This patch addresses two distinct races:

- Until now, the driver started the chip for Tx regardless of errors
  pending in the status register. Not good if an error occured while
  we were queueing packets -- the chip counter had not been reset,
  so Tx died. (We can't reliably get an interrupt for every error
  condition)

- The Rhine-II (when under load) frequently produces a Tx descriptor
  write-back race error. Failing to handle this means waiting for the
  netdev watchdog. Fixed.

  In addition, we must wait for the Tx engine to turn off on error
  conditions before we scavenge the descriptor entries. Failing to do
  so will typically lead to performance going down to about 10%: Burst,
  timeout, burst, timeout.. (again, with a Rhine-II under load).

Since Linus and Jeff raised the issue of PCI posted writes, I cleaned up
wait_for_reset() some more. Experiments show that with MMIO, a reset may
indeed take seemingly longer -- that is fixed by flushing that buffer.

Also, the driver now polls the appropriate register while waiting for the
reset to finish.

Closes bugzilla #307

It turns out that this bug is only on the pre-production models,
but as a few of them have 'escaped' the labs to the hands of
a few lucky developers, we still need it.

Uses CONFIG_ but doesn't include config.h

As per bugzilla #179

ENTRY implies ALIGN.

was this actually ever used? :)

This patch fixes bugzilla bug #398,
http://bugme.osdl.org/show_bug.cgi?id=398, PNP API breakage in the
ad1848 sound driver.

Fix problem with devfs when unloading the dm module.

dm-ioctl.c: deregister the misc device, and its associated symlink
*before* removing the /dev/mapper dir.  [Alasdair Kergon]

Use the correct size for "name" in register_with_devfs().

During Al Viro's devfs cleanup a few versions ago, this function was
rewritten, and the "name" string added. The 32-byte size is not large
enough to prevent a possible buffer overflow in the sprintf() call,
since the hash cell can have a name up to 128 characters.

Return correct error codes from dm_table_add_target().  [Kevin Corry]

Another fix for the __LOW macro.

When dm_table and dm_target structures are initialized, the "limits" fields
(struct io_restrictions) are initialized to zero (e.g. in dm_table_add_target()
in dm-table.c). However, zero is not a useable value in these fields. The
request queue will never let an I/O through, regardless of how small it might
be, if max_sectors is set to zero (see generic_make_request in ll_rw_blk.c).
This change to the __LOW() macro sets these fields correctly when they are
first initialized.  [Kevin Corry]

Allow slashes ('/') within a DM device name, but not at the beginning.

Devfs will automatically create all necessary sub-directories if a name
with embedded slashes is registered.  [Kevin Corry]

Silly mistake in error path when an unknown target type is requested.

Fix __LOW macro.  [Kevin Corry]

When reloading a device the ioctl interface was forgetting to drop a
reference on the new table.

into home.transmeta.com:/home/torvalds/v2.5/linux

In 2.5.63 I get the following compile error in drivers/scsi/sym53c416.c:


<--  snip  -->

...
  gcc -Wp,-MD,drivers/scsi/.sym53c416.o.d -D__KERNEL__ -Iinclude -Wall
-Wstrict-prototypes -Wno-trigraphs -O2 -fno-strict-aliasing -fno-common
-pipe -mpreferred-stack-boundary=2 -march=k6
-Iinclude/asm-i386/mach-default -nostdinc -iwithprefix include
-DKBUILD_BASENAME=sym53c416 -DKBUILD_MODNAME=sym53c416 -c -o
drivers/scsi/sym53c416.o drivers/scsi/sym53c416.c
drivers/scsi/sym53c416.c: In function `sym53c416_detect':
drivers/scsi/sym53c416.c:682: too many arguments to function
`pnp_activate_dev'
make[2]: *** [drivers/scsi/sym53c416.o] Error 1

<--  snip  -->


The following patch fixes it:

In 2.5.63 I get the following compile error in drivers/scsi/g_NCR5380.c:

<--  snip  -->

...
  gcc -Wp,-MD,drivers/scsi/.g_NCR5380.o.d -D__KERNEL__ -Iinclude -Wall
-Wstrict-prototypes -Wno-trigraphs -O2 -fno-strict-aliasing -fno-common
-pipe -mpreferred-stack-boundary=2 -march=k6
-Iinclude/asm-i386/mach-default -nostdinc -iwithprefix include
-DKBUILD_BASENAME=g_NCR5380 -DKBUILD_MODNAME=g_NCR5380 -c -o
drivers/scsi/g_NCR5380.o drivers/scsi/g_NCR5380.c
drivers/scsi/g_NCR5380.c: In function `generic_NCR5380_detect':
drivers/scsi/g_NCR5380.c:326: too many arguments to function
`pnp_activate_dev'
...
make[2]: *** [drivers/scsi/g_NCR5380.o] Error 1

<--  snip  -->


The following patch fixes it:

Patrick pointed out that I could get into a ABBA issue with holding
list_lock while calling scsi_eh_scmd_add which takes host_lock.
A call to scsi_prep_fn already holds the queue_lock and then makes a
call to scsi_get_command which takes list_lock.

-andmike
--
Michael Anderson
andmike@us.ibm.com

=====
name:		00_scsi_set_device_offline.diff
version:	2003-02-26.15:18:19-0800
against:	scsi-misc-2.5

 scsi.c |   15 +++++++++------
 1 files changed, 9 insertions(+), 6 deletions(-)

=====

into home.transmeta.com:/home/torvalds/v2.5/linux

Against 2.5.63, untested

regards
john

On Wed, Feb 26, 2003 at 08:20:05PM +0100, Christoph Hellwig wrote:
> This patch updates nsp_cs to use scsi_add_host / scsi_remove_host when
> compiledfor Linux 2.5 which allows to get rid of the scsi host list
> walks and is a preparation for allowing to support multiple cards of
> this type.

Sorry, I sent you the wrong paetch, here's the right one:

Against 2.5.63, untested

regards
john

Against 2.5.63, untested

regards
john

Untested, against 2.5.63

regards
john

Untested, 2.5.63

I remember having this submitted a while ago, but here's the code
again, this time with the untested aic79xx bits.