Re: Proposal to fix PGD/PMD lazy deallocation

Benjamin C.R. LaHaise (blah@kvack.org)
Tue, 9 Jun 1998 20:43:35 -0400 (EDT)


On Sat, 6 Jun 1998, Perry Harrington wrote:

> I exchanged a couple of emails with Alan today, he enlightened me
> to a few things. I have an idea to fix the lazy deallocation of
> page tables.
>
> The idea involves 2 changes:
>
> First, we'd add a pointer to the task_struct and allocate a page
> of RAM upon task creation. The page of RAM would be a table
> which retained usage counts for PGD's, making them a fast cinch
> to clean up on the x86 architecture.

Noooooo!!!!! I don't feel like having an extra 240KB of RAM wasted on my
system when it can be easily reclaimed from elsewhere (this works on
architechures where sizeof(pgd/pmd) >= PAGE_SIZE -- this is the case for
all architechures, isn't it?): each pgd/pmd has a corresponding (struct
page), which has an 'atomic_t count'. Now, how about making use of count
in the same fashion as with user level pages? I'm thinking ahead here:
it would be quite helpful for any future where page tables are
swappable...

-ben

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu