This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [PATCH 3/5] range stepping: gdb
- From: Pedro Alves <palves at redhat dot com>
- To: Eli Zaretskii <eliz at gnu dot org>
- Cc: gdb-patches at sourceware dot org
- Date: Wed, 15 May 2013 13:39:05 +0100
- Subject: Re: [PATCH 3/5] range stepping: gdb
- References: <20130514191026 dot 13213 dot 39574 dot stgit at brno dot lan> <20130514191047 dot 13213 dot 8476 dot stgit at brno dot lan> <83k3n173ao dot fsf at gnu dot org> <5193621C dot 50603 at redhat dot com> <83ppws5w00 dot fsf at gnu dot org>
On 05/15/2013 12:21 PM, Eli Zaretskii wrote:
>> Date: Wed, 15 May 2013 11:23:24 +0100
>> From: Pedro Alves <palves@redhat.com>
>> CC: gdb-patches@sourceware.org
>>
>>>> +@var{end} is the address of the first instruction beyond the step
>>>> +range, and @strong{not} the address of the last instruction within it.
>>>> +(This has the property that @var{start} == @var{end} single-steps
>>>> +once, and only once, even if the instruction at @var{start} jumps to
>>>> +@var{end}.)
>>>
>>> This sentence in parentheses got me completely confused. Before
>>> reading it, I thought I understood what is this about; now I don't.
>>> In particular, if START is equal to END, then how in the world could
>>> the instruction at START jump to END?
>>
>> Sorry, I had that typo in the gdbserver code as well, fixed it
>> there, but missed this one.
>>
>> It should read, even if the instruction at @var{start} jumps to @var{start}.
>>
>> vCont;r first steps, then checks. IOW:
>>
>> vCont ;r ADDR1,ADDR1
>>
>> is equivalent to (and could be thought to supersede):
>>
>> vCont ;s
>>
>>> And if END is excluded from the
>>> range, then why when START equals END do we step at all? Please
>>> explain.
>>
>> It's just a design decision. I recall at least one target I saw I worked
>> with that supported range stepping, and it didn't even a distinction
>> between range vs no-range step commands. The way to do a single step
>> was to pass both addresses the same. I find it a better design than
>> requiring the target do one current-address check _before_ stepping,
>> and another _after_ single-stepping.
>
> Doesn't this mean that these two use cases are explicit exceptions
> from the rule that END is excluded?
Nope. There's no exception.
With:
vCont ;r START,END
#1 - The stub single-steps the thread.
#2 - Once the thread stops, the stub checks whether the thread
stopped in the [START,END) range. If so, goto #1.
It not, goto #3.
#3 - The stub reports to gdb that the thread stopped stepping.
If it happens that START and END are the same, then #2 always
goes to #3.
When I said:
"(This has the property that @var{start} == @var{end} single-steps
once, and only once, even if the instruction at @var{start} jumps to
@var{start}.)"
I was trying to clarify the case of the instruction at START being:
jump START
Then,
vCont ;r START,START
always single-steps once, and only once, instead of
continuously single-stepping that instruction without
reporting to GDB.
> If so, we should describe them as
> exceptions, not use them as evidence for the rule (which they
> evidently violate).
>
> Or did I misunderstand again?
--
Pedro Alves