into nuts.ninka.net:/home/davem/src/BK/net-2.5

As it has to happen before pktsched_init, that is called from
net_dev_init that is a subsys_initcall, making it the same
init level as netlink_proto_init, that ends up being called
_after_ net_dev_init, so when pktsched_init is called it finds
rtnetlink_links[PF_UNSPEC] as null and therefore not sets
the ->dumpit entry for RTM_GETQDISC (and the others too):
b00m, rtnetlink_rcv sends a failure message to tc.

into nuts.ninka.net:/home/davem/src/BK/net-2.5

After all calls to tcp_tw_deschedule we had a call to tcp_timewait_kill,
move it to the end of tcp_tw_deschedule and unexport tcp_timewait_kill,
making it static.

This fixes a oops that happens when we map a framebuffer device to a
non-existant console.

set_con2fb_map wasn't testing to see the VC we where mapping to actually
exist.  Now it does.

I also added code to fbcon_cursor to reset the hotspot if it was changed
by userland.

This kills off warnings about unused variables.

This removes EDID support for VESA.  The EDID code needs more
developement which can be done on the side.

The results where mixed. It worked for some but not for others.

The font size needs to be set for all terminals.

This was the bug that was causing dual head (vga and mda) to lock up.

into home.transmeta.com:/home/torvalds/v2.5/linux

into home.transmeta.com:/home/torvalds/v2.5/linux

into home.transmeta.com:/home/torvalds/v2.5/linux

into home.transmeta.com:/home/torvalds/v2.5/linux

into home.transmeta.com:/home/torvalds/v2.5/linux

into tetrachloride.(none):/mnt/raid/src/kernel/2.5/agpgart

into nuts.ninka.net:/home/davem/src/BK/bluetooth-2.5

into kernel.bkbits.net:/home/davem/net-2.5

Quite a few suspicious places here that pass kernel pointers
to the internal ioctl engine. 

declarations.

Much help provided by Rusty Russell in fixing device leak
and TOS modification handling bugs.

into nuts.ninka.net:/home/davem/src/BK/net-2.5

pointer. The integrated graphics AGP things don't have one.

into tetrachloride.(none):/mnt/raid/src/kernel/2.5/agpgart

Closes bugzilla #646

into home.transmeta.com:/home/torvalds/v2.5/linux

into raven.il.steeleye.com:/home/jejb/BK/scsi-for-linus-2.5

From Alex Tomas

We started using ext3_dx_readdir() for all dir_index filesystems, because
we want to return entries in hash order always, so that readdir with a
partial read + new entry added before next readdir won't be crazy.

So we now need to free the structure at filp->pricate_data even against
non-indexed directories.