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

Stopping an isochronous reception DMA context takes two loop iterations
in context_stop on several controllers (JMicron, NEC, VIA).  But there
is no extra delay necessary between these two reg_read trials; the MMIO
reads themselves are slow enough.  Hence bring back the behavior from
before commit dd6254e5 "firewire: ohci:
remove superfluous posted write flushes" on these controllers by means
of an "if (i)" condition.

Isochronous context stop is performed in preemptible contexts (and only
rarely), hence this change is of little impact.  (Besides, Agere and TI
controllers always, or almost always, have the context stopped already
at the first ContextControl read.)

More important is asynchronous transmit context stop, which is performed
while local interrupts are disabled (on the two AT DMAs in
bus_reset_tasklet, i.e. after a self-ID-complete event).  In my
experience with several controllers, tested with a usermode AT-request
transmitter as well as with FTP transmission over firewire-net, the AT
contexts were luckily already stopped at the first ContextControl read,
i.e. never required another MMIO read let alone mdelay.  A possible
explanation for this is that the controllers which I tested perhaps stop
AT DMA before they perform the self-ID reception DMA.

But we cannot be sure about that and should keep the interrupts-disabled
busy loop as short as possible.  Hence, query the ContextControl
register in 1000 udelay(10) intervals instead of 10 udelay(1000)
intervals.  I understand from an estimation by Clemens Ladisch that
stopping a busy DMA context should take microseconds or at worst tens of
microseconds, not milliseconds.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

The call to flush_writes() in context_stop() is superfluous because
another register read is done immediately afterwards.

The call to flush_writes() in ar_context_run() does not need to be done
individually for each AR context, so move it to ohci_enable().  This
also makes ohci_enable() clearer because it no longer depends on a side
effect of ar_context_run() to flush its own register writes.

Finally, the setting of a context's wake bit does not need to be flushed
because neither the driver logic nor the API require the CPU to wait for
this action.  This removes the last MMIO reads from the packet queueing
code paths.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

When queueing iso packets, the run time is dominated by the two
MMIO accesses that set the DMA context's wake bit.  Because most
drivers submit packets in batches, we can save much time by
removing all but the last wakeup.

The internal kernel API is changed to require a call to
fw_iso_context_queue_flush() after a batch of queued packets.
The user space API does not change, so one call to
FW_CDEV_IOC_QUEUE_ISO must specify multiple packets to take
advantage of this optimization.

In my measurements, this patch reduces the time needed to queue
fifty skip packets from userspace to one sixth on a 2.5 GHz CPU,
or to one third at 800 MHz.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Current implementation of ohci_set_config_rom() uses a deferred
bus reset via fw_schedule_bus_reset(). If clients add multiple
unit descriptors to the config_rom in quick succession, the
deferred bus reset may not have fired before succeeding update
requests have come in. This can lead to an incorrect partial
update of the config_rom for both addition and removal of
config_rom descriptors, as the ohci_set_config_rom() routine
will return -EBUSY if a previous pending update has not been
completed yet; the requested update just gets dropped on the floor.

This patch recognizes that the "in-flight" update can be modified
until it has been processed by the bus-reset, and the locking
in the bus_reset_tasklet ensures that the update is done atomically
with respect to modifications made by ohci_set_config_rom(). The
-EBUSY error case is simply removed.

[Stefan R:  The bug always existed at least theoretically.  But it
became easy to trigger since 2.6.36 commit 02d37bed "firewire: core:
integrate software-forced bus resets with bus management" which
introduced long mandatory delays between janitorial bus resets.]
Signed-off-by: NBenjamin Buchalter <bj@mhlabs.com>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> (trivial style changes)
Cc: <stable@kernel.org> # 2.6.36.y and newer

When z==2, the condition "key == 2" is superfluous because it cannot
occur without "b == 3", as a descriptor with b!=3 and key==2 would be
an OUTPUT_MORE_IMMEDIATE descriptor which cannot be used alone.

Also remove magic numbers and needless computations on the b field.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

For AT packet payloads of up to eight bytes, we have enough unused space
in the DMA descriptors list so that we can put a copy of the payload
there and thus avoid having to create a separate streaming DMA mapping
for the payload buffer.

In a CPU-bound microbenchmark that just sends 8-byte packets, bandwidth
was measured to increase by 5.7 %, from 1009 KB/s to 1067 KB/s.  In
practice, the only performance-sensitive usage of small asynchronous
packets is the SBP-2 driver's write to the ORB_POINTER register during
SCSI command submission.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

OHCI 1.1 5.7.3 not only forbids enabling or starting any DMA contexts
before the linkEnable bit is set, but also explicitly warns of undefined
behaviour if this order is violated.

Don't violate it then.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

It seems drivers/firewire/ohci.c is making some optimistic assumptions
about struct fw_ohci and that member "card" will always remain the first
member of the struct.
Plus it's probably going to confuse a lot of static code analyzers too.

So I wonder if there is a good reason not to free the ohci struct just
like it was allocated instead of the tricky &ohci->card way?
Signed-off-by: NOleg Drokin <green@linuxhacker.ru>

It is perhaps just a rudiment from before mainline submission of the
driver.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Since commit 82b662dc "flush AT contexts after bus reset for OHCI 1.2",
the driver takes care of any AT packets that were enqueued during a bus
reset phase.  The check from commit 76f73ca1 is therefore no longer
necessary and the MMIO read can be avoided.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

If a misguided program tried to start an isochronous context before it
has queued any packets, the call would appear to succeed, but the
context would not actually go into the running state, and the OHCI
controller would then raise an unrecoverableError interrupt because the
first Z value is zero and thus invalid.  The driver logs such errors,
but there is no mechanism to report this back to the program.

Add an explicit check so that this error can be returned synchronously.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

To prevent the iso packet callback from being called after
fw_iso_context_stop() has returned, make sure that the
context's tasklet has finished executing before that.

This fixes access-after-free bugs that have so far been
observed only in the upcoming snd-firewire-speakers driver,
but can theoretically also happen in the firedtv driver.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

When a DMA context goes into the dead state (and the controller thus
stops working correctly), logging this error and the controller's error
code might be helpful for debugging.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

PAGE_KERNEL_RO is not available on all architectures, so its use
in the new AR code broke compilation on sparc64.

Because the read-only mapping was just a debugging aid, just use
PAGE_KERNEL instead.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>

James Bottomley wrote:
> On Thu, 2011-01-13 at 08:27 +0100, Clemens Ladisch wrote:
>> firewire: ohci: fix compilation on arches without PAGE_KERNEL_RO, e.g. sparc
>>
>> PAGE_KERNEL_RO is not available on all architectures, so its use in the
>> new AR code broke compilation on sparc64.
>>
>> Because the R/O mapping is only used to catch drivers that try to write
>> to the reception buffer and not actually required for correct operation,
>> we can just use a normal PAGE_KERNEL mapping where _RO is not available.
[...]
>> +/*
>> + * For archs where PAGE_KERNEL_RO is not supported;
>> + * mapping the AR buffers readonly for the CPU is just a debugging aid.
>> + */
>> +#ifndef PAGE_KERNEL_RO
>> +#define PAGE_KERNEL_RO PAGE_KERNEL
>> +#endif
>
> This might cause interesting issues on sparc64 if it ever acquired a
> PAGE_KERNEL_RO.  Sparc64 has extern pgprot_t for it's PAGE_KERNEL types
> rather than #defines, so the #ifdef check wouldn't see this.
>
> I think either PAGE_PROT_RO becomes part of our arch API (so all
> architectures are forced to add it), or, if it's not part of the API,
> ohci isn't entitled to use it.  The latter seems simplest since you have
> no real use for write protection anyway.
Reported-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

"firewire: ohci: restart iso DMA contexts on resume from low power mode"
added the flag struct context.active and "firewire: ohci: cache the
context run bit" added struct context.running.

These flags contain the same information; combine them.
Also, normalize whitespace in pci_resume().
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

The DMA context run control bit is entirely controlled by software, so
it is safe to cache it.  This allows the driver to avoid doing an
additional MMIO read when queueing an AT packet.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Add comments
  - on why bus_reset_tasklet flushes AT queues,
  - that commit 76f73ca1 can possibly be reverted now.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>
Acked-by: NJarod Wilson <jarod@redhat.com>

The OHCI 1.2 (draft) specification, clause 7.2.3.3, allows and
recommends that, after a bus reset, the controller does not flush all
the packets in the AT queues.  Therefore, the driver has to do this
itself.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Restore iso channels DMA so that iso channels could continue to work
after resume from RAM/disk.
Signed-off-by: NMaxim Levitsky <maximlevitsky@gmail.com>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Some lousy BIOSes, e.g. my Aspire 5720 BIOS forget to restore the GUID
register on resume from RAM.

Fix that by setting it to the last value that was read from it.
Signed-off-by: NMaxim Levitsky <maximlevitsky@gmail.com>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

The buffers used for the selfIDs packets and the AR request and response
descriptors end up using three pages because dma_alloc_coherent()
allocates at least one page per call.  However, these data structures
would all fit into 4 KB, so we can save space by using a common buffer
for them.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

When the isochRx/isochTx bit is clear, we do not need to read the
corresponding iso interrupt event register.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Change the header of PHY packets to be sent to include a pseudo
transaction code.  This makes the header consistent with that of
received PHY packets, and allows at_context_queue_packet() and
log_ar_at_event() to see the packet type directly instead of having
to deduce it from the header length or even from the header contents.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

To remove the error information from the controller's queue and to allow
more posted writes, the driver has to read the failed posted write
address before clearing the postedWriteErr interrupt bit.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>

(Stefan R:) The spec is somewhat fuzzy about the actual requirements.
To err on the safe side, let's do these two read accesses.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Make sure that interrupt event clear bit writes are executed before the
interrupt handler returns.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Add proper error handling for the context_init() calls.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Move the AR DMA descriptors out of the buffer pages, and map the buffer
pages linearly into the kernel's address space.  This allows the driver
to ignore any page boundaries in the DMA data and thus to avoid any
copying around of packet payloads.

This fixes the bug where S800 packets that are so big (> 4080 bytes)
that they can be split over three pages were not handled correctly.

Due to the changed algorithm, we can now use arbitrarily many buffer
pages, which improves performance because the controller can more easily
unload its DMA FIFO.

Furthermore, using streaming DMA mappings should improve perfomance on
architectures where coherent DMA mappings are not cacheable.  Even on
other architectures, the caching behaviour should be improved slightly
because the CPU no longer writes to the buffer pages.

v2: Detect the last filled buffer page by searching the descriptor's
    residual count value fields in order (like in the old code), instead
    of going backwards through the transfer status fields; it looks as
    if some controllers do not set the latter correctly.

v3: Fix an old resume bug that would now make the handler run into
    a BUG_ON, and replace that check with more useful error handling.
    Increase the buffer size for better performance with non-TI chips.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>

Maxim Levitsky writes:
    Works almost perfectly.  I can still see RCODE_BUSY errors
    sometimes, not very often though.  64K here eliminates these errors
    completely.  This is most likely due to nouveau drivers and lowest
    perf level I use to lower card temperature.  That increases
    latencies too much I think.  Besides that the IO is just perfect.
Tested-by: NMaxim Levitsky <maximlevitsky@gmail.com>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Agere FW643 rev 06, listed as "11c1:5901 (rev 06) (prog-if 10 [OHCI])",
produced SBP-2 I/O errors since kernel 2.6.36.  Disabling MSI fixes it.

Since MSI work on Agere FW643-E (same vendor and device ID, but rev 07),
introduce a device revision field into firewire-ohci's quirks list so
that different quirks can be defined for older and newer revisions.
Reported-by: NJonathan Isom <jeisom@gmail.com>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>
Cc: <stable@kernel.org> # 2.6.36.y

"VIA Technologies, Inc. VT6315 Series Firewire Controller [1106:3403]"
does not generate any interrupts if Message Signaled Interrupts were
enabled.  This is a regression since kernel 2.6.36 in which MSI support
was added to firewire-ohci.  Hence blacklist MSI on all VIA controllers.
Reported-by: NRobin Cook <rcook@wyrms.net>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>
Cc: <stable@kernel.org> # 2.6.36.y

If the controller is storing a split packet and therefore changing
d->res_count to zero between the two reads by the driver, we end up with
an end pointer that is not at a packet boundary, and therefore overflow
the buffer when handling the split packet.

To fix this, read the field once, atomically.  The compiler usually
merges the two reads anyway, but for correctness, we have to enforce it.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Tested-by: NMaxim Levitsky <maximlevitsky@gmail.com>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

Freeing an AR buffer page just to allocate a new page immediately
afterwards is not only a pointless effort but also dangerous because
the allocation can fail, which would result in an oops later.

Split ar_context_add_page() into two functions so that we can reuse
the old page directly.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Tested-by: NMaxim Levitsky <maximlevitsky@gmail.com>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

When handling an AR buffer that has been completely filled, we assumed
that its descriptor will not be read by the controller and can be
overwritten.  However, when the last received packet happens to end at
the end of the buffer, the controller might not yet have moved on to the
next buffer and might read the branch address later.  If we overwrite
and free the page before that, the DMA context will either go dead
because of an invalid Z value, or go off into some random memory.

To fix this, ensure that the descriptor does not get overwritten by
using only the actual buffer instead of the entire page for reassembling
the split packet.  Furthermore, to avoid freeing the page too early,
move on to the next buffer only when some data in it guarantees that the
controller has moved on.

This should eliminate the remaining firewire-net problems.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Cc: 2.6.22-2.6.36 <stable@kernel.org>
Tested-by: NMaxim Levitsky <maximlevitsky@gmail.com>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

When the controller had to split a received asynchronous packet into two
buffers, the driver tries to reassemble it by copying both parts into
the first page.  However, if size + rest > PAGE_SIZE, i.e., if the yet
unhandled packets before the split packet, the split packet itself, and
any received packets after the split packet are together larger than one
page, then the memory after the first page would get overwritten.

To fix this, do not try to copy the data of all unhandled packets at
once, but copy the possibly needed data every time when handling
a packet.

This gets rid of most of the infamous crashes and data corruptions when
using firewire-net.
Signed-off-by: NClemens Ladisch <clemens@ladisch.de>
Cc: 2.6.22-2.6.36 <stable@kernel.org>
Tested-by: NMaxim Levitsky <maximlevitsky@gmail.com>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> (cast PAGE_SIZE to size_t)

Revert commit 54672386
"firewire: ohci: fix up configuration of TI chips".
It caused massive slow-down and data corruption with a TSB82AA2 based
StarTech EC1394B2 ExpressCard and FireWire 800 harddisks.

https://bugs.launchpad.net/ubuntu/+source/linux/+bug/657081
http://thread.gmane.org/gmane.linux.kernel.firewire.user/4013

The fact that some card EEPROMs do not program these enhancements may be
related to TSB81BA3 phy chip errata, if not to bugs of TSB82AA2 itself.
We could re-add these configuration steps, but only conditional on a
whitelist of cards on which these enhancements bring a proven positive
effect.
Reported-and-tested-by: NEric Shattow <lucent@gmail.com>
Cc: Clemens Ladisch <clemens@ladisch.de>
Cc: <stable@kernel.org> 2.6.35
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

The Ricoh FireWire controllers appear to have the non-atomic cycle
timer register access bug, so, activate the driver workaround by
default.

The behaviour was observed on:
Ricoh Co Ltd R5C552 IEEE 1394 Controller [1180:0552] and
Ricoh Co Ltd R5C832 IEEE 1394 Controller [1180:0832] (rev 04).
Signed-off-by: NHeikki Lindholm <holin@iki.fi>
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

VIA VT6306, VIA VT6308, and NEC OrangeLink controllers do not write
packet event codes for received PHY packets (or perhaps write
evt_no_status, hard to tell).  Work around it by overwriting the
packet's ACK by ack_complete, so that upper layers that listen to PHY
packet reception get to see these packets.

(Also tested:  TI TSB82AA2, TI TSB43AB22/A, TI XIO2213A, Agere FW643,
JMicron JMB381 --- these do not exhibit this bug.)

Clemens proposed a quirks flag for that, IOW whitelist known misbehaving
controllers for this workaround.  Though to me it seems harmless enough
to enable for all controllers.

The log_ar_at_event() debug log will continue to show the original
status from the DMA unit.

Reported-by: Clemens Ladisch <clemens@ladisch.de> (VT6308)
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

This adds the DMA context programming and userspace ABI for multichannel
reception, i.e. for listening on multiple channel numbers by means of a
single DMA context.

The use case is reception of more streams than there are IR DMA units
offered by the link layer.  This is already implemented by the older
ohci1394 + ieee1394 + raw1394 stack.  And as discussed recently on
linux1394-devel, this feature is occasionally used in practice.

The big drawbacks of this mode are that buffer layout and interrupt
generation necessarily differ from single-channel reception:  Headers
and trailers are not stripped from packets, packets are not aligned with
buffer chunks, interrupts are per buffer chunk, not per packet.

These drawbacks also cause a rather hefty code footprint to support this
rarely used OHCI-1394 feature.  (367 lines added, among them 94 lines of
added userspace ABI documentation.)

This implementation enforces that a multichannel reception context may
only listen to channels to which no single-channel context on the same
link layer is presently listening to.  OHCI-1394 would allow to overlay
single-channel contexts by the multi-channel context, but this would be
a departure from the present first-come-first-served policy of IR
context creation.

The implementation is heavily based on an earlier one by Jay Fenlason.
Thanks Jay.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

firewire-ohci keeps book of which isochronous channels are occupied by
IR DMA contexts, so that there cannot be more than one context listening
to a certain channel.

If IR context creation failed due to an out-of-memory condition, this
bookkeeping leaked a channel.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

When we append to a DMA program, we need to ensure that the order in
which initialization of the new descriptors and update of the
branch_address of the old tail descriptor, as seen by the PCI device,
happen as intended.
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>

This extends the FW_CDEV_IOC_SEND_PHY_PACKET ioctl() for /dev/fw* to be
useful for ping time measurements.  One application for it would be gap
count optimization in userspace that is based on ping times rather than
hop count.  (The latter is implemented in firewire-core itself but is
not applicable to beta PHYs that act as repeater.)
Signed-off-by: NStefan Richter <stefanr@s5r6.in-berlin.de>