What can introduce very high latency inside the broken Fast Ethernet controller?

[m4rtin]

I have a fairly old 10BASE-T/100BASE-TX NIC for PCI bus with Realtek RTL8139D chip:



At first I had very high(up to 2000ms) latency with this NIC even to hosts in my local area network(latency should be <1ms). In addition, packet delay variation was very high fluctuating from 20-30ms to 2000ms. Now if I load the driver:

Code: [Select]

router:~# modprobe -v 8139too
insmod /lib/modules/3.2.0-4-686-pae/kernel/drivers/net/ethernet/realtek/8139too.ko
router:~#

..the card does not respond at all:

Code: [Select]

[  131.363894] 8139too 0000:00:0d.0: Chip not responding, ignoring board
[  131.363992] 8139too: probe of 0000:00:0d.0 failed with error -5

Other Network Interface Cards in the very same PCI slot work just fine. Looks like the RTL8139D Fast Ethernet controller is faulty. I am curious what happens inside the chip if it is half-broken and introduces such high latency? I understand that this would require exact schematics of the chip(obviously proprietary), but an educated guess from someone who has experience with chip design is also fine

[wraper]

Probably crystal oscillator frequency is off or data in 93C46 is corrupted.

[mazurov]

Try to turn off auto-negotiation on both sides of the link.

[Shock]

Clean the jack and slot pins with alcohol and a cotton bud etc.

Use a loop back connector when testing unless you like wasting time.
Don't use an OS use something like the vendors floppy diags or if your low on options a generic one.

You might be right and it's faulty but if you didn't know the above or things like how ATX power and interrupts work then likely your jumping to the wrong conclusion.

[janoc]

Just bin it. These Realtek chipsets are known for their bugs and problems - google  "8139too" and see for yourself. They are used because they are dirt cheap, but often require some dirty driver hacks working around problems with the silicon itself.

[rob77]

8139too: probe of 0000:00:0d.0 failed with error -5

i would bet on the PCI connector contacts that log message tells you the probe of PCI device failed. if cleaning the connector doesn't help then desolder the usable parts (crystal, LDO, eeprom, that 1Kv ceramic cap, ferrite beads) and trash the rest