Re: [rfc 2/6] dma-remap: add additional atomic pools to map to gfp mask

[Christoph Hellwig <hch@lst.de>]

On Sun, Mar 01, 2020 at 04:05:13PM -0800, David Rientjes wrote:
> The single atomic pool is allocated from the lowest zone possible since
> it is guaranteed to be applicable for any DMA allocation.
> 
> Devices may allocate through the DMA API but not have a strict reliance
> on GFP_DMA memory.  Since the atomic pool will be used for all
> non-blockable allocations, returning all memory from ZONE_DMA may
> unnecessarily deplete the zone.
> 
> Provision for multiple atomic pools that will map to the optimal gfp
> mask of the device.  These will be wired up in a subsequent patch.
> 
> Signed-off-by: David Rientjes <rientjes@google.com>
> ---
>  kernel/dma/remap.c | 75 +++++++++++++++++++++++++++-------------------
>  1 file changed, 45 insertions(+), 30 deletions(-)
> 
> diff --git a/kernel/dma/remap.c b/kernel/dma/remap.c
> --- a/kernel/dma/remap.c
> +++ b/kernel/dma/remap.c
> @@ -100,6 +100,8 @@ void dma_common_free_remap(void *cpu_addr, size_t size)
>  
>  #ifdef CONFIG_DMA_DIRECT_REMAP
>  static struct gen_pool *atomic_pool __ro_after_init;
> +static struct gen_pool *atomic_pool_dma32 __ro_after_init;
> +static struct gen_pool *atomic_pool_normal __ro_after_init;

Maybe rename atomic_pool as well as it really kinda looks like the
default at the moment?

>  
>  #define DEFAULT_DMA_COHERENT_POOL_SIZE  SZ_256K
>  static size_t atomic_pool_size __initdata = DEFAULT_DMA_COHERENT_POOL_SIZE;
> @@ -111,66 +113,79 @@ static int __init early_coherent_pool(char *p)
>  }
>  early_param("coherent_pool", early_coherent_pool);
>  
> -static gfp_t dma_atomic_pool_gfp(void)
> +static int __init __dma_atomic_pool_init(struct gen_pool **pool,
> +					 size_t pool_size, gfp_t gfp)
>  {

Can this just return the pool and return NULL (or an ERR_PTR) on
failure?