[PATCH 0/2] nvme: check identifiers in order of precedence

[Bryan Gurney]

After commit e2724cb9f0c40 ("nvme: fix the check for duplicate unique
identifiers"), some users are seeing a regression where namespaces on a
PCI-Express NVMe SSD supporting namespace management are failing to have
device nodes created, with the kernel reporting the error "duplicate IDs
in subsystem", despite having at least one unique identifier:

nvme0n1:nguid   : 36434430528018200025384100000001   
nvme0n2:nguid   : 36434430528018200025384100000002
nvme0n3:nguid   : 36434430528018200025384100000003   
nvme0n4:nguid   : 36434430528018200025384100000004
nvme0n5:nguid   : 36434430528018200025384100000005   

nvme0n1:eui64   : 00253898110036df
nvme0n2:eui64   : 00253898110036df
nvme0n3:eui64   : 00253898110036df
nvme0n4:eui64   : 00253898110036df
nvme0n5:eui64   : 00253898110036df

This patch set ensures that there is a unique identifier, in the order
of precedence specified by the NVMe specification (UUID, then NGUID,
then EUI64).  This avoids the issue that has been reported by customers,
on multiple drive models, where the old algorithm would impose too
stringent of a check, as many drives will report a unique NGUID value,
and a non-unique EUI64 value.

This patch set has been tested by an affected customer of Red Hat
Enterprise Linux, and was verified to have restored the creation of the
device nodes of NVMe namespaces on drives with the affected firmware.

Bryan Gurney (2):
  nvme: check duplicate unique identifiers in order
  nvme: print warning on non-unique namespace identifier

 drivers/nvme/host/core.c | 41 +++++++++++++++++++++++++++++++---------
 1 file changed, 32 insertions(+), 9 deletions(-)

-- 
2.49.0