Re: [RFC PATCH 1/1] arm64: implement live patching

On 6/17/21 4:29 AM, nobuta.keiya@fujitsu.com wrote:

quoted

It's my understanding that the two pieces of work required to enable live
patching on arm are in flight upstream;
- Reliable stack traces as implemented by Madhavan T. Venkataraman [1]
- Objtool as implemented by Julien Thierry [2]

This is the remaining part required to enable live patching on arm.
Based on work by Torsten Duwe [3]

Allocate a task flag used to represent the patch pending state for the
task. Also implement generic functions klp_arch_set_pc() &
klp_get_ftrace_location().

In klp_arch_set_pc() it is sufficient to set regs->pc as in
ftrace_common_return() the return address is loaded from the stack.

ldr     x9, [sp, #S_PC]
<snip>
ret     x9

In klp_get_ftrace_location() it is necessary to advance the address by
AARCH64_INSN_SIZE (4) to point to the BL in the callsite as 2 nops were
placed at the start of the function, one to be patched to save the LR and
another to be patched to branch to the ftrace call, and
klp_get_ftrace_location() is expected to return the address of the BL. It
may also be necessary to advance the address by another AARCH64_INSN_SIZE
if CONFIG_ARM64_BTI_KERNEL is enabled due to the instruction placed at the
branch target to satisfy BTI,

Signed-off-by: Suraj Jitindar Singh <redacted>

[1] https://lkml.org/lkml/2021/5/26/1212
[2] https://lkml.org/lkml/2021/3/3/1135
[3] https://lkml.org/lkml/2018/10/26/536
---

AFAIU Madhavan's patch series linked in the above [1] is currently awaiting
review by Mark Rutland. It seems that not only this patch series but also the
implementation of arch_stack_walk_reliable() at the below link is required
to enable livepatch.

Yes. I have a patch ready for that. But I can submit that only after the previous
series has been accepted.

Thanks

Madhavan