[PATCH v7 0/6] libata-scsi: multi-LUN ATAPI device support

[Phil Pemberton]

Some ATAPI devices expose more than one logical unit behind a single ATA
target: Panasonic/COMPAQ PD/CD combo drives (LUN 0 = CD-ROM, LUN 1 =
PD), and Nakamichi CD changers (one LUN per disc slot, up to 7).

libata has historically hard-coded shost->max_lun = 1, so the SCSI
layer never scans past LUN 0 on any ATA-attached device.  This series
lifts that restriction for ATAPI devices gated by BLIST_FORCELUN.

Changes since v6
================

Addressing review from Hannes Reinecke (v6 review) and automated
analysis (Sashiko):

Patch 1: MODULE_PARM_DESC clarified to describe the parameter as a
count of LUNs (not a maximum LUN number), and the range corrected from
"1..7" to "1..8" (eight LUN slots, values 0..7).

Patch 2:
  - Drop the nr_luns companion field.  The ATAPI_MAX_LUN constant (8)
    is the correct fixed bound for all iterations; nr_luns adds
    complexity without benefit since sdev[] slots not populated are
    always NULL.  Use ATAPI_MAX_LUN throughout (Hannes Reinecke).
  - Fix ata_port_detach UAF window: clear dev->sdev[lun] to NULL before
    the spin_unlock, then pass the saved pointer to scsi_remove_device()
    (Hannes Reinecke).
  - ata_scsi_sdev_destroy: trigger ATA-level detach only when all LUN
    slots are NULL, not only on LUN 0 destruction.
  - ZPODD: iterate all LUN slots in zpodd_enable_run_wake(),
    zpodd_post_poweron(), and zpodd_wake_dev() (Hannes Reinecke).
  - ata_scsi_dev_rescan: snapshot all LUN sdevs under the spinlock
    before releasing it; release remaining refs on early exit paths.

Patch 3:
  - atapi_xlat(): always clear CDB byte 1 bits 7:5 first, then set them
    only for non-zero LUNs.  This avoids silently zeroing those bits on
    every LUN-0 command regardless of LUN, and separates the "zero for
    LUN 0" from the "encode LUN" paths.
  - On WARN_ON_ONCE overflow, set scmd->result = DID_ERROR before
    returning so the SCSI layer does not treat the aborted command as
    a success.

Patch 4:
  - Move the BLIST_NO_LUN_1F -> pdt_1f_for_no_lun assignment from
    scsi_add_lun() to scsi_probe_and_add_lun(), immediately before the
    PDT=0x1f check.  scsi_add_lun() is called after that check, so the
    v6 placement was too late to suppress LUN 0 if it returned
    PDT=0x1f.

Patch 5:
  - Call ata_scsi_assign_ofnode() before scsi_device_put(), so the
    reference to dev->sdev[0] is still held during the OF node
    assignment.

Patches 3/6, 4/6, and 5/6 carry Reviewed-by from Hannes Reinecke
(unchanged from v6).

Series structure
================

  1/6  ata: libata-scsi: add atapi_max_lun module parameter
  2/6  ata: libata-scsi: convert dev->sdev to per-LUN array
  3/6  ata: libata-scsi: route non-zero LUN commands for multi-LUN ATAPI
  4/6  scsi: add BLIST_NO_LUN_1F blacklist flag
  5/6  ata: libata-scsi: probe additional LUNs for multi-LUN ATAPI devices
  6/6  scsi: scsi_devinfo: add COMPAQ PD-1 multi-LUN ATAPI device quirk

Testing
=======

Tested on real hardware (Panasonic/COMPAQ LF-1195C on Intel ICH5 PATA):
  [x] Boot with CD inserted: sr0 attaches, mount and read files
  [x] Boot with PD inserted: sda attaches at correct capacity
      (1298496 x 512 B = 634 MiB)
  [x] All seven LUNs scanned (atapi_max_lun=7); LUNs 2..6 correctly
      report PDT 0x1f and are silently skipped
  [x] Single-LUN ATAPI CD-ROM (LITE-ON iHAS124): no regression,
      only LUN 0 scanned

Known limitations
=================

Media-change events are not propagated across LUNs of a SINGLELUN
multi-LUN device.  The SCSI layer's UA handling is per-sdev.  On the
PD/CD combo, swapping media and then accessing the other LUN may return
stale capacity until a manual rescan:

  echo 1 > /sys/class/scsi_device/H:0:0:1/device/rescan

A follow-up patch to propagate media-change events to sibling LUNs is
deferred to keep this series focused on the LUN-scanning core.

Phil Pemberton (6):
  ata: libata-scsi: add atapi_max_lun module parameter
  ata: libata-scsi: convert dev->sdev to per-LUN array
  ata: libata-scsi: route non-zero LUN commands for multi-LUN ATAPI
  scsi: add BLIST_NO_LUN_1F blacklist flag
  ata: libata-scsi: probe additional LUNs for multi-LUN ATAPI devices
  scsi: scsi_devinfo: add COMPAQ PD-1 multi-LUN ATAPI device quirk

 drivers/ata/libata-acpi.c   |   9 +-
 drivers/ata/libata-core.c   |  16 ++-
 drivers/ata/libata-scsi.c   | 226 ++++++++++++++++++++++++------------
 drivers/ata/libata-zpodd.c  |  27 ++++-
 drivers/ata/libata.h        |   1 +
 drivers/scsi/scsi_devinfo.c |   2 +
 drivers/scsi/scsi_scan.c    |   3 +
 include/linux/libata.h      |  11 +-
 include/scsi/scsi_devinfo.h |   6 +-
 9 files changed, 210 insertions(+), 91 deletions(-)

base-commit: a3f75e5e6b023958c92ad03fa2e68e047b6169c4
--
2.43.0