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
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 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 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 * @gcc_pwr_off_mask: Mask to maintain the channels that can be turned off.
559 * @gen_dmac: the struct for generic registers values to represent u8500/8540
563 spinlock_t interrupt_lock;
564 spinlock_t execmd_lock;
566 void __iomem *virtbase;
569 phys_addr_t phy_start;
570 resource_size_t phy_size;
572 int num_memcpy_chans;
575 struct device_dma_parameters dma_parms;
576 struct dma_device dma_both;
577 struct dma_device dma_slave;
578 struct dma_device dma_memcpy;
579 struct d40_chan *phy_chans;
580 struct d40_chan *log_chans;
581 struct d40_chan **lookup_log_chans;
582 struct d40_chan **lookup_phy_chans;
583 struct stedma40_platform_data *plat_data;
584 struct regulator *lcpa_regulator;
585 /* Physical half channels */
586 struct d40_phy_res *phy_res;
587 struct d40_lcla_pool lcla_pool;
590 resource_size_t lcpa_size;
591 struct kmem_cache *desc_slab;
592 u32 reg_val_backup[BACKUP_REGS_SZ];
593 u32 reg_val_backup_v4[BACKUP_REGS_SZ_MAX];
594 u32 *reg_val_backup_chan;
595 u16 gcc_pwr_off_mask;
596 struct d40_gen_dmac gen_dmac;
599 static struct device *chan2dev(struct d40_chan *d40c)
601 return &d40c->chan.dev->device;
604 static bool chan_is_physical(struct d40_chan *chan)
606 return chan->log_num == D40_PHY_CHAN;
609 static bool chan_is_logical(struct d40_chan *chan)
611 return !chan_is_physical(chan);
614 static void __iomem *chan_base(struct d40_chan *chan)
616 return chan->base->virtbase + D40_DREG_PCBASE +
617 chan->phy_chan->num * D40_DREG_PCDELTA;
620 #define d40_err(dev, format, arg...) \
621 dev_err(dev, "[%s] " format, __func__, ## arg)
623 #define chan_err(d40c, format, arg...) \
624 d40_err(chan2dev(d40c), format, ## arg)
626 static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d,
629 bool is_log = chan_is_logical(d40c);
634 align = sizeof(struct d40_log_lli);
636 align = sizeof(struct d40_phy_lli);
639 base = d40d->lli_pool.pre_alloc_lli;
640 d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
641 d40d->lli_pool.base = NULL;
643 d40d->lli_pool.size = lli_len * 2 * align;
645 base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
646 d40d->lli_pool.base = base;
648 if (d40d->lli_pool.base == NULL)
653 d40d->lli_log.src = PTR_ALIGN(base, align);
654 d40d->lli_log.dst = d40d->lli_log.src + lli_len;
656 d40d->lli_pool.dma_addr = 0;
658 d40d->lli_phy.src = PTR_ALIGN(base, align);
659 d40d->lli_phy.dst = d40d->lli_phy.src + lli_len;
661 d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev,
666 if (dma_mapping_error(d40c->base->dev,
667 d40d->lli_pool.dma_addr)) {
668 kfree(d40d->lli_pool.base);
669 d40d->lli_pool.base = NULL;
670 d40d->lli_pool.dma_addr = 0;
678 static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d)
680 if (d40d->lli_pool.dma_addr)
681 dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr,
682 d40d->lli_pool.size, DMA_TO_DEVICE);
684 kfree(d40d->lli_pool.base);
685 d40d->lli_pool.base = NULL;
686 d40d->lli_pool.size = 0;
687 d40d->lli_log.src = NULL;
688 d40d->lli_log.dst = NULL;
689 d40d->lli_phy.src = NULL;
690 d40d->lli_phy.dst = NULL;
693 static int d40_lcla_alloc_one(struct d40_chan *d40c,
694 struct d40_desc *d40d)
700 spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
703 * Allocate both src and dst at the same time, therefore the half
704 * start on 1 since 0 can't be used since zero is used as end marker.
706 for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
707 int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
709 if (!d40c->base->lcla_pool.alloc_map[idx]) {
710 d40c->base->lcla_pool.alloc_map[idx] = d40d;
717 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
722 static int d40_lcla_free_all(struct d40_chan *d40c,
723 struct d40_desc *d40d)
729 if (chan_is_physical(d40c))
732 spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
734 for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
735 int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
737 if (d40c->base->lcla_pool.alloc_map[idx] == d40d) {
738 d40c->base->lcla_pool.alloc_map[idx] = NULL;
740 if (d40d->lcla_alloc == 0) {
747 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
753 static void d40_desc_remove(struct d40_desc *d40d)
755 list_del(&d40d->node);
758 static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
760 struct d40_desc *desc = NULL;
762 if (!list_empty(&d40c->client)) {
766 list_for_each_entry_safe(d, _d, &d40c->client, node) {
767 if (async_tx_test_ack(&d->txd)) {
770 memset(desc, 0, sizeof(*desc));
777 desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT);
780 INIT_LIST_HEAD(&desc->node);
785 static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
788 d40_pool_lli_free(d40c, d40d);
789 d40_lcla_free_all(d40c, d40d);
790 kmem_cache_free(d40c->base->desc_slab, d40d);
793 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
795 list_add_tail(&desc->node, &d40c->active);
798 static void d40_phy_lli_load(struct d40_chan *chan, struct d40_desc *desc)
800 struct d40_phy_lli *lli_dst = desc->lli_phy.dst;
801 struct d40_phy_lli *lli_src = desc->lli_phy.src;
802 void __iomem *base = chan_base(chan);
804 writel(lli_src->reg_cfg, base + D40_CHAN_REG_SSCFG);
805 writel(lli_src->reg_elt, base + D40_CHAN_REG_SSELT);
806 writel(lli_src->reg_ptr, base + D40_CHAN_REG_SSPTR);
807 writel(lli_src->reg_lnk, base + D40_CHAN_REG_SSLNK);
809 writel(lli_dst->reg_cfg, base + D40_CHAN_REG_SDCFG);
810 writel(lli_dst->reg_elt, base + D40_CHAN_REG_SDELT);
811 writel(lli_dst->reg_ptr, base + D40_CHAN_REG_SDPTR);
812 writel(lli_dst->reg_lnk, base + D40_CHAN_REG_SDLNK);
815 static void d40_desc_done(struct d40_chan *d40c, struct d40_desc *desc)
817 list_add_tail(&desc->node, &d40c->done);
820 static void d40_log_lli_to_lcxa(struct d40_chan *chan, struct d40_desc *desc)
822 struct d40_lcla_pool *pool = &chan->base->lcla_pool;
823 struct d40_log_lli_bidir *lli = &desc->lli_log;
824 int lli_current = desc->lli_current;
825 int lli_len = desc->lli_len;
826 bool cyclic = desc->cyclic;
827 int curr_lcla = -EINVAL;
829 bool use_esram_lcla = chan->base->plat_data->use_esram_lcla;
833 * We may have partially running cyclic transfers, in case we did't get
834 * enough LCLA entries.
836 linkback = cyclic && lli_current == 0;
839 * For linkback, we need one LCLA even with only one link, because we
840 * can't link back to the one in LCPA space
842 if (linkback || (lli_len - lli_current > 1)) {
844 * If the channel is expected to use only soft_lli don't
845 * allocate a lcla. This is to avoid a HW issue that exists
846 * in some controller during a peripheral to memory transfer
847 * that uses linked lists.
849 if (!(chan->phy_chan->use_soft_lli &&
850 chan->dma_cfg.dir == DMA_DEV_TO_MEM))
851 curr_lcla = d40_lcla_alloc_one(chan, desc);
853 first_lcla = curr_lcla;
857 * For linkback, we normally load the LCPA in the loop since we need to
858 * link it to the second LCLA and not the first. However, if we
859 * couldn't even get a first LCLA, then we have to run in LCPA and
862 if (!linkback || curr_lcla == -EINVAL) {
863 unsigned int flags = 0;
865 if (curr_lcla == -EINVAL)
866 flags |= LLI_TERM_INT;
868 d40_log_lli_lcpa_write(chan->lcpa,
869 &lli->dst[lli_current],
870 &lli->src[lli_current],
879 for (; lli_current < lli_len; lli_current++) {
880 unsigned int lcla_offset = chan->phy_chan->num * 1024 +
882 struct d40_log_lli *lcla = pool->base + lcla_offset;
883 unsigned int flags = 0;
886 if (lli_current + 1 < lli_len)
887 next_lcla = d40_lcla_alloc_one(chan, desc);
889 next_lcla = linkback ? first_lcla : -EINVAL;
891 if (cyclic || next_lcla == -EINVAL)
892 flags |= LLI_TERM_INT;
894 if (linkback && curr_lcla == first_lcla) {
895 /* First link goes in both LCPA and LCLA */
896 d40_log_lli_lcpa_write(chan->lcpa,
897 &lli->dst[lli_current],
898 &lli->src[lli_current],
903 * One unused LCLA in the cyclic case if the very first
906 d40_log_lli_lcla_write(lcla,
907 &lli->dst[lli_current],
908 &lli->src[lli_current],
912 * Cache maintenance is not needed if lcla is
915 if (!use_esram_lcla) {
916 dma_sync_single_range_for_device(chan->base->dev,
917 pool->dma_addr, lcla_offset,
918 2 * sizeof(struct d40_log_lli),
921 curr_lcla = next_lcla;
923 if (curr_lcla == -EINVAL || curr_lcla == first_lcla) {
930 desc->lli_current = lli_current;
933 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
935 if (chan_is_physical(d40c)) {
936 d40_phy_lli_load(d40c, d40d);
937 d40d->lli_current = d40d->lli_len;
939 d40_log_lli_to_lcxa(d40c, d40d);
942 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
944 return list_first_entry_or_null(&d40c->active, struct d40_desc, node);
947 /* remove desc from current queue and add it to the pending_queue */
948 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
950 d40_desc_remove(desc);
951 desc->is_in_client_list = false;
952 list_add_tail(&desc->node, &d40c->pending_queue);
955 static struct d40_desc *d40_first_pending(struct d40_chan *d40c)
957 return list_first_entry_or_null(&d40c->pending_queue, struct d40_desc,
961 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
963 return list_first_entry_or_null(&d40c->queue, struct d40_desc, node);
966 static struct d40_desc *d40_first_done(struct d40_chan *d40c)
968 return list_first_entry_or_null(&d40c->done, struct d40_desc, node);
971 static int d40_psize_2_burst_size(bool is_log, int psize)
974 if (psize == STEDMA40_PSIZE_LOG_1)
977 if (psize == STEDMA40_PSIZE_PHY_1)
985 * The dma only supports transmitting packages up to
986 * STEDMA40_MAX_SEG_SIZE * data_width, where data_width is stored in Bytes.
988 * Calculate the total number of dma elements required to send the entire sg list.
990 static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
993 u32 max_w = max(data_width1, data_width2);
994 u32 min_w = min(data_width1, data_width2);
995 u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE * min_w, max_w);
997 if (seg_max > STEDMA40_MAX_SEG_SIZE)
1000 if (!IS_ALIGNED(size, max_w))
1003 if (size <= seg_max)
1006 dmalen = size / seg_max;
1007 if (dmalen * seg_max < size)
1013 static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
1014 u32 data_width1, u32 data_width2)
1016 struct scatterlist *sg;
1021 for_each_sg(sgl, sg, sg_len, i) {
1022 ret = d40_size_2_dmalen(sg_dma_len(sg),
1023 data_width1, data_width2);
1031 static int __d40_execute_command_phy(struct d40_chan *d40c,
1032 enum d40_command command)
1036 void __iomem *active_reg;
1038 unsigned long flags;
1041 if (command == D40_DMA_STOP) {
1042 ret = __d40_execute_command_phy(d40c, D40_DMA_SUSPEND_REQ);
1047 spin_lock_irqsave(&d40c->base->execmd_lock, flags);
1049 if (d40c->phy_chan->num % 2 == 0)
1050 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1052 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1054 if (command == D40_DMA_SUSPEND_REQ) {
1055 status = (readl(active_reg) &
1056 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1057 D40_CHAN_POS(d40c->phy_chan->num);
1059 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1063 wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
1064 writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
1067 if (command == D40_DMA_SUSPEND_REQ) {
1069 for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
1070 status = (readl(active_reg) &
1071 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1072 D40_CHAN_POS(d40c->phy_chan->num);
1076 * Reduce the number of bus accesses while
1077 * waiting for the DMA to suspend.
1081 if (status == D40_DMA_STOP ||
1082 status == D40_DMA_SUSPENDED)
1086 if (i == D40_SUSPEND_MAX_IT) {
1088 "unable to suspend the chl %d (log: %d) status %x\n",
1089 d40c->phy_chan->num, d40c->log_num,
1097 spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
1101 static void d40_term_all(struct d40_chan *d40c)
1103 struct d40_desc *d40d;
1104 struct d40_desc *_d;
1106 /* Release completed descriptors */
1107 while ((d40d = d40_first_done(d40c))) {
1108 d40_desc_remove(d40d);
1109 d40_desc_free(d40c, d40d);
1112 /* Release active descriptors */
1113 while ((d40d = d40_first_active_get(d40c))) {
1114 d40_desc_remove(d40d);
1115 d40_desc_free(d40c, d40d);
1118 /* Release queued descriptors waiting for transfer */
1119 while ((d40d = d40_first_queued(d40c))) {
1120 d40_desc_remove(d40d);
1121 d40_desc_free(d40c, d40d);
1124 /* Release pending descriptors */
1125 while ((d40d = d40_first_pending(d40c))) {
1126 d40_desc_remove(d40d);
1127 d40_desc_free(d40c, d40d);
1130 /* Release client owned descriptors */
1131 if (!list_empty(&d40c->client))
1132 list_for_each_entry_safe(d40d, _d, &d40c->client, node) {
1133 d40_desc_remove(d40d);
1134 d40_desc_free(d40c, d40d);
1137 /* Release descriptors in prepare queue */
1138 if (!list_empty(&d40c->prepare_queue))
1139 list_for_each_entry_safe(d40d, _d,
1140 &d40c->prepare_queue, node) {
1141 d40_desc_remove(d40d);
1142 d40_desc_free(d40c, d40d);
1145 d40c->pending_tx = 0;
1148 static void __d40_config_set_event(struct d40_chan *d40c,
1149 enum d40_events event_type, u32 event,
1152 void __iomem *addr = chan_base(d40c) + reg;
1156 switch (event_type) {
1158 case D40_DEACTIVATE_EVENTLINE:
1160 writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
1161 | ~D40_EVENTLINE_MASK(event), addr);
1164 case D40_SUSPEND_REQ_EVENTLINE:
1165 status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1166 D40_EVENTLINE_POS(event);
1168 if (status == D40_DEACTIVATE_EVENTLINE ||
1169 status == D40_SUSPEND_REQ_EVENTLINE)
1172 writel((D40_SUSPEND_REQ_EVENTLINE << D40_EVENTLINE_POS(event))
1173 | ~D40_EVENTLINE_MASK(event), addr);
1175 for (tries = 0 ; tries < D40_SUSPEND_MAX_IT; tries++) {
1177 status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1178 D40_EVENTLINE_POS(event);
1182 * Reduce the number of bus accesses while
1183 * waiting for the DMA to suspend.
1187 if (status == D40_DEACTIVATE_EVENTLINE)
1191 if (tries == D40_SUSPEND_MAX_IT) {
1193 "unable to stop the event_line chl %d (log: %d)"
1194 "status %x\n", d40c->phy_chan->num,
1195 d40c->log_num, status);
1199 case D40_ACTIVATE_EVENTLINE:
1201 * The hardware sometimes doesn't register the enable when src and dst
1202 * event lines are active on the same logical channel. Retry to ensure
1203 * it does. Usually only one retry is sufficient.
1207 writel((D40_ACTIVATE_EVENTLINE <<
1208 D40_EVENTLINE_POS(event)) |
1209 ~D40_EVENTLINE_MASK(event), addr);
1211 if (readl(addr) & D40_EVENTLINE_MASK(event))
1216 dev_dbg(chan2dev(d40c),
1217 "[%s] workaround enable S%cLNK (%d tries)\n",
1218 __func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
1224 case D40_ROUND_EVENTLINE:
1231 static void d40_config_set_event(struct d40_chan *d40c,
1232 enum d40_events event_type)
1234 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1236 /* Enable event line connected to device (or memcpy) */
1237 if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) ||
1238 (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
1239 __d40_config_set_event(d40c, event_type, event,
1240 D40_CHAN_REG_SSLNK);
1242 if (d40c->dma_cfg.dir != DMA_DEV_TO_MEM)
1243 __d40_config_set_event(d40c, event_type, event,
1244 D40_CHAN_REG_SDLNK);
1247 static u32 d40_chan_has_events(struct d40_chan *d40c)
1249 void __iomem *chanbase = chan_base(d40c);
1252 val = readl(chanbase + D40_CHAN_REG_SSLNK);
1253 val |= readl(chanbase + D40_CHAN_REG_SDLNK);
1259 __d40_execute_command_log(struct d40_chan *d40c, enum d40_command command)
1261 unsigned long flags;
1264 void __iomem *active_reg;
1266 if (d40c->phy_chan->num % 2 == 0)
1267 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1269 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1272 spin_lock_irqsave(&d40c->phy_chan->lock, flags);
1276 case D40_DMA_SUSPEND_REQ:
1278 active_status = (readl(active_reg) &
1279 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1280 D40_CHAN_POS(d40c->phy_chan->num);
1282 if (active_status == D40_DMA_RUN)
1283 d40_config_set_event(d40c, D40_SUSPEND_REQ_EVENTLINE);
1285 d40_config_set_event(d40c, D40_DEACTIVATE_EVENTLINE);
1287 if (!d40_chan_has_events(d40c) && (command == D40_DMA_STOP))
1288 ret = __d40_execute_command_phy(d40c, command);
1294 d40_config_set_event(d40c, D40_ACTIVATE_EVENTLINE);
1295 ret = __d40_execute_command_phy(d40c, command);
1298 case D40_DMA_SUSPENDED:
1303 spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
1307 static int d40_channel_execute_command(struct d40_chan *d40c,
1308 enum d40_command command)
1310 if (chan_is_logical(d40c))
1311 return __d40_execute_command_log(d40c, command);
1313 return __d40_execute_command_phy(d40c, command);
1316 static u32 d40_get_prmo(struct d40_chan *d40c)
1318 static const unsigned int phy_map[] = {
1319 [STEDMA40_PCHAN_BASIC_MODE]
1320 = D40_DREG_PRMO_PCHAN_BASIC,
1321 [STEDMA40_PCHAN_MODULO_MODE]
1322 = D40_DREG_PRMO_PCHAN_MODULO,
1323 [STEDMA40_PCHAN_DOUBLE_DST_MODE]
1324 = D40_DREG_PRMO_PCHAN_DOUBLE_DST,
1326 static const unsigned int log_map[] = {
1327 [STEDMA40_LCHAN_SRC_PHY_DST_LOG]
1328 = D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG,
1329 [STEDMA40_LCHAN_SRC_LOG_DST_PHY]
1330 = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY,
1331 [STEDMA40_LCHAN_SRC_LOG_DST_LOG]
1332 = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG,
1335 if (chan_is_physical(d40c))
1336 return phy_map[d40c->dma_cfg.mode_opt];
1338 return log_map[d40c->dma_cfg.mode_opt];
1341 static void d40_config_write(struct d40_chan *d40c)
1346 /* Odd addresses are even addresses + 4 */
1347 addr_base = (d40c->phy_chan->num % 2) * 4;
1348 /* Setup channel mode to logical or physical */
1349 var = ((u32)(chan_is_logical(d40c)) + 1) <<
1350 D40_CHAN_POS(d40c->phy_chan->num);
1351 writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
1353 /* Setup operational mode option register */
1354 var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num);
1356 writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
1358 if (chan_is_logical(d40c)) {
1359 int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS)
1360 & D40_SREG_ELEM_LOG_LIDX_MASK;
1361 void __iomem *chanbase = chan_base(d40c);
1363 /* Set default config for CFG reg */
1364 writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG);
1365 writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG);
1367 /* Set LIDX for lcla */
1368 writel(lidx, chanbase + D40_CHAN_REG_SSELT);
1369 writel(lidx, chanbase + D40_CHAN_REG_SDELT);
1371 /* Clear LNK which will be used by d40_chan_has_events() */
1372 writel(0, chanbase + D40_CHAN_REG_SSLNK);
1373 writel(0, chanbase + D40_CHAN_REG_SDLNK);
1377 static u32 d40_residue(struct d40_chan *d40c)
1381 if (chan_is_logical(d40c))
1382 num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
1383 >> D40_MEM_LCSP2_ECNT_POS;
1385 u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT);
1386 num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK)
1387 >> D40_SREG_ELEM_PHY_ECNT_POS;
1390 return num_elt * d40c->dma_cfg.dst_info.data_width;
1393 static bool d40_tx_is_linked(struct d40_chan *d40c)
1397 if (chan_is_logical(d40c))
1398 is_link = readl(&d40c->lcpa->lcsp3) & D40_MEM_LCSP3_DLOS_MASK;
1400 is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK)
1401 & D40_SREG_LNK_PHYS_LNK_MASK;
1406 static int d40_pause(struct dma_chan *chan)
1408 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
1410 unsigned long flags;
1412 if (d40c->phy_chan == NULL) {
1413 chan_err(d40c, "Channel is not allocated!\n");
1420 spin_lock_irqsave(&d40c->lock, flags);
1421 pm_runtime_get_sync(d40c->base->dev);
1423 res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1425 pm_runtime_mark_last_busy(d40c->base->dev);
1426 pm_runtime_put_autosuspend(d40c->base->dev);
1427 spin_unlock_irqrestore(&d40c->lock, flags);
1431 static int d40_resume(struct dma_chan *chan)
1433 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
1435 unsigned long flags;
1437 if (d40c->phy_chan == NULL) {
1438 chan_err(d40c, "Channel is not allocated!\n");
1445 spin_lock_irqsave(&d40c->lock, flags);
1446 pm_runtime_get_sync(d40c->base->dev);
1448 /* If bytes left to transfer or linked tx resume job */
1449 if (d40_residue(d40c) || d40_tx_is_linked(d40c))
1450 res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1452 pm_runtime_mark_last_busy(d40c->base->dev);
1453 pm_runtime_put_autosuspend(d40c->base->dev);
1454 spin_unlock_irqrestore(&d40c->lock, flags);
1458 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1460 struct d40_chan *d40c = container_of(tx->chan,
1463 struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1464 unsigned long flags;
1465 dma_cookie_t cookie;
1467 spin_lock_irqsave(&d40c->lock, flags);
1468 cookie = dma_cookie_assign(tx);
1469 d40_desc_queue(d40c, d40d);
1470 spin_unlock_irqrestore(&d40c->lock, flags);
1475 static int d40_start(struct d40_chan *d40c)
1477 return d40_channel_execute_command(d40c, D40_DMA_RUN);
1480 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1482 struct d40_desc *d40d;
1485 /* Start queued jobs, if any */
1486 d40d = d40_first_queued(d40c);
1491 pm_runtime_get_sync(d40c->base->dev);
1494 /* Remove from queue */
1495 d40_desc_remove(d40d);
1497 /* Add to active queue */
1498 d40_desc_submit(d40c, d40d);
1500 /* Initiate DMA job */
1501 d40_desc_load(d40c, d40d);
1504 err = d40_start(d40c);
1513 /* called from interrupt context */
1514 static void dma_tc_handle(struct d40_chan *d40c)
1516 struct d40_desc *d40d;
1518 /* Get first active entry from list */
1519 d40d = d40_first_active_get(d40c);
1526 * If this was a paritially loaded list, we need to reloaded
1527 * it, and only when the list is completed. We need to check
1528 * for done because the interrupt will hit for every link, and
1529 * not just the last one.
1531 if (d40d->lli_current < d40d->lli_len
1532 && !d40_tx_is_linked(d40c)
1533 && !d40_residue(d40c)) {
1534 d40_lcla_free_all(d40c, d40d);
1535 d40_desc_load(d40c, d40d);
1536 (void) d40_start(d40c);
1538 if (d40d->lli_current == d40d->lli_len)
1539 d40d->lli_current = 0;
1542 d40_lcla_free_all(d40c, d40d);
1544 if (d40d->lli_current < d40d->lli_len) {
1545 d40_desc_load(d40c, d40d);
1547 (void) d40_start(d40c);
1551 if (d40_queue_start(d40c) == NULL) {
1554 pm_runtime_mark_last_busy(d40c->base->dev);
1555 pm_runtime_put_autosuspend(d40c->base->dev);
1558 d40_desc_remove(d40d);
1559 d40_desc_done(d40c, d40d);
1563 tasklet_schedule(&d40c->tasklet);
1567 static void dma_tasklet(unsigned long data)
1569 struct d40_chan *d40c = (struct d40_chan *) data;
1570 struct d40_desc *d40d;
1571 unsigned long flags;
1572 bool callback_active;
1573 struct dmaengine_desc_callback cb;
1575 spin_lock_irqsave(&d40c->lock, flags);
1577 /* Get first entry from the done list */
1578 d40d = d40_first_done(d40c);
1580 /* Check if we have reached here for cyclic job */
1581 d40d = d40_first_active_get(d40c);
1582 if (d40d == NULL || !d40d->cyclic)
1587 dma_cookie_complete(&d40d->txd);
1590 * If terminating a channel pending_tx is set to zero.
1591 * This prevents any finished active jobs to return to the client.
1593 if (d40c->pending_tx == 0) {
1594 spin_unlock_irqrestore(&d40c->lock, flags);
1598 /* Callback to client */
1599 callback_active = !!(d40d->txd.flags & DMA_PREP_INTERRUPT);
1600 dmaengine_desc_get_callback(&d40d->txd, &cb);
1602 if (!d40d->cyclic) {
1603 if (async_tx_test_ack(&d40d->txd)) {
1604 d40_desc_remove(d40d);
1605 d40_desc_free(d40c, d40d);
1606 } else if (!d40d->is_in_client_list) {
1607 d40_desc_remove(d40d);
1608 d40_lcla_free_all(d40c, d40d);
1609 list_add_tail(&d40d->node, &d40c->client);
1610 d40d->is_in_client_list = true;
1616 if (d40c->pending_tx)
1617 tasklet_schedule(&d40c->tasklet);
1619 spin_unlock_irqrestore(&d40c->lock, flags);
1621 if (callback_active)
1622 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->gen_dmac.il_size];
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;
1760 /* Logical channel */
1762 if (phy->allocated_src == D40_ALLOC_PHY)
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);
1774 if (phy->allocated_dst == D40_ALLOC_PHY)
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);
1788 spin_unlock_irqrestore(&phy->lock, flags);
1791 spin_unlock_irqrestore(&phy->lock, flags);
1795 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1798 unsigned long flags;
1799 bool is_free = false;
1801 spin_lock_irqsave(&phy->lock, flags);
1802 if (!log_event_line) {
1803 phy->allocated_dst = D40_ALLOC_FREE;
1804 phy->allocated_src = D40_ALLOC_FREE;
1809 /* Logical channel */
1811 phy->allocated_src &= ~BIT(log_event_line);
1812 if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1813 phy->allocated_src = D40_ALLOC_FREE;
1815 phy->allocated_dst &= ~BIT(log_event_line);
1816 if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1817 phy->allocated_dst = D40_ALLOC_FREE;
1820 is_free = ((phy->allocated_src | phy->allocated_dst) ==
1824 spin_unlock_irqrestore(&phy->lock, flags);
1829 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1831 int dev_type = d40c->dma_cfg.dev_type;
1834 struct d40_phy_res *phys;
1840 bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1842 phys = d40c->base->phy_res;
1843 num_phy_chans = d40c->base->num_phy_chans;
1845 if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
1846 log_num = 2 * dev_type;
1848 } else if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
1849 d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1850 /* dst event lines are used for logical memcpy */
1851 log_num = 2 * dev_type + 1;
1856 event_group = D40_TYPE_TO_GROUP(dev_type);
1857 event_line = D40_TYPE_TO_EVENT(dev_type);
1860 if (d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1861 /* Find physical half channel */
1862 if (d40c->dma_cfg.use_fixed_channel) {
1863 i = d40c->dma_cfg.phy_channel;
1864 if (d40_alloc_mask_set(&phys[i], is_src,
1869 for (i = 0; i < num_phy_chans; i++) {
1870 if (d40_alloc_mask_set(&phys[i], is_src,
1877 for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1878 int phy_num = j + event_group * 2;
1879 for (i = phy_num; i < phy_num + 2; i++) {
1880 if (d40_alloc_mask_set(&phys[i],
1890 d40c->phy_chan = &phys[i];
1891 d40c->log_num = D40_PHY_CHAN;
1897 /* Find logical channel */
1898 for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1899 int phy_num = j + event_group * 2;
1901 if (d40c->dma_cfg.use_fixed_channel) {
1902 i = d40c->dma_cfg.phy_channel;
1904 if ((i != phy_num) && (i != phy_num + 1)) {
1905 dev_err(chan2dev(d40c),
1906 "invalid fixed phy channel %d\n", i);
1910 if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1911 is_log, first_phy_user))
1914 dev_err(chan2dev(d40c),
1915 "could not allocate fixed phy channel %d\n", i);
1920 * Spread logical channels across all available physical rather
1921 * than pack every logical channel at the first available phy
1925 for (i = phy_num; i < phy_num + 2; i++) {
1926 if (d40_alloc_mask_set(&phys[i], is_src,
1932 for (i = phy_num + 1; i >= phy_num; i--) {
1933 if (d40_alloc_mask_set(&phys[i], is_src,
1943 d40c->phy_chan = &phys[i];
1944 d40c->log_num = log_num;
1948 d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1950 d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1956 static int d40_config_memcpy(struct d40_chan *d40c)
1958 dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1960 if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1961 d40c->dma_cfg = dma40_memcpy_conf_log;
1962 d40c->dma_cfg.dev_type = dma40_memcpy_channels[d40c->chan.chan_id];
1964 d40_log_cfg(&d40c->dma_cfg,
1965 &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
1967 } else if (dma_has_cap(DMA_MEMCPY, cap) &&
1968 dma_has_cap(DMA_SLAVE, cap)) {
1969 d40c->dma_cfg = dma40_memcpy_conf_phy;
1971 /* Generate interrrupt at end of transfer or relink. */
1972 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_TIM_POS);
1974 /* Generate interrupt on error. */
1975 d40c->src_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
1976 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
1979 chan_err(d40c, "No memcpy\n");
1986 static int d40_free_dma(struct d40_chan *d40c)
1990 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1991 struct d40_phy_res *phy = d40c->phy_chan;
1994 /* Terminate all queued and active transfers */
1998 chan_err(d40c, "phy == null\n");
2002 if (phy->allocated_src == D40_ALLOC_FREE &&
2003 phy->allocated_dst == D40_ALLOC_FREE) {
2004 chan_err(d40c, "channel already free\n");
2008 if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2009 d40c->dma_cfg.dir == DMA_MEM_TO_MEM)
2011 else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2014 chan_err(d40c, "Unknown direction\n");
2018 pm_runtime_get_sync(d40c->base->dev);
2019 res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2021 chan_err(d40c, "stop failed\n");
2025 d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2027 if (chan_is_logical(d40c))
2028 d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2030 d40c->base->lookup_phy_chans[phy->num] = NULL;
2033 pm_runtime_mark_last_busy(d40c->base->dev);
2034 pm_runtime_put_autosuspend(d40c->base->dev);
2038 d40c->phy_chan = NULL;
2039 d40c->configured = false;
2042 pm_runtime_mark_last_busy(d40c->base->dev);
2043 pm_runtime_put_autosuspend(d40c->base->dev);
2047 static bool d40_is_paused(struct d40_chan *d40c)
2049 void __iomem *chanbase = chan_base(d40c);
2050 bool is_paused = false;
2051 unsigned long flags;
2052 void __iomem *active_reg;
2054 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2056 spin_lock_irqsave(&d40c->lock, flags);
2058 if (chan_is_physical(d40c)) {
2059 if (d40c->phy_chan->num % 2 == 0)
2060 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2062 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2064 status = (readl(active_reg) &
2065 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2066 D40_CHAN_POS(d40c->phy_chan->num);
2067 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2073 if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2074 d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
2075 status = readl(chanbase + D40_CHAN_REG_SDLNK);
2076 } else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
2077 status = readl(chanbase + D40_CHAN_REG_SSLNK);
2079 chan_err(d40c, "Unknown direction\n");
2083 status = (status & D40_EVENTLINE_MASK(event)) >>
2084 D40_EVENTLINE_POS(event);
2086 if (status != D40_DMA_RUN)
2089 spin_unlock_irqrestore(&d40c->lock, flags);
2094 static u32 stedma40_residue(struct dma_chan *chan)
2096 struct d40_chan *d40c =
2097 container_of(chan, struct d40_chan, chan);
2099 unsigned long flags;
2101 spin_lock_irqsave(&d40c->lock, flags);
2102 bytes_left = d40_residue(d40c);
2103 spin_unlock_irqrestore(&d40c->lock, flags);
2109 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2110 struct scatterlist *sg_src, struct scatterlist *sg_dst,
2111 unsigned int sg_len, dma_addr_t src_dev_addr,
2112 dma_addr_t dst_dev_addr)
2114 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2115 struct stedma40_half_channel_info *src_info = &cfg->src_info;
2116 struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2119 ret = d40_log_sg_to_lli(sg_src, sg_len,
2122 chan->log_def.lcsp1,
2123 src_info->data_width,
2124 dst_info->data_width);
2126 ret = d40_log_sg_to_lli(sg_dst, sg_len,
2129 chan->log_def.lcsp3,
2130 dst_info->data_width,
2131 src_info->data_width);
2133 return ret < 0 ? ret : 0;
2137 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2138 struct scatterlist *sg_src, struct scatterlist *sg_dst,
2139 unsigned int sg_len, dma_addr_t src_dev_addr,
2140 dma_addr_t dst_dev_addr)
2142 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2143 struct stedma40_half_channel_info *src_info = &cfg->src_info;
2144 struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2145 unsigned long flags = 0;
2149 flags |= LLI_CYCLIC | LLI_TERM_INT;
2151 ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2153 virt_to_phys(desc->lli_phy.src),
2155 src_info, dst_info, flags);
2157 ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2159 virt_to_phys(desc->lli_phy.dst),
2161 dst_info, src_info, flags);
2163 dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2164 desc->lli_pool.size, DMA_TO_DEVICE);
2166 return ret < 0 ? ret : 0;
2169 static struct d40_desc *
2170 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2171 unsigned int sg_len, unsigned long dma_flags)
2173 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2174 struct d40_desc *desc;
2177 desc = d40_desc_get(chan);
2181 desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2182 cfg->dst_info.data_width);
2183 if (desc->lli_len < 0) {
2184 chan_err(chan, "Unaligned size\n");
2188 ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2190 chan_err(chan, "Could not allocate lli\n");
2194 desc->lli_current = 0;
2195 desc->txd.flags = dma_flags;
2196 desc->txd.tx_submit = d40_tx_submit;
2198 dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2203 d40_desc_free(chan, desc);
2207 static struct dma_async_tx_descriptor *
2208 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2209 struct scatterlist *sg_dst, unsigned int sg_len,
2210 enum dma_transfer_direction direction, unsigned long dma_flags)
2212 struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2213 dma_addr_t src_dev_addr = 0;
2214 dma_addr_t dst_dev_addr = 0;
2215 struct d40_desc *desc;
2216 unsigned long flags;
2219 if (!chan->phy_chan) {
2220 chan_err(chan, "Cannot prepare unallocated channel\n");
2224 spin_lock_irqsave(&chan->lock, flags);
2226 desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2230 if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2231 desc->cyclic = true;
2233 if (direction == DMA_DEV_TO_MEM)
2234 src_dev_addr = chan->runtime_addr;
2235 else if (direction == DMA_MEM_TO_DEV)
2236 dst_dev_addr = chan->runtime_addr;
2238 if (chan_is_logical(chan))
2239 ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2240 sg_len, src_dev_addr, dst_dev_addr);
2242 ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2243 sg_len, src_dev_addr, dst_dev_addr);
2246 chan_err(chan, "Failed to prepare %s sg job: %d\n",
2247 chan_is_logical(chan) ? "log" : "phy", ret);
2252 * add descriptor to the prepare queue in order to be able
2253 * to free them later in terminate_all
2255 list_add_tail(&desc->node, &chan->prepare_queue);
2257 spin_unlock_irqrestore(&chan->lock, flags);
2263 d40_desc_free(chan, desc);
2264 spin_unlock_irqrestore(&chan->lock, flags);
2268 bool stedma40_filter(struct dma_chan *chan, void *data)
2270 struct stedma40_chan_cfg *info = data;
2271 struct d40_chan *d40c =
2272 container_of(chan, struct d40_chan, chan);
2276 err = d40_validate_conf(d40c, info);
2278 d40c->dma_cfg = *info;
2280 err = d40_config_memcpy(d40c);
2283 d40c->configured = true;
2287 EXPORT_SYMBOL(stedma40_filter);
2289 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2291 bool realtime = d40c->dma_cfg.realtime;
2292 bool highprio = d40c->dma_cfg.high_priority;
2294 u32 event = D40_TYPE_TO_EVENT(dev_type);
2295 u32 group = D40_TYPE_TO_GROUP(dev_type);
2296 u32 bit = BIT(event);
2298 struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2300 rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2302 * Due to a hardware bug, in some cases a logical channel triggered by
2303 * a high priority destination event line can generate extra packet
2306 * The workaround is to not set the high priority level for the
2307 * destination event lines that trigger logical channels.
2309 if (!src && chan_is_logical(d40c))
2312 prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2314 /* Destination event lines are stored in the upper halfword */
2318 writel(bit, d40c->base->virtbase + prioreg + group * 4);
2319 writel(bit, d40c->base->virtbase + rtreg + group * 4);
2322 static void d40_set_prio_realtime(struct d40_chan *d40c)
2324 if (d40c->base->rev < 3)
2327 if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) ||
2328 (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2329 __d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, true);
2331 if ((d40c->dma_cfg.dir == DMA_MEM_TO_DEV) ||
2332 (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2333 __d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, false);
2336 #define D40_DT_FLAGS_MODE(flags) ((flags >> 0) & 0x1)
2337 #define D40_DT_FLAGS_DIR(flags) ((flags >> 1) & 0x1)
2338 #define D40_DT_FLAGS_BIG_ENDIAN(flags) ((flags >> 2) & 0x1)
2339 #define D40_DT_FLAGS_FIXED_CHAN(flags) ((flags >> 3) & 0x1)
2340 #define D40_DT_FLAGS_HIGH_PRIO(flags) ((flags >> 4) & 0x1)
2342 static struct dma_chan *d40_xlate(struct of_phandle_args *dma_spec,
2343 struct of_dma *ofdma)
2345 struct stedma40_chan_cfg cfg;
2349 memset(&cfg, 0, sizeof(struct stedma40_chan_cfg));
2352 dma_cap_set(DMA_SLAVE, cap);
2354 cfg.dev_type = dma_spec->args[0];
2355 flags = dma_spec->args[2];
2357 switch (D40_DT_FLAGS_MODE(flags)) {
2358 case 0: cfg.mode = STEDMA40_MODE_LOGICAL; break;
2359 case 1: cfg.mode = STEDMA40_MODE_PHYSICAL; break;
2362 switch (D40_DT_FLAGS_DIR(flags)) {
2364 cfg.dir = DMA_MEM_TO_DEV;
2365 cfg.dst_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2368 cfg.dir = DMA_DEV_TO_MEM;
2369 cfg.src_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2373 if (D40_DT_FLAGS_FIXED_CHAN(flags)) {
2374 cfg.phy_channel = dma_spec->args[1];
2375 cfg.use_fixed_channel = true;
2378 if (D40_DT_FLAGS_HIGH_PRIO(flags))
2379 cfg.high_priority = true;
2381 return dma_request_channel(cap, stedma40_filter, &cfg);
2384 /* DMA ENGINE functions */
2385 static int d40_alloc_chan_resources(struct dma_chan *chan)
2388 unsigned long flags;
2389 struct d40_chan *d40c =
2390 container_of(chan, struct d40_chan, chan);
2392 spin_lock_irqsave(&d40c->lock, flags);
2394 dma_cookie_init(chan);
2396 /* If no dma configuration is set use default configuration (memcpy) */
2397 if (!d40c->configured) {
2398 err = d40_config_memcpy(d40c);
2400 chan_err(d40c, "Failed to configure memcpy channel\n");
2405 err = d40_allocate_channel(d40c, &is_free_phy);
2407 chan_err(d40c, "Failed to allocate channel\n");
2408 d40c->configured = false;
2412 pm_runtime_get_sync(d40c->base->dev);
2414 d40_set_prio_realtime(d40c);
2416 if (chan_is_logical(d40c)) {
2417 if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2418 d40c->lcpa = d40c->base->lcpa_base +
2419 d40c->dma_cfg.dev_type * D40_LCPA_CHAN_SIZE;
2421 d40c->lcpa = d40c->base->lcpa_base +
2422 d40c->dma_cfg.dev_type *
2423 D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2425 /* Unmask the Global Interrupt Mask. */
2426 d40c->src_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2427 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2430 dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2431 chan_is_logical(d40c) ? "logical" : "physical",
2432 d40c->phy_chan->num,
2433 d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2437 * Only write channel configuration to the DMA if the physical
2438 * resource is free. In case of multiple logical channels
2439 * on the same physical resource, only the first write is necessary.
2442 d40_config_write(d40c);
2444 pm_runtime_mark_last_busy(d40c->base->dev);
2445 pm_runtime_put_autosuspend(d40c->base->dev);
2446 spin_unlock_irqrestore(&d40c->lock, flags);
2450 static void d40_free_chan_resources(struct dma_chan *chan)
2452 struct d40_chan *d40c =
2453 container_of(chan, struct d40_chan, chan);
2455 unsigned long flags;
2457 if (d40c->phy_chan == NULL) {
2458 chan_err(d40c, "Cannot free unallocated channel\n");
2462 spin_lock_irqsave(&d40c->lock, flags);
2464 err = d40_free_dma(d40c);
2467 chan_err(d40c, "Failed to free channel\n");
2468 spin_unlock_irqrestore(&d40c->lock, flags);
2471 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2475 unsigned long dma_flags)
2477 struct scatterlist dst_sg;
2478 struct scatterlist src_sg;
2480 sg_init_table(&dst_sg, 1);
2481 sg_init_table(&src_sg, 1);
2483 sg_dma_address(&dst_sg) = dst;
2484 sg_dma_address(&src_sg) = src;
2486 sg_dma_len(&dst_sg) = size;
2487 sg_dma_len(&src_sg) = size;
2489 return d40_prep_sg(chan, &src_sg, &dst_sg, 1,
2490 DMA_MEM_TO_MEM, dma_flags);
2493 static struct dma_async_tx_descriptor *
2494 d40_prep_memcpy_sg(struct dma_chan *chan,
2495 struct scatterlist *dst_sg, unsigned int dst_nents,
2496 struct scatterlist *src_sg, unsigned int src_nents,
2497 unsigned long dma_flags)
2499 if (dst_nents != src_nents)
2502 return d40_prep_sg(chan, src_sg, dst_sg, src_nents,
2503 DMA_MEM_TO_MEM, dma_flags);
2506 static struct dma_async_tx_descriptor *
2507 d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2508 unsigned int sg_len, enum dma_transfer_direction direction,
2509 unsigned long dma_flags, void *context)
2511 if (!is_slave_direction(direction))
2514 return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2517 static struct dma_async_tx_descriptor *
2518 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2519 size_t buf_len, size_t period_len,
2520 enum dma_transfer_direction direction, unsigned long flags)
2522 unsigned int periods = buf_len / period_len;
2523 struct dma_async_tx_descriptor *txd;
2524 struct scatterlist *sg;
2527 sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2531 for (i = 0; i < periods; i++) {
2532 sg_dma_address(&sg[i]) = dma_addr;
2533 sg_dma_len(&sg[i]) = period_len;
2534 dma_addr += period_len;
2537 sg[periods].offset = 0;
2538 sg_dma_len(&sg[periods]) = 0;
2539 sg[periods].page_link =
2540 ((unsigned long)sg | 0x01) & ~0x02;
2542 txd = d40_prep_sg(chan, sg, sg, periods, direction,
2543 DMA_PREP_INTERRUPT);
2550 static enum dma_status d40_tx_status(struct dma_chan *chan,
2551 dma_cookie_t cookie,
2552 struct dma_tx_state *txstate)
2554 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2555 enum dma_status ret;
2557 if (d40c->phy_chan == NULL) {
2558 chan_err(d40c, "Cannot read status of unallocated channel\n");
2562 ret = dma_cookie_status(chan, cookie, txstate);
2563 if (ret != DMA_COMPLETE && txstate)
2564 dma_set_residue(txstate, stedma40_residue(chan));
2566 if (d40_is_paused(d40c))
2572 static void d40_issue_pending(struct dma_chan *chan)
2574 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2575 unsigned long flags;
2577 if (d40c->phy_chan == NULL) {
2578 chan_err(d40c, "Channel is not allocated!\n");
2582 spin_lock_irqsave(&d40c->lock, flags);
2584 list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2586 /* Busy means that queued jobs are already being processed */
2588 (void) d40_queue_start(d40c);
2590 spin_unlock_irqrestore(&d40c->lock, flags);
2593 static int d40_terminate_all(struct dma_chan *chan)
2595 unsigned long flags;
2596 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2599 if (d40c->phy_chan == NULL) {
2600 chan_err(d40c, "Channel is not allocated!\n");
2604 spin_lock_irqsave(&d40c->lock, flags);
2606 pm_runtime_get_sync(d40c->base->dev);
2607 ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2609 chan_err(d40c, "Failed to stop channel\n");
2612 pm_runtime_mark_last_busy(d40c->base->dev);
2613 pm_runtime_put_autosuspend(d40c->base->dev);
2615 pm_runtime_mark_last_busy(d40c->base->dev);
2616 pm_runtime_put_autosuspend(d40c->base->dev);
2620 spin_unlock_irqrestore(&d40c->lock, flags);
2625 dma40_config_to_halfchannel(struct d40_chan *d40c,
2626 struct stedma40_half_channel_info *info,
2631 if (chan_is_logical(d40c)) {
2633 psize = STEDMA40_PSIZE_LOG_16;
2634 else if (maxburst >= 8)
2635 psize = STEDMA40_PSIZE_LOG_8;
2636 else if (maxburst >= 4)
2637 psize = STEDMA40_PSIZE_LOG_4;
2639 psize = STEDMA40_PSIZE_LOG_1;
2642 psize = STEDMA40_PSIZE_PHY_16;
2643 else if (maxburst >= 8)
2644 psize = STEDMA40_PSIZE_PHY_8;
2645 else if (maxburst >= 4)
2646 psize = STEDMA40_PSIZE_PHY_4;
2648 psize = STEDMA40_PSIZE_PHY_1;
2651 info->psize = psize;
2652 info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2657 /* Runtime reconfiguration extension */
2658 static int d40_set_runtime_config(struct dma_chan *chan,
2659 struct dma_slave_config *config)
2661 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2662 struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2663 enum dma_slave_buswidth src_addr_width, dst_addr_width;
2664 dma_addr_t config_addr;
2665 u32 src_maxburst, dst_maxburst;
2668 if (d40c->phy_chan == NULL) {
2669 chan_err(d40c, "Channel is not allocated!\n");
2673 src_addr_width = config->src_addr_width;
2674 src_maxburst = config->src_maxburst;
2675 dst_addr_width = config->dst_addr_width;
2676 dst_maxburst = config->dst_maxburst;
2678 if (config->direction == DMA_DEV_TO_MEM) {
2679 config_addr = config->src_addr;
2681 if (cfg->dir != DMA_DEV_TO_MEM)
2682 dev_dbg(d40c->base->dev,
2683 "channel was not configured for peripheral "
2684 "to memory transfer (%d) overriding\n",
2686 cfg->dir = DMA_DEV_TO_MEM;
2688 /* Configure the memory side */
2689 if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2690 dst_addr_width = src_addr_width;
2691 if (dst_maxburst == 0)
2692 dst_maxburst = src_maxburst;
2694 } else if (config->direction == DMA_MEM_TO_DEV) {
2695 config_addr = config->dst_addr;
2697 if (cfg->dir != DMA_MEM_TO_DEV)
2698 dev_dbg(d40c->base->dev,
2699 "channel was not configured for memory "
2700 "to peripheral transfer (%d) overriding\n",
2702 cfg->dir = DMA_MEM_TO_DEV;
2704 /* Configure the memory side */
2705 if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2706 src_addr_width = dst_addr_width;
2707 if (src_maxburst == 0)
2708 src_maxburst = dst_maxburst;
2710 dev_err(d40c->base->dev,
2711 "unrecognized channel direction %d\n",
2716 if (config_addr <= 0) {
2717 dev_err(d40c->base->dev, "no address supplied\n");
2721 if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2722 dev_err(d40c->base->dev,
2723 "src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2731 if (src_maxburst > 16) {
2733 dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2734 } else if (dst_maxburst > 16) {
2736 src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2739 /* Only valid widths are; 1, 2, 4 and 8. */
2740 if (src_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2741 src_addr_width > DMA_SLAVE_BUSWIDTH_8_BYTES ||
2742 dst_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2743 dst_addr_width > DMA_SLAVE_BUSWIDTH_8_BYTES ||
2744 !is_power_of_2(src_addr_width) ||
2745 !is_power_of_2(dst_addr_width))
2748 cfg->src_info.data_width = src_addr_width;
2749 cfg->dst_info.data_width = dst_addr_width;
2751 ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2756 ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2761 /* Fill in register values */
2762 if (chan_is_logical(d40c))
2763 d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2765 d40_phy_cfg(cfg, &d40c->src_def_cfg, &d40c->dst_def_cfg);
2767 /* These settings will take precedence later */
2768 d40c->runtime_addr = config_addr;
2769 d40c->runtime_direction = config->direction;
2770 dev_dbg(d40c->base->dev,
2771 "configured channel %s for %s, data width %d/%d, "
2772 "maxburst %d/%d elements, LE, no flow control\n",
2773 dma_chan_name(chan),
2774 (config->direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2775 src_addr_width, dst_addr_width,
2776 src_maxburst, dst_maxburst);
2781 /* Initialization functions */
2783 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2784 struct d40_chan *chans, int offset,
2788 struct d40_chan *d40c;
2790 INIT_LIST_HEAD(&dma->channels);
2792 for (i = offset; i < offset + num_chans; i++) {
2795 d40c->chan.device = dma;
2797 spin_lock_init(&d40c->lock);
2799 d40c->log_num = D40_PHY_CHAN;
2801 INIT_LIST_HEAD(&d40c->done);
2802 INIT_LIST_HEAD(&d40c->active);
2803 INIT_LIST_HEAD(&d40c->queue);
2804 INIT_LIST_HEAD(&d40c->pending_queue);
2805 INIT_LIST_HEAD(&d40c->client);
2806 INIT_LIST_HEAD(&d40c->prepare_queue);
2808 tasklet_init(&d40c->tasklet, dma_tasklet,
2809 (unsigned long) d40c);
2811 list_add_tail(&d40c->chan.device_node,
2816 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2818 if (dma_has_cap(DMA_SLAVE, dev->cap_mask))
2819 dev->device_prep_slave_sg = d40_prep_slave_sg;
2821 if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2822 dev->device_prep_dma_memcpy = d40_prep_memcpy;
2825 * This controller can only access address at even
2826 * 32bit boundaries, i.e. 2^2
2828 dev->copy_align = DMAENGINE_ALIGN_4_BYTES;
2831 if (dma_has_cap(DMA_SG, dev->cap_mask))
2832 dev->device_prep_dma_sg = d40_prep_memcpy_sg;
2834 if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2835 dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2837 dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2838 dev->device_free_chan_resources = d40_free_chan_resources;
2839 dev->device_issue_pending = d40_issue_pending;
2840 dev->device_tx_status = d40_tx_status;
2841 dev->device_config = d40_set_runtime_config;
2842 dev->device_pause = d40_pause;
2843 dev->device_resume = d40_resume;
2844 dev->device_terminate_all = d40_terminate_all;
2845 dev->dev = base->dev;
2848 static int __init d40_dmaengine_init(struct d40_base *base,
2849 int num_reserved_chans)
2853 d40_chan_init(base, &base->dma_slave, base->log_chans,
2854 0, base->num_log_chans);
2856 dma_cap_zero(base->dma_slave.cap_mask);
2857 dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2858 dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2860 d40_ops_init(base, &base->dma_slave);
2862 err = dma_async_device_register(&base->dma_slave);
2865 d40_err(base->dev, "Failed to register slave channels\n");
2869 d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2870 base->num_log_chans, base->num_memcpy_chans);
2872 dma_cap_zero(base->dma_memcpy.cap_mask);
2873 dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2874 dma_cap_set(DMA_SG, base->dma_memcpy.cap_mask);
2876 d40_ops_init(base, &base->dma_memcpy);
2878 err = dma_async_device_register(&base->dma_memcpy);
2882 "Failed to register memcpy only channels\n");
2886 d40_chan_init(base, &base->dma_both, base->phy_chans,
2887 0, num_reserved_chans);
2889 dma_cap_zero(base->dma_both.cap_mask);
2890 dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2891 dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2892 dma_cap_set(DMA_SG, base->dma_both.cap_mask);
2893 dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2895 d40_ops_init(base, &base->dma_both);
2896 err = dma_async_device_register(&base->dma_both);
2900 "Failed to register logical and physical capable channels\n");
2905 dma_async_device_unregister(&base->dma_memcpy);
2907 dma_async_device_unregister(&base->dma_slave);
2912 /* Suspend resume functionality */
2913 #ifdef CONFIG_PM_SLEEP
2914 static int dma40_suspend(struct device *dev)
2916 struct platform_device *pdev = to_platform_device(dev);
2917 struct d40_base *base = platform_get_drvdata(pdev);
2920 ret = pm_runtime_force_suspend(dev);
2924 if (base->lcpa_regulator)
2925 ret = regulator_disable(base->lcpa_regulator);
2929 static int dma40_resume(struct device *dev)
2931 struct platform_device *pdev = to_platform_device(dev);
2932 struct d40_base *base = platform_get_drvdata(pdev);
2935 if (base->lcpa_regulator) {
2936 ret = regulator_enable(base->lcpa_regulator);
2941 return pm_runtime_force_resume(dev);
2946 static void dma40_backup(void __iomem *baseaddr, u32 *backup,
2947 u32 *regaddr, int num, bool save)
2951 for (i = 0; i < num; i++) {
2952 void __iomem *addr = baseaddr + regaddr[i];
2955 backup[i] = readl_relaxed(addr);
2957 writel_relaxed(backup[i], addr);
2961 static void d40_save_restore_registers(struct d40_base *base, bool save)
2965 /* Save/Restore channel specific registers */
2966 for (i = 0; i < base->num_phy_chans; i++) {
2970 if (base->phy_res[i].reserved)
2973 addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA;
2974 idx = i * ARRAY_SIZE(d40_backup_regs_chan);
2976 dma40_backup(addr, &base->reg_val_backup_chan[idx],
2977 d40_backup_regs_chan,
2978 ARRAY_SIZE(d40_backup_regs_chan),
2982 /* Save/Restore global registers */
2983 dma40_backup(base->virtbase, base->reg_val_backup,
2984 d40_backup_regs, ARRAY_SIZE(d40_backup_regs),
2987 /* Save/Restore registers only existing on dma40 v3 and later */
2988 if (base->gen_dmac.backup)
2989 dma40_backup(base->virtbase, base->reg_val_backup_v4,
2990 base->gen_dmac.backup,
2991 base->gen_dmac.backup_size,
2995 static int dma40_runtime_suspend(struct device *dev)
2997 struct platform_device *pdev = to_platform_device(dev);
2998 struct d40_base *base = platform_get_drvdata(pdev);
3000 d40_save_restore_registers(base, true);
3002 /* Don't disable/enable clocks for v1 due to HW bugs */
3004 writel_relaxed(base->gcc_pwr_off_mask,
3005 base->virtbase + D40_DREG_GCC);
3010 static int dma40_runtime_resume(struct device *dev)
3012 struct platform_device *pdev = to_platform_device(dev);
3013 struct d40_base *base = platform_get_drvdata(pdev);
3015 d40_save_restore_registers(base, false);
3017 writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
3018 base->virtbase + D40_DREG_GCC);
3023 static const struct dev_pm_ops dma40_pm_ops = {
3024 SET_LATE_SYSTEM_SLEEP_PM_OPS(dma40_suspend, dma40_resume)
3025 SET_RUNTIME_PM_OPS(dma40_runtime_suspend,
3026 dma40_runtime_resume,
3030 /* Initialization functions. */
3032 static int __init d40_phy_res_init(struct d40_base *base)
3035 int num_phy_chans_avail = 0;
3037 int odd_even_bit = -2;
3038 int gcc = D40_DREG_GCC_ENA;
3040 val[0] = readl(base->virtbase + D40_DREG_PRSME);
3041 val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3043 for (i = 0; i < base->num_phy_chans; i++) {
3044 base->phy_res[i].num = i;
3045 odd_even_bit += 2 * ((i % 2) == 0);
3046 if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3047 /* Mark security only channels as occupied */
3048 base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3049 base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3050 base->phy_res[i].reserved = true;
3051 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3053 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3058 base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3059 base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3060 base->phy_res[i].reserved = false;
3061 num_phy_chans_avail++;
3063 spin_lock_init(&base->phy_res[i].lock);
3066 /* Mark disabled channels as occupied */
3067 for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3068 int chan = base->plat_data->disabled_channels[i];
3070 base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3071 base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3072 base->phy_res[chan].reserved = true;
3073 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3075 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3077 num_phy_chans_avail--;
3080 /* Mark soft_lli channels */
3081 for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3082 int chan = base->plat_data->soft_lli_chans[i];
3084 base->phy_res[chan].use_soft_lli = true;
3087 dev_info(base->dev, "%d of %d physical DMA channels available\n",
3088 num_phy_chans_avail, base->num_phy_chans);
3090 /* Verify settings extended vs standard */
3091 val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3093 for (i = 0; i < base->num_phy_chans; i++) {
3095 if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3096 (val[0] & 0x3) != 1)
3098 "[%s] INFO: channel %d is misconfigured (%d)\n",
3099 __func__, i, val[0] & 0x3);
3101 val[0] = val[0] >> 2;
3105 * To keep things simple, Enable all clocks initially.
3106 * The clocks will get managed later post channel allocation.
3107 * The clocks for the event lines on which reserved channels exists
3108 * are not managed here.
3110 writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3111 base->gcc_pwr_off_mask = gcc;
3113 return num_phy_chans_avail;
3116 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
3118 struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3119 struct clk *clk = NULL;
3120 void __iomem *virtbase = NULL;
3121 struct resource *res = NULL;
3122 struct d40_base *base = NULL;
3123 int num_log_chans = 0;
3125 int num_memcpy_chans;
3126 int clk_ret = -EINVAL;
3132 clk = clk_get(&pdev->dev, NULL);
3134 d40_err(&pdev->dev, "No matching clock found\n");
3138 clk_ret = clk_prepare_enable(clk);
3140 d40_err(&pdev->dev, "Failed to prepare/enable clock\n");
3144 /* Get IO for DMAC base address */
3145 res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
3149 if (request_mem_region(res->start, resource_size(res),
3150 D40_NAME " I/O base") == NULL)
3153 virtbase = ioremap(res->start, resource_size(res));
3157 /* This is just a regular AMBA PrimeCell ID actually */
3158 for (pid = 0, i = 0; i < 4; i++)
3159 pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
3161 for (cid = 0, i = 0; i < 4; i++)
3162 cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
3165 if (cid != AMBA_CID) {
3166 d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
3169 if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3170 d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
3171 AMBA_MANF_BITS(pid),
3177 * DB8500ed has revision 0
3179 * DB8500v1 has revision 2
3180 * DB8500v2 has revision 3
3181 * AP9540v1 has revision 4
3182 * DB8540v1 has revision 4
3184 rev = AMBA_REV_BITS(pid);
3186 d40_err(&pdev->dev, "hardware revision: %d is not supported", rev);
3190 /* The number of physical channels on this HW */
3191 if (plat_data->num_of_phy_chans)
3192 num_phy_chans = plat_data->num_of_phy_chans;
3194 num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3196 /* The number of channels used for memcpy */
3197 if (plat_data->num_of_memcpy_chans)
3198 num_memcpy_chans = plat_data->num_of_memcpy_chans;
3200 num_memcpy_chans = ARRAY_SIZE(dma40_memcpy_channels);
3202 num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY;
3204 dev_info(&pdev->dev,
3205 "hardware rev: %d @ %pa with %d physical and %d logical channels\n",
3206 rev, &res->start, num_phy_chans, num_log_chans);
3208 base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
3209 (num_phy_chans + num_log_chans + num_memcpy_chans) *
3210 sizeof(struct d40_chan), GFP_KERNEL);
3217 base->num_memcpy_chans = num_memcpy_chans;
3218 base->num_phy_chans = num_phy_chans;
3219 base->num_log_chans = num_log_chans;
3220 base->phy_start = res->start;
3221 base->phy_size = resource_size(res);
3222 base->virtbase = virtbase;
3223 base->plat_data = plat_data;
3224 base->dev = &pdev->dev;
3225 base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3226 base->log_chans = &base->phy_chans[num_phy_chans];
3228 if (base->plat_data->num_of_phy_chans == 14) {
3229 base->gen_dmac.backup = d40_backup_regs_v4b;
3230 base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3231 base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3232 base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3233 base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3234 base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3235 base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3236 base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3237 base->gen_dmac.il = il_v4b;
3238 base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3239 base->gen_dmac.init_reg = dma_init_reg_v4b;
3240 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3242 if (base->rev >= 3) {
3243 base->gen_dmac.backup = d40_backup_regs_v4a;
3244 base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3246 base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3247 base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3248 base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3249 base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3250 base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3251 base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3252 base->gen_dmac.il = il_v4a;
3253 base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3254 base->gen_dmac.init_reg = dma_init_reg_v4a;
3255 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3258 base->phy_res = kzalloc(num_phy_chans * sizeof(struct d40_phy_res),
3263 base->lookup_phy_chans = kzalloc(num_phy_chans *
3264 sizeof(struct d40_chan *),
3266 if (!base->lookup_phy_chans)
3269 base->lookup_log_chans = kzalloc(num_log_chans *
3270 sizeof(struct d40_chan *),
3272 if (!base->lookup_log_chans)
3275 base->reg_val_backup_chan = kmalloc(base->num_phy_chans *
3276 sizeof(d40_backup_regs_chan),
3278 if (!base->reg_val_backup_chan)
3281 base->lcla_pool.alloc_map =
3282 kzalloc(num_phy_chans * sizeof(struct d40_desc *)
3283 * D40_LCLA_LINK_PER_EVENT_GRP, GFP_KERNEL);
3284 if (!base->lcla_pool.alloc_map)
3287 base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3288 0, SLAB_HWCACHE_ALIGN,
3290 if (base->desc_slab == NULL)
3297 clk_disable_unprepare(clk);
3303 release_mem_region(res->start,
3304 resource_size(res));
3309 kfree(base->lcla_pool.alloc_map);
3310 kfree(base->reg_val_backup_chan);
3311 kfree(base->lookup_log_chans);
3312 kfree(base->lookup_phy_chans);
3313 kfree(base->phy_res);
3320 static void __init d40_hw_init(struct d40_base *base)
3324 u32 prmseo[2] = {0, 0};
3325 u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3328 struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3329 u32 reg_size = base->gen_dmac.init_reg_size;
3331 for (i = 0; i < reg_size; i++)
3332 writel(dma_init_reg[i].val,
3333 base->virtbase + dma_init_reg[i].reg);
3335 /* Configure all our dma channels to default settings */
3336 for (i = 0; i < base->num_phy_chans; i++) {
3338 activeo[i % 2] = activeo[i % 2] << 2;
3340 if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3342 activeo[i % 2] |= 3;
3346 /* Enable interrupt # */
3347 pcmis = (pcmis << 1) | 1;
3349 /* Clear interrupt # */
3350 pcicr = (pcicr << 1) | 1;
3352 /* Set channel to physical mode */
3353 prmseo[i % 2] = prmseo[i % 2] << 2;
3358 writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3359 writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3360 writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3361 writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3363 /* Write which interrupt to enable */
3364 writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3366 /* Write which interrupt to clear */
3367 writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3369 /* These are __initdata and cannot be accessed after init */
3370 base->gen_dmac.init_reg = NULL;
3371 base->gen_dmac.init_reg_size = 0;
3374 static int __init d40_lcla_allocate(struct d40_base *base)
3376 struct d40_lcla_pool *pool = &base->lcla_pool;
3377 unsigned long *page_list;
3382 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3383 * To full fill this hardware requirement without wasting 256 kb
3384 * we allocate pages until we get an aligned one.
3386 page_list = kmalloc(sizeof(unsigned long) * MAX_LCLA_ALLOC_ATTEMPTS,
3394 /* Calculating how many pages that are required */
3395 base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3397 for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3398 page_list[i] = __get_free_pages(GFP_KERNEL,
3399 base->lcla_pool.pages);
3400 if (!page_list[i]) {
3402 d40_err(base->dev, "Failed to allocate %d pages.\n",
3403 base->lcla_pool.pages);
3406 for (j = 0; j < i; j++)
3407 free_pages(page_list[j], base->lcla_pool.pages);
3411 if ((virt_to_phys((void *)page_list[i]) &
3412 (LCLA_ALIGNMENT - 1)) == 0)
3416 for (j = 0; j < i; j++)
3417 free_pages(page_list[j], base->lcla_pool.pages);
3419 if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3420 base->lcla_pool.base = (void *)page_list[i];
3423 * After many attempts and no succees with finding the correct
3424 * alignment, try with allocating a big buffer.
3427 "[%s] Failed to get %d pages @ 18 bit align.\n",
3428 __func__, base->lcla_pool.pages);
3429 base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3430 base->num_phy_chans +
3433 if (!base->lcla_pool.base_unaligned) {
3438 base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3442 pool->dma_addr = dma_map_single(base->dev, pool->base,
3443 SZ_1K * base->num_phy_chans,
3445 if (dma_mapping_error(base->dev, pool->dma_addr)) {
3451 writel(virt_to_phys(base->lcla_pool.base),
3452 base->virtbase + D40_DREG_LCLA);
3458 static int __init d40_of_probe(struct platform_device *pdev,
3459 struct device_node *np)
3461 struct stedma40_platform_data *pdata;
3462 int num_phy = 0, num_memcpy = 0, num_disabled = 0;
3465 pdata = devm_kzalloc(&pdev->dev,
3466 sizeof(struct stedma40_platform_data),
3471 /* If absent this value will be obtained from h/w. */
3472 of_property_read_u32(np, "dma-channels", &num_phy);
3474 pdata->num_of_phy_chans = num_phy;
3476 list = of_get_property(np, "memcpy-channels", &num_memcpy);
3477 num_memcpy /= sizeof(*list);
3479 if (num_memcpy > D40_MEMCPY_MAX_CHANS || num_memcpy <= 0) {
3481 "Invalid number of memcpy channels specified (%d)\n",
3485 pdata->num_of_memcpy_chans = num_memcpy;
3487 of_property_read_u32_array(np, "memcpy-channels",
3488 dma40_memcpy_channels,
3491 list = of_get_property(np, "disabled-channels", &num_disabled);
3492 num_disabled /= sizeof(*list);
3494 if (num_disabled >= STEDMA40_MAX_PHYS || num_disabled < 0) {
3496 "Invalid number of disabled channels specified (%d)\n",
3501 of_property_read_u32_array(np, "disabled-channels",
3502 pdata->disabled_channels,
3504 pdata->disabled_channels[num_disabled] = -1;
3506 pdev->dev.platform_data = pdata;
3511 static int __init d40_probe(struct platform_device *pdev)
3513 struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3514 struct device_node *np = pdev->dev.of_node;
3516 struct d40_base *base;
3517 struct resource *res;
3518 int num_reserved_chans;
3523 if (d40_of_probe(pdev, np)) {
3525 goto report_failure;
3528 d40_err(&pdev->dev, "No pdata or Device Tree provided\n");
3529 goto report_failure;
3533 base = d40_hw_detect_init(pdev);
3535 goto report_failure;
3537 num_reserved_chans = d40_phy_res_init(base);
3539 platform_set_drvdata(pdev, base);
3541 spin_lock_init(&base->interrupt_lock);
3542 spin_lock_init(&base->execmd_lock);
3544 /* Get IO for logical channel parameter address */
3545 res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
3548 d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
3551 base->lcpa_size = resource_size(res);
3552 base->phy_lcpa = res->start;
3554 if (request_mem_region(res->start, resource_size(res),
3555 D40_NAME " I/O lcpa") == NULL) {
3557 d40_err(&pdev->dev, "Failed to request LCPA region %pR\n", res);
3561 /* We make use of ESRAM memory for this. */
3562 val = readl(base->virtbase + D40_DREG_LCPA);
3563 if (res->start != val && val != 0) {
3564 dev_warn(&pdev->dev,
3565 "[%s] Mismatch LCPA dma 0x%x, def %pa\n",
3566 __func__, val, &res->start);
3568 writel(res->start, base->virtbase + D40_DREG_LCPA);
3570 base->lcpa_base = ioremap(res->start, resource_size(res));
3571 if (!base->lcpa_base) {
3573 d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
3576 /* If lcla has to be located in ESRAM we don't need to allocate */
3577 if (base->plat_data->use_esram_lcla) {
3578 res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3583 "No \"lcla_esram\" memory resource\n");
3586 base->lcla_pool.base = ioremap(res->start,
3587 resource_size(res));
3588 if (!base->lcla_pool.base) {
3590 d40_err(&pdev->dev, "Failed to ioremap LCLA region\n");
3593 writel(res->start, base->virtbase + D40_DREG_LCLA);
3596 ret = d40_lcla_allocate(base);
3598 d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
3603 spin_lock_init(&base->lcla_pool.lock);
3605 base->irq = platform_get_irq(pdev, 0);
3607 ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3609 d40_err(&pdev->dev, "No IRQ defined\n");
3613 if (base->plat_data->use_esram_lcla) {
3615 base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3616 if (IS_ERR(base->lcpa_regulator)) {
3617 d40_err(&pdev->dev, "Failed to get lcpa_regulator\n");
3618 ret = PTR_ERR(base->lcpa_regulator);
3619 base->lcpa_regulator = NULL;
3623 ret = regulator_enable(base->lcpa_regulator);
3626 "Failed to enable lcpa_regulator\n");
3627 regulator_put(base->lcpa_regulator);
3628 base->lcpa_regulator = NULL;
3633 writel_relaxed(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3635 pm_runtime_irq_safe(base->dev);
3636 pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3637 pm_runtime_use_autosuspend(base->dev);
3638 pm_runtime_mark_last_busy(base->dev);
3639 pm_runtime_set_active(base->dev);
3640 pm_runtime_enable(base->dev);
3642 ret = d40_dmaengine_init(base, num_reserved_chans);
3646 base->dev->dma_parms = &base->dma_parms;
3647 ret = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3649 d40_err(&pdev->dev, "Failed to set dma max seg size\n");
3656 ret = of_dma_controller_register(np, d40_xlate, NULL);
3659 "could not register of_dma_controller\n");
3662 dev_info(base->dev, "initialized\n");
3666 kmem_cache_destroy(base->desc_slab);
3668 iounmap(base->virtbase);
3670 if (base->lcla_pool.base && base->plat_data->use_esram_lcla) {
3671 iounmap(base->lcla_pool.base);
3672 base->lcla_pool.base = NULL;
3675 if (base->lcla_pool.dma_addr)
3676 dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3677 SZ_1K * base->num_phy_chans,
3680 if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3681 free_pages((unsigned long)base->lcla_pool.base,
3682 base->lcla_pool.pages);
3684 kfree(base->lcla_pool.base_unaligned);
3687 release_mem_region(base->phy_lcpa,
3689 if (base->phy_start)
3690 release_mem_region(base->phy_start,
3693 clk_disable_unprepare(base->clk);
3697 if (base->lcpa_regulator) {
3698 regulator_disable(base->lcpa_regulator);
3699 regulator_put(base->lcpa_regulator);
3702 kfree(base->lcla_pool.alloc_map);
3703 kfree(base->lookup_log_chans);
3704 kfree(base->lookup_phy_chans);
3705 kfree(base->phy_res);
3708 d40_err(&pdev->dev, "probe failed\n");
3712 static const struct of_device_id d40_match[] = {
3713 { .compatible = "stericsson,dma40", },
3717 static struct platform_driver d40_driver = {
3720 .pm = &dma40_pm_ops,
3721 .of_match_table = d40_match,
3725 static int __init stedma40_init(void)
3727 return platform_driver_probe(&d40_driver, d40_probe);
3729 subsys_initcall(stedma40_init);