Re: 2nd posting: exceptions and asm() blocks on x86

["John S. Yates, Jr." <jyates at netezza dot com>]

"Falk Hueffner" <falk.hueffner@student.uni-tuebingen.de> writes:
>
> "John S. Yates, Jr." <jyates@netezza.com> writes:
> >
> > If your explanation is correct, that is that a call to an
> > exception-raising, noreturn function is viewed as changing
> > control flow, then what is gcc's stance on using
> > instructions that raise exceptions within asm() blocks?
>
> Is it possible that you're confusing C++ exception and
> hardware traps?  Those two concepts are in no way related with
> g++. Some C++ compilers catch hardware traps and make C++
> exceptions of them, but g++ doesn't.  Therefore, you cannot
> throw C++ exceptions from asm at all.

The fact that g++ does not on its own turn arithmetic traps into
C++ exceptions does not preclude a user from doing so.  This is
exactly what we do in the system on which I work:  We implement
multi-terabyte databases including billions of rows on massively
parallel hardware by converting SQL queries into C++, compiling
the result and broadcasting the generate binary to all nodes.
Given arbitrary content in a database and arbitrary aggregation
within a query one has to expect exceptional cases.  When these
arise it is not appropriate to crash the system or to simply kill
a process.  We have to abort the query, roll back transactions,
etc.  The only sane way of doing this is to turn all hardware
traps into an exceptions.


> "Falk Hueffner" <falk.hueffner@student.uni-tuebingen.de> writes:
>
> It's an expected limitation. asm's are not supposed to
> change the control flow. The manual could probably more
> explicit about this...

I find this assertion hard to believe.  The gcc manual shows
an explicit example of using an asm() to make a call.  If your
assertion is truly gcc's reasoned design intent then it implies
that any author of an asm() containing a call is responsible
for ensuring that the transitive closure of all code reachable
from that call is and remains free of exception possibilities
for all time.  Surely such drastic consequences from a design
choice would have merited a cautionary comment of some sort...

I therefore put it to you that this is not simply a deficiency
in gcc documentation but rather a legitimate bug.  True, a fix
might require additional syntax...  But as things stand asm()
is a festering gotcha.


> "Falk Hueffner" <falk.hueffner@student.uni-tuebingen.de> writes:
>
> Maybe you could try to generate the required .eh_frames
> manually, I think the Linux kernel does this somewhere.

I am desperate and willing to try anything.  Searching the web
based-on your suggestion I found the following article:

http://www.tldp.org/LDP/khg/HyperNews/get/devices/exceptions.html

I believe I could indeed do something similar if I knew enough
about the contents of .eh_frames.  So far though I have had no
luck looking for documentation.  Any suggestions?

/john
--
John S. Yates, Jr.   508 665-6897 (voice)
Netezza Inc          508 665-6811 (fax)
200 Crossing Blvd.   Framingham, MA 01701