RFA: frame id enhancement
J. Johnston
jjohnstn@redhat.com
Thu Oct 16 21:49:00 GMT 2003
Andrew Cagney wrote:
>>
>>> It's the reverse of infrun.c:2383 where the inferior is falling out
>>> of a singnal trampoline, I think the assumptions again hold.
>>>
>>> infrun.c:2641: if (!(frame_id_inner (current_frame, step_frame_id)))
>>>
>>> "Trust me" there's no value add. While the comment reads:
>>> /* In the case where we just stepped out of a function into the
>>> middle of a line of the caller, continue stepping, but
>>> step_frame_id must be modified to current frame */
>>> The test also updates step_frame_id when switching between frameless
>>> stackless leaf function. The extra test wouldn't fix that problem.
>>> I'll try to remember to add some comments to that code.
>
>
> I've done this.
>
>> Ok, that simplifies things. I have included a revised patch that
>> allows for the wild-card scenario.
>
>
> We're going to need more comments so that the next person better
> understands what is going on:
>
> + /* The frame's special address. This shall be constant through out the
> + lifetime of the frame. This is used for architectures that may have
> + frames that have the same stack_addr and code_addr but are distinct
> + due to some other qualification (e.g. the ia64 uses a register
> + stack which is distinct from the memory stack). */
> + CORE_ADDR special_addr;
>
> can you expand this definition to to note that the value isn't ordered,
> and that zero is treated as a wild card (its mentioned further down but
> I think here, at the definition, is better). For the ia64, is/can the
> second area be described as a register spill area rather than a stack?
> If the word "stack" can be avoided, the rationale for "special" being
> un-ordered is stronger.
>
It "is" a register stack on the ia64. Registers r32 - r127 for any frame all
come from this area. It gets bumped up by a special alloc() instruction. I'm
not sure I would call it unordered. It may be better to say that it is treated
as unordered. That would make the comments below much simpler - i.e. the
special_addr field is treated as unordered so it is never used to determine
order when comparing frames.
I can easily add the zero/wildcard comment.
> For:
>
> NOTE: Given frameless functions A and B, where A calls B (and hence
> B is inner-to A). The relationships: !eq(A,B); !eq(B,A);
> !inner(A,B); !inner(B,A); all hold. This is because, while B is
> inner to A, B is not strictly inner to A (being frameless, they
> have the same .base value). */
>
> an update is needed, suggest something like:
>
> NOTE:
>
> Given stackless functions A and B, where A calls B (and hence
> B is inner-to A). The relationships: !eq(A,B); !eq(B,A);
> !inner(A,B); !inner(B,A); all hold.
>
> This is because, while B is
> inner-to A, B is not strictly inner-to A. Being stackless, they
> have an identical .stack_addr value, and differ only by their
> unordered .code_addr .special_addr values.
>
> Because frame_id_inner is only used as a safety net (e.g.,
> detect a corrupt stack) the lack of strictness is not a problem.
> Code needing to determine an exact relationship between two frames
> must instead use frame_id_eq and frame_id_unwind. For instance,
> in the above, to determine that A stepped-into B, the equation
> "A.id != B.id && A.id == id_unwind (B)" can be used.
>
>
> and a similar update to:
>
> frame_id_inner (struct frame_id l, struct frame_id r)
> {
> int inner;
> if (l.stack_addr == 0 || r.stack_addr == 0)
> /* Like NaN, any operation involving an invalid ID always fails. */
> inner = 0;
> else
> /* Only return non-zero when strictly inner than. Note that, per
> comment in "frame.h", there is some fuzz here. Frameless
> functions are not strictly inner than (same .stack but
> different .code). */
> inner = INNER_THAN (l.stack_addr, r.stack_addr);
>
> I can't think of a word better than "special", so I guess special it is :-)
>
> Andrew
>
>
>
>
More information about the Gdb-patches
mailing list