Re: Current status about arm64 livepatch support

[Cc-ing live-patching mailing list which might also be interested in the 
progress of arm64 support]

On 3/13/20 12:22 PM, Mark Rutland wrote:

On Fri, Mar 13, 2020 at 10:27:23AM +0800, Xiao Yang wrote:

quoted

Hi,

Ping.

Best Regards,
Xiao Yang

On 2020/3/4 15:18, Xiao Yang wrote:

quoted

Hi Torsten,

Sorry to bother you.

I focus on arm64 livepatch support recently and saw that you have tried
to implement it by:
-------------------------------------------------------------------------------
http://lists.infradead.org/pipermail/linux-arm-kernel/2018-October/609126.html
http://lists.infradead.org/pipermail/linux-arm-kernel/2018-October/609124.html
http://lists.infradead.org/pipermail/linux-arm-kernel/2018-October/609125.html
-------------------------------------------------------------------------------

This patch set seems to be blocked because of some issues, but your
another patch set inlcuding the first one "arm64: implement ftrace with
regs" has been merged into upstream kernel:
-------------------------------------------------------------------------------
http://lists.infradead.org/pipermail/linux-arm-kernel/2019-February/631104.html
http://lists.infradead.org/pipermail/linux-arm-kernel/2019-February/631107.html
http://lists.infradead.org/pipermail/linux-arm-kernel/2019-February/631105.html
http://lists.infradead.org/pipermail/linux-arm-kernel/2019-February/631106.html
http://lists.infradead.org/pipermail/linux-arm-kernel/2019-February/631114.html
--------------------------------------------------------------------------------

Could you tell me current status about arm64 livepatch support?
For example:
1) Are you(or someone) still working on arm64 livepatch support?
2) Are there some unresolved problems about arm64 livepatch support?
     What are they?
3) Will you send a newer version for arm64 livepatch support recently?

1) I beleive a few people are working on portions of this.

2) I believe that some work is necessary.

    Julien Thierry has done some work on objtool, which is necessary to
    check ensure that sequences (including assembly functions) manipulate
    the stack, and calls/returns as we expect. Mark Brown has been
    converting our assembly to use modern annotations which objtool
    consumes when checking this.

I've recently started working on the arm64 objtool again and saw the 
work to use new annotations by Mark B. which is very helpful, thanks for 
that. I've rebased the objtool work on them and working on solving the 
new/remaining objtool warnings.

I've also reworked the arm64 decoder. I'm not sure yet when I'll be able 
to post a new version but it's coming!

    There might be additional assembly work necessary for this, depending
    on any deecisions we make for objtool.

    For reliable stack tracing we may need to rework some assemvly and/or
    rework the stack tracing code. That will likely depend on the objtool
    bits.

There is one thing I'll be introducing in the next arm64 objtool 
patchset which are unwind_hints (inspired from 
arch/x86/include/asm/unhind_hints.h) which are annotation indicating in 
which state we expect the stack to be when entering assembly code or 
fiddling with stack registers in the middle of assembly code.

I haven't finished the work on that yet.

Cheers,

-- 
Julien Thierry