correct spelling of FALLTHROUGH

Use mtx_init() to initialize stack-based mutexes

mtx_init() ensures the mutex' lock_object has static storage duration.
This makes it safe to re-enable WITNESS with stack-based mutexes.

OK bluhm@ miod@

Disable witness for mutexes created on the stack which allows
the ddb command 'show witness' to succeed without panicking.
Leaving witness enabled on these mutexes saves a pointer to
struct lock_type on the stack which gets clobbered resulting
in a panic in witness_ddb_display_descendants().

okay miod@

constify SCSI adapter entry points

ok krw@

Since dlg@'s 2009 scsi midlayer refactoring the 'struct scsi_generic *cmd'
member of 'struct scsi_xfer' has always been pointed at the 'struct scsi_generic
cmdstore' member of the same instance. So nuke 'cmdstore' and remove the '*'
from cmd. Take the address of cmd as required by the various casts.

No intentional functional change.

luna88k test by aoyama@, sparc64 test by jmatthew@

Identification of 2009's last *cmd use and ok jmatthew@

Rename [READ|WRITE]_BIG to [READ|WRITE]_10. Rename struct scsi_rw_big to
struct scsi_rw_10.

ok gnezdo@ jmatthew@ (who also did sparc64 compile test)

Turning on various scsi drivers' *DEBUG options reveals that this has
rarely (if ever) been done.

Fix many printf format errors to calm clang and gcc on amd64, i386,
hppa. Missing #include, complaints if 'option <blah>DEBUG' is used in
config files, etc. All in debug code.

Move remaining scsi bus initialization info from "prototype scsi link"
fields to struct scsibus_attach_args. Nuke the struct scsi_link *
(saa_sc_link) in scaibus_attach_args.

Explicitly initialize each field in scsibus_attach_args variables.

Move the adapter related items (luns, adapter, adapter_target,
adapter_buswidth, adapter_softc) from struct scsi_link to struct
scsibus_attach_args.

Additional compile tests by jmatthew@ (sparc64) and aoyam@ (luna88k).

Access adapter softc via link->bus->sb_adapter_softc.

In sparc64 autoconf access 'luns' via sb_luns and 'adapter_buswidth'
via sb_adapter_buswidth.

Removes last post-config uses of the copies of bus related
information in scsi_link.

Where we have a pointer (sc) to the relevant adapter_softc, use
&sc->sc_dev to get a device pointer rather than using the void *
whence we obtained 'sc'.

Remove some pointless casting of void * when assigning one
(scsi_link.adapter_softc) to a pointer of a particular softc type.

Nuke trailing whitespace.

No need to bzero()/memset() 'struct scsibus_attach_args' variables
immediately before initializing the only field in the struct.

Use SDEV_NO_ADAPTER_TARGET instead of -1 to indicate
the adapter is not a target on the bus.

ok dlg@ as part of a larger diff.

Simplify some goto'ing that jumps a single line or
does nothing at all.

The first one spotted as CID 1452976.

Reads good to millert@

ok mpi@

Nuke *_minphys() functions that either simply apply MAXPHYS or do
nothing at all. MAXPHYS will be applied in minphys() and nothing at
all, well, doesn't do anything.

Also remove any '#define <blah> MAXPHYS' statements used solely to
disguise MAXPHYS in said functions.

Shuffle some names around to make reading the code less headache
inducing. Rename scsi_adapter member 'scsi_minphys' to 'dev_minphys'
to reflect what it is supposed to do. Use consistent naming convention
(<dev>_minphys) for the actual device functions.

No functional change.

Drivers that implement their own *minphys() don't need to call the
system minphys(). scsi_minphys() will do that and cd/sd/st will call
scsi_minphys().

ok jmatthew@ as part of larger diff

Use a consistant idiom/format when declaring scsi_adapter structures
in drivers. Terse one liners, NULLs instead of 0's, explicitly specify
all members, etc.

Nuke #ifdef notyet blocks related to the scsi_adapter in aic.

No intentional functional change.

ok tedu@

Convert infinite sleeps to {m,t}sleep_nsec(9).

ok kn@

apply background init progress fix from r1.59 of mfii.c
(reverting most of r1.60 of mfi.c, oddly enough)

ok dlg@
tested on a SAS2008 (perc h310)

return ENOTTY rather than EINVAL to indicate an ioctl hasn't been handled

ok dlg@ deraadt@ kettenis@

If you use sys/param.h, you don't need sys/types.h

copy the mbox into the dcmd struct, not the pointer to the mbox.

makes bioctl work on mfi(4) again.

while here memset a few mboxes before using them.

bioctl breakage reported by jason george

represent the mbox layout with union instead of an array of bytes.

memcpying uint16_ts into inconsistently addressed offsets is hard
to read, and this makes future work easier to implement.

tested on mfi(4) and mfii(4)

sizes for free(); ok sthen

Initial addition of ``Patrol Read'' support in bio(4), biocto(8), and
mfi(4). Based on FreeBSD, but done without mfiutil(8).

OK deraadt@

Prevent splassert from firing during sd_flush which runs "cold"

While mfi(4) should pass SCSI transfer flags (e.g. SCSI_POLL and
SCSI_NOSLEEP) down to the management function, make it at least
use "cold" consistently for now.

ok dlg

Remove some includes include-what-you-use claims don't
have any direct symbols used. Tested for indirect use by compiling
amd64/i386/sparc64 kernels.

ok tedu@ deraadt@

properly set sgl related flags on skinny controllers. this fixes
io against physical disks on my dell perc h310s. logical disks still
work fine.

ok deraadt@ yasuoka@

Fix mfi ioctl to set drive state properly.
diff from Tsubai Masanari

test dlg jmatthew
ok dlg

Add support DIOC{G,S}CACHE to mfi(4). Cache of disk on mfi(4) will
be enabled by default. Also if the controller doesn't have any RAM,
the cache on the physical disk will be enabled.

discuss with jsg tsubai
ok dlg

another handful of bcopy -> memcpy because there is no overlap

remove uneeded proc.h includes
ok mpi@ kspillner@

mark the interrupt handler mpsafe, and give up biglock in the scsi cmd
submission paths. take biglock again when calling back into the scsi
stack.

tested on a variety of cards here, including the skinny mfis on the
firewall next to mpsafe myx.

Create a function which loads sgd in the mfi_iop_ops struct so that skinny
adapters can use "IEEE sgl".

tested dlg yasuoka
ok dlg jsg

Some reallocarray() use; review Jean-Philippe Ouellet, patrick keshishian
ok tedu

add a size argument to free. will be used soon, but for now default to 0.
after discussions with beck deraadt kettenis.

format string fixes for bus_addr_t and bus_size_t

bus_addr_t and bus_size_t are u_long everywhere

ok kettenis@

revert previous commit; didn't intend to commit those bits

Avoid printing the "nvram corrupt" message for onboard 2200s found on Sun
hardware.

ok dlg@, jmatthew@

mfi_refresh_sensors calls mfi_ioctl_vol which calls mfi_bio_getitall, which
replaces pointers in the softc and all sorts of fun stuff. this protects
the call to bio_getitall with the sc_lock the ioctl path uses, so we dont
get panics like those reported on bugs@ recently.

ok deraadt@

Add support for displaying the cache write policy of RAID volumes to bioctl(8)
and make mfi(4) pass up the necessary information. Adding support for other
RAID controllers is left as an excercise to the reader.

ok deraadt@

Include the 'state of health' field in the bbu ok sensor value.
The dell perc 6/i sets this to zero (meaning bad) when it
decides the battery doesn't have enough capacity for it to
allow write back mode, but doesn't set any of the other flags
to tell us why.

ok dlg@, tested by sthen@

enw->new in comment;
spotted by krw, missed by jsg, and forwarded From: Alexey Suslikov

fix some leaks in error paths
ok fgsch@ krw@ kettenis@

break out the battery status into excruciating detail. makes it harder for
$VENDOR to complain about a lack of diagnostic material from their crappy
mgmt tools if you're not running an operating system they support.

always attach the battery sensor if the adapter lets us query it.
newly replaced batteries dont show up immediately.

provide a sensor for the battery backup unit (bbu) on mfi boards that
support it.

now i can tell (a bit better) why io might be slow on some of my boxes.

dont need to walk alldevs during sensor creation looking for the logical
disk scsibus when we record that in mfi_attach.

also happens to get rid of a logic error that let it come out of the loop
on alldevs with a reference to something that wasnt the logical disk
scsibus and use that.

use MR_DCMD consistently. some MD_DCMD was in there somehow.

david imhoff points out that the error handling in mfi_init_ccb gets the
index to which ccb to free wrong.

this takes the logic from mpi/mpii/mfii which pulls the ready ccbs off
the free list rather than rely on the current index from the for loop
that we failed inside of.

thanks to david for finding this out.

Make sure we don't sleep on autoconf.

ok mikeb

add support for "physical devices" on skinny controllers.

these controllers let you specify disks that should be accessed directly
rather than be part of logical volumes. apparently you can plug in things
like tape drives too.

anyway, physical devices are exposed via a separate scsibus attached to
mfi.

ok mikeb@

fix mfi_poll by having it do delays between loops, and breaking on
timeouts. how embarrassment.

implement mfi_exec for doing sync executions of commands in process
contexts and cut mfi_mgmt over to it.

rework mfi_poll so it will do all the completions on a ccb including
calling the done handler.

this makes it consistently complete scsi io with all the proper error
checking.

stop passing mfi_softc around as a member of the ccb, consistently give it
as the first argument to functions. like mfii.

scrub ccbs when we're about to use them, not when we put them back
on the free list. ccbs can be reused, so we dont want old state screwing up
new commands.

move the dma syncs for the ccb data into ccb_done, so we can do it once
rather than put the code everywhere.

fix up the dma sync for the command frame in mfi_done.

use scsi_get_link to find out the name of a child device for bioctl rather
than snoop scsi commands and copy it in an io path.

no functional change, but the code has one less XXX now.

expose the max number of logical volumes the hardware supports as the width
of its scsibus. only advertise 1 lun on logical volumes instead of letting
the midlayer fix it up to 8.

give every target on the bus max_cmds openings. iopools means they will
properly share access to them.

this in particular is useful on skinny controllers which only advertise 31
command slots. if you have 16 volumes, theyll only get 1 opening each with
the old maths. this way round the ones that are busy will share the slots.

tested on a perc5 with two volumes and hard workloads.

print the boards name (eg, perc 5/i) according to the firmware like mfii.

ok deraadt@

move knowledge of the location of the inbound doorbell out of code in
transition_firmware into a member of the iop structures.

ok mikeb@ haesbaert@

get rid of the last foo_lo and foo_hi bits i could find in the hardware
structures and use htole64 instead of htole32(addr >> 32); htole32(addr);

gets rid of "handy" stack variables to get the dva to 64bits.

sprinkle some more byte swaps for things that should have it.

tested on a perc5 (xscale)

do appropriate bus_dmamapy_syncs around the pcq/completion ring.

tested on a perc5 (xscale)

replace uint32_ts for hi and lo addresses in the mfi_init_ hardware
descriptors with single uint64_ts. theyre not some weird middle
endian thing. this makes the code more readable.

add some missing htole32s to the code while here.

tested on a perc5 (xscale).

the producer/consumer values are used by the device, which is little
endian. do appropriate byteswapping for when the host cpu uses them.

tested on a perc5 (xscale) in an amd64.

add support for skinny variants.

from Alexey Suslikov
reviewed by marco

Backout a bunch of my SCSI commits from c2k11. At least one of these
is causing problems when trying to boot sparc64 from an isp(4).

Verified to fix the sparc64/isp(4) regression by krw@; ok deraadt@

First batch of converting SCSI HBAs from setting saa_targets and
saa_luns instead of adapter_buswidth and luns in the prototype link.

ok dlg@, miod@

add all 64 bit dma plumbing but stick with 32 for now by default.

ok deraadt

make sure the upper 32 bits are always written to instead of assuming it
is 0.

ok deraadt.

kill some stray spaces

rename wiat channel to something less than 8 chars.

prompted by deraadt

use dma_alloc for transient management commands.

prodded and ok deraadt

cut mfi over to using iopools.

this gets rid of NO_CCB, provides better scheduling of io between busy
logical volumes on the same controller, and makes the ioctl and sensor
paths more reliable.

this is the least intrusive version of this change i could come up with. it
only took 4 or 5 goes...

ok marco@
deraadt@ says this is a good time for this change.

no use setting cmd in mfi_ioctl_setstate if you never use it.

Use SSD_ERRCODE_CURRENT instead of magic 0x70.

ok dlg@ matthew@

the scsi completion code thinks that if you dont get MFI_STAT_OK
back from the chip on a command then the command completely failed.
MFI_STAT_SCSI_DONE_WITH_ERROR really means the command completed
fine, but there's some sense data too. this tweaks the handling to
be more appropriate, as per the linux and solaris drivers.

timed out waiting for beck@
putting this in cos its obviously more correct than the current code.

SCSI HBA drivers are supposed to get their SCSI CDBs from xs->cmd, not
xs->cmdstore.

ok krw@ a while back (contingent upon testing), and deraadt@ says this
has been in snapshots for weeks now.

Change scsibus(4)'s scsi_link array to an SLIST to save memory on
sparsely populated buses.

ok dlg@, krw@

Use SLIST for the ccb free list.

ok krw, discontent grunts from marco.

Remove all adapter-specific 'struct scsi_device's. They are never used. First
step in elminating 'struct scsi_device' entirely.

Spotted and initial diff from matthew@.

ok matthew@ dlg@ deraadt@ marco@ miod@

dont pass the dev_t from the scsi device drivers into the midlayer for
ioctl requests, and dont pass the proc pointers around for any ioctl
requests in scsi land at all. neither were used, so trim the fat.

ok krw@ marco@

New scsi code seems to be stable. Pluck previously identified
low-hanging splbio/splx pairs that are no longer needed and see if
this reveals any hidden scsi flaws.

ok dlg@

Use a temporary variable for now to sidestep -Wbounded checking when
copying vendor[8]/product[16]/revision[4] out of struct scsi_inquiry_data
together with one memcopy.

ok krw

use BUS_DMA_ZERO on alloc instead of bzeroing after.

ok dlg@, marco@

grrr var names in protos

sprinkle some bus_dmamap_sync around frame access.
prompted by oga

Change the scsi_cmd function member of scsi_adapter from int to
void. Use XS_NO_CCB error in the scsi command (xs) to report the
NO_CCB condition. Eliminates all SUCCESSFULLY_QUEUED and COMPLETE
confusion and untangles the midlayer from the adapter a bit more.

Eyes and some fixes by miod@

There may be some compile issues on little used (i.e. I don't have
any) drivers but the change is mechanical and thus easy to remedy.

ok dlg@

Zap all setting of ITSDONE in drivers that don't look at it. Nobody
else cares so it's just noise. Drivers that actually look at ITSDONE
are unchanged.

ok marco@ (for his files) dlg@ beck@

rename ccb_xs to ccb_cookie and change its type from a scsi_xfer * to a
void *. this lets us pass things other than scsi_xfers to ccb completion
handlers.

protect the ccb free queue with a mutex rather than splbio

change M_WAITOK --> M_WAITOK|M_CANFAIL

with input from marco@

ok marco@ krw@

this was ok looooooooong time ago but lost in my mailbox

Add READ(16) & WRITE(16) commands.
ok krw

MFI_D_IOCTL should be commented out. No binary change.

scrub more fields in the ccb when returning them to the free list after
theyve been used, in particular the mfi header flags which has a bit
that specifies if a command should be completed via the interrupt path.

if we use a ccb during boot we set that bit, but it isnt necessarily
cleared by things that use it later on. this means a ccb we expected to
complete via an interrupt never actually generates an interrupt or appears
in the reply queue. this obviously stalls the io.

spello

when we steal the SYNC CACHE command in mfi_scsi_cmd we still have to
mark the xs as done and call scsi_done for it rather than just returning
COMPLETE.

fix a small race in mfi_mgmt between the checking of a ccbs completion and
the sleep waiting for the completion. it is possible to get the interrupt
completing the command just before the tsleep, which will never get a
wakeup because the interrupt with the wakeup has already happened.

Extend the scsi_adapter minphys() callback to take a struct scsi_link *
as additional argument. This will allow intermediate layers between
scsi devices such as sd and scsi host adapters to take appropriate
action if necessary.

Add support for next generation megaraid sas raid controllers.
From Jim Giannoules <gortag@gmail.com>

Generalize function to obtain all details to do bio stuff. This is in
preparation to start showing unsused disks which will require all this
information to do the math.

While at it fix a size of hotspare volume bug.

Fix bug that when there are no volumes but there are hot spares it would
no show the hotspares.

Fix output when reporting on multivolume disk groups.

Reported by mitja@muzenic.net

Move some include stuff around

Space

Fix PR5645
ok dlg
tested beck and others

Initialize 'ssc' to NULL since we test it later. Shouldn't ever be
NULL at the point of test unless the device list is confused, but
be consistantly paranoid if we're going to be paranoid.

Return NO_CCB and not TRY_AGAIN_LATER when running out of ccb's.
Eliminates 'not queued' errors reported on the mailing lists.

Tested by Michael (belenus at bsdmail dot de).

ok marco@

add support for the mfi controllers with the ppc iops, aka, LSI SAS1078 or
Dell PERC6.

committed off a machine booting and rooting off a perc6. tested on a perc5
by marco@
ok marco@

Set the return value instead of returning right away so that the DMA
memory is actually synched.

ok dlg@

sync dma mem back to the cpu on poll failure.

from mickey via pr5714

despatch -> dispatch

disparate spellink fix from mickey

More scsi_done() at SPLBIO.

M_ZERO changes

ok dlg@

rework sensor tasks to use the kernels generic workq rather than a special
kernel thread of its own. the api has changed (which will be fixed in the
manpage shortly) so all the users of sensor tasks that i can find have
been fixed too.

noone tested, so its going in to force people to run with it.
"put it in" deraadt@

sprinkle some SMALL_KERNEL to permit bio/softraid to run on boot media
ok marco@ then suggested deraadt@

split userland & kernel struct sensor/sensordev so that the addition
of new fields in the future is less disruptive. This is done similar
to how struct proc is handled for ps(1). ok jmc (man page changes)
tested fkr simon, and more suggestions from millert

bus_space does byte swapping to cope with host endianness for us, so remove
an unnecessary htole32().

Fix subtle x ? y : z bug when calling mfi_create_sgl. This worked because
during runtime the bug condition does not happen.

ok krw joris

Remove double semicolons.

From Pierre Riteau <pierre dot riteau at free dot fr>.

OK marco@.

adapt to new two-level sensor api; Constantine A. Murenin

give scsi controllers a real attach args to fill in when attaching scsibus.

ok miod@ marco@ deraadt@

Add sensor thread for individual sd devices.

tested dlg
ok dlg deraadt

when determining the size of the volumes and disks we want how many
bytes each of them has, not the number of kilobytes. this solves the
"why am i dividing by 2" question. the answer is "you dont, you multiply
by 512".

make dmesg printing look a little bit nicer.

Factor for disk size is /2. Odd but tested on all mfis I have.

Everytime one forgets an argument in a printf-like function call, God kills
a kitten. Commiters, please think of the kittens when working on code.

Make a BGI (Back Ground Init) a scrub instead of rebuild.

Add hot spare support.

Add blinking for disks.

Move a int i; to top of function so that it compiles on non c99 compilers.
Pointed out by miod@

The enclosure id modifier didn't reflect reality so move it back to whatever
the raid card uses.

And add time and percentage to builds and check consistencies.

Add last bit of physcal disk stuff.
Remove bogus memset.
Hotspares still don't work.

Add physical device and read config structures.
Add initial part of BIOCDISK

commited with unused variable

Remove some debug cruft.

Add first half of physical disk data for bio.

Add support for raid 50 and 10 in bio.

Add mail box to mfi_mgmt to handle sub commands more gracefully.
Remove some debug crud that snuck in the TEST_UNIT_READY case.
Finish other half of BIOCVOL primitive.
Add several structures to fill out the BIOCVOL primitive.

Add first half of BIOCVOL primitive.

Do a giant flush of disk and raid-card memory when SYNCHONIZE_CACHE is called.

pretty up some debug statements.

Add bio alarm function.
Redo mfi_mgmt to deal with zero data commands.

Add generic management function.
Redo ccb_state flags so that we can actually sleep on it.
Redo mfi_get_info to use new mgmt function.
Add rw_locks for mgmt interface.
Make BIOCINQ dynamic.

Oops, helps to use the right structure field.
Use actual nr of physical disks present.

Add first BIOCINQ primitive.

Remove some debug cruft and pretty up some debug messages.

Kill splbio/splx in the interrupt handler.

And we have lift off! Was able to create a filesystem and write to the
filesystem.

Collapse both completion routines into one.
Fixup scsi error reporting in the completion routine.

Add completion routines.
Fix some bugs inside the interrupt handler.

Add mfi_get_info to obtain all kinds of fun information from the fw.
Print out whole info structure when debug is enabled.
Add a nice print with useful info during attach.

Refactor scsi path so that we can have a single sgl function for io and non-io
traffic.
Redo parameters on functions that were passing too many parameters around.

Nuke some { }

Move !xs->data case into mfi_start_xs.
Return proper sense data when a polling command fails.

Fix some more htole32 screw-ups.
Colapse all little SCSI functions into a single generic one.
Fix a bug in the SGE calculation.

This brings the driver up to post:
mfi0 at pci12 dev 14 function 0 "Dell PERC SAS" rev 0x00: Dell PERC 5/e irq 5
scsibus3 at mfi0: 2 targets
sd1 at scsibus3 targ 0 lun 0: <DELL, PERC 5/E Adapter, 1.00> SCSI5 0/direct fixed
sd1: 237824MB, 237824 cyl, 64 head, 32 sec, 512 bytes/sec, 487063552 sec total
sd2 at scsibus3 targ 1 lun 0: <DELL, PERC 5/E Adapter, 1.00> SCSI5 0/direct fixed
sd2: 34176MB, 34176 cyl, 64 head, 32 sec, 512 bytes/sec, 69992448 sec total

Fix some htole32 I missed.
Sprinkle more debug.

Add READ CAPACITY.

Oops PREREAD not PREWRITE.

Add TUR and make it send it directly to firmware.
Sprinkle some extra debug.

Clear all relevant ccb fields when mfi_put_ccb frees a ccb.

Oops don't send command twice to firmware when polling.

Add INQUIRY command.
Add sgl magic.

Add some airplane code.

add bio hooks.
add scsi entry point for io.

Fix some debug prints.

typo

Add mfi_poll and use the new shiny function to setup all the hardware queues.
Lots of ideas from the FreeBSD driver.

Fix various printf() issues: too many arguments, not enough arguments, argument
order reversed, wrong modifiers. ok deraadt@ marco@ mickey@

marco hates typedefs as much as i do, so he asked me to clean mfi like i
just did with ami. this gets rid of the typedef for the ccb lists.

also uses TAILQ_FIRST to get ccbs off the freelist. this is a proactive
fix for a lint teary.

It helps to initialize a queue before usage.

Actually add ccbs to the queue.
Fix the error path for mfi_init_ccb.
Alter dmamap for io.

Sprinkle more debug. Remove 64 byte boundary; it doesn't seem to like it.
Hack around 64 byte boundary for the frames for now.

Setup ccbs and clean up ccb structure.

Oops, make mfi_allocmem always use 64 byte boundaries since frames have to
have that alignment. It doesn't matter for the other consumers of the
function.

Add frame memory and sglist allocation.

Make memory access use macro.

typo

Add get and put ccb functions. Shuffle some more definitions.

Nicer print when context is invalid.

Shuffle some #defines.
Fix comments.
Add human readable enable intrrupt #define.

Add interrupt handler. Idea borrowed from FreeBSD since it is much nicer
than the Linux one.

Replace bus_space functions with generic ones that do barriers.

Add memory allocator functions.

Print state in debug. Change return vales to 1 instead of ENXIO.

Make a better debug mechanism.

Enable mfi_transition_fw() original code from FreeBSD.

Add fw transition logic.

Add skeleton driver for MegaRAID SAS.

Use mtx_init() to initialize stack-based mutexes

mtx_init() ensures the mutex' lock_object has static storage duration.
This makes it safe to re-enable WITNESS with stack-based mutexes.

OK bluhm@ miod@

Disable witness for mutexes created on the stack which allows
the ddb command 'show witness' to succeed without panicking.
Leaving witness enabled on these mutexes saves a pointer to
struct lock_type on the stack which gets clobbered resulting
in a panic in witness_ddb_display_descendants().

okay miod@

constify SCSI adapter entry points

ok krw@

Since dlg@'s 2009 scsi midlayer refactoring the 'struct scsi_generic *cmd'
member of 'struct scsi_xfer' has always been pointed at the 'struct scsi_generic
cmdstore' member of the same instance. So nuke 'cmdstore' and remove the '*'
from cmd. Take the address of cmd as required by the various casts.

No intentional functional change.

luna88k test by aoyama@, sparc64 test by jmatthew@

Identification of 2009's last *cmd use and ok jmatthew@

Rename [READ|WRITE]_BIG to [READ|WRITE]_10. Rename struct scsi_rw_big to
struct scsi_rw_10.

ok gnezdo@ jmatthew@ (who also did sparc64 compile test)

Turning on various scsi drivers' *DEBUG options reveals that this has
rarely (if ever) been done.

Fix many printf format errors to calm clang and gcc on amd64, i386,
hppa. Missing #include, complaints if 'option <blah>DEBUG' is used in
config files, etc. All in debug code.

Move remaining scsi bus initialization info from "prototype scsi link"
fields to struct scsibus_attach_args. Nuke the struct scsi_link *
(saa_sc_link) in scaibus_attach_args.

Explicitly initialize each field in scsibus_attach_args variables.

Move the adapter related items (luns, adapter, adapter_target,
adapter_buswidth, adapter_softc) from struct scsi_link to struct
scsibus_attach_args.

Additional compile tests by jmatthew@ (sparc64) and aoyam@ (luna88k).

Access adapter softc via link->bus->sb_adapter_softc.

In sparc64 autoconf access 'luns' via sb_luns and 'adapter_buswidth'
via sb_adapter_buswidth.

Removes last post-config uses of the copies of bus related
information in scsi_link.

Where we have a pointer (sc) to the relevant adapter_softc, use
&sc->sc_dev to get a device pointer rather than using the void *
whence we obtained 'sc'.

Remove some pointless casting of void * when assigning one
(scsi_link.adapter_softc) to a pointer of a particular softc type.

Nuke trailing whitespace.

No need to bzero()/memset() 'struct scsibus_attach_args' variables
immediately before initializing the only field in the struct.

Use SDEV_NO_ADAPTER_TARGET instead of -1 to indicate
the adapter is not a target on the bus.

ok dlg@ as part of a larger diff.

Simplify some goto'ing that jumps a single line or
does nothing at all.

The first one spotted as CID 1452976.

Reads good to millert@

ok mpi@

Nuke *_minphys() functions that either simply apply MAXPHYS or do
nothing at all. MAXPHYS will be applied in minphys() and nothing at
all, well, doesn't do anything.

Also remove any '#define <blah> MAXPHYS' statements used solely to
disguise MAXPHYS in said functions.

Shuffle some names around to make reading the code less headache
inducing. Rename scsi_adapter member 'scsi_minphys' to 'dev_minphys'
to reflect what it is supposed to do. Use consistent naming convention
(<dev>_minphys) for the actual device functions.

No functional change.

Drivers that implement their own *minphys() don't need to call the
system minphys(). scsi_minphys() will do that and cd/sd/st will call
scsi_minphys().

ok jmatthew@ as part of larger diff

Use a consistant idiom/format when declaring scsi_adapter structures
in drivers. Terse one liners, NULLs instead of 0's, explicitly specify
all members, etc.

Nuke #ifdef notyet blocks related to the scsi_adapter in aic.

No intentional functional change.

ok tedu@

Convert infinite sleeps to {m,t}sleep_nsec(9).

ok kn@

apply background init progress fix from r1.59 of mfii.c
(reverting most of r1.60 of mfi.c, oddly enough)

ok dlg@
tested on a SAS2008 (perc h310)

return ENOTTY rather than EINVAL to indicate an ioctl hasn't been handled

ok dlg@ deraadt@ kettenis@

If you use sys/param.h, you don't need sys/types.h

copy the mbox into the dcmd struct, not the pointer to the mbox.

makes bioctl work on mfi(4) again.

while here memset a few mboxes before using them.

bioctl breakage reported by jason george

represent the mbox layout with union instead of an array of bytes.

memcpying uint16_ts into inconsistently addressed offsets is hard
to read, and this makes future work easier to implement.

tested on mfi(4) and mfii(4)

sizes for free(); ok sthen

Initial addition of ``Patrol Read'' support in bio(4), biocto(8), and
mfi(4). Based on FreeBSD, but done without mfiutil(8).

OK deraadt@

Prevent splassert from firing during sd_flush which runs "cold"

While mfi(4) should pass SCSI transfer flags (e.g. SCSI_POLL and
SCSI_NOSLEEP) down to the management function, make it at least
use "cold" consistently for now.

ok dlg

Remove some includes include-what-you-use claims don't
have any direct symbols used. Tested for indirect use by compiling
amd64/i386/sparc64 kernels.

ok tedu@ deraadt@

properly set sgl related flags on skinny controllers. this fixes
io against physical disks on my dell perc h310s. logical disks still
work fine.

ok deraadt@ yasuoka@

Fix mfi ioctl to set drive state properly.
diff from Tsubai Masanari

test dlg jmatthew
ok dlg

Add support DIOC{G,S}CACHE to mfi(4). Cache of disk on mfi(4) will
be enabled by default. Also if the controller doesn't have any RAM,
the cache on the physical disk will be enabled.

discuss with jsg tsubai
ok dlg

another handful of bcopy -> memcpy because there is no overlap

remove uneeded proc.h includes
ok mpi@ kspillner@

mark the interrupt handler mpsafe, and give up biglock in the scsi cmd
submission paths. take biglock again when calling back into the scsi
stack.

tested on a variety of cards here, including the skinny mfis on the
firewall next to mpsafe myx.

Create a function which loads sgd in the mfi_iop_ops struct so that skinny
adapters can use "IEEE sgl".

tested dlg yasuoka
ok dlg jsg

Some reallocarray() use; review Jean-Philippe Ouellet, patrick keshishian
ok tedu

add a size argument to free. will be used soon, but for now default to 0.
after discussions with beck deraadt kettenis.

format string fixes for bus_addr_t and bus_size_t

bus_addr_t and bus_size_t are u_long everywhere

ok kettenis@

revert previous commit; didn't intend to commit those bits

Avoid printing the "nvram corrupt" message for onboard 2200s found on Sun
hardware.

ok dlg@, jmatthew@

mfi_refresh_sensors calls mfi_ioctl_vol which calls mfi_bio_getitall, which
replaces pointers in the softc and all sorts of fun stuff. this protects
the call to bio_getitall with the sc_lock the ioctl path uses, so we dont
get panics like those reported on bugs@ recently.

ok deraadt@

Add support for displaying the cache write policy of RAID volumes to bioctl(8)
and make mfi(4) pass up the necessary information. Adding support for other
RAID controllers is left as an excercise to the reader.

ok deraadt@

Include the 'state of health' field in the bbu ok sensor value.
The dell perc 6/i sets this to zero (meaning bad) when it
decides the battery doesn't have enough capacity for it to
allow write back mode, but doesn't set any of the other flags
to tell us why.

ok dlg@, tested by sthen@

enw->new in comment;
spotted by krw, missed by jsg, and forwarded From: Alexey Suslikov

fix some leaks in error paths
ok fgsch@ krw@ kettenis@

break out the battery status into excruciating detail. makes it harder for
$VENDOR to complain about a lack of diagnostic material from their crappy
mgmt tools if you're not running an operating system they support.

always attach the battery sensor if the adapter lets us query it.
newly replaced batteries dont show up immediately.

provide a sensor for the battery backup unit (bbu) on mfi boards that
support it.

now i can tell (a bit better) why io might be slow on some of my boxes.

dont need to walk alldevs during sensor creation looking for the logical
disk scsibus when we record that in mfi_attach.

also happens to get rid of a logic error that let it come out of the loop
on alldevs with a reference to something that wasnt the logical disk
scsibus and use that.

use MR_DCMD consistently. some MD_DCMD was in there somehow.

david imhoff points out that the error handling in mfi_init_ccb gets the
index to which ccb to free wrong.

this takes the logic from mpi/mpii/mfii which pulls the ready ccbs off
the free list rather than rely on the current index from the for loop
that we failed inside of.

thanks to david for finding this out.

Make sure we don't sleep on autoconf.

ok mikeb

add support for "physical devices" on skinny controllers.

these controllers let you specify disks that should be accessed directly
rather than be part of logical volumes. apparently you can plug in things
like tape drives too.

anyway, physical devices are exposed via a separate scsibus attached to
mfi.

ok mikeb@

fix mfi_poll by having it do delays between loops, and breaking on
timeouts. how embarrassment.

implement mfi_exec for doing sync executions of commands in process
contexts and cut mfi_mgmt over to it.

rework mfi_poll so it will do all the completions on a ccb including
calling the done handler.

this makes it consistently complete scsi io with all the proper error
checking.

stop passing mfi_softc around as a member of the ccb, consistently give it
as the first argument to functions. like mfii.

scrub ccbs when we're about to use them, not when we put them back
on the free list. ccbs can be reused, so we dont want old state screwing up
new commands.

move the dma syncs for the ccb data into ccb_done, so we can do it once
rather than put the code everywhere.

fix up the dma sync for the command frame in mfi_done.

use scsi_get_link to find out the name of a child device for bioctl rather
than snoop scsi commands and copy it in an io path.

no functional change, but the code has one less XXX now.

expose the max number of logical volumes the hardware supports as the width
of its scsibus. only advertise 1 lun on logical volumes instead of letting
the midlayer fix it up to 8.

give every target on the bus max_cmds openings. iopools means they will
properly share access to them.

this in particular is useful on skinny controllers which only advertise 31
command slots. if you have 16 volumes, theyll only get 1 opening each with
the old maths. this way round the ones that are busy will share the slots.

tested on a perc5 with two volumes and hard workloads.

print the boards name (eg, perc 5/i) according to the firmware like mfii.

ok deraadt@

move knowledge of the location of the inbound doorbell out of code in
transition_firmware into a member of the iop structures.

ok mikeb@ haesbaert@

get rid of the last foo_lo and foo_hi bits i could find in the hardware
structures and use htole64 instead of htole32(addr >> 32); htole32(addr);

gets rid of "handy" stack variables to get the dva to 64bits.

sprinkle some more byte swaps for things that should have it.

tested on a perc5 (xscale)

do appropriate bus_dmamapy_syncs around the pcq/completion ring.

tested on a perc5 (xscale)

replace uint32_ts for hi and lo addresses in the mfi_init_ hardware
descriptors with single uint64_ts. theyre not some weird middle
endian thing. this makes the code more readable.

add some missing htole32s to the code while here.

tested on a perc5 (xscale).

the producer/consumer values are used by the device, which is little
endian. do appropriate byteswapping for when the host cpu uses them.

tested on a perc5 (xscale) in an amd64.

add support for skinny variants.

from Alexey Suslikov
reviewed by marco

Backout a bunch of my SCSI commits from c2k11. At least one of these
is causing problems when trying to boot sparc64 from an isp(4).

Verified to fix the sparc64/isp(4) regression by krw@; ok deraadt@

First batch of converting SCSI HBAs from setting saa_targets and
saa_luns instead of adapter_buswidth and luns in the prototype link.

ok dlg@, miod@

add all 64 bit dma plumbing but stick with 32 for now by default.

ok deraadt

make sure the upper 32 bits are always written to instead of assuming it
is 0.

ok deraadt.

kill some stray spaces

rename wiat channel to something less than 8 chars.

prompted by deraadt

use dma_alloc for transient management commands.

prodded and ok deraadt

cut mfi over to using iopools.

this gets rid of NO_CCB, provides better scheduling of io between busy
logical volumes on the same controller, and makes the ioctl and sensor
paths more reliable.

this is the least intrusive version of this change i could come up with. it
only took 4 or 5 goes...

ok marco@
deraadt@ says this is a good time for this change.

no use setting cmd in mfi_ioctl_setstate if you never use it.

Use SSD_ERRCODE_CURRENT instead of magic 0x70.

ok dlg@ matthew@

the scsi completion code thinks that if you dont get MFI_STAT_OK
back from the chip on a command then the command completely failed.
MFI_STAT_SCSI_DONE_WITH_ERROR really means the command completed
fine, but there's some sense data too. this tweaks the handling to
be more appropriate, as per the linux and solaris drivers.

timed out waiting for beck@
putting this in cos its obviously more correct than the current code.

SCSI HBA drivers are supposed to get their SCSI CDBs from xs->cmd, not
xs->cmdstore.

ok krw@ a while back (contingent upon testing), and deraadt@ says this
has been in snapshots for weeks now.

Change scsibus(4)'s scsi_link array to an SLIST to save memory on
sparsely populated buses.

ok dlg@, krw@

Use SLIST for the ccb free list.

ok krw, discontent grunts from marco.

Remove all adapter-specific 'struct scsi_device's. They are never used. First
step in elminating 'struct scsi_device' entirely.

Spotted and initial diff from matthew@.

ok matthew@ dlg@ deraadt@ marco@ miod@

dont pass the dev_t from the scsi device drivers into the midlayer for
ioctl requests, and dont pass the proc pointers around for any ioctl
requests in scsi land at all. neither were used, so trim the fat.

ok krw@ marco@

New scsi code seems to be stable. Pluck previously identified
low-hanging splbio/splx pairs that are no longer needed and see if
this reveals any hidden scsi flaws.

ok dlg@

Use a temporary variable for now to sidestep -Wbounded checking when
copying vendor[8]/product[16]/revision[4] out of struct scsi_inquiry_data
together with one memcopy.

ok krw

use BUS_DMA_ZERO on alloc instead of bzeroing after.

ok dlg@, marco@

grrr var names in protos

sprinkle some bus_dmamap_sync around frame access.
prompted by oga

Change the scsi_cmd function member of scsi_adapter from int to
void. Use XS_NO_CCB error in the scsi command (xs) to report the
NO_CCB condition. Eliminates all SUCCESSFULLY_QUEUED and COMPLETE
confusion and untangles the midlayer from the adapter a bit more.

Eyes and some fixes by miod@

There may be some compile issues on little used (i.e. I don't have
any) drivers but the change is mechanical and thus easy to remedy.

ok dlg@

Zap all setting of ITSDONE in drivers that don't look at it. Nobody
else cares so it's just noise. Drivers that actually look at ITSDONE
are unchanged.

ok marco@ (for his files) dlg@ beck@

rename ccb_xs to ccb_cookie and change its type from a scsi_xfer * to a
void *. this lets us pass things other than scsi_xfers to ccb completion
handlers.

protect the ccb free queue with a mutex rather than splbio

change M_WAITOK --> M_WAITOK|M_CANFAIL

with input from marco@

ok marco@ krw@

this was ok looooooooong time ago but lost in my mailbox

Add READ(16) & WRITE(16) commands.
ok krw

MFI_D_IOCTL should be commented out. No binary change.

scrub more fields in the ccb when returning them to the free list after
theyve been used, in particular the mfi header flags which has a bit
that specifies if a command should be completed via the interrupt path.

if we use a ccb during boot we set that bit, but it isnt necessarily
cleared by things that use it later on. this means a ccb we expected to
complete via an interrupt never actually generates an interrupt or appears
in the reply queue. this obviously stalls the io.

spello

when we steal the SYNC CACHE command in mfi_scsi_cmd we still have to
mark the xs as done and call scsi_done for it rather than just returning
COMPLETE.

fix a small race in mfi_mgmt between the checking of a ccbs completion and
the sleep waiting for the completion. it is possible to get the interrupt
completing the command just before the tsleep, which will never get a
wakeup because the interrupt with the wakeup has already happened.

Extend the scsi_adapter minphys() callback to take a struct scsi_link *
as additional argument. This will allow intermediate layers between
scsi devices such as sd and scsi host adapters to take appropriate
action if necessary.

Add support for next generation megaraid sas raid controllers.
From Jim Giannoules <gortag@gmail.com>

Generalize function to obtain all details to do bio stuff. This is in
preparation to start showing unsused disks which will require all this
information to do the math.

While at it fix a size of hotspare volume bug.

Fix bug that when there are no volumes but there are hot spares it would
no show the hotspares.

Fix output when reporting on multivolume disk groups.

Reported by mitja@muzenic.net

Move some include stuff around

Space

Fix PR5645
ok dlg
tested beck and others

Initialize 'ssc' to NULL since we test it later. Shouldn't ever be
NULL at the point of test unless the device list is confused, but
be consistantly paranoid if we're going to be paranoid.

Return NO_CCB and not TRY_AGAIN_LATER when running out of ccb's.
Eliminates 'not queued' errors reported on the mailing lists.

Tested by Michael (belenus at bsdmail dot de).

ok marco@

add support for the mfi controllers with the ppc iops, aka, LSI SAS1078 or
Dell PERC6.

committed off a machine booting and rooting off a perc6. tested on a perc5
by marco@
ok marco@

Set the return value instead of returning right away so that the DMA
memory is actually synched.

ok dlg@

sync dma mem back to the cpu on poll failure.

from mickey via pr5714

despatch -> dispatch

disparate spellink fix from mickey

More scsi_done() at SPLBIO.

M_ZERO changes

ok dlg@

rework sensor tasks to use the kernels generic workq rather than a special
kernel thread of its own. the api has changed (which will be fixed in the
manpage shortly) so all the users of sensor tasks that i can find have
been fixed too.

noone tested, so its going in to force people to run with it.
"put it in" deraadt@

sprinkle some SMALL_KERNEL to permit bio/softraid to run on boot media
ok marco@ then suggested deraadt@

split userland & kernel struct sensor/sensordev so that the addition
of new fields in the future is less disruptive. This is done similar
to how struct proc is handled for ps(1). ok jmc (man page changes)
tested fkr simon, and more suggestions from millert

bus_space does byte swapping to cope with host endianness for us, so remove
an unnecessary htole32().

Fix subtle x ? y : z bug when calling mfi_create_sgl. This worked because
during runtime the bug condition does not happen.

ok krw joris

Remove double semicolons.

From Pierre Riteau <pierre dot riteau at free dot fr>.

OK marco@.

adapt to new two-level sensor api; Constantine A. Murenin

give scsi controllers a real attach args to fill in when attaching scsibus.

ok miod@ marco@ deraadt@

Add sensor thread for individual sd devices.

tested dlg
ok dlg deraadt

when determining the size of the volumes and disks we want how many
bytes each of them has, not the number of kilobytes. this solves the
"why am i dividing by 2" question. the answer is "you dont, you multiply
by 512".

make dmesg printing look a little bit nicer.

Factor for disk size is /2. Odd but tested on all mfis I have.

Everytime one forgets an argument in a printf-like function call, God kills
a kitten. Commiters, please think of the kittens when working on code.

Make a BGI (Back Ground Init) a scrub instead of rebuild.

Add hot spare support.

Add blinking for disks.

Move a int i; to top of function so that it compiles on non c99 compilers.
Pointed out by miod@

The enclosure id modifier didn't reflect reality so move it back to whatever
the raid card uses.

And add time and percentage to builds and check consistencies.

Add last bit of physcal disk stuff.
Remove bogus memset.
Hotspares still don't work.

Add physical device and read config structures.
Add initial part of BIOCDISK

commited with unused variable

Remove some debug cruft.

Add first half of physical disk data for bio.

Add support for raid 50 and 10 in bio.

Add mail box to mfi_mgmt to handle sub commands more gracefully.
Remove some debug crud that snuck in the TEST_UNIT_READY case.
Finish other half of BIOCVOL primitive.
Add several structures to fill out the BIOCVOL primitive.

Add first half of BIOCVOL primitive.

Do a giant flush of disk and raid-card memory when SYNCHONIZE_CACHE is called.

pretty up some debug statements.

Add bio alarm function.
Redo mfi_mgmt to deal with zero data commands.

Add generic management function.
Redo ccb_state flags so that we can actually sleep on it.
Redo mfi_get_info to use new mgmt function.
Add rw_locks for mgmt interface.
Make BIOCINQ dynamic.

Oops, helps to use the right structure field.
Use actual nr of physical disks present.

Add first BIOCINQ primitive.

Remove some debug cruft and pretty up some debug messages.

Kill splbio/splx in the interrupt handler.

And we have lift off! Was able to create a filesystem and write to the
filesystem.

Collapse both completion routines into one.
Fixup scsi error reporting in the completion routine.

Add completion routines.
Fix some bugs inside the interrupt handler.

Add mfi_get_info to obtain all kinds of fun information from the fw.
Print out whole info structure when debug is enabled.
Add a nice print with useful info during attach.

Refactor scsi path so that we can have a single sgl function for io and non-io
traffic.
Redo parameters on functions that were passing too many parameters around.

Nuke some { }

Move !xs->data case into mfi_start_xs.
Return proper sense data when a polling command fails.

Fix some more htole32 screw-ups.
Colapse all little SCSI functions into a single generic one.
Fix a bug in the SGE calculation.

This brings the driver up to post:
mfi0 at pci12 dev 14 function 0 "Dell PERC SAS" rev 0x00: Dell PERC 5/e irq 5
scsibus3 at mfi0: 2 targets
sd1 at scsibus3 targ 0 lun 0: <DELL, PERC 5/E Adapter, 1.00> SCSI5 0/direct fixed
sd1: 237824MB, 237824 cyl, 64 head, 32 sec, 512 bytes/sec, 487063552 sec total
sd2 at scsibus3 targ 1 lun 0: <DELL, PERC 5/E Adapter, 1.00> SCSI5 0/direct fixed
sd2: 34176MB, 34176 cyl, 64 head, 32 sec, 512 bytes/sec, 69992448 sec total

Fix some htole32 I missed.
Sprinkle more debug.

Add READ CAPACITY.

Oops PREREAD not PREWRITE.

Add TUR and make it send it directly to firmware.
Sprinkle some extra debug.

Clear all relevant ccb fields when mfi_put_ccb frees a ccb.

Oops don't send command twice to firmware when polling.

Add INQUIRY command.
Add sgl magic.

Add some airplane code.

add bio hooks.
add scsi entry point for io.

Fix some debug prints.

typo

Add mfi_poll and use the new shiny function to setup all the hardware queues.
Lots of ideas from the FreeBSD driver.

Fix various printf() issues: too many arguments, not enough arguments, argument
order reversed, wrong modifiers. ok deraadt@ marco@ mickey@

marco hates typedefs as much as i do, so he asked me to clean mfi like i
just did with ami. this gets rid of the typedef for the ccb lists.

also uses TAILQ_FIRST to get ccbs off the freelist. this is a proactive
fix for a lint teary.

It helps to initialize a queue before usage.

Actually add ccbs to the queue.
Fix the error path for mfi_init_ccb.
Alter dmamap for io.

Sprinkle more debug. Remove 64 byte boundary; it doesn't seem to like it.
Hack around 64 byte boundary for the frames for now.

Setup ccbs and clean up ccb structure.

Oops, make mfi_allocmem always use 64 byte boundaries since frames have to
have that alignment. It doesn't matter for the other consumers of the
function.

Add frame memory and sglist allocation.

Make memory access use macro.

typo

Add get and put ccb functions. Shuffle some more definitions.

Nicer print when context is invalid.

Shuffle some #defines.
Fix comments.
Add human readable enable intrrupt #define.

Add interrupt handler. Idea borrowed from FreeBSD since it is much nicer
than the Linux one.

Replace bus_space functions with generic ones that do barriers.

Add memory allocator functions.

Print state in debug. Change return vales to 1 instead of ENXIO.

Make a better debug mechanism.

Enable mfi_transition_fw() original code from FreeBSD.

Add fw transition logic.

Add skeleton driver for MegaRAID SAS.

Disable witness for mutexes created on the stack which allows
the ddb command 'show witness' to succeed without panicking.
Leaving witness enabled on these mutexes saves a pointer to
struct lock_type on the stack which gets clobbered resulting
in a panic in witness_ddb_display_descendants().

okay miod@

constify SCSI adapter entry points

ok krw@

Since dlg@'s 2009 scsi midlayer refactoring the 'struct scsi_generic *cmd'
member of 'struct scsi_xfer' has always been pointed at the 'struct scsi_generic
cmdstore' member of the same instance. So nuke 'cmdstore' and remove the '*'
from cmd. Take the address of cmd as required by the various casts.

No intentional functional change.

luna88k test by aoyama@, sparc64 test by jmatthew@

Identification of 2009's last *cmd use and ok jmatthew@

Rename [READ|WRITE]_BIG to [READ|WRITE]_10. Rename struct scsi_rw_big to
struct scsi_rw_10.

ok gnezdo@ jmatthew@ (who also did sparc64 compile test)

Turning on various scsi drivers' *DEBUG options reveals that this has
rarely (if ever) been done.

Fix many printf format errors to calm clang and gcc on amd64, i386,
hppa. Missing #include, complaints if 'option <blah>DEBUG' is used in
config files, etc. All in debug code.

Move remaining scsi bus initialization info from "prototype scsi link"
fields to struct scsibus_attach_args. Nuke the struct scsi_link *
(saa_sc_link) in scaibus_attach_args.

Explicitly initialize each field in scsibus_attach_args variables.

Move the adapter related items (luns, adapter, adapter_target,
adapter_buswidth, adapter_softc) from struct scsi_link to struct
scsibus_attach_args.

Additional compile tests by jmatthew@ (sparc64) and aoyam@ (luna88k).

Access adapter softc via link->bus->sb_adapter_softc.

In sparc64 autoconf access 'luns' via sb_luns and 'adapter_buswidth'
via sb_adapter_buswidth.

Removes last post-config uses of the copies of bus related
information in scsi_link.

Where we have a pointer (sc) to the relevant adapter_softc, use
&sc->sc_dev to get a device pointer rather than using the void *
whence we obtained 'sc'.

Remove some pointless casting of void * when assigning one
(scsi_link.adapter_softc) to a pointer of a particular softc type.

Nuke trailing whitespace.

No need to bzero()/memset() 'struct scsibus_attach_args' variables
immediately before initializing the only field in the struct.

Use SDEV_NO_ADAPTER_TARGET instead of -1 to indicate
the adapter is not a target on the bus.

ok dlg@ as part of a larger diff.

Simplify some goto'ing that jumps a single line or
does nothing at all.

The first one spotted as CID 1452976.

Reads good to millert@

ok mpi@

Nuke *_minphys() functions that either simply apply MAXPHYS or do
nothing at all. MAXPHYS will be applied in minphys() and nothing at
all, well, doesn't do anything.

Also remove any '#define <blah> MAXPHYS' statements used solely to
disguise MAXPHYS in said functions.

Shuffle some names around to make reading the code less headache
inducing. Rename scsi_adapter member 'scsi_minphys' to 'dev_minphys'
to reflect what it is supposed to do. Use consistent naming convention
(<dev>_minphys) for the actual device functions.

No functional change.

Drivers that implement their own *minphys() don't need to call the
system minphys(). scsi_minphys() will do that and cd/sd/st will call
scsi_minphys().

ok jmatthew@ as part of larger diff

Use a consistant idiom/format when declaring scsi_adapter structures
in drivers. Terse one liners, NULLs instead of 0's, explicitly specify
all members, etc.

Nuke #ifdef notyet blocks related to the scsi_adapter in aic.

No intentional functional change.

ok tedu@

Convert infinite sleeps to {m,t}sleep_nsec(9).

ok kn@

apply background init progress fix from r1.59 of mfii.c
(reverting most of r1.60 of mfi.c, oddly enough)

ok dlg@
tested on a SAS2008 (perc h310)

return ENOTTY rather than EINVAL to indicate an ioctl hasn't been handled

ok dlg@ deraadt@ kettenis@

If you use sys/param.h, you don't need sys/types.h

copy the mbox into the dcmd struct, not the pointer to the mbox.

makes bioctl work on mfi(4) again.

while here memset a few mboxes before using them.

bioctl breakage reported by jason george

represent the mbox layout with union instead of an array of bytes.

memcpying uint16_ts into inconsistently addressed offsets is hard
to read, and this makes future work easier to implement.

tested on mfi(4) and mfii(4)

sizes for free(); ok sthen

Initial addition of ``Patrol Read'' support in bio(4), biocto(8), and
mfi(4). Based on FreeBSD, but done without mfiutil(8).

OK deraadt@

Prevent splassert from firing during sd_flush which runs "cold"

While mfi(4) should pass SCSI transfer flags (e.g. SCSI_POLL and
SCSI_NOSLEEP) down to the management function, make it at least
use "cold" consistently for now.

ok dlg

Remove some includes include-what-you-use claims don't
have any direct symbols used. Tested for indirect use by compiling
amd64/i386/sparc64 kernels.

ok tedu@ deraadt@

properly set sgl related flags on skinny controllers. this fixes
io against physical disks on my dell perc h310s. logical disks still
work fine.

ok deraadt@ yasuoka@

Fix mfi ioctl to set drive state properly.
diff from Tsubai Masanari

test dlg jmatthew
ok dlg

Add support DIOC{G,S}CACHE to mfi(4). Cache of disk on mfi(4) will
be enabled by default. Also if the controller doesn't have any RAM,
the cache on the physical disk will be enabled.

discuss with jsg tsubai
ok dlg

another handful of bcopy -> memcpy because there is no overlap

remove uneeded proc.h includes
ok mpi@ kspillner@

mark the interrupt handler mpsafe, and give up biglock in the scsi cmd
submission paths. take biglock again when calling back into the scsi
stack.

tested on a variety of cards here, including the skinny mfis on the
firewall next to mpsafe myx.

Create a function which loads sgd in the mfi_iop_ops struct so that skinny
adapters can use "IEEE sgl".

tested dlg yasuoka
ok dlg jsg

Some reallocarray() use; review Jean-Philippe Ouellet, patrick keshishian
ok tedu

add a size argument to free. will be used soon, but for now default to 0.
after discussions with beck deraadt kettenis.

format string fixes for bus_addr_t and bus_size_t

bus_addr_t and bus_size_t are u_long everywhere

ok kettenis@

revert previous commit; didn't intend to commit those bits

Avoid printing the "nvram corrupt" message for onboard 2200s found on Sun
hardware.

ok dlg@, jmatthew@

mfi_refresh_sensors calls mfi_ioctl_vol which calls mfi_bio_getitall, which
replaces pointers in the softc and all sorts of fun stuff. this protects
the call to bio_getitall with the sc_lock the ioctl path uses, so we dont
get panics like those reported on bugs@ recently.

ok deraadt@

Add support for displaying the cache write policy of RAID volumes to bioctl(8)
and make mfi(4) pass up the necessary information. Adding support for other
RAID controllers is left as an excercise to the reader.

ok deraadt@

Include the 'state of health' field in the bbu ok sensor value.
The dell perc 6/i sets this to zero (meaning bad) when it
decides the battery doesn't have enough capacity for it to
allow write back mode, but doesn't set any of the other flags
to tell us why.

ok dlg@, tested by sthen@

enw->new in comment;
spotted by krw, missed by jsg, and forwarded From: Alexey Suslikov

fix some leaks in error paths
ok fgsch@ krw@ kettenis@

break out the battery status into excruciating detail. makes it harder for
$VENDOR to complain about a lack of diagnostic material from their crappy
mgmt tools if you're not running an operating system they support.

always attach the battery sensor if the adapter lets us query it.
newly replaced batteries dont show up immediately.

provide a sensor for the battery backup unit (bbu) on mfi boards that
support it.

now i can tell (a bit better) why io might be slow on some of my boxes.

dont need to walk alldevs during sensor creation looking for the logical
disk scsibus when we record that in mfi_attach.

also happens to get rid of a logic error that let it come out of the loop
on alldevs with a reference to something that wasnt the logical disk
scsibus and use that.

use MR_DCMD consistently. some MD_DCMD was in there somehow.

david imhoff points out that the error handling in mfi_init_ccb gets the
index to which ccb to free wrong.

this takes the logic from mpi/mpii/mfii which pulls the ready ccbs off
the free list rather than rely on the current index from the for loop
that we failed inside of.

thanks to david for finding this out.

Make sure we don't sleep on autoconf.

ok mikeb

add support for "physical devices" on skinny controllers.

these controllers let you specify disks that should be accessed directly
rather than be part of logical volumes. apparently you can plug in things
like tape drives too.

anyway, physical devices are exposed via a separate scsibus attached to
mfi.

ok mikeb@

fix mfi_poll by having it do delays between loops, and breaking on
timeouts. how embarrassment.

implement mfi_exec for doing sync executions of commands in process
contexts and cut mfi_mgmt over to it.

rework mfi_poll so it will do all the completions on a ccb including
calling the done handler.

this makes it consistently complete scsi io with all the proper error
checking.

stop passing mfi_softc around as a member of the ccb, consistently give it
as the first argument to functions. like mfii.

scrub ccbs when we're about to use them, not when we put them back
on the free list. ccbs can be reused, so we dont want old state screwing up
new commands.