From: Chris Mason <mason@suse.com>

There's a race in loopback setup, it's easiest to trigger with one or more
procs doing loopback mounts at the same time.  The problem is that
fs/block_dev.c:do_open() only calls bdev_set_size on the first open.
Picture two procs:

proc1: mount -o loop file1 mnt1
proc2: mount -o loop file2 mnt2

proc1                   proc2
open /dev/loop0                         # bd_openers now 1
do_open
 bd_set_size(bdev, 0)                   # loop unbound, so bdev size is 0
                        open /dev/loop0 # bd_openers now 2
loop_set_fd                             # disk capacity now correct, but
				        # bdev not updated
mount /dev/loop0 /mnt
do_open

Because bd_openers != 0 for the last do_open, bd_set_size is not called
again and a size of 0 is used.  This eventually leads to an oops when the
loop device is unmounted, because fsync_bdev calls block_write_full_page
who decides every page on the block device is outside i_size and unmaps
them.

When ext2 or reiserfs try to sync a metadata buffer, we get an oops on
because the buffers are no longer mapped.

The patch below changes loop_set_fd and loop_clr_fd to also manipulate the
size of the block device, which fixes things for me.

From: Arjan van de Ven <arjanv@redhat.com>

The patch below (written by Al Viro) solves a nasty chicken-and-egg issue
for operating system installers (well at least anaconda but the problem
domain is not exclusive to that)

The basic problem is this:

- The small first stage installer locates the image file of the second
  stage installer (which has X and all the graphical stuff); this image can
  be on the same CD, but it can come via NFS, http or ftp or ...  as well.

- The first stage installer loop-back mounts this image and gives control
  to the second stage installer by calling some binary there.

- The graphical installer then asks the user all those questions and
  starts installing packages.  Again the packages can come from the CD but
  also from NFS or http or ...

Now in case of a CD install, once all requested packages from the first CD
are installed, the installer wants to unmount and eject the CD and prompt
the user to put CD 2 in.......  EXCEPT that the unmount can't work since
the installer is actually running from a loopback mount of this cd.

The solution is a "LOOP_CHANGE_FD" ioctl, where basically the installer
copies the image to the harddisk (which can only be done late since only
late the target harddisk is mkfs'd) and then magically switches the backing
store FD from underneath the loop device to the one on the target harddisk
(and thus unbusying the CD mount).

This is obviously only allowed if the size of the new image is identical
and if the loop image is read-only in the first place.  It's the
responsibility of root to make sure the contents is the same (but that's of
the give-root-enough-rope kind)

From: Sam Ravnborg <sam@ravnborg.org>

Make the difference between 'make clean' and 'make distclean/mrproper' more
explicit.

make clean now removes all generated files except .config* and .version.
The result is much easier to understand now.

make clean deletes all generated files (except .config* and .version).
make mrproper deletes configuration and all temporary files left by patch,
editors and the like.

Example output:
> make mrproper
  CLEAN   init
  CLEAN   usr
  CLEAN   scripts/kconfig
  CLEAN   scripts
  CLEAN   .tmp_versions include/config
  CLEAN   include/asm-i386/asm_offsets.h include/linux/autoconf.h include/linux/version.h include/asm .tmp_versions
  CLEAN   .version .config

Form the list of files/directories deleted during make clean, removed all
references that is no longer relevant for the current kernel.

From: William Lee Irwin III <wli@holomorphy.com>

Using PAGE_SIZE rather than 4096 so that mmap() granularity is honored by
whatever non-i386 architectures use MMAP_PAGE_ZERO.

From: Ville Nuorvala <vnuorval@tcs.hut.fi>

There are a few kernel-only things in compiler.h which should have been
placed inside __KERNEL__.

From: "Chen, Kenneth W" <kenneth.w.chen@intel.com>

The only path to get to del_timer call in __generic_unplug_device() is when
blk_remove_plug() returns 1, and in that case it already removed the
unplug_timer.

Patch to remove this redundant call.

From: <j-nomura@ce.jp.nec.com>

The attached patch is NUMA-aware zonelist builder patch, which sorts
zonelist in the order that near-node first, far-node last.  In lse-tech and
linux-ia64, where most of NUMA people resides, no objections are raised so
far.

The patch adds NUMA-specific version of build_zonelists which calls
find_next_best_node to select the next-nearest node to add to zonelist.

The patch has no effect on flat NUMA platform.

From: Sam Ravnborg <sam@ravnborg.org>

The EXTRA_CFLAGS assignments in the following files are a left-over from
the early 2.5 days where the source was not compiled from the root of the
source tree.

Removing these wrong assignments fixes
http://bugme.osdl.org/show_bug.cgi?id=2210

A script named 'kernel' in the .. directory no longer halt compilation.

From: Ben Fennema <bfennema@falcon.csc.calpoly.edu>

- added udf 2.5 #defines

- fixed prealloc discard race

- fixed several bugs in inode_getblk

- added S_IFSOCK support

- fix unicode encoding bug

- change partition allocation from kmalloc to vmalloc for large
  allocations

From: James Morris <jmorris@redhat.com>

selinux is currently inspecting the filesystem name ("nfs" vs "coda" vs
watever) to work out whether it needs to hanbdle binary mount data.

Eliminate all that by adding a flag to file_system_type.fs_flags.

dm-stripe.c: The stripe width must be at least the page size.

From: Joe Thornber <thornber@redhat.com>

List targets ioctl.  [Patrick Caulfield]

From: Joe Thornber <thornber@redhat.com>

Fill in missing queue limitations when table is complete instead of enforcing
the "default" limits on every dm device.  Problem noticed by Mike Christie.

[Christophe Saout]

From: Joe Thornber <thornber@redhat.com>

Audit for list_for_each_*entry*

From: Joe Thornber <thornber@redhat.com>

Add an endio method to targets.  This method is allowed to request another
shot at failed ios (think multipath).  Context can be passed between the map
method and the endio method.

From: Herbert Xu <herbert@gondor.apana.org.au>

By allowing X86_MCE_NONFATAL to be a module, it can be included in
distribution kernels without upsetting those with strange hardware.

From: "H. Peter Anvin" <hpa@zytor.com>

This patch cleans up the very early memory setup on the i386 platform.  In
particular, it removes the hard-coded 8 MB limit completely by dynamically
creating the early-boot pagetables rather than having them hard coded.

While I was at it, I changed head.S so that it always sets up a local GDT;
this means among other things that SMP and VISWS are no longer special
cases, and is conceptually cleaner to boot.  The VISWS people have
confirmed it works on VISWS.

It also uses a separate entrypoint for non-boot processors since this is
completely kernel-internal anyway.  This eliminates the need to set %bx on
boot.  (If you think this is a bad idea I can eliminate this change; it
just seemed cleaner to me to do it this way.)

Additionally, zero bss with rep;stosl rather that rep;stosb.

From: "Randy.Dunlap" <rddunlap@osdl.org>

From: Luiz Fernando Capitulino <lcapitulino@prefeitura.sp.gov.br>

remove ifdef/endif in rtc_generic_init().
use returned error code;

From: Brian Gerst <bgerst@didntduck.org>

Drivers should all be converted to use ioremap() or isa_*() by now.

If there's nothing available and the file is O_NONBLOCK, return -EAGAIN.

This is a bit grubby - really we should push the file* down into do_syslog()
and handle it inside the spinlock.

From: john stultz <johnstul@us.ibm.com>

In developing the ia64-cyclone patch, which implements a cyclone based time
interpolator, I found the following bug which could cause time
inconsistencies.

In update_wall_time_one_tick(), which is called each timer interrupt, we
call time_interpolator_update(delta_nsec) where delta_nsec is approximately
NSEC_PER_SEC/HZ.  This directly correlates with the changes to xtime which
occurs in update_wall_time_one_tick().

However in update_wall_time(), on a second overflow, we again call
time_interpolator_update(NSEC_PER_SEC).  However while the components of
xtime are being changed, the overall value of xtime does not (nsec is
decremented NSEC_PER_SEC and sec is incremented).  Thus this call to
time_interpolator_update is incorrect.

This patch removes the incorrect call to time_interpolator_update and was
found to resolve the time inconsistencies I had seen while developing the
ia64-cyclone patch.

From: James Simmons <jsimmons@infradead.org>

This patch fixes fb_console_init from being called twice.  I still need to
fix set_con2fb but this helps but this is still important to get in.

From: Niraj Kumar <niraj17@iitbombay.org>

This patch adds read-only support for ufs2 (used in FreeBSD 5.x) variant of
ufs filesystem.  For filesystem specific tools, see
http://ufs-linux.sourceforge.com .

From: Suparna Bhattacharya <suparna@in.ibm.com>

From: Ram Pai <linuxram@us.ibm.com>

Pipelined readahead behaviour is suitable for sequential reads, but not for
large random reads (typical of database workloads), where lazy readahead
provides a big performance boost.

One option (suggested by Andrew Morton) would be to have the application
pass hints to turn off readahead by setting the readahead window to zero
using posix_fadvise64(POSIX_FADV_RANDOM), and to special-case that in
do_generic_mapping_read to completely bypass the readahead logic and
instead read in all the pages needed directly.

This was the idea I started with.  But then I thought, we can do a still
better job ?  How about adapting the readahead algorithm to lazy-read or
non-lazy-read based on the past i/o patterns ?

The overall idea is to keep track of average number of contiguous pages
accessed in a file.  If the average at any given time is above ra->pages
the pattern is sequential.  If not the pattern is random.  If pattern is
sequential do non-lazy-readahead( read as soon as the first page in the
active window is touched) else do lazy-readahead.

I have studied the behaviour of this patch using my user-level simulator.
It adapts pretty well.

Note from Suparna: This appears to bring streaming AIO read performance for
large (64KB) random AIO reads back to sane values (since the lazy readahead
backout in the mainline).

From: <viro@parcelfarce.linux.theplanet.co.uk>

cramfs and freevxfs explicitly mark themselves readonly (as other r/o fs
do).

afs marked noatime (ACKed by maintainer)

filesystems that do not do update_atime() in their ->readdir() had been
explicitly marked nodiratime.  NOTE: cifs, coda and ncpfs almost certainly
need full noatime as we currently have in nfs and afs.

update_atime() call shifted to callers of ->readdir() and out of
->readdir() instances.  Bugs caught:

  dcache_readdir() updated atime only if it reached EOF.

  bfs_readdir() - ditto.

  qnx4_readdir() - ditto.

From: Brian Gerst <bgerst@didntduck.org>

Remove stubs for sys_ioperm for non-x86 arches, using sys_ni_syscall
instead where applicable.  Support for sys_ioperm is unconditionally no for
non-x86 arches.

From: Anton Blanchard <anton@samba.org>

We were hitting problems on machines with cpu_possible != cpu_online when
NUMA was enabled.  The debug checks would trip during scheduler init
because we iterate through all possible cpus whereas we only set up NUMA
information for online cpus.

Longer term we should have a cpu_up hook which sets up its NUMA information
but for now we initalise all possible cpus and memory to node 0.

From: Arjan van de Ven <arjanv@redhat.com>

Unfortunatly a large portion of the oops reports lack the basic
information about what kernel version the oops is for; it's trivial to just
print this in the oops as well to improve the usefulness of bugreports...

Fix a typo in the code that prevents lockup on config space access
to sleeping devices on ppc32/G5. Please apply.

Thanks to some great debugging work by Olaf Hering and Marcus Meissner
it has been noticed that the current ppc64 syscall code is corrupting
4 bytes past errno. Why we even bothered to set errno beats me, its
unusable in the kernel.

Since we had to reinstate the inline syscall code we can go back to
using it for those few syscalls that we call. Especially now with
Randy's syscall prototype cleanup we should be calling them directly
but we can do that sometime later.

This small patch builds on top of the blk_rq_map_user() patch just sent,
and enables us to easily support DMA for CDROMREADAUDIO cdda extraction.
It's quite amazing how much cool stuff you can with the new block layer
:-)

Patch has intelligent fall back from multi frame dma to single frame
dma, and further to old-style pio ripping in case of hardware problems.

This patch allows you to map a request with user data for io, similarly
to what you can do with bio_map_user() already to a bio. However, this
goes one step further and populates the request so the user only has to
fill in the cdb (almost) and put it on the queue for execution. Patch
converts sg_io() to use it, next patch I'll send adapts cdrom layer to
use it for zero copy cdda dma extraction.

While playing with udev, I discovered that the virtual console
devices on iSeries had there minor numbers off by one i.e. /dev/tty1
was minor 2!

This fixes it.

From Takashi Iwai

kobj_lookup had a 64bit bug, which caused the request of a unknown
character device to burn CPU instead of failing quickly.

The biggest new feature is fixed 32bit vsyscall (SYSCALL+SYSENTER)
support, mostly from Jakub Jelinek.  This increases 32bit syscall
performance greatly (latency halved and better).  The SYSENTER for Intel
support required some infrastructure changes, but seems to work now too.

The 64bit vsyscall vtime() just references xtime.tv_sec now.  This
should make it a lot faster too.

A fix for some Intel IA32e systems.  Also a few long standing bugs in
NMI like exception handlers were fixed.

And a lot of other bug fixes.

Full changeLog:
 - Clean up 32bit address room limit handling, fix 3gb personality
 - Move memcpy_{from,to}io export to ksyms.c file. This seems to work
   around a toolchain bug (Andreas Gruenbacher)
 - Update defconfig
 - ACPI merges from i386 (SBF should work now, acpi=strict)
 - Implement mmconfig support based on i386 code (untested)
 - Fix i386/x86-64 pci source file sharing
 - Implement ptrace access for 32bit vsyscall page
 - Always initialize all 32bit SYSENTER/SYSCALL MSRs.
 - Export run time cache line size to generic kernel
 - Remove explicit CPUID in ia32 syscall code
 - Fill in most of boot_cpu_data early
 - Remove unused PER_LINUX32 setup
 - Fix syscall trace in fast 32bit calls (Suresh B. Siddha)
 - Tighten first line of the oops again.
 - Set up ptrace registers correctly for debug,ss,double fault exceptions
 - Fix 64bit bug in sys_time64
 - Optimize time syscall/vsyscall to only read xtime
 - Fix csum_partial_copy_nocheck
 - Remove last traces of FPU emulation
 - Check properly for rescheduling in exceptions with own stack
 - Harden exception stack entries (#SS,#NMI,#MC,#DF,#DB) against bogus GS
 - Use an exception stack for machine checks
 - Handle TIF_SINGLESTEP properly in kernel exit
 - Add exception stack for debug handler
 - Disable X86_HT for Opteron optimized builds because it pulls in ACPI_BOOT
 - Fix CONFIG_ACPI_BOOT compilation without CONFIG_ACPI
 - Fix eflags handling in SYSENTER path (Jakub Jelinek)
 - Use atomic counter for enable/disable_hlt
 - Support 32bit SYSENTER vsyscall too (Jakub Jelinek)
 - Don't redefine Dprintk
 - Change some cpu/apic id arrays to char
 - Support arbitary cpu<->apicid in hard_smp_processor_id (Surresh B Sidda)
 - Move K8 erratum #100 workaround into slow path of page fault handler.
 - Fix 32bit cdrom direct access ioctls (Jens Axboe)
 - Enable 32bit vsyscalls by default
 - Fix 32bit vsyscalls (Jakub Jelinek)

This makes the temperature control code more robust, putting less
pressure on i2c, and work around occasional misconfiguration of the ADC
chips leading to incorrect temperature readings.

This patch fixes the code that workaround lockups when accessing
the config space of devices on K2 when they are shut down. The
code was there but in the wrong place ;) And a typo prevented the
ohci1394 version of it from working.

into ppc970.osdl.org:/home/torvalds/v2.5/linux

When adding an additional device, a variable which has been removed
is accessed.  Adding the first device or a 4-port card works.
Adding a second device was not done in PCI hot plug testing.  Now
tested on IA32.