Re: [PATCH 1/2] dma-mapping: add a dma_ops_bypass flag to struct device

Robin Murphy Mon, 23 Mar 2020 05:16:08 -0700

On 2020-03-20 2:16 pm, Christoph Hellwig wrote:

Several IOMMU drivers have a bypass mode where they can use a direct
mapping if the devices DMA mask is large enough.  Add generic support
to the core dma-mapping code to do that to switch those drivers to
a common solution.

Hmm, this is _almost_, but not quite the same as the case where driversare managing their own IOMMU mappings, but still need to use streamingDMA for cache maintenance on the underlying pages. For that we need theops bypass to be a "true" bypass and also avoid SWIOTLB regardless ofthe device's DMA mask. That behaviour should in fact be fine for theopportunistic bypass case here as well, since the mask being "bigenough" implies by definition that this should never need to bounce either.

For the (hopefully less common) third case where, due to groups or useroverrides, we end up giving an identity DMA domain to a device withlimited DMA masks which _does_ need SWIOTLB, I'd like to think we cansolve that by not giving the device IOMMU DMA ops in the first place,such that it never needs to engage the bypass mechanism at all.


Thoughts?

Robin.

Signed-off-by: Christoph Hellwig <h...@lst.de>
---
  include/linux/device.h      |  6 ++++++
  include/linux/dma-mapping.h | 30 ++++++++++++++++++------------
  kernel/dma/mapping.c        | 36 +++++++++++++++++++++++++++---------
  3 files changed, 51 insertions(+), 21 deletions(-)

diff --git a/include/linux/device.h b/include/linux/device.h
index 0cd7c647c16c..09be8bb2c4a6 100644
--- a/include/linux/device.h
+++ b/include/linux/device.h
@@ -525,6 +525,11 @@ struct dev_links_info {
   *              sync_state() callback.
   * @dma_coherent: this particular device is dma coherent, even if the
   *            architecture supports non-coherent devices.
+ * @dma_ops_bypass: If set to %true then the dma_ops are bypassed for the
+ *             streaming DMA operations (->map_* / ->unmap_* / ->sync_*),
+ *             and optionall (if the coherent mask is large enough) also
+ *             for dma allocations.  This flag is managed by the dma ops
+ *             instance from ->dma_supported.
   *
   * At the lowest level, every device in a Linux system is represented by an
   * instance of struct device. The device structure contains the information
@@ -625,6 +630,7 @@ struct device {
      defined(CONFIG_ARCH_HAS_SYNC_DMA_FOR_CPU_ALL)
        bool                    dma_coherent:1;
  #endif
+       bool                    dma_ops_bypass : 1;
  };

static inline struct device *kobj_to_dev(struct kobject *kobj)

diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
index 330ad58fbf4d..c3af0cf5e435 100644
--- a/include/linux/dma-mapping.h
+++ b/include/linux/dma-mapping.h
@@ -188,9 +188,15 @@ static inline int dma_mmap_from_global_coherent(struct 
vm_area_struct *vma,
  }
  #endif /* CONFIG_DMA_DECLARE_COHERENT */

-static inline bool dma_is_direct(const struct dma_map_ops *ops)

+/*
+ * Check if the devices uses a direct mapping for streaming DMA operations.
+ * This allows IOMMU drivers to set a bypass mode if the DMA mask is large
+ * enough.
+ */
+static inline bool dma_map_direct(struct device *dev,
+               const struct dma_map_ops *ops)
  {
-       return likely(!ops);
+       return likely(!ops) || dev->dma_ops_bypass;
  }

/*

@@ -279,7 +285,7 @@ static inline dma_addr_t dma_map_page_attrs(struct device 
*dev,
        dma_addr_t addr;