/* $OpenBSD: ncr53c9x.c,v 1.6 1998/05/28 22:07:52 jason Exp $ */ /* $NetBSD: ncr53c9x.c,v 1.26 1998/05/26 23:17:34 thorpej Exp $ */ /* * Copyright (c) 1996 Charles M. Hannum. 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. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by Charles M. Hannum. * 4. 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 ``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 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. */ /* * Copyright (c) 1994 Peter Galbavy * Copyright (c) 1995 Paul Kranenburg * 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. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by Peter Galbavy * 4. 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 ``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 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. */ /* * Based on aic6360 by Jarle Greipsland * * Acknowledgements: Many of the algorithms used in this driver are * inspired by the work of Julian Elischer (julian@tfs.com) and * Charles Hannum (mycroft@duality.gnu.ai.mit.edu). Thanks a million! */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include int ncr53c9x_debug = 0; /*NCR_SHOWPHASE|NCR_SHOWMISC|NCR_SHOWTRAC|NCR_SHOWCMDS;*/ /*static*/ void ncr53c9x_readregs __P((struct ncr53c9x_softc *)); /*static*/ void ncr53c9x_select __P((struct ncr53c9x_softc *, struct ncr53c9x_ecb *)); /*static*/ int ncr53c9x_reselect __P((struct ncr53c9x_softc *, int)); /*static*/ void ncr53c9x_scsi_reset __P((struct ncr53c9x_softc *)); /*static*/ void ncr53c9x_init __P((struct ncr53c9x_softc *, int)); /*static*/ int ncr53c9x_poll __P((struct ncr53c9x_softc *, struct scsi_xfer *, int)); /*static*/ void ncr53c9x_sched __P((struct ncr53c9x_softc *)); /*static*/ void ncr53c9x_done __P((struct ncr53c9x_softc *, struct ncr53c9x_ecb *)); /*static*/ void ncr53c9x_msgin __P((struct ncr53c9x_softc *)); /*static*/ void ncr53c9x_msgout __P((struct ncr53c9x_softc *)); /*static*/ void ncr53c9x_timeout __P((void *arg)); /*static*/ void ncr53c9x_abort __P((struct ncr53c9x_softc *, struct ncr53c9x_ecb *)); /*static*/ void ncr53c9x_dequeue __P((struct ncr53c9x_softc *, struct ncr53c9x_ecb *)); void ncr53c9x_sense __P((struct ncr53c9x_softc *, struct ncr53c9x_ecb *)); void ncr53c9x_free_ecb __P((struct ncr53c9x_softc *, struct ncr53c9x_ecb *, int)); struct ncr53c9x_ecb *ncr53c9x_get_ecb __P((struct ncr53c9x_softc *, int)); static inline int ncr53c9x_stp2cpb __P((struct ncr53c9x_softc *, int)); static inline void ncr53c9x_setsync __P((struct ncr53c9x_softc *, struct ncr53c9x_tinfo *)); /* * Names for the NCR53c9x variants, correspnding to the variant tags * in ncr53c9xvar.h. */ const char *ncr53c9x_variant_names[] = { "ESP100", "ESP100A", "ESP200", "NCR53C94", "NCR53C96", "ESP406", "FAS408", "FAS216", }; /* * Attach this instance, and then all the sub-devices */ void ncr53c9x_attach(sc, adapter, dev) struct ncr53c9x_softc *sc; struct scsi_adapter *adapter; struct scsi_device *dev; { /* * Allocate SCSI message buffers. * Front-ends can override allocation to avoid alignment * handling in the DMA engines. Note that that ncr53c9x_msgout() * can request a 1 byte DMA transfer. */ if (sc->sc_omess == NULL) sc->sc_omess = malloc(NCR_MAX_MSG_LEN, M_DEVBUF, M_NOWAIT); if (sc->sc_imess == NULL) sc->sc_imess = malloc(NCR_MAX_MSG_LEN+1, M_DEVBUF, M_NOWAIT); if (sc->sc_omess == NULL || sc->sc_imess == NULL) { printf("out of memory\n"); return; } /* * Note, the front-end has set us up to print the chip variation. */ if (sc->sc_rev >= NCR_VARIANT_MAX) { printf("\n%s: unknown variant %d, devices not attached\n", sc->sc_dev.dv_xname, sc->sc_rev); return; } printf(": %s, %dMHz, SCSI ID %d\n", ncr53c9x_variant_names[sc->sc_rev], sc->sc_freq, sc->sc_id); sc->sc_ccf = FREQTOCCF(sc->sc_freq); /* The value *must not* be == 1. Make it 2 */ if (sc->sc_ccf == 1) sc->sc_ccf = 2; /* * The recommended timeout is 250ms. This register is loaded * with a value calculated as follows, from the docs: * * (timout period) x (CLK frequency) * reg = ------------------------------------- * 8192 x (Clock Conversion Factor) * * Since CCF has a linear relation to CLK, this generally computes * to the constant of 153. */ sc->sc_timeout = ((250 * 1000) * sc->sc_freq) / (8192 * sc->sc_ccf); /* CCF register only has 3 bits; 0 is actually 8 */ sc->sc_ccf &= 7; /* Reset state & bus */ sc->sc_cfflags = sc->sc_dev.dv_cfdata->cf_flags; sc->sc_state = 0; ncr53c9x_init(sc, 1); /* * fill in the prototype scsi_link. */ sc->sc_link.adapter_softc = sc; sc->sc_link.adapter_target = sc->sc_id; sc->sc_link.adapter = adapter; sc->sc_link.device = dev; sc->sc_link.openings = 2; /* * Now try to attach all the sub-devices */ config_found(&sc->sc_dev, &sc->sc_link, scsiprint); /* * Enable interupts from the SCSI core */ if ((sc->sc_rev == NCR_VARIANT_ESP406) || (sc->sc_rev == NCR_VARIANT_FAS408)) { NCR_PIOREGS(sc); NCR_WRITE_REG(sc, NCR_CFG5, NCRCFG5_SINT | NCR_READ_REG(sc, NCR_CFG5)); NCR_SCSIREGS(sc); } } /* * This is the generic esp reset function. It does not reset the SCSI bus, * only this controllers, but kills any on-going commands, and also stops * and resets the DMA. * * After reset, registers are loaded with the defaults from the attach * routine above. */ void ncr53c9x_reset(sc) struct ncr53c9x_softc *sc; { /* reset DMA first */ NCRDMA_RESET(sc); /* reset SCSI chip */ NCRCMD(sc, NCRCMD_RSTCHIP); NCRCMD(sc, NCRCMD_NOP); DELAY(500); /* do these backwards, and fall through */ switch (sc->sc_rev) { case NCR_VARIANT_ESP406: case NCR_VARIANT_FAS408: NCR_SCSIREGS(sc); case NCR_VARIANT_FAS216: case NCR_VARIANT_NCR53C94: case NCR_VARIANT_NCR53C96: case NCR_VARIANT_ESP200: sc->sc_features |= NCR_F_HASCFG3; NCR_WRITE_REG(sc, NCR_CFG3, sc->sc_cfg3); case NCR_VARIANT_ESP100A: NCR_WRITE_REG(sc, NCR_CFG2, sc->sc_cfg2); case NCR_VARIANT_ESP100: NCR_WRITE_REG(sc, NCR_CFG1, sc->sc_cfg1); NCR_WRITE_REG(sc, NCR_CCF, sc->sc_ccf); NCR_WRITE_REG(sc, NCR_SYNCOFF, 0); NCR_WRITE_REG(sc, NCR_TIMEOUT, sc->sc_timeout); break; default: printf("%s: unknown revision code, assuming ESP100\n", sc->sc_dev.dv_xname); NCR_WRITE_REG(sc, NCR_CFG1, sc->sc_cfg1); NCR_WRITE_REG(sc, NCR_CCF, sc->sc_ccf); NCR_WRITE_REG(sc, NCR_SYNCOFF, 0); NCR_WRITE_REG(sc, NCR_TIMEOUT, sc->sc_timeout); } } /* * Reset the SCSI bus, but not the chip */ void ncr53c9x_scsi_reset(sc) struct ncr53c9x_softc *sc; { (*sc->sc_glue->gl_dma_stop)(sc); printf("%s: resetting SCSI bus\n", sc->sc_dev.dv_xname); NCRCMD(sc, NCRCMD_RSTSCSI); } /* * Initialize esp state machine */ void ncr53c9x_init(sc, doreset) struct ncr53c9x_softc *sc; int doreset; { struct ncr53c9x_ecb *ecb; int r; NCR_TRACE(("[NCR_INIT(%d)] ", doreset)); if (sc->sc_state == 0) { /* First time through; initialize. */ TAILQ_INIT(&sc->ready_list); TAILQ_INIT(&sc->nexus_list); TAILQ_INIT(&sc->free_list); sc->sc_nexus = NULL; ecb = sc->sc_ecb; bzero(ecb, sizeof(sc->sc_ecb)); for (r = 0; r < sizeof(sc->sc_ecb) / sizeof(*ecb); r++) { TAILQ_INSERT_TAIL(&sc->free_list, ecb, chain); ecb++; } bzero(sc->sc_tinfo, sizeof(sc->sc_tinfo)); } else { /* Cancel any active commands. */ sc->sc_state = NCR_CLEANING; if ((ecb = sc->sc_nexus) != NULL) { ecb->xs->error = XS_TIMEOUT; ncr53c9x_done(sc, ecb); } while ((ecb = sc->nexus_list.tqh_first) != NULL) { ecb->xs->error = XS_TIMEOUT; ncr53c9x_done(sc, ecb); } } /* * reset the chip to a known state */ ncr53c9x_reset(sc); sc->sc_phase = sc->sc_prevphase = INVALID_PHASE; for (r = 0; r < 8; r++) { struct ncr53c9x_tinfo *ti = &sc->sc_tinfo[r]; /* XXX - config flags per target: low bits: no reselect; high bits: no synch */ ti->flags = ((sc->sc_minsync && !(sc->sc_cfflags & (1<<(r+8)))) ? T_NEGOTIATE : 0) | ((sc->sc_cfflags & (1<period = sc->sc_minsync; ti->offset = 0; } if (doreset) { sc->sc_state = NCR_SBR; NCRCMD(sc, NCRCMD_RSTSCSI); } else { sc->sc_state = NCR_IDLE; ncr53c9x_sched(sc); } } /* * Read the NCR registers, and save their contents for later use. * NCR_STAT, NCR_STEP & NCR_INTR are mostly zeroed out when reading * NCR_INTR - so make sure it is the last read. * * I think that (from reading the docs) most bits in these registers * only make sense when he DMA CSR has an interrupt showing. Call only * if an interrupt is pending. */ __inline__ void ncr53c9x_readregs(sc) struct ncr53c9x_softc *sc; { sc->sc_espstat = NCR_READ_REG(sc, NCR_STAT); /* Only the stepo bits are of interest */ sc->sc_espstep = NCR_READ_REG(sc, NCR_STEP) & NCRSTEP_MASK; sc->sc_espintr = NCR_READ_REG(sc, NCR_INTR); if (sc->sc_glue->gl_clear_latched_intr != NULL) (*sc->sc_glue->gl_clear_latched_intr)(sc); /* * Determine the SCSI bus phase, return either a real SCSI bus phase * or some pseudo phase we use to detect certain exceptions. */ sc->sc_phase = (sc->sc_espintr & NCRINTR_DIS) ? /* Disconnected */ BUSFREE_PHASE : sc->sc_espstat & NCRSTAT_PHASE; NCR_MISC(("regs[intr=%02x,stat=%02x,step=%02x] ", sc->sc_espintr, sc->sc_espstat, sc->sc_espstep)); } /* * Convert Synchronous Transfer Period to chip register Clock Per Byte value. */ static inline int ncr53c9x_stp2cpb(sc, period) struct ncr53c9x_softc *sc; int period; { int v; v = (sc->sc_freq * period) / 250; if (ncr53c9x_cpb2stp(sc, v) < period) /* Correct round-down error */ v++; return (v); } static inline void ncr53c9x_setsync(sc, ti) struct ncr53c9x_softc *sc; struct ncr53c9x_tinfo *ti; { u_char syncoff, synctp, cfg3 = sc->sc_cfg3; if (ti->flags & T_SYNCMODE) { syncoff = ti->offset; synctp = ncr53c9x_stp2cpb(sc, ti->period); if (sc->sc_features & NCR_F_FASTSCSI) { /* * If the period is 200ns or less (ti->period <= 50), * put the chip in Fast SCSI mode. */ if (ti->period <= 50) cfg3 |= NCRCFG3_FSCSI; } } else { syncoff = 0; synctp = 0; } if (sc->sc_features & NCR_F_HASCFG3) NCR_WRITE_REG(sc, NCR_CFG3, cfg3); NCR_WRITE_REG(sc, NCR_SYNCOFF, syncoff); NCR_WRITE_REG(sc, NCR_SYNCTP, synctp); } int ncr53c9x_dmaselect = 0; /* * Send a command to a target, set the driver state to NCR_SELECTING * and let the caller take care of the rest. * * Keeping this as a function allows me to say that this may be done * by DMA instead of programmed I/O soon. */ void ncr53c9x_select(sc, ecb) struct ncr53c9x_softc *sc; struct ncr53c9x_ecb *ecb; { struct scsi_link *sc_link = ecb->xs->sc_link; int target = sc_link->target; int lun = sc_link->lun; struct ncr53c9x_tinfo *ti = &sc->sc_tinfo[target]; int tiflags = ti->flags; u_char *cmd; int clen; NCR_TRACE(("[ncr53c9x_select(t%d,l%d,cmd:%x)] ", target, lun, ecb->cmd.cmd.opcode)); sc->sc_state = NCR_SELECTING; /* * Schedule the timeout now, the first time we will go away * expecting to come back due to an interrupt, because it is * always possible that the interrupt may never happen. */ if ((ecb->xs->flags & SCSI_POLL) == 0) timeout(ncr53c9x_timeout, ecb, (ecb->timeout * hz) / 1000); /* * The docs say the target register is never reset, and I * can't think of a better place to set it */ NCR_WRITE_REG(sc, NCR_SELID, target); ncr53c9x_setsync(sc, ti); if (ncr53c9x_dmaselect && (tiflags & T_NEGOTIATE) == 0) { size_t dmasize; ecb->cmd.id = MSG_IDENTIFY(lun, (tiflags & T_RSELECTOFF)?0:1); /* setup DMA transfer for command */ dmasize = clen = ecb->clen + 1; sc->sc_cmdlen = clen; sc->sc_cmdp = (caddr_t)&ecb->cmd; NCRDMA_SETUP(sc, &sc->sc_cmdp, &sc->sc_cmdlen, 0, &dmasize); /* Program the SCSI counter */ NCR_WRITE_REG(sc, NCR_TCL, dmasize); NCR_WRITE_REG(sc, NCR_TCM, dmasize >> 8); if (sc->sc_cfg2 & NCRCFG2_FE) { NCR_WRITE_REG(sc, NCR_TCH, dmasize >> 16); } /* load the count in */ NCRCMD(sc, NCRCMD_NOP|NCRCMD_DMA); /* And get the targets attention */ NCRCMD(sc, NCRCMD_SELATN | NCRCMD_DMA); NCRDMA_GO(sc); return; } /* * Who am I. This is where we tell the target that we are * happy for it to disconnect etc. */ NCR_WRITE_REG(sc, NCR_FIFO, MSG_IDENTIFY(lun, (tiflags & T_RSELECTOFF)?0:1)); if (ti->flags & T_NEGOTIATE) { /* Arbitrate, select and stop after IDENTIFY message */ NCRCMD(sc, NCRCMD_SELATNS); return; } /* Now the command into the FIFO */ cmd = (u_char *)&ecb->cmd.cmd; clen = ecb->clen; while (clen--) NCR_WRITE_REG(sc, NCR_FIFO, *cmd++); /* And get the targets attention */ NCRCMD(sc, NCRCMD_SELATN); } void ncr53c9x_free_ecb(sc, ecb, flags) struct ncr53c9x_softc *sc; struct ncr53c9x_ecb *ecb; int flags; { int s; s = splbio(); ecb->flags = 0; TAILQ_INSERT_HEAD(&sc->free_list, ecb, chain); /* * If there were none, wake anybody waiting for one to come free, * starting with queued entries. */ if (ecb->chain.tqe_next == 0) wakeup(&sc->free_list); splx(s); } struct ncr53c9x_ecb * ncr53c9x_get_ecb(sc, flags) struct ncr53c9x_softc *sc; int flags; { struct ncr53c9x_ecb *ecb; int s; s = splbio(); while ((ecb = sc->free_list.tqh_first) == NULL && (flags & SCSI_NOSLEEP) == 0) tsleep(&sc->free_list, PRIBIO, "especb", 0); if (ecb) { TAILQ_REMOVE(&sc->free_list, ecb, chain); ecb->flags |= ECB_ALLOC; } splx(s); return (ecb); } /* * DRIVER FUNCTIONS CALLABLE FROM HIGHER LEVEL DRIVERS */ /* * Start a SCSI-command * This function is called by the higher level SCSI-driver to queue/run * SCSI-commands. */ int ncr53c9x_scsi_cmd(xs) struct scsi_xfer *xs; { struct scsi_link *sc_link = xs->sc_link; struct ncr53c9x_softc *sc = sc_link->adapter_softc; struct ncr53c9x_ecb *ecb; int s, flags; NCR_TRACE(("[ncr53c9x_scsi_cmd] ")); NCR_CMDS(("[0x%x, %d]->%d ", (int)xs->cmd->opcode, xs->cmdlen, sc_link->target)); flags = xs->flags; if ((ecb = ncr53c9x_get_ecb(sc, flags)) == NULL) return (TRY_AGAIN_LATER); /* Initialize ecb */ ecb->xs = xs; ecb->timeout = xs->timeout; if (flags & SCSI_RESET) { ecb->flags |= ECB_RESET; ecb->clen = 0; ecb->dleft = 0; } else { bcopy(xs->cmd, &ecb->cmd.cmd, xs->cmdlen); ecb->clen = xs->cmdlen; ecb->daddr = xs->data; ecb->dleft = xs->datalen; } ecb->stat = 0; s = splbio(); TAILQ_INSERT_TAIL(&sc->ready_list, ecb, chain); if (sc->sc_state == NCR_IDLE) ncr53c9x_sched(sc); splx(s); if ((flags & SCSI_POLL) == 0) return (SUCCESSFULLY_QUEUED); /* Not allowed to use interrupts, use polling instead */ if (ncr53c9x_poll(sc, xs, ecb->timeout)) { ncr53c9x_timeout(ecb); if (ncr53c9x_poll(sc, xs, ecb->timeout)) ncr53c9x_timeout(ecb); } return (COMPLETE); } /* * Used when interrupt driven I/O isn't allowed, e.g. during boot. */ int ncr53c9x_poll(sc, xs, count) struct ncr53c9x_softc *sc; struct scsi_xfer *xs; int count; { NCR_TRACE(("[ncr53c9x_poll] ")); while (count) { if (NCRDMA_ISINTR(sc)) { ncr53c9x_intr(sc); } #if alternatively if (NCR_READ_REG(sc, NCR_STAT) & NCRSTAT_INT) ncr53c9x_intr(sc); #endif if ((xs->flags & ITSDONE) != 0) return (0); if (sc->sc_state == NCR_IDLE) { NCR_TRACE(("[ncr53c9x_poll: rescheduling] ")); ncr53c9x_sched(sc); } DELAY(1000); count--; } return (1); } /* * LOW LEVEL SCSI UTILITIES */ /* * Schedule a scsi operation. This has now been pulled out of the interrupt * handler so that we may call it from ncr53c9x_scsi_cmd and ncr53c9x_done. * This may save us an unecessary interrupt just to get things going. * Should only be called when state == NCR_IDLE and at bio pl. */ void ncr53c9x_sched(sc) struct ncr53c9x_softc *sc; { struct ncr53c9x_ecb *ecb; struct scsi_link *sc_link; struct ncr53c9x_tinfo *ti; NCR_TRACE(("[ncr53c9x_sched] ")); if (sc->sc_state != NCR_IDLE) panic("ncr53c9x_sched: not IDLE (state=%d)", sc->sc_state); /* * Find first ecb in ready queue that is for a target/lunit * combinations that is not busy. */ for (ecb = sc->ready_list.tqh_first; ecb; ecb = ecb->chain.tqe_next) { sc_link = ecb->xs->sc_link; ti = &sc->sc_tinfo[sc_link->target]; if ((ti->lubusy & (1 << sc_link->lun)) == 0) { TAILQ_REMOVE(&sc->ready_list, ecb, chain); sc->sc_nexus = ecb; ncr53c9x_select(sc, ecb); break; } else NCR_MISC(("%d:%d busy\n", sc_link->target, sc_link->lun)); } } void ncr53c9x_sense(sc, ecb) struct ncr53c9x_softc *sc; struct ncr53c9x_ecb *ecb; { struct scsi_xfer *xs = ecb->xs; struct scsi_link *sc_link = xs->sc_link; struct ncr53c9x_tinfo *ti = &sc->sc_tinfo[sc_link->target]; struct scsi_sense *ss = (void *)&ecb->cmd.cmd; NCR_MISC(("requesting sense ")); /* Next, setup a request sense command block */ bzero(ss, sizeof(*ss)); ss->opcode = REQUEST_SENSE; ss->byte2 = sc_link->lun << 5; ss->length = sizeof(struct scsi_sense_data); ecb->clen = sizeof(*ss); ecb->daddr = (char *)&xs->sense; ecb->dleft = sizeof(struct scsi_sense_data); ecb->flags |= ECB_SENSE; ecb->timeout = NCR_SENSE_TIMEOUT; ti->senses++; if (ecb->flags & ECB_NEXUS) ti->lubusy &= ~(1 << sc_link->lun); if (ecb == sc->sc_nexus) { ecb->flags &= ~ECB_NEXUS; ncr53c9x_select(sc, ecb); } else { ncr53c9x_dequeue(sc, ecb); TAILQ_INSERT_HEAD(&sc->ready_list, ecb, chain); if (sc->sc_state == NCR_IDLE) ncr53c9x_sched(sc); } } /* * POST PROCESSING OF SCSI_CMD (usually current) */ void ncr53c9x_done(sc, ecb) struct ncr53c9x_softc *sc; struct ncr53c9x_ecb *ecb; { struct scsi_xfer *xs = ecb->xs; struct scsi_link *sc_link = xs->sc_link; struct ncr53c9x_tinfo *ti = &sc->sc_tinfo[sc_link->target]; NCR_TRACE(("[ncr53c9x_done(error:%x)] ", xs->error)); untimeout(ncr53c9x_timeout, ecb); /* * Now, if we've come here with no error code, i.e. we've kept the * initial XS_NOERROR, and the status code signals that we should * check sense, we'll need to set up a request sense cmd block and * push the command back into the ready queue *before* any other * commands for this target/lunit, else we lose the sense info. * We don't support chk sense conditions for the request sense cmd. */ if (xs->error == XS_NOERROR) { xs->status = ecb->stat; if ((ecb->flags & ECB_ABORT) != 0) { xs->error = XS_TIMEOUT; } else if ((ecb->flags & ECB_SENSE) != 0) { xs->error = XS_SENSE; } else if ((ecb->stat & ST_MASK) == SCSI_CHECK) { /* First, save the return values */ xs->resid = ecb->dleft; ncr53c9x_sense(sc, ecb); return; } else { xs->resid = ecb->dleft; } } xs->flags |= ITSDONE; #ifdef NCR53C9X_DEBUG if (ncr53c9x_debug & NCR_SHOWMISC) { if (xs->resid != 0) printf("resid=%d ", xs->resid); if (xs->error == XS_SENSE) printf("sense=0x%02x\n", xs->sense.error_code); else printf("error=%d\n", xs->error); } #endif /* * Remove the ECB from whatever queue it's on. */ if (ecb->flags & ECB_NEXUS) ti->lubusy &= ~(1 << sc_link->lun); if (ecb == sc->sc_nexus) { sc->sc_nexus = NULL; if (sc->sc_state != NCR_CLEANING) { sc->sc_state = NCR_IDLE; ncr53c9x_sched(sc); } } else ncr53c9x_dequeue(sc, ecb); ncr53c9x_free_ecb(sc, ecb, xs->flags); ti->cmds++; scsi_done(xs); } void ncr53c9x_dequeue(sc, ecb) struct ncr53c9x_softc *sc; struct ncr53c9x_ecb *ecb; { if (ecb->flags & ECB_NEXUS) { TAILQ_REMOVE(&sc->nexus_list, ecb, chain); ecb->flags &= ~ECB_NEXUS; } else { TAILQ_REMOVE(&sc->ready_list, ecb, chain); } } /* * INTERRUPT/PROTOCOL ENGINE */ /* * Schedule an outgoing message by prioritizing it, and asserting * attention on the bus. We can only do this when we are the initiator * else there will be an illegal command interrupt. */ #define ncr53c9x_sched_msgout(m) \ do { \ NCR_MISC(("ncr53c9x_sched_msgout %d ", m)); \ NCRCMD(sc, NCRCMD_SETATN); \ sc->sc_flags |= NCR_ATN; \ sc->sc_msgpriq |= (m); \ } while (0) int ncr53c9x_reselect(sc, message) struct ncr53c9x_softc *sc; int message; { u_char selid, target, lun; struct ncr53c9x_ecb *ecb; struct scsi_link *sc_link; struct ncr53c9x_tinfo *ti; /* * The SCSI chip made a snapshot of the data bus while the reselection * was being negotiated. This enables us to determine which target did * the reselect. */ selid = sc->sc_selid & ~(1 << sc->sc_id); if (selid & (selid - 1)) { printf("%s: reselect with invalid selid %02x;" " sending DEVICE RESET\n", sc->sc_dev.dv_xname, selid); goto reset; } /* * Search wait queue for disconnected cmd * The list should be short, so I haven't bothered with * any more sophisticated structures than a simple * singly linked list. */ target = ffs(selid) - 1; lun = message & 0x07; for (ecb = sc->nexus_list.tqh_first; ecb != NULL; ecb = ecb->chain.tqe_next) { sc_link = ecb->xs->sc_link; if (sc_link->target == target && sc_link->lun == lun) break; } if (ecb == NULL) { printf("%s: reselect from target %d lun %d with no nexus;" " sending ABORT\n", sc->sc_dev.dv_xname, target, lun); goto abort; } /* Make this nexus active again. */ TAILQ_REMOVE(&sc->nexus_list, ecb, chain); sc->sc_state = NCR_CONNECTED; sc->sc_nexus = ecb; ti = &sc->sc_tinfo[target]; #ifdef NCR53C9X_DEBUG if ((ti->lubusy & (1 << lun)) == 0) { printf("%s: reselect: target %d, lun %d: should be busy\n", sc->sc_dev.dv_xname, target, lun); ti->lubusy |= (1 << lun); } #endif ncr53c9x_setsync(sc, ti); if (ecb->flags & ECB_RESET) ncr53c9x_sched_msgout(SEND_DEV_RESET); else if (ecb->flags & ECB_ABORT) ncr53c9x_sched_msgout(SEND_ABORT); /* Do an implicit RESTORE POINTERS. */ sc->sc_dp = ecb->daddr; sc->sc_dleft = ecb->dleft; return (0); reset: ncr53c9x_sched_msgout(SEND_DEV_RESET); return (1); abort: ncr53c9x_sched_msgout(SEND_ABORT); return (1); } #define IS1BYTEMSG(m) (((m) != 1 && (m) < 0x20) || (m) & 0x80) #define IS2BYTEMSG(m) (((m) & 0xf0) == 0x20) #define ISEXTMSG(m) ((m) == 1) /* * Get an incoming message as initiator. * * The SCSI bus must already be in MESSAGE_IN_PHASE and there is a * byte in the FIFO */ void ncr53c9x_msgin(sc) register struct ncr53c9x_softc *sc; { register int v; NCR_TRACE(("[ncr53c9x_msgin(curmsglen:%ld)] ", (long)sc->sc_imlen)); if ((NCR_READ_REG(sc, NCR_FFLAG) & NCRFIFO_FF) == 0) { printf("%s: msgin: no msg byte available\n", sc->sc_dev.dv_xname); return; } /* * Prepare for a new message. A message should (according * to the SCSI standard) be transmitted in one single * MESSAGE_IN_PHASE. If we have been in some other phase, * then this is a new message. */ if (sc->sc_prevphase != MESSAGE_IN_PHASE) { sc->sc_flags &= ~NCR_DROP_MSGI; sc->sc_imlen = 0; } v = NCR_READ_REG(sc, NCR_FIFO); NCR_MISC(("", v)); #if 0 if (sc->sc_state == NCR_RESELECTED && sc->sc_imlen == 0) { /* * Which target is reselecting us? (The ID bit really) */ sc->sc_selid = v; NCR_MISC(("selid=0x%2x ", sc->sc_selid)); return; } #endif sc->sc_imess[sc->sc_imlen] = v; /* * If we're going to reject the message, don't bother storing * the incoming bytes. But still, we need to ACK them. */ if ((sc->sc_flags & NCR_DROP_MSGI)) { NCRCMD(sc, NCRCMD_MSGOK); printf("", sc->sc_imess[sc->sc_imlen]); return; } if (sc->sc_imlen >= NCR_MAX_MSG_LEN) { ncr53c9x_sched_msgout(SEND_REJECT); sc->sc_flags |= NCR_DROP_MSGI; } else { sc->sc_imlen++; /* * This testing is suboptimal, but most * messages will be of the one byte variety, so * it should not effect performance * significantly. */ if (sc->sc_imlen == 1 && IS1BYTEMSG(sc->sc_imess[0])) goto gotit; if (sc->sc_imlen == 2 && IS2BYTEMSG(sc->sc_imess[0])) goto gotit; if (sc->sc_imlen >= 3 && ISEXTMSG(sc->sc_imess[0]) && sc->sc_imlen == sc->sc_imess[1] + 2) goto gotit; } /* Ack what we have so far */ NCRCMD(sc, NCRCMD_MSGOK); return; gotit: NCR_MSGS(("gotmsg(%x)", sc->sc_imess[0])); /* * Now we should have a complete message (1 byte, 2 byte * and moderately long extended messages). We only handle * extended messages which total length is shorter than * NCR_MAX_MSG_LEN. Longer messages will be amputated. */ switch (sc->sc_state) { struct ncr53c9x_ecb *ecb; struct ncr53c9x_tinfo *ti; case NCR_CONNECTED: ecb = sc->sc_nexus; ti = &sc->sc_tinfo[ecb->xs->sc_link->target]; switch (sc->sc_imess[0]) { case MSG_CMDCOMPLETE: NCR_MSGS(("cmdcomplete ")); if (sc->sc_dleft < 0) { struct scsi_link *sc_link = ecb->xs->sc_link; printf("%s: %ld extra bytes from %d:%d\n", sc->sc_dev.dv_xname, -(long)sc->sc_dleft, sc_link->target, sc_link->lun); sc->sc_dleft = 0; } ecb->dleft = (ecb->flags & ECB_TENTATIVE_DONE) ? 0 : sc->sc_dleft; if ((ecb->flags & ECB_SENSE) == 0) ecb->xs->resid = ecb->dleft; sc->sc_state = NCR_CMDCOMPLETE; break; case MSG_MESSAGE_REJECT: NCR_MSGS(("msg reject (msgout=%x) ", sc->sc_msgout)); switch (sc->sc_msgout) { case SEND_SDTR: sc->sc_flags &= ~NCR_SYNCHNEGO; ti->flags &= ~(T_NEGOTIATE | T_SYNCMODE); ncr53c9x_setsync(sc, ti); break; case SEND_INIT_DET_ERR: goto abort; } break; case MSG_NOOP: NCR_MSGS(("noop ")); break; case MSG_DISCONNECT: NCR_MSGS(("disconnect ")); ti->dconns++; sc->sc_state = NCR_DISCONNECT; /* * Mark the fact that all bytes have moved. The * target may not bother to do a SAVE POINTERS * at this stage. This flag will set the residual * count to zero on MSG COMPLETE. */ if (sc->sc_dleft == 0) ecb->flags |= ECB_TENTATIVE_DONE; break; case MSG_SAVEDATAPOINTER: NCR_MSGS(("save datapointer ")); ecb->daddr = sc->sc_dp; ecb->dleft = sc->sc_dleft; break; case MSG_RESTOREPOINTERS: NCR_MSGS(("restore datapointer ")); sc->sc_dp = ecb->daddr; sc->sc_dleft = ecb->dleft; break; case MSG_EXTENDED: NCR_MSGS(("extended(%x) ", sc->sc_imess[2])); switch (sc->sc_imess[2]) { case MSG_EXT_SDTR: NCR_MSGS(("SDTR period %d, offset %d ", sc->sc_imess[3], sc->sc_imess[4])); if (sc->sc_imess[1] != 3) goto reject; ti->period = sc->sc_imess[3]; ti->offset = sc->sc_imess[4]; ti->flags &= ~T_NEGOTIATE; if (sc->sc_minsync == 0 || ti->offset == 0 || ti->period > 124) { printf("%s:%d: async\n", "esp", ecb->xs->sc_link->target); if ((sc->sc_flags&NCR_SYNCHNEGO) == 0) { /* * target initiated negotiation */ ti->offset = 0; ti->flags &= ~T_SYNCMODE; ncr53c9x_sched_msgout( SEND_SDTR); } else { /* we are async */ ti->flags &= ~T_SYNCMODE; } } else { int r = 250/ti->period; int s = (100*250)/ti->period - 100*r; int p; p = ncr53c9x_stp2cpb(sc, ti->period); ti->period = ncr53c9x_cpb2stp(sc, p); #ifdef NCR53C9X_DEBUG sc_print_addr(ecb->xs->sc_link); printf("max sync rate %d.%02dMb/s\n", r, s); #endif if ((sc->sc_flags&NCR_SYNCHNEGO) == 0) { /* * target initiated negotiation */ if (ti->period < sc->sc_minsync) ti->period = sc->sc_minsync; if (ti->offset > 15) ti->offset = 15; ti->flags &= ~T_SYNCMODE; ncr53c9x_sched_msgout( SEND_SDTR); } else { /* we are sync */ ti->flags |= T_SYNCMODE; } } sc->sc_flags &= ~NCR_SYNCHNEGO; ncr53c9x_setsync(sc, ti); break; default: printf("%s: unrecognized MESSAGE EXTENDED;" " sending REJECT\n", sc->sc_dev.dv_xname); goto reject; } break; default: NCR_MSGS(("ident ")); printf("%s: unrecognized MESSAGE; sending REJECT\n", sc->sc_dev.dv_xname); reject: ncr53c9x_sched_msgout(SEND_REJECT); break; } break; case NCR_RESELECTED: if (!MSG_ISIDENTIFY(sc->sc_imess[0])) { printf("%s: reselect without IDENTIFY;" " sending DEVICE RESET\n", sc->sc_dev.dv_xname); goto reset; } (void) ncr53c9x_reselect(sc, sc->sc_imess[0]); break; default: printf("%s: unexpected MESSAGE IN; sending DEVICE RESET\n", sc->sc_dev.dv_xname); reset: ncr53c9x_sched_msgout(SEND_DEV_RESET); break; abort: ncr53c9x_sched_msgout(SEND_ABORT); break; } /* Ack last message byte */ NCRCMD(sc, NCRCMD_MSGOK); /* Done, reset message pointer. */ sc->sc_flags &= ~NCR_DROP_MSGI; sc->sc_imlen = 0; } /* * Send the highest priority, scheduled message */ void ncr53c9x_msgout(sc) register struct ncr53c9x_softc *sc; { struct ncr53c9x_tinfo *ti; struct ncr53c9x_ecb *ecb; size_t size; NCR_TRACE(("[ncr53c9x_msgout(priq:%x, prevphase:%x)]", sc->sc_msgpriq, sc->sc_prevphase)); /* * XXX - the NCR_ATN flag is not in sync with the actual ATN * condition on the SCSI bus. The 53c9x chip * automatically turns off ATN before sending the * message byte. (see also the comment below in the * default case when picking out a message to send) */ if (sc->sc_flags & NCR_ATN) { if (sc->sc_prevphase != MESSAGE_OUT_PHASE) { new: NCRCMD(sc, NCRCMD_FLUSH); DELAY(1); sc->sc_msgoutq = 0; sc->sc_omlen = 0; } } else { if (sc->sc_prevphase == MESSAGE_OUT_PHASE) { ncr53c9x_sched_msgout(sc->sc_msgoutq); goto new; } else { printf("%s at line %d: unexpected MESSAGE OUT phase\n", sc->sc_dev.dv_xname, __LINE__); } } if (sc->sc_omlen == 0) { /* Pick up highest priority message */ sc->sc_msgout = sc->sc_msgpriq & -sc->sc_msgpriq; sc->sc_msgoutq |= sc->sc_msgout; sc->sc_msgpriq &= ~sc->sc_msgout; sc->sc_omlen = 1; /* "Default" message len */ switch (sc->sc_msgout) { case SEND_SDTR: ecb = sc->sc_nexus; ti = &sc->sc_tinfo[ecb->xs->sc_link->target]; sc->sc_omess[0] = MSG_EXTENDED; sc->sc_omess[1] = 3; sc->sc_omess[2] = MSG_EXT_SDTR; sc->sc_omess[3] = ti->period; sc->sc_omess[4] = ti->offset; sc->sc_omlen = 5; if ((sc->sc_flags & NCR_SYNCHNEGO) == 0) { ti->flags |= T_SYNCMODE; ncr53c9x_setsync(sc, ti); } break; case SEND_IDENTIFY: if (sc->sc_state != NCR_CONNECTED) { printf("%s at line %d: no nexus\n", sc->sc_dev.dv_xname, __LINE__); } ecb = sc->sc_nexus; sc->sc_omess[0] = MSG_IDENTIFY(ecb->xs->sc_link->lun, 0); break; case SEND_DEV_RESET: sc->sc_flags |= NCR_ABORTING; sc->sc_omess[0] = MSG_BUS_DEV_RESET; ecb = sc->sc_nexus; ti = &sc->sc_tinfo[ecb->xs->sc_link->target]; ti->flags &= ~T_SYNCMODE; ti->flags |= T_NEGOTIATE; break; case SEND_PARITY_ERROR: sc->sc_omess[0] = MSG_PARITY_ERROR; break; case SEND_ABORT: sc->sc_flags |= NCR_ABORTING; sc->sc_omess[0] = MSG_ABORT; break; case SEND_INIT_DET_ERR: sc->sc_omess[0] = MSG_INITIATOR_DET_ERR; break; case SEND_REJECT: sc->sc_omess[0] = MSG_MESSAGE_REJECT; break; default: /* * We normally do not get here, since the chip * automatically turns off ATN before the last * byte of a message is sent to the target. * However, if the target rejects our (multi-byte) * message early by switching to MSG IN phase * ATN remains on, so the target may return to * MSG OUT phase. If there are no scheduled messages * left we send a NO-OP. * * XXX - Note that this leaves no useful purpose for * the NCR_ATN flag. */ sc->sc_flags &= ~NCR_ATN; sc->sc_omess[0] = MSG_NOOP; break; } sc->sc_omp = sc->sc_omess; } /* (re)send the message */ size = min(sc->sc_omlen, sc->sc_maxxfer); NCRDMA_SETUP(sc, &sc->sc_omp, &sc->sc_omlen, 0, &size); /* Program the SCSI counter */ NCR_WRITE_REG(sc, NCR_TCL, size); NCR_WRITE_REG(sc, NCR_TCM, size >> 8); if (sc->sc_cfg2 & NCRCFG2_FE) { NCR_WRITE_REG(sc, NCR_TCH, size >> 16); } /* Load the count in and start the message-out transfer */ NCRCMD(sc, NCRCMD_NOP|NCRCMD_DMA); NCRCMD(sc, NCRCMD_TRANS|NCRCMD_DMA); NCRDMA_GO(sc); } /* * This is the most critical part of the driver, and has to know * how to deal with *all* error conditions and phases from the SCSI * bus. If there are no errors and the DMA was active, then call the * DMA pseudo-interrupt handler. If this returns 1, then that was it * and we can return from here without further processing. * * Most of this needs verifying. */ int sdebug = 0; int ncr53c9x_intr(sc) register struct ncr53c9x_softc *sc; { register struct ncr53c9x_ecb *ecb; register struct scsi_link *sc_link; struct ncr53c9x_tinfo *ti; size_t size; int nfifo; NCR_TRACE(("[ncr53c9x_intr] ")); if (!NCRDMA_ISINTR(sc)) return (0); again: /* and what do the registers day... */ ncr53c9x_readregs(sc); sc->sc_intrcnt.ev_count++; /* * Command are classed as errors. A disconnect is a * valid condition, and we let the code check is the * "NCR_BUSFREE_OK" flag was set before declaring it * and error. * * Also, the status register tells us about "Gross * Errors" and "Parity errors". Only the Gross Error * is really bad, and the parity errors are dealt * with later * * TODO * If there are too many parity error, go to slow * cable mode ? */ /* SCSI Reset */ if (sc->sc_espintr & NCRINTR_SBR) { if (NCR_READ_REG(sc, NCR_FFLAG) & NCRFIFO_FF) { NCRCMD(sc, NCRCMD_FLUSH); DELAY(1); } if (sc->sc_state != NCR_SBR) { printf("%s: SCSI bus reset\n", sc->sc_dev.dv_xname); ncr53c9x_init(sc, 0); /* Restart everything */ return (1); } #if 0 /*XXX*/ printf("\n", sc->sc_espintr, sc->sc_espstat, sc->sc_espstep); #endif if (sc->sc_nexus) panic("%s: nexus in reset state", sc->sc_dev.dv_xname); goto sched; } ecb = sc->sc_nexus; #define NCRINTR_ERR (NCRINTR_SBR|NCRINTR_ILL) if (sc->sc_espintr & NCRINTR_ERR || sc->sc_espstat & NCRSTAT_GE) { if (sc->sc_espstat & NCRSTAT_GE) { /* Gross Error; no target ? */ if (NCR_READ_REG(sc, NCR_FFLAG) & NCRFIFO_FF) { NCRCMD(sc, NCRCMD_FLUSH); DELAY(1); } if (sc->sc_state == NCR_CONNECTED || sc->sc_state == NCR_SELECTING) { ecb->xs->error = XS_TIMEOUT; ncr53c9x_done(sc, ecb); } return (1); } if (sc->sc_espintr & NCRINTR_ILL) { if (sc->sc_flags & NCR_EXPECT_ILLCMD) { /* * Eat away "Illegal command" interrupt * on a ESP100 caused by a re-selection * while we were trying to select * another target. */ #ifdef DEBUG printf("%s: ESP100 work-around activated\n", sc->sc_dev.dv_xname); #endif sc->sc_flags &= ~NCR_EXPECT_ILLCMD; return (1); } /* illegal command, out of sync ? */ printf("%s: illegal command: 0x%x " "(state %d, phase %x, prevphase %x)\n", sc->sc_dev.dv_xname, sc->sc_lastcmd, sc->sc_state, sc->sc_phase, sc->sc_prevphase); if (NCR_READ_REG(sc, NCR_FFLAG) & NCRFIFO_FF) { NCRCMD(sc, NCRCMD_FLUSH); DELAY(1); } ncr53c9x_init(sc, 1); /* Restart everything */ return (1); } } sc->sc_flags &= ~NCR_EXPECT_ILLCMD; /* * Call if DMA is active. * * If DMA_INTR returns true, then maybe go 'round the loop * again in case there is no more DMA queued, but a phase * change is expected. */ if (NCRDMA_ISACTIVE(sc)) { int r = NCRDMA_INTR(sc); if (r == -1) { printf("%s: DMA error; resetting\n", sc->sc_dev.dv_xname); ncr53c9x_init(sc, 1); } /* If DMA active here, then go back to work... */ if (NCRDMA_ISACTIVE(sc)) return (1); if ((sc->sc_espstat & NCRSTAT_TC) == 0) { /* * DMA not completed. If we can not find a * acceptable explanation, print a diagnostic. */ if (sc->sc_state == NCR_SELECTING) /* * This can happen if we are reselected * while using DMA to select a target. */ /*void*/; else if (sc->sc_prevphase == MESSAGE_OUT_PHASE){ /* * Our (multi-byte) message (eg SDTR) * was interrupted by the target to * send a MSG REJECT. * Print diagnostic if current phase * is not MESSAGE IN. */ if (sc->sc_phase != MESSAGE_IN_PHASE) printf("%s: !TC on MSG OUT" " [intr %x, stat %x, step %d]" " prevphase %x, resid %x\n", sc->sc_dev.dv_xname, sc->sc_espintr, sc->sc_espstat, sc->sc_espstep, sc->sc_prevphase, sc->sc_omlen); } else if (sc->sc_dleft == 0) { /* * The DMA operation was started for * a DATA transfer. Print a diagnostic * if the DMA counter and TC bit * appear to be out of sync. */ printf("%s: !TC on DATA XFER" " [intr %x, stat %x, step %d]" " prevphase %x, resid %x\n", sc->sc_dev.dv_xname, sc->sc_espintr, sc->sc_espstat, sc->sc_espstep, sc->sc_prevphase, ecb?ecb->dleft:-1); } } } /* * check for less serious errors */ if (sc->sc_espstat & NCRSTAT_PE) { printf("%s: SCSI bus parity error\n", sc->sc_dev.dv_xname); if (sc->sc_prevphase == MESSAGE_IN_PHASE) ncr53c9x_sched_msgout(SEND_PARITY_ERROR); else ncr53c9x_sched_msgout(SEND_INIT_DET_ERR); } if (sc->sc_espintr & NCRINTR_DIS) { NCR_MISC(("", sc->sc_espintr,sc->sc_espstat,sc->sc_espstep)); if (NCR_READ_REG(sc, NCR_FFLAG) & NCRFIFO_FF) { NCRCMD(sc, NCRCMD_FLUSH); DELAY(1); } /* * This command must (apparently) be issued within * 250mS of a disconnect. So here you are... */ NCRCMD(sc, NCRCMD_ENSEL); switch (sc->sc_state) { case NCR_RESELECTED: goto sched; case NCR_SELECTING: ecb->xs->error = XS_SELTIMEOUT; goto finish; case NCR_CONNECTED: if ((sc->sc_flags & NCR_SYNCHNEGO)) { #ifdef NCR53C9X_DEBUG if (ecb) sc_print_addr(ecb->xs->sc_link); printf("sync nego not completed!\n"); #endif ti = &sc->sc_tinfo[ecb->xs->sc_link->target]; sc->sc_flags &= ~NCR_SYNCHNEGO; ti->flags &= ~(T_NEGOTIATE | T_SYNCMODE); } /* it may be OK to disconnect */ if ((sc->sc_flags & NCR_ABORTING) == 0) { /* * Section 5.1.1 of the SCSI 2 spec * suggests issuing a REQUEST SENSE * following an unexpected disconnect. * Some devices go into a contingent * allegiance condition when * disconnecting, and this is necessary * to clean up their state. */ printf("%s: unexpected disconnect; ", sc->sc_dev.dv_xname); if (ecb->flags & ECB_SENSE) { printf("resetting\n"); goto reset; } printf("sending REQUEST SENSE\n"); untimeout(ncr53c9x_timeout, ecb); ncr53c9x_sense(sc, ecb); goto out; } ecb->xs->error = XS_TIMEOUT; goto finish; case NCR_DISCONNECT: TAILQ_INSERT_HEAD(&sc->nexus_list, ecb, chain); sc->sc_nexus = NULL; goto sched; case NCR_CMDCOMPLETE: goto finish; } } switch (sc->sc_state) { case NCR_SBR: printf("%s: waiting for SCSI Bus Reset to happen\n", sc->sc_dev.dv_xname); return (1); case NCR_RESELECTED: /* * we must be continuing a message ? */ if (sc->sc_phase != MESSAGE_IN_PHASE) { printf("%s: target didn't identify\n", sc->sc_dev.dv_xname); ncr53c9x_init(sc, 1); return (1); } printf("<>"); #if XXXX ncr53c9x_msgin(sc); if (sc->sc_state != NCR_CONNECTED) { /* IDENTIFY fail?! */ printf("%s: identify failed\n", sc->sc_dev.dv_xname); ncr53c9x_init(sc, 1); return (1); } #endif break; case NCR_IDLE: case NCR_SELECTING: sc->sc_msgpriq = sc->sc_msgout = sc->sc_msgoutq = 0; sc->sc_flags = 0; ecb = sc->sc_nexus; if (ecb != NULL && (ecb->flags & ECB_NEXUS)) { sc_print_addr(ecb->xs->sc_link); printf("ECB_NEXUS while in state %x\n", sc->sc_state); } if (sc->sc_espintr & NCRINTR_RESEL) { /* * If we're trying to select a * target ourselves, push our command * back into the ready list. */ if (sc->sc_state == NCR_SELECTING) { NCR_MISC(("backoff selector ")); untimeout(ncr53c9x_timeout, ecb); sc_link = ecb->xs->sc_link; ti = &sc->sc_tinfo[sc_link->target]; TAILQ_INSERT_HEAD(&sc->ready_list, ecb, chain); ecb = sc->sc_nexus = NULL; } sc->sc_state = NCR_RESELECTED; if (sc->sc_phase != MESSAGE_IN_PHASE) { /* * Things are seriously fucked up. * Pull the brakes, i.e. reset */ printf("%s: target didn't identify\n", sc->sc_dev.dv_xname); ncr53c9x_init(sc, 1); return (1); } /* * The C90 only inhibits FIFO writes until * reselection is complete, instead of * waiting until the interrupt status register * has been read. So, if the reselect happens * while we were entering a command bytes (for * another target) some of those bytes can * appear in the FIFO here, after the * interrupt is taken. */ nfifo = NCR_READ_REG(sc,NCR_FFLAG) & NCRFIFO_FF; if (nfifo < 2 || (nfifo > 2 && sc->sc_rev != NCR_VARIANT_ESP100)) { printf("%s: RESELECT: " "%d bytes in FIFO! " "[intr %x, stat %x, step %d, prevphase %x]\n", sc->sc_dev.dv_xname, nfifo, sc->sc_espintr, sc->sc_espstat, sc->sc_espstep, sc->sc_prevphase); ncr53c9x_init(sc, 1); return (1); } sc->sc_selid = NCR_READ_REG(sc, NCR_FIFO); NCR_MISC(("selid=0x%2x ", sc->sc_selid)); /* Handle identify message */ ncr53c9x_msgin(sc); if (nfifo != 2) { /* * Note: this should not happen * with `dmaselect' on. */ sc->sc_flags |= NCR_EXPECT_ILLCMD; NCRCMD(sc, NCRCMD_FLUSH); } else if (ncr53c9x_dmaselect && sc->sc_rev == NCR_VARIANT_ESP100) { sc->sc_flags |= NCR_EXPECT_ILLCMD; } if (sc->sc_state != NCR_CONNECTED) { /* IDENTIFY fail?! */ printf("%s: identify failed\n", sc->sc_dev.dv_xname); ncr53c9x_init(sc, 1); return (1); } goto shortcut; /* ie. next phase expected soon */ } #define NCRINTR_DONE (NCRINTR_FC|NCRINTR_BS) if ((sc->sc_espintr & NCRINTR_DONE) == NCRINTR_DONE) { /* * Arbitration won; examine the `step' register * to determine how far the selection could progress. */ ecb = sc->sc_nexus; if (!ecb) panic("esp: no nexus"); sc_link = ecb->xs->sc_link; ti = &sc->sc_tinfo[sc_link->target]; switch (sc->sc_espstep) { case 0: /* * The target did not respond with a * message out phase - probably an old * device that doesn't recognize ATN. * Clear ATN and just continue, the * target should be in the command * phase. * XXXX check for command phase? */ NCRCMD(sc, NCRCMD_RSTATN); break; case 1: if ((ti->flags & T_NEGOTIATE) == 0) { printf("%s: step 1 & !NEG\n", sc->sc_dev.dv_xname); goto reset; } if (sc->sc_phase != MESSAGE_OUT_PHASE) { printf("%s: !MSGOUT\n", sc->sc_dev.dv_xname); goto reset; } /* Start negotiating */ ti->period = sc->sc_minsync; ti->offset = 15; sc->sc_flags |= NCR_SYNCHNEGO; ncr53c9x_sched_msgout(SEND_SDTR); break; case 3: /* * Grr, this is supposed to mean * "target left command phase prematurely". * It seems to happen regularly when * sync mode is on. * Look at FIFO to see if command went out. * (Timing problems?) */ if (ncr53c9x_dmaselect) { if (sc->sc_cmdlen == 0) /* Hope for the best.. */ break; } else if ((NCR_READ_REG(sc, NCR_FFLAG) & NCRFIFO_FF) == 0) { /* Hope for the best.. */ break; } printf("(%s:%d:%d): selection failed;" " %d left in FIFO " "[intr %x, stat %x, step %d]\n", sc->sc_dev.dv_xname, sc_link->target, sc_link->lun, NCR_READ_REG(sc, NCR_FFLAG) & NCRFIFO_FF, sc->sc_espintr, sc->sc_espstat, sc->sc_espstep); NCRCMD(sc, NCRCMD_FLUSH); ncr53c9x_sched_msgout(SEND_ABORT); return (1); case 2: /* Select stuck at Command Phase */ NCRCMD(sc, NCRCMD_FLUSH); case 4: if (ncr53c9x_dmaselect && sc->sc_cmdlen != 0) printf("(%s:%d:%d): select; " "%d left in DMA buffer " "[intr %x, stat %x, step %d]\n", sc->sc_dev.dv_xname, sc_link->target, sc_link->lun, sc->sc_cmdlen, sc->sc_espintr, sc->sc_espstat, sc->sc_espstep); /* So far, everything went fine */ break; } ecb->flags |= ECB_NEXUS; ti->lubusy |= (1 << sc_link->lun); sc->sc_prevphase = INVALID_PHASE; /* ?? */ /* Do an implicit RESTORE POINTERS. */ sc->sc_dp = ecb->daddr; sc->sc_dleft = ecb->dleft; sc->sc_state = NCR_CONNECTED; break; } else { printf("%s: unexpected status after select" ": [intr %x, stat %x, step %x]\n", sc->sc_dev.dv_xname, sc->sc_espintr, sc->sc_espstat, sc->sc_espstep); NCRCMD(sc, NCRCMD_FLUSH); DELAY(1); goto reset; } if (sc->sc_state == NCR_IDLE) { printf("%s: stray interrupt\n", sc->sc_dev.dv_xname); return (0); } break; case NCR_CONNECTED: if (sc->sc_flags & NCR_ICCS) { /* "Initiate Command Complete Steps" in progress */ u_char msg; sc->sc_flags &= ~NCR_ICCS; if (!(sc->sc_espintr & NCRINTR_DONE)) { printf("%s: ICCS: " ": [intr %x, stat %x, step %x]\n", sc->sc_dev.dv_xname, sc->sc_espintr, sc->sc_espstat, sc->sc_espstep); } if ((NCR_READ_REG(sc, NCR_FFLAG) & NCRFIFO_FF) != 2) { int i = (NCR_READ_REG(sc, NCR_FFLAG) & NCRFIFO_FF) - 2; while (i--) (void) NCR_READ_REG(sc, NCR_FIFO); } ecb->stat = NCR_READ_REG(sc, NCR_FIFO); msg = NCR_READ_REG(sc, NCR_FIFO); NCR_PHASE(("", ecb->stat, msg)); if (msg == MSG_CMDCOMPLETE) { ecb->dleft = (ecb->flags & ECB_TENTATIVE_DONE) ? 0 : sc->sc_dleft; if ((ecb->flags & ECB_SENSE) == 0) ecb->xs->resid = ecb->dleft; sc->sc_state = NCR_CMDCOMPLETE; } else printf("%s: STATUS_PHASE: msg %d\n", sc->sc_dev.dv_xname, msg); NCRCMD(sc, NCRCMD_MSGOK); goto shortcut; /* ie. wait for disconnect */ } break; default: panic("%s: invalid state: %d", sc->sc_dev.dv_xname, sc->sc_state); } /* * Driver is now in state NCR_CONNECTED, i.e. we * have a current command working the SCSI bus. */ if (sc->sc_state != NCR_CONNECTED || ecb == NULL) { panic("esp no nexus"); } switch (sc->sc_phase) { case MESSAGE_OUT_PHASE: NCR_PHASE(("MESSAGE_OUT_PHASE ")); ncr53c9x_msgout(sc); sc->sc_prevphase = MESSAGE_OUT_PHASE; break; case MESSAGE_IN_PHASE: NCR_PHASE(("MESSAGE_IN_PHASE ")); sc->sc_prevphase = MESSAGE_IN_PHASE; if (sc->sc_espintr & NCRINTR_BS) { NCRCMD(sc, NCRCMD_FLUSH); sc->sc_flags |= NCR_WAITI; NCRCMD(sc, NCRCMD_TRANS); } else if (sc->sc_espintr & NCRINTR_FC) { if ((sc->sc_flags & NCR_WAITI) == 0) { printf("%s: MSGIN: unexpected FC bit: " "[intr %x, stat %x, step %x]\n", sc->sc_dev.dv_xname, sc->sc_espintr, sc->sc_espstat, sc->sc_espstep); } sc->sc_flags &= ~NCR_WAITI; ncr53c9x_msgin(sc); } else { printf("%s: MSGIN: weird bits: " "[intr %x, stat %x, step %x]\n", sc->sc_dev.dv_xname, sc->sc_espintr, sc->sc_espstat, sc->sc_espstep); } goto shortcut; /* i.e. expect data to be ready */ break; case COMMAND_PHASE: /* * Send the command block. Normally we don't see this * phase because the SEL_ATN command takes care of * all this. However, we end up here if either the * target or we wanted to exchange some more messages * first (e.g. to start negotiations). */ NCR_PHASE(("COMMAND_PHASE 0x%02x (%d) ", ecb->cmd.cmd.opcode, ecb->clen)); if (NCR_READ_REG(sc, NCR_FFLAG) & NCRFIFO_FF) { NCRCMD(sc, NCRCMD_FLUSH); DELAY(1); } if (ncr53c9x_dmaselect) { size_t size; /* setup DMA transfer for command */ size = ecb->clen; sc->sc_cmdlen = size; sc->sc_cmdp = (caddr_t)&ecb->cmd.cmd; NCRDMA_SETUP(sc, &sc->sc_cmdp, &sc->sc_cmdlen, 0, &size); /* Program the SCSI counter */ NCR_WRITE_REG(sc, NCR_TCL, size); NCR_WRITE_REG(sc, NCR_TCM, size >> 8); if (sc->sc_cfg2 & NCRCFG2_FE) { NCR_WRITE_REG(sc, NCR_TCH, size >> 16); } /* load the count in */ NCRCMD(sc, NCRCMD_NOP|NCRCMD_DMA); /* start the command transfer */ NCRCMD(sc, NCRCMD_TRANS | NCRCMD_DMA); NCRDMA_GO(sc); } else { u_char *cmd = (u_char *)&ecb->cmd.cmd; int i; /* Now the command into the FIFO */ for (i = 0; i < ecb->clen; i++) NCR_WRITE_REG(sc, NCR_FIFO, *cmd++); NCRCMD(sc, NCRCMD_TRANS); } sc->sc_prevphase = COMMAND_PHASE; break; case DATA_OUT_PHASE: NCR_PHASE(("DATA_OUT_PHASE [%ld] ",(long)sc->sc_dleft)); NCRCMD(sc, NCRCMD_FLUSH); size = min(sc->sc_dleft, sc->sc_maxxfer); NCRDMA_SETUP(sc, &sc->sc_dp, &sc->sc_dleft, 0, &size); sc->sc_prevphase = DATA_OUT_PHASE; goto setup_xfer; case DATA_IN_PHASE: NCR_PHASE(("DATA_IN_PHASE ")); if (sc->sc_rev == NCR_VARIANT_ESP100) NCRCMD(sc, NCRCMD_FLUSH); size = min(sc->sc_dleft, sc->sc_maxxfer); NCRDMA_SETUP(sc, &sc->sc_dp, &sc->sc_dleft, 1, &size); sc->sc_prevphase = DATA_IN_PHASE; setup_xfer: /* Target returned to data phase: wipe "done" memory */ ecb->flags &= ~ECB_TENTATIVE_DONE; /* Program the SCSI counter */ NCR_WRITE_REG(sc, NCR_TCL, size); NCR_WRITE_REG(sc, NCR_TCM, size >> 8); if (sc->sc_cfg2 & NCRCFG2_FE) { NCR_WRITE_REG(sc, NCR_TCH, size >> 16); } /* load the count in */ NCRCMD(sc, NCRCMD_NOP|NCRCMD_DMA); /* * Note that if `size' is 0, we've already transceived * all the bytes we want but we're still in DATA PHASE. * Apparently, the device needs padding. Also, a * transfer size of 0 means "maximum" to the chip * DMA logic. */ NCRCMD(sc, (size==0?NCRCMD_TRPAD:NCRCMD_TRANS)|NCRCMD_DMA); NCRDMA_GO(sc); return (1); case STATUS_PHASE: NCR_PHASE(("STATUS_PHASE ")); sc->sc_flags |= NCR_ICCS; NCRCMD(sc, NCRCMD_ICCS); sc->sc_prevphase = STATUS_PHASE; break; case INVALID_PHASE: break; default: printf("%s: unexpected bus phase; resetting\n", sc->sc_dev.dv_xname); goto reset; } out: return (1); reset: ncr53c9x_init(sc, 1); goto out; finish: ncr53c9x_done(sc, ecb); goto out; sched: sc->sc_state = NCR_IDLE; ncr53c9x_sched(sc); goto out; shortcut: /* * The idea is that many of the SCSI operations take very little * time, and going away and getting interrupted is too high an * overhead to pay. For example, selecting, sending a message * and command and then doing some work can be done in one "pass". * * The delay is a heuristic. It is 2 when at 20Mhz, 2 at 25Mhz and 1 * at 40Mhz. This needs testing. */ DELAY(50/sc->sc_freq); if (NCRDMA_ISINTR(sc)) goto again; goto out; } void ncr53c9x_abort(sc, ecb) struct ncr53c9x_softc *sc; struct ncr53c9x_ecb *ecb; { /* 2 secs for the abort */ ecb->timeout = NCR_ABORT_TIMEOUT; ecb->flags |= ECB_ABORT; if (ecb == sc->sc_nexus) { /* * If we're still selecting, the message will be scheduled * after selection is complete. */ if (sc->sc_state == NCR_CONNECTED) ncr53c9x_sched_msgout(SEND_ABORT); /* * Reschedule timeout. First, cancel a queued timeout (if any) * in case someone decides to call ncr53c9x_abort() from * elsewhere. */ untimeout(ncr53c9x_timeout, ecb); timeout(ncr53c9x_timeout, ecb, (ecb->timeout * hz) / 1000); } else { /* The command should be on the nexus list */ if ((ecb->flags & ECB_NEXUS) == 0) { sc_print_addr(ecb->xs->sc_link); printf("ncr53c9x_abort: not NEXUS\n"); ncr53c9x_init(sc, 1); } /* * Just leave the command on the nexus list. * XXX - what choice do we have but to reset the SCSI * eventually? */ if (sc->sc_state == NCR_IDLE) ncr53c9x_sched(sc); } } void ncr53c9x_timeout(arg) void *arg; { struct ncr53c9x_ecb *ecb = arg; struct scsi_xfer *xs = ecb->xs; struct scsi_link *sc_link = xs->sc_link; struct ncr53c9x_softc *sc = sc_link->adapter_softc; struct ncr53c9x_tinfo *ti = &sc->sc_tinfo[sc_link->target]; int s; sc_print_addr(sc_link); printf("%s: timed out [ecb %p (flags 0x%x, dleft %x, stat %x)], " "", sc->sc_dev.dv_xname, ecb, ecb->flags, ecb->dleft, ecb->stat, sc->sc_state, sc->sc_nexus, NCR_READ_REG(sc, NCR_STAT), sc->sc_phase, sc->sc_prevphase, (long)sc->sc_dleft, sc->sc_msgpriq, sc->sc_msgout, NCRDMA_ISACTIVE(sc) ? "DMA active" : ""); #if NCR53C9X_DEBUG > 1 printf("TRACE: %s.", ecb->trace); #endif s = splbio(); if (ecb->flags & ECB_ABORT) { /* abort timed out */ printf(" AGAIN\n"); ncr53c9x_init(sc, 1); } else { /* abort the operation that has timed out */ printf("\n"); xs->error = XS_TIMEOUT; ncr53c9x_abort(sc, ecb); /* Disable sync mode if stuck in a data phase */ if (ecb == sc->sc_nexus && (ti->flags & T_SYNCMODE) != 0 && (sc->sc_phase & (MSGI|CDI)) == 0) { sc_print_addr(sc_link); printf("sync negotiation disabled\n"); sc->sc_cfflags |= (1<<(sc_link->target+8)); } } splx(s); }