RFR(S): 8050978: Fix bad field access check in C1 and C2

[Lindenmaier, Goetz]

Hi,

This fixes an error doing field access checks in C1 and C2.
Please review and test the change.  We please need a sponsor.
http://cr.openjdk.java.net/~goetz/webrevs/8050978-fieldCheck/webrev-01/

This should be included in 8u20, too.

JCK8 test vm/constantpool/accessControl/accessControl004/accessControl00402m3/accessControl00402m3.html fails with -Xbatch -Xcomp due to bad field access check in C1 and C2

Precondition:
-------------

Consider the following class hierarchy:

    A
  / \
 B1 B2

A declares a field "aa" which both B1 and B2 inherit.

Despite aa is declared in a super class of B1, methods in B1 might not access the field aa of an object of class B2:

class B1 extends A {
  m(B2 b2) {
    ...
    x = b2.aa; // !!! Access not allowed
  }
}

This is checked by the test mentioned above.

Problem:
--------

ciField::will_link() used by C1 and C2 does the access check using the canonical_holder (which is A in this case) and thus the access erroneously succeeds.

Fix:
----

In ciField::ciField(), just before the canonical holder is stored into the _holder variable (and which is used by ciField::will_link()) perform an additional access check with the holder declared in the class file. If this check fails, store the declared holder instead and ciField::will_link() will bail out compilation for this field later on. Then, the interpreter will throw an PrivilegedAccessException at runtime.

Ways to reproduce:
------------------

Run the above JCK test with

C2 only: -XX:-TieredCompilation -Xbatch -Xcomp

or

with C1: -Xbatch -Xcomp -XX:-Inline

Best regards,
  Andreas and Goetz