When aac_command_thread detects an adapter event (AifCmdDriverNotify or
AifCmdEventNotify) it calls aac_handle_aif. This routine sets a flag,
calls fib_adapter_complete, and returns. The bad news is that after the
return, aac_command_thread continues to process the command and calls
fib_adapter_complete again.

Under some circumstances this causes the driver to take the device
offline. In my case, it happens with a Dell CERC SATA with a RAID 5 in
the "building" state:

aacraid: Host adapter reset request. SCSI hang ?
aacraid: Host adapter appears dead
scsi: Device offlined - not ready after error recovery: host 0 channel 0
id 0 lun 0
SCSI error : <0 0 0 0> return code = 0x6000000
end_request: I/O error, dev sda, sector 976537592

Mark Salyzyn says the intent is for aac_handle_aif to perform some
plug-n-play actions based on the adapter event, and return, leaving the
command completion to the caller.

The attached patch solves the problem by removing aac_handle_aif
entirely, since it is wrong, and there is currently no code in the
driver to actually do anything with these events.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

This patch

- removes support for 2.2.x and 2.4.x without the full dma API
  (<= 2.4.13 or 2.4.9rh)
- makes sure we don't acquire or release unessecary locks around
  ->scsi_done - it's a small BH/softirq that doesn't care about
  the callers lock state
- cleans up the 2.4 vs 2.6 compat code a little
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

gdth is the fourth-highest stack user (for a single function)
in 2.6.10-rc3-bk-recent (sizes on x86-32).

Reduce stack usage in gdth driver:
reduce ioc_rescan() from 1564 to 52 bytes;
reduce ioc_hdrlist() from 1528 to 24 bytes;
reduce gdth_get_info() from 1076 to 300 bytes;
Signed-off-by: Randy Dunlap <rddunlap@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Just tripped over a bug report for the SUSE kernel where gdth would
crash on a 32G opteron, turned out that the gdth_internal_copy() sg
handling was really buggy. After fixing this I wanted to do the same for
mainline, but I can see that a vain attempt was already made to fix it.
Unfortunately it wasn't complete, and on top of that there's room for
improvement.

The current code is buggy on highmem, as page_address() will not yield a
valid kernel address causing a NULL pointer dereference. The current
code also doesn't unmap the sg list if it sees a NULL sl->page. In fact,
the whole sg mapping looks really strange, why on earth would you be
mapping the sglist for dma when you are only going to copy from it?

This patch corrects both errors - correctly maps in the page, and kills
the pci_map_sg/pci_unmap_sg calls completely. If someone could test
this, that would be great.
Signed-off-by: Jens Axboe <axboe@suse.de>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

From: 	Willem Riede <osst@riede.org>

adds sysfs support to osst. This enables hotplug and udev to manage
the osst /dev nodes, which is a real necessity on installations that
use a dynamic /dev, such as Fedora Core 3.
signed-off-by: Willem Riede <osst@riede.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

From: 	Willem Riede <osst@riede.org>

important error handling improvements that I've made as the result of
problem reports.
signed-off-by: Willem Riede <osst@riede.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Make style changes that are the equivalent of recent changes to st,
such as using 'struct osst_tape' where we used to have 'OS_Scsi_Tape'
as a typedef. Osst behavior is not affected by this patch.
signed-off-by: Willem Riede <osst@riede.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Description: Fix error paths to handle SCSI targets
that reject SCSI aborts and resets and subsequently
complete SCSI commands.  There are targets in the field
that currently exhibit this behaviour, particularly in
the case of bad media (a CD drive got stuck for a LOOONG
time on a read op.)  We previously ignore the status
on aborts and resets under the mistaken belief that
whether they worked or not, the command response was
never going to show up.
Signed-off-by: Dave Boutcher <boutcher@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

This code has been problematic for a while and still contained a leg
where free_event_struct was called....followed by a reference to the
event_struct.  Restructure to make the code cleaner and fix the
dangling pointer reference.
Signed-off-by: Dave Boutcher <boutcher@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Fix a bug where we could fall out of our delay loop and then forget to
scan for drives.
Signed-off-by: Dave Boutcher <boutcher@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Description: Limit the size of I/O requests sent by the
ibmvscsi adapter.  With better I/O scheduling (and thus larger
requests) we were breaking some servers.

Updated based on comments from Jens Axboe and James
Bottomley to not specify max I/O sectors as a module
parameter, and subsequently not needlessly store
the value as a static variable.
Signed-off-by: Dave Boutcher <boutcher@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

There's been a problem reported where a WD Ultra3 drive reports that it
has an echo buffer of length 255 and then returns ILLEGAL REQUEST when
anyone tries to use it.  This causes DV to treat this as a retraining
error and eventually drop back to async.

The attached fix makes the DV code identify the ILLEGAL REQUEST
condition and configure the drive using the read only DV tests instead.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Some miscellaneous cleanups for the Zalon driver:

 - Remove unused definitions of sync_scsi_data_for_cpu and
   sync_scsi_data_for_device
 - Fill in dev->irq in the zalon driver
 - Request the interrupt in the name of the driver, not the bus address
 - Change the driver name to look better in sysfs
 - Call ncr53c8xx_exit() in zalon7xx_exit()
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Inline lasi700.h into lasi700.c to cut down on the size of the
drivers/scsi directory.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

sym2 version 2.1.18n:
 - Prevent querying for DT clocking on a single ended bus
 - Check the U3EN bit instead of the ULTRA3 bit
 - Only use PPR if SDTR is incapable of negotiating the desired options or
   speed
 - minsync bugfix (James Bottomley)
 - Always calculate what negotiation to perform inside sym_prepare_nego()
 - Delete unused SYM_OPT_HANDLE_IO_TIMEOUT and SYM_CONF_TIMEOUT_ORDER_MAX
   code (Christoph Hellwig)
 - Use SCSI-3 message names instead of SCSI-2 names
 - Remove private definitions of PCI IDs
 - Reorganise DMA mask setup
 - Fix comment tpyo
 - Make some needlessly global code static (Adrian Bunk)
 - Reorder some functions to eliminate predeclaration
 - Use memset instead of bzero
 - Consolidate and abstract SPARC's special IRQ printing
 - Convert hcb_p to struct sym_hcb *
 - Remove cam_ccb_p and cam_scsiio_p typedefs
 - Treat PA-RISC firmware as if it were a type of NVRAM
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Instead of doing an attribute override.

Ack'd by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Allow drivers to override reclaim_buffers in an OS-independent 
way by passing drm_device_t* as first parameter, like in the
BSD version.

From: Felix Kuehling <fxhuehl@gmx.de>
Signed-off-by: Dave Airlie <airlied@linux.ie>

From: Jon Smirl <jonsmirl@gmail.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>

From: Keith Withwell <keithw@tungstengraphics.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>

sparse complains about them:
drivers/char/drm/sis_ds.h:88:12: warning: dubious one-bit signed bitfield
drivers/char/drm/sis_ds.h:89:16: warning: dubious one-bit signed bitfield
Signed-off-by: Randy Dunlap <rddunlap@osdl.org>
Signed-off-by: Dave Airlie <airlied@linux.ie>

into starflyer.(none):/foo/airlied/bitkeeper/drm-latest

From: Stefan Dirsch <sndirsch@suse.de>
Signed-off-by: Dave Airlie <airlied@linux.ie>

into starflyer.(none):/foo/airlied/bitkeeper/drm-latest

into ppc970.osdl.org:/home/torvalds/v2.6/linux

Patch from Deepak Saxena

Patch adds implementation details for IXP46x CPUs and new IXDPG465
reference platform. 

Signed-off-by: Deepak Saxena
Signed-off-by: Russell King

Patch from Deepak Saxena

Supersedes 2226/1

According to the IXP Programmer's Reference Manual, a read from
IRQ_STATUS can only return '1' for IRQ sources that have been
explicitly enabled in IRQ_ENABLE. So if we never enable 'invalid'
IRQ sources, we don't actually have to mask off IRQ_STATUS with
IXP2000_VALID_IRQ_MASK.

Signed-off-by: Deepak Saxena
Signed-off-by: Russell King

into ppc970.osdl.org:/home/torvalds/v2.6/linux

From: Jason Gaston <jason.d.gaston@intel.com>

This patch adds the Intel ICH7 DID's to the pci_ids.h file
and updates the piix driver and related files for PATA support.

bart: this patch also adds PIRQ support
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>

- nothing clever here: the most noticeable change is the change of
  returned value for (*init_setup) in struct ide_pci_device_s which
  goes from void to int. Anything else is editing and checking for
  errors in the output of the compiler;
- pci_disable_device() added for the error path in pci_init_sgiioc4();
- BUG() removed in amd74xx_probe(): good old printk() is enough.
Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>

- no need to overwrite the status code returned by the pci_xyz() functions;
- jump into the new century and use DMA_32BIT_MASK;
- misc cleanup in the error paths. It should not add a huge value in
  ide_pci_enable() due to the FIXME comment but it should not bite too hard
  either.
Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>

ide_setup_pci_controller() puts the device in a PCI enabled state.
The patch adds a small helper to balance it when things go wrong.

Actually the helper does not *exactly* balance the setup: if it can
not do a better job, ide_setup_pci_controller() may only enable some
BARS whereas the new counterpart will try to disable everything.
Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>

- Change the return value and the prototype of do_ide_setup_pci_device
  Due to lack of appropriate return status code, the current clients of
  do_ide_setup_pci_device() can not distinguish a failing invocation
  from a successfull one. The patch modify do_ide_setup_pci_device() so
  that it can propagate some of the errors from the lower layers.

- Make ide_setup_pci_device() aware of the change and propagate the news
  itself. I only gave a quick sight to create_proc_ide_interfaces() (and
  ide_remove_proc_entries()) but they do seem sane and it should not matter
  if it fails or not.

- ide_setup_pci_devices(): mostly the same thing than ide_setup_pci_device().
  do not look trivially suspect.
Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>

From: Alan Cox <alan@redhat.com>

This patch updates ide/pci/generic.c to fix the incorrect returns
causing PCI devices to be left reserved wrongly by the driver.

From: Francois Romieu <romieu@fr.zoreil.com>

Use -ENODEV instead of -EAGAIN.
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>

From: Alan Cox <alan@lxorguk.ukuu.org.uk>

Several IDE drivers return positive values as errors in the PCI setup
code. Unfortunately the PCI layer considers positive values as success
so the driver skips the device but still claims it and things then go
downhill.

This fixes the IT8172 driver.

From: Francois Romieu <romieu@fr.zoreil.com>

Use -ENODEV instead of -EAGAIN.
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>

This fixes a build failure that happens when you don't select IPV6.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

DMA should be already released by ide_unregister()
(unless interface is still busy).
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>

/proc interface entries should be already unregistered
by ide_unregister() (unless interface is still busy).
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>

into bkbits.net:/repos/d/drm/drm-2.6