Hello, Jeff, Jens and Albert.
This is the second take of generic NCQ completion/error-handling
patchset. Changes are...
* Generic special command helper (ata_qc_exec_special) added. Now
all internal commands run with timeout (request sense, read log,
set xfer...)
* New non-NCQ error handling to make EH paths more unified with NCQ
case and ATAPI handling less hacky. Now _all_ errors (NCQ or not)
except for internal commands are posted to EH thread and handled
there. As was in ATAPI EH, all failed qc's are finished by EH
thread.
* NCQ helpers can now deal with ATAPI errors.
This patchset is composed of the following parts.
patches #01-#03: implement ata_qc_exec_special, new EH framework
and new non-NCQ error handling.
patches #04-#08: implement NCQ helpers and convert & fix AHCI.
patches #09-#11: misc stuff
Currently this patchset only converts sata_sil, ata_piix and ahci.
All other drivers won't compile.
[ Start of patch descriptions ]
01_libata_implement-ata_qc_exec_special.patch
: implement ata_qc_exec_special()
This patch implements ata_qc_exec_special() function which
generalizes libata internal special command execution and adds
timeout. All internal commands are converted. Timeout
constants are added to libata.h and, while at it, removed
unused ATA_TMOUT_EDD.
02_libata_new-error-handling.patch
: implement new EH framework and convert non-ncq EH
This patch implements new EH framework and converts non-ncq EH
to use it. Now errors and timeouts are all handled by EH
handler inside EH thread.
* All failed commands are posted to EH by using ata_qc_error()
without qc-completion. All timed out commands are posted to
EH, too. On entry to EH handler, all active qc's are either
failed or timed out qc's, and they are all. Also it's
guaranteed that once EH is started, only EH can finish or
retry it. Normal or spurious interrupts during recovery
don't affect failed qc's.
* EH handles error and determines whether to retry or fail
qc's. It is responsible for setting error information to
notify upper layer if it's gonna fail a command. (Jeff,
this should make implementing error classes you've talked
about easier. All error handling is done inside EH and we
can just set sense data appropriately and eh-complete the
commands.)
* After EH is complete, operation resumes.
The following changes are worth noting.
* ->eng_timeout renamed to ->error_handler
* ata_eh_qc_complete(), ata_eh_qc_retry() added
* __ata_qc_complete() used to deal with resource freeing and
completing wait for special cmds. As this patch removes the
only use of __ata_qc_complete(), it's moved into
ata_qc_free() and ata_qc_free() deals only with resource
freeing. New __ata_qc_complete() is defined to be used by
ata_eh_qc_*() functions (internal use only).
* After allocated, all commands are either freed with
ata_qc_free() if it can't be issued or completed with
ata[_eh]_qc_complete(). No half-completion anymore.
* As filling upper layer error info is the reponsibility of
the recovery handler now, qc->complete_fn() doesn't need
drv_stat argument. Also, as all commands are completed
fully at once, there's no need for int return value. This
leaves very little functionality to qc->complete_fn,
currently only ATAPI inquiry result tempering. I think just
inlining that part and removing this callback will make the
code easier.
* Although it looks like a lot of changes, from a device's
point of view, nothing changes. Only software structure is
changed.
This patch is separated out only to make changes incremental.
I haven't really tested this patch alone. Please test with
the following NCQ EH patches applied.
03_libata_convert-drivers-to-new-eh.patch
: convert sata_sil and ata_piix to use new EH
This patch converts sata_sil and ata_piix to use new EH.
04_libata_implement-sactive.patch
: implement ap->sactive
This patch implements generic ap->sactive handling for NCQ
commands.
05_libata_add-drv_err-to-ata_to_sense_error.patch
: add drv_err argument to ata_to_sense_error()
During NCQ error handling, drv_stat and drv_err values are
obtained from log page 10h. This patch adds drv_err argument
to ata_to_sense_error() such that it can be used with values
obtained from log page 10h.
06_libata_implement-ncq-helpers.patch
: implement generic NCQ helpers
This patch implements generic NCQ completion/error-handling
helpers.
07_libata_convert-ahci-to-new-eh.patch
: convert ahci driver to use new NCQ helpers
This patch converts ahci driver to use new NCQ helpers.
08_libata_ahci-stop-dma-before-resetting.patch
: stop dma before reset
AHCI 1.1 mandates stopping dma before issueing COMMRESET. The
original code didn't and it resulted in occasional lockup of
the controller during EH recovery. This patch fixes the
problem.
09_libata_remove-unused-eh-functions.patch
: remove unused functions
This patch removes ata_scsi_requeue(),
ata_scsi_block_requests() and ata_scsi_unblock_requests().
10_libata_ahci-atapi.patch
: add ATAPI support to ahci
This patch adds ATAPI support to ahci driver. This currently
doesn't work. I'll write a reply to this thread to tell more
about this. However, it at least shows that NCQ ATAPI error
handling works.
11_libata_debug.patch
: debug stuff
As usual, debug messages and error triggers.
[ End of patch descriptions ]
Please let me know what you guys think.
Thanks.
--
tejun
-
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
