[PATCH v2 12/13] arm64: add support for relocatable kernel

On 8 January 2016 at 11:17, James Morse [off-list ref] wrote:

Hi Ard!

On 30/12/15 15:26, Ard Biesheuvel wrote:

quoted

This adds support for runtime relocation of the kernel Image, by
building it as a PIE (ET_DYN) executable and applying the dynamic
relocations in the early boot code.

Signed-off-by: Ard Biesheuvel <redacted>
---

quoted

diff --git a/arch/arm64/kernel/head.S b/arch/arm64/kernel/head.S
index 01a33e42ed70..ab582ee58b58 100644
--- a/arch/arm64/kernel/head.S
+++ b/arch/arm64/kernel/head.S

@@ -243,6 +253,16 @@ ENDPROC(stext)
 preserve_boot_args:
      mov     x21, x0                         // x21=FDT

+#ifdef CONFIG_ARM64_RELOCATABLE_KERNEL
+     /*
+      * Mask off the bits of the random value supplied in x1 so it can serve
+      * as a KASLR displacement value which will move the kernel image to a
+      * random offset in the lower half of the VMALLOC area.
+      */
+     mov     x23, #(1 << (VA_BITS - 2)) - 1
+     and     x23, x23, x1, lsl #SWAPPER_BLOCK_SHIFT
+#endif

I've managed to make this fail to boot by providing a seed that caused
the kernel to overlap a 1G boundary on a 4K system.

Ah, yes. Thanks for spotting that.

(It looks like your v3 may have the same issue - but I haven't tested it.)

Yes, it does. It probably makes sense to sacrifice some entropy bits
and simply round up the kaslr offset to a log2 upper bound of the
kernel Image size, rather than hacking up some logic in assembly to
test whether we are crossing a PMD/PUD boundary

quoted

+
      adr_l   x0, boot_args                   // record the contents of
      stp     x21, x1, [x0]                   // x0 .. x3 at kernel entry
      stp     x2, x3, [x0, #16]

Thanks!

James