Revision a45dc2d2b8d1afa57c91dcfac224e50ffcd3f805 authored by Benny Halevy on 13 September 2010, 19:32:19 UTC, committed by Jens Axboe on 21 September 2010, 09:49:17 UTC
This bug was introduced in 7b6d91daee5cac6402186ff224c3af39d79f4a0e "block: unify flags for struct bio and struct request" Cc: Boaz Harrosh <bharrosh@panasas.com> Signed-off-by: Benny Halevy <bhalevy@panasas.com> Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
1 parent b0722cb
dtc3x80.txt
README file for the Linux DTC3180/3280 scsi driver.
by Ray Van Tassle (rayvt@comm.mot.com) March 1996
Based on the generic & core NCR5380 code by Drew Eckhard
SCSI device driver for the DTC 3180/3280.
Data Technology Corp---a division of Qume.
The 3280 has a standard floppy interface.
The 3180 does not. Otherwise, they are identical.
The DTC3x80 does not support DMA but it does have Pseudo-DMA which is
supported by the driver.
Its DTC406 scsi chip is supposedly compatible with the NCR 53C400.
It is memory mapped, uses an IRQ, but no dma or io-port. There is
internal DMA, between SCSI bus and an on-chip 128-byte buffer. Double
buffering is done automagically by the chip. Data is transferred
between the on-chip buffer and CPU/RAM via memory moves.
The driver detects the possible memory addresses (jumper selectable):
CC00, DC00, C800, and D800
The possible IRQ's (jumper selectable) are:
IRQ 10, 11, 12, 15
Parity is supported by the chip, but not by this driver.
Information can be obtained from /proc/scsi/dtc3c80/N.
Note on interrupts:
The documentation says that it can be set to interrupt whenever the
on-chip buffer needs CPU attention. I couldn't get this to work. So
the driver polls for data-ready in the pseudo-DMA transfer routine.
The interrupt support routines in the NCR3280.c core modules handle
scsi disconnect/reconnect, and this (mostly) works. However..... I
have tested it with 4 totally different hard drives (both SCSI-1 and
SCSI-2), and one CDROM drive. Interrupts works great for all but one
specific hard drive. For this one, the driver will eventually hang in
the transfer state. I have tested with: "dd bs=4k count=2k
of=/dev/null if=/dev/sdb". It reads ok for a while, then hangs.
After beating my head against this for a couple of weeks, getting
nowhere, I give up. So.....This driver does NOT use interrupts, even
if you have the card jumpered to an IRQ. Probably nobody will ever
care.
![swh spinner](/static/img/swh-spinner.gif)
Computing file changes ...