Re: [RFA] Resubmit process record and replay, 6/10

[teawater <teawater at gmail dot com>]

Sorry I forget a big part that need it.
When GDB work in replay mode, P record will set regs and memory in
record_wait. All of them can't be record.

So what about set not_record flag to record_wait in replay mode,
record_insert_breakpoint and record_remove_breakpoint.

And about the name of this flag, do you have some idea on it?


Thanks,
Hui


On Wed, Nov 26, 2008 at 02:22, Michael Snyder <msnyder@vmware.com> wrote:
> teawater wrote:
>>
>> On Tue, Nov 25, 2008 at 03:16, Michael Snyder <msnyder@vmware.com> wrote:
>>>
>>> teawater wrote:
>>>>
>>>> Hi Michael,
>>>>
>>>> About "record_not_record_set", It set record_not_record to let P
>>>> record doesn't record the memory and registers control behaviors of
>>>> GDB in function record_store_registers and record_xfer_partial.
>>>>
>>>> So I think the name "record_not_record_set" and
>>>> "record_skip_recording" are not very clear.
>>>> Could you please give me some advices on it?
>>>
>>> Yeah, that's pretty much the way I understood it.
>>>
>>> It sets a one-time flag that says "omit (skip) recording
>>> registers and memory that would otherwise be recorded".
>>>
>>> And if I understand correctly, this is to avoid adding
>>> changes to the record log that are made by gdb when it
>>> resumes the target.  It's only called from "proceed()".
>>>
>>> I'm not completely clear on what those changes are.
>>> Is gdb modifying the PC?  Or are you perhaps trying to
>>> avoid recording breakpoints?
>>
>> I think avoid recording breakpoints is the main affect.
>> Another function is help deal with displaced step. Of course, P record
>> and displaced step will not work together now.
>>
>> I think I add "record_not_record" function is because I want
>> record_store_registers and record_xfer_partial just record the user
>> level change, not for others.
>> What do you think about it?
>
> OK, so if we ignore displaced stepping for now, then can we
> limit the issue to breakpoints?
>
> Breakpoint writes will all pass through functions called
> memory_insert_breakpoint and memory_remove_breakpoint (mem-break.c).
>
> So what we want to do is get the information from there into
> record.c.  I guess you could do pretty much what you are doing
> now, only call the access function from mem-break.c instead of
> from infrun.  It would help to localize it and make its meaning
> clear.
>
> Maybe call it "dont_record_memory_breakpoint" or something like that.
>
>