Need reviewers for changes in JVM initialization code (added check for -XX:StackShadowPages option)
tom.rodriguez at oracle.com
Wed Jul 28 10:12:58 PDT 2010
>> Tom Rodriguez said the following on 07/28/10 04:49:
>>> I think you could put this check into arguments.cpp since I all platforms would require a positive number for the StackShadowPages. The same should be true of StackRedPages and StackYellowPages. Actually they all should be required to greater than 0 I think. Other than that I don't see other obvious constraints on the values. That bug report doesn't really show other problematic values, though I'm not sure I follow the point of the guarantee that's failing either. What does the min stack size have to do with the number of guard pages? I would expect it to be checking against ThreadStackSize and returning an error if it was too small like the other places that check against TheadStackSize.
> It's a simple sanity check that the minimum allowed stack size can actually accommodate a stack with all the requested guard pages. We need a minimum allowed, as well as a default (ThreadStackSize) for when stacks are programmatically requested.
Why should it be a guarantee instead of a simple check that returns an error code like the stack size check below it? It sort of make sense as a guarantee if the various Stack*Pages flags have their default value since we're not bootstrappable in that case but since they are adjustable by a user returning a real error message seems to make more sense.
>>> On Jul 27, 2010, at 8:15 AM, Pavel Tisnovsky wrote:
>>>> Hi all,
>>>> can anybody please review two quite simple changes in JVM initialization code? Webrev is available at:
>>>> When -XX:StackShadowPages is set to negative integer or zero value, JVM segfaulted on Linux and, according to http://bugs.sun.com/view_bug.do?bug_id=6885308, hangs up on Solaris (although I only check this issue on Linux)
>>>> I also would like to add more check for -XX:StackRedPages, -XX:StackShadowPages and -XX:StackYellowPages options to avoid issue described in the bug report mentioned above (http://bugs.sun.com/view_bug.do?bug_id=6885308) but I'm unable to find relevant information about proper conditions (it may depends on page sizes, VM stack size etc.). Any ideas?
>>>> Pavel Tisnovsky
More information about the hotspot-dev