Skip to contentSkip to conversation

hw csum failure introduced in linux-4.2

Bridge Álvaro Fernández Rojas 3 participants 7 messages Jan 22, 2016 – Feb 15, 2016 Ethernet bridge

lore
Share & tools
lore
#0 lore raw

From linux-4.2 there's a "hw csum failure" for bridges using hardware checksums when IPv6 multicast traffic is received.

Here are some reports:
https://lkml.org/lkml/2015/12/1/699
https://bugzilla.redhat.com/show_bug.cgi?id=1266601
https://discourse.osmc.tv/t/hw-csum-failure/9990
https://github.com/raspberrypi/linux/issues/1259

I think it may have been introduced in commit 9afd85c9e4552b276e2f4cfefd622bdeeffbbf26 (net: Export IGMP/MLD message validation code) but I haven't done any tests to confirm it. Any help on this issue would be really appreciated.

Regards,
Álvaro.

#1 lore raw
Quoted 9 lines · On Fri, 22 Jan 2016 22:03:55 +0100 Álvaro Fernández Rojas <[email protected]>…

From linux-4.2 there's a "hw csum failure" for bridges using hardware
checksums when IPv6 multicast traffic is received.

Here are some reports:
https://lkml.org/lkml/2015/12/1/699
https://bugzilla.redhat.com/show_bug.cgi?id=1266601
https://discourse.osmc.tv/t/hw-csum-failure/9990
https://github.com/raspberrypi/linux/issues/1259

Quoted 7 lines

I think it may have been introduced in commit 9afd85c9e4552b276e2f4cfefd622bdeeffbbf26 (net: Export IGMP/MLD message validation code) but I haven't done any tests to confirm it. Any help on this issue would be really appreciated.

Regards,
Álvaro.

What hardware? You missed the kernel bugzilla? If it was reported there I would have forwarded it to [email protected].

Can you reproduce it with 4.4?

#2 lore raw

Yes, I could reproduce it on OpenWrt using linux-4.4 on a Raspberry Pi 2.
https://gist.github.com/Noltari/5b1cfdecce5ed4bc08fd

No, I didn't open any issue on the kernel bugzilla.
If you want me to report it somewhere else I don't have any problem with it.

Regards,
Álvaro.

Quoted 23 lines · Stephen Hemminger
Quoted 9 lines · On Fri, 22 Jan 2016 22:03:55 +0100 Álvaro Fernández Rojas <[email protected]>…

From linux-4.2 there's a "hw csum failure" for bridges using hardware
checksums when IPv6 multicast traffic is received.

Here are some reports:
https://lkml.org/lkml/2015/12/1/699
https://bugzilla.redhat.com/show_bug.cgi?id=1266601
https://discourse.osmc.tv/t/hw-csum-failure/9990
https://github.com/raspberrypi/linux/issues/1259

Quoted 7 lines

I think it may have been introduced in commit 9afd85c9e4552b276e2f4cfefd622bdeeffbbf26 (net: Export IGMP/MLD message validation code) but I haven't done any tests to confirm it. Any help on this issue would be really appreciated.

Regards,
Álvaro.

What hardware? You missed the kernel bugzilla? If it was reported there I would have forwarded it to [email protected].

Can you reproduce it with 4.4?

#3 lore raw

Hi Alvaro,

Did you have a chance to narrow it down to that commit yet? Would be great if you could check whether reverting the following commits helps:

c2d4fbd2163e ("bridge: fix igmpv3 / mldv2 report parsing") a516993f0ac1 ("net: fix wrong skb_get() usage / crash in IGMP/MLD parsing code") 9afd85c9e455 ("net: Export IGMP/MLD message validation code")

Or reverting all of the above plus:

3c9e4f870012 ("bridge: multicast: call skb_checksum_{simple_, }validate")

Will try to get my hands on a Pi 2, too.

Regards, Linus

PS: 4.4.0 did not have c2d4fb yet, that was added with 4.4.1. While missing it does cause trouble, it should not cause a csum error.

Quoted 34 lines · Álvaro Fernández Rojas

Yes, I could reproduce it on OpenWrt using linux-4.4 on a Raspberry Pi 2.
https://gist.github.com/Noltari/5b1cfdecce5ed4bc08fd

No, I didn't open any issue on the kernel bugzilla.
If you want me to report it somewhere else I don't have any problem with it.

Regards,
Álvaro.

Quoted 23 lines · Stephen Hemminger
Quoted 9 lines · On Fri, 22 Jan 2016 22:03:55 +0100 Álvaro Fernández Rojas <[email protected]>…

From linux-4.2 there's a "hw csum failure" for bridges using hardware
checksums when IPv6 multicast traffic is received.

Here are some reports:
https://lkml.org/lkml/2015/12/1/699
https://bugzilla.redhat.com/show_bug.cgi?id=1266601
https://discourse.osmc.tv/t/hw-csum-failure/9990
https://github.com/raspberrypi/linux/issues/1259

Quoted 7 lines

I think it may have been introduced in commit 9afd85c9e4552b276e2f4cfefd622bdeeffbbf26 (net: Export IGMP/MLD message validation code) but I haven't done any tests to confirm it. Any help on this issue would be really appreciated.

Regards,
Álvaro.

What hardware? You missed the kernel bugzilla? If it was reported there I would have forwarded it to [email protected].

Can you reproduce it with 4.4?

#4 lore raw

Ok, was able to borrow a Raspi 2 and indeed I can reproduce the issue there.

Reverting these patches helps:

Quoted 3 lines · Linus Lüssing

c2d4fbd2163e ("bridge: fix igmpv3 / mldv2 report parsing") a516993f0ac1 ("net: fix wrong skb_get() usage / crash in IGMP/MLD parsing code") 9afd85c9e455 ("net: Export IGMP/MLD message validation code")

I'm on it.

#5 lore raw

Hi Linus,

Thanks for taking the time to look into this issue because unfortunately I don't have that time right now. BTW, the error should also be reproducible on any Raspberry Pi v1 with ethernet.

Regards,
Álvaro.

Quoted 10 lines · Linus Lüssing

Ok, was able to borrow a Raspi 2 and indeed I can reproduce the issue there.

Reverting these patches helps:

Quoted 3 lines · Linus Lüssing

c2d4fbd2163e ("bridge: fix igmpv3 / mldv2 report parsing") a516993f0ac1 ("net: fix wrong skb_get() usage / crash in IGMP/MLD parsing code") 9afd85c9e455 ("net: Export IGMP/MLD message validation code")

I'm on it.

#6 lore raw
In reply to Linus Lüssing
Quoted 2 lines · Linus Lüssing

PS: 4.4.0 did not have c2d4fb yet, that was added with 4.4.1. While missing it does cause trouble, it should not cause a csum error.

Just for the record: I was wrong here, that patch was part of v4.4 already and even v4.3.