On Mon, Jan 14, 2013 at 10:27:21AM +0000, Gleb Natapov wrote:

On Tue, Jan 08, 2013 at 01:38:48PM -0500, Christoffer Dall wrote:

quoted

Add a method (hyp_idmap_setup) to populate a hyp pgd with an
identity mapping of the code contained in the .hyp.idmap.text
section.

Offer a method to drop this identity mapping through
hyp_idmap_teardown.

Make all the above depend on CONFIG_ARM_VIRT_EXT and CONFIG_ARM_LPAE.

Cc: Will Deacon <redacted>
Reviewed-by: Marcelo Tosatti <redacted>
Signed-off-by: Marc Zyngier <redacted>
Signed-off-by: Christoffer Dall <redacted>
---
 arch/arm/include/asm/idmap.h                |    1 +
 arch/arm/include/asm/pgtable-3level-hwdef.h |    1 +
 arch/arm/kernel/vmlinux.lds.S               |    6 +++
 arch/arm/mm/idmap.c                         |   54 ++++++++++++++++++++++-----
 4 files changed, 50 insertions(+), 12 deletions(-)

[...]

quoted

-static void identity_mapping_add(pgd_t *pgd, unsigned long addr, unsigned long end)
+static void identity_mapping_add(pgd_t *pgd, const char *text_start,
+				 const char *text_end, unsigned long prot)
 {
-	unsigned long prot, next;
+	unsigned long addr, end;
+	unsigned long next;
+
+	addr = virt_to_phys(text_start);
+	end = virt_to_phys(text_end);

How does this work with phys addresses greater than 32bit (with
LPAE)? This was the same before the patch too, but I am still
curious. Since __virt_to_phys() returns unsigned long kernel cannot be
put in high memory, right?

Well, AArch32 (arch/arm/) only supports 32-bit virtual addresses by virtue
of the fact that our registers are only 32 bits wide, so we can't
identity-map physical addresses above the 4GB boundary.

You may want to look at the keystone patches from TI for insight about
kernels at high (>32-bit) addresses, although I've not seen any activity
around that for some time now (which is a pity, because the code-patching
stuff was in a good shape).

Will