AROS loses network connection, "no buffer space available"

[retrofaza]

1. launch Shell and type the command "ping google.com"
2. launch OWB and go to the aros archive page. Download some files.

While the files are being downloaded, AROS loses its connection to the Internet, and ping starts sending the message "no buffer space available".

This behavior occurs especially on:

• nvidianet.device (tested on ASUS M2N68-AM SE2)
• pcnet32.device (tested on VirtualBox 7.0.12)

I also manage to trigger this bug on rtl8139.device (ASUS M2N68-AM SE2 + pci network card), but not every time, easier when I download several files simultaneously and browse the web at the same time.

Despite trying, I have not been able to trigger this bug on ASUS M2N68-AM SE2 with USB internet (usbrndis.device).

[ncafferkey]

Try "ping -f" to your router to perhaps reproduce this problem more easily. E.g. ping -f 192.168.0.1

I would be good to know also if this bug definitely needs both ping and OWB or if just one of them can trigger it.

[retrofaza]

Using ping is not necessary to trigger the bug. Recently I discovered that ping gives the message "no buffer space available". However, the bug also occurs when I use only OWB or when I use, for example, only MPlayer with Internet radio. Today I checked "ping -f" and on nvidianet.device alone this command without other programs also triggers the bug.

[ncafferkey]

Is this a regression with a particular build (sorry, I haven't been keeping up with recent test builds)?

[retrofaza]

I don't think this is a regression in recent builds. This behavior occurs in the latest builds, but also in Icaros Desktop 2.3 from 2018.

[ncafferkey]

I wonder if this is connected to the USB problems with nVidia chipsets. Are you using any USB devices on the M2N68-AM? I tried "ping -f" on mine today with Icaros 2.3 and it ran for three hours without errors (apart from 4% packet loss, which isn't too bad for a ping flood). I don't have any USB devices connected, not even keyboard or mouse.

[retrofaza]

I don't have any USB devices connected. Are you pinging the router or the external domain? In my case it is much faster to trigger a bug when I ping an external domain (e.g. google.com).

[ncafferkey]

I was pinging a router. I pinged google.com just now but only for five minutes (as it's not considered polite to ping flood the internet). How long do you typically have to wait before getting errors?

[retrofaza]

In the case of nvidianet.device pinging google.com with -f crashes after a few minutes, and sometimes even after less than a minute

[retrofaza]

I recorded a video (VirtualBox, icaros desktop, pcnet32.device). Bug under two minutes.

https://drive.google.com/file/d/19kUkYJeS56FYPEX7aKFmDd0TQqYIDa1s/view?usp=sharing

[deadwood2]

I can confirm this behavior with AROS One 2.4 under VirtualBox 7.0.12. In my case it takes less then a minute with ping -f google.com to start getting "No buffer" errors.

My setup: computer (Linux Mint 21) which runs VBox is connected via ethernet to router which uses LTE modem to access Internet

[retrofaza]

In this video, I am turning on the ASUS M2N68-AM SE2. The only thing I have connected to my computer is:

• ps/2 mouse
• ps/2 keyboard
• IDE HDD
  System: Latest build of AROS x86 (ABIv0 20220318-1)
  Driver: nvidianet.device

https://drive.google.com/file/d/1i8izgPxhsPKSIcZOzoqw800Atb_K0etL/view?usp=sharing

[ncafferkey]

Thanks for the video. I can also reproduce the error easily with pcnet.device in VirtualBox (with ping -f alone, against my router, in just a few seconds).

However, I can't reproduce with nvidianet.device unfortunately. I wonder what the difference is, as I think I have the same motherboard as you. What's the exact version and file size of your nvidianet.device?

[retrofaza]

23304 bytes. The same as in the latest buid:
https://github.com/deadwood2/AROS/releases/download/ABIv0_20220318-1/AROS-20220318-1-pc-i386-boot-iso.zip

[ncafferkey]

I retested with the version from that build but there's no difference for me. Do you have any PCI cards installed?

[retrofaza]

As I wrote before: only mouse and ps/2 keyboard, no other accessories or extensions

BIOS: 1801 (latest)

[ncafferkey]

Have you tried Icaros on that machine, or only on VirtualBox?

[retrofaza]

I tested icaros only on VirtualBox.

[ncafferkey]

I just realised my board probably isn't an SE2. Can you post a screengrab of the Ethernet information from PCITool please?

[deadwood2]

FYI, I tried my hardware M2N68-AM SE2 and it takes less then 10 seconds to get "No buffer" error with ping -f google.com using nvidianet.device.

Here is PCIInfo:

Driver Name: pcipc.hidd
Direct bus: yes
IÒ base: 0x00000000
Hardware info: IA32-native Direct Access PCI Bus Controller
Description: Bridge Other
Vendor name: NVIDIA Corporation
Product name: MCP61 Ethernet
Subsystem: ASUSTeK Computer Inc. : M4N68T series motherboard
Used by: nvidianet.device
VendorID: 0x10de
ProductID: 0x03ef
RevisionID: 0xa2
Class: 0x06
Subclass: 0x80
Interface: 0x00
ROM Base: N/A
ROM Size: N/A
IRQ: 10 (A)
Status: IO: yes, MEM: yes, Master: yes, PaletteSnoop: no, 66MHz capable: yes
RangeList_0: Bar0: Memory at 0xdfffd000 (size 0x1000)
RangeList_1: Bar1: I/O at 0xe480 (size 0x8)
RangeList_2: Bar2: I/O at 0xe000 (size 0x8)
RangeList_3: Bar3: I/O at 0xdc00 (size 0x4)
RangeList_4: Bar4: I/O at 0xffa0 (size 0x10)
RangeList_5: Bar5: Memory at 0xdfffc000 (size 0x1000)

Edit: this is with clean 20220318-1 release installed.

[ncafferkey]

Mine's an MCP67 with product ID 0x054c. I may look into acquiring an SE2 or another motherboard with MCP61 in order to debug it.

Overall, these errors look to be caused by individual driver bugs rather than a general networking bug. I think a lot of our network drivers behave poorly when stress tested unfortunately. We may want to open bug reports for individual drivers to replace this one.

[retrofaza]

[deadwood2]

Two new bugs related to specific configurations created.