Re: allocate_string_data memory corruption

[Stefan Monnier]

>> He has a certain Emacs Lisp setup that he can use to crash Emacs
>> reproducibly, when hyperthreading is turned on.  Without
>> hyperthreading, he has gotten one or two strange bus errors, but these
>> seem to be difficult to reproduce, and we're not sure if they are
>> real.

> Hmm... bus errors are quite uncommon on i386, are you 100% sure that
> we're looking at a software problem here?  It's possible that this
> user's machine has faulty RAM chips that cause random memory corruption,
> and that HyperThreading worsens the problem because it increases memory
> pressure.

> Can the problem be reproduced on several machines?  If not, could the
> user run memtest86+ on that one machine?

> If this is a problem specific to HT, it strikes me as suspicious that we
> got only *one* report given the number of people who use the CVS version
> of Emacs.

> (For the record, my main work machine has HyperThreading and Emacs
> hasn't crashed there for many months now.)

I still believe in the race-condition-with-signal-handler, because I'm
absolutely sure there are bugs left in that area (which is why I came up
with the SYNC_INPUT patch in the first place).


        Stefan