I would suggest you go back and read my original email once more, now that
you realize that you had simply not understood the difference between
physical page pinning and virtual page pinning.
Seriously.
Now that you understand why I called the code buggy, maybe you realize
that calling the code "insane and misdesigned" is actually not overly
rude: it's just an accurate representation of the state of the code.
And if you read the mail once more, you'll also notice that every single
derogatory remark was about the _code_, not you.
Oh, and I did ask you for an explanation for why we shouldn't just remove
it. There can't be all that many users.
Because quite frankly, if you apparently want to keep the vma around, the
code is going to get way more complex and ugly. You may be able to avoid
some of the _worst_ crap if you require that user pointers have to always
be u64-aligned. Yes, that's a very ugly and non-intuitive requirement for
a read() interface, but probably better than the alternative.
Or maybe just do the double buffering, and limiting pagemap reads to
fairly small chunks at a time.
Linus
--
