2 * Copyright (C) Ericsson AB 2007-2008
3 * Copyright (C) ST-Ericsson SA 2008-2010
6 * License terms: GNU General Public License (GPL) version 2
9 #include <linux/dma-mapping.h>
10 #include <linux/kernel.h>
11 #include <linux/slab.h>
12 #include <linux/export.h>
13 #include <linux/dmaengine.h>
14 #include <linux/platform_device.h>
15 #include <linux/clk.h>
16 #include <linux/delay.h>
17 #include <linux/log2.h>
19 #include <linux/pm_runtime.h>
20 #include <linux/err.h>
22 #include <linux/of_dma.h>
23 #include <linux/amba/bus.h>
24 #include <linux/regulator/consumer.h>
25 #include <linux/platform_data/dma-ste-dma40.h>
27 #include "dmaengine.h"
28 #include "ste_dma40_ll.h"
30 #define D40_NAME "dma40"
32 #define D40_PHY_CHAN -1
34 /* For masking out/in 2 bit channel positions */
35 #define D40_CHAN_POS(chan) (2 * (chan / 2))
36 #define D40_CHAN_POS_MASK(chan) (0x3 << D40_CHAN_POS(chan))
38 /* Maximum iterations taken before giving up suspending a channel */
39 #define D40_SUSPEND_MAX_IT 500
42 #define DMA40_AUTOSUSPEND_DELAY 100
44 /* Hardware requirement on LCLA alignment */
45 #define LCLA_ALIGNMENT 0x40000
47 /* Max number of links per event group */
48 #define D40_LCLA_LINK_PER_EVENT_GRP 128
49 #define D40_LCLA_END D40_LCLA_LINK_PER_EVENT_GRP
51 /* Max number of logical channels per physical channel */
52 #define D40_MAX_LOG_CHAN_PER_PHY 32
54 /* Attempts before giving up to trying to get pages that are aligned */
55 #define MAX_LCLA_ALLOC_ATTEMPTS 256
57 /* Bit markings for allocation map */
58 #define D40_ALLOC_FREE BIT(31)
59 #define D40_ALLOC_PHY BIT(30)
60 #define D40_ALLOC_LOG_FREE 0
62 #define D40_MEMCPY_MAX_CHANS 8
64 /* Reserved event lines for memcpy only. */
65 #define DB8500_DMA_MEMCPY_EV_0 51
66 #define DB8500_DMA_MEMCPY_EV_1 56
67 #define DB8500_DMA_MEMCPY_EV_2 57
68 #define DB8500_DMA_MEMCPY_EV_3 58
69 #define DB8500_DMA_MEMCPY_EV_4 59
70 #define DB8500_DMA_MEMCPY_EV_5 60
72 static int dma40_memcpy_channels[] = {
73 DB8500_DMA_MEMCPY_EV_0,
74 DB8500_DMA_MEMCPY_EV_1,
75 DB8500_DMA_MEMCPY_EV_2,
76 DB8500_DMA_MEMCPY_EV_3,
77 DB8500_DMA_MEMCPY_EV_4,
78 DB8500_DMA_MEMCPY_EV_5,
81 /* Default configuration for physcial memcpy */
82 static const struct stedma40_chan_cfg dma40_memcpy_conf_phy = {
83 .mode = STEDMA40_MODE_PHYSICAL,
84 .dir = DMA_MEM_TO_MEM,
86 .src_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
87 .src_info.psize = STEDMA40_PSIZE_PHY_1,
88 .src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
90 .dst_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
91 .dst_info.psize = STEDMA40_PSIZE_PHY_1,
92 .dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
95 /* Default configuration for logical memcpy */
96 static const struct stedma40_chan_cfg dma40_memcpy_conf_log = {
97 .mode = STEDMA40_MODE_LOGICAL,
98 .dir = DMA_MEM_TO_MEM,
100 .src_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
101 .src_info.psize = STEDMA40_PSIZE_LOG_1,
102 .src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
104 .dst_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
105 .dst_info.psize = STEDMA40_PSIZE_LOG_1,
106 .dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
110 * enum 40_command - The different commands and/or statuses.
112 * @D40_DMA_STOP: DMA channel command STOP or status STOPPED,
113 * @D40_DMA_RUN: The DMA channel is RUNNING of the command RUN.
114 * @D40_DMA_SUSPEND_REQ: Request the DMA to SUSPEND as soon as possible.
115 * @D40_DMA_SUSPENDED: The DMA channel is SUSPENDED.
120 D40_DMA_SUSPEND_REQ = 2,
121 D40_DMA_SUSPENDED = 3
125 * enum d40_events - The different Event Enables for the event lines.
127 * @D40_DEACTIVATE_EVENTLINE: De-activate Event line, stopping the logical chan.
128 * @D40_ACTIVATE_EVENTLINE: Activate the Event line, to start a logical chan.
129 * @D40_SUSPEND_REQ_EVENTLINE: Requesting for suspending a event line.
130 * @D40_ROUND_EVENTLINE: Status check for event line.
134 D40_DEACTIVATE_EVENTLINE = 0,
135 D40_ACTIVATE_EVENTLINE = 1,
136 D40_SUSPEND_REQ_EVENTLINE = 2,
137 D40_ROUND_EVENTLINE = 3
141 * These are the registers that has to be saved and later restored
142 * when the DMA hw is powered off.
143 * TODO: Add save/restore of D40_DREG_GCC on dma40 v3 or later, if that works.
145 static u32 d40_backup_regs[] = {
154 #define BACKUP_REGS_SZ ARRAY_SIZE(d40_backup_regs)
157 * since 9540 and 8540 has the same HW revision
158 * use v4a for 9540 or ealier
159 * use v4b for 8540 or later
161 * DB8500ed has revision 0
162 * DB8500v1 has revision 2
163 * DB8500v2 has revision 3
164 * AP9540v1 has revision 4
165 * DB8540v1 has revision 4
166 * TODO: Check if all these registers have to be saved/restored on dma40 v4a
168 static u32 d40_backup_regs_v4a[] = {
187 #define BACKUP_REGS_SZ_V4A ARRAY_SIZE(d40_backup_regs_v4a)
189 static u32 d40_backup_regs_v4b[] = {
212 #define BACKUP_REGS_SZ_V4B ARRAY_SIZE(d40_backup_regs_v4b)
214 static u32 d40_backup_regs_chan[] = {
225 #define BACKUP_REGS_SZ_MAX ((BACKUP_REGS_SZ_V4A > BACKUP_REGS_SZ_V4B) ? \
226 BACKUP_REGS_SZ_V4A : BACKUP_REGS_SZ_V4B)
229 * struct d40_interrupt_lookup - lookup table for interrupt handler
231 * @src: Interrupt mask register.
232 * @clr: Interrupt clear register.
233 * @is_error: true if this is an error interrupt.
234 * @offset: start delta in the lookup_log_chans in d40_base. If equals to
235 * D40_PHY_CHAN, the lookup_phy_chans shall be used instead.
237 struct d40_interrupt_lookup {
245 static struct d40_interrupt_lookup il_v4a[] = {
246 {D40_DREG_LCTIS0, D40_DREG_LCICR0, false, 0},
247 {D40_DREG_LCTIS1, D40_DREG_LCICR1, false, 32},
248 {D40_DREG_LCTIS2, D40_DREG_LCICR2, false, 64},
249 {D40_DREG_LCTIS3, D40_DREG_LCICR3, false, 96},
250 {D40_DREG_LCEIS0, D40_DREG_LCICR0, true, 0},
251 {D40_DREG_LCEIS1, D40_DREG_LCICR1, true, 32},
252 {D40_DREG_LCEIS2, D40_DREG_LCICR2, true, 64},
253 {D40_DREG_LCEIS3, D40_DREG_LCICR3, true, 96},
254 {D40_DREG_PCTIS, D40_DREG_PCICR, false, D40_PHY_CHAN},
255 {D40_DREG_PCEIS, D40_DREG_PCICR, true, D40_PHY_CHAN},
258 static struct d40_interrupt_lookup il_v4b[] = {
259 {D40_DREG_CLCTIS1, D40_DREG_CLCICR1, false, 0},
260 {D40_DREG_CLCTIS2, D40_DREG_CLCICR2, false, 32},
261 {D40_DREG_CLCTIS3, D40_DREG_CLCICR3, false, 64},
262 {D40_DREG_CLCTIS4, D40_DREG_CLCICR4, false, 96},
263 {D40_DREG_CLCTIS5, D40_DREG_CLCICR5, false, 128},
264 {D40_DREG_CLCEIS1, D40_DREG_CLCICR1, true, 0},
265 {D40_DREG_CLCEIS2, D40_DREG_CLCICR2, true, 32},
266 {D40_DREG_CLCEIS3, D40_DREG_CLCICR3, true, 64},
267 {D40_DREG_CLCEIS4, D40_DREG_CLCICR4, true, 96},
268 {D40_DREG_CLCEIS5, D40_DREG_CLCICR5, true, 128},
269 {D40_DREG_CPCTIS, D40_DREG_CPCICR, false, D40_PHY_CHAN},
270 {D40_DREG_CPCEIS, D40_DREG_CPCICR, true, D40_PHY_CHAN},
274 * struct d40_reg_val - simple lookup struct
276 * @reg: The register.
277 * @val: The value that belongs to the register in reg.
284 static __initdata struct d40_reg_val dma_init_reg_v4a[] = {
285 /* Clock every part of the DMA block from start */
286 { .reg = D40_DREG_GCC, .val = D40_DREG_GCC_ENABLE_ALL},
288 /* Interrupts on all logical channels */
289 { .reg = D40_DREG_LCMIS0, .val = 0xFFFFFFFF},
290 { .reg = D40_DREG_LCMIS1, .val = 0xFFFFFFFF},
291 { .reg = D40_DREG_LCMIS2, .val = 0xFFFFFFFF},
292 { .reg = D40_DREG_LCMIS3, .val = 0xFFFFFFFF},
293 { .reg = D40_DREG_LCICR0, .val = 0xFFFFFFFF},
294 { .reg = D40_DREG_LCICR1, .val = 0xFFFFFFFF},
295 { .reg = D40_DREG_LCICR2, .val = 0xFFFFFFFF},
296 { .reg = D40_DREG_LCICR3, .val = 0xFFFFFFFF},
297 { .reg = D40_DREG_LCTIS0, .val = 0xFFFFFFFF},
298 { .reg = D40_DREG_LCTIS1, .val = 0xFFFFFFFF},
299 { .reg = D40_DREG_LCTIS2, .val = 0xFFFFFFFF},
300 { .reg = D40_DREG_LCTIS3, .val = 0xFFFFFFFF}
302 static __initdata struct d40_reg_val dma_init_reg_v4b[] = {
303 /* Clock every part of the DMA block from start */
304 { .reg = D40_DREG_GCC, .val = D40_DREG_GCC_ENABLE_ALL},
306 /* Interrupts on all logical channels */
307 { .reg = D40_DREG_CLCMIS1, .val = 0xFFFFFFFF},
308 { .reg = D40_DREG_CLCMIS2, .val = 0xFFFFFFFF},
309 { .reg = D40_DREG_CLCMIS3, .val = 0xFFFFFFFF},
310 { .reg = D40_DREG_CLCMIS4, .val = 0xFFFFFFFF},
311 { .reg = D40_DREG_CLCMIS5, .val = 0xFFFFFFFF},
312 { .reg = D40_DREG_CLCICR1, .val = 0xFFFFFFFF},
313 { .reg = D40_DREG_CLCICR2, .val = 0xFFFFFFFF},
314 { .reg = D40_DREG_CLCICR3, .val = 0xFFFFFFFF},
315 { .reg = D40_DREG_CLCICR4, .val = 0xFFFFFFFF},
316 { .reg = D40_DREG_CLCICR5, .val = 0xFFFFFFFF},
317 { .reg = D40_DREG_CLCTIS1, .val = 0xFFFFFFFF},
318 { .reg = D40_DREG_CLCTIS2, .val = 0xFFFFFFFF},
319 { .reg = D40_DREG_CLCTIS3, .val = 0xFFFFFFFF},
320 { .reg = D40_DREG_CLCTIS4, .val = 0xFFFFFFFF},
321 { .reg = D40_DREG_CLCTIS5, .val = 0xFFFFFFFF}
325 * struct d40_lli_pool - Structure for keeping LLIs in memory
327 * @base: Pointer to memory area when the pre_alloc_lli's are not large
328 * enough, IE bigger than the most common case, 1 dst and 1 src. NULL if
329 * pre_alloc_lli is used.
330 * @dma_addr: DMA address, if mapped
331 * @size: The size in bytes of the memory at base or the size of pre_alloc_lli.
332 * @pre_alloc_lli: Pre allocated area for the most common case of transfers,
333 * one buffer to one buffer.
335 struct d40_lli_pool {
339 /* Space for dst and src, plus an extra for padding */
340 u8 pre_alloc_lli[3 * sizeof(struct d40_phy_lli)];
344 * struct d40_desc - A descriptor is one DMA job.
346 * @lli_phy: LLI settings for physical channel. Both src and dst=
347 * points into the lli_pool, to base if lli_len > 1 or to pre_alloc_lli if
348 * lli_len equals one.
349 * @lli_log: Same as above but for logical channels.
350 * @lli_pool: The pool with two entries pre-allocated.
351 * @lli_len: Number of llis of current descriptor.
352 * @lli_current: Number of transferred llis.
353 * @lcla_alloc: Number of LCLA entries allocated.
354 * @txd: DMA engine struct. Used for among other things for communication
357 * @is_in_client_list: true if the client owns this descriptor.
358 * @cyclic: true if this is a cyclic job
360 * This descriptor is used for both logical and physical transfers.
364 struct d40_phy_lli_bidir lli_phy;
366 struct d40_log_lli_bidir lli_log;
368 struct d40_lli_pool lli_pool;
373 struct dma_async_tx_descriptor txd;
374 struct list_head node;
376 bool is_in_client_list;
381 * struct d40_lcla_pool - LCLA pool settings and data.
383 * @base: The virtual address of LCLA. 18 bit aligned.
384 * @base_unaligned: The orignal kmalloc pointer, if kmalloc is used.
385 * This pointer is only there for clean-up on error.
386 * @pages: The number of pages needed for all physical channels.
387 * Only used later for clean-up on error
388 * @lock: Lock to protect the content in this struct.
389 * @alloc_map: big map over which LCLA entry is own by which job.
391 struct d40_lcla_pool {
394 void *base_unaligned;
397 struct d40_desc **alloc_map;
401 * struct d40_phy_res - struct for handling eventlines mapped to physical
404 * @lock: A lock protection this entity.
405 * @reserved: True if used by secure world or otherwise.
406 * @num: The physical channel number of this entity.
407 * @allocated_src: Bit mapped to show which src event line's are mapped to
408 * this physical channel. Can also be free or physically allocated.
409 * @allocated_dst: Same as for src but is dst.
410 * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as
412 * @use_soft_lli: To mark if the linked lists of channel are managed by SW.
426 * struct d40_chan - Struct that describes a channel.
428 * @lock: A spinlock to protect this struct.
429 * @log_num: The logical number, if any of this channel.
430 * @pending_tx: The number of pending transfers. Used between interrupt handler
432 * @busy: Set to true when transfer is ongoing on this channel.
433 * @phy_chan: Pointer to physical channel which this instance runs on. If this
434 * point is NULL, then the channel is not allocated.
435 * @chan: DMA engine handle.
436 * @tasklet: Tasklet that gets scheduled from interrupt context to complete a
437 * transfer and call client callback.
438 * @client: Cliented owned descriptor list.
439 * @pending_queue: Submitted jobs, to be issued by issue_pending()
440 * @active: Active descriptor.
441 * @done: Completed jobs
442 * @queue: Queued jobs.
443 * @prepare_queue: Prepared jobs.
444 * @dma_cfg: The client configuration of this dma channel.
445 * @configured: whether the dma_cfg configuration is valid
446 * @base: Pointer to the device instance struct.
447 * @src_def_cfg: Default cfg register setting for src.
448 * @dst_def_cfg: Default cfg register setting for dst.
449 * @log_def: Default logical channel settings.
450 * @lcpa: Pointer to dst and src lcpa settings.
451 * @runtime_addr: runtime configured address.
452 * @runtime_direction: runtime configured direction.
454 * This struct can either "be" a logical or a physical channel.
461 struct d40_phy_res *phy_chan;
462 struct dma_chan chan;
463 struct tasklet_struct tasklet;
464 struct list_head client;
465 struct list_head pending_queue;
466 struct list_head active;
467 struct list_head done;
468 struct list_head queue;
469 struct list_head prepare_queue;
470 struct stedma40_chan_cfg dma_cfg;
472 struct d40_base *base;
473 /* Default register configurations */
476 struct d40_def_lcsp log_def;
477 struct d40_log_lli_full *lcpa;
478 /* Runtime reconfiguration */
479 dma_addr_t runtime_addr;
480 enum dma_transfer_direction runtime_direction;
484 * struct d40_gen_dmac - generic values to represent u8500/u8540 DMA
487 * @backup: the pointer to the registers address array for backup
488 * @backup_size: the size of the registers address array for backup
489 * @realtime_en: the realtime enable register
490 * @realtime_clear: the realtime clear register
491 * @high_prio_en: the high priority enable register
492 * @high_prio_clear: the high priority clear register
493 * @interrupt_en: the interrupt enable register
494 * @interrupt_clear: the interrupt clear register
495 * @il: the pointer to struct d40_interrupt_lookup
496 * @il_size: the size of d40_interrupt_lookup array
497 * @init_reg: the pointer to the struct d40_reg_val
498 * @init_reg_size: the size of d40_reg_val array
500 struct d40_gen_dmac {
509 struct d40_interrupt_lookup *il;
511 struct d40_reg_val *init_reg;
516 * struct d40_base - The big global struct, one for each probe'd instance.
518 * @interrupt_lock: Lock used to make sure one interrupt is handle a time.
519 * @execmd_lock: Lock for execute command usage since several channels share
520 * the same physical register.
521 * @dev: The device structure.
522 * @virtbase: The virtual base address of the DMA's register.
523 * @rev: silicon revision detected.
524 * @clk: Pointer to the DMA clock structure.
525 * @phy_start: Physical memory start of the DMA registers.
526 * @phy_size: Size of the DMA register map.
527 * @irq: The IRQ number.
528 * @num_memcpy_chans: The number of channels used for memcpy (mem-to-mem
530 * @num_phy_chans: The number of physical channels. Read from HW. This
531 * is the number of available channels for this driver, not counting "Secure
532 * mode" allocated physical channels.
533 * @num_log_chans: The number of logical channels. Calculated from
535 * @dma_both: dma_device channels that can do both memcpy and slave transfers.
536 * @dma_slave: dma_device channels that can do only do slave transfers.
537 * @dma_memcpy: dma_device channels that can do only do memcpy transfers.
538 * @phy_chans: Room for all possible physical channels in system.
539 * @log_chans: Room for all possible logical channels in system.
540 * @lookup_log_chans: Used to map interrupt number to logical channel. Points
541 * to log_chans entries.
542 * @lookup_phy_chans: Used to map interrupt number to physical channel. Points
543 * to phy_chans entries.
544 * @plat_data: Pointer to provided platform_data which is the driver
546 * @lcpa_regulator: Pointer to hold the regulator for the esram bank for lcla.
547 * @phy_res: Vector containing all physical channels.
548 * @lcla_pool: lcla pool settings and data.
549 * @lcpa_base: The virtual mapped address of LCPA.
550 * @phy_lcpa: The physical address of the LCPA.
551 * @lcpa_size: The size of the LCPA area.
552 * @desc_slab: cache for descriptors.
553 * @reg_val_backup: Here the values of some hardware registers are stored
554 * before the DMA is powered off. They are restored when the power is back on.
555 * @reg_val_backup_v4: Backup of registers that only exits on dma40 v3 and
557 * @reg_val_backup_chan: Backup data for standard channel parameter registers.
558 * @regs_interrupt: Scratch space for registers during interrupt.
559 * @gcc_pwr_off_mask: Mask to maintain the channels that can be turned off.
560 * @gen_dmac: the struct for generic registers values to represent u8500/8540
564 spinlock_t interrupt_lock;
565 spinlock_t execmd_lock;
567 void __iomem *virtbase;
570 phys_addr_t phy_start;
571 resource_size_t phy_size;
573 int num_memcpy_chans;
576 struct device_dma_parameters dma_parms;
577 struct dma_device dma_both;
578 struct dma_device dma_slave;
579 struct dma_device dma_memcpy;
580 struct d40_chan *phy_chans;
581 struct d40_chan *log_chans;
582 struct d40_chan **lookup_log_chans;
583 struct d40_chan **lookup_phy_chans;
584 struct stedma40_platform_data *plat_data;
585 struct regulator *lcpa_regulator;
586 /* Physical half channels */
587 struct d40_phy_res *phy_res;
588 struct d40_lcla_pool lcla_pool;
591 resource_size_t lcpa_size;
592 struct kmem_cache *desc_slab;
593 u32 reg_val_backup[BACKUP_REGS_SZ];
594 u32 reg_val_backup_v4[BACKUP_REGS_SZ_MAX];
595 u32 *reg_val_backup_chan;
597 u16 gcc_pwr_off_mask;
598 struct d40_gen_dmac gen_dmac;
601 static struct device *chan2dev(struct d40_chan *d40c)
603 return &d40c->chan.dev->device;
606 static bool chan_is_physical(struct d40_chan *chan)
608 return chan->log_num == D40_PHY_CHAN;
611 static bool chan_is_logical(struct d40_chan *chan)
613 return !chan_is_physical(chan);
616 static void __iomem *chan_base(struct d40_chan *chan)
618 return chan->base->virtbase + D40_DREG_PCBASE +
619 chan->phy_chan->num * D40_DREG_PCDELTA;
622 #define d40_err(dev, format, arg...) \
623 dev_err(dev, "[%s] " format, __func__, ## arg)
625 #define chan_err(d40c, format, arg...) \
626 d40_err(chan2dev(d40c), format, ## arg)
628 static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d,
631 bool is_log = chan_is_logical(d40c);
636 align = sizeof(struct d40_log_lli);
638 align = sizeof(struct d40_phy_lli);
641 base = d40d->lli_pool.pre_alloc_lli;
642 d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
643 d40d->lli_pool.base = NULL;
645 d40d->lli_pool.size = lli_len * 2 * align;
647 base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
648 d40d->lli_pool.base = base;
650 if (d40d->lli_pool.base == NULL)
655 d40d->lli_log.src = PTR_ALIGN(base, align);
656 d40d->lli_log.dst = d40d->lli_log.src + lli_len;
658 d40d->lli_pool.dma_addr = 0;
660 d40d->lli_phy.src = PTR_ALIGN(base, align);
661 d40d->lli_phy.dst = d40d->lli_phy.src + lli_len;
663 d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev,
668 if (dma_mapping_error(d40c->base->dev,
669 d40d->lli_pool.dma_addr)) {
670 kfree(d40d->lli_pool.base);
671 d40d->lli_pool.base = NULL;
672 d40d->lli_pool.dma_addr = 0;
680 static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d)
682 if (d40d->lli_pool.dma_addr)
683 dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr,
684 d40d->lli_pool.size, DMA_TO_DEVICE);
686 kfree(d40d->lli_pool.base);
687 d40d->lli_pool.base = NULL;
688 d40d->lli_pool.size = 0;
689 d40d->lli_log.src = NULL;
690 d40d->lli_log.dst = NULL;
691 d40d->lli_phy.src = NULL;
692 d40d->lli_phy.dst = NULL;
695 static int d40_lcla_alloc_one(struct d40_chan *d40c,
696 struct d40_desc *d40d)
702 spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
705 * Allocate both src and dst at the same time, therefore the half
706 * start on 1 since 0 can't be used since zero is used as end marker.
708 for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
709 int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
711 if (!d40c->base->lcla_pool.alloc_map[idx]) {
712 d40c->base->lcla_pool.alloc_map[idx] = d40d;
719 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
724 static int d40_lcla_free_all(struct d40_chan *d40c,
725 struct d40_desc *d40d)
731 if (chan_is_physical(d40c))
734 spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
736 for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
737 int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
739 if (d40c->base->lcla_pool.alloc_map[idx] == d40d) {
740 d40c->base->lcla_pool.alloc_map[idx] = NULL;
742 if (d40d->lcla_alloc == 0) {
749 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
755 static void d40_desc_remove(struct d40_desc *d40d)
757 list_del(&d40d->node);
760 static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
762 struct d40_desc *desc = NULL;
764 if (!list_empty(&d40c->client)) {
768 list_for_each_entry_safe(d, _d, &d40c->client, node) {
769 if (async_tx_test_ack(&d->txd)) {
772 memset(desc, 0, sizeof(*desc));
779 desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT);
782 INIT_LIST_HEAD(&desc->node);
787 static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
790 d40_pool_lli_free(d40c, d40d);
791 d40_lcla_free_all(d40c, d40d);
792 kmem_cache_free(d40c->base->desc_slab, d40d);
795 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
797 list_add_tail(&desc->node, &d40c->active);
800 static void d40_phy_lli_load(struct d40_chan *chan, struct d40_desc *desc)
802 struct d40_phy_lli *lli_dst = desc->lli_phy.dst;
803 struct d40_phy_lli *lli_src = desc->lli_phy.src;
804 void __iomem *base = chan_base(chan);
806 writel(lli_src->reg_cfg, base + D40_CHAN_REG_SSCFG);
807 writel(lli_src->reg_elt, base + D40_CHAN_REG_SSELT);
808 writel(lli_src->reg_ptr, base + D40_CHAN_REG_SSPTR);
809 writel(lli_src->reg_lnk, base + D40_CHAN_REG_SSLNK);
811 writel(lli_dst->reg_cfg, base + D40_CHAN_REG_SDCFG);
812 writel(lli_dst->reg_elt, base + D40_CHAN_REG_SDELT);
813 writel(lli_dst->reg_ptr, base + D40_CHAN_REG_SDPTR);
814 writel(lli_dst->reg_lnk, base + D40_CHAN_REG_SDLNK);
817 static void d40_desc_done(struct d40_chan *d40c, struct d40_desc *desc)
819 list_add_tail(&desc->node, &d40c->done);
822 static void d40_log_lli_to_lcxa(struct d40_chan *chan, struct d40_desc *desc)
824 struct d40_lcla_pool *pool = &chan->base->lcla_pool;
825 struct d40_log_lli_bidir *lli = &desc->lli_log;
826 int lli_current = desc->lli_current;
827 int lli_len = desc->lli_len;
828 bool cyclic = desc->cyclic;
829 int curr_lcla = -EINVAL;
831 bool use_esram_lcla = chan->base->plat_data->use_esram_lcla;
835 * We may have partially running cyclic transfers, in case we did't get
836 * enough LCLA entries.
838 linkback = cyclic && lli_current == 0;
841 * For linkback, we need one LCLA even with only one link, because we
842 * can't link back to the one in LCPA space
844 if (linkback || (lli_len - lli_current > 1)) {
846 * If the channel is expected to use only soft_lli don't
847 * allocate a lcla. This is to avoid a HW issue that exists
848 * in some controller during a peripheral to memory transfer
849 * that uses linked lists.
851 if (!(chan->phy_chan->use_soft_lli &&
852 chan->dma_cfg.dir == DMA_DEV_TO_MEM))
853 curr_lcla = d40_lcla_alloc_one(chan, desc);
855 first_lcla = curr_lcla;
859 * For linkback, we normally load the LCPA in the loop since we need to
860 * link it to the second LCLA and not the first. However, if we
861 * couldn't even get a first LCLA, then we have to run in LCPA and
864 if (!linkback || curr_lcla == -EINVAL) {
865 unsigned int flags = 0;
867 if (curr_lcla == -EINVAL)
868 flags |= LLI_TERM_INT;
870 d40_log_lli_lcpa_write(chan->lcpa,
871 &lli->dst[lli_current],
872 &lli->src[lli_current],
881 for (; lli_current < lli_len; lli_current++) {
882 unsigned int lcla_offset = chan->phy_chan->num * 1024 +
884 struct d40_log_lli *lcla = pool->base + lcla_offset;
885 unsigned int flags = 0;
888 if (lli_current + 1 < lli_len)
889 next_lcla = d40_lcla_alloc_one(chan, desc);
891 next_lcla = linkback ? first_lcla : -EINVAL;
893 if (cyclic || next_lcla == -EINVAL)
894 flags |= LLI_TERM_INT;
896 if (linkback && curr_lcla == first_lcla) {
897 /* First link goes in both LCPA and LCLA */
898 d40_log_lli_lcpa_write(chan->lcpa,
899 &lli->dst[lli_current],
900 &lli->src[lli_current],
905 * One unused LCLA in the cyclic case if the very first
908 d40_log_lli_lcla_write(lcla,
909 &lli->dst[lli_current],
910 &lli->src[lli_current],
914 * Cache maintenance is not needed if lcla is
917 if (!use_esram_lcla) {
918 dma_sync_single_range_for_device(chan->base->dev,
919 pool->dma_addr, lcla_offset,
920 2 * sizeof(struct d40_log_lli),
923 curr_lcla = next_lcla;
925 if (curr_lcla == -EINVAL || curr_lcla == first_lcla) {
931 desc->lli_current = lli_current;
934 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
936 if (chan_is_physical(d40c)) {
937 d40_phy_lli_load(d40c, d40d);
938 d40d->lli_current = d40d->lli_len;
940 d40_log_lli_to_lcxa(d40c, d40d);
943 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
945 return list_first_entry_or_null(&d40c->active, struct d40_desc, node);
948 /* remove desc from current queue and add it to the pending_queue */
949 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
951 d40_desc_remove(desc);
952 desc->is_in_client_list = false;
953 list_add_tail(&desc->node, &d40c->pending_queue);
956 static struct d40_desc *d40_first_pending(struct d40_chan *d40c)
958 return list_first_entry_or_null(&d40c->pending_queue, struct d40_desc,
962 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
964 return list_first_entry_or_null(&d40c->queue, struct d40_desc, node);
967 static struct d40_desc *d40_first_done(struct d40_chan *d40c)
969 return list_first_entry_or_null(&d40c->done, struct d40_desc, node);
972 static int d40_psize_2_burst_size(bool is_log, int psize)
975 if (psize == STEDMA40_PSIZE_LOG_1)
978 if (psize == STEDMA40_PSIZE_PHY_1)
986 * The dma only supports transmitting packages up to
987 * STEDMA40_MAX_SEG_SIZE * data_width, where data_width is stored in Bytes.
989 * Calculate the total number of dma elements required to send the entire sg list.
991 static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
994 u32 max_w = max(data_width1, data_width2);
995 u32 min_w = min(data_width1, data_width2);
996 u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE * min_w, max_w);
998 if (seg_max > STEDMA40_MAX_SEG_SIZE)
1001 if (!IS_ALIGNED(size, max_w))
1004 if (size <= seg_max)
1007 dmalen = size / seg_max;
1008 if (dmalen * seg_max < size)
1014 static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
1015 u32 data_width1, u32 data_width2)
1017 struct scatterlist *sg;
1022 for_each_sg(sgl, sg, sg_len, i) {
1023 ret = d40_size_2_dmalen(sg_dma_len(sg),
1024 data_width1, data_width2);
1032 static int __d40_execute_command_phy(struct d40_chan *d40c,
1033 enum d40_command command)
1037 void __iomem *active_reg;
1039 unsigned long flags;
1042 if (command == D40_DMA_STOP) {
1043 ret = __d40_execute_command_phy(d40c, D40_DMA_SUSPEND_REQ);
1048 spin_lock_irqsave(&d40c->base->execmd_lock, flags);
1050 if (d40c->phy_chan->num % 2 == 0)
1051 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1053 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1055 if (command == D40_DMA_SUSPEND_REQ) {
1056 status = (readl(active_reg) &
1057 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1058 D40_CHAN_POS(d40c->phy_chan->num);
1060 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1064 wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
1065 writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
1068 if (command == D40_DMA_SUSPEND_REQ) {
1070 for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
1071 status = (readl(active_reg) &
1072 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1073 D40_CHAN_POS(d40c->phy_chan->num);
1077 * Reduce the number of bus accesses while
1078 * waiting for the DMA to suspend.
1082 if (status == D40_DMA_STOP ||
1083 status == D40_DMA_SUSPENDED)
1087 if (i == D40_SUSPEND_MAX_IT) {
1089 "unable to suspend the chl %d (log: %d) status %x\n",
1090 d40c->phy_chan->num, d40c->log_num,
1098 spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
1102 static void d40_term_all(struct d40_chan *d40c)
1104 struct d40_desc *d40d;
1105 struct d40_desc *_d;
1107 /* Release completed descriptors */
1108 while ((d40d = d40_first_done(d40c))) {
1109 d40_desc_remove(d40d);
1110 d40_desc_free(d40c, d40d);
1113 /* Release active descriptors */
1114 while ((d40d = d40_first_active_get(d40c))) {
1115 d40_desc_remove(d40d);
1116 d40_desc_free(d40c, d40d);
1119 /* Release queued descriptors waiting for transfer */
1120 while ((d40d = d40_first_queued(d40c))) {
1121 d40_desc_remove(d40d);
1122 d40_desc_free(d40c, d40d);
1125 /* Release pending descriptors */
1126 while ((d40d = d40_first_pending(d40c))) {
1127 d40_desc_remove(d40d);
1128 d40_desc_free(d40c, d40d);
1131 /* Release client owned descriptors */
1132 if (!list_empty(&d40c->client))
1133 list_for_each_entry_safe(d40d, _d, &d40c->client, node) {
1134 d40_desc_remove(d40d);
1135 d40_desc_free(d40c, d40d);
1138 /* Release descriptors in prepare queue */
1139 if (!list_empty(&d40c->prepare_queue))
1140 list_for_each_entry_safe(d40d, _d,
1141 &d40c->prepare_queue, node) {
1142 d40_desc_remove(d40d);
1143 d40_desc_free(d40c, d40d);
1146 d40c->pending_tx = 0;
1149 static void __d40_config_set_event(struct d40_chan *d40c,
1150 enum d40_events event_type, u32 event,
1153 void __iomem *addr = chan_base(d40c) + reg;
1157 switch (event_type) {
1159 case D40_DEACTIVATE_EVENTLINE:
1161 writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
1162 | ~D40_EVENTLINE_MASK(event), addr);
1165 case D40_SUSPEND_REQ_EVENTLINE:
1166 status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1167 D40_EVENTLINE_POS(event);
1169 if (status == D40_DEACTIVATE_EVENTLINE ||
1170 status == D40_SUSPEND_REQ_EVENTLINE)
1173 writel((D40_SUSPEND_REQ_EVENTLINE << D40_EVENTLINE_POS(event))
1174 | ~D40_EVENTLINE_MASK(event), addr);
1176 for (tries = 0 ; tries < D40_SUSPEND_MAX_IT; tries++) {
1178 status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1179 D40_EVENTLINE_POS(event);
1183 * Reduce the number of bus accesses while
1184 * waiting for the DMA to suspend.
1188 if (status == D40_DEACTIVATE_EVENTLINE)
1192 if (tries == D40_SUSPEND_MAX_IT) {
1194 "unable to stop the event_line chl %d (log: %d)"
1195 "status %x\n", d40c->phy_chan->num,
1196 d40c->log_num, status);
1200 case D40_ACTIVATE_EVENTLINE:
1202 * The hardware sometimes doesn't register the enable when src and dst
1203 * event lines are active on the same logical channel. Retry to ensure
1204 * it does. Usually only one retry is sufficient.
1208 writel((D40_ACTIVATE_EVENTLINE <<
1209 D40_EVENTLINE_POS(event)) |
1210 ~D40_EVENTLINE_MASK(event), addr);
1212 if (readl(addr) & D40_EVENTLINE_MASK(event))
1217 dev_dbg(chan2dev(d40c),
1218 "[%s] workaround enable S%cLNK (%d tries)\n",
1219 __func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
1225 case D40_ROUND_EVENTLINE:
1232 static void d40_config_set_event(struct d40_chan *d40c,
1233 enum d40_events event_type)
1235 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1237 /* Enable event line connected to device (or memcpy) */
1238 if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) ||
1239 (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
1240 __d40_config_set_event(d40c, event_type, event,
1241 D40_CHAN_REG_SSLNK);
1243 if (d40c->dma_cfg.dir != DMA_DEV_TO_MEM)
1244 __d40_config_set_event(d40c, event_type, event,
1245 D40_CHAN_REG_SDLNK);
1248 static u32 d40_chan_has_events(struct d40_chan *d40c)
1250 void __iomem *chanbase = chan_base(d40c);
1253 val = readl(chanbase + D40_CHAN_REG_SSLNK);
1254 val |= readl(chanbase + D40_CHAN_REG_SDLNK);
1260 __d40_execute_command_log(struct d40_chan *d40c, enum d40_command command)
1262 unsigned long flags;
1265 void __iomem *active_reg;
1267 if (d40c->phy_chan->num % 2 == 0)
1268 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1270 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1273 spin_lock_irqsave(&d40c->phy_chan->lock, flags);
1277 case D40_DMA_SUSPEND_REQ:
1279 active_status = (readl(active_reg) &
1280 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1281 D40_CHAN_POS(d40c->phy_chan->num);
1283 if (active_status == D40_DMA_RUN)
1284 d40_config_set_event(d40c, D40_SUSPEND_REQ_EVENTLINE);
1286 d40_config_set_event(d40c, D40_DEACTIVATE_EVENTLINE);
1288 if (!d40_chan_has_events(d40c) && (command == D40_DMA_STOP))
1289 ret = __d40_execute_command_phy(d40c, command);
1295 d40_config_set_event(d40c, D40_ACTIVATE_EVENTLINE);
1296 ret = __d40_execute_command_phy(d40c, command);
1299 case D40_DMA_SUSPENDED:
1304 spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
1308 static int d40_channel_execute_command(struct d40_chan *d40c,
1309 enum d40_command command)
1311 if (chan_is_logical(d40c))
1312 return __d40_execute_command_log(d40c, command);
1314 return __d40_execute_command_phy(d40c, command);
1317 static u32 d40_get_prmo(struct d40_chan *d40c)
1319 static const unsigned int phy_map[] = {
1320 [STEDMA40_PCHAN_BASIC_MODE]
1321 = D40_DREG_PRMO_PCHAN_BASIC,
1322 [STEDMA40_PCHAN_MODULO_MODE]
1323 = D40_DREG_PRMO_PCHAN_MODULO,
1324 [STEDMA40_PCHAN_DOUBLE_DST_MODE]
1325 = D40_DREG_PRMO_PCHAN_DOUBLE_DST,
1327 static const unsigned int log_map[] = {
1328 [STEDMA40_LCHAN_SRC_PHY_DST_LOG]
1329 = D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG,
1330 [STEDMA40_LCHAN_SRC_LOG_DST_PHY]
1331 = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY,
1332 [STEDMA40_LCHAN_SRC_LOG_DST_LOG]
1333 = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG,
1336 if (chan_is_physical(d40c))
1337 return phy_map[d40c->dma_cfg.mode_opt];
1339 return log_map[d40c->dma_cfg.mode_opt];
1342 static void d40_config_write(struct d40_chan *d40c)
1347 /* Odd addresses are even addresses + 4 */
1348 addr_base = (d40c->phy_chan->num % 2) * 4;
1349 /* Setup channel mode to logical or physical */
1350 var = ((u32)(chan_is_logical(d40c)) + 1) <<
1351 D40_CHAN_POS(d40c->phy_chan->num);
1352 writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
1354 /* Setup operational mode option register */
1355 var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num);
1357 writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
1359 if (chan_is_logical(d40c)) {
1360 int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS)
1361 & D40_SREG_ELEM_LOG_LIDX_MASK;
1362 void __iomem *chanbase = chan_base(d40c);
1364 /* Set default config for CFG reg */
1365 writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG);
1366 writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG);
1368 /* Set LIDX for lcla */
1369 writel(lidx, chanbase + D40_CHAN_REG_SSELT);
1370 writel(lidx, chanbase + D40_CHAN_REG_SDELT);
1372 /* Clear LNK which will be used by d40_chan_has_events() */
1373 writel(0, chanbase + D40_CHAN_REG_SSLNK);
1374 writel(0, chanbase + D40_CHAN_REG_SDLNK);
1378 static u32 d40_residue(struct d40_chan *d40c)
1382 if (chan_is_logical(d40c))
1383 num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
1384 >> D40_MEM_LCSP2_ECNT_POS;
1386 u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT);
1387 num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK)
1388 >> D40_SREG_ELEM_PHY_ECNT_POS;
1391 return num_elt * d40c->dma_cfg.dst_info.data_width;
1394 static bool d40_tx_is_linked(struct d40_chan *d40c)
1398 if (chan_is_logical(d40c))
1399 is_link = readl(&d40c->lcpa->lcsp3) & D40_MEM_LCSP3_DLOS_MASK;
1401 is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK)
1402 & D40_SREG_LNK_PHYS_LNK_MASK;
1407 static int d40_pause(struct dma_chan *chan)
1409 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
1411 unsigned long flags;
1413 if (d40c->phy_chan == NULL) {
1414 chan_err(d40c, "Channel is not allocated!\n");
1421 spin_lock_irqsave(&d40c->lock, flags);
1422 pm_runtime_get_sync(d40c->base->dev);
1424 res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1426 pm_runtime_mark_last_busy(d40c->base->dev);
1427 pm_runtime_put_autosuspend(d40c->base->dev);
1428 spin_unlock_irqrestore(&d40c->lock, flags);
1432 static int d40_resume(struct dma_chan *chan)
1434 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
1436 unsigned long flags;
1438 if (d40c->phy_chan == NULL) {
1439 chan_err(d40c, "Channel is not allocated!\n");
1446 spin_lock_irqsave(&d40c->lock, flags);
1447 pm_runtime_get_sync(d40c->base->dev);
1449 /* If bytes left to transfer or linked tx resume job */
1450 if (d40_residue(d40c) || d40_tx_is_linked(d40c))
1451 res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1453 pm_runtime_mark_last_busy(d40c->base->dev);
1454 pm_runtime_put_autosuspend(d40c->base->dev);
1455 spin_unlock_irqrestore(&d40c->lock, flags);
1459 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1461 struct d40_chan *d40c = container_of(tx->chan,
1464 struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1465 unsigned long flags;
1466 dma_cookie_t cookie;
1468 spin_lock_irqsave(&d40c->lock, flags);
1469 cookie = dma_cookie_assign(tx);
1470 d40_desc_queue(d40c, d40d);
1471 spin_unlock_irqrestore(&d40c->lock, flags);
1476 static int d40_start(struct d40_chan *d40c)
1478 return d40_channel_execute_command(d40c, D40_DMA_RUN);
1481 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1483 struct d40_desc *d40d;
1486 /* Start queued jobs, if any */
1487 d40d = d40_first_queued(d40c);
1492 pm_runtime_get_sync(d40c->base->dev);
1495 /* Remove from queue */
1496 d40_desc_remove(d40d);
1498 /* Add to active queue */
1499 d40_desc_submit(d40c, d40d);
1501 /* Initiate DMA job */
1502 d40_desc_load(d40c, d40d);
1505 err = d40_start(d40c);
1514 /* called from interrupt context */
1515 static void dma_tc_handle(struct d40_chan *d40c)
1517 struct d40_desc *d40d;
1519 /* Get first active entry from list */
1520 d40d = d40_first_active_get(d40c);
1527 * If this was a paritially loaded list, we need to reloaded
1528 * it, and only when the list is completed. We need to check
1529 * for done because the interrupt will hit for every link, and
1530 * not just the last one.
1532 if (d40d->lli_current < d40d->lli_len
1533 && !d40_tx_is_linked(d40c)
1534 && !d40_residue(d40c)) {
1535 d40_lcla_free_all(d40c, d40d);
1536 d40_desc_load(d40c, d40d);
1537 (void) d40_start(d40c);
1539 if (d40d->lli_current == d40d->lli_len)
1540 d40d->lli_current = 0;
1543 d40_lcla_free_all(d40c, d40d);
1545 if (d40d->lli_current < d40d->lli_len) {
1546 d40_desc_load(d40c, d40d);
1548 (void) d40_start(d40c);
1552 if (d40_queue_start(d40c) == NULL) {
1555 pm_runtime_mark_last_busy(d40c->base->dev);
1556 pm_runtime_put_autosuspend(d40c->base->dev);
1559 d40_desc_remove(d40d);
1560 d40_desc_done(d40c, d40d);
1564 tasklet_schedule(&d40c->tasklet);
1568 static void dma_tasklet(unsigned long data)
1570 struct d40_chan *d40c = (struct d40_chan *) data;
1571 struct d40_desc *d40d;
1572 unsigned long flags;
1573 bool callback_active;
1574 struct dmaengine_desc_callback cb;
1576 spin_lock_irqsave(&d40c->lock, flags);
1578 /* Get first entry from the done list */
1579 d40d = d40_first_done(d40c);
1581 /* Check if we have reached here for cyclic job */
1582 d40d = d40_first_active_get(d40c);
1583 if (d40d == NULL || !d40d->cyclic)
1584 goto check_pending_tx;
1588 dma_cookie_complete(&d40d->txd);
1591 * If terminating a channel pending_tx is set to zero.
1592 * This prevents any finished active jobs to return to the client.
1594 if (d40c->pending_tx == 0) {
1595 spin_unlock_irqrestore(&d40c->lock, flags);
1599 /* Callback to client */
1600 callback_active = !!(d40d->txd.flags & DMA_PREP_INTERRUPT);
1601 dmaengine_desc_get_callback(&d40d->txd, &cb);
1603 if (!d40d->cyclic) {
1604 if (async_tx_test_ack(&d40d->txd)) {
1605 d40_desc_remove(d40d);
1606 d40_desc_free(d40c, d40d);
1607 } else if (!d40d->is_in_client_list) {
1608 d40_desc_remove(d40d);
1609 d40_lcla_free_all(d40c, d40d);
1610 list_add_tail(&d40d->node, &d40c->client);
1611 d40d->is_in_client_list = true;
1617 if (d40c->pending_tx)
1618 tasklet_schedule(&d40c->tasklet);
1620 spin_unlock_irqrestore(&d40c->lock, flags);
1622 if (callback_active)
1623 dmaengine_desc_callback_invoke(&cb, NULL);
1627 /* Rescue manouver if receiving double interrupts */
1628 if (d40c->pending_tx > 0)
1630 spin_unlock_irqrestore(&d40c->lock, flags);
1633 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1639 struct d40_chan *d40c;
1640 unsigned long flags;
1641 struct d40_base *base = data;
1642 u32 *regs = base->regs_interrupt;
1643 struct d40_interrupt_lookup *il = base->gen_dmac.il;
1644 u32 il_size = base->gen_dmac.il_size;
1646 spin_lock_irqsave(&base->interrupt_lock, flags);
1648 /* Read interrupt status of both logical and physical channels */
1649 for (i = 0; i < il_size; i++)
1650 regs[i] = readl(base->virtbase + il[i].src);
1654 chan = find_next_bit((unsigned long *)regs,
1655 BITS_PER_LONG * il_size, chan + 1);
1657 /* No more set bits found? */
1658 if (chan == BITS_PER_LONG * il_size)
1661 row = chan / BITS_PER_LONG;
1662 idx = chan & (BITS_PER_LONG - 1);
1664 if (il[row].offset == D40_PHY_CHAN)
1665 d40c = base->lookup_phy_chans[idx];
1667 d40c = base->lookup_log_chans[il[row].offset + idx];
1671 * No error because this can happen if something else
1672 * in the system is using the channel.
1678 writel(BIT(idx), base->virtbase + il[row].clr);
1680 spin_lock(&d40c->lock);
1682 if (!il[row].is_error)
1683 dma_tc_handle(d40c);
1685 d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1686 chan, il[row].offset, idx);
1688 spin_unlock(&d40c->lock);
1691 spin_unlock_irqrestore(&base->interrupt_lock, flags);
1696 static int d40_validate_conf(struct d40_chan *d40c,
1697 struct stedma40_chan_cfg *conf)
1700 bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1703 chan_err(d40c, "Invalid direction.\n");
1707 if ((is_log && conf->dev_type > d40c->base->num_log_chans) ||
1708 (!is_log && conf->dev_type > d40c->base->num_phy_chans) ||
1709 (conf->dev_type < 0)) {
1710 chan_err(d40c, "Invalid device type (%d)\n", conf->dev_type);
1714 if (conf->dir == DMA_DEV_TO_DEV) {
1716 * DMAC HW supports it. Will be added to this driver,
1717 * in case any dma client requires it.
1719 chan_err(d40c, "periph to periph not supported\n");
1723 if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1724 conf->src_info.data_width !=
1725 d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1726 conf->dst_info.data_width) {
1728 * The DMAC hardware only supports
1729 * src (burst x width) == dst (burst x width)
1732 chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1739 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1740 bool is_src, int log_event_line, bool is_log,
1743 unsigned long flags;
1744 spin_lock_irqsave(&phy->lock, flags);
1746 *first_user = ((phy->allocated_src | phy->allocated_dst)
1750 /* Physical interrupts are masked per physical full channel */
1751 if (phy->allocated_src == D40_ALLOC_FREE &&
1752 phy->allocated_dst == D40_ALLOC_FREE) {
1753 phy->allocated_dst = D40_ALLOC_PHY;
1754 phy->allocated_src = D40_ALLOC_PHY;
1757 goto not_found_unlock;
1760 /* Logical channel */
1762 if (phy->allocated_src == D40_ALLOC_PHY)
1763 goto not_found_unlock;
1765 if (phy->allocated_src == D40_ALLOC_FREE)
1766 phy->allocated_src = D40_ALLOC_LOG_FREE;
1768 if (!(phy->allocated_src & BIT(log_event_line))) {
1769 phy->allocated_src |= BIT(log_event_line);
1772 goto not_found_unlock;
1774 if (phy->allocated_dst == D40_ALLOC_PHY)
1775 goto not_found_unlock;
1777 if (phy->allocated_dst == D40_ALLOC_FREE)
1778 phy->allocated_dst = D40_ALLOC_LOG_FREE;
1780 if (!(phy->allocated_dst & BIT(log_event_line))) {
1781 phy->allocated_dst |= BIT(log_event_line);
1786 spin_unlock_irqrestore(&phy->lock, flags);
1789 spin_unlock_irqrestore(&phy->lock, flags);
1793 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1796 unsigned long flags;
1797 bool is_free = false;
1799 spin_lock_irqsave(&phy->lock, flags);
1800 if (!log_event_line) {
1801 phy->allocated_dst = D40_ALLOC_FREE;
1802 phy->allocated_src = D40_ALLOC_FREE;
1807 /* Logical channel */
1809 phy->allocated_src &= ~BIT(log_event_line);
1810 if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1811 phy->allocated_src = D40_ALLOC_FREE;
1813 phy->allocated_dst &= ~BIT(log_event_line);
1814 if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1815 phy->allocated_dst = D40_ALLOC_FREE;
1818 is_free = ((phy->allocated_src | phy->allocated_dst) ==
1821 spin_unlock_irqrestore(&phy->lock, flags);
1826 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1828 int dev_type = d40c->dma_cfg.dev_type;
1831 struct d40_phy_res *phys;
1837 bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1839 phys = d40c->base->phy_res;
1840 num_phy_chans = d40c->base->num_phy_chans;
1842 if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
1843 log_num = 2 * dev_type;
1845 } else if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
1846 d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1847 /* dst event lines are used for logical memcpy */
1848 log_num = 2 * dev_type + 1;
1853 event_group = D40_TYPE_TO_GROUP(dev_type);
1854 event_line = D40_TYPE_TO_EVENT(dev_type);
1857 if (d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1858 /* Find physical half channel */
1859 if (d40c->dma_cfg.use_fixed_channel) {
1860 i = d40c->dma_cfg.phy_channel;
1861 if (d40_alloc_mask_set(&phys[i], is_src,
1866 for (i = 0; i < num_phy_chans; i++) {
1867 if (d40_alloc_mask_set(&phys[i], is_src,
1874 for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1875 int phy_num = j + event_group * 2;
1876 for (i = phy_num; i < phy_num + 2; i++) {
1877 if (d40_alloc_mask_set(&phys[i],
1887 d40c->phy_chan = &phys[i];
1888 d40c->log_num = D40_PHY_CHAN;
1894 /* Find logical channel */
1895 for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1896 int phy_num = j + event_group * 2;
1898 if (d40c->dma_cfg.use_fixed_channel) {
1899 i = d40c->dma_cfg.phy_channel;
1901 if ((i != phy_num) && (i != phy_num + 1)) {
1902 dev_err(chan2dev(d40c),
1903 "invalid fixed phy channel %d\n", i);
1907 if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1908 is_log, first_phy_user))
1911 dev_err(chan2dev(d40c),
1912 "could not allocate fixed phy channel %d\n", i);
1917 * Spread logical channels across all available physical rather
1918 * than pack every logical channel at the first available phy
1922 for (i = phy_num; i < phy_num + 2; i++) {
1923 if (d40_alloc_mask_set(&phys[i], is_src,
1929 for (i = phy_num + 1; i >= phy_num; i--) {
1930 if (d40_alloc_mask_set(&phys[i], is_src,
1940 d40c->phy_chan = &phys[i];
1941 d40c->log_num = log_num;
1945 d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1947 d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1953 static int d40_config_memcpy(struct d40_chan *d40c)
1955 dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1957 if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1958 d40c->dma_cfg = dma40_memcpy_conf_log;
1959 d40c->dma_cfg.dev_type = dma40_memcpy_channels[d40c->chan.chan_id];
1961 d40_log_cfg(&d40c->dma_cfg,
1962 &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
1964 } else if (dma_has_cap(DMA_MEMCPY, cap) &&
1965 dma_has_cap(DMA_SLAVE, cap)) {
1966 d40c->dma_cfg = dma40_memcpy_conf_phy;
1968 /* Generate interrrupt at end of transfer or relink. */
1969 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_TIM_POS);
1971 /* Generate interrupt on error. */
1972 d40c->src_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
1973 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
1976 chan_err(d40c, "No memcpy\n");
1983 static int d40_free_dma(struct d40_chan *d40c)
1987 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1988 struct d40_phy_res *phy = d40c->phy_chan;
1991 /* Terminate all queued and active transfers */
1995 chan_err(d40c, "phy == null\n");
1999 if (phy->allocated_src == D40_ALLOC_FREE &&
2000 phy->allocated_dst == D40_ALLOC_FREE) {
2001 chan_err(d40c, "channel already free\n");
2005 if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2006 d40c->dma_cfg.dir == DMA_MEM_TO_MEM)
2008 else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2011 chan_err(d40c, "Unknown direction\n");
2015 pm_runtime_get_sync(d40c->base->dev);
2016 res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2018 chan_err(d40c, "stop failed\n");
2019 goto mark_last_busy;
2022 d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2024 if (chan_is_logical(d40c))
2025 d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2027 d40c->base->lookup_phy_chans[phy->num] = NULL;
2030 pm_runtime_mark_last_busy(d40c->base->dev);
2031 pm_runtime_put_autosuspend(d40c->base->dev);
2035 d40c->phy_chan = NULL;
2036 d40c->configured = false;
2038 pm_runtime_mark_last_busy(d40c->base->dev);
2039 pm_runtime_put_autosuspend(d40c->base->dev);
2043 static bool d40_is_paused(struct d40_chan *d40c)
2045 void __iomem *chanbase = chan_base(d40c);
2046 bool is_paused = false;
2047 unsigned long flags;
2048 void __iomem *active_reg;
2050 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2052 spin_lock_irqsave(&d40c->lock, flags);
2054 if (chan_is_physical(d40c)) {
2055 if (d40c->phy_chan->num % 2 == 0)
2056 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2058 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2060 status = (readl(active_reg) &
2061 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2062 D40_CHAN_POS(d40c->phy_chan->num);
2063 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2068 if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2069 d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
2070 status = readl(chanbase + D40_CHAN_REG_SDLNK);
2071 } else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
2072 status = readl(chanbase + D40_CHAN_REG_SSLNK);
2074 chan_err(d40c, "Unknown direction\n");
2078 status = (status & D40_EVENTLINE_MASK(event)) >>
2079 D40_EVENTLINE_POS(event);
2081 if (status != D40_DMA_RUN)
2084 spin_unlock_irqrestore(&d40c->lock, flags);
2089 static u32 stedma40_residue(struct dma_chan *chan)
2091 struct d40_chan *d40c =
2092 container_of(chan, struct d40_chan, chan);
2094 unsigned long flags;
2096 spin_lock_irqsave(&d40c->lock, flags);
2097 bytes_left = d40_residue(d40c);
2098 spin_unlock_irqrestore(&d40c->lock, flags);
2104 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2105 struct scatterlist *sg_src, struct scatterlist *sg_dst,
2106 unsigned int sg_len, dma_addr_t src_dev_addr,
2107 dma_addr_t dst_dev_addr)
2109 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2110 struct stedma40_half_channel_info *src_info = &cfg->src_info;
2111 struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2114 ret = d40_log_sg_to_lli(sg_src, sg_len,
2117 chan->log_def.lcsp1,
2118 src_info->data_width,
2119 dst_info->data_width);
2121 ret = d40_log_sg_to_lli(sg_dst, sg_len,
2124 chan->log_def.lcsp3,
2125 dst_info->data_width,
2126 src_info->data_width);
2128 return ret < 0 ? ret : 0;
2132 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2133 struct scatterlist *sg_src, struct scatterlist *sg_dst,
2134 unsigned int sg_len, dma_addr_t src_dev_addr,
2135 dma_addr_t dst_dev_addr)
2137 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2138 struct stedma40_half_channel_info *src_info = &cfg->src_info;
2139 struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2140 unsigned long flags = 0;
2144 flags |= LLI_CYCLIC | LLI_TERM_INT;
2146 ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2148 virt_to_phys(desc->lli_phy.src),
2150 src_info, dst_info, flags);
2152 ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2154 virt_to_phys(desc->lli_phy.dst),
2156 dst_info, src_info, flags);
2158 dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2159 desc->lli_pool.size, DMA_TO_DEVICE);
2161 return ret < 0 ? ret : 0;
2164 static struct d40_desc *
2165 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2166 unsigned int sg_len, unsigned long dma_flags)
2168 struct stedma40_chan_cfg *cfg;
2169 struct d40_desc *desc;
2172 desc = d40_desc_get(chan);
2176 cfg = &chan->dma_cfg;
2177 desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2178 cfg->dst_info.data_width);
2179 if (desc->lli_len < 0) {
2180 chan_err(chan, "Unaligned size\n");
2184 ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2186 chan_err(chan, "Could not allocate lli\n");
2190 desc->lli_current = 0;
2191 desc->txd.flags = dma_flags;
2192 desc->txd.tx_submit = d40_tx_submit;
2194 dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2198 d40_desc_free(chan, desc);
2202 static struct dma_async_tx_descriptor *
2203 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2204 struct scatterlist *sg_dst, unsigned int sg_len,
2205 enum dma_transfer_direction direction, unsigned long dma_flags)
2207 struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2208 dma_addr_t src_dev_addr;
2209 dma_addr_t dst_dev_addr;
2210 struct d40_desc *desc;
2211 unsigned long flags;
2214 if (!chan->phy_chan) {
2215 chan_err(chan, "Cannot prepare unallocated channel\n");
2219 spin_lock_irqsave(&chan->lock, flags);
2221 desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2225 if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2226 desc->cyclic = true;
2230 if (direction == DMA_DEV_TO_MEM)
2231 src_dev_addr = chan->runtime_addr;
2232 else if (direction == DMA_MEM_TO_DEV)
2233 dst_dev_addr = chan->runtime_addr;
2235 if (chan_is_logical(chan))
2236 ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2237 sg_len, src_dev_addr, dst_dev_addr);
2239 ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2240 sg_len, src_dev_addr, dst_dev_addr);
2243 chan_err(chan, "Failed to prepare %s sg job: %d\n",
2244 chan_is_logical(chan) ? "log" : "phy", ret);
2249 * add descriptor to the prepare queue in order to be able
2250 * to free them later in terminate_all
2252 list_add_tail(&desc->node, &chan->prepare_queue);
2254 spin_unlock_irqrestore(&chan->lock, flags);
2258 d40_desc_free(chan, desc);
2260 spin_unlock_irqrestore(&chan->lock, flags);
2264 bool stedma40_filter(struct dma_chan *chan, void *data)
2266 struct stedma40_chan_cfg *info = data;
2267 struct d40_chan *d40c =
2268 container_of(chan, struct d40_chan, chan);
2272 err = d40_validate_conf(d40c, info);
2274 d40c->dma_cfg = *info;
2276 err = d40_config_memcpy(d40c);
2279 d40c->configured = true;
2283 EXPORT_SYMBOL(stedma40_filter);
2285 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2287 bool realtime = d40c->dma_cfg.realtime;
2288 bool highprio = d40c->dma_cfg.high_priority;
2290 u32 event = D40_TYPE_TO_EVENT(dev_type);
2291 u32 group = D40_TYPE_TO_GROUP(dev_type);
2292 u32 bit = BIT(event);
2294 struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2296 rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2298 * Due to a hardware bug, in some cases a logical channel triggered by
2299 * a high priority destination event line can generate extra packet
2302 * The workaround is to not set the high priority level for the
2303 * destination event lines that trigger logical channels.
2305 if (!src && chan_is_logical(d40c))
2308 prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2310 /* Destination event lines are stored in the upper halfword */
2314 writel(bit, d40c->base->virtbase + prioreg + group * 4);
2315 writel(bit, d40c->base->virtbase + rtreg + group * 4);
2318 static void d40_set_prio_realtime(struct d40_chan *d40c)
2320 if (d40c->base->rev < 3)
2323 if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) ||
2324 (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2325 __d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, true);
2327 if ((d40c->dma_cfg.dir == DMA_MEM_TO_DEV) ||
2328 (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2329 __d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, false);
2332 #define D40_DT_FLAGS_MODE(flags) ((flags >> 0) & 0x1)
2333 #define D40_DT_FLAGS_DIR(flags) ((flags >> 1) & 0x1)
2334 #define D40_DT_FLAGS_BIG_ENDIAN(flags) ((flags >> 2) & 0x1)
2335 #define D40_DT_FLAGS_FIXED_CHAN(flags) ((flags >> 3) & 0x1)
2336 #define D40_DT_FLAGS_HIGH_PRIO(flags) ((flags >> 4) & 0x1)
2338 static struct dma_chan *d40_xlate(struct of_phandle_args *dma_spec,
2339 struct of_dma *ofdma)
2341 struct stedma40_chan_cfg cfg;
2345 memset(&cfg, 0, sizeof(struct stedma40_chan_cfg));
2348 dma_cap_set(DMA_SLAVE, cap);
2350 cfg.dev_type = dma_spec->args[0];
2351 flags = dma_spec->args[2];
2353 switch (D40_DT_FLAGS_MODE(flags)) {
2354 case 0: cfg.mode = STEDMA40_MODE_LOGICAL; break;
2355 case 1: cfg.mode = STEDMA40_MODE_PHYSICAL; break;
2358 switch (D40_DT_FLAGS_DIR(flags)) {
2360 cfg.dir = DMA_MEM_TO_DEV;
2361 cfg.dst_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2364 cfg.dir = DMA_DEV_TO_MEM;
2365 cfg.src_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2369 if (D40_DT_FLAGS_FIXED_CHAN(flags)) {
2370 cfg.phy_channel = dma_spec->args[1];
2371 cfg.use_fixed_channel = true;
2374 if (D40_DT_FLAGS_HIGH_PRIO(flags))
2375 cfg.high_priority = true;
2377 return dma_request_channel(cap, stedma40_filter, &cfg);
2380 /* DMA ENGINE functions */
2381 static int d40_alloc_chan_resources(struct dma_chan *chan)
2384 unsigned long flags;
2385 struct d40_chan *d40c =
2386 container_of(chan, struct d40_chan, chan);
2388 spin_lock_irqsave(&d40c->lock, flags);
2390 dma_cookie_init(chan);
2392 /* If no dma configuration is set use default configuration (memcpy) */
2393 if (!d40c->configured) {
2394 err = d40_config_memcpy(d40c);
2396 chan_err(d40c, "Failed to configure memcpy channel\n");
2397 goto mark_last_busy;
2401 err = d40_allocate_channel(d40c, &is_free_phy);
2403 chan_err(d40c, "Failed to allocate channel\n");
2404 d40c->configured = false;
2405 goto mark_last_busy;
2408 pm_runtime_get_sync(d40c->base->dev);
2410 d40_set_prio_realtime(d40c);
2412 if (chan_is_logical(d40c)) {
2413 if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2414 d40c->lcpa = d40c->base->lcpa_base +
2415 d40c->dma_cfg.dev_type * D40_LCPA_CHAN_SIZE;
2417 d40c->lcpa = d40c->base->lcpa_base +
2418 d40c->dma_cfg.dev_type *
2419 D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2421 /* Unmask the Global Interrupt Mask. */
2422 d40c->src_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2423 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2426 dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2427 chan_is_logical(d40c) ? "logical" : "physical",
2428 d40c->phy_chan->num,
2429 d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2433 * Only write channel configuration to the DMA if the physical
2434 * resource is free. In case of multiple logical channels
2435 * on the same physical resource, only the first write is necessary.
2438 d40_config_write(d40c);
2440 pm_runtime_mark_last_busy(d40c->base->dev);
2441 pm_runtime_put_autosuspend(d40c->base->dev);
2442 spin_unlock_irqrestore(&d40c->lock, flags);
2446 static void d40_free_chan_resources(struct dma_chan *chan)
2448 struct d40_chan *d40c =
2449 container_of(chan, struct d40_chan, chan);
2451 unsigned long flags;
2453 if (d40c->phy_chan == NULL) {
2454 chan_err(d40c, "Cannot free unallocated channel\n");
2458 spin_lock_irqsave(&d40c->lock, flags);
2460 err = d40_free_dma(d40c);
2463 chan_err(d40c, "Failed to free channel\n");
2464 spin_unlock_irqrestore(&d40c->lock, flags);
2467 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2471 unsigned long dma_flags)
2473 struct scatterlist dst_sg;
2474 struct scatterlist src_sg;
2476 sg_init_table(&dst_sg, 1);
2477 sg_init_table(&src_sg, 1);
2479 sg_dma_address(&dst_sg) = dst;
2480 sg_dma_address(&src_sg) = src;
2482 sg_dma_len(&dst_sg) = size;
2483 sg_dma_len(&src_sg) = size;
2485 return d40_prep_sg(chan, &src_sg, &dst_sg, 1,
2486 DMA_MEM_TO_MEM, dma_flags);
2489 static struct dma_async_tx_descriptor *
2490 d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2491 unsigned int sg_len, enum dma_transfer_direction direction,
2492 unsigned long dma_flags, void *context)
2494 if (!is_slave_direction(direction))
2497 return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2500 static struct dma_async_tx_descriptor *
2501 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2502 size_t buf_len, size_t period_len,
2503 enum dma_transfer_direction direction, unsigned long flags)
2505 unsigned int periods = buf_len / period_len;
2506 struct dma_async_tx_descriptor *txd;
2507 struct scatterlist *sg;
2510 sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2514 for (i = 0; i < periods; i++) {
2515 sg_dma_address(&sg[i]) = dma_addr;
2516 sg_dma_len(&sg[i]) = period_len;
2517 dma_addr += period_len;
2520 sg_chain(sg, periods + 1, sg);
2522 txd = d40_prep_sg(chan, sg, sg, periods, direction,
2523 DMA_PREP_INTERRUPT);
2530 static enum dma_status d40_tx_status(struct dma_chan *chan,
2531 dma_cookie_t cookie,
2532 struct dma_tx_state *txstate)
2534 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2535 enum dma_status ret;
2537 if (d40c->phy_chan == NULL) {
2538 chan_err(d40c, "Cannot read status of unallocated channel\n");
2542 ret = dma_cookie_status(chan, cookie, txstate);
2543 if (ret != DMA_COMPLETE && txstate)
2544 dma_set_residue(txstate, stedma40_residue(chan));
2546 if (d40_is_paused(d40c))
2552 static void d40_issue_pending(struct dma_chan *chan)
2554 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2555 unsigned long flags;
2557 if (d40c->phy_chan == NULL) {
2558 chan_err(d40c, "Channel is not allocated!\n");
2562 spin_lock_irqsave(&d40c->lock, flags);
2564 list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2566 /* Busy means that queued jobs are already being processed */
2568 (void) d40_queue_start(d40c);
2570 spin_unlock_irqrestore(&d40c->lock, flags);
2573 static int d40_terminate_all(struct dma_chan *chan)
2575 unsigned long flags;
2576 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2579 if (d40c->phy_chan == NULL) {
2580 chan_err(d40c, "Channel is not allocated!\n");
2584 spin_lock_irqsave(&d40c->lock, flags);
2586 pm_runtime_get_sync(d40c->base->dev);
2587 ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2589 chan_err(d40c, "Failed to stop channel\n");
2592 pm_runtime_mark_last_busy(d40c->base->dev);
2593 pm_runtime_put_autosuspend(d40c->base->dev);
2595 pm_runtime_mark_last_busy(d40c->base->dev);
2596 pm_runtime_put_autosuspend(d40c->base->dev);
2600 spin_unlock_irqrestore(&d40c->lock, flags);
2605 dma40_config_to_halfchannel(struct d40_chan *d40c,
2606 struct stedma40_half_channel_info *info,
2611 if (chan_is_logical(d40c)) {
2613 psize = STEDMA40_PSIZE_LOG_16;
2614 else if (maxburst >= 8)
2615 psize = STEDMA40_PSIZE_LOG_8;
2616 else if (maxburst >= 4)
2617 psize = STEDMA40_PSIZE_LOG_4;
2619 psize = STEDMA40_PSIZE_LOG_1;
2622 psize = STEDMA40_PSIZE_PHY_16;
2623 else if (maxburst >= 8)
2624 psize = STEDMA40_PSIZE_PHY_8;
2625 else if (maxburst >= 4)
2626 psize = STEDMA40_PSIZE_PHY_4;
2628 psize = STEDMA40_PSIZE_PHY_1;
2631 info->psize = psize;
2632 info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2637 /* Runtime reconfiguration extension */
2638 static int d40_set_runtime_config(struct dma_chan *chan,
2639 struct dma_slave_config *config)
2641 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2642 struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2643 enum dma_slave_buswidth src_addr_width, dst_addr_width;
2644 dma_addr_t config_addr;
2645 u32 src_maxburst, dst_maxburst;
2648 if (d40c->phy_chan == NULL) {
2649 chan_err(d40c, "Channel is not allocated!\n");
2653 src_addr_width = config->src_addr_width;
2654 src_maxburst = config->src_maxburst;
2655 dst_addr_width = config->dst_addr_width;
2656 dst_maxburst = config->dst_maxburst;
2658 if (config->direction == DMA_DEV_TO_MEM) {
2659 config_addr = config->src_addr;
2661 if (cfg->dir != DMA_DEV_TO_MEM)
2662 dev_dbg(d40c->base->dev,
2663 "channel was not configured for peripheral "
2664 "to memory transfer (%d) overriding\n",
2666 cfg->dir = DMA_DEV_TO_MEM;
2668 /* Configure the memory side */
2669 if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2670 dst_addr_width = src_addr_width;
2671 if (dst_maxburst == 0)
2672 dst_maxburst = src_maxburst;
2674 } else if (config->direction == DMA_MEM_TO_DEV) {
2675 config_addr = config->dst_addr;
2677 if (cfg->dir != DMA_MEM_TO_DEV)
2678 dev_dbg(d40c->base->dev,
2679 "channel was not configured for memory "
2680 "to peripheral transfer (%d) overriding\n",
2682 cfg->dir = DMA_MEM_TO_DEV;
2684 /* Configure the memory side */
2685 if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2686 src_addr_width = dst_addr_width;
2687 if (src_maxburst == 0)
2688 src_maxburst = dst_maxburst;
2690 dev_err(d40c->base->dev,
2691 "unrecognized channel direction %d\n",
2696 if (config_addr <= 0) {
2697 dev_err(d40c->base->dev, "no address supplied\n");
2701 if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2702 dev_err(d40c->base->dev,
2703 "src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2711 if (src_maxburst > 16) {
2713 dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2714 } else if (dst_maxburst > 16) {
2716 src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2719 /* Only valid widths are; 1, 2, 4 and 8. */
2720 if (src_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2721 src_addr_width > DMA_SLAVE_BUSWIDTH_8_BYTES ||
2722 dst_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2723 dst_addr_width > DMA_SLAVE_BUSWIDTH_8_BYTES ||
2724 !is_power_of_2(src_addr_width) ||
2725 !is_power_of_2(dst_addr_width))
2728 cfg->src_info.data_width = src_addr_width;
2729 cfg->dst_info.data_width = dst_addr_width;
2731 ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2736 ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2741 /* Fill in register values */
2742 if (chan_is_logical(d40c))
2743 d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2745 d40_phy_cfg(cfg, &d40c->src_def_cfg, &d40c->dst_def_cfg);
2747 /* These settings will take precedence later */
2748 d40c->runtime_addr = config_addr;
2749 d40c->runtime_direction = config->direction;
2750 dev_dbg(d40c->base->dev,
2751 "configured channel %s for %s, data width %d/%d, "
2752 "maxburst %d/%d elements, LE, no flow control\n",
2753 dma_chan_name(chan),
2754 (config->direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2755 src_addr_width, dst_addr_width,
2756 src_maxburst, dst_maxburst);
2761 /* Initialization functions */
2763 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2764 struct d40_chan *chans, int offset,
2768 struct d40_chan *d40c;
2770 INIT_LIST_HEAD(&dma->channels);
2772 for (i = offset; i < offset + num_chans; i++) {
2775 d40c->chan.device = dma;
2777 spin_lock_init(&d40c->lock);
2779 d40c->log_num = D40_PHY_CHAN;
2781 INIT_LIST_HEAD(&d40c->done);
2782 INIT_LIST_HEAD(&d40c->active);
2783 INIT_LIST_HEAD(&d40c->queue);
2784 INIT_LIST_HEAD(&d40c->pending_queue);
2785 INIT_LIST_HEAD(&d40c->client);
2786 INIT_LIST_HEAD(&d40c->prepare_queue);
2788 tasklet_init(&d40c->tasklet, dma_tasklet,
2789 (unsigned long) d40c);
2791 list_add_tail(&d40c->chan.device_node,
2796 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2798 if (dma_has_cap(DMA_SLAVE, dev->cap_mask)) {
2799 dev->device_prep_slave_sg = d40_prep_slave_sg;
2800 dev->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
2803 if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2804 dev->device_prep_dma_memcpy = d40_prep_memcpy;
2805 dev->directions = BIT(DMA_MEM_TO_MEM);
2807 * This controller can only access address at even
2808 * 32bit boundaries, i.e. 2^2
2810 dev->copy_align = DMAENGINE_ALIGN_4_BYTES;
2813 if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2814 dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2816 dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2817 dev->device_free_chan_resources = d40_free_chan_resources;
2818 dev->device_issue_pending = d40_issue_pending;
2819 dev->device_tx_status = d40_tx_status;
2820 dev->device_config = d40_set_runtime_config;
2821 dev->device_pause = d40_pause;
2822 dev->device_resume = d40_resume;
2823 dev->device_terminate_all = d40_terminate_all;
2824 dev->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
2825 dev->dev = base->dev;
2828 static int __init d40_dmaengine_init(struct d40_base *base,
2829 int num_reserved_chans)
2833 d40_chan_init(base, &base->dma_slave, base->log_chans,
2834 0, base->num_log_chans);
2836 dma_cap_zero(base->dma_slave.cap_mask);
2837 dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2838 dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2840 d40_ops_init(base, &base->dma_slave);
2842 err = dmaenginem_async_device_register(&base->dma_slave);
2845 d40_err(base->dev, "Failed to register slave channels\n");
2849 d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2850 base->num_log_chans, base->num_memcpy_chans);
2852 dma_cap_zero(base->dma_memcpy.cap_mask);
2853 dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2855 d40_ops_init(base, &base->dma_memcpy);
2857 err = dmaenginem_async_device_register(&base->dma_memcpy);
2861 "Failed to register memcpy only channels\n");
2865 d40_chan_init(base, &base->dma_both, base->phy_chans,
2866 0, num_reserved_chans);
2868 dma_cap_zero(base->dma_both.cap_mask);
2869 dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2870 dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2871 dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2873 d40_ops_init(base, &base->dma_both);
2874 err = dmaenginem_async_device_register(&base->dma_both);
2878 "Failed to register logical and physical capable channels\n");
2886 /* Suspend resume functionality */
2887 #ifdef CONFIG_PM_SLEEP
2888 static int dma40_suspend(struct device *dev)
2890 struct d40_base *base = dev_get_drvdata(dev);
2893 ret = pm_runtime_force_suspend(dev);
2897 if (base->lcpa_regulator)
2898 ret = regulator_disable(base->lcpa_regulator);
2902 static int dma40_resume(struct device *dev)
2904 struct d40_base *base = dev_get_drvdata(dev);
2907 if (base->lcpa_regulator) {
2908 ret = regulator_enable(base->lcpa_regulator);
2913 return pm_runtime_force_resume(dev);
2918 static void dma40_backup(void __iomem *baseaddr, u32 *backup,
2919 u32 *regaddr, int num, bool save)
2923 for (i = 0; i < num; i++) {
2924 void __iomem *addr = baseaddr + regaddr[i];
2927 backup[i] = readl_relaxed(addr);
2929 writel_relaxed(backup[i], addr);
2933 static void d40_save_restore_registers(struct d40_base *base, bool save)
2937 /* Save/Restore channel specific registers */
2938 for (i = 0; i < base->num_phy_chans; i++) {
2942 if (base->phy_res[i].reserved)
2945 addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA;
2946 idx = i * ARRAY_SIZE(d40_backup_regs_chan);
2948 dma40_backup(addr, &base->reg_val_backup_chan[idx],
2949 d40_backup_regs_chan,
2950 ARRAY_SIZE(d40_backup_regs_chan),
2954 /* Save/Restore global registers */
2955 dma40_backup(base->virtbase, base->reg_val_backup,
2956 d40_backup_regs, ARRAY_SIZE(d40_backup_regs),
2959 /* Save/Restore registers only existing on dma40 v3 and later */
2960 if (base->gen_dmac.backup)
2961 dma40_backup(base->virtbase, base->reg_val_backup_v4,
2962 base->gen_dmac.backup,
2963 base->gen_dmac.backup_size,
2967 static int dma40_runtime_suspend(struct device *dev)
2969 struct d40_base *base = dev_get_drvdata(dev);
2971 d40_save_restore_registers(base, true);
2973 /* Don't disable/enable clocks for v1 due to HW bugs */
2975 writel_relaxed(base->gcc_pwr_off_mask,
2976 base->virtbase + D40_DREG_GCC);
2981 static int dma40_runtime_resume(struct device *dev)
2983 struct d40_base *base = dev_get_drvdata(dev);
2985 d40_save_restore_registers(base, false);
2987 writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
2988 base->virtbase + D40_DREG_GCC);
2993 static const struct dev_pm_ops dma40_pm_ops = {
2994 SET_LATE_SYSTEM_SLEEP_PM_OPS(dma40_suspend, dma40_resume)
2995 SET_RUNTIME_PM_OPS(dma40_runtime_suspend,
2996 dma40_runtime_resume,
3000 /* Initialization functions. */
3002 static int __init d40_phy_res_init(struct d40_base *base)
3005 int num_phy_chans_avail = 0;
3007 int odd_even_bit = -2;
3008 int gcc = D40_DREG_GCC_ENA;
3010 val[0] = readl(base->virtbase + D40_DREG_PRSME);
3011 val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3013 for (i = 0; i < base->num_phy_chans; i++) {
3014 base->phy_res[i].num = i;
3015 odd_even_bit += 2 * ((i % 2) == 0);
3016 if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3017 /* Mark security only channels as occupied */
3018 base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3019 base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3020 base->phy_res[i].reserved = true;
3021 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3023 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3028 base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3029 base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3030 base->phy_res[i].reserved = false;
3031 num_phy_chans_avail++;
3033 spin_lock_init(&base->phy_res[i].lock);
3036 /* Mark disabled channels as occupied */
3037 for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3038 int chan = base->plat_data->disabled_channels[i];
3040 base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3041 base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3042 base->phy_res[chan].reserved = true;
3043 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3045 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3047 num_phy_chans_avail--;
3050 /* Mark soft_lli channels */
3051 for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3052 int chan = base->plat_data->soft_lli_chans[i];
3054 base->phy_res[chan].use_soft_lli = true;
3057 dev_info(base->dev, "%d of %d physical DMA channels available\n",
3058 num_phy_chans_avail, base->num_phy_chans);
3060 /* Verify settings extended vs standard */
3061 val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3063 for (i = 0; i < base->num_phy_chans; i++) {
3065 if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3066 (val[0] & 0x3) != 1)
3068 "[%s] INFO: channel %d is misconfigured (%d)\n",
3069 __func__, i, val[0] & 0x3);
3071 val[0] = val[0] >> 2;
3075 * To keep things simple, Enable all clocks initially.
3076 * The clocks will get managed later post channel allocation.
3077 * The clocks for the event lines on which reserved channels exists
3078 * are not managed here.
3080 writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3081 base->gcc_pwr_off_mask = gcc;
3083 return num_phy_chans_avail;
3086 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
3088 struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3090 void __iomem *virtbase;
3091 struct resource *res;
3092 struct d40_base *base;
3095 int num_memcpy_chans;
3096 int clk_ret = -EINVAL;
3102 clk = clk_get(&pdev->dev, NULL);
3104 d40_err(&pdev->dev, "No matching clock found\n");
3105 goto check_prepare_enabled;
3108 clk_ret = clk_prepare_enable(clk);
3110 d40_err(&pdev->dev, "Failed to prepare/enable clock\n");
3111 goto disable_unprepare;
3114 /* Get IO for DMAC base address */
3115 res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
3117 goto disable_unprepare;
3119 if (request_mem_region(res->start, resource_size(res),
3120 D40_NAME " I/O base") == NULL)
3121 goto release_region;
3123 virtbase = ioremap(res->start, resource_size(res));
3125 goto release_region;
3127 /* This is just a regular AMBA PrimeCell ID actually */
3128 for (pid = 0, i = 0; i < 4; i++)
3129 pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
3131 for (cid = 0, i = 0; i < 4; i++)
3132 cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
3135 if (cid != AMBA_CID) {
3136 d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
3139 if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3140 d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
3141 AMBA_MANF_BITS(pid),
3147 * DB8500ed has revision 0
3149 * DB8500v1 has revision 2
3150 * DB8500v2 has revision 3
3151 * AP9540v1 has revision 4
3152 * DB8540v1 has revision 4
3154 rev = AMBA_REV_BITS(pid);
3156 d40_err(&pdev->dev, "hardware revision: %d is not supported", rev);
3160 /* The number of physical channels on this HW */
3161 if (plat_data->num_of_phy_chans)
3162 num_phy_chans = plat_data->num_of_phy_chans;
3164 num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3166 /* The number of channels used for memcpy */
3167 if (plat_data->num_of_memcpy_chans)
3168 num_memcpy_chans = plat_data->num_of_memcpy_chans;
3170 num_memcpy_chans = ARRAY_SIZE(dma40_memcpy_channels);
3172 num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY;
3174 dev_info(&pdev->dev,
3175 "hardware rev: %d @ %pa with %d physical and %d logical channels\n",
3176 rev, &res->start, num_phy_chans, num_log_chans);
3178 base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
3179 (num_phy_chans + num_log_chans + num_memcpy_chans) *
3180 sizeof(struct d40_chan), GFP_KERNEL);
3187 base->num_memcpy_chans = num_memcpy_chans;
3188 base->num_phy_chans = num_phy_chans;
3189 base->num_log_chans = num_log_chans;
3190 base->phy_start = res->start;
3191 base->phy_size = resource_size(res);
3192 base->virtbase = virtbase;
3193 base->plat_data = plat_data;
3194 base->dev = &pdev->dev;
3195 base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3196 base->log_chans = &base->phy_chans[num_phy_chans];
3198 if (base->plat_data->num_of_phy_chans == 14) {
3199 base->gen_dmac.backup = d40_backup_regs_v4b;
3200 base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3201 base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3202 base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3203 base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3204 base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3205 base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3206 base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3207 base->gen_dmac.il = il_v4b;
3208 base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3209 base->gen_dmac.init_reg = dma_init_reg_v4b;
3210 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3212 if (base->rev >= 3) {
3213 base->gen_dmac.backup = d40_backup_regs_v4a;
3214 base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3216 base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3217 base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3218 base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3219 base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3220 base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3221 base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3222 base->gen_dmac.il = il_v4a;
3223 base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3224 base->gen_dmac.init_reg = dma_init_reg_v4a;
3225 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3228 base->phy_res = kcalloc(num_phy_chans,
3229 sizeof(*base->phy_res),
3234 base->lookup_phy_chans = kcalloc(num_phy_chans,
3235 sizeof(*base->lookup_phy_chans),
3237 if (!base->lookup_phy_chans)
3240 base->lookup_log_chans = kcalloc(num_log_chans,
3241 sizeof(*base->lookup_log_chans),
3243 if (!base->lookup_log_chans)
3244 goto free_phy_chans;
3246 base->reg_val_backup_chan = kmalloc_array(base->num_phy_chans,
3247 sizeof(d40_backup_regs_chan),
3249 if (!base->reg_val_backup_chan)
3250 goto free_log_chans;
3252 base->lcla_pool.alloc_map = kcalloc(num_phy_chans
3253 * D40_LCLA_LINK_PER_EVENT_GRP,
3254 sizeof(*base->lcla_pool.alloc_map),
3256 if (!base->lcla_pool.alloc_map)
3257 goto free_backup_chan;
3259 base->regs_interrupt = kmalloc_array(base->gen_dmac.il_size,
3260 sizeof(*base->regs_interrupt),
3262 if (!base->regs_interrupt)
3265 base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3266 0, SLAB_HWCACHE_ALIGN,
3268 if (base->desc_slab == NULL)
3274 kfree(base->regs_interrupt);
3276 kfree(base->lcla_pool.alloc_map);
3278 kfree(base->reg_val_backup_chan);
3280 kfree(base->lookup_log_chans);
3282 kfree(base->lookup_phy_chans);
3284 kfree(base->phy_res);
3290 release_mem_region(res->start, resource_size(res));
3291 check_prepare_enabled:
3294 clk_disable_unprepare(clk);
3300 static void __init d40_hw_init(struct d40_base *base)
3304 u32 prmseo[2] = {0, 0};
3305 u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3308 struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3309 u32 reg_size = base->gen_dmac.init_reg_size;
3311 for (i = 0; i < reg_size; i++)
3312 writel(dma_init_reg[i].val,
3313 base->virtbase + dma_init_reg[i].reg);
3315 /* Configure all our dma channels to default settings */
3316 for (i = 0; i < base->num_phy_chans; i++) {
3318 activeo[i % 2] = activeo[i % 2] << 2;
3320 if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3322 activeo[i % 2] |= 3;
3326 /* Enable interrupt # */
3327 pcmis = (pcmis << 1) | 1;
3329 /* Clear interrupt # */
3330 pcicr = (pcicr << 1) | 1;
3332 /* Set channel to physical mode */
3333 prmseo[i % 2] = prmseo[i % 2] << 2;
3338 writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3339 writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3340 writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3341 writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3343 /* Write which interrupt to enable */
3344 writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3346 /* Write which interrupt to clear */
3347 writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3349 /* These are __initdata and cannot be accessed after init */
3350 base->gen_dmac.init_reg = NULL;
3351 base->gen_dmac.init_reg_size = 0;
3354 static int __init d40_lcla_allocate(struct d40_base *base)
3356 struct d40_lcla_pool *pool = &base->lcla_pool;
3357 unsigned long *page_list;
3362 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3363 * To full fill this hardware requirement without wasting 256 kb
3364 * we allocate pages until we get an aligned one.
3366 page_list = kmalloc_array(MAX_LCLA_ALLOC_ATTEMPTS,
3372 /* Calculating how many pages that are required */
3373 base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3375 for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3376 page_list[i] = __get_free_pages(GFP_KERNEL,
3377 base->lcla_pool.pages);
3378 if (!page_list[i]) {
3380 d40_err(base->dev, "Failed to allocate %d pages.\n",
3381 base->lcla_pool.pages);
3384 for (j = 0; j < i; j++)
3385 free_pages(page_list[j], base->lcla_pool.pages);
3386 goto free_page_list;
3389 if ((virt_to_phys((void *)page_list[i]) &
3390 (LCLA_ALIGNMENT - 1)) == 0)
3394 for (j = 0; j < i; j++)
3395 free_pages(page_list[j], base->lcla_pool.pages);
3397 if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3398 base->lcla_pool.base = (void *)page_list[i];
3401 * After many attempts and no succees with finding the correct
3402 * alignment, try with allocating a big buffer.
3405 "[%s] Failed to get %d pages @ 18 bit align.\n",
3406 __func__, base->lcla_pool.pages);
3407 base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3408 base->num_phy_chans +
3411 if (!base->lcla_pool.base_unaligned) {
3413 goto free_page_list;
3416 base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3420 pool->dma_addr = dma_map_single(base->dev, pool->base,
3421 SZ_1K * base->num_phy_chans,
3423 if (dma_mapping_error(base->dev, pool->dma_addr)) {
3426 goto free_page_list;
3429 writel(virt_to_phys(base->lcla_pool.base),
3430 base->virtbase + D40_DREG_LCLA);
3437 static int __init d40_of_probe(struct platform_device *pdev,
3438 struct device_node *np)
3440 struct stedma40_platform_data *pdata;
3441 int num_phy = 0, num_memcpy = 0, num_disabled = 0;
3444 pdata = devm_kzalloc(&pdev->dev, sizeof(*pdata), GFP_KERNEL);
3448 /* If absent this value will be obtained from h/w. */
3449 of_property_read_u32(np, "dma-channels", &num_phy);
3451 pdata->num_of_phy_chans = num_phy;
3453 list = of_get_property(np, "memcpy-channels", &num_memcpy);
3454 num_memcpy /= sizeof(*list);
3456 if (num_memcpy > D40_MEMCPY_MAX_CHANS || num_memcpy <= 0) {
3458 "Invalid number of memcpy channels specified (%d)\n",
3462 pdata->num_of_memcpy_chans = num_memcpy;
3464 of_property_read_u32_array(np, "memcpy-channels",
3465 dma40_memcpy_channels,
3468 list = of_get_property(np, "disabled-channels", &num_disabled);
3469 num_disabled /= sizeof(*list);
3471 if (num_disabled >= STEDMA40_MAX_PHYS || num_disabled < 0) {
3473 "Invalid number of disabled channels specified (%d)\n",
3478 of_property_read_u32_array(np, "disabled-channels",
3479 pdata->disabled_channels,
3481 pdata->disabled_channels[num_disabled] = -1;
3483 pdev->dev.platform_data = pdata;
3488 static int __init d40_probe(struct platform_device *pdev)
3490 struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3491 struct device_node *np = pdev->dev.of_node;
3493 struct d40_base *base;
3494 struct resource *res;
3495 int num_reserved_chans;
3500 if (d40_of_probe(pdev, np)) {
3502 goto report_failure;
3505 d40_err(&pdev->dev, "No pdata or Device Tree provided\n");
3506 goto report_failure;
3510 base = d40_hw_detect_init(pdev);
3512 goto report_failure;
3514 num_reserved_chans = d40_phy_res_init(base);
3516 platform_set_drvdata(pdev, base);
3518 spin_lock_init(&base->interrupt_lock);
3519 spin_lock_init(&base->execmd_lock);
3521 /* Get IO for logical channel parameter address */
3522 res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
3525 d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
3528 base->lcpa_size = resource_size(res);
3529 base->phy_lcpa = res->start;
3531 if (request_mem_region(res->start, resource_size(res),
3532 D40_NAME " I/O lcpa") == NULL) {
3534 d40_err(&pdev->dev, "Failed to request LCPA region %pR\n", res);
3538 /* We make use of ESRAM memory for this. */
3539 val = readl(base->virtbase + D40_DREG_LCPA);
3540 if (res->start != val && val != 0) {
3541 dev_warn(&pdev->dev,
3542 "[%s] Mismatch LCPA dma 0x%x, def %pa\n",
3543 __func__, val, &res->start);
3545 writel(res->start, base->virtbase + D40_DREG_LCPA);
3547 base->lcpa_base = ioremap(res->start, resource_size(res));
3548 if (!base->lcpa_base) {
3550 d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
3553 /* If lcla has to be located in ESRAM we don't need to allocate */
3554 if (base->plat_data->use_esram_lcla) {
3555 res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3560 "No \"lcla_esram\" memory resource\n");
3563 base->lcla_pool.base = ioremap(res->start,
3564 resource_size(res));
3565 if (!base->lcla_pool.base) {
3567 d40_err(&pdev->dev, "Failed to ioremap LCLA region\n");
3570 writel(res->start, base->virtbase + D40_DREG_LCLA);
3573 ret = d40_lcla_allocate(base);
3575 d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
3580 spin_lock_init(&base->lcla_pool.lock);
3582 base->irq = platform_get_irq(pdev, 0);
3584 ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3586 d40_err(&pdev->dev, "No IRQ defined\n");
3590 if (base->plat_data->use_esram_lcla) {
3592 base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3593 if (IS_ERR(base->lcpa_regulator)) {
3594 d40_err(&pdev->dev, "Failed to get lcpa_regulator\n");
3595 ret = PTR_ERR(base->lcpa_regulator);
3596 base->lcpa_regulator = NULL;
3600 ret = regulator_enable(base->lcpa_regulator);
3603 "Failed to enable lcpa_regulator\n");
3604 regulator_put(base->lcpa_regulator);
3605 base->lcpa_regulator = NULL;
3610 writel_relaxed(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3612 pm_runtime_irq_safe(base->dev);
3613 pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3614 pm_runtime_use_autosuspend(base->dev);
3615 pm_runtime_mark_last_busy(base->dev);
3616 pm_runtime_set_active(base->dev);
3617 pm_runtime_enable(base->dev);
3619 ret = d40_dmaengine_init(base, num_reserved_chans);
3623 base->dev->dma_parms = &base->dma_parms;
3624 ret = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3626 d40_err(&pdev->dev, "Failed to set dma max seg size\n");
3633 ret = of_dma_controller_register(np, d40_xlate, NULL);
3636 "could not register of_dma_controller\n");
3639 dev_info(base->dev, "initialized\n");
3642 kmem_cache_destroy(base->desc_slab);
3644 iounmap(base->virtbase);
3646 if (base->lcla_pool.base && base->plat_data->use_esram_lcla) {
3647 iounmap(base->lcla_pool.base);
3648 base->lcla_pool.base = NULL;
3651 if (base->lcla_pool.dma_addr)
3652 dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3653 SZ_1K * base->num_phy_chans,
3656 if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3657 free_pages((unsigned long)base->lcla_pool.base,
3658 base->lcla_pool.pages);
3660 kfree(base->lcla_pool.base_unaligned);
3663 release_mem_region(base->phy_lcpa,
3665 if (base->phy_start)
3666 release_mem_region(base->phy_start,
3669 clk_disable_unprepare(base->clk);
3673 if (base->lcpa_regulator) {
3674 regulator_disable(base->lcpa_regulator);
3675 regulator_put(base->lcpa_regulator);
3678 kfree(base->lcla_pool.alloc_map);
3679 kfree(base->lookup_log_chans);
3680 kfree(base->lookup_phy_chans);
3681 kfree(base->phy_res);
3684 d40_err(&pdev->dev, "probe failed\n");
3688 static const struct of_device_id d40_match[] = {
3689 { .compatible = "stericsson,dma40", },
3693 static struct platform_driver d40_driver = {
3696 .pm = &dma40_pm_ops,
3697 .of_match_table = d40_match,
3701 static int __init stedma40_init(void)
3703 return platform_driver_probe(&d40_driver, d40_probe);
3705 subsys_initcall(stedma40_init);