Plug the two-megabyte-per-day memory leak.

I happened to spot this kfree(of complete garbage) - it is on an
oh-we-raced-retry path which is obviously exceedingly rare,

From: Peter Bergner <bergner@vnet.ibm.com>

In fs/binfmt_elf.c:load_elf_binary() (both 2.6 and 2.4), there is some
minimal checking whether the interpreter it's about to load/run is a valid
ELF file, but it fails to check whether the interpreter is of the correct
arch.  We ran into this when a borked powerpc64-linux toolchain set the
interpreter on our 64-bit app to our 32-bit ld.so.  Executing the app
caused the kernel to really chew up memory.  I'm assuming x86_64 and
sparc64 might possibly see the same behavior.

Note I'm not sure of the history behind INTERPRETER_AOUT, so I added the
test for INTERPRETER_ELF so as not to change it's behavior in case someone
still relies on it.

As an aside, it seems the elf_check_arch() macros should really be checking
for more than a valid e_machine value.  I'd think checking one or more of
the e_ident[EI_CLASS], e_ident[EI_DATA] and e_ident[EI_OSABI] values would
be required as well, no?

From: Jesper Juhl <juhl-lkml@dif.dk>

Fix a C99ism.

From: "Kurtis D. Rader" <kdrader@us.ibm.com>

http://bugme.osdl.org/show_bug.cgi?id=1365

The digi_acceleport.c USB serial driver has a bogus "address of" operator
that results in BUGs.

The problem is that digi_wakeup_write_lock() takes a pointer to a struct
usb_serial_port.  However, what gets passed is a pointer to a pointer to a
struct usb_serial_port.

OK, I give up.  Kill all the might_sleep warnings from the early boot
process.

There seems to be no header file which declares system_running.

From: Jan Kara <jack@ucw.cz>

attached patch should fix a quota locking problem causing deadlock (when
inode was being released from icache and it caused newly created quota
structure to be written).

From: Hugh Dickins <hugh@veritas.com>

mm/filemap.c's generic_file_aio_write_nolock changed SetPageReferenced
to mark_page_accessed in -test3: now follow that in shmem_file_write.

From: Hugh Dickins <hugh@veritas.com>

mm/shmem.c was converted to i_size_read in -test1, and the remaining
references to a file's naked i_size are safely protected by i_sem;
but surely shmem_file_write must use i_size_write to update i_size.

From: Hugh Dickins <hugh@veritas.com>

If it's possible for a tmpfs page beyond i_size to remain in cache until
shmem_truncate repeats truncate_inode_pages, then shmem_writepage's
BUG_ON(index >= info->next_index) cannot be completely safe.  But it's a
useful check in a fragile area, so retain it when not in shmem_truncate.

From: Hugh Dickins <hugh@veritas.com>

Extend use of that SHMEM_PAGEIN flag to where shmem_getpage adds a page
to the cache.  It couldn't have caused a BUG_ON(inode->i_blocks), but if
i_size is reduced (from another cpu) the instant after shmem_swp_alloc
checks it, shmem_getpage could insert a page into the cache just after
truncate_inode_pages has passed through cleaning it, leaving stale data
(which may mysteriously reappear if the file is later extended).

Easily fixed for tmpfs, using the mechanism just added for swapoff; and
probably more important there, since its read from swap can insert non-0
data.  But is there not a similar issue, a tiny window, in filemap.c?
if truncate_inode_pages comes in between checking i_size and adding new
page to cache.  Not worth getting excited, but something to beware of.

From: Hugh Dickins <hugh@veritas.com>

On 23 July, Sergey S. Kostyliov <rathamahata@php4.ru> reported a tmpfs
BUG_ON(inode->i_blocks) during swapoff: my last version of the fix to
swapoff/truncate race was inadequate, since I_FREEING might get set or
i_size be reduced (from another cpu) the instant after it's tested here.

So revert to the previous version of the fix, shmem_truncate calling
truncate_inode_pages again, if pages still left in cache; but avoid the
recall in usual cases of partial truncation, by having a "pagein" flag
to indicate when recall might be necessary.  (Since those flags already
use VM_ACCOUNT and VM_LOCKED, must redefine another VM_flag for this.)
Sergey and 2.4-aa have run fine with this for a couple of months.

From: Hugh Dickins <hugh@veritas.com>

LTP tests the filesystem on /tmp: many failures when tmpfs because it missed
the way directories hand down their gid.  Also fix ramfs and hugetlbfs.

From: Hugh Dickins <hugh@veritas.com>

LTP tests the filesystem on /tmp: there are many failures when using tmpfs
because simple_lookup forgot to reject filenames longer than the NAME_MAX
tmpfs declares in its statfs.  This also fixes ramfs and hugetlbfs.

This driver is taking uinitialised stack gunk from the pdev[] array and
feeding it into pci_read_config_byte() and crashing when modprobed with no
hardware present.

Fix it to not index past the initialised members of pdev[].

We don't know if this driver works.

From: Stephen Hemminger <shemminger@osdl.org>

The following will prevent adjtime from causing time regression.  It delays
starting the adjtime mechanism for one tick, and keeps gettimeofday inside
the window.

Only fixes i386, but changes to other arch would be similar.

Running a simple clock test program and playing with adjtime demonstrates
that this fixes the problem (and 2.6.0-test6 is broken).  But given the
fragile nature of the timer code, it should go through some more testing
before inclusion.

Silence a bogus "may be used uninitialised" warning.  It only affects
architectures which use the tlb_finish_mmu() args.

Sync this up with 2.4:

 ChangeSet@1.404.2.2  2002-05-06 21:30:10-03:00  hch@infradead.org
 [PATCH] memsetup fixes (again)

 The mem= fixes from Red Hat's tree had a small bug:
 if mem= was not actually used with the additional features, but
 int plain old way, is used the value as the size of memory it
 wants, not the upper limit.  The problem with this is that there
 is a small difference due to memory holes.

 I had one report of a person using mem= to reduce memory size for
 a broken i386 chipset thaty only supports 64MB cached and the rest
 as mtd/slram device for swap.  I got broken as the boundaries changed.

From: Jens Axboe <axboe@suse.de>

The command 'eject /dev/scd0' sends a START_STOP command to the device with
the data direction set to SCSI_DATA_WRITE but a transfer length of zero.
This causes a problem for some code paths.

From: "V. Rajesh" <vrajesh@eecs.umich.edu>

If a vma is already present in an i_mmap list of a mapping,
then it is racy to update the vm_start, vm_end, and vm_pgoff
members of the vma without holding the mapping's i_shared_sem. 
This is because the updates can race with invalidate_mmap_range_list.

I audited all the places that assign vm_start, vm_end, and vm_pgoff.
AFAIK, the following is the list of questionable places:

1) This patch fixes the racy split_vma. Kernel 2.4 does the
   right thing, but the following changesets introduced a race.

   http://linux.bkbits.net:8080/linux-2.5/patch@1.536.34.4
   http://linux.bkbits.net:8080/linux-2.5/patch@1.536.34.5

   You can use the patch and programs in the following URL to
   trigger the race.

  http://www-personal.engin.umich.edu/~vrajesh/linux/truncate-race/

2) This patch also locks a small racy window in vma_merge.

3) In few cases vma_merge and do_mremap expand a vma by adding 
   extra length to vm_end without holding i_shared_sem. I think
   that's fine.

4) In arch/sparc64, vm_end is updated without holding i_shared_sem.
   Check make_hugetlb_page_present.  I hope that is fine, but
   I am not sure.

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

The "irq_vector[]" array is indexed by the sum of all RTEs in all I/O
APICs, and is not necessarily limited by the x86 CPU irq vector inputs.

In fact, the irq vector index would overflow on big machines with lots
of IO APIC's, causing the boot to fail.

So grow the array for the big SMP boxes, keeping the default the same as
before (and shrink the vector entry size down to a 8-bit value, since
that's the size of the actual CPU vector entry).

into redhat.com:/spare/repo/libata-2.5-merge

Intel ICH5 (production)
ServerWorks / Apple K2 (beta)
VIA (beta)
Silicon Image 3112 (broken!)
Various Promise (alpha/beta)

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

Patch from Alexander Schulz

This patch adds the define for PCIMEM_BASE so that the shark kernel
compiles again.

Patch from Dave Jiang

The fix allows IOP321 based platforms to boot all the way instead of
blank screen after "starting kernel...".  Inherited from Deepak's
earlier patch. 

CMD640 driver also supports VLB version of the chipset, therefore fix
drivers/ide/Makefile to include pci/ subdir even if CONFIG_BLK_DEV_IDEPCI=n.

 - remove cruft, memset() i2c-client structures in tda9840, tea6420,
    tea6415c driver, otherwise i2c_register()/kobject() segfaults later on

 - remove cruft, add I2C_ADAP_CLASS_TV_ANALOG identifier for analog tv
   i2c handler

 - DVB networking uses big endian crc32, so change all occurences of
   crc32_le to crc32_be
 - fix usage of firmware location Kconfig option in tda1004x frontend
   driver
 - add missing VBI line decoding initialization to saa7113 code for
   av7110 driver
 - make av7110 firmware static and *not* __initdata, so recover_arm()
   can work in case the driver is compiled in statically

  http://bugme.osdl.org/show_bug.cgi?id=1185

  use X86 ACPI specific version of eisa_set_level_irq()
  http://bugzilla.kernel.org/show_bug.cgi?id=1390

This fixes initrd with devfs.  With that combination the late-boot code
does temporary mount of devfs over rootfs /dev, which made /dev/initrd
inaccessible.  For setups without devfs that didn't happen. 

The fix is trivial - put the file in question outside of /dev; IOW,
we simply replace "/dev/initrd" with "/initrd.image" in init/*.

Confirmed to fix the problem by Valdis Kletnieks

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

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

... as if "nohide" is in used, it will be the wrong filehandle, and
returning a filehandle is optional anyway.