Simplify VirtualMemory by removing unused ReserveAligned method.

Simplify VirtualMemory by removing unused ReserveAligned method.

Also add the ability to truncate without actually unmapping.
Note: We never create holes in the underlying reservation.

This CL is in preparation for using a thin wrapper around VirtualMemory to verify our upcoming concurrent write barrier.

R=iposva@google.com

Committed: https://code.google.com/p/dart/source/detail?r=41017

[koda, 2014-10-09 00:25:30]

For background, in the special debugging write barrier verification mode, every
audited object pointer write into the heap will be duplicated at a fixed offset
from the original address.

[Ivan Posva, 2014-10-09 07:01:13]

iposva@google.com changed reviewers:
+ iposva@google.com

[Ivan Posva, 2014-10-09 07:01:14]

LGTM

-Ivan

https://codereview.chromium.org/644453003/diff/80001/runtime/vm/virtual_memory.h
File runtime/vm/virtual_memory.h (right):

https://codereview.chromium.org/644453003/diff/80001/runtime/vm/virtual_memor...
runtime/vm/virtual_memory.h:52: return ReserveInternal(size);
Why this extra indirection?

https://codereview.chromium.org/644453003/diff/80001/runtime/vm/virtual_memor...
runtime/vm/virtual_memory.h:64: void Truncate(intptr_t new_size, bool try_unmap
= true);
When would it be beneficial to truncate without unmapping?

[koda, 2014-10-09 15:58:52]

https://codereview.chromium.org/644453003/diff/80001/runtime/vm/virtual_memory.h
File runtime/vm/virtual_memory.h (right):

https://codereview.chromium.org/644453003/diff/80001/runtime/vm/virtual_memor...
runtime/vm/virtual_memory.h:52: return ReserveInternal(size);
On 2014/10/09 07:01:14, Ivan Posva wrote:
> Why this extra indirection?

To allow for inserting a bit of OS-independent logic in the future without
having to update all the OS-specific files (e.g., the write barrier
verification).

https://codereview.chromium.org/644453003/diff/80001/runtime/vm/virtual_memor...
runtime/vm/virtual_memory.h:64: void Truncate(intptr_t new_size, bool try_unmap
= true);
On 2014/10/09 07:01:13, Ivan Posva wrote:
> When would it be beneficial to truncate without unmapping?

I will use it to implement the writer barrier verification mode with minimal
changes to the code using VirtualMemory. Each VirtualMemory will have a trailing
"shadow" area, and pointer writes will be duplicated at a fixed offset.

[koda, 2014-10-09 15:59:35]

Committed patchset #5 (id:80001) manually as r41017 (presubmit successful).