arm64: mm: Flip kernel VA space
authorSteve Capper <steve.capper@arm.com>
Wed, 7 Aug 2019 15:55:14 +0000 (16:55 +0100)
committerWill Deacon <will@kernel.org>
Fri, 9 Aug 2019 10:16:51 +0000 (11:16 +0100)
commit14c127c957c1c6070647c171e72f06e0db275ebf
tree6f5e57a4dd8c8cbe895b01513c2bd6a1cb3ddcff
parent9cb1c5ddd2c432834dd4f40c0170d6b639e8e5c3
arm64: mm: Flip kernel VA space

In order to allow for a KASAN shadow that changes size at boot time, one
must fix the KASAN_SHADOW_END for both 48 & 52-bit VAs and "grow" the
start address. Also, it is highly desirable to maintain the same
function addresses in the kernel .text between VA sizes. Both of these
requirements necessitate us to flip the kernel address space halves s.t.
the direct linear map occupies the lower addresses.

This patch puts the direct linear map in the lower addresses of the
kernel VA range and everything else in the higher ranges.

We need to adjust:
 *) KASAN shadow region placement logic,
 *) KASAN_SHADOW_OFFSET computation logic,
 *) virt_to_phys, phys_to_virt checks,
 *) page table dumper.

These are all small changes, that need to take place atomically, so they
are bundled into this commit.

As part of the re-arrangement, a guard region of 2MB (to preserve
alignment for fixed map) is added after the vmemmap. Otherwise the
vmemmap could intersect with IS_ERR pointers.

Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Steve Capper <steve.capper@arm.com>
Signed-off-by: Will Deacon <will@kernel.org>
arch/arm64/Makefile
arch/arm64/include/asm/memory.h
arch/arm64/include/asm/pgtable.h
arch/arm64/kernel/hibernate.c
arch/arm64/mm/dump.c
arch/arm64/mm/init.c
arch/arm64/mm/kasan_init.c
arch/arm64/mm/mmu.c