review (S) for 6970683: improvements to hs_err output
tom.rodriguez at oracle.com
Tue Oct 12 10:02:18 PDT 2010
On Oct 8, 2010, at 9:03 AM, Coleen Phillimore wrote:
> This change looks good to me. I missed when print_location() was added for the registers, but it does a lot of things that we used to consider unsafe from the error handler.
It does, though it seems like it's handled properly by the existing logic for restarting failures during dumping. Many of the routines do fairly trivial printing. If we find an oop then much more complicated printing occurs though mostly more interesting than resource allocation is really going to occur. Stack overflow conditions are likely the most dangerous possibility.
> And I didn't like all the white space. I wonder if you should check for Universe::is_initialized() in vmError before calling this?
> Can you attach an "after" version of hs_err? I guess I can get one of my own pretty easily after you check this in, but I'd like to see one first.
I'll generate a couple after fixing the review comments and send them out.
> Lastly, what sort of problems can you diagnose from the code cache bounds?
I've wanted it several times to make sure that values that appeared to be code cache values actually were. Also the counts of how full the code cache is can be useful.
> Tom Rodriguez wrote:
>> 6970683: improvements to hs_err output
>> There are a few things missing from the hs_err dump that would be
>> useful. First we don't dump the sparc L and I registers. Second some
>> information about the size and contents of the code cache would be
>> useful. Third we should dump a larger region around the faulting
>> instruction. Additionally the new register to memory mapping output
>> can crash which stops us from getting the stack and instructions at
>> the faulting pc, so I moved it into it's own section. block_start
>> would assert in some cases so I augmented existing logic to just
>> return null. I also changed the formatting to remove all the extra
>> whitespace and made some of the output more compact and eliminated
>> most of the useless whitespace.
More information about the hotspot-dev