Received-SPF: pass (google.com: domain of linux-kernel+bounces-66300-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
Date: Thu, 15 Feb 2024 06:06:42 +0100
From: Christoph Hellwig <hch@lst.de>
To: Robin Murphy <robin.murphy@arm.com>
Cc: Alexander Lobakin <aleksander.lobakin@intel.com>,
	"David S. Miller" <davem@davemloft.net>,
	Eric Dumazet <edumazet@google.com>,
	Jakub Kicinski <kuba@kernel.org>, Paolo Abeni <pabeni@redhat.com>,
	Christoph Hellwig <hch@lst.de>,
	Marek Szyprowski <m.szyprowski@samsung.com>,
	Joerg Roedel <joro@8bytes.org>, Will Deacon <will@kernel.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	"Rafael J. Wysocki" <rafael@kernel.org>,
	Magnus Karlsson <magnus.karlsson@intel.com>,
	Maciej Fijalkowski <maciej.fijalkowski@intel.com>,
	Alexander Duyck <alexanderduyck@fb.com>, bpf@vger.kernel.org,
	netdev@vger.kernel.org, iommu@lists.linux.dev,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH net-next v3 1/7] dma: compile-out DMA sync op calls
 when not used
Message-ID: <20240215050642.GA4861@lst.de>
References: <20240214162201.4168778-1-aleksander.lobakin@intel.com> <20240214162201.4168778-2-aleksander.lobakin@intel.com> <893ad3a4-ba24-43cf-8200-b8cd7742622d@arm.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <893ad3a4-ba24-43cf-8200-b8cd7742622d@arm.com>
User-Agent: Mutt/1.5.17 (2007-11-01)

On Wed, Feb 14, 2024 at 05:20:50PM +0000, Robin Murphy wrote:
>>   +config DMA_NEED_SYNC
>> +	def_bool ARCH_HAS_SYNC_DMA_FOR_DEVICE || ARCH_HAS_SYNC_DMA_FOR_CPU || \
>> +		 ARCH_HAS_SYNC_DMA_FOR_CPU_ALL || DMA_OPS || SWIOTLB
>
> I'm not sure DMA_OPS belongs here - several architectures have non-trivial 
> ops without syncs, e.g. Alpha.

True, but peeking through the ops is a bit hard.  And I don't think it's
worth optimizing the dma sync performance on Alpha :)

>> +static inline void __dma_sync_single_for_cpu(struct device *dev,
>> +		dma_addr_t addr, size_t size, enum dma_data_direction dir)
>
> To me it would feel more logical to put all the wrappers inside the #ifdef 
> CONFIG_HAS_DMA and not touch these stubs at all (what does it mean to skip 
> an inline no-op?). Or in fact, if dma_skip_sync() is constant false for 
> !HAS_DMA, then we could also just make the external function declarations 
> unconditional and remove the stubs. Not a critical matter though, and I 
> defer to whatever Christoph thinks is most maintainable.

Your idea sounds reasonable to me, but I don't have a strong preference.

>> +static inline bool dma_need_sync(struct device *dev, dma_addr_t dma_addr)
>> +{
>> +	return !dma_skip_sync(dev) ? __dma_need_sync(dev, dma_addr) : false;
>> +}
>
> That's a bit of a mind-bender... is it actually just
>
> 	return !dma_skip_sync(dev) && __dma_need_sync(dev, dma_addr);
>
> ?

That looks a lot more readable for sure.

> (I do still think the negative flag makes it all a little harder to follow 
> in general than a positive "device needs to consider syncs" flag would.)

Probably.