lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20220814053124.fsj3bqamrqyfoiey@google.com>
Date:   Sat, 13 Aug 2022 22:31:24 -0700
From:   Fangrui Song <maskray@...gle.com>
To:     Dmitrii Bundin <dmitrii.bundin.a@...il.com>,
        Masahiro Yamada <masahiroy@...nel.org>
Cc:     Andrew Morton <akpm@...ux-foundation.org>,
        Michal Marek <michal.lkml@...kovi.net>,
        Nick Desaulniers <ndesaulniers@...gle.com>,
        Nathan Chancellor <nathan@...nel.org>,
        "Peter Zijlstra (Intel)" <peterz@...radead.org>,
        Kees Cook <keescook@...omium.org>,
        Josh Poimboeuf <jpoimboe@...nel.org>,
        Dan Williams <dan.j.williams@...el.com>,
        Isabella Basso <isabbasso@...eup.net>,
        Eric Dumazet <edumazet@...gle.com>,
        Vlastimil Babka <vbabka@...e.cz>,
        Rasmus Villemoes <linux@...musvillemoes.dk>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Linux Kbuild mailing list <linux-kbuild@...r.kernel.org>
Subject: Re: [PATCH v2] kbuild: add configurable debug info level

On 2022-08-14, Masahiro Yamada wrote:
>+CC: Fangrui Song <maskray@...gle.com>
>
>
>
>
>On Sun, Aug 14, 2022 at 9:25 AM Dmitrii Bundin
><dmitrii.bundin.a@...il.com> wrote:
>>
>> Provides a way to configure debug info level (-glevel).
>> Debug level 3 includes extra information such as macro definitions. With
>> level 3 enabled it's possible to expand macros right from the debugging
>> session in gdb simplifying debugging when complicated macros involved.
>> The default level is set to 2 to not change the default build behavior.

GCC -g3 generates macro information (in the .debug_macro section).

In Clang, -g = -g2 = -g3. To generate macro information,
specify -fdebug-macro.
The different choice is known in the initial implementation https://reviews.llvm.org/D16135 .

Not generating macro information for -g3 (i.e. diverging from GCC
behavior) makes some sense to me: -fstandalone-debug will probably be
more suitable as -g3 (it retains some type debug info for C++ (the code
after https://github.com/llvm/llvm-project/blob/b2f31cac28c8a03ceb908b544f5790f4f9f2d9ab/clang/lib/CodeGen/CGDebugInfo.cpp#L2497-L2499).

>> Signed-off-by: Dmitrii Bundin <dmitrii.bundin.a@...il.com>
>> ---
>>
>> Changes in v2: https://lore.kernel.org/all/20220804223504.4739-1-dmitrii.bundin.a@gmail.com/
>>   - Replace hardcoded -g3 with a configurable debug info level
>>
>>  lib/Kconfig.debug      | 11 +++++++++++
>>  scripts/Makefile.debug |  2 +-
>>  2 files changed, 12 insertions(+), 1 deletion(-)
>>
>> diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
>> index 2e24db4bff19..a17c12c20290 100644
>> --- a/lib/Kconfig.debug
>> +++ b/lib/Kconfig.debug
>> @@ -304,6 +304,17 @@ config DEBUG_INFO_REDUCED
>>           DEBUG_INFO build and compile times are reduced too.
>>           Only works with newer gcc versions.
>>
>> +config DEBUG_INFO_LEVEL
>> +       int "Debug info level"
>> +       range 0 3
>> +       default "2"
>> +       help
>> +         Sets the level of how much debug information to generate (-glevel).
>> +         Level 1 produces minimal debug information without including information
>> +         about local variables. Level 3 includes extra information like macro
>> +         definitions. Setting up level 3 will require significantly more disk
>> +         space and increase built time. Level 0 produces no debug information.
>> +
>
>
>
>We already have CONFIG_DEBUG_INFO_NONE to
>disable the debug info.
>
>
>The combination of CONFIG_DEBUG_INFO=y and
>CONFIG_DEBUG_INFO_LEVEL=0  (-g0)
>would emulate CONFIG_DEBUG_INFO_NONE ?
>
>
>
>Using 'int' does not look sensible to me.
>
>
>
>
>
>>  config DEBUG_INFO_COMPRESSED
>>         bool "Compressed debugging information"
>>         depends on $(cc-option,-gz=zlib)
>> diff --git a/scripts/Makefile.debug b/scripts/Makefile.debug
>> index 9f39b0130551..28beffc42e71 100644
>> --- a/scripts/Makefile.debug
>> +++ b/scripts/Makefile.debug
>> @@ -3,7 +3,7 @@ DEBUG_CFLAGS    :=
>>  ifdef CONFIG_DEBUG_INFO_SPLIT
>>  DEBUG_CFLAGS   += -gsplit-dwarf
>>  else
>> -DEBUG_CFLAGS   += -g
>> +DEBUG_CFLAGS   += -g$(CONFIG_DEBUG_INFO_LEVEL)
>>  endif
>>
>>  ifndef CONFIG_AS_IS_LLVM
>> --
>> 2.17.1
>>
>
>
>I want to consult Fangrui Song for this part.
>
>
>With this Makefile code, CONFIG_DEBUG_INFO_SPLIT
>takes the presidency over CONFIG_DEBUG_INFO_LEVEL.
>
>
>When CONFIG_DEBUG_INFO_SPLIT is enabled (-gsplit-dwarf),
>it always uses the default -g2 level.
>CONFIG_DEBUG_INFO_LEVEL is just ignored silently.
>
>
>
>It might be sensible in older GCC/Clang behavior because
>-gsplit-dwarf implied -g2.
>
>
>But, with this commit:
>https://reviews.llvm.org/D80391
>
>-gsplit-dwarf and -g<level> are orthogonal
>for GCC 11+/Clang 12+, correct?

Correct.

>I think "splitting debug files" and "debug level"
>should be controlled independently.
>(but it depends on the compiler version, if I understood correctly)

Before GCC 11 and Clang 12, -gsplit-dwarf implied -g2 (older
-gsplit-dwarf is like today's `-gsplit-dwarf -g2`).

GCC 11 and Clang 12 (https://reviews.llvm.org/D80391) have changed
-gsplit-dwarf to not imply -g2.

For a group of -g0 -g1 -g2, the last option wins.  Therefore,

-g0 -gsplit-dwarf => debug info in GCC<11 and Clang<12
-g0 -gsplit-dwarf => no debug info in GCC>=11 and Clang>=12

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ