aha1542: remove dead code
authorOndrej Zary <linux@rainbow-software.org>
Fri, 6 Feb 2015 22:11:23 +0000 (23:11 +0100)
committerJames Bottomley <JBottomley@Odin.com>
Fri, 10 Apr 2015 01:03:11 +0000 (18:03 -0700)
Signed-off-by: Ondrej Zary <linux@rainbow-software.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: James Bottomley <JBottomley@Odin.com>
drivers/scsi/aha1542.c
drivers/scsi/aha1542.h

index b7a62da..506e33e 100644 (file)
@@ -570,13 +570,6 @@ static int aha1542_queuecommand_lck(Scsi_Cmnd * SCpnt, void (*done) (Scsi_Cmnd *
 
        if (*cmd == REQUEST_SENSE) {
                /* Don't do the command - we have the sense data already */
-#if 0
-               /* scsi_request_sense() provides a buffer of size 256,
-                  so there is no reason to expect equality */
-               if (bufflen != SCSI_SENSE_BUFFERSIZE)
-                       printk(KERN_CRIT "aha1542: Wrong buffer length supplied "
-                              "for request sense (%d)\n", bufflen);
-#endif
                SCpnt->result = 0;
                done(SCpnt);
                return 0;
@@ -1095,24 +1088,14 @@ static int aha1542_restart(struct Scsi_Host *shost)
 {
        int i;
        int count = 0;
-#if 0
-       unchar ahacmd = CMD_START_SCSI;
-#endif
 
        for (i = 0; i < AHA1542_MAILBOXES; i++)
                if (HOSTDATA(shost)->SCint[i] &&
                    !(HOSTDATA(shost)->SCint[i]->device->soft_reset)) {
-#if 0
-                       HOSTDATA(shost)->mb[i].status = 1;      /* Indicate ready to restart... */
-#endif
                        count++;
                }
        printk(KERN_DEBUG "Potential to restart %d stalled commands...\n", count);
-#if 0
-       /* start scsi command */
-       if (count)
-               aha1542_out(shost->io_port, &ahacmd, 1);
-#endif
+
        return 0;
 }
 
@@ -1177,39 +1160,6 @@ static int aha1542_dev_reset(Scsi_Cmnd * SCpnt)
                "Trying device reset for target\n");
 
        return SUCCESS;
-
-
-#ifdef ERIC_neverdef
-       /* 
-        * With the 1542 we apparently never get an interrupt to
-        * acknowledge a device reset being sent.  Then again, Leonard
-        * says we are doing this wrong in the first place...
-        *
-        * Take a wait and see attitude.  If we get spurious interrupts,
-        * then the device reset is doing something sane and useful, and
-        * we will wait for the interrupt to post completion.
-        */
-       printk(KERN_WARNING "Sent BUS DEVICE RESET to target %d\n", SCpnt->target);
-
-       /*
-        * Free the command block for all commands running on this 
-        * target... 
-        */
-       for (i = 0; i < AHA1542_MAILBOXES; i++) {
-               if (HOSTDATA(SCpnt->host)->SCint[i] &&
-                   HOSTDATA(SCpnt->host)->SCint[i]->target == SCpnt->target) {
-                       Scsi_Cmnd *SCtmp;
-                       SCtmp = HOSTDATA(SCpnt->host)->SCint[i];
-                       kfree(SCtmp->host_scribble);
-                       SCtmp->host_scribble = NULL;
-                       HOSTDATA(SCpnt->host)->SCint[i] = NULL;
-                       HOSTDATA(SCpnt->host)->mb[i].status = 0;
-               }
-       }
-       return SUCCESS;
-
-       return FAILED;
-#endif                         /* ERIC_neverdef */
 }
 
 static int aha1542_bus_reset(Scsi_Cmnd * SCpnt)
@@ -1344,197 +1294,6 @@ fail:
        return FAILED;
 }
 
-#if 0
-/*
- * These are the old error handling routines.  They are only temporarily
- * here while we play with the new error handling code.
- */
-static int aha1542_old_abort(Scsi_Cmnd * SCpnt)
-{
-#if 0
-       unchar ahacmd = CMD_START_SCSI;
-       unsigned long flags;
-       struct mailbox *mb;
-       int mbi, mbo, i;
-
-       printk(KERN_DEBUG "In aha1542_abort: %x %x\n",
-              inb(STATUS(SCpnt->host->io_port)),
-              inb(INTRFLAGS(SCpnt->host->io_port)));
-
-       spin_lock_irqsave(&aha1542_lock, flags);
-       mb = HOSTDATA(SCpnt->host)->mb;
-       mbi = HOSTDATA(SCpnt->host)->aha1542_last_mbi_used + 1;
-       if (mbi >= 2 * AHA1542_MAILBOXES)
-               mbi = AHA1542_MAILBOXES;
-
-       do {
-               if (mb[mbi].status != 0)
-                       break;
-               mbi++;
-               if (mbi >= 2 * AHA1542_MAILBOXES)
-                       mbi = AHA1542_MAILBOXES;
-       } while (mbi != HOSTDATA(SCpnt->host)->aha1542_last_mbi_used);
-       spin_unlock_irqrestore(&aha1542_lock, flags);
-
-       if (mb[mbi].status) {
-               printk(KERN_ERR "Lost interrupt discovered on irq %d - attempting to recover\n",
-                      SCpnt->host->irq);
-               aha1542_intr_handle(SCpnt->host, NULL);
-               return 0;
-       }
-       /* OK, no lost interrupt.  Try looking to see how many pending commands
-          we think we have. */
-
-       for (i = 0; i < AHA1542_MAILBOXES; i++)
-               if (HOSTDATA(SCpnt->host)->SCint[i]) {
-                       if (HOSTDATA(SCpnt->host)->SCint[i] == SCpnt) {
-                               printk(KERN_ERR "Timed out command pending for %s\n",
-                                      SCpnt->request->rq_disk ?
-                                      SCpnt->request->rq_disk->disk_name : "?"
-                                      );
-                               if (HOSTDATA(SCpnt->host)->mb[i].status) {
-                                       printk(KERN_ERR "OGMB still full - restarting\n");
-                                       aha1542_out(SCpnt->host->io_port, &ahacmd, 1);
-                               };
-                       } else
-                               printk(KERN_ERR "Other pending command %s\n",
-                                      SCpnt->request->rq_disk ?
-                                      SCpnt->request->rq_disk->disk_name : "?"
-                                      );
-               }
-#endif
-
-       DEB(printk("aha1542_abort\n"));
-#if 0
-       spin_lock_irqsave(&aha1542_lock, flags);
-       for (mbo = 0; mbo < AHA1542_MAILBOXES; mbo++) {
-               if (SCpnt == HOSTDATA(SCpnt->host)->SCint[mbo]) {
-                       mb[mbo].status = 2;     /* Abort command */
-                       aha1542_out(SCpnt->host->io_port, &ahacmd, 1);  /* start scsi command */
-                       spin_unlock_irqrestore(&aha1542_lock, flags);
-                       break;
-               }
-       }
-       if (AHA1542_MAILBOXES == mbo)
-               spin_unlock_irqrestore(&aha1542_lock, flags);
-#endif
-       return SCSI_ABORT_SNOOZE;
-}
-
-/* We do not implement a reset function here, but the upper level code
-   assumes that it will get some kind of response for the command in
-   SCpnt.  We must oblige, or the command will hang the scsi system.
-   For a first go, we assume that the 1542 notifies us with all of the
-   pending commands (it does implement soft reset, after all). */
-
-static int aha1542_old_reset(Scsi_Cmnd * SCpnt, unsigned int reset_flags)
-{
-       unchar ahacmd = CMD_START_SCSI;
-       int i;
-
-       /*
-        * See if a bus reset was suggested.
-        */
-       if (reset_flags & SCSI_RESET_SUGGEST_BUS_RESET) {
-               /* 
-                * This does a scsi reset for all devices on the bus.
-                * In principle, we could also reset the 1542 - should
-                * we do this?  Try this first, and we can add that later
-                * if it turns out to be useful.
-                */
-               outb(HRST | SCRST, CONTROL(SCpnt->host->io_port));
-
-               /*
-                * Wait for the thing to settle down a bit.  Unfortunately
-                * this is going to basically lock up the machine while we
-                * wait for this to complete.  To be 100% correct, we need to
-                * check for timeout, and if we are doing something like this
-                * we are pretty desperate anyways.
-                */
-               WAIT(STATUS(SCpnt->host->io_port),
-               STATMASK, INIT | IDLE, STST | DIAGF | INVDCMD | DF | CDF);
-
-               /*
-                * We need to do this too before the 1542 can interact with
-                * us again.
-                */
-               setup_mailboxes(SCpnt->host->io_port, SCpnt->host);
-
-               /*
-                * Now try to pick up the pieces.  Restart all commands
-                * that are currently active on the bus, and reset all of
-                * the datastructures.  We have some time to kill while
-                * things settle down, so print a nice message.
-                */
-               printk(KERN_WARNING "Sent BUS RESET to scsi host %d\n", SCpnt->host->host_no);
-
-               for (i = 0; i < AHA1542_MAILBOXES; i++)
-                       if (HOSTDATA(SCpnt->host)->SCint[i] != NULL) {
-                               Scsi_Cmnd *SCtmp;
-                               SCtmp = HOSTDATA(SCpnt->host)->SCint[i];
-                               SCtmp->result = DID_RESET << 16;
-                               kfree(SCtmp->host_scribble);
-                               SCtmp->host_scribble = NULL;
-                               printk(KERN_WARNING "Sending DID_RESET for target %d\n", SCpnt->target);
-                               SCtmp->scsi_done(SCpnt);
-
-                               HOSTDATA(SCpnt->host)->SCint[i] = NULL;
-                               HOSTDATA(SCpnt->host)->mb[i].status = 0;
-                       }
-               /*
-                * Now tell the mid-level code what we did here.  Since
-                * we have restarted all of the outstanding commands,
-                * then report SUCCESS.
-                */
-               return (SCSI_RESET_SUCCESS | SCSI_RESET_BUS_RESET);
-fail:
-               printk(KERN_CRIT "aha1542.c: Unable to perform hard reset.\n");
-               printk(KERN_CRIT "Power cycle machine to reset\n");
-               return (SCSI_RESET_ERROR | SCSI_RESET_BUS_RESET);
-
-
-       } else {
-               /* This does a selective reset of just the one device */
-               /* First locate the ccb for this command */
-               for (i = 0; i < AHA1542_MAILBOXES; i++)
-                       if (HOSTDATA(SCpnt->host)->SCint[i] == SCpnt) {
-                               HOSTDATA(SCpnt->host)->ccb[i].op = 0x81;        /* BUS DEVICE RESET */
-                               /* Now tell the 1542 to flush all pending commands for this target */
-                               aha1542_out(SCpnt->host->io_port, &ahacmd, 1);
-
-                               /* Here is the tricky part.  What to do next.  Do we get an interrupt
-                                  for the commands that we aborted with the specified target, or
-                                  do we generate this on our own?  Try it without first and see
-                                  what happens */
-                               printk(KERN_WARNING "Sent BUS DEVICE RESET to target %d\n", SCpnt->target);
-
-                               /* If the first does not work, then try the second.  I think the
-                                  first option is more likely to be correct. Free the command
-                                  block for all commands running on this target... */
-                               for (i = 0; i < AHA1542_MAILBOXES; i++)
-                                       if (HOSTDATA(SCpnt->host)->SCint[i] &&
-                                           HOSTDATA(SCpnt->host)->SCint[i]->target == SCpnt->target) {
-                                               Scsi_Cmnd *SCtmp;
-                                               SCtmp = HOSTDATA(SCpnt->host)->SCint[i];
-                                               SCtmp->result = DID_RESET << 16;
-                                               kfree(SCtmp->host_scribble);
-                                               SCtmp->host_scribble = NULL;
-                                               printk(KERN_WARNING "Sending DID_RESET for target %d\n", SCpnt->target);
-                                               SCtmp->scsi_done(SCpnt);
-
-                                               HOSTDATA(SCpnt->host)->SCint[i] = NULL;
-                                               HOSTDATA(SCpnt->host)->mb[i].status = 0;
-                                       }
-                               return SCSI_RESET_SUCCESS;
-                       }
-       }
-       /* No active command at this time, so this means that each time we got
-          some kind of response the last time through.  Tell the mid-level code
-          to request sense information in order to decide what to do next. */
-       return SCSI_RESET_PUNT;
-}
-#endif    /* end of big comment block around old_abort + old_reset */
-
 static int aha1542_biosparam(struct scsi_device *sdev,
                struct block_device *bdev, sector_t capacity, int *ip)
 {
index 76e6abc..af91125 100644 (file)
@@ -100,12 +100,6 @@ static inline void any2scsi(u8 *p, u32 v)
 
 #define scsi2int(up) ( (((long)*(up)) << 16) + (((long)(up)[1]) << 8) + ((long)(up)[2]) )
 
-#define xany2scsi(up, p)       \
-(up)[0] = ((long)(p)) >> 24;   \
-(up)[1] = ((long)(p)) >> 16;   \
-(up)[2] = ((long)(p)) >> 8;    \
-(up)[3] = ((long)(p));
-
 #define xscsi2int(up) ( (((long)(up)[0]) << 24) + (((long)(up)[1]) << 16) \
                      + (((long)(up)[2]) <<  8) +  ((long)(up)[3]) )