Reducing Boot Time in Embedded Linux Systems

Using some reasonably simple techniques, you may be able to reduce dramatically the boot time of your embedded Linux system.

It is no secret that Linux has won the race in the embedded device marketplace. Tremendous advantages in Linux have broken almost every barrier to entry for using Linux on embedded systems across a wide variety of processor architectures. Today's developers are not asking, “Should I use Linux for my embedded system?”, but instead are asking questions like, “How can I get more performance out of my embedded Linux design?” Reducing boot time has become one of the more interesting discussions taking place in the embedded Linux community.

As it turns out, it is relatively easy to save substantial time on system boot. Without a significant expenditure of engineering resources, savings of more than 80% are possible with certain system configurations. Of course, there is a point of diminishing returns. The graph of engineering effort against boot time would rapidly approach infinite effort as time reduced into the milliseconds and lower.

Fast Boot Requires Definition

Before you can measure boot time, you must define what it means. (I introduce measurement techniques later in this article.) Most often, your customers or end users provide, or at least influence, the definition. The type of product you design certainly impacts your definition. Most systems that appear to boot very quickly actually are just providing early feedback to users in the form of graphical banners, audible feedback, animation or some combination thereof. You as the system designer must specify what it means for your embedded device to be booted and exactly what the user experience will be during power-on.

Do you define boot time as the time from power-on to playing your favorite music? Or, maybe you design big iron, and boot time eats into your annual “five-nines” reliability budget. A cellular radio node controller that takes two minutes to boot eats up almost half your annual downtime budget! Yet, many systems we perceive as fast boot systems are not actually booting from power-on. Consider a popular cell-phone design, such as the BlackBerry Curve. The only time these systems perform a full boot is when the battery is removed and replaced. Power “on” is actually a resume from a low-power system state that largely preserves its current operational status.

It Starts with the Hardware

Although it may seem trivial to mention, sound hardware design is a fundamental component of a fast boot system. Many aspects of hardware design can have a marked influence on 1) how quickly your first bits of code get to execute and 2) how quickly that code can be read out of a nonvolatile storage device during initial boot. Pay particular attention to power-on reset circuitry and initial hardware strapping, which provides default timings for external buses and chip selects on certain processors. It is not uncommon to find “conservative” values being employed here that often can be improved upon.

Your overall hardware architecture will set the stage for what performance you will be able to achieve. Choice of processor, clock speed, choice of nonvolatile storage used for boot images and many other factors will influence how fast your design can fetch and execute its startup image (usually a bootloader) and then go on to load and execute an operating system. Your hardware choices at design time must be carefully considered if single-digit boot times are part of your product requirements.

Typical Boot Sequence

To understand where time is being spent, it helps to visualize the boot sequence of a typical embedded Linux system. Figure 1 shows the basic sequence.

Figure 1. Typical Boot Sequence of Events

Upon power-on, the hardware needs time for voltages (and often clocks) to stabilize and for reset to be released. The first code executed upon release of reset depends on the hardware architecture and processor, but often it is your bootloader running from nonvolatile memory, such as NOR Flash. A small section of code performs some low-level initialization that includes the memory controller and typically copies itself into DRAM for further execution. This copy operation can consume a significant portion of boot time. It is easy to see that keeping the bootloader small and simple (the KISS principle) will help keep boot time to a minimum. The bootloader's primary responsibility after hardware initialization is to locate, load and pass control to your Linux kernel. Once the kernel has completed its own initialization, it must locate and mount a root filesystem. Your root filesystem will contain a set of initialization scripts as well as your own applications. There are numerous opportunities for optimization in all of these steps, as I explain below.


Free Dummies Books
Continuous Engineering


  • What continuous engineering is
  • How to continuously improve complex product designs
  • How to anticipate and respond to markets and clients
  • How to get the most out of your engineering resources

Get your free book now

Sponsored by IBM

Free Dummies Books
Service Virtualization

Learn to:

  • Define service virtualization
  • Select the most beneficial services to virtualize
  • Improve your traditional approach to testing
  • Deliver higher-quality software faster

Get your free book now

Sponsored by IBM