[RFC,v2,09/21] dma-mapping: provide callbacks to link/unlink HMM PFNs to specific IOVA

Commit Message

Leon Romanovsky Sept. 12, 2024, 11:15 a.m. UTC

From: Leon Romanovsky <[e-mail]>

Introduce new DMA link/unlink API to provide a way for HMM users to link
pages to already preallocated IOVA.

Signed-off-by: Leon Romanovsky <[e-mail]>
---
 include/linux/dma-mapping.h |  15 ++++++
 kernel/dma/mapping.c        | 102 ++++++++++++++++++++++++++++++++++++
 2 files changed, 117 insertions(+)

Patch

diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
index bb541f8944e5..8c2a468c5420 100644
--- a/include/linux/dma-mapping.h
+++ b/include/linux/dma-mapping.h
@@ -123,6 +123,10 @@  static inline int dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 int dma_alloc_iova_unaligned(struct dma_iova_state *state, phys_addr_t phys,
 			     size_t size);
 void dma_free_iova(struct dma_iova_state *state);
+dma_addr_t dma_hmm_link_page(struct dma_iova_state *state, unsigned long *pfn,
+			     dma_addr_t dma_offset);
+void dma_hmm_unlink_page(struct dma_iova_state *state, unsigned long *pfn,
+			 dma_addr_t dma_offset);
 
 dma_addr_t dma_map_page_attrs(struct device *dev, struct page *page,
 		size_t offset, size_t size, enum dma_data_direction dir,
@@ -189,6 +193,17 @@  static inline int dma_alloc_iova_unaligned(struct dma_iova_state *state,
 static inline void dma_free_iova(struct dma_iova_state *state)
 {
 }
+static inline dma_addr_t dma_hmm_link_page(struct dma_iova_state *state,
+					   unsigned long *pfn,
+					   dma_addr_t dma_offset)
+{
+	return DMA_MAPPING_ERROR;
+}
+static inline void dma_hmm_unlink_page(struct dma_iova_state *state,
+				       unsigned long *pfn,
+				       dma_addr_t dma_offset)
+{
+}
 static inline dma_addr_t dma_map_page_attrs(struct device *dev,
 		struct page *page, size_t offset, size_t size,
 		enum dma_data_direction dir, unsigned long attrs)
diff --git a/kernel/dma/mapping.c b/kernel/dma/mapping.c
index 39fac8c21643..5354ddc3ac03 100644
--- a/kernel/dma/mapping.c
+++ b/kernel/dma/mapping.c
@@ -17,6 +17,7 @@ 
 #include <linux/slab.h>
 #include <linux/vmalloc.h>
 #include <linux/cc_platform.h>
+#include <linux/hmm.h>
 #include "debug.h"
 #include "direct.h"
 
@@ -1084,3 +1085,104 @@  void dma_unlink_range_attrs(struct dma_iova_state *state, unsigned long attrs)
 			       state->dir, attrs);
 }
 EXPORT_SYMBOL_GPL(dma_unlink_range_attrs);
+
+/**
+ * dma_hmm_link_page - Link a physical HMM page to DMA address
+ * @state: IOVA state
+ * @pfn: HMM PFN
+ * @dma_offset: DMA offset form which this page needs to be linked
+ *
+ * dma_alloc_iova() allocates IOVA based on the size specified by their use in
+ * iova->size. Call this function after IOVA allocation to link whole @page
+ * to get the DMA address. Note that very first call to this function
+ * will have @dma_offset set to 0 in the IOVA space allocated from
+ * dma_alloc_iova(). For subsequent calls to this function on same @iova,
+ * @dma_offset needs to be advanced by the caller with the size of previous
+ * page that was linked + DMA address returned for the previous page that was
+ * linked by this function.
+ */
+dma_addr_t dma_hmm_link_page(struct dma_iova_state *state, unsigned long *pfn,
+			     dma_addr_t dma_offset)
+{
+	struct device *dev = state->dev;
+	struct page *page = hmm_pfn_to_page(*pfn);
+	phys_addr_t phys = page_to_phys(page);
+	bool coherent = dev_is_dma_coherent(dev);
+	dma_addr_t addr;
+	int ret;
+
+	if (*pfn & HMM_PFN_DMA_MAPPED)
+		/*
+		 * We are in this flow when there is a need to resync flags,
+		 * for example when page was already linked in prefetch call
+		 * with READ flag and now we need to add WRITE flag
+		 *
+		 * This page was already programmed to HW and we don't want/need
+		 * to unlink and link it again just to resync flags.
+		 *
+		 * The DMA address calculation below is based on the fact that
+		 * HMM doesn't work with swiotlb.
+		 */
+		return (state->addr) ? state->addr + dma_offset :
+				       phys_to_dma(dev, phys);
+
+	state->range_size = dma_offset;
+
+	/*
+	 * The below check is based on assumption that HMM range users
+	 * don't work with swiotlb and hence can be or in direct mode
+	 * or in IOMMU mode.
+	 */
+	if (!use_dma_iommu(dev)) {
+		if (!coherent)
+			arch_sync_dma_for_device(phys, PAGE_SIZE, state->dir);
+
+		addr = phys_to_dma(dev, phys);
+		goto done;
+	}
+
+	ret = dma_start_range(state);
+	if (ret)
+		return DMA_MAPPING_ERROR;
+
+	addr = dma_link_range(state, phys, PAGE_SIZE);
+	dma_end_range(state);
+	if (dma_mapping_error(state->dev, addr))
+		return addr;
+
+done:
+	kmsan_handle_dma(page, 0, PAGE_SIZE, state->dir);
+	*pfn |= HMM_PFN_DMA_MAPPED;
+	return addr;
+}
+EXPORT_SYMBOL_GPL(dma_hmm_link_page);
+
+/**
+ * dma_hmm_unlink_page - Unlink a physical HMM page from DMA address
+ * @state: IOVA state
+ * @pfn: HMM PFN
+ * @dma_offset: DMA offset form which this page needs to be unlinked
+ *              from the IOVA space
+ */
+void dma_hmm_unlink_page(struct dma_iova_state *state, unsigned long *pfn,
+			 dma_addr_t dma_offset)
+{
+	struct device *dev = state->dev;
+	struct page *page;
+	phys_addr_t phys;
+
+	*pfn &= ~HMM_PFN_DMA_MAPPED;
+
+	if (!use_dma_iommu(dev)) {
+		page = hmm_pfn_to_page(*pfn);
+		phys = page_to_phys(page);
+
+		dma_direct_sync_single_for_cpu(dev, phys_to_dma(dev, phys),
+					       PAGE_SIZE, state->dir);
+		return;
+	}
+
+	iommu_dma_unlink_range(dev, state->addr + dma_offset, PAGE_SIZE,
+			       state->dir, 0);
+}
+EXPORT_SYMBOL_GPL(dma_hmm_unlink_page);