Hi Jan,
On 21/01/2020 11:02, Jan Beulich wrote:
On 21.01.2020 10:49, Wei Xu wrote:
Add __acpi_unmap_table function for ARM and invoke it at acpi_os_unmap_memory
to make sure the related fixmap has been cleared before using it for a
different mapping.
How can it possibly be that this is needed for Arm only?
Let me give some background (I will let Wei dealing with the rest of the
patches). On Arm, I made the decision to forbid a mapping replacement in
the page-tables code. This means that if you want to re-use the same
fixmap, then you need to clear it first.
The reason for the requirement is quite simple. On Arm, you need to use
a break-before-make sequence any time you replace a valid entry by
another valid entry (there is a couple of case where it is not needed).
As the sequence name suggests it, we will have a small window where the
virtual address will point to nothing. This may result to an abort if
another CPU is accessing the address at the same time.
In the fixmap case below, this should never happen. But now imagine
shattering a superpage...
So rather than trying to allow in some cases the modification of a
mapping, we just forbid for everything but permission changes. This is
much simpler to reason and a much saner interface.
Similarly, I think this is much saner to call have a clear_fixmap() for
each set_fixmap().
Cheers,
--
Julien Grall
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel
