pids: fix fork_idle() to setup ->pids correctly

copy_process(pid => &init_struct_pid) doesn't do attach_pid/etc.

It shouldn't, but this means that the idle threads run with the wrong
pids copied from the caller's task_struct. In x86 case the caller is
either kernel_init() thread or keventd.

In particular, this means that after the series of cpu_up/cpu_down an
idle thread (which never exits) can run with .pid pointing to nowhere.

Change fork_idle() to initialize idle->pids[] correctly. We only set
.pid = &init_struct_pid but do not add .node to list, INIT_TASK() does
the same for the boot-cpu idle thread (swapper).

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Cedric Le Goater <clg@fr.ibm.com>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Herbert Poetzl <herbert@13thfloor.at>
Cc: Mathias Krause <Mathias.Krause@secunet.com>
Acked-by: Roland McGrath <roland@redhat.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: Sukadev Bhattiprolu <sukadev@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pids: init_struct_pid.tasks should never see the swapper process

"statically initialize struct pid for swapper" commit 820e45db says:

Statically initialize a struct pid for the swapper process (pid_t == 0)
and attach it to init_task.  This is needed so task_pid(), task_pgrp()
and task_session() interfaces work on the swapper process also.

OK, but:

- it doesn't make sense to add init_task.pids[].node into
  init_struct_pid.tasks[], and in fact this just wrong.

  idle threads are special, they shouldn't be visible on any
  global list. In particular do_each_pid_task(init_struct_pid)
  shouldn't see swapper.

  This is the actual reason why kill(0, SIGKILL) from /sbin/init
  (which starts with 0,0 special pids) crashes the kernel. The
  signal sent to pgid/sid == 0 must never see idle threads, even
  if the previous patch fixed the crash itself.

- we have other idle threads running on the non-boot CPUs, see
  the next patch.

Change INIT_STRUCT_PID/INIT_PID_LINK to create the empty/unhashed
hlist_head/hlist_node. Like any other idle thread swapper can never exit,
so detach_pid()->__hlist_del() is not possible, but we could change
INIT_PID_LINK() to set pprev = &next if needed.

All we need is the valid swapper->pids[].pid == &init_struct_pid.

Reported-by: Mathias Krause <mathias.krause@secunet.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Cedric Le Goater <clg@fr.ibm.com>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Herbert Poetzl <herbert@13thfloor.at>
Cc: Mathias Krause <Mathias.Krause@secunet.com>
Acked-by: Roland McGrath <roland@redhat.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: Sukadev Bhattiprolu <sukadev@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

INIT_TASK() should initialize ->thread_group list

The trivial /sbin/init doing

int main(void)
{
kill(0, SIGKILL)
}

crashes the kernel.

This happens because __kill_pgrp_info(init_struct_pid) also sends SIGKILL
to the swapper process which runs with the uninitialized ->thread_group.

Change INIT_TASK() to initialize ->thread_group properly.

Note: the real problem is that the swapper process must not be visible to
signals, see the next patch. But this change is right anyway and fixes
the crash.

Reported-and-tested-by: Mathias Krause <mathias.krause@secunet.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Cedric Le Goater <clg@fr.ibm.com>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Herbert Poetzl <herbert@13thfloor.at>
Cc: Mathias Krause <Mathias.Krause@secunet.com>
Acked-by: Roland McGrath <roland@redhat.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Acked-by: Sukadev Bhattiprolu <sukadev@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pids: increase pid_max based on num_possible_cpus

On a system with a substantial number of processors, the early default
pid_max of 32k will not be enough.  A system with 1664 CPU's, there are
25163 processes started before the login prompt.  It's estimated that with
2048 CPU's we will pass the 32k limit.  With 4096, we'll reach that limit
very early during the boot cycle, and processes would stall waiting for an
available pid.

This patch increases the early maximum number of pids available, and
increases the minimum number of pids that can be set during runtime.

[akpm@linux-foundation.org: fix warnings]
Signed-off-by: Hedi Berriche <hedi@sgi.com>
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Robin Holt <holt@sgi.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Greg KH <gregkh@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: John Stoffel <john@stoffel.org>
Cc: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rapidio: fix maintenance access to higher memory areas

Fix the maintenance access functions to farend RapidIO devices.
1. Fixed shift of the given offset, to open the maintenance window
2. Mask offset to limit access to the opened maintenance window
3. Added extended destid part to rowtear register, required for 16bit mode

This method is matching maintenance transactions generation described
by Freescale in the appnote AN2932. With this modification full access
to a 16MB maintenance window is possible, this patch is required for
IDT cps switches. For easier handling of the access routines, the
access was limited to aligned memory regions. This should be no problem
because all registers are 32bit wide.

Signed-off-by: Thomas Moll <thomas.moll@sysgo.com>
Tested-by: Alexandre Bounine <alexandre.bounine@idt.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rapidio: use default route value for CPS switches

Fix to use correct default value for routing table entries.

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Cc: Thomas Moll <thomas.moll@sysgo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rapidio: add switch domain routines

Add switch specific domain routines required for 16-bit routing support in
switches with hierarchical implementation of routing tables.

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Cc: Thomas Moll <thomas.moll@sysgo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rapidio: modify initialization of switch operations

Modify the way how RapidIO switch operations are declared.  Multiple
assignments through the linker script replaced by single initialization
call.

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Cc: Thomas Moll <thomas.moll@sysgo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rapidio: add debug configuration option

Add debug configuration option for RapidIO subsystem.

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Cc: Thomas Moll <thomas.moll@sysgo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rapidio: fix typos and minor edits

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Cc: Thomas Moll <thomas.moll@sysgo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rapidio: add enabling SRIO port RX and TX

Add the functionality to enable Input receiver and Output transmitter of
every port, to allow non-maintenance traffic.

Signed-off-by: Thomas Moll <thomas.moll@sysgo.com>
Signed-off-by: Alexandre Bounine <abounine@tundra.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rapidio, powerpc/85xx: Add MChk handler for SRIO port

Add Machine Check exception handling into RapidIO port driver for
Freescale SoCs (MPC85xx).

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Tested-by: Thomas Moll <thomas.moll@sysgo.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rapidio, powerpc/85xx: add Port-Write message handler for SRIO port

Add RapidIO Port-Write message handler for Freescale SoCs with RapidIO
port.

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Tested-by: Thomas Moll <thomas.moll@sysgo.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Kumar Gala <galak@kernel.crashing.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>

rapidio: add Port-Write handling for EM

Add RapidIO Port-Write message handling in the context of Error
   Management Extensions Specification Rev.1.3.

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Tested-by: Thomas Moll <thomas.moll@sysgo.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rapidio: add switch locking during discovery

Add switch access locking during RapidIO discovery.  Access lock is
required when reading switch routing table contents due to indexed
mechanism of RT addressing.

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Tested-by: Thomas Moll <thomas.moll@sysgo.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rapidio: add IDT CPS/TSI switches

Extentions to RapidIO switch support:

1. modify switch route operation declarations to allow using single
   switch-specific file for family of switches that share the same route
   table operations.

2. add standard route table operations for switches that that support
   route table manipulation registers as defined in the Rev.1.3 of RapidIO
   specification.

3. add clear-route-table operation for switches

4. add CPSxx and TSIxxx families of RapidIO switches

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Tested-by: Thomas Moll <thomas.moll@sysgo.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/char/applicom.c: use memdup_user

Use memdup_user when user data is immediately copied into the
allocated region.

The semantic patch that makes this change is as follows:
(http://coccinelle.lip6.fr/)

// <smpl>
@@
expression from,to,size,flag;
position p;
identifier l1,l2;
@@

-  to = \(kmalloc@p\|kzalloc@p\)(size,flag);
+  to = memdup_user(from,size);
   if (
-      to==NULL
+      IS_ERR(to)
                 || ...) {
   <+... when != goto l1;
-  -ENOMEM
+  PTR_ERR(to)
   ...+>
   }
-  if (copy_from_user(to, from, size) != 0) {
-    <+... when != goto l2;
-    -EFAULT
-    ...+>
-  }
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/char/ppdev.c: use kasprintf

kasprintf combines kmalloc and sprintf, and takes care of the size
calculation itself.

The semantic patch that makes this change is as follows:
(http://coccinelle.lip6.fr/)

// <smpl>
@@
expression a,flag;
expression list args;
statement S;
@@

  a =
-  \(kmalloc\|kzalloc\)(...,flag)
+  kasprintf(flag,args)
  <... when != a
  if (a == NULL || ...) S
  ...>
- sprintf(a,args);
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Cc: Michael Buesch <mb@bu3sch.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

char drivers: RAM oops/panic logger

Ramoops, like mtdoops, can log oops/panic information but in RAM.  It can
be used with persistent RAM for systems without flash support.  In
addition, for this systems, with this driver, it's no more needed add to
the kernel the mtd subsystem with advantage in footprint.

It can be used in a very easy way with persistent RAM for systems without
flash support.  For these systems, with this driver, it is no longer
required to cinlude mtd subsystem with an advantage in footprint.  In
addition, you can save flash space and store this information only in RAM.

Signed-off-by: Marco Stornelli <marco.stornelli@gmail.com>
Cc: Simon Kagstrom <simon.kagstrom@netinsight.net>
Cc: David Woodhouse <David.Woodhouse@intel.com>
Cc; Anders Grafstrom <anders.grafstrom@netinsight.net>
Cc: Yuasa Yoichi <yuasa@linux-mips.org>
Cc: Jamie Lokier <jamie@shareable.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipmi: handle run_to_completion properly in deliver_recv_msg()

If run_to_completion flag is set, it means that we are running in a
single-threaded mode, and thus no locks are held.

This fixes a deadlock when IPMI notifier is being called during panic.

Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Acked-by: Corey Minyard <minyard@acm.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipmi: update driver to use dev_printk and its constructs

Update core IPMI driver printk()'s with dev_printk(), and its constructs,
to provide additional device topology information.

An example of the additional device topology for a PNP device -
  ipmi_si 00:02: probing via ACPI
  ipmi_si 00:02: [io  0x0ca2-0x0ca3] regsize 1 spacing 1 irq 0
  ipmi_si 00:02: Found new BMC (man_id: 0x00000b, prod_id: 0x0000, ...
  ipmi_si 00:02: IPMI kcs interface initialized

and for a PCI device -
  ipmi_si 0000:01:04.6: probing via PCI
  ipmi_si 0000:01:04.6: PCI INT A -> GSI 21 (level, low) -> IRQ 21
  ipmi_si 0000:01:04.6: [mem 0xf1ef0000-0xf1ef00ff] regsize 1 spaci...
  ipmi_si 0000:01:04.6: IPMI kcs interface initialized

[minyard@acm.org: rework to fix rejects, extended it a bit]
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Myron Stowe <myron.stowe@hp.com>
Signed-off-by: Corey Minyard <minyard@acm.org>
Cc: Zhao Yakui <yakui.zhao@intel.com>
Cc: Len Brown <lenb@kernel.org>
Cc: Bjorn Helgaas <bjorn.helgaas@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipmi: convert tracking of the ACPI device pointer to a PNP device

Convert PNP patch (git 9e368fa011d4e0aa050db348d69514900520e40b) to
maintain a pointer to a PNP device, 'pnp_dev', instead of the ACPI device,
'acpi_dev', that is currently being tracked with PNP based IPMI device
discovery.

Signed-off-by: Myron Stowe <myron.stowe@hp.com>
Acked-by: Zhao Yakui <yakui.zhao@intel.com>
Acked-by: Corey Minyard <minyard@acm.org>
Cc: Len Brown <lenb@kernel.org>
Cc: Bjorn Helgaas <bjorn.helgaas@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipmi: change timeout and event poll to one second

The timeouts in IPMI are in the 1-5 second range in message handling, so a
1 second timeout is a reasonable thing to do.  This should help with
reducing power consumption on idle systems.

Signed-off-by: Corey Minyard <cminyard@mvista.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipmi: attempt to register multiple SIs of the same type

Some odd systems may have multiple BMCs, and we want to be able to support
them.  Let's make the assumption that if a system legitimately has
multiple BMCs then each BMC's SI will be of the same type, and also that
we won't see multiple SIs of the same type unless we have multiple BMCs.
If these hold true then we should register all SIs of the same type.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Corey Minyard <cminyard@mvista.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipmi: reduce polling

We can reasonably alter the poll rate depending on whether we're
performing a transaction or merely waiting for an event.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Corey Minyard <cminyard@mvista.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipmi: reduce polling when interrupts are available

If we're not currently in the middle of a transaction, and if we have
interrupts, there's no real reason to poll the controller more frequently
than the core IPMI code does.  Set the interrupt_disabled flag
appropriately as the interrupt state changes, and make the timeout code
reset itself only if the transaction is incomplete or we have no
interrupts.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Corey Minyard <cminyard@mvista.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipmi: change device discovery order

The ipmi spec provides an ordering for si discovery.  Change the driver to
match, with the exception of preferring smbios to SPMI as HPs (at least)
contain accurate information in the former but not the latter.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Corey Minyard <cminyard@mvista.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipmi: only register one si per bmc

Only register one si per bmc.  Use any user-provided devices first,
followed by the first device with an irq, followed by the first device
discovered.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Corey Minyard <cminyard@mvista.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipmi: split device discovery and registration

The ipmi spec indicates that we should only make use of one si per bmc, so
separate device discovery and registration to make that possible.

[thenzl@redhat.com: fix mutex use]
Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Corey Minyard <cminyard@mvista.com>
Signed-off-by: Tomas Henzl <thenzl@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipmi: change addr_source to an enum rather than strings

Switch from a char* to an enum to identify the address source of SIs,
making it easier to handle them appropriately during registration.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Corey Minyard <cminyard@mvista.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc/sem.c: use ERR_CAST

Use ERR_CAST(x) rather than ERR_PTR(PTR_ERR(x)).  The former makes more
clear what is the purpose of the operation, which otherwise looks like a
no-op.

The semantic patch that makes this change is as follows:
(http://coccinelle.lip6.fr/)

// <smpl>
@@
type T;
T x;
identifier f;
@@

T f (...) { <+...
- ERR_PTR(PTR_ERR(x))
+ x
 ...+> }

@@
expression x;
@@

- ERR_PTR(PTR_ERR(x))
+ ERR_CAST(x)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc/sem.c: update description of the implementation

ipc/sem.c begins with a 15 year old description about bugs in the initial
implementation in Linux-1.0.  The patch replaces that with a top level
description of the current code.

A TODO could be derived from this text:

The opengroup man page for semop() does not mandate FIFO.  Thus there is
no need for a semaphore array list of pending operations.

If

- this list is removed
- the per-semaphore array spinlock is removed (possible if there is no
  list to protect)
- sem_otime is moved into the semaphores and calculated on demand during
  semctl()

then the array would be read-mostly - which would significantly improve
scaling for applications that use semaphore arrays with lots of entries.

The price would be expensive semctl() calls:

for(i=0;i<sma->sem_nsems;i++) spin_lock(sma->sem_lock);
<do stuff>
for(i=0;i<sma->sem_nsems;i++) spin_unlock(sma->sem_lock);

I'm not sure if the complexity is worth the effort, thus here is the
documentation of the current behavior first.

Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Zach Brown <zach.brown@oracle.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc/sem.c: cacheline align the ipc spinlock for semaphores

Cacheline align the spinlock for sysv semaphores.  Without the patch, the
spinlock and sem_otime [written by every semop that modified the array]
and sem_base [read in the hot path of try_atomic_semop()] can be in the
same cacheline.

Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Zach Brown <zach.brown@oracle.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc/sem.c: move wake_up_process out of the spinlock section

The wake-up part of semtimedop() consists out of two steps:

- the right tasks must be identified.
- they must be woken up.

Right now, both steps run while the array spinlock is held.  This patch
reorders the code and moves the actual wake_up_process() behind the point
where the spinlock is dropped.

The code also moves setting sem->sem_otime to one place: It does not make
sense to set the last modify time multiple times.

[akpm@linux-foundation.org: repair kerneldoc]
[akpm@linux-foundation.org: fix uninitialised retval]
Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Zach Brown <zach.brown@oracle.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc/sem.c: optimize update_queue() for bulk wakeup calls

The following series of patches tries to fix the spinlock contention
reported by Chris Mason - his benchmark exposes problems of the current
code:

- In the worst case, the algorithm used by update_queue() is O(N^2).
  Bulk wake-up calls can enter this worst case.  The patch series fix
  that.

  Note that the benchmark app doesn't expose the problem, it just should
  be fixed: Real world apps might do the wake-ups in another order than
  perfect FIFO.

- The part of the code that runs within the semaphore array spinlock is
  significantly larger than necessary.

  The patch series fixes that.  This change is responsible for the main
  improvement.

- The cacheline with the spinlock is also used for a variable that is
  read in the hot path (sem_base) and for a variable that is unnecessarily
  written to multiple times (sem_otime).  The last step of the series
  cacheline-aligns the spinlock.

This patch:

The SysV semaphore code allows to perform multiple operations on all
semaphores in the array as atomic operations.  After a modification,
update_queue() checks which of the waiting tasks can complete.

The algorithm that is used to identify the tasks is O(N^2) in the worst
case.  For some cases, it is simple to avoid the O(N^2).

The patch adds a detection logic for some cases, especially for the case
of an array where all sleeping tasks are single sembuf operations and a
multi-sembuf operation is used to wake up multiple tasks.

A big database application uses that approach.

The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of
the patch breaks that.

[akpm@linux-foundation.org: make do_smart_update() static]
Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Zach Brown <zach.brown@oracle.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

idr: fix backtrack logic in idr_remove_all

Currently idr_remove_all will fail with a use after free error if
idr::layers is bigger than 2, which on 32 bit systems corresponds to items
more than 1024.  This is due to stepping back too many levels during
backtracking.  For simplicity let's assume that IDR_BITS=1 -> we have 2
nodes at each level below the root node and each leaf node stores two IDs.
 (In reality for 32 bit systems IDR_BITS=5, with 32 nodes at each sub-root
level and 32 IDs in each leaf node).  The sequence of freeing the nodes at
the moment is as follows:

layer
1 ->                       a(7)
2 ->            b(3)                  c(5)
3 ->        d(1)   e(2)           f(4)    g(6)

Until step 4 things go fine, but then node c is freed, whereas node g
should be freed first.  Since node c contains the pointer to node g we'll
have a use after free error at step 6.

How many levels we step back after visiting the leaf nodes is currently
determined by the msb of the id we are currently visiting:

Step
1.          node d with IDs 0,1 is freed, current ID is advanced to 2.
            msb of the current ID bit 1. This means we need to step back
            1 level to node b and take the next sibling, node e.
2-3.        node e with IDs 2,3 is freed, current ID is 4, msb is bit 2.
            This means we need to step back 2 levels to node a, freeing
            node b on the way.
4-5.        node f with IDs 4,5 is freed, current ID is 6, msb is still
            bit 2. This means we again need to step back 2 levels to node
            a and free c on the way.
6.          We should visit node g, but its pointer is not available as
            node c was freed.

The fix changes how we determine the number of levels to step back.
Instead of deducting this merely from the msb of the current ID, we should
really check if advancing the ID causes an overflow to a bit position
corresponding to a given layer.  In the above example overflow from bit 0
to bit 1 should mean stepping back 1 level.  Overflow from bit 1 to bit 2
should mean stepping back 2 levels and so on.

The fix was tested with IDs up to 1 << 20, which corresponds to 4 layers
on 32 bit systems.

Signed-off-by: Imre Deak <imre.deak@nokia.com>
Reviewed-by: Tejun Heo <tj@kernel.org>
Cc: Eric Paris <eparis@redhat.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: <stable@kernel.org> [2.6.34.1]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpuhotplug: do not need cpu_hotplug_begin() when CONFIG_HOTPLUG_CPU=n

Since when CONFIG_HOTPLUG_CPU=n, get_online_cpus() do nothing, so we don't
need cpu_hotplug_begin() either.

This patch moves cpu_hotplug_begin()/cpu_hotplug_done() into the code
block of CONFIG_HOTPLUG_CPU=y.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Gautham R Shenoy <ego@in.ibm.com>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fault-injection: add CPU notifier error injection module

I used this module to test the series of modification to the cpu notifiers
code.

Example1: inject CPU offline error (-1 == -EPERM)

# modprobe cpu-notifier-error-inject cpu_down_prepare_error=-1
# echo 0 > /sys/devices/system/cpu/cpu1/online
bash: echo: write error: Operation not permitted

Example2: inject CPU online error (-2 == -ENOENT)

# modprobe cpu-notifier-error-inject cpu_up_prepare_error=-2
# echo 1 > /sys/devices/system/cpu/cpu1/online
bash: echo: write error: No such file or directory

[akpm@linux-foundation.org: fix Kconfig help text]
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

md: convert cpu notifier to return encapsulate errno value

By the previous modification, the cpu notifier can return encapsulate
errno value.  This converts the cpu notifiers for raid5.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

s390: convert cpu notifier to return encapsulate errno value

By the previous modification, the cpu notifier can return encapsulate
errno value.  This converts the cpu notifiers for s390.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.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>

ehca: convert cpu notifier to return encapsulate errno value

By the previous modification, the cpu notifier can return encapsulate
errno value. This converts the cpu notifiers for ehca.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
Cc: Christoph Raisch <raisch@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iucv: convert cpu notifier to return encapsulate errno value

By the previous modification, the cpu notifier can return encapsulate
errno value.  This converts the cpu notifiers for iucv.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Ursula Braun <ursula.braun@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

slab: convert cpu notifier to return encapsulate errno value

By the previous modification, the cpu notifier can return encapsulate
errno value.  This converts the cpu notifiers for slab.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel/: convert cpu notifier to return encapsulate errno value

By the previous modification, the cpu notifier can return encapsulate
errno value.  This converts the cpu notifiers for kernel/*.c

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

topology: convert cpu notifier to return encapsulate errno value

By the previous modification, the cpu notifier can return encapsulate
errno value.  This converts the cpu notifiers for topology.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

x86: convert cpu notifier to return encapsulate errno value

By the previous modification, the cpu notifier can return encapsulate
errno value.  This converts the cpu notifiers for msr, cpuid, and
therm_throt.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

notifier: change notifier_from_errno(0) to return NOTIFY_OK

This changes notifier_from_errno(0) to be NOTIFY_OK instead of
NOTIFY_STOP_MASK | NOTIFY_OK.

Currently, the notifiers which return encapsulated errno value have to
do something like this:

err = do_something(); // returns -errno
if (err)
return notifier_from_errno(err);
else
return NOTIFY_OK;

This change makes the above code simple:

err = do_something(); // returns -errno

return return notifier_from_errno(err);

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpu-hotplug: return better errno on cpu hotplug failure

Currently, onlining or offlining a CPU failure by one of the cpu notifiers
error always cause -EINVAL error.  (i.e.  writing 0 or 1 to
/sys/devices/system/cpu/cpuX/online gets EINVAL)

To get better error reporting rather than always getting -EINVAL, This
changes cpu_notify() to return -errno value with notifier_to_errno() and
fix the callers.  Now that cpu notifiers can return encapsulate errno
value.

Currently, all cpu hotplug notifiers return NOTIFY_OK, NOTIFY_BAD, or
NOTIFY_DONE.  So cpu_notify() can returns 0 or -EPERM with this change for
now.

(notifier_to_errno(NOTIFY_OK) == 0, notifier_to_errno(NOTIFY_DONE) == 0,
notifier_to_errno(NOTIFY_BAD) == -EPERM)

Forthcoming patches convert several cpu notifiers to return encapsulate
errno value with notifier_from_errno().

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.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>

cpu-hotplug: introduce cpu_notify(), __cpu_notify(), cpu_notify_nofail()

No functional change.  These are just wrappers of
raw_cpu_notifier_call_chain.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.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>

kcore: add _text to KCORE_TEXT

Extend KCORE_TEXT to cover the pages between _text and _stext, to allow
examining some important page table pages.

`readelf -a` output on x86_64 before and after patch:
  Type           Offset             VirtAddr           PhysAddr
before    LOAD           0x00007fff8100c000 0xffffffff81009000 0x0000000000000000
after     LOAD           0x00007fff81003000 0xffffffff81000000 0x0000000000000000

The newly covered pages are:

0xffffffff81000000 <startup_64> etc.
0xffffffff81001000 <init_level4_pgt>
0xffffffff81002000 <level3_ident_pgt>
0xffffffff81003000 <level3_kernel_pgt>
0xffffffff81004000 <level2_fixmap_pgt>
0xffffffff81005000 <level1_fixmap_pgt>
0xffffffff81006000 <level2_ident_pgt>
0xffffffff81007000 <level2_kernel_pgt>
0xffffffff81008000 <level2_spare_pgt>

Before patch, /proc/kcore shows outdated contents for the above page
table pages, for example:

(gdb) p level3_ident_pgt
$1 = {<text variable, no debug info>} 0xffffffff81002000 <level3_ident_pgt>
(gdb) p/x *((pud_t *)&level3_ident_pgt)@512
$2 = {{pud = 0x1006063}, {pud = 0x0} <repeats 511 times>}

while the real content is:

root@hp /home/wfg# hexdump -s 0x1002000 -n 4096 /dev/mem
1002000 6063 0100 0000 0000 8067 0000 0000 0000
1002010 0000 0000 0000 0000 0000 0000 0000 0000
*
1003000

That is, on a x86_64 box with 2GB memory, we can see first-1GB / full-2GB
identity mapping before/after patch:

(gdb) p/x *((pud_t *)&level3_ident_pgt)@512
before  $1 = {{pud = 0x1006063}, {pud = 0x0} <repeats 511 times>}
after   $1 = {{pud = 0x1006063}, {pud = 0x8067}, {pud = 0x0} <repeats 510 times>}

Obviously the content before patch is wrong.

Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

proc: remove obsolete comments

A quick test shows these comments are obsolete, so just remove them.

Signed-off-by: WANG Cong <amwang@redhat.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

proc: cleanup: remove unused assignments

I removed 3 unused assignments.  The first two get reset on the first
statement of their functions.  For "err" in root.c we don't return an
error and we don't use the variable again.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

proc: turn signal_struct->count into "int nr_threads"

No functional changes, just s/atomic_t count/int nr_threads/.

With the recent changes this counter has a single user, get_nr_threads()
And, none of its callers need the really accurate number of threads, not
to mention each caller obviously races with fork/exit.  It is only used to
report this value to the user-space, except first_tid() uses it to avoid
the unnecessary while_each_thread() loop in the unlikely case.

It is a bit sad we need a word in struct signal_struct for this, perhaps
we can change get_nr_threads() to approximate the number of threads using
signal->live and kill ->nr_threads later.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

keyctl_session_to_parent(): use thread_group_empty() to check singlethreadness

No functional changes.

keyctl_session_to_parent() is the only user of signal->count which needs
the correct value.  Change it to use thread_group_empty() instead, this
must be strictly equivalent under tasklist, and imho looks better.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: David Howells <dhowells@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

proc_sched_show_task(): use get_nr_threads()

Trivial, use get_nr_threads() helper to read signal->count which we are
going to change.

Like other callers, proc_sched_show_task() doesn't need the exactly
precise nr_threads.

David said:

: Note that get_nr_threads() isn't completely equivalent (it can return 0
: where proc_sched_show_task() will display a 1).  But I don't think this
: should be a problem.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: David Howells <dhowells@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

proc: get_nr_threads() doesn't need ->siglock any longer

Now that task->signal can't go away get_nr_threads() doesn't need
->siglock to read signal->count.

Also, make it inline, move into sched.h, and convert 2 other proc users of
signal->count to use this (now trivial) helper.

Henceforth get_nr_threads() is the only valid user of signal->count, we
are ready to turn it into "int nr_threads" or, perhaps, kill it.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: David Howells <dhowells@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

check_unshare_flags: kill the bogus CLONE_SIGHAND/sig->count check

check_unshare_flags(CLONE_SIGHAND) adds CLONE_THREAD to *flags_ptr if the
task is multithreaded to ensure unshare_thread() will fail.

Not only this is a bit strange way to return the error, this is absolutely
meaningless.  If signal->count > 1 then sighand->count must be also > 1,
and unshare_sighand() will fail anyway.

In fact, all CLONE_THREAD/SIGHAND/VM checks inside sys_unshare() do not
look right.  Fortunately this code doesn't really work anyway.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

exit: move taskstats_tgid_free() from __exit_signal() to free_signal_struct()

Move taskstats_tgid_free() from __exit_signal() to free_signal_struct().

This way signal->stats never points to nowhere and we can read ->stats
lockless.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kill the obsolete thread_group_cputime_free() helper

Kill the empty thread_group_cputime_free() helper.  It was needed to free
the per-cpu data which we no longer have.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

exit: __exit_signal: use thread_group_leader() consistently

Cleanup:

- Add the boolean, group_dead = thread_group_leader(), for clarity.

- Do not test/set sig == NULL to detect the all-dead case, use this
  boolean.

- Pass this boolen to __unhash_process() and use it instead of another
  thread_group_leader() call which needs ->group_leader.

  This can be considered as microoptimization, but hopefully this also
  allows us do do other cleanups later.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

signals: kill the awful task_rq_unlock_wait() hack

Now that task->signal can't go away we can revert the horrible hack added
by ad474caca3e2a0550b7ce0706527ad5ab389a4d4 ("fix for
account_group_exec_runtime(), make sure ->signal can't be freed under
rq->lock").

And we can do more cleanups sched_stats.h/posix-cpu-timers.c later.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Alan Cox <alan@linux.intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

signals: clear signal->tty when the last thread exits

When the last thread exits signal->tty is freed, but the pointer is not
cleared and points to nowhere.

This is OK.  Nobody should use signal->tty lockless, and it is no longer
possible to take ->siglock.  However this looks wrong even if correct, and
the nice OOPS is better than subtle and hard to find bugs.

Change __exit_signal() to clear signal->tty under ->siglock.

Note: __exit_signal() needs more cleanups.  It should not check "sig !=
NULL" to detect the all-dead case and we have the same issues with
signal->stats.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Alan Cox <alan@linux.intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

signals: make task_struct->signal immutable/refcountable

We have a lot of problems with accessing task_struct->signal, it can
"disappear" at any moment.  Even current can't use its ->signal safely
after exit_notify().  ->siglock helps, but it is not convenient, not
always possible, and sometimes it makes sense to use task->signal even
after this task has already dead.

This patch adds the reference counter, sigcnt, into signal_struct.  This
reference is owned by task_struct and it is dropped in
__put_task_struct().  Perhaps it makes sense to export
get/put_signal_struct() later, but currently I don't see the immediate
reason.

Rename __cleanup_signal() to free_signal_struct() and unexport it.  With
the previous changes it does nothing except kmem_cache_free().

Change __exit_signal() to not clear/free ->signal, it will be freed when
the last reference to any thread in the thread group goes away.

Note:
- when the last thead exits signal->tty can point to nowhere, see
  the next patch.

- with or without this patch signal_struct->count should go away,
  or at least it should be "int nr_threads" for fs/proc. This will
  be addressed later.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Alan Cox <alan@linux.intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fork/exit: move tty_kref_put() outside of __cleanup_signal()

tty_kref_put() has two callsites in copy_process() paths,

1. if copy_process() suceeds it is called before we copy
   signal->tty from parent

2. otherwise it is called from __cleanup_signal() under
   bad_fork_cleanup_signal: label

In both cases tty_kref_put() is not right and unneeded because we don't
have the balancing tty_kref_get().  Fortunately, this is harmless because
this can only happen without CLONE_THREAD, and in this case signal->tty
must be NULL.

Remove tty_kref_put() from copy_process() and __cleanup_signal(), and
change another caller of __cleanup_signal(), __exit_signal(), to call
tty_kref_put() by hand.

I hope this change makes sense by itself, but it is also needed to make
->signal refcountable.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Alan Cox <alan@linux.intel.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ia64: ptrace_attach_sync_user_rbs: avoid "task->signal != NULL" checks

Preparation to make task->signal immutable, no functional changes.

It doesn't matter which pointer we check under tasklist to ensure the task
was not released, ->signal or ->sighand.  But we are going to make
->signal refcountable, change the code to use ->sighand.

Note: this code doesn't need this check and tasklist_lock at all, it
should be converted to use lock_task_sighand().  And, the code under
SIGNAL_STOP_STOPPED check looks wrong.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Stanislaw Gruszka <sgruszka@redhat.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

posix-cpu-timers: avoid "task->signal != NULL" checks

Preparation to make task->signal immutable, no functional changes.

posix-cpu-timers.c checks task->signal != NULL to ensure this task is
alive and didn't pass __exit_signal().  This is correct but we are going
to change the lifetime rules for ->signal and never reset this pointer.

Change the code to check ->sighand instead, it doesn't matter which
pointer we check under tasklist, they both are cleared simultaneously.

As Roland pointed out, some of these changes are not strictly needed and
probably it makes sense to revert them later, when ->signal will be pinned
to task_struct.  But this patch tries to ensure the subsequent changes in
fork/exit can't make any visible impact on posix cpu timers.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: Stanislaw Gruszka <sgruszka@redhat.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

exit: avoid sig->count in __exit_signal() to detect the group-dead case

Change __exit_signal() to check thread_group_leader() instead of
atomic_dec_and_test(&sig->count).  This must be equivalent, the group
leader must be released only after all other threads have exited and
passed __exit_signal().

Henceforth sig->count is not actually used, except in fs/proc for
get_nr_threads/etc.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

exit: avoid sig->count in de_thread/__exit_signal synchronization

de_thread() and __exit_signal() use signal_struct->count/notify_count for
synchronization.  We can simplify the code and use ->notify_count only.
Instead of comparing these two counters, we can change de_thread() to set
->notify_count = nr_of_sub_threads, then change __exit_signal() to
dec-and-test this counter and notify group_exit_task.

Note that __exit_signal() checks "notify_count > 0" just for symmetry with
exit_notify(), we could just check it is != 0.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

exit: change zap_other_threads() to count sub-threads

Change zap_other_threads() to return the number of other sub-threads found
on ->thread_group list.

Other changes are cosmetic:

- change the code to use while_each_thread() helper

- remove the obsolete comment about SIGKILL/SIGSTOP

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

exit: exit_notify() can trust signal->notify_count < 0

signal_struct->count in its current form must die.

- it has no reasons to be atomic_t

- it looks like a reference counter, but it is not

- otoh, we really need to make task->signal refcountable, just look at
  the extremely ugly task_rq_unlock_wait() called from __exit_signals().

- we should change the lifetime rules for task->signal, it should be
  pinned to task_struct.  We have a lot of code which can be simplified
  after that.

- it is not needed!  while the code is correct, any usage of this
  counter is artificial, except fs/proc uses it correctly to show the
  number of threads.

This series removes the usage of sig->count from exit pathes.

This patch:

Now that Veaceslav changed copy_signal() to use zalloc(), exit_notify()
can just check notify_count < 0 to ensure the execing sub-threads needs
the notification from us.  No need to do other checks, notify_count != 0
must always mean ->group_exit_task != NULL is waiting for us.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: shift down_write(mmap_sem) into coredump_wait()

- move the cprm.mm_flags checks up, before we take mmap_sem

- move down_write(mmap_sem) and ->core_state check from do_coredump()
  to coredump_wait()

This simplifies the code and makes the locking symmetrical.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Neil Horman <nhorman@tuxdriver.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: factor out put_cred() calls

Given that do_coredump() calls put_cred() on exit path, it is a bit ugly
to do put_cred() + "goto fail" twice, just add the new "fail_creds" label.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Neil Horman <nhorman@tuxdriver.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: cleanup "ispipe" code

- kill "int dump_count", argv_split(argcp) accepts argcp == NULL.

- move "int dump_count" under " if (ispipe)" branch, fail_dropcount
  can check ispipe.

- move "char **helper_argv" as well, change the code to do argv_free()
  right after call_usermodehelper_fns().

- If call_usermodehelper_fns() fails goto close_fail label instead
  of closing the file by hand.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Neil Horman <nhorman@tuxdriver.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: factor out the not-ispipe file checks

do_coredump() does a lot of file checks after it opens the file or calls
usermode helper.  But all of these checks are only needed in !ispipe case.

Move this code into the "else" branch and kill the ugly repetitive ispipe
checks.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Neil Horman <nhorman@tuxdriver.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

call_usermodehelper: UMH_WAIT_EXEC ignores kernel_thread() failure

UMH_WAIT_EXEC should report the error if kernel_thread() fails, like
UMH_WAIT_PROC does.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

call_usermodehelper: simplify/fix UMH_NO_WAIT case

__call_usermodehelper(UMH_NO_WAIT) has 2 problems:

- if kernel_thread() fails, call_usermodehelper_freeinfo()
  is not called.

- for unknown reason UMH_NO_WAIT has UMH_WAIT_PROC logic,
  we spawn yet another thread which waits until the user
  mode application exits.

Change the UMH_NO_WAIT code to use ____call_usermodehelper() instead of
wait_for_helper(), and do call_usermodehelper_freeinfo() unconditionally.
We can rely on CLONE_VFORK, do_fork(CLONE_VFORK) until the child exits or
execs.

With or without this patch UMH_NO_WAIT does not report the error if
kernel_thread() fails, this is correct since the caller doesn't wait for
result.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

wait_for_helper: SIGCHLD from user-space can lead to use-after-free

1. wait_for_helper() calls allow_signal(SIGCHLD) to ensure the child
   can't autoreap itself.

   However, this means that a spurious SIGCHILD from user-space can
   set TIF_SIGPENDING and:

    - kernel_thread() or sys_wait4() can fail due to signal_pending()

    - worse, wait4() can fail before ____call_usermodehelper() execs
      or exits. In this case the caller may kfree(subprocess_info)
      while the child still uses this memory.

   Change the code to use SIG_DFL instead of magic "(void __user *)2"
   set by allow_signal(). This means that SIGCHLD won't be delivered,
   yet the child won't autoreap itsefl.

   The problem is minor, only root can send a signal to this kthread.

2. If sys_wait4(&ret) fails it doesn't populate "ret", in this case
   wait_for_helper() reports a random value from uninitialized var.

   With this patch sys_wait4() should never fail, but still it makes
   sense to initialize ret = -ECHILD so that the caller can notice
   the problem.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

call_usermodehelper: no need to unblock signals

____call_usermodehelper() correctly calls flush_signal_handlers() to set
SIG_DFL, but sigemptyset(->blocked) and recalc_sigpending() are not
needed.

This kthread was forked by workqueue thread, all signals must be unblocked
and ignored, no pending signal is possible.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

umh: creds: kill subprocess_info->cred logic

Now that nobody ever changes subprocess_info->cred we can kill this member
and related code.  ____call_usermodehelper() always runs in the context of
freshly forked kernel thread, it has the proper ->cred copied from its
parent kthread, keventd.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

umh: creds: convert call_usermodehelper_keys() to use subprocess_info->init()

call_usermodehelper_keys() uses call_usermodehelper_setkeys() to change
subprocess_info->cred in advance.  Now that we have info->init() we can
change this code to set tgcred->session_keyring in context of execing
kernel thread.

Note: since currently call_usermodehelper_keys() is never called with
UMH_NO_WAIT, call_usermodehelper_keys()->key_get() and umh_keys_cleanup()
are not really needed, we could rely on install_session_keyring_to_cred()
which does key_get() on success.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

exec: replace call_usermodehelper_pipe with use of umh init function and resolve limit

The first patch in this series introduced an init function to the
call_usermodehelper api so that processes could be customized by caller.
This patch takes advantage of that fact, by customizing the helper in
do_coredump to create the pipe and set its core limit to one (for our
recusrsion check).  This lets us clean up the previous uglyness in the
usermodehelper internals and factor call_usermodehelper out entirely.
While I'm at it, we can also modify the helper setup to look for a core
limit value of 1 rather than zero for our recursion check

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kmod: add init function to usermodehelper

About 6 months ago, I made a set of changes to how the core-dump-to-a-pipe
feature in the kernel works.  We had reports of several races, including
some reports of apps bypassing our recursion check so that a process that
was forked as part of a core_pattern setup could infinitely crash and
refork until the system crashed.

We fixed those by improving our recursion checks.  The new check basically
refuses to fork a process if its core limit is zero, which works well.

Unfortunately, I've been getting grief from maintainer of user space
programs that are inserted as the forked process of core_pattern.  They
contend that in order for their programs (such as abrt and apport) to
work, all the running processes in a system must have their core limits
set to a non-zero value, to which I say 'yes'.  I did this by design, and
think thats the right way to do things.

But I've been asked to ease this burden on user space enough times that I
thought I would take a look at it.  The first suggestion was to make the
recursion check fail on a non-zero 'special' number, like one.  That way
the core collector process could set its core size ulimit to 1, and enable
the kernel's recursion detection.  This isn't a bad idea on the surface,
but I don't like it since its opt-in, in that if a program like abrt or
apport has a bug and fails to set such a core limit, we're left with a
recursively crashing system again.

So I've come up with this.  What I've done is modify the
call_usermodehelper api such that an extra parameter is added, a function
pointer which will be called by the user helper task, after it forks, but
before it exec's the required process.  This will give the caller the
opportunity to get a call back in the processes context, allowing it to do
whatever it needs to to the process in the kernel prior to exec-ing the
user space code.  In the case of do_coredump, this callback is ues to set
the core ulimit of the helper process to 1.  This elimnates the opt-in
problem that I had above, as it allows the ulimit for core sizes to be set
to the value of 1, which is what the recursion check looks for in
do_coredump.

This patch:

Create new function call_usermodehelper_fns() and allow it to assign both
an init and cleanup function, as we'll as arbitrary data.

The init function is called from the context of the forked process and
allows for customization of the helper process prior to calling exec.  Its
return code gates the continuation of the process, or causes its exit.
Also add an arbitrary data pointer to the subprocess_info struct allowing
for data to be passed from the caller to the new process, and the
subsequent cleanup process

Also, use this patch to cleanup the cleanup function.  It currently takes
an argp and envp pointer for freeing, which is ugly.  Lets instead just
make the subprocess_info structure public, and pass that to the cleanup
and init routines

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

signals: check_kill_permission(): don't check creds if same_thread_group()

Andrew Tridgell reports that aio_read(SIGEV_SIGNAL) can fail if the
notification from the helper thread races with setresuid(), see
http://samba.org/~tridge/junkcode/aio_uid.c

This happens because check_kill_permission() doesn't permit sending a
signal to the task with the different cred->xids.  But there is not any
security reason to check ->cred's when the task sends a signal (private or
group-wide) to its sub-thread.  Whatever we do, any thread can bypass all
security checks and send SIGKILL to all threads, or it can block a signal
SIG and do kill(gettid(), SIG) to deliver this signal to another
sub-thread.  Not to mention that CLONE_THREAD implies CLONE_VM.

Change check_kill_permission() to avoid the credentials check when the
sender and the target are from the same thread group.

Also, move "cred = current_cred()" down to avoid calling get_current()
twice.

Note: David Howells pointed out we could relax this even more, the
CLONE_SIGHAND (without CLONE_THREAD) case probably does not need
these checks too.

Roland said:
: The glibc (libpthread) that does set*id across threads has
: been in use for a while (2.3.4?), probably in distro's using kernels as old
: or older than any active -stable streams.  In the race in question, this
: kernel bug is breaking valid POSIX application expectations.

Reported-by: Andrew Tridgell <tridge@samba.org>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Acked-by: David Howells <dhowells@redhat.com>
Cc: Eric Paris <eparis@parisplace.org>
Cc: Jakub Jelinek <jakub@redhat.com>
Cc: James Morris <jmorris@namei.org>
Cc: Roland McGrath <roland@redhat.com>
Cc: Stephen Smalley <sds@tycho.nsa.gov>
Cc: <stable@kernel.org> [all kernel versions]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ptrace: PTRACE_GETFDPIC: fix the unsafe usage of child->mm

Now that Mike Frysinger unified the FDPIC ptrace code, we can fix the
unsafe usage of child->mm in ptrace_request(PTRACE_GETFDPIC).

We have the reference to task_struct, and ptrace_check_attach() verified
the tracee is stopped.  But nothing can protect from SIGKILL after that,
we must not assume child->mm != NULL.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Mike Frysinger <vapier.adi@gmail.com>
Acked-by: David Howells <dhowells@redhat.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Greg Ungerer <gerg@snapgear.com>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ptrace: unify FDPIC implementations

The Blackfin/FRV/SuperH guys all have the same exact FDPIC ptrace code in
their arch handlers (since they were probably copied & pasted).  Since
these ptrace interfaces are an arch independent aspect of the FDPIC code,
unify them in the common ptrace code so new FDPIC ports don't need to copy
and paste this fundamental stuff yet again.

Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Acked-by: Roland McGrath <roland@redhat.com>
Acked-by: David Howells <dhowells@redhat.com>
Acked-by: Paul Mundt <lethal@linux-sh.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpusets: randomize node rotor used in cpuset_mem_spread_node()

Some workloads that create a large number of small files tend to assign
too many pages to node 0 (multi-node systems).  Part of the reason is that
the rotor (in cpuset_mem_spread_node()) used to assign nodes starts at
node 0 for newly created tasks.

This patch changes the rotor to be initialized to a random node number of
the cpuset.

[akpm@linux-foundation.org: fix layout]
[Lee.Schermerhorn@hp.com: Define stub numa_random() for !NUMA configuration]
Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Paul Menage <menage@google.com>
Cc: Jack Steiner <steiner@sgi.com>
Cc: Robin Holt <holt@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpusets: new round-robin rotor for SLAB allocations

We have observed several workloads running on multi-node systems where
memory is assigned unevenly across the nodes in the system.  There are
numerous reasons for this but one is the round-robin rotor in
cpuset_mem_spread_node().

For example, a simple test that writes a multi-page file will allocate
pages on nodes 0 2 4 6 ...  Odd nodes are skipped.  (Sometimes it
allocates on odd nodes & skips even nodes).

An example is shown below.  The program "lfile" writes a file consisting
of 10 pages.  The program then mmaps the file & uses get_mempolicy(...,
MPOL_F_NODE) to determine the nodes where the file pages were allocated.
The output is shown below:

# ./lfile
 allocated on nodes: 2 4 6 0 1 2 6 0 2

There is a single rotor that is used for allocating both file pages & slab
pages.  Writing the file allocates both a data page & a slab page
(buffer_head).  This advances the RR rotor 2 nodes for each page
allocated.

A quick confirmation seems to confirm this is the cause of the uneven
allocation:

# echo 0 >/dev/cpuset/memory_spread_slab
# ./lfile
 allocated on nodes: 6 7 8 9 0 1 2 3 4 5

This patch introduces a second rotor that is used for slab allocations.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Acked-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Paul Menage <menage@google.com>
Cc: Jack Steiner <steiner@sgi.com>
Cc: Robin Holt <holt@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: clean up memory thresholds

Introduce struct mem_cgroup_thresholds.  It helps to reduce number of
checks of thresholds type (memory or mem+swap).

[akpm@linux-foundation.org: repair comment]
Signed-off-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Phil Carmody <ext-phil.2.carmody@nokia.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroups: make cftype.unregister_event() void-returning

Since we are unable to handle an error returned by
cftype.unregister_event() properly, let's make the callback
void-returning.

mem_cgroup_unregister_event() has been rewritten to be a "never fail"
function.  On mem_cgroup_usage_register_event() we save old buffer for
thresholds array and reuse it in mem_cgroup_usage_unregister_event() to
avoid allocation.

Signed-off-by: Kirill A. Shutemov <kirill@shutemov.name>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Phil Carmody <ext-phil.2.carmody@nokia.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: fix mis-accounting of file mapped racy with migration

FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.

Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.

Now, at migrating mapped file, events happen in following sequence.

 1. allocate a new page.
 2. get memcg of an old page.
 3. charge ageinst a new page before migration. But at this point,
    no changes to new page's page_cgroup, no commit for the charge.
    (IOW, PCG_USED bit is not set.)
 4. page migration replaces radix-tree, old-page and new-page.
 5. page migration remaps the new page if the old page was mapped.
 6. Here, the new page is unlocked.
 7. memcg commits the charge for newpage, Mark the new page's page_cgroup
    as PCG_USED.

Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
 for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
 not on LRU or page_cgroup->mem_cgroup is NULL.)

We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.

BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
  - the page is really freed before remap.
  - migration fails but it's freed before remap
or .....corner cases.

New migration sequence with memcg is:

 1. allocate a new page.
 2. mark PageCgroupMigration to the old page.
 3. charge against a new page onto the old page's memcg. (here, new page's pc
    is marked as PageCgroupUsed.)
 4. page migration replaces radix-tree, page table, etc...
 5. At remapping, new page's page_cgroup is now makrked as "USED"
    We can catch 0->1 event and FILE_MAPPED will be properly updated.

    And we can catch SWAPOUT event after unlock this and freeing this
    page by unmap() can be caught.

 7. Clear PageCgroupMigration of the old page.

So, FILE_MAPPED will be correctly updated.

Then, for what MIGRATION flag is ?
  Without it, at migration failure, we may have to charge old page again
  because it may be fully unmapped. "charge" means that we have to dive into
  memory reclaim or something complated. So, it's better to avoid
  charge it again. Before this patch, __commit_charge() was working for
  both of the old/new page and fixed up all. But this technique has some
  racy condtion around FILE_MAPPED and SWAPOUT etc...
  Now, the kernel use MIGRATION flag and don't uncharge old page until
  the end of migration.

I hope this change will make memcg's page migration much simpler.  This
page migration has caused several troubles.  Worth to add a flag for
simplification.

Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol - uninitialised return value

Only an out of memory error will cause ret to be set.

Signed-off-by: Phil Carmody <ext-phil.2.carmody@nokia.com>
Acked-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove unnecessary use of atomic

The bottom 4 hunks are atomically changing memory to which there are no
aliases as it's freshly allocated, so there's no need to use atomic
operations.

The other hunks are just atomic_read and atomic_set, and do not involve
any read-modify-write.  The use of atomic_{read,set} doesn't prevent a
read/write or write/write race, so if a race were possible (I'm not saying
one is), then it would still be there even with atomic_set.

See:
http://digitalvampire.org/blog/index.php/2007/05/13/atomic-cargo-cults/

Signed-off-by: Phil Carmody <ext-phil.2.carmody@nokia.com>
Acked-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: make oom killer a no-op when no killable task can be found

It's pointless to try to kill current if select_bad_process() did not find
an eligible task to kill in mem_cgroup_out_of_memory() since it's
guaranteed that current is a member of the memcg that is oom and it is, by
definition, unkillable.

Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: update documentation

Some information are old, and I think current document doesn't work as "a
guide for users".  We need summary of all of our controls, at least.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Randy Dunlap <randy.dunlap@oracle.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: move charge of file pages

This patch adds support for moving charge of file pages, which include
normal file, tmpfs file and swaps of tmpfs file.  It's enabled by setting
bit 1 of <target cgroup>/memory.move_charge_at_immigrate.

Unlike the case of anonymous pages, file pages(and swaps) in the range
mmapped by the task will be moved even if the task hasn't done page fault,
i.e.  they might not be the task's "RSS", but other task's "RSS" that maps
the same file.  And mapcount of the page is ignored(the page can be moved
even if page_mapcount(page) > 1).  So, conditions that the page/swap
should be met to be moved is that it must be in the range mmapped by the
target task and it must be charged to the old cgroup.

[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: fix warning]
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: clean up move charge

This patch cleans up move charge code by:

- define functions to handle pte for each types, and make
  is_target_pte_for_mc() cleaner.

- instead of checking the MOVE_CHARGE_TYPE_ANON bit, define a function
  that checks the bit.

Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: oom kill disable and oom status

This adds a feature to disable oom-killer for memcg, if disabled, of
course, tasks under memcg will stop.

But now, we have oom-notifier for memcg.  And the world around memcg is
not under out-of-memory.  memcg's out-of-memory just shows memcg hits
limit.  Then, administrator or management daemon can recover the situation
by

- kill some process
- enlarge limit, add more swap.
- migrate some tasks
- remove file cache on tmps (difficult ?)

Unlike oom-killer, you can take enough information before killing tasks.
(by gcore, or, ps etc.)

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: oom notifier

Considering containers or other resource management softwares in userland,
event notification of OOM in memcg should be implemented.  Now, memcg has
"threshold" notifier which uses eventfd, we can make use of it for oom
notification.

This patch adds oom notification eventfd callback for memcg.  The usage is
very similar to threshold notifier, but control file is memory.oom_control
and no arguments other than eventfd is required.

% cgroup_event_notifier /cgroup/A/memory.oom_control dummy
(About cgroup_event_notifier, see Documentation/cgroup/)

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: David Rientjes <rientjes@google.com>
Cc: Davide Libenzi <davidel@xmailserver.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: oom wakeup filter

memcg's oom waitqueue is a system-wide wait_queue (for handling
hierarchy.) So, it's better to add custom wake function and do filtering
in wake up path.

This patch adds a filtering feature for waking up oom-waiters.  Hierarchy
is properly handled.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Documentation/cgroups/cgroups.txt: fix reference to "numtasks"

Signed-off-by: Trevor Woerner <twoerner@gmail.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>