Discussion:
[PATCH v2 01/12] spi: spi-mem: Extend the SPI mem interface to set a custom memory name
Frieder Schrempf
2018-07-05 11:14:57 UTC
Permalink
When porting (Q)SPI controller drivers from the MTD layer to the SPI
layer, the naming scheme for the memory devices changes. To be able
to keep compatibility with the old drivers naming scheme, a function
is added to the SPI mem interface to let controller drivers set a
custom name for the memory.

Example for the FSL QSPI driver:

Name with the old driver: 21e0000.qspi,
or with multiple devices: 21e0000.qspi-0, 21e0000.qspi-1, ...

Name with the new driver without spi_mem_get_name: spi4.0

Signed-off-by: Frieder Schrempf <***@exceet.de>
---
Changes in v2:
==============
* Add a name field to struct spi_mem and fill it while probing

drivers/spi/spi-mem.c | 30 ++++++++++++++++++++++++++++++
include/linux/spi/spi-mem.h | 7 ++++++-
2 files changed, 36 insertions(+), 1 deletion(-)

diff --git a/drivers/spi/spi-mem.c b/drivers/spi/spi-mem.c
index 990770d..048101c 100644
--- a/drivers/spi/spi-mem.c
+++ b/drivers/spi/spi-mem.c
@@ -311,6 +311,26 @@ int spi_mem_exec_op(struct spi_mem *mem, const struct spi_mem_op *op)
EXPORT_SYMBOL_GPL(spi_mem_exec_op);

/**
+ * spi_mem_get_name() - Let drivers using the SPI mem interface specify a
+ * custom name for the memory to avoid compatibility
+ * issues with ported drivers.
+ * @mem: the SPI memory
+ *
+ * When porting (Q)SPI controller drivers from the MTD layer to the SPI
+ * layer, the naming scheme for the memory devices changes. To be able to
+ * keep compatibility with the old drivers naming scheme, this function can
+ * be used to get a custom name from the controller driver.
+ * If no custom name is available, the name of the SPI device is returned.
+ *
+ * Return: a char array that contains the name for the flash memory
+ */
+const char *spi_mem_get_name(struct spi_mem *mem)
+{
+ return mem->name;
+}
+EXPORT_SYMBOL_GPL(spi_mem_get_name);
+
+/**
* spi_mem_adjust_op_size() - Adjust the data size of a SPI mem operation to
* match controller limitations
* @mem: the SPI memory
@@ -344,6 +364,7 @@ static inline struct spi_mem_driver *to_spi_mem_drv(struct device_driver *drv)
static int spi_mem_probe(struct spi_device *spi)
{
struct spi_mem_driver *memdrv = to_spi_mem_drv(spi->dev.driver);
+ struct spi_controller *ctlr = spi->controller;
struct spi_mem *mem;

mem = devm_kzalloc(&spi->dev, sizeof(*mem), GFP_KERNEL);
@@ -351,6 +372,15 @@ static int spi_mem_probe(struct spi_device *spi)
return -ENOMEM;

mem->spi = spi;
+
+ if (ctlr->mem_ops && ctlr->mem_ops->get_name)
+ mem->name = ctlr->mem_ops->get_name(mem);
+ else
+ mem->name = dev_name(&spi->dev);
+
+ if (IS_ERR_OR_NULL(mem->name))
+ return PTR_ERR(mem->name);
+
spi_set_drvdata(spi, mem);

return memdrv->probe(mem);
diff --git a/include/linux/spi/spi-mem.h b/include/linux/spi/spi-mem.h
index 4fa34a2..c3f82d1 100644
--- a/include/linux/spi/spi-mem.h
+++ b/include/linux/spi/spi-mem.h
@@ -124,7 +124,8 @@ struct spi_mem_op {
/**
* struct spi_mem - describes a SPI memory device
* @spi: the underlying SPI device
- * @drvpriv: spi_mem_drviver private data
+ * @drvpriv: spi_mem_driver private data
+ * @name: name of the SPI memory device
*
* Extra information that describe the SPI memory device and may be needed by
* the controller to properly handle this device should be placed here.
@@ -135,6 +136,7 @@ struct spi_mem_op {
struct spi_mem {
struct spi_device *spi;
void *drvpriv;
+ char *name;
};

/**
@@ -178,6 +180,7 @@ struct spi_controller_mem_ops {
const struct spi_mem_op *op);
int (*exec_op)(struct spi_mem *mem,
const struct spi_mem_op *op);
+ const char *(*get_name)(struct spi_mem *mem);
};

/**
@@ -236,6 +239,8 @@ bool spi_mem_supports_op(struct spi_mem *mem,
int spi_mem_exec_op(struct spi_mem *mem,
const struct spi_mem_op *op);

+const char *spi_mem_get_name(struct spi_mem *mem);
+
int spi_mem_driver_register_with_owner(struct spi_mem_driver *drv,
struct module *owner);
--
2.7.4
Frieder Schrempf
2018-07-05 11:15:02 UTC
Permalink
The FSL QSPI driver was moved to the SPI framework and it now
acts as a SPI controller. Therefore the subnodes need to set
spi-[rx/tx]-bus-width = <4>, so quad mode is used just as before.

Also the properties 'bus-num', 'fsl,spi-num-chipselects' and
'fsl,spi-flash-chipselects' were never read by the driver and
can be removed.

The 'reg' properties are adjusted to reflect the what bus and
chipselect the flash is connected to, as the new driver needs
this information.

The property 'fsl,qspi-has-second-chip' is not needed anymore
and will be removed after the old driver was disabled to avoid
breaking ls1021a-moxa-uc-8410a.dts.

Signed-off-by: Frieder Schrempf <***@exceet.de>
---
arch/arm/boot/dts/imx6sx-sdb-reva.dts | 8 ++++++--
arch/arm/boot/dts/imx6sx-sdb.dts | 8 ++++++--
arch/arm/boot/dts/imx6ul-14x14-evk.dtsi | 2 ++
arch/arm/boot/dts/ls1021a-moxa-uc-8410a.dts | 5 ++---
4 files changed, 16 insertions(+), 7 deletions(-)

diff --git a/arch/arm/boot/dts/imx6sx-sdb-reva.dts b/arch/arm/boot/dts/imx6sx-sdb-reva.dts
index e3533e7..1a6f680 100644
--- a/arch/arm/boot/dts/imx6sx-sdb-reva.dts
+++ b/arch/arm/boot/dts/imx6sx-sdb-reva.dts
@@ -131,13 +131,17 @@
#size-cells = <1>;
compatible = "spansion,s25fl128s", "jedec,spi-nor";
spi-max-frequency = <66000000>;
+ spi-rx-bus-width = <4>;
+ spi-tx-bus-width = <4>;
};

- flash1: ***@1 {
- reg = <1>;
+ flash1: ***@2 {
+ reg = <2>;
#address-cells = <1>;
#size-cells = <1>;
compatible = "spansion,s25fl128s", "jedec,spi-nor";
spi-max-frequency = <66000000>;
+ spi-rx-bus-width = <4>;
+ spi-tx-bus-width = <4>;
};
};
diff --git a/arch/arm/boot/dts/imx6sx-sdb.dts b/arch/arm/boot/dts/imx6sx-sdb.dts
index 6dd9beb..9acfda8 100644
--- a/arch/arm/boot/dts/imx6sx-sdb.dts
+++ b/arch/arm/boot/dts/imx6sx-sdb.dts
@@ -117,15 +117,19 @@
#size-cells = <1>;
compatible = "micron,n25q256a", "jedec,spi-nor";
spi-max-frequency = <29000000>;
+ spi-rx-bus-width = <4>;
+ spi-tx-bus-width = <4>;
reg = <0>;
};

- flash1: ***@1 {
+ flash1: ***@2 {
#address-cells = <1>;
#size-cells = <1>;
compatible = "micron,n25q256a", "jedec,spi-nor";
spi-max-frequency = <29000000>;
- reg = <1>;
+ spi-rx-bus-width = <4>;
+ spi-tx-bus-width = <4>;
+ reg = <2>;
};
};

diff --git a/arch/arm/boot/dts/imx6ul-14x14-evk.dtsi b/arch/arm/boot/dts/imx6ul-14x14-evk.dtsi
index 32a0723..c2c9a2a 100644
--- a/arch/arm/boot/dts/imx6ul-14x14-evk.dtsi
+++ b/arch/arm/boot/dts/imx6ul-14x14-evk.dtsi
@@ -176,6 +176,8 @@
#size-cells = <1>;
compatible = "micron,n25q256a";
spi-max-frequency = <29000000>;
+ spi-rx-bus-width = <4>;
+ spi-tx-bus-width = <4>;
reg = <0>;
};
};
diff --git a/arch/arm/boot/dts/ls1021a-moxa-uc-8410a.dts b/arch/arm/boot/dts/ls1021a-moxa-uc-8410a.dts
index d01f64b..6a83f30 100644
--- a/arch/arm/boot/dts/ls1021a-moxa-uc-8410a.dts
+++ b/arch/arm/boot/dts/ls1021a-moxa-uc-8410a.dts
@@ -203,9 +203,6 @@
};

&qspi {
- bus-num = <0>;
- fsl,spi-num-chipselects = <2>;
- fsl,spi-flash-chipselects = <0>;
fsl,qspi-has-second-chip;
status = "okay";

@@ -214,6 +211,8 @@
#address-cells = <1>;
#size-cells = <1>;
spi-max-frequency = <20000000>;
+ spi-rx-bus-width = <4>;
+ spi-tx-bus-width = <4>;
reg = <0>;

***@0 {
--
2.7.4
Frieder Schrempf
2018-07-05 11:15:01 UTC
Permalink
Adjust the documentation of the new SPI memory interface based
driver to reflect the new drivers settings.

Signed-off-by: Frieder Schrempf <***@exceet.de>
---
Changes in v2:
==============
* Split the moving and editing of the dt-bindings in two patches

.../devicetree/bindings/spi/spi-fsl-qspi.txt | 22 ++++++++++----------
1 file changed, 11 insertions(+), 11 deletions(-)

diff --git a/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt b/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
index 483e9cf..8b4eed7 100644
--- a/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
+++ b/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
@@ -3,9 +3,8 @@
Required properties:
- compatible : Should be "fsl,vf610-qspi", "fsl,imx6sx-qspi",
"fsl,imx7d-qspi", "fsl,imx6ul-qspi",
- "fsl,ls1021a-qspi"
+ "fsl,ls1021a-qspi", "fsl,ls2080a-qspi"
or
- "fsl,ls2080a-qspi" followed by "fsl,ls1021a-qspi",
"fsl,ls1043a-qspi" followed by "fsl,ls1021a-qspi"
- reg : the first contains the register location and length,
the second contains the memory mapping address and length
@@ -15,14 +14,15 @@ Required properties:
- clock-names : Should contain the name of the clocks: "qspi_en" and "qspi".

Optional properties:
- - fsl,qspi-has-second-chip: The controller has two buses, bus A and bus B.
- Each bus can be connected with two NOR flashes.
- Most of the time, each bus only has one NOR flash
- connected, this is the default case.
- But if there are two NOR flashes connected to the
- bus, you should enable this property.
- (Please check the board's schematic.)
- - big-endian : That means the IP register is big endian
+ - big-endian : That means the IP registers format is big endian
+
+Required SPI slave node properties:
+ - reg: There are two buses (A and B) with two chip selects each.
+ This encodes to which bus and CS the flash is connected:
+ <0>: Bus A, CS 0
+ <1>: Bus A, CS 1
+ <2>: Bus B, CS 0
+ <3>: Bus B, CS 1

Example:

@@ -40,7 +40,7 @@ qspi0: ***@40044000 {
};
};

-Example showing the usage of two SPI NOR devices:
+Example showing the usage of two SPI NOR devices on bus A:

&qspi2 {
pinctrl-names = "default";
--
2.7.4
Rob Herring
2018-07-11 16:05:21 UTC
Permalink
Post by Frieder Schrempf
Adjust the documentation of the new SPI memory interface based
driver to reflect the new drivers settings.
Bindings shouldn't change (other than new properties) due to driver
changes.
Post by Frieder Schrempf
---
==============
* Split the moving and editing of the dt-bindings in two patches
.../devicetree/bindings/spi/spi-fsl-qspi.txt | 22 ++++++++++----------
1 file changed, 11 insertions(+), 11 deletions(-)
diff --git a/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt b/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
index 483e9cf..8b4eed7 100644
--- a/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
+++ b/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
@@ -3,9 +3,8 @@
- compatible : Should be "fsl,vf610-qspi", "fsl,imx6sx-qspi",
"fsl,imx7d-qspi", "fsl,imx6ul-qspi",
- "fsl,ls1021a-qspi"
+ "fsl,ls1021a-qspi", "fsl,ls2080a-qspi"
or
- "fsl,ls2080a-qspi" followed by "fsl,ls1021a-qspi",
"fsl,ls1043a-qspi" followed by "fsl,ls1021a-qspi"
So the 2080a h/w was compatible with the 1021a h/w, but now it is not?
How did the h/w change?
Post by Frieder Schrempf
- reg : the first contains the register location and length,
the second contains the memory mapping address and length
- clock-names : Should contain the name of the clocks: "qspi_en" and "qspi".
- - fsl,qspi-has-second-chip: The controller has two buses, bus A and bus B.
- Each bus can be connected with two NOR flashes.
- Most of the time, each bus only has one NOR flash
- connected, this is the default case.
- But if there are two NOR flashes connected to the
- bus, you should enable this property.
- (Please check the board's schematic.)
You can't just remove properties without explanation. Why is this no
longer needed? What about backwards compatibility with existing dtbs?
Post by Frieder Schrempf
- - big-endian : That means the IP register is big endian
+ - big-endian : That means the IP registers format is big endian
This is a standard property so it doesn't really need to be redefined
here, but just reference the definition.
Post by Frieder Schrempf
+
+ - reg: There are two buses (A and B) with two chip selects each.
+ <0>: Bus A, CS 0
+ <1>: Bus A, CS 1
+ <2>: Bus B, CS 0
+ <3>: Bus B, CS 1
@@ -40,7 +40,7 @@ qspi0: ***@40044000 {
};
};
&qspi2 {
pinctrl-names = "default";
--
2.7.4
Frieder Schrempf
2018-07-12 08:13:57 UTC
Permalink
Hi Rob,
Post by Rob Herring
Post by Frieder Schrempf
Adjust the documentation of the new SPI memory interface based
driver to reflect the new drivers settings.
Bindings shouldn't change (other than new properties) due to driver
changes.
Right, I added an explanation below, why I think the changes are necessary.
Post by Rob Herring
Post by Frieder Schrempf
---
==============
* Split the moving and editing of the dt-bindings in two patches
.../devicetree/bindings/spi/spi-fsl-qspi.txt | 22 ++++++++++----------
1 file changed, 11 insertions(+), 11 deletions(-)
diff --git a/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt b/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
index 483e9cf..8b4eed7 100644
--- a/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
+++ b/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
@@ -3,9 +3,8 @@
- compatible : Should be "fsl,vf610-qspi", "fsl,imx6sx-qspi",
"fsl,imx7d-qspi", "fsl,imx6ul-qspi",
- "fsl,ls1021a-qspi"
+ "fsl,ls1021a-qspi", "fsl,ls2080a-qspi"
or
- "fsl,ls2080a-qspi" followed by "fsl,ls1021a-qspi",
"fsl,ls1043a-qspi" followed by "fsl,ls1021a-qspi"
So the 2080a h/w was compatible with the 1021a h/w, but now it is not?
How did the h/w change?
I guess this should be posted as a separate fix. Formerly there was only
"fsl,ls1021a-qspi" handled in the driver and the bindings here claimed
that "fsl,ls2080a-qspi" is compatible.

Some time ago a separate entry for "fsl,ls2080a-qspi" was added to the
driver [1] and it adds a quirk, that is not set for "fsl,ls1021a-qspi".
That's why I concluded, that these two are actually not compatible.
Post by Rob Herring
Post by Frieder Schrempf
- reg : the first contains the register location and length,
the second contains the memory mapping address and length
- clock-names : Should contain the name of the clocks: "qspi_en" and "qspi".
- - fsl,qspi-has-second-chip: The controller has two buses, bus A and bus B.
- Each bus can be connected with two NOR flashes.
- Most of the time, each bus only has one NOR flash
- connected, this is the default case.
- But if there are two NOR flashes connected to the
- bus, you should enable this property.
- (Please check the board's schematic.)
You can't just remove properties without explanation. Why is this no
longer needed? What about backwards compatibility with existing dtbs?
You're right, the explanation is missing here.

The "old" driver was using this property to select one of two dual chip
setups (two chips on one bus or two chips on separate buses). And it
used the order in which the subnodes are defined in the dt to select the
CS, the chip is connected to.

Both methods are wrong and in fact the "reg" property should be used to
determine which bus and CS a chip is connected to. This also enables us
to use different setups than just single chip, or symmetric dual chip.

So the porting of the driver from the MTD to the SPI framework actually
enforces the use of the "reg" properties and makes
"fsl,qspi-has-second-chip" superfluous.

As all boards that have "fsl,qspi-has-second-chip" set, also have
correct "reg" properties, the removal of this property shouldn't lead to
any incompatibilities.

The only compatibility issues I can see are with imx6sx-sdb.dts and
imx6sx-sdb-reva.dts, which have their reg properties set incorrectly
(see explanation here: [2]), all other boards should stay compatible.
Post by Rob Herring
Post by Frieder Schrempf
- - big-endian : That means the IP register is big endian
+ - big-endian : That means the IP registers format is big endian
This is a standard property so it doesn't really need to be redefined
here, but just reference the definition.
So I will change that to:

big-endian : See common-properties.txt for a definition

Thanks,
Frieder

[1]
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/drivers/mtd/spi-nor/fsl-quadspi.c?h=v4.18-rc4&id=d728a7ea9037c2df085bf9494d56e90d0ff69d7d
[2] https://patchwork.ozlabs.org/patch/922817/#1925445
Post by Rob Herring
Post by Frieder Schrempf
+
+ - reg: There are two buses (A and B) with two chip selects each.
+ <0>: Bus A, CS 0
+ <1>: Bus A, CS 1
+ <2>: Bus B, CS 0
+ <3>: Bus B, CS 1
@@ -40,7 +40,7 @@ qspi0: ***@40044000 {
};
};
&qspi2 {
pinctrl-names = "default";
--
2.7.4
Rob Herring
2018-07-12 15:20:45 UTC
Permalink
On Thu, Jul 12, 2018 at 2:14 AM Frieder Schrempf
Post by Frieder Schrempf
Hi Rob,
Post by Rob Herring
Post by Frieder Schrempf
Adjust the documentation of the new SPI memory interface based
driver to reflect the new drivers settings.
Bindings shouldn't change (other than new properties) due to driver
changes.
Right, I added an explanation below, why I think the changes are necessary.
Post by Rob Herring
Post by Frieder Schrempf
---
==============
* Split the moving and editing of the dt-bindings in two patches
.../devicetree/bindings/spi/spi-fsl-qspi.txt | 22 ++++++++++----------
1 file changed, 11 insertions(+), 11 deletions(-)
diff --git a/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt b/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
index 483e9cf..8b4eed7 100644
--- a/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
+++ b/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
@@ -3,9 +3,8 @@
- compatible : Should be "fsl,vf610-qspi", "fsl,imx6sx-qspi",
"fsl,imx7d-qspi", "fsl,imx6ul-qspi",
- "fsl,ls1021a-qspi"
+ "fsl,ls1021a-qspi", "fsl,ls2080a-qspi"
or
- "fsl,ls2080a-qspi" followed by "fsl,ls1021a-qspi",
"fsl,ls1043a-qspi" followed by "fsl,ls1021a-qspi"
So the 2080a h/w was compatible with the 1021a h/w, but now it is not?
How did the h/w change?
I guess this should be posted as a separate fix. Formerly there was only
"fsl,ls1021a-qspi" handled in the driver and the bindings here claimed
that "fsl,ls2080a-qspi" is compatible.
Some time ago a separate entry for "fsl,ls2080a-qspi" was added to the
driver [1] and it adds a quirk, that is not set for "fsl,ls1021a-qspi".
That's why I concluded, that these two are actually not compatible.
So before the driver change, the driver didn't work at all on the
ls2080a? If so, then removing is appropriate. If not, then removing
breaks all kernel versions before the change if you use a newer DT.
Post by Frieder Schrempf
Post by Rob Herring
Post by Frieder Schrempf
- reg : the first contains the register location and length,
the second contains the memory mapping address and length
- clock-names : Should contain the name of the clocks: "qspi_en" and "qspi".
- - fsl,qspi-has-second-chip: The controller has two buses, bus A and bus B.
- Each bus can be connected with two NOR flashes.
- Most of the time, each bus only has one NOR flash
- connected, this is the default case.
- But if there are two NOR flashes connected to the
- bus, you should enable this property.
- (Please check the board's schematic.)
You can't just remove properties without explanation. Why is this no
longer needed? What about backwards compatibility with existing dtbs?
You're right, the explanation is missing here.
The "old" driver was using this property to select one of two dual chip
setups (two chips on one bus or two chips on separate buses). And it
used the order in which the subnodes are defined in the dt to select the
CS, the chip is connected to.
Both methods are wrong and in fact the "reg" property should be used to
determine which bus and CS a chip is connected to. This also enables us
to use different setups than just single chip, or symmetric dual chip.
So the porting of the driver from the MTD to the SPI framework actually
enforces the use of the "reg" properties and makes
"fsl,qspi-has-second-chip" superfluous.
As all boards that have "fsl,qspi-has-second-chip" set, also have
correct "reg" properties, the removal of this property shouldn't lead to
any incompatibilities.
The only compatibility issues I can see are with imx6sx-sdb.dts and
imx6sx-sdb-reva.dts, which have their reg properties set incorrectly
(see explanation here: [2]), all other boards should stay compatible.
Add this to the commit msg.
Post by Frieder Schrempf
Post by Rob Herring
Post by Frieder Schrempf
- - big-endian : That means the IP register is big endian
+ - big-endian : That means the IP registers format is big endian
This is a standard property so it doesn't really need to be redefined
here, but just reference the definition.
big-endian : See common-properties.txt for a definition
You can drop "for a definition"

Rob
Frieder Schrempf
2018-07-16 07:04:12 UTC
Permalink
Hi Rob,
Post by Rob Herring
On Thu, Jul 12, 2018 at 2:14 AM Frieder Schrempf
Post by Frieder Schrempf
Hi Rob,
Post by Rob Herring
Post by Frieder Schrempf
Adjust the documentation of the new SPI memory interface based
driver to reflect the new drivers settings.
Bindings shouldn't change (other than new properties) due to driver
changes.
Right, I added an explanation below, why I think the changes are necessary.
Post by Rob Herring
Post by Frieder Schrempf
---
==============
* Split the moving and editing of the dt-bindings in two patches
.../devicetree/bindings/spi/spi-fsl-qspi.txt | 22 ++++++++++----------
1 file changed, 11 insertions(+), 11 deletions(-)
diff --git a/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt b/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
index 483e9cf..8b4eed7 100644
--- a/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
+++ b/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
@@ -3,9 +3,8 @@
- compatible : Should be "fsl,vf610-qspi", "fsl,imx6sx-qspi",
"fsl,imx7d-qspi", "fsl,imx6ul-qspi",
- "fsl,ls1021a-qspi"
+ "fsl,ls1021a-qspi", "fsl,ls2080a-qspi"
or
- "fsl,ls2080a-qspi" followed by "fsl,ls1021a-qspi",
"fsl,ls1043a-qspi" followed by "fsl,ls1021a-qspi"
So the 2080a h/w was compatible with the 1021a h/w, but now it is not?
How did the h/w change?
I guess this should be posted as a separate fix. Formerly there was only
"fsl,ls1021a-qspi" handled in the driver and the bindings here claimed
that "fsl,ls2080a-qspi" is compatible.
Some time ago a separate entry for "fsl,ls2080a-qspi" was added to the
driver [1] and it adds a quirk, that is not set for "fsl,ls1021a-qspi".
That's why I concluded, that these two are actually not compatible.
So before the driver change, the driver didn't work at all on the
ls2080a? If so, then removing is appropriate. If not, then removing
breaks all kernel versions before the change if you use a newer DT.
Before the driver change it couldn't work fully on the ls2080a, as the
quirk was missing. At least that's what I deduce from the driver.
Post by Rob Herring
Post by Frieder Schrempf
Post by Rob Herring
Post by Frieder Schrempf
- reg : the first contains the register location and length,
the second contains the memory mapping address and length
- clock-names : Should contain the name of the clocks: "qspi_en" and "qspi".
- - fsl,qspi-has-second-chip: The controller has two buses, bus A and bus B.
- Each bus can be connected with two NOR flashes.
- Most of the time, each bus only has one NOR flash
- connected, this is the default case.
- But if there are two NOR flashes connected to the
- bus, you should enable this property.
- (Please check the board's schematic.)
You can't just remove properties without explanation. Why is this no
longer needed? What about backwards compatibility with existing dtbs?
You're right, the explanation is missing here.
The "old" driver was using this property to select one of two dual chip
setups (two chips on one bus or two chips on separate buses). And it
used the order in which the subnodes are defined in the dt to select the
CS, the chip is connected to.
Both methods are wrong and in fact the "reg" property should be used to
determine which bus and CS a chip is connected to. This also enables us
to use different setups than just single chip, or symmetric dual chip.
So the porting of the driver from the MTD to the SPI framework actually
enforces the use of the "reg" properties and makes
"fsl,qspi-has-second-chip" superfluous.
As all boards that have "fsl,qspi-has-second-chip" set, also have
correct "reg" properties, the removal of this property shouldn't lead to
any incompatibilities.
The only compatibility issues I can see are with imx6sx-sdb.dts and
imx6sx-sdb-reva.dts, which have their reg properties set incorrectly
(see explanation here: [2]), all other boards should stay compatible.
Add this to the commit msg.
Ok.
Post by Rob Herring
Post by Frieder Schrempf
Post by Rob Herring
Post by Frieder Schrempf
- - big-endian : That means the IP register is big endian
+ - big-endian : That means the IP registers format is big endian
This is a standard property so it doesn't really need to be redefined
here, but just reference the definition.
big-endian : See common-properties.txt for a definition
You can drop "for a definition"
Ok.

Thanks,
Frieder
Frieder Schrempf
2018-07-05 11:15:07 UTC
Permalink
After switching to the new FSL QSPI driver the property
'fsl,qspi-has-second-chip' is not needed anymore.

The driver now uses the 'reg' property to determine the bus and
the chipselect.

Signed-off-by: Frieder Schrempf <***@exceet.de>
---
arch/arm64/boot/dts/freescale/fsl-ls1046a.dtsi | 1 -
1 file changed, 1 deletion(-)

diff --git a/arch/arm64/boot/dts/freescale/fsl-ls1046a.dtsi b/arch/arm64/boot/dts/freescale/fsl-ls1046a.dtsi
index 136ebfa..871189e 100644
--- a/arch/arm64/boot/dts/freescale/fsl-ls1046a.dtsi
+++ b/arch/arm64/boot/dts/freescale/fsl-ls1046a.dtsi
@@ -247,7 +247,6 @@
clock-names = "qspi_en", "qspi";
clocks = <&clockgen 4 1>, <&clockgen 4 1>;
big-endian;
- fsl,qspi-has-second-chip;
status = "disabled";
};
--
2.7.4
Frieder Schrempf
2018-07-05 11:15:08 UTC
Permalink
The driver was ported to the SPI framework so it can be used as
a generic SPI memory driver and not only for SPI NOR.
Reflect this transition in the MAINTAINERS file.

Signed-off-by: Frieder Schrempf <***@exceet.de>
---
MAINTAINERS | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/MAINTAINERS b/MAINTAINERS
index 63700ce..14039e0 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -5746,9 +5746,9 @@ F: Documentation/devicetree/bindings/ptp/ptp-qoriq.txt

FREESCALE QUAD SPI DRIVER
M: Han Xu <***@nxp.com>
-L: linux-***@lists.infradead.org
+L: linux-***@vger.kernel.org
S: Maintained
-F: drivers/mtd/spi-nor/fsl-quadspi.c
+F: drivers/spi/spi-fsl-qspi.c

FREESCALE QUICC ENGINE LIBRARY
M: Qiang Zhao <***@nxp.com>
--
2.7.4
Frieder Schrempf
2018-07-05 11:15:06 UTC
Permalink
After switching to the new FSL QSPI driver the property
'fsl,qspi-has-second-chip' is not needed anymore.

The driver now uses the 'reg' property to determine the bus and
the chipselect.

Signed-off-by: Frieder Schrempf <***@exceet.de>
---
arch/arm/boot/dts/ls1021a-moxa-uc-8410a.dts | 1 -
1 file changed, 1 deletion(-)

diff --git a/arch/arm/boot/dts/ls1021a-moxa-uc-8410a.dts b/arch/arm/boot/dts/ls1021a-moxa-uc-8410a.dts
index 6a83f30..d3a1a73 100644
--- a/arch/arm/boot/dts/ls1021a-moxa-uc-8410a.dts
+++ b/arch/arm/boot/dts/ls1021a-moxa-uc-8410a.dts
@@ -203,7 +203,6 @@
};

&qspi {
- fsl,qspi-has-second-chip;
status = "okay";

flash: ***@0 {
--
2.7.4
Frieder Schrempf
2018-07-05 11:14:58 UTC
Permalink
By calling spi_mem_get_name(), the driver of the (Q)SPI controller can
set a custom name for the memory device if necessary.
This is useful to keep mtdparts compatible when controller drivers are
ported from the MTD to the SPI layer.

Signed-off-by: Frieder Schrempf <***@exceet.de>
---
drivers/mtd/devices/m25p80.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/mtd/devices/m25p80.c b/drivers/mtd/devices/m25p80.c
index e84563d..482f0ef 100644
--- a/drivers/mtd/devices/m25p80.c
+++ b/drivers/mtd/devices/m25p80.c
@@ -199,8 +199,7 @@ static int m25p_probe(struct spi_mem *spimem)
hwcaps.mask |= SNOR_HWCAPS_READ_1_2_2;
}

- if (data && data->name)
- nor->mtd.name = data->name;
+ nor->mtd.name = spi_mem_get_name(spimem);

/* For some (historical?) reason many platforms provide two different
* names in flash_platform_data: "name" and "type". Quite often name is
--
2.7.4
Boris Brezillon
2018-07-05 12:56:28 UTC
Permalink
On Thu, 5 Jul 2018 13:14:58 +0200
Post by Frieder Schrempf
By calling spi_mem_get_name(), the driver of the (Q)SPI controller can
set a custom name for the memory device if necessary.
This is useful to keep mtdparts compatible when controller drivers are
ported from the MTD to the SPI layer.
---
drivers/mtd/devices/m25p80.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/drivers/mtd/devices/m25p80.c b/drivers/mtd/devices/m25p80.c
index e84563d..482f0ef 100644
--- a/drivers/mtd/devices/m25p80.c
+++ b/drivers/mtd/devices/m25p80.c
@@ -199,8 +199,7 @@ static int m25p_probe(struct spi_mem *spimem)
hwcaps.mask |= SNOR_HWCAPS_READ_1_2_2;
}
- if (data && data->name)
- nor->mtd.name = data->name;
+ nor->mtd.name = spi_mem_get_name(spimem);
Hm, not sure you can do that without breaking setup that pass a
flash_platform_data object with a name.

Also, you should keep the name extracted from the label property if any
(this is done earlier, when spi_nor_set_flash_node() is called).

if (data && data->name)
nor->mtd.name = data->name;

if (!nor->mtd.name)
nor->mtd.name = spi_mem_get_name(spimem);
Post by Frieder Schrempf
/* For some (historical?) reason many platforms provide two different
* names in flash_platform_data: "name" and "type". Quite often name is
Frieder Schrempf
2018-07-05 13:06:01 UTC
Permalink
Hi Boris,
Post by Boris Brezillon
On Thu, 5 Jul 2018 13:14:58 +0200
Post by Frieder Schrempf
By calling spi_mem_get_name(), the driver of the (Q)SPI controller can
set a custom name for the memory device if necessary.
This is useful to keep mtdparts compatible when controller drivers are
ported from the MTD to the SPI layer.
---
drivers/mtd/devices/m25p80.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/drivers/mtd/devices/m25p80.c b/drivers/mtd/devices/m25p80.c
index e84563d..482f0ef 100644
--- a/drivers/mtd/devices/m25p80.c
+++ b/drivers/mtd/devices/m25p80.c
@@ -199,8 +199,7 @@ static int m25p_probe(struct spi_mem *spimem)
hwcaps.mask |= SNOR_HWCAPS_READ_1_2_2;
}
- if (data && data->name)
- nor->mtd.name = data->name;
+ nor->mtd.name = spi_mem_get_name(spimem);
Hm, not sure you can do that without breaking setup that pass a
flash_platform_data object with a name.
Also, you should keep the name extracted from the label property if any
(this is done earlier, when spi_nor_set_flash_node() is called).
if (data && data->name)
nor->mtd.name = data->name;
if (!nor->mtd.name)
nor->mtd.name = spi_mem_get_name(spimem);
I missed both facts, that data could be passed with a name assigned
already and that spi_nor_set_flash_node() might assign a name from dt.
I should have looked at the code more closely.

Thanks for pointing that out. I'll fix it.

Frieder
Post by Boris Brezillon
Post by Frieder Schrempf
/* For some (historical?) reason many platforms provide two different
* names in flash_platform_data: "name" and "type". Quite often name is
Frieder Schrempf
2018-07-05 11:15:05 UTC
Permalink
There's a new driver using the SPI memory interface of the SPI framework
at spi/spi-fsl-qspi.c, which can be used together with m25p80.c to
replace the functionality of this SPI NOR driver.

Signed-off-by: Frieder Schrempf <***@exceet.de>
---
drivers/mtd/spi-nor/Kconfig | 9 -
drivers/mtd/spi-nor/Makefile | 1 -
drivers/mtd/spi-nor/fsl-quadspi.c | 1217 --------------------------------
3 files changed, 1227 deletions(-)

diff --git a/drivers/mtd/spi-nor/Kconfig b/drivers/mtd/spi-nor/Kconfig
index 6cc9c92..d1ca307 100644
--- a/drivers/mtd/spi-nor/Kconfig
+++ b/drivers/mtd/spi-nor/Kconfig
@@ -59,15 +59,6 @@ config SPI_CADENCE_QUADSPI
device with a Cadence QSPI controller and want to access the
Flash as an MTD device.

-config SPI_FSL_QUADSPI
- tristate "Freescale Quad SPI controller"
- depends on ARCH_MXC || SOC_LS1021A || ARCH_LAYERSCAPE || COMPILE_TEST
- depends on HAS_IOMEM
- help
- This enables support for the Quad SPI controller in master mode.
- This controller does not support generic SPI. It only supports
- SPI NOR.
-
config SPI_HISI_SFC
tristate "Hisilicon SPI-NOR Flash Controller(SFC)"
depends on ARCH_HISI || COMPILE_TEST
diff --git a/drivers/mtd/spi-nor/Makefile b/drivers/mtd/spi-nor/Makefile
index f4c61d2..3f160c2e3 100644
--- a/drivers/mtd/spi-nor/Makefile
+++ b/drivers/mtd/spi-nor/Makefile
@@ -3,7 +3,6 @@ obj-$(CONFIG_MTD_SPI_NOR) += spi-nor.o
obj-$(CONFIG_SPI_ASPEED_SMC) += aspeed-smc.o
obj-$(CONFIG_SPI_ATMEL_QUADSPI) += atmel-quadspi.o
obj-$(CONFIG_SPI_CADENCE_QUADSPI) += cadence-quadspi.o
-obj-$(CONFIG_SPI_FSL_QUADSPI) += fsl-quadspi.o
obj-$(CONFIG_SPI_HISI_SFC) += hisi-sfc.o
obj-$(CONFIG_MTD_MT81xx_NOR) += mtk-quadspi.o
obj-$(CONFIG_SPI_NXP_SPIFI) += nxp-spifi.o
diff --git a/drivers/mtd/spi-nor/fsl-quadspi.c b/drivers/mtd/spi-nor/fsl-quadspi.c
deleted file mode 100644
index 7d9620c..0000000
--- a/drivers/mtd/spi-nor/fsl-quadspi.c
+++ /dev/null
@@ -1,1217 +0,0 @@
-/*
- * Freescale QuadSPI driver.
- *
- * Copyright (C) 2013 Freescale Semiconductor, Inc.
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 of the License, or
- * (at your option) any later version.
- */
-#include <linux/kernel.h>
-#include <linux/module.h>
-#include <linux/interrupt.h>
-#include <linux/errno.h>
-#include <linux/platform_device.h>
-#include <linux/sched.h>
-#include <linux/delay.h>
-#include <linux/io.h>
-#include <linux/clk.h>
-#include <linux/err.h>
-#include <linux/of.h>
-#include <linux/of_device.h>
-#include <linux/timer.h>
-#include <linux/jiffies.h>
-#include <linux/completion.h>
-#include <linux/mtd/mtd.h>
-#include <linux/mtd/partitions.h>
-#include <linux/mtd/spi-nor.h>
-#include <linux/mutex.h>
-#include <linux/pm_qos.h>
-#include <linux/sizes.h>
-
-/* Controller needs driver to swap endian */
-#define QUADSPI_QUIRK_SWAP_ENDIAN (1 << 0)
-/* Controller needs 4x internal clock */
-#define QUADSPI_QUIRK_4X_INT_CLK (1 << 1)
-/*
- * TKT253890, Controller needs driver to fill txfifo till 16 byte to
- * trigger data transfer even though extern data will not transferred.
- */
-#define QUADSPI_QUIRK_TKT253890 (1 << 2)
-/* Controller cannot wake up from wait mode, TKT245618 */
-#define QUADSPI_QUIRK_TKT245618 (1 << 3)
-
-/* The registers */
-#define QUADSPI_MCR 0x00
-#define QUADSPI_MCR_RESERVED_SHIFT 16
-#define QUADSPI_MCR_RESERVED_MASK (0xF << QUADSPI_MCR_RESERVED_SHIFT)
-#define QUADSPI_MCR_MDIS_SHIFT 14
-#define QUADSPI_MCR_MDIS_MASK (1 << QUADSPI_MCR_MDIS_SHIFT)
-#define QUADSPI_MCR_CLR_TXF_SHIFT 11
-#define QUADSPI_MCR_CLR_TXF_MASK (1 << QUADSPI_MCR_CLR_TXF_SHIFT)
-#define QUADSPI_MCR_CLR_RXF_SHIFT 10
-#define QUADSPI_MCR_CLR_RXF_MASK (1 << QUADSPI_MCR_CLR_RXF_SHIFT)
-#define QUADSPI_MCR_DDR_EN_SHIFT 7
-#define QUADSPI_MCR_DDR_EN_MASK (1 << QUADSPI_MCR_DDR_EN_SHIFT)
-#define QUADSPI_MCR_END_CFG_SHIFT 2
-#define QUADSPI_MCR_END_CFG_MASK (3 << QUADSPI_MCR_END_CFG_SHIFT)
-#define QUADSPI_MCR_SWRSTHD_SHIFT 1
-#define QUADSPI_MCR_SWRSTHD_MASK (1 << QUADSPI_MCR_SWRSTHD_SHIFT)
-#define QUADSPI_MCR_SWRSTSD_SHIFT 0
-#define QUADSPI_MCR_SWRSTSD_MASK (1 << QUADSPI_MCR_SWRSTSD_SHIFT)
-
-#define QUADSPI_IPCR 0x08
-#define QUADSPI_IPCR_SEQID_SHIFT 24
-#define QUADSPI_IPCR_SEQID_MASK (0xF << QUADSPI_IPCR_SEQID_SHIFT)
-
-#define QUADSPI_BUF0CR 0x10
-#define QUADSPI_BUF1CR 0x14
-#define QUADSPI_BUF2CR 0x18
-#define QUADSPI_BUFXCR_INVALID_MSTRID 0xe
-
-#define QUADSPI_BUF3CR 0x1c
-#define QUADSPI_BUF3CR_ALLMST_SHIFT 31
-#define QUADSPI_BUF3CR_ALLMST_MASK (1 << QUADSPI_BUF3CR_ALLMST_SHIFT)
-#define QUADSPI_BUF3CR_ADATSZ_SHIFT 8
-#define QUADSPI_BUF3CR_ADATSZ_MASK (0xFF << QUADSPI_BUF3CR_ADATSZ_SHIFT)
-
-#define QUADSPI_BFGENCR 0x20
-#define QUADSPI_BFGENCR_PAR_EN_SHIFT 16
-#define QUADSPI_BFGENCR_PAR_EN_MASK (1 << (QUADSPI_BFGENCR_PAR_EN_SHIFT))
-#define QUADSPI_BFGENCR_SEQID_SHIFT 12
-#define QUADSPI_BFGENCR_SEQID_MASK (0xF << QUADSPI_BFGENCR_SEQID_SHIFT)
-
-#define QUADSPI_BUF0IND 0x30
-#define QUADSPI_BUF1IND 0x34
-#define QUADSPI_BUF2IND 0x38
-#define QUADSPI_SFAR 0x100
-
-#define QUADSPI_SMPR 0x108
-#define QUADSPI_SMPR_DDRSMP_SHIFT 16
-#define QUADSPI_SMPR_DDRSMP_MASK (7 << QUADSPI_SMPR_DDRSMP_SHIFT)
-#define QUADSPI_SMPR_FSDLY_SHIFT 6
-#define QUADSPI_SMPR_FSDLY_MASK (1 << QUADSPI_SMPR_FSDLY_SHIFT)
-#define QUADSPI_SMPR_FSPHS_SHIFT 5
-#define QUADSPI_SMPR_FSPHS_MASK (1 << QUADSPI_SMPR_FSPHS_SHIFT)
-#define QUADSPI_SMPR_HSENA_SHIFT 0
-#define QUADSPI_SMPR_HSENA_MASK (1 << QUADSPI_SMPR_HSENA_SHIFT)
-
-#define QUADSPI_RBSR 0x10c
-#define QUADSPI_RBSR_RDBFL_SHIFT 8
-#define QUADSPI_RBSR_RDBFL_MASK (0x3F << QUADSPI_RBSR_RDBFL_SHIFT)
-
-#define QUADSPI_RBCT 0x110
-#define QUADSPI_RBCT_WMRK_MASK 0x1F
-#define QUADSPI_RBCT_RXBRD_SHIFT 8
-#define QUADSPI_RBCT_RXBRD_USEIPS (0x1 << QUADSPI_RBCT_RXBRD_SHIFT)
-
-#define QUADSPI_TBSR 0x150
-#define QUADSPI_TBDR 0x154
-#define QUADSPI_SR 0x15c
-#define QUADSPI_SR_IP_ACC_SHIFT 1
-#define QUADSPI_SR_IP_ACC_MASK (0x1 << QUADSPI_SR_IP_ACC_SHIFT)
-#define QUADSPI_SR_AHB_ACC_SHIFT 2
-#define QUADSPI_SR_AHB_ACC_MASK (0x1 << QUADSPI_SR_AHB_ACC_SHIFT)
-
-#define QUADSPI_FR 0x160
-#define QUADSPI_FR_TFF_MASK 0x1
-
-#define QUADSPI_SFA1AD 0x180
-#define QUADSPI_SFA2AD 0x184
-#define QUADSPI_SFB1AD 0x188
-#define QUADSPI_SFB2AD 0x18c
-#define QUADSPI_RBDR 0x200
-
-#define QUADSPI_LUTKEY 0x300
-#define QUADSPI_LUTKEY_VALUE 0x5AF05AF0
-
-#define QUADSPI_LCKCR 0x304
-#define QUADSPI_LCKER_LOCK 0x1
-#define QUADSPI_LCKER_UNLOCK 0x2
-
-#define QUADSPI_RSER 0x164
-#define QUADSPI_RSER_TFIE (0x1 << 0)
-
-#define QUADSPI_LUT_BASE 0x310
-
-/*
- * The definition of the LUT register shows below:
- *
- * ---------------------------------------------------
- * | INSTR1 | PAD1 | OPRND1 | INSTR0 | PAD0 | OPRND0 |
- * ---------------------------------------------------
- */
-#define OPRND0_SHIFT 0
-#define PAD0_SHIFT 8
-#define INSTR0_SHIFT 10
-#define OPRND1_SHIFT 16
-
-/* Instruction set for the LUT register. */
-#define LUT_STOP 0
-#define LUT_CMD 1
-#define LUT_ADDR 2
-#define LUT_DUMMY 3
-#define LUT_MODE 4
-#define LUT_MODE2 5
-#define LUT_MODE4 6
-#define LUT_FSL_READ 7
-#define LUT_FSL_WRITE 8
-#define LUT_JMP_ON_CS 9
-#define LUT_ADDR_DDR 10
-#define LUT_MODE_DDR 11
-#define LUT_MODE2_DDR 12
-#define LUT_MODE4_DDR 13
-#define LUT_FSL_READ_DDR 14
-#define LUT_FSL_WRITE_DDR 15
-#define LUT_DATA_LEARN 16
-
-/*
- * The PAD definitions for LUT register.
- *
- * The pad stands for the lines number of IO[0:3].
- * For example, the Quad read need four IO lines, so you should
- * set LUT_PAD4 which means we use four IO lines.
- */
-#define LUT_PAD1 0
-#define LUT_PAD2 1
-#define LUT_PAD4 2
-
-/* Oprands for the LUT register. */
-#define ADDR24BIT 0x18
-#define ADDR32BIT 0x20
-
-/* Macros for constructing the LUT register. */
-#define LUT0(ins, pad, opr) \
- (((opr) << OPRND0_SHIFT) | ((LUT_##pad) << PAD0_SHIFT) | \
- ((LUT_##ins) << INSTR0_SHIFT))
-
-#define LUT1(ins, pad, opr) (LUT0(ins, pad, opr) << OPRND1_SHIFT)
-
-/* other macros for LUT register. */
-#define QUADSPI_LUT(x) (QUADSPI_LUT_BASE + (x) * 4)
-#define QUADSPI_LUT_NUM 64
-
-/* SEQID -- we can have 16 seqids at most. */
-#define SEQID_READ 0
-#define SEQID_WREN 1
-#define SEQID_WRDI 2
-#define SEQID_RDSR 3
-#define SEQID_SE 4
-#define SEQID_CHIP_ERASE 5
-#define SEQID_PP 6
-#define SEQID_RDID 7
-#define SEQID_WRSR 8
-#define SEQID_RDCR 9
-#define SEQID_EN4B 10
-#define SEQID_BRWR 11
-
-#define QUADSPI_MIN_IOMAP SZ_4M
-
-enum fsl_qspi_devtype {
- FSL_QUADSPI_VYBRID,
- FSL_QUADSPI_IMX6SX,
- FSL_QUADSPI_IMX7D,
- FSL_QUADSPI_IMX6UL,
- FSL_QUADSPI_LS1021A,
- FSL_QUADSPI_LS2080A,
-};
-
-struct fsl_qspi_devtype_data {
- enum fsl_qspi_devtype devtype;
- int rxfifo;
- int txfifo;
- int ahb_buf_size;
- int driver_data;
-};
-
-static const struct fsl_qspi_devtype_data vybrid_data = {
- .devtype = FSL_QUADSPI_VYBRID,
- .rxfifo = 128,
- .txfifo = 64,
- .ahb_buf_size = 1024,
- .driver_data = QUADSPI_QUIRK_SWAP_ENDIAN,
-};
-
-static const struct fsl_qspi_devtype_data imx6sx_data = {
- .devtype = FSL_QUADSPI_IMX6SX,
- .rxfifo = 128,
- .txfifo = 512,
- .ahb_buf_size = 1024,
- .driver_data = QUADSPI_QUIRK_4X_INT_CLK
- | QUADSPI_QUIRK_TKT245618,
-};
-
-static const struct fsl_qspi_devtype_data imx7d_data = {
- .devtype = FSL_QUADSPI_IMX7D,
- .rxfifo = 512,
- .txfifo = 512,
- .ahb_buf_size = 1024,
- .driver_data = QUADSPI_QUIRK_TKT253890
- | QUADSPI_QUIRK_4X_INT_CLK,
-};
-
-static const struct fsl_qspi_devtype_data imx6ul_data = {
- .devtype = FSL_QUADSPI_IMX6UL,
- .rxfifo = 128,
- .txfifo = 512,
- .ahb_buf_size = 1024,
- .driver_data = QUADSPI_QUIRK_TKT253890
- | QUADSPI_QUIRK_4X_INT_CLK,
-};
-
-static struct fsl_qspi_devtype_data ls1021a_data = {
- .devtype = FSL_QUADSPI_LS1021A,
- .rxfifo = 128,
- .txfifo = 64,
- .ahb_buf_size = 1024,
- .driver_data = 0,
-};
-
-static const struct fsl_qspi_devtype_data ls2080a_data = {
- .devtype = FSL_QUADSPI_LS2080A,
- .rxfifo = 128,
- .txfifo = 64,
- .ahb_buf_size = 1024,
- .driver_data = QUADSPI_QUIRK_TKT253890,
-};
-
-
-#define FSL_QSPI_MAX_CHIP 4
-struct fsl_qspi {
- struct spi_nor nor[FSL_QSPI_MAX_CHIP];
- void __iomem *iobase;
- void __iomem *ahb_addr;
- u32 memmap_phy;
- u32 memmap_offs;
- u32 memmap_len;
- struct clk *clk, *clk_en;
- struct device *dev;
- struct completion c;
- const struct fsl_qspi_devtype_data *devtype_data;
- u32 nor_size;
- u32 nor_num;
- u32 clk_rate;
- unsigned int chip_base_addr; /* We may support two chips. */
- bool has_second_chip;
- bool big_endian;
- struct mutex lock;
- struct pm_qos_request pm_qos_req;
-};
-
-static inline int needs_swap_endian(struct fsl_qspi *q)
-{
- return q->devtype_data->driver_data & QUADSPI_QUIRK_SWAP_ENDIAN;
-}
-
-static inline int needs_4x_clock(struct fsl_qspi *q)
-{
- return q->devtype_data->driver_data & QUADSPI_QUIRK_4X_INT_CLK;
-}
-
-static inline int needs_fill_txfifo(struct fsl_qspi *q)
-{
- return q->devtype_data->driver_data & QUADSPI_QUIRK_TKT253890;
-}
-
-static inline int needs_wakeup_wait_mode(struct fsl_qspi *q)
-{
- return q->devtype_data->driver_data & QUADSPI_QUIRK_TKT245618;
-}
-
-/*
- * R/W functions for big- or little-endian registers:
- * The qSPI controller's endian is independent of the CPU core's endian.
- * So far, although the CPU core is little-endian but the qSPI have two
- * versions for big-endian and little-endian.
- */
-static void qspi_writel(struct fsl_qspi *q, u32 val, void __iomem *addr)
-{
- if (q->big_endian)
- iowrite32be(val, addr);
- else
- iowrite32(val, addr);
-}
-
-static u32 qspi_readl(struct fsl_qspi *q, void __iomem *addr)
-{
- if (q->big_endian)
- return ioread32be(addr);
- else
- return ioread32(addr);
-}
-
-/*
- * An IC bug makes us to re-arrange the 32-bit data.
- * The following chips, such as IMX6SLX, have fixed this bug.
- */
-static inline u32 fsl_qspi_endian_xchg(struct fsl_qspi *q, u32 a)
-{
- return needs_swap_endian(q) ? __swab32(a) : a;
-}
-
-static inline void fsl_qspi_unlock_lut(struct fsl_qspi *q)
-{
- qspi_writel(q, QUADSPI_LUTKEY_VALUE, q->iobase + QUADSPI_LUTKEY);
- qspi_writel(q, QUADSPI_LCKER_UNLOCK, q->iobase + QUADSPI_LCKCR);
-}
-
-static inline void fsl_qspi_lock_lut(struct fsl_qspi *q)
-{
- qspi_writel(q, QUADSPI_LUTKEY_VALUE, q->iobase + QUADSPI_LUTKEY);
- qspi_writel(q, QUADSPI_LCKER_LOCK, q->iobase + QUADSPI_LCKCR);
-}
-
-static irqreturn_t fsl_qspi_irq_handler(int irq, void *dev_id)
-{
- struct fsl_qspi *q = dev_id;
- u32 reg;
-
- /* clear interrupt */
- reg = qspi_readl(q, q->iobase + QUADSPI_FR);
- qspi_writel(q, reg, q->iobase + QUADSPI_FR);
-
- if (reg & QUADSPI_FR_TFF_MASK)
- complete(&q->c);
-
- dev_dbg(q->dev, "QUADSPI_FR : 0x%.8x:0x%.8x\n", q->chip_base_addr, reg);
- return IRQ_HANDLED;
-}
-
-static void fsl_qspi_init_lut(struct fsl_qspi *q)
-{
- void __iomem *base = q->iobase;
- int rxfifo = q->devtype_data->rxfifo;
- u32 lut_base;
- int i;
-
- struct spi_nor *nor = &q->nor[0];
- u8 addrlen = (nor->addr_width == 3) ? ADDR24BIT : ADDR32BIT;
- u8 read_op = nor->read_opcode;
- u8 read_dm = nor->read_dummy;
-
- fsl_qspi_unlock_lut(q);
-
- /* Clear all the LUT table */
- for (i = 0; i < QUADSPI_LUT_NUM; i++)
- qspi_writel(q, 0, base + QUADSPI_LUT_BASE + i * 4);
-
- /* Read */
- lut_base = SEQID_READ * 4;
-
- qspi_writel(q, LUT0(CMD, PAD1, read_op) | LUT1(ADDR, PAD1, addrlen),
- base + QUADSPI_LUT(lut_base));
- qspi_writel(q, LUT0(DUMMY, PAD1, read_dm) |
- LUT1(FSL_READ, PAD4, rxfifo),
- base + QUADSPI_LUT(lut_base + 1));
-
- /* Write enable */
- lut_base = SEQID_WREN * 4;
- qspi_writel(q, LUT0(CMD, PAD1, SPINOR_OP_WREN),
- base + QUADSPI_LUT(lut_base));
-
- /* Page Program */
- lut_base = SEQID_PP * 4;
-
- qspi_writel(q, LUT0(CMD, PAD1, nor->program_opcode) |
- LUT1(ADDR, PAD1, addrlen),
- base + QUADSPI_LUT(lut_base));
- qspi_writel(q, LUT0(FSL_WRITE, PAD1, 0),
- base + QUADSPI_LUT(lut_base + 1));
-
- /* Read Status */
- lut_base = SEQID_RDSR * 4;
- qspi_writel(q, LUT0(CMD, PAD1, SPINOR_OP_RDSR) |
- LUT1(FSL_READ, PAD1, 0x1),
- base + QUADSPI_LUT(lut_base));
-
- /* Erase a sector */
- lut_base = SEQID_SE * 4;
-
- qspi_writel(q, LUT0(CMD, PAD1, nor->erase_opcode) |
- LUT1(ADDR, PAD1, addrlen),
- base + QUADSPI_LUT(lut_base));
-
- /* Erase the whole chip */
- lut_base = SEQID_CHIP_ERASE * 4;
- qspi_writel(q, LUT0(CMD, PAD1, SPINOR_OP_CHIP_ERASE),
- base + QUADSPI_LUT(lut_base));
-
- /* READ ID */
- lut_base = SEQID_RDID * 4;
- qspi_writel(q, LUT0(CMD, PAD1, SPINOR_OP_RDID) |
- LUT1(FSL_READ, PAD1, 0x8),
- base + QUADSPI_LUT(lut_base));
-
- /* Write Register */
- lut_base = SEQID_WRSR * 4;
- qspi_writel(q, LUT0(CMD, PAD1, SPINOR_OP_WRSR) |
- LUT1(FSL_WRITE, PAD1, 0x2),
- base + QUADSPI_LUT(lut_base));
-
- /* Read Configuration Register */
- lut_base = SEQID_RDCR * 4;
- qspi_writel(q, LUT0(CMD, PAD1, SPINOR_OP_RDCR) |
- LUT1(FSL_READ, PAD1, 0x1),
- base + QUADSPI_LUT(lut_base));
-
- /* Write disable */
- lut_base = SEQID_WRDI * 4;
- qspi_writel(q, LUT0(CMD, PAD1, SPINOR_OP_WRDI),
- base + QUADSPI_LUT(lut_base));
-
- /* Enter 4 Byte Mode (Micron) */
- lut_base = SEQID_EN4B * 4;
- qspi_writel(q, LUT0(CMD, PAD1, SPINOR_OP_EN4B),
- base + QUADSPI_LUT(lut_base));
-
- /* Enter 4 Byte Mode (Spansion) */
- lut_base = SEQID_BRWR * 4;
- qspi_writel(q, LUT0(CMD, PAD1, SPINOR_OP_BRWR),
- base + QUADSPI_LUT(lut_base));
-
- fsl_qspi_lock_lut(q);
-}
-
-/* Get the SEQID for the command */
-static int fsl_qspi_get_seqid(struct fsl_qspi *q, u8 cmd)
-{
- switch (cmd) {
- case SPINOR_OP_READ_1_1_4:
- return SEQID_READ;
- case SPINOR_OP_WREN:
- return SEQID_WREN;
- case SPINOR_OP_WRDI:
- return SEQID_WRDI;
- case SPINOR_OP_RDSR:
- return SEQID_RDSR;
- case SPINOR_OP_SE:
- return SEQID_SE;
- case SPINOR_OP_CHIP_ERASE:
- return SEQID_CHIP_ERASE;
- case SPINOR_OP_PP:
- return SEQID_PP;
- case SPINOR_OP_RDID:
- return SEQID_RDID;
- case SPINOR_OP_WRSR:
- return SEQID_WRSR;
- case SPINOR_OP_RDCR:
- return SEQID_RDCR;
- case SPINOR_OP_EN4B:
- return SEQID_EN4B;
- case SPINOR_OP_BRWR:
- return SEQID_BRWR;
- default:
- if (cmd == q->nor[0].erase_opcode)
- return SEQID_SE;
- dev_err(q->dev, "Unsupported cmd 0x%.2x\n", cmd);
- break;
- }
- return -EINVAL;
-}
-
-static int
-fsl_qspi_runcmd(struct fsl_qspi *q, u8 cmd, unsigned int addr, int len)
-{
- void __iomem *base = q->iobase;
- int seqid;
- u32 reg, reg2;
- int err;
-
- init_completion(&q->c);
- dev_dbg(q->dev, "to 0x%.8x:0x%.8x, len:%d, cmd:%.2x\n",
- q->chip_base_addr, addr, len, cmd);
-
- /* save the reg */
- reg = qspi_readl(q, base + QUADSPI_MCR);
-
- qspi_writel(q, q->memmap_phy + q->chip_base_addr + addr,
- base + QUADSPI_SFAR);
- qspi_writel(q, QUADSPI_RBCT_WMRK_MASK | QUADSPI_RBCT_RXBRD_USEIPS,
- base + QUADSPI_RBCT);
- qspi_writel(q, reg | QUADSPI_MCR_CLR_RXF_MASK, base + QUADSPI_MCR);
-
- do {
- reg2 = qspi_readl(q, base + QUADSPI_SR);
- if (reg2 & (QUADSPI_SR_IP_ACC_MASK | QUADSPI_SR_AHB_ACC_MASK)) {
- udelay(1);
- dev_dbg(q->dev, "The controller is busy, 0x%x\n", reg2);
- continue;
- }
- break;
- } while (1);
-
- /* trigger the LUT now */
- seqid = fsl_qspi_get_seqid(q, cmd);
- qspi_writel(q, (seqid << QUADSPI_IPCR_SEQID_SHIFT) | len,
- base + QUADSPI_IPCR);
-
- /* Wait for the interrupt. */
- if (!wait_for_completion_timeout(&q->c, msecs_to_jiffies(1000))) {
- dev_err(q->dev,
- "cmd 0x%.2x timeout, addr@%.8x, FR:0x%.8x, SR:0x%.8x\n",
- cmd, addr, qspi_readl(q, base + QUADSPI_FR),
- qspi_readl(q, base + QUADSPI_SR));
- err = -ETIMEDOUT;
- } else {
- err = 0;
- }
-
- /* restore the MCR */
- qspi_writel(q, reg, base + QUADSPI_MCR);
-
- return err;
-}
-
-/* Read out the data from the QUADSPI_RBDR buffer registers. */
-static void fsl_qspi_read_data(struct fsl_qspi *q, int len, u8 *rxbuf)
-{
- u32 tmp;
- int i = 0;
-
- while (len > 0) {
- tmp = qspi_readl(q, q->iobase + QUADSPI_RBDR + i * 4);
- tmp = fsl_qspi_endian_xchg(q, tmp);
- dev_dbg(q->dev, "chip addr:0x%.8x, rcv:0x%.8x\n",
- q->chip_base_addr, tmp);
-
- if (len >= 4) {
- *((u32 *)rxbuf) = tmp;
- rxbuf += 4;
- } else {
- memcpy(rxbuf, &tmp, len);
- break;
- }
-
- len -= 4;
- i++;
- }
-}
-
-/*
- * If we have changed the content of the flash by writing or erasing,
- * we need to invalidate the AHB buffer. If we do not do so, we may read out
- * the wrong data. The spec tells us reset the AHB domain and Serial Flash
- * domain at the same time.
- */
-static inline void fsl_qspi_invalid(struct fsl_qspi *q)
-{
- u32 reg;
-
- reg = qspi_readl(q, q->iobase + QUADSPI_MCR);
- reg |= QUADSPI_MCR_SWRSTHD_MASK | QUADSPI_MCR_SWRSTSD_MASK;
- qspi_writel(q, reg, q->iobase + QUADSPI_MCR);
-
- /*
- * The minimum delay : 1 AHB + 2 SFCK clocks.
- * Delay 1 us is enough.
- */
- udelay(1);
-
- reg &= ~(QUADSPI_MCR_SWRSTHD_MASK | QUADSPI_MCR_SWRSTSD_MASK);
- qspi_writel(q, reg, q->iobase + QUADSPI_MCR);
-}
-
-static ssize_t fsl_qspi_nor_write(struct fsl_qspi *q, struct spi_nor *nor,
- u8 opcode, unsigned int to, u32 *txbuf,
- unsigned count)
-{
- int ret, i, j;
- u32 tmp;
-
- dev_dbg(q->dev, "to 0x%.8x:0x%.8x, len : %d\n",
- q->chip_base_addr, to, count);
-
- /* clear the TX FIFO. */
- tmp = qspi_readl(q, q->iobase + QUADSPI_MCR);
- qspi_writel(q, tmp | QUADSPI_MCR_CLR_TXF_MASK, q->iobase + QUADSPI_MCR);
-
- /* fill the TX data to the FIFO */
- for (j = 0, i = ((count + 3) / 4); j < i; j++) {
- tmp = fsl_qspi_endian_xchg(q, *txbuf);
- qspi_writel(q, tmp, q->iobase + QUADSPI_TBDR);
- txbuf++;
- }
-
- /* fill the TXFIFO upto 16 bytes for i.MX7d */
- if (needs_fill_txfifo(q))
- for (; i < 4; i++)
- qspi_writel(q, tmp, q->iobase + QUADSPI_TBDR);
-
- /* Trigger it */
- ret = fsl_qspi_runcmd(q, opcode, to, count);
-
- if (ret == 0)
- return count;
-
- return ret;
-}
-
-static void fsl_qspi_set_map_addr(struct fsl_qspi *q)
-{
- int nor_size = q->nor_size;
- void __iomem *base = q->iobase;
-
- qspi_writel(q, nor_size + q->memmap_phy, base + QUADSPI_SFA1AD);
- qspi_writel(q, nor_size * 2 + q->memmap_phy, base + QUADSPI_SFA2AD);
- qspi_writel(q, nor_size * 3 + q->memmap_phy, base + QUADSPI_SFB1AD);
- qspi_writel(q, nor_size * 4 + q->memmap_phy, base + QUADSPI_SFB2AD);
-}
-
-/*
- * There are two different ways to read out the data from the flash:
- * the "IP Command Read" and the "AHB Command Read".
- *
- * The IC guy suggests we use the "AHB Command Read" which is faster
- * then the "IP Command Read". (What's more is that there is a bug in
- * the "IP Command Read" in the Vybrid.)
- *
- * After we set up the registers for the "AHB Command Read", we can use
- * the memcpy to read the data directly. A "missed" access to the buffer
- * causes the controller to clear the buffer, and use the sequence pointed
- * by the QUADSPI_BFGENCR[SEQID] to initiate a read from the flash.
- */
-static void fsl_qspi_init_ahb_read(struct fsl_qspi *q)
-{
- void __iomem *base = q->iobase;
- int seqid;
-
- /* AHB configuration for access buffer 0/1/2 .*/
- qspi_writel(q, QUADSPI_BUFXCR_INVALID_MSTRID, base + QUADSPI_BUF0CR);
- qspi_writel(q, QUADSPI_BUFXCR_INVALID_MSTRID, base + QUADSPI_BUF1CR);
- qspi_writel(q, QUADSPI_BUFXCR_INVALID_MSTRID, base + QUADSPI_BUF2CR);
- /*
- * Set ADATSZ with the maximum AHB buffer size to improve the
- * read performance.
- */
- qspi_writel(q, QUADSPI_BUF3CR_ALLMST_MASK |
- ((q->devtype_data->ahb_buf_size / 8)
- << QUADSPI_BUF3CR_ADATSZ_SHIFT),
- base + QUADSPI_BUF3CR);
-
- /* We only use the buffer3 */
- qspi_writel(q, 0, base + QUADSPI_BUF0IND);
- qspi_writel(q, 0, base + QUADSPI_BUF1IND);
- qspi_writel(q, 0, base + QUADSPI_BUF2IND);
-
- /* Set the default lut sequence for AHB Read. */
- seqid = fsl_qspi_get_seqid(q, q->nor[0].read_opcode);
- qspi_writel(q, seqid << QUADSPI_BFGENCR_SEQID_SHIFT,
- q->iobase + QUADSPI_BFGENCR);
-}
-
-/* This function was used to prepare and enable QSPI clock */
-static int fsl_qspi_clk_prep_enable(struct fsl_qspi *q)
-{
- int ret;
-
- ret = clk_prepare_enable(q->clk_en);
- if (ret)
- return ret;
-
- ret = clk_prepare_enable(q->clk);
- if (ret) {
- clk_disable_unprepare(q->clk_en);
- return ret;
- }
-
- if (needs_wakeup_wait_mode(q))
- pm_qos_add_request(&q->pm_qos_req, PM_QOS_CPU_DMA_LATENCY, 0);
-
- return 0;
-}
-
-/* This function was used to disable and unprepare QSPI clock */
-static void fsl_qspi_clk_disable_unprep(struct fsl_qspi *q)
-{
- if (needs_wakeup_wait_mode(q))
- pm_qos_remove_request(&q->pm_qos_req);
-
- clk_disable_unprepare(q->clk);
- clk_disable_unprepare(q->clk_en);
-
-}
-
-/* We use this function to do some basic init for spi_nor_scan(). */
-static int fsl_qspi_nor_setup(struct fsl_qspi *q)
-{
- void __iomem *base = q->iobase;
- u32 reg;
- int ret;
-
- /* disable and unprepare clock to avoid glitch pass to controller */
- fsl_qspi_clk_disable_unprep(q);
-
- /* the default frequency, we will change it in the future. */
- ret = clk_set_rate(q->clk, 66000000);
- if (ret)
- return ret;
-
- ret = fsl_qspi_clk_prep_enable(q);
- if (ret)
- return ret;
-
- /* Reset the module */
- qspi_writel(q, QUADSPI_MCR_SWRSTSD_MASK | QUADSPI_MCR_SWRSTHD_MASK,
- base + QUADSPI_MCR);
- udelay(1);
-
- /* Init the LUT table. */
- fsl_qspi_init_lut(q);
-
- /* Disable the module */
- qspi_writel(q, QUADSPI_MCR_MDIS_MASK | QUADSPI_MCR_RESERVED_MASK,
- base + QUADSPI_MCR);
-
- reg = qspi_readl(q, base + QUADSPI_SMPR);
- qspi_writel(q, reg & ~(QUADSPI_SMPR_FSDLY_MASK
- | QUADSPI_SMPR_FSPHS_MASK
- | QUADSPI_SMPR_HSENA_MASK
- | QUADSPI_SMPR_DDRSMP_MASK), base + QUADSPI_SMPR);
-
- /* Enable the module */
- qspi_writel(q, QUADSPI_MCR_RESERVED_MASK | QUADSPI_MCR_END_CFG_MASK,
- base + QUADSPI_MCR);
-
- /* clear all interrupt status */
- qspi_writel(q, 0xffffffff, q->iobase + QUADSPI_FR);
-
- /* enable the interrupt */
- qspi_writel(q, QUADSPI_RSER_TFIE, q->iobase + QUADSPI_RSER);
-
- return 0;
-}
-
-static int fsl_qspi_nor_setup_last(struct fsl_qspi *q)
-{
- unsigned long rate = q->clk_rate;
- int ret;
-
- if (needs_4x_clock(q))
- rate *= 4;
-
- /* disable and unprepare clock to avoid glitch pass to controller */
- fsl_qspi_clk_disable_unprep(q);
-
- ret = clk_set_rate(q->clk, rate);
- if (ret)
- return ret;
-
- ret = fsl_qspi_clk_prep_enable(q);
- if (ret)
- return ret;
-
- /* Init the LUT table again. */
- fsl_qspi_init_lut(q);
-
- /* Init for AHB read */
- fsl_qspi_init_ahb_read(q);
-
- return 0;
-}
-
-static const struct of_device_id fsl_qspi_dt_ids[] = {
- { .compatible = "fsl,vf610-qspi", .data = &vybrid_data, },
- { .compatible = "fsl,imx6sx-qspi", .data = &imx6sx_data, },
- { .compatible = "fsl,imx7d-qspi", .data = &imx7d_data, },
- { .compatible = "fsl,imx6ul-qspi", .data = &imx6ul_data, },
- { .compatible = "fsl,ls1021a-qspi", .data = (void *)&ls1021a_data, },
- { .compatible = "fsl,ls2080a-qspi", .data = &ls2080a_data, },
- { /* sentinel */ }
-};
-MODULE_DEVICE_TABLE(of, fsl_qspi_dt_ids);
-
-static void fsl_qspi_set_base_addr(struct fsl_qspi *q, struct spi_nor *nor)
-{
- q->chip_base_addr = q->nor_size * (nor - q->nor);
-}
-
-static int fsl_qspi_read_reg(struct spi_nor *nor, u8 opcode, u8 *buf, int len)
-{
- int ret;
- struct fsl_qspi *q = nor->priv;
-
- ret = fsl_qspi_runcmd(q, opcode, 0, len);
- if (ret)
- return ret;
-
- fsl_qspi_read_data(q, len, buf);
- return 0;
-}
-
-static int fsl_qspi_write_reg(struct spi_nor *nor, u8 opcode, u8 *buf, int len)
-{
- struct fsl_qspi *q = nor->priv;
- int ret;
-
- if (!buf) {
- ret = fsl_qspi_runcmd(q, opcode, 0, 1);
- if (ret)
- return ret;
-
- if (opcode == SPINOR_OP_CHIP_ERASE)
- fsl_qspi_invalid(q);
-
- } else if (len > 0) {
- ret = fsl_qspi_nor_write(q, nor, opcode, 0,
- (u32 *)buf, len);
- if (ret > 0)
- return 0;
- } else {
- dev_err(q->dev, "invalid cmd %d\n", opcode);
- ret = -EINVAL;
- }
-
- return ret;
-}
-
-static ssize_t fsl_qspi_write(struct spi_nor *nor, loff_t to,
- size_t len, const u_char *buf)
-{
- struct fsl_qspi *q = nor->priv;
- ssize_t ret = fsl_qspi_nor_write(q, nor, nor->program_opcode, to,
- (u32 *)buf, len);
-
- /* invalid the data in the AHB buffer. */
- fsl_qspi_invalid(q);
- return ret;
-}
-
-static ssize_t fsl_qspi_read(struct spi_nor *nor, loff_t from,
- size_t len, u_char *buf)
-{
- struct fsl_qspi *q = nor->priv;
- u8 cmd = nor->read_opcode;
-
- /* if necessary,ioremap buffer before AHB read, */
- if (!q->ahb_addr) {
- q->memmap_offs = q->chip_base_addr + from;
- q->memmap_len = len > QUADSPI_MIN_IOMAP ? len : QUADSPI_MIN_IOMAP;
-
- q->ahb_addr = ioremap_nocache(
- q->memmap_phy + q->memmap_offs,
- q->memmap_len);
- if (!q->ahb_addr) {
- dev_err(q->dev, "ioremap failed\n");
- return -ENOMEM;
- }
- /* ioremap if the data requested is out of range */
- } else if (q->chip_base_addr + from < q->memmap_offs
- || q->chip_base_addr + from + len >
- q->memmap_offs + q->memmap_len) {
- iounmap(q->ahb_addr);
-
- q->memmap_offs = q->chip_base_addr + from;
- q->memmap_len = len > QUADSPI_MIN_IOMAP ? len : QUADSPI_MIN_IOMAP;
- q->ahb_addr = ioremap_nocache(
- q->memmap_phy + q->memmap_offs,
- q->memmap_len);
- if (!q->ahb_addr) {
- dev_err(q->dev, "ioremap failed\n");
- return -ENOMEM;
- }
- }
-
- dev_dbg(q->dev, "cmd [%x],read from %p, len:%zd\n",
- cmd, q->ahb_addr + q->chip_base_addr + from - q->memmap_offs,
- len);
-
- /* Read out the data directly from the AHB buffer.*/
- memcpy(buf, q->ahb_addr + q->chip_base_addr + from - q->memmap_offs,
- len);
-
- return len;
-}
-
-static int fsl_qspi_erase(struct spi_nor *nor, loff_t offs)
-{
- struct fsl_qspi *q = nor->priv;
- int ret;
-
- dev_dbg(nor->dev, "%dKiB at 0x%08x:0x%08x\n",
- nor->mtd.erasesize / 1024, q->chip_base_addr, (u32)offs);
-
- ret = fsl_qspi_runcmd(q, nor->erase_opcode, offs, 0);
- if (ret)
- return ret;
-
- fsl_qspi_invalid(q);
- return 0;
-}
-
-static int fsl_qspi_prep(struct spi_nor *nor, enum spi_nor_ops ops)
-{
- struct fsl_qspi *q = nor->priv;
- int ret;
-
- mutex_lock(&q->lock);
-
- ret = fsl_qspi_clk_prep_enable(q);
- if (ret)
- goto err_mutex;
-
- fsl_qspi_set_base_addr(q, nor);
- return 0;
-
-err_mutex:
- mutex_unlock(&q->lock);
- return ret;
-}
-
-static void fsl_qspi_unprep(struct spi_nor *nor, enum spi_nor_ops ops)
-{
- struct fsl_qspi *q = nor->priv;
-
- fsl_qspi_clk_disable_unprep(q);
- mutex_unlock(&q->lock);
-}
-
-static int fsl_qspi_probe(struct platform_device *pdev)
-{
- const struct spi_nor_hwcaps hwcaps = {
- .mask = SNOR_HWCAPS_READ_1_1_4 |
- SNOR_HWCAPS_PP,
- };
- struct device_node *np = pdev->dev.of_node;
- struct device *dev = &pdev->dev;
- struct fsl_qspi *q;
- struct resource *res;
- struct spi_nor *nor;
- struct mtd_info *mtd;
- int ret, i = 0;
-
- q = devm_kzalloc(dev, sizeof(*q), GFP_KERNEL);
- if (!q)
- return -ENOMEM;
-
- q->nor_num = of_get_child_count(dev->of_node);
- if (!q->nor_num || q->nor_num > FSL_QSPI_MAX_CHIP)
- return -ENODEV;
-
- q->dev = dev;
- q->devtype_data = of_device_get_match_data(dev);
- if (!q->devtype_data)
- return -ENODEV;
- platform_set_drvdata(pdev, q);
-
- /* find the resources */
- res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "QuadSPI");
- q->iobase = devm_ioremap_resource(dev, res);
- if (IS_ERR(q->iobase))
- return PTR_ERR(q->iobase);
-
- q->big_endian = of_property_read_bool(np, "big-endian");
- res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
- "QuadSPI-memory");
- if (!devm_request_mem_region(dev, res->start, resource_size(res),
- res->name)) {
- dev_err(dev, "can't request region for resource %pR\n", res);
- return -EBUSY;
- }
-
- q->memmap_phy = res->start;
-
- /* find the clocks */
- q->clk_en = devm_clk_get(dev, "qspi_en");
- if (IS_ERR(q->clk_en))
- return PTR_ERR(q->clk_en);
-
- q->clk = devm_clk_get(dev, "qspi");
- if (IS_ERR(q->clk))
- return PTR_ERR(q->clk);
-
- ret = fsl_qspi_clk_prep_enable(q);
- if (ret) {
- dev_err(dev, "can not enable the clock\n");
- goto clk_failed;
- }
-
- /* find the irq */
- ret = platform_get_irq(pdev, 0);
- if (ret < 0) {
- dev_err(dev, "failed to get the irq: %d\n", ret);
- goto irq_failed;
- }
-
- ret = devm_request_irq(dev, ret,
- fsl_qspi_irq_handler, 0, pdev->name, q);
- if (ret) {
- dev_err(dev, "failed to request irq: %d\n", ret);
- goto irq_failed;
- }
-
- ret = fsl_qspi_nor_setup(q);
- if (ret)
- goto irq_failed;
-
- if (of_get_property(np, "fsl,qspi-has-second-chip", NULL))
- q->has_second_chip = true;
-
- mutex_init(&q->lock);
-
- /* iterate the subnodes. */
- for_each_available_child_of_node(dev->of_node, np) {
- /* skip the holes */
- if (!q->has_second_chip)
- i *= 2;
-
- nor = &q->nor[i];
- mtd = &nor->mtd;
-
- nor->dev = dev;
- spi_nor_set_flash_node(nor, np);
- nor->priv = q;
-
- if (q->nor_num > 1 && !mtd->name) {
- int spiflash_idx;
-
- ret = of_property_read_u32(np, "reg", &spiflash_idx);
- if (!ret) {
- mtd->name = devm_kasprintf(dev, GFP_KERNEL,
- "%s-%d",
- dev_name(dev),
- spiflash_idx);
- if (!mtd->name) {
- ret = -ENOMEM;
- goto mutex_failed;
- }
- } else {
- dev_warn(dev, "reg property is missing\n");
- }
- }
-
- /* fill the hooks */
- nor->read_reg = fsl_qspi_read_reg;
- nor->write_reg = fsl_qspi_write_reg;
- nor->read = fsl_qspi_read;
- nor->write = fsl_qspi_write;
- nor->erase = fsl_qspi_erase;
-
- nor->prepare = fsl_qspi_prep;
- nor->unprepare = fsl_qspi_unprep;
-
- ret = of_property_read_u32(np, "spi-max-frequency",
- &q->clk_rate);
- if (ret < 0)
- goto mutex_failed;
-
- /* set the chip address for READID */
- fsl_qspi_set_base_addr(q, nor);
-
- ret = spi_nor_scan(nor, NULL, &hwcaps);
- if (ret)
- goto mutex_failed;
-
- ret = mtd_device_register(mtd, NULL, 0);
- if (ret)
- goto mutex_failed;
-
- /* Set the correct NOR size now. */
- if (q->nor_size == 0) {
- q->nor_size = mtd->size;
-
- /* Map the SPI NOR to accessiable address */
- fsl_qspi_set_map_addr(q);
- }
-
- /*
- * The TX FIFO is 64 bytes in the Vybrid, but the Page Program
- * may writes 265 bytes per time. The write is working in the
- * unit of the TX FIFO, not in the unit of the SPI NOR's page
- * size.
- *
- * So shrink the spi_nor->page_size if it is larger then the
- * TX FIFO.
- */
- if (nor->page_size > q->devtype_data->txfifo)
- nor->page_size = q->devtype_data->txfifo;
-
- i++;
- }
-
- /* finish the rest init. */
- ret = fsl_qspi_nor_setup_last(q);
- if (ret)
- goto last_init_failed;
-
- fsl_qspi_clk_disable_unprep(q);
- return 0;
-
-last_init_failed:
- for (i = 0; i < q->nor_num; i++) {
- /* skip the holes */
- if (!q->has_second_chip)
- i *= 2;
- mtd_device_unregister(&q->nor[i].mtd);
- }
-mutex_failed:
- mutex_destroy(&q->lock);
-irq_failed:
- fsl_qspi_clk_disable_unprep(q);
-clk_failed:
- dev_err(dev, "Freescale QuadSPI probe failed\n");
- return ret;
-}
-
-static int fsl_qspi_remove(struct platform_device *pdev)
-{
- struct fsl_qspi *q = platform_get_drvdata(pdev);
- int i;
-
- for (i = 0; i < q->nor_num; i++) {
- /* skip the holes */
- if (!q->has_second_chip)
- i *= 2;
- mtd_device_unregister(&q->nor[i].mtd);
- }
-
- /* disable the hardware */
- qspi_writel(q, QUADSPI_MCR_MDIS_MASK, q->iobase + QUADSPI_MCR);
- qspi_writel(q, 0x0, q->iobase + QUADSPI_RSER);
-
- mutex_destroy(&q->lock);
-
- if (q->ahb_addr)
- iounmap(q->ahb_addr);
-
- return 0;
-}
-
-static int fsl_qspi_suspend(struct platform_device *pdev, pm_message_t state)
-{
- return 0;
-}
-
-static int fsl_qspi_resume(struct platform_device *pdev)
-{
- int ret;
- struct fsl_qspi *q = platform_get_drvdata(pdev);
-
- ret = fsl_qspi_clk_prep_enable(q);
- if (ret)
- return ret;
-
- fsl_qspi_nor_setup(q);
- fsl_qspi_set_map_addr(q);
- fsl_qspi_nor_setup_last(q);
-
- fsl_qspi_clk_disable_unprep(q);
-
- return 0;
-}
-
-static struct platform_driver fsl_qspi_driver = {
- .driver = {
- .name = "fsl-quadspi",
- .of_match_table = fsl_qspi_dt_ids,
- },
- .probe = fsl_qspi_probe,
- .remove = fsl_qspi_remove,
- .suspend = fsl_qspi_suspend,
- .resume = fsl_qspi_resume,
-};
-module_platform_driver(fsl_qspi_driver);
-
-MODULE_DESCRIPTION("Freescale QuadSPI Controller Driver");
-MODULE_AUTHOR("Freescale Semiconductor Inc.");
-MODULE_LICENSE("GPL v2");
--
2.7.4
Frieder Schrempf
2018-07-05 11:15:03 UTC
Permalink
The FSL QSPI driver was moved to the SPI framework and it now
acts as a SPI controller. Therefore the subnodes need to set
spi-[rx/tx]-bus-width = <4>, so quad mode is used just as before.

Also the properties 'num-cs' and 'bus-num' were never read by the
driver and can be removed.

The property 'fsl,qspi-has-second-chip' is not needed anymore
and will be removed after the old driver was disabled to avoid
breaking fsl-ls1046a-rdb.dts.

Signed-off-by: Frieder Schrempf <***@exceet.de>
---
arch/arm64/boot/dts/freescale/fsl-ls1043a-qds.dts | 3 ++-
arch/arm64/boot/dts/freescale/fsl-ls1046a-qds.dts | 4 ++--
arch/arm64/boot/dts/freescale/fsl-ls1046a-rdb.dts | 6 ++++--
arch/arm64/boot/dts/freescale/fsl-ls208xa-qds.dtsi | 4 ++++
4 files changed, 12 insertions(+), 5 deletions(-)

diff --git a/arch/arm64/boot/dts/freescale/fsl-ls1043a-qds.dts b/arch/arm64/boot/dts/freescale/fsl-ls1043a-qds.dts
index 6341281..31e7b31 100644
--- a/arch/arm64/boot/dts/freescale/fsl-ls1043a-qds.dts
+++ b/arch/arm64/boot/dts/freescale/fsl-ls1043a-qds.dts
@@ -170,7 +170,6 @@
};

&qspi {
- bus-num = <0>;
status = "okay";

qflash0: ***@0 {
@@ -178,6 +177,8 @@
#address-cells = <1>;
#size-cells = <1>;
spi-max-frequency = <20000000>;
+ spi-rx-bus-width = <4>;
+ spi-tx-bus-width = <4>;
reg = <0>;
};
};
diff --git a/arch/arm64/boot/dts/freescale/fsl-ls1046a-qds.dts b/arch/arm64/boot/dts/freescale/fsl-ls1046a-qds.dts
index 434383b..dc10105 100644
--- a/arch/arm64/boot/dts/freescale/fsl-ls1046a-qds.dts
+++ b/arch/arm64/boot/dts/freescale/fsl-ls1046a-qds.dts
@@ -198,8 +198,6 @@
};

&qspi {
- num-cs = <2>;
- bus-num = <0>;
status = "okay";

qflash0: ***@0 {
@@ -207,6 +205,8 @@
#address-cells = <1>;
#size-cells = <1>;
spi-max-frequency = <20000000>;
+ spi-rx-bus-width = <4>;
+ spi-tx-bus-width = <4>;
reg = <0>;
};
};
diff --git a/arch/arm64/boot/dts/freescale/fsl-ls1046a-rdb.dts b/arch/arm64/boot/dts/freescale/fsl-ls1046a-rdb.dts
index 5dc2782..1848c33 100644
--- a/arch/arm64/boot/dts/freescale/fsl-ls1046a-rdb.dts
+++ b/arch/arm64/boot/dts/freescale/fsl-ls1046a-rdb.dts
@@ -136,8 +136,6 @@
};

&qspi {
- num-cs = <2>;
- bus-num = <0>;
status = "okay";

qflash0: ***@0 {
@@ -145,6 +143,8 @@
#address-cells = <1>;
#size-cells = <1>;
spi-max-frequency = <20000000>;
+ spi-rx-bus-width = <4>;
+ spi-tx-bus-width = <4>;
reg = <0>;
};

@@ -153,6 +153,8 @@
#address-cells = <1>;
#size-cells = <1>;
spi-max-frequency = <20000000>;
+ spi-rx-bus-width = <4>;
+ spi-tx-bus-width = <4>;
reg = <1>;
};
};
diff --git a/arch/arm64/boot/dts/freescale/fsl-ls208xa-qds.dtsi b/arch/arm64/boot/dts/freescale/fsl-ls208xa-qds.dtsi
index 1de6188..fc62ed9 100644
--- a/arch/arm64/boot/dts/freescale/fsl-ls208xa-qds.dtsi
+++ b/arch/arm64/boot/dts/freescale/fsl-ls208xa-qds.dtsi
@@ -170,6 +170,8 @@
#size-cells = <1>;
compatible = "st,m25p80";
spi-max-frequency = <20000000>;
+ spi-rx-bus-width = <4>;
+ spi-tx-bus-width = <4>;
reg = <0>;
};
flash2: ***@2 {
@@ -177,6 +179,8 @@
#size-cells = <1>;
compatible = "st,m25p80";
spi-max-frequency = <20000000>;
+ spi-rx-bus-width = <4>;
+ spi-tx-bus-width = <4>;
reg = <2>;
};
};
--
2.7.4
Frieder Schrempf
2018-07-05 11:14:59 UTC
Permalink
This driver is derived from the SPI NOR driver at
mtd/spi-nor/fsl-quadspi.c. It uses the new SPI memory interface
of the SPI framework to issue flash memory operations to up to
four connected flash chips (2 buses with 2 CS each).

The controller does not support generic SPI messages.

Signed-off-by: Frieder Schrempf <***@exceet.de>
---
Changes in v2:
==============
* Add Yogesh Gaur and Suresh Gupta as authors
* Use GENMASK() for generating bitmasks
* Use callback functions for read/write of registers
* Attach the seq variable to the selected CS
* Avoid using conditional in read/write loop
* Avoid infinite loop and use a timeout instead
* Return error pointer when allocation in fsl_qspi_get_name() fails
* Remove redundant iounmap()
* Put suspend()/resume() in struct dev_pm_ops instead of struct platform_driver

drivers/spi/Kconfig | 11 +
drivers/spi/Makefile | 1 +
drivers/spi/spi-fsl-qspi.c | 954 ++++++++++++++++++++++++++++++++++++++++
3 files changed, 966 insertions(+)

diff --git a/drivers/spi/Kconfig b/drivers/spi/Kconfig
index ad5d68e..5e95ffc 100644
--- a/drivers/spi/Kconfig
+++ b/drivers/spi/Kconfig
@@ -251,6 +251,17 @@ config SPI_FSL_LPSPI
help
This enables Freescale i.MX LPSPI controllers in master mode.

+config SPI_FSL_QSPI
+ tristate "Freescale QSPI controller"
+ depends on ARCH_MXC || SOC_LS1021A || ARCH_LAYERSCAPE || COMPILE_TEST
+ depends on HAS_IOMEM
+ help
+ This enables support for the Quad SPI controller in master mode.
+ Up to four flash chips can be connected on two buses with two
+ chipselects each.
+ This controller does not support generic SPI messages. It only
+ supports the high-level SPI memory interface.
+
config SPI_GPIO
tristate "GPIO-based bitbanging SPI Master"
depends on GPIOLIB || COMPILE_TEST
diff --git a/drivers/spi/Makefile b/drivers/spi/Makefile
index cb1f437..a8f7fda 100644
--- a/drivers/spi/Makefile
+++ b/drivers/spi/Makefile
@@ -43,6 +43,7 @@ obj-$(CONFIG_SPI_FSL_DSPI) += spi-fsl-dspi.o
obj-$(CONFIG_SPI_FSL_LIB) += spi-fsl-lib.o
obj-$(CONFIG_SPI_FSL_ESPI) += spi-fsl-espi.o
obj-$(CONFIG_SPI_FSL_LPSPI) += spi-fsl-lpspi.o
+obj-$(CONFIG_SPI_FSL_QSPI) += spi-fsl-qspi.o
obj-$(CONFIG_SPI_FSL_SPI) += spi-fsl-spi.o
obj-$(CONFIG_SPI_GPIO) += spi-gpio.o
obj-$(CONFIG_SPI_IMG_SPFI) += spi-img-spfi.o
diff --git a/drivers/spi/spi-fsl-qspi.c b/drivers/spi/spi-fsl-qspi.c
new file mode 100644
index 0000000..3e17209
--- /dev/null
+++ b/drivers/spi/spi-fsl-qspi.c
@@ -0,0 +1,954 @@
+// SPDX-License-Identifier: GPL-2.0+
+
+/*
+ * Freescale QuadSPI driver.
+ *
+ * Copyright (C) 2013 Freescale Semiconductor, Inc.
+ * Copyright (C) 2018 Bootlin
+ * Copyright (C) 2018 Exceet Electronics GmbH
+ *
+ * Transition to SPI MEM interface:
+ * Author:
+ * Boris Brezillion <***@bootlin.com>
+ * Frieder Schrempf <***@exceet.de>
+ * Yogesh Gaur <***@nxp.com>
+ * Suresh Gupta <***@nxp.com>
+ *
+ * Based on the original fsl-quadspi.c spi-nor driver:
+ * Author: Freescale Semiconductor, Inc.
+ *
+ */
+
+#include <linux/bitops.h>
+#include <linux/clk.h>
+#include <linux/completion.h>
+#include <linux/delay.h>
+#include <linux/err.h>
+#include <linux/errno.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/jiffies.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/of.h>
+#include <linux/of_device.h>
+#include <linux/platform_device.h>
+#include <linux/pm_qos.h>
+#include <linux/sizes.h>
+
+#include <linux/spi/spi.h>
+#include <linux/spi/spi-mem.h>
+
+/*
+ * The driver only uses one single LUT entry, that is updated on
+ * each call of exec_op(). Index 0 is preset at boot with a basic
+ * read operation, so let's use the last entry (15).
+ */
+#define SEQID_LUT 15
+
+/* Registers used by the driver */
+#define QUADSPI_MCR 0x00
+#define QUADSPI_MCR_RESERVED_MASK GENMASK(19, 16)
+#define QUADSPI_MCR_MDIS_MASK BIT(14)
+#define QUADSPI_MCR_CLR_TXF_MASK BIT(11)
+#define QUADSPI_MCR_CLR_RXF_MASK BIT(10)
+#define QUADSPI_MCR_DDR_EN_MASK BIT(7)
+#define QUADSPI_MCR_END_CFG_MASK GENMASK(3, 2)
+#define QUADSPI_MCR_SWRSTHD_MASK BIT(1)
+#define QUADSPI_MCR_SWRSTSD_MASK BIT(0)
+
+#define QUADSPI_IPCR 0x08
+#define QUADSPI_IPCR_SEQID_SHIFT 24
+
+#define QUADSPI_BUF3CR 0x1c
+#define QUADSPI_BUF3CR_ALLMST_MASK BIT(31)
+#define QUADSPI_BUF3CR_ADATSZ_SHIFT 8
+#define QUADSPI_BUF3CR_ADATSZ_MASK (0xFF << QUADSPI_BUF3CR_ADATSZ_SHIFT)
+
+#define QUADSPI_BFGENCR 0x20
+#define QUADSPI_BFGENCR_SEQID_SHIFT 12
+
+#define QUADSPI_BUF0IND 0x30
+#define QUADSPI_BUF1IND 0x34
+#define QUADSPI_BUF2IND 0x38
+#define QUADSPI_SFAR 0x100
+
+#define QUADSPI_SMPR 0x108
+#define QUADSPI_SMPR_DDRSMP_MASK GENMASK(18, 16)
+#define QUADSPI_SMPR_FSDLY_MASK BIT(6)
+#define QUADSPI_SMPR_FSPHS_MASK BIT(5)
+#define QUADSPI_SMPR_HSENA_MASK BIT(0)
+
+#define QUADSPI_RBCT 0x110
+#define QUADSPI_RBCT_WMRK_MASK GENMASK(4, 0)
+#define QUADSPI_RBCT_RXBRD_USEIPS BIT(8)
+
+#define QUADSPI_TBDR 0x154
+
+#define QUADSPI_SR 0x15c
+#define QUADSPI_SR_IP_ACC_MASK BIT(1)
+#define QUADSPI_SR_AHB_ACC_MASK BIT(2)
+
+#define QUADSPI_FR 0x160
+#define QUADSPI_FR_TFF_MASK BIT(0)
+
+#define QUADSPI_SPTRCLR 0x16c
+#define QUADSPI_SPTRCLR_IPPTRC BIT(8)
+#define QUADSPI_SPTRCLR_BFPTRC BIT(0)
+
+#define QUADSPI_SFA1AD 0x180
+#define QUADSPI_SFA2AD 0x184
+#define QUADSPI_SFB1AD 0x188
+#define QUADSPI_SFB2AD 0x18c
+#define QUADSPI_RBDR(x) (0x200 + ((x) * 4))
+
+#define QUADSPI_LUTKEY 0x300
+#define QUADSPI_LUTKEY_VALUE 0x5AF05AF0
+
+#define QUADSPI_LCKCR 0x304
+#define QUADSPI_LCKER_LOCK BIT(0)
+#define QUADSPI_LCKER_UNLOCK BIT(1)
+
+#define QUADSPI_RSER 0x164
+#define QUADSPI_RSER_TFIE BIT(0)
+
+#define QUADSPI_LUT_BASE 0x310
+#define QUADSPI_LUT_OFFSET (SEQID_LUT * 4 * 4)
+#define QUADSPI_LUT_REG(idx) \
+ (QUADSPI_LUT_BASE + QUADSPI_LUT_OFFSET + (idx) * 4)
+
+/* Instruction set for the LUT register */
+#define LUT_STOP 0
+#define LUT_CMD 1
+#define LUT_ADDR 2
+#define LUT_DUMMY 3
+#define LUT_MODE 4
+#define LUT_MODE2 5
+#define LUT_MODE4 6
+#define LUT_FSL_READ 7
+#define LUT_FSL_WRITE 8
+#define LUT_JMP_ON_CS 9
+#define LUT_ADDR_DDR 10
+#define LUT_MODE_DDR 11
+#define LUT_MODE2_DDR 12
+#define LUT_MODE4_DDR 13
+#define LUT_FSL_READ_DDR 14
+#define LUT_FSL_WRITE_DDR 15
+#define LUT_DATA_LEARN 16
+
+/*
+ * The PAD definitions for LUT register.
+ *
+ * The pad stands for the number of IO lines [0:3].
+ * For example, the quad read needs four IO lines,
+ * so you should use LUT_PAD(4).
+ */
+#define LUT_PAD(x) (fls(x) - 1)
+
+/*
+ * Macro for constructing the LUT entries with the following
+ * register layout:
+ *
+ * ---------------------------------------------------
+ * | INSTR1 | PAD1 | OPRND1 | INSTR0 | PAD0 | OPRND0 |
+ * ---------------------------------------------------
+ */
+#define LUT_DEF(idx, ins, pad, opr) \
+ ((((ins) << 10) | ((pad) << 8) | (opr)) << (((idx) % 2) * 16))
+
+/* Controller needs driver to swap endianness */
+#define QUADSPI_QUIRK_SWAP_ENDIAN BIT(0)
+
+/* Controller needs 4x internal clock */
+#define QUADSPI_QUIRK_4X_INT_CLK BIT(1)
+
+/*
+ * TKT253890, the controller needs the driver to fill the txfifo with
+ * 16 bytes at least to trigger a data transfer, even though the extra
+ * data won't be transferred.
+ */
+#define QUADSPI_QUIRK_TKT253890 BIT(2)
+
+/* TKT245618, the controller cannot wake up from wait mode */
+#define QUADSPI_QUIRK_TKT245618 BIT(3)
+
+enum fsl_qspi_devtype {
+ FSL_QUADSPI_VYBRID,
+ FSL_QUADSPI_IMX6SX,
+ FSL_QUADSPI_IMX7D,
+ FSL_QUADSPI_IMX6UL,
+ FSL_QUADSPI_LS1021A,
+ FSL_QUADSPI_LS2080A,
+};
+
+struct fsl_qspi_devtype_data {
+ enum fsl_qspi_devtype devtype;
+ unsigned int rxfifo;
+ unsigned int txfifo;
+ unsigned int ahb_buf_size;
+ unsigned int quirks;
+};
+
+static const struct fsl_qspi_devtype_data vybrid_data = {
+ .devtype = FSL_QUADSPI_VYBRID,
+ .rxfifo = SZ_128,
+ .txfifo = SZ_64,
+ .ahb_buf_size = SZ_1K,
+ .quirks = QUADSPI_QUIRK_SWAP_ENDIAN,
+};
+
+static const struct fsl_qspi_devtype_data imx6sx_data = {
+ .devtype = FSL_QUADSPI_IMX6SX,
+ .rxfifo = SZ_128,
+ .txfifo = SZ_512,
+ .ahb_buf_size = SZ_1K,
+ .quirks = QUADSPI_QUIRK_4X_INT_CLK | QUADSPI_QUIRK_TKT245618,
+};
+
+static const struct fsl_qspi_devtype_data imx7d_data = {
+ .devtype = FSL_QUADSPI_IMX7D,
+ .rxfifo = SZ_512,
+ .txfifo = SZ_512,
+ .ahb_buf_size = SZ_1K,
+ .quirks = QUADSPI_QUIRK_TKT253890 | QUADSPI_QUIRK_4X_INT_CLK,
+};
+
+static const struct fsl_qspi_devtype_data imx6ul_data = {
+ .devtype = FSL_QUADSPI_IMX6UL,
+ .rxfifo = SZ_128,
+ .txfifo = SZ_512,
+ .ahb_buf_size = SZ_1K,
+ .quirks = QUADSPI_QUIRK_TKT253890 | QUADSPI_QUIRK_4X_INT_CLK,
+};
+
+static const struct fsl_qspi_devtype_data ls1021a_data = {
+ .devtype = FSL_QUADSPI_LS1021A,
+ .rxfifo = SZ_128,
+ .txfifo = SZ_64,
+ .ahb_buf_size = SZ_1K,
+ .quirks = 0,
+};
+
+static const struct fsl_qspi_devtype_data ls2080a_data = {
+ .devtype = FSL_QUADSPI_LS2080A,
+ .rxfifo = SZ_128,
+ .txfifo = SZ_64,
+ .ahb_buf_size = SZ_1K,
+ .quirks = QUADSPI_QUIRK_TKT253890,
+};
+
+struct fsl_qspi {
+ void __iomem *iobase;
+ void __iomem *ahb_addr;
+ u32 memmap_phy;
+ struct clk *clk, *clk_en;
+ struct device *dev;
+ struct completion c;
+ const struct fsl_qspi_devtype_data *devtype_data;
+ bool big_endian;
+ struct mutex lock;
+ struct pm_qos_request pm_qos_req;
+ int selected;
+ u8 seq;
+ void (*write)(u32 val, void __iomem *addr);
+ u32 (*read)(void __iomem *addr);
+};
+
+static inline int needs_swap_endian(struct fsl_qspi *q)
+{
+ return q->devtype_data->quirks & QUADSPI_QUIRK_SWAP_ENDIAN;
+}
+
+static inline int needs_4x_clock(struct fsl_qspi *q)
+{
+ return q->devtype_data->quirks & QUADSPI_QUIRK_4X_INT_CLK;
+}
+
+static inline int needs_fill_txfifo(struct fsl_qspi *q)
+{
+ return q->devtype_data->quirks & QUADSPI_QUIRK_TKT253890;
+}
+
+static inline int needs_wakeup_wait_mode(struct fsl_qspi *q)
+{
+ return q->devtype_data->quirks & QUADSPI_QUIRK_TKT245618;
+}
+
+/*
+ * An IC bug makes it necessary to rearrange the 32-bit data.
+ * Later chips, such as IMX6SLX, have fixed this bug.
+ */
+static inline u32 fsl_qspi_endian_xchg(struct fsl_qspi *q, u32 a)
+{
+ return needs_swap_endian(q) ? __swab32(a) : a;
+}
+
+static void qspi_writel_be(u32 val, void __iomem *addr)
+{
+ iowrite32be(val, addr);
+}
+
+static void qspi_writel(u32 val, void __iomem *addr)
+{
+ iowrite32(val, addr);
+}
+
+static u32 qspi_readl_be(void __iomem *addr)
+{
+ return ioread32be(addr);
+}
+
+static u32 qspi_readl(void __iomem *addr)
+{
+ return ioread32(addr);
+}
+
+static irqreturn_t fsl_qspi_irq_handler(int irq, void *dev_id)
+{
+ struct fsl_qspi *q = dev_id;
+ u32 reg;
+
+ /* clear interrupt */
+ reg = q->read(q->iobase + QUADSPI_FR);
+ q->write(reg, q->iobase + QUADSPI_FR);
+
+ if (reg & QUADSPI_FR_TFF_MASK)
+ complete(&q->c);
+
+ dev_dbg(q->dev, "QUADSPI_FR : 0x%.8x:0x%.8x\n", 0, reg);
+ return IRQ_HANDLED;
+}
+
+static int fsl_qspi_check_buswidth(struct fsl_qspi *q, u8 width)
+{
+ switch (width) {
+ case 1:
+ case 2:
+ case 4:
+ return 0;
+ }
+
+ return -ENOTSUPP;
+}
+
+static bool fsl_qspi_supports_op(struct spi_mem *mem,
+ const struct spi_mem_op *op)
+{
+ struct fsl_qspi *q = spi_controller_get_devdata(mem->spi->master);
+ int ret;
+
+ ret = fsl_qspi_check_buswidth(q, op->cmd.buswidth);
+
+ if (op->addr.nbytes)
+ ret |= fsl_qspi_check_buswidth(q, op->addr.buswidth);
+
+ if (op->dummy.nbytes)
+ ret |= fsl_qspi_check_buswidth(q, op->dummy.buswidth);
+
+ if (op->data.nbytes)
+ ret |= fsl_qspi_check_buswidth(q, op->data.buswidth);
+
+ if (ret)
+ return false;
+
+ /*
+ * The number of instructions needed for the op, needs
+ * to fit into a single LUT entry.
+ */
+ if (op->addr.nbytes +
+ (op->dummy.nbytes ? 1:0) +
+ (op->data.nbytes ? 1:0) > 6)
+ return false;
+
+ /* Max 64 dummy clock cycles supported */
+ if (op->dummy.nbytes * 8 / op->dummy.buswidth > 64)
+ return false;
+
+ /* Max data length, check controller limits and alignment */
+ if (op->data.dir == SPI_MEM_DATA_IN &&
+ (op->data.nbytes > q->devtype_data->ahb_buf_size ||
+ (op->data.nbytes > q->devtype_data->rxfifo - 4 &&
+ !IS_ALIGNED(op->data.nbytes, 8))))
+ return false;
+
+ if (op->data.dir == SPI_MEM_DATA_OUT &&
+ op->data.nbytes > q->devtype_data->txfifo)
+ return false;
+
+ return true;
+}
+
+static void fsl_qspi_prepare_lut(struct fsl_qspi *q,
+ const struct spi_mem_op *op)
+{
+ void __iomem *base = q->iobase;
+ u32 lutval[4] = {};
+ int lutidx = 1, i;
+
+ lutval[0] |= LUT_DEF(0, LUT_CMD, LUT_PAD(op->cmd.buswidth),
+ op->cmd.opcode);
+
+ /*
+ * For some unknown reason, using LUT_ADDR doesn't work in some
+ * cases (at least with only one byte long addresses), so
+ * let's use LUT_MODE to write the address bytes one by one
+ */
+ for (i = 0; i < op->addr.nbytes; i++) {
+ u8 addrbyte = op->addr.val >> (8 * (op->addr.nbytes - i - 1));
+
+ lutval[lutidx / 2] |= LUT_DEF(lutidx, LUT_MODE,
+ LUT_PAD(op->addr.buswidth),
+ addrbyte);
+ lutidx++;
+ }
+
+ if (op->dummy.nbytes) {
+ lutval[lutidx / 2] |= LUT_DEF(lutidx, LUT_DUMMY,
+ LUT_PAD(op->dummy.buswidth),
+ op->dummy.nbytes * 8 /
+ op->dummy.buswidth);
+ lutidx++;
+ }
+
+ if (op->data.nbytes) {
+ lutval[lutidx / 2] |= LUT_DEF(lutidx,
+ op->data.dir == SPI_MEM_DATA_IN ?
+ LUT_FSL_READ : LUT_FSL_WRITE,
+ LUT_PAD(op->data.buswidth),
+ 0);
+ lutidx++;
+ }
+
+ lutval[lutidx / 2] |= LUT_DEF(lutidx, LUT_STOP, 0, 0);
+
+ /* unlock LUT */
+ q->write(QUADSPI_LUTKEY_VALUE, q->iobase + QUADSPI_LUTKEY);
+ q->write(QUADSPI_LCKER_UNLOCK, q->iobase + QUADSPI_LCKCR);
+
+ /* fill LUT */
+ for (i = 0; i < ARRAY_SIZE(lutval); i++)
+ q->write(lutval[i], base + QUADSPI_LUT_REG(i));
+
+ /* lock LUT */
+ q->write(QUADSPI_LUTKEY_VALUE, q->iobase + QUADSPI_LUTKEY);
+ q->write(QUADSPI_LCKER_LOCK, q->iobase + QUADSPI_LCKCR);
+}
+
+static int fsl_qspi_clk_prep_enable(struct fsl_qspi *q)
+{
+ int ret;
+
+ ret = clk_prepare_enable(q->clk_en);
+ if (ret)
+ return ret;
+
+ ret = clk_prepare_enable(q->clk);
+ if (ret) {
+ clk_disable_unprepare(q->clk_en);
+ return ret;
+ }
+
+ if (needs_wakeup_wait_mode(q))
+ pm_qos_add_request(&q->pm_qos_req, PM_QOS_CPU_DMA_LATENCY, 0);
+
+ return 0;
+}
+
+static void fsl_qspi_clk_disable_unprep(struct fsl_qspi *q)
+{
+ if (needs_wakeup_wait_mode(q))
+ pm_qos_remove_request(&q->pm_qos_req);
+
+ clk_disable_unprepare(q->clk);
+ clk_disable_unprepare(q->clk_en);
+}
+
+static void fsl_qspi_select_mem(struct fsl_qspi *q, struct spi_device *spi)
+{
+ unsigned long rate = spi->max_speed_hz;
+ int ret, i;
+ u32 map_addr;
+
+ if (q->selected == spi->chip_select)
+ return;
+
+ /*
+ * In HW there can be a maximum of four chips on two buses with
+ * two chip selects on each bus. We use four chip selects in SW
+ * to differentiate between the four chips.
+ * We use the SFA1AD, SFA2AD, SFB1AD, SFB2AD registers to select
+ * the chip we want to access.
+ */
+ for (i = 0; i < 4; i++) {
+ if (i < spi->chip_select)
+ map_addr = q->memmap_phy;
+ else
+ map_addr = q->memmap_phy +
+ 2 * q->devtype_data->ahb_buf_size;
+
+ q->write(map_addr, q->iobase + QUADSPI_SFA1AD + (i * 4));
+ }
+
+ if (needs_4x_clock(q))
+ rate *= 4;
+
+ fsl_qspi_clk_disable_unprep(q);
+
+ ret = clk_set_rate(q->clk, rate);
+ if (ret)
+ return;
+
+ ret = fsl_qspi_clk_prep_enable(q);
+ if (ret)
+ return;
+
+ q->selected = spi->chip_select;
+}
+
+static void fsl_qspi_read_ahb(struct fsl_qspi *q, const struct spi_mem_op *op)
+{
+ /*
+ * We want to avoid needing to invalidate the cache by issueing
+ * a reset to the AHB and Serial Flash domain, as this needs
+ * time. So we change the address on each read to trigger an
+ * actual read operation on the flash. The actual address for
+ * the flash memory is set by programming the LUT.
+ */
+ memcpy_fromio(op->data.buf.in,
+ q->ahb_addr +
+ (((q->seq & (1 << q->selected)) == 0 ? 0:1) *
+ q->devtype_data->ahb_buf_size),
+ op->data.nbytes);
+
+ q->seq ^= 1 << q->selected;
+}
+
+static void fsl_qspi_fill_txfifo(struct fsl_qspi *q,
+ const struct spi_mem_op *op)
+{
+ void __iomem *base = q->iobase;
+ int i;
+ u32 val;
+
+ for (i = 0; i < ALIGN_DOWN(op->data.nbytes, 4); i += 4) {
+ memcpy(&val, op->data.buf.out + i, 4);
+ val = fsl_qspi_endian_xchg(q, val);
+ q->write(val, base + QUADSPI_TBDR);
+ }
+
+ if (i < op->data.nbytes) {
+ memcpy(&val, op->data.buf.out + i, op->data.nbytes - i);
+ val = fsl_qspi_endian_xchg(q, val);
+ q->write(val, base + QUADSPI_TBDR);
+ }
+
+ if (needs_fill_txfifo(q)) {
+ for (i = op->data.nbytes; i < 16; i += 4)
+ q->write(0, base + QUADSPI_TBDR);
+ }
+}
+
+static void fsl_qspi_read_rxfifo(struct fsl_qspi *q,
+ const struct spi_mem_op *op)
+{
+ void __iomem *base = q->iobase;
+ int i;
+ u8 *buf = op->data.buf.in;
+ u32 val;
+
+ for (i = 0; i < ALIGN_DOWN(op->data.nbytes, 4); i += 4) {
+ val = q->read(base + QUADSPI_RBDR(i / 4));
+ val = fsl_qspi_endian_xchg(q, val);
+ memcpy(buf + i, &val, 4);
+ }
+
+ if (i < op->data.nbytes) {
+ val = q->read(base + QUADSPI_RBDR(i / 4));
+ val = fsl_qspi_endian_xchg(q, val);
+ memcpy(buf + i, &val, op->data.nbytes - i);
+ }
+}
+
+static int fsl_qspi_do_op(struct fsl_qspi *q, const struct spi_mem_op *op)
+{
+ void __iomem *base = q->iobase;
+ int err = 0;
+
+ init_completion(&q->c);
+
+ /*
+ * Always start the sequence at the same index since we update
+ * the LUT at each exec_op() call. And also specify the DATA
+ * length, since it's has not been specified in the LUT.
+ */
+ q->write(op->data.nbytes |
+ (SEQID_LUT << QUADSPI_IPCR_SEQID_SHIFT),
+ base + QUADSPI_IPCR);
+
+ /* Wait for the interrupt. */
+ if (!wait_for_completion_timeout(&q->c, msecs_to_jiffies(1000)))
+ err = -ETIMEDOUT;
+
+ if (!err && op->data.nbytes && op->data.dir == SPI_MEM_DATA_IN)
+ fsl_qspi_read_rxfifo(q, op);
+
+ return err;
+}
+
+static int fsl_qspi_exec_op(struct spi_mem *mem, const struct spi_mem_op *op)
+{
+ struct fsl_qspi *q = spi_controller_get_devdata(mem->spi->master);
+ void __iomem *base = q->iobase;
+ int err = 0;
+ unsigned int timeout = 1000;
+
+ mutex_lock(&q->lock);
+
+ /* wait for the controller being ready */
+ do {
+ u32 status;
+
+ status = q->read(base + QUADSPI_SR);
+ if (status &
+ (QUADSPI_SR_IP_ACC_MASK | QUADSPI_SR_AHB_ACC_MASK)) {
+ udelay(1);
+ dev_dbg(q->dev, "The controller is busy, 0x%x\n",
+ status);
+ continue;
+ }
+ break;
+ } while (--timeout);
+
+ fsl_qspi_select_mem(q, mem->spi);
+
+ q->write(q->memmap_phy, base + QUADSPI_SFAR);
+
+ q->write(q->read(base + QUADSPI_MCR) |
+ QUADSPI_MCR_CLR_RXF_MASK | QUADSPI_MCR_CLR_TXF_MASK,
+ base + QUADSPI_MCR);
+
+ q->write(QUADSPI_SPTRCLR_BFPTRC | QUADSPI_SPTRCLR_IPPTRC,
+ base + QUADSPI_SPTRCLR);
+
+ fsl_qspi_prepare_lut(q, op);
+
+ /*
+ * If we have large chunks of data, we read them through the AHB bus
+ * by accessing the mapped memory. In all other cases we use
+ * IP commands to access the flash.
+ */
+ if (op->data.nbytes > (q->devtype_data->rxfifo - 4) &&
+ op->data.dir == SPI_MEM_DATA_IN) {
+ fsl_qspi_read_ahb(q, op);
+ } else {
+ q->write(QUADSPI_RBCT_WMRK_MASK | QUADSPI_RBCT_RXBRD_USEIPS,
+ base + QUADSPI_RBCT);
+
+ if (op->data.nbytes && op->data.dir == SPI_MEM_DATA_OUT)
+ fsl_qspi_fill_txfifo(q, op);
+
+ err = fsl_qspi_do_op(q, op);
+ }
+
+ mutex_unlock(&q->lock);
+
+ return err;
+}
+
+static int fsl_qspi_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
+{
+ struct fsl_qspi *q = spi_controller_get_devdata(mem->spi->master);
+
+ if (op->data.dir == SPI_MEM_DATA_OUT) {
+ if (op->data.nbytes > q->devtype_data->txfifo)
+ op->data.nbytes = q->devtype_data->txfifo;
+ } else {
+ if (op->data.nbytes > q->devtype_data->ahb_buf_size)
+ op->data.nbytes = q->devtype_data->ahb_buf_size;
+ else if (op->data.nbytes > (q->devtype_data->rxfifo - 4))
+ op->data.nbytes = ALIGN_DOWN(op->data.nbytes, 8);
+ }
+
+ return 0;
+}
+
+static int fsl_qspi_default_setup(struct fsl_qspi *q)
+{
+ void __iomem *base = q->iobase;
+ u32 reg;
+ int ret;
+
+ /* disable and unprepare clock to avoid glitch pass to controller */
+ fsl_qspi_clk_disable_unprep(q);
+
+ /* the default frequency, we will change it later if necessary. */
+ ret = clk_set_rate(q->clk, 66000000);
+ if (ret)
+ return ret;
+
+ ret = fsl_qspi_clk_prep_enable(q);
+ if (ret)
+ return ret;
+
+ /* Reset the module */
+ q->write(QUADSPI_MCR_SWRSTSD_MASK | QUADSPI_MCR_SWRSTHD_MASK,
+ base + QUADSPI_MCR);
+ udelay(1);
+
+ /* Disable the module */
+ q->write(QUADSPI_MCR_MDIS_MASK | QUADSPI_MCR_RESERVED_MASK,
+ base + QUADSPI_MCR);
+
+ reg = q->read(base + QUADSPI_SMPR);
+ q->write(reg & ~(QUADSPI_SMPR_FSDLY_MASK
+ | QUADSPI_SMPR_FSPHS_MASK
+ | QUADSPI_SMPR_HSENA_MASK
+ | QUADSPI_SMPR_DDRSMP_MASK), base + QUADSPI_SMPR);
+
+ /* We only use the buffer3 for AHB read */
+ q->write(0, base + QUADSPI_BUF0IND);
+ q->write(0, base + QUADSPI_BUF1IND);
+ q->write(0, base + QUADSPI_BUF2IND);
+
+ q->write(SEQID_LUT << QUADSPI_BFGENCR_SEQID_SHIFT,
+ q->iobase + QUADSPI_BFGENCR);
+ q->write(QUADSPI_RBCT_WMRK_MASK, base + QUADSPI_RBCT);
+ q->write(QUADSPI_BUF3CR_ALLMST_MASK |
+ ((q->devtype_data->ahb_buf_size / 8) <<
+ QUADSPI_BUF3CR_ADATSZ_SHIFT),
+ base + QUADSPI_BUF3CR);
+
+ q->selected = -1;
+ q->seq = 0;
+
+ /* Enable the module */
+ q->write(QUADSPI_MCR_RESERVED_MASK | QUADSPI_MCR_END_CFG_MASK,
+ base + QUADSPI_MCR);
+
+ /* clear all interrupt status */
+ q->write(0xffffffff, q->iobase + QUADSPI_FR);
+
+ /* enable the interrupt */
+ q->write(QUADSPI_RSER_TFIE, q->iobase + QUADSPI_RSER);
+
+ return 0;
+}
+
+static const char *fsl_qspi_get_name(struct spi_mem *mem)
+{
+ struct fsl_qspi *q = spi_controller_get_devdata(mem->spi->master);
+ struct device *dev = &mem->spi->dev;
+ const char *name;
+
+ /*
+ * In order to keep mtdparts compatible with the old MTD driver at
+ * mtd/spi-nor/fsl-quadspi.c, we set a custom name derived from the
+ * platform_device of the controller.
+ */
+ if (of_get_available_child_count(q->dev->of_node) == 1)
+ return dev_name(q->dev);
+
+ name = devm_kasprintf(dev, GFP_KERNEL,
+ "%s-%d", dev_name(q->dev),
+ mem->spi->chip_select);
+
+ if (!name) {
+ dev_err(dev, "failed to get memory for custom flash name\n");
+ return ERR_PTR(-ENOMEM);
+ }
+
+ return name;
+}
+
+static const struct spi_controller_mem_ops fsl_qspi_mem_ops = {
+ .adjust_op_size = fsl_qspi_adjust_op_size,
+ .supports_op = fsl_qspi_supports_op,
+ .exec_op = fsl_qspi_exec_op,
+ .get_name = fsl_qspi_get_name,
+};
+
+static int fsl_qspi_probe(struct platform_device *pdev)
+{
+ struct spi_controller *ctlr;
+ struct device *dev = &pdev->dev;
+ struct device_node *np = dev->of_node;
+ struct resource *res;
+ struct fsl_qspi *q;
+ int ret;
+
+ ctlr = spi_alloc_master(&pdev->dev, sizeof(*q));
+ if (!ctlr)
+ return -ENOMEM;
+
+ ctlr->mode_bits = SPI_RX_DUAL | SPI_RX_QUAD |
+ SPI_TX_DUAL | SPI_TX_QUAD;
+
+ q = spi_controller_get_devdata(ctlr);
+ q->dev = dev;
+ q->devtype_data = of_device_get_match_data(dev);
+ if (!q->devtype_data) {
+ ret = -ENODEV;
+ goto err_put_ctrl;
+ }
+
+ platform_set_drvdata(pdev, q);
+
+ /* find the resources */
+ res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "QuadSPI");
+ q->iobase = devm_ioremap_resource(dev, res);
+ if (IS_ERR(q->iobase)) {
+ ret = PTR_ERR(q->iobase);
+ goto err_put_ctrl;
+ }
+
+ /*
+ * R/W functions for big- or little-endian registers:
+ * The QSPI controller's endianness is independent of
+ * the CPU core's endianness. So far, although the CPU
+ * core is little-endian the QSPI controller can use
+ * big-endian or little-endian.
+ */
+ if (of_property_read_bool(np, "big-endian")) {
+ q->write = qspi_writel_be;
+ q->read = qspi_readl_be;
+ } else {
+ q->write = qspi_writel;
+ q->read = qspi_readl;
+ }
+
+ res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
+ "QuadSPI-memory");
+ q->ahb_addr = devm_ioremap_resource(dev, res);
+ if (IS_ERR(q->ahb_addr)) {
+ ret = PTR_ERR(q->ahb_addr);
+ goto err_put_ctrl;
+ }
+
+ q->memmap_phy = res->start;
+
+ /* find the clocks */
+ q->clk_en = devm_clk_get(dev, "qspi_en");
+ if (IS_ERR(q->clk_en)) {
+ ret = PTR_ERR(q->clk_en);
+ goto err_put_ctrl;
+ }
+
+ q->clk = devm_clk_get(dev, "qspi");
+ if (IS_ERR(q->clk)) {
+ ret = PTR_ERR(q->clk);
+ goto err_put_ctrl;
+ }
+
+ ret = fsl_qspi_clk_prep_enable(q);
+ if (ret) {
+ dev_err(dev, "can not enable the clock\n");
+ goto err_put_ctrl;
+ }
+
+ /* find the irq */
+ ret = platform_get_irq(pdev, 0);
+ if (ret < 0) {
+ dev_err(dev, "failed to get the irq: %d\n", ret);
+ goto err_disable_clk;
+ }
+
+ ret = devm_request_irq(dev, ret,
+ fsl_qspi_irq_handler, 0, pdev->name, q);
+ if (ret) {
+ dev_err(dev, "failed to request irq: %d\n", ret);
+ goto err_disable_clk;
+ }
+
+ mutex_init(&q->lock);
+
+ ctlr->bus_num = -1;
+ ctlr->num_chipselect = 4;
+ ctlr->mem_ops = &fsl_qspi_mem_ops;
+
+ fsl_qspi_default_setup(q);
+
+ ctlr->dev.of_node = np;
+
+ ret = spi_register_controller(ctlr);
+ if (ret)
+ goto err_destroy_mutex;
+
+ return 0;
+
+err_destroy_mutex:
+ mutex_destroy(&q->lock);
+
+err_disable_clk:
+ fsl_qspi_clk_disable_unprep(q);
+
+err_put_ctrl:
+ spi_controller_put(ctlr);
+
+ dev_err(dev, "Freescale QuadSPI probe failed\n");
+ return ret;
+}
+
+static int fsl_qspi_remove(struct platform_device *pdev)
+{
+ struct fsl_qspi *q = platform_get_drvdata(pdev);
+
+ /* disable the hardware */
+ q->write(QUADSPI_MCR_MDIS_MASK, q->iobase + QUADSPI_MCR);
+ q->write(0x0, q->iobase + QUADSPI_RSER);
+
+ fsl_qspi_clk_disable_unprep(q);
+
+ mutex_destroy(&q->lock);
+
+ return 0;
+}
+
+static int fsl_qspi_suspend(struct device *dev)
+{
+ return 0;
+}
+
+static int fsl_qspi_resume(struct device *dev)
+{
+ struct fsl_qspi *q = dev_get_drvdata(dev);
+
+ fsl_qspi_default_setup(q);
+
+ return 0;
+}
+
+static const struct of_device_id fsl_qspi_dt_ids[] = {
+ { .compatible = "fsl,vf610-qspi", .data = &vybrid_data, },
+ { .compatible = "fsl,imx6sx-qspi", .data = &imx6sx_data, },
+ { .compatible = "fsl,imx7d-qspi", .data = &imx7d_data, },
+ { .compatible = "fsl,imx6ul-qspi", .data = &imx6ul_data, },
+ { .compatible = "fsl,ls1021a-qspi", .data = &ls1021a_data, },
+ { .compatible = "fsl,ls2080a-qspi", .data = &ls2080a_data, },
+ { /* sentinel */ }
+};
+MODULE_DEVICE_TABLE(of, fsl_qspi_dt_ids);
+
+static const struct dev_pm_ops fsl_qspi_pm_ops = {
+ .suspend = fsl_qspi_suspend,
+ .resume = fsl_qspi_resume,
+};
+
+static struct platform_driver fsl_qspi_driver = {
+ .driver = {
+ .name = "fsl-quadspi",
+ .of_match_table = fsl_qspi_dt_ids,
+ .pm = &fsl_qspi_pm_ops,
+ },
+ .probe = fsl_qspi_probe,
+ .remove = fsl_qspi_remove,
+};
+module_platform_driver(fsl_qspi_driver);
+
+MODULE_DESCRIPTION("Freescale QuadSPI Controller Driver");
+MODULE_AUTHOR("Freescale Semiconductor Inc.");
+MODULE_AUTHOR("Boris Brezillion <***@bootlin.com>");
+MODULE_AUTHOR("Frieder Schrempf <***@exceet.de>");
+MODULE_AUTHOR("Yogesh Gaur <***@nxp.com>");
+MODULE_AUTHOR("Suresh Gupta <***@nxp.com>");
+MODULE_LICENSE("GPL v2");
--
2.7.4
Frieder Schrempf
2018-08-02 13:09:03 UTC
Permalink
Ping.

I'm not sure if my message below went out to you at all. At least I
can't find it in the ML archive.

I still hope someone can help with the questions below.

Meanwhile for the second point I did some tests myself with one chip on
each of the two buses and it worked fine with my latest v2 patches.
So I'm not sure at all why Yogesh has problems with his setup (two chips
on the first bus).

Thanks,
Frieder
Hi guys from NXP,
we still have some issues/questions concerning the FSL QSPI IP, that
somehow block the new driver under the SPI memory framework [1].
If anyone of you could comment on these, or help finding solutions, this
would be very much appreciated.
1. The SPI NOR driver was using a reset of the flash and AHB domain to
invalidate the AHB buffer [2]. But this needs quite a lot of time, so we
have a hack to use two regions in memory and switch between them
alternately to invalidate the cache [3]. As this is not so nice, do you
know of any other possibility to invalidate the flash?
2. We tried to reuse the mapped memory to access different chips on
different CS, by switching the QUADSPI_SFA1AD, QUADSPI_SFA2AD, etc.
accordingly [4]. This doesn't work as expected, but Yogesh found out,
that it works with a fixed memory map for each CS and by adding a
ioremap(), as the SPI NOR driver does [5]. Can someone explain this
behavior and why the ioremap is needed?
3. In case of a NOR page program operation, the driver is expected to
write the full page, even if it's larger than the TX buffer size. Boris
explained this here: [6].
So we need to find a way of triggering a refill of the TX buffer by CPU
or DMA if possible.
Thank you in advance for your help,
Frieder
[1] https://patchwork.ozlabs.org/cover/939864/
[2]
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/mtd/spi-nor/fsl-quadspi.c?h=v4.18-rc4#n591
[3]
https://github.com/fschrempf/linux/blob/fsl-qspi-next-2/drivers/spi/spi-fsl-qspi.c#L511
[4]
https://github.com/fschrempf/linux/blob/fsl-qspi-next-2/drivers/spi/spi-fsl-qspi.c#L476
[5]
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/mtd/spi-nor/fsl-quadspi.c?h=v4.18-rc4#n906
[6] https://patchwork.ozlabs.org/patch/928677/#1950278
Han Xu
2018-08-02 21:58:48 UTC
Permalink
-----Original Message-----
Sent: Thursday, August 2, 2018 8:09 AM
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Ping.
I'm not sure if my message below went out to you at all. At least I can't find it
in the ML archive.
I still hope someone can help with the questions below.
Meanwhile for the second point I did some tests myself with one chip on
each of the two buses and it worked fine with my latest v2 patches.
So I'm not sure at all why Yogesh has problems with his setup (two chips on
the first bus).
Tried to test the v2 patch set on i.MX6SX SDB board but get the memory map failure.

[ 1.298633] fsl-quadspi 21e4000.qspi: ioremap failed for resource [mem 0x70000000-0x7fffffff]
[ 1.307330] fsl-quadspi 21e4000.qspi: Freescale QuadSPI probe failed
[ 1.313922] fsl-quadspi: probe of 21e4000.qspi failed with error -12

This is the reason why dynamic ioremap added in previous driver, please refer to
https://patchwork.ozlabs.org/patch/503655/

I cannot really test the functionality in this case, will get another platform without such issue and retry.
Thanks,
Frieder
Hi guys from NXP,
we still have some issues/questions concerning the FSL QSPI IP, that
somehow block the new driver under the SPI memory framework [1].
If anyone of you could comment on these, or help finding solutions,
this would be very much appreciated.
1. The SPI NOR driver was using a reset of the flash and AHB domain to
invalidate the AHB buffer [2]. But this needs quite a lot of time, so
we have a hack to use two regions in memory and switch between them
alternately to invalidate the cache [3]. As this is not so nice, do
you know of any other possibility to invalidate the flash?
2. We tried to reuse the mapped memory to access different chips on
different CS, by switching the QUADSPI_SFA1AD, QUADSPI_SFA2AD, etc.
accordingly [4]. This doesn't work as expected, but Yogesh found out,
that it works with a fixed memory map for each CS and by adding a
ioremap(), as the SPI NOR driver does [5]. Can someone explain this
behavior and why the ioremap is needed?
3. In case of a NOR page program operation, the driver is expected to
write the full page, even if it's larger than the TX buffer size.
Boris explained this here: [6].
So we need to find a way of triggering a refill of the TX buffer by
CPU or DMA if possible.
Thank you in advance for your help,
Frieder
[1]
https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpat
chwork.ozlabs.org%2Fcover%2F939864%2F&amp;data=02%7C01%7Chan.xu
%40nxp.
com%7C567dbf7050bc4b38dbea08d5f87925db%7C686ea1d3bc2b4c6fa92cd9
9c5c301
635%7C0%7C0%7C636688121577481337&amp;sdata=d%2BRazwVOV0z0wJ%
2BHrOthM6a
50W0o8omAUOQ%2F3Ji%2Fs6M%3D&amp;reserved=0
[2]
https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgit
.kernel.org%2Fpub%2Fscm%2Flinux%2Fkernel%2Fgit%2Ftorvalds%2Flinux.g
it%
2Ftree%2Fdrivers%2Fmtd%2Fspi-nor%2Ffsl-quadspi.c%3Fh%3Dv4.18-
rc4%23n59
1&amp;data=02%7C01%7Chan.xu%40nxp.com%7C567dbf7050bc4b38dbea0
8d5f87925
db%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C63668812157748
1337&amp
;sdata=48yLzR8tBHRgLh1rN1Nw1L45%2FNCe4m6idmzvpbi5jpc%3D&amp;re
served=0
[3]
https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgit
hub.com%2Ffschrempf%2Flinux%2Fblob%2Ffsl-qspi-next-
2%2Fdrivers%2Fspi%2
Fspi-fsl-
qspi.c%23L511&amp;data=02%7C01%7Chan.xu%40nxp.com%7C567dbf705
0bc4b38dbea08d5f87925db%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%
7C0%7C6
36688121577481337&amp;sdata=lf0Xwxr6RYrs5YHIIf9TO5RbyEgtr1qMus7p0
vfUgh
c%3D&amp;reserved=0
[4]
https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgit
hub.com%2Ffschrempf%2Flinux%2Fblob%2Ffsl-qspi-next-
2%2Fdrivers%2Fspi%2
Fspi-fsl-
qspi.c%23L476&amp;data=02%7C01%7Chan.xu%40nxp.com%7C567dbf705
0bc4b38dbea08d5f87925db%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%
7C0%7C6
36688121577481337&amp;sdata=NYmMKaTsff7%2FZ%2BO7nRjnsIWDFDrAN
dXZfgpn0H
b6Nmw%3D&amp;reserved=0
[5]
https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgit
.kernel.org%2Fpub%2Fscm%2Flinux%2Fkernel%2Fgit%2Ftorvalds%2Flinux.g
it%
2Ftree%2Fdrivers%2Fmtd%2Fspi-nor%2Ffsl-quadspi.c%3Fh%3Dv4.18-
rc4%23n90
6&amp;data=02%7C01%7Chan.xu%40nxp.com%7C567dbf7050bc4b38dbea0
8d5f87925
db%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C63668812157748
1337&amp
;sdata=BSLjT%2BlwxocGJT3eJ1pUIYxsrBP%2FCdSt0Mthb%2Bx3ySE%3D&am
p;reserv
ed=0
[6]
https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpat
chwork.ozlabs.org%2Fpatch%2F928677%2F%231950278&amp;data=02%7C0
1%7Chan
.xu%40nxp.com%7C567dbf7050bc4b38dbea08d5f87925db%7C686ea1d3bc
2b4c6fa92
cd99c5c301635%7C0%7C0%7C636688121577491349&amp;sdata=LPUHvPVnX
Qvp7OfIl
yH26FADDXqJI7%2B%2FsUClsfrGx7k%3D&amp;reserved=0
Boris Brezillon
2018-08-04 13:37:28 UTC
Permalink
Hi Han,

On Thu, 2 Aug 2018 21:58:48 +0000
Post by Han Xu
-----Original Message-----
Sent: Thursday, August 2, 2018 8:09 AM
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Ping.
I'm not sure if my message below went out to you at all. At least I can't find it
in the ML archive.
I still hope someone can help with the questions below.
Meanwhile for the second point I did some tests myself with one chip on
each of the two buses and it worked fine with my latest v2 patches.
So I'm not sure at all why Yogesh has problems with his setup (two chips on
the first bus).
Tried to test the v2 patch set on i.MX6SX SDB board but get the memory map failure.
[ 1.298633] fsl-quadspi 21e4000.qspi: ioremap failed for resource [mem 0x70000000-0x7fffffff]
[ 1.307330] fsl-quadspi 21e4000.qspi: Freescale QuadSPI probe failed
[ 1.313922] fsl-quadspi: probe of 21e4000.qspi failed with error -12
This is the reason why dynamic ioremap added in previous driver, please refer to
https://patchwork.ozlabs.org/patch/503655/
We can reduce the size of the iomap to 2k * 4, since this is all we use
currently. Can you try to change the size of the ioremap call to 16k and
tell us if it works.

Unrelated to this issue, we still have 2 questions left unanswered:

1/ is there an easy way to invalidate AHB buffers? I mean, not
something that implies a full reset + several milliseconds of delay
after the reset. Right now we trick the caching logic by mapping a
portion that is twice the size of the buffer and switching from one
sub-portion to this other to trigger a real read on each read
access, but that's hack-ish, and I'd be surprised if HW
engineers hadn't planned for this "manual AHB buffer flush" case.

2/ if we use DMA, do you know what happens when the TX FIFO runs out
of data while the TX request is not finished yet. In PIO mode, it
seems the engine sends garbage on the bus when that happens, and we
definitely don't want that.

While #1 is not blocking us, #2 is if we don't have those patches
[1][2] applied, and Marek wanted to be sure there was no other
ways to solve the "TX FIFO starvation" issue before considering these
changes. So that'd be great if someone from NXP could have a look/ask
around and give us answers to those 2 questions.

Thanks,

Boris

[1]http://patchwork.ozlabs.org/patch/928677/
[2]http://patchwork.ozlabs.org/patch/928678/
Frieder Schrempf
2018-09-03 09:02:19 UTC
Permalink
Hi Han,
Post by Boris Brezillon
Hi Han,
On Thu, 2 Aug 2018 21:58:48 +0000
Post by Han Xu
-----Original Message-----
Sent: Thursday, August 2, 2018 8:09 AM
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Ping.
I'm not sure if my message below went out to you at all. At least I can't find it
in the ML archive.
I still hope someone can help with the questions below.
Meanwhile for the second point I did some tests myself with one chip on
each of the two buses and it worked fine with my latest v2 patches.
So I'm not sure at all why Yogesh has problems with his setup (two chips on
the first bus).
Tried to test the v2 patch set on i.MX6SX SDB board but get the memory map failure.
[ 1.298633] fsl-quadspi 21e4000.qspi: ioremap failed for resource [mem 0x70000000-0x7fffffff]
[ 1.307330] fsl-quadspi 21e4000.qspi: Freescale QuadSPI probe failed
[ 1.313922] fsl-quadspi: probe of 21e4000.qspi failed with error -12
This is the reason why dynamic ioremap added in previous driver, please refer to
https://patchwork.ozlabs.org/patch/503655/
We can reduce the size of the iomap to 2k * 4, since this is all we use
currently. Can you try to change the size of the ioremap call to 16k and
tell us if it works.
Were you able to test with the reduced iomap size?
It would be great to know if it works on your board.

Thanks,
Frieder
Post by Boris Brezillon
1/ is there an easy way to invalidate AHB buffers? I mean, not
something that implies a full reset + several milliseconds of delay
after the reset. Right now we trick the caching logic by mapping a
portion that is twice the size of the buffer and switching from one
sub-portion to this other to trigger a real read on each read
access, but that's hack-ish, and I'd be surprised if HW
engineers hadn't planned for this "manual AHB buffer flush" case.
2/ if we use DMA, do you know what happens when the TX FIFO runs out
of data while the TX request is not finished yet. In PIO mode, it
seems the engine sends garbage on the bus when that happens, and we
definitely don't want that.
While #1 is not blocking us, #2 is if we don't have those patches
[1][2] applied, and Marek wanted to be sure there was no other
ways to solve the "TX FIFO starvation" issue before considering these
changes. So that'd be great if someone from NXP could have a look/ask
around and give us answers to those 2 questions.
Thanks,
Boris
[1]http://patchwork.ozlabs.org/patch/928677/
[2]http://patchwork.ozlabs.org/patch/928678/
Han Xu
2018-09-12 17:04:21 UTC
Permalink
-----Original Message-----
Sent: Monday, September 3, 2018 4:02 AM
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Hi Han,
Post by Boris Brezillon
Hi Han,
On Thu, 2 Aug 2018 21:58:48 +0000
Post by Han Xu
-----Original Message-----
Sent: Thursday, August 2, 2018 8:09 AM
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Ping.
I'm not sure if my message below went out to you at all. At least I
can't find it in the ML archive.
I still hope someone can help with the questions below.
Meanwhile for the second point I did some tests myself with one chip
on each of the two buses and it worked fine with my latest v2 patches.
So I'm not sure at all why Yogesh has problems with his setup (two
chips on the first bus).
Tried to test the v2 patch set on i.MX6SX SDB board but get the memory
map failure.
Post by Boris Brezillon
Post by Han Xu
[ 1.298633] fsl-quadspi 21e4000.qspi: ioremap failed for resource [mem
0x70000000-0x7fffffff]
Post by Boris Brezillon
Post by Han Xu
[ 1.307330] fsl-quadspi 21e4000.qspi: Freescale QuadSPI probe failed
[ 1.313922] fsl-quadspi: probe of 21e4000.qspi failed with error -12
This is the reason why dynamic ioremap added in previous driver, please
refer to
https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpa
tchwork.ozlabs.org%2Fpatch%2F503655%2F&amp;data=02%7C01%7Chan.xu
%40nx
p.com%7C9f45a8b666d3478f065408d6117bf524%7C686ea1d3bc2b4c6fa92cd9
9c5c
301635%7C0%7C0%7C636715621426190473&amp;sdata=XWPfWe%2Fu2ePW
mNPe179D0
Post by Boris Brezillon
Post by Han Xu
vjTp6eLp0%2FJRF2vRayDwug%3D&amp;reserved=0
We can reduce the size of the iomap to 2k * 4, since this is all we
use currently. Can you try to change the size of the ioremap call to
16k and tell us if it works.
Were you able to test with the reduced iomap size?
It would be great to know if it works on your board.
Thanks,
Frieder
Test the code on i.MX6SX sabreauto board with two micron n25q256a chips on two CS.
First issue found is __div0 kernel dump with these code
/* Max 64 dummy clock cycles supported */
if (op->dummy.nbytes * 8 / op->dummy.buswidth > 64)
dummy.buswidth was not set during read id.

Second issue is the second part failed to be probed, tried both buswidth 4 and buswidth 1.

[ 1.364979] m25p80 spi5.0: found n25q256a, expected m25p80
[ 1.370986] m25p80 spi5.0: n25q256a (32768 Kbytes)
[ 1.381020] m25p80 spi5.1: unrecognized JEDEC id bytes: ff, ff, ff

These are the DT settings:
&qspi1 {
pinctrl-names = "default";
pinctrl-0 = <&pinctrl_qspi1_1>;
status = "okay";

flash0: ***@0 {
¦ #address-cells = <1>;
¦ #size-cells = <1>;
¦ compatible = "micron,m25p80";
¦ spi-max-frequency = <29000000>;
spi-rx-bus-width = <4>;
spi-tx-bus-width = <4>;
¦ reg = <0>;
};

flash1: ***@1 {
¦ #address-cells = <1>;
¦ #size-cells = <1>;
¦ compatible = "micron,m25p80";
¦ spi-max-frequency = <29000000>;
spi-rx-bus-width = <4>;
spi-tx-bus-width = <4>;
¦ reg = <1>;
};
};
Post by Boris Brezillon
1/ is there an easy way to invalidate AHB buffers? I mean, not
something that implies a full reset + several milliseconds of delay
after the reset. Right now we trick the caching logic by mapping a
portion that is twice the size of the buffer and switching from one
sub-portion to this other to trigger a real read on each read
access, but that's hack-ish, and I'd be surprised if HW
engineers hadn't planned for this "manual AHB buffer flush" case.
2/ if we use DMA, do you know what happens when the TX FIFO runs out
of data while the TX request is not finished yet. In PIO mode, it
seems the engine sends garbage on the bus when that happens, and we
definitely don't want that.
While #1 is not blocking us, #2 is if we don't have those patches
[1][2] applied, and Marek wanted to be sure there was no other ways to
solve the "TX FIFO starvation" issue before considering these changes.
So that'd be great if someone from NXP could have a look/ask around
and give us answers to those 2 questions.
Thanks,
Boris
[1]https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fp
atchwork.ozlabs.org%2Fpatch%2F928677%2F&amp;data=02%7C01%7Chan.x
u%40nx
p.com%7C9f45a8b666d3478f065408d6117bf524%7C686ea1d3bc2b4c6fa92cd9
9c5c3
01635%7C0%7C0%7C636715621426190473&amp;sdata=p%2FJBfxNd2Swlmrrr
H9Xk2R2
Post by Boris Brezillon
DDl%2BshfXdUoyXAy4bXBc%3D&amp;reserved=0
[2]https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fp
atchwork.ozlabs.org%2Fpatch%2F928678%2F&amp;data=02%7C01%7Chan.x
u%40nx
p.com%7C9f45a8b666d3478f065408d6117bf524%7C686ea1d3bc2b4c6fa92cd9
9c5c3
01635%7C0%7C0%7C636715621426190473&amp;sdata=xo2l3Ld3n9mkO5PTW
5DOqP7u0
Post by Boris Brezillon
%2Bgch0cXs7jEjHVokqQ%3D&amp;reserved=0
Frieder Schrempf
2018-09-12 18:40:39 UTC
Permalink
Hi Han,
Post by Han Xu
-----Original Message-----
Sent: Monday, September 3, 2018 4:02 AM
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Hi Han,
Post by Boris Brezillon
Hi Han,
On Thu, 2 Aug 2018 21:58:48 +0000
Post by Han Xu
-----Original Message-----
Sent: Thursday, August 2, 2018 8:09 AM
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Ping.
I'm not sure if my message below went out to you at all. At least I
can't find it in the ML archive.
I still hope someone can help with the questions below.
Meanwhile for the second point I did some tests myself with one chip
on each of the two buses and it worked fine with my latest v2 patches.
So I'm not sure at all why Yogesh has problems with his setup (two
chips on the first bus).
Tried to test the v2 patch set on i.MX6SX SDB board but get the memory
map failure.
Post by Boris Brezillon
Post by Han Xu
[ 1.298633] fsl-quadspi 21e4000.qspi: ioremap failed for resource [mem
0x70000000-0x7fffffff]
Post by Boris Brezillon
Post by Han Xu
[ 1.307330] fsl-quadspi 21e4000.qspi: Freescale QuadSPI probe failed
[ 1.313922] fsl-quadspi: probe of 21e4000.qspi failed with error -12
This is the reason why dynamic ioremap added in previous driver, please
refer to
https://smex12-5-en-ctp.trendmicro.com:443/wis/clicktime/v1/query?url=https%3a%2f%2femea01.safelinks.protection.outlook.com%2f%3furl%3dhttps%253A%252F%252Fpa&umid=d6cc1014-1848-42fb-92fd-9626d45c8050&auth=541e45255b6517100d80c2b1b80b6933b203c492-5aa8e1977a9db94300a9f61f5446e7a21b175f56
tchwork.ozlabs.org%2Fpatch%2F503655%2F&amp;data=02%7C01%7Chan.xu
%40nx
p.com%7C9f45a8b666d3478f065408d6117bf524%7C686ea1d3bc2b4c6fa92cd9
9c5c
301635%7C0%7C0%7C636715621426190473&amp;sdata=XWPfWe%2Fu2ePW
mNPe179D0
Post by Boris Brezillon
Post by Han Xu
vjTp6eLp0%2FJRF2vRayDwug%3D&amp;reserved=0
We can reduce the size of the iomap to 2k * 4, since this is all we
use currently. Can you try to change the size of the ioremap call to
16k and tell us if it works.
Were you able to test with the reduced iomap size?
It would be great to know if it works on your board.
Thanks,
Frieder
Test the code on i.MX6SX sabreauto board with two micron n25q256a chips on two CS.
First issue found is __div0 kernel dump with these code
/* Max 64 dummy clock cycles supported */
if (op->dummy.nbytes * 8 / op->dummy.buswidth > 64)
dummy.buswidth was not set during read id.
First, thank you for coming back to this and doing the test.
I'm currently not sure about the reason for this, but I guess Boris will
figure it out easily ;)
Post by Han Xu
Second issue is the second part failed to be probed, tried both buswidth 4 and buswidth 1.
[ 1.364979] m25p80 spi5.0: found n25q256a, expected m25p80
[ 1.370986] m25p80 spi5.0: n25q256a (32768 Kbytes)
[ 1.381020] m25p80 spi5.1: unrecognized JEDEC id bytes: ff, ff, ff
&qspi1 {
pinctrl-names = "default";
pinctrl-0 = <&pinctrl_qspi1_1>;
status = "okay";
¦ #address-cells = <1>;
¦ #size-cells = <1>;
¦ compatible = "micron,m25p80";
¦ spi-max-frequency = <29000000>;
spi-rx-bus-width = <4>;
spi-tx-bus-width = <4>;
¦ reg = <0>;
};
¦ #address-cells = <1>;
¦ #size-cells = <1>;
¦ compatible = "micron,m25p80";
¦ spi-max-frequency = <29000000>;
spi-rx-bus-width = <4>;
spi-tx-bus-width = <4>;
¦ reg = <1>;
};
};
First, I think you should add "jedec,spi-nor" to your compatible properties.

Second, are you sure, that the two chips are both on QSPIA using the two
chip selects?
I have no schematics of the board, but if I look at the devicetree in
the linux-imx kernel [1] it seems to me that one chip is on QSPIA CS0
and the other on QSPIB CS0.
If this is the case, then you have to set reg = <2> for the second chip.

Thanks,
Frieder

[1]
http://git.freescale.com/git/cgit.cgi/imx/linux-imx.git/tree/arch/arm/boot/dts/imx6sx-sabreauto.dts?h=imx_4.9.11_1.0.0_ga#n771
Post by Han Xu
Post by Boris Brezillon
1/ is there an easy way to invalidate AHB buffers? I mean, not
something that implies a full reset + several milliseconds of delay
after the reset. Right now we trick the caching logic by mapping a
portion that is twice the size of the buffer and switching from one
sub-portion to this other to trigger a real read on each read
access, but that's hack-ish, and I'd be surprised if HW
engineers hadn't planned for this "manual AHB buffer flush" case.
2/ if we use DMA, do you know what happens when the TX FIFO runs out
of data while the TX request is not finished yet. In PIO mode, it
seems the engine sends garbage on the bus when that happens, and we
definitely don't want that.
While #1 is not blocking us, #2 is if we don't have those patches
[1][2] applied, and Marek wanted to be sure there was no other ways to
solve the "TX FIFO starvation" issue before considering these changes.
So that'd be great if someone from NXP could have a look/ask around
and give us answers to those 2 questions.
Thanks,
Boris
Han Xu
2018-09-12 21:04:19 UTC
Permalink
-----Original Message-----
Sent: Wednesday, September 12, 2018 1:41 PM
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Hi Han,
Post by Han Xu
-----Original Message-----
Sent: Monday, September 3, 2018 4:02 AM
Gaur
Post by Han Xu
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Hi Han,
Post by Boris Brezillon
Hi Han,
On Thu, 2 Aug 2018 21:58:48 +0000
Post by Han Xu
-----Original Message-----
Sent: Thursday, August 2, 2018 8:09 AM
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Ping.
I'm not sure if my message below went out to you at all. At least
I can't find it in the ML archive.
I still hope someone can help with the questions below.
Meanwhile for the second point I did some tests myself with one
chip on each of the two buses and it worked fine with my latest v2
patches.
Post by Han Xu
Post by Boris Brezillon
Post by Han Xu
So I'm not sure at all why Yogesh has problems with his setup (two
chips on the first bus).
Tried to test the v2 patch set on i.MX6SX SDB board but get the memory
map failure.
Post by Boris Brezillon
Post by Han Xu
[ 1.298633] fsl-quadspi 21e4000.qspi: ioremap failed for resource
[mem
Post by Han Xu
0x70000000-0x7fffffff]
Post by Boris Brezillon
Post by Han Xu
[ 1.307330] fsl-quadspi 21e4000.qspi: Freescale QuadSPI probe failed
[ 1.313922] fsl-quadspi: probe of 21e4000.qspi failed with error -12
This is the reason why dynamic ioremap added in previous driver, please
refer to
https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fsm
Post by Han Xu
ex12-5-en-
ctp.trendmicro.com%3A443%2Fwis%2Fclicktime%2Fv1%2Fquery%3Fu
rl%3Dhttps%253a%252f%252femea01.safelinks.protection.outlook.com%252
f
Post by Han Xu
%253furl%253dhttps%25253A%25252F%25252Fpa%26umid%3Dd6cc1014-
1848-42fb
Post by Han Xu
-92fd-
9626d45c8050%26auth%3D541e45255b6517100d80c2b1b80b6933b203c492-
5aa8e1977a9db94300a9f61f5446e7a21b175f56&amp;data=02%7C01%7Chan.x
u%40
nxp.com%7C9d43e9b29a82419a36c408d618df3dc3%7C686ea1d3bc2b4c6fa92c
d99c
5c301635%7C0%7C0%7C636723744426896223&amp;sdata=Y32I9H9adPn%2Bn
lcICuV
Post by Han Xu
M8Uoozsig%2BM3F0rNAhkIF5ZE%3D&amp;reserved=0
tchwork.ozlabs.org%2Fpatch%2F503655%2F&amp;data=02%7C01%7Chan.xu
Post by Han Xu
%40nx
p.com%7C9f45a8b666d3478f065408d6117bf524%7C686ea1d3bc2b4c6fa92cd9
Post by Han Xu
9c5c
301635%7C0%7C0%7C636715621426190473&amp;sdata=XWPfWe%2Fu2ePW
Post by Han Xu
mNPe179D0
Post by Boris Brezillon
Post by Han Xu
vjTp6eLp0%2FJRF2vRayDwug%3D&amp;reserved=0
We can reduce the size of the iomap to 2k * 4, since this is all we
use currently. Can you try to change the size of the ioremap call to
16k and tell us if it works.
Were you able to test with the reduced iomap size?
It would be great to know if it works on your board.
Thanks,
Frieder
Test the code on i.MX6SX sabreauto board with two micron n25q256a chips
on two CS.
Post by Han Xu
First issue found is __div0 kernel dump with these code
/* Max 64 dummy clock cycles supported */ if (op->dummy.nbytes * 8 /
op->dummy.buswidth > 64) dummy.buswidth was not set during read id.
First, thank you for coming back to this and doing the test.
I'm currently not sure about the reason for this, but I guess Boris will figure it
out easily ;)
Post by Han Xu
Second issue is the second part failed to be probed, tried both buswidth 4
and buswidth 1.
Post by Han Xu
[ 1.364979] m25p80 spi5.0: found n25q256a, expected m25p80
[ 1.370986] m25p80 spi5.0: n25q256a (32768 Kbytes)
[ 1.381020] m25p80 spi5.1: unrecognized JEDEC id bytes: ff, ff, ff
&qspi1 {
pinctrl-names = "default";
pinctrl-0 = <&pinctrl_qspi1_1>;
status = "okay";
¦ #address-cells = <1>;
¦ #size-cells = <1>;
¦ compatible = "micron,m25p80";
¦ spi-max-frequency = <29000000>;
spi-rx-bus-width = <4>;
spi-tx-bus-width = <4>;
¦ reg = <0>;
};
¦ #address-cells = <1>;
¦ #size-cells = <1>;
¦ compatible = "micron,m25p80";
¦ spi-max-frequency = <29000000>;
spi-rx-bus-width = <4>;
spi-tx-bus-width = <4>;
¦ reg = <1>;
};
};
First, I think you should add "jedec,spi-nor" to your compatible properties.
Second, are you sure, that the two chips are both on QSPIA using the two
chip selects?
I have no schematics of the board, but if I look at the devicetree in the linux-
imx kernel [1] it seems to me that one chip is on QSPIA CS0 and the other on
QSPIB CS0.
If this is the case, then you have to set reg = <2> for the second chip.
Yes, you are right, the second chip connects to QSPIB CS0, with the DT change, both of them work fine with __div0 issue workaround.
Thanks,
Frieder
[1]
https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fgit.f
reescale.com%2Fgit%2Fcgit.cgi%2Fimx%2Flinux-
imx.git%2Ftree%2Farch%2Farm%2Fboot%2Fdts%2Fimx6sx-
sabreauto.dts%3Fh%3Dimx_4.9.11_1.0.0_ga%23n771&amp;data=02%7C01%
7Chan.xu%40nxp.com%7C9d43e9b29a82419a36c408d618df3dc3%7C686ea1d
3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C636723744426896223&amp;sdata
=038DdgMLjO23Io3CTHoSVrJL2Ho%2BgxyHlx9%2BLMrLWkQ%3D&amp;reser
ved=0
Post by Han Xu
Post by Boris Brezillon
1/ is there an easy way to invalidate AHB buffers? I mean, not
something that implies a full reset + several milliseconds of delay
after the reset. Right now we trick the caching logic by mapping a
portion that is twice the size of the buffer and switching from one
sub-portion to this other to trigger a real read on each read
access, but that's hack-ish, and I'd be surprised if HW
engineers hadn't planned for this "manual AHB buffer flush" case.
2/ if we use DMA, do you know what happens when the TX FIFO runs
out
Post by Han Xu
Post by Boris Brezillon
of data while the TX request is not finished yet. In PIO mode, it
seems the engine sends garbage on the bus when that happens, and
we
Post by Han Xu
Post by Boris Brezillon
definitely don't want that.
While #1 is not blocking us, #2 is if we don't have those patches
[1][2] applied, and Marek wanted to be sure there was no other ways
to solve the "TX FIFO starvation" issue before considering these changes.
So that'd be great if someone from NXP could have a look/ask around
and give us answers to those 2 questions.
Thanks,
Boris
Frieder Schrempf
2018-09-13 07:00:47 UTC
Permalink
Hi Han,
Post by Han Xu
-----Original Message-----
Sent: Wednesday, September 12, 2018 1:41 PM
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Hi Han,
Post by Han Xu
-----Original Message-----
Sent: Monday, September 3, 2018 4:02 AM
Gaur
Post by Han Xu
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Hi Han,
Post by Boris Brezillon
Hi Han,
On Thu, 2 Aug 2018 21:58:48 +0000
Post by Han Xu
-----Original Message-----
Sent: Thursday, August 2, 2018 8:09 AM
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Ping.
I'm not sure if my message below went out to you at all. At least
I can't find it in the ML archive.
I still hope someone can help with the questions below.
Meanwhile for the second point I did some tests myself with one
chip on each of the two buses and it worked fine with my latest v2
patches.
Post by Han Xu
Post by Boris Brezillon
Post by Han Xu
So I'm not sure at all why Yogesh has problems with his setup (two
chips on the first bus).
Tried to test the v2 patch set on i.MX6SX SDB board but get the memory
map failure.
Post by Boris Brezillon
Post by Han Xu
[ 1.298633] fsl-quadspi 21e4000.qspi: ioremap failed for resource
[mem
Post by Han Xu
0x70000000-0x7fffffff]
Post by Boris Brezillon
Post by Han Xu
[ 1.307330] fsl-quadspi 21e4000.qspi: Freescale QuadSPI probe failed
[ 1.313922] fsl-quadspi: probe of 21e4000.qspi failed with error -12
This is the reason why dynamic ioremap added in previous driver, please
refer to
https://smex12-5-en-ctp.trendmicro.com:443/wis/clicktime/v1/query?url=https%3a%2f%2femea01.safelinks.protection.outlook.com%2f%3furl%3dhttps%253A%252F%252Fsm&umid=aac77072-ee97-4b8a-b8b9-55765e276aec&auth=78a0452d0eda3018cc3487ba1bec995089e109a2-cefb4275e4aea35bf0c475f6bd6da9e9a8c877a0
Post by Han Xu
ex12-5-en-
ctp.trendmicro.com%3A443%2Fwis%2Fclicktime%2Fv1%2Fquery%3Fu
rl%3Dhttps%253a%252f%252femea01.safelinks.protection.outlook.com%252
f
Post by Han Xu
%253furl%253dhttps%25253A%25252F%25252Fpa%26umid%3Dd6cc1014-
1848-42fb
Post by Han Xu
-92fd-
9626d45c8050%26auth%3D541e45255b6517100d80c2b1b80b6933b203c492-
5aa8e1977a9db94300a9f61f5446e7a21b175f56&amp;data=02%7C01%7Chan.x
u%40
nxp.com%7C9d43e9b29a82419a36c408d618df3dc3%7C686ea1d3bc2b4c6fa92c
d99c
5c301635%7C0%7C0%7C636723744426896223&amp;sdata=Y32I9H9adPn%2Bn
lcICuV
Post by Han Xu
M8Uoozsig%2BM3F0rNAhkIF5ZE%3D&amp;reserved=0
tchwork.ozlabs.org%2Fpatch%2F503655%2F&amp;data=02%7C01%7Chan.xu
Post by Han Xu
%40nx
p.com%7C9f45a8b666d3478f065408d6117bf524%7C686ea1d3bc2b4c6fa92cd9
Post by Han Xu
9c5c
301635%7C0%7C0%7C636715621426190473&amp;sdata=XWPfWe%2Fu2ePW
Post by Han Xu
mNPe179D0
Post by Boris Brezillon
Post by Han Xu
vjTp6eLp0%2FJRF2vRayDwug%3D&amp;reserved=0
We can reduce the size of the iomap to 2k * 4, since this is all we
use currently. Can you try to change the size of the ioremap call to
16k and tell us if it works.
Were you able to test with the reduced iomap size?
It would be great to know if it works on your board.
Thanks,
Frieder
Test the code on i.MX6SX sabreauto board with two micron n25q256a chips
on two CS.
Post by Han Xu
First issue found is __div0 kernel dump with these code
/* Max 64 dummy clock cycles supported */ if (op->dummy.nbytes * 8 /
op->dummy.buswidth > 64) dummy.buswidth was not set during read id.
First, thank you for coming back to this and doing the test.
I'm currently not sure about the reason for this, but I guess Boris will figure it
out easily ;)
Ok, on a closer look it is obvious that the reason is this commit [1].
My last test was before this change, when dummy.buswidth was still set
to 1 even if dummy.n_bytes is 0.
Now both are 0 and I need to handle this in the FSL QSPI driver.
Post by Han Xu
Post by Han Xu
Second issue is the second part failed to be probed, tried both buswidth 4
and buswidth 1.
Post by Han Xu
[ 1.364979] m25p80 spi5.0: found n25q256a, expected m25p80
[ 1.370986] m25p80 spi5.0: n25q256a (32768 Kbytes)
[ 1.381020] m25p80 spi5.1: unrecognized JEDEC id bytes: ff, ff, ff
&qspi1 {
pinctrl-names = "default";
pinctrl-0 = <&pinctrl_qspi1_1>;
status = "okay";
¦ #address-cells = <1>;
¦ #size-cells = <1>;
¦ compatible = "micron,m25p80";
¦ spi-max-frequency = <29000000>;
spi-rx-bus-width = <4>;
spi-tx-bus-width = <4>;
¦ reg = <0>;
};
¦ #address-cells = <1>;
¦ #size-cells = <1>;
¦ compatible = "micron,m25p80";
¦ spi-max-frequency = <29000000>;
spi-rx-bus-width = <4>;
spi-tx-bus-width = <4>;
¦ reg = <1>;
};
};
First, I think you should add "jedec,spi-nor" to your compatible properties.
Second, are you sure, that the two chips are both on QSPIA using the two
chip selects?
I have no schematics of the board, but if I look at the devicetree in the linux-
imx kernel [1] it seems to me that one chip is on QSPIA CS0 and the other on
QSPIB CS0.
If this is the case, then you have to set reg = <2> for the second chip.
Yes, you are right, the second chip connects to QSPIB CS0, with the DT change, both of them work fine with __div0 issue workaround.
Ok, great. Thanks for your test.

As Yogesh reported problems with his LS1088ARDB board, where both chips
are connected to QSPIA, I was hoping someone could confirm this by
testing on a similar setup.

So if you have a board around, that uses this setup, it would be great
if you can try on that. If not we have to find other ways to investigate
this.

Thanks,
Frieder

[1]
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit?id=9882b5375df532acb2c2399a90d882461112e612
Post by Han Xu
Thanks,
Frieder
[1]
https://smex12-5-en-ctp.trendmicro.com:443/wis/clicktime/v1/query?url=https%3a%2f%2femea01.safelinks.protection.outlook.com%2f%3furl%3dhttp%253A%252F%252Fgit.f&umid=aac77072-ee97-4b8a-b8b9-55765e276aec&auth=78a0452d0eda3018cc3487ba1bec995089e109a2-54ce7888db3c2cb56ff686a840c4d0fa657a3cdb
reescale.com%2Fgit%2Fcgit.cgi%2Fimx%2Flinux-
imx.git%2Ftree%2Farch%2Farm%2Fboot%2Fdts%2Fimx6sx-
sabreauto.dts%3Fh%3Dimx_4.9.11_1.0.0_ga%23n771&amp;data=02%7C01%
7Chan.xu%40nxp.com%7C9d43e9b29a82419a36c408d618df3dc3%7C686ea1d
3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C636723744426896223&amp;sdata
=038DdgMLjO23Io3CTHoSVrJL2Ho%2BgxyHlx9%2BLMrLWkQ%3D&amp;reser
ved=0
Post by Han Xu
Post by Boris Brezillon
1/ is there an easy way to invalidate AHB buffers? I mean, not
something that implies a full reset + several milliseconds of delay
after the reset. Right now we trick the caching logic by mapping a
portion that is twice the size of the buffer and switching from one
sub-portion to this other to trigger a real read on each read
access, but that's hack-ish, and I'd be surprised if HW
engineers hadn't planned for this "manual AHB buffer flush" case.
2/ if we use DMA, do you know what happens when the TX FIFO runs
out
Post by Han Xu
Post by Boris Brezillon
of data while the TX request is not finished yet. In PIO mode, it
seems the engine sends garbage on the bus when that happens, and
we
Post by Han Xu
Post by Boris Brezillon
definitely don't want that.
While #1 is not blocking us, #2 is if we don't have those patches
[1][2] applied, and Marek wanted to be sure there was no other ways
to solve the "TX FIFO starvation" issue before considering these changes.
So that'd be great if someone from NXP could have a look/ask around
and give us answers to those 2 questions.
Thanks,
Boris
______________________________________________________
Linux MTD discussion mailing list
Lukasz Majewski
2018-09-18 22:42:47 UTC
Permalink
Dear All,

Maybe I do jump a bit off topic here, but...

I've read through the following thread:
https://patchwork.ozlabs.org/patch/939885/

And it mentions some issues with reading AHB content (buffers) in
fsl-quadspi.c driver discovered when new QuadSPI driver was developed.

I do have a setup with qspi0 having two SPI memories connected (2x16
MiB), and I'm wondering if anybody has some more info regarding:

(What's more is that there is a bug in
* the "IP Command Read" in the Vybrid.) found here:
https://elixir.bootlin.com/linux/v4.19-rc4/source/drivers/mtd/spi-nor/fsl-quadspi.c#L671

I've googled for some errata or known issues for vybryd's QSPI IP block
(vf610) but without luck so far ...

Thanks in advance,
Post by Boris Brezillon
Hi Han,
Post by Han Xu
-----Original Message-----
Sent: Wednesday, September 12, 2018 1:41 PM
QuadSPI controller
Hi Han,
Post by Han Xu
-----Original Message-----
Sent: Monday, September 3, 2018 4:02 AM
Gaur
Post by Han Xu
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Hi Han,
Post by Boris Brezillon
Hi Han,
On Thu, 2 Aug 2018 21:58:48 +0000
Post by Han Xu
-----Original Message-----
Sent: Thursday, August 2, 2018 8:09 AM
Subject: Re: Questions about the Freescale/NXP QuadSPI
controller
Ping.
I'm not sure if my message below went out to you at all. At
least I can't find it in the ML archive.
I still hope someone can help with the questions below.
Meanwhile for the second point I did some tests myself with
one chip on each of the two buses and it worked fine with my
latest v2
patches.
Post by Han Xu
Post by Boris Brezillon
Post by Han Xu
So I'm not sure at all why Yogesh has problems with his setup
(two chips on the first bus).
Tried to test the v2 patch set on i.MX6SX SDB board but get the memory
map failure.
Post by Boris Brezillon
Post by Han Xu
[ 1.298633] fsl-quadspi 21e4000.qspi: ioremap failed for resource
[mem
Post by Han Xu
0x70000000-0x7fffffff]
Post by Boris Brezillon
Post by Han Xu
[ 1.307330] fsl-quadspi 21e4000.qspi: Freescale QuadSPI
probe failed [ 1.313922] fsl-quadspi: probe of 21e4000.qspi
failed with error -12
This is the reason why dynamic ioremap added in previous driver, please
refer to
https://smex12-5-en-ctp.trendmicro.com:443/wis/clicktime/v1/query?url=https%3a%2f%2femea01.safelinks.protection.outlook.com%2f%3furl%3dhttps%253A%252F%252Fsm&umid=aac77072-ee97-4b8a-b8b9-55765e276aec&auth=78a0452d0eda3018cc3487ba1bec995089e109a2-cefb4275e4aea35bf0c475f6bd6da9e9a8c877a0
Post by Han Xu
ex12-5-en-
ctp.trendmicro.com%3A443%2Fwis%2Fclicktime%2Fv1%2Fquery%3Fu
rl%3Dhttps%253a%252f%252femea01.safelinks.protection.outlook.com%252
f
Post by Han Xu
%253furl%253dhttps%25253A%25252F%25252Fpa%26umid%3Dd6cc1014-
1848-42fb
Post by Han Xu
-92fd-
9626d45c8050%26auth%3D541e45255b6517100d80c2b1b80b6933b203c492-
5aa8e1977a9db94300a9f61f5446e7a21b175f56&amp;data=02%7C01%7Chan.x
u%40
nxp.com%7C9d43e9b29a82419a36c408d618df3dc3%7C686ea1d3bc2b4c6fa92c
d99c
5c301635%7C0%7C0%7C636723744426896223&amp;sdata=Y32I9H9adPn%2Bn
lcICuV
Post by Han Xu
M8Uoozsig%2BM3F0rNAhkIF5ZE%3D&amp;reserved=0
tchwork.ozlabs.org%2Fpatch%2F503655%2F&amp;data=02%7C01%7Chan.xu
Post by Han Xu
%40nx
p.com%7C9f45a8b666d3478f065408d6117bf524%7C686ea1d3bc2b4c6fa92cd9
Post by Han Xu
9c5c
301635%7C0%7C0%7C636715621426190473&amp;sdata=XWPfWe%2Fu2ePW
Post by Han Xu
mNPe179D0
Post by Boris Brezillon
Post by Han Xu
vjTp6eLp0%2FJRF2vRayDwug%3D&amp;reserved=0
We can reduce the size of the iomap to 2k * 4, since this is
all we use currently. Can you try to change the size of the
ioremap call to 16k and tell us if it works.
Were you able to test with the reduced iomap size?
It would be great to know if it works on your board.
Thanks,
Frieder
Test the code on i.MX6SX sabreauto board with two micron n25q256a chips
on two CS.
Post by Han Xu
First issue found is __div0 kernel dump with these code
/* Max 64 dummy clock cycles supported */ if (op->dummy.nbytes *
8 / op->dummy.buswidth > 64) dummy.buswidth was not set during
read id.
First, thank you for coming back to this and doing the test.
I'm currently not sure about the reason for this, but I guess
Boris will figure it out easily ;)
Ok, on a closer look it is obvious that the reason is this commit
[1]. My last test was before this change, when dummy.buswidth was
still set to 1 even if dummy.n_bytes is 0.
Now both are 0 and I need to handle this in the FSL QSPI driver.
Post by Han Xu
Post by Han Xu
Second issue is the second part failed to be probed, tried both buswidth 4
and buswidth 1.
Post by Han Xu
[ 1.364979] m25p80 spi5.0: found n25q256a, expected m25p80
[ 1.370986] m25p80 spi5.0: n25q256a (32768 Kbytes)
[ 1.381020] m25p80 spi5.1: unrecognized JEDEC id bytes: ff, ff, ff
&qspi1 {
pinctrl-names = "default";
pinctrl-0 = <&pinctrl_qspi1_1>;
status = "okay";
Š #address-cells = <1>;
Š #size-cells = <1>;
Š compatible = "micron,m25p80";
Š spi-max-frequency = <29000000>;
spi-rx-bus-width = <4>;
spi-tx-bus-width = <4>;
Š reg = <0>;
};
Š #address-cells = <1>;
Š #size-cells = <1>;
Š compatible = "micron,m25p80";
Š spi-max-frequency = <29000000>;
spi-rx-bus-width = <4>;
spi-tx-bus-width = <4>;
Š reg = <1>;
};
};
First, I think you should add "jedec,spi-nor" to your compatible properties.
Second, are you sure, that the two chips are both on QSPIA using
the two chip selects?
I have no schematics of the board, but if I look at the devicetree
in the linux- imx kernel [1] it seems to me that one chip is on
QSPIA CS0 and the other on QSPIB CS0.
If this is the case, then you have to set reg = <2> for the second chip.
Yes, you are right, the second chip connects to QSPIB CS0, with the
DT change, both of them work fine with __div0 issue workaround.
Ok, great. Thanks for your test.
As Yogesh reported problems with his LS1088ARDB board, where both
chips are connected to QSPIA, I was hoping someone could confirm this
by testing on a similar setup.
So if you have a board around, that uses this setup, it would be
great if you can try on that. If not we have to find other ways to
investigate this.
Thanks,
Frieder
[1]
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit?id=9882b5375df532acb2c2399a90d882461112e612
Post by Han Xu
Thanks,
Frieder
[1]
https://smex12-5-en-ctp.trendmicro.com:443/wis/clicktime/v1/query?url=https%3a%2f%2femea01.safelinks.protection.outlook.com%2f%3furl%3dhttp%253A%252F%252Fgit.f&umid=aac77072-ee97-4b8a-b8b9-55765e276aec&auth=78a0452d0eda3018cc3487ba1bec995089e109a2-54ce7888db3c2cb56ff686a840c4d0fa657a3cdb
reescale.com%2Fgit%2Fcgit.cgi%2Fimx%2Flinux-
imx.git%2Ftree%2Farch%2Farm%2Fboot%2Fdts%2Fimx6sx-
sabreauto.dts%3Fh%3Dimx_4.9.11_1.0.0_ga%23n771&amp;data=02%7C01%
7Chan.xu%40nxp.com%7C9d43e9b29a82419a36c408d618df3dc3%7C686ea1d
3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C636723744426896223&amp;sdata
=038DdgMLjO23Io3CTHoSVrJL2Ho%2BgxyHlx9%2BLMrLWkQ%3D&amp;reser
ved=0
Post by Han Xu
Post by Boris Brezillon
Unrelated to this issue, we still have 2 questions left
1/ is there an easy way to invalidate AHB buffers? I mean, not
something that implies a full reset + several
milliseconds of delay after the reset. Right now we trick the
caching logic by mapping a portion that is twice the size of
the buffer and switching from one sub-portion to this other to
trigger a real read on each read access, but that's hack-ish,
and I'd be surprised if HW engineers hadn't planned for this
"manual AHB buffer flush" case.
2/ if we use DMA, do you know what happens when the TX FIFO runs
out
Post by Han Xu
Post by Boris Brezillon
of data while the TX request is not finished yet. In PIO
mode, it seems the engine sends garbage on the bus when that
happens, and
we
Post by Han Xu
Post by Boris Brezillon
definitely don't want that.
While #1 is not blocking us, #2 is if we don't have those
patches [1][2] applied, and Marek wanted to be sure there was
no other ways to solve the "TX FIFO starvation" issue before
considering these changes. So that'd be great if someone from
NXP could have a look/ask around and give us answers to those 2
questions.
Thanks,
Boris
______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: ***@denx.de
Frieder Schrempf
2018-09-19 06:49:14 UTC
Permalink
Hi Lukasz,
Post by Lukasz Majewski
Dear All,
Maybe I do jump a bit off topic here, but...
https://patchwork.ozlabs.org/patch/939885/
And it mentions some issues with reading AHB content (buffers) in
fsl-quadspi.c driver discovered when new QuadSPI driver was developed.
The only setup with two chips that is known to be problematic with the
new driver, is when both chips are connected to the same bus (e.g.
QSPIA) using separate chip selects.

Does your board use this kind of setup, or are the two chips connected
two different buses (QSPIA and QSPIB)?

Have you tested the new driver? It would be great to receive some more
feedback.
Post by Lukasz Majewski
I do have a setup with qspi0 having two SPI memories connected (2x16
(What's more is that there is a bug in
https://elixir.bootlin.com/linux/v4.19-rc4/source/drivers/mtd/spi-nor/fsl-quadspi.c#L671
I've googled for some errata or known issues for vybryd's QSPI IP block
(vf610) but without luck so far ...
Unfortunately I don't know the background for this comment. Is your
board using VF610? Do you experience problems?

Regards,
Frieder
Lukasz Majewski
2018-09-19 11:02:11 UTC
Permalink
Hi Frieder,
Post by Frieder Schrempf
Hi Lukasz,
Post by Lukasz Majewski
Dear All,
Maybe I do jump a bit off topic here, but...
https://patchwork.ozlabs.org/patch/939885/
And it mentions some issues with reading AHB content (buffers) in
fsl-quadspi.c driver discovered when new QuadSPI driver was
developed.
The only setup with two chips that is known to be problematic with
the new driver, is when both chips are connected to the same bus
(e.g. QSPIA) using separate chip selects.
I'm using QSPI0 controller, with two memories connected to QSPI0_A and
QSPI0_B lines.
Post by Frieder Schrempf
Does your board use this kind of setup, or are the two chips
connected two different buses (QSPIA and QSPIB)?
Have you tested the new driver? It would be great to receive some
more feedback.
I will check (test) this new driver. No problem.
Post by Frieder Schrempf
Post by Lukasz Majewski
I do have a setup with qspi0 having two SPI memories connected (2x16
(What's more is that there is a bug in
https://elixir.bootlin.com/linux/v4.19-rc4/source/drivers/mtd/spi-nor/fsl-quadspi.c#L671
I've googled for some errata or known issues for vybryd's QSPI IP
block (vf610) but without luck so far ...
Unfortunately I don't know the background for this comment.
The comment was added by some Freescale employee when the driver was
added to Linux (by Huang - CC'ed).
Post by Frieder Schrempf
Is your
board using VF610?
Yes, it uses vf610 (A5 + M4 cores).
Post by Frieder Schrempf
Do you experience problems?
I've already observed following issue:

For the current QSPI ML driver (fsl-quadspi.c - 4.19-rc3) only half
of the AHB buffer is valid. When I do read the whole one - I do see read
data corruption (on UBI or raw memory).

This was pointed out in the patch:
https://patchwork.ozlabs.org/patch/675401/

Unfortunately, I did not found any info regarding this problem (in
NXP's errata or other doc).

I will check if this issue shows up on new patches.

Thanks in advance for your help.
Post by Frieder Schrempf
Regards,
Frieder
______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: ***@denx.de
Huang Shijie
2018-09-20 01:17:59 UTC
Permalink
Post by Boris Brezillon
Hi Frieder,
Post by Frieder Schrempf
Hi Lukasz,
Post by Lukasz Majewski
Dear All,
Maybe I do jump a bit off topic here, but...
https://patchwork.ozlabs.org/patch/939885/
And it mentions some issues with reading AHB content (buffers) in
fsl-quadspi.c driver discovered when new QuadSPI driver was developed.
The only setup with two chips that is known to be problematic with
the new driver, is when both chips are connected to the same bus
(e.g. QSPIA) using separate chip selects.
I'm using QSPI0 controller, with two memories connected to QSPI0_A and
QSPI0_B lines.
Post by Frieder Schrempf
Does your board use this kind of setup, or are the two chips
connected two different buses (QSPIA and QSPIB)?
Have you tested the new driver? It would be great to receive some
more feedback.
I will check (test) this new driver. No problem.
Post by Frieder Schrempf
Post by Lukasz Majewski
I do have a setup with qspi0 having two SPI memories connected (2x16
(What's more is that there is a bug in
https://elixir.bootlin.com/linux/v4.19-rc4/source/drivers/mtd/spi-nor/fsl-quadspi.c#L671
I've googled for some errata or known issues for vybryd's QSPI IP
block (vf610) but without luck so far ...
Unfortunately I don't know the background for this comment.
The comment was added by some Freescale employee when the driver was
added to Linux (by Huang - CC'ed).
I was told that there was a hardware bug in the vybryd when we use the "IP Command Read":
If you use the "IP Command Read", you will not get the correct data.

So I added that comment.

I suggest that do not use the "IP Command Read" for vybryd.

Thanks
Huang Shijie
Post by Boris Brezillon
Post by Frieder Schrempf
Is your
board using VF610?
Yes, it uses vf610 (A5 + M4 cores).
Lukasz Majewski
2018-09-20 15:00:32 UTC
Permalink
Hi Frieder,
Post by Boris Brezillon
Hi Frieder,
Post by Frieder Schrempf
Hi Lukasz,
Post by Lukasz Majewski
Dear All,
Maybe I do jump a bit off topic here, but...
https://patchwork.ozlabs.org/patch/939885/
And it mentions some issues with reading AHB content (buffers) in
fsl-quadspi.c driver discovered when new QuadSPI driver was developed.
The only setup with two chips that is known to be problematic with
the new driver, is when both chips are connected to the same bus
(e.g. QSPIA) using separate chip selects.
I'm using QSPI0 controller, with two memories connected to QSPI0_A and
QSPI0_B lines.
Post by Frieder Schrempf
Does your board use this kind of setup, or are the two chips
connected two different buses (QSPIA and QSPIB)?
Have you tested the new driver? It would be great to receive some
more feedback.
I will check (test) this new driver. No problem.
I've ported your patch on v4.19-rc3.

I had to fix the "div0 issue" by adding:
if (op->dummy.buswidth &&
(op->dummy.nbytes * 8 / op->dummy.buswidth > 64))

In fsl_qspi_supports_op() function.


Unfortunately, on Vybrid vf610 when testing I do see corruption of read
data from the mtd device:

~# hexdump aaa.img
0000000 464c eec0 baa5 c5ff 7b99 4dfb e0b6 8a2e
0000010 e98e 5265 683c a635 c069 e402 303f d936
0000020 c243 01a7 7064 fce8 e3a9 200a 7e85 28bc
0000030 4296 a30e 1bb4 88d4 b456 b4a6 f3aa 8cff
0000040 ffff ffff ffff ffff ffff ffff ffff ffff
*
0000100

~# hexdump data_test.img
0000000 464c eec0 baa5 c5ff 7b99 4dfb e0b6 8a2e
0000010 e98e 5265 683c a635 c069 e402 303f d936
0000020 c243 01a7 7064 fce8 e3a9 200a 7e85 28bc
0000030 4296 a30e 1bb4 88d4 b456 b4a6 f3aa 8cff
0000040 01c9 462d 0a43 f893 0e42 67f1 57f0 787c
0000050 49c0 fb2a e514 e954 1d21 affa bac4 38f1
0000060 1ca5 ec46 77eb a854 285b 8e21 12d7 f377
0000070 b441 2baf ee33 596c 98b6 e71a c1cb 876c


Test case:
flash_erase /dev/mtd3 0 1
dd if=data_test.img of=/dev/mtd3
dd if=/dev/mtd3 of=aaa.img bs=1 count=256


From the code I see that the AHB buffer has 1024B. It is accessed in 8
bytes packs.

This seems like some cache or HW issue....

In the old driver, to fix this problem one needed to
1. Reduce the RX fifo size from 128B to 64B

2. Disable (set to 0) the AHB transfer's BUF3CR ADATSZ field - the
transfer size is the same as in LUT.

(Original patch: https://patchwork.ozlabs.org/patch/675401/)
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
I do have a setup with qspi0 having two SPI memories connected
(2x16 MiB), and I'm wondering if anybody has some more info
(What's more is that there is a bug in
^^^^^^^^^^^^^^^^^^^^^^^^ - It would be handy to have the exact
explanation of this issue.
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
https://elixir.bootlin.com/linux/v4.19-rc4/source/drivers/mtd/spi-nor/fsl-quadspi.c#L67
I've googled for some errata or known issues for vybryd's QSPI IP
block (vf610) but without luck so far ...
Unfortunately I don't know the background for this comment.
The comment was added by some Freescale employee when the driver was
added to Linux (by Huang - CC'ed).
Post by Frieder Schrempf
Is your
board using VF610?
Yes, it uses vf610 (A5 + M4 cores).
Post by Frieder Schrempf
Do you experience problems?
For the current QSPI ML driver (fsl-quadspi.c - 4.19-rc3) only half
of the AHB buffer is valid. When I do read the whole one - I do see
read data corruption (on UBI or raw memory).
https://patchwork.ozlabs.org/patch/675401/
Unfortunately, I did not found any info regarding this problem (in
NXP's errata or other doc).
I will check if this issue shows up on new patches.
Thanks in advance for your help.
Post by Frieder Schrempf
Regards,
Frieder
______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: ***@denx.de
Frieder Schrempf
2018-09-20 15:41:52 UTC
Permalink
Hi Lukasz,
Post by Boris Brezillon
Hi Frieder,
Post by Boris Brezillon
Hi Frieder,
Post by Frieder Schrempf
Hi Lukasz,
Post by Lukasz Majewski
Dear All,
Maybe I do jump a bit off topic here, but...
https://patchwork.ozlabs.org/patch/939885/
And it mentions some issues with reading AHB content (buffers) in
fsl-quadspi.c driver discovered when new QuadSPI driver was
developed.
The only setup with two chips that is known to be problematic with
the new driver, is when both chips are connected to the same bus
(e.g. QSPIA) using separate chip selects.
I'm using QSPI0 controller, with two memories connected to QSPI0_A and
QSPI0_B lines.
Post by Frieder Schrempf
Does your board use this kind of setup, or are the two chips
connected two different buses (QSPIA and QSPIB)?
Have you tested the new driver? It would be great to receive some
more feedback.
I will check (test) this new driver. No problem.
I've ported your patch on v4.19-rc3.
if (op->dummy.buswidth &&
(op->dummy.nbytes * 8 / op->dummy.buswidth > 64))
In fsl_qspi_supports_op() function.
Unfortunately, on Vybrid vf610 when testing I do see corruption of read
~# hexdump aaa.img
0000000 464c eec0 baa5 c5ff 7b99 4dfb e0b6 8a2e
0000010 e98e 5265 683c a635 c069 e402 303f d936
0000020 c243 01a7 7064 fce8 e3a9 200a 7e85 28bc
0000030 4296 a30e 1bb4 88d4 b456 b4a6 f3aa 8cff
0000040 ffff ffff ffff ffff ffff ffff ffff ffff
*
0000100
~# hexdump data_test.img
0000000 464c eec0 baa5 c5ff 7b99 4dfb e0b6 8a2e
0000010 e98e 5265 683c a635 c069 e402 303f d936
0000020 c243 01a7 7064 fce8 e3a9 200a 7e85 28bc
0000030 4296 a30e 1bb4 88d4 b456 b4a6 f3aa 8cff
0000040 01c9 462d 0a43 f893 0e42 67f1 57f0 787c
0000050 49c0 fb2a e514 e954 1d21 affa bac4 38f1
0000060 1ca5 ec46 77eb a854 285b 8e21 12d7 f377
0000070 b441 2baf ee33 596c 98b6 e71a c1cb 876c
flash_erase /dev/mtd3 0 1
dd if=data_test.img of=/dev/mtd3
dd if=/dev/mtd3 of=aaa.img bs=1 count=256
From the code I see that the AHB buffer has 1024B. It is accessed in 8
bytes packs.
This seems like some cache or HW issue....
In the old driver, to fix this problem one needed to
1. Reduce the RX fifo size from 128B to 64B
2. Disable (set to 0) the AHB transfer's BUF3CR ADATSZ field - the
transfer size is the same as in LUT.
(Original patch: https://patchwork.ozlabs.org/patch/675401/)
Thanks a lot for doing the test. It seems like Vybrid needs some special
handling to work around this problem.
So this means the current driver has never worked for Vybrid, as the
mentioned patch was never merged!?

As these issues seem to be specific to Vybrid and as they seem to exist
in the current driver too, I would like to fix them after we have a
first version of the new driver that works for the other platforms.

I hope we can figure out the other issues soon.

Thanks,
Frieder
Post by Boris Brezillon
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
I do have a setup with qspi0 having two SPI memories connected
(2x16 MiB), and I'm wondering if anybody has some more info
(What's more is that there is a bug in
^^^^^^^^^^^^^^^^^^^^^^^^ - It would be handy to have the exact
explanation of this issue.
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
https://elixir.bootlin.com/linux/v4.19-rc4/source/drivers/mtd/spi-nor/fsl-quadspi.c#L67
I've googled for some errata or known issues for vybryd's QSPI IP
block (vf610) but without luck so far ...
Unfortunately I don't know the background for this comment.
The comment was added by some Freescale employee when the driver was
added to Linux (by Huang - CC'ed).
Post by Frieder Schrempf
Is your
board using VF610?
Yes, it uses vf610 (A5 + M4 cores).
Post by Frieder Schrempf
Do you experience problems?
For the current QSPI ML driver (fsl-quadspi.c - 4.19-rc3) only half
of the AHB buffer is valid. When I do read the whole one - I do see
read data corruption (on UBI or raw memory).
https://patchwork.ozlabs.org/patch/675401/
Unfortunately, I did not found any info regarding this problem (in
NXP's errata or other doc).
I will check if this issue shows up on new patches.
Thanks in advance for your help.
Post by Frieder Schrempf
Regards,
Frieder
______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Lukasz Majewski
2018-09-20 20:37:23 UTC
Permalink
Hi Frieder,
Post by Frieder Schrempf
Hi Lukasz,
Post by Boris Brezillon
Hi Frieder,
Post by Boris Brezillon
Hi Frieder,
Post by Frieder Schrempf
Hi Lukasz,
Post by Lukasz Majewski
Dear All,
Maybe I do jump a bit off topic here, but...
https://patchwork.ozlabs.org/patch/939885/
And it mentions some issues with reading AHB content (buffers) in
fsl-quadspi.c driver discovered when new QuadSPI driver was developed.
The only setup with two chips that is known to be problematic with
the new driver, is when both chips are connected to the same bus
(e.g. QSPIA) using separate chip selects.
I'm using QSPI0 controller, with two memories connected to QSPI0_A
and QSPI0_B lines.
Post by Frieder Schrempf
Does your board use this kind of setup, or are the two chips
connected two different buses (QSPIA and QSPIB)?
Have you tested the new driver? It would be great to receive some
more feedback.
I will check (test) this new driver. No problem.
I've ported your patch on v4.19-rc3.
if (op->dummy.buswidth &&
(op->dummy.nbytes * 8 / op->dummy.buswidth > 64))
In fsl_qspi_supports_op() function.
Unfortunately, on Vybrid vf610 when testing I do see corruption of
~# hexdump aaa.img
0000000 464c eec0 baa5 c5ff 7b99 4dfb e0b6 8a2e
0000010 e98e 5265 683c a635 c069 e402 303f d936
0000020 c243 01a7 7064 fce8 e3a9 200a 7e85 28bc
0000030 4296 a30e 1bb4 88d4 b456 b4a6 f3aa 8cff
0000040 ffff ffff ffff ffff ffff ffff ffff ffff
*
0000100
~# hexdump data_test.img
0000000 464c eec0 baa5 c5ff 7b99 4dfb e0b6 8a2e
0000010 e98e 5265 683c a635 c069 e402 303f d936
0000020 c243 01a7 7064 fce8 e3a9 200a 7e85 28bc
0000030 4296 a30e 1bb4 88d4 b456 b4a6 f3aa 8cff
0000040 01c9 462d 0a43 f893 0e42 67f1 57f0 787c
0000050 49c0 fb2a e514 e954 1d21 affa bac4 38f1
0000060 1ca5 ec46 77eb a854 285b 8e21 12d7 f377
0000070 b441 2baf ee33 596c 98b6 e71a c1cb 876c
flash_erase /dev/mtd3 0 1
dd if=data_test.img of=/dev/mtd3
dd if=/dev/mtd3 of=aaa.img bs=1 count=256
From the code I see that the AHB buffer has 1024B. It is accessed
in 8 bytes packs.
This seems like some cache or HW issue....
In the old driver, to fix this problem one needed to
1. Reduce the RX fifo size from 128B to 64B
2. Disable (set to 0) the AHB transfer's BUF3CR ADATSZ field - the
transfer size is the same as in LUT.
(Original patch: https://patchwork.ozlabs.org/patch/675401/)
Thanks a lot for doing the test. It seems like Vybrid needs some
special handling to work around this problem.
I'm not sure if it is special - it looks like there may be some HW
issue with Vybrid.
Post by Frieder Schrempf
So this means the current driver has never worked for Vybrid, as the
mentioned patch was never merged!?
At least on my setup. You cannot use MTD memory for anything as you
cannot read correctly data from it.

I'm wondering if any other vybrid users (with QSPI) also experience
such problems.
Post by Frieder Schrempf
As these issues seem to be specific to Vybrid and as they seem to
exist in the current driver too, I would like to fix them after we
have a first version of the new driver that works for the other
platforms.
It is fine for me - I hope that I will find solution soon (and rebase
it on top of your patch).
Post by Frieder Schrempf
I hope we can figure out the other issues soon.
I would be very happy if anybody from NXP could look on this issue and
forward it to HW engineers for some insight.

(Vybryd has some old "chip errata document":
https://community.nxp.com/message/1059701 , but it mentions other issue
with QSPI).
Post by Frieder Schrempf
Thanks,
Frieder
Post by Boris Brezillon
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
I do have a setup with qspi0 having two SPI memories connected
(2x16 MiB), and I'm wondering if anybody has some more info
(What's more is that there is a bug in
^^^^^^^^^^^^^^^^^^^^^^^^ - It would be handy to have the
exact explanation of this issue.
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
https://elixir.bootlin.com/linux/v4.19-rc4/source/drivers/mtd/spi-nor/fsl-quadspi.c#L67
I've googled for some errata or known issues for vybryd's QSPI IP
block (vf610) but without luck so far ...
Unfortunately I don't know the background for this comment.
The comment was added by some Freescale employee when the driver
was added to Linux (by Huang - CC'ed).
Post by Frieder Schrempf
Is your
board using VF610?
Yes, it uses vf610 (A5 + M4 cores).
Post by Frieder Schrempf
Do you experience problems?
For the current QSPI ML driver (fsl-quadspi.c - 4.19-rc3) only half
of the AHB buffer is valid. When I do read the whole one - I do see
read data corruption (on UBI or raw memory).
https://patchwork.ozlabs.org/patch/675401/
Unfortunately, I did not found any info regarding this problem (in
NXP's errata or other doc).
I will check if this issue shows up on new patches.
Thanks in advance for your help.
Post by Frieder Schrempf
Regards,
Frieder
______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang
Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell,
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang
Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell,
______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: ***@denx.de
Lukasz Majewski
2018-09-20 22:13:23 UTC
Permalink
Hi Frieder,

Please find some more comments regarding the new spi-fsl-qspi.c driver.
Post by Frieder Schrempf
Hi Lukasz,
Post by Boris Brezillon
Hi Frieder,
Post by Boris Brezillon
Hi Frieder,
Post by Frieder Schrempf
Hi Lukasz,
Post by Lukasz Majewski
Dear All,
Maybe I do jump a bit off topic here, but...
https://patchwork.ozlabs.org/patch/939885/
And it mentions some issues with reading AHB content (buffers) in
fsl-quadspi.c driver discovered when new QuadSPI driver was developed.
The only setup with two chips that is known to be problematic with
the new driver, is when both chips are connected to the same bus
(e.g. QSPIA) using separate chip selects.
I'm using QSPI0 controller, with two memories connected to QSPI0_A
and QSPI0_B lines.
Post by Frieder Schrempf
Does your board use this kind of setup, or are the two chips
connected two different buses (QSPIA and QSPIB)?
Have you tested the new driver? It would be great to receive some
more feedback.
I will check (test) this new driver. No problem.
I've ported your patch on v4.19-rc3.
if (op->dummy.buswidth &&
(op->dummy.nbytes * 8 / op->dummy.buswidth > 64))
In fsl_qspi_supports_op() function.
Unfortunately, on Vybrid vf610 when testing I do see corruption of
~# hexdump aaa.img
0000000 464c eec0 baa5 c5ff 7b99 4dfb e0b6 8a2e
0000010 e98e 5265 683c a635 c069 e402 303f d936
0000020 c243 01a7 7064 fce8 e3a9 200a 7e85 28bc
0000030 4296 a30e 1bb4 88d4 b456 b4a6 f3aa 8cff
0000040 ffff ffff ffff ffff ffff ffff ffff ffff
*
0000100
~# hexdump data_test.img
0000000 464c eec0 baa5 c5ff 7b99 4dfb e0b6 8a2e
0000010 e98e 5265 683c a635 c069 e402 303f d936
0000020 c243 01a7 7064 fce8 e3a9 200a 7e85 28bc
0000030 4296 a30e 1bb4 88d4 b456 b4a6 f3aa 8cff
0000040 01c9 462d 0a43 f893 0e42 67f1 57f0 787c
0000050 49c0 fb2a e514 e954 1d21 affa bac4 38f1
0000060 1ca5 ec46 77eb a854 285b 8e21 12d7 f377
0000070 b441 2baf ee33 596c 98b6 e71a c1cb 876c
flash_erase /dev/mtd3 0 1
dd if=data_test.img of=/dev/mtd3
dd if=/dev/mtd3 of=aaa.img bs=1 count=256
From the code I see that the AHB buffer has 1024B. It is accessed
in 8 bytes packs.
This seems like some cache or HW issue....
In the old driver, to fix this problem one needed to
1. Reduce the RX fifo size from 128B to 64B
2. Disable (set to 0) the AHB transfer's BUF3CR ADATSZ field - the
transfer size is the same as in LUT.
(Original patch: https://patchwork.ozlabs.org/patch/675401/)
Thanks a lot for doing the test. It seems like Vybrid needs some
special handling to work around this problem.
So this means the current driver has never worked for Vybrid, as the
mentioned patch was never merged!?
As these issues seem to be specific to Vybrid and as they seem to
exist in the current driver too, I would like to fix them after we
have a first version of the new driver that works for the other
platforms.
I hope we can figure out the other issues soon.
Thanks,
Frieder
Post by Boris Brezillon
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
I do have a setup with qspi0 having two SPI memories connected
(2x16 MiB), and I'm wondering if anybody has some more info
(What's more is that there is a bug in
^^^^^^^^^^^^^^^^^^^^^^^^ - It would be handy to have the
exact explanation of this issue.
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
https://elixir.bootlin.com/linux/v4.19-rc4/source/drivers/mtd/spi-nor/fsl-quadspi.c#L671
^^^^^^^^^^^^ - [1]


I've looked a bit closer to the new spi-fsl-qspi.c driver and it looks
like it uses fsl_qspi_exec_op() and then calls fsl_qspi_do_op().

In the last function we setup the QUADSPI_IPCR register to initiate the
transfer between SPI-NOR and the internal RX buffer.

Please correct me if I'm wrong but it seems to me like we are using "IP
Command Read" approach here.

If this is true - what was the rationale to use it in the new driver, as
in the old one the "AHB Command Read" approach was recommended (i.e.
faster) [1]?

To be even more problematic the info in link [1] states that on vybrid
there is a bug on the "IP Command Read" HW block and it should be
avoided.
Post by Frieder Schrempf
Post by Boris Brezillon
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
I've googled for some errata or known issues for vybryd's QSPI IP
block (vf610) but without luck so far ...
Unfortunately I don't know the background for this comment.
The comment was added by some Freescale employee when the driver
was added to Linux (by Huang - CC'ed).
Post by Frieder Schrempf
Is your
board using VF610?
Yes, it uses vf610 (A5 + M4 cores).
Post by Frieder Schrempf
Do you experience problems?
For the current QSPI ML driver (fsl-quadspi.c - 4.19-rc3) only half
of the AHB buffer is valid. When I do read the whole one - I do see
read data corruption (on UBI or raw memory).
https://patchwork.ozlabs.org/patch/675401/
Unfortunately, I did not found any info regarding this problem (in
NXP's errata or other doc).
I will check if this issue shows up on new patches.
Thanks in advance for your help.
Post by Frieder Schrempf
Regards,
Frieder
______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang
Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell,
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang
Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell,
______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: ***@denx.de
Frieder Schrempf
2018-09-25 06:49:20 UTC
Permalink
Hi Lukasz,
Post by Boris Brezillon
Hi Frieder,
Please find some more comments regarding the new spi-fsl-qspi.c driver.
Post by Frieder Schrempf
Hi Lukasz,
Post by Boris Brezillon
Hi Frieder,
Post by Boris Brezillon
Hi Frieder,
Post by Frieder Schrempf
Hi Lukasz,
Post by Lukasz Majewski
Dear All,
Maybe I do jump a bit off topic here, but...
https://patchwork.ozlabs.org/patch/939885/
And it mentions some issues with reading AHB content (buffers) in
fsl-quadspi.c driver discovered when new QuadSPI driver was developed.
The only setup with two chips that is known to be problematic with
the new driver, is when both chips are connected to the same bus
(e.g. QSPIA) using separate chip selects.
I'm using QSPI0 controller, with two memories connected to QSPI0_A
and QSPI0_B lines.
Post by Frieder Schrempf
Does your board use this kind of setup, or are the two chips
connected two different buses (QSPIA and QSPIB)?
Have you tested the new driver? It would be great to receive some
more feedback.
I will check (test) this new driver. No problem.
I've ported your patch on v4.19-rc3.
if (op->dummy.buswidth &&
(op->dummy.nbytes * 8 / op->dummy.buswidth > 64))
In fsl_qspi_supports_op() function.
Unfortunately, on Vybrid vf610 when testing I do see corruption of
~# hexdump aaa.img
0000000 464c eec0 baa5 c5ff 7b99 4dfb e0b6 8a2e
0000010 e98e 5265 683c a635 c069 e402 303f d936
0000020 c243 01a7 7064 fce8 e3a9 200a 7e85 28bc
0000030 4296 a30e 1bb4 88d4 b456 b4a6 f3aa 8cff
0000040 ffff ffff ffff ffff ffff ffff ffff ffff
*
0000100
~# hexdump data_test.img
0000000 464c eec0 baa5 c5ff 7b99 4dfb e0b6 8a2e
0000010 e98e 5265 683c a635 c069 e402 303f d936
0000020 c243 01a7 7064 fce8 e3a9 200a 7e85 28bc
0000030 4296 a30e 1bb4 88d4 b456 b4a6 f3aa 8cff
0000040 01c9 462d 0a43 f893 0e42 67f1 57f0 787c
0000050 49c0 fb2a e514 e954 1d21 affa bac4 38f1
0000060 1ca5 ec46 77eb a854 285b 8e21 12d7 f377
0000070 b441 2baf ee33 596c 98b6 e71a c1cb 876c
flash_erase /dev/mtd3 0 1
dd if=data_test.img of=/dev/mtd3
dd if=/dev/mtd3 of=aaa.img bs=1 count=256
From the code I see that the AHB buffer has 1024B. It is accessed
in 8 bytes packs.
This seems like some cache or HW issue....
In the old driver, to fix this problem one needed to
1. Reduce the RX fifo size from 128B to 64B
2. Disable (set to 0) the AHB transfer's BUF3CR ADATSZ field - the
transfer size is the same as in LUT.
(Original patch: https://patchwork.ozlabs.org/patch/675401/)
Thanks a lot for doing the test. It seems like Vybrid needs some
special handling to work around this problem.
So this means the current driver has never worked for Vybrid, as the
mentioned patch was never merged!?
As these issues seem to be specific to Vybrid and as they seem to
exist in the current driver too, I would like to fix them after we
have a first version of the new driver that works for the other
platforms.
I hope we can figure out the other issues soon.
Thanks,
Frieder
Post by Boris Brezillon
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
I do have a setup with qspi0 having two SPI memories connected
(2x16 MiB), and I'm wondering if anybody has some more info
(What's more is that there is a bug in
^^^^^^^^^^^^^^^^^^^^^^^^ - It would be handy to have the
exact explanation of this issue.
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
https://elixir.bootlin.com/linux/v4.19-rc4/source/drivers/mtd/spi-nor/fsl-quadspi.c#L671
^^^^^^^^^^^^ - [1]
I've looked a bit closer to the new spi-fsl-qspi.c driver and it looks
like it uses fsl_qspi_exec_op() and then calls fsl_qspi_do_op().
In the last function we setup the QUADSPI_IPCR register to initiate the
transfer between SPI-NOR and the internal RX buffer.
Please correct me if I'm wrong but it seems to me like we are using "IP
Command Read" approach here.
"IP Read" is only used if data size is smaller then RX FIFO size (128
bytes for Vybrid). For bigger chunks of data "AHB Read" is used.
Post by Boris Brezillon
If this is true - what was the rationale to use it in the new driver, as
in the old one the "AHB Command Read" approach was recommended (i.e.
faster) [1]?
As explained above we only use "IP Read" for small chunks of data. To
speed up the tranfer for bigger chunks we do use "AHB Read".
And in the future we plan to add an interface for mapping memory, so we
can use the "AHB Read" even more efficiently.
Post by Boris Brezillon
To be even more problematic the info in link [1] states that on vybrid
there is a bug on the "IP Command Read" HW block and it should be
avoided.
If "IP Read" should be avoided completely for Vybrid, we have to add
some kind of quirk.

Thanks for you comments,
Frieder
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Boris Brezillon
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
I've googled for some errata or known issues for vybryd's QSPI IP
block (vf610) but without luck so far ...
Unfortunately I don't know the background for this comment.
The comment was added by some Freescale employee when the driver
was added to Linux (by Huang - CC'ed).
Post by Frieder Schrempf
Is your
board using VF610?
Yes, it uses vf610 (A5 + M4 cores).
Post by Frieder Schrempf
Do you experience problems?
For the current QSPI ML driver (fsl-quadspi.c - 4.19-rc3) only half
of the AHB buffer is valid. When I do read the whole one - I do see
read data corruption (on UBI or raw memory).
https://patchwork.ozlabs.org/patch/675401/
Unfortunately, I did not found any info regarding this problem (in
NXP's errata or other doc).
I will check if this issue shows up on new patches.
Thanks in advance for your help.
Post by Frieder Schrempf
Regards,
Frieder
______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang
Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell,
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang
Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell,
______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Lukasz Majewski
2018-09-25 08:53:12 UTC
Permalink
Hi Frieder,
Post by Frieder Schrempf
Hi Lukasz,
Post by Boris Brezillon
Hi Frieder,
Please find some more comments regarding the new spi-fsl-qspi.c driver.
Post by Frieder Schrempf
Hi Lukasz,
Post by Boris Brezillon
Hi Frieder,
Post by Boris Brezillon
Hi Frieder,
Post by Frieder Schrempf
Hi Lukasz,
Post by Lukasz Majewski
Dear All,
Maybe I do jump a bit off topic here, but...
https://patchwork.ozlabs.org/patch/939885/
And it mentions some issues with reading AHB content (buffers)
in fsl-quadspi.c driver discovered when new QuadSPI driver was
developed.
The only setup with two chips that is known to be problematic
with the new driver, is when both chips are connected to the
same bus (e.g. QSPIA) using separate chip selects.
I'm using QSPI0 controller, with two memories connected to
QSPI0_A and QSPI0_B lines.
Post by Frieder Schrempf
Does your board use this kind of setup, or are the two chips
connected two different buses (QSPIA and QSPIB)?
Have you tested the new driver? It would be great to receive
some more feedback.
I will check (test) this new driver. No problem.
I've ported your patch on v4.19-rc3.
if (op->dummy.buswidth &&
(op->dummy.nbytes * 8 / op->dummy.buswidth > 64))
In fsl_qspi_supports_op() function.
Unfortunately, on Vybrid vf610 when testing I do see corruption of
~# hexdump aaa.img
0000000 464c eec0 baa5 c5ff 7b99 4dfb e0b6 8a2e
0000010 e98e 5265 683c a635 c069 e402 303f d936
0000020 c243 01a7 7064 fce8 e3a9 200a 7e85 28bc
0000030 4296 a30e 1bb4 88d4 b456 b4a6 f3aa 8cff
0000040 ffff ffff ffff ffff ffff ffff ffff ffff
*
0000100
~# hexdump data_test.img
0000000 464c eec0 baa5 c5ff 7b99 4dfb e0b6 8a2e
0000010 e98e 5265 683c a635 c069 e402 303f d936
0000020 c243 01a7 7064 fce8 e3a9 200a 7e85 28bc
0000030 4296 a30e 1bb4 88d4 b456 b4a6 f3aa 8cff
0000040 01c9 462d 0a43 f893 0e42 67f1 57f0 787c
0000050 49c0 fb2a e514 e954 1d21 affa bac4 38f1
0000060 1ca5 ec46 77eb a854 285b 8e21 12d7 f377
0000070 b441 2baf ee33 596c 98b6 e71a c1cb 876c
flash_erase /dev/mtd3 0 1
dd if=data_test.img of=/dev/mtd3
dd if=/dev/mtd3 of=aaa.img bs=1 count=256
From the code I see that the AHB buffer has 1024B. It is
accessed in 8 bytes packs.
This seems like some cache or HW issue....
In the old driver, to fix this problem one needed to
1. Reduce the RX fifo size from 128B to 64B
2. Disable (set to 0) the AHB transfer's BUF3CR ADATSZ field - the
transfer size is the same as in LUT.
(Original patch: https://patchwork.ozlabs.org/patch/675401/)
Thanks a lot for doing the test. It seems like Vybrid needs some
special handling to work around this problem.
So this means the current driver has never worked for Vybrid, as
the mentioned patch was never merged!?
As these issues seem to be specific to Vybrid and as they seem to
exist in the current driver too, I would like to fix them after we
have a first version of the new driver that works for the other
platforms.
I hope we can figure out the other issues soon.
Thanks,
Frieder
Post by Boris Brezillon
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
I do have a setup with qspi0 having two SPI memories connected
(2x16 MiB), and I'm wondering if anybody has some more info
(What's more is that there is a bug in
^^^^^^^^^^^^^^^^^^^^^^^^ - It would be handy to have the
exact explanation of this issue.
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
https://elixir.bootlin.com/linux/v4.19-rc4/source/drivers/mtd/spi-nor/fsl-quadspi.c#L671
^^^^^^^^^^^^ - [1]
I've looked a bit closer to the new spi-fsl-qspi.c driver and it
looks like it uses fsl_qspi_exec_op() and then calls
fsl_qspi_do_op().
In the last function we setup the QUADSPI_IPCR register to initiate
the transfer between SPI-NOR and the internal RX buffer.
Please correct me if I'm wrong but it seems to me like we are using
"IP Command Read" approach here.
"IP Read" is only used if data size is smaller then RX FIFO size (128
bytes for Vybrid).
I would like to just point out that the "legacy" (in-kernel) driver
(fsl_quadspi.c) is using AHB transfers for even small chunks of data
(even RX FIFO size).
Post by Frieder Schrempf
For bigger chunks of data "AHB Read" is used.
Post by Boris Brezillon
If this is true - what was the rationale to use it in the new
driver, as in the old one the "AHB Command Read" approach was
recommended (i.e. faster) [1]?
As explained above we only use "IP Read" for small chunks of data. To
speed up the tranfer for bigger chunks we do use "AHB Read".
And in the future we plan to add an interface for mapping memory, so
we can use the "AHB Read" even more efficiently.
It looks like I do read the AHB internal QUADSPI buffer before it is
filled up completely.
(I've asked the NXP community support about it,
but no reply so far https://community.nxp.com/thread/485139).
Post by Frieder Schrempf
Post by Boris Brezillon
To be even more problematic the info in link [1] states that on
vybrid there is a bug on the "IP Command Read" HW block and it
should be avoided.
If "IP Read" should be avoided completely for Vybrid, we have to add
some kind of quirk.
Yes, I think so. According to the in-code comment the "IP Read" mode
shall be avoided as it has some bugs.
Post by Frieder Schrempf
Thanks for you comments,
Frieder
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Boris Brezillon
Post by Boris Brezillon
Post by Frieder Schrempf
Post by Lukasz Majewski
I've googled for some errata or known issues for vybryd's QSPI
IP block (vf610) but without luck so far ...
Unfortunately I don't know the background for this comment.
The comment was added by some Freescale employee when the driver
was added to Linux (by Huang - CC'ed).
Post by Frieder Schrempf
Is your
board using VF610?
Yes, it uses vf610 (A5 + M4 cores).
Post by Frieder Schrempf
Do you experience problems?
For the current QSPI ML driver (fsl-quadspi.c - 4.19-rc3) only
half of the AHB buffer is valid. When I do read the whole one -
I do see read data corruption (on UBI or raw memory).
https://patchwork.ozlabs.org/patch/675401/
Unfortunately, I did not found any info regarding this problem
(in NXP's errata or other doc).
I will check if this issue shows up on new patches.
Thanks in advance for your help.
Post by Frieder Schrempf
Regards,
Frieder
______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang
Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang
Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell,
______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang
Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell,
Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: ***@denx.de
Yogesh Narayan Gaur
2018-09-06 11:11:57 UTC
Permalink
Hi Boris,
-----Original Message-----
Sent: Saturday, August 4, 2018 7:07 PM
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Hi Han,
On Thu, 2 Aug 2018 21:58:48 +0000
Post by Han Xu
-----Original Message-----
Sent: Thursday, August 2, 2018 8:09 AM
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
Ping.
I'm not sure if my message below went out to you at all. At least I
can't find it in the ML archive.
I still hope someone can help with the questions below.
Meanwhile for the second point I did some tests myself with one chip
on each of the two buses and it worked fine with my latest v2 patches.
So I'm not sure at all why Yogesh has problems with his setup (two
chips on the first bus).
Tried to test the v2 patch set on i.MX6SX SDB board but get the memory map
failure.
Post by Han Xu
[ 1.298633] fsl-quadspi 21e4000.qspi: ioremap failed for resource [mem
0x70000000-0x7fffffff]
Post by Han Xu
[ 1.307330] fsl-quadspi 21e4000.qspi: Freescale QuadSPI probe failed
[ 1.313922] fsl-quadspi: probe of 21e4000.qspi failed with error -12
This is the reason why dynamic ioremap added in previous driver, please refer to
https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpat
chwork.ozlabs.org%2Fpatch%2F503655%2F&amp;data=02%7C01%7Cyogeshnar
ayan
Post by Han Xu
.gaur%40nxp.com%7Caed89ff1b0ac4936acd408d5fa0f7303%7C686ea1d3bc2
b4c6fa
92cd99c5c301635%7C0%7C0%7C636689866643724038&amp;sdata=lZnBc0m%
2BOp8yY
Post by Han Xu
JBFcNBEa2HSoNMhmJjeM9cIoGeVs0E%3D&amp;reserved=0
We can reduce the size of the iomap to 2k * 4, since this is all we use currently.
Can you try to change the size of the ioremap call to 16k and tell us if it works.
1/ is there an easy way to invalidate AHB buffers? I mean, not
something that implies a full reset + several milliseconds of delay
after the reset. Right now we trick the caching logic by mapping a
portion that is twice the size of the buffer and switching from one
sub-portion to this other to trigger a real read on each read
access, but that's hack-ish, and I'd be surprised if HW
engineers hadn't planned for this "manual AHB buffer flush" case.
I had a discussion with the HW IP owner, they said that IP command and AHB command are two separate sets of accessing method to flash.
Memory coherency between IP and AHB command can't be maintained by the hardware.
So after every IP data write command (write, erase, write reg) AHB RX buffer needs to be flushed.

Software Reset is the safest way to achieve this.
Adding of the delay in several millisecond after setting of the Reset Bits is too conservative, can be tried with the reduced value.
2/ if we use DMA, do you know what happens when the TX FIFO runs out
of data while the TX request is not finished yet. In PIO mode, it
seems the engine sends garbage on the bus when that happens, and we
definitely don't want that.
For this query, they have said TX FIFO is the max limit, if data send more than TX FIFO size then it would results in un-defined behavior and there would be data corruption.

--
Regards
Yogesh Gaur.
While #1 is not blocking us, #2 is if we don't have those patches [1][2] applied,
and Marek wanted to be sure there was no other ways to solve the "TX FIFO
starvation" issue before considering these changes. So that'd be great if
someone from NXP could have a look/ask around and give us answers to those 2
questions.
Thanks,
Boris
[1]https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpatc
hwork.ozlabs.org%2Fpatch%2F928677%2F&amp;data=02%7C01%7Cyogeshnara
yan.gaur%40nxp.com%7Caed89ff1b0ac4936acd408d5fa0f7303%7C686ea1d3bc
2b4c6fa92cd99c5c301635%7C0%7C0%7C636689866643724038&amp;sdata=Iw
nETfWcL%2BmDkgFPgAjn73C33a877%2BsZz37qhUdLG0I%3D&amp;reserved=0
[2]https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpatc
hwork.ozlabs.org%2Fpatch%2F928678%2F&amp;data=02%7C01%7Cyogeshnara
yan.gaur%40nxp.com%7Caed89ff1b0ac4936acd408d5fa0f7303%7C686ea1d3bc
2b4c6fa92cd99c5c301635%7C0%7C0%7C636689866643724038&amp;sdata=thf
I9yDC2epen9Zp18ekku%2FVLsjUIdX6EIPENBbY8xE%3D&amp;reserved=0
Boris Brezillon
2018-09-06 11:36:00 UTC
Permalink
On Thu, 6 Sep 2018 11:11:57 +0000
Post by Frieder Schrempf
Hi Boris,
-----Original Message-----
Sent: Saturday, August 4, 2018 7:07 PM
QuadSPI controller
Hi Han,
On Thu, 2 Aug 2018 21:58:48 +0000
Post by Han Xu
-----Original Message-----
Sent: Thursday, August 2, 2018 8:09 AM
Subject: Re: Questions about the Freescale/NXP QuadSPI
controller
Ping.
I'm not sure if my message below went out to you at all. At
least I can't find it in the ML archive.
I still hope someone can help with the questions below.
Meanwhile for the second point I did some tests myself with one
chip on each of the two buses and it worked fine with my latest
v2 patches. So I'm not sure at all why Yogesh has problems with
his setup (two chips on the first bus).
Tried to test the v2 patch set on i.MX6SX SDB board but get the memory map
failure.
Post by Han Xu
[ 1.298633] fsl-quadspi 21e4000.qspi: ioremap failed for resource [mem
0x70000000-0x7fffffff]
Post by Han Xu
[ 1.307330] fsl-quadspi 21e4000.qspi: Freescale QuadSPI probe
failed [ 1.313922] fsl-quadspi: probe of 21e4000.qspi failed
with error -12
This is the reason why dynamic ioremap added in previous driver, please refer to
https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpat
chwork.ozlabs.org%2Fpatch%2F503655%2F&amp;data=02%7C01%7Cyogeshnar
ayan
Post by Han Xu
.gaur%40nxp.com%7Caed89ff1b0ac4936acd408d5fa0f7303%7C686ea1d3bc2
b4c6fa
92cd99c5c301635%7C0%7C0%7C636689866643724038&amp;sdata=lZnBc0m%
2BOp8yY
Post by Han Xu
JBFcNBEa2HSoNMhmJjeM9cIoGeVs0E%3D&amp;reserved=0
We can reduce the size of the iomap to 2k * 4, since this is all we
use currently. Can you try to change the size of the ioremap call
to 16k and tell us if it works.
1/ is there an easy way to invalidate AHB buffers? I mean, not
something that implies a full reset + several milliseconds of
delay after the reset. Right now we trick the caching logic by
mapping a portion that is twice the size of the buffer and
switching from one sub-portion to this other to trigger a real read
on each read access, but that's hack-ish, and I'd be surprised if HW
engineers hadn't planned for this "manual AHB buffer flush" case.
I had a discussion with the HW IP owner, they said that IP command
and AHB command are two separate sets of accessing method to flash.
Memory coherency between IP and AHB command can't be maintained by
the hardware. So after every IP data write command (write, erase,
write reg) AHB RX buffer needs to be flushed.
Software Reset is the safest way to achieve this.
Adding of the delay in several millisecond after setting of the Reset
Bits is too conservative, can be tried with the reduced value.
Do you know exactly how many cycles/nanoseconds we should wait? Is
there a status reg that says when the block is ready to be used again?
Post by Frieder Schrempf
2/ if we use DMA, do you know what happens when the TX FIFO runs out
of data while the TX request is not finished yet. In PIO mode, it
seems the engine sends garbage on the bus when that happens, and
we definitely don't want that.
For this query, they have said TX FIFO is the max limit, if data send
more than TX FIFO size then it would results in un-defined behavior
and there would be data corruption.
Okay.
Marek, I guess we have a good reason to accept patch [1] now.

[1]http://patchwork.ozlabs.org/patch/928677/
Yogesh Narayan Gaur
2018-09-06 12:22:56 UTC
Permalink
Hi Boris,
-----Original Message-----
Sent: Thursday, September 6, 2018 5:06 PM
Subject: Re: Questions about the Freescale/NXP QuadSPI controller
On Thu, 6 Sep 2018 11:11:57 +0000
Post by Frieder Schrempf
Hi Boris,
-----Original Message-----
Sent: Saturday, August 4, 2018 7:07 PM
Gaur
Post by Frieder Schrempf
QuadSPI controller
Hi Han,
On Thu, 2 Aug 2018 21:58:48 +0000
Post by Han Xu
-----Original Message-----
Sent: Thursday, August 2, 2018 8:09 AM
Subject: Re: Questions about the Freescale/NXP QuadSPI
controller
Ping.
I'm not sure if my message below went out to you at all. At
least I can't find it in the ML archive.
I still hope someone can help with the questions below.
Meanwhile for the second point I did some tests myself with one
chip on each of the two buses and it worked fine with my latest
v2 patches. So I'm not sure at all why Yogesh has problems with
his setup (two chips on the first bus).
Tried to test the v2 patch set on i.MX6SX SDB board but get the memory map
failure.
Post by Han Xu
[ 1.298633] fsl-quadspi 21e4000.qspi: ioremap failed for resource [mem
0x70000000-0x7fffffff]
Post by Han Xu
[ 1.307330] fsl-quadspi 21e4000.qspi: Freescale QuadSPI probe
failed [ 1.313922] fsl-quadspi: probe of 21e4000.qspi failed
with error -12
This is the reason why dynamic ioremap added in previous driver,
please refer to
https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2
Fpat
chwork.ozlabs.org%2Fpatch%2F503655%2F&amp;data=02%7C01%7Cyogeshnar
Post by Frieder Schrempf
ayan
Post by Han Xu
.gaur%40nxp.com%7Caed89ff1b0ac4936acd408d5fa0f7303%7C686ea1d3
bc2
Post by Frieder Schrempf
b4c6fa
92cd99c5c301635%7C0%7C0%7C636689866643724038&amp;sdata=lZnBc0m%
Post by Frieder Schrempf
2BOp8yY
Post by Han Xu
JBFcNBEa2HSoNMhmJjeM9cIoGeVs0E%3D&amp;reserved=0
We can reduce the size of the iomap to 2k * 4, since this is all we
use currently. Can you try to change the size of the ioremap call to
16k and tell us if it works.
1/ is there an easy way to invalidate AHB buffers? I mean, not
something that implies a full reset + several milliseconds of
delay after the reset. Right now we trick the caching logic by
mapping a portion that is twice the size of the buffer and switching
from one sub-portion to this other to trigger a real read on each
read access, but that's hack-ish, and I'd be surprised if HW
engineers hadn't planned for this "manual AHB buffer flush" case.
I had a discussion with the HW IP owner, they said that IP command and
AHB command are two separate sets of accessing method to flash.
Memory coherency between IP and AHB command can't be maintained by the
hardware. So after every IP data write command (write, erase, write
reg) AHB RX buffer needs to be flushed.
Software Reset is the safest way to achieve this.
Adding of the delay in several millisecond after setting of the Reset
Bits is too conservative, can be tried with the reduced value.
Do you know exactly how many cycles/nanoseconds we should wait? Is there a
status reg that says when the block is ready to be used again?
There is no specific cycles being mentioned in document and this was the design issue to not having these bits as w1c or any reg which provides the status.
Needs to explicitly set the RESET bits, wait for some time and then reset the RESET bit with 0.
We need to come-out with nanosecond number with the experiment.
Post by Frieder Schrempf
2/ if we use DMA, do you know what happens when the TX FIFO runs out
of data while the TX request is not finished yet. In PIO mode, it
seems the engine sends garbage on the bus when that happens, and
we definitely don't want that.
For this query, they have said TX FIFO is the max limit, if data send
more than TX FIFO size then it would results in un-defined behavior
and there would be data corruption.
Okay.
Marek, I guess we have a good reason to accept patch [1] now.
Thanks

Regards
Yogesh Gaur
[1]https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpatc
hwork.ozlabs.org%2Fpatch%2F928677%2F&amp;data=02%7C01%7Cyogeshnara
yan.gaur%40nxp.com%7Cf8c6c7f0083c4ddf0e0708d613ecec3c%7C686ea1d3bc
2b4c6fa92cd99c5c301635%7C0%7C0%7C636718305631257433&amp;sdata=3sI
wpOn4pXDNsqfPSpJqLXKerll%2BoJ1i%2FV3WMY9cx5s%3D&amp;reserved=0
Frieder Schrempf
2018-07-05 11:15:00 UTC
Permalink
Move the documentation of the old SPI NOR driver to the place of the new
SPI memory interface based driver.

Signed-off-by: Frieder Schrempf <***@exceet.de>
---
Changes in v2:
==============
* Split the moving and editing of the dt-bindings in two patches

.../devicetree/bindings/mtd/fsl-quadspi.txt | 65 --------------------
.../devicetree/bindings/spi/spi-fsl-qspi.txt | 65 ++++++++++++++++++++
2 files changed, 65 insertions(+), 65 deletions(-)

diff --git a/Documentation/devicetree/bindings/mtd/fsl-quadspi.txt b/Documentation/devicetree/bindings/mtd/fsl-quadspi.txt
deleted file mode 100644
index 483e9cf..0000000
--- a/Documentation/devicetree/bindings/mtd/fsl-quadspi.txt
+++ /dev/null
@@ -1,65 +0,0 @@
-* Freescale Quad Serial Peripheral Interface(QuadSPI)
-
-Required properties:
- - compatible : Should be "fsl,vf610-qspi", "fsl,imx6sx-qspi",
- "fsl,imx7d-qspi", "fsl,imx6ul-qspi",
- "fsl,ls1021a-qspi"
- or
- "fsl,ls2080a-qspi" followed by "fsl,ls1021a-qspi",
- "fsl,ls1043a-qspi" followed by "fsl,ls1021a-qspi"
- - reg : the first contains the register location and length,
- the second contains the memory mapping address and length
- - reg-names: Should contain the reg names "QuadSPI" and "QuadSPI-memory"
- - interrupts : Should contain the interrupt for the device
- - clocks : The clocks needed by the QuadSPI controller
- - clock-names : Should contain the name of the clocks: "qspi_en" and "qspi".
-
-Optional properties:
- - fsl,qspi-has-second-chip: The controller has two buses, bus A and bus B.
- Each bus can be connected with two NOR flashes.
- Most of the time, each bus only has one NOR flash
- connected, this is the default case.
- But if there are two NOR flashes connected to the
- bus, you should enable this property.
- (Please check the board's schematic.)
- - big-endian : That means the IP register is big endian
-
-Example:
-
-qspi0: ***@40044000 {
- compatible = "fsl,vf610-qspi";
- reg = <0x40044000 0x1000>, <0x20000000 0x10000000>;
- reg-names = "QuadSPI", "QuadSPI-memory";
- interrupts = <0 24 IRQ_TYPE_LEVEL_HIGH>;
- clocks = <&clks VF610_CLK_QSPI0_EN>,
- <&clks VF610_CLK_QSPI0>;
- clock-names = "qspi_en", "qspi";
-
- flash0: ***@0 {
- ....
- };
-};
-
-Example showing the usage of two SPI NOR devices:
-
-&qspi2 {
- pinctrl-names = "default";
- pinctrl-0 = <&pinctrl_qspi2>;
- status = "okay";
-
- flash0: ***@0 {
- #address-cells = <1>;
- #size-cells = <1>;
- compatible = "micron,n25q256a", "jedec,spi-nor";
- spi-max-frequency = <29000000>;
- reg = <0>;
- };
-
- flash1: ***@1 {
- #address-cells = <1>;
- #size-cells = <1>;
- compatible = "micron,n25q256a", "jedec,spi-nor";
- spi-max-frequency = <29000000>;
- reg = <1>;
- };
-};
diff --git a/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt b/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
new file mode 100644
index 0000000..483e9cf
--- /dev/null
+++ b/Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
@@ -0,0 +1,65 @@
+* Freescale Quad Serial Peripheral Interface(QuadSPI)
+
+Required properties:
+ - compatible : Should be "fsl,vf610-qspi", "fsl,imx6sx-qspi",
+ "fsl,imx7d-qspi", "fsl,imx6ul-qspi",
+ "fsl,ls1021a-qspi"
+ or
+ "fsl,ls2080a-qspi" followed by "fsl,ls1021a-qspi",
+ "fsl,ls1043a-qspi" followed by "fsl,ls1021a-qspi"
+ - reg : the first contains the register location and length,
+ the second contains the memory mapping address and length
+ - reg-names: Should contain the reg names "QuadSPI" and "QuadSPI-memory"
+ - interrupts : Should contain the interrupt for the device
+ - clocks : The clocks needed by the QuadSPI controller
+ - clock-names : Should contain the name of the clocks: "qspi_en" and "qspi".
+
+Optional properties:
+ - fsl,qspi-has-second-chip: The controller has two buses, bus A and bus B.
+ Each bus can be connected with two NOR flashes.
+ Most of the time, each bus only has one NOR flash
+ connected, this is the default case.
+ But if there are two NOR flashes connected to the
+ bus, you should enable this property.
+ (Please check the board's schematic.)
+ - big-endian : That means the IP register is big endian
+
+Example:
+
+qspi0: ***@40044000 {
+ compatible = "fsl,vf610-qspi";
+ reg = <0x40044000 0x1000>, <0x20000000 0x10000000>;
+ reg-names = "QuadSPI", "QuadSPI-memory";
+ interrupts = <0 24 IRQ_TYPE_LEVEL_HIGH>;
+ clocks = <&clks VF610_CLK_QSPI0_EN>,
+ <&clks VF610_CLK_QSPI0>;
+ clock-names = "qspi_en", "qspi";
+
+ flash0: ***@0 {
+ ....
+ };
+};
+
+Example showing the usage of two SPI NOR devices:
+
+&qspi2 {
+ pinctrl-names = "default";
+ pinctrl-0 = <&pinctrl_qspi2>;
+ status = "okay";
+
+ flash0: ***@0 {
+ #address-cells = <1>;
+ #size-cells = <1>;
+ compatible = "micron,n25q256a", "jedec,spi-nor";
+ spi-max-frequency = <29000000>;
+ reg = <0>;
+ };
+
+ flash1: ***@1 {
+ #address-cells = <1>;
+ #size-cells = <1>;
+ compatible = "micron,n25q256a", "jedec,spi-nor";
+ spi-max-frequency = <29000000>;
+ reg = <1>;
+ };
+};
--
2.7.4
Rob Herring
2018-07-11 15:54:04 UTC
Permalink
Post by Frieder Schrempf
Move the documentation of the old SPI NOR driver to the place of the new
SPI memory interface based driver.
---
==============
* Split the moving and editing of the dt-bindings in two patches
.../devicetree/bindings/mtd/fsl-quadspi.txt | 65 --------------------
.../devicetree/bindings/spi/spi-fsl-qspi.txt | 65 ++++++++++++++++++++
2 files changed, 65 insertions(+), 65 deletions(-)
Next time, use the -M option so the move shows any real diffs.

Acked-by: Rob Herring <***@kernel.org>
Frieder Schrempf
2018-07-12 08:11:20 UTC
Permalink
Hi Rob,
Post by Rob Herring
Post by Frieder Schrempf
Move the documentation of the old SPI NOR driver to the place of the new
SPI memory interface based driver.
---
==============
* Split the moving and editing of the dt-bindings in two patches
.../devicetree/bindings/mtd/fsl-quadspi.txt | 65 --------------------
.../devicetree/bindings/spi/spi-fsl-qspi.txt | 65 ++++++++++++++++++++
2 files changed, 65 insertions(+), 65 deletions(-)
Next time, use the -M option so the move shows any real diffs.
Ok, I'll keep that in mind.

Thanks,
Frieder
Frieder Schrempf
2018-07-05 11:15:04 UTC
Permalink
The new driver at spi/spi-fsl-qspi.c replaces the old SPI NOR driver
at mtd/fsl-quadspi.c. Switch to the new driver in the defconfigs.

Signed-off-by: Frieder Schrempf <***@exceet.de>
---
arch/arm/configs/imx_v6_v7_defconfig | 2 +-
arch/arm/configs/multi_v7_defconfig | 2 +-
2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/arm/configs/imx_v6_v7_defconfig b/arch/arm/configs/imx_v6_v7_defconfig
index f70507a..d07a535 100644
--- a/arch/arm/configs/imx_v6_v7_defconfig
+++ b/arch/arm/configs/imx_v6_v7_defconfig
@@ -111,7 +111,6 @@ CONFIG_MTD_NAND_GPMI_NAND=y
CONFIG_MTD_NAND_VF610_NFC=y
CONFIG_MTD_NAND_MXC=y
CONFIG_MTD_SPI_NOR=y
-CONFIG_SPI_FSL_QUADSPI=y
CONFIG_MTD_UBI=y
CONFIG_MTD_UBI_FASTMAP=y
CONFIG_MTD_UBI_BLOCK=y
@@ -199,6 +198,7 @@ CONFIG_I2C_ALGOPCA=m
CONFIG_I2C_GPIO=y
CONFIG_I2C_IMX=y
CONFIG_SPI=y
+CONFIG_SPI_FSL_QSPI=y
CONFIG_SPI_GPIO=y
CONFIG_SPI_IMX=y
CONFIG_SPI_FSL_DSPI=y
diff --git a/arch/arm/configs/multi_v7_defconfig b/arch/arm/configs/multi_v7_defconfig
index 8f6be19..da4db3b 100644
--- a/arch/arm/configs/multi_v7_defconfig
+++ b/arch/arm/configs/multi_v7_defconfig
@@ -189,7 +189,6 @@ CONFIG_MTD_NAND_BRCMNAND=y
CONFIG_MTD_NAND_VF610_NFC=y
CONFIG_MTD_NAND_DAVINCI=y
CONFIG_MTD_SPI_NOR=y
-CONFIG_SPI_FSL_QUADSPI=m
CONFIG_MTD_UBI=y
CONFIG_BLK_DEV_LOOP=y
CONFIG_BLK_DEV_RAM=y
@@ -377,6 +376,7 @@ CONFIG_SPI_BCM2835=y
CONFIG_SPI_BCM2835AUX=y
CONFIG_SPI_CADENCE=y
CONFIG_SPI_DAVINCI=y
+CONFIG_SPI_FSL_QSPI=m
CONFIG_SPI_GPIO=m
CONFIG_SPI_FSL_DSPI=m
CONFIG_SPI_OMAP24XX=y
--
2.7.4
Boris Brezillon
2018-07-05 12:39:48 UTC
Permalink
Hi Frieder,

On Thu, 5 Jul 2018 13:14:57 +0200
Post by Frieder Schrempf
When porting (Q)SPI controller drivers from the MTD layer to the SPI
layer, the naming scheme for the memory devices changes. To be able
to keep compatibility with the old drivers naming scheme, a function
is added to the SPI mem interface to let controller drivers set a
custom name for the memory.
Name with the old driver: 21e0000.qspi,
or with multiple devices: 21e0000.qspi-0, 21e0000.qspi-1, ...
Name with the new driver without spi_mem_get_name: spi4.0
---
==============
* Add a name field to struct spi_mem and fill it while probing
drivers/spi/spi-mem.c | 30 ++++++++++++++++++++++++++++++
include/linux/spi/spi-mem.h | 7 ++++++-
2 files changed, 36 insertions(+), 1 deletion(-)
diff --git a/drivers/spi/spi-mem.c b/drivers/spi/spi-mem.c
index 990770d..048101c 100644
--- a/drivers/spi/spi-mem.c
+++ b/drivers/spi/spi-mem.c
@@ -311,6 +311,26 @@ int spi_mem_exec_op(struct spi_mem *mem, const struct spi_mem_op *op)
EXPORT_SYMBOL_GPL(spi_mem_exec_op);
/**
+ * spi_mem_get_name() - Let drivers using the SPI mem interface specify a
+ * custom name for the memory to avoid compatibility
+ * issues with ported drivers.
You're not describing what this function does, but why it was
introduced. Not sure the spi-mem user/driver cares about that, these are
just internal details. Probably something you should put in the
->get_name() hook doc.
Post by Frieder Schrempf
+ *
+ * When porting (Q)SPI controller drivers from the MTD layer to the SPI
+ * layer, the naming scheme for the memory devices changes. To be able to
+ * keep compatibility with the old drivers naming scheme, this function can
+ * be used to get a custom name from the controller driver.
+ * If no custom name is available, the name of the SPI device is returned.
Ditto. Just say that this function returns the spi-mem device name which
can be used by the upper layer if they need to expose a user-friendly
name.
Post by Frieder Schrempf
+ *
+ * Return: a char array that contains the name for the flash memory
^ a string?

Return: a string containing the name of the memory device to be
used by the spi-mem user
Post by Frieder Schrempf
+ */
+const char *spi_mem_get_name(struct spi_mem *mem)
+{
+ return mem->name;
+}
+EXPORT_SYMBOL_GPL(spi_mem_get_name);
+
+/**
* spi_mem_adjust_op_size() - Adjust the data size of a SPI mem operation to
* match controller limitations
@@ -344,6 +364,7 @@ static inline struct spi_mem_driver *to_spi_mem_drv(struct device_driver *drv)
static int spi_mem_probe(struct spi_device *spi)
{
struct spi_mem_driver *memdrv = to_spi_mem_drv(spi->dev.driver);
+ struct spi_controller *ctlr = spi->controller;
struct spi_mem *mem;
mem = devm_kzalloc(&spi->dev, sizeof(*mem), GFP_KERNEL);
@@ -351,6 +372,15 @@ static int spi_mem_probe(struct spi_device *spi)
return -ENOMEM;
mem->spi = spi;
+
+ if (ctlr->mem_ops && ctlr->mem_ops->get_name)
+ mem->name = ctlr->mem_ops->get_name(mem);
+ else
+ mem->name = dev_name(&spi->dev);
+
+ if (IS_ERR_OR_NULL(mem->name))
+ return PTR_ERR(mem->name);
+
spi_set_drvdata(spi, mem);
return memdrv->probe(mem);
diff --git a/include/linux/spi/spi-mem.h b/include/linux/spi/spi-mem.h
index 4fa34a2..c3f82d1 100644
--- a/include/linux/spi/spi-mem.h
+++ b/include/linux/spi/spi-mem.h
@@ -124,7 +124,8 @@ struct spi_mem_op {
/**
* struct spi_mem - describes a SPI memory device
Can you fix this typo in a separate patch?
Post by Frieder Schrempf
*
* Extra information that describe the SPI memory device and may be needed by
* the controller to properly handle this device should be placed here.
@@ -135,6 +136,7 @@ struct spi_mem_op {
struct spi_mem {
struct spi_device *spi;
void *drvpriv;
+ char *name;
};
/**
@@ -178,6 +180,7 @@ struct spi_controller_mem_ops {
const struct spi_mem_op *op);
int (*exec_op)(struct spi_mem *mem,
const struct spi_mem_op *op);
+ const char *(*get_name)(struct spi_mem *mem);
You forgot to document this hook. Also, it's important to mention that
if the name returned by this function is dynamically allocated, it
should be allocated with devm_xxxx(), because we don't have a
->free_name() function.
Post by Frieder Schrempf
};
/**
@@ -236,6 +239,8 @@ bool spi_mem_supports_op(struct spi_mem *mem,
int spi_mem_exec_op(struct spi_mem *mem,
const struct spi_mem_op *op);
+const char *spi_mem_get_name(struct spi_mem *mem);
+
int spi_mem_driver_register_with_owner(struct spi_mem_driver *drv,
struct module *owner);
Thanks,

Boris
Frieder Schrempf
2018-07-05 12:50:49 UTC
Permalink
Hi Boris,
Post by Boris Brezillon
Hi Frieder,
On Thu, 5 Jul 2018 13:14:57 +0200
Post by Frieder Schrempf
When porting (Q)SPI controller drivers from the MTD layer to the SPI
layer, the naming scheme for the memory devices changes. To be able
to keep compatibility with the old drivers naming scheme, a function
is added to the SPI mem interface to let controller drivers set a
custom name for the memory.
Name with the old driver: 21e0000.qspi,
or with multiple devices: 21e0000.qspi-0, 21e0000.qspi-1, ...
Name with the new driver without spi_mem_get_name: spi4.0
---
==============
* Add a name field to struct spi_mem and fill it while probing
drivers/spi/spi-mem.c | 30 ++++++++++++++++++++++++++++++
include/linux/spi/spi-mem.h | 7 ++++++-
2 files changed, 36 insertions(+), 1 deletion(-)
diff --git a/drivers/spi/spi-mem.c b/drivers/spi/spi-mem.c
index 990770d..048101c 100644
--- a/drivers/spi/spi-mem.c
+++ b/drivers/spi/spi-mem.c
@@ -311,6 +311,26 @@ int spi_mem_exec_op(struct spi_mem *mem, const struct spi_mem_op *op)
EXPORT_SYMBOL_GPL(spi_mem_exec_op);
/**
+ * spi_mem_get_name() - Let drivers using the SPI mem interface specify a
+ * custom name for the memory to avoid compatibility
+ * issues with ported drivers.
You're not describing what this function does, but why it was
introduced. Not sure the spi-mem user/driver cares about that, these are
just internal details. Probably something you should put in the
->get_name() hook doc.
Ok. You're right.
Post by Boris Brezillon
Post by Frieder Schrempf
+ *
+ * When porting (Q)SPI controller drivers from the MTD layer to the SPI
+ * layer, the naming scheme for the memory devices changes. To be able to
+ * keep compatibility with the old drivers naming scheme, this function can
+ * be used to get a custom name from the controller driver.
+ * If no custom name is available, the name of the SPI device is returned.
Ditto. Just say that this function returns the spi-mem device name which
can be used by the upper layer if they need to expose a user-friendly
name.
Ok.
Post by Boris Brezillon
Post by Frieder Schrempf
+ *
+ * Return: a char array that contains the name for the flash memory
^ a string?
Return: a string containing the name of the memory device to be
used by the spi-mem user
Ok.
Post by Boris Brezillon
Post by Frieder Schrempf
+ */
+const char *spi_mem_get_name(struct spi_mem *mem)
+{
+ return mem->name;
+}
+EXPORT_SYMBOL_GPL(spi_mem_get_name);
+
+/**
* spi_mem_adjust_op_size() - Adjust the data size of a SPI mem operation to
* match controller limitations
@@ -344,6 +364,7 @@ static inline struct spi_mem_driver *to_spi_mem_drv(struct device_driver *drv)
static int spi_mem_probe(struct spi_device *spi)
{
struct spi_mem_driver *memdrv = to_spi_mem_drv(spi->dev.driver);
+ struct spi_controller *ctlr = spi->controller;
struct spi_mem *mem;
mem = devm_kzalloc(&spi->dev, sizeof(*mem), GFP_KERNEL);
@@ -351,6 +372,15 @@ static int spi_mem_probe(struct spi_device *spi)
return -ENOMEM;
mem->spi = spi;
+
+ if (ctlr->mem_ops && ctlr->mem_ops->get_name)
+ mem->name = ctlr->mem_ops->get_name(mem);
+ else
+ mem->name = dev_name(&spi->dev);
+
+ if (IS_ERR_OR_NULL(mem->name))
+ return PTR_ERR(mem->name);
+
spi_set_drvdata(spi, mem);
return memdrv->probe(mem);
diff --git a/include/linux/spi/spi-mem.h b/include/linux/spi/spi-mem.h
index 4fa34a2..c3f82d1 100644
--- a/include/linux/spi/spi-mem.h
+++ b/include/linux/spi/spi-mem.h
@@ -124,7 +124,8 @@ struct spi_mem_op {
/**
* struct spi_mem - describes a SPI memory device
Can you fix this typo in a separate patch?
Ok. I'll move it to a separate patch.
Post by Boris Brezillon
Post by Frieder Schrempf
*
* Extra information that describe the SPI memory device and may be needed by
* the controller to properly handle this device should be placed here.
@@ -135,6 +136,7 @@ struct spi_mem_op {
struct spi_mem {
struct spi_device *spi;
void *drvpriv;
+ char *name;
};
/**
@@ -178,6 +180,7 @@ struct spi_controller_mem_ops {
const struct spi_mem_op *op);
int (*exec_op)(struct spi_mem *mem,
const struct spi_mem_op *op);
+ const char *(*get_name)(struct spi_mem *mem);
You forgot to document this hook. Also, it's important to mention that
if the name returned by this function is dynamically allocated, it
should be allocated with devm_xxxx(), because we don't have a
->free_name() function.
Indeed. I'll add the documentation.

Thanks,
Frieder
Post by Boris Brezillon
Post by Frieder Schrempf
};
/**
@@ -236,6 +239,8 @@ bool spi_mem_supports_op(struct spi_mem *mem,
int spi_mem_exec_op(struct spi_mem *mem,
const struct spi_mem_op *op);
+const char *spi_mem_get_name(struct spi_mem *mem);
+
int spi_mem_driver_register_with_owner(struct spi_mem_driver *drv,
struct module *owner);
Thanks,
Boris
Yogesh Narayan Gaur
2018-07-06 05:08:51 UTC
Permalink
Hi Frieder,

These patch series also depends on the patches [1][2] for the controller which can't write nor->page_size bytes in single write command step.
Thus, have to invoke Write sequence again and before sending WRITE command, need to send WREN command.

[1] https://patchwork.ozlabs.org/patch/928677/
[2] https://patchwork.ozlabs.org/patch/928678/

--
Regards
Yogesh Gaur.
-----Original Message-----
Sent: Thursday, July 5, 2018 4:45 PM
Subject: [PATCH v2 00/12] Port the FSL QSPI driver to the SPI framework
Now that the SPI memory interface was introduced by Boris [1], it is possible to
move drivers from mtd/spi-nor to the SPI framework in order to use them for
different type of SPI memory chips.
Patch 1 adds a function spi_mem_get_name() to the SPI memory interface and a
->name field to struct spi_mem.
Patch 2 uses it in m25p80.c to make it possible for SPI controller drivers to
provide a custom naming scheme for the flash chip.
This is needed to avoid breaking compatibility of mtdparts when switching from
the old to the new driver.
Patch 3 adds a driver for the Freescale QSPI controller to the SPI framework.
Together with m25p80.c it can be used to interface SPI NOR flashes just as the
old driver did. For this to work properly a few minor changes to the devicetrees
are necessary (see patches 5 to 7).
Patch 8 changes the defconfigs to use the new driver and patch 9 removes the
old driver.
Patch 10 and 11 remove 'fsl,qspi-has-second-chip' from the devicetrees.
Patch 12 adjusts the MAINTAINERS file.
The read performance of the new driver is almost the same or even better than
the old driver, depending on the block size.
The write performance is a bit slower on average (~10-15%).
The new driver was also tested with the SPI NAND framework [2] and a Winbond
W25M02GV flash.
If someone has a board that uses both chips selects and/or both busses, it would
be really nice to have the driver be tested on such a setup.
[1]
https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgit.ker
nel.org%2Fpub%2Fscm%2Flinux%2Fkernel%2Fgit%2Fbroonie%2Fspi.git%2Fcom
mit%2F%3Fh%3Dfor-
4.18%26id%3Dc36ff266dc82f4ae797a6f3513c6ffa344f7f1c7&data=02%7C01%7
Cyogeshnarayan.gaur%40nxp.com%7C77846565197d44d7413008d5e268ccb2%
7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C636663862101876439&
sdata=Y8dyVYysfK3Q6vDXUMEBs%2BBygMxh9qM%2ByWdtk9hGyQU%3D&rese
rved=0
[2]
https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpatch
work.ozlabs.org%2Fcover%2F913728%2F&data=02%7C01%7Cyogeshnarayan.g
aur%40nxp.com%7C77846565197d44d7413008d5e268ccb2%7C686ea1d3bc2b4
c6fa92cd99c5c301635%7C0%7C0%7C636663862101876439&sdata=VUE5kVY7V
DW0a3PKmQvsUC9TyHZUhhntgM1SJC3EUWU%3D&reserved=0
==============
* Rebase on top of nand/next
* Add a name field to struct spi_mem and fill it while probing
* Add Yogesh Gaur and Suresh Gupta as authors
* Use GENMASK() for generating bitmasks
* Use callback functions for read/write of registers
* Attach the seq variable to the selected CS
* Avoid using conditional in read/write loop
* Avoid infinite loop and use a timeout instead
* Return error pointer when allocation in fsl_qspi_get_name() fails
* Remove redundant iounmap()
* Put suspend()/resume() in struct dev_pm_ops instead of struct platform_driver
* Split the moving and editing of the dt-bindings in two patches
spi: spi-mem: Extend the SPI mem interface to set a custom memory name
mtd: m25p80: Call spi_mem_get_name() to let controller set a custom
name
spi: Add a driver for the Freescale/NXP QuadSPI controller
dt-bindings: spi: Move the bindings for the FSL QSPI driver
dt-bindings: spi: Adjust the bindings for the FSL QSPI driver
ARM: dts: Reflect change of FSL QSPI driver and remove unused
properties
arm64: dts: Reflect change of FSL QSPI driver and remove unused
properties
ARM: defconfig: Use the new FSL QSPI driver under the SPI framework
mtd: fsl-quadspi: Remove the driver as it was replaced by
spi-fsl-qspi.c
ARM: dts: ls1021a: Remove fsl,qspi-has-second-chip as it is not used
ARM64: dts: ls1046a: Remove fsl,qspi-has-second-chip as it is not used
MAINTAINERS: Move the Freescale QSPI driver to the SPI framework
.../devicetree/bindings/mtd/fsl-quadspi.txt | 65 -
.../devicetree/bindings/spi/spi-fsl-qspi.txt | 65 +
MAINTAINERS | 4 +-
arch/arm/boot/dts/imx6sx-sdb-reva.dts | 8 +-
arch/arm/boot/dts/imx6sx-sdb.dts | 8 +-
arch/arm/boot/dts/imx6ul-14x14-evk.dtsi | 2 +
arch/arm/boot/dts/ls1021a-moxa-uc-8410a.dts | 6 +-
arch/arm/configs/imx_v6_v7_defconfig | 2 +-
arch/arm/configs/multi_v7_defconfig | 2 +-
.../boot/dts/freescale/fsl-ls1043a-qds.dts | 3 +-
.../boot/dts/freescale/fsl-ls1046a-qds.dts | 4 +-
.../boot/dts/freescale/fsl-ls1046a-rdb.dts | 6 +-
arch/arm64/boot/dts/freescale/fsl-ls1046a.dtsi | 1 -
.../boot/dts/freescale/fsl-ls208xa-qds.dtsi | 4 +
drivers/mtd/devices/m25p80.c | 3 +-
drivers/mtd/spi-nor/Kconfig | 9 -
drivers/mtd/spi-nor/Makefile | 1 -
drivers/mtd/spi-nor/fsl-quadspi.c | 1217 ------------------
drivers/spi/Kconfig | 11 +
drivers/spi/Makefile | 1 +
drivers/spi/spi-fsl-qspi.c | 954 ++++++++++++++
drivers/spi/spi-mem.c | 30 +
include/linux/spi/spi-mem.h | 7 +-
23 files changed, 1100 insertions(+), 1313 deletions(-) delete mode 100644
Documentation/devicetree/bindings/mtd/fsl-quadspi.txt
create mode 100644 Documentation/devicetree/bindings/spi/spi-fsl-qspi.txt
delete mode 100644 drivers/mtd/spi-nor/fsl-quadspi.c create mode 100644
drivers/spi/spi-fsl-qspi.c
--
2.7.4
Boris Brezillon
2018-10-31 13:40:55 UTC
Permalink
Hi Frieder, Yogesh,

On Thu, 5 Jul 2018 13:14:56 +0200
Now that the SPI memory interface was introduced by Boris [1], it is
possible to move drivers from mtd/spi-nor to the SPI framework in order
to use them for different type of SPI memory chips.
Patch 1 adds a function spi_mem_get_name() to the SPI memory interface
and a ->name field to struct spi_mem.
Patch 2 uses it in m25p80.c to make it possible for SPI controller
drivers to provide a custom naming scheme for the flash chip.
This is needed to avoid breaking compatibility of mtdparts when switching
from the old to the new driver.
Patch 3 adds a driver for the Freescale QSPI controller to the SPI
framework. Together with m25p80.c it can be used to interface SPI
NOR flashes just as the old driver did. For this to work properly a few
minor changes to the devicetrees are necessary (see patches 5 to 7).
Patch 8 changes the defconfigs to use the new driver and patch 9 removes
the old driver.
Patch 10 and 11 remove 'fsl,qspi-has-second-chip' from the devicetrees.
Patch 12 adjusts the MAINTAINERS file.
The read performance of the new driver is almost the same or even better
than the old driver, depending on the block size.
The write performance is a bit slower on average (~10-15%).
The new driver was also tested with the SPI NAND framework [2] and a
Winbond W25M02GV flash.
If someone has a board that uses both chips selects and/or both busses,
it would be really nice to have the driver be tested on such a setup.
Any progress on this front? Yogesh, can you please remind us the
remaining issues? I'd really like to make some progress, otherwise the
conversion to spi-mem will take ages.

Regards,

Boris
Schrempf Frieder
2018-10-31 13:54:29 UTC
Permalink
Hi Boris,
Post by Boris Brezillon
Hi Frieder, Yogesh,
On Thu, 5 Jul 2018 13:14:56 +0200
Now that the SPI memory interface was introduced by Boris [1], it is
possible to move drivers from mtd/spi-nor to the SPI framework in order
to use them for different type of SPI memory chips.
Patch 1 adds a function spi_mem_get_name() to the SPI memory interface
and a ->name field to struct spi_mem.
Patch 2 uses it in m25p80.c to make it possible for SPI controller
drivers to provide a custom naming scheme for the flash chip.
This is needed to avoid breaking compatibility of mtdparts when switching
from the old to the new driver.
Patch 3 adds a driver for the Freescale QSPI controller to the SPI
framework. Together with m25p80.c it can be used to interface SPI
NOR flashes just as the old driver did. For this to work properly a few
minor changes to the devicetrees are necessary (see patches 5 to 7).
Patch 8 changes the defconfigs to use the new driver and patch 9 removes
the old driver.
Patch 10 and 11 remove 'fsl,qspi-has-second-chip' from the devicetrees.
Patch 12 adjusts the MAINTAINERS file.
The read performance of the new driver is almost the same or even better
than the old driver, depending on the block size.
The write performance is a bit slower on average (~10-15%).
The new driver was also tested with the SPI NAND framework [2] and a
Winbond W25M02GV flash.
If someone has a board that uses both chips selects and/or both busses,
it would be really nice to have the driver be tested on such a setup.
Any progress on this front? Yogesh, can you please remind us the
remaining issues? I'd really like to make some progress, otherwise the
conversion to spi-mem will take ages.
I definitely want to continue this. I just did not have the time to work
on it.

I think the only remaining blocking issues is the one that Yogesh
reported while testing with two chips on the same bus.

Han and I have both successfully tested with two chips, but on separate
buses.

Regards,
Frieder
Boris Brezillon
2018-10-31 14:31:25 UTC
Permalink
On Wed, 31 Oct 2018 13:54:29 +0000
Post by Frieder Schrempf
Hi Boris,
Post by Boris Brezillon
Hi Frieder, Yogesh,
On Thu, 5 Jul 2018 13:14:56 +0200
Now that the SPI memory interface was introduced by Boris [1], it is
possible to move drivers from mtd/spi-nor to the SPI framework in order
to use them for different type of SPI memory chips.
Patch 1 adds a function spi_mem_get_name() to the SPI memory interface
and a ->name field to struct spi_mem.
Patch 2 uses it in m25p80.c to make it possible for SPI controller
drivers to provide a custom naming scheme for the flash chip.
This is needed to avoid breaking compatibility of mtdparts when switching
from the old to the new driver.
Patch 3 adds a driver for the Freescale QSPI controller to the SPI
framework. Together with m25p80.c it can be used to interface SPI
NOR flashes just as the old driver did. For this to work properly a few
minor changes to the devicetrees are necessary (see patches 5 to 7).
Patch 8 changes the defconfigs to use the new driver and patch 9 removes
the old driver.
Patch 10 and 11 remove 'fsl,qspi-has-second-chip' from the devicetrees.
Patch 12 adjusts the MAINTAINERS file.
The read performance of the new driver is almost the same or even better
than the old driver, depending on the block size.
The write performance is a bit slower on average (~10-15%).
The new driver was also tested with the SPI NAND framework [2] and a
Winbond W25M02GV flash.
If someone has a board that uses both chips selects and/or both busses,
it would be really nice to have the driver be tested on such a setup.
Any progress on this front? Yogesh, can you please remind us the
remaining issues? I'd really like to make some progress, otherwise the
conversion to spi-mem will take ages.
I definitely want to continue this. I just did not have the time to work
on it.
I think the only remaining blocking issues is the one that Yogesh
reported while testing with two chips on the same bus.
Maybe you can send a new version rebased on v4.20-rc1 (when it's out)
and push it somewhere so that Yogesh can test it (again). Yogesh, can we
please make some progress on this? If you really have a bug, that'd be
great to have a serious investigation on what is causing this bug. The
explanation we had so far were not really helpful in understanding the
problem.
Yogesh Narayan Gaur
2018-10-31 16:03:55 UTC
Permalink
Hi Frieder,
-----Original Message-----
Sent: Wednesday, October 31, 2018 8:01 PM
Subject: Re: [PATCH v2 00/12] Port the FSL QSPI driver to the SPI framework
On Wed, 31 Oct 2018 13:54:29 +0000
Post by Frieder Schrempf
Hi Boris,
Post by Boris Brezillon
Hi Frieder, Yogesh,
On Thu, 5 Jul 2018 13:14:56 +0200
Now that the SPI memory interface was introduced by Boris [1], it
is possible to move drivers from mtd/spi-nor to the SPI framework
in order to use them for different type of SPI memory chips.
Patch 1 adds a function spi_mem_get_name() to the SPI memory
interface and a ->name field to struct spi_mem.
Patch 2 uses it in m25p80.c to make it possible for SPI controller
drivers to provide a custom naming scheme for the flash chip.
This is needed to avoid breaking compatibility of mtdparts when
switching from the old to the new driver.
Patch 3 adds a driver for the Freescale QSPI controller to the SPI
framework. Together with m25p80.c it can be used to interface SPI
NOR flashes just as the old driver did. For this to work properly a
few minor changes to the devicetrees are necessary (see patches 5 to 7).
Patch 8 changes the defconfigs to use the new driver and patch 9
removes the old driver.
Patch 10 and 11 remove 'fsl,qspi-has-second-chip' from the devicetrees.
Patch 12 adjusts the MAINTAINERS file.
The read performance of the new driver is almost the same or even
better than the old driver, depending on the block size.
The write performance is a bit slower on average (~10-15%).
The new driver was also tested with the SPI NAND framework [2] and
a Winbond W25M02GV flash.
If someone has a board that uses both chips selects and/or both
busses, it would be really nice to have the driver be tested on such a setup.
Any progress on this front? Yogesh, can you please remind us the
remaining issues? I'd really like to make some progress, otherwise
the conversion to spi-mem will take ages.
I definitely want to continue this. I just did not have the time to
work on it.
I think the only remaining blocking issues is the one that Yogesh
reported while testing with two chips on the same bus.
Maybe you can send a new version rebased on v4.20-rc1 (when it's out) and
push it somewhere so that Yogesh can test it (again).
In next version, please take care of the comments being raised by Boris in NXP FlexSPI controller driver patch series. E.g.
1. Redefine all mask fields and use BIT(X) for single bit set/unset.
2. Remove function pointer hooks for ->read/->write function.
3. Remove dependency from 'endianess' property from DTS node.
4. Usage of readl_poll_timeout() instead of busy looping.

--
Regards
Yogesh Gaur

Yogesh, can we please
make some progress on this? If you really have a bug, that'd be great to have a
serious investigation on what is causing this bug. The explanation we had so far
were not really helpful in understanding the problem.
Schrempf Frieder
2018-10-31 16:09:29 UTC
Permalink
Hi Yogesh,
Post by Boris Brezillon
Hi Frieder,
-----Original Message-----
Sent: Wednesday, October 31, 2018 8:01 PM
Subject: Re: [PATCH v2 00/12] Port the FSL QSPI driver to the SPI framework
On Wed, 31 Oct 2018 13:54:29 +0000
Post by Frieder Schrempf
Hi Boris,
Post by Boris Brezillon
Hi Frieder, Yogesh,
On Thu, 5 Jul 2018 13:14:56 +0200
Now that the SPI memory interface was introduced by Boris [1], it
is possible to move drivers from mtd/spi-nor to the SPI framework
in order to use them for different type of SPI memory chips.
Patch 1 adds a function spi_mem_get_name() to the SPI memory
interface and a ->name field to struct spi_mem.
Patch 2 uses it in m25p80.c to make it possible for SPI controller
drivers to provide a custom naming scheme for the flash chip.
This is needed to avoid breaking compatibility of mtdparts when
switching from the old to the new driver.
Patch 3 adds a driver for the Freescale QSPI controller to the SPI
framework. Together with m25p80.c it can be used to interface SPI
NOR flashes just as the old driver did. For this to work properly a
few minor changes to the devicetrees are necessary (see patches 5 to 7).
Patch 8 changes the defconfigs to use the new driver and patch 9
removes the old driver.
Patch 10 and 11 remove 'fsl,qspi-has-second-chip' from the devicetrees.
Patch 12 adjusts the MAINTAINERS file.
The read performance of the new driver is almost the same or even
better than the old driver, depending on the block size.
The write performance is a bit slower on average (~10-15%).
The new driver was also tested with the SPI NAND framework [2] and
a Winbond W25M02GV flash.
If someone has a board that uses both chips selects and/or both
busses, it would be really nice to have the driver be tested on such a setup.
Any progress on this front? Yogesh, can you please remind us the
remaining issues? I'd really like to make some progress, otherwise
the conversion to spi-mem will take ages.
I definitely want to continue this. I just did not have the time to
work on it.
I think the only remaining blocking issues is the one that Yogesh
reported while testing with two chips on the same bus.
Maybe you can send a new version rebased on v4.20-rc1 (when it's out) and
push it somewhere so that Yogesh can test it (again).
In next version, please take care of the comments being raised by Boris in NXP FlexSPI controller driver patch series. E.g.
1. Redefine all mask fields and use BIT(X) for single bit set/unset.
2. Remove function pointer hooks for ->read/->write function.
3. Remove dependency from 'endianess' property from DTS node.
4. Usage of readl_poll_timeout() instead of busy looping.
Thanks for reminding me. I have these on my list.

Thanks,
Frieder
Post by Boris Brezillon
--
Regards
Yogesh Gaur
Yogesh, can we please
make some progress on this? If you really have a bug, that'd be great to have a
serious investigation on what is causing this bug. The explanation we had so far
were not really helpful in understanding the problem.
Schrempf Frieder
2018-11-08 08:15:55 UTC
Permalink
This post might be inappropriate. Click to display it.
Boris Brezillon
2018-11-08 08:19:57 UTC
Permalink
This post might be inappropriate. Click to display it.
Schrempf Frieder
2018-11-08 08:35:28 UTC
Permalink
Post by Boris Brezillon
Hi Frieder,
On Thu, 8 Nov 2018 08:15:55 +0000
Post by Schrempf Frieder
Hi Boris, hi Yogesh,
Post by Boris Brezillon
On Wed, 31 Oct 2018 13:54:29 +0000
Post by Frieder Schrempf
Hi Boris,
Post by Boris Brezillon
Hi Frieder, Yogesh,
On Thu, 5 Jul 2018 13:14:56 +0200
Now that the SPI memory interface was introduced by Boris [1], it is
possible to move drivers from mtd/spi-nor to the SPI framework in order
to use them for different type of SPI memory chips.
Patch 1 adds a function spi_mem_get_name() to the SPI memory interface
and a ->name field to struct spi_mem.
Patch 2 uses it in m25p80.c to make it possible for SPI controller
drivers to provide a custom naming scheme for the flash chip.
This is needed to avoid breaking compatibility of mtdparts when switching
from the old to the new driver.
Patch 3 adds a driver for the Freescale QSPI controller to the SPI
framework. Together with m25p80.c it can be used to interface SPI
NOR flashes just as the old driver did. For this to work properly a few
minor changes to the devicetrees are necessary (see patches 5 to 7).
Patch 8 changes the defconfigs to use the new driver and patch 9 removes
the old driver.
Patch 10 and 11 remove 'fsl,qspi-has-second-chip' from the devicetrees.
Patch 12 adjusts the MAINTAINERS file.
The read performance of the new driver is almost the same or even better
than the old driver, depending on the block size.
The write performance is a bit slower on average (~10-15%).
The new driver was also tested with the SPI NAND framework [2] and a
Winbond W25M02GV flash.
If someone has a board that uses both chips selects and/or both busses,
it would be really nice to have the driver be tested on such a setup.
Any progress on this front? Yogesh, can you please remind us the
remaining issues? I'd really like to make some progress, otherwise the
conversion to spi-mem will take ages.
I definitely want to continue this. I just did not have the time to work
on it.
I think the only remaining blocking issues is the one that Yogesh
reported while testing with two chips on the same bus.
Maybe you can send a new version rebased on v4.20-rc1 (when it's out)
and push it somewhere so that Yogesh can test it (again). Yogesh, can we
please make some progress on this? If you really have a bug, that'd be
great to have a serious investigation on what is causing this bug. The
explanation we had so far were not really helpful in understanding the
problem.
I have sent a v4 that is based on v4.20-rc1. I only applied small fixes
and cosmetic changes.
There is still the hack to avoid AHB buffer invalidation delay (not sure
if we should keep it for now)
IIRC, Yogesh proposed to replace this hack by a RESET+smaller-delay.
Maybe you can try that.
Right, I will do some experiments.
Post by Boris Brezillon
Post by Schrempf Frieder
and we still need to figure out the
reported problems with two chips on one bus.
As Yogesh is the only one how has a hardware setup for this, maybe you
can retry and provide some debugging info?
Unfortunately I'll probably do not have time to switch to the dirmap API
now, so I'd rather do that as a second step, as it probably won't solve
our problems either.
Totally agree with that.
Thanks for sending a v4.
Boris
Schrempf Frieder
2018-11-08 08:57:19 UTC
Permalink
Post by Boris Brezillon
Hi Frieder,
On Thu, 8 Nov 2018 08:15:55 +0000
Post by Schrempf Frieder
Hi Boris, hi Yogesh,
Post by Boris Brezillon
On Wed, 31 Oct 2018 13:54:29 +0000
Post by Frieder Schrempf
Hi Boris,
Post by Boris Brezillon
Hi Frieder, Yogesh,
On Thu, 5 Jul 2018 13:14:56 +0200
Now that the SPI memory interface was introduced by Boris [1], it is
possible to move drivers from mtd/spi-nor to the SPI framework in order
to use them for different type of SPI memory chips.
Patch 1 adds a function spi_mem_get_name() to the SPI memory interface
and a ->name field to struct spi_mem.
Patch 2 uses it in m25p80.c to make it possible for SPI controller
drivers to provide a custom naming scheme for the flash chip.
This is needed to avoid breaking compatibility of mtdparts when switching
from the old to the new driver.
Patch 3 adds a driver for the Freescale QSPI controller to the SPI
framework. Together with m25p80.c it can be used to interface SPI
NOR flashes just as the old driver did. For this to work properly a few
minor changes to the devicetrees are necessary (see patches 5 to 7).
Patch 8 changes the defconfigs to use the new driver and patch 9 removes
the old driver.
Patch 10 and 11 remove 'fsl,qspi-has-second-chip' from the devicetrees.
Patch 12 adjusts the MAINTAINERS file.
The read performance of the new driver is almost the same or even better
than the old driver, depending on the block size.
The write performance is a bit slower on average (~10-15%).
The new driver was also tested with the SPI NAND framework [2] and a
Winbond W25M02GV flash.
If someone has a board that uses both chips selects and/or both busses,
it would be really nice to have the driver be tested on such a setup.
Any progress on this front? Yogesh, can you please remind us the
remaining issues? I'd really like to make some progress, otherwise the
conversion to spi-mem will take ages.
I definitely want to continue this. I just did not have the time to work
on it.
I think the only remaining blocking issues is the one that Yogesh
reported while testing with two chips on the same bus.
Maybe you can send a new version rebased on v4.20-rc1 (when it's out)
and push it somewhere so that Yogesh can test it (again). Yogesh, can we
please make some progress on this? If you really have a bug, that'd be
great to have a serious investigation on what is causing this bug. The
explanation we had so far were not really helpful in understanding the
problem.
I have sent a v4 that is based on v4.20-rc1. I only applied small fixes
and cosmetic changes.
I forgot to add, that I have pushed the v4 here if someone needs it:

https://github.com/fschrempf/linux/commits/fsl-qspi-next-4
Post by Boris Brezillon
Post by Schrempf Frieder
There is still the hack to avoid AHB buffer invalidation delay (not sure
if we should keep it for now)
IIRC, Yogesh proposed to replace this hack by a RESET+smaller-delay.
Maybe you can try that.
Post by Schrempf Frieder
and we still need to figure out the
reported problems with two chips on one bus.
As Yogesh is the only one how has a hardware setup for this, maybe you
can retry and provide some debugging info?
Unfortunately I'll probably do not have time to switch to the dirmap API
now, so I'd rather do that as a second step, as it probably won't solve
our problems either.
Totally agree with that.
Thanks for sending a v4.
Boris
Loading...