documentation: Cover requirements controlling stall warnings
authorPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Wed, 7 Oct 2015 22:35:35 +0000 (15:35 -0700)
committerPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Sat, 5 Dec 2015 20:32:19 +0000 (12:32 -0800)
This commit adds verbiage on boot and sysfs parameters that can be
used to control RCU CPU stall warnings, both to change the timeout
and to suppress these warnings entirely.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Documentation/RCU/Design/Requirements/Requirements.html
Documentation/RCU/Design/Requirements/Requirements.htmlx

index 871f627b77138357e676fd39f052020f6db824ea..cc5b587c0ec51f7cac4208aeb3903439c8cbd397 100644 (file)
@@ -1618,12 +1618,35 @@ guard against mishaps and misuse:
        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.
+       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.
index a544db4646c6cfb36db3e6de753b34766d1834f3..23524d75a3c38017e090f7f82f57846feb1dd7a3 100644 (file)
@@ -1777,12 +1777,35 @@ guard against mishaps and misuse:
        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.
+       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.