NetBSD/sys/dev/pci/mly.c

2460 lines
61 KiB
C

/* $NetBSD: mly.c,v 1.27 2006/05/14 21:45:00 elad Exp $ */
/*-
* Copyright (c) 2001 The NetBSD Foundation, Inc.
* All rights reserved.
*
* This code is derived from software contributed to The NetBSD Foundation
* by Andrew Doran, Thor Lancelot Simon, and Eric Haszlakiewicz.
*
* 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 NetBSD
* Foundation, Inc. and its contributors.
* 4. Neither the name of The NetBSD Foundation 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 NETBSD FOUNDATION, INC. 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 FOUNDATION 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.
*/
/*-
* Copyright (c) 2000, 2001 Michael Smith
* Copyright (c) 2000 BSDi
* 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.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* from FreeBSD: mly.c,v 1.8 2001/07/14 00:12:22 msmith Exp
*/
/*
* Driver for the Mylex AcceleRAID and eXtremeRAID family with v6 firmware.
*
* TODO:
*
* o Make mly->mly_btl a hash, then MLY_BTL_RESCAN becomes a SIMPLEQ.
* o Handle FC and multiple LUNs.
* o Fix mmbox usage.
* o Fix transfer speed fudge.
*/
#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: mly.c,v 1.27 2006/05/14 21:45:00 elad Exp $");
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/device.h>
#include <sys/kernel.h>
#include <sys/queue.h>
#include <sys/buf.h>
#include <sys/endian.h>
#include <sys/conf.h>
#include <sys/malloc.h>
#include <sys/ioctl.h>
#include <sys/scsiio.h>
#include <sys/kthread.h>
#include <uvm/uvm_extern.h>
#include <machine/bus.h>
#include <dev/scsipi/scsi_all.h>
#include <dev/scsipi/scsipi_all.h>
#include <dev/scsipi/scsiconf.h>
#include <dev/pci/pcireg.h>
#include <dev/pci/pcivar.h>
#include <dev/pci/pcidevs.h>
#include <dev/pci/mlyreg.h>
#include <dev/pci/mlyio.h>
#include <dev/pci/mlyvar.h>
#include <dev/pci/mly_tables.h>
static void mly_attach(struct device *, struct device *, void *);
static int mly_match(struct device *, struct cfdata *, void *);
static const struct mly_ident *mly_find_ident(struct pci_attach_args *);
static int mly_fwhandshake(struct mly_softc *);
static int mly_flush(struct mly_softc *);
static int mly_intr(void *);
static void mly_shutdown(void *);
static int mly_alloc_ccbs(struct mly_softc *);
static void mly_check_event(struct mly_softc *);
static void mly_complete_event(struct mly_softc *, struct mly_ccb *);
static void mly_complete_rescan(struct mly_softc *, struct mly_ccb *);
static int mly_dmamem_alloc(struct mly_softc *, int, bus_dmamap_t *,
caddr_t *, bus_addr_t *, bus_dma_segment_t *);
static void mly_dmamem_free(struct mly_softc *, int, bus_dmamap_t,
caddr_t, bus_dma_segment_t *);
static int mly_enable_mmbox(struct mly_softc *);
static void mly_fetch_event(struct mly_softc *);
static int mly_get_controllerinfo(struct mly_softc *);
static int mly_get_eventstatus(struct mly_softc *);
static int mly_ioctl(struct mly_softc *, struct mly_cmd_ioctl *,
void **, size_t, void *, size_t *);
static void mly_padstr(char *, const char *, int);
static void mly_process_event(struct mly_softc *, struct mly_event *);
static void mly_release_ccbs(struct mly_softc *);
static int mly_scan_btl(struct mly_softc *, int, int);
static void mly_scan_channel(struct mly_softc *, int);
static void mly_thread(void *);
static void mly_thread_create(void *);
static int mly_ccb_alloc(struct mly_softc *, struct mly_ccb **);
static void mly_ccb_complete(struct mly_softc *, struct mly_ccb *);
static void mly_ccb_enqueue(struct mly_softc *, struct mly_ccb *);
static void mly_ccb_free(struct mly_softc *, struct mly_ccb *);
static int mly_ccb_map(struct mly_softc *, struct mly_ccb *);
static int mly_ccb_poll(struct mly_softc *, struct mly_ccb *, int);
static int mly_ccb_submit(struct mly_softc *, struct mly_ccb *);
static void mly_ccb_unmap(struct mly_softc *, struct mly_ccb *);
static int mly_ccb_wait(struct mly_softc *, struct mly_ccb *, int);
static void mly_get_xfer_mode(struct mly_softc *, int,
struct scsipi_xfer_mode *);
static void mly_scsipi_complete(struct mly_softc *, struct mly_ccb *);
static int mly_scsipi_ioctl(struct scsipi_channel *, u_long, caddr_t,
int, struct proc *);
static void mly_scsipi_minphys(struct buf *);
static void mly_scsipi_request(struct scsipi_channel *,
scsipi_adapter_req_t, void *);
static int mly_user_command(struct mly_softc *, struct mly_user_command *);
static int mly_user_health(struct mly_softc *, struct mly_user_health *);
extern struct cfdriver mly_cd;
CFATTACH_DECL(mly, sizeof(struct mly_softc),
mly_match, mly_attach, NULL, NULL);
dev_type_open(mlyopen);
dev_type_close(mlyclose);
dev_type_ioctl(mlyioctl);
const struct cdevsw mly_cdevsw = {
mlyopen, mlyclose, noread, nowrite, mlyioctl,
nostop, notty, nopoll, nommap, nokqfilter,
};
struct mly_ident {
u_short vendor;
u_short product;
u_short subvendor;
u_short subproduct;
int hwif;
const char *desc;
} static const mly_ident[] = {
{
PCI_VENDOR_MYLEX,
PCI_PRODUCT_MYLEX_EXTREMERAID,
PCI_VENDOR_MYLEX,
0x0040,
MLY_HWIF_STRONGARM,
"eXtremeRAID 2000"
},
{
PCI_VENDOR_MYLEX,
PCI_PRODUCT_MYLEX_EXTREMERAID,
PCI_VENDOR_MYLEX,
0x0030,
MLY_HWIF_STRONGARM,
"eXtremeRAID 3000"
},
{
PCI_VENDOR_MYLEX,
PCI_PRODUCT_MYLEX_ACCELERAID,
PCI_VENDOR_MYLEX,
0x0050,
MLY_HWIF_I960RX,
"AcceleRAID 352"
},
{
PCI_VENDOR_MYLEX,
PCI_PRODUCT_MYLEX_ACCELERAID,
PCI_VENDOR_MYLEX,
0x0052,
MLY_HWIF_I960RX,
"AcceleRAID 170"
},
{
PCI_VENDOR_MYLEX,
PCI_PRODUCT_MYLEX_ACCELERAID,
PCI_VENDOR_MYLEX,
0x0054,
MLY_HWIF_I960RX,
"AcceleRAID 160"
},
};
static void *mly_sdh;
/*
* Try to find a `mly_ident' entry corresponding to this board.
*/
static const struct mly_ident *
mly_find_ident(struct pci_attach_args *pa)
{
const struct mly_ident *mpi, *maxmpi;
pcireg_t reg;
mpi = mly_ident;
maxmpi = mpi + sizeof(mly_ident) / sizeof(mly_ident[0]);
if (PCI_CLASS(pa->pa_class) == PCI_CLASS_I2O)
return (NULL);
for (; mpi < maxmpi; mpi++) {
if (PCI_VENDOR(pa->pa_id) != mpi->vendor ||
PCI_PRODUCT(pa->pa_id) != mpi->product)
continue;
if (mpi->subvendor == 0x0000)
return (mpi);
reg = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_SUBSYS_ID_REG);
if (PCI_VENDOR(reg) == mpi->subvendor &&
PCI_PRODUCT(reg) == mpi->subproduct)
return (mpi);
}
return (NULL);
}
/*
* Match a supported board.
*/
static int
mly_match(struct device *parent, struct cfdata *cfdata, void *aux)
{
return (mly_find_ident(aux) != NULL);
}
/*
* Attach a supported board.
*/
static void
mly_attach(struct device *parent, struct device *self, void *aux)
{
struct pci_attach_args *pa;
struct mly_softc *mly;
struct mly_ioctl_getcontrollerinfo *mi;
const struct mly_ident *ident;
pci_chipset_tag_t pc;
pci_intr_handle_t ih;
bus_space_handle_t memh, ioh;
bus_space_tag_t memt, iot;
pcireg_t reg;
const char *intrstr;
int ior, memr, i, rv, state;
struct scsipi_adapter *adapt;
struct scsipi_channel *chan;
mly = (struct mly_softc *)self;
pa = aux;
pc = pa->pa_pc;
ident = mly_find_ident(pa);
state = 0;
mly->mly_dmat = pa->pa_dmat;
mly->mly_hwif = ident->hwif;
printf(": Mylex %s\n", ident->desc);
/*
* Map the PCI register window.
*/
memr = -1;
ior = -1;
for (i = 0x10; i <= 0x14; i += 4) {
reg = pci_conf_read(pa->pa_pc, pa->pa_tag, i);
if (PCI_MAPREG_TYPE(reg) == PCI_MAPREG_TYPE_IO) {
if (ior == -1 && PCI_MAPREG_IO_SIZE(reg) != 0)
ior = i;
} else {
if (memr == -1 && PCI_MAPREG_MEM_SIZE(reg) != 0)
memr = i;
}
}
if (memr != -1)
if (pci_mapreg_map(pa, memr, PCI_MAPREG_TYPE_MEM, 0,
&memt, &memh, NULL, NULL))
memr = -1;
if (ior != -1)
if (pci_mapreg_map(pa, ior, PCI_MAPREG_TYPE_IO, 0,
&iot, &ioh, NULL, NULL))
ior = -1;
if (memr != -1) {
mly->mly_iot = memt;
mly->mly_ioh = memh;
} else if (ior != -1) {
mly->mly_iot = iot;
mly->mly_ioh = ioh;
} else {
printf("%s: can't map i/o or memory space\n", self->dv_xname);
return;
}
/*
* Enable the device.
*/
reg = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
pci_conf_write(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG,
reg | PCI_COMMAND_MASTER_ENABLE);
/*
* Map and establish the interrupt.
*/
if (pci_intr_map(pa, &ih)) {
printf("%s: can't map interrupt\n", self->dv_xname);
return;
}
intrstr = pci_intr_string(pc, ih);
mly->mly_ih = pci_intr_establish(pc, ih, IPL_BIO, mly_intr, mly);
if (mly->mly_ih == NULL) {
printf("%s: can't establish interrupt", self->dv_xname);
if (intrstr != NULL)
printf(" at %s", intrstr);
printf("\n");
return;
}
if (intrstr != NULL)
printf("%s: interrupting at %s\n", mly->mly_dv.dv_xname,
intrstr);
/*
* Take care of interface-specific tasks.
*/
switch (mly->mly_hwif) {
case MLY_HWIF_I960RX:
mly->mly_doorbell_true = 0x00;
mly->mly_cmd_mailbox = MLY_I960RX_COMMAND_MAILBOX;
mly->mly_status_mailbox = MLY_I960RX_STATUS_MAILBOX;
mly->mly_idbr = MLY_I960RX_IDBR;
mly->mly_odbr = MLY_I960RX_ODBR;
mly->mly_error_status = MLY_I960RX_ERROR_STATUS;
mly->mly_interrupt_status = MLY_I960RX_INTERRUPT_STATUS;
mly->mly_interrupt_mask = MLY_I960RX_INTERRUPT_MASK;
break;
case MLY_HWIF_STRONGARM:
mly->mly_doorbell_true = 0xff;
mly->mly_cmd_mailbox = MLY_STRONGARM_COMMAND_MAILBOX;
mly->mly_status_mailbox = MLY_STRONGARM_STATUS_MAILBOX;
mly->mly_idbr = MLY_STRONGARM_IDBR;
mly->mly_odbr = MLY_STRONGARM_ODBR;
mly->mly_error_status = MLY_STRONGARM_ERROR_STATUS;
mly->mly_interrupt_status = MLY_STRONGARM_INTERRUPT_STATUS;
mly->mly_interrupt_mask = MLY_STRONGARM_INTERRUPT_MASK;
break;
}
/*
* Allocate and map the scatter/gather lists.
*/
rv = mly_dmamem_alloc(mly, MLY_SGL_SIZE * MLY_MAX_CCBS,
&mly->mly_sg_dmamap, (caddr_t *)&mly->mly_sg,
&mly->mly_sg_busaddr, &mly->mly_sg_seg);
if (rv) {
printf("%s: unable to allocate S/G maps\n",
mly->mly_dv.dv_xname);
goto bad;
}
state++;
/*
* Allocate and map the memory mailbox.
*/
rv = mly_dmamem_alloc(mly, sizeof(struct mly_mmbox),
&mly->mly_mmbox_dmamap, (caddr_t *)&mly->mly_mmbox,
&mly->mly_mmbox_busaddr, &mly->mly_mmbox_seg);
if (rv) {
printf("%s: unable to allocate mailboxes\n",
mly->mly_dv.dv_xname);
goto bad;
}
state++;
/*
* Initialise per-controller queues.
*/
SLIST_INIT(&mly->mly_ccb_free);
SIMPLEQ_INIT(&mly->mly_ccb_queue);
/*
* Disable interrupts before we start talking to the controller.
*/
mly_outb(mly, mly->mly_interrupt_mask, MLY_INTERRUPT_MASK_DISABLE);
/*
* Wait for the controller to come ready, handshaking with the
* firmware if required. This is typically only necessary on
* platforms where the controller BIOS does not run.
*/
if (mly_fwhandshake(mly)) {
printf("%s: unable to bring controller online\n",
mly->mly_dv.dv_xname);
goto bad;
}
/*
* Allocate initial command buffers, obtain controller feature
* information, and then reallocate command buffers, since we'll
* know how many we want.
*/
if (mly_alloc_ccbs(mly)) {
printf("%s: unable to allocate CCBs\n",
mly->mly_dv.dv_xname);
goto bad;
}
state++;
if (mly_get_controllerinfo(mly)) {
printf("%s: unable to retrieve controller info\n",
mly->mly_dv.dv_xname);
goto bad;
}
mly_release_ccbs(mly);
if (mly_alloc_ccbs(mly)) {
printf("%s: unable to allocate CCBs\n",
mly->mly_dv.dv_xname);
state--;
goto bad;
}
/*
* Get the current event counter for health purposes, populate the
* initial health status buffer.
*/
if (mly_get_eventstatus(mly)) {
printf("%s: unable to retrieve event status\n",
mly->mly_dv.dv_xname);
goto bad;
}
/*
* Enable memory-mailbox mode.
*/
if (mly_enable_mmbox(mly)) {
printf("%s: unable to enable memory mailbox\n",
mly->mly_dv.dv_xname);
goto bad;
}
/*
* Print a little information about the controller.
*/
mi = mly->mly_controllerinfo;
printf("%s: %d physical channel%s, firmware %d.%02d-%d-%02d "
"(%02d%02d%02d%02d), %dMB RAM\n", mly->mly_dv.dv_xname,
mi->physical_channels_present,
(mi->physical_channels_present) > 1 ? "s" : "",
mi->fw_major, mi->fw_minor, mi->fw_turn, mi->fw_build,
mi->fw_century, mi->fw_year, mi->fw_month, mi->fw_day,
le16toh(mi->memory_size));
/*
* Register our `shutdownhook'.
*/
if (mly_sdh == NULL)
shutdownhook_establish(mly_shutdown, NULL);
/*
* Clear any previous BTL information. For each bus that scsipi
* wants to scan, we'll receive the SCBUSIOLLSCAN ioctl and retrieve
* all BTL info at that point.
*/
memset(&mly->mly_btl, 0, sizeof(mly->mly_btl));
mly->mly_nchans = mly->mly_controllerinfo->physical_channels_present +
mly->mly_controllerinfo->virtual_channels_present;
/*
* Attach to scsipi.
*/
adapt = &mly->mly_adapt;
memset(adapt, 0, sizeof(*adapt));
adapt->adapt_dev = &mly->mly_dv;
adapt->adapt_nchannels = mly->mly_nchans;
adapt->adapt_openings = mly->mly_ncmds - MLY_CCBS_RESV;
adapt->adapt_max_periph = mly->mly_ncmds - MLY_CCBS_RESV;
adapt->adapt_request = mly_scsipi_request;
adapt->adapt_minphys = mly_scsipi_minphys;
adapt->adapt_ioctl = mly_scsipi_ioctl;
for (i = 0; i < mly->mly_nchans; i++) {
chan = &mly->mly_chans[i];
memset(chan, 0, sizeof(*chan));
chan->chan_adapter = adapt;
chan->chan_bustype = &scsi_bustype;
chan->chan_channel = i;
chan->chan_ntargets = MLY_MAX_TARGETS;
chan->chan_nluns = MLY_MAX_LUNS;
chan->chan_id = mly->mly_controllerparam->initiator_id;
chan->chan_flags = SCSIPI_CHAN_NOSETTLE;
config_found(&mly->mly_dv, chan, scsiprint);
}
/*
* Now enable interrupts...
*/
mly_outb(mly, mly->mly_interrupt_mask, MLY_INTERRUPT_MASK_ENABLE);
/*
* Finally, create our monitoring thread.
*/
kthread_create(mly_thread_create, mly);
mly->mly_state |= MLY_STATE_INITOK;
return;
bad:
if (state > 2)
mly_release_ccbs(mly);
if (state > 1)
mly_dmamem_free(mly, sizeof(struct mly_mmbox),
mly->mly_mmbox_dmamap, (caddr_t)mly->mly_mmbox,
&mly->mly_mmbox_seg);
if (state > 0)
mly_dmamem_free(mly, MLY_SGL_SIZE * MLY_MAX_CCBS,
mly->mly_sg_dmamap, (caddr_t)mly->mly_sg,
&mly->mly_sg_seg);
}
/*
* Scan all possible devices on the specified channel.
*/
static void
mly_scan_channel(struct mly_softc *mly, int bus)
{
int s, target;
for (target = 0; target < MLY_MAX_TARGETS; target++) {
s = splbio();
if (!mly_scan_btl(mly, bus, target)) {
tsleep(&mly->mly_btl[bus][target], PRIBIO, "mlyscan",
0);
}
splx(s);
}
}
/*
* Shut down all configured `mly' devices.
*/
static void
mly_shutdown(void *cookie)
{
struct mly_softc *mly;
int i;
for (i = 0; i < mly_cd.cd_ndevs; i++) {
if ((mly = device_lookup(&mly_cd, i)) == NULL)
continue;
if (mly_flush(mly))
printf("%s: unable to flush cache\n",
mly->mly_dv.dv_xname);
}
}
/*
* Fill in the mly_controllerinfo and mly_controllerparam fields in the
* softc.
*/
static int
mly_get_controllerinfo(struct mly_softc *mly)
{
struct mly_cmd_ioctl mci;
int rv;
/*
* Build the getcontrollerinfo ioctl and send it.
*/
memset(&mci, 0, sizeof(mci));
mci.sub_ioctl = MDACIOCTL_GETCONTROLLERINFO;
rv = mly_ioctl(mly, &mci, (void **)&mly->mly_controllerinfo,
sizeof(*mly->mly_controllerinfo), NULL, NULL);
if (rv != 0)
return (rv);
/*
* Build the getcontrollerparameter ioctl and send it.
*/
memset(&mci, 0, sizeof(mci));
mci.sub_ioctl = MDACIOCTL_GETCONTROLLERPARAMETER;
rv = mly_ioctl(mly, &mci, (void **)&mly->mly_controllerparam,
sizeof(*mly->mly_controllerparam), NULL, NULL);
return (rv);
}
/*
* Rescan a device, possibly as a consequence of getting an event which
* suggests that it may have changed. Must be called with interrupts
* blocked.
*/
static int
mly_scan_btl(struct mly_softc *mly, int bus, int target)
{
struct mly_ccb *mc;
struct mly_cmd_ioctl *mci;
int rv;
if (target == mly->mly_controllerparam->initiator_id) {
mly->mly_btl[bus][target].mb_flags = MLY_BTL_PROTECTED;
return (EIO);
}
/* Don't re-scan if a scan is already in progress. */
if ((mly->mly_btl[bus][target].mb_flags & MLY_BTL_SCANNING) != 0)
return (EBUSY);
/* Get a command. */
if ((rv = mly_ccb_alloc(mly, &mc)) != 0)
return (rv);
/* Set up the data buffer. */
mc->mc_data = malloc(sizeof(union mly_devinfo),
M_DEVBUF, M_NOWAIT|M_ZERO);
mc->mc_flags |= MLY_CCB_DATAIN;
mc->mc_complete = mly_complete_rescan;
/*
* Build the ioctl.
*/
mci = (struct mly_cmd_ioctl *)&mc->mc_packet->ioctl;
mci->opcode = MDACMD_IOCTL;
mci->timeout = 30 | MLY_TIMEOUT_SECONDS;
memset(&mci->param, 0, sizeof(mci->param));
if (MLY_BUS_IS_VIRTUAL(mly, bus)) {
mc->mc_length = sizeof(struct mly_ioctl_getlogdevinfovalid);
mci->data_size = htole32(mc->mc_length);
mci->sub_ioctl = MDACIOCTL_GETLOGDEVINFOVALID;
_lto3l(MLY_LOGADDR(0, MLY_LOGDEV_ID(mly, bus, target)),
mci->addr);
} else {
mc->mc_length = sizeof(struct mly_ioctl_getphysdevinfovalid);
mci->data_size = htole32(mc->mc_length);
mci->sub_ioctl = MDACIOCTL_GETPHYSDEVINFOVALID;
_lto3l(MLY_PHYADDR(0, bus, target, 0), mci->addr);
}
/*
* Dispatch the command.
*/
if ((rv = mly_ccb_map(mly, mc)) != 0) {
free(mc->mc_data, M_DEVBUF);
mly_ccb_free(mly, mc);
return(rv);
}
mly->mly_btl[bus][target].mb_flags |= MLY_BTL_SCANNING;
mly_ccb_enqueue(mly, mc);
return (0);
}
/*
* Handle the completion of a rescan operation.
*/
static void
mly_complete_rescan(struct mly_softc *mly, struct mly_ccb *mc)
{
struct mly_ioctl_getlogdevinfovalid *ldi;
struct mly_ioctl_getphysdevinfovalid *pdi;
struct mly_cmd_ioctl *mci;
struct mly_btl btl, *btlp;
struct scsipi_xfer_mode xm;
int bus, target, rescan;
u_int tmp;
mly_ccb_unmap(mly, mc);
/*
* Recover the bus and target from the command. We need these even
* in the case where we don't have a useful response.
*/
mci = (struct mly_cmd_ioctl *)&mc->mc_packet->ioctl;
tmp = _3ltol(mci->addr);
rescan = 0;
if (mci->sub_ioctl == MDACIOCTL_GETLOGDEVINFOVALID) {
bus = MLY_LOGDEV_BUS(mly, MLY_LOGADDR_DEV(tmp));
target = MLY_LOGDEV_TARGET(mly, MLY_LOGADDR_DEV(tmp));
} else {
bus = MLY_PHYADDR_CHANNEL(tmp);
target = MLY_PHYADDR_TARGET(tmp);
}
btlp = &mly->mly_btl[bus][target];
/* The default result is 'no device'. */
memset(&btl, 0, sizeof(btl));
btl.mb_flags = MLY_BTL_PROTECTED;
/* If the rescan completed OK, we have possibly-new BTL data. */
if (mc->mc_status != 0)
goto out;
if (mc->mc_length == sizeof(*ldi)) {
ldi = (struct mly_ioctl_getlogdevinfovalid *)mc->mc_data;
tmp = le32toh(ldi->logical_device_number);
if (MLY_LOGDEV_BUS(mly, tmp) != bus ||
MLY_LOGDEV_TARGET(mly, tmp) != target) {
#ifdef MLYDEBUG
printf("%s: WARNING: BTL rescan (logical) for %d:%d "
"returned data for %d:%d instead\n",
mly->mly_dv.dv_xname, bus, target,
MLY_LOGDEV_BUS(mly, tmp),
MLY_LOGDEV_TARGET(mly, tmp));
#endif
goto out;
}
btl.mb_flags = MLY_BTL_LOGICAL | MLY_BTL_TQING;
btl.mb_type = ldi->raid_level;
btl.mb_state = ldi->state;
} else if (mc->mc_length == sizeof(*pdi)) {
pdi = (struct mly_ioctl_getphysdevinfovalid *)mc->mc_data;
if (pdi->channel != bus || pdi->target != target) {
#ifdef MLYDEBUG
printf("%s: WARNING: BTL rescan (physical) for %d:%d "
" returned data for %d:%d instead\n",
mly->mly_dv.dv_xname,
bus, target, pdi->channel, pdi->target);
#endif
goto out;
}
btl.mb_flags = MLY_BTL_PHYSICAL;
btl.mb_type = MLY_DEVICE_TYPE_PHYSICAL;
btl.mb_state = pdi->state;
btl.mb_speed = pdi->speed;
btl.mb_width = pdi->width;
if (pdi->state != MLY_DEVICE_STATE_UNCONFIGURED)
btl.mb_flags |= MLY_BTL_PROTECTED;
if (pdi->command_tags != 0)
btl.mb_flags |= MLY_BTL_TQING;
} else {
printf("%s: BTL rescan result invalid\n", mly->mly_dv.dv_xname);
goto out;
}
/* Decide whether we need to rescan the device. */
if (btl.mb_flags != btlp->mb_flags ||
btl.mb_speed != btlp->mb_speed ||
btl.mb_width != btlp->mb_width)
rescan = 1;
out:
*btlp = btl;
if (rescan && (btl.mb_flags & MLY_BTL_PROTECTED) == 0) {
xm.xm_target = target;
mly_get_xfer_mode(mly, bus, &xm);
/* XXX SCSI mid-layer rescan goes here. */
}
/* Wake anybody waiting on the device to be rescanned. */
wakeup(btlp);
free(mc->mc_data, M_DEVBUF);
mly_ccb_free(mly, mc);
}
/*
* Get the current health status and set the 'next event' counter to suit.
*/
static int
mly_get_eventstatus(struct mly_softc *mly)
{
struct mly_cmd_ioctl mci;
struct mly_health_status *mh;
int rv;
/* Build the gethealthstatus ioctl and send it. */
memset(&mci, 0, sizeof(mci));
mh = NULL;
mci.sub_ioctl = MDACIOCTL_GETHEALTHSTATUS;
rv = mly_ioctl(mly, &mci, (void *)&mh, sizeof(*mh), NULL, NULL);
if (rv)
return (rv);
/* Get the event counter. */
mly->mly_event_change = le32toh(mh->change_counter);
mly->mly_event_waiting = le32toh(mh->next_event);
mly->mly_event_counter = le32toh(mh->next_event);
/* Save the health status into the memory mailbox */
memcpy(&mly->mly_mmbox->mmm_health.status, mh, sizeof(*mh));
bus_dmamap_sync(mly->mly_dmat, mly->mly_mmbox_dmamap,
offsetof(struct mly_mmbox, mmm_health),
sizeof(mly->mly_mmbox->mmm_health),
BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
free(mh, M_DEVBUF);
return (0);
}
/*
* Enable memory mailbox mode.
*/
static int
mly_enable_mmbox(struct mly_softc *mly)
{
struct mly_cmd_ioctl mci;
u_int8_t *sp;
u_int64_t tmp;
int rv;
/* Build the ioctl and send it. */
memset(&mci, 0, sizeof(mci));
mci.sub_ioctl = MDACIOCTL_SETMEMORYMAILBOX;
/* Set buffer addresses. */
tmp = mly->mly_mmbox_busaddr + offsetof(struct mly_mmbox, mmm_command);
mci.param.setmemorymailbox.command_mailbox_physaddr = htole64(tmp);
tmp = mly->mly_mmbox_busaddr + offsetof(struct mly_mmbox, mmm_status);
mci.param.setmemorymailbox.status_mailbox_physaddr = htole64(tmp);
tmp = mly->mly_mmbox_busaddr + offsetof(struct mly_mmbox, mmm_health);
mci.param.setmemorymailbox.health_buffer_physaddr = htole64(tmp);
/* Set buffer sizes - abuse of data_size field is revolting. */
sp = (u_int8_t *)&mci.data_size;
sp[0] = (sizeof(union mly_cmd_packet) * MLY_MMBOX_COMMANDS) >> 10;
sp[1] = (sizeof(union mly_status_packet) * MLY_MMBOX_STATUS) >> 10;
mci.param.setmemorymailbox.health_buffer_size =
sizeof(union mly_health_region) >> 10;
rv = mly_ioctl(mly, &mci, NULL, 0, NULL, NULL);
if (rv)
return (rv);
mly->mly_state |= MLY_STATE_MMBOX_ACTIVE;
return (0);
}
/*
* Flush all pending I/O from the controller.
*/
static int
mly_flush(struct mly_softc *mly)
{
struct mly_cmd_ioctl mci;
/* Build the ioctl */
memset(&mci, 0, sizeof(mci));
mci.sub_ioctl = MDACIOCTL_FLUSHDEVICEDATA;
mci.param.deviceoperation.operation_device =
MLY_OPDEVICE_PHYSICAL_CONTROLLER;
/* Pass it off to the controller */
return (mly_ioctl(mly, &mci, NULL, 0, NULL, NULL));
}
/*
* Perform an ioctl command.
*
* If (data) is not NULL, the command requires data transfer to the
* controller. If (*data) is NULL the command requires data transfer from
* the controller, and we will allocate a buffer for it.
*/
static int
mly_ioctl(struct mly_softc *mly, struct mly_cmd_ioctl *ioctl, void **data,
size_t datasize, void *sense_buffer,
size_t *sense_length)
{
struct mly_ccb *mc;
struct mly_cmd_ioctl *mci;
u_int8_t status;
int rv;
mc = NULL;
if ((rv = mly_ccb_alloc(mly, &mc)) != 0)
goto bad;
/*
* Copy the ioctl structure, but save some important fields and then
* fixup.
*/
mci = &mc->mc_packet->ioctl;
ioctl->sense_buffer_address = htole64(mci->sense_buffer_address);
ioctl->maximum_sense_size = mci->maximum_sense_size;
*mci = *ioctl;
mci->opcode = MDACMD_IOCTL;
mci->timeout = 30 | MLY_TIMEOUT_SECONDS;
/* Handle the data buffer. */
if (data != NULL) {
if (*data == NULL) {
/* Allocate data buffer */
mc->mc_data = malloc(datasize, M_DEVBUF, M_NOWAIT);
mc->mc_flags |= MLY_CCB_DATAIN;
} else {
mc->mc_data = *data;
mc->mc_flags |= MLY_CCB_DATAOUT;
}
mc->mc_length = datasize;
mc->mc_packet->generic.data_size = htole32(datasize);
}
/* Run the command. */
if (datasize > 0)
if ((rv = mly_ccb_map(mly, mc)) != 0)
goto bad;
rv = mly_ccb_poll(mly, mc, 30000);
if (datasize > 0)
mly_ccb_unmap(mly, mc);
if (rv != 0)
goto bad;
/* Clean up and return any data. */
status = mc->mc_status;
if (status != 0)
printf("mly_ioctl: command status %d\n", status);
if (mc->mc_sense > 0 && sense_buffer != NULL) {
memcpy(sense_buffer, mc->mc_packet, mc->mc_sense);
*sense_length = mc->mc_sense;
goto bad;
}
/* Should we return a data pointer? */
if (data != NULL && *data == NULL)
*data = mc->mc_data;
/* Command completed OK. */
rv = (status != 0 ? EIO : 0);
bad:
if (mc != NULL) {
/* Do we need to free a data buffer we allocated? */
if (rv != 0 && mc->mc_data != NULL &&
(data == NULL || *data == NULL))
free(mc->mc_data, M_DEVBUF);
mly_ccb_free(mly, mc);
}
return (rv);
}
/*
* Check for event(s) outstanding in the controller.
*/
static void
mly_check_event(struct mly_softc *mly)
{
bus_dmamap_sync(mly->mly_dmat, mly->mly_mmbox_dmamap,
offsetof(struct mly_mmbox, mmm_health),
sizeof(mly->mly_mmbox->mmm_health),
BUS_DMASYNC_POSTWRITE | BUS_DMASYNC_POSTREAD);
/*
* The controller may have updated the health status information, so
* check for it here. Note that the counters are all in host
* memory, so this check is very cheap. Also note that we depend on
* checking on completion
*/
if (le32toh(mly->mly_mmbox->mmm_health.status.change_counter) !=
mly->mly_event_change) {
mly->mly_event_change =
le32toh(mly->mly_mmbox->mmm_health.status.change_counter);
mly->mly_event_waiting =
le32toh(mly->mly_mmbox->mmm_health.status.next_event);
/* Wake up anyone that might be interested in this. */
wakeup(&mly->mly_event_change);
}
bus_dmamap_sync(mly->mly_dmat, mly->mly_mmbox_dmamap,
offsetof(struct mly_mmbox, mmm_health),
sizeof(mly->mly_mmbox->mmm_health),
BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
if (mly->mly_event_counter != mly->mly_event_waiting)
mly_fetch_event(mly);
}
/*
* Fetch one event from the controller. If we fail due to resource
* starvation, we'll be retried the next time a command completes.
*/
static void
mly_fetch_event(struct mly_softc *mly)
{
struct mly_ccb *mc;
struct mly_cmd_ioctl *mci;
int s;
u_int32_t event;
/* Get a command. */
if (mly_ccb_alloc(mly, &mc))
return;
/* Set up the data buffer. */
mc->mc_data = malloc(sizeof(struct mly_event), M_DEVBUF,
M_NOWAIT|M_ZERO);
mc->mc_length = sizeof(struct mly_event);
mc->mc_flags |= MLY_CCB_DATAIN;
mc->mc_complete = mly_complete_event;
/*
* Get an event number to fetch. It's possible that we've raced
* with another context for the last event, in which case there will
* be no more events.
*/
s = splbio();
if (mly->mly_event_counter == mly->mly_event_waiting) {
splx(s);
free(mc->mc_data, M_DEVBUF);
mly_ccb_free(mly, mc);
return;
}
event = mly->mly_event_counter++;
splx(s);
/*
* Build the ioctl.
*
* At this point we are committed to sending this request, as it
* will be the only one constructed for this particular event
* number.
*/
mci = (struct mly_cmd_ioctl *)&mc->mc_packet->ioctl;
mci->opcode = MDACMD_IOCTL;
mci->data_size = htole32(sizeof(struct mly_event));
_lto3l(MLY_PHYADDR(0, 0, (event >> 16) & 0xff, (event >> 24) & 0xff),
mci->addr);
mci->timeout = 30 | MLY_TIMEOUT_SECONDS;
mci->sub_ioctl = MDACIOCTL_GETEVENT;
mci->param.getevent.sequence_number_low = htole16(event & 0xffff);
/*
* Submit the command.
*/
if (mly_ccb_map(mly, mc) != 0)
goto bad;
mly_ccb_enqueue(mly, mc);
return;
bad:
printf("%s: couldn't fetch event %u\n", mly->mly_dv.dv_xname, event);
free(mc->mc_data, M_DEVBUF);
mly_ccb_free(mly, mc);
}
/*
* Handle the completion of an event poll.
*/
static void
mly_complete_event(struct mly_softc *mly, struct mly_ccb *mc)
{
struct mly_event *me;
me = (struct mly_event *)mc->mc_data;
mly_ccb_unmap(mly, mc);
mly_ccb_free(mly, mc);
/* If the event was successfully fetched, process it. */
if (mc->mc_status == SCSI_OK)
mly_process_event(mly, me);
else
printf("%s: unable to fetch event; status = 0x%x\n",
mly->mly_dv.dv_xname, mc->mc_status);
free(me, M_DEVBUF);
/* Check for another event. */
mly_check_event(mly);
}
/*
* Process a controller event. Called with interrupts blocked (i.e., at
* interrupt time).
*/
static void
mly_process_event(struct mly_softc *mly, struct mly_event *me)
{
struct scsi_sense_data *ssd;
int bus, target, event, class, action;
const char *fp, *tp;
ssd = (struct scsi_sense_data *)&me->sense[0];
/*
* Errors can be reported using vendor-unique sense data. In this
* case, the event code will be 0x1c (Request sense data present),
* the sense key will be 0x09 (vendor specific), the MSB of the ASC
* will be set, and the actual event code will be a 16-bit value
* comprised of the ASCQ (low byte) and low seven bits of the ASC
* (low seven bits of the high byte).
*/
if (le32toh(me->code) == 0x1c &&
SSD_SENSE_KEY(ssd->flags) == SKEY_VENDOR_SPECIFIC &&
(ssd->asc & 0x80) != 0) {
event = ((int)(ssd->asc & ~0x80) << 8) +
ssd->ascq;
} else
event = le32toh(me->code);
/* Look up event, get codes. */
fp = mly_describe_code(mly_table_event, event);
/* Quiet event? */
class = fp[0];
#ifdef notyet
if (isupper(class) && bootverbose)
class = tolower(class);
#endif
/* Get action code, text string. */
action = fp[1];
tp = fp + 3;
/*
* Print some information about the event.
*
* This code uses a table derived from the corresponding portion of
* the Linux driver, and thus the parser is very similar.
*/
switch (class) {
case 'p':
/*
* Error on physical drive.
*/
printf("%s: physical device %d:%d %s\n", mly->mly_dv.dv_xname,
me->channel, me->target, tp);
if (action == 'r')
mly->mly_btl[me->channel][me->target].mb_flags |=
MLY_BTL_RESCAN;
break;
case 'l':
case 'm':
/*
* Error on logical unit, or message about logical unit.
*/
bus = MLY_LOGDEV_BUS(mly, me->lun);
target = MLY_LOGDEV_TARGET(mly, me->lun);
printf("%s: logical device %d:%d %s\n", mly->mly_dv.dv_xname,
bus, target, tp);
if (action == 'r')
mly->mly_btl[bus][target].mb_flags |= MLY_BTL_RESCAN;
break;
case 's':
/*
* Report of sense data.
*/
if ((SSD_SENSE_KEY(ssd->flags) == SKEY_NO_SENSE ||
SSD_SENSE_KEY(ssd->flags) == SKEY_NOT_READY) &&
ssd->asc == 0x04 &&
(ssd->ascq == 0x01 ||
ssd->ascq == 0x02)) {
/* Ignore NO_SENSE or NOT_READY in one case */
break;
}
/*
* XXX Should translate this if SCSIVERBOSE.
*/
printf("%s: physical device %d:%d %s\n", mly->mly_dv.dv_xname,
me->channel, me->target, tp);
printf("%s: sense key %d asc %02x ascq %02x\n",
mly->mly_dv.dv_xname, SSD_SENSE_KEY(ssd->flags),
ssd->asc, ssd->ascq);
printf("%s: info %x%x%x%x csi %x%x%x%x\n",
mly->mly_dv.dv_xname, ssd->info[0], ssd->info[1],
ssd->info[2], ssd->info[3], ssd->csi[0],
ssd->csi[1], ssd->csi[2],
ssd->csi[3]);
if (action == 'r')
mly->mly_btl[me->channel][me->target].mb_flags |=
MLY_BTL_RESCAN;
break;
case 'e':
printf("%s: ", mly->mly_dv.dv_xname);
printf(tp, me->target, me->lun);
break;
case 'c':
printf("%s: controller %s\n", mly->mly_dv.dv_xname, tp);
break;
case '?':
printf("%s: %s - %d\n", mly->mly_dv.dv_xname, tp, event);
break;
default:
/* Probably a 'noisy' event being ignored. */
break;
}
}
/*
* Create the monitoring thread. Called after the standard kernel threads
* have been created.
*/
static void
mly_thread_create(void *cookie)
{
struct mly_softc *mly;
int rv;
mly = cookie;
rv = kthread_create1(mly_thread, mly, &mly->mly_thread, "%s",
mly->mly_dv.dv_xname);
if (rv != 0)
printf("%s: unable to create thread (%d)\n",
mly->mly_dv.dv_xname, rv);
}
/*
* Perform periodic activities.
*/
static void
mly_thread(void *cookie)
{
struct mly_softc *mly;
struct mly_btl *btl;
int s, bus, target, done;
mly = (struct mly_softc *)cookie;
for (;;) {
/* Check for new events. */
mly_check_event(mly);
/* Re-scan up to 1 device. */
s = splbio();
done = 0;
for (bus = 0; bus < mly->mly_nchans && !done; bus++) {
for (target = 0; target < MLY_MAX_TARGETS; target++) {
/* Perform device rescan? */
btl = &mly->mly_btl[bus][target];
if ((btl->mb_flags & MLY_BTL_RESCAN) != 0) {
btl->mb_flags ^= MLY_BTL_RESCAN;
mly_scan_btl(mly, bus, target);
done = 1;
break;
}
}
}
splx(s);
/* Sleep for N seconds. */
tsleep(mly_thread, PWAIT, "mlyzzz",
hz * MLY_PERIODIC_INTERVAL);
}
}
/*
* Submit a command to the controller and poll on completion. Return
* non-zero on timeout.
*/
static int
mly_ccb_poll(struct mly_softc *mly, struct mly_ccb *mc, int timo)
{
int rv;
if ((rv = mly_ccb_submit(mly, mc)) != 0)
return (rv);
for (timo *= 10; timo != 0; timo--) {
if ((mc->mc_flags & MLY_CCB_COMPLETE) != 0)
break;
mly_intr(mly);
DELAY(100);
}
return (timo == 0);
}
/*
* Submit a command to the controller and sleep on completion. Return
* non-zero on timeout.
*/
static int
mly_ccb_wait(struct mly_softc *mly, struct mly_ccb *mc, int timo)
{
int rv, s;
mly_ccb_enqueue(mly, mc);
s = splbio();
if ((mc->mc_flags & MLY_CCB_COMPLETE) != 0) {
splx(s);
return (0);
}
rv = tsleep(mc, PRIBIO, "mlywccb", timo * hz / 1000);
splx(s);
return (rv);
}
/*
* If a CCB is specified, enqueue it. Pull CCBs off the software queue in
* the order that they were enqueued and try to submit their command blocks
* to the controller for execution.
*/
void
mly_ccb_enqueue(struct mly_softc *mly, struct mly_ccb *mc)
{
int s;
s = splbio();
if (mc != NULL)
SIMPLEQ_INSERT_TAIL(&mly->mly_ccb_queue, mc, mc_link.simpleq);
while ((mc = SIMPLEQ_FIRST(&mly->mly_ccb_queue)) != NULL) {
if (mly_ccb_submit(mly, mc))
break;
SIMPLEQ_REMOVE_HEAD(&mly->mly_ccb_queue, mc_link.simpleq);
}
splx(s);
}
/*
* Deliver a command to the controller.
*/
static int
mly_ccb_submit(struct mly_softc *mly, struct mly_ccb *mc)
{
union mly_cmd_packet *pkt;
int s, off;
mc->mc_packet->generic.command_id = htole16(mc->mc_slot);
bus_dmamap_sync(mly->mly_dmat, mly->mly_pkt_dmamap,
mc->mc_packetphys - mly->mly_pkt_busaddr,
sizeof(union mly_cmd_packet),
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
s = splbio();
/*
* Do we have to use the hardware mailbox?
*/
if ((mly->mly_state & MLY_STATE_MMBOX_ACTIVE) == 0) {
/*
* Check to see if the controller is ready for us.
*/
if (mly_idbr_true(mly, MLY_HM_CMDSENT)) {
splx(s);
return (EBUSY);
}
/*
* It's ready, send the command.
*/
mly_outl(mly, mly->mly_cmd_mailbox,
(u_int64_t)mc->mc_packetphys & 0xffffffff);
mly_outl(mly, mly->mly_cmd_mailbox + 4,
(u_int64_t)mc->mc_packetphys >> 32);
mly_outb(mly, mly->mly_idbr, MLY_HM_CMDSENT);
} else {
pkt = &mly->mly_mmbox->mmm_command[mly->mly_mmbox_cmd_idx];
off = (caddr_t)pkt - (caddr_t)mly->mly_mmbox;
bus_dmamap_sync(mly->mly_dmat, mly->mly_mmbox_dmamap,
off, sizeof(mly->mly_mmbox->mmm_command[0]),
BUS_DMASYNC_POSTWRITE | BUS_DMASYNC_POSTREAD);
/* Check to see if the next index is free yet. */
if (pkt->mmbox.flag != 0) {
splx(s);
return (EBUSY);
}
/* Copy in new command */
memcpy(pkt->mmbox.data, mc->mc_packet->mmbox.data,
sizeof(pkt->mmbox.data));
/* Copy flag last. */
pkt->mmbox.flag = mc->mc_packet->mmbox.flag;
bus_dmamap_sync(mly->mly_dmat, mly->mly_mmbox_dmamap,
off, sizeof(mly->mly_mmbox->mmm_command[0]),
BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
/* Signal controller and update index. */
mly_outb(mly, mly->mly_idbr, MLY_AM_CMDSENT);
mly->mly_mmbox_cmd_idx =
(mly->mly_mmbox_cmd_idx + 1) % MLY_MMBOX_COMMANDS;
}
splx(s);
return (0);
}
/*
* Pick up completed commands from the controller and handle accordingly.
*/
int
mly_intr(void *cookie)
{
struct mly_ccb *mc;
union mly_status_packet *sp;
u_int16_t slot;
int forus, off;
struct mly_softc *mly;
mly = cookie;
forus = 0;
/*
* Pick up hardware-mailbox commands.
*/
if (mly_odbr_true(mly, MLY_HM_STSREADY)) {
slot = mly_inw(mly, mly->mly_status_mailbox);
if (slot < MLY_SLOT_MAX) {
mc = mly->mly_ccbs + (slot - MLY_SLOT_START);
mc->mc_status =
mly_inb(mly, mly->mly_status_mailbox + 2);
mc->mc_sense =
mly_inb(mly, mly->mly_status_mailbox + 3);
mc->mc_resid =
mly_inl(mly, mly->mly_status_mailbox + 4);
mly_ccb_complete(mly, mc);
} else {
/* Slot 0xffff may mean "extremely bogus command". */
printf("%s: got HM completion for illegal slot %u\n",
mly->mly_dv.dv_xname, slot);
}
/* Unconditionally acknowledge status. */
mly_outb(mly, mly->mly_odbr, MLY_HM_STSREADY);
mly_outb(mly, mly->mly_idbr, MLY_HM_STSACK);
forus = 1;
}
/*
* Pick up memory-mailbox commands.
*/
if (mly_odbr_true(mly, MLY_AM_STSREADY)) {
for (;;) {
sp = &mly->mly_mmbox->mmm_status[mly->mly_mmbox_sts_idx];
off = (caddr_t)sp - (caddr_t)mly->mly_mmbox;
bus_dmamap_sync(mly->mly_dmat, mly->mly_mmbox_dmamap,
off, sizeof(mly->mly_mmbox->mmm_command[0]),
BUS_DMASYNC_POSTWRITE | BUS_DMASYNC_POSTREAD);
/* Check for more status. */
if (sp->mmbox.flag == 0)
break;
/* Get slot number. */
slot = le16toh(sp->status.command_id);
if (slot < MLY_SLOT_MAX) {
mc = mly->mly_ccbs + (slot - MLY_SLOT_START);
mc->mc_status = sp->status.status;
mc->mc_sense = sp->status.sense_length;
mc->mc_resid = le32toh(sp->status.residue);
mly_ccb_complete(mly, mc);
} else {
/*
* Slot 0xffff may mean "extremely bogus
* command".
*/
printf("%s: got AM completion for illegal "
"slot %u at %d\n", mly->mly_dv.dv_xname,
slot, mly->mly_mmbox_sts_idx);
}
/* Clear and move to next index. */
sp->mmbox.flag = 0;
mly->mly_mmbox_sts_idx =
(mly->mly_mmbox_sts_idx + 1) % MLY_MMBOX_STATUS;
}
/* Acknowledge that we have collected status value(s). */
mly_outb(mly, mly->mly_odbr, MLY_AM_STSREADY);
forus = 1;
}
/*
* Run the queue.
*/
if (forus && ! SIMPLEQ_EMPTY(&mly->mly_ccb_queue))
mly_ccb_enqueue(mly, NULL);
return (forus);
}
/*
* Process completed commands
*/
static void
mly_ccb_complete(struct mly_softc *mly, struct mly_ccb *mc)
{
void (*complete)(struct mly_softc *, struct mly_ccb *);
bus_dmamap_sync(mly->mly_dmat, mly->mly_pkt_dmamap,
mc->mc_packetphys - mly->mly_pkt_busaddr,
sizeof(union mly_cmd_packet),
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
complete = mc->mc_complete;
mc->mc_flags |= MLY_CCB_COMPLETE;
/*
* Call completion handler or wake up sleeping consumer.
*/
if (complete != NULL)
(*complete)(mly, mc);
else
wakeup(mc);
}
/*
* Allocate a command.
*/
int
mly_ccb_alloc(struct mly_softc *mly, struct mly_ccb **mcp)
{
struct mly_ccb *mc;
int s;
s = splbio();
mc = SLIST_FIRST(&mly->mly_ccb_free);
if (mc != NULL)
SLIST_REMOVE_HEAD(&mly->mly_ccb_free, mc_link.slist);
splx(s);
*mcp = mc;
return (mc == NULL ? EAGAIN : 0);
}
/*
* Release a command back to the freelist.
*/
void
mly_ccb_free(struct mly_softc *mly, struct mly_ccb *mc)
{
int s;
/*
* Fill in parts of the command that may cause confusion if a
* consumer doesn't when we are later allocated.
*/
mc->mc_data = NULL;
mc->mc_flags = 0;
mc->mc_complete = NULL;
mc->mc_private = NULL;
mc->mc_packet->generic.command_control = 0;
/*
* By default, we set up to overwrite the command packet with sense
* information.
*/
mc->mc_packet->generic.sense_buffer_address =
htole64(mc->mc_packetphys);
mc->mc_packet->generic.maximum_sense_size =
sizeof(union mly_cmd_packet);
s = splbio();
SLIST_INSERT_HEAD(&mly->mly_ccb_free, mc, mc_link.slist);
splx(s);
}
/*
* Allocate and initialize command and packet structures.
*
* If the controller supports fewer than MLY_MAX_CCBS commands, limit our
* allocation to that number. If we don't yet know how many commands the
* controller supports, allocate a very small set (suitable for initialization
* purposes only).
*/
static int
mly_alloc_ccbs(struct mly_softc *mly)
{
struct mly_ccb *mc;
int i, rv;
if (mly->mly_controllerinfo == NULL)
mly->mly_ncmds = MLY_CCBS_RESV;
else {
i = le16toh(mly->mly_controllerinfo->maximum_parallel_commands);
mly->mly_ncmds = min(MLY_MAX_CCBS, i);
}
/*
* Allocate enough space for all the command packets in one chunk
* and map them permanently into controller-visible space.
*/
rv = mly_dmamem_alloc(mly,
mly->mly_ncmds * sizeof(union mly_cmd_packet),
&mly->mly_pkt_dmamap, (caddr_t *)&mly->mly_pkt,
&mly->mly_pkt_busaddr, &mly->mly_pkt_seg);
if (rv)
return (rv);
mly->mly_ccbs = malloc(sizeof(struct mly_ccb) * mly->mly_ncmds,
M_DEVBUF, M_NOWAIT|M_ZERO);
for (i = 0; i < mly->mly_ncmds; i++) {
mc = mly->mly_ccbs + i;
mc->mc_slot = MLY_SLOT_START + i;
mc->mc_packet = mly->mly_pkt + i;
mc->mc_packetphys = mly->mly_pkt_busaddr +
(i * sizeof(union mly_cmd_packet));
rv = bus_dmamap_create(mly->mly_dmat, MLY_MAX_XFER,
MLY_MAX_SEGS, MLY_MAX_XFER, 0,
BUS_DMA_NOWAIT | BUS_DMA_ALLOCNOW,
&mc->mc_datamap);
if (rv) {
mly_release_ccbs(mly);
return (rv);
}
mly_ccb_free(mly, mc);
}
return (0);
}
/*
* Free all the storage held by commands.
*
* Must be called with all commands on the free list.
*/
static void
mly_release_ccbs(struct mly_softc *mly)
{
struct mly_ccb *mc;
/* Throw away command buffer DMA maps. */
while (mly_ccb_alloc(mly, &mc) == 0)
bus_dmamap_destroy(mly->mly_dmat, mc->mc_datamap);
/* Release CCB storage. */
free(mly->mly_ccbs, M_DEVBUF);
/* Release the packet storage. */
mly_dmamem_free(mly, mly->mly_ncmds * sizeof(union mly_cmd_packet),
mly->mly_pkt_dmamap, (caddr_t)mly->mly_pkt, &mly->mly_pkt_seg);
}
/*
* Map a command into controller-visible space.
*/
static int
mly_ccb_map(struct mly_softc *mly, struct mly_ccb *mc)
{
struct mly_cmd_generic *gen;
struct mly_sg_entry *sg;
bus_dma_segment_t *ds;
int flg, nseg, rv;
#ifdef DIAGNOSTIC
/* Don't map more than once. */
if ((mc->mc_flags & MLY_CCB_MAPPED) != 0)
panic("mly_ccb_map: already mapped");
mc->mc_flags |= MLY_CCB_MAPPED;
/* Does the command have a data buffer? */
if (mc->mc_data == NULL)
panic("mly_ccb_map: no data buffer");
#endif
rv = bus_dmamap_load(mly->mly_dmat, mc->mc_datamap, mc->mc_data,
mc->mc_length, NULL, BUS_DMA_NOWAIT | BUS_DMA_STREAMING |
((mc->mc_flags & MLY_CCB_DATAIN) != 0 ?
BUS_DMA_READ : BUS_DMA_WRITE));
if (rv != 0)
return (rv);
gen = &mc->mc_packet->generic;
/*
* Can we use the transfer structure directly?
*/
if ((nseg = mc->mc_datamap->dm_nsegs) <= 2) {
mc->mc_sgoff = -1;
sg = &gen->transfer.direct.sg[0];
} else {
mc->mc_sgoff = (mc->mc_slot - MLY_SLOT_START) *
MLY_MAX_SEGS;
sg = mly->mly_sg + mc->mc_sgoff;
gen->command_control |= MLY_CMDCTL_EXTENDED_SG_TABLE;
gen->transfer.indirect.entries[0] = htole16(nseg);
gen->transfer.indirect.table_physaddr[0] =
htole64(mly->mly_sg_busaddr +
(mc->mc_sgoff * sizeof(struct mly_sg_entry)));
}
/*
* Fill the S/G table.
*/
for (ds = mc->mc_datamap->dm_segs; nseg != 0; nseg--, sg++, ds++) {
sg->physaddr = htole64(ds->ds_addr);
sg->length = htole64(ds->ds_len);
}
/*
* Sync up the data map.
*/
if ((mc->mc_flags & MLY_CCB_DATAIN) != 0)
flg = BUS_DMASYNC_PREREAD;
else /* if ((mc->mc_flags & MLY_CCB_DATAOUT) != 0) */ {
gen->command_control |= MLY_CMDCTL_DATA_DIRECTION;
flg = BUS_DMASYNC_PREWRITE;
}
bus_dmamap_sync(mly->mly_dmat, mc->mc_datamap, 0, mc->mc_length, flg);
/*
* Sync up the chained S/G table, if we're using one.
*/
if (mc->mc_sgoff == -1)
return (0);
bus_dmamap_sync(mly->mly_dmat, mly->mly_sg_dmamap, mc->mc_sgoff,
MLY_SGL_SIZE, BUS_DMASYNC_PREWRITE);
return (0);
}
/*
* Unmap a command from controller-visible space.
*/
static void
mly_ccb_unmap(struct mly_softc *mly, struct mly_ccb *mc)
{
int flg;
#ifdef DIAGNOSTIC
if ((mc->mc_flags & MLY_CCB_MAPPED) == 0)
panic("mly_ccb_unmap: not mapped");
mc->mc_flags &= ~MLY_CCB_MAPPED;
#endif
if ((mc->mc_flags & MLY_CCB_DATAIN) != 0)
flg = BUS_DMASYNC_POSTREAD;
else /* if ((mc->mc_flags & MLY_CCB_DATAOUT) != 0) */
flg = BUS_DMASYNC_POSTWRITE;
bus_dmamap_sync(mly->mly_dmat, mc->mc_datamap, 0, mc->mc_length, flg);
bus_dmamap_unload(mly->mly_dmat, mc->mc_datamap);
if (mc->mc_sgoff == -1)
return;
bus_dmamap_sync(mly->mly_dmat, mly->mly_sg_dmamap, mc->mc_sgoff,
MLY_SGL_SIZE, BUS_DMASYNC_POSTWRITE);
}
/*
* Adjust the size of each I/O before it passes to the SCSI layer.
*/
static void
mly_scsipi_minphys(struct buf *bp)
{
if (bp->b_bcount > MLY_MAX_XFER)
bp->b_bcount = MLY_MAX_XFER;
minphys(bp);
}
/*
* Start a SCSI command.
*/
static void
mly_scsipi_request(struct scsipi_channel *chan, scsipi_adapter_req_t req,
void *arg)
{
struct mly_ccb *mc;
struct mly_cmd_scsi_small *ss;
struct scsipi_xfer *xs;
struct scsipi_periph *periph;
struct mly_softc *mly;
struct mly_btl *btl;
int s, tmp;
mly = (void *)chan->chan_adapter->adapt_dev;
switch (req) {
case ADAPTER_REQ_RUN_XFER:
xs = arg;
periph = xs->xs_periph;
btl = &mly->mly_btl[chan->chan_channel][periph->periph_target];
s = splbio();
tmp = btl->mb_flags;
splx(s);
/*
* Check for I/O attempt to a protected or non-existant
* device.
*/
if ((tmp & MLY_BTL_PROTECTED) != 0) {
xs->error = XS_SELTIMEOUT;
scsipi_done(xs);
break;
}
#ifdef DIAGNOSTIC
/* XXX Increase if/when we support large SCSI commands. */
if (xs->cmdlen > MLY_CMD_SCSI_SMALL_CDB) {
printf("%s: cmd too large\n", mly->mly_dv.dv_xname);
xs->error = XS_DRIVER_STUFFUP;
scsipi_done(xs);
break;
}
#endif
if (mly_ccb_alloc(mly, &mc)) {
xs->error = XS_RESOURCE_SHORTAGE;
scsipi_done(xs);
break;
}
/* Build the command. */
mc->mc_data = xs->data;
mc->mc_length = xs->datalen;
mc->mc_complete = mly_scsipi_complete;
mc->mc_private = xs;
/* Build the packet for the controller. */
ss = &mc->mc_packet->scsi_small;
ss->opcode = MDACMD_SCSI;
#ifdef notdef
/*
* XXX FreeBSD does this, but it doesn't fix anything,
* XXX and appears potentially harmful.
*/
ss->command_control |= MLY_CMDCTL_DISABLE_DISCONNECT;
#endif
ss->data_size = htole32(xs->datalen);
_lto3l(MLY_PHYADDR(0, chan->chan_channel,
periph->periph_target, periph->periph_lun), ss->addr);
if (xs->timeout < 60 * 1000)
ss->timeout = xs->timeout / 1000 |
MLY_TIMEOUT_SECONDS;
else if (xs->timeout < 60 * 60 * 1000)
ss->timeout = xs->timeout / (60 * 1000) |
MLY_TIMEOUT_MINUTES;
else
ss->timeout = xs->timeout / (60 * 60 * 1000) |
MLY_TIMEOUT_HOURS;
ss->maximum_sense_size = sizeof(xs->sense);
ss->cdb_length = xs->cmdlen;
memcpy(ss->cdb, xs->cmd, xs->cmdlen);
if (mc->mc_length != 0) {
if ((xs->xs_control & XS_CTL_DATA_OUT) != 0)
mc->mc_flags |= MLY_CCB_DATAOUT;
else /* if ((xs->xs_control & XS_CTL_DATA_IN) != 0) */
mc->mc_flags |= MLY_CCB_DATAIN;
if (mly_ccb_map(mly, mc) != 0) {
xs->error = XS_DRIVER_STUFFUP;
mly_ccb_free(mly, mc);
scsipi_done(xs);
break;
}
}
/*
* Give the command to the controller.
*/
if ((xs->xs_control & XS_CTL_POLL) != 0) {
if (mly_ccb_poll(mly, mc, xs->timeout + 5000)) {
xs->error = XS_REQUEUE;
if (mc->mc_length != 0)
mly_ccb_unmap(mly, mc);
mly_ccb_free(mly, mc);
scsipi_done(xs);
}
} else
mly_ccb_enqueue(mly, mc);
break;
case ADAPTER_REQ_GROW_RESOURCES:
/*
* Not supported.
*/
break;
case ADAPTER_REQ_SET_XFER_MODE:
/*
* We can't change the transfer mode, but at least let
* scsipi know what the adapter has negotiated.
*/
mly_get_xfer_mode(mly, chan->chan_channel, arg);
break;
}
}
/*
* Handle completion of a SCSI command.
*/
static void
mly_scsipi_complete(struct mly_softc *mly, struct mly_ccb *mc)
{
struct scsipi_xfer *xs;
struct scsipi_channel *chan;
struct scsipi_inquiry_data *inq;
struct mly_btl *btl;
int target, sl, s;
const char *p;
xs = mc->mc_private;
xs->status = mc->mc_status;
/*
* XXX The `resid' value as returned by the controller appears to be
* bogus, so we always set it to zero. Is it perhaps the transfer
* count?
*/
xs->resid = 0; /* mc->mc_resid; */
if (mc->mc_length != 0)
mly_ccb_unmap(mly, mc);
switch (mc->mc_status) {
case SCSI_OK:
/*
* In order to report logical device type and status, we
* overwrite the result of the INQUIRY command to logical
* devices.
*/
if (xs->cmd->opcode == INQUIRY) {
chan = xs->xs_periph->periph_channel;
target = xs->xs_periph->periph_target;
btl = &mly->mly_btl[chan->chan_channel][target];
s = splbio();
if ((btl->mb_flags & MLY_BTL_LOGICAL) != 0) {
inq = (struct scsipi_inquiry_data *)xs->data;
mly_padstr(inq->vendor, "MYLEX", 8);
p = mly_describe_code(mly_table_device_type,
btl->mb_type);
mly_padstr(inq->product, p, 16);
p = mly_describe_code(mly_table_device_state,
btl->mb_state);
mly_padstr(inq->revision, p, 4);
}
splx(s);
}
xs->error = XS_NOERROR;
break;
case SCSI_CHECK:
sl = mc->mc_sense;
if (sl > sizeof(xs->sense.scsi_sense))
sl = sizeof(xs->sense.scsi_sense);
memcpy(&xs->sense.scsi_sense, mc->mc_packet, sl);
xs->error = XS_SENSE;
break;
case SCSI_BUSY:
case SCSI_QUEUE_FULL:
xs->error = XS_BUSY;
break;
default:
printf("%s: unknown SCSI status 0x%x\n",
mly->mly_dv.dv_xname, xs->status);
xs->error = XS_DRIVER_STUFFUP;
break;
}
mly_ccb_free(mly, mc);
scsipi_done(xs);
}
/*
* Notify scsipi about a target's transfer mode.
*/
static void
mly_get_xfer_mode(struct mly_softc *mly, int bus, struct scsipi_xfer_mode *xm)
{
struct mly_btl *btl;
int s;
btl = &mly->mly_btl[bus][xm->xm_target];
xm->xm_mode = 0;
s = splbio();
if ((btl->mb_flags & MLY_BTL_PHYSICAL) != 0) {
if (btl->mb_speed == 0) {
xm->xm_period = 0;
xm->xm_offset = 0;
} else {
xm->xm_period = 12; /* XXX */
xm->xm_offset = 8; /* XXX */
xm->xm_mode |= PERIPH_CAP_SYNC; /* XXX */
}
switch (btl->mb_width) {
case 32:
xm->xm_mode = PERIPH_CAP_WIDE32;
break;
case 16:
xm->xm_mode = PERIPH_CAP_WIDE16;
break;
default:
xm->xm_mode = 0;
break;
}
} else /* ((btl->mb_flags & MLY_BTL_LOGICAL) != 0) */ {
xm->xm_mode = PERIPH_CAP_WIDE16 | PERIPH_CAP_SYNC;
xm->xm_period = 12;
xm->xm_offset = 8;
}
if ((btl->mb_flags & MLY_BTL_TQING) != 0)
xm->xm_mode |= PERIPH_CAP_TQING;
splx(s);
scsipi_async_event(&mly->mly_chans[bus], ASYNC_EVENT_XFER_MODE, xm);
}
/*
* ioctl hook; used here only to initiate low-level rescans.
*/
static int
mly_scsipi_ioctl(struct scsipi_channel *chan, u_long cmd, caddr_t data,
int flag, struct proc *p)
{
struct mly_softc *mly;
int rv;
mly = (struct mly_softc *)chan->chan_adapter->adapt_dev;
switch (cmd) {
case SCBUSIOLLSCAN:
mly_scan_channel(mly, chan->chan_channel);
rv = 0;
break;
default:
rv = ENOTTY;
break;
}
return (rv);
}
/*
* Handshake with the firmware while the card is being initialized.
*/
static int
mly_fwhandshake(struct mly_softc *mly)
{
u_int8_t error, param0, param1;
int spinup;
spinup = 0;
/* Set HM_STSACK and let the firmware initialize. */
mly_outb(mly, mly->mly_idbr, MLY_HM_STSACK);
DELAY(1000); /* too short? */
/* If HM_STSACK is still true, the controller is initializing. */
if (!mly_idbr_true(mly, MLY_HM_STSACK))
return (0);
printf("%s: controller initialization started\n",
mly->mly_dv.dv_xname);
/*
* Spin waiting for initialization to finish, or for a message to be
* delivered.
*/
while (mly_idbr_true(mly, MLY_HM_STSACK)) {
/* Check for a message */
if (!mly_error_valid(mly))
continue;
error = mly_inb(mly, mly->mly_error_status) & ~MLY_MSG_EMPTY;
param0 = mly_inb(mly, mly->mly_cmd_mailbox);
param1 = mly_inb(mly, mly->mly_cmd_mailbox + 1);
switch (error) {
case MLY_MSG_SPINUP:
if (!spinup) {
printf("%s: drive spinup in progress\n",
mly->mly_dv.dv_xname);
spinup = 1;
}
break;
case MLY_MSG_RACE_RECOVERY_FAIL:
printf("%s: mirror race recovery failed - \n",
mly->mly_dv.dv_xname);
printf("%s: one or more drives offline\n",
mly->mly_dv.dv_xname);
break;
case MLY_MSG_RACE_IN_PROGRESS:
printf("%s: mirror race recovery in progress\n",
mly->mly_dv.dv_xname);
break;
case MLY_MSG_RACE_ON_CRITICAL:
printf("%s: mirror race recovery on critical drive\n",
mly->mly_dv.dv_xname);
break;
case MLY_MSG_PARITY_ERROR:
printf("%s: FATAL MEMORY PARITY ERROR\n",
mly->mly_dv.dv_xname);
return (ENXIO);
default:
printf("%s: unknown initialization code 0x%x\n",
mly->mly_dv.dv_xname, error);
break;
}
}
return (0);
}
/*
* Space-fill a character string
*/
static void
mly_padstr(char *dst, const char *src, int len)
{
while (len-- > 0) {
if (*src != '\0')
*dst++ = *src++;
else
*dst++ = ' ';
}
}
/*
* Allocate DMA safe memory.
*/
static int
mly_dmamem_alloc(struct mly_softc *mly, int size, bus_dmamap_t *dmamap,
caddr_t *kva, bus_addr_t *paddr, bus_dma_segment_t *seg)
{
int rseg, rv, state;
state = 0;
if ((rv = bus_dmamem_alloc(mly->mly_dmat, size, PAGE_SIZE, 0,
seg, 1, &rseg, BUS_DMA_NOWAIT)) != 0) {
printf("%s: dmamem_alloc = %d\n", mly->mly_dv.dv_xname, rv);
goto bad;
}
state++;
if ((rv = bus_dmamem_map(mly->mly_dmat, seg, 1, size, kva,
BUS_DMA_NOWAIT | BUS_DMA_COHERENT)) != 0) {
printf("%s: dmamem_map = %d\n", mly->mly_dv.dv_xname, rv);
goto bad;
}
state++;
if ((rv = bus_dmamap_create(mly->mly_dmat, size, size, 1, 0,
BUS_DMA_NOWAIT, dmamap)) != 0) {
printf("%s: dmamap_create = %d\n", mly->mly_dv.dv_xname, rv);
goto bad;
}
state++;
if ((rv = bus_dmamap_load(mly->mly_dmat, *dmamap, *kva, size,
NULL, BUS_DMA_NOWAIT)) != 0) {
printf("%s: dmamap_load = %d\n", mly->mly_dv.dv_xname, rv);
goto bad;
}
*paddr = (*dmamap)->dm_segs[0].ds_addr;
memset(*kva, 0, size);
return (0);
bad:
if (state > 2)
bus_dmamap_destroy(mly->mly_dmat, *dmamap);
if (state > 1)
bus_dmamem_unmap(mly->mly_dmat, *kva, size);
if (state > 0)
bus_dmamem_free(mly->mly_dmat, seg, 1);
return (rv);
}
/*
* Free DMA safe memory.
*/
static void
mly_dmamem_free(struct mly_softc *mly, int size, bus_dmamap_t dmamap,
caddr_t kva, bus_dma_segment_t *seg)
{
bus_dmamap_unload(mly->mly_dmat, dmamap);
bus_dmamap_destroy(mly->mly_dmat, dmamap);
bus_dmamem_unmap(mly->mly_dmat, kva, size);
bus_dmamem_free(mly->mly_dmat, seg, 1);
}
/*
* Accept an open operation on the control device.
*/
int
mlyopen(dev_t dev, int flag, int mode, struct lwp *l)
{
struct mly_softc *mly;
if ((mly = device_lookup(&mly_cd, minor(dev))) == NULL)
return (ENXIO);
if ((mly->mly_state & MLY_STATE_INITOK) == 0)
return (ENXIO);
if ((mly->mly_state & MLY_STATE_OPEN) != 0)
return (EBUSY);
mly->mly_state |= MLY_STATE_OPEN;
return (0);
}
/*
* Accept the last close on the control device.
*/
int
mlyclose(dev_t dev, int flag, int mode, struct lwp *l)
{
struct mly_softc *mly;
mly = device_lookup(&mly_cd, minor(dev));
mly->mly_state &= ~MLY_STATE_OPEN;
return (0);
}
/*
* Handle control operations.
*/
int
mlyioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct lwp *l)
{
struct mly_softc *mly;
int rv;
if (securelevel >= 2)
return (EPERM);
mly = device_lookup(&mly_cd, minor(dev));
switch (cmd) {
case MLYIO_COMMAND:
rv = mly_user_command(mly, (void *)data);
break;
case MLYIO_HEALTH:
rv = mly_user_health(mly, (void *)data);
break;
default:
rv = ENOTTY;
break;
}
return (rv);
}
/*
* Execute a command passed in from userspace.
*
* The control structure contains the actual command for the controller, as
* well as the user-space data pointer and data size, and an optional sense
* buffer size/pointer. On completion, the data size is adjusted to the
* command residual, and the sense buffer size to the size of the returned
* sense data.
*/
static int
mly_user_command(struct mly_softc *mly, struct mly_user_command *uc)
{
struct mly_ccb *mc;
int rv, mapped;
if ((rv = mly_ccb_alloc(mly, &mc)) != 0)
return (rv);
mapped = 0;
mc->mc_data = NULL;
/*
* Handle data size/direction.
*/
if ((mc->mc_length = abs(uc->DataTransferLength)) != 0) {
if (mc->mc_length > MAXPHYS) {
rv = EINVAL;
goto out;
}
mc->mc_data = malloc(mc->mc_length, M_DEVBUF, M_WAITOK);
if (mc->mc_data == NULL) {
rv = ENOMEM;
goto out;
}
if (uc->DataTransferLength > 0) {
mc->mc_flags |= MLY_CCB_DATAIN;
memset(mc->mc_data, 0, mc->mc_length);
}
if (uc->DataTransferLength < 0) {
mc->mc_flags |= MLY_CCB_DATAOUT;
rv = copyin(uc->DataTransferBuffer, mc->mc_data,
mc->mc_length);
if (rv != 0)
goto out;
}
if ((rv = mly_ccb_map(mly, mc)) != 0)
goto out;
mapped = 1;
}
/* Copy in the command and execute it. */
memcpy(mc->mc_packet, &uc->CommandMailbox, sizeof(uc->CommandMailbox));
if ((rv = mly_ccb_wait(mly, mc, 60000)) != 0)
goto out;
/* Return the data to userspace. */
if (uc->DataTransferLength > 0) {
rv = copyout(mc->mc_data, uc->DataTransferBuffer,
mc->mc_length);
if (rv != 0)
goto out;
}
/* Return the sense buffer to userspace. */
if (uc->RequestSenseLength > 0 && mc->mc_sense > 0) {
rv = copyout(mc->mc_packet, uc->RequestSenseBuffer,
min(uc->RequestSenseLength, mc->mc_sense));
if (rv != 0)
goto out;
}
/* Return command results to userspace (caller will copy out). */
uc->DataTransferLength = mc->mc_resid;
uc->RequestSenseLength = min(uc->RequestSenseLength, mc->mc_sense);
uc->CommandStatus = mc->mc_status;
rv = 0;
out:
if (mapped)
mly_ccb_unmap(mly, mc);
if (mc->mc_data != NULL)
free(mc->mc_data, M_DEVBUF);
mly_ccb_free(mly, mc);
return (rv);
}
/*
* Return health status to userspace. If the health change index in the
* user structure does not match that currently exported by the controller,
* we return the current status immediately. Otherwise, we block until
* either interrupted or new status is delivered.
*/
static int
mly_user_health(struct mly_softc *mly, struct mly_user_health *uh)
{
struct mly_health_status mh;
int rv, s;
/* Fetch the current health status from userspace. */
rv = copyin(uh->HealthStatusBuffer, &mh, sizeof(mh));
if (rv != 0)
return (rv);
/* spin waiting for a status update */
s = splbio();
if (mly->mly_event_change == mh.change_counter)
rv = tsleep(&mly->mly_event_change, PRIBIO | PCATCH,
"mlyhealth", 0);
splx(s);
if (rv == 0) {
/*
* Copy the controller's health status buffer out (there is
* a race here if it changes again).
*/
rv = copyout(&mly->mly_mmbox->mmm_health.status,
uh->HealthStatusBuffer, sizeof(uh->HealthStatusBuffer));
}
return (rv);
}