Why won't my server see or boot a 4Kn SAS drive?

1. 1.List block devices with their logical and physical sector sizes. If the drive is not in the list at all, the kernel never received it from the controller, and the sector format is not yet the question.

   lsblk -o NAME,SIZE,LOG-SEC,PHY-SEC,MODEL

   ReadingLOG-SEC 4096 with PHY-SEC 4096 is 4Kn, 512 with 4096 is 512e, 512 with 512 is 512n. lsblk -t prints the same sector columns with alignment.
2. 2.If the drive is missing, rule out power and the drive itself before blaming 4Kn. In one Unraid thread, 8TB 4Kn SAS drives behind two LSI 9300-8i cards and expanders were not detected, and the cause was the 3.3V power-disable pins, not the sector format. In a Dell thread, a used 4Kn SAS drive that read about 1.8TB on three controllers was identified as an EMC Unity array part, which a Dell moderator said was not supported in PowerEdge.
3. 3.Read the sizes from the block layer.

   blockdev --getss --getpbsz --getsize64 /dev/sdX

   Reading--getss is the logical sector size, --getpbsz the physical sector size, --getsize64 the capacity in bytes. Ignore --getbsz: it is a kernel-internal block size and does not describe the device.
4. 4.The same two values are in sysfs.

   cat /sys/block/sdX/queue/logical_block_size /sys/block/sdX/queue/physical_block_size

   ReadingThe first value is the smallest unit the device can address (4096 on 4Kn, 512 on 512e). The second is the smallest unit it writes atomically (4096 on both 512e and 4Kn).
5. 5.Ask the drive directly with READ CAPACITY(16). Use --long: the 10-byte default caps block addresses at 2**32 - 2.

   sg_readcap --long /dev/sdX

   ReadingA block size of 4096 bytes is a 4Kn drive; 512 is 512e or 512n. The output also gives the number of logical blocks per physical block. Block count times block size should match the capacity on the label.
6. 6.fdisk shows the same sizes in its header.

   fdisk -l /dev/sdX

   ReadingThe sector size line shows logical and physical sector size as the kernel reports them. Leave -b off so the header shows what the kernel reports; the man page reserves it for very old kernels, disk images, or overriding the kernel's default.
7. 7.On Windows, query a volume on the disk. Replace C: with that volume's letter.

   fsutil fsinfo sectorinfo C:

   ReadingLogicalBytesPerSector 4096 with PhysicalBytesPerSectorForAtomicity 4096 is 4Kn; 512 with 4096 is 512e. On older Windows, fsutil fsinfo ntfsinfo C: gives 'Bytes Per Sector' and 'Bytes per Physical Sector' (4096/4096 is 4K native, 512/4096 is 512E, 512/512 is 512n). 'Not Supported' in the physical field means the storage driver did not answer the query.

1. 1.Compare the capacity the host reports with the label.

   sg_readcap --long /dev/sdX

   ReadingA figure about one-eighth of the label, as in a Dell thread where a PERC H700 showed a 4TB 4Kn drive as 465.25GB, is consistent with 4096-byte blocks being counted as 512-byte blocks somewhere in the path. That is our arithmetic, not a vendor statement.
2. 2.Check whether your controller is documented for 4 KB sector drives. Dell's PERC 9 User's Guide lists the PERC H330, H730, H730P, H730P MX, H830, FD33xS and FD33xD. We found no Dell document stating 4Kn support for the H310, H700, H710 or HBA330, or for PERC 10, and no Broadcom document on 4Kn support or minimum firmware for LSI HBAs. A Dell community representative said controllers before PERC 9 do not support 4K drives and show them as Unsupported, and community members said PERC 9 was the first generation to support 4Kn; those are forum statements, not Dell documents, and one H700 owner found some 4Kn models worked while others did not.
3. 3.Supermicro states that 4Kn is supported only with SAS2 and SAS3 controllers with new firmware, and in UEFI mode; it does not name the firmware versions in the FAQ we read. On an X9SRA-F it found a 4Kn drive was not supported by the onboard Intel RSTe and LSI RAID firmware.
4. 4.On a PERC 9, do not add a 4Kn drive to a virtual disk built from 512-byte drives. Dell allows 512n and 512e drives in one virtual disk, but not 512-byte and 4 KB native drives together.
5. 5.If the controller is not documented for 4Kn, there are three ways forward: a controller that is, switching the drive to 512-byte sectors where the model supports it (path p5, which erases the drive), or a drive that presents 512-byte logical sectors (512e or 512n).

1. 1.Set the system to boot in UEFI mode. Dell, Intel and Supermicro each state that 4K sector drives, or RAID volumes built from them, boot only in UEFI mode; Intel notes that the MBR is fixed to 512 bytes per sector. Supermicro's first remedy for an install that boots to a blinking cursor is UEFI boot with an EFI partition present.
2. 2.This matches the threads. An Ubuntu install to an 8TB 4Kn drive in BIOS mode completed without error, then booted to a blinking underscore, while UEFI installs worked. A FreeBSD install to a 4Kn SAS drive behind an onboard LSI SAS3008 looped on reboot in CSM mode and booted once the boot environment was set to EFI only.
3. 3.Check the controller's option ROM setting as well as the boot mode. Supermicro says the Avago legacy option ROM supports only 512-byte sector drives. On an X10DRL-C with an onboard LSI 3108, a RAID 5 of 4Kn drives installed under UEFI but had no boot entry until Onboard SAS OPROM was changed from Legacy to EFI, under Advanced >> PCIe/PCI/PnP Configuration >> Onboard SAS OPROM.
4. 4.The Proxmox VE installer refuses legacy BIOS installs to 4Kn disks; its developers state that legacy BIOS boot on 4Kn disks does not work. Start the installer in UEFI mode instead.
5. 5.If the system cannot boot UEFI, Supermicro's other remedies are to use non-4Kn drives for the OS or to put the OS on a SATA DOM.
6. 6.For virtual machines on vSphere 9.0 and later, Broadcom states that guests cannot boot from 4Kn virtual disks under BIOS firmware; select EFI firmware during installation.

1. 1.On ESXi, list each device's block sizes and format type.

   esxcli storage core device capacity list

   ReadingPhysical Blocksize, Logical Blocksize and Format Type per device. '4K' is a 4Kn device; '512e' and '512n' are as named. '4K SWE' means ESX is emulating 512-byte sectors on a local 4Kn HDD, which is how ESX 9.0 presents one; Broadcom says such a disk supports 512n VMDKs only.
2. 2.Check your ESXi version against Broadcom's two articles, which describe support before 9.0 differently. KB 327012 says vSphere and vSAN 6.7/7.x support 4K native hard disk drives, with VMFS 6 required, no RDM on 4Kn HDDs and 512n VMDKs presented to guests, and that 6.0 and earlier were not designed for 4Kn or 512e direct-attached drives. KB 453249 says 4Kn devices are not natively supported before ESXi 9.0, that ESXi supports only local 4Kn SAS and SATA HDDs, that 4Kn SSD, NVMe and 4Kn RDM are not supported before 9.0, and that its 512-byte emulation on 4Kn HDDs can raise write latency; its resolution is to reformat to 512 or upgrade to ESXi 9. Both allow 4Kn hard drives on pre-9.0 ESXi in some form and both rule out RDM on them; KB 327012 covers only direct-attached HDDs and says nothing about 4Kn flash. We print both and do not choose between them on native support.
3. 3.If the disks are 4Kn SSDs behind a RAID controller, Intel states that RAID arrays built from 4Kn SSDs will not be detected by ESXi, and its remedy is 512n or 512e SSDs or hard disks.
4. 4.If a datastore shows in the ESXi Host Client but cannot be selected while deploying vCenter Server, Broadcom traces it to a 4Kn disk, and esxcli shows Format Type 4K for it. Its workaround deletes the datastore and reformats the disk to 512e, which loses everything on that datastore, so move its contents first.
5. 5.On Windows, check the version. Microsoft supports 4Kn on Windows 8 and Windows Server 2012 and later, including installing to and booting from a 4Kn disk, and 512e from Windows 7 SP1 and Server 2008 R2 SP1 (earlier with KB 982018 or KB 2553708). Windows XP, Server 2003 and Server 2003 R2 support neither. Sector sizes other than 512 or 4K are not supported.
6. 6.Microsoft lists two known issues on drives whose logical sector is not 512 bytes: Windows system image backup may fail with 0x8078002A, and creating a VHD on a 4Kn drive in Server 2008 R2 fails with 'Incorrect function'.

1. 1.Find out whether the model can change sector size at all. Seagate says support is per drive: check the data sheet, the product manual or the Features Supported list in SeaChest's identify output. On Windows the device is -d PD1 rather than /dev/sgN. We found no WD or Toshiba document on changing sector size.

   SeaChest_Format -d /dev/sgN -i

   ReadingRead-only. If Features Supported, the data sheet and the manual do not show that the drive can change sector size, stop here and use p2's other options.
2. 2.This erases all data on the drive, and Seagate warns that any interruption while formatting may render the drive inoperable. Copy off anything you need, Seagate does not recommend running it over USB, stop smartd and any out-of-band management first, and do not interrupt it: a SATA drive may take up to 5 hours to return. The change may affect every LUN or namespace on the device. Seagate's example sets 4096; for a host that needs 512-byte sectors, set 512.
   Destructive. This command destroys what is on the device and cannot be undone. Confirm the device path twice and have a backup of anything you still need before running it.

   SeaChest_Format -d /dev/sgN --setSectorSize 512 --confirm this-will-erase-data-and-may-render-the-drive-inoperable

   ReadingWhen it returns, confirm the new size with sg_readcap --long or lsblk. In one TrueNAS thread, Exos X16 drives switched from 512e to 4Kn with SeaChest_Lite reported success, then two were inaccessible after reboot with read and CRC errors; no recovery was confirmed.
3. 3.This destroys all data on the drive. On SAS and other SCSI disks that offer 512 as an alternate logical block size, sg_format can set it. A format can take days on disks over 10TB. You can abort with Ctrl-C during the 15-second countdown; once the format starts, Ctrl-C only ends the tool and the format continues, and a reset or power cycle will probably stop it partway and leave the disk needing another format. Leave it until it finishes.
   Destructive. This command destroys what is on the device and cannot be undone. Confirm the device path twice and have a backup of anything you still need before running it.

   sg_format --format --size=512 /dev/sgN

   ReadingConfirm afterwards with sg_readcap --long. The 520-byte sector page linked below covers sg_format, its progress output and its failure modes in detail.

We found no Broadcom document stating 4Kn support or a minimum firmware for SAS2008 cards. One serverbuilds.net reply claims all SAS2008 HBAs can use 4Kn drives but cannot boot from them; the asker never confirmed it, so treat it as a forum claim and check with lsblk and sg_readcap --long once the drive is connected.

OpenZFS sets each vdev's ashift (2^ashift is the smallest IO) from the largest physical sector size its drives report when the vdev is created (ZFS on Linux also accepts ashift= at creation), and it cannot be changed afterwards; OpenZFS also notes that many devices misreport. We have not read OpenZFS text on what happens when a 4Kn disk replaces a member of a vdev created with 512-byte sectors, so we do not state an outcome. Check each drive's PHY-SEC and LOG-SEC before you create or extend a vdev.

Microsoft notes that writes smaller than 4 KB to a 512e drive trigger a read-modify-write, because its physical sector is 4 KB while it addresses in 512-byte units. Western Digital describes 512e as existing so new drives work with operating systems that do not support 4Kn. We hold no measurement of the cost on any model.

We found no WD or Toshiba document on changing sector size; Western Digital's knowledge base defines the formats but gives no procedure. On SAS drives, sg_format --format --size=512 works only where the drive offers 512 as a logical block size, and it destroys all data.