home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!auspex-gw!guy
- From: guy@Auspex.COM (Guy Harris)
- Newsgroups: comp.arch
- Subject: Re: 64-bit CPU vs 2 x 32-bit CPUs
- Message-ID: <13795@auspex-gw.auspex.com>
- Date: 29 Jul 92 05:59:05 GMT
- References: <13754@auspex-gw.auspex.com> <712296941snx@ananke.stgt.sub.org>
- Sender: news@auspex-gw.auspex.com
- Organization: Auspex Systems, Santa Clara
- Lines: 45
- Nntp-Posting-Host: bootme.auspex.com
-
- > > Unless, for example, your shared libraries are built in
- > > position-independent form; SunOS 4.x/SVR4 shared libraries don't require
- > > system-wide unique virtual addresses, for example.
- >
- >How does SunOS handle references between different shared libraries?
- >Vector tables in memory local to the process?
-
- Yup. Calls made by a shareable object to other routines (even to
- routines that are static to a module, at least with the SunOS bundled
- compiler; hopefully, other compilers are smarter about that) go through
- the Procedure Linkage Table, which is a process-local jump table.
-
- > > Shared memories don't necessarily require them, either.
- >
- >Of course systems can be designed to go without having the same virtual
- >memory address for shared memory (maybe Unixes are designed that way),
-
- The UNIX systems I know of *allow* you to ask that shared memory
- (whether it's SV shared memory or "mmap"ped shared memory) be put at a
- specific virtual address, assuming you know what address to give it
- (said address, of course, must be unused by the process running your
- program). They also allow you to ask the system to choose an address.
-
- > > Umm, the machine running on my desk has a 32-bit virtual address space,
- > > and the OS on that machine supports memory-mapped files.
- >
- > > I would not be surprised if a movable mapping window like that weren't a
- > > pain, though.
-
- My phrasing was a bit more ornate than it should have been, and I ended
- up saying the opposite of what I wanted to say, namely that I wouldn't
- be surprised if having to use a movable mapping window *WAS* a pain -
- i.e., I suspect it *is* a pain.
-
- In any case:
-
- >Easier as moving to 64-bit addresses (well, 43 ot 55 bits or whatever
- >is implemented in the first chips)?
-
- I suspect "easier as" may be a transliterated German usage, with which
- I'm unfortunately unfamiliar; do you mean that the using such a window
- is easier than (to use a English - or, at least, US English - phrasing)
- moving to 64-bit addresses, or that moving to 64-bit chips is easier
- than using a movable mapping window, or that mapping large files will
- get easier as we move to 64-bit addresses?
-