Re: [PATCH 4.4 35/46] ravb: unmap descriptors when freeing rings

[Ben Hutchings]

On Thu, 2017-06-15 at 19:52 +0200, Greg Kroah-Hartman wrote:
> 4.4-stable review patch.  If anyone has any objections, please let me know.
> 
> ------------------
> 
> From: Kazuya Mizuguchi <kazuya.mizuguchi.ks@xxxxxxxxxxx>
> 
> 
> [ Upstream commit a47b70ea86bdeb3091341f5ae3ef580f1a1ad822 ]
> 
> "swiotlb buffer is full" errors occur after repeated initialisation of a
> device - f.e. suspend/resume or ip link set up/down. This is because memory
> mapped using dma_map_single() in ravb_ring_format() and ravb_start_xmit()
> is not released.  Resolve this problem by unmapping descriptors when
> freeing rings.

This should be followed by:

commit 79514ef670e9e575a1fe36922268c439d0f0ca8a
Author: Eugeniu Rosca <erosca@xxxxxxxxxxxxxx>
Date:   Tue Jun 6 00:08:10 2017 +0200

    ravb: Fix use-after-free on `ifconfig eth0 down`

But also, this loop looks wrong:

[...]
>  	if (priv->rx_ring[q]) {
> +		for (i = 0; i < priv->num_rx_ring[q]; i++) {
> +			struct ravb_ex_rx_desc *desc = &priv->rx_ring[q][i];
> +
> +			if (!dma_mapping_error(ndev->dev.parent,
> +					       le32_to_cpu(desc->dptr)))
> +				dma_unmap_single(ndev->dev.parent,
> +						 le32_to_cpu(desc->dptr),
> +						 PKT_BUF_SZ,
> +						 DMA_FROM_DEVICE);
> +		}
[...]

It's possible that the driver hasn't filled (or attempted to fill or
refill) every RX descriptor, so this could result in a double-unmap.  I
think this needs to use cur_rx and dirty_rx to determine which
descriptors to process.

Ben.

-- 
Ben Hutchings
Software Developer, Codethink Ltd.