Re: [PATCH 3/5] fork: Remove assumption that vm_area->nr_pages equals to THREAD_SIZE

[Linus Walleij <linus.walleij@linaro.org>]

On Wed, May 7, 2025 at 3:53 AM Andrew Morton <akpm@linux-foundation.org> wrote:

> > -     BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
> > -
> > -     for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
> > +     for (i = 0; i < vm_area->nr_pages; i++) {
>
> Well, constants are fast and small.  I assume this will add a few
> instructions and a bit of runtime overhead?

That is true.

Pashas original patch has a local variable instead of comparing
vm_area->nr_pages, so I went back to that to ascertain we
are not dereferencing the struct on every iteration of the
loop, but will instead use a register.

For this first instance the assignment of that variable should be no
more time consuming than the deleted BUG_ON() dereference so
it is +/- 0.

For the other two instances I like the readability this gives, the
local variable restored there too.

Maybe those two should have also has BUG_ON() guards
actually. vm_area->nr_pages should be the trusted, run-time
value.

I'll resend what I have so you can decide.

Yours,
Linus Walleij