Return-Path: <linux-kernel-owner@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
        id S945272AbcJaRxF (ORCPT <rfc822;w@1wt.eu>);
        Mon, 31 Oct 2016 13:53:05 -0400
Received: from userp1040.oracle.com ([156.151.31.81]:24993 "EHLO
        userp1040.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
        with ESMTP id S938549AbcJaRxC (ORCPT
        <rfc822;linux-kernel@vger.kernel.org>);
        Mon, 31 Oct 2016 13:53:02 -0400
Date: Mon, 31 Oct 2016 13:52:52 -0400
From: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
To: Geert Uytterhoeven <geert+renesas@glider.be>
Cc: Jonathan Corbet <corbet@lwn.net>, Magnus Damm <magnus.damm@gmail.com>,
        linux-kernel@vger.kernel.org, linux-doc@vger.kernel.org,
        iommu@lists.linux-foundation.org, linux-renesas-soc@vger.kernel.org
Subject: Re: [PATCH 2/2] swiotlb: Add swiotlb=nobounce debug option
Message-ID: <20161031175252.GA6952@char.us.oracle.com>
References: <1477928704-10611-1-git-send-email-geert+renesas@glider.be>
 <1477928704-10611-3-git-send-email-geert+renesas@glider.be>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1477928704-10611-3-git-send-email-geert+renesas@glider.be>
User-Agent: Mutt/1.6.2 (2016-07-01)
X-Source-IP: aserv0022.oracle.com [141.146.126.234]
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org
Content-Length: 3548
Lines: 104

On Mon, Oct 31, 2016 at 04:45:04PM +0100, Geert Uytterhoeven wrote:
> On architectures like arm64, swiotlb is tied intimately to the core
> architecture DMA support. In addition, ZONE_DMA cannot be disabled.
> 
> To aid debugging and catch devices not supporting DMA to memory outside
> the 32-bit address space, add a kernel command line option
> "swiotlb=nobounce", which disables the use of bounce buffers.
> If specified, trying to map memory that cannot be used with DMA will
> fail, and a warning will be printed (rate-limited).

I would make the 'swiotlb_force' an enum. And then instead of this
being 'nobounce' just do the inverse of 'force', that is the
'noforce' would trigger this no bounce effect.

So:

enum {
	NORMAL,		/* Default - depending on the hardware DMA mask and such. */
	FORCE,		/* swiotlb=force */
	NO_FORCE,	/* swiotlb=noforce */
}
> 
> Note that io_tlb_nslabs is set to 1, which is the minimal supported
> value.
> 
> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
> ---
>  Documentation/kernel-parameters.txt |  3 ++-
>  lib/swiotlb.c                       | 19 +++++++++++++++++--
>  2 files changed, 19 insertions(+), 3 deletions(-)
> 
> diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
> index 37babf91f2cb6de2..38556cdceabaf087 100644
> --- a/Documentation/kernel-parameters.txt
> +++ b/Documentation/kernel-parameters.txt
> @@ -3998,10 +3998,11 @@ bytes respectively. Such letter suffixes can also be entirely omitted.
>  			it if 0 is given (See Documentation/cgroup-v1/memory.txt)
>  
>  	swiotlb=	[ARM,IA-64,PPC,MIPS,X86]
> -			Format: { <int> | force }
> +			Format: { <int> | force | nobounce }
>  			<int> -- Number of I/O TLB slabs
>  			force -- force using of bounce buffers even if they
>  			         wouldn't be automatically used by the kernel
> +			nobounce -- Never use bounce buffers (for debugging)
>  
>  	switches=	[HW,M68k]
>  
> diff --git a/lib/swiotlb.c b/lib/swiotlb.c
> index 6ce764410ae475cc..4550e6b516c2a4c0 100644
> --- a/lib/swiotlb.c
> +++ b/lib/swiotlb.c
> @@ -54,6 +54,7 @@
>  #define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT)
>  
>  int swiotlb_force;
> +static int swiotlb_nobounce;
>  
>  /*
>   * Used to do a quick range check in swiotlb_tbl_unmap_single and
> @@ -106,8 +107,12 @@
>  	}
>  	if (*str == ',')
>  		++str;
> -	if (!strcmp(str, "force"))
> +	if (!strcmp(str, "force")) {
>  		swiotlb_force = 1;
> +	} else if (!strcmp(str, "nobounce")) {
> +		swiotlb_nobounce = 1;
> +		io_tlb_nslabs = 1;
> +	}
>  
>  	return 0;
>  }
> @@ -541,8 +546,15 @@ phys_addr_t swiotlb_tbl_map_single(struct device *hwdev,
>  map_single(struct device *hwdev, phys_addr_t phys, size_t size,
>  	   enum dma_data_direction dir)
>  {
> -	dma_addr_t start_dma_addr = phys_to_dma(hwdev, io_tlb_start);
> +	dma_addr_t start_dma_addr;
> +
> +	if (swiotlb_nobounce) {
> +		dev_warn_ratelimited(hwdev, "Cannot do DMA to address %pa\n",
> +				     &phys);
> +		return SWIOTLB_MAP_ERROR;
> +	}
>  
> +	start_dma_addr = phys_to_dma(hwdev, io_tlb_start);
>  	return swiotlb_tbl_map_single(hwdev, start_dma_addr, phys, size, dir);
>  }
>  
> @@ -707,6 +719,9 @@ void swiotlb_tbl_sync_single(struct device *hwdev, phys_addr_t tlb_addr,
>  swiotlb_full(struct device *dev, size_t size, enum dma_data_direction dir,
>  	     int do_panic)
>  {
> +	if (swiotlb_nobounce)
> +		return;
> +
>  	/*
>  	 * Ran out of IOMMU space for this operation. This is very bad.
>  	 * Unfortunately the drivers cannot handle this operation properly.
> -- 
> 1.9.1
>