mjhjmtu

I have a SSD drive that uses 8K block size. On x86_64 Linux, ext4/xfs won't support blocksize larger than Linux PAGE_SIZE, which is 4K.

I have tried using LVM and mdadm RAID0 with stripping set to 8K hoping that the drive will be sent 8K sizeed IO requests, but unfortunately all IO requests are 4K only.

How can I can setup the drive in such a way that the file system is 4K based (to keep ext4/xfs happy) but the requests sent to the drive are 8K?

Or is there some code that I can write to make this happen? Perhaps play around with /dev/mapper? I'm open to ideas and suggestions.

The SSD drive I'm using is an eval unit that has not been released yet. Works under Windows, being tested on Linux. The following data is returned for Identify namespace and controller data commands

Identify: Namespace data
------------------------
Size in GiB: 4000
NSZE: Size in 8K blocks: 524288000
NCAP: Capacity in 8K blocks: 524288000
NUSE: Utilization: 524288000
NSFEAT:
 deallocate supported: 0
 namespace atomic writes supported: 0
 thin provisioning supported: 0
NLBAF: number of LBA formats: 0 (0 based)
FLBAS: formatted LBA size
 metadata at end of LBA: 0
 supported LBA format: 0
MC: metadata capabilities
 metadata in separate buffer supported: 0
 metadata in extended LBA supported: 0
DPC: end to end data protection
 prot info in last 8 bytes of metadata: 0
 prot info in first 8 bytes of metadata 0
 prot info type 3 supported: 0
 prot info type 2 supported: 0
 prot info type 1 supported: 0
 DPS: end to end data protection setting
 protection info in first 8 bytes: 0
 protection info type: 0
NMIC: namespace accessible by 2+ controllers 0
RESCAP: reservation capabilities
 NS supports exclusive all access: 0
 NS supports write exclusive all: 0
 NS supports exclusive all reg only: 0
 NS supports write exclusive reg only: 0
 NS supports exclusive access res: 0
 NS supports write exclusive res: 0
 NS supports persist through pow loss 0
FPI: format progress indicator
 FPI supported: 0
 percentage yet to be formatted: 0
NAWUN: NS write atomic unit normal: 0
NAWUPF NS write atomic unit power fail: 0
NACWU: NS atomic compare and write unit: 0
NABSN: NS atomic boundary size normal: 0
NABO: NS atomic boundary offset: 0
NABSPF NS atomic boundary size power fail: 0
NVMCAP capacity in bytes: 0
NSGUID 0x00000000000000000000000000000000
EUI64: IEEE extended UID: 0x0000000000000000
LBA format 00: 0xd0000
LBA format 01: 0x0
LBA format 02: 0x0
LBA format 03: 0x0
LBA format 04: 0x0
LBA format 05: 0x0
LBA format 06: 0x0
LBA format 07: 0x0
LBA format 08: 0x0
LBA format 09: 0x0
LBA format 10: 0x0
LBA format 11: 0x0
LBA format 12: 0x0
LBA format 13: 0x0
LBA format 14: 0x0
LBA format 15: 0x0

Identify: Controller data
-------------------------
VID: PCI vendor ID: deleted
SSVID: PCI subsystem vendor ID: deleted
SN: Serial number: 1 
MN: Model number: deleted 
FR: Firmware revision: 3C408
RAB: Recommended arbitration burst: 16
IEEE: IEEE OUI: 0x002508
CMIC: 
 Controller associated with SR-IOV: 0
 Controller associated with PCIE: 1
 NVMSS contains multiple controllers: 0
 NVMSS contains multiple PCIE ports: 0
MDTS: Max data transfer size: 5
CNTLID Controller ID: 0x1
VER: NVMe version: 0x10200
RTD3R: RTD3 resume latency: 1
RTD3E: RTD3 entry latency: 1
OAES: Optional async events supported
 NS attrib changed event supported: 0
OACS: Optional admin command support
 NS management supported: 0
 FW commit & image download supported: 0
 Format NVMD cmd supported: 0
 Security send/recv supported: 0
ABCL: Abort command limit: 3
AERL: Async event request limit: 3
FRMW: Firmware updates
 Requires reset to activate FW: 1
 Number of FW slots: 1
 FW slot 1 is read only: 0
LPA: Log page attributes
 Command effects log page supported: 0
 SMART log reported on per NS basis: 0
ELPE: Error log pages entries stored in dev: 31
NPSS: Number of power states supported: 0 (0 based)
AVSCC: Admin vendor specific cmd confg
 Uses vendor specific admin vend cmds: 0
APSTA: Autonomous power state transitions sup 0
WCTEMP Warning composite temp threshold: 343
CCTEMP Critical composite temp threshold: 383
MTFA: Max time for FW activation: 0
HMPRE: Host memory buffer preferred size: 0
HMMIN: Host memory buffer minimum size: 0
TNVMCAP Total NVM capacity: 0
UNVMCAP Total unallocated NVM capacity: 0
RPMBS: Replay protected memory block support
 Access size: 0
 Total size: 0
 Authentication method: 0
 Number of RPMB units: 0
SQES: Submission queue entry size
 Maximum entry size: 6
 Minimum entry size: 6
CQES: Completion queue entry size
 Maximum entry size: 4
 Minimum entry size: 4
NN: Number of namespaces: 1
ONCS: Optional NVM command support
 Reservations supported: 0
 Save field in set features supported: 0
 Write zeros supported: 0
 DSM supported: 1
 Write uncorrectable supported: 0
 Compare command supported: 0
FUSES: Fused operations supported: 0
FNA: Format NVM attributes
 Cryptographic erase supported: 0
 Cryptographic erase applies to all NS: 0
 Format applies to all NS: 0
VWC: Volatile write cache present: 0
AWUN: Atomic write unit normal size: 0 (0 based)
AWUPF: Atomic write unit power fail size: 0 (0 based)
NVSCC: NVM vendor specific cmd config
 Uses vendor specific NVM vend cmds: 0
ACWU: Atomic write and compare unit: 0
SGLS: SGL support
 SGL length can be larger than data buf 0
 Byte aligned contiguous buf supported 0
 SGL bit bucket supported: 0
 SGLs for for NVM command set supported 0