]> git.kernelconcepts.de Git - karo-tx-linux.git/commit
scsi: lpfc: NVME Initiator: Base modifications
authorJames Smart <jsmart2021@gmail.com>
Sun, 12 Feb 2017 21:52:30 +0000 (13:52 -0800)
committerMartin K. Petersen <martin.petersen@oracle.com>
Wed, 22 Feb 2017 23:41:43 +0000 (18:41 -0500)
commit895427bd012ce5814fc9888c7c0ee9de44761833
tree307a6d5500f676e5df31b8120a3c5986d0636eba
parent1d9d5a9879ad493ee7cf75987df1f365c61fefe5
scsi: lpfc: NVME Initiator: Base modifications

NVME Initiator: Base modifications

This patch adds base modifications for NVME initiator support.

The base modifications consist of:
- Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as
  rings as well) as implementation now widely varies between the two.
- Addition of configuration modes:
   SCSI initiator only; NVME initiator only; NVME target only; and
   SCSI and NVME initiator.
   The configuration mode drives overall adapter configuration,
   offloads enabled, and resource splits.
   NVME support is only available on SLI-4 devices and newer fw.
- Implements the following based on configuration mode:
  - Exchange resources are split by protocol; Obviously, if only
     1 mode, then no split occurs. Default is 50/50. module attribute
     allows tuning.
  - Pools and config parameters are separated per-protocol
  - Each protocol has it's own set of queues, but share interrupt
    vectors.
     SCSI:
       SLI3 devices have few queues and the original style of queue
         allocation remains.
       SLI4 devices piggy back on an "io-channel" concept that
         eventually needs to merge with scsi-mq/blk-mq support (it is
 underway).  For now, the paradigm continues as it existed
 prior. io channel allocates N msix and N WQs (N=4 default)
 and either round robins or uses cpu # modulo N for scheduling.
 A bunch of module parameters allow the configuration to be
 tuned.
     NVME (initiator):
       Allocates an msix per cpu (or whatever pci_alloc_irq_vectors
         gets)
       Allocates a WQ per cpu, and maps the WQs to msix on a WQ #
         modulo msix vector count basis.
       Module parameters exist to cap/control the config if desired.
  - Each protocol has its own buffer and dma pools.

I apologize for the size of the patch.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
----
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
22 files changed:
drivers/scsi/lpfc/lpfc.h
drivers/scsi/lpfc/lpfc_attr.c
drivers/scsi/lpfc/lpfc_bsg.c
drivers/scsi/lpfc/lpfc_crtn.h
drivers/scsi/lpfc/lpfc_debugfs.c
drivers/scsi/lpfc/lpfc_debugfs.h
drivers/scsi/lpfc/lpfc_els.c
drivers/scsi/lpfc/lpfc_hbadisc.c
drivers/scsi/lpfc/lpfc_hw.h
drivers/scsi/lpfc/lpfc_hw4.h
drivers/scsi/lpfc/lpfc_init.c
drivers/scsi/lpfc/lpfc_logmsg.h
drivers/scsi/lpfc/lpfc_mbox.c
drivers/scsi/lpfc/lpfc_mem.c
drivers/scsi/lpfc/lpfc_nportdisc.c
drivers/scsi/lpfc/lpfc_nvme.h [new file with mode: 0644]
drivers/scsi/lpfc/lpfc_scsi.c
drivers/scsi/lpfc/lpfc_scsi.h
drivers/scsi/lpfc/lpfc_sli.c
drivers/scsi/lpfc/lpfc_sli.h
drivers/scsi/lpfc/lpfc_sli4.h
drivers/scsi/lpfc/lpfc_vport.c