reduce lock congestion and improve SMP scalability of the SCSI/ATA stack,
preparing the ground for the coming next GEOM direct dispatch support.
Replace big per-SIM locks with bunch of smaller ones:
- per-LUN locks to protect device and peripheral drivers state;
- per-target locks to protect list of LUNs on target;
- per-bus locks to protect reference counting;
- per-send queue locks to protect queue of CCBs to be sent;
- per-done queue locks to protect queue of completed CCBs;
- remaining per-SIM locks now protect only HBA driver internals.
While holding LUN lock it is allowed (while not recommended for performance
reasons) to take SIM lock. The opposite acquisition order is forbidden.
All the other locks are leaf locks, that can be taken anywhere, but should
not be cascaded. Many functions, such as: xpt_action(), xpt_done(),
xpt_async(), xpt_create_path(), etc. are no longer require (but allow) SIM
lock to be held.
To keep compatibility and solve cases where SIM lock can't be dropped, all
xpt_async() calls in addition to xpt_done() calls are queued to completion
threads for async processing in clean environment without SIM lock held.
Instead of single CAM SWI thread, used for commands completion processing
before, use multiple (depending on number of CPUs) threads. Load balanced
between them using "hash" of the device B:T:L address.
HBA drivers that can drop SIM lock during completion processing and have
sufficient number of completion threads to efficiently scale to multiple
CPUs can use new function xpt_done_direct() to avoid extra context switch.
Make ahci(4) driver to use this mechanism depending on hardware setup.
Sponsored by: iXsystems, Inc.
MFC after: 2 months
- Remove two excessive and slow register reads from isp_intr(). Instead
of rereading value every time, assume that registers contain what we have
written there.
- Avoid sequential search through 4096 array elements when looking for
command tag. Use hash of lists to store active tags separately from free
ones and so greatly speedup the searches.
Reviewed by: mjacob
driver.
This tells consumers up the stack the maximum I/O size that the
controller can handle.
The I/O size is bounded by the number of scatter/gather segments
the controller can handle and the page size. For an amd64 system,
it works out to around 5MB.
Reviewed by: mjacob
MFC after: 3 days
Sponsored by: Spectra Logic
command register. The lazy BAR allocation code in FreeBSD sometimes
disables this bit when it detects a range conflict, and will re-enable
it on demand when a driver allocates the BAR. Thus, the bit is no longer
a reliable indication of capability, and should not be checked. This
results in the elimination of a lot of code from drivers, and also gives
the opportunity to simplify a lot of drivers to use a helper API to set
the busmaster enable bit.
This changes fixes some recent reports of disk controllers and their
associated drives/enclosures disappearing during boot.
Submitted by: jhb
Reviewed by: jfv, marius, achadd, achim
MFC after: 1 day
a mailbox command and which registers to copy back in when
the command completes, the bits being set need to not only
specify what bits you want to add from the default from the
table but also what bits you want *subtract* (mask) from the
default from the table.
A failing ISP2200 command pointed this out.
Much appreciation to: marius, who persisted and narrowed down what
the failure delta was, and shamed me into actually fixing it.
MFC after: 1 week
Stop abusing xpt_periph in random plases that really have no periph related
to CCB, for example, bus scanning. NULL value is fine in such cases and it
is correctly logged in debug messages as "noperiph". If at some point we
need some real XPT periphs (alike to pmpX now), quite likely they will be
per-bus, and not a single global instance as xpt_periph now.
might have been enabled for them- now that we use all 32 bits of handle.
Fast Posting doesn't pass the full 32 bits.
Noticed by: Bugs in NetBSD. Only a NetBSD user might actually still use such old hardware.
MFC after: 1 week
every architecture's busdma_machdep.c. It is done by unifying the
bus_dmamap_load_buffer() routines so that they may be called from MI
code. The MD busdma is then given a chance to do any final processing
in the complete() callback.
The cam changes unify the bus_dmamap_load* handling in cam drivers.
The arm and mips implementations are updated to track virtual
addresses for sync(). Previously this was done in a type specific
way. Now it is done in a generic way by recording the list of
virtuals in the map.
Submitted by: jeff (sponsored by EMC/Isilon)
Reviewed by: kan (previous version), scottl,
mjacob (isp(4), no objections for target mode changes)
Discussed with: ian (arm changes)
Tested by: marius (sparc64), mips (jmallet), isci(4) on x86 (jharris),
amd64 (Fabian Keil <freebsd-listen@fabiankeil.de>)
CCB at a time outstanding reliable. It's not there yet, but this
is the direction to go in so might as well commit. So far,
multiple at a time CCBs work (see ISP_INTERNAL_TARGET test mode),
but it fails if there are more downstream than the SIM wants
to handle and SRR is sort of confused when this happens, plus
it is not entirely quite clear what one does if a CCB/CTIO fails
and you have more in flight (that don't fail, say) and more queued
up at the SIM level that haven't been started yet.
Some of this is driven because there apparently is no flow control
to requeue XPT_CONTINUE_IO requests like there are for XPT_SCSI_IO
requests. It is also more driven in that the few target mode
periph drivers there are are not really set up for handling pushback-
heck most of them don't even check for errors (and what would they
really do with them anyway? It's the initiator's problem, really....).
The data transfer arithmetic has been worked over again to handle
multiple outstanding commands, so you have a notion of what's been
moved already as well as what's currently in flight. It turns that
this led to uncovering a REPORT_LUNS bug in the ISP_INTERNAL_TARGET
code which was sending back 24 bytes of rpl data instead of the
specified 16. What happened furthermore here is that sending back
16 bytes and reporting an overrun of 8 bytes made the initiator
(running FC-Tape aware f/w) mad enough to request, and keep
requesting, another FCP response (I guess it didn't like the answer
so kept asking for it again).
Sponsored by: Spectralogic
MFC after: 1 month
a tinderbox myself and caught the error.
Change to isp_send_cmd needs a final ecmd argument.
Sponsored by: Spectralogic
MFC after: 1 month
X-MFC: 238869
MISC CHANGES
Add a new async event- ISP_TARGET_NOTIFY_ACK, that will guarantee
eventual delivery of a NOTIFY ACK. This is tons better than just
ignoring the return from isp_notify_ack and hoping for the best.
Clean up the lower level lun enable code to be a bit more sensible.
Fix a botch in isp_endcmd which was messing up the sense data.
Fix notify ack for SRR to use a sensible error code in the case
of a reject.
Clean up and make clear what kind of firmware we've loaded and
what capabilities it has.
-----------
FULL (252 byte) SENSE DATA
In CTIOs for the ISP, there's only a limimted amount of space
to load SENSE DATA for associated CHECK CONDITIONS (24 or 26
bytes). This makes it difficult to send full SENSE DATA that can
be up to 252 bytes.
Implement MODE 2 responses which have us build the FCP Response
in system memory which the ISP will put onto the wire directly.
On the initiator side, the same problem occurs in that a command
status response only has a limited amount of space for SENSE DATA.
This data is supplemented by status continuation responses that
the ISP pushes onto the response queue after the status response.
We now pull them all together so that full sense data can be
returned to the periph driver.
This is supported on 23XX, 24XX and 25XX cards.
This is also preparation for doing >16 byte CDBs.
-----------
FC TAPE
Implement full FC-TAPE on both initiator and target mode side. This
capability is driven by firmware loaded, board type, board NVRAM
settings, or hint configuration options to enable or disable. This
is supported for 23XX, 24XX and 25XX cards.
On the initiator side, we pretty much just have to generate a command
reference number for each command we send out. This is FCP-4 compliant
in that we do this per ITL nexus to generate the allowed 1 thru 255
CRN.
In order to support the target side of FC-TAPE, we now pay attention
to more of the PRLI word 3 parameters which will tell us whether
an initiator wants confirmed responses. While we're at it, we'll
pay attention to the initiator view too and report it.
On sending back CTIOs, we will notice whether the initiator wants
confirmed responses and we'll set up flags to do so.
If a response or data frame is lost the initiator sends us an SRR
(Sequence Retransmit Request) ELS which shows up as an SRR notify
and all outstanding CTIOs are nuked with SRR Received status. The
SRR notify contains the offset that the initiator wants us to restart
the data transfer from or to retransmit the response frame.
If the ISP driver still has the CCB around for which the data segment
or response applies, it will retransmit.
However, we typically don't know about a lost data frame until we
send the FCP Response and the initiator totes up counters for data
moved and notices missing segments. In this case we've already
completed the data CCBs already and sent themn back up to the periph
driver. Because there's no really clean mechanism yet in CAM to
handle this, a hack has been put into place to complete the CTIO
CCB with the CAM_MESSAGE_RECV status which will have a MODIFY DATA
POINTER extended message in it. The internal ISP target groks this
and ctl(8) will be modified to deal with this as well.
At any rate, the data is retransmitted and an an FCP response is
sent. The whole point here is to successfully complete a command
so that you don't have to depend on ULP (SCSI) to have to recover,
which in the case of tape is not really possible (hence the name
FC-TAPE).
Sponsored by: Spectralogic
MFC after: 1 month
not by some hint setting. Do more preparations for FC-Tape.
Clean up resource counting for 24XX or later chipsets so
we find out after EXEC_FIRMWARE what is actually supported.
Set target mode exchange count based upon whether or not
we are supporting simultaneous target/initiator mode. Clean
up some old (pre-24XX) xfwoption and zfwoption issues.
Sponsored by: Spectralogic
MFC after: 3 days
and crosschecks against firmware documentation. We now check and report
FC firmware attributes and at least are now prepared for the upper 48 bits
of f/w attributes (which are probably for the 8100 or later cards). This
involed changing how inbits and outbits are calculated for varios commands,
hopefully clearer and cleaner. This also caused me to clean up the actual
mailbox register usage. Finally, we are now unconditionally using a CRN
for initiator mode.
A longstanding issue with the 2400/2500 is that they do *not* support
a "Prefer PTP followed by loop", which explains why enabling that
caused the f/w to crash.
A slightly more invasive change is to let the firmware load entirely
drive whether multi_id support is enabled or not.
Sponsored by: Spectralogic
MFC after: 1 week
Make the default role NONE if target mode is selected. This
allows ctl(8) to switch to/from target mode via knob settings.
If we default to role 'none', this causes a reset of the
24XX f/w which then causes initiators to wake up and notice
when we come online.
Reviewed by: kdm
MFC after: 2 weeks
Sponsored by: Spectralogic
- in destroy_lun_state() assert hold == 1 instead of 0, as it should
receive hold taken by the create_lun_state() or get_lun_statep() before;
- fix hold count leak inside rls_lun_statep() that also fired above assert;
- in destroy_lun_state() use SIM bus number instead of SIM path id for
ISP_GET_PC_ADDR(), as it was before r196008;
- make isp_disable_lun() to set status in CCB;
- make isp_target_mark_aborted() set status into the proper CCB.
Reviewed by: mjacob
Sponsored by: iXsystems, inc.
MFC after: 1 month
is actually broken, or needs a BIOS upgrade for 64 bit loads, but this uncovered
a couple of misplaced opcode definitions and some missing continual mbox command
cases, so might as well update them here.
- fix other errors introduced when committing r226436
- add 'function' to a sentence where it makes sense
Submitted by: delphij
Submitted by: dougb
Submitted by: jhb
Approved by: dougb
Approved by: jhb
Zero any sense not transferred by the device as the SCSI specification
mandates that any untransferred data should be assumed to be zero.
Reviewed by: ken
CAM.
Desriptor sense is a new sense data format that originated in SPC-3. Among
other things, it allows for an 8-byte info field, which is necessary to
pass back block numbers larger than 4 bytes.
This change adds a number of new functions to scsi_all.c (and therefore
libcam) that abstract out most access to sense data.
This includes a bump of CAM_VERSION, because the CCB ABI has changed.
Userland programs that use the CAM pass(4) driver will need to be
recompiled.
camcontrol.c: Change uses of scsi_extract_sense() to use
scsi_extract_sense_len().
Use scsi_get_sks() instead of accessing sense key specific
data directly.
scsi_modes: Update the control mode page to the latest version (SPC-4).
scsi_cmds.c,
scsi_target.c: Change references to struct scsi_sense_data to struct
scsi_sense_data_fixed. This should be changed to allow the
user to specify fixed or descriptor sense, and then use
scsi_set_sense_data() to build the sense data.
ps3cdrom.c: Use scsi_set_sense_data() instead of setting sense data
manually.
cam_periph.c: Use scsi_extract_sense_len() instead of using
scsi_extract_sense() or accessing sense data directly.
cam_ccb.h: Bump the CAM_VERSION from 0x15 to 0x16. The change of
struct scsi_sense_data from 32 to 252 bytes changes the
size of struct ccb_scsiio, but not the size of union ccb.
So the version must be bumped to prevent structure
mis-matches.
scsi_all.h: Lots of updated SCSI sense data and other structures.
Add function prototypes for the new sense data functions.
Take out the inline implementation of scsi_extract_sense().
It is now too large to put in a header file.
Add macros to calculate whether fields are present and
filled in fixed and descriptor sense data
scsi_all.c: In scsi_op_desc(), allow the user to pass in NULL inquiry
data, and we'll assume a direct access device in that case.
Changed the SCSI RESERVED sense key name and description
to COMPLETED, as it is now defined in the spec.
Change the error recovery action for a number of read errors
to prevent lots of retries when the drive has said that the
block isn't accessible. This speeds up reconstruction of
the block by any RAID software running on top of the drive
(e.g. ZFS).
In scsi_sense_desc(), allow for invalid sense key numbers.
This allows calling this routine without checking the input
values first.
Change scsi_error_action() to use scsi_extract_sense_len(),
and handle things when invalid asc/ascq values are
encountered.
Add a new routine, scsi_desc_iterate(), that will call the
supplied function for every descriptor in descriptor format
sense data.
Add scsi_set_sense_data(), and scsi_set_sense_data_va(),
which build descriptor and fixed format sense data. They
currently default to fixed format sense data.
Add a number of scsi_get_*() functions, which get different
types of sense data fields from either fixed or descriptor
format sense data, if the data is present.
Add a number of scsi_*_sbuf() functions, which print
formatted versions of various sense data fields. These
functions work for either fixed or descriptor sense.
Add a number of scsi_sense_*_sbuf() functions, which have a
standard calling interface and print the indicated field.
These functions take descriptors only.
Add scsi_sense_desc_sbuf(), which will print a formatted
version of the given sense descriptor.
Pull out a majority of the scsi_sense_sbuf() function and
put it into scsi_sense_only_sbuf(). This allows callers
that don't use struct ccb_scsiio to easily utilize the
printing routines. Revamp that function to handle
descriptor sense and use the new sense fetching and
printing routines.
Move scsi_extract_sense() into scsi_all.c, and implement it
in terms of the new function, scsi_extract_sense_len().
The _len() version takes a length (which should be the
sense length - residual) and can indicate which fields are
present and valid in the sense data.
Add a couple of new scsi_get_*() routines to get the sense
key, asc, and ascq only.
mly.c: Rename struct scsi_sense_data to struct
scsi_sense_data_fixed.
sbp_targ.c: Use the new sense fetching routines to get sense data
instead of accessing it directly.
sbp.c: Change the firewire/SCSI sense data transformation code to
use struct scsi_sense_data_fixed instead of struct
scsi_sense_data. This should be changed later to use
scsi_set_sense_data().
ciss.c: Calculate the sense residual properly. Use
scsi_get_sense_key() to fetch the sense key.
mps_sas.c,
mpt_cam.c: Set the sense residual properly.
iir.c: Use scsi_set_sense_data() instead of building sense data by
hand.
iscsi_subr.c: Use scsi_extract_sense_len() instead of grabbing sense data
directly.
umass.c: Use scsi_set_sense_data() to build sense data.
Grab the sense key using scsi_get_sense_key().
Calculate the sense residual properly.
isp_freebsd.h: Use scsi_get_*() routines to grab asc, ascq, and sense key
values.
Calculate and set the sense residual.
MFC after: 3 days
Sponsored by: Spectra Logic Corporation
We also revive loop down freezes. We also externaliz within isp
isp_prt_endcmd so something outside the core module can print
something about a command completing. Also some work in progress to
assist in handling timed out commands better.
Partially Sponsored by: Panasas
Approved by: re (kib)
MFC after: 1 month
once. Use taskqueues to do the actual work.
Fix an offset line.
Fix isp_prt so that prints from just one buffer, which makes it
appear cleanly cleanly in logs on SMP systems.
Approved by: re (kib)
MFC after: 1 month
- Allocate coherent DMA memory for the request/response queue area and
and the FC scratch area.
These changes allow isp(4) to work properly on sparc64 with usage of the
IOMMU streaming buffers enabled.
Approved by: mjacob
MFC after: 2 weeks
