2 * Copyright (c) 2013 Linaro Ltd.
3 * Copyright (c) 2013 Hisilicon Limited.
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 as
7 * published by the Free Software Foundation.
9 #include <linux/sched.h>
10 #include <linux/device.h>
11 #include <linux/dmaengine.h>
12 #include <linux/init.h>
13 #include <linux/interrupt.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/platform_device.h>
17 #include <linux/slab.h>
18 #include <linux/spinlock.h>
19 #include <linux/of_device.h>
21 #include <linux/clk.h>
22 #include <linux/of_dma.h>
26 #define DRIVER_NAME "k3-dma"
28 #define DMA_MAX_SIZE 0x1ffc
34 #define INT_TC1_MASK 0x18
35 #define INT_ERR1_MASK 0x20
36 #define INT_ERR2_MASK 0x24
37 #define INT_TC1_RAW 0x600
38 #define INT_ERR1_RAW 0x608
39 #define INT_ERR2_RAW 0x610
42 #define CX_CUR_CNT 0x704
49 #define AXI_CFG_DEFAULT 0x201201
51 #define CX_LLI_CHAIN_EN 0x2
53 #define CX_CFG_MEM2PER (0x1 << 2)
54 #define CX_CFG_PER2MEM (0x2 << 2)
55 #define CX_CFG_SRCINCR (0x1 << 31)
56 #define CX_CFG_DSTINCR (0x1 << 30)
67 struct k3_dma_desc_sw {
68 struct virt_dma_desc vd;
69 dma_addr_t desc_hw_lli;
72 struct k3_desc_hw desc_hw[0];
79 struct virt_dma_chan vc;
80 struct k3_dma_phy *phy;
81 struct list_head node;
82 enum dma_transfer_direction dir;
84 enum dma_status status;
90 struct k3_dma_chan *vchan;
91 struct k3_dma_desc_sw *ds_run;
92 struct k3_dma_desc_sw *ds_done;
96 struct dma_device slave;
98 struct tasklet_struct task;
100 struct list_head chan_pending;
101 struct k3_dma_phy *phy;
102 struct k3_dma_chan *chans;
108 #define to_k3_dma(dmadev) container_of(dmadev, struct k3_dma_dev, slave)
110 static struct k3_dma_chan *to_k3_chan(struct dma_chan *chan)
112 return container_of(chan, struct k3_dma_chan, vc.chan);
115 static void k3_dma_pause_dma(struct k3_dma_phy *phy, bool on)
120 val = readl_relaxed(phy->base + CX_CFG);
122 writel_relaxed(val, phy->base + CX_CFG);
124 val = readl_relaxed(phy->base + CX_CFG);
126 writel_relaxed(val, phy->base + CX_CFG);
130 static void k3_dma_terminate_chan(struct k3_dma_phy *phy, struct k3_dma_dev *d)
134 k3_dma_pause_dma(phy, false);
136 val = 0x1 << phy->idx;
137 writel_relaxed(val, d->base + INT_TC1_RAW);
138 writel_relaxed(val, d->base + INT_ERR1_RAW);
139 writel_relaxed(val, d->base + INT_ERR2_RAW);
142 static void k3_dma_set_desc(struct k3_dma_phy *phy, struct k3_desc_hw *hw)
144 writel_relaxed(hw->lli, phy->base + CX_LLI);
145 writel_relaxed(hw->count, phy->base + CX_CNT);
146 writel_relaxed(hw->saddr, phy->base + CX_SRC);
147 writel_relaxed(hw->daddr, phy->base + CX_DST);
148 writel_relaxed(AXI_CFG_DEFAULT, phy->base + AXI_CFG);
149 writel_relaxed(hw->config, phy->base + CX_CFG);
152 static u32 k3_dma_get_curr_cnt(struct k3_dma_dev *d, struct k3_dma_phy *phy)
156 cnt = readl_relaxed(d->base + CX_CUR_CNT + phy->idx * 0x10);
161 static u32 k3_dma_get_curr_lli(struct k3_dma_phy *phy)
163 return readl_relaxed(phy->base + CX_LLI);
166 static u32 k3_dma_get_chan_stat(struct k3_dma_dev *d)
168 return readl_relaxed(d->base + CH_STAT);
171 static void k3_dma_enable_dma(struct k3_dma_dev *d, bool on)
174 /* set same priority */
175 writel_relaxed(0x0, d->base + CH_PRI);
178 writel_relaxed(0xffff, d->base + INT_TC1_MASK);
179 writel_relaxed(0xffff, d->base + INT_ERR1_MASK);
180 writel_relaxed(0xffff, d->base + INT_ERR2_MASK);
183 writel_relaxed(0x0, d->base + INT_TC1_MASK);
184 writel_relaxed(0x0, d->base + INT_ERR1_MASK);
185 writel_relaxed(0x0, d->base + INT_ERR2_MASK);
189 static irqreturn_t k3_dma_int_handler(int irq, void *dev_id)
191 struct k3_dma_dev *d = (struct k3_dma_dev *)dev_id;
192 struct k3_dma_phy *p;
193 struct k3_dma_chan *c;
194 u32 stat = readl_relaxed(d->base + INT_STAT);
195 u32 tc1 = readl_relaxed(d->base + INT_TC1);
196 u32 err1 = readl_relaxed(d->base + INT_ERR1);
197 u32 err2 = readl_relaxed(d->base + INT_ERR2);
203 if (likely(tc1 & BIT(i))) {
209 spin_lock_irqsave(&c->vc.lock, flags);
210 vchan_cookie_complete(&p->ds_run->vd);
211 p->ds_done = p->ds_run;
212 spin_unlock_irqrestore(&c->vc.lock, flags);
216 if (unlikely((err1 & BIT(i)) || (err2 & BIT(i))))
217 dev_warn(d->slave.dev, "DMA ERR\n");
220 writel_relaxed(irq_chan, d->base + INT_TC1_RAW);
221 writel_relaxed(err1, d->base + INT_ERR1_RAW);
222 writel_relaxed(err2, d->base + INT_ERR2_RAW);
225 tasklet_schedule(&d->task);
231 static int k3_dma_start_txd(struct k3_dma_chan *c)
233 struct k3_dma_dev *d = to_k3_dma(c->vc.chan.device);
234 struct virt_dma_desc *vd = vchan_next_desc(&c->vc);
239 if (BIT(c->phy->idx) & k3_dma_get_chan_stat(d))
243 struct k3_dma_desc_sw *ds =
244 container_of(vd, struct k3_dma_desc_sw, vd);
246 * fetch and remove request from vc->desc_issued
247 * so vc->desc_issued only contains desc pending
249 list_del(&ds->vd.node);
251 c->phy->ds_done = NULL;
253 k3_dma_set_desc(c->phy, &ds->desc_hw[0]);
256 c->phy->ds_done = NULL;
257 c->phy->ds_run = NULL;
261 static void k3_dma_tasklet(unsigned long arg)
263 struct k3_dma_dev *d = (struct k3_dma_dev *)arg;
264 struct k3_dma_phy *p;
265 struct k3_dma_chan *c, *cn;
266 unsigned pch, pch_alloc = 0;
268 /* check new dma request of running channel in vc->desc_issued */
269 list_for_each_entry_safe(c, cn, &d->slave.channels, vc.chan.device_node) {
270 spin_lock_irq(&c->vc.lock);
272 if (p && p->ds_done) {
273 if (k3_dma_start_txd(c)) {
274 /* No current txd associated with this channel */
275 dev_dbg(d->slave.dev, "pchan %u: free\n", p->idx);
276 /* Mark this channel free */
281 spin_unlock_irq(&c->vc.lock);
284 /* check new channel request in d->chan_pending */
285 spin_lock_irq(&d->lock);
286 for (pch = 0; pch < d->dma_channels; pch++) {
289 if (p->vchan == NULL && !list_empty(&d->chan_pending)) {
290 c = list_first_entry(&d->chan_pending,
291 struct k3_dma_chan, node);
292 /* remove from d->chan_pending */
293 list_del_init(&c->node);
294 pch_alloc |= 1 << pch;
295 /* Mark this channel allocated */
298 dev_dbg(d->slave.dev, "pchan %u: alloc vchan %p\n", pch, &c->vc);
301 spin_unlock_irq(&d->lock);
303 for (pch = 0; pch < d->dma_channels; pch++) {
304 if (pch_alloc & (1 << pch)) {
308 spin_lock_irq(&c->vc.lock);
310 spin_unlock_irq(&c->vc.lock);
316 static int k3_dma_alloc_chan_resources(struct dma_chan *chan)
321 static void k3_dma_free_chan_resources(struct dma_chan *chan)
323 struct k3_dma_chan *c = to_k3_chan(chan);
324 struct k3_dma_dev *d = to_k3_dma(chan->device);
327 spin_lock_irqsave(&d->lock, flags);
328 list_del_init(&c->node);
329 spin_unlock_irqrestore(&d->lock, flags);
331 vchan_free_chan_resources(&c->vc);
335 static enum dma_status k3_dma_tx_status(struct dma_chan *chan,
336 dma_cookie_t cookie, struct dma_tx_state *state)
338 struct k3_dma_chan *c = to_k3_chan(chan);
339 struct k3_dma_dev *d = to_k3_dma(chan->device);
340 struct k3_dma_phy *p;
341 struct virt_dma_desc *vd;
346 ret = dma_cookie_status(&c->vc.chan, cookie, state);
347 if (ret == DMA_SUCCESS)
350 spin_lock_irqsave(&c->vc.lock, flags);
355 * If the cookie is on our issue queue, then the residue is
358 vd = vchan_find_desc(&c->vc, cookie);
360 bytes = container_of(vd, struct k3_dma_desc_sw, vd)->size;
361 } else if ((!p) || (!p->ds_run)) {
364 struct k3_dma_desc_sw *ds = p->ds_run;
365 u32 clli = 0, index = 0;
367 bytes = k3_dma_get_curr_cnt(d, p);
368 clli = k3_dma_get_curr_lli(p);
369 index = (clli - ds->desc_hw_lli) / sizeof(struct k3_desc_hw);
370 for (; index < ds->desc_num; index++) {
371 bytes += ds->desc_hw[index].count;
373 if (!ds->desc_hw[index].lli)
377 spin_unlock_irqrestore(&c->vc.lock, flags);
378 dma_set_residue(state, bytes);
382 static void k3_dma_issue_pending(struct dma_chan *chan)
384 struct k3_dma_chan *c = to_k3_chan(chan);
385 struct k3_dma_dev *d = to_k3_dma(chan->device);
388 spin_lock_irqsave(&c->vc.lock, flags);
389 /* add request to vc->desc_issued */
390 if (vchan_issue_pending(&c->vc)) {
393 if (list_empty(&c->node)) {
394 /* if new channel, add chan_pending */
395 list_add_tail(&c->node, &d->chan_pending);
396 /* check in tasklet */
397 tasklet_schedule(&d->task);
398 dev_dbg(d->slave.dev, "vchan %p: issued\n", &c->vc);
401 spin_unlock(&d->lock);
403 dev_dbg(d->slave.dev, "vchan %p: nothing to issue\n", &c->vc);
404 spin_unlock_irqrestore(&c->vc.lock, flags);
407 static void k3_dma_fill_desc(struct k3_dma_desc_sw *ds, dma_addr_t dst,
408 dma_addr_t src, size_t len, u32 num, u32 ccfg)
410 if ((num + 1) < ds->desc_num)
411 ds->desc_hw[num].lli = ds->desc_hw_lli + (num + 1) *
412 sizeof(struct k3_desc_hw);
413 ds->desc_hw[num].lli |= CX_LLI_CHAIN_EN;
414 ds->desc_hw[num].count = len;
415 ds->desc_hw[num].saddr = src;
416 ds->desc_hw[num].daddr = dst;
417 ds->desc_hw[num].config = ccfg;
420 static struct dma_async_tx_descriptor *k3_dma_prep_memcpy(
421 struct dma_chan *chan, dma_addr_t dst, dma_addr_t src,
422 size_t len, unsigned long flags)
424 struct k3_dma_chan *c = to_k3_chan(chan);
425 struct k3_dma_desc_sw *ds;
432 num = DIV_ROUND_UP(len, DMA_MAX_SIZE);
433 ds = kzalloc(sizeof(*ds) + num * sizeof(ds->desc_hw[0]), GFP_ATOMIC);
435 dev_dbg(chan->device->dev, "vchan %p: kzalloc fail\n", &c->vc);
438 ds->desc_hw_lli = __virt_to_phys((unsigned long)&ds->desc_hw[0]);
444 /* default is memtomem, without calling device_control */
445 c->ccfg = CX_CFG_SRCINCR | CX_CFG_DSTINCR | CX_CFG_EN;
446 c->ccfg |= (0xf << 20) | (0xf << 24); /* burst = 16 */
447 c->ccfg |= (0x3 << 12) | (0x3 << 16); /* width = 64 bit */
451 copy = min_t(size_t, len, DMA_MAX_SIZE);
452 k3_dma_fill_desc(ds, dst, src, copy, num++, c->ccfg);
454 if (c->dir == DMA_MEM_TO_DEV) {
456 } else if (c->dir == DMA_DEV_TO_MEM) {
465 ds->desc_hw[num-1].lli = 0; /* end of link */
466 return vchan_tx_prep(&c->vc, &ds->vd, flags);
469 static struct dma_async_tx_descriptor *k3_dma_prep_slave_sg(
470 struct dma_chan *chan, struct scatterlist *sgl, unsigned int sglen,
471 enum dma_transfer_direction dir, unsigned long flags, void *context)
473 struct k3_dma_chan *c = to_k3_chan(chan);
474 struct k3_dma_desc_sw *ds;
475 size_t len, avail, total = 0;
476 struct scatterlist *sg;
477 dma_addr_t addr, src = 0, dst = 0;
483 for_each_sg(sgl, sg, sglen, i) {
484 avail = sg_dma_len(sg);
485 if (avail > DMA_MAX_SIZE)
486 num += DIV_ROUND_UP(avail, DMA_MAX_SIZE) - 1;
489 ds = kzalloc(sizeof(*ds) + num * sizeof(ds->desc_hw[0]), GFP_ATOMIC);
491 dev_dbg(chan->device->dev, "vchan %p: kzalloc fail\n", &c->vc);
494 ds->desc_hw_lli = __virt_to_phys((unsigned long)&ds->desc_hw[0]);
498 for_each_sg(sgl, sg, sglen, i) {
499 addr = sg_dma_address(sg);
500 avail = sg_dma_len(sg);
504 len = min_t(size_t, avail, DMA_MAX_SIZE);
506 if (dir == DMA_MEM_TO_DEV) {
509 } else if (dir == DMA_DEV_TO_MEM) {
514 k3_dma_fill_desc(ds, dst, src, len, num++, c->ccfg);
521 ds->desc_hw[num-1].lli = 0; /* end of link */
523 return vchan_tx_prep(&c->vc, &ds->vd, flags);
526 static int k3_dma_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
529 struct k3_dma_chan *c = to_k3_chan(chan);
530 struct k3_dma_dev *d = to_k3_dma(chan->device);
531 struct dma_slave_config *cfg = (void *)arg;
532 struct k3_dma_phy *p = c->phy;
534 u32 maxburst = 0, val = 0;
535 enum dma_slave_buswidth width = DMA_SLAVE_BUSWIDTH_UNDEFINED;
539 case DMA_SLAVE_CONFIG:
542 c->dir = cfg->direction;
543 if (c->dir == DMA_DEV_TO_MEM) {
544 c->ccfg = CX_CFG_DSTINCR;
545 c->dev_addr = cfg->src_addr;
546 maxburst = cfg->src_maxburst;
547 width = cfg->src_addr_width;
548 } else if (c->dir == DMA_MEM_TO_DEV) {
549 c->ccfg = CX_CFG_SRCINCR;
550 c->dev_addr = cfg->dst_addr;
551 maxburst = cfg->dst_maxburst;
552 width = cfg->dst_addr_width;
555 case DMA_SLAVE_BUSWIDTH_1_BYTE:
556 case DMA_SLAVE_BUSWIDTH_2_BYTES:
557 case DMA_SLAVE_BUSWIDTH_4_BYTES:
558 case DMA_SLAVE_BUSWIDTH_8_BYTES:
565 c->ccfg |= (val << 12) | (val << 16);
567 if ((maxburst == 0) || (maxburst > 16))
571 c->ccfg |= (val << 20) | (val << 24);
572 c->ccfg |= CX_CFG_MEM2PER | CX_CFG_EN;
574 /* specific request line */
575 c->ccfg |= c->vc.chan.chan_id << 4;
578 case DMA_TERMINATE_ALL:
579 dev_dbg(d->slave.dev, "vchan %p: terminate all\n", &c->vc);
581 /* Prevent this channel being scheduled */
583 list_del_init(&c->node);
584 spin_unlock(&d->lock);
586 /* Clear the tx descriptor lists */
587 spin_lock_irqsave(&c->vc.lock, flags);
588 vchan_get_all_descriptors(&c->vc, &head);
590 /* vchan is assigned to a pchan - stop the channel */
591 k3_dma_terminate_chan(p, d);
594 p->ds_run = p->ds_done = NULL;
596 spin_unlock_irqrestore(&c->vc.lock, flags);
597 vchan_dma_desc_free_list(&c->vc, &head);
601 dev_dbg(d->slave.dev, "vchan %p: pause\n", &c->vc);
602 if (c->status == DMA_IN_PROGRESS) {
603 c->status = DMA_PAUSED;
605 k3_dma_pause_dma(p, false);
608 list_del_init(&c->node);
609 spin_unlock(&d->lock);
615 dev_dbg(d->slave.dev, "vchan %p: resume\n", &c->vc);
616 spin_lock_irqsave(&c->vc.lock, flags);
617 if (c->status == DMA_PAUSED) {
618 c->status = DMA_IN_PROGRESS;
620 k3_dma_pause_dma(p, true);
621 } else if (!list_empty(&c->vc.desc_issued)) {
623 list_add_tail(&c->node, &d->chan_pending);
624 spin_unlock(&d->lock);
627 spin_unlock_irqrestore(&c->vc.lock, flags);
635 static void k3_dma_free_desc(struct virt_dma_desc *vd)
637 struct k3_dma_desc_sw *ds =
638 container_of(vd, struct k3_dma_desc_sw, vd);
643 static struct of_device_id k3_pdma_dt_ids[] = {
644 { .compatible = "hisilicon,k3-dma-1.0", },
647 MODULE_DEVICE_TABLE(of, k3_pdma_dt_ids);
649 static struct dma_chan *k3_of_dma_simple_xlate(struct of_phandle_args *dma_spec,
650 struct of_dma *ofdma)
652 struct k3_dma_dev *d = ofdma->of_dma_data;
653 unsigned int request = dma_spec->args[0];
655 if (request > d->dma_requests)
658 return dma_get_slave_channel(&(d->chans[request].vc.chan));
661 static int k3_dma_probe(struct platform_device *op)
663 struct k3_dma_dev *d;
664 const struct of_device_id *of_id;
665 struct resource *iores;
668 iores = platform_get_resource(op, IORESOURCE_MEM, 0);
672 d = devm_kzalloc(&op->dev, sizeof(*d), GFP_KERNEL);
676 d->base = devm_ioremap_resource(&op->dev, iores);
678 return PTR_ERR(d->base);
680 of_id = of_match_device(k3_pdma_dt_ids, &op->dev);
682 of_property_read_u32((&op->dev)->of_node,
683 "dma-channels", &d->dma_channels);
684 of_property_read_u32((&op->dev)->of_node,
685 "dma-requests", &d->dma_requests);
688 d->clk = devm_clk_get(&op->dev, NULL);
689 if (IS_ERR(d->clk)) {
690 dev_err(&op->dev, "no dma clk\n");
691 return PTR_ERR(d->clk);
694 irq = platform_get_irq(op, 0);
695 ret = devm_request_irq(&op->dev, irq,
696 k3_dma_int_handler, IRQF_DISABLED, DRIVER_NAME, d);
700 /* init phy channel */
701 d->phy = devm_kzalloc(&op->dev,
702 d->dma_channels * sizeof(struct k3_dma_phy), GFP_KERNEL);
706 for (i = 0; i < d->dma_channels; i++) {
707 struct k3_dma_phy *p = &d->phy[i];
710 p->base = d->base + i * 0x40;
713 INIT_LIST_HEAD(&d->slave.channels);
714 dma_cap_set(DMA_SLAVE, d->slave.cap_mask);
715 dma_cap_set(DMA_MEMCPY, d->slave.cap_mask);
716 d->slave.dev = &op->dev;
717 d->slave.device_alloc_chan_resources = k3_dma_alloc_chan_resources;
718 d->slave.device_free_chan_resources = k3_dma_free_chan_resources;
719 d->slave.device_tx_status = k3_dma_tx_status;
720 d->slave.device_prep_dma_memcpy = k3_dma_prep_memcpy;
721 d->slave.device_prep_slave_sg = k3_dma_prep_slave_sg;
722 d->slave.device_issue_pending = k3_dma_issue_pending;
723 d->slave.device_control = k3_dma_control;
724 d->slave.copy_align = DMA_ALIGN;
725 d->slave.chancnt = d->dma_requests;
727 /* init virtual channel */
728 d->chans = devm_kzalloc(&op->dev,
729 d->dma_requests * sizeof(struct k3_dma_chan), GFP_KERNEL);
730 if (d->chans == NULL)
733 for (i = 0; i < d->dma_requests; i++) {
734 struct k3_dma_chan *c = &d->chans[i];
736 c->status = DMA_IN_PROGRESS;
737 INIT_LIST_HEAD(&c->node);
738 c->vc.desc_free = k3_dma_free_desc;
739 vchan_init(&c->vc, &d->slave);
742 /* Enable clock before accessing registers */
743 ret = clk_prepare_enable(d->clk);
745 dev_err(&op->dev, "clk_prepare_enable failed: %d\n", ret);
749 k3_dma_enable_dma(d, true);
751 ret = dma_async_device_register(&d->slave);
755 ret = of_dma_controller_register((&op->dev)->of_node,
756 k3_of_dma_simple_xlate, d);
758 goto of_dma_register_fail;
760 spin_lock_init(&d->lock);
761 INIT_LIST_HEAD(&d->chan_pending);
762 tasklet_init(&d->task, k3_dma_tasklet, (unsigned long)d);
763 platform_set_drvdata(op, d);
764 dev_info(&op->dev, "initialized\n");
768 of_dma_register_fail:
769 dma_async_device_unregister(&d->slave);
773 static int k3_dma_remove(struct platform_device *op)
775 struct k3_dma_chan *c, *cn;
776 struct k3_dma_dev *d = platform_get_drvdata(op);
778 dma_async_device_unregister(&d->slave);
779 of_dma_controller_free((&op->dev)->of_node);
781 list_for_each_entry_safe(c, cn, &d->slave.channels, vc.chan.device_node) {
782 list_del(&c->vc.chan.device_node);
783 tasklet_kill(&c->vc.task);
785 tasklet_kill(&d->task);
786 clk_disable_unprepare(d->clk);
790 static int k3_dma_suspend(struct device *dev)
792 struct k3_dma_dev *d = dev_get_drvdata(dev);
795 stat = k3_dma_get_chan_stat(d);
797 dev_warn(d->slave.dev,
798 "chan %d is running fail to suspend\n", stat);
801 k3_dma_enable_dma(d, false);
802 clk_disable_unprepare(d->clk);
806 static int k3_dma_resume(struct device *dev)
808 struct k3_dma_dev *d = dev_get_drvdata(dev);
811 ret = clk_prepare_enable(d->clk);
813 dev_err(d->slave.dev, "clk_prepare_enable failed: %d\n", ret);
816 k3_dma_enable_dma(d, true);
820 SIMPLE_DEV_PM_OPS(k3_dma_pmops, k3_dma_suspend, k3_dma_resume);
822 static struct platform_driver k3_pdma_driver = {
825 .owner = THIS_MODULE,
827 .of_match_table = k3_pdma_dt_ids,
829 .probe = k3_dma_probe,
830 .remove = k3_dma_remove,
833 module_platform_driver(k3_pdma_driver);
835 MODULE_DESCRIPTION("Hisilicon k3 DMA Driver");
836 MODULE_ALIAS("platform:k3dma");
837 MODULE_LICENSE("GPL v2");