Dependent on CONFIG_SMP the num_*_cpus() functions return unsigned or
signed values. Let them always return unsigned values to avoid strange
casts.
Fixes at least one warning:
kernel/kprobes.c: In function 'register_kretprobe':
kernel/kprobes.c:1038: warning: comparison of distinct pointer types lacks a cast
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
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>
Currently, rcu_needs_cpu() simply checks whether the current CPU
has an outstanding RCU callback, which means that the last CPU
to go into dyntick-idle mode might wait a few ticks for the
relevant grace periods to complete. However, if all the other
CPUs are in dyntick-idle mode, and if this CPU is in a quiescent
state (which it is for RCU-bh and RCU-sched any time that we are
considering going into dyntick-idle mode), then the grace period
is instantly complete.
This patch therefore repeatedly invokes the RCU grace-period
machinery in order to force any needed grace periods to complete
quickly. It does so a limited number of times in order to
prevent starvation by an RCU callback function that might pass
itself to call_rcu().
However, if any CPU other than the current one is not in
dyntick-idle mode, fall back to simply checking (with fix to bug
noted by Lai Jiangshan). Also, take advantage of last
grace-period forcing, the opportunity to do so noted by Steve
Rostedt. And apply simplified #ifdef condition suggested by
Frederic Weisbecker.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: laijs@cn.fujitsu.com
Cc: dipankar@in.ibm.com
Cc: mathieu.desnoyers@polymtl.ca
Cc: josh@joshtriplett.org
Cc: dvhltc@us.ibm.com
Cc: niv@us.ibm.com
Cc: peterz@infradead.org
Cc: rostedt@goodmis.org
Cc: Valdis.Kletnieks@vt.edu
Cc: dhowells@redhat.com
LKML-Reference: <1266887105-1528-15-git-send-email-paulmck@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Since (e761b77: cpu hotplug, sched: Introduce cpu_active_map and redo
sched domain managment) we have cpu_active_mask which is suppose to rule
scheduler migration and load-balancing, except it never (fully) did.
The particular problem being solved here is a crash in try_to_wake_up()
where select_task_rq() ends up selecting an offline cpu because
select_task_rq_fair() trusts the sched_domain tree to reflect the
current state of affairs, similarly select_task_rq_rt() trusts the
root_domain.
However, the sched_domains are updated from CPU_DEAD, which is after the
cpu is taken offline and after stop_machine is done. Therefore it can
race perfectly well with code assuming the domains are right.
Cure this by building the domains from cpu_active_mask on
CPU_DOWN_PREPARE.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The new ones have pretty kerneldoc. Move the old ones to the end to
avoid confusing people.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: benh@kernel.crashing.org
We're not forcing removal of the old cpu_ functions, but we might as
well delete the now-unused ones.
Especially CPUMASK_ALLOC and friends. I actually got a phone call (!)
from a hacker who thought I had introduced them as the new cpumask
API. He seemed bewildered that I had lost all taste.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: benh@kernel.crashing.org
(Thanks to Al Viro for reminding me of this, via Ingo)
CPU_MASK_ALL is the (deprecated) "all bits set" cpumask, defined as so:
#define CPU_MASK_ALL (cpumask_t) { { ... } }
Taking the address of such a temporary is questionable at best,
unfortunately 321a8e9d (cpumask: add CPU_MASK_ALL_PTR macro) added
CPU_MASK_ALL_PTR:
#define CPU_MASK_ALL_PTR (&CPU_MASK_ALL)
Which formalizes this practice. One day gcc could bite us over this
usage (though we seem to have gotten away with it so far).
Now all callers are removed, we kill it.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Ingo Molnar <mingo@elte.hu>
Reported-by: Al Viro <viro@zeniv.linux.org.uk>
Cc: Mike Travis <travis@sgi.com>
This patch can remove spinlock from struct call_function_data, the
reasons are below:
1: add a new interface for cpumask named cpumask_test_and_clear_cpu(),
it can atomically test and clear specific cpu, we can use it instead
of cpumask_test_cpu() and cpumask_clear_cpu() and no need data->lock
to protect those in generic_smp_call_function_interrupt().
2: in smp_call_function_many(), after csd_lock() return, the current's
cfd_data is deleted from call_function list, so it not have race
between other cpus, then cfs_data is only used in
smp_call_function_many() that must disable preemption and not from
a hardware interrupthandler or from a bottom half handler to call,
only the correspond cpu can use it, so it not have race in current
cpu, no need cfs_data->lock to protect it.
3: after 1 and 2, cfs_data->lock is only use to protect cfs_data->refs in
generic_smp_call_function_interrupt(), so we can define cfs_data->refs
to atomic_t, and no need cfs_data->lock any more.
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Peter Zijlstra <peterz@infradead.org>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
[akpm@linux-foundation.org: use atomic_dec_return()]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When 'and'ing two bitmasks (where 'andnot' is a variation on it), some
cases want to know whether the result is the empty set or not. In
particular, the TLB IPI sending code wants to do cpumask operations and
determine if there are any CPU's left in the final set.
So this just makes the bitmask (and cpumask) functions return a boolean
for whether the result has any bits set.
Cc: stable@kernel.org (2.6.30, needed by TLB shootdown fix)
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Impact: new debug CONFIG options
This helps find unconverted code. It currently breaks compile horribly,
but we never wanted a flag day so that's expected.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
They're only for use in boot/cpu hotplug code anyway, and this avoids
the use of deprecated cpu_*_map.
Stephen Rothwell points out that gcc 4.2.4 (on powerpc at least)
didn't like the cast away of const anyway:
include/linux/cpumask.h: In function 'set_cpu_possible':
include/linux/cpumask.h:1052: warning: passing argument 2 of 'cpumask_set_cpu' discards qualifiers from pointer target type
So this kills two birds with one stone.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Changes:
1) cpumask_t to struct cpumask,
2) cpus_weight_nr to cpumask_weight,
3) cpu_isset to cpumask_test_cpu,
4) ->bits to cpumask_bits()
5) cpu_*_map to cpu_*_mask.
6) for_each_cpu_mask_nr to for_each_cpu
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Impact: cleanup
This implements the obsolescent cpu_online_map in terms of
cpu_online_mask, rather than the other way around. Same for the other
maps.
The documentation comments are also updated to refer to _mask rather
than _map.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Mike Travis <travis@sgi.com>
Impact: New API
This will be needed in x86 code to allocate the domain and old_domain
cpumasks on the same node as where the containing irq_cfg struct is
allocated.
(Also fixes double-dump_stack on rare CONFIG_DEBUG_PER_CPU_MAPS case)
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au> (re-impl alloc_cpumask_var)
Impact: futureproof as we convert more code to new APIs
The old cpumask operators treat all NR_CPUS bits as relevent, the new
ones use nr_cpumask_bits. For large NR_CPUS and small nr_cpu_ids, this
makes a difference.
However, mixing the two can cause problems with undefined bits. An
arch which sets CONFIG_CPUMASK_OFFSTACK should have converted across
to the new operators, so it's safe in that case.
(Thanks to Stephen Rothwell for bisecting the initial unused-bits bug,
and Mike Travis for this solution).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Mike Travis <travis@sgi.com>
Impact: change calling convention of existing cpumask APIs
Most cpumask functions started with cpus_: these have been replaced by
cpumask_ ones which take struct cpumask pointers as expected.
These four functions don't have good replacement names; fortunately
they're rarely used, so we just change them over.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Mike Travis <travis@sgi.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: paulus@samba.org
Cc: mingo@redhat.com
Cc: tony.luck@intel.com
Cc: ralf@linux-mips.org
Cc: Greg Kroah-Hartman <gregkh@suse.de>
Cc: cl@linux-foundation.org
Cc: srostedt@redhat.com
Impact: cleanup
Clean up based on feedback from Andrew Morton and others:
- change to inline functions instead of macros
- add __init to bootmem method
- add a missing debug check
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: introduce new APIs
We want to deprecate cpumasks on the stack, as we are headed for
gynormous numbers of CPUs. Eventually, we want to head towards an
undefined 'struct cpumask' so they can never be declared on stack.
1) New cpumask functions which take pointers instead of copies.
(cpus_* -> cpumask_*)
2) Several new helpers to reduce requirements for temporary cpumasks
(cpumask_first_and, cpumask_next_and, cpumask_any_and)
3) Helpers for declaring cpumasks on or offstack for large NR_CPUS
(cpumask_var_t, alloc_cpumask_var and free_cpumask_var)
4) 'struct cpumask' for explicitness and to mark new-style code.
5) Make iterator functions stop at nr_cpu_ids (a runtime constant),
not NR_CPUS for time efficiency and for smaller dynamic allocations
in future.
6) cpumask_copy() so we can allocate less than a full cpumask eventually
(for alloc_cpumask_var), and so we can eliminate the 'struct cpumask'
definition eventually.
7) work_on_cpu() helper for doing task on a CPU, rather than saving old
cpumask for current thread and manipulating it.
8) smp_call_function_many() which is smp_call_function_mask() except
taking a cpumask pointer.
Note that this patch simply introduces the new functions and leaves
the obsolescent ones in place. This is to simplify the transition
patches.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
when you take the address of the result. Noticed on a sparc64 compile
using a version 3.4.5 cross compiler.
kernel/time/tick-common.c: In function `tick_check_new_device':
kernel/time/tick-common.c:210: error: invalid lvalue in unary `&'
...
Just make it a regular expression.
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
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/2008/7/27/156http://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>
If an arch doesn't define cpumask_of_cpu_map, create a generic
statically-initialized one for them. This allows removal of the buggy
cpumask_of_cpu() macro (&cpumask_of_cpu() gives address of
out-of-scope var).
An arch with NR_CPUS of 4096 probably wants to allocate this itself
based on the actual number of CPUs, since otherwise they're using 2MB
of rodata (1024 cpus means 128k). That's what
CONFIG_HAVE_CPUMASK_OF_CPU_MAP is for (only x86/64 does so at the
moment).
In future as we support more CPUs, we'll need to resort to a
get_cpu_map()/put_cpu_map() allocation scheme.
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Jack Steiner <steiner@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* 'sched/for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
sched: hrtick_enabled() should use cpu_active()
sched, x86: clean up hrtick implementation
sched: fix build error, provide partition_sched_domains() unconditionally
sched: fix warning in inc_rt_tasks() to not declare variable 'rq' if it's not needed
cpu hotplug: Make cpu_active_map synchronization dependency clear
cpu hotplug, sched: Introduce cpu_active_map and redo sched domain managment (take 2)
sched: rework of "prioritize non-migratable tasks over migratable ones"
sched: reduce stack size in isolated_cpu_setup()
Revert parts of "ftrace: do not trace scheduler functions"
Fixed up conflicts in include/asm-x86/thread_info.h (due to the
TIF_SINGLESTEP unification vs TIF_HRTICK_RESCHED removal) and
kernel/sched_fair.c (due to cpu_active_map vs for_each_cpu_mask_nr()
introduction).
* Rename CPUMASK_VAR --> CPUMASK_PTR (and simplify)
* Fix a semantic error in CPUMASK_ALLOC
* Add a bit of commentry to cpumask.h
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* Provide a generic set of CPUMASK_ALLOC macros patterned after the
SCHED_CPUMASK_ALLOC macros. This is used where multiple cpumask_t
variables are declared on the stack to reduce the amount of stack
space required.
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* This patch replaces the dangerous lvalue version of cpumask_of_cpu
with new cpumask_of_cpu_ptr macros. These are patterned after the
node_to_cpumask_ptr macros.
In general terms, if there is a cpumask_of_cpu_map[] then a pointer to
the cpumask_of_cpu_map[cpu] entry is used. The cpumask_of_cpu_map
is provided when there is a large NR_CPUS count, reducing
greatly the amount of code generated and stack space used for
cpumask_of_cpu(). The pointer to the cpumask_t value is needed for
calling set_cpus_allowed_ptr() to reduce the amount of stack space
needed to pass the cpumask_t value.
If there isn't a cpumask_of_cpu_map[], then a temporary variable is
declared and filled in with value from cpumask_of_cpu(cpu) as well as
a pointer variable pointing to this temporary variable. Afterwards,
the pointer is used to reference the cpumask value. The compiler
will optimize out the extra dereference through the pointer as well
as the stack space used for the pointer, resulting in identical code.
A good example of the orthogonal usages is in net/sunrpc/svc.c:
case SVC_POOL_PERCPU:
{
unsigned int cpu = m->pool_to[pidx];
cpumask_of_cpu_ptr(cpumask, cpu);
*oldmask = current->cpus_allowed;
set_cpus_allowed_ptr(current, cpumask);
return 1;
}
case SVC_POOL_PERNODE:
{
unsigned int node = m->pool_to[pidx];
node_to_cpumask_ptr(nodecpumask, node);
*oldmask = current->cpus_allowed;
set_cpus_allowed_ptr(current, nodecpumask);
return 1;
}
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This is based on Linus' idea of creating cpu_active_map that prevents
scheduler load balancer from migrating tasks to the cpu that is going
down.
It allows us to simplify domain management code and avoid unecessary
domain rebuilds during cpu hotplug event handling.
Please ignore the cpusets part for now. It needs some more work in order
to avoid crazy lock nesting. Although I did simplfy and unify domain
reinitialization logic. We now simply call partition_sched_domains() in
all the cases. This means that we're using exact same code paths as in
cpusets case and hence the test below cover cpusets too.
Cpuset changes to make rebuild_sched_domains() callable from various
contexts are in the separate patch (right next after this one).
This not only boots but also easily handles
while true; do make clean; make -j 8; done
and
while true; do on-off-cpu 1; done
at the same time.
(on-off-cpu 1 simple does echo 0/1 > /sys/.../cpu1/online thing).
Suprisingly the box (dual-core Core2) is quite usable. In fact I'm typing
this on right now in gnome-terminal and things are moving just fine.
Also this is running with most of the debug features enabled (lockdep,
mutex, etc) no BUG_ONs or lockdep complaints so far.
I believe I addressed all of the Dmitry's comments for original Linus'
version. I changed both fair and rt balancer to mask out non-active cpus.
And replaced cpu_is_offline() with !cpu_active() in the main scheduler
code where it made sense (to me).
Signed-off-by: Max Krasnyanskiy <maxk@qualcomm.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Gregory Haskins <ghaskins@novell.com>
Cc: dmitry.adamushko@gmail.com
Cc: pj@sgi.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In linux-next there is a commit ("x86: Add performance variants of cpumask
operators") which, as part of the 4096 cpu support work adds some new APIs
for dealing with cpu masks. Add trivial versions of these now so that
subsystems can update in a timely manner and avoid conflicts in linux-next
and the next merge window.
Cc: Mike Travis <travis@sgi.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The for_each_cpu_mask loop is used quite often in the kernel. It
makes use of two functions: first_cpu and next_cpu. This patch
changes for_each_cpu_mask to use only the latter. Because next_cpu
finds the next eligible cpu _after_ the given one, the iteration
variable has to be initialized to -1 and next_cpu has to be
called with this value before the first iteration. An x86_64
defconfig kernel (from sched/latest) is about 2500 bytes smaller
with this patch applied:
text data bss dec hex filename
6222517 917952 749932 7890401 7865e1 vmlinux.orig
6219922 917952 749932 7887806 785bbe vmlinux
The same size reduction is seen for defconfig+MAXSMP
text data bss dec hex filename
6241772 2563968 1492716 10298456 9d2458 vmlinux.orig
6239211 2563968 1492716 10295895 9d1a57 vmlinux
Signed-off-by: Alexander van Heukelum <heukelum@fastmail.fm>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
* Increase performance for systems with large count NR_CPUS by limiting
the range of the cpumask operators that loop over the bits in a cpumask_t
variable. This removes a large amount of wasted cpu cycles.
* Add performance variants of the cpumask operators:
int cpus_weight_nr(mask) Same using nr_cpu_ids instead of NR_CPUS
int first_cpu_nr(mask) Number lowest set bit, or nr_cpu_ids
int next_cpu_nr(cpu, mask) Next cpu past 'cpu', or nr_cpu_ids
for_each_cpu_mask_nr(cpu, mask) for-loop cpu over mask using nr_cpu_ids
* Modify following to use performance variants:
#define num_online_cpus() cpus_weight_nr(cpu_online_map)
#define num_possible_cpus() cpus_weight_nr(cpu_possible_map)
#define num_present_cpus() cpus_weight_nr(cpu_present_map)
#define for_each_possible_cpu(cpu) for_each_cpu_mask_nr((cpu), ...)
#define for_each_online_cpu(cpu) for_each_cpu_mask_nr((cpu), ...)
#define for_each_present_cpu(cpu) for_each_cpu_mask_nr((cpu), ...)
* Comment added to include/linux/cpumask.h:
Note: The alternate operations with the suffix "_nr" are used
to limit the range of the loop to nr_cpu_ids instead of
NR_CPUS when NR_CPUS > 64 for performance reasons.
If NR_CPUS is <= 64 then most assembler bitmask
operators execute faster with a constant range, so
the operator will continue to use NR_CPUS.
Another consideration is that nr_cpu_ids is initialized
to NR_CPUS and isn't lowered until the possible cpus are
discovered (including any disabled cpus). So early uses
will span the entire range of NR_CPUS.
(The net effect is that for systems with 64 or less CPU's there are no
functional changes.)
For inclusion into sched-devel/latest tree.
Based on:
git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6.git
+ sched-devel/latest .../mingo/linux-2.6-sched-devel.git
Cc: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <clameter@sgi.com>
Reviewed-by: Paul Jackson <pj@sgi.com>
Reviewed-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
They aren't used. They were briefly used as part of some other patches to
provide an alternative format for displaying some /proc and /sys cpumasks.
They probably should have been removed when those other patches were dropped,
in favor of a different solution.
Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: "Mike Travis" <travis@sgi.com>
Cc: "Bert Wesarg" <bert.wesarg@googlemail.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: WANG Cong <xiyou.wangcong@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The following adds two more bitmap operators, bitmap_onto() and bitmap_fold(),
with the usual cpumask and nodemask wrappers.
The bitmap_onto() operator computes one bitmap relative to another. If the
n-th bit in the origin mask is set, then the m-th bit of the destination mask
will be set, where m is the position of the n-th set bit in the relative mask.
The bitmap_fold() operator folds a bitmap into a second that has bit m set iff
the input bitmap has some bit n set, where m == n mod sz, for the specified sz
value.
There are two substantive changes between this patch and its
predecessor bitmap_relative:
1) Renamed bitmap_relative() to be bitmap_onto().
2) Added bitmap_fold().
The essential motivation for bitmap_onto() is to provide a mechanism for
converting a cpuset-relative CPU or Node mask to an absolute mask. Cpuset
relative masks are written as if the current task were in a cpuset whose CPUs
or Nodes were just the consecutive ones numbered 0..N-1, for some N. The
bitmap_onto() operator is provided in anticipation of adding support for the
first such cpuset relative mask, by the mbind() and set_mempolicy() system
calls, using a planned flag of MPOL_F_RELATIVE_NODES. These bitmap operators
(and their nodemask wrappers, in particular) will be used in code that
converts the user specified cpuset relative memory policy to a specific system
node numbered policy, given the current mems_allowed of the tasks cpuset.
Such cpuset relative mempolicies will address two deficiencies
of the existing interface between cpusets and mempolicies:
1) A task cannot at present reliably establish a cpuset
relative mempolicy because there is an essential race
condition, in that the tasks cpuset may be changed in
between the time the task can query its cpuset placement,
and the time the task can issue the applicable mbind or
set_memplicy system call.
2) A task cannot at present establish what cpuset relative
mempolicy it would like to have, if it is in a smaller
cpuset than it might have mempolicy preferences for,
because the existing interface only allows specifying
mempolicies for nodes currently allowed by the cpuset.
Cpuset relative mempolicies are useful for tasks that don't distinguish
particularly between one CPU or Node and another, but only between how many of
each are allowed, and the proper placement of threads and memory pages on the
various CPUs and Nodes available.
The motivation for the added bitmap_fold() can be seen in the following
example.
Let's say an application has specified some mempolicies that presume 16 memory
nodes, including say a mempolicy that specified MPOL_F_RELATIVE_NODES (cpuset
relative) nodes 12-15. Then lets say that application is crammed into a
cpuset that only has 8 memory nodes, 0-7. If one just uses bitmap_onto(),
this mempolicy, mapped to that cpuset, would ignore the requested relative
nodes above 7, leaving it empty of nodes. That's not good; better to fold the
higher nodes down, so that some nodes are included in the resulting mapped
mempolicy. In this case, the mempolicy nodes 12-15 are taken modulo 8 (the
weight of the mems_allowed of the confining cpuset), resulting in a mempolicy
specifying nodes 4-7.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: <kosaki.motohiro@jp.fujitsu.com>
Cc: <ray-lk@madrabbit.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* Here is a simple patch to use an allocated array of cpumasks to
represent cpumask_of_cpu() instead of constructing one on the stack.
It's based on the Kconfig option "HAVE_CPUMASK_OF_CPU_MAP" which is
currently only set for x86_64 SMP. Otherwise the the existing
cpumask_of_cpu() is used but has been changed to produce an lvalue
so a pointer to it can be used.
Cc: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* Add a static cpumask_t variable "CPU_MASK_ALL_PTR" to use as
a pointer reference to CPU_MASK_ALL. This reduces where possible
the instances where CPU_MASK_ALL allocates and fills a large
array on the stack. Used only if NR_CPUS > BITS_PER_LONG.
* Change init/main.c to use new set_cpus_allowed_ptr().
Depends on:
[sched-devel]: sched: add new set_cpus_allowed_ptr function
Cc: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Add a new function cpumask_scnprintf_len() to return the number of
characters needed to display "len" cpumask bits. The current method
of allocating NR_CPUS bytes is incorrect as what's really needed is
9 characters per 32-bit word of cpumask bits (8 hex digits plus the
seperator [','] or the terminating NULL.) This function provides the
caller the means to allocate the correct string length.
Cc: Paul Jackson <pj@sgi.com>
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
make randconfig bootup testing found that the cpufreq code
crashes on bootup, if the powernow-k8 driver is enabled and
if maxcpus=1 passed on the boot line to a !CONFIG_HOTPLUG_CPU
kernel.
First lockdep found out that there's an inconsistent unlock
sequence:
=====================================
[ BUG: bad unlock balance detected! ]
-------------------------------------
swapper/1 is trying to release lock (&per_cpu(cpu_policy_rwsem, cpu)) at:
[<ffffffff806ffd8e>] unlock_policy_rwsem_write+0x3c/0x42
but there are no more locks to release!
Call Trace:
[<ffffffff806ffd8e>] unlock_policy_rwsem_write+0x3c/0x42
[<ffffffff80251c29>] print_unlock_inbalance_bug+0x104/0x12c
[<ffffffff80252f3a>] mark_held_locks+0x56/0x94
[<ffffffff806ffd8e>] unlock_policy_rwsem_write+0x3c/0x42
[<ffffffff807008b6>] cpufreq_add_dev+0x2a8/0x5c4
...
then shortly afterwards the cpufreq code crashed on an assert:
------------[ cut here ]------------
kernel BUG at drivers/cpufreq/cpufreq.c:1068!
invalid opcode: 0000 [1] SMP
[...]
Call Trace:
[<ffffffff805145d6>] sysdev_driver_unregister+0x5b/0x91
[<ffffffff806ff520>] cpufreq_register_driver+0x15d/0x1a2
[<ffffffff80cc0596>] powernowk8_init+0x86/0x94
[...]
---[ end trace 1e9219be2b4431de ]---
the bug was caused by maxcpus=1 bootup, which brought up the
secondary core as !cpu_online() but !cpu_is_offline() either,
which on on !CONFIG_HOTPLUG_CPU is always 0 (include/linux/cpu.h):
/* CPUs don't go offline once they're online w/o CONFIG_HOTPLUG_CPU */
static inline int cpu_is_offline(int cpu) { return 0; }
but the cpufreq code uses cpu_online() and cpu_is_offline() in
a mixed way - the low-level drivers use cpu_online(), while
the cpufreq core uses cpu_is_offline(). This opened up the
possibility to add the non-initialized sysdev device of the
secondary core:
cpufreq-core: trying to register driver powernow-k8
cpufreq-core: adding CPU 0
powernow-k8: BIOS error - no PSB or ACPI _PSS objects
cpufreq-core: initialization failed
cpufreq-core: adding CPU 1
cpufreq-core: initialization failed
which then blew up. The fix is to make cpu_is_offline() always
the negation of cpu_online(). With that fix applied the kernel
boots up fine without crashing:
Calling initcall 0xffffffff80cc0510: powernowk8_init+0x0/0x94()
powernow-k8: Found 1 AMD Athlon(tm) 64 X2 Dual Core Processor 3800+ processors (1 cpu cores) (version 2.20.00)
powernow-k8: BIOS error - no PSB or ACPI _PSS objects
initcall 0xffffffff80cc0510: powernowk8_init+0x0/0x94() returned -19.
initcall 0xffffffff80cc0510 ran for 19 msecs: powernowk8_init+0x0/0x94()
Calling initcall 0xffffffff80cc328f: init_lapic_nmi_sysfs+0x0/0x39()
We could fix this by making CPU enumeration aware of max_cpus, but that
would be more fragile IMO, and the cpu_online(cpu) != cpu_is_offline(cpu)
possibility was quite confusing and a continuous source of bugs too.
Most distributions have kernels with CPU hotplug enabled, so this bug
remained hidden for a long time.
Bug forensics:
The broken cpu_is_offline() API variant was introduced via:
commit a59d2e4e6977e7b94e003c96a41f07e96cddc340
Author: Rusty Russell <rusty@rustcorp.com.au>
Date: Mon Mar 8 06:06:03 2004 -0800
[PATCH] minor cleanups for hotplug CPUs
( this predates linux-2.6.git, this commit is available from Thomas's
historic git tree. )
Then 1.5 years later the cpufreq code made use of it:
commit c32b6b8e52
Author: Ashok Raj <ashok.raj@intel.com>
Date: Sun Oct 30 14:59:54 2005 -0800
[PATCH] create and destroy cpufreq sysfs entries based on cpu notifiers
+ if (cpu_is_offline(cpu))
+ return 0;
which is a correct use of the subtly broken new API. v2.6.15 then
shipped with this bug included.
then it took two more years for random-kernel qa to hit it.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We frequently need the maximum number of possible processors in order to
allocate arrays for all processors. So far this was done using
highest_possible_processor_id(). However, we do need the number of
processors not the highest id. Moreover the number was so far dynamically
calculated on each invokation. The number of possible processors does not
change when the system is running. We can therefore calculate that number
once.
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Frederik Deweerdt <frederik.deweerdt@gmail.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
lib/bitmap.c:bitmap_parse() is a library function that received as input a
user buffer. This seemed to have originated from the way the write_proc
function of the /proc filesystem operates.
This has been reworked to not use kmalloc and eliminates a lot of
get_user() overhead by performing one access_ok before using __get_user().
We need to test if we are in kernel or user space (is_user) and access the
buffer differently. We cannot use __get_user() to access kernel addresses
in all cases, for example in architectures with separate address space for
kernel and user.
This function will be useful for other uses as well; for example, taking
input for /sysfs instead of /proc, so it was changed to accept kernel
buffers. We have this use for the Linux UWB project, as part as the
upcoming bandwidth allocator code.
Only a few routines used this function and they were changed too.
Signed-off-by: Reinette Chatre <reinette.chatre@linux.intel.com>
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Joe Korty <joe.korty@ccur.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
On UP, this:
cpumask_t mask = node_to_cpumask(numa_node_id());
for_each_cpu_mask(cpu, mask)
does this:
mm/readahead.c: In function `node_readahead_aging':
mm/readahead.c:850: warning: unused variable `mask'
which is unpleasantly fixed by this:
Acked-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Convert a few stragglers over to for_each_possible_cpu(), remove
for_each_cpu().
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
for_each_cpu() is a for-loop over cpu_possible_map. for_each_online_cpu is
for-loop cpu over cpu_online_map. .....for_each_cpu() is not sufficiently
explicit and can lead to mistakes.
This patch adds for_each_possible_cpu() in preparation for the removal of
for_each_cpu().
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
text data bss dec hex filename
before: 3605597 1363528 363328 5332453 515de5 vmlinux
after: 3605295 1363612 363200 5332107 515c8b vmlinux
218 bytes saved.
Also, optimise any_online_cpu() out of existence on CONFIG_SMP=n.
This function seems inefficient. Can't we simply AND the two masks, then use
find_first_bit()?
Cc: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Shrinks the only caller (net/bridge/netfilter/ebtables.c) by 174 bytes.
Also, optimise highest_possible_processor_id() out of existence on
CONFIG_SMP=n.
Cc: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Initialising cpu_possible_map to all-ones with CONFIG_HOTPLUG_CPU means that
a) All for_each_cpu() loops will iterate across all NR_CPUS CPUs, rather
than over possible ones. That can be quite expensive.
b) Soon we'll be allocating per-cpu areas only for possible CPUs. So with
CPU_MASK_ALL, we'll be wasting memory.
I also switched voyager over to not use CPU_MASK_ALL in the non-CPU-hotplug
case. Should be OK..
I note that parisc is also using CPU_MASK_ALL. Suggest that it stop doing
that.
Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Paul Jackson <pj@sgi.com>
Cc: Ashok Raj <ashok.raj@intel.com>
Cc: Zwane Mwaikambo <zwane@linuxpower.ca>
Cc: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Heiko Carstens