Add MXCSR save and restore to upcall stubs for non-windows platforms

…e and added test (openjdk#11)

* JVM-1754: [PEA] Bad graph detected in build_loop_late

After materialization, we used to replace all appearances of the old cooked
oop with the new cooked oop along JVMState. This isn't correct if the
old cooked oop is alive in the map of an ancester.

When C2 encounters a real call, it needs to aggregate debug edges from
all ancestors. If we replace the old cook oop in materization before, we
may end up with a bad graph. The late control of the new cooked oop is
higher than its early control in dominator tree.

* Enhance PEAState::print_on().

dump more information for VirtualState.

* Deep copy VirtualObject in its clone function.

This patch also changes VirtualObject::_nfields from unsigned to signed and makes
it in release build.

* Extend VirtualState to record diverging state.

if is_singleton() returns true, VirtualState is merged.
merged VirtualStates are records in _diverged.

We normalize "diverging" VirtualState. if fields are all identical, we
store 'this' rather than the real pointer of VirtualState.

* JVM-1617: [PEA] Merge all virtual objects

This patch emits PhiNode for the field which has multiple values.

* Reset virtualObject when it copys to a brand new  basic block.

* Add the unit test MergeAllVirts

* Fix release build.

* Fix linux-x86-hs-optimized build

This build is fastdebug build but also optimized, so ASSERT is not
defined. we don't define validate() if ASSERT is not defined.

* Dump more information for VirtualObject.

* Add a new test for JVM-1798.

* JVM-1798: fix VirtualObject

We insert phiNode at merging point rather than staging divergent
VirtualObjects.

* Add yet another test to cover MergeAllVirts;

---------

Co-authored-by: Xin Liu <xxinliu@amazon.com>

It was added to the Kernel in 5.18

https://riscv.org/news/2022/03/risc-v-gets-sv57-based-virtual-memory-other-improvements-for-linux-5-18-michael-larabel-phoronix/

It doesn't work as expected. Local testing didn't catch the failing cases. We really need the proper test suite..

Reverting for now, and once we've proper CI, I'll rework the original patch and make it work.

This reverts commit 5be1125.

… data.

Co-authored-by: Martin Balao Alonso <mbalao@redhat.com>
Co-authored-by: Francisco Ferrari Bihurriet <fferrari@redhat.com>

Make a PKCS openjdk#11 library supporting only CKM_HKDF_DERIVE or CKM_HKDF_DATA
to also register the HKDF algorithms. Notice that the corresponding HMAC
mechanism is still required. In case only one of the mechanisms is
available, KDF::deriveKey or KDF::deriveData will fail at run time.

If the user disables one of CKM_HKDF_DERIVE or CKM_HKDF_DATA through
sun.security.pkcs11.Config, also refrain from using it.

Co-authored-by: Martin Balao Alonso <mbalao@redhat.com>
Co-authored-by: Francisco Ferrari Bihurriet <fferrari@redhat.com>

Update README.md

… fix naming

- Remove unused method deactivate_region_at() from g1CollectedHeap.hpp
- Remove unused method should_uncommit() from g1HeapRegion.hpp
- Remove duplicate log message in request_heap_shrink()
- Change request_heap_shrink() return type from bool to void
- Move _heap_evaluation_task initialization to constructor initializer list
- Remove unnecessary assert and init log message
- Inline record_activity() method as update_last_access_timestamp()
- Rename current_heap variable to current_capacity for clarity

Addresses issues #1, #2, openjdk#3, openjdk#4, openjdk#10, openjdk#11, openjdk#14, openjdk#18, openjdk#26 from Thomas review

… fix naming

- Remove unused method deactivate_region_at() from g1CollectedHeap.hpp
- Remove unused method should_uncommit() from g1HeapRegion.hpp
- Remove duplicate log message in request_heap_shrink()
- Change request_heap_shrink() return type from bool to void
- Move _heap_evaluation_task initialization to constructor initializer list
- Remove unnecessary assert and init log message
- Inline record_activity() method as update_last_access_timestamp()
- Rename current_heap variable to current_capacity for clarity

Addresses issues #1, #2, openjdk#3, openjdk#4, openjdk#10, openjdk#11, openjdk#14, openjdk#18, openjdk#26 from Thomas review

Update copyright year.

Backport BSD related fixes from mainline