by Konrad Rzeszutek Wilk

[permalink] [raw]

Subject: Re: [PATCH v9 10/13] xen/arm/arm64: introduce xen_arch_need_swiotlb

On Fri, Nov 21, 2014 at 04:31:31PM +0000, Stefano Stabellini wrote:
> On Wed, 12 Nov 2014, Stefano Stabellini wrote:
> > Introduce an arch specific function to find out whether a particular dma
> > mapping operation needs to bounce on the swiotlb buffer.
> >
> > On ARM and ARM64, if the page involved is a foreign page and the device
> > is not coherent, we need to bounce because at unmap time we cannot
> > execute any required cache maintenance operations (we don't know how to
> > find the pfn from the mfn).
> >
> > No change of behaviour for x86.
> >
> > Signed-off-by: Stefano Stabellini <[email protected]>
> > Reviewed-by: David Vrabel <[email protected]>
> > Reviewed-by: Catalin Marinas <[email protected]>
> > Acked-by: Ian Campbell <[email protected]>
> > Acked-by: Konrad Rzeszutek Wilk <[email protected]>
>
> I am thinking of asking a backport of this patch to 3.16+
>
> The catch is that is_device_dma_coherent is not available on older
> kernels, so I'll have to change the arm implementation of
> xen_arch_need_swiotlb to:
>
> +bool xen_arch_need_swiotlb(struct device *dev,
> + unsigned long pfn,
> + unsigned long mfn)
> +{
> + return pfn != mfn;
> +}
> +
>
> It is going to make things slower but it is going to fix the issue with
> cache flushing buffers for non-coherent devices.
>
> Konrad, are you OK with that?

Looks fine.
>
>
> > Changes in v6:
> > - fix ts.
> >
> > Changes in v5:
> > - fix indentation.
> > ---
> > arch/arm/include/asm/xen/page.h | 4 ++++
> > arch/arm/xen/mm.c | 7 +++++++
> > arch/x86/include/asm/xen/page.h | 7 +++++++
> > drivers/xen/swiotlb-xen.c | 5 ++++-
> > 4 files changed, 22 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/arm/include/asm/xen/page.h b/arch/arm/include/asm/xen/page.h
> > index 135c24a..68c739b 100644
> > --- a/arch/arm/include/asm/xen/page.h
> > +++ b/arch/arm/include/asm/xen/page.h
> > @@ -107,4 +107,8 @@ static inline bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
> > #define xen_remap(cookie, size) ioremap_cache((cookie), (size))
> > #define xen_unmap(cookie) iounmap((cookie))
> >
> > +bool xen_arch_need_swiotlb(struct device *dev,
> > + unsigned long pfn,
> > + unsigned long mfn);
> > +
> > #endif /* _ASM_ARM_XEN_PAGE_H */
> > diff --git a/arch/arm/xen/mm.c b/arch/arm/xen/mm.c
> > index ab700e1..28ebf3e 100644
> > --- a/arch/arm/xen/mm.c
> > +++ b/arch/arm/xen/mm.c
> > @@ -100,6 +100,13 @@ void __xen_dma_sync_single_for_device(struct device *hwdev,
> > __xen_dma_page_cpu_to_dev(hwdev, handle, size, dir);
> > }
> >
> > +bool xen_arch_need_swiotlb(struct device *dev,
> > + unsigned long pfn,
> > + unsigned long mfn)
> > +{
> > + return ((pfn != mfn) && !is_device_dma_coherent(dev));
> > +}
> > +
> > int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
> > unsigned int address_bits,
> > dma_addr_t *dma_handle)
> > diff --git a/arch/x86/include/asm/xen/page.h b/arch/x86/include/asm/xen/page.h
> > index c949923..f58ef6c 100644
> > --- a/arch/x86/include/asm/xen/page.h
> > +++ b/arch/x86/include/asm/xen/page.h
> > @@ -236,4 +236,11 @@ void make_lowmem_page_readwrite(void *vaddr);
> > #define xen_remap(cookie, size) ioremap((cookie), (size));
> > #define xen_unmap(cookie) iounmap((cookie))
> >
> > +static inline bool xen_arch_need_swiotlb(struct device *dev,
> > + unsigned long pfn,
> > + unsigned long mfn)
> > +{
> > + return false;
> > +}
> > +
> > #endif /* _ASM_X86_XEN_PAGE_H */
> > diff --git a/drivers/xen/swiotlb-xen.c b/drivers/xen/swiotlb-xen.c
> > index ad2c5eb..3725ee4 100644
> > --- a/drivers/xen/swiotlb-xen.c
> > +++ b/drivers/xen/swiotlb-xen.c
> > @@ -399,7 +399,9 @@ dma_addr_t xen_swiotlb_map_page(struct device *dev, struct page *page,
> > * buffering it.
> > */
> > if (dma_capable(dev, dev_addr, size) &&
> > - !range_straddles_page_boundary(phys, size) && !swiotlb_force) {
> > + !range_straddles_page_boundary(phys, size) &&
> > + !xen_arch_need_swiotlb(dev, PFN_DOWN(phys), PFN_DOWN(dev_addr)) &&
> > + !swiotlb_force) {
> > /* we are not interested in the dma_addr returned by
> > * xen_dma_map_page, only in the potential cache flushes executed
> > * by the function. */
> > @@ -557,6 +559,7 @@ xen_swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl,
> > dma_addr_t dev_addr = xen_phys_to_bus(paddr);
> >
> > if (swiotlb_force ||
> > + xen_arch_need_swiotlb(hwdev, PFN_DOWN(paddr), PFN_DOWN(dev_addr)) ||
> > !dma_capable(hwdev, dev_addr, sg->length) ||
> > range_straddles_page_boundary(paddr, sg->length)) {
> > phys_addr_t map = swiotlb_tbl_map_single(hwdev,
> > --
> > 1.7.10.4
> >