Skip to contentSkip to conversation

dmaengine: ste_dma40: Fix numerous accumulated bugs

PATCH Superseded Linus Walleij 3 participants 36 messages Sep 20 – Sep 22

Subsystems (2)

v5 of 8 Superseded Latest is v8 →

lore
Share & tools
lore

Linus Walleij

author maintainer
#0 lore raw

v5 cover dmaengine: ste_dma40: Fix numerous accumulated bugs

This series fixes 26 DMA40 bugs.

12 were found while reviewing the Ux500 LCLA SRAM power-domain conversion.

The cyclic transfer residue bug was exposed by Ux500 audio playback.

13 more issues were identified during review and hardware-manual audit.

The method taken is: whenever Sashiko complains: fix the bug it complains about if possible.

This has been boot tested on the Samsung Skomer device: DMA for MMC, wireless SDIO and UART still works after these patches, and DMA for audio started working.

Assisted-by LLM

---
Changes in v5:
- Respin to fix some further corner cases found by Sashiko as well
as can be done.
- Resynchronize cyclic callback tracking after failed hardware-pointer samples and limit unchanged-pointer interrupts to one callback. DMA40 has only one latched terminal-count bit, so extra periods coalesced while the pointer is unavailable and exact one-or-more full-buffer laps cannot be recovered without risking spurious callbacks.
- Count DMA40 status from channels owned by other SoC cores as handled without acknowledging it, so only status-less interrupts return IRQ_NONE and foreign DMA traffic cannot trigger spurious-IRQ disabling.
This is REALLY FRINGE but let's do our best anyway!
- Link to v4: https://lore.kernel.org/r/[email protected]

Changes in v4:
- Rework cyclic-transfer residue reporting to use the current memory-side hardware pointer and reject periods that need more than one LLI.
- Add cyclic callback recovery when terminal-count interrupts coalesce.
- Retire all issued descriptors and release software channel state when a runtime PM resume fails, while avoiding inaccessible hardware registers.
- Validate physical event IDs against the variant event-group limit rather than the physical channel count, as confirmed by the DB8500 manual.
- Link to v3: https://lore.kernel.org/r/[email protected]

Changes in v3:
- Add a fix so physical channels advertise their existing cyclic support.
- Add a cyclic-transfer residue fix so DMAengine PCM reports the correct hardware pointer and Ux500 audio playback does not stop with -EIO.
- In patch 3, preserve cookie completion order when the next queued transfer fails to start and retire failed cyclic descriptors.
- In patch 4, enable runtime PM before requesting the IRQ so pending interrupts can be acknowledged during probe.
- In patch 5, keep valid interrupt handling with CONFIG_PM disabled and only drop a runtime PM reference when one was acquired.
- Add checked runtime PM resume handling to channel operations.
- Add an IRQ status patch returning IRQ_NONE when no DMA40 interrupt was acknowledged.
- In patch 8, keep the IRQ disabled until hardware initialization has configured and cleared the DMA40 interrupt state.
- In the DMA registration unwind patch, free the IRQ before unregistering the DMA devices and drain initialized tasklets before releasing storage.
- Add a fix releasing the LCPA SRAM node reference after reading it.
- Move the disabled-channels binding restoration to its own series.
- Store memcpy channel mappings per controller and check the property read.
- Add validation for disabled physical channel indexes.
- Reject DMA specifiers that do not contain exactly three cells.
- Reject transfer direction changes after channel allocation so logical channel resource masks are released correctly.
- In the event-group bounds patch, use variant-specific limits so DB8540 event group 4 remains available.
- Add fixed-channel fixes that search later physical blocks and validate configured physical channel indexes.
- Move the generic DMAengine debugfs naming fix to its own series.
- Use Assisted-by: LLM for the existing assistance trailers.
- Rebase onto v7.3-rc1.
- Link to v2: https://lore.kernel.org/r/[email protected]

Changes in v2:
- In patch 1, complete the failed-start descriptor through the normal tasklet path and drop the runtime PM reference instead of freeing the submitted descriptor directly.
- Add an IRQ fix to avoid register access when DMA40 is runtime suspended.
- Add a probe ordering fix so DMA40 hardware is initialized before DMAengine devices are registered.
- Add a probe unwind fix so DMAengine registrations are released before freeing IRQ and LCLA resources.
- Add an LCLA allocation fix so __get_free_pages() and free_pages() use an allocation order instead of a raw page count.
- Add a probe unwind fix so ESRAM LCLA mappings are not released with free_pages().
- Add a device tree parsing fix so memcpy-channels cannot overflow the memcpy channel array.
- Add a dev_type bounds fix so derived event groups cannot overflow
phy_res or the priority/realtime register window.
- Add validation for fallback memcpy configurations so memcpy-channels entries cannot bypass the dev_type bounds checks.
- Add a DMAengine debugfs naming fix so drivers registering several DMAengine devices for one parent device do not trigger duplicate-name warnings.
- Link to v1: https://lore.kernel.org/r/[email protected]

---
Linus Walleij (23):
dmaengine: ste_dma40: Fix physical cyclic capability
dmaengine: ste_dma40: Fix cyclic transfer residue
dmaengine: ste_dma40: Recover coalesced cyclic callbacks
dmaengine: ste_dma40: Fix failed start cleanup
dmaengine: ste_dma40: Fix probe runtime PM disable
dmaengine: ste_dma40: Check runtime PM in IRQ
dmaengine: ste_dma40: Handle runtime PM resume errors
dmaengine: ste_dma40: Return IRQ_NONE without interrupt status
dmaengine: ste_dma40: Init hardware before registration
dmaengine: ste_dma40: Fix probe IRQ leak
dmaengine: ste_dma40: Fix DMA registration unwind
dmaengine: ste_dma40: Fix LCLA allocation order
dmaengine: ste_dma40: Fix probe LCLA free
dmaengine: ste_dma40: Put the LCPA SRAM node
dmaengine: ste_dma40: Fix memcpy channel parsing
dmaengine: ste_dma40: Validate disabled channel indexes
dmaengine: ste_dma40: Validate DMA specifier length
dmaengine: ste_dma40: Reject direction changes after allocation
dmaengine: ste_dma40: Fix logical channel bounds check
dmaengine: ste_dma40: Fix event group bounds
dmaengine: ste_dma40: Search all blocks for fixed logical channels
dmaengine: ste_dma40: Validate fixed physical channel indexes
dmaengine: ste_dma40: Validate memcpy configuration

 drivers/dma/ste_dma40.c | 537 +++++++++++++++++++++++++++++++++++++-----------
 1 file changed, 419 insertions(+), 118 deletions(-)

---
base-commit: cee9395acd8043be0644b25c34bfa86623f2b935
change-id: 20260820-dma40-fixes-b99af66002bf

Best regards,

Signature · 1 line
Linus Walleij <[email protected]>
author maintainer
#1 lore raw

PATCH v5 1/23 dmaengine: ste_dma40: Fix physical cyclic capability

d40_dmaengine_init() configures dma_both for physical channels that can handle both slave and memcpy transfers. Physical DMA40 channel setup has supported cyclic LLIs since cyclic transfer support was added, but the DMA_CYCLIC capability is set on dma_slave a second time instead of dma_both.

Set DMA_CYCLIC on dma_both so d40_ops_init() installs device_prep_dma_cyclic and physical channels advertise cyclic support.

Fixes 0c842b551063 ("dma40: cyclic xfer support")
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/dma/ste_dma40.c +1 -1
index 0d9ffa3e2663..e4d689c9eba8 100644
@@ -2897,7 +2897,7 @@ static int __init d40_dmaengine_init(struct d40_base *base,
 	dma_cap_zero(base->dma_both.cap_mask);
 	dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
 	dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
-	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
+	dma_cap_set(DMA_CYCLIC, base->dma_both.cap_mask);
 
 	d40_ops_init(base, &base->dma_both);
 	err = dmaenginem_async_device_register(&base->dma_both);
Signature · 1 line
2.55.0
Reviewed-by on v5 1/23
#2 lore raw

Re: [PATCH v5 01/23] dmaengine: ste_dma40: Fix physical cyclic capability

Quoted 6 lines · Linus Walleij

d40_dmaengine_init() configures dma_both for physical channels that can handle both slave and memcpy transfers. Physical DMA40 channel setup has supported cyclic LLIs since cyclic transfer support was added, but the DMA_CYCLIC capability is set on dma_slave a second time instead of dma_both.

Quoted 7 lines

Set DMA_CYCLIC on dma_both so d40_ops_init() installs device_prep_dma_cyclic and physical channels advertise cyclic support.

Fixes 0c842b551063 ("dma40: cyclic xfer support")
Assisted-by LLM

---

Missed my review by tags at v3
https://lore.kernel.org/all/aq2XrwzJDF8yZCbz@SMW015318/

Quoted 20 lines
 drivers/dma/ste_dma40.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/dma/ste_dma40.c +1 -2
index 0d9ffa3e2663..e4d689c9eba8 100644
@@ -2897,7 +2897,7 @@ static int __init d40_dmaengine_init(struct d40_base *base,
 	dma_cap_zero(base->dma_both.cap_mask);
 	dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
 	dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
-	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
+	dma_cap_set(DMA_CYCLIC, base->dma_both.cap_mask);
 
 	d40_ops_init(base, &base->dma_both);
 	err = dmaenginem_async_device_register(&base->dma_both);
 
--

2.55.0

author maintainer
#3 lore raw

PATCH v5 2/23 dmaengine: ste_dma40: Fix cyclic transfer residue

In reply to Linus Walleij

DMA40 reads residue from the element count of the currently active LLI. For a cyclic transfer this reports at most one period, not the bytes remaining until the cyclic buffer wraps.

Once DMA40 advertises burst granularity, DMAengine PCM uses this residue directly. For a four-period PCM buffer it consequently reports the hardware pointer near three periods after every period interrupt. ALSA eventually stops playback with -EIO although DMA period callbacks continue.

Calculate cyclic residue from the current memory-side hardware pointer instead. Read the destination pointer for capture and the source pointer for playback. Sample the split logical channel pointer coherently and retain the last valid residue during relink transitions.

This avoids counting terminal-count interrupts, which races with hardware advancing to the next LLI and cannot account for coalesced interrupt status. Also reject cyclic periods that expand into multiple LLIs because logical cyclic LLIs each request a terminal-count interrupt and would generate more than one callback per period.

Reject invalid cyclic geometries before dividing or constructing the scatterlist as well.

Fixes 15c606686541 ("dmaengine: ste_dma40: indicate granularity on channels")
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 104 +++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 99 insertions(+), 5 deletions(-)
drivers/dma/ste_dma40.c +99 -5
index e4d689c9eba8..c9983e600daf 100644
@@ -378,6 +378,9 @@ struct d40_lli_pool {
  * @lli_len: Number of llis of current descriptor.
  * @lli_current: Number of transferred llis.
  * @lcla_alloc: Number of LCLA entries allocated.
+ * @cyclic_dma_addr: Start address of the cyclic buffer.
+ * @cyclic_buf_len: Length of the cyclic buffer.
+ * @cyclic_residue: Last valid cyclic residue sample.
  * @txd: DMA engine struct. Used for among other things for communication
  * during a transfer.
  * @node: List entry.
@@ -396,6 +399,9 @@ struct d40_desc {
 	int				 lli_len;
 	int				 lli_current;
 	int				 lcla_alloc;
+	dma_addr_t			 cyclic_dma_addr;
+	size_t				 cyclic_buf_len;
+	size_t				 cyclic_residue;
 
 	struct dma_async_tx_descriptor	 txd;
 	struct list_head		 node;
@@ -1420,6 +1426,64 @@ static u32 d40_residue(struct d40_chan *d40c)
 	return num_elt * d40c->dma_cfg.dst_info.data_width;
 }
 
+static bool d40_current_addr(struct d40_chan *d40c, dma_addr_t *addr)
+{
+	bool dst = d40c->dma_cfg.dir == DMA_DEV_TO_MEM;
+	void __iomem *high_reg;
+	void __iomem *low_reg;
+	u32 low;
+	u32 high;
+	u32 check;
+	int i;
+
+	if (chan_is_physical(d40c)) {
+		*addr = readl(chan_base(d40c) +
+			      (dst ? D40_CHAN_REG_SDPTR : D40_CHAN_REG_SSPTR));
+		return true;
+	}
+
+	if (dst) {
+		low_reg = &d40c->lcpa->lcsp2;
+		high_reg = &d40c->lcpa->lcsp3;
+	} else {
+		low_reg = &d40c->lcpa->lcsp0;
+		high_reg = &d40c->lcpa->lcsp1;
+	}
+
+	for (i = 0; i < 3; i++) {
+		high = readl(high_reg) & D40_MEM_LCSP1_SPTR_MASK;
+		low = readl(low_reg) & D40_MEM_LCSP0_SPTR_MASK;
+		check = readl(high_reg) & D40_MEM_LCSP1_SPTR_MASK;
+		if (high == check) {
+			*addr = low | high;
+			return true;
+		}
+	}
+
+	return false;
+}
+
+static bool d40_cyclic_offset(struct d40_chan *d40c, struct d40_desc *d40d,
+			      size_t *offset)
+{
+	dma_addr_t current_addr;
+	dma_addr_t current_offset;
+	int i;
+
+	for (i = 0; i < 3; i++) {
+		if (!d40_current_addr(d40c, &current_addr))
+			continue;
+
+		current_offset = current_addr - d40d->cyclic_dma_addr;
+		if (current_offset <= d40d->cyclic_buf_len) {
+			*offset = current_offset;
+			return true;
+		}
+	}
+
+	return false;
+}
+
 static bool d40_tx_is_linked(struct d40_chan *d40c)
 {
 	bool is_link;
@@ -1566,6 +1630,7 @@ static void dma_tc_handle(struct d40_chan *d40c)
 			if (d40d->lli_current == d40d->lli_len)
 				d40d->lli_current = 0;
 		}
+
 	} else {
 		d40_lcla_free_all(d40c, d40d);
 
@@ -2108,15 +2173,26 @@ static bool d40_is_paused(struct d40_chan *d40c)
 
 }
 
-static u32 stedma40_residue(struct dma_chan *chan)
+static u32 stedma40_residue(struct dma_chan *chan, dma_cookie_t cookie)
 {
 	struct d40_chan *d40c =
 		container_of(chan, struct d40_chan, chan);
+	struct d40_desc *d40d;
+	size_t offset;
 	u32 bytes_left;
 	unsigned long flags;
 
 	spin_lock_irqsave(&d40c->lock, flags);
-	bytes_left = d40_residue(d40c);
+	d40d = d40_first_active_get(d40c);
+	if (d40d && d40d->txd.cookie == cookie && d40d->cyclic &&
+	    d40d->cyclic_buf_len) {
+		if (d40_cyclic_offset(d40c, d40d, &offset))
+			d40d->cyclic_residue = d40d->cyclic_buf_len - offset;
+		bytes_left = d40d->cyclic_residue;
+	} else {
+		bytes_left = d40_residue(d40c);
+	}
+
 	spin_unlock_irqrestore(&d40c->lock, flags);
 
 	return bytes_left;
@@ -2246,8 +2322,13 @@ d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
 	if (desc == NULL)
 		goto unlock;
 
-	if (sg_next(&sg_src[sg_len - 1]) == sg_src)
+	if (sg_next(&sg_src[sg_len - 1]) == sg_src) {
 		desc->cyclic = true;
+		if (desc->lli_len != sg_len) {
+			chan_err(chan, "Cyclic periods must fit in one LLI\n");
+			goto free_desc;
+		}
+	}
 
 	src_dev_addr = 0;
 	dst_dev_addr = 0;
@@ -2524,11 +2605,18 @@ dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
 		     size_t buf_len, size_t period_len,
 		     enum dma_transfer_direction direction, unsigned long flags)
 {
-	unsigned int periods = buf_len / period_len;
+	unsigned int periods;
 	struct dma_async_tx_descriptor *txd;
+	struct d40_desc *desc;
 	struct scatterlist *sg;
+	dma_addr_t buf_addr = dma_addr;
 	int i;
 
+	if (!buf_len || !period_len || buf_len % period_len)
+		return NULL;
+
+	periods = buf_len / period_len;
+
 	sg = kzalloc_objs(struct scatterlist, periods + 1, GFP_NOWAIT);
 	if (!sg)
 		return NULL;
@@ -2543,6 +2631,12 @@ dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
 
 	txd = d40_prep_sg(chan, sg, sg, periods, direction,
 			  DMA_PREP_INTERRUPT);
+	if (txd) {
+		desc = container_of(txd, struct d40_desc, txd);
+		desc->cyclic_dma_addr = buf_addr;
+		desc->cyclic_buf_len = buf_len;
+		desc->cyclic_residue = buf_len;
+	}
 
 	kfree(sg);
 
@@ -2563,7 +2657,7 @@ static enum dma_status d40_tx_status(struct dma_chan *chan,
 
 	ret = dma_cookie_status(chan, cookie, txstate);
 	if (ret != DMA_COMPLETE && txstate)
-		dma_set_residue(txstate, stedma40_residue(chan));
+		dma_set_residue(txstate, stedma40_residue(chan, cookie));
 
 	if (d40_is_paused(d40c))
 		ret = DMA_PAUSED;
Signature · 1 line
2.55.0
#4 lore raw

Re: [PATCH v5 02/23] dmaengine: ste_dma40: Fix cyclic transfer residue

Quoted 76 lines · Linus Walleij

DMA40 reads residue from the element count of the currently active LLI. For a cyclic transfer this reports at most one period, not the bytes remaining until the cyclic buffer wraps.

Once DMA40 advertises burst granularity, DMAengine PCM uses this residue directly. For a four-period PCM buffer it consequently reports the hardware pointer near three periods after every period interrupt. ALSA eventually stops playback with -EIO although DMA period callbacks continue.

Calculate cyclic residue from the current memory-side hardware pointer instead. Read the destination pointer for capture and the source pointer for playback. Sample the split logical channel pointer coherently and retain the last valid residue during relink transitions.

This avoids counting terminal-count interrupts, which races with hardware advancing to the next LLI and cannot account for coalesced interrupt status. Also reject cyclic periods that expand into multiple LLIs because logical cyclic LLIs each request a terminal-count interrupt and would generate more than one callback per period.

Reject invalid cyclic geometries before dividing or constructing the scatterlist as well.

Fixes 15c606686541 ("dmaengine: ste_dma40: indicate granularity on channels")
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 104 +++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 99 insertions(+), 5 deletions(-)
drivers/dma/ste_dma40.c +21 -0
index e4d689c9eba8..c9983e600daf 100644
@@ -378,6 +378,9 @@ struct d40_lli_pool {
  * @lli_len: Number of llis of current descriptor.
  * @lli_current: Number of transferred llis.
  * @lcla_alloc: Number of LCLA entries allocated.
+ * @cyclic_dma_addr: Start address of the cyclic buffer.
+ * @cyclic_buf_len: Length of the cyclic buffer.
+ * @cyclic_residue: Last valid cyclic residue sample.
  * @txd: DMA engine struct. Used for among other things for communication
  * during a transfer.
  * @node: List entry.
@@ -396,6 +399,9 @@ struct d40_desc {
 	int				 lli_len;
 	int				 lli_current;
 	int				 lcla_alloc;
+	dma_addr_t			 cyclic_dma_addr;
+	size_t				 cyclic_buf_len;
+	size_t				 cyclic_residue;
 
 	struct dma_async_tx_descriptor	 txd;
 	struct list_head		 node;
@@ -1420,6 +1426,64 @@ static u32 d40_residue(struct d40_chan *d40c)
 	return num_elt * d40c->dma_cfg.dst_info.data_width;
 }
 
+static bool d40_current_addr(struct d40_chan *d40c, dma_addr_t *addr)
+{
+	bool dst = d40c->dma_cfg.dir == DMA_DEV_TO_MEM;
+	void __iomem *high_reg;
+	void __iomem *low_reg;
+	u32 low;
+	u32 high;
+	u32 check;
+	int i;
+
+	if (chan_is_physical(d40c)) {
+		*addr = readl(chan_base(d40c) +
+			      (dst ? D40_CHAN_REG_SDPTR : D40_CHAN_REG_SSPTR));
+		return true;
+	}

drivers/dma/ste_dma40.c static u32 d40_residue(struct d40_chan *d40c) · line 1453

Quoted 10 lines
drivers/dma/ste_dma40.c +10 -0
+
+	if (dst) {
+		low_reg = &d40c->lcpa->lcsp2;
+		high_reg = &d40c->lcpa->lcsp3;
+	} else {
+		low_reg = &d40c->lcpa->lcsp0;
+		high_reg = &d40c->lcpa->lcsp1;
+	}
+
+	for (i = 0; i < 3; i++) {

what's means of 3, retry counter? can you define macro for it.

Quoted 37 lines
drivers/dma/ste_dma40.c +33 -0
+		high = readl(high_reg) & D40_MEM_LCSP1_SPTR_MASK;
+		low = readl(low_reg) & D40_MEM_LCSP0_SPTR_MASK;
+		check = readl(high_reg) & D40_MEM_LCSP1_SPTR_MASK;
+		if (high == check) {
+			*addr = low | high;
+			return true;
+		}
+	}
+
+	return false;
+}
+
+static bool d40_cyclic_offset(struct d40_chan *d40c, struct d40_desc *d40d,
+			      size_t *offset)
+{
+	dma_addr_t current_addr;
+	dma_addr_t current_offset;
+	int i;
+
+	for (i = 0; i < 3; i++) {
+		if (!d40_current_addr(d40c, &current_addr))
+			continue;
+
+		current_offset = current_addr - d40d->cyclic_dma_addr;
+		if (current_offset <= d40d->cyclic_buf_len) {
+			*offset = current_offset;
+			return true;
+		}
+	}
+
+	return false;
+}
+
 static bool d40_tx_is_linked(struct d40_chan *d40c)
 {
 	bool is_link;

drivers/dma/ste_dma40.c static void dma_tc_handle(struct d40_chan *d40c) · line 1633

Quoted 4 lines
drivers/dma/ste_dma40.c +1 -0
@@ -1566,6 +1630,7 @@ static void dma_tc_handle(struct d40_chan *d40c)
 			if (d40d->lli_current == d40d->lli_len)
 				d40d->lli_current = 0;
 		}
+

Nit: clean this unnessary changes.

Frank

Quoted 93 lines
drivers/dma/ste_dma40.c +34 -6
 	} else {
 		d40_lcla_free_all(d40c, d40d);
 
@@ -2108,15 +2173,26 @@ static bool d40_is_paused(struct d40_chan *d40c)
 
 }
 
-static u32 stedma40_residue(struct dma_chan *chan)
+static u32 stedma40_residue(struct dma_chan *chan, dma_cookie_t cookie)
 {
 	struct d40_chan *d40c =
 		container_of(chan, struct d40_chan, chan);
+	struct d40_desc *d40d;
+	size_t offset;
 	u32 bytes_left;
 	unsigned long flags;
 
 	spin_lock_irqsave(&d40c->lock, flags);
-	bytes_left = d40_residue(d40c);
+	d40d = d40_first_active_get(d40c);
+	if (d40d && d40d->txd.cookie == cookie && d40d->cyclic &&
+	    d40d->cyclic_buf_len) {
+		if (d40_cyclic_offset(d40c, d40d, &offset))
+			d40d->cyclic_residue = d40d->cyclic_buf_len - offset;
+		bytes_left = d40d->cyclic_residue;
+	} else {
+		bytes_left = d40_residue(d40c);
+	}
+
 	spin_unlock_irqrestore(&d40c->lock, flags);
 
 	return bytes_left;
@@ -2246,8 +2322,13 @@ d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
 	if (desc == NULL)
 		goto unlock;
 
-	if (sg_next(&sg_src[sg_len - 1]) == sg_src)
+	if (sg_next(&sg_src[sg_len - 1]) == sg_src) {
 		desc->cyclic = true;
+		if (desc->lli_len != sg_len) {
+			chan_err(chan, "Cyclic periods must fit in one LLI\n");
+			goto free_desc;
+		}
+	}
 
 	src_dev_addr = 0;
 	dst_dev_addr = 0;
@@ -2524,11 +2605,18 @@ dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
 		     size_t buf_len, size_t period_len,
 		     enum dma_transfer_direction direction, unsigned long flags)
 {
-	unsigned int periods = buf_len / period_len;
+	unsigned int periods;
 	struct dma_async_tx_descriptor *txd;
+	struct d40_desc *desc;
 	struct scatterlist *sg;
+	dma_addr_t buf_addr = dma_addr;
 	int i;
 
+	if (!buf_len || !period_len || buf_len % period_len)
+		return NULL;
+
+	periods = buf_len / period_len;
+
 	sg = kzalloc_objs(struct scatterlist, periods + 1, GFP_NOWAIT);
 	if (!sg)
 		return NULL;
@@ -2543,6 +2631,12 @@ dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
 
 	txd = d40_prep_sg(chan, sg, sg, periods, direction,
 			  DMA_PREP_INTERRUPT);
+	if (txd) {
+		desc = container_of(txd, struct d40_desc, txd);
+		desc->cyclic_dma_addr = buf_addr;
+		desc->cyclic_buf_len = buf_len;
+		desc->cyclic_residue = buf_len;
+	}
 
 	kfree(sg);
 
@@ -2563,7 +2657,7 @@ static enum dma_status d40_tx_status(struct dma_chan *chan,
 
 	ret = dma_cookie_status(chan, cookie, txstate);
 	if (ret != DMA_COMPLETE && txstate)
-		dma_set_residue(txstate, stedma40_residue(chan));
+		dma_set_residue(txstate, stedma40_residue(chan, cookie));
 
 	if (d40_is_paused(d40c))
 		ret = DMA_PAUSED;
 
--

2.55.0

author maintainer
#5 lore raw

PATCH v5 3/23 dmaengine: ste_dma40: Recover coalesced cyclic callbacks

In reply to Linus Walleij

DMA40 exposes one terminal-count status bit per channel. If more than one cyclic period completes before the interrupt handler clears the bit, the events coalesce and the driver schedules only one callback. Short audio periods can consequently lose period notifications.

Use the current memory-side pointer to find the period boundary reached since callbacks were last queued. Add every elapsed period to pending_tx so the tasklet invokes one callback for each of them.

If the pointer cannot be sampled, report one callback and mark the callback position invalid. At the next successful sample, resynchronize instead of deriving a count from stale state. Any additional periods coalesced while the pointer is unavailable cannot be recovered reliably.

DMA40 exposes one latched terminal-count status bit, not an event counter. If the pointer remains at the same boundary, hardware cannot distinguish a repeated status from an exact full-buffer lap. In this ambiguous case, report one callback rather than potentially enqueueing a whole buffer of spurious callbacks. Exact full laps, including multiple laps, therefore cannot be recovered by software.

The preceding cyclic-residue fix ensures that every cyclic period fits in one LLI, so each boundary corresponds to one client callback.

Fixes 0c842b551063 ("dma40: cyclic xfer support")
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 62 ++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 61 insertions(+), 1 deletion(-)
drivers/dma/ste_dma40.c +61 -1
index c9983e600daf..1f9e8c7249b1 100644
@@ -381,11 +381,14 @@ struct d40_lli_pool {
  * @cyclic_dma_addr: Start address of the cyclic buffer.
  * @cyclic_buf_len: Length of the cyclic buffer.
  * @cyclic_residue: Last valid cyclic residue sample.
+ * @cyclic_period_len: Length of one cyclic period.
+ * @cyclic_callback_pos: Position after the callbacks already queued.
  * @txd: DMA engine struct. Used for among other things for communication
  * during a transfer.
  * @node: List entry.
  * @is_in_client_list: true if the client owns this descriptor.
  * @cyclic: true if this is a cyclic job
+ * @cyclic_callback_pos_valid: Whether cyclic_callback_pos is reliable.
  *
  * This descriptor is used for both logical and physical transfers.
  */
@@ -402,12 +405,15 @@ struct d40_desc {
 	dma_addr_t			 cyclic_dma_addr;
 	size_t				 cyclic_buf_len;
 	size_t				 cyclic_residue;
+	size_t				 cyclic_period_len;
+	size_t				 cyclic_callback_pos;
 
 	struct dma_async_tx_descriptor	 txd;
 	struct list_head		 node;
 
 	bool				 is_in_client_list;
 	bool				 cyclic;
+	bool				 cyclic_callback_pos_valid;
 };
 
 /**
@@ -1484,6 +1490,55 @@ static bool d40_cyclic_offset(struct d40_chan *d40c, struct d40_desc *d40d,
 	return false;
 }
 
+static unsigned int d40_cyclic_periods_elapsed(struct d40_chan *d40c,
+					       struct d40_desc *d40d)
+{
+	size_t current_pos;
+	size_t offset;
+	unsigned int periods;
+
+	if (!d40d->cyclic_period_len)
+		return 1;
+
+	if (!d40_cyclic_offset(d40c, d40d, &offset)) {
+		d40d->cyclic_callback_pos_valid = false;
+		return 1;
+	}
+
+	current_pos = rounddown(offset, d40d->cyclic_period_len);
+	if (!d40_residue(d40c) && current_pos != offset)
+		current_pos += d40d->cyclic_period_len;
+	if (current_pos == d40d->cyclic_buf_len)
+		current_pos = 0;
+
+	if (!d40d->cyclic_callback_pos_valid) {
+		/*
+		 * One callback was reported without a reliable pointer.
+		 * Resynchronize instead of deriving periods from stale state.
+		 */
+		periods = 1;
+	} else if (current_pos > d40d->cyclic_callback_pos) {
+		periods = (current_pos - d40d->cyclic_callback_pos) /
+			d40d->cyclic_period_len;
+	} else if (current_pos < d40d->cyclic_callback_pos) {
+		periods = (d40d->cyclic_buf_len -
+			d40d->cyclic_callback_pos + current_pos) /
+			d40d->cyclic_period_len;
+	} else {
+		/*
+		 * The TC status is a single latched bit. An unchanged pointer
+		 * cannot distinguish a complete lap from a repeated interrupt,
+		 * so do not amplify it into a buffer's worth of callbacks.
+		 */
+		periods = 1;
+	}
+
+	d40d->cyclic_callback_pos = current_pos;
+	d40d->cyclic_callback_pos_valid = true;
+
+	return periods;
+}
+
 static bool d40_tx_is_linked(struct d40_chan *d40c)
 {
 	bool is_link;
@@ -1606,6 +1661,7 @@ static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
 static void dma_tc_handle(struct d40_chan *d40c)
 {
 	struct d40_desc *d40d;
+	unsigned int callbacks = 1;
 
 	/* Get first active entry from list */
 	d40d = d40_first_active_get(d40c);
@@ -1631,6 +1687,7 @@ static void dma_tc_handle(struct d40_chan *d40c)
 				d40d->lli_current = 0;
 		}
 
+		callbacks = d40_cyclic_periods_elapsed(d40c, d40d);
 	} else {
 		d40_lcla_free_all(d40c, d40d);
 
@@ -1651,7 +1708,7 @@ static void dma_tc_handle(struct d40_chan *d40c)
 		d40_desc_done(d40c, d40d);
 	}
 
-	d40c->pending_tx++;
+	d40c->pending_tx += callbacks;
 	tasklet_schedule(&d40c->tasklet);
 
 }
@@ -2636,6 +2693,9 @@ dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
 		desc->cyclic_dma_addr = buf_addr;
 		desc->cyclic_buf_len = buf_len;
 		desc->cyclic_residue = buf_len;
+		desc->cyclic_period_len = period_len;
+		desc->cyclic_callback_pos = 0;
+		desc->cyclic_callback_pos_valid = true;
 	}
 
 	kfree(sg);
Signature · 1 line
2.55.0
#6 lore raw

Re: [PATCH v5 03/23] dmaengine: ste_dma40: Recover coalesced cyclic callbacks

Quoted 72 lines · Linus Walleij

DMA40 exposes one terminal-count status bit per channel. If more than one cyclic period completes before the interrupt handler clears the bit, the events coalesce and the driver schedules only one callback. Short audio periods can consequently lose period notifications.

Use the current memory-side pointer to find the period boundary reached since callbacks were last queued. Add every elapsed period to pending_tx so the tasklet invokes one callback for each of them.

If the pointer cannot be sampled, report one callback and mark the callback position invalid. At the next successful sample, resynchronize instead of deriving a count from stale state. Any additional periods coalesced while the pointer is unavailable cannot be recovered reliably.

DMA40 exposes one latched terminal-count status bit, not an event counter. If the pointer remains at the same boundary, hardware cannot distinguish a repeated status from an exact full-buffer lap. In this ambiguous case, report one callback rather than potentially enqueueing a whole buffer of spurious callbacks. Exact full laps, including multiple laps, therefore cannot be recovered by software.

The preceding cyclic-residue fix ensures that every cyclic period fits in one LLI, so each boundary corresponds to one client callback.

Fixes 0c842b551063 ("dma40: cyclic xfer support")
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 62 ++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 61 insertions(+), 1 deletion(-)
drivers/dma/ste_dma40.c +8 -0
index c9983e600daf..1f9e8c7249b1 100644
@@ -381,11 +381,14 @@ struct d40_lli_pool {
  * @cyclic_dma_addr: Start address of the cyclic buffer.
  * @cyclic_buf_len: Length of the cyclic buffer.
  * @cyclic_residue: Last valid cyclic residue sample.
+ * @cyclic_period_len: Length of one cyclic period.
+ * @cyclic_callback_pos: Position after the callbacks already queued.
  * @txd: DMA engine struct. Used for among other things for communication
  * during a transfer.
  * @node: List entry.
  * @is_in_client_list: true if the client owns this descriptor.
  * @cyclic: true if this is a cyclic job
+ * @cyclic_callback_pos_valid: Whether cyclic_callback_pos is reliable.
  *
  * This descriptor is used for both logical and physical transfers.
  */
@@ -402,12 +405,15 @@ struct d40_desc {
 	dma_addr_t			 cyclic_dma_addr;
 	size_t				 cyclic_buf_len;
 	size_t				 cyclic_residue;
+	size_t				 cyclic_period_len;
+	size_t				 cyclic_callback_pos;
 
 	struct dma_async_tx_descriptor	 txd;
 	struct list_head		 node;
 
 	bool				 is_in_client_list;
 	bool				 cyclic;
+	bool				 cyclic_callback_pos_valid;
 };
 
 /**
@@ -1484,6 +1490,55 @@ static bool d40_cyclic_offset(struct d40_chan *d40c, struct d40_desc *d40d,
 	return false;
 }
 
+static unsigned int d40_cyclic_periods_elapsed(struct d40_chan *d40c,
+					       struct d40_desc *d40d)

drivers/dma/ste_dma40.c static bool d40_cyclic_offset(struct d40_chan *d40c, struct d40_desc *d40d, · line 1504

Quoted 10 lines
drivers/dma/ste_dma40.c +10 -0
+{
+	size_t current_pos;
+	size_t offset;
+	unsigned int periods;
+
+	if (!d40d->cyclic_period_len)
+		return 1;
+
+	if (!d40_cyclic_offset(d40c, d40d, &offset)) {
+		d40d->cyclic_callback_pos_valid = false;

looks like cyclic_callback_pos_valid is not useful. if d40_cyclic_offset() failure, always return 1.

Logically, can't reconfigized n*period_len. assume period is 4. if there are not additional register to recorder counter, you can't distringiush, 1, 5, 9, ... irq elapsed.

So we have to assume never elapse more than peroid.

Frank

Quoted 79 lines
drivers/dma/ste_dma40.c +43 -2
+		return 1;
+	}
+
+	current_pos = rounddown(offset, d40d->cyclic_period_len);
+	if (!d40_residue(d40c) && current_pos != offset)
+		current_pos += d40d->cyclic_period_len;
+	if (current_pos == d40d->cyclic_buf_len)
+		current_pos = 0;
+
+	if (!d40d->cyclic_callback_pos_valid) {
+		/*
+		 * One callback was reported without a reliable pointer.
+		 * Resynchronize instead of deriving periods from stale state.
+		 */
+		periods = 1;
+	} else if (current_pos > d40d->cyclic_callback_pos) {
+		periods = (current_pos - d40d->cyclic_callback_pos) /
+			d40d->cyclic_period_len;
+	} else if (current_pos < d40d->cyclic_callback_pos) {
+		periods = (d40d->cyclic_buf_len -
+			d40d->cyclic_callback_pos + current_pos) /
+			d40d->cyclic_period_len;
+	} else {
+		/*
+		 * The TC status is a single latched bit. An unchanged pointer
+		 * cannot distinguish a complete lap from a repeated interrupt,
+		 * so do not amplify it into a buffer's worth of callbacks.
+		 */
+		periods = 1;
+	}
+
+	d40d->cyclic_callback_pos = current_pos;
+	d40d->cyclic_callback_pos_valid = true;
+
+	return periods;
+}
+
 static bool d40_tx_is_linked(struct d40_chan *d40c)
 {
 	bool is_link;
@@ -1606,6 +1661,7 @@ static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
 static void dma_tc_handle(struct d40_chan *d40c)
 {
 	struct d40_desc *d40d;
+	unsigned int callbacks = 1;
 
 	/* Get first active entry from list */
 	d40d = d40_first_active_get(d40c);
@@ -1631,6 +1687,7 @@ static void dma_tc_handle(struct d40_chan *d40c)
 				d40d->lli_current = 0;
 		}
 
+		callbacks = d40_cyclic_periods_elapsed(d40c, d40d);
 	} else {
 		d40_lcla_free_all(d40c, d40d);
 
@@ -1651,7 +1708,7 @@ static void dma_tc_handle(struct d40_chan *d40c)
 		d40_desc_done(d40c, d40d);
 	}
 
-	d40c->pending_tx++;
+	d40c->pending_tx += callbacks;
 	tasklet_schedule(&d40c->tasklet);
 
 }
@@ -2636,6 +2693,9 @@ dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
 		desc->cyclic_dma_addr = buf_addr;
 		desc->cyclic_buf_len = buf_len;
 		desc->cyclic_residue = buf_len;
+		desc->cyclic_period_len = period_len;
+		desc->cyclic_callback_pos = 0;
+		desc->cyclic_callback_pos_valid = true;
 	}
 
 	kfree(sg);
 
--

2.55.0

author maintainer
#7 lore raw

Re: [PATCH v5 03/23] dmaengine: ste_dma40: Recover coalesced cyclic callbacks

Quoted 16 lines · Frank Li

drivers/dma/ste_dma40.c static bool d40_cyclic_offset(struct d40_chan *d40c, struct d40_desc *d40d, · line 1504

Quoted 12 lines
drivers/dma/ste_dma40.c +12 -0
+static unsigned int d40_cyclic_periods_elapsed(struct d40_chan *d40c,
+                                            struct d40_desc *d40d)
+{
+     size_t current_pos;
+     size_t offset;
+     unsigned int periods;
+
+     if (!d40d->cyclic_period_len)
+             return 1;
+
+     if (!d40_cyclic_offset(d40c, d40d, &offset)) {
+             d40d->cyclic_callback_pos_valid = false;

looks like cyclic_callback_pos_valid is not useful. if d40_cyclic_offset() failure, always return 1.

Quoted 5 lines

Logically, can't reconfigized n*period_len. assume period is 4. if there are not additional register to recorder counter, you can't distringiush, 1, 5, 9, ... irq elapsed.

So we have to assume never elapse more than peroid.

I'll try to make it clearer in the code, but I think it can detect the minimum buffer counts that we can see, if we have for example 4 buffer counts (we should see 4) if we are one period off that means we could have had 1, 5, 9 .. etc periods. If we are two periods off we could have had 2, 6, 10 ... etc periods.

If we wrapped around the whoe buffer one or several times there is no way we can see that :/

We report 2 missed periods even if it was actually 6 we missed for example, I'm just hoping it will mostly be 2 if it ever happens. It's not like this is a problem I have seen in practice, it's driven by Sashiko reviews.

Maybe this patch should be dropped altogether, it's just me chatting with Sashiko about things I don't think happens much.

Yours,
Linus Walleij

author maintainer
#8 lore raw

PATCH v5 4/23 dmaengine: ste_dma40: Fix failed start cleanup

In reply to Linus Walleij

If d40_start() fails after a queued descriptor has been moved to the active list, d40_queue_start() currently returns NULL without unwinding the transfer state or clearing the channel busy flag.

Fix this pre-existing error path by completing the descriptor through the normal tasklet path, clearing the busy flag, balancing the runtime PM reference and returning an error pointer to distinguish the failure from the no-work case. Do not free the descriptor directly, since it has already been submitted and has a DMA cookie.

When starting the next queued transfer from the completion handler, put the completed descriptor on the done list first. This preserves FIFO completion order if the new transfer fails to start and prevents the completed cookie from moving backwards.

Use done-list membership rather than the cyclic flag to identify terminal descriptors in the tasklet. A cyclic descriptor that failed to start is then completed and removed instead of remaining permanently at the head of the done list.

Fixes 7d83a854a1a4 ("dma40: remove "hardware link with previous jobs" code")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 25 +++++++++++++++++--------
 1 file changed, 17 insertions(+), 8 deletions(-)
drivers/dma/ste_dma40.c +17 -8
index 1f9e8c7249b1..280a0d2f16c5 100644
@@ -1650,8 +1650,15 @@ static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
 		/* Start dma job */
 		err = d40_start(d40c);
 
-		if (err)
-			return NULL;
+		if (err) {
+			d40_desc_remove(d40d);
+			d40_desc_done(d40c, d40d);
+			d40c->pending_tx++;
+			d40c->busy = false;
+			pm_runtime_put_autosuspend(d40c->base->dev);
+			tasklet_schedule(&d40c->tasklet);
+			return ERR_PTR(err);
+		}
 	}
 
 	return d40d;
@@ -1698,14 +1705,14 @@ static void dma_tc_handle(struct d40_chan *d40c)
 			return;
 		}
 
+		d40_desc_remove(d40d);
+		d40_desc_done(d40c, d40d);
+
 		if (d40_queue_start(d40c) == NULL) {
 			d40c->busy = false;
 
 			pm_runtime_put_autosuspend(d40c->base->dev);
 		}
-
-		d40_desc_remove(d40d);
-		d40_desc_done(d40c, d40d);
 	}
 
 	d40c->pending_tx += callbacks;
@@ -1719,20 +1726,22 @@ static void dma_tasklet(struct tasklet_struct *t)
 	struct d40_desc *d40d;
 	unsigned long flags;
 	bool callback_active;
+	bool from_done;
 	struct dmaengine_desc_callback cb;
 
 	spin_lock_irqsave(&d40c->lock, flags);
 
 	/* Get first entry from the done list */
 	d40d = d40_first_done(d40c);
-	if (d40d == NULL) {
+	from_done = !!d40d;
+	if (!from_done) {
 		/* Check if we have reached here for cyclic job */
 		d40d = d40_first_active_get(d40c);
 		if (d40d == NULL || !d40d->cyclic)
 			goto check_pending_tx;
 	}
 
-	if (!d40d->cyclic)
+	if (from_done)
 		dma_cookie_complete(&d40d->txd);
 
 	/*
@@ -1748,7 +1757,7 @@ static void dma_tasklet(struct tasklet_struct *t)
 	callback_active = !!(d40d->txd.flags & DMA_PREP_INTERRUPT);
 	dmaengine_desc_get_callback(&d40d->txd, &cb);
 
-	if (!d40d->cyclic) {
+	if (from_done) {
 		if (async_tx_test_ack(&d40d->txd)) {
 			d40_desc_remove(d40d);
 			d40_desc_free(d40c, d40d);
Signature · 1 line
2.55.0
Reviewed-by on v5 4/23
#9 lore raw

Re: [PATCH v5 04/23] dmaengine: ste_dma40: Fix failed start cleanup

Quoted 20 lines · Linus Walleij

If d40_start() fails after a queued descriptor has been moved to the active list, d40_queue_start() currently returns NULL without unwinding the transfer state or clearing the channel busy flag.

Fix this pre-existing error path by completing the descriptor through the normal tasklet path, clearing the busy flag, balancing the runtime PM reference and returning an error pointer to distinguish the failure from the no-work case. Do not free the descriptor directly, since it has already been submitted and has a DMA cookie.

When starting the next queued transfer from the completion handler, put the completed descriptor on the done list first. This preserves FIFO completion order if the new transfer fails to start and prevents the completed cookie from moving backwards.

Use done-list membership rather than the cyclic flag to identify terminal descriptors in the tasklet. A cyclic descriptor that failed to start is then completed and removed instead of remaining permanently at the head of the done list.

Quoted 6 lines
Fixes 7d83a854a1a4 ("dma40: remove "hardware link with previous jobs" code")
Reported-by [email protected]
Assisted-by LLM

---

Quoted 81 lines
 drivers/dma/ste_dma40.c | 25 +++++++++++++++++--------
 1 file changed, 17 insertions(+), 8 deletions(-)
drivers/dma/ste_dma40.c +17 -9
index 1f9e8c7249b1..280a0d2f16c5 100644
@@ -1650,8 +1650,15 @@ static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
 		/* Start dma job */
 		err = d40_start(d40c);
 
-		if (err)
-			return NULL;
+		if (err) {
+			d40_desc_remove(d40d);
+			d40_desc_done(d40c, d40d);
+			d40c->pending_tx++;
+			d40c->busy = false;
+			pm_runtime_put_autosuspend(d40c->base->dev);
+			tasklet_schedule(&d40c->tasklet);
+			return ERR_PTR(err);
+		}
 	}
 
 	return d40d;
@@ -1698,14 +1705,14 @@ static void dma_tc_handle(struct d40_chan *d40c)
 			return;
 		}
 
+		d40_desc_remove(d40d);
+		d40_desc_done(d40c, d40d);
+
 		if (d40_queue_start(d40c) == NULL) {
 			d40c->busy = false;
 
 			pm_runtime_put_autosuspend(d40c->base->dev);
 		}
-
-		d40_desc_remove(d40d);
-		d40_desc_done(d40c, d40d);
 	}
 
 	d40c->pending_tx += callbacks;
@@ -1719,20 +1726,22 @@ static void dma_tasklet(struct tasklet_struct *t)
 	struct d40_desc *d40d;
 	unsigned long flags;
 	bool callback_active;
+	bool from_done;
 	struct dmaengine_desc_callback cb;
 
 	spin_lock_irqsave(&d40c->lock, flags);
 
 	/* Get first entry from the done list */
 	d40d = d40_first_done(d40c);
-	if (d40d == NULL) {
+	from_done = !!d40d;
+	if (!from_done) {
 		/* Check if we have reached here for cyclic job */
 		d40d = d40_first_active_get(d40c);
 		if (d40d == NULL || !d40d->cyclic)
 			goto check_pending_tx;
 	}
 
-	if (!d40d->cyclic)
+	if (from_done)
 		dma_cookie_complete(&d40d->txd);
 
 	/*
@@ -1748,7 +1757,7 @@ static void dma_tasklet(struct tasklet_struct *t)
 	callback_active = !!(d40d->txd.flags & DMA_PREP_INTERRUPT);
 	dmaengine_desc_get_callback(&d40d->txd, &cb);
 
-	if (!d40d->cyclic) {
+	if (from_done) {
 		if (async_tx_test_ack(&d40d->txd)) {
 			d40_desc_remove(d40d);
 			d40_desc_free(d40c, d40d);
 
--

2.55.0

author maintainer
#10 lore raw

PATCH v5 5/23 dmaengine: ste_dma40: Fix probe runtime PM disable

In reply to Linus Walleij

Some d40_probe() error paths jump to destroy_cache before runtime PM has been enabled for the DMA controller device. The label unconditionally calls pm_runtime_disable(), which increments disable_depth even though this probe attempt never enabled runtime PM.

Track whether this probe attempt enabled runtime PM before disabling it on the error path. This is not about a later deferred-probe retry, since the driver is registered with platform_driver_probe(); it keeps the probe unwind balanced.

The interrupt handler uses pm_runtime_get_if_active() and cannot acknowledge a pending interrupt while runtime PM is disabled. Request the IRQ only after enabling runtime PM so the handler cannot enter an unacknowledged interrupt loop during probe.

Fixes 0618c077a8c2 ("dmaengine: ste_dma40: Fix PM disable depth imbalance in d40_probe")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 17 ++++++++++-------
 1 file changed, 10 insertions(+), 7 deletions(-)
drivers/dma/ste_dma40.c +10 -7
index 280a0d2f16c5..1d02226ab5ff 100644
@@ -3661,6 +3661,7 @@ static int __init d40_probe(struct platform_device *pdev)
 	struct resource *res;
 	struct resource res_lcpa;
 	int num_reserved_chans;
+	bool runtime_pm_enabled = false;
 	u32 val;
 	int ret;
 
@@ -3748,12 +3749,6 @@ static int __init d40_probe(struct platform_device *pdev)
 		goto destroy_cache;
 	}
 
-	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
-	if (ret) {
-		d40_err(dev, "No IRQ defined\n");
-		goto destroy_cache;
-	}
-
 	if (base->plat_data->use_esram_lcla) {
 
 		base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
@@ -3782,6 +3777,13 @@ static int __init d40_probe(struct platform_device *pdev)
 	pm_runtime_mark_last_busy(base->dev);
 	pm_runtime_set_active(base->dev);
 	pm_runtime_enable(base->dev);
+	runtime_pm_enabled = true;
+
+	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
+	if (ret) {
+		d40_err(dev, "No IRQ defined\n");
+		goto destroy_cache;
+	}
 
 	ret = d40_dmaengine_init(base, num_reserved_chans);
 	if (ret)
@@ -3817,7 +3819,8 @@ static int __init d40_probe(struct platform_device *pdev)
 		regulator_disable(base->lcpa_regulator);
 		regulator_put(base->lcpa_regulator);
 	}
-	pm_runtime_disable(base->dev);
+	if (runtime_pm_enabled)
+		pm_runtime_disable(base->dev);
 
  report_failure:
 	d40_err(dev, "probe failed\n");
Signature · 1 line
2.55.0
#11 lore raw

Re: [PATCH v5 05/23] dmaengine: ste_dma40: Fix probe runtime PM disable

Quoted 50 lines · Linus Walleij

Some d40_probe() error paths jump to destroy_cache before runtime PM has been enabled for the DMA controller device. The label unconditionally calls pm_runtime_disable(), which increments disable_depth even though this probe attempt never enabled runtime PM.

Track whether this probe attempt enabled runtime PM before disabling it on the error path. This is not about a later deferred-probe retry, since the driver is registered with platform_driver_probe(); it keeps the probe unwind balanced.

The interrupt handler uses pm_runtime_get_if_active() and cannot acknowledge a pending interrupt while runtime PM is disabled. Request the IRQ only after enabling runtime PM so the handler cannot enter an unacknowledged interrupt loop during probe.

Fixes 0618c077a8c2 ("dmaengine: ste_dma40: Fix PM disable depth imbalance in d40_probe")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 17 ++++++++++-------
 1 file changed, 10 insertions(+), 7 deletions(-)
drivers/dma/ste_dma40.c +1 -6
index 280a0d2f16c5..1d02226ab5ff 100644
@@ -3661,6 +3661,7 @@ static int __init d40_probe(struct platform_device *pdev)
 	struct resource *res;
 	struct resource res_lcpa;
 	int num_reserved_chans;
+	bool runtime_pm_enabled = false;
 	u32 val;
 	int ret;
 
@@ -3748,12 +3749,6 @@ static int __init d40_probe(struct platform_device *pdev)
 		goto destroy_cache;
 	}
 
-	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
-	if (ret) {
-		d40_err(dev, "No IRQ defined\n");
-		goto destroy_cache;
-	}
-
 	if (base->plat_data->use_esram_lcla) {
 
 		base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");

drivers/dma/ste_dma40.c static int __init d40_probe(struct platform_device *pdev) · line 3780

Quoted 4 lines
drivers/dma/ste_dma40.c +1 -0
@@ -3782,6 +3777,13 @@ static int __init d40_probe(struct platform_device *pdev)
 	pm_runtime_mark_last_busy(base->dev);
 	pm_runtime_set_active(base->dev);
 	pm_runtime_enable(base->dev);
+	runtime_pm_enabled = true;

Does devm_pm_runtime_set_active_enabled() help your case? I have not other driver need this kindle varible.

Frank

Quoted 23 lines
drivers/dma/ste_dma40.c +8 -2
+
+	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
+	if (ret) {
+		d40_err(dev, "No IRQ defined\n");
+		goto destroy_cache;
+	}
 
 	ret = d40_dmaengine_init(base, num_reserved_chans);
 	if (ret)
@@ -3817,7 +3819,8 @@ static int __init d40_probe(struct platform_device *pdev)
 		regulator_disable(base->lcpa_regulator);
 		regulator_put(base->lcpa_regulator);
 	}
-	pm_runtime_disable(base->dev);
+	if (runtime_pm_enabled)
+		pm_runtime_disable(base->dev);
 
  report_failure:
 	d40_err(dev, "probe failed\n");
 
--

2.55.0

author maintainer
#12 lore raw

PATCH v5 6/23 dmaengine: ste_dma40: Check runtime PM in IRQ

In reply to Linus Walleij

d40_handle_interrupt() reads DMA40 interrupt registers unconditionally. A spurious interrupt can arrive while the device is runtime suspended, after dma40_runtime_suspend() has disabled the GCC clock.

Avoid touching the registers unless the device is runtime active by taking a conditional runtime PM reference. Return IRQ_NONE when the device is suspended, and drop the reference after handling an active interrupt.

When CONFIG_PM is disabled, pm_runtime_get_if_active() returns -EINVAL even though the registers remain accessible. Keep handling interrupts in that configuration and only drop the runtime PM reference when one was acquired.

Fixes 7fb3e75e1833 ("dmaengine/ste_dma40: support pm in dma40")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/dma/ste_dma40.c +8 -0
index 1d02226ab5ff..58128a980847 100644
@@ -1798,6 +1798,11 @@ static irqreturn_t d40_handle_interrupt(int irq, void *data)
 	u32 *regs = base->regs_interrupt;
 	struct d40_interrupt_lookup *il = base->gen_dmac.il;
 	u32 il_size = base->gen_dmac.il_size;
+	int ret;
+
+	ret = pm_runtime_get_if_active(base->dev);
+	if (IS_ENABLED(CONFIG_PM) && ret <= 0)
+		return IRQ_NONE;
 
 	spin_lock(&base->interrupt_lock);
 
@@ -1846,6 +1851,9 @@ static irqreturn_t d40_handle_interrupt(int irq, void *data)
 
 	spin_unlock(&base->interrupt_lock);
 
+	if (ret > 0)
+		pm_runtime_put_autosuspend(base->dev);
+
 	return IRQ_HANDLED;
 }
 
Signature · 1 line
2.55.0
Reviewed-by on v5 6/23
#13 lore raw

Re: [PATCH v5 06/23] dmaengine: ste_dma40: Check runtime PM in IRQ

Quoted 12 lines · Linus Walleij

d40_handle_interrupt() reads DMA40 interrupt registers unconditionally. A spurious interrupt can arrive while the device is runtime suspended, after dma40_runtime_suspend() has disabled the GCC clock.

Avoid touching the registers unless the device is runtime active by taking a conditional runtime PM reference. Return IRQ_NONE when the device is suspended, and drop the reference after handling an active interrupt.

When CONFIG_PM is disabled, pm_runtime_get_if_active() returns -EINVAL even though the registers remain accessible. Keep handling interrupts in that configuration and only drop the runtime PM reference when one was acquired.

Quoted 6 lines
Fixes 7fb3e75e1833 ("dmaengine/ste_dma40: support pm in dma40")
Reported-by [email protected]
Assisted-by LLM

---

Quoted 33 lines
 drivers/dma/ste_dma40.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/dma/ste_dma40.c +8 -1
index 1d02226ab5ff..58128a980847 100644
@@ -1798,6 +1798,11 @@ static irqreturn_t d40_handle_interrupt(int irq, void *data)
 	u32 *regs = base->regs_interrupt;
 	struct d40_interrupt_lookup *il = base->gen_dmac.il;
 	u32 il_size = base->gen_dmac.il_size;
+	int ret;
+
+	ret = pm_runtime_get_if_active(base->dev);
+	if (IS_ENABLED(CONFIG_PM) && ret <= 0)
+		return IRQ_NONE;
 
 	spin_lock(&base->interrupt_lock);
 
@@ -1846,6 +1851,9 @@ static irqreturn_t d40_handle_interrupt(int irq, void *data)
 
 	spin_unlock(&base->interrupt_lock);
 
+	if (ret > 0)
+		pm_runtime_put_autosuspend(base->dev);
+
 	return IRQ_HANDLED;
 }
 
 
--

2.55.0

author maintainer
#14 lore raw

PATCH v5 7/23 dmaengine: ste_dma40: Handle runtime PM resume errors

In reply to Linus Walleij

Several channel operations use pm_runtime_get_sync() and access DMA40 registers without checking whether runtime resume succeeded. If resume fails, the registers may be inaccessible. pm_runtime_get_sync() also increments the usage counter on failure, making error unwinding easy to unbalance.

Use pm_runtime_resume_and_get() and avoid register access when resume fails. Acquire the runtime PM reference before allocating a channel so failure needs no channel-allocation rollback.

If a queued transfer cannot be started because resume failed, retire all issued descriptors through the normal tasklet path. Since dma_async_issue_pending() cannot return an error, leaving them queued would make clients wait indefinitely for callbacks.

Termination and channel release must also clean up software state when the controller cannot resume. Always release descriptors and the outstanding busy reference, and release channel allocation state when freeing the channel. Skip only the hardware stop that requires register access.

Fixes 7fb3e75e1833 ("dmaengine/ste_dma40: support pm in dma40")
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 67 ++++++++++++++++++++++++++++++++++++-------------
 1 file changed, 50 insertions(+), 17 deletions(-)
drivers/dma/ste_dma40.c +50 -17
index 58128a980847..bb052d3028cc 100644
@@ -1567,11 +1567,14 @@ static int d40_pause(struct dma_chan *chan)
 		return 0;
 
 	spin_lock_irqsave(&d40c->lock, flags);
-	pm_runtime_get_sync(d40c->base->dev);
+	res = pm_runtime_resume_and_get(d40c->base->dev);
+	if (res < 0)
+		goto unlock;
 
 	res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
 
 	pm_runtime_put_autosuspend(d40c->base->dev);
+ unlock:
 	spin_unlock_irqrestore(&d40c->lock, flags);
 	return res;
 }
@@ -1591,13 +1594,16 @@ static int d40_resume(struct dma_chan *chan)
 		return 0;
 
 	spin_lock_irqsave(&d40c->lock, flags);
-	pm_runtime_get_sync(d40c->base->dev);
+	res = pm_runtime_resume_and_get(d40c->base->dev);
+	if (res < 0)
+		goto unlock;
 
 	/* If bytes left to transfer or linked tx resume job */
 	if (d40_residue(d40c) || d40_tx_is_linked(d40c))
 		res = d40_channel_execute_command(d40c, D40_DMA_RUN);
 
 	pm_runtime_put_autosuspend(d40c->base->dev);
+ unlock:
 	spin_unlock_irqrestore(&d40c->lock, flags);
 	return res;
 }
@@ -1634,8 +1640,20 @@ static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
 
 	if (d40d != NULL) {
 		if (!d40c->busy) {
+			err = pm_runtime_resume_and_get(d40c->base->dev);
+			if (err < 0) {
+				chan_err(d40c, "Failed to resume DMA: %d\n",
+					 err);
+				do {
+					d40_desc_remove(d40d);
+					d40_desc_done(d40c, d40d);
+					d40c->pending_tx++;
+					d40d = d40_first_queued(d40c);
+				} while (d40d);
+				tasklet_schedule(&d40c->tasklet);
+				return ERR_PTR(err);
+			}
 			d40c->busy = true;
-			pm_runtime_get_sync(d40c->base->dev);
 		}
 
 		/* Remove from queue */
@@ -2150,11 +2168,9 @@ static int d40_free_dma(struct d40_chan *d40c)
 	int res = 0;
 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
 	struct d40_phy_res *phy = d40c->phy_chan;
+	bool pm_acquired = false;
 	bool is_src;
 
-	/* Terminate all queued and active transfers */
-	d40_term_all(d40c);
-
 	if (phy == NULL) {
 		chan_err(d40c, "phy == null\n");
 		return -EINVAL;
@@ -2176,13 +2192,21 @@ static int d40_free_dma(struct d40_chan *d40c)
 		return -EINVAL;
 	}
 
-	pm_runtime_get_sync(d40c->base->dev);
+	/* Terminate all queued and active transfers */
+	d40_term_all(d40c);
+
+	res = pm_runtime_resume_and_get(d40c->base->dev);
+	if (res < 0)
+		goto release_channel;
+	pm_acquired = true;
+
 	res = d40_channel_execute_command(d40c, D40_DMA_STOP);
 	if (res) {
 		chan_err(d40c, "stop failed\n");
 		goto mark_last_busy;
 	}
 
+ release_channel:
 	d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
 
 	if (chan_is_logical(d40c))
@@ -2197,7 +2221,8 @@ static int d40_free_dma(struct d40_chan *d40c)
 	d40c->phy_chan = NULL;
 	d40c->configured = false;
  mark_last_busy:
-	pm_runtime_put_autosuspend(d40c->base->dev);
+	if (pm_acquired)
+		pm_runtime_put_autosuspend(d40c->base->dev);
 	return res;
 }
 
@@ -2572,10 +2597,14 @@ static int d40_alloc_chan_resources(struct dma_chan *chan)
 		err = d40_config_memcpy(d40c);
 		if (err) {
 			chan_err(d40c, "Failed to configure memcpy channel\n");
-			goto mark_last_busy;
+			goto unlock;
 		}
 	}
 
+	err = pm_runtime_resume_and_get(d40c->base->dev);
+	if (err < 0)
+		goto unlock;
+
 	err = d40_allocate_channel(d40c, &is_free_phy);
 	if (err) {
 		chan_err(d40c, "Failed to allocate channel\n");
@@ -2583,8 +2612,6 @@ static int d40_alloc_chan_resources(struct dma_chan *chan)
 		goto mark_last_busy;
 	}
 
-	pm_runtime_get_sync(d40c->base->dev);
-
 	d40_set_prio_realtime(d40c);
 
 	if (chan_is_logical(d40c)) {
@@ -2616,6 +2643,7 @@ static int d40_alloc_chan_resources(struct dma_chan *chan)
 		d40_config_write(d40c);
  mark_last_busy:
 	pm_runtime_put_autosuspend(d40c->base->dev);
+ unlock:
 	spin_unlock_irqrestore(&d40c->lock, flags);
 	return err;
 }
@@ -2767,6 +2795,7 @@ static int d40_terminate_all(struct dma_chan *chan)
 {
 	unsigned long flags;
 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
+	bool pm_acquired = false;
 	int ret;
 
 	if (d40c->phy_chan == NULL) {
@@ -2776,19 +2805,23 @@ static int d40_terminate_all(struct dma_chan *chan)
 
 	spin_lock_irqsave(&d40c->lock, flags);
 
-	pm_runtime_get_sync(d40c->base->dev);
-	ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
-	if (ret)
-		chan_err(d40c, "Failed to stop channel\n");
+	ret = pm_runtime_resume_and_get(d40c->base->dev);
+	if (ret >= 0) {
+		pm_acquired = true;
+		ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
+		if (ret)
+			chan_err(d40c, "Failed to stop channel\n");
+	}
 
 	d40_term_all(d40c);
-	pm_runtime_put_autosuspend(d40c->base->dev);
+	if (pm_acquired)
+		pm_runtime_put_autosuspend(d40c->base->dev);
 	if (d40c->busy)
 		pm_runtime_put_autosuspend(d40c->base->dev);
 	d40c->busy = false;
 
 	spin_unlock_irqrestore(&d40c->lock, flags);
-	return 0;
+	return ret;
 }
 
 static int
Signature · 1 line
2.55.0
#15 lore raw

Re: [PATCH v5 07/23] dmaengine: ste_dma40: Handle runtime PM resume errors

Quoted 191 lines · Linus Walleij

Several channel operations use pm_runtime_get_sync() and access DMA40 registers without checking whether runtime resume succeeded. If resume fails, the registers may be inaccessible. pm_runtime_get_sync() also increments the usage counter on failure, making error unwinding easy to unbalance.

Use pm_runtime_resume_and_get() and avoid register access when resume fails. Acquire the runtime PM reference before allocating a channel so failure needs no channel-allocation rollback.

If a queued transfer cannot be started because resume failed, retire all issued descriptors through the normal tasklet path. Since dma_async_issue_pending() cannot return an error, leaving them queued would make clients wait indefinitely for callbacks.

Termination and channel release must also clean up software state when the controller cannot resume. Always release descriptors and the outstanding busy reference, and release channel allocation state when freeing the channel. Skip only the hardware stop that requires register access.

Fixes 7fb3e75e1833 ("dmaengine/ste_dma40: support pm in dma40")
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 67 ++++++++++++++++++++++++++++++++++++-------------
 1 file changed, 50 insertions(+), 17 deletions(-)
drivers/dma/ste_dma40.c +47 -15
index 58128a980847..bb052d3028cc 100644
@@ -1567,11 +1567,14 @@ static int d40_pause(struct dma_chan *chan)
 		return 0;
 
 	spin_lock_irqsave(&d40c->lock, flags);
-	pm_runtime_get_sync(d40c->base->dev);
+	res = pm_runtime_resume_and_get(d40c->base->dev);
+	if (res < 0)
+		goto unlock;
 
 	res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
 
 	pm_runtime_put_autosuspend(d40c->base->dev);
+ unlock:
 	spin_unlock_irqrestore(&d40c->lock, flags);
 	return res;
 }
@@ -1591,13 +1594,16 @@ static int d40_resume(struct dma_chan *chan)
 		return 0;
 
 	spin_lock_irqsave(&d40c->lock, flags);
-	pm_runtime_get_sync(d40c->base->dev);
+	res = pm_runtime_resume_and_get(d40c->base->dev);
+	if (res < 0)
+		goto unlock;
 
 	/* If bytes left to transfer or linked tx resume job */
 	if (d40_residue(d40c) || d40_tx_is_linked(d40c))
 		res = d40_channel_execute_command(d40c, D40_DMA_RUN);
 
 	pm_runtime_put_autosuspend(d40c->base->dev);
+ unlock:
 	spin_unlock_irqrestore(&d40c->lock, flags);
 	return res;
 }
@@ -1634,8 +1640,20 @@ static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
 
 	if (d40d != NULL) {
 		if (!d40c->busy) {
+			err = pm_runtime_resume_and_get(d40c->base->dev);
+			if (err < 0) {
+				chan_err(d40c, "Failed to resume DMA: %d\n",
+					 err);
+				do {
+					d40_desc_remove(d40d);
+					d40_desc_done(d40c, d40d);
+					d40c->pending_tx++;
+					d40d = d40_first_queued(d40c);
+				} while (d40d);
+				tasklet_schedule(&d40c->tasklet);
+				return ERR_PTR(err);
+			}
 			d40c->busy = true;
-			pm_runtime_get_sync(d40c->base->dev);
 		}
 
 		/* Remove from queue */
@@ -2150,11 +2168,9 @@ static int d40_free_dma(struct d40_chan *d40c)
 	int res = 0;
 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
 	struct d40_phy_res *phy = d40c->phy_chan;
+	bool pm_acquired = false;
 	bool is_src;
 
-	/* Terminate all queued and active transfers */
-	d40_term_all(d40c);
-
 	if (phy == NULL) {
 		chan_err(d40c, "phy == null\n");
 		return -EINVAL;
@@ -2176,13 +2192,21 @@ static int d40_free_dma(struct d40_chan *d40c)
 		return -EINVAL;
 	}
 
-	pm_runtime_get_sync(d40c->base->dev);
+	/* Terminate all queued and active transfers */
+	d40_term_all(d40c);
+
+	res = pm_runtime_resume_and_get(d40c->base->dev);
+	if (res < 0)
+		goto release_channel;
+	pm_acquired = true;
+
 	res = d40_channel_execute_command(d40c, D40_DMA_STOP);
 	if (res) {
 		chan_err(d40c, "stop failed\n");
 		goto mark_last_busy;
 	}
 
+ release_channel:
 	d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
 
 	if (chan_is_logical(d40c))
@@ -2197,7 +2221,8 @@ static int d40_free_dma(struct d40_chan *d40c)
 	d40c->phy_chan = NULL;
 	d40c->configured = false;
  mark_last_busy:
-	pm_runtime_put_autosuspend(d40c->base->dev);
+	if (pm_acquired)
+		pm_runtime_put_autosuspend(d40c->base->dev);
 	return res;
 }
 
@@ -2572,10 +2597,14 @@ static int d40_alloc_chan_resources(struct dma_chan *chan)
 		err = d40_config_memcpy(d40c);
 		if (err) {
 			chan_err(d40c, "Failed to configure memcpy channel\n");
-			goto mark_last_busy;
+			goto unlock;
 		}
 	}
 
+	err = pm_runtime_resume_and_get(d40c->base->dev);
+	if (err < 0)
+		goto unlock;
+
 	err = d40_allocate_channel(d40c, &is_free_phy);
 	if (err) {
 		chan_err(d40c, "Failed to allocate channel\n");
@@ -2583,8 +2612,6 @@ static int d40_alloc_chan_resources(struct dma_chan *chan)
 		goto mark_last_busy;
 	}
 
-	pm_runtime_get_sync(d40c->base->dev);
-
 	d40_set_prio_realtime(d40c);
 
 	if (chan_is_logical(d40c)) {
@@ -2616,6 +2643,7 @@ static int d40_alloc_chan_resources(struct dma_chan *chan)
 		d40_config_write(d40c);
  mark_last_busy:
 	pm_runtime_put_autosuspend(d40c->base->dev);
+ unlock:
 	spin_unlock_irqrestore(&d40c->lock, flags);
 	return err;
 }
@@ -2767,6 +2795,7 @@ static int d40_terminate_all(struct dma_chan *chan)
 {
 	unsigned long flags;
 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
+	bool pm_acquired = false;
 	int ret;
 
 	if (d40c->phy_chan == NULL) {
@@ -2776,19 +2805,23 @@ static int d40_terminate_all(struct dma_chan *chan)
 
 	spin_lock_irqsave(&d40c->lock, flags);
 
-	pm_runtime_get_sync(d40c->base->dev);
-	ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
-	if (ret)
-		chan_err(d40c, "Failed to stop channel\n");
+	ret = pm_runtime_resume_and_get(d40c->base->dev);
+	if (ret >= 0) {
+		pm_acquired = true;
+		ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
+		if (ret)
+			chan_err(d40c, "Failed to stop channel\n");
+	}
 
 	d40_term_all(d40c);

drivers/dma/ste_dma40.c static int d40_terminate_all(struct dma_chan *chan) · line 2824

Quoted 10 lines
drivers/dma/ste_dma40.c +3 -2
-	pm_runtime_put_autosuspend(d40c->base->dev);
+	if (pm_acquired)
+		pm_runtime_put_autosuspend(d40c->base->dev);
 	if (d40c->busy)
 		pm_runtime_put_autosuspend(d40c->base->dev);
 	d40c->busy = false;
 
 	spin_unlock_irqrestore(&d40c->lock, flags);
-	return 0;
+	return ret;

pm_acquired is not necesary

	ret = pm_runtime_resume_and_get(d40c->base->dev);
	if (ret >= ) {
		ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
		if (ret)
			...
		pm_runtime_put_autosuspend(d40c->base->dev);
	}

	d40_term_all(d40c); /* In your patch, d40_term_all(d40c) can be call
without acquire runtime pm
	...

Frank

Quoted 7 lines
drivers/dma/ste_dma40.c +0 -1
 }
 
 static int
 
--

2.55.0

author maintainer
#16 lore raw

PATCH v5 8/23 dmaengine: ste_dma40: Return IRQ_NONE without interrupt status

In reply to Linus Walleij

d40_handle_interrupt() returns IRQ_HANDLED even when no terminal-count or error status bit is set. If the line remains asserted without matching status, reporting it as handled prevents the generic interrupt code from detecting the stuck interrupt.

Track whether the handler observes any DMA40 status. Return IRQ_NONE only when none is present, allowing the generic spurious interrupt detector to disable a faulty status-less interrupt line.

DMA40 channels can be owned by other SoC cores. Their status still explains why the interrupt fired, but Linux must not acknowledge it. Treat foreign status as handled and leave its acknowledgment to the owning core, while acknowledging and dispatching only registered Linux channels.

Fixes 8d318a50b3d7 ("DMAENGINE: Support for ST-Ericssons DMA40 block v3")
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
drivers/dma/ste_dma40.c +8 -1
index bb052d3028cc..4380f9a1b035 100644
@@ -1807,6 +1807,7 @@ static void dma_tasklet(struct tasklet_struct *t)
 
 static irqreturn_t d40_handle_interrupt(int irq, void *data)
 {
+	irqreturn_t handled = IRQ_NONE;
 	int i;
 	u32 idx;
 	u32 row;
@@ -1840,6 +1841,12 @@ static irqreturn_t d40_handle_interrupt(int irq, void *data)
 		row = chan / BITS_PER_LONG;
 		idx = chan & (BITS_PER_LONG - 1);
 
+		/*
+		 * Status for a channel owned by another core still explains
+		 * the interrupt, but only ACK Linux-owned channels below.
+		 */
+		handled = IRQ_HANDLED;
+
 		if (il[row].offset == D40_PHY_CHAN)
 			d40c = base->lookup_phy_chans[idx];
 		else
@@ -1872,7 +1879,7 @@ static irqreturn_t d40_handle_interrupt(int irq, void *data)
 	if (ret > 0)
 		pm_runtime_put_autosuspend(base->dev);
 
-	return IRQ_HANDLED;
+	return handled;
 }
 
 static int d40_validate_conf(struct d40_chan *d40c,
Signature · 1 line
2.55.0
#17 lore raw

Re: [PATCH v5 08/23] dmaengine: ste_dma40: Return IRQ_NONE without interrupt status

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] dmaengine: ste_dma40: IRQ_NONE on AMP-shared line disables IRQ --

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] dmaengine: ste_dma40: IRQ_NONE on AMP-shared line disables IRQ
--

commit 2ed162fa364e4bf7289280c1ba5e44ef8ed65fc3
Author: Linus Walleij <[email protected]>
Subject: dmaengine: ste_dma40: Return IRQ_NONE without interrupt status

This commit changes the DMA40 interrupt handler to return IRQ_NONE when no status bits are set. It aims to allow the generic spurious interrupt detector to disable a faulty status-less interrupt line while still treating foreign status as handled.

Quoted 9 lines
drivers/dma/ste_dma40.c +1 -0
index bb052d3028cc2..4380f9a1b035c 100644
@@ -1807,6 +1807,7 @@ static void dma_tasklet(struct tasklet_struct *t)
 
 static irqreturn_t d40_handle_interrupt(int irq, void *data)
 {
+	irqreturn_t handled = IRQ_NONE;

[ ... ]

drivers/dma/ste_dma40.c static irqreturn_t d40_handle_interrupt(int irq, void *data) · line 1882

Quoted 6 lines
drivers/dma/ste_dma40.c +1 -1
@@ -1872,7 +1879,7 @@ static irqreturn_t d40_handle_interrupt(int irq, void *data)
 	if (ret > 0)
 		pm_runtime_put_autosuspend(base->dev);
 
-	return IRQ_HANDLED;
+	return handled;

[Severity: High]
Does returning IRQ_NONE here risk disabling the DMA40 interrupt line in an Asymmetric Multi-Processing (AMP) setup?

In an AMP setup where both Linux and another SoC core share the DMA40 interrupt line and status registers, both cores receive the IRQ when an interrupt fires for the other core's channel. If the faster core (e.g., an RTOS) clears the status register before the slower Linux handler reads it, Linux will read 0 bits set, and handled will remain IRQ_NONE.

If the other core processes many interrupts, such as audio streams, Linux could quickly accumulate 99,900 'ghost' interrupts within a 100,000 interrupt window. This would cause the generic Linux spurious interrupt detector to permanently disable the shared IRQ line, breaking all Linux-owned DMA channels. Returning IRQ_NONE for status-less interrupts appears to be incompatible with AMP-shared interrupt lines.

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=8
author maintainer
#18 lore raw

Re: [PATCH v5 08/23] dmaengine: ste_dma40: Return IRQ_NONE without interrupt status

Hello Sashiko,

Quoted 3 lines · On Sun, Sep 20, 2026 at 9:10 PM <[email protected]> wrote:

[Severity: High]
Does returning IRQ_NONE here risk disabling the DMA40 interrupt line in an Asymmetric Multi-Processing (AMP) setup?

You didn't really read the cover letter did you?

This is FRINGE, the other CPU cores on the system (ux500) where this is used do *NOT* use the DMA40 engine.

Nevertheless I have done my best to circumvent this.

I think the series is good to go.

Yours,
Linus Walleij

#19 lore raw

Re: [PATCH v5 08/23] dmaengine: ste_dma40: Return IRQ_NONE without interrupt status

In reply to Linus Walleij

On Sun, Sep 20, 2026 at 08:59:43PM +0200, Linus Walleij wrote:

suggested subject:

return IRQ_NONE when no interrupt is pending

Quoted 4 lines

d40_handle_interrupt() returns IRQ_HANDLED even when no terminal-count or error status bit is set. If the line remains asserted without matching status, reporting it as handled prevents the generic interrupt code from detecting the stuck interrupt.

Looks like it is debug function. The major purpose is return IRQ_NONE should be for irq sharing?

Frank

Quoted 55 lines

Track whether the handler observes any DMA40 status. Return IRQ_NONE only when none is present, allowing the generic spurious interrupt detector to disable a faulty status-less interrupt line.

DMA40 channels can be owned by other SoC cores. Their status still explains why the interrupt fired, but Linux must not acknowledge it. Treat foreign status as handled and leave its acknowledgment to the owning core, while acknowledging and dispatching only registered Linux channels.

Fixes 8d318a50b3d7 ("DMAENGINE: Support for ST-Ericssons DMA40 block v3")
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
drivers/dma/ste_dma40.c +8 -2
index bb052d3028cc..4380f9a1b035 100644
@@ -1807,6 +1807,7 @@ static void dma_tasklet(struct tasklet_struct *t)
 
 static irqreturn_t d40_handle_interrupt(int irq, void *data)
 {
+	irqreturn_t handled = IRQ_NONE;
 	int i;
 	u32 idx;
 	u32 row;
@@ -1840,6 +1841,12 @@ static irqreturn_t d40_handle_interrupt(int irq, void *data)
 		row = chan / BITS_PER_LONG;
 		idx = chan & (BITS_PER_LONG - 1);
 
+		/*
+		 * Status for a channel owned by another core still explains
+		 * the interrupt, but only ACK Linux-owned channels below.
+		 */
+		handled = IRQ_HANDLED;
+
 		if (il[row].offset == D40_PHY_CHAN)
 			d40c = base->lookup_phy_chans[idx];
 		else
@@ -1872,7 +1879,7 @@ static irqreturn_t d40_handle_interrupt(int irq, void *data)
 	if (ret > 0)
 		pm_runtime_put_autosuspend(base->dev);
 
-	return IRQ_HANDLED;
+	return handled;
 }
 
 static int d40_validate_conf(struct d40_chan *d40c,
 
--

2.55.0

author maintainer
#20 lore raw

Re: [PATCH v5 08/23] dmaengine: ste_dma40: Return IRQ_NONE without interrupt status

Hi Frank,

Quoted 6 lines · Frank Li

On Sun, Sep 20, 2026 at 08:59:43PM +0200, Linus Walleij wrote:

suggested subject:

return IRQ_NONE when no interrupt is pending

Quoted 7 lines
Quoted 4 lines

d40_handle_interrupt() returns IRQ_HANDLED even when no terminal-count or error status bit is set. If the line remains asserted without matching status, reporting it as handled prevents the generic interrupt code from detecting the stuck interrupt.

Looks like it is debug function. The major purpose is return IRQ_NONE should be for irq sharing?

That's the most usual case, but it just generally means "that IRQ was not for me / caused by my hardware" so it is something you can/should emit even for non-shared IRQs in case you realized you are certain the hardware didn't raise it.

Then it will appear in dmesg as "spurious IRQ" I think.

Yours,
Linus Walleij

author maintainer
#21 lore raw

PATCH v5 9/23 dmaengine: ste_dma40: Init hardware before registration

In reply to Linus Walleij

d40_probe() enables the interrupt handler and registers DMAengine devices before calling d40_hw_init(). An interrupt pending from the bootloader can therefore reach the handler while the hardware interrupt state is not initialized and channel lookup entries are empty. The handler deliberately does not acknowledge an interrupt for an unknown channel, so a level IRQ can retrigger continuously.

Request the IRQ with IRQF_NO_AUTOEN, then initialize the hardware and set the DMA segment limit. Enable the IRQ before registering DMAengine devices. This ensures the handler and clients only observe initialized hardware while still letting request_irq() fail before hardware interrupts are enabled.

Fixes 8d318a50b3d7 ("DMAENGINE: Support for ST-Ericssons DMA40 block v3")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)
drivers/dma/ste_dma40.c +7 -5
index 4380f9a1b035..7cbcf8433fe0 100644
@@ -3827,19 +3827,21 @@ static int __init d40_probe(struct platform_device *pdev)
 	pm_runtime_enable(base->dev);
 	runtime_pm_enabled = true;
 
-	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
+	ret = request_irq(base->irq, d40_handle_interrupt, IRQF_NO_AUTOEN,
+			  D40_NAME, base);
 	if (ret) {
 		d40_err(dev, "No IRQ defined\n");
 		goto destroy_cache;
 	}
 
-	ret = d40_dmaengine_init(base, num_reserved_chans);
-	if (ret)
-		goto destroy_cache;
-
 	dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
 
 	d40_hw_init(base);
+	enable_irq(base->irq);
+
+	ret = d40_dmaengine_init(base, num_reserved_chans);
+	if (ret)
+		goto destroy_cache;
 
 	ret = of_dma_controller_register(np, d40_xlate, NULL);
 	if (ret) {
Signature · 1 line
2.55.0
author maintainer
#22 lore raw

PATCH v5 10/23 dmaengine: ste_dma40: Fix probe IRQ leak

In reply to Linus Walleij

d40_probe() registers the hardware interrupt before several later probe steps that can fail. Those error paths jump to destroy_cache without freeing the IRQ, leaving the handler registered after probe resources have been released.

Track successful IRQ registration and free the IRQ on later probe failure.

Fixes 8d318a50b3d7 ("DMAENGINE: Support for ST-Ericssons DMA40 block v3")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/dma/ste_dma40.c +4 -0
index 7cbcf8433fe0..ba529bd91752 100644
@@ -3710,6 +3710,7 @@ static int __init d40_probe(struct platform_device *pdev)
 	struct resource res_lcpa;
 	int num_reserved_chans;
 	bool runtime_pm_enabled = false;
+	bool irq_requested = false;
 	u32 val;
 	int ret;
 
@@ -3833,6 +3834,7 @@ static int __init d40_probe(struct platform_device *pdev)
 		d40_err(dev, "No IRQ defined\n");
 		goto destroy_cache;
 	}
+	irq_requested = true;
 
 	dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
 
@@ -3869,6 +3871,8 @@ static int __init d40_probe(struct platform_device *pdev)
 		regulator_disable(base->lcpa_regulator);
 		regulator_put(base->lcpa_regulator);
 	}
+	if (irq_requested)
+		free_irq(base->irq, base);
 	if (runtime_pm_enabled)
 		pm_runtime_disable(base->dev);
 
Signature · 1 line
2.55.0
author maintainer
#23 lore raw

PATCH v5 11/23 dmaengine: ste_dma40: Fix DMA registration unwind

In reply to Linus Walleij

d40_dmaengine_init() registers DMAengine devices using devres-managed unregister actions. If probe fails after one of those registrations, the DMAengine devices stay visible until devres unwinds after d40_probe() returns.

The channel tasklets are also initialized before each DMAengine device is registered. An interrupt can therefore have queued a tasklet by the time a later registration step fails, but unregistering the DMAengine devices does not drain that tasklet before probe-owned storage is released.

Add tasklet cleanup actions to the DMAengine registration devres group. On failure, free the IRQ before releasing the group so no new tasklets can be scheduled, then unregister the DMAengine devices and drain their tasklets before freeing the remaining probe resources. Keep the managed registrations and cleanup actions in place on successful probe by removing only the temporary group markers.

Fixes 42ae6f1695be ("dmaengine: ste_dma40: Remove platform data")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 46 ++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 44 insertions(+), 2 deletions(-)
drivers/dma/ste_dma40.c +44 -2
index ba529bd91752..52d39883da51 100644
@@ -3033,6 +3033,18 @@ static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
 	}
 }
 
+static void d40_kill_tasklets(void *data)
+{
+	struct dma_device *dma = data;
+	struct d40_chan *d40c;
+	struct dma_chan *chan;
+
+	list_for_each_entry(chan, &dma->channels, device_node) {
+		d40c = container_of(chan, struct d40_chan, chan);
+		tasklet_kill(&d40c->tasklet);
+	}
+}
+
 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
 {
 	if (dma_has_cap(DMA_SLAVE, dev->cap_mask)) {
@@ -3079,6 +3091,11 @@ static int __init d40_dmaengine_init(struct d40_base *base,
 
 	d40_ops_init(base, &base->dma_slave);
 
+	err = devm_add_action_or_reset(base->dev, d40_kill_tasklets,
+				       &base->dma_slave);
+	if (err)
+		goto exit;
+
 	err = dmaenginem_async_device_register(&base->dma_slave);
 
 	if (err) {
@@ -3094,6 +3111,11 @@ static int __init d40_dmaengine_init(struct d40_base *base,
 
 	d40_ops_init(base, &base->dma_memcpy);
 
+	err = devm_add_action_or_reset(base->dev, d40_kill_tasklets,
+				       &base->dma_memcpy);
+	if (err)
+		goto exit;
+
 	err = dmaenginem_async_device_register(&base->dma_memcpy);
 
 	if (err) {
@@ -3111,6 +3133,12 @@ static int __init d40_dmaengine_init(struct d40_base *base,
 	dma_cap_set(DMA_CYCLIC, base->dma_both.cap_mask);
 
 	d40_ops_init(base, &base->dma_both);
+
+	err = devm_add_action_or_reset(base->dev, d40_kill_tasklets,
+				       &base->dma_both);
+	if (err)
+		goto exit;
+
 	err = dmaenginem_async_device_register(&base->dma_both);
 
 	if (err) {
@@ -3708,6 +3736,7 @@ static int __init d40_probe(struct platform_device *pdev)
 	struct d40_base *base;
 	struct resource *res;
 	struct resource res_lcpa;
+	void *dmaenginem_reg_group;
 	int num_reserved_chans;
 	bool runtime_pm_enabled = false;
 	bool irq_requested = false;
@@ -3841,20 +3870,33 @@ static int __init d40_probe(struct platform_device *pdev)
 	d40_hw_init(base);
 	enable_irq(base->irq);
 
+	dmaenginem_reg_group = devres_open_group(dev, NULL, GFP_KERNEL);
+	if (!dmaenginem_reg_group) {
+		ret = -ENOMEM;
+		goto destroy_cache;
+	}
+
 	ret = d40_dmaengine_init(base, num_reserved_chans);
 	if (ret)
-		goto destroy_cache;
+		goto release_dmaenginem;
 
 	ret = of_dma_controller_register(np, d40_xlate, NULL);
 	if (ret) {
 		dev_err(dev,
 			"could not register of_dma_controller\n");
-		goto destroy_cache;
+		goto release_dmaenginem;
 	}
+	devres_remove_group(dev, dmaenginem_reg_group);
 
 	dev_info(base->dev, "initialized\n");
 	return 0;
 
+ release_dmaenginem:
+	if (irq_requested) {
+		free_irq(base->irq, base);
+		irq_requested = false;
+	}
+	devres_release_group(dev, dmaenginem_reg_group);
  destroy_cache:
 	if (base->lcla_pool.dma_addr)
 		dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
Signature · 1 line
2.55.0
author maintainer
#24 lore raw

PATCH v5 12/23 dmaengine: ste_dma40: Fix LCLA allocation order

In reply to Linus Walleij

d40_lcla_allocate() calculates the number of pages needed for LCLA, but passes that raw page count as the allocation order to __get_free_pages(). The same value is later passed to free_pages().

Store the allocation order with get_order() instead, and use a separate byte size for allocation diagnostics, fallback kmalloc() sizing and DMA mapping.

Fixes 508849ade23c ("DMAENGINE: ste_dma40: allocate LCLA dynamically")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 33 ++++++++++++++++-----------------
 1 file changed, 16 insertions(+), 17 deletions(-)
drivers/dma/ste_dma40.c +16 -17
index 52d39883da51..33cd93e90bc4 100644
@@ -423,8 +423,8 @@ struct d40_desc {
  * @dma_addr: DMA address, if mapped
  * @base_unaligned: The original kmalloc pointer, if kmalloc is used.
  * This pointer is only there for clean-up on error.
- * @pages: The number of pages needed for all physical channels.
- * Only used later for clean-up on error
+ * @alloc_order: Order used for the LCLA page allocation.
+ * Only used later for clean-up on error.
  * @lock: Lock to protect the content in this struct.
  * @alloc_map: big map over which LCLA entry is own by which job.
  */
@@ -432,7 +432,7 @@ struct d40_lcla_pool {
 	void		*base;
 	dma_addr_t	dma_addr;
 	void		*base_unaligned;
-	int		 pages;
+	unsigned int	 alloc_order;
 	spinlock_t	 lock;
 	struct d40_desc	**alloc_map;
 };
@@ -3597,6 +3597,7 @@ static void __init d40_hw_init(struct d40_base *base)
 static int __init d40_lcla_allocate(struct d40_base *base)
 {
 	struct d40_lcla_pool *pool = &base->lcla_pool;
+	size_t lcla_size = SZ_1K * base->num_phy_chans;
 	unsigned long *page_list;
 	int i, j;
 	int ret;
@@ -3612,20 +3613,20 @@ static int __init d40_lcla_allocate(struct d40_base *base)
 	if (!page_list)
 		return -ENOMEM;
 
-	/* Calculating how many pages that are required */
-	base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
+	base->lcla_pool.alloc_order = get_order(lcla_size);
 
 	for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
 		page_list[i] = __get_free_pages(GFP_KERNEL,
-						base->lcla_pool.pages);
+						base->lcla_pool.alloc_order);
 		if (!page_list[i]) {
 
-			d40_err(base->dev, "Failed to allocate %d pages.\n",
-				base->lcla_pool.pages);
+			d40_err(base->dev, "Failed to allocate %zu bytes.\n",
+				lcla_size);
 			ret = -ENOMEM;
 
 			for (j = 0; j < i; j++)
-				free_pages(page_list[j], base->lcla_pool.pages);
+				free_pages(page_list[j],
+					   base->lcla_pool.alloc_order);
 			goto free_page_list;
 		}
 
@@ -3635,7 +3636,7 @@ static int __init d40_lcla_allocate(struct d40_base *base)
 	}
 
 	for (j = 0; j < i; j++)
-		free_pages(page_list[j], base->lcla_pool.pages);
+		free_pages(page_list[j], base->lcla_pool.alloc_order);
 
 	if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
 		base->lcla_pool.base = (void *)page_list[i];
@@ -3645,10 +3646,9 @@ static int __init d40_lcla_allocate(struct d40_base *base)
 		 * alignment, try with allocating a big buffer.
 		 */
 		dev_warn(base->dev,
-			 "[%s] Failed to get %d pages @ 18 bit align.\n",
-			 __func__, base->lcla_pool.pages);
-		base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
-							 base->num_phy_chans +
+			 "[%s] Failed to get %zu bytes @ 18 bit align.\n",
+			 __func__, lcla_size);
+		base->lcla_pool.base_unaligned = kmalloc(lcla_size +
 							 LCLA_ALIGNMENT,
 							 GFP_KERNEL);
 		if (!base->lcla_pool.base_unaligned) {
@@ -3660,8 +3660,7 @@ static int __init d40_lcla_allocate(struct d40_base *base)
 						 LCLA_ALIGNMENT);
 	}
 
-	pool->dma_addr = dma_map_single(base->dev, pool->base,
-					SZ_1K * base->num_phy_chans,
+	pool->dma_addr = dma_map_single(base->dev, pool->base, lcla_size,
 					DMA_TO_DEVICE);
 	if (dma_mapping_error(base->dev, pool->dma_addr)) {
 		pool->dma_addr = 0;
@@ -3905,7 +3904,7 @@ static int __init d40_probe(struct platform_device *pdev)
 
 	if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
 		free_pages((unsigned long)base->lcla_pool.base,
-			   base->lcla_pool.pages);
+			   base->lcla_pool.alloc_order);
 
 	kfree(base->lcla_pool.base_unaligned);
 
Signature · 1 line
2.55.0
author maintainer
#25 lore raw

PATCH v5 13/23 dmaengine: ste_dma40: Fix probe LCLA free

In reply to Linus Walleij

When LCLA is placed in ESRAM, d40_probe() stores a devm_ioremap() address in lcla_pool.base and leaves base_unaligned unset. The destroy_cache error path can then pass the ioremap address to free_pages().

Only free lcla_pool.base with free_pages() when the driver allocated the LCLA pool from normal memory. The ESRAM mapping is devm-managed.

Fixes 339f5041089a ("dmaengine: ste_dma40: Use managed resources")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/dma/ste_dma40.c +2 -1
index 33cd93e90bc4..77ae6b28610b 100644
@@ -3902,7 +3902,8 @@ static int __init d40_probe(struct platform_device *pdev)
 				 SZ_1K * base->num_phy_chans,
 				 DMA_TO_DEVICE);
 
-	if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
+	if (!base->plat_data->use_esram_lcla &&
+	    !base->lcla_pool.base_unaligned && base->lcla_pool.base)
 		free_pages((unsigned long)base->lcla_pool.base,
 			   base->lcla_pool.alloc_order);
 
Signature · 1 line
2.55.0
author maintainer
#26 lore raw

PATCH v5 14/23 dmaengine: ste_dma40: Put the LCPA SRAM node

In reply to Linus Walleij

of_parse_phandle() takes a reference to the LCPA SRAM node, but d40_probe() does not release it after translating the node into a resource. This leaks the node reference for the lifetime of the system.

Put the node immediately after of_address_to_resource() so both the success and error paths release the reference.

Fixes 5a1a3b9c19dd ("dmaengine: ste_dma40: Get LCPA SRAM from SRAM node")
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 1 +
 1 file changed, 1 insertion(+)
drivers/dma/ste_dma40.c +1 -0
index 77ae6b28610b..02ccad255a75 100644
@@ -3767,6 +3767,7 @@ static int __init d40_probe(struct platform_device *pdev)
 	}
 	/* This is no device so read the address directly from the node */
 	ret = of_address_to_resource(np_lcpa, 0, &res_lcpa);
+	of_node_put(np_lcpa);
 	if (ret) {
 		dev_err(dev, "no LCPA SRAM resource\n");
 		goto report_failure;
Signature · 1 line
2.55.0
author maintainer
#27 lore raw

PATCH v5 15/23 dmaengine: ste_dma40: Fix memcpy channel parsing

In reply to Linus Walleij

d40_of_probe() validates the memcpy-channels property against D40_MEMCPY_MAX_CHANS, but reads the property directly into a smaller global array. A long property can therefore overwrite adjacent data. The mutable global also lets a later DMA40 instance replace the memcpy channel mapping used for future allocations on an earlier instance.

Store the mapping in the per-device platform data, validate the property against that storage, and check the property read result. The property is required and d40_probe() always populates the platform data, so remove the obsolete global mapping and fallback.

Fixes a7dacb68b35a ("dmaengine: ste_dma40: Allow memcpy channels to be configured from DT")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 45 ++++++++++++++-------------------------------
 1 file changed, 14 insertions(+), 31 deletions(-)
drivers/dma/ste_dma40.c +14 -31
index 02ccad255a75..48dccd432ff4 100644
@@ -28,6 +28,8 @@
 #include "ste_dma40.h"
 #include "ste_dma40_ll.h"
 
+#define D40_MEMCPY_MAX_CHANS	8
+
 /**
  * struct stedma40_platform_data - Configuration struct for the dma device.
  *
@@ -41,6 +43,7 @@
  * to use SoftLLI.
  * @use_esram_lcla: flag for mapping the lcla into esram region
  * @num_of_memcpy_chans: The number of channels reserved for memcpy.
+ * @memcpy_channels: The event lines used for memcpy.
  * @num_of_phy_chans: The number of physical channels implemented in HW.
  * 0 means reading the number of channels from DMA HW but this is only valid
  * for 'multiple of 4' channels, like 8.
@@ -51,6 +54,7 @@ struct stedma40_platform_data {
 	int				 num_of_soft_lli_chans;
 	bool				 use_esram_lcla;
 	int				 num_of_memcpy_chans;
+	u32				 memcpy_channels[D40_MEMCPY_MAX_CHANS];
 	int				 num_of_phy_chans;
 };
 
@@ -86,25 +90,6 @@ struct stedma40_platform_data {
 #define D40_ALLOC_PHY		BIT(30)
 #define D40_ALLOC_LOG_FREE	0
 
-#define D40_MEMCPY_MAX_CHANS	8
-
-/* Reserved event lines for memcpy only. */
-#define DB8500_DMA_MEMCPY_EV_0	51
-#define DB8500_DMA_MEMCPY_EV_1	56
-#define DB8500_DMA_MEMCPY_EV_2	57
-#define DB8500_DMA_MEMCPY_EV_3	58
-#define DB8500_DMA_MEMCPY_EV_4	59
-#define DB8500_DMA_MEMCPY_EV_5	60
-
-static int dma40_memcpy_channels[] = {
-	DB8500_DMA_MEMCPY_EV_0,
-	DB8500_DMA_MEMCPY_EV_1,
-	DB8500_DMA_MEMCPY_EV_2,
-	DB8500_DMA_MEMCPY_EV_3,
-	DB8500_DMA_MEMCPY_EV_4,
-	DB8500_DMA_MEMCPY_EV_5,
-};
-
 /* Default configuration for physical memcpy */
 static const struct stedma40_chan_cfg dma40_memcpy_conf_phy = {
 	.mode = STEDMA40_MODE_PHYSICAL,
@@ -2145,7 +2130,8 @@ static int d40_config_memcpy(struct d40_chan *d40c)
 
 	if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
 		d40c->dma_cfg = dma40_memcpy_conf_log;
-		d40c->dma_cfg.dev_type = dma40_memcpy_channels[d40c->chan.chan_id];
+		d40c->dma_cfg.dev_type =
+			d40c->base->plat_data->memcpy_channels[d40c->chan.chan_id];
 
 		d40_log_cfg(&d40c->dma_cfg,
 			    &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
@@ -3428,12 +3414,7 @@ static int __init d40_hw_detect_init(struct platform_device *pdev,
 	num_phy_chans = min(num_phy_chans, STEDMA40_MAX_PHYS);
 
 	/* The number of channels used for memcpy */
-	if (plat_data->num_of_memcpy_chans)
-		num_memcpy_chans = plat_data->num_of_memcpy_chans;
-	else
-		num_memcpy_chans = ARRAY_SIZE(dma40_memcpy_channels);
-
-	num_memcpy_chans = min(num_memcpy_chans, D40_MEMCPY_MAX_CHANS);
+	num_memcpy_chans = plat_data->num_of_memcpy_chans;
 	num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY;
 
 	dev_info(dev,
@@ -3682,6 +3663,7 @@ static int __init d40_of_probe(struct device *dev,
 	struct stedma40_platform_data *pdata;
 	int num_phy = 0, num_memcpy = 0, num_disabled = 0;
 	const __be32 *list;
+	int ret;
 
 	pdata = devm_kzalloc(dev, sizeof(*pdata), GFP_KERNEL);
 	if (!pdata)
@@ -3695,18 +3677,19 @@ static int __init d40_of_probe(struct device *dev,
 	list = of_get_property(np, "memcpy-channels", &num_memcpy);
 	num_memcpy /= sizeof(*list);
 
-	if (num_memcpy > D40_MEMCPY_MAX_CHANS || num_memcpy <= 0) {
+	if (num_memcpy > ARRAY_SIZE(pdata->memcpy_channels) ||
+	    num_memcpy <= 0) {
 		d40_err(dev,
 			"Invalid number of memcpy channels specified (%d)\n",
 			num_memcpy);
 		return -EINVAL;
 	}
+	ret = of_property_read_u32_array(np, "memcpy-channels",
+					 pdata->memcpy_channels, num_memcpy);
+	if (ret)
+		return ret;
 	pdata->num_of_memcpy_chans = num_memcpy;
 
-	of_property_read_u32_array(np, "memcpy-channels",
-				   dma40_memcpy_channels,
-				   num_memcpy);
-
 	list = of_get_property(np, "disabled-channels", &num_disabled);
 	num_disabled /= sizeof(*list);
 
Signature · 1 line
2.55.0
author maintainer
#28 lore raw

PATCH v5 16/23 dmaengine: ste_dma40: Validate disabled channel indexes

In reply to Linus Walleij

d40_of_probe() checks how many entries disabled-channels contains, but not the channel numbers themselves. d40_phy_res_init() later uses every value as an index into base->phy_res, whose size is the number of channels found in this DMA40 instance. An out-of-range value can therefore write beyond the allocation.

Validate each disabled channel against the detected hardware channel count before allocating and initializing the channel resources.

Fixes 499c2bc3cc89 ("dmaengine: ste_dma40: Fetch disabled channels from DT")
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 9 +++++++++
 1 file changed, 9 insertions(+)
drivers/dma/ste_dma40.c +9 -0
index 48dccd432ff4..8ae236d8270f 100644
@@ -3413,6 +3413,15 @@ static int __init d40_hw_detect_init(struct platform_device *pdev,
 
 	num_phy_chans = min(num_phy_chans, STEDMA40_MAX_PHYS);
 
+	for (i = 0; plat_data->disabled_channels[i] != -1; i++) {
+		int chan = plat_data->disabled_channels[i];
+
+		if (chan < 0 || chan >= num_phy_chans) {
+			dev_err(dev, "Invalid disabled channel %d\n", chan);
+			return -EINVAL;
+		}
+	}
+
 	/* The number of channels used for memcpy */
 	num_memcpy_chans = plat_data->num_of_memcpy_chans;
 	num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY;
Signature · 1 line
2.55.0
author maintainer
#29 lore raw

PATCH v5 17/23 dmaengine: ste_dma40: Validate DMA specifier length

In reply to Linus Walleij

The DMA40 binding requires three cells, but d40_xlate() reads args[0], args[1], and args[2] without checking args_count. A malformed provider node can specify fewer cells, leaving some of these values uninitialized when the OF DMA core invokes the translation callback.

Reject specifiers that do not contain exactly three cells before reading the argument array.

Fixes fa332de5c6b3 ("dmaengine: ste_dma40: Supply full Device Tree parsing support")
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/dma/ste_dma40.c +3 -0
index 8ae236d8270f..9db4a325b36e 100644
@@ -2538,6 +2538,9 @@ static struct dma_chan *d40_xlate(struct of_phandle_args *dma_spec,
 	dma_cap_mask_t cap;
 	u32 flags;
 
+	if (dma_spec->args_count != 3)
+		return NULL;
+
 	memset(&cfg, 0, sizeof(struct stedma40_chan_cfg));
 
 	dma_cap_zero(cap);
Signature · 1 line
2.55.0
author maintainer
#30 lore raw

PATCH v5 18/23 dmaengine: ste_dma40: Reject direction changes after allocation

In reply to Linus Walleij

Logical channel allocation uses the configured direction to select the source or destination allocation mask, derive the logical channel number, and choose the LCPA location.

d40_set_runtime_config_write() nevertheless overwrites the configured direction when a transfer is prepared for the opposite direction. d40_free_dma() then clears the wrong allocation mask, leaking the original resource and potentially releasing one used by another client.

Reject directions that differ from the direction used during allocation and propagate runtime configuration errors to callers. Skip slave runtime configuration for memcpy transfers, which also use d40_prep_sg() but do not require it.

Fixes 95e1400fa131 ("DMAENGINE: add runtime slave config to DMA40 v3")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 29 ++++++++++++++---------------
 1 file changed, 14 insertions(+), 15 deletions(-)
drivers/dma/ste_dma40.c +14 -15
index 9db4a325b36e..2966a8167c22 100644
@@ -2406,7 +2406,13 @@ d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
 		return NULL;
 	}
 
-	d40_set_runtime_config_write(dchan, &chan->slave_config, direction);
+	if (direction != DMA_MEM_TO_MEM) {
+		ret = d40_set_runtime_config_write(dchan,
+						   &chan->slave_config,
+						   direction);
+		if (ret)
+			return NULL;
+	}
 
 	spin_lock_irqsave(&chan->lock, flags);
 
@@ -2880,6 +2886,13 @@ static int d40_set_runtime_config_write(struct dma_chan *chan,
 		return -EINVAL;
 	}
 
+	if (direction != cfg->dir) {
+		chan_err(d40c,
+			 "transfer direction %d differs from allocated direction %d\n",
+			 direction, cfg->dir);
+		return -EINVAL;
+	}
+
 	src_addr_width = config->src_addr_width;
 	src_maxburst = config->src_maxburst;
 	dst_addr_width = config->dst_addr_width;
@@ -2888,13 +2901,6 @@ static int d40_set_runtime_config_write(struct dma_chan *chan,
 	if (direction == DMA_DEV_TO_MEM) {
 		config_addr = config->src_addr;
 
-		if (cfg->dir != DMA_DEV_TO_MEM)
-			dev_dbg(d40c->base->dev,
-				"channel was not configured for peripheral "
-				"to memory transfer (%d) overriding\n",
-				cfg->dir);
-		cfg->dir = DMA_DEV_TO_MEM;
-
 		/* Configure the memory side */
 		if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
 			dst_addr_width = src_addr_width;
@@ -2904,13 +2910,6 @@ static int d40_set_runtime_config_write(struct dma_chan *chan,
 	} else if (direction == DMA_MEM_TO_DEV) {
 		config_addr = config->dst_addr;
 
-		if (cfg->dir != DMA_MEM_TO_DEV)
-			dev_dbg(d40c->base->dev,
-				"channel was not configured for memory "
-				"to peripheral transfer (%d) overriding\n",
-				cfg->dir);
-		cfg->dir = DMA_MEM_TO_DEV;
-
 		/* Configure the memory side */
 		if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
 			src_addr_width = dst_addr_width;
Signature · 1 line
2.55.0
author maintainer
#31 lore raw

PATCH v5 19/23 dmaengine: ste_dma40: Fix logical channel bounds check

In reply to Linus Walleij

d40_validate_conf() checks the raw dev_type against num_log_chans, but d40_allocate_channel() derives the lookup_log_chans index as either 2 * dev_type or 2 * dev_type + 1.

Validate the dev_type against the derived logical channel index limit so channel allocation cannot write past lookup_log_chans.

Fixes 26955c07dcf3 ("dmaengine: ste_dma40: Amalgamate DMA source and destination channel numbers")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)
drivers/dma/ste_dma40.c +14 -3
index 2966a8167c22..b4dff597fa7c 100644
@@ -1872,15 +1872,26 @@ static int d40_validate_conf(struct d40_chan *d40c,
 {
 	int res = 0;
 	bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
+	bool invalid_dev_type = conf->dev_type < 0;
 
 	if (!conf->dir) {
 		chan_err(d40c, "Invalid direction.\n");
 		res = -EINVAL;
 	}
 
-	if ((is_log && conf->dev_type > d40c->base->num_log_chans)  ||
-	    (!is_log && conf->dev_type > d40c->base->num_phy_chans) ||
-	    (conf->dev_type < 0)) {
+	if (!invalid_dev_type && is_log) {
+		int max_dev_type;
+
+		if (conf->dir == DMA_DEV_TO_MEM)
+			max_dev_type = DIV_ROUND_UP(d40c->base->num_log_chans, 2);
+		else
+			max_dev_type = d40c->base->num_log_chans / 2;
+
+		invalid_dev_type = conf->dev_type >= max_dev_type;
+	}
+
+	if (invalid_dev_type ||
+	    (!is_log && conf->dev_type > d40c->base->num_phy_chans)) {
 		chan_err(d40c, "Invalid device type (%d)\n", conf->dev_type);
 		res = -EINVAL;
 	}
Signature · 1 line
2.55.0
author maintainer
#32 lore raw

PATCH v5 20/23 dmaengine: ste_dma40: Fix event group bounds

In reply to Linus Walleij

The dev_type validation can allow values whose derived event group has no matching physical channel pair. d40_allocate_channel() then indexes phy_res with j + event_group * 2, and __d40_set_prio_rt() uses the same group to select priority and realtime registers.

The physical-mode validation also compares dev_type with the number of physical channels. However, dev_type identifies an event line: DB8500 has eight physical channels but 64 source and 64 destination event lines. The event group determines which physical channel pair can serve an event, so the physical channel count is not a valid dev_type limit.

Reject dev_type values outside the hardware event-group range, remove the incorrect physical channel count limit, and stop the physical-channel search before a partial final channel group can index past phy_res.

Fixes 26955c07dcf3 ("dmaengine: ste_dma40: Amalgamate DMA source and destination channel numbers")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 26 ++++++++++++++++++++++----
 1 file changed, 22 insertions(+), 4 deletions(-)
drivers/dma/ste_dma40.c +22 -4
index b4dff597fa7c..13d1592bb3a4 100644
@@ -79,6 +79,10 @@ struct stedma40_platform_data {
 #define D40_LCLA_LINK_PER_EVENT_GRP 128
 #define D40_LCLA_END D40_LCLA_LINK_PER_EVENT_GRP
 
+/* Number of event groups per hardware register layout */
+#define D40_EVENT_GROUPS_V4A 4
+#define D40_EVENT_GROUPS_V4B 5
+
 /* Max number of logical channels per physical channel */
 #define D40_MAX_LOG_CHAN_PER_PHY 32
 
@@ -519,6 +523,7 @@ struct d40_chan {
  * @high_prio_clear: the high priority clear register
  * @interrupt_en: the interrupt enable register
  * @interrupt_clear: the interrupt clear register
+ * @num_event_groups: number of supported event groups
  * @il: the pointer to struct d40_interrupt_lookup
  * @il_size: the size of d40_interrupt_lookup array
  * @init_reg: the pointer to the struct d40_reg_val
@@ -533,6 +538,7 @@ struct d40_gen_dmac {
 	u32				 high_prio_clear;
 	u32				 interrupt_en;
 	u32				 interrupt_clear;
+	u32				 num_event_groups;
 	struct d40_interrupt_lookup	*il;
 	u32				 il_size;
 	struct d40_reg_val		*init_reg;
@@ -1874,6 +1880,11 @@ static int d40_validate_conf(struct d40_chan *d40c,
 	bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
 	bool invalid_dev_type = conf->dev_type < 0;
 
+	if (!invalid_dev_type &&
+	    D40_TYPE_TO_GROUP(conf->dev_type) >=
+	    d40c->base->gen_dmac.num_event_groups)
+		invalid_dev_type = true;
+
 	if (!conf->dir) {
 		chan_err(d40c, "Invalid direction.\n");
 		res = -EINVAL;
@@ -1890,8 +1901,7 @@ static int d40_validate_conf(struct d40_chan *d40c,
 		invalid_dev_type = conf->dev_type >= max_dev_type;
 	}
 
-	if (invalid_dev_type ||
-	    (!is_log && conf->dev_type > d40c->base->num_phy_chans)) {
+	if (invalid_dev_type) {
 		chan_err(d40c, "Invalid device type (%d)\n", conf->dev_type);
 		res = -EINVAL;
 	}
@@ -2056,8 +2066,12 @@ static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
 				}
 			}
 		} else
-			for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
+			for (j = 0; j < d40c->base->num_phy_chans;
+			     j += D40_GROUP_SIZE) {
 				int phy_num = j  + event_group * 2;
+				if (phy_num + 1 >= num_phy_chans)
+					break;
+
 				for (i = phy_num; i < phy_num + 2; i++) {
 					if (d40_alloc_mask_set(&phys[i],
 							       is_src,
@@ -2077,8 +2091,10 @@ static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
 		return -EINVAL;
 
 	/* Find logical channel */
-	for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
+	for (j = 0; j < d40c->base->num_phy_chans; j += D40_GROUP_SIZE) {
 		int phy_num = j + event_group * 2;
+		if (phy_num + 1 >= num_phy_chans)
+			break;
 
 		if (d40c->dma_cfg.use_fixed_channel) {
 			i = d40c->dma_cfg.phy_channel;
@@ -3461,6 +3477,7 @@ static int __init d40_hw_detect_init(struct platform_device *pdev,
 	base->log_chans = &base->phy_chans[num_phy_chans];
 
 	if (base->plat_data->num_of_phy_chans == 14) {
+		base->gen_dmac.num_event_groups = D40_EVENT_GROUPS_V4B;
 		base->gen_dmac.backup = d40_backup_regs_v4b;
 		base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
 		base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
@@ -3474,6 +3491,7 @@ static int __init d40_hw_detect_init(struct platform_device *pdev,
 		base->gen_dmac.init_reg = dma_init_reg_v4b;
 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
 	} else {
+		base->gen_dmac.num_event_groups = D40_EVENT_GROUPS_V4A;
 		if (base->rev >= 3) {
 			base->gen_dmac.backup = d40_backup_regs_v4a;
 			base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
Signature · 1 line
2.55.0
author maintainer
#33 lore raw

PATCH v5 21/23 dmaengine: ste_dma40: Search all blocks for fixed logical channels

In reply to Linus Walleij

Fixed logical channel allocation scans physical channels in blocks of eight. When the requested physical channel does not belong to the first block, d40_allocate_channel() returns -EINVAL instead of checking later blocks.

Skip nonmatching blocks so controllers with more than eight physical channels can allocate fixed logical channels from the later blocks.

Fixes 5cd326fd27da ("dmaengine/ste_dma40: allow fixed physical channel")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)
drivers/dma/ste_dma40.c +2 -5
index 13d1592bb3a4..41c7561314c1 100644
@@ -2099,11 +2099,8 @@ static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
 		if (d40c->dma_cfg.use_fixed_channel) {
 			i = d40c->dma_cfg.phy_channel;
 
-			if ((i != phy_num) && (i != phy_num + 1)) {
-				dev_err(chan2dev(d40c),
-					"invalid fixed phy channel %d\n", i);
-				return -EINVAL;
-			}
+			if (i != phy_num && i != phy_num + 1)
+				continue;
 
 			if (d40_alloc_mask_set(&phys[i], is_src, event_line,
 					       is_log, first_phy_user))
Signature · 1 line
2.55.0
author maintainer
#34 lore raw

PATCH v5 22/23 dmaengine: ste_dma40: Validate fixed physical channel indexes

In reply to Linus Walleij

Fixed physical memcpy allocation uses dma_cfg.phy_channel directly as an index into phy_res when use_fixed_channel is set. d40_validate_conf() validates dev_type but not the fixed physical channel, allowing an invalid configuration to access memory outside the array.

Validate the fixed physical channel centrally before accepting the channel configuration.

Fixes f000df8c5a0e ("dmaengine: ste_dma40: support fixed physical channel allocation")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/dma/ste_dma40.c +8 -0
index 41c7561314c1..a3159a657040 100644
@@ -1906,6 +1906,14 @@ static int d40_validate_conf(struct d40_chan *d40c,
 		res = -EINVAL;
 	}
 
+	if (conf->use_fixed_channel &&
+	    (conf->phy_channel < 0 ||
+	     conf->phy_channel >= d40c->base->num_phy_chans)) {
+		chan_err(d40c, "Invalid physical channel (%d)\n",
+			 conf->phy_channel);
+		res = -EINVAL;
+	}
+
 	if (conf->dir == DMA_DEV_TO_DEV) {
 		/*
 		 * DMAC HW supports it. Will be added to this driver,
Signature · 1 line
2.55.0
author maintainer
#35 lore raw

PATCH v5 23/23 dmaengine: ste_dma40: Validate memcpy configuration

In reply to Linus Walleij

d40_config_memcpy() builds a default memcpy configuration without passing it through d40_validate_conf(). A dev_type supplied through the memcpy-channels device tree property can therefore bypass the bounds checks added for client configurations.

The generic DMA direction enum assigns zero to DMA_MEM_TO_MEM, unlike DMA40's old private enum. Allow memory-to-memory directions in the validator so checking the generated configuration does not reject all memcpy channels.

Validate the generated memcpy configuration before deriving logical channel registers or allocating the channel.

Fixes 2c2b62d5d911 ("dmaengine: ste_dma40: Replace ST-E's home-brew DMA direction defs with generic ones")
Reported-by [email protected]
Assisted-by LLM
---
 drivers/dma/ste_dma40.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
drivers/dma/ste_dma40.c +11 -1
index a3159a657040..57d556e46402 100644
@@ -1885,7 +1885,8 @@ static int d40_validate_conf(struct d40_chan *d40c,
 	    d40c->base->gen_dmac.num_event_groups)
 		invalid_dev_type = true;
 
-	if (!conf->dir) {
+	if (conf->dir != DMA_MEM_TO_MEM &&
+	    !is_slave_direction(conf->dir)) {
 		chan_err(d40c, "Invalid direction.\n");
 		res = -EINVAL;
 	}
@@ -2159,12 +2160,17 @@ static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
 static int d40_config_memcpy(struct d40_chan *d40c)
 {
 	dma_cap_mask_t cap = d40c->chan.device->cap_mask;
+	int ret;
 
 	if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
 		d40c->dma_cfg = dma40_memcpy_conf_log;
 		d40c->dma_cfg.dev_type =
 			d40c->base->plat_data->memcpy_channels[d40c->chan.chan_id];
 
+		ret = d40_validate_conf(d40c, &d40c->dma_cfg);
+		if (ret)
+			return ret;
+
 		d40_log_cfg(&d40c->dma_cfg,
 			    &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
 
@@ -2172,6 +2178,10 @@ static int d40_config_memcpy(struct d40_chan *d40c)
 		   dma_has_cap(DMA_SLAVE, cap)) {
 		d40c->dma_cfg = dma40_memcpy_conf_phy;
 
+		ret = d40_validate_conf(d40c, &d40c->dma_cfg);
+		if (ret)
+			return ret;
+
 		/* Generate interrupt at end of transfer or relink. */
 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_TIM_POS);
 
Signature · 1 line
2.55.0