From: Andrey Panin <pazke@donpac.ru>

this trivial patch changes mailing list address for visws subarch support
along with some occurences of my old email addresses.

We need to unconditionally brelse() the buffer in there, because
journal_remove_journal_head() leaves a ref behind.

release_buffer_page() does that.  Call it all the time because we can usually
strip the buffers and free the page even if it was not marked buffer_freed().

Mainly affects data=journal mode

ext3_block_truncate_page() is calling grab_cache_page() inside a JBD
transaction.  This is wrong, because transactions nest inside lock_page().

The deadlock is against shrink_list->ext3_journalled_writepage->journal_start.

This was not noticed before because we never used to journal writepage() data
in journalled-data mode.  And because the deadlock against
generic_file_write() is covered up by i_sem.

Rework things so that we lock the page prior to starting a transaction.

From: David Mosberger <davidm@napali.hpl.hp.com>

compat_sys_old_getrlimit() depends on sys_old_getrlimit() and the patch
below updates the guarding #ifdef accordingly.

From: Andrew Theurer <habanero@us.ibm.com>

This patch ensures that when node loads are compared, the load value is
normalised.  Without this, load balance across nodes of dissimilar cpu
counts can cause unfairness and sometimes lower overall performance.

For example, a 2 node system with 4 cpus in the first node and 2 cpus in
the second.  A workload with 6 running tasks would have 3 tasks running on
one node and 3 on the other, leaving one cpu idle in the first node and two
tasks sharing a cpu in the second node.  The patch would ensure that 4
tasks run in the first node and 2 in the second.

I ran some kernel compiles comparing this patch on a 2 node 4 cpu/2 cpu
system to show the benefits.  Without the patch I got 140 second elapsed
time.  With the patch I get 132 seconds (6% better).

Although it is not very common to have nodes with dissimilar cpu counts, it
is already happening.  PPC64 systems with partitioning have this happen,
and I expect it to be more common on ia32 as partitioning becomes more
common.

From: Chris Heath <chris@heathens.co.nz>

Here's a column counting bug that was lurking in a corner of n_tty.c.

From: Zwane Mwaikambo <zwane@linuxpower.ca>

smbfs tends to oops over a null server->ops->getattr in smb_proc_getattr().

Urban says:

 In 2.5 the server->ops is initialized when smbfs gets a connection, to
 match whatever the server is capable of or not. This happens after the
 mount syscall so smbfs is then mounted but not usable.

 Not sure if smb_lookup is always called before any other operation that
 uses server->ops. If it is then it would be enough to have a test there.

 Otherwise I will just change all users of server->ops to verify the
 pointer first and return -EIO or something.

Al Viro says (paraphrasing)

 ugh, we need mount2().

It doesn't look to me like we'll be getting mount2() in the 2.6 timeframe.
Zwane's patch implement's Urban's workaround.

From: Maneesh Soni <maneesh@in.ibm.com>

- hpfs_unlink() can race with lockless d_lookup(), as we can have situations
  where d_lookup() has successfully looked-up a dentry and at the sametime
  hpfs_unlink()--->d_drop() has dropped it. Taking the per dentry lock
  before checking the d_count in hpfs_unlink() solves this race condition.

From: Maneesh Soni <maneesh@in.ibm.com>

- nfs_unlink() can race with lockless d_lookup() as d_lookup() can
  successfully lookup a dentry for which nfs_unlink() can assume that no one
  else is using and can go ahead and do nfs_safe_remove() on it. By using
  per dentry lock, it is solved as we d_lookup() will fail the lookup for
  unhashed dentries.

From: Maneesh Soni <maneesh@in.ibm.com>

- d_invalidate() can incorrectly return success instead of returning -EBUSY
  as we can have situations where lockless d_lookup has found a dentry
  successfully before d_invalidate drops it

From: Ingo Molnar

Apparently our thread-creation performance has gone down the tubes again,
because the mm.free_area_cache search heuristic broke.

The initial, more naive hole-cache patch helped the testcode in the
beginning.  Then after some point glibc started creating a 'small hole' in
the vmas, which hole was _below_ the thread stacks, and which hole thus
prevented the intended operation of the cache.

The new code solves the problem by relaxing the 'smallest address hole cache'
rule a bit, the cache is now not re-set at every get_unmapped_area() time,
it's only re-set during unmaps.  It's also re-set if there are no allocatable
mappings at all - ie.  correctness is not affected.

From: David Mosberger <davidm@napali.hpl.hp.com>

Give the ia32 vsyscall DSO the same name as ia64's.

From: Stephen Smalley <sds@epoch.ncsc.mil>

This patch adds an AT_SECURE auxv entry to pass a boolean flag indicating
whether "secure mode" should be enabled (i.e.  sanitize the environment,
initial descriptors, etc) and allows each security module to specify the
flag value via a new hook.

New userland can then simply obey this flag when present rather than
applying other methods of deciding (sample patch for glibc-2.3.2 can be
found at http://www.cs.utah.edu/~sds/glibc-secureexec.patch).

This change enables security modules like SELinux to request secure mode
upon changes to other security attributes (e.g.  capabilities,
roles/domains, etc) in addition to uid/gid changes or even to completely
override the legacy logic.

The legacy decision algorithm is preserved in the default hook functions
for the dummy and capability security modules.

Credit for the idea of adding an AT_SECURE auxv entry goes to Roland
McGrath.

- unused var warning in pnp_add_irq_resource()

- unused var in fd_mcs.c (Adrian Bunk)

- remove dead code in i810_main.c (a u8 cannot exceed 255) (Adrian Bunk)

- don't truncate dma_addr_t's in gdth.c (Jes Sorensen)

- vgastate.c needs vmalloc.h (Geert Uytterhoeven <geert@linux-m68k.org>)

- revert bogus test from rd_open() - it was a 2.4 forward-port, and 2.5
  doesn't need it.  (The gendisks aren't registered, we cannot get there
  anyway).

From: Robert Love <rml@tech9.net>

Basically, the problem is that setscheduler() does not set need_resched
when needed.  There are two basic cases where this is needed:

	- the task is running, but now it is no longer the highest
	  priority task on the rq
	- the task is not running, but now it is the highest
	  priority task on the rq

In either case, we need to set need_resched to invoke the scheduler.

sysrq-T currently displays the same stack trace for every process.

Teach show_stack() to look in the passed task_struct first if the caller did
not pass in a stack address.

From: Mikael Pettersson <mikpe@csd.uu.se>

2.5.73 removed the workaround needed to prevent gcc-2.95.x from
miscompiling spinlocks on UP.  It turns out that some versions of
gcc-2.95 still do have problems with empty structs, so re-introduce
the workaround.

into kroah.com:/home/greg/linux/BK/pci-2.5

Useful for testing hotplug issues with pci drivers.

Added needed release function, now all pci hotplug drivers need to implement
it...

 - Combine pci_alloc_primary_bus_parented into pci_scan_bus_parented.
 - Move the EXPORT_SYMBOL for pci_root_buses up to its definition.
 - Don't EXPORT_SYMBOL pci_scan_bus since it's a static inline.
 - Add the pci_domain_nr() to the sysfs name for this bus.

I noticed we have a couple of redundancies in drivers/pci/Makefile,
have a patch...

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

Add missing CONFIG_TC35805 entry to Kconfig.
Update CONFIG_NET_SB1250_MAC Kconfig entry.
Minor cosmetic updates to gt96100eth.

Submitted by Ralf Baechle <ralf@linux-mips.org>

[ Registers of the world, unite! ]

This makes the IO-APIC data structures use unions, so that we can
cleanly access them both as flat "raw" values, and as the bitmap
sub-entries.

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

This error is handled in the signal delivery code and should never be
returned from a syscall unless a signal is pending.  Grepping seems to
indicate that that is in fact the case (but not for ERESTARTSYS, but
that is another problem).

Ignore earlier patch -- this one locks and free's as appropriate.
Tested on 2.5.72 with SMP.

Of course, it begs the question why have two (now three) versions of drivers for
the same hardware...

dev_ioctl already checks capable(CAP_NET_ADMIN), so no need to do so in
drivers.

net_device is no longer allocated as part of the driver's private structure,
instead it's allocated via alloc_netdev. compile tested only since no hardware
against 2.5.73-bk


-daniel

===== smc91c92_cs.c 1.18 vs edited =====