There is a DMA map/ unmap imbalance whenever a block write request
packet is sent and then dequeued with ohci_cancel_packet. The latter
may happen frequently if the AR resp tasklet is executed before the AT
req tasklet for the same transaction.
Add the missing dma_unmap_single. This fixes
https://bugzilla.redhat.com/show_bug.cgi?id=475156
Reported-by: Emmanuel Kowalski
Tested-by: Emmanuel Kowalski
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Add another model ID of a broken firmware to prevent early I/O errors
by acesses at the end of the disk. Reported at linux1394-user,
http://marc.info/?t=122670842900002
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
1: There is a small race between queue_delayed_work() and its
corresponding kref_get(). Do the kref_get first, and _put it again
if the queue_delayed_work() failed, so there is no chance of the
kref going to zero while the work is scheduled.
2: An SBP2_LOGOUT_REQUEST could be sent out with a login_id full of
garbage. Initialize it to an invalid value so we can tell if we
ever got a valid login_id.
3: The node ID and generation may have changed but the new values may
not yet have been recorded in lu and tgt when the final logout is
attempted. Use the latest values from the device in
sbp2_release_target().
Signed-off-by: Jay Fenlason <fenlason@redhat.com>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
This optimizes firewire-sbp2's device probe for the case that the local
node and the SBP-2 node were discovered at the same time. In this case,
fw-core's bus management work and fw-sbp2's login and SCSI probe work
are scheduled in parallel (in the globally shared workqueue and in
fw-sbp2's workqueue, respectively). The bus reset from fw-core may then
disturb and extremely delay the login and SCSI probe because the latter
fails with several command timeouts and retries and has to be retried
from scratch.
We avoid this particular situation of sbp2_login() and fw_card_bm_work()
running in parallel by delaying the first sbp2_login() a little bit.
This is meant to be a short-term fix for
https://bugzilla.redhat.com/show_bug.cgi?id=466679. In the long run,
the SCSI probe, i.e. fw-sbp2's call of __scsi_add_device(), should be
parallelized with sbp2_reconnect().
Problem reported and fix tested and confirmed by Alex Kanavin.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
The transmit and receive context dma memory was not being freed on
module removal. Neither was the config rom memory. Fix that.
The ab->next assignment is pure paranoia.
Signed-off-by: Jay Fenlason <fenlason@redhat.com>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
With the bus_resets patch applied, it is easy to see this memory leak
by repeatedly resetting the firewire bus while running slabtop in
another window. Just watch kmalloc-32 grow and grow...
Signed-off-by: Jay Fenlason <fenlason@redhat.com>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
The "color" is used during the topology building after a bus reset,
hovever in "struct fw_node"s it is stored in a u8, but in struct fw_card
it is stored in an int. When the value wraps in one struct, but not
the other, disaster strikes.
Signed-off-by: Jay Fenlason <fenlason@redhat.com>
Fixes http://bugzilla.kernel.org/show_bug.cgi?id=10922.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Reported by Jay Fenlason: ioctl() did not return as intended
- the size of data read into ioctl_send_request,
- the number of datagrams enqueued by ioctl_queue_iso.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
queuecommand() looked at the remote and local node IDs before it read
the bus generation. The corresponding race with sbp2_reconnect updating
these data was probably impossible to happen though because the current
code blocks the SCSI layer during reconnection. However, better safe
than sorry, especially if someone later improves the code to not block
the SCSI layer.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
1. We don't need to round the SBP-2 segment size limit down to a
multiple of 4 kB (0xffff -> 0xf000). It is only necessary to
ensure quadlet alignment (0xffff -> 0xfffc).
2. Use dma_set_max_seg_size() to tell the DMA mapping infrastructure
and the block IO layer about the restriction. This way we can
remove the size checks and segment splitting in the queuecommand
path.
This assumes that no other code in the firewire stack uses
dma_map_sg() with conflicting requirements. It furthermore assumes
that the controller device's platform actually allows us to set the
segment size to our liking. Assert the latter with a BUG_ON().
3. Also use blk_queue_max_segment_size() to tell the block IO layer
about it. It cannot know it because our scsi_add_host() does not
point to the FireWire controller's device.
Thanks to Grant Grundler and FUJITA Tomonori for advice.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Share code between fw_send_request + wait_for_completion callers.
Signed-off-by: Jay Fenlason <fenlason@redhat.com>
Addendum:
Removes an unnecessary struct and an ununsed retry loop.
Calls it fw_run_transaction() instead of fw_send_request_sync().
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Acked-by: Kristian Høgsberg <krh@redhat.com>
Recently, a bug having to do with the alignment of transaction response
data was fixed. However, some apps such as libdc1394 relied on the
presence of that bug in order to function correctly. In order to stay
compatible with old versions of those apps, this patch preserves the bug
in cases where it is harmless to normal operation (such as the single
quadlet read) due to a simple duplication of data. This guarantees
maximum compatability for those users who are using the old app with the
fixed kernel.
Signed-off-by: David Moore <dcm@acm.org>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Add per-device dma_mapping_ops support for CONFIG_X86_64 as POWER
architecture does:
This enables us to cleanly fix the Calgary IOMMU issue that some devices
are not behind the IOMMU (http://lkml.org/lkml/2008/5/8/423).
I think that per-device dma_mapping_ops support would be also helpful for
KVM people to support PCI passthrough but Andi thinks that this makes it
difficult to support the PCI passthrough (see the above thread). So I
CC'ed this to KVM camp. Comments are appreciated.
A pointer to dma_mapping_ops to struct dev_archdata is added. If the
pointer is non NULL, DMA operations in asm/dma-mapping.h use it. If it's
NULL, the system-wide dma_ops pointer is used as before.
If it's useful for KVM people, I plan to implement a mechanism to register
a hook called when a new pci (or dma capable) device is created (it works
with hot plugging). It enables IOMMUs to set up an appropriate
dma_mapping_ops per device.
The major obstacle is that dma_mapping_error doesn't take a pointer to the
device unlike other DMA operations. So x86 can't have dma_mapping_ops per
device. Note all the POWER IOMMUs use the same dma_mapping_error function
so this is not a problem for POWER but x86 IOMMUs use different
dma_mapping_error functions.
The first patch adds the device argument to dma_mapping_error. The patch
is trivial but large since it touches lots of drivers and dma-mapping.h in
all the architecture.
This patch:
dma_mapping_error() doesn't take a pointer to the device unlike other DMA
operations. So we can't have dma_mapping_ops per device.
Note that POWER already has dma_mapping_ops per device but all the POWER
IOMMUs use the same dma_mapping_error function. x86 IOMMUs use device
argument.
[akpm@linux-foundation.org: fix sge]
[akpm@linux-foundation.org: fix svc_rdma]
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: fix bnx2x]
[akpm@linux-foundation.org: fix s2io]
[akpm@linux-foundation.org: fix pasemi_mac]
[akpm@linux-foundation.org: fix sdhci]
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: fix sparc]
[akpm@linux-foundation.org: fix ibmvscsi]
Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Muli Ben-Yehuda <muli@il.ibm.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Avi Kivity <avi@qumranet.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Use only statically allocated data for PHY config packet transmission.
With the previous incarnation, some data wouldn't be freed if the packet
transmit callback was never called.
A theoretical drawback now is that, in PCs with more than one card,
card A may complete() for a waiter on card B. But this is highly
unlikely and its impact not serious. Bus manager B may reset bus B
before the PHY config went out, but the next phy config on B should be
fine. However, with a timeout of 100ms, this situation is close to
impossible.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Isochronous reception in dualbuffer mode is reportedly broken with
TI TSB43AB22A on x86-64. Descriptor addresses above 2G have been
determined as the trigger:
https://bugzilla.redhat.com/show_bug.cgi?id=435550
Two fixes are possible:
- pci_set_consistent_dma_mask(pdev, DMA_31BIT_MASK);
at least when IR descriptors are allocated, or
- simply don't use dualbuffer.
This fix implements the latter workaround.
But we keep using dualbuffer on x86-32 which won't give us highmen (and
thus physical addresses outside the 31bit range) in coherent DMA memory
allocations. Right now we could for example also whitelist PPC32, but
DMA mapping implementation details are expected to change there.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Signed-off-by: Jarod Wilson <jwilson@redhat.com>
There will be 4 padding bytes in struct fw_cdev_event_response on some platforms
The member:__u32 data will point to these padding bytes. While queue the
response and data in complete_transaction in fw-cdev.c, it will queue like this:
|response(excluding padding bytes)|4 padding bytes|4 padding bytes|data.
It queue 4 extra bytes. That is to say it use "&response + sizeof(response)"
while other place of kernel and userspace library use "&response + offsetof
(typeof(response), data)". So it will lost the last 4 bytes of data. This patch
can fix it while not changing the struct definition.
Signed-off-by: JiSheng Zhang <jszhang3@mail.ustc.edu.cn>
This fixes responses to outbound block read requests on 64bit architectures.
Tested on i686, x86-64, and x86-64 with i686 userland, using firecontrol and
gscanbus.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
After card->done and card->work are completed, any remaining pending
request would be a bug. We cannot safely complete a transaction at
that point anymore.
IOW card users must not drop their last fw_card reference (usually
indirect references through fw_device references) before their last
outbound transaction through that card was finished.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
- better name for a function argument
- removal of a local variable which became unnecessary after
"fully initialize fw_transaction before marking it pending"
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
In theory, card->flush_timer could already access a transaction between
fw_send_request()'s spin_unlock_irqrestore and the rest of what happens
in fw_send_request(). This would happen if the process which sends the
request is preempted and put to sleep right after spin_unlock_irqrestore
for longer than 100ms.
Therefore we fill in everything in struct fw_transaction at which the
flush_timer might look at before we lift the lock.
To do: Ensure that the timer does not pick up the transaction before
the time of the AT request event plus split transaction timeout.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Reported by Jay Fenlason: A bus reset tasklet may call
fw_flush_transactions and touch transactions (call their callback which
will free them) while the context which submitted the transaction is
still inserting it into the transmission queue.
A simple solution to this problem is to _not_ "flush" the transactions
because of a bus reset (complete the transcations as 'cancelled'). They
will now simply time out (completed as 'cancelled' by the split-timeout
timer).
Jay Fenlason thought of this fix too but I was quicker to type it out.
:-)
Background:
Contexts which access an instance of struct fw_transaction are:
1. the submitter, until it inserted the packet which is embedded in the
transaction into the AT req DMA,
2. the AsReqTrContext tasklet when the request packet was acked by the
responder node or transmission to the responder failed,
3. the AsRspRcvContext tasklet when it found a request which matched
an incoming response,
4. the card->flush_timer when it picks up timed-out transactions to
cancel them,
5. the bus reset tasklet when it cancels transactions (this access is
eliminated by this patch),
6. a process which shuts down an fw_card (unregisters it from fw-core
when the controller is unbound from fw-ohci) --- although in this
case there shouldn't really be any transactions anymore because we
wait until all card users finished their business with the card.
All of these contexts run concurrently (except for the 6th, presumably).
The 1st is safe against the 2nd and 3rd because of the way how a request
packet is carefully submitted to the hardware. A race between 2nd and
3rd has been fixed a while ago (bug 9617). The 4th is almost safe
against 1st, 2nd, 3rd; there are issues with it if huge scheduling
latencies occur, to be fixed separately. The 5th looks safe against
2nd, 3rd, and 4th but is unsafe against 1st. Maybe this could be fixed
with an explicit state variable in struct fw_transaction. But this
would require fw_transaction to be rewritten as only dynamically
allocatable object with reference counting --- not a good solution if we
also can simply kill this 5th accessing context (replace it by the 4th).
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Contrary to a comment in the source, request->ack of a broadcast write
request can be ACK_PENDING. Hence the existing check is insufficient.
Debug dmesg before:
AR spd 0 tl 00, ffc0 -> ffff, ack_pending , QW req, fffff0000234 = ffffffff
AT spd 0 tl 00, ffff -> ffc0, ack_complete, W resp
And the requesting node (linux1394) reports an unsolicited response.
Debug dmesg after:
AR spd 0 tl 00, ffc0 -> ffff, ack_pending , QW req, fffff0000234 = ffffffff
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
This is a functionally equivalent replacement of the current reference
counting of struct fw_card instances. It only converts it to common
idioms as suggested by Kristian Høgsberg:
- struct kref replaces atomic_t as the counter.
- wait_for_completion is used to wait for all card users to complete.
BTW, it may make sense to count card->flush_timer and card->work as
card users too.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
This instructs sd_mod to send START STOP UNIT on suspend and resume,
and on driver unbinding or unloading (including when the system is shut
down).
We don't do this though if multiple initiators may log in to the target.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Tested-by: Tino Keitel <tino.keitel@gmx.de>
Reported by Tino Keitel: PL-3507 with firmware from Prolific does not
spin down the disk on START STOP UNIT with power condition = 0 and start
= 0. It does however work with power condition = 2 or 3.
Also found while investigating this: DViCO Momobay CX-1 and FX-3A (TI
TSB42AA9/A based) become unresponsive after START STOP UNIT with power
condition = 0 and start = 0. They stay responsive if power condition is
set when stopping the motor.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Tested-by: Tino Keitel <tino.keitel@gmx.de>
There is a small off-by-one bug in firewire-sbp2. This causes problems
when a device exports multiple LUN Directories. I found it when trying
to talk to a SONY DVD Jukebox.
Signed-off-by: Richard Sharpe <realrichardsharpe@gmail.com>
Acked-by: Kristian Høgsberg <krh@redhat.com>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> (op. order, changelog)
Emphasize the recommendation to build only one stack.
Trim the prompts to better fit into short attention spans.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
If the low-level driver failed to initialize a card properly without
noticing it, fw-core was blocked indefinitely when trying to send a
PHY config packet. This hung up the events kernel thread, e.g. locked
up keyboard input.
https://bugzilla.redhat.com/show_bug.cgi?id=444694https://bugzilla.redhat.com/show_bug.cgi?id=446763
This problem was introduced between 2.6.25 and 2.6.26-rc1 by commit
2a0a259049 "firewire: wait until PHY
configuration packet was transmitted (fix bus reset loop)".
The solution is to wait with timeout. I tested it with 7 different
working controllers and 1 non-working controller. On the working ones,
the packet callback complete()s usually --- but not always --- before a
timeout of 10ms. Hence I chose a safer timeout of 100ms.
On the few tests with the non-working controller ALi M5271, PHY config
packet transmission always timed out so far. (Fw-ohci needs to be fixed
for this controller independently of this deadline fix. Often the core
doesn't even attempt to send a phy config because not even self ID
reception works.)
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
The messages which can be enabled by fw-ohci's debug module parameter
are changed from KERN_DEBUG to KERN_NOTICE level and uniformly prefixed
with "firewire_ohci: ". This further simplifies communication with
users when we ask them to capture debug messages.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Callers of fill_bus_reset_event() have to take card->lock. Otherwise
access to node data may oops if node removal is in progress.
A lockless alternative would be
- event->local_node_id = card->local_node->node_id;
+ tmp = fw_node_get(card->local_node);
+ event->local_node_id = tmp->node_id;
+ fw_node_put(tmp);
and ditto with the other node pointers which fill_bus_reset_event()
accesses. But I went the locked route because one of the two callers
already holds the lock. As a bonus, we don't need the memory barrier
anymore because device->generation and device->node_id are written in
a card->lock protected section.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Signed-off-by: Kristian Høgsberg <krh@redhat.com>
OHCI 1.1 clause 5.10 requires that selfIDBufferPtr is valid when a 1 is
written into LinkControl.rcvSelfID.
This driver bug has so far not been known to cause harm because most
chips obviously accept a later selfIDBufferPtr write, at least before
HCControl.linkEnable is written.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Signed-off-by: Jarod Wilson <jwilson@redhat.com>
Signed-off-by: Kristian Høgsberg <krh@redhat.com>
We want the rcvPhyPkt bit in LinkControl off before we start using the
chip. However, the spec says that the reset value of it is undefined.
Hence switch it explicitly off.
https://bugzilla.redhat.com/show_bug.cgi?id=244576#c48 shows that for
example the nForce2 integrated FireWire controller seems to have it on
by default.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Signed-off-by: Jarod Wilson <jwilson@redhat.com>
header_length and payload_length are filled with random data if an
unknown tcode was read from the AR buffer (i.e. if the AR buffer
contained invalid data).
We still need a better strategy to recover from this, but at least
handle_ar_packet now doesn't return out of bound buffer addresses
anymore.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
BUG() at this place is wrong. (Unless if the low level driver would
already do higher-level input validation of incoming request headers.)
Invalid incoming requests or bugs in the controller which corrupt the
AR-req buffer needlessly crashed the box because this is run in tasklet
context.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
If userspace ignores the POLLERR bit from poll(), and only attempts to
read() the device when POLLIN is set, it can still make ioctl() calls on
a device that has been removed from the system. The node_id and
generation returned by GET_INFO will be outdated, but INITIATE_BUS_RESET
would still cause a bus reset, and GET_CYCLE_TIMER will return data.
And if you guess the correct generation to use, you can send requests to
a different device on the bus, and get responses back.
This patch prevents open, ioctl, compat_ioctl, and mmap against shutdown
devices.
Signed-off-by: Jay Fenlason <fenlason@redhat.com>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
- struct scsi_cmnd had a 16 bytes command buffer of its own.
This is an unnecessary duplication and copy of request's
cmd. It is probably left overs from the time that scsi_cmnd
could function without a request attached. So clean that up.
- Once above is done, few places, apart from scsi-ml, needed
adjustments due to changing the data type of scsi_cmnd->cmnd.
- Lots of drivers still use MAX_COMMAND_SIZE. So I have left
that #define but equate it to BLK_MAX_CDB. The way I see it
and is reflected in the patch below is.
MAX_COMMAND_SIZE - means: The longest fixed-length (*) SCSI CDB
as per the SCSI standard and is not related
to the implementation.
BLK_MAX_CDB. - The allocated space at the request level
- I have audit all ISA drivers and made sure none use ->cmnd in a DMA
Operation. Same audit was done by Andi Kleen.
(*)fixed-length here means commands that their size can be determined
by their opcode and the CDB does not carry a length specifier, (unlike
the VARIABLE_LENGTH_CMD(0x7f) command). This is actually not exactly
true and the SCSI standard also defines extended commands and
vendor specific commands that can be bigger than 16 bytes. The kernel
will support these using the same infrastructure used for VARLEN CDB's.
So in effect MAX_COMMAND_SIZE means the maximum size command
scsi-ml supports without specifying a cmd_len by ULD's
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
None of these files use any of the functionality promised by
asm/semaphore.h. It's possible that they rely on it dragging in some
unrelated header file, but I can't build all these files, so we'll have
fix any build failures as they come up.
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Fix: The fact that nodes had different gap counts would be overlooked
if the bus manager code would pick gap count 63 because of beta
repeaters or because of very large hop counts. In this case, the bus
manager code would miss that it actually has to send the PHY config
packet with gap count 63.
Related trivial changes: Use bool for an int used as bool, touch up
some comments.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
We now exit fw_send_phy_config /after/ the PHY config packet has been
transmitted, instead of before. A subsequent fw_core_initiate_bus_reset
will therefore not overlap with the transmission. This is meant to make
the send PHY config packet + reset bus routine more deterministic.
Fixes bus reset loop and eventual panic with
- VIA VT6307 + IOGEAR hub + Unibrain Fire-i camera
http://bugzilla.kernel.org/show_bug.cgi?id=10128
- JMicron card
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Signed-off-by: Jarod Wilson <jwilson@redhat.com>
Stefan Richter