Skip to contentSkip to conversation

net: atlantic: fix overwritten return value in Aquantia driver

PATCH In review Tian 4 participants 5 messages Jul 29, 2025

Subsystems (3)
lore
Share & tools
lore

Tian

author
#0 lore raw

PATCH v2 1/1 net: atlantic: fix overwritten return value in Aquantia driver

From: Tian Liu <[email protected]>

In hw_atl_utils.c and hw_atl_utils_fw2x.c, a return value is first assigned to err, but then later calls overwrite it unconditionally. As a result, any earlier failure is lost if the later call succeeds. This may cause the driver to proceed when it should have stopped.

This patch uses err |= instead of assignment, so that any earlier error is preserved even if later calls succeed. This is safe because all involved functions return standard negative error codes (e.g. -EIO, -ETIMEDOUT, etc). OR-ing such values does not convert them into success, and preserves the indication that an error occurred.

Fixes 6a7f2277313b4 ("net: aquantia: replace AQ_HW_WAIT_FOR with readx_poll_timeout_atomic")

Changes in v2:
- Add full name in Signed-off-by
- Update subject to include [PATCH net v2]
- Add Fixes: tag
- Expand commit message to explain why OR-ing errors is safe

---
 drivers/net/ethernet/aquantia/atlantic/hw_atl/hw_atl_utils.c  | 2 +-
 .../net/ethernet/aquantia/atlantic/hw_atl/hw_atl_utils_fw2x.c | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)
drivers/net/ethernet/aquantia/atlantic/hw_atl/hw_atl_utils.c +1 -1
index 7e88d7234b14..372f30e296ec 100644
@@ -492,7 +492,7 @@ static int hw_atl_utils_init_ucp(struct aq_hw_s *self,
 					self, self->mbox_addr,
 					self->mbox_addr != 0U,
 					1000U, 10000U);
-	err = readx_poll_timeout_atomic(aq_fw1x_rpc_get, self,
+	err |= readx_poll_timeout_atomic(aq_fw1x_rpc_get, self,
 					self->rpc_addr,
 					self->rpc_addr != 0U,
 					1000U, 100000U);
drivers/net/ethernet/aquantia/atlantic/hw_atl/hw_atl_utils_fw2x.c +2 -2
index 4d4cfbc91e19..45b7720fd49e 100644
@@ -102,12 +102,12 @@ static int aq_fw2x_init(struct aq_hw_s *self)
 					self->mbox_addr != 0U,
 					1000U, 10000U);
 
-	err = readx_poll_timeout_atomic(aq_fw2x_rpc_get,
+	err |= readx_poll_timeout_atomic(aq_fw2x_rpc_get,
 					self, self->rpc_addr,
 					self->rpc_addr != 0U,
 					1000U, 100000U);
 
-	err = aq_fw2x_settings_get(self, &self->settings_addr);
+	err |= aq_fw2x_settings_get(self, &self->settings_addr);
 
 	return err;
 }
Signature · 1 line
2.39.5 (Apple Git-154)
#1 lore raw
Quoted 5 lines

This patch uses err |= instead of assignment, so that any earlier error is preserved even if later calls succeed. This is safe because all involved functions return standard negative error codes (e.g. -EIO, -ETIMEDOUT, etc). OR-ing such values does not convert them into success, and preserves the indication that an error occurred.

22 | 110 = 132 = ERFKILL

How easy will it be for somebody to debug that?

Andrew

reviewer
#2 lore raw
Quoted 9 lines · Andrew Lunn
Quoted 5 lines

This patch uses err |= instead of assignment, so that any earlier error is preserved even if later calls succeed. This is safe because all involved functions return standard negative error codes (e.g. -EIO, -ETIMEDOUT, etc). OR-ing such values does not convert them into success, and preserves the indication that an error occurred.

22 | 110 = 132 = ERFKILL

How easy will it be for somebody to debug that?

Hi Andrew and Tian,

I agree that this does not seem to be a good approach.

Looking over the code, it seems that we can just bail out if an error occurs. Returning error codes as they arise.

Does this seem reasonable to you?

drivers/net/ethernet/aquantia/atlantic/hw_atl/hw_atl_utils.c +3 -0
index 7e88d7234b14..6c490f61ce5c 100644
@@ -492,6 +492,9 @@ static int hw_atl_utils_init_ucp(struct aq_hw_s *self,
 					self, self->mbox_addr,
 					self->mbox_addr != 0U,
 					1000U, 10000U);
+	if (err)
+		return err;
+
 	err = readx_poll_timeout_atomic(aq_fw1x_rpc_get, self,
 					self->rpc_addr,
 					self->rpc_addr != 0U,
drivers/net/ethernet/aquantia/atlantic/hw_atl/hw_atl_utils_fw2x.c +4 -0
index 4d4cfbc91e19..48f7c863c04b 100644
@@ -101,11 +101,15 @@ static int aq_fw2x_init(struct aq_hw_s *self)
 					self, self->mbox_addr,
 					self->mbox_addr != 0U,
 					1000U, 10000U);
+	if (err)
+		return err;
 
 	err = readx_poll_timeout_atomic(aq_fw2x_rpc_get,
 					self, self->rpc_addr,
 					self->rpc_addr != 0U,
 					1000U, 100000U);
+	if (err)
+		return err;
 
 	err = aq_fw2x_settings_get(self, &self->settings_addr);
 
reviewer
#3 lore raw
In reply to Tian
Quoted 7 lines · Tian

From: Tian Liu <[email protected]>

In hw_atl_utils.c and hw_atl_utils_fw2x.c, a return value is first assigned to err, but then later calls overwrite it unconditionally. As a result, any earlier failure is lost if the later call succeeds. This may cause the driver to proceed when it should have stopped.

Quoted 7 lines

This patch uses err |= instead of assignment, so that any earlier error is preserved even if later calls succeed. This is safe because all involved functions return standard negative error codes (e.g. -EIO, -ETIMEDOUT, etc). OR-ing such values does not convert them into success, and preserves the indication that an error occurred.

Fixes 6a7f2277313b4 ("net: aquantia: replace AQ_HW_WAIT_FOR with readx_poll_timeout_atomic")

nit: No blank line here

I do agree that the Fixes tag is correct in the case of hw_atl_utils_fw2x.c

But, in the case of hw_atl_utils.c, it seems to me that the correct Fixes tag would be as follows, because prior to this commit the value of err was not overwritten:

Fixes e7b5f97e6574 ("net: atlantic: check rpc result and wait for rpc address")

Given the different commits that are being fixed I would suggest splitting the patch into two, one per file that is being updated.

Quoted 2 lines

...

Signature · 1 line
pw-bot: changes-requested
#4 lore raw
In reply to Tian