[PATCH] nvme: Implement NS Optimial IO Boundary from 1.3 Spec

[PATCH] nvme: Implement NS Optimial IO Boundary from 1.3 Spec

[Scott Bauer]

The NVMe 1.3 spec introduces Namespace Optimal IO Boundaries (NOIOB).
This feature standardizes the stripe mechanism we currently have quirks for.
This patch implements the necessary logic to handle this new feature.

Signed-off-by: Scott Bauer <scott.bauer at intel.com>
---
 drivers/nvme/host/core.c | 7 +++++++
 include/linux/nvme.h     | 2 +-
 2 files changed, 8 insertions(+), 1 deletion(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index d5e0906..69d2798 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -1000,6 +1000,7 @@ static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
 	u8 lbaf, pi_type;
 	u16 old_ms;
 	unsigned short bs;
+	unsigned int iob;
 
 	old_ms = ns->ms;
 	lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
@@ -1027,6 +1028,12 @@ static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
 
 	ns->pi_type = pi_type;
 	blk_queue_logical_block_size(ns->queue, bs);
+	if (ns->ctrl->vs >= NVME_VS(1, 3, 0) && le16_to_cpu(id->noiob)) {
+		/* Convert from controller logical blocks to 512 sectors. */
+		iob = ((unsigned int)le16_to_cpu(id->noiob)) << (ns->lba_shift - 9);
+		if (is_power_of_2(iob))
+			blk_queue_chunk_sectors(ns->queue, iob);
+	}
 
 	if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
 		nvme_init_integrity(ns);
diff --git a/include/linux/nvme.h b/include/linux/nvme.h
index b625bac..d19b193 100644
--- a/include/linux/nvme.h
+++ b/include/linux/nvme.h
@@ -274,7 +274,7 @@ struct nvme_id_ns {
 	__le16			nabsn;
 	__le16			nabo;
 	__le16			nabspf;
-	__u16			rsvd46;
+	__le16			noiob;
 	__u8			nvmcap[16];
 	__u8			rsvd64[40];
 	__u8			nguid[16];
-- 
2.7.4

[PATCH] nvme: Implement NS Optimial IO Boundary from 1.3 Spec

[Keith Busch]

On Fri, May 12, 2017@02:29:05PM -0600, Scott Bauer wrote:
> The NVMe 1.3 spec introduces Namespace Optimal IO Boundaries (NOIOB).
> This feature standardizes the stripe mechanism we currently have quirks for.
> This patch implements the necessary logic to handle this new feature.
> 
> Signed-off-by: Scott Bauer <scott.bauer at intel.com>

Thanks Scott, looks good to me.

Reviewed-by: Keith Busch <keith.busch at intel.com>

> +	if (ns->ctrl->vs >= NVME_VS(1, 3, 0) && le16_to_cpu(id->noiob)) {
> +		/* Convert from controller logical blocks to 512 sectors. */
> +		iob = ((unsigned int)le16_to_cpu(id->noiob)) << (ns->lba_shift - 9);
> +		if (is_power_of_2(iob))
> +			blk_queue_chunk_sectors(ns->queue, iob);
> +	}

Just want to point out that if a Namespace has this set, then magically
doesn't after a rescan, we're stuck with the previous chunk size. Even if
that does happen, the only harm is a command may be split into two smaller
ones. Not a big deal, and that'll never happen in real life anyway.

[PATCH] nvme: Implement NS Optimial IO Boundary from 1.3 Spec

[Christoph Hellwig]

Looks fine,

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

Can you also add support to nvme-cli?