Re: [PATCH 05/12] dmaengine: Add STM32 DMA3 support

Amelie Delaunay <amelie.delaunay@xxxxxxxxxxx> · Mon, 20 May 2024 17:42:00 +0200

On 5/17/24 16:57, Frank Li wrote:
On Fri, May 17, 2024 at 11:42:17AM +0200, Amelie Delaunay wrote:
On 5/16/24 19:09, Frank Li wrote:
On Thu, May 16, 2024 at 05:25:58PM +0200, Amelie Delaunay wrote:
On 5/15/24 20:56, Frank Li wrote:
On Tue, Apr 23, 2024 at 02:32:55PM +0200, Amelie Delaunay wrote:
STM32 DMA3 driver supports the 3 hardware configurations of the STM32 DMA3
controller:
...
+	writel_relaxed(hwdesc->cdar, ddata->base + STM32_DMA3_CDAR(id));
+	writel_relaxed(hwdesc->cllr, ddata->base + STM32_DMA3_CLLR(id));
+
+	/* Clear any pending interrupts */
+	csr = readl_relaxed(ddata->base + STM32_DMA3_CSR(id));
+	if (csr & CSR_ALL_F)
+		writel_relaxed(csr, ddata->base + STM32_DMA3_CFCR(id));
+
+	stm32_dma3_chan_dump_reg(chan);
+
+	ccr = readl_relaxed(ddata->base + STM32_DMA3_CCR(id));
+	writel_relaxed(ccr | CCR_EN, ddata->base + STM32_DMA3_CCR(id));

This one should use writel instead of writel_relaxed because it need
dma_wmb() as barrier for preious write complete.

Frank

ddata->base is Device memory type thanks to ioremap() use, so it is strongly
ordered and non-cacheable.
DMA3 is outside CPU cluster, its registers are accessible through AHB bus.
dma_wmb() (in case of writel instead of writel_relaxed) is useless in that
case: it won't ensure the propagation on the bus is complete, and it will
have impacts on the system.
That's why CCR register is written once,  then it is read before CCR_EN is
set and being written again, with _relaxed(), because registers are behind a
bus, and ioremapped with Device memory type which ensures it is strongly
ordered and non-cacheable.

regardless memory map, writel_relaxed() just make sure io write and read is
orderred, not necessary order with other memory access. only readl and
writel make sure order with other memory read/write.

1. Write src_addr to descriptor
2. dma_wmb()
3. Write "ready" to descriptor
4. enable channel or doorbell by write a register.

if 4 use writel_relaxe(). because 3 write to DDR, which difference place of
mmio, 4 may happen before 3.  Your can refer axi order model.

4 have to use ONE writel(), to make sure 3 already write to DDR.

You need use at least one writel() to make sure all nornmal memory finish.

+    writel_relaxed(chan->swdesc->ccr, ddata->base + STM32_DMA3_CCR(id));
+    writel_relaxed(hwdesc->ctr1, ddata->base + STM32_DMA3_CTR1(id));
+    writel_relaxed(hwdesc->ctr2, ddata->base + STM32_DMA3_CTR2(id));
+    writel_relaxed(hwdesc->cbr1, ddata->base + STM32_DMA3_CBR1(id));
+    writel_relaxed(hwdesc->csar, ddata->base + STM32_DMA3_CSAR(id));
+    writel_relaxed(hwdesc->cdar, ddata->base + STM32_DMA3_CDAR(id));
+    writel_relaxed(hwdesc->cllr, ddata->base + STM32_DMA3_CLLR(id));

These writel_relaxed() are from descriptors to DMA3 registers (descriptors
being prepared "a long time ago" during _prep_).

You can't depend on "a long time ago" during _prep_. If later your driver
run at fast CPU. The execute time will be short.

All dma_map_sg and dma_alloc_coherence ... need at least one writel() to
make sure previous write actually reach to DDR.

Some data may not really reach DDR, when DMA already start transfer

Please ref linux kernel document:
	Documentation/memory-barriers.txt, line 1948.

In your issue_pending(), call this function to enable channel. So need
at least one writel().

As I said previously, DMA3 registers are outside CPU cluster, accessible
through AHB bus, and ddata->base to address registers is ioremapped as
Device memory type, non-cacheable and strongly ordered.

arch/arm/include/asm/io.h:
/*
* ioremap() and friends.
*
* ioremap() takes a resource address, and size.  Due to the ARM memory
* types, it is important to use the correct ioremap() function as each
* mapping has specific properties.
*
* Function		Memory type	Cacheability	Cache hint
* *ioremap()*		*Device*		*n/a*		*n/a*
* ioremap_cache()	Normal		Writeback	Read allocate
* ioremap_wc()		Normal		Non-cacheable	n/a
* ioremap_wt()		Normal		Non-cacheable	n/a
*
* All device mappings have the following properties:
* - no access speculation
* - no repetition (eg, on return from an exception)
* - number, order and size of accesses are maintained
* - unaligned accesses are "unpredictable"
* - writes may be delayed before they hit the endpoint device

On our platforms, we know that to ensure the writes have hit the endpoint
device (aka DMA3 registers), a read have to be done before.
And that's what is done before enabling the channel:

+    ccr = readl_relaxed(ddata->base + STM32_DMA3_CCR(id));
+    writel_relaxed(ccr | CCR_EN, ddata->base + STM32_DMA3_CCR(id));

If there was an issue in this part of the code, it means channel would be
started while it is wrongly programmed. In that case, DMA3 would raise a
User Setting Error interrupt and disable the channel. User Setting Error is
managed in this driver (USEF/stm32_dma3_check_user_setting()). And we never
had reached a situation.

I am not talking about registers IO read/write order. read_releax() and
write_relax() can guarantee IO read and write ordered. But not guarantee
order with normal memory's read and write.

Thank you for rephrasing. It wasn't clear to be that you were talking 
about the descriptors being read by the DMA controller when the channel 
is enabled.

I looked more closely at the Documentation/core-api/dma-api-howto.rst 
and Documentation/core-api/dma-api.rst and I noted "Consistent DMA 
memory does not preclude the usage of proper memory barriers." and "make 
sure to flush the processor's write buffers before telling devices to 
read that memory".
As DMA3 descriptors are allocated using dma_pool (consistent DMA 
mappings), indeed, there is a barrier missing to ensure CPU's write 
buffers are flushed.

But instead of using writel in stm32_dma3_chan_start() called by 
issue_pending(), as descriptors can be reused (DMA_CTRL_REUSE), I prefer 
to add an explicit __io_wmb() in stm32_dma3_chan_prep_hwdesc() function, 
when the last descriptor is written.

I'm going to push a v3 with discussed updates. Please then comment on 
the latest patchset version.

Amelie

You have not met problem doesn't means your code is correct. Please check
document:
	Documentation/memory-barriers.txt

+
+	chan->dma_status = DMA_IN_PROGRESS;
+
+	dev_dbg(chan2dev(chan), "vchan %pK: started\n", &chan->vchan);
+}
+
+static int stm32_dma3_chan_suspend(struct stm32_dma3_chan *chan, bool susp)
+{
+	struct stm32_dma3_ddata *ddata = to_stm32_dma3_ddata(chan);
+	u32 csr, ccr = readl_relaxed(ddata->base + STM32_DMA3_CCR(chan->id)) & ~CCR_EN;
+	int ret = 0;
+
+	if (susp)
+		ccr |= CCR_SUSP;
+	else
+		ccr &= ~CCR_SUSP;
+
+	writel_relaxed(ccr, ddata->base + STM32_DMA3_CCR(chan->id));
+
+	if (susp) {
+		ret = readl_relaxed_poll_timeout_atomic(ddata->base + STM32_DMA3_CSR(chan->id), csr,
+							csr & CSR_SUSPF, 1, 10);
+		if (!ret)
+			writel_relaxed(CFCR_SUSPF, ddata->base + STM32_DMA3_CFCR(chan->id));
+
+		stm32_dma3_chan_dump_reg(chan);
+	}
+
+	return ret;
+}
+
+static void stm32_dma3_chan_reset(struct stm32_dma3_chan *chan)
+{
+	struct stm32_dma3_ddata *ddata = to_stm32_dma3_ddata(chan);
+	u32 ccr = readl_relaxed(ddata->base + STM32_DMA3_CCR(chan->id)) & ~CCR_EN;
+
+	writel_relaxed(ccr |= CCR_RESET, ddata->base + STM32_DMA3_CCR(chan->id));
+}
+
+static int stm32_dma3_chan_stop(struct stm32_dma3_chan *chan)
+{
+	struct stm32_dma3_ddata *ddata = to_stm32_dma3_ddata(chan);
+	u32 ccr;
+	int ret = 0;
+
+	chan->dma_status = DMA_COMPLETE;
+
+	/* Disable interrupts */
+	ccr = readl_relaxed(ddata->base + STM32_DMA3_CCR(chan->id));
+	writel_relaxed(ccr & ~(CCR_ALLIE | CCR_EN), ddata->base + STM32_DMA3_CCR(chan->id));
+
+	if (!(ccr & CCR_SUSP) && (ccr & CCR_EN)) {
+		/* Suspend the channel */
+		ret = stm32_dma3_chan_suspend(chan, true);
+		if (ret)
+			dev_warn(chan2dev(chan), "%s: timeout, data might be lost\n", __func__);
+	}
+
+	/*
+	 * Reset the channel: this causes the reset of the FIFO and the reset of the channel
+	 * internal state, the reset of CCR_EN and CCR_SUSP bits.
+	 */
+	stm32_dma3_chan_reset(chan);
+
+	return ret;
+}
+
+static void stm32_dma3_chan_complete(struct stm32_dma3_chan *chan)
+{
+	if (!chan->swdesc)
+		return;
+
+	vchan_cookie_complete(&chan->swdesc->vdesc);
+	chan->swdesc = NULL;
+	stm32_dma3_chan_start(chan);
+}
+
+static irqreturn_t stm32_dma3_chan_irq(int irq, void *devid)
+{
+	struct stm32_dma3_chan *chan = devid;
+	struct stm32_dma3_ddata *ddata = to_stm32_dma3_ddata(chan);
+	u32 misr, csr, ccr;
+
+	spin_lock(&chan->vchan.lock);
+
+	misr = readl_relaxed(ddata->base + STM32_DMA3_MISR);
+	if (!(misr & MISR_MIS(chan->id))) {
+		spin_unlock(&chan->vchan.lock);
+		return IRQ_NONE;
+	}
+
+	csr = readl_relaxed(ddata->base + STM32_DMA3_CSR(chan->id));
+	ccr = readl_relaxed(ddata->base + STM32_DMA3_CCR(chan->id)) & CCR_ALLIE;
+
+	if (csr & CSR_TCF && ccr & CCR_TCIE) {
+		if (chan->swdesc->cyclic)
+			vchan_cyclic_callback(&chan->swdesc->vdesc);
+		else
+			stm32_dma3_chan_complete(chan);
+	}
+
+	if (csr & CSR_USEF && ccr & CCR_USEIE) {
+		dev_err(chan2dev(chan), "User setting error\n");
+		chan->dma_status = DMA_ERROR;
+		/* CCR.EN automatically cleared by HW */
+		stm32_dma3_check_user_setting(chan);
+		stm32_dma3_chan_reset(chan);
+	}
+
+	if (csr & CSR_ULEF && ccr & CCR_ULEIE) {
+		dev_err(chan2dev(chan), "Update link transfer error\n");
+		chan->dma_status = DMA_ERROR;
+		/* CCR.EN automatically cleared by HW */
+		stm32_dma3_chan_reset(chan);
+	}
+
+	if (csr & CSR_DTEF && ccr & CCR_DTEIE) {
+		dev_err(chan2dev(chan), "Data transfer error\n");
+		chan->dma_status = DMA_ERROR;
+		/* CCR.EN automatically cleared by HW */
+		stm32_dma3_chan_reset(chan);
+	}
+
+	/*
+	 * Half Transfer Interrupt may be disabled but Half Transfer Flag can be set,
+	 * ensure HTF flag to be cleared, with other flags.
+	 */
+	csr &= (ccr | CCR_HTIE);
+
+	if (csr)
+		writel_relaxed(csr, ddata->base + STM32_DMA3_CFCR(chan->id));
+
+	spin_unlock(&chan->vchan.lock);
+
+	return IRQ_HANDLED;
+}
+
+static int stm32_dma3_alloc_chan_resources(struct dma_chan *c)
+{
+	struct stm32_dma3_chan *chan = to_stm32_dma3_chan(c);
+	struct stm32_dma3_ddata *ddata = to_stm32_dma3_ddata(chan);
+	u32 id = chan->id, csemcr, ccid;
+	int ret;
+
+	ret = pm_runtime_resume_and_get(ddata->dma_dev.dev);
+	if (ret < 0)
+		return ret;

It doesn't prefer runtime pm get at alloc dma chan, many client driver
doesn't actual user dma when allocate dma chan.

Ideally, resume get when issue_pending. Please refer pl330.c.

You may add runtime pm later after enablement patch.

Frank

To well balance clock enable/disable, if pm_runtime_resume_and_get() (rather
than pm_runtime_get_sync() which doesn't decrement the counter in case of
error) is used when issue_pending, it means pm_runtime_put_sync() should be
done when transfer ends.

terminate_all is not always called, so put_sync can't be used only there, it
should be conditionnally used in terminate_all, but also in interrupt
handler, on error events and on transfer completion event, provided that it
is the last transfer complete event (last item of the linked-list).

For clients with high transfer rate, it means a lot of clock enable/disable.
Moreover, DMA3 clock is managed by Secure OS. So it means a lot of
non-secure/secure world transitions.

I prefer to keep the implementation as it is for now, and possibly propose
runtime pm improvement later, with autosuspend.

Autosuspend is perfered. we try to use pm_runtime_get/put at channel alloc
/free before, but this solution are rejected by community.

you can leave clock on for this enablement patch and add runtime pm later
time.

Frank

Current implementation leaves the clock off if no channel is requested. It
also disables the clock if platform is suspended.
I just took example from what is done in stm32 drivers.

I have further patches, not proposed in this series which adds a basic
support of DMA3. There will be improvements, including runtime pm, in next
series.

Amelie

Amelie

+
+	/* Ensure the channel is free */
+	if (chan->semaphore_mode &&
+	    readl_relaxed(ddata->base + STM32_DMA3_CSEMCR(chan->id)) & CSEMCR_SEM_MUTEX) {
+		ret = -EBUSY;
+		goto err_put_sync;
+	}
+
+	chan->lli_pool = dmam_pool_create(dev_name(&c->dev->device), c->device->dev,
+					  sizeof(struct stm32_dma3_hwdesc),
+					  __alignof__(struct stm32_dma3_hwdesc), 0);
+	if (!chan->lli_pool) {
+		dev_err(chan2dev(chan), "Failed to create LLI pool\n");
+		ret = -ENOMEM;
+		goto err_put_sync;
+	}
+
+	/* Take the channel semaphore */
+	if (chan->semaphore_mode) {
+		writel_relaxed(CSEMCR_SEM_MUTEX, ddata->base + STM32_DMA3_CSEMCR(id));
+		csemcr = readl_relaxed(ddata->base + STM32_DMA3_CSEMCR(id));
+		ccid = FIELD_GET(CSEMCR_SEM_CCID, csemcr);
+		/* Check that the channel is well taken */
+		if (ccid != CCIDCFGR_CID1) {
+			dev_err(chan2dev(chan), "Not under CID1 control (in-use by CID%d)\n", ccid);
+			ret = -EPERM;
+			goto err_pool_destroy;
+		}
+		dev_dbg(chan2dev(chan), "Under CID1 control (semcr=0x%08x)\n", csemcr);
+	}
+
+	return 0;
+
+err_pool_destroy:
+	dmam_pool_destroy(chan->lli_pool);
+	chan->lli_pool = NULL;
+
+err_put_sync:
+	pm_runtime_put_sync(ddata->dma_dev.dev);
+
+	return ret;
+}
+
+static void stm32_dma3_free_chan_resources(struct dma_chan *c)
+{
+	struct stm32_dma3_chan *chan = to_stm32_dma3_chan(c);
+	struct stm32_dma3_ddata *ddata = to_stm32_dma3_ddata(chan);
+	unsigned long flags;
+
+	/* Ensure channel is in idle state */
+	spin_lock_irqsave(&chan->vchan.lock, flags);
+	stm32_dma3_chan_stop(chan);
+	chan->swdesc = NULL;
+	spin_unlock_irqrestore(&chan->vchan.lock, flags);
+
+	vchan_free_chan_resources(to_virt_chan(c));
+
+	dmam_pool_destroy(chan->lli_pool);
+	chan->lli_pool = NULL;
+
+	/* Release the channel semaphore */
+	if (chan->semaphore_mode)
+		writel_relaxed(0, ddata->base + STM32_DMA3_CSEMCR(chan->id));
+
+	pm_runtime_put_sync(ddata->dma_dev.dev);
+
+	/* Reset configuration */
+	memset(&chan->dt_config, 0, sizeof(chan->dt_config));
+	memset(&chan->dma_config, 0, sizeof(chan->dma_config));
+}
+
+static struct dma_async_tx_descriptor *stm32_dma3_prep_slave_sg(struct dma_chan *c,
+								struct scatterlist *sgl,
+								unsigned int sg_len,
+								enum dma_transfer_direction dir,
+								unsigned long flags, void *context)
+{
+	struct stm32_dma3_chan *chan = to_stm32_dma3_chan(c);
+	struct stm32_dma3_swdesc *swdesc;
+	struct scatterlist *sg;
+	size_t len;
+	dma_addr_t sg_addr, dev_addr, src, dst;
+	u32 i, j, count, ctr1, ctr2;
+	int ret;
+
+	count = sg_len;
+	for_each_sg(sgl, sg, sg_len, i) {
+		len = sg_dma_len(sg);
+		if (len > STM32_DMA3_MAX_BLOCK_SIZE)
+			count += DIV_ROUND_UP(len, STM32_DMA3_MAX_BLOCK_SIZE) - 1;
+	}
+
+	swdesc = stm32_dma3_chan_desc_alloc(chan, count);
+	if (!swdesc)
+		return NULL;
+
+	/* sg_len and i correspond to the initial sgl; count and j correspond to the hwdesc LL */
+	j = 0;
+	for_each_sg(sgl, sg, sg_len, i) {
+		sg_addr = sg_dma_address(sg);
+		dev_addr = (dir == DMA_MEM_TO_DEV) ? chan->dma_config.dst_addr :
+						     chan->dma_config.src_addr;
+		len = sg_dma_len(sg);
+
+		do {
+			size_t chunk = min_t(size_t, len, STM32_DMA3_MAX_BLOCK_SIZE);
+
+			if (dir == DMA_MEM_TO_DEV) {
+				src = sg_addr;
+				dst = dev_addr;
+
+				ret = stm32_dma3_chan_prep_hw(chan, dir, &swdesc->ccr, &ctr1, &ctr2,
+							      src, dst, chunk);
+
+				if (FIELD_GET(CTR1_DINC, ctr1))
+					dev_addr += chunk;
+			} else { /* (dir == DMA_DEV_TO_MEM || dir == DMA_MEM_TO_MEM) */
+				src = dev_addr;
+				dst = sg_addr;
+
+				ret = stm32_dma3_chan_prep_hw(chan, dir, &swdesc->ccr, &ctr1, &ctr2,
+							      src, dst, chunk);
+
+				if (FIELD_GET(CTR1_SINC, ctr1))
+					dev_addr += chunk;
+			}
+
+			if (ret)
+				goto err_desc_free;
+
+			stm32_dma3_chan_prep_hwdesc(chan, swdesc, j, src, dst, chunk,
+						    ctr1, ctr2, j == (count - 1), false);
+
+			sg_addr += chunk;
+			len -= chunk;
+			j++;
+		} while (len);
+	}
+
+	/* Enable Error interrupts */
+	swdesc->ccr |= CCR_USEIE | CCR_ULEIE | CCR_DTEIE;
+	/* Enable Transfer state interrupts */
+	swdesc->ccr |= CCR_TCIE;
+
+	swdesc->cyclic = false;
+
+	return vchan_tx_prep(&chan->vchan, &swdesc->vdesc, flags);
+
+err_desc_free:
+	stm32_dma3_chan_desc_free(chan, swdesc);
+
+	return NULL;
+}
+
+static void stm32_dma3_caps(struct dma_chan *c, struct dma_slave_caps *caps)
+{
+	struct stm32_dma3_chan *chan = to_stm32_dma3_chan(c);
+
+	if (!chan->fifo_size) {
+		caps->max_burst = 0;
+		caps->src_addr_widths &= ~BIT(DMA_SLAVE_BUSWIDTH_8_BYTES);
+		caps->dst_addr_widths &= ~BIT(DMA_SLAVE_BUSWIDTH_8_BYTES);
+	} else {
+		/* Burst transfer should not exceed half of the fifo size */
+		caps->max_burst = chan->max_burst;
+		if (caps->max_burst < DMA_SLAVE_BUSWIDTH_8_BYTES) {
+			caps->src_addr_widths &= ~BIT(DMA_SLAVE_BUSWIDTH_8_BYTES);
+			caps->dst_addr_widths &= ~BIT(DMA_SLAVE_BUSWIDTH_8_BYTES);
+		}
+	}
+}
+
+static int stm32_dma3_config(struct dma_chan *c, struct dma_slave_config *config)
+{
+	struct stm32_dma3_chan *chan = to_stm32_dma3_chan(c);
+
+	memcpy(&chan->dma_config, config, sizeof(*config));
+
+	return 0;
+}
+
+static int stm32_dma3_terminate_all(struct dma_chan *c)
+{
+	struct stm32_dma3_chan *chan = to_stm32_dma3_chan(c);
+	unsigned long flags;
+	LIST_HEAD(head);
+
+	spin_lock_irqsave(&chan->vchan.lock, flags);
+
+	if (chan->swdesc) {
+		vchan_terminate_vdesc(&chan->swdesc->vdesc);
+		chan->swdesc = NULL;
+	}
+
+	stm32_dma3_chan_stop(chan);
+
+	vchan_get_all_descriptors(&chan->vchan, &head);
+
+	spin_unlock_irqrestore(&chan->vchan.lock, flags);
+	vchan_dma_desc_free_list(&chan->vchan, &head);
+
+	dev_dbg(chan2dev(chan), "vchan %pK: terminated\n", &chan->vchan);
+
+	return 0;
+}
+
+static void stm32_dma3_synchronize(struct dma_chan *c)
+{
+	struct stm32_dma3_chan *chan = to_stm32_dma3_chan(c);
+
+	vchan_synchronize(&chan->vchan);
+}
+
+static void stm32_dma3_issue_pending(struct dma_chan *c)
+{
+	struct stm32_dma3_chan *chan = to_stm32_dma3_chan(c);
+	unsigned long flags;
+
+	spin_lock_irqsave(&chan->vchan.lock, flags);
+
+	if (vchan_issue_pending(&chan->vchan) && !chan->swdesc) {
+		dev_dbg(chan2dev(chan), "vchan %pK: issued\n", &chan->vchan);
+		stm32_dma3_chan_start(chan);
+	}
+
+	spin_unlock_irqrestore(&chan->vchan.lock, flags);
+}
+
+static bool stm32_dma3_filter_fn(struct dma_chan *c, void *fn_param)
+{
+	struct stm32_dma3_chan *chan = to_stm32_dma3_chan(c);
+	struct stm32_dma3_ddata *ddata = to_stm32_dma3_ddata(chan);
+	struct stm32_dma3_dt_conf *conf = fn_param;
+	u32 mask, semcr;
+	int ret;
+
+	dev_dbg(c->device->dev, "%s(%s): req_line=%d ch_conf=%08x tr_conf=%08x\n",
+		__func__, dma_chan_name(c), conf->req_line, conf->ch_conf, conf->tr_conf);
+
+	if (!of_property_read_u32(c->device->dev->of_node, "dma-channel-mask", &mask))
+		if (!(mask & BIT(chan->id)))
+			return false;
+
+	ret = pm_runtime_resume_and_get(ddata->dma_dev.dev);
+	if (ret < 0)
+		return false;
+	semcr = readl_relaxed(ddata->base + STM32_DMA3_CSEMCR(chan->id));
+	pm_runtime_put_sync(ddata->dma_dev.dev);
+
+	/* Check if chan is free */
+	if (semcr & CSEMCR_SEM_MUTEX)
+		return false;
+
+	/* Check if chan fifo fits well */
+	if (FIELD_GET(STM32_DMA3_DT_FIFO, conf->ch_conf) != chan->fifo_size)
+		return false;
+
+	return true;
+}
+
+static struct dma_chan *stm32_dma3_of_xlate(struct of_phandle_args *dma_spec, struct of_dma *ofdma)
+{
+	struct stm32_dma3_ddata *ddata = ofdma->of_dma_data;
+	dma_cap_mask_t mask = ddata->dma_dev.cap_mask;
+	struct stm32_dma3_dt_conf conf;
+	struct stm32_dma3_chan *chan;
+	struct dma_chan *c;
+
+	if (dma_spec->args_count < 3) {
+		dev_err(ddata->dma_dev.dev, "Invalid args count\n");
+		return NULL;
+	}
+
+	conf.req_line = dma_spec->args[0];
+	conf.ch_conf = dma_spec->args[1];
+	conf.tr_conf = dma_spec->args[2];
+
+	if (conf.req_line >= ddata->dma_requests) {
+		dev_err(ddata->dma_dev.dev, "Invalid request line\n");
+		return NULL;
+	}
+
+	/* Request dma channel among the generic dma controller list */
+	c = dma_request_channel(mask, stm32_dma3_filter_fn, &conf);
+	if (!c) {
+		dev_err(ddata->dma_dev.dev, "No suitable channel found\n");
+		return NULL;
+	}
+
+	chan = to_stm32_dma3_chan(c);
+	chan->dt_config = conf;
+
+	return c;
+}
+
+static u32 stm32_dma3_check_rif(struct stm32_dma3_ddata *ddata)
+{
+	u32 chan_reserved, mask = 0, i, ccidcfgr, invalid_cid = 0;
+
+	/* Reserve Secure channels */
+	chan_reserved = readl_relaxed(ddata->base + STM32_DMA3_SECCFGR);
+
+	/*
+	 * CID filtering must be configured to ensure that the DMA3 channel will inherit the CID of
+	 * the processor which is configuring and using the given channel.
+	 * In case CID filtering is not configured, dma-channel-mask property can be used to
+	 * specify available DMA channels to the kernel.
+	 */
+	of_property_read_u32(ddata->dma_dev.dev->of_node, "dma-channel-mask", &mask);
+
+	/* Reserve !CID-filtered not in dma-channel-mask, static CID != CID1, CID1 not allowed */
+	for (i = 0; i < ddata->dma_channels; i++) {
+		ccidcfgr = readl_relaxed(ddata->base + STM32_DMA3_CCIDCFGR(i));
+
+		if (!(ccidcfgr & CCIDCFGR_CFEN)) { /* !CID-filtered */
+			invalid_cid |= BIT(i);
+			if (!(mask & BIT(i))) /* Not in dma-channel-mask */
+				chan_reserved |= BIT(i);
+		} else { /* CID-filtered */
+			if (!(ccidcfgr & CCIDCFGR_SEM_EN)) { /* Static CID mode */
+				if (FIELD_GET(CCIDCFGR_SCID, ccidcfgr) != CCIDCFGR_CID1)
+					chan_reserved |= BIT(i);
+			} else { /* Semaphore mode */
+				if (!FIELD_GET(CCIDCFGR_SEM_WLIST_CID1, ccidcfgr))
+					chan_reserved |= BIT(i);
+				ddata->chans[i].semaphore_mode = true;
+			}
+		}
+		dev_dbg(ddata->dma_dev.dev, "chan%d: %s mode, %s\n", i,
+			!(ccidcfgr & CCIDCFGR_CFEN) ? "!CID-filtered" :
+			ddata->chans[i].semaphore_mode ? "Semaphore" : "Static CID",
+			(chan_reserved & BIT(i)) ? "denied" :
+			mask & BIT(i) ? "force allowed" : "allowed");
+	}
+
+	if (invalid_cid)
+		dev_warn(ddata->dma_dev.dev, "chan%*pbl have invalid CID configuration\n",
+			 ddata->dma_channels, &invalid_cid);
+
+	return chan_reserved;
+}
+
+static const struct of_device_id stm32_dma3_of_match[] = {
+	{ .compatible = "st,stm32-dma3", },
+	{ /* sentinel */},
+};
+MODULE_DEVICE_TABLE(of, stm32_dma3_of_match);
+
+static int stm32_dma3_probe(struct platform_device *pdev)
+{
+	struct device_node *np = pdev->dev.of_node;
+	struct stm32_dma3_ddata *ddata;
+	struct reset_control *reset;
+	struct stm32_dma3_chan *chan;
+	struct dma_device *dma_dev;
+	u32 master_ports, chan_reserved, i, verr;
+	u64 hwcfgr;
+	int ret;
+
+	ddata = devm_kzalloc(&pdev->dev, sizeof(*ddata), GFP_KERNEL);
+	if (!ddata)
+		return -ENOMEM;
+	platform_set_drvdata(pdev, ddata);
+
+	dma_dev = &ddata->dma_dev;
+
+	ddata->base = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(ddata->base))
+		return PTR_ERR(ddata->base);
+
+	ddata->clk = devm_clk_get(&pdev->dev, NULL);
+	if (IS_ERR(ddata->clk))
+		return dev_err_probe(&pdev->dev, PTR_ERR(ddata->clk), "Failed to get clk\n");
+
+	reset = devm_reset_control_get_optional(&pdev->dev, NULL);
+	if (IS_ERR(reset))
+		return dev_err_probe(&pdev->dev, PTR_ERR(reset), "Failed to get reset\n");
+
+	ret = clk_prepare_enable(ddata->clk);
+	if (ret)
+		return dev_err_probe(&pdev->dev, ret, "Failed to enable clk\n");
+
+	reset_control_reset(reset);
+
+	INIT_LIST_HEAD(&dma_dev->channels);
+
+	dma_cap_set(DMA_SLAVE, dma_dev->cap_mask);
+	dma_cap_set(DMA_PRIVATE, dma_dev->cap_mask);
+	dma_dev->dev = &pdev->dev;
+	/*
+	 * This controller supports up to 8-byte buswidth depending on the port used and the
+	 * channel, and can only access address at even boundaries, multiple of the buswidth.
+	 */
+	dma_dev->copy_align = DMAENGINE_ALIGN_8_BYTES;
+	dma_dev->src_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
+				   BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
+				   BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) |
+				   BIT(DMA_SLAVE_BUSWIDTH_8_BYTES);
+	dma_dev->dst_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
+				   BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
+				   BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) |
+				   BIT(DMA_SLAVE_BUSWIDTH_8_BYTES);
+	dma_dev->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV) | BIT(DMA_MEM_TO_MEM);
+
+	dma_dev->descriptor_reuse = true;
+	dma_dev->max_sg_burst = STM32_DMA3_MAX_SEG_SIZE;
+	dma_dev->residue_granularity = DMA_RESIDUE_GRANULARITY_DESCRIPTOR;
+	dma_dev->device_alloc_chan_resources = stm32_dma3_alloc_chan_resources;
+	dma_dev->device_free_chan_resources = stm32_dma3_free_chan_resources;
+	dma_dev->device_prep_slave_sg = stm32_dma3_prep_slave_sg;
+	dma_dev->device_caps = stm32_dma3_caps;
+	dma_dev->device_config = stm32_dma3_config;
+	dma_dev->device_terminate_all = stm32_dma3_terminate_all;
+	dma_dev->device_synchronize = stm32_dma3_synchronize;
+	dma_dev->device_tx_status = dma_cookie_status;
+	dma_dev->device_issue_pending = stm32_dma3_issue_pending;
+
+	/* if dma_channels is not modified, get it from hwcfgr1 */
+	if (of_property_read_u32(np, "dma-channels", &ddata->dma_channels)) {
+		hwcfgr = readl_relaxed(ddata->base + STM32_DMA3_HWCFGR1);
+		ddata->dma_channels = FIELD_GET(G_NUM_CHANNELS, hwcfgr);
+	}
+
+	/* if dma_requests is not modified, get it from hwcfgr2 */
+	if (of_property_read_u32(np, "dma-requests", &ddata->dma_requests)) {
+		hwcfgr = readl_relaxed(ddata->base + STM32_DMA3_HWCFGR2);
+		ddata->dma_requests = FIELD_GET(G_MAX_REQ_ID, hwcfgr) + 1;
+	}
+
+	/* G_MASTER_PORTS, G_M0_DATA_WIDTH_ENC, G_M1_DATA_WIDTH_ENC in HWCFGR1 */
+	hwcfgr = readl_relaxed(ddata->base + STM32_DMA3_HWCFGR1);
+	master_ports = FIELD_GET(G_MASTER_PORTS, hwcfgr);
+
+	ddata->ports_max_dw[0] = FIELD_GET(G_M0_DATA_WIDTH_ENC, hwcfgr);
+	if (master_ports == AXI64 || master_ports == AHB32) /* Single master port */
+		ddata->ports_max_dw[1] = DW_INVALID;
+	else /* Dual master ports */
+		ddata->ports_max_dw[1] = FIELD_GET(G_M1_DATA_WIDTH_ENC, hwcfgr);
+
+	ddata->chans = devm_kcalloc(&pdev->dev, ddata->dma_channels, sizeof(*ddata->chans),
+				    GFP_KERNEL);
+	if (!ddata->chans) {
+		ret = -ENOMEM;
+		goto err_clk_disable;
+	}
+
+	chan_reserved = stm32_dma3_check_rif(ddata);
+
+	if (chan_reserved == GENMASK(ddata->dma_channels - 1, 0)) {
+		ret = -ENODEV;
+		dev_err_probe(&pdev->dev, ret, "No channel available, abort registration\n");
+		goto err_clk_disable;
+	}
+
+	/* G_FIFO_SIZE x=0..7 in HWCFGR3 and G_FIFO_SIZE x=8..15 in HWCFGR4 */
+	hwcfgr = readl_relaxed(ddata->base + STM32_DMA3_HWCFGR3);
+	hwcfgr |= ((u64)readl_relaxed(ddata->base + STM32_DMA3_HWCFGR4)) << 32;
+
+	for (i = 0; i < ddata->dma_channels; i++) {
+		if (chan_reserved & BIT(i))
+			continue;
+
+		chan = &ddata->chans[i];
+		chan->id = i;
+		chan->fifo_size = get_chan_hwcfg(i, G_FIFO_SIZE(i), hwcfgr);
+		/* If chan->fifo_size > 0 then half of the fifo size, else no burst when no FIFO */
+		chan->max_burst = (chan->fifo_size) ? (1 << (chan->fifo_size + 1)) / 2 : 0;
+		chan->vchan.desc_free = stm32_dma3_chan_vdesc_free;
+
+		vchan_init(&chan->vchan, dma_dev);
+	}
+
+	ret = dmaenginem_async_device_register(dma_dev);
+	if (ret)
+		goto err_clk_disable;
+
+	for (i = 0; i < ddata->dma_channels; i++) {
+		if (chan_reserved & BIT(i))
+			continue;
+
+		ret = platform_get_irq(pdev, i);
+		if (ret < 0)
+			goto err_clk_disable;
+
+		chan = &ddata->chans[i];
+		chan->irq = ret;
+
+		ret = devm_request_irq(&pdev->dev, chan->irq, stm32_dma3_chan_irq, 0,
+				       dev_name(chan2dev(chan)), chan);
+		if (ret) {
+			dev_err_probe(&pdev->dev, ret, "Failed to request channel %s IRQ\n",
+				      dev_name(chan2dev(chan)));
+			goto err_clk_disable;
+		}
+	}
+
+	ret = of_dma_controller_register(np, stm32_dma3_of_xlate, ddata);
+	if (ret) {
+		dev_err_probe(&pdev->dev, ret, "Failed to register controller\n");
+		goto err_clk_disable;
+	}
+
+	verr = readl_relaxed(ddata->base + STM32_DMA3_VERR);
+
+	pm_runtime_set_active(&pdev->dev);
+	pm_runtime_enable(&pdev->dev);
+	pm_runtime_get_noresume(&pdev->dev);
+	pm_runtime_put(&pdev->dev);
+
+	dev_info(&pdev->dev, "STM32 DMA3 registered rev:%lu.%lu\n",
+		 FIELD_GET(VERR_MAJREV, verr), FIELD_GET(VERR_MINREV, verr));
+
+	return 0;
+
+err_clk_disable:
+	clk_disable_unprepare(ddata->clk);
+
+	return ret;
+}
+
+static void stm32_dma3_remove(struct platform_device *pdev)
+{
+	pm_runtime_disable(&pdev->dev);
+}
+
+static int stm32_dma3_runtime_suspend(struct device *dev)
+{
+	struct stm32_dma3_ddata *ddata = dev_get_drvdata(dev);
+
+	clk_disable_unprepare(ddata->clk);
+
+	return 0;
+}
+
+static int stm32_dma3_runtime_resume(struct device *dev)
+{
+	struct stm32_dma3_ddata *ddata = dev_get_drvdata(dev);
+	int ret;
+
+	ret = clk_prepare_enable(ddata->clk);
+	if (ret)
+		dev_err(dev, "Failed to enable clk: %d\n", ret);
+
+	return ret;
+}
+
+static const struct dev_pm_ops stm32_dma3_pm_ops = {
+	SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend, pm_runtime_force_resume)
+	RUNTIME_PM_OPS(stm32_dma3_runtime_suspend, stm32_dma3_runtime_resume, NULL)
+};
+
+static struct platform_driver stm32_dma3_driver = {
+	.probe = stm32_dma3_probe,
+	.remove_new = stm32_dma3_remove,
+	.driver = {
+		.name = "stm32-dma3",
+		.of_match_table = stm32_dma3_of_match,
+		.pm = pm_ptr(&stm32_dma3_pm_ops),
+	},
+};
+
+static int __init stm32_dma3_init(void)
+{
+	return platform_driver_register(&stm32_dma3_driver);
+}
+
+subsys_initcall(stm32_dma3_init);
+
+MODULE_DESCRIPTION("STM32 DMA3 controller driver");
+MODULE_AUTHOR("Amelie Delaunay <amelie.delaunay@xxxxxxxxxxx>");
+MODULE_LICENSE("GPL");
--
2.25.1