[PATCH v7 2/6] iommu/dma: Add support for non-strict mode

On 2018-09-18 6:10 PM, Will Deacon wrote:

Hi Robin,

On Fri, Sep 14, 2018 at 03:30:20PM +0100, Robin Murphy wrote:

quoted

From: Zhen Lei <redacted>

1. Save the related domain pointer in struct iommu_dma_cookie, make iovad
    capable call domain->ops->flush_iotlb_all to flush TLB.
2. During the iommu domain initialization phase, base on domain->non_strict
    field to check whether non-strict mode is supported or not. If so, call
    init_iova_flush_queue to register iovad->flush_cb callback.
3. All unmap(contains iova-free) APIs will finally invoke __iommu_dma_unmap
    -->iommu_dma_free_iova. If the domain is non-strict, call queue_iova to
    put off iova freeing, and omit iommu_tlb_sync operation.

Hmm, this is basically just a commentary on the code. Please could you write
it more in terms of the problem that's being solved?

Sure - I intentionally kept a light touch when it came to the 
documentation and commit messages in this rework (other than patch #1 
where I eventually remembered the original reasoning and that it wasn't 
a bug). If we're more-or-less happy with the shape of the technical side 
I'll make sure to take a final pass through v8 to tidy up all the prose.

quoted

Signed-off-by: Zhen Lei <redacted>
[rm: convert raw boolean to domain attribute]
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
---
  drivers/iommu/dma-iommu.c | 29 ++++++++++++++++++++++++++++-
  include/linux/iommu.h     |  1 +
  2 files changed, 29 insertions(+), 1 deletion(-)

diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
index 511ff9a1d6d9..092e6926dc3c 100644
--- a/drivers/iommu/dma-iommu.c
+++ b/drivers/iommu/dma-iommu.c

@@ -55,6 +55,9 @@ struct iommu_dma_cookie {
  	};
  	struct list_head		msi_page_list;
  	spinlock_t			msi_lock;
+
+	/* Only be assigned in non-strict mode, otherwise it's NULL */
+	struct iommu_domain		*domain;
  };
  
  static inline size_t cookie_msi_granule(struct iommu_dma_cookie *cookie)

@@ -257,6 +260,17 @@ static int iova_reserve_iommu_regions(struct device *dev,
  	return ret;
  }
  
+static void iommu_dma_flush_iotlb_all(struct iova_domain *iovad)
+{
+	struct iommu_dma_cookie *cookie;
+	struct iommu_domain *domain;
+
+	cookie = container_of(iovad, struct iommu_dma_cookie, iovad);
+	domain = cookie->domain;
+
+	domain->ops->flush_iotlb_all(domain);

Can we rely on this function pointer being non-NULL? I think it would
be better to call iommu_flush_tlb_all(cookie->domain) instead.

Yeah, that's deliberate - in fact got as far as writing that change, 
then undid it as I realised that although the attribute conversion got 
rid of the explicit ops->flush_iotlb_all check, it still makes zero 
sense for an IOMMU driver to claim to support the flush queue attribute 
without also providing the relevant callback, so I do actually want this 
to blow up rather than silently do nothing if that assumption isn't met.

quoted

+}
+
  /**
   * iommu_dma_init_domain - Initialise a DMA mapping domain
   * @domain: IOMMU domain previously prepared by iommu_get_dma_cookie()

@@ -275,6 +289,7 @@ int iommu_dma_init_domain(struct iommu_domain *domain, dma_addr_t base,
  	struct iommu_dma_cookie *cookie = domain->iova_cookie;
  	struct iova_domain *iovad = &cookie->iovad;
  	unsigned long order, base_pfn, end_pfn;
+	int attr = 1;

Do we actually need to initialise this?

Oops, no, that's a left-over from the turned-out-messier-that-I-thought 
v6 implementation.

Thanks,
Robin.