additional errors on several xp tests on buildbot

[derekbruening]

From bruen...@google.com on February 02, 2011 09:24:27

the xp bot is not matching my xp machine, mainly due to additional errors:

cs2bug =>   stderr failed to match :::Dr\.Memory::: 1 unique, 1 total uninitialized
      access\(es\)?, found :::Dr.Memory::: 5 unique, 5 total uninitialized
      access(es) instead
patterns =>   stderr failed to match :::Dr\.Memory::: 0 unique, 0 total unaddressable
      access\(es\)?, found :::Dr.Memory::: 1 unique, 1 total unaddressable
      access(es) instead
suppress =>   stderr failed to match :::Dr\.Memory::: 0 unique, 0 total unaddressable
      access\(es\)?, found :::Dr.Memory::: 6 unique, 6 total unaddressable
      access(es) instead
wincrtdbg =>   stderr failed to match :::Dr\.Memory::: 7 unique, 7 total unaddressable
      access\(es\)?, found :::Dr.Memory::: 4 unique, 4 total unaddressable
      access(es) instead
malloccrt =>   stderr failed to match :::Dr\.Memory::: 1 unique, 20 total unaddressable
      access\(es\)?, found :::Dr.Memory::: 1 unique, 1 total unaddressable
      access(es) instead

Original issue: http://code.google.com/p/drmemory/issues/detail?id=286

[derekbruening]

From bruen...@google.com on July 22, 2011 07:54:03

pasting in info from long-ago r23 :

issue #286: windows thread creation false positives: unaddr, uninit, and leaks
Fixed multiple false positives reported for each thread creation on Windows:

• set shadow defined for new thread TEB by querying post-NtCreateThread
  (TEB is allocated by kernel so we never see its allocation)
• set shadow unaddr for thread TEB dealloc on thread exit
• set full range of kernel-written stack on Ki routine
• set below-TOS to defined and beyond-TOS to unaddr in thread init,
  which required augmenting issue handle shared memory written externally #117/PR 395156 to support getting
  the mcontext on (non-primary thread) thread init
• fixed bug in interception buffer mapping where the jmp back was also
  considered on the original app code side, causing the bb after an
  interception to be hidden from the client, resulting in false positives
  b/c Dr. Memory did not see some KiAPC code!
• set register shadow values based on NtContine CONTEXT
• suppressed RtlAllocateActivationContextStack leak that shows
  up with _beginthreadex() under the assumption that it's a false positive:
  but see PR 575445 where suppressed leaks still show up by default!

xref issue #445 , issue #506 : other activation context leaks