u-boot/doc/README.ubispl
Thomas Gleixner 6f4e7d3c75 spl: Lightweight UBI and UBI fastmap support
Booting a payload out of NAND FLASH from the SPL is a crux today, as
it requires hard partioned FLASH. Not a brilliant idea with the
reliability of todays NAND FLASH chips.

The upstream UBI + UBI fastmap implementation which is about to
brought to u-boot is too heavy weight for SPLs as it provides way more
functionality than needed for a SPL and does not even fit into the
restricted SPL areas which are loaded from the SoC boot ROM.

So this provides a fast and lightweight implementation of UBI scanning
and UBI fastmap attach. The scan and logical to physical block mapping
code is developed from scratch, while the fastmap implementation is
lifted from the linux kernel source and stripped down to fit the SPL
needs.

The text foot print on the board which I used for development is:

6854	0	0	6854	1abd
drivers/mtd/ubispl/built-in.o

Attaching a NAND chip with 4096 physical eraseblocks (4 blocks are
reserved for the SPL) takes:

In full scan mode:      1172ms
In fastmap mode:          95ms

The code requires quite some storage. The largest and unknown part of
it is the number of fastmap blocks to read. Therefor the data
structure is not put into the BSS. The code requires a pointer to free
memory handed in which is initialized by the UBI attach code itself.

See doc/README.ubispl for further information on how to use it.

This shares the ubi-media.h and crc32 implementation of drivers/mtd/ubi
There is no way to share the fastmap code, as UBISPL only utilizes the
slightly modified functions ubi_attach_fastmap() and ubi_scan_fastmap()
from the original kernel ubi fastmap implementation.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ladislav Michl <ladis@linux-mips.org>
Acked-by: Heiko Schocher <hs@denx.de>
Reviewed-by: Tom Rini <trini@konsulko.com>
2016-07-22 09:53:00 -04:00

142 lines
4.1 KiB
Plaintext

Lightweight UBI and UBI fastmap support
# Copyright (C) Thomas Gleixner <tglx@linutronix.de>
#
# SPDX-License-Identifier: GPL 2.0+ BSD-3-Clause
Scans the UBI information and loads the requested static volumes into
memory.
Configuration Options:
CONFIG_SPL_UBI
Enables the SPL UBI support
CONFIG_SPL_UBI_MAX_VOL_LEBS
The maximum number of logical eraseblocks which a static volume
to load can contain. Used for sizing the scan data structure
CONFIG_SPL_UBI_MAX_PEB_SIZE
The maximum physical erase block size. Either a compile time
constant or runtime detection. Used for sizing the scan data
structure
CONFIG_SPL_UBI_MAX_PEBS
The maximum physical erase block count. Either a compile time
constant or runtime detection. Used for sizing the scan data
structure
CONFIG_SPL_UBI_VOL_IDS
The maximum volume ids which can be loaded. Used for sizing the
scan data structure.
Usage notes:
In the board config file define for example:
#define CONFIG_SPL_UBI
#define CONFIG_SPL_UBI_MAX_VOL_LEBS 256
#define CONFIG_SPL_UBI_MAX_PEB_SIZE (256*1024)
#define CONFIG_SPL_UBI_MAX_PEBS 4096
#define CONFIG_SPL_UBI_VOL_IDS 8
The size requirement is roughly as follows:
2k for the basic data structure
+ CONFIG_SPL_UBI_VOL_IDS * CONFIG_SPL_UBI_MAX_VOL_LEBS * 8
+ CONFIG_SPL_UBI_MAX_PEBS * 64
+ CONFIG_SPL_UBI_MAX_PEB_SIZE * UBI_FM_MAX_BLOCKS
The last one is big, but I really don't care in that stage. Real world
implementations only use the first couple of blocks, but the code
handles up to UBI_FM_MAX_BLOCKS.
Given the above configuration example the requirement is about 5M
which is usually not a problem to reserve in the RAM along with the
other areas like the kernel/dts load address.
So something like this will do the trick:
#define SPL_FINFO_ADDR 0x80800000
#define SPL_DTB_LOAD_ADDR 0x81800000
#define SPL_KERNEL_LOAD_ADDR 0x82000000
In the board file, implement the following:
static struct ubispl_load myvolumes[] = {
{
.vol_id = 0, /* kernel volume */
.load_addr = (void *)SPL_KERNEL_LOAD_ADDR,
},
{
.vol_id = 1, /* DT blob */
.load_addr = (void *)SPL_DTB_LOAD_ADDR,
}
};
int spl_start_uboot(void)
{
struct ubispl_info info;
info.ubi = (struct ubi_scan_info *) SPL_FINFO_ADDR;
info.fastmap = 1;
info.read = nand_spl_read_flash;
#if COMPILE_TIME_DEFINED
/*
* MY_NAND_NR_SPL_PEBS is the number of physical erase blocks
* in the FLASH which are reserved for the SPL. Think about
* mtd partitions:
*
* part_spl { .start = 0, .end = 4 }
* part_ubi { .start = 4, .end = NR_PEBS }
*/
info.peb_offset = MY_NAND_NR_SPL_PEBS;
info.peb_size = CONFIG_SYS_NAND_BLOCK_SIZE;
info.vid_offset = MY_NAND_UBI_VID_OFFS;
info.leb_start = MY_NAND_UBI_DATA_OFFS;
info.peb_count = MY_NAND_UBI_NUM_PEBS;
#else
get_flash_info(&flash_info);
info.peb_offset = MY_NAND_NR_SPL_PEBS;
info.peb_size = flash_info.peb_size;
/*
* The VID and Data offset depend on the capability of the
* FLASH chip to do subpage writes.
*
* If the flash chip supports subpage writes, then the VID
* header starts at the second subpage. So for 2k pages size
* with 4 subpages the VID offset is 512. The DATA offset is 2k.
*
* If the flash chip does not support subpage writes then the
* VID offset is FLASH_PAGE_SIZE and the DATA offset
* 2 * FLASH_PAGE_SIZE
*/
info.vid_offset = flash_info.vid_offset;
info.leb_start = flash_info.data_offset;
/*
* The flash reports the total number of erase blocks, so
* we need to subtract the number of blocks which are reserved
* for the SPL itself and not managed by UBI.
*/
info.peb_count = flash_info.peb_count - MY_NAND_NR_SPL_PEBS;
#endif
ret = ubispl_load_volumes(&info, myvolumes, ARRAY_SIZE(myvolumes);
....
}
Note: you can load any payload that way. You can even load u-boot from
UBI, so the only non UBI managed FLASH area is the one which is
reserved for the SPL itself and read from the SoC ROM.
And you can do fallback scenarios:
if (ubispl_load_volumes(&info, volumes0, ARRAY_SIZE(volumes0)))
if (ubispl_load_volumes(&info, volumes1, ARRAY_SIZE(volumes1)))
ubispl_load_volumes(&info, vol_uboot, ARRAY_SIZE(vol_uboot));