[PATCH net-next] drivers: net: davinci_mdio: prevent sprious timeout

[PATCH net-next] drivers: net: davinci_mdio: prevent sprious timeout

[Sekhar Nori]

A well timed kernel preemption in the time_after() loop
in wait_for_idle() can result in a spurious timeout
error to be returned.

Fix it by checking for status of hardware before returning
timeout error.

Signed-off-by: Sekhar Nori <nsekhar@ti.com>
---
The issue has not been personally observed by me, but has
been reported by users. Sending for next-next given the
non-critical nature. There is seems to be no easy way to
reproduce this.

 drivers/net/ethernet/ti/davinci_mdio.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/ti/davinci_mdio.c b/drivers/net/ethernet/ti/davinci_mdio.c
index 3c33f4504d8e..4fbd04fd38cf 100644
--- a/drivers/net/ethernet/ti/davinci_mdio.c
+++ b/drivers/net/ethernet/ti/davinci_mdio.c
@@ -231,10 +231,16 @@ static inline int wait_for_idle(struct davinci_mdio_data *data)
 
 	while (time_after(timeout, jiffies)) {
 		if (__raw_readl(&regs->control) & CONTROL_IDLE)
-			return 0;
+			goto out;
 	}
-	dev_err(data->dev, "timed out waiting for idle\n");
-	return -ETIMEDOUT;
+
+	if (!(__raw_readl(&regs->control) & CONTROL_IDLE)) {
+		dev_err(data->dev, "timed out waiting for idle\n");
+		return -ETIMEDOUT;
+	}
+
+out:
+	return 0;
 }
 
 static int davinci_mdio_read(struct mii_bus *bus, int phy_id, int phy_reg)
-- 
2.16.2

[PATCH net-next] drivers: net: davinci_mdio: prevent sprious timeout

[Sekhar Nori]

A well timed kernel preemption in the time_after() loop
in wait_for_idle() can result in a spurious timeout
error to be returned.

Fix it by checking for status of hardware before returning
timeout error.

Signed-off-by: Sekhar Nori <nsekhar@ti.com>
---
The issue has not been personally observed by me, but has
been reported by users. Sending for next-next given the
non-critical nature. There is seems to be no easy way to
reproduce this.

 drivers/net/ethernet/ti/davinci_mdio.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/ti/davinci_mdio.c b/drivers/net/ethernet/ti/davinci_mdio.c
index 3c33f4504d8e..4fbd04fd38cf 100644
--- a/drivers/net/ethernet/ti/davinci_mdio.c
+++ b/drivers/net/ethernet/ti/davinci_mdio.c
@@ -231,10 +231,16 @@ static inline int wait_for_idle(struct davinci_mdio_data *data)
 
 	while (time_after(timeout, jiffies)) {
 		if (__raw_readl(&regs->control) & CONTROL_IDLE)
-			return 0;
+			goto out;
 	}
-	dev_err(data->dev, "timed out waiting for idle\n");
-	return -ETIMEDOUT;
+
+	if (!(__raw_readl(&regs->control) & CONTROL_IDLE)) {
+		dev_err(data->dev, "timed out waiting for idle\n");
+		return -ETIMEDOUT;
+	}
+
+out:
+	return 0;
 }
 
 static int davinci_mdio_read(struct mii_bus *bus, int phy_id, int phy_reg)
-- 
2.16.2

Re: [PATCH net-next] drivers: net: davinci_mdio: prevent sprious timeout

[Andrew Lunn]

On Tue, May 08, 2018 at 01:56:38PM +0530, Sekhar Nori wrote:
> A well timed kernel preemption in the time_after() loop
> in wait_for_idle() can result in a spurious timeout
> error to be returned.
> 
> Fix it by checking for status of hardware before returning
> timeout error.
> 
> Signed-off-by: Sekhar Nori <nsekhar@ti.com>

I've seen this with other drivers as well.

I suggest you make use of readx_poll_timeout(), or one of its
cousins. They get this right.

	 Andrew

Re: [PATCH net-next] drivers: net: davinci_mdio: prevent sprious timeout

[Sekhar Nori]

On Tuesday 08 May 2018 06:18 PM, Andrew Lunn wrote:
> On Tue, May 08, 2018 at 01:56:38PM +0530, Sekhar Nori wrote:
>> A well timed kernel preemption in the time_after() loop
>> in wait_for_idle() can result in a spurious timeout
>> error to be returned.
>>
>> Fix it by checking for status of hardware before returning
>> timeout error.
>>
>> Signed-off-by: Sekhar Nori <nsekhar@ti.com>
> 
> I've seen this with other drivers as well.
> 
> I suggest you make use of readx_poll_timeout(), or one of its
> cousins. They get this right.

Thanks for pointing me to these. I somehow thought these helpers are
only available with regmap.

Sending a version using readl_poll_timeout(). I know __raw_readl() is
used elsewhere in the driver, but I think that should be cleaned up
sometime to use readl_relaxed() at least. So not sure if there is
benefit in persisting with existing accessor.

Thanks,
Sekhar

Re: [PATCH net-next] drivers: net: davinci_mdio: prevent sprious timeout

[Sekhar Nori]

On Tuesday 08 May 2018 06:18 PM, Andrew Lunn wrote:
> On Tue, May 08, 2018 at 01:56:38PM +0530, Sekhar Nori wrote:
>> A well timed kernel preemption in the time_after() loop
>> in wait_for_idle() can result in a spurious timeout
>> error to be returned.
>>
>> Fix it by checking for status of hardware before returning
>> timeout error.
>>
>> Signed-off-by: Sekhar Nori <nsekhar@ti.com>
> 
> I've seen this with other drivers as well.
> 
> I suggest you make use of readx_poll_timeout(), or one of its
> cousins. They get this right.

Thanks for pointing me to these. I somehow thought these helpers are
only available with regmap.

Sending a version using readl_poll_timeout(). I know __raw_readl() is
used elsewhere in the driver, but I think that should be cleaned up
sometime to use readl_relaxed() at least. So not sure if there is
benefit in persisting with existing accessor.

Thanks,
Sekhar