Problem with FreeBSD for AMD64 & Mylex AcceleRAID

[Vyacheslav I.]

Hi!

I used to use the FreeBSD on i386 architecture together with the controller Mylex AcceleRAID 170.There was RAID 0+1 configured on it out of 5 discs (Enhanced Mirroring).

# pciconf -lvc
...
mly0 at pci0:6:1:0:        class=0x010400 card=0x00521069 chip=0x00501069 rev=0x02 hdr=0x00
    vendor     = 'Mylex Corp'
    device     = 'AcceleRAID Disk Array'
    class      = mass storage
    subclass   = RAID
    cap 01[80] = powerspec 2  supports D0 D3  current D0

I used this configuration on i386 architecture with the following versions FreeBSD: 4.x, 5.x, 6.x and 7.0. It always worked perfectly. But recently I got needed  8 GB RAM so I had to install the version FreeBSD for AMD64. First I tried the 7.0, and after all - RELENG_7_1 dated 12.12.2008. Both these systems work with the files system located on RAID unstablе! Meanwhile the same hardware on i386 architecture works with RELENG_7_1 perfectly.

Hardware: MB INTEL DG33FB, CPU Intel(R) Core(TM)2 Quad CPU @ 2.40 GHz, RAM 8 GB.

The base system is installed on a separate IDE disc.

# df -h
Filesystem     Size    Used   Avail Capacity  Mounted on
/dev/ad4s1a    496M    111M    345M    24%    /
devfs          1.0K    1.0K      0B   100%    /dev
/dev/ad4s1d    3.9G     22K    3.6G     0%    /tmp
/dev/ad4s1f    333G     52G    255G    17%    /usr
/dev/ad4s1e     15G     41M     14G     0%    /var
/dev/da0s1d   70.7G    410K   70.1G     0%    /mnt/da0/d

I carried out the following test by creating 3 big sized files of 1 GB each, and then deleting them:
# cd /mnt/da0/d
# dd if=/dev/zero of=test1 bs=1024k count=1024
# dd if=/dev/zero of=test2 bs=1024k count=1024
# dd if=/dev/zero of=test5 bs=1024k count=1024
# rm test*

After that I tried to unmount the files system, but I get a message of the kernel panic:
# cd /
# umount /mnt/da0/d

...
bad block 123456789, ino 176
dev = da0s1d, block = 4, fs = /mnt/da0/d
panic: ffs_blkfree: freeing free block
cpuid = 3
Uptime: 34 min...

The test was successful when using the i386 architecture with the same hardware

Taking a piece of advice of my friend, I used a dirty patch:
=== [code] ===
diff -Nru src.orig/sys/cam/scsi/scsi_da.c src/sys/cam/scsi/scsi_da.c
--- src.orig/sys/cam/scsi/scsi_da.c     2008-12-10 10:01:40.000000000 +0800
+++ src/sys/cam/scsi/scsi_da.c  2008-12-19 12:18:27.000000000 +0800
@@ -1187,7 +1187,7 @@
        if (match != NULL)
                softc->quirks = ((struct da_quirk_entry *)match)->quirks;
        else
-               softc->quirks = DA_Q_NONE;
+               softc->quirks = DA_Q_NO_SYNC_CACHE; // Dirty hack for AMD64

        /* Check if the SIM does not want 6 byte commands */
        xpt_setup_ccb(&cpi.ccb_h, periph->path, /*priority*/1);
=== [/code] ===

This patch solved the problem and the systems stopped being panicking. But I assume that this solution is wrong as the hardware works on i386 architecture without this patch well.