Date: Thu, 4 Jan 2018 01:53:25 -0800
From: Christoph Hellwig <hch@infradead.org>
To: Christian =?iso-8859-1?Q?K=F6nig?= 
        <ckoenig.leichtzumerken@gmail.com>
Cc: konrad.wilk@oracle.com, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] swiotlb: suppress warning when __GFP_NOWARN is set v4
Message-ID: <20180104095325.GB13091@infradead.org>
References: <20180104093233.16311-1-christian.koenig@amd.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20180104093233.16311-1-christian.koenig@amd.com>
User-Agent: Mutt/1.9.1 (2017-09-22)
Sender: linux-kernel-owner@vger.kernel.org

This seems to collide with my dma direct/swiotlb series posted recently.

> +++ b/lib/swiotlb.c
> @@ -490,11 +490,11 @@ static void swiotlb_bounce(phys_addr_t orig_addr, phys_addr_t tlb_addr,
>  	}
>  }
>  
> -phys_addr_t swiotlb_tbl_map_single(struct device *hwdev,
> -				   dma_addr_t tbl_dma_addr,
> -				   phys_addr_t orig_addr, size_t size,
> -				   enum dma_data_direction dir,
> -				   unsigned long attrs)
> +static phys_addr_t tbl_map_single(struct device *hwdev,
> +				  dma_addr_t tbl_dma_addr,
> +				  phys_addr_t orig_addr, size_t size,
> +				  enum dma_data_direction dir,
> +				  unsigned long attrs, bool warn)

We already have DMA_ATTR_NO_WARN which can be passed in attrs.  Please
use it instead of reinventing it.

>  swiotlb_alloc_coherent(struct device *hwdev, size_t size,
>  		       dma_addr_t *dma_handle, gfp_t flags)
>  {
> +	bool warn = !(flags & __GFP_NOWARN);
>  	dma_addr_t dev_addr;
>  	void *ret;
>  	int order = get_order(size);
> @@ -739,7 +750,7 @@ swiotlb_alloc_coherent(struct device *hwdev, size_t size,
>  		 * will grab memory from the lowest available address range.
>  		 */
>  		phys_addr_t paddr = map_single(hwdev, 0, size,
> -					       DMA_FROM_DEVICE, 0);
> +					       DMA_FROM_DEVICE, 0, warn);

Note: in my above series swiotlb_alloc_coherent is going away, and
replaced with a swiotlb_alloc that takes a dma_attrs argument.

Using that for passing the nowarn flag is the right way to go instead
of using __GFP_NOWARN.