RR 6313816: SA: jstack -m fails on Win32 : UnalignedAddressException

[Poonam Bajaj]

Hi Kevin,

The changes look good.

Thanks,
Poonam

On 5/7/2013 5:36 PM, Kevin Walls wrote:
> Hi -
>
> I'd like to get another review from an openjdk user/reviewer/...
>
> Thanks
> Kevin
>
>
> On 03/05/13 09:32, Kevin Walls wrote:
>> Thanks Staffan, thanks Peter!
>>
>>
>> On 02/05/13 14:11, Peter Allwin wrote:
>>> Hi Kevin,
>>>
>>> Looks good!
>>>
>>> I've been looking into improving the quality of native stack traces 
>>> on Windows using IDebugControl::GetStackTrace. This has the 
>>> advantage of handling some FPO frames on x86 (a common source of 
>>> these exceptions), as well as supporting AMD64 (walking the stack 
>>> using RBP does not work at all on Windows/X64). There are still some 
>>> issues to iron out before this change is ready, but your fix is a 
>>> good start to what I'd like to accomplish.
>>>
>>>
>>> Thanks,
>>>
>>> Peter
>>>
>>>> ------------------------------------------------------------------------
>>>>
>>>> 	Kevin Walls <mailto:kevin.walls at oracle.com>
>>>> Tuesday, April 30, 2013 7:02 PM
>>>>
>>>>
>>>> Hi,
>>>>
>>>> I'd like some review of this suggestion to fix:
>>>> 6313816 SA windows unaligned address exception
>>>>
>>>> webrev:
>>>> http://cr.openjdk.java.net/~kevinw/6313816/webrev/
>>>>
>>>> This bug is a long-standing annoyance, mainly shown by using jstack -m
>>>>
>>>> Already fixed on Linux with additional sanity checks on ebp/rbp 
>>>> values in
>>>> e.g. LinuxX86CFrame.sender()
>>>>
>>>> On Windows there isn't such a convenient place to put these checks, 
>>>> as they
>>>> call getAddressValue() which is only part of the specific debugger 
>>>> interfaces
>>>> such as LinuxDebugger, WindbgDebugger, etc.. and on Windows we use
>>>> X86CFrame and AMD64CFrame.  These are constructed with only a
>>>> CDebugger, there is no "WinXXXFrame" class - perhaps there should be!
>>>>
>>>> That's what  I've tried here, by renaming e.g. X86CFrame to 
>>>> WinX86CFrame,
>>>> and adding the extra sanity checks as we already have in 
>>>> LinuxX86CFrame.
>>>>
>>>> As files have move thed diff isn't obvious, but in common with some 
>>>> previous changes
>>>> the new bits are x86 and x64 versions of these:
>>>>
>>>> + // Check alignment of rbp
>>>> + if ( dbg.getAddressValue(rbp) % ADDRESS_SIZE != 0) {
>>>> + return null;
>>>> + }
>>>> +
>>>>
>>>> and
>>>>
>>>> - if (nextRBP == null) {
>>>> + if (nextRBP == null || nextRBP.lessThanOrEqual(rbp)) {
>>>>
>>>>
>>>> This additional symmetry between the platforms seems to make sense, 
>>>> and
>>>> should help any future changes.
>>>>
>>>> Thanks
>>>> Kevin
>>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/serviceability-dev/attachments/20130516/d2d3e90c/attachment-0001.html 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: compose-unknown-contact.jpg
Type: image/jpeg
Size: 770 bytes
Desc: not available
Url : http://mail.openjdk.java.net/pipermail/serviceability-dev/attachments/20130516/d2d3e90c/compose-unknown-contact-0001.jpg