similar to: [PATCH v2] virtio_ring: use alloc_pages_node for NUMA-aware allocation

Hi Shile, Thank you for the patch! Yet something to improve: [auto build test ERROR on linus/master] [also build test ERROR on v5.8-rc5 next-20200716] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in https://git-scm.com/docs/git-format-patch] url:

Hi Shile, Thank you for the patch! Yet something to improve: [auto build test ERROR on linus/master] [also build test ERROR on v5.8-rc5 next-20200716] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in https://git-scm.com/docs/git-format-patch] url:

This API is the equivalent of alloc_pages, except that the returned memory is guaranteed to be DMA addressable by the passed in device. The implementation will also be used to provide a more sensible replacement for DMA_ATTR_NON_CONSISTENT flag. Additionally dma_alloc_noncoherent is switched over to use dma_alloc_pages as its backend. Signed-off-by: Christoph Hellwig <hch at lst.de> ---

Add a new API to allocate and free pages that are guaranteed to be addressable by a device, but otherwise behave like pages allocated by alloc_pages. The intended APIs to sync them for use with the device and cpu are dma_sync_single_for_{device,cpu} that are also used for streaming mappings. Switch all drivers over to this new API, but keep the usage of the crufty dma_cache_sync API for now,

We should only assign intel_dma_ops to devices which will actually use the iommu and let the default fall back dma_direct_* functions handle all other devices. This won't change any behaviour but will just use the generic implementations for direct mapped devices rather than intel specific ones. Signed-off-by: Tom Murphy <murphyt7 at tcd.ie> --- drivers/iommu/intel-iommu.c | 52

The jazzdma ops implement support for a very basic IOMMU. Thus we really should not use the dma-direct code that takes physical address limits into account. This survived through the great MIPS DMA ops cleanup mostly because I was lazy, but now it is time to fully split the implementations. Signed-off-by: Christoph Hellwig <hch at lst.de> --- arch/mips/jazz/jazzdma.c | 32

Eliminate kernel version checks from i386/kernel/pci-dma.c as a trivial first pass cleanup towards merging. Signed-off-by: Chris Wright <chrisw@osdl.org> --- linux-2.6.11-xen-sparse/arch/xen/i386/kernel/pci-dma.c | 29 ----------------- 1 files changed, 29 deletions(-) ===== arch/xen/i386/kernel/pci-dma.c 1.20 vs edited ===== ---

After switching DMA attributes to unsigned long it is easier to just compare the bits. Signed-off-by: Krzysztof Kozlowski <k.kozlowski at samsung.com> --- Documentation/DMA-API.txt | 4 +-- arch/arc/mm/dma.c | 4 +-- arch/arm/mm/dma-mapping.c | 36 ++++++++++++-------------- arch/arm/xen/mm.c

Split out subsystem specific changes for easier reviews. This will be squashed with main commit. Signed-off-by: Krzysztof Kozlowski <k.kozlowski at samsung.com> --- drivers/xen/swiotlb-xen.c | 14 +++++++------- include/xen/swiotlb-xen.h | 12 ++++++------ 2 files changed, 13 insertions(+), 13 deletions(-) diff --git a/drivers/xen/swiotlb-xen.c b/drivers/xen/swiotlb-xen.c index

Split out subsystem specific changes for easier reviews. This will be squashed with main commit. Signed-off-by: Krzysztof Kozlowski <k.kozlowski at samsung.com> --- arch/arc/mm/dma.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/arch/arc/mm/dma.c b/arch/arc/mm/dma.c index 73d7e4c75b7d..3d1f467d1792 100644 --- a/arch/arc/mm/dma.c +++ b/arch/arc/mm/dma.c @@

Split out subsystem specific changes for easier reviews. This will be squashed with main commit. Signed-off-by: Krzysztof Kozlowski <k.kozlowski at samsung.com> --- arch/m68k/kernel/dma.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/arch/m68k/kernel/dma.c b/arch/m68k/kernel/dma.c index cbc78b4117b5..8cf97cbadc91 100644 --- a/arch/m68k/kernel/dma.c +++

Split out subsystem specific changes for easier reviews. This will be squashed with main commit. Signed-off-by: Krzysztof Kozlowski <k.kozlowski at samsung.com> --- arch/cris/arch-v32/drivers/pci/dma.c | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-) diff --git a/arch/cris/arch-v32/drivers/pci/dma.c b/arch/cris/arch-v32/drivers/pci/dma.c index 8d5efa58cce1..1f0636793f0c

We are going to require dma_ops for several common drivers, even for systems that do have an identity mapping. Lets provide some minimal no-op dma_ops that can be used for that purpose. Signed-off-by: Christian Borntraeger <borntraeger at de.ibm.com> --- include/linux/dma-mapping.h | 2 ++ lib/Makefile | 1 + lib/dma-noop.c | 75

We are going to require dma_ops for several common drivers, even for systems that do have an identity mapping. Lets provide some minimal no-op dma_ops that can be used for that purpose. Signed-off-by: Christian Borntraeger <borntraeger at de.ibm.com> --- include/linux/dma-mapping.h | 2 ++ lib/Makefile | 2 +- lib/dma-noop.c | 77

We are going to require dma_ops for several common drivers, even for systems that do have an identity mapping. Lets provide some minimal no-op dma_ops that can be used for that purpose. Signed-off-by: Christian Borntraeger <borntraeger at de.ibm.com> Reviewed-by: Joerg Roedel <jroedel at suse.de> --- include/linux/dma-mapping.h | 2 ++ lib/Makefile | 1 +

> +/* > + * Virtio direct mapping DMA API operations structure > + * > + * This defines DMA API structure for all virtio devices which would not > + * either bring in their own DMA OPS from architecture or they would not > + * like to use architecture specific IOMMU based DMA OPS because QEMU > + * expects GPA instead of an IOVA in absence of VIRTIO_F_IOMMU_PLATFORM. > + */

Current implementation of DMA API inside virtio core calls device's DMA OPS callback functions when the flag VIRTIO_F_IOMMU_PLATFORM flag is set. But in absence of the flag, virtio core falls back calling basic transformation of the incoming SG addresses as GPA. Going forward virtio should only call DMA API based transformations generating either GPA or IOVA depending on QEMU expectations

The dma-mapping core and the implementations do not change the DMA attributes passed by pointer. Thus the pointer can point to const data. However the attributes do not have to be a bitfield. Instead unsigned long will do fine: 1. This is just simpler. Both in terms of reading the code and setting attributes. Instead of initializing local attributes on the stack and passing pointer to

Split out subsystem specific changes for easier reviews. This will be squashed with main commit. Signed-off-by: Krzysztof Kozlowski <k.kozlowski at samsung.com> --- arch/h8300/kernel/dma.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/arch/h8300/kernel/dma.c b/arch/h8300/kernel/dma.c index eeb13d3f2424..3651da045806 100644 --- a/arch/h8300/kernel/dma.c +++

These patches are not ready to be merged because I was unable to measure a performance improvement. I'm publishing them so they are archived in case someone picks up this work again in the future. The goal of these patches is to allocate virtqueues and driver state from the device's NUMA node for optimal memory access latency. Only guests with a vNUMA topology and virtio devices spread