RFR(S): 7069863: G1: SIGSEGV running SPECjbb2011 and -UseBiasedLocking

[John Cuthbertson]

Hi Everyone,

A new webrev based upon the feed back from Vladimir can be found at: 
http://cr.openjdk.java.net/~johnc/7069863/webrev.1/

I'm looking for one more reviewer.

Thanks,

JohnC

On 07/27/11 12:06, Vladimir Kozlov wrote:
> g1CollectedHeap.cpp:
>
>   // When compressed oops are used the preferred heap base is calculated
>   // by subtracting the requested size from the 32Gb boundary and using
>   // the result as the base address for heap reservation. If the size is
>   // not aligned to HeapRegion::GrainBytes passed into the 
> ReservedHeapSpace
>   // constructor then the base of the actual reserved heap may end up
>   // differing from the requested base address. If this happens then we
>   // could end up using a non-optimal compressed oops mode.
>
>
> virtualspace.cpp:
>
>  213   // prefix_align == suffix_align).
>                        ^ prefix_align < suffix_align
>
> ReservedSpace::ReservedSpace() also misses the call to 
> failed_to_reserve_as_requested() after the code which handle wrong 
> alignment.
>
> Add assert into ReservedSpace::initialize()
>
>   // Assert that if noaccess_prefix is used, it is the same as alignment.
>   assert(noaccess_prefix == 0 ||
>          noaccess_prefix == alignment, "noaccess prefix wrong");
>
> Why we have next code? Should we do this alignment adjustment at the 
> beginning of the method?
>
> 355   _alignment = MAX2(alignment, (size_t) os::vm_page_size());
>
> Vladimir
>
> John Cuthbertson wrote:
>> Hi Everyone,
>>
>> Can I have a couple of volunteers look over these changes - the 
>> webrev can be found at: 
>> http://cr.openjdk.java.net/~johnc/7069863/webrev.0/
>>
>> The issue was caused by an implicit null check in generated code not 
>> firing. The implicit null check did not trap as a result of a 
>> mismatch between the compressed oops mode and the calculated heap 
>> base. Also the page below heap base was not being protected. The 
>> mismatch between the compressed oops mode and heap base was the 
>> result of the G1 heap initialization code passing in a total heap 
>> size (G1 heap and perm) that was not a multiple of the alignment 
>> passed into the ReservedSpace constructor. Hence the preferred heap 
>> base address was not correctly aligned causing the ReservedSpace to 
>> be allocated at an address other than the preferred heap base. The 
>> page below the heap base was not protected because the G1 heap 
>> initialization code was calling the wrong ReservedSpace constructor.
>>
>> Testing: the failing test case; various small tests with various 
>> collectors, large heaps, and +PrintCompressedOopsMode.
>>
>> Many thanks to Vladimir and Igor for helping to diagnose the issue.
>>
>> Thanks,
>>
>> JohnC