Re: [uml-devel] swap entry and newpage bit

[Jeff D. <jd...@ka...>]

rc...@im... said:
> 4. the page that this pte was previously set is added to swap cache
> indexed by the --> 0x3307c <--  address, ie, page->index = 0x3307c. 

This seems like a bug.  The low pte bits are arch-specific, and as far as
you're concerned, they should be considered random.  The arch can fiddle
those bits at any time for any reason.  Why aren't you using 
page->index == SWP_OFFSET(pte)?

> 4. we read this page from disk. The correct block on disk is computed
> correctly since SWP_TYPE and SWP_OFFSET macros strips out those last
> two bits, but it returns _bogus_ data. 

Because the page had never been written out to disk in the first place, I
take it?

> Only user-level apps have ptes set to swap entries, so in order to one
> of these ptes to fault an address, it must have run for a while.

At the very least, flush_tlb_page should have been called when the pte
was marked as being swapped.  That should have unmapped the page and cleared
the newpage bit.

				Jeff