Re: PCN0 problem

[Rin Okuyama <rokuyama.rk%gmail.com@localhost>]

Hi,

On 2021/07/29 23:55, David Brownlee wrote:
> On Thu, 29 Jul 2021 at 15:38, T <bobs%thelibertytree.org@localhost> wrote:
> > On Sat, 26 Jun 2021 21:42:48 -0000 (UTC), T wrote:
> >
> > > Hello,
> > > I have a problem with the pcn0 NIC on the following machine:
> > > https://dmesgd.nycbug.org/index.cgi?do=view&id=6120 The integrated pcn0
> > > causes the machine to become unresponsive if interacting with it through
> > > NetBSD. Two specific examples I can think of are:
> > > 1) When I originally obtained the system, I used the firmware net boot
> > > functionality to load the sysinst image over the network and tried to
> > > install packages from a NFS share. This would initiate and load into
> > > sysinst, but then the system would become completely unresponsive during
> > > the unpacking phase of the install, as in the screen would become frozen
> > > on output related to unpacking the sets. Pressing the NUM_lock key would
> > > not toggle the keyboard LED for it and no user interaction seemed to
> > > work (remote ping, SSH, keyboard, debugger key sequence). I started
> > > using NetBSD with this machine with one of the later version 7 releases,
> > > from which this example originates.
> > >
> > > 2) Currently, with 9.2_STABLE, I can reproduce this problem by just
> > > calling 'dhcpcd pcn0', which will cause the same unresponsive behavior
> > > as mentioned above. I have worked around this issue by using an
> > > alternative add-on NIC, but interacting with the integrated pcn NIC has
> > > always been an issue since the original example.
> > >
> > > The system firmware does not seem to have any problems with the NIC when
> > > using built-in device tests or net boot functionality.
> > >
> > > I was wondering if anyone else has had issues with this driver (likely
> > > being used on a different port) and if the problem is likely not a
> > > driver issue, but specific to port-prep? It has been mentioned to me on
> > > IRC that I might need to use something like KGDB to troubleshoot the
> > > issue, but that would need to be added to the port, because it does not
> > > currently exist.
> >
> > I did some additional testing using the firmware net boot functionality
> > to test sysinst.fs images for 6.2, 7.0, 7.1, 7.2, 8.0, 8.1, and 8.2 .
> > These all appeared to work fine when using the installer "configure
> > network" utility menu option and ping utility. The 9.0 sysinst.fs is too
> > large to work on this system so I had to switch over to the smaller
> > sysinst_small.fs image. This version did not work like the others, and
> > had a noticeable problem with the "configure network" option in the
> > utility menu. Attempting to select a NIC "media" would not accept values.
> > In prior versions, this had a populated default of (autoconfigure), but
> > was missing for 9.0 . Trying the sysinst_small.fs image for 8.2 worked
> > fine and appeared to function the same as the regular sysinst.fs image.
>
> Some random thoughts:
>
> - It sounds like the pcn has issues with all versions, but its much
> more serious with NetBSD 9.x?
> - Is there any version which works reliably (maybe boot earlier
> sysinst.fs, nfs mount a remote share and tar tzf the images to try to
> reproduce)
> - Could it be related to a combination of pcn and other (eg disk)
> activity - so tar tzf may not trigger but extracting would
> - Does a current sysinst image from
> http://nyftp.netbsd.org/pub/NetBSD-daily/HEAD/latest/ also show the
> issue
> - If there is an earlier version that works 100% I'd suggest checking
> the difference in dmesg between that and the first version to show the
> issue

(1) VirtualBox can optionally emulate Am79c970A, and it works fine as
    far as I can see with NetBSD/amd64-current:

pcn0 at pci0 dev 3 function 0: AMD PCnet-PCI Ethernet
pcn0: Am79c970A PCnet-PCI II rev 0, Ethernet address 08:00:27:7f:3d:2e
pcn0: interrupting at ioapic0 pin 19
pcn0: 10base5, 10base5-FDX, 10baseT, 10baseT-FDX, auto, auto-FDX

    Full dmesg is provided here:

http://www.netbsd.org/~rin/dmesg_vb_20210730.txt

(2) I have an Am79c971-based PCI NIC. This card works fine on a
    powerpc/ibm4xx machine running NetBSD/evbppc 9.99.86:

pcn0 at pci0 dev 3 function 0: AMD PCnet-PCI Ethernet
pcn0: Am79c971 PCnet-FAST rev 6, Ethernet address 00:06:29:ac:a5:f8
pcn0: interrupting at irq 30
nsphy0 at pcn0 phy 1: DP83840 10/100 media interface, rev. 1
nsphy0: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, auto
ukphy0 at pcn0 phy 31: Advanced Micro Devices Am79C972 internal 10BASE-T interface (OUI 0x000058, model 0x0001), rev. 1
ukphy0: 10baseT, 10baseT-FDX, auto

    I'm not sure why two MIIs are attached here, but it is linked up as
    100baseTX-FDX and scores ~20Mbps (RX) / ~30Mbps (TX) for iperf3:

# ifconfig -v pcn0
pcn0: flags=0x8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500
        address: 00:06:29:ac:a5:f8
        media: Ethernet autoselect (100baseTX full-duplex)
        status: active
        input: 39837 packets, 29957040 bytes, 2593 multicasts, 1695 errors
        output: 37320 packets, 41695557 bytes, 5 multicasts
        inet6 fe80::206:29ff:feac:a5f8%pcn0/64 flags 0 scopeid 0x2
        inet 192.168.10.181/24 broadcast 192.168.10.255 flags 0

    Full dmesg is uploaded here:

http://www.netbsd.org/~rin/dmesg_ibm4xx_20210730.txt

Thanks,
rin