kvm pveoi support

The feature allowing reuse of an inherited constructor is not available
on in gcc 4.7.

kvm provides a way to EOI without exiting; detect its availability and use it.

Works with xapic, and allows us to hook eoi().

kvm's pv_eoi functionality fails if any vmexits are taken before the EOI
(likely due to a bug).  Work around this by deferring the handler until after
the interrupt has been EOIed.  Since level-triggered interrupts must be acked
at the device level prior to the EOI, add an additional handler to be called
prior to the EOI.

Make sure percpu data is aligned correctly as some hardware features expect it.

RhinoCLI relied on a bunch of global Java variables such as _cx and _args.
This was not only ugly, it also prevents us from running multiple instances
of the CLI on the same JVM - e.g., to support multiple telnet connections.

There isn't actually a need for these JVM-wide global variables. At most,
we need to global use variables in the Javascript interpreter (so each
instance of the interpreter would have its own copy).

This patch puts main's arguments in a new global-per-javascript-interpreter
variable "mainargs" instead of the global-for-entire-JVM _args. "_cx" isn't
needed at all: one of uses was to for Java to convert a String[] into the
equivalent Javascript alternative - but Java should just return String[]
and let Javascript worry about handling that (it seems to work just fine
without change). A second use was for returning an exit code, but a more
appropriate methods to do the same thing without global variables exist.

Based on FreeBSD virtio code
Provides a x7 boost for rx netperf

allows the dhcp worker to see incoming dhcp packets using a simple hook.
no reason to make this kind of hook generic (a BPF alternative) as there
are already many protocol specific hooks in the networking stack.

if we ever need to do more hooking, the pfil interface may be used or we
can implement a generic solution, but it's not reasonable to do so for
a single user (dhcp).

Use DHCP to discover an IP address per each interface, DHCP packets are being hooked
in the networking stack in ip_input and queued for deffered processing using a
dhcp worker thread. Sending dhcp packet is done directly over ethrernet (building
IP and UDP).

There's still alot to be done: setting up lease time, timeouts, more error handling
but it's possible to implement these later on.

Facilities for capturing call stack on arbitrary tracepoints.

New command: 'perf callstack <tracepoint>'

Although we are not expecting to have many interfaces, it is still useful
to print the interface name when we fail to set its parameters. (For starters,
we may very well have plenty, and it is also good for debugging)

Replace the lambda-based with_lock API with a block-based API.

The current with_lock() has the downside of using relatively large lambdas.
Lambdas need extra syntax (and thought) to capture variables, and change
the meaning of 'return', 'break', and 'continue'.

The new with_lock (named WITH_LOCK, since it's a macro) accepts a statement
instead of a lambda:

   WITH_LOCK(mtx) {
       do_something();
       WITH_LOCK(preempt_lock) {
          do_something_else();
       }
   }

In a WITH_LOCK context, any return, break, continue, or exception which exits
the block will cause the lock's unlock() method to be called.

  sprintf(fmt, ...) - returns a std::stream
  fprintf(os, fmt, ...) - prints to a std::ostream&

Easier than the usualy method of constructing a temporary ostringstream
and formatting into that.

Use:

  perf callstack tracepoint

to list frequent callstacks for a tracepoint (from 'perf list')

The collector can attach to a tracepoint and collect backtraces; later
the traces can be dumped as a histogram.

Needed for safe backtrace.

A backtrace() implementation which is safe for use in wierd contexts like
interrupts.  Needs -fno-omit-frame-pointer, but doesn't crash if some object
is compiled without it.