hmm pud-entry callback locking?

hmm pud-entry callback locking?

[Thomas Hellström (VMware)]

Hi, Jerome,

I was asked by Kirill to try to unify the pagewalk pud_entry and 
pmd_entry callbacks. The only user of the pagewalk pud-entry is 
currently hmm.

But the pagewalk code call pud_entry only for huge puds with the 
page-table lock held, whereas the hmm callback appears to assume it gets 
called unconditionally without the page-table lock held?

Could you shed some light into this?

Thanks,
Thomas

Re: hmm pud-entry callback locking?

[Jerome Glisse]

On Sat, Oct 05, 2019 at 02:29:40PM +0200, Thomas Hellström (VMware) wrote:
> Hi, Jerome,
> 
> I was asked by Kirill to try to unify the pagewalk pud_entry and pmd_entry
> callbacks. The only user of the pagewalk pud-entry is currently hmm.
> 
> But the pagewalk code call pud_entry only for huge puds with the page-table
> lock held, whereas the hmm callback appears to assume it gets called
> unconditionally without the page-table lock held?
> 
> Could you shed some light into this?

I think in my mind they were already unified :) I think easiest thing is
to remove the hmm pud walker, this is not a big deal this will break huge
pud for now, we can re-add this to hmm once you unified them.

Cheers,
Jérôme