[PATCH v3 0/2] Fix CPU stall in xilinx_dma_poll_timeout caused by passing delay_us=0

[PATCH v3 0/2] Fix CPU stall in xilinx_dma_poll_timeout caused by passing delay_us=0

[Alex Bereza]

Signed-off-by: Alex Bereza <alex@bereza.email>
---
Changes in v3:
- patch 1/2:
  - Fix commit message: remove blank line between tags
- patch 2/2: nothing
- Link to v2: https://patch.msgid.link/20260401-fix-atomic-poll-timeout-regression-v2-0-68a265e3770f@bereza.email

Changes in v2:
- Fixed the Fixes: tags as suggested by Geert Uytterhoeven
  <geert+renesas@glider.be> - thanks!
- Split the renaming of XILINX_DMA_LOOP_COUNT into separate commit
- Link to v1: https://patch.msgid.link/20260331-fix-atomic-poll-timeout-regression-v1-1-5b7bd96eaca0@bereza.email

---
Alex Bereza (2):
      dmaengine: xilinx_dma: Fix CPU stall in xilinx_dma_poll_timeout
      dmaengine: xilinx_dma: Rename XILINX_DMA_LOOP_COUNT

 drivers/dma/xilinx/xilinx_dma.c | 26 ++++++++++++++++----------
 1 file changed, 16 insertions(+), 10 deletions(-)
---
base-commit: b7560798466a07d9c3fb011698e92c335ab28baf
change-id: 20260330-fix-atomic-poll-timeout-regression-4f4e3baf3fd7

Best regards,
--  
Alex Bereza <alex@bereza.email>

[PATCH v3 1/2] dmaengine: xilinx_dma: Fix CPU stall in xilinx_dma_poll_timeout

[Alex Bereza]

Currently when calling xilinx_dma_poll_timeout with delay_us=0 and a
condition that is never fulfilled, the CPU busy-waits for prolonged time
and the timeout triggers only with a massive delay causing a CPU stall.

This happens due to a huge underestimation of wall clock time in
poll_timeout_us_atomic. Commit 7349a69cf312 ("iopoll: Do not use
timekeeping in read_poll_timeout_atomic()") changed the behavior to no
longer use ktime_get at the expense of underestimation of wall clock
time which appears to be very large for delay_us=0. Instead of timing
out after approximately XILINX_DMA_LOOP_COUNT microseconds, the timeout
takes XILINX_DMA_LOOP_COUNT * 1000 * (time that the overhead of the for
loop in poll_timeout_us_atomic takes) which is in the range of several
minutes for XILINX_DMA_LOOP_COUNT=1000000. Fix this by using a non-zero
value for delay_us. Use delay_us=10 to keep the delay in the hot path of
starting DMA transfers minimal but still avoid CPU stalls in case of
unexpected hardware failures.

One-off measurement with delay_us=0 causes the cpu to busy wait around 7
minutes in the timeout case. After applying this patch with delay_us=10
the measured timeout was 1053428 microseconds which is roughly
equivalent to the expected 1000000 microseconds specified in
XILINX_DMA_LOOP_COUNT.

Add a constant XILINX_DMA_POLL_DELAY_US for delay_us value.

Fixes: 9495f2648287 ("dmaengine: xilinx_vdma: Use readl_poll_timeout instead of do while loop's")
Fixes: 7349a69cf312 ("iopoll: Do not use timekeeping in read_poll_timeout_atomic()")
Signed-off-by: Alex Bereza <alex@bereza.email>
---
 drivers/dma/xilinx/xilinx_dma.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

diff --git a/drivers/dma/xilinx/xilinx_dma.c b/drivers/dma/xilinx/xilinx_dma.c
index 02a05f215614..345a738bab2c 100644
--- a/drivers/dma/xilinx/xilinx_dma.c
+++ b/drivers/dma/xilinx/xilinx_dma.c
@@ -167,6 +167,8 @@
 
 /* Delay loop counter to prevent hardware failure */
 #define XILINX_DMA_LOOP_COUNT		1000000
+/* Delay between polls (avoid a delay of 0 to prevent CPU stalls) */
+#define XILINX_DMA_POLL_DELAY_US	10
 
 /* AXI DMA Specific Registers/Offsets */
 #define XILINX_DMA_REG_SRCDSTADDR	0x18
@@ -1332,7 +1334,8 @@ static int xilinx_dma_stop_transfer(struct xilinx_dma_chan *chan)
 
 	/* Wait for the hardware to halt */
 	return xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMASR, val,
-				       val & XILINX_DMA_DMASR_HALTED, 0,
+				       val & XILINX_DMA_DMASR_HALTED,
+				       XILINX_DMA_POLL_DELAY_US,
 				       XILINX_DMA_LOOP_COUNT);
 }
 
@@ -1347,7 +1350,8 @@ static int xilinx_cdma_stop_transfer(struct xilinx_dma_chan *chan)
 	u32 val;
 
 	return xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMASR, val,
-				       val & XILINX_DMA_DMASR_IDLE, 0,
+				       val & XILINX_DMA_DMASR_IDLE,
+				       XILINX_DMA_POLL_DELAY_US,
 				       XILINX_DMA_LOOP_COUNT);
 }
 
@@ -1364,7 +1368,8 @@ static void xilinx_dma_start(struct xilinx_dma_chan *chan)
 
 	/* Wait for the hardware to start */
 	err = xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMASR, val,
-				      !(val & XILINX_DMA_DMASR_HALTED), 0,
+				      !(val & XILINX_DMA_DMASR_HALTED),
+				      XILINX_DMA_POLL_DELAY_US,
 				      XILINX_DMA_LOOP_COUNT);
 
 	if (err) {
@@ -1780,7 +1785,8 @@ static int xilinx_dma_reset(struct xilinx_dma_chan *chan)
 
 	/* Wait for the hardware to finish reset */
 	err = xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMACR, tmp,
-				      !(tmp & XILINX_DMA_DMACR_RESET), 0,
+				      !(tmp & XILINX_DMA_DMACR_RESET),
+				      XILINX_DMA_POLL_DELAY_US,
 				      XILINX_DMA_LOOP_COUNT);
 
 	if (err) {

-- 
2.53.0

[PATCH v3 2/2] dmaengine: xilinx_dma: Rename XILINX_DMA_LOOP_COUNT

[Alex Bereza]

Rename XILINX_DMA_LOOP_COUNT to XILINX_DMA_POLL_TIMEOUT_US because the
former is incorrect. It is a timeout value for polling various register
bits in microseconds. It is not a loop count.

Signed-off-by: Alex Bereza <alex@bereza.email>
---
 drivers/dma/xilinx/xilinx_dma.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/drivers/dma/xilinx/xilinx_dma.c b/drivers/dma/xilinx/xilinx_dma.c
index 345a738bab2c..253c27fd1a0e 100644
--- a/drivers/dma/xilinx/xilinx_dma.c
+++ b/drivers/dma/xilinx/xilinx_dma.c
@@ -165,8 +165,8 @@
 #define XILINX_DMA_FLUSH_MM2S		2
 #define XILINX_DMA_FLUSH_BOTH		1
 
-/* Delay loop counter to prevent hardware failure */
-#define XILINX_DMA_LOOP_COUNT		1000000
+/* Timeout for polling various registers */
+#define XILINX_DMA_POLL_TIMEOUT_US	1000000
 /* Delay between polls (avoid a delay of 0 to prevent CPU stalls) */
 #define XILINX_DMA_POLL_DELAY_US	10
 
@@ -1336,7 +1336,7 @@ static int xilinx_dma_stop_transfer(struct xilinx_dma_chan *chan)
 	return xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMASR, val,
 				       val & XILINX_DMA_DMASR_HALTED,
 				       XILINX_DMA_POLL_DELAY_US,
-				       XILINX_DMA_LOOP_COUNT);
+				       XILINX_DMA_POLL_TIMEOUT_US);
 }
 
 /**
@@ -1352,7 +1352,7 @@ static int xilinx_cdma_stop_transfer(struct xilinx_dma_chan *chan)
 	return xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMASR, val,
 				       val & XILINX_DMA_DMASR_IDLE,
 				       XILINX_DMA_POLL_DELAY_US,
-				       XILINX_DMA_LOOP_COUNT);
+				       XILINX_DMA_POLL_TIMEOUT_US);
 }
 
 /**
@@ -1370,7 +1370,7 @@ static void xilinx_dma_start(struct xilinx_dma_chan *chan)
 	err = xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMASR, val,
 				      !(val & XILINX_DMA_DMASR_HALTED),
 				      XILINX_DMA_POLL_DELAY_US,
-				      XILINX_DMA_LOOP_COUNT);
+				      XILINX_DMA_POLL_TIMEOUT_US);
 
 	if (err) {
 		dev_err(chan->dev, "Cannot start channel %p: %x\n",
@@ -1787,7 +1787,7 @@ static int xilinx_dma_reset(struct xilinx_dma_chan *chan)
 	err = xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMACR, tmp,
 				      !(tmp & XILINX_DMA_DMACR_RESET),
 				      XILINX_DMA_POLL_DELAY_US,
-				      XILINX_DMA_LOOP_COUNT);
+				      XILINX_DMA_POLL_TIMEOUT_US);
 
 	if (err) {
 		dev_err(chan->dev, "reset timeout, cr %x, sr %x\n",

-- 
2.53.0

RE: [PATCH v3 0/2] Fix CPU stall in xilinx_dma_poll_timeout caused by passing delay_us=0

[Gupta, Suraj]

[AMD Official Use Only - AMD Internal Distribution Only]

> -----Original Message-----
> From: Alex Bereza <alex@bereza.email>
> Sent: Wednesday, April 1, 2026 4:27 PM
> To: Vinod Koul <vkoul@kernel.org>; Frank Li <Frank.Li@kernel.org>; Simek,
> Michal <michal.simek@amd.com>; Geert Uytterhoeven
> <geert+renesas@glider.be>; Ulf Hansson <ulf.hansson@linaro.org>; Arnd
> Bergmann <arnd@arndb.de>; Tony Lindgren <tony@atomide.com>; Rao,
> Appana Durga Kedareswara <appana.durga.kedareswara.rao@amd.com>
> Cc: dmaengine@vger.kernel.org; linux-arm-kernel@lists.infradead.org; linux-
> kernel@vger.kernel.org; Alex Bereza <alex@bereza.email>
> Subject: [PATCH v3 0/2] Fix CPU stall in xilinx_dma_poll_timeout caused by
> passing delay_us=0
>
> [You don't often get email from alex@bereza.email. Learn why this is important
> at https://aka.ms/LearnAboutSenderIdentification ]
>
> Caution: This message originated from an External Source. Use proper caution
> when opening attachments, clicking links, or responding.
>
>
> Signed-off-by: Alex Bereza <alex@bereza.email>
> ---

For the series, Reviewed-by: Suraj Gupta <suraj.gupta2@amd.com>

Thanks!
Suraj

> Changes in v3:
> - patch 1/2:
>   - Fix commit message: remove blank line between tags
> - patch 2/2: nothing
> - Link to v2: https://patch.msgid.link/20260401-fix-atomic-poll-timeout-
> regression-v2-0-68a265e3770f@bereza.email
>
> Changes in v2:
> - Fixed the Fixes: tags as suggested by Geert Uytterhoeven
>   <geert+renesas@glider.be> - thanks!
> - Split the renaming of XILINX_DMA_LOOP_COUNT into separate commit
> - Link to v1: https://patch.msgid.link/20260331-fix-atomic-poll-timeout-
> regression-v1-1-5b7bd96eaca0@bereza.email
>
> ---
> Alex Bereza (2):
>       dmaengine: xilinx_dma: Fix CPU stall in xilinx_dma_poll_timeout
>       dmaengine: xilinx_dma: Rename XILINX_DMA_LOOP_COUNT
>
>  drivers/dma/xilinx/xilinx_dma.c | 26 ++++++++++++++++----------
>  1 file changed, 16 insertions(+), 10 deletions(-)
> ---
> base-commit: b7560798466a07d9c3fb011698e92c335ab28baf
> change-id: 20260330-fix-atomic-poll-timeout-regression-4f4e3baf3fd7
>
> Best regards,
> --
> Alex Bereza <alex@bereza.email>
>

Re: [PATCH v3 2/2] dmaengine: xilinx_dma: Rename XILINX_DMA_LOOP_COUNT

[Frank Li]

On Wed, Apr 01, 2026 at 12:56:33PM +0200, Alex Bereza wrote:
> Rename XILINX_DMA_LOOP_COUNT to XILINX_DMA_POLL_TIMEOUT_US because the
> former is incorrect. It is a timeout value for polling various register
> bits in microseconds. It is not a loop count.

Rename XILINX_DMA_LOOP_COUNT to XILINX_DMA_POLL_TIMEOUT_US because it is a
timeout value, not a loop count for polling register in microseconds.

No functional changes.

Frank

>
> Signed-off-by: Alex Bereza <alex@bereza.email>
> ---
>  drivers/dma/xilinx/xilinx_dma.c | 12 ++++++------
>  1 file changed, 6 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/dma/xilinx/xilinx_dma.c b/drivers/dma/xilinx/xilinx_dma.c
> index 345a738bab2c..253c27fd1a0e 100644
> --- a/drivers/dma/xilinx/xilinx_dma.c
> +++ b/drivers/dma/xilinx/xilinx_dma.c
> @@ -165,8 +165,8 @@
>  #define XILINX_DMA_FLUSH_MM2S		2
>  #define XILINX_DMA_FLUSH_BOTH		1
>
> -/* Delay loop counter to prevent hardware failure */
> -#define XILINX_DMA_LOOP_COUNT		1000000
> +/* Timeout for polling various registers */
> +#define XILINX_DMA_POLL_TIMEOUT_US	1000000
>  /* Delay between polls (avoid a delay of 0 to prevent CPU stalls) */
>  #define XILINX_DMA_POLL_DELAY_US	10
>
> @@ -1336,7 +1336,7 @@ static int xilinx_dma_stop_transfer(struct xilinx_dma_chan *chan)
>  	return xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMASR, val,
>  				       val & XILINX_DMA_DMASR_HALTED,
>  				       XILINX_DMA_POLL_DELAY_US,
> -				       XILINX_DMA_LOOP_COUNT);
> +				       XILINX_DMA_POLL_TIMEOUT_US);
>  }
>
>  /**
> @@ -1352,7 +1352,7 @@ static int xilinx_cdma_stop_transfer(struct xilinx_dma_chan *chan)
>  	return xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMASR, val,
>  				       val & XILINX_DMA_DMASR_IDLE,
>  				       XILINX_DMA_POLL_DELAY_US,
> -				       XILINX_DMA_LOOP_COUNT);
> +				       XILINX_DMA_POLL_TIMEOUT_US);
>  }
>
>  /**
> @@ -1370,7 +1370,7 @@ static void xilinx_dma_start(struct xilinx_dma_chan *chan)
>  	err = xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMASR, val,
>  				      !(val & XILINX_DMA_DMASR_HALTED),
>  				      XILINX_DMA_POLL_DELAY_US,
> -				      XILINX_DMA_LOOP_COUNT);
> +				      XILINX_DMA_POLL_TIMEOUT_US);
>
>  	if (err) {
>  		dev_err(chan->dev, "Cannot start channel %p: %x\n",
> @@ -1787,7 +1787,7 @@ static int xilinx_dma_reset(struct xilinx_dma_chan *chan)
>  	err = xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMACR, tmp,
>  				      !(tmp & XILINX_DMA_DMACR_RESET),
>  				      XILINX_DMA_POLL_DELAY_US,
> -				      XILINX_DMA_LOOP_COUNT);
> +				      XILINX_DMA_POLL_TIMEOUT_US);
>
>  	if (err) {
>  		dev_err(chan->dev, "reset timeout, cr %x, sr %x\n",
>
> --
> 2.53.0
>

Re: [PATCH v3 1/2] dmaengine: xilinx_dma: Fix CPU stall in xilinx_dma_poll_timeout

[Frank Li]

On Wed, Apr 01, 2026 at 12:56:32PM +0200, Alex Bereza wrote:
> Currently when calling xilinx_dma_poll_timeout with delay_us=0 and a
> condition that is never fulfilled, the CPU busy-waits for prolonged time
> and the timeout triggers only with a massive delay causing a CPU stall.
>
> This happens due to a huge underestimation of wall clock time in
> poll_timeout_us_atomic. Commit 7349a69cf312 ("iopoll: Do not use
> timekeeping in read_poll_timeout_atomic()") changed the behavior to no
> longer use ktime_get at the expense of underestimation of wall clock
> time which appears to be very large for delay_us=0. Instead of timing
> out after approximately XILINX_DMA_LOOP_COUNT microseconds, the timeout
> takes XILINX_DMA_LOOP_COUNT * 1000 * (time that the overhead of the for
> loop in poll_timeout_us_atomic takes) which is in the range of several
> minutes for XILINX_DMA_LOOP_COUNT=1000000. Fix this by using a non-zero
> value for delay_us. Use delay_us=10 to keep the delay in the hot path of
> starting DMA transfers minimal but still avoid CPU stalls in case of
> unexpected hardware failures.
>
> One-off measurement with delay_us=0 causes the cpu to busy wait around 7
> minutes in the timeout case. After applying this patch with delay_us=10
> the measured timeout was 1053428 microseconds which is roughly
> equivalent to the expected 1000000 microseconds specified in
> XILINX_DMA_LOOP_COUNT.
>
> Add a constant XILINX_DMA_POLL_DELAY_US for delay_us value.
>
> Fixes: 9495f2648287 ("dmaengine: xilinx_vdma: Use readl_poll_timeout instead of do while loop's")
> Fixes: 7349a69cf312 ("iopoll: Do not use timekeeping in read_poll_timeout_atomic()")
> Signed-off-by: Alex Bereza <alex@bereza.email>
> ---

Reviewed-by: Frank Li <Frank.Li@nxp.com>
>  drivers/dma/xilinx/xilinx_dma.c | 14 ++++++++++----
>  1 file changed, 10 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/dma/xilinx/xilinx_dma.c b/drivers/dma/xilinx/xilinx_dma.c
> index 02a05f215614..345a738bab2c 100644
> --- a/drivers/dma/xilinx/xilinx_dma.c
> +++ b/drivers/dma/xilinx/xilinx_dma.c
> @@ -167,6 +167,8 @@
>
>  /* Delay loop counter to prevent hardware failure */
>  #define XILINX_DMA_LOOP_COUNT		1000000
> +/* Delay between polls (avoid a delay of 0 to prevent CPU stalls) */
> +#define XILINX_DMA_POLL_DELAY_US	10
>
>  /* AXI DMA Specific Registers/Offsets */
>  #define XILINX_DMA_REG_SRCDSTADDR	0x18
> @@ -1332,7 +1334,8 @@ static int xilinx_dma_stop_transfer(struct xilinx_dma_chan *chan)
>
>  	/* Wait for the hardware to halt */
>  	return xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMASR, val,
> -				       val & XILINX_DMA_DMASR_HALTED, 0,
> +				       val & XILINX_DMA_DMASR_HALTED,
> +				       XILINX_DMA_POLL_DELAY_US,
>  				       XILINX_DMA_LOOP_COUNT);
>  }
>
> @@ -1347,7 +1350,8 @@ static int xilinx_cdma_stop_transfer(struct xilinx_dma_chan *chan)
>  	u32 val;
>
>  	return xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMASR, val,
> -				       val & XILINX_DMA_DMASR_IDLE, 0,
> +				       val & XILINX_DMA_DMASR_IDLE,
> +				       XILINX_DMA_POLL_DELAY_US,
>  				       XILINX_DMA_LOOP_COUNT);
>  }
>
> @@ -1364,7 +1368,8 @@ static void xilinx_dma_start(struct xilinx_dma_chan *chan)
>
>  	/* Wait for the hardware to start */
>  	err = xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMASR, val,
> -				      !(val & XILINX_DMA_DMASR_HALTED), 0,
> +				      !(val & XILINX_DMA_DMASR_HALTED),
> +				      XILINX_DMA_POLL_DELAY_US,
>  				      XILINX_DMA_LOOP_COUNT);
>
>  	if (err) {
> @@ -1780,7 +1785,8 @@ static int xilinx_dma_reset(struct xilinx_dma_chan *chan)
>
>  	/* Wait for the hardware to finish reset */
>  	err = xilinx_dma_poll_timeout(chan, XILINX_DMA_REG_DMACR, tmp,
> -				      !(tmp & XILINX_DMA_DMACR_RESET), 0,
> +				      !(tmp & XILINX_DMA_DMACR_RESET),
> +				      XILINX_DMA_POLL_DELAY_US,
>  				      XILINX_DMA_LOOP_COUNT);
>
>  	if (err) {
>
> --
> 2.53.0
>