|| ||Jeremy Fitzhardinge <email@example.com>|
|| ||Ingo Molnar <firstname.lastname@example.org>|
|| ||[PATCH 0 of 7] x86: lay groundwork for Xen domain 0 support|
|| ||Sun, 07 Sep 2008 15:21:12 -0700|
|| ||email@example.com, "H. Peter Anvin" <firstname.lastname@example.org>,
Xen Devel <email@example.com>,
Andi Kleen <firstname.lastname@example.org>|
This series begins to lay the groundwork for Xen domain 0 support.
Domain 0 is much like a normal Xen domain, but it is also allowed to
have direct access to the underlying hardware (including both IO space
and various BIOS tables), so it can run device drivers, etc. This
means that we need to be able to distinguish between a Xen domain's
normal pseudo-physical address space, and the real machine physical
- Adds a new pte flag - _PAGE_IOMAP - used to indicate that a mapping
is intended to map an IO device, and the physical address is
actually a real machine physical address rather than a
pseudo-physical address. This is exposed as PAGE_KERNEL_IO and so
on. ioremap() and early_ioremap() are modified to set this flag,
as they (should) always be used to map IO devices and not other
memory. By default __supported_pte_mask masks this flag out, so it
won't end up in the final pagetable. The Xen (and any other) code
which cares about this flag unmask it from __supported_pte_mask.
- Add early_memremap() to deal with the cases where early_ioremap()
actually being used to map normal memory.
- Remove a bogus check in x86-64's implementation of
set_pte_vaddr_pud(), which prevents memory mappings from being
- Convert __acpi_map_table to use early_ioremap(), rather than have
its own implementation.
- Make __acpi_map_table always map the memory rather than assuming
that the linear map maps some of the acpi tables. This won't be
true in the virtual case, and always mapping doesn't hurt in the
I've tested these patches on both 32 and 64 native booting, and it all
works for me.