Re: [PATCH net 1/3] net: mvpp2: fix the dma_mask and coherent_dma_mask settings for PPv2.2

[Antoine Tenart <antoine.tenart@free-electrons.com>]

Hi David,

On Mon, Sep 18, 2017 at 05:18:58PM -0700, David Miller wrote:
> From: Antoine Tenart <antoine.tenart@free-electrons.com>
> Date: Mon, 18 Sep 2017 15:04:06 +0200
> 
> > The dev->dma_mask usually points to dev->coherent_dma_mask. This is an
> > issue as setting both of them will override the other. This is
> > problematic here as the PPv2 driver uses a 32-bit-mask for coherent
> > accesses (txq, rxq, bm) and a 40-bit mask for all other accesses due to
> > an hardware limitation.
> > 
> > This can lead to a memory remap for all dma_map_single() calls when
> > dealing with memory above 4GB.
> > 
> > Fixes: 2067e0a13cfe ("net: mvpp2: set dma mask and coherent dma mask on PPv2.2")
> > Reported-by: Stefan Chulski <stefanc@marvell.com>
> > Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
> 
> I surrmise that if the platform has made dev->dma_mask point to
> &dev->coherent_dma_mask, it is because it does not allow the two
> settings to be set separately.

That's also the default when the platform does not allocate dma_mask.
You have a point, that could be because it's not supported. But I don't
know what would be a good check then.

> By rearranging the pointer, you are bypassing that, and probably
> breaking things or creating a situation that the DMA mapping
> layer is not expecting.
> 
> I want to know more about the situations where dma_mask is set to
> point to &coherent_dma_mask and how that is supposed to work.

>From what I see in other parts of the kernel, dma_mask points to
&coherent_dma_mask by default and having two different values for these
two masks isn't a use case for many drivers.

Antoine

-- 
Antoine Ténart, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com