Re: x86_64 RIP-relative addressing bug

[Roland McGrath <roland at redhat dot com>]

> You're very likely right about the relative hassles, I'm certainly no
> expert.  On this single point though, isn't it true that it would be
> sufficient to copy & modify only the parent nodes of the specific page
> table leaf node, and not the entire rest of the tree?  

Yes, that is what I meant in fact.  However, note that the kernel part of
the table usually uses big pages, which means contiguous 2MB regions
indicated by special leaf entries in what would ordinarily be the slot for
the parent of the leaves.  So, either you copy 2MB regions to tweak the one
byte, or else as you copy & mutate to reach the 4KB leaf page you need to
change, you have to allocate the new intermediate level of page tables.
But I said I wasn't going to get into the gory details, and now I've only
left out two or three of them. ;-)