Re: how does mmap() work?

["Mark S. Waterbury" <mark_s_waterbury@xxxxxxxxx>]

Might I ask, why do you "care" how the mmap() API works?  Is it not enough

to know the purpose of the API, the parameters (the interface), and the results

that it provides, in order to make good use of this facility?

 

If you really want to know how a particular mmap works, you can always look

at the source code for any Linux kernel to see how it is implemented.  In effect,

they manipulate the page tables to "map" the contents of the file into the virtual

address space of the running process.

 

Since OS/400 is vastly different, primarily because there are no separate

address spaces "per process", but instead, there is a single large address space,

the so-called "single level store", we can speculate that, somehow, OS/400 would

implement mmap() as follows: it would first create a teraspace into which to

"map" the contents of the file, and then, somehow, set up the corresponding

page table entries to map the relevant data. Now, again, not knowing exactly

how this is implemented "under the covers", this could mean that OS/400 must

also set up the page tables so that, upon "first reference" to each page of data,

the appropriate data from the actual underlying file can be copied into the relevant

pages in virtual storage, etc.  Also, I do not know if this mmap() API can work

with the old "DB2" file system (aka. the QSYS file system files).  Since it is a

"Unix" API, it probably only works for IFS stream files.

 

I hope that helps...

 

Again, why do you "care" about "how it works" versus "what it does" for you?

 

Regards,

 

Mark S. Waterbury

----- Original Message -----

From: Steve Richter

To: MI400@midrange.com

Sent: Tuesday, May 22, 2001 11:12 AM

Subject: Re: how does mmap() work?

 

 

Are teraspaces implemented the same way?

 

====> we are still waiting for Alexei to tell us how.

MY guess: we won't get any info.

 

I think he said he did not know.  I hope he or someone else can tell us who does know.