[PATCH] spi: sunxi: wait for TX/RX fifo reset done

[PATCH] spi: sunxi: wait for TX/RX fifo reset done

[Yixun Lan]

Once reset SPI TX or RX fifo, the underlying hardware need to take
some time to actually settle down, the two bits will automatically
clear to 0, so use a poll mechanism to check status bits to make sure
it's done correctly.

Signed-off-by: Yixun Lan <dlan@gentoo.org>
---
On Cubie A7A board which using A733 SoC, we encoutered a SPI nor flash
timeout issue, it turns out that the SPI fifo reset take a few time to
settle down, Add a loop to poll the status.

This was the error message shows on A7A board once this issue happened.

=> sf probe
ERROR: sun4i_spi: Timeout transferring data
Failed to initialize SPI flash at 0:0 (error -2)
---
 drivers/spi/spi-sunxi.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/spi/spi-sunxi.c b/drivers/spi/spi-sunxi.c
index e00532a371b..cf41905c7b7 100644
--- a/drivers/spi/spi-sunxi.c
+++ b/drivers/spi/spi-sunxi.c
@@ -347,7 +347,7 @@ static int sun4i_spi_xfer(struct udevice *dev, unsigned int bitlen,
 	struct sun4i_spi_priv *priv = dev_get_priv(bus);
 	struct dm_spi_slave_plat *slave_plat = dev_get_parent_plat(dev);
 
-	u32 len = bitlen / 8;
+	u32 rst, val, len = bitlen / 8;
 	u8 nbytes;
 	int ret;
 
@@ -363,8 +363,11 @@ static int sun4i_spi_xfer(struct udevice *dev, unsigned int bitlen,
 		sun4i_spi_set_cs(bus, slave_plat->cs[0], true);
 
 	/* Reset FIFOs */
-	setbits_le32(SPI_REG(priv, SPI_FCR), SPI_BIT(priv, SPI_FCR_RF_RST) |
-		     SPI_BIT(priv, SPI_FCR_TF_RST));
+	rst = SPI_BIT(priv, SPI_FCR_RF_RST) | SPI_BIT(priv, SPI_FCR_TF_RST);
+	setbits_le32(SPI_REG(priv, SPI_FCR), rst);
+	ret = readl_poll_timeout(SPI_REG(priv, SPI_FCR), val, !(rst & val), 20);
+	if (ret)
+		return -EBUSY;
 
 	while (len) {
 		/* Setup the transfer now... */

---
base-commit: 88dc2788777babfd6322fa655df549a019aa1e69
change-id: 20260220-02-spi-fifo-reset-25e371314c3f

Best regards,
-- 
Yixun Lan <dlan@gentoo.org>

Re: [PATCH] spi: sunxi: wait for TX/RX fifo reset done

[Jernej Škrabec]

Dne torek, 21. april 2026 ob 06:47:50 Srednjeevropski poletni čas je Yixun Lan napisal(a):
> Once reset SPI TX or RX fifo, the underlying hardware need to take
> some time to actually settle down, the two bits will automatically
> clear to 0, so use a poll mechanism to check status bits to make sure
> it's done correctly.
> 
> Signed-off-by: Yixun Lan <dlan@gentoo.org>
> ---
> On Cubie A7A board which using A733 SoC, we encoutered a SPI nor flash
> timeout issue, it turns out that the SPI fifo reset take a few time to
> settle down, Add a loop to poll the status.
> 
> This was the error message shows on A7A board once this issue happened.
> 
> => sf probe
> ERROR: sun4i_spi: Timeout transferring data
> Failed to initialize SPI flash at 0:0 (error -2)

Reviewed-by: Jernej Skrabec <jernej.skrabec@gmail.com>

Best regards,
Jernej

Re: [PATCH] spi: sunxi: wait for TX/RX fifo reset done

[Andre Przywara]

Hi,

On 4/21/26 06:47, Yixun Lan wrote:
> Once reset SPI TX or RX fifo, the underlying hardware need to take
> some time to actually settle down, the two bits will automatically
> clear to 0, so use a poll mechanism to check status bits to make sure
> it's done correctly.

Ah, interesting, thanks for posting this!
I looked into some manuals, and it seems like this self-clearing 
property is already used in the A10, so it's fine to use unconditionally.

> 
> Signed-off-by: Yixun Lan <dlan@gentoo.org>

Acked-by: Andre Przywara <andre.przywara@arm.com>

If you don't mind, I would pull the below paragraph into the commit 
message, since it's useful to have in the git history.

Cheers,
Andre.

P.S. Just curious if we need a similar fix for Linux, or are we saved by 
the Linux code spending more time in setup before doing a transfer?

> ---
> On Cubie A7A board which using A733 SoC, we encoutered a SPI nor flash
> timeout issue, it turns out that the SPI fifo reset take a few time to
> settle down, Add a loop to poll the status.
> 
> This was the error message shows on A7A board once this issue happened.
> 
> => sf probe
> ERROR: sun4i_spi: Timeout transferring data
> Failed to initialize SPI flash at 0:0 (error -2)
> ---
>   drivers/spi/spi-sunxi.c | 9 ++++++---
>   1 file changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/spi/spi-sunxi.c b/drivers/spi/spi-sunxi.c
> index e00532a371b..cf41905c7b7 100644
> --- a/drivers/spi/spi-sunxi.c
> +++ b/drivers/spi/spi-sunxi.c
> @@ -347,7 +347,7 @@ static int sun4i_spi_xfer(struct udevice *dev, unsigned int bitlen,
>   	struct sun4i_spi_priv *priv = dev_get_priv(bus);
>   	struct dm_spi_slave_plat *slave_plat = dev_get_parent_plat(dev);
>   
> -	u32 len = bitlen / 8;
> +	u32 rst, val, len = bitlen / 8;
>   	u8 nbytes;
>   	int ret;
>   
> @@ -363,8 +363,11 @@ static int sun4i_spi_xfer(struct udevice *dev, unsigned int bitlen,
>   		sun4i_spi_set_cs(bus, slave_plat->cs[0], true);
>   
>   	/* Reset FIFOs */
> -	setbits_le32(SPI_REG(priv, SPI_FCR), SPI_BIT(priv, SPI_FCR_RF_RST) |
> -		     SPI_BIT(priv, SPI_FCR_TF_RST));
> +	rst = SPI_BIT(priv, SPI_FCR_RF_RST) | SPI_BIT(priv, SPI_FCR_TF_RST);
> +	setbits_le32(SPI_REG(priv, SPI_FCR), rst);
> +	ret = readl_poll_timeout(SPI_REG(priv, SPI_FCR), val, !(rst & val), 20);
> +	if (ret)
> +		return -EBUSY;
>   
>   	while (len) {
>   		/* Setup the transfer now... */
> 
> ---
> base-commit: 88dc2788777babfd6322fa655df549a019aa1e69
> change-id: 20260220-02-spi-fifo-reset-25e371314c3f
> 
> Best regards,

Re: [PATCH] spi: sunxi: wait for TX/RX fifo reset done

[Yixun Lan]

Hi Andre,

On 11:54 Thu 30 Apr     , Andre Przywara wrote:
> Hi,
> 
> On 4/21/26 06:47, Yixun Lan wrote:
> > Once reset SPI TX or RX fifo, the underlying hardware need to take
> > some time to actually settle down, the two bits will automatically
> > clear to 0, so use a poll mechanism to check status bits to make sure
> > it's done correctly.
> 
> Ah, interesting, thanks for posting this!
> I looked into some manuals, and it seems like this self-clearing 
> property is already used in the A10, so it's fine to use unconditionally.
> 
> > 
> > Signed-off-by: Yixun Lan <dlan@gentoo.org>
> 
> Acked-by: Andre Przywara <andre.przywara@arm.com>
> 
> If you don't mind, I would pull the below paragraph into the commit 
> message, since it's useful to have in the git history.
> 
No, feel free to adjust.. And thanks
I'd assume you will handle all this? so no need from my side to post new version

> Cheers,
> Andre.
> 
> P.S. Just curious if we need a similar fix for Linux, or are we saved by 
> the Linux code spending more time in setup before doing a transfer?
> 
I've not looked into Linux/Kernel side, since this is a hardware feature,
so yes, we should do something similar (to be safe)

> > ---
> > On Cubie A7A board which using A733 SoC, we encoutered a SPI nor flash
> > timeout issue, it turns out that the SPI fifo reset take a few time to
> > settle down, Add a loop to poll the status.
> > 
> > This was the error message shows on A7A board once this issue happened.
> > 
> > => sf probe
> > ERROR: sun4i_spi: Timeout transferring data
> > Failed to initialize SPI flash at 0:0 (error -2)
> > ---
> >   drivers/spi/spi-sunxi.c | 9 ++++++---
> >   1 file changed, 6 insertions(+), 3 deletions(-)
> > 
> > diff --git a/drivers/spi/spi-sunxi.c b/drivers/spi/spi-sunxi.c
> > index e00532a371b..cf41905c7b7 100644
> > --- a/drivers/spi/spi-sunxi.c
> > +++ b/drivers/spi/spi-sunxi.c
> > @@ -347,7 +347,7 @@ static int sun4i_spi_xfer(struct udevice *dev, unsigned int bitlen,
> >   	struct sun4i_spi_priv *priv = dev_get_priv(bus);
> >   	struct dm_spi_slave_plat *slave_plat = dev_get_parent_plat(dev);
> >   
> > -	u32 len = bitlen / 8;
> > +	u32 rst, val, len = bitlen / 8;
> >   	u8 nbytes;
> >   	int ret;
> >   
> > @@ -363,8 +363,11 @@ static int sun4i_spi_xfer(struct udevice *dev, unsigned int bitlen,
> >   		sun4i_spi_set_cs(bus, slave_plat->cs[0], true);
> >   
> >   	/* Reset FIFOs */
> > -	setbits_le32(SPI_REG(priv, SPI_FCR), SPI_BIT(priv, SPI_FCR_RF_RST) |
> > -		     SPI_BIT(priv, SPI_FCR_TF_RST));
> > +	rst = SPI_BIT(priv, SPI_FCR_RF_RST) | SPI_BIT(priv, SPI_FCR_TF_RST);
> > +	setbits_le32(SPI_REG(priv, SPI_FCR), rst);
> > +	ret = readl_poll_timeout(SPI_REG(priv, SPI_FCR), val, !(rst & val), 20);
> > +	if (ret)
> > +		return -EBUSY;
> >   
> >   	while (len) {
> >   		/* Setup the transfer now... */
> > 
> > ---
> > base-commit: 88dc2788777babfd6322fa655df549a019aa1e69
> > change-id: 20260220-02-spi-fifo-reset-25e371314c3f
> > 
> > Best regards,
> 

-- 
Yixun Lan (dlan)