RFR (XS) 8148813 - Windows os::check_heap needs more information

[Ioi Lam]

On 6/21/16 10:38 PM, Ioi Lam wrote:
>
>
> On 6/21/16 10:19 PM, David Holmes wrote:
>> Hi Ioi,
>>
>> On 22/06/2016 2:51 PM, Ioi Lam wrote:
>>> HI,
>>>
>>> Please review a small Windows fix:
>>>
>>> https://bugs.openjdk.java.net/browse/JDK-8148813
>>> http://cr.openjdk.java.net/~iklam/jdk9/8148813_windows_check_heap.v01/
>>>
>>> We have seen strange heap corruption fatal errors that produced mdmp
>>> files (windows crash dumps),
>>> e.g. https://bugs.openjdk.java.net/browse/JDK-8147481
>>>
>>> However, when we load these mdmp files in WinDBG, "!heap -triage"
>>> shows no heap corruption.
>>>
>>> These are probably spurious errors reported by HeapWalk. To properly
>>> diagnose such errors, this patch saves a few of the last seen
>>> PROCESS_HEAP_ENTRY's into a ring buffer, which can later
>>> be examined with a debugger.
>>
>> I don't understand the use of count or the magic 42. ??
>>
> Oops, that was some debug code that shouldn't be there. Here's a fixed 
> version
>
> http://cr.openjdk.java.net/~iklam/jdk9/8148813_windows_check_heap.v02/
>

BTW, I left the count in the output so we can see how many blocks we 
have walked before the heap corruption was detected. This info is also 
useful for diagnosing the spurious errors. The output looks like this:

C heap has been corrupted (time: 1 allocations)
corrupted block near address 0x356d30, length 32, count 43

Thanks
- Ioi