Problem reported/corrected by Michael Reed [mdr@sgi.com]:

  We have a 2312 that puts the 8.00.00b12-k qla2xxx driver
  into an infinite loop with a solid interrupt, 0x8008,
  hcsr = 0x7430, risc status register 0x40008110,
  immediately after enabling parity on the board.

  It looks as though the early out test in
  qla_isr.c:qla2x00_intr_handler() against stat is broken.
  HSR_RISC_PAUSED is set.  Also, there's a stale mailbox
  completion flagged (stat&0xff) which will short circuit
  the default case in the switch if it got that far.
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

  Restructure qla2x00_start_scsi() so that PCI mappings are
  done after we've verified command list and request queue
  resource availability.
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

On Tuesday, June 22, 2004 12:07 AM, Arjan van de Ven wrote:
> On Tue, 2004-06-22 at 07:49, Andrew Vasquez wrote:
>
> > diff -Nru a/drivers/scsi/qla2xxx/qla_iocb.c
> b/drivers/scsi/qla2xxx/qla_iocb.c
> > --- a/drivers/scsi/qla2xxx/qla_iocb.c	2004-06-21 15:36:37 -07:00
> > +++ b/drivers/scsi/qla2xxx/qla_iocb.c	2004-06-21 15:36:37 -07:00 @@
> >  	-43,7 +43,7 @@ else if (cmd->sc_data_direction ==
> >  	DMA_FROM_DEVICE)  		cflags = CF_READ; else {
> > -		switch (cmd->data_cmnd[0]) {
> > +		switch (cmd->cmnd[0]) {
> >  		case WRITE_6:
> >  		case WRITE_10:
> >  		case WRITE_12:
>
> this sounds wrong.
>

Yes, it is _wrong_!   This code is baggage left behind from earlier
drivers when we were 'asked' to work around certain applications'
inability to properly construct commands.

> Is there any reason qlogic drivers can't use/trust the midlayer
> instead of doing qla2x00_get_cmd_direction() to reverse engineer to
> direction ??

Sure.  Especially considering an earlier thread:

	http://marc.theaimsgroup.com/?t=108727983000002&r=1&w=2Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

  Problem reported/corrected by Tony Battersby [tonyb@cybernetics.com]:

  o The qla2xxx driver calculates data transfer residuals
    for some but not all circumstances.  The patch below
    adds data transfer residual calculations for the
    remaining cases.

  o Interrogate the proper cmnd[] array when determining
    the transfer direction.
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

  Add memory barriers to ensure that all load operations have
  completed before the (MMIO) write to the ISP's registers.
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

While reviewing tmscsim, I noticed something, I didn't quite like /
understand. The driver takes the host_lock (with irqsave) at the entry to
the ISR, and releases it at the exit. And inside the ISR there are
potentially long busy-waits... Like

 	int ctr = 6000000; /* only try for about a second */
 	while( --ctr && !((dstate = DC390_read8 (DMA_Status)) &
 				DMA_XFER_DONE) && pSRB->SGToBeXferLen );

The attached patch is attempting to fix those places. Not sure if this is
a proper fix though. In my understanding, after looking through the code,
the host_lock is used to protect host-specific data and host-registers.
The ->queuecommand is already called with it help, so, one just,
basically, have to protect other contexts - interrupt, timer,... So, looks
mostly right.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

On Sun, 23 May 2004, Christoph Hellwig wrote:
> these leaks.  Maybe also  merge dc390_init and dc390_init_one?
>
> Similarly I think DC390_release should be merged into dc390_remove_one.

Attached. Also fixed some __init / __devinit and __exit / __devexit
attributes. Although, would be good to have something like

#ifdef CONFIG_HOTPLUG_PCI
#define __pcidevinit
#define __pcidevinitdata
#define __pcidevexit
#define __pcidevexitdata
#else
#define __pcidevinit __init
#define __pcidevinitdata __initdata
#define __pcidevexit __exit
#define __pcidevexitdata __exitdata
#endif
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Allows 32 bit apps to use ioctls in a 64 bit kernel.

Signed-off-by Mark Haverkamp <markh@osdl.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

 + minor cleanups
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

avoid obsolete APIs
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Avoid old APIs and fix a bad bug were the irq handler argument was
derefenced as struct Scsi_Host * instead of Adpater * which made the
driver instantly crash on SMP.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

replace obsolete typedefs and scsi_to_pci_dma_dir, use proper includes
and re-order includes correctly.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

From: Joel Soete <soete.joel@tiscali.be>

This patch makes the ncr driver capable of accepting segments which
are larger than the block size.  It has been tested against IOMMU
merging on parisc.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

From: Matthew Wilcox <willy@debian.org>

- remove all of the PCI chips (8xx series) leaving only 720
- pull out stand alone initialisation code (driver becomes a chip only
  driver for NCR Q720 and Zalon)
- remove most vestiges of PCI support (chip driver should be bus neutral)
- tidy up typedefs
- remove simulated intfly (may need putting back for 770)
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

the patch below finishes off the removal of the obsolete hosts.h usage, and
now that the users are gone, it's safe to turn on the warning to
prevent accidental future use... Please apply
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Some SCSI-IDE cdrom drives take up to 15 seconds to respond following
an abort being issued to them. This patch changes ipr to only send
a cancel all to a device as part of request sense processing when
the device is running tagged command queueing.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

The following patch fixes a hang that occurs when an abort
is issued by the midlayer. The hang is the result of overriding
a field in the union of the ipr_cmd struct. It is fixed by moving
this pointer outside of the union, since it is not mutually exclusive
with the other fields in the union.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

into mulgrave.(none):/home/jejb/BK/scsi-misc-2.6

The BLIST_REPORTLUN2 blacklist item has exactly the behavior that
the MYLEX DACARMRB (and SGI TP9100, which is really a 2Gb upgrade
to the DACARMRB) need.

These devices use a PQ of 1 instead of 3 for unconfigured luns,
which means that every RAID has 32 luns, many or most of which
are just phantoms.

However, it does support Report Luns, and it returns the correct
data.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

The object of this infrastructure is to give HBAs early warning that
error handling is about to happen and also provide them with the
opportunity to do something about it.

It introduces the extra template callback:

eh_timed_out()

which scsi_times_out() will call if it is populated to notify the LLD
that an outstanding command took a timeout.

There are three possible returns:

EH_HANDLED:	I've fixed the problem, please complete the command for me
(as soon as the timer fires, scsi_done will do nothing, so the timer
itself will call a special version of scsi_done that doesn't check the
timer).

EH_NOT_HANDLED:	Invoke error recovery as normal

EH_RESET_TIMER:	The command will complete, reset the timer to its
original value and start it ticking again.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Based on work by Christoph Hellwig and Luben Tuikov

I've been going through Mark's changes with a fine toothcomb and this merges
most of them. Its tested on 64bit SMP hardware and seems to be fine. There
are a couple of Mark's changes I've left out for now but there isnt really
an easy way to break down the changes further.

This fixes a whole host of problems including random hangs under high load
Signed-off-by: Alan Cox <alan@redhat.com>
        
Original contribution under GPL from Adaptec, updates checking by Red Hat
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

into ppc970.osdl.org:/home/torvalds/v2.6/linux

This is a fix for when we encounter an alias during insert. When that
happens we move the request to dispatch, but it may happen that this
request is also the ->last_merge hint. So we may attempt to merge with
this later, when it's either in progress or already freed. Rearrange the
logic a bit so we clear the merge hint there as well. It looks more
complex than it is, the only real code change is the addition of a
cfq_remove_merge_hints() in cfq_dispatch_sort().
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

It's possible under unlucky circumstances for this race to trigger. I
described it with a comment in the code.
Signed-Off-By: Jens Axboe <axboe@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

This makes the CFQ tunables available in sysfs, like AS and deadline.
Signed-off-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Noted by Alexey Dobriyan.

into ppc970.osdl.org:/home/torvalds/v2.6/linux

This cset adds the code to handle the hardware vector floating point
unit found on some ARM926 and later CPUs.  The hardware provides
an implementation for the common cases, and bounces exceptions for
other cases, which have to be handled in software, and signalling
SIGFPE as appropriate.

Remove unused queued_signals global accounting.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Add a user_struct pointer to the sigqueue structure.  Charge sigqueue
allocation and destruction to the user_struct rather than a global pool.  This
per user rlimit accounting obsoletes the global queued_signals accouting.

The patch as charges the sigqueue struct allocation to the queue that it's
pending on (the receiver of the signal).  So the owner of the queue is charged
for whoever writes to it (much like quota for a 777 file).

The patch started out charging the task which allocated the sigqueue struct.
In most cases, these are always the same user (permission for sending a
signal), so those cases are moot.  In the cases where it isn't the same user,
it's a privileged user sending a signal to another user.

It seems wrong to charge the allocation to the privleged user, when the other
user could block receipt as long as it feels.  The flipside is, someone else
can fill your queue (expectation is that someone else is privileged).  I think
it's right the way it is.  The change to revert is very small.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Update send_signal() api to allow passing the task receiving the signal.  This
is necessary to ensure signals generated out of process context can be charged
to the correct user.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>