This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Remove unnecessary IFUNC dispatch for __memset_chk.


On Mon, Aug 17, 2015 at 10:01 AM, Alexander Monakov <amonakov@ispras.ru> wrote:
>
>
> On Mon, 17 Aug 2015, H.J. Lu wrote:
>
>> On Mon, Aug 17, 2015 at 9:33 AM, Alexander Monakov <amonakov@ispras.ru> wrote:
>> > On Mon, 17 Aug 2015, H.J. Lu wrote:
>> >> On Mon, Aug 17, 2015 at 8:54 AM, Alexander Monakov <amonakov@ispras.ru> wrote:
>> >> > On Tue, 11 Aug 2015, H.J. Lu wrote:
>> >> >
>> >> >> On Tue, Aug 11, 2015 at 1:30 PM, OndÅej BÃlka <neleai@seznam.cz> wrote:
>> >> >> > On Tue, Aug 11, 2015 at 12:55:31PM -0700, H.J. Lu wrote:
>> >> >> >> On Tue, Aug 11, 2015 at 12:41 PM, Alexander Monakov <amonakov@ispras.ru> wrote:
>> >> >> >> > On Tue, 11 Aug 2015, H.J. Lu wrote:
>> >> >> >> >> > Didn't know its explicitly forbidden to use nonstatic resolvers. Do we document that requirement somewhere?
>> >> >> >> >> > If not then we should add it to documentation.
>> >> >> >> >>
>> >> >> >> >> It makes no senses for IFUNC selector to return the address of
>> >> >> >> >> a global function, which may be preempted at run-time.
>> >> >> >> >
>> >> >> >> > That would be imposing a policy while providing a mechanism.
>> >> >> >> >
>> >> >> >> > I think there are plausible scenarios where returning a global function from
>> >> >> >> > an ifunc resolver would be natural.
>> >> >> >> >
>> >> >> >> > To give a specific example, imagine an ifunc symbol
>> >> >> >> > 'printf_fortified_opportunistically', which resolves to 'printf_chk' if it is
>> >> >> >> > available in the global namespace, or normal 'printf' otherwise.
>> >> >> >> >
>> >> >> >> > Such a symbol could be in any DSO, not necessarily libc, and thus it has to be
>> >> >> >> > able to pick up global symbol definitions.
>> >> >> >> >
>> >> >> >>
>> >> >> >> By the nature of preemption, you have NO control over which definition
>> >> >> >> will be used at run-time.  It can be literally anything, including segfault.
>> >> >> >>
>> >> >> > Could you document that somewhere? Sorry for earlier confusion where I
>> >> >> > thought you meant resolver must be static. Its about that users could
>> >> >> > have similar ideas like making symbol global for silly reasons like
>> >> >> >  be able to profile variant with LD_PRELOAD.
>> >> >>
>> >> >> Something like:
>> >> >>
>> >> >> If the returned function has non-local binding, the run-time behavior
>> >> >> is undefined
>> >> >> as it may be preempted at run-time.
>> >> >
>> >> > I still don't understand why you point that out specifically for IFUNC.
>> >> > Symbol preemption is normal for ELF symbol lookups.  To me it sounds like
>> >> > saying that "with LD_PRELOAD, the run-time behavior is undefined, as some
>> >> > symbols are preempted at run-time".
>> >> >
>> >>
>> >> IFUNC selector should be deterministic at run-time on a given hardware.
>> >> If the function returned by IFUNC selector can be preempted, IFUNC isn't
>> >> the appropriate approach for this use case.
>> >
>> > You appear to be arguing from a standpoint that IFUNC is *exclusively* for
>> > choosing a CPU-specific variant.  Why do you reject the possibility of using
>> > IFUNC for different purposes?  I gave one example earlier.
>> >
>>
>> I said " IFUNC selector should be deterministic at run-time on a given
>> hardware."
>> You can use it for other purpose as long as it is deterministic.  The
>> whole IFUNC
>> implementation on x86 is built on this assumption.
>
> Earlier you rejected the idea that a resolver can use an externally visible
> symbol.  I just don't see why you say it might be non-deterministic.  One
> can inspect the executable, LD_PRELOAD'ed modules, etc. to see whether they
> are going to provide an interposing definition of a symbol or not.  What is
> non-deterministic in there?

Due to the way how IFUNC is implemented, dependency on
external relocation may lead to undefined behavior.  We tried
to support external relocations in IFUNC selector:

https://sourceware.org/bugzilla/show_bug.cgi?id=13302

But external relocation on  function address returned by IFUNC
selector isn't supported.

-- 
H.J.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]