Bug#297529: kernel-image-2.6.8-2-686: via_velocity: receiver hang after setting mtu >1500

[Anders Boström <anders@netinsight.se>]

Package: kernel-image-2.6.8-2-686
Version: 2.6.8-13
Severity: normal

The via velocity GE-driver isn't working after setting MTU to higher
then 1500 (and setting it to lower than 1500 isn't allowed by the
driver).

Only a few frames go through after setting the mtu to higher then 1500
(ifconfig ethX mtu 1600), and in some cases no frames at all. The size
of the frames don't seem to make any difference. Ethereal indicates
that the receiver stops working. Sending of frames still works.

When this has occured must the interface be taken down and then up in
order to get it working again. Just setting the mtu to 1500 again do
not solve the problem. Also, the kernel load was 0.67 when this
occured, on an idle system. The load went down to <0.05 after taking
the interface down.

One oddity in the driver code is this:

--------------------------------------------------------------------------
static int velocity_open(struct net_device *dev)
{
        struct velocity_info *vptr = dev->priv;
        int ret;

        vptr->rx_buf_sz = (dev->mtu <= 1504 ? PKT_BUF_SZ : dev->mtu + 32);

        ret = velocity_init_rings(vptr);
--------------------------------------------------------------------------
static int velocity_change_mtu(struct net_device *dev, int new_mtu)
{
        struct velocity_info *vptr = dev->priv;
        unsigned long flags;
        int oldmtu = dev->mtu;
        int ret = 0;

        if ((new_mtu < VELOCITY_MIN_MTU) || new_mtu > (VELOCITY_MAX_MTU)) {
                VELOCITY_PRT(MSG_LEVEL_ERR, KERN_NOTICE "%s: Invalid MTU.\n", 
                                vptr->dev->name);
                return -EINVAL;
        }

        if (new_mtu != oldmtu) {
                spin_lock_irqsave(&vptr->lock, flags);

                netif_stop_queue(dev);
                velocity_shutdown(vptr);

                velocity_free_td_ring(vptr);
                velocity_free_rd_ring(vptr);

                dev->mtu = new_mtu;
                if (new_mtu > 8192)
                        vptr->rx_buf_sz = 9 * 1024;
                else if (new_mtu > 4096)
                        vptr->rx_buf_sz = 8192;
                else
                        vptr->rx_buf_sz = 4 * 1024;

                ret = velocity_init_rd_ring(vptr);
--------------------------------------------------------------------------

And PKT_BUF_SZ is set to 1540. The resulting size of the rd's isn't
the same after changing the mtu to first 1600 and then back to 1500 as
when set to 1500 in open.

lspci -vv:

0000:02:04.0 Ethernet controller: VIA Technologies, Inc. VT6120/VT6121/VT6122 Gigabit Ethernet Adapter (rev 11)
        Subsystem: VIA Technologies, Inc.: Unknown device 0110
        Control: I/O+ Mem+ BusMaster- SpecCycle- MemWINV+ VGASnoop- ParErr- Stepping- SERR+ FastB2B-
        Status: Cap+ 66MHz+ UDF- FastB2B- ParErr- DEVSEL=medium >TAbort- <TAbort- <MAbort- >SERR- <PERR-
        Interrupt: pin A routed to IRQ 201
        Region 0: I/O ports at 4000 [size=256]
        Region 1: Memory at e0201000 (32-bit, non-prefetchable) [size=256]
        Capabilities: [50] Power Management version 2
                Flags: PMEClk- DSI- D1+ D2+ AuxCurrent=0mA PME(D0+,D1+,D2+,D3hot+,D3cold+)
                Status: D3 PME-Enable- DSel=0 DScale=0 PME-


/ Anders

-- System Information:
Debian Release: 3.1
  APT prefers testing
  APT policy: (500, 'testing'), (50, 'unstable')
Architecture: i386 (i686)
Kernel: Linux 2.6.8-2-686
Locale: LANG=C, LC_CTYPE=C (charmap=ISO-8859-1) (ignored: LC_ALL set to sv_SE)

Versions of packages kernel-image-2.6.8-2-686 depends on:
ii  coreutils [fileutils]         5.2.1-2    The GNU core utilities
ii  initrd-tools                  0.1.77     tools to create initrd image for p
ii  module-init-tools             3.2-pre1-2 tools for managing Linux kernel mo

-- no debconf information