2 * driver/dma/ste_dma40.c
4 * Copyright (C) ST-Ericsson 2007-2010
5 * License terms: GNU General Public License (GPL) version 2
6 * Author: Per Friden <per.friden@stericsson.com>
7 * Author: Jonas Aaberg <jonas.aberg@stericsson.com>
11 #include <linux/kernel.h>
12 #include <linux/slab.h>
13 #include <linux/dmaengine.h>
14 #include <linux/platform_device.h>
15 #include <linux/clk.h>
16 #include <linux/delay.h>
18 #include <plat/ste_dma40.h>
20 #include "ste_dma40_ll.h"
22 #define D40_NAME "dma40"
24 #define D40_PHY_CHAN -1
26 /* For masking out/in 2 bit channel positions */
27 #define D40_CHAN_POS(chan) (2 * (chan / 2))
28 #define D40_CHAN_POS_MASK(chan) (0x3 << D40_CHAN_POS(chan))
30 /* Maximum iterations taken before giving up suspending a channel */
31 #define D40_SUSPEND_MAX_IT 500
33 /* Hardware requirement on LCLA alignment */
34 #define LCLA_ALIGNMENT 0x40000
35 /* Attempts before giving up to trying to get pages that are aligned */
36 #define MAX_LCLA_ALLOC_ATTEMPTS 256
38 /* Bit markings for allocation map */
39 #define D40_ALLOC_FREE (1 << 31)
40 #define D40_ALLOC_PHY (1 << 30)
41 #define D40_ALLOC_LOG_FREE 0
43 /* Hardware designer of the block */
44 #define D40_HW_DESIGNER 0x8
47 * enum 40_command - The different commands and/or statuses.
49 * @D40_DMA_STOP: DMA channel command STOP or status STOPPED,
50 * @D40_DMA_RUN: The DMA channel is RUNNING of the command RUN.
51 * @D40_DMA_SUSPEND_REQ: Request the DMA to SUSPEND as soon as possible.
52 * @D40_DMA_SUSPENDED: The DMA channel is SUSPENDED.
57 D40_DMA_SUSPEND_REQ = 2,
62 * struct d40_lli_pool - Structure for keeping LLIs in memory
64 * @base: Pointer to memory area when the pre_alloc_lli's are not large
65 * enough, IE bigger than the most common case, 1 dst and 1 src. NULL if
66 * pre_alloc_lli is used.
67 * @size: The size in bytes of the memory at base or the size of pre_alloc_lli.
68 * @pre_alloc_lli: Pre allocated area for the most common case of transfers,
69 * one buffer to one buffer.
74 /* Space for dst and src, plus an extra for padding */
75 u8 pre_alloc_lli[3 * sizeof(struct d40_phy_lli)];
79 * struct d40_desc - A descriptor is one DMA job.
81 * @lli_phy: LLI settings for physical channel. Both src and dst=
82 * points into the lli_pool, to base if lli_len > 1 or to pre_alloc_lli if
84 * @lli_log: Same as above but for logical channels.
85 * @lli_pool: The pool with two entries pre-allocated.
86 * @lli_len: Number of llis of current descriptor.
87 * @lli_count: Number of transfered llis.
88 * @lli_tx_len: Max number of LLIs per transfer, there can be
89 * many transfer for one descriptor.
90 * @txd: DMA engine struct. Used for among other things for communication
93 * @dir: The transfer direction of this job.
94 * @is_in_client_list: true if the client owns this descriptor.
96 * This descriptor is used for both logical and physical transfers.
101 struct d40_phy_lli_bidir lli_phy;
103 struct d40_log_lli_bidir lli_log;
105 struct d40_lli_pool lli_pool;
110 struct dma_async_tx_descriptor txd;
111 struct list_head node;
113 enum dma_data_direction dir;
114 bool is_in_client_list;
118 * struct d40_lcla_pool - LCLA pool settings and data.
120 * @base: The virtual address of LCLA. 18 bit aligned.
121 * @base_unaligned: The orignal kmalloc pointer, if kmalloc is used.
122 * This pointer is only there for clean-up on error.
123 * @pages: The number of pages needed for all physical channels.
124 * Only used later for clean-up on error
125 * @lock: Lock to protect the content in this struct.
126 * @alloc_map: Bitmap mapping between physical channel and LCLA entries.
127 * @num_blocks: The number of entries of alloc_map. Equals to the
128 * number of physical channels.
130 struct d40_lcla_pool {
132 void *base_unaligned;
140 * struct d40_phy_res - struct for handling eventlines mapped to physical
143 * @lock: A lock protection this entity.
144 * @num: The physical channel number of this entity.
145 * @allocated_src: Bit mapped to show which src event line's are mapped to
146 * this physical channel. Can also be free or physically allocated.
147 * @allocated_dst: Same as for src but is dst.
148 * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as
149 * event line number. Both allocated_src and allocated_dst can not be
150 * allocated to a physical channel, since the interrupt handler has then
151 * no way of figure out which one the interrupt belongs to.
163 * struct d40_chan - Struct that describes a channel.
165 * @lock: A spinlock to protect this struct.
166 * @log_num: The logical number, if any of this channel.
167 * @completed: Starts with 1, after first interrupt it is set to dma engine's
169 * @pending_tx: The number of pending transfers. Used between interrupt handler
171 * @busy: Set to true when transfer is ongoing on this channel.
172 * @phy_chan: Pointer to physical channel which this instance runs on. If this
173 * point is NULL, then the channel is not allocated.
174 * @chan: DMA engine handle.
175 * @tasklet: Tasklet that gets scheduled from interrupt context to complete a
176 * transfer and call client callback.
177 * @client: Cliented owned descriptor list.
178 * @active: Active descriptor.
179 * @queue: Queued jobs.
180 * @dma_cfg: The client configuration of this dma channel.
181 * @base: Pointer to the device instance struct.
182 * @src_def_cfg: Default cfg register setting for src.
183 * @dst_def_cfg: Default cfg register setting for dst.
184 * @log_def: Default logical channel settings.
185 * @lcla: Space for one dst src pair for logical channel transfers.
186 * @lcpa: Pointer to dst and src lcpa settings.
188 * This struct can either "be" a logical or a physical channel.
193 /* ID of the most recent completed transfer */
197 struct d40_phy_res *phy_chan;
198 struct dma_chan chan;
199 struct tasklet_struct tasklet;
200 struct list_head client;
201 struct list_head active;
202 struct list_head queue;
203 struct stedma40_chan_cfg dma_cfg;
204 struct d40_base *base;
205 /* Default register configurations */
208 struct d40_def_lcsp log_def;
209 struct d40_lcla_elem lcla;
210 struct d40_log_lli_full *lcpa;
211 /* Runtime reconfiguration */
212 dma_addr_t runtime_addr;
213 enum dma_data_direction runtime_direction;
217 * struct d40_base - The big global struct, one for each probe'd instance.
219 * @interrupt_lock: Lock used to make sure one interrupt is handle a time.
220 * @execmd_lock: Lock for execute command usage since several channels share
221 * the same physical register.
222 * @dev: The device structure.
223 * @virtbase: The virtual base address of the DMA's register.
224 * @rev: silicon revision detected.
225 * @clk: Pointer to the DMA clock structure.
226 * @phy_start: Physical memory start of the DMA registers.
227 * @phy_size: Size of the DMA register map.
228 * @irq: The IRQ number.
229 * @num_phy_chans: The number of physical channels. Read from HW. This
230 * is the number of available channels for this driver, not counting "Secure
231 * mode" allocated physical channels.
232 * @num_log_chans: The number of logical channels. Calculated from
234 * @dma_both: dma_device channels that can do both memcpy and slave transfers.
235 * @dma_slave: dma_device channels that can do only do slave transfers.
236 * @dma_memcpy: dma_device channels that can do only do memcpy transfers.
237 * @phy_chans: Room for all possible physical channels in system.
238 * @log_chans: Room for all possible logical channels in system.
239 * @lookup_log_chans: Used to map interrupt number to logical channel. Points
240 * to log_chans entries.
241 * @lookup_phy_chans: Used to map interrupt number to physical channel. Points
242 * to phy_chans entries.
243 * @plat_data: Pointer to provided platform_data which is the driver
245 * @phy_res: Vector containing all physical channels.
246 * @lcla_pool: lcla pool settings and data.
247 * @lcpa_base: The virtual mapped address of LCPA.
248 * @phy_lcpa: The physical address of the LCPA.
249 * @lcpa_size: The size of the LCPA area.
250 * @desc_slab: cache for descriptors.
253 spinlock_t interrupt_lock;
254 spinlock_t execmd_lock;
256 void __iomem *virtbase;
259 phys_addr_t phy_start;
260 resource_size_t phy_size;
264 struct dma_device dma_both;
265 struct dma_device dma_slave;
266 struct dma_device dma_memcpy;
267 struct d40_chan *phy_chans;
268 struct d40_chan *log_chans;
269 struct d40_chan **lookup_log_chans;
270 struct d40_chan **lookup_phy_chans;
271 struct stedma40_platform_data *plat_data;
272 /* Physical half channels */
273 struct d40_phy_res *phy_res;
274 struct d40_lcla_pool lcla_pool;
277 resource_size_t lcpa_size;
278 struct kmem_cache *desc_slab;
282 * struct d40_interrupt_lookup - lookup table for interrupt handler
284 * @src: Interrupt mask register.
285 * @clr: Interrupt clear register.
286 * @is_error: true if this is an error interrupt.
287 * @offset: start delta in the lookup_log_chans in d40_base. If equals to
288 * D40_PHY_CHAN, the lookup_phy_chans shall be used instead.
290 struct d40_interrupt_lookup {
298 * struct d40_reg_val - simple lookup struct
300 * @reg: The register.
301 * @val: The value that belongs to the register in reg.
308 static int d40_pool_lli_alloc(struct d40_desc *d40d,
309 int lli_len, bool is_log)
315 align = sizeof(struct d40_log_lli);
317 align = sizeof(struct d40_phy_lli);
320 base = d40d->lli_pool.pre_alloc_lli;
321 d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
322 d40d->lli_pool.base = NULL;
324 d40d->lli_pool.size = ALIGN(lli_len * 2 * align, align);
326 base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
327 d40d->lli_pool.base = base;
329 if (d40d->lli_pool.base == NULL)
334 d40d->lli_log.src = PTR_ALIGN((struct d40_log_lli *) base,
336 d40d->lli_log.dst = PTR_ALIGN(d40d->lli_log.src + lli_len,
339 d40d->lli_phy.src = PTR_ALIGN((struct d40_phy_lli *)base,
341 d40d->lli_phy.dst = PTR_ALIGN(d40d->lli_phy.src + lli_len,
344 d40d->lli_phy.src_addr = virt_to_phys(d40d->lli_phy.src);
345 d40d->lli_phy.dst_addr = virt_to_phys(d40d->lli_phy.dst);
351 static void d40_pool_lli_free(struct d40_desc *d40d)
353 kfree(d40d->lli_pool.base);
354 d40d->lli_pool.base = NULL;
355 d40d->lli_pool.size = 0;
356 d40d->lli_log.src = NULL;
357 d40d->lli_log.dst = NULL;
358 d40d->lli_phy.src = NULL;
359 d40d->lli_phy.dst = NULL;
360 d40d->lli_phy.src_addr = 0;
361 d40d->lli_phy.dst_addr = 0;
364 static dma_cookie_t d40_assign_cookie(struct d40_chan *d40c,
365 struct d40_desc *desc)
367 dma_cookie_t cookie = d40c->chan.cookie;
372 d40c->chan.cookie = cookie;
373 desc->txd.cookie = cookie;
378 static void d40_desc_remove(struct d40_desc *d40d)
380 list_del(&d40d->node);
383 static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
388 if (!list_empty(&d40c->client)) {
389 list_for_each_entry_safe(d, _d, &d40c->client, node)
390 if (async_tx_test_ack(&d->txd)) {
391 d40_pool_lli_free(d);
396 d = kmem_cache_alloc(d40c->base->desc_slab, GFP_NOWAIT);
398 memset(d, 0, sizeof(struct d40_desc));
399 INIT_LIST_HEAD(&d->node);
405 static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
407 kmem_cache_free(d40c->base->desc_slab, d40d);
410 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
412 list_add_tail(&desc->node, &d40c->active);
415 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
419 if (list_empty(&d40c->active))
422 d = list_first_entry(&d40c->active,
428 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
430 list_add_tail(&desc->node, &d40c->queue);
433 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
437 if (list_empty(&d40c->queue))
440 d = list_first_entry(&d40c->queue,
446 /* Support functions for logical channels */
448 static int d40_lcla_id_get(struct d40_chan *d40c)
452 struct d40_log_lli *lcla_lidx_base =
453 d40c->base->lcla_pool.base + d40c->phy_chan->num * 1024;
455 int lli_per_log = d40c->base->plat_data->llis_per_log;
458 if (d40c->lcla.src_id >= 0 && d40c->lcla.dst_id >= 0)
461 if (d40c->base->lcla_pool.num_blocks > 32)
464 spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
466 for (i = 0; i < d40c->base->lcla_pool.num_blocks; i++) {
467 if (!(d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num] &
469 d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num] |=
475 if (src_id >= d40c->base->lcla_pool.num_blocks)
478 for (; i < d40c->base->lcla_pool.num_blocks; i++) {
479 if (!(d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num] &
481 d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num] |=
488 if (dst_id == src_id)
491 d40c->lcla.src_id = src_id;
492 d40c->lcla.dst_id = dst_id;
493 d40c->lcla.dst = lcla_lidx_base + dst_id * lli_per_log + 1;
494 d40c->lcla.src = lcla_lidx_base + src_id * lli_per_log + 1;
496 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
499 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
504 static int d40_channel_execute_command(struct d40_chan *d40c,
505 enum d40_command command)
508 void __iomem *active_reg;
513 spin_lock_irqsave(&d40c->base->execmd_lock, flags);
515 if (d40c->phy_chan->num % 2 == 0)
516 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
518 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
520 if (command == D40_DMA_SUSPEND_REQ) {
521 status = (readl(active_reg) &
522 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
523 D40_CHAN_POS(d40c->phy_chan->num);
525 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
529 wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
530 writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
533 if (command == D40_DMA_SUSPEND_REQ) {
535 for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
536 status = (readl(active_reg) &
537 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
538 D40_CHAN_POS(d40c->phy_chan->num);
542 * Reduce the number of bus accesses while
543 * waiting for the DMA to suspend.
547 if (status == D40_DMA_STOP ||
548 status == D40_DMA_SUSPENDED)
552 if (i == D40_SUSPEND_MAX_IT) {
553 dev_err(&d40c->chan.dev->device,
554 "[%s]: unable to suspend the chl %d (log: %d) status %x\n",
555 __func__, d40c->phy_chan->num, d40c->log_num,
563 spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
567 static void d40_term_all(struct d40_chan *d40c)
569 struct d40_desc *d40d;
572 /* Release active descriptors */
573 while ((d40d = d40_first_active_get(d40c))) {
574 d40_desc_remove(d40d);
576 /* Return desc to free-list */
577 d40_desc_free(d40c, d40d);
580 /* Release queued descriptors waiting for transfer */
581 while ((d40d = d40_first_queued(d40c))) {
582 d40_desc_remove(d40d);
584 /* Return desc to free-list */
585 d40_desc_free(d40c, d40d);
588 spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
590 d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num] &=
591 (~(0x1 << d40c->lcla.dst_id));
592 d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num] &=
593 (~(0x1 << d40c->lcla.src_id));
595 d40c->lcla.src_id = -1;
596 d40c->lcla.dst_id = -1;
598 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
600 d40c->pending_tx = 0;
604 static void d40_config_set_event(struct d40_chan *d40c, bool do_enable)
609 /* Notice, that disable requires the physical channel to be stopped */
611 val = D40_ACTIVATE_EVENTLINE;
613 val = D40_DEACTIVATE_EVENTLINE;
615 spin_lock_irqsave(&d40c->phy_chan->lock, flags);
617 /* Enable event line connected to device (or memcpy) */
618 if ((d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) ||
619 (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH)) {
620 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
622 writel((val << D40_EVENTLINE_POS(event)) |
623 ~D40_EVENTLINE_MASK(event),
624 d40c->base->virtbase + D40_DREG_PCBASE +
625 d40c->phy_chan->num * D40_DREG_PCDELTA +
628 if (d40c->dma_cfg.dir != STEDMA40_PERIPH_TO_MEM) {
629 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
631 writel((val << D40_EVENTLINE_POS(event)) |
632 ~D40_EVENTLINE_MASK(event),
633 d40c->base->virtbase + D40_DREG_PCBASE +
634 d40c->phy_chan->num * D40_DREG_PCDELTA +
638 spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
641 static u32 d40_chan_has_events(struct d40_chan *d40c)
645 val = readl(d40c->base->virtbase + D40_DREG_PCBASE +
646 d40c->phy_chan->num * D40_DREG_PCDELTA +
649 val |= readl(d40c->base->virtbase + D40_DREG_PCBASE +
650 d40c->phy_chan->num * D40_DREG_PCDELTA +
655 static void d40_config_write(struct d40_chan *d40c)
660 /* Odd addresses are even addresses + 4 */
661 addr_base = (d40c->phy_chan->num % 2) * 4;
662 /* Setup channel mode to logical or physical */
663 var = ((u32)(d40c->log_num != D40_PHY_CHAN) + 1) <<
664 D40_CHAN_POS(d40c->phy_chan->num);
665 writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
667 /* Setup operational mode option register */
668 var = ((d40c->dma_cfg.channel_type >> STEDMA40_INFO_CH_MODE_OPT_POS) &
669 0x3) << D40_CHAN_POS(d40c->phy_chan->num);
671 writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
673 if (d40c->log_num != D40_PHY_CHAN) {
674 /* Set default config for CFG reg */
675 writel(d40c->src_def_cfg,
676 d40c->base->virtbase + D40_DREG_PCBASE +
677 d40c->phy_chan->num * D40_DREG_PCDELTA +
679 writel(d40c->dst_def_cfg,
680 d40c->base->virtbase + D40_DREG_PCBASE +
681 d40c->phy_chan->num * D40_DREG_PCDELTA +
684 /* Set LIDX for lcla */
685 writel((d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS) &
686 D40_SREG_ELEM_LOG_LIDX_MASK,
687 d40c->base->virtbase + D40_DREG_PCBASE +
688 d40c->phy_chan->num * D40_DREG_PCDELTA +
691 writel((d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS) &
692 D40_SREG_ELEM_LOG_LIDX_MASK,
693 d40c->base->virtbase + D40_DREG_PCBASE +
694 d40c->phy_chan->num * D40_DREG_PCDELTA +
700 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
702 if (d40d->lli_phy.dst && d40d->lli_phy.src) {
703 d40_phy_lli_write(d40c->base->virtbase,
707 } else if (d40d->lli_log.dst && d40d->lli_log.src) {
708 struct d40_log_lli *src = d40d->lli_log.src;
709 struct d40_log_lli *dst = d40d->lli_log.dst;
712 src += d40d->lli_count;
713 dst += d40d->lli_count;
714 s = d40_log_lli_write(d40c->lcpa,
715 d40c->lcla.src, d40c->lcla.dst,
717 d40c->base->plat_data->llis_per_log);
719 /* If s equals to zero, the job is not linked */
721 (void) dma_map_single(d40c->base->dev, d40c->lcla.src,
722 s * sizeof(struct d40_log_lli),
724 (void) dma_map_single(d40c->base->dev, d40c->lcla.dst,
725 s * sizeof(struct d40_log_lli),
729 d40d->lli_count += d40d->lli_tx_len;
732 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
734 struct d40_chan *d40c = container_of(tx->chan,
737 struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
740 spin_lock_irqsave(&d40c->lock, flags);
742 tx->cookie = d40_assign_cookie(d40c, d40d);
744 d40_desc_queue(d40c, d40d);
746 spin_unlock_irqrestore(&d40c->lock, flags);
751 static int d40_start(struct d40_chan *d40c)
753 if (d40c->base->rev == 0) {
756 if (d40c->log_num != D40_PHY_CHAN) {
757 err = d40_channel_execute_command(d40c,
758 D40_DMA_SUSPEND_REQ);
764 if (d40c->log_num != D40_PHY_CHAN)
765 d40_config_set_event(d40c, true);
767 return d40_channel_execute_command(d40c, D40_DMA_RUN);
770 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
772 struct d40_desc *d40d;
775 /* Start queued jobs, if any */
776 d40d = d40_first_queued(d40c);
781 /* Remove from queue */
782 d40_desc_remove(d40d);
784 /* Add to active queue */
785 d40_desc_submit(d40c, d40d);
787 /* Initiate DMA job */
788 d40_desc_load(d40c, d40d);
791 err = d40_start(d40c);
800 /* called from interrupt context */
801 static void dma_tc_handle(struct d40_chan *d40c)
803 struct d40_desc *d40d;
808 /* Get first active entry from list */
809 d40d = d40_first_active_get(d40c);
814 if (d40d->lli_count < d40d->lli_len) {
816 d40_desc_load(d40c, d40d);
818 (void) d40_start(d40c);
822 if (d40_queue_start(d40c) == NULL)
826 tasklet_schedule(&d40c->tasklet);
830 static void dma_tasklet(unsigned long data)
832 struct d40_chan *d40c = (struct d40_chan *) data;
833 struct d40_desc *d40d_fin;
835 dma_async_tx_callback callback;
836 void *callback_param;
838 spin_lock_irqsave(&d40c->lock, flags);
840 /* Get first active entry from list */
841 d40d_fin = d40_first_active_get(d40c);
843 if (d40d_fin == NULL)
846 d40c->completed = d40d_fin->txd.cookie;
849 * If terminating a channel pending_tx is set to zero.
850 * This prevents any finished active jobs to return to the client.
852 if (d40c->pending_tx == 0) {
853 spin_unlock_irqrestore(&d40c->lock, flags);
857 /* Callback to client */
858 callback = d40d_fin->txd.callback;
859 callback_param = d40d_fin->txd.callback_param;
861 if (async_tx_test_ack(&d40d_fin->txd)) {
862 d40_pool_lli_free(d40d_fin);
863 d40_desc_remove(d40d_fin);
864 /* Return desc to free-list */
865 d40_desc_free(d40c, d40d_fin);
867 if (!d40d_fin->is_in_client_list) {
868 d40_desc_remove(d40d_fin);
869 list_add_tail(&d40d_fin->node, &d40c->client);
870 d40d_fin->is_in_client_list = true;
876 if (d40c->pending_tx)
877 tasklet_schedule(&d40c->tasklet);
879 spin_unlock_irqrestore(&d40c->lock, flags);
881 if (callback && (d40d_fin->txd.flags & DMA_PREP_INTERRUPT))
882 callback(callback_param);
887 /* Rescue manouver if receiving double interrupts */
888 if (d40c->pending_tx > 0)
890 spin_unlock_irqrestore(&d40c->lock, flags);
893 static irqreturn_t d40_handle_interrupt(int irq, void *data)
895 static const struct d40_interrupt_lookup il[] = {
896 {D40_DREG_LCTIS0, D40_DREG_LCICR0, false, 0},
897 {D40_DREG_LCTIS1, D40_DREG_LCICR1, false, 32},
898 {D40_DREG_LCTIS2, D40_DREG_LCICR2, false, 64},
899 {D40_DREG_LCTIS3, D40_DREG_LCICR3, false, 96},
900 {D40_DREG_LCEIS0, D40_DREG_LCICR0, true, 0},
901 {D40_DREG_LCEIS1, D40_DREG_LCICR1, true, 32},
902 {D40_DREG_LCEIS2, D40_DREG_LCICR2, true, 64},
903 {D40_DREG_LCEIS3, D40_DREG_LCICR3, true, 96},
904 {D40_DREG_PCTIS, D40_DREG_PCICR, false, D40_PHY_CHAN},
905 {D40_DREG_PCEIS, D40_DREG_PCICR, true, D40_PHY_CHAN},
909 u32 regs[ARRAY_SIZE(il)];
913 struct d40_chan *d40c;
915 struct d40_base *base = data;
917 spin_lock_irqsave(&base->interrupt_lock, flags);
919 /* Read interrupt status of both logical and physical channels */
920 for (i = 0; i < ARRAY_SIZE(il); i++)
921 regs[i] = readl(base->virtbase + il[i].src);
925 chan = find_next_bit((unsigned long *)regs,
926 BITS_PER_LONG * ARRAY_SIZE(il), chan + 1);
928 /* No more set bits found? */
929 if (chan == BITS_PER_LONG * ARRAY_SIZE(il))
932 row = chan / BITS_PER_LONG;
933 idx = chan & (BITS_PER_LONG - 1);
936 writel(1 << idx, base->virtbase + il[row].clr);
938 if (il[row].offset == D40_PHY_CHAN)
939 d40c = base->lookup_phy_chans[idx];
941 d40c = base->lookup_log_chans[il[row].offset + idx];
942 spin_lock(&d40c->lock);
944 if (!il[row].is_error)
948 "[%s] IRQ chan: %ld offset %d idx %d\n",
949 __func__, chan, il[row].offset, idx);
951 spin_unlock(&d40c->lock);
954 spin_unlock_irqrestore(&base->interrupt_lock, flags);
960 static int d40_validate_conf(struct d40_chan *d40c,
961 struct stedma40_chan_cfg *conf)
964 u32 dst_event_group = D40_TYPE_TO_GROUP(conf->dst_dev_type);
965 u32 src_event_group = D40_TYPE_TO_GROUP(conf->src_dev_type);
966 bool is_log = (conf->channel_type & STEDMA40_CHANNEL_IN_OPER_MODE)
967 == STEDMA40_CHANNEL_IN_LOG_MODE;
970 dev_err(&d40c->chan.dev->device, "[%s] Invalid direction.\n",
975 if (conf->dst_dev_type != STEDMA40_DEV_DST_MEMORY &&
976 d40c->base->plat_data->dev_tx[conf->dst_dev_type] == 0 &&
977 d40c->runtime_addr == 0) {
979 dev_err(&d40c->chan.dev->device,
980 "[%s] Invalid TX channel address (%d)\n",
981 __func__, conf->dst_dev_type);
985 if (conf->src_dev_type != STEDMA40_DEV_SRC_MEMORY &&
986 d40c->base->plat_data->dev_rx[conf->src_dev_type] == 0 &&
987 d40c->runtime_addr == 0) {
988 dev_err(&d40c->chan.dev->device,
989 "[%s] Invalid RX channel address (%d)\n",
990 __func__, conf->src_dev_type);
994 if (conf->dir == STEDMA40_MEM_TO_PERIPH &&
995 dst_event_group == STEDMA40_DEV_DST_MEMORY) {
996 dev_err(&d40c->chan.dev->device, "[%s] Invalid dst\n",
1001 if (conf->dir == STEDMA40_PERIPH_TO_MEM &&
1002 src_event_group == STEDMA40_DEV_SRC_MEMORY) {
1003 dev_err(&d40c->chan.dev->device, "[%s] Invalid src\n",
1008 if (src_event_group == STEDMA40_DEV_SRC_MEMORY &&
1009 dst_event_group == STEDMA40_DEV_DST_MEMORY && is_log) {
1010 dev_err(&d40c->chan.dev->device,
1011 "[%s] No event line\n", __func__);
1015 if (conf->dir == STEDMA40_PERIPH_TO_PERIPH &&
1016 (src_event_group != dst_event_group)) {
1017 dev_err(&d40c->chan.dev->device,
1018 "[%s] Invalid event group\n", __func__);
1022 if (conf->dir == STEDMA40_PERIPH_TO_PERIPH) {
1024 * DMAC HW supports it. Will be added to this driver,
1025 * in case any dma client requires it.
1027 dev_err(&d40c->chan.dev->device,
1028 "[%s] periph to periph not supported\n",
1036 static bool d40_alloc_mask_set(struct d40_phy_res *phy, bool is_src,
1037 int log_event_line, bool is_log)
1039 unsigned long flags;
1040 spin_lock_irqsave(&phy->lock, flags);
1042 /* Physical interrupts are masked per physical full channel */
1043 if (phy->allocated_src == D40_ALLOC_FREE &&
1044 phy->allocated_dst == D40_ALLOC_FREE) {
1045 phy->allocated_dst = D40_ALLOC_PHY;
1046 phy->allocated_src = D40_ALLOC_PHY;
1052 /* Logical channel */
1054 if (phy->allocated_src == D40_ALLOC_PHY)
1057 if (phy->allocated_src == D40_ALLOC_FREE)
1058 phy->allocated_src = D40_ALLOC_LOG_FREE;
1060 if (!(phy->allocated_src & (1 << log_event_line))) {
1061 phy->allocated_src |= 1 << log_event_line;
1066 if (phy->allocated_dst == D40_ALLOC_PHY)
1069 if (phy->allocated_dst == D40_ALLOC_FREE)
1070 phy->allocated_dst = D40_ALLOC_LOG_FREE;
1072 if (!(phy->allocated_dst & (1 << log_event_line))) {
1073 phy->allocated_dst |= 1 << log_event_line;
1080 spin_unlock_irqrestore(&phy->lock, flags);
1083 spin_unlock_irqrestore(&phy->lock, flags);
1087 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1090 unsigned long flags;
1091 bool is_free = false;
1093 spin_lock_irqsave(&phy->lock, flags);
1094 if (!log_event_line) {
1095 /* Physical interrupts are masked per physical full channel */
1096 phy->allocated_dst = D40_ALLOC_FREE;
1097 phy->allocated_src = D40_ALLOC_FREE;
1102 /* Logical channel */
1104 phy->allocated_src &= ~(1 << log_event_line);
1105 if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1106 phy->allocated_src = D40_ALLOC_FREE;
1108 phy->allocated_dst &= ~(1 << log_event_line);
1109 if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1110 phy->allocated_dst = D40_ALLOC_FREE;
1113 is_free = ((phy->allocated_src | phy->allocated_dst) ==
1117 spin_unlock_irqrestore(&phy->lock, flags);
1122 static int d40_allocate_channel(struct d40_chan *d40c)
1127 struct d40_phy_res *phys;
1132 bool is_log = (d40c->dma_cfg.channel_type &
1133 STEDMA40_CHANNEL_IN_OPER_MODE)
1134 == STEDMA40_CHANNEL_IN_LOG_MODE;
1137 phys = d40c->base->phy_res;
1139 if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1140 dev_type = d40c->dma_cfg.src_dev_type;
1141 log_num = 2 * dev_type;
1143 } else if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1144 d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1145 /* dst event lines are used for logical memcpy */
1146 dev_type = d40c->dma_cfg.dst_dev_type;
1147 log_num = 2 * dev_type + 1;
1152 event_group = D40_TYPE_TO_GROUP(dev_type);
1153 event_line = D40_TYPE_TO_EVENT(dev_type);
1156 if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1157 /* Find physical half channel */
1158 for (i = 0; i < d40c->base->num_phy_chans; i++) {
1160 if (d40_alloc_mask_set(&phys[i], is_src,
1165 for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1166 int phy_num = j + event_group * 2;
1167 for (i = phy_num; i < phy_num + 2; i++) {
1168 if (d40_alloc_mask_set(&phys[i],
1177 d40c->phy_chan = &phys[i];
1178 d40c->log_num = D40_PHY_CHAN;
1184 /* Find logical channel */
1185 for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1186 int phy_num = j + event_group * 2;
1188 * Spread logical channels across all available physical rather
1189 * than pack every logical channel at the first available phy
1193 for (i = phy_num; i < phy_num + 2; i++) {
1194 if (d40_alloc_mask_set(&phys[i], is_src,
1195 event_line, is_log))
1199 for (i = phy_num + 1; i >= phy_num; i--) {
1200 if (d40_alloc_mask_set(&phys[i], is_src,
1201 event_line, is_log))
1209 d40c->phy_chan = &phys[i];
1210 d40c->log_num = log_num;
1214 d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1216 d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1222 static int d40_config_memcpy(struct d40_chan *d40c)
1224 dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1226 if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1227 d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_log;
1228 d40c->dma_cfg.src_dev_type = STEDMA40_DEV_SRC_MEMORY;
1229 d40c->dma_cfg.dst_dev_type = d40c->base->plat_data->
1230 memcpy[d40c->chan.chan_id];
1232 } else if (dma_has_cap(DMA_MEMCPY, cap) &&
1233 dma_has_cap(DMA_SLAVE, cap)) {
1234 d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_phy;
1236 dev_err(&d40c->chan.dev->device, "[%s] No memcpy\n",
1245 static int d40_free_dma(struct d40_chan *d40c)
1250 struct d40_phy_res *phy = d40c->phy_chan;
1253 struct d40_desc *_d;
1256 /* Terminate all queued and active transfers */
1259 /* Release client owned descriptors */
1260 if (!list_empty(&d40c->client))
1261 list_for_each_entry_safe(d, _d, &d40c->client, node) {
1262 d40_pool_lli_free(d);
1264 /* Return desc to free-list */
1265 d40_desc_free(d40c, d);
1269 dev_err(&d40c->chan.dev->device, "[%s] phy == null\n",
1274 if (phy->allocated_src == D40_ALLOC_FREE &&
1275 phy->allocated_dst == D40_ALLOC_FREE) {
1276 dev_err(&d40c->chan.dev->device, "[%s] channel already free\n",
1281 if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1282 d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1283 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1285 } else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1286 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1289 dev_err(&d40c->chan.dev->device,
1290 "[%s] Unknown direction\n", __func__);
1294 res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1296 dev_err(&d40c->chan.dev->device, "[%s] suspend failed\n",
1301 if (d40c->log_num != D40_PHY_CHAN) {
1302 /* Release logical channel, deactivate the event line */
1304 d40_config_set_event(d40c, false);
1305 d40c->base->lookup_log_chans[d40c->log_num] = NULL;
1308 * Check if there are more logical allocation
1309 * on this phy channel.
1311 if (!d40_alloc_mask_free(phy, is_src, event)) {
1312 /* Resume the other logical channels if any */
1313 if (d40_chan_has_events(d40c)) {
1314 res = d40_channel_execute_command(d40c,
1317 dev_err(&d40c->chan.dev->device,
1318 "[%s] Executing RUN command\n",
1326 (void) d40_alloc_mask_free(phy, is_src, 0);
1329 /* Release physical channel */
1330 res = d40_channel_execute_command(d40c, D40_DMA_STOP);
1332 dev_err(&d40c->chan.dev->device,
1333 "[%s] Failed to stop channel\n", __func__);
1336 d40c->phy_chan = NULL;
1337 /* Invalidate channel type */
1338 d40c->dma_cfg.channel_type = 0;
1339 d40c->base->lookup_phy_chans[phy->num] = NULL;
1344 static int d40_pause(struct dma_chan *chan)
1346 struct d40_chan *d40c =
1347 container_of(chan, struct d40_chan, chan);
1349 unsigned long flags;
1351 spin_lock_irqsave(&d40c->lock, flags);
1353 res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1355 if (d40c->log_num != D40_PHY_CHAN) {
1356 d40_config_set_event(d40c, false);
1357 /* Resume the other logical channels if any */
1358 if (d40_chan_has_events(d40c))
1359 res = d40_channel_execute_command(d40c,
1364 spin_unlock_irqrestore(&d40c->lock, flags);
1368 static bool d40_is_paused(struct d40_chan *d40c)
1370 bool is_paused = false;
1371 unsigned long flags;
1372 void __iomem *active_reg;
1376 spin_lock_irqsave(&d40c->lock, flags);
1378 if (d40c->log_num == D40_PHY_CHAN) {
1379 if (d40c->phy_chan->num % 2 == 0)
1380 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1382 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1384 status = (readl(active_reg) &
1385 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1386 D40_CHAN_POS(d40c->phy_chan->num);
1387 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1393 if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1394 d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM)
1395 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1396 else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM)
1397 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1399 dev_err(&d40c->chan.dev->device,
1400 "[%s] Unknown direction\n", __func__);
1403 status = d40_chan_has_events(d40c);
1404 status = (status & D40_EVENTLINE_MASK(event)) >>
1405 D40_EVENTLINE_POS(event);
1407 if (status != D40_DMA_RUN)
1410 spin_unlock_irqrestore(&d40c->lock, flags);
1416 static bool d40_tx_is_linked(struct d40_chan *d40c)
1420 if (d40c->log_num != D40_PHY_CHAN)
1421 is_link = readl(&d40c->lcpa->lcsp3) & D40_MEM_LCSP3_DLOS_MASK;
1423 is_link = readl(d40c->base->virtbase + D40_DREG_PCBASE +
1424 d40c->phy_chan->num * D40_DREG_PCDELTA +
1425 D40_CHAN_REG_SDLNK) &
1426 D40_SREG_LNK_PHYS_LNK_MASK;
1430 static u32 d40_residue(struct d40_chan *d40c)
1434 if (d40c->log_num != D40_PHY_CHAN)
1435 num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
1436 >> D40_MEM_LCSP2_ECNT_POS;
1438 num_elt = (readl(d40c->base->virtbase + D40_DREG_PCBASE +
1439 d40c->phy_chan->num * D40_DREG_PCDELTA +
1440 D40_CHAN_REG_SDELT) &
1441 D40_SREG_ELEM_PHY_ECNT_MASK) >>
1442 D40_SREG_ELEM_PHY_ECNT_POS;
1443 return num_elt * (1 << d40c->dma_cfg.dst_info.data_width);
1446 static int d40_resume(struct dma_chan *chan)
1448 struct d40_chan *d40c =
1449 container_of(chan, struct d40_chan, chan);
1451 unsigned long flags;
1453 spin_lock_irqsave(&d40c->lock, flags);
1455 if (d40c->base->rev == 0)
1456 if (d40c->log_num != D40_PHY_CHAN) {
1457 res = d40_channel_execute_command(d40c,
1458 D40_DMA_SUSPEND_REQ);
1462 /* If bytes left to transfer or linked tx resume job */
1463 if (d40_residue(d40c) || d40_tx_is_linked(d40c)) {
1464 if (d40c->log_num != D40_PHY_CHAN)
1465 d40_config_set_event(d40c, true);
1466 res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1470 spin_unlock_irqrestore(&d40c->lock, flags);
1474 static u32 stedma40_residue(struct dma_chan *chan)
1476 struct d40_chan *d40c =
1477 container_of(chan, struct d40_chan, chan);
1479 unsigned long flags;
1481 spin_lock_irqsave(&d40c->lock, flags);
1482 bytes_left = d40_residue(d40c);
1483 spin_unlock_irqrestore(&d40c->lock, flags);
1488 /* Public DMA functions in addition to the DMA engine framework */
1490 int stedma40_set_psize(struct dma_chan *chan,
1494 struct d40_chan *d40c =
1495 container_of(chan, struct d40_chan, chan);
1496 unsigned long flags;
1498 spin_lock_irqsave(&d40c->lock, flags);
1500 if (d40c->log_num != D40_PHY_CHAN) {
1501 d40c->log_def.lcsp1 &= ~D40_MEM_LCSP1_SCFG_PSIZE_MASK;
1502 d40c->log_def.lcsp3 &= ~D40_MEM_LCSP1_SCFG_PSIZE_MASK;
1503 d40c->log_def.lcsp1 |= src_psize <<
1504 D40_MEM_LCSP1_SCFG_PSIZE_POS;
1505 d40c->log_def.lcsp3 |= dst_psize <<
1506 D40_MEM_LCSP1_SCFG_PSIZE_POS;
1510 if (src_psize == STEDMA40_PSIZE_PHY_1)
1511 d40c->src_def_cfg &= ~(1 << D40_SREG_CFG_PHY_PEN_POS);
1513 d40c->src_def_cfg |= 1 << D40_SREG_CFG_PHY_PEN_POS;
1514 d40c->src_def_cfg &= ~(STEDMA40_PSIZE_PHY_16 <<
1515 D40_SREG_CFG_PSIZE_POS);
1516 d40c->src_def_cfg |= src_psize << D40_SREG_CFG_PSIZE_POS;
1519 if (dst_psize == STEDMA40_PSIZE_PHY_1)
1520 d40c->dst_def_cfg &= ~(1 << D40_SREG_CFG_PHY_PEN_POS);
1522 d40c->dst_def_cfg |= 1 << D40_SREG_CFG_PHY_PEN_POS;
1523 d40c->dst_def_cfg &= ~(STEDMA40_PSIZE_PHY_16 <<
1524 D40_SREG_CFG_PSIZE_POS);
1525 d40c->dst_def_cfg |= dst_psize << D40_SREG_CFG_PSIZE_POS;
1528 spin_unlock_irqrestore(&d40c->lock, flags);
1531 EXPORT_SYMBOL(stedma40_set_psize);
1533 struct dma_async_tx_descriptor *stedma40_memcpy_sg(struct dma_chan *chan,
1534 struct scatterlist *sgl_dst,
1535 struct scatterlist *sgl_src,
1536 unsigned int sgl_len,
1537 unsigned long dma_flags)
1540 struct d40_desc *d40d;
1541 struct d40_chan *d40c = container_of(chan, struct d40_chan,
1543 unsigned long flags;
1545 if (d40c->phy_chan == NULL) {
1546 dev_err(&d40c->chan.dev->device,
1547 "[%s] Unallocated channel.\n", __func__);
1548 return ERR_PTR(-EINVAL);
1551 spin_lock_irqsave(&d40c->lock, flags);
1552 d40d = d40_desc_get(d40c);
1557 d40d->lli_len = sgl_len;
1558 d40d->lli_tx_len = d40d->lli_len;
1559 d40d->txd.flags = dma_flags;
1561 if (d40c->log_num != D40_PHY_CHAN) {
1562 if (d40d->lli_len > d40c->base->plat_data->llis_per_log)
1563 d40d->lli_tx_len = d40c->base->plat_data->llis_per_log;
1567 * Check if there is space available in lcla. If not,
1568 * split list into 1-length and run only in lcpa
1571 if (d40_lcla_id_get(d40c) != 0)
1572 d40d->lli_tx_len = 1;
1574 if (d40_pool_lli_alloc(d40d, sgl_len, true) < 0) {
1575 dev_err(&d40c->chan.dev->device,
1576 "[%s] Out of memory\n", __func__);
1580 (void) d40_log_sg_to_lli(d40c->lcla.src_id,
1584 d40c->log_def.lcsp1,
1585 d40c->dma_cfg.src_info.data_width,
1587 d40c->base->plat_data->llis_per_log);
1589 (void) d40_log_sg_to_lli(d40c->lcla.dst_id,
1593 d40c->log_def.lcsp3,
1594 d40c->dma_cfg.dst_info.data_width,
1596 d40c->base->plat_data->llis_per_log);
1600 if (d40_pool_lli_alloc(d40d, sgl_len, false) < 0) {
1601 dev_err(&d40c->chan.dev->device,
1602 "[%s] Out of memory\n", __func__);
1606 res = d40_phy_sg_to_lli(sgl_src,
1610 d40d->lli_phy.src_addr,
1612 d40c->dma_cfg.src_info.data_width,
1613 d40c->dma_cfg.src_info.psize);
1618 res = d40_phy_sg_to_lli(sgl_dst,
1622 d40d->lli_phy.dst_addr,
1624 d40c->dma_cfg.dst_info.data_width,
1625 d40c->dma_cfg.dst_info.psize);
1630 (void) dma_map_single(d40c->base->dev, d40d->lli_phy.src,
1631 d40d->lli_pool.size, DMA_TO_DEVICE);
1634 dma_async_tx_descriptor_init(&d40d->txd, chan);
1636 d40d->txd.tx_submit = d40_tx_submit;
1638 spin_unlock_irqrestore(&d40c->lock, flags);
1642 spin_unlock_irqrestore(&d40c->lock, flags);
1645 EXPORT_SYMBOL(stedma40_memcpy_sg);
1647 bool stedma40_filter(struct dma_chan *chan, void *data)
1649 struct stedma40_chan_cfg *info = data;
1650 struct d40_chan *d40c =
1651 container_of(chan, struct d40_chan, chan);
1655 err = d40_validate_conf(d40c, info);
1657 d40c->dma_cfg = *info;
1659 err = d40_config_memcpy(d40c);
1663 EXPORT_SYMBOL(stedma40_filter);
1665 /* DMA ENGINE functions */
1666 static int d40_alloc_chan_resources(struct dma_chan *chan)
1669 unsigned long flags;
1670 struct d40_chan *d40c =
1671 container_of(chan, struct d40_chan, chan);
1673 spin_lock_irqsave(&d40c->lock, flags);
1675 d40c->completed = chan->cookie = 1;
1678 * If no dma configuration is set (channel_type == 0)
1679 * use default configuration (memcpy)
1681 if (d40c->dma_cfg.channel_type == 0) {
1682 err = d40_config_memcpy(d40c);
1684 dev_err(&d40c->chan.dev->device,
1685 "[%s] Failed to configure memcpy channel\n",
1690 is_free_phy = (d40c->phy_chan == NULL);
1692 err = d40_allocate_channel(d40c);
1694 dev_err(&d40c->chan.dev->device,
1695 "[%s] Failed to allocate channel\n", __func__);
1699 /* Fill in basic CFG register values */
1700 d40_phy_cfg(&d40c->dma_cfg, &d40c->src_def_cfg,
1701 &d40c->dst_def_cfg, d40c->log_num != D40_PHY_CHAN);
1703 if (d40c->log_num != D40_PHY_CHAN) {
1704 d40_log_cfg(&d40c->dma_cfg,
1705 &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
1707 if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM)
1708 d40c->lcpa = d40c->base->lcpa_base +
1709 d40c->dma_cfg.src_dev_type * D40_LCPA_CHAN_SIZE;
1711 d40c->lcpa = d40c->base->lcpa_base +
1712 d40c->dma_cfg.dst_dev_type *
1713 D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
1717 * Only write channel configuration to the DMA if the physical
1718 * resource is free. In case of multiple logical channels
1719 * on the same physical resource, only the first write is necessary.
1722 d40_config_write(d40c);
1724 spin_unlock_irqrestore(&d40c->lock, flags);
1728 static void d40_free_chan_resources(struct dma_chan *chan)
1730 struct d40_chan *d40c =
1731 container_of(chan, struct d40_chan, chan);
1733 unsigned long flags;
1735 if (d40c->phy_chan == NULL) {
1736 dev_err(&d40c->chan.dev->device,
1737 "[%s] Cannot free unallocated channel\n", __func__);
1742 spin_lock_irqsave(&d40c->lock, flags);
1744 err = d40_free_dma(d40c);
1747 dev_err(&d40c->chan.dev->device,
1748 "[%s] Failed to free channel\n", __func__);
1749 spin_unlock_irqrestore(&d40c->lock, flags);
1752 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
1756 unsigned long dma_flags)
1758 struct d40_desc *d40d;
1759 struct d40_chan *d40c = container_of(chan, struct d40_chan,
1761 unsigned long flags;
1764 if (d40c->phy_chan == NULL) {
1765 dev_err(&d40c->chan.dev->device,
1766 "[%s] Channel is not allocated.\n", __func__);
1767 return ERR_PTR(-EINVAL);
1770 spin_lock_irqsave(&d40c->lock, flags);
1771 d40d = d40_desc_get(d40c);
1774 dev_err(&d40c->chan.dev->device,
1775 "[%s] Descriptor is NULL\n", __func__);
1779 d40d->txd.flags = dma_flags;
1781 dma_async_tx_descriptor_init(&d40d->txd, chan);
1783 d40d->txd.tx_submit = d40_tx_submit;
1785 if (d40c->log_num != D40_PHY_CHAN) {
1787 if (d40_pool_lli_alloc(d40d, 1, true) < 0) {
1788 dev_err(&d40c->chan.dev->device,
1789 "[%s] Out of memory\n", __func__);
1793 d40d->lli_tx_len = 1;
1795 d40_log_fill_lli(d40d->lli_log.src,
1799 d40c->log_def.lcsp1,
1800 d40c->dma_cfg.src_info.data_width,
1803 d40_log_fill_lli(d40d->lli_log.dst,
1807 d40c->log_def.lcsp3,
1808 d40c->dma_cfg.dst_info.data_width,
1813 if (d40_pool_lli_alloc(d40d, 1, false) < 0) {
1814 dev_err(&d40c->chan.dev->device,
1815 "[%s] Out of memory\n", __func__);
1819 err = d40_phy_fill_lli(d40d->lli_phy.src,
1822 d40c->dma_cfg.src_info.psize,
1826 d40c->dma_cfg.src_info.data_width,
1831 err = d40_phy_fill_lli(d40d->lli_phy.dst,
1834 d40c->dma_cfg.dst_info.psize,
1838 d40c->dma_cfg.dst_info.data_width,
1844 (void) dma_map_single(d40c->base->dev, d40d->lli_phy.src,
1845 d40d->lli_pool.size, DMA_TO_DEVICE);
1848 spin_unlock_irqrestore(&d40c->lock, flags);
1852 dev_err(&d40c->chan.dev->device,
1853 "[%s] Failed filling in PHY LLI\n", __func__);
1854 d40_pool_lli_free(d40d);
1856 spin_unlock_irqrestore(&d40c->lock, flags);
1860 static int d40_prep_slave_sg_log(struct d40_desc *d40d,
1861 struct d40_chan *d40c,
1862 struct scatterlist *sgl,
1863 unsigned int sg_len,
1864 enum dma_data_direction direction,
1865 unsigned long dma_flags)
1867 dma_addr_t dev_addr = 0;
1870 if (d40_pool_lli_alloc(d40d, sg_len, true) < 0) {
1871 dev_err(&d40c->chan.dev->device,
1872 "[%s] Out of memory\n", __func__);
1876 d40d->lli_len = sg_len;
1877 if (d40d->lli_len <= d40c->base->plat_data->llis_per_log)
1878 d40d->lli_tx_len = d40d->lli_len;
1880 d40d->lli_tx_len = d40c->base->plat_data->llis_per_log;
1884 * Check if there is space available in lcla.
1885 * If not, split list into 1-length and run only
1888 if (d40_lcla_id_get(d40c) != 0)
1889 d40d->lli_tx_len = 1;
1891 if (direction == DMA_FROM_DEVICE)
1892 if (d40c->runtime_addr)
1893 dev_addr = d40c->runtime_addr;
1895 dev_addr = d40c->base->plat_data->dev_rx[d40c->dma_cfg.src_dev_type];
1896 else if (direction == DMA_TO_DEVICE)
1897 if (d40c->runtime_addr)
1898 dev_addr = d40c->runtime_addr;
1900 dev_addr = d40c->base->plat_data->dev_tx[d40c->dma_cfg.dst_dev_type];
1905 total_size = d40_log_sg_to_dev(&d40c->lcla,
1909 d40c->dma_cfg.src_info.data_width,
1910 d40c->dma_cfg.dst_info.data_width,
1912 dev_addr, d40d->lli_tx_len,
1913 d40c->base->plat_data->llis_per_log);
1921 static int d40_prep_slave_sg_phy(struct d40_desc *d40d,
1922 struct d40_chan *d40c,
1923 struct scatterlist *sgl,
1924 unsigned int sgl_len,
1925 enum dma_data_direction direction,
1926 unsigned long dma_flags)
1928 dma_addr_t src_dev_addr;
1929 dma_addr_t dst_dev_addr;
1932 if (d40_pool_lli_alloc(d40d, sgl_len, false) < 0) {
1933 dev_err(&d40c->chan.dev->device,
1934 "[%s] Out of memory\n", __func__);
1938 d40d->lli_len = sgl_len;
1939 d40d->lli_tx_len = sgl_len;
1941 if (direction == DMA_FROM_DEVICE) {
1943 if (d40c->runtime_addr)
1944 src_dev_addr = d40c->runtime_addr;
1946 src_dev_addr = d40c->base->plat_data->dev_rx[d40c->dma_cfg.src_dev_type];
1947 } else if (direction == DMA_TO_DEVICE) {
1948 if (d40c->runtime_addr)
1949 dst_dev_addr = d40c->runtime_addr;
1951 dst_dev_addr = d40c->base->plat_data->dev_tx[d40c->dma_cfg.dst_dev_type];
1956 res = d40_phy_sg_to_lli(sgl,
1960 d40d->lli_phy.src_addr,
1962 d40c->dma_cfg.src_info.data_width,
1963 d40c->dma_cfg.src_info.psize);
1967 res = d40_phy_sg_to_lli(sgl,
1971 d40d->lli_phy.dst_addr,
1973 d40c->dma_cfg.dst_info.data_width,
1974 d40c->dma_cfg.dst_info.psize);
1978 (void) dma_map_single(d40c->base->dev, d40d->lli_phy.src,
1979 d40d->lli_pool.size, DMA_TO_DEVICE);
1983 static struct dma_async_tx_descriptor *d40_prep_slave_sg(struct dma_chan *chan,
1984 struct scatterlist *sgl,
1985 unsigned int sg_len,
1986 enum dma_data_direction direction,
1987 unsigned long dma_flags)
1989 struct d40_desc *d40d;
1990 struct d40_chan *d40c = container_of(chan, struct d40_chan,
1992 unsigned long flags;
1995 if (d40c->phy_chan == NULL) {
1996 dev_err(&d40c->chan.dev->device,
1997 "[%s] Cannot prepare unallocated channel\n", __func__);
1998 return ERR_PTR(-EINVAL);
2001 if (d40c->dma_cfg.pre_transfer)
2002 d40c->dma_cfg.pre_transfer(chan,
2003 d40c->dma_cfg.pre_transfer_data,
2006 spin_lock_irqsave(&d40c->lock, flags);
2007 d40d = d40_desc_get(d40c);
2008 spin_unlock_irqrestore(&d40c->lock, flags);
2013 if (d40c->log_num != D40_PHY_CHAN)
2014 err = d40_prep_slave_sg_log(d40d, d40c, sgl, sg_len,
2015 direction, dma_flags);
2017 err = d40_prep_slave_sg_phy(d40d, d40c, sgl, sg_len,
2018 direction, dma_flags);
2020 dev_err(&d40c->chan.dev->device,
2021 "[%s] Failed to prepare %s slave sg job: %d\n",
2023 d40c->log_num != D40_PHY_CHAN ? "log" : "phy", err);
2027 d40d->txd.flags = dma_flags;
2029 dma_async_tx_descriptor_init(&d40d->txd, chan);
2031 d40d->txd.tx_submit = d40_tx_submit;
2036 static enum dma_status d40_tx_status(struct dma_chan *chan,
2037 dma_cookie_t cookie,
2038 struct dma_tx_state *txstate)
2040 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2041 dma_cookie_t last_used;
2042 dma_cookie_t last_complete;
2045 if (d40c->phy_chan == NULL) {
2046 dev_err(&d40c->chan.dev->device,
2047 "[%s] Cannot read status of unallocated channel\n",
2052 last_complete = d40c->completed;
2053 last_used = chan->cookie;
2055 if (d40_is_paused(d40c))
2058 ret = dma_async_is_complete(cookie, last_complete, last_used);
2060 dma_set_tx_state(txstate, last_complete, last_used,
2061 stedma40_residue(chan));
2066 static void d40_issue_pending(struct dma_chan *chan)
2068 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2069 unsigned long flags;
2071 if (d40c->phy_chan == NULL) {
2072 dev_err(&d40c->chan.dev->device,
2073 "[%s] Channel is not allocated!\n", __func__);
2077 spin_lock_irqsave(&d40c->lock, flags);
2079 /* Busy means that pending jobs are already being processed */
2081 (void) d40_queue_start(d40c);
2083 spin_unlock_irqrestore(&d40c->lock, flags);
2086 /* Runtime reconfiguration extension */
2087 static void d40_set_runtime_config(struct dma_chan *chan,
2088 struct dma_slave_config *config)
2090 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2091 struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2092 enum dma_slave_buswidth config_addr_width;
2093 dma_addr_t config_addr;
2094 u32 config_maxburst;
2095 enum stedma40_periph_data_width addr_width;
2098 if (config->direction == DMA_FROM_DEVICE) {
2099 dma_addr_t dev_addr_rx =
2100 d40c->base->plat_data->dev_rx[cfg->src_dev_type];
2102 config_addr = config->src_addr;
2104 dev_dbg(d40c->base->dev,
2105 "channel has a pre-wired RX address %08x "
2106 "overriding with %08x\n",
2107 dev_addr_rx, config_addr);
2108 if (cfg->dir != STEDMA40_PERIPH_TO_MEM)
2109 dev_dbg(d40c->base->dev,
2110 "channel was not configured for peripheral "
2111 "to memory transfer (%d) overriding\n",
2113 cfg->dir = STEDMA40_PERIPH_TO_MEM;
2115 config_addr_width = config->src_addr_width;
2116 config_maxburst = config->src_maxburst;
2118 } else if (config->direction == DMA_TO_DEVICE) {
2119 dma_addr_t dev_addr_tx =
2120 d40c->base->plat_data->dev_tx[cfg->dst_dev_type];
2122 config_addr = config->dst_addr;
2124 dev_dbg(d40c->base->dev,
2125 "channel has a pre-wired TX address %08x "
2126 "overriding with %08x\n",
2127 dev_addr_tx, config_addr);
2128 if (cfg->dir != STEDMA40_MEM_TO_PERIPH)
2129 dev_dbg(d40c->base->dev,
2130 "channel was not configured for memory "
2131 "to peripheral transfer (%d) overriding\n",
2133 cfg->dir = STEDMA40_MEM_TO_PERIPH;
2135 config_addr_width = config->dst_addr_width;
2136 config_maxburst = config->dst_maxburst;
2139 dev_err(d40c->base->dev,
2140 "unrecognized channel direction %d\n",
2145 switch (config_addr_width) {
2146 case DMA_SLAVE_BUSWIDTH_1_BYTE:
2147 addr_width = STEDMA40_BYTE_WIDTH;
2149 case DMA_SLAVE_BUSWIDTH_2_BYTES:
2150 addr_width = STEDMA40_HALFWORD_WIDTH;
2152 case DMA_SLAVE_BUSWIDTH_4_BYTES:
2153 addr_width = STEDMA40_WORD_WIDTH;
2155 case DMA_SLAVE_BUSWIDTH_8_BYTES:
2156 addr_width = STEDMA40_DOUBLEWORD_WIDTH;
2159 dev_err(d40c->base->dev,
2160 "illegal peripheral address width "
2162 config->src_addr_width);
2166 if (config_maxburst >= 16)
2167 psize = STEDMA40_PSIZE_LOG_16;
2168 else if (config_maxburst >= 8)
2169 psize = STEDMA40_PSIZE_LOG_8;
2170 else if (config_maxburst >= 4)
2171 psize = STEDMA40_PSIZE_LOG_4;
2173 psize = STEDMA40_PSIZE_LOG_1;
2175 /* Set up all the endpoint configs */
2176 cfg->src_info.data_width = addr_width;
2177 cfg->src_info.psize = psize;
2178 cfg->src_info.endianess = STEDMA40_LITTLE_ENDIAN;
2179 cfg->src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2180 cfg->dst_info.data_width = addr_width;
2181 cfg->dst_info.psize = psize;
2182 cfg->dst_info.endianess = STEDMA40_LITTLE_ENDIAN;
2183 cfg->dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2185 /* These settings will take precedence later */
2186 d40c->runtime_addr = config_addr;
2187 d40c->runtime_direction = config->direction;
2188 dev_dbg(d40c->base->dev,
2189 "configured channel %s for %s, data width %d, "
2190 "maxburst %d bytes, LE, no flow control\n",
2191 dma_chan_name(chan),
2192 (config->direction == DMA_FROM_DEVICE) ? "RX" : "TX",
2197 static int d40_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
2200 unsigned long flags;
2201 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2203 if (d40c->phy_chan == NULL) {
2204 dev_err(&d40c->chan.dev->device,
2205 "[%s] Channel is not allocated!\n", __func__);
2210 case DMA_TERMINATE_ALL:
2211 spin_lock_irqsave(&d40c->lock, flags);
2213 spin_unlock_irqrestore(&d40c->lock, flags);
2216 return d40_pause(chan);
2218 return d40_resume(chan);
2219 case DMA_SLAVE_CONFIG:
2220 d40_set_runtime_config(chan,
2221 (struct dma_slave_config *) arg);
2227 /* Other commands are unimplemented */
2231 /* Initialization functions */
2233 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2234 struct d40_chan *chans, int offset,
2238 struct d40_chan *d40c;
2240 INIT_LIST_HEAD(&dma->channels);
2242 for (i = offset; i < offset + num_chans; i++) {
2245 d40c->chan.device = dma;
2247 /* Invalidate lcla element */
2248 d40c->lcla.src_id = -1;
2249 d40c->lcla.dst_id = -1;
2251 spin_lock_init(&d40c->lock);
2253 d40c->log_num = D40_PHY_CHAN;
2255 INIT_LIST_HEAD(&d40c->active);
2256 INIT_LIST_HEAD(&d40c->queue);
2257 INIT_LIST_HEAD(&d40c->client);
2259 tasklet_init(&d40c->tasklet, dma_tasklet,
2260 (unsigned long) d40c);
2262 list_add_tail(&d40c->chan.device_node,
2267 static int __init d40_dmaengine_init(struct d40_base *base,
2268 int num_reserved_chans)
2272 d40_chan_init(base, &base->dma_slave, base->log_chans,
2273 0, base->num_log_chans);
2275 dma_cap_zero(base->dma_slave.cap_mask);
2276 dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2278 base->dma_slave.device_alloc_chan_resources = d40_alloc_chan_resources;
2279 base->dma_slave.device_free_chan_resources = d40_free_chan_resources;
2280 base->dma_slave.device_prep_dma_memcpy = d40_prep_memcpy;
2281 base->dma_slave.device_prep_slave_sg = d40_prep_slave_sg;
2282 base->dma_slave.device_tx_status = d40_tx_status;
2283 base->dma_slave.device_issue_pending = d40_issue_pending;
2284 base->dma_slave.device_control = d40_control;
2285 base->dma_slave.dev = base->dev;
2287 err = dma_async_device_register(&base->dma_slave);
2291 "[%s] Failed to register slave channels\n",
2296 d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2297 base->num_log_chans, base->plat_data->memcpy_len);
2299 dma_cap_zero(base->dma_memcpy.cap_mask);
2300 dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2302 base->dma_memcpy.device_alloc_chan_resources = d40_alloc_chan_resources;
2303 base->dma_memcpy.device_free_chan_resources = d40_free_chan_resources;
2304 base->dma_memcpy.device_prep_dma_memcpy = d40_prep_memcpy;
2305 base->dma_memcpy.device_prep_slave_sg = d40_prep_slave_sg;
2306 base->dma_memcpy.device_tx_status = d40_tx_status;
2307 base->dma_memcpy.device_issue_pending = d40_issue_pending;
2308 base->dma_memcpy.device_control = d40_control;
2309 base->dma_memcpy.dev = base->dev;
2311 * This controller can only access address at even
2312 * 32bit boundaries, i.e. 2^2
2314 base->dma_memcpy.copy_align = 2;
2316 err = dma_async_device_register(&base->dma_memcpy);
2320 "[%s] Failed to regsiter memcpy only channels\n",
2325 d40_chan_init(base, &base->dma_both, base->phy_chans,
2326 0, num_reserved_chans);
2328 dma_cap_zero(base->dma_both.cap_mask);
2329 dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2330 dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2332 base->dma_both.device_alloc_chan_resources = d40_alloc_chan_resources;
2333 base->dma_both.device_free_chan_resources = d40_free_chan_resources;
2334 base->dma_both.device_prep_dma_memcpy = d40_prep_memcpy;
2335 base->dma_both.device_prep_slave_sg = d40_prep_slave_sg;
2336 base->dma_both.device_tx_status = d40_tx_status;
2337 base->dma_both.device_issue_pending = d40_issue_pending;
2338 base->dma_both.device_control = d40_control;
2339 base->dma_both.dev = base->dev;
2340 base->dma_both.copy_align = 2;
2341 err = dma_async_device_register(&base->dma_both);
2345 "[%s] Failed to register logical and physical capable channels\n",
2351 dma_async_device_unregister(&base->dma_memcpy);
2353 dma_async_device_unregister(&base->dma_slave);
2358 /* Initialization functions. */
2360 static int __init d40_phy_res_init(struct d40_base *base)
2363 int num_phy_chans_avail = 0;
2365 int odd_even_bit = -2;
2367 val[0] = readl(base->virtbase + D40_DREG_PRSME);
2368 val[1] = readl(base->virtbase + D40_DREG_PRSMO);
2370 for (i = 0; i < base->num_phy_chans; i++) {
2371 base->phy_res[i].num = i;
2372 odd_even_bit += 2 * ((i % 2) == 0);
2373 if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
2374 /* Mark security only channels as occupied */
2375 base->phy_res[i].allocated_src = D40_ALLOC_PHY;
2376 base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
2378 base->phy_res[i].allocated_src = D40_ALLOC_FREE;
2379 base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
2380 num_phy_chans_avail++;
2382 spin_lock_init(&base->phy_res[i].lock);
2385 /* Mark disabled channels as occupied */
2386 for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
2387 base->phy_res[i].allocated_src = D40_ALLOC_PHY;
2388 base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
2389 num_phy_chans_avail--;
2392 dev_info(base->dev, "%d of %d physical DMA channels available\n",
2393 num_phy_chans_avail, base->num_phy_chans);
2395 /* Verify settings extended vs standard */
2396 val[0] = readl(base->virtbase + D40_DREG_PRTYP);
2398 for (i = 0; i < base->num_phy_chans; i++) {
2400 if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
2401 (val[0] & 0x3) != 1)
2403 "[%s] INFO: channel %d is misconfigured (%d)\n",
2404 __func__, i, val[0] & 0x3);
2406 val[0] = val[0] >> 2;
2409 return num_phy_chans_avail;
2412 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
2414 static const struct d40_reg_val dma_id_regs[] = {
2416 { .reg = D40_DREG_PERIPHID0, .val = 0x0040},
2417 { .reg = D40_DREG_PERIPHID1, .val = 0x0000},
2419 * D40_DREG_PERIPHID2 Depends on HW revision:
2420 * MOP500/HREF ED has 0x0008,
2422 * HREF V1 has 0x0028
2424 { .reg = D40_DREG_PERIPHID3, .val = 0x0000},
2427 { .reg = D40_DREG_CELLID0, .val = 0x000d},
2428 { .reg = D40_DREG_CELLID1, .val = 0x00f0},
2429 { .reg = D40_DREG_CELLID2, .val = 0x0005},
2430 { .reg = D40_DREG_CELLID3, .val = 0x00b1}
2432 struct stedma40_platform_data *plat_data;
2433 struct clk *clk = NULL;
2434 void __iomem *virtbase = NULL;
2435 struct resource *res = NULL;
2436 struct d40_base *base = NULL;
2437 int num_log_chans = 0;
2443 clk = clk_get(&pdev->dev, NULL);
2446 dev_err(&pdev->dev, "[%s] No matching clock found\n",
2453 /* Get IO for DMAC base address */
2454 res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
2458 if (request_mem_region(res->start, resource_size(res),
2459 D40_NAME " I/O base") == NULL)
2462 virtbase = ioremap(res->start, resource_size(res));
2466 /* HW version check */
2467 for (i = 0; i < ARRAY_SIZE(dma_id_regs); i++) {
2468 if (dma_id_regs[i].val !=
2469 readl(virtbase + dma_id_regs[i].reg)) {
2471 "[%s] Unknown hardware! Expected 0x%x at 0x%x but got 0x%x\n",
2475 readl(virtbase + dma_id_regs[i].reg));
2480 /* Get silicon revision and designer */
2481 val = readl(virtbase + D40_DREG_PERIPHID2);
2483 if ((val & D40_DREG_PERIPHID2_DESIGNER_MASK) !=
2486 "[%s] Unknown designer! Got %x wanted %x\n",
2487 __func__, val & D40_DREG_PERIPHID2_DESIGNER_MASK,
2492 rev = (val & D40_DREG_PERIPHID2_REV_MASK) >>
2493 D40_DREG_PERIPHID2_REV_POS;
2495 /* The number of physical channels on this HW */
2496 num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
2498 dev_info(&pdev->dev, "hardware revision: %d @ 0x%x\n",
2501 plat_data = pdev->dev.platform_data;
2503 /* Count the number of logical channels in use */
2504 for (i = 0; i < plat_data->dev_len; i++)
2505 if (plat_data->dev_rx[i] != 0)
2508 for (i = 0; i < plat_data->dev_len; i++)
2509 if (plat_data->dev_tx[i] != 0)
2512 base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
2513 (num_phy_chans + num_log_chans + plat_data->memcpy_len) *
2514 sizeof(struct d40_chan), GFP_KERNEL);
2517 dev_err(&pdev->dev, "[%s] Out of memory\n", __func__);
2523 base->num_phy_chans = num_phy_chans;
2524 base->num_log_chans = num_log_chans;
2525 base->phy_start = res->start;
2526 base->phy_size = resource_size(res);
2527 base->virtbase = virtbase;
2528 base->plat_data = plat_data;
2529 base->dev = &pdev->dev;
2530 base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
2531 base->log_chans = &base->phy_chans[num_phy_chans];
2533 base->phy_res = kzalloc(num_phy_chans * sizeof(struct d40_phy_res),
2538 base->lookup_phy_chans = kzalloc(num_phy_chans *
2539 sizeof(struct d40_chan *),
2541 if (!base->lookup_phy_chans)
2544 if (num_log_chans + plat_data->memcpy_len) {
2546 * The max number of logical channels are event lines for all
2547 * src devices and dst devices
2549 base->lookup_log_chans = kzalloc(plat_data->dev_len * 2 *
2550 sizeof(struct d40_chan *),
2552 if (!base->lookup_log_chans)
2555 base->lcla_pool.alloc_map = kzalloc(num_phy_chans * sizeof(u32),
2557 if (!base->lcla_pool.alloc_map)
2560 base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
2561 0, SLAB_HWCACHE_ALIGN,
2563 if (base->desc_slab == NULL)
2576 release_mem_region(res->start,
2577 resource_size(res));
2582 kfree(base->lcla_pool.alloc_map);
2583 kfree(base->lookup_log_chans);
2584 kfree(base->lookup_phy_chans);
2585 kfree(base->phy_res);
2592 static void __init d40_hw_init(struct d40_base *base)
2595 static const struct d40_reg_val dma_init_reg[] = {
2596 /* Clock every part of the DMA block from start */
2597 { .reg = D40_DREG_GCC, .val = 0x0000ff01},
2599 /* Interrupts on all logical channels */
2600 { .reg = D40_DREG_LCMIS0, .val = 0xFFFFFFFF},
2601 { .reg = D40_DREG_LCMIS1, .val = 0xFFFFFFFF},
2602 { .reg = D40_DREG_LCMIS2, .val = 0xFFFFFFFF},
2603 { .reg = D40_DREG_LCMIS3, .val = 0xFFFFFFFF},
2604 { .reg = D40_DREG_LCICR0, .val = 0xFFFFFFFF},
2605 { .reg = D40_DREG_LCICR1, .val = 0xFFFFFFFF},
2606 { .reg = D40_DREG_LCICR2, .val = 0xFFFFFFFF},
2607 { .reg = D40_DREG_LCICR3, .val = 0xFFFFFFFF},
2608 { .reg = D40_DREG_LCTIS0, .val = 0xFFFFFFFF},
2609 { .reg = D40_DREG_LCTIS1, .val = 0xFFFFFFFF},
2610 { .reg = D40_DREG_LCTIS2, .val = 0xFFFFFFFF},
2611 { .reg = D40_DREG_LCTIS3, .val = 0xFFFFFFFF}
2614 u32 prmseo[2] = {0, 0};
2615 u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
2619 for (i = 0; i < ARRAY_SIZE(dma_init_reg); i++)
2620 writel(dma_init_reg[i].val,
2621 base->virtbase + dma_init_reg[i].reg);
2623 /* Configure all our dma channels to default settings */
2624 for (i = 0; i < base->num_phy_chans; i++) {
2626 activeo[i % 2] = activeo[i % 2] << 2;
2628 if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
2630 activeo[i % 2] |= 3;
2634 /* Enable interrupt # */
2635 pcmis = (pcmis << 1) | 1;
2637 /* Clear interrupt # */
2638 pcicr = (pcicr << 1) | 1;
2640 /* Set channel to physical mode */
2641 prmseo[i % 2] = prmseo[i % 2] << 2;
2646 writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
2647 writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
2648 writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
2649 writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
2651 /* Write which interrupt to enable */
2652 writel(pcmis, base->virtbase + D40_DREG_PCMIS);
2654 /* Write which interrupt to clear */
2655 writel(pcicr, base->virtbase + D40_DREG_PCICR);
2659 static int __init d40_lcla_allocate(struct d40_base *base)
2661 unsigned long *page_list;
2666 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
2667 * To full fill this hardware requirement without wasting 256 kb
2668 * we allocate pages until we get an aligned one.
2670 page_list = kmalloc(sizeof(unsigned long) * MAX_LCLA_ALLOC_ATTEMPTS,
2678 /* Calculating how many pages that are required */
2679 base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
2681 for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
2682 page_list[i] = __get_free_pages(GFP_KERNEL,
2683 base->lcla_pool.pages);
2684 if (!page_list[i]) {
2687 "[%s] Failed to allocate %d pages.\n",
2688 __func__, base->lcla_pool.pages);
2690 for (j = 0; j < i; j++)
2691 free_pages(page_list[j], base->lcla_pool.pages);
2695 if ((virt_to_phys((void *)page_list[i]) &
2696 (LCLA_ALIGNMENT - 1)) == 0)
2700 for (j = 0; j < i; j++)
2701 free_pages(page_list[j], base->lcla_pool.pages);
2703 if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
2704 base->lcla_pool.base = (void *)page_list[i];
2706 /* After many attempts, no succees with finding the correct
2707 * alignment try with allocating a big buffer */
2709 "[%s] Failed to get %d pages @ 18 bit align.\n",
2710 __func__, base->lcla_pool.pages);
2711 base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
2712 base->num_phy_chans +
2715 if (!base->lcla_pool.base_unaligned) {
2720 base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
2724 writel(virt_to_phys(base->lcla_pool.base),
2725 base->virtbase + D40_DREG_LCLA);
2731 static int __init d40_probe(struct platform_device *pdev)
2735 struct d40_base *base;
2736 struct resource *res = NULL;
2737 int num_reserved_chans;
2740 base = d40_hw_detect_init(pdev);
2745 num_reserved_chans = d40_phy_res_init(base);
2747 platform_set_drvdata(pdev, base);
2749 spin_lock_init(&base->interrupt_lock);
2750 spin_lock_init(&base->execmd_lock);
2752 /* Get IO for logical channel parameter address */
2753 res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
2757 "[%s] No \"lcpa\" memory resource\n",
2761 base->lcpa_size = resource_size(res);
2762 base->phy_lcpa = res->start;
2764 if (request_mem_region(res->start, resource_size(res),
2765 D40_NAME " I/O lcpa") == NULL) {
2768 "[%s] Failed to request LCPA region 0x%x-0x%x\n",
2769 __func__, res->start, res->end);
2773 /* We make use of ESRAM memory for this. */
2774 val = readl(base->virtbase + D40_DREG_LCPA);
2775 if (res->start != val && val != 0) {
2776 dev_warn(&pdev->dev,
2777 "[%s] Mismatch LCPA dma 0x%x, def 0x%x\n",
2778 __func__, val, res->start);
2780 writel(res->start, base->virtbase + D40_DREG_LCPA);
2782 base->lcpa_base = ioremap(res->start, resource_size(res));
2783 if (!base->lcpa_base) {
2786 "[%s] Failed to ioremap LCPA region\n",
2791 ret = d40_lcla_allocate(base);
2793 dev_err(&pdev->dev, "[%s] Failed to allocate LCLA area\n",
2798 spin_lock_init(&base->lcla_pool.lock);
2800 base->lcla_pool.num_blocks = base->num_phy_chans;
2802 base->irq = platform_get_irq(pdev, 0);
2804 ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
2807 dev_err(&pdev->dev, "[%s] No IRQ defined\n", __func__);
2811 err = d40_dmaengine_init(base, num_reserved_chans);
2817 dev_info(base->dev, "initialized\n");
2822 if (base->desc_slab)
2823 kmem_cache_destroy(base->desc_slab);
2825 iounmap(base->virtbase);
2826 if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
2827 free_pages((unsigned long)base->lcla_pool.base,
2828 base->lcla_pool.pages);
2829 if (base->lcla_pool.base_unaligned)
2830 kfree(base->lcla_pool.base_unaligned);
2832 release_mem_region(base->phy_lcpa,
2834 if (base->phy_start)
2835 release_mem_region(base->phy_start,
2838 clk_disable(base->clk);
2842 kfree(base->lcla_pool.alloc_map);
2843 kfree(base->lookup_log_chans);
2844 kfree(base->lookup_phy_chans);
2845 kfree(base->phy_res);
2849 dev_err(&pdev->dev, "[%s] probe failed\n", __func__);
2853 static struct platform_driver d40_driver = {
2855 .owner = THIS_MODULE,
2860 int __init stedma40_init(void)
2862 return platform_driver_probe(&d40_driver, d40_probe);
2864 arch_initcall(stedma40_init);