[PATCH v3] fpga: mgr: altera-ps-spi: enable usage on non-dt platforms

[PATCH v3] fpga: mgr: altera-ps-spi: enable usage on non-dt platforms

[Anatolij Gustschin]

Driver probing fails on non-dt platforms since of_match_device()
always returns NULL here. Add device names and matching driver
data to the spi_device_id table. This allows driver binding to
dynamically added PS-SPI devices (e.g. added via spi_new_device()
after hot-plugging).

Signed-off-by: Anatolij Gustschin <agust@denx.de>
---
Changes in v3:
 - don't use pointers to device data struct in driver_data

Changes in v2:
 - removed not necessary braces {}
 - dropped not needed !id check

 drivers/fpga/altera-ps-spi.c | 36 +++++++++++++++++++++++++++++++-----
 1 file changed, 31 insertions(+), 5 deletions(-)

diff --git a/drivers/fpga/altera-ps-spi.c b/drivers/fpga/altera-ps-spi.c
index 33aafda50af5..78d809194b7d 100644
--- a/drivers/fpga/altera-ps-spi.c
+++ b/drivers/fpga/altera-ps-spi.c
@@ -75,6 +75,11 @@ static struct altera_ps_data a10_data = {
 	.t_st2ck_us = 10, /* min(t_ST2CK) */
 };
 
+static const struct altera_ps_data *altera_ps_data_map[] = {
+	&c5_data,
+	&a10_data,
+};
+
 static const struct of_device_id of_ef_match[] = {
 	{ .compatible = "altr,fpga-passive-serial", .data = &c5_data },
 	{ .compatible = "altr,fpga-arria10-passive-serial", .data = &a10_data },
@@ -234,6 +239,19 @@ static const struct fpga_manager_ops altera_ps_ops = {
 	.write_complete = altera_ps_write_complete,
 };
 
+static const struct altera_ps_data *id_to_data(const struct spi_device_id *id)
+{
+	const struct altera_ps_data *data;
+	unsigned int i;
+
+	for (i = 0; i < ARRAY_SIZE(altera_ps_data_map); i++) {
+		data = altera_ps_data_map[i];
+		if (data && data->devtype == id->driver_data)
+			return data;
+	}
+	return NULL;
+}
+
 static int altera_ps_probe(struct spi_device *spi)
 {
 	struct altera_ps_conf *conf;
@@ -244,11 +262,17 @@ static int altera_ps_probe(struct spi_device *spi)
 	if (!conf)
 		return -ENOMEM;
 
-	of_id = of_match_device(of_ef_match, &spi->dev);
-	if (!of_id)
-		return -ENODEV;
+	if (spi->dev.of_node) {
+		of_id = of_match_device(of_ef_match, &spi->dev);
+		if (!of_id)
+			return -ENODEV;
+		conf->data = of_id->data;
+	} else {
+		conf->data = id_to_data(spi_get_device_id(spi));
+		if (!conf->data)
+			return -ENODEV;
+	}
 
-	conf->data = of_id->data;
 	conf->spi = spi;
 	conf->config = devm_gpiod_get(&spi->dev, "nconfig", GPIOD_OUT_LOW);
 	if (IS_ERR(conf->config)) {
@@ -294,7 +318,9 @@ static int altera_ps_remove(struct spi_device *spi)
 }
 
 static const struct spi_device_id altera_ps_spi_ids[] = {
-	{"cyclone-ps-spi", 0},
+	{ "cyclone-ps-spi", CYCLONE5 },
+	{ "fpga-passive-serial", CYCLONE5 },
+	{ "fpga-arria10-passive-serial", ARRIA10 },
 	{}
 };
 MODULE_DEVICE_TABLE(spi, altera_ps_spi_ids);
-- 
2.17.1

Re: [PATCH v3] fpga: mgr: altera-ps-spi: enable usage on non-dt platforms

[Anatolij Gustschin]

Hi Alan,

On Thu,  1 Nov 2018 22:22:31 +0100
Anatolij Gustschin agust@denx.de wrote:

>Driver probing fails on non-dt platforms since of_match_device()
>always returns NULL here. Add device names and matching driver
>data to the spi_device_id table. This allows driver binding to
>dynamically added PS-SPI devices (e.g. added via spi_new_device()
>after hot-plugging).
>
>Signed-off-by: Anatolij Gustschin <agust@denx.de>
>---
>Changes in v3:
> - don't use pointers to device data struct in driver_data

Ping. Could this be queued for -next?

Thanks,
Anatolij

Re: [PATCH v3] fpga: mgr: altera-ps-spi: enable usage on non-dt platforms

[Alan Tull]

On Thu, Nov 1, 2018 at 4:22 PM Anatolij Gustschin <agust@denx.de> wrote:

Hi Anatolij,

>
> Driver probing fails on non-dt platforms since of_match_device()
> always returns NULL here. Add device names and matching driver
> data to the spi_device_id table. This allows driver binding to
> dynamically added PS-SPI devices (e.g. added via spi_new_device()
> after hot-plugging).

Need to update the header to explain what has been added since v2
(added spi ids and an array and code to map spi ids to data).

>
> Signed-off-by: Anatolij Gustschin <agust@denx.de>
> ---
> Changes in v3:
>  - don't use pointers to device data struct in driver_data
>
> Changes in v2:
>  - removed not necessary braces {}
>  - dropped not needed !id check
>
>  drivers/fpga/altera-ps-spi.c | 36 +++++++++++++++++++++++++++++++-----
>  1 file changed, 31 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/fpga/altera-ps-spi.c b/drivers/fpga/altera-ps-spi.c
> index 33aafda50af5..78d809194b7d 100644
> --- a/drivers/fpga/altera-ps-spi.c
> +++ b/drivers/fpga/altera-ps-spi.c
> @@ -75,6 +75,11 @@ static struct altera_ps_data a10_data = {
>         .t_st2ck_us = 10, /* min(t_ST2CK) */
>  };
>
> +static const struct altera_ps_data *altera_ps_data_map[] = {

Please add a comment that array index is enum altera_ps_devtype or
something like that.  Make it easy for future generations to do the
right thing.  :)

> +       &c5_data,
> +       &a10_data,
> +};
> +
>  static const struct of_device_id of_ef_match[] = {
>         { .compatible = "altr,fpga-passive-serial", .data = &c5_data },
>         { .compatible = "altr,fpga-arria10-passive-serial", .data = &a10_data },
> @@ -234,6 +239,19 @@ static const struct fpga_manager_ops altera_ps_ops = {
>         .write_complete = altera_ps_write_complete,
>  };
>
> +static const struct altera_ps_data *id_to_data(const struct spi_device_id *id)
> +{
> +       const struct altera_ps_data *data;
> +       unsigned int i;
> +
> +       for (i = 0; i < ARRAY_SIZE(altera_ps_data_map); i++) {

At this point, id->driver_data == CYCLONE5 or ARRIA10 (more added in
the future) so we don't need a loop to search the mapping array for a
matching id.     Would something like the following work instead of
this for loop?  I understand you wanted to do some checking, this
accomplishes some checking but will exit NULL if someone extends the
array wrongly.

kernel_ulong_t devtype = id->driver_data;

/* someone added a altera_ps_devtype without adding to the map array */
if (devtype >= ARRAY_SIZE(altera_ps_data_map))
   return NULL;

data = altera_ps_data_map[devtype];
if (!data || data->devtype != devtype)
    return NULL;

return data;

> +               data = altera_ps_data_map[i];
> +               if (data && data->devtype == id->driver_data)
> +                       return data;
> +       }
> +       return NULL;
> +}
> +
>  static int altera_ps_probe(struct spi_device *spi)
>  {
>         struct altera_ps_conf *conf;
> @@ -244,11 +262,17 @@ static int altera_ps_probe(struct spi_device *spi)
>         if (!conf)
>                 return -ENOMEM;
>
> -       of_id = of_match_device(of_ef_match, &spi->dev);
> -       if (!of_id)
> -               return -ENODEV;
> +       if (spi->dev.of_node) {
> +               of_id = of_match_device(of_ef_match, &spi->dev);
> +               if (!of_id)
> +                       return -ENODEV;
> +               conf->data = of_id->data;
> +       } else {
> +               conf->data = id_to_data(spi_get_device_id(spi));
> +               if (!conf->data)
> +                       return -ENODEV;
> +       }
>
> -       conf->data = of_id->data;
>         conf->spi = spi;
>         conf->config = devm_gpiod_get(&spi->dev, "nconfig", GPIOD_OUT_LOW);
>         if (IS_ERR(conf->config)) {
> @@ -294,7 +318,9 @@ static int altera_ps_remove(struct spi_device *spi)
>  }
>
>  static const struct spi_device_id altera_ps_spi_ids[] = {
> -       {"cyclone-ps-spi", 0},
> +       { "cyclone-ps-spi", CYCLONE5 },
> +       { "fpga-passive-serial", CYCLONE5 },
> +       { "fpga-arria10-passive-serial", ARRIA10 },
>         {}
>  };
>  MODULE_DEVICE_TABLE(spi, altera_ps_spi_ids);
> --
> 2.17.1

Thanks,
Alan
>

Re: [PATCH v3] fpga: mgr: altera-ps-spi: enable usage on non-dt platforms

[Anatolij Gustschin]

Hi Alan,

On Thu, 15 Nov 2018 16:24:21 -0600
Alan Tull atull@kernel.org wrote:
...
>Need to update the header to explain what has been added since v2
>(added spi ids and an array and code to map spi ids to data).

Done in v4.

...
>> +static const struct altera_ps_data *altera_ps_data_map[] = {  
>
>Please add a comment that array index is enum altera_ps_devtype or
>something like that.  Make it easy for future generations to do the
>right thing.  :)

Done in v4.

...
>> +static const struct altera_ps_data *id_to_data(const struct spi_device_id *id)
>> +{
>> +       const struct altera_ps_data *data;
>> +       unsigned int i;
>> +
>> +       for (i = 0; i < ARRAY_SIZE(altera_ps_data_map); i++) {  
>
>At this point, id->driver_data == CYCLONE5 or ARRIA10 (more added in
>the future) so we don't need a loop to search the mapping array for a
>matching id.     Would something like the following work instead of
>this for loop?  I understand you wanted to do some checking, this
>accomplishes some checking but will exit NULL if someone extends the
>array wrongly.

Yes, this works.

>kernel_ulong_t devtype = id->driver_data;
>
>/* someone added a altera_ps_devtype without adding to the map array */
>if (devtype >= ARRAY_SIZE(altera_ps_data_map))
>   return NULL;
>
>data = altera_ps_data_map[devtype];
>if (!data || data->devtype != devtype)
>    return NULL;
>
>return data;

Okay, v4 uses this code.

Thanks,
Anatolij