/* $NetBSD: ncr5380sbc.c,v 1.3 1996/02/10 00:11:48 christos Exp $ */ /* * Copyright (c) 1995 David Jones, Gordon W. Ross * Copyright (c) 1994 Jarle Greipsland * 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. The name of the authors may not be used to endorse or promote products * derived from this software without specific prior written permission. * 4. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by * David Jones and Gordon Ross * * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``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 AUTHORS 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. */ /* * This is a machine-independent driver for the NCR5380 * SCSI Bus Controller (SBC), also known as the Am5380. * * This code should work with any memory-mapped 5380, * and can be shared by multiple adapters that address * the 5380 with different register offset spacings. * (This can happen on the atari, for example.) * For porting/design info. see: ncr5380.doc * * Credits, history: * * David Jones is the author of most of the code that now * appears in this file, and was the architect of the * current overall structure (MI/MD code separation, etc.) * * Gordon Ross integrated the message phase code, added lots of * comments about what happens when and why (re. SCSI spec.), * debugged some reentrance problems, and added several new * "hooks" needed for the Sun3 "si" adapters. * * The message in/out code was taken nearly verbatim from * the aic6360 driver by Jarle Greipsland. * * Several other NCR5380 drivers were used for reference * while developing this driver, including work by: * The Alice Group (mac68k port) namely: * Allen K. Briggs, Chris P. Caputo, Michael L. Finch, * Bradley A. Grantham, and Lawrence A. Kesteloot * Michael L. Hitch (amiga drivers: sci.c) * Leo Weppelman (atari driver: ncr5380.c) * There are others too. Thanks, everyone. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #ifndef DEBUG #define DEBUG XXX #endif #include #include static int ncr5380_wait_req __P((struct ncr5380_softc *)); static int ncr5380_wait_not_req __P((struct ncr5380_softc *)); static void ncr5380_sched __P((struct ncr5380_softc *)); static void ncr5380_done __P((struct ncr5380_softc *)); static int ncr5380_select __P((struct ncr5380_softc *, struct sci_req *)); static void ncr5380_reselect __P((struct ncr5380_softc *)); static int ncr5380_msg_in __P((struct ncr5380_softc *)); static int ncr5380_msg_out __P((struct ncr5380_softc *)); static int ncr5380_data_xfer __P((struct ncr5380_softc *, int)); static int ncr5380_command __P((struct ncr5380_softc *)); static int ncr5380_status __P((struct ncr5380_softc *)); static void ncr5380_machine __P((struct ncr5380_softc *)); /* * Action flags returned by the info_tranfer functions: * (These determine what happens next.) */ #define ACT_CONTINUE 0x00 /* No flags: expect another phase */ #define ACT_DISCONNECT 0x01 /* Target is disconnecting */ #define ACT_CMD_DONE 0x02 /* Need to call scsi_done() */ #define ACT_RESET_BUS 0x04 /* Need bus reset (cmd timeout) */ #define ACT_WAIT_DMA 0x10 /* Wait for DMA to complete */ /***************************************************************** * Debugging stuff *****************************************************************/ #ifndef DDB /* This is used only in recoverable places. */ #define Debugger() printf("Debug: ncr5380.c:%d\n", __LINE__) #endif #ifdef DEBUG #define NCR_DBG_BREAK 1 #define NCR_DBG_CMDS 2 int ncr5380_debug = 0; #define NCR_BREAK() \ do { if (ncr5380_debug & NCR_DBG_BREAK) Debugger(); } while (0) static void ncr5380_show_scsi_cmd __P((struct scsi_xfer *)); static void ncr5380_show_sense __P((struct scsi_xfer *)); #else /* DEBUG */ #define NCR_BREAK() /* nada */ #define ncr5380_show_scsi_cmd(xs) /* nada */ #define ncr5380_show_sense(xs) /* nada */ #endif /* DEBUG */ static char * phase_names[8] = { "DATA_OUT", "DATA_IN", "COMMAND", "STATUS", "UNSPEC1", "UNSPEC2", "MSG_OUT", "MSG_IN", }; /***************************************************************** * Actual chip control *****************************************************************/ /* * XXX: These timeouts might need to be tuned... */ /* This one is used when waiting for a phase change. (X100uS.) */ int ncr5380_wait_phase_timo = 1000 * 10 * 300; /* 5 min. */ /* These are used in the following inline functions. */ int ncr5380_wait_req_timo = 1000 * 50; /* X2 = 100 mS. */ int ncr5380_wait_nrq_timo = 1000 * 25; /* X2 = 50 mS. */ /* Return zero on success. */ static __inline__ int ncr5380_wait_req(sc) struct ncr5380_softc *sc; { register int timo = ncr5380_wait_req_timo; for (;;) { if (*sc->sci_bus_csr & SCI_BUS_REQ) { timo = 0; /* return 0 */ break; } if (--timo < 0) break; /* return -1 */ delay(2); } return (timo); } /* Return zero on success. */ static __inline__ int ncr5380_wait_not_req(sc) struct ncr5380_softc *sc; { register int timo = ncr5380_wait_nrq_timo; for (;;) { if ((*sc->sci_bus_csr & SCI_BUS_REQ) == 0) { timo = 0; /* return 0 */ break; } if (--timo < 0) break; /* return -1 */ delay(2); } return (timo); } /* Ask the target for a MSG_OUT phase. */ static __inline__ void ncr_sched_msgout(sc, msg_code) struct ncr5380_softc *sc; int msg_code; { /* First time, raise ATN line. */ if (sc->sc_msgpriq == 0) { register u_char icmd; icmd = *sc->sci_icmd & SCI_ICMD_RMASK; *sc->sci_icmd = icmd | SCI_ICMD_ATN; delay(2); } sc->sc_msgpriq |= msg_code; } int ncr5380_pio_out(sc, phase, count, data) struct ncr5380_softc *sc; int phase, count; unsigned char *data; { register u_char icmd; register int resid; register int error; icmd = *(sc->sci_icmd) & SCI_ICMD_RMASK; icmd |= SCI_ICMD_DATA; *sc->sci_icmd = icmd; resid = count; while (resid > 0) { if (!SCI_BUSY(sc)) { NCR_TRACE("pio_out: lost BSY, resid=%d\n", resid); break; } if (ncr5380_wait_req(sc)) { NCR_TRACE("pio_out: no REQ, resid=%d\n", resid); break; } if (SCI_BUS_PHASE(*sc->sci_bus_csr) != phase) break; /* Put the data on the bus. */ *sc->sci_odata = *data++; /* Tell the target it's there. */ icmd |= SCI_ICMD_ACK; *sc->sci_icmd = icmd; /* Wait for target to get it. */ error = ncr5380_wait_not_req(sc); /* OK, it's got it (or we gave up waiting). */ icmd &= ~SCI_ICMD_ACK; *sc->sci_icmd = icmd; if (error) { NCR_TRACE("pio_out: stuck REQ, resid=%d\n", resid); break; } --resid; } /* Stop driving the data bus. */ icmd &= ~SCI_ICMD_DATA; *sc->sci_icmd = icmd; return (count - resid); } int ncr5380_pio_in(sc, phase, count, data) struct ncr5380_softc *sc; int phase, count; unsigned char *data; { register u_char icmd; register int resid; register int error; icmd = *(sc->sci_icmd) & SCI_ICMD_RMASK; resid = count; while (resid > 0) { if (!SCI_BUSY(sc)) { NCR_TRACE("pio_in: lost BSY, resid=%d\n", resid); break; } if (ncr5380_wait_req(sc)) { NCR_TRACE("pio_in: no REQ, resid=%d\n", resid); break; } /* A phase change is not valid until AFTER REQ rises! */ if (SCI_BUS_PHASE(*sc->sci_bus_csr) != phase) break; /* Read the data bus. */ *data++ = *sc->sci_data; /* Tell target we got it. */ icmd |= SCI_ICMD_ACK; *sc->sci_icmd = icmd; /* Wait for target to drop REQ... */ error = ncr5380_wait_not_req(sc); /* OK, we can drop ACK. */ icmd &= ~SCI_ICMD_ACK; *sc->sci_icmd = icmd; if (error) { NCR_TRACE("pio_in: stuck REQ, resid=%d\n", resid); break; } --resid; } return (count - resid); } void ncr5380_init(sc) struct ncr5380_softc *sc; { int i, j; #ifdef DEBUG ncr5380_debug_sc = sc; #endif for (i = 0; i < SCI_OPENINGS; i++) sc->sc_ring[i].sr_xs = NULL; for (i = 0; i < 8; i++) for (j = 0; j < 8; j++) sc->sc_matrix[i][j] = NULL; sc->sc_link.openings = 2; /* XXX - Not SCI_OPENINGS */ sc->sc_prevphase = PHASE_INVALID; sc->sc_state = NCR_IDLE; *sc->sci_tcmd = PHASE_INVALID; *sc->sci_icmd = 0; *sc->sci_mode = 0; *sc->sci_sel_enb = 0; SCI_CLR_INTR(sc); /* XXX: Enable reselect interrupts... */ *sc->sci_sel_enb = 0x80; /* Another hack (Er.. hook!) for the sun3 si: */ if (sc->sc_intr_on) { NCR_TRACE("init: intr ON\n", 0); sc->sc_intr_on(sc); } } void ncr5380_reset_scsibus(sc) struct ncr5380_softc *sc; { NCR_TRACE("reset_scsibus, cur=0x%x\n", (long) sc->sc_current); *sc->sci_icmd = SCI_ICMD_RST; delay(500); *sc->sci_icmd = 0; *sc->sci_mode = 0; *sc->sci_tcmd = PHASE_INVALID; SCI_CLR_INTR(sc); /* XXX - Need long delay here! */ delay(100000); /* XXX - Need to cancel disconnected requests. */ } /* * Interrupt handler for the SCSI Bus Controller (SBC) * This may also called for a DMA timeout (at splbio). */ int ncr5380_intr(sc) struct ncr5380_softc *sc; { int claimed = 0; /* * Do not touch SBC regs here unless sc_current == NULL * or it will complain about "register conflict" errors. * Instead, just let ncr5380_machine() deal with it. */ NCR_TRACE("intr: top, state=%d\n", sc->sc_state); if (sc->sc_state == NCR_IDLE) { /* * Might be reselect. ncr5380_reselect() will check, * and set up the connection if so. This will verify * that sc_current == NULL at the beginning... */ /* Another hack (Er.. hook!) for the sun3 si: */ if (sc->sc_intr_off) { NCR_TRACE("intr: for reselect, intr off\n", 0); sc->sc_intr_off(sc); } ncr5380_reselect(sc); } /* * The remaining documented interrupt causes are phase mismatch and * disconnect. In addition, the sunsi controller may produce a state * where SCI_CSR_DONE is false, yet DMA is complete. * * The procedure in all these cases is to let ncr5380_machine() * figure out what to do next. */ if (sc->sc_state & NCR_WORKING) { NCR_TRACE("intr: call machine, cur=0x%x\n", (long) sc->sc_current); /* This will usually free-up the nexus. */ ncr5380_machine(sc); NCR_TRACE("intr: machine done, cur=0x%x\n", (long) sc->sc_current); claimed = 1; } /* Maybe we can run some commands now... */ if (sc->sc_state == NCR_IDLE) { NCR_TRACE("intr: call sched, cur=0x%x\n", (long) sc->sc_current); ncr5380_sched(sc); NCR_TRACE("intr: sched done, cur=0x%x\n", (long) sc->sc_current); } return claimed; } /* * Abort the current command (i.e. due to timeout) */ void ncr5380_abort(sc) struct ncr5380_softc *sc; { /* * Finish it now. If DMA is in progress, we * can not call ncr_sched_msgout() because * that hits the SBC (avoid DMA conflict). */ /* Another hack (Er.. hook!) for the sun3 si: */ if (sc->sc_intr_off) { NCR_TRACE("abort: intr off\n", 0); sc->sc_intr_off(sc); } sc->sc_state |= NCR_ABORTING; if ((sc->sc_state & NCR_DOINGDMA) == 0) { ncr_sched_msgout(sc, SEND_ABORT); } NCR_TRACE("abort: call machine, cur=0x%x\n", (long) sc->sc_current); ncr5380_machine(sc); NCR_TRACE("abort: machine done, cur=0x%x\n", (long) sc->sc_current); /* Another hack (Er.. hook!) for the sun3 si: */ if (sc->sc_intr_on) { NCR_TRACE("abort: intr ON\n", 0); sc->sc_intr_on(sc); } } /* * Timeout handler, scheduled for each SCSI command. */ void ncr5380_cmd_timeout(arg) void *arg; { struct sci_req *sr = arg; struct scsi_xfer *xs; struct scsi_link *sc_link; struct ncr5380_softc *sc; int s; s = splbio(); /* Get all our variables... */ xs = sr->sr_xs; if (xs == NULL) { printf("ncr5380_cmd_timeout: no scsi_xfer\n"); goto out; } sc_link = xs->sc_link; sc = sc_link->adapter_softc; printf("%s: cmd timeout, targ=%d, lun=%d\n", sc->sc_dev.dv_xname, sr->sr_target, sr->sr_lun); /* * Mark the overdue job as failed, and arrange for * ncr5380_machine to terminate it. If the victim * is the current job, call ncr5380_machine() now. * Otherwise arrange for ncr5380_sched() to do it. */ sr->sr_flags |= SR_OVERDUE; if (sc->sc_current == sr) { NCR_TRACE("cmd_tmo: call abort, sr=0x%x\n", (long) sr); ncr5380_abort(sc); } else { /* * The driver may be idle, or busy with another job. * Arrange for ncr5380_sched() to do the deed. */ NCR_TRACE("cmd_tmo: clear matrix, t/l=0x%02x\n", (sr->sr_target << 4) | sr->sr_lun); sc->sc_matrix[sr->sr_target][sr->sr_lun] = NULL; } /* * We may have aborted the current job, or may have * already been idle. In either case, we should now * be idle, so try to start another job. */ if (sc->sc_state == NCR_IDLE) { NCR_TRACE("cmd_tmo: call sched, cur=0x%x\n", (long) sc->sc_current); ncr5380_sched(sc); NCR_TRACE("cmd_tmo: sched done, cur=0x%x\n", (long) sc->sc_current); } out: splx(s); } /***************************************************************** * Interface to higher level *****************************************************************/ /* * Enter a new SCSI command into the "issue" queue, and * if there is work to do, start it going. * * WARNING: This can be called recursively! * (see comment in ncr5380_done) */ int ncr5380_scsi_cmd(xs) struct scsi_xfer *xs; { struct ncr5380_softc *sc; struct sci_req *sr; int s, rv, i, flags; extern int cold; /* XXX */ sc = xs->sc_link->adapter_softc; flags = xs->flags; /* * XXX: Hack: During autoconfig, force polling mode. * Needed as long as sdsize() can be called while cold, * otherwise timeouts will never call back (grumble). */ if (cold) flags |= SCSI_POLL; if (sc->sc_flags & NCR5380_FORCE_POLLING) flags |= SCSI_POLL; if (flags & SCSI_DATA_UIO) panic("ncr5380: scsi data uio requested"); s = splbio(); if (flags & SCSI_POLL) { /* Terminate any current command. */ sr = sc->sc_current; if (sr) { printf("%s: polled request aborting %d/%d\n", sc->sc_dev.dv_xname, sr->sr_target, sr->sr_lun); ncr5380_abort(sc); } if (sc->sc_state != NCR_IDLE) { panic("ncr5380_scsi_cmd: polled request, abort failed"); } } /* * Find lowest empty slot in ring buffer. * XXX: What about "fairness" and cmd order? */ for (i = 0; i < SCI_OPENINGS; i++) if (sc->sc_ring[i].sr_xs == NULL) goto new; rv = TRY_AGAIN_LATER; NCR_TRACE("scsi_cmd: no openings, rv=%d\n", rv); goto out; new: /* Create queue entry */ sr = &sc->sc_ring[i]; sr->sr_xs = xs; sr->sr_target = xs->sc_link->target; sr->sr_lun = xs->sc_link->lun; sr->sr_dma_hand = NULL; sr->sr_dataptr = xs->data; sr->sr_datalen = xs->datalen; sr->sr_flags = (flags & SCSI_POLL) ? SR_IMMED : 0; sr->sr_status = -1; /* no value */ sc->sc_ncmds++; rv = SUCCESSFULLY_QUEUED; NCR_TRACE("scsi_cmd: new sr=0x%x\n", (long)sr); if (flags & SCSI_POLL) { /* Force this new command to be next. */ sc->sc_rr = i; } /* * If we were idle, run some commands... */ if (sc->sc_state == NCR_IDLE) { NCR_TRACE("scsi_cmd: call sched, cur=0x%x\n", (long) sc->sc_current); ncr5380_sched(sc); NCR_TRACE("scsi_cmd: sched done, cur=0x%x\n", (long) sc->sc_current); } if (flags & SCSI_POLL) { /* Make sure ncr5380_sched() finished it. */ if ((xs->flags & ITSDONE) == 0) panic("ncr5380_scsi_cmd: poll didn't finish"); rv = COMPLETE; } out: splx(s); return (rv); } /* * POST PROCESSING OF SCSI_CMD (usually current) * Called by ncr5380_sched(), ncr5380_machine() */ static void ncr5380_done(sc) struct ncr5380_softc *sc; { struct sci_req *sr; struct scsi_xfer *xs; #ifdef DIAGNOSTIC if (sc->sc_state == NCR_IDLE) panic("ncr5380_done: state=idle"); if (sc->sc_current == NULL) panic("ncr5380_done: current=0"); #endif sr = sc->sc_current; xs = sr->sr_xs; NCR_TRACE("done: top, cur=0x%x\n", (long) sc->sc_current); /* * Clean up DMA resources for this command. */ if (sr->sr_dma_hand) { NCR_TRACE("done: dma_free, dh=0x%x\n", (long) sr->sr_dma_hand); (*sc->sc_dma_free)(sc); } #ifdef DIAGNOSTIC if (sr->sr_dma_hand) panic("ncr5380_done: dma free did not"); #endif if (sc->sc_state & NCR_ABORTING) { NCR_TRACE("done: aborting, error=%d\n", xs->error); if (xs->error == XS_NOERROR) xs->error = XS_TIMEOUT; } NCR_TRACE("done: check error=%d\n", (long) xs->error); /* If error is already set, ignore sr_status value. */ if (xs->error != XS_NOERROR) goto finish; NCR_TRACE("done: check status=%d\n", sr->sr_status); switch (sr->sr_status) { case SCSI_OK: /* 0 */ if (sr->sr_flags & SR_SENSE) { if (ncr5380_debug & NCR_DBG_CMDS) { ncr5380_show_sense(xs); } xs->error = XS_SENSE; } break; case SCSI_CHECK: if (sr->sr_flags & SR_SENSE) { /* Sense command also asked for sense? */ printf("ncr5380_done: sense asked for sense\n"); NCR_BREAK(); xs->error = XS_DRIVER_STUFFUP; break; } sr->sr_flags |= SR_SENSE; NCR_TRACE("done: get sense, sr=0x%x\n", (long) sr); /* * Leave queued, but clear sc_current so we start over * with selection. Guaranteed to get the same request. */ sc->sc_state = NCR_IDLE; sc->sc_current = NULL; sc->sc_matrix[sr->sr_target][sr->sr_lun] = NULL; return; /* XXX */ case SCSI_BUSY: xs->error = XS_BUSY; break; case -1: /* This is our "impossible" initial value. */ /* fallthrough */ default: printf("%s: target %d, bad status=%d\n", sc->sc_dev.dv_xname, sr->sr_target, sr->sr_status); xs->error = XS_DRIVER_STUFFUP; break; } finish: NCR_TRACE("done: finish, error=%d\n", xs->error); /* * Dequeue the finished command, but don't clear sc_state until * after the call to scsi_done(), because that may call back to * ncr5380_scsi_cmd() - unwanted recursion! * * Keeping sc->sc_state != idle terminates the recursion. */ #ifdef DIAGNOSTIC if ((sc->sc_state & NCR_WORKING) == 0) panic("ncr5380_done: bad state"); #endif /* Clear our pointers to the request. */ sc->sc_current = NULL; sc->sc_matrix[sr->sr_target][sr->sr_lun] = NULL; untimeout(ncr5380_cmd_timeout, sr); /* Make the request free. */ sr->sr_xs = NULL; sc->sc_ncmds--; /* Tell common SCSI code it is done. */ xs->flags |= ITSDONE; scsi_done(xs); sc->sc_state = NCR_IDLE; /* Now ncr5380_sched() may be called again. */ } /* * Schedule a SCSI operation. This routine should return * only after it achieves one of the following conditions: * Busy (sc->sc_state != NCR_IDLE) * No more work can be started. */ static void ncr5380_sched(sc) struct ncr5380_softc *sc; { struct sci_req *sr; struct scsi_xfer *xs; int target, lun; int error, i; /* Another hack (Er.. hook!) for the sun3 si: */ if (sc->sc_intr_off) { NCR_TRACE("sched: top, intr off\n", 0); sc->sc_intr_off(sc); } next_job: /* * Grab the next job from queue. Must be idle. */ #ifdef DIAGNOSTIC if (sc->sc_state != NCR_IDLE) panic("ncr5380_sched: not idle"); if (sc->sc_current) panic("ncr5380_sched: current set"); #endif /* * Always start the search where we last looked. * The REQUEST_SENSE logic depends on this to * choose the same job as was last picked, so it * can just clear sc_current and reschedule. * (Avoids loss of "contingent allegiance".) */ i = sc->sc_rr; sr = NULL; do { if (sc->sc_ring[i].sr_xs) { target = sc->sc_ring[i].sr_target; lun = sc->sc_ring[i].sr_lun; if (sc->sc_matrix[target][lun] == NULL) { sc->sc_matrix[target][lun] = sr = &sc->sc_ring[i]; sc->sc_rr = i; break; } } i++; if (i == SCI_OPENINGS) i = 0; } while (i != sc->sc_rr); if (sr == NULL) { NCR_TRACE("sched: no work, cur=0x%x\n", (long) sc->sc_current); /* Another hack (Er.. hook!) for the sun3 si: */ if (sc->sc_intr_on) { NCR_TRACE("sched: ret, intr ON\n", 0); sc->sc_intr_on(sc); } return; /* No more work to do. */ } NCR_TRACE("sched: select for t/l=0x%02x\n", (sr->sr_target << 4) | sr->sr_lun); sc->sc_state = NCR_WORKING; error = ncr5380_select(sc, sr); if (sc->sc_current) { /* Lost the race! reselected out from under us! */ /* Work with the reselected job. */ if (sr->sr_flags & SR_IMMED) { printf("%s: reselected while polling (abort)\n", sc->sc_dev.dv_xname); /* Abort the reselected job. */ sc->sc_state |= NCR_ABORTING; sc->sc_msgpriq |= SEND_ABORT; } sr = sc->sc_current; xs = sr->sr_xs; NCR_TRACE("sched: reselect, new sr=0x%x\n", (long)sr); goto have_nexus; } /* Normal selection result */ sc->sc_current = sr; /* connected */ xs = sr->sr_xs; /* * Initialize pointers, etc. for this job */ sc->sc_dataptr = sr->sr_dataptr; sc->sc_datalen = sr->sr_datalen; sc->sc_prevphase = PHASE_INVALID; sc->sc_msgpriq = SEND_IDENTIFY; sc->sc_msgoutq = 0; sc->sc_msgout = 0; NCR_TRACE("sched: select rv=%d\n", error); switch (error) { case XS_NOERROR: break; case XS_BUSY: /* XXX - Reset and try again. */ printf("%s: SCSI bus busy, resetting...\n", sc->sc_dev.dv_xname); ncr5380_reset_scsibus(sc); /* fallthrough */ case XS_SELTIMEOUT: default: xs->error = error; /* from select */ NCR_TRACE("sched: call done, sr=0x%x\n", (long)sr); ncr5380_done(sc); /* Paranoia: clear everything. */ sc->sc_dataptr = NULL; sc->sc_datalen = 0; sc->sc_prevphase = PHASE_INVALID; sc->sc_msgpriq = 0; sc->sc_msgoutq = 0; sc->sc_msgout = 0; goto next_job; } /* * Selection was successful. Normally, this means * we are starting a new command. However, this * might be the termination of an overdue job. */ if (sr->sr_flags & SR_OVERDUE) { NCR_TRACE("sched: overdue, sr=0x%x\n", (long)sr); sc->sc_state |= NCR_ABORTING; sc->sc_msgpriq |= SEND_ABORT; goto have_nexus; } /* * This may be the continuation of some job that * completed with a "check condition" code. */ if (sr->sr_flags & SR_SENSE) { NCR_TRACE("sched: get sense, sr=0x%x\n", (long)sr); /* Do not allocate DMA, nor set timeout. */ goto have_nexus; } /* * OK, we are starting a new command. * Initialize and allocate resources for the new command. * Device reset is special (only uses MSG_OUT phase). * Normal commands start in MSG_OUT phase where we will * send and IDENDIFY message, and then expect CMD phase. */ if (ncr5380_debug & NCR_DBG_CMDS) { printf("ncr5380_sched: begin, target=%d, LUN=%d\n", xs->sc_link->target, xs->sc_link->lun); ncr5380_show_scsi_cmd(xs); } if (xs->flags & SCSI_RESET) { NCR_TRACE("sched: cmd=reset, sr=0x%x\n", (long)sr); /* Not an error, so do not set NCR_ABORTING */ sc->sc_msgpriq |= SEND_DEV_RESET; goto have_nexus; } #ifdef DIAGNOSTIC if ((xs->flags & (SCSI_DATA_IN | SCSI_DATA_OUT)) == 0) { if (sc->sc_dataptr) { printf("%s: ptr but no data in/out flags?\n"); NCR_BREAK(); sc->sc_dataptr = NULL; } } #endif /* Allocate DMA space (maybe) */ if (sc->sc_dataptr && sc->sc_dma_alloc && (sc->sc_datalen >= sc->sc_min_dma_len)) { NCR_TRACE("sched: dma_alloc, len=%d\n", sc->sc_datalen); (*sc->sc_dma_alloc)(sc); } /* * Initialization hook called just after select, * at the beginning of COMMAND phase. * (but AFTER the DMA allocation is done) * * The evil Sun "si" adapter (OBIO variant) needs some * setup done to the DMA engine BEFORE the target puts * the SCSI bus into any DATA phase. */ if (sr->sr_dma_hand && sc->sc_dma_setup) { NCR_TRACE("sched: dma_setup, dh=0x%x\n", (long) sr->sr_dma_hand); sc->sc_dma_setup(sc); } /* * Schedule a timeout for the job we are starting. */ if ((sr->sr_flags & SR_IMMED) == 0) { i = (xs->timeout * hz) / 1000; NCR_TRACE("sched: set timeout=%d\n", i); timeout(ncr5380_cmd_timeout, sr, i); } have_nexus: NCR_TRACE("sched: call machine, cur=0x%x\n", (long) sc->sc_current); ncr5380_machine(sc); NCR_TRACE("sched: machine done, cur=0x%x\n", (long) sc->sc_current); /* * What state did ncr5380_machine() leave us in? * Hopefully it sometimes completes a job... */ if (sc->sc_state == NCR_IDLE) goto next_job; return; /* Have work in progress. */ } /* * Reselect handler: checks for reselection, and if we are being * reselected, it sets up sc->sc_current. * * We are reselected when: * SEL is TRUE * IO is TRUE * BSY is FALSE */ void ncr5380_reselect(sc) struct ncr5380_softc *sc; { struct sci_req *sr; int target, lun, phase, timo; u_char bus, data, icmd, msg; #ifdef DIAGNOSTIC /* * Note: sc_state will be "idle" when ncr5380_intr() * calls, or "working" when ncr5380_select() calls. * (So don't test that in this DIAGNOSTIC) */ if (sc->sc_current) panic("ncr5380_reselect: current set"); #endif /* * First, check the select line. * (That has to be set first.) */ bus = *(sc->sci_bus_csr); if ((bus & SCI_BUS_SEL) == 0) { /* Not a selection or reselection. */ return; } /* * The target will assert BSY first (for bus arbitration), * then raise SEL, and finally drop BSY. Only then is the * data bus required to have valid selection ID bits set. * Wait for: SEL==1, BSY==0 before reading the data bus. * While this theoretically can happen, we are aparently * never fast enough to get here before BSY drops. */ timo = ncr5380_wait_nrq_timo; for (;;) { if ((bus & SCI_BUS_BSY) == 0) break; /* Probably never get here... */ if (--timo <= 0) { printf("%s: reselect, BSY stuck, bus=0x%x\n", sc->sc_dev.dv_xname, bus); /* Not much we can do. Reset the bus. */ ncr5380_reset_scsibus(sc); return; } delay(2); bus = *(sc->sci_bus_csr); /* If SEL went away, forget it. */ if ((bus & SCI_BUS_SEL) == 0) return; /* Still have SEL, check BSY. */ } NCR_TRACE("reselect, valid data after %d loops\n", ncr5380_wait_nrq_timo - timo); /* * Good. We have SEL=1 and BSY=0. Now wait for a * "bus settle delay" before we sample the data bus */ delay(2); data = *(sc->sci_data) & 0xFF; /* Parity check is implicit in data validation below. */ /* * Is this a reselect (I/O == 1) or have we been * selected as a target? (I/O == 0) */ if ((bus & SCI_BUS_IO) == 0) { printf("%s: selected as target, data=0x%x\n", sc->sc_dev.dv_xname, data); /* Not much we can do. Reset the bus. */ ncr5380_reset_scsibus(sc); return; } /* * OK, this is a reselection. */ for (target = 0; target < 7; target++) if (data & (1 << target)) break; if ((data & 0x7F) != (1 << target)) { /* No selecting ID? or >2 IDs on bus? */ printf("%s: bad reselect, data=0x%x\n", sc->sc_dev.dv_xname, data); return; } NCR_TRACE("reselect: target=0x%x\n", target); /* Raise BSY to acknowledge target reselection. */ *(sc->sci_icmd) = SCI_ICMD_BSY; /* Wait for target to drop SEL. */ timo = ncr5380_wait_nrq_timo; for (;;) { bus = *(sc->sci_bus_csr); if ((bus & SCI_BUS_SEL) == 0) break; /* success */ if (--timo <= 0) { printf("%s: reselect, SEL stuck, bus=0x%x\n", sc->sc_dev.dv_xname, bus); NCR_BREAK(); /* assume connected (fail later if not) */ break; } delay(2); } /* Now we drop BSY, and we are connected. */ *(sc->sci_icmd) = 0; *sc->sci_sel_enb = 0; SCI_CLR_INTR(sc); /* * At this point the target should send an IDENTIFY message, * which will permit us to determine the reselecting LUN. * If not, we assume LUN 0. */ lun = 0; /* Wait for REQ before reading bus phase. */ if (ncr5380_wait_req(sc)) { printf("%s: reselect, no REQ\n", sc->sc_dev.dv_xname); /* Try to send an ABORT message. */ goto abort; } phase = SCI_BUS_PHASE(*sc->sci_bus_csr); if (phase != PHASE_MSG_IN) { printf("%s: reselect, phase=%d\n", sc->sc_dev.dv_xname, phase); goto abort; } /* Ack. the change to PHASE_MSG_IN */ *(sc->sci_tcmd) = PHASE_MSG_IN; /* Peek at the message byte without consuming it! */ msg = *(sc->sci_data); if ((msg & 0x80) == 0) { printf("%s: reselect, not identify, msg=%d\n", sc->sc_dev.dv_xname, msg); goto abort; } lun = msg & 7; /* We now know target/LUN. Do we have the request? */ sr = sc->sc_matrix[target][lun]; if (sr) { /* We now have a nexus. */ sc->sc_state |= NCR_WORKING; sc->sc_current = sr; NCR_TRACE("reselect: resume sr=0x%x\n", (long)sr); /* Implicit restore pointers message */ sc->sc_dataptr = sr->sr_dataptr; sc->sc_datalen = sr->sr_datalen; sc->sc_prevphase = PHASE_INVALID; sc->sc_msgpriq = 0; sc->sc_msgoutq = 0; sc->sc_msgout = 0; /* * Another hack for the Sun3 "si", which needs * some setup done to its DMA engine before the * target puts the SCSI bus into any DATA phase. */ if (sr->sr_dma_hand && sc->sc_dma_setup) { NCR_TRACE("reselect: call DMA setup, dh=0x%x\n", (long) sr->sr_dma_hand); sc->sc_dma_setup(sc); } /* Now consume the IDENTIFY message. */ ncr5380_pio_in(sc, PHASE_MSG_IN, 1, &msg); return; } printf("%s: phantom reselect: target=%d, LUN=%d\n", sc->sc_dev.dv_xname, target, lun); abort: /* * Try to send an ABORT message. This makes us * temporarily busy, but no current command... */ sc->sc_state |= NCR_ABORTING; /* Raise ATN, delay, raise ACK... */ icmd = SCI_ICMD_ATN; *sc->sci_icmd = icmd; delay(2); /* Now consume the IDENTIFY message. */ ncr5380_pio_in(sc, PHASE_MSG_IN, 1, &msg); /* Finally try to send the ABORT. */ sc->sc_prevphase = PHASE_INVALID; sc->sc_msgpriq = SEND_ABORT; ncr5380_msg_out(sc); *(sc->sci_tcmd) = PHASE_INVALID; *sc->sci_sel_enb = 0; SCI_CLR_INTR(sc); *sc->sci_sel_enb = 0x80; sc->sc_state &= ~NCR_ABORTING; } /* * Select target: xs is the transfer that we are selecting for. * sc->sc_current should be NULL. * * Returns: * sc->sc_current != NULL ==> we were reselected (race!) * XS_NOERROR ==> selection worked * XS_BUSY ==> lost arbitration * XS_SELTIMEOUT ==> no response to selection */ static int ncr5380_select(sc, sr) struct ncr5380_softc *sc; struct sci_req *sr; { int timo; u_char bus, data, icmd; /* Check for reselect */ ncr5380_reselect(sc); if (sc->sc_current) { NCR_TRACE("select: reselect, cur=0x%x\n", (long) sc->sc_current); return XS_BUSY; /* reselected */ } /* * Set phase bits to 0, otherwise the 5380 won't drive the bus during * selection. */ *sc->sci_tcmd = PHASE_DATA_OUT; *sc->sci_icmd = icmd = 0; *sc->sci_mode = 0; /* * Arbitrate for the bus. The 5380 takes care of the * time-critical bus interactions. We set our ID bit * in the output data register and set MODE_ARB. The * 5380 watches for the required "bus free" period. * If and when the "bus free" period is detected, the * 5380 drives BSY, drives the data bus, and sets the * "arbitration in progress" (AIP) bit to let us know * arbitration has started (and that it asserts BSY). * We then wait for one arbitration delay (2.2uS) and * check the ICMD_LST bit, which will be set if some * other target drives SEL during arbitration. */ *(sc->sci_odata) = 0x80; /* OUR_ID */ *(sc->sci_mode) = SCI_MODE_ARB; /* Wait for ICMD_AIP. */ timo = ncr5380_wait_req_timo; for (;;) { if (*(sc->sci_icmd) & SCI_ICMD_AIP) break; if (--timo <= 0) { /* Did not see any "bus free" period. */ *sc->sci_mode = 0; NCR_TRACE("select: bus busy, rc=%d\n", XS_BUSY); return XS_BUSY; } delay(2); } NCR_TRACE("select: have AIP after %d loops\n", ncr5380_wait_req_timo - timo); /* Got AIP. Wait one arbitration delay (2.2 uS.) */ delay(3); /* Check for ICMD_LST */ if (*(sc->sci_icmd) & SCI_ICMD_LST) { /* Some other target asserted SEL. */ *sc->sci_mode = 0; NCR_TRACE("select: lost one, rc=%d\n", XS_BUSY); ncr5380_reselect(sc); /* XXX */ return XS_BUSY; } /* * No other device has declared itself the winner. * The spec. says to check for higher IDs, but we * are always the highest (ID=7) so don't bother. * We can now declare victory by asserting SEL. * * Note that the 5380 is asserting BSY because we * asked it to do arbitration. We will now hold * BSY directly so we can turn off ARB mode. */ icmd = (SCI_ICMD_BSY | SCI_ICMD_SEL); *sc->sci_icmd = icmd; /* * "The SCSI device that wins arbitration shall wait * at least a bus clear delay plus a bus settle delay * after asserting the SEL signal before changing * any [other] signal." (1.2uS. total) */ delay(2); #if 1 /* * XXX: Check one last time to see if we really * XXX: did win arbitration. (too paranoid?) */ if (*(sc->sci_icmd) & SCI_ICMD_LST) { *sc->sci_icmd = 0; *sc->sci_mode = 0; NCR_TRACE("select: lost two, rc=%d\n", XS_BUSY); return XS_BUSY; } #endif /* Leave ARB mode Now that we drive BSY+SEL */ *sc->sci_mode = 0; *sc->sci_sel_enb = 0; /* * Arbitration is complete. Now do selection: * Drive the data bus with the ID bits for both * the host and target. Also set ATN now, to * ask the target for a messgae out phase. */ data = 0x80 | (1 << sr->sr_target); *(sc->sci_odata) = data; icmd |= (SCI_ICMD_DATA | SCI_ICMD_ATN); *(sc->sci_icmd) = icmd; delay(2); /* two deskew delays. */ /* De-assert BSY (targets sample the data now). */ icmd &= ~SCI_ICMD_BSY; *(sc->sci_icmd) = icmd; delay(3); /* Bus settle delay. */ /* * Wait for the target to assert BSY. * SCSI spec. says wait for 250 mS. */ for (timo = 25000;;) { if (*sc->sci_bus_csr & SCI_BUS_BSY) goto success; if (--timo <= 0) break; delay(10); } /* * There is no reaction from the target. Start the selection * timeout procedure. We release the databus but keep SEL+ATN * asserted. After that we wait a 'selection abort time' (200 * usecs) and 2 deskew delays (90 ns) and check BSY again. * When BSY is asserted, we assume the selection succeeded, * otherwise we release the bus. */ icmd &= ~SCI_ICMD_DATA; *(sc->sci_icmd) = icmd; delay(201); if ((*sc->sci_bus_csr & SCI_BUS_BSY) == 0) { /* Really no device on bus */ *sc->sci_tcmd = PHASE_INVALID; *sc->sci_icmd = 0; *sc->sci_mode = 0; *sc->sci_sel_enb = 0; SCI_CLR_INTR(sc); *sc->sci_sel_enb = 0x80; NCR_TRACE("select: device down, rc=%d\n", XS_SELTIMEOUT); return XS_SELTIMEOUT; } success: /* * The target is now driving BSY, so we can stop * driving SEL and the data bus (keep ATN true). * Configure the ncr5380 to monitor BSY, parity. */ icmd &= ~(SCI_ICMD_DATA | SCI_ICMD_SEL); *sc->sci_icmd = icmd; /* XXX - Make parity checking optional? */ *sc->sci_mode = (SCI_MODE_MONBSY | SCI_MODE_PAR_CHK); return XS_NOERROR; } /***************************************************************** * Functions to handle each info. transfer phase: *****************************************************************/ /* * The message system: * * This is a revamped message system that now should easier accomodate * new messages, if necessary. * * Currently we accept these messages: * IDENTIFY (when reselecting) * COMMAND COMPLETE # (expect bus free after messages marked #) * NOOP * MESSAGE REJECT * SYNCHRONOUS DATA TRANSFER REQUEST * SAVE DATA POINTER * RESTORE POINTERS * DISCONNECT # * * We may send these messages in prioritized order: * BUS DEVICE RESET # if SCSI_RESET & xs->flags (or in weird sits.) * MESSAGE PARITY ERROR par. err. during MSGI * MESSAGE REJECT If we get a message we don't know how to handle * ABORT # send on errors * INITIATOR DETECTED ERROR also on errors (SCSI2) (during info xfer) * IDENTIFY At the start of each transfer * SYNCHRONOUS DATA TRANSFER REQUEST if appropriate * NOOP if nothing else fits the bill ... */ #define IS1BYTEMSG(m) (((m) != 0x01 && (m) < 0x20) || (m) >= 0x80) #define IS2BYTEMSG(m) (((m) & 0xf0) == 0x20) #define ISEXTMSG(m) ((m) == 0x01) /* * Precondition: * The SCSI bus is already in the MSGI phase and there is a message byte * on the bus, along with an asserted REQ signal. * * Our return value determines whether our caller, ncr5380_machine() * will expect to see another REQ (and possibly phase change). */ static int ncr5380_msg_in(sc) register struct ncr5380_softc *sc; { struct sci_req *sr = sc->sc_current; int n, phase, timo; int act_flags; register u_char icmd; /* acknowledge phase change */ *sc->sci_tcmd = PHASE_MSG_IN; act_flags = ACT_CONTINUE; icmd = *sc->sci_icmd & SCI_ICMD_RMASK; if (sc->sc_prevphase == PHASE_MSG_IN) { /* This is a continuation of the previous message. */ n = sc->sc_imp - sc->sc_imess; NCR_TRACE("msg_in: continuation, n=%d\n", n); goto nextbyte; } /* This is a new MESSAGE IN phase. Clean up our state. */ sc->sc_state &= ~NCR_DROP_MSGIN; nextmsg: n = 0; sc->sc_imp = &sc->sc_imess[n]; nextbyte: /* * Read a whole message, but don't ack the last byte. If we reject the * message, we have to assert ATN during the message transfer phase * itself. */ for (;;) { /* * Read a message byte. * First, check BSY, REQ, phase... */ if (!SCI_BUSY(sc)) { NCR_TRACE("msg_in: lost BSY, n=%d\n", n); /* XXX - Assume the command completed? */ act_flags |= (ACT_DISCONNECT | ACT_CMD_DONE); return (act_flags); } if (ncr5380_wait_req(sc)) { NCR_TRACE("msg_in: BSY but no REQ, n=%d\n", n); /* Just let ncr5380_machine() handle it... */ return (act_flags); } phase = SCI_BUS_PHASE(*sc->sci_bus_csr); if (phase != PHASE_MSG_IN) { /* * Target left MESSAGE IN, probably because it * a) noticed our ATN signal, or * b) ran out of messages. */ return (act_flags); } /* Still in MESSAGE IN phase, and REQ is asserted. */ if (*sc->sci_csr & SCI_CSR_PERR) { ncr_sched_msgout(sc, SEND_PARITY_ERROR); sc->sc_state |= NCR_DROP_MSGIN; } /* Gather incoming message bytes if needed. */ if ((sc->sc_state & NCR_DROP_MSGIN) == 0) { if (n >= NCR_MAX_MSG_LEN) { ncr_sched_msgout(sc, SEND_REJECT); sc->sc_state |= NCR_DROP_MSGIN; } else { *sc->sc_imp++ = *sc->sci_data; n++; /* * This testing is suboptimal, but most * messages will be of the one byte variety, so * it should not affect performance * significantly. */ if (n == 1 && IS1BYTEMSG(sc->sc_imess[0])) goto have_msg; if (n == 2 && IS2BYTEMSG(sc->sc_imess[0])) goto have_msg; if (n >= 3 && ISEXTMSG(sc->sc_imess[0]) && n == sc->sc_imess[1] + 2) goto have_msg; } } /* * If we reach this spot we're either: * a) in the middle of a multi-byte message, or * b) dropping bytes. */ /* Ack the last byte read. */ icmd |= SCI_ICMD_ACK; *sc->sci_icmd = icmd; if (ncr5380_wait_not_req(sc)) { NCR_TRACE("msg_in: drop, stuck REQ, n=%d\n", n); act_flags |= ACT_RESET_BUS; } icmd &= ~SCI_ICMD_ACK; *sc->sci_icmd = icmd; if (act_flags != ACT_CONTINUE) return (act_flags); /* back to nextbyte */ } have_msg: /* We now have a complete message. Parse it. */ switch (sc->sc_imess[0]) { case MSG_CMDCOMPLETE: NCR_TRACE("msg_in: CMDCOMPLETE\n", 0); /* Target is about to disconnect. */ act_flags |= (ACT_DISCONNECT | ACT_CMD_DONE); break; case MSG_DISCONNECT: NCR_TRACE("msg_in: DISCONNECT\n", 0); /* Target is about to disconnect. */ act_flags |= ACT_DISCONNECT; break; case MSG_PARITY_ERROR: NCR_TRACE("msg_in: PARITY_ERROR\n", 0); /* Resend the last message. */ ncr_sched_msgout(sc, sc->sc_msgout); break; case MSG_MESSAGE_REJECT: /* The target rejects the last message we sent. */ NCR_TRACE("msg_in: got reject for 0x%x\n", sc->sc_msgout); switch (sc->sc_msgout) { case SEND_IDENTIFY: /* Really old target controller? */ /* XXX ... */ break; case SEND_INIT_DET_ERR: goto abort; } break; case MSG_NOOP: NCR_TRACE("msg_in: NOOP\n", 0); break; case MSG_SAVEDATAPOINTER: NCR_TRACE("msg_in: SAVE_PTRS\n", 0); sr->sr_dataptr = sc->sc_dataptr; sr->sr_datalen = sc->sc_datalen; break; case MSG_RESTOREPOINTERS: NCR_TRACE("msg_in: RESTORE_PTRS\n", 0); sc->sc_dataptr = sr->sr_dataptr; sc->sc_datalen = sr->sr_datalen; break; case MSG_EXTENDED: switch (sc->sc_imess[2]) { case MSG_EXT_SDTR: case MSG_EXT_WDTR: /* The ncr5380 can not do synchronous mode. */ goto reject; default: printf("%s: unrecognized MESSAGE EXTENDED; sending REJECT\n", sc->sc_dev.dv_xname); NCR_BREAK(); goto reject; } break; default: NCR_TRACE("msg_in: eh? imsg=0x%x\n", sc->sc_imess[0]); printf("%s: unrecognized MESSAGE; sending REJECT\n", sc->sc_dev.dv_xname); NCR_BREAK(); /* fallthrough */ reject: ncr_sched_msgout(sc, SEND_REJECT); break; abort: sc->sc_state |= NCR_ABORTING; ncr_sched_msgout(sc, SEND_ABORT); break; } /* Ack the last byte read. */ icmd |= SCI_ICMD_ACK; *sc->sci_icmd = icmd; if (ncr5380_wait_not_req(sc)) { NCR_TRACE("msg_in: last, stuck REQ, n=%d\n", n); act_flags |= ACT_RESET_BUS; } icmd &= ~SCI_ICMD_ACK; *sc->sci_icmd = icmd; /* Go get the next message, if any. */ if (act_flags == ACT_CONTINUE) goto nextmsg; return (act_flags); } /* * The message out (and in) stuff is a bit complicated: * If the target requests another message (sequence) without * having changed phase in between it really asks for a * retransmit, probably due to parity error(s). * The following messages can be sent: * IDENTIFY @ These 4 stem from SCSI command activity * SDTR @ * WDTR @ * DEV_RESET @ * REJECT if MSGI doesn't make sense * PARITY_ERROR if parity error while in MSGI * INIT_DET_ERR if parity error while not in MSGI * ABORT if INIT_DET_ERR rejected * NOOP if asked for a message and there's nothing to send * * Note that we call this one with (sc_current == NULL) * when sending ABORT for unwanted reselections. */ static int ncr5380_msg_out(sc) register struct ncr5380_softc *sc; { struct sci_req *sr = sc->sc_current; int n, phase, resel; int progress, act_flags; register u_char icmd; /* acknowledge phase change */ *sc->sci_tcmd = PHASE_MSG_OUT; progress = 0; /* did we send any messages? */ act_flags = ACT_CONTINUE; /* * Set ATN. If we're just sending a trivial 1-byte message, * we'll clear ATN later on anyway. Also drive the data bus. */ icmd = *sc->sci_icmd & SCI_ICMD_RMASK; icmd |= (SCI_ICMD_ATN | SCI_ICMD_DATA); *sc->sci_icmd = icmd; if (sc->sc_prevphase == PHASE_MSG_OUT) { if (sc->sc_omp == sc->sc_omess) { /* * This is a retransmission. * * We get here if the target stayed in MESSAGE OUT * phase. Section 5.1.9.2 of the SCSI 2 spec indicates * that all of the previously transmitted messages must * be sent again, in the same order. Therefore, we * requeue all the previously transmitted messages, and * start again from the top. Our simple priority * scheme keeps the messages in the right order. */ sc->sc_msgpriq |= sc->sc_msgoutq; NCR_TRACE("msg_out: retrans priq=0x%x\n", sc->sc_msgpriq); } else { /* This is a continuation of the previous message. */ n = sc->sc_omp - sc->sc_omess; NCR_TRACE("msg_out: continuation, n=%d\n", n); goto nextbyte; } } /* No messages transmitted so far. */ sc->sc_msgoutq = 0; nextmsg: /* Pick up highest priority message. */ sc->sc_msgout = sc->sc_msgpriq & -sc->sc_msgpriq; sc->sc_msgpriq &= ~sc->sc_msgout; sc->sc_msgoutq |= sc->sc_msgout; /* Build the outgoing message data. */ switch (sc->sc_msgout) { case SEND_IDENTIFY: NCR_TRACE("msg_out: SEND_IDENTIFY\n", 0); if (sr == NULL) { printf("%s: SEND_IDENTIFY while not connected; sending NOOP\n", sc->sc_dev.dv_xname); NCR_BREAK(); goto noop; } resel = (sc->sc_flags & NCR5380_PERMIT_RESELECT) ? 1 : 0; resel &= (sr->sr_flags & (SR_IMMED | SR_SENSE)) ? 0 : 1; sc->sc_omess[0] = MSG_IDENTIFY(sr->sr_lun, resel); n = 1; break; case SEND_DEV_RESET: NCR_TRACE("msg_out: SEND_DEV_RESET\n", 0); /* Expect disconnect after this! */ /* XXX: Kill jobs for this target? */ act_flags |= (ACT_DISCONNECT | ACT_CMD_DONE); sc->sc_omess[0] = MSG_BUS_DEV_RESET; n = 1; break; case SEND_REJECT: NCR_TRACE("msg_out: SEND_REJECT\n", 0); sc->sc_omess[0] = MSG_MESSAGE_REJECT; n = 1; break; case SEND_PARITY_ERROR: NCR_TRACE("msg_out: SEND_PARITY_ERROR\n", 0); sc->sc_omess[0] = MSG_PARITY_ERROR; n = 1; break; case SEND_INIT_DET_ERR: NCR_TRACE("msg_out: SEND_INIT_DET_ERR\n", 0); sc->sc_omess[0] = MSG_INITIATOR_DET_ERR; n = 1; break; case SEND_ABORT: NCR_TRACE("msg_out: SEND_ABORT\n", 0); /* Expect disconnect after this! */ /* XXX: Set error flag? */ act_flags |= (ACT_DISCONNECT | ACT_CMD_DONE); sc->sc_omess[0] = MSG_ABORT; n = 1; break; case 0: printf("%s: unexpected MESSAGE OUT; sending NOOP\n", sc->sc_dev.dv_xname); NCR_BREAK(); noop: NCR_TRACE("msg_out: send NOOP\n", 0); sc->sc_omess[0] = MSG_NOOP; n = 1; break; default: printf("%s: weird MESSAGE OUT; sending NOOP\n", sc->sc_dev.dv_xname); NCR_BREAK(); goto noop; } sc->sc_omp = &sc->sc_omess[n]; nextbyte: /* Send message bytes. */ while (n > 0) { /* * Send a message byte. * First check BSY, REQ, phase... */ if (!SCI_BUSY(sc)) { NCR_TRACE("msg_out: lost BSY, n=%d\n", n); goto out; } if (ncr5380_wait_req(sc)) { NCR_TRACE("msg_out: no REQ, n=%d\n", n); goto out; } phase = SCI_BUS_PHASE(*sc->sci_bus_csr); if (phase != PHASE_MSG_OUT) { /* * Target left MESSAGE OUT, possibly to reject * our message. */ NCR_TRACE("msg_out: new phase=%d\n", phase); goto out; } /* Yes, we can send this message byte. */ --n; /* Clear ATN before last byte if this is the last message. */ if (n == 0 && sc->sc_msgpriq == 0) { icmd &= ~SCI_ICMD_ATN; *sc->sci_icmd = icmd; /* 2 deskew delays */ delay(2); /* XXX */ } /* Put data on the bus. */ *sc->sci_odata = *--sc->sc_omp; /* Raise ACK to tell target data is on the bus. */ icmd |= SCI_ICMD_ACK; *sc->sci_icmd = icmd; /* Wait for REQ to be negated. */ if (ncr5380_wait_not_req(sc)) { NCR_TRACE("msg_out: stuck REQ, n=%d\n", n); act_flags |= ACT_RESET_BUS; } /* Finally, drop ACK. */ icmd &= ~SCI_ICMD_ACK; *sc->sci_icmd = icmd; /* Stuck bus or something... */ if (act_flags & ACT_RESET_BUS) goto out; } progress++; /* We get here only if the entire message has been transmitted. */ if (sc->sc_msgpriq != 0) { /* There are more outgoing messages. */ goto nextmsg; } /* * The last message has been transmitted. We need to remember the last * message transmitted (in case the target switches to MESSAGE IN phase * and sends a MESSAGE REJECT), and the list of messages transmitted * this time around (in case the target stays in MESSAGE OUT phase to * request a retransmit). */ out: /* Stop driving the data bus. */ icmd &= ~SCI_ICMD_DATA; *sc->sci_icmd = icmd; if (!progress) act_flags |= ACT_RESET_BUS; return (act_flags); } /* * Handle command phase. */ static int ncr5380_command(sc) struct ncr5380_softc *sc; { struct sci_req *sr = sc->sc_current; struct scsi_xfer *xs = sr->sr_xs; struct scsi_sense rqs; int len; /* acknowledge phase change */ *sc->sci_tcmd = PHASE_COMMAND; if (sr->sr_flags & SR_SENSE) { rqs.opcode = REQUEST_SENSE; rqs.byte2 = xs->sc_link->lun << 5; rqs.length = sizeof(xs->sense); rqs.unused[0] = rqs.unused[1] = rqs.control = 0; len = ncr5380_pio_out(sc, PHASE_COMMAND, sizeof(rqs), (u_char *)&rqs); } else { /* Assume command can be sent in one go. */ /* XXX: Do this using DMA, and get a phase change intr? */ len = ncr5380_pio_out(sc, PHASE_COMMAND, xs->cmdlen, (u_char *)xs->cmd); } if (len != xs->cmdlen) { #ifdef DEBUG printf("ncr5380_command: short transfer: wanted %d got %d.\n", xs->cmdlen, len); ncr5380_show_scsi_cmd(xs); NCR_BREAK(); #endif if (len < 6) { xs->error = XS_DRIVER_STUFFUP; sc->sc_state |= NCR_ABORTING; ncr_sched_msgout(sc, SEND_ABORT); } } return ACT_CONTINUE; } /* * Handle either data_in or data_out */ static int ncr5380_data_xfer(sc, phase) struct ncr5380_softc *sc; int phase; { struct sci_req *sr = sc->sc_current; struct scsi_xfer *xs = sr->sr_xs; int expected_phase; int i, len; if (sr->sr_flags & SR_SENSE) { NCR_TRACE("data_xfer: get sense, sr=0x%x\n", (long)sr); if (phase != PHASE_DATA_IN) { printf("%s: sense phase error\n", sc->sc_dev.dv_xname); goto abort; } /* acknowledge phase change */ *sc->sci_tcmd = PHASE_DATA_IN; len = ncr5380_pio_in(sc, phase, sizeof(xs->sense), (u_char *)&xs->sense); return ACT_CONTINUE; } /* * When aborting a command, disallow any data phase. */ if (sc->sc_state & NCR_ABORTING) { printf("%s: aborting, but phase=%s (reset)\n", sc->sc_dev.dv_xname, phase_names[phase & 7]); return ACT_RESET_BUS; /* XXX */ } /* Validate expected phase (data_in or data_out) */ expected_phase = (xs->flags & SCSI_DATA_OUT) ? PHASE_DATA_OUT : PHASE_DATA_IN; if (phase != expected_phase) { printf("%s: data phase error\n", sc->sc_dev.dv_xname); goto abort; } /* Make sure we have some data to move. */ if (sc->sc_datalen <= 0) { printf("%s: can not transfer more data\n", sc->sc_dev.dv_xname); goto abort; } /* * Attempt DMA only if dma_alloc gave us a DMA handle AND * there is enough left to transfer so DMA is worth while. */ if (sr->sr_dma_hand && (sc->sc_datalen >= sc->sc_min_dma_len)) { /* * OK, really start DMA. Note, the MD start function * is responsible for setting the TCMD register, etc. * (Acknowledge the phase change there, not here.) */ NCR_TRACE("data_xfer: dma_start, dh=0x%x\n", (long) sr->sr_dma_hand); (*sc->sc_dma_start)(sc); return ACT_WAIT_DMA; } /* * Doing PIO for data transfer. (Possibly "Pseudo DMA") * XXX: Do PDMA functions need to set tcmd later? */ NCR_TRACE("data_xfer: doing PIO, len=%d\n", sc->sc_datalen); /* acknowledge phase change */ *sc->sci_tcmd = phase; /* XXX: OK for PDMA? */ if (phase == PHASE_DATA_OUT) { len = (*sc->sc_pio_out)(sc, phase, sc->sc_datalen, sc->sc_dataptr); } else { len = (*sc->sc_pio_in) (sc, phase, sc->sc_datalen, sc->sc_dataptr); } sc->sc_dataptr += len; sc->sc_datalen -= len; NCR_TRACE("data_xfer: did PIO, resid=%d\n", sc->sc_datalen); return (ACT_CONTINUE); abort: sc->sc_state |= NCR_ABORTING; ncr_sched_msgout(sc, SEND_ABORT); return (ACT_CONTINUE); } static int ncr5380_status(sc) struct ncr5380_softc *sc; { int len; u_char status; struct sci_req *sr = sc->sc_current; struct scsi_xfer *xs = sr->sr_xs; /* acknowledge phase change */ *sc->sci_tcmd = PHASE_STATUS; len = ncr5380_pio_in(sc, PHASE_STATUS, 1, &status); if (len) { sr->sr_status = status; } else { printf("ncr5380_status: none?\n"); } return ACT_CONTINUE; } /* * This is the big state machine that follows SCSI phase changes. * This is somewhat like a co-routine. It will do a SCSI command, * and exit if the command is complete, or if it must wait, i.e. * for DMA to complete or for reselect to resume the job. * * The bus must be selected, and we need to know which command is * being undertaken. */ static void ncr5380_machine(sc) struct ncr5380_softc *sc; { struct sci_req *sr; struct scsi_xfer *xs; int act_flags, phase, timo; #ifdef DIAGNOSTIC if (sc->sc_state == NCR_IDLE) panic("ncr5380_machine: state=idle"); if (sc->sc_current == NULL) panic("ncr5380_machine: no current cmd"); #endif sr = sc->sc_current; xs = sr->sr_xs; act_flags = ACT_CONTINUE; /* * This will be called by ncr5380_intr() when DMA is * complete. Must stop DMA before touching the 5380 or * there will be "register conflict" errors. */ if (sc->sc_state & NCR_DOINGDMA) { /* Pick-up where where we left off... */ goto dma_done; } next_phase: if (!SCI_BUSY(sc)) { /* Unexpected disconnect */ printf("ncr5380_machine: unexpected disconnect.\n"); xs->error = XS_DRIVER_STUFFUP; act_flags |= (ACT_DISCONNECT | ACT_CMD_DONE); goto do_actions; } /* * Wait for REQ before reading the phase. * Need to wait longer than usual here, because * some devices are just plain slow... */ timo = ncr5380_wait_phase_timo; for (;;) { if (*sc->sci_bus_csr & SCI_BUS_REQ) break; if (--timo <= 0) { if (sc->sc_state & NCR_ABORTING) { printf("%s: no REQ while aborting, reset\n", sc->sc_dev.dv_xname); act_flags |= ACT_RESET_BUS; goto do_actions; } printf("%s: no REQ for next phase, abort\n", sc->sc_dev.dv_xname); sc->sc_state |= NCR_ABORTING; ncr_sched_msgout(sc, SEND_ABORT); goto next_phase; } delay(100); } phase = SCI_BUS_PHASE(*sc->sci_bus_csr); NCR_TRACE("machine: phase=%s\n", (long) phase_names[phase & 7]); /* * We assume that the device knows what it's doing, * so any phase is good. */ #if 0 /* * XXX: Do not ACK the phase yet! do it later... * XXX: ... each phase routine does that itself. * In particular, DMA needs it done LATER. */ *sc->sci_tcmd = phase; /* acknowledge phase change */ #endif switch (phase) { case PHASE_DATA_OUT: case PHASE_DATA_IN: act_flags = ncr5380_data_xfer(sc, phase); break; case PHASE_COMMAND: act_flags = ncr5380_command(sc); break; case PHASE_STATUS: act_flags = ncr5380_status(sc); break; case PHASE_MSG_OUT: act_flags = ncr5380_msg_out(sc); break; case PHASE_MSG_IN: act_flags = ncr5380_msg_in(sc); break; default: printf("ncr5380_machine: Unexpected phase 0x%x\n", phase); sc->sc_state |= NCR_ABORTING; ncr_sched_msgout(sc, SEND_ABORT); goto next_phase; } /* switch */ sc->sc_prevphase = phase; do_actions: __asm("_ncr5380_actions:"); if (act_flags & ACT_WAIT_DMA) { act_flags &= ~ACT_WAIT_DMA; /* Wait for DMA to complete (polling, or interrupt). */ if ((sr->sr_flags & SR_IMMED) == 0) { NCR_TRACE("machine: wait for DMA intr.\n", 0); return; /* will resume at dma_done */ } /* Busy-wait for it to finish. */ NCR_TRACE("machine: dma_poll, dh=0x%x\n", (long) sr->sr_dma_hand); (*sc->sc_dma_poll)(sc); dma_done: /* Return here after interrupt. */ if (sr->sr_flags & SR_OVERDUE) sc->sc_state |= NCR_ABORTING; NCR_TRACE("machine: dma_stop, dh=0x%x\n", (long) sr->sr_dma_hand); (*sc->sc_dma_stop)(sc); SCI_CLR_INTR(sc); /* XXX */ /* * While DMA is running we can not touch the SBC, * so various places just set NCR_ABORTING and * expect us the "kick it" when DMA is done. */ if (sc->sc_state & NCR_ABORTING) { ncr_sched_msgout(sc, SEND_ABORT); } } /* * Check for parity error. * XXX - better place to check? */ if (*(sc->sci_csr) & SCI_CSR_PERR) { printf("%s: parity error!\n", sc->sc_dev.dv_xname); /* XXX: sc->sc_state |= NCR_ABORTING; */ ncr_sched_msgout(sc, SEND_PARITY_ERROR); } if (act_flags == ACT_CONTINUE) goto next_phase; /* All other actions "break" from the loop. */ NCR_TRACE("machine: act_flags=0x%x\n", act_flags); if (act_flags & ACT_RESET_BUS) { act_flags |= ACT_CMD_DONE; /* * Reset the SCSI bus, usually due to a timeout. * The error code XS_TIMEOUT allows retries. */ sc->sc_state |= NCR_ABORTING; printf("%s: reset SCSI bus for TID=%d LUN=%d\n", sc->sc_dev.dv_xname, sr->sr_target, sr->sr_lun); ncr5380_reset_scsibus(sc); } if (act_flags & ACT_CMD_DONE) { act_flags |= ACT_DISCONNECT; /* Need to call scsi_done() */ /* XXX: from the aic6360 driver, but why? */ if (sc->sc_datalen < 0) { printf("%s: %d extra bytes from %d:%d\n", sc->sc_dev.dv_xname, -sc->sc_datalen, sr->sr_target, sr->sr_lun); sc->sc_datalen = 0; } xs->resid = sc->sc_datalen; /* Note: this will clear sc_current */ NCR_TRACE("machine: call done, cur=0x%x\n", (long)sr); ncr5380_done(sc); } if (act_flags & ACT_DISCONNECT) { /* * The device has dropped BSY (or will soon). * Return and let ncr5380_sched() do its thing. */ *sc->sci_icmd = 0; *sc->sci_mode = 0; *sc->sci_tcmd = PHASE_INVALID; *sc->sci_sel_enb = 0; SCI_CLR_INTR(sc); *sc->sci_sel_enb = 0x80; if ((act_flags & ACT_CMD_DONE) == 0) { __asm("_ncr5380_disconnected:"); NCR_TRACE("machine: discon, cur=0x%x\n", (long)sr); } /* * We may be here due to a disconnect message, * in which case we did NOT call ncr5380_done, * and we need to clear sc_current. */ sc->sc_state = NCR_IDLE; sc->sc_current = NULL; /* Paranoia: clear everything. */ sc->sc_dataptr = NULL; sc->sc_datalen = 0; sc->sc_prevphase = PHASE_INVALID; sc->sc_msgpriq = 0; sc->sc_msgoutq = 0; sc->sc_msgout = 0; /* Our caller will re-enable interrupts. */ } } #ifdef DEBUG static void ncr5380_show_scsi_cmd(xs) struct scsi_xfer *xs; { u_char *b = (u_char *) xs->cmd; int i = 0; if ( ! ( xs->flags & SCSI_RESET ) ) { printf("si(%d:%d:%d)-", xs->sc_link->scsibus, xs->sc_link->target, xs->sc_link->lun); while (i < xs->cmdlen) { if (i) printf(","); printf("%x",b[i++]); } printf("-\n"); } else { printf("si(%d:%d:%d)-RESET-\n", xs->sc_link->scsibus, xs->sc_link->target, xs->sc_link->lun); } } static void ncr5380_show_sense(xs) struct scsi_xfer *xs; { u_char *b = (u_char *)&xs->sense; int i; printf("sense:"); for (i = 0; i < sizeof(xs->sense); i++) printf(" %02x", b[i]); printf("\n"); } int ncr5380_traceidx = 0; #define TRACE_MAX 1024 struct trace_ent { char *msg; long val; } ncr5380_tracebuf[TRACE_MAX]; void ncr5380_trace(msg, val) char *msg; long val; { register struct trace_ent *tr; register int s; s = splhigh(); tr = &ncr5380_tracebuf[ncr5380_traceidx]; ncr5380_traceidx++; if (ncr5380_traceidx >= TRACE_MAX) ncr5380_traceidx = 0; tr->msg = msg; tr->val = val; splx(s); } #ifdef DDB void ncr5380_clear_trace() { ncr5380_traceidx = 0; bzero((char*) ncr5380_tracebuf, sizeof(ncr5380_tracebuf)); } void ncr5380_show_trace() { struct trace_ent *tr; int idx; idx = ncr5380_traceidx; do { tr = &ncr5380_tracebuf[idx]; idx++; if (idx >= TRACE_MAX) idx = 0; if (tr->msg) db_printf(tr->msg, tr->val); } while (idx != ncr5380_traceidx); } void ncr5380_show_req(sr) struct sci_req *sr; { struct scsi_xfer *xs = sr->sr_xs; db_printf("TID=%d ", sr->sr_target); db_printf("LUN=%d ", sr->sr_lun); db_printf("dh=0x%x ", sr->sr_dma_hand); db_printf("dptr=0x%x ", sr->sr_dataptr); db_printf("dlen=0x%x ", sr->sr_datalen); db_printf("flags=%d ", sr->sr_flags); db_printf("stat=%d ", sr->sr_status); if (xs == NULL) { db_printf("(xs=NULL)\n"); return; } db_printf("\n"); #ifdef SCSIDEBUG show_scsi_xs(xs); #else db_printf("xs=0x%x\n", xs); #endif } void ncr5380_show_state() { struct ncr5380_softc *sc; struct sci_req *sr; int i, j, k; sc = ncr5380_debug_sc; if (sc == NULL) { db_printf("ncr5380_debug_sc == NULL\n"); return; } db_printf("sc_ncmds=%d\n", sc->sc_ncmds); k = -1; /* which is current? */ for (i = 0; i < SCI_OPENINGS; i++) { sr = &sc->sc_ring[i]; if (sr->sr_xs) { if (sr == sc->sc_current) k = i; db_printf("req %d: (sr=0x%x)", i, (long)sr); ncr5380_show_req(sr); } } db_printf("sc_rr=%d, current=%d\n", sc->sc_rr, k); db_printf("Active request matrix:\n"); for(i = 0; i < 8; i++) { /* targets */ for (j = 0; j < 8; j++) { /* LUN */ sr = sc->sc_matrix[i][j]; if (sr) { db_printf("TID=%d LUN=%d sr=0x%x\n", i, j, (long)sr); } } } db_printf("sc_state=0x%x\n", sc->sc_state); db_printf("sc_current=0x%x\n", sc->sc_current); db_printf("sc_dataptr=0x%x\n", sc->sc_dataptr); db_printf("sc_datalen=0x%x\n", sc->sc_datalen); db_printf("sc_prevphase=%d\n", sc->sc_prevphase); db_printf("sc_msgpriq=0x%x\n", sc->sc_msgpriq); } #endif /* DDB */ #endif /* DEBUG */