Message-Id: <20241002225614.774bdd0742a826557f142d0e@kernel.org>
Date: Wed, 2 Oct 2024 22:56:14 +0900
From: Masami Hiramatsu (Google) <mhiramat@...nel.org>
To: Ian Rogers <irogers@...gle.com>
Cc: Namhyung Kim <namhyung@...nel.org>, Peter Zijlstra
 <peterz@...radead.org>, Ingo Molnar <mingo@...hat.com>, Arnaldo Carvalho de
 Melo <acme@...nel.org>, Mark Rutland <mark.rutland@....com>, Alexander
 Shishkin <alexander.shishkin@...ux.intel.com>, Jiri Olsa
 <jolsa@...nel.org>, Adrian Hunter <adrian.hunter@...el.com>, Kan Liang
 <kan.liang@...ux.intel.com>, John Garry <john.g.garry@...cle.com>, Will
 Deacon <will@...nel.org>, James Clark <james.clark@...aro.org>, Mike Leach
 <mike.leach@...aro.org>, Leo Yan <leo.yan@...ux.dev>, Guo Ren
 <guoren@...nel.org>, Paul Walmsley <paul.walmsley@...ive.com>, Palmer
 Dabbelt <palmer@...belt.com>, Albert Ou <aou@...s.berkeley.edu>, Nick
 Terrell <terrelln@...com>, Guilherme Amadio <amadio@...too.org>, Changbin
 Du <changbin.du@...wei.com>, "Steinar H. Gunderson" <sesse@...gle.com>,
 Aditya Gupta <adityag@...ux.ibm.com>, Athira Rajeev
 <atrajeev@...ux.vnet.ibm.com>, Masahiro Yamada <masahiroy@...nel.org>,
 Huacai Chen <chenhuacai@...nel.org>, Bibo Mao <maobibo@...ngson.cn>, Kajol
 Jain <kjain@...ux.ibm.com>, Anup Patel <anup@...infault.org>, Shenlin Liang
 <liangshenlin@...incomputing.com>, Atish Patra <atishp@...osinc.com>,
 Oliver Upton <oliver.upton@...ux.dev>, Chen Pei <cp0613@...ux.alibaba.com>,
 Dima Kogan <dima@...retsauce.net>, Alexander Lobakin
 <aleksander.lobakin@...el.com>, "David S. Miller" <davem@...emloft.net>,
 Przemek Kitszel <przemyslaw.kitszel@...el.com>,
 linux-kernel@...r.kernel.org, linux-perf-users@...r.kernel.org, Yang Jihong
 <yangjihong@...edance.com>
Subject: Re: [PATCH v1 11/11] perf build: Rename PERF_HAVE_DWARF_REGS to
 PERF_HAVE_LIBDW_REGS

On Tue, 1 Oct 2024 18:31:43 -0700
Ian Rogers <irogers@...gle.com> wrote:

> On Tue, Oct 1, 2024 at 4:29 PM Masami Hiramatsu <mhiramat@...nel.org> wrote:
> >
> > On Tue, 1 Oct 2024 16:17:34 -0700
> > Ian Rogers <irogers@...gle.com> wrote:
> >
> > > On Tue, Oct 1, 2024 at 4:10 PM Namhyung Kim <namhyung@...nel.org> wrote:
> > > >
> > > > On Mon, Sep 30, 2024 at 09:02:36PM -0700, Ian Rogers wrote:
> > > > > On Sat, Sep 28, 2024 at 7:35 PM Masami Hiramatsu <mhiramat@...nel.org> wrote:
> > > > > >
> > > > > > On Fri, 27 Sep 2024 11:15:21 -0700
> > > > > > Ian Rogers <irogers@...gle.com> wrote:
> > > > > >
> > > > > > > On Fri, Sep 27, 2024 at 10:16 AM Namhyung Kim <namhyung@...nel.org> wrote:
> > > > > > > >
> > > > > > > > On Thu, Sep 26, 2024 at 12:55:18PM -0700, Ian Rogers wrote:
> > > > > > > > > On Thu, Sep 26, 2024 at 12:40 PM Namhyung Kim <namhyung@...nel.org> wrote:
> > > > > > > > > >
> > > > > > > > > > On Thu, Sep 26, 2024 at 05:47:16AM -0700, Ian Rogers wrote:
> > > > > > > > > > > On Wed, Sep 25, 2024 at 8:27 PM Namhyung Kim <namhyung@...nel.org> wrote:
> > > > > > > > > > > >
> > > > > > > > > > > > On Tue, Sep 24, 2024 at 09:04:18AM -0700, Ian Rogers wrote:
> > > > > > > > > > > > > The name dwarf can imply libunwind support, whereas
> > > > > > > > > > > > > PERF_HAVE_DWARF_REGS is only enabled with libdw support. Rename to
> > > > > > > > > > > > > make it clearer there is a libdw connection.
> > > > > > > > > > > >
> > > > > > > > > > > > While it only covers libdw, I think the idea of this macro is whether
> > > > > > > > > > > > the arch has register mappings defined in DWARF standard.  So I think
> > > > > > > > > > > > it's better to keep the name for this case.
> > > > > > > > > > >
> > > > > > > > > > > How can the dwarf standard exist for an arch but not define registers?
> > > > > > > > > >
> > > > > > > > > > I meant it's about the arch code in the perf tools to have the mapping,
> > > > > > > > > > not the DWARF standard itself.
> > > > > > > > >
> > > > > > > > > But we guard those definitions behind having libdw:
> > > > > > > > > https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/arch/x86/Makefile?h=perf-tools-next#n3
> > > > > > > > > So we only have the regs if libdw is present, not if dwarf is in use
> > > > > > > > > for libunwind/libdw. Hence wanting to be specific that they are just a
> > > > > > > > > libdw and not a dwarf thing. Trying to use the regs in libunwind code
> > > > > > > > > would be broken. That could change but I wanted to make the code clear
> > > > > > > > > for the way things are at the moment.
> > > > > > > >
> > > > > > > > I understand your point but calling it LIBDW_REGS looks unnatural to me.
> > > > > > >
> > > > > > > I don't follow. Wouldn't it be unnatural to see PERF_HAVE_DWARF_REGS
> > > > > > > in libunwind code but you are to some how know that the code only had
> > > > > > > meaning if libdw was present? I don't like the implication that DWARF
> > > > > > > means LIBDW as throughout the code it doesn't. I think the name
> > > > > > > PERF_HAVE_LIBDW_REGS better captures how the code is, makes the code
> > > > > > > more intention revealing and so readable, etc.
> > > > > >
> > > > > > I agree with Namhyung this point. dwarf-regs is defined only by the
> > > > > > DWARF standard, not libdw only. The standard encode registers by a digit
> > > > > > number and the dwarf-regs decode the number to actual register name.
> > > > >
> > > > > The code is not making a statement about the DWARF standard, take arch/csky:
> > > > > https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/arch/csky/Makefile?h=perf-tools-next
> > > > > ```
> > > > > # SPDX-License-Identifier: GPL-2.0-only
> > > > > ifndef NO_DWARF
> > > > > PERF_HAVE_DWARF_REGS := 1
> > > > > endif
> > > > > ```
> > > > > in the patch series NO_DWARF becomes NO_LIBDW, so it is now:
> > > > > ```
> > > > > # SPDX-License-Identifier: GPL-2.0-only
> > > > > ifndef NO_LIBDW
> > > > > PERF_HAVE_DWARF_REGS := 1
> > > > > endif
> > > > > ```
> > > > > So the Makefile says that PERF_HAVE_DWARF_REGS is dependent on having
> > > > > NO_LIBDW, that is having libdw implies PERF_HAVE_DWARF_REGS is defined
> > > > > for csky.
> > > >
> > > > I think this is totally fine and we can change the condition later if
> > > > needed.
> > > >
> > > > After all, I don't think it's a big deal.  Let's just call DWARF
> > > > registers DWARF_REGS. :)
> > >
> > > The define is called PERF_HAVE_DWARF_REGS, notice the HAVE, but we're
> > > not setting it while supporting call-graph dwarf with libunwind. It is
> > > actively confusing.
> >
> > Does libunwind requires the dwarf regs? I think the dwarf regs information
> > is only needed for analyzing dwarf register assignment, not stack unwinding.
> 
> So you are saying the #define is guarding a libdw feature?
> perf record/report --call-graph=dwarf is supported with either libdw
> or libunwind. The dwarf support in the tool may come from more sources
> hence wanting in this patch set to be clear what variable is guarding
> what. PERF_HAVE_DWARF_REGS is set to 1 for a specific set of
> architectures and only when libdw is present. The variable is saying
> that libdw supports the notion of registers needed for the #define
> HAVE_DWARF_SUPPORT that patch 9 in the series renamed to
> HAVE_LIBDW_SUPPORT. So I want the makefile variable
> PERF_HAVE_LIBDW_REGS to guard the #define HAVE_LIBDW_SUPPORT, rather
> than what is being argued by yourself and Namhyung that the #define
> HAVE_LIBDW_SUPPORT be guarded by a variable called
> PERF_HAVE_DWARF_REGS and that is only set when NO_LIBDW isn't set.

It will be only used with the libdw, but I don't care.
"HAVE_DWARF_REG" (internal config, just indicates the arch implemented
feature) simply means there is `arch/XXX/util/dwarf-regs.c`.
Also the APIs provided by the dwarf-regs.c are still based on DWARF
standard, which defines registers by number like DW_OP_reg[0-31].
So the mapping of these suffix number and actual register must be
defined for each architecture.

That is why I had introduced dwarf-regs.c and call it "dwarf"-regs.
Even if the implementation depends on libdw, this dwarf-regs.c is
still based on DWARF standard.

> We've made a digression into the name dwarf for a reason I can't
> fathom, at best it is inconsistent. Having dwarf registers is like
> having a bright sun or numeric numbers, it is a truism (playing devils
> advocate maybe if there were an ELF file format for postscript we
> could have a dwarf specification without registers). Anyway, I'm
> trying to connect the dots that libdw support controls the libdw type
> variables and defines hence not wanting 10 out of 11 patches applied.

Oh, wait, I think we can apply other patches. I just don't like this
patch. I think the other patches are good. But this is 

Thank you,

> 
> Thanks,
> Ian
> 
> > Thank you,
> >
> > >
> > > Thanks,
> > > Ian
> > >
> > > > Thanks,
> > > > Namhyung
> > > >
> > > > >
> > > > > Dwarf in the code base implies libdw, libunwind and potentially other
> > > > > dwarf capable things like llvm. If we don't have libdw then NO_LIBDW
> > > > > will be set and PERF_HAVE_DWARF_REGS won't be set. That is the more
> > > > > general dwarf thing will not be set because of missing libdw. This
> > > > > goes contrary to wanting this to be true whenever a dwarf thing is
> > > > > present - something that reflecting what the standard says would
> > > > > achieve.
> > > > >
> > > > > In the code base right now PERF_HAVE_DWARF_REGS isn't a dwarf
> > > > > dependent thing, it is a libdw dependent thing, this is why
> > > > > PERF_HAVE_LIBDW_REGS is a more intention revealing name as it makes
> > > > > the connection explicit.
> > > > >
> > > > > We could change the code so that in Makefile.config we set something like:
> > > > > ```
> > > > > ...
> > > > > ifndef NO_LIBDW
> > > > >   PERF_HAVE_DWARF := 1
> > > > > ...
> > > > > ```
> > > > > and in the arch/.../Makefiles change them to be:
> > > > > ```
> > > > > if PERF_HAVE_DWARF
> > > > > PERF_HAVE_DWARF_REGS := 1
> > > > > endif
> > > > > ```
> > > > > but this is going beyond the clean up this patch series was trying to
> > > > > achieve. I also don't know of an architecture where dwarf is present
> > > > > but registers are not, so having a definition for this case feels
> > > > > redundant.
> > > > >
> > > > > Thanks,
> > > > > Ian
> > > > >
> > > > > > Actually, there is a histrical reason I had called it is DWARF. I used to
> > > > > > use "libdwarf", and switched to "libdw" for required features. So I know
> > > > > > there are more than 1 implementation of DWARF library, and the libdwarf
> > > > > > also uses the same operation number because it depends on the same standard.
> > > > > >
> > > > > > https://github.com/davea42/libdwarf-code/blob/main/src/lib/libdwarf/dwarf.h#L809
> > > > > >
> > > > > > So I think we'd better keep it call as DWARF_REGS.
> > > > > >
> > > > > > Thank you,
> > > > > >
> > > > > > >
> > > > > > > Thanks,
> > > > > > > Ian
> > > > > > >
> > > > > >
> > > > > >
> > > > > > --
> > > > > > Masami Hiramatsu (Google) <mhiramat@...nel.org>
> > >
> >
> >
> > --
> > Masami Hiramatsu (Google) <mhiramat@...nel.org>
> 


-- 
Masami Hiramatsu (Google) <mhiramat@...nel.org>

Powered by blists - more mailing lists