[linux-2.6-block.git] / Documentation / RCU / Design / Requirements / Requirements.htmlx

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
        "http://www.w3.org/TR/html4/loose.dtd">
        <html>
        <head><title>A Tour Through RCU's Requirements [LWN.net]</title>
        <meta HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=utf-8">

<h1>A Tour Through RCU's Requirements</h1>

<p>Copyright IBM Corporation, 2015</p>
<p>Author: Paul E.&nbsp;McKenney</p>
<p><i>The initial version of this document appeared in the
<a href="https://lwn.net/">LWN</a> articles
<a href="https://lwn.net/Articles/652156/">here</a>,
<a href="https://lwn.net/Articles/652677/">here</a>, and
<a href="https://lwn.net/Articles/653326/">here</a>.</i></p>

<h2>Introduction</h2>

<p>
Read-copy update (RCU) is a synchronization mechanism that is often
used as a replacement for reader-writer locking.
RCU is unusual in that updaters do not block readers,
which means that RCU's read-side primitives can be exceedingly fast
and scalable.
In addition, updaters can make useful forward progress concurrently
with readers.
However, all this concurrency between RCU readers and updaters does raise
the question of exactly what RCU readers are doing, which in turn
raises the question of exactly what RCU's requirements are.

<p>
This document therefore summarizes RCU's requirements, and can be thought
of as an informal, high-level specification for RCU.
It is important to understand that RCU's specification is primarily
empirical in nature;
in fact, I learned about many of these requirements the hard way.
This situation might cause some consternation, however, not only
has this learning process been a lot of fun, but it has also been
a great privilege to work with so many people willing to apply
technologies in interesting new ways.

<p>
All that aside, here are the categories of currently known RCU requirements:
</p>

<ol>
<li>	<a href="#Fundamental Requirements">
	Fundamental Requirements</a>
<li>	<a href="#Fundamental Non-Requirements">Fundamental Non-Requirements</a>
<li>	<a href="#Parallelism Facts of Life">
	Parallelism Facts of Life</a>
<li>	<a href="#Quality-of-Implementation Requirements">
	Quality-of-Implementation Requirements</a>
<li>	<a href="#Linux Kernel Complications">
	Linux Kernel Complications</a>
<li>	<a href="#Software-Engineering Requirements">
	Software-Engineering Requirements</a>
<li>	<a href="#Other RCU Flavors">
	Other RCU Flavors</a>
<li>	<a href="#Possible Future Changes">
	Possible Future Changes</a>
</ol>

<p>
This is followed by a <a href="#Summary">summary</a>,
which is in turn followed by the inevitable
<a href="#Answers to Quick Quizzes">answers to the quick quizzes</a>.

<h2><a name="Fundamental Requirements">Fundamental Requirements</a></h2>

<p>
RCU's fundamental requirements are the closest thing RCU has to hard
mathematical requirements.
These are:

<ol>
<li>	<a href="#Grace-Period Guarantee">
	Grace-Period Guarantee</a>
<li>	<a href="#Publish-Subscribe Guarantee">
	Publish-Subscribe Guarantee</a>
<li>	<a href="#RCU Primitives Guaranteed to Execute Unconditionally">
	RCU Primitives Guaranteed to Execute Unconditionally</a>
<li>	<a href="#Guaranteed Read-to-Write Upgrade">
	Guaranteed Read-to-Write Upgrade</a>
</ol>

<h3><a name="Grace-Period Guarantee">Grace-Period Guarantee</a></h3>

<p>
RCU's grace-period guarantee is unusual in being premeditated:
Jack Slingwine and I had this guarantee firmly in mind when we started
work on RCU (then called &ldquo;rclock&rdquo;) in the early 1990s.
That said, the past two decades of experience with RCU have produced
a much more detailed understanding of this guarantee.

<p>
RCU's grace-period guarantee allows updaters to wait for the completion
of all pre-existing RCU read-side critical sections.
An RCU read-side critical section
begins with the marker <tt>rcu_read_lock()</tt> and ends with
the marker <tt>rcu_read_unlock()</tt>.
These markers may be nested, and RCU treats a nested set as one
big RCU read-side critical section.
Production-quality implementations of <tt>rcu_read_lock()</tt> and
<tt>rcu_read_unlock()</tt> are extremely lightweight, and in
fact have exactly zero overhead in Linux kernels built for production
use with <tt>CONFIG_PREEMPT=n</tt>.

<p>
This guarantee allows ordering to be enforced with extremely low
overhead to readers, for example:

<blockquote>
<pre>
 1 int x, y;
 2
 3 void thread0(void)
 4 {
 5   rcu_read_lock();
 6   r1 = READ_ONCE(x);
 7   r2 = READ_ONCE(y);
 8   rcu_read_unlock();
 9 }
10
11 void thread1(void)
12 {
13   WRITE_ONCE(x, 1);
14   synchronize_rcu();
15   WRITE_ONCE(y, 1);
16 }
</pre>
</blockquote>

<p>
Because the <tt>synchronize_rcu()</tt> on line&nbsp;14 waits for
all pre-existing readers, any instance of <tt>thread0()</tt> that
loads a value of zero from <tt>x</tt> must complete before
<tt>thread1()</tt> stores to <tt>y</tt>, so that instance must
also load a value of zero from <tt>y</tt>.
Similarly, any instance of <tt>thread0()</tt> that loads a value of
one from <tt>y</tt> must have started after the
<tt>synchronize_rcu()</tt> started, and must therefore also load
a value of one from <tt>x</tt>.
Therefore, the outcome:
<blockquote>
<pre>
(r1 == 0 &amp;&amp; r2 == 1)
</pre>
</blockquote>
cannot happen.

<p>@@QQ@@
Wait a minute!
You said that updaters can make useful forward progress concurrently
with readers, but pre-existing readers will block
<tt>synchronize_rcu()</tt>!!!
Just who are you trying to fool???
<p>@@QQA@@
First, if updaters do not wish to be blocked by readers, they can use
<tt>call_rcu()</tt> or <tt>kfree_rcu()</tt>, which will
be discussed later.
Second, even when using <tt>synchronize_rcu()</tt>, the other
update-side code does run concurrently with readers, whether pre-existing
or not.
<p>@@QQE@@

<p>
This scenario resembles one of the first uses of RCU in
<a href="https://en.wikipedia.org/wiki/DYNIX">DYNIX/ptx</a>,
which managed a distributed lock manager's transition into
a state suitable for handling recovery from node failure,
more or less as follows:

<blockquote>
<pre>
 1 #define STATE_NORMAL        0
 2 #define STATE_WANT_RECOVERY 1
 3 #define STATE_RECOVERING    2
 4 #define STATE_WANT_NORMAL   3
 5
 6 int state = STATE_NORMAL;
 7
 8 void do_something_dlm(void)
 9 {
10   int state_snap;
11
12   rcu_read_lock();
13   state_snap = READ_ONCE(state);
14   if (state_snap == STATE_NORMAL)
15     do_something();
16   else
17     do_something_carefully();
18   rcu_read_unlock();
19 }
20
21 void start_recovery(void)
22 {
23   WRITE_ONCE(state, STATE_WANT_RECOVERY);
24   synchronize_rcu();
25   WRITE_ONCE(state, STATE_RECOVERING);
26   recovery();
27   WRITE_ONCE(state, STATE_WANT_NORMAL);
28   synchronize_rcu();
29   WRITE_ONCE(state, STATE_NORMAL);
30 }
</pre>
</blockquote>

<p>
The RCU read-side critical section in <tt>do_something_dlm()</tt>
works with the <tt>synchronize_rcu()</tt> in <tt>start_recovery()</tt>
to guarantee that <tt>do_something()</tt> never runs concurrently
with <tt>recovery()</tt>, but with little or no synchronization
overhead in <tt>do_something_dlm()</tt>.

<p>@@QQ@@
Why is the <tt>synchronize_rcu()</tt> on line&nbsp;28 needed?
<p>@@QQA@@
Without that extra grace period, memory reordering could result in
<tt>do_something_dlm()</tt> executing <tt>do_something()</tt>
concurrently with the last bits of <tt>recovery()</tt>.
<p>@@QQE@@

<p>
In order to avoid fatal problems such as deadlocks,
an RCU read-side critical section must not contain calls to
<tt>synchronize_rcu()</tt>.
Similarly, an RCU read-side critical section must not
contain anything that waits, directly or indirectly, on completion of
an invocation of <tt>synchronize_rcu()</tt>.

<p>
Although RCU's grace-period guarantee is useful in and of itself, with
<a href="https://lwn.net/Articles/573497/">quite a few use cases</a>,
it would be good to be able to use RCU to coordinate read-side
access to linked data structures.
For this, the grace-period guarantee is not sufficient, as can
be seen in function <tt>add_gp_buggy()</tt> below.
We will look at the reader's code later, but in the meantime, just think of
the reader as locklessly picking up the <tt>gp</tt> pointer,
and, if the value loaded is non-<tt>NULL</tt>, locklessly accessing the
<tt>-&gt;a</tt> and <tt>-&gt;b</tt> fields.

<blockquote>
<pre>
 1 bool add_gp_buggy(int a, int b)
 2 {
 3   p = kmalloc(sizeof(*p), GFP_KERNEL);
 4   if (!p)
 5     return -ENOMEM;
 6   spin_lock(&amp;gp_lock);
 7   if (rcu_access_pointer(gp)) {
 8     spin_unlock(&amp;gp_lock);
 9     return false;
10   }
11   p-&gt;a = a;
12   p-&gt;b = a;
13   gp = p; /* ORDERING BUG */
14   spin_unlock(&amp;gp_lock);
15   return true;
16 }
</pre>
</blockquote>

<p>
The problem is that both the compiler and weakly ordered CPUs are within
their rights to reorder this code as follows:

<blockquote>
<pre>
 1 bool add_gp_buggy_optimized(int a, int b)
 2 {
 3   p = kmalloc(sizeof(*p), GFP_KERNEL);
 4   if (!p)
 5     return -ENOMEM;
 6   spin_lock(&amp;gp_lock);
 7   if (rcu_access_pointer(gp)) {
 8     spin_unlock(&amp;gp_lock);
 9     return false;
10   }
<b>11   gp = p; /* ORDERING BUG */
12   p-&gt;a = a;
13   p-&gt;b = a;</b>
14   spin_unlock(&amp;gp_lock);
15   return true;
16 }
</pre>
</blockquote>

<p>
If an RCU reader fetches <tt>gp</tt> just after
<tt>add_gp_buggy_optimized</tt> executes line&nbsp;11,
it will see garbage in the <tt>-&gt;a</tt> and <tt>-&gt;b</tt>
fields.
And this is but one of many ways in which compiler and hardware optimizations
could cause trouble.
Therefore, we clearly need some way to prevent the compiler and the CPU from
reordering in this manner, which brings us to the publish-subscribe
guarantee discussed in the next section.

<h3><a name="Publish-Subscribe Guarantee">Publish/Subscribe Guarantee</a></h3>

<p>
RCU's publish-subscribe guarantee allows data to be inserted
into a linked data structure without disrupting RCU readers.
The updater uses <tt>rcu_assign_pointer()</tt> to insert the
new data, and readers use <tt>rcu_dereference()</tt> to
access data, whether new or old.
The following shows an example of insertion:

<blockquote>
<pre>
 1 bool add_gp(int a, int b)
 2 {
 3   p = kmalloc(sizeof(*p), GFP_KERNEL);
 4   if (!p)
 5     return -ENOMEM;
 6   spin_lock(&amp;gp_lock);
 7   if (rcu_access_pointer(gp)) {
 8     spin_unlock(&amp;gp_lock);
 9     return false;
10   }
11   p-&gt;a = a;
12   p-&gt;b = a;
13   rcu_assign_pointer(gp, p);
14   spin_unlock(&amp;gp_lock);
15   return true;
16 }
</pre>
</blockquote>

<p>
The <tt>rcu_assign_pointer()</tt> on line&nbsp;13 is conceptually
equivalent to a simple assignment statement, but also guarantees
that its assignment will
happen after the two assignments in lines&nbsp;11 and&nbsp;12,
similar to the C11 <tt>memory_order_release</tt> store operation.
It also prevents any number of &ldquo;interesting&rdquo; compiler
optimizations, for example, the use of <tt>gp</tt> as a scratch
location immediately preceding the assignment.

<p>@@QQ@@
But <tt>rcu_assign_pointer()</tt> does nothing to prevent the
two assignments to <tt>p-&gt;a</tt> and <tt>p-&gt;b</tt>
from being reordered.
Can't that also cause problems?
<p>@@QQA@@
No, it cannot.
The readers cannot see either of these two fields until
the assignment to <tt>gp</tt>, by which time both fields are
fully initialized.
So reordering the assignments
to <tt>p-&gt;a</tt> and <tt>p-&gt;b</tt> cannot possibly
cause any problems.
<p>@@QQE@@

<p>
It is tempting to assume that the reader need not do anything special
to control its accesses to the RCU-protected data,
as shown in <tt>do_something_gp_buggy()</tt> below:

<blockquote>
<pre>
 1 bool do_something_gp_buggy(void)
 2 {
 3   rcu_read_lock();
 4   p = gp;  /* OPTIMIZATIONS GALORE!!! */
 5   if (p) {
 6     do_something(p-&gt;a, p-&gt;b);
 7     rcu_read_unlock();
 8     return true;
 9   }
10   rcu_read_unlock();
11   return false;
12 }
</pre>
</blockquote>

<p>
However, this temptation must be resisted because there are a
surprisingly large number of ways that the compiler
(to say nothing of
<a href="https://h71000.www7.hp.com/wizard/wiz_2637.html">DEC Alpha CPUs</a>)
can trip this code up.
For but one example, if the compiler were short of registers, it
might choose to refetch from <tt>gp</tt> rather than keeping
a separate copy in <tt>p</tt> as follows:

<blockquote>
<pre>
 1 bool do_something_gp_buggy_optimized(void)
 2 {
 3   rcu_read_lock();
 4   if (gp) { /* OPTIMIZATIONS GALORE!!! */
<b> 5     do_something(gp-&gt;a, gp-&gt;b);</b>
 6     rcu_read_unlock();
 7     return true;
 8   }
 9   rcu_read_unlock();
10   return false;
11 }
</pre>
</blockquote>

<p>
If this function ran concurrently with a series of updates that
replaced the current structure with a new one,
the fetches of <tt>gp-&gt;a</tt>
and <tt>gp-&gt;b</tt> might well come from two different structures,
which could cause serious confusion.
To prevent this (and much else besides), <tt>do_something_gp()</tt> uses
<tt>rcu_dereference()</tt> to fetch from <tt>gp</tt>:

<blockquote>
<pre>
 1 bool do_something_gp(void)
 2 {
 3   rcu_read_lock();
 4   p = rcu_dereference(gp);
 5   if (p) {
 6     do_something(p-&gt;a, p-&gt;b);
 7     rcu_read_unlock();
 8     return true;
 9   }
10   rcu_read_unlock();
11   return false;
12 }
</pre>
</blockquote>

<p>
The <tt>rcu_dereference()</tt> uses volatile casts and (for DEC Alpha)
memory barriers in the Linux kernel.
Should a
<a href="http://www.rdrop.com/users/paulmck/RCU/consume.2015.07.13a.pdf">high-quality implementation of C11 <tt>memory_order_consume</tt> [PDF]</a>
ever appear, then <tt>rcu_dereference()</tt> could be implemented
as a <tt>memory_order_consume</tt> load.
Regardless of the exact implementation, a pointer fetched by
<tt>rcu_dereference()</tt> may not be used outside of the
outermost RCU read-side critical section containing that
<tt>rcu_dereference()</tt>, unless protection of
the corresponding data element has been passed from RCU to some
other synchronization mechanism, most commonly locking or
<a href="https://www.kernel.org/doc/Documentation/RCU/rcuref.txt">reference counting</a>.

<p>
In short, updaters use <tt>rcu_assign_pointer()</tt> and readers
use <tt>rcu_dereference()</tt>, and these two RCU API elements
work together to ensure that readers have a consistent view of
newly added data elements.

<p>
Of course, it is also necessary to remove elements from RCU-protected
data structures, for example, using the following process:

<ol>
<li>	Remove the data element from the enclosing structure.
<li>	Wait for all pre-existing RCU read-side critical sections
	to complete (because only pre-existing readers can possibly have
	a reference to the newly removed data element).
<li>	At this point, only the updater has a reference to the
	newly removed data element, so it can safely reclaim
	the data element, for example, by passing it to <tt>kfree()</tt>.
</ol>

This process is implemented by <tt>remove_gp_synchronous()</tt>:

<blockquote>
<pre>
 1 bool remove_gp_synchronous(void)
 2 {
 3   struct foo *p;
 4
 5   spin_lock(&amp;gp_lock);
 6   p = rcu_access_pointer(gp);
 7   if (!p) {
 8     spin_unlock(&amp;gp_lock);
 9     return false;
10   }
11   rcu_assign_pointer(gp, NULL);
12   spin_unlock(&amp;gp_lock);
13   synchronize_rcu();
14   kfree(p);
15   return true;
16 }
</pre>
</blockquote>

<p>
This function is straightforward, with line&nbsp;13 waiting for a grace
period before line&nbsp;14 frees the old data element.
This waiting ensures that readers will reach line&nbsp;7 of
<tt>do_something_gp()</tt> before the data element referenced by
<tt>p</tt> is freed.
The <tt>rcu_access_pointer()</tt> on line&nbsp;6 is similar to
<tt>rcu_dereference()</tt>, except that:

<ol>
<li>	The value returned by <tt>rcu_access_pointer()</tt>
	cannot be dereferenced.
	If you want to access the value pointed to as well as
	the pointer itself, use <tt>rcu_dereference()</tt>
	instead of <tt>rcu_access_pointer()</tt>.
<li>	The call to <tt>rcu_access_pointer()</tt> need not be
	protected.
	In contrast, <tt>rcu_dereference()</tt> must either be
	within an RCU read-side critical section or in a code
	segment where the pointer cannot change, for example, in
	code protected by the corresponding update-side lock.
</ol>

<p>@@QQ@@
Without the <tt>rcu_dereference()</tt> or the
<tt>rcu_access_pointer()</tt>, what destructive optimizations
might the compiler make use of?
<p>@@QQA@@
Let's start with what happens to <tt>do_something_gp()</tt>
if it fails to use <tt>rcu_dereference()</tt>.
It could reuse a value formerly fetched from this same pointer.
It could also fetch the pointer from <tt>gp</tt> in a byte-at-a-time
manner, resulting in <i>load tearing</i>, in turn resulting a bytewise
mash-up of two distince pointer values.
It might even use value-speculation optimizations, where it makes a wrong
guess, but by the time it gets around to checking the value, an update
has changed the pointer to match the wrong guess.
Too bad about any dereferences that returned pre-initialization garbage
in the meantime!

<p>
For <tt>remove_gp_synchronous()</tt>, as long as all modifications
to <tt>gp</tt> are carried out while holding <tt>gp_lock</tt>,
the above optimizations are harmless.
However,
with <tt>CONFIG_SPARSE_RCU_POINTER=y</tt>,
<tt>sparse</tt> will complain if you
define <tt>gp</tt> with <tt>__rcu</tt> and then
access it without using
either <tt>rcu_access_pointer()</tt> or <tt>rcu_dereference()</tt>.
<p>@@QQE@@

<p>
This simple linked-data-structure scenario clearly demonstrates the need
for RCU's stringent memory-ordering guarantees on systems with more than
one CPU:

<ol>
<li>	Each CPU that has an RCU read-side critical section that
	begins before <tt>synchronize_rcu()</tt> starts is
	guaranteed to execute a full memory barrier between the time
	that the RCU read-side critical section ends and the time that
	<tt>synchronize_rcu()</tt> returns.
	Without this guarantee, a pre-existing RCU read-side critical section
	might hold a reference to the newly removed <tt>struct foo</tt>
	after the <tt>kfree()</tt> on line&nbsp;14 of
	<tt>remove_gp_synchronous()</tt>.
<li>	Each CPU that has an RCU read-side critical section that ends
	after <tt>synchronize_rcu()</tt> returns is guaranteed
	to execute a full memory barrier between the time that
	<tt>synchronize_rcu()</tt> begins and the time that the RCU
	read-side critical section begins.
	Without this guarantee, a later RCU read-side critical section
	running after the <tt>kfree()</tt> on line&nbsp;14 of
	<tt>remove_gp_synchronous()</tt> might
	later run <tt>do_something_gp()</tt> and find the
	newly deleted <tt>struct foo</tt>.
<li>	If the task invoking <tt>synchronize_rcu()</tt> remains
	on a given CPU, then that CPU is guaranteed to execute a full
	memory barrier sometime during the execution of
	<tt>synchronize_rcu()</tt>.
	This guarantee ensures that the <tt>kfree()</tt> on
	line&nbsp;14 of <tt>remove_gp_synchronous()</tt> really does
	execute after the removal on line&nbsp;11.
<li>	If the task invoking <tt>synchronize_rcu()</tt> migrates
	among a group of CPUs during that invocation, then each of the
	CPUs in that group is guaranteed to execute a full memory barrier
	sometime during the execution of <tt>synchronize_rcu()</tt>.
	This guarantee also ensures that the <tt>kfree()</tt> on
	line&nbsp;14 of <tt>remove_gp_synchronous()</tt> really does
	execute after the removal on
	line&nbsp;11, but also in the case where the thread executing the
	<tt>synchronize_rcu()</tt> migrates in the meantime.
</ol>

<p>@@QQ@@
Given that multiple CPUs can start RCU read-side critical sections
at any time without any ordering whatsoever, how can RCU possibly tell whether
or not a given RCU read-side critical section starts before a
given instance of <tt>synchronize_rcu()</tt>?
<p>@@QQA@@
If RCU cannot tell whether or not a given
RCU read-side critical section starts before a
given instance of <tt>synchronize_rcu()</tt>,
then it must assume that the RCU read-side critical section
started first.
In other words, a given instance of <tt>synchronize_rcu()</tt>
can avoid waiting on a given RCU read-side critical section only
if it can prove that <tt>synchronize_rcu()</tt> started first.
<p>@@QQE@@

<p>@@QQ@@
The first and second guarantees require unbelievably strict ordering!
Are all these memory barriers <i> really</i> required?
<p>@@QQA@@
Yes, they really are required.
To see why the first guarantee is required, consider the following
sequence of events:

<ol>
<li>	CPU 1: <tt>rcu_read_lock()</tt>
<li>	CPU 1: <tt>q = rcu_dereference(gp);
	/* Very likely to return p. */</tt>
<li>	CPU 0: <tt>list_del_rcu(p);</tt>
<li>	CPU 0: <tt>synchronize_rcu()</tt> starts.
<li>	CPU 1: <tt>do_something_with(q-&gt;a);
	/* No smp_mb(), so might happen after kfree(). */</tt>
<li>	CPU 1: <tt>rcu_read_unlock()</tt>
<li>	CPU 0: <tt>synchronize_rcu()</tt> returns.
<li>	CPU 0: <tt>kfree(p);</tt>
</ol>

<p>
Therefore, there absolutely must be a full memory barrier between the
end of the RCU read-side critical section and the end of the
grace period.

<p>
The sequence of events demonstrating the necessity of the second rule
is roughly similar:

<ol>
<li>	CPU 0: <tt>list_del_rcu(p);</tt>
<li>	CPU 0: <tt>synchronize_rcu()</tt> starts.
<li>	CPU 1: <tt>rcu_read_lock()</tt>
<li>	CPU 1: <tt>q = rcu_dereference(gp);
	/* Might return p if no memory barrier. */</tt>
<li>	CPU 0: <tt>synchronize_rcu()</tt> returns.
<li>	CPU 0: <tt>kfree(p);</tt>
<li>	CPU 1: <tt>do_something_with(q-&gt;a); /* Boom!!! */</tt>
<li>	CPU 1: <tt>rcu_read_unlock()</tt>
</ol>

<p>
And similarly, without a memory barrier between the beginning of the
grace period and the beginning of the RCU read-side critical section,
CPU&nbsp;1 might end up accessing the freelist.

<p>
The &ldquo;as if&rdquo; rule of course applies, so that any implementation
that acts as if the appropriate memory barriers were in place is a
correct implementation.
That said, it is much easier to fool yourself into believing that you have
adhered to the as-if rule than it is to actually adhere to it!
<p>@@QQE@@

<p>
In short, RCU's publish-subscribe guarantee is provided by the combination
of <tt>rcu_assign_pointer()</tt> and <tt>rcu_dereference()</tt>.
This guarantee allows data elements to be safely added to RCU-protected
linked data structures without disrupting RCU readers.
This guarantee can be used in combination with the grace-period
guarantee to also allow data elements to be removed from RCU-protected
linked data structures, again without disrupting RCU readers.

<p>
This guarantee was only partially premeditated.
DYNIX/ptx used an explicit memory barrier for publication, but had nothing
resembling <tt>rcu_dereference()</tt> for subscription, nor did it
have anything resembling the <tt>smp_read_barrier_depends()</tt>
that was later subsumed into <tt>rcu_dereference()</tt>.
The need for these operations made itself known quite suddenly at a
late-1990s meeting with the DEC Alpha architects, back in the days when
DEC was still a free-standing company.
It took the Alpha architects a good hour to convince me that any sort
of barrier would ever be needed, and it then took me a good <i>two</i> hours
to convince them that their documentation did not make this point clear.
More recent work with the C and C++ standards committees have provided
much education on tricks and traps from the compiler.
In short, compilers were much less tricky in the early 1990s, but in
2015, don't even think about omitting <tt>rcu_dereference()</tt>!

<h3><a name="RCU Primitives Guaranteed to Execute Unconditionally">RCU Primitives Guaranteed to Execute Unconditionally</a></h3>

<p>
The common-case RCU primitives are unconditional.
They are invoked, they do their job, and they return, with no possibility
of error, and no need to retry.
This is a key RCU design philosophy.

<p>
However, this philosophy is pragmatic rather than pigheaded.
If someone comes up with a good justification for a particular conditional
RCU primitive, it might well be implemented and added.
After all, this guarantee was reverse-engineered, not premeditated.
The unconditional nature of the RCU primitives was initially an
accident of implementation, and later experience with synchronization
primitives with conditional primitives caused me to elevate this
accident to a guarantee.
Therefore, the justification for adding a conditional primitive to
RCU would need to be based on detailed and compelling use cases.

<h3><a name="Guaranteed Read-to-Write Upgrade">Guaranteed Read-to-Write Upgrade</a></h3>

<p>
As far as RCU is concerned, it is always possible to carry out an
update within an RCU read-side critical section.
For example, that RCU read-side critical section might search for
a given data element, and then might acquire the update-side
spinlock in order to update that element, all while remaining
in that RCU read-side critical section.
Of course, it is necessary to exit the RCU read-side critical section
before invoking <tt>synchronize_rcu()</tt>, however, this
inconvenience can be avoided through use of the
<tt>call_rcu()</tt> and <tt>kfree_rcu()</tt> API members
described later in this document.

<p>@@QQ@@
But how does the upgrade-to-write operation exclude other readers?
<p>@@QQA@@
It doesn't, just like normal RCU updates, which also do not exclude
RCU readers.
<p>@@QQE@@

<p>
This guarantee allows lookup code to be shared between read-side
and update-side code, and was premeditated, appearing in the earliest
DYNIX/ptx RCU documentation.

<h2><a name="Fundamental Non-Requirements">Fundamental Non-Requirements</a></h2>

<p>
RCU provides extremely lightweight readers, and its read-side guarantees,
though quite useful, are correspondingly lightweight.
It is therefore all too easy to assume that RCU is guaranteeing more
than it really is.
Of course, the list of things that RCU does not guarantee is infinitely
long, however, the following sections list a few non-guarantees that
have caused confusion.
Except where otherwise noted, these non-guarantees were premeditated.

<ol>
<li>	<a href="#Readers Impose Minimal Ordering">
	Readers Impose Minimal Ordering</a>
<li>	<a href="#Readers Do Not Exclude Updaters">
	Readers Do Not Exclude Updaters</a>
<li>	<a href="#Updaters Only Wait For Old Readers">
	Updaters Only Wait For Old Readers</a>
<li>	<a href="#Grace Periods Don't Partition Read-Side Critical Sections">
	Grace Periods Don't Partition Read-Side Critical Sections</a>
<li>	<a href="#Read-Side Critical Sections Don't Partition Grace Periods">
	Read-Side Critical Sections Don't Partition Grace Periods</a>
<li>	<a href="#Disabling Preemption Does Not Block Grace Periods">
	Disabling Preemption Does Not Block Grace Periods</a>
</ol>

<h3><a name="Readers Impose Minimal Ordering">Readers Impose Minimal Ordering</a></h3>

<p>
Reader-side markers such as <tt>rcu_read_lock()</tt> and
<tt>rcu_read_unlock()</tt> provide absolutely no ordering guarantees
except through their interaction with the grace-period APIs such as
<tt>synchronize_rcu()</tt>.
To see this, consider the following pair of threads:

<blockquote>
<pre>
 1 void thread0(void)
 2 {
 3   rcu_read_lock();
 4   WRITE_ONCE(x, 1);
 5   rcu_read_unlock();
 6   rcu_read_lock();
 7   WRITE_ONCE(y, 1);
 8   rcu_read_unlock();
 9 }
10
11 void thread1(void)
12 {
13   rcu_read_lock();
14   r1 = READ_ONCE(y);
15   rcu_read_unlock();
16   rcu_read_lock();
17   r2 = READ_ONCE(x);
18   rcu_read_unlock();
19 }
</pre>
</blockquote>

<p>
After <tt>thread0()</tt> and <tt>thread1()</tt> execute
concurrently, it is quite possible to have

<blockquote>
<pre>
(r1 == 1 &amp;&amp; r2 == 0)
</pre>
</blockquote>

(that is, <tt>y</tt> appears to have been assigned before <tt>x</tt>),
which would not be possible if <tt>rcu_read_lock()</tt> and
<tt>rcu_read_unlock()</tt> had much in the way of ordering
properties.
But they do not, so the CPU is within its rights
to do significant reordering.
This is by design:  Any significant ordering constraints would slow down
these fast-path APIs.

<p>@@QQ@@
Can't the compiler also reorder this code?
<p>@@QQA@@
No, the volatile casts in <tt>READ_ONCE()</tt> and
<tt>WRITE_ONCE()</tt> prevent the compiler from reordering in
this particular case.
<p>@@QQE@@

<h3><a name="Readers Do Not Exclude Updaters">Readers Do Not Exclude Updaters</a></h3>

<p>
Neither <tt>rcu_read_lock()</tt> nor <tt>rcu_read_unlock()</tt>
exclude updates.
All they do is to prevent grace periods from ending.
The following example illustrates this:

<blockquote>
<pre>
 1 void thread0(void)
 2 {
 3   rcu_read_lock();
 4   r1 = READ_ONCE(y);
 5   if (r1) {
 6     do_something_with_nonzero_x();
 7     r2 = READ_ONCE(x);
 8     WARN_ON(!r2); /* BUG!!! */
 9   }
10   rcu_read_unlock();
11 }
12
13 void thread1(void)
14 {
15   spin_lock(&amp;my_lock);
16   WRITE_ONCE(x, 1);
17   WRITE_ONCE(y, 1);
18   spin_unlock(&amp;my_lock);
19 }
</pre>
</blockquote>

<p>
If the <tt>thread0()</tt> function's <tt>rcu_read_lock()</tt>
excluded the <tt>thread1()</tt> function's update,
the <tt>WARN_ON()</tt> could never fire.
But the fact is that <tt>rcu_read_lock()</tt> does not exclude
much of anything aside from subsequent grace periods, of which
<tt>thread1()</tt> has none, so the
<tt>WARN_ON()</tt> can and does fire.

<h3><a name="Updaters Only Wait For Old Readers">Updaters Only Wait For Old Readers</a></h3>

<p>
It might be tempting to assume that after <tt>synchronize_rcu()</tt>
completes, there are no readers executing.
This temptation must be avoided because
new readers can start immediately after <tt>synchronize_rcu()</tt>
starts, and <tt>synchronize_rcu()</tt> is under no
obligation to wait for these new readers.

<p>@@QQ@@
Suppose that synchronize_rcu() did wait until all readers had completed.
Would the updater be able to rely on this?
<p>@@QQA@@
No.
Even if <tt>synchronize_rcu()</tt> were to wait until
all readers had completed, a new reader might start immediately after
<tt>synchronize_rcu()</tt> completed.
Therefore, the code following
<tt>synchronize_rcu()</tt> cannot rely on there being no readers
in any case.
<p>@@QQE@@

<h3><a name="Grace Periods Don't Partition Read-Side Critical Sections">
Grace Periods Don't Partition Read-Side Critical Sections</a></h3>

<p>
It is tempting to assume that if any part of one RCU read-side critical
section precedes a given grace period, and if any part of another RCU
read-side critical section follows that same grace period, then all of
the first RCU read-side critical section must precede all of the second.
However, this just isn't the case: A single grace period does not
partition the set of RCU read-side critical sections.
An example of this situation can be illustrated as follows, where
<tt>x</tt>, <tt>y</tt>, and <tt>z</tt> are initially all zero:

<blockquote>
<pre>
 1 void thread0(void)
 2 {
 3   rcu_read_lock();
 4   WRITE_ONCE(a, 1);
 5   WRITE_ONCE(b, 1);
 6   rcu_read_unlock();
 7 }
 8
 9 void thread1(void)
10 {
11   r1 = READ_ONCE(a);
12   synchronize_rcu();
13   WRITE_ONCE(c, 1);
14 }
15
16 void thread2(void)
17 {
18   rcu_read_lock();
19   r2 = READ_ONCE(b);
20   r3 = READ_ONCE(c);
21   rcu_read_unlock();
22 }
</pre>
</blockquote>

<p>
It turns out that the outcome:

<blockquote>
<pre>
(r1 == 1 &amp;&amp; r2 == 0 &amp;&amp; r3 == 1)
</pre>
</blockquote>

is entirely possible.
The following figure show how this can happen, with each circled
<tt>QS</tt> indicating the point at which RCU recorded a
<i>quiescent state</i> for each thread, that is, a state in which
RCU knows that the thread cannot be in the midst of an RCU read-side
critical section that started before the current grace period:

<p><img src="GPpartitionReaders1.svg" alt="GPpartitionReaders1.svg" width="60%"></p>

<p>
If it is necessary to partition RCU read-side critical sections in this
manner, it is necessary to use two grace periods, where the first
grace period is known to end before the second grace period starts:

<blockquote>
<pre>
 1 void thread0(void)
 2 {
 3   rcu_read_lock();
 4   WRITE_ONCE(a, 1);
 5   WRITE_ONCE(b, 1);
 6   rcu_read_unlock();
 7 }
 8
 9 void thread1(void)
10 {
11   r1 = READ_ONCE(a);
12   synchronize_rcu();
13   WRITE_ONCE(c, 1);
14 }
15
16 void thread2(void)
17 {
18   r2 = READ_ONCE(c);
19   synchronize_rcu();
20   WRITE_ONCE(d, 1);
21 }
22
23 void thread3(void)
24 {
25   rcu_read_lock();
26   r3 = READ_ONCE(b);
27   r4 = READ_ONCE(d);
28   rcu_read_unlock();
29 }
</pre>
</blockquote>

<p>
Here, if <tt>(r1 == 1)</tt>, then
<tt>thread0()</tt>'s write to <tt>b</tt> must happen
before the end of <tt>thread1()</tt>'s grace period.
If in addition <tt>(r4 == 1)</tt>, then
<tt>thread3()</tt>'s read from <tt>b</tt> must happen
after the beginning of <tt>thread2()</tt>'s grace period.
If it is also the case that <tt>(r2 == 1)</tt>, then the
end of <tt>thread1()</tt>'s grace period must precede the
beginning of <tt>thread2()</tt>'s grace period.
This mean that the two RCU read-side critical sections cannot overlap,
guaranteeing that <tt>(r3 == 1)</tt>.
As a result, the outcome:

<blockquote>
<pre>
(r1 == 1 &amp;&amp; r2 == 1 &amp;&amp; r3 == 0 &amp;&amp; r4 == 1)
</pre>
</blockquote>

cannot happen.

<p>
This non-requirement was also non-premeditated, but became apparent
when studying RCU's interaction with memory ordering.

<h3><a name="Read-Side Critical Sections Don't Partition Grace Periods">
Read-Side Critical Sections Don't Partition Grace Periods</a></h3>

<p>
It is also tempting to assume that if an RCU read-side critical section
happens between a pair of grace periods, then those grace periods cannot
overlap.
However, this temptation leads nowhere good, as can be illustrated by
the following, with all variables initially zero:

<blockquote>
<pre>
 1 void thread0(void)
 2 {
 3   rcu_read_lock();
 4   WRITE_ONCE(a, 1);
 5   WRITE_ONCE(b, 1);
 6   rcu_read_unlock();
 7 }
 8
 9 void thread1(void)
10 {
11   r1 = READ_ONCE(a);
12   synchronize_rcu();
13   WRITE_ONCE(c, 1);
14 }
15
16 void thread2(void)
17 {
18   rcu_read_lock();
19   WRITE_ONCE(d, 1);
20   r2 = READ_ONCE(c);
21   rcu_read_unlock();
22 }
23
24 void thread3(void)
25 {
26   r3 = READ_ONCE(d);
27   synchronize_rcu();
28   WRITE_ONCE(e, 1);
29 }
30
31 void thread4(void)
32 {
33   rcu_read_lock();
34   r4 = READ_ONCE(b);
35   r5 = READ_ONCE(e);
36   rcu_read_unlock();
37 }
</pre>
</blockquote>

<p>
In this case, the outcome:

<blockquote>
<pre>
(r1 == 1 &amp;&amp; r2 == 1 &amp;&amp; r3 == 1 &amp;&amp; r4 == 0 &amp&amp; r5 == 1)
</pre>
</blockquote>

is entirely possible, as illustrated below:

<p><img src="ReadersPartitionGP1.svg" alt="ReadersPartitionGP1.svg" width="100%"></p>

<p>
Again, an RCU read-side critical section can overlap almost all of a
given grace period, just so long as it does not overlap the entire
grace period.
As a result, an RCU read-side critical section cannot partition a pair
of RCU grace periods.

<p>@@QQ@@
How long a sequence of grace periods, each separated by an RCU read-side
critical section, would be required to partition the RCU read-side
critical sections at the beginning and end of the chain?
<p>@@QQA@@
In theory, an infinite number.
In practice, an unknown number that is sensitive to both implementation
details and timing considerations.
Therefore, even in practice, RCU users must abide by the theoretical rather
than the practical answer.
<p>@@QQE@@

<h3><a name="Disabling Preemption Does Not Block Grace Periods">
Disabling Preemption Does Not Block Grace Periods</a></h3>

<p>
There was a time when disabling preemption on any given CPU would block
subsequent grace periods.
However, this was an accident of implementation and is not a requirement.
And in the current Linux-kernel implementation, disabling preemption
on a given CPU in fact does not block grace periods, as Oleg Nesterov
<a href="https://lkml.kernel.org/g/20150614193825.GA19582@redhat.com">demonstrated</a>.

<p>
If you need a preempt-disable region to block grace periods, you need to add
<tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>, for example
as follows:

<blockquote>
<pre>
 1 preempt_disable();
 2 rcu_read_lock();
 3 do_something();
 4 rcu_read_unlock();
 5 preempt_enable();
 6
 7 /* Spinlocks implicitly disable preemption. */
 8 spin_lock(&amp;mylock);
 9 rcu_read_lock();
10 do_something();
11 rcu_read_unlock();
12 spin_unlock(&amp;mylock);
</pre>
</blockquote>

<p>
In theory, you could enter the RCU read-side critical section first,
but it is more efficient to keep the entire RCU read-side critical
section contained in the preempt-disable region as shown above.
Of course, RCU read-side critical sections that extend outside of
preempt-disable regions will work correctly, but such critical sections
can be preempted, which forces <tt>rcu_read_unlock()</tt> to do
more work.
And no, this is <i>not</i> an invitation to enclose all of your RCU
read-side critical sections within preempt-disable regions, because
doing so would degrade real-time response.

<p>
This non-requirement appeared with preemptible RCU.
If you need a grace period that waits on non-preemptible code regions, use
<a href="#Sched Flavor">RCU-sched</a>.

<h2><a name="Parallelism Facts of Life">Parallelism Facts of Life</a></h2>

<p>
These parallelism facts of life are by no means specific to RCU, but
the RCU implementation must abide by them.
They therefore bear repeating:

<ol>
<li>	Any CPU or task may be delayed at any time,
	and any attempts to avoid these delays by disabling
	preemption, interrupts, or whatever are completely futile.
	This is most obvious in preemptible user-level
	environments and in virtualized environments (where
	a given guest OS's VCPUs can be preempted at any time by
	the underlying hypervisor), but can also happen in bare-metal
	environments due to ECC errors, NMIs, and other hardware
	events.
	Although a delay of more than about 20 seconds can result
	in splats, the RCU implementation is obligated to use
	algorithms that can tolerate extremely long delays, but where
	&ldquo;extremely long&rdquo; is not long enough to allow
	wrap-around when incrementing a 64-bit counter.
<li>	Both the compiler and the CPU can reorder memory accesses.
	Where it matters, RCU must use compiler directives and
	memory-barrier instructions to preserve ordering.
<li>	Conflicting writes to memory locations in any given cache line
	will result in expensive cache misses.
	Greater numbers of concurrent writes and more-frequent
	concurrent writes will result in more dramatic slowdowns.
	RCU is therefore obligated to use algorithms that have
	sufficient locality to avoid significant performance and
	scalability problems.
<li>	As a rough rule of thumb, only one CPU's worth of processing
	may be carried out under the protection of any given exclusive
	lock.
	RCU must therefore use scalable locking designs.
<li>	Counters are finite, especially on 32-bit systems.
	RCU's use of counters must therefore tolerate counter wrap,
	or be designed such that counter wrap would take way more
	time than a single system is likely to run.
	An uptime of ten years is quite possible, a runtime
	of a century much less so.
	As an example of the latter, RCU's dyntick-idle nesting counter
	allows 54 bits for interrupt nesting level (this counter
	is 64 bits even on a 32-bit system).
	Overflowing this counter requires 2<sup>54</sup>
	half-interrupts on a given CPU without that CPU ever going idle.
	If a half-interrupt happened every microsecond, it would take
	570 years of runtime to overflow this counter, which is currently
	believed to be an acceptably long time.
<li>	Linux systems can have thousands of CPUs running a single
	Linux kernel in a single shared-memory environment.
	RCU must therefore pay close attention to high-end scalability.
</ol>

<p>
This last parallelism fact of life means that RCU must pay special
attention to the preceding facts of life.
The idea that Linux might scale to systems with thousands of CPUs would
have been met with some skepticism in the 1990s, but these requirements
would have otherwise have been unsurprising, even in the early 1990s.

<h2><a name="Quality-of-Implementation Requirements">Quality-of-Implementation Requirements</a></h2>

<p>
These sections list quality-of-implementation requirements.
Although an RCU implementation that ignores these requirements could
still be used, it would likely be subject to limitations that would
make it inappropriate for industrial-strength production use.
Classes of quality-of-implementation requirements are as follows:

<ol>
<li>	<a href="#Specialization">Specialization</a>
<li>	<a href="#Performance and Scalability">Performance and Scalability</a>
<li>	<a href="#Composability">Composability</a>
<li>	<a href="#Corner Cases">Corner Cases</a>
</ol>

<p>
These classes is covered in the following sections.

<h3><a name="Specialization">Specialization</a></h3>

<p>
RCU is and always has been intended primarily for read-mostly situations, as
illustrated by the following figure.
This means that RCU's read-side primitives are optimized, often at the
expense of its update-side primitives.

<p><img src="RCUApplicability.svg" alt="RCUApplicability.svg" width="70%"></p>

<p>
This focus on read-mostly situations means that RCU must interoperate
with other synchronization primitives.
For example, the <tt>add_gp()</tt> and <tt>remove_gp_synchronous()</tt>
examples discussed earlier use RCU to protect readers and locking to
coordinate updaters.
However, the need extends much farther, requiring that a variety of
synchronization primitives be legal within RCU read-side critical sections,
including spinlocks, sequence locks, atomic operations, reference
counters, and memory barriers.

<p>@@QQ@@
What about sleeping locks?
<p>@@QQA@@
These are forbidden within Linux-kernel RCU read-side critical sections
because it is not legal to place a quiescent state (in this case,
voluntary context switch) within an RCU read-side critical section.
However, sleeping locks may be used within userspace RCU read-side critical
sections, and also within Linux-kernel sleepable RCU
<a href="#Sleepable RCU">(SRCU)</a>
read-side critical sections.
In addition, the -rt patchset turns spinlocks into a sleeping locks so
that the corresponding critical sections can be preempted, which
also means that these sleeplockified spinlocks (but not other sleeping locks!)
may be acquire within -rt-Linux-kernel RCU read-side critical sections.

<p>
Note that it <i>is</i> legal for a normal RCU read-side critical section
to conditionally acquire a sleeping locks (as in <tt>mutex_trylock()</tt>),
but only as long as it does not loop indefinitely attempting to
conditionally acquire that sleeping locks.
The key point is that things like <tt>mutex_trylock()</tt>
either return with the mutex held, or return an error indication if
the mutex was not immediately available.
Either way, <tt>mutex_trylock()</tt> returns immediately without sleeping.
<p>@@QQE@@

<p>
It often comes as a surprise that many algorithms do not require a
consistent view of data, but many can function in that mode,
with network routing being the poster child.
Internet routing algorithms take significant time to propagate
updates, so that by the time an update arrives at a given system,
that system has been sending network traffic the wrong way for
a considerable length of time.
Having a few threads continue to send traffic the wrong way for a
few more milliseconds is clearly not a problem:  In the worst case,
TCP retransmissions will eventually get the data where it needs to go.
In general, when tracking the state of the universe outside of the
computer, some level of inconsistency must be tolerated due to
speed-of-light delays if nothing else.

<p>
Furthermore, uncertainty about external state is inherent in many cases.
For example, a pair of veternarians might use heartbeat to determine
whether or not a given cat was alive.
But how long should they wait after the last heartbeat to decide that
the cat is in fact dead?
Waiting less than 400 milliseconds makes no sense because this would
mean that a relaxed cat would be considered to cycle between death
and life more than 100 times per minute.
Moreover, just as with human beings, a cat's heart might stop for
some period of time, so the exact wait period is a judgment call.
One of our pair of veternarians might wait 30 seconds before pronouncing
the cat dead, while the other might insist on waiting a full minute.
The two veternarians would then disagree on the state of the cat during
the final 30 seconds of the minute following the last heartbeat, as
fancifully illustrated below:

<p><img src="2013-08-is-it-dead.png" alt="2013-08-is-it-dead.png" width="431"></p>

<p>
Interestingly enough, this same situation applies to hardware.
When push comes to shove, how do we tell whether or not some
external server has failed?
We send messages to it periodically, and declare it failed if we
don't receive a response within a given period of time.
Policy decisions can usually tolerate short
periods of inconsistency.
The policy was decided some time ago, and is only now being put into
effect, so a few milliseconds of delay is normally inconsequential.

<p>
However, there are algorithms that absolutely must see consistent data.
For example, the translation between a user-level SystemV semaphore
ID to the corresponding in-kernel data structure is protected by RCU,
but it is absolutely forbidden to update a semaphore that has just been
removed.
In the Linux kernel, this need for consistency is accommodated by acquiring
spinlocks located in the in-kernel data structure from within
the RCU read-side critical section, and this is indicated by the
green box in the figure above.
Many other techniques may be used, and are in fact used within the
Linux kernel.

<p>
In short, RCU is not required to maintain consistency, and other
mechanisms may be used in concert with RCU when consistency is required.
RCU's specialization allows it to do its job extremely well, and its
ability to interoperate with other synchronization mechanisms allows
the right mix of synchronization tools to be used for a given job.

<h3><a name="Performance and Scalability">Performance and Scalability</a></h3>

<p>
Energy efficiency is a critical component of performance today,
and Linux-kernel RCU implementations must therefore avoid unnecessarily
awakening idle CPUs.
I cannot claim that this requirement was premeditated.
In fact, I learned of it during a telephone conversation in which I
was given &ldquo;frank and open&rdquo; feedback on the importance
of energy efficiency in battery-powered systems and on specific
energy-efficiency shortcomings of the Linux-kernel RCU implementation.
In my experience, the battery-powered embedded community will consider
any unnecessary wakeups to be extremely unfriendly acts.
So much so that mere Linux-kernel-mailing-list posts are
insufficient to vent their ire.

<p>
Memory consumption is not particularly important for in most
situations, and has become decreasingly
so as memory sizes have expanded and memory
costs have plummeted.
However, as I learned from Matt Mackall's
<a href="http://elinux.org/Linux_Tiny-FAQ">bloatwatch</a>
efforts, memory footprint is critically important on single-CPU systems with
non-preemptible (<tt>CONFIG_PREEMPT=n</tt>) kernels, and thus
<a href="https://lkml.kernel.org/g/20090113221724.GA15307@linux.vnet.ibm.com">tiny RCU</a>
was born.
Josh Triplett has since taken over the small-memory banner with his
<a href="https://tiny.wiki.kernel.org/">Linux kernel tinification</a>
project, which resulted in
<a href="#Sleepable RCU">SRCU</a>
becoming optional for those kernels not needing it.

<p>
The remaining performance requirements are, for the most part,
unsurprising.
For example, in keeping with RCU's read-side specialization,
<tt>rcu_dereference()</tt> should have negligible overhead (for
example, suppression of a few minor compiler optimizations).
Similarly, in non-preemptible environments, <tt>rcu_read_lock()</tt> and
<tt>rcu_read_unlock()</tt> should have exactly zero overhead.

<p>
In preemptible environments, in the case where the RCU read-side
critical section was not preempted (as will be the case for the
highest-priority real-time process), <tt>rcu_read_lock()</tt> and
<tt>rcu_read_unlock()</tt> should have minimal overhead.
In particular, they should not contain atomic read-modify-write
operations, memory-barrier instructions, preemption disabling,
interrupt disabling, or backwards branches.
However, in the case where the RCU read-side critical section was preempted,
<tt>rcu_read_unlock()</tt> may acquire spinlocks and disable interrupts.
This is why it is better to nest an RCU read-side critical section
within a preempt-disable region than vice versa, at least in cases
where that critical section is short enough to avoid unduly degrading
real-time latencies.

<p>
The <tt>synchronize_rcu()</tt> grace-period-wait primitive is
optimized for throughput.
It may therefore incur several milliseconds of latency in addition to
the duration of the longest RCU read-side critical section.
On the other hand, multiple concurrent invocations of
<tt>synchronize_rcu()</tt> are required to use batching optimizations
so that they can be satisfied by a single underlying grace-period-wait
operation.
For example, in the Linux kernel, it is not unusual for a single
grace-period-wait operation to serve more than
<a href="https://www.usenix.org/conference/2004-usenix-annual-technical-conference/making-rcu-safe-deep-sub-millisecond-response">1,000 separate invocations</a>
of <tt>synchronize_rcu()</tt>, thus amortizing the per-invocation
overhead down to nearly zero.
However, the grace-period optimization is also required to avoid
measurable degradation of real-time scheduling and interrupt latencies.

<p>
In some cases, the multi-millisecond <tt>synchronize_rcu()</tt>
latencies are unacceptable.
In these cases, <tt>synchronize_rcu_expedited()</tt> may be used
instead, reducing the grace-period latency down to a few tens of
microseconds on small systems, at least in cases where the RCU read-side
critical sections are short.
There are currently no special latency requirements for
<tt>synchronize_rcu_expedited()</tt> on large systems, but,
consistent with the empirical nature of the RCU specification,
that is subject to change.
However, there most definitely are scalability requirements:
A storm of <tt>synchronize_rcu_expedited()</tt> invocations on 4096
CPUs should at least make reasonable forward progress.
In return for its shorter latencies, <tt>synchronize_rcu_expedited()</tt>
is permitted to impose modest degradation of real-time latency
on non-idle online CPUs.
That said, it will likely be necessary to take further steps to reduce this
degradation, hopefully to roughly that of a scheduling-clock interrupt.

<p>
There are a number of situations where even
<tt>synchronize_rcu_expedited()</tt>'s reduced grace-period
latency is unacceptable.
In these situations, the asynchronous <tt>call_rcu()</tt> can be
used in place of <tt>synchronize_rcu()</tt> as follows:

<blockquote>
<pre>
 1 struct foo {
 2   int a;
 3   int b;
 4   struct rcu_head rh;
 5 };
 6
 7 static void remove_gp_cb(struct rcu_head *rhp)
 8 {
 9   struct foo *p = container_of(rhp, struct foo, rh);
10
11   kfree(p);
12 }
13
14 bool remove_gp_asynchronous(void)
15 {
16   struct foo *p;
17
18   spin_lock(&amp;gp_lock);
19   p = rcu_dereference(gp);
20   if (!p) {
21     spin_unlock(&amp;gp_lock);
22     return false;
23   }
24   rcu_assign_pointer(gp, NULL);
25   call_rcu(&amp;p-&gt;rh, remove_gp_cb);
26   spin_unlock(&amp;gp_lock);
27   return true;
28 }
</pre>
</blockquote>

<p>
A definition of <tt>struct foo</tt> is finally needed, and appears
on lines&nbsp;1-5.
The function <tt>remove_gp_cb()</tt> is passed to <tt>call_rcu()</tt>
on line&nbsp;25, and will be invoked after the end of a subsequent
grace period.
This gets the same effect as <tt>remove_gp_synchronous()</tt>,
but without forcing the updater to wait for a grace period to elapse.
The <tt>call_rcu()</tt> function may be used in a number of
situations where neither <tt>synchronize_rcu()</tt> nor
<tt>synchronize_rcu_expedited()</tt> would be legal,
including within preempt-disable code, <tt>local_bh_disable()</tt> code,
interrupt-disable code, and interrupt handlers.
However, even <tt>call_rcu()</tt> is illegal within NMI handlers.
The callback function (<tt>remove_gp_cb()</tt> in this case) will be
executed within softirq (software interrupt) environment within the
Linux kernel,
either within a real softirq handler or under the protection
of <tt>local_bh_disable()</tt>.
In both the Linux kernel and in userspace, it is bad practice to
write an RCU callback function that takes too long.
Long-running operations should be relegated to separate threads or
(in the Linux kernel) workqueues.

<p>@@QQ@@
Why does line&nbsp;19 use <tt>rcu_access_pointer()</tt>?
After all, <tt>call_rcu()</tt> on line&nbsp;25 stores into the
structure, which would interact badly with concurrent insertions.
Doesn't this mean that <tt>rcu_dereference()</tt> is required?
<p>@@QQA@@
Presumably the <tt>-&gt;gp_lock</tt> acquired on line&nbsp;18 excludes
any changes, including any insertions that <tt>rcu_dereference()</tt>
would protect against.
Therefore, any insertions will be delayed until after <tt>-&gt;gp_lock</tt>
is released on line&nbsp;25, which in turn means that
<tt>rcu_access_pointer()</tt> suffices.
<p>@@QQE@@

<p>
However, all that <tt>remove_gp_cb()</tt> is doing is
invoking <tt>kfree()</tt> on the data element.
This is a common idiom, and is supported by <tt>kfree_rcu()</tt>,
which allows &ldquo;fire and forget&rdquo; operation as shown below:

<blockquote>
<pre>
 1 struct foo {
 2   int a;
 3   int b;
 4   struct rcu_head rh;
 5 };
 6
 7 bool remove_gp_faf(void)
 8 {
 9   struct foo *p;
10
11   spin_lock(&amp;gp_lock);
12   p = rcu_dereference(gp);
13   if (!p) {
14     spin_unlock(&amp;gp_lock);
15     return false;
16   }
17   rcu_assign_pointer(gp, NULL);
18   kfree_rcu(p, rh);
19   spin_unlock(&amp;gp_lock);
20   return true;
21 }
</pre>
</blockquote>

<p>
Note that <tt>remove_gp_faf()</tt> simply invokes
<tt>kfree_rcu()</tt> and proceeds, without any need to pay any
further attention to the subsequent grace period and <tt>kfree()</tt>.
It is permissible to invoke <tt>kfree_rcu()</tt> from the same
environments as for <tt>call_rcu()</tt>.
Interestingly enough, DYNIX/ptx had the equivalents of
<tt>call_rcu()</tt> and <tt>kfree_rcu()</tt>, but not
<tt>synchronize_rcu()</tt>.
This was due to the fact that RCU was not heavily used within DYNIX/ptx,
so the very few places that needed something like
<tt>synchronize_rcu()</tt> simply open-coded it.

<p>@@QQ@@
Earlier it was claimed that <tt>call_rcu()</tt> and
<tt>kfree_rcu()</tt> allowed updaters to avoid being blocked
by readers.
But how can that be correct, given that the invocation of the callback
and the freeing of the memory (respectively) must still wait for
a grace period to elapse?
<p>@@QQA@@
We could define things this way, but keep in mind that this sort of
definition would say that updates in garbage-collected languages
cannot complete until the next time the garbage collector runs,
which does not seem at all reasonable.
The key point is that in most cases, an updater using either
<tt>call_rcu()</tt> or <tt>kfree_rcu()</tt> can proceed to the
next update as soon as it has invoked <tt>call_rcu()</tt> or
<tt>kfree_rcu()</tt>, without having to wait for a subsequent
grace period.
<p>@@QQE@@

<p>
But what if the updater must wait for the completion of code to be
executed after the end of the grace period, but has other tasks
that can be carried out in the meantime?
The polling-style <tt>get_state_synchronize_rcu()</tt> and
<tt>cond_synchronize_rcu()</tt> functions may be used for this
purpose, as shown below:

<blockquote>
<pre>
 1 bool remove_gp_poll(void)
 2 {
 3   struct foo *p;
 4   unsigned long s;
 5
 6   spin_lock(&amp;gp_lock);
 7   p = rcu_access_pointer(gp);
 8   if (!p) {
 9     spin_unlock(&amp;gp_lock);
10     return false;
11   }
12   rcu_assign_pointer(gp, NULL);
13   spin_unlock(&amp;gp_lock);
14   s = get_state_synchronize_rcu();
15   do_something_while_waiting();
16   cond_synchronize_rcu(s);
17   kfree(p);
18   return true;
19 }
</pre>
</blockquote>

<p>
On line&nbsp;14, <tt>get_state_synchronize_rcu()</tt> obtains a
&ldquo;cookie&rdquo; from RCU,
then line&nbsp;15 carries out other tasks,
and finally, line&nbsp;16 returns immediately if a grace period has
elapsed in the meantime, but otherwise waits as required.
The need for <tt>get_state_synchronize_rcu</tt> and
<tt>cond_synchronize_rcu()</tt> has appeared quite recently,
so it is too early to tell whether they will stand the test of time.

<p>
RCU thus provides a range of tools to allow updaters to strike the
required tradeoff between latency, flexibility and CPU overhead.

<h3><a name="Composability">Composability</a></h3>

<p>
Composability has received much attention in recent years, perhaps in part
due to the collision of multicore hardware with object-oriented techniques
designed in single-threaded environments for single-threaded use.
And in theory, RCU read-side critical sections may be composed, and in
fact may be nested arbitrarily deeply.
In practice, as with all real-world implementations of composable
constructs, there are limitations.

<p>
Implementations of RCU for which <tt>rcu_read_lock()</tt>
and <tt>rcu_read_unlock()</tt> generate no code, such as
Linux-kernel RCU when <tt>CONFIG_PREEMPT=n</tt>, can be
nested arbitrarily deeply.
After all, there is no overhead.
Except that if all these instances of <tt>rcu_read_lock()</tt>
and <tt>rcu_read_unlock()</tt> are visible to the compiler,
compilation will eventually fail due to exhausting memory,
mass storage, or user patience, whichever comes first.
If the nesting is not visible to the compiler, as is the case with
mutually recursive functions each in its own translation unit,
stack overflow will result.
If the nesting takes the form of loops, either the control variable
will overflow or (in the Linux kernel) you will get an RCU CPU stall warning.
Nevertheless, this class of RCU implementations is one
of the most composable constructs in existence.

<p>
RCU implementations that explicitly track nesting depth
are limited by the nesting-depth counter.
For example, the Linux kernel's preemptible RCU limits nesting to
<tt>INT_MAX</tt>.
This should suffice for almost all practical purposes.
That said, a consecutive pair of RCU read-side critical sections
between which there is an operation that waits for a grace period
cannot be enclosed in another RCU read-side critical section.
This is because it is not legal to wait for a grace period within
an RCU read-side critical section:  To do so would result either
in deadlock or
in RCU implicitly splitting the enclosing RCU read-side critical
section, neither of which is conducive to a long-lived and prosperous
kernel.

<p>
It is worth noting that RCU is not alone in limiting composability.
For example, many transactional-memory implementations prohibit
composing a pair of transactions separated by an irrevocable
operation (for example, a network receive operation).
For another example, lock-based critical sections can be composed
surprisingly freely, but only if deadlock is avoided.

<p>
In short, although RCU read-side critical sections are highly composable,
care is required in some situations, just as is the case for any other
composable synchronization mechanism.

<h3><a name="Corner Cases">Corner Cases</a></h3>

<p>
A given RCU workload might have an endless and intense stream of
RCU read-side critical sections, perhaps even so intense that there
was never a point in time during which there was not at least one
RCU read-side critical section in flight.
RCU cannot allow this situation to block grace periods:  As long as
all the RCU read-side critical sections are finite, grace periods
must also be finite.

<p>
That said, preemptible RCU implementations could potentially result
in RCU read-side critical sections being preempted for long durations,
which has the effect of creating a long-duration RCU read-side
critical section.
This situation can arise only in heavily loaded systems, but systems using
real-time priorities are of course more vulnerable.
Therefore, RCU priority boosting is provided to help deal with this
case.
That said, the exact requirements on RCU priority boosting will likely
evolve as more experience accumulates.

<p>
Other workloads might have very high update rates.
Although one can argue that such workloads should instead use
something other than RCU, the fact remains that RCU must
handle such workloads gracefully.
This requirement is another factor driving batching of grace periods,
but it is also the driving force behind the checks for large numbers
of queued RCU callbacks in the <tt>call_rcu()</tt> code path.
Finally, high update rates should not delay RCU read-side critical
sections, although some read-side delays can occur when using
<tt>synchronize_rcu_expedited()</tt>, courtesy of this function's use
of <tt>try_stop_cpus()</tt>.
(In the future, <tt>synchronize_rcu_expedited()</tt> will be
converted to use lighter-weight inter-processor interrupts (IPIs),
but this will still disturb readers, though to a much smaller degree.)

<p>
Although all three of these corner cases were understood in the early
1990s, a simple user-level test consisting of <tt>close(open(path))</tt>
in a tight loop
in the early 2000s suddenly provided a much deeper appreciation of the
high-update-rate corner case.
This test also motivated addition of some RCU code to react to high update
rates, for example, if a given CPU finds itself with more than 10,000
RCU callbacks queued, it will cause RCU to take evasive action by
more aggressively starting grace periods and more aggressively forcing
completion of grace-period processing.
This evasive action causes the grace period to complete more quickly,
but at the cost of restricting RCU's batching optimizations, thus
increasing the CPU overhead incurred by that grace period.

<h2><a name="Software-Engineering Requirements">
Software-Engineering Requirements</a></h2>

<p>
Between Murphy's Law and &ldquo;To err is human&rdquo;, it is necessary to
guard against mishaps and misuse:

<ol>
<li>	It is all too easy to forget to use <tt>rcu_read_lock()</tt>
	everywhere that it is needed, so kernels built with
	<tt>CONFIG_PROVE_RCU=y</tt> will spat if
	<tt>rcu_dereference()</tt> is used outside of an
	RCU read-side critical section.
	Update-side code can use <tt>rcu_dereference_protected()</tt>,
	which takes a
	<a href="https://lwn.net/Articles/371986/">lockdep expression</a>
	to indicate what is providing the protection.
	If the indicated protection is not provided, a lockdep splat
	is emitted.

	<p>
	Code shared between readers and updaters can use
	<tt>rcu_dereference_check()</tt>, which also takes a
	lockdep expression, and emits a lockdep splat if neither
	<tt>rcu_read_lock()</tt> nor the indicated protection
	is in place.
	In addition, <tt>rcu_dereference_raw()</tt> is used in those
	(hopefully rare) cases where the required protection cannot
	be easily described.
	Finally, <tt>rcu_read_lock_held()</tt> is provided to
	allow a function to verify that it has been invoked within
	an RCU read-side critical section.
	I was made aware of this set of requirements shortly after Thomas
	Gleixner audited a number of RCU uses.
<li>	A given function might wish to check for RCU-related preconditions
	upon entry, before using any other RCU API.
	The <tt>rcu_lockdep_assert()</tt> does this job,
	asserting the expression in kernels having lockdep enabled
	and doing nothing otherwise.
<li>	It is also easy to forget to use <tt>rcu_assign_pointer()</tt>
	and <tt>rcu_dereference()</tt>, perhaps (incorrectly)
	substituting a simple assignment.
	To catch this sort of error, a given RCU-protected pointer may be
	tagged with <tt>__rcu</tt>, after which running sparse
	with <tt>CONFIG_SPARSE_RCU_POINTER=y</tt> will complain
	about simple-assignment accesses to that pointer.
	Arnd Bergmann made me aware of this requirement, and also
	supplied the needed
	<a href="https://lwn.net/Articles/376011/">patch series</a>.
<li>	Kernels built with <tt>CONFIG_DEBUG_OBJECTS_RCU_HEAD=y</tt>
	will splat if a data element is passed to <tt>call_rcu()</tt>
	twice in a row, without a grace period in between.
	(This error is similar to a double free.)
	The corresponding <tt>rcu_head</tt> structures that are
	dynamically allocated are automatically tracked, but
	<tt>rcu_head</tt> structures allocated on the stack
	must be initialized with <tt>init_rcu_head_on_stack()</tt>
	and cleaned up with <tt>destroy_rcu_head_on_stack()</tt>.
	Similarly, statically allocated non-stack <tt>rcu_head</tt>
	structures must be initialized with <tt>init_rcu_head()</tt>
	and cleaned up with <tt>destroy_rcu_head()</tt>.
	Mathieu Desnoyers made me aware of this requirement, and also
	supplied the needed
	<a href="https://lkml.kernel.org/g/20100319013024.GA28456@Krystal">patch</a>.
<li>	An infinite loop in an RCU read-side critical section will
	eventually trigger an RCU CPU stall warning splat, with
	the duration of &ldquo;eventually&rdquo; being controlled by the
	<tt>RCU_CPU_STALL_TIMEOUT</tt> <tt>Kconfig</tt> option, or,
	alternatively, by the
	<tt>rcupdate.rcu_cpu_stall_timeout</tt> boot/sysfs
	parameter.
	However, RCU is not obligated to produce this splat
	unless there is a grace period waiting on that particular
	RCU read-side critical section.
	<p>
	Some extreme workloads might intentionally delay
	RCU grace periods, and systems running those workloads can
	be booted with <tt>rcupdate.rcu_cpu_stall_suppress</tt>
	to suppress the splats.
	This kernel parameter may also be set via <tt>sysfs</tt>.
	Furthermore, RCU CPU stall warnings are counter-productive
	during sysrq dumps and during panics.
	RCU therefore supplies the <tt>rcu_sysrq_start()</tt> and
	<tt>rcu_sysrq_end()</tt> API members to be called before
	and after long sysrq dumps.
	RCU also supplies the <tt>rcu_panic()</tt> notifier that is
	automatically invoked at the beginning of a panic to suppress
	further RCU CPU stall warnings.

	<p>
	This requirement made itself known in the early 1990s, pretty
	much the first time that it was necessary to debug a CPU stall.
	That said, the initial implementation in DYNIX/ptx was quite
	generic in comparison with that of Linux.
<li>	Although it would be very good to detect pointers leaking out
	of RCU read-side critical sections, there is currently no
	good way of doing this.
	One complication is the need to distinguish between pointers
	leaking and pointers that have been handed off from RCU to
	some other synchronization mechanism, for example, reference
	counting.
<li>	In kernels built with <tt>CONFIG_RCU_TRACE=y</tt>, RCU-related
	information is provided via both debugfs and event tracing.
<li>	Open-coded use of <tt>rcu_assign_pointer()</tt> and
	<tt>rcu_dereference()</tt> to create typical linked
	data structures can be surprisingly error-prone.
	Therefore, RCU-protected
	<a href="https://lwn.net/Articles/609973/#RCU List APIs">linked lists</a>
	and, more recently, RCU-protected
	<a href="https://lwn.net/Articles/612100/">hash tables</a>
	are available.
	Many other special-purpose RCU-protected data structures are
	available in the Linux kernel and the userspace RCU library.
<li>	Some linked structures are created at compile time, but still
	require <tt>__rcu</tt> checking.
	The <tt>RCU_POINTER_INITIALIZER()</tt> macro serves this
	purpose.
<li>	It is not necessary to use <tt>rcu_assign_pointer()</tt>
	when creating linked structures that are to be published via
	a single external pointer.
	The <tt>RCU_INIT_POINTER()</tt> macro is provided for
	this task and also for assigning <tt>NULL</tt> pointers
	at runtime.
</ol>

<p>
This not a hard-and-fast list:  RCU's diagnostic capabilities will
continue to be guided by the number and type of usage bugs found
in real-world RCU usage.

<h2><a name="Linux Kernel Complications">Linux Kernel Complications</a></h2>

<p>
The Linux kernel provides an interesting environment for all kinds of
software, including RCU.
Some of the relevant points of interest are as follows:

<ol>
<li>	<a href="#Configuration">Configuration</a>.
<li>	<a href="#Firmware Interface">Firmware Interface</a>.
<li>	<a href="#Early Boot">Early Boot</a>.
<li>	<a href="#Interrupts and NMIs">
	Interrupts and non-maskable interrupts (NMIs)</a>.
<li>	<a href="#Loadable Modules">Loadable Modules</a>.
<li>	<a href="#Hotplug CPU">Hotplug CPU</a>.
<li>	<a href="#Scheduler and RCU">Scheduler and RCU</a>.
<li>	<a href="#Tracing and RCU">Tracing and RCU</a>.
<li>	<a href="#Energy Efficiency">Energy Efficiency</a>.
<li>	<a href="#Memory Efficiency">Memory Efficiency</a>.
<li>	<a href="#Performance, Scalability, Response Time, and Reliability">
	Performance, Scalability, Response Time, and Reliability</a>.
</ol>

<p>
This list is probably incomplete, but it does give a feel for the
most notable Linux-kernel complications.
Each of the following sections covers one of the above topics.

<h3><a name="Configuration">Configuration</a></h3>

<p>
RCU's goal is automatic configuration, so that almost nobody
needs to worry about RCU's <tt>Kconfig</tt> options.
And for almost all users, RCU does in fact work well
&ldquo;out of the box.&rdquo;

<p>
However, there are specialized use cases that are handled by
kernel boot parameters and <tt>Kconfig</tt> options.
Unfortunately, the <tt>Kconfig</tt> system will explicitly ask users
about new <tt>Kconfig</tt> options, which requires almost all of them
be hidden behind a <tt>CONFIG_RCU_EXPERT</tt> <tt>Kconfig</tt> option.

<p>
This all should be quite obvious, but the fact remains that
Linus Torvalds recently had to
<a href="https://lkml.kernel.org/g/CA+55aFy4wcCwaL4okTs8wXhGZ5h-ibecy_Meg9C4MNQrUnwMcg@mail.gmail.com">remind</a>
me of this requirement.

<h3><a name="Firmware Interface">Firmware Interface</a></h3>

<p>
In many cases, kernel obtains information about the system from the
firmware, and sometimes things are lost in translation.
Or the translation is accurate, but the original message is bogus.

<p>
For example, some systems' firmware overreports the number of CPUs,
sometimes by a large factor.
If RCU naively believed the firmware, as it used to do,
it would create too many per-CPU kthreads.
Although the resulting system will still run correctly, the extra
kthreads needlessly consume memory and can cause confusion
when they show up in <tt>ps</tt> listings.

<p>
RCU must therefore wait for a given CPU to actually come online before
it can allow itself to believe that the CPU actually exists.
The resulting &ldquo;ghost CPUs&rdquo; (which are never going to
come online) cause a number of
<a href="https://paulmck.livejournal.com/37494.html">interesting complications</a>.

<h3><a name="Early Boot">Early Boot</a></h3>

<p>
The Linux kernel's boot sequence is an interesting process,
and RCU is used early, even before <tt>rcu_init()</tt>
is invoked.
In fact, a number of RCU's primitives can be used as soon as the
initial task's <tt>task_struct</tt> is available and the
boot CPU's per-CPU variables are set up.
The read-side primitives (<tt>rcu_read_lock()</tt>,
<tt>rcu_read_unlock()</tt>, <tt>rcu_dereference()</tt>,
and <tt>rcu_access_pointer()</tt>) will operate normally very early on,
as will <tt>rcu_assign_pointer()</tt>.

<p>
Although <tt>call_rcu()</tt> may be invoked at any
time during boot, callbacks are not guaranteed to be invoked until after
the scheduler is fully up and running.
This delay in callback invocation is due to the fact that RCU does not
invoke callbacks until it is fully initialized, and this full initialization
cannot occur until after the scheduler has initialized itself to the
point where RCU can spawn and run its kthreads.
In theory, it would be possible to invoke callbacks earlier,
however, this is not a panacea because there would be severe restrictions
on what operations those callbacks could invoke.

<p>
Perhaps surprisingly, <tt>synchronize_rcu()</tt>,
<a href="#Bottom-Half Flavor"><tt>synchronize_rcu_bh()</tt></a>
(<a href="#Bottom-Half Flavor">discussed below</a>),
and
<a href="#Sched Flavor"><tt>synchronize_sched()</tt></a>
will all operate normally
during very early boot, the reason being that there is only one CPU
and preemption is disabled.
This means that the call <tt>synchronize_rcu()</tt> (or friends)
itself is a quiescent
state and thus a grace period, so the early-boot implementation can
be a no-op.

<p>
Both <tt>synchronize_rcu_bh()</tt> and <tt>synchronize_sched()</tt>
continue to operate normally through the remainder of boot, courtesy
of the fact that preemption is disabled across their RCU read-side
critical sections and also courtesy of the fact that there is still
only one CPU.
However, once the scheduler starts initializing, preemption is enabled.
There is still only a single CPU, but the fact that preemption is enabled
means that the no-op implementation of <tt>synchronize_rcu()</tt> no
longer works in <tt>CONFIG_PREEMPT=y</tt> kernels.
Therefore, as soon as the scheduler starts initializing, the early-boot
fastpath is disabled.
This means that <tt>synchronize_rcu()</tt> switches to its runtime
mode of operation where it posts callbacks, which in turn means that
any call to <tt>synchronize_rcu()</tt> will block until the corresponding
callback is invoked.
Unfortunately, the callback cannot be invoked until RCU's runtime
grace-period machinery is up and running, which cannot happen until
the scheduler has initialized itself sufficiently to allow RCU's
kthreads to be spawned.
Therefore, invoking <tt>synchronize_rcu()</tt> during scheduler
initialization can result in deadlock.

<p>@@QQ@@
So what happens with <tt>synchronize_rcu()</tt> during
scheduler initialization for <tt>CONFIG_PREEMPT=n</tt>
kernels?
<p>@@QQA@@
In <tt>CONFIG_PREEMPT=n</tt> kernel, <tt>synchronize_rcu()</tt>
maps directly to <tt>synchronize_sched()</tt>.
Therefore, <tt>synchronize_rcu()</tt> works normally throughout
boot in <tt>CONFIG_PREEMPT=n</tt> kernels.
However, your code must also work in <tt>CONFIG_PREEMPT=y</tt> kernels,
so it is still necessary to avoid invoking <tt>synchronize_rcu()</tt>
during scheduler initialization.
<p>@@QQE@@

<p>
I learned of these boot-time requirements as a result of a series of
system hangs.

<h3><a name="Interrupts and NMIs">Interrupts and NMIs</a></h3>

<p>
The Linux kernel has interrupts, and RCU read-side critical sections are
legal within interrupt handlers and within interrupt-disabled regions
of code, as are invocations of <tt>call_rcu()</tt>.

<p>
Some Linux-kernel architectures can enter an interrupt handler from
non-idle process context, and then just never leave it, instead stealthily
transitioning back to process context.
This trick is sometimes used to invoke system calls from inside the kernel.
These &ldquo;half-interrupts&rdquo; mean that RCU has to be very careful
about how it counts interrupt nesting levels.
I learned of this requirement the hard way during a rewrite
of RCU's dyntick-idle code.

<p>
The Linux kernel has non-maskable interrupts (NMIs), and
RCU read-side critical sections are legal within NMI handlers.
Thankfully, RCU update-side primitives, including
<tt>call_rcu()</tt>, are prohibited within NMI handlers.

<p>
The name notwithstanding, some Linux-kernel architectures
can have nested NMIs, which RCU must handle correctly.
Andy Lutomirski
<a href="https://lkml.kernel.org/g/CALCETrXLq1y7e_dKFPgou-FKHB6Pu-r8+t-6Ds+8=va7anBWDA@mail.gmail.com">surprised me</a>
with this requirement;
he also kindly surprised me with
<a href="https://lkml.kernel.org/g/CALCETrXSY9JpW3uE6H8WYk81sg56qasA2aqmjMPsq5dOtzso=g@mail.gmail.com">an algorithm</a>
that meets this requirement.

<h3><a name="Loadable Modules">Loadable Modules</a></h3>

<p>
The Linux kernel has loadable modules, and these modules can
also be unloaded.
After a given module has been unloaded, any attempt to call
one of its functions results in a segmentation fault.
The module-unload functions must therefore cancel any
delayed calls to loadable-module functions, for example,
any outstanding <tt>mod_timer()</tt> must be dealt with
via <tt>del_timer_sync()</tt> or similar.

<p>
Unfortunately, there is no way to cancel an RCU callback;
once you invoke <tt>call_rcu()</tt>, the callback function is
going to eventually be invoked, unless the system goes down first.
Because it is normally considered socially irresponsible to crash the system
in response to a module unload request, we need some other way
to deal with in-flight RCU callbacks.

<p>
RCU therefore provides
<tt><a href="https://lwn.net/Articles/217484/">rcu_barrier()</a></tt>,
which waits until all in-flight RCU callbacks have been invoked.
If a module uses <tt>call_rcu()</tt>, its exit function should therefore
prevent any future invocation of <tt>call_rcu()</tt>, then invoke
<tt>rcu_barrier()</tt>.
In theory, the underlying module-unload code could invoke
<tt>rcu_barrier()</tt> unconditionally, but in practice this would
incur unacceptable latencies.

<p>
Nikita Danilov noted this requirement for an analogous filesystem-unmount
situation, and Dipankar Sarma incorporated <tt>rcu_barrier()</tt> into RCU.
The need for <tt>rcu_barrier()</tt> for module unloading became
apparent later.

<h3><a name="Hotplug CPU">Hotplug CPU</a></h3>

<p>
The Linux kernel supports CPU hotplug, which means that CPUs
can come and go.
It is of course illegal to use any RCU API member from an offline CPU.
This requirement was present from day one in DYNIX/ptx, but
on the other hand, the Linux kernel's CPU-hotplug implementation
is &ldquo;interesting.&rdquo;

<p>
The Linux-kernel CPU-hotplug implementation has notifiers that
are used to allow the various kernel subsystems (including RCU)
to respond appropriately to a given CPU-hotplug operation.
Most RCU operations may be invoked from CPU-hotplug notifiers,
including even normal synchronous grace-period operations
such as <tt>synchronize_rcu()</tt>.
However, expedited grace-period operations such as
<tt>synchronize_rcu_expedited()</tt> are not supported,
due to the fact that current implementations block CPU-hotplug
operations, which could result in deadlock.

<p>
In addition, all-callback-wait operations such as
<tt>rcu_barrier()</tt> are also not supported, due to the
fact that there are phases of CPU-hotplug operations where
the outgoing CPU's callbacks will not be invoked until after
the CPU-hotplug operation ends, which could also result in deadlock.

<h3><a name="Scheduler and RCU">Scheduler and RCU</a></h3>

<p>
RCU depends on the scheduler, and the scheduler uses RCU to
protect some of its data structures.
This means the scheduler is forbidden from acquiring
the runqueue locks and the priority-inheritance locks
in the middle of an outermost RCU read-side critical section unless
it also releases them before exiting that same
RCU read-side critical section.
This same prohibition also applies to any lock that is acquired
while holding any lock to which this prohibition applies.
Violating this rule results in deadlock.

<p>
For RCU's part, the preemptible-RCU <tt>rcu_read_unlock()</tt>
implementation must be written carefully to avoid similar deadlocks.
In particular, <tt>rcu_read_unlock()</tt> must tolerate an
interrupt where the interrupt handler invokes both
<tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>.
This possibility requires <tt>rcu_read_unlock()</tt> to use
negative nesting levels to avoid destructive recursion via
interrupt handler's use of RCU.

<p>
This pair of mutual scheduler-RCU requirements came as a
<a href="https://lwn.net/Articles/453002/">complete surprise</a>.

<p>
As noted above, RCU makes use of kthreads, and it is necessary to
avoid excessive CPU-time accumulation by these kthreads.
This requirement was no surprise, but RCU's violation of it
when running context-switch-heavy workloads when built with
<tt>CONFIG_NO_HZ_FULL=y</tt>
<a href="http://www.rdrop.com/users/paulmck/scalability/paper/BareMetal.2015.01.15b.pdf">did come as a surprise [PDF]</a>.
RCU has made good progress towards meeting this requirement, even
for context-switch-have <tt>CONFIG_NO_HZ_FULL=y</tt> workloads,
but there is room for further improvement.

<h3><a name="Tracing and RCU">Tracing and RCU</a></h3>

<p>
It is possible to use tracing on RCU code, but tracing itself
uses RCU.
For this reason, <tt>rcu_dereference_raw_notrace()</tt>
is provided for use by tracing, which avoids the destructive
recursion that could otherwise ensue.
This API is also used by virtualization in some architectures,
where RCU readers execute in environments in which tracing
cannot be used.
The tracing folks both located the requirement and provided the
needed fix, so this surprise requirement was relatively painless.

<h3><a name="Energy Efficiency">Energy Efficiency</a></h3>

<p>
Interrupting idle CPUs is considered socially unacceptable,
especially by people with battery-powered embedded systems.
RCU therefore conserves energy by detecting which CPUs are
idle, including tracking CPUs that have been interrupted from idle.
This is a large part of the energy-efficiency requirement,
so I learned of this via an irate phone call.

<p>
Because RCU avoids interrupting idle CPUs, it is illegal to
execute an RCU read-side critical section on an idle CPU.
(Kernels built with <tt>CONFIG_PROVE_RCU=y</tt> will splat
if you try it.)
The <tt>RCU_NONIDLE()</tt> macro and <tt>_rcuidle</tt>
event tracing is provided to work around this restriction.
In addition, <tt>rcu_is_watching()</tt> may be used to
test whether or not it is currently legal to run RCU read-side
critical sections on this CPU.
I learned of the need for diagnostics on the one hand
and <tt>RCU_NONIDLE()</tt> on the other while inspecting
idle-loop code.
Steven Rostedt supplied <tt>_rcuidle</tt> event tracing,
which is used quite heavily in the idle loop.

<p>
It is similarly socially unacceptable to interrupt an
<tt>nohz_full</tt> CPU running in userspace.
RCU must therefore track <tt>nohz_full</tt> userspace
execution.
And in
<a href="https://lwn.net/Articles/558284/"><tt>CONFIG_NO_HZ_FULL_SYSIDLE=y</tt></a>
kernels, RCU must separately track idle CPUs on the one hand and
CPUs that are either idle or executing in userspace on the other.
In both cases, RCU must be able to sample state at two points in
time, and be able to determine whether or not some other CPU spent
any time idle and/or executing in userspace.

<p>
These energy-efficiency requirements have proven quite difficult to
understand and to meet, for example, there have been more than five
clean-sheet rewrites of RCU's energy-efficiency code, the last of
which was finally able to demonstrate
<a href="http://www.rdrop.com/users/paulmck/realtime/paper/AMPenergy.2013.04.19a.pdf">real energy savings running on real hardware [PDF]</a>.
As noted earlier,
I learned of many of these requirements via angry phone calls:
Flaming me on the Linux-kernel mailing list was apparently not
sufficient to fully vent their ire at RCU's energy-efficiency bugs!

<h3><a name="Memory Efficiency">Memory Efficiency</a></h3>

<p>
Although small-memory non-realtime systems can simply use Tiny RCU,
code size is only one aspect of memory efficiency.
Another aspect is the size of the <tt>rcu_head</tt> structure
used by <tt>call_rcu()</tt> and <tt>kfree_rcu()</tt>.
Although this structure contains nothing more than a pair of pointers,
it does appear in many RCU-protected data structures, including
some that are size critical.
The <tt>page</tt> structure is a case in point, as evidenced by
the many occurrences of the <tt>union</tt> keyword within that structure.

<p>
This need for memory efficiency is one reason that RCU uses hand-crafted
singly linked lists to track the <tt>rcu_head</tt> structures that
are waiting for a grace period to elapse.
It is also the reason why <tt>rcu_head</tt> structures do not contain
debug information, such as fields tracking the file and line of the
<tt>call_rcu()</tt> or <tt>kfree_rcu()</tt> that posted them.
Although this information might appear in debug-only kernel builds at some
point, in the meantime, the <tt>-&gt;func</tt> field will often provide
the needed debug information.

<p>
However, in some cases, the need for memory efficiency leads to even
more extreme measures.
Returning to the <tt>page</tt> structure, the <tt>rcu_head</tt> field
shares storage with a great many other structures that are used at
various points in the corresponding page's lifetime.
In order to correctly resolve certain
<a href="https://lkml.kernel.org/g/1439976106-137226-1-git-send-email-kirill.shutemov@linux.intel.com">race conditions</a>,
the Linux kernel's memory-management subsystem needs a particular bit
to remain zero during all phases of grace-period processing,
and that bit happens to map to the bottom bit of the
<tt>rcu_head</tt> structure's <tt>-&gt;next</tt> field.
RCU makes this guarantee as long as <tt>call_rcu()</tt>
is used to post the callback, as opposed to <tt>kfree_rcu()</tt>
or some future &ldquo;lazy&rdquo;
variant of <tt>call_rcu()</tt> that might one day be created for
energy-efficiency purposes.

<h3><a name="Performance, Scalability, Response Time, and Reliability">
Performance, Scalability, Response Time, and Reliability</a></h3>

<p>
Expanding on the
<a href="#Performance and Scalability">earlier discussion</a>,
RCU is used heavily by hot code paths in performance-critical
portions of the Linux kernel's networking, security, virtualization,
and scheduling code paths.
RCU must therefore use efficient implementations, especially in its
read-side primitives.
To that end, it would be good if preemptible RCU's implementation
of <tt>rcu_read_lock()</tt> could be inlined, however, doing
this requires resolving <tt>#include</tt> issues with the
<tt>task_struct</tt> structure.

<p>
The Linux kernel supports hardware configurations with up to
4096 CPUs, which means that RCU must be extremely scalable.
Algorithms that involve frequent acquisitions of global locks or
frequent atomic operations on global variables simply cannot be
tolerated within the RCU implementation.
RCU therefore makes heavy use of a combining tree based on the
<tt>rcu_node</tt> structure.
RCU is required to tolerate all CPUs continuously invoking any
combination of RCU's runtime primitives with minimal per-operation
overhead.
In fact, in many cases, increasing load must <i>decrease</i> the
per-operation overhead, witness the batching optimizations for
<tt>synchronize_rcu()</tt>, <tt>call_rcu()</tt>,
<tt>synchronize_rcu_expedited()</tt>, and <tt>rcu_barrier()</tt>.
As a general rule, RCU must cheerfully accept whatever the
rest of the Linux kernel decides to throw at it.

<p>
The Linux kernel is used for real-time workloads, especially
in conjunction with the
<a href="https://rt.wiki.kernel.org/index.php/Main_Page">-rt patchset</a>.
The real-time-latency response requirements are such that the
traditional approach of disabling preemption across RCU
read-side critical sections is inappropriate.
Kernels built with <tt>CONFIG_PREEMPT=y</tt> therefore
use an RCU implementation that allows RCU read-side critical
sections to be preempted.
This requirement made its presence known after users made it
clear that an earlier
<a href="https://lwn.net/Articles/107930/">real-time patch</a>
did not meet their needs, in conjunction with some
<a href="https://lkml.kernel.org/g/20050318002026.GA2693@us.ibm.com">RCU issues</a>
encountered by a very early version of the -rt patchset.

<p>
In addition, RCU must make do with a sub-100-microsecond real-time latency
budget.
In fact, on smaller systems with the -rt patchset, the Linux kernel
provides sub-20-microsecond real-time latencies for the whole kernel,
including RCU.
RCU's scalability and latency must therefore be sufficient for
these sorts of configurations.
To my surprise, the sub-100-microsecond real-time latency budget
<a href="http://www.rdrop.com/users/paulmck/realtime/paper/bigrt.2013.01.31a.LCA.pdf">
applies to even the largest systems [PDF]</a>,
up to and including systems with 4096 CPUs.
This real-time requirement motivated the grace-period kthread, which
also simplified handling of a number of race conditions.

<p>
Finally, RCU's status as a synchronization primitive means that
any RCU failure can result in arbitrary memory corruption that can be
extremely difficult to debug.
This means that RCU must be extremely reliable, which in
practice also means that RCU must have an aggressive stress-test
suite.
This stress-test suite is called <tt>rcutorture</tt>.

<p>
Although the need for <tt>rcutorture</tt> was no surprise,
the current immense popularity of the Linux kernel is posing
interesting&mdash;and perhaps unprecedented&mdash;validation
challenges.
To see this, keep in mind that there are well over one billion
instances of the Linux kernel running today, given Android
smartphones, Linux-powered televisions, and servers.
This number can be expected to increase sharply with the advent of
the celebrated Internet of Things.

<p>
Suppose that RCU contains a race condition that manifests on average
once per million years of runtime.
This bug will be occurring about three times per <i>day</i> across
the installed base.
RCU could simply hide behind hardware error rates, given that no one
should really expect their smartphone to last for a million years.
However, anyone taking too much comfort from this thought should
consider the fact that in most jurisdictions, a successful multi-year
test of a given mechanism, which might include a Linux kernel,
suffices for a number of types of safety-critical certifications.
In fact, rumor has it that the Linux kernel is already being used
in production for safety-critical applications.
I don't know about you, but I would feel quite bad if a bug in RCU
killed someone.
Which might explain my recent focus on validation and verification.

<h2><a name="Other RCU Flavors">Other RCU Flavors</a></h2>

<p>
One of the more surprising things about RCU is that there are now
no fewer than five <i>flavors</i>, or API families.
In addition, the primary flavor that has been the sole focus up to
this point has two different implementations, non-preemptible and
preemptible.
The other four flavors are listed below, with requirements for each
described in a separate section.

<ol>
<li>	<a href="#Bottom-Half Flavor">Bottom-Half Flavor</a>
<li>	<a href="#Sched Flavor">Sched Flavor</a>
<li>	<a href="#Sleepable RCU">Sleepable RCU</a>
<li>	<a href="#Tasks RCU">Tasks RCU</a>
</ol>

<h3><a name="Bottom-Half Flavor">Bottom-Half Flavor</a></h3>

<p>
The softirq-disable (AKA &ldquo;bottom-half&rdquo;,
hence the &ldquo;_bh&rdquo; abbreviations)
flavor of RCU, or <i>RCU-bh</i>, was developed by
Dipankar Sarma to provide a flavor of RCU that could withstand the
network-based denial-of-service attacks researched by Robert
Olsson.
These attacks placed so much networking load on the system
that some of the CPUs never exited softirq execution,
which in turn prevented those CPUs from ever executing a context switch,
which, in the RCU implementation of that time, prevented grace periods
from ever ending.
The result was an out-of-memory condition and a system hang.

<p>
The solution was the creation of RCU-bh, which does
<tt>local_bh_disable()</tt>
across its read-side critical sections, and which uses the transition
from one type of softirq processing to another as a quiescent state
in addition to context switch, idle, user mode, and offline.
This means that RCU-bh grace periods can complete even when some of
the CPUs execute in softirq indefinitely, thus allowing algorithms
based on RCU-bh to withstand network-based denial-of-service attacks.

<p>
Because
<tt>rcu_read_lock_bh()</tt> and <tt>rcu_read_unlock_bh()</tt>
disable and re-enable softirq handlers, any attempt to start a softirq
handlers during the
RCU-bh read-side critical section will be deferred.
In this case, <tt>rcu_read_unlock_bh()</tt>
will invoke softirq processing, which can take considerable time.
One can of course argue that this softirq overhead should be associated
with the code following the RCU-bh read-side critical section rather
than <tt>rcu_read_unlock_bh()</tt>, but the fact
is that most profiling tools cannot be expected to make this sort
of fine distinction.
For example, suppose that a three-millisecond-long RCU-bh read-side
critical section executes during a time of heavy networking load.
There will very likely be an attempt to invoke at least one softirq
handler during that three milliseconds, but any such invocation will
be delayed until the time of the <tt>rcu_read_unlock_bh()</tt>.
This can of course make it appear at first glance as if
<tt>rcu_read_unlock_bh()</tt> was executing very slowly.

<p>
The
<a href="https://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-bh API</a>
includes
<tt>rcu_read_lock_bh()</tt>,
<tt>rcu_read_unlock_bh()</tt>,
<tt>rcu_dereference_bh()</tt>,
<tt>rcu_dereference_bh_check()</tt>,
<tt>synchronize_rcu_bh()</tt>,
<tt>synchronize_rcu_bh_expedited()</tt>,
<tt>call_rcu_bh()</tt>,
<tt>rcu_barrier_bh()</tt>, and
<tt>rcu_read_lock_bh_held()</tt>.

<h3><a name="Sched Flavor">Sched Flavor</a></h3>

<p>
Before preemptible RCU, waiting for an RCU grace period had the
side effect of also waiting for all pre-existing interrupt
and NMI handlers.
However, there are legitimate preemptible-RCU implementations that
do not have this property, given that any point in the code outside
of an RCU read-side critical section can be a quiescent state.
Therefore, <i>RCU-sched</i> was created, which follows &ldquo;classic&rdquo;
RCU in that an RCU-sched grace period waits for for pre-existing
interrupt and NMI handlers.
In kernels built with <tt>CONFIG_PREEMPT=n</tt>, the RCU and RCU-sched
APIs have identical implementations, while kernels built with
<tt>CONFIG_PREEMPT=y</tt> provide a separate implementation for each.

<p>
Note well that in <tt>CONFIG_PREEMPT=y</tt> kernels,
<tt>rcu_read_lock_sched()</tt> and <tt>rcu_read_unlock_sched()</tt>
disable and re-enable preemption, respectively.
This means that if there was a preemption attempt during the
RCU-sched read-side critical section, <tt>rcu_read_unlock_sched()</tt>
will enter the scheduler, with all the latency and overhead entailed.
Just as with <tt>rcu_read_unlock_bh()</tt>, this can make it look
as if <tt>rcu_read_unlock_sched()</tt> was executing very slowly.
However, the highest-priority task won't be preempted, so that task
will enjoy low-overhead <tt>rcu_read_unlock_sched()</tt> invocations.

<p>
The
<a href="https://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-sched API</a>
includes
<tt>rcu_read_lock_sched()</tt>,
<tt>rcu_read_unlock_sched()</tt>,
<tt>rcu_read_lock_sched_notrace()</tt>,
<tt>rcu_read_unlock_sched_notrace()</tt>,
<tt>rcu_dereference_sched()</tt>,
<tt>rcu_dereference_sched_check()</tt>,
<tt>synchronize_sched()</tt>,
<tt>synchronize_rcu_sched_expedited()</tt>,
<tt>call_rcu_sched()</tt>,
<tt>rcu_barrier_sched()</tt>, and
<tt>rcu_read_lock_sched_held()</tt>.
However, anything that disables preemption also marks an RCU-sched
read-side critical section, including
<tt>preempt_disable()</tt> and <tt>preempt_enable()</tt>,
<tt>local_irq_save()</tt> and <tt>local_irq_restore()</tt>,
and so on.

<h3><a name="Sleepable RCU">Sleepable RCU</a></h3>

<p>
For well over a decade, someone saying &ldquo;I need to block within
an RCU read-side critical section&rdquo; was a reliable indication
that this someone did not understand RCU.
After all, if you are always blocking in an RCU read-side critical
section, you can probably afford to use a higher-overhead synchronization
mechanism.
However, that changed with the advent of the Linux kernel's notifiers,
whose RCU read-side critical
sections almost never sleep, but sometimes need to.
This resulted in the introduction of
<a href="https://lwn.net/Articles/202847/">sleepable RCU</a>,
or <i>SRCU</i>.

<p>
SRCU allows different domains to be defined, with each such domain
defined by an instance of an <tt>srcu_struct</tt> structure.
A pointer to this structure must be passed in to each SRCU function,
for example, <tt>synchronize_srcu(&amp;ss)</tt>, where
<tt>ss</tt> is the <tt>srcu_struct</tt> structure.
The key benefit of these domains is that a slow SRCU reader in one
domain does not delay an SRCU grace period in some other domain.
That said, one consequence of these domains is that read-side code
must pass a &ldquo;cookie&rdquo; from <tt>srcu_read_lock()</tt>
to <tt>srcu_read_unlock()</tt>, for example, as follows:

<blockquote>
<pre>
 1 int idx;
 2
 3 idx = srcu_read_lock(&amp;ss);
 4 do_something();
 5 srcu_read_unlock(&amp;ss, idx);
</pre>
</blockquote>

<p>
As noted above, it is legal to block within SRCU read-side critical sections,
however, with great power comes great responsibility.
If you block forever in one of a given domain's SRCU read-side critical
sections, then that domain's grace periods will also be blocked forever.
Of course, one good way to block forever is to deadlock, which can
happen if any operation in a given domain's SRCU read-side critical
section can block waiting, either directly or indirectly, for that domain's
grace period to elapse.
For example, this results in a self-deadlock:

<blockquote>
<pre>
 1 int idx;
 2
 3 idx = srcu_read_lock(&amp;ss);
 4 do_something();
 5 synchronize_srcu(&amp;ss);
 6 srcu_read_unlock(&amp;ss, idx);
</pre>
</blockquote>

<p>
However, if line&nbsp;5 acquired a mutex that was held across
a <tt>synchronize_srcu()</tt> for domain <tt>ss</tt>,
deadlock would still be possible.
Furthermore, if line&nbsp;5 acquired a mutex that was held across
a <tt>synchronize_srcu()</tt> for some other domain <tt>ss1</tt>,
and if an <tt>ss1</tt>-domain SRCU read-side critical section
acquired another mutex that was held across as <tt>ss</tt>-domain
<tt>synchronize_srcu()</tt>,
deadlock would again be possible.
Such a deadlock cycle could extend across an arbitrarily large number
of different SRCU domains.
Again, with great power comes great responsibility.

<p>
Unlike the other RCU flavors, SRCU read-side critical sections can
run on idle and even offline CPUs.
This ability requires that <tt>srcu_read_lock()</tt> and
<tt>srcu_read_unlock()</tt> contain memory barriers, which means
that SRCU readers will run a bit slower than would RCU readers.
It also motivates the <tt>smp_mb__after_srcu_read_unlock()</tt>
API, which, in combination with <tt>srcu_read_unlock()</tt>,
guarantees a full memory barrier.

<p>
The
<a href="https://lwn.net/Articles/609973/#RCU Per-Flavor API Table">SRCU API</a>
includes
<tt>srcu_read_lock()</tt>,
<tt>srcu_read_unlock()</tt>,
<tt>srcu_dereference()</tt>,
<tt>srcu_dereference_check()</tt>,
<tt>synchronize_srcu()</tt>,
<tt>synchronize_srcu_expedited()</tt>,
<tt>call_srcu()</tt>,
<tt>srcu_barrier()</tt>, and
<tt>srcu_read_lock_held()</tt>.
It also includes
<tt>DEFINE_SRCU()</tt>,
<tt>DEFINE_STATIC_SRCU()</tt>, and
<tt>init_srcu_struct()</tt>
APIs for defining and initializing <tt>srcu_struct</tt> structures.

<h3><a name="Tasks RCU">Tasks RCU</a></h3>

<p>
Some forms of tracing use &ldquo;tramopolines&rdquo; to handle the
binary rewriting required to install different types of probes.
It would be good to be able to free old trampolines, which sounds
like a job for some form of RCU.
However, because it is necessary to be able to install a trace
anywhere in the code, it is not possible to use read-side markers
such as <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>.
In addition, it does not work to have these markers in the trampoline
itself, because there would need to be instructions following
<tt>rcu_read_unlock()</tt>.
Although <tt>synchronize_rcu()</tt> would guarantee that execution
reached the <tt>rcu_read_unlock()</tt>, it would not be able to
guarantee that execution had completely left the trampoline.

<p>
The solution, in the form of
<a href="https://lwn.net/Articles/607117/"><i>Tasks RCU</i></a>,
is to have implicit
read-side critical sections that are delimited by voluntary context
switches, that is, calls to <tt>schedule()</tt>,
<tt>cond_resched_rcu_qs()</tt>, and
<tt>synchronize_rcu_tasks()</tt>.
In addition, transitions to and from userspace execution also delimit
tasks-RCU read-side critical sections.

<p>
The tasks-RCU API is quite compact, consisting only of
<tt>call_rcu_tasks()</tt>,
<tt>synchronize_rcu_tasks()</tt>, and
<tt>rcu_barrier_tasks()</tt>.

<h2><a name="Possible Future Changes">Possible Future Changes</a></h2>

<p>
One of the tricks that RCU uses to attain update-side scalability is
to increase grace-period latency with increasing numbers of CPUs.
If this becomes a serious problem, it will be necessary to rework the
grace-period state machine so as to avoid the need for the additional
latency.

<p>
Expedited grace periods scan the CPUs, so their latency and overhead
increases with increasing numbers of CPUs.
If this becomes a serious problem on large systems, it will be necessary
to do some redesign to avoid this scalability problem.

<p>
RCU disables CPU hotplug in a few places, perhaps most notably in the
expedited grace-period and <tt>rcu_barrier()</tt> operations.
If there is a strong reason to use expedited grace periods in CPU-hotplug
notifiers, it will be necessary to avoid disabling CPU hotplug.
This would introduce some complexity, so there had better be a <i>very</i>
good reason.

<p>
The tradeoff between grace-period latency on the one hand and interruptions
of other CPUs on the other hand may need to be re-examined.
The desire is of course for zero grace-period latency as well as zero
interprocessor interrupts undertaken during an expedited grace period
operation.
While this ideal is unlikely to be achievable, it is quite possible that
further improvements can be made.

<p>
The multiprocessor implementations of RCU use a combining tree that
groups CPUs so as to reduce lock contention and increase cache locality.
However, this combining tree does not spread its memory across NUMA
nodes nor does it align the CPU groups with hardware features such
as sockets or cores.
Such spreading and alignment is currently believed to be unnecessary
because the hotpath read-side primitives do not access the combining
tree, nor does <tt>call_rcu()</tt> in the common case.
If you believe that your architecture needs such spreading and alignment,
then your architecture should also benefit from the
<tt>rcutree.rcu_fanout_leaf</tt> boot parameter, which can be set
to the number of CPUs in a socket, NUMA node, or whatever.
If the number of CPUs is too large, use a fraction of the number of
CPUs.
If the number of CPUs is a large prime number, well, that certainly
is an &ldquo;interesting&rdquo; architectural choice!
More flexible arrangements might be considered, but only if
<tt>rcutree.rcu_fanout_leaf</tt> has proven inadequate, and only
if the inadequacy has been demonstrated by a carefully run and
realistic system-level workload.

<p>
Please note that arrangements that require RCU to remap CPU numbers will
require extremely good demonstration of need and full exploration of
alternatives.

<p>
There is an embarrassingly large number of flavors of RCU, and this
number has been increasing over time.
Perhaps it will be possible to combine some at some future date.

<p>
RCU's various kthreads are reasonably recent additions.
It is quite likely that adjustments will be required to more gracefully
handle extreme loads.
It might also be necessary to be able to relate CPU utilization by
RCU's kthreads and softirq handlers to the code that instigated this
CPU utilization.
For example, RCU callback overhead might be charged back to the
originating <tt>call_rcu()</tt> instance, though probably not
in production kernels.

<h2><a name="Summary">Summary</a></h2>

<p>
This document has presented more than two decade's worth of RCU
requirements.
Given that the requirements keep changing, this will not be the last
word on this subject, but at least it serves to get an important
subset of the requirements set forth.

<h2><a name="Acknowledgments">Acknowledgments</a></h2>

I am grateful to Steven Rostedt, Lai Jiangshan, Ingo Molnar,
Oleg Nesterov, Borislav Petkov, Peter Zijlstra, Boqun Feng, and
Andy Lutomirski for their help in rendering
this article human readable, and to Michelle Rankin for her support
of this effort.
Other contributions are acknowledged in the Linux kernel's git archive.
The cartoon is copyright (c) 2013 by Melissa Broussard,
and is provided
under the terms of the Creative Commons Attribution-Share Alike 3.0
United States license.

<p>@@QQAL@@

</body></html>