[PATCHv3 2/2] nvme: remove virtual boundary for sgl capable devices

[Keith Busch <kbusch@meta.com>]

From: Keith Busch <kbusch@kernel.org>

The nvme virtual boundary is only required for the PRP format. Devices
that can use SGL for DMA don't need it for IO queues. Drop reporting it
for such PCIe devices; fabrics controllers will continue to use the
limit as they currently don't report any boundary requirements.

Applications may continue to align to the same virtual boundaries for
optimization purposes if they want, and the driver will continue to
decide whether to use the PRP format the same as before if the IO allows
it.

Signed-off-by: Keith Busch <kbusch@kernel.org>
---
 drivers/nvme/host/core.c | 21 ++++++++++++++++-----
 drivers/nvme/host/pci.c  | 16 +++++++++++++---
 2 files changed, 29 insertions(+), 8 deletions(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 812c1565114fd..cce1fbf2becd8 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -2065,13 +2065,24 @@ static u32 nvme_max_drv_segments(struct nvme_ctrl *ctrl)
 }
 
 static void nvme_set_ctrl_limits(struct nvme_ctrl *ctrl,
-		struct queue_limits *lim)
+		struct queue_limits *lim, bool admin)
 {
 	lim->max_hw_sectors = ctrl->max_hw_sectors;
 	lim->max_segments = min_t(u32, USHRT_MAX,
 		min_not_zero(nvme_max_drv_segments(ctrl), ctrl->max_segments));
 	lim->max_integrity_segments = ctrl->max_integrity_segments;
-	lim->virt_boundary_mask = NVME_CTRL_PAGE_SIZE - 1;
+
+	/*
+	 * The virtual boundary mask is not necessary for PCI controllers that
+	 * support SGL for DMA. It's only necessary when using PRP. Admin
+	 * queues only support PRP, and fabrics drivers currently don't report
+	 * what boundaries they require, so set the virtual boundary for
+	 * either.
+	 */
+	if (!nvme_ctrl_sgl_supported(ctrl) || admin ||
+	    ctrl->ops->flags & NVME_F_FABRICS)
+		lim->virt_boundary_mask = NVME_CTRL_PAGE_SIZE - 1;
+
 	lim->max_segment_size = UINT_MAX;
 	lim->dma_alignment = 3;
 }
@@ -2173,7 +2184,7 @@ static int nvme_update_ns_info_generic(struct nvme_ns *ns,
 	int ret;
 
 	lim = queue_limits_start_update(ns->disk->queue);
-	nvme_set_ctrl_limits(ns->ctrl, &lim);
+	nvme_set_ctrl_limits(ns->ctrl, &lim, false);
 
 	memflags = blk_mq_freeze_queue(ns->disk->queue);
 	ret = queue_limits_commit_update(ns->disk->queue, &lim);
@@ -2377,7 +2388,7 @@ static int nvme_update_ns_info_block(struct nvme_ns *ns,
 	ns->head->lba_shift = id->lbaf[lbaf].ds;
 	ns->head->nuse = le64_to_cpu(id->nuse);
 	capacity = nvme_lba_to_sect(ns->head, le64_to_cpu(id->nsze));
-	nvme_set_ctrl_limits(ns->ctrl, &lim);
+	nvme_set_ctrl_limits(ns->ctrl, &lim, false);
 	nvme_configure_metadata(ns->ctrl, ns->head, id, nvm, info);
 	nvme_set_chunk_sectors(ns, id, &lim);
 	if (!nvme_update_disk_info(ns, id, &lim))
@@ -3580,7 +3591,7 @@ static int nvme_init_identify(struct nvme_ctrl *ctrl)
 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
 
 	lim = queue_limits_start_update(ctrl->admin_q);
-	nvme_set_ctrl_limits(ctrl, &lim);
+	nvme_set_ctrl_limits(ctrl, &lim, true);
 	ret = queue_limits_commit_update(ctrl->admin_q, &lim);
 	if (ret)
 		goto out_free;
diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index d8a9dee55de33..6e848aa4531b6 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -619,9 +619,19 @@ static inline enum nvme_use_sgl nvme_pci_use_sgls(struct nvme_dev *dev,
 	struct nvme_queue *nvmeq = req->mq_hctx->driver_data;
 
 	if (nvmeq->qid && nvme_ctrl_sgl_supported(&dev->ctrl)) {
-		if (nvme_req(req)->flags & NVME_REQ_USERCMD)
-			return SGL_FORCED;
-		if (req->nr_integrity_segments > 1)
+		/*
+		 * A request with page gaps within the controller's mask can
+		 * not use the PRP format.
+		 *
+		 * We force user commands to use SGL because that lets the
+		 * device validate the requested transfer lengths.
+		 *
+		 * Multiple integrity segments must use SGL as that's the only
+		 * way to describe such a command.
+		 */
+		if (req->page_gap & (NVME_CTRL_PAGE_SIZE - 1) ||
+		    nvme_req(req)->flags & NVME_REQ_USERCMD ||
+		    req->nr_integrity_segments > 1)
 			return SGL_FORCED;
 		return SGL_SUPPORTED;
 	}
-- 
2.47.3