From: Moore, Eric Dean <Emoore@lsil.com>

Here is list of fix's.
* added new PCI API support
* added ACPI support
* added CONFIG_LBA, READ16, WRITE16 support
* underun fix
* chain buffer free list not being init properly
* reduce task management 
        (abort=2sec,reset bus=5sec, timeout=10sec)
* Hot plug fix's requested from Christoph Hellwig, and several
others from the linux-scsi@ list.

> This patch is against the sg driver found in lk 2.6.1-bk6 .
> It is an update on the sg patch sent in the "devices with more
> than one node" thread:
> http://marc.theaimsgroup.com/?l=linux-scsi&m=107415435300783&w=2

Attached is an updated patch adding the cdev_unmap() call
which Kai Makisara reported as required.

Noticed by: Guennadi Liakhovetski <g.liakhovetski@gmx.de>
Fixed By: Juergen E. Fischer <fischer@linux-buechse.de>

On Sat, Jan 24, 2004 at 15:02:12 +0100, Guennadi Liakhovetski wrote:
> Correct and forgive me if I am wrong, it looks like you are now poking at
> ports without requesting them - in *_porttest. Is it a good idea?

Ouch.  No, of course not, you are right.  Thanks for pointing.  I
thought I had moved the port testing to aha152x_probe_one...

Following patch cleans that up.

The port test during autoconfiguration was also done without requesting
the region first and there was a bug in auto configuration of a TC1550
controller (both even before the patch).

attached my patch for the aha152x driver.  Tested for pcmcia and not
pcmcia and various devices.

changes in the aha152x driver:
- gather code that is not used by PCMCIA at the end
- move request_region for !PCMCIA case to detection
- migration to new scsi host api (remove legacy code)
- free host scribble before scsi_done
- fix error handling
- one isapnp device added to id_table

and in the pcmcia stub module:
- default to synchronous transfers
- release_region hack removed
- let core module do the scsi host api calls.

The patch at the end of this message fixes the following problems:

- cdev allocation moved outside st_dev_arr_lock, prevents the debugging
error messages reported by Mike and Mike
- cdev_unmap() added before cdev_del(), prevents oops (and kernel data
corruption) in case someone tries to use a device after removing the
module
- better error handling for failures in cdev allocation

As far as I am concerned, the patch is tested but testing by others might
be useful :-)

into hopper.(none):/usr/src/kernel/linux-2.6

into home.osdl.org:/home/torvalds/v2.5/linux

into home.osdl.org:/home/torvalds/v2.5/linux

tty->driver_data or state->port may end up being NULL in uart_close.
Make sure that we correctly clean up in this case, rather than
oopsing.

There is no way that tty can be NULL in uart_put_char() and
uart_write().  Eliminate these redundant tests.

From: Jim Collette <jim@hamachi.net>

There's an exit path in i8042_interrupt() which forgets to rearm the timer.
It can make the mouse die when X is started.

into home.osdl.org:/home/torvalds/v2.5/linux

into home.osdl.org:/home/torvalds/v2.5/linux

Current driver doesn't do shared irq properly.  When testing on
a laptop here irq 3 get shared between pcmcia slot and tty/IRDA

Andreas Happe <andreashappe@gmx.net> writes:
> my notebook (hp/compaq nx7000) still crashes when using 8139cp (runs
> rock solid with 8139too driver). The computer just locks up, there is no
> dmesg output. This has happened since I've got this laptop (around
> november '03).

It seems 8139cp.c has the race condition of rx_poll and interrupt.

NOTE, since I don't have this device, patch is untested. Sorry.

into kernel.bkbits.net:/home/davem/irda-2.6

into home.osdl.org:/home/torvalds/v2.5/linux

I got a Wacom Graphire3 for my birthday and unfortunately it didn't
work. After some playing around, I noticed the 2.6 kernel needs a few
small modifications to make it work.

This simple patch adds support for the Wacom Graphire 3.

into home.osdl.org:/home/torvalds/v2.5/linux

Fix this:

drivers/scsi/aha1542.c:74: warning: unsigned int format, different type arg (arg 5)

From: 	Justin T. Gibbs <gibbs@scsiguy.com>

From: 	Petri Koistinen <petri.koistinen@iki.fi>

Here is an update to use pci_set_consistent_dma_mask.  But since
dma_alloc_coherent uses dma_mask instead of consistent one, I left in
setting dma_mask as well until the alloc routine changes.

Avoid a possible timer deletion race.

From: James Cross <jscross@veritas.com>

The RAW_GETBIND compatibility ioctl call does convert properly between the
32bit/64bit version of raw_config_request due to a trivial error, and the
ioctl call fails.

From: Dave Jones <davej@redhat.com>

Fix typo in the recent 2.4 DMI sync-up.

From: Jean Delvare <khali@linux-fr.org>

There are 24 members of this array.

From: Bartlomiej Zolnierkiewicz <B.Zolnierkiewicz@elka.pw.edu.pl>

From: Benjamin Herrenschmidt <benh@kernel.crashing.org>

Fix initialization of the PDC20270/1 chipsets on the Xserve Apple machines.

From: Arkadiusz Miskiewicz <arekm@pld-linux.org>

Both drivers/ide/legacy and drivers/ide/ppc broke because they can no
longer include drivers/ide/timing.h.  Fix.

(Acked by Bart)

x86-64 using drivers/Kconfig requires some minor changes. Mostly to disable drivers
that do not work.

- Mark paride bpck6 not 64bit clean
- Disable I2O on 64bit
- Mark PNP dependent on ISA
- Mark NSP32 driver as not 64bit clean

Mainly lots of bug fixes and a few minor features. One change is that
it uses drivers/Kconfig now like i386. This requires a few minor changes in
outside Kconfig files which I am sending separately.

 - Tighten/fix some code in NUMA node discovery
 - Fix oopses in threaded 32bit coredumps and read correct registers.
 - Merge with 2.6.2rc1
 - Sync arch/x86_64/Kconfig with i386. Uses drivers/Kconfig now.
 - Remove bcopy export
 - Fix check for signal stack for 32bit signals
 - Fix bcopy and exit prototypes for gcc 3.4
 - Fix asm contraint in usercopy.c for gcc 3.4
 - Use rt_sigreturn, not sigreturn for rt sigreturns.
 - Pass si_fault address to 32bit
 - Truncate si_error to 16bit in 32bit emulation to match i386
 - Move IA32 flag switching for 32bit executables to flush_thread
   (code copied from ppc64/sparc64)
 - Print exception trace for strace too, share code.
 - Default to 3GB address space for a.out executables
 - Fix security hole in ptrace. Also fixes some problems with 32bit gdb.
 - Sync mmap address selection algorithm with mm/mmap.c version
 - Disable a.out coredumps completely
 - Fix bug in sigaltstack 32bit emulation. Kylix IDE now works.
 - Move errata 93/BIOS workaround into fault handler.  This should work
   around USB legacy BIOS bugs too, although not completely (we cannot fix
   faults injected by SMM into user space 64bit processes)
 - Quieten some unimplemented 32bit syscall warnings and avoid repeated
   warnings.
 - Set LDT segment limit correctly (fixes problems with some modify_ldt
   user)
 - Remove obsolete ldt rw lock.
 - Remove sys32_modify_ldt. The standard sys_modify_ldt is equivalent.
 - Remove traces of old kgdb support
 - Merge CFI changes from Jim Houston and some other smaller changes The
   kernel assembly functions are described with dwarf2 unwind
   information now, which makes it easier for debuggers to make sense of
   stack backtraces.  The code is only enabled with CONFIG_DEBUG_INFO.
   Note this implies that when you use CONFIG_DEBUG_INFO you may need
   an binutils update.
 - defconfig updated
 - Readd sleep support code (Pavel Machek)
 - Drop fusion and flush workarounds from IOMMU code
 - Add iommu=nofullflush option
 - Rewrite 32bit emulation for siginfo conversion (Joe Korty)
 - Allow remapping of scatterlists after unmap. This fixes some problems
   with the SCSI layer retrying previously mapped sg lists when iommu
   merging was enabled (it's disabled now by default)
 - Port HPET rtc device emulation code from i386

into home.osdl.org:/home/torvalds/v2.5/linux

When sending a multicast and using looping back a copy to the
local machine, the interface filter checks can be done before the
source address is specified. For an INCLUDE filter, this won't match
the allowed sources and the packets won't be delivered locally,
even when the ultimate source address chosen is in the allowed list.

The patch below fixes the filter checks for both IGMPv3 and MLDv2
to only apply when a source address is available.

Thanks to Steven Hessing for reporting the problem and providing
a test case for reproducing it.