Skip to contentSkip to conversation

omap_gpmc: Do not default to HAM1 when SW ECC is selected

U-Boot In review Adam YH Lee 4 participants 6 messages Feb 18, 2015 – Mar 6, 2015

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

The ECC scheme selection algorithm in OMAP GPMC appears to be left untested when BCH8 handling code was added. Running 'nandecc sw' defaults to HAM1 even if the board is using another scheme (ex. OMAP_ECC_BCH8_CODE_HW_DETECTION_SW on OMAP3). This results in unrecoverable ECC errors when reading data. This commit fixes the behavior by checking for CONFIG_BCH and using the scheme defined by CONFIG_NAND_OMAP_ECCSCHEME in the board configuration file.

This has been tested on Gumstix Overo (OMAP3).

---
 drivers/mtd/nand/omap_gpmc.c | 5 +++++
 1 file changed, 5 insertions(+)
drivers/mtd/nand/omap_gpmc.c +5 -0
index fc64f48..5daf932 100644
@@ -901,8 +901,13 @@ int __maybe_unused omap_nand_switch_ecc(uint32_t hardware, uint32_t eccstrength)
 			return -EINVAL;
 		}
 	} else {
+		#ifdef CONFIG_BCH
+		err = omap_select_ecc_scheme(nand, CONFIG_NAND_OMAP_ECCSCHEME,
+					mtd->writesize, mtd->oobsize);
+		#else
 		err = omap_select_ecc_scheme(nand, OMAP_ECC_HAM1_CODE_SW,
 					mtd->writesize, mtd->oobsize);
+		#endif
 	}
 
 	/* Update NAND handling after ECC mode switch */
Signature · 1 line
1.9.1
#1 lore raw

Hi Adam,

Quoted 19 lines · Adam YH Lee

The ECC scheme selection algorithm in OMAP GPMC appears to be left untested when BCH8 handling code was added. Running 'nandecc sw' defaults to HAM1 even if the board is using another scheme (ex. OMAP_ECC_BCH8_CODE_HW_DETECTION_SW on OMAP3). This results in unrecoverable ECC errors when reading data. This commit fixes the behavior by checking for CONFIG_BCH and using the scheme defined by CONFIG_NAND_OMAP_ECCSCHEME in the board configuration file.

This has been tested on Gumstix Overo (OMAP3).

---
 drivers/mtd/nand/omap_gpmc.c | 5 +++++
 1 file changed, 5 insertions(+)
drivers/mtd/nand/omap_gpmc.c +0 -0
index fc64f48..5daf932 100644

drivers/mtd/nand/omap_gpmc.c int __maybe_unused omap_nand_switch_ecc(uint32_t hardware, uint32_t eccstrength) · line 909

Quoted 9 lines
drivers/mtd/nand/omap_gpmc.c +4 -0
@@ -901,8 +901,13 @@ int __maybe_unused omap_nand_switch_ecc(uint32_t hardware, uint32_t eccstrength)
 			return -EINVAL;
 		}
 	} else {
+		#ifdef CONFIG_BCH
+		err = omap_select_ecc_scheme(nand, CONFIG_NAND_OMAP_ECCSCHEME,
+					mtd->writesize, mtd->oobsize);
+		#else
 		err = omap_select_ecc_scheme(nand, OMAP_ECC_HAM1_CODE_SW,
 					mtd->writesize, mtd->oobsize);

Couldn't we just use the CONFIG_NAND_OMAP_ECCSCHEME instead of OMAP_ECC_HAM1_CODE_SW here and omit the CONFIG_BCH define?

On the other hand I think the whole function omap_nand_switch_ecc() is wrong. AFAIR it should only be used on omap3 aka am35xx/dm37xx devices witch the nandecc command.
These SoC do not have a ELM. Therefore I decided to say BCH8 on those devices is always 'HW ECC' when introduced BCH8 on omap3 first. Nowadays it is the so called BCH8_CODE_HW_DETECTION_SW. Coming from this mindset the right solution is to use some detection if the ELM is supported or not and switch in the HW part of omap_nand_switch_ecc() between OMAP_ECC_BCH8_CODE_HW and OMAP_ECC_BCH8_CODE_HW_DETECTION_SW.

Best regards

Andreas Bie?mann

author
#2 lore raw

Hi Andreas, for OMAP3 and AM35xx boards, it would have been ok omitting the CONFIG_BCH check and simply use CONFIG_NAND_OMAP_ECCSCHEME. Those boards use the ecc scheme config already. However I just wasn't 100% sure if I could rely on this config for all TI OMAP/AM based boards. I know OMAP3
and AM35xx board configs already have CONFIG_NAND_OMAP_ECCSCHEME.
Should I check for other OMAP and AM series? The original ecc detection
function
seems to be written with an assumption that config is nonexistent - hence defaulting
to HAM1.

That said, I agree that the whole omap_nand_switch_ecc() could be improved. However, to me, the confusion arised from the fact that OMAP3 can do BCH8 hw ECC
calculation but needs software to do the correction. Hence my patch changes the
software part of the omap_nand_switch_ecc(), and not the hardware part.

Just to clarify, what you are saying is that I should leave the software part as it was (defaulting
to HAM1), and in the hardware part I should check for ELM support and choose a BCH8
scheme accordingly, regardless of what's defined by CONFIG_NAND_OMAP_ECCSCHEME?

In other words, I will run 'nandecc hw' to enable BCH8?

Let me know,

Thanks,

Adam

Quoted 56 lines · Andreas Bie?mann

Hi Adam,

Quoted 1 line · Adam YH Lee

The ECC scheme selection algorithm in OMAP GPMC appears to be left

untested when

Quoted 1 line

BCH8 handling code was added. Running 'nandecc sw' defaults to HAM1 even

if

Quoted 1 line

the board is using another scheme (ex.

OMAP_ECC_BCH8_CODE_HW_DETECTION_SW on

Quoted 1 line

OMAP3). This results in unrecoverable ECC errors when reading data. This

commit

Quoted 1 line

fixes the behavior by checking for CONFIG_BCH and using the scheme

defined by

drivers/mtd/nand/omap_gpmc.c int __maybe_unused omap_nand_switch_ecc(uint32_t

Quoted 14 lines

CONFIG_NAND_OMAP_ECCSCHEME in the board configuration file.

This has been tested on Gumstix Overo (OMAP3).

---
 drivers/mtd/nand/omap_gpmc.c | 5 +++++
 1 file changed, 5 insertions(+)
drivers/mtd/nand/omap_gpmc.c +0 -0
index fc64f48..5daf932 100644
@@ -901,8 +901,13 @@ int __maybe_unused omap_nand_switch_ecc(uint32_t

hardware, uint32_t eccstrength)

drivers/mtd/nand/omap_gpmc.c int __maybe_unused omap_nand_switch_ecc(uint32_t hardware, uint32_t eccstrength)

Quoted 5 lines
drivers/mtd/nand/omap_gpmc.c +2 -0
                      return -EINVAL;
              }
      } else {
+             #ifdef CONFIG_BCH
+             err = omap_select_ecc_scheme(nand,

CONFIG_NAND_OMAP_ECCSCHEME,

drivers/mtd/nand/omap_gpmc.c int __maybe_unused omap_nand_switch_ecc(uint32_t hardware, uint32_t eccstrength) · line 909

Quoted 4 lines
drivers/mtd/nand/omap_gpmc.c +2 -0
+                                     mtd->writesize, mtd->oobsize);
+             #else
              err = omap_select_ecc_scheme(nand, OMAP_ECC_HAM1_CODE_SW,
                                      mtd->writesize, mtd->oobsize);

Couldn't we just use the CONFIG_NAND_OMAP_ECCSCHEME instead of OMAP_ECC_HAM1_CODE_SW here and omit the CONFIG_BCH define?

On the other hand I think the whole function omap_nand_switch_ecc() is wrong. AFAIR it should only be used on omap3 aka am35xx/dm37xx devices witch the nandecc command.
These SoC do not have a ELM. Therefore I decided to say BCH8 on those devices is always 'HW ECC' when introduced BCH8 on omap3 first. Nowadays it is the so called BCH8_CODE_HW_DETECTION_SW. Coming from this mindset the right solution is to use some detection if the ELM is supported or not and switch in the HW part of omap_nand_switch_ecc() between OMAP_ECC_BCH8_CODE_HW and OMAP_ECC_BCH8_CODE_HW_DETECTION_SW.

Best regards

Andreas Bie?mann

author
#3 lore raw

Had a conversation with Ash @ Gumstix and he pointed out relying on CONFIG_NAND_OMAP_ECCSCHEME could be dangerous as it could be anything other than the two SW ECC schemes available for OMAP3.

Also it looks like making a selection between OMAP_ECC_BCH8_CODE_HW and OMAP_ECC_BCH8_CODE_HW_DETECTION_SW may not make sense as the latter clearly requires CONFIG_BCH, which enables "software" library for BCH.

Quoted 98 lines · Adam Lee

Hi Andreas, for OMAP3 and AM35xx boards, it would have been ok omitting the
CONFIG_BCH check and simply use CONFIG_NAND_OMAP_ECCSCHEME. Those boards use the ecc scheme config already. However I just wasn't 100% sure if I could rely on this config for all TI OMAP/AM based boards. I know OMAP3
and AM35xx board configs already have CONFIG_NAND_OMAP_ECCSCHEME.
Should I check for other OMAP and AM series? The original ecc detection
function
seems to be written with an assumption that config is nonexistent - hence defaulting
to HAM1.

That said, I agree that the whole omap_nand_switch_ecc() could be improved. However, to me, the confusion arised from the fact that OMAP3 can do BCH8 hw ECC
calculation but needs software to do the correction. Hence my patch changes the
software part of the omap_nand_switch_ecc(), and not the hardware part.

Just to clarify, what you are saying is that I should leave the software part as it was (defaulting
to HAM1), and in the hardware part I should check for ELM support and choose a BCH8
scheme accordingly, regardless of what's defined by CONFIG_NAND_OMAP_ECCSCHEME?

In other words, I will run 'nandecc hw' to enable BCH8?

Let me know,

Thanks,

Adam

Quoted 56 lines · Andreas Bie?mann

Hi Adam,

Quoted 1 line · Adam YH Lee

The ECC scheme selection algorithm in OMAP GPMC appears to be left

untested when

Quoted 1 line

BCH8 handling code was added. Running 'nandecc sw' defaults to HAM1

even if

Quoted 1 line

the board is using another scheme (ex.

OMAP_ECC_BCH8_CODE_HW_DETECTION_SW on

Quoted 1 line

OMAP3). This results in unrecoverable ECC errors when reading data.

This commit

Quoted 1 line

fixes the behavior by checking for CONFIG_BCH and using the scheme

defined by

drivers/mtd/nand/omap_gpmc.c int __maybe_unused omap_nand_switch_ecc(uint32_t

Quoted 14 lines

CONFIG_NAND_OMAP_ECCSCHEME in the board configuration file.

This has been tested on Gumstix Overo (OMAP3).

---
 drivers/mtd/nand/omap_gpmc.c | 5 +++++
 1 file changed, 5 insertions(+)
drivers/mtd/nand/omap_gpmc.c +0 -0
index fc64f48..5daf932 100644
@@ -901,8 +901,13 @@ int __maybe_unused omap_nand_switch_ecc(uint32_t

hardware, uint32_t eccstrength)

drivers/mtd/nand/omap_gpmc.c int __maybe_unused omap_nand_switch_ecc(uint32_t hardware, uint32_t eccstrength)

Quoted 5 lines
drivers/mtd/nand/omap_gpmc.c +2 -0
                      return -EINVAL;
              }
      } else {
+             #ifdef CONFIG_BCH
+             err = omap_select_ecc_scheme(nand,

CONFIG_NAND_OMAP_ECCSCHEME,

drivers/mtd/nand/omap_gpmc.c int __maybe_unused omap_nand_switch_ecc(uint32_t hardware, uint32_t eccstrength) · line 909

Quoted 4 lines
drivers/mtd/nand/omap_gpmc.c +2 -0
+                                     mtd->writesize, mtd->oobsize);
+             #else
              err = omap_select_ecc_scheme(nand, OMAP_ECC_HAM1_CODE_SW,
                                      mtd->writesize, mtd->oobsize);

Couldn't we just use the CONFIG_NAND_OMAP_ECCSCHEME instead of OMAP_ECC_HAM1_CODE_SW here and omit the CONFIG_BCH define?

On the other hand I think the whole function omap_nand_switch_ecc() is wrong. AFAIR it should only be used on omap3 aka am35xx/dm37xx devices witch the nandecc command.
These SoC do not have a ELM. Therefore I decided to say BCH8 on those devices is always 'HW ECC' when introduced BCH8 on omap3 first. Nowadays it is the so called BCH8_CODE_HW_DETECTION_SW. Coming from this mindset the right solution is to use some detection if the ELM is supported or not and switch in the HW part of omap_nand_switch_ecc() between OMAP_ECC_BCH8_CODE_HW and OMAP_ECC_BCH8_CODE_HW_DETECTION_SW.

Best regards

Andreas Bie?mann

#4 lore raw

PATCH 1/1 omap: gpmc: 'nandecc sw' can use HAM1 or BCH8

In reply to Andreas Bießmann

The 'nandecc sw' command selects a software-based error correction algorithm. By default, this is OMAP_ECC_HAM1_CODE_SW but some platforms use OMAP_ECC_BCH8_CODE_HW_DETECTION_SW as their software-based correction algorithm. Allow a user to be specific e.g.

 # nandecc sw <hamming|bch8>
where 'hamming' is still the default.

Note: we don't just use CONFIG_NAND_OMAP_ECCSCHEME as it might be set
to a hardware-based ECC scheme---a little strange when the user has requested 'sw' ECC.

---
 arch/arm/cpu/armv7/omap3/board.c | 11 ++++++++++-
 drivers/mtd/nand/omap_gpmc.c     | 12 +++++++++++-
 2 files changed, 21 insertions(+), 2 deletions(-)
arch/arm/cpu/armv7/omap3/board.c +10 -1
index 667e77f..b4e29ab 100644
@@ -327,7 +327,16 @@ static int do_switch_ecc(cmd_tbl_t * cmdtp, int flag, int argc, char * const arg
 				goto usage;
 		}
 	} else if (strncmp(argv[1], "sw", 2) == 0) {
-		omap_nand_switch_ecc(0, 0);
+		if (argc == 2) {
+			omap_nand_switch_ecc(0, 1);
+		} else {
+			if (strncmp(argv[2], "hamming", 7) == 0)
+				omap_nand_switch_ecc(0, 1);
+			else if (strncmp(argv[2], "bch8", 4) == 0)
+				omap_nand_switch_ecc(0, 8);
+			else
+				goto usage;
+		}
 	} else {
 		goto usage;
 	}
drivers/mtd/nand/omap_gpmc.c +11 -1
index db1599e..d6a28e6 100644
@@ -794,8 +794,18 @@ int __maybe_unused omap_nand_switch_ecc(uint32_t hardware, uint32_t eccstrength)
 			return -EINVAL;
 		}
 	} else {
-		err = omap_select_ecc_scheme(nand, OMAP_ECC_HAM1_CODE_SW,
+		if (eccstrength == 1) {
+			err = omap_select_ecc_scheme(nand,
+					OMAP_ECC_HAM1_CODE_SW,
+					mtd->writesize, mtd->oobsize);
+		} else if (eccstrength == 8) {
+			err = omap_select_ecc_scheme(nand,
+					OMAP_ECC_BCH8_CODE_HW_DETECTION_SW,
 					mtd->writesize, mtd->oobsize);
+		} else {
+			printf("nand: error: unsupported ECC scheme\n");
+			return -EINVAL;
+		}
 	}
 
 	/* Update NAND handling after ECC mode switch */
Signature · 1 line
2.1.0
#5 lore raw

[U-Boot] omap: gpmc: 'nandecc sw' can use HAM1 or BCH8

Quoted 4 lines · Ash Charles

The 'nandecc sw' command selects a software-based error correction algorithm. By default, this is OMAP_ECC_HAM1_CODE_SW but some platforms use OMAP_ECC_BCH8_CODE_HW_DETECTION_SW as their software-based correction algorithm. Allow a user to be specific e.g.

Quoted 8 lines
 # nandecc sw <hamming|bch8>
where 'hamming' is still the default.

Note: we don't just use CONFIG_NAND_OMAP_ECCSCHEME as it might be set
to a hardware-based ECC scheme---a little strange when the user has requested 'sw' ECC.

Applied to u-boot/master, thanks!

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20150306/b5e2f10d/attachment.sig>