Re: [PATCH v2 2/2] dmaengine: ma35d1-pdma: Add MA35D1 PDMA driver
From: Shan-Chun Hung
Date: Tue Nov 17 2026 - 02:05:18 EST
Hi Frank,
Thanks for your review.
On 11/17/2026 5:14 AM, Frank Li wrote:
On Mon, Nov 09, 2026 at 01:29:00PM +0800, Shan-Chun Hung wrote:I will use dma_alloc_coherent() for the hardware descriptors.
Add a driver for the PDMA controller found in the Nuvoton MA35D1 SoC....
It supports up to 10 channels with memory-to-memory, slave scatter-gather
and cyclic transfers.
Signed-off-by: Shan-Chun Hung <shanchun1218@xxxxxxxxx>
---
+
+static struct ma35d1_desc *ma35d1_desc_alloc(struct ma35d1_chan *ch, u32 sg_len)
+{
+ struct ma35d1_desc *d;
+
+ d = kzalloc(struct_size(d, sg, sg_len), GFP_ATOMIC);
+ if (!d)
+ return NULL;
Generally, we use dma_alloc_coherence() or dma pool for desc. There are
extra complex if use kzalloc and dma_map_single(), which will cause second
bounce buffer at swiotlb.
+
+ d->sg_len = sg_len;
+ d->dma_len = sizeof(*d->sg) * sg_len;
+ d->dma_addr = dma_map_single(ch->dev, d->sg, d->dma_len, DMA_TO_DEVICE);
+ if (dma_mapping_error(ch->dev, d->dma_addr)) {
+ dev_err(ch->dev, "dma mapping error for channel\n");
+ kfree(d);
+ return NULL;
+ }
+
+ return d;
+}
+
+static struct dma_async_tx_descriptor *
+ma35d1_desc_submit(struct ma35d1_chan *ch, struct ma35d1_desc *d, bool cyclic,
+ unsigned long tx_flags)
+{
+ dma_sync_single_for_device(ch->dev, d->dma_addr, d->dma_len, DMA_TO_DEVICE);
+ d->cyclic = cyclic;
+ d->ch = ch;
+
+ return vchan_tx_prep(&ch->vc, &d->vd, tx_flags);
+}
+
+static struct dma_async_tx_descriptor *
+ma35d1_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dst, dma_addr_t src,
+ size_t len, unsigned long tx_flags)
+{
+ int i;
+ size_t tlen, max_chunk;
+ struct ma35d1_chan *ch = to_ma35d1_pdma_chan(chan);
+ struct ma35d1_desc *d;
+ u32 sg_len, width = 1;
+
+ if (ch->request != 0) {
+ dev_err(ch->dev, "%s: channel is bound to a peripheral request\n", __func__);
+ return NULL;
+ }
+
+ if (!len) {
+ dev_err(ch->dev, "invalid memcpy length\n");
+ return NULL;
+ }
+
+ if (IS_ALIGNED(src, 4) && IS_ALIGNED(dst, 4) && IS_ALIGNED(len, 4))
+ width = 4;
where are field copy_align, you check it here.
The controller supports byte-wide transfers, so the minimum alignment
is one byte. I will set copy_align to DMAENGINE_ALIGN_1_BYTE explicitly.
The check selects a 4-byte transfer width when the source, destination,
and length are aligned; otherwise, the driver uses a 1-byte width.
The prep function does not submit the descriptor.
+
+ max_chunk = (size_t)PDMA_MAX_CHAN_BEATS * width;
+
+ sg_len = DIV_ROUND_UP_ULL(len, max_chunk);
+ d = ma35d1_desc_alloc(ch, sg_len);
+ if (!d)
+ return NULL;
+
+ for (i = 0; i < sg_len; i++) {
+ tlen = min_t(size_t, len, max_chunk);
+ d->sg[i].src = src;
+ d->sg[i].dst = dst;
+ d->sg[i].ctl = PDMA_WIDTH(width) | PDMA_TXCNT(tlen / width) | PDMA_TBINTDIS |
+ PDMA_OP_SCATTER;
+ d->sg[i].next = d->dma_addr + PDMA_CH_OFFSET(i + 1);
+ src += tlen;
+ dst += tlen;
+ len -= tlen;
+ }
+ d->sg[sg_len - 1].ctl &= ~(PDMA_OP_MSK | PDMA_TBINTDIS);
+ d->sg[sg_len - 1].ctl |= PDMA_OP_BASIC;
+
+ return ma35d1_desc_submit(ch, d, false, tx_flags);
Prep() don't submit request, dmaengine_submit() do it, move it submit()
call back.
ma35d1_desc_prep_done() (renamed from ma35d1_desc_submit()) prepares
the descriptor and calls vchan_tx_prep(), which sets tx_submit to
vchan_tx_submit. The actual submission happens when the client calls
dmaengine_submit().
I renamed the helper to avoid confusion; there is no functional change.
I will update the interrupt handler accordingly.>
+}
+
+static struct dma_async_tx_descriptor *
+ma35d1_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
+ u32 sg_len, enum dma_transfer_direction dir,
+ unsigned long tx_flags, void *context)
+{
+ int i, width;
+ struct ma35d1_chan *ch = to_ma35d1_pdma_chan(chan);
+ struct ma35d1_desc *d;
+ struct scatterlist *sgent;
+
+ width = ma35d1_slave_width(ch, dir);
+ if (width < 0)
+ return NULL;
+
+ if (!sg_len) {
+ dev_err(ch->dev, "SG entry length is zero\n");
+ return NULL;
+ }
+
+ for_each_sg(sgl, sgent, sg_len, i) {
+ unsigned int len = sg_dma_len(sgent);
+
+ if (!len || len % width || len / width > PDMA_MAX_CHAN_BEATS) {
+ dev_err(ch->dev, "invalid SG segment length %u\n", len);
+ return NULL;
+ }
+
+ if (!IS_ALIGNED(sg_dma_address(sgent), width)) {
+ dev_err(ch->dev, "SG address not aligned to DMA width\n");
+ return NULL;
+ }
+ }
+
+ d = ma35d1_desc_alloc(ch, sg_len);
+ if (!d)
+ return NULL;
+
+ for_each_sg(sgl, sgent, sg_len, i) {
+ if (dir == DMA_MEM_TO_DEV) {
+ d->sg[i].ctl = PDMA_WIDTH(width) | PDMA_DAFIX;
+ d->sg[i].src = sg_dma_address(sgent);
+ d->sg[i].dst = ch->cfg.dst_addr;
+ } else {
+ d->sg[i].ctl = PDMA_WIDTH(width) | PDMA_SAFIX;
+ d->sg[i].src = ch->cfg.src_addr;
+ d->sg[i].dst = sg_dma_address(sgent);
+ }
+ d->sg[i].ctl |= PDMA_TXCNT(sg_dma_len(sgent) / width) | PDMA_TBINTDIS |
+ PDMA_TXTYPE | PDMA_OP_SCATTER;
+ d->sg[i].next = d->dma_addr + PDMA_CH_OFFSET(i + 1);
+ }
+
+ d->sg[sg_len - 1].ctl &= ~(PDMA_OP_MSK | PDMA_TBINTDIS);
+ d->sg[sg_len - 1].ctl |= PDMA_OP_BASIC;
+
+ return ma35d1_desc_submit(ch, d, false, tx_flags);
+}
+
+static struct dma_async_tx_descriptor *
+ma35d1_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr, size_t buf_len,
+ size_t period_len, enum dma_transfer_direction dir,
+ unsigned long flags)
+{
+ int i, sg_len, width;
+ size_t offset = 0;
+ struct ma35d1_chan *ch = to_ma35d1_pdma_chan(chan);
+ struct ma35d1_desc *d;
+
+ width = ma35d1_slave_width(ch, dir);
+ if (width < 0)
+ return NULL;
+
+ if (!buf_len || !period_len || period_len / width > PDMA_MAX_CHAN_BEATS) {
+ dev_err(ch->dev, "Invalid buffer/period length\n");
+ return NULL;
+ }
+
+ if (buf_len % period_len) {
+ dev_err(ch->dev, "Buffer length not multiple of period length\n");
+ return NULL;
+ }
+
+ if (period_len % width) {
+ dev_err(ch->dev, "Period length must be a multiple of DMA width\n");
+ return NULL;
+ }
+
+ if (!IS_ALIGNED(dma_addr, width)) {
+ dev_err(ch->dev, "Buffer address not aligned to DMA width\n");
+ return NULL;
+ }
+
+ sg_len = buf_len / period_len;
+ d = ma35d1_desc_alloc(ch, sg_len);
+ if (!d)
+ return NULL;
+
+ /* Split the buffer into period size chunks. */
+ for (offset = 0, i = 0; offset < buf_len; offset += period_len, i++) {
+ if (dir == DMA_MEM_TO_DEV) {
+ d->sg[i].ctl = PDMA_WIDTH(width) | PDMA_TXCNT(period_len / width) |
+ PDMA_DAFIX;
+ d->sg[i].src = dma_addr + offset;
+ d->sg[i].dst = ch->cfg.dst_addr;
+ } else {
+ d->sg[i].ctl = PDMA_WIDTH(width) | PDMA_TXCNT(period_len / width) |
+ PDMA_SAFIX;
+ d->sg[i].src = ch->cfg.src_addr;
+ d->sg[i].dst = dma_addr + offset;
+ }
+ d->sg[i].ctl |= PDMA_TXTYPE | PDMA_OP_SCATTER;
+ d->sg[i].next = d->dma_addr + PDMA_CH_OFFSET(i + 1);
+ }
+ d->sg[sg_len - 1].next = d->dma_addr;
+
+ return ma35d1_desc_submit(ch, d, true, flags);
+}
+
+static struct dma_chan *ma35d1_of_xlate(struct of_phandle_args *dma_spec, struct of_dma *of)
+{
+ struct ma35d1_pdmadev *pdmadev = of->of_dma_data;
+ struct dma_chan *chan;
+ unsigned int chan_id, reqsel;
+
+ if (dma_spec->args_count != 2)
+ return NULL;
+
+ chan_id = dma_spec->args[0];
+ reqsel = dma_spec->args[1];
+ if (chan_id >= PDMA_MAX_CHANS || reqsel > GENMASK(7, 0))
+ return NULL;
+
+ chan = dma_get_slave_channel(&pdmadev->channels[chan_id].vc.chan);
+ if (chan)
+ pdmadev->channels[chan_id].request = reqsel;
+
+ return chan;
+}
+
+static void ma35d1_free_chan_resources(struct dma_chan *chan)
+{
+ struct ma35d1_chan *ch = to_ma35d1_pdma_chan(chan);
+
+ ma35d1_terminate_all(chan);
+ vchan_free_chan_resources(&ch->vc);
+ ch->request = 0;
+ memset(&ch->cfg, 0, sizeof(ch->cfg));
+}
+
+static void ma35d1_pdma_start(struct ma35d1_chan *ch)
+{
+ struct ma35d1_desc *d = ch->desc;
+
+ if (d->sg_len == 1 && !d->cyclic) {
+ writel(d->sg[0].src, ch->base + PDMA_CH_OFFSET(ch->id) + PDMA_DSCT_SA);
+ writel(d->sg[0].dst, ch->base + PDMA_CH_OFFSET(ch->id) + PDMA_DSCT_DA);
+ writel(d->sg[0].ctl, ch->base + PDMA_CH_OFFSET(ch->id) + PDMA_DSCT_CTL);
+ } else {
+ writel(d->dma_addr, ch->base + PDMA_CH_OFFSET(ch->id) + PDMA_DSCT_NEXT);
+ ma35d1_pdma_reg_clear(ch, PDMA_CH_OFFSET(ch->id) + PDMA_DSCT_CTL, PDMA_OP_MSK);
+ ma35d1_pdma_reg_set(ch, PDMA_CH_OFFSET(ch->id) + PDMA_DSCT_CTL, PDMA_OP_SCATTER);
+ }
+
+ /* Memory-to-memory has no peripheral request, so trigger it by software. */
+ if (ch->request == 0)
+ writel(BIT(ch->id), ch->base + PDMA_SWREQ);
+}
+
+static void ma35d1_pdma_start_desc(struct dma_chan *chan)
+{
+ struct ma35d1_chan *ch = to_ma35d1_pdma_chan(chan);
+ struct virt_dma_desc *vd;
+
+ vd = vchan_next_desc(&ch->vc);
+ if (!vd) {
+ ch->desc = NULL;
+ return;
+ }
+
+ list_del(&vd->node);
+ ch->desc = to_ma35d1_pdma_desc(vd);
+ ma35d1_pdma_setup_channel(ch);
+ ma35d1_pdma_start(ch);
+}
+
+static void ma35d1_issue_pending(struct dma_chan *chan)
+{
+ struct ma35d1_chan *ch = to_ma35d1_pdma_chan(chan);
+ unsigned long flags;
+
+ spin_lock_irqsave(&ch->vc.lock, flags);
+
+ if (vchan_issue_pending(&ch->vc) && !ch->desc)
+ ma35d1_pdma_start_desc(chan);
+
+ spin_unlock_irqrestore(&ch->vc.lock, flags);
+}
+
+static void ma35d1_pdma_complete(struct ma35d1_chan *ch)
+{
+ if (!ch->desc)
+ return;
+
+ vchan_cookie_complete(&ch->desc->vd);
+ ma35d1_pdma_start_desc(&ch->vc.chan);
+}
+
+static irqreturn_t ma35d1_pdma_interrupt(int irq, void *devid)
+{
+ int i;
+ bool handled = false;
+ struct ma35d1_pdmadev *pdmadev = devid;
+ struct ma35d1_chan *ch;
+
+ for (i = 0; i < PDMA_MAX_CHANS; i++) {
+ unsigned long flags;
+ bool done, abort;
+
+ ch = &pdmadev->channels[i];
+
+ spin_lock_irqsave(&ch->vc.lock, flags);
+
+ done = readl(ch->base + PDMA_TDSTS) & BIT(i);
+ abort = readl(ch->base + PDMA_ABTSTS) & BIT(i);
+
+ if (done || abort) {
+ handled = true;
+
+ /* Ack both flags first so a new descriptor is not completed twice. */
+ writel(BIT(i), ch->base + PDMA_TDSTS);
+ writel(BIT(i), ch->base + PDMA_ABTSTS);
done = readl(ch->base + PDMA_TDSTS);
abort = readl(ch->base + PDMA_ABTSTS);
writel(done, ch->base + PDMA_TDSTS);
writel(abort, ch->base + PDMA_ABTSTS);
for_each_bit(done | abort) {
...
}
I will move the IRQ request after register device().+
+ if (ch->desc) {
+ if (abort) {
+ ch->desc->vd.tx_result.result = DMA_TRANS_ABORTED;
+ ma35d1_pdma_complete(ch);
+ } else if (ch->desc->cyclic) {
+ vchan_cyclic_callback(&ch->desc->vd);
+ } else {
+ ma35d1_pdma_complete(ch);
+ }
+ }
+ }
+
+ spin_unlock_irqrestore(&ch->vc.lock, flags);
+ }
+
+ return handled ? IRQ_HANDLED : IRQ_NONE;
+}
+
+static int ma35d1_pdma_probe(struct platform_device *pdev)
+{
+ int i, ret;
+ struct clk *pdma_clk;
+ struct reset_control *rst;
+ struct device *dev = &pdev->dev;
+ struct device_node *node = dev->of_node;
+ struct ma35d1_chan *ch;
+ struct ma35d1_pdmadev *pdmadev;
+ void __iomem *reg_base;
+
+ pdma_clk = devm_clk_get_enabled(dev, NULL);
+ if (IS_ERR(pdma_clk))
+ return dev_err_probe(dev, PTR_ERR(pdma_clk), "Failed to get clk\n");
+
+ rst = devm_reset_control_get_optional_exclusive_deasserted(dev, NULL);
+ if (IS_ERR(rst))
+ return dev_err_probe(dev, PTR_ERR(rst), "Failed to get reset\n");
+
+ ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
+ if (ret)
+ return ret;
+
+ pdmadev = devm_kzalloc(dev, sizeof(*pdmadev), GFP_KERNEL);
+ if (!pdmadev)
+ return -ENOMEM;
+
+ pdmadev->irq = platform_get_irq(pdev, 0);
+ if (pdmadev->irq < 0)
+ return pdmadev->irq;
+
+ reg_base = devm_platform_ioremap_resource(pdev, 0);
+ if (IS_ERR(reg_base))
+ return dev_err_probe(dev, PTR_ERR(reg_base),
+ "Failed to map pdma global registers\n");
+
+ dma_cap_zero(pdmadev->ddev.cap_mask);
+ dma_cap_set(DMA_SLAVE, pdmadev->ddev.cap_mask);
+ dma_cap_set(DMA_PRIVATE, pdmadev->ddev.cap_mask);
+ dma_cap_set(DMA_MEMCPY, pdmadev->ddev.cap_mask);
+ dma_cap_set(DMA_CYCLIC, pdmadev->ddev.cap_mask);
+
+ pdmadev->ddev.dev = dev;
+ pdmadev->ddev.src_addr_widths = PDMA_SLAVE_BUSWIDTHS;
+ pdmadev->ddev.dst_addr_widths = PDMA_SLAVE_BUSWIDTHS;
+ pdmadev->ddev.directions = BIT(DMA_MEM_TO_DEV) | BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_MEM);
+ pdmadev->ddev.device_prep_dma_memcpy = ma35d1_prep_dma_memcpy;
+ pdmadev->ddev.device_prep_slave_sg = ma35d1_prep_slave_sg;
+ pdmadev->ddev.device_prep_dma_cyclic = ma35d1_prep_dma_cyclic;
+ pdmadev->ddev.device_free_chan_resources = ma35d1_free_chan_resources;
+ pdmadev->ddev.device_issue_pending = ma35d1_issue_pending;
+ pdmadev->ddev.device_tx_status = dma_cookie_status;
+ pdmadev->ddev.device_config = ma35d1_slave_config;
+ pdmadev->ddev.device_terminate_all = ma35d1_terminate_all;
+ pdmadev->ddev.device_synchronize = ma35d1_synchronize;
+
+ spin_lock_init(&pdmadev->lock);
+ INIT_LIST_HEAD(&pdmadev->ddev.channels);
+ for (i = 0; i < PDMA_MAX_CHANS; i++) {
+ ch = &pdmadev->channels[i];
+ ch->dev = dev;
+ ch->pdmadev = pdmadev;
+ ch->id = i;
+ ch->base = reg_base;
+ ch->vc.desc_free = ma35d1_pdma_desc_free;
+ vchan_init(&ch->vc, &pdmadev->ddev);
+ }
+
+ ret = devm_request_irq(dev, pdmadev->irq, ma35d1_pdma_interrupt, 0, pdev->name, pdmadev);
+ if (ret)
+ return dev_err_probe(dev, ret, "Failed to request IRQ\n");
Move irq request after register device()
I will use dmaenginem_async_device_register().
+
+ platform_set_drvdata(pdev, pdmadev);
+
+ ret = dma_async_device_register(&pdmadev->ddev);
dmaenginem_async_device_register();
I will use devm_of_dma_controller_register().
+ if (ret)
+ return ret;
+
+ ret = of_dma_controller_register(node, ma35d1_of_xlate, pdmadev);
devm_of_dma_controller_register();
Frank
+ if (ret) {
+ dma_async_device_unregister(&pdmadev->ddev);
+ for (i = 0; i < PDMA_MAX_CHANS; i++)
+ tasklet_kill(&pdmadev->channels[i].vc.task);
+ return dev_err_probe(dev, ret, "of_dma_controller_register failed\n");
+ }
+
+ return 0;
+}
+
+static void ma35d1_pdma_remove(struct platform_device *pdev)
+{
+ struct ma35d1_pdmadev *pdmadev = platform_get_drvdata(pdev);
+ void __iomem *base = pdmadev->channels[0].base;
+ int i;
+
+ /* Stop new requests before the hardware is quiesced. */
+ of_dma_controller_free(pdev->dev.of_node);
+ dma_async_device_unregister(&pdmadev->ddev);
+
+ writel(0, base + PDMA_CHCTL);
+ writel(0, base + PDMA_INTEN);
+ synchronize_irq(pdmadev->irq);
+
+ for (i = 0; i < PDMA_MAX_CHANS; i++)
+ tasklet_kill(&pdmadev->channels[i].vc.task);
+}
+
+static const struct of_device_id ma35d1_pdma_match[] = {
+ { .compatible = "nuvoton,ma35d1-pdma" },
+ { /* sentinel */ }
+};
+MODULE_DEVICE_TABLE(of, ma35d1_pdma_match);
+
+static struct platform_driver ma35d1_pdma_driver = {
+ .probe = ma35d1_pdma_probe,
+ .remove = ma35d1_pdma_remove,
+ .driver = {
+ .name = "ma35d1-pdma",
+ .of_match_table = ma35d1_pdma_match,
+ },
+};
+module_platform_driver(ma35d1_pdma_driver);
+
+MODULE_DESCRIPTION("PDMA controller driver for Nuvoton MA35D1");
+MODULE_AUTHOR("Shan-Chun Hung <shanchun1218@xxxxxxxxx>");
+MODULE_LICENSE("GPL");
--
2.25.1
Best Regards,
Shan-Chun