Most operating systems have implemented realtime signals incorrectly | Hacker News Reader

Most operating systems have implemented realtime signals incorrectly

zinascii.com

105 pointsrzezeski23 comments

Since they all do such a spectacular job of delivering different (buggy) results, I have to believe support was 'roll-your-own' on each platform. There's no reference implementation of POSIX features? No test cases? I guess there are now.

I helped deliver the 1st POSIX implementation, the year it was specified, on an old real-time OS called CTOS (BTOS if you bought it from Burroughs/Unisys). I wonder if we got it right...

There's test suites: http://www.opengroup.org/testing/downloads.html , though I don't know what they cover - posix is highly componentized.

Happy (if you could call it that!) to report AIX 7.1 works correctly :)

    RT sig 50 received with sival 0
    RT sig 50 received with sival 1
    RT sig 50 received with sival 2
    RT sig 51 received with sival 0
    RT sig 51 received with sival 1
    RT sig 51 received with sival 2
    RT sig 52 received with sival 0
    RT sig 52 received with sival 1
    RT sig 52 received with sival 2

Solaris 11 still appears backwards:

    RT sig 43 received with sival 0
    RT sig 43 received with sival 1
    RT sig 43 received with sival 2
    RT sig 42 received with sival 0
    RT sig 42 received with sival 1
    RT sig 42 received with sival 2
    RT sig 41 received with sival 0
    RT sig 41 received with sival 1
    RT sig 41 received with sival 2

I have seen realtime signals before but never quite know what the intended use of them would be. They are relegated in my head to the same category as POSIX message queues -- remember for trivia question purposes.

So anyone used real-time signals and why did you pick them instead of something else?

The intended use case is people trying to write realtime systems - that is, systems with a latency guarantee, like antilock brakes or flight control systems - in UNIX.

However, no sensible person would write such a thing in a general-purpose multi-user operating system. Realtime is hard, see the Toyota mess: http://www.viva64.com/en/a/0083/

"Move fast and break things" is not a great motto for automotive control software.

How about games, where, e.g., sound effects have a maximum acceptable delay w.r.t. the physics that generates them?

Signals are not a great mechanism anyway, it turned out.

Well, any sort of process control too (as in, controlling an assembly line or a chemical plant).

According to POSIX.1b RT signals are used for other features defined in .1b. An excerpt from "POSIX.4 Programmers Guide":

> So far, we've talked about real-time signals sent by the sigqueue function. However, most POSIX.4 signals are sent as a result of three other facilities in POSIX.4. We'll cover these facilities in detail in their own chapters, but for now, you need to know that realtime signals can be generated as a result of expiration of a POSIX.4 timer, completion of asynchronous I/O, and by arrival of a message on an empty message queue.

Linux reserves 2-3 RT signals for use in its threading library:

http://man7.org/linux/man-pages/man7/signal.7.html

I remember studying POSIX message queues in the first OS class I took at the university. Nowadays, I have no I idea if anybody is actually using them.

Minor nitpick: The HP-UX version that you are likely referring to is HP-UX 11i which is also known as HP-UX 11.11. 11.11i is not an HP-UX version number.

As a point of curiosity, this version is still employed in many nuclear plants.

Conclusion: hardly anyone uses realtime signals, and if they do they have no way of testing that they actually get the desired preemption behavior.

(I've used them once to workaround per-thread signal masking not behaving well for non-realtime signals)

I would not have that take-away.. I would instead say that probably most of the users of the signal features described in the article are not concerned with making their code cross-platform.

> (I've used them once to workaround per-thread signal masking not behaving well for non-realtime signals)

Do you remember what the misbehavior was? I ask because I actually work with these kinds of things.

> The kernel delivers the the signals in the correct order only to have userland invert them on the stack.

So does it mean that it's not a kernel bug at all? Just the libc? Then it's not surprising that different kernels, all using the same libc, have the same bug?

These different kernels do not use the same libc. In particular, BSD derivatives are using a BSD libc, while Linux and friends are using glibc (or some cousin). I'm not certain which libc the others are using, but they are likely separate forks as well.

All kernels do not use the same libc. In fact, Linux is different in the fact that it doesn't ship a libc with the kernel. Illumos, the OS I investigated, goes by the philosophy that you _have_ to ship libc with your kernel, and thus they have their own libc.

My diagnosis for Illumos may not hold up on other OSs, they may invert priority for other reasons. In any case the spec is fairly clear that a signal is considered delivered when the "appropriate action" is taken. In this case the appropriate action is to invoke the user's handler, which is above libc on the stack.

Finally, in case you missed it, I witnessed 4 unique orderings across the systems I tested. They were not the same across the board.

Here's another POSIXism that everybody gets wrong: command lookup for sh; follow this algorithm for executing a command with no slashes, consider what would happen if "echo" is not in your path. Try it out on any shell and see that it just runs the "echo" builtin despite it being required to return 127.

Text below taken from: http://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3...

If the command name matches the name of a special built-in utility, that special built-in utility shall be invoked.

If the command name matches the name of a function known to this shell, the function shall be invoked as described in Function Definition Command. If the implementation has provided a standard utility in the form of a function, it shall not be recognized at this point. It shall be invoked in conjunction with the path search in step 1d.

If the command name matches the name of a utility listed in the following table, that utility shall be invoked.

alias bg cd command

false fc fg getopts

jobs kill newgrp pwd

read true umask unalias

wait

Otherwise, the command shall be searched for using the PATH environment variable as described in XBD Environment Variables :

If the search is successful:

<snip>

If the search is unsuccessful, the command shall fail with an exit status of 127 and the shell shall write an error message.

Step 1.d.i.a:

If the system has implemented the utility as a regular built-in or as a shell function, it shall be invoked at this point in the path search.

1.d.i only happens if the search in PATH is successful.

I think the logic behind this was to ensure that as shells add more built-ins scripts that depend on an executable failing when not in the path won't break.

Quoting from your post:

"If the command name matches the name of a special built-in utility, that special built-in utility shall be invoked."

Let echo be a "special built-in utility", and sh is compliant.

"special built-in utilities" are specifically listed elsewhere. echo is not among them:

http://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3...

A lawyerly reading still permits echo to qualify as a special built-in utility.

Two relevant excerpts: "The following "special built-in" utilities shall be supported in the shell command language ..."

"... The term "built-in" implies that the shell can execute the utility directly and does not need to search for it. An implementation may choose to make any utility a built-in; however, the special built-in utilities described here differ from regular built-in utilities ..."

(the two differences are irrelevant to this discussion as they do not mention PATH resolution behaviors)

These quotes taken together do not preclude the shell adding a built-in utility and assigning it the "special" status. The standard does not say that only the listed utilities are special; it says that the listed utilities shall be special. Adding a normal built-in utility (not "special") not on the list is explicitly condoned.

It is my belief that echo being a "special built-in utility" and thus having magic PATH behavior is POSIX compliant, although the standard is not 100% clear on this.

Reply on news.ycombinator.com