Skip to contentSkip to conversation

usb: gadget: u_ether: Re-attach netif device to mirror detachment

PATCH In review Richard Acayan 7 participants 8 messages Dec 18, 2023 – Jan 21, 2024

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 usb: gadget: u_ether: Re-attach netif device to mirror detachment

In 6.7-rc1, there was a netif_device_detach call added to the gether_disconnect function. This clears the __LINK_STATE_PRESENT bit of the netif device and suppresses pings (ICMP messages) and TCP connection requests from the connected host. If userspace temporarily disconnects the gadget, such as by temporarily removing configuration in the gadget configfs interface, network activity should continue to be processed when the gadget is re-connected. Mirror the netif_device_detach call with a netif_device_attach call in gether_connect to fix re-connecting gadgets.

---
 drivers/usb/gadget/function/u_ether.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/usb/gadget/function/u_ether.c +2 -0
index 9d1c40c152d8..3c5a6f6ac341 100644
@@ -1163,6 +1163,8 @@ struct net_device *gether_connect(struct gether *link)
 		if (netif_running(dev->net))
 			eth_start(dev, GFP_ATOMIC);
 
+		netif_device_attach(dev->net);
+
 	/* on error, disable any endpoints  */
 	} else {
 		(void) usb_ep_disable(link->out_ep);
Signature · 1 line
2.43.0
Tested-by on v1 1/1
#1 lore raw
Quoted 9 lines · On Montag, 18. Dezember 2023 17:45:33 CET Richard Acayan wrote:

In 6.7-rc1, there was a netif_device_detach call added to the gether_disconnect function. This clears the __LINK_STATE_PRESENT bit of the netif device and suppresses pings (ICMP messages) and TCP connection requests from the connected host. If userspace temporarily disconnects the gadget, such as by temporarily removing configuration in the gadget configfs interface, network activity should continue to be processed when the gadget is re-connected. Mirror the netif_device_detach call with a netif_device_attach call in gether_connect to fix re-connecting gadgets.

(+Cc Thorsten Leemhuis)

This appears to fix the regression on a 6.7-rc5-based build for qcom-msm8974pro-fairphone-fp2, that the NCM network gadget doesn't work. I've also heard reports from qcom-sdm845 and a PXA1908-based phone (if I see this correctly) about issues on 6.7.

In postmarketOS on the device side the usb0 interface doesn't get the IP address assigned correctly it seems, but it seems to behave a bit inconsistently - but always broken.

Anyways, with this patch everything looks good again. I hope this makes it for 6.7 final still.

Quoted 21 lines
---
 drivers/usb/gadget/function/u_ether.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/usb/gadget/function/u_ether.c +2 -0
index 9d1c40c152d8..3c5a6f6ac341 100644
@@ -1163,6 +1163,8 @@ struct net_device *gether_connect(struct gether *link)
 if (netif_running(dev->net))
 			eth_start(dev, GFP_ATOMIC);
 
+		netif_device_attach(dev->net);
+
 	/* on error, disable any endpoints  */
 	} else {
 		(void) usb_ep_disable(link->out_ep);
#2 lore raw
Quoted 18 lines · Luca Weiss
Quoted 9 lines · On Montag, 18. Dezember 2023 17:45:33 CET Richard Acayan wrote:

In 6.7-rc1, there was a netif_device_detach call added to the gether_disconnect function. This clears the __LINK_STATE_PRESENT bit of the netif device and suppresses pings (ICMP messages) and TCP connection requests from the connected host. If userspace temporarily disconnects the gadget, such as by temporarily removing configuration in the gadget configfs interface, network activity should continue to be processed when the gadget is re-connected. Mirror the netif_device_detach call with a netif_device_attach call in gether_connect to fix re-connecting gadgets.

(+Cc Thorsten Leemhuis)

This appears to fix the regression on a 6.7-rc5-based build for qcom-msm8974pro-fairphone-fp2, that the NCM network gadget doesn't work. I've also heard reports from qcom-sdm845 and a PXA1908-based phone (if I see this correctly) about issues on 6.7.

Quoted 6 lines

In postmarketOS on the device side the usb0 interface doesn't get the IP address assigned correctly it seems, but it seems to behave a bit inconsistently - but always broken.

Anyways, with this patch everything looks good again. I hope this makes it for 6.7 final still.

It will have to wait until 6.8-rc1, sorry.

Tested-by on v1 1/1
#3 lore raw
In reply to Richard Acayan
Quoted 9 lines · Richard Acayan

In 6.7-rc1, there was a netif_device_detach call added to the gether_disconnect function. This clears the __LINK_STATE_PRESENT bit of the netif device and suppresses pings (ICMP messages) and TCP connection requests from the connected host. If userspace temporarily disconnects the gadget, such as by temporarily removing configuration in the gadget configfs interface, network activity should continue to be processed when the gadget is re-connected. Mirror the netif_device_detach call with a netif_device_attach call in gether_connect to fix re-connecting gadgets.

This fixes the NCM gadget on the PXA1908 phone Luca mentioned on v6.7-rc7.

Regards,
Duje

#4 lore raw
In reply to Richard Acayan
Quoted 10 lines · Richard Acayan

In 6.7-rc1, there was a netif_device_detach call added to the gether_disconnect function. This clears the __LINK_STATE_PRESENT bit of the netif device and suppresses pings (ICMP messages) and TCP connection requests from the connected host. If userspace temporarily disconnects the gadget, such as by temporarily removing configuration in the gadget configfs interface, network activity should continue to be processed when the gadget is re-connected. Mirror the netif_device_detach call with a netif_device_attach call in gether_connect to fix re-connecting gadgets.

Quoted 3 lines

Thanks for the report. To be sure the issue doesn't fall through the cracks unnoticed, I'm adding it to regzbot, the Linux kernel regression tracking bot:

#regzbot ^introduced f49449fbc21e
#regzbot title usb: gadget: u_ether: network gadgets don't work #regzbot fix: usb: gadget: u_ether: Re-attach netif device to mirror detachment
#regzbot ignore-activity

Ciao, Thorsten (wearing his 'the Linux kernel's regression tracker' hat) --
Everything you wanna know about Linux kernel regression tracking:
https://linux-regtracking.leemhuis.info/about/#tldr
That page also explains what to do if mails like this annoy you.

Tested-by on v1 1/1
#5 lore raw
In reply to Richard Acayan

Hi,

Op 18-12-2023 om 17:45 schreef Richard Acayan:

Quoted 22 lines

In 6.7-rc1, there was a netif_device_detach call added to the gether_disconnect function. This clears the __LINK_STATE_PRESENT bit of the netif device and suppresses pings (ICMP messages) and TCP connection requests from the connected host. If userspace temporarily disconnects the gadget, such as by temporarily removing configuration in the gadget configfs interface, network activity should continue to be processed when the gadget is re-connected. Mirror the netif_device_detach call with a netif_device_attach call in gether_connect to fix re-connecting gadgets.

---
  drivers/usb/gadget/function/u_ether.c | 2 ++
  1 file changed, 2 insertions(+)
drivers/usb/gadget/function/u_ether.c +0 -0
index 9d1c40c152d8..3c5a6f6ac341 100644

drivers/usb/gadget/function/u_ether.c struct net_device *gether_connect(struct gether *link) · line 1170

Quoted 8 lines
drivers/usb/gadget/function/u_ether.c +2 -0
@@ -1163,6 +1163,8 @@ struct net_device *gether_connect(struct gether *link)
  		if (netif_running(dev->net))
  			eth_start(dev, GFP_ATOMIC);
 
+		netif_device_attach(dev->net);
+
  	/* on error, disable any endpoints  */
  	} else {
  		(void) usb_ep_disable(link->out_ep);

This works mrfld (Intel Edison Arduino) using configfs with v6.7.0. Tested using iperf3 -s on mrfld, iperf3 --bidir -c edison-usb
[ ID][Role] Interval Transfer Bitrate Retr
[ 5][TX-C] 0.00-10.00 sec 130 MBytes 109 Mbits/sec 0
sender
[ 5][TX-C] 0.00-9.99 sec 129 MBytes 108 Mbits/sec
receiver
[ 7][RX-C] 0.00-10.00 sec 167 MBytes 140 Mbits/sec 0
sender
[ 7][RX-C] 0.00-9.99 sec 166 MBytes 139 Mbits/sec
receiver

and
iperf3 -c edison-usb

[  5]   0.00-10.00  sec   247 MBytes   207 Mbits/sec    0             sender
[  5]   0.00-9.99   sec   246 MBytes   206 Mbits/sec receiver
Tested-by on v1 1/1
#6 lore raw
In reply to Richard Acayan
Quoted 9 lines · Richard Acayan

In 6.7-rc1, there was a netif_device_detach call added to the gether_disconnect function. This clears the __LINK_STATE_PRESENT bit of the netif device and suppresses pings (ICMP messages) and TCP connection requests from the connected host. If userspace temporarily disconnects the gadget, such as by temporarily removing configuration in the gadget configfs interface, network activity should continue to be processed when the gadget is re-connected. Mirror the netif_device_detach call with a netif_device_attach call in gether_connect to fix re-connecting gadgets.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
Tested-by
#7 lore raw
Quoted 11 lines · Andy Shevchenko
Quoted 9 lines · Richard Acayan

In 6.7-rc1, there was a netif_device_detach call added to the gether_disconnect function. This clears the __LINK_STATE_PRESENT bit of the netif device and suppresses pings (ICMP messages) and TCP connection requests from the connected host. If userspace temporarily disconnects the gadget, such as by temporarily removing configuration in the gadget configfs interface, network activity should continue to be processed when the gadget is re-connected. Mirror the netif_device_detach call with a netif_device_attach call in gether_connect to fix re-connecting gadgets.

Quoted 1 line

Sorry, now correct tag.

Signature · 2 lines
With Best Regards,
Andy Shevchenko