[PATCH v3 0/2] block: Enable proper MMIO memory handling for P2P DMA

[PATCH v3 0/2] block: Enable proper MMIO memory handling for P2P DMA

[Leon Romanovsky]

Changelog:
v3:
 * Encoded p2p map type in IOD flags instead of DMA attributes.
 * Removed REQ_P2PDMA flag from block layer.
 * Simplified map_phys conversion patch.
v2: https://lore.kernel.org/all/20251020-block-with-mmio-v2-0-147e9f93d8d4@nvidia.com/
 * Added Chirstoph's Reviewed-by tag for first patch.
 * Squashed patches
 * Stored DMA MMIO attribute in NVMe IOD flags variable instead of block layer.
v1: https://patch.msgid.link/20251017-block-with-mmio-v1-0-3f486904db5e@nvidia.com
 * Reordered patches.
 * Dropped patch which tried to unify unmap flow.
 * Set MMIO flag separately for data and integrity payloads.
v0: https://lore.kernel.org/all/cover.1760369219.git.leon@kernel.org/

----------------------------------------------------------------------

This patch series improves block layer and NVMe driver support for MMIO
memory regions, particularly for peer-to-peer (P2P) DMA transfers that
go through the host bridge.

The series addresses a critical gap where P2P transfers through the host
bridge (PCI_P2PDMA_MAP_THRU_HOST_BRIDGE) were not properly marked as
MMIO memory, leading to potential issues with:

- Inappropriate CPU cache synchronization operations on MMIO regions
- Incorrect DMA mapping/unmapping that doesn't respect MMIO semantics  
- Missing IOMMU configuration for MMIO memory handling

This work is extracted from the larger DMA physical API improvement
series [1] and focuses specifically on block layer and NVMe requirements
for MMIO memory support.

Thanks

[1] https://lore.kernel.org/all/cover.1757423202.git.leonro@nvidia.com/

Leon Romanovsky (2):
      nvme-pci: migrate to dma_map_phys instead of map_page
      block-dma: properly take MMIO path

 block/blk-mq-dma.c            | 19 ++++++----
 drivers/nvme/host/pci.c       | 86 +++++++++++++++++++++++++++++++++++--------
 include/linux/bio-integrity.h |  1 -
 include/linux/blk-integrity.h | 14 -------
 include/linux/blk-mq-dma.h    | 28 +++++++-------
 include/linux/blk_types.h     |  2 -
 6 files changed, 95 insertions(+), 55 deletions(-)
---
base-commit: 3a8660878839faadb4f1a6dd72c3179c1df56787
change-id: 20251016-block-with-mmio-02acf4285427

Best regards,
--  
Leon Romanovsky <leonro@nvidia.com>

[PATCH v3 1/2] nvme-pci: migrate to dma_map_phys instead of map_page

[Leon Romanovsky]

From: Leon Romanovsky <leonro@nvidia.com>

After introduction of dma_map_phys(), there is no need to convert
from physical address to struct page in order to map page. So let's
use it directly.

Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 block/blk-mq-dma.c      |  4 ++--
 drivers/nvme/host/pci.c | 25 +++++++++++++------------
 2 files changed, 15 insertions(+), 14 deletions(-)

diff --git a/block/blk-mq-dma.c b/block/blk-mq-dma.c
index 449950029872..4ba7b0323da4 100644
--- a/block/blk-mq-dma.c
+++ b/block/blk-mq-dma.c
@@ -93,8 +93,8 @@ static bool blk_dma_map_bus(struct blk_dma_iter *iter, struct phys_vec *vec)
 static bool blk_dma_map_direct(struct request *req, struct device *dma_dev,
 		struct blk_dma_iter *iter, struct phys_vec *vec)
 {
-	iter->addr = dma_map_page(dma_dev, phys_to_page(vec->paddr),
-			offset_in_page(vec->paddr), vec->len, rq_dma_dir(req));
+	iter->addr = dma_map_phys(dma_dev, vec->paddr, vec->len,
+			rq_dma_dir(req), 0);
 	if (dma_mapping_error(dma_dev, iter->addr)) {
 		iter->status = BLK_STS_RESOURCE;
 		return false;
diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index c916176bd9f0..002412431940 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -685,20 +685,20 @@ static void nvme_free_descriptors(struct request *req)
 	}
 }
 
-static void nvme_free_prps(struct request *req)
+static void nvme_free_prps(struct request *req, unsigned int attrs)
 {
 	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
 	struct nvme_queue *nvmeq = req->mq_hctx->driver_data;
 	unsigned int i;
 
 	for (i = 0; i < iod->nr_dma_vecs; i++)
-		dma_unmap_page(nvmeq->dev->dev, iod->dma_vecs[i].addr,
-				iod->dma_vecs[i].len, rq_dma_dir(req));
+		dma_unmap_phys(nvmeq->dev->dev, iod->dma_vecs[i].addr,
+			       iod->dma_vecs[i].len, rq_dma_dir(req), attrs);
 	mempool_free(iod->dma_vecs, nvmeq->dev->dmavec_mempool);
 }
 
 static void nvme_free_sgls(struct request *req, struct nvme_sgl_desc *sge,
-		struct nvme_sgl_desc *sg_list)
+		struct nvme_sgl_desc *sg_list, unsigned int attrs)
 {
 	struct nvme_queue *nvmeq = req->mq_hctx->driver_data;
 	enum dma_data_direction dir = rq_dma_dir(req);
@@ -707,13 +707,14 @@ static void nvme_free_sgls(struct request *req, struct nvme_sgl_desc *sge,
 	unsigned int i;
 
 	if (sge->type == (NVME_SGL_FMT_DATA_DESC << 4)) {
-		dma_unmap_page(dma_dev, le64_to_cpu(sge->addr), len, dir);
+		dma_unmap_phys(dma_dev, le64_to_cpu(sge->addr), len, dir,
+			       attrs);
 		return;
 	}
 
 	for (i = 0; i < len / sizeof(*sg_list); i++)
-		dma_unmap_page(dma_dev, le64_to_cpu(sg_list[i].addr),
-			le32_to_cpu(sg_list[i].length), dir);
+		dma_unmap_phys(dma_dev, le64_to_cpu(sg_list[i].addr),
+			le32_to_cpu(sg_list[i].length), dir, attrs);
 }
 
 static void nvme_unmap_metadata(struct request *req)
@@ -734,10 +735,10 @@ static void nvme_unmap_metadata(struct request *req)
 	if (!blk_rq_integrity_dma_unmap(req, dma_dev, &iod->meta_dma_state,
 					iod->meta_total_len)) {
 		if (nvme_pci_cmd_use_meta_sgl(&iod->cmd))
-			nvme_free_sgls(req, sge, &sge[1]);
+			nvme_free_sgls(req, sge, &sge[1], 0);
 		else
-			dma_unmap_page(dma_dev, iod->meta_dma,
-				       iod->meta_total_len, dir);
+			dma_unmap_phys(dma_dev, iod->meta_dma,
+				       iod->meta_total_len, dir, 0);
 	}
 
 	if (iod->meta_descriptor)
@@ -762,9 +763,9 @@ static void nvme_unmap_data(struct request *req)
 	if (!blk_rq_dma_unmap(req, dma_dev, &iod->dma_state, iod->total_len)) {
 		if (nvme_pci_cmd_use_sgl(&iod->cmd))
 			nvme_free_sgls(req, iod->descriptors[0],
-				       &iod->cmd.common.dptr.sgl);
+				       &iod->cmd.common.dptr.sgl, 0);
 		else
-			nvme_free_prps(req);
+			nvme_free_prps(req, 0);
 	}
 
 	if (iod->nr_descriptors)

-- 
2.51.0

[PATCH v3 2/2] block-dma: properly take MMIO path

[Leon Romanovsky]

From: Leon Romanovsky <leonro@nvidia.com>

In commit eadaa8b255f3 ("dma-mapping: introduce new DMA attribute to
indicate MMIO memory"), DMA_ATTR_MMIO attribute was added to describe
MMIO addresses, which require to avoid any memory cache flushing, as
an outcome of the discussion pointed in Link tag below.

In case of PCI_P2PDMA_MAP_THRU_HOST_BRIDGE transfer, blk-mq-dm logic
treated this as regular page and relied on "struct page" DMA flow.
That flow performs CPU cache flushing, which shouldn't be done here,
and doesn't set IOMMU_MMIO flag in DMA-IOMMU case.

As a solution, let's encode peer-to-peer transaction type in NVMe IOD
flags variable and provide it to blk-mq-dma API.

Link: https://lore.kernel.org/all/f912c446-1ae9-4390-9c11-00dce7bf0fd3@arm.com/
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 block/blk-mq-dma.c            | 17 +++++++----
 drivers/nvme/host/pci.c       | 69 ++++++++++++++++++++++++++++++++++++++-----
 include/linux/bio-integrity.h |  1 -
 include/linux/blk-integrity.h | 14 ---------
 include/linux/blk-mq-dma.h    | 28 ++++++++----------
 include/linux/blk_types.h     |  2 --
 6 files changed, 85 insertions(+), 46 deletions(-)

diff --git a/block/blk-mq-dma.c b/block/blk-mq-dma.c
index 4ba7b0323da4..98554929507a 100644
--- a/block/blk-mq-dma.c
+++ b/block/blk-mq-dma.c
@@ -93,8 +93,13 @@ static bool blk_dma_map_bus(struct blk_dma_iter *iter, struct phys_vec *vec)
 static bool blk_dma_map_direct(struct request *req, struct device *dma_dev,
 		struct blk_dma_iter *iter, struct phys_vec *vec)
 {
+	unsigned int attrs = 0;
+
+	if (iter->p2pdma.map == PCI_P2PDMA_MAP_THRU_HOST_BRIDGE)
+		attrs |= DMA_ATTR_MMIO;
+
 	iter->addr = dma_map_phys(dma_dev, vec->paddr, vec->len,
-			rq_dma_dir(req), 0);
+			rq_dma_dir(req), attrs);
 	if (dma_mapping_error(dma_dev, iter->addr)) {
 		iter->status = BLK_STS_RESOURCE;
 		return false;
@@ -109,14 +114,18 @@ static bool blk_rq_dma_map_iova(struct request *req, struct device *dma_dev,
 {
 	enum dma_data_direction dir = rq_dma_dir(req);
 	unsigned int mapped = 0;
+	unsigned int attrs = 0;
 	int error;
 
 	iter->addr = state->addr;
 	iter->len = dma_iova_size(state);
 
+	if (iter->p2pdma.map == PCI_P2PDMA_MAP_THRU_HOST_BRIDGE)
+		attrs |= DMA_ATTR_MMIO;
+
 	do {
 		error = dma_iova_link(dma_dev, state, vec->paddr, mapped,
-				vec->len, dir, 0);
+				vec->len, dir, attrs);
 		if (error)
 			break;
 		mapped += vec->len;
@@ -174,10 +183,6 @@ static bool blk_dma_map_iter_start(struct request *req, struct device *dma_dev,
 	switch (pci_p2pdma_state(&iter->p2pdma, dma_dev,
 				 phys_to_page(vec.paddr))) {
 	case PCI_P2PDMA_MAP_BUS_ADDR:
-		if (iter->iter.is_integrity)
-			bio_integrity(req->bio)->bip_flags |= BIP_P2P_DMA;
-		else
-			req->cmd_flags |= REQ_P2PDMA;
 		return blk_dma_map_bus(iter, &vec);
 	case PCI_P2PDMA_MAP_THRU_HOST_BRIDGE:
 		/*
diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index 002412431940..dbf1a924d7bf 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -260,8 +260,20 @@ enum nvme_iod_flags {
 	/* single segment dma mapping */
 	IOD_SINGLE_SEGMENT	= 1U << 2,
 
+	/* Data payload contains p2p memory */
+	IOD_DATA_P2P		= 1U << 3,
+
+	/* Metadata contains p2p memory */
+	IOD_META_P2P		= 1U << 4,
+
+	/* Data payload contains MMIO memory */
+	IOD_DATA_MMIO		= 1U << 5,
+
+	/* Metadata contains MMIO memory */
+	IOD_META_MMIO		= 1U << 6,
+
 	/* Metadata using non-coalesced MPTR */
-	IOD_SINGLE_META_SEGMENT	= 1U << 5,
+	IOD_SINGLE_META_SEGMENT	= 1U << 7,
 };
 
 struct nvme_dma_vec {
@@ -720,10 +732,12 @@ static void nvme_free_sgls(struct request *req, struct nvme_sgl_desc *sge,
 static void nvme_unmap_metadata(struct request *req)
 {
 	struct nvme_queue *nvmeq = req->mq_hctx->driver_data;
+	enum pci_p2pdma_map_type map = PCI_P2PDMA_MAP_NONE;
 	enum dma_data_direction dir = rq_dma_dir(req);
 	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
 	struct device *dma_dev = nvmeq->dev->dev;
 	struct nvme_sgl_desc *sge = iod->meta_descriptor;
+	unsigned int attrs = 0;
 
 	if (iod->flags & IOD_SINGLE_META_SEGMENT) {
 		dma_unmap_page(dma_dev, iod->meta_dma,
@@ -732,13 +746,20 @@ static void nvme_unmap_metadata(struct request *req)
 		return;
 	}
 
-	if (!blk_rq_integrity_dma_unmap(req, dma_dev, &iod->meta_dma_state,
-					iod->meta_total_len)) {
+	if (iod->flags & IOD_META_P2P)
+		map = PCI_P2PDMA_MAP_BUS_ADDR;
+	if (iod->flags & IOD_META_MMIO) {
+		map = PCI_P2PDMA_MAP_THRU_HOST_BRIDGE;
+		attrs |= DMA_ATTR_MMIO;
+	}
+
+	if (!blk_rq_dma_unmap(req, dma_dev, &iod->meta_dma_state,
+			      iod->meta_total_len, map)) {
 		if (nvme_pci_cmd_use_meta_sgl(&iod->cmd))
-			nvme_free_sgls(req, sge, &sge[1], 0);
+			nvme_free_sgls(req, sge, &sge[1], attrs);
 		else
 			dma_unmap_phys(dma_dev, iod->meta_dma,
-				       iod->meta_total_len, dir, 0);
+				       iod->meta_total_len, dir, attrs);
 	}
 
 	if (iod->meta_descriptor)
@@ -748,9 +769,11 @@ static void nvme_unmap_metadata(struct request *req)
 
 static void nvme_unmap_data(struct request *req)
 {
+	enum pci_p2pdma_map_type map = PCI_P2PDMA_MAP_NONE;
 	struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
 	struct nvme_queue *nvmeq = req->mq_hctx->driver_data;
 	struct device *dma_dev = nvmeq->dev->dev;
+	unsigned int attrs = 0;
 
 	if (iod->flags & IOD_SINGLE_SEGMENT) {
 		static_assert(offsetof(union nvme_data_ptr, prp1) ==
@@ -760,12 +783,20 @@ static void nvme_unmap_data(struct request *req)
 		return;
 	}
 
-	if (!blk_rq_dma_unmap(req, dma_dev, &iod->dma_state, iod->total_len)) {
+	if (iod->flags & IOD_DATA_P2P)
+		map = PCI_P2PDMA_MAP_BUS_ADDR;
+	if (iod->flags & IOD_DATA_MMIO) {
+		map = PCI_P2PDMA_MAP_THRU_HOST_BRIDGE;
+		attrs |= DMA_ATTR_MMIO;
+	}
+
+	if (!blk_rq_dma_unmap(req, dma_dev, &iod->dma_state, iod->total_len,
+			      map)) {
 		if (nvme_pci_cmd_use_sgl(&iod->cmd))
 			nvme_free_sgls(req, iod->descriptors[0],
-				       &iod->cmd.common.dptr.sgl, 0);
+				       &iod->cmd.common.dptr.sgl, attrs);
 		else
-			nvme_free_prps(req, 0);
+			nvme_free_prps(req, attrs);
 	}
 
 	if (iod->nr_descriptors)
@@ -1036,6 +1067,17 @@ static blk_status_t nvme_map_data(struct request *req)
 	if (!blk_rq_dma_map_iter_start(req, dev->dev, &iod->dma_state, &iter))
 		return iter.status;
 
+	switch (iter.p2pdma.map) {
+	case PCI_P2PDMA_MAP_BUS_ADDR:
+		iod->flags |= IOD_DATA_P2P;
+		break;
+	case PCI_P2PDMA_MAP_THRU_HOST_BRIDGE:
+		iod->flags |= IOD_DATA_MMIO;
+		break;
+	default:
+		return BLK_STS_RESOURCE;
+	}
+
 	if (use_sgl == SGL_FORCED ||
 	    (use_sgl == SGL_SUPPORTED &&
 	     (sgl_threshold && nvme_pci_avg_seg_size(req) >= sgl_threshold)))
@@ -1058,6 +1100,17 @@ static blk_status_t nvme_pci_setup_meta_sgls(struct request *req)
 						&iod->meta_dma_state, &iter))
 		return iter.status;
 
+	switch (iter.p2pdma.map) {
+	case PCI_P2PDMA_MAP_BUS_ADDR:
+		iod->flags |= IOD_META_P2P;
+		break;
+	case PCI_P2PDMA_MAP_THRU_HOST_BRIDGE:
+		iod->flags |= IOD_META_MMIO;
+		break;
+	default:
+		return BLK_STS_RESOURCE;
+	}
+
 	if (blk_rq_dma_map_coalesce(&iod->meta_dma_state))
 		entries = 1;
 
diff --git a/include/linux/bio-integrity.h b/include/linux/bio-integrity.h
index 851254f36eb3..0a25716820fe 100644
--- a/include/linux/bio-integrity.h
+++ b/include/linux/bio-integrity.h
@@ -13,7 +13,6 @@ enum bip_flags {
 	BIP_CHECK_GUARD		= 1 << 5, /* guard check */
 	BIP_CHECK_REFTAG	= 1 << 6, /* reftag check */
 	BIP_CHECK_APPTAG	= 1 << 7, /* apptag check */
-	BIP_P2P_DMA		= 1 << 8, /* using P2P address */
 };
 
 struct bio_integrity_payload {
diff --git a/include/linux/blk-integrity.h b/include/linux/blk-integrity.h
index b659373788f6..b9e6376b5e36 100644
--- a/include/linux/blk-integrity.h
+++ b/include/linux/blk-integrity.h
@@ -28,14 +28,6 @@ static inline bool queue_limits_stack_integrity_bdev(struct queue_limits *t,
 #ifdef CONFIG_BLK_DEV_INTEGRITY
 int blk_rq_map_integrity_sg(struct request *, struct scatterlist *);
 
-static inline bool blk_rq_integrity_dma_unmap(struct request *req,
-		struct device *dma_dev, struct dma_iova_state *state,
-		size_t mapped_len)
-{
-	return blk_dma_unmap(req, dma_dev, state, mapped_len,
-			bio_integrity(req->bio)->bip_flags & BIP_P2P_DMA);
-}
-
 int blk_rq_count_integrity_sg(struct request_queue *, struct bio *);
 int blk_rq_integrity_map_user(struct request *rq, void __user *ubuf,
 			      ssize_t bytes);
@@ -124,12 +116,6 @@ static inline int blk_rq_map_integrity_sg(struct request *q,
 {
 	return 0;
 }
-static inline bool blk_rq_integrity_dma_unmap(struct request *req,
-		struct device *dma_dev, struct dma_iova_state *state,
-		size_t mapped_len)
-{
-	return false;
-}
 static inline int blk_rq_integrity_map_user(struct request *rq,
 					    void __user *ubuf,
 					    ssize_t bytes)
diff --git a/include/linux/blk-mq-dma.h b/include/linux/blk-mq-dma.h
index 51829958d872..cb88fc791fbd 100644
--- a/include/linux/blk-mq-dma.h
+++ b/include/linux/blk-mq-dma.h
@@ -16,13 +16,13 @@ struct blk_dma_iter {
 	/* Output address range for this iteration */
 	dma_addr_t			addr;
 	u32				len;
+	struct pci_p2pdma_map_state	p2pdma;
 
 	/* Status code. Only valid when blk_rq_dma_map_iter_* returned false */
 	blk_status_t			status;
 
 	/* Internal to blk_rq_dma_map_iter_* */
 	struct blk_map_iter		iter;
-	struct pci_p2pdma_map_state	p2pdma;
 };
 
 bool blk_rq_dma_map_iter_start(struct request *req, struct device *dma_dev,
@@ -43,36 +43,34 @@ static inline bool blk_rq_dma_map_coalesce(struct dma_iova_state *state)
 }
 
 /**
- * blk_dma_unmap - try to DMA unmap a request
+ * blk_rq_dma_unmap - try to DMA unmap a request
  * @req:	request to unmap
  * @dma_dev:	device to unmap from
  * @state:	DMA IOVA state
  * @mapped_len: number of bytes to unmap
- * @is_p2p:	true if mapped with PCI_P2PDMA_MAP_BUS_ADDR
+ * @map:	peer-to-peer mapping type
  *
  * Returns %false if the callers need to manually unmap every DMA segment
  * mapped using @iter or %true if no work is left to be done.
  */
-static inline bool blk_dma_unmap(struct request *req, struct device *dma_dev,
-		struct dma_iova_state *state, size_t mapped_len, bool is_p2p)
+static inline bool blk_rq_dma_unmap(struct request *req, struct device *dma_dev,
+		struct dma_iova_state *state, size_t mapped_len,
+		enum pci_p2pdma_map_type map)
 {
-	if (is_p2p)
+	if (map == PCI_P2PDMA_MAP_BUS_ADDR)
 		return true;
 
 	if (dma_use_iova(state)) {
+		unsigned int attrs = 0;
+
+		if (map == PCI_P2PDMA_MAP_THRU_HOST_BRIDGE)
+			attrs |= DMA_ATTR_MMIO;
+
 		dma_iova_destroy(dma_dev, state, mapped_len, rq_dma_dir(req),
-				 0);
+				 attrs);
 		return true;
 	}
 
 	return !dma_need_unmap(dma_dev);
 }
-
-static inline bool blk_rq_dma_unmap(struct request *req, struct device *dma_dev,
-		struct dma_iova_state *state, size_t mapped_len)
-{
-	return blk_dma_unmap(req, dma_dev, state, mapped_len,
-				req->cmd_flags & REQ_P2PDMA);
-}
-
 #endif /* BLK_MQ_DMA_H */
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index 8e8d1cc8b06c..1f8e429c4c77 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -381,7 +381,6 @@ enum req_flag_bits {
 	__REQ_DRV,		/* for driver use */
 	__REQ_FS_PRIVATE,	/* for file system (submitter) use */
 	__REQ_ATOMIC,		/* for atomic write operations */
-	__REQ_P2PDMA,		/* contains P2P DMA pages */
 	/*
 	 * Command specific flags, keep last:
 	 */
@@ -414,7 +413,6 @@ enum req_flag_bits {
 #define REQ_DRV		(__force blk_opf_t)(1ULL << __REQ_DRV)
 #define REQ_FS_PRIVATE	(__force blk_opf_t)(1ULL << __REQ_FS_PRIVATE)
 #define REQ_ATOMIC	(__force blk_opf_t)(1ULL << __REQ_ATOMIC)
-#define REQ_P2PDMA	(__force blk_opf_t)(1ULL << __REQ_P2PDMA)
 
 #define REQ_NOUNMAP	(__force blk_opf_t)(1ULL << __REQ_NOUNMAP)
 

-- 
2.51.0

Re: [PATCH v3 2/2] block-dma: properly take MMIO path

[Christoph Hellwig]

> +	switch (iter.p2pdma.map) {
> +	case PCI_P2PDMA_MAP_BUS_ADDR:
> +		iod->flags |= IOD_DATA_P2P;
> +		break;
> +	case PCI_P2PDMA_MAP_THRU_HOST_BRIDGE:
> +		iod->flags |= IOD_DATA_MMIO;
> +		break;
> +	default:
> +		return BLK_STS_RESOURCE;

I almost wonder if we should just the pci_p2pdma_map_type values into
place.  But that's a future cleanup, I'd rather get this going now.

> +static inline bool blk_rq_dma_unmap(struct request *req, struct device *dma_dev,
> +		struct dma_iova_state *state, size_t mapped_len,
> +		enum pci_p2pdma_map_type map)
>  {
> -	if (is_p2p)
> +	if (map == PCI_P2PDMA_MAP_BUS_ADDR)
>  		return true;
>  
>  	if (dma_use_iova(state)) {
> +		unsigned int attrs = 0;
> +
> +		if (map == PCI_P2PDMA_MAP_THRU_HOST_BRIDGE)
> +			attrs |= DMA_ATTR_MMIO;
> +
>  		dma_iova_destroy(dma_dev, state, mapped_len, rq_dma_dir(req),
> -				 0);
> +				 attrs);

The only thing in req that is used now is the data directrion.  I'd be
almost tempted to just pass that and lift this to dma-mapping.h.

But I guess we could just do that in a follow on to not drag in
another subsystem.

Otherwise looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH v3 2/2] block-dma: properly take MMIO path

[Leon Romanovsky]

On Mon, Oct 27, 2025 at 08:49:22AM +0100, Christoph Hellwig wrote:
> > +	switch (iter.p2pdma.map) {
> > +	case PCI_P2PDMA_MAP_BUS_ADDR:
> > +		iod->flags |= IOD_DATA_P2P;
> > +		break;
> > +	case PCI_P2PDMA_MAP_THRU_HOST_BRIDGE:
> > +		iod->flags |= IOD_DATA_MMIO;
> > +		break;
> > +	default:
> > +		return BLK_STS_RESOURCE;
> 
> I almost wonder if we should just the pci_p2pdma_map_type values into
> place.  But that's a future cleanup, I'd rather get this going now.

I thought about it, but decided to use flags as more space efficient.
PCI_P2PDMA_MAP_BUS_ADDR == 3 and PCI_P2PDMA_MAP_THRU_HOST_BRIDGE == 4.
It means that will need to occupy 6 bits (extra u8 IOD field), while
this flags encoding takes only 4 bits (without extra IOD field).

> 
> > +static inline bool blk_rq_dma_unmap(struct request *req, struct device *dma_dev,
> > +		struct dma_iova_state *state, size_t mapped_len,
> > +		enum pci_p2pdma_map_type map)
> >  {
> > -	if (is_p2p)
> > +	if (map == PCI_P2PDMA_MAP_BUS_ADDR)
> >  		return true;
> >  
> >  	if (dma_use_iova(state)) {
> > +		unsigned int attrs = 0;
> > +
> > +		if (map == PCI_P2PDMA_MAP_THRU_HOST_BRIDGE)
> > +			attrs |= DMA_ATTR_MMIO;
> > +
> >  		dma_iova_destroy(dma_dev, state, mapped_len, rq_dma_dir(req),
> > -				 0);
> > +				 attrs);
> 
> The only thing in req that is used now is the data directrion.  I'd be
> almost tempted to just pass that and lift this to dma-mapping.h.
> 
> But I guess we could just do that in a follow on to not drag in
> another subsystem.
> 
> Otherwise looks good:
> 
> Reviewed-by: Christoph Hellwig <hch@lst.de>

Thanks

Re: [PATCH v3 2/2] block-dma: properly take MMIO path

[Leon Romanovsky]

On Mon, Oct 27, 2025 at 09:57:38AM +0200, Leon Romanovsky wrote:
> On Mon, Oct 27, 2025 at 08:49:22AM +0100, Christoph Hellwig wrote:
> > > +	switch (iter.p2pdma.map) {
> > > +	case PCI_P2PDMA_MAP_BUS_ADDR:
> > > +		iod->flags |= IOD_DATA_P2P;
> > > +		break;
> > > +	case PCI_P2PDMA_MAP_THRU_HOST_BRIDGE:
> > > +		iod->flags |= IOD_DATA_MMIO;
> > > +		break;
> > > +	default:
> > > +		return BLK_STS_RESOURCE;
> > 
> > I almost wonder if we should just the pci_p2pdma_map_type values into
> > place.  But that's a future cleanup, I'd rather get this going now.
> 
> I thought about it, but decided to use flags as more space efficient.
> PCI_P2PDMA_MAP_BUS_ADDR == 3 and PCI_P2PDMA_MAP_THRU_HOST_BRIDGE == 4.
> It means that will need to occupy 6 bits (extra u8 IOD field), while
> this flags encoding takes only 4 bits (without extra IOD field).

BTW, one can reorganize pci_p2pdma_map_type to make sure that
PCI_P2PDMA_MAP_BUS_ADDR will be 2 and PCI_P2PDMA_MAP_THRU_HOST_BRIDGE
will be 3, so it will be possible to take 2 bits per-type instead of 3.

diff --git a/include/linux/pci-p2pdma.h b/include/linux/pci-p2pdma.h
index 951f81a38f3a..2280e5a55b1f 100644
--- a/include/linux/pci-p2pdma.h
+++ b/include/linux/pci-p2pdma.h
@@ -112,14 +112,6 @@ enum pci_p2pdma_map_type {
         */
        PCI_P2PDMA_MAP_NONE,

-       /*
-        * PCI_P2PDMA_MAP_NOT_SUPPORTED: Indicates the transaction will
-        * traverse the host bridge and the host bridge is not in the
-        * allowlist. DMA Mapping routines should return an error when
-        * this is returned.
-        */
-       PCI_P2PDMA_MAP_NOT_SUPPORTED,
-
        /*
         * PCI_P2PDMA_MAP_BUS_ADDR: Indicates that two devices can talk to
         * each other directly through a PCI switch and the transaction will
@@ -136,6 +128,14 @@ enum pci_p2pdma_map_type {
         * case of IOMMUs) should be used to program the DMA engine.
         */
        PCI_P2PDMA_MAP_THRU_HOST_BRIDGE,
+
+       /*
+        * PCI_P2PDMA_MAP_NOT_SUPPORTED: Indicates the transaction will
+        * traverse the host bridge and the host bridge is not in the
+        * allowlist. DMA Mapping routines should return an error when
+        * this is returned.
+        */
+       PCI_P2PDMA_MAP_NOT_SUPPORTED,
 };

 struct pci_p2pdma_map_state {

Thanks

Re: [PATCH v3 2/2] block-dma: properly take MMIO path

[Christoph Hellwig]

On Mon, Oct 27, 2025 at 10:11:42AM +0200, Leon Romanovsky wrote:
> BTW, one can reorganize pci_p2pdma_map_type to make sure that
> PCI_P2PDMA_MAP_BUS_ADDR will be 2 and PCI_P2PDMA_MAP_THRU_HOST_BRIDGE
> will be 3, so it will be possible to take 2 bits per-type instead of 3.

Maybe.  I'd be happy to leave this as-is and then see if we can optimize
things once we grow more users.

Re: [PATCH v3 2/2] block-dma: properly take MMIO path

[Leon Romanovsky]

On Mon, Oct 27, 2025 at 09:16:58AM +0100, Christoph Hellwig wrote:
> On Mon, Oct 27, 2025 at 10:11:42AM +0200, Leon Romanovsky wrote:
> > BTW, one can reorganize pci_p2pdma_map_type to make sure that
> > PCI_P2PDMA_MAP_BUS_ADDR will be 2 and PCI_P2PDMA_MAP_THRU_HOST_BRIDGE
> > will be 3, so it will be possible to take 2 bits per-type instead of 3.
> 
> Maybe.  I'd be happy to leave this as-is and then see if we can optimize
> things once we grow more users.

Completely agree, thanks for the patience and review.

Re: [PATCH v3 2/2] block-dma: properly take MMIO path

[Keith Busch]

On Mon, Oct 27, 2025 at 09:30:21AM +0200, Leon Romanovsky wrote:
> @@ -732,13 +746,20 @@ static void nvme_unmap_metadata(struct request *req)
>  		return;
>  	}
>  
> -	if (!blk_rq_integrity_dma_unmap(req, dma_dev, &iod->meta_dma_state,
> -					iod->meta_total_len)) {
> +	if (iod->flags & IOD_META_P2P)
> +		map = PCI_P2PDMA_MAP_BUS_ADDR;
> +	if (iod->flags & IOD_META_MMIO) {
> +		map = PCI_P2PDMA_MAP_THRU_HOST_BRIDGE;
> +		attrs |= DMA_ATTR_MMIO;
> +	}

This should be an 'else if' no?

> @@ -760,12 +783,20 @@ static void nvme_unmap_data(struct request *req)
>  		return;
>  	}
>  
> -	if (!blk_rq_dma_unmap(req, dma_dev, &iod->dma_state, iod->total_len)) {
> +	if (iod->flags & IOD_DATA_P2P)
> +		map = PCI_P2PDMA_MAP_BUS_ADDR;
> +	if (iod->flags & IOD_DATA_MMIO) {
> +		map = PCI_P2PDMA_MAP_THRU_HOST_BRIDGE;
> +		attrs |= DMA_ATTR_MMIO;
> +	}

Same here.

> @@ -1036,6 +1067,17 @@ static blk_status_t nvme_map_data(struct request *req)
>  	if (!blk_rq_dma_map_iter_start(req, dev->dev, &iod->dma_state, &iter))
>  		return iter.status;
>  
> +	switch (iter.p2pdma.map) {
> +	case PCI_P2PDMA_MAP_BUS_ADDR:
> +		iod->flags |= IOD_DATA_P2P;
> +		break;
> +	case PCI_P2PDMA_MAP_THRU_HOST_BRIDGE:
> +		iod->flags |= IOD_DATA_MMIO;
> +		break;
> +	default:
> +		return BLK_STS_RESOURCE;
> +	}

Why is it an error if the data isn't a P2PDMA type?

> @@ -1058,6 +1100,17 @@ static blk_status_t nvme_pci_setup_meta_sgls(struct request *req)
>  						&iod->meta_dma_state, &iter))
>  		return iter.status;
>  
> +	switch (iter.p2pdma.map) {
> +	case PCI_P2PDMA_MAP_BUS_ADDR:
> +		iod->flags |= IOD_META_P2P;
> +		break;
> +	case PCI_P2PDMA_MAP_THRU_HOST_BRIDGE:
> +		iod->flags |= IOD_META_MMIO;
> +		break;
> +	default:
> +		return BLK_STS_RESOURCE;
> +	}

Same here.

Re: [PATCH v3 2/2] block-dma: properly take MMIO path

[Leon Romanovsky]

On Wed, Nov 12, 2025 at 09:38:07AM -0500, Keith Busch wrote:
> On Mon, Oct 27, 2025 at 09:30:21AM +0200, Leon Romanovsky wrote:
> > @@ -732,13 +746,20 @@ static void nvme_unmap_metadata(struct request *req)
> >  		return;
> >  	}
> >  
> > -	if (!blk_rq_integrity_dma_unmap(req, dma_dev, &iod->meta_dma_state,
> > -					iod->meta_total_len)) {
> > +	if (iod->flags & IOD_META_P2P)
> > +		map = PCI_P2PDMA_MAP_BUS_ADDR;
> > +	if (iod->flags & IOD_META_MMIO) {
> > +		map = PCI_P2PDMA_MAP_THRU_HOST_BRIDGE;
> > +		attrs |= DMA_ATTR_MMIO;
> > +	}
> 
> This should be an 'else if' no?

We can't have both IOD_META_P2P and IOD_META_MMIO. It can be only one or
IOD_META_P2P or IOD_META_MMIO. In this case "else if' is not necessary.

> 

<...>

> > @@ -1036,6 +1067,17 @@ static blk_status_t nvme_map_data(struct request *req)
> >  	if (!blk_rq_dma_map_iter_start(req, dev->dev, &iod->dma_state, &iter))
> >  		return iter.status;
> >  
> > +	switch (iter.p2pdma.map) {
> > +	case PCI_P2PDMA_MAP_BUS_ADDR:
> > +		iod->flags |= IOD_DATA_P2P;
> > +		break;
> > +	case PCI_P2PDMA_MAP_THRU_HOST_BRIDGE:
> > +		iod->flags |= IOD_DATA_MMIO;
> > +		break;
> > +	default:
> > +		return BLK_STS_RESOURCE;
> > +	}
> 
> Why is it an error if the data isn't a P2PDMA type?

This is probably bug, I was supposed to handle PCI_P2PDMA_MAP_NONE case too.

Thanks

Re: [PATCH v3 2/2] block-dma: properly take MMIO path

[Keith Busch]

On Wed, Nov 12, 2025 at 05:16:04PM +0200, Leon Romanovsky wrote:
> On Wed, Nov 12, 2025 at 09:38:07AM -0500, Keith Busch wrote:
> > On Mon, Oct 27, 2025 at 09:30:21AM +0200, Leon Romanovsky wrote:
> > > @@ -732,13 +746,20 @@ static void nvme_unmap_metadata(struct request *req)
> > >  		return;
> > >  	}
> > >  
> > > -	if (!blk_rq_integrity_dma_unmap(req, dma_dev, &iod->meta_dma_state,
> > > -					iod->meta_total_len)) {
> > > +	if (iod->flags & IOD_META_P2P)
> > > +		map = PCI_P2PDMA_MAP_BUS_ADDR;
> > > +	if (iod->flags & IOD_META_MMIO) {
> > > +		map = PCI_P2PDMA_MAP_THRU_HOST_BRIDGE;
> > > +		attrs |= DMA_ATTR_MMIO;
> > > +	}
> > 
> > This should be an 'else if' no?
> 
> We can't have both IOD_META_P2P and IOD_META_MMIO. It can be only one or
> IOD_META_P2P or IOD_META_MMIO. In this case "else if' is not necessary.

I get that and I'm not saying this is a bug, but it just looks odd.
You're checking the conditions as if they're independent of each other
when they're not. If the first 'if' is true, there's no need to test the
second one as we know it's false, but this code flow will test it
anyway.

Re: [PATCH v3 2/2] block-dma: properly take MMIO path

[Leon Romanovsky]

On Wed, Nov 12, 2025 at 10:23:19AM -0500, Keith Busch wrote:
> On Wed, Nov 12, 2025 at 05:16:04PM +0200, Leon Romanovsky wrote:
> > On Wed, Nov 12, 2025 at 09:38:07AM -0500, Keith Busch wrote:
> > > On Mon, Oct 27, 2025 at 09:30:21AM +0200, Leon Romanovsky wrote:
> > > > @@ -732,13 +746,20 @@ static void nvme_unmap_metadata(struct request *req)
> > > >  		return;
> > > >  	}
> > > >  
> > > > -	if (!blk_rq_integrity_dma_unmap(req, dma_dev, &iod->meta_dma_state,
> > > > -					iod->meta_total_len)) {
> > > > +	if (iod->flags & IOD_META_P2P)
> > > > +		map = PCI_P2PDMA_MAP_BUS_ADDR;
> > > > +	if (iod->flags & IOD_META_MMIO) {
> > > > +		map = PCI_P2PDMA_MAP_THRU_HOST_BRIDGE;
> > > > +		attrs |= DMA_ATTR_MMIO;
> > > > +	}
> > > 
> > > This should be an 'else if' no?
> > 
> > We can't have both IOD_META_P2P and IOD_META_MMIO. It can be only one or
> > IOD_META_P2P or IOD_META_MMIO. In this case "else if' is not necessary.
> 
> I get that and I'm not saying this is a bug, but it just looks odd.

Sorry, I didn't get the intent.

Thanks

> You're checking the conditions as if they're independent of each other
> when they're not. If the first 'if' is true, there's no need to test the
> second one as we know it's false, but this code flow will test it
> anyway.

Re: [PATCH v3 0/2] block: Enable proper MMIO memory handling for P2P DMA

[Leon Romanovsky]

On Mon, Oct 27, 2025 at 09:30:19AM +0200, Leon Romanovsky wrote:

<...>

> ----------------------------------------------------------------------
> 
> This patch series improves block layer and NVMe driver support for MMIO
> memory regions, particularly for peer-to-peer (P2P) DMA transfers that
> go through the host bridge.
> 
> The series addresses a critical gap where P2P transfers through the host
> bridge (PCI_P2PDMA_MAP_THRU_HOST_BRIDGE) were not properly marked as
> MMIO memory, leading to potential issues with:
> 
> - Inappropriate CPU cache synchronization operations on MMIO regions
> - Incorrect DMA mapping/unmapping that doesn't respect MMIO semantics  
> - Missing IOMMU configuration for MMIO memory handling
> 
> This work is extracted from the larger DMA physical API improvement
> series [1] and focuses specifically on block layer and NVMe requirements
> for MMIO memory support.
> 
> Thanks
> 
> [1] https://lore.kernel.org/all/cover.1757423202.git.leonro@nvidia.com/
> 
> Leon Romanovsky (2):
>       nvme-pci: migrate to dma_map_phys instead of map_page
>       block-dma: properly take MMIO path

Hi,

Kind reminder.

Thanks