Re: [RFC 1/1] swiotlb: Reduce calls to swiotlb_find_pool()

[Petr Tesařík <petr@xxxxxxxxxxx>]

On Thu,  6 Jun 2024 20:14:21 -0700
mhkelley58@xxxxxxxxx wrote:

> From: Michael Kelley <mhklinux@xxxxxxxxxxx>
> 
> With CONFIG_SWIOTLB_DYNAMIC enabled, each round-trip map/unmap pair
> in the swiotlb results in 6 calls to swiotlb_find_pool(). In multiple
> places, the pool is found and used in one function, and then must
> found again in the next function that is called because only the
> tlb_addr is passed as an argument. These are the six call sites:
> 
> dma_direct_map_page:
> 1. swiotlb_map->swiotlb_tbl_map_single->swiotlb_bounce
> 
> dma_direct_unmap_page:
> 2. dma_direct_sync_single_for_cpu->is_swiotlb_buffer
> 3. dma_direct_sync_single_for_cpu->swiotlb_sync_single_for_cpu->
>       swiotlb_bounce
> 4. is_swiotlb_buffer
> 5. swiotlb_tbl_unmap_single->swiotlb_del_transient
> 6. swiotlb_tbl_unmap_single->swiotlb_release_slots
> 
> Reduce the number of calls by finding the pool at a higher level, and
> passing it as an argument instead of searching again. A key change is
> for is_swiotlb_buffer() to return a pool pointer instead of a boolean,
> and then pass this pool pointer to subsequent swiotlb functions.
> With these changes, a round-trip map/unmap pair requires only 2 calls
> to swiotlb_find_pool():
> 
> dma_direct_unmap_page:
> 1. dma_direct_sync_single_for_cpu->is_swiotlb_buffer
> 2. is_swiotlb_buffer
> 
> These changes come from noticing the inefficiencies in a code review,
> not from performance measurements. With CONFIG_SWIOTLB_DYNAMIC,
> swiotlb_find_pool() is not trivial, and it uses an RCU read lock,
> so avoiding the redundant calls helps performance in a hot path.
> When CONFIG_SWIOTLB_DYNAMIC is *not* set, the code size reduction
> is minimal and the perf benefits are likely negligible, but no
> harm is done.
> 
> No functional change is intended.
> 
> Signed-off-by: Michael Kelley <mhklinux@xxxxxxxxxxx>
> ---
> This patch trades off making many of the core swiotlb APIs take
> an additional argument in order to avoid duplicating calls to
> swiotlb_find_pool(). The current code seems rather wasteful in
> making 6 calls per round-trip, but I'm happy to accept others'
> judgment as to whether getting rid of the waste is worth the
> additional code complexity.
> 
>  drivers/iommu/dma-iommu.c | 27 ++++++++++++++------
>  drivers/xen/swiotlb-xen.c | 25 +++++++++++-------
>  include/linux/swiotlb.h   | 54 +++++++++++++++++++++------------------
>  kernel/dma/direct.c       | 12 ++++++---
>  kernel/dma/direct.h       | 18 ++++++++-----
>  kernel/dma/swiotlb.c      | 43 ++++++++++++++++---------------
>  6 files changed, 106 insertions(+), 73 deletions(-)
> 
> diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
> index f731e4b2a417..ab6bc37ecf90 100644
> --- a/drivers/iommu/dma-iommu.c
> +++ b/drivers/iommu/dma-iommu.c
> @@ -1073,6 +1073,7 @@ static void iommu_dma_sync_single_for_cpu(struct device 
> *dev,
>               dma_addr_t dma_handle, size_t size, enum dma_data_direction dir)
>  {
>       phys_addr_t phys;
> +     struct io_tlb_pool *pool;
>  
>       if (dev_is_dma_coherent(dev) && !dev_use_swiotlb(dev, size, dir))
>               return;
> @@ -1081,21 +1082,25 @@ static void iommu_dma_sync_single_for_cpu(struct 
> device *dev,
>       if (!dev_is_dma_coherent(dev))
>               arch_sync_dma_for_cpu(phys, size, dir);
>  
> -     if (is_swiotlb_buffer(dev, phys))
> -             swiotlb_sync_single_for_cpu(dev, phys, size, dir);
> +     pool = is_swiotlb_buffer(dev, phys);
> +     if (pool)
> +             swiotlb_sync_single_for_cpu(dev, phys, size, dir, pool);
>  }
>  
>  static void iommu_dma_sync_single_for_device(struct device *dev,
>               dma_addr_t dma_handle, size_t size, enum dma_data_direction dir)
>  {
>       phys_addr_t phys;
> +     struct io_tlb_pool *pool;
>  
>       if (dev_is_dma_coherent(dev) && !dev_use_swiotlb(dev, size, dir))
>               return;
>  
>       phys = iommu_iova_to_phys(iommu_get_dma_domain(dev), dma_handle);
> -     if (is_swiotlb_buffer(dev, phys))
> -             swiotlb_sync_single_for_device(dev, phys, size, dir);
> +
> +     pool = is_swiotlb_buffer(dev, phys);
> +     if (pool)
> +             swiotlb_sync_single_for_device(dev, phys, size, dir, pool);
>  
>       if (!dev_is_dma_coherent(dev))
>               arch_sync_dma_for_device(phys, size, dir);
> @@ -1189,8 +1194,12 @@ static dma_addr_t iommu_dma_map_page(struct device 
> *dev, struct page *page,
>               arch_sync_dma_for_device(phys, size, dir);
>  
>       iova = __iommu_dma_map(dev, phys, size, prot, dma_mask);
> -     if (iova == DMA_MAPPING_ERROR && is_swiotlb_buffer(dev, phys))
> -             swiotlb_tbl_unmap_single(dev, phys, size, dir, attrs);
> +     if (iova == DMA_MAPPING_ERROR) {
> +             struct io_tlb_pool *pool = is_swiotlb_buffer(dev, phys);
> +
> +             if (pool)
> +                     swiotlb_tbl_unmap_single(dev, phys, size, dir, attrs, 
> pool);
> +     }
>       return iova;
>  }
>  
> @@ -1199,6 +1208,7 @@ static void iommu_dma_unmap_page(struct device *dev, 
> dma_addr_t dma_handle,
>  {
>       struct iommu_domain *domain = iommu_get_dma_domain(dev);
>       phys_addr_t phys;
> +     struct io_tlb_pool *pool;
>  
>       phys = iommu_iova_to_phys(domain, dma_handle);
>       if (WARN_ON(!phys))
> @@ -1209,8 +1219,9 @@ static void iommu_dma_unmap_page(struct device *dev, 
> dma_addr_t dma_handle,
>  
>       __iommu_dma_unmap(dev, dma_handle, size);
>  
> -     if (unlikely(is_swiotlb_buffer(dev, phys)))
> -             swiotlb_tbl_unmap_single(dev, phys, size, dir, attrs);
> +     pool = is_swiotlb_buffer(dev, phys);
> +     if (unlikely(pool))
> +             swiotlb_tbl_unmap_single(dev, phys, size, dir, attrs, pool);
>  }
>  
>  /*
> diff --git a/drivers/xen/swiotlb-xen.c b/drivers/xen/swiotlb-xen.c
> index 6579ae3f6dac..7af8c8466e1d 100644
> --- a/drivers/xen/swiotlb-xen.c
> +++ b/drivers/xen/swiotlb-xen.c
> @@ -88,7 +88,7 @@ static inline int range_straddles_page_boundary(phys_addr_t 
> p, size_t size)
>       return 0;
>  }
>  
> -static int is_xen_swiotlb_buffer(struct device *dev, dma_addr_t dma_addr)
> +static struct io_tlb_pool *is_xen_swiotlb_buffer(struct device *dev, 
> dma_addr_t dma_addr)
>  {
>       unsigned long bfn = XEN_PFN_DOWN(dma_to_phys(dev, dma_addr));
>       unsigned long xen_pfn = bfn_to_local_pfn(bfn);
> @@ -100,7 +100,7 @@ static int is_xen_swiotlb_buffer(struct device *dev, 
> dma_addr_t dma_addr)
>        */
>       if (pfn_valid(PFN_DOWN(paddr)))
>               return is_swiotlb_buffer(dev, paddr);
> -     return 0;
> +     return NULL;
>  }
>  
>  #ifdef CONFIG_X86
> @@ -228,7 +228,8 @@ static dma_addr_t xen_swiotlb_map_page(struct device 
> *dev, struct page *page,
>        */
>       if (unlikely(!dma_capable(dev, dev_addr, size, true))) {
>               swiotlb_tbl_unmap_single(dev, map, size, dir,
> -                             attrs | DMA_ATTR_SKIP_CPU_SYNC);
> +                             attrs | DMA_ATTR_SKIP_CPU_SYNC,
> +                             swiotlb_find_pool(dev, map));
>               return DMA_MAPPING_ERROR;
>       }
>  
> @@ -254,6 +255,7 @@ static void xen_swiotlb_unmap_page(struct device *hwdev, 
> dma_addr_t dev_addr,
>               size_t size, enum dma_data_direction dir, unsigned long attrs)
>  {
>       phys_addr_t paddr = xen_dma_to_phys(hwdev, dev_addr);
> +     struct io_tlb_pool *pool;
>  
>       BUG_ON(dir == DMA_NONE);
>  
> @@ -265,8 +267,9 @@ static void xen_swiotlb_unmap_page(struct device *hwdev, 
> dma_addr_t dev_addr,
>       }
>  
>       /* NOTE: We use dev_addr here, not paddr! */
> -     if (is_xen_swiotlb_buffer(hwdev, dev_addr))
> -             swiotlb_tbl_unmap_single(hwdev, paddr, size, dir, attrs);
> +     pool = is_xen_swiotlb_buffer(hwdev, dev_addr);
> +     if (pool)
> +             swiotlb_tbl_unmap_single(hwdev, paddr, size, dir, attrs, pool);
>  }
>  
>  static void
> @@ -274,6 +277,7 @@ xen_swiotlb_sync_single_for_cpu(struct device *dev, 
> dma_addr_t dma_addr,
>               size_t size, enum dma_data_direction dir)
>  {
>       phys_addr_t paddr = xen_dma_to_phys(dev, dma_addr);
> +     struct io_tlb_pool *pool;
>  
>       if (!dev_is_dma_coherent(dev)) {
>               if (pfn_valid(PFN_DOWN(dma_to_phys(dev, dma_addr))))
> @@ -282,8 +286,9 @@ xen_swiotlb_sync_single_for_cpu(struct device *dev, 
> dma_addr_t dma_addr,
>                       xen_dma_sync_for_cpu(dev, dma_addr, size, dir);
>       }
>  
> -     if (is_xen_swiotlb_buffer(dev, dma_addr))
> -             swiotlb_sync_single_for_cpu(dev, paddr, size, dir);
> +     pool = is_xen_swiotlb_buffer(dev, dma_addr);
> +     if (pool)
> +             swiotlb_sync_single_for_cpu(dev, paddr, size, dir, pool);
>  }
>  
>  static void
> @@ -291,9 +296,11 @@ xen_swiotlb_sync_single_for_device(struct device *dev, 
> dma_addr_t dma_addr,
>               size_t size, enum dma_data_direction dir)
>  {
>       phys_addr_t paddr = xen_dma_to_phys(dev, dma_addr);
> +     struct io_tlb_pool *pool;
>  
> -     if (is_xen_swiotlb_buffer(dev, dma_addr))
> -             swiotlb_sync_single_for_device(dev, paddr, size, dir);
> +     pool = is_xen_swiotlb_buffer(dev, dma_addr);
> +     if (pool)
> +             swiotlb_sync_single_for_device(dev, paddr, size, dir, pool);
>  
>       if (!dev_is_dma_coherent(dev)) {
>               if (pfn_valid(PFN_DOWN(dma_to_phys(dev, dma_addr))))
> diff --git a/include/linux/swiotlb.h b/include/linux/swiotlb.h
> index 14bc10c1bb23..ce8651949123 100644
> --- a/include/linux/swiotlb.h
> +++ b/include/linux/swiotlb.h
> @@ -42,24 +42,6 @@ int swiotlb_init_late(size_t size, gfp_t gfp_mask,
>       int (*remap)(void *tlb, unsigned long nslabs));
>  extern void __init swiotlb_update_mem_attributes(void);
>  
> -phys_addr_t swiotlb_tbl_map_single(struct device *hwdev, phys_addr_t phys,
> -             size_t mapping_size,
> -             unsigned int alloc_aligned_mask, enum dma_data_direction dir,
> -             unsigned long attrs);
> -
> -extern void swiotlb_tbl_unmap_single(struct device *hwdev,
> -                                  phys_addr_t tlb_addr,
> -                                  size_t mapping_size,
> -                                  enum dma_data_direction dir,
> -                                  unsigned long attrs);
> -
> -void swiotlb_sync_single_for_device(struct device *dev, phys_addr_t tlb_addr,
> -             size_t size, enum dma_data_direction dir);
> -void swiotlb_sync_single_for_cpu(struct device *dev, phys_addr_t tlb_addr,
> -             size_t size, enum dma_data_direction dir);
> -dma_addr_t swiotlb_map(struct device *dev, phys_addr_t phys,
> -             size_t size, enum dma_data_direction dir, unsigned long attrs);
> -
>  #ifdef CONFIG_SWIOTLB
>  
>  /**
> @@ -168,12 +150,12 @@ static inline struct io_tlb_pool 
> *swiotlb_find_pool(struct device *dev,
>   * * %true if @paddr points into a bounce buffer
>   * * %false otherwise
>   */
> -static inline bool is_swiotlb_buffer(struct device *dev, phys_addr_t paddr)
> +static inline struct io_tlb_pool *is_swiotlb_buffer(struct device *dev, 
> phys_addr_t paddr)
>  {
>       struct io_tlb_mem *mem = dev->dma_io_tlb_mem;
>  
>       if (!mem)
> -             return false;
> +             return NULL;
>  
>  #ifdef CONFIG_SWIOTLB_DYNAMIC
>       /*
> @@ -187,10 +169,13 @@ static inline bool is_swiotlb_buffer(struct device 
> *dev, phys_addr_t paddr)
>        * This barrier pairs with smp_mb() in swiotlb_find_slots().
>        */
>       smp_rmb();
> -     return READ_ONCE(dev->dma_uses_io_tlb) &&
> -             swiotlb_find_pool(dev, paddr);
> +     if (READ_ONCE(dev->dma_uses_io_tlb))
> +             return swiotlb_find_pool(dev, paddr);
> +     return NULL;
>  #else
> -     return paddr >= mem->defpool.start && paddr < mem->defpool.end;
> +     if (paddr >= mem->defpool.start && paddr < mem->defpool.end)
> +             return &mem->defpool;

Why are we open-coding swiotlb_find_pool() here? It does not make a
difference now, but if swiotlb_find_pool() were to change, both places
would have to be updated.

Does it save a reload from dev->dma_io_tlb_mem? IOW is the compiler
unable to optimize it away?

What about this (functionally identical) variant:

#ifdef CONFIG_SWIOTLB_DYNAMIC
        smp_rmb();
        if (!READ_ONCE(dev->dma_uses_io_tlb))
                return NULL;
#else
        if (paddr < mem->defpool.start || paddr >= mem->defpool.end);
                return NULL;
#endif

        return swiotlb_find_pool(dev, paddr);

Petr T