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 v2 3/5] powerpc/mm: Allow more than 16 low slices
To:     Christophe LEROY <christophe.leroy@c-s.fr>,
        Benjamin Herrenschmidt <benh@kernel.crashing.org>,
        Paul Mackerras <paulus@samba.org>,
        Michael Ellerman <mpe@ellerman.id.au>,
        Scott Wood <oss@buserror.net>
Cc:     linux-kernel@vger.kernel.org, linuxppc-dev@lists.ozlabs.org
References: <49148d07955d3e5f963cedf9adcfcc37c3e03ef4.1516179904.git.christophe.leroy@c-s.fr>
 <1c9752ac98fd3278ef448e2553053c287af42b3f.1516179904.git.christophe.leroy@c-s.fr>
 <87po66z1w2.fsf@linux.vnet.ibm.com>
 <baf3a2db-3522-8419-0eda-ef62ad13d76e@c-s.fr>
From:   "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Date:   Fri, 19 Jan 2018 14:36:22 +0530
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
 Thunderbird/52.5.2
MIME-Version: 1.0
In-Reply-To: <baf3a2db-3522-8419-0eda-ef62ad13d76e@c-s.fr>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Message-Id: <26a6d845-cf38-d8c6-5f54-a57ddb2a5a77@linux.vnet.ibm.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


On 01/19/2018 02:29 PM, Christophe LEROY wrote:
> 
> 
> Le 19/01/2018 à 09:30, Aneesh Kumar K.V a écrit :
>> Christophe Leroy <christophe.leroy@c-s.fr> writes:
>>
>>> While the implementation of the "slices" address space allows
>>> a significant amount of high slices, it limits the number of
>>> low slices to 16 due to the use of a single u64 low_slices_psize
>>> element in struct mm_context_t
>>>
>>> On the 8xx, the minimum slice size is the size of the area
>>> covered by a single PMD entry, ie 4M in 4K pages mode and 64M in
>>> 16K pages mode. This means we could have resp. up to 1024 and 64
>>> slices.
>>>
>>> In order to override this limitation, this patch switches the
>>> handling of low_slices to BITMAPs as done already for high_slices.
>>
>> Does it have a performance impact. When we switched high_slices
>> that was one of the question asked. Now with a topdown search we should
>> mostly be using the high_slices. But it will good to get numbers for
>> ppc64 for this change.
> 
> It should have almost no performance impact at all, because all bitmap 
> functions used a simplified way when the number of bits is small and 
> constant:
> 
> -    ret->low_slices = 0;
> +    slice_bitmap_zero(ret->low_slices, SLICE_NUM_LOW);
> 
> 
> static inline void bitmap_zero(unsigned long *dst, unsigned int nbits)
> {
>      if (small_const_nbits(nbits))
>          *dst = 0UL;
>      else {
>          unsigned int len = BITS_TO_LONGS(nbits) * sizeof(unsigned long);
>          memset(dst, 0, len);
>      }
> }
> 
> 
> 
> -    dst->low_slices |= src->low_slices;
> +    slice_bitmap_or(dst->low_slices, dst->low_slices, src->low_slices,
> +            SLICE_NUM_LOW);
> 
> 
> static inline void bitmap_or(unsigned long *dst, const unsigned long *src1,
>              const unsigned long *src2, unsigned int nbits)
> {
>      if (small_const_nbits(nbits))
>          *dst = *src1 | *src2;
>      else
>          __bitmap_or(dst, src1, src2, nbits);
> }
> 
> 
>

may be capture that in commit message saying since we are 64 bit on 
ppc64 there is no impact there?

-aneesh