linux-kernel - Re: [PATCH 1/2] of: device: Allow DMA range map to be set before of_dma_configure

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <ddf44e96-187b-91c0-822d-ade4f1e5be2b@arm.com>
Date:   Sat, 16 Jan 2021 14:57:43 +0000
From:   Robin Murphy <robin.murphy@....com>
To:     Paul Kocialkowski <paul.kocialkowski@...tlin.com>,
        devicetree@...r.kernel.org, linux-kernel@...r.kernel.org,
        linux-arm-kernel@...ts.infradead.org
Cc:     Jernej Skrabec <jernej.skrabec@...l.net>,
        Daniel Vetter <daniel.vetter@...ll.ch>,
        Maxime Ripard <mripard@...nel.org>,
        Chen-Yu Tsai <wens@...e.org>, Rob Herring <robh+dt@...nel.org>,
        Thomas Petazzoni <thomas.petazzoni@...tlin.com>,
        Frank Rowand <frowand.list@...il.com>
Subject: Re: [PATCH 1/2] of: device: Allow DMA range map to be set before
 of_dma_configure_id

On 2021-01-15 17:58, Paul Kocialkowski wrote:
> A mechanism was recently introduced for the sunxi architecture where
> the DMA offset for specific devices (under the MBUS) is set by a common
> driver (sunxi_mbus). This driver calls dma_direct_set_offset to set
> the device's dma_range_map manually.
> 
> However this information was overwritten by of_dma_configure_id, which
> obtains the map from of_dma_get_range (or keeps it NULL when it fails
> and the force_dma argument is true, which is the case for platform
> devices).
> 
> As a result, the dma_range_map was always overwritten and the mechanism
> could not correctly take effect.
> 
> This adds a check to ensure that no previous DMA range map is
> overwritten and prints a warning when the map was already set while
> also being available from dt. In this case, the map that was already
> set is kept.

Hang on, the hard-coded offset is only intended to be installed when 
there *isn't* anything described in DT, in which case of_dma_get_range() 
should always bail out early without touching it anyway. This sounds 
like something's not quite right in the MBUS driver, so I don't think 
working around it in core code is really the right thing to do.

Do you have a case where one of the relevant devices inherits a 
"dma-ranges" via the regular hierarchy without indirecting via an 
"interconnects" reference? Currently you're only checking for the 
latter, so that would be one way things could go awry (although to be a 
problem, said "dma-ranges" would also have to encode something *other* 
than the appropriate MBUS offset, which implies an incorrect or at least 
inaccurately-structured DT as well).

Robin.

> Fixes: b4bdc4fbf8d0 ("soc: sunxi: Deal with the MBUS DMA offsets in a central place")
> Signed-off-by: Paul Kocialkowski <paul.kocialkowski@...tlin.com>
> ---
>   drivers/of/device.c | 9 ++++++++-
>   1 file changed, 8 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/of/device.c b/drivers/of/device.c
> index aedfaaafd3e7..db1b8634c2c7 100644
> --- a/drivers/of/device.c
> +++ b/drivers/of/device.c
> @@ -181,7 +181,14 @@ int of_dma_configure_id(struct device *dev, struct device_node *np,
>   
>   	arch_setup_dma_ops(dev, dma_start, size, iommu, coherent);
>   
> -	dev->dma_range_map = map;
> +	if (!dev->dma_range_map) {
> +		dev->dma_range_map = map;
> +	} else if (map) {
> +		dev_warn(dev,
> +			 "DMA range map was already set, ignoring range map from dt\n");
> +		kfree(map);
> +	}
> +
>   	return 0;
>   }
>   EXPORT_SYMBOL_GPL(of_dma_configure_id);
>