\input{configpres}

\section{Basics}

\title{Linux Fastboot}
\maketitle

\subsection{Motivation}
\begin{frame}
\frametitle{Motivation}
\begin{itemize}
\item always greater requirements for energy saving functions
\item runtime of multimedia devices
\item automotive applications
\pause
\item \textbf{Solution:} power off when idle
\pause
\item \textbf{But:} users do not like ''waiting''
\end{itemize}
\end{frame}

\subsection{Theoretical Basics}
\begin{frame}
\frametitle{The first step: Requirements Definition}
\begin{itemize}
\item What is the maximum allowed boot time?
\item Which functionality must be availalbe after this time?
\item speed vs. flexibility
\end{itemize}
\end{frame}

\begin{frame}
\frametitle{Boot Process}
\includegraphics[height=0.7\textheight]{images/boot_overview.png}
\end{frame}

\begin{frame}
\frametitle{Actors in the Boot Process}
\begin{itemize}
\item hardware reset
\item bootloader
\item operating system (load drivers, mount filesystems)
\item start scripts / applications
\end{itemize}
\end{frame}

\begin{frame}
\frametitle{Critical Hardware Components}
\begin{itemize}
\item power supply
\item reset logic
\item boot logic / boot order
\item connection to the boot medium
\item connection to required peripherals
\end{itemize}
\textbf{Important: The hardware is an essential part of a fastboot concept!!}
\end{frame}

\begin{frame}
\frametitle{Bootloader}
\begin{itemize}
\item ''base configuration'' of the CPU
\item modifies the device tree and/or sets up the ''ATAGs structure''
\item flushes the caches
\item disables the MMU
\end{itemize}
\end{frame}

\begin{frame}
\frametitle{The Linux Kernel}
\begin{itemize}
\item many features for boot time optimization
\item very flexible
\item compression
\item possible to initialize in parallel
\item approx. 150ms - 250ms to mount a filesystem
\end{itemize}
\end{frame}

\begin{frame}
\frametitle{The Application}
\begin{itemize}
\item for systems, this usually has the most potential for optimizing
\item start scripts
\item linking
\end{itemize}
\end{frame}

\section{Optimization}
\subsection{Bootloader}
\begin{frame}[fragile]
\frametitle{Bootloader Optimization (U-Boot) 1}
remove features that are not needed in the field:
\begin{lstlisting}
/* include/configs/boardname.h */
[...]
#include <config_cmd_default.h>
#undef CONFIG_CMD_NET
[...]
\end{lstlisting}
\end{frame}

\begin{frame}[fragile]
\frametitle{Bootloader Optimization (U-Boot) 2}
disable verification of the kernel image:
\begin{verbatim}
setenv verify n
\end{verbatim}
disable output on the U-Boot console:
\begin{verbatim}
setenv silent 1
\end{verbatim}
do not wait for user input:
\begin{verbatim}
setenv bootdelay 0
\end{verbatim}
\end{frame}

\begin{frame}
\frametitle{Bootloader Optimization (IPL)}
\begin{itemize}
\item replace a ''general purpose'' bootloader with an optimized IPL
\item IPL also as a basis for update concepts and ''rescue kernels''
\end{itemize}
\end{frame}

\subsection{Kernel}
\begin{frame}[fragile]
\frametitle{Kernel Optimization: Components}
\begin{itemize}
\item configuration and build
\item compression
\item boot parameters
\item driver initialization
\item root filesystem
\end{itemize}
\end{frame}

\begin{frame}[fragile]
\frametitle{Kernel Optimization: Configuration}
\begin{verbatim}
General setup  --->
Kernel compression mode -->
\end{verbatim}
\begin{itemize}
\item LZO for embedded systems very interesting
\item copy vs. decompress
\item for storage medium with direct addressable read access, ''excecute in place (XIP)'' is available
\end{itemize}
\end{frame}

\begin{frame}[fragile]
\frametitle{Kernel Optimization: Command Line Parameter}
\begin{itemize}
\item delay loop calibration: ''lpj=''; for ARM9 systems, this can save >100ms
\item parameters available for runtime analysis of the boot process: ''initcall\_debug'', ''printk.time=1''
\end{itemize}
\end{frame}

\begin{frame}[fragile]
\frametitle{Kernel Optimization: Delay Loop}
\begin{verbatim}
...
Calibrating delay loop...
   99.12 BogoMIPS (lpj=495616)
...
\end{verbatim}
\end{frame}

\begin{frame}[fragile]
\frametitle{Kernel Optimization: initcall\_debug}
\begin{verbatim}
calling  populate_rootfs+0x0/0x1f8 @ 1
initcall populate_rootfs+0x0/0x1f8
        returned 0 after 281 usecs
calling  timer_init_sysfs+0x0/0x38 @ 1
initcall timer_init_sysfs+0x0/0x38
        returned 0 after 590 usecs
calling  leds_init+0x0/0x3c @ 1
initcall leds_init+0x0/0x3c
        returned 0 after 488 usecs
calling  leds_init+0x0/0x50 @ 1
initcall leds_init+0x0/0x50
        returned -19 after 1 usecs
calling  fpe_init+0x0/0x78 @ 1
\end{verbatim}
\end{frame}

\begin{frame}[fragile]
\frametitle{Kernel Optimization: printk.time}
\begin{verbatim}
[    0.800000] mice: PS/2 mouse device 
[    0.810000] rtc-at91sam9 at91_rtt.0:
[    0.820000] rtc-at91sam9 at91_rtt.0: 
[    0.830000] TCP cubic registered
[    0.830000] NET: Registered protocol 
[    0.830000] rtc-at91sam9 at91_rtt.0:
[    2.610000] VFS: Mounted root (jffs2 
[    2.610000] Freeing init memory: 116K
\end{verbatim}
\end{frame}

\subsection{File Systems}
\begin{frame}[fragile]
\frametitle{Initramfs}
\begin{verbatim}
dir /dev 755 0 0
nod /dev/console 644 0 0 c 5 1
nod /dev/loop0 644 0 0 b 7 0
dir /bin 755 1000 1000
slink /bin/sh busybox 777 0 0
file /bin/busybox initfs/busybox 755 0 0
[...]
dir /proc 755 0 0
dir /sys 755 0 0
dir /mnt 755 0 0
\end{verbatim}
\end{frame}

\begin{frame}
\frametitle{Ubifs}
\begin{itemize}
\item for Linux systems, this is \textbf{the} flash filesystem to choose
\item fast mount times
\item power-fail safe
\item \dots
\end{itemize}
\end{frame}

\subsection{Application}

\begin{frame}
\frametitle{Optimizing the Application}
\begin{itemize}
\item analyze the start processes with bootchartd
\item possibly starting the application directly with init=
\item dynamic vs. static linking
\item prelink for dynamically linked applications
\end{itemize}
\end{frame}

\begin{frame}[fragile]
\frametitle{Start Script Tasks in the Application}
\begin{verbatim}
ret = mount("sysfs", "/sys",
              "sysfs", 0, NULL);

if(ret < 0)
      perror("Can't mount sysfs\n");
\end{verbatim}
\end{frame}

\begin{frame}[fragile]
\frametitle{Dynamic Linking}
\begin{enumerate}
\item DT\_RPATH section in ELF executables
\item paths that are specified in LD\_LIBRARY\_PATH
\item DT\_RUNPATH section in ELF executables
\item dynamic loader cache: /etc/ld.so.cache
\item default paths /lib, /usr/lib, \dots
\end{enumerate}
\end{frame}

\begin{frame}[fragile]
\frametitle{Dynamic Linking: Analysis}
\begin{verbatim}
$ LD_DEBUG=libs ls
   3082: find library=librt.so.1 [0];
         searching
   3082: search cache=/etc/ld.so.cache
   3082: trying file=/lib/librt.so.1
\end{verbatim}
\end{frame}

\section{Example}
\subsection{Analysis of a Test System}

\begin{frame}
\frametitle{Test System}
\begin{itemize}
\item ARM9 CPU from the Atmel AT91 series
\item starting point: busybox image from the Angstrom distribution
\item storage medium: NAND flash
\item test application: toggling a GPIO via sysfs interface
\end{itemize}
\end{frame}

\begin{frame}
\frametitle{AT91 Controller Boot Process}
\includegraphics[height=0.7\textheight]{images/boot_at91_overview.png}
\end{frame}

\begin{frame}
\frametitle{The AT91 Family Boot Modes}
\begin{itemize}
\item RomBOOT: boot via an internal boot logic
\item boot via chip-select 0 (CS0) of the external bus interface
\end{itemize}
\end{frame}

\begin{frame}
\frametitle{AT91 RomBOOT}
\includegraphics[height=0.7\textheight]{images/boot_at91_romboot.png}
\end{frame}

\begin{frame}
\frametitle{Power-On Behavior / Power Supply}
\includegraphics[height=0.7\textheight]{images/at91_netzteil.png}
\end{frame}

\begin{frame}
\frametitle{Reset Behavior}
\includegraphics[height=0.7\textheight]{images/at91_reset.png}
\end{frame}

\begin{frame}
\frametitle{RomBOOT}
\includegraphics[height=0.7\textheight]{images/at91_romboot_measure.png}
\end{frame}

\begin{frame}
\frametitle{Summary / Hardware Optimization}
\begin{itemize}
\item use the internal oscillator for slowclock: >1s savings
\item boot from CS0: ~100ms - 150ms savings
\end{itemize}
\end{frame}

\subsection{Optimizing the Test System}

\begin{frame}
\frametitle{Boot Time Measurement via GPIO}
\includegraphics[width=10cm]{images/gpio_measure.png}
\end{frame}

\begin{frame}
\frametitle{Parts of the Boot Time Measurement}
\begin{itemize}
\item from bootstrap until U-Boot
\item from U-Boot until early boot code in the kernel (also includes copying the kernel into RAM and unpacking the kernel)
\item from kernel until application (also includes mounting the filesystem)
\end{itemize}
\end{frame}

\begin{frame}
\frametitle{Initial Boot Times}
\begin{table}[h]
\centering
\begin{tabular}{ | c | c | }
\hline
Boot Part & Time \\
\hline 
bootstrap - U-Boot  & --- \\
\hline
U-Boot - kernel  & 6.5s  \\
\hline
kernel - application & 4.5s \\
\hline
\textbf{Total} & \textbf{11s} \\
\hline
\end{tabular}
\end{table}
\end{frame}

\begin{frame}
\frametitle{U-Boot without Netzwerk Support}
\begin{table}[h]
\centering
\begin{tabular}{ | c | c | }
\hline
Boot Part & Time \\
\hline 
bootstrap - U-Boot  & --- \\
\hline
U-Boot - kernel  & 4.25s  \\
\hline
kernel - application & 4.5s \\
\hline
\textbf{Total} & \textbf{8.75s} \\
\hline
\end{tabular}
\end{table}
\end{frame}

\begin{frame}
\frametitle{U-Boot verify=n}
\begin{table}[h]
\centering
\begin{tabular}{ | c | c | }
\hline
Boot Part & Time \\
\hline 
bootstrap - U-Boot  & --- \\
\hline
U-Boot - kernel  & 3.89s  \\
\hline
kernel - application & 4.5s \\
\hline
\textbf{Total} & \textbf{8.39s} \\
\hline
\end{tabular}
\end{table}
\end{frame}

\begin{frame}
\frametitle{Kernel ''Trimming''}
\begin{table}[h]
\centering
\begin{tabular}{ | c | c | }
\hline
Boot Part & Time \\
\hline 
bootstrap - U-Boot  & --- \\
\hline
U-Boot - kernel  & 3.77s  \\
\hline
kernel - application & 4.33s \\
\hline
\textbf{Total} & \textbf{8.1s} \\
\hline
\end{tabular}
\end{table}
\end{frame}

\begin{frame}
\frametitle{Analyze the Start Scripts: bootchartd}
\includegraphics[height=0.85\textheight]{images/bootchart.png}
\end{frame}

\begin{frame}
\frametitle{Optimizing the Start Scripts}
\begin{table}[h]
\centering
\begin{tabular}{ | c | c | }
\hline
Boot Part & Time \\
\hline 
bootstrap - U-Boot  & --- \\
\hline
U-Boot - kernel  & 3.77s  \\
\hline
kernel - application & 3.61s \\
\hline
\textbf{Total} & \textbf{7.38s} \\
\hline
\end{tabular}
\end{table}
\end{frame}

\begin{frame}
\frametitle{Initramfs Compressed with LZO}
The application is started directly with the init= parameter.
\begin{table}[h]
\centering
\begin{tabular}{ | c | c | }
\hline
Boot Part & Time \\
\hline 
bootstrap - U-Boot  & --- \\
\hline
U-Boot - kernel  & 3.79s  \\
\hline
kernel - application & 0.372s \\
\hline
\textbf{Total} & \textbf{4.162s} \\
\hline
\end{tabular}
\end{table}
\end{frame}

\begin{frame}
\frametitle{Modified AT91 Bootstrap}
AT91 bootstrap starts Linux directly, i.e. the system boots without U-Boot.
\begin{table}[h]
\centering
\begin{tabular}{ | c | c | }
\hline
Boot Part & Time \\
\hline 
bootstrap - kernel  & 676ms  \\
\hline
kernel - application & 584ms \\
\hline
\textbf{Total} & \textbf{1.260s} \\
\hline
\end{tabular}
\end{table}
\end{frame}

\begin{frame}
\frametitle{lpj=}
\begin{table}[h]
\centering
\begin{tabular}{ | c | c | }
\hline
Boot Part & Time \\
\hline 
bootstrap - kernel  & 676ms  \\
\hline
kernel - application & 384ms \\
\hline
\textbf{Total} & \textbf{1.060s} \\
\hline
\end{tabular}
\end{table}
\end{frame}

\begin{frame}
\frametitle{No Output on the Serial Interface (quiet)}
\begin{table}[!h]
\centering
\begin{tabular}{ | c | c | }
\hline
Boot Part & Time \\
\hline 
bootstrap - kernel  & 524ms  \\
\hline
kernel - application & 212ms \\
\hline
\textbf{Total} & \textbf{736ms} \\
\hline
\end{tabular}
\end{table}
\end{frame}

\begin{frame}
\frametitle{Kernel Image Compressed with LZO}
\begin{table}[h]
\centering
\begin{tabular}{ | c | c | }
\hline
Boot Part & Time \\
\hline 
bootstrap - kernel  & 444ms  \\
\hline
kernel - application & 212ms \\
\hline
\textbf{Total} & \textbf{656ms} \\
\hline
\end{tabular}
\end{table}
\end{frame}

\begin{frame}
\frametitle{Boot Times After the Optimizations}
\includegraphics[height=0.7\textheight]{images/at91_ipl_quiet_lpj_lzo.png}
\end{frame}

\begin{frame}
\frametitle{Summary}
\begin{itemize}
\item Linux provides an optimal platform for anyone that wants a modern operating system, yet also needs to boot in under 1s
\item using just the simple optimizations can save multiple seconds
\item the hardware is an essential part of a fastboot concept
\item the fastboot concept is mostly platform independent
\end{itemize}
\end{frame}

\input{tailpres}