NIC chipsets

[Razor]

Regarding the RL8139:

Here are some comments from the FreeBSD driver:

 * The RealTek 8139 PCI NIC redefines the meaning of 'low end.' This is
 * probably the worst PCI ethernet controller ever made, with the possible
 * exception of the FEAST chip made by SMC. The 8139 supports bus-master
 * DMA, but it has a terrible interface that nullifies any performance
 * gains that bus-master DMA usually offers.
 *
 * For transmission, the chip offers a series of four TX descriptor
 * registers. Each transmit frame must be in a contiguous buffer, aligned
 * on a longword (32-bit) boundary. This means we almost always have to
 * do mbuf copies in order to transmit a frame, except in the unlikely
 * case where a) the packet fits into a single mbuf, and b) the packet
 * is 32-bit aligned within the mbuf's data area. The presence of only
 * four descriptor registers means that we can never have more than four
 * packets queued for transmission at any one time.
 *
 * Reception is not much better. The driver has to allocate a single large
 * buffer area (up to 64K in size) into which the chip will DMA received
 * frames. Because we don't know where within this region received packets
 * will begin or end, we have no choice but to copy data from the buffer
 * area into mbufs in order to pass the packets up to the higher protocol
 * levels.

and further down:

/*
* Here's a totally undocumented fact for you. When the
* RealTek chip is in the process of copying a packet into
* RAM for you, the length will be 0xfff0. If you spot a
* packet header with this value, you need to stop. The
* datasheet makes absolutely no mention of this and
* RealTek should be shot for this.
*/

You can check the driver here.