svn commit: r212076 - head/lib/libthr/thread

[David Xu]

John Baldwin wrote:
> On Thursday, September 02, 2010 7:36:08 am David Xu wrote:
>   
>> John Baldwin wrote:
>>     
>>> On Thursday, September 02, 2010 6:22:18 am Kostik Belousov wrote:
>>>   
>>>       
>>>> On Thu, Sep 02, 2010 at 05:35:39PM +0000, David Xu wrote:
>>>>     
>>>>         
>>>>> Kostik Belousov wrote:
>>>>>
>>>>>       
>>>>>           
>>>>>>> the tf_err may not be equal to ksi_addr! This may need to be fixed.
>>>>>>>
>>>>>>>           
>>>>>>>               
>>>>>> The change was introduced by
>>>>>> r151316 | davidxu | 2005-10-14
>>>>>>      /* Old FreeBSD-style arguments. */
>>>>>> -     sf.sf_siginfo = code;
>>>>>> -     sf.sf_addr = regs->tf_err;
>>>>>> +     sf.sf_siginfo = ksi->ksi_code;
>>>>>> +     sf.sf_addr = (register_t)ksi->ksi_addr;
>>>>>>      sf.sf_ahu.sf_handler = catcher;
>>>>>>
>>>>>>         
>>>>>>             
>>>>> :(
>>>>>       
>>>>>           
>>>> The rollback looks straightforward. I explicitely decided to not change
>>>> any architecture that is not i386.
>>>>     
>>>>         
>>> It may not be this simple.  At one point we had a "feature" where we 
>>>       
> trashed 
>   
>>> tf_err in the trapframe to store the address so it could be passed to 
>>>       
> sendsig 
>   
>>> for this purpose.  I think once we started using ksi_addr here we removed 
>>>       
> the 
>   
>>> trashing of tf_err as it was no longer necessary.
>>>
>>>   
>>>       
>> I saw RELENG_5 saved it at td_md.md_fault_address, but I really can not 
>> remember
>> which version has such "feature". ;-)
>> By the way, I still can not find a  modern program uses this old style 
>> and wants to
>> know the exact fault address.
>> Kostik Belousov fixed the error I had made when I was implementing 
>> signal queue.
>>     
>
> See revision 170688 for example:
>
> r170688 | jhb | 2007-06-13 18:37:48 -0400 (Wed, 13 Jun 2007) | 6 lines
>
> Don't clobber tf_err with the eva from a page fault as the page fault
> address is saved in ksi_addr already.
>
> PR:             i386/101379
> Submitted by:   Tijl Coosemans : tijl ulyssis org
>
> Index: trap.c
> ===================================================================
> --- trap.c      (revision 170687)
> +++ trap.c      (revision 170688)
> @@ -785,9 +785,6 @@
>                 return (-1);
>         }
>  
> -       /* kludge to pass faulting virtual address to sendsig */
> -       frame->tf_err = eva;
> -
>         return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
>  }
>
> Given that, I think the existing code is correct and that there is no bug to 
> be fixed.
>
>   
Thanks for clarifying, I am really confused when Kostik Belousov mailed me,
I wish I had not made the mistake, after all it was 4 years 10 months 
ago, so I
can not remember exactly what had happened at that time.