is
 685                  * mapped in esram
 686                  */
 687                 if (!use_esram_lcla) {
 688                         dma_sync_single_range_for_device(chan->base->dev,
 689                                                 pool->dma_addr, lcla_offset,
 690                                                 2 * sizeof(struct d40_log_lli),
 691                                                 DMA_TO_DEVICE);
 692                 }
 693                 curr_lcla = next_lcla;
 694
 695                 if (curr_lcla == -EINVAL || curr_lcla == first_lcla) {
 696                         lli_current++;
 697                         break;
 698                 }
 699         }
 700
 701 out:
 702         desc->lli_current = lli_current;
 703 }
 704
 705 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
 706 {
 707         if (chan_is_physical(d40c)) {
 708                 d40_phy_lli_load(d40c, d40d);
 709                 d40d->lli_current = d40d->lli_len;
 710         } else
 711                 d40_log_lli_to_lcxa(d40c, d40d);
 712 }
 713
 714 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
 715 {
 716         struct d40_desc *d;
 717
 718         if (list_empty(&d40c->active))
 719                 return NULL;
 720
 721         d = list_first_entry(&d40c->active,
 722                              struct d40_desc,
 723                              node);
 724         return d;
 725 }
 726
 727 /* remove desc from current queue and add it to the pending_queue */
 728 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
 729 {
 730         d40_desc_remove(desc);
 731         desc->is_in_client_list = false;
 732         list_add_tail(&desc->node, &d40c->pending_queue);
 733 }
 734
 735 static struct d40_desc *d40_first_pending(struct d40_chan *d40c)
 736 {
 737         struct d40_desc *d;
 738
 739         if (list_empty(&d40c->pending_queue))
 740                 return NULL;
 741
 742         d = list_first_entry(&d40c->pending_queue,
 743                              struct d40_desc,
 744                              node);
 745         return d;
 746 }
 747
 748 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
 749 {
 750         struct d40_desc *d;
 751
 752         if (list_empty(&d40c->queue))
 753                 return NULL;
 754
 755         d = list_first_entry(&d40c->queue,
 756                              struct d40_desc,
 757                              node);
 758         return d;
 759 }
 760
 761 static int d40_psize_2_burst_size(bool is_log, int psize)
 762 {
 763         if (is_log) {
 764                 if (psize == STEDMA40_PSIZE_LOG_1)
 765                         return 1;
 766         } else {
 767                 if (psize == STEDMA40_PSIZE_PHY_1)
 768                         return 1;
 769         }
 770
 771         return 2 << psize;
 772 }
 773
 774 /*
 775  * The dma only supports transmitting packages up to
 776  * STEDMA40_MAX_SEG_SIZE << data_width. Calculate the total number of
 777  * dma elements required to send the entire sg list
 778  */
 779 static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
 780 {
 781         int dmalen;
 782         u32 max_w = max(data_width1, data_width2);
 783         u32 min_w = min(data_width1, data_width2);
 784         u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE << min_w, 1 << max_w);
 785
 786         if (seg_max > STEDMA40_MAX_SEG_SIZE)
 787                 seg_max -= (1 << max_w);
 788
 789         if (!IS_ALIGNED(size, 1 << max_w))
 790                 return -EINVAL;
 791
 792         if (size <= seg_max)
 793                 dmalen = 1;
 794         else {
 795                 dmalen = size / seg_max;
 796                 if (dmalen * seg_max < size)
 797                         dmalen++;
 798         }
 799         return dmalen;
 800 }
 801
 802 static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
 803                            u32 data_width1, u32 data_width2)
 804 {
 805         struct scatterlist *sg;
 806         int i;
 807         int len = 0;
 808         int ret;
 809
 810         for_each_sg(sgl, sg, sg_len, i) {
 811                 ret = d40_size_2_dmalen(sg_dma_len(sg),
 812                                         data_width1, data_width2);
 813                 if (ret < 0)
 814                         return ret;
 815                 len += ret;
 816         }
 817         return len;
 818 }
 819
 820
 821 #ifdef CONFIG_PM
 822 static void dma40_backup(void __iomem *baseaddr, u32 *backup,
 823                          u32 *regaddr, int num, bool save)
 824 {
 825         int i;
 826
 827         for (i = 0; i < num; i++) {
 828                 void __iomem *addr = baseaddr + regaddr[i];
 829
 830                 if (save)
 831                         backup[i] = readl_relaxed(addr);
 832                 else
 833                         writel_relaxed(backup[i], addr);
 834         }
 835 }
 836
 837 static void d40_save_restore_registers(struct d40_base *base, bool save)
 838 {
 839         int i;
 840
 841         /* Save/Restore channel specific registers */
 842         for (i = 0; i < base->num_phy_chans; i++) {
 843                 void __iomem *addr;
 844                 int idx;
 845
 846                 if (base->phy_res[i].reserved)
 847                         continue;
 848
 849                 addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA;
 850                 idx = i * ARRAY_SIZE(d40_backup_regs_chan);
 851
 852                 dma40_backup(addr, &base->reg_val_backup_chan[idx],
 853                              d40_backup_regs_chan,
 854                              ARRAY_SIZE(d40_backup_regs_chan),
 855                              save);
 856         }
 857
 858         /* Save/Restore global registers */
 859         dma40_backup(base->virtbase, base->reg_val_backup,
 860                      d40_backup_regs, ARRAY_SIZE(d40_backup_regs),
 861                      save);
 862
 863         /* Save/Restore registers only existing on dma40 v3 and later */
 864         if (base->rev >= 3)
 865                 dma40_backup(base->virtbase, base->reg_val_backup_v3,
 866                              d40_backup_regs_v3,
 867                              ARRAY_SIZE(d40_backup_regs_v3),
 868                              save);
 869 }
 870 #else
 871 static void d40_save_restore_registers(struct d40_base *base, bool save)
 872 {
 873 }
 874 #endif
 875
 876 static int d40_channel_execute_command(struct d40_chan *d40c,
 877                                        enum d40_command command)
 878 {
 879         u32 status;
 880         int i;
 881         void __iomem *active_reg;
 882         int ret = 0;
 883         unsigned long flags;
 884         u32 wmask;
 885
 886         spin_lock_irqsave(&d40c->base->execmd_lock, flags);
 887
 888         if (d40c->phy_chan->num % 2 == 0)
 889                 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
 890         else
 891                 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
 892
 893         if (command == D40_DMA_SUSPEND_REQ) {
 894                 status = (readl(active_reg) &
 895                           D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
 896                         D40_CHAN_POS(d40c->phy_chan->num);
 897
 898                 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
 899                         goto done;
 900         }
 901
 902         wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
 903         writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
 904                active_reg);
 905
 906         if (command == D40_DMA_SUSPEND_REQ) {
 907
 908                 for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
 909                         status = (readl(active_reg) &
 910                                   D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
 911                                 D40_CHAN_POS(d40c->phy_chan->num);
 912
 913                         cpu_relax();
 914                         /*
 915                          * Reduce the number of bus accesses while
 916                          * waiting for the DMA to suspend.
 917                          */
 918                         udelay(3);
 919
 920                         if (status == D40_DMA_STOP ||
 921                             status == D40_DMA_SUSPENDED)
 922                                 break;
 923                 }
 924
 925                 if (i == D40_SUSPEND_MAX_IT) {
 926                         chan_err(d40c,
 927                                 "unable to suspend the chl %d (log: %d) status %x\n",
 928                                 d40c->phy_chan->num, d40c->log_num,
 929                                 status);
 930                         dump_stack();
 931                         ret = -EBUSY;
 932                 }
 933
 934         }
 935 done:
 936         spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
 937         return ret;
 938 }
 939
 940 static void d40_term_all(struct d40_chan *d40c)
 941 {
 942         struct d40_desc *d40d;
 943         struct d40_desc *_d;
 944
 945         /* Release active descriptors */
 946         while ((d40d = d40_first_active_get(d40c))) {
 947                 d40_desc_remove(d40d);
 948                 d40_desc_free(d40c, d40d);
 949         }
 950
 951         /* Release queued descriptors waiting for transfer */
 952         while ((d40d = d40_first_queued(d40c))) {
 953                 d40_desc_remove(d40d);
 954                 d40_desc_free(d40c, d40d);
 955         }
 956
 957         /* Release pending descriptors */
 958         while ((d40d = d40_first_pending(d40c))) {
 959                 d40_desc_remove(d40d);
 960                 d40_desc_free(d40c, d40d);
 961         }
 962
 963         /* Release client owned descriptors */
 964         if (!list_empty(&d40c->client))
 965                 list_for_each_entry_safe(d40d, _d, &d40c->client, node) {
 966                         d40_desc_remove(d40d);
 967                         d40_desc_free(d40c, d40d);
 968                 }
 969
 970         /* Release descriptors in prepare queue */
 971         if (!list_empty(&d40c->prepare_queue))
 972                 list_for_each_entry_safe(d40d, _d,
 973                                          &d40c->prepare_queue, node) {
 974                         d40_desc_remove(d40d);
 975                         d40_desc_free(d40c, d40d);
 976                 }
 977
 978         d40c->pending_tx = 0;
 979         d40c->busy = false;
 980 }
 981
 982 static void __d40_config_set_event(struct d40_chan *d40c, bool enable,
 983                                    u32 event, int reg)
 984 {
 985         void __iomem *addr = chan_base(d40c) + reg;
 986         int tries;
 987
 988         if (!enable) {
 989                 writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
 990                        | ~D40_EVENTLINE_MASK(event), addr);
 991                 return;
 992         }
 993
 994         /*
 995          * The hardware sometimes doesn't register the enable when src and dst
 996          * event lines are active on the same logical channel.  Retry to ensure
 997          * it does.  Usually only one retry is sufficient.
 998          */
 999         tries = 100;
1000         while (--tries) {
1001                 writel((D40_ACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
1002                        | ~D40_EVENTLINE_MASK(event), addr);
1003
1004                 if (readl(addr) & D40_EVENTLINE_MASK(event))
1005                         break;
1006         }
1007
1008         if (tries != 99)
1009                 dev_dbg(chan2dev(d40c),
1010                         "[%s] workaround enable S%cLNK (%d tries)\n",
1011                         __func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
1012                         100 - tries);
1013
1014         WARN_ON(!tries);
1015 }
1016
1017 static void d40_config_set_event(struct d40_chan *d40c, bool do_enable)
1018 {
1019         unsigned long flags;
1020
1021         spin_lock_irqsave(&d40c->phy_chan->lock, flags);
1022
1023         /* Enable event line connected to device (or memcpy) */
1024         if ((d40c->dma_cfg.dir ==  STEDMA40_PERIPH_TO_MEM) ||
1025             (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH)) {
1026                 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1027
1028                 __d40_config_set_event(d40c, do_enable, event,
1029                                        D40_CHAN_REG_SSLNK);
1030         }
1031
1032         if (d40c->dma_cfg.dir !=  STEDMA40_PERIPH_TO_MEM) {
1033                 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1034
1035                 __d40_config_set_event(d40c, do_enable, event,
1036                                        D40_CHAN_REG_SDLNK);
1037         }
1038
1039         spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
1040 }
1041
1042 static u32 d40_chan_has_events(struct d40_chan *d40c)
1043 {
1044         void __iomem *chanbase = chan_base(d40c);
1045         u32 val;
1046
1047         val = readl(chanbase + D40_CHAN_REG_SSLNK);
1048         val |= readl(chanbase + D40_CHAN_REG_SDLNK);
1049
1050         return val;
1051 }
1052
1053 static u32 d40_get_prmo(struct d40_chan *d40c)
1054 {
1055         static const unsigned int phy_map[] = {
1056                 [STEDMA40_PCHAN_BASIC_MODE]
1057                         = D40_DREG_PRMO_PCHAN_BASIC,
1058                 [STEDMA40_PCHAN_MODULO_MODE]
1059                         = D40_DREG_PRMO_PCHAN_MODULO,
1060                 [STEDMA40_PCHAN_DOUBLE_DST_MODE]
1061                         = D40_DREG_PRMO_PCHAN_DOUBLE_DST,
1062         };
1063         static const unsigned int log_map[] = {
1064                 [STEDMA40_LCHAN_SRC_PHY_DST_LOG]
1065                         = D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG,
1066                 [STEDMA40_LCHAN_SRC_LOG_DST_PHY]
1067                         = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY,
1068                 [STEDMA40_LCHAN_SRC_LOG_DST_LOG]
1069                         = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG,
1070         };
1071
1072         if (chan_is_physical(d40c))
1073                 return phy_map[d40c->dma_cfg.mode_opt];
1074         else
1075                 return log_map[d40c->dma_cfg.mode_opt];
1076 }
1077
1078 static void d40_config_write(struct d40_chan *d40c)
1079 {
1080         u32 addr_base;
1081         u32 var;
1082
1083         /* Odd addresses are even addresses + 4 */
1084         addr_base = (d40c->phy_chan->num % 2) * 4;
1085         /* Setup channel mode to logical or physical */
1086         var = ((u32)(chan_is_logical(d40c)) + 1) <<
1087                 D40_CHAN_POS(d40c->phy_chan->num);
1088         writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
1089
1090         /* Setup operational mode option register */
1091         var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num);
1092
1093         writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
1094
1095         if (chan_is_logical(d40c)) {
1096                 int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS)
1097                            & D40_SREG_ELEM_LOG_LIDX_MASK;
1098                 void __iomem *chanbase = chan_base(d40c);
1099
1100                 /* Set default config for CFG reg */
1101                 writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG);
1102                 writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG);
1103
1104                 /* Set LIDX for lcla */
1105                 writel(lidx, chanbase + D40_CHAN_REG_SSELT);
1106                 writel(lidx, chanbase + D40_CHAN_REG_SDELT);
1107
1108                 /* Clear LNK which will be used by d40_chan_has_events() */
1109                 writel(0, chanbase + D40_CHAN_REG_SSLNK);
1110                 writel(0, chanbase + D40_CHAN_REG_SDLNK);
1111         }
1112 }
1113
1114 static u32 d40_residue(struct d40_chan *d40c)
1115 {
1116         u32 num_elt;
1117
1118         if (chan_is_logical(d40c))
1119                 num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
1120                         >> D40_MEM_LCSP2_ECNT_POS;
1121         else {
1122                 u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT);
1123                 num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK)
1124                           >> D40_SREG_ELEM_PHY_ECNT_POS;
1125         }
1126
1127         return num_elt * (1 << d40c->dma_cfg.dst_info.data_width);
1128 }
1129
1130 static bool d40_tx_is_linked(struct d40_chan *d40c)
1131 {
1132         bool is_link;
1133
1134         if (chan_is_logical(d40c))
1135                 is_link = readl(&d40c->lcpa->lcsp3) &  D40_MEM_LCSP3_DLOS_MASK;
1136         else
1137                 is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK)
1138                           & D40_SREG_LNK_PHYS_LNK_MASK;
1139
1140         return is_link;
1141 }
1142
1143 static int d40_pause(struct d40_chan *d40c)
1144 {
1145         int res = 0;
1146         unsigned long flags;
1147
1148         if (!d40c->busy)
1149                 return 0;
1150
1151         pm_runtime_get_sync(d40c->base->dev);
1152         spin_lock_irqsave(&d40c->lock, flags);
1153
1154         res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1155         if (res == 0) {
1156                 if (chan_is_logical(d40c)) {
1157                         d40_config_set_event(d40c, false);
1158                         /* Resume the other logical channels if any */
1159                         if (d40_chan_has_events(d40c))
1160                                 res = d40_channel_execute_command(d40c,
1161                                                                   D40_DMA_RUN);
1162                 }
1163         }
1164         pm_runtime_mark_last_busy(d40c->base->dev);
1165         pm_runtime_put_autosuspend(d40c->base->dev);
1166         spin_unlock_irqrestore(&d40c->lock, flags);
1167         return res;
1168 }
1169
1170 static int d40_resume(struct d40_chan *d40c)
1171 {
1172         int res = 0;
1173         unsigned long flags;
1174
1175         if (!d40c->busy)
1176                 return 0;
1177
1178         spin_lock_irqsave(&d40c->lock, flags);
1179         pm_runtime_get_sync(d40c->base->dev);
1180         if (d40c->base->rev == 0)
1181                 if (chan_is_logical(d40c)) {
1182                         res = d40_channel_execute_command(d40c,
1183                                                           D40_DMA_SUSPEND_REQ);
1184                         goto no_suspend;
1185                 }
1186
1187         /* If bytes left to transfer or linked tx resume job */
1188         if (d40_residue(d40c) || d40_tx_is_linked(d40c)) {
1189
1190                 if (chan_is_logical(d40c))
1191                         d40_config_set_event(d40c, true);
1192
1193                 res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1194         }
1195
1196 no_suspend:
1197         pm_runtime_mark_last_busy(d40c->base->dev);
1198         pm_runtime_put_autosuspend(d40c->base->dev);
1199         spin_unlock_irqrestore(&d40c->lock, flags);
1200         return res;
1201 }
1202
1203 static int d40_terminate_all(struct d40_chan *chan)
1204 {
1205         unsigned long flags;
1206         int ret = 0;
1207
1208         ret = d40_pause(chan);
1209         if (!ret && chan_is_physical(chan))
1210                 ret = d40_channel_execute_command(chan, D40_DMA_STOP);
1211
1212         spin_lock_irqsave(&chan->lock, flags);
1213         d40_term_all(chan);
1214         spin_unlock_irqrestore(&chan->lock, flags);
1215
1216         return ret;
1217 }
1218
1219 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1220 {
1221         struct d40_chan *d40c = container_of(tx->chan,
1222                                              struct d40_chan,
1223                                              chan);
1224         struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1225         unsigned long flags;
1226
1227         spin_lock_irqsave(&d40c->lock, flags);
1228
1229         d40c->chan.cookie++;
1230
1231         if (d40c->chan.cookie < 0)
1232                 d40c->chan.cookie = 1;
1233
1234         d40d->txd.cookie = d40c->chan.cookie;
1235
1236         d40_desc_queue(d40c, d40d);
1237
1238         spin_unlock_irqrestore(&d40c->lock, flags);
1239
1240         return tx->cookie;
1241 }
1242
1243 static int d40_start(struct d40_chan *d40c)
1244 {
1245         if (d40c->base->rev == 0) {
1246                 int err;
1247
1248                 if (chan_is_logical(d40c)) {
1249                         err = d40_channel_execute_command(d40c,
1250                                                           D40_DMA_SUSPEND_REQ);
1251                         if (err)
1252                                 return err;
1253                 }
1254         }
1255
1256         if (chan_is_logical(d40c))
1257                 d40_config_set_event(d40c, true);
1258
1259         return d40_channel_execute_command(d40c, D40_DMA_RUN);
1260 }
1261
1262 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1263 {
1264         struct d40_desc *d40d;
1265         int err;
1266
1267         /* Start queued jobs, if any */
1268         d40d = d40_first_queued(d40c);
1269
1270         if (d40d != NULL) {
1271                 if (!d40c->busy)
1272                         d40c->busy = true;
1273
1274                 pm_runtime_get_sync(d40c->base->dev);
1275
1276                 /* Remove from queue */
1277                 d40_desc_remove(d40d);
1278
1279                 /* Add to active queue */
1280                 d40_desc_submit(d40c, d40d);
1281
1282                 /* Initiate DMA job */
1283                 d40_desc_load(d40c, d40d);
1284
1285                 /* Start dma job */
1286                 err = d40_start(d40c);
1287
1288                 if (err)
1289                         return NULL;
1290         }
1291
1292         return d40d;
1293 }
1294
1295 /* called from interrupt context */
1296 static void dma_tc_handle(struct d40_chan *d40c)
1297 {
1298         struct d40_desc *d40d;
1299
1300         /* Get first active entry from list */
1301         d40d = d40_first_active_get(d40c);
1302
1303         if (d40d == NULL)
1304                 return;
1305
1306         if (d40d->cyclic) {
1307                 /*
1308                  * If this was a paritially loaded list, we need to reloaded
1309                  * it, and only when the list is completed.  We need to check
1310                  * for done because the interrupt will hit for every link, and
1311                  * not just the last one.
1312                  */
1313                 if (d40d->lli_current < d40d->lli_len
1314                     && !d40_tx_is_linked(d40c)
1315                     && !d40_residue(d40c)) {
1316                         d40_lcla_free_all(d40c, d40d);
1317                         d40_desc_load(d40c, d40d);
1318                         (void) d40_start(d40c);
1319
1320                         if (d40d->lli_current == d40d->lli_len)
1321                                 d40d->lli_current = 0;
1322                 }
1323         } else {
1324                 d40_lcla_free_all(d40c, d40d);
1325
1326                 if (d40d->lli_current < d40d->lli_len) {
1327                         d40_desc_load(d40c, d40d);
1328                         /* Start dma job */
1329                         (void) d40_start(d40c);
1330                         return;
1331                 }
1332
1333                 if (d40_queue_start(d40c) == NULL)
1334                         d40c->busy = false;
1335                 pm_runtime_mark_last_busy(d40c->base->dev);
1336                 pm_runtime_put_autosuspend(d40c->base->dev);
1337         }
1338
1339         d40c->pending_tx++;
1340         tasklet_schedule(&d40c->tasklet);
1341
1342 }
1343
1344 static void dma_tasklet(unsigned long data)
1345 {
1346         struct d40_chan *d40c = (struct d40_chan *) data;
1347         struct d40_desc *d40d;
1348         unsigned long flags;
1349         dma_async_tx_callback callback;
1350         void *callback_param;
1351
1352         spin_lock_irqsave(&d40c->lock, flags);
1353
1354         /* Get first active entry from list */
1355         d40d = d40_first_active_get(d40c);
1356         if (d40d == NULL)
1357                 goto err;
1358
1359         if (!d40d->cyclic)
1360                 d40c->completed = d40d->txd.cookie;
1361
1362         /*
1363          * If terminating a channel pending_tx is set to zero.
1364          * This prevents any finished active jobs to return to the client.
1365          */
1366         if (d40c->pending_tx == 0) {
1367                 spin_unlock_irqrestore(&d40c->lock, flags);
1368                 return;
1369         }
1370
1371         /* Callback to client */
1372         callback = d40d->txd.callback;
1373         callback_param = d40d->txd.callback_param;
1374
1375         if (!d40d->cyclic) {
1376                 if (async_tx_test_ack(&d40d->txd)) {
1377                         d40_desc_remove(d40d);
1378                         d40_desc_free(d40c, d40d);
1379                 } else {
1380                         if (!d40d->is_in_client_list) {
1381                                 d40_desc_remove(d40d);
1382                                 d40_lcla_free_all(d40c, d40d);
1383                                 list_add_tail(&d40d->node, &d40c->client);
1384                                 d40d->is_in_client_list = true;
1385                         }
1386                 }
1387         }
1388
1389         d40c->pending_tx--;
1390
1391         if (d40c->pending_tx)
1392                 tasklet_schedule(&d40c->tasklet);
1393
1394         spin_unlock_irqrestore(&d40c->lock, flags);
1395
1396         if (callback && (d40d->txd.flags & DMA_PREP_INTERRUPT))
1397                 callback(callback_param);
1398
1399         return;
1400
1401  err:
1402         /* Rescue manoeuvre if receiving double interrupts */
1403         if (d40c->pending_tx > 0)
1404                 d40c->pending_tx--;
1405         spin_unlock_irqrestore(&d40c->lock, flags);
1406 }
1407
1408 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1409 {
1410         static const struct d40_interrupt_lookup il[] = {
1411                 {D40_DREG_LCTIS0, D40_DREG_LCICR0, false,  0},
1412                 {D40_DREG_LCTIS1, D40_DREG_LCICR1, false, 32},
1413                 {D40_DREG_LCTIS2, D40_DREG_LCICR2, false, 64},
1414                 {D40_DREG_LCTIS3, D40_DREG_LCICR3, false, 96},
1415                 {D40_DREG_LCEIS0, D40_DREG_LCICR0, true,   0},
1416                 {D40_DREG_LCEIS1, D40_DREG_LCICR1, true,  32},
1417                 {D40_DREG_LCEIS2, D40_DREG_LCICR2, true,  64},
1418                 {D40_DREG_LCEIS3, D40_DREG_LCICR3, true,  96},
1419                 {D40_DREG_PCTIS,  D40_DREG_PCICR,  false, D40_PHY_CHAN},
1420                 {D40_DREG_PCEIS,  D40_DREG_PCICR,  true,  D40_PHY_CHAN},
1421         };
1422
1423         int i;
1424         u32 regs[ARRAY_SIZE(il)];
1425         u32 idx;
1426         u32 row;
1427         long chan = -1;
1428         struct d40_chan *d40c;
1429         unsigned long flags;
1430         struct d40_base *base = data;
1431
1432         spin_lock_irqsave(&base->interrupt_lock, flags);
1433
1434         /* Read interrupt status of both logical and physical channels */
1435         for (i = 0; i < ARRAY_SIZE(il); i++)
1436                 regs[i] = readl(base->virtbase + il[i].src);
1437
1438         for (;;) {
1439
1440                 chan = find_next_bit((unsigned long *)regs,
1441                                      BITS_PER_LONG * ARRAY_SIZE(il), chan + 1);
1442
1443                 /* No more set bits found? */
1444                 if (chan == BITS_PER_LONG * ARRAY_SIZE(il))
1445                         break;
1446
1447                 row = chan / BITS_PER_LONG;
1448                 idx = chan & (BITS_PER_LONG - 1);
1449
1450                 /* ACK interrupt */
1451                 writel(1 << idx, base->virtbase + il[row].clr);
1452
1453                 if (il[row].offset == D40_PHY_CHAN)
1454                         d40c = base->lookup_phy_chans[idx];
1455                 else
1456                         d40c = base->lookup_log_chans[il[row].offset + idx];
1457                 spin_lock(&d40c->lock);
1458
1459                 if (!il[row].is_error)
1460                         dma_tc_handle(d40c);
1461                 else
1462                         d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1463                                 chan, il[row].offset, idx);
1464
1465                 spin_unlock(&d40c->lock);
1466         }
1467
1468         spin_unlock_irqrestore(&base->interrupt_lock, flags);
1469
1470         return IRQ_HANDLED;
1471 }
1472
1473 static int d40_validate_conf(struct d40_chan *d40c,
1474                              struct stedma40_chan_cfg *conf)
1475 {
1476         int res = 0;
1477         u32 dst_event_group = D40_TYPE_TO_GROUP(conf->dst_dev_type);
1478         u32 src_event_group = D40_TYPE_TO_GROUP(conf->src_dev_type);
1479         bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1480
1481         if (!conf->dir) {
1482                 chan_err(d40c, "Invalid direction.\n");
1483                 res = -EINVAL;
1484         }
1485
1486         if (conf->dst_dev_type != STEDMA40_DEV_DST_MEMORY &&
1487             d40c->base->plat_data->dev_tx[conf->dst_dev_type] == 0 &&
1488             d40c->runtime_addr == 0) {
1489
1490                 chan_err(d40c, "Invalid TX channel address (%d)\n",
1491                          conf->dst_dev_type);
1492                 res = -EINVAL;
1493         }
1494
1495         if (conf->src_dev_type != STEDMA40_DEV_SRC_MEMORY &&
1496             d40c->base->plat_data->dev_rx[conf->src_dev_type] == 0 &&
1497             d40c->runtime_addr == 0) {
1498                 chan_err(d40c, "Invalid RX channel address (%d)\n",
1499                         conf->src_dev_type);
1500                 res = -EINVAL;
1501         }
1502
1503         if (conf->dir == STEDMA40_MEM_TO_PERIPH &&
1504             dst_event_group == STEDMA40_DEV_DST_MEMORY) {
1505                 chan_err(d40c, "Invalid dst\n");
1506                 res = -EINVAL;
1507         }
1508
1509         if (conf->dir == STEDMA40_PERIPH_TO_MEM &&
1510             src_event_group == STEDMA40_DEV_SRC_MEMORY) {
1511                 chan_err(d40c, "Invalid src\n");
1512                 res = -EINVAL;
1513         }
1514
1515         if (src_event_group == STEDMA40_DEV_SRC_MEMORY &&
1516             dst_event_group == STEDMA40_DEV_DST_MEMORY && is_log) {
1517                 chan_err(d40c, "No event line\n");
1518                 res = -EINVAL;
1519         }
1520
1521         if (conf->dir == STEDMA40_PERIPH_TO_PERIPH &&
1522             (src_event_group != dst_event_group)) {
1523                 chan_err(d40c, "Invalid event group\n");
1524                 res = -EINVAL;
1525         }
1526
1527         if (conf->dir == STEDMA40_PERIPH_TO_PERIPH) {
1528                 /*
1529                  * DMAC HW supports it. Will be added to this driver,
1530                  * in case any dma client requires it.
1531                  */
1532                 chan_err(d40c, "periph to periph not supported\n");
1533                 res = -EINVAL;
1534         }
1535
1536         if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1537             (1 << conf->src_info.data_width) !=
1538             d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1539             (1 << conf->dst_info.data_width)) {
1540                 /*
1541                  * The DMAC hardware only supports
1542                  * src (burst x width) == dst (burst x width)
1543                  */
1544
1545                 chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1546                 res = -EINVAL;
1547         }
1548
1549         return res;
1550 }
1551
1552 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1553                                bool is_src, int log_event_line, bool is_log,
1554                                bool *first_user)
1555 {
1556         unsigned long flags;
1557         spin_lock_irqsave(&phy->lock, flags);
1558
1559         *first_user = ((phy->allocated_src | phy->allocated_dst)
1560                         == D40_ALLOC_FREE);
1561
1562         if (!is_log) {
1563                 /* Physical interrupts are masked per physical full channel */
1564                 if (phy->allocated_src == D40_ALLOC_FREE &&
1565                     phy->allocated_dst == D40_ALLOC_FREE) {
1566                         phy->allocated_dst = D40_ALLOC_PHY;
1567                         phy->allocated_src = D40_ALLOC_PHY;
1568                         goto found;
1569                 } else
1570                         goto not_found;
1571         }
1572
1573         /* Logical channel */
1574         if (is_src) {
1575                 if (phy->allocated_src == D40_ALLOC_PHY)
1576                         goto not_found;
1577
1578                 if (phy->allocated_src == D40_ALLOC_FREE)
1579                         phy->allocated_src = D40_ALLOC_LOG_FREE;
1580
1581                 if (!(phy->allocated_src & (1 << log_event_line))) {
1582                         phy->allocated_src |= 1 << log_event_line;
1583                         goto found;
1584                 } else
1585                         goto not_found;
1586         } else {
1587                 if (phy->allocated_dst == D40_ALLOC_PHY)
1588                         goto not_found;
1589
1590                 if (phy->allocated_dst == D40_ALLOC_FREE)
1591                         phy->allocated_dst = D40_ALLOC_LOG_FREE;
1592
1593                 if (!(phy->allocated_dst & (1 << log_event_line))) {
1594                         phy->allocated_dst |= 1 << log_event_line;
1595                         goto found;
1596                 } else
1597                         goto not_found;
1598         }
1599
1600 not_found:
1601         spin_unlock_irqrestore(&phy->lock, flags);
1602         return false;
1603 found:
1604         spin_unlock_irqrestore(&phy->lock, flags);
1605         return true;
1606 }
1607
1608 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1609                                int log_event_line)
1610 {
1611         unsigned long flags;
1612         bool is_free = false;
1613
1614         spin_lock_irqsave(&phy->lock, flags);
1615         if (!log_event_line) {
1616                 phy->allocated_dst = D40_ALLOC_FREE;
1617                 phy->allocated_src = D40_ALLOC_FREE;
1618                 is_free = true;
1619                 goto out;
1620         }
1621
1622         /* Logical channel */
1623         if (is_src) {
1624                 phy->allocated_src &= ~(1 << log_event_line);
1625                 if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1626                         phy->allocated_src = D40_ALLOC_FREE;
1627         } else {
1628                 phy->allocated_dst &= ~(1 << log_event_line);
1629                 if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1630                         phy->allocated_dst = D40_ALLOC_FREE;
1631         }
1632
1633         is_free = ((phy->allocated_src | phy->allocated_dst) ==
1634                    D40_ALLOC_FREE);
1635
1636 out:
1637         spin_unlock_irqrestore(&phy->lock, flags);
1638
1639         return is_free;
1640 }
1641
1642 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1643 {
1644         int dev_type;
1645         int event_group;
1646         int event_line;
1647         struct d40_phy_res *phys;
1648         int i;
1649         int j;
1650         int log_num;
1651         bool is_src;
1652         bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1653
1654         phys = d40c->base->phy_res;
1655
1656         if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1657                 dev_type = d40c->dma_cfg.src_dev_type;
1658                 log_num = 2 * dev_type;
1659                 is_src = true;
1660         } else if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1661                    d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1662                 /* dst event lines are used for logical memcpy */
1663                 dev_type = d40c->dma_cfg.dst_dev_type;
1664                 log_num = 2 * dev_type + 1;
1665                 is_src = false;
1666         } else
1667                 return -EINVAL;
1668
1669         event_group = D40_TYPE_TO_GROUP(dev_type);
1670         event_line = D40_TYPE_TO_EVENT(dev_type);
1671
1672         if (!is_log) {
1673                 if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1674                         /* Find physical half channel */
1675                         for (i = 0; i < d40c->base->num_phy_chans; i++) {
1676
1677                                 if (d40_alloc_mask_set(&phys[i], is_src,
1678                                                        0, is_log,
1679                                                        first_phy_user))
1680                                         goto found_phy;
1681                         }
1682                 } else
1683                         for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1684                                 int phy_num = j  + event_group * 2;
1685                                 for (i = phy_num; i < phy_num + 2; i++) {
1686                                         if (d40_alloc_mask_set(&phys[i],
1687                                                                is_src,
1688                                                                0,
1689                                                                is_log,
1690                                                                first_phy_user))
1691                                                 goto found_phy;
1692                                 }
1693                         }
1694                 return -EINVAL;
1695 found_phy:
1696                 d40c->phy_chan = &phys[i];
1697                 d40c->log_num = D40_PHY_CHAN;
1698                 goto out;
1699         }
1700         if (dev_type == -1)
1701                 return -EINVAL;
1702
1703         /* Find logical channel */
1704         for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1705                 int phy_num = j + event_group * 2;
1706
1707                 if (d40c->dma_cfg.use_fixed_channel) {
1708                         i = d40c->dma_cfg.phy_channel;
1709
1710                         if ((i != phy_num) && (i != phy_num + 1)) {
1711                                 dev_err(chan2dev(d40c),
1712                                         "invalid fixed phy channel %d\n", i);
1713                                 return -EINVAL;
1714                         }
1715
1716                         if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1717                                                is_log, first_phy_user))
1718                                 goto found_log;
1719
1720                         dev_err(chan2dev(d40c),
1721                                 "could not allocate fixed phy channel %d\n", i);
1722                         return -EINVAL;
1723                 }
1724
1725                 /*
1726                  * Spread logical channels across all available physical rather
1727                  * than pack every logical channel at the first available phy
1728                  * channels.
1729                  */
1730                 if (is_src) {
1731                         for (i = phy_num; i < phy_num + 2; i++) {
1732                                 if (d40_alloc_mask_set(&phys[i], is_src,
1733                                                        event_line, is_log,
1734                                                        first_phy_user))
1735                                         goto found_log;
1736                         }
1737                 } else {
1738                         for (i = phy_num + 1; i >= phy_num; i--) {
1739                                 if (d40_alloc_mask_set(&phys[i], is_src,
1740                                                        event_line, is_log,
1741                                                        first_phy_user))
1742                                         goto found_log;
1743                         }
1744                 }
1745         }
1746         return -EINVAL;
1747
1748 found_log:
1749         d40c->phy_chan = &phys[i];
1750         d40c->log_num = log_num;
1751 out:
1752
1753         if (is_log)
1754                 d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1755         else
1756                 d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1757
1758         return 0;
1759
1760 }
1761
1762 static int d40_config_memcpy(struct d40_chan *d40c)
1763 {
1764         dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1765
1766         if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1767                 d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_log;
1768                 d40c->dma_cfg.src_dev_type = STEDMA40_DEV_SRC_MEMORY;
1769                 d40c->dma_cfg.dst_dev_type = d40c->base->plat_data->
1770                         memcpy[d40c->chan.chan_id];
1771
1772         } else if (dma_has_cap(DMA_MEMCPY, cap) &&
1773                    dma_has_cap(DMA_SLAVE, cap)) {
1774                 d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_phy;
1775         } else {
1776                 chan_err(d40c, "No memcpy\n");
1777                 return -EINVAL;
1778         }
1779
1780         return 0;
1781 }
1782
1783
1784 static int d40_free_dma(struct d40_chan *d40c)
1785 {
1786
1787         int res = 0;
1788         u32 event;
1789         struct d40_phy_res *phy = d40c->phy_chan;
1790         bool is_src;
1791
1792         /* Terminate all queued and active transfers */
1793         d40_term_all(d40c);
1794
1795         if (phy == NULL) {
1796                 chan_err(d40c, "phy == null\n");
1797                 return -EINVAL;
1798         }
1799
1800         if (phy->allocated_src == D40_ALLOC_FREE &&
1801             phy->allocated_dst == D40_ALLOC_FREE) {
1802                 chan_err(d40c, "channel already free\n");
1803                 return -EINVAL;
1804         }
1805
1806         if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1807             d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1808                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1809                 is_src = false;
1810         } else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1811                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1812                 is_src = true;
1813         } else {
1814                 chan_err(d40c, "Unknown direction\n");
1815                 return -EINVAL;
1816         }
1817
1818         pm_runtime_get_sync(d40c->base->dev);
1819         res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1820         if (res) {
1821                 chan_err(d40c, "suspend failed\n");
1822                 goto out;
1823         }
1824
1825         if (chan_is_logical(d40c)) {
1826                 /* Release logical channel, deactivate the event line */
1827
1828                 d40_config_set_event(d40c, false);
1829                 d40c->base->lookup_log_chans[d40c->log_num] = NULL;
1830
1831                 /*
1832                  * Check if there are more logical allocation
1833                  * on this phy channel.
1834                  */
1835                 if (!d40_alloc_mask_free(phy, is_src, event)) {
1836                         /* Resume the other logical channels if any */
1837                         if (d40_chan_has_events(d40c)) {
1838                                 res = d40_channel_execute_command(d40c,
1839                                                                   D40_DMA_RUN);
1840                                 if (res)
1841                                         chan_err(d40c,
1842                                                 "Executing RUN command\n");
1843                         }
1844                         goto out;
1845                 }
1846         } else {
1847                 (void) d40_alloc_mask_free(phy, is_src, 0);
1848         }
1849
1850         /* Release physical channel */
1851         res = d40_channel_execute_command(d40c, D40_DMA_STOP);
1852         if (res) {
1853                 chan_err(d40c, "Failed to stop channel\n");
1854                 goto out;
1855         }
1856
1857         if (d40c->busy) {
1858                 pm_runtime_mark_last_busy(d40c->base->dev);
1859                 pm_runtime_put_autosuspend(d40c->base->dev);
1860         }
1861
1862         d40c->busy = false;
1863         d40c->phy_chan = NULL;
1864         d40c->configured = false;
1865         d40c->base->lookup_phy_chans[phy->num] = NULL;
1866 out:
1867
1868         pm_runtime_mark_last_busy(d40c->base->dev);
1869         pm_runtime_put_autosuspend(d40c->base->dev);
1870         return res;
1871 }
1872
1873 static bool d40_is_paused(struct d40_chan *d40c)
1874 {
1875         void __iomem *chanbase = chan_base(d40c);
1876         bool is_paused = false;
1877         unsigned long flags;
1878         void __iomem *active_reg;
1879         u32 status;
1880         u32 event;
1881
1882         spin_lock_irqsave(&d40c->lock, flags);
1883
1884         if (chan_is_physical(d40c)) {
1885                 if (d40c->phy_chan->num % 2 == 0)
1886                         active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1887                 else
1888                         active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1889
1890                 status = (readl(active_reg) &
1891                           D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1892                         D40_CHAN_POS(d40c->phy_chan->num);
1893                 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1894                         is_paused = true;
1895
1896                 goto _exit;
1897         }
1898
1899         if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1900             d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1901                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1902                 status = readl(chanbase + D40_CHAN_REG_SDLNK);
1903         } else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1904                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1905                 status = readl(chanbase + D40_CHAN_REG_SSLNK);
1906         } else {
1907                 chan_err(d40c, "Unknown direction\n");
1908                 goto _exit;
1909         }
1910
1911         status = (status & D40_EVENTLINE_MASK(event)) >>
1912                 D40_EVENTLINE_POS(event);
1913
1914         if (status != D40_DMA_RUN)
1915                 is_paused = true;
1916 _exit:
1917         spin_unlock_irqrestore(&d40c->lock, flags);
1918         return is_paused;
1919
1920 }
1921
1922
1923 static u32 stedma40_residue(struct dma_chan *chan)
1924 {
1925         struct d40_chan *d40c =
1926                 container_of(chan, struct d40_chan, chan);
1927         u32 bytes_left;
1928         unsigned long flags;
1929
1930         spin_lock_irqsave(&d40c->lock, flags);
1931         bytes_left = d40_residue(d40c);
1932         spin_unlock_irqrestore(&d40c->lock, flags);
1933
1934         return bytes_left;
1935 }
1936
1937 static int
1938 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
1939                 struct scatterlist *sg_src, struct scatterlist *sg_dst,
1940                 unsigned int sg_len, dma_addr_t src_dev_addr,
1941                 dma_addr_t dst_dev_addr)
1942 {
1943         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
1944         struct stedma40_half_channel_info *src_info = &cfg->src_info;
1945         struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
1946         int ret;
1947
1948         ret = d40_log_sg_to_lli(sg_src, sg_len,
1949                                 src_dev_addr,
1950                                 desc->lli_log.src,
1951                                 chan->log_def.lcsp1,
1952                                 src_info->data_width,
1953                                 dst_info->data_width);
1954
1955         ret = d40_log_sg_to_lli(sg_dst, sg_len,
1956                                 dst_dev_addr,
1957                                 desc->lli_log.dst,
1958                                 chan->log_def.lcsp3,
1959                                 dst_info->data_width,
1960                                 src_info->data_width);
1961
1962         return ret < 0 ? ret : 0;
1963 }
1964
1965 static int
1966 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
1967                 struct scatterlist *sg_src, struct scatterlist *sg_dst,
1968                 unsigned int sg_len, dma_addr_t src_dev_addr,
1969                 dma_addr_t dst_dev_addr)
1970 {
1971         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
1972         struct stedma40_half_channel_info *src_info = &cfg->src_info;
1973         struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
1974         unsigned long flags = 0;
1975         int ret;
1976
1977         if (desc->cyclic)
1978                 flags |= LLI_CYCLIC | LLI_TERM_INT;
1979
1980         ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
1981                                 desc->lli_phy.src,
1982                                 virt_to_phys(desc->lli_phy.src),
1983                                 chan->src_def_cfg,
1984                                 src_info, dst_info, flags);
1985
1986         ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
1987                                 desc->lli_phy.dst,
1988                                 virt_to_phys(desc->lli_phy.dst),
1989                                 chan->dst_def_cfg,
1990                                 dst_info, src_info, flags);
1991
1992         dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
1993                                    desc->lli_pool.size, DMA_TO_DEVICE);
1994
1995         return ret < 0 ? ret : 0;
1996 }
1997
1998
1999 static struct d40_desc *
2000 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2001               unsigned int sg_len, unsigned long dma_flags)
2002 {
2003         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2004         struct d40_desc *desc;
2005         int ret;
2006
2007         desc = d40_desc_get(chan);
2008         if (!desc)
2009                 return NULL;
2010
2011         desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2012                                         cfg->dst_info.data_width);
2013         if (desc->lli_len < 0) {
2014                 chan_err(chan, "Unaligned size\n");
2015                 goto err;
2016         }
2017
2018         ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2019         if (ret < 0) {
2020                 chan_err(chan, "Could not allocate lli\n");
2021                 goto err;
2022         }
2023
2024
2025         desc->lli_current = 0;
2026         desc->txd.flags = dma_flags;
2027         desc->txd.tx_submit = d40_tx_submit;
2028
2029         dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2030
2031         return desc;
2032
2033 err:
2034         d40_desc_free(chan, desc);
2035         return NULL;
2036 }
2037
2038 static dma_addr_t
2039 d40_get_dev_addr(struct d40_chan *chan, enum dma_transfer_direction direction)
2040 {
2041         struct stedma40_platform_data *plat = chan->base->plat_data;
2042         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2043         dma_addr_t addr = 0;
2044
2045         if (chan->runtime_addr)
2046                 return chan->runtime_addr;
2047
2048         if (direction == DMA_DEV_TO_MEM)
2049                 addr = plat->dev_rx[cfg->src_dev_type];
2050         else if (direction == DMA_MEM_TO_DEV)
2051                 addr = plat->dev_tx[cfg->dst_dev_type];
2052
2053         return addr;
2054 }
2055
2056 static struct dma_async_tx_descriptor *
2057 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2058             struct scatterlist *sg_dst, unsigned int sg_len,
2059             enum dma_transfer_direction direction, unsigned long dma_flags)
2060 {
2061         struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2062         dma_addr_t src_dev_addr = 0;
2063         dma_addr_t dst_dev_addr = 0;
2064         struct d40_desc *desc;
2065         unsigned long flags;
2066         int ret;
2067
2068         if (!chan->phy_chan) {
2069                 chan_err(chan, "Cannot prepare unallocated channel\n");
2070                 return NULL;
2071         }
2072
2073
2074         spin_lock_irqsave(&chan->lock, flags);
2075
2076         desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2077         if (desc == NULL)
2078                 goto err;
2079
2080         if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2081                 desc->cyclic = true;
2082
2083         if (direction != DMA_NONE) {
2084                 dma_addr_t dev_addr = d40_get_dev_addr(chan, direction);
2085
2086                 if (direction == DMA_DEV_TO_MEM)
2087                         src_dev_addr = dev_addr;
2088                 else if (direction == DMA_MEM_TO_DEV)
2089                         dst_dev_addr = dev_addr;
2090         }
2091
2092         if (chan_is_logical(chan))
2093                 ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2094                                       sg_len, src_dev_addr, dst_dev_addr);
2095         else
2096                 ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2097                                       sg_len, src_dev_addr, dst_dev_addr);
2098
2099         if (ret) {
2100                 chan_err(chan, "Failed to prepare %s sg job: %d\n",
2101                          chan_is_logical(chan) ? "log" : "phy", ret);
2102                 goto err;
2103         }
2104
2105         /*
2106          * add descriptor to the prepare queue in order to be able
2107          * to free them later in terminate_all
2108          */
2109         list_add_tail(&desc->node, &chan->prepare_queue);
2110
2111         spin_unlock_irqrestore(&chan->lock, flags);
2112
2113         return &desc->txd;
2114
2115 err:
2116         if (desc)
2117                 d40_desc_free(chan, desc);
2118         spin_unlock_irqrestore(&chan->lock, flags);
2119         return NULL;
2120 }
2121
2122 bool stedma40_filter(struct dma_chan *chan, void *data)
2123 {
2124         struct stedma40_chan_cfg *info = data;
2125         struct d40_chan *d40c =
2126                 container_of(chan, struct d40_chan, chan);
2127         int err;
2128
2129         if (data) {
2130                 err = d40_validate_conf(d40c, info);
2131                 if (!err)
2132                         d40c->dma_cfg = *info;
2133         } else
2134                 err = d40_config_memcpy(d40c);
2135
2136         if (!err)
2137                 d40c->configured = true;
2138
2139         return err == 0;
2140 }
2141 EXPORT_SYMBOL(stedma40_filter);
2142
2143 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2144 {
2145         bool realtime = d40c->dma_cfg.realtime;
2146         bool highprio = d40c->dma_cfg.high_priority;
2147         u32 prioreg = highprio ? D40_DREG_PSEG1 : D40_DREG_PCEG1;
2148         u32 rtreg = realtime ? D40_DREG_RSEG1 : D40_DREG_RCEG1;
2149         u32 event = D40_TYPE_TO_EVENT(dev_type);
2150         u32 group = D40_TYPE_TO_GROUP(dev_type);
2151         u32 bit = 1 << event;
2152
2153         /* Destination event lines are stored in the upper halfword */
2154         if (!src)
2155                 bit <<= 16;
2156
2157         writel(bit, d40c->base->virtbase + prioreg + group * 4);
2158         writel(bit, d40c->base->virtbase + rtreg + group * 4);
2159 }
2160
2161 static void d40_set_prio_realtime(struct d40_chan *d40c)
2162 {
2163         if (d40c->base->rev < 3)
2164                 return;
2165
2166         if ((d40c->dma_cfg.dir ==  STEDMA40_PERIPH_TO_MEM) ||
2167             (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH))
2168                 __d40_set_prio_rt(d40c, d40c->dma_cfg.src_dev_type, true);
2169
2170         if ((d40c->dma_cfg.dir ==  STEDMA40_MEM_TO_PERIPH) ||
2171             (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH))
2172                 __d40_set_prio_rt(d40c, d40c->dma_cfg.dst_dev_type, false);
2173 }
2174
2175 /* DMA ENGINE functions */
2176 static int d40_alloc_chan_resources(struct dma_chan *chan)
2177 {
2178         int err;
2179         unsigned long flags;
2180         struct d40_chan *d40c =
2181                 container_of(chan, struct d40_chan, chan);
2182         bool is_free_phy;
2183         spin_lock_irqsave(&d40c->lock, flags);
2184
2185         d40c->completed = chan->cookie = 1;
2186
2187         /* If no dma configuration is set use default configuration (memcpy) */
2188         if (!d40c->configured) {
2189                 err = d40_config_memcpy(d40c);
2190                 if (err) {
2191                         chan_err(d40c, "Failed to configure memcpy channel\n");
2192                         goto fail;
2193                 }
2194         }
2195
2196         err = d40_allocate_channel(d40c, &is_free_phy);
2197         if (err) {
2198                 chan_err(d40c, "Failed to allocate channel\n");
2199                 d40c->configured = false;
2200                 goto fail;
2201         }
2202
2203         pm_runtime_get_sync(d40c->base->dev);
2204         /* Fill in basic CFG register values */
2205         d40_phy_cfg(&d40c->dma_cfg, &d40c->src_def_cfg,
2206                     &d40c->dst_def_cfg, chan_is_logical(d40c));
2207
2208         d40_set_prio_realtime(d40c);
2209
2210         if (chan_is_logical(d40c)) {
2211                 d40_log_cfg(&d40c->dma_cfg,
2212                             &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2213
2214                 if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM)
2215                         d40c->lcpa = d40c->base->lcpa_base +
2216                           d40c->dma_cfg.src_dev_type * D40_LCPA_CHAN_SIZE;
2217                 else
2218                         d40c->lcpa = d40c->base->lcpa_base +
2219                           d40c->dma_cfg.dst_dev_type *
2220                           D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2221         }
2222
2223         dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2224                  chan_is_logical(d40c) ? "logical" : "physical",
2225                  d40c->phy_chan->num,
2226                  d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2227
2228
2229         /*
2230          * Only write channel configuration to the DMA if the physical
2231          * resource is free. In case of multiple logical channels
2232          * on the same physical resource, only the first write is necessary.
2233          */
2234         if (is_free_phy)
2235                 d40_config_write(d40c);
2236 fail:
2237         pm_runtime_mark_last_busy(d40c->base->dev);
2238         pm_runtime_put_autosuspend(d40c->base->dev);
2239         spin_unlock_irqrestore(&d40c->lock, flags);
2240         return err;
2241 }
2242
2243 static void d40_free_chan_resources(struct dma_chan *chan)
2244 {
2245         struct d40_chan *d40c =
2246                 container_of(chan, struct d40_chan, chan);
2247         int err;
2248         unsigned long flags;
2249
2250         if (d40c->phy_chan == NULL) {
2251                 chan_err(d40c, "Cannot free unallocated channel\n");
2252                 return;
2253         }
2254
2255
2256         spin_lock_irqsave(&d40c->lock, flags);
2257
2258         err = d40_free_dma(d40c);
2259
2260         if (err)
2261                 chan_err(d40c, "Failed to free channel\n");
2262         spin_unlock_irqrestore(&d40c->lock, flags);
2263 }
2264
2265 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2266                                                        dma_addr_t dst,
2267                                                        dma_addr_t src,
2268                                                        size_t size,
2269                                                        unsigned long dma_flags)
2270 {
2271         struct scatterlist dst_sg;
2272         struct scatterlist src_sg;
2273
2274         sg_init_table(&dst_sg, 1);
2275         sg_init_table(&src_sg, 1);
2276
2277         sg_dma_address(&dst_sg) = dst;
2278         sg_dma_address(&src_sg) = src;
2279
2280         sg_dma_len(&dst_sg) = size;
2281         sg_dma_len(&src_sg) = size;
2282
2283         return d40_prep_sg(chan, &src_sg, &dst_sg, 1, DMA_NONE, dma_flags);
2284 }
2285
2286 static struct dma_async_tx_descriptor *
2287 d40_prep_memcpy_sg(struct dma_chan *chan,
2288                    struct scatterlist *dst_sg, unsigned int dst_nents,
2289                    struct scatterlist *src_sg, unsigned int src_nents,
2290                    unsigned long dma_flags)
2291 {
2292         if (dst_nents != src_nents)
2293                 return NULL;
2294
2295         return d40_prep_sg(chan, src_sg, dst_sg, src_nents, DMA_NONE, dma_flags);
2296 }
2297
2298 static struct dma_async_tx_descriptor *d40_prep_slave_sg(struct dma_chan *chan,
2299                                                          struct scatterlist *sgl,
2300                                                          unsigned int sg_len,
2301                                                          enum dma_transfer_direction direction,
2302                                                          unsigned long dma_flags)
2303 {
2304         if (direction != DMA_DEV_TO_MEM && direction != DMA_MEM_TO_DEV)
2305                 return NULL;
2306
2307         return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2308 }
2309
2310 static struct dma_async_tx_descriptor *
2311 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2312                      size_t buf_len, size_t period_len,
2313                      enum dma_transfer_direction direction)
2314 {
2315         unsigned int periods = buf_len / period_len;
2316         struct dma_async_tx_descriptor *txd;
2317         struct scatterlist *sg;
2318         int i;
2319
2320         sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2321         for (i = 0; i < periods; i++) {
2322                 sg_dma_address(&sg[i]) = dma_addr;
2323                 sg_dma_len(&sg[i]) = period_len;
2324                 dma_addr += period_len;
2325         }
2326
2327         sg[periods].offset = 0;
2328         sg[periods].length = 0;
2329         sg[periods].page_link =
2330                 ((unsigned long)sg | 0x01) & ~0x02;
2331
2332         txd = d40_prep_sg(chan, sg, sg, periods, direction,
2333                           DMA_PREP_INTERRUPT);
2334
2335         kfree(sg);
2336
2337         return txd;
2338 }
2339
2340 static enum dma_status d40_tx_status(struct dma_chan *chan,
2341                                      dma_cookie_t cookie,
2342                                      struct dma_tx_state *txstate)
2343 {
2344         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2345         dma_cookie_t last_used;
2346         dma_cookie_t last_complete;
2347         int ret;
2348
2349         if (d40c->phy_chan == NULL) {
2350                 chan_err(d40c, "Cannot read status of unallocated channel\n");
2351                 return -EINVAL;
2352         }
2353
2354         last_complete = d40c->completed;
2355         last_used = chan->cookie;
2356
2357         if (d40_is_paused(d40c))
2358                 ret = DMA_PAUSED;
2359         else
2360                 ret = dma_async_is_complete(cookie, last_complete, last_used);
2361
2362         dma_set_tx_state(txstate, last_complete, last_used,
2363                          stedma40_residue(chan));
2364
2365         return ret;
2366 }
2367
2368 static void d40_issue_pending(struct dma_chan *chan)
2369 {
2370         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2371         unsigned long flags;
2372
2373         if (d40c->phy_chan == NULL) {
2374                 chan_err(d40c, "Channel is not allocated!\n");
2375                 return;
2376         }
2377
2378         spin_lock_irqsave(&d40c->lock, flags);
2379
2380         list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2381
2382         /* Busy means that queued jobs are already being processed */
2383         if (!d40c->busy)
2384                 (void) d40_queue_start(d40c);
2385
2386         spin_unlock_irqrestore(&d40c->lock, flags);
2387 }
2388
2389 static int
2390 dma40_config_to_halfchannel(struct d40_chan *d40c,
2391                             struct stedma40_half_channel_info *info,
2392                             enum dma_slave_buswidth width,
2393                             u32 maxburst)
2394 {
2395         enum stedma40_periph_data_width addr_width;
2396         int psize;
2397
2398         switch (width) {
2399         case DMA_SLAVE_BUSWIDTH_1_BYTE:
2400                 addr_width = STEDMA40_BYTE_WIDTH;
2401                 break;
2402         case DMA_SLAVE_BUSWIDTH_2_BYTES:
2403                 addr_width = STEDMA40_HALFWORD_WIDTH;
2404                 break;
2405         case DMA_SLAVE_BUSWIDTH_4_BYTES:
2406                 addr_width = STEDMA40_WORD_WIDTH;
2407                 break;
2408         case DMA_SLAVE_BUSWIDTH_8_BYTES:
2409                 addr_width = STEDMA40_DOUBLEWORD_WIDTH;
2410                 break;
2411         default:
2412                 dev_err(d40c->base->dev,
2413                         "illegal peripheral address width "
2414                         "requested (%d)\n",
2415                         width);
2416                 return -EINVAL;
2417         }
2418
2419         if (chan_is_logical(d40c)) {
2420                 if (maxburst >= 16)
2421                         psize = STEDMA40_PSIZE_LOG_16;
2422                 else if (maxburst >= 8)
2423                         psize = STEDMA40_PSIZE_LOG_8;
2424                 else if (maxburst >= 4)
2425                         psize = STEDMA40_PSIZE_LOG_4;
2426                 else
2427                         psize = STEDMA40_PSIZE_LOG_1;
2428         } else {
2429                 if (maxburst >= 16)
2430                         psize = STEDMA40_PSIZE_PHY_16;
2431                 else if (maxburst >= 8)
2432                         psize = STEDMA40_PSIZE_PHY_8;
2433                 else if (maxburst >= 4)
2434                         psize = STEDMA40_PSIZE_PHY_4;
2435                 else
2436                         psize = STEDMA40_PSIZE_PHY_1;
2437         }
2438
2439         info->data_width = addr_width;
2440         info->psize = psize;
2441         info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2442
2443         return 0;
2444 }
2445
2446 /* Runtime reconfiguration extension */
2447 static int d40_set_runtime_config(struct dma_chan *chan,
2448                                   struct dma_slave_config *config)
2449 {
2450         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2451         struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2452         enum dma_slave_buswidth src_addr_width, dst_addr_width;
2453         dma_addr_t config_addr;
2454         u32 src_maxburst, dst_maxburst;
2455         int ret;
2456
2457         src_addr_width = config->src_addr_width;
2458         src_maxburst = config->src_maxburst;
2459         dst_addr_width = config->dst_addr_width;
2460         dst_maxburst = config->dst_maxburst;
2461
2462         if (config->direction == DMA_DEV_TO_MEM) {
2463                 dma_addr_t dev_addr_rx =
2464                         d40c->base->plat_data->dev_rx[cfg->src_dev_type];
2465
2466                 config_addr = config->src_addr;
2467                 if (dev_addr_rx)
2468                         dev_dbg(d40c->base->dev,
2469                                 "channel has a pre-wired RX address %08x "
2470                                 "overriding with %08x\n",
2471                                 dev_addr_rx, config_addr);
2472                 if (cfg->dir != STEDMA40_PERIPH_TO_MEM)
2473                         dev_dbg(d40c->base->dev,
2474                                 "channel was not configured for peripheral "
2475                                 "to memory transfer (%d) overriding\n",
2476                                 cfg->dir);
2477                 cfg->dir = STEDMA40_PERIPH_TO_MEM;
2478
2479                 /* Configure the memory side */
2480                 if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2481                         dst_addr_width = src_addr_width;
2482                 if (dst_maxburst == 0)
2483                         dst_maxburst = src_maxburst;
2484
2485         } else if (config->direction == DMA_MEM_TO_DEV) {
2486                 dma_addr_t dev_addr_tx =
2487                         d40c->base->plat_data->dev_tx[cfg->dst_dev_type];
2488
2489                 config_addr = config->dst_addr;
2490                 if (dev_addr_tx)
2491                         dev_dbg(d40c->base->dev,
2492                                 "channel has a pre-wired TX address %08x "
2493                                 "overriding with %08x\n",
2494                                 dev_addr_tx, config_addr);
2495                 if (cfg->dir != STEDMA40_MEM_TO_PERIPH)
2496                         dev_dbg(d40c->base->dev,
2497                                 "channel was not configured for memory "
2498                                 "to peripheral transfer (%d) overriding\n",
2499                                 cfg->dir);
2500                 cfg->dir = STEDMA40_MEM_TO_PERIPH;
2501
2502                 /* Configure the memory side */
2503                 if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2504                         src_addr_width = dst_addr_width;
2505                 if (src_maxburst == 0)
2506                         src_maxburst = dst_maxburst;
2507         } else {
2508                 dev_err(d40c->base->dev,
2509                         "unrecognized channel direction %d\n",
2510                         config->direction);
2511                 return -EINVAL;
2512         }
2513
2514         if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2515                 dev_err(d40c->base->dev,
2516                         "src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2517                         src_maxburst,
2518                         src_addr_width,
2519                         dst_maxburst,
2520                         dst_addr_width);
2521                 return -EINVAL;
2522         }
2523
2524         ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2525                                           src_addr_width,
2526                                           src_maxburst);
2527         if (ret)
2528                 return ret;
2529
2530         ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2531                                           dst_addr_width,
2532                                           dst_maxburst);
2533         if (ret)
2534                 return ret;
2535
2536         /* Fill in register values */
2537         if (chan_is_logical(d40c))
2538                 d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2539         else
2540                 d40_phy_cfg(cfg, &d40c->src_def_cfg,
2541                             &d40c->dst_def_cfg, false);
2542
2543         /* These settings will take precedence later */
2544         d40c->runtime_addr = config_addr;
2545         d40c->runtime_direction = config->direction;
2546         dev_dbg(d40c->base->dev,
2547                 "configured channel %s for %s, data width %d/%d, "
2548                 "maxburst %d/%d elements, LE, no flow control\n",
2549                 dma_chan_name(chan),
2550                 (config->direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2551                 src_addr_width, dst_addr_width,
2552                 src_maxburst, dst_maxburst);
2553
2554         return 0;
2555 }
2556
2557 static int d40_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
2558                        unsigned long arg)
2559 {
2560         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2561
2562         if (d40c->phy_chan == NULL) {
2563                 chan_err(d40c, "Channel is not allocated!\n");
2564                 return -EINVAL;
2565         }
2566
2567         switch (cmd) {
2568         case DMA_TERMINATE_ALL:
2569                 return d40_terminate_all(d40c);
2570         case DMA_PAUSE:
2571                 return d40_pause(d40c);
2572         case DMA_RESUME:
2573                 return d40_resume(d40c);
2574         case DMA_SLAVE_CONFIG:
2575                 return d40_set_runtime_config(chan,
2576                         (struct dma_slave_config *) arg);
2577         default:
2578                 break;
2579         }
2580
2581         /* Other commands are unimplemented */
2582         return -ENXIO;
2583 }
2584
2585 /* Initialization functions */
2586
2587 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2588                                  struct d40_chan *chans, int offset,
2589                                  int num_chans)
2590 {
2591         int i = 0;
2592         struct d40_chan *d40c;
2593
2594         INIT_LIST_HEAD(&dma->channels);
2595
2596         for (i = offset; i < offset + num_chans; i++) {
2597                 d40c = &chans[i];
2598                 d40c->base = base;
2599                 d40c->chan.device = dma;
2600
2601                 spin_lock_init(&d40c->lock);
2602
2603                 d40c->log_num = D40_PHY_CHAN;
2604
2605                 INIT_LIST_HEAD(&d40c->active);
2606                 INIT_LIST_HEAD(&d40c->queue);
2607                 INIT_LIST_HEAD(&d40c->pending_queue);
2608                 INIT_LIST_HEAD(&d40c->client);
2609                 INIT_LIST_HEAD(&d40c->prepare_queue);
2610
2611                 tasklet_init(&d40c->tasklet, dma_tasklet,
2612                              (unsigned long) d40c);
2613
2614                 list_add_tail(&d40c->chan.device_node,
2615                               &dma->channels);
2616         }
2617 }
2618
2619 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2620 {
2621         if (dma_has_cap(DMA_SLAVE, dev->cap_mask))
2622                 dev->device_prep_slave_sg = d40_prep_slave_sg;
2623
2624         if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2625                 dev->device_prep_dma_memcpy = d40_prep_memcpy;
2626
2627                 /*
2628                  * This controller can only access address at even
2629                  * 32bit boundaries, i.e. 2^2
2630                  */
2631                 dev->copy_align = 2;
2632         }
2633
2634         if (dma_has_cap(DMA_SG, dev->cap_mask))
2635                 dev->device_prep_dma_sg = d40_prep_memcpy_sg;
2636
2637         if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2638                 dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2639
2640         dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2641         dev->device_free_chan_resources = d40_free_chan_resources;
2642         dev->device_issue_pending = d40_issue_pending;
2643         dev->device_tx_status = d40_tx_status;
2644         dev->device_control = d40_control;
2645         dev->dev = base->dev;
2646 }
2647
2648 static int __init d40_dmaengine_init(struct d40_base *base,
2649                                      int num_reserved_chans)
2650 {
2651         int err ;
2652
2653         d40_chan_init(base, &base->dma_slave, base->log_chans,
2654                       0, base->num_log_chans);
2655
2656         dma_cap_zero(base->dma_slave.cap_mask);
2657         dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2658         dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2659
2660         d40_ops_init(base, &base->dma_slave);
2661
2662         err = dma_async_device_register(&base->dma_slave);
2663
2664         if (err) {
2665                 d40_err(base->dev, "Failed to register slave channels\n");
2666                 goto failure1;
2667         }
2668
2669         d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2670                       base->num_log_chans, base->plat_data->memcpy_len);
2671
2672         dma_cap_zero(base->dma_memcpy.cap_mask);
2673         dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2674         dma_cap_set(DMA_SG, base->dma_memcpy.cap_mask);
2675
2676         d40_ops_init(base, &base->dma_memcpy);
2677
2678         err = dma_async_device_register(&base->dma_memcpy);
2679
2680         if (err) {
2681                 d40_err(base->dev,
2682                         "Failed to regsiter memcpy only channels\n");
2683                 goto failure2;
2684         }
2685
2686         d40_chan_init(base, &base->dma_both, base->phy_chans,
2687                       0, num_reserved_chans);
2688
2689         dma_cap_zero(base->dma_both.cap_mask);
2690         dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2691         dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2692         dma_cap_set(DMA_SG, base->dma_both.cap_mask);
2693         dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2694
2695         d40_ops_init(base, &base->dma_both);
2696         err = dma_async_device_register(&base->dma_both);
2697
2698         if (err) {
2699                 d40_err(base->dev,
2700                         "Failed to register logical and physical capable channels\n");
2701                 goto failure3;
2702         }
2703         return 0;
2704 failure3:
2705         dma_async_device_unregister(&base->dma_memcpy);
2706 failure2:
2707         dma_async_device_unregister(&base->dma_slave);
2708 failure1:
2709         return err;
2710 }
2711
2712 /* Suspend resume functionality */
2713 #ifdef CONFIG_PM
2714 static int dma40_pm_suspend(struct device *dev)
2715 {
2716         struct platform_device *pdev = to_platform_device(dev);
2717         struct d40_base *base = platform_get_drvdata(pdev);
2718         int ret = 0;
2719         if (!pm_runtime_suspended(dev))
2720                 return -EBUSY;
2721
2722         if (base->lcpa_regulator)
2723                 ret = regulator_disable(base->lcpa_regulator);
2724         return ret;
2725 }
2726
2727 static int dma40_runtime_suspend(struct device *dev)
2728 {
2729         struct platform_device *pdev = to_platform_device(dev);
2730         struct d40_base *base = platform_get_drvdata(pdev);
2731
2732         d40_save_restore_registers(base, true);
2733
2734         /* Don't disable/enable clocks for v1 due to HW bugs */
2735         if (base->rev != 1)
2736                 writel_relaxed(base->gcc_pwr_off_mask,
2737                                base->virtbase + D40_DREG_GCC);
2738
2739         return 0;
2740 }
2741
2742 static int dma40_runtime_resume(struct device *dev)
2743 {
2744         struct platform_device *pdev = to_platform_device(dev);
2745         struct d40_base *base = platform_get_drvdata(pdev);
2746
2747         if (base->initialized)
2748                 d40_save_restore_registers(base, false);
2749
2750         writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
2751                        base->virtbase + D40_DREG_GCC);
2752         return 0;
2753 }
2754
2755 static int dma40_resume(struct device *dev)
2756 {
2757         struct platform_device *pdev = to_platform_device(dev);
2758         struct d40_base *base = platform_get_drvdata(pdev);
2759         int ret = 0;
2760
2761         if (base->lcpa_regulator)
2762                 ret = regulator_enable(base->lcpa_regulator);
2763
2764         return ret;
2765 }
2766
2767 static const struct dev_pm_ops dma40_pm_ops = {
2768         .suspend                = dma40_pm_suspend,
2769         .runtime_suspend        = dma40_runtime_suspend,
2770         .runtime_resume         = dma40_runtime_resume,
2771         .resume                 = dma40_resume,
2772 };
2773 #define DMA40_PM_OPS    (&dma40_pm_ops)
2774 #else
2775 #define DMA40_PM_OPS    NULL
2776 #endif
2777
2778 /* Initialization functions. */
2779
2780 static int __init d40_phy_res_init(struct d40_base *base)
2781 {
2782         int i;
2783         int num_phy_chans_avail = 0;
2784         u32 val[2];
2785         int odd_even_bit = -2;
2786         int gcc = D40_DREG_GCC_ENA;
2787
2788         val[0] = readl(base->virtbase + D40_DREG_PRSME);
2789         val[1] = readl(base->virtbase + D40_DREG_PRSMO);
2790
2791         for (i = 0; i < base->num_phy_chans; i++) {
2792                 base->phy_res[i].num = i;
2793                 odd_even_bit += 2 * ((i % 2) == 0);
2794                 if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
2795                         /* Mark security only channels as occupied */
2796                         base->phy_res[i].allocated_src = D40_ALLOC_PHY;
2797                         base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
2798                         base->phy_res[i].reserved = true;
2799                         gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
2800                                                        D40_DREG_GCC_SRC);
2801                         gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
2802                                                        D40_DREG_GCC_DST);
2803
2804
2805                 } else {
2806                         base->phy_res[i].allocated_src = D40_ALLOC_FREE;
2807                         base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
2808                         base->phy_res[i].reserved = false;
2809                         num_phy_chans_avail++;
2810                 }
2811                 spin_lock_init(&base->phy_res[i].lock);
2812         }
2813
2814         /* Mark disabled channels as occupied */
2815         for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
2816                 int chan = base->plat_data->disabled_channels[i];
2817
2818                 base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
2819                 base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
2820                 base->phy_res[chan].reserved = true;
2821                 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
2822                                                D40_DREG_GCC_SRC);
2823                 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
2824                                                D40_DREG_GCC_DST);
2825                 num_phy_chans_avail--;
2826         }
2827
2828         dev_info(base->dev, "%d of %d physical DMA channels available\n",
2829                  num_phy_chans_avail, base->num_phy_chans);
2830
2831         /* Verify settings extended vs standard */
2832         val[0] = readl(base->virtbase + D40_DREG_PRTYP);
2833
2834         for (i = 0; i < base->num_phy_chans; i++) {
2835
2836                 if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
2837                     (val[0] & 0x3) != 1)
2838                         dev_info(base->dev,
2839                                  "[%s] INFO: channel %d is misconfigured (%d)\n",
2840                                  __func__, i, val[0] & 0x3);
2841
2842                 val[0] = val[0] >> 2;
2843         }
2844
2845         /*
2846          * To keep things simple, Enable all clocks initially.
2847          * The clocks will get managed later post channel allocation.
2848          * The clocks for the event lines on which reserved channels exists
2849          * are not managed here.
2850          */
2851         writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
2852         base->gcc_pwr_off_mask = gcc;
2853
2854         return num_phy_chans_avail;
2855 }
2856
2857 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
2858 {
2859         struct stedma40_platform_data *plat_data;
2860         struct clk *clk = NULL;
2861         void __iomem *virtbase = NULL;
2862         struct resource *res = NULL;
2863         struct d40_base *base = NULL;
2864         int num_log_chans = 0;
2865         int num_phy_chans;
2866         int i;
2867         u32 pid;
2868         u32 cid;
2869         u8 rev;
2870
2871         clk = clk_get(&pdev->dev, NULL);
2872
2873         if (IS_ERR(clk)) {
2874                 d40_err(&pdev->dev, "No matching clock found\n");
2875                 goto failure;
2876         }
2877
2878         clk_enable(clk);
2879
2880         /* Get IO for DMAC base address */
2881         res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
2882         if (!res)
2883                 goto failure;
2884
2885         if (request_mem_region(res->start, resource_size(res),
2886                                D40_NAME " I/O base") == NULL)
2887                 goto failure;
2888
2889         virtbase = ioremap(res->start, resource_size(res));
2890         if (!virtbase)
2891                 goto failure;
2892
2893         /* This is just a regular AMBA PrimeCell ID actually */
2894         for (pid = 0, i = 0; i < 4; i++)
2895                 pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
2896                         & 255) << (i * 8);
2897         for (cid = 0, i = 0; i < 4; i++)
2898                 cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
2899                         & 255) << (i * 8);
2900
2901         if (cid != AMBA_CID) {
2902                 d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
2903                 goto failure;
2904         }
2905         if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
2906                 d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
2907                         AMBA_MANF_BITS(pid),
2908                         AMBA_VENDOR_ST);
2909                 goto failure;
2910         }
2911         /*
2912          * HW revision:
2913          * DB8500ed has revision 0
2914          * ? has revision 1
2915          * DB8500v1 has revision 2
2916          * DB8500v2 has revision 3
2917          */
2918         rev = AMBA_REV_BITS(pid);
2919
2920         /* The number of physical channels on this HW */
2921         num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
2922
2923         dev_info(&pdev->dev, "hardware revision: %d @ 0x%x\n",
2924                  rev, res->start);
2925
2926         plat_data = pdev->dev.platform_data;
2927
2928         /* Count the number of logical channels in use */
2929         for (i = 0; i < plat_data->dev_len; i++)
2930                 if (plat_data->dev_rx[i] != 0)
2931                         num_log_chans++;
2932
2933         for (i = 0; i < plat_data->dev_len; i++)
2934                 if (plat_data->dev_tx[i] != 0)
2935                         num_log_chans++;
2936
2937         base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
2938                        (num_phy_chans + num_log_chans + plat_data->memcpy_len) *
2939                        sizeof(struct d40_chan), GFP_KERNEL);
2940
2941         if (base == NULL) {
2942                 d40_err(&pdev->dev, "Out of memory\n");
2943                 goto failure;
2944         }
2945
2946         base->rev = rev;
2947         base->clk = clk;
2948         base->num_phy_chans = num_phy_chans;
2949         base->num_log_chans = num_log_chans;
2950         base->phy_start = res->start;
2951         base->phy_size = resource_size(res);
2952         base->virtbase = virtbase;
2953         base->plat_data = plat_data;
2954         base->dev = &pdev->dev;
2955         base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
2956         base->log_chans = &base->phy_chans[num_phy_chans];
2957
2958         base->phy_res = kzalloc(num_phy_chans * sizeof(struct d40_phy_res),
2959                                 GFP_KERNEL);
2960         if (!base->phy_res)
2961                 goto failure;
2962
2963         base->lookup_phy_chans = kzalloc(num_phy_chans *
2964                                          sizeof(struct d40_chan *),
2965                                          GFP_KERNEL);
2966         if (!base->lookup_phy_chans)
2967                 goto failure;
2968
2969         if (num_log_chans + plat_data->memcpy_len) {
2970                 /*
2971                  * The max number of logical channels are event lines for all
2972                  * src devices and dst devices
2973                  */
2974                 base->lookup_log_chans = kzalloc(plat_data->dev_len * 2 *
2975                                                  sizeof(struct d40_chan *),
2976                                                  GFP_KERNEL);
2977                 if (!base->lookup_log_chans)
2978                         goto failure;
2979         }
2980
2981         base->reg_val_backup_chan = kmalloc(base->num_phy_chans *
2982                                             sizeof(d40_backup_regs_chan),
2983                                             GFP_KERNEL);
2984         if (!base->reg_val_backup_chan)
2985                 goto failure;
2986
2987         base->lcla_pool.alloc_map =
2988                 kzalloc(num_phy_chans * sizeof(struct d40_desc *)
2989                         * D40_LCLA_LINK_PER_EVENT_GRP, GFP_KERNEL);
2990         if (!base->lcla_pool.alloc_map)
2991                 goto failure;
2992
2993         base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
2994                                             0, SLAB_HWCACHE_ALIGN,
2995                                             NULL);
2996         if (base->desc_slab == NULL)
2997                 goto failure;
2998
2999         return base;
3000
3001 failure:
3002         if (!IS_ERR(clk)) {
3003                 clk_disable(clk);
3004                 clk_put(clk);
3005         }
3006         if (virtbase)
3007                 iounmap(virtbase);
3008         if (res)
3009                 release_mem_region(res->start,
3010                                    resource_size(res));
3011         if (virtbase)
3012                 iounmap(virtbase);
3013
3014         if (base) {
3015                 kfree(base->lcla_pool.alloc_map);
3016                 kfree(base->lookup_log_chans);
3017                 kfree(base->lookup_phy_chans);
3018                 kfree(base->phy_res);
3019                 kfree(base);
3020         }
3021
3022         return NULL;
3023 }
3024
3025 static void __init d40_hw_init(struct d40_base *base)
3026 {
3027
3028         static struct d40_reg_val dma_init_reg[] = {
3029                 /* Clock every part of the DMA block from start */
3030                 { .reg = D40_DREG_GCC,    .val = D40_DREG_GCC_ENABLE_ALL},
3031
3032                 /* Interrupts on all logical channels */
3033                 { .reg = D40_DREG_LCMIS0, .val = 0xFFFFFFFF},
3034                 { .reg = D40_DREG_LCMIS1, .val = 0xFFFFFFFF},
3035                 { .reg = D40_DREG_LCMIS2, .val = 0xFFFFFFFF},
3036                 { .reg = D40_DREG_LCMIS3, .val = 0xFFFFFFFF},
3037                 { .reg = D40_DREG_LCICR0, .val = 0xFFFFFFFF},
3038                 { .reg = D40_DREG_LCICR1, .val = 0xFFFFFFFF},
3039                 { .reg = D40_DREG_LCICR2, .val = 0xFFFFFFFF},
3040                 { .reg = D40_DREG_LCICR3, .val = 0xFFFFFFFF},
3041                 { .reg = D40_DREG_LCTIS0, .val = 0xFFFFFFFF},
3042                 { .reg = D40_DREG_LCTIS1, .val = 0xFFFFFFFF},
3043                 { .reg = D40_DREG_LCTIS2, .val = 0xFFFFFFFF},
3044                 { .reg = D40_DREG_LCTIS3, .val = 0xFFFFFFFF}
3045         };
3046         int i;
3047         u32 prmseo[2] = {0, 0};
3048         u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3049         u32 pcmis = 0;
3050         u32 pcicr = 0;
3051
3052         for (i = 0; i < ARRAY_SIZE(dma_init_reg); i++)
3053                 writel(dma_init_reg[i].val,
3054                        base->virtbase + dma_init_reg[i].reg);
3055
3056         /* Configure all our dma channels to default settings */
3057         for (i = 0; i < base->num_phy_chans; i++) {
3058
3059                 activeo[i % 2] = activeo[i % 2] << 2;
3060
3061                 if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3062                     == D40_ALLOC_PHY) {
3063                         activeo[i % 2] |= 3;
3064                         continue;
3065                 }
3066
3067                 /* Enable interrupt # */
3068                 pcmis = (pcmis << 1) | 1;
3069
3070                 /* Clear interrupt # */
3071                 pcicr = (pcicr << 1) | 1;
3072
3073                 /* Set channel to physical mode */
3074                 prmseo[i % 2] = prmseo[i % 2] << 2;
3075                 prmseo[i % 2] |= 1;
3076
3077         }
3078
3079         writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3080         writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3081         writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3082         writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3083
3084         /* Write which interrupt to enable */
3085         writel(pcmis, base->virtbase + D40_DREG_PCMIS);
3086
3087         /* Write which interrupt to clear */
3088         writel(pcicr, base->virtbase + D40_DREG_PCICR);
3089
3090 }
3091
3092 static int __init d40_lcla_allocate(struct d40_base *base)
3093 {
3094         struct d40_lcla_pool *pool = &base->lcla_pool;
3095         unsigned long *page_list;
3096         int i, j;
3097         int ret = 0;
3098
3099         /*
3100          * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3101          * To full fill this hardware requirement without wasting 256 kb
3102          * we allocate pages until we get an aligned one.
3103          */
3104         page_list = kmalloc(sizeof(unsigned long) * MAX_LCLA_ALLOC_ATTEMPTS,
3105                             GFP_KERNEL);
3106
3107         if (!page_list) {
3108                 ret = -ENOMEM;
3109                 goto failure;
3110         }
3111
3112         /* Calculating how many pages that are required */
3113         base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3114
3115         for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3116                 page_list[i] = __get_free_pages(GFP_KERNEL,
3117                                                 base->lcla_pool.pages);
3118                 if (!page_list[i]) {
3119
3120                         d40_err(base->dev, "Failed to allocate %d pages.\n",
3121                                 base->lcla_pool.pages);
3122
3123                         for (j = 0; j < i; j++)
3124                                 free_pages(page_list[j], base->lcla_pool.pages);
3125                         goto failure;
3126                 }
3127
3128                 if ((virt_to_phys((void *)page_list[i]) &
3129                      (LCLA_ALIGNMENT - 1)) == 0)
3130                         break;
3131         }
3132
3133         for (j = 0; j < i; j++)
3134                 free_pages(page_list[j], base->lcla_pool.pages);
3135
3136         if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3137                 base->lcla_pool.base = (void *)page_list[i];
3138         } else {
3139                 /*
3140                  * After many attempts and no succees with finding the correct
3141                  * alignment, try with allocating a big buffer.
3142                  */
3143                 dev_warn(base->dev,
3144                          "[%s] Failed to get %d pages @ 18 bit align.\n",
3145                          __func__, base->lcla_pool.pages);
3146                 base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3147                                                          base->num_phy_chans +
3148                                                          LCLA_ALIGNMENT,
3149                                                          GFP_KERNEL);
3150                 if (!base->lcla_pool.base_unaligned) {
3151                         ret = -ENOMEM;
3152                         goto failure;
3153                 }
3154
3155                 base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3156                                                  LCLA_ALIGNMENT);
3157         }
3158
3159         pool->dma_addr = dma_map_single(base->dev, pool->base,
3160                                         SZ_1K * base->num_phy_chans,
3161                                         DMA_TO_DEVICE);
3162         if (dma_mapping_error(base->dev, pool->dma_addr)) {
3163                 pool->dma_addr = 0;
3164                 ret = -ENOMEM;
3165                 goto failure;
3166         }
3167
3168         writel(virt_to_phys(base->lcla_pool.base),
3169                base->virtbase + D40_DREG_LCLA);
3170 failure:
3171         kfree(page_list);
3172         return ret;
3173 }
3174
3175 static int __init d40_probe(struct platform_device *pdev)
3176 {
3177         int err;
3178         int ret = -ENOENT;
3179         struct d40_base *base;
3180         struct resource *res = NULL;
3181         int num_reserved_chans;
3182         u32 val;
3183
3184         base = d40_hw_detect_init(pdev);
3185
3186         if (!base)
3187                 goto failure;
3188
3189         num_reserved_chans = d40_phy_res_init(base);
3190
3191         platform_set_drvdata(pdev, base);
3192
3193         spin_lock_init(&base->interrupt_lock);
3194         spin_lock_init(&base->execmd_lock);
3195
3196         /* Get IO for logical channel parameter address */
3197         res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
3198         if (!res) {
3199                 ret = -ENOENT;
3200                 d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
3201                 goto failure;
3202         }
3203         base->lcpa_size = resource_size(res);
3204         base->phy_lcpa = res->start;
3205
3206         if (request_mem_region(res->start, resource_size(res),
3207                                D40_NAME " I/O lcpa") == NULL) {
3208                 ret = -EBUSY;
3209                 d40_err(&pdev->dev,
3210                         "Failed to request LCPA region 0x%x-0x%x\n",
3211                         res->start, res->end);
3212                 goto failure;
3213         }
3214
3215         /* We make use of ESRAM memory for this. */
3216         val = readl(base->virtbase + D40_DREG_LCPA);
3217         if (res->start != val && val != 0) {
3218                 dev_warn(&pdev->dev,
3219                          "[%s] Mismatch LCPA dma 0x%x, def 0x%x\n",
3220                          __func__, val, res->start);
3221         } else
3222                 writel(res->start, base->virtbase + D40_DREG_LCPA);
3223
3224         base->lcpa_base = ioremap(res->start, resource_size(res));
3225         if (!base->lcpa_base) {
3226                 ret = -ENOMEM;
3227                 d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
3228                 goto failure;
3229         }
3230         /* If lcla has to be located in ESRAM we don't need to allocate */
3231         if (base->plat_data->use_esram_lcla) {
3232                 res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3233                                                         "lcla_esram");
3234                 if (!res) {
3235                         ret = -ENOENT;
3236                         d40_err(&pdev->dev,
3237                                 "No \"lcla_esram\" memory resource\n");
3238                         goto failure;
3239                 }
3240                 base->lcla_pool.base = ioremap(res->start,
3241                                                 resource_size(res));
3242                 if (!base->lcla_pool.base) {
3243                         ret = -ENOMEM;
3244                         d40_err(&pdev->dev, "Failed to ioremap LCLA region\n");
3245                         goto failure;
3246                 }
3247                 writel(res->start, base->virtbase + D40_DREG_LCLA);
3248
3249         } else {
3250                 ret = d40_lcla_allocate(base);
3251                 if (ret) {
3252                         d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
3253                         goto failure;
3254                 }
3255         }
3256
3257         spin_lock_init(&base->lcla_pool.lock);
3258
3259         base->irq = platform_get_irq(pdev, 0);
3260
3261         ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3262         if (ret) {
3263                 d40_err(&pdev->dev, "No IRQ defined\n");
3264                 goto failure;
3265         }
3266
3267         pm_runtime_irq_safe(base->dev);
3268         pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3269         pm_runtime_use_autosuspend(base->dev);
3270         pm_runtime_enable(base->dev);
3271         pm_runtime_resume(base->dev);
3272
3273         if (base->plat_data->use_esram_lcla) {
3274
3275                 base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3276                 if (IS_ERR(base->lcpa_regulator)) {
3277                         d40_err(&pdev->dev, "Failed to get lcpa_regulator\n");
3278                         base->lcpa_regulator = NULL;
3279                         goto failure;
3280                 }
3281
3282                 ret = regulator_enable(base->lcpa_regulator);
3283                 if (ret) {
3284                         d40_err(&pdev->dev,
3285                                 "Failed to enable lcpa_regulator\n");
3286                         regulator_put(base->lcpa_regulator);
3287                         base->lcpa_regulator = NULL;
3288                         goto failure;
3289                 }
3290         }
3291
3292         base->initialized = true;
3293         err = d40_dmaengine_init(base, num_reserved_chans);
3294         if (err)
3295                 goto failure;
3296
3297         d40_hw_init(base);
3298
3299         dev_info(base->dev, "initialized\n");
3300         return 0;
3301
3302 failure:
3303         if (base) {
3304                 if (base->desc_slab)
3305                         kmem_cache_destroy(base->desc_slab);
3306                 if (base->virtbase)
3307                         iounmap(base->virtbase);
3308
3309                 if (base->lcla_pool.base && base->plat_data->use_esram_lcla) {
3310                         iounmap(base->lcla_pool.base);
3311                         base->lcla_pool.base = NULL;
3312                 }
3313
3314                 if (base->lcla_pool.dma_addr)
3315                         dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3316                                          SZ_1K * base->num_phy_chans,
3317                                          DMA_TO_DEVICE);
3318
3319                 if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3320                         free_pages((unsigned long)base->lcla_pool.base,
3321                                    base->lcla_pool.pages);
3322
3323                 kfree(base->lcla_pool.base_unaligned);
3324
3325                 if (base->phy_lcpa)
3326                         release_mem_region(base->phy_lcpa,
3327                                            base->lcpa_size);
3328                 if (base->phy_start)
3329                         release_mem_region(base->phy_start,
3330                                            base->phy_size);
3331                 if (base->clk) {
3332                         clk_disable(base->clk);
3333                         clk_put(base->clk);
3334                 }
3335
3336                 if (base->lcpa_regulator) {
3337                         regulator_disable(base->lcpa_regulator);
3338                         regulator_put(base->lcpa_regulator);
3339                 }
3340
3341                 kfree(base->lcla_pool.alloc_map);
3342                 kfree(base->lookup_log_chans);
3343                 kfree(base->lookup_phy_chans);
3344                 kfree(base->phy_res);
3345                 kfree(base);
3346         }
3347
3348         d40_err(&pdev->dev, "probe failed\n");
3349         return ret;
3350 }
3351
3352 static struct platform_driver d40_driver = {
3353         .driver = {
3354                 .owner = THIS_MODULE,
3355                 .name  = D40_NAME,
3356                 .pm = DMA40_PM_OPS,
3357         },
3358 };
3359
3360 static int __init stedma40_init(void)
3361 {
3362         return platform_driver_probe(&d40_driver, d40_probe);
3363 }
3364 subsys_initcall(stedma40_init);