Skip to contentSkip to conversation

mtd: core: Handle unsupported OTP operations

PATCH Superseded Aapo Vienamo 2 participants 9 messages Mar 7, 2024 – Mar 13, 2024 Memory technology devices (MTD)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover mtd: core: Handle unsupported OTP operations

Make MTD core tolerate SPI controllers without OTP support and report an error from MTD core if OTP initialization fails early.

These changes address the issue that occurs when an OTP capable SPI NOR device is initialized with the Intel SPI controller. The limited supported opcode set of the SPI controller leads to failure in the OTP initialization, which in turn fails the probe for the MTD device. By allowing the MTD core to tolerate unsupported OTP, the rest of the MTD functionality remains intact even if OTP initialization is unsupported.

Aapo Vienamo (2):
mtd: core: Report error if first mtd_otp_size() call fails in

    mtd_otp_nvmem_add()

mtd: core: Don't fail mtd_device_parse_register() if OTP is
unsupported

 drivers/mtd/mtdcore.c | 14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)
Signature · 1 line
2.41.0
#1 lore raw

PATCH 1/2 mtd: core: Report error if first mtd_otp_size() call fails in mtd_otp_nvmem_add()

Jump to the error reporting code in mtd_otp_nvmem_add() if the mtd_otp_size() call fails. Without this fix, the error is not logged.

Fixes 4b361cfa8624 ("mtd: core: add OTP nvmem provider support")
---
 drivers/mtd/mtdcore.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/mtd/mtdcore.c +4 -2
index 5887feb347a4..c365c97e7232 100644
@@ -956,8 +956,10 @@ static int mtd_otp_nvmem_add(struct mtd_info *mtd)
 
 	if (mtd->_get_user_prot_info && mtd->_read_user_prot_reg) {
 		size = mtd_otp_size(mtd, true);
-		if (size < 0)
-			return size;
+		if (size < 0) {
+			err = size;
+			goto err;
+		}
 
 		if (size > 0) {
 			nvmem = mtd_otp_nvmem_register(mtd, "user-otp", size,
Signature · 1 line
2.41.0
Reviewed-by on v1 1/2
#2 lore raw

Re: [PATCH 1/2] mtd: core: Report error if first mtd_otp_size() call fails in mtd_otp_nvmem_add()

Quoted 6 lines · Aapo Vienamo

Jump to the error reporting code in mtd_otp_nvmem_add() if the mtd_otp_size() call fails. Without this fix, the error is not logged.

Fixes 4b361cfa8624 ("mtd: core: add OTP nvmem provider support")

Not sure this qualifies as a Fixes patch, I'll leave that to the MTD mainterainer.

Anyways,

-michael

#3 lore raw

PATCH 2/2 mtd: core: Don't fail mtd_device_parse_register() if OTP is unsupported

In reply to Aapo Vienamo

Handle the case where -EOPNOTSUPP is returned from OTP driver.

This addresses an issue that occurs with the Intel SPI flash controller, which has a limited supported opcode set. Whilst the OTP functionality is not available due to this restriction, other parts of the MTD functionality of the device are intact. This change allows the driver to gracefully handle the restriction by allowing the supported functionality to remain available instead of failing the probe altogether.

---
 drivers/mtd/mtdcore.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/mtd/mtdcore.c +7 -1
index c365c97e7232..1cfc8bb5187d 100644
@@ -1054,8 +1054,14 @@ int mtd_device_parse_register(struct mtd_info *mtd, const char * const *types,
 
 	mtd_set_dev_defaults(mtd);
 
+	/*
+	 * Don't abort MTD init if OTP functionality is unsupported. The
+	 * cleanup of the OTP init is contained within mtd_otp_nvmem_add().
+	 * Omitting goto out here is safe since the cleanup code there
+	 * should be no-ops.
+	 */
 	ret = mtd_otp_nvmem_add(mtd);
-	if (ret)
+	if (ret && ret != -EOPNOTSUPP)
 		goto out;
 
 	if (IS_ENABLED(CONFIG_MTD_PARTITIONED_MASTER)) {
Signature · 1 line
2.41.0
#4 lore raw

Re: [PATCH 2/2] mtd: core: Don't fail mtd_device_parse_register() if OTP is unsupported

Quoted 22 lines · Aapo Vienamo

Handle the case where -EOPNOTSUPP is returned from OTP driver.

This addresses an issue that occurs with the Intel SPI flash controller, which has a limited supported opcode set. Whilst the OTP functionality is not available due to this restriction, other parts of the MTD functionality of the device are intact. This change allows the driver to gracefully handle the restriction by allowing the supported functionality to remain available instead of failing the probe altogether.

---
 drivers/mtd/mtdcore.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/mtd/mtdcore.c +0 -0
index c365c97e7232..1cfc8bb5187d 100644

drivers/mtd/mtdcore.c int mtd_device_parse_register(struct mtd_info *mtd, const char * const *types, · line 1062

Quoted 9 lines
drivers/mtd/mtdcore.c +6 -0
@@ -1054,8 +1054,14 @@ int mtd_device_parse_register(struct mtd_info *mtd, const char * const *types,
 
 	mtd_set_dev_defaults(mtd);
 
+	/*
+	 * Don't abort MTD init if OTP functionality is unsupported. The
+	 * cleanup of the OTP init is contained within mtd_otp_nvmem_add().
+	 * Omitting goto out here is safe since the cleanup code there
+	 * should be no-ops.
+	 */

Only if that's true for both the factory and user OTP area. Also, you'll print an error message for EOPNOTSUPP, although that is not really an error. Is that intended?

drivers/mtd/mtdcore.c int mtd_device_parse_register(struct mtd_info *mtd, const char * const *types, · line 1064

Quoted 3 lines
drivers/mtd/mtdcore.c +1 -1
 	ret = mtd_otp_nvmem_add(mtd);
-	if (ret)
+	if (ret && ret != -EOPNOTSUPP)

Maybe there is a better way to handle this, like controller capabilities instead of putting these EOPNOTSUPP checks everywhere? I'm not sure.

-michael

Quoted 3 lines
drivers/mtd/mtdcore.c +0 -0
 		goto out;
 
 	if (IS_ENABLED(CONFIG_MTD_PARTITIONED_MASTER)) {
#5 lore raw

Re: [PATCH 2/2] mtd: core: Don't fail mtd_device_parse_register() if OTP is unsupported

Hi Michael

Quoted 10 lines · Michael Walle

drivers/mtd/mtdcore.c int mtd_device_parse_register(struct mtd_info *mtd, const char * const *types, · line 1062

Quoted 7 lines · Aapo Vienamo

Handle the case where -EOPNOTSUPP is returned from OTP driver.

drivers/mtd/mtdcore.c +6 -0
+	/*
+	 * Don't abort MTD init if OTP functionality is unsupported. The
+	 * cleanup of the OTP init is contained within mtd_otp_nvmem_add().
+	 * Omitting goto out here is safe since the cleanup code there
+	 * should be no-ops.
+	 */

Only if that's true for both the factory and user OTP area.

I'm not sure I follow. I'm not seeing a path in mtd_otp_nvmem_add() that would result in either mtd->otp_user_nvmem or mtd->otp_factor_nvmem needing to be cleaned up by the caller, if an error is returned, if that's what you are referring to.

Quoted 2 lines

Also, you'll print an error message for EOPNOTSUPP, although that is not really an error. Is that intended?

Well, when we hit this, the functionality of the SPI memory itself is degraded in the sense that the OTP functionality is not available. What would you suggest?

Quoted 8 lines

drivers/mtd/mtdcore.c int mtd_device_parse_register(struct mtd_info *mtd, const char * const *types, · line 1064

Quoted 4 lines
drivers/mtd/mtdcore.c +1 -1
 	ret = mtd_otp_nvmem_add(mtd);
-	if (ret)
+	if (ret && ret != -EOPNOTSUPP)

Maybe there is a better way to handle this, like controller capabilities instead of putting these EOPNOTSUPP checks everywhere? I'm not sure.

Trying to come up with clear semantics for a capabilities flag to solve this is difficult. The issue is that on the SPI controller side, the limitation stems from the really strict set of opcodes that are allowed. For example, we already hit an error with the 0x35 (read configuration register) not being on the set of allowed opcodes. While this instruction is used by the OTP code, it's not a strictly OTP specific operation.

If there was a flag that would signal OTP support, I think it would have be defined as "the controller supports all operations that are performed by the OTP code", which sounds brittle. The other way around would be to have a really fine-grained set of flags that the MTD core would check, but that feels tedious and error prone as well.

Best,
Aapo

#6 lore raw

Re: [PATCH 2/2] mtd: core: Don't fail mtd_device_parse_register() if OTP is unsupported

Hi,

Quoted 12 lines · Aapo Vienamo
Quoted 10 lines · Michael Walle

drivers/mtd/mtdcore.c int mtd_device_parse_register(struct mtd_info *mtd, const char * const *types, · line 1062

Quoted 7 lines · Aapo Vienamo

Handle the case where -EOPNOTSUPP is returned from OTP driver.

drivers/mtd/mtdcore.c +6 -0
+	/*
+	 * Don't abort MTD init if OTP functionality is unsupported. The
+	 * cleanup of the OTP init is contained within mtd_otp_nvmem_add().
+	 * Omitting goto out here is safe since the cleanup code there
+	 * should be no-ops.
+	 */

Only if that's true for both the factory and user OTP area.

Quoted 4 lines

I'm not sure I follow. I'm not seeing a path in mtd_otp_nvmem_add() that would result in either mtd->otp_user_nvmem or mtd->otp_factor_nvmem needing to be cleaned up by the caller, if an error is returned, if that's what you are referring to.

Yes you're right, sorry for the noise.

Quoted 7 lines
Quoted 3 lines

Also, you'll print an error message for EOPNOTSUPP, although that is not really an error. Is that intended?

Well, when we hit this, the functionality of the SPI memory itself is degraded in the sense that the OTP functionality is not available. What would you suggest?

But it's not really an error, I mean, we are ignoring that one on purpose now :) I'd just guard it with "if (ret != -EOPNOTSUPP)".

Quoted 9 lines
Quoted 8 lines

drivers/mtd/mtdcore.c int mtd_device_parse_register(struct mtd_info *mtd, const char * const *types, · line 1064

Quoted 4 lines
drivers/mtd/mtdcore.c +1 -1
 	ret = mtd_otp_nvmem_add(mtd);
-	if (ret)
+	if (ret && ret != -EOPNOTSUPP)

Maybe there is a better way to handle this, like controller capabilities instead of putting these EOPNOTSUPP checks everywhere? I'm not sure.

Quoted 7 lines

Trying to come up with clear semantics for a capabilities flag to solve this is difficult. The issue is that on the SPI controller side, the limitation stems from the really strict set of opcodes that are allowed. For example, we already hit an error with the 0x35 (read configuration register) not being on the set of allowed opcodes. While this instruction is used by the OTP code, it's not a strictly OTP specific operation.

I see. It's just that due to this (very) restricted SPI contoller all this EOPNOTSUPP handling is creeping into more an more places in spi-nor core and now mtdcore :)

Anyway, I don't have any better idea right now. So I think this is fine.

-michael

Quoted 5 lines

If there was a flag that would signal OTP support, I think it would have be defined as "the controller supports all operations that are performed by the OTP code", which sounds brittle. The other way around would be to have a really fine-grained set of flags that the MTD core would check, but that feels tedious and error prone as well.

#7 lore raw

Re: [PATCH 2/2] mtd: core: Don't fail mtd_device_parse_register() if OTP is unsupported

Quoted 11 lines · Michael Walle
Quoted 7 lines · Aapo Vienamo
Quoted 2 lines · Michael Walle

Also, you'll print an error message for EOPNOTSUPP, although that is not really an error. Is that intended?

Well, when we hit this, the functionality of the SPI memory itself is degraded in the sense that the OTP functionality is not available. What would you suggest?

But it's not really an error, I mean, we are ignoring that one on purpose now :) I'd just guard it with "if (ret != -EOPNOTSUPP)".

To clarify, are you suggesting a modification like this to the code at the end of mtd_otp_nvmem_add()?

err:

	nvmem_unregister(...);
	if (ret != EOPNOTSUPP)
		return dev_err_probe(...);
	return 0;

Best,
Aapo

#8 lore raw

Re: [PATCH 2/2] mtd: core: Don't fail mtd_device_parse_register() if OTP is unsupported

Hi,

Quoted 13 lines · Aapo Vienamo
Quoted 11 lines · Michael Walle
Quoted 7 lines · Aapo Vienamo
Quoted 2 lines · Michael Walle

Also, you'll print an error message for EOPNOTSUPP, although that is not really an error. Is that intended?

Well, when we hit this, the functionality of the SPI memory itself is degraded in the sense that the OTP functionality is not available. What would you suggest?

But it's not really an error, I mean, we are ignoring that one on purpose now :) I'd just guard it with "if (ret != -EOPNOTSUPP)".

Quoted 8 lines

To clarify, are you suggesting a modification like this to the code at the end of mtd_otp_nvmem_add()?

err:

	nvmem_unregister(...);
	if (ret != EOPNOTSUPP)
		return dev_err_probe(...);
	return 0;

Yes, either this variant, then you don't need to fix the caller or return EOPNOTSUPP but just don't print the message.

-michael