Naive question about grant/map

[Joshua LeVasseur]

On Dec 4, 2003, at 21:41, Jonathan S. Shapiro wrote:

>
> I am looking for a sequence in which A maps to B and B ??? to C and the
> result would look like:
>
>    A -> B
>    A -> C
>
> Is there such an operation?
>
> Also, I forgot to ask: is there some reason why such an operation would
> be inadvisable?


If you look at the mapping database as a *cache*, then your scenario 
becomes one which describes the cache behavior, and thus isn't so 
important in light of the semantics of user-level correctness.  In any 
cache, it is possible for the entries to disappear, and they must be 
resupplied when needed later.  The page fault IPC informs of cache 
misses in the mapping database.  Upon page fault, the mappings must be 
reconstructed.

The grant operation can be viewed as an operation which removes an 
entry from the cache.  If B accesses the mapping which it just dropped 
(via granting to C), A will receive a page fault.  And at user-level, A 
should know that it has mapped to B, and thus can decide to resupply 
the mapping to B.

When you consider the mapping database from the cache perspective, it 
isn't so upsetting to have an undirected unmap operation.  An unmap 
brutally removes all entries, and thus it becomes necessary to 
repopulate the cache.  And obviously, to repopulate the cache, A must 
know about its outstanding mappings.  And this is where the cache model 
can break down; how can A know that it should have a mapping to C, 
after B granted the mapping to C?  A user-level protocol must be used 
to inform A that it has the duty of maintaining mappings to C.  But the 
grant operation probably should be carefully used; more in the case to 
support transparent interposition, so that A thinks it maps to Y, but 
really maps to Z due to some indirection.  And then Z transparently 
forwards the mapping to Y via grant(), to construct the original 
mapping hierarchy intended by A.

-Josh