in if_tlp_pci.c... The name code really needs to be fixed so that it does
not have a buffer overflow again:
1. the cardbus and eisa code don't initialize sc_name; it probably should?
2. tlp_attach() seems to print the name and the ethernet address, but
the pci front end seems to do the same.
3. some of the name determination code in the pci front-end does not guarantee
nul terminated strings.
Really, all this name code is just bloat and another place for bugs to hide.
than a max size Ethernet frame without getting a frame-too-long error. VPC
seems to be adding 4 zeros at the end of every frame. Detect this condition
and simply truncate the packet to a max size Ethernet frame.
I now have no problems with networking on NetBSD inside Virtual PC 6.
TULIPF_BLE /* data is big endian */
TULIPF_DBO /* descriptor is big endian */
These setting is required for Big-endian bus front-end (such as APbus for
newsmips) to work with tulip. Also,
sc_maxburst
member is added to the tulip_softc to limit the maximum burst length.
This member defaults to zero which means no restriction for burst length.
98715* series.
The MX98715AEC-[C,E] use a different location in the serial eerom for
LED control, and programming it with the original location's values
caused unpredictable behavior.
Also, start integrating fixes where media changes on an adapter
under load may fail. There's more work to be done here, but I need
to sort out our internal changes a little more carefully.
cardbus_get_capability() [mirror change already made to if_tlp_pci.c]
- if_tlp_cardbus.c: If we don't find an ISV SROM, try to grab the
Ethernet address from the CIS.
- if_tlp_cardbus.c: set CardBus cards to store-and-forward mode from
the get-go.
- Put the TxThresh tables in tulipvar.h, and use them in the CardBus
and PCI front-ends to go to store-and-forward mode.
- Document the Xircom X3201-3 clone a little more.
timeout()/untimeout() API:
- Clients supply callout handle storage, thus eliminating problems of
resource allocation.
- Insertion and removal of callouts is constant time, important as
this facility is used quite a lot in the kernel.
The old timeout()/untimeout() API has been removed from the kernel.
more efficient on genuine DEC chips, but some clones apparently Lose Badly
if you use chain mode. To further complicate things, some clones *only*
do chain mode, so make this whole mess conditional.
- The Xircom doesn't bring itself out of reset; we have to do it.
- The Xircom requires transmit buffers to be aligned to 4 bytes, too.
Partially from Rafal Boni (though the Tx buffer alignment issue is
handled much differently in this version).
this consistently, and it doesn't always work even when the chip
supports it.
- Make sure things DMA'd to the chip that the chip interprets are in
little-endian mode.
add support for the DECchip 21142/21143 SIA/SYM media. Make the Macronix
98713 behave like a 21140A, and the 98713A and later behave like a 21143.
Nuke all of the Macronix-specific Nway code, as it will be identical to
the 21143 Nway code once it is written.
- 82C115 has a 128-bit multicast hash table, not 512-bit.
- Correct the way the MAC address is read from the SROM, after re-reading
the MX98715A Application Note.
Other semi-related changes:
- Differentiate between MX98715 and MX98715A.
- Improve the Macronix link-up/link-down detection.
cegger