preempt-rt: more rtla stuff

Signed-off-by: Maxime Chevallier <[email protected]>

slides/realtime-linux-benchmarking/realtime-linux-benchmarking.tex

@@ -115,6 +115,18 @@ \section{Testing and Benchmarking}
 	\end{itemize}
 \end{frame}
 
+\begin{frame}
+	\frametitle{rtla}
+	\textbf{R}eal\textbf{T}ime \textbf{L}inux \textbf{A}nalysis tool
+	\begin{itemize}
+		\item Developped by Daniel Bristot de Oliveira
+		\item High-level interface to the \code{timerlat} and \code{osnoise} tracers
+		\item \code{rtla osnoise|timerlat top|hist} gives high-level view of noise and latencies
+		\item Can generate histograms, that can then be visualized
+	\end{itemize}
+
+\end{frame}
+
 \begin{frame}
 	\frametitle{osnoise}
 	\begin{itemize}
@@ -145,6 +157,19 @@ \section{Testing and Benchmarking}
 	%%% Schematics
 \end{frame}
 
+\begin{frame}
+	\frametitle{rtla autoanalysis}
+	\begin{itemize}
+		\item Running \code{timerlat -a <us>} will trigger the auto-analysis mode
+		\item A threshold is set with \code{-a}
+		\item Measurement will stop if a latency higher than the threshold is detected
+		\item Timerlat then prints a stack trace with the cause of the latency
+		\item Can identify if the latency comes from a \textbf{blocking} or an \textbf{interference}
+		\item Can identify the \textbf{task} of \textbf{interrupt} at the origin of the latency
+		\item Can identify if the \textbf{hardware} itself is the culprit
+	\end{itemize}
+\end{frame}
+
 \begin{frame}
 	\frametitle{kernelshark}
 	\begin{itemize}
@@ -157,18 +182,6 @@ \section{Testing and Benchmarking}
 	\end{itemize}
 \end{frame}
 
-\begin{frame}
-	\frametitle{rtla}
-	\textbf{R}eal\textbf{T}ime \textbf{L}inux \textbf{A}nalysis tool
-	\begin{itemize}
-		\item Developed by Daniel Bristot de Oliveira
-		\item High-level interface to the \code{timerlat} and \code{osnoise} tracers
-		\item \code{rtla osnoise|timerlat top|hist} gives high-level view of noise and latencies
-		\item Can generate histograms, that can then be visualized
-	\end{itemize}
-
-\end{frame}
-
 \begin{frame}
 	\frametitle{cyclictest}
 	\begin{itemize}

slides/realtime-linux-configuration/realtime-linux-configuration.tex

@@ -71,6 +71,7 @@ \section{Configuring the system}
 				\item \code{while read i; do /bin/echo $i; done < /dev/cpuset/tasks > /dev/cpuset/nontrt-set/tasks}
 				\item \code{/bin/echo $$ > /dev/cpuset/rt-set/tasks}
 			\end{itemize}
+		\item You can run tasks in a given set with \code{cgexec -g cpuset:rt-set ...}
 	\end{itemize}
 \end{frame}
 
@@ -85,9 +86,27 @@ \section{Configuring the system}
 		\item IRQs can be pinned to CPUs by tweaking \code{/proc/irq/XX/smp_affinity}
 		\item The \code{irqbalance} tool monitors and distributes the irq affinty to spread the load across CPUs
 		\item Use the \code{IRQBALANCE_BANNED_CPUS} environment variable to make \code{irqbalance} ignore some CPUs
+		\item The \code{irqaffinity} cmdline parameter can also be used
 	\end{itemize}
 \end{frame}
 
+\begin{frame}
+	\frametitle{RCU Callbacks and Workqueues}
+	\begin{center}\textbf{RCU}\end{center}
+	\begin{itemize}
+		\item \textbf{R}ead \textbf{C}opy \textbf{U}pdate
+		\item Synchronisation mechanism that can deferred object reclamation
+		\item Deffered reclamation can be executed on any CPU \code{RCU callbacks}
+		\item We can preven CPU cores from running RCU callbacks with \code{rcu_nocbs=<cpus> rcu_nocb_poll}
+	\end{itemize}
+	\begin{center}\textbf{Workqueues}\end{center}
+	\begin{itemize}
+		\item Deferred execution mechanism
+		\item Can be pinned to CPUs in \code{/sys/devices/virtual/workqueue/cpumask}
+	\end{itemize}
+\end{frame}
+%TODO RCU/workqueues
+
 % scheduling classes
 \input{../common/scheduling-classes.tex}