Re: Bug 205201 - overflow of DMA mask and bus mask

[Robin Murphy]

On 05/11/2019 16:28, Christoph Hellwig wrote:
> On Tue, Nov 05, 2019 at 08:56:27AM +0100, Christian Zigotzky wrote:
> > Hi All,
> >
> > We still have DMA problems with some PCI devices. Since the PowerPC updates
> > 4.21-1 [1] we need to decrease the RAM to 3500MB (mem=3500M) if we want to
> > work with our PCI devices. The FSL P5020 and P5040 have these problems
> > currently.
> >
> > Error message:
> >
> > [ÂÂ 25.654852] bttv 1000:04:05.0: overflow 0x00000000fe077000+4096 of DMA
> > mask ffffffff bus mask df000000

Hmm, that bus mask looks pretty wacky - are you able to figure out where 
that's coming from?

Robin.

> > All 5.x Linux kernels can't initialize a SCSI PCI card anymore so booting
> > of a Linux userland isn't possible.
> >
> > PLEASE check the DMA changes in the PowerPC updates 4.21-1 [1]. The kernel
> > 4.20 works with all PCI devices without limitation of RAM.
>
> Can you send me the .config and a dmesg?  And in the meantime try the
> patch below?
>
> ---
>  From 4d659b7311bd4141fdd3eeeb80fa2d7602ea01d4 Mon Sep 17 00:00:00 2001
> From: Nicolas Saenz Julienne <nsaenzjulienne@xxxxxxx>
> Date: Fri, 18 Oct 2019 13:00:43 +0200
> Subject: dma-direct: check for overflows on 32 bit DMA addresses
>
> As seen on the new Raspberry Pi 4 and sta2x11's DMA implementation it is
> possible for a device configured with 32 bit DMA addresses and a partial
> DMA mapping located at the end of the address space to overflow. It
> happens when a higher physical address, not DMAable, is translated to
> it's DMA counterpart.
>
> For example the Raspberry Pi 4, configurable up to 4 GB of memory, has
> an interconnect capable of addressing the lower 1 GB of physical memory
> with a DMA offset of 0xc0000000. It transpires that, any attempt to
> translate physical addresses higher than the first GB will result in an
> overflow which dma_capable() can't detect as it only checks for
> addresses bigger then the maximum allowed DMA address.
>
> Fix this by verifying in dma_capable() if the DMA address range provided
> is at any point lower than the minimum possible DMA address on the bus.
>
> Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@xxxxxxx>
> ---
>   include/linux/dma-direct.h | 8 ++++++++
>   1 file changed, 8 insertions(+)
>
> diff --git a/include/linux/dma-direct.h b/include/linux/dma-direct.h
> index adf993a3bd58..6ad9e9ea7564 100644
> --- a/include/linux/dma-direct.h
> +++ b/include/linux/dma-direct.h
> @@ -3,6 +3,7 @@
>   #define _LINUX_DMA_DIRECT_H 1
>   
>   #include <linux/dma-mapping.h>
> +#include <linux/memblock.h> /* for min_low_pfn */
>   #include <linux/mem_encrypt.h>
>   
>   #ifdef CONFIG_ARCH_HAS_PHYS_TO_DMA
> @@ -27,6 +28,13 @@ static inline bool dma_capable(struct device *dev, dma_addr_t addr, size_t size)
>   	if (!dev->dma_mask)
>   		return false;
>   
> +#ifndef CONFIG_ARCH_DMA_ADDR_T_64BIT
> +	/* Check if DMA address overflowed */
> +	if (min(addr, addr + size - 1) <
> +		__phys_to_dma(dev, (phys_addr_t)(min_low_pfn << PAGE_SHIFT)))
> +		return false;
> +#endif
> +
>   	return addr + size - 1 <=
>   		min_not_zero(*dev->dma_mask, dev->bus_dma_mask);
>   }