Debugging a process, component, or crash dump

Attaching to a process by filters

Running a process on Fuchsia is more complicated than in other systems because there isn't a default environment (see “A note about launcher environments” below).

The only way to reliably debug all types of processes is to create a filter via “attach” and start it the normal way you would start that process outside of the debugger. The debugger supports the following filter criteria.

  • Component moniker (v2 components).
  • Component URL (v2 components).
  • Process name, exact match or partial match.

The “attach” command can interpret the input differently according to the format. For example, attach /core/feedback will be interpreted as a component moniker and thus creates a filter that matches all processes in that component, while attach fuchsia-boot:///#meta/archivist.cm will be interpreted as a component URL and could possibly match processes in multiple components.

For convenience, it's allowed to only pass the last part of a component URL to the “attach” command. So attach archivist.cm will function the same as attach fuchsia-boot:///#meta/archivist.cm.

If the input doesn‘t look like some component, it’ll be interpreted as a process name. The launcher of a process could set the process name freely. To check what this is, use “ps” (either in the debugger or from a system shell) with it running.

Note: only the first 31 bytes of the name are included in the Zircon process description.

This example sets a pending breakpoint on main to stop at the beginning of execution, and waits for processes in “my_app.cm” component to start:

[zxdb] attach my_app.cm
Waiting for process matching "my_app.cm".

[zxdb] break main
Created Breakpoint 1 @ main
Pending: No current matches for location. It will be matched against new
         processes and shared libraries.

Then run the process the way you would in normal use (directly on the command line, via fx test, via ffx component run fuchsia-pkg://..., or another way. The debugger should then immediately break on main (it may take some time to load symbols so you may see a delay before showing the source code):

Attached Process 1 [Running] koid=51590 my_app.cmx
🛑 on bp 1 main(…) • main.cc:222
   220 }
   221
 â–¶ 222 int main(int argc, const char* argv[]) {
   223   foo::CommandLineOptions options;
   224   cmdline::Status status = ParseCommandLine(argc, argv, &options);

You can then do basic commands that are similar to GDB:

next
step
print argv[1]
continue
quit

A note about launcher environments

The following loader environments all have different capabilities (in order from least capable to most capable):

  • The debugger's run <file name> command (base system process stuff).
  • The system console or fx shell (adds some libraries).
  • The base component environment via the shell‘s run and the debugger’s run -c <package url> (adds component capabilities).
  • The test environment via fx test.
  • The user environment when launched from a “story” (adds high-level services like scenic).

This panoply of environments is why the debugger can't have a simple “run” command that always works.

Directly launching components

Note: only v1 components can be directly launched in the debugger.

Components that can be executed with the console command run fuchsia-pkg://... can be loaded in the debugger with the following command, substituting your component's URL:

[zxdb] run -c fuchsia-pkg://fuchsia.com/your\_app#meta/your\_app.cmx

Not all components can be launched this way since most higher-level services won‘t be accessible: if you can’t do run ... from the system console, it won‘t work from the debugger either. Note also that fx test is a different environment. According to your test’s dependencies, it may or may not work from the debugger's run command.

Attaching to an existing process

You can attach to most running processes given the process’ koid (the kernel object ID that, when applied to a process, is equivalent to a process ID on other systems). You can get the koid by running ps on the target Fuchsia system or use zxdb's built-in ps command:

[zxdb] ps
j: 1030 root
  j: 1079 zircon-drivers
    p: 1926 driver_host:sys
...

In this listing, “j:” indicates a job (a container for processes) and “p:” indicates a process. The number following the type prefix is the object's koid.

Then to attach:

[zxdb] attach 3517
Process 1 Running koid=1249 pwrbtn-monitor

When you’re done, you can choose to detach (keep running) or kill (terminate) the process.

Attaching to processes in specific jobs

It's possible to limit the scope of a filter to specific jobs. In this case, the filter can be empty which means attaching to all applied processes. For example,

[zxdb] ps
 j: 1033 root
   j: 2057
     p: 2274 foobar.cm
     j: 2301
       p: 2307 foo
       p: 2318 bar

attach -j 2301 foo will only attach to process 2307. attach -j 2057 will attach to all 3 processes.

Debugging drivers

It's not currently possible to set up the debugger early enough in system startup to debug most driver initialization. And since zxdb itself uses the network and filesystem, no drivers associated with network communication can be debuged, and neither can many drivers associated with storage or other critical system functions.

Driver debugging support is tracked in issue 5456.

You can debug running drivers by attaching like any other process (see “Attaching to an existing process”). You can delay initialization to allow time to attach by adding a busyloop at the beginning of your code:

volatile bool stop = false;
while (!stop) {}

To break out of the loop after attaching, either set the variable to true:

[zxdb] print stop = true
true
[zxdb] continue

Or jump to the line after the loop:

[zxdb] jump <line #>
[zxdb] continue

Debugging crash dumps

Zxdb supports loading a minidump generated by a crash report. Use the “opendump” verb and supply the local file name of the dump. The debugger must not be attached to another dump or a running system (use “disconnect” first if so).

[zxdb] opendump upload_file_minidump-e71256ba30163a0.dmp
Opening dump file
Dump loaded successfully.

Now the thread, stack, and memory commands can be used to inspect the state of the program. Use disconnect to close the dump.

It's also possible to use the ffx debug core command, for example:

ffx debug core upload_file_minidump-e71256ba30163a0.dmp

Downloading symbols

To tell zxdb to look for debug symbols for your core dump in a GCS URL and download those symbols automatically, run the following command, substituting the location of your symbols:

zxdb --symbol-server gs://my-bucket-name/namespace

Most users should automatically have the option set, with the server pointed to a bucket containing symbols for all release builds.

The first time you use the symbol server, you will have to authenticate using the auth command. The authentication flow will require you to complete part of the authentication in your browser.

[zxdb] auth
To authenticate, please supply an authentication token. You can retrieve a token from:

https://accounts.google.com/o/oauth2/v2/< very long URL omitted >

Once you've retrieved a token, run 'auth <token>'

[zxdb] auth 4/hAF-pASODIFUASDIFUASODIUFSADF329827349872V6
Successfully authenticated with gs://fuchsia-artifacts-release/debug

Debugging multiple processes

You can debug many arbitrary processes at the same time. Attaching or running when a process is already running in the debugger will just create a new one in parallel.

Recall from the “Interaction model” section you can list the current processes with:

[zxdb] process
  # State       Koid Name
â–¶ 1 Running     1249 pwrbtn-monitor
  2 Not running 7235 pwrbtn-monitor

Select one of those as the default by providing its index (not koid):

[zxdb] process 2

Or apply commands to a specific process (even if it’s not the default) with:

[zxdb] process 2 pause