[Bug 3533] HotSpot generates code with unaligned stack, crashes on SSE operations

[bugzilla-daemon at icedtea.classpath.org]

https://icedtea.classpath.org/bugzilla/show_bug.cgi?id=3533

--- Comment #18 from Maciej S. Szmigiero <mail at maciej.szmigiero.name> ---
(In reply to Andrew John Hughes from comment #16)
(..)
> On the other hand, distros tend to build binaries for the lowest common
> denominator. These are the Fedora GCC 8 build flags:
> 
> -g -pipe -Wformat -Wno-cpp -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2
> -Wp,-D_GLIBCXX_ASSERTIONS  -fstack-p\
> rotector-strong -grecord-gcc-switches
> -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1
> -specs=/usr/lib/rpm/redhat/redhat-annobin-\
> cc1 -m32 -march=i686 -mtune=generic -fasynchronous-unwind-tables
> -fstack-clash-protection -mcet -fcf-protection -std=gnu++98 -W\
> no-error -fno-delete-null-pointer-checks -fno-lifetime-dse
> 
> https://koji.fedoraproject.org/koji/buildinfo?buildID=1055554
> https://kojipkgs.fedoraproject.org//packages/java-1.8.0-openjdk/1.8.0.161/9.
> b14.fc29/data/logs/i686/build.log
> 
> Oracle's binaries will be built in a similar way, so I suggest the problem
> there is a collision with optimisations in glibc.

If Fedora and Oracle build HotSpot with -march=i686 then GCC shouldn't generate
any SSE operations there on its own.

In this case the problem can only happen due to SSE operations in libraries
called from JVM.
Since glibc is probably the most heavily used external library in the system
SSE code there has the highest chance to trigger this bug.

But this requires that glibc was built with SSE operations in the first place,
so again this does not happen if the whole system was compiled with
-march=i686.

> > > It's also still not clear to me how to reproduce this.
> > 
> > I guess you want to reproduce the actual crash, not just a misaligned stack
> > .
> > You can try to rebuild icedtea with GCC 7.3.0 using
> > {C,XX}FLAGS="-O2 -march=pentium-m".
> > 
> > As far as I can remember, this on its own generated only occasional JVM
> > crashes,
> > for example when compiling a large number of java files.
> 
> I'll look at adding that to a Fedora x86 build. Locally, I've used
> -march=core2 (Intel Core 2 CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3
> and SSSE3 instruction set support) since I installed Gentoo a decade ago and
> not seen problems.

Is your Gentoo setup 32-bit?
64-bit ABI does much less parameter passing on the stack, also the 64-bit
HotSpot code generator often exercises a bit different code paths than the
32-bit one.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/distro-pkg-dev/attachments/20180321/c2d6c6bf/attachment-0001.html>