Re: [PATCH 7/7] DWARF: add the config option

From: hpa@zytor.com
To: Ingo Molnar <mingo@kernel.org>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>,
	linux-kernel@vger.kernel.org, Jiri Slaby <jslaby@suse.cz>,
	Andrew Morton <akpm@linux-foundation.org>,
	live-patching@vger.kernel.org,
	Thomas Gleixner <tglx@linutronix.de>,
	Ingo Molnar <mingo@redhat.com>,
	the arch/x86 maintainers <x86@kernel.org>,
	Andy Lutomirski <luto@kernel.org>, Jiri Kosina <jikos@kernel.org>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Peter Zijlstra <a.p.zijlstra@chello.nl>
Subject: Re: [PATCH 7/7] DWARF: add the config option
Date: Fri, 26 May 2017 12:16:09 -0700	[thread overview]
Message-ID: <C652445D-1B84-4146-A2EA-DB820100BA78@zytor.com> (raw)
In-Reply-To: <20170523054906.q2x2hib73fo7wibh@gmail.com>

On May 22, 2017 10:49:06 PM PDT, Ingo Molnar <mingo@kernel.org> wrote:
>
>* H. Peter Anvin <hpa@zytor.com> wrote:
>
>> On 05/22/17 04:12, Ingo Molnar wrote:
>> \>>
>> >> This construct might be useful for other arches, which is why I
>called
>> >> it "FP" instead of "BP".  But then I ruined that with the last 3
>:-)
>> > 
>> > Please call it BP - 'FP' can easily be read as floating-point,
>making it all 
>> > super-confusing. We should use canonical x86 register names and
>ordering - even
>> > if not all registers are used straight away.
>> > 
>> 
>> Seriously, I suspect that at the end of the day we will have
>reinvented
>> DWARF.
>
>Absolutely - the main difference is:
>
>- the debug-info implementation is _internal_ to the kernel so it can
>be fixed
>instead of "oh, wait 2 years for the toolchain to fix this particular
>bug, work
>it around in the kernel meanwhile" kind of crazy flow and dependencies.
>I.e. 
>the debug-info generation and parsing code is both part of the kernel
>Git tree 
>   and can be iterated (and fixed) at once with.
>
>- the debug-info is auto-generated for assembly as well, leaving
>assembly code 
>   maintainable.
>
>- the debug-info has a sane data structure designed for robustness and 
>   compactness
>
>So even if it's a subset of the existing complexity of dwarf et al we
>are still 
>literally infinitely better off with this model.
>
>Thanks,
>
>	Ingo

This assumes that it actually ends up being feasible for objtool to do so.

It is worth noting that using DWARF for unwinding vs auto-generating the unwind information are independent issues.

Another option is to use (or postprocess) the compiler-generated DWARF for C modules and pursue autogeneration only for assembly modules, which ought to be a much easier problem and is less dependent on discovering new compiler-generated patterns.
-- 
Sent from my Android device with K-9 Mail. Please excuse my brevity.