5 ms·

Glibc and the kernel user-space API

52 points by Tsiolkovsky 14y ago

g3orge 14y ago
I always wanted to say that lwn's effort to write good articles (like this one) is shadowed by their weak website. There isn't for example an index of similar articles like this one or introductions to various kernel modules/structures. Maybe everyone but me can find them.
- corbet 14y ago
  LWN kernel index at https://lwn.net/Kernel/Index/ https://lwn.net/Kernel/Index/ Yes, the LWN site needs a lot of work. It's just hard to take the time to do that at the expense of the content itself, but some things are in the works...slowly...
  - g3orge 14y ago
    ok, thanks for the answer. btw, which kernel development book are you currently suggesting??
    - corbet 14y ago
      Linux Device Drivers, of course! Love's "Linux Kernel Development" is less obsolete currently, but I do still think that LDD has a lot of useful information not found elsewhere. Available in all the usual places or downloadable from http://lwn.net/Kernel/LDD3/ http://lwn.net/Kernel/LDD3/
planckscnst 14y ago
I like the idea of a header file with macros distributed with the kernel for providing basic syscall functions. Is there any reason that couldn't work?
- JoachimSchipper 14y ago
  Type-safety is hard with macros; you can't use a macro as a function pointer, for the obvious reason that it's not a function; such macros would tend to trample all over libc functionality, and glibc is not the only libc; the kernel and userland should arguably be separate. It wouldn't be too hard to create (generate) a "libwrapsyscall" that just provided syscalls; if this is worth doing, it's worth doing properly.
  - binarycrusader 14y ago
    I would think that the new generic selection macro capability in C11 would help here: http://www.robertgamble.net/2012/01/c11-generic-selections.html I haven't taken the time to prototype an implementation, but it seems like at least some cases could be solved.
  - throwaway54-762 14y ago
    How about 'static inline' functions distributed along with the libc header files? You get type / number of arguments correctness and no ABI bloat, at the cost of having (minimal) code in header files and per-object code duplication (possibly solved by an intelligent cc/ld).
revelation 14y ago
Classic drepper. Theres a syscall here that maybe the layer between OS and program should expose - Never
- DrJokepu 14y ago
  Didn't Ulrich Drepper retire from glibc development quite a while ago?
  - revelation 14y ago
    Yes, but I wasn't so much conjuring up his response, I'm quoting him directly from the linked bug report :)

52 points by Tsiolkovsky 14y ago

I always wanted to say that lwn's effort to write good articles (like this one) is shadowed by their weak website. There isn't for example an index of similar articles like this one or introductions to various kernel modules/structures. Maybe everyone but me can find them.

corbet 14y ago
LWN kernel index at https://lwn.net/Kernel/Index/ https://lwn.net/Kernel/Index/ Yes, the LWN site needs a lot of work. It's just hard to take the time to do that at the expense of the content itself, but some things are in the works...slowly...
- g3orge 14y ago
  ok, thanks for the answer. btw, which kernel development book are you currently suggesting??
  - corbet 14y ago
    Linux Device Drivers, of course! Love's "Linux Kernel Development" is less obsolete currently, but I do still think that LDD has a lot of useful information not found elsewhere. Available in all the usual places or downloadable from http://lwn.net/Kernel/LDD3/ http://lwn.net/Kernel/LDD3/

planckscnst 14y ago

I like the idea of a header file with macros distributed with the kernel for providing basic syscall functions. Is there any reason that couldn't work?

JoachimSchipper 14y ago
Type-safety is hard with macros; you can't use a macro as a function pointer, for the obvious reason that it's not a function; such macros would tend to trample all over libc functionality, and glibc is not the only libc; the kernel and userland should arguably be separate. It wouldn't be too hard to create (generate) a "libwrapsyscall" that just provided syscalls; if this is worth doing, it's worth doing properly.
- binarycrusader 14y ago
  I would think that the new generic selection macro capability in C11 would help here: http://www.robertgamble.net/2012/01/c11-generic-selections.html I haven't taken the time to prototype an implementation, but it seems like at least some cases could be solved.
- throwaway54-762 14y ago
  How about 'static inline' functions distributed along with the libc header files? You get type / number of arguments correctness and no ABI bloat, at the cost of having (minimal) code in header files and per-object code duplication (possibly solved by an intelligent cc/ld).

revelation 14y ago

Classic drepper. Theres a syscall here that maybe the layer between OS and program should expose - Never

DrJokepu 14y ago
Didn't Ulrich Drepper retire from glibc development quite a while ago?
- revelation 14y ago
  Yes, but I wasn't so much conjuring up his response, I'm quoting him directly from the linked bug report :)