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: [RFC WIP PATCH] powerpc/32: system call implement entry/exit
 logic in C
To:     Nicholas Piggin <npiggin@gmail.com>,
        Benjamin Herrenschmidt <benh@kernel.crashing.org>,
        Michael Ellerman <mpe@ellerman.id.au>, msuchanek@suse.de,
        Paul Mackerras <paulus@samba.org>
Cc:     linux-kernel@vger.kernel.org, linuxppc-dev@lists.ozlabs.org
References: <bbc0a09cfaf523bc00893253a7101362c98b31eb.1585667934.git.christophe.leroy@c-s.fr>
 <059c1abd-6be2-25ea-83e0-dcd411b7951b@c-s.fr>
 <1585898897.1jwur86s6a.astroid@bobo.none>
From:   Christophe Leroy <christophe.leroy@c-s.fr>
Message-ID: <fa79eac6-49e6-bbb0-5faa-7d52662a0cbe@c-s.fr>
Date:   Sun, 5 Apr 2020 20:53:00 +0200
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:68.0) Gecko/20100101
 Thunderbird/68.6.0
MIME-Version: 1.0
In-Reply-To: <1585898897.1jwur86s6a.astroid@bobo.none>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: fr
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


Le 03/04/2020 à 09:33, Nicholas Piggin a écrit :
> Christophe Leroy's on April 1, 2020 9:48 pm:
>>
>>
>> Le 31/03/2020 à 17:22, Christophe Leroy a écrit :
>>> That's first try to port PPC64 syscall entry/exit logic in C to PPC32.
>>> I've do the minimum to get it work. I have not reworked calls
>>> to sys_fork() and friends for instance.
>>>
>>> For the time being, it seems to work more or less but:
>>> - ping reports EINVAL on recvfrom
>>> - strace shows NULL instead of strings in call like open() for instance.
>>
>> For the two above problems, that's because system_call_exception()
>> doesn't set orig_gpr3 whereas DoSycall() does in entry_32.S . Is that
>> only done on PPC32 ?
>>
>> With the following line at the begining of system_call_exception(), it
>> works perfectly:
>>
>> 	regs->orig_gpr3 = r3;
> 
> Oh great, nice work. We should be able to make some simple helpers or
> move some things a bit to reduce the amount of ifdefs in the C code.
> It doesn't look too bad though.
> 
>> I will now focus on performance to see if we can do something about it.
> 
> What's the performance difference between current asm code just with
> always saving NVGPRS vs C?

Done new measurement and sent a series. lower values now because the 
time accounting was done twice as it was still in the ASM part.

Before the series, 311 cycles for a null_syscall
If adding SAVE_NVGPRS to the entry macro, 335 cycles

First patch: 353 cycles ie +13,5%

After a few changes, including conditional saving of non volatile 
registers, I get 325 cycles that is only +4,5%. I thing that's acceptable.

Do you see a problem with still saving non volatile registers only when 
required ?

Christophe