Ran into a couple small issues with the patch. First,
blk_queue_resize_tags wasn't getting exported, so I sent Jens a patch to
fix this, which he has now sent upstream. Second, the comment above
blk_queue_resize_tags says the queue_lock must be held when calling this
routine. Attached is an updated patch that grabs the lock. Other than
that, it looks good.

James Bottomley wrote:
> On Tue, 2004-08-03 at 10:50, Brian King wrote:
>>Currently, it is possible to call scsi_activate_tcq with a small queue depth,
>>then later call scsi_adjust_queue_depth with a larger queue depth. This results
>>in the scsi layer having a larger queue depth than the block layer knows about.
>>This results in these additional commands being issued as untagged ops rather than
>>tagged ops. This patch changes scsi_activate_tcq to call blk_queue_init_tags with
>>the maximum supported number of tags so this cannot occur.
>
>
> Sorry, been away at conferences with not enough time to remember what
> went on here.
>
> The reason it looks the way it does is historical...when the blk layer
> tcq interfaces were created, there was no way to resize the queue.  Jens
> later added resize (for me) and I forgot to incorporate it into the
> code.
>
> Another small point is that the max number of tags can be greater than
> 256.  256 is a SPI limit only (and even the qla1280, a SPI card which
> could use the tag as its global queue index would take > 256).  The
> limit in scsi_adjust_queue_depth has long since been obsoleted by our
> dynamic command allocation.
>
> I think the attached should work correctly (as long as it compiles...I
> coded it up on the flight home).
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

The non unique SCSI ioctls:

SCSI_IOCTL_SEND_COMMAND
SCSI_IOCTL_TEST_UNIT_READY
SCSI_IOCTL_BENCHMARK_COMMAND
SCSI_IOCTL_SYNC
SCSI_IOCTL_START_UNIT
SCSI_IOCTL_STOP_UNIT

Have been deprecated for a while now.  We should make the kernel
complain when a program actually uses one of them, so users have had
adequate warning before they are removed.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

drivers/scsi/megaraid/megaraid_mm.c: In function `mraid_mm_init':
drivers/scsi/megaraid/megaraid_mm.c:1069: `mraid_mm_compat_ioctl' undeclared (first use in this function)
drivers/scsi/megaraid/megaraid_mm.c:1069: (Each undeclared identifier is reported only once
drivers/scsi/megaraid/megaraid_mm.c:1069: for each function it appears in.)
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

make it universally include linux/ioctl32.h for
its compat ioctl.

This fixes a build failure on parisc64
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

From: Mukker, Atul <Atulm@lsil.com>

This is a complete re-write to replace the now obsoleted megaraid2
driver.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

This patch against scsi-misc-2.6 reorders a call to scsi_forget_host in
the scsi_remove_host function. This removes the error message
"Synchronizing SCSI cache..." on rmmod of a scsi host driver module as
shown in example output 1 and 2 below.

I had previous created a much more complicated patch to work around this
problem. In response Christoph mention a simpler solution (pointer to
thread below) which is this patch.

http://marc.theaimsgroup.com/?l=linux-scsi&m=108720042210331&w=2

Example output.

1.) Before patch.

elm:~# scsi1 : scsi_debug, version 1.73 [20040518], dev_size_mb=8,
opts=0x0
  Vendor: Linux     Model: scsi_debug        Rev: 0004
  Type:   Direct-Access                      ANSI SCSI revision: 03
SCSI device sdc: 16384 512-byte hdwr sectors (8 MB)
SCSI device sdc: drive cache: write back
 sdc: unknown partition table
Attached scsi disk sdc at scsi1, channel 0, id 0, lun 0
Attached scsi generic sg4 at scsi1, channel 0, id 0, lun 0,  type 0
Synchronizing SCSI cache for disk sdc: <4>FAILED
  status = 0, message = 00, host = 1, driver = 00

2.) After patch.

elm:~# scsi1 : scsi_debug, version 1.73 [20040518], dev_size_mb=8, opts=0x0
  Vendor: Linux     Model: scsi_debug        Rev: 0004
  Type:   Direct-Access                      ANSI SCSI revision: 03
SCSI device sdc: 16384 512-byte hdwr sectors (8 MB)
SCSI device sdc: drive cache: write back
 sdc: unknown partition table
Attached scsi disk sdc at scsi1, channel 0, id 0, lun 0
Attached scsi generic sg4 at scsi1, channel 0, id 0, lun 0,  type 0
Synchronizing SCSI cache for disk sdc:
Signed-off-by: Mike Anderson <andmike@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Highlights of this release:
- Patch provided by Christoph Hellwig to remove the isense code.
- Fix compile errors when CONFIG_PROC_FS=n.
- A fix for the module parameter "mptscsih" which was not being exported.
- The port of the 2.05.17 thru 2.05.23 of the lk 2.4 mpt driver.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Switch sd.c device number allocation to use idr, this limits memory
usage for lowend setups and highend setups are only limited by the
available names now.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

From: 	Kenn Humborg <kenn@linux.ie>

round line 2470 in NCR5380.c, if a command fails, and AUTOSENSE is on,
the current command is converted into a REQUEST_SENSE command and
put at the top of the request queue.  Part of this setup, is setting
up cmd->SCp to point at the sense_buffer, so that the received sense
data goes straight into the sense buffer.

Around line 1730, NCR5380_select() calls initialize_SCp() which ends
up clobbering cmd->SCp, and the sense data get written into the 
original request buffer.

This is what happens if cmd->use_sg = 0.  I haven't looked at what
happens if use_sg is non-zero, but I expect the effect will be the same.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Here are some changes from Adaptec for the AACRAID driver.

Change the pae_support flag to dac_support.  This was suggested sometime
last year and never made it in.
Fix some cast problems in pci_set_dma_mask.  Was casting the mask to
dma_addr_t when it should be a u64.
Fix SCp.ptr problem, should be SCp.dma_handle.

Applies to the scsi-misc bk tree.

Signed-off by: Mark Haverkamp <markh@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

> Andrew Vasquez wrote:
>
> >
> >This allocation should be done with GFP_ATOMIC flags.  The attached
> >patch should apply cleanly to any recent kernel
> >
>
> and seems to work fine.
>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

I got the following warning in 2.6.8-rc3-mm2:

<--  snip  -->

...
  CC      drivers/scsi/gdth.o
drivers/scsi/gdth.c:622:1: warning: "__devinitdata" redefined
In file included from include/linux/moduleparam.h:4,
                 from include/linux/module.h:20,
                 from drivers/scsi/gdth.c:375:
include/linux/init.h:227:1: warning: this is the location of the
previous definition
...

<--  snip  -->


The #define in question seems bogus, and the following patch simply
removes it:
Signed-off-by: Adrian Bunk <bunk@fs.tum.de>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Here is a patch from Adaptec for new adapter support.
Signed-off-by: Mark Haverkamp <markh@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

From: Nishanth Aravamudan <nacc@us.ibm.com>
Subject: [PATCH] scsi/ipr: replace schedule_timeout() with msleep()
Description: Use msleep() instead of schedule_timeout() to
guarantee the task delays for the desired time.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

The following patch adds an adapter response code to the
table in the driver to prevent an error from being logged
when this response is received.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

This fixes ipr to only allow eh initiated start_unit commands
following erp to disk devices.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

The following patch fixes a hang in i/o that could occur if an
adapter was ever declared "dead" by the ipr driver due to repeated
failed attempts to revive the card. The patch now allows ipr to fail
new commands when in this state rather than endlessly return
SCSI_MLQUEUE_HOST_BUSY.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

From: Mika Kukkonen <mika@osdl.org>
The following patch fixes some sparse warnings.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

This patch correctly enables tagged command queuing for the ipr driver.
I had misinterpreted scsi_mid_low_api.txt. The following patch has the
appropriate scsi_activate_tcq/scsi_deactivate_tcq calls.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

From: Alan Cox <alan@redhat.com>
Subject: PATCH: Fix assorted dma_addr_t typing errors in ipr driver
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Add a couple new devices supported by the ipr device driver. This patch
has already been sent to pciids-devel@lists.sourceforge.net.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

The following patch adds my email address to the block comment
at the start of each file of the ipr driver to make it easier to find.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

The following patch converts ipr to use a kref rather than a kobject to
do reference counting on an internal data structure. The added overhead
of a kobject is not needed.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

From: Francois Romieu <romieu@fr.zoreil.com>
Subject: [PATCH 2.6.7-mm3] ipr: minor fixes and assorted nit

- balance pci_enable_device() with pci_disable_device() where appropriate;
- pci_release_regions() replaces release_mem_region();
- ipr_alloc_mem() can not simply issue a call to ipr_free_mem() when
  something goes wrong as it would lead to pci_free_consistent() of
  unset data. Let ipr_alloc_mem() carefully release whatever it has
  allocated instead;
- no need to memset(..., 0, ...) an area returned by pci_alloc_consistent;
- ipr_probe_ioa:
  + DMA_32BIT_MASK for all;
  + error path rework (includes bug fix when ipr_alloc_mem fails);
- ipr_init() can fail: return adequate status code.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

This patch adds support for 2 new ipr adapters.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

The ipr device driver runs tagged queuing with QERR=1. When a check
condition occurs on a device running tagged queuing, the other queued
commands get aborted by the ipr driver and will be returned by the
adapter with the IPR_IOASC_ABORTED_CMD_TERM_BY_HOST response. These
commands should be retried by the midlayer and their retry counter not
decremented.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Bump driver version.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Currently, ABORT_TASK does not work properly on ipr adapters. It results in
timeouts and eh ends up getting escalated to eh_host_reset to recover.
The adapters are supposed to be fixed to properly handle this at some
point in the future, but until that updated microcode is available,
we need to send a cancel all to the device instead.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

From: Pawel Sikora <pluto@pld-linux.org>
The first parameter for sector_div must be sector_t rather than int.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

I might be wrong, but it looks like the attached patch has still not been
applied, although it is pretty important - it fixes error handling in
tmscsim. E.g., it was reported, that this patch fixes a problem with
tmscsim in 2.6.7 with a SCSI-3 scanner, that doesn't support REPORT LUNS
command.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Two scsi drivers do not compile on systems without ISA/PCI. Therefore
allyesconfig breaks on hardware like s390.
Signed-off-by: Christian Bornträger <linux-kernel@borntraeger.net>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Use correct __setup function type and parameters.
Return result from __setup function.
Signed-off-by: Randy Dunlap <rddunlap@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Use correct __setup function type and parameters.
Return result from __setup function.
Don't confuse <ints> and <interrupts>.
Signed-off-by: Randy Dunlap <rddunlap@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

The SCSI tree as included in 2.6.8-rc2-mm2 only removes the inline's
from the functions prototypes, but the part of my original patch that
also removes the inline's from the functions was lost.
Signed-off-by: Adrian Bunk <bunk@fs.tum.de>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Fix imm for IMM_DEBUG (changed device id struct).
Signed-off-by: Randy Dunlap <rddunlap@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

The patch below lets AIC7{9,X}XX_BUILD_FIRMWARE depend on
!PREVENT_FIRMWARE_BUILD.
Signed-off-by: Adrian Bunk <bunk@fs.tum.de>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

From: "J.A. Magallon" <jamagallon@able.es>

This allows to build aic with db4.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

From: Joe Korty <joe.korty@ccur.com>

Fix occasional PCI bus parity errors on the Dell PowerEdge 4600 during
boot.

Symptoms: The LCD display would turn orange and display "PCI SYSTEM E13F5",
and the following message would appear in /var/log/dmesg: "Uhhuh.  NMI
received.  Dazed and confused, but trying to continue".

By inserting a PCI card with a PDC20268 IDE controller and attaching to
that a Sony DRU-510A DVD RW burner with an unloaded tray, the failure can
be made to happen on every boot.

Cause: The aic7xxx driver was resetting the onboard AIC7891 SCSI controller
while waiting for a previous reset to complete.  This second reset confuses
the controller causing it to put bad data onto the PCI bus.

This is a backport of a RedHat 2.4.21-15.ELsmp fix.  A letter discussing
this problem, or one very close to it, may be found at:

   http://lists.us.dell.com/pipermail/linux-poweredge/2003-May/025010.htmlSigned-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

From: Andi Kleen <ak@muc.de>

Trivial gcc-3.5 build fixes.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>