| Message ID | 20231207114209.14595-1-jtornosm@redhat.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [v5] net: usb: ax88179_178a: avoid failed operations when device is disconnected | expand |
On Thu, Dec 07, 2023 at 12:42:09PM +0100, Jose Ignacio Tornos Martinez wrote: > When the device is disconnected we get the following messages showing > failed operations: > Nov 28 20:22:11 localhost kernel: usb 2-3: USB disconnect, device number 2 > Nov 28 20:22:11 localhost kernel: ax88179_178a 2-3:1.0 enp2s0u3: unregister 'ax88179_178a' usb-0000:02:00.0-3, ASIX AX88179 USB 3.0 Gigabit Ethernet > Nov 28 20:22:11 localhost kernel: ax88179_178a 2-3:1.0 enp2s0u3: Failed to read reg index 0x0002: -19 > Nov 28 20:22:11 localhost kernel: ax88179_178a 2-3:1.0 enp2s0u3: Failed to write reg index 0x0002: -19 > Nov 28 20:22:11 localhost kernel: ax88179_178a 2-3:1.0 enp2s0u3 (unregistered): Failed to write reg index 0x0002: -19 > Nov 28 20:22:11 localhost kernel: ax88179_178a 2-3:1.0 enp2s0u3 (unregistered): Failed to write reg index 0x0001: -19 > Nov 28 20:22:11 localhost kernel: ax88179_178a 2-3:1.0 enp2s0u3 (unregistered): Failed to write reg index 0x0002: -19 > > The reason is that although the device is detached, normal stop and > unbind operations are commanded from the driver. These operations are > not necessary in this situation, so avoid these logs when the device is > detached if the result of the operation is -ENODEV and if the new flag > informing about the disconnecting status is enabled. > > cc: stable@vger.kernel.org > Fixes: e2ca90c276e1f ("ax88179_178a: ASIX AX88179_178A USB 3.0/2.0 to gigabit ethernet adapter driver") > Signed-off-by: Jose Ignacio Tornos Martinez <jtornosm@redhat.com> > --- > V1 -> V2: > - Follow the suggestions from Alan Stern and Oliver Neukum to check the > result of the operations (-ENODEV) and not the internal state of the USB > layer (USB_STATE_NOTATTACHED). > V2 -> V3 > - Add cc: stable line in the signed-off-by area. > V3 -> V4 > - Follow the suggestions from Oliver Neukum to use only one flag when > disconnecting and include barriers to avoid memory ordering issues. > V4 -> V5 > - Fix my misundestanding and follow the suggestion from Alan Stern to > syncronize and not order the flag. I did not suggest that you should synchronize anything. What I said was that the problem you faced was one of synchronization, not of ordering. In fact, you should not try to synchronize -- ultimately it is impossible to do. The only way to get true synchronization is to prevent the user from disconnecting the device and bringing the interface down at the same time, and obviously the kernel cannot do this. > drivers/net/usb/ax88179_178a.c | 47 +++++++++++++++++++++++++++++----- > 1 file changed, 40 insertions(+), 7 deletions(-) > > diff --git a/drivers/net/usb/ax88179_178a.c b/drivers/net/usb/ax88179_178a.c > index 4ea0e155bb0d..e07614799f75 100644 > --- a/drivers/net/usb/ax88179_178a.c > +++ b/drivers/net/usb/ax88179_178a.c > @@ -173,6 +173,8 @@ struct ax88179_data { > u8 in_pm; > u32 wol_supported; > u32 wolopts; > + u8 disconnecting; > + struct mutex lock; > }; > > struct ax88179_int_data { > @@ -208,6 +210,8 @@ static int __ax88179_read_cmd(struct usbnet *dev, u8 cmd, u16 value, u16 index, > { > int ret; > int (*fn)(struct usbnet *, u8, u8, u16, u16, void *, u16); > + struct ax88179_data *ax179_data = dev->driver_priv; > + u8 disconnecting; > > BUG_ON(!dev); > > @@ -219,9 +223,14 @@ static int __ax88179_read_cmd(struct usbnet *dev, u8 cmd, u16 value, u16 index, > ret = fn(dev, cmd, USB_DIR_IN | USB_TYPE_VENDOR | USB_RECIP_DEVICE, > value, index, data, size); > > - if (unlikely(ret < 0)) > - netdev_warn(dev->net, "Failed to read reg index 0x%04x: %d\n", > - index, ret); > + if (unlikely(ret < 0)) { > + mutex_lock(&ax179_data->lock); > + disconnecting = ax179_data->disconnecting; > + mutex_unlock(&ax179_data->lock); Clearly you don't understand how mutexes work. Using a mutex like this accomplishes nothing. > + if (!(ret == -ENODEV && disconnecting)) > + netdev_warn(dev->net, "Failed to read reg index 0x%04x: %d\n", > + index, ret); > + } > > return ret; > } > @@ -231,6 +240,8 @@ static int __ax88179_write_cmd(struct usbnet *dev, u8 cmd, u16 value, u16 index, > { > int ret; > int (*fn)(struct usbnet *, u8, u8, u16, u16, const void *, u16); > + struct ax88179_data *ax179_data = dev->driver_priv; > + u8 disconnecting; > > BUG_ON(!dev); > > @@ -242,9 +253,14 @@ static int __ax88179_write_cmd(struct usbnet *dev, u8 cmd, u16 value, u16 index, > ret = fn(dev, cmd, USB_DIR_OUT | USB_TYPE_VENDOR | USB_RECIP_DEVICE, > value, index, data, size); > > - if (unlikely(ret < 0)) > - netdev_warn(dev->net, "Failed to write reg index 0x%04x: %d\n", > - index, ret); > + if (unlikely(ret < 0)) { > + mutex_lock(&ax179_data->lock); > + disconnecting = ax179_data->disconnecting; > + mutex_unlock(&ax179_data->lock); Same here. > + if (!(ret == -ENODEV && disconnecting)) > + netdev_warn(dev->net, "Failed to write reg index 0x%04x: %d\n", > + index, ret); > + } > > return ret; > } > @@ -492,6 +508,22 @@ static int ax88179_resume(struct usb_interface *intf) > return usbnet_resume(intf); > } > > +static void ax88179_disconnect(struct usb_interface *intf) > +{ > + struct usbnet *dev = usb_get_intfdata(intf); > + struct ax88179_data *ax179_data; > + > + if (!dev) > + return; > + > + ax179_data = dev->driver_priv; > + mutex_lock(&ax179_data->lock); > + ax179_data->disconnecting = 1; > + mutex_unlock(&ax179_data->lock); And again, this does nothing. Imagine a situation where the user brings the interface down on CPU 1, while at the same time CPU 2 handles a disconnection. Then the following timeline of events can occur: CPU 1 CPU 2 --------------------------- ----------------------------- The stop operation calls __ax88179_read_cmd() The device is unplugged The routine tries to communicate with the device and gets a -ENODEV error because the device is gone The USB core calls ax88179_disconnect() The routine locks the mutex, reads ax179_data->disconnecting = 0, and unlocks the mutex ax88179_disconnect() locks the mutex, sets ax179_data->disconnecting = 1, and unlocks the mutex The routine prints an error message As you can see, using the mutex does not prevent the problem from occurring. You should not add the mutex at all; it serves no purpose. Alan Stern
diff --git a/drivers/net/usb/ax88179_178a.c b/drivers/net/usb/ax88179_178a.c index 4ea0e155bb0d..e07614799f75 100644 --- a/drivers/net/usb/ax88179_178a.c +++ b/drivers/net/usb/ax88179_178a.c @@ -173,6 +173,8 @@ struct ax88179_data { u8 in_pm; u32 wol_supported; u32 wolopts; + u8 disconnecting; + struct mutex lock; }; struct ax88179_int_data { @@ -208,6 +210,8 @@ static int __ax88179_read_cmd(struct usbnet *dev, u8 cmd, u16 value, u16 index, { int ret; int (*fn)(struct usbnet *, u8, u8, u16, u16, void *, u16); + struct ax88179_data *ax179_data = dev->driver_priv; + u8 disconnecting; BUG_ON(!dev); @@ -219,9 +223,14 @@ static int __ax88179_read_cmd(struct usbnet *dev, u8 cmd, u16 value, u16 index, ret = fn(dev, cmd, USB_DIR_IN | USB_TYPE_VENDOR | USB_RECIP_DEVICE, value, index, data, size); - if (unlikely(ret < 0)) - netdev_warn(dev->net, "Failed to read reg index 0x%04x: %d\n", - index, ret); + if (unlikely(ret < 0)) { + mutex_lock(&ax179_data->lock); + disconnecting = ax179_data->disconnecting; + mutex_unlock(&ax179_data->lock); + if (!(ret == -ENODEV && disconnecting)) + netdev_warn(dev->net, "Failed to read reg index 0x%04x: %d\n", + index, ret); + } return ret; } @@ -231,6 +240,8 @@ static int __ax88179_write_cmd(struct usbnet *dev, u8 cmd, u16 value, u16 index, { int ret; int (*fn)(struct usbnet *, u8, u8, u16, u16, const void *, u16); + struct ax88179_data *ax179_data = dev->driver_priv; + u8 disconnecting; BUG_ON(!dev); @@ -242,9 +253,14 @@ static int __ax88179_write_cmd(struct usbnet *dev, u8 cmd, u16 value, u16 index, ret = fn(dev, cmd, USB_DIR_OUT | USB_TYPE_VENDOR | USB_RECIP_DEVICE, value, index, data, size); - if (unlikely(ret < 0)) - netdev_warn(dev->net, "Failed to write reg index 0x%04x: %d\n", - index, ret); + if (unlikely(ret < 0)) { + mutex_lock(&ax179_data->lock); + disconnecting = ax179_data->disconnecting; + mutex_unlock(&ax179_data->lock); + if (!(ret == -ENODEV && disconnecting)) + netdev_warn(dev->net, "Failed to write reg index 0x%04x: %d\n", + index, ret); + } return ret; } @@ -492,6 +508,22 @@ static int ax88179_resume(struct usb_interface *intf) return usbnet_resume(intf); } +static void ax88179_disconnect(struct usb_interface *intf) +{ + struct usbnet *dev = usb_get_intfdata(intf); + struct ax88179_data *ax179_data; + + if (!dev) + return; + + ax179_data = dev->driver_priv; + mutex_lock(&ax179_data->lock); + ax179_data->disconnecting = 1; + mutex_unlock(&ax179_data->lock); + + usbnet_disconnect(intf); +} + static void ax88179_get_wol(struct net_device *net, struct ethtool_wolinfo *wolinfo) { @@ -1274,6 +1306,7 @@ static int ax88179_bind(struct usbnet *dev, struct usb_interface *intf) ax179_data = kzalloc(sizeof(*ax179_data), GFP_KERNEL); if (!ax179_data) return -ENOMEM; + mutex_init(&ax179_data->lock); dev->driver_priv = ax179_data; @@ -1906,7 +1939,7 @@ static struct usb_driver ax88179_178a_driver = { .suspend = ax88179_suspend, .resume = ax88179_resume, .reset_resume = ax88179_resume, - .disconnect = usbnet_disconnect, + .disconnect = ax88179_disconnect, .supports_autosuspend = 1, .disable_hub_initiated_lpm = 1, };
When the device is disconnected we get the following messages showing failed operations: Nov 28 20:22:11 localhost kernel: usb 2-3: USB disconnect, device number 2 Nov 28 20:22:11 localhost kernel: ax88179_178a 2-3:1.0 enp2s0u3: unregister 'ax88179_178a' usb-0000:02:00.0-3, ASIX AX88179 USB 3.0 Gigabit Ethernet Nov 28 20:22:11 localhost kernel: ax88179_178a 2-3:1.0 enp2s0u3: Failed to read reg index 0x0002: -19 Nov 28 20:22:11 localhost kernel: ax88179_178a 2-3:1.0 enp2s0u3: Failed to write reg index 0x0002: -19 Nov 28 20:22:11 localhost kernel: ax88179_178a 2-3:1.0 enp2s0u3 (unregistered): Failed to write reg index 0x0002: -19 Nov 28 20:22:11 localhost kernel: ax88179_178a 2-3:1.0 enp2s0u3 (unregistered): Failed to write reg index 0x0001: -19 Nov 28 20:22:11 localhost kernel: ax88179_178a 2-3:1.0 enp2s0u3 (unregistered): Failed to write reg index 0x0002: -19 The reason is that although the device is detached, normal stop and unbind operations are commanded from the driver. These operations are not necessary in this situation, so avoid these logs when the device is detached if the result of the operation is -ENODEV and if the new flag informing about the disconnecting status is enabled. cc: stable@vger.kernel.org Fixes: e2ca90c276e1f ("ax88179_178a: ASIX AX88179_178A USB 3.0/2.0 to gigabit ethernet adapter driver") Signed-off-by: Jose Ignacio Tornos Martinez <jtornosm@redhat.com> --- V1 -> V2: - Follow the suggestions from Alan Stern and Oliver Neukum to check the result of the operations (-ENODEV) and not the internal state of the USB layer (USB_STATE_NOTATTACHED). V2 -> V3 - Add cc: stable line in the signed-off-by area. V3 -> V4 - Follow the suggestions from Oliver Neukum to use only one flag when disconnecting and include barriers to avoid memory ordering issues. V4 -> V5 - Fix my misundestanding and follow the suggestion from Alan Stern to syncronize and not order the flag. drivers/net/usb/ax88179_178a.c | 47 +++++++++++++++++++++++++++++----- 1 file changed, 40 insertions(+), 7 deletions(-)