2005-04-17 06:20:36 +08:00
|
|
|
/* CPU control.
|
|
|
|
|
* (C) 2001, 2002, 2003, 2004 Rusty Russell
|
|
|
|
|
*
|
|
|
|
|
* This code is licenced under the GPL.
|
|
|
|
|
*/
|
|
|
|
|
#include <linux/proc_fs.h>
|
|
|
|
|
#include <linux/smp.h>
|
|
|
|
|
#include <linux/init.h>
|
|
|
|
|
#include <linux/notifier.h>
|
|
|
|
|
#include <linux/sched.h>
|
|
|
|
|
#include <linux/unistd.h>
|
|
|
|
|
#include <linux/cpu.h>
|
cpu: introduce clear_tasks_mm_cpumask() helper
Many architectures clear tasks' mm_cpumask like this:
read_lock(&tasklist_lock);
for_each_process(p) {
if (p->mm)
cpumask_clear_cpu(cpu, mm_cpumask(p->mm));
}
read_unlock(&tasklist_lock);
Depending on the context, the code above may have several problems,
such as:
1. Working with task->mm w/o getting mm or grabing the task lock is
dangerous as ->mm might disappear (exit_mm() assigns NULL under
task_lock(), so tasklist lock is not enough).
2. Checking for process->mm is not enough because process' main
thread may exit or detach its mm via use_mm(), but other threads
may still have a valid mm.
This patch implements a small helper function that does things
correctly, i.e.:
1. We take the task's lock while whe handle its mm (we can't use
get_task_mm()/mmput() pair as mmput() might sleep);
2. To catch exited main thread case, we use find_lock_task_mm(),
which walks up all threads and returns an appropriate task
(with task lock held).
Also, Per Peter Zijlstra's idea, now we don't grab tasklist_lock in
the new helper, instead we take the rcu read lock. We can do this
because the function is called after the cpu is taken down and marked
offline, so no new tasks will get this cpu set in their mm mask.
Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Cc: Richard Weinberger <richard@nod.at>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Mike Frysinger <vapier@gentoo.org>
Cc: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-06-01 07:26:22 +08:00
|
|
|
#include <linux/oom.h>
|
|
|
|
|
#include <linux/rcupdate.h>
|
2011-05-24 02:51:41 +08:00
|
|
|
#include <linux/export.h>
|
2012-06-01 07:26:26 +08:00
|
|
|
#include <linux/bug.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
#include <linux/kthread.h>
|
|
|
|
|
#include <linux/stop_machine.h>
|
2006-06-26 15:24:32 +08:00
|
|
|
#include <linux/mutex.h>
|
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h
percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 16:04:11 +08:00
|
|
|
#include <linux/gfp.h>
|
2011-11-03 07:59:25 +08:00
|
|
|
#include <linux/suspend.h>
|
2014-03-11 04:34:03 +08:00
|
|
|
#include <linux/lockdep.h>
|
2015-03-30 17:29:19 +08:00
|
|
|
#include <linux/tick.h>
|
2015-07-06 01:12:30 +08:00
|
|
|
#include <linux/irq.h>
|
2014-06-06 20:40:17 +08:00
|
|
|
#include <trace/events/power.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2012-04-20 21:05:44 +08:00
|
|
|
#include "smpboot.h"
|
|
|
|
|
|
2008-12-13 18:49:41 +08:00
|
|
|
#ifdef CONFIG_SMP
|
2008-12-30 06:35:14 +08:00
|
|
|
/* Serializes the updates to cpu_online_mask, cpu_present_mask */
|
2006-07-24 03:12:16 +08:00
|
|
|
static DEFINE_MUTEX(cpu_add_remove_lock);
|
2016-02-27 02:43:23 +08:00
|
|
|
bool cpuhp_tasks_frozen;
|
|
|
|
|
EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2010-05-27 05:43:36 +08:00
|
|
|
/*
|
CPU hotplug: Provide lockless versions of callback registration functions
The following method of CPU hotplug callback registration is not safe
due to the possibility of an ABBA deadlock involving the cpu_add_remove_lock
and the cpu_hotplug.lock.
get_online_cpus();
for_each_online_cpu(cpu)
init_cpu(cpu);
register_cpu_notifier(&foobar_cpu_notifier);
put_online_cpus();
The deadlock is shown below:
CPU 0 CPU 1
----- -----
Acquire cpu_hotplug.lock
[via get_online_cpus()]
CPU online/offline operation
takes cpu_add_remove_lock
[via cpu_maps_update_begin()]
Try to acquire
cpu_add_remove_lock
[via register_cpu_notifier()]
CPU online/offline operation
tries to acquire cpu_hotplug.lock
[via cpu_hotplug_begin()]
*** DEADLOCK! ***
The problem here is that callback registration takes the locks in one order
whereas the CPU hotplug operations take the same locks in the opposite order.
To avoid this issue and to provide a race-free method to register CPU hotplug
callbacks (along with initialization of already online CPUs), introduce new
variants of the callback registration APIs that simply register the callbacks
without holding the cpu_add_remove_lock during the registration. That way,
we can avoid the ABBA scenario. However, we will need to hold the
cpu_add_remove_lock throughout the entire critical section, to protect updates
to the callback/notifier chain.
This can be achieved by writing the callback registration code as follows:
cpu_maps_update_begin(); [ or cpu_notifier_register_begin(); see below ]
for_each_online_cpu(cpu)
init_cpu(cpu);
/* This doesn't take the cpu_add_remove_lock */
__register_cpu_notifier(&foobar_cpu_notifier);
cpu_maps_update_done(); [ or cpu_notifier_register_done(); see below ]
Note that we can't use get_online_cpus() here instead of cpu_maps_update_begin()
because the cpu_hotplug.lock is dropped during the invocation of CPU_POST_DEAD
notifiers, and hence get_online_cpus() cannot provide the necessary
synchronization to protect the callback/notifier chains against concurrent
reads and writes. On the other hand, since the cpu_add_remove_lock protects
the entire hotplug operation (including CPU_POST_DEAD), we can use
cpu_maps_update_begin/done() to guarantee proper synchronization.
Also, since cpu_maps_update_begin/done() is like a super-set of
get/put_online_cpus(), the former naturally protects the critical sections
from concurrent hotplug operations.
Since the names cpu_maps_update_begin/done() don't make much sense in CPU
hotplug callback registration scenarios, we'll introduce new APIs named
cpu_notifier_register_begin/done() and map them to cpu_maps_update_begin/done().
In summary, introduce the lockless variants of un/register_cpu_notifier() and
also export the cpu_notifier_register_begin/done() APIs for use by modules.
This way, we provide a race-free way to register hotplug callbacks as well as
perform initialization for the CPUs that are already online.
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@kernel.org>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-11 04:34:14 +08:00
|
|
|
* The following two APIs (cpu_maps_update_begin/done) must be used when
|
|
|
|
|
* attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
|
|
|
|
|
* The APIs cpu_notifier_register_begin/done() must be used to protect CPU
|
|
|
|
|
* hotplug callback (un)registration performed using __register_cpu_notifier()
|
|
|
|
|
* or __unregister_cpu_notifier().
|
2010-05-27 05:43:36 +08:00
|
|
|
*/
|
|
|
|
|
void cpu_maps_update_begin(void)
|
|
|
|
|
{
|
|
|
|
|
mutex_lock(&cpu_add_remove_lock);
|
|
|
|
|
}
|
CPU hotplug: Provide lockless versions of callback registration functions
The following method of CPU hotplug callback registration is not safe
due to the possibility of an ABBA deadlock involving the cpu_add_remove_lock
and the cpu_hotplug.lock.
get_online_cpus();
for_each_online_cpu(cpu)
init_cpu(cpu);
register_cpu_notifier(&foobar_cpu_notifier);
put_online_cpus();
The deadlock is shown below:
CPU 0 CPU 1
----- -----
Acquire cpu_hotplug.lock
[via get_online_cpus()]
CPU online/offline operation
takes cpu_add_remove_lock
[via cpu_maps_update_begin()]
Try to acquire
cpu_add_remove_lock
[via register_cpu_notifier()]
CPU online/offline operation
tries to acquire cpu_hotplug.lock
[via cpu_hotplug_begin()]
*** DEADLOCK! ***
The problem here is that callback registration takes the locks in one order
whereas the CPU hotplug operations take the same locks in the opposite order.
To avoid this issue and to provide a race-free method to register CPU hotplug
callbacks (along with initialization of already online CPUs), introduce new
variants of the callback registration APIs that simply register the callbacks
without holding the cpu_add_remove_lock during the registration. That way,
we can avoid the ABBA scenario. However, we will need to hold the
cpu_add_remove_lock throughout the entire critical section, to protect updates
to the callback/notifier chain.
This can be achieved by writing the callback registration code as follows:
cpu_maps_update_begin(); [ or cpu_notifier_register_begin(); see below ]
for_each_online_cpu(cpu)
init_cpu(cpu);
/* This doesn't take the cpu_add_remove_lock */
__register_cpu_notifier(&foobar_cpu_notifier);
cpu_maps_update_done(); [ or cpu_notifier_register_done(); see below ]
Note that we can't use get_online_cpus() here instead of cpu_maps_update_begin()
because the cpu_hotplug.lock is dropped during the invocation of CPU_POST_DEAD
notifiers, and hence get_online_cpus() cannot provide the necessary
synchronization to protect the callback/notifier chains against concurrent
reads and writes. On the other hand, since the cpu_add_remove_lock protects
the entire hotplug operation (including CPU_POST_DEAD), we can use
cpu_maps_update_begin/done() to guarantee proper synchronization.
Also, since cpu_maps_update_begin/done() is like a super-set of
get/put_online_cpus(), the former naturally protects the critical sections
from concurrent hotplug operations.
Since the names cpu_maps_update_begin/done() don't make much sense in CPU
hotplug callback registration scenarios, we'll introduce new APIs named
cpu_notifier_register_begin/done() and map them to cpu_maps_update_begin/done().
In summary, introduce the lockless variants of un/register_cpu_notifier() and
also export the cpu_notifier_register_begin/done() APIs for use by modules.
This way, we provide a race-free way to register hotplug callbacks as well as
perform initialization for the CPUs that are already online.
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@kernel.org>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-11 04:34:14 +08:00
|
|
|
EXPORT_SYMBOL(cpu_notifier_register_begin);
|
2010-05-27 05:43:36 +08:00
|
|
|
|
|
|
|
|
void cpu_maps_update_done(void)
|
|
|
|
|
{
|
|
|
|
|
mutex_unlock(&cpu_add_remove_lock);
|
|
|
|
|
}
|
CPU hotplug: Provide lockless versions of callback registration functions
The following method of CPU hotplug callback registration is not safe
due to the possibility of an ABBA deadlock involving the cpu_add_remove_lock
and the cpu_hotplug.lock.
get_online_cpus();
for_each_online_cpu(cpu)
init_cpu(cpu);
register_cpu_notifier(&foobar_cpu_notifier);
put_online_cpus();
The deadlock is shown below:
CPU 0 CPU 1
----- -----
Acquire cpu_hotplug.lock
[via get_online_cpus()]
CPU online/offline operation
takes cpu_add_remove_lock
[via cpu_maps_update_begin()]
Try to acquire
cpu_add_remove_lock
[via register_cpu_notifier()]
CPU online/offline operation
tries to acquire cpu_hotplug.lock
[via cpu_hotplug_begin()]
*** DEADLOCK! ***
The problem here is that callback registration takes the locks in one order
whereas the CPU hotplug operations take the same locks in the opposite order.
To avoid this issue and to provide a race-free method to register CPU hotplug
callbacks (along with initialization of already online CPUs), introduce new
variants of the callback registration APIs that simply register the callbacks
without holding the cpu_add_remove_lock during the registration. That way,
we can avoid the ABBA scenario. However, we will need to hold the
cpu_add_remove_lock throughout the entire critical section, to protect updates
to the callback/notifier chain.
This can be achieved by writing the callback registration code as follows:
cpu_maps_update_begin(); [ or cpu_notifier_register_begin(); see below ]
for_each_online_cpu(cpu)
init_cpu(cpu);
/* This doesn't take the cpu_add_remove_lock */
__register_cpu_notifier(&foobar_cpu_notifier);
cpu_maps_update_done(); [ or cpu_notifier_register_done(); see below ]
Note that we can't use get_online_cpus() here instead of cpu_maps_update_begin()
because the cpu_hotplug.lock is dropped during the invocation of CPU_POST_DEAD
notifiers, and hence get_online_cpus() cannot provide the necessary
synchronization to protect the callback/notifier chains against concurrent
reads and writes. On the other hand, since the cpu_add_remove_lock protects
the entire hotplug operation (including CPU_POST_DEAD), we can use
cpu_maps_update_begin/done() to guarantee proper synchronization.
Also, since cpu_maps_update_begin/done() is like a super-set of
get/put_online_cpus(), the former naturally protects the critical sections
from concurrent hotplug operations.
Since the names cpu_maps_update_begin/done() don't make much sense in CPU
hotplug callback registration scenarios, we'll introduce new APIs named
cpu_notifier_register_begin/done() and map them to cpu_maps_update_begin/done().
In summary, introduce the lockless variants of un/register_cpu_notifier() and
also export the cpu_notifier_register_begin/done() APIs for use by modules.
This way, we provide a race-free way to register hotplug callbacks as well as
perform initialization for the CPUs that are already online.
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@kernel.org>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-11 04:34:14 +08:00
|
|
|
EXPORT_SYMBOL(cpu_notifier_register_done);
|
2010-05-27 05:43:36 +08:00
|
|
|
|
2010-06-01 19:15:11 +08:00
|
|
|
static RAW_NOTIFIER_HEAD(cpu_chain);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2006-09-26 14:32:48 +08:00
|
|
|
/* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
|
|
|
|
|
* Should always be manipulated under cpu_add_remove_lock
|
|
|
|
|
*/
|
|
|
|
|
static int cpu_hotplug_disabled;
|
|
|
|
|
|
2010-05-27 05:43:36 +08:00
|
|
|
#ifdef CONFIG_HOTPLUG_CPU
|
|
|
|
|
|
2008-01-26 04:08:01 +08:00
|
|
|
static struct {
|
|
|
|
|
struct task_struct *active_writer;
|
2014-12-12 17:11:44 +08:00
|
|
|
/* wait queue to wake up the active_writer */
|
|
|
|
|
wait_queue_head_t wq;
|
|
|
|
|
/* verifies that no writer will get active while readers are active */
|
|
|
|
|
struct mutex lock;
|
2008-01-26 04:08:01 +08:00
|
|
|
/*
|
|
|
|
|
* Also blocks the new readers during
|
|
|
|
|
* an ongoing cpu hotplug operation.
|
|
|
|
|
*/
|
2014-12-12 17:11:44 +08:00
|
|
|
atomic_t refcount;
|
2014-03-11 04:34:03 +08:00
|
|
|
|
|
|
|
|
#ifdef CONFIG_DEBUG_LOCK_ALLOC
|
|
|
|
|
struct lockdep_map dep_map;
|
|
|
|
|
#endif
|
2009-06-23 12:18:12 +08:00
|
|
|
} cpu_hotplug = {
|
|
|
|
|
.active_writer = NULL,
|
2014-12-12 17:11:44 +08:00
|
|
|
.wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq),
|
2009-06-23 12:18:12 +08:00
|
|
|
.lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
|
2014-03-11 04:34:03 +08:00
|
|
|
#ifdef CONFIG_DEBUG_LOCK_ALLOC
|
|
|
|
|
.dep_map = {.name = "cpu_hotplug.lock" },
|
|
|
|
|
#endif
|
2009-06-23 12:18:12 +08:00
|
|
|
};
|
2008-01-26 04:08:01 +08:00
|
|
|
|
2014-03-11 04:34:03 +08:00
|
|
|
/* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */
|
|
|
|
|
#define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map)
|
2014-08-26 11:25:06 +08:00
|
|
|
#define cpuhp_lock_acquire_tryread() \
|
|
|
|
|
lock_map_acquire_tryread(&cpu_hotplug.dep_map)
|
2014-03-11 04:34:03 +08:00
|
|
|
#define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map)
|
|
|
|
|
#define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map)
|
|
|
|
|
|
2014-10-23 05:51:49 +08:00
|
|
|
|
2008-01-26 04:08:02 +08:00
|
|
|
void get_online_cpus(void)
|
2005-11-29 05:43:46 +08:00
|
|
|
{
|
2008-01-26 04:08:01 +08:00
|
|
|
might_sleep();
|
|
|
|
|
if (cpu_hotplug.active_writer == current)
|
2006-07-24 03:12:16 +08:00
|
|
|
return;
|
2014-03-11 04:34:03 +08:00
|
|
|
cpuhp_lock_acquire_read();
|
2008-01-26 04:08:01 +08:00
|
|
|
mutex_lock(&cpu_hotplug.lock);
|
2014-12-12 17:11:44 +08:00
|
|
|
atomic_inc(&cpu_hotplug.refcount);
|
2008-01-26 04:08:01 +08:00
|
|
|
mutex_unlock(&cpu_hotplug.lock);
|
2005-11-29 05:43:46 +08:00
|
|
|
}
|
2008-01-26 04:08:02 +08:00
|
|
|
EXPORT_SYMBOL_GPL(get_online_cpus);
|
2005-11-09 13:34:24 +08:00
|
|
|
|
2008-01-26 04:08:02 +08:00
|
|
|
void put_online_cpus(void)
|
2005-11-29 05:43:46 +08:00
|
|
|
{
|
2014-12-12 17:11:44 +08:00
|
|
|
int refcount;
|
|
|
|
|
|
2008-01-26 04:08:01 +08:00
|
|
|
if (cpu_hotplug.active_writer == current)
|
2006-07-24 03:12:16 +08:00
|
|
|
return;
|
2012-10-09 07:28:20 +08:00
|
|
|
|
2014-12-12 17:11:44 +08:00
|
|
|
refcount = atomic_dec_return(&cpu_hotplug.refcount);
|
|
|
|
|
if (WARN_ON(refcount < 0)) /* try to fix things up */
|
|
|
|
|
atomic_inc(&cpu_hotplug.refcount);
|
|
|
|
|
|
|
|
|
|
if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq))
|
|
|
|
|
wake_up(&cpu_hotplug.wq);
|
2012-10-09 07:28:20 +08:00
|
|
|
|
2014-03-11 04:34:03 +08:00
|
|
|
cpuhp_lock_release();
|
2008-01-26 04:08:01 +08:00
|
|
|
|
2005-11-29 05:43:46 +08:00
|
|
|
}
|
2008-01-26 04:08:02 +08:00
|
|
|
EXPORT_SYMBOL_GPL(put_online_cpus);
|
2005-11-29 05:43:46 +08:00
|
|
|
|
2008-01-26 04:08:01 +08:00
|
|
|
/*
|
|
|
|
|
* This ensures that the hotplug operation can begin only when the
|
|
|
|
|
* refcount goes to zero.
|
|
|
|
|
*
|
|
|
|
|
* Note that during a cpu-hotplug operation, the new readers, if any,
|
|
|
|
|
* will be blocked by the cpu_hotplug.lock
|
|
|
|
|
*
|
2008-04-29 16:00:29 +08:00
|
|
|
* Since cpu_hotplug_begin() is always called after invoking
|
|
|
|
|
* cpu_maps_update_begin(), we can be sure that only one writer is active.
|
2008-01-26 04:08:01 +08:00
|
|
|
*
|
|
|
|
|
* Note that theoretically, there is a possibility of a livelock:
|
|
|
|
|
* - Refcount goes to zero, last reader wakes up the sleeping
|
|
|
|
|
* writer.
|
|
|
|
|
* - Last reader unlocks the cpu_hotplug.lock.
|
|
|
|
|
* - A new reader arrives at this moment, bumps up the refcount.
|
|
|
|
|
* - The writer acquires the cpu_hotplug.lock finds the refcount
|
|
|
|
|
* non zero and goes to sleep again.
|
|
|
|
|
*
|
|
|
|
|
* However, this is very difficult to achieve in practice since
|
2008-01-26 04:08:02 +08:00
|
|
|
* get_online_cpus() not an api which is called all that often.
|
2008-01-26 04:08:01 +08:00
|
|
|
*
|
|
|
|
|
*/
|
ACPI / processor: Acquire writer lock to update CPU maps
CPU system maps are protected with reader/writer locks. The reader
lock, get_online_cpus(), assures that the maps are not updated while
holding the lock. The writer lock, cpu_hotplug_begin(), is used to
udpate the cpu maps along with cpu_maps_update_begin().
However, the ACPI processor handler updates the cpu maps without
holding the the writer lock.
acpi_map_lsapic() is called from acpi_processor_hotadd_init() to
update cpu_possible_mask and cpu_present_mask. acpi_unmap_lsapic()
is called from acpi_processor_remove() to update cpu_possible_mask.
Currently, they are either unprotected or protected with the reader
lock, which is not correct.
For example, the get_online_cpus() below is supposed to assure that
cpu_possible_mask is not changed while the code is iterating with
for_each_possible_cpu().
get_online_cpus();
for_each_possible_cpu(cpu) {
:
}
put_online_cpus();
However, this lock has no protection with CPU hotplug since the ACPI
processor handler does not use the writer lock when it updates
cpu_possible_mask. The reader lock does not serialize within the
readers.
This patch protects them with the writer lock with cpu_hotplug_begin()
along with cpu_maps_update_begin(), which must be held before calling
cpu_hotplug_begin(). It also protects arch_register_cpu() /
arch_unregister_cpu(), which creates / deletes a sysfs cpu device
interface. For this purpose it changes cpu_hotplug_begin() and
cpu_hotplug_done() to global and exports them in cpu.h.
Signed-off-by: Toshi Kani <toshi.kani@hp.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2013-08-12 23:45:53 +08:00
|
|
|
void cpu_hotplug_begin(void)
|
2008-01-26 04:08:01 +08:00
|
|
|
{
|
2014-12-12 17:11:44 +08:00
|
|
|
DEFINE_WAIT(wait);
|
2008-04-29 16:00:29 +08:00
|
|
|
|
2014-12-12 17:11:44 +08:00
|
|
|
cpu_hotplug.active_writer = current;
|
2014-03-11 04:34:03 +08:00
|
|
|
cpuhp_lock_acquire();
|
2014-12-12 17:11:44 +08:00
|
|
|
|
2008-04-29 16:00:29 +08:00
|
|
|
for (;;) {
|
|
|
|
|
mutex_lock(&cpu_hotplug.lock);
|
2014-12-12 17:11:44 +08:00
|
|
|
prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE);
|
|
|
|
|
if (likely(!atomic_read(&cpu_hotplug.refcount)))
|
|
|
|
|
break;
|
2008-01-26 04:08:01 +08:00
|
|
|
mutex_unlock(&cpu_hotplug.lock);
|
|
|
|
|
schedule();
|
|
|
|
|
}
|
2014-12-12 17:11:44 +08:00
|
|
|
finish_wait(&cpu_hotplug.wq, &wait);
|
2008-01-26 04:08:01 +08:00
|
|
|
}
|
|
|
|
|
|
ACPI / processor: Acquire writer lock to update CPU maps
CPU system maps are protected with reader/writer locks. The reader
lock, get_online_cpus(), assures that the maps are not updated while
holding the lock. The writer lock, cpu_hotplug_begin(), is used to
udpate the cpu maps along with cpu_maps_update_begin().
However, the ACPI processor handler updates the cpu maps without
holding the the writer lock.
acpi_map_lsapic() is called from acpi_processor_hotadd_init() to
update cpu_possible_mask and cpu_present_mask. acpi_unmap_lsapic()
is called from acpi_processor_remove() to update cpu_possible_mask.
Currently, they are either unprotected or protected with the reader
lock, which is not correct.
For example, the get_online_cpus() below is supposed to assure that
cpu_possible_mask is not changed while the code is iterating with
for_each_possible_cpu().
get_online_cpus();
for_each_possible_cpu(cpu) {
:
}
put_online_cpus();
However, this lock has no protection with CPU hotplug since the ACPI
processor handler does not use the writer lock when it updates
cpu_possible_mask. The reader lock does not serialize within the
readers.
This patch protects them with the writer lock with cpu_hotplug_begin()
along with cpu_maps_update_begin(), which must be held before calling
cpu_hotplug_begin(). It also protects arch_register_cpu() /
arch_unregister_cpu(), which creates / deletes a sysfs cpu device
interface. For this purpose it changes cpu_hotplug_begin() and
cpu_hotplug_done() to global and exports them in cpu.h.
Signed-off-by: Toshi Kani <toshi.kani@hp.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2013-08-12 23:45:53 +08:00
|
|
|
void cpu_hotplug_done(void)
|
2008-01-26 04:08:01 +08:00
|
|
|
{
|
|
|
|
|
cpu_hotplug.active_writer = NULL;
|
|
|
|
|
mutex_unlock(&cpu_hotplug.lock);
|
2014-03-11 04:34:03 +08:00
|
|
|
cpuhp_lock_release();
|
2008-01-26 04:08:01 +08:00
|
|
|
}
|
2010-05-27 05:43:36 +08:00
|
|
|
|
2013-06-13 05:04:36 +08:00
|
|
|
/*
|
|
|
|
|
* Wait for currently running CPU hotplug operations to complete (if any) and
|
|
|
|
|
* disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
|
|
|
|
|
* the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
|
|
|
|
|
* hotplug path before performing hotplug operations. So acquiring that lock
|
|
|
|
|
* guarantees mutual exclusion from any currently running hotplug operations.
|
|
|
|
|
*/
|
|
|
|
|
void cpu_hotplug_disable(void)
|
|
|
|
|
{
|
|
|
|
|
cpu_maps_update_begin();
|
2015-08-05 15:52:46 +08:00
|
|
|
cpu_hotplug_disabled++;
|
2013-06-13 05:04:36 +08:00
|
|
|
cpu_maps_update_done();
|
|
|
|
|
}
|
2015-08-05 15:52:47 +08:00
|
|
|
EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
|
2013-06-13 05:04:36 +08:00
|
|
|
|
|
|
|
|
void cpu_hotplug_enable(void)
|
|
|
|
|
{
|
|
|
|
|
cpu_maps_update_begin();
|
2015-08-05 15:52:46 +08:00
|
|
|
WARN_ON(--cpu_hotplug_disabled < 0);
|
2013-06-13 05:04:36 +08:00
|
|
|
cpu_maps_update_done();
|
|
|
|
|
}
|
2015-08-05 15:52:47 +08:00
|
|
|
EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
|
ACPI / processor: Acquire writer lock to update CPU maps
CPU system maps are protected with reader/writer locks. The reader
lock, get_online_cpus(), assures that the maps are not updated while
holding the lock. The writer lock, cpu_hotplug_begin(), is used to
udpate the cpu maps along with cpu_maps_update_begin().
However, the ACPI processor handler updates the cpu maps without
holding the the writer lock.
acpi_map_lsapic() is called from acpi_processor_hotadd_init() to
update cpu_possible_mask and cpu_present_mask. acpi_unmap_lsapic()
is called from acpi_processor_remove() to update cpu_possible_mask.
Currently, they are either unprotected or protected with the reader
lock, which is not correct.
For example, the get_online_cpus() below is supposed to assure that
cpu_possible_mask is not changed while the code is iterating with
for_each_possible_cpu().
get_online_cpus();
for_each_possible_cpu(cpu) {
:
}
put_online_cpus();
However, this lock has no protection with CPU hotplug since the ACPI
processor handler does not use the writer lock when it updates
cpu_possible_mask. The reader lock does not serialize within the
readers.
This patch protects them with the writer lock with cpu_hotplug_begin()
along with cpu_maps_update_begin(), which must be held before calling
cpu_hotplug_begin(). It also protects arch_register_cpu() /
arch_unregister_cpu(), which creates / deletes a sysfs cpu device
interface. For this purpose it changes cpu_hotplug_begin() and
cpu_hotplug_done() to global and exports them in cpu.h.
Signed-off-by: Toshi Kani <toshi.kani@hp.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2013-08-12 23:45:53 +08:00
|
|
|
#endif /* CONFIG_HOTPLUG_CPU */
|
2010-05-27 05:43:36 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/* Need to know about CPUs going up/down? */
|
2015-07-20 02:06:22 +08:00
|
|
|
int register_cpu_notifier(struct notifier_block *nb)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2006-10-17 15:10:35 +08:00
|
|
|
int ret;
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_maps_update_begin();
|
2006-10-17 15:10:35 +08:00
|
|
|
ret = raw_notifier_chain_register(&cpu_chain, nb);
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_maps_update_done();
|
2006-10-17 15:10:35 +08:00
|
|
|
return ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2006-06-27 17:54:08 +08:00
|
|
|
|
2015-07-20 02:06:22 +08:00
|
|
|
int __register_cpu_notifier(struct notifier_block *nb)
|
CPU hotplug: Provide lockless versions of callback registration functions
The following method of CPU hotplug callback registration is not safe
due to the possibility of an ABBA deadlock involving the cpu_add_remove_lock
and the cpu_hotplug.lock.
get_online_cpus();
for_each_online_cpu(cpu)
init_cpu(cpu);
register_cpu_notifier(&foobar_cpu_notifier);
put_online_cpus();
The deadlock is shown below:
CPU 0 CPU 1
----- -----
Acquire cpu_hotplug.lock
[via get_online_cpus()]
CPU online/offline operation
takes cpu_add_remove_lock
[via cpu_maps_update_begin()]
Try to acquire
cpu_add_remove_lock
[via register_cpu_notifier()]
CPU online/offline operation
tries to acquire cpu_hotplug.lock
[via cpu_hotplug_begin()]
*** DEADLOCK! ***
The problem here is that callback registration takes the locks in one order
whereas the CPU hotplug operations take the same locks in the opposite order.
To avoid this issue and to provide a race-free method to register CPU hotplug
callbacks (along with initialization of already online CPUs), introduce new
variants of the callback registration APIs that simply register the callbacks
without holding the cpu_add_remove_lock during the registration. That way,
we can avoid the ABBA scenario. However, we will need to hold the
cpu_add_remove_lock throughout the entire critical section, to protect updates
to the callback/notifier chain.
This can be achieved by writing the callback registration code as follows:
cpu_maps_update_begin(); [ or cpu_notifier_register_begin(); see below ]
for_each_online_cpu(cpu)
init_cpu(cpu);
/* This doesn't take the cpu_add_remove_lock */
__register_cpu_notifier(&foobar_cpu_notifier);
cpu_maps_update_done(); [ or cpu_notifier_register_done(); see below ]
Note that we can't use get_online_cpus() here instead of cpu_maps_update_begin()
because the cpu_hotplug.lock is dropped during the invocation of CPU_POST_DEAD
notifiers, and hence get_online_cpus() cannot provide the necessary
synchronization to protect the callback/notifier chains against concurrent
reads and writes. On the other hand, since the cpu_add_remove_lock protects
the entire hotplug operation (including CPU_POST_DEAD), we can use
cpu_maps_update_begin/done() to guarantee proper synchronization.
Also, since cpu_maps_update_begin/done() is like a super-set of
get/put_online_cpus(), the former naturally protects the critical sections
from concurrent hotplug operations.
Since the names cpu_maps_update_begin/done() don't make much sense in CPU
hotplug callback registration scenarios, we'll introduce new APIs named
cpu_notifier_register_begin/done() and map them to cpu_maps_update_begin/done().
In summary, introduce the lockless variants of un/register_cpu_notifier() and
also export the cpu_notifier_register_begin/done() APIs for use by modules.
This way, we provide a race-free way to register hotplug callbacks as well as
perform initialization for the CPUs that are already online.
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@kernel.org>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-11 04:34:14 +08:00
|
|
|
{
|
|
|
|
|
return raw_notifier_chain_register(&cpu_chain, nb);
|
|
|
|
|
}
|
|
|
|
|
|
2016-02-27 02:43:23 +08:00
|
|
|
static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call,
|
2010-05-27 05:43:28 +08:00
|
|
|
int *nr_calls)
|
|
|
|
|
{
|
2016-02-27 02:43:23 +08:00
|
|
|
unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0;
|
|
|
|
|
void *hcpu = (void *)(long)cpu;
|
|
|
|
|
|
2010-05-27 05:43:29 +08:00
|
|
|
int ret;
|
|
|
|
|
|
2016-02-27 02:43:23 +08:00
|
|
|
ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call,
|
2010-05-27 05:43:28 +08:00
|
|
|
nr_calls);
|
2010-05-27 05:43:29 +08:00
|
|
|
|
|
|
|
|
return notifier_to_errno(ret);
|
2010-05-27 05:43:28 +08:00
|
|
|
}
|
|
|
|
|
|
2016-02-27 02:43:23 +08:00
|
|
|
static int cpu_notify(unsigned long val, unsigned int cpu)
|
2010-05-27 05:43:28 +08:00
|
|
|
{
|
2016-02-27 02:43:23 +08:00
|
|
|
return __cpu_notify(val, cpu, -1, NULL);
|
2010-05-27 05:43:28 +08:00
|
|
|
}
|
|
|
|
|
|
2016-02-27 02:43:24 +08:00
|
|
|
/* Notifier wrappers for transitioning to state machine */
|
|
|
|
|
static int notify_prepare(unsigned int cpu)
|
|
|
|
|
{
|
|
|
|
|
int nr_calls = 0;
|
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
|
|
ret = __cpu_notify(CPU_UP_PREPARE, cpu, -1, &nr_calls);
|
|
|
|
|
if (ret) {
|
|
|
|
|
nr_calls--;
|
|
|
|
|
printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n",
|
|
|
|
|
__func__, cpu);
|
|
|
|
|
__cpu_notify(CPU_UP_CANCELED, cpu, nr_calls, NULL);
|
|
|
|
|
}
|
|
|
|
|
return ret;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int notify_online(unsigned int cpu)
|
|
|
|
|
{
|
|
|
|
|
cpu_notify(CPU_ONLINE, cpu);
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int bringup_cpu(unsigned int cpu)
|
|
|
|
|
{
|
|
|
|
|
struct task_struct *idle = idle_thread_get(cpu);
|
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
|
|
/* Arch-specific enabling code. */
|
|
|
|
|
ret = __cpu_up(cpu, idle);
|
|
|
|
|
if (ret) {
|
|
|
|
|
cpu_notify(CPU_UP_CANCELED, cpu);
|
|
|
|
|
return ret;
|
|
|
|
|
}
|
|
|
|
|
BUG_ON(!cpu_online(cpu));
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
2010-05-28 01:32:08 +08:00
|
|
|
#ifdef CONFIG_HOTPLUG_CPU
|
|
|
|
|
|
2016-02-27 02:43:23 +08:00
|
|
|
static void cpu_notify_nofail(unsigned long val, unsigned int cpu)
|
2010-05-27 05:43:28 +08:00
|
|
|
{
|
2016-02-27 02:43:23 +08:00
|
|
|
BUG_ON(cpu_notify(val, cpu));
|
2010-05-27 05:43:28 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
EXPORT_SYMBOL(register_cpu_notifier);
|
CPU hotplug: Provide lockless versions of callback registration functions
The following method of CPU hotplug callback registration is not safe
due to the possibility of an ABBA deadlock involving the cpu_add_remove_lock
and the cpu_hotplug.lock.
get_online_cpus();
for_each_online_cpu(cpu)
init_cpu(cpu);
register_cpu_notifier(&foobar_cpu_notifier);
put_online_cpus();
The deadlock is shown below:
CPU 0 CPU 1
----- -----
Acquire cpu_hotplug.lock
[via get_online_cpus()]
CPU online/offline operation
takes cpu_add_remove_lock
[via cpu_maps_update_begin()]
Try to acquire
cpu_add_remove_lock
[via register_cpu_notifier()]
CPU online/offline operation
tries to acquire cpu_hotplug.lock
[via cpu_hotplug_begin()]
*** DEADLOCK! ***
The problem here is that callback registration takes the locks in one order
whereas the CPU hotplug operations take the same locks in the opposite order.
To avoid this issue and to provide a race-free method to register CPU hotplug
callbacks (along with initialization of already online CPUs), introduce new
variants of the callback registration APIs that simply register the callbacks
without holding the cpu_add_remove_lock during the registration. That way,
we can avoid the ABBA scenario. However, we will need to hold the
cpu_add_remove_lock throughout the entire critical section, to protect updates
to the callback/notifier chain.
This can be achieved by writing the callback registration code as follows:
cpu_maps_update_begin(); [ or cpu_notifier_register_begin(); see below ]
for_each_online_cpu(cpu)
init_cpu(cpu);
/* This doesn't take the cpu_add_remove_lock */
__register_cpu_notifier(&foobar_cpu_notifier);
cpu_maps_update_done(); [ or cpu_notifier_register_done(); see below ]
Note that we can't use get_online_cpus() here instead of cpu_maps_update_begin()
because the cpu_hotplug.lock is dropped during the invocation of CPU_POST_DEAD
notifiers, and hence get_online_cpus() cannot provide the necessary
synchronization to protect the callback/notifier chains against concurrent
reads and writes. On the other hand, since the cpu_add_remove_lock protects
the entire hotplug operation (including CPU_POST_DEAD), we can use
cpu_maps_update_begin/done() to guarantee proper synchronization.
Also, since cpu_maps_update_begin/done() is like a super-set of
get/put_online_cpus(), the former naturally protects the critical sections
from concurrent hotplug operations.
Since the names cpu_maps_update_begin/done() don't make much sense in CPU
hotplug callback registration scenarios, we'll introduce new APIs named
cpu_notifier_register_begin/done() and map them to cpu_maps_update_begin/done().
In summary, introduce the lockless variants of un/register_cpu_notifier() and
also export the cpu_notifier_register_begin/done() APIs for use by modules.
This way, we provide a race-free way to register hotplug callbacks as well as
perform initialization for the CPUs that are already online.
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@kernel.org>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-11 04:34:14 +08:00
|
|
|
EXPORT_SYMBOL(__register_cpu_notifier);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-07-20 02:06:22 +08:00
|
|
|
void unregister_cpu_notifier(struct notifier_block *nb)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_maps_update_begin();
|
2006-10-17 15:10:35 +08:00
|
|
|
raw_notifier_chain_unregister(&cpu_chain, nb);
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_maps_update_done();
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
EXPORT_SYMBOL(unregister_cpu_notifier);
|
|
|
|
|
|
2015-07-20 02:06:22 +08:00
|
|
|
void __unregister_cpu_notifier(struct notifier_block *nb)
|
CPU hotplug: Provide lockless versions of callback registration functions
The following method of CPU hotplug callback registration is not safe
due to the possibility of an ABBA deadlock involving the cpu_add_remove_lock
and the cpu_hotplug.lock.
get_online_cpus();
for_each_online_cpu(cpu)
init_cpu(cpu);
register_cpu_notifier(&foobar_cpu_notifier);
put_online_cpus();
The deadlock is shown below:
CPU 0 CPU 1
----- -----
Acquire cpu_hotplug.lock
[via get_online_cpus()]
CPU online/offline operation
takes cpu_add_remove_lock
[via cpu_maps_update_begin()]
Try to acquire
cpu_add_remove_lock
[via register_cpu_notifier()]
CPU online/offline operation
tries to acquire cpu_hotplug.lock
[via cpu_hotplug_begin()]
*** DEADLOCK! ***
The problem here is that callback registration takes the locks in one order
whereas the CPU hotplug operations take the same locks in the opposite order.
To avoid this issue and to provide a race-free method to register CPU hotplug
callbacks (along with initialization of already online CPUs), introduce new
variants of the callback registration APIs that simply register the callbacks
without holding the cpu_add_remove_lock during the registration. That way,
we can avoid the ABBA scenario. However, we will need to hold the
cpu_add_remove_lock throughout the entire critical section, to protect updates
to the callback/notifier chain.
This can be achieved by writing the callback registration code as follows:
cpu_maps_update_begin(); [ or cpu_notifier_register_begin(); see below ]
for_each_online_cpu(cpu)
init_cpu(cpu);
/* This doesn't take the cpu_add_remove_lock */
__register_cpu_notifier(&foobar_cpu_notifier);
cpu_maps_update_done(); [ or cpu_notifier_register_done(); see below ]
Note that we can't use get_online_cpus() here instead of cpu_maps_update_begin()
because the cpu_hotplug.lock is dropped during the invocation of CPU_POST_DEAD
notifiers, and hence get_online_cpus() cannot provide the necessary
synchronization to protect the callback/notifier chains against concurrent
reads and writes. On the other hand, since the cpu_add_remove_lock protects
the entire hotplug operation (including CPU_POST_DEAD), we can use
cpu_maps_update_begin/done() to guarantee proper synchronization.
Also, since cpu_maps_update_begin/done() is like a super-set of
get/put_online_cpus(), the former naturally protects the critical sections
from concurrent hotplug operations.
Since the names cpu_maps_update_begin/done() don't make much sense in CPU
hotplug callback registration scenarios, we'll introduce new APIs named
cpu_notifier_register_begin/done() and map them to cpu_maps_update_begin/done().
In summary, introduce the lockless variants of un/register_cpu_notifier() and
also export the cpu_notifier_register_begin/done() APIs for use by modules.
This way, we provide a race-free way to register hotplug callbacks as well as
perform initialization for the CPUs that are already online.
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@kernel.org>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-11 04:34:14 +08:00
|
|
|
{
|
|
|
|
|
raw_notifier_chain_unregister(&cpu_chain, nb);
|
|
|
|
|
}
|
|
|
|
|
EXPORT_SYMBOL(__unregister_cpu_notifier);
|
|
|
|
|
|
2012-06-01 07:26:26 +08:00
|
|
|
/**
|
|
|
|
|
* clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
|
|
|
|
|
* @cpu: a CPU id
|
|
|
|
|
*
|
|
|
|
|
* This function walks all processes, finds a valid mm struct for each one and
|
|
|
|
|
* then clears a corresponding bit in mm's cpumask. While this all sounds
|
|
|
|
|
* trivial, there are various non-obvious corner cases, which this function
|
|
|
|
|
* tries to solve in a safe manner.
|
|
|
|
|
*
|
|
|
|
|
* Also note that the function uses a somewhat relaxed locking scheme, so it may
|
|
|
|
|
* be called only for an already offlined CPU.
|
|
|
|
|
*/
|
cpu: introduce clear_tasks_mm_cpumask() helper
Many architectures clear tasks' mm_cpumask like this:
read_lock(&tasklist_lock);
for_each_process(p) {
if (p->mm)
cpumask_clear_cpu(cpu, mm_cpumask(p->mm));
}
read_unlock(&tasklist_lock);
Depending on the context, the code above may have several problems,
such as:
1. Working with task->mm w/o getting mm or grabing the task lock is
dangerous as ->mm might disappear (exit_mm() assigns NULL under
task_lock(), so tasklist lock is not enough).
2. Checking for process->mm is not enough because process' main
thread may exit or detach its mm via use_mm(), but other threads
may still have a valid mm.
This patch implements a small helper function that does things
correctly, i.e.:
1. We take the task's lock while whe handle its mm (we can't use
get_task_mm()/mmput() pair as mmput() might sleep);
2. To catch exited main thread case, we use find_lock_task_mm(),
which walks up all threads and returns an appropriate task
(with task lock held).
Also, Per Peter Zijlstra's idea, now we don't grab tasklist_lock in
the new helper, instead we take the rcu read lock. We can do this
because the function is called after the cpu is taken down and marked
offline, so no new tasks will get this cpu set in their mm mask.
Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Cc: Richard Weinberger <richard@nod.at>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Mike Frysinger <vapier@gentoo.org>
Cc: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-06-01 07:26:22 +08:00
|
|
|
void clear_tasks_mm_cpumask(int cpu)
|
|
|
|
|
{
|
|
|
|
|
struct task_struct *p;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* This function is called after the cpu is taken down and marked
|
|
|
|
|
* offline, so its not like new tasks will ever get this cpu set in
|
|
|
|
|
* their mm mask. -- Peter Zijlstra
|
|
|
|
|
* Thus, we may use rcu_read_lock() here, instead of grabbing
|
|
|
|
|
* full-fledged tasklist_lock.
|
|
|
|
|
*/
|
2012-06-01 07:26:26 +08:00
|
|
|
WARN_ON(cpu_online(cpu));
|
cpu: introduce clear_tasks_mm_cpumask() helper
Many architectures clear tasks' mm_cpumask like this:
read_lock(&tasklist_lock);
for_each_process(p) {
if (p->mm)
cpumask_clear_cpu(cpu, mm_cpumask(p->mm));
}
read_unlock(&tasklist_lock);
Depending on the context, the code above may have several problems,
such as:
1. Working with task->mm w/o getting mm or grabing the task lock is
dangerous as ->mm might disappear (exit_mm() assigns NULL under
task_lock(), so tasklist lock is not enough).
2. Checking for process->mm is not enough because process' main
thread may exit or detach its mm via use_mm(), but other threads
may still have a valid mm.
This patch implements a small helper function that does things
correctly, i.e.:
1. We take the task's lock while whe handle its mm (we can't use
get_task_mm()/mmput() pair as mmput() might sleep);
2. To catch exited main thread case, we use find_lock_task_mm(),
which walks up all threads and returns an appropriate task
(with task lock held).
Also, Per Peter Zijlstra's idea, now we don't grab tasklist_lock in
the new helper, instead we take the rcu read lock. We can do this
because the function is called after the cpu is taken down and marked
offline, so no new tasks will get this cpu set in their mm mask.
Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Cc: Richard Weinberger <richard@nod.at>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Mike Frysinger <vapier@gentoo.org>
Cc: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-06-01 07:26:22 +08:00
|
|
|
rcu_read_lock();
|
|
|
|
|
for_each_process(p) {
|
|
|
|
|
struct task_struct *t;
|
|
|
|
|
|
2012-06-01 07:26:26 +08:00
|
|
|
/*
|
|
|
|
|
* Main thread might exit, but other threads may still have
|
|
|
|
|
* a valid mm. Find one.
|
|
|
|
|
*/
|
cpu: introduce clear_tasks_mm_cpumask() helper
Many architectures clear tasks' mm_cpumask like this:
read_lock(&tasklist_lock);
for_each_process(p) {
if (p->mm)
cpumask_clear_cpu(cpu, mm_cpumask(p->mm));
}
read_unlock(&tasklist_lock);
Depending on the context, the code above may have several problems,
such as:
1. Working with task->mm w/o getting mm or grabing the task lock is
dangerous as ->mm might disappear (exit_mm() assigns NULL under
task_lock(), so tasklist lock is not enough).
2. Checking for process->mm is not enough because process' main
thread may exit or detach its mm via use_mm(), but other threads
may still have a valid mm.
This patch implements a small helper function that does things
correctly, i.e.:
1. We take the task's lock while whe handle its mm (we can't use
get_task_mm()/mmput() pair as mmput() might sleep);
2. To catch exited main thread case, we use find_lock_task_mm(),
which walks up all threads and returns an appropriate task
(with task lock held).
Also, Per Peter Zijlstra's idea, now we don't grab tasklist_lock in
the new helper, instead we take the rcu read lock. We can do this
because the function is called after the cpu is taken down and marked
offline, so no new tasks will get this cpu set in their mm mask.
Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Cc: Richard Weinberger <richard@nod.at>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Mike Frysinger <vapier@gentoo.org>
Cc: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-06-01 07:26:22 +08:00
|
|
|
t = find_lock_task_mm(p);
|
|
|
|
|
if (!t)
|
|
|
|
|
continue;
|
|
|
|
|
cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
|
|
|
|
|
task_unlock(t);
|
|
|
|
|
}
|
|
|
|
|
rcu_read_unlock();
|
|
|
|
|
}
|
|
|
|
|
|
2014-06-25 16:19:55 +08:00
|
|
|
static inline void check_for_tasks(int dead_cpu)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2014-06-25 16:19:55 +08:00
|
|
|
struct task_struct *g, *p;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-09-10 21:07:50 +08:00
|
|
|
read_lock(&tasklist_lock);
|
|
|
|
|
for_each_process_thread(g, p) {
|
2014-06-25 16:19:55 +08:00
|
|
|
if (!p->on_rq)
|
|
|
|
|
continue;
|
|
|
|
|
/*
|
|
|
|
|
* We do the check with unlocked task_rq(p)->lock.
|
|
|
|
|
* Order the reading to do not warn about a task,
|
|
|
|
|
* which was running on this cpu in the past, and
|
|
|
|
|
* it's just been woken on another cpu.
|
|
|
|
|
*/
|
|
|
|
|
rmb();
|
|
|
|
|
if (task_cpu(p) != dead_cpu)
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n",
|
|
|
|
|
p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags);
|
2015-09-10 21:07:50 +08:00
|
|
|
}
|
|
|
|
|
read_unlock(&tasklist_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Take this CPU down. */
|
2015-07-20 02:06:22 +08:00
|
|
|
static int take_cpu_down(void *_param)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2016-02-27 02:43:23 +08:00
|
|
|
int err, cpu = smp_processor_id();
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
|
/* Ensure this CPU doesn't handle any more interrupts. */
|
|
|
|
|
err = __cpu_disable();
|
|
|
|
|
if (err < 0)
|
2005-06-26 05:54:50 +08:00
|
|
|
return err;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2016-02-27 02:43:23 +08:00
|
|
|
cpu_notify(CPU_DYING, cpu);
|
2015-04-03 08:37:24 +08:00
|
|
|
/* Give up timekeeping duties */
|
|
|
|
|
tick_handover_do_timer();
|
2013-01-31 20:11:14 +08:00
|
|
|
/* Park the stopper thread */
|
2016-02-27 02:43:23 +08:00
|
|
|
stop_machine_park(cpu);
|
2005-06-26 05:54:50 +08:00
|
|
|
return 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
2006-09-26 14:32:48 +08:00
|
|
|
/* Requires cpu_add_remove_lock to be held */
|
2015-07-20 02:06:22 +08:00
|
|
|
static int _cpu_down(unsigned int cpu, int tasks_frozen)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2007-05-09 17:34:04 +08:00
|
|
|
int err, nr_calls = 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2006-09-26 14:32:48 +08:00
|
|
|
if (num_online_cpus() == 1)
|
|
|
|
|
return -EBUSY;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2006-09-26 14:32:48 +08:00
|
|
|
if (!cpu_online(cpu))
|
|
|
|
|
return -EINVAL;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_hotplug_begin();
|
2011-03-23 07:34:07 +08:00
|
|
|
|
2016-02-27 02:43:23 +08:00
|
|
|
cpuhp_tasks_frozen = tasks_frozen;
|
|
|
|
|
|
|
|
|
|
err = __cpu_notify(CPU_DOWN_PREPARE, cpu, -1, &nr_calls);
|
2010-05-27 05:43:29 +08:00
|
|
|
if (err) {
|
2007-10-18 18:05:12 +08:00
|
|
|
nr_calls--;
|
2016-02-27 02:43:23 +08:00
|
|
|
__cpu_notify(CPU_DOWN_FAILED, cpu, nr_calls, NULL);
|
2014-06-05 07:11:17 +08:00
|
|
|
pr_warn("%s: attempt to take down CPU %u failed\n",
|
|
|
|
|
__func__, cpu);
|
2007-05-09 17:34:03 +08:00
|
|
|
goto out_release;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
2013-10-11 20:38:20 +08:00
|
|
|
/*
|
|
|
|
|
* By now we've cleared cpu_active_mask, wait for all preempt-disabled
|
|
|
|
|
* and RCU users of this state to go away such that all new such users
|
|
|
|
|
* will observe it.
|
|
|
|
|
*
|
|
|
|
|
* For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might
|
2015-06-11 04:34:41 +08:00
|
|
|
* not imply sync_sched(), so wait for both.
|
2013-11-13 11:10:56 +08:00
|
|
|
*
|
|
|
|
|
* Do sync before park smpboot threads to take care the rcu boost case.
|
2013-10-11 20:38:20 +08:00
|
|
|
*/
|
2015-06-11 04:34:41 +08:00
|
|
|
if (IS_ENABLED(CONFIG_PREEMPT))
|
|
|
|
|
synchronize_rcu_mult(call_rcu, call_rcu_sched);
|
|
|
|
|
else
|
|
|
|
|
synchronize_rcu();
|
2013-10-11 20:38:20 +08:00
|
|
|
|
2013-11-13 11:10:56 +08:00
|
|
|
smpboot_park_threads(cpu);
|
|
|
|
|
|
2013-10-11 20:38:20 +08:00
|
|
|
/*
|
2015-07-06 01:12:30 +08:00
|
|
|
* Prevent irq alloc/free while the dying cpu reorganizes the
|
|
|
|
|
* interrupt affinities.
|
2013-10-11 20:38:20 +08:00
|
|
|
*/
|
2015-07-06 01:12:30 +08:00
|
|
|
irq_lock_sparse();
|
2013-10-11 20:38:20 +08:00
|
|
|
|
2015-07-06 01:12:30 +08:00
|
|
|
/*
|
|
|
|
|
* So now all preempt/rcu users must observe !cpu_active().
|
|
|
|
|
*/
|
2016-02-27 02:43:23 +08:00
|
|
|
err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu));
|
2008-07-29 01:16:29 +08:00
|
|
|
if (err) {
|
2005-04-17 06:20:36 +08:00
|
|
|
/* CPU didn't die: tell everyone. Can't complain. */
|
2016-02-27 02:43:23 +08:00
|
|
|
cpu_notify_nofail(CPU_DOWN_FAILED, cpu);
|
2015-07-06 01:12:30 +08:00
|
|
|
irq_unlock_sparse();
|
2010-03-15 17:10:23 +08:00
|
|
|
goto out_release;
|
2006-10-29 01:38:57 +08:00
|
|
|
}
|
2008-07-29 01:16:29 +08:00
|
|
|
BUG_ON(cpu_online(cpu));
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2010-11-14 02:32:29 +08:00
|
|
|
/*
|
|
|
|
|
* The migration_call() CPU_DYING callback will have removed all
|
|
|
|
|
* runnable tasks from the cpu, there's only the idle task left now
|
|
|
|
|
* that the migration thread is done doing the stop_machine thing.
|
2010-11-20 03:37:53 +08:00
|
|
|
*
|
|
|
|
|
* Wait for the stop thread to go away.
|
2010-11-14 02:32:29 +08:00
|
|
|
*/
|
2015-01-29 06:09:43 +08:00
|
|
|
while (!per_cpu(cpu_dead_idle, cpu))
|
2010-11-20 03:37:53 +08:00
|
|
|
cpu_relax();
|
2015-01-29 06:09:43 +08:00
|
|
|
smp_mb(); /* Read from cpu_dead_idle before __cpu_die(). */
|
|
|
|
|
per_cpu(cpu_dead_idle, cpu) = false;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-07-06 01:12:30 +08:00
|
|
|
/* Interrupts are moved away from the dying cpu, reenable alloc/free */
|
|
|
|
|
irq_unlock_sparse();
|
|
|
|
|
|
2015-03-30 17:29:19 +08:00
|
|
|
hotplug_cpu__broadcast_tick_pull(cpu);
|
2005-04-17 06:20:36 +08:00
|
|
|
/* This actually kills the CPU. */
|
|
|
|
|
__cpu_die(cpu);
|
|
|
|
|
|
|
|
|
|
/* CPU is completely dead: tell everyone. Too late to complain. */
|
2015-04-03 08:38:05 +08:00
|
|
|
tick_cleanup_dead_cpu(cpu);
|
2016-02-27 02:43:23 +08:00
|
|
|
cpu_notify_nofail(CPU_DEAD, cpu);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
|
check_for_tasks(cpu);
|
|
|
|
|
|
2007-05-09 17:34:03 +08:00
|
|
|
out_release:
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_hotplug_done();
|
2010-05-27 05:43:28 +08:00
|
|
|
if (!err)
|
2016-02-27 02:43:23 +08:00
|
|
|
cpu_notify_nofail(CPU_POST_DEAD, cpu);
|
2006-09-26 14:32:48 +08:00
|
|
|
return err;
|
|
|
|
|
}
|
|
|
|
|
|
2015-07-20 02:06:22 +08:00
|
|
|
int cpu_down(unsigned int cpu)
|
2006-09-26 14:32:48 +08:00
|
|
|
{
|
2008-12-22 19:36:30 +08:00
|
|
|
int err;
|
2006-09-26 14:32:48 +08:00
|
|
|
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_maps_update_begin();
|
2008-07-15 19:43:49 +08:00
|
|
|
|
|
|
|
|
if (cpu_hotplug_disabled) {
|
2006-09-26 14:32:48 +08:00
|
|
|
err = -EBUSY;
|
2008-07-15 19:43:49 +08:00
|
|
|
goto out;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
err = _cpu_down(cpu, 0);
|
2006-09-26 14:32:48 +08:00
|
|
|
|
2008-07-15 19:43:49 +08:00
|
|
|
out:
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_maps_update_done();
|
2005-04-17 06:20:36 +08:00
|
|
|
return err;
|
|
|
|
|
}
|
2008-04-29 14:35:56 +08:00
|
|
|
EXPORT_SYMBOL(cpu_down);
|
2005-04-17 06:20:36 +08:00
|
|
|
#endif /*CONFIG_HOTPLUG_CPU*/
|
|
|
|
|
|
2015-04-12 23:06:55 +08:00
|
|
|
/*
|
|
|
|
|
* Unpark per-CPU smpboot kthreads at CPU-online time.
|
|
|
|
|
*/
|
|
|
|
|
static int smpboot_thread_call(struct notifier_block *nfb,
|
|
|
|
|
unsigned long action, void *hcpu)
|
|
|
|
|
{
|
|
|
|
|
int cpu = (long)hcpu;
|
|
|
|
|
|
|
|
|
|
switch (action & ~CPU_TASKS_FROZEN) {
|
|
|
|
|
|
2015-04-16 03:45:41 +08:00
|
|
|
case CPU_DOWN_FAILED:
|
2015-04-12 23:06:55 +08:00
|
|
|
case CPU_ONLINE:
|
|
|
|
|
smpboot_unpark_threads(cpu);
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
default:
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return NOTIFY_OK;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static struct notifier_block smpboot_thread_notifier = {
|
|
|
|
|
.notifier_call = smpboot_thread_call,
|
|
|
|
|
.priority = CPU_PRI_SMPBOOT,
|
|
|
|
|
};
|
|
|
|
|
|
2015-04-28 06:47:58 +08:00
|
|
|
void smpboot_thread_init(void)
|
2015-04-12 23:06:55 +08:00
|
|
|
{
|
|
|
|
|
register_cpu_notifier(&smpboot_thread_notifier);
|
|
|
|
|
}
|
|
|
|
|
|
2006-09-26 14:32:48 +08:00
|
|
|
/* Requires cpu_add_remove_lock to be held */
|
2013-06-20 02:53:51 +08:00
|
|
|
static int _cpu_up(unsigned int cpu, int tasks_frozen)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2012-04-21 08:08:50 +08:00
|
|
|
struct task_struct *idle;
|
2016-02-27 02:43:24 +08:00
|
|
|
int ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_hotplug_begin();
|
2012-04-20 21:05:44 +08:00
|
|
|
|
2012-10-23 07:30:54 +08:00
|
|
|
if (cpu_online(cpu) || !cpu_present(cpu)) {
|
|
|
|
|
ret = -EINVAL;
|
|
|
|
|
goto out;
|
|
|
|
|
}
|
|
|
|
|
|
2012-04-21 08:08:50 +08:00
|
|
|
idle = idle_thread_get(cpu);
|
|
|
|
|
if (IS_ERR(idle)) {
|
|
|
|
|
ret = PTR_ERR(idle);
|
2012-04-20 21:05:44 +08:00
|
|
|
goto out;
|
2012-04-21 08:08:50 +08:00
|
|
|
}
|
2012-04-20 21:05:44 +08:00
|
|
|
|
2016-02-27 02:43:24 +08:00
|
|
|
cpuhp_tasks_frozen = tasks_frozen;
|
|
|
|
|
|
2012-07-16 18:42:36 +08:00
|
|
|
ret = smpboot_create_threads(cpu);
|
|
|
|
|
if (ret)
|
|
|
|
|
goto out;
|
|
|
|
|
|
2016-02-27 02:43:24 +08:00
|
|
|
ret = notify_prepare(cpu);
|
|
|
|
|
if (ret)
|
|
|
|
|
goto out;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2016-02-27 02:43:24 +08:00
|
|
|
ret = bringup_cpu(cpu);
|
|
|
|
|
if (ret)
|
|
|
|
|
goto out;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2016-02-27 02:43:24 +08:00
|
|
|
notify_online(cpu);
|
2012-04-20 21:05:44 +08:00
|
|
|
out:
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_hotplug_done();
|
2006-09-26 14:32:48 +08:00
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
|
}
|
|
|
|
|
|
2013-06-20 02:53:51 +08:00
|
|
|
int cpu_up(unsigned int cpu)
|
2006-09-26 14:32:48 +08:00
|
|
|
{
|
|
|
|
|
int err = 0;
|
2010-05-25 05:32:41 +08:00
|
|
|
|
2009-01-01 07:42:28 +08:00
|
|
|
if (!cpu_possible(cpu)) {
|
2014-06-05 07:11:17 +08:00
|
|
|
pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
|
|
|
|
|
cpu);
|
2010-03-06 05:42:38 +08:00
|
|
|
#if defined(CONFIG_IA64)
|
2014-06-05 07:11:17 +08:00
|
|
|
pr_err("please check additional_cpus= boot parameter\n");
|
2007-10-19 14:40:47 +08:00
|
|
|
#endif
|
|
|
|
|
return -EINVAL;
|
|
|
|
|
}
|
2006-09-26 14:32:48 +08:00
|
|
|
|
2013-11-13 07:07:25 +08:00
|
|
|
err = try_online_node(cpu_to_node(cpu));
|
|
|
|
|
if (err)
|
|
|
|
|
return err;
|
2010-05-25 05:32:41 +08:00
|
|
|
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_maps_update_begin();
|
2008-07-15 19:43:49 +08:00
|
|
|
|
|
|
|
|
if (cpu_hotplug_disabled) {
|
2006-09-26 14:32:48 +08:00
|
|
|
err = -EBUSY;
|
2008-07-15 19:43:49 +08:00
|
|
|
goto out;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
err = _cpu_up(cpu, 0);
|
|
|
|
|
|
|
|
|
|
out:
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_maps_update_done();
|
2006-09-26 14:32:48 +08:00
|
|
|
return err;
|
|
|
|
|
}
|
2011-12-12 13:54:45 +08:00
|
|
|
EXPORT_SYMBOL_GPL(cpu_up);
|
2006-09-26 14:32:48 +08:00
|
|
|
|
2007-08-31 14:56:29 +08:00
|
|
|
#ifdef CONFIG_PM_SLEEP_SMP
|
2009-01-01 07:42:28 +08:00
|
|
|
static cpumask_var_t frozen_cpus;
|
2006-09-26 14:32:48 +08:00
|
|
|
|
|
|
|
|
int disable_nonboot_cpus(void)
|
|
|
|
|
{
|
2010-05-28 04:16:22 +08:00
|
|
|
int cpu, first_cpu, error = 0;
|
2006-09-26 14:32:48 +08:00
|
|
|
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_maps_update_begin();
|
2009-01-01 07:42:28 +08:00
|
|
|
first_cpu = cpumask_first(cpu_online_mask);
|
2009-12-17 01:04:32 +08:00
|
|
|
/*
|
|
|
|
|
* We take down all of the non-boot CPUs in one shot to avoid races
|
2006-09-26 14:32:48 +08:00
|
|
|
* with the userspace trying to use the CPU hotplug at the same time
|
|
|
|
|
*/
|
2009-01-01 07:42:28 +08:00
|
|
|
cpumask_clear(frozen_cpus);
|
2009-11-25 20:31:39 +08:00
|
|
|
|
2014-06-05 07:11:17 +08:00
|
|
|
pr_info("Disabling non-boot CPUs ...\n");
|
2006-09-26 14:32:48 +08:00
|
|
|
for_each_online_cpu(cpu) {
|
|
|
|
|
if (cpu == first_cpu)
|
|
|
|
|
continue;
|
2014-06-06 20:40:17 +08:00
|
|
|
trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
|
2007-05-09 17:35:10 +08:00
|
|
|
error = _cpu_down(cpu, 1);
|
2014-06-06 20:40:17 +08:00
|
|
|
trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
|
2009-11-18 08:22:13 +08:00
|
|
|
if (!error)
|
2009-01-01 07:42:28 +08:00
|
|
|
cpumask_set_cpu(cpu, frozen_cpus);
|
2009-11-18 08:22:13 +08:00
|
|
|
else {
|
2014-06-05 07:11:17 +08:00
|
|
|
pr_err("Error taking CPU%d down: %d\n", cpu, error);
|
2006-09-26 14:32:48 +08:00
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
}
|
2009-07-01 10:31:07 +08:00
|
|
|
|
2015-08-05 15:52:46 +08:00
|
|
|
if (!error)
|
2006-09-26 14:32:48 +08:00
|
|
|
BUG_ON(num_online_cpus() > 1);
|
2015-08-05 15:52:46 +08:00
|
|
|
else
|
2014-06-05 07:11:17 +08:00
|
|
|
pr_err("Non-boot CPUs are not disabled\n");
|
2015-08-05 15:52:46 +08:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Make sure the CPUs won't be enabled by someone else. We need to do
|
|
|
|
|
* this even in case of failure as all disable_nonboot_cpus() users are
|
|
|
|
|
* supposed to do enable_nonboot_cpus() on the failure path.
|
|
|
|
|
*/
|
|
|
|
|
cpu_hotplug_disabled++;
|
|
|
|
|
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_maps_update_done();
|
2006-09-26 14:32:48 +08:00
|
|
|
return error;
|
|
|
|
|
}
|
|
|
|
|
|
2009-08-20 09:05:36 +08:00
|
|
|
void __weak arch_enable_nonboot_cpus_begin(void)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void __weak arch_enable_nonboot_cpus_end(void)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
|
2015-07-20 02:06:22 +08:00
|
|
|
void enable_nonboot_cpus(void)
|
2006-09-26 14:32:48 +08:00
|
|
|
{
|
|
|
|
|
int cpu, error;
|
|
|
|
|
|
|
|
|
|
/* Allow everyone to use the CPU hotplug again */
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_maps_update_begin();
|
2015-08-05 15:52:46 +08:00
|
|
|
WARN_ON(--cpu_hotplug_disabled < 0);
|
2009-01-01 07:42:28 +08:00
|
|
|
if (cpumask_empty(frozen_cpus))
|
2007-04-02 14:49:49 +08:00
|
|
|
goto out;
|
2006-09-26 14:32:48 +08:00
|
|
|
|
2014-06-05 07:11:17 +08:00
|
|
|
pr_info("Enabling non-boot CPUs ...\n");
|
2009-08-20 09:05:36 +08:00
|
|
|
|
|
|
|
|
arch_enable_nonboot_cpus_begin();
|
|
|
|
|
|
2009-01-01 07:42:28 +08:00
|
|
|
for_each_cpu(cpu, frozen_cpus) {
|
2014-06-06 20:40:17 +08:00
|
|
|
trace_suspend_resume(TPS("CPU_ON"), cpu, true);
|
2007-05-09 17:35:10 +08:00
|
|
|
error = _cpu_up(cpu, 1);
|
2014-06-06 20:40:17 +08:00
|
|
|
trace_suspend_resume(TPS("CPU_ON"), cpu, false);
|
2006-09-26 14:32:48 +08:00
|
|
|
if (!error) {
|
2014-06-05 07:11:17 +08:00
|
|
|
pr_info("CPU%d is up\n", cpu);
|
2006-09-26 14:32:48 +08:00
|
|
|
continue;
|
|
|
|
|
}
|
2014-06-05 07:11:17 +08:00
|
|
|
pr_warn("Error taking CPU%d up: %d\n", cpu, error);
|
2006-09-26 14:32:48 +08:00
|
|
|
}
|
2009-08-20 09:05:36 +08:00
|
|
|
|
|
|
|
|
arch_enable_nonboot_cpus_end();
|
|
|
|
|
|
2009-01-01 07:42:28 +08:00
|
|
|
cpumask_clear(frozen_cpus);
|
2007-04-02 14:49:49 +08:00
|
|
|
out:
|
2008-01-26 04:08:01 +08:00
|
|
|
cpu_maps_update_done();
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2009-01-01 07:42:28 +08:00
|
|
|
|
2011-11-16 04:59:31 +08:00
|
|
|
static int __init alloc_frozen_cpus(void)
|
2009-01-01 07:42:28 +08:00
|
|
|
{
|
|
|
|
|
if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
|
|
|
|
|
return -ENOMEM;
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
core_initcall(alloc_frozen_cpus);
|
2011-11-03 07:59:25 +08:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* When callbacks for CPU hotplug notifications are being executed, we must
|
|
|
|
|
* ensure that the state of the system with respect to the tasks being frozen
|
|
|
|
|
* or not, as reported by the notification, remains unchanged *throughout the
|
|
|
|
|
* duration* of the execution of the callbacks.
|
|
|
|
|
* Hence we need to prevent the freezer from racing with regular CPU hotplug.
|
|
|
|
|
*
|
|
|
|
|
* This synchronization is implemented by mutually excluding regular CPU
|
|
|
|
|
* hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
|
|
|
|
|
* Hibernate notifications.
|
|
|
|
|
*/
|
|
|
|
|
static int
|
|
|
|
|
cpu_hotplug_pm_callback(struct notifier_block *nb,
|
|
|
|
|
unsigned long action, void *ptr)
|
|
|
|
|
{
|
|
|
|
|
switch (action) {
|
|
|
|
|
|
|
|
|
|
case PM_SUSPEND_PREPARE:
|
|
|
|
|
case PM_HIBERNATION_PREPARE:
|
2013-06-13 05:04:36 +08:00
|
|
|
cpu_hotplug_disable();
|
2011-11-03 07:59:25 +08:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case PM_POST_SUSPEND:
|
|
|
|
|
case PM_POST_HIBERNATION:
|
2013-06-13 05:04:36 +08:00
|
|
|
cpu_hotplug_enable();
|
2011-11-03 07:59:25 +08:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
default:
|
|
|
|
|
return NOTIFY_DONE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return NOTIFY_OK;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
2011-11-16 04:59:31 +08:00
|
|
|
static int __init cpu_hotplug_pm_sync_init(void)
|
2011-11-03 07:59:25 +08:00
|
|
|
{
|
2012-11-14 03:32:43 +08:00
|
|
|
/*
|
|
|
|
|
* cpu_hotplug_pm_callback has higher priority than x86
|
|
|
|
|
* bsp_pm_callback which depends on cpu_hotplug_pm_callback
|
|
|
|
|
* to disable cpu hotplug to avoid cpu hotplug race.
|
|
|
|
|
*/
|
2011-11-03 07:59:25 +08:00
|
|
|
pm_notifier(cpu_hotplug_pm_callback, 0);
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
core_initcall(cpu_hotplug_pm_sync_init);
|
|
|
|
|
|
2007-08-31 14:56:29 +08:00
|
|
|
#endif /* CONFIG_PM_SLEEP_SMP */
|
2008-05-30 02:17:02 +08:00
|
|
|
|
2008-09-07 22:57:22 +08:00
|
|
|
/**
|
|
|
|
|
* notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
|
|
|
|
|
* @cpu: cpu that just started
|
|
|
|
|
*
|
|
|
|
|
* This function calls the cpu_chain notifiers with CPU_STARTING.
|
|
|
|
|
* It must be called by the arch code on the new cpu, before the new cpu
|
|
|
|
|
* enables interrupts and before the "boot" cpu returns from __cpu_up().
|
|
|
|
|
*/
|
2013-06-20 02:53:51 +08:00
|
|
|
void notify_cpu_starting(unsigned int cpu)
|
2008-09-07 22:57:22 +08:00
|
|
|
{
|
2016-02-27 02:43:23 +08:00
|
|
|
cpu_notify(CPU_STARTING, cpu);
|
2008-09-07 22:57:22 +08:00
|
|
|
}
|
|
|
|
|
|
2008-05-30 02:17:02 +08:00
|
|
|
#endif /* CONFIG_SMP */
|
2008-07-25 09:21:29 +08:00
|
|
|
|
cpu masks: optimize and clean up cpumask_of_cpu()
Clean up and optimize cpumask_of_cpu(), by sharing all the zero words.
Instead of stupidly generating all possible i=0...NR_CPUS 2^i patterns
creating a huge array of constant bitmasks, realize that the zero words
can be shared.
In other words, on a 64-bit architecture, we only ever need 64 of these
arrays - with a different bit set in one single world (with enough zero
words around it so that we can create any bitmask by just offsetting in
that big array). And then we just put enough zeroes around it that we
can point every single cpumask to be one of those things.
So when we have 4k CPU's, instead of having 4k arrays (of 4k bits each,
with one bit set in each array - 2MB memory total), we have exactly 64
arrays instead, each 8k bits in size (64kB total).
And then we just point cpumask(n) to the right position (which we can
calculate dynamically). Once we have the right arrays, getting
"cpumask(n)" ends up being:
static inline const cpumask_t *get_cpu_mask(unsigned int cpu)
{
const unsigned long *p = cpu_bit_bitmap[1 + cpu % BITS_PER_LONG];
p -= cpu / BITS_PER_LONG;
return (const cpumask_t *)p;
}
This brings other advantages and simplifications as well:
- we are not wasting memory that is just filled with a single bit in
various different places
- we don't need all those games to re-create the arrays in some dense
format, because they're already going to be dense enough.
if we compile a kernel for up to 4k CPU's, "wasting" that 64kB of memory
is a non-issue (especially since by doing this "overlapping" trick we
probably get better cache behaviour anyway).
[ mingo@elte.hu:
Converted Linus's mails into a commit. See:
http://lkml.org/lkml/2008/7/27/156
http://lkml.org/lkml/2008/7/28/320
Also applied a family filter - which also has the side-effect of leaving
out the bits where Linus calls me an idio... Oh, never mind ;-)
]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-29 02:32:33 +08:00
|
|
|
/*
|
|
|
|
|
* cpu_bit_bitmap[] is a special, "compressed" data structure that
|
|
|
|
|
* represents all NR_CPUS bits binary values of 1<<nr.
|
|
|
|
|
*
|
2009-01-01 07:42:28 +08:00
|
|
|
* It is used by cpumask_of() to get a constant address to a CPU
|
cpu masks: optimize and clean up cpumask_of_cpu()
Clean up and optimize cpumask_of_cpu(), by sharing all the zero words.
Instead of stupidly generating all possible i=0...NR_CPUS 2^i patterns
creating a huge array of constant bitmasks, realize that the zero words
can be shared.
In other words, on a 64-bit architecture, we only ever need 64 of these
arrays - with a different bit set in one single world (with enough zero
words around it so that we can create any bitmask by just offsetting in
that big array). And then we just put enough zeroes around it that we
can point every single cpumask to be one of those things.
So when we have 4k CPU's, instead of having 4k arrays (of 4k bits each,
with one bit set in each array - 2MB memory total), we have exactly 64
arrays instead, each 8k bits in size (64kB total).
And then we just point cpumask(n) to the right position (which we can
calculate dynamically). Once we have the right arrays, getting
"cpumask(n)" ends up being:
static inline const cpumask_t *get_cpu_mask(unsigned int cpu)
{
const unsigned long *p = cpu_bit_bitmap[1 + cpu % BITS_PER_LONG];
p -= cpu / BITS_PER_LONG;
return (const cpumask_t *)p;
}
This brings other advantages and simplifications as well:
- we are not wasting memory that is just filled with a single bit in
various different places
- we don't need all those games to re-create the arrays in some dense
format, because they're already going to be dense enough.
if we compile a kernel for up to 4k CPU's, "wasting" that 64kB of memory
is a non-issue (especially since by doing this "overlapping" trick we
probably get better cache behaviour anyway).
[ mingo@elte.hu:
Converted Linus's mails into a commit. See:
http://lkml.org/lkml/2008/7/27/156
http://lkml.org/lkml/2008/7/28/320
Also applied a family filter - which also has the side-effect of leaving
out the bits where Linus calls me an idio... Oh, never mind ;-)
]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-29 02:32:33 +08:00
|
|
|
* mask value that has a single bit set only.
|
|
|
|
|
*/
|
2008-07-25 09:21:29 +08:00
|
|
|
|
cpu masks: optimize and clean up cpumask_of_cpu()
Clean up and optimize cpumask_of_cpu(), by sharing all the zero words.
Instead of stupidly generating all possible i=0...NR_CPUS 2^i patterns
creating a huge array of constant bitmasks, realize that the zero words
can be shared.
In other words, on a 64-bit architecture, we only ever need 64 of these
arrays - with a different bit set in one single world (with enough zero
words around it so that we can create any bitmask by just offsetting in
that big array). And then we just put enough zeroes around it that we
can point every single cpumask to be one of those things.
So when we have 4k CPU's, instead of having 4k arrays (of 4k bits each,
with one bit set in each array - 2MB memory total), we have exactly 64
arrays instead, each 8k bits in size (64kB total).
And then we just point cpumask(n) to the right position (which we can
calculate dynamically). Once we have the right arrays, getting
"cpumask(n)" ends up being:
static inline const cpumask_t *get_cpu_mask(unsigned int cpu)
{
const unsigned long *p = cpu_bit_bitmap[1 + cpu % BITS_PER_LONG];
p -= cpu / BITS_PER_LONG;
return (const cpumask_t *)p;
}
This brings other advantages and simplifications as well:
- we are not wasting memory that is just filled with a single bit in
various different places
- we don't need all those games to re-create the arrays in some dense
format, because they're already going to be dense enough.
if we compile a kernel for up to 4k CPU's, "wasting" that 64kB of memory
is a non-issue (especially since by doing this "overlapping" trick we
probably get better cache behaviour anyway).
[ mingo@elte.hu:
Converted Linus's mails into a commit. See:
http://lkml.org/lkml/2008/7/27/156
http://lkml.org/lkml/2008/7/28/320
Also applied a family filter - which also has the side-effect of leaving
out the bits where Linus calls me an idio... Oh, never mind ;-)
]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-29 02:32:33 +08:00
|
|
|
/* cpu_bit_bitmap[0] is empty - so we can back into it */
|
2011-03-23 07:34:07 +08:00
|
|
|
#define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
|
cpu masks: optimize and clean up cpumask_of_cpu()
Clean up and optimize cpumask_of_cpu(), by sharing all the zero words.
Instead of stupidly generating all possible i=0...NR_CPUS 2^i patterns
creating a huge array of constant bitmasks, realize that the zero words
can be shared.
In other words, on a 64-bit architecture, we only ever need 64 of these
arrays - with a different bit set in one single world (with enough zero
words around it so that we can create any bitmask by just offsetting in
that big array). And then we just put enough zeroes around it that we
can point every single cpumask to be one of those things.
So when we have 4k CPU's, instead of having 4k arrays (of 4k bits each,
with one bit set in each array - 2MB memory total), we have exactly 64
arrays instead, each 8k bits in size (64kB total).
And then we just point cpumask(n) to the right position (which we can
calculate dynamically). Once we have the right arrays, getting
"cpumask(n)" ends up being:
static inline const cpumask_t *get_cpu_mask(unsigned int cpu)
{
const unsigned long *p = cpu_bit_bitmap[1 + cpu % BITS_PER_LONG];
p -= cpu / BITS_PER_LONG;
return (const cpumask_t *)p;
}
This brings other advantages and simplifications as well:
- we are not wasting memory that is just filled with a single bit in
various different places
- we don't need all those games to re-create the arrays in some dense
format, because they're already going to be dense enough.
if we compile a kernel for up to 4k CPU's, "wasting" that 64kB of memory
is a non-issue (especially since by doing this "overlapping" trick we
probably get better cache behaviour anyway).
[ mingo@elte.hu:
Converted Linus's mails into a commit. See:
http://lkml.org/lkml/2008/7/27/156
http://lkml.org/lkml/2008/7/28/320
Also applied a family filter - which also has the side-effect of leaving
out the bits where Linus calls me an idio... Oh, never mind ;-)
]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-29 02:32:33 +08:00
|
|
|
#define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
|
|
|
|
|
#define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
|
|
|
|
|
#define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
|
2008-07-25 09:21:29 +08:00
|
|
|
|
cpu masks: optimize and clean up cpumask_of_cpu()
Clean up and optimize cpumask_of_cpu(), by sharing all the zero words.
Instead of stupidly generating all possible i=0...NR_CPUS 2^i patterns
creating a huge array of constant bitmasks, realize that the zero words
can be shared.
In other words, on a 64-bit architecture, we only ever need 64 of these
arrays - with a different bit set in one single world (with enough zero
words around it so that we can create any bitmask by just offsetting in
that big array). And then we just put enough zeroes around it that we
can point every single cpumask to be one of those things.
So when we have 4k CPU's, instead of having 4k arrays (of 4k bits each,
with one bit set in each array - 2MB memory total), we have exactly 64
arrays instead, each 8k bits in size (64kB total).
And then we just point cpumask(n) to the right position (which we can
calculate dynamically). Once we have the right arrays, getting
"cpumask(n)" ends up being:
static inline const cpumask_t *get_cpu_mask(unsigned int cpu)
{
const unsigned long *p = cpu_bit_bitmap[1 + cpu % BITS_PER_LONG];
p -= cpu / BITS_PER_LONG;
return (const cpumask_t *)p;
}
This brings other advantages and simplifications as well:
- we are not wasting memory that is just filled with a single bit in
various different places
- we don't need all those games to re-create the arrays in some dense
format, because they're already going to be dense enough.
if we compile a kernel for up to 4k CPU's, "wasting" that 64kB of memory
is a non-issue (especially since by doing this "overlapping" trick we
probably get better cache behaviour anyway).
[ mingo@elte.hu:
Converted Linus's mails into a commit. See:
http://lkml.org/lkml/2008/7/27/156
http://lkml.org/lkml/2008/7/28/320
Also applied a family filter - which also has the side-effect of leaving
out the bits where Linus calls me an idio... Oh, never mind ;-)
]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-29 02:32:33 +08:00
|
|
|
const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
|
|
|
|
|
|
|
|
|
|
MASK_DECLARE_8(0), MASK_DECLARE_8(8),
|
|
|
|
|
MASK_DECLARE_8(16), MASK_DECLARE_8(24),
|
|
|
|
|
#if BITS_PER_LONG > 32
|
|
|
|
|
MASK_DECLARE_8(32), MASK_DECLARE_8(40),
|
|
|
|
|
MASK_DECLARE_8(48), MASK_DECLARE_8(56),
|
2008-07-25 09:21:29 +08:00
|
|
|
#endif
|
|
|
|
|
};
|
cpu masks: optimize and clean up cpumask_of_cpu()
Clean up and optimize cpumask_of_cpu(), by sharing all the zero words.
Instead of stupidly generating all possible i=0...NR_CPUS 2^i patterns
creating a huge array of constant bitmasks, realize that the zero words
can be shared.
In other words, on a 64-bit architecture, we only ever need 64 of these
arrays - with a different bit set in one single world (with enough zero
words around it so that we can create any bitmask by just offsetting in
that big array). And then we just put enough zeroes around it that we
can point every single cpumask to be one of those things.
So when we have 4k CPU's, instead of having 4k arrays (of 4k bits each,
with one bit set in each array - 2MB memory total), we have exactly 64
arrays instead, each 8k bits in size (64kB total).
And then we just point cpumask(n) to the right position (which we can
calculate dynamically). Once we have the right arrays, getting
"cpumask(n)" ends up being:
static inline const cpumask_t *get_cpu_mask(unsigned int cpu)
{
const unsigned long *p = cpu_bit_bitmap[1 + cpu % BITS_PER_LONG];
p -= cpu / BITS_PER_LONG;
return (const cpumask_t *)p;
}
This brings other advantages and simplifications as well:
- we are not wasting memory that is just filled with a single bit in
various different places
- we don't need all those games to re-create the arrays in some dense
format, because they're already going to be dense enough.
if we compile a kernel for up to 4k CPU's, "wasting" that 64kB of memory
is a non-issue (especially since by doing this "overlapping" trick we
probably get better cache behaviour anyway).
[ mingo@elte.hu:
Converted Linus's mails into a commit. See:
http://lkml.org/lkml/2008/7/27/156
http://lkml.org/lkml/2008/7/28/320
Also applied a family filter - which also has the side-effect of leaving
out the bits where Linus calls me an idio... Oh, never mind ;-)
]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-29 02:32:33 +08:00
|
|
|
EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
|
2008-11-05 10:39:10 +08:00
|
|
|
|
|
|
|
|
const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
|
|
|
|
|
EXPORT_SYMBOL(cpu_all_bits);
|
2008-12-30 06:35:14 +08:00
|
|
|
|
|
|
|
|
#ifdef CONFIG_INIT_ALL_POSSIBLE
|
2016-01-21 07:00:19 +08:00
|
|
|
struct cpumask __cpu_possible_mask __read_mostly
|
2016-01-21 07:00:16 +08:00
|
|
|
= {CPU_BITS_ALL};
|
2008-12-30 06:35:14 +08:00
|
|
|
#else
|
2016-01-21 07:00:19 +08:00
|
|
|
struct cpumask __cpu_possible_mask __read_mostly;
|
2008-12-30 06:35:14 +08:00
|
|
|
#endif
|
2016-01-21 07:00:19 +08:00
|
|
|
EXPORT_SYMBOL(__cpu_possible_mask);
|
2008-12-30 06:35:14 +08:00
|
|
|
|
2016-01-21 07:00:19 +08:00
|
|
|
struct cpumask __cpu_online_mask __read_mostly;
|
|
|
|
|
EXPORT_SYMBOL(__cpu_online_mask);
|
2008-12-30 06:35:14 +08:00
|
|
|
|
2016-01-21 07:00:19 +08:00
|
|
|
struct cpumask __cpu_present_mask __read_mostly;
|
|
|
|
|
EXPORT_SYMBOL(__cpu_present_mask);
|
2008-12-30 06:35:14 +08:00
|
|
|
|
2016-01-21 07:00:19 +08:00
|
|
|
struct cpumask __cpu_active_mask __read_mostly;
|
|
|
|
|
EXPORT_SYMBOL(__cpu_active_mask);
|
2008-12-30 06:35:16 +08:00
|
|
|
|
|
|
|
|
void init_cpu_present(const struct cpumask *src)
|
|
|
|
|
{
|
2016-01-21 07:00:16 +08:00
|
|
|
cpumask_copy(&__cpu_present_mask, src);
|
2008-12-30 06:35:16 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void init_cpu_possible(const struct cpumask *src)
|
|
|
|
|
{
|
2016-01-21 07:00:16 +08:00
|
|
|
cpumask_copy(&__cpu_possible_mask, src);
|
2008-12-30 06:35:16 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void init_cpu_online(const struct cpumask *src)
|
|
|
|
|
{
|
2016-01-21 07:00:16 +08:00
|
|
|
cpumask_copy(&__cpu_online_mask, src);
|
2008-12-30 06:35:16 +08:00
|
|
|
}
|