freebsd-dev/sys/dev/isp/isp_tpublic.h
Matt Jacob 5d57194434 Spring MegaChange #1.
----

Make a device for each ISP- really usable only with devfs and add an ioctl
entry point (this can be used to (re)set debug levels, reset the HBA,
rescan the fabric, issue lips, etc).

----

Add in a kernel thread for Fibre Channel cards. The purpose of this
thread is to be woken up to clean up after Fibre Channel events
block things.  Basically, any FC event that casts doubt on the
location or identify of FC devices blocks the queues. When, and
if, we get the PORT DATABASE CHANGED or NAME SERVER DATABASE CHANGED
async event, we activate the kthread which will then, in full thread
context, re-evaluate the local loop and/or the fabric. When it's
satisfied that things are stable, it can then release the blocked
queues and let commands flow again.

The prior mechanism was a lazy evaluation. That is, the next command
to come down the pipe after change events would pay the full price
for re-evaluation. And if this was done off of a softcall, it really
could hang up the system.

These changes brings the FreeBSD port more in line with the Solaris,
Linux and NetBSD ports. It also, more importantly, gets us being
more proactive about topology changes which could then be reflected
upwards to CAM so that the periph driver can be informed sooner
rather than later when things arrive or depart.

---

Add in the (correct) usage of locking macros- we now have lock transition
macros which allow us to transition from holding the CAM lock (Giant)
and grabbing the softc lock and vice versa. Switch over to having this
HBA do real locking. Some folks claim this won't be a win. They're right.
But you have to start somewhere, and this will begin to teach us how
to DTRT for HBAs, etc.

--

Start putting in prototype 2300 support.  Add back in LIP
and Loop Reset as async events that each platform will handle.
Add in another int_bogus instrumentation point.

Do some more substantial target mode cleanups.

MFC after:	8 weeks
2001-05-28 21:20:43 +00:00

335 lines
14 KiB
C

/* $FreeBSD$ */
/*
* Qlogic ISP Host Adapter Public Target Interface Structures && Routines
*---------------------------------------
* Copyright (c) 2000 by Matthew Jacob
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions, and the following disclaimer,
* without modification, immediately at the beginning of the file.
* 2. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* Matthew Jacob
* Feral Software
* mjacob@feral.com
*/
/*
* Required software target mode message and event handling structures.
*
* The message and event structures are used by the MI layer
* to propagate messages and events upstream.
*/
#ifndef IN_MSGLEN
#define IN_MSGLEN 8
#endif
typedef struct {
void * nt_hba; /* HBA tag */
u_int64_t nt_iid; /* inititator id */
u_int64_t nt_tgt; /* target id */
u_int64_t nt_lun; /* logical unit */
u_int8_t nt_bus; /* bus */
u_int8_t nt_tagtype; /* tag type */
u_int16_t nt_tagval; /* tag value */
u_int8_t nt_msg[IN_MSGLEN]; /* message content */
} tmd_msg_t;
typedef struct {
void * ev_hba; /* HBA tag */
u_int16_t ev_bus; /* bus */
u_int16_t ev_event; /* type of async event */
} tmd_event_t;
/*
* Suggested Software Target Mode Command Handling structure.
*
* A note about terminology:
*
* MD stands for "Machine Dependent".
*
* This driver is structured in three layers: Outer MD, core, and inner MD.
* The latter also is bus dependent (i.e., is cognizant of PCI bus issues
* as well as platform issues).
*
*
* "Outer Layer" means "Other Module"
*
* Some additional module that actually implements SCSI target command
* policy is the recipient of incoming commands and the source of the
* disposition for them.
*
* The command structure below is one suggested possible MD command structure,
* but since the handling of thbis is entirely in the MD layer, there is
* no explicit or implicit requirement that it be used.
*
* The cd_private tag should be used by the MD layer to keep a free list
* of these structures. Code outside of this driver can then use this
* as an to identify it's own unit structures. That is, when not on the MD
* layer's freelist, the MD layer should shove into it the identifier
* that the outer layer has for it- passed in on an initial QIN_HBA_REG
* call (see below).
*
* The cd_hba tag is a tag that uniquely identifies the HBA this target
* mode command is coming from. The outer layer has to pass this back
* unchanged to avoid chaos.
*
* The cd_iid, cd_tgt, cd_lun and cd_bus tags are used to identify the
* id of the initiator who sent us a command, the target claim to be, the
* lun on the target we claim to be, and the bus instance (for multiple
* bus host adapters) that this applies to (consider it an extra Port
* parameter). The iid, tgt and lun values are deliberately chosen to be
* fat so that, for example, World Wide Names can be used instead of
* the units that the Qlogic firmware uses (in the case where the MD
* layer maintains a port database, for example).
*
* The cd_tagtype field specifies what kind of command tag has been
* sent with the command. The cd_tagval is the tag's value (low 16
* bits). It also contains (in the upper 16 bits) any command handle.
*
*
* N.B.: when the MD layer sends this command to outside software
* the outside software likely *MUST* return the same cd_tagval that
* was in place because this value is likely what the Qlogic f/w uses
* to identify a command.
*
* The cd_cdb contains storage for the passed in command descriptor block.
* This is the maximum size we can get out of the Qlogic f/w. There's no
* passed in length because whoever decodes the command to act upon it
* will know what the appropriate length is.
*
* The tag cd_lflags are the flags set by the MD driver when it gets
* command incoming or when it needs to inform any outside entities
* that the last requested action failed.
*
* The tag cd_hflags should be set by any outside software to indicate
* the validity of sense and status fields (defined below) and to indicate
* the direction data is expected to move. It is an error to have both
* CDFH_DATA_IN and CDFH_DATA_OUT set.
*
* If the CDFH_STSVALID flag is set, the command should be completed (after
* sending any data and/or status). If CDFH_SNSVALID is set and the MD layer
* can also handle sending the associated sense data (either back with an
* FCP RESPONSE IU for Fibre Channel or otherwise automatically handling a
* REQUEST SENSE from the initator for this target/lun), the MD layer will
* set the CDFL_SENTSENSE flag on successful transmission of the sense data.
* It is an error for the CDFH_SNSVALID bit to be set and CDFH_STSVALID not
* to be set. It is an error for the CDFH_SNSVALID be set and the associated
* SCSI status (cd_scsi_status) not be set to CHECK CONDITON.
*
* The tag cd_data points to a data segment to either be filled or
* read from depending on the direction of data movement. The tag
* is undefined if no data direction is set. The MD layer and outer
* layers must agree on the meaning of cd_data.
*
* The tag cd_totlen is the total data amount expected to be moved
* over the life of the command. It *may* be set by the MD layer, possibly
* from the datalen field of an FCP CMND IU unit. If it shows up in the outer
* layers set to zero and the CDB indicates data should be moved, the outer
* layer should set it to the amount expected to be moved.
*
* The tag cd_resid should be the total residual of data not transferred.
* The outer layers need to set this at the begining of command processing
* to equal cd_totlen. As data is successfully moved, this value is decreased.
* At the end of a command, any nonzero residual indicates the number of bytes
* requested but not moved. XXXXXXXXXXXXXXXXXXXXXXX TOO VAGUE!!!
*
* The tag cd_xfrlen is the length of the currently active data transfer.
* This allows several interations between any outside software and the
* MD layer to move data.
*
* The reason that total length and total residual have to be tracked
* is that fibre channel FCP DATA IU units have to have a relative
* offset field.
*
* N.B.: there is no necessary 1-to-1 correspondence between any one
* data transfer segment and the number of CTIOs that will be generated
* satisfy the current data transfer segment. It's not also possible to
* predict how big a transfer can be before it will be 'too big'. Be
* reasonable- a 64KB transfer is 'reasonable'. A 1MB transfer may not
* be. A 32MB transfer is unreasonable. The problem here has to do with
* how CTIOs can be used to map passed data pointers. In systems which
* have page based scatter-gather requirements, each PAGESIZEd chunk will
* consume one data segment descriptor- you get 3 or 4 of them per CTIO.
* The size of the REQUEST QUEUE you drop a CTIO onto is finite (typically
* it's 256, but on some systems it's even smaller, and note you have to
* sure this queue with the initiator side of this driver).
*
* The tags cd_sense and cd_scsi_status are pretty obvious.
*
* The tag cd_error is to communicate between the MD layer and outer software
* the current error conditions.
*
* The tag cd_reserved pads out the structure to 128 bytes. The first
* half of the pad area is reserved to the MD layer, and the second half
* may be used by outer layers, for scratch purposes.
*/
#ifndef _LP64
#if defined(__alpha__) || defined(__sparcv9cpu) || defined(__sparc_v9__)
#define _LP64
#endif
#endif
#ifndef _TMD_PAD_LEN
#ifdef _LP64
#define _TMD_PAD_LEN 12
#else
#define _TMD_PAD_LEN 24
#endif
#endif
#ifndef ATIO_CDBLEN
#define ATIO_CDBLEN 26
#endif
#ifndef QLTM_SENSELEN
#define QLTM_SENSELEN 18
#endif
typedef struct tmd_cmd {
void * cd_private; /* layer private data */
void * cd_hba; /* HBA tag */
void * cd_data; /* 'pointer' to data */
u_int64_t cd_iid; /* initiator ID */
u_int64_t cd_tgt; /* target id */
u_int64_t cd_lun; /* logical unit */
u_int8_t cd_bus; /* bus */
u_int8_t cd_tagtype; /* tag type */
u_int32_t cd_tagval; /* tag value */
u_int8_t cd_cdb[ATIO_CDBLEN]; /* Command */
u_int8_t cd_lflags; /* flags lower level sets */
u_int8_t cd_hflags; /* flags higher level sets */
u_int32_t cd_totlen; /* total data requirement */
u_int32_t cd_resid; /* total data residual */
u_int32_t cd_xfrlen; /* current data requirement */
int32_t cd_error; /* current error */
u_int8_t cd_sense[QLTM_SENSELEN];
u_int16_t cd_scsi_status; /* closing SCSI status */
u_int8_t cd_reserved[_TMD_PAD_LEN];
} tmd_cmd_t;
#define CDFL_SNSVALID 0x01 /* sense data (from f/w) valid */
#define CDFL_NODISC 0x02 /* disconnects disabled */
#define CDFL_SENTSENSE 0x04 /* last action sent sense data */
#define CDFL_SENTSTATUS 0x08 /* last action sent status */
#define CDFL_ERROR 0x10 /* last action ended in error */
#define CDFL_BUSY 0x40 /* this command is not on a free list */
#define CDFL_PRIVATE_0 0x80 /* private layer flags */
#define CDFH_SNSVALID 0x01 /* sense data valid */
#define CDFH_STSVALID 0x02 /* status valid */
#define CDFH_NODATA 0x00 /* no data transfer expected */
#define CDFH_DATA_IN 0x04 /* target (us) -> initiator (them) */
#define CDFH_DATA_OUT 0x08 /* initiator (them) -> target (us) */
#define CDFH_DATA_MASK 0x0C /* mask to cover data direction */
#define CDFH_PRIVATE_0 0x80 /* private layer flags */
/*
* Action codes set by the Qlogic MD target driver for
* the external layer to figure out what to do with.
*/
typedef enum {
QOUT_HBA_REG=0, /* the argument is a pointer to a hba_register_t */
QOUT_TMD_START, /* the argument is a pointer to a tmd_cmd_t */
QOUT_TMD_DONE, /* the argument is a pointer to a tmd_cmd_t */
QOUT_TEVENT, /* the argument is a pointer to a tmd_event_t */
QOUT_TMSG, /* the argument is a pointer to a tmd_msg_t */
QOUT_HBA_UNREG /* the argument is a pointer to a hba_register_t */
} tact_e;
/*
* Action codes set by the external layer for the
* MD Qlogic driver to figure out what to do with.
*/
typedef enum {
QIN_HBA_REG=6, /* the argument is a pointer to a hba_register_t */
QIN_ENABLE, /* the argument is a pointer to a tmd_cmd_t */
QIN_DISABLE, /* the argument is a pointer to a tmd_cmd_t */
QIN_TMD_CONT, /* the argument is a pointer to a tmd_cmd_t */
QIN_TMD_FIN, /* the argument is a pointer to a done tmd_cmd_t */
QIN_HBA_UNREG /* the argument is a pointer to a hba_register_t */
} qact_e;
/*
* A word about the START/CONT/DONE/FIN dance:
*
* When the HBA is enabled for receiving commands, one may show up
* without notice. When that happens, the Qlogic target mode driver
* gets a tmd_cmd_t, fills it with the info that just arrived, and
* calls the outer layer with a QOUT_TMD_START code and pointer to
* the tmd_cmd_t.
*
* The outer layer decodes the command, fetches data, prepares stuff,
* whatever, and starts by passing back the pointer with a QIN_TMD_CONT
* code which causes the Qlogic target mode driver to generate CTIOs to
* satisfy whatever action needs to be taken. When those CTIOs complete,
* the Qlogic target driver sends the pointer to the cmd_tmd_t back with
* a QOUT_TMD_DONE code. This repeats for as long as necessary.
*
* The outer layer signals it wants to end the command by settings within
* the tmd_cmd_t itself. When the final QIN_TMD_CONT is reported completed,
* the outer layer frees the tmd_cmd_t by sending the pointer to it
* back with a QIN_TMD_FIN code.
*
* The graph looks like:
*
* QOUT_TMD_START -> [ QIN_TMD_CONT -> QOUT_TMD_DONE ] * -> QIN_TMD_FIN.
*
*/
/*
* A word about ENABLE/DISABLE: the argument is a pointer to an tmd_cmd_t
* with cd_hba, cd_bus, cd_tgt and cd_lun filled out. If an error occurs
* in either enabling or disabling the described lun, cd_lflags is set
* with CDFL_ERROR.
*
* Logical unit zero must be the first enabled and the last disabled.
*/
/*
* Target handler functions.
* The MD target handler function (the outer layer calls this)
* should be be prototyped like:
*
* void target_action(qact_e, void *arg)
*
* The outer layer target handler function (the MD layer calls this)
* should be be prototyped like:
*
* void system_action(tact_e, void *arg)
*/
/*
* This structure is used to register to other software modules the
* binding of an HBA identifier, driver name and instance and the
* lun width capapbilities of this target driver. It's up to each
* platform to figure out how it wants to do this, but a typical
* sequence would be for the MD layer to find some external module's
* entry point and start by sending a QOUT_HBA_REG with info filled
* in, and the external module to call back with a QIN_HBA_REG that
* passes back the corresponding information.
*/
typedef struct {
void * r_identity;
char r_name[8];
int r_inst;
int r_lunwidth;
int r_buswidth;
void (*r_action)(int, void *);
} hba_register_t;