openwrt/staging/blogic.git
19 years agoMerge by hand (conflicts in sd.c)
James Bottomley [Tue, 6 Sep 2005 22:52:54 +0000 (17:52 -0500)]
Merge by hand (conflicts in sd.c)

19 years ago[SCSI] quieten messages on scsi_execute commands
James Bottomley [Sun, 4 Sep 2005 16:32:05 +0000 (11:32 -0500)]
[SCSI] quieten messages on scsi_execute commands

scsi_io_completion() can be a bit noisy about certain conditions.
Previously this wasn't a problem for internally generated commands,
since they never hit it.  However, since we do all SCSI commands via
bios, now they do.  user CD testers like magicdev are now getting not
ready messages every time they touch the CD to see if there's anything
in it.

Fix this by making all scsi_execute commands REQ_QUIET and making
scsi_finish_io() not say anything for REQ_QUIET.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] unexport scsi_add_timer/scsi_delete_timer
Christoph Hellwig [Tue, 6 Sep 2005 12:15:37 +0000 (14:15 +0200)]
[SCSI] unexport scsi_add_timer/scsi_delete_timer

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] switch EH thread startup to the kthread API
Christoph Hellwig [Tue, 6 Sep 2005 12:04:26 +0000 (14:04 +0200)]
[SCSI] switch EH thread startup to the kthread API

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] fix SCSI_IOCTL_PROBE_HOST
Christoph Hellwig [Tue, 6 Sep 2005 12:03:44 +0000 (14:03 +0200)]
[SCSI] fix SCSI_IOCTL_PROBE_HOST

This returns always false with new-style drivers right now.  Make it
return always true instead, as a host must be present if we are able
to call the ioctl (without a host attached there would be no device
node to call on..)

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] Universal Xport no attach blacklist
Anton Blanchard [Tue, 6 Sep 2005 08:08:14 +0000 (18:08 +1000)]
[SCSI] Universal Xport no attach blacklist

On Fri, Dec 13, 2002 at 12:24:39AM +1100, Anton Blanchard wrote:

> We tested 2.5.51 on a ppc64 box, qlogic 2312 and a fastt700 array. I
> had CONFIG_SCSI_REPORT_LUNS and unfortunately it thought the management
> LUN was a disk:
>
>   Vendor: IBM       Model: Universal Xport   Rev: 0520
>   Type:   Direct-Access                      ANSI SCSI revision: 03
>
> ...
>
> SCSI device sdaj: drive cache: write through
> SCSI device sdaj: 40960 512-byte hdwr sectors (21 MB)
>  sdaj: unknown partition table
> Attached scsi disk sdaj at scsi2, channel 0, id 0, lun 31
>
> ...
>
> end_request: I/O error, dev sdaj, sector 0

Three years later...

It looks like SGI use the same FC vendor and they already have a
workaround for this issue. The following patch adds the IBM version of
it.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] sd: pause in sd_spinup_disk for slow USB devices
Alan Stern [Tue, 12 Jul 2005 14:45:17 +0000 (10:45 -0400)]
[SCSI] sd: pause in sd_spinup_disk for slow USB devices

This patch adds a delay tailored for USB flash devices that are slow to
initialize their firmware.  The symptom is a repeated Unit Attention with
ASC=0x28 (Not Ready to Ready transition).  The patch will wait for up to 5
seconds for such devices to become ready.  Normal devices won't send the
repeated Unit Attention sense key and hence won't trigger the patch.

This fixes a problem with James Roberts-Thomson's USB device, and I've
seen several reports of other devices exhibiting the same symptoms --
presumably they will be helped as well.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] return success after retries in scsi_eh_tur
Alan Stern [Wed, 30 Mar 2005 20:05:45 +0000 (15:05 -0500)]
[SCSI] return success after retries in scsi_eh_tur

The problem lies in the way the error handler uses TEST UNIT READY to
tell whether error recovery has succeeded.  The scsi_eh_tur function
gives up after one round of retrying; after that it decides that more
error recovery is needed.

However TUR is liable to report sense data indicating a retry is needed
when in fact error recovery has succeeded.  A typical example might be
SK=2, ASC=4, ASCQ=1 (Logical unit in process of becoming ready).  The mere
fact that we were able to get a sensible reply to the TUR should indicate
that the device is working well enough to stop error recovery.

I ran across a case back in January where this happened.  A CD-ROM drive
timed out the INQUIRY command, and a device reset fixed the blockage.
But then the drive kept responding with 2/4/1 -- because it was spinning
up I suppose -- until the error handler gave up and placed it offline.
If the initial INQUIRY had received the 2/4/1 instead, everything would
have worked okay.  It doesn't seem reasonable for things to fail just
because the error handler had started running.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] ibmvscsi: handle large scatter/gather lists
James Bottomley [Tue, 6 Sep 2005 22:11:54 +0000 (17:11 -0500)]
[SCSI] ibmvscsi: handle large scatter/gather lists

The maximum size of a scatter-gather list that the current IBM VSCSI
Client can handle is 10.  This patch adds large scatter-gather support
to the client so that it is capable of handling up to SG_ALL(255)
number of requests in the scatter-gather list.

Signed-off-by: Linda Xie <lxie@us.ibm.com>
Acked by: Dave C Boutcher <sleddog@us.ibm.com>

Rejections fixed up and
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] qla2xxx: Update version number to 8.01.00-k.
Andrew Vasquez [Sat, 27 Aug 2005 02:10:30 +0000 (19:10 -0700)]
[SCSI] qla2xxx: Update version number to 8.01.00-k.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] qla2xxx: Stop firmware execution at unintialization time.
Andrew Vasquez [Sat, 27 Aug 2005 02:10:20 +0000 (19:10 -0700)]
[SCSI] qla2xxx: Stop firmware execution at unintialization time.

On ISP24xx parts, stop execution of firmware during ISP
tear-down.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] qla2xxx: Replace schedule_timeout().
Andrew Vasquez [Sat, 27 Aug 2005 02:10:10 +0000 (19:10 -0700)]
[SCSI] qla2xxx: Replace schedule_timeout().

From: Nishanth Aravamudan <nacc@us.ibm.com>

Replace schedule_timeout() with
msleep()/msleep_interruptible() as appropriate, to guarantee the task
delays as expected.

Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] qla2xxx: Remove bad call to fc_remove_host() during probe failure.
Andrew Vasquez [Sat, 27 Aug 2005 02:10:00 +0000 (19:10 -0700)]
[SCSI] qla2xxx: Remove bad call to fc_remove_host() during probe failure.

fc_remove_host() should only be called after a scsi_host has
been successfully added via scsi_add_host() -- any failures
while qla2xxx probing would result in an incorrect call to
fc_remove_host() during cleanup.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] qla2xxx: Add host attributes.
Andrew Vasquez [Sat, 27 Aug 2005 02:09:40 +0000 (19:09 -0700)]
[SCSI] qla2xxx: Add host attributes.

Export additional host information via the shost_attrs member in
the scsi_host template.  Attributes include: driver version,
firmware version, ISP serial number, ISP type, ISP product ID,
HBA model name, HBA model description, PCI interconnect
information, and HBA port state.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] qla2xxx: Add change_queue_depth/type() API support.
Andrew Vasquez [Sat, 27 Aug 2005 02:09:30 +0000 (19:09 -0700)]
[SCSI] qla2xxx: Add change_queue_depth/type() API support.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] qla2xxx: Remove redundant call to pci_unmap_sg().
Andrew Vasquez [Sat, 27 Aug 2005 02:09:20 +0000 (19:09 -0700)]
[SCSI] qla2xxx: Remove redundant call to pci_unmap_sg().

In a corner-case failure where the request-q does not
contain enough entries for a given request, pci_unmap_sg()
would be called twice.  Remove direct call and let the
failure-path logic handle the unmapping.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] qla2xxx: Remove RISC pause/release barriers during flash manipulation.
Andrew Vasquez [Sat, 27 Aug 2005 02:09:10 +0000 (19:09 -0700)]
[SCSI] qla2xxx: Remove RISC pause/release barriers during flash manipulation.

Remove unnecessary RISC pause/release barriers during
ISP24xx flash manipulation.  The ISP24xx can arbitrate flash
access requests during RISC executions.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] qla2xxx: Correct LED scheme definition.
Andrew Vasquez [Sat, 27 Aug 2005 02:09:00 +0000 (19:09 -0700)]
[SCSI] qla2xxx: Correct LED scheme definition.

Original implementation used an overloaded bit in the EFI
parameters.  The correct bit is BIT_4 of the special_options
section of NVRAM.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] qla2xxx: Simplify redundant target/device reset logic.
Andrew Vasquez [Sat, 27 Aug 2005 02:08:50 +0000 (19:08 -0700)]
[SCSI] qla2xxx: Simplify redundant target/device reset logic.

Remove redundant qla2x00_target_reset() function in favour of
the equivalent qla2x00_device_reset().  Update callers of
old function.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] qla2xxx: Correct domain/area exclusion logic.
Andrew Vasquez [Sat, 27 Aug 2005 02:08:40 +0000 (19:08 -0700)]
[SCSI] qla2xxx: Correct domain/area exclusion logic.

In an FL topology, limit port recognition to those devices
not within the same area and domain of the ISP.  The
firmware will recogonize such devices during local-loop
discovery.

Some devices may respond to a PLOGI before they have
completed their fabric login or they may not be a public
device. In this case they will report:

        domain == 00
        area == 00
        alpa == <XX>

which is valid. Exclude such devices from local loop
discovery.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] qla2xxx: Add FDMI support.
Andrew Vasquez [Sat, 27 Aug 2005 02:08:30 +0000 (19:08 -0700)]
[SCSI] qla2xxx: Add FDMI support.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] qla2xxx: Export class-of-service (COS) information.
Andrew Vasquez [Sat, 27 Aug 2005 02:08:10 +0000 (19:08 -0700)]
[SCSI] qla2xxx: Export class-of-service (COS) information.

Export COS information for the fc_host and fc_remote_port
objects added by the driver.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] qla2xxx: Use dma_get_required_mask() in determining the 'ideal' DMA mask.
Andrew Vasquez [Sat, 27 Aug 2005 02:08:00 +0000 (19:08 -0700)]
[SCSI] qla2xxx: Use dma_get_required_mask() in determining the 'ideal' DMA mask.

In order to efficiently utilise the ISP's IOCB
request-queue, use the dma_get_required_mask() function to
determine the use of command-type 2 or 3 IOCBs when queueing
SCSI commands.  This applies to ISP2[123]xx chips only, as
the ISP24xx uses command-type 7 IOCBs which use 64bit DSDs.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] fusion: whitespace fixes
Christoph Hellwig [Thu, 18 Aug 2005 14:24:53 +0000 (16:24 +0200)]
[SCSI] fusion: whitespace fixes

Acked by: Moore, Eric Dean <Eric.Moore@lsil.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] fusion: endianess fixes
Christoph Hellwig [Thu, 18 Aug 2005 14:25:44 +0000 (16:25 +0200)]
[SCSI] fusion: endianess fixes

Assorted endianess fixes.  I'll work on full endianess annotations
later.

Acked by: Moore, Eric Dean <Eric.Moore@lsil.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] fusion: update LSI headers
Christoph Hellwig [Thu, 18 Aug 2005 14:24:26 +0000 (16:24 +0200)]
[SCSI] fusion: update LSI headers

Acked by: Moore, Eric Dean <Eric.Moore@lsil.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] fusion: extended config header support
Christoph Hellwig [Thu, 18 Aug 2005 14:26:15 +0000 (16:26 +0200)]
[SCSI] fusion: extended config header support

Acked by: Moore, Eric Dean <Eric.Moore@lsil.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] aic7xxx: remove aiclib.c
Christoph Hellwig [Fri, 19 Aug 2005 16:57:13 +0000 (18:57 +0200)]
[SCSI] aic7xxx: remove aiclib.c

#include of C files and macro tricks to rename symbols are evil and just
cause trouble.  Let's doublicate the two functions as they're going to
go away soon enough anyway.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] aacraid: bad BUG_ON fix
Mark Haverkamp [Thu, 1 Sep 2005 15:19:23 +0000 (08:19 -0700)]
[SCSI] aacraid: bad BUG_ON fix

This was noticed by Doug Bazamic and the fix found by Mark Salyzyn at
Adaptec.

There was an error in the BUG_ON() statement that validated the
calculated fib size which can cause the driver to panic.

Signed-off-by: Mark Haverkamp <markh@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] sg direct io/mmap oops, st sync
Douglas Gilbert [Thu, 1 Sep 2005 11:50:02 +0000 (21:50 +1000)]
[SCSI] sg direct io/mmap oops, st sync

This patch adopts the same solution as proposed by Kai M. in
a post titled: "[PATCH] SCSI tape signed/unsigned fix".
The fix is in a function that the sg driver borrowed from
the st driver so its maintenance is a little easier if
the functions remain the same after the fix.

   - change nr_pages type from unsigned to signed so errors
     from get_user_pages() call are properly handled

Signed-off-by: Douglas Gilbert <dougg@torque.net>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years agoThis patch fixes in st.c the bug in the signed/unsigned int comparison
James Bottomley [Thu, 1 Sep 2005 00:27:22 +0000 (20:27 -0400)]
This patch fixes in st.c the bug in the signed/unsigned int comparison
reported by Doug Gilbert and fixed by him in sg.c (see [PATCH] sg direct
io/mmap oops). Doug fixed the comparison in sg.c. This fix for st.c does not
touch the comparison but makes both arguments signed to remove the
problem. The new code is adapted from linux/fs/bio.c.

Signed-off-by: Kai Makisara <kai.makisara@kolumbus.fi>
Rejections fixed up and

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] embryonic RAID class
James Bottomley [Tue, 16 Aug 2005 23:27:34 +0000 (18:27 -0500)]
[SCSI] embryonic RAID class

The idea behind a RAID class is to provide a uniform interface to all
RAID subsystems (both hardware and software) in the kernel.

To do that, I've made this class a transport class that's entirely
subsystem independent (although the matching routines have to match per
subsystem, as you'll see looking at the code).  I put it in the scsi
subdirectory purely because I needed somewhere to play with it, but it's
not a scsi specific module.

I used a fusion raid card as the test bed for this; with that kind of
card, this is the type of class output you get:

jejb@titanic> ls -l /sys/class/raid_devices/20\:0\:0\:0/
total 0
lrwxrwxrwx  1 root root     0 Aug 16 17:21 component-0 -> ../../../devices/pci0000:80/0000:80:04.0/host20/target20:1:0/20:1:0:0/
lrwxrwxrwx  1 root root     0 Aug 16 17:21 component-1 -> ../../../devices/pci0000:80/0000:80:04.0/host20/target20:1:1/20:1:1:0/
lrwxrwxrwx  1 root root     0 Aug 16 17:21 device -> ../../../devices/pci0000:80/0000:80:04.0/host20/target20:0:0/20:0:0:0/
-r--r--r--  1 root root 16384 Aug 16 17:21 level
-r--r--r--  1 root root 16384 Aug 16 17:21 resync
-r--r--r--  1 root root 16384 Aug 16 17:21 state

So it's really simple: for a SCSI device representing a hardware raid,
it shows the raid level, the array state, the resync % complete (if the
state is resyncing) and the underlying components of the RAID (these are
exposed in fusion on the virtual channel 1).

As you can see, this type of information can be exported by almost
anything, including software raid.

The more difficult trick, of course, is going to be getting it to
perform configuration type actions with writable attributes.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] attribute container final klist fixes
James Bottomley [Sun, 28 Aug 2005 14:13:17 +0000 (09:13 -0500)]
[SCSI] attribute container final klist fixes

Since the attribute container deletes from a klist while it's walking
it, it is vulnerable to the problem (and fix) here:

http://marc.theaimsgroup.com/?l=linux-scsi&m=112485448830217

The attached fixes this (but won't compile without the above).

It also fixes the logical reversal in the traversal loop which meant
that we were never actually traversing the loop to hit this bug in the
first place.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] correct attribute_container list usage
James Bottomley [Mon, 22 Aug 2005 15:06:19 +0000 (10:06 -0500)]
[SCSI] correct attribute_container list usage

One of the changes in the attribute_container code in the scsi-misc tree
was to add a lock to protect the list of devices per container.  This,
unfortunately, leads to potential scheduling while atomic problems if
there's a sleep in the function called by a trigger.

The correct solution is to use the kernel klist infrastructure instead
which allows lockless traversal of a list.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years agoMerge by hand (conflicts in sr.c)
James Bottomley [Sun, 28 Aug 2005 16:40:00 +0000 (11:40 -0500)]
Merge by hand (conflicts in sr.c)

19 years ago[SCSI] fix sense buffer length handling problem
James Bottomley [Tue, 9 Aug 2005 16:55:36 +0000 (11:55 -0500)]
[SCSI] fix sense buffer length handling problem

The new bio code was incorrectly converted from stack allocated to
kmalloc'd buffer handling.  There are two places where it incorrectly
uses sizeof(*sense) to get the size of the sense buffer.  This
actually produces one, so no sense data was ever getting back, causing
failure in things like disk spin up.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] fix 3ware raid emulated commands
James Bottomley [Tue, 28 Jun 2005 14:18:21 +0000 (09:18 -0500)]
[SCSI] fix 3ware raid emulated commands

The 3ware emulated commands all expect they are executing in the
use_sg == 0 case, which isn't true either in the block layer rework or
an SG_IO ioctl.

Fix this by adding the correct kmapping of the first element in the sg
list.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] comment cleanup for spi_execute
Christoph Hellwig [Sun, 19 Jun 2005 09:32:53 +0000 (11:32 +0200)]
[SCSI] comment cleanup for spi_execute

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] fix C syntax problem in scsi_lib.c
akpm@osdl.org [Sun, 26 Jun 2005 07:12:51 +0000 (00:12 -0700)]
[SCSI] fix C syntax problem in scsi_lib.c

Older gcc's require variable definitions at the beginning of a block.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] convert ch to use scsi_execute_req
James Bottomley [Mon, 13 Jun 2005 03:37:10 +0000 (22:37 -0500)]
[SCSI] convert ch to use scsi_execute_req

I also tinkered with it's sense recognition routines to make them take
scsi_sense_hdr structures instead of raw sense data.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] convert sr to scsi_execute_req
James Bottomley [Mon, 13 Jun 2005 03:21:29 +0000 (22:21 -0500)]
[SCSI] convert sr to scsi_execute_req

This follows almost the identical model to sd, except that there's one
ioctl which returns raw sense data, so it had to use scsi_execute()
instead.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] convert sd to scsi_execute_req (and update the scsi_execute_req API)
James Bottomley [Sun, 28 Aug 2005 16:33:52 +0000 (11:33 -0500)]
[SCSI] convert sd to scsi_execute_req (and update the scsi_execute_req API)

This one removes struct scsi_request entirely from sd.  In the process,
I noticed we have no callers of scsi_wait_req who don't immediately
normalise the sense, so I updated the API to make it take a struct
scsi_sense_hdr instead of simply a big sense buffer.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] convert SPI transport class to scsi_execute
James Bottomley [Sun, 28 Aug 2005 16:31:14 +0000 (11:31 -0500)]
[SCSI] convert SPI transport class to scsi_execute

This one's slightly more difficult.  The transport class uses
REQ_FAILFAST, so another interface (scsi_execute) had to be invented to
take the extra flag.  Also, the sense functions are shifted around to
allow spi_execute to place data directly into a struct scsi_sense_hdr.
With this change, there's probably a lot of unnecessary sense buffer
allocation going on which we can fix later.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] convert the remaining mid-layer pieces to scsi_execute_req
James Bottomley [Sun, 28 Aug 2005 16:27:01 +0000 (11:27 -0500)]
[SCSI] convert the remaining mid-layer pieces to scsi_execute_req

After this, we just have some drivers, all the ULDs and the SPI
transport class using scsi_wait_req().

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years agoMerge HEAD from ../scsi-misc-2.6-tmp
James Bottomley [Sun, 28 Aug 2005 16:18:35 +0000 (11:18 -0500)]
Merge HEAD from ../scsi-misc-2.6-tmp

19 years ago[SCSI] sr.c: Fix getting wrong size
Pete Zaitcev [Wed, 6 Jul 2005 01:18:08 +0000 (18:18 -0700)]
[SCSI] sr.c: Fix getting wrong size

Here's the problem. Try to do this on 2.6.12:
- Kill udev and HAL
- Insert a CD-ROM into a SCSI or USB CD-ROM drive
- Run dd if=/dev/scd0
- cat /sys/block/sr0/size
- Eject the CD, insert a different one
- Run dd if=/dev/scd0
This is likely to do "access beyond the end of device", if you let it
- cat /sys/block/sr0/size
This shows the size of a previous CD, even though dd was supposed
to revalidate the device.
- Run dd if=/dev/scd0
The second run of dd works correctly!

The bug was introduced in 2.5.31, when Al fixes the recursive opens
in partitioning. Before, the code worked like this:
- Block layer called cdrom_open directly
- cdrom_open called sr_open
- sr_open called check_disk_change
- check_disk_change called sr_media_change
- sr_media_change did cd->needs_disk_change=1
- before returning sr_open tested cd->needs_disk_change
  and called get_sector_size.

In 2.6.12, the check_disk_change is called from cdrom_open only. Thus:
- Block layer calls sr_bd_open
- sr_bd_open calls cdrom_open
- cdrom_open calls sr_open
- sr_open tests cd->needs_disk_change, which wasn't set yet; returns
- cdrom_open calls check_disk_change
- check_disk_change calls sr_media_change
- sr_media_change does cd->needs_disk_change=1, but nobody cares

Acked by: Alexander Viro <aviro@redhat.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] ibmvscsi timeout fix
Dave C Boutcher [Mon, 22 Aug 2005 19:38:26 +0000 (14:38 -0500)]
[SCSI] ibmvscsi timeout fix

This patch fixes a long term borkenness in
ibmvscsi where we were using the wrong timeout
field from the scsi command (and using the
wrong units.)  Now broken by the fact that the
scsi_cmnd timeout field is gone entirely.
This only worked before because all the SCSI
targets assumed that 0 was default.

Signed-off-by: Dave Boutcher <boutcher@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] drivers/scsi/constants.c should include scsi_dbg.h
Adrian Bunk [Sat, 20 Aug 2005 19:27:27 +0000 (21:27 +0200)]
[SCSI] drivers/scsi/constants.c should include scsi_dbg.h

C files should include the files with the prototypes for their global
functions.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] add missing attribute container function prototype
James Bottomley [Tue, 16 Aug 2005 22:26:10 +0000 (17:26 -0500)]
[SCSI] add missing attribute container function prototype

attribute_container_classdev_to_container is an exported function of the
attribute_container.c file.  However, there's no prototype for it.  Now
I actually want to use it, so add one.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] ibmvscsi eh locking
Dave C Boutcher [Mon, 15 Aug 2005 21:52:58 +0000 (16:52 -0500)]
[SCSI] ibmvscsi eh locking

With the removal of the spinlocking around eh calls, we need to add a
little more locking back in, otherwise we do some naked list
manipulation.

Signed-off-by: Dave Boutcher <boutcher@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] aacraid: Fix aacraid probe breakage (updated)
Mark Haverkamp [Mon, 15 Aug 2005 17:50:24 +0000 (10:50 -0700)]
[SCSI] aacraid: Fix aacraid probe breakage (updated)

This patch fixes the bad assumption of the aacraid driver with use_sg.
I used the 3w-xxxx driver fix as a guide for this.

Signed-off-by: Mark Haverkamp <markh@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] fix transport class corner case after rework
James Bottomley [Mon, 15 Aug 2005 21:13:19 +0000 (16:13 -0500)]
[SCSI] fix transport class corner case after rework

If your transport class sets the ATTRIBUTE_CONTAINER_NO_CLASSDEVS flag,
then its configure method never gets called.  This patch fixes that so
that the configure method is called with a NULL classdev.

Also remove a spurious inverted comma in the transport_class comments.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] use scatter lists for all block pc requests and simplify hw handlers
James Bottomley [Wed, 15 Jun 2005 23:48:29 +0000 (18:48 -0500)]
[SCSI] use scatter lists for all block pc requests and simplify hw handlers

Original From: Mike Christie <michaelc@cs.wisc.edu>

Add scsi_execute_req() as a replacement for scsi_wait_req()

Fixed up various pieces (added REQ_SPECIAL and caught req use after
free)

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years agoupdate scsi_wait_req to new format for blk_rq_map_kern()
James Bottomley [Wed, 15 Jun 2005 23:16:09 +0000 (18:16 -0500)]
update scsi_wait_req to new format for blk_rq_map_kern()

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years ago[SCSI] use scatter lists for all block pc requests and simplify hw handlers
James Bottomley [Sun, 5 Jun 2005 07:07:14 +0000 (02:07 -0500)]
[SCSI] use scatter lists for all block pc requests and simplify hw handlers

Here's the proof of concept for this one.  It converts scsi_wait_req to
do correct REQ_BLOCK_PC submission (and works nicely in my setup).

The final goal should be to eliminate struct scsi_request, but that
can't be done until the character submission paths of sg and st are also
modified.

There's some loss of functionality to this: retries are no longer
controllable (except by setting REQ_FASTFAIL) and the wait_req API needs
to be altered, but it looks very nice.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
19 years agofix mismerge in ll_rw_blk.c
James Bottomley [Sun, 28 Aug 2005 15:43:07 +0000 (10:43 -0500)]
fix mismerge in ll_rw_blk.c

19 years agoMerge refs/heads/upstream-fixes from master.kernel.org:/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 28 Aug 2005 01:05:14 +0000 (18:05 -0700)]
Merge refs/heads/upstream-fixes from /linux/kernel/git/jgarzik/netdev-2.6

19 years ago[PATCH] Remove race between con_open and con_close
Paul Mackerras [Sat, 27 Aug 2005 23:40:01 +0000 (09:40 +1000)]
[PATCH] Remove race between con_open and con_close

[ Same race and same patch also by Steven Rostedt <rostedt@goodmis.org> ]

I have a laptop (G3 powerbook) which will pretty reliably hit a race
between con_open and con_close late in the boot process and oops in
vt_ioctl due to tty->driver_data being NULL.

What happens is this: process A opens /dev/tty6; it comes into
con_open() (drivers/char/vt.c) and assign a non-NULL value to
tty->driver_data.  Then process A closes that and concurrently process
B opens /dev/tty6.  Process A gets through con_close() and clears
tty->driver_data, since tty->count == 1.  However, before process A
can decrement tty->count, we switch to process B (e.g. at the
down(&tty_sem) call at drivers/char/tty_io.c line 1626).

So process B gets to run and comes into con_open with tty->count == 2,
as tty->count is incremented (in init_dev) before con_open is called.
Because tty->count != 1, we don't set tty->driver_data.  Then when the
process tries to do anything with that fd, it oopses.

The simple and effective fix for this is to test tty->driver_data
rather than tty->count in con_open.  The testing and setting of
tty->driver_data is serialized with respect to the clearing of
tty->driver_data in con_close by the console_sem.  We can't get a
situation where con_open sees tty->driver_data != NULL and then
con_close on a different fd clears tty->driver_data, because
tty->count is incremented before con_open is called.  Thus this patch
eliminates the race, and in fact with this patch my laptop doesn't
oops.

Signed-off-by: Paul Mackerras <paulus@samba.org>
[ Same patch
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
  in http://marc.theaimsgroup.com/?l=linux-kernel&m=112450820432121&w=2 ]
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] zfcp: add rports to enable scsi_add_device to work again
Andreas Herrmann [Sat, 27 Aug 2005 18:07:54 +0000 (11:07 -0700)]
[PATCH] zfcp: add rports to enable scsi_add_device to work again

This patch fixes a severe problem with 2.6.13-rc7.

Due to recent SCSI changes it is not possible to add any LUNs to the zfcp
device driver anymore.  With registration of remote ports this is fixed.

Signed-off-by: Andreas Herrmann <aherrman@de.ibm.com>
Acked-by: James Bottomley <jejb@steeleye.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] sg.c: fix a memory leak in devices seq_file implementation
Jan Blunck [Sat, 27 Aug 2005 18:07:52 +0000 (11:07 -0700)]
[PATCH] sg.c: fix a memory leak in devices seq_file implementation

I know that scsi procfs is legacy code but this is a fix for a memory leak.

While reading through sg.c I realized that the implementation of
/proc/scsi/sg/devices with seq_file is leaking memory due to freeing the
pointer returned by the next() iterator method.  Since next() might return
NULL or an error this is wrong.  This patch fixes it through using the
seq_files private field for holding the reference to the iterator object.

Here is a small bash script to trigger the leak. Use slabtop to watch
the size-32 usage grow and grow.

#!/bin/sh

while true; do
cat /proc/scsi/sg/devices > /dev/null
done

Signed-off-by: Jan Blunck <j.blunck@tu-harburg.de>
Acked-by: James Bottomley <James.Bottomley@steeleye.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] fix for race problem in DVB USB drivers (dibusb)
Patrick Boettcher [Sat, 27 Aug 2005 17:30:30 +0000 (19:30 +0200)]
[PATCH] fix for race problem in DVB USB drivers (dibusb)

Fixed race between submitting streaming URBs in the driver and starting
the actual transfer in hardware (demodulator and USB controller) which
sometimes lead to garbled data transfers. URBs are now submitted first,
then the transfer is enabled. Dibusb devices and clones are now fully
functional again.

Signed-off-by: Patrick Boettcher <pb@linuxtv.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Fix capifs bug in initialization error path.
James Morris [Sat, 27 Aug 2005 11:47:06 +0000 (13:47 +0200)]
[PATCH] Fix capifs bug in initialization error path.

This fixes a bug in the capifs initialization code, where the
filesystem is not unregistered if kern_mount() fails.

Signed-off-by: James Morris <jmorris@namei.org>
Signed-off-by: Karsten Keil <kkeil@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] acpi_shutdown: Only prepare for power off on power_off
Eric W. Biederman [Sat, 27 Aug 2005 06:56:18 +0000 (00:56 -0600)]
[PATCH] acpi_shutdown: Only prepare for power off on power_off

When acpi_sleep_prepare was moved into a shutdown method we
started calling it for all shutdowns.

It appears this triggers some systems to power off on reboot.

Avoid this by only calling acpi_sleep_prepare if we are going to power
off the system.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] mmaper_kern.c fixes [buffer overruns]
Al Viro [Sat, 27 Aug 2005 05:48:15 +0000 (06:48 +0100)]
[PATCH] mmaper_kern.c fixes [buffer overruns]

 - copy_from_user() can fail; ->write() must check its return value.

 - severe buffer overruns both in ->read() and ->write() - lseek to the
   end (i.e.  to mmapper_size) and

if (count + *ppos > mmapper_size)
count = count + *ppos - mmapper_size;

   will do absolutely nothing.  Then it will call

copy_to_user(buf,&v_buf[*ppos],count);

   with obvious results (similar for ->write()).

   Fixed by turning read to simple_read_from_buffer() and by doing
   normal limiting of count in ->write().

 - gratitious lock_kernel() in ->mmap() - it's useless there.

 - lots of gratuitous includes.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Fix 6pack setting of MAC address
Ralf Baechle [Wed, 24 Aug 2005 17:01:33 +0000 (18:01 +0100)]
[PATCH] Fix 6pack setting of MAC address

Don't check type of sax25_family; dev_set_mac_address has already done
that before and anyway, the type to check against would have been
ARPHRD_AX25.  We only got away because AF_AX25 and ARPHRD_AX25 both happen
to be defined to the same value.

Don't check sax25_ndigis either; it's value is insignificant for the
purpose of setting the MAC address and the check has shown to break
some application software for no good reason.

Signed-off-by: Ralf Baechle DL5RB <ralf@linux-mips.org>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
19 years ago[PATCH] 6pack Timer initialization
Ralf Baechle [Thu, 25 Aug 2005 18:38:30 +0000 (19:38 +0100)]
[PATCH] 6pack Timer initialization

I dropped the timer initialization bits by accident when sending the
p-persistence fix.  This patch gets the driver to work again on halfduplex
links.

Signed-off-by: Ralf Baechle DL5RB <ralf@linux-mips.org>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
19 years ago[PATCH] Fix oops in sysfs_hash_and_remove_file()
James Bottomley [Sat, 27 Aug 2005 01:34:17 +0000 (18:34 -0700)]
[PATCH] Fix oops in sysfs_hash_and_remove_file()

The problem arises if an entity in sysfs is created and removed without
ever having been made completely visible.  In SCSI this is triggered by
removing a device while it's initialising.

The problem appears to be that because it was never made visible in sysfs,
the sysfs dentry has a null d_inode which oopses when a reference is made
to it.  The solution is simply to check d_inode and assume the object was
never made visible (and thus doesn't need deleting) if it's NULL.

(akpm: possibly a stopgap for 2.6.13 scsi problems.  May not be the
long-term fix)

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: clear the 'recovery' flags when starting an md array.
NeilBrown [Sat, 27 Aug 2005 01:34:16 +0000 (18:34 -0700)]
[PATCH] md: clear the 'recovery' flags when starting an md array.

It's possible for this to still have flags in it and a previous instance
has been stopped, and that confused the new array using the same mddev.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: create a MODULE_ALIAS for md corresponding to its block major number.
NeilBrown [Sat, 27 Aug 2005 01:34:15 +0000 (18:34 -0700)]
[PATCH] md: create a MODULE_ALIAS for md corresponding to its block major number.

I just discovered this is needed for module auto-loading.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] IB: fix use-after-free in user verbs cleanup
Roland Dreier [Sat, 27 Aug 2005 01:34:14 +0000 (18:34 -0700)]
[PATCH] IB: fix use-after-free in user verbs cleanup

Fix a use-after-free bug in userspace verbs cleanup: we can't touch
mr->device after we free mr by calling ib_dereg_mr().

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] arm: fix IXP4xx flash resource range
Deepak Saxena [Sat, 27 Aug 2005 01:34:11 +0000 (18:34 -0700)]
[PATCH] arm: fix IXP4xx flash resource range

We are currently reserving one byte more than actually needed by the flash
device and overlapping into the next I/O expansion bus window.  This a)
causes us to allocate an extra page of VM due to ARM ioremap() alignment
code and b) could cause problems if another driver tries to request the
next expansion bus window.

Signed-off-by: Deepak Saxena <dsaxena@plexity.net>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] x86_64: Tell VM about holes in nodes
Andi Kleen [Sat, 27 Aug 2005 01:34:10 +0000 (18:34 -0700)]
[PATCH] x86_64: Tell VM about holes in nodes

Some nodes can have large holes on x86-64.

This fixes problems with the VM allowing too many dirty pages because it
overestimates the number of available RAM in a node.  In extreme cases you
can end up with all RAM filled with dirty pages which can lead to deadlocks
and other nasty behaviour.

This patch just tells the VM about the known holes from e820.  Reserved
(like the kernel text or mem_map) is still not taken into account, but that
should be only a few percent error now.

Small detail is that the flat setup uses the NUMA free_area_init_node() now
too because it offers more flexibility.

(akpm: lotsa thanks to Martin for working this problem out)

Cc: Martin Bligh <mbligh@mbligh.org>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] I2C hwmon: kfree fixes
Mark M. Hoffman [Sat, 27 Aug 2005 01:34:08 +0000 (18:34 -0700)]
[PATCH] I2C hwmon: kfree fixes

This patch fixes several instances of hwmon drivers kfree'ing the "wrong"
pointer; the existing code works somewhat by accident.

(akpm: plucked from Greg's queue based on lkml discussion.  Finishes off the
patch from Jon Corbet)

Signed-off-by: Mark M. Hoffman <mhoffman@lightlink.com>
Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: Fix issue with gcc 4.0 compiled kernels
Anton Blanchard [Sat, 27 Aug 2005 01:34:07 +0000 (18:34 -0700)]
[PATCH] ppc64: Fix issue with gcc 4.0 compiled kernels

I recently had a BUG_ON() go off spuriously on a gcc 4.0 compiled kernel.
It turns out gcc-4.0 was removing a sign extension while earlier gcc
versions would not.  Thinking this to be a compiler bug, I submitted a
report:

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23422

It turns out we need to cast the input in order to tell gcc to sign extend
it.

Thanks to Andrew Pinski for his help on this bug.

Signed-off-by: Anton Blanchard <anton@samba.org>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] completely disable cpu_exclusive sched domain
Paul Jackson [Thu, 25 Aug 2005 19:47:56 +0000 (12:47 -0700)]
[PATCH] completely disable cpu_exclusive sched domain

At the suggestion of Nick Piggin and Dinakar, totally disable
the facility to allow cpu_exclusive cpusets to define dynamic
sched domains in Linux 2.6.13, in order to avoid problems
first reported by John Hawkes (corrupt sched data structures
and kernel oops).

This has been built for ppc64, i386, ia64, x86_64, sparc, alpha.
It has been built, booted and tested for cpuset functionality
on an SN2 (ia64).

Dinakar or Nick - could you verify that it for sure does avoid
the problems Hawkes reported.  Hawkes is out of town, and I don't
have the recipe to reproduce what he found.

Signed-off-by: Paul Jackson <pj@sgi.com>
Acked-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] undo partial cpu_exclusive sched domain disabling
Paul Jackson [Thu, 25 Aug 2005 19:47:50 +0000 (12:47 -0700)]
[PATCH] undo partial cpu_exclusive sched domain disabling

The partial disabling of Dinakar's new facility to allow
cpu_exclusive cpusets to define dynamic sched domains
doesn't go far enough.  At the suggestion of Nick Piggin
and Dinakar, let us instead totally disable this facility
for 2.6.13, in order to avoid problems first reported
by John Hawkes (corrupt sched data structures and kernel oops).

This patch removes the partial disabling code in 2.6.13-rc7,
in anticipation of the next patch, which will totally disable
it instead.

Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years agoMerge HEAD from master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6.git
Linus Torvalds [Fri, 26 Aug 2005 23:32:31 +0000 (16:32 -0700)]
Merge HEAD from /linux/kernel/git/davem/net-2.6.git

19 years ago[PATCH] hwmon: Off-by-one error in fscpos driver
Jean Delvare [Thu, 25 Aug 2005 16:43:37 +0000 (18:43 +0200)]
[PATCH] hwmon: Off-by-one error in fscpos driver

Coverity uncovered an off-by-one error in the fscpos driver, in function
set_temp_reset(). Writing to the temp3_reset sysfs file will lead to an
array overrun, in turn causing an I2C write to a random register of the
FSC Poseidon chip. Additionally, writing to temp1_reset and temp2_reset
will not work as expected. The fix is straightforward.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc32 8xx: fix m8xx_ide_init() #ifdef
Marcelo Tosatti [Tue, 23 Aug 2005 20:20:44 +0000 (17:20 -0300)]
[PATCH] ppc32 8xx: fix m8xx_ide_init() #ifdef

Be more precise on deciding whether to call m8xx_ide_init() at
m8xx_setup.c:platform_init().

Compilation fails if CONFIG_BLK_DEV_IDE is defined but
CONFIG_BLK_DEV_MPC8xx_IDE isnt.

Signed-off-by: Marcelo Tosatti <marcelo.tosatti@cyclades.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] late spinlock initialization in ieee1394/ohci
Al Viro [Thu, 25 Aug 2005 22:13:14 +0000 (23:13 +0100)]
[PATCH] late spinlock initialization in ieee1394/ohci

spinlock used in irq handler should be initialized before registering
irq, even if we know that our device has interrupts disabled; handler
is registered shared and taking spinlock is done unconditionally.  As
it is, we can and do get oopsen on boot for some configuration, depending
on irq routing - I've got a reproducer.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Ben Collins <bcollins@debian.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] bogus function type in qdio
Al Viro [Thu, 25 Aug 2005 22:03:35 +0000 (23:03 +0100)]
[PATCH] bogus function type in qdio

In qdio_get_micros() volatile in return type is plain noise (even with old
gccisms it would make no sense - noreturn function returning __u64 is a
bit odd ;-)

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] bogus iounmap() in emac
Al Viro [Thu, 25 Aug 2005 21:59:48 +0000 (22:59 +0100)]
[PATCH] bogus iounmap() in emac

Dumb typo: iounmap(&local_pointer_variable).

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] drivers/hwmon/*: kfree() correct pointers
Alexey Dobriyan [Thu, 25 Aug 2005 21:49:14 +0000 (01:49 +0400)]
[PATCH] drivers/hwmon/*: kfree() correct pointers

The adm9240 driver, in adm9240_detect(), allocates a structure.  The
error path attempts to kfree() ->client field of it (second one),
resulting in an oops (or slab corruption) if the hardware is not present.

->client field in adm1026, adm1031, smsc47b397 and smsc47m1 is the first in
${HWMON}_data structure, but fix them too.

Signed-off-by: Jonathan Corbet <corbet@lwn.net
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Fix oops in fs/locks.c on close of file with pending locks
Steve French [Fri, 26 Aug 2005 19:42:59 +0000 (14:42 -0500)]
[PATCH] Fix oops in fs/locks.c on close of file with pending locks

The recent change to locks_remove_flock code in fs/locks.c changes how
byte range locks are removed from closing files, which shows up a bug in
cifs.

The assumption in the cifs code was that the close call sent to the
server would remove any pending locks on the server on this file, but
that is no longer safe as the fs/locks.c code on the client wants unlock
of 0 to PATH_MAX to remove all locks (at least from this client, it is
not possible AFAIK to remove all locks from other clients made to the
server copy of the file).

Note that cifs locks are different from posix locks - and it is not
possible to map posix locks perfectly on the wire yet, due to
restrictions of the cifs network protocol, even to Samba without adding
a new request type to the network protocol (which we plan to do for
Samba 3.0.21 within a few months), but the local client will have the
correct, posix view, of the lock in most cases.

The correct fix for cifs for this would involve a bigger change than I
would like to do this late in the 2.6.13-rc cycle - and would involve
cifs keeping track of all unmerged (uncoalesced) byte range locks for
each remote inode and scanning that list to remove locks that intersect
or fall wholly within the range - locks that intersect may have to be
reaquired with the smaller, remaining range.

Signed-off-by: Steve French <sfrench@us.ibm.com>
Signed-off-by: Dave Kleikamp <shaggy@austin.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hppfs: fix symlink error path
Paolo 'Blaisorblade' Giarrusso [Fri, 26 Aug 2005 14:57:53 +0000 (16:57 +0200)]
[PATCH] hppfs: fix symlink error path

While touching this code I noticed the error handling is bogus, so I
fixed it up.

I've removed the IS_ERR(proc_dentry) check, which will never trigger and
is clearly a typo: we must check proc_file instead.

Signed-off-by: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Fixup symlink function pointers for hppfs [for 2.6.13]
Paolo 'Blaisorblade' Giarrusso [Fri, 26 Aug 2005 14:57:44 +0000 (16:57 +0200)]
[PATCH] Fixup symlink function pointers for hppfs [for 2.6.13]

Update hppfs for the symlink functions prototype change.

Yes, I know the code I leave there is still _bogus_, see next patch for
this.

Signed-off-by: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Document idr_get_new_above() semantics, update inotify
John McCutchan [Fri, 26 Aug 2005 18:02:04 +0000 (14:02 -0400)]
[PATCH] Document idr_get_new_above() semantics, update inotify

There is an off by one problem with idr_get_new_above.

The comment and function name suggest that it will return an id >
starting_id, but it actually returned an id >= starting_id, and kernel
callers other than inotify treated it as such.

The patch below fixes the comment, and fixes inotifys usage.  The
function name still doesn't match the behaviour, but it never did.

Signed-off-by: John McCutchan <ttb@tentacle.dhs.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years agoIgnore disabled ROM resources at setup
Linus Torvalds [Fri, 26 Aug 2005 17:49:22 +0000 (10:49 -0700)]
Ignore disabled ROM resources at setup

Writing even a disabled value seems to mess up some matrox graphics
cards.  It may be a card-related issue, but we may also be writing
reserved low bits in the result.

This was a fall-out of switching x86 over to the generic PCI resource
allocation code, and needs more debugging.  In particular, the old x86
code defaulted to not doing any resource allocations at all for ROM
resources.

In the meantime, this has been reported to make X happier by Helge
Hafting <helgehaf@aitel.hist.no>.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years agoOnly pre-allocate 256 bytes of cardbio IO range
Linus Torvalds [Fri, 26 Aug 2005 17:40:10 +0000 (10:40 -0700)]
Only pre-allocate 256 bytes of cardbio IO range

It may seem small, but most cards need much less, if any, and this not
only makes the code adhere to the comment, it seems to fix a boot-time
lockup on a ThinkPad 380XD laptop reported by Tero Roponen <teanropo@cc.jyu.fi>

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[TG3]: Fix ethtool loopback test lockup
Michael Chan [Thu, 25 Aug 2005 22:31:41 +0000 (15:31 -0700)]
[TG3]: Fix ethtool loopback test lockup

The tg3_abort_hw() call in tg3_test_loopback() is causing lockups on
some devices. tg3_abort_hw() disables the memory arbiter, causing
tg3_reset_hw() to hang when it tries to write the pre-reset signature.
tg3_abort_hw() should only be called after the pre-reset signature has
been written. This is all done in tg3_reset_hw() so the tg3_abort_hw()
call is unnecessary and can be removed.

[ Also bump driver version and release date. -DaveM ]

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 years ago[PATCH] cfq-iosched.c: minor fixes
Jens Axboe [Wed, 24 Aug 2005 12:57:54 +0000 (14:57 +0200)]
[PATCH] cfq-iosched.c: minor fixes

One critical fix and two minor fixes for 2.6.13-rc7:

- Max depth must currently be 2 to allow barriers to function on SCSI
- Prefer sync request over async in choosing the next request
- Never allow async request to preempt or disturb the "anticipation" for
  a single cfq process context. This is as-designed, the code right now
  is buggy in that area.

Signed-off-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Export pcibios_bus_to_resource
Keith Owens [Wed, 24 Aug 2005 06:06:25 +0000 (16:06 +1000)]
[PATCH] Export pcibios_bus_to_resource

pcibios_bus_to_resource is exported on all architectures except ia64
and sparc.  Add exports for the two missing architectures.  Needed when
Yenta socket support is compiled as a module.

Signed-off-by: Keith Owens <kaos@sgi.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] x86_64: update defconfig - reenable fusion
Andi Kleen [Wed, 24 Aug 2005 05:37:37 +0000 (07:37 +0200)]
[PATCH] x86_64: update defconfig - reenable fusion

I mistakedly disabled fusion support in an earlier update. Fusion
is commonly used on many x86-64 systems, so this was a problem.
This patch fixes that.

Signed-off-by: And Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: Export machine_power_off for therm_pm72 module
Benjamin Herrenschmidt [Wed, 24 Aug 2005 04:18:53 +0000 (14:18 +1000)]
[PATCH] ppc64: Export machine_power_off for therm_pm72 module

This patch puts back the export of machine_power_off() that was removed
by some janitor as it's used for emergency shutdown by the G5 thermal
control driver. Wether that driver should use kernel_power_off() instead
is debatable and a post-2.6.13 decision. In the meantime, please commit
that patch that fixes the driver for now.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] cpu_exclusive sched domains build fix
Paul Jackson [Wed, 24 Aug 2005 11:15:10 +0000 (04:15 -0700)]
[PATCH] cpu_exclusive sched domains build fix

As reported by Paul Mackerras <paulus@samba.org>, the previous patch
"cpu_exclusive sched domains fix" broke the ppc64 build with
CONFIC_CPUSET, yielding error messages:

kernel/cpuset.c: In function 'update_cpu_domains':
kernel/cpuset.c:648: error: invalid lvalue in unary '&'
kernel/cpuset.c:648: error: invalid lvalue in unary '&'

On some arch's, the node_to_cpumask() is a function, returning
a cpumask_t.  But the for_each_cpu_mask() requires an lvalue mask.

The following patch fixes this build failure by making a copy
of the cpumask_t on the stack.

Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] m68k: fix broken macros causing compile errors
Andreas Schwab [Wed, 24 Aug 2005 15:36:21 +0000 (17:36 +0200)]
[PATCH] m68k: fix broken macros causing compile errors

Add parens around macro parameters.

Signed-off-by: Andreas Schwab <schwab@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[FIB_TRIE]: Don't ignore negative results from fib_semantic_match
Patrick McHardy [Wed, 24 Aug 2005 05:06:09 +0000 (22:06 -0700)]
[FIB_TRIE]: Don't ignore negative results from fib_semantic_match

When a semantic match occurs either success, not found or an error
(for matching unreachable routes/blackholes) is returned. fib_trie
ignores the errors and looks for a different matching route. Treat
results other than "no match" as success and end lookup.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 years agoLinux v2.6.13-rc7
Linus Torvalds [Wed, 24 Aug 2005 03:39:14 +0000 (20:39 -0700)]
Linux v2.6.13-rc7

Too many changes to release a final 2.6.13.

19 years ago[PATCH] Fix IXP4xx CLOCK_TICK_RATE
Deepak Saxena [Tue, 23 Aug 2005 20:30:29 +0000 (13:30 -0700)]
[PATCH] Fix IXP4xx CLOCK_TICK_RATE

As pointed out in the following thread, the CLOCK_TICK_RATE setting for
IXP4xx is incorrect b/c the HW ignores the lowest 2 bits of the LATCH
value.

   http://lists.arm.linux.org.uk/pipermail/linux-arm-kernel/2005-August/030950.html

Tnx to George Anziger and Egil Hjelmeland for finding the issue.

Signed-off-by: Deepak Saxena <dsaxena@plexity.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] cpu_exclusive sched domains on partial nodes temp fix
Paul Jackson [Tue, 23 Aug 2005 08:04:27 +0000 (01:04 -0700)]
[PATCH] cpu_exclusive sched domains on partial nodes temp fix

This keeps the kernel/cpuset.c routine update_cpu_domains() from
invoking the sched.c routine partition_sched_domains() if the cpuset in
question doesn't fall on node boundaries.

I have boot tested this on an SN2, and with the help of a couple of ad
hoc printk's, determined that it does indeed avoid calling the
partition_sched_domains() routine on partial nodes.

I did not directly verify that this avoids setting up bogus sched
domains or avoids the oops that Hawkes saw.

This patch imposes a silent artificial constraint on which cpusets can
be used to define dynamic sched domains.

This patch should allow proceeding with this new feature in 2.6.13 for
the configurations in which it is useful (node alligned sched domains)
while avoiding trying to setup sched domains in the less useful cases
that can cause the kernel corruption and oops.

Signed-off-by: Paul Jackson <pj@sgi.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Dinakar Guniguntala <dino@in.ibm.com>
Acked-by: John Hawkes <hawkes@sgi.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>