nvme: support 64 LBA formats for NVM and ZNS command set

Format LBA size (FLBAS) is updated to have:
Bit 3:0 as least significant 4 bits for format index
Bit 6:5 as most significant 2 bits for format index

NVMe format command fields are updated accordingly.

Add a new helper function to fetch the correct format index.
Update examples and unit test files accordingly.

Signed-off-by: Ankit Kumar <ankit.kumar@samsung.com>
Change-Id: I2d6d9045b9d65ae91cb18843ca75b59cc27ed2f2
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/16515
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>

CHANGELOG.md

@@ -18,6 +18,11 @@ New function `spdk_env_get_main_core` was added.
 New `spdk_nvmf_request_copy_to/from_buf()` APIs have been added, which support
 iovecs, unlike the deprecated `spdk_nvmf_request_get_data()`.
 
+### nvme
+
+New API `spdk_nvme_ns_get_format_index` was added to calculate the exact format index, that
+was used to format the namespace.
+
 ## v23.01
 
 ### accel

examples/nvme/identify/identify.c

@@ -791,7 +791,7 @@ get_and_print_zns_zone_report(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *q
 	uint64_t handled_zones = 0;
 	uint64_t slba = 0;
 	size_t zdes = 0;
-	uint32_t zds, zrs;
+	uint32_t zds, zrs, format_index;
 	int rc = 0;
 
 	outstanding_commands = 0;
@@ -801,7 +801,9 @@ get_and_print_zns_zone_report(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *q
 
 	zrs = sizeof(struct spdk_nvme_zns_zone_report);
 	zds = sizeof(struct spdk_nvme_zns_zone_desc);
-	zdes = nsdata_zns->lbafe[nsdata->flbas.format].zdes * 64;
+
+	format_index = spdk_nvme_ns_get_format_index(nsdata);
+	zdes = nsdata_zns->lbafe[format_index].zdes * 64;
 
 	report_bufsize = spdk_nvme_ns_get_max_io_xfer_size(ns);
 	report_buf = calloc(1, report_bufsize);
@@ -964,7 +966,7 @@ print_namespace(struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_ns *ns)
 	uint32_t				i;
 	uint32_t				flags;
 	char					uuid_str[SPDK_UUID_STRING_LEN];
-	uint32_t				blocksize;
+	uint32_t				blocksize, format_index;
 	enum spdk_nvme_dealloc_logical_block_read_value	dlfeat_read_value;
 
 	cdata = spdk_nvme_ctrlr_get_data(ctrlr);
@@ -1017,13 +1019,14 @@ print_namespace(struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_ns *ns)
 		printf("Protection Information Transferred as: %s\n",
 		       nsdata->dps.md_start ? "First 8 Bytes" : "Last 8 Bytes");
 	}
-	if (nsdata->lbaf[nsdata->flbas.format].ms > 0) {
+	format_index = spdk_nvme_ns_get_format_index(nsdata);
+	if (nsdata->lbaf[format_index].ms > 0) {
 		printf("Metadata Transferred as:               %s\n",
 		       nsdata->flbas.extended ? "Extended Data LBA" : "Separate Metadata Buffer");
 	}
 	printf("Namespace Sharing Capabilities:        %s\n",
 	       nsdata->nmic.can_share ? "Multiple Controllers" : "Private");
-	blocksize = 1 << nsdata->lbaf[nsdata->flbas.format].lbads;
+	blocksize = 1 << nsdata->lbaf[format_index].lbads;
 	printf("Size (in LBAs):                        %lld (%lldGiB)\n",
 	       (long long)nsdata->nsze,
 	       (long long)nsdata->nsze * blocksize / 1024 / 1024 / 1024);
@@ -1103,7 +1106,7 @@ print_namespace(struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_ns *ns)
 
 	printf("Number of LBA Formats:                 %d\n", nsdata->nlbaf + 1);
 	printf("Current LBA Format:                    LBA Format #%02d\n",
-	       nsdata->flbas.format);
+	       format_index);
 	for (i = 0; i <= nsdata->nlbaf; i++) {
 		printf("LBA Format #%02d: Data Size: %5d  Metadata Size: %5d\n",
 		       i, 1 << nsdata->lbaf[i].lbads, nsdata->lbaf[i].ms);

examples/nvme/nvme_manage/nvme_manage.c

@@ -161,8 +161,10 @@ display_namespace(struct spdk_nvme_ns *ns)
 {
 	const struct spdk_nvme_ns_data		*nsdata;
 	uint32_t				i;
+	uint32_t				format_index;
 
 	nsdata = spdk_nvme_ns_get_data(ns);
+	format_index = spdk_nvme_ns_get_format_index(nsdata);
 
 	printf("Namespace ID:%d\n", spdk_nvme_ns_get_id(ns));
 
@@ -182,7 +184,7 @@ display_namespace(struct spdk_nvme_ns *ns)
 	}
 	printf("Number of LBA Formats:       %d\n", nsdata->nlbaf + 1);
 	printf("Current LBA Format:          LBA Format #%02d\n",
-	       nsdata->flbas.format);
+	       format_index);
 	for (i = 0; i <= nsdata->nlbaf; i++)
 		printf("LBA Format #%02d: Data Size: %5d  Metadata Size: %5d\n",
 		       i, 1 << nsdata->lbaf[i].lbads, nsdata->lbaf[i].ms);
@@ -467,7 +469,8 @@ ns_manage_add(struct dev *device, uint64_t ns_size, uint64_t ns_capacity, int ns
 
 	ndata->nsze = ns_size;
 	ndata->ncap = ns_capacity;
-	ndata->flbas.format = ns_lbasize;
+	ndata->flbas.format = ns_lbasize & 0xF;
+	ndata->flbas.msb_format = (ns_lbasize >> 4) & 0x3;
 	if (SPDK_NVME_FMT_NVM_PROTECTION_DISABLE != ns_dps_type) {
 		ndata->dps.pit = ns_dps_type;
 		ndata->dps.md_start = ns_dps_location;
@@ -501,11 +504,12 @@ nvme_manage_format(struct dev *device, int ns_id, int ses, int pi, int pil, int
 	int ret = 0;
 	struct spdk_nvme_format format = {};
 
-	format.lbaf	= lbaf;
+	format.lbaf	= lbaf & 0xF;
 	format.ms	= ms;
 	format.pi	= pi;
 	format.pil	= pil;
 	format.ses	= ses;
+	format.lbafu	= (lbaf >> 4) & 0x3;
 	ret = spdk_nvme_ctrlr_format(device->ctrlr, ns_id, &format);
 	if (ret) {
 		fprintf(stdout, "nvme format: Failed\n");

include/spdk/nvme.h

@@ -2851,6 +2851,18 @@ enum spdk_nvme_pi_type spdk_nvme_ns_get_pi_type(struct spdk_nvme_ns *ns);
  */
 uint32_t spdk_nvme_ns_get_md_size(struct spdk_nvme_ns *ns);
 
+/**
+ * Get the format index of the given namespace.
+ *
+ * This function is thread safe and can be called at any point while the controller
+ * is attached to the SPDK NVMe driver.
+ *
+ * \param nsdata pointer to the NVMe namespace data.
+ *
+ * \return the format index of the given namespace.
+ */
+uint32_t spdk_nvme_ns_get_format_index(const struct spdk_nvme_ns_data *nsdata);
+
 /**
  * Check whether if the namespace can support extended LBA when end-to-end data
  * protection enabled.

include/spdk/nvme_spec.h

@@ -2569,9 +2569,12 @@ struct spdk_nvme_ns_data {
 
 	/** formatted lba size */
 	struct {
-		uint8_t		format    : 4;
+		/** LSB for Format index */
-		uint8_t		extended  : 1;
+		uint8_t		format     : 4;
-		uint8_t		reserved2 : 3;
+		uint8_t		extended   : 1;
+		/** MSB for Format index, to be ignored if nlbaf <= 16 */
+		uint8_t		msb_format : 2;
+		uint8_t		reserved2  : 1;
 	} flbas;
 
 	/** metadata capabilities */
@@ -2772,9 +2775,7 @@ struct spdk_nvme_ns_data {
 		uint32_t	rp	  : 2;
 
 		uint32_t	reserved6 : 6;
-	} lbaf[16];
+	} lbaf[64];
-
-	uint8_t			reserved6[192];
 
 	uint8_t			vendor_specific[3712];
 };
@@ -2835,9 +2836,7 @@ struct spdk_nvme_zns_ns_data {
 		uint64_t	zdes : 8;
 
 		uint64_t	reserved15 : 56;
-	} lbafe[16];
+	} lbafe[64];
-
-	uint8_t			reserved3072[768];
 
 	uint8_t			vendor_specific[256];
 };
@@ -3545,13 +3544,21 @@ enum spdk_nvme_metadata_setting {
 	SPDK_NVME_FMT_NVM_METADATA_TRANSFER_AS_LBA	= 0x1,
 };
 
+/* Format - Command Dword 10 */
 struct spdk_nvme_format {
+	/* LBA format lower (LSB 4 bits of format index), also called lbafl in 2.0 spec */
 	uint32_t	lbaf		: 4;
+	/* Metadata settings, also called mset in 2.0 spec */
 	uint32_t	ms		: 1;
+	/* Protection information */
 	uint32_t	pi		: 3;
+	/* Protection information location */
 	uint32_t	pil		: 1;
+	/* Secure erase settings */
 	uint32_t	ses		: 3;
-	uint32_t	reserved	: 20;
+	/* LBA format upper (MSB 2 bits of format index) */
+	uint32_t	lbafu		: 2;
+	uint32_t	reserved	: 18;
 };
 SPDK_STATIC_ASSERT(sizeof(struct spdk_nvme_format) == 4, "Incorrect size");
 

lib/nvme/nvme_ns.c

@@ -21,15 +21,17 @@ void
 nvme_ns_set_identify_data(struct spdk_nvme_ns *ns)
 {
 	struct spdk_nvme_ns_data	*nsdata;
+	uint32_t			format_index;
 
 	nsdata = _nvme_ns_get_data(ns);
 
 	ns->flags = 0x0000;
+	format_index = spdk_nvme_ns_get_format_index(nsdata);
 
-	ns->sector_size = 1 << nsdata->lbaf[nsdata->flbas.format].lbads;
+	ns->sector_size = 1 << nsdata->lbaf[format_index].lbads;
 	ns->extended_lba_size = ns->sector_size;
 
-	ns->md_size = nsdata->lbaf[nsdata->flbas.format].ms;
+	ns->md_size = nsdata->lbaf[format_index].ms;
 	if (nsdata->flbas.extended) {
 		ns->flags |= SPDK_NVME_NS_EXTENDED_LBA_SUPPORTED;
 		ns->extended_lba_size += ns->md_size;
@@ -80,7 +82,7 @@ nvme_ns_set_identify_data(struct spdk_nvme_ns *ns)
 	}
 
 	ns->pi_type = SPDK_NVME_FMT_NVM_PROTECTION_DISABLE;
-	if (nsdata->lbaf[nsdata->flbas.format].ms && nsdata->dps.pit) {
+	if (nsdata->lbaf[format_index].ms && nsdata->dps.pit) {
 		ns->flags |= SPDK_NVME_NS_DPS_PI_SUPPORTED;
 		ns->pi_type = nsdata->dps.pit;
 	}
@@ -318,6 +320,16 @@ spdk_nvme_ns_get_md_size(struct spdk_nvme_ns *ns)
 	return ns->md_size;
 }
 
+uint32_t
+spdk_nvme_ns_get_format_index(const struct spdk_nvme_ns_data *nsdata)
+{
+	if (nsdata->nlbaf < 16) {
+		return nsdata->flbas.format;
+	} else {
+		return ((nsdata->flbas.msb_format << 4) + nsdata->flbas.format);
+	}
+}
+
 const struct spdk_nvme_ns_data *
 spdk_nvme_ns_get_data(struct spdk_nvme_ns *ns)
 {

lib/nvme/nvme_zns.c

@@ -17,8 +17,11 @@ spdk_nvme_zns_ns_get_zone_size_sectors(struct spdk_nvme_ns *ns)
 {
 	const struct spdk_nvme_zns_ns_data *nsdata_zns = spdk_nvme_zns_ns_get_data(ns);
 	const struct spdk_nvme_ns_data *nsdata = spdk_nvme_ns_get_data(ns);
+	uint32_t format_index;
 
-	return nsdata_zns->lbafe[nsdata->flbas.format].zsze;
+	format_index = spdk_nvme_ns_get_format_index(nsdata);
+
+	return nsdata_zns->lbafe[format_index].zsze;
 }
 
 uint64_t

lib/nvme/spdk_nvme.map

@@ -136,6 +136,7 @@
 	spdk_nvme_ns_get_size;
 	spdk_nvme_ns_get_pi_type;
 	spdk_nvme_ns_get_md_size;
+	spdk_nvme_ns_get_format_index;
 	spdk_nvme_ns_supports_extended_lba;
 	spdk_nvme_ns_supports_compare;
 	spdk_nvme_ns_get_dealloc_logical_block_read_value;

lib/nvmf/ctrlr.c

@@ -2618,7 +2618,7 @@ spdk_nvmf_ctrlr_identify_ns(struct spdk_nvmf_ctrlr *ctrlr,
 {
 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
 	struct spdk_nvmf_ns *ns;
-	uint32_t max_num_blocks;
+	uint32_t max_num_blocks, format_index;
 	enum spdk_nvme_ana_state ana_state;
 
 	ns = _nvmf_subsystem_get_ns_safe(subsystem, cmd->nsid, rsp);
@@ -2629,9 +2629,12 @@ spdk_nvmf_ctrlr_identify_ns(struct spdk_nvmf_ctrlr *ctrlr,
 	nvmf_bdev_ctrlr_identify_ns(ns, nsdata, ctrlr->dif_insert_or_strip);
 
 	assert(ctrlr->admin_qpair);
+
+	format_index = spdk_nvme_ns_get_format_index(nsdata);
+
 	/* Due to bug in the Linux kernel NVMe driver we have to set noiob no larger than mdts */
 	max_num_blocks = ctrlr->admin_qpair->transport->opts.max_io_size /
-			 (1U << nsdata->lbaf[nsdata->flbas.format].lbads);
+			 (1U << nsdata->lbaf[format_index].lbads);
 	if (nsdata->noiob > max_num_blocks) {
 		nsdata->noiob = max_num_blocks;
 	}

lib/nvmf/ctrlr_bdev.c

@@ -129,6 +129,7 @@ nvmf_bdev_ctrlr_identify_ns(struct spdk_nvmf_ns *ns, struct spdk_nvme_ns_data *n
 	nsdata->nuse = num_blocks;
 	nsdata->nlbaf = 0;
 	nsdata->flbas.format = 0;
+	nsdata->flbas.msb_format = 0;
 	nsdata->nacwu = spdk_bdev_get_acwu(bdev) - 1; /* nacwu is 0-based */
 	if (!dif_insert_or_strip) {
 		nsdata->lbaf[0].ms = spdk_bdev_get_md_size(bdev);

test/unit/lib/nvme/nvme_ns.c/nvme_ns_ut.c

@@ -253,6 +253,7 @@ test_nvme_ns_set_identify_data(void)
 	ns.nsdata.nsrescap.raw = 1;
 	ns.nsdata.dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_TYPE1;
 	ns.nsdata.flbas.format = 0;
+	ns.nsdata.flbas.msb_format = 0;
 	ns.nsdata.lbaf[0].lbads = 9;
 	ns.nsdata.lbaf[0].ms = 8;
 

test/unit/lib/nvmf/ctrlr.c/ctrlr_ut.c

@@ -210,6 +210,9 @@ DEFINE_STUB(spdk_bdev_get_max_open_zones, uint32_t, (const struct spdk_bdev *bde
 DEFINE_STUB(spdk_bdev_get_zone_size, uint64_t, (const struct spdk_bdev *bdev), ZONE_SIZE);
 DEFINE_STUB(spdk_bdev_is_zoned, bool, (const struct spdk_bdev *bdev), false);
 
+DEFINE_STUB(spdk_nvme_ns_get_format_index, uint32_t,
+	    (const struct spdk_nvme_ns_data *nsdata), 0);
+
 int
 spdk_nvmf_qpair_disconnect(struct spdk_nvmf_qpair *qpair, nvmf_qpair_disconnect_cb cb_fn, void *ctx)
 {
@@ -229,6 +232,7 @@ nvmf_bdev_ctrlr_identify_ns(struct spdk_nvmf_ns *ns, struct spdk_nvme_ns_data *n
 	nsdata->nuse = num_blocks;
 	nsdata->nlbaf = 0;
 	nsdata->flbas.format = 0;
+	nsdata->flbas.msb_format = 0;
 	nsdata->lbaf[0].lbads = spdk_u32log2(512);
 }
 

test/unit/lib/nvmf/ctrlr_bdev.c/ctrlr_bdev_ut.c

@@ -488,6 +488,7 @@ test_nvmf_bdev_ctrlr_identify_ns(void)
 	CU_ASSERT(nsdata.nuse == 10);
 	CU_ASSERT(nsdata.nlbaf == 0);
 	CU_ASSERT(nsdata.flbas.format == 0);
+	CU_ASSERT(nsdata.flbas.msb_format == 0);
 	CU_ASSERT(nsdata.nacwu == 0);
 	CU_ASSERT(nsdata.lbaf[0].lbads == spdk_u32log2(4096));
 	CU_ASSERT(nsdata.lbaf[0].ms == 512);
@@ -516,6 +517,7 @@ test_nvmf_bdev_ctrlr_identify_ns(void)
 	CU_ASSERT(nsdata.nuse == 10);
 	CU_ASSERT(nsdata.nlbaf == 0);
 	CU_ASSERT(nsdata.flbas.format == 0);
+	CU_ASSERT(nsdata.flbas.msb_format == 0);
 	CU_ASSERT(nsdata.nacwu == 0);
 	CU_ASSERT(nsdata.lbaf[0].lbads == spdk_u32log2(4096));
 	CU_ASSERT(nsdata.noiob == SPDK_BDEV_IO_NUM_CHILD_IOV);

test/unit/lib/nvmf/tcp.c/tcp_ut.c

@@ -235,6 +235,9 @@ DEFINE_STUB(spdk_bdev_get_max_open_zones, uint32_t,
 DEFINE_STUB(spdk_bdev_is_zoned, bool, (const struct spdk_bdev *bdev), false);
 DEFINE_STUB(spdk_bdev_get_zone_size, uint64_t, (const struct spdk_bdev *bdev), 0);
 
+DEFINE_STUB(spdk_nvme_ns_get_format_index, uint32_t,
+	    (const struct spdk_nvme_ns_data *nsdata), 0);
+
 struct spdk_io_channel *
 spdk_accel_get_io_channel(void)
 {
@@ -339,6 +342,7 @@ nvmf_bdev_ctrlr_identify_ns(struct spdk_nvmf_ns *ns, struct spdk_nvme_ns_data *n
 	nsdata->nuse = num_blocks;
 	nsdata->nlbaf = 0;
 	nsdata->flbas.format = 0;
+	nsdata->flbas.msb_format = 0;
 	nsdata->lbaf[0].lbads = spdk_u32log2(512);
 }