Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH 10/10] powerpc: select DYNAMIC_DEBUG_RELATIVE_POINTERS for
 PPC64
To:     Andrew Morton <akpm@linux-foundation.org>,
        Christophe Leroy <christophe.leroy@c-s.fr>
Cc:     linuxppc-dev@lists.ozlabs.org, Jason Baron <jbaron@akamai.com>,
        linux-kernel@vger.kernel.org
References: <20190409212517.7321-1-linux@rasmusvillemoes.dk>
 <20190409212517.7321-11-linux@rasmusvillemoes.dk>
 <b5304c4c-736a-fd8b-e025-4ea040f5c346@c-s.fr>
 <20190423123611.1f7276cc4eb9f7a7005899a9@linux-foundation.org>
From:   Rasmus Villemoes <linux@rasmusvillemoes.dk>
Message-ID: <0fbffd5d-7e2a-1b12-5c94-b75f3f5d7ac1@rasmusvillemoes.dk>
Date:   Wed, 24 Apr 2019 08:46:40 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.6.1
MIME-Version: 1.0
In-Reply-To: <20190423123611.1f7276cc4eb9f7a7005899a9@linux-foundation.org>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 23/04/2019 21.36, Andrew Morton wrote:
> On Tue, 23 Apr 2019 17:37:33 +0200 Christophe Leroy <christophe.leroy@c-s.fr> wrote:
> 
>>> --- a/arch/powerpc/Kconfig
>>> +++ b/arch/powerpc/Kconfig
>>> @@ -155,6 +155,7 @@ config PPC
>>>   	select BUILDTIME_EXTABLE_SORT
>>>   	select CLONE_BACKWARDS
>>>   	select DCACHE_WORD_ACCESS		if PPC64 && CPU_LITTLE_ENDIAN
>>> +	select DYNAMIC_DEBUG_RELATIVE_POINTERS	if PPC64
>>
>> Why only PPC64 ? Won't it work with PPC32 ? Or would it be 
>> counter-performant on 32 bits ?
> 
> Ditto arm and i386?
> 

It's pointless on 32-bit platforms - I'm replacing absolute const char*
pointers with a relative s32 offset from the _ddebug descriptor, so if
sizeof(void*)==4 there's no gain.

And yes, the current implementation also wouldn't work out-of-the-box
for 32-bit platforms, since the asm needs to know how to properly
initialize a whole struct _ddebug, which (often) contains a static_key,
which in turn contains a pointer member, which both affects its size as
well as its placement inside _ddebug. The C code in dynamic_debug.c
would likely Just Work, but there's no point in complicating the asm
part for no gain, so there are static_assert()s in place to ensure
BITS_PER_LONG==64 (as well as checking the various offsetof()s etc.).

[I don't think performance matters at all, it's one extra addition to
access these fields, and that is only done in the rare cases where
someone interacts with the dynamic_debug/control sysfs file, and when
one of the activated pr_debug()s is actually hit (so a few extra
instructions drown in the printk overhead).]

I do now see that PPC64 does not select GENERIC_BUG_RELATIVE_POINTERS,
so maybe this scheme simply doesn't work on PPC64, or nobody has done
the work to reduce the sizeof(struct bug_entry) on PPC64? As I said,
I've only compile-tested arm64 and ppc64.

Rasmus