Skip to contentSkip to conversation

fwu: Mark failed bank as invalid during rollback

PATCH Superseded Kory Maincent 2 participants 5 messages Jan 14 – Jan 15

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fwu: Mark failed bank as invalid during rollback

When boot_idx differs from active_idx at boot time, it indicates a rollback scenario where the firmware update failed and the system reverted to the previous working bank. In this case, mark the failed bank (active_idx) as invalid to prevent future boot attempts from that bank.

This change adds a new fwu_invalid_bank() function that sets the bank state to FWU_BANK_INVALID and syncs the metadata. This is only supported in FWU metadata version 2; version 1 provides a no-op stub.

Without this change, a failed bank would remain in a valid state and could potentially be selected for boot again, leading to repeated boot failures.

---
 include/fwu.h            |  8 ++++++++
 lib/fwu_updates/fwu.c    |  1 +
 lib/fwu_updates/fwu_v1.c | 10 ++++++++++
 lib/fwu_updates/fwu_v2.c | 19 +++++++++++++++++++
 4 files changed, 38 insertions(+)
include/fwu.h +8 -0
index e7bd1d492af..e4f3a062e36 100644
@@ -415,6 +415,14 @@ int fwu_get_mdata_size(uint32_t *mdata_size);
  */
 int fwu_state_machine_updates(bool trial_state, uint32_t update_index);
 
+/**
+ * fwu_invalid_bank() - Invalid bank state of a specific bank
+ * @index: Index of the bank
+ *
+ * Only supported in version 2 of FWU metadata.
+ */
+void fwu_invalid_bank(uint32_t index);
+
 /**
  * fwu_init() - FWU specific initialisations
  *
lib/fwu_updates/fwu.c +1 -0
index 0f5ef2ba515..e54c19b0c64 100644
@@ -766,6 +766,7 @@ static int fwu_boottime_checks(void)
 	if (boot_idx != active_idx) {
 		log_info("Boot idx %u is not matching active idx %u, changing active_idx\n",
 			 boot_idx, active_idx);
+		fwu_invalid_bank(active_idx);
 		ret = fwu_set_active_index(boot_idx);
 		if (!ret)
 			boottime_check = 1;
lib/fwu_updates/fwu_v1.c +10 -0
index 974abf216f6..c9ac1ec244d 100644
@@ -136,6 +136,16 @@ int fwu_get_mdata_size(uint32_t *mdata_size)
 	return 0;
 }
 
+/**
+ * fwu_invalid_bank() - Invalid bank state of a specific bank
+ * @index: Index of the bank
+ *
+ * Not supported.
+ */
+void fwu_invalid_bank(uint32_t index)
+{
+}
+
 /**
  * fwu_init() - FWU specific initialisations
  *
lib/fwu_updates/fwu_v2.c +19 -0
index 159315b45b9..3c07adfd4e0 100644
@@ -194,6 +194,25 @@ int fwu_state_machine_updates(bool trial_state, uint32_t update_index)
 		fwu_bank_state_update(0, update_index);
 }
 
+/**
+ * fwu_invalid_bank() - Invalid bank state of a specific bank
+ * @index: Index of the bank
+ */
+void fwu_invalid_bank(uint32_t index)
+{
+	struct fwu_data *data = fwu_get_data();
+	struct fwu_mdata *mdata;
+	int ret;
+
+	mdata = data->fwu_mdata;
+	data->bank_state[index] = FWU_BANK_INVALID;
+	mdata->bank_state[index] = FWU_BANK_INVALID;
+
+	ret = fwu_sync_mdata(mdata, BOTH_PARTS);
+	if (ret)
+		log_err("Unable to invalid bank %u\n", index);
+}
+
 /**
  * fwu_get_mdata_size() - Get the FWU metadata size
  * @mdata_size: Size of the metadata structure
Signature · 1 line
2.43.0
#1 lore raw

Hi Kory,

Quoted 7 lines · Kory Maincent

When boot_idx differs from active_idx at boot time, it indicates a rollback scenario where the firmware update failed and the system reverted to the previous working bank. In this case, mark the failed bank (active_idx) as invalid to prevent future boot attempts from that bank.

Quoted 7 lines

This change adds a new fwu_invalid_bank() function that sets the bank state to FWU_BANK_INVALID and syncs the metadata. This is only supported in FWU metadata version 2; version 1 provides a no-op stub.

Without this change, a failed bank would remain in a valid state and could potentially be selected for boot again, leading to repeated boot failures.

How? Doesn't the code already switch the metadata to point to the 'boot_idx' bank as active?
What could happen here I guess, is the selected active bank to go bad somehow and the firmware would try the secondary bank, which contains an invalid firmware. But in that case the board will fail to boot regardless. Is there a corner case I am missing?

Thanks
/Ilias

Quoted 94 lines
---
 include/fwu.h            |  8 ++++++++
 lib/fwu_updates/fwu.c    |  1 +
 lib/fwu_updates/fwu_v1.c | 10 ++++++++++
 lib/fwu_updates/fwu_v2.c | 19 +++++++++++++++++++
 4 files changed, 38 insertions(+)
include/fwu.h +8 -0
index e7bd1d492af..e4f3a062e36 100644
@@ -415,6 +415,14 @@ int fwu_get_mdata_size(uint32_t *mdata_size);
  */
 int fwu_state_machine_updates(bool trial_state, uint32_t update_index);
 
+/**
+ * fwu_invalid_bank() - Invalid bank state of a specific bank
+ * @index: Index of the bank
+ *
+ * Only supported in version 2 of FWU metadata.
+ */
+void fwu_invalid_bank(uint32_t index);
+
 /**
  * fwu_init() - FWU specific initialisations
  *
lib/fwu_updates/fwu.c +1 -0
index 0f5ef2ba515..e54c19b0c64 100644
@@ -766,6 +766,7 @@ static int fwu_boottime_checks(void)
        if (boot_idx != active_idx) {
                log_info("Boot idx %u is not matching active idx %u, changing active_idx\n",
                         boot_idx, active_idx);
+               fwu_invalid_bank(active_idx);
                ret = fwu_set_active_index(boot_idx);
                if (!ret)
                        boottime_check = 1;
lib/fwu_updates/fwu_v1.c +10 -0
index 974abf216f6..c9ac1ec244d 100644
@@ -136,6 +136,16 @@ int fwu_get_mdata_size(uint32_t *mdata_size)
        return 0;
 }
 
+/**
+ * fwu_invalid_bank() - Invalid bank state of a specific bank
+ * @index: Index of the bank
+ *
+ * Not supported.
+ */
+void fwu_invalid_bank(uint32_t index)
+{
+}
+
 /**
  * fwu_init() - FWU specific initialisations
  *
lib/fwu_updates/fwu_v2.c +19 -1
index 159315b45b9..3c07adfd4e0 100644
@@ -194,6 +194,25 @@ int fwu_state_machine_updates(bool trial_state, uint32_t update_index)
                fwu_bank_state_update(0, update_index);
 }
 
+/**
+ * fwu_invalid_bank() - Invalid bank state of a specific bank
+ * @index: Index of the bank
+ */
+void fwu_invalid_bank(uint32_t index)
+{
+       struct fwu_data *data = fwu_get_data();
+       struct fwu_mdata *mdata;
+       int ret;
+
+       mdata = data->fwu_mdata;
+       data->bank_state[index] = FWU_BANK_INVALID;
+       mdata->bank_state[index] = FWU_BANK_INVALID;
+
+       ret = fwu_sync_mdata(mdata, BOTH_PARTS);
+       if (ret)
+               log_err("Unable to invalid bank %u\n", index);
+}
+
 /**
  * fwu_get_mdata_size() - Get the FWU metadata size
  * @mdata_size: Size of the metadata structure
--

2.43.0

#2 lore raw
Quoted 18 lines · On Wed, 14 Jan 2026 15:54:07 +0200 Ilias Apalodimas…

Hi Kory,

Quoted 7 lines · Kory Maincent

When boot_idx differs from active_idx at boot time, it indicates a rollback scenario where the firmware update failed and the system reverted to the previous working bank. In this case, mark the failed bank (active_idx) as invalid to prevent future boot attempts from that bank.

Quoted 7 lines

This change adds a new fwu_invalid_bank() function that sets the bank state to FWU_BANK_INVALID and syncs the metadata. This is only supported in FWU metadata version 2; version 1 provides a no-op stub.

Without this change, a failed bank would remain in a valid state and could potentially be selected for boot again, leading to repeated boot failures.

Quoted 6 lines

How? Doesn't the code already switch the metadata to point to the 'boot_idx' bank as active?
What could happen here I guess, is the selected active bank to go bad somehow and the firmware would try the secondary bank, which contains an invalid firmware. But in that case the board will fail to boot regardless. Is there a corner case I am missing?

I think I have faced a corner case, but I can't recall what it was. :/ Anyway we should change the bank state to invalid if we don't manage to boot one bank right. We should provide to the user the information that one bank is in an invalid state if that is the case. What do you think?

Regards,

Signature · 3 lines
Köry Maincent, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com
#3 lore raw

Hi Kory,

Quoted 22 lines · Kory Maincent
Quoted 18 lines · On Wed, 14 Jan 2026 15:54:07 +0200 Ilias Apalodimas…

Hi Kory,

Quoted 14 lines · Kory Maincent

When boot_idx differs from active_idx at boot time, it indicates a rollback scenario where the firmware update failed and the system reverted to the previous working bank. In this case, mark the failed bank (active_idx) as invalid to prevent future boot attempts from that bank.

This change adds a new fwu_invalid_bank() function that sets the bank state to FWU_BANK_INVALID and syncs the metadata. This is only supported in FWU metadata version 2; version 1 provides a no-op stub.

Without this change, a failed bank would remain in a valid state and could potentially be selected for boot again, leading to repeated boot failures.

Quoted 8 lines
Quoted 6 lines

How? Doesn't the code already switch the metadata to point to the 'boot_idx' bank as active?
What could happen here I guess, is the selected active bank to go bad somehow and the firmware would try the secondary bank, which contains an invalid firmware. But in that case the board will fail to boot regardless. Is there a corner case I am missing?

I think I have faced a corner case, but I can't recall what it was. :/

I think this patch has value if you have 3 banks. In that case, you won't have to try and boot the invalid one, before booting a correct one.

Quoted 3 lines

Anyway we should change the bank state to invalid if we don't manage to boot one bank right. We should provide to the user the information that one bank is in an invalid state if that is the case. What do you think?

Yes. The invalid state is already described in the spec so I think this is fine. However, instead of adding a new function to invalidate the state, can't we tweak fwu_bank_state_update() instead?
Instead of a boolean for trial state, pass an the actual state and update it while keeping the same checks.

Regards
/Ilias

Quoted 6 lines

Regards,
--
Köry Maincent, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com

#4 lore raw
Quoted 33 lines · On Thu, 15 Jan 2026 11:36:29 +0200 Ilias Apalodimas…

Hi Kory,

Quoted 30 lines · Kory Maincent
Quoted 25 lines · On Wed, 14 Jan 2026 15:54:07 +0200 Ilias Apalodimas…

Hi Kory,

Quoted 7 lines · Kory Maincent

When boot_idx differs from active_idx at boot time, it indicates a rollback scenario where the firmware update failed and the system reverted to the previous working bank. In this case, mark the failed bank (active_idx) as invalid to prevent future boot attempts from that bank.

Quoted 7 lines

This change adds a new fwu_invalid_bank() function that sets the bank state to FWU_BANK_INVALID and syncs the metadata. This is only supported in FWU metadata version 2; version 1 provides a no-op stub.

Without this change, a failed bank would remain in a valid state and could potentially be selected for boot again, leading to repeated boot failures.

How? Doesn't the code already switch the metadata to point to the 'boot_idx' bank as active?
What could happen here I guess, is the selected active bank to go bad somehow and the firmware would try the secondary bank, which contains an invalid firmware. But in that case the board will fail to boot regardless. Is there a corner case I am missing?

Quoted 5 lines
Quoted 1 line

I think I have faced a corner case, but I can't recall what it was. :/

I think this patch has value if you have 3 banks. In that case, you won't have to try and boot the invalid one, before booting a correct one.

Good point.

Quoted 9 lines
Quoted 3 lines

Anyway we should change the bank state to invalid if we don't manage to boot one bank right. We should provide to the user the information that one bank is in an invalid state if that is the case. What do you think?

Yes. The invalid state is already described in the spec so I think this is fine. However, instead of adding a new function to invalidate the state, can't we tweak fwu_bank_state_update() instead?
Instead of a boolean for trial state, pass an the actual state and update it while keeping the same checks.

Indeed, we could do that. I will prepare a V2 accordingly.

Regards,

Signature · 3 lines
Köry Maincent, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com