Re: Regression in ptrace (Wine) starting with 2.6.33-rc1

[Frederic Weisbecker <fweisbec@gmail.com>]

On Mon, Feb 15, 2010 at 11:47:24AM -0800, Roland McGrath wrote:
> > - The other one with 'locally'/'globally' enabled breakpoints.
> 
> There is no "local/global" enablement.  That distinction is meaningless
> given the way the kernel uses the hardware.  Which of those bits you set
> has no material effect on the watchpoint/trap behavior.  



Yeah.


 
> The only regression is in the observed bit pattern read back from dr7.
> To be 100% compatible, the hw_breakpoint ptrace-compatibility front-end
> should record the state of the useless bits to report back, so the only
> differences from the bit pattern written are whatever ones the real
> hardware would have shown from writing dr7 and reading it back.



Agreed. We have to match the previous ABI, it's a regression.

We have the NULL breakpoint address thing fixed (will push it to Ingo).
We now need to fix the local/global flag storage.

The fastest way to do so is to keep a per thread dr7 variable.
I'm looking at it and will send a fix soon.

We can think about something proper later.

Thanks.