[PATCH] lightnvm: introduce ADDR_PADDED to signal padding

[PATCH] lightnvm: introduce ADDR_PADDED to signal padding

[Javier González]

When REQ_FUA or REQ_FLUSH are sent from the block layer, a target might
need to pad some sectors in order to meet device constrains. ADDR_PADDED
allows to signal that such sector has been padded. It can be used to
mark both lbas and ppas.

Signed-off-by: Javier González <javier@cnexlabs.com>
---
 include/linux/lightnvm.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/include/linux/lightnvm.h b/include/linux/lightnvm.h
index 37cda94..5c48207 100644
--- a/include/linux/lightnvm.h
+++ b/include/linux/lightnvm.h
@@ -214,6 +214,7 @@ struct nvm_tgt_instance {
 };
 
 #define ADDR_EMPTY (~0ULL)
+#define ADDR_PADDED (~0ULL - 1)
 
 #define NVM_VERSION_MAJOR 1
 #define NVM_VERSION_MINOR 0
-- 
2.5.0

[PATCH 2/2] lightnvm: enable metadata to be sent to device

[Javier González]

Enable metadata to be sent to the device through the metadata field on
the physical rw nvme command. When a single ppa is sent to the device, a
64-bit integer can be sent as metadata; when a ppa list is sent, a
64-bit integer list mapping to the ppa list can be used to send
metadata.

Signed-off-by: Javier González <javier@cnexlabs.com>
---
 drivers/nvme/host/lightnvm.c | 1 +
 include/linux/lightnvm.h     | 8 ++++++--
 2 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/drivers/nvme/host/lightnvm.c b/drivers/nvme/host/lightnvm.c
index 4956de3..7886c2c 100644
--- a/drivers/nvme/host/lightnvm.c
+++ b/drivers/nvme/host/lightnvm.c
@@ -504,6 +504,7 @@ static inline void nvme_nvm_rqtocmd(struct request *rq, struct nvm_rq *rqd,
 	c->ph_rw.opcode = rqd->opcode;
 	c->ph_rw.nsid = cpu_to_le32(ns->ns_id);
 	c->ph_rw.spba = cpu_to_le64(rqd->ppa_addr.ppa);
+	c->ph_rw.metadata = cpu_to_le64(rqd->metadata);
 	c->ph_rw.control = cpu_to_le16(rqd->flags);
 	c->ph_rw.length = cpu_to_le16(rqd->nr_pages - 1);
 
diff --git a/include/linux/lightnvm.h b/include/linux/lightnvm.h
index 208887f..f0936e3 100644
--- a/include/linux/lightnvm.h
+++ b/include/linux/lightnvm.h
@@ -237,8 +237,12 @@ struct nvm_rq {
 
 	struct ppa_addr *ppa_list;
 
-	void *metadata;
-	dma_addr_t dma_metadata;
+	union {
+		u64 metadata;
+		dma_addr_t dma_meta_list;
+	};
+
+	u64 *meta_list;
 
 	struct completion *wait;
 	nvm_end_io_fn *end_io;
-- 
2.5.0

[PATCH 1/2] lightnvm: free metadata dma on rrpc when needed

[Javier González]

rrpc has assumed that medatada is always sent as a dma buffer to the
device. When this is not true, do not attempt to free a non-allocated
dma region.

Signed-off-by: Javier González <javier@cnexlabs.com>
---
 drivers/lightnvm/rrpc.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/drivers/lightnvm/rrpc.c b/drivers/lightnvm/rrpc.c
index a9a5fcc..ba789c3 100644
--- a/drivers/lightnvm/rrpc.c
+++ b/drivers/lightnvm/rrpc.c
@@ -707,10 +707,12 @@ static void rrpc_end_io(struct nvm_rq *rqd)
 
 	rrpc_unlock_rq(rrpc, rqd);
 
-	if (npages > 1)
+	if (npages > 1) {
 		nvm_dev_dma_free(rrpc->dev, rqd->ppa_list, rqd->dma_ppa_list);
-	if (rqd->metadata)
-		nvm_dev_dma_free(rrpc->dev, rqd->metadata, rqd->dma_metadata);
+		if (rqd->meta_list)
+			nvm_dev_dma_free(rrpc->dev, rqd->meta_list,
+							rqd->dma_meta_list);
+	}
 
 	mempool_free(rqd, rrpc->rq_pool);
 }
-- 
2.5.0