Date: Tue, 18 Jun 96 09:47:00 PDT From: Duncan Barclay <Duncan.Barclay@pa-consulting.com> To: send-pr <FreeBSD-gnats-submit@freebsd.org> Subject: i386/1331: changes and bug in ft driver Message-ID: <31C6E014@SMTPGATE.PA-CONSULTING.COM> Resent-Message-ID: <199606180900.CAA11456@freefall.freebsd.org>
next in thread | raw e-mail | index | archive | help
>Number: 1331
>Category: i386
>Synopsis: changes and bug in ft driver
>Confidential: no
>Severity: non-critical
>Priority: medium
>Responsible: freebsd-bugs
>State: open
>Class: sw-bug
>Submitter-Id: current-users
>Arrival-Date: Tue Jun 18 02:00:02 PDT 1996
>Last-Modified:
>Originator: Duncan Barclay &
>Organization:
PA-Consulting
>Release: FreeBSD 2.1-STABLE i386
>Environment:
486DX33, 8MB RAM UMC motherboard.
Connor TapeStor 800AT 800MB (Compressed) Travan Tapedrive
>Description:
From: Duncan Barclay
To: Poul-Henning Kamp
Subject: ft driver
Date: 11 June 1996 08:46
Hi Poul
I posted the below for -hackers a while back. You requested the
diffs but I forgot to send them to you. I remembered!
I have discovered some problems in the ft (floppy tape) driver code
when using my Connor Tapestore800 drive. I am running 2.0.5-R (but
am backing up prior to installing 2.1-R)
1) The list of recognised tape formats is limited. ft doesnt
like Travan cartridges. I have added the right
numbers to the table
but am not posting them as they changed when I
reformatted my cart using Connor's DOS software. I
dont want to start confusing people!
Someone who has the right specs should
update the entries.
2) One I fixed the above, all things were going okay using
the travan cart and cpio.
After a while the machine appeared to hang (I was in
single user mode at the
time, however later tests shows the process
to be waiting on bdone).
As I had hacked /sbin/ft to print
the segment number I continued to play a little seeking
the tape to that segment and kept
getting the hangs. Eventually I discovered that the
drive took a little longer than the code waits
for to respond when it moves the head a few tracks.
>How-To-Repeat:
in sys/i386/isa/ft.c
Specifics:
ACMD_BLOCKIO (or similar) calls ACMD_READID to find out
where the head is.
ACMD_READID issues read id commands to the drive.
If the first read id command returns garbage
then ACMD_READID then calls ACMD_SEEKSTS to stop the tape moving.
ACMD_SEEKSTS waits for the tape to stop and
to become ready, timing out after 10seconds.
If this timeout occurs then control goes back to
ACMD_READID and ACMD_READID should do a
further 4 retries.
It does not. It hangs. ***Important this is the bug***
My fix for _my_ tape drive is to up the timeout
specified by ACMD_READID in the call to ACMD_SEEKSTS
to 90seconds. In fact by drive takes about
12seconds to become ready. THIS IS NOT A REAL FIX. I think
that the retry stuff in ACMD_READID is broken (a comment
in the header of ft.c seems to indicate that
hangs did occur and may have been fixed but not tested).
I am willing to have ago at fixing this, but:
I have never written a device driver.
I have no time for 6months.
I have no idea how QIC tapes work at the s/w level.
The state machine in ft.c is big and it would take a
long time to trace out what I need to know.
I seem to remember someone saying they
were going to maintain this code a while back.
>Fix:
The code is the same in 2.0.5R and 2.1R so these are diffs between my
fixes and the 2.1R live filesystem CDROM (keep the Live Filesystem, it is
excellent)
*** ft.c Thu Apr 4 16:34:51 1996
--- /cdrom/usr/src/sys/i386/isa/ft.c Mon May 29 18:01:41 1995
***************
*** 91,97 ****
/* Enable or disable debugging messages. */
#define FTDBGALL 0 /* 1 if you want everything */
! /*#define DPRT(a) printf a */
#define DPRT(a)
/* Constants private to the driver */
--- 91,97 ----
/* Enable or disable debugging messages. */
#define FTDBGALL 0 /* 1 if you want everything */
! /*#define DPRT(a) printf a */
#define DPRT(a)
/* Constants private to the driver */
***************
*** 161,168 ****
{ 3, 2, "QIC-500", "307.5/550", 0, 0, 0, 0, 0 },
/* ??? */
{ 3, 3, "QIC-500", "295/900", 0, 0, 0, 0, 0 },
/* ??? */
{ 3, 4, "QIC-500", "1100/550", 0, 0, 0, 0, 0 },
/* ??? */
! { 3, 5, "QIC-500", "1100/900", 0, 0, 0, 0, 0 },
/* ??? */
! { 2,11, "TR-1", "750/550", 36, 369, 11808, 128, 32640 }
/* dmlb */
};
#define NGEOM (sizeof(ftgtbl) / sizeof(QIC_Geom))
--- 161,167 ----
{ 3, 2, "QIC-500", "307.5/550", 0, 0, 0, 0, 0 },
/* ??? */
{ 3, 3, "QIC-500", "295/900", 0, 0, 0, 0, 0 },
/* ??? */
{ 3, 4, "QIC-500", "1100/550", 0, 0, 0, 0, 0 },
/* ??? */
! { 3, 5, "QIC-500", "1100/900", 0, 0, 0, 0, 0 }
/* ??? */
};
#define NGEOM (sizeof(ftgtbl) / sizeof(QIC_Geom))
***************
*** 766,772 ****
CALL_ACMD(2, ACMD_STATUS, QC_STATUS, 8, 0);
case 2:
if ((async_ret & async_arg1) != 0) goto complete;
- DPRT (("seeksts retry, %d left\n", async_retries-1));
if (--async_retries == 0) {
DPRT(("ft%d: acmd_seeksts retries exceeded\n", ftu));
goto complete;
--- 765,770 ----
***************
*** 807,813 ****
* 4 microstep head back to 0
* 5 fail
*/
- DPRT (("readid retry number %d\n", errcnt+1));
if (++errcnt >= 5) {
DPRT(("ft%d: acmd_readid errcnt
exceeded\n", fdcu));
async_ret = -2;
--- 805,810 ----
***************
*** 816,825 ****
}
if (errcnt == 1) {
ft->moving = 0;
! CALL_ACMD(4, ACMD_SEEKSTS, QC_STOP,
QS_READY, 90);
} else {
ft->moving = 0;
! CALL_ACMD(4, ACMD_SEEKSTS,
QC_STPAUSE, QS_READY, 90);
}
DPRT(("readid retry %d...\n", errcnt));
async_state = 0;
--- 813,822 ----
}
if (errcnt == 1) {
ft->moving = 0;
! CALL_ACMD(4, ACMD_SEEKSTS, QC_STOP,
QS_READY, 0);
} else {
ft->moving = 0;
! CALL_ACMD(4, ACMD_SEEKSTS,
QC_STPAUSE, QS_READY, 0);
}
DPRT(("readid retry %d...\n", errcnt));
async_state = 0;
***************
*** 842,848 ****
async_ret = retpos+1;
goto complete;
case 4:
- DPRT (("readid state 4\n"));
CALL_ACMD(5, ACMD_SEEK, QC_FORWARD, 0, 0);
case 5:
ft->moving = 1;
--- 839,844 ----
***************
*** 1835,1844 ****
* XXX - This doesn't seem to work on my Colorado Jumbo 250...
* if it works on your drive, I'd sure like to hear about it.
*/
! /*
! * dmlb - But it does work on my Connor Tapestore 800.
! */
! #if 1
/* Report drive status */
for (sts = -1, tries = 0; sts < 0 && tries < 3; tries++)
sts = qic_status(ftu, QC_TSTATUS, 8);
--- 1831,1837 ----
* XXX - This doesn't seem to work on my Colorado Jumbo 250...
* if it works on your drive, I'd sure like to hear about it.
*/
! #if 0
/* Report drive status */
for (sts = -1, tries = 0; sts < 0 && tries < 3; tries++)
sts = qic_status(ftu, QC_TSTATUS, 8);
*** ftape.h Tue Apr 2 18:11:23 1996
--- /cdrom/usr/src/sys/sys/ftape.h Sat Aug 13 11:52:48 1994
***************
*** 39,45 ****
#define QCV_ECCSIZE 3072 /* Bytes ecc eats */
#define QCV_ECCBLKS 3 /* Blocks ecc eats */
#define QCV_NFMT 3 /* Number of tape formats */
! #define QCV_NLEN 11 /* Number of tape lengths */
#define QCV_HDRMAGIC 0xaa55aa55 /* Magic for header segment */
#define QCV_FSMAGIC 0x33cc33cc /* Magic for fileset */
--- 39,45 ----
#define QCV_ECCSIZE 3072 /* Bytes ecc eats */
#define QCV_ECCBLKS 3 /* Blocks ecc eats */
#define QCV_NFMT 3 /* Number of tape formats */
! #define QCV_NLEN 5 /* Number of tape lengths */
#define QCV_HDRMAGIC 0xaa55aa55 /* Magic for header segment */
#define QCV_FSMAGIC 0x33cc33cc /* Magic for fileset */
>Audit-Trail:
>Unformatted:
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?31C6E014>