[PATCH] mtd: nand: Add support for reading ooblayout from device tree

[PATCH] mtd: nand: Add support for reading ooblayout from device tree

[Paul Cercueil]

By specifying the properties "mtd-oob-ecc" and "mtd-oob-free", it is
now possible to specify from devicetree where the ECC data is located
inside the OOB region.

Signed-off-by: Paul Cercueil <paul@crapouillou.net>
---
 Documentation/devicetree/bindings/mtd/nand.txt |  7 +++++
 drivers/mtd/nand/raw/nand_base.c               | 42 ++++++++++++++++++++++++++
 2 files changed, 49 insertions(+)

diff --git a/Documentation/devicetree/bindings/mtd/nand.txt b/Documentation/devicetree/bindings/mtd/nand.txt
index 8bb11d809429..118ea92787cb 100644
--- a/Documentation/devicetree/bindings/mtd/nand.txt
+++ b/Documentation/devicetree/bindings/mtd/nand.txt
@@ -45,6 +45,13 @@ Optional NAND chip properties:
 		     as reliable as possible.
 - nand-rb: shall contain the native Ready/Busy ids.
 
+- nand-oob-ecc: <offset, length> couples of integers, specifying the offset
+		     and length of the ECC data in the OOB region. There can be more
+		     than one couple.
+- nand-oob-free: <offset, length> couples of integers, specifying the offset
+		     and length of a free-to-use area in the OOB region. There can be
+		     more than one couple.
+
 The ECC strength and ECC step size properties define the correction capability
 of a controller. Together, they say a controller can correct "{strength} bit
 errors per {size} bytes".
diff --git a/drivers/mtd/nand/raw/nand_base.c b/drivers/mtd/nand/raw/nand_base.c
index 72f3a89da513..c905531effb0 100644
--- a/drivers/mtd/nand/raw/nand_base.c
+++ b/drivers/mtd/nand/raw/nand_base.c
@@ -213,6 +213,43 @@ static const struct mtd_ooblayout_ops nand_ooblayout_lp_hamming_ops = {
 	.free = nand_ooblayout_free_lp_hamming,
 };
 
+static int nand_oob_of(struct device_node *np, int section,
+		       struct mtd_oob_region *oobregion, const char *prop)
+{
+	int ret = of_property_read_u32_index(np, prop,
+			section * 2, &oobregion->offset);
+	if (ret == -EOVERFLOW)
+		return -ERANGE; /* We're done */
+	if (ret)
+		return ret;
+
+	ret = of_property_read_u32_index(np, prop,
+			section * 2 + 1, &oobregion->length);
+	if (ret == -EOVERFLOW)
+		return -EINVAL; /* We must have an even number of integers */
+
+	return ret;
+}
+
+static int nand_ooblayout_ecc_of(struct mtd_info *mtd, int section,
+				 struct mtd_oob_region *oobregion)
+{
+	return nand_oob_of(mtd->dev.of_node, section,
+			oobregion, "nand-oob-ecc");
+}
+
+static int nand_ooblayout_free_of(struct mtd_info *mtd, int section,
+				 struct mtd_oob_region *oobregion)
+{
+	return nand_oob_of(mtd->dev.of_node, section,
+			oobregion, "nand-oob-free");
+}
+
+static const struct mtd_ooblayout_ops nand_ooblayout_of_ops = {
+	.ecc = nand_ooblayout_ecc_of,
+	.free = nand_ooblayout_free_of,
+};
+
 static int check_offs_len(struct mtd_info *mtd,
 					loff_t ofs, uint64_t len)
 {
@@ -5843,6 +5880,11 @@ static int nand_dt_init(struct nand_chip *chip)
 	if (of_property_read_bool(dn, "nand-ecc-maximize"))
 		chip->ecc.options |= NAND_ECC_MAXIMIZE;
 
+	if (!chip->mtd.ooblayout &&
+				of_property_read_bool(dn, "nand-oob-ecc") &&
+				of_property_read_bool(dn, "nand-oob-free"))
+		chip->mtd.ooblayout = &nand_ooblayout_of_ops;
+
 	return 0;
 }
 
-- 
2.11.0

Re: [PATCH] mtd: nand: Add support for reading ooblayout from device tree

[Boris Brezillon]

Hi Paul,

On Fri, 11 May 2018 23:29:12 +0200
Paul Cercueil <paul@crapouillou.net> wrote:

> By specifying the properties "mtd-oob-ecc" and "mtd-oob-free", it is
> now possible to specify from devicetree where the ECC data is located
> inside the OOB region.

Why would we want to do that? I mean, ECC/free regions are ECC
controller dependent (and NAND chip dependent for the OOB size part),
so there's no reason to describe it in the DT. And more importantly,
people are likely to get it wrong.

I'm curious, why do you need that?

Regards,

Boris

> 
> Signed-off-by: Paul Cercueil <paul@crapouillou.net>
> ---
>  Documentation/devicetree/bindings/mtd/nand.txt |  7 +++++
>  drivers/mtd/nand/raw/nand_base.c               | 42 ++++++++++++++++++++++++++
>  2 files changed, 49 insertions(+)
> 
> diff --git a/Documentation/devicetree/bindings/mtd/nand.txt b/Documentation/devicetree/bindings/mtd/nand.txt
> index 8bb11d809429..118ea92787cb 100644
> --- a/Documentation/devicetree/bindings/mtd/nand.txt
> +++ b/Documentation/devicetree/bindings/mtd/nand.txt
> @@ -45,6 +45,13 @@ Optional NAND chip properties:
>  		     as reliable as possible.
>  - nand-rb: shall contain the native Ready/Busy ids.
>  
> +- nand-oob-ecc: <offset, length> couples of integers, specifying the offset
> +		     and length of the ECC data in the OOB region. There can be more
> +		     than one couple.
> +- nand-oob-free: <offset, length> couples of integers, specifying the offset
> +		     and length of a free-to-use area in the OOB region. There can be
> +		     more than one couple.
> +
>  The ECC strength and ECC step size properties define the correction capability
>  of a controller. Together, they say a controller can correct "{strength} bit
>  errors per {size} bytes".
> diff --git a/drivers/mtd/nand/raw/nand_base.c b/drivers/mtd/nand/raw/nand_base.c
> index 72f3a89da513..c905531effb0 100644
> --- a/drivers/mtd/nand/raw/nand_base.c
> +++ b/drivers/mtd/nand/raw/nand_base.c
> @@ -213,6 +213,43 @@ static const struct mtd_ooblayout_ops nand_ooblayout_lp_hamming_ops = {
>  	.free = nand_ooblayout_free_lp_hamming,
>  };
>  
> +static int nand_oob_of(struct device_node *np, int section,
> +		       struct mtd_oob_region *oobregion, const char *prop)
> +{
> +	int ret = of_property_read_u32_index(np, prop,
> +			section * 2, &oobregion->offset);
> +	if (ret == -EOVERFLOW)
> +		return -ERANGE; /* We're done */
> +	if (ret)
> +		return ret;
> +
> +	ret = of_property_read_u32_index(np, prop,
> +			section * 2 + 1, &oobregion->length);
> +	if (ret == -EOVERFLOW)
> +		return -EINVAL; /* We must have an even number of integers */
> +
> +	return ret;
> +}
> +
> +static int nand_ooblayout_ecc_of(struct mtd_info *mtd, int section,
> +				 struct mtd_oob_region *oobregion)
> +{
> +	return nand_oob_of(mtd->dev.of_node, section,
> +			oobregion, "nand-oob-ecc");
> +}
> +
> +static int nand_ooblayout_free_of(struct mtd_info *mtd, int section,
> +				 struct mtd_oob_region *oobregion)
> +{
> +	return nand_oob_of(mtd->dev.of_node, section,
> +			oobregion, "nand-oob-free");
> +}
> +
> +static const struct mtd_ooblayout_ops nand_ooblayout_of_ops = {
> +	.ecc = nand_ooblayout_ecc_of,
> +	.free = nand_ooblayout_free_of,
> +};
> +
>  static int check_offs_len(struct mtd_info *mtd,
>  					loff_t ofs, uint64_t len)
>  {
> @@ -5843,6 +5880,11 @@ static int nand_dt_init(struct nand_chip *chip)
>  	if (of_property_read_bool(dn, "nand-ecc-maximize"))
>  		chip->ecc.options |= NAND_ECC_MAXIMIZE;
>  
> +	if (!chip->mtd.ooblayout &&
> +				of_property_read_bool(dn, "nand-oob-ecc") &&
> +				of_property_read_bool(dn, "nand-oob-free"))
> +		chip->mtd.ooblayout = &nand_ooblayout_of_ops;
> +
>  	return 0;
>  }
>  

Re: [PATCH] mtd: nand: Add support for reading ooblayout from device tree

[Paul Cercueil]

Le sam. 12 mai 2018 à 20:02, Boris Brezillon 
<boris.brezillon@bootlin.com> a écrit :
> On Sat, 12 May 2018 19:42:49 +0200
> Paul Cercueil <paul@crapouillou.net> wrote:
> 
>>  >>  >>  My motivation is to get rid of this (move it to devicetree):
>>  >>  >>
>>  >>  >>
>>  >> 
>> https://elixir.bootlin.com/linux/latest/source/arch/mips/jz4740/board-qi_lb60.c#L93
>>  >>  >>  And enable the support of other boards with custom OOB 
>> layouts.
>>  >>  >
>>  >>  > Can you list the different layouts you have? I'm pretty sure
>>  >> there's a
>>  >>  > pattern. Maybe we can even deduce the layout from the page 
>> size
>>  >> or OOB
>>  >>  > size.
>>  >>
>>  >>  This is the other layout I have for another ingenic device:
>>  >>
>>  >> 
>> http://projects.qi-hardware.com/index.php/p/qi-kernel/source/tree/od-2011-09-18/arch/mips/jz4740/board-a320.c#L125
>>  >>
>>  >>  Page size and OOB size are the same between these two devices.
>>  >
>>  > Indeed. Do you know if there are other kind of layouts in the 
>> wild?
>> 
>>  I'm getting a new board in a few weeks, I'll be able to check that 
>> out.
>> 
>>  > Note that <layout-id> can be a string, so if each each board is
>>  > defining its own layout, you could specify the board name here.
>>  > Otherwise, if you just have those 2 patterns, you can just name 
>> them
>>  > "contiguous" and "interleaved".
>> 
>>  I don't like the idea of adding board-specific data inside the 
>> driver...
>>  I'd prefer to use the method I used in this patch, but inside the
>>  jz4740-nand driver, if you're OK with it.
> 
> Please don't. Encoding such detailed description in the DT has almost
> always proven to be poor choice. Also, people are likely to get it
> wrong, and then you'll have to fix all DTs, while, with a single 
> unique
> ID representing the layout, you can
> - re-use existing layouts easily without having to describe everything
>   again in the DT
> - fix the driver without getting in trouble with people who claim
>   that DT is a stable ABI and don't want to update their DT
> 
> So, please just pick user-friendly IDs and add layout definitions in
> the driver. If you don't want to leak board info in the driver, then
> don't name the layout with the board name. BTW, I still hope you'll 
> only
> have 2 kind of layouts (contiguous and interleaved).

Alright. Thanks for the insights.

-Paul

Re: [PATCH] mtd: nand: Add support for reading ooblayout from device tree

[Boris Brezillon]

On Sat, 12 May 2018 19:42:49 +0200
Paul Cercueil <paul@crapouillou.net> wrote:

> >>  >>  My motivation is to get rid of this (move it to devicetree):
> >>  >>
> >>  >>   
> >> https://elixir.bootlin.com/linux/latest/source/arch/mips/jz4740/board-qi_lb60.c#L93  
> >>  >>  And enable the support of other boards with custom OOB layouts.  
> >>  >
> >>  > Can you list the different layouts you have? I'm pretty sure   
> >> there's a  
> >>  > pattern. Maybe we can even deduce the layout from the page size   
> >> or OOB  
> >>  > size.  
> >> 
> >>  This is the other layout I have for another ingenic device:
> >>  
> >> http://projects.qi-hardware.com/index.php/p/qi-kernel/source/tree/od-2011-09-18/arch/mips/jz4740/board-a320.c#L125
> >> 
> >>  Page size and OOB size are the same between these two devices.  
> > 
> > Indeed. Do you know if there are other kind of layouts in the wild?  
> 
> I'm getting a new board in a few weeks, I'll be able to check that out.
> 
> > Note that <layout-id> can be a string, so if each each board is
> > defining its own layout, you could specify the board name here.
> > Otherwise, if you just have those 2 patterns, you can just name them
> > "contiguous" and "interleaved".  
> 
> I don't like the idea of adding board-specific data inside the driver...
> I'd prefer to use the method I used in this patch, but inside the
> jz4740-nand driver, if you're OK with it.

Please don't. Encoding such detailed description in the DT has almost
always proven to be poor choice. Also, people are likely to get it
wrong, and then you'll have to fix all DTs, while, with a single unique
ID representing the layout, you can
- re-use existing layouts easily without having to describe everything
  again in the DT
- fix the driver without getting in trouble with people who claim
  that DT is a stable ABI and don't want to update their DT

So, please just pick user-friendly IDs and add layout definitions in
the driver. If you don't want to leak board info in the driver, then
don't name the layout with the board name. BTW, I still hope you'll only
have 2 kind of layouts (contiguous and interleaved).

Re: [PATCH] mtd: nand: Add support for reading ooblayout from device tree

[Paul Cercueil]

>>  >>  My motivation is to get rid of this (move it to devicetree):
>>  >>
>>  >> 
>> https://elixir.bootlin.com/linux/latest/source/arch/mips/jz4740/board-qi_lb60.c#L93
>>  >>  And enable the support of other boards with custom OOB layouts.
>>  >
>>  > Can you list the different layouts you have? I'm pretty sure 
>> there's a
>>  > pattern. Maybe we can even deduce the layout from the page size 
>> or OOB
>>  > size.
>> 
>>  This is the other layout I have for another ingenic device:
>>  
>> http://projects.qi-hardware.com/index.php/p/qi-kernel/source/tree/od-2011-09-18/arch/mips/jz4740/board-a320.c#L125
>> 
>>  Page size and OOB size are the same between these two devices.
> 
> Indeed. Do you know if there are other kind of layouts in the wild?

I'm getting a new board in a few weeks, I'll be able to check that out.

> Note that <layout-id> can be a string, so if each each board is
> defining its own layout, you could specify the board name here.
> Otherwise, if you just have those 2 patterns, you can just name them
> "contiguous" and "interleaved".

I don't like the idea of adding board-specific data inside the driver...
I'd prefer to use the method I used in this patch, but inside the
jz4740-nand driver, if you're OK with it.

Thanks,
-Paul

Re: [PATCH] mtd: nand: Add support for reading ooblayout from device tree

[Boris Brezillon]

On Sat, 12 May 2018 11:38:26 -0300
Paul Cercueil <paul@crapouillou.net> wrote:

> Le sam. 12 mai 2018 à 10:42, Boris Brezillon 
> <boris.brezillon@bootlin.com> a écrit :
> > On Sat, 12 May 2018 08:55:40 -0300
> > Paul Cercueil <paul@crapouillou.net> wrote:
> >   
> >>  Hi Boris,
> >> 
> >>  Le 12 mai 2018 02:55, Boris Brezillon <boris.brezillon@bootlin.com> 
> >> a écrit :  
> >>  >
> >>  > Hi Paul,
> >>  >
> >>  > On Fri, 11 May 2018 23:29:12 +0200
> >>  > Paul Cercueil <paul@crapouillou.net> wrote:
> >>  >  
> >>  > > By specifying the properties "mtd-oob-ecc" and "mtd-oob-free",   
> >> it is  
> >>  > > now possible to specify from devicetree where the ECC data is   
> >> located  
> >>  > > inside the OOB region.  
> >>  >
> >>  > Why would we want to do that? I mean, ECC/free regions are ECC
> >>  > controller dependent (and NAND chip dependent for the OOB size   
> >> part),  
> >>  > so there's no reason to describe it in the DT. And more   
> >> importantly,  
> >>  > people are likely to get it wrong.
> >>  >
> >>  > I'm curious, why do you need that?  
> >> 
> >>  Good question.
> >> 
> >>  The reason is that some SoCs have no ECC controller.
> >>  The various boards for these SoCs then all use a different layout.  
> > 
> > Okay. Still think defining the layouts in the DT is a bad idea. We
> > can add a jz4740 specific property to define the layout id
> > (ingenic,nand-oob-layout = <layout-id>), but not a generic way to
> > define custom layouts for all kind of NAND controller.  
> 
> Okay.
> 
> >> 
> >>  My motivation is to get rid of this (move it to devicetree):
> >>  
> >> https://elixir.bootlin.com/linux/latest/source/arch/mips/jz4740/board-qi_lb60.c#L93
> >>  And enable the support of other boards with custom OOB layouts.  
> > 
> > Can you list the different layouts you have? I'm pretty sure there's a
> > pattern. Maybe we can even deduce the layout from the page size or OOB
> > size.  
> 
> This is the other layout I have for another ingenic device:
> http://projects.qi-hardware.com/index.php/p/qi-kernel/source/tree/od-2011-09-18/arch/mips/jz4740/board-a320.c#L125
> 
> Page size and OOB size are the same between these two devices.

Indeed. Do you know if there are other kind of layouts in the wild?
Note that <layout-id> can be a string, so if each each board is
defining its own layout, you could specify the board name here.
Otherwise, if you just have those 2 patterns, you can just name them
"contiguous" and "interleaved".

Re: [PATCH] mtd: nand: Add support for reading ooblayout from device tree

[Paul Cercueil]

Le sam. 12 mai 2018 à 10:42, Boris Brezillon 
<boris.brezillon@bootlin.com> a écrit :
> On Sat, 12 May 2018 08:55:40 -0300
> Paul Cercueil <paul@crapouillou.net> wrote:
> 
>>  Hi Boris,
>> 
>>  Le 12 mai 2018 02:55, Boris Brezillon <boris.brezillon@bootlin.com> 
>> a écrit :
>>  >
>>  > Hi Paul,
>>  >
>>  > On Fri, 11 May 2018 23:29:12 +0200
>>  > Paul Cercueil <paul@crapouillou.net> wrote:
>>  >
>>  > > By specifying the properties "mtd-oob-ecc" and "mtd-oob-free", 
>> it is
>>  > > now possible to specify from devicetree where the ECC data is 
>> located
>>  > > inside the OOB region.
>>  >
>>  > Why would we want to do that? I mean, ECC/free regions are ECC
>>  > controller dependent (and NAND chip dependent for the OOB size 
>> part),
>>  > so there's no reason to describe it in the DT. And more 
>> importantly,
>>  > people are likely to get it wrong.
>>  >
>>  > I'm curious, why do you need that?
>> 
>>  Good question.
>> 
>>  The reason is that some SoCs have no ECC controller.
>>  The various boards for these SoCs then all use a different layout.
> 
> Okay. Still think defining the layouts in the DT is a bad idea. We
> can add a jz4740 specific property to define the layout id
> (ingenic,nand-oob-layout = <layout-id>), but not a generic way to
> define custom layouts for all kind of NAND controller.

Okay.

>> 
>>  My motivation is to get rid of this (move it to devicetree):
>>  
>> https://elixir.bootlin.com/linux/latest/source/arch/mips/jz4740/board-qi_lb60.c#L93
>>  And enable the support of other boards with custom OOB layouts.
> 
> Can you list the different layouts you have? I'm pretty sure there's a
> pattern. Maybe we can even deduce the layout from the page size or OOB
> size.

This is the other layout I have for another ingenic device:
http://projects.qi-hardware.com/index.php/p/qi-kernel/source/tree/od-2011-09-18/arch/mips/jz4740/board-a320.c#L125

Page size and OOB size are the same between these two devices.

-Paul

Re: [PATCH] mtd: nand: Add support for reading ooblayout from device tree

[Boris Brezillon]

On Sat, 12 May 2018 08:55:40 -0300
Paul Cercueil <paul@crapouillou.net> wrote:

> Hi Boris,
> 
> Le 12 mai 2018 02:55, Boris Brezillon <boris.brezillon@bootlin.com> a écrit :
> >
> > Hi Paul, 
> >
> > On Fri, 11 May 2018 23:29:12 +0200 
> > Paul Cercueil <paul@crapouillou.net> wrote: 
> >  
> > > By specifying the properties "mtd-oob-ecc" and "mtd-oob-free", it is 
> > > now possible to specify from devicetree where the ECC data is located 
> > > inside the OOB region.   
> >
> > Why would we want to do that? I mean, ECC/free regions are ECC 
> > controller dependent (and NAND chip dependent for the OOB size part), 
> > so there's no reason to describe it in the DT. And more importantly, 
> > people are likely to get it wrong. 
> >
> > I'm curious, why do you need that?   
> 
> Good question.
> 
> The reason is that some SoCs have no ECC controller.
> The various boards for these SoCs then all use a different layout.

Okay. Still think defining the layouts in the DT is a bad idea. We
can add a jz4740 specific property to define the layout id
(ingenic,nand-oob-layout = <layout-id>), but not a generic way to
define custom layouts for all kind of NAND controller.

> 
> My motivation is to get rid of this (move it to devicetree):
> https://elixir.bootlin.com/linux/latest/source/arch/mips/jz4740/board-qi_lb60.c#L93
> And enable the support of other boards with custom OOB layouts.

Can you list the different layouts you have? I'm pretty sure there's a
pattern. Maybe we can even deduce the layout from the page size or OOB
size.