/* * Copyright (c) 1995 Daniel Widenfalk * Copyright (c) 1994 Christian E. Hopps * Copyright (c) 1990 The Regents of the University of California. * All rights reserved. * * This code is derived from software contributed to Berkeley by * Van Jacobson of Lawrence Berkeley Laboratory. * * 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 the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)scsi.c 7.5 (Berkeley) 5/4/91 */ /* * AMIGA Emulex FAS216 scsi adaptor driver */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include void sfasinitialize __P((struct sfas_softc *)); u_int sfas_minphys __P((struct buf *bp)); int sfas_scsicmd __P((struct scsi_xfer *xs)); int sfas_donextcmd __P((struct sfas_softc *dev, struct sfas_pending *pendp)); void sfas_scsidone __P((struct sfas_softc *dev, struct scsi_xfer *xs, int stat)); void sfasintr __P((struct sfas_softc *dev)); void sfasiwait __P((struct sfas_softc *dev)); void sfasreset __P((struct sfas_softc *dev, int how)); int sfasselect __P((struct sfas_softc *dev, struct sfas_pending *pendp, unsigned char *cbuf, int clen, unsigned char *buf, int len, int mode)); void sfasicmd __P((struct sfas_softc *dev, struct sfas_pending *pendp)); /* * Initialize these to make 'em patchable. Defaults to enable sync and discon. */ u_char sfas_inhibit_sync[8] = { 0, 0, 0, 0, 0, 0, 0, 0 }; u_char sfas_inhibit_disc[8] = { 0, 0, 0, 0, 0, 0, 0, 0 }; #ifdef DEBUG #define QPRINTF(a) if (sfas_debug > 1) printf a int sfas_debug = 0; #else #define QPRINTF #endif /* * default minphys routine for sfas based controllers */ u_int sfas_minphys(bp) struct buf *bp; { /* * No max transfer at this level. */ return (minphys(bp)); } /* * Initialize the nexus structs. */ void sfas_init_nexus(dev, nexus) struct sfas_softc *dev; struct nexus *nexus; { bzero(nexus, sizeof(struct nexus)); nexus->state = SFAS_NS_IDLE; nexus->period = 200; nexus->offset = 0; nexus->syncper = 5; nexus->syncoff = 0; nexus->config3 = dev->sc_config3 & ~SFAS_CFG3_FASTSCSI; } void sfasinitialize(dev) struct sfas_softc *dev; { sfas_regmap_p rp; u_int *pte, page; int i; dev->sc_led_status = 0; TAILQ_INIT(&dev->sc_xs_pending); TAILQ_INIT(&dev->sc_xs_free); /* * Initialize the sfas_pending structs and link them into the free list. We * have to set vm_link_data.pages to 0 or the vm FIX won't work. */ for(i=0; isc_xs_store[i].vm_link_data.pages = 0; #endif TAILQ_INSERT_TAIL(&dev->sc_xs_free, &dev->sc_xs_store[i], link); } /* * Calculate the correct clock conversion factor 2 <= factor <= 8, i.e. set * the factor to clock_freq / 5 (int). */ if (dev->sc_clock_freq <= 10) dev->sc_clock_conv_fact = 2; if (dev->sc_clock_freq <= 40) dev->sc_clock_conv_fact = 2+((dev->sc_clock_freq-10)/5); else panic("sfasinitialize: Clock frequence too high"); /* Setup and save the basic configuration registers */ dev->sc_config1 = (dev->sc_host_id & SFAS_CFG1_BUS_ID_MASK); dev->sc_config2 = SFAS_CFG2_FEATURES_ENABLE; dev->sc_config3 = (dev->sc_clock_freq > 25 ? SFAS_CFG3_FASTCLK : 0); /* Precalculate timeout value and clock period. */ dev->sc_timeout_val = 1+dev->sc_timeout*dev->sc_clock_freq/ (7.682*dev->sc_clock_conv_fact); dev->sc_clock_period = 1000/dev->sc_clock_freq; sfasreset(dev, 1 | 2); /* Reset Chip and Bus */ dev->sc_units_disconnected = 0; dev->sc_msg_in_len = 0; dev->sc_msg_out_len = 0; dev->sc_flags = 0; for(i=0; i<8; i++) sfas_init_nexus(dev, &dev->sc_nexus[i]); /* * Setup bump buffer. If dev->sc_bump_pa has the upper bits set, we should * allocate z2-mem else we can allocate "any" memory. This code should check * that the bump-buffer is LW aligned, but I think alloc_z2mem/kmem_alloc * does that. */ if (dev->sc_bump_pa & 0xFF000000) { dev->sc_bump_va = (u_char *)alloc_z2mem(dev->sc_bump_sz); if (isztwomem(dev->sc_bump_va)) dev->sc_bump_pa = kvtop(dev->sc_bump_va); else dev->sc_bump_pa = (vm_offset_t) PREP_DMA_MEM(dev->sc_bump_va); } else { dev->sc_bump_va = (u_char *)kmem_alloc(kernel_map, dev->sc_bump_sz); dev->sc_bump_pa = kvtop(dev->sc_bump_va); } /* * Setup pages to noncachable, that way we don't have to flush the cache * every time we need "bumped" transfer. */ pte = kvtopte(dev->sc_bump_va); page= (u_int)dev->sc_bump_pa & PG_FRAME; *pte = PG_V | PG_RW | PG_CI | page; TBIAS(); printf(": dmabuf 0x%x", dev->sc_bump_pa); /* * FIX * The scsi drivers tend to allocate buffers from the kernel stacks. When the * kernel goes to sleep, it does a contect-switch thus removing the mapping * to the stack. To work around this we allocate MAXPHYS+alignment bytes * of virtual memory to which we can later map physical memory to. */ #ifdef SFAS_NEED_VM_PATCH vm_map_lock(kernel_map); /* Locate available space. */ if (vm_map_findspace(kernel_map, 0, MAXPHYS+NBPG, (vm_offset_t *)&dev->sc_vm_link)) { vm_map_unlock(kernel_map); panic("SFAS_SCSICMD: No VM space available."); } else { int offset; /* * Map space to virtual memory in kernel_map. This vm will always be available * to us during interrupt time. */ offset = (vm_offset_t)dev->sc_vm_link - VM_MIN_KERNEL_ADDRESS; printf(" vmlnk %x", dev->sc_vm_link); vm_object_reference(kernel_object); vm_map_insert(kernel_map, kernel_object, offset, (vm_offset_t)dev->sc_vm_link, (vm_offset_t)dev->sc_vm_link+(MAXPHYS+NBPG)); vm_map_unlock(kernel_map); } dev->sc_vm_link_pages = 0; #endif } #ifdef SFAS_NEED_VM_PATCH /* * Remove our memory-FIX mapping */ void sfas_unlink_vm_link(dev) struct sfas_softc *dev; { if (dev->sc_flags & SFAS_HAS_VM_LINK) { physunaccess(dev->sc_vm_link, dev->sc_vm_link_pages*NBPG); dev->sc_vm_link_pages = 0; dev->sc_flags &= ~SFAS_HAS_VM_LINK; } } /* * Setup a physical-to-virtual mapping to work around the above mentioned * bug in the scsi drivers */ void sfas_link_vm_link(dev, vm_link_data) struct sfas_softc *dev; struct vm_link_data *vm_link_data; { int i; if (dev->sc_flags & SFAS_HAS_VM_LINK) sfas_unlink_vm_link(dev); dev->sc_vm_link_pages = vm_link_data->pages; if (vm_link_data->pages) { for(i=0; ipages; i++) physaccess(dev->sc_vm_link+i*NBPG, vm_link_data->pa[i], NBPG, PG_CI); dev->sc_flags |= SFAS_HAS_VM_LINK; } } #endif /* * used by specific sfas controller */ int sfas_scsicmd(struct scsi_xfer *xs) { struct sfas_softc *dev; struct scsi_link *slp; struct sfas_pending *pendp; int flags, s, target; #ifdef SFAS_NEED_VM_PATCH struct vm_link_data vm_link_data; #endif slp = xs->sc_link; dev = slp->adapter_softc; flags = xs->flags; target = slp->target; if (flags & SCSI_DATA_UIO) panic("sfas: scsi data uio requested"); if ((flags & SCSI_POLL) && (dev->sc_flags & SFAS_ACTIVE)) panic("sfas_scsicmd: busy"); /* Get hold of a sfas_pending block. */ s = splbio(); pendp = dev->sc_xs_free.tqh_first; if (pendp == NULL) { splx(s); return(TRY_AGAIN_LATER); } TAILQ_REMOVE(&dev->sc_xs_free, pendp, link); pendp->xs = xs; splx(s); #ifdef SFAS_NEED_VM_PATCH pendp->vm_link_data.offset = 0; pendp->vm_link_data.pages = 0; /* * We need our FIX vm-link if: * 1) We are NOT using polled IO. * 2) Out data source/destination is not in the u-stack area. */ if (!(flags & SCSI_POLL) && ( #ifdef M68040 (cpu040 && ((vm_offset_t)xs->data >= 0xFFFC0000)) && #endif ((vm_offset_t)xs->data >= 0xFF000000))) { vm_offset_t sva; short n; /* Extract and store the physical adresses of the data block */ sva = (vm_offset_t)xs->data & PG_FRAME; pendp->vm_link_data.offset = (vm_offset_t)xs->data & PGOFSET; pendp->vm_link_data.pages = round_page(xs->data+xs->datalen- sva)/NBPG; for(n=0; nvm_link_data.pages; n++) pendp->vm_link_data.pa[n] = kvtop(sva + n*NBPG); } #endif /* If the chip if busy OR the unit is busy, we have to wait for out turn. */ if ((dev->sc_flags & SFAS_ACTIVE) || (dev->sc_nexus[target].flags & SFAS_NF_UNIT_BUSY)) { s = splbio(); TAILQ_INSERT_TAIL(&dev->sc_xs_pending, pendp, link); splx(s); } else sfas_donextcmd(dev, pendp); return((flags & SCSI_POLL) ? COMPLETE : SUCCESSFULLY_QUEUED); } /* * Actually select the unit, whereby the whole scsi-process is started. */ int sfas_donextcmd(dev, pendp) struct sfas_softc *dev; struct sfas_pending *pendp; { int s; /* * Special case for scsi unit reset. I think this is waterproof. We first * select the unit during splbio. We then cycle through the generated * interrupts until the interrupt routine signals that the unit has * acknowledged the reset. After that we have to wait a reset to select * delay before anything else can happend. */ if (pendp->xs->flags & SCSI_RESET) { struct nexus *nexus; s = splbio(); while(!sfasselect(dev, pendp, 0, 0, 0, 0, SFAS_SELECT_K)) { splx(s); delay(10); s = splbio(); } nexus = dev->sc_cur_nexus; while(nexus->flags & SFAS_NF_UNIT_BUSY) { sfasiwait(dev); sfasintr(dev); } nexus->flags |= SFAS_NF_UNIT_BUSY; splx(s); sfasreset(dev, 0); s = splbio(); nexus->flags &= ~SFAS_NF_UNIT_BUSY; splx(s); } /* * If we are polling, go to splbio and perform the command, else we poke * the scsi-bus via sfasgo to get the interrupt machine going. */ if (pendp->xs->flags & SCSI_POLL) { s = splbio(); sfasicmd(dev, pendp); TAILQ_INSERT_TAIL(&dev->sc_xs_free, pendp, link); splx(s); } else { sfasgo(dev, pendp); return; } } void sfas_scsidone(dev, xs, stat) struct sfas_softc *dev; struct scsi_xfer *xs; int stat; { struct sfas_pending *pendp; int s; xs->status = stat; if (stat == 0) xs->resid = 0; else { switch(stat) { case SCSI_CHECK: /* If we get here we have valid sense data. Faults during * sense is handeled elsewhere and will generate a * XS_DRIVER_STUFFUP. */ xs->error = XS_SENSE; break; case SCSI_BUSY: xs->error = XS_BUSY; break; case -1: xs->error = XS_DRIVER_STUFFUP; QPRINTF(("sfas_scsicmd() bad %x\n", stat)); break; default: xs->error = XS_TIMEOUT; break; } } xs->flags |= ITSDONE; /* Steal the next command from the queue so that one unit can't hog the bus. */ s = splbio(); pendp = dev->sc_xs_pending.tqh_first; while(pendp) { if (!(dev->sc_nexus[pendp->xs->sc_link->target].flags & SFAS_NF_UNIT_BUSY)) break; pendp = pendp->link.tqe_next; } if (pendp != NULL) { TAILQ_REMOVE(&dev->sc_xs_pending, pendp, link); } splx(s); scsi_done(xs); if (pendp) sfas_donextcmd(dev, pendp); } /* * There are two kinds of reset: * 1) CHIP-bus reset. This also implies a SCSI-bus reset. * 2) SCSI-bus reset. * After the appropriate resets have been performed we wait a reset to select * delay time. */ void sfasreset(dev, how) struct sfas_softc *dev; int how; { sfas_regmap_p rp; int i, s; rp = dev->sc_fas; if (how & 1) { for(i=0; i<8; i++) sfas_init_nexus(dev, &dev->sc_nexus[i]); *rp->sfas_command = SFAS_CMD_RESET_CHIP; delay(1); *rp->sfas_command = SFAS_CMD_NOP; *rp->sfas_config1 = dev->sc_config1; *rp->sfas_config2 = dev->sc_config2; *rp->sfas_config3 = dev->sc_config3; *rp->sfas_timeout = dev->sc_timeout_val; *rp->sfas_clkconv = dev->sc_clock_conv_fact & SFAS_CLOCK_CONVERSION_MASK; } if (how & 2) { for(i=0; i<8; i++) sfas_init_nexus(dev, &dev->sc_nexus[i]); s = splbio(); *rp->sfas_command = SFAS_CMD_RESET_SCSI_BUS; delay(100); /* Skip interrupt generated by RESET_SCSI_BUS */ while(*rp->sfas_status & SFAS_STAT_INTERRUPT_PENDING) { dev->sc_status = *rp->sfas_status; dev->sc_interrupt = *rp->sfas_interrupt; delay(100); } dev->sc_status = *rp->sfas_status; dev->sc_interrupt = *rp->sfas_interrupt; splx(s); } if (dev->sc_config_flags & SFAS_SLOW_START) delay(4*250000); /* RESET to SELECT DELAY*4 for slow devices */ else delay(250000); /* RESET to SELECT DELAY */ } /* * Save active data pointers to the nexus block currently active. */ void sfas_save_pointers(dev) struct sfas_softc *dev; { struct nexus *nx; nx = dev->sc_cur_nexus; if (nx) { nx->cur_link = dev->sc_cur_link; nx->max_link = dev->sc_max_link; nx->buf = dev->sc_buf; nx->len = dev->sc_len; nx->dma_len = dev->sc_dma_len; nx->dma_buf = dev->sc_dma_buf; nx->dma_blk_flg = dev->sc_dma_blk_flg; nx->dma_blk_len = dev->sc_dma_blk_len; nx->dma_blk_ptr = dev->sc_dma_blk_ptr; } } /* * Restore data pointers from the currently active nexus block. */ void sfas_restore_pointers(dev) struct sfas_softc *dev; { struct nexus *nx; nx = dev->sc_cur_nexus; if (nx) { dev->sc_cur_link = nx->cur_link; dev->sc_max_link = nx->max_link; dev->sc_buf = nx->buf; dev->sc_len = nx->len; dev->sc_dma_len = nx->dma_len; dev->sc_dma_buf = nx->dma_buf; dev->sc_dma_blk_flg = nx->dma_blk_flg; dev->sc_dma_blk_len = nx->dma_blk_len; dev->sc_dma_blk_ptr = nx->dma_blk_ptr; dev->sc_chain = nx->dma; dev->sc_unit = (nx->lun_unit & 0x0F); dev->sc_lun = (nx->lun_unit & 0xF0) >> 4; } } /* * sfasiwait is used during interrupt and polled IO to wait for an event from * the FAS chip. This function MUST NOT BE CALLED without interrupt disabled. */ void sfasiwait(dev) struct sfas_softc *dev; { sfas_regmap_p rp; /* * If SFAS_DONT_WAIT is set, we have already grabbed the interrupt info * elsewhere. So we don't have to wait for it. */ if (dev->sc_flags & SFAS_DONT_WAIT) { dev->sc_flags &= ~SFAS_DONT_WAIT; return; } rp = dev->sc_fas; /* Wait for FAS chip to signal an interrupt. */ while(!(*rp->sfas_status & SFAS_STAT_INTERRUPT_PENDING)) delay(1); /* Grab interrupt info from chip. */ dev->sc_status = *rp->sfas_status; dev->sc_interrupt = *rp->sfas_interrupt; if (dev->sc_interrupt & SFAS_INT_RESELECTED) { dev->sc_resel[0] = *rp->sfas_fifo; dev->sc_resel[1] = *rp->sfas_fifo; } } /* * Transfer info to/from device. sfas_ixfer uses polled IO+sfasiwait so the * rules that apply to sfasiwait also applies here. */ void sfas_ixfer(dev) struct sfas_softc *dev; { sfas_regmap_p rp; u_char *buf; int len, mode, phase; rp = dev->sc_fas; buf = dev->sc_buf; len = dev->sc_len; /* * Decode the scsi phase to determine whether we are reading or writing. * mode == 1 => READ, mode == 0 => WRITE */ phase = dev->sc_status & SFAS_STAT_PHASE_MASK; mode = (phase == SFAS_PHASE_DATA_IN); while(len && ((dev->sc_status & SFAS_STAT_PHASE_MASK) == phase)) if (mode) { *rp->sfas_command = SFAS_CMD_TRANSFER_INFO; sfasiwait(dev); *buf++ = *rp->sfas_fifo; len--; } else { len--; *rp->sfas_fifo = *buf++; *rp->sfas_command = SFAS_CMD_TRANSFER_INFO; sfasiwait(dev); } /* Update buffer pointers to reflect the sent/recieved data. */ dev->sc_buf = buf; dev->sc_len = len; /* * Since the last sfasiwait will be a phase-change, we can't wait for it * again later, so we have to signal that. */ dev->sc_flags |= SFAS_DONT_WAIT; } /* * Build a Synchronous Data Transfer Request message */ void sfas_build_sdtrm(dev, period, offset) struct sfas_softc *dev; int period; int offset; { dev->sc_msg_out[0] = 0x01; dev->sc_msg_out[1] = 0x03; dev->sc_msg_out[2] = 0x01; dev->sc_msg_out[3] = period/4; dev->sc_msg_out[4] = offset; dev->sc_msg_out_len= 5; } /* * Arbitate the scsi bus and select the unit */ int sfas_select_unit(dev, target) struct sfas_softc *dev; short target; { sfas_regmap_p rp; struct nexus *nexus; int s, retcode, i; u_char cmd; s = splbio(); /* Do this at splbio so that we won't be disturbed. */ retcode = 0; nexus = &dev->sc_nexus[target]; /* * Check if the chip is busy. If not the we mark it as so and hope that nobody * reselects us until we have grabbed the bus. */ if (!(dev->sc_flags & SFAS_ACTIVE) && !dev->sc_sel_nexus) { dev->sc_flags |= SFAS_ACTIVE; rp = dev->sc_fas; *rp->sfas_syncper = nexus->syncper; *rp->sfas_syncoff = nexus->syncoff; *rp->sfas_config3 = nexus->config3; *rp->sfas_config1 = dev->sc_config1; *rp->sfas_timeout = dev->sc_timeout_val; *rp->sfas_dest_id = target; /* If nobody has stolen the bus, we can send a select command to the chip. */ if (!(*rp->sfas_status & SFAS_STAT_INTERRUPT_PENDING)) { *rp->sfas_fifo = nexus->ID; if ((nexus->flags & (SFAS_NF_DO_SDTR | SFAS_NF_RESET)) || (dev->sc_msg_out_len != 0)) cmd = SFAS_CMD_SEL_ATN_STOP; else { for(i=0; iclen; i++) *rp->sfas_fifo = nexus->cbuf[i]; cmd = SFAS_CMD_SEL_ATN; } dev->sc_sel_nexus = nexus; *rp->sfas_command = cmd; retcode = 1; } } splx(s); return(retcode); } /* * Grab the nexus if available else return 0. */ struct nexus * sfas_arbitate_target(dev, target) struct sfas_softc *dev; int target; { struct nexus *nexus; int s; /* * This is realy simple. Raise interrupt level to splbio. Grab the nexus and * leave. */ nexus = &dev->sc_nexus[target]; s = splbio(); if (nexus->flags & SFAS_NF_UNIT_BUSY) nexus = 0; else nexus->flags |= SFAS_NF_UNIT_BUSY; splx(s); return(nexus); } /* * Setup a nexus for use. Initializes command, buffer pointers and dma chain. */ void sfas_setup_nexus(dev, nexus, pendp, cbuf, clen, buf, len, mode) struct sfas_softc *dev; struct nexus *nexus; struct sfas_pending *pendp; unsigned char *cbuf; int clen; unsigned char *buf; int len; int mode; { char sync, target, lun; target = pendp->xs->sc_link->target; lun = pendp->xs->sc_link->lun; /* * Adopt mode to reflect the config flags. * If we can't use DMA we can't use synch transfer. Also check the * sfas_inhibit_xxx[target] flags. */ if ((dev->sc_config_flags & (SFAS_NO_SYNCH | SFAS_NO_DMA)) || sfas_inhibit_sync[target]) mode &= ~SFAS_SELECT_S; if ((dev->sc_config_flags & SFAS_NO_RESELECT) || sfas_inhibit_disc[target]) mode &= ~SFAS_SELECT_R; nexus->xs = pendp->xs; #ifdef SFAS_NEED_VM_PATCH nexus->vm_link_data = pendp->vm_link_data; #endif /* Setup the nexus struct. */ nexus->ID = ((mode & SFAS_SELECT_R) ? 0xC0 : 0x80) | lun; nexus->clen = clen; bcopy(cbuf, nexus->cbuf, nexus->clen); nexus->cbuf[1] |= lun << 5; /* Fix the lun bits */ nexus->cur_link = 0; nexus->dma_len = 0; nexus->dma_buf = 0; nexus->dma_blk_len = 0; nexus->dma_blk_ptr = 0; nexus->len = len; nexus->buf = buf; nexus->lun_unit = (lun << 4) | target; nexus->state = SFAS_NS_SELECTED; /* We must keep these flags. All else must be zero. */ nexus->flags &= SFAS_NF_UNIT_BUSY | SFAS_NF_REQUEST_SENSE | SFAS_NF_SYNC_TESTED | SFAS_NF_SELECT_ME; /* * If we are requesting sense, reflect that in the flags so that we can handle * error in sense data correctly */ if (nexus->flags & SFAS_NF_REQUEST_SENSE) { nexus->flags &= ~SFAS_NF_REQUEST_SENSE; nexus->flags |= SFAS_NF_SENSING; } if (mode & SFAS_SELECT_I) nexus->flags |= SFAS_NF_IMMEDIATE; if (mode & SFAS_SELECT_K) nexus->flags |= SFAS_NF_RESET; sync = ((mode & SFAS_SELECT_S) ? 1 : 0); /* We can't use sync during polled IO. */ if (sync && (mode & SFAS_SELECT_I)) sync = 0; if (!sync && ((nexus->flags & SFAS_NF_SYNC_TESTED) && (nexus->offset != 0))) { /* * If the scsi unit is set to synch transfer and we don't want * that, we have to renegotiate. */ nexus->flags |= SFAS_NF_DO_SDTR; nexus->period = 200; nexus->offset = 0; } else if (sync && !(nexus->flags & SFAS_NF_SYNC_TESTED)) { /* * If the scsi unit is not set to synch transfer and we want * that, we have to negotiate. This should realy base the * period on the clock frequence rather than just check if * >25Mhz */ nexus->flags |= SFAS_NF_DO_SDTR; nexus->period = ((dev->sc_clock_freq>25) ? 100 : 200); nexus->offset = 8; /* If the user has a long cable, we want to limit the period */ if ((nexus->period == 100) && (dev->sc_config_flags & SFAS_SLOW_CABLE)) nexus->period = 200; } /* * Fake a dma-block for polled IO. This way we can use the same code to handle * reselection. Much nicer this way. */ if ((mode & SFAS_SELECT_I) || (dev->sc_config_flags & SFAS_NO_DMA)) { nexus->dma[0].ptr = (vm_offset_t)buf; nexus->dma[0].len = len; nexus->dma[0].flg = SFAS_CHAIN_PRG; nexus->max_link = 1; } else { #ifdef SFAS_NEED_VM_PATCH if (nexus->vm_link_data.pages) sfas_link_vm_link(dev, &nexus->vm_link_data); #endif nexus->max_link = dev->sc_build_dma_chain(dev, nexus->dma, buf, len); } /* Flush the caches. (If needed) */ if (cpu040 && len && !(mode & SFAS_SELECT_I)) dma_cachectl(buf, len); } int sfasselect(dev, pendp, cbuf, clen, buf, len, mode) struct sfas_softc *dev; struct sfas_pending *pendp; unsigned char *cbuf; int clen; unsigned char *buf; int len; int mode; { struct nexus *nexus; /* Get the nexus struct. */ nexus = sfas_arbitate_target(dev, pendp->xs->sc_link->target); if (nexus == NULL) return(0); /* Setup the nexus struct. */ sfas_setup_nexus(dev, nexus, pendp, cbuf, clen, buf, len, mode); /* Post it to the interrupt machine. */ sfas_select_unit(dev, pendp->xs->sc_link->target); return(1); } void sfas_request_sense(dev, nexus) struct sfas_softc *dev; struct nexus *nexus; { struct scsi_xfer *xs; struct sfas_pending pend; struct scsi_sense rqs; int stat, mode; xs = nexus->xs; /* Fake a sfas_pending structure. */ pend.vm_link_data.pages = 0; pend.xs = xs; rqs.opcode = REQUEST_SENSE; rqs.byte2 = xs->sc_link->lun << 5; #ifdef not_yet rqs.length=xs->req_sense_length?xs->req_sense_length:sizeof(xs->sense); #else rqs.length=sizeof(xs->sense); #endif rqs.unused[0] = rqs.unused[1] = rqs.control = 0; /* * If we are requesting sense during polled IO, we have to sense with polled * IO too. */ mode = SFAS_SELECT_RS; if (nexus->flags & SFAS_NF_IMMEDIATE) mode = SFAS_SELECT_I; /* Setup the nexus struct for sensing. */ sfas_setup_nexus(dev, nexus, &pend, (char *)&rqs, sizeof(rqs), (char *)&xs->sense, rqs.length, mode); /* Post it to the interrupt machine. */ sfas_select_unit(dev, xs->sc_link->target); } int sfasgo(dev, pendp) struct sfas_softc *dev; struct sfas_pending *pendp; { int s; char *buf; buf = pendp->xs->data; /* * If we need the vm FIX, make buf reflect that. */ #ifdef SFAS_NEED_VM_PATCH if (pendp->vm_link_data.pages) buf = dev->sc_vm_link + pendp->vm_link_data.offset; #endif if (sfasselect(dev, pendp, (char *)pendp->xs->cmd, pendp->xs->cmdlen, buf, pendp->xs->datalen, SFAS_SELECT_RS)) { /* * We got the command going so the sfas_pending struct is now * free to reuse. */ s = splbio(); TAILQ_INSERT_TAIL(&dev->sc_xs_free, pendp, link); splx(s); } else { /* * We couldn't make the command fly so we have to wait. The * struct MUST be inserted at the head to keep the order of * the commands. */ s = splbio(); TAILQ_INSERT_HEAD(&dev->sc_xs_pending, pendp, link); splx(s); } return(0); } /* * Part one of the interrupt machine. Error checks and reselection test. * We don't know if we have an active nexus here! */ int sfas_pretests(dev, rp) struct sfas_softc *dev; sfas_regmap_p rp; { struct nexus *nexus; int i, s; if (dev->sc_interrupt & SFAS_INT_SCSI_RESET_DETECTED) { /* * Cleanup and notify user. Lets hope that this is all we * have to do */ for(i=0; i<8; i++) { if (dev->sc_nexus[i].xs) sfas_scsidone(dev, dev->sc_nexus[i].xs, -2); sfas_init_nexus(dev, &dev->sc_nexus[i]); } printf("sfasintr: SCSI-RESET detected!"); return(-1); } if (dev->sc_interrupt & SFAS_INT_ILLEGAL_COMMAND) { /* Something went terrible wrong! Dump some data and panic! */ printf("FIFO:"); while(*rp->sfas_fifo_flags & SFAS_FIFO_COUNT_MASK) printf(" %x", *rp->sfas_fifo); printf("\n"); printf("CMD: %x\n", *rp->sfas_command); panic("sfasintr: ILLEGAL COMMAND!"); } if (dev->sc_interrupt & SFAS_INT_RESELECTED) { /* We were reselected. Set the chip as busy */ s = splbio(); dev->sc_flags |= SFAS_ACTIVE; if (dev->sc_sel_nexus) { dev->sc_sel_nexus->flags |= SFAS_NF_SELECT_ME; dev->sc_sel_nexus = 0; } splx(s); if (dev->sc_units_disconnected) { /* Find out who reselected us. */ dev->sc_resel[0] &= ~(1<sc_host_id); for(i=0; i<8; i++) if (dev->sc_resel[0] & (1<sc_nexus[i].state == SFAS_NS_DISCONNECTED) { /* * This unit had disconnected, so we reconnect * it. */ dev->sc_cur_nexus = &dev->sc_nexus[i]; nexus = dev->sc_cur_nexus; *rp->sfas_syncper = nexus->syncper; *rp->sfas_syncoff = nexus->syncoff; *rp->sfas_config3 = nexus->config3; *rp->sfas_dest_id = i & 7; dev->sc_units_disconnected--; dev->sc_msg_in_len= 0; #ifdef SFAS_NEED_VM_PATCH if (nexus->vm_link_data.pages) sfas_link_vm_link(dev, &nexus->vm_link_data); #endif /* Restore active pointers. */ sfas_restore_pointers(dev); nexus->state = SFAS_NS_RESELECTED; *rp->sfas_command = SFAS_CMD_MESSAGE_ACCEPTED; return(1); } } /* Somehow we got an illegal reselection. Dump and panic. */ printf("sfasintr: resel[0] %x resel[1] %x disconnected %d\n", dev->sc_resel[0], dev->sc_resel[1], dev->sc_units_disconnected); panic("sfasintr: Unexpected reselection!"); } return(0); } /* * Part two of the interrupt machine. Handle disconnection and post command * processing. We know that we have an active nexus here. */ int sfas_midaction(dev, rp, nexus) struct sfas_softc *dev; sfas_regmap_p rp; struct nexus *nexus; { int i, left, len, s; u_char status, msg; if (dev->sc_interrupt & SFAS_INT_DISCONNECT) { s = splbio(); dev->sc_cur_nexus = 0; /* Mark chip as busy and clean up the chip FIFO. */ dev->sc_flags &= ~SFAS_ACTIVE; *rp->sfas_command = SFAS_CMD_FLUSH_FIFO; #ifdef SFAS_NEED_VM_PATCH sfas_unlink_vm_link(dev); #endif /* Let the nexus state reflect what we have to do. */ switch(nexus->state) { case SFAS_NS_SELECTED: dev->sc_sel_nexus = 0; nexus->flags &= ~SFAS_NF_SELECT_ME; /* * We were trying to select the unit. Probably no unit * at this ID. */ nexus->xs->resid = dev->sc_len; nexus->status = -2; nexus->flags &= ~SFAS_NF_UNIT_BUSY; nexus->state = SFAS_NS_FINISHED; break; case SFAS_NS_SENSE: /* * Oops! We have to request sense data from this unit. * Do so. */ dev->sc_led(dev, 0); nexus->flags |= SFAS_NF_REQUEST_SENSE; sfas_request_sense(dev, nexus); break; case SFAS_NS_DONE: /* All done. */ nexus->xs->resid = dev->sc_len; nexus->flags &= ~SFAS_NF_UNIT_BUSY; nexus->state = SFAS_NS_FINISHED; dev->sc_led(dev, 0); break; case SFAS_NS_DISCONNECTING: /* * We have recieved a DISCONNECT message, so we are * doing a normal disconnection. */ nexus->state = SFAS_NS_DISCONNECTED; dev->sc_units_disconnected++; break; case SFAS_NS_RESET: /* * We were reseting this SCSI-unit. Clean up the * nexus struct. */ dev->sc_led(dev, 0); sfas_init_nexus(dev, nexus); break; default: /* * Unexpected disconnection! Cleanup and exit. This * shouldn't cause any problems. */ printf("sfasintr: Unexpected disconnection\n"); printf("sfasintr: u %x s %d p %d f %x c %x\n", nexus->lun_unit, nexus->state, dev->sc_status & SFAS_STAT_PHASE_MASK, nexus->flags, nexus->cbuf[0]); nexus->xs->resid = dev->sc_len; nexus->flags &= ~SFAS_NF_UNIT_BUSY; nexus->state = SFAS_NS_FINISHED; nexus->status = -3; dev->sc_led(dev, 0); break; } /* * If we have disconnected units, we MUST enable reselection * within 250ms. */ if (dev->sc_units_disconnected && !(dev->sc_flags & SFAS_ACTIVE)) *rp->sfas_command = SFAS_CMD_ENABLE_RESEL; splx(s); /* Select the first pre-initialized nexus we find. */ for(i=0; i<8; i++) if (dev->sc_nexus[i].flags & SFAS_NF_SELECT_ME) if (sfas_select_unit(dev, i) == 2) break; /* Does any unit need sense data? */ for(i=0; i<8; i++) if (dev->sc_nexus[i].flags & SFAS_NF_REQUEST_SENSE) { sfas_request_sense(dev, &dev->sc_nexus[i]); break; } /* We are done with this nexus! */ if (nexus->state == SFAS_NS_FINISHED) sfas_scsidone(dev, nexus->xs, nexus->status); return(1); } switch(nexus->state) { case SFAS_NS_SELECTED: dev->sc_cur_nexus = nexus; dev->sc_sel_nexus = 0; nexus->flags &= ~SFAS_NF_SELECT_ME; /* * We have selected a unit. Setup chip, restore pointers and * light the led. */ *rp->sfas_syncper = nexus->syncper; *rp->sfas_syncoff = nexus->syncoff; *rp->sfas_config3 = nexus->config3; sfas_restore_pointers(dev); if (!(nexus->flags & SFAS_NF_SENSING)) nexus->status = 0xFF; dev->sc_msg_in[0] = 0xFF; dev->sc_msg_in_len= 0; dev->sc_led(dev, 1); break; case SFAS_NS_DATA_IN: case SFAS_NS_DATA_OUT: /* We have transfered data. */ if (dev->sc_dma_len) if (dev->sc_cur_link < dev->sc_max_link) { /* * Clean up dma and at the same time get how * many bytes that were NOT transfered. */ left = dev->sc_setup_dma(dev, 0, 0, SFAS_DMA_CLEAR); len = dev->sc_dma_len; if (nexus->state == SFAS_NS_DATA_IN) { /* * If we were bumping we may have had an odd length * which means that there may be bytes left in the * fifo. We also need to move the data from the * bump buffer to the actual memory. */ if (dev->sc_dma_buf == dev->sc_bump_pa) { while((*rp->sfas_fifo_flags&SFAS_FIFO_COUNT_MASK) && left) dev->sc_bump_va[len-(left--)] = *rp->sfas_fifo; bcopy(dev->sc_bump_va, dev->sc_buf, len-left); } } else { /* Count any unsent bytes and flush them. */ left+= *rp->sfas_fifo_flags & SFAS_FIFO_COUNT_MASK; *rp->sfas_command = SFAS_CMD_FLUSH_FIFO; } /* * Update pointers/length to reflect the transfered * data. */ dev->sc_len -= len-left; dev->sc_buf += len-left; dev->sc_dma_buf += len-left; dev->sc_dma_len = left; dev->sc_dma_blk_ptr += len-left; dev->sc_dma_blk_len -= len-left; /* * If it was the end of a dma block, we select the * next to begin with. */ if (!dev->sc_dma_blk_len) dev->sc_cur_link++; } break; case SFAS_NS_STATUS: /* * If we were not sensing, grab the status byte. If we were * sensing and we got a bad status, let the user know. */ status = *rp->sfas_fifo; msg = *rp->sfas_fifo; if (!(nexus->flags & SFAS_NF_SENSING)) nexus->status = status; else if (status != 0) nexus->status = -1; /* * Preload the command complete message. Handeled in * sfas_postaction. */ dev->sc_msg_in[0] = msg; dev->sc_msg_in_len = 1; nexus->flags |= SFAS_NF_HAS_MSG; break; default: break; } return(0); } /* * Part three of the interrupt machine. Handle phase changes (and repeated * phase passes). We know that we have an active nexus here. */ int sfas_postaction(dev, rp, nexus) struct sfas_softc *dev; sfas_regmap_p rp; struct nexus *nexus; { int i, left, len; u_char cmd; short offset, period; cmd = 0; switch(dev->sc_status & SFAS_STAT_PHASE_MASK) { case SFAS_PHASE_DATA_OUT: case SFAS_PHASE_DATA_IN: if ((dev->sc_status & SFAS_STAT_PHASE_MASK) == SFAS_PHASE_DATA_OUT) nexus->state = SFAS_NS_DATA_OUT; else nexus->state = SFAS_NS_DATA_IN; /* Make DMA ready to accept new data. Load active pointers * from the DMA block. */ dev->sc_setup_dma(dev, 0, 0, SFAS_DMA_CLEAR); if (dev->sc_cur_link < dev->sc_max_link) { if (!dev->sc_dma_blk_len) { dev->sc_dma_blk_ptr = dev->sc_chain[dev->sc_cur_link].ptr; dev->sc_dma_blk_len = dev->sc_chain[dev->sc_cur_link].len; dev->sc_dma_blk_flg = dev->sc_chain[dev->sc_cur_link].flg; } /* We should use polled IO here. */ if (dev->sc_dma_blk_flg == SFAS_CHAIN_PRG) { sfas_ixfer(dev); dev->sc_cur_link++; dev->sc_dma_len = 0; break; } else if (dev->sc_dma_blk_flg == SFAS_CHAIN_BUMP) len = dev->sc_dma_blk_len; else len = dev->sc_need_bump(dev, dev->sc_dma_blk_ptr, dev->sc_dma_blk_len); /* * If len != 0 we must bump the data, else we just DMA it * straight into memory. */ if (len) { dev->sc_dma_buf = dev->sc_bump_pa; dev->sc_dma_len = len; if (nexus->state == SFAS_NS_DATA_OUT) bcopy(dev->sc_buf, dev->sc_bump_va, dev->sc_dma_len); } else { dev->sc_dma_buf = dev->sc_dma_blk_ptr; dev->sc_dma_len = dev->sc_dma_blk_len; } /* Load DMA with adress and length of transfer. */ dev->sc_setup_dma(dev, dev->sc_dma_buf, dev->sc_dma_len, ((nexus->state == SFAS_NS_DATA_OUT) ? SFAS_DMA_WRITE : SFAS_DMA_READ)); cmd = SFAS_CMD_TRANSFER_INFO | SFAS_CMD_DMA; } else { /* * Hmmm, the unit wants more info than we have or has * more than we want. Let the chip handle that. */ *rp->sfas_tc_low = 256; *rp->sfas_tc_mid = 0; *rp->sfas_tc_high = 0; cmd = SFAS_CMD_TRANSFER_PAD; } break; case SFAS_PHASE_COMMAND: /* The scsi unit wants the command, send it. */ nexus->state = SFAS_NS_SVC; *rp->sfas_command = SFAS_CMD_FLUSH_FIFO; for(i=0; i<5; i++); for(i=0; iclen; i++) *rp->sfas_fifo = nexus->cbuf[i]; cmd = SFAS_CMD_TRANSFER_INFO; break; case SFAS_PHASE_STATUS: /* * We've got status phase. Request status and command * complete message. */ nexus->state = SFAS_NS_STATUS; cmd = SFAS_CMD_COMMAND_COMPLETE; break; case SFAS_PHASE_MESSAGE_OUT: /* * Either the scsi unit wants us to send a message or we have * asked for it by seting the ATN bit. */ nexus->state = SFAS_NS_MSG_OUT; *rp->sfas_command = SFAS_CMD_FLUSH_FIFO; if (nexus->flags & SFAS_NF_DO_SDTR) { /* Send a Synchronous Data Transfer Request. */ sfas_build_sdtrm(dev, nexus->period, nexus->offset); nexus->flags |= SFAS_NF_SDTR_SENT; nexus->flags &= ~SFAS_NF_DO_SDTR; } else if (nexus->flags & SFAS_NF_RESET) { /* Send a reset scsi unit message. */ dev->sc_msg_out[0] = 0x0C; dev->sc_msg_out_len = 1; nexus->state = SFAS_NS_RESET; nexus->flags &= ~SFAS_NF_RESET; } else if (dev->sc_msg_out_len == 0) { /* Don't know what to send so we send a NOP message. */ dev->sc_msg_out[0] = 0x08; dev->sc_msg_out_len = 1; } cmd = SFAS_CMD_TRANSFER_INFO; for(i=0; isc_msg_out_len; i++) *rp->sfas_fifo = dev->sc_msg_out[i]; dev->sc_msg_out_len = 0; break; case SFAS_PHASE_MESSAGE_IN: /* Receive a message from the scsi unit. */ nexus->state = SFAS_NS_MSG_IN; while(!(nexus->flags & SFAS_NF_HAS_MSG)) { *rp->sfas_command = SFAS_CMD_TRANSFER_INFO; sfasiwait(dev); dev->sc_msg_in[dev->sc_msg_in_len++] = *rp->sfas_fifo; /* Check if we got all the bytes in the message. */ if (dev->sc_msg_in[0] >= 0x80) ; else if (dev->sc_msg_in[0] >= 0x30) ; else if (((dev->sc_msg_in[0] >= 0x20) && (dev->sc_msg_in_len == 2)) || ((dev->sc_msg_in[0] != 0x01) && (dev->sc_msg_in_len == 1))) { nexus->flags |= SFAS_NF_HAS_MSG; break; } else { if (dev->sc_msg_in_len >= 2) if ((dev->sc_msg_in[1]+2) == dev->sc_msg_in_len) { nexus->flags |= SFAS_NF_HAS_MSG; break; } } *rp->sfas_command = SFAS_CMD_MESSAGE_ACCEPTED; sfasiwait(dev); if ((dev->sc_status & SFAS_STAT_PHASE_MASK) != SFAS_PHASE_MESSAGE_IN) break; } cmd = SFAS_CMD_MESSAGE_ACCEPTED; if (nexus->flags & SFAS_NF_HAS_MSG) { /* We have a message. Decode it. */ switch(dev->sc_msg_in[0]) { case 0x00: /* COMMAND COMPLETE */ if ((nexus->status == SCSI_CHECK) && !(nexus->flags & SFAS_NF_SENSING)) nexus->state = SFAS_NS_SENSE; else nexus->state = SFAS_NS_DONE; break; case 0x04: /* DISCONNECT */ nexus->state = SFAS_NS_DISCONNECTING; break; case 0x02: /* SAVE DATA POINTER */ sfas_save_pointers(dev); break; case 0x03: /* RESTORE DATA POINTERS */ sfas_restore_pointers(dev); break; case 0x07: /* MESSAGE REJECT */ /* * If we had sent a SDTR and we got a message * reject, the scsi docs say that we must go * to async transfer. */ if (nexus->flags & SFAS_NF_SDTR_SENT) { nexus->flags &= ~SFAS_NF_SDTR_SENT; nexus->config3 &= ~SFAS_CFG3_FASTSCSI; nexus->syncper = 5; nexus->syncoff = 0; *rp->sfas_syncper = nexus->syncper; *rp->sfas_syncoff = nexus->syncoff; *rp->sfas_config3 = nexus->config3; } else /* * Something was rejected but we don't know * what! PANIC! */ panic("sfasintr: Unknown message rejected!"); break; case 0x08: /* MO OPERATION */ break; case 0x01: /* EXTENDED MESSAGE */ switch(dev->sc_msg_in[2]) { case 0x01:/* SYNC. DATA TRANSFER REQUEST */ /* Decode the SDTR message. */ period = 4*dev->sc_msg_in[3]; offset = dev->sc_msg_in[4]; /* * Make sure that the specs are within * chip limits. Note that if we * initiated the negotiation the specs * WILL be withing chip limits. If it * was the scsi unit that initiated * the negotiation, the specs may be * to high. */ if (offset > 16) offset = 16; if ((period < 200) && (dev->sc_clock_freq <= 25)) period = 200; if (offset == 0) period = 5*dev->sc_clock_period; nexus->syncper = period/ dev->sc_clock_period; nexus->syncoff = offset; if (period < 200) nexus->config3 |= SFAS_CFG3_FASTSCSI; else nexus->config3 &=~SFAS_CFG3_FASTSCSI; nexus->flags |= SFAS_NF_SYNC_TESTED; *rp->sfas_syncper = nexus->syncper; *rp->sfas_syncoff = nexus->syncoff; *rp->sfas_config3 = nexus->config3; /* * Hmmm, it seems that the scsi unit * initiated sync negotiation, so lets * reply acording to scsi-2 standard. */ if (!(nexus->flags& SFAS_NF_SDTR_SENT)) { if ((dev->sc_config_flags & SFAS_NO_SYNCH) || (dev->sc_config_flags & SFAS_NO_DMA) || sfas_inhibit_sync[ nexus->lun_unit & 7]) { period = 200; offset = 0; } nexus->offset = offset; nexus->period = period; nexus->flags |= SFAS_NF_DO_SDTR; *rp->sfas_command = SFAS_CMD_SET_ATN; } nexus->flags &= ~SFAS_NF_SDTR_SENT; break; case 0x00: /* MODIFY DATA POINTERS */ case 0x02: /* EXTENDED IDENTIFY (SCSI-1) */ case 0x03: /* WIDE DATA TRANSFER REQUEST */ default: /* Reject any unhandeled messages. */ dev->sc_msg_out[0] = 0x07; dev->sc_msg_out_len = 1; *rp->sfas_command = SFAS_CMD_SET_ATN; cmd = SFAS_CMD_MESSAGE_ACCEPTED; break; } break; default: /* Reject any unhandeled messages. */ dev->sc_msg_out[0] = 0x07; dev->sc_msg_out_len = 1; *rp->sfas_command = SFAS_CMD_SET_ATN; cmd = SFAS_CMD_MESSAGE_ACCEPTED; break; } nexus->flags &= ~SFAS_NF_HAS_MSG; dev->sc_msg_in_len = 0; } break; default: printf("SFASINTR: UNKNOWN PHASE! phase: %d\n", dev->sc_status & SFAS_STAT_PHASE_MASK); dev->sc_led(dev, 0); sfas_scsidone(dev, nexus->xs, -4); return(-1); } if (cmd) *rp->sfas_command = cmd; return(0); } /* * Stub for interrupt machine. */ void sfasintr(dev) struct sfas_softc *dev; { sfas_regmap_p rp; struct nexus *nexus; int s; rp = dev->sc_fas; if (!sfas_pretests(dev, rp)) { nexus = dev->sc_cur_nexus; if (nexus == NULL) nexus = dev->sc_sel_nexus; if (nexus) if (!sfas_midaction(dev, rp, nexus)) sfas_postaction(dev, rp, nexus); } } /* * sfasicmd is used to perform IO when we can't use interrupts. sfasicmd * emulates the normal environment by waiting for the chip and calling * sfasintr. */ void sfasicmd(dev, pendp) struct sfas_softc *dev; struct sfas_pending *pendp; { sfas_regmap_p rp; struct nexus *nexus; nexus = &dev->sc_nexus[pendp->xs->sc_link->target]; rp = dev->sc_fas; if (!sfasselect(dev, pendp, (char *)pendp->xs->cmd, pendp->xs->cmdlen, (char *)pendp->xs->data, pendp->xs->datalen, SFAS_SELECT_I)) panic("sfasicmd: Couldn't select unit"); while(nexus->state != SFAS_NS_FINISHED) { sfasiwait(dev); sfasintr(dev); } nexus->flags &= ~SFAS_NF_SYNC_TESTED; }