Aren
/
NetBSD
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix data-corruption bug on async targets: 

*  When we are transferring in DATA (in asc_dma_in) and the target
  is an async device, there is sometimes an extra byte in the FIFO.
  If so, we need to drain that byte out of the fifo, but if and only
  if the target is async.  See also the comments in asc_dma_in()
  in the related Mach mk84 asc driver (scsi_53C94_hdw.c), which
  has an identical fix but applied in more restrictive conditions
  than we need, with async *disk* targets, as well as async tapes.

* Add a watchdog and timeout active SCSI requests, to eliminate any
  potential for deadlock due to applying the fix above on newer
  silicon versions of the 53c94 which may not have the above problem.
  Should use the MI scsi per-target timeout instead, when available.
This commit is contained in:
jonathan 1996-09-29 03:02:19 +00:00
parent c347bc09f1
commit 9b84ee7f35
1 changed files with 56 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

61
sys/dev/tc/asc.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: asc.c,v 1.28 1996/09/25 21:07:46 jonathan Exp $ */
/* $NetBSD: asc.c,v 1.29 1996/09/29 03:02:19 jonathan Exp $ */
/*-
* Copyright (c) 1992, 1993
@ -128,6 +128,7 @@
#include <sys/buf.h>
#include <sys/conf.h>
#include <sys/errno.h>
#include <sys/kernel.h>
#include <sys/device.h>
#include <sys/reboot.h>
@ -408,6 +409,7 @@ script_t asc_scripts[] = {
*/
static void asc_reset __P((asc_softc_t asc, asc_regmap_t *regs));
static void asc_startcmd __P((asc_softc_t asc, int target));
static void asc_timeout __P((void *arg));
extern struct cfdriver asc_cd;
struct cfdriver asc_cd = {
@ -598,6 +600,12 @@ asc_start(scsicmd)
splx(s);
}
asc->cmd[sdp->sd_drive] = scsicmd;
/*
* Kludge: use a 60 second timeout if data is being transfered,
* otherwise use a 30 minute timeout.
*/
timeout(asc_timeout, scsicmd, hz * (scsicmd->buflen == 0 ?
1800 : 60));
asc_startcmd(asc, sdp->sd_drive);
splx(s);
}
@ -1315,6 +1323,7 @@ asc_end(asc, status, ss, ir)
scsicmd = asc->cmd[target];
asc->cmd[target] = (ScsiCmd *)0;
state = &asc->st[target];
untimeout(asc_timeout, scsicmd);
#ifdef DEBUG
if (asc_debug > 1) {
@ -1408,13 +1417,36 @@ asc_dma_in(asc, status, ss, ir)
state->buflen -= len;
}
#ifdef DEBUG
if (!(state->flags & DMA_IN_PROGRESS) &&
(regs->asc_flags & ASC_FLAGS_FIFO_CNT) != 0) {
printf("asc_dma_in: FIFO count %x flags %x\n",
regs->asc_flags, state->flags);
volatile int async_fifo_junk = 0;
/*
* If the target is asynchronous, the FIFO contains
* a byte of garbage. (see the Mach mk84 53c94 driver,
* where this occurs on tk-50s and exabytes.)
* It also occurs on asynch disks like SCSI-1 disks.
* Recover by reading the byte of junk from the fifo if,
* and only if, the target is async. If the target is
* synch, there is no junk, and reading the fifo
* deadlocks our SCSI state machine.
*/
if (state->sync_offset == 0)
async_fifo_junk = regs->asc_fifo;
#ifdef DEBUG
printf("%s: asc_dma_in: FIFO count %x flags %x sync_offset %d",
asc->sc_dev.dv_xname, regs->asc_flags,
state->flags, state->sync_offset);
if (state->sync_offset != 0)
printf("\n");
else
printf(" unexpected fifo data %x\n", async_fifo_junk);
#ifdef DIAGNOSTIC
asc_DumpLog("asc_dma_in");
#endif /* DIAGNOSTIC */
#endif /* DEBUG */
}
#endif
/* setup to start reading the next chunk */
len = state->buflen;
#ifdef DEBUG
@ -2011,6 +2043,25 @@ asc_disconnect(asc, status, ss, ir)
}
void
asc_timeout(arg)
void *arg;
{
int s = splbio();
ScsiCmd *scsicmd = (ScsiCmd *) arg;
printf("asc_timeout: cmd %p drive %d\n", scsicmd, scsicmd->sd->sd_drive);
#ifdef DEBUG
asc_DumpLog("asc_timeout");
#endif
#if 0
panic("asc_timeout");
#else
boot(4, NULL); /* XXX */
#endif
}
#ifdef DEBUG
void
asc_DumpLog(str)