ARM: v3.13-rc1: USB regression

[Aaro Koskinen]

Hi,

On Mon, Nov 25, 2013 at 05:18:53PM +0000, Russell King - ARM Linux wrote:
> On Mon, Nov 25, 2013 at 06:33:02PM +0200, Aaro Koskinen wrote:
> > Hi,
> > 
> > On Sun, Nov 24, 2013 at 10:43:59PM +0000, Russell King - ARM Linux wrote:
> > > On Mon, Nov 25, 2013 at 12:22:47AM +0200, Aaro Koskinen wrote:
> > > > [   33.967324] ohci ohci: Coherent DMA mask 0xffffffff (pfn 0xe0000-0xe0000) covers a smaller range of system memory than the DMA zone pfn 0x0-0x100000
> > > > 
> > > > I bisected this to 4dcfa60071b3d23f0181f27d8519f12e37cefbb9 (ARM: DMA-API:
> > > > better handing of DMA masks for coherent allocations). Reverting that
> > > > commit makes the USB work again fine.
> > 
> > [...]
> > 
> > > Better would be:
> > > 
> > > #define __arch_dma_to_pfn(dev, addr)	\
> > > 	({ unsigned long pfn = (addr) >> PAGE_SHIFT;		\
> > > 	   if (is_lbus_device(dev))				\
> > > 		pfn += PHYS_PFN_OFFSET - 			\
> > > 			(OMAP1510_LB_OFFSET >> PAGE_SHIFT);	\
> > > 	   pfn;							\
> > > 	})
> > > 
> > > Can you try that in arch/arm/mach-omap1/include/mach/memory.h please?
> > 
> > Still doesn't work:
> > 
> > [   33.878790] ohci ohci: Coherent DMA mask 0xffffffff (pfn 0xfffe0000-0xe0000) covers a smaller range of system memory than the DMA zone pfn 0x0-0x100000
> > [   33.894019] ohci ohci: can't setup: -12
> 
> Well, that looks technically better, rather unfortunate that we end up
> going to negative PFNs though.
> 
> Without that change, could you try this instead please:

This works, thanks.

A.

>  arch/arm/mm/dma-mapping.c |    9 +++++++--
>  1 files changed, 7 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
> index f0ea0134e5a3..a18cfc53f445 100644
> --- a/arch/arm/mm/dma-mapping.c
> +++ b/arch/arm/mm/dma-mapping.c
> @@ -9,6 +9,7 @@
>   *
>   *  DMA uncached mapping support.
>   */
> +#include <linux/bootmem.h>
>  #include <linux/module.h>
>  #include <linux/mm.h>
>  #include <linux/gfp.h>
> @@ -162,6 +163,8 @@ static u64 get_coherent_dma_mask(struct device *dev)
>  	u64 mask = (u64)DMA_BIT_MASK(32);
>  
>  	if (dev) {
> +		unsigned long max_dma_pfn;
> +
>  		mask = dev->coherent_dma_mask;
>  
>  		/*
> @@ -173,6 +176,8 @@ static u64 get_coherent_dma_mask(struct device *dev)
>  			return 0;
>  		}
>  
> +		max_dma_pfn = min(max_pfn, arm_dma_pfn_limit);
> +
>  		/*
>  		 * If the mask allows for more memory than we can address,
>  		 * and we actually have that much memory, then fail the
> @@ -180,7 +185,7 @@ static u64 get_coherent_dma_mask(struct device *dev)
>  		 */
>  		if (sizeof(mask) != sizeof(dma_addr_t) &&
>  		    mask > (dma_addr_t)~0 &&
> -		    dma_to_pfn(dev, ~0) > arm_dma_pfn_limit) {
> +		    dma_to_pfn(dev, ~0) > max_dma_pfn) {
>  			dev_warn(dev, "Coherent DMA mask %#llx is larger than dma_addr_t allows\n",
>  				 mask);
>  			dev_warn(dev, "Driver did not use or check the return value from dma_set_coherent_mask()?\n");
> @@ -192,7 +197,7 @@ static u64 get_coherent_dma_mask(struct device *dev)
>  		 * fits within the allowable addresses which we can
>  		 * allocate.
>  		 */
> -		if (dma_to_pfn(dev, mask) < arm_dma_pfn_limit) {
> +		if (dma_to_pfn(dev, mask) < max_dma_pfn) {
>  			dev_warn(dev, "Coherent DMA mask %#llx (pfn %#lx-%#lx) covers a smaller range of system memory than the DMA zone pfn 0x0-%#lx\n",
>  				 mask,
>  				 dma_to_pfn(dev, 0), dma_to_pfn(dev, mask) + 1,
>