Notes for hacking on Zircon

This file contains a random collection of notes for hacking on Zircon.

Building and testing

To ensure changes don't impact any of the builds it is recommended that that one tests all targets, with gcc and clang, and in both release mode and debug mode. This can all be executed with the buildall script:

./scripts/buildall -q -c -r

From the zircon shell run k ut all to execute all kernel tests, and runtests to execute all userspace tests.

Syscall generation

Syscall support is generated from system/public/zircon/syscalls.abigen. A host tool called abigen consumes that file and produces output for both the kernel and userspace in a variety of languages. This output includes C or C++ headers for both the kernel and userspace, syscall entry points, other language bindings, and so on.

This tool is invoked as a part of the build, rather than checking in its output.

Terminal navigation and keyboard shortcuts

  • Alt+Tab switches between VTs
  • Alt+F{1,2,...} switches directly to a VT
  • Alt+Up/Down scrolls up and down by lines
  • Shift+PgUp/PgDown scrolls up and down by half page
  • Ctrl+Alt+Delete reboots

Kernel panics

Since the kernel can't reliably draw to a framebuffer when the GPU is enabled, the system will reboot by default if the kernel crashes or panics.

If the kernel crashes and the system reboots, the log from the kernel panic will appear at /boot/log/last-panic.txt, suitable for viewing, downloading, etc.

Please attach your last-panic.txt and zircon.elf files to any kernel panic bugs you file.

If there's a last-panic.txt, that indicates that this is the first successful boot since a kernel panic occurred.

It is not "sticky" -- if you reboot cleanly, it will be gone, and if you crash again it will be replaced.

To disable reboot-on-panic, pass the kernel commandline argument kernel.halt-on-panic=true.

Low level kernel development

For kernel development it's not uncommon to need to monitor or break things before the gfxconsole comes up.

To force-enable log output to the legacy serial console on an x64 machine, pass "kernel.serial=legacy". For other serial configurations, see the kernel.serial docs in kernel_cmdline.md.

To enable the early console before the graphical console comes up use the gfxconsole.early cmdline option. More information can be found in kernel_cmdline.md. Enabling startup.keep-log-visiblewill ensure that the kernel log stays visible if the gfxconsole comes up after boot. To disable the gfxconsole entirely you can disable the video driver it is binding to via driver.<driver name>.disable. On a skylake system, all these options together would look something like:

$ tools/build-x86/bootserver build-x86/zircon.bin -- gfxconsole.early driver.intel-i915-display.disable

To directly output to the console rather than buffering it (useful in the event of kernel freezes) you can enable ENABLE_KERNEL_LL_DEBUG in your build like so:

fx set ... --args='zircon_extra_args={kernel_extra_defines=["ENABLE_KERNEL_LL_DEBUG"]}'

There is also a kernel cmdline parameter kernel.bypass-debuglog, which can be set to true to force output to the console instead of buffering it. The reason we have both a compile switch and a cmdline parameter is to facilitate prints in the kernel before cmdline is parsed to be forced to go to the console. The compile switch setting overrides the cmdline parameter (if both are present). Note that both the compile switch and the cmdline parameter have the side effect of disabling irq driven uart Tx.

Changing the compiler optimization level of a module

You can override the default -On level for a module by defining in its build arguments:

opt_level := <n>

Requesting a backtrace

From within a user process

For debugging purposes, the system crashlogger can print backtraces by request. It requires modifying your source, but in the absence of a debugger, or as a general builtin debug mechanism, this can be useful.

#include <lib/backtrace-request/backtrace-request.h>

void my_function() {
  backtrace_request();
}

When backtrace\_request is called, it causes an exception used by debuggers for breakpoint handling. If a debugger is not attached, the system crashlogger will process the exception, print a backtrace, and then resume the thread.

From a kernel thread

#include <kernel/thread.h>

void my_function() {
  thread_print_backtrace(get_current_thread(), __GET_FRAME(0));
}

Exporting debug data during boot

To support testing the system during early boot, there is a mechanism to export data files from the kernel to the /boot filesystem. To export a data file, create a VMO, give it a name, and pass it to userboot with handle_info of type PA_VMO_DEBUG_FILE (and argument 0). Then userboot will automatically pass it through to devmgr, and devmgr will export the VMO as a file at the path

/boot/kernel/<name-of-vmo>

This mechanism is used by the entropy collector quality tests to export relatively large (~1 Mbit) files full of random data.

How to deprecate #define constants

One can create a deprecated typedef and have the constant definition cast to that type. The ensuing warning/error will include the name of the deprecated typedef.

typedef int ZX_RESUME_NOT_HANDLED_DEPRECATION __attribute__((deprecated));
#define ZX_RESUME_NOT_HANDLED ((ZX_RESUME_NOT_HANDLED_DEPRECATION)(2))