home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!crdgw1!rdsunx.crd.ge.com!ariel!davidsen
- From: davidsen@ariel.crd.GE.COM (william E Davidsen)
- Newsgroups: comp.os.linux
- Subject: Re: Memory mapped files?
- Message-ID: <1992Aug13.143944.17119@crd.ge.com>
- Date: 13 Aug 92 14:39:44 GMT
- References: <1992Aug12.215042.5114@murdoch.acc.Virginia.EDU> <1992Aug13.054653.29996@colorado.edu>
- Sender: usenet@crd.ge.com (Required for NNTP)
- Reply-To: davidsen@crd.ge.com (bill davidsen)
- Organization: GE Corporate R&D Center, Schenectady NY
- Lines: 37
- Nntp-Posting-Host: ariel.crd.ge.com
-
- In article <1992Aug13.054653.29996@colorado.edu>, drew@kinglear.cs.colorado.edu (Drew Eckhardt) writes:
- | In article <1992Aug12.215042.5114@murdoch.acc.Virginia.EDU> lfoard@Turing.ORG (Lawrence C. Foard) writes:
- | >Is there any soft of project underway to implement memory mapped files
- | >in Linux? It seems that the shared library code is pretty close except
- | >that it doesn't support writting pages back to the library file.
- | >
- | >Has anyone looked into what would be involved?
- |
- | Ideally, some one will implement something along the lines of a
- | vnode based vm system, where each virtual memory object has a mapping
- | and certain characteristics associated with it.
-
- How about for a start someone fix mmap() to work? As the notes say "it
- works well enough for X," but it sure isn't a generally working
- implementation.
-
- | That mapping could be to a file, to a device, etc. Ideally, a file mapping
- | would be a mapping to blocks as specified by the file's bmap, with those
- | blocks shared with buffercache. Shared libararies / mmap / sysV
- | shared memory semmantics could be implemented on top of this
- | internal structure, quite easily.
-
- There is a question of what mechanism mmap uses now (I don't have
- source code at work).
-
- | Not 4G per process. You still want room for kernel space in the
- | same page table, so you can keep most of the same code in place
- | for kernel<>user space transfers, with kernel and user space in
- | different x86 segments. Linus has something going with
- | > 3G per process, with the remainder allocated to the kernel -
-
- I guess HURD will have 1GB for the user, 1 for the kernel, and 2 for
- the new release of emacs ;-)
-
- --
- bill davidsen, GE Corp. R&D Center; Box 8; Schenectady NY 12345
- I admit that when I was in school I wrote COBOL. But I didn't compile.
-