[U-Boot] [PATCH v13 0/7] SPI-NAND support (third batch)

Miquel Raynal miquel.raynal at bootlin.com
Mon Oct 1 13:43:24 UTC 2018

During the last months, Boris Brezillon shared his work to support
serial flashes within Linux. First, he delivered (and merged) a new
layer called spi-mem. He also initiated in Linux MTD subsystem the move
of all 'raw' NAND related code to a raw/ subdirectory, adding at the
same time a NAND core that would be shared with all NAND devices. Then,
he contributed a generic SPI-NAND driver, making use of this NAND core,
as well as some vendor code to drive a few chips.

On top of this work, I made some cleanups in the MTD layer and added an
'mtd' U-Boot command to handle all sort of MTD devices. This should
become the default command instead of having one per flash flavor
('sf', 'nand', 'spi-nand' ?).

The series has been tested on an Ocelot board PCB123 (VSC7514),
featuring a Macronix SPI NAND chip.

TL;DR: the series contains (stripped version since ~30 patches have
already been taken):
- Support for spi-nand devices in mtdparts.
- Generics mtdparts/mtdids parsers.
- A new 'mtd' command.
- A note to set mtdparts command legacy.

To test your SPI-NAND device with U-Boot, you can test someting like:

> setenv mtdparts 'spi-nand0:1m(foo),-(bar)'
> setenv mtdids 'spi-nand0=spi0.0' # spi0.0 is Linux MTD name for this device
> ubi part bar         # create a static UBI volume in the bar partition
> mtd list             # show the current MTD devices/partitions


NB1: If UBI refuses to attach, verify the partition is epty with
     # mtd erase bar

NB2: If your U-Boot crashes and you are using a non SPI-NAND device,
     don't forget to probe your device *first* (sf probe, ...).

Changes since v12:
* Dropped the patch removing a DM-helper without users that was breaking
  the build when included from a file that can be compiled either with
  and without DM support. Instead, added a forward declaration of struct
  udevice, then enclosed the declarations in a #if defined(CONFIG_DM)
  conditional with dummy helpers on the other side.
* Build test with Travis CI:

Changes since v11:
* Due to the number of small changes I had to do to handle the various
  build failures reported by Travis, I decided to send a full v12 so
  that all the patches can be replaced and are known to compile
* Adding a patch to drop an unused helper in mtd_uboot.c that in some
  cases causes build issues (if compiled without DM support enabled).
* Travis CI status (build success except for one unrelated issue):

Changes since v10:
* Only two patches sent independently with small fixes to handle build
  failures for some configurations (reported by Travis)

Changes since v9:
* mtd_search_alternate_name() is moved in mtd_uboot.c (generic code,
  everybody wants to use mtdids).
* mtd_parse_partitions() is still in mtdparts.c because it depends on
  partitions support, but the header file declaring it
  (include/linux/mtd/partitions.h) also has a dummy function if
* Typo corrected in mtd_parse_partitions prototype
* Added Boris' R-b tags.

Changes since v8 (called v7 by mistake):
* Moved most of the generic logic to the core (mtd_uboot.c) so that it
  can be reused by other parts of U-Boot without depending on anything
  else than the MTD core itself.
* Removed the "#ifdef CONFIG_MTD" around mtd_probe_devices() calls now
  that the code is in the core.
* Created an helper for partitions deletion (as there is one to
  parse/create partition objects).
* Drop "cmd: mtdparts: try to probe the MTD devices as a fallback" to
  actually deprecate the command.
* Enhanced a bit the text in Kconfig about deprecating mtdparts.
* Fixed checkpatch.pl warnings in the mtdparts driver.

* The logic in the mtd_is_aligned_with_*() functions was wrong, inversed
  it so the result is what the reader expects (the logic in the end was
  valid, though).
* Discriminate two path: for NAND-like devices (pages, OOB, raw mode,
  bad blocks) and the others.
* Add the "bad" feature to display all bad blocks (if any).
* Fix some typos/display strings.
* Moved the refcounting put_mtd_device() below in the code as suggested
  by Boris.
* Avoid writing empty pages by default (add an option to force the
* Always skip bad blocks.
* Lightened the code by creating helpers.

Changes since v7:
* Added Stefan R-b tags.
* Used tolower() as suggested by Stefan in lib/strto.c.
* Added a mention saying that we continue to abuse the kB suffix in
  lib/strto.c by not making a difference between kB and kiB which both
  mean "* 1024" for us. This has been added in the commit log.
* Fixed various bugs in the mtd command and in the parsers, they should
  be pretty stable now (tested with two MTD devices probed and several
  different partition configurations).
* mtdids is not deprecated at all as opposed at what I wrote in
  v6. U-Boot still uses its own MTD names (being <type><idx> like nor0
  or spi-nand1) and mtdids is still needed to do the glue.
* Removed a useless patch removing a Kconfig entry I added myself since
  during a rebase someone else also added this symbol.

Changes since v6:
* Squashed Boris' fixes about build issues in Travis.
* Merged Stefan Roese's contributions.
* cmd: mtdparts:
  * added fallthrough comments in switch blocks deriving sizes.
  * balanced debug messages to compare freed and allocated partitions.
  * move mtdparts_init() declaration in mtd/partitions.h to be used in
    mtd command as well.
  * introduced a "compatibility" function to return what the MTD device
    name could be thanks to the mtdids environment variable (which is
    useless with the new partitioning logic).
* mtd: mtdpart:
  * reworked the way partitions are created/handled.
* cmd: mtd:
  * implement a partitioning logic that takes the best out of the MTD
    layer without re-inventing the wheel. Partitions are all freed
    correctly when the value of mtdparts changes.
  * balance get_mtd_device() by put_mtd_device() calls.
  * prevent memory leak by reorganizing the code a bit in do_mtd.
  * do not reconstruct partition list when a partition is already in use
    (complain about that instead).
  * add support for the legacy mtdids (not needed anymore if mtdparts
    directly uses the MTD device names as in Linux.
* cmd: ubi:
  * clean and simplify the partition handling a bit to make use the best
    use of the MTD layer as well and keep the code simple.

Changes since v5:
* Included Boris fixup about the build issues.
* Added Rb/Ab tags from Jagan on patchs 20/21.

Changes since v4:
* Added Jagan's Acked-by tags to every patch related to the
  SPI-mem/SPI-NAND addition.
* Rebased on top of master.

Changes since v3:
* Fixed the debug messages in spi-mem to print either Rx or Tx data.
* Fixed a Kconfig error that prevented to build mtdparts with plain
* Fixed a compilation error due to the above error that prevented one
  file to be compiled.
* Adapted the mtd command to probe MTD partitions also.
* Declared mtd_probe_devices() in a header so mtdparts or UBI could
  use it too (to probe all devices and MTD partitions in a clean way).
* As I worked on mtdparts, I found annoying and completely useless the
  fact that we need to prefix the environment variable with
  "mtdparts=". Canceled this obligation.
* Added one patch to allow spi-nand devices to be recognized by mtdparts
  (this is purely useless but needed to be done in order to use this
* Removed useless definitions of MTD device types in UBI code.
* Wrote a generic mtdparts environment variable parser, used by the mtd
* Used the mtd_probe_devices() function from get_mtd_info() in
  cmd/mtdparts.c to be sure the desired partition really does not exist
  (otherwise it will be probed and then found).

Changes since v2:
* Rebased on u-boot master branch.
* Removed extra-parenthesis in
  "mtd: Fallback to ->_read/write() when ->_read/write_oob() is missing"
* s/fiels/files/ in "mtd: move NAND fiels into a raw/ subdirectory"
* Do not describe generic SPI device properties in SPI NAND bindings.
* Changes in the mtd command:
  * Printing more information in 'mtd list' (device type, device
  * Switch to do_div() instead of '(u32)value64b % value32b' which only
    worked because value32b was a power of 2.
  * Removed erase.chip option.
  * By default, erase/read/write happen on the full MTD device while a
    dump will only work on a single page.

Changes since v1:
* Fixed the nand_memorg structure of the MX35LF2GE4AB chip.
* Added Reviewed-by tags from Jagan.
* Backported and squashed two patches fixing things in the SPI NAND core
  received on the Linux ML.
* Backported more changes in mtdcore.c from Linux.
* Added a patch to add a fallback on mtd->_read/_write() in mtdcore.c
  when mtd->_read/write_oob() is not supported.
* Removed the DT changes, useless as the DTs are not available in
  mainline yet.
* Addressed Boris/Stefan comments on the 'mtd' command.
* Added support for multi-pages OOB read/write.

Miquel Raynal (7):
  mtd: uclass: add probe function
  mtd: mtdpart: add a generic mtdparts-like parser
  mtd: uboot: search for an equivalent MTD name with the mtdids
  mtd: mtdpart: implement proper partition handling
  cmd: mtd: add 'mtd' command
  cmd: ubi: clean the partition handling
  cmd: mtdparts: describe as legacy

 cmd/Kconfig                    |  18 +-
 cmd/Makefile                   |   1 +
 cmd/mtd.c                      | 473 +++++++++++++++++++++++++
 cmd/ubi.c                      |  96 ++---
 drivers/mtd/Makefile           |   2 +-
 drivers/mtd/mtd-uclass.c       |  16 +
 drivers/mtd/mtd_uboot.c        | 224 +++++++++++-
 drivers/mtd/mtdcore.c          |   2 +
 drivers/mtd/mtdpart.c          | 623 ++++++++++++++++++++-------------
 include/linux/mtd/mtd.h        |  37 ++
 include/linux/mtd/partitions.h |  22 +-
 include/mtd.h                  |  18 +
 12 files changed, 1217 insertions(+), 315 deletions(-)
 create mode 100644 cmd/mtd.c


More information about the U-Boot mailing list