Patchwork [V2] mtd: spi-nor: add Kconfig option to disable 4K sectors

login
register
mail settings
Submitter Rafał Miłecki
Date Aug. 17, 2014, 9:27 a.m.
Message ID <1408267646-6643-1-git-send-email-zajec5@gmail.com>
Download mbox | patch
Permalink /patch/380565/
State Accepted
Headers show

Comments

Rafał Miłecki - Aug. 17, 2014, 9:27 a.m.
Current situation with 4K sectors is quite messy. First of all, some
MTD "users" don't work with such small size. An example may be UBIFS
which requires 15 KiB erase blocks as a minimum. In theory spi-nor
should provide multiple erase regions and MTD "users" should use the
one they need. Unforunately that is not implemented.

In the result our flashes database in spi-nor is hackish. For some
flashes we pretend they don't support 4K sectors just because some
distribution uses UBIFS on it. This ofc leads to conflicts, like
Samsung using w25q128 with 4K sectors vs. OpenWrt requiring it to
pretend it's 64 KiB blocks only.

My idea (plan?) for fixing this situation:
1) Use real hw info (this requires a way for disabling 4K for now)
2) Provide detailed info about erase regions
3) Make UBIFS work with devices that support 4K sectors

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
---
V2: Add extra info about performance in Kconfig. Thanks Kevin.
    Fix typo s/users/uses/ in commit message.
---
 drivers/mtd/spi-nor/Kconfig   | 14 ++++++++++++++
 drivers/mtd/spi-nor/spi-nor.c |  5 ++++-
 2 files changed, 18 insertions(+), 1 deletion(-)
Brian Norris - Sept. 28, 2014, 12:28 a.m.
On Sun, Aug 17, 2014 at 11:27:26AM +0200, Rafał Miłecki wrote:
> Current situation with 4K sectors is quite messy. First of all, some
> MTD "users" don't work with such small size. An example may be UBIFS
> which requires 15 KiB erase blocks as a minimum. In theory spi-nor
> should provide multiple erase regions and MTD "users" should use the
> one they need. Unforunately that is not implemented.
> 
> In the result our flashes database in spi-nor is hackish. For some
> flashes we pretend they don't support 4K sectors just because some
> distribution uses UBIFS on it. This ofc leads to conflicts, like
> Samsung using w25q128 with 4K sectors vs. OpenWrt requiring it to
> pretend it's 64 KiB blocks only.
> 
> My idea (plan?) for fixing this situation:
> 1) Use real hw info (this requires a way for disabling 4K for now)
> 2) Provide detailed info about erase regions
> 3) Make UBIFS work with devices that support 4K sectors
> 
> Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
> ---
> V2: Add extra info about performance in Kconfig. Thanks Kevin.
>     Fix typo s/users/uses/ in commit message.

Pushed to l2-mtd.git. Thanks!

Now we should probably try to handle (1), at least, so that the data
table holds fully accurate info about the flash.

Brian
Rafał Miłecki - Sept. 28, 2014, 8:29 p.m.
On 28 September 2014 02:28, Brian Norris <computersforpeace@gmail.com> wrote:
> On Sun, Aug 17, 2014 at 11:27:26AM +0200, Rafał Miłecki wrote:
>> Current situation with 4K sectors is quite messy. First of all, some
>> MTD "users" don't work with such small size. An example may be UBIFS
>> which requires 15 KiB erase blocks as a minimum. In theory spi-nor
>> should provide multiple erase regions and MTD "users" should use the
>> one they need. Unforunately that is not implemented.
>>
>> In the result our flashes database in spi-nor is hackish. For some
>> flashes we pretend they don't support 4K sectors just because some
>> distribution uses UBIFS on it. This ofc leads to conflicts, like
>> Samsung using w25q128 with 4K sectors vs. OpenWrt requiring it to
>> pretend it's 64 KiB blocks only.
>>
>> My idea (plan?) for fixing this situation:
>> 1) Use real hw info (this requires a way for disabling 4K for now)
>> 2) Provide detailed info about erase regions
>> 3) Make UBIFS work with devices that support 4K sectors
>>
>> Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
>> ---
>> V2: Add extra info about performance in Kconfig. Thanks Kevin.
>>     Fix typo s/users/uses/ in commit message.
>
> Pushed to l2-mtd.git. Thanks!

Ooops, for some reason the whole diff is part of the commit message :|
http://git.infradead.org/l2-mtd.git/commitdiff/f83cd098095c0e2967e15914f3f9521bb122bc0a


> Now we should probably try to handle (1), at least, so that the data
> table holds fully accurate info about the flash.

I agree, will try to collect real info and submit a patch!
Brian Norris - Sept. 28, 2014, 9:17 p.m.
On Sun, Sep 28, 2014 at 10:29:42PM +0200, Rafał Miłecki wrote:
> On 28 September 2014 02:28, Brian Norris <computersforpeace@gmail.com> wrote:
> > Pushed to l2-mtd.git. Thanks!
> 
> Ooops, for some reason the whole diff is part of the commit message :|
> http://git.infradead.org/l2-mtd.git/commitdiff/f83cd098095c0e2967e15914f3f9521bb122bc0a

Oops indeed. Not sure how I did that. Fixed now, thanks for reporting.

Brian

Patch

diff --git a/drivers/mtd/spi-nor/Kconfig b/drivers/mtd/spi-nor/Kconfig
index f8acfa4..abab223 100644
--- a/drivers/mtd/spi-nor/Kconfig
+++ b/drivers/mtd/spi-nor/Kconfig
@@ -7,6 +7,20 @@  menuconfig MTD_SPI_NOR
 
 if MTD_SPI_NOR
 
+config MTD_SPI_NOR_USE_4K_SECTORS
+	bool "Use small 4096 B erase sectors"
+	default y
+	help
+	  Many flash memories support erasing small (4096 B) sectors. Depending
+	  on the usage this feature may provide performance gain in comparison
+	  to erasing whole blocks (32/64 KiB).
+	  Changing small part of flash content is usually faster with small
+	  sectors. On the other hand erasing should be faster when using 64 KiB
+	  block instead of 16 × 4 KiB sectors.
+
+	  Please note that some tools/drivers/filesystems may not work with
+	  4096 B erase size (e.g. UBIFS requires 15 KiB as a minimum).
+
 config SPI_FSL_QUADSPI
 	tristate "Freescale Quad SPI controller"
 	depends on ARCH_MXC
diff --git a/drivers/mtd/spi-nor/spi-nor.c b/drivers/mtd/spi-nor/spi-nor.c
index ba3f61e..ae16aa2 100644
--- a/drivers/mtd/spi-nor/spi-nor.c
+++ b/drivers/mtd/spi-nor/spi-nor.c
@@ -987,6 +987,7 @@  int spi_nor_scan(struct spi_nor *nor, const struct spi_device_id *id,
 	    nor->wait_till_ready == spi_nor_wait_till_ready)
 		nor->wait_till_ready = spi_nor_wait_till_fsr_ready;
 
+#ifdef CONFIG_MTD_SPI_NOR_USE_4K_SECTORS
 	/* prefer "small sector" erase if possible */
 	if (info->flags & SECT_4K) {
 		nor->erase_opcode = SPINOR_OP_BE_4K;
@@ -994,7 +995,9 @@  int spi_nor_scan(struct spi_nor *nor, const struct spi_device_id *id,
 	} else if (info->flags & SECT_4K_PMC) {
 		nor->erase_opcode = SPINOR_OP_BE_4K_PMC;
 		mtd->erasesize = 4096;
-	} else {
+	} else
+#endif
+	{
 		nor->erase_opcode = SPINOR_OP_SE;
 		mtd->erasesize = info->sector_size;
 	}