[PATCH] Annotate our UB to sneak it past the sanitizers

[Justin Bogner via llvm-commits]

Alexey Samsonov <vonosmas at gmail.com> writes:
> LGTM. I'd probably keep the amount of annotations to a minimum for now.

r262683. Thanks!

> On Fri, Feb 26, 2016 at 5:09 PM, Justin Bogner <mail at justinbogner.com>
> wrote:
>
>> Alexey Samsonov <vonosmas at gmail.com> writes:
>> > On Fri, Feb 26, 2016 at 2:56 PM, Justin Bogner <mail at justinbogner.com>
>> wrote:
>> >> Alexey Samsonov <vonosmas at gmail.com> writes:
>> >>> What are the failures you observe? We're running an UBSan bootstrap on
>> our
>> >>> buildbot, and it's usually green:
>> >>>
>> >>
>> http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-bootstrap/builds/11059
>> >>
>> >> Is that running without any blacklists?
>> >
>> > Hm, not really. cmake -DLLVM_USE_SANITIZER=Undefined expands to
>> > "-fsanitize=undefined -fno-sanitize=vptr,function",
>> > but this specific error seem to be coming from -fsanitize=object-size.
>> >
>> >> These are both the undefined
>> >> behaviour in ilist_node and how we use it, ie:
>> >>
>> >>   runtime error: downcast of address XXX with insufficient space for an
>> object of type 'llvm::MachineBasicBlock'
>> >>
>> >> This is because "Sentinel" in both of these functions is an
>> >> ilist_half_node, which is certainly not large enough.
>> >>
>> >> It's certainly been necessary to blacklist MachineFunction on darwin
>> >> since at least October, though the bug existed and wasn't caught by
>> >> ubsan before that:
>> >>
>> >>   http://lists.llvm.org/pipermail/llvm-dev/2015-October/091115.html
>> >>
>> >> The one in MemorySSA just started hitting a month or so ago.
>> >>
>> >>> On Thu, Feb 25, 2016 at 2:42 PM, Kostya Serebryany <kcc at google.com>
>> wrote:
>> >>>> Sounds very scary.
>> >>>> At the very least we should specify which flavor of UB we allow
>> >>>> here (not just all of it)
>> >>
>> >> The macro expands to no_sanitize("undefined") right now - does
>> >> no_sanitize take a more specific argument?
>> >>
>> >>> +1 I would prefer LLVM_NO_SANITIZE("kind") macro.
>> >>
>> >> Sure. I'll update the patch.
>> >
>> > Yes, you can use smth. like
>> >   __attribute__((no_sanitize("signed-integer-overflow")))
>>
>> Here's an updated patch. This ones refers to the PR, renames the macro
>> LLVM_NO_SANITIZE() and calls it with "object-size" instead of
>> "undefined".
>>
>> Optionally, I could annotate the 7 other cases of this UB that aren't
>> currently caught by running check-all under ubsan. I'm not sure which
>> way's better - thoughts?
>>
>>