[v5,0/6] block: partition table OF support

Hi,
this is an initial proposal to complete support for manually defining
partition table.

Some background on this. Many OEM on embedded device (modem, router...)
are starting to migrate from NOR/NAND flash to eMMC. The reason for this
is that OEM are starting to require more and more space for the firmware
and price difference is becoming so little that using eMMC is only benefits
and no cons.

Given these reason, OEM are also using very custom way to provide a
partition table and doesn't relay on common method like writing a table
on the eMMC.

One way that is commonly used is to hardcode the partition table and
pass it to the system via various way (cmdline, special glue driver,
block2mtd...)
This way is also used on Android where the partition table
is passed from the bootloader via cmdline.

One reason to use this method is to save space on the device and to
permit more flexibility on partition handling.

What this series does is complete support for this feature.
It's possible to use the cmdline to define a partition table similar
to how it's done for MTD but this is problematic for a number of device
where tweaking the cmdline is not possible. This series adds OF support
to make it possible to define a partition table in the Device Tree.

We implement a similar schema to the MTD fixed-partition, where we define
a "label" and a "reg" with "offset" and "size".

A new block partition parser is introduced that check if the disk device
have an OF node attached and check if a fixed-partition table is defined.

block driver can use the device_add_of_disk() function to register a new
disk and attach a fwnode to it for usage with the OF parser.

This permits flexibility from the driver side to implement the partitions
node in different nodes across different block devices.

If a correct node is found, then partition table is filled. cmdline will
still have priority to this new parser.

Some block device also implement boot1 and boot2 additional disk. Similar
to the cmdline parser, these disk can have OF support using the
"partitions-boot1" and "partitions-boot2" additional node. Also eMMC
gp 1/2/3/4 disk are supported.

It's also completed support for declaring partition as read-only as this
feature was introduced but never finished in the cmdline parser.

I hope this solution is better accepted as downstream this is becoming
a real problem with a growing number of strange solution for the simple
task of providing a fixed partition table.

Changes v5:
- Introduce device_add_of_disk() function
- Detach eMMC special disk from OF block partition code and move
  parsing to eMMC block driver (as requested by Christoph)
- Rework OF block partition to use the device disk device_node
- Extend support for eMMC GP1/2/3/4
- Rename boot0/1 to boot1/2
- Drop strends patch (unused now)
Changes v4:
- Fix wrong description and title in Kconfig
- Validate reg len with addr and size cells
- Drop offset 0 constraint (not needed)
- Rework bytes to sector conversion
- Follow common logic with ignore partitions after state->limit
- Better handle device_node put
- Add suggested strends string helper
Changes v3:
- Out of RFC
- Drop partition schema generalization and simplify it
- Require fixed-partitions compatible to adapt to MTD schema
- Make label property optional and fallback to node name
Changes v2:
- Reference bytes in DT instead of Sector Size
- Validate offset and size after Sector Size conversion
- Limit boot0 and boot1 to eMMC and add comments about JEDEC spec
- Generalize MTD partition schema and introduce block partitions schema
- Add missing code to actually attach the OF parser to block partition core
- Add reviewed by tag for read-only patch

Christian Marangi (6):
  block: add support for defining read-only partitions
  docs: block: Document support for read-only partition in cmdline part
  block: introduce device_add_of_disk()
  mmc: block: attach partitions fwnode if found in mmc-card
  block: add support for partition table defined in OF
  dt-bindings: mmc: Document support for partition table in mmc-card

 Documentation/block/cmdline-partition.rst     |   5 +-
 .../devicetree/bindings/mmc/mmc-card.yaml     |  52 ++++++++
 block/blk.h                                   |   1 +
 block/genhd.c                                 |  21 +++-
 block/partitions/Kconfig                      |   9 ++
 block/partitions/Makefile                     |   1 +
 block/partitions/check.h                      |   1 +
 block/partitions/cmdline.c                    |   3 +
 block/partitions/core.c                       |   6 +
 block/partitions/of.c                         | 116 ++++++++++++++++++
 drivers/mmc/core/block.c                      |  55 ++++++++-
 include/linux/blkdev.h                        |   3 +
 12 files changed, 269 insertions(+), 4 deletions(-)
 create mode 100644 block/partitions/of.c

Thanks,

this looks much better.  A few minor nitpicks, though:

> -int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
> -				 const struct attribute_group **groups)
> +static int __device_add_disk(struct device *parent, struct gendisk *disk,
> +			     const struct attribute_group **groups,
> +			     struct fwnode_handle *fwnode)

I don't think we need a separate helper if device_add_disk simply
wraps the OF version by passing a NULL fwnode.

> +int __must_check device_add_of_disk(struct device *parent, struct gendisk *disk,
> +				    const struct attribute_group **groups,
> +				    struct fwnode_handle *fwnode)
> +{
> +	return __device_add_disk(parent, disk, groups, fwnode);
> +}

I'd name this as add_disk_fwnode as the of in device_add_of_disk
reads as in add the device of the disk, and the fwnode is what gets
passed.  The device_ is a bit redundant and just there for historic
reasons as the original add_disk predates the device model.

Can you also add a kerneldoc comment for the new helper?

> +EXPORT_SYMBOL(device_add_of_disk);

EXPORT_SYMBO_GPL, please.

On Wed, Oct 02, 2024 at 01:40:58AM -0700, Christoph Hellwig wrote:
> Thanks,
> 
> this looks much better.  A few minor nitpicks, though:
>

Very happy you like it, yes I wasn't sure what was the correct way to
introduce the helper. If you notice in the blkdev.h we have also add_disk()
that is a static inline wrapper for device_add_disk().

Wonder if device_add_disk() should have the same treatement? No idea if
it would cause problem with symbol with external modules, that is why I
used the wrapper.

> > -int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
> > -				 const struct attribute_group **groups)
> > +static int __device_add_disk(struct device *parent, struct gendisk *disk,
> > +			     const struct attribute_group **groups,
> > +			     struct fwnode_handle *fwnode)
> 
> I don't think we need a separate helper if device_add_disk simply
> wraps the OF version by passing a NULL fwnode.
> 
> > +int __must_check device_add_of_disk(struct device *parent, struct gendisk *disk,
> > +				    const struct attribute_group **groups,
> > +				    struct fwnode_handle *fwnode)
> > +{
> > +	return __device_add_disk(parent, disk, groups, fwnode);
> > +}
> 
> I'd name this as add_disk_fwnode as the of in device_add_of_disk
> reads as in add the device of the disk, and the fwnode is what gets
> passed.  The device_ is a bit redundant and just there for historic
> reasons as the original add_disk predates the device model.
> 
> Can you also add a kerneldoc comment for the new helper?
> 

sure! I will wait the usual 24h to respin this.

> > +EXPORT_SYMBOL(device_add_of_disk);
> 
> EXPORT_SYMBO_GPL, please.
> 

ack.

On Wed, Oct 02, 2024 at 10:46:46AM +0200, Christian Marangi wrote:
> Very happy you like it, yes I wasn't sure what was the correct way to
> introduce the helper. If you notice in the blkdev.h we have also add_disk()
> that is a static inline wrapper for device_add_disk().
> 
> Wonder if device_add_disk() should have the same treatement? No idea if
> it would cause problem with symbol with external modules, that is why I
> used the wrapper.

We could make it an inline wrapper, but it's not in a high performance
path so there isn't really much of a point in doing so.  I don't
remember why it was done for add_disk.