[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <2d7e424f13682295f655faf9b8c8f1cf@overdrivepizza.com>
Date: Wed, 20 Apr 2022 15:14:02 -0700
From: Joao Moreira <joao@...rdrivepizza.com>
To: Kees Cook <keescook@...omium.org>
Cc: linux-kernel@...r.kernel.org, linux-hardening@...r.kernel.org,
peterz@...radead.org, jpoimboe@...hat.com,
andrew.cooper3@...rix.com, samitolvanen@...gle.com,
mark.rutland@....com, hjl.tools@...il.com,
alyssa.milburn@...ux.intel.com, ndesaulniers@...gle.com,
gabriel.gomes@...ux.intel.com, rick.p.edgecombe@...el.com
Subject: Re: [RFC PATCH 10/11] linux/interrupt: Fix prototype matching
property
>> Fix this CFI policy violation by removing the function pointer union
>> in
>> the tasklet struct.
>
> The good news is that tasklet is on the way out the door[1], so this
> may
> quickly become a non-issue, but also to that end, this fix is hardly a
> problem for a deprecated API...
You are right, sorry for the noise. I looked a bit further and the
problem I saw was actually caused by a compiler bug fusing similar
instructions/basic blocks. It was fixed when I later stumbled on the
problem again and added the following lines (668 and 669 in
llvm/lib/CodeGen/MachineInstr.cpp) to the compiler, but without properly
realizing what was actually behind the previous issue. Hopefully this is
at least a good heads-up about possible pitfalls to other people (@Sami)
implementing CFI in the compiler.
https://github.com/lvwr/llvm-project/commit/0a22ca42877fd156ce95145b11f29c642092dbb7#diff-92843a1f037a9a1e56f92242c4e1746a1166a6b7044ad47a0b4fd2f4b1c6a359R668-R669
Powered by blists - more mailing lists