nosy_start/stop_snoop() are always only called by the ioctl method, i.e.
with IRQs enabled.  packet_handler() and bus_reset_handler() are always
only called by the IRQ handler.  Hence neither one needs to track IRQ
flags.

To underline the call context of packet_handler() and
bus_reset_handler(), rename these functions to *_irq_handler().
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

nosy_stop_snoop() would blow up the second time it was called without
nosy_start_snoop() in between.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

The required serialization of NOSY_IOC_START and NOSY_IOC_STOP is
already provided by the client_list_lock.

NOSY_IOC_FILTER does not really require serialization since accesses
to tcode_mask are atomic on any sane CPU architecture.  Nevertheless,
make it explicit that we want this to be atomic by means of
client_list_lock (which also surrounds the other tcode_mask access in
the IRQ handler).  While we are at it, change the type of tcode_mask to
u32 for consistency with the user API.

NOSY_IOC_GET_STATS does not require serialization against itself.  But
there is a bug here regarding concurrent updates of the two counters
by the IRQ handler.  Fix it by taking the client_list_lock in this ioctl
too.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Extend copyright note to 2007, c.f. Kristian's git log.

Includes:
  - replace some <asm/*.h> by <linux/*.h>
  - add required indirectly included <linux/spinlock.h>
  - order alphabetically

Coding style related changes:
  - change to utf8
  - normalize whitespace
  - normalize comment style
  - remove usages of __FUNCTION__
  - remove an unnecessary cast from void *

Const and static declarations:
  - driver_name is not const in pci_driver.name, drop const qualifier
  - driver_name can be taken from KBUILD_MODNAME
  - the global variable minors[] can and should be static
  - constify struct file_operations instance

Data types:
  - Remove unused struct member struct packet.code.  struct packet is
    only used for driver-internal bookkeeping; it does not appear on the
    wire or in DMA programs or the userspace ABI.  Hence the unused
    member .code can be removed without worries.

Preprocessor macros:
  - unroll a preprocessor macro that containd a return
  - use list_for_each_entry

Printk:
  - add missing terminating \n in some format strings
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

This adds the traffic sniffer driver for Texas Instruments PCILynx/
PCILynx2 based cards.  The use cases for nosy are analysis of
nonstandard protocols and as an aid in development of drivers,
applications, or firmwares.

Author of the driver is Kristian Høgsberg.  Known contributers are
Jody McIntyre and Jonathan Woithe.

Nosy programs PCILynx chips to operate in promiscuous mode, which is a
feature that is not found in OHCI-1394 controllers.  Hence, only special
hardware as mentioned in the Kconfig help text is suitable for nosy.

This is only the kernelspace part of nosy.  There is a userspace
interface to it, called nosy-dump, proposed to be added into the tools/
subdirectory of the kernel sources in a subsequent change.  Kernelspace
and userspave component of nosy communicate via a 'misc' character
device file called /dev/nosy with a simple ioctl() and read() based
protocol, as described by nosy-user.h.

The files added here are taken from
git://anongit.freedesktop.org/~krh/nosy commit ee29be97 (2009-11-10)
with the following changes by Stefan Richter:
  - Kconfig and Makefile hunks are written from scratch.
  - Commented out version printk in nosy.c.
  - Included missing <linux/sched.h>, reported by Stephen Rothwell.

"git shortlog nosy{-user.h,.c,.h}" from nosy's git repository:

Jonathan Woithe (2):
      Nosy updates for recent kernels
      Fix uninitialised memory (needed for 2.6.31 kernel)

Kristian Høgsberg (5):
      Pull over nosy from mercurial repo.
      Use a misc device instead.
      Add simple AV/C decoder.
      Don't break down on big payloads.
      Set parent device for misc device.

As a low-level IEEE 1394 driver, its files are placed into
drivers/firewire/ although nosy is not part of the firewire driver
stack.

I am aware of the following literature from Texas Instruments about
PCILynx programming:
      SCPA020A - PCILynx 1394 to PCI Bus Interface TSB12LV21BPGF
                 Functional Specification
      SLLA023  - Initialization and Asynchronous Programming of the
                 TSB12LV21A 1394 Device
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>
Acked-by: NKristian Høgsberg <krh@bitplanet.net>

Add a loop to wait for the controller to finish a locally-initiated CSR
lock operation.  Google shows some occurrences of the "swap not done
yet" message which might indicate that some OHCI controllers are not
fast enough to do the lock/swap in the time needed for one PCI access.

This also correctly handles the case where the lock operation did not
finish, instead of silently returning an uninitialized value.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

We must compute the offset from the CSR register base with the
full 48 address bits to prevent matching with addresses whose
lower 32 bits happen to be equal with one of the specially
handled registers.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Returning -EIO for all errors would not allow clients to determine if
the resource allocation process itself failed, or if the resources are
not available.  (The latter information is needed by CMP to synchronize
restoring of overlayed connections after a bus reset.)
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

If there is a permanent error condition when communicating with the IRM,
after the sixth error, the retry variable will be decremented to -1.
If, in this case, the bits in channels_mask are not yet exhausted, the
next channel is retried 2^32 times.

To fix this, check that retry is never decremented beyond zero.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

A userspace client got to see uninitialized stack-allocated memory if it
specified an _IOC_READ type of ioctl and an argument size larger than
expected by firewire-core's ioctl handlers (but not larger than the
core's union ioctl_arg).

Fix this by clearing the requested buffer size to zero, but only at _IOR
ioctls.  This way, there is almost no runtime penalty to legitimate
ioctls.  The only legitimate _IOR is FW_CDEV_IOC_GET_CYCLE_TIMER with 12
or 16 bytes to memset.

[Another way to fix this would be strict checking of argument size (and
possibly direction) vs. command number.  However, we then need a lookup
table, and we need to allow for slight size deviations in case of 32bit
userland on 64bit kernel.]
Reported-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

The definition of struct fw_cdev_iso_packet seems to imply that the
header_length must be quadlet-aligned, and in fact, specifying an
unaligned header has never really worked when using multiple packet
structures, because the position of the next control word is computed by
rounding the header_length _down_, so the last one to three bytes of the
header would overlap the next control word.

To avoid this problem, check that the header length is properly aligned.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

In receive contexts, reject packets with header_length==0.  This would
be an instruction to queue zero packets which would not make sense.

This prevents a division by zero in the OHCI driver.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h

percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files.  percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.

percpu.h -> slab.h dependency is about to be removed.  Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability.  As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.

  http://userweb.kernel.org/~tj/misc/slabh-sweep.py

The script does the followings.

* Scan files for gfp and slab usages and update includes such that
  only the necessary includes are there.  ie. if only gfp is used,
  gfp.h, if slab is used, slab.h.

* When the script inserts a new include, it looks at the include
  blocks and try to put the new include such that its order conforms
  to its surrounding.  It's put in the include block which contains
  core kernel includes, in the same order that the rest are ordered -
  alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
  doesn't seem to be any matching order.

* If the script can't find a place to put a new include (mostly
  because the file doesn't have fitting include block), it prints out
  an error message indicating which .h file needs to be added to the
  file.

The conversion was done in the following steps.

1. The initial automatic conversion of all .c files updated slightly
   over 4000 files, deleting around 700 includes and adding ~480 gfp.h
   and ~3000 slab.h inclusions.  The script emitted errors for ~400
   files.

2. Each error was manually checked.  Some didn't need the inclusion,
   some needed manual addition while adding it to implementation .h or
   embedding .c file was more appropriate for others.  This step added
   inclusions to around 150 files.

3. The script was run again and the output was compared to the edits
   from #2 to make sure no file was left behind.

4. Several build tests were done and a couple of problems were fixed.
   e.g. lib/decompress_*.c used malloc/free() wrappers around slab
   APIs requiring slab.h to be added manually.

5. The script was run on all .h files but without automatically
   editing them as sprinkling gfp.h and slab.h inclusions around .h
   files could easily lead to inclusion dependency hell.  Most gfp.h
   inclusion directives were ignored as stuff from gfp.h was usually
   wildly available and often used in preprocessor macros.  Each
   slab.h inclusion directive was examined and added manually as
   necessary.

6. percpu.h was updated not to include slab.h.

7. Build test were done on the following configurations and failures
   were fixed.  CONFIG_GCOV_KERNEL was turned off for all tests (as my
   distributed build env didn't work with gcov compiles) and a few
   more options had to be turned off depending on archs to make things
   build (like ipr on powerpc/64 which failed due to missing writeq).

   * x86 and x86_64 UP and SMP allmodconfig and a custom test config.
   * powerpc and powerpc64 SMP allmodconfig
   * sparc and sparc64 SMP allmodconfig
   * ia64 SMP allmodconfig
   * s390 SMP allmodconfig
   * alpha SMP allmodconfig
   * um on x86_64 SMP allmodconfig

8. percpu.h modifications were reverted so that it could be applied as
   a separate patch and serve as bisection point.

Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: NTejun Heo <tj@kernel.org>
Guess-its-ok-by: NChristoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>

The driver match strategy was:
  - Match vendor/model/specifier/version of the unit directory.
  - If that was a miss, match vendor from the root directory and
    model/specifier/version of the unit directory.

This was inconsistent with how the modalias string was constructed
until recently (take vendor/model from root directory and specifier/
version from unit directory).  It was also inconsistent with how it is
done since the parent commit:
  - Use vendor/model/specifier/version of the unit directory if possible,
  - fall back to one or more of vendor/model/specifier/version from the
    root directory depending on which ones are not present at the unit
    directory.

Fix this inconsistency by sharing the ROM scanner function between
modalias printer function and driver match function.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

The modalias string of devices that represent units on a FireWire node
did not show Module_ID entries within unit directories.  This was
because firewire-core searched only the root directory of the
configuration ROM for a Model_ID entry.

We now search first the root directory, then the unit directory.  IOW
honor a unit directory's Model_ID if present, otherwise fall back to the
root directory's model ID (if present).

Furthermore, apply the same change to Vendor_ID.  This had the same
issue but it was less apparent because most devices provide Vendor_ID
only in the root directory.

And finally, also use this strategy for the remaining two IDs in the
modalias, Specifier_ID and Version.  It does not actually make sense to
look for them elsewhere than in the unit directory because they are
mandatory there.  However, a uniform search order simplifies the
implementation and has no adverse affect in practice.

Side notes:
  - The older counterpart of this, nodemgr.c of ieee1394, looked for
    Vendor_ID first in the root directory, then in the unit directory,
    and for Model_ID only in the unit directory.
  - There is a single mainline driver which requires Vendor_ID and
    Model_ID --- the firedtv driver.  This one worked because FireDTVs
    provide Vendor_ID in the root directory and Model_ID identically in
    root directory and unit directory.
  - Apart from firedtv, there are currently no drivers known to me
    (including userspace drivers) that look at the Vendor_ID or Model_ID
    of the modalias.
Reported-by: NMaciej Żenczykowski <zenczykowski@gmail.com>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Among the many entries in the TSB12LV22 errata list (TI literature
number SLLS312) is the following:

  PCI Slave reads of the Cycle Timer register may occasionally get an
  incorrect value.
  Software may be able to validate value by reading the register
  multiple times rapidly and evaluating for a reasonable difference.

Signed-off-by: Clemens Ladisch <clemens@ladisch.de> (untested)
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> (added #define)

If the bandwidth allocation fails, the error must be returned in
*channel regardless of whether the channel allocation succeeded.
Checking for c >= 0 is not correct if no channel allocation was
requested, in which case this part of the code is reached with
c == -EINVAL.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

In the future, we are going to be changing the lock type for struct
device (once we get the lockdep infrastructure properly worked out)  To
make that changeover easier, and to possibly burry the lock in a
different part of struct device, let's create some functions to lock and
unlock a device so that no out-of-core code needs to be changed in the
future.

This patch creates the device_lock/unlock/trylock() functions, and
converts all in-tree users to them.

Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Jean Delvare <khali@linux-fr.org>
Cc: Dave Young <hidave.darkstar@gmail.com>
Cc: Ming Lei <tom.leiming@gmail.com>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Phil Carmody <ext-phil.2.carmody@nokia.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Cornelia Huck <cornelia.huck@de.ibm.com>
Cc: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Len Brown <len.brown@intel.com>
Cc: Magnus Damm <damm@igel.co.jp>
Cc: Alan Stern <stern@rowland.harvard.edu>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Cc: Stefan Richter <stefanr@s5r6.in-berlin.de>
Cc: David Brownell <dbrownell@users.sourceforge.net>
Cc: Vegard Nossum <vegard.nossum@gmail.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: Alex Chiang <achiang@hp.com>
Cc: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andrew Patterson <andrew.patterson@hp.com>
Cc: Yu Zhao <yu.zhao@intel.com>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Samuel Ortiz <sameo@linux.intel.com>
Cc: Wolfram Sang <w.sang@pengutronix.de>
Cc: CHENG Renquan <rqcheng@smu.edu.sg>
Cc: Oliver Neukum <oliver@neukum.org>
Cc: Frans Pop <elendil@planet.nl>
Cc: David Vrabel <david.vrabel@csr.com>
Cc: Kay Sievers <kay.sievers@vrfy.org>
Cc: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@suse.de>

The block layer calling convention is blk_queue_<limit name>.
blk_queue_max_sectors predates this practice, leading to some confusion.
Rename the function to appropriately reflect that its intended use is to
set max_hw_sectors.

Also introduce a temporary wrapper for backwards compability.  This can
be removed after the merge window is closed.
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

by the number of available isochronous DMA contexts and active quirks
which is occasionally useful information.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

This bug was present in firewire-ohci since day one:  The number of
available isochronous receive DMA contexts was mixed up with that of
available isochronous transmit DMA contexts.

This is harmless on a few chips which offer the same number of contexts
in both directions, but most chips nowadays implement only the standard
minimum of 4 IR contexts, but 8 IT contexts.  If a user attempted to run
a lot of IR contexts at once, results with more than four were therefore
unpredictable.  I suppose the controller would simply refuse to start
DMA of any unimplemented context.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

This way, we can advise users of precompiled kernel packages to test
existing quirk fixes on chips which have not been listed yet, without
them having to build a kernel from source.

Note, to use this feature on a machine with more than one controller,
steps like these are necessary:
# lspci | grep 1394
# ls /sys/bus/pci/drivers/firewire_ohci/
# echo -n "0000:03:02.0" > /sys/bus/pci/drivers/firewire_ohci/unbind
# echo 2 > /sys/module/firewire_ohci/parameters/quirks
# echo -n "0000:03:02.0" > /sys/bus/pci/drivers/firewire_ohci/bind
# echo 0 > /sys/module/firewire_ohci/parameters/quirks

The parameter can also be used to switch off quirk flags that were
hardwired into firewire-ohci's quirks table.  Simply specify a non-zero
quirks value but without any known flags, e.g. 0x100.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

We don't have a lot of quirks to take into account (especially since
dual-buffer IR is out of the picture), but still, a table-based approach
is more organized than a series of if () clauses.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

The config_rom struct members are only accessed during relatively
infrequent self-ID-complete interrupts and only if the local config ROM
was changed, while the ar_, at_, ir_, it_ members are used very
frequently during I/O.  Hence move the config_rom members further down.

More importantly, make the huge self_id_buffer member the last one; this
is only accessed in self-ID-complete interrupts.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

This code was no longer used since 2.6.33, "firewire: ohci: always use
packet-per-buffer mode for isochronous reception" commit 090699c0.  If
anybody needs this code in the future for special purposes, it can be
brought back in.  But it must not be re-enabled by default; drivers
(kernelspace or userspace drivers) should only get this mode if they
explicitly request it.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

from array of char to union of structs.  I already used a union to size
the buffer which holds ioctl arguments; more consequent is to define it
as an instance of this union in the first place.

Also rename several local variables from "request" to "a"(rgument) since
the term request can be mistaken to mean a transaction subaction, e.g.
an instance of struct fw_request.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

The system time from CLOCK_REALTIME is not monotonic, hence problematic
for the main user of the FW_CDEV_IOC_GET_CYCLE_TIMER ioctl.  This issue
exists in its successor ABI, i.e. raw1394, too.
http://subversion.ffado.org/ticket/242

We now offer an alternative ioctl which lets the caller choose between
CLOCK_REALTIME, CLOCK_MONOTONIC, and CLOCK_MONOTONIC_RAW as source of
the local time, very similar to the clock_gettime libc function.  The
format of the local time return value matches that of clock_gettime
(seconds and nanoseconds, instead of a single microseconds value from
the existing ioctl).
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

according to what it really does.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

If a device exposes a sparsely populated configuration ROM,
firewire-core's sysfs interface and character device file interface
showed random data in the gaps between config ROM blocks.  Fix this by
zero-initialization of the config ROM reader's scratch buffer.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

The stack size of 16 was artificially chosen and may be too small in
extreme cases.  A device won't be accessible then.

Since it doesn't really matter to the slab allocator whether we ask for
1088 bytes or 2048 bytes of scratch memory, just allocate 2048 bytes for
the sum of temporary config ROM image and stack, and we will never ever
overflow the stack (because there simply can't be more stack items than
ROM entries).
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

It never happened yet, but better safe than sorry:  If a device's config
ROM contains a block which overlaps the boundary at 0xfffff00007ff, just
ignore that one block instead of refusing to add the device
representation.  That way, upper layers (kernelspace or userspace
drivers) might still be able to use the device to some degree.

That's better than total inaccessibility of the device.  Worse, the core
would have logged only a generic "giving up on config rom" message which
could only be debugged by feeding a firewire-ohci debug logging session
through a config ROM interpreter, IOW would likely remain undiagnosed.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

The Panasonic AG-DV2500 tape deck contains an invalid entry in its
configuration ROM root directory:  A leaf pointer with the undefined key
ID 0 and an offset that points way out of the standard config ROM area.
This caused firewire-core to dismiss the device with the generic log
message "giving up on config rom for node id...", after which it was of
course impossible to access the tape deck with dvgrab or any other
program.  https://bugzilla.redhat.com/show_bug.cgi?id=449252#c29

The fix is to simply ignore this invalid ROM entry and proceed to read
the valid rest of the ROM.  There is a catch though:  When the kernel
later iterates over the ROM, it would be nasty having to check again for
such too large ROM offsets.  Therefore we manipulate the defective or
unsupported ROM entry to become a harmless immediate entry that won't
have any side effects later (an entry with the value 0x00000000).

Reported-by: George Chriss
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

The current implementation of Bus_Time read access was buggy since it
did not ensure that Bus_Time.second_count_hi and second_count_lo came
from the same 128 seconds period.
Reported-by: NHåkan Johansson <f96hajo@chalmers.se>

Instead of a fix, remove Bus_Time register support altogether.  The spec
requires all cycle master capable nodes to implement this (all Linux
nodes are cycle master capable) while it also says that it "may" be
initialized by the bus manager or by the IRM standing in for a bus
manager.  (Neither Linux' firewire-core nor ieee1394 nodemgr implement
this.)

Since we cannot rely on Bus_Time having been initialized by a bus
manager, it is better to return an error instead of a nonsensical value
on a read request to Bus_Time.

Alternatively, we could fix the Bus_Time read integrity bug _and_
implement (a) cycle master's write support of the register as well as
(b) bus manager's Bus_Time initialization service, i.e. preservation of
the Bus_Time when the cycle master node of a bus changes.  However, that
would be quite some code for a feature that is unreliable to begin with
and very likely unused in practice.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

ohci:  Break out of the retry loop if too many attempts were necessary.
This may theoretically happen if the chip is fatally defective or if the
get_cycle_timer ioctl was performed after a CardBus controller was
ejected.

Also micro-optimize the loop by re-using the last two register reads in
the next iteration, remove a questionable inline keyword, and shuffle a
comment around.

core:  ioctl_get_cycle_timer() is always called with interrupts on,
therefore local_irq_save() can be replaced by local_irq_disable().
Disabled local IRQs imply disabled preemption, hence preempt_disable()
can be removed.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Discussed in "read_cycle_timer backwards for sub-cycle 0000, 0001",
http://thread.gmane.org/gmane.linux.kernel.firewire.devel/13704

Known bad controllers:
  ALi M5271, listed by lspci as M5253 [10b9:5253]
  NEC OrangeLink [1033:00cd] (rev 03)
  NEC uPD72874 [1033:00f2] (rev 01)
  VIA VT6306 [1106:3044] (rev 46)
  VIA VT6308P, listed by lspci as rev c0
Reported-by: NPieter Palmers <pieterp@joow.be>
Reported-by: NHåkan Johansson <f96hajo@chalmers.se>
Reported-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

VIA controllers sometimes return an inconsistent value when reading the
isochronous cycle timer register.  To work around this, read the
register multiple times and add consistency checks.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Reported-by: NPieter Palmers <pieterp@joow.be>
Reported-by: NHåkan Johansson <f96hajo@chalmers.se>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

In isochronous transmit DMA descriptors, link the skip address pointer
back to the descriptor itself.  When a cycle is lost, the controller
will send the packet in the next cycle, instead of terminating the
entire DMA program.

There are two reasons for this:

* This behaviour is compatible with the old IEEE1394 stack.  Old
  applications would not expect the DMA program to stop in this case.

* Since the OHCI driver does not report any uncompleted packets, the
  context would stop silently; clients would not have any chance to
  detect and handle this error without a watchdog timer.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>

Pieter Palmers notes:

"The reason I added this retry behavior to the old stack is because some
cards now and then fail to send a packet (e.g. the o2micro card in my
dell laptop).  I couldn't figure out why exactly this happens, my best
guess is that the card cannot fetch the payload data on time.  This
happens much more frequently when sending large packets, which leads me
to suspect that there are some contention issues with the DMA that fills
the transmit FIFO.

In the old stack it was a pretty critical issue as it resulted in a
freeze of the userspace application.

The omission of a packet doesn't necessarily have to be an issue.  E.g.
in IEC61883 streams the DBC field can be used to detect discontinuities
in the stream.  So as long as the other side doesn't bail when no
[packet] is present in a cycle, there is not really a problem.

I'm not convinced though that retrying is the proper solution, but it is
simple and effective for what it had to do.  And I think there are no
reasons not to do it this way.  Userspace can still detect this by
checking the cycle the descriptor was sent in."

Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> (changelog, comment)

In the transmit path of firewire-net (IPv4 over 1394), the following
race condition may occur:
  - The networking soft IRQ inserts a datagram into the 1394 async
    request transmit DMA.
  - The 1394 async transmit completion tasklet runs to finish cleaning
    up (unlink datagram from list of pending ones, release skb and
    outbound 1394 transaction object) --- before the networking soft IRQ
    had a chance to proceed and add the datagram to the list of pending
    datagrams.

This caused a panic in the 1394 async transmit completion tasklet when
it dereferenced unitialized list heads:
http://bugzilla.kernel.org/show_bug.cgi?id=15077

The fix is to add checks in the tx soft IRQ and in the tasklet to
determine which of these two is the last referrer to the transaction
object.  Then handle the cleanup of the object by the last referrer
rather than assuming that the tasklet is always the last one.

There is another similar race:  Between said tasklet and fwnet_close,
i.e. at ifdown.  However, that race is much less likely to occur in
practice and shall be fixed in a separate update.
Reported-by: NИлья Басин <basinilya@gmail.com>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Unsurprisingly, Texas Instruments TSB43AB23 exhibits the same behaviour
as TSB43AB22/A in dual buffer IR DMA mode:  If descriptors are located
at physical addresses above the 31 bit address range (2 GB), the
controller will overwrite random memory.  With luck, this merely
prevents video reception.  With only a little less luck, the machine
crashes.

We use the same workaround here as with TSB43AB22/A:  Switch off the
dual buffer capability flag and use packet-per-buffer IR DMA instead.
Another possible workaround would be to limit the coherent DMA mask to
31 bits.

In Linux 2.6.33, this change serves effectively only as documentation
since dual buffer mode is not used for any controller anymore.  But
somebody might want to re-enable it in the future to make use of
features of dual buffer DMA that are not available in packet-per-buffer
mode.

In Linux 2.6.32 and older, this update is vital for anyone with this
controller, more than 2 GB RAM, a 64 bit kernel, and FireWire video or
audio applications.

We have at least four reports:
http://bugzilla.kernel.org/show_bug.cgi?id=13808
http://marc.info/?l=linux1394-user&m=126154279004083
https://bugzilla.redhat.com/show_bug.cgi?id=552142
http://marc.info/?l=linux1394-user&m=126432246128386

Reported-by: Paul Johnson
Reported-by: Ronneil Camara
Reported-by: G Zornetzer
Reported-by: Mark Thompson
Cc: stable@kernel.org
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>