[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20200305120738.GA31049@leoy-ThinkPad-X240s>
Date: Thu, 5 Mar 2020 20:07:38 +0800
From: Leo Yan <leo.yan@...aro.org>
To: "Naveen N. Rao" <naveen.n.rao@...ux.vnet.ibm.com>
Cc: Arnaldo Carvalho de Melo <acme@...nel.org>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
Allison Randal <allison@...utok.net>,
Hendrik Brueckner <brueckner@...ux.vnet.ibm.com>,
Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
Enrico Weigelt <info@...ux.net>,
John Garry <john.garry@...wei.com>,
Jiri Olsa <jolsa@...hat.com>,
Kate Stewart <kstewart@...uxfoundation.org>,
linux-kernel@...r.kernel.org, Mark Rutland <mark.rutland@....com>,
Mathieu Poirier <mathieu.poirier@...aro.org>,
Mike Leach <mike.leach@...aro.org>,
Ingo Molnar <mingo@...hat.com>,
Namhyung Kim <namhyung@...nel.org>,
Peter Zijlstra <peterz@...radead.org>,
Thomas Gleixner <tglx@...utronix.de>,
Thomas Richter <tmricht@...ux.vnet.ibm.com>
Subject: Re: [PATCH v2] perf symbols: Consolidate symbol fixup issue
Hi Naveen,
On Thu, Mar 05, 2020 at 04:32:30PM +0530, Naveen N. Rao wrote:
> Leo Yan wrote:
> > After copying Arm64's perf archive with object files and perf.data file
> > to x86 laptop, the x86's perf kernel symbol resolution fails. It
> > outputs 'unknown' for all symbols parsing.
> >
> > This issue is root caused by the function elf__needs_adjust_symbols(),
> > x86 perf tool uses one weak version, Arm64 (and powerpc) has rewritten
> > their own version. elf__needs_adjust_symbols() decides if need to parse
> > symbols with the relative offset address; but x86 building uses the weak
> > function which misses to check for the elf type 'ET_DYN', so that it
> > cannot parse symbols in Arm DSOs due to the wrong result from
> > elf__needs_adjust_symbols().
> >
> > The DSO parsing should not depend on any specific architecture perf
> > building; e.g. x86 perf tool can parse Arm and Arm64 DSOs, vice versa.
> > So this patch changes elf__needs_adjust_symbols() as a common function
> > and removes the arch specific functions for Arm64 and powerpc.
> >
> > In the common elf__needs_adjust_symbols(), it checks elf header and if
> > the machine type is one of Arm64/ppc/ppc64, it checks extra condition
> > for 'ET_DYN'. Finally, the Arm64 DSO can be parsed properly with x86's
> > perf tool.
> >
> > Before:
> >
> > # perf script
> > main 3258 1 branches: 0 [unknown] ([unknown]) => ffff800010c4665c [unknown] ([kernel.kallsyms])
> > main 3258 1 branches: ffff800010c46670 [unknown] ([kernel.kallsyms]) => ffff800010c4eaec [unknown] ([kernel.kallsyms])
> > main 3258 1 branches: ffff800010c4eaec [unknown] ([kernel.kallsyms]) => ffff800010c4eb00 [unknown] ([kernel.kallsyms])
> > main 3258 1 branches: ffff800010c4eb08 [unknown] ([kernel.kallsyms]) => ffff800010c4e780 [unknown] ([kernel.kallsyms])
> > main 3258 1 branches: ffff800010c4e7a0 [unknown] ([kernel.kallsyms]) => ffff800010c4eeac [unknown] ([kernel.kallsyms])
> > main 3258 1 branches: ffff800010c4eebc [unknown] ([kernel.kallsyms]) => ffff800010c4ed80 [unknown] ([kernel.kallsyms])
> >
> > After:
> >
> > # perf script
> > main 3258 1 branches: 0 [unknown] ([unknown]) => ffff800010c4665c coresight_timeout+0x54 ([kernel.kallsyms])
> > main 3258 1 branches: ffff800010c46670 coresight_timeout+0x68 ([kernel.kallsyms]) => ffff800010c4eaec etm4_enable_hw+0x3cc ([kernel.kallsyms])
> > main 3258 1 branches: ffff800010c4eaec etm4_enable_hw+0x3cc ([kernel.kallsyms]) => ffff800010c4eb00 etm4_enable_hw+0x3e0 ([kernel.kallsyms])
> > main 3258 1 branches: ffff800010c4eb08 etm4_enable_hw+0x3e8 ([kernel.kallsyms]) => ffff800010c4e780 etm4_enable_hw+0x60 ([kernel.kallsyms])
> > main 3258 1 branches: ffff800010c4e7a0 etm4_enable_hw+0x80 ([kernel.kallsyms]) => ffff800010c4eeac etm4_enable+0x2d4 ([kernel.kallsyms])
> > main 3258 1 branches: ffff800010c4eebc etm4_enable+0x2e4 ([kernel.kallsyms]) => ffff800010c4ed80 etm4_enable+0x1a8 ([kernel.kallsyms])
> >
>
> I am not able to reproduce this since powerpc64 kernels are not being built
> as ET_EXEC anymore.
Thanks for reviewing!
Based on the context, I think you mean powerpc64 kernels are not being
built as ET_DYN anymore (and now change to ET_EXEC).
> > v2: Fixed Arm64 and powerpc native building.
> >
> > Reported-by: Mike Leach <mike.leach@...aro.org>
> > Signed-off-by: Leo Yan <leo.yan@...aro.org>
> > ---
> > tools/perf/arch/arm64/util/Build | 1 -
> > tools/perf/arch/arm64/util/sym-handling.c | 19 -------------------
> > tools/perf/arch/powerpc/util/Build | 1 -
> > tools/perf/arch/powerpc/util/sym-handling.c | 10 ----------
> > tools/perf/util/symbol-elf.c | 8 +++++++-
> > 5 files changed, 7 insertions(+), 32 deletions(-)
> > delete mode 100644 tools/perf/arch/arm64/util/sym-handling.c
> >
> > diff --git a/tools/perf/arch/arm64/util/Build b/tools/perf/arch/arm64/util/Build
> > index 0a7782c61209..789956f76d85 100644
> > --- a/tools/perf/arch/arm64/util/Build
> > +++ b/tools/perf/arch/arm64/util/Build
> > @@ -1,6 +1,5 @@
> > perf-y += header.o
> > perf-y += perf_regs.o
> > -perf-y += sym-handling.o
> > perf-$(CONFIG_DWARF) += dwarf-regs.o
> > perf-$(CONFIG_LOCAL_LIBUNWIND) += unwind-libunwind.o
> > perf-$(CONFIG_LIBDW_DWARF_UNWIND) += unwind-libdw.o
> > diff --git a/tools/perf/arch/arm64/util/sym-handling.c b/tools/perf/arch/arm64/util/sym-handling.c
> > deleted file mode 100644
> > index 8dfa3e5229f1..000000000000
> > --- a/tools/perf/arch/arm64/util/sym-handling.c
> > +++ /dev/null
> > @@ -1,19 +0,0 @@
> > -// SPDX-License-Identifier: GPL-2.0-only
> > -/*
> > - *
> > - * Copyright (C) 2015 Naveen N. Rao, IBM Corporation
> > - */
> > -
> > -#include "symbol.h" // for the elf__needs_adjust_symbols() prototype
> > -#include <stdbool.h>
> > -
> > -#ifdef HAVE_LIBELF_SUPPORT
> > -#include <gelf.h>
> > -
> > -bool elf__needs_adjust_symbols(GElf_Ehdr ehdr)
> > -{
> > - return ehdr.e_type == ET_EXEC ||
> > - ehdr.e_type == ET_REL ||
> > - ehdr.e_type == ET_DYN;
> > -}
> > -#endif
> > diff --git a/tools/perf/arch/powerpc/util/Build b/tools/perf/arch/powerpc/util/Build
> > index 7cf0b8803097..e5c9504f8586 100644
> > --- a/tools/perf/arch/powerpc/util/Build
> > +++ b/tools/perf/arch/powerpc/util/Build
> > @@ -1,5 +1,4 @@
> > perf-y += header.o
> > -perf-y += sym-handling.o
> > perf-y += kvm-stat.o
> > perf-y += perf_regs.o
> > perf-y += mem-events.o
> > diff --git a/tools/perf/arch/powerpc/util/sym-handling.c b/tools/perf/arch/powerpc/util/sym-handling.c
> > index abb7a12d8f93..0856b32f9e08 100644
> > --- a/tools/perf/arch/powerpc/util/sym-handling.c
> > +++ b/tools/perf/arch/powerpc/util/sym-handling.c
> > @@ -10,16 +10,6 @@
> > #include "probe-event.h"
> > #include "probe-file.h"
> >
> > -#ifdef HAVE_LIBELF_SUPPORT
> > -bool elf__needs_adjust_symbols(GElf_Ehdr ehdr)
> > -{
> > - return ehdr.e_type == ET_EXEC ||
> > - ehdr.e_type == ET_REL ||
> > - ehdr.e_type == ET_DYN;
> > -}
> > -
> > -#endif
> > -
> > int arch__choose_best_symbol(struct symbol *syma,
> > struct symbol *symb __maybe_unused)
> > {
> > diff --git a/tools/perf/util/symbol-elf.c b/tools/perf/util/symbol-elf.c
> > index 1965aefccb02..ee788ac67415 100644
> > --- a/tools/perf/util/symbol-elf.c
> > +++ b/tools/perf/util/symbol-elf.c
> > @@ -704,8 +704,14 @@ void symsrc__destroy(struct symsrc *ss)
> > close(ss->fd);
> > }
> >
> > -bool __weak elf__needs_adjust_symbols(GElf_Ehdr ehdr)
> > +bool elf__needs_adjust_symbols(GElf_Ehdr ehdr)
> > {
> > + if (ehdr.e_machine == EM_AARCH64 ||
> > + ehdr.e_machine == EM_PPC ||
> > + ehdr.e_machine == EM_PPC64)
> > + return ehdr.e_type == ET_EXEC || ehdr.e_type == ET_REL ||
> > + ehdr.e_type == ET_DYN;
> > +
> > return ehdr.e_type == ET_EXEC || ehdr.e_type == ET_REL;
>
> Patch looks good to me. However:
Can I add your review tag?
> This is only used for checking kernel, so I wonder if we can simply include
> check for ET_DYN across all architectures? This would only matter if there
> are architectures building their kernel as ET_DYN that _don't_ want to
> adjust symbols.
Seems only Arm64 enables the link option '-share' for LDFLAGS_vmlinux;
I confirmed with below command:
$ find arch -name 'Makefile' -exec grep -n '\-share' {} + | grep vmlinux
arch/arm64/Makefile:21:LDFLAGS_vmlinux += -shared -Bsymbolic -z notext -z norelro \
Also reviewed the output for searching '\-share' for all Makefiles under
'arch' folder, many architectures use it for vdso but only Arm64 enables
'-share' for vmlinux linkage. If so, your suggestion is valid and we
can simply include check for ET_DYN for all archs (and it's better to
add comment for this).
I'd like to wait a bit for anyone has other ideas, and if no objection
will send out new patch for this.
Thanks,
Leo
Powered by blists - more mailing lists