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/kernel.h>
10 #include <linux/slab.h>
11 #include <linux/dmaengine.h>
12 #include <linux/platform_device.h>
13 #include <linux/clk.h>
14 #include <linux/delay.h>
15 #include <linux/err.h>
17 #include <plat/ste_dma40.h>
19 #include "ste_dma40_ll.h"
21 #define D40_NAME "dma40"
23 #define D40_PHY_CHAN -1
25 /* For masking out/in 2 bit channel positions */
26 #define D40_CHAN_POS(chan) (2 * (chan / 2))
27 #define D40_CHAN_POS_MASK(chan) (0x3 << D40_CHAN_POS(chan))
29 /* Maximum iterations taken before giving up suspending a channel */
30 #define D40_SUSPEND_MAX_IT 500
32 /* Hardware requirement on LCLA alignment */
33 #define LCLA_ALIGNMENT 0x40000
35 /* Max number of links per event group */
36 #define D40_LCLA_LINK_PER_EVENT_GRP 128
37 #define D40_LCLA_END D40_LCLA_LINK_PER_EVENT_GRP
39 /* Attempts before giving up to trying to get pages that are aligned */
40 #define MAX_LCLA_ALLOC_ATTEMPTS 256
42 /* Bit markings for allocation map */
43 #define D40_ALLOC_FREE (1 << 31)
44 #define D40_ALLOC_PHY (1 << 30)
45 #define D40_ALLOC_LOG_FREE 0
47 /* Hardware designer of the block */
48 #define D40_HW_DESIGNER 0x8
51 * enum 40_command - The different commands and/or statuses.
53 * @D40_DMA_STOP: DMA channel command STOP or status STOPPED,
54 * @D40_DMA_RUN: The DMA channel is RUNNING of the command RUN.
55 * @D40_DMA_SUSPEND_REQ: Request the DMA to SUSPEND as soon as possible.
56 * @D40_DMA_SUSPENDED: The DMA channel is SUSPENDED.
61 D40_DMA_SUSPEND_REQ = 2,
66 * struct d40_lli_pool - Structure for keeping LLIs in memory
68 * @base: Pointer to memory area when the pre_alloc_lli's are not large
69 * enough, IE bigger than the most common case, 1 dst and 1 src. NULL if
70 * pre_alloc_lli is used.
71 * @dma_addr: DMA address, if mapped
72 * @size: The size in bytes of the memory at base or the size of pre_alloc_lli.
73 * @pre_alloc_lli: Pre allocated area for the most common case of transfers,
74 * one buffer to one buffer.
80 /* Space for dst and src, plus an extra for padding */
81 u8 pre_alloc_lli[3 * sizeof(struct d40_phy_lli)];
85 * struct d40_desc - A descriptor is one DMA job.
87 * @lli_phy: LLI settings for physical channel. Both src and dst=
88 * points into the lli_pool, to base if lli_len > 1 or to pre_alloc_lli if
90 * @lli_log: Same as above but for logical channels.
91 * @lli_pool: The pool with two entries pre-allocated.
92 * @lli_len: Number of llis of current descriptor.
93 * @lli_current: Number of transfered llis.
94 * @lcla_alloc: Number of LCLA entries allocated.
95 * @txd: DMA engine struct. Used for among other things for communication
98 * @is_in_client_list: true if the client owns this descriptor.
101 * This descriptor is used for both logical and physical transfers.
105 struct d40_phy_lli_bidir lli_phy;
107 struct d40_log_lli_bidir lli_log;
109 struct d40_lli_pool lli_pool;
114 struct dma_async_tx_descriptor txd;
115 struct list_head node;
117 bool is_in_client_list;
121 * struct d40_lcla_pool - LCLA pool settings and data.
123 * @base: The virtual address of LCLA. 18 bit aligned.
124 * @base_unaligned: The orignal kmalloc pointer, if kmalloc is used.
125 * This pointer is only there for clean-up on error.
126 * @pages: The number of pages needed for all physical channels.
127 * Only used later for clean-up on error
128 * @lock: Lock to protect the content in this struct.
129 * @alloc_map: big map over which LCLA entry is own by which job.
131 struct d40_lcla_pool {
134 void *base_unaligned;
137 struct d40_desc **alloc_map;
141 * struct d40_phy_res - struct for handling eventlines mapped to physical
144 * @lock: A lock protection this entity.
145 * @num: The physical channel number of this entity.
146 * @allocated_src: Bit mapped to show which src event line's are mapped to
147 * this physical channel. Can also be free or physically allocated.
148 * @allocated_dst: Same as for src but is dst.
149 * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as
162 * struct d40_chan - Struct that describes a channel.
164 * @lock: A spinlock to protect this struct.
165 * @log_num: The logical number, if any of this channel.
166 * @completed: Starts with 1, after first interrupt it is set to dma engine's
168 * @pending_tx: The number of pending transfers. Used between interrupt handler
170 * @busy: Set to true when transfer is ongoing on this channel.
171 * @phy_chan: Pointer to physical channel which this instance runs on. If this
172 * point is NULL, then the channel is not allocated.
173 * @chan: DMA engine handle.
174 * @tasklet: Tasklet that gets scheduled from interrupt context to complete a
175 * transfer and call client callback.
176 * @client: Cliented owned descriptor list.
177 * @active: Active descriptor.
178 * @queue: Queued jobs.
179 * @dma_cfg: The client configuration of this dma channel.
180 * @configured: whether the dma_cfg configuration is valid
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;
205 struct d40_base *base;
206 /* Default register configurations */
209 struct d40_def_lcsp log_def;
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 * @log_chans: Room for all possible logical channels in system.
238 * @lookup_log_chans: Used to map interrupt number to logical channel. Points
239 * to log_chans entries.
240 * @lookup_phy_chans: Used to map interrupt number to physical channel. Points
241 * to phy_chans entries.
242 * @plat_data: Pointer to provided platform_data which is the driver
244 * @phy_res: Vector containing all physical channels.
245 * @lcla_pool: lcla pool settings and data.
246 * @lcpa_base: The virtual mapped address of LCPA.
247 * @phy_lcpa: The physical address of the LCPA.
248 * @lcpa_size: The size of the LCPA area.
249 * @desc_slab: cache for descriptors.
252 spinlock_t interrupt_lock;
253 spinlock_t execmd_lock;
255 void __iomem *virtbase;
258 phys_addr_t phy_start;
259 resource_size_t phy_size;
263 struct dma_device dma_both;
264 struct dma_device dma_slave;
265 struct dma_device dma_memcpy;
266 struct d40_chan *phy_chans;
267 struct d40_chan *log_chans;
268 struct d40_chan **lookup_log_chans;
269 struct d40_chan **lookup_phy_chans;
270 struct stedma40_platform_data *plat_data;
271 /* Physical half channels */
272 struct d40_phy_res *phy_res;
273 struct d40_lcla_pool lcla_pool;
276 resource_size_t lcpa_size;
277 struct kmem_cache *desc_slab;
281 * struct d40_interrupt_lookup - lookup table for interrupt handler
283 * @src: Interrupt mask register.
284 * @clr: Interrupt clear register.
285 * @is_error: true if this is an error interrupt.
286 * @offset: start delta in the lookup_log_chans in d40_base. If equals to
287 * D40_PHY_CHAN, the lookup_phy_chans shall be used instead.
289 struct d40_interrupt_lookup {
297 * struct d40_reg_val - simple lookup struct
299 * @reg: The register.
300 * @val: The value that belongs to the register in reg.
307 static struct device *chan2dev(struct d40_chan *d40c)
309 return &d40c->chan.dev->device;
312 static bool chan_is_physical(struct d40_chan *chan)
314 return chan->log_num == D40_PHY_CHAN;
317 static bool chan_is_logical(struct d40_chan *chan)
319 return !chan_is_physical(chan);
322 static void __iomem *chan_base(struct d40_chan *chan)
324 return chan->base->virtbase + D40_DREG_PCBASE +
325 chan->phy_chan->num * D40_DREG_PCDELTA;
328 #define d40_err(dev, format, arg...) \
329 dev_err(dev, "[%s] " format, __func__, ## arg)
331 #define chan_err(d40c, format, arg...) \
332 d40_err(chan2dev(d40c), format, ## arg)
334 static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d,
337 bool is_log = chan_is_logical(d40c);
342 align = sizeof(struct d40_log_lli);
344 align = sizeof(struct d40_phy_lli);
347 base = d40d->lli_pool.pre_alloc_lli;
348 d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
349 d40d->lli_pool.base = NULL;
351 d40d->lli_pool.size = lli_len * 2 * align;
353 base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
354 d40d->lli_pool.base = base;
356 if (d40d->lli_pool.base == NULL)
361 d40d->lli_log.src = PTR_ALIGN(base, align);
362 d40d->lli_log.dst = d40d->lli_log.src + lli_len;
364 d40d->lli_pool.dma_addr = 0;
366 d40d->lli_phy.src = PTR_ALIGN(base, align);
367 d40d->lli_phy.dst = d40d->lli_phy.src + lli_len;
369 d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev,
374 if (dma_mapping_error(d40c->base->dev,
375 d40d->lli_pool.dma_addr)) {
376 kfree(d40d->lli_pool.base);
377 d40d->lli_pool.base = NULL;
378 d40d->lli_pool.dma_addr = 0;
386 static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d)
388 if (d40d->lli_pool.dma_addr)
389 dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr,
390 d40d->lli_pool.size, DMA_TO_DEVICE);
392 kfree(d40d->lli_pool.base);
393 d40d->lli_pool.base = NULL;
394 d40d->lli_pool.size = 0;
395 d40d->lli_log.src = NULL;
396 d40d->lli_log.dst = NULL;
397 d40d->lli_phy.src = NULL;
398 d40d->lli_phy.dst = NULL;
401 static int d40_lcla_alloc_one(struct d40_chan *d40c,
402 struct d40_desc *d40d)
409 spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
411 p = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP;
414 * Allocate both src and dst at the same time, therefore the half
415 * start on 1 since 0 can't be used since zero is used as end marker.
417 for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
418 if (!d40c->base->lcla_pool.alloc_map[p + i]) {
419 d40c->base->lcla_pool.alloc_map[p + i] = d40d;
426 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
431 static int d40_lcla_free_all(struct d40_chan *d40c,
432 struct d40_desc *d40d)
438 if (chan_is_physical(d40c))
441 spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
443 for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
444 if (d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num *
445 D40_LCLA_LINK_PER_EVENT_GRP + i] == d40d) {
446 d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num *
447 D40_LCLA_LINK_PER_EVENT_GRP + i] = NULL;
449 if (d40d->lcla_alloc == 0) {
456 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
462 static void d40_desc_remove(struct d40_desc *d40d)
464 list_del(&d40d->node);
467 static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
469 struct d40_desc *desc = NULL;
471 if (!list_empty(&d40c->client)) {
475 list_for_each_entry_safe(d, _d, &d40c->client, node)
476 if (async_tx_test_ack(&d->txd)) {
477 d40_pool_lli_free(d40c, d);
480 memset(desc, 0, sizeof(*desc));
486 desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT);
489 INIT_LIST_HEAD(&desc->node);
494 static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
497 d40_pool_lli_free(d40c, d40d);
498 d40_lcla_free_all(d40c, d40d);
499 kmem_cache_free(d40c->base->desc_slab, d40d);
502 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
504 list_add_tail(&desc->node, &d40c->active);
507 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
509 int curr_lcla = -EINVAL, next_lcla;
511 if (chan_is_physical(d40c)) {
512 d40_phy_lli_write(d40c->base->virtbase,
516 d40d->lli_current = d40d->lli_len;
519 if ((d40d->lli_len - d40d->lli_current) > 1)
520 curr_lcla = d40_lcla_alloc_one(d40c, d40d);
522 d40_log_lli_lcpa_write(d40c->lcpa,
523 &d40d->lli_log.dst[d40d->lli_current],
524 &d40d->lli_log.src[d40d->lli_current],
528 for (; d40d->lli_current < d40d->lli_len; d40d->lli_current++) {
529 unsigned int lcla_offset = d40c->phy_chan->num * 1024 +
531 struct d40_lcla_pool *pool = &d40c->base->lcla_pool;
532 struct d40_log_lli *lcla = pool->base + lcla_offset;
534 if (d40d->lli_current + 1 < d40d->lli_len)
535 next_lcla = d40_lcla_alloc_one(d40c, d40d);
539 d40_log_lli_lcla_write(lcla,
540 &d40d->lli_log.dst[d40d->lli_current],
541 &d40d->lli_log.src[d40d->lli_current],
544 dma_sync_single_range_for_device(d40c->base->dev,
545 pool->dma_addr, lcla_offset,
546 2 * sizeof(struct d40_log_lli),
549 curr_lcla = next_lcla;
551 if (curr_lcla == -EINVAL) {
560 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
564 if (list_empty(&d40c->active))
567 d = list_first_entry(&d40c->active,
573 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
575 list_add_tail(&desc->node, &d40c->queue);
578 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
582 if (list_empty(&d40c->queue))
585 d = list_first_entry(&d40c->queue,
591 static int d40_psize_2_burst_size(bool is_log, int psize)
594 if (psize == STEDMA40_PSIZE_LOG_1)
597 if (psize == STEDMA40_PSIZE_PHY_1)
605 * The dma only supports transmitting packages up to
606 * STEDMA40_MAX_SEG_SIZE << data_width. Calculate the total number of
607 * dma elements required to send the entire sg list
609 static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
612 u32 max_w = max(data_width1, data_width2);
613 u32 min_w = min(data_width1, data_width2);
614 u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE << min_w, 1 << max_w);
616 if (seg_max > STEDMA40_MAX_SEG_SIZE)
617 seg_max -= (1 << max_w);
619 if (!IS_ALIGNED(size, 1 << max_w))
625 dmalen = size / seg_max;
626 if (dmalen * seg_max < size)
632 static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
633 u32 data_width1, u32 data_width2)
635 struct scatterlist *sg;
640 for_each_sg(sgl, sg, sg_len, i) {
641 ret = d40_size_2_dmalen(sg_dma_len(sg),
642 data_width1, data_width2);
650 /* Support functions for logical channels */
652 static int d40_channel_execute_command(struct d40_chan *d40c,
653 enum d40_command command)
657 void __iomem *active_reg;
662 spin_lock_irqsave(&d40c->base->execmd_lock, flags);
664 if (d40c->phy_chan->num % 2 == 0)
665 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
667 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
669 if (command == D40_DMA_SUSPEND_REQ) {
670 status = (readl(active_reg) &
671 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
672 D40_CHAN_POS(d40c->phy_chan->num);
674 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
678 wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
679 writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
682 if (command == D40_DMA_SUSPEND_REQ) {
684 for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
685 status = (readl(active_reg) &
686 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
687 D40_CHAN_POS(d40c->phy_chan->num);
691 * Reduce the number of bus accesses while
692 * waiting for the DMA to suspend.
696 if (status == D40_DMA_STOP ||
697 status == D40_DMA_SUSPENDED)
701 if (i == D40_SUSPEND_MAX_IT) {
703 "unable to suspend the chl %d (log: %d) status %x\n",
704 d40c->phy_chan->num, d40c->log_num,
712 spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
716 static void d40_term_all(struct d40_chan *d40c)
718 struct d40_desc *d40d;
720 /* Release active descriptors */
721 while ((d40d = d40_first_active_get(d40c))) {
722 d40_desc_remove(d40d);
723 d40_desc_free(d40c, d40d);
726 /* Release queued descriptors waiting for transfer */
727 while ((d40d = d40_first_queued(d40c))) {
728 d40_desc_remove(d40d);
729 d40_desc_free(d40c, d40d);
733 d40c->pending_tx = 0;
737 static void __d40_config_set_event(struct d40_chan *d40c, bool enable,
740 void __iomem *addr = chan_base(d40c) + reg;
744 writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
745 | ~D40_EVENTLINE_MASK(event), addr);
750 * The hardware sometimes doesn't register the enable when src and dst
751 * event lines are active on the same logical channel. Retry to ensure
752 * it does. Usually only one retry is sufficient.
756 writel((D40_ACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
757 | ~D40_EVENTLINE_MASK(event), addr);
759 if (readl(addr) & D40_EVENTLINE_MASK(event))
764 dev_dbg(chan2dev(d40c),
765 "[%s] workaround enable S%cLNK (%d tries)\n",
766 __func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
772 static void d40_config_set_event(struct d40_chan *d40c, bool do_enable)
776 spin_lock_irqsave(&d40c->phy_chan->lock, flags);
778 /* Enable event line connected to device (or memcpy) */
779 if ((d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) ||
780 (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH)) {
781 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
783 __d40_config_set_event(d40c, do_enable, event,
787 if (d40c->dma_cfg.dir != STEDMA40_PERIPH_TO_MEM) {
788 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
790 __d40_config_set_event(d40c, do_enable, event,
794 spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
797 static u32 d40_chan_has_events(struct d40_chan *d40c)
799 void __iomem *chanbase = chan_base(d40c);
802 val = readl(chanbase + D40_CHAN_REG_SSLNK);
803 val |= readl(chanbase + D40_CHAN_REG_SDLNK);
808 static u32 d40_get_prmo(struct d40_chan *d40c)
810 static const unsigned int phy_map[] = {
811 [STEDMA40_PCHAN_BASIC_MODE]
812 = D40_DREG_PRMO_PCHAN_BASIC,
813 [STEDMA40_PCHAN_MODULO_MODE]
814 = D40_DREG_PRMO_PCHAN_MODULO,
815 [STEDMA40_PCHAN_DOUBLE_DST_MODE]
816 = D40_DREG_PRMO_PCHAN_DOUBLE_DST,
818 static const unsigned int log_map[] = {
819 [STEDMA40_LCHAN_SRC_PHY_DST_LOG]
820 = D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG,
821 [STEDMA40_LCHAN_SRC_LOG_DST_PHY]
822 = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY,
823 [STEDMA40_LCHAN_SRC_LOG_DST_LOG]
824 = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG,
827 if (chan_is_physical(d40c))
828 return phy_map[d40c->dma_cfg.mode_opt];
830 return log_map[d40c->dma_cfg.mode_opt];
833 static void d40_config_write(struct d40_chan *d40c)
838 /* Odd addresses are even addresses + 4 */
839 addr_base = (d40c->phy_chan->num % 2) * 4;
840 /* Setup channel mode to logical or physical */
841 var = ((u32)(chan_is_logical(d40c)) + 1) <<
842 D40_CHAN_POS(d40c->phy_chan->num);
843 writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
845 /* Setup operational mode option register */
846 var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num);
848 writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
850 if (chan_is_logical(d40c)) {
851 int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS)
852 & D40_SREG_ELEM_LOG_LIDX_MASK;
853 void __iomem *chanbase = chan_base(d40c);
855 /* Set default config for CFG reg */
856 writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG);
857 writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG);
859 /* Set LIDX for lcla */
860 writel(lidx, chanbase + D40_CHAN_REG_SSELT);
861 writel(lidx, chanbase + D40_CHAN_REG_SDELT);
865 static u32 d40_residue(struct d40_chan *d40c)
869 if (chan_is_logical(d40c))
870 num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
871 >> D40_MEM_LCSP2_ECNT_POS;
873 u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT);
874 num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK)
875 >> D40_SREG_ELEM_PHY_ECNT_POS;
878 return num_elt * (1 << d40c->dma_cfg.dst_info.data_width);
881 static bool d40_tx_is_linked(struct d40_chan *d40c)
885 if (chan_is_logical(d40c))
886 is_link = readl(&d40c->lcpa->lcsp3) & D40_MEM_LCSP3_DLOS_MASK;
888 is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK)
889 & D40_SREG_LNK_PHYS_LNK_MASK;
894 static int d40_pause(struct dma_chan *chan)
896 struct d40_chan *d40c =
897 container_of(chan, struct d40_chan, chan);
904 spin_lock_irqsave(&d40c->lock, flags);
906 res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
908 if (chan_is_logical(d40c)) {
909 d40_config_set_event(d40c, false);
910 /* Resume the other logical channels if any */
911 if (d40_chan_has_events(d40c))
912 res = d40_channel_execute_command(d40c,
917 spin_unlock_irqrestore(&d40c->lock, flags);
921 static int d40_resume(struct dma_chan *chan)
923 struct d40_chan *d40c =
924 container_of(chan, struct d40_chan, chan);
931 spin_lock_irqsave(&d40c->lock, flags);
933 if (d40c->base->rev == 0)
934 if (chan_is_logical(d40c)) {
935 res = d40_channel_execute_command(d40c,
936 D40_DMA_SUSPEND_REQ);
940 /* If bytes left to transfer or linked tx resume job */
941 if (d40_residue(d40c) || d40_tx_is_linked(d40c)) {
943 if (chan_is_logical(d40c))
944 d40_config_set_event(d40c, true);
946 res = d40_channel_execute_command(d40c, D40_DMA_RUN);
950 spin_unlock_irqrestore(&d40c->lock, flags);
954 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
956 struct d40_chan *d40c = container_of(tx->chan,
959 struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
962 spin_lock_irqsave(&d40c->lock, flags);
966 if (d40c->chan.cookie < 0)
967 d40c->chan.cookie = 1;
969 d40d->txd.cookie = d40c->chan.cookie;
971 d40_desc_queue(d40c, d40d);
973 spin_unlock_irqrestore(&d40c->lock, flags);
978 static int d40_start(struct d40_chan *d40c)
980 if (d40c->base->rev == 0) {
983 if (chan_is_logical(d40c)) {
984 err = d40_channel_execute_command(d40c,
985 D40_DMA_SUSPEND_REQ);
991 if (chan_is_logical(d40c))
992 d40_config_set_event(d40c, true);
994 return d40_channel_execute_command(d40c, D40_DMA_RUN);
997 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
999 struct d40_desc *d40d;
1002 /* Start queued jobs, if any */
1003 d40d = d40_first_queued(d40c);
1008 /* Remove from queue */
1009 d40_desc_remove(d40d);
1011 /* Add to active queue */
1012 d40_desc_submit(d40c, d40d);
1014 /* Initiate DMA job */
1015 d40_desc_load(d40c, d40d);
1018 err = d40_start(d40c);
1027 /* called from interrupt context */
1028 static void dma_tc_handle(struct d40_chan *d40c)
1030 struct d40_desc *d40d;
1032 /* Get first active entry from list */
1033 d40d = d40_first_active_get(d40c);
1038 d40_lcla_free_all(d40c, d40d);
1040 if (d40d->lli_current < d40d->lli_len) {
1041 d40_desc_load(d40c, d40d);
1043 (void) d40_start(d40c);
1047 if (d40_queue_start(d40c) == NULL)
1051 tasklet_schedule(&d40c->tasklet);
1055 static void dma_tasklet(unsigned long data)
1057 struct d40_chan *d40c = (struct d40_chan *) data;
1058 struct d40_desc *d40d;
1059 unsigned long flags;
1060 dma_async_tx_callback callback;
1061 void *callback_param;
1063 spin_lock_irqsave(&d40c->lock, flags);
1065 /* Get first active entry from list */
1066 d40d = d40_first_active_get(d40c);
1071 d40c->completed = d40d->txd.cookie;
1074 * If terminating a channel pending_tx is set to zero.
1075 * This prevents any finished active jobs to return to the client.
1077 if (d40c->pending_tx == 0) {
1078 spin_unlock_irqrestore(&d40c->lock, flags);
1082 /* Callback to client */
1083 callback = d40d->txd.callback;
1084 callback_param = d40d->txd.callback_param;
1086 if (async_tx_test_ack(&d40d->txd)) {
1087 d40_pool_lli_free(d40c, d40d);
1088 d40_desc_remove(d40d);
1089 d40_desc_free(d40c, d40d);
1091 if (!d40d->is_in_client_list) {
1092 d40_desc_remove(d40d);
1093 d40_lcla_free_all(d40c, d40d);
1094 list_add_tail(&d40d->node, &d40c->client);
1095 d40d->is_in_client_list = true;
1101 if (d40c->pending_tx)
1102 tasklet_schedule(&d40c->tasklet);
1104 spin_unlock_irqrestore(&d40c->lock, flags);
1106 if (callback && (d40d->txd.flags & DMA_PREP_INTERRUPT))
1107 callback(callback_param);
1112 /* Rescue manouver if receiving double interrupts */
1113 if (d40c->pending_tx > 0)
1115 spin_unlock_irqrestore(&d40c->lock, flags);
1118 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1120 static const struct d40_interrupt_lookup il[] = {
1121 {D40_DREG_LCTIS0, D40_DREG_LCICR0, false, 0},
1122 {D40_DREG_LCTIS1, D40_DREG_LCICR1, false, 32},
1123 {D40_DREG_LCTIS2, D40_DREG_LCICR2, false, 64},
1124 {D40_DREG_LCTIS3, D40_DREG_LCICR3, false, 96},
1125 {D40_DREG_LCEIS0, D40_DREG_LCICR0, true, 0},
1126 {D40_DREG_LCEIS1, D40_DREG_LCICR1, true, 32},
1127 {D40_DREG_LCEIS2, D40_DREG_LCICR2, true, 64},
1128 {D40_DREG_LCEIS3, D40_DREG_LCICR3, true, 96},
1129 {D40_DREG_PCTIS, D40_DREG_PCICR, false, D40_PHY_CHAN},
1130 {D40_DREG_PCEIS, D40_DREG_PCICR, true, D40_PHY_CHAN},
1134 u32 regs[ARRAY_SIZE(il)];
1138 struct d40_chan *d40c;
1139 unsigned long flags;
1140 struct d40_base *base = data;
1142 spin_lock_irqsave(&base->interrupt_lock, flags);
1144 /* Read interrupt status of both logical and physical channels */
1145 for (i = 0; i < ARRAY_SIZE(il); i++)
1146 regs[i] = readl(base->virtbase + il[i].src);
1150 chan = find_next_bit((unsigned long *)regs,
1151 BITS_PER_LONG * ARRAY_SIZE(il), chan + 1);
1153 /* No more set bits found? */
1154 if (chan == BITS_PER_LONG * ARRAY_SIZE(il))
1157 row = chan / BITS_PER_LONG;
1158 idx = chan & (BITS_PER_LONG - 1);
1161 writel(1 << idx, base->virtbase + il[row].clr);
1163 if (il[row].offset == D40_PHY_CHAN)
1164 d40c = base->lookup_phy_chans[idx];
1166 d40c = base->lookup_log_chans[il[row].offset + idx];
1167 spin_lock(&d40c->lock);
1169 if (!il[row].is_error)
1170 dma_tc_handle(d40c);
1172 d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1173 chan, il[row].offset, idx);
1175 spin_unlock(&d40c->lock);
1178 spin_unlock_irqrestore(&base->interrupt_lock, flags);
1183 static int d40_validate_conf(struct d40_chan *d40c,
1184 struct stedma40_chan_cfg *conf)
1187 u32 dst_event_group = D40_TYPE_TO_GROUP(conf->dst_dev_type);
1188 u32 src_event_group = D40_TYPE_TO_GROUP(conf->src_dev_type);
1189 bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1192 chan_err(d40c, "Invalid direction.\n");
1196 if (conf->dst_dev_type != STEDMA40_DEV_DST_MEMORY &&
1197 d40c->base->plat_data->dev_tx[conf->dst_dev_type] == 0 &&
1198 d40c->runtime_addr == 0) {
1200 chan_err(d40c, "Invalid TX channel address (%d)\n",
1201 conf->dst_dev_type);
1205 if (conf->src_dev_type != STEDMA40_DEV_SRC_MEMORY &&
1206 d40c->base->plat_data->dev_rx[conf->src_dev_type] == 0 &&
1207 d40c->runtime_addr == 0) {
1208 chan_err(d40c, "Invalid RX channel address (%d)\n",
1209 conf->src_dev_type);
1213 if (conf->dir == STEDMA40_MEM_TO_PERIPH &&
1214 dst_event_group == STEDMA40_DEV_DST_MEMORY) {
1215 chan_err(d40c, "Invalid dst\n");
1219 if (conf->dir == STEDMA40_PERIPH_TO_MEM &&
1220 src_event_group == STEDMA40_DEV_SRC_MEMORY) {
1221 chan_err(d40c, "Invalid src\n");
1225 if (src_event_group == STEDMA40_DEV_SRC_MEMORY &&
1226 dst_event_group == STEDMA40_DEV_DST_MEMORY && is_log) {
1227 chan_err(d40c, "No event line\n");
1231 if (conf->dir == STEDMA40_PERIPH_TO_PERIPH &&
1232 (src_event_group != dst_event_group)) {
1233 chan_err(d40c, "Invalid event group\n");
1237 if (conf->dir == STEDMA40_PERIPH_TO_PERIPH) {
1239 * DMAC HW supports it. Will be added to this driver,
1240 * in case any dma client requires it.
1242 chan_err(d40c, "periph to periph not supported\n");
1246 if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1247 (1 << conf->src_info.data_width) !=
1248 d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1249 (1 << conf->dst_info.data_width)) {
1251 * The DMAC hardware only supports
1252 * src (burst x width) == dst (burst x width)
1255 chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1262 static bool d40_alloc_mask_set(struct d40_phy_res *phy, bool is_src,
1263 int log_event_line, bool is_log)
1265 unsigned long flags;
1266 spin_lock_irqsave(&phy->lock, flags);
1268 /* Physical interrupts are masked per physical full channel */
1269 if (phy->allocated_src == D40_ALLOC_FREE &&
1270 phy->allocated_dst == D40_ALLOC_FREE) {
1271 phy->allocated_dst = D40_ALLOC_PHY;
1272 phy->allocated_src = D40_ALLOC_PHY;
1278 /* Logical channel */
1280 if (phy->allocated_src == D40_ALLOC_PHY)
1283 if (phy->allocated_src == D40_ALLOC_FREE)
1284 phy->allocated_src = D40_ALLOC_LOG_FREE;
1286 if (!(phy->allocated_src & (1 << log_event_line))) {
1287 phy->allocated_src |= 1 << log_event_line;
1292 if (phy->allocated_dst == D40_ALLOC_PHY)
1295 if (phy->allocated_dst == D40_ALLOC_FREE)
1296 phy->allocated_dst = D40_ALLOC_LOG_FREE;
1298 if (!(phy->allocated_dst & (1 << log_event_line))) {
1299 phy->allocated_dst |= 1 << log_event_line;
1306 spin_unlock_irqrestore(&phy->lock, flags);
1309 spin_unlock_irqrestore(&phy->lock, flags);
1313 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1316 unsigned long flags;
1317 bool is_free = false;
1319 spin_lock_irqsave(&phy->lock, flags);
1320 if (!log_event_line) {
1321 phy->allocated_dst = D40_ALLOC_FREE;
1322 phy->allocated_src = D40_ALLOC_FREE;
1327 /* Logical channel */
1329 phy->allocated_src &= ~(1 << log_event_line);
1330 if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1331 phy->allocated_src = D40_ALLOC_FREE;
1333 phy->allocated_dst &= ~(1 << log_event_line);
1334 if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1335 phy->allocated_dst = D40_ALLOC_FREE;
1338 is_free = ((phy->allocated_src | phy->allocated_dst) ==
1342 spin_unlock_irqrestore(&phy->lock, flags);
1347 static int d40_allocate_channel(struct d40_chan *d40c)
1352 struct d40_phy_res *phys;
1357 bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1359 phys = d40c->base->phy_res;
1361 if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1362 dev_type = d40c->dma_cfg.src_dev_type;
1363 log_num = 2 * dev_type;
1365 } else if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1366 d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1367 /* dst event lines are used for logical memcpy */
1368 dev_type = d40c->dma_cfg.dst_dev_type;
1369 log_num = 2 * dev_type + 1;
1374 event_group = D40_TYPE_TO_GROUP(dev_type);
1375 event_line = D40_TYPE_TO_EVENT(dev_type);
1378 if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1379 /* Find physical half channel */
1380 for (i = 0; i < d40c->base->num_phy_chans; i++) {
1382 if (d40_alloc_mask_set(&phys[i], is_src,
1387 for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1388 int phy_num = j + event_group * 2;
1389 for (i = phy_num; i < phy_num + 2; i++) {
1390 if (d40_alloc_mask_set(&phys[i],
1399 d40c->phy_chan = &phys[i];
1400 d40c->log_num = D40_PHY_CHAN;
1406 /* Find logical channel */
1407 for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1408 int phy_num = j + event_group * 2;
1410 * Spread logical channels across all available physical rather
1411 * than pack every logical channel at the first available phy
1415 for (i = phy_num; i < phy_num + 2; i++) {
1416 if (d40_alloc_mask_set(&phys[i], is_src,
1417 event_line, is_log))
1421 for (i = phy_num + 1; i >= phy_num; i--) {
1422 if (d40_alloc_mask_set(&phys[i], is_src,
1423 event_line, is_log))
1431 d40c->phy_chan = &phys[i];
1432 d40c->log_num = log_num;
1436 d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1438 d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1444 static int d40_config_memcpy(struct d40_chan *d40c)
1446 dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1448 if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1449 d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_log;
1450 d40c->dma_cfg.src_dev_type = STEDMA40_DEV_SRC_MEMORY;
1451 d40c->dma_cfg.dst_dev_type = d40c->base->plat_data->
1452 memcpy[d40c->chan.chan_id];
1454 } else if (dma_has_cap(DMA_MEMCPY, cap) &&
1455 dma_has_cap(DMA_SLAVE, cap)) {
1456 d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_phy;
1458 chan_err(d40c, "No memcpy\n");
1466 static int d40_free_dma(struct d40_chan *d40c)
1471 struct d40_phy_res *phy = d40c->phy_chan;
1474 struct d40_desc *_d;
1477 /* Terminate all queued and active transfers */
1480 /* Release client owned descriptors */
1481 if (!list_empty(&d40c->client))
1482 list_for_each_entry_safe(d, _d, &d40c->client, node) {
1483 d40_pool_lli_free(d40c, d);
1485 d40_desc_free(d40c, d);
1489 chan_err(d40c, "phy == null\n");
1493 if (phy->allocated_src == D40_ALLOC_FREE &&
1494 phy->allocated_dst == D40_ALLOC_FREE) {
1495 chan_err(d40c, "channel already free\n");
1499 if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1500 d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1501 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1503 } else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1504 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1507 chan_err(d40c, "Unknown direction\n");
1511 res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1513 chan_err(d40c, "suspend failed\n");
1517 if (chan_is_logical(d40c)) {
1518 /* Release logical channel, deactivate the event line */
1520 d40_config_set_event(d40c, false);
1521 d40c->base->lookup_log_chans[d40c->log_num] = NULL;
1524 * Check if there are more logical allocation
1525 * on this phy channel.
1527 if (!d40_alloc_mask_free(phy, is_src, event)) {
1528 /* Resume the other logical channels if any */
1529 if (d40_chan_has_events(d40c)) {
1530 res = d40_channel_execute_command(d40c,
1534 "Executing RUN command\n");
1541 (void) d40_alloc_mask_free(phy, is_src, 0);
1544 /* Release physical channel */
1545 res = d40_channel_execute_command(d40c, D40_DMA_STOP);
1547 chan_err(d40c, "Failed to stop channel\n");
1550 d40c->phy_chan = NULL;
1551 d40c->configured = false;
1552 d40c->base->lookup_phy_chans[phy->num] = NULL;
1557 static bool d40_is_paused(struct d40_chan *d40c)
1559 void __iomem *chanbase = chan_base(d40c);
1560 bool is_paused = false;
1561 unsigned long flags;
1562 void __iomem *active_reg;
1566 spin_lock_irqsave(&d40c->lock, flags);
1568 if (chan_is_physical(d40c)) {
1569 if (d40c->phy_chan->num % 2 == 0)
1570 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1572 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1574 status = (readl(active_reg) &
1575 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1576 D40_CHAN_POS(d40c->phy_chan->num);
1577 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1583 if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1584 d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1585 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1586 status = readl(chanbase + D40_CHAN_REG_SDLNK);
1587 } else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1588 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1589 status = readl(chanbase + D40_CHAN_REG_SSLNK);
1591 chan_err(d40c, "Unknown direction\n");
1595 status = (status & D40_EVENTLINE_MASK(event)) >>
1596 D40_EVENTLINE_POS(event);
1598 if (status != D40_DMA_RUN)
1601 spin_unlock_irqrestore(&d40c->lock, flags);
1607 static u32 stedma40_residue(struct dma_chan *chan)
1609 struct d40_chan *d40c =
1610 container_of(chan, struct d40_chan, chan);
1612 unsigned long flags;
1614 spin_lock_irqsave(&d40c->lock, flags);
1615 bytes_left = d40_residue(d40c);
1616 spin_unlock_irqrestore(&d40c->lock, flags);
1622 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
1623 struct scatterlist *sg_src, struct scatterlist *sg_dst,
1624 unsigned int sg_len, enum dma_data_direction direction,
1625 dma_addr_t dev_addr)
1627 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
1628 struct stedma40_half_channel_info *src_info = &cfg->src_info;
1629 struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
1631 if (direction == DMA_NONE) {
1633 (void) d40_log_sg_to_lli(sg_src, sg_len,
1635 chan->log_def.lcsp1,
1636 src_info->data_width,
1637 dst_info->data_width);
1639 (void) d40_log_sg_to_lli(sg_dst, sg_len,
1641 chan->log_def.lcsp3,
1642 dst_info->data_width,
1643 src_info->data_width);
1645 unsigned int total_size;
1647 total_size = d40_log_sg_to_dev(sg_src, sg_len,
1650 src_info->data_width,
1651 dst_info->data_width,
1652 direction, dev_addr);
1661 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
1662 struct scatterlist *sg_src, struct scatterlist *sg_dst,
1663 unsigned int sg_len, enum dma_data_direction direction,
1664 dma_addr_t dev_addr)
1666 dma_addr_t src_dev_addr = direction == DMA_FROM_DEVICE ? dev_addr : 0;
1667 dma_addr_t dst_dev_addr = direction == DMA_TO_DEVICE ? dev_addr : 0;
1668 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
1669 struct stedma40_half_channel_info *src_info = &cfg->src_info;
1670 struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
1673 ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
1675 virt_to_phys(desc->lli_phy.src),
1677 src_info->data_width,
1678 dst_info->data_width,
1681 ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
1683 virt_to_phys(desc->lli_phy.dst),
1685 dst_info->data_width,
1686 src_info->data_width,
1689 dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
1690 desc->lli_pool.size, DMA_TO_DEVICE);
1692 return ret < 0 ? ret : 0;
1696 static struct d40_desc *
1697 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
1698 unsigned int sg_len, unsigned long dma_flags)
1700 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
1701 struct d40_desc *desc;
1704 desc = d40_desc_get(chan);
1708 desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
1709 cfg->dst_info.data_width);
1710 if (desc->lli_len < 0) {
1711 chan_err(chan, "Unaligned size\n");
1715 ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
1717 chan_err(chan, "Could not allocate lli\n");
1722 desc->lli_current = 0;
1723 desc->txd.flags = dma_flags;
1724 desc->txd.tx_submit = d40_tx_submit;
1726 dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
1731 d40_desc_free(chan, desc);
1735 struct dma_async_tx_descriptor *stedma40_memcpy_sg(struct dma_chan *chan,
1736 struct scatterlist *sgl_dst,
1737 struct scatterlist *sgl_src,
1738 unsigned int sgl_len,
1739 unsigned long dma_flags)
1741 struct d40_desc *d40d;
1742 struct d40_chan *d40c = container_of(chan, struct d40_chan,
1744 unsigned long flags;
1746 if (d40c->phy_chan == NULL) {
1747 chan_err(d40c, "Unallocated channel.\n");
1748 return ERR_PTR(-EINVAL);
1751 spin_lock_irqsave(&d40c->lock, flags);
1753 d40d = d40_prep_desc(d40c, sgl_dst, sgl_len, dma_flags);
1757 if (chan_is_logical(d40c)) {
1758 d40_prep_sg_log(d40c, d40d, sgl_src, sgl_dst,
1759 sgl_len, DMA_NONE, 0);
1761 d40_prep_sg_phy(d40c, d40d, sgl_src, sgl_dst,
1762 sgl_len, DMA_NONE, 0);
1765 spin_unlock_irqrestore(&d40c->lock, flags);
1770 d40_desc_free(d40c, d40d);
1771 spin_unlock_irqrestore(&d40c->lock, flags);
1774 EXPORT_SYMBOL(stedma40_memcpy_sg);
1776 bool stedma40_filter(struct dma_chan *chan, void *data)
1778 struct stedma40_chan_cfg *info = data;
1779 struct d40_chan *d40c =
1780 container_of(chan, struct d40_chan, chan);
1784 err = d40_validate_conf(d40c, info);
1786 d40c->dma_cfg = *info;
1788 err = d40_config_memcpy(d40c);
1791 d40c->configured = true;
1795 EXPORT_SYMBOL(stedma40_filter);
1797 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
1799 bool realtime = d40c->dma_cfg.realtime;
1800 bool highprio = d40c->dma_cfg.high_priority;
1801 u32 prioreg = highprio ? D40_DREG_PSEG1 : D40_DREG_PCEG1;
1802 u32 rtreg = realtime ? D40_DREG_RSEG1 : D40_DREG_RCEG1;
1803 u32 event = D40_TYPE_TO_EVENT(dev_type);
1804 u32 group = D40_TYPE_TO_GROUP(dev_type);
1805 u32 bit = 1 << event;
1807 /* Destination event lines are stored in the upper halfword */
1811 writel(bit, d40c->base->virtbase + prioreg + group * 4);
1812 writel(bit, d40c->base->virtbase + rtreg + group * 4);
1815 static void d40_set_prio_realtime(struct d40_chan *d40c)
1817 if (d40c->base->rev < 3)
1820 if ((d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) ||
1821 (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH))
1822 __d40_set_prio_rt(d40c, d40c->dma_cfg.src_dev_type, true);
1824 if ((d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH) ||
1825 (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH))
1826 __d40_set_prio_rt(d40c, d40c->dma_cfg.dst_dev_type, false);
1829 /* DMA ENGINE functions */
1830 static int d40_alloc_chan_resources(struct dma_chan *chan)
1833 unsigned long flags;
1834 struct d40_chan *d40c =
1835 container_of(chan, struct d40_chan, chan);
1837 spin_lock_irqsave(&d40c->lock, flags);
1839 d40c->completed = chan->cookie = 1;
1841 /* If no dma configuration is set use default configuration (memcpy) */
1842 if (!d40c->configured) {
1843 err = d40_config_memcpy(d40c);
1845 chan_err(d40c, "Failed to configure memcpy channel\n");
1849 is_free_phy = (d40c->phy_chan == NULL);
1851 err = d40_allocate_channel(d40c);
1853 chan_err(d40c, "Failed to allocate channel\n");
1857 /* Fill in basic CFG register values */
1858 d40_phy_cfg(&d40c->dma_cfg, &d40c->src_def_cfg,
1859 &d40c->dst_def_cfg, chan_is_logical(d40c));
1861 d40_set_prio_realtime(d40c);
1863 if (chan_is_logical(d40c)) {
1864 d40_log_cfg(&d40c->dma_cfg,
1865 &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
1867 if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM)
1868 d40c->lcpa = d40c->base->lcpa_base +
1869 d40c->dma_cfg.src_dev_type * D40_LCPA_CHAN_SIZE;
1871 d40c->lcpa = d40c->base->lcpa_base +
1872 d40c->dma_cfg.dst_dev_type *
1873 D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
1877 * Only write channel configuration to the DMA if the physical
1878 * resource is free. In case of multiple logical channels
1879 * on the same physical resource, only the first write is necessary.
1882 d40_config_write(d40c);
1884 spin_unlock_irqrestore(&d40c->lock, flags);
1888 static void d40_free_chan_resources(struct dma_chan *chan)
1890 struct d40_chan *d40c =
1891 container_of(chan, struct d40_chan, chan);
1893 unsigned long flags;
1895 if (d40c->phy_chan == NULL) {
1896 chan_err(d40c, "Cannot free unallocated channel\n");
1901 spin_lock_irqsave(&d40c->lock, flags);
1903 err = d40_free_dma(d40c);
1906 chan_err(d40c, "Failed to free channel\n");
1907 spin_unlock_irqrestore(&d40c->lock, flags);
1910 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
1914 unsigned long dma_flags)
1916 struct scatterlist dst_sg;
1917 struct scatterlist src_sg;
1919 sg_init_table(&dst_sg, 1);
1920 sg_init_table(&src_sg, 1);
1922 sg_dma_address(&dst_sg) = dst;
1923 sg_dma_address(&src_sg) = src;
1925 sg_dma_len(&dst_sg) = size;
1926 sg_dma_len(&src_sg) = size;
1928 return stedma40_memcpy_sg(chan, &dst_sg, &src_sg, 1, dma_flags);
1931 static struct dma_async_tx_descriptor *
1932 d40_prep_sg(struct dma_chan *chan,
1933 struct scatterlist *dst_sg, unsigned int dst_nents,
1934 struct scatterlist *src_sg, unsigned int src_nents,
1935 unsigned long dma_flags)
1937 if (dst_nents != src_nents)
1940 return stedma40_memcpy_sg(chan, dst_sg, src_sg, dst_nents, dma_flags);
1944 d40_get_dev_addr(struct d40_chan *chan, enum dma_data_direction direction)
1946 struct stedma40_platform_data *plat = chan->base->plat_data;
1947 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
1950 if (chan->runtime_addr)
1951 return chan->runtime_addr;
1953 if (direction == DMA_FROM_DEVICE)
1954 addr = plat->dev_rx[cfg->src_dev_type];
1955 else if (direction == DMA_TO_DEVICE)
1956 addr = plat->dev_tx[cfg->dst_dev_type];
1961 static struct dma_async_tx_descriptor *d40_prep_slave_sg(struct dma_chan *chan,
1962 struct scatterlist *sgl,
1963 unsigned int sg_len,
1964 enum dma_data_direction direction,
1965 unsigned long dma_flags)
1967 struct d40_desc *d40d;
1968 struct d40_chan *d40c = container_of(chan, struct d40_chan,
1970 dma_addr_t dev_addr;
1971 unsigned long flags;
1974 if (d40c->phy_chan == NULL) {
1975 chan_err(d40c, "Cannot prepare unallocated channel\n");
1976 return ERR_PTR(-EINVAL);
1979 if (direction != DMA_FROM_DEVICE && direction != DMA_TO_DEVICE)
1982 spin_lock_irqsave(&d40c->lock, flags);
1984 d40d = d40_prep_desc(d40c, sgl, sg_len, dma_flags);
1988 dev_addr = d40_get_dev_addr(d40c, direction);
1990 if (chan_is_logical(d40c))
1991 err = d40_prep_sg_log(d40c, d40d, sgl, NULL,
1992 sg_len, direction, dev_addr);
1994 err = d40_prep_sg_phy(d40c, d40d, sgl, NULL,
1995 sg_len, direction, dev_addr);
1998 chan_err(d40c, "Failed to prepare %s slave sg job: %d\n",
1999 chan_is_logical(d40c) ? "log" : "phy", err);
2003 spin_unlock_irqrestore(&d40c->lock, flags);
2008 d40_desc_free(d40c, d40d);
2009 spin_unlock_irqrestore(&d40c->lock, flags);
2013 static enum dma_status d40_tx_status(struct dma_chan *chan,
2014 dma_cookie_t cookie,
2015 struct dma_tx_state *txstate)
2017 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2018 dma_cookie_t last_used;
2019 dma_cookie_t last_complete;
2022 if (d40c->phy_chan == NULL) {
2023 chan_err(d40c, "Cannot read status of unallocated channel\n");
2027 last_complete = d40c->completed;
2028 last_used = chan->cookie;
2030 if (d40_is_paused(d40c))
2033 ret = dma_async_is_complete(cookie, last_complete, last_used);
2035 dma_set_tx_state(txstate, last_complete, last_used,
2036 stedma40_residue(chan));
2041 static void d40_issue_pending(struct dma_chan *chan)
2043 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2044 unsigned long flags;
2046 if (d40c->phy_chan == NULL) {
2047 chan_err(d40c, "Channel is not allocated!\n");
2051 spin_lock_irqsave(&d40c->lock, flags);
2053 /* Busy means that pending jobs are already being processed */
2055 (void) d40_queue_start(d40c);
2057 spin_unlock_irqrestore(&d40c->lock, flags);
2060 /* Runtime reconfiguration extension */
2061 static void d40_set_runtime_config(struct dma_chan *chan,
2062 struct dma_slave_config *config)
2064 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2065 struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2066 enum dma_slave_buswidth config_addr_width;
2067 dma_addr_t config_addr;
2068 u32 config_maxburst;
2069 enum stedma40_periph_data_width addr_width;
2072 if (config->direction == DMA_FROM_DEVICE) {
2073 dma_addr_t dev_addr_rx =
2074 d40c->base->plat_data->dev_rx[cfg->src_dev_type];
2076 config_addr = config->src_addr;
2078 dev_dbg(d40c->base->dev,
2079 "channel has a pre-wired RX address %08x "
2080 "overriding with %08x\n",
2081 dev_addr_rx, config_addr);
2082 if (cfg->dir != STEDMA40_PERIPH_TO_MEM)
2083 dev_dbg(d40c->base->dev,
2084 "channel was not configured for peripheral "
2085 "to memory transfer (%d) overriding\n",
2087 cfg->dir = STEDMA40_PERIPH_TO_MEM;
2089 config_addr_width = config->src_addr_width;
2090 config_maxburst = config->src_maxburst;
2092 } else if (config->direction == DMA_TO_DEVICE) {
2093 dma_addr_t dev_addr_tx =
2094 d40c->base->plat_data->dev_tx[cfg->dst_dev_type];
2096 config_addr = config->dst_addr;
2098 dev_dbg(d40c->base->dev,
2099 "channel has a pre-wired TX address %08x "
2100 "overriding with %08x\n",
2101 dev_addr_tx, config_addr);
2102 if (cfg->dir != STEDMA40_MEM_TO_PERIPH)
2103 dev_dbg(d40c->base->dev,
2104 "channel was not configured for memory "
2105 "to peripheral transfer (%d) overriding\n",
2107 cfg->dir = STEDMA40_MEM_TO_PERIPH;
2109 config_addr_width = config->dst_addr_width;
2110 config_maxburst = config->dst_maxburst;
2113 dev_err(d40c->base->dev,
2114 "unrecognized channel direction %d\n",
2119 switch (config_addr_width) {
2120 case DMA_SLAVE_BUSWIDTH_1_BYTE:
2121 addr_width = STEDMA40_BYTE_WIDTH;
2123 case DMA_SLAVE_BUSWIDTH_2_BYTES:
2124 addr_width = STEDMA40_HALFWORD_WIDTH;
2126 case DMA_SLAVE_BUSWIDTH_4_BYTES:
2127 addr_width = STEDMA40_WORD_WIDTH;
2129 case DMA_SLAVE_BUSWIDTH_8_BYTES:
2130 addr_width = STEDMA40_DOUBLEWORD_WIDTH;
2133 dev_err(d40c->base->dev,
2134 "illegal peripheral address width "
2136 config->src_addr_width);
2140 if (chan_is_logical(d40c)) {
2141 if (config_maxburst >= 16)
2142 psize = STEDMA40_PSIZE_LOG_16;
2143 else if (config_maxburst >= 8)
2144 psize = STEDMA40_PSIZE_LOG_8;
2145 else if (config_maxburst >= 4)
2146 psize = STEDMA40_PSIZE_LOG_4;
2148 psize = STEDMA40_PSIZE_LOG_1;
2150 if (config_maxburst >= 16)
2151 psize = STEDMA40_PSIZE_PHY_16;
2152 else if (config_maxburst >= 8)
2153 psize = STEDMA40_PSIZE_PHY_8;
2154 else if (config_maxburst >= 4)
2155 psize = STEDMA40_PSIZE_PHY_4;
2156 else if (config_maxburst >= 2)
2157 psize = STEDMA40_PSIZE_PHY_2;
2159 psize = STEDMA40_PSIZE_PHY_1;
2162 /* Set up all the endpoint configs */
2163 cfg->src_info.data_width = addr_width;
2164 cfg->src_info.psize = psize;
2165 cfg->src_info.big_endian = false;
2166 cfg->src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2167 cfg->dst_info.data_width = addr_width;
2168 cfg->dst_info.psize = psize;
2169 cfg->dst_info.big_endian = false;
2170 cfg->dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2172 /* Fill in register values */
2173 if (chan_is_logical(d40c))
2174 d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2176 d40_phy_cfg(cfg, &d40c->src_def_cfg,
2177 &d40c->dst_def_cfg, false);
2179 /* These settings will take precedence later */
2180 d40c->runtime_addr = config_addr;
2181 d40c->runtime_direction = config->direction;
2182 dev_dbg(d40c->base->dev,
2183 "configured channel %s for %s, data width %d, "
2184 "maxburst %d bytes, LE, no flow control\n",
2185 dma_chan_name(chan),
2186 (config->direction == DMA_FROM_DEVICE) ? "RX" : "TX",
2191 static int d40_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
2194 unsigned long flags;
2195 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2197 if (d40c->phy_chan == NULL) {
2198 chan_err(d40c, "Channel is not allocated!\n");
2203 case DMA_TERMINATE_ALL:
2204 spin_lock_irqsave(&d40c->lock, flags);
2206 spin_unlock_irqrestore(&d40c->lock, flags);
2209 return d40_pause(chan);
2211 return d40_resume(chan);
2212 case DMA_SLAVE_CONFIG:
2213 d40_set_runtime_config(chan,
2214 (struct dma_slave_config *) arg);
2220 /* Other commands are unimplemented */
2224 /* Initialization functions */
2226 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2227 struct d40_chan *chans, int offset,
2231 struct d40_chan *d40c;
2233 INIT_LIST_HEAD(&dma->channels);
2235 for (i = offset; i < offset + num_chans; i++) {
2238 d40c->chan.device = dma;
2240 spin_lock_init(&d40c->lock);
2242 d40c->log_num = D40_PHY_CHAN;
2244 INIT_LIST_HEAD(&d40c->active);
2245 INIT_LIST_HEAD(&d40c->queue);
2246 INIT_LIST_HEAD(&d40c->client);
2248 tasklet_init(&d40c->tasklet, dma_tasklet,
2249 (unsigned long) d40c);
2251 list_add_tail(&d40c->chan.device_node,
2256 static int __init d40_dmaengine_init(struct d40_base *base,
2257 int num_reserved_chans)
2261 d40_chan_init(base, &base->dma_slave, base->log_chans,
2262 0, base->num_log_chans);
2264 dma_cap_zero(base->dma_slave.cap_mask);
2265 dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2267 base->dma_slave.device_alloc_chan_resources = d40_alloc_chan_resources;
2268 base->dma_slave.device_free_chan_resources = d40_free_chan_resources;
2269 base->dma_slave.device_prep_dma_memcpy = d40_prep_memcpy;
2270 base->dma_slave.device_prep_dma_sg = d40_prep_sg;
2271 base->dma_slave.device_prep_slave_sg = d40_prep_slave_sg;
2272 base->dma_slave.device_tx_status = d40_tx_status;
2273 base->dma_slave.device_issue_pending = d40_issue_pending;
2274 base->dma_slave.device_control = d40_control;
2275 base->dma_slave.dev = base->dev;
2277 err = dma_async_device_register(&base->dma_slave);
2280 d40_err(base->dev, "Failed to register slave channels\n");
2284 d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2285 base->num_log_chans, base->plat_data->memcpy_len);
2287 dma_cap_zero(base->dma_memcpy.cap_mask);
2288 dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2289 dma_cap_set(DMA_SG, base->dma_slave.cap_mask);
2291 base->dma_memcpy.device_alloc_chan_resources = d40_alloc_chan_resources;
2292 base->dma_memcpy.device_free_chan_resources = d40_free_chan_resources;
2293 base->dma_memcpy.device_prep_dma_memcpy = d40_prep_memcpy;
2294 base->dma_slave.device_prep_dma_sg = d40_prep_sg;
2295 base->dma_memcpy.device_prep_slave_sg = d40_prep_slave_sg;
2296 base->dma_memcpy.device_tx_status = d40_tx_status;
2297 base->dma_memcpy.device_issue_pending = d40_issue_pending;
2298 base->dma_memcpy.device_control = d40_control;
2299 base->dma_memcpy.dev = base->dev;
2301 * This controller can only access address at even
2302 * 32bit boundaries, i.e. 2^2
2304 base->dma_memcpy.copy_align = 2;
2306 err = dma_async_device_register(&base->dma_memcpy);
2310 "Failed to regsiter memcpy only channels\n");
2314 d40_chan_init(base, &base->dma_both, base->phy_chans,
2315 0, num_reserved_chans);
2317 dma_cap_zero(base->dma_both.cap_mask);
2318 dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2319 dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2320 dma_cap_set(DMA_SG, base->dma_slave.cap_mask);
2322 base->dma_both.device_alloc_chan_resources = d40_alloc_chan_resources;
2323 base->dma_both.device_free_chan_resources = d40_free_chan_resources;
2324 base->dma_both.device_prep_dma_memcpy = d40_prep_memcpy;
2325 base->dma_slave.device_prep_dma_sg = d40_prep_sg;
2326 base->dma_both.device_prep_slave_sg = d40_prep_slave_sg;
2327 base->dma_both.device_tx_status = d40_tx_status;
2328 base->dma_both.device_issue_pending = d40_issue_pending;
2329 base->dma_both.device_control = d40_control;
2330 base->dma_both.dev = base->dev;
2331 base->dma_both.copy_align = 2;
2332 err = dma_async_device_register(&base->dma_both);
2336 "Failed to register logical and physical capable channels\n");
2341 dma_async_device_unregister(&base->dma_memcpy);
2343 dma_async_device_unregister(&base->dma_slave);
2348 /* Initialization functions. */
2350 static int __init d40_phy_res_init(struct d40_base *base)
2353 int num_phy_chans_avail = 0;
2355 int odd_even_bit = -2;
2357 val[0] = readl(base->virtbase + D40_DREG_PRSME);
2358 val[1] = readl(base->virtbase + D40_DREG_PRSMO);
2360 for (i = 0; i < base->num_phy_chans; i++) {
2361 base->phy_res[i].num = i;
2362 odd_even_bit += 2 * ((i % 2) == 0);
2363 if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
2364 /* Mark security only channels as occupied */
2365 base->phy_res[i].allocated_src = D40_ALLOC_PHY;
2366 base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
2368 base->phy_res[i].allocated_src = D40_ALLOC_FREE;
2369 base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
2370 num_phy_chans_avail++;
2372 spin_lock_init(&base->phy_res[i].lock);
2375 /* Mark disabled channels as occupied */
2376 for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
2377 int chan = base->plat_data->disabled_channels[i];
2379 base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
2380 base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
2381 num_phy_chans_avail--;
2384 dev_info(base->dev, "%d of %d physical DMA channels available\n",
2385 num_phy_chans_avail, base->num_phy_chans);
2387 /* Verify settings extended vs standard */
2388 val[0] = readl(base->virtbase + D40_DREG_PRTYP);
2390 for (i = 0; i < base->num_phy_chans; i++) {
2392 if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
2393 (val[0] & 0x3) != 1)
2395 "[%s] INFO: channel %d is misconfigured (%d)\n",
2396 __func__, i, val[0] & 0x3);
2398 val[0] = val[0] >> 2;
2401 return num_phy_chans_avail;
2404 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
2406 static const struct d40_reg_val dma_id_regs[] = {
2408 { .reg = D40_DREG_PERIPHID0, .val = 0x0040},
2409 { .reg = D40_DREG_PERIPHID1, .val = 0x0000},
2411 * D40_DREG_PERIPHID2 Depends on HW revision:
2412 * DB8500ed has 0x0008,
2414 * DB8500v1 has 0x0028
2415 * DB8500v2 has 0x0038
2417 { .reg = D40_DREG_PERIPHID3, .val = 0x0000},
2420 { .reg = D40_DREG_CELLID0, .val = 0x000d},
2421 { .reg = D40_DREG_CELLID1, .val = 0x00f0},
2422 { .reg = D40_DREG_CELLID2, .val = 0x0005},
2423 { .reg = D40_DREG_CELLID3, .val = 0x00b1}
2425 struct stedma40_platform_data *plat_data;
2426 struct clk *clk = NULL;
2427 void __iomem *virtbase = NULL;
2428 struct resource *res = NULL;
2429 struct d40_base *base = NULL;
2430 int num_log_chans = 0;
2436 clk = clk_get(&pdev->dev, NULL);
2439 d40_err(&pdev->dev, "No matching clock found\n");
2445 /* Get IO for DMAC base address */
2446 res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
2450 if (request_mem_region(res->start, resource_size(res),
2451 D40_NAME " I/O base") == NULL)
2454 virtbase = ioremap(res->start, resource_size(res));
2458 /* HW version check */
2459 for (i = 0; i < ARRAY_SIZE(dma_id_regs); i++) {
2460 if (dma_id_regs[i].val !=
2461 readl(virtbase + dma_id_regs[i].reg)) {
2463 "Unknown hardware! Expected 0x%x at 0x%x but got 0x%x\n",
2466 readl(virtbase + dma_id_regs[i].reg));
2471 /* Get silicon revision and designer */
2472 val = readl(virtbase + D40_DREG_PERIPHID2);
2474 if ((val & D40_DREG_PERIPHID2_DESIGNER_MASK) !=
2476 d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
2477 val & D40_DREG_PERIPHID2_DESIGNER_MASK,
2482 rev = (val & D40_DREG_PERIPHID2_REV_MASK) >>
2483 D40_DREG_PERIPHID2_REV_POS;
2485 /* The number of physical channels on this HW */
2486 num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
2488 dev_info(&pdev->dev, "hardware revision: %d @ 0x%x\n",
2491 plat_data = pdev->dev.platform_data;
2493 /* Count the number of logical channels in use */
2494 for (i = 0; i < plat_data->dev_len; i++)
2495 if (plat_data->dev_rx[i] != 0)
2498 for (i = 0; i < plat_data->dev_len; i++)
2499 if (plat_data->dev_tx[i] != 0)
2502 base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
2503 (num_phy_chans + num_log_chans + plat_data->memcpy_len) *
2504 sizeof(struct d40_chan), GFP_KERNEL);
2507 d40_err(&pdev->dev, "Out of memory\n");
2513 base->num_phy_chans = num_phy_chans;
2514 base->num_log_chans = num_log_chans;
2515 base->phy_start = res->start;
2516 base->phy_size = resource_size(res);
2517 base->virtbase = virtbase;
2518 base->plat_data = plat_data;
2519 base->dev = &pdev->dev;
2520 base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
2521 base->log_chans = &base->phy_chans[num_phy_chans];
2523 base->phy_res = kzalloc(num_phy_chans * sizeof(struct d40_phy_res),
2528 base->lookup_phy_chans = kzalloc(num_phy_chans *
2529 sizeof(struct d40_chan *),
2531 if (!base->lookup_phy_chans)
2534 if (num_log_chans + plat_data->memcpy_len) {
2536 * The max number of logical channels are event lines for all
2537 * src devices and dst devices
2539 base->lookup_log_chans = kzalloc(plat_data->dev_len * 2 *
2540 sizeof(struct d40_chan *),
2542 if (!base->lookup_log_chans)
2546 base->lcla_pool.alloc_map = kzalloc(num_phy_chans *
2547 sizeof(struct d40_desc *) *
2548 D40_LCLA_LINK_PER_EVENT_GRP,
2550 if (!base->lcla_pool.alloc_map)
2553 base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
2554 0, SLAB_HWCACHE_ALIGN,
2556 if (base->desc_slab == NULL)
2569 release_mem_region(res->start,
2570 resource_size(res));
2575 kfree(base->lcla_pool.alloc_map);
2576 kfree(base->lookup_log_chans);
2577 kfree(base->lookup_phy_chans);
2578 kfree(base->phy_res);
2585 static void __init d40_hw_init(struct d40_base *base)
2588 static const struct d40_reg_val dma_init_reg[] = {
2589 /* Clock every part of the DMA block from start */
2590 { .reg = D40_DREG_GCC, .val = 0x0000ff01},
2592 /* Interrupts on all logical channels */
2593 { .reg = D40_DREG_LCMIS0, .val = 0xFFFFFFFF},
2594 { .reg = D40_DREG_LCMIS1, .val = 0xFFFFFFFF},
2595 { .reg = D40_DREG_LCMIS2, .val = 0xFFFFFFFF},
2596 { .reg = D40_DREG_LCMIS3, .val = 0xFFFFFFFF},
2597 { .reg = D40_DREG_LCICR0, .val = 0xFFFFFFFF},
2598 { .reg = D40_DREG_LCICR1, .val = 0xFFFFFFFF},
2599 { .reg = D40_DREG_LCICR2, .val = 0xFFFFFFFF},
2600 { .reg = D40_DREG_LCICR3, .val = 0xFFFFFFFF},
2601 { .reg = D40_DREG_LCTIS0, .val = 0xFFFFFFFF},
2602 { .reg = D40_DREG_LCTIS1, .val = 0xFFFFFFFF},
2603 { .reg = D40_DREG_LCTIS2, .val = 0xFFFFFFFF},
2604 { .reg = D40_DREG_LCTIS3, .val = 0xFFFFFFFF}
2607 u32 prmseo[2] = {0, 0};
2608 u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
2612 for (i = 0; i < ARRAY_SIZE(dma_init_reg); i++)
2613 writel(dma_init_reg[i].val,
2614 base->virtbase + dma_init_reg[i].reg);
2616 /* Configure all our dma channels to default settings */
2617 for (i = 0; i < base->num_phy_chans; i++) {
2619 activeo[i % 2] = activeo[i % 2] << 2;
2621 if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
2623 activeo[i % 2] |= 3;
2627 /* Enable interrupt # */
2628 pcmis = (pcmis << 1) | 1;
2630 /* Clear interrupt # */
2631 pcicr = (pcicr << 1) | 1;
2633 /* Set channel to physical mode */
2634 prmseo[i % 2] = prmseo[i % 2] << 2;
2639 writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
2640 writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
2641 writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
2642 writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
2644 /* Write which interrupt to enable */
2645 writel(pcmis, base->virtbase + D40_DREG_PCMIS);
2647 /* Write which interrupt to clear */
2648 writel(pcicr, base->virtbase + D40_DREG_PCICR);
2652 static int __init d40_lcla_allocate(struct d40_base *base)
2654 struct d40_lcla_pool *pool = &base->lcla_pool;
2655 unsigned long *page_list;
2660 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
2661 * To full fill this hardware requirement without wasting 256 kb
2662 * we allocate pages until we get an aligned one.
2664 page_list = kmalloc(sizeof(unsigned long) * MAX_LCLA_ALLOC_ATTEMPTS,
2672 /* Calculating how many pages that are required */
2673 base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
2675 for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
2676 page_list[i] = __get_free_pages(GFP_KERNEL,
2677 base->lcla_pool.pages);
2678 if (!page_list[i]) {
2680 d40_err(base->dev, "Failed to allocate %d pages.\n",
2681 base->lcla_pool.pages);
2683 for (j = 0; j < i; j++)
2684 free_pages(page_list[j], base->lcla_pool.pages);
2688 if ((virt_to_phys((void *)page_list[i]) &
2689 (LCLA_ALIGNMENT - 1)) == 0)
2693 for (j = 0; j < i; j++)
2694 free_pages(page_list[j], base->lcla_pool.pages);
2696 if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
2697 base->lcla_pool.base = (void *)page_list[i];
2700 * After many attempts and no succees with finding the correct
2701 * alignment, try with allocating a big buffer.
2704 "[%s] Failed to get %d pages @ 18 bit align.\n",
2705 __func__, base->lcla_pool.pages);
2706 base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
2707 base->num_phy_chans +
2710 if (!base->lcla_pool.base_unaligned) {
2715 base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
2719 pool->dma_addr = dma_map_single(base->dev, pool->base,
2720 SZ_1K * base->num_phy_chans,
2722 if (dma_mapping_error(base->dev, pool->dma_addr)) {
2728 writel(virt_to_phys(base->lcla_pool.base),
2729 base->virtbase + D40_DREG_LCLA);
2735 static int __init d40_probe(struct platform_device *pdev)
2739 struct d40_base *base;
2740 struct resource *res = NULL;
2741 int num_reserved_chans;
2744 base = d40_hw_detect_init(pdev);
2749 num_reserved_chans = d40_phy_res_init(base);
2751 platform_set_drvdata(pdev, base);
2753 spin_lock_init(&base->interrupt_lock);
2754 spin_lock_init(&base->execmd_lock);
2756 /* Get IO for logical channel parameter address */
2757 res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
2760 d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
2763 base->lcpa_size = resource_size(res);
2764 base->phy_lcpa = res->start;
2766 if (request_mem_region(res->start, resource_size(res),
2767 D40_NAME " I/O lcpa") == NULL) {
2770 "Failed to request LCPA region 0x%x-0x%x\n",
2771 res->start, res->end);
2775 /* We make use of ESRAM memory for this. */
2776 val = readl(base->virtbase + D40_DREG_LCPA);
2777 if (res->start != val && val != 0) {
2778 dev_warn(&pdev->dev,
2779 "[%s] Mismatch LCPA dma 0x%x, def 0x%x\n",
2780 __func__, val, res->start);
2782 writel(res->start, base->virtbase + D40_DREG_LCPA);
2784 base->lcpa_base = ioremap(res->start, resource_size(res));
2785 if (!base->lcpa_base) {
2787 d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
2791 ret = d40_lcla_allocate(base);
2793 d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
2797 spin_lock_init(&base->lcla_pool.lock);
2799 base->irq = platform_get_irq(pdev, 0);
2801 ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
2803 d40_err(&pdev->dev, "No IRQ defined\n");
2807 err = d40_dmaengine_init(base, num_reserved_chans);
2813 dev_info(base->dev, "initialized\n");
2818 if (base->desc_slab)
2819 kmem_cache_destroy(base->desc_slab);
2821 iounmap(base->virtbase);
2823 if (base->lcla_pool.dma_addr)
2824 dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
2825 SZ_1K * base->num_phy_chans,
2828 if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
2829 free_pages((unsigned long)base->lcla_pool.base,
2830 base->lcla_pool.pages);
2832 kfree(base->lcla_pool.base_unaligned);
2835 release_mem_region(base->phy_lcpa,
2837 if (base->phy_start)
2838 release_mem_region(base->phy_start,
2841 clk_disable(base->clk);
2845 kfree(base->lcla_pool.alloc_map);
2846 kfree(base->lookup_log_chans);
2847 kfree(base->lookup_phy_chans);
2848 kfree(base->phy_res);
2852 d40_err(&pdev->dev, "probe failed\n");
2856 static struct platform_driver d40_driver = {
2858 .owner = THIS_MODULE,
2863 static int __init stedma40_init(void)
2865 return platform_driver_probe(&d40_driver, d40_probe);
2867 arch_initcall(stedma40_init);