Running tests is now done by editing the command line in build.mak

With virtio we must use dynamic allocation for the correctness of
virt_to_phys

Now that we actually look for DT_NEEDED libraries in the right place we
fail loading it because we don't support STT_GNU_IFUNC symbols.  But
by now we already have a feature complete libm implementation in our
tree so we don't actually need it anymore.

They don't cause problems for other tests anymore.

One test so far: java.lang.System.nanoTime()

We're only about 500% slower than native.

Note that the glue code to set up the class path is written in Java.

Needed by Java formatter.

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

they seem to break running new tests after them.

The test code assumes /tests contains shared objects.

This standardizes our method of running a payload: run a shared object with
argc/argv.  No special treatment for the jvm.

breaks the build

After my recent test suite changes tests are expected in /tests

Style requires manual edititing of loader.cc to run, no automated way
to pick up whole directories in the manifest, and we won't clear the
environment before running the next .so

Mirror the jre file layout closely, otherwise it gets confused.

Add a new hello_world test payload that can be used to test OSv functionality
more fine grained.  For now the binary loaded remains hardcoded in loader.cc
and still defaults to libjvm.so.

Change the directory structure a little to conform with jvm expecations
(removing server/ from libjvm.so).

Don't load it from the glibc library.