freebsd-dev/sys/dev/ata/ata-raid.c
Scott Long 316ec49abd Some kernel threads try to do significant work, and the default KSTACK_PAGES
doesn't give them enough stack to do much before blowing away the pcb.
This adds MI and MD code to allow the allocation of an alternate kstack
who's size can be speficied when calling kthread_create.  Passing the
value 0 prevents the alternate kstack from being created.  Note that the
ia64 MD code is missing for now, and PowerPC was only partially written
due to the pmap.c being incomplete there.
Though this patch does not modify anything to make use of the alternate
kstack, acpi and usb are good candidates.

Reviewed by:	jake, peter, jhb
2002-10-02 07:44:29 +00:00

1452 lines
41 KiB
C

/*-
* Copyright (c) 2000,2001,2002 Søren Schmidt <sos@FreeBSD.org>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer,
* without modification, immediately at the beginning of the file.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* $FreeBSD$
*/
#include "opt_ata.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/ata.h>
#include <sys/kernel.h>
#include <sys/proc.h>
#include <sys/malloc.h>
#include <sys/bio.h>
#include <sys/bus.h>
#include <sys/conf.h>
#include <sys/disk.h>
#include <sys/devicestat.h>
#include <sys/cons.h>
#include <sys/unistd.h>
#include <sys/kthread.h>
#include <machine/bus.h>
#include <sys/rman.h>
#include <dev/ata/ata-all.h>
#include <dev/ata/ata-disk.h>
#include <dev/ata/ata-raid.h>
/* device structures */
static d_open_t aropen;
static d_strategy_t arstrategy;
static struct cdevsw ar_cdevsw = {
/* open */ aropen,
/* close */ nullclose,
/* read */ physread,
/* write */ physwrite,
/* ioctl */ noioctl,
/* poll */ nopoll,
/* mmap */ nommap,
/* strategy */ arstrategy,
/* name */ "ar",
/* maj */ 157,
/* dump */ nodump,
/* psize */ nopsize,
/* flags */ D_DISK,
};
static struct cdevsw ardisk_cdevsw;
/* prototypes */
static void ar_attach_raid(struct ar_softc *, int);
static void ar_done(struct bio *);
static void ar_config_changed(struct ar_softc *, int);
static void ar_rebuild(void *);
static int ar_highpoint_read_conf(struct ad_softc *, struct ar_softc **);
static int ar_highpoint_write_conf(struct ar_softc *);
static int ar_promise_read_conf(struct ad_softc *, struct ar_softc **, int);
static int ar_promise_write_conf(struct ar_softc *);
static int ar_rw(struct ad_softc *, u_int32_t, int, caddr_t, int);
static struct ata_device *ar_locate_disk(int);
/* internal vars */
static struct ar_softc **ar_table = NULL;
static MALLOC_DEFINE(M_AR, "AR driver", "ATA RAID driver");
int
ata_raiddisk_attach(struct ad_softc *adp)
{
struct ar_softc *rdp;
int array, disk;
if (ar_table) {
for (array = 0; array < MAX_ARRAYS; array++) {
if (!(rdp = ar_table[array]) || !rdp->flags)
continue;
for (disk = 0; disk < rdp->total_disks; disk++) {
if ((rdp->disks[disk].flags & AR_DF_ASSIGNED) &&
rdp->disks[disk].device == adp->device) {
ata_prtdev(rdp->disks[disk].device,
"inserted into ar%d disk%d as spare\n",
array, disk);
rdp->disks[disk].flags |= (AR_DF_PRESENT | AR_DF_SPARE);
AD_SOFTC(rdp->disks[disk])->flags = AD_F_RAID_SUBDISK;
ar_config_changed(rdp, 1);
return 1;
}
}
}
}
if (!ar_table)
ar_table = malloc(sizeof(struct ar_soft *) * MAX_ARRAYS,
M_AR, M_NOWAIT | M_ZERO);
if (!ar_table) {
ata_prtdev(adp->device, "no memory for ATA raid array\n");
return 0;
}
switch(adp->device->channel->chiptype) {
case 0x4d33105a: case 0x4d38105a: case 0x4d30105a:
case 0x0d30105a: case 0x4d68105a: case 0x6268105a:
case 0x4d69105a: case 0x5275105a: case 0x6269105a:
case 0x7275105a:
/* test RAID bit in PCI reg XXX */
return (ar_promise_read_conf(adp, ar_table, 0));
case 0x00041103: case 0x00051103: case 0x00081103:
return (ar_highpoint_read_conf(adp, ar_table));
default:
return (ar_promise_read_conf(adp, ar_table, 1));
}
return 0;
}
int
ata_raiddisk_detach(struct ad_softc *adp)
{
struct ar_softc *rdp;
int array, disk;
if (ar_table) {
for (array = 0; array < MAX_ARRAYS; array++) {
if (!(rdp = ar_table[array]) || !rdp->flags)
continue;
for (disk = 0; disk < rdp->total_disks; disk++) {
if (rdp->disks[disk].device == adp->device) {
ata_prtdev(rdp->disks[disk].device,
"deleted from ar%d disk%d\n", array, disk);
rdp->disks[disk].flags &= ~(AR_DF_PRESENT | AR_DF_ONLINE);
AD_SOFTC(rdp->disks[disk])->flags &= ~AD_F_RAID_SUBDISK;
ar_config_changed(rdp, 1);
return 1;
}
}
}
}
return 0;
}
void
ata_raid_attach()
{
struct ar_softc *rdp;
int array;
if (!ar_table)
return;
for (array = 0; array < MAX_ARRAYS; array++) {
if (!(rdp = ar_table[array]) || !rdp->flags)
continue;
ar_attach_raid(rdp, 0);
}
}
static void
ar_attach_raid(struct ar_softc *rdp, int update)
{
dev_t dev;
int disk;
ar_config_changed(rdp, update);
dev = disk_create(rdp->lun, &rdp->disk, 0, &ar_cdevsw, &ardisk_cdevsw);
dev->si_drv1 = rdp;
dev->si_iosize_max = 256 * DEV_BSIZE;
rdp->dev = dev;
printf("ar%d: %lluMB <ATA ", rdp->lun, (unsigned long long)
(rdp->total_sectors / ((1024L * 1024L) / DEV_BSIZE)));
switch (rdp->flags & (AR_F_RAID0 | AR_F_RAID1 | AR_F_SPAN)) {
case AR_F_RAID0:
printf("RAID0 "); break;
case AR_F_RAID1:
printf("RAID1 "); break;
case AR_F_SPAN:
printf("SPAN "); break;
case (AR_F_RAID0 | AR_F_RAID1):
printf("RAID0+1 "); break;
default:
printf("unknown 0x%x> ", rdp->flags);
return;
}
printf("array> [%d/%d/%d] status: ",
rdp->cylinders, rdp->heads, rdp->sectors);
switch (rdp->flags & (AR_F_DEGRADED | AR_F_READY)) {
case AR_F_READY:
printf("READY");
break;
case (AR_F_DEGRADED | AR_F_READY):
printf("DEGRADED");
break;
default:
printf("BROKEN");
break;
}
printf(" subdisks:\n");
for (disk = 0; disk < rdp->total_disks; disk++) {
if (rdp->disks[disk].flags & AR_DF_PRESENT) {
if (rdp->disks[disk].flags & AR_DF_ONLINE)
printf(" %d READY ", disk);
else if (rdp->disks[disk].flags & AR_DF_SPARE)
printf(" %d SPARE ", disk);
else
printf(" %d FREE ", disk);
ad_print(AD_SOFTC(rdp->disks[disk]));
}
else if (rdp->disks[disk].flags & AR_DF_ASSIGNED)
printf(" %d DOWN\n", disk);
else
printf(" %d INVALID no RAID config info on this disk\n", disk);
}
}
int
ata_raid_create(struct raid_setup *setup)
{
struct ata_device *atadev;
struct ar_softc *rdp;
int array, disk;
int ctlr = 0, disk_size = 0, total_disks = 0;
if (!ar_table)
ar_table = malloc(sizeof(struct ar_soft *) * MAX_ARRAYS,
M_AR, M_NOWAIT | M_ZERO);
if (!ar_table) {
printf("ar: no memory for ATA raid array\n");
return 0;
}
for (array = 0; array < MAX_ARRAYS; array++) {
if (!ar_table[array])
break;
}
if (array >= MAX_ARRAYS)
return ENOSPC;
if (!(rdp = (struct ar_softc*)malloc(sizeof(struct ar_softc), M_AR,
M_NOWAIT | M_ZERO))) {
printf("ar%d: failed to allocate raid config storage\n", array);
return ENOMEM;
}
for (disk = 0; disk < setup->total_disks; disk++) {
if ((atadev = ar_locate_disk(setup->disks[disk]))) {
rdp->disks[disk].device = atadev;
if (AD_SOFTC(rdp->disks[disk])->flags & AD_F_RAID_SUBDISK) {
setup->disks[disk] = -1;
free(rdp, M_AR);
return EBUSY;
}
switch (rdp->disks[disk].device->channel->chiptype & 0xffff) {
case 0x1103:
ctlr |= AR_F_HIGHPOINT_RAID;
rdp->disks[disk].disk_sectors =
AD_SOFTC(rdp->disks[disk])->total_secs;
break;
default:
ctlr |= AR_F_FREEBSD_RAID;
/* FALLTHROUGH */
case 0x105a:
ctlr |= AR_F_PROMISE_RAID;
rdp->disks[disk].disk_sectors =
PR_LBA(AD_SOFTC(rdp->disks[disk]));
break;
}
if (rdp->flags & (AR_F_PROMISE_RAID|AR_F_HIGHPOINT_RAID) &&
(rdp->flags & (AR_F_PROMISE_RAID|AR_F_HIGHPOINT_RAID)) !=
(ctlr & (AR_F_PROMISE_RAID|AR_F_HIGHPOINT_RAID))) {
free(rdp, M_AR);
return EXDEV;
}
else
rdp->flags |= ctlr;
if (disk_size)
disk_size = min(rdp->disks[disk].disk_sectors, disk_size);
else
disk_size = rdp->disks[disk].disk_sectors;
rdp->disks[disk].flags =
(AR_DF_PRESENT | AR_DF_ASSIGNED | AR_DF_ONLINE);
total_disks++;
}
else {
setup->disks[disk] = -1;
free(rdp, M_AR);
return ENXIO;
}
}
if (!total_disks) {
free(rdp, M_AR);
return ENODEV;
}
switch (setup->type) {
case 1:
rdp->flags |= AR_F_RAID0;
break;
case 2:
rdp->flags |= AR_F_RAID1;
if (total_disks != 2) {
free(rdp, M_AR);
return EPERM;
}
break;
case 3:
rdp->flags |= (AR_F_RAID0 | AR_F_RAID1);
if (total_disks % 2 != 0) {
free(rdp, M_AR);
return EPERM;
}
break;
case 4:
rdp->flags |= AR_F_SPAN;
break;
}
for (disk = 0; disk < total_disks; disk++)
AD_SOFTC(rdp->disks[disk])->flags = AD_F_RAID_SUBDISK;
rdp->lun = array;
if (rdp->flags & AR_F_RAID0) {
int bit = 0;
while (setup->interleave >>= 1)
bit++;
if (rdp->flags & AR_F_PROMISE_RAID)
rdp->interleave = min(max(2, 1 << bit), 2048);
if (rdp->flags & AR_F_HIGHPOINT_RAID)
rdp->interleave = min(max(32, 1 << bit), 128);
}
rdp->total_disks = total_disks;
rdp->width = total_disks / ((rdp->flags & AR_F_RAID1) ? 2 : 1);
rdp->total_sectors = disk_size * rdp->width;
rdp->heads = 255;
rdp->sectors = 63;
rdp->cylinders = rdp->total_sectors / (255 * 63);
if (rdp->flags & AR_F_PROMISE_RAID) {
rdp->offset = 0;
rdp->reserved = 63;
}
if (rdp->flags & AR_F_HIGHPOINT_RAID) {
rdp->offset = HPT_LBA + 1;
rdp->reserved = HPT_LBA + 1;
}
rdp->lock_start = rdp->lock_end = 0xffffffff;
rdp->flags |= AR_F_READY;
ar_table[array] = rdp;
ar_attach_raid(rdp, 1);
setup->unit = array;
return 0;
}
int
ata_raid_delete(int array)
{
struct ar_softc *rdp;
int disk;
if (!ar_table) {
printf("ar: no memory for ATA raid array\n");
return 0;
}
if (!(rdp = ar_table[array]))
return ENXIO;
rdp->flags &= ~AR_F_READY;
for (disk = 0; disk < rdp->total_disks; disk++) {
if ((rdp->disks[disk].flags&AR_DF_PRESENT) && rdp->disks[disk].device) {
AD_SOFTC(rdp->disks[disk])->flags &= ~AD_F_RAID_SUBDISK;
ata_drawerleds(rdp->disks[disk].device, ATA_LED_GREEN);
rdp->disks[disk].flags = 0;
}
}
if (rdp->flags & AR_F_PROMISE_RAID)
ar_promise_write_conf(rdp);
else
ar_highpoint_write_conf(rdp);
disk_invalidate(&rdp->disk);
disk_destroy(rdp->dev);
free(rdp, M_AR);
ar_table[array] = NULL;
return 0;
}
int
ata_raid_status(int array, struct raid_status *status)
{
struct ar_softc *rdp;
int i;
if (!ar_table || !(rdp = ar_table[array]))
return ENXIO;
switch (rdp->flags & (AR_F_RAID0 | AR_F_RAID1 | AR_F_SPAN)) {
case AR_F_RAID0:
status->type = AR_RAID0;
break;
case AR_F_RAID1:
status->type = AR_RAID1;
break;
case AR_F_RAID0 | AR_F_RAID1:
status->type = AR_RAID0 | AR_RAID1;
break;
case AR_F_SPAN:
status->type = AR_SPAN;
break;
}
status->total_disks = rdp->total_disks;
for (i = 0; i < rdp->total_disks; i++ ) {
if ((rdp->disks[i].flags & AR_DF_PRESENT) && rdp->disks[i].device)
status->disks[i] = AD_SOFTC(rdp->disks[i])->lun;
else
status->disks[i] = -1;
}
status->interleave = rdp->interleave;
status->status = 0;
if (rdp->flags & AR_F_READY)
status->status |= AR_READY;
if (rdp->flags & AR_F_DEGRADED)
status->status |= AR_DEGRADED;
if (rdp->flags & AR_F_REBUILDING) {
status->status |= AR_REBUILDING;
status->progress = 100*rdp->lock_start/(rdp->total_sectors/rdp->width);
}
return 0;
}
int
ata_raid_rebuild(int array)
{
struct ar_softc *rdp;
if (!ar_table || !(rdp = ar_table[array]))
return ENXIO;
if (rdp->flags & AR_F_REBUILDING)
return EBUSY;
return kthread_create(ar_rebuild, rdp, &rdp->pid, RFNOWAIT, 0,
"rebuilding ar%d", array);
}
static int
aropen(dev_t dev, int flags, int fmt, struct thread *td)
{
struct ar_softc *rdp = dev->si_drv1;
rdp->disk.d_sectorsize = DEV_BSIZE;
rdp->disk.d_mediasize = (off_t)rdp->total_sectors * DEV_BSIZE;
rdp->disk.d_fwsectors = rdp->sectors;
rdp->disk.d_fwheads = rdp->heads;
return 0;
}
static void
arstrategy(struct bio *bp)
{
struct ar_softc *rdp = bp->bio_dev->si_drv1;
int blkno, count, chunk, lba, lbs, tmplba;
int drv = 0, change = 0;
caddr_t data;
if (!(rdp->flags & AR_F_READY)) {
bp->bio_flags |= BIO_ERROR;
bp->bio_error = EIO;
biodone(bp);
return;
}
bp->bio_resid = bp->bio_bcount;
blkno = bp->bio_pblkno;
data = bp->bio_data;
for (count = howmany(bp->bio_bcount, DEV_BSIZE); count > 0;
count -= chunk, blkno += chunk, data += (chunk * DEV_BSIZE)) {
struct ar_buf *buf1, *buf2;
switch (rdp->flags & (AR_F_RAID0 | AR_F_RAID1 | AR_F_SPAN)) {
case AR_F_SPAN:
lba = blkno;
while (lba >= AD_SOFTC(rdp->disks[drv])->total_secs-rdp->reserved)
lba -= AD_SOFTC(rdp->disks[drv++])->total_secs-rdp->reserved;
chunk = min(AD_SOFTC(rdp->disks[drv])->total_secs-rdp->reserved-lba,
count);
break;
case AR_F_RAID0:
case AR_F_RAID0 | AR_F_RAID1:
tmplba = blkno / rdp->interleave;
chunk = blkno % rdp->interleave;
if (tmplba == rdp->total_sectors / rdp->interleave) {
lbs = (rdp->total_sectors-(tmplba*rdp->interleave))/rdp->width;
drv = chunk / lbs;
lba = ((tmplba/rdp->width)*rdp->interleave) + chunk%lbs;
chunk = min(count, lbs);
}
else {
drv = tmplba % rdp->width;
lba = ((tmplba / rdp->width) * rdp->interleave) + chunk;
chunk = min(count, rdp->interleave - chunk);
}
break;
case AR_F_RAID1:
drv = 0;
lba = blkno;
chunk = count;
break;
default:
printf("ar%d: unknown array type in arstrategy\n", rdp->lun);
bp->bio_flags |= BIO_ERROR;
bp->bio_error = EIO;
biodone(bp);
return;
}
buf1 = malloc(sizeof(struct ar_buf), M_AR, M_NOWAIT | M_ZERO);
buf1->bp.bio_pblkno = lba;
if ((buf1->drive = drv) > 0)
buf1->bp.bio_pblkno += rdp->offset;
buf1->bp.bio_caller1 = (void *)rdp;
buf1->bp.bio_bcount = chunk * DEV_BSIZE;
buf1->bp.bio_data = data;
buf1->bp.bio_cmd = bp->bio_cmd;
buf1->bp.bio_flags = bp->bio_flags;
buf1->bp.bio_done = ar_done;
buf1->org = bp;
switch (rdp->flags & (AR_F_RAID0 | AR_F_RAID1 | AR_F_SPAN)) {
case AR_F_SPAN:
case AR_F_RAID0:
if ((rdp->disks[buf1->drive].flags &
(AR_DF_PRESENT|AR_DF_ONLINE))==(AR_DF_PRESENT|AR_DF_ONLINE) &&
!AD_SOFTC(rdp->disks[buf1->drive])->dev->si_disk) {
rdp->disks[buf1->drive].flags &= ~AR_DF_ONLINE;
ar_config_changed(rdp, 1);
free(buf1, M_AR);
bp->bio_flags |= BIO_ERROR;
bp->bio_error = EIO;
biodone(bp);
return;
}
buf1->bp.bio_dev = AD_SOFTC(rdp->disks[buf1->drive])->dev;
AR_STRATEGY((struct bio *)buf1);
break;
case AR_F_RAID1:
case AR_F_RAID0 | AR_F_RAID1:
if (rdp->flags & AR_F_REBUILDING && bp->bio_cmd == BIO_WRITE) {
if ((bp->bio_pblkno >= rdp->lock_start &&
bp->bio_pblkno < rdp->lock_end) ||
((bp->bio_pblkno + chunk) > rdp->lock_start &&
(bp->bio_pblkno + chunk) <= rdp->lock_end)) {
tsleep(rdp, PRIBIO, "arwait", 0);
}
}
if ((rdp->disks[buf1->drive].flags &
(AR_DF_PRESENT|AR_DF_ONLINE))==(AR_DF_PRESENT|AR_DF_ONLINE) &&
!AD_SOFTC(rdp->disks[buf1->drive])->dev->si_disk) {
rdp->disks[buf1->drive].flags &= ~AR_DF_ONLINE;
change = 1;
}
if ((rdp->disks[buf1->drive + rdp->width].flags &
(AR_DF_PRESENT|AR_DF_ONLINE))==(AR_DF_PRESENT|AR_DF_ONLINE) &&
!AD_SOFTC(rdp->disks[buf1->drive + rdp->width])->dev->si_disk) {
rdp->disks[buf1->drive + rdp->width].flags &= ~AR_DF_ONLINE;
change = 1;
}
if (change)
ar_config_changed(rdp, 1);
if (!(rdp->flags & AR_F_READY)) {
free(buf1, M_AR);
bp->bio_flags |= BIO_ERROR;
bp->bio_error = EIO;
biodone(bp);
return;
}
if (bp->bio_cmd == BIO_READ) {
if ((buf1->bp.bio_pblkno <
(rdp->disks[buf1->drive].last_lba - AR_PROXIMITY) ||
buf1->bp.bio_pblkno >
(rdp->disks[buf1->drive].last_lba + AR_PROXIMITY) ||
!(rdp->disks[buf1->drive].flags & AR_DF_ONLINE)) &&
(rdp->disks[buf1->drive+rdp->width].flags & AR_DF_ONLINE))
buf1->drive = buf1->drive + rdp->width;
}
if (bp->bio_cmd == BIO_WRITE) {
if ((rdp->disks[buf1->drive+rdp->width].flags & AR_DF_ONLINE) ||
((rdp->flags & AR_F_REBUILDING) &&
(rdp->disks[buf1->drive+rdp->width].flags & AR_DF_SPARE) &&
buf1->bp.bio_pblkno < rdp->lock_start)) {
if ((rdp->disks[buf1->drive].flags & AR_DF_ONLINE) ||
((rdp->flags & AR_F_REBUILDING) &&
(rdp->disks[buf1->drive].flags & AR_DF_SPARE) &&
buf1->bp.bio_pblkno < rdp->lock_start)) {
buf2 = malloc(sizeof(struct ar_buf), M_AR, M_NOWAIT);
bcopy(buf1, buf2, sizeof(struct ar_buf));
buf1->mirror = buf2;
buf2->mirror = buf1;
buf2->drive = buf1->drive + rdp->width;
buf2->bp.bio_dev =
AD_SOFTC(rdp->disks[buf2->drive])->dev;
AR_STRATEGY((struct bio *)buf2);
rdp->disks[buf2->drive].last_lba =
buf2->bp.bio_pblkno + chunk;
}
else
buf1->drive = buf1->drive + rdp->width;
}
}
buf1->bp.bio_dev = AD_SOFTC(rdp->disks[buf1->drive])->dev;
AR_STRATEGY((struct bio *)buf1);
rdp->disks[buf1->drive].last_lba = buf1->bp.bio_pblkno + chunk;
break;
default:
printf("ar%d: unknown array type in arstrategy\n", rdp->lun);
}
}
}
static void
ar_done(struct bio *bp)
{
struct ar_softc *rdp = (struct ar_softc *)bp->bio_caller1;
struct ar_buf *buf = (struct ar_buf *)bp;
switch (rdp->flags & (AR_F_RAID0 | AR_F_RAID1 | AR_F_SPAN)) {
case AR_F_SPAN:
case AR_F_RAID0:
if (buf->bp.bio_flags & BIO_ERROR) {
rdp->disks[buf->drive].flags &= ~AR_DF_ONLINE;
ar_config_changed(rdp, 1);
buf->org->bio_flags |= BIO_ERROR;
buf->org->bio_error = EIO;
biodone(buf->org);
}
else {
buf->org->bio_resid -= buf->bp.bio_bcount;
if (buf->org->bio_resid == 0)
biodone(buf->org);
}
break;
case AR_F_RAID1:
case AR_F_RAID0 | AR_F_RAID1:
if (buf->bp.bio_flags & BIO_ERROR) {
rdp->disks[buf->drive].flags &= ~AR_DF_ONLINE;
ar_config_changed(rdp, 1);
if (rdp->flags & AR_F_READY) {
if (buf->bp.bio_cmd == BIO_READ) {
if (buf->drive < rdp->width)
buf->drive = buf->drive + rdp->width;
else
buf->drive = buf->drive - rdp->width;
buf->bp.bio_dev = AD_SOFTC(rdp->disks[buf->drive])->dev;
buf->bp.bio_flags = buf->org->bio_flags;
buf->bp.bio_error = 0;
AR_STRATEGY((struct bio *)buf);
return;
}
if (buf->bp.bio_cmd == BIO_WRITE) {
if (buf->flags & AB_F_DONE) {
buf->org->bio_resid -= buf->bp.bio_bcount;
if (buf->org->bio_resid == 0)
biodone(buf->org);
}
else
buf->mirror->flags |= AB_F_DONE;
}
}
else {
buf->org->bio_flags |= BIO_ERROR;
buf->org->bio_error = EIO;
biodone(buf->org);
}
}
else {
if (buf->bp.bio_cmd == BIO_WRITE) {
if (buf->mirror && !(buf->flags & AB_F_DONE)){
buf->mirror->flags |= AB_F_DONE;
break;
}
}
buf->org->bio_resid -= buf->bp.bio_bcount;
if (buf->org->bio_resid == 0)
biodone(buf->org);
}
break;
default:
printf("ar%d: unknown array type in ar_done\n", rdp->lun);
}
free(buf, M_AR);
}
static void
ar_config_changed(struct ar_softc *rdp, int writeback)
{
int disk, flags;
flags = rdp->flags;
rdp->flags |= AR_F_READY;
rdp->flags &= ~AR_F_DEGRADED;
for (disk = 0; disk < rdp->total_disks; disk++)
if (!(rdp->disks[disk].flags & AR_DF_PRESENT))
rdp->disks[disk].flags &= ~AR_DF_ONLINE;
for (disk = 0; disk < rdp->total_disks; disk++) {
switch (rdp->flags & (AR_F_RAID0 | AR_F_RAID1 | AR_F_SPAN)) {
case AR_F_SPAN:
case AR_F_RAID0:
if (!(rdp->disks[disk].flags & AR_DF_ONLINE)) {
rdp->flags &= ~AR_F_READY;
printf("ar%d: ERROR - array broken\n", rdp->lun);
}
break;
case AR_F_RAID1:
case AR_F_RAID0 | AR_F_RAID1:
if (disk < rdp->width) {
if (!(rdp->disks[disk].flags & AR_DF_ONLINE) &&
!(rdp->disks[disk + rdp->width].flags & AR_DF_ONLINE)) {
rdp->flags &= ~AR_F_READY;
printf("ar%d: ERROR - array broken\n", rdp->lun);
}
else if (((rdp->disks[disk].flags & AR_DF_ONLINE) &&
!(rdp->disks
[disk + rdp->width].flags & AR_DF_ONLINE))||
(!(rdp->disks[disk].flags & AR_DF_ONLINE) &&
(rdp->disks
[disk + rdp->width].flags & AR_DF_ONLINE))) {
rdp->flags |= AR_F_DEGRADED;
if (!(flags & AR_F_DEGRADED))
printf("ar%d: WARNING - mirror lost\n", rdp->lun);
}
}
break;
}
if ((rdp->disks[disk].flags&AR_DF_PRESENT) && rdp->disks[disk].device) {
if (rdp->disks[disk].flags & AR_DF_ONLINE)
ata_drawerleds(rdp->disks[disk].device, ATA_LED_GREEN);
else
ata_drawerleds(rdp->disks[disk].device, ATA_LED_RED);
}
}
if (writeback) {
if (rdp->flags & AR_F_PROMISE_RAID)
ar_promise_write_conf(rdp);
if (rdp->flags & AR_F_HIGHPOINT_RAID)
ar_highpoint_write_conf(rdp);
}
}
static void
ar_rebuild(void *arg)
{
struct ar_softc *rdp = arg;
int disk, s, count = 0, error = 0;
caddr_t buffer;
if ((rdp->flags & (AR_F_READY|AR_F_DEGRADED)) != (AR_F_READY|AR_F_DEGRADED))
kthread_exit(EEXIST);
for (disk = 0; disk < rdp->total_disks; disk++) {
if (((rdp->disks[disk].flags&(AR_DF_PRESENT|AR_DF_ONLINE|AR_DF_SPARE))==
(AR_DF_PRESENT | AR_DF_SPARE)) && rdp->disks[disk].device) {
if (AD_SOFTC(rdp->disks[disk])->total_secs <
rdp->disks[disk].disk_sectors) {
ata_prtdev(rdp->disks[disk].device,
"disk capacity too small for this RAID config\n");
#if 0
rdp->disks[disk].flags &= ~AR_DF_SPARE;
AD_SOFTC(rdp->disks[disk])->flags &= ~AD_F_RAID_SUBDISK;
#endif
continue;
}
ata_drawerleds(rdp->disks[disk].device, ATA_LED_ORANGE);
count++;
}
}
if (!count)
kthread_exit(ENODEV);
/* setup start conditions */
s = splbio();
rdp->lock_start = 0;
rdp->lock_end = rdp->lock_start + 256;
rdp->flags |= AR_F_REBUILDING;
splx(s);
buffer = malloc(256 * DEV_BSIZE, M_AR, M_NOWAIT | M_ZERO);
/* now go copy entire disk(s) */
while (rdp->lock_end < (rdp->total_sectors / rdp->width)) {
int size = min(256, (rdp->total_sectors / rdp->width) - rdp->lock_end);
for (disk = 0; disk < rdp->width; disk++) {
struct ad_softc *adp;
if (((rdp->disks[disk].flags & AR_DF_ONLINE) &&
(rdp->disks[disk + rdp->width].flags & AR_DF_ONLINE)) ||
((rdp->disks[disk].flags & AR_DF_ONLINE) &&
!(rdp->disks[disk + rdp->width].flags & AR_DF_SPARE)) ||
((rdp->disks[disk + rdp->width].flags & AR_DF_ONLINE) &&
!(rdp->disks[disk].flags & AR_DF_SPARE)))
continue;
if (rdp->disks[disk].flags & AR_DF_ONLINE)
adp = AD_SOFTC(rdp->disks[disk]);
else
adp = AD_SOFTC(rdp->disks[disk + rdp->width]);
if ((error = ar_rw(adp, rdp->lock_start,
size * DEV_BSIZE, buffer, AR_READ | AR_WAIT)))
break;
if (rdp->disks[disk].flags & AR_DF_ONLINE)
adp = AD_SOFTC(rdp->disks[disk + rdp->width]);
else
adp = AD_SOFTC(rdp->disks[disk]);
if ((error = ar_rw(adp, rdp->lock_start,
size * DEV_BSIZE, buffer, AR_WRITE | AR_WAIT)))
break;
}
if (error) {
wakeup(rdp);
free(buffer, M_AR);
kthread_exit(error);
}
s = splbio();
rdp->lock_start = rdp->lock_end;
rdp->lock_end = rdp->lock_start + size;
splx(s);
wakeup(rdp);
sprintf(rdp->pid->p_comm, "rebuilding ar%d %lld%%", rdp->lun,
(unsigned long long)(100 * rdp->lock_start /
(rdp->total_sectors / rdp->width)));
}
free(buffer, M_AR);
for (disk = 0; disk < rdp->total_disks; disk++) {
if ((rdp->disks[disk].flags&(AR_DF_PRESENT|AR_DF_ONLINE|AR_DF_SPARE))==
(AR_DF_PRESENT | AR_DF_SPARE)) {
rdp->disks[disk].flags &= ~AR_DF_SPARE;
rdp->disks[disk].flags |= (AR_DF_ASSIGNED | AR_DF_ONLINE);
}
}
s = splbio();
rdp->lock_start = 0xffffffff;
rdp->lock_end = 0xffffffff;
rdp->flags &= ~AR_F_REBUILDING;
splx(s);
ar_config_changed(rdp, 1);
kthread_exit(0);
}
static int
ar_highpoint_read_conf(struct ad_softc *adp, struct ar_softc **raidp)
{
struct highpoint_raid_conf *info;
struct ar_softc *raid = NULL;
int array, disk_number = 0, retval = 0;
if (!(info = (struct highpoint_raid_conf *)
malloc(sizeof(struct highpoint_raid_conf), M_AR, M_NOWAIT | M_ZERO)))
return retval;
if (ar_rw(adp, HPT_LBA, sizeof(struct highpoint_raid_conf),
(caddr_t)info, AR_READ | AR_WAIT)) {
if (bootverbose)
printf("ar: HighPoint read conf failed\n");
goto highpoint_out;
}
/* check if this is a HighPoint RAID struct */
if (info->magic != HPT_MAGIC_OK && info->magic != HPT_MAGIC_BAD) {
if (bootverbose)
printf("ar: HighPoint check1 failed\n");
goto highpoint_out;
}
/* is this disk defined, or an old leftover/spare ? */
if (!info->magic_0) {
if (bootverbose)
printf("ar: HighPoint check2 failed\n");
goto highpoint_out;
}
/* now convert HighPoint config info into our generic form */
for (array = 0; array < MAX_ARRAYS; array++) {
if (!raidp[array]) {
raidp[array] =
(struct ar_softc*)malloc(sizeof(struct ar_softc), M_AR,
M_NOWAIT | M_ZERO);
if (!raidp[array]) {
printf("ar%d: failed to allocate raid config storage\n", array);
goto highpoint_out;
}
}
raid = raidp[array];
if (raid->flags & AR_F_PROMISE_RAID)
continue;
switch (info->type) {
case HPT_T_RAID0:
if ((info->order & (HPT_O_RAID0|HPT_O_OK))==(HPT_O_RAID0|HPT_O_OK))
goto highpoint_raid1;
if (info->order & (HPT_O_RAID0 | HPT_O_RAID1))
goto highpoint_raid01;
if (raid->magic_0 && raid->magic_0 != info->magic_0)
continue;
raid->magic_0 = info->magic_0;
raid->flags |= AR_F_RAID0;
raid->interleave = 1 << info->stripe_shift;
disk_number = info->disk_number;
if (!(info->order & HPT_O_OK))
info->magic = 0; /* mark bad */
break;
case HPT_T_RAID1:
highpoint_raid1:
if (raid->magic_0 && raid->magic_0 != info->magic_0)
continue;
raid->magic_0 = info->magic_0;
raid->flags |= AR_F_RAID1;
disk_number = (info->disk_number > 0);
break;
case HPT_T_RAID01_RAID0:
highpoint_raid01:
if (info->order & HPT_O_RAID0) {
if ((raid->magic_0 && raid->magic_0 != info->magic_0) ||
(raid->magic_1 && raid->magic_1 != info->magic_1))
continue;
raid->magic_0 = info->magic_0;
raid->magic_1 = info->magic_1;
raid->flags |= (AR_F_RAID0 | AR_F_RAID1);
raid->interleave = 1 << info->stripe_shift;
disk_number = info->disk_number;
}
else {
if (raid->magic_1 && raid->magic_1 != info->magic_1)
continue;
raid->magic_1 = info->magic_1;
raid->flags |= (AR_F_RAID0 | AR_F_RAID1);
raid->interleave = 1 << info->stripe_shift;
disk_number = info->disk_number + info->array_width;
if (!(info->order & HPT_O_RAID1))
info->magic = 0; /* mark bad */
}
break;
case HPT_T_SPAN:
if (raid->magic_0 && raid->magic_0 != info->magic_0)
continue;
raid->magic_0 = info->magic_0;
raid->flags |= AR_F_SPAN;
disk_number = info->disk_number;
break;
default:
printf("ar%d: HighPoint unknown RAID type 0x%02x\n",
array, info->type);
goto highpoint_out;
}
raid->flags |= AR_F_HIGHPOINT_RAID;
raid->disks[disk_number].device = adp->device;
raid->disks[disk_number].flags = (AR_DF_PRESENT | AR_DF_ASSIGNED);
raid->lun = array;
if (info->magic == HPT_MAGIC_OK) {
raid->disks[disk_number].flags |= AR_DF_ONLINE;
raid->flags |= AR_F_READY;
raid->width = info->array_width;
raid->heads = 255;
raid->sectors = 63;
raid->cylinders = info->total_sectors / (63 * 255);
raid->total_sectors = info->total_sectors;
raid->offset = HPT_LBA + 1;
raid->reserved = HPT_LBA + 1;
raid->lock_start = raid->lock_end = info->rebuild_lba;
raid->disks[disk_number].disk_sectors =
info->total_sectors / info->array_width;
}
else
raid->disks[disk_number].flags &= ~ AR_DF_ONLINE;
if ((raid->flags & AR_F_RAID0) && (raid->total_disks < raid->width))
raid->total_disks = raid->width;
if (disk_number >= raid->total_disks)
raid->total_disks = disk_number + 1;
retval = 1;
break;
}
highpoint_out:
free(info, M_AR);
return retval;
}
static int
ar_highpoint_write_conf(struct ar_softc *rdp)
{
struct highpoint_raid_conf *config;
struct timeval timestamp;
int disk;
microtime(&timestamp);
rdp->magic_0 = timestamp.tv_sec + 2;
rdp->magic_1 = timestamp.tv_sec;
for (disk = 0; disk < rdp->total_disks; disk++) {
if (!(config = (struct highpoint_raid_conf *)
malloc(sizeof(struct highpoint_raid_conf),
M_AR, M_NOWAIT | M_ZERO))) {
printf("ar%d: Highpoint write conf failed\n", rdp->lun);
return -1;
}
if ((rdp->disks[disk].flags & (AR_DF_PRESENT | AR_DF_ONLINE)) ==
(AR_DF_PRESENT | AR_DF_ONLINE))
config->magic = HPT_MAGIC_OK;
if (rdp->disks[disk].flags & AR_DF_ASSIGNED) {
config->magic_0 = rdp->magic_0;
strcpy(config->name_1, "FreeBSD");
}
config->disk_number = disk;
switch (rdp->flags & (AR_F_RAID0 | AR_F_RAID1 | AR_F_SPAN)) {
case AR_F_RAID0:
config->type = HPT_T_RAID0;
strcpy(config->name_2, "RAID 0");
if (rdp->disks[disk].flags & AR_DF_ONLINE)
config->order = HPT_O_OK;
break;
case AR_F_RAID1:
config->type = HPT_T_RAID0;
strcpy(config->name_2, "RAID 1");
config->disk_number = (disk < rdp->width) ? disk : disk + 5;
config->order = HPT_O_RAID0 | HPT_O_OK;
break;
case AR_F_RAID0 | AR_F_RAID1:
config->type = HPT_T_RAID01_RAID0;
strcpy(config->name_2, "RAID 0+1");
if (rdp->disks[disk].flags & AR_DF_ONLINE) {
if (disk < rdp->width) {
config->order = (HPT_O_RAID0 | HPT_O_RAID1);
config->magic_0 = rdp->magic_0 - 1;
}
else {
config->order = HPT_O_RAID1;
config->disk_number -= rdp->width;
}
}
else
config->magic_0 = rdp->magic_0 - 1;
config->magic_1 = rdp->magic_1;
break;
case AR_F_SPAN:
config->type = HPT_T_SPAN;
strcpy(config->name_2, "SPAN");
break;
}
config->array_width = rdp->width;
config->stripe_shift = (rdp->width > 1) ? (ffs(rdp->interleave)-1) : 0;
config->total_sectors = rdp->total_sectors;
config->rebuild_lba = rdp->lock_start;
if (rdp->disks[disk].device && rdp->disks[disk].device->driver &&
!(rdp->disks[disk].device->flags & ATA_D_DETACHING)) {
if (ar_rw(AD_SOFTC(rdp->disks[disk]), HPT_LBA,
sizeof(struct highpoint_raid_conf),
(caddr_t)config, AR_WRITE)) {
printf("ar%d: Highpoint write conf failed\n", rdp->lun);
return -1;
}
}
}
return 0;
}
static int
ar_promise_read_conf(struct ad_softc *adp, struct ar_softc **raidp, int local)
{
struct promise_raid_conf *info;
struct ar_softc *raid;
u_int32_t magic, cksum, *ckptr;
int array, count, disk, disksum = 0, retval = 0;
if (!(info = (struct promise_raid_conf *)
malloc(sizeof(struct promise_raid_conf), M_AR, M_NOWAIT | M_ZERO)))
return retval;
if (ar_rw(adp, PR_LBA(adp), sizeof(struct promise_raid_conf),
(caddr_t)info, AR_READ | AR_WAIT)) {
if (bootverbose)
printf("ar: %s read conf failed\n", local ? "FreeBSD" : "Promise");
goto promise_out;
}
/* check if this is a Promise RAID struct (or our local one) */
if (local) {
if (strncmp(info->promise_id, ATA_MAGIC, sizeof(ATA_MAGIC))) {
if (bootverbose)
printf("ar: FreeBSD check1 failed\n");
goto promise_out;
}
}
else {
if (strncmp(info->promise_id, PR_MAGIC, sizeof(PR_MAGIC))) {
if (bootverbose)
printf("ar: Promise check1 failed\n");
goto promise_out;
}
}
/* check if the checksum is OK */
for (cksum = 0, ckptr = (int32_t *)info, count = 0; count < 511; count++)
cksum += *ckptr++;
if (cksum != *ckptr) {
if (bootverbose)
printf("ar: %s check2 failed\n", local ? "FreeBSD" : "Promise");
goto promise_out;
}
/* now convert Promise config info into our generic form */
if (info->raid.integrity != PR_I_VALID) {
if (bootverbose)
printf("ar: %s check3 failed\n", local ? "FreeBSD" : "Promise");
goto promise_out;
}
for (array = 0; array < MAX_ARRAYS; array++) {
if (!raidp[array]) {
raidp[array] =
(struct ar_softc*)malloc(sizeof(struct ar_softc), M_AR,
M_NOWAIT | M_ZERO);
if (!raidp[array]) {
printf("ar%d: failed to allocate raid config storage\n", array);
goto promise_out;
}
}
raid = raidp[array];
if (raid->flags & AR_F_HIGHPOINT_RAID)
continue;
magic = (adp->device->channel->chiptype >> 16) |
(info->raid.array_number << 16);
if (raid->flags & AR_F_PROMISE_RAID && magic != raid->magic_0)
continue;
/* update our knowledge about the array config based on generation */
if (!info->raid.generation || info->raid.generation > raid->generation){
raid->generation = info->raid.generation;
raid->flags = AR_F_PROMISE_RAID;
if (local)
raid->flags |= AR_F_FREEBSD_RAID;
raid->magic_0 = magic;
raid->lun = array;
if ((info->raid.status &
(PR_S_VALID | PR_S_ONLINE | PR_S_INITED | PR_S_READY)) ==
(PR_S_VALID | PR_S_ONLINE | PR_S_INITED | PR_S_READY)) {
raid->flags |= AR_F_READY;
if (info->raid.status & PR_S_DEGRADED)
raid->flags |= AR_F_DEGRADED;
}
else
raid->flags &= ~AR_F_READY;
switch (info->raid.type) {
case PR_T_RAID0:
raid->flags |= AR_F_RAID0;
break;
case PR_T_RAID1:
raid->flags |= AR_F_RAID1;
if (info->raid.array_width > 1)
raid->flags |= AR_F_RAID0;
break;
case PR_T_SPAN:
raid->flags |= AR_F_SPAN;
break;
default:
printf("ar%d: %s unknown RAID type 0x%02x\n",
array, local ? "FreeBSD" : "Promise", info->raid.type);
goto promise_out;
}
raid->interleave = 1 << info->raid.stripe_shift;
raid->width = info->raid.array_width;
raid->total_disks = info->raid.total_disks;
raid->heads = info->raid.heads + 1;
raid->sectors = info->raid.sectors;
raid->cylinders = info->raid.cylinders + 1;
raid->total_sectors = info->raid.total_sectors;
raid->offset = 0;
raid->reserved = 63;
raid->lock_start = raid->lock_end = info->raid.rebuild_lba;
/* convert disk flags to our internal types */
for (disk = 0; disk < info->raid.total_disks; disk++) {
raid->disks[disk].flags = 0;
disksum += info->raid.disk[disk].flags;
if (info->raid.disk[disk].flags & PR_F_ONLINE)
raid->disks[disk].flags |= AR_DF_ONLINE;
if (info->raid.disk[disk].flags & PR_F_ASSIGNED)
raid->disks[disk].flags |= AR_DF_ASSIGNED;
if (info->raid.disk[disk].flags & PR_F_SPARE) {
raid->disks[disk].flags &= ~AR_DF_ONLINE;
raid->disks[disk].flags |= AR_DF_SPARE;
}
if (info->raid.disk[disk].flags & (PR_F_REDIR | PR_F_DOWN))
raid->disks[disk].flags &= ~AR_DF_ONLINE;
}
if (!disksum) {
free(raidp[array], M_AR);
raidp[array] = NULL;
goto promise_out;
}
}
if (raid->disks[info->raid.disk_number].flags && adp->device) {
raid->disks[info->raid.disk_number].device = adp->device;
raid->disks[info->raid.disk_number].flags |= AR_DF_PRESENT;
raid->disks[info->raid.disk_number].disk_sectors =
info->raid.disk_sectors;
retval = 1;
}
break;
}
promise_out:
free(info, M_AR);
return retval;
}
static int
ar_promise_write_conf(struct ar_softc *rdp)
{
struct promise_raid_conf *config;
struct timeval timestamp;
u_int32_t *ckptr;
int count, disk, drive;
int local = rdp->flags & AR_F_FREEBSD_RAID;
rdp->generation++;
microtime(&timestamp);
for (disk = 0; disk < rdp->total_disks; disk++) {
if (!(config = (struct promise_raid_conf *)
malloc(sizeof(struct promise_raid_conf), M_AR, M_NOWAIT))) {
printf("ar%d: %s write conf failed\n",
rdp->lun, local ? "FreeBSD" : "Promise");
return -1;
}
for (count = 0; count < sizeof(struct promise_raid_conf); count++)
*(((u_int8_t *)config) + count) = 255 - (count % 256);
if (local)
bcopy(ATA_MAGIC, config->promise_id, sizeof(ATA_MAGIC));
else
bcopy(PR_MAGIC, config->promise_id, sizeof(PR_MAGIC));
config->dummy_0 = 0x00020000;
config->magic_0 = PR_MAGIC0(rdp->disks[disk]) | timestamp.tv_sec;
config->magic_1 = timestamp.tv_sec >> 16;
config->magic_2 = timestamp.tv_sec;
config->raid.integrity = PR_I_VALID;
config->raid.disk_number = disk;
if (rdp->disks[disk].flags & AR_DF_PRESENT && rdp->disks[disk].device) {
config->raid.channel = rdp->disks[disk].device->channel->unit;
config->raid.device = (rdp->disks[disk].device->unit != 0);
if (AD_SOFTC(rdp->disks[disk])->dev->si_disk)
config->raid.disk_sectors = PR_LBA(AD_SOFTC(rdp->disks[disk]));
/*config->raid.disk_offset*/
}
config->raid.magic_0 = config->magic_0;
config->raid.rebuild_lba = rdp->lock_start;
config->raid.generation = rdp->generation;
if (rdp->flags & AR_F_READY) {
config->raid.flags = (PR_F_VALID | PR_F_ASSIGNED | PR_F_ONLINE);
config->raid.status =
(PR_S_VALID | PR_S_ONLINE | PR_S_INITED | PR_S_READY);
if (rdp->flags & AR_F_DEGRADED)
config->raid.status |= PR_S_DEGRADED;
else
config->raid.status |= PR_S_FUNCTIONAL;
}
else {
config->raid.flags = PR_F_DOWN;
config->raid.status = 0;
}
switch (rdp->flags & (AR_F_RAID0 | AR_F_RAID1 | AR_F_SPAN)) {
case AR_F_RAID0:
config->raid.type = PR_T_RAID0;
break;
case AR_F_RAID1:
config->raid.type = PR_T_RAID1;
break;
case AR_F_RAID0 | AR_F_RAID1:
config->raid.type = PR_T_RAID1;
break;
case AR_F_SPAN:
config->raid.type = PR_T_SPAN;
break;
}
config->raid.total_disks = rdp->total_disks;
config->raid.stripe_shift = ffs(rdp->interleave) - 1;
config->raid.array_width = rdp->width;
config->raid.array_number = rdp->lun;
config->raid.total_sectors = rdp->total_sectors;
config->raid.cylinders = rdp->cylinders - 1;
config->raid.heads = rdp->heads - 1;
config->raid.sectors = rdp->sectors;
config->raid.magic_1 = (u_int64_t)config->magic_2<<16 | config->magic_1;
bzero(&config->raid.disk, 8 * 12);
for (drive = 0; drive < rdp->total_disks; drive++) {
config->raid.disk[drive].flags = 0;
if (rdp->disks[drive].flags & AR_DF_PRESENT)
config->raid.disk[drive].flags |= PR_F_VALID;
if (rdp->disks[drive].flags & AR_DF_ASSIGNED)
config->raid.disk[drive].flags |= PR_F_ASSIGNED;
if (rdp->disks[drive].flags & AR_DF_ONLINE)
config->raid.disk[drive].flags |= PR_F_ONLINE;
else
if (rdp->disks[drive].flags & AR_DF_PRESENT)
config->raid.disk[drive].flags = (PR_F_REDIR | PR_F_DOWN);
if (rdp->disks[drive].flags & AR_DF_SPARE)
config->raid.disk[drive].flags |= PR_F_SPARE;
config->raid.disk[drive].dummy_0 = 0x0;
if (rdp->disks[drive].device) {
config->raid.disk[drive].channel =
rdp->disks[drive].device->channel->unit;
config->raid.disk[drive].device =
(rdp->disks[drive].device->unit != 0);
}
config->raid.disk[drive].magic_0 =
PR_MAGIC0(rdp->disks[drive]) | timestamp.tv_sec;
}
config->checksum = 0;
for (ckptr = (int32_t *)config, count = 0; count < 511; count++)
config->checksum += *ckptr++;
if (rdp->disks[disk].device && rdp->disks[disk].device->driver &&
!(rdp->disks[disk].device->flags & ATA_D_DETACHING)) {
if (ar_rw(AD_SOFTC(rdp->disks[disk]),
PR_LBA(AD_SOFTC(rdp->disks[disk])),
sizeof(struct promise_raid_conf),
(caddr_t)config, AR_WRITE)) {
printf("ar%d: %s write conf failed\n",
rdp->lun, local ? "FreeBSD" : "Promise");
return -1;
}
}
}
return 0;
}
static void
ar_rw_done(struct bio *bp)
{
free(bp->bio_data, M_AR);
free(bp, M_AR);
}
static int
ar_rw(struct ad_softc *adp, u_int32_t lba, int count, caddr_t data, int flags)
{
struct bio *bp;
int retry = 0, error = 0;
if (!(bp = (struct bio *)malloc(sizeof(struct bio), M_AR, M_NOWAIT|M_ZERO)))
return 1;
bp->bio_dev = adp->dev;
bp->bio_data = data;
bp->bio_pblkno = lba;
bp->bio_bcount = count;
if (flags & AR_READ)
bp->bio_cmd = BIO_READ;
if (flags & AR_WRITE)
bp->bio_cmd = BIO_WRITE;
if (flags & AR_WAIT)
bp->bio_done = (void *)wakeup;
else
bp->bio_done = ar_rw_done;
AR_STRATEGY(bp);
if (flags & AR_WAIT) {
while ((retry++ < (15*hz/10)) && (error = !(bp->bio_flags & BIO_DONE)))
error = tsleep(bp, PRIBIO, "arrw", 10);
if (!error && bp->bio_flags & BIO_ERROR)
error = bp->bio_error;
free(bp, M_AR);
}
return error;
}
static struct ata_device *
ar_locate_disk(int diskno)
{
struct ata_channel *ch;
int ctlr;
for (ctlr = 0; ctlr < devclass_get_maxunit(ata_devclass); ctlr++) {
if (!(ch = devclass_get_softc(ata_devclass, ctlr)))
continue;
if (ch->devices & ATA_ATA_MASTER)
if (ch->device[MASTER].driver &&
((struct ad_softc *)(ch->device[MASTER].driver))->lun == diskno)
return &ch->device[MASTER];
if (ch->devices & ATA_ATA_SLAVE)
if (ch->device[SLAVE].driver &&
((struct ad_softc *)(ch->device[SLAVE].driver))->lun == diskno)
return &ch->device[SLAVE];
}
return NULL;
}