mtd: spi-nor: write support for minor aligned partitions
Subsystems (2)
PATCH 1/1 mtd: spi-nor: write support for minor aligned partitions
Do not prevent writing to mtd partitions where a partition boundary sits on a minor erasesize boundary.
This addresses a FIXME that has been present since the start of the linux git history:
/* Doesn't start on a boundary of major erase size */
/* FIXME: Let it be writable if it is on a boundary of
* _minor_ erase size though */
Allow a uniform erase region spi-nor device to be configured to use the non-uniform erase regions code path for an erase with:CONFIG_MTD_SPI_NOR_USE_VARIABLE_ERASE=y
On supporting hardware (SECT_4K: majority of current SPI-NOR device) provide the facility for an erase to use the least number of SPI-NOR operations, as well as access to 4K erase without requiring CONFIG_MTD_SPI_NOR_USE_4K_SECTORS
Introduce erasesize_minor to the mtd struct,
the smallest erasesize supported by the device
On existing devices, this is useful where write support is wanted for data on a 4K partition, such as some u-boot-env partitions, or RouterBoot soft_config, while still netting the performance benefits of using 64K sectors
Performance:
time mtd erase firmware
OpenWrt 5.10 ramips MT7621 w25q128jv 0xfc0000 partition length
Without this patch
MTD_SPI_NOR_USE_4K_SECTORS=y |n
real 2m 11.66s |0m 50.86s user 0m 0.00s |0m 0.00s sys 1m 56.20s |0m 50.80s
With this patch
MTD_SPI_NOR_USE_VARIABLE_ERASE=n|y |4K_SECTORS=y
real 0m 51.68s |0m 50.85s |2m 12.89s user 0m 0.00s |0m 0.00s |0m 0.01s sys 0m 46.94s |0m 50.38s |2m 12.46s
---
Have not tested on variable erase regions device.
checkpatch does not like the printk(KERN_WARNING
these should be changed separately beforehand?
Changes RFC -> v1:
Fix uninitialized variable smatch warning
--- drivers/mtd/mtdpart.c | 52 ++++++++++++++++++++++++++++--------- drivers/mtd/spi-nor/Kconfig | 10 +++++++ drivers/mtd/spi-nor/core.c | 10 +++++-- include/linux/mtd/mtd.h | 2 ++ 4 files changed, 60 insertions(+), 14 deletions(-)
drivers/mtd/mtdpart.c +40 -12
@@ -38,10 +38,11 @@ static struct mtd_info *allocate_partition(struct mtd_info *parent,struct mtd_info *master =mtd_get_master (parent);int wr_alignment = (parent->flags &MTD_NO_ERASE ) ? master->writesize : master->erasesize; +int wr_alignment_minor =0 ;u64 parent_size =mtd_is_partition (parent) ? parent->part.size : parent->size;struct mtd_info *child; -u32 remainder; +u32 remainder, remainder_minor;char *name;u64 tmp;
@@ -143,6 +144,7 @@ static struct mtd_info *allocate_partition(struct mtd_info *parent,int i, max = parent->numeraseregions;u64 end = child->part.offset + child->part.size;struct mtd_erase_region_info *regions = parent->eraseregions; +uint32_t erasesize_minor = child->erasesize;/* Find the first erase regions which is part of this * partition. */
@@ -153,15 +155,24 @@ static struct mtd_info *allocate_partition(struct mtd_info *parent,if (i >0 ) i--; -/* Pick biggest erasesize */ for (; i < max && regions[i].offset < end; i++) { +/* Pick biggest erasesize */ if (child->erasesize < regions[i].erasesize) child->erasesize = regions[i].erasesize; +/* Pick smallest non-zero erasesize */ +if ((erasesize_minor > regions[i].erasesize) && (regions[i].erasesize >0 )) + erasesize_minor = regions[i].erasesize; } + +if (erasesize_minor < child->erasesize) + child->erasesize_minor = erasesize_minor; +BUG_ON (child->erasesize ==0 ); }else {/* Single erase size */ child->erasesize = master->erasesize; +if (master->erasesize_minor) + child->erasesize_minor = master->erasesize_minor; }/*
@@ -169,26 +180,43 @@ static struct mtd_info *allocate_partition(struct mtd_info *parent,* exposes several regions with different erasesize. Adjust * wr_alignment accordingly. */ -if (!(child->flags &MTD_NO_ERASE )) +if (!(child->flags &MTD_NO_ERASE )) { wr_alignment = child->erasesize; +if (IS_ENABLED (CONFIG_MTD_SPI_NOR_USE_VARIABLE_ERASE ) && child->erasesize_minor) + wr_alignment_minor = child->erasesize_minor; + } tmp =mtd_get_master_ofs (child,0 ); remainder =do_div (tmp, wr_alignment);if ((child->flags &MTD_WRITEABLE ) && remainder) { -/* Doesn't start on a boundary of major erase size */ -/* FIXME: Let it be writable if it is on a boundary of -* _minor_ erase size though */ - child->flags &= ~MTD_WRITEABLE ; -printk (KERN_WARNING "mtd: partition \"%s\" doesn't start on an erase/write block boundary -- force read-only\n" , - part->name); +if (wr_alignment_minor) { + tmp =mtd_get_master_ofs (child,0 ); + remainder_minor =do_div (tmp, wr_alignment_minor); +if (remainder_minor ==0 ) + child->erasesize = child->erasesize_minor; + } + +if ((!wr_alignment_minor) || (wr_alignment_minor && remainder_minor !=0 )) { + child->flags &= ~MTD_WRITEABLE ; +printk (KERN_WARNING "mtd: partition \"%s\" doesn't start on an erase/write block boundary -- force read-only\n" , + part->name); + } } tmp =mtd_get_master_ofs (child,0 ) + child->part.size; remainder =do_div (tmp, wr_alignment);if ((child->flags &MTD_WRITEABLE ) && remainder) { - child->flags &= ~MTD_WRITEABLE ; -printk (KERN_WARNING "mtd: partition \"%s\" doesn't end on an erase/write block -- force read-only\n" , - part->name); +if (wr_alignment_minor) { + tmp =mtd_get_master_ofs (child,0 ) + child->part.size; + remainder_minor =do_div (tmp, wr_alignment_minor); +if (remainder_minor ==0 ) + child->erasesize = child->erasesize_minor; + } +if ((!wr_alignment_minor) || (wr_alignment_minor && remainder_minor !=0 )) { + child->flags &= ~MTD_WRITEABLE ; +printk (KERN_WARNING "mtd: partition \"%s\" doesn't end on an erase/write block -- force read-only\n" , + part->name); + } } child->size = child->part.size;
drivers/mtd/spi-nor/Kconfig +10 -0
@@ -10,6 +10,16 @@ menuconfig MTD_SPI_NOR if MTD_SPI_NOR +config MTD_SPI_NOR_USE_VARIABLE_ERASE + bool "Disable uniform_erase to allow use of all hardware supported erasesizes" + depends on !MTD_SPI_NOR_USE_4K_SECTORS + default n + help + Allow mixed use of all hardware supported erasesizes, + by forcing spi_nor to use the multiple eraseregions code path. + For example: A 68K erase will use one 64K erase, and one 4K erase + on supporting hardware. + config MTD_SPI_NOR_USE_4K_SECTORS bool "Use small 4096 B erase sectors" default y
drivers/mtd/spi-nor/core.c +8 -2
@@ -1262,6 +1262,8 @@ static u8 spi_nor_convert_3to4_erase(u8 opcode)static bool spi_nor_has_uniform_erase (const struct spi_nor *nor) { +if (IS_ENABLED (CONFIG_MTD_SPI_NOR_USE_VARIABLE_ERASE )) +return false;return !!nor->params->erase_map.uniform_erase_type; }
@@ -2381,6 +2383,7 @@ static int spi_nor_select_erase(struct spi_nor *nor) {struct spi_nor_erase_map *map = &nor->params->erase_map;const struct spi_nor_erase_type *erase =NULL ; +const struct spi_nor_erase_type *erase_minor =NULL ;struct mtd_info *mtd = &nor->mtd;u32 wanted_size = nor->info->sector_size;int i;
@@ -2413,8 +2416,9 @@ static int spi_nor_select_erase(struct spi_nor *nor)*/ for (i =SNOR_ERASE_TYPE_MAX -1 ; i >=0 ; i--) {if (map->erase_type[i].size) { - erase = &map->erase_type[i]; -break ; +if (!erase) + erase = &map->erase_type[i]; + erase_minor = &map->erase_type[i]; } }
@@ -2422,6 +2426,8 @@ static int spi_nor_select_erase(struct spi_nor *nor)return -EINVAL ; mtd->erasesize = erase->size; +if (erase_minor && erase_minor->size < erase->size) + mtd->erasesize_minor = erase_minor->size;return 0 ; }
include/linux/mtd/mtd.h +2 -0
@@ -243,6 +243,8 @@ struct mtd_info {* information below if they desire */ uint32_t erasesize; +/* "Minor" (smallest) erase size supported by the whole device */ +uint32_t erasesize_minor;/* Minimal writable flash unit size. In case of NOR flash it is 1 (even * though individual bits can be cleared), in case of NAND flash it is * one NAND page (or half, or one-fourths of it), in case of ECC-ed NOR
Signature · 1 line
2.31.1
Quoted 12 lines · John Thomson
Do not prevent writing to mtd partitions where a partition boundary sits on a minor erasesize boundary.
This addresses a FIXME that has been present since the start of the linux git history:
/* Doesn't start on a boundary of major erase size */
/* FIXME: Let it be writable if it is on a boundary of
* _minor_ erase size though */Allow a uniform erase region spi-nor device to be configured to use the non-uniform erase regions code path for an erase with:
CONFIG_MTD_SPI_NOR_USE_VARIABLE_ERASE=y
Quoted 7 lines
On supporting hardware (
SECT_4K: majority of current SPI-NOR device) provide the facility for an erase to use the least number of SPI-NOR operations, as well as access to 4K erase without requiringCONFIG_MTD_SPI_NOR_USE_4K_SECTORSIntroduce
erasesize_minorto the mtd struct,
the smallest erasesize supported by the device
Instead of having just a major and minor erase size, wouldn't it make more sense to have a list of all supported erases on a sector? That is, instead of hard coding two erase size instead of one, how about generalizing the erase machinery to allow any number of erase sizes on a sector and then choosing the most efficient one on run time?
For example, imagine a device can support 4K, 64K, and 256K erases on each sector. Then you should be able to use 2 x 4K erases if you want to erase 8K, 2 x 64K + 1 x 4K erases for 132K and so on.
I am not very familiar with MTD erase machinery so I would like someone else to chime in, but I think my suggestion above is the better way forward.
Signature · 3 lines
Regards, Pratyush Yadav Texas Instruments Inc.
Quoted 27 lines · Pratyush Yadav
Quoted 12 lines · John Thomson
Do not prevent writing to mtd partitions where a partition boundary sits on a minor erasesize boundary.
This addresses a FIXME that has been present since the start of the linux git history:
/* Doesn't start on a boundary of major erase size */
/* FIXME: Let it be writable if it is on a boundary of
* _minor_ erase size though */Allow a uniform erase region spi-nor device to be configured to use the non-uniform erase regions code path for an erase with:
CONFIG_MTD_SPI_NOR_USE_VARIABLE_ERASE=yQuoted 7 lines
On supporting hardware (
SECT_4K: majority of current SPI-NOR device) provide the facility for an erase to use the least number of SPI-NOR operations, as well as access to 4K erase without requiringCONFIG_MTD_SPI_NOR_USE_4K_SECTORSIntroduce
erasesize_minorto the mtd struct,
the smallest erasesize supported by the deviceInstead of having just a major and minor erase size, wouldn't it make more sense to have a list of all supported erases on a sector? That is, instead of hard coding two erase size instead of one, how about generalizing the erase machinery to allow any number of erase sizes on a sector and then choosing the most efficient one on run time?
Quoted 3 lines
For example, imagine a device can support 4K, 64K, and 256K erases on each sector. Then you should be able to use 2 x 4K erases if you want to erase 8K, 2 x 64K + 1 x 4K erases for 132K and so on.
Thank for you the feedback Pratyush,
Yes, this is what I am attempting to do.
mtdpart only sets if the partition can be writeable, it does not do the erasing. Due to this, I thought that only the smallest erasesize should matter to mtdpart. The erase is carried out by the mtd device, so for an SPI-NOR device, spi_nor_erase_multi_sectors can be used to select and action the best combination of erases. This is done by this patch when CONFIG_MTD_SPI_NOR_USE_VARIABLE_ERASE=y
Cheers,
Signature · 1 line
John Thomson