Re: [PATCH] slab.h: Avoid using & for logical and of booleans

From: Vlastimil Babka <vbabka@suse.cz>
To: David Laight <David.Laight@ACULAB.COM>,
	'Bart Van Assche' <bvanassche@acm.org>,
	Andrew Morton <akpm@linux-foundation.org>
Cc: "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	Mel Gorman <mgorman@techsingularity.net>,
	Christoph Lameter <cl@linux.com>, Roman Gushchin <guro@fb.com>
Subject: Re: [PATCH] slab.h: Avoid using & for logical and of booleans
Date: Tue, 6 Nov 2018 11:22:27 +0100	[thread overview]
Message-ID: <e44e6c8b-e4e4-e7cb-a5ca-88e9559eb0d7@suse.cz> (raw)
In-Reply-To: <62188a351f2249188ce654ee03c894b1@AcuMS.aculab.com>

On 11/6/18 11:08 AM, David Laight wrote:
> From: Bart Van Assche
>> Sent: 05 November 2018 20:40
>>
>> This patch suppresses the following sparse warning:
>>
>> ./include/linux/slab.h:332:43: warning: dubious: x & !y

BTW, I wonder why the warnings appeared only now, after maybe months in
linux-next. Don't the various automated testing bots run sparse also on
linux-next?

>>
>> Fixes: 1291523f2c1d ("mm, slab/slub: introduce kmalloc-reclaimable caches")
>> Cc: Vlastimil Babka <vbabka@suse.cz>
>> Cc: Mel Gorman <mgorman@techsingularity.net>
>> Cc: Christoph Lameter <cl@linux.com>
>> Cc: Roman Gushchin <guro@fb.com>
>> Signed-off-by: Bart Van Assche <bvanassche@acm.org>
>> ---
>>  include/linux/slab.h | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/include/linux/slab.h b/include/linux/slab.h
>> index 918f374e7156..97d0599ddb7b 100644
>> --- a/include/linux/slab.h
>> +++ b/include/linux/slab.h
>> @@ -329,7 +329,7 @@ static __always_inline enum kmalloc_cache_type kmalloc_type(gfp_t flags)
>>  	 * If an allocation is both __GFP_DMA and __GFP_RECLAIMABLE, return
>>  	 * KMALLOC_DMA and effectively ignore __GFP_RECLAIMABLE
>>  	 */
>> -	return type_dma + (is_reclaimable & !is_dma) * KMALLOC_RECLAIM;
>> +	return type_dma + is_reclaimable * !is_dma * KMALLOC_RECLAIM;
> 
> ISTM that changing is_dma and is_reclaimable from int to bool will stop the bleating.
> 
> It is also strange that this code is trying so hard here to avoid conditional instructions

I primarily wanted to avoid branches in a hot path, not cmovs. Note
those are also not "free" (latency-wise) if the result of cmov is
immediately used for further computation.

> and then uses several to generate the boolean values in the first place.

I'm not sure where exactly?

> OTOH I'd probably write:
> 	int gfp_dma = 0;
> 
> #ifdef CONFIG_ZONE_DMA
> 	gfp_dma = __GFP_DMA;
> #endif
> 
> 	return flags & gfp_dma ? KMALLOC_DMA : flags & __GFP_RECLAIMABLE ? KMALLOC_RECLAIM : 0;

I'm not opposed to this. Christoph might :)

> 
> That might generate cmovs, but is may be better to put unlikely() around both
> conditional expressions. Or redo as:
> 
> 	return !unlikely(flags & (dfp_dma | __GFP_RECLAIMABLE)) ? 0 : flags & gfp_dma ? KMALLOC_DMA : KMALLOC_RECLAIM;

I guess it should be structured so that the fast path is for gfp without
both __GFP_DMA and __GFP_RECLAIMABLE, with a single test+branch. IIRC
that's what Christoph originally requested.

> 	David
> 
> -
> Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
> Registration No: 1397386 (Wales)
> 