mbox series

[v4,0/8] ATA PASS-THROUGH sense data fixes

Message ID 20240701195758.1045917-1-ipylypiv@google.com
Headers show
Series ATA PASS-THROUGH sense data fixes | expand

Message

Igor Pylypiv July 1, 2024, 7:57 p.m. UTC 
This patch series is fixing a few ATA PASS-THROUGH issues:
1. Not reporting "ATA Status Return sense data descriptor" / "Fixed format
   sense data" when ATA_QCFLAG_SENSE_VALID is set.
2. Generating "fake" sk/asc/ascq based on ATA status/error registers when
   ATA_QCFLAG_SENSE_VALID is set and CK_COND=1.
3. Fixed format sense data was using incorrect field offsets for ATA
   PASS-THROUGH commands.
4. Using qc->result_tf in ATA sense data generation functions without
   checking if qc->result_tf contains a valid data.

Changes since v1:
Thanks Damien and Niklas for the reviews!

- Squashed two v1 patches 2/4 and 3/4 into one patch with a different
  implementation.
- Added 'Cc: stable@vger.kernel.org' tags to patches that are fixing bugs.
- Reordered patches with the 'Cc: stable@vger.kernel.org' tag to be applied
  first in order to simplify backports to stable releases.
- Restored the buffer memset in atapi_eh_request_sense().
- Updated declaration order in v1 patch 4/4.
- Added a patch to cleanup unused ATA device id in ata_to_sense_error().
- Updated fill_result_tf() to set ATA_QCFLAG_RTF_FILLED after populating
  the result taskfile. Removed now redundant flag sets/checks from ahci.
- Updated ATA sense data generation functions to return early if result_tf
  is not filled. Added WARN_ON_ONCE checks to generate a warning when
  ATA_QCFLAG_RTF_FILLED is not set and libata needs to generate sense data.

Changes since v2:
- Moved v2 2/6 patch (fixed ATA PT offsets) to be the first one in v3.
- Removed unused variable 'sb' from ata_gen_passthru_sense().
- Removed WARN_ON_ONCE checks and added ata_dev_dbg() logs instead.
- Removed the Fixes tag from v2 4/6 patch because the patch is doing
  a cleanup and is not fixing any bugs.

Changes since v3:
- Changed "RTF" to "result TF" in the log messages to make it more clear.
  Removed capitalization and punctuation to be consistent with existing logs.
- Marked the stable tag of the v3 2/6 patch with the 4.19+ LTS version.
- Added a patch to honour the D_SENSE bit for CK_COND=1 and no error commands.
- Fixed a bug in the v3 5/6 patch which caused the qc->result_tf.flags field
  assignment to be skipped if ahci_qc_ncq_fill_rtf() was executed.
- Added a patch proposed by Niklas to improve the readability of the complex
  ATA PASS-THROUGH handling in ata_scsi_qc_complete().

Igor Pylypiv (8):
  ata: libata-scsi: Fix offsets for the fixed format sense data
  ata: libata-scsi: Do not overwrite valid sense data when CK_COND=1
  ata: libata-scsi: Honour the D_SENSE bit for CK_COND=1 and no error
  ata: libata-scsi: Remove redundant sense_buffer memsets
  ata: libata-scsi: Do not pass ATA device id to ata_to_sense_error()
  ata: libata-core: Set ATA_QCFLAG_RTF_FILLED in fill_result_tf()
  ata: libata-scsi: Check ATA_QCFLAG_RTF_FILLED before using result_tf
  ata: libata-scsi: Make ata_scsi_qc_complete() more readable

 drivers/ata/libahci.c     |  12 +--
 drivers/ata/libata-core.c |   8 ++
 drivers/ata/libata-scsi.c | 209 +++++++++++++++++++++-----------------
 3 files changed, 128 insertions(+), 101 deletions(-)