RTC: add periodic irq support to rtc-cmos

Adds support for periodic irq enabling in rtc-cmos.  This could be used by
the ALSA driver and is already being tested with the zaptel ztdummy module.

Signed-off-by: Alessandro Zummo <a.zummo@towertech.it>
Cc: David Brownell <david-b@pacbell.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

knfsd: clean up EX_RDONLY

Share a little common code, reverse the arguments for consistency, drop the
unnecessary "inline", and lowercase the name.

Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Acked-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

knfsd: move EX_RDONLY out of header

EX_RDONLY is only called in one place; just put it there.

Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Acked-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

nfsd: remove unnecessary NULL checks from nfsd_cross_mnt

We can now assume that rqst_exp_get_by_name() does not return NULL; so clean
up some unnecessary checks.

Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Acked-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

nfsd: return errors, not NULL, from export functions

I converted the various export-returning functions to return -ENOENT instead
of NULL, but missed a few cases.

This particular case could cause actual bugs in the case of a krb5 client that
doesn't match any ip-based client and that is trying to access a filesystem
not exported to krb5 clients.

Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Acked-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

nfsd: fix possible read-ahead cache and export table corruption

The value of nperbucket calculated here is too small--we should be rounding up
instead of down--with the result that the index j in the following loop can
overflow the raparm_hash array.  At least in my case, the next thing in memory
turns out to be export_table, so the symptoms I see are crashes caused by the
appearance of four zeroed-out export entries in the first bucket of the hash
table of exports (which were actually entries in the readahead cache, a
pointer to which had been written to the export table in this initialization
code).

It looks like the bug was probably introduced with commit
fce1456a19f5c08b688c29f00ef90fdfa074c79b ("knfsd: make the readahead params
cache SMP-friendly").

Cc: <stable@kernel.org>
Cc: Greg Banks <gnb@melbourne.sgi.com>
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Acked-by: NeilBrown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

move page writeback acounting out of macros

page-writeback accounting is presently performed in the page-flags macros.
This is inconsistent and a bit ugly and makes it awkward to implement
per-backing_dev under-writeback page accounting.

So move this accounting down to the callsite(s).

Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

timer.c: cleanup recently introduced whitespace damage

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

timekeeping: fixup shadow variable argument

clocksource_adjust() has a clock argument, which shadows the file global clock
variable.  Fix this up.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: john stultz <johnstul@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

m68knommu: remove is_in_rom() function

Remove is_in_rom() function.  It doesn't actually serve the purpose it was
intended to.  If you look at the use of it _access_ok() (which is the only use
of it) then it is obvious that most of memory is marked as access_ok.  No
point having is_in_rom() then, so remove it.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

m68knommu: start dump from exception stack

In die_if_kernel() start the stack dump at the exception-time SP, not at the
SP with all the saved registers; the stack below exception-time sp contains
only exception-saved values and is already printed in details just before.

Signed-off-by: Philippe De Muyter <phdm@macqel.be>
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

m68knommu: generic irq handling

Change the m68knommu irq handling to use the generic irq framework.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: use set_compound_page_dtor

Use appropriate accessor function to set compound page destructor
function.

Cc: William Irwin <wli@holomorphy.com>
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Acked-by: Adam Litke <agl@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Remove nid_lock from alloc_fresh_huge_page

The fix to that race in alloc_fresh_huge_page() which could give an illegal
node ID did not need nid_lock at all: the fix was to replace static int nid
by static int prev_nid and do the work on local int nid.  nid_lock did make
sure that racers strictly roundrobin the nodes, but that's not something we
need to enforce strictly.  Kill nid_lock.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

isdn/sc: compile breakage re check_reset()

There is check_reset() -- global function in drivers/isdn/sc/
There is check_reset -- variable holding module param in aacraid driver.

On allyesconfig they clash with:

  LD      drivers/built-in.o
drivers/isdn/built-in.o: In function `check_reset':
: multiple definition of `check_reset'
drivers/scsi/built-in.o:(.data+0xe458): first defined here
ld: Warning: size of symbol `check_reset' changed from 4 in drivers/scsi/built-in.o to 219 in drivers/isdn/built-in.o
ld: Warning: type of symbol `check_reset' changed from 1 to 2 in drivers/isdn/built-in.o

Rename the former.

Signed-off-by: Alexey Dobriyan <adobriyan@sw.ru>
Cc: Karsten Keil <kkeil@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmalloc_32 should use GFP_KERNEL

I've noticed lots of failures of vmalloc_32 on machines where it
shouldn't have failed unless it was doing an atomic operation.

Looking closely, I noticed that:

#if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
#define GFP_VMALLOC32 GFP_DMA32
#elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
#define GFP_VMALLOC32 GFP_DMA
#else
#define GFP_VMALLOC32 GFP_KERNEL
#endif

Which seems to be incorrect, it should always -or- in the DMA flags
on top of GFP_KERNEL, thus this patch.

This fixes frequent errors launchin X with the nouveau DRM for example.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Andi Kleen <ak@suse.de>
Cc: Dave Airlie <airlied@linux.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

FRV: work around a possible compiler bug

Work around a possible bug in the FRV compiler.

What appears to be happening is that gcc resolves the
__builtin_constant_p() in kmalloc() to true, but then fails to reduce the
therefore constant conditions in the if-statements it guards to constant
results.

When compiling with -O2 or -Os, one single spurious error crops up in
cpuup_callback() in mm/slab.c.  This can be avoided by making the memsize
variable const.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

dequeue_huge_page() warning fix

mm/hugetlb.c: In function `dequeue_huge_page':
mm/hugetlb.c:72: warning: 'nid' might be used uninitialized in this function

Cc: Christoph Lameter <clameter@sgi.com>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <hermes@gibson.dropbear.id.au>
Cc: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

some kmalloc/memset ->kzalloc (tree wide)

Transform some calls to kmalloc/memset to a single kzalloc (or kcalloc).

Here is a short excerpt of the semantic patch performing
this transformation:

@@
type T2;
expression x;
identifier f,fld;
expression E;
expression E1,E2;
expression e1,e2,e3,y;
statement S;
@@

 x =
- kmalloc
+ kzalloc
  (E1,E2)
  ...  when != \(x->fld=E;\|y=f(...,x,...);\|f(...,x,...);\|x=E;\|while(...) S\|for(e1;e2;e3) S\)
- memset((T2)x,0,E1);

@@
expression E1,E2,E3;
@@

- kzalloc(E1 * E2,E3)
+ kcalloc(E1,E2,E3)

[akpm@linux-foundation.org: get kcalloc args the right way around]
Signed-off-by: Yoann Padioleau <padator@wanadoo.fr>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Acked-by: Russell King <rmk@arm.linux.org.uk>
Cc: Bryan Wu <bryan.wu@analog.com>
Acked-by: Jiri Slaby <jirislaby@gmail.com>
Cc: Dave Airlie <airlied@linux.ie>
Acked-by: Roland Dreier <rolandd@cisco.com>
Cc: Jiri Kosina <jkosina@suse.cz>
Acked-by: Dmitry Torokhov <dtor@mail.ru>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Acked-by: Pierre Ossman <drzeus-list@drzeus.cx>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: "David S. Miller" <davem@davemloft.net>
Acked-by: Greg KH <greg@kroah.com>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: "Antonino A. Daplas" <adaplas@pol.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

stacktrace: fix header file for !CONFIG_STACKTRACE

The print_stack_trace macro in stacktrace.h has a wrong number of
arguments, fix it.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arjan van de Ven <arjan@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lockdep debugging: give stacktrace for init_error

When I started adding support for lockdep to 64-bit powerpc, I got a
lockdep_init_error and with this patch was able to pinpoint why and where
to put lockdep_init().  Let's support this generally for others adding
lockdep support to their architecture.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arjan van de Ven <arjan@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lockstat: better class name representation

optionally add class->name_version and class->subclass to the class name

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lockstat: measure lock bouncing

    __acquire
        |
       lock _____
        |        \
        |    __contended
        |         |
        |        wait
        | _______/
        |/
        |
   __acquired
        |
   __release
        |
     unlock

We measure acquisition and contention bouncing.

This is done by recording a cpu stamp in each lock instance.

Contention bouncing requires the cpu stamp to be set on acquisition. Hence we
move __acquired into the generic path.

__acquired is then used to measure acquisition bouncing by comparing the
current cpu with the old stamp before replacing it.

__contended is used to measure contention bouncing (only useful for preemptable
locks)

[akpm@linux-foundation.org: cleanups]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lockdep: fixup sk_callback_lock annotation

the two init sites resulted in inconsistend names for the lock class.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lockdep: various fixes

 - update the copyright notices
 - use the default hash function
 - fix a thinko in a BUILD_BUG_ON
 - add a WARN_ON to spot inconsitent naming
 - fix a termination issue in /proc/lock_stat

[akpm@linux-foundation.org: cleanups]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lockstat: hook into spinlock_t, rwlock_t, rwsem and mutex

Call the new lockstat tracking functions from the various lock primitives.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Jason Baron <jbaron@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lockstat: human readability tweaks

Present all this fancy new lock statistics information:

*warning, _wide_ output ahead*

(output edited for purpose of brevity)

 # cat /proc/lock_stat
lock_stat version 0.1
-----------------------------------------------------------------------------------------------------------------------------------------------------------------
                              class name    contentions   waittime-min   waittime-max waittime-total   acquisitions   holdtime-min   holdtime-max holdtime-total
-----------------------------------------------------------------------------------------------------------------------------------------------------------------

                         &inode->i_mutex:         14458           6.57      398832.75     2469412.23        6768876           0.34    11398383.65   339410830.89
                         ---------------
                         &inode->i_mutex           4486          [<ffffffff802a08f9>] pipe_wait+0x86/0x8d
                         &inode->i_mutex              0          [<ffffffff802a01e8>] pipe_write_fasync+0x29/0x5d
                         &inode->i_mutex              0          [<ffffffff802a0e18>] pipe_read+0x74/0x3a5
                         &inode->i_mutex              0          [<ffffffff802a1a6a>] do_lookup+0x81/0x1ae

.................................................................................................................................................................

              &inode->i_data.tree_lock-W:           491           0.27          62.47         493.89        2477833           0.39         468.89     1146584.25
              &inode->i_data.tree_lock-R:            65           0.44           4.27          48.78       26288792           0.36         184.62    10197458.24
              --------------------------
                &inode->i_data.tree_lock             46          [<ffffffff80277095>] __do_page_cache_readahead+0x69/0x24f
                &inode->i_data.tree_lock             31          [<ffffffff8026f9fb>] add_to_page_cache+0x31/0xba
                &inode->i_data.tree_lock              0          [<ffffffff802770ee>] __do_page_cache_readahead+0xc2/0x24f
                &inode->i_data.tree_lock              0          [<ffffffff8026f6e4>] find_get_page+0x1a/0x58

.................................................................................................................................................................

                      proc_inum_idr.lock:             0           0.00           0.00           0.00             36           0.00          65.60         148.26
                        proc_subdir_lock:             0           0.00           0.00           0.00        3049859           0.00         106.81     1563212.42
                        shrinker_rwsem-W:             0           0.00           0.00           0.00              5           0.00           1.73           3.68
                        shrinker_rwsem-R:             0           0.00           0.00           0.00            633           2.57         246.57       10909.76

'contentions' and 'acquisitions' are the number of such events measured (since
the last reset). The waittime- and holdtime- (min, max, total) numbers are
presented in microseconds.

If there are any contention points, the lock class is presented in the block
format (as i_mutex and tree_lock above), otherwise a single line of output is
presented.

The output is sorted on absolute number of contentions (read + write), this
should get the worst offenders presented first, so that:

 # grep : /proc/lock_stat | head

will quickly show who's bad.

The stats can be reset using:

 # echo 0 > /proc/lock_stat

[bunk@stusta.de: make 2 functions static]
[akpm@linux-foundation.org: fix printk warning]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Jason Baron <jbaron@redhat.com>
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lockstat: core infrastructure

Introduce the core lock statistics code.

Lock statistics provides lock wait-time and hold-time (as well as the count
of corresponding contention and acquisitions events). Also, the first few
call-sites that encounter contention are tracked.

Lock wait-time is the time spent waiting on the lock. This provides insight
into the locking scheme, that is, a heavily contended lock is indicative of
a too coarse locking scheme.

Lock hold-time is the duration the lock was held, this provides a reference for
the wait-time numbers, so they can be put into perspective.

  1)
    lock
  2)
    ... do stuff ..
    unlock
  3)

The time between 1 and 2 is the wait-time. The time between 2 and 3 is the
hold-time.

The lockdep held-lock tracking code is reused, because it already collects locks
into meaningful groups (classes), and because it is an existing infrastructure
for lock instrumentation.

Currently lockdep tracks lock acquisition with two hooks:

  lock()
    lock_acquire()
    _lock()

 ... code protected by lock ...

  unlock()
    lock_release()
    _unlock()

We need to extend this with two more hooks, in order to measure contention.

  lock_contended() - used to measure contention events
  lock_acquired()  - completion of the contention

These are then placed the following way:

  lock()
    lock_acquire()
    if (!_try_lock())
      lock_contended()
      _lock()
      lock_acquired()

 ... do locked stuff ...

  unlock()
    lock_release()
    _unlock()

(Note: the try_lock() 'trick' is used to avoid instrumenting all platform
       dependent lock primitive implementations.)

It is also possible to toggle the two lockdep features at runtime using:

  /proc/sys/kernel/prove_locking
  /proc/sys/kernel/lock_stat

(esp. turning off the O(n^2) prove_locking functionaliy can help)

[akpm@linux-foundation.org: build fixes]
[akpm@linux-foundation.org: nuke unneeded ifdefs]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Jason Baron <jbaron@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lockdep: reduce the ifdeffery

Move code around to get fewer but larger #ifdef sections.  Break some
in-function #ifdefs out into their own functions.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lockdep: sanitise CONFIG_PROVE_LOCKING

Ensure that all of the lock dependency tracking code is under
CONFIG_PROVE_LOCKING.  This allows us to use the held lock tracking code for
other purposes.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Jason Baron <jbaron@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fix raw_spinlock_t vs lockdep

Use the lockdep infrastructure to track lock contention and other lock
statistics.

It tracks lock contention events, and the first four unique call-sites that
encountered contention.

It also measures lock wait-time and hold-time in nanoseconds. The minimum and
maximum times are tracked, as well as a total (which together with the number
of event can give the avg).

All statistics are done per lock class, per write (exclusive state) and per read
(shared state).

The statistics are collected per-cpu, so that the collection overhead is
minimized via having no global cachemisses.

This new lock statistics feature is independent of the lock dependency checking
traditionally done by lockdep; it just shares the lock tracking code. It is
also possible to enable both and runtime disabled either component - thereby
avoiding the O(n^2) lock chain walks for instance.

This patch:

raw_spinlock_t should not use lockdep (and doesn't) since lockdep itself
relies on it.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: update module information

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: remove statistics counters from /proc/fs/coda

Similar information can easily be obtained with strace -c.

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: remove struct coda_sb_info

The sb_info structure only contains a single pointer to the character device,
there is no need for the added indirection.

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: cleanup downcall handler

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: cleanup coda_lookup, use dsplice_alias

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: ignore returned values when upcalls return errors

Venus returns an ENOENT error on open, so we shouldn't try to grab the
filehandle for the returned fd.

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: replace upc_alloc/upc_free with kmalloc/kfree

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: avoid lockdep warning in coda_readdir

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: block signals during upcall processing

We ignore signals for about 30 seconds to give userspace a chance to see the
upcall.  As we did not block signals we ended up in a busy loop for the
remainder of the period when a signal is received.

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: cleanup for upcall handling path

Make the code that processes upcall responses more straightforward, uncovered
at least one bad assumption.  We trusted that vc_inuse would be 0 when upcalls
are aborted, however the device may have been reopened.

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: cleanup /dev/cfs open and close handling

- Make sure device index is not a negative number.
- Unlink queued requests when the device is closed to avoid passing them
  to the next opener.

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: use ilookup5

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: coda doesn't track atime

Set MS_NOATIME flag to avoid unnecessary calls when the coda inode is
accessed.

Also, set statfs.f_bsize to 4k.  1k is obviously too small for the suggested
IO size.

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: allow removal of busy directories

A directory without children may still be busy when it is the cwd for some
process.  We can safely remove such a directory because the VFS prevents
further operations.  Also we don't need to call d_delete as it is already
called in vfs_rmdir.

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: fix nlink updates for directories

The Coda client sets the directory link count to 1 when it isn't sure how many
subdirectories we have.  In this case we shouldn't change the link count in
the kernel when a subdirectory is created or removed.

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: correctly invalidate cached access rights

Change the epoch value to forces a refresh instead of clearing the cached
rights mask and block all further accesses to the object.

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: do not grab an uninitialized fd when the open upcall returns an error

When open fails the fd in the response is uninitialized and we ended up taking
a reference on the file struct and never released it.

Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Use --build-id ld option

This change passes the --build-id when linking the kernel and when linking
modules, if ld supports it.  This is a new GNU ld option that synthesizes an
ELF note section inside the read-only data.  The note in this section contains
unique identifying bits called the "build ID", which are generated so as to be
different for any two linked ELF files that aren't identical.  The build ID
can be recovered from stripped files, memory dumps, etc.  and used to look up
the original program built, locate debuginfo or other details or history
associated with it.  For normal program linking, the compiler passes
--build-id to ld by default, but the option is needed when using ld directly
as we do.

Signed-off-by: Roland McGrath <roland@redhat.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Add /sys/kernel/notes

This patch adds the /sys/kernel/notes magic file.  Reading this delivers the
contents of the kernel's .notes section.  This lets userland easily glean any
detailed information about the running kernel's build that was stored there at
compile time.

Signed-off-by: Roland McGrath <roland@redhat.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

s390: Put allocated ELF notes in read-only data segment

This changes the s390 linker script to use the asm-generic NOTES macro so that
ELF note sections with SHF_ALLOC set are linked into the kernel image along
with other read-only data.  The PT_NOTE also points to their location.

This paves the way for putting useful build-time information into ELF notes
that can be found easily later in a kernel memory dump.

Signed-off-by: Roland McGrath <roland@redhat.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

powerpc: Put allocated ELF notes in read-only data segment

This changes the powerpc linker script to use the asm-generic NOTES macro so
that ELF note sections with SHF_ALLOC set are linked into the kernel image
along with other read-only data.  The PT_NOTE also points to their location.

This paves the way for putting useful build-time information into ELF notes
that can be found easily later in a kernel memory dump.

Signed-off-by: Roland McGrath <roland@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

alpha: Put allocated ELF notes in read-only data segment

This changes the alpha linker script to use the asm-generic NOTES macro so
that ELF note sections with SHF_ALLOC set are linked into the kernel image
along with other read-only data.  The PT_NOTE also points to their location.

This paves the way for putting useful build-time information into ELF notes
that can be found easily later in a kernel memory dump.

Signed-off-by: Roland McGrath <roland@redhat.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

x86_64: Put allocated ELF notes in read-only data segment

This changes the x86_64 linker script to use the asm-generic NOTES macro so
that ELF note sections with SHF_ALLOC set are linked into the kernel image
along with other read-only data.  The PT_NOTE also points to their location.

This paves the way for putting useful build-time information into ELF notes
that can be found easily later in a kernel memory dump.

Signed-off-by: Roland McGrath <roland@redhat.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

i386: Put allocated ELF notes in read-only data segment

This changes the i386 linker script and the asm-generic macro it uses so that
ELF note sections with SHF_ALLOC set are linked into the kernel image along
with other read-only data.  The PT_NOTE also points to their location.

This paves the way for putting useful build-time information into ELF notes
that can be found easily later in a kernel memory dump.

Signed-off-by: Roland McGrath <roland@redhat.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fix ext4/JBD2 build warnings

Looking at the current linus-git tree jbd_debug() define in
include/linux/jbd2.h

extern u8 journal_enable_debug;

#define jbd_debug(n, f, a...)                                           \
        do {                                                            \
                if ((n) <= journal_enable_debug) {                      \
                        printk (KERN_DEBUG "(%s, %d): %s: ",            \
                                __FILE__, __LINE__, __FUNCTION__);      \
                        printk (f, ## a);                               \
                }                                                       \
        } while (0)
> fs/ext4/inode.c: In function â\80\98ext4_write_inodeâ\80\99:
> fs/ext4/inode.c:2906: warning: comparison is always true due to limited
> range of data type
>
> fs/jbd2/recovery.c: In function â\80\98jbd2_journal_recoverâ\80\99:
> fs/jbd2/recovery.c:254: warning: comparison is always true due to
> limited range of data type
> fs/jbd2/recovery.c:257: warning: comparison is always true due to
> limited range of data type
>
> fs/jbd2/recovery.c: In function â\80\98jbd2_journal_skip_recoveryâ\80\99:
> fs/jbd2/recovery.c:301: warning: comparison is always true due to
> limited range of data type
>
Noticed all warnings are occurs when the debug level is 0. Then found
the "jbd2: Move jbd2-debug file to debugfs" patch
http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=0f49d5d019afa4e94253bfc92f0daca3badb990b

changed the jbd2_journal_enable_debug from int type to u8, makes the
jbd_debug comparision is always true when the debugging level is 0. Thus
the compile warning occurs.

Thought about changing the jbd2_journal_enable_debug data type back to
int, but can't, because the jbd2-debug is moved to debug fs, where
calling debugfs_create_u8() to create the debugfs entry needs the value
to be u8 type.

Even if we changed the data type back to int, the code is still buggy,
kernel should not print jbd2 debug message if the
jbd2_journal_enable_debug is set to 0. But this is not the case.

The fix is change the level of debugging to 1. The same should fixed in
ext3/JBD, but currently ext3 jbd-debug via /proc fs is broken, so we
probably should fix it all together.

Signed-off-by: Mingming Cao <cmm@us.ibm.com>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: Theodore Tso <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

update checkpatch.pl to version 0.08

This version brings a number of new checks, and a number of bug
fixes.  Of note:

  - warnings for multiple assignments per line
  - warnings for multiple declarations per line
  - checks for single statement blocks with braces

This patch includes an update for feature-removal-schedule.txt to
better target checks.

Andy Whitcroft (12):
      Version: 0.08
      only apply printk checks where there is a string literal
      allow suppression of errors for when no patch is found
      warn about multiple assignments
      warn on declaration of multiple variables
      check for kfree() with needless null check
      check for single statement braced blocks
      check for aggregate initialisation on the next line
      handle the => operator
      check for spaces between function name and open parenthesis
      move to explicit Check: entries in feature-removal-schedule.txt
      handle pointer attributes

Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Typo: fro -> from

Signed-off-by: Rolf Eike Beer <eike-kernel@sf-tec.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

add POSIX clocks and timers maintainer

Update the MAINTAINERS file: Thomas Gleixner has been the de-facto
maintainer of POSIX timers and clocks for quite some time.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel/relay.c: make functions static

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Cc: Tom Zanussi <zanussi@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump masking: documentation for /proc/pid/coredump_filter

This patch adds the documentation for /proc/<pid>/coredump_filter.

Signed-off-by: Hidehiro Kawai <hidehiro.kawai.ez@hitachi.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: David Howells <dhowells@redhat.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: "Randy.Dunlap" <rdunlap@xenotime.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump masking: ELF-FDPIC: enable core dump filtering

This patch enables core dump filtering for ELF-FDPIC-formatted core file.

Signed-off-by: Hidehiro Kawai <hidehiro.kawai.ez@hitachi.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: David Howells <dhowells@redhat.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump masking: ELF-FDPIC: remove an unused argument

This patch removes an unused argument from elf_fdpic_dump_segments().

Signed-off-by: Hidehiro Kawai <hidehiro.kawai.ez@hitachi.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: David Howells <dhowells@redhat.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump masking: ELF: enable core dump filtering

This patch enables core dump filtering for ELF-formatted core file.

Signed-off-by: Hidehiro Kawai <hidehiro.kawai.ez@hitachi.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: David Howells <dhowells@redhat.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump masking: add an interface for core dump filter

This patch adds an interface to set/reset flags which determines each memory
segment should be dumped or not when a core file is generated.

/proc/<pid>/coredump_filter file is provided to access the flags.  You can
change the flag status for a particular process by writing to or reading from
the file.

The flag status is inherited to the child process when it is created.

Signed-off-by: Hidehiro Kawai <hidehiro.kawai.ez@hitachi.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: David Howells <dhowells@redhat.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump masking: reimplementation of dumpable using two flags

This patch changes mm_struct.dumpable to a pair of bit flags.

set_dumpable() converts three-value dumpable to two flags and stores it into
lower two bits of mm_struct.flags instead of mm_struct.dumpable.
get_dumpable() behaves in the opposite way.

[akpm@linux-foundation.org: export set_dumpable]
Signed-off-by: Hidehiro Kawai <hidehiro.kawai.ez@hitachi.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: David Howells <dhowells@redhat.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump masking: bound suid_dumpable sysctl

This patch series is version 5 of the core dump masking feature, which
controls which VMAs should be dumped based on their memory types and
per-process flags.

I adopted most of Andrew's suggestion at the previous version.  He also
suggested using system call instead of /proc/<pid>/ interface, I decided to
use the latter continuously because adding new system call with pid argument
will give a big impact on the kernel.

You can access the per-process flags via /proc/<pid>/coredump_filter
interface.  coredump_filter represents a bitmask of memory types, and if a bit
is set, VMAs of corresponding memory type are written into a core file when
the process is dumped.  The bitmask is inherited from the parent process when
a process is created.

The original purpose is to avoid longtime system slowdown when a number of
processes which share a huge shared memory are dumped at the same time.  To
achieve this purpose, this patch series adds an ability to suppress dumping
anonymous shared memory for specified processes.  In this version, three other
memory types are also supported.

Here are the coredump_filter bits:
  bit 0: anonymous private memory
  bit 1: anonymous shared memory
  bit 2: file-backed private memory
  bit 3: file-backed shared memory

The default value of coredump_filter is 0x3.  This means the new core dump
routine has the same behavior as conventional behavior by default.

In this version, coredump_filter bits and mm.dumpable are merged into
mm.flags, and it is accessed by atomic bitops.

The supported core file formats are ELF and ELF-FDPIC.  ELF has been tested,
but ELF-FDPIC has not been built and tested because I don't have the test
environment.

This patch limits a value of suid_dumpable sysctl to the range of 0 to 2.

Signed-off-by: Hidehiro Kawai <hidehiro.kawai.ez@hitachi.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: David Howells <dhowells@redhat.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

docbook: don't reference file without kernel-doc

Remove include/linux/rmap.h from kernel-api.tmpl since it no longer
contains kernel-doc.  Fixes this warning:

Warning(linux-2.6.22//include/linux/rmap.h): no structured comments found

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel-doc: fix leading dot in man-mode output

If a parameter description begins with a '.', this indicates a "request"
for "man" mode output (*roff), so it needs special handling.

Problem case is in include/asm-i386/atomic.h for function
atomic_add_unless():
 * @u: ...unless v is equal to u.
This parameter description is currently not printed in man mode output.

[akpm@linux-foundation.org: cleanup]
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel-doc: strip C99 comments

Strip C99-style comments from the input stream.
/*...*/ comments are already stripped.
C99 comments confuse the kernel-doc script.

Also update some comments.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel-doc: fix unnamed struct/union warning

Fix kernel-doc warning:
Warning(linux-2.6.22-rc2-git2/include/linux/skbuff.h:316): No description found for parameter '}'

which is caused by nested anonymous structs/unions ending with:
  };
};

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel-doc: add tools doc in Makefile

Add kernel-doc tools info in Makefile.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs: remove path_walk export

Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@suse.de>
Cc: Michael Halcrow <mhalcrow@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs: mark link_path_walk static

Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@suse.de>
Cc: Michael Halcrow <mhalcrow@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

nfsctl: use vfs_path_lookup

use vfs_path_lookup instead of open-coding the necessary functionality.

Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Acked-by: NeilBrown <neilb@suse.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Michael Halcrow <mhalcrow@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sunrpc: use vfs_path_lookup

use vfs_path_lookup instead of open-coding the necessary functionality.

Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Acked-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Christoph Hellwig <hch@lst.de>
Cc: Neil Brown <neilb@suse.de>
Cc: Michael Halcrow <mhalcrow@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs: introduce vfs_path_lookup

Stackable file systems, among others, frequently need to lookup paths or
path components starting from an arbitrary point in the namespace
(identified by a dentry and a vfsmount).  Currently, such file systems use
lookup_one_len, which is frowned upon [1] as it does not pass the lookup
intent along; not passing a lookup intent, for example, can trigger BUG_ON's
when stacking on top of NFSv4.

The first patch introduces a new lookup function to allow lookup starting
from an arbitrary point in the namespace.  This approach has been suggested
by Christoph Hellwig [2].

The second patch changes sunrpc to use vfs_path_lookup.

The third patch changes nfsctl.c to use vfs_path_lookup.

The fourth patch marks link_path_walk static.

The fifth, and last patch, unexports path_walk because it is no longer
unnecessary to call it directly, and using the new vfs_path_lookup is
cleaner.

For example, the following snippet of code, looks up "some/path/component"
in a directory pointed to by parent_{dentry,vfsmnt}:

err = vfs_path_lookup(parent_dentry, parent_vfsmnt,
      "some/path/component", 0, &nd);
if (!err) {
/* exits */

...

/* once done, release the references */
path_release(&nd);
} else if (err == -ENOENT) {
/* doesn't exist */
} else {
/* other error */
}

VFS functions such as lookup_create can be used on the nameidata structure
to pass the create intent to the file system.

Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@suse.de>
Cc: Michael Halcrow <mhalcrow@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: variable length argument support

Remove the arg+env limit of MAX_ARG_PAGES by copying the strings directly from
the old mm into the new mm.

We create the new mm before the binfmt code runs, and place the new stack at
the very top of the address space.  Once the binfmt code runs and figures out
where the stack should be, we move it downwards.

It is a bit peculiar in that we have one task with two mm's, one of which is
inactive.

[a.p.zijlstra@chello.nl: limit stack size]
Signed-off-by: Ollie Wild <aaw@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: <linux-arch@vger.kernel.org>
Cc: Hugh Dickins <hugh@veritas.com>
[bunk@stusta.de: unexport bprm_mm_init]
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

audit: rework execve audit

The purpose of audit_bprm() is to log the argv array to a userspace daemon at
the end of the execve system call.  Since user-space hasn't had time to run,
this array is still in pristine state on the process' stack; so no need to
copy it, we can just grab it from there.

In order to minimize the damage to audit_log_*() copy each string into a
temporary kernel buffer first.

Currently the audit code requires that the full argument vector fits in a
single packet.  So currently it does clip the argv size to a (sysctl) limit,
but only when execve auditing is enabled.

If the audit protocol gets extended to allow for multiple packets this check
can be removed.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ollie Wild <aaw@google.com>
Cc: <linux-audit@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arch: personality independent stack top

New arch macro STACK_TOP_MAX it gives the larges valid stack address for the
architecture in question.

It differs from STACK_TOP in that it will not distinguish between
personalities but will always return the largest possible address.

This is used to create the initial stack on execve, which we will move down to
the proper location once the binfmt code has figured out where that is.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ollie Wild <aaw@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

use the new percpu interface for shared data

Currently most of the per cpu data, which is accessed by different cpus,
has a ____cacheline_aligned_in_smp attribute.  Move all this data to the
new per cpu shared data section: .data.percpu.shared_aligned.

This will seperate the percpu data which is referenced frequently by other
cpus from the local only percpu data.

Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Acked-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

define new percpu interface for shared data

per cpu data section contains two types of data.  One set which is
exclusively accessed by the local cpu and the other set which is per cpu,
but also shared by remote cpus.  In the current kernel, these two sets are
not clearely separated out.  This can potentially cause the same data
cacheline shared between the two sets of data, which will result in
unnecessary bouncing of the cacheline between cpus.

One way to fix the problem is to cacheline align the remotely accessed per
cpu data, both at the beginning and at the end.  Because of the padding at
both ends, this will likely cause some memory wastage and also the
interface to achieve this is not clean.

This patch:

Moves the remotely accessed per cpu data (which is currently marked
as ____cacheline_aligned_in_smp) into a different section, where all the data
elements are cacheline aligned. And as such, this differentiates the local
only data and remotely accessed data cleanly.

Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Acked-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: <linux-arch@vger.kernel.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

jprobes: make jprobes a little safer for users

I realise jprobes are a razor-blades-included type of interface, but that
doesn't mean we can't try and make them safer to use.  This guy I know once
wrote code like this:

struct jprobe jp = { .kp.symbol_name = "foo", .entry = "jprobe_foo" };

And then his kernel exploded. Oops.

This patch adds an arch hook, arch_deref_entry_point() (I don't like it
either) which takes the void * in a struct jprobe, and gives back the text
address that it represents.

We can then use that in register_jprobe() to check that the entry point we're
passed is actually in the kernel text, rather than just some random value.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Cc: Prasanna S Panchamukhi <prasanna@in.ibm.com>
Acked-by: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

jprobes: remove JPROBE_ENTRY()

AFAICT now that jprobe.entry is a void *, JPROBE_ENTRY doesn't do anything
useful - so remove it ..

I've left a do-nothing version so that out-of-tree jprobes code will still
compile without modifications.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Cc: Prasanna S Panchamukhi <prasanna@in.ibm.com>
Acked-by: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

jprobes: make struct jprobe.entry a void *

Currently jprobe.entry is a kprobe_opcode_t *, but that's a lie.  On some
platforms it doesn't point to an opcode at all, it points to a function
descriptor.

It's really a pointer to something that the arch code can turn into a function
entry point.  And that's what actually happens, none of the generic code ever
looks at jprobe.entry, it's only ever dereferenced by arch code.

So just make it a void *.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Cc: Prasanna S Panchamukhi <prasanna@in.ibm.com>
Acked-by: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

readahead: sanify file_ra_state names

Rename some file_ra_state variables and remove some accessors.

It results in much simpler code.
Kudos to Rusty!

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

readahead: split ondemand readahead interface into two functions

Split ondemand readahead interface into two functions.  I think this makes it
a little clearer for non-readahead experts (like Rusty).

Internally they both call ondemand_readahead(), but the page argument is
changed to an obvious boolean flag.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: share PG_readahead and PG_reclaim

Share the same page flag bit for PG_readahead and PG_reclaim.

One is used only on file reads, another is only for emergency writes.  One
is used mostly for fresh/young pages, another is for old pages.

Combinations of possible interactions are:

a) clear PG_reclaim => implicit clear of PG_readahead
it will delay an asynchronous readahead into a synchronous one
it actually does _good_ for readahead:
the pages will be reclaimed soon, it's readahead thrashing!
in this case, synchronous readahead makes more sense.

b) clear PG_readahead => implicit clear of PG_reclaim
one(and only one) page will not be reclaimed in time
it can be avoided by checking PageWriteback(page) in readahead first

c) set PG_reclaim => implicit set of PG_readahead
will confuse readahead and make it restart the size rampup process
it's a trivial problem, and can mostly be avoided by checking
PageWriteback(page) first in readahead

d) set PG_readahead => implicit set of PG_reclaim
PG_readahead will never be set on already cached pages.
PG_reclaim will always be cleared on dirtying a page.
so not a problem.

In summary,
a)   we get better behavior
b,d) possible interactions can be avoided
c)   racy condition exists that might affect readahead, but the chance
     is _really_ low, and the hurt on readahead is trivial.

Compound pages also use PG_reclaim, but for now they do not interact with
reclaim/readahead code.

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

readahead: pass real splice size

Pass real splice size to page_cache_readahead_ondemand().

The splice code works in chunks of 16 pages internally.  The readahead code
should be told of the overall splice size, instead of the internal chunk size.
 Otherwize bad things may happen.  Imagine some 17-page random splice reads.
The code before this patch will result in two readahead calls: readahead(16);
readahead(1); That leads to one 16-page I/O and one 32-page I/O: one extra I/O
and 31 readahead miss pages.

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

readahead: move synchronous readahead call out of splice loop

Move synchronous page_cache_readahead_ondemand() call out of splice loop.

This avoids one pointless page allocation/insertion in case of non-zero
ra_pages, or many pointless readahead calls in case of zero ra_pages.

Note that if a user sets ra_pages to less than PIPE_BUFFERS=16 pages, he will
not get expected readahead behavior anyway.  The splice code works in batches
of 16 pages, which can be taken as another form of synchronous readahead.

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

readahead: remove the old algorithm

Remove the old readahead algorithm.

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Cc: Steven Pratt <slpratt@austin.ibm.com>
Cc: Ram Pai <linuxram@us.ibm.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

readahead: convert ext3/ext4 invocations

Convert ext3/ext4 dir reads to use on-demand readahead.

Readahead for dirs operates _not_ on file level, but on blockdev level.  This
makes a difference when the data blocks are not continuous.  And the read
routine is somehow opaque: there's no handy info about the status of current
page.  So a simplified call scheme is employed: to call into readahead
whenever the current page falls out of readahead windows.

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Cc: Steven Pratt <slpratt@austin.ibm.com>
Cc: Ram Pai <linuxram@us.ibm.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

readahead: convert splice invocations

Convert splice reads to use on-demand readahead.

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Cc: Steven Pratt <slpratt@austin.ibm.com>
Cc: Ram Pai <linuxram@us.ibm.com>
Cc: Jens Axboe <axboe@suse.de>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

readahead: convert filemap invocations

Convert filemap reads to use on-demand readahead.

The new call scheme is to
- call readahead on non-cached page
- call readahead on look-ahead page
- update prev_index when finished with the read request

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Cc: Steven Pratt <slpratt@austin.ibm.com>
Cc: Ram Pai <linuxram@us.ibm.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

readahead: on-demand readahead logic

This is a minimal readahead algorithm that aims to replace the current one.
It is more flexible and reliable, while maintaining almost the same behavior
and performance.  Also it is full integrated with adaptive readahead.

It is designed to be called on demand:
- on a missing page, to do synchronous readahead
- on a lookahead page, to do asynchronous readahead

In this way it eliminated the awkward workarounds for cache hit/miss,
readahead thrashing, retried read, and unaligned read.  It also adopts the
data structure introduced by adaptive readahead, parameterizes readahead
pipelining with `lookahead_index', and reduces the current/ahead windows to
one single window.

HEURISTICS

The logic deals with four cases:

- sequential-next
found a consistent readahead window, so push it forward

- random
standalone small read, so read as is

- sequential-first
create a new readahead window for a sequential/oversize request

- lookahead-clueless
hit a lookahead page not associated with the readahead window,
so create a new readahead window and ramp it up

In each case, three parameters are determined:

- readahead index: where the next readahead begins
- readahead size:  how much to readahead
- lookahead size:  when to do the next readahead (for pipelining)

BEHAVIORS

The old behaviors are maximally preserved for trivial sequential/random reads.
Notable changes are:

- It no longer imposes strict sequential checks.
  It might help some interleaved cases, and clustered random reads.
  It does introduce risks of a random lookahead hit triggering an
  unexpected readahead. But in general it is more likely to do good
  than to do evil.

- Interleaved reads are supported in a minimal way.
  Their chances of being detected and proper handled are still low.

- Readahead thrashings are better handled.
  The current readahead leads to tiny average I/O sizes, because it
  never turn back for the thrashed pages.  They have to be fault in
  by do_generic_mapping_read() one by one.  Whereas the on-demand
  readahead will redo readahead for them.

OVERHEADS

The new code reduced the overheads of

- excessively calling the readahead routine on small sized reads
  (the current readahead code insists on seeing all requests)

- doing a lot of pointless page-cache lookups for small cached files
  (the current readahead only turns itself off after 256 cache hits,
  unfortunately most files are < 1MB, so never see that chance)

That accounts for speedup of
- 0.3% on 1-page sequential reads on sparse file
- 1.2% on 1-page cache hot sequential reads
- 3.2% on 256-page cache hot sequential reads
- 1.3% on cache hot `tar /lib`

However, it does introduce one extra page-cache lookup per cache miss, which
impacts random reads slightly. That's 1% overheads for 1-page random reads on
sparse file.

PERFORMANCE

The basic benchmark setup is
- 2.6.20 kernel with on-demand readahead
- 1MB max readahead size
- 2.9GHz Intel Core 2 CPU
- 2GB memory
- 160G/8M Hitachi SATA II 7200 RPM disk

The benchmarks show that
- it maintains the same performance for trivial sequential/random reads
- sysbench/OLTP performance on MySQL gains up to 8%
- performance on readahead thrashing gains up to 3 times

iozone throughput (KB/s): roughly the same
==========================================
iozone -c -t1 -s 4096m -r 64k

       2.6.20          on-demand      gain
first run
  "  Initial write "   61437.27        64521.53      +5.0%
  "        Rewrite "   47893.02        48335.20      +0.9%
  "           Read "   62111.84        62141.49      +0.0%
  "        Re-read "   62242.66        62193.17      -0.1%
  "   Reverse Read "   50031.46        49989.79      -0.1%
  "    Stride read "    8657.61         8652.81      -0.1%
  "    Random read "   13914.28        13898.23      -0.1%
  " Mixed workload "   19069.27        19033.32      -0.2%
  "   Random write "   14849.80        14104.38      -5.0%
  "         Pwrite "   62955.30        65701.57      +4.4%
  "          Pread "   62209.99        62256.26      +0.1%

second run
  "  Initial write "   60810.31        66258.69      +9.0%
  "        Rewrite "   49373.89        57833.66     +17.1%
  "           Read "   62059.39        62251.28      +0.3%
  "        Re-read "   62264.32        62256.82      -0.0%
  "   Reverse Read "   49970.96        50565.72      +1.2%
  "    Stride read "    8654.81         8638.45      -0.2%
  "    Random read "   13901.44        13949.91      +0.3%
  " Mixed workload "   19041.32        19092.04      +0.3%
  "   Random write "   14019.99        14161.72      +1.0%
  "         Pwrite "   64121.67        68224.17      +6.4%
  "          Pread "   62225.08        62274.28      +0.1%

In summary, writes are unstable, reads are pretty close on average:

  access pattern  2.6.20  on-demand   gain
   Read  62085.61  62196.38  +0.2%
Re-read  62253.49  62224.99  -0.0%
   Reverse Read  50001.21  50277.75  +0.6%
    Stride read   8656.21   8645.63  -0.1%
    Random read  13907.86  13924.07  +0.1%
   Mixed workload  19055.29  19062.68  +0.0%
  Pread  62217.53  62265.27  +0.1%

aio-stress: roughly the same
============================
aio-stress -l -s4096 -r128 -t1 -o1 knoppix511-dvd-cn.iso
aio-stress -l -s4096 -r128 -t1 -o3 knoppix511-dvd-cn.iso

2.6.20      on-demand  delta
sequential  92.57s      92.54s    -0.0%
random 311.87s     312.15s    +0.1%

sysbench fileio: roughly the same
=================================
sysbench --test=fileio --file-io-mode=async --file-test-mode=rndrw \
 --file-total-size=4G --file-block-size=64K \
 --num-threads=001 --max-requests=10000 --max-time=900 run

threads    2.6.20   on-demand    delta
first run
      1   59.1974s    59.2262s  +0.0%
      2   58.0575s    58.2269s  +0.3%
      4   48.0545s    47.1164s  -2.0%
      8   41.0684s    41.2229s  +0.4%
     16   35.8817s    36.4448s  +1.6%
     32   32.6614s    32.8240s  +0.5%
     64   23.7601s    24.1481s  +1.6%
    128   24.3719s    23.8225s  -2.3%
    256   23.2366s    22.0488s  -5.1%

second run
      1   59.6720s    59.5671s  -0.2%
      8   41.5158s    41.9541s  +1.1%
     64   25.0200s    23.9634s  -4.2%
    256   22.5491s    20.9486s  -7.1%

Note that the numbers are not very stable because of the writes.
The overall performance is close when we sum all seconds up:

                sum all up               495.046s    491.514s   -0.7%

sysbench oltp (trans/sec): up to 8% gain
========================================
sysbench --test=oltp --oltp-table-size=10000000 --oltp-read-only \
 --mysql-socket=/var/run/mysqld/mysqld.sock \
 --mysql-user=root --mysql-password=readahead \
 --num-threads=064 --max-requests=10000 --max-time=900 run

10000-transactions run
threads    2.6.20   on-demand    gain
      1     62.81       64.56   +2.8%
      2     67.97       70.93   +4.4%
      4     81.81       85.87   +5.0%
      8     94.60       97.89   +3.5%
     16     99.07      104.68   +5.7%
     32     95.93      104.28   +8.7%
     64     96.48      103.68   +7.5%
5000-transactions run
      1     48.21       48.65   +0.9%
      8     68.60       70.19   +2.3%
     64     70.57       74.72   +5.9%
2000-transactions run
      1     37.57       38.04   +1.3%
      2     38.43       38.99   +1.5%
      4     45.39       46.45   +2.3%
      8     51.64       52.36   +1.4%
     16     54.39       55.18   +1.5%
     32     52.13       54.49   +4.5%
     64     54.13       54.61   +0.9%

That's interesting results. Some investigations show that
- MySQL is accessing the db file non-uniformly: some parts are
  more hot than others
- It is mostly doing 4-page random reads, and sometimes doing two
  reads in a row, the latter one triggers a 16-page readahead.
- The on-demand readahead leaves many lookahead pages (flagged
  PG_readahead) there. Many of them will be hit, and trigger
  more readahead pages. Which might save more seeks.
- Naturally, the readahead windows tend to lie in hot areas,
  and the lookahead pages in hot areas is more likely to be hit.
- The more overall read density, the more possible gain.

That also explains the adaptive readahead tricks for clustered random reads.

readahead thrashing: 3 times better
===================================
We boot kernel with "mem=128m single", and start a 100KB/s stream on every
second, until reaching 200 streams.

      max throughput     min avg I/O size
2.6.20:            5MB/s               16KB
on-demand:        15MB/s              140KB

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Cc: Steven Pratt <slpratt@austin.ibm.com>
Cc: Ram Pai <linuxram@us.ibm.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

readahead: data structure and routines

Extend struct file_ra_state to support the on-demand readahead logic.  Also
define some helpers for it.

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Cc: Steven Pratt <slpratt@austin.ibm.com>
Cc: Ram Pai <linuxram@us.ibm.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

readahead: MIN_RA_PAGES/MAX_RA_PAGES macros

Define two convenient macros for read-ahead:
- MAX_RA_PAGES: rounded down counterpart of VM_MAX_READAHEAD
- MIN_RA_PAGES: rounded _up_ counterpart of VM_MIN_READAHEAD

Note that the rounded up MIN_RA_PAGES will work flawlessly with _large_
page sizes like 64k.

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Cc: Steven Pratt <slpratt@austin.ibm.com>
Cc: Ram Pai <linuxram@us.ibm.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

readahead: add look-ahead support to __do_page_cache_readahead()

Add look-ahead support to __do_page_cache_readahead().

It works by
- mark the Nth backwards page with PG_readahead,
(which instructs the page's first reader to invoke readahead)
- and only do the marking for newly allocated pages.
(to prevent blindly doing readahead on already cached pages)

Look-ahead is a technique to achieve I/O pipelining:

While the application is working through a chunk of cached pages, the kernel
reads-ahead the next chunk of pages _before_ time of need.  It effectively
hides low level I/O latencies to high level applications.

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Cc: Steven Pratt <slpratt@austin.ibm.com>
Cc: Ram Pai <linuxram@us.ibm.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

readahead: introduce PG_readahead

Introduce a new page flag: PG_readahead.

It acts as a look-ahead mark, which tells the page reader: Hey, it's time to
invoke the read-ahead logic.  For the sake of I/O pipelining, don't wait until
it runs out of cached pages!

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Cc: Steven Pratt <slpratt@austin.ibm.com>
Cc: Ram Pai <linuxram@us.ibm.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

AIO sparse fix (type of ki_flags)

Fix type issue reported by latest 'sparse': kiocb.ki_flags should be
"unsigned long" (not "long"), to match bitop type signature.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Benjamin LaHaise <bcrl@kvack.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>