o [FEATURE] New hashbin locking scheme for irqueue
o [FEATURE] Get rid of old broken hashbin locking schemes
o [FEATURE] Lock hasbins while enumerating it in various places
o [CORRECT] Remove all remaining "save_flags(flags);cli();"
o [CORRECT] Fix two "return with spinlock" found by Stanford checker

o [CORRECT] Make sure LAP address is sane, which mean not NULL,
  not BROADCAST and not already in use by another link.

o [CRITICA] Remove correct IAS Attribute/Object even if name is dup'ed
o [CORRECT] Make irqueue 64 bit compliant (__u32 -> long)
o [FEATURE] Don't use random handle for IrLMP handle, use self
  Remove dependancy on random generator in stack init

it has just one function nowdays, and that one would be much better
static in scsi.c

update DAC960 and umem to new bio_endio() completions

Make BIO_UPTODATE a sticky good flag, only clearing it on error.

I pondered leaving the BIO_UPTODATE clearing up to the bi_end_io()
functions, but decided against that and left it in bio_endio() as
before.

I also added a bit of debug in bio_endio() to detect callers specifying
more than bi_size i/o complete.  And to top it all of, bio_endio()
documentation.  This is mainly to clear my back when people bitch about
using bio_endio(), I can say RTFM without worry :-)

This renames the svc_getlong and svc_putlong macros as svc_getu32 and
svc_putu32.  this is simple clean up and is obviously correct.  it was
part of the patch that implements stricter type checking for rpc auth
flavors.

This implements stricter type checking for rpc auth flavors.  it is a
prerequisite for RPC GSSAPI and its authentication pseudoflavors.
please apply it.

This renames *reconn* to *conn* since the same code now handles both
initial TCP connect, and TCP reconnection, and corrects some comments.
against 2.5.36, requires earlier patch (1/1).

This provides clean up and bug fixes for the RPC layer's TCP socket
connection management logic.  Trond, Alexey, and DaveM have seen this
patch.  i've been running it for several weeks here and feel it is ready
for wider testing.  these two patches are prerequisites for further
clean-ups and fixes for RPC over TCP.

bugs fixed:
+  TCP connection establishment now times out after 60 seconds instead of
   hanging for ten minutes.  60 seconds is more in line with how long a
   server takes to reboot.
+  on a soft-mounted file system, TCP reconnections now time out and fail
   the RPC request, like most other NFS clients, instead of hanging the
   NFS client until the server comes back.
+  on hard-mounted file systems, the RPC layer now delays 15 seconds
   before retrying after a failed connection attempt instead of retrying
   as soon as it can.
+  TCP connection error recovery is now more verbose so users can see why
   their NFS sessions are hung.  this can be tuned with future patches if
   it is unreasonably noisy.
+  the TCP connect logic is cleaned up so adding checks for new errnos
   is easier.
+  the same code now handles both initial connection and reconnection.
   the original initial connection code did not have comprehensive error
   handling.
+  some obscure design elements are now documented in comments.
+  kfree was used by mistake in xprt_destroy.

This patch consolidates zap_thread into reparent_thread, and fixes
__will_become_orphaned_pgrp to check p->real_parent instead of
p->parent.

XFS now needs to use bio_add_page() as not to build bio's of the wrong
size. Also makes the bio build nicer.

Seems James introduced this one, it's not needed though. Will take a
look at eh thread now...

missed updating xfs and jfs to new bi_end_io handling

Make bio->bi_end_io() take bytes_done and actual error as argument. This
enables partial completion of bio's, which is important for latency
reasons (bio can be huge, for slow media we want page-by-page
completions).

I think I got most of the bi_end_io() functions out there, but I might
have missed a few. For the record, if you don't care about partial
completions and just want to be notified when the entire bio completes,
add a

	if (bio->bi_size)
		return 1;

to the top of your bi_end_io(). It should return 0 on completion.
bio_endio() will decrement bio->bi_size appropriately, it's recommended
for people to go through that. Otherwise they will have to control
BIO_UPTODATE and bi_size decrement themselves, there's really no reason
to do that. I've deliberately avoided doing any functional changes to
any of the end_io functions, as I think that would only make the patch
more complex. It's simple right now, but this being i/o paths I prefer
(as usual) to be careful and take small steps. The mpage_end_io_read()
do-vecs-at-the-time change can come right after this, for instance.

This makes mpage use bio_add_page().  It has the added advantage that
users don't need to have deep knowledge about what the different bio
fields mean, nor does it have to abuse some of them while building the
bio.

This is bio_add_page(), 100% identical to the version I sent out for
comments earlier this week.  With the previous queue restriction patch,
this guarentees that we can always add a page worth of data to the bio.
bio_add_page() returns 0 on success, and 1 on failure.  Either the page
is added completely, or the attempt is aborted.

bio_add_page() uses the normal queue restrictions to determine whether
we an add the page or not. if a queue has further restrictions, it can
define a q->merge_bvec_fn() to further impose limits.

Patch also includes changes to ll_rw_kio(), if for nothing else to
demonstrate how to use this piece of instrastructure.

Impose sane limits on queue restrictions. that means:

 - q->max_sectors must be able to at least span a page
 - q->max_phys_segments and q->max_hw_segments must be at least 1
 - q->max_segment_size must be at least page sized
 - q->seg_boundary_mask must be at least PAGE_CACHE_SIZE-1

Art Haas did these.

cyclades.c doesn't compile cleanly: the problem is that older -dj
patches contained patches to both cyclades.c and cyclades.h but only the
former one went into your tree.  This syncs up the header file too.

  Here's a resend of C99 patches for the files in fs.

The old form of designated initializers are obsolete: we need to
replace them with the ISO C forms before 2.6.  Gcc has always supported
both forms anyway.

  I suck.  Lvalues continue to haunt me.  This works for me.

  BTW, I prefer to have bug reports cc'd to linux-kernel, so the
  results are archived.  Plus, public humiliation is good for the
  soul.

  This is a trivial patch already applied in the -ac tree for the 2.4.19 kernel.
  Patch for i2c-philips-par.c free() the memory in the case of problems loading
  the adapter.

   <linux/smp_lock.h> has the normal idempotent construction.

   The attached file removes the second #include.

  Here's a C99 designated initializer patch for fs/freevxfs. The
  patch is against 2.5.35.

into home.transmeta.com:/home/torvalds/v2.5/linux

Display the total slab memory in /proc/meminfo.  Handy while we play
with the slab pruning code.

This info is also available via /proc/slabinfo, but I think this
convenience is worth the extra few lines.

Patch from Christoph Hellwig.

Move the buffer_head-based IO functions out of ll_rw_blk.c and into
fs/buffer.c.  So the buffer IO functions are all in buffer.c, and
ll_rw_blk.c knows nothing about buffer_heads.

This patch has been acked by Jens.

From Anton Altaparmakov

The below bk patch against your current bk repository adds an export
for fs/buffer.c::unmap_underlying_metadata() to modules.

It is required to make ntfs compile as a module in the current kernel.

Note that NTFS doesn't currently dirty bufferheads of the underlying
blockdevice and as such could probably live without calling
unmap_underlying_metadata() but as soon as we enable writing to
compressed files it will do so and considering some of the
infrastructure code is already present I would rather not rip it out
now.

Patch by Christoph Hellwig.

This patch is cleanup of mmap.c with the side-effect of making the lock
order of mapping->i_shared_lock and mm->page_table_lock consistant.

__remove_shared_vm_struct is merged into its only caller,
remove_shared_vm_struct, which now also does the job of
{lock,unlock}_vma_mappings itself, to avoid duplicate checks.

Similarly, vma_link is updated to take and release i_shared_lock
itself, fixing the lock order.  split_vma is updated to use
insert_vm_struct instead of using __insert_vm_struct and taking
page_table_lock and i_shared_lock itself (wrongly).  __insert_vm_struct
is removed as it is unused now.

Patch from Christoph Hellwig.  250 lines of code go away.

The three syscalls madvice/mlock/mprotect all change attributes of
memory regions.  These attributes are stored in vm_area_structs (VMAs).

The current code for those syscalls does this by having four
subroutines each, for changing a whole VMA, one for just setting new
flags if a full VMA is affected, one that creates a new VMA in the
right part of the old one and sets the flags there, one for the left
part, and one that splits of both the left and the right part and sets
the new flags in the middle.

This patch makes those routines create new VMAs on the left and on te
right hand first and then always setting the flags in the old VMA.
Together with using a library function (split_vma) to to the split this
simplies the code and makes it more readable.  As a side affect it
fixes some VM accounting bugs observed by Hugh Dickins.

Patch from Christoph Hellwig moves the madvise implementation out of
filemap.c and into its own .c file.  No other changes are made.

into home.transmeta.com:/home/torvalds/v2.5/linux

into home.transmeta.com:/home/torvalds/v2.5/linux

Andrew removed a flush_dcache_page in his kmap_atomic generic_file_*
changes.

Doing that sort of corrupts data on some platforms.