ethernet r8169 crash on Asus F2A85-M/CSM

[dch24]

This is a new install on an Asus F2A85-M/CSM, looking around I found [topic=952878]ethernet module crashes with IOMMU enabled[/topic], which the author marked solved but only because he got sound working. He never actually fixed the ethernet problem.

I get:

Code: Select all

[    0.356526] AMD-Vi: Found IOMMU at 0000:00:00.2 cap 0x40
[    0.356527] AMD-Vi:  Extended features:  PreF PPR GT IA
[    0.356530] AMD-Vi: Interrupt remapping enabled
[    0.363498] AMD-Vi: Using passthrough domain for device 0000:01:00.0
[    0.364101] AMD-Vi: Lazy IO/TLB flushing enabled
[    4.242992] r8169 Gigabit Ethernet driver 2.3LK-NAPI loaded
[    4.243348] r8169 0000:04:00.0: irq 62 for MSI/MSI-X
[    4.243625] r8169 0000:04:00.0 eth0: RTL8168f/8111f at 0xffffc9001126e000, 50:46:5d:90:d5:76, XID 08000800 IRQ 62
[    4.243632] r8169 0000:04:00.0 eth0: jumbo features [frames: 9200 bytes, tx checksumming: ko]
[    7.513449] r8169 0000:04:00.0 eth0: link down
[    7.513527] r8169 0000:04:00.0 eth0: link down
[    9.978426] r8169 0000:04:00.0 eth0: link up

Until I do a large transfer or just use netcat to run the TX queue at 100%, then the device stops responding and I have to reboot to get it back. I also get this message in the kernel:

Code: Select all

[   59.582220] AMD-Vi: Event logged [IO_PAGE_FAULT device=04:00.0 domain=0x001b address=0x0000000000003000 flags=0x0050]

My first thought was to disable the iommu with "iommu=off" but then running the TX queue at 100% causes a kernel panic.

I filed https://bugzilla.kernel.org/show_bug.cgi?id=55841 thinking this was a kernel driver problem but after I thought some more it occurred to me to ask here first.

[Anon-E-moose]

actually when I turned on the "iommu=pt" kernel boot flag the network started working fine.

AMD (don't know about Intel) doesn't work well if iommu is turned on and the pt flag is not set.

Passthrough is working perfectly fine.

So try it.

Best of luck

[dch24]

the "iommu=pt" kernel boot flag the network started working fine

That worked!

Ok, thanks! I'll put this info on the kernel bugzilla.