The Atmel SAM E70 flash has a 16-byte write block size. Increase the
NVS_BLOCK_SIZE a bit and take some margin. This might also improve the
performances by reducing the calls to the flash driver when moving data
during garbage collection.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
When the write block size is bigger than sizeof(nvs_ate), which is 8
bytes, we should not read or write more than the ATE. The
_nvs_flash_al_wrt() function will take care of padding the write with
0xff up to write_block_size. Of course the addresses should still be
incremented by write_block_size.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
VLA are usually not recommended and are a MISRA C violation. Replace
fs->write_block_size by NVS_BLOCK_SIZE as we now have a check at
initialization that ensures that fs->write_block_size <= NVS_BLOCK_SIZE.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
In case the write block size is bigger than NVS_BLOCK_SIZE, some
functions end up in an endless loop. Detect the unsupported cases
at initialization.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
We had the if in the wrong place causing some Kconfigs to be set even if
filesystem was not configured in.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Under GNU C, sizeof(void) = 1. This commit merely makes it explicit u8.
Pointer arithmetics over void types is:
* A GNU C extension
* Not supported by Clang
* Illegal across all ISO C standards
See also: https://gcc.gnu.org/onlinedocs/gcc/Pointer-Arith.html
Signed-off-by: Mark Ruvald Pedersen <mped@oticon.com>
New shell implementation is on the way. For now old one and all
references are kept to be gradually replaced by new shell.
Signed-off-by: Krzysztof Chruscinski <krzysztof.chruscinski@nordicsemi.no>
In case a write to the flash failed, do not leave the flash unprotected.
Always call flash_write_protection_set in that case.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Avoid unsetting and setting the flash protection if there is nothing to
write to the flash. This happens for example when deleting data from the
flash using nvs_delete.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
In case a sector is not empty nor properly closed (ie it never contains
8 times 0xff nor 0x00), the _nvs_prev_ate will loop indefinitely and
will start adressing memory outside of the flash area.
Fix that by stopping the loop when the address matches the beginning of
the sector.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Instead of forcing the crc8 entry to 0xff for the crc8 computation, just
ignore this field in the computation as it is the last one. This avoid
having to set it back to the original value for _nvs_ate_crc8_check.
Add a build assertion to ensure crc8 is kept last.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
The allocation table entry should be as small as possible in the flash,
so declare it as packed to avoid that the compiler pads it.
Note that this doesn't change anything on ARM, but it might help for
other (future) architectures.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
- Add a missing plural.
- Use a comma to separate the sector number with the offset to not
confuse that with a range.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
The return of memset is never checked. This patch explicitly ignore
the return to avoid MISRA-C violations.
The only directory excluded directory was ext/* since it contains
only imported code.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
The current code computes the block-aligned len by ANDing the len with
~write_block_size instead of ~(write_block_size - 1).
In addition the compute value can be 0 (for lengths that are less than
the block size), so the first flash write might have to be skipped.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
On flash NVS was stored one entry after another including the metadata
of each entry. This has the disadvantage that when an incomplete write
is performed (e.g. due to power failure) the complete sector had to be
rewritten to get a completely functional system.
The present rewrite changed the storage in flash of the data. For each
sector the data is now written as follows: the data itself at the
beginning of the sector (one after the other), the metadata (id, length,
data offset in the sector, and a crc of the metadata) is written from
the end of the sector. The metadata is of fixed size (8 byte) and for
a sector that is completely occupied a metadata entry of all zeros is
used.
Writing data to flash always is done by:
1. Writing the data,
2. Writing the metadata.
If an incomplete write is done NVS will ignore this incomplete write.
At the same time the following improvements were done:
1. NVS now support 65536 sectors of each 65536 byte.
2. The sector size no longer requires to be a power of 2 (but it
still needs to be a multiple of the flash erase page size).
3. NVS now also keeps track of the free space available.
Signed-off-by: Laczen JMS <laczenjms@gmail.com>
The nvs module has some disadvantages for larger block size. The data
header and slot are taking up to much space. A rewrite is proposed that
reduces the used storage space for systems with write block size > 4.
The data storage in flash is now one unit consisting of: data_length,
data_id, data and data_length again in a multiple of the write block
size. The data_length at the end is used to validate the correctness of
the flash write and also allows to travel backwards in the filesystem.
As a comparison, on a system with block size 8 byte, a 32 bit values
now fits 1 block including the metadata (length and id). This used to
be 3 blocks.
The data_length will occupy 1 byte if the data length is less than 128
byte, it will occupy 2 byte if the data length is 128 byte or more. The
data length is limited to 16383 byte.
Each write to flash is verified by a read back of the data.
The read performance is improved because reading is done backwards so
the latest items are found first.
When the filesystem is locked it can be unlocked by calling
reinit(), this will clear flash and setup everything for storage.
add sample documentation - README.rst
Update dtsi to include erase_block_size, use erase_block_size in sample
Update prj.conf to include CONFIG_MPU_ALLOW_FLASH_WRITE
Signed-off-by: Laczen JMS <laczenjms@gmail.com>
This patch removes the need for application build script code to
explicitly link 'app' with a filesystem implementation.
It does this by introducing a zephyr interface library called 'FS'
that contains the usage requirements for linking with the filesystem
library subsys__fs and using Kconfig to default to linking the 'app'
library with this interface library.
Signed-off-by: Sebastian Bøe <sebastian.boe@nordicsemi.no>
Bool symbols implicitly default to 'n'.
A 'default n' can make sense e.g. in a Kconfig.defconfig file, if you
want to override a 'default y' on the base definition of the symbol. It
isn't used like that on any of these symbols though.
Remove some 'default ""' properties on string symbols too.
Also make definitions more consistent by converting some
config FOO
<type>
prompt "foo"
definitions to a shorter form:
config FOO
<type> "foo"
This shorthand works for int/hex/string symbols too, not just for bool
symbols.
Signed-off-by: Ulf Magnusson <Ulf.Magnusson@nordicsemi.no>
_nvs_sector_is_used() never uses the offset argument. As a consequence,
it only check the first sector of the flash. Fix that.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
The two functions that compute the crc16 when writing (nvs_append_close)
and when reading (nvs_check_crc) currently assume that the flash is
also mapped in read mode at address 0. This is not true on all SoCs, and
even less on an SPI flash.
Fix this by adding a new nvs_compute_crc() function which compute the
CRC16 of an entry using the flash using nvs_flash_read, in blocks of
write_block_size. This might not be the optimal size, but it keeps the
stack usage small.
Use this function in both nvs_append_close() and nvs_check_crc() instead
of accessing the flash from address 0.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
GCC complains that last_entry.len and last_entry.data_addr might be
uninitialized in _nvs_gc. Fix that.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Now that the flash writes are padded up to the write block size, there
is no need to have explicit padding fields in the _nvs_sector_hdr and
_nvs_data_slt structure. This allow to save space when the write block
size equals to 1 or 2
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Writing more than the source buffer means that some random data,
possibly coming from the stack, ends-up in the flash. This could be
a security issue.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Reading more than the destination buffer means that data is overwritten
possibly on the stack. This causes unpredictable behaviours like
crashes.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Some SoCs do not allow shorter writes than the write block size, usually
when they have ECC memory. This patch first write all data in multiple
of the write block size and then do a last write with the data padded.
It uses 0xff as the padding byte to avoid wearing-out the flash.
nvs_append_close() is slightly modified to compute the crc16 that will
be put in the slot over the size defined in the header, to match the way
it is checked on read.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Zephyr supports fatfs, nffs and fcb as storage layer. fatfs and nffs
are less suited for application in memory restricted IC's. fcb has a
smaller footprint but has a complex api.
The proposed module is a module with a even smaller footprint compared
to fcb and a simple interface for reading and writing entries. The
module provides wear levelling of flash. This allows the module to be
used not only to store configuration settings but to store device state
(e.g. state of a light switch over reboots) of a zephyr device.
Fixes buffer overflow by introducing maximum read length in nvs_read()
and nvs_read_hist().
Fixes nvs_write() not to reflash the same data. Allows the user to do
call nvs_write() for all defined entries without worries about flash
wear.
Fixes garbage collection error where wrong data could be copied.
Add nvs_delete() to allow deleting a stored entry. A deleted entry will
not be copied to a new flash sector
Include flash wear information in the README.md documentation
0/25 Update module after reviewers remarks, added documentation to
nvs.h, removed README.md by nvs.rst in doc/subsystems folder
04/26 Update module after reviewers remarks, updated nvs.rst, added more
documentation to samples/subsys/nvs/src/main.c, updated doxygen info
in nvs.h (hope this time it works).
04/26 Update subsystems.rst to include nvs.restart
04/27 Updated nvs.c and nvs.h to avoid a possible flash deletion loop
when the file system is full.
04/29 Updated nvs_write to detect and ignore deletes of non-existing
items
05/06 Update NVS module to return standard error codes, removed low
level API, added configuration options. NVS now uses the board dts to
determine the flash storage location (FLASH_AREA_STORAGE_OFFSET).
05/06 Update nvs.rst. Updated intendation and added intermediate
variables in nvs.c to make the code easier to read.
05/06 Update nvs.rst.
05/07 Update nvs.rst
05/08 Changed the API to a more standard file system API.
05/08 Removed cnt_max from nvs_read() as it is not used.
05/08 Removed #ifdef(CONFIG_NVS_LOG) from nvs_priv.h, now the module can
be build with debugging off.
05/09 Removed configuration options for SECTOR_SIZE, SECTOR_COUNT and
MAX_ELEM_SIZE. It is now easy to support multiple NVS filesystems on
one or multiple devices. Changed logging to support newlib systems.
Thanks to Olivier Martin for reporting and proposed changes.
Signed-off-by: Laczen JMS <laczenjms@gmail.com>
Add support for multiple instances of a file system by
making use of mount point as the disk volume name which
is used by the file system library while formatting or
mounting a disk.
Also moved out file system specific data structures from
public fs.h header and handled them in corresponding
file system interface files by introducing open files and
open directories concept which is already being used in
NFFS interface module. Now it is extended to FatFs as well.
Signed-off-by: Ramakrishna Pallala <ramakrishna.pallala@intel.com>
Extend storage_dev type beyond 'struct device' by changing
the variable type to void pointer. This will be needed if we
have additional transactional layers b/w filesystem and the
actual backend device.
Signed-off-by: Ramakrishna Pallala <ramakrishna.pallala@intel.com>
This is bug-fix for issue #7311
FCB: CRC write size in append_finish doesn't honor
flash min write size
This patch changes write size to the minimum supported
write size.
Signed-off-by: Andrzej Puzdrowski <andrzej.puzdrowski@nordicsemi.no>
FS_FLASH_MAP_STORAGE keyword enables the storage partition,
but it was depend on flash_map module which is unused by
NFFS. This patch makes it independent thanks
to it is possible to enable the storage partition
without flash_map module.
FS_FLASH_MAP_STORAGE was renamed to
CONFIG_FS_FLASH_STORAGE_PARTITION
as it is independent for flash_map.
Signed-off-by: Andrzej Puzdrowski <andrzej.puzdrowski@nordicsemi.no>
The NFFS partition at the end of flash is also useful for any other
file system or even the Flash Circular Buffer (FCB). Rename the
partition from 'nffs_partition' to 'storage_partition' and make it
depend on a new hidden Kconfig entry which the relevant users will
select (such as NFFS and FCB).
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The implementation for the tell() primitive was attempting to unlock
the NFFS mutex after returning an error code.
Coverity-Id: 185283
Signed-off-by: Leandro Pereira <leandro.pereira@intel.com>
With the introduce of VFS the typedef for fs_file_t & fs_dir_t don't
exist anymore so we need to use 'struct fs_dir_t' or 'struct fs_file_t'.
Fix up some places that got missed in the VFS conversion.
Signed-off-by: Kumar Gala <kumar.gala@linaro.org>
Add support for Virtual File system Switch (VFS) by
introducing mount point concept to Zephyr. This allows
the applications to mount multiple file systems at
different mount points (ex: "/fatfs" and "/nffs"). The
mount point structure contains all the necessary info
required to instantiate, mount and operate on file system.
Decouple applications from directly accessing individual
file systems API's or internal functions by introducing
file system registration mechanism in VFS.
Move the file system defination and mount responsibility
to application so that application can decide which file system
to use and where to mount.
Signed-off-by: Ramakrishna Pallala <ramakrishna.pallala@intel.com>
It is faster to operate directly on flash_area pointer instead
of fetch it all the time using the fcb flash area id.
Also as f_area_id was needed only for get appropriate flash_area
pointer, so it is better to pass it only while initialization
the fcb and not store it in fcb instance data at all.
Signed-off-by: Andrzej Puzdrowski <andrzej.puzdrowski@nordicsemi.no>
Aurelien Jarno