Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Wed, 20 Oct 2021 15:24:19 +0200
From:   Jiri Olsa <jolsa@redhat.com>
To:     Lexi Shao <shaolexi@huawei.com>
Cc:     linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org,
        acme@kernel.org, mark.rutland@arm.com, peterz@infradead.org,
        mingo@redhat.com, alexander.shishkin@linux.intel.com,
        namhyung@kernel.org, qiuxi1@huawei.com, nixiaoming@huawei.com,
        wangbing6@huawei.com
Subject: Re: [PATCH] perf script: Show binary offsets for userspace addr
Message-ID: <YXAYg8N1vikCsOnU@krava>
References: <20211019072417.122576-1-shaolexi@huawei.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20211019072417.122576-1-shaolexi@huawei.com>
Precedence: bulk

On Tue, Oct 19, 2021 at 03:24:17PM +0800, Lexi Shao wrote:
> Show binary offsets for userspace addr with map in perf script output
> with callchain.
> 
> In commit 19610184693c("perf script: Show virtual addresses instead of
> offsets"), the addr shown in perf script output with callchain is changed
> from binary offsets to virtual address to fix the incorrectness when
> displaying symbol offset.
> 
> This is inconvenient in scenario that the binary is stripped and
> symbol cannot be resolved. If someone wants to further resolve symbols for
> specific binaries later, he would need an extra step to translate virtual
> address to binary offset with mapping information recorded in perf.data,
> which can be difficult for people not familiar with perf.
> 
> This patch modifies function sample__fprintf_callchain to print binary
> offset for userspace addr with dsos, and virtual address otherwise. It
> does not affect symbol offset calculation so symoff remains correct.
> 
> Before applying this patch:
> test  1512    78.711307:     533129 cycles:
> 	aaaae0da07f4 [unknown] (/tmp/test)
> 	aaaae0da0704 [unknown] (/tmp/test)
> 	ffffbe9f7ef4 __libc_start_main+0xe4 (/lib64/libc-2.31.so)
> 
> After this patch:
> test  1519   111.330127:     406953 cycles:
> 	7f4 [unknown] (/tmp/test)
> 	704 [unknown] (/tmp/test)
> 	20ef4 __libc_start_main+0xe4 (/lib64/libc-2.31.so)
> 
> Fixes: 19610184693c("perf script: Show virtual addresses instead of offsets")
> 
> Signed-off-by: Lexi Shao <shaolexi@huawei.com>

makes sense to me

Acked-/Tested-by Jiri Olsa <jolsa@redhat.com>

should we have a way to display these callchain detail in perf report output as well?

thanks,
jirka

> ---
>  tools/perf/util/evsel_fprintf.c | 12 +++++++++---
>  1 file changed, 9 insertions(+), 3 deletions(-)
> 
> diff --git a/tools/perf/util/evsel_fprintf.c b/tools/perf/util/evsel_fprintf.c
> index bfedd7b23521..8c2ea8001329 100644
> --- a/tools/perf/util/evsel_fprintf.c
> +++ b/tools/perf/util/evsel_fprintf.c
> @@ -11,6 +11,7 @@
>  #include "strlist.h"
>  #include "symbol.h"
>  #include "srcline.h"
> +#include "dso.h"
>  
>  static int comma_fprintf(FILE *fp, bool *first, const char *fmt, ...)
>  {
> @@ -144,12 +145,17 @@ int sample__fprintf_callchain(struct perf_sample *sample, int left_alignment,
>  			if (print_arrow && !first)
>  				printed += fprintf(fp, " <-");
>  
> -			if (print_ip)
> -				printed += fprintf(fp, "%c%16" PRIx64, s, node->ip);
> -
>  			if (map)
>  				addr = map->map_ip(map, node->ip);
>  
> +			if (print_ip) {
> +				/* Show binary offset for userspace addr */
> +				if (map && !map->dso->kernel)
> +					printed += fprintf(fp, "%c%16" PRIx64, s, addr);
> +				else
> +					printed += fprintf(fp, "%c%16" PRIx64, s, node->ip);
> +			}
> +
>  			if (print_sym) {
>  				printed += fprintf(fp, " ");
>  				node_al.addr = addr;
> -- 
> 2.12.3
>