Linux-insides: The initcall mechanism

Linux-insides: The initcall mechanism(github.com)

76 points by 0xAX 10 years ago | 8 comments

gshrikant 10 years ago |

The Linux-Insides series is a great resource for learning the details of the Linux implementation and hopefully would help more people contribute to Linux by providing more documentation.

An idea I have is to have a similar series for basic device driver structure and making smaller changes in those places to fix minor annoyances and understand better the OS-machine interaction. Does anyone know of a similar (existing) resource (LDD apart, I guess that book is pretty outdated now.)?

[1] Linux Device Drivers - the O'Reilly book by gregkh and Jim Corbet.

rwmj 10 years ago |

This is quite a coincidence. Last week I've been trying to track down why our Linux kernel takes so long to boot (where "so long" = about 1 second), and using "ignore_loglevel initcall_debug" has been a lot of help.

Here is a typical trace showing initcall timings. It's helpful to download it and use `less -r' to view so you can see the ANSI colours:

http://oirase.annexia.org/tmp/min-seabios.txt

riteshkpr 10 years ago | |

can you explain, what change you made and what improvements you observed?

d33 10 years ago |

Absolutely amazing.

One thing that struck my attention is how little comments there are in Linux at times. Consider this:

https://github.com/0xAX/linux-insides/blob/master/Initializa... https://github.com/torvalds/linux/blob/097f70b3c4d84ffccca15...

Not a single line of commentary for something that definitely is not a trivial instruction.

I'm getting the impression that this kind of documentation should be built directly into Linux's source code, creating a document that is self-documenting and both human- and machine-readable.

On a separate note, it would be really nice if Gitlab supported something like ctags for browser-based code navigation...

sytse 10 years ago | |

Great idea to use ctags for code navigation, I created an issue https://gitlab.com/gitlab-org/gitlab-ee/issues/436

d33 10 years ago | | |

Actually I meant Github, but I'd love to see it in Gitlab as well - being able to explore open source projects with ctags using either would be absolutely awesome.

alwaysdownvoted 10 years ago |

From init/main.c:

"The Bourne shell can be used instead of init..."

An easy way to avoid systemd if the user is more comfortable with using a BSD-style /etc/rc approach to starting services?

cyphar 10 years ago | |

The Bourne shell probably doesn't deal with reparented processes properly (at least, that's my experience with working on Docker (PID 1 is a weird process). Not to mention that your shell doesn't propogate signals to children so you'd need quite a bit of shell scripting to make it act like a real init system.

The main reason bash is the fallback init is because it's a good way to fix a borked setup. And "real men don't use init, they just need a root shell" (Linus).