Hi Andy,
On Thu, Sep 03, 2020 at 07:52:35AM -0700, Andy Lutomirski wrote:
> Does this mean we can get rid of arch_sync_kernel_mappings()? Or
> should we consider adding some locking to make it non-racy again?
Well, removing arch_sync_kernel_mappings() would mean to re-introduce
vmalloc_sync_all() calls all over the place, I am not in favour for
that.
I also thought about locking, but that is not easily doable without
destroying performance/scalability of the vmalloc alloc/free path for
other architectures too. It _could_ be done, but the effort is large and
touches a lot of generic page-table allocation code just for x86-32.
This seemed not worth it while thinking about it.
Regards,
Joerg
