UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
/*
|
2010-08-08 21:58:20 +02:00
|
|
|
|
* Copyright © International Business Machines Corp., 2006
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
*
|
|
|
|
|
* This program is free software; you can redistribute it and/or modify
|
|
|
|
|
* it under the terms of the GNU General Public License as published by
|
|
|
|
|
* the Free Software Foundation; either version 2 of the License, or
|
|
|
|
|
* (at your option) any later version.
|
|
|
|
|
*
|
|
|
|
|
* This program is distributed in the hope that it will be useful,
|
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
|
|
|
|
|
* the GNU General Public License for more details.
|
|
|
|
|
*
|
|
|
|
|
* You should have received a copy of the GNU General Public License
|
|
|
|
|
* along with this program; if not, write to the Free Software
|
|
|
|
|
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
|
|
|
*
|
|
|
|
|
* Author: Artem Bityutskiy (Битюцкий Артём)
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#ifndef __UBI_USER_H__
|
|
|
|
|
#define __UBI_USER_H__
|
|
|
|
|
|
2009-02-26 00:51:41 +01:00
|
|
|
|
#include <linux/types.h>
|
|
|
|
|
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
/*
|
2007-12-18 17:22:16 +01:00
|
|
|
|
* UBI device creation (the same as MTD device attachment)
|
|
|
|
|
* ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
|
|
|
|
*
|
|
|
|
|
* MTD devices may be attached using %UBI_IOCATT ioctl command of the UBI
|
|
|
|
|
* control device. The caller has to properly fill and pass
|
|
|
|
|
* &struct ubi_attach_req object - UBI will attach the MTD device specified in
|
|
|
|
|
* the request and return the newly created UBI device number as the ioctl
|
|
|
|
|
* return value.
|
|
|
|
|
*
|
|
|
|
|
* UBI device deletion (the same as MTD device detachment)
|
|
|
|
|
* ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
|
|
|
|
*
|
|
|
|
|
* An UBI device maybe deleted with %UBI_IOCDET ioctl command of the UBI
|
|
|
|
|
* control device.
|
|
|
|
|
*
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
* UBI volume creation
|
|
|
|
|
* ~~~~~~~~~~~~~~~~~~~
|
|
|
|
|
*
|
2009-01-07 15:10:58 +01:00
|
|
|
|
* UBI volumes are created via the %UBI_IOCMKVOL ioctl command of UBI character
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
* device. A &struct ubi_mkvol_req object has to be properly filled and a
|
2009-01-07 15:10:58 +01:00
|
|
|
|
* pointer to it has to be passed to the ioctl.
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
*
|
|
|
|
|
* UBI volume deletion
|
|
|
|
|
* ~~~~~~~~~~~~~~~~~~~
|
|
|
|
|
*
|
2009-01-07 15:10:58 +01:00
|
|
|
|
* To delete a volume, the %UBI_IOCRMVOL ioctl command of the UBI character
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
* device should be used. A pointer to the 32-bit volume ID hast to be passed
|
2009-01-07 15:10:58 +01:00
|
|
|
|
* to the ioctl.
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
*
|
|
|
|
|
* UBI volume re-size
|
|
|
|
|
* ~~~~~~~~~~~~~~~~~~
|
|
|
|
|
*
|
2009-01-07 15:10:58 +01:00
|
|
|
|
* To re-size a volume, the %UBI_IOCRSVOL ioctl command of the UBI character
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
* device should be used. A &struct ubi_rsvol_req object has to be properly
|
2009-01-07 15:10:58 +01:00
|
|
|
|
* filled and a pointer to it has to be passed to the ioctl.
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
*
|
2008-07-13 20:47:47 +02:00
|
|
|
|
* UBI volumes re-name
|
|
|
|
|
* ~~~~~~~~~~~~~~~~~~~
|
|
|
|
|
*
|
|
|
|
|
* To re-name several volumes atomically at one go, the %UBI_IOCRNVOL command
|
|
|
|
|
* of the UBI character device should be used. A &struct ubi_rnvol_req object
|
2009-01-07 15:10:58 +01:00
|
|
|
|
* has to be properly filled and a pointer to it has to be passed to the ioctl.
|
2008-07-13 20:47:47 +02:00
|
|
|
|
*
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
* UBI volume update
|
|
|
|
|
* ~~~~~~~~~~~~~~~~~
|
|
|
|
|
*
|
2009-01-07 15:10:58 +01:00
|
|
|
|
* Volume update should be done via the %UBI_IOCVOLUP ioctl command of the
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
* corresponding UBI volume character device. A pointer to a 64-bit update
|
2009-01-07 15:10:58 +01:00
|
|
|
|
* size should be passed to the ioctl. After this, UBI expects user to write
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
* this number of bytes to the volume character device. The update is finished
|
|
|
|
|
* when the claimed number of bytes is passed. So, the volume update sequence
|
|
|
|
|
* is something like:
|
|
|
|
|
*
|
|
|
|
|
* fd = open("/dev/my_volume");
|
|
|
|
|
* ioctl(fd, UBI_IOCVOLUP, &image_size);
|
|
|
|
|
* write(fd, buf, image_size);
|
|
|
|
|
* close(fd);
|
2008-01-24 15:15:14 +01:00
|
|
|
|
*
|
2009-01-07 15:10:58 +01:00
|
|
|
|
* Logical eraseblock erase
|
2008-01-24 15:15:14 +01:00
|
|
|
|
* ~~~~~~~~~~~~~~~~~~~~~~~~
|
|
|
|
|
*
|
2009-01-07 15:10:58 +01:00
|
|
|
|
* To erase a logical eraseblock, the %UBI_IOCEBER ioctl command of the
|
|
|
|
|
* corresponding UBI volume character device should be used. This command
|
|
|
|
|
* unmaps the requested logical eraseblock, makes sure the corresponding
|
|
|
|
|
* physical eraseblock is successfully erased, and returns.
|
|
|
|
|
*
|
|
|
|
|
* Atomic logical eraseblock change
|
|
|
|
|
* ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
|
|
|
|
*
|
|
|
|
|
* Atomic logical eraseblock change operation is called using the %UBI_IOCEBCH
|
|
|
|
|
* ioctl command of the corresponding UBI volume character device. A pointer to
|
|
|
|
|
* a &struct ubi_leb_change_req object has to be passed to the ioctl. Then the
|
|
|
|
|
* user is expected to write the requested amount of bytes (similarly to what
|
|
|
|
|
* should be done in case of the "volume update" ioctl).
|
|
|
|
|
*
|
|
|
|
|
* Logical eraseblock map
|
|
|
|
|
* ~~~~~~~~~~~~~~~~~~~~~
|
|
|
|
|
*
|
|
|
|
|
* To map a logical eraseblock to a physical eraseblock, the %UBI_IOCEBMAP
|
|
|
|
|
* ioctl command should be used. A pointer to a &struct ubi_map_req object is
|
|
|
|
|
* expected to be passed. The ioctl maps the requested logical eraseblock to
|
|
|
|
|
* a physical eraseblock and returns. Only non-mapped logical eraseblocks can
|
|
|
|
|
* be mapped. If the logical eraseblock specified in the request is already
|
|
|
|
|
* mapped to a physical eraseblock, the ioctl fails and returns error.
|
|
|
|
|
*
|
|
|
|
|
* Logical eraseblock unmap
|
|
|
|
|
* ~~~~~~~~~~~~~~~~~~~~~~~~
|
|
|
|
|
*
|
|
|
|
|
* To unmap a logical eraseblock to a physical eraseblock, the %UBI_IOCEBUNMAP
|
|
|
|
|
* ioctl command should be used. The ioctl unmaps the logical eraseblocks,
|
|
|
|
|
* schedules corresponding physical eraseblock for erasure, and returns. Unlike
|
|
|
|
|
* the "LEB erase" command, it does not wait for the physical eraseblock being
|
|
|
|
|
* erased. Note, the side effect of this is that if an unclean reboot happens
|
|
|
|
|
* after the unmap ioctl returns, you may find the LEB mapped again to the same
|
|
|
|
|
* physical eraseblock after the UBI is run again.
|
|
|
|
|
*
|
|
|
|
|
* Check if logical eraseblock is mapped
|
|
|
|
|
* ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
|
|
|
|
*
|
|
|
|
|
* To check if a logical eraseblock is mapped to a physical eraseblock, the
|
|
|
|
|
* %UBI_IOCEBISMAP ioctl command should be used. It returns %0 if the LEB is
|
|
|
|
|
* not mapped, and %1 if it is mapped.
|
2009-01-27 10:11:46 +01:00
|
|
|
|
*
|
|
|
|
|
* Set an UBI volume property
|
|
|
|
|
* ~~~~~~~~~~~~~~~~~~~~~~~~~
|
|
|
|
|
*
|
|
|
|
|
* To set an UBI volume property the %UBI_IOCSETPROP ioctl command should be
|
2011-03-15 15:25:38 +01:00
|
|
|
|
* used. A pointer to a &struct ubi_set_vol_prop_req object is expected to be
|
2009-01-27 10:11:46 +01:00
|
|
|
|
* passed. The object describes which property should be set, and to which value
|
|
|
|
|
* it should be set.
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
/*
|
2007-12-18 17:22:16 +01:00
|
|
|
|
* When a new UBI volume or UBI device is created, users may either specify the
|
|
|
|
|
* volume/device number they want to create or to let UBI automatically assign
|
|
|
|
|
* the number using these constants.
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
*/
|
|
|
|
|
#define UBI_VOL_NUM_AUTO (-1)
|
2007-12-18 17:22:16 +01:00
|
|
|
|
#define UBI_DEV_NUM_AUTO (-1)
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
|
|
|
|
|
/* Maximum volume name length */
|
|
|
|
|
#define UBI_MAX_VOLUME_NAME 127
|
|
|
|
|
|
2009-01-07 15:10:58 +01:00
|
|
|
|
/* ioctl commands of UBI character devices */
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
|
|
|
|
|
#define UBI_IOC_MAGIC 'o'
|
|
|
|
|
|
|
|
|
|
/* Create an UBI volume */
|
|
|
|
|
#define UBI_IOCMKVOL _IOW(UBI_IOC_MAGIC, 0, struct ubi_mkvol_req)
|
|
|
|
|
/* Remove an UBI volume */
|
2009-02-26 00:51:41 +01:00
|
|
|
|
#define UBI_IOCRMVOL _IOW(UBI_IOC_MAGIC, 1, __s32)
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
/* Re-size an UBI volume */
|
|
|
|
|
#define UBI_IOCRSVOL _IOW(UBI_IOC_MAGIC, 2, struct ubi_rsvol_req)
|
2008-07-13 20:47:47 +02:00
|
|
|
|
/* Re-name volumes */
|
|
|
|
|
#define UBI_IOCRNVOL _IOW(UBI_IOC_MAGIC, 3, struct ubi_rnvol_req)
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
|
2009-01-07 15:10:58 +01:00
|
|
|
|
/* ioctl commands of the UBI control character device */
|
2007-12-18 17:22:16 +01:00
|
|
|
|
|
|
|
|
|
#define UBI_CTRL_IOC_MAGIC 'o'
|
|
|
|
|
|
|
|
|
|
/* Attach an MTD device */
|
|
|
|
|
#define UBI_IOCATT _IOW(UBI_CTRL_IOC_MAGIC, 64, struct ubi_attach_req)
|
|
|
|
|
/* Detach an MTD device */
|
2009-02-26 00:51:41 +01:00
|
|
|
|
#define UBI_IOCDET _IOW(UBI_CTRL_IOC_MAGIC, 65, __s32)
|
2007-12-18 17:22:16 +01:00
|
|
|
|
|
2009-01-07 15:10:58 +01:00
|
|
|
|
/* ioctl commands of UBI volume character devices */
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
|
|
|
|
|
#define UBI_VOL_IOC_MAGIC 'O'
|
|
|
|
|
|
|
|
|
|
/* Start UBI volume update */
|
2009-02-26 00:51:41 +01:00
|
|
|
|
#define UBI_IOCVOLUP _IOW(UBI_VOL_IOC_MAGIC, 0, __s64)
|
2009-01-07 15:10:58 +01:00
|
|
|
|
/* LEB erasure command, used for debugging, disabled by default */
|
2009-02-26 00:51:41 +01:00
|
|
|
|
#define UBI_IOCEBER _IOW(UBI_VOL_IOC_MAGIC, 1, __s32)
|
2009-01-07 15:10:58 +01:00
|
|
|
|
/* Atomic LEB change command */
|
2009-02-26 00:51:41 +01:00
|
|
|
|
#define UBI_IOCEBCH _IOW(UBI_VOL_IOC_MAGIC, 2, __s32)
|
2009-01-07 15:10:58 +01:00
|
|
|
|
/* Map LEB command */
|
2009-01-05 14:44:11 +01:00
|
|
|
|
#define UBI_IOCEBMAP _IOW(UBI_VOL_IOC_MAGIC, 3, struct ubi_map_req)
|
2009-01-07 15:10:58 +01:00
|
|
|
|
/* Unmap LEB command */
|
2009-02-26 00:51:41 +01:00
|
|
|
|
#define UBI_IOCEBUNMAP _IOW(UBI_VOL_IOC_MAGIC, 4, __s32)
|
2009-01-07 15:10:58 +01:00
|
|
|
|
/* Check if LEB is mapped command */
|
2009-02-26 00:51:41 +01:00
|
|
|
|
#define UBI_IOCEBISMAP _IOR(UBI_VOL_IOC_MAGIC, 5, __s32)
|
2009-01-27 10:11:46 +01:00
|
|
|
|
/* Set an UBI volume property */
|
2011-03-15 15:25:38 +01:00
|
|
|
|
#define UBI_IOCSETVOLPROP _IOW(UBI_VOL_IOC_MAGIC, 6, \
|
|
|
|
|
struct ubi_set_vol_prop_req)
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
|
2007-12-18 17:22:16 +01:00
|
|
|
|
/* Maximum MTD device name length supported by UBI */
|
|
|
|
|
#define MAX_UBI_MTD_NAME_LEN 127
|
|
|
|
|
|
2008-07-13 20:47:47 +02:00
|
|
|
|
/* Maximum amount of UBI volumes that can be re-named at one go */
|
|
|
|
|
#define UBI_MAX_RNVOL 32
|
|
|
|
|
|
2008-01-24 15:15:14 +01:00
|
|
|
|
/*
|
|
|
|
|
* UBI data type hint constants.
|
|
|
|
|
*
|
|
|
|
|
* UBI_LONGTERM: long-term data
|
|
|
|
|
* UBI_SHORTTERM: short-term data
|
|
|
|
|
* UBI_UNKNOWN: data persistence is unknown
|
|
|
|
|
*
|
|
|
|
|
* These constants are used when data is written to UBI volumes in order to
|
|
|
|
|
* help the UBI wear-leveling unit to find more appropriate physical
|
|
|
|
|
* eraseblocks.
|
|
|
|
|
*/
|
|
|
|
|
enum {
|
|
|
|
|
UBI_LONGTERM = 1,
|
|
|
|
|
UBI_SHORTTERM = 2,
|
|
|
|
|
UBI_UNKNOWN = 3,
|
|
|
|
|
};
|
|
|
|
|
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
/*
|
|
|
|
|
* UBI volume type constants.
|
|
|
|
|
*
|
|
|
|
|
* @UBI_DYNAMIC_VOLUME: dynamic volume
|
|
|
|
|
* @UBI_STATIC_VOLUME: static volume
|
|
|
|
|
*/
|
|
|
|
|
enum {
|
|
|
|
|
UBI_DYNAMIC_VOLUME = 3,
|
2008-01-24 15:15:14 +01:00
|
|
|
|
UBI_STATIC_VOLUME = 4,
|
2007-12-18 17:22:16 +01:00
|
|
|
|
};
|
|
|
|
|
|
2009-01-27 10:11:46 +01:00
|
|
|
|
/*
|
2011-03-15 15:37:57 +01:00
|
|
|
|
* UBI set volume property ioctl constants.
|
2009-01-27 10:11:46 +01:00
|
|
|
|
*
|
2011-03-15 15:37:57 +01:00
|
|
|
|
* @UBI_VOL_PROP_DIRECT_WRITE: allow (any non-zero value) or disallow (value 0)
|
|
|
|
|
* user to directly write and erase individual
|
|
|
|
|
* eraseblocks on dynamic volumes
|
2009-01-27 10:11:46 +01:00
|
|
|
|
*/
|
|
|
|
|
enum {
|
2011-03-15 15:37:57 +01:00
|
|
|
|
UBI_VOL_PROP_DIRECT_WRITE = 1,
|
2009-01-27 10:11:46 +01:00
|
|
|
|
};
|
|
|
|
|
|
2007-12-18 17:22:16 +01:00
|
|
|
|
/**
|
|
|
|
|
* struct ubi_attach_req - attach MTD device request.
|
|
|
|
|
* @ubi_num: UBI device number to create
|
|
|
|
|
* @mtd_num: MTD device number to attach
|
|
|
|
|
* @vid_hdr_offset: VID header offset (use defaults if %0)
|
|
|
|
|
* @padding: reserved for future, not used, has to be zeroed
|
|
|
|
|
*
|
|
|
|
|
* This data structure is used to specify MTD device UBI has to attach and the
|
|
|
|
|
* parameters it has to use. The number which should be assigned to the new UBI
|
2008-01-24 15:15:14 +01:00
|
|
|
|
* device is passed in @ubi_num. UBI may automatically assign the number if
|
2007-12-18 17:22:16 +01:00
|
|
|
|
* @UBI_DEV_NUM_AUTO is passed. In this case, the device number is returned in
|
|
|
|
|
* @ubi_num.
|
|
|
|
|
*
|
|
|
|
|
* Most applications should pass %0 in @vid_hdr_offset to make UBI use default
|
|
|
|
|
* offset of the VID header within physical eraseblocks. The default offset is
|
|
|
|
|
* the next min. I/O unit after the EC header. For example, it will be offset
|
|
|
|
|
* 512 in case of a 512 bytes page NAND flash with no sub-page support. Or
|
|
|
|
|
* it will be 512 in case of a 2KiB page NAND flash with 4 512-byte sub-pages.
|
|
|
|
|
*
|
|
|
|
|
* But in rare cases, if this optimizes things, the VID header may be placed to
|
2008-07-18 12:19:52 +02:00
|
|
|
|
* a different offset. For example, the boot-loader might do things faster if
|
|
|
|
|
* the VID header sits at the end of the first 2KiB NAND page with 4 sub-pages.
|
|
|
|
|
* As the boot-loader would not normally need to read EC headers (unless it
|
|
|
|
|
* needs UBI in RW mode), it might be faster to calculate ECC. This is weird
|
|
|
|
|
* example, but it real-life example. So, in this example, @vid_hdr_offer would
|
|
|
|
|
* be 2KiB-64 bytes = 1984. Note, that this position is not even 512-bytes
|
|
|
|
|
* aligned, which is OK, as UBI is clever enough to realize this is 4th
|
|
|
|
|
* sub-page of the first page and add needed padding.
|
2007-12-18 17:22:16 +01:00
|
|
|
|
*/
|
|
|
|
|
struct ubi_attach_req {
|
2009-02-26 00:51:41 +01:00
|
|
|
|
__s32 ubi_num;
|
|
|
|
|
__s32 mtd_num;
|
|
|
|
|
__s32 vid_hdr_offset;
|
|
|
|
|
__s8 padding[12];
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* struct ubi_mkvol_req - volume description data structure used in
|
2007-12-18 17:22:16 +01:00
|
|
|
|
* volume creation requests.
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
* @vol_id: volume number
|
|
|
|
|
* @alignment: volume alignment
|
|
|
|
|
* @bytes: volume size in bytes
|
|
|
|
|
* @vol_type: volume type (%UBI_DYNAMIC_VOLUME or %UBI_STATIC_VOLUME)
|
2007-12-18 17:22:16 +01:00
|
|
|
|
* @padding1: reserved for future, not used, has to be zeroed
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
* @name_len: volume name length
|
2007-12-18 17:22:16 +01:00
|
|
|
|
* @padding2: reserved for future, not used, has to be zeroed
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
* @name: volume name
|
|
|
|
|
*
|
2008-01-24 15:15:14 +01:00
|
|
|
|
* This structure is used by user-space programs when creating new volumes. The
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
* @used_bytes field is only necessary when creating static volumes.
|
|
|
|
|
*
|
|
|
|
|
* The @alignment field specifies the required alignment of the volume logical
|
|
|
|
|
* eraseblock. This means, that the size of logical eraseblocks will be aligned
|
|
|
|
|
* to this number, i.e.,
|
|
|
|
|
* (UBI device logical eraseblock size) mod (@alignment) = 0.
|
|
|
|
|
*
|
|
|
|
|
* To put it differently, the logical eraseblock of this volume may be slightly
|
|
|
|
|
* shortened in order to make it properly aligned. The alignment has to be
|
|
|
|
|
* multiple of the flash minimal input/output unit, or %1 to utilize the entire
|
|
|
|
|
* available space of logical eraseblocks.
|
|
|
|
|
*
|
|
|
|
|
* The @alignment field may be useful, for example, when one wants to maintain
|
|
|
|
|
* a block device on top of an UBI volume. In this case, it is desirable to fit
|
|
|
|
|
* an integer number of blocks in logical eraseblocks of this UBI volume. With
|
|
|
|
|
* alignment it is possible to update this volume using plane UBI volume image
|
|
|
|
|
* BLOBs, without caring about how to properly align them.
|
|
|
|
|
*/
|
|
|
|
|
struct ubi_mkvol_req {
|
2009-02-26 00:51:41 +01:00
|
|
|
|
__s32 vol_id;
|
|
|
|
|
__s32 alignment;
|
|
|
|
|
__s64 bytes;
|
|
|
|
|
__s8 vol_type;
|
|
|
|
|
__s8 padding1;
|
|
|
|
|
__s16 name_len;
|
|
|
|
|
__s8 padding2[4];
|
2007-12-18 17:22:16 +01:00
|
|
|
|
char name[UBI_MAX_VOLUME_NAME + 1];
|
2011-03-28 09:04:09 +02:00
|
|
|
|
} __packed;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* struct ubi_rsvol_req - a data structure used in volume re-size requests.
|
|
|
|
|
* @vol_id: ID of the volume to re-size
|
|
|
|
|
* @bytes: new size of the volume in bytes
|
|
|
|
|
*
|
|
|
|
|
* Re-sizing is possible for both dynamic and static volumes. But while dynamic
|
|
|
|
|
* volumes may be re-sized arbitrarily, static volumes cannot be made to be
|
2008-10-16 19:02:37 +02:00
|
|
|
|
* smaller than the number of bytes they bear. To arbitrarily shrink a static
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
* volume, it must be wiped out first (by means of volume update operation with
|
|
|
|
|
* zero number of bytes).
|
|
|
|
|
*/
|
|
|
|
|
struct ubi_rsvol_req {
|
2009-02-26 00:51:41 +01:00
|
|
|
|
__s64 bytes;
|
|
|
|
|
__s32 vol_id;
|
2011-03-28 09:04:09 +02:00
|
|
|
|
} __packed;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
|
2008-07-13 20:47:47 +02:00
|
|
|
|
/**
|
|
|
|
|
* struct ubi_rnvol_req - volumes re-name request.
|
|
|
|
|
* @count: count of volumes to re-name
|
|
|
|
|
* @padding1: reserved for future, not used, has to be zeroed
|
|
|
|
|
* @vol_id: ID of the volume to re-name
|
|
|
|
|
* @name_len: name length
|
|
|
|
|
* @padding2: reserved for future, not used, has to be zeroed
|
|
|
|
|
* @name: new volume name
|
|
|
|
|
*
|
|
|
|
|
* UBI allows to re-name up to %32 volumes at one go. The count of volumes to
|
|
|
|
|
* re-name is specified in the @count field. The ID of the volumes to re-name
|
|
|
|
|
* and the new names are specified in the @vol_id and @name fields.
|
|
|
|
|
*
|
|
|
|
|
* The UBI volume re-name operation is atomic, which means that should power cut
|
|
|
|
|
* happen, the volumes will have either old name or new name. So the possible
|
|
|
|
|
* use-cases of this command is atomic upgrade. Indeed, to upgrade, say, volumes
|
|
|
|
|
* A and B one may create temporary volumes %A1 and %B1 with the new contents,
|
|
|
|
|
* then atomically re-name A1->A and B1->B, in which case old %A and %B will
|
|
|
|
|
* be removed.
|
|
|
|
|
*
|
|
|
|
|
* If it is not desirable to remove old A and B, the re-name request has to
|
|
|
|
|
* contain 4 entries: A1->A, A->A1, B1->B, B->B1, in which case old A1 and B1
|
|
|
|
|
* become A and B, and old A and B will become A1 and B1.
|
|
|
|
|
*
|
|
|
|
|
* It is also OK to request: A1->A, A1->X, B1->B, B->Y, in which case old A1
|
|
|
|
|
* and B1 become A and B, and old A and B become X and Y.
|
|
|
|
|
*
|
|
|
|
|
* In other words, in case of re-naming into an existing volume name, the
|
|
|
|
|
* existing volume is removed, unless it is re-named as well at the same
|
|
|
|
|
* re-name request.
|
|
|
|
|
*/
|
|
|
|
|
struct ubi_rnvol_req {
|
2009-02-26 00:51:41 +01:00
|
|
|
|
__s32 count;
|
|
|
|
|
__s8 padding1[12];
|
2008-07-13 20:47:47 +02:00
|
|
|
|
struct {
|
2009-02-26 00:51:41 +01:00
|
|
|
|
__s32 vol_id;
|
|
|
|
|
__s16 name_len;
|
|
|
|
|
__s8 padding2[2];
|
2008-07-13 20:47:47 +02:00
|
|
|
|
char name[UBI_MAX_VOLUME_NAME + 1];
|
|
|
|
|
} ents[UBI_MAX_RNVOL];
|
2011-03-28 09:04:09 +02:00
|
|
|
|
} __packed;
|
2008-07-13 20:47:47 +02:00
|
|
|
|
|
2008-01-24 15:15:14 +01:00
|
|
|
|
/**
|
2009-01-07 15:10:58 +01:00
|
|
|
|
* struct ubi_leb_change_req - a data structure used in atomic LEB change
|
|
|
|
|
* requests.
|
2008-01-24 15:15:14 +01:00
|
|
|
|
* @lnum: logical eraseblock number to change
|
|
|
|
|
* @bytes: how many bytes will be written to the logical eraseblock
|
|
|
|
|
* @dtype: data type (%UBI_LONGTERM, %UBI_SHORTTERM, %UBI_UNKNOWN)
|
|
|
|
|
* @padding: reserved for future, not used, has to be zeroed
|
|
|
|
|
*/
|
|
|
|
|
struct ubi_leb_change_req {
|
2009-02-26 00:51:41 +01:00
|
|
|
|
__s32 lnum;
|
|
|
|
|
__s32 bytes;
|
|
|
|
|
__s8 dtype;
|
|
|
|
|
__s8 padding[7];
|
2011-03-28 09:04:09 +02:00
|
|
|
|
} __packed;
|
2008-01-24 15:15:14 +01:00
|
|
|
|
|
2009-01-05 14:44:11 +01:00
|
|
|
|
/**
|
2009-01-07 15:10:58 +01:00
|
|
|
|
* struct ubi_map_req - a data structure used in map LEB requests.
|
2009-01-05 14:44:11 +01:00
|
|
|
|
* @lnum: logical eraseblock number to unmap
|
|
|
|
|
* @dtype: data type (%UBI_LONGTERM, %UBI_SHORTTERM, %UBI_UNKNOWN)
|
|
|
|
|
* @padding: reserved for future, not used, has to be zeroed
|
|
|
|
|
*/
|
|
|
|
|
struct ubi_map_req {
|
2009-02-26 00:51:41 +01:00
|
|
|
|
__s32 lnum;
|
|
|
|
|
__s8 dtype;
|
|
|
|
|
__s8 padding[3];
|
2011-03-28 09:04:09 +02:00
|
|
|
|
} __packed;
|
2009-01-05 14:44:11 +01:00
|
|
|
|
|
2009-01-27 10:11:46 +01:00
|
|
|
|
|
|
|
|
|
/**
|
2011-03-15 15:37:57 +01:00
|
|
|
|
* struct ubi_set_vol_prop_req - a data structure used to set an UBI volume
|
2011-03-15 15:25:38 +01:00
|
|
|
|
* property.
|
|
|
|
|
* @property: property to set (%UBI_VOL_PROP_DIRECT_WRITE)
|
2009-01-27 10:11:46 +01:00
|
|
|
|
* @padding: reserved for future, not used, has to be zeroed
|
|
|
|
|
* @value: value to set
|
|
|
|
|
*/
|
2011-03-15 15:25:38 +01:00
|
|
|
|
struct ubi_set_vol_prop_req {
|
2011-03-28 09:12:25 +02:00
|
|
|
|
__u8 property;
|
|
|
|
|
__u8 padding[7];
|
|
|
|
|
__u64 value;
|
2011-03-28 09:04:09 +02:00
|
|
|
|
} __packed;
|
2009-01-27 10:11:46 +01:00
|
|
|
|
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 10:22:22 +02:00
|
|
|
|
#endif /* __UBI_USER_H__ */
|