perform various operations on a controller. Specifically, for each mpt(4)
device, create a character device in devfs which accepts ioctl requests for
reading and writing configuration pages and performing RAID actions.
MFC after: 1 week
Reviewed by: scottl
abstraction as the RAID and CAM modules, making it nearly impossible
for enough initialization to be done in time for the RAID module to
know whether to attach. On top of this, no reset was being done on
the controller on attach, in violation of the spec. Additionally,
the port enable step was being deferred to the end of the attach
process, long after it should have been done to ensure reliable
operation from the controller. Fix all of these with a few hacks
to force the "attach" and "enable" steps of the core module early
on, and ensure that a reset and port enable also happens early on.
In the future, the driver needs to be refactored to eliminate the
core module abstraction, clean up withe reset/enable steps, and
defer event messages until all of the modules are available to
recieve them.
it's been printing out scary messages about "Unhanded Event Notify Frame"
that are needlessly worrisome to users. Change this warning to only print
out at an elevated debugging level.
to kproc_xxx as they actually make whole processes.
Thos makes way for us to add REAL kthread_create() and friends
that actually make theads. it turns out that most of these
calls actually end up being moved back to the thread version
when it's added. but we need to make this cosmetic change first.
I'd LOVE to do this rename in 7.0 so that we can eventually MFC the
new kthread_xxx() calls.
the mfi(4) LSI MegaSAS RAID card. Looking at the Linux driver for the
mpt(4) it should be 0x0062 and not 0x0060. Tested with an mfi card
of this device id.
Approved by: re (bmah)
Reviewed by: scottl
MFC after: 3 days
now takes a device_t to be the parent of the bus that is being created.
Most SIMs have been updated with a reasonable argument, but a few exceptions
just pass NULL for now. This argument isn't used yet and the newbus
integration likely won't be ready until after 7.0-RELEASE.
mpt.h:
Add support for reading extended configuration pages.
mpt_cam.c:
Do a top level topology scan on the SAS controller. If any SATA
device are discovered in this scan, send a passthrough FIS to set
the write cache. This is controllable through the following
tunable at boot:
hw.mpt.enable_sata_wc:
-1 = Do not configure, use the controller default
0 = Disable the write cache
1 = Enable the write cache
The default is -1. This tunable is just a hack and may be
deprecated in the future.
Turning on the write cache alleviates the write performance problems with
SATA that many people have observed. It is not recommend for those who
value data reliability! I cannot stress this strongly enough. However,
it is useful in certain circumstances, and it brings the performence in line
with what a generic SATA controller running under the FreeBSD ATA driver
provides (and the ATA driver has had the WC enabled by default for years).
use to synchornize and protect all data objects that are used for that
SIM. Drivers that are not yet MPSAFE register Giant and operate as
usual. RIght now, no drivers are MPSAFE, though a few will be changed
in the coming week as this work settles down.
The driver API has changed, so all CAM drivers will need to be recompiled.
The userland API has not changed, so tools like camcontrol do not need to
be recompiled.
attachment of new devices that arrive (and we notice them
via async Fibre Channel events). We've always had the
right thing (of sorts) happen when devices go away- this
is the corollary function that makes multipath failover
actually work.
MFC after: 2 weeks
read wasn't flagging the SYNC mode was enabled. The temp
values for offset and sync period were uint8_t, but were
being assigned and shifted from a uint32_t value.
This didn't show up in testing because a random number
of 1030 cards set a bit that says "honor BIOS negotiation",
which means this whole code path was skipped.
This should clear up at least some of the negotation
issues that have been seen.
Fix things to use the LSI-Logic Fusion Library mask and shift names for
offset and sync, no matter how awkward they are, in preference to just
plain numbers.
This allows us to play nicely on SANs when we have target mode
enabled in f/w but have neither the scsi_targbh enabled or
scsi_targ with a target enabled.
the CAM_NEW_TRAN_CODE that has been in the tree for some years now.
This first step consists solely of adding to or correcting
CAM_NEW_TRAN_CODE pieces in the kernel source tree such
that a both a GENERIC (at least on i386) and a LINT build
with CAM_NEW_TRAN_CODE as an option will compile correctly
and run (at least with some the h/w I have).
After a short settle time, the other pieces (making
CAM_NEW_TRAN_CODE the default and updating libcam
and camcontrol) will be brought in.
This will be an incompatible change in that the size of structures
related to XPT_PATH_INQ and XPT_{GET,SET}_TRAN_SETTINGS change
in both size and content. However, basic system operation and
basic system utilities work well enough with this change.
Reviewed by: freebsd-scsi and specific stakeholders
Unfortunately, the QUEUE FULL event only tells you Bus && Target.
It doesn't tell you lun. In order for the XPT_REL_SIMQ action to
work, we have to have a real lun. But which one? For now, just
iterate over MPT_MAX_LUNS.
Practically speaking, this is only going to be happening for lower
quality SAS or SATA drives behind the SAS controller, which means
only lun 0, so it's not so bad.
Helpful Reminder Nagging from: John Baldwin, Fred Whiteside
MFC after: 5 days
is to able to be called after *all* attach and enable events are done.
We establish a SYSINIT hook to call this handler. The current usage for it
is to add scsi target resources *after* all enables are done. There seems
to be some dependencies between different halves of a dual-port with respect
to target mode.
Put in more meaningful event messages for some events- in particular
QUEUE FULL events so we can see what the queue depth was when the
IOC sent us this message.
MFC after: 1 week
desired role configuration instead of existing role. This gets
us out of the mess where we configured a role of NONE (or were
LAN only, for example), but didn't continue to attach the CAM
module (because we had neither initiator nor target role
set). Unfortunately, the code that rewrites NVRAM to match
actual to desired role only works if the CAM module attaches.
MFC after: 2 weeks
mark it as timed out. Don't try and free the config
request for read_cfg_header that times out because
it's still active. Put in code for the config reply
handler that will then free up timed out requests.
Fix the FC_PRIMITIVE_SEND completion to not try
and free a command twice. Dunno how this possibly
could have been working for awhile.
MFC after: 2 weeks
out ELS buffers but *don't* hang out commands,
we hang folks on the SAN because the LSI-Logic
f/w apparently sends back BUSY or QFULL or some
darn thing.
If we add command buffers, we have to respond to
them sensibly even if we don't have any upstream
listeners (scsi_targ or scsi_targ_bh), so put in
some local command reponse stuff.
MFC after: 2 weeks
actually go write the config page. This fixes the long standing
problem about updating NVRAM on Fibre Channel cards and seems
so far to not break SPI config page writes.
Put back role setting into mpt. That is, you can set a desired role
for mpt as a hint. On the next reboot, it'll pick that up and redo
the NVRAM settings appropriately and warn you that this won't take
effect until the next reboot. This saves people the step of having
to find a BIOS utilities disk to set target and/or initiator role
for the MPT cards.
Don't enable/disable I/O space except for SAS adapters.
This fixes a problem with VMware 4.5 Workstation.
Fix an egregious bug introduced to target mode so it actually
will not panic when you first enable a lun.
Minor fixes:
Take more infor from port facts and configuration pages.
MFC after: 1 week
Clean out the abortive start to homegrown, per-mpt,
Domain Validation. This should really be done at a
higher level.
Use the PIM_SEQSCAN flag for U320- this seems to correct
cases of being unable to consistently negotiate U320 in
the cases where I'd seen this before.
Between this and other recent checkins, this driver is
pretty close to being ready for MFC.
Reviewed by: scottl, ken, scsi@
MFC after: 1 week
fixing speed negotiation.
Also fix the mpt_execute_req function to actually
match mpt_execute_req_a64. This may explain why
i386 users were having more grief.
can see the results of SPI negotiation w/o being overwhelmed
with other crap).
+ For U320 devices, check against both Settings *and* DV flags before
deciding whether we need to skip actual SPI settings for a device.
+ Go back to creating a 'physical disk' side of a raid/passthru bus that
is limited to the number of maximum physical disks. Actually, this isn't
probably *quite* right yet for one RAID volume, and if we ever end up
with finding a device that supports more than one RAID volume (not likely),
it probably won't quite be right either.
The problem here is that the creating of this 'physical' passthru sim is
just a cheap way to leverage off the CAM midlayer to do our negotiation
for us on the subentities that make up a RAID volume. It almost causes
more trouble than it is worth because we have to remember which side
we're talking to in terms of forming commands and which target ids are
real and so on. Bleah.
+ Skip trying to actually do SPI settings for the RAID volumes on the
real side of the raid/passthru bus pair- this just confuses the issue.
The underlying real physical devices will have the negotiation performed
and the Raid volume will inherit the resultant settings. At the sime time,
non-RAID devices can be on the same real bus, so *do* perform negotiations
with them.
+ At the end of doing all of the settings twiddling, *ahem*, remember to
go update the settings on the card itself (dunno how this got nuked).
At this point, negotiations *seem* to be being done (again) correctly for
both RAID volumes and their subentities. And they seem to be *mostly*
now right for other non-RAID entities on the same bus (I ended up with
3 out of 8 other disks still at narror/async- haven't the slightest
idea why yes).
Finally, negotiations on a normal bus seem to work (again).
There's still more work coming into this area, but we're in the
final stretch.
the passed target id is one of the RAID VolumeID. This result
is used to decide whether to try and do actual SPI negotiations
on the real side of the raid/passthru bus pair. The reason we
check this is that we can have both RAID volumes and real devices
on the same bus.
lost one set to a peninsula power failure last night. After
this, I can see both submembers and the raid volumes again,
but speed negotiation is still broken.
Add a mpt_raid_free_mem function to centralize the resource
reclaim and fixed a small memory leak.
Remove restriction on number of targets for systems with IM enabled-
you can have setups that have both IM volumes as well as other devices.
Fix target id selection for passthru and nonpastrhu cases.
Move complete command dumpt to MPT_PRT_DEBUG1 level so that just
setting debug level gets mostly informative albeit less verbose
dumping.
state structure. This field is only for CCBs that are associated with
actions that are occurring on the HBA (i.e., XPT_CONT_IO actions).
This way we also don't get confused when the upstream listener stalls
try and look at a CCB which has already been freed (by CAM).
+ Add boatloads of KASSERTs and *really* check out more locking
issues (to catch recursions when we actually go to real locking
in CAM soon). The KASSERTs also caught lots of other issues like
using commands that were put back on free lists, etc.
+ Target mode: role setting is derived directly from port capabilities.
There is no need to set a role any more. Some target mode resources
are allocated early on (ELS), but target command buffer allocation
is deferred until the first lun enable.
+ Fix some breakages I introduced with target mode in that some commands
are *repeating* commands. That is, the reply shows up but the command
isn't really done (we don't free it). We still need to take it off the
pending list because when we resubmit it, bad things then happen.
+ Fix more of the way that timed out commands and bus reset is done. The
actual TMF response code was being ignored.
+ For SPI, honor BIOS settings. This doesn't quite fix the problems we've
seen where we can't seem to (re)negotiate U320 on all drives but avoids
it instead by letting us honor the BIOS settings. I'm sure this is not
quite right and will have to change again soon.
There's something strange going on with async events. They seem
to be be treated differently for different Fusion implementations.
Some will really tell you when it's okay to free the request that
started them. Some won't. Very disconcerting.
This is particularily bad when the chip (FC in this case) tells you
in the reply that it's not a continuation reply, which means you
can free the request that its associated with. However, if you do
that, I've found that additional async event replies come back for
that message context after you freed it. Very Bad Things Happen.
Put in a reply register debounce. Warn about out of range context
indices. Use more MPILIB defines where possible. Replace bzero with
memset. Add tons more KASSERTS. Do a *lot* more request free list
auditting and serial number usages. Get rid of the warning about
the short IOC Facts Reply. Go back to 16 bits of context index.
Do a lot more target state auditting as well. Make a tag out
of not only the ioindex but the request index as well and worry
less about keeping a full serial number.
the error on sparc64 hadn't changed since the last checkin, pass
LINT on other platforms and mpt doesn't work on sparc64 anyway
and the tinderbox build didn't work for me in a cross build case
on my main build machine (which runs RELENG_6). Sigh. Still
need to try harder.
A) Fibre Channel Target Mode support mostly works
(SAS/SPI won't be too far behind). I'd say that
this probably works just about as well as isp(4)
does right now. Still, it and isp(4) and the whole
target mode stack need a bit of tightening.
B) The startup sequence has been changed so that
after all attaches are done, a set of enable functions
are called. The idea here is that the attaches do
whatever needs to be done *prior* to a port being
enabled and the enables do what need to be done for
enabling stuff for a port after it's been enabled.
This means that we also have events handled by their
proper handlers as we start up.
C) Conditional code that means that this driver goes
back all the way to RELENG_4 in terms of support.
D) Quite a lot of little nitty bug fixes- some discovered
by doing RELENG_4 support. We've been living under Giant
*waaaayyyyy* too long and it's made some of us (me) sloppy.
E) Some shutdown hook stuff that makes sure we don't blow
up during a reboot (like by the arrival of a new command
from an initiator).
There's been some testing and LINT checking, but not as
complete as would be liked. Regression testing with Fusion
RAID instances has not been possible. Caveat Emptor.
Sponsored by: LSI-Logic.
mpt_soft_reset more than once. And to wait for MPT_DB_STATE_READY
twice. I mean, this is crucial- give the IOC a chance to get
ready.
If mpt_reset is called to reinit things, and we succeed, make
sure to re-enable interrupts. This is what has mostly led to
system lockup after having to hard reset the chip. Also, if
we think that interrupts aren't function in mpt_cam_timeout,
for goodness sake, turn them on again.
In read_cfg_header, return distinguishing errnos so the caller
can decide what's an error. It's *not* an error to fail to
read a RAID page from a non-RAID capable device like the FC929X.
Some whitespace fixes (removing spaces from ends of lines).
not hang the system for 5 seconds. If a TMF doesn't complete within,
oh, say 500ms, that's enough.
Put in a printout to catch mpt_recover_commands being activated with
no commands.
*both* SAS and FC, not just SAS.
b) Don't tell the chip we want it to do FIFO signalling if we actually
don't set up the address where the FIFO signal is supposed to be written
(oops).
