On Wed, Jun 16, 2021 at 11:54 AM Claire Chang <tientzu@xxxxxxxxxxxx> wrote:
>
> Add the functions, swiotlb_{alloc,free} to support the memory allocation
> from restricted DMA pool.
>
> The restricted DMA pool is preferred if available.
>
> Note that since coherent allocation needs remapping, one must set up
> another device coherent pool by shared-dma-pool and use
> dma_alloc_from_dev_coherent instead for atomic coherent allocation.
>
> Signed-off-by: Claire Chang <tientzu@xxxxxxxxxxxx>
> Reviewed-by: Christoph Hellwig <hch@xxxxxx>
> ---
> include/linux/swiotlb.h | 15 +++++++++++++
> kernel/dma/direct.c | 50 ++++++++++++++++++++++++++++++-----------
> kernel/dma/swiotlb.c | 45 +++++++++++++++++++++++++++++++++++--
> 3 files changed, 95 insertions(+), 15 deletions(-)
>
> diff --git a/include/linux/swiotlb.h b/include/linux/swiotlb.h
> index efcd56e3a16c..2d5ec670e064 100644
> --- a/include/linux/swiotlb.h
> +++ b/include/linux/swiotlb.h
> @@ -156,4 +156,19 @@ static inline void swiotlb_adjust_size(unsigned long size)
> extern void swiotlb_print_info(void);
> extern void swiotlb_set_max_segment(unsigned int);
>
> +#ifdef CONFIG_DMA_RESTRICTED_POOL
> +struct page *swiotlb_alloc(struct device *dev, size_t size);
> +bool swiotlb_free(struct device *dev, struct page *page, size_t size);
> +#else
> +static inline struct page *swiotlb_alloc(struct device *dev, size_t size)
> +{
> + return NULL;
> +}
> +static inline bool swiotlb_free(struct device *dev, struct page *page,
> + size_t size)
> +{
> + return false;
> +}
> +#endif /* CONFIG_DMA_RESTRICTED_POOL */
> +
> #endif /* __LINUX_SWIOTLB_H */
> diff --git a/kernel/dma/direct.c b/kernel/dma/direct.c
> index 3713461d6fe0..da0e09621230 100644
> --- a/kernel/dma/direct.c
> +++ b/kernel/dma/direct.c
> @@ -75,6 +75,15 @@ static bool dma_coherent_ok(struct device *dev, phys_addr_t phys, size_t size)
> min_not_zero(dev->coherent_dma_mask, dev->bus_dma_limit);
> }
>
> +static void __dma_direct_free_pages(struct device *dev, struct page *page,
> + size_t size)
> +{
> + if (IS_ENABLED(CONFIG_DMA_RESTRICTED_POOL) &&
> + swiotlb_free(dev, page, size))
> + return;
> + dma_free_contiguous(dev, page, size);
> +}
> +
> static struct page *__dma_direct_alloc_pages(struct device *dev, size_t size,
> gfp_t gfp)
> {
> @@ -86,7 +95,16 @@ static struct page *__dma_direct_alloc_pages(struct device *dev, size_t size,
>
> gfp |= dma_direct_optimal_gfp_mask(dev, dev->coherent_dma_mask,
> &phys_limit);
> - page = dma_alloc_contiguous(dev, size, gfp);
> + if (IS_ENABLED(CONFIG_DMA_RESTRICTED_POOL)) {
> + page = swiotlb_alloc(dev, size);
> + if (page && !dma_coherent_ok(dev, page_to_phys(page), size)) {
> + __dma_direct_free_pages(dev, page, size);
> + return NULL;
> + }
> + }
> +
> + if (!page)
> + page = dma_alloc_contiguous(dev, size, gfp);
> if (page && !dma_coherent_ok(dev, page_to_phys(page), size)) {
> dma_free_contiguous(dev, page, size);
> page = NULL;
> @@ -142,7 +160,7 @@ void *dma_direct_alloc(struct device *dev, size_t size,
> gfp |= __GFP_NOWARN;
>
> if ((attrs & DMA_ATTR_NO_KERNEL_MAPPING) &&
> - !force_dma_unencrypted(dev)) {
> + !force_dma_unencrypted(dev) && !is_dev_swiotlb_force(dev)) {
> page = __dma_direct_alloc_pages(dev, size, gfp & ~__GFP_ZERO);
> if (!page)
> return NULL;
> @@ -155,18 +173,23 @@ void *dma_direct_alloc(struct device *dev, size_t size,
> }
>
> if (!IS_ENABLED(CONFIG_ARCH_HAS_DMA_SET_UNCACHED) &&
> - !IS_ENABLED(CONFIG_DMA_DIRECT_REMAP) &&
> - !dev_is_dma_coherent(dev))
> + !IS_ENABLED(CONFIG_DMA_DIRECT_REMAP) && !dev_is_dma_coherent(dev) &&
> + !is_dev_swiotlb_force(dev))
> return arch_dma_alloc(dev, size, dma_handle, gfp, attrs);
Just noticed that after propagating swiotlb_force setting into
io_tlb_default_mem->force, the memory allocation behavior for
swiotlb_force will change (i.e. always skipping arch_dma_alloc and
dma_direct_alloc_from_pool).
>
> /*
> * Remapping or decrypting memory may block. If either is required and
> * we can't block, allocate the memory from the atomic pools.
> + * If restricted DMA (i.e., is_dev_swiotlb_force) is required, one must
> + * set up another device coherent pool by shared-dma-pool and use
> + * dma_alloc_from_dev_coherent instead.
> */
> if (IS_ENABLED(CONFIG_DMA_COHERENT_POOL) &&
> !gfpflags_allow_blocking(gfp) &&
> (force_dma_unencrypted(dev) ||
> - (IS_ENABLED(CONFIG_DMA_DIRECT_REMAP) && !dev_is_dma_coherent(dev))))
> + (IS_ENABLED(CONFIG_DMA_DIRECT_REMAP) &&
> + !dev_is_dma_coherent(dev))) &&
> + !is_dev_swiotlb_force(dev))
> return dma_direct_alloc_from_pool(dev, size, dma_handle, gfp);
And here.
>
> /* we always manually zero the memory once we are done */
> @@ -237,7 +260,7 @@ void *dma_direct_alloc(struct device *dev, size_t size,
> return NULL;
> }
> out_free_pages:
> - dma_free_contiguous(dev, page, size);
> + __dma_direct_free_pages(dev, page, size);
> return NULL;
> }
>
> @@ -247,15 +270,15 @@ void dma_direct_free(struct device *dev, size_t size,
> unsigned int page_order = get_order(size);
>
> if ((attrs & DMA_ATTR_NO_KERNEL_MAPPING) &&
> - !force_dma_unencrypted(dev)) {
> + !force_dma_unencrypted(dev) && !is_dev_swiotlb_force(dev)) {
> /* cpu_addr is a struct page cookie, not a kernel address */
> dma_free_contiguous(dev, cpu_addr, size);
> return;
> }
>
> if (!IS_ENABLED(CONFIG_ARCH_HAS_DMA_SET_UNCACHED) &&
> - !IS_ENABLED(CONFIG_DMA_DIRECT_REMAP) &&
> - !dev_is_dma_coherent(dev)) {
> + !IS_ENABLED(CONFIG_DMA_DIRECT_REMAP) && !dev_is_dma_coherent(dev) &&
> + !is_dev_swiotlb_force(dev)) {
> arch_dma_free(dev, size, cpu_addr, dma_addr, attrs);
> return;
> }
> @@ -273,7 +296,7 @@ void dma_direct_free(struct device *dev, size_t size,
> else if (IS_ENABLED(CONFIG_ARCH_HAS_DMA_CLEAR_UNCACHED))
> arch_dma_clear_uncached(cpu_addr, size);
>
> - dma_free_contiguous(dev, dma_direct_to_page(dev, dma_addr), size);
> + __dma_direct_free_pages(dev, dma_direct_to_page(dev, dma_addr), size);
> }
>
> struct page *dma_direct_alloc_pages(struct device *dev, size_t size,
> @@ -283,7 +306,8 @@ struct page *dma_direct_alloc_pages(struct device *dev, size_t size,
> void *ret;
>
> if (IS_ENABLED(CONFIG_DMA_COHERENT_POOL) &&
> - force_dma_unencrypted(dev) && !gfpflags_allow_blocking(gfp))
> + force_dma_unencrypted(dev) && !gfpflags_allow_blocking(gfp) &&
> + !is_dev_swiotlb_force(dev))
> return dma_direct_alloc_from_pool(dev, size, dma_handle, gfp);
>
> page = __dma_direct_alloc_pages(dev, size, gfp);
> @@ -310,7 +334,7 @@ struct page *dma_direct_alloc_pages(struct device *dev, size_t size,
> *dma_handle = phys_to_dma_direct(dev, page_to_phys(page));
> return page;
> out_free_pages:
> - dma_free_contiguous(dev, page, size);
> + __dma_direct_free_pages(dev, page, size);
> return NULL;
> }
>
> @@ -329,7 +353,7 @@ void dma_direct_free_pages(struct device *dev, size_t size,
> if (force_dma_unencrypted(dev))
> set_memory_encrypted((unsigned long)vaddr, 1 << page_order);
>
> - dma_free_contiguous(dev, page, size);
> + __dma_direct_free_pages(dev, page, size);
> }
>
> #if defined(CONFIG_ARCH_HAS_SYNC_DMA_FOR_DEVICE) || \
> diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
> index fec4934b9926..6ad85b48f101 100644
> --- a/kernel/dma/swiotlb.c
> +++ b/kernel/dma/swiotlb.c
> @@ -462,8 +462,9 @@ static int swiotlb_find_slots(struct device *dev, phys_addr_t orig_addr,
>
> index = wrap = wrap_index(mem, ALIGN(mem->index, stride));
> do {
> - if ((slot_addr(tbl_dma_addr, index) & iotlb_align_mask) !=
> - (orig_addr & iotlb_align_mask)) {
> + if (orig_addr &&
> + (slot_addr(tbl_dma_addr, index) & iotlb_align_mask) !=
> + (orig_addr & iotlb_align_mask)) {
> index = wrap_index(mem, index + 1);
> continue;
> }
> @@ -702,3 +703,43 @@ static int __init swiotlb_create_default_debugfs(void)
> late_initcall(swiotlb_create_default_debugfs);
>
> #endif
> +
> +#ifdef CONFIG_DMA_RESTRICTED_POOL
> +struct page *swiotlb_alloc(struct device *dev, size_t size)
> +{
> + struct io_tlb_mem *mem = dev->dma_io_tlb_mem;
> + phys_addr_t tlb_addr;
> + int index;
> +
> + /*
> + * Skip io_tlb_default_mem since swiotlb_alloc doesn't support atomic
> + * coherent allocation. Otherwise might break existing devices.
> + * One must set up another device coherent pool by shared-dma-pool and
> + * use dma_alloc_from_dev_coherent instead for atomic coherent
> + * allocation to avoid memory remapping.
> + */
> + if (!mem || mem == io_tlb_default_mem)
> + return NULL;
> +
> + index = swiotlb_find_slots(dev, 0, size);
> + if (index == -1)
> + return NULL;
> +
> + tlb_addr = slot_addr(mem->start, index);
> +
> + return pfn_to_page(PFN_DOWN(tlb_addr));
> +}
> +
> +bool swiotlb_free(struct device *dev, struct page *page, size_t size)
> +{
> + phys_addr_t tlb_addr = page_to_phys(page);
> +
> + if (!is_swiotlb_buffer(dev, tlb_addr))
> + return false;
> +
> + swiotlb_release_slots(dev, tlb_addr);
> +
> + return true;
> +}
> +
> +#endif /* CONFIG_DMA_RESTRICTED_POOL */
> --
> 2.32.0.272.g935e593368-goog
>
