Re: Message passing in user-land

[rreale]

On Tue, Jul 16, 2002 at 12:44:07PM -0700, James Morrison wrote:
> 
> 
>  Ok, so it looks like every process maintains it's own set of queues.  So
> I am a server process and I setup 3 queues for clients to write in.  Now,
> how do the clients know where to write the data so I can find it?  Previously,
> you said it would be the memory managers job.  So, I write to some pre-defined
> queue of the memory manager, to say people can write msg's to me.  Now, any
> process can ask the memory manager where to put msg's to me.  If this is what
> you are thinking, it sort of makes sense.  But assumming we have a 
> process/task
> boundary where each process has it's own virtual address space how can it 
> break
> this barrier without the kernel's help?  I can sort of wrap my brain around
> this,
> but I keep comming back to crazy libc hacks.

You are right, but we need the kernel's help only once during the whole
process' lifetime, i.e. when the new process is created and the shared 
memory area chosen by the memory manager.  During the process creation,
the address of this area could be written in some well-known location
of the process' address space.  This way there's no need for the process
itself to query the memory manager each time it wants to send a message.

Roberto