kernel/linux.git/drivers/ata/libata-eh.c, branch v6.3

ata: scsi: rename flag ATA_QCFLAG_FAILED to ATA_QCFLAG_EH

2023-01-04T04:36:26+00:00

The name ATA_QCFLAG_FAILED is misleading since it does not mean that a QC completed in error, or that it didn't complete at all. It means that libata decided to schedule EH for the QC, so the QC is now owned by the libata error handler (EH). The normal execution path is responsible for not accessing a QC owned by EH. libata core enforces the rule by returning NULL from ata_qc_from_tag() for QCs owned by EH. It is quite easy to mistake that a QC marked with ATA_QCFLAG_FAILED was an error. However, a QC that was actually an error is instead indicated by having qc->err_mask set. E.g. when we have a NCQ error, we abort all QCs, which currently will mark all QCs as ATA_QCFLAG_FAILED. However, it will only be a single QC that is an error (i.e. has qc->err_mask set). Rename ATA_QCFLAG_FAILED to ATA_QCFLAG_EH to more clearly highlight that this flag simply means that a QC is now owned by EH. This new name will not mislead to think that the QC was an error (which is instead indicated by having qc->err_mask set). This also makes it more obvious that the EH code skips all QCs that do not have ATA_QCFLAG_EH set (rather than ATA_QCFLAG_FAILED), since the EH code should simply only care about QCs that are owned by EH itself. Signed-off-by: Niklas Cassel Reviewed-by: John Garry Signed-off-by: Damien Le Moal

ata: libata-eh: Cleanup ata_scsi_cmd_error_handler()

2023-01-04T04:25:28+00:00

If ap->ops->error_handler is NULL just return. This patch also fixes some comment style issue. Signed-off-by: Wenchao Hao Reviewed-by: Niklas Cassel Signed-off-by: Damien Le Moal

ata: libata: fix commands incorrectly not getting retried during NCQ error

2022-11-19T00:41:52+00:00

A NCQ error means that the device has aborted processing of all active commands. To get the single NCQ command that caused the NCQ error, host software has to read the NCQ error log, which also takes the device out of error state. When the device encounters a NCQ error, we receive an error interrupt from the HBA, and call ata_do_link_abort() to mark all outstanding commands on the link as ATA_QCFLAG_FAILED (which means that these commands are owned by libata EH), and then call ata_qc_complete() on them. ata_qc_complete() will call fill_result_tf() for all commands marked as ATA_QCFLAG_FAILED. The taskfile is simply the latest status/error as seen from the device's perspective. The taskfile will have ATA_ERR set in the status field and ATA_ABORTED set in the error field. When we fill the current taskfile values for all outstanding commands, that means that qc->result_tf will have ATA_ERR set for all commands owned by libata EH. When ata_eh_link_autopsy() later analyzes all commands owned by libata EH, it will call ata_eh_analyze_tf(), which will check if qc->result_tf has ATA_ERR set, if it does, it will set qc->err_mask (which marks the command as an error). When ata_eh_finish() later calls __ata_qc_complete() on all commands owned by libata EH, it will call qc->complete_fn() (ata_scsi_qc_complete()), ata_scsi_qc_complete() will call ata_gen_ata_sense() to generate sense data if qc->err_mask is set. This means that we will generate sense data for commands that should not have any sense data set. Having sense data set for the non-failed commands will cause SCSI to finish these commands instead of retrying them. While this incorrect behavior has existed for a long time, this first became a problem once we started reading the correct taskfile register in commit 4ba09d202657 ("ata: libahci: read correct status and error field for NCQ commands"). Before this commit, NCQ commands would read the taskfile values received from the last non-NCQ command completion, which most likely did not have ATA_ERR set, since the last non-NCQ command was most likely not an error. Fix this by changing ata_eh_analyze_ncq_error() to mark all non-failed commands as ATA_QCFLAG_RETRY, and change the loop in ata_eh_link_autopsy() to skip commands marked as ATA_QCFLAG_RETRY. While at it, make sure that we clear ATA_ERR and any error bits for all commands except the actual command that caused the NCQ error, so that no other libata code will be able to misinterpret these commands as errors. Fixes: 4ba09d202657 ("ata: libahci: read correct status and error field for NCQ commands") Signed-off-by: Niklas Cassel Signed-off-by: Damien Le Moal

ata: make use of ata_port_is_frozen() helper

2022-10-18T04:53:27+00:00

Clean up the code by making use of the newly introduced ata_port_is_frozen() helper function. Signed-off-by: Niklas Cassel Signed-off-by: Damien Le Moal

ata: libata: fetch sense data for ATA devices supporting sense reporting

2022-10-17T02:31:52+00:00

Currently, the sense data reporting feature set is enabled for all ATA devices which supports the feature set (ata_id_has_sense_reporting()), see ata_dev_config_sense_reporting(). However, even if sense data reporting is enabled, and the device indicates that sense data is available, the sense data is only fetched for ATA ZAC devices. For regular ATA devices, the available sense data is never fetched, it is simply ignored. Instead, libata will use the ERROR + STATUS fields and map them to a very generic and reduced set of sense data, see ata_gen_ata_sense() and ata_to_sense_error(). When sense data reporting was first implemented, regular ATA devices did fetch the sense data from the device. However, this was restricted to only ATA ZAC devices in commit ca156e006add ("libata: don't request sense data on !ZAC ATA devices"). With recent changes related to sense data and NCQ autosense, we want to, once again, fetch the sense data for all ATA devices supporting sense reporting. ata_gen_ata_sense() should only be used for devices that don't support the sense data reporting feature set. hopefully the features will be more robust this time around. It is not just ZAC, many new ATA features, e.g. Command Duration Limits, relies on working NCQ autosense and sense data. Therefore, it is not really an option to avoid fetching the sense data forever. If we encounter a device that is misbehaving because the sense data is actually fetched, then that device should be quirked such that it never enables the sense data reporting feature set in the first place, since such a device is obviously not compliant with the specification. The order in which we will try to add sense data to a scsi_cmnd: 1) NCQ autosense (if supported) - ata_eh_analyze_ncq_error() 2) REQUEST SENSE DATA EXT (if supported) - ata_eh_request_sense() 3) error + status field translation - ata_gen_ata_sense(), called by ata_scsi_qc_complete() if neither 1) or 2) is supported. Signed-off-by: Niklas Cassel Signed-off-by: Damien Le Moal

ata: libata: only set sense valid flag if sense data is valid

2022-10-17T02:31:52+00:00

While this shouldn't be needed if all devices that claim that they support NCQ autosense (ata_id_has_ncq_autosense()) and/or the sense data reporting feature (ata_id_has_sense_reporting()), actually supported those features. However, there might be some old ATA devices that either have these bits set, even when they don't support those features, or they simply return malformed data when using those features. These devices should be quirked, but in order to try to minimize the impact for the users of these such devices, it was suggested by Damien Le Moal that it might be a good idea to sanity check the sense data received from the device. If the sense data looks bogus, then the sense data is never added to the scsi_cmnd command. Introduce a new function, ata_scsi_sense_is_valid(), and use it in all places where sense data is received from the device. Suggested-by: Damien Le Moal Signed-off-by: Niklas Cassel Signed-off-by: Damien Le Moal

ata: libata: clarify when ata_eh_request_sense() will be called

2022-10-17T02:31:52+00:00

ata_eh_request_sense() returns early when flag ATA_QCFLAG_SENSE_VALID is set. However, since the call to ata_eh_request_sense() is guarded by a ATA_SENSE bit conditional, the logical conclusion for the reader is that all checks are performed at the call site. Highlight the fact that the sense data will not be fetched if flag ATA_QCFLAG_SENSE_VALID is already set by adding an additional check to the existing guarding conditional. No functional change. Additionally, add a comment explaining that ata_eh_analyze_tf() will only fetch the sense data if: -It was a non-NCQ command that failed, or -It was a NCQ command that failed, but the sense data was not included in the NCQ command error log (i.e. NCQ autosense is not supported). Signed-off-by: Niklas Cassel Signed-off-by: Damien Le Moal

Merge tag 'ata-6.1-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/dlemoal/libata

2022-10-07T17:48:49+00:00

Pull ata updates from Damien Le Moal: - Print the timeout value for internal command failures due to a timeout (from Tomas) - Improve parameter names in ata_dev_set_feature() to clarify this function use (from Niklas) - Improve the ahci driver low power mode setting initialization to allow more flexibility for the user (from Rafael) - Several patches to remove redundant variables in libata-core, libata-eh and the pata_macio driver and to fix typos in comments (from Jinpeng, Shaomin, Ye) - Some code simplifications and macro renaming (for clarity) in various functions of libata-core (from me) - Add a missing check for a potential failure of sata_scr_read() in sata_print_link_status() (from Li) - Cleanup of libata Kconfig PATA_PLATFORM and PATA_OF_PLATFORM options (from Lukas) - Cleanups of ata dt-bindings and improvements of libahci_platform, ahci and libahci code (from Serge) - New driver for Synopsys AHCI SATA controllers, based of the generic ahci code (from Serge). One compilation warning fix is added for this driver (from me) - Several fixes to macros used to discover a drive capabilities to be consistent with the ACS specifications (from Niklas) - A couple of simplifcations to some libata functions, removing unnecessary arguments (from Niklas) - An improvements to libata-eh code to avoid unnecessary link reset when revalidating a drive after a failed command. In practice, this extra, unneeded reset, reset does not cause any arm beyond slightly slowing down error recovery (from Niklas) * tag 'ata-6.1-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/dlemoal/libata: (45 commits) ata: libata-eh: avoid needless hard reset when revalidating link ata: libata: drop superfluous ata_eh_analyze_tf() parameter ata: libata: drop superfluous ata_eh_request_sense() parameter ata: fix ata_id_has_dipm() ata: fix ata_id_has_ncq_autosense() ata: fix ata_id_has_devslp() ata: fix ata_id_sense_reporting_enabled() and ata_id_has_sense_reporting() ata: libata-eh: Remove the unneeded result variable ata: ahci_st: Enable compile test ata: ahci_st: Fix compilation warning MAINTAINERS: Add maintainers for DWC AHCI SATA driver ata: ahci-dwc: Add Baikal-T1 AHCI SATA interface support ata: ahci-dwc: Add platform-specific quirks support dt-bindings: ata: ahci: Add Baikal-T1 AHCI SATA controller DT schema ata: ahci: Add DWC AHCI SATA controller support ata: libahci_platform: Add function returning a clock-handle by id dt-bindings: ata: ahci: Add DWC AHCI SATA controller DT schema ata: ahci: Introduce firmware-specific caps initialization ata: ahci: Convert __ahci_port_base to accepting hpriv as arguments ata: libahci: Don't read AHCI version twice in the save-config method ...

ata: libata-eh: avoid needless hard reset when revalidating link

2022-09-24T06:31:00+00:00

Performing a revalidation on a AHCI controller supporting LPM, while using a lpm mode of e.g. med_power_with_dip (hipm + dipm) or medium_power (hipm), will currently always lead to a hard reset. The expected behavior is that a hard reset is only performed when revalidate fails, because the properties of the drive has changed. A revalidate performed after e.g. a NCQ error, or such a simple thing as disabling write-caching (hdparm -W 0 /dev/sda), should succeed on the first try (and should therefore not cause the link to be reset). This unwarranted hard reset happens because ata_phys_link_offline() returns true for a link that is in deep sleep. Thus the call to ata_phys_link_offline() in ata_eh_revalidate_and_attach() will cause the revalidation to fail, which causes ata_eh_handle_dev_fail() to be called, which will set ehc->i.action |= ATA_EH_RESET, such that the link is reset before retrying revalidation. When the link is reset, the link is reestablished, so when ata_eh_revalidate_and_attach() is called the second time, directly after the link has been reset, ata_phys_link_offline() will return false, and the revalidation will succeed. Looking at "8.3.1.3 HBA Initiated" in the AHCI 1.3.1 specification, it is clear the when host software writes a new command to memory, by setting a bit in the PxCI/PxSACT HBA port registers, the HBA will automatically bring back the link before sending out the Command FIS. However, simply reading a SCR (like ata_phys_link_offline() does), will not cause the HBA to automatically bring back the link. As long as hipm is enabled, the HBA will put an idle link into deep sleep. Avoid this needless hard reset on revalidation by temporarily disabling hipm, by setting the LPM mode to ATA_LPM_MAX_POWER. After revalidation is complete, ata_eh_recover() will restore the link policy by setting the LPM mode to ap->target_lpm_policy. Signed-off-by: Niklas Cassel Signed-off-by: Damien Le Moal

ata: libata: drop superfluous ata_eh_analyze_tf() parameter

2022-09-21T02:18:36+00:00

The parameter can easily be derived from struct ata_queued_cmd. Signed-off-by: Niklas Cassel Signed-off-by: Damien Le Moal