into penguin.transmeta.com:/home/penguin/torvalds/repositories/kernel/linux

into penguin.transmeta.com:/home/penguin/torvalds/repositories/kernel/linux

CONFIG_SERIAL_TX3912_CONSOLE, CONFIG_AU1000_SERIAL_CONSOLE,
CONFIG_AU1000_UART, CONFIG_EUROTECH_WDT to drivers/char/Config.help.

from e1000 net driver.

that accidentally crept back into driver.

into mandrakesoft.com:/spare/repo/net-drivers-2.5

into mandrakesoft.com:/spare/repo/misc-2.5

hardware limits) to the 8139cp net driver.

	- removed all usage of port->sem as the usb serial core now does this.

	- reworked urb handling, getting rid of lots of code now that we have
	  proper urb reference counting.
	- removed port locks as the usb-serial core now does this.
	- added support for the Palm m515 thanks to SilaS

	- cleaned up some whitespace issues
	- changed MOD_INC logic for the generic driver
	- the port->sem lock is now taken by the serial core, not the individual
	  usb-serial drivers.  This is to reduce races.

  	Added support for the HP Jornada.

Basically, the patch turns switching off FSBR into a lazy operation with
the assumption there will be another transfer shortly afterwards. This
works wonders for usb-storage for instance.

  
    - Moves 8 functions from usb.[hc] to hcd.[hc]
    - Also moves some data structures and types
    - Now usbdevfs and "old" HCDs #include "hcd.h"
    - Minor tweaks to the "hcd" layer (one less FIXME)
    - Minor kernel doc and comment cleanups
  
Basically this continues moving the HCD-only functionality
out of the way of normal USB device drivers.  Converging
"usb_bus" and "usb_hcd" (later!) will be a bit easier too.

I did basic sanity tests, there's little to break ... :)
  
There are still a few functions in usb.c that aren't for
general driver use.  They're mostly for enumeration,
in areas where the hub driver and HCD root hubs
need to do various kinds of magic.  It wasn't clear
how to decouple those, they can certainly wait.

  
This fixes a bug in the audio driver which came from an
incorrect conversion from static to dynamic URB allocation.
It's against 2.5.5
  
I noticed this while trying to see exactly how ISO transfers
get used.  The bug is that while originally the driver statically
allocated several structures {urb + N * iso packet descriptors},
the update forgot to allocate the ISO descriptors.
    
Likely not many folk noticed this on 32 bit machines, where
sizeof urb == 92, because kmalloc rounds that up to 128,
adding 36 bytes of external padding.  The ISO descriptors
took up 32 bytes of that, which "just happened" to already
have been allocated but unused.

  
This is minor cleanup; pulls #includes out of files
that aren't intended to compile by themselves.

ehci bandwidth recording
  
Here's a minor update to the EHCI interrupt scheduler,
recording the bandwidth used by an URB for usbfs.

  
This restores a line someone deleted, which
affects hotplugging.  Basically this restores
correct/previous behavior:  the HID driver only
matches HID devices, not every device that
ever connects.

	- changed the minor number the auerswald driver was using, as it was found
	  out that this number was already in use by another USB driver!

Written by Irene Zubarev, Tong Yu, Jyoti Shah, Chuck Cole, and me.

	- changed proc entry creation to use the proper parent directory variable.

	- pcihpfs cleanup, removing unneeded file operations.
	- Added facility to have the files change their timestamps if the data
	  within the file changes.

into penguin.transmeta.com:/home/penguin/torvalds/repositories/kernel/linux

This patch allows my qlogic pcmcia scsi card to compile again, as broken
by the recent scsi layer changes. 

Despite davem's apparent suggestion that it'd be better to rewrite the
driver, i'd rather not _just_ right now :)

	/ Brett Pemberton

Numerous bugfixes brought forward from 2.4. I added some quick bio fixes
to mtdblock.c, which seemed to work for me whilst testing JFFS2 changes.

Forward ports from 2.4, Various janitor bits, and some fixes by me to
make the thing work again in 2.5.  I munged the MTDRAM driver to work
also (seperate patch to follow), and it seems to work.

David Woodhouse gave this the once over, and approved the changes.
Complete changelog below:

o   Don't create two slabcaches with the same name.
o   Don't corrupt eraseblock lists on mount
o   Don't mark nodes obsolete during mount
o   __attribute__((packed)) on the node definitions.
o   Fix up() without down() in jffs2_readdir().
o   Fix duplicate version number usage - s/highest_version++/++highest_version/
o   Fix (i.e. implement) mtime/ctime on directories.
    maybe too busy with the bk stuff
o   Don't allow hardlinks of directories.
o   s/(mode&S_IFMT)==S_IFLNK/S_ISLNK(mode)/ et al to keep Al happy.
o   Fix for garbage-collection of holes, where we used to write nodes out
    with csize/dsize swapped. Workarounds for existing such brokenness.
o   Improve wear levelling by rotating node lists on mount, to avoid starting
     at one end of the flash every time.
o   Remember to get internal inode-semaphore on symlink operations.

Enable NFS over TCP via config option

Limit number of active tcp connections to an RPC service

If a connection comes in and that results in number of
connections being more than 5 times the number of threads,
then we close a connection.

We randomly drop with the oldest or the newest connection.

Thus if we are flooded with connection requests, some will
get in and hopefully stay long enough to service at least one request.

Declare response sizes for nfs/lockd requests

This allows sndbuf reservation to be more accurate.

For lockd we just say "0" for now, meaning assume the max.
This could be improved, but it isn't critical.

Make sure there is alway adequate sndbuf space for replies.

We keep track of how much space might be needed for replies
and never dequeue a request unless there is adequate space
for a maximal reply.  We assume each request will generate a maximal
sized reply until the request is partly decoded.
Each RPC program/procedure can specify the maximum size
of a reply to the precedure (though they don't yet).

The wspace callback is used to enqueue sockets that may be waiting
for sndbuf space to become available.

As there should always be enough buffer space to the full
reply, the only reason that sock_sendmsg could block is due
to a kmalloc delay.  As this is likely to be fairly quick (and if
it isn't the server is clagged anyway) we remove the MSG_DONTWAIT
flag, but set a 30 second timeout on waiting.  If the wait
ever times out, we close the connection.  If it doesn't we can
be sure that we did a complete write.


When a request completes, we make sure that the space
used for the reply does not exceed the space reserved.  This
is an internal consistancy check.

This patchs sets the sndbuf and rcvbuf sizes for all sockets
used for rpc service.  This size if dependant on the servers bufsize (S) and
partially on the number of threads (N).

For UDP
    sndbuf == 5*S
    rcvbuf == (N+2)*S
for TCP
    sndbuf == N*S
    rcvbuf == 3*S

see code for rationale (in comments).

Close idle rpc/tcp sockets

We split the list of sv_allsocks into two, one
of permanent sockets (udp, tcp listener) and one
of temporary sockets (tcp data).

Whenever we complete a successful receive on a temp socket,
it gets pushed to the end of the list.

Whenever a thread wants to do something, it first checks if
the oldest temp socket has not has a receive for 6 mintutes
(should possibly be configurable).  It so, we simulate
a close.

Finally we make sure that threads wake up every few minutes
so that if the server is completely idle, all temp
sockets will get closed.

Detect and close tcp connections that we cannot work with.

If an rpc fragment that arrives on a tcp connection
is non-terminal or too large for our buffer, then we
have to close the connection.
Also, if a write fails on a tcp connection, we close
the connection.

Tidy up SMP locking for svc_sock

sk_lock is not necessary and is now removed.
The only things that were happening under sk_lock but
not the more global sv_lock were testing and setting
some of the flags: sk_busy, sk_conn, sk_data etc.

These have been changed to bits in a flags word which are atomically
set and tested.

Also, by establishing some simple rules about that must
be done after setting these flags, the locking is not needed.

With this patch sk_conn and sk_data are now flags, not counts (sk_data
was already a flag for udp).  They are set if there might be
a connection or data, and only clear when we are sure there aren't
(or when we are about to check if there is).

svc_sock_accepted becomes identical to svc_sock_recieved and
so is discarded in favour of the latter.

sk_rqstp was never used and is now gone.

Get sunrpc to use module_init properly

rpc_register_sysctl was called in several places.
Now only once.

Change sunrpc to use more list.h lists

The sunrpc client code uses home-grown doubly linked
lists to group
   - idle server threads
   - pending server sockets
   - waiting rpc tasks
   - all rpc tasks.

This patch converts all of these lists to <linux/list.h> lists and
also makes the list of all server sockets for a particular server into
a list.h list instead of a single-link list.

Possibly the least obvious change is replacing RPC_INIT_WAITQ
with RPC_WAITQ and INIT_RPC_WAITQ.  These follow the model of
LIST_HEAD and INIT_LIST_HEAD defined in list.h and are needed
to initialise the list_head in the rpc_waitq properly.

Remove last unneeded bit of BKL from knfsd

Remove BKL from nfsservctl systemcall.  All
components have their own locking.
Also remove it from the body of nfsd threads.  Keep
it for final thread destruction though.