[PATCH 0/2] x86: fixes for the allow_coherent rewrite (tip/x86/iommu)

[PATCH 0/2] x86: fixes for the allow_coherent rewrite (tip/x86/iommu)

[FUJITA Tomonori]

This is the final patchset for the allow_coherent rewrite (hopefully).

[PATCH 1/2] x86: avoid unnecessary low zone allocation in Calgary's alloc_coherent

[FUJITA Tomonori]

x86's common alloc_coherent (dma_alloc_coherent in dma-mapping.h) sets
up the gfp flag according to the device dma_mask but Calgary doesn't
need it because of virtual mappings. This patch avoids unnecessary low
zone allocation.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Alexis Bruemmer <alexisb@us.ibm.com>
Cc: Muli Ben-Yehuda <muli@il.ibm.com>
---
 arch/x86/kernel/pci-calgary_64.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/arch/x86/kernel/pci-calgary_64.c b/arch/x86/kernel/pci-calgary_64.c
index 8415d92..fe7695e 100644
--- a/arch/x86/kernel/pci-calgary_64.c
+++ b/arch/x86/kernel/pci-calgary_64.c
@@ -491,6 +491,8 @@ static void* calgary_alloc_coherent(struct device *dev, size_t size,
 	npages = size >> PAGE_SHIFT;
 	order = get_order(size);
 
+	flag &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
+
 	/* alloc enough pages (and possibly more) */
 	ret = (void *)__get_free_pages(flag, order);
 	if (!ret)
-- 
1.5.4.2

[PATCH 2/2] x86: gart alloc_coherent does virtual mapppings only when necessary

[FUJITA Tomonori]

gart alloc_coherent need to do virtual mapppings only when an
allocated buffer is not DMA-capable for a device.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kernel/pci-gart_64.c |   10 +++++++++-
 1 files changed, 9 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c
index 1b0c412..47abe43 100644
--- a/arch/x86/kernel/pci-gart_64.c
+++ b/arch/x86/kernel/pci-gart_64.c
@@ -488,15 +488,23 @@ gart_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_addr,
 		    gfp_t flag)
 {
 	void *vaddr;
+	dma_addr_t paddr;
 	unsigned long align_mask;
+	u64 dma_mask = dma_alloc_coherent_mask(dev, flag);
 
 	vaddr = (void *)__get_free_pages(flag | __GFP_ZERO, get_order(size));
 	if (!vaddr)
 		return NULL;
 
+	paddr = virt_to_phys(vaddr);
+	if (is_buffer_dma_capable(dma_mask, paddr, size)) {
+		*dma_addr = paddr;
+		return vaddr;
+	}
+
 	align_mask = (1UL << get_order(size)) - 1;
 
-	*dma_addr = dma_map_area(dev, __pa(vaddr), size, DMA_BIDIRECTIONAL,
+	*dma_addr = dma_map_area(dev, paddr, size, DMA_BIDIRECTIONAL,
 				 align_mask);
 	flush_gart();
 
-- 
1.5.4.2

Re: [PATCH 2/2] x86: gart alloc_coherent does virtual mapppings only when necessary

[Joerg Roedel]

Looks good to me.

On Thu, Sep 11, 2008 at 11:08:48PM +0900, FUJITA Tomonori wrote:
> gart alloc_coherent need to do virtual mapppings only when an
> allocated buffer is not DMA-capable for a device.
> 
> Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
> Cc: Joerg Roedel <joerg.roedel@amd.com>

Acked-by: Joerg Roedel <joerg.roedel@amd.com>

> ---
>  arch/x86/kernel/pci-gart_64.c |   10 +++++++++-
>  1 files changed, 9 insertions(+), 1 deletions(-)
> 
> diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c
> index 1b0c412..47abe43 100644
> --- a/arch/x86/kernel/pci-gart_64.c
> +++ b/arch/x86/kernel/pci-gart_64.c
> @@ -488,15 +488,23 @@ gart_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_addr,
>  		    gfp_t flag)
>  {
>  	void *vaddr;
> +	dma_addr_t paddr;
>  	unsigned long align_mask;
> +	u64 dma_mask = dma_alloc_coherent_mask(dev, flag);
>  
>  	vaddr = (void *)__get_free_pages(flag | __GFP_ZERO, get_order(size));
>  	if (!vaddr)
>  		return NULL;
>  
> +	paddr = virt_to_phys(vaddr);
> +	if (is_buffer_dma_capable(dma_mask, paddr, size)) {
> +		*dma_addr = paddr;
> +		return vaddr;
> +	}
> +
>  	align_mask = (1UL << get_order(size)) - 1;
>  
> -	*dma_addr = dma_map_area(dev, __pa(vaddr), size, DMA_BIDIRECTIONAL,
> +	*dma_addr = dma_map_area(dev, paddr, size, DMA_BIDIRECTIONAL,
>  				 align_mask);
>  	flush_gart();
>  
> -- 
> 1.5.4.2
> 
> 

-- 
           |           AMD Saxony Limited Liability Company & Co. KG
 Operating |         Wilschdorfer Landstr. 101, 01109 Dresden, Germany
 System    |                  Register Court Dresden: HRA 4896
 Research  |              General Partner authorized to represent:
 Center    |             AMD Saxony LLC (Wilmington, Delaware, US)
           | General Manager of AMD Saxony LLC: Dr. Hans-R. Deppe, Thomas McCoy

Re: [PATCH 1/2] x86: avoid unnecessary low zone allocation in Calgary's alloc_coherent

[Muli Ben-Yehuda]

On Thu, Sep 11, 2008 at 11:08:47PM +0900, FUJITA Tomonori wrote:
> x86's common alloc_coherent (dma_alloc_coherent in dma-mapping.h) sets
> up the gfp flag according to the device dma_mask but Calgary doesn't
> need it because of virtual mappings. This patch avoids unnecessary low
> zone allocation.
> 
> Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
> Cc: Alexis Bruemmer <alexisb@us.ibm.com>
> Cc: Muli Ben-Yehuda <muli@il.ibm.com>

Looks good,

Acked-by: Muli Ben-Yehuda <muli@il.ibm.com>

Cheers,
Muli
-- 
Workshop on I/O Virtualization (WIOV '08)
Co-located with OSDI '08, Dec 2008, San Diego, CA
http://www.usenix.org/wiov08

Re: [PATCH 2/2] x86: gart alloc_coherent does virtual mapppings only when necessary

[Ingo Molnar]

* Joerg Roedel <joerg.roedel@amd.com> wrote:

> Looks good to me.
> 
> On Thu, Sep 11, 2008 at 11:08:48PM +0900, FUJITA Tomonori wrote:
> > gart alloc_coherent need to do virtual mapppings only when an
> > allocated buffer is not DMA-capable for a device.
> > 
> > Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
> > Cc: Joerg Roedel <joerg.roedel@amd.com>
> 
> Acked-by: Joerg Roedel <joerg.roedel@amd.com>

applied to tip/x86/iommu, thanks.

	Ingo

Re: [PATCH 1/2] x86: avoid unnecessary low zone allocation in Calgary's alloc_coherent

[Ingo Molnar]

* Muli Ben-Yehuda <muli@il.ibm.com> wrote:

> Looks good,
> 
> Acked-by: Muli Ben-Yehuda <muli@il.ibm.com>

applied to tip/x86/iommu, thanks.

	Ingo