httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ralf S. Engelschall" <...@engelschall.com>
Subject Linux and Shared Memory.. hell!
Date Mon, 01 May 2000 11:03:05 GMT

I'm not a Linux kernel expert, but perhaps someone of us can shed some
light on me. For MM 1.1.2 I now had to override the Linux decision
hierarchy for determining the shared memory facility, because

    1. Linux since many years since to have a broken MMANON support.
       I never got this running. On a kernel 2.3.99 box it hangs
       the process and causes a "ps" to hang later, too.
       This means MM's --with-shm=MMANON cannot be used.

    2. Linux (kernel 2.0) today showed to dislike even the
       classic mmap() on /dev/zero variant. It allways returned with
       EINVAL, independent how I prepare /dev/zero (ftruncate, lseek,
       write, etc.) or what mapping size I use.
       This means MM's --with-shm=MMZERO cannot be used.

    3. Linux (seems like kernel 2.2 or 2.3) requires a shm filesystem to
       be mounted by the system admin in order to use IPC SHM.
       This means MM's --with-shm=IPCSHM cannot be used all the time.

    4. Linux lacks the POSIX shm_open stuff.
       This means MM's --with-shm=MMPOSX cannot be used at all.

As a result, it looks like only --with-shm=MMFILE is left for using on
Linux - which I dislike, because it requires a temporary file. So, can
this be that the situation is such bad on Linux? Dislikes Linux the way
MM tries to create the shared memory segments under MMANON and MMZERO or
why is it failing? Look at mm_core.c if you need details.

                                       Ralf S. Engelschall
                                       rse@engelschall.com
                                       www.engelschall.com

Mime
View raw message