this_cpu operations
-------------------

this_cpu operations are a way of optimizing access to per cpu
variables associated with the *currently* executing processor. This is
done through the use of segment registers (or a dedicated register where
the cpu permanently stored the beginning of the per cpu area for a
specific processor).

this_cpu operations add a per cpu variable offset to the processor
specific per cpu base and encode that operation in the instruction
operating on the per cpu variable.

This means that there are no atomicity issues between the calculation of
the offset and the operation on the data. Therefore it is not
necessary to disable preemption or interrupts to ensure that the
processor is not changed between the calculation of the address and
the operation on the data.

Read-modify-write operations are of particular interest. Frequently
processors have special lower latency instructions that can operate
without the typical synchronization overhead, but still provide some
sort of relaxed atomicity guarantees. The x86, for example, can execute
RMW (Read Modify Write) instructions like inc/dec/cmpxchg without the
lock prefix and the associated latency penalty.

Access to the variable without the lock prefix is not synchronized but
synchronization is not necessary since we are dealing with per cpu
data specific to the currently executing processor. Only the current
processor should be accessing that variable and therefore there are no
concurrency issues with other processors in the system.

Please note that accesses by remote processors to a per cpu area are
exceptional situations and may impact performance and/or correctness
(remote write operations) of local RMW operations via this_cpu_*.

The main use of the this_cpu operations has been to optimize counter
operations.

The following this_cpu() operations with implied preemption protection
are defined. These operations can be used without worrying about
preemption and interrupts.

        this_cpu_read(pcp)
        this_cpu_write(pcp, val)
        this_cpu_add(pcp, val)
        this_cpu_and(pcp, val)
        this_cpu_or(pcp, val)
        this_cpu_add_return(pcp, val)
        this_cpu_xchg(pcp, nval)
        this_cpu_cmpxchg(pcp, oval, nval)
        this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
        this_cpu_sub(pcp, val)
        this_cpu_inc(pcp)
        this_cpu_dec(pcp)
        this_cpu_sub_return(pcp, val)
        this_cpu_inc_return(pcp)
        this_cpu_dec_return(pcp)


Inner working of this_cpu operations
------------------------------------

On x86 the fs: or the gs: segment registers contain the base of the
per cpu area. It is then possible to simply use the segment override
to relocate a per cpu relative address to the proper per cpu area for
the processor. So the relocation to the per cpu base is encoded in the
instruction via a segment register prefix.

For example:

        DEFINE_PER_CPU(int, x);
        int z;

        z = this_cpu_read(x);

results in a single instruction

        mov ax, gs:[x]

instead of a sequence of calculation of the address and then a fetch
from that address which occurs with the per cpu operations. Before
this_cpu_ops such sequence also required preempt disable/enable to
prevent the kernel from moving the thread to a different processor
while the calculation is performed.

Consider the following this_cpu operation:

        this_cpu_inc(x)

The above results in the following single instruction (no lock prefix!)

        inc gs:[x]

instead of the following operations required if there is no segment
register:

        int *y;
        int cpu;

        cpu = get_cpu();
        y = per_cpu_ptr(&x, cpu);
        (*y)++;
        put_cpu();

Note that these operations can only be used on per cpu data that is
reserved for a specific processor. Without disabling preemption in the
surrounding code this_cpu_inc() will only guarantee that one of the
per cpu counters is correctly incremented. However, there is no
guarantee that the OS will not move the process directly before or
after the this_cpu instruction is executed. In general this means that
the value of the individual counters for each processor are
meaningless. The sum of all the per cpu counters is the only value
that is of interest.

Per cpu variables are used for performance reasons. Bouncing cache
lines can be avoided if multiple processors concurrently go through
the same code paths.  Since each processor has its own per cpu
variables no concurrent cache line updates take place. The price that
has to be paid for this optimization is the need to add up the per cpu
counters when the value of a counter is needed.


Special operations:
-------------------

        y = this_cpu_ptr(&x)

Takes the offset of a per cpu variable (&x !) and returns the address
of the per cpu variable that belongs to the currently executing
processor.  this_cpu_ptr avoids multiple steps that the common
get_cpu/put_cpu sequence requires. No processor number is
available. Instead, the offset of the local per cpu area is simply
added to the per cpu offset.

Note that this operation is usually used in a code segment when
preemption has been disabled. The pointer is then used to
access local per cpu data in a critical section. When preemption
is re-enabled this pointer is usually no longer useful since it may
no longer point to per cpu data of the current processor.


Per cpu variables and offsets
-----------------------------

Per cpu variables have *offsets* to the beginning of the per cpu
area. They do not have addresses although they look like that in the
code. Offsets cannot be directly dereferenced. The offset must be
added to a base pointer of a per cpu area of a processor in order to
form a valid address.

Therefore the use of x or &x outside of the context of per cpu
operations is invalid and will generally be treated like a NULL
pointer dereference.

        DEFINE_PER_CPU(int, x);

In the context of per cpu operations the above implies that x is a per
cpu variable. Most this_cpu operations take a cpu variable.

        int __percpu *p = &x;

&x and hence p is the *offset* of a per cpu variable. this_cpu_ptr()
takes the offset of a per cpu variable which makes this look a bit
strange.


Operations on a field of a per cpu structure
--------------------------------------------

Let's say we have a percpu structure

        struct s {
                int n,m;
        };

        DEFINE_PER_CPU(struct s, p);


Operations on these fields are straightforward

        this_cpu_inc(p.m)

        z = this_cpu_cmpxchg(p.m, 0, 1);


If we have an offset to struct s:

        struct s __percpu *ps = &p;

        this_cpu_dec(ps->m);

        z = this_cpu_inc_return(ps->n);


The calculation of the pointer may require the use of this_cpu_ptr()
if we do not make use of this_cpu ops later to manipulate fields:

        struct s *pp;

        pp = this_cpu_ptr(&p);

        pp->m--;

        z = pp->n++;


Variants of this_cpu ops
-------------------------

this_cpu ops are interrupt safe. Some architectures do not support
these per cpu local operations. In that case the operation must be
replaced by code that disables interrupts, then does the operations
that are guaranteed to be atomic and then re-enable interrupts. Doing
so is expensive. If there are other reasons why the scheduler cannot
change the processor we are executing on then there is no reason to
disable interrupts. For that purpose the following __this_cpu operations
are provided.

These operations have no guarantee against concurrent interrupts or
preemption. If a per cpu variable is not used in an interrupt context
and the scheduler cannot preempt, then they are safe. If any interrupts
still occur while an operation is in progress and if the interrupt too
modifies the variable, then RMW actions can not be guaranteed to be
safe.

        __this_cpu_read(pcp)
        __this_cpu_write(pcp, val)
        __this_cpu_add(pcp, val)
        __this_cpu_and(pcp, val)
        __this_cpu_or(pcp, val)
        __this_cpu_add_return(pcp, val)
        __this_cpu_xchg(pcp, nval)
        __this_cpu_cmpxchg(pcp, oval, nval)
        __this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
        __this_cpu_sub(pcp, val)
        __this_cpu_inc(pcp)
        __this_cpu_dec(pcp)
        __this_cpu_sub_return(pcp, val)
        __this_cpu_inc_return(pcp)
        __this_cpu_dec_return(pcp)


Will increment x and will not fall-back to code that disables
interrupts on platforms that cannot accomplish atomicity through
address relocation and a Read-Modify-Write operation in the same
instruction.


&this_cpu_ptr(pp)->n vs this_cpu_ptr(&pp->n)
--------------------------------------------

The first operation takes the offset and forms an address and then
adds the offset of the n field. This may result in two add
instructions emitted by the compiler.

The second one first adds the two offsets and then does the
relocation.  IMHO the second form looks cleaner and has an easier time
with (). The second form also is consistent with the way
this_cpu_read() and friends are used.


Remote access to per cpu data
------------------------------

Per cpu data structures are designed to be used by one cpu exclusively.
If you use the variables as intended, this_cpu_ops() are guaranteed to
be "atomic" as no other CPU has access to these data structures.

There are special cases where you might need to access per cpu data
structures remotely. It is usually safe to do a remote read access
and that is frequently done to summarize counters. Remote write access
something which could be problematic because this_cpu ops do not
have lock semantics. A remote write may interfere with a this_cpu
RMW operation.

Remote write accesses to percpu data structures are highly discouraged
unless absolutely necessary. Please consider using an IPI to wake up
the remote CPU and perform the update to its per cpu area.

To access per-cpu data structure remotely, typically the per_cpu_ptr()
function is used:


        DEFINE_PER_CPU(struct data, datap);

        struct data *p = per_cpu_ptr(&datap, cpu);

This makes it explicit that we are getting ready to access a percpu
area remotely.

You can also do the following to convert the datap offset to an address

        struct data *p = this_cpu_ptr(&datap);

but, passing of pointers calculated via this_cpu_ptr to other cpus is
unusual and should be avoided.

Remote access are typically only for reading the status of another cpus
per cpu data. Write accesses can cause unique problems due to the
relaxed synchronization requirements for this_cpu operations.

One example that illustrates some concerns with write operations is
the following scenario that occurs because two per cpu variables
share a cache-line but the relaxed synchronization is applied to
only one process updating the cache-line.

Consider the following example


        struct test {
                atomic_t a;
                int b;
        };

        DEFINE_PER_CPU(struct test, onecacheline);

There is some concern about what would happen if the field 'a' is updated
remotely from one processor and the local processor would use this_cpu ops
to update field b. Care should be taken that such simultaneous accesses to
data within the same cache line are avoided. Also costly synchronization
may be necessary. IPIs are generally recommended in such scenarios instead
of a remote write to the per cpu area of another processor.

Even in cases where the remote writes are rare, please bear in
mind that a remote write will evict the cache line from the processor
that most likely will access it. If the processor wakes up and finds a
missing local cache line of a per cpu area, its performance and hence
the wake up times will be affected.

Christoph Lameter, August 4th, 2014
Pranith Kumar, Aug 2nd, 2014