Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Mon, 22 May 2023 20:07:29 +0800
From:   Leo Yan <leo.yan@linaro.org>
To:     James Clark <james.clark@arm.com>
Cc:     Arnaldo Carvalho de Melo <acme@kernel.org>,
        John Garry <john.g.garry@oracle.com>,
        Will Deacon <will@kernel.org>,
        Mike Leach <mike.leach@linaro.org>,
        Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        Mark Rutland <mark.rutland@arm.com>,
        Alexander Shishkin <alexander.shishkin@linux.intel.com>,
        Jiri Olsa <jolsa@kernel.org>,
        Namhyung Kim <namhyung@kernel.org>,
        Ian Rogers <irogers@google.com>,
        Adrian Hunter <adrian.hunter@intel.com>,
        Guo Ren <guoren@kernel.org>,
        Paul Walmsley <paul.walmsley@sifive.com>,
        Palmer Dabbelt <palmer@dabbelt.com>,
        Albert Ou <aou@eecs.berkeley.edu>,
        Eric Lin <eric.lin@sifive.com>,
        Kan Liang <kan.liang@linux.intel.com>,
        Qi Liu <liuqi115@huawei.com>,
        Sandipan Das <sandipan.das@amd.com>,
        linux-kernel@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
        linux-perf-users@vger.kernel.org, linux-csky@vger.kernel.org,
        linux-riscv@lists.infradead.org
Subject: Re: [PATCH v1 2/5] perf parse-regs: Introduce functions
 arch__reg_{ip|sp}()
Message-ID: <20230522120729.GB1826292@leoy-yangtze.lan>
References: <20230520025537.1811986-1-leo.yan@linaro.org>
 <20230520025537.1811986-3-leo.yan@linaro.org>
 <839836e8-9600-9249-dcdb-e29519335141@arm.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <839836e8-9600-9249-dcdb-e29519335141@arm.com>
Precedence: bulk

On Mon, May 22, 2023 at 09:57:25AM +0100, James Clark wrote:
> 
> 
> On 20/05/2023 03:55, Leo Yan wrote:
> > Ideally, we want util/perf_regs.c to be general enough and doesn't bind
> > with specific architecture.
> > 
> > But since util/perf_regs.c uses the macros PERF_REG_IP and PERF_REG_SP
> > which are defined by architecture, thus util/perf_regs.c is dependent on
> > architecture header (see util/perf_regs.h includes "<perf_regs.h>", here
> > perf_regs.h is architecture specific header).
> > 
> > As a step to generalize util/perf_regs.c, this commit introduces weak
> > functions arch__reg_ip() and arch__reg_sp() and every architecture can
> > define their own functions; thus, util/perf_regs.c doesn't need to use
> > PERF_REG_IP and PERF_REG_SP anymore.
> > 
> > This is a preparation to get rid of architecture specific header from
> > util/perf_regs.h.
> > 
> > Signed-off-by: Leo Yan <leo.yan@linaro.org>
> > ---
> [...]
> >  
> > -#define DWARF_MINIMAL_REGS ((1ULL << PERF_REG_IP) | (1ULL << PERF_REG_SP))
> > +#define DWARF_MINIMAL_REGS ((1ULL << arch__reg_ip()) | (1ULL << arch__reg_sp()))
> >  
> >  const char *perf_reg_name(int id, const char *arch);
> >  int perf_reg_value(u64 *valp, struct regs_dump *regs, int id);
> > diff --git a/tools/perf/util/unwind-libdw.c b/tools/perf/util/unwind-libdw.c
> > index bdccfc511b7e..f308f2ea512b 100644
> > --- a/tools/perf/util/unwind-libdw.c
> > +++ b/tools/perf/util/unwind-libdw.c
> > @@ -252,7 +252,7 @@ int unwind__get_entries(unwind_entry_cb_t cb, void *arg,
> >  	if (!ui->dwfl)
> >  		goto out;
> >  
> > -	err = perf_reg_value(&ip, &data->user_regs, PERF_REG_IP);
> > +	err = perf_reg_value(&ip, &data->user_regs, arch__reg_ip());
> 
> Shouldn't it be more like this, because the weak symbols are a compile
> time thing and it's supposed to support cross arch unwinding at runtime
> (assuming something containing the arch from the file is passed down,
> like we did with perf_reg_name()):
> 
>   char *arch = perf_env__arch(evsel__env(evsel));
>   err = perf_reg_value(&ip, &data->user_regs, arch__reg_ip(arch));

Thanks for pointing out, James.

Agreed that we need to return the IP and SP register based on the
arch.  I will look into more details and spin for a new patch set for
this.

> Now I'm wondering how cross unwinding ever worked because I see
> libunwind also has something hard coded too:
> 
>   #define LIBUNWIND__ARCH_REG_SP PERF_REG_SP

Yeah, I also used arch__reg_sp() to replace PERF_REG_SP; but as you
suggestion, we should fix this with passing 'arch' parameter for
getting SP register based on arch.

Another important thing is to find a good test for cross unwinding.
Maybe I can use tools/perf/tests/shell/record.sh, function
test_register_capture() for testing registers, if you have any other
suggesion, please let me know.

Thanks,
Leo