Re: 2.6.28-git8: tg3 doesn't work due to firmware not loading (-git7 is ok)

On Sun, 11 Jan 2009 19:10:36 EST, Valdis.Kletnieks@vt.edu said:

In the dmesg I see during early bootup:

[    0.694230] loop: module loaded
[    0.694249] tg3.c:v3.97 (December 10, 2008)
[    0.694261] vendor=8086 device=27d4
[    0.694265] tg3 0000:09:00.0: PCI INT A -> GSI 18 (level, low) -> IRQ 18
[    0.694274] tg3 0000:09:00.0: setting latency timer to 64
[    0.696087] tg3 0000:09:00.0: wake-up capability disabled by ACPI
[    0.696094] tg3 0000:09:00.0: PME# disabled
[    0.702276] tg3 0000:09:00.0: firmware: using built-in firmware tigon/tg3_tso.bin
[    0.702287] vendor=8086 device=27d4
[    0.702288] tg3 0000:09:00.0: PCI INT A disabled
[    0.702512] console [netcon0] enabled
[    0.702515] netconsole: network logging started
[    0.702575] Driver 'sd' needs updating - please use bus_type methods

but once we get to userspace, 'ifconfig' or 'ip link show' have *zero*
about an eth0 device.

For comparison, the dmesg if I revert your patch:

[    0.696638] loop: module loaded
[    0.696658] tg3.c:v3.97 (December 10, 2008)
[    0.696670] vendor=8086 device=27d4
[    0.696674] tg3 0000:09:00.0: PCI INT A -> GSI 18 (level, low) -> IRQ 18
[    0.696683] tg3 0000:09:00.0: setting latency timer to 64
[    0.698063] tg3 0000:09:00.0: wake-up capability disabled by ACPI
[    0.698070] tg3 0000:09:00.0: PME# disabled
[    0.704276] tg3 0000:09:00.0: firmware: using built-in firmware tigon/tg3_tso.bin
[    0.704445] eth0: Tigon3 [partno(BCM5752KFBG) rev 6002] (PCI Express) MAC address 00:15:c5:c8:33:4e
[    0.704448] eth0: attached PHY is 5752 (10/100/1000Base-T Ethernet) (WireSpeed[1])
[    0.704451] eth0: RXcsums[1] LinkChgREG[0] MIirq[0] ASF[0] TSOcap[1]
[    0.704453] eth0: dma_rwctrl[76180000] dma_mask[64-bit]
[    0.704653] console [netcon0] enabled
[    0.704656] netconsole: network logging started
[    0.704718] Driver 'sd' needs updating - please use bus_type methods

So it looks like the patch is failing to finish initialization of the card.
Damned if *I* can see what's breaking it, the conversion to use a helper
function tg3_request_firmware seems sane enough....

Damn.  I wonder if netconsole's initialization is turning around and stomping
on everything?  This looks suspicious:

static int tg3_open(struct net_device *dev)
 {
        struct tg3 *tp = netdev_priv(dev);
        int err;
 
+       if (tp->fw_needed) {

Do we know for sure that tp-> struct is the same one we set up back in
tg3_request_firmware, *and* that tg3_open() doesn't get called before
tg3_init_one() (which would result in an uninitialized fw_needed)?