The following race condition still existed:
P1 P2
cancel_work_sync()
hci_uart_tx_wakeup()
hci_uart_write_work()
hci_uart_dequeue()
clear_bit(HCI_UART_PROTO_READY)
hci_unregister_dev(hdev)
hci_free_dev(hdev)
hu->proto->close(hu)
kfree(hu)
access to hdev and hu
Cancelling the work after clearing the HCI_UART_PROTO_READY bit avoids
this as any hci_uart_tx_wakeup() issued after the flag is cleared will
detect that and not schedule further work.
Signed-off-by: Ronald Tschalär <ronald@innovation.ch>
Reviewed-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Commit dec2c92880 ("Bluetooth: hci_ldisc:
Use rwlocking to avoid closing proto races") introduced locks in
hci_ldisc that are held while calling the proto functions. These locks
are rwlock's, and hence do not allow sleeping while they are held.
However, the proto functions that hci_bcm registers use mutexes and
hence need to be able to sleep.
In more detail: hci_uart_tty_receive() and hci_uart_dequeue() both
acquire the rwlock, after which they call proto->recv() and
proto->dequeue(), respectively. In the case of hci_bcm these point to
bcm_recv() and bcm_dequeue(). The latter both acquire the
bcm_device_lock, which is a mutex, so doing so results in a call to
might_sleep(). But since we're holding a rwlock in hci_ldisc, that
results in the following BUG (this for the dequeue case - a similar
one for the receive case is omitted for brevity):
BUG: sleeping function called from invalid context at kernel/locking/mutex.c
in_atomic(): 1, irqs_disabled(): 0, pid: 7303, name: kworker/7:3
INFO: lockdep is turned off.
CPU: 7 PID: 7303 Comm: kworker/7:3 Tainted: G W OE 4.13.2+ #17
Hardware name: Apple Inc. MacBookPro13,3/Mac-A5C67F76ED83108C, BIOS MBP133.8
Workqueue: events hci_uart_write_work [hci_uart]
Call Trace:
dump_stack+0x8e/0xd6
___might_sleep+0x164/0x250
__might_sleep+0x4a/0x80
__mutex_lock+0x59/0xa00
? lock_acquire+0xa3/0x1f0
? lock_acquire+0xa3/0x1f0
? hci_uart_write_work+0xd3/0x160 [hci_uart]
mutex_lock_nested+0x1b/0x20
? mutex_lock_nested+0x1b/0x20
bcm_dequeue+0x21/0xc0 [hci_uart]
hci_uart_write_work+0xe6/0x160 [hci_uart]
process_one_work+0x253/0x6a0
worker_thread+0x4d/0x3b0
kthread+0x133/0x150
We can't replace the mutex in hci_bcm, because there are other calls
there that might sleep. Therefore this replaces the rwlock's in
hci_ldisc with rw_semaphore's (which allow sleeping). This is a safer
approach anyway as it reduces the restrictions on the proto callbacks.
Also, because acquiring write-lock is very rare compared to acquiring
the read-lock, the percpu variant of rw_semaphore is used.
Lastly, because hci_uart_tx_wakeup() may be called from an IRQ context,
we can't block (sleep) while trying acquire the read lock there, so we
use the trylock variant.
Signed-off-by: Ronald Tschalär <ronald@innovation.ch>
Reviewed-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Fix a NULL pointer deref (hu->tty) when calling hci_uart_set_flow_control
on hci_uart-s using serdev.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Fixed alignment of all block comments.
Found using checkpatch
Signed-off-by: Derek Robson <robsonde@gmail.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
When HCI_UART_PROTO_READY is in the set state, the Data Link protocol
layer (proto) is bound to the HCI UART driver. This state allows the
registered proto function pointers to be used by the HCI UART driver.
When unbinding (closing) the Data Link protocol layer, the proto
function pointers much be prevented from being used immediately before
running the proto close function pointer. Otherwise, there is a risk
that a proto non-close function pointer is used during or after the
proto close function pointer is used. The consequences are likely to
be a kernel crash because the proto close function pointer will free
resources used in the Data Link protocol layer.
Therefore, add a reader writer lock (rwlock) solution to prevent the
close proto function pointer from running by using write_lock_irqsave()
whilst the other proto function pointers are protected using
read_lock(). This means HCI_UART_PROTO_READY can safely be cleared
in the knowledge that no proto function pointers are running.
When flag HCI_UART_PROTO_READY is put into the clear state,
proto close function pointer can safely be run. Note
flag HCI_UART_PROTO_SET being in the set state prevents the proto
open function pointer from being run so there is no race condition
between proto open and close function pointers.
Signed-off-by: Dean Jenkins <Dean_Jenkins@mentor.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Before attempting to schedule a work-item onto hu->write_work in
hci_uart_tx_wakeup(), check that the Data Link protocol layer is
still bound to the HCI UART driver.
Failure to perform this protocol check causes a race condition between
the work queue hu->write_work running hci_uart_write_work() and the
Data Link protocol layer being unbound (closed) in hci_uart_tty_close().
Note hci_uart_tty_close() does have a "cancel_work_sync(&hu->write_work)"
but it is ineffective because it cannot prevent work-items being added
to hu->write_work after cancel_work_sync() has run.
Therefore, add a check for HCI_UART_PROTO_READY into hci_uart_tx_wakeup()
which prevents scheduling of the work queue when HCI_UART_PROTO_READY
is in the clear state. However, note a small race condition remains
because the hci_uart_tx_wakeup() thread can run in parallel with the
hci_uart_tty_close() thread so it is possible that a schedule of
hu->write_work can occur when HCI_UART_PROTO_READY is cleared. A complete
solution needs locking of the threads which is implemented in a future
commit.
Signed-off-by: Dean Jenkins <Dean_Jenkins@mentor.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Before attempting to dequeue a Data Link protocol encapsulated message,
check that the Data Link protocol is still bound to the HCI UART driver.
This makes the code consistent with the usage of the other proto
function pointers.
Therefore, add a check for HCI_UART_PROTO_READY into hci_uart_dequeue()
and return NULL if the Data Link protocol is not bound.
This is needed for robustness as there is a scheduling race condition.
hci_uart_write_work() is scheduled to run via work queue hu->write_work
from hci_uart_tx_wakeup(). Therefore, there is a delay between
scheduling hci_uart_write_work() to run and hci_uart_dequeue() running
whereby the Data Link protocol layer could become unbound during the
scheduling delay. In this case, without the check, the call to the
unbound Data Link protocol layer dequeue function can crash.
It is noted that hci_uart_tty_close() has a
"cancel_work_sync(&hu->write_work)" statement but this only reduces
the window of the race condition because it is possible for a new
work-item to be added to work queue hu->write_work after the call to
cancel_work_sync(). For example, Data Link layer retransmissions can
be added to the work queue after the cancel_work_sync() has finished.
Signed-off-by: Dean Jenkins <Dean_Jenkins@mentor.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Before attempting to send a HCI message, check that the Data Link
protocol is still bound to the HCI UART driver. This makes the code
consistent with the usage of the other proto function pointers.
Therefore, add a check for HCI_UART_PROTO_READY into hci_uart_send_frame()
and return -EUNATCH if the Data Link protocol is not bound.
This also allows hci_send_frame() to report the error of an unbound
Data Link protocol layer. Therefore, it assists with diagnostics into
why HCI messages are being sent when the Data Link protocol is not
bound and avoids potential crashes.
Signed-off-by: Dean Jenkins <Dean_Jenkins@mentor.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Ensure that HCI_UART_PROTO_READY is cleared before close(hu) is
called which closes the Data Link protocol layer.
Therefore, add the missing bit clear of HCI_UART_PROTO_READY to
hci_uart_init_work() so that the flag is cleared when
hci_register_dev fails.
Without the fix, the functions of the Data Link protocol layer could
potentially be accessed after that layer has been closed. This
could lead to a crash as memory would have been freed in that layer.
Signed-off-by: Dean Jenkins <Dean_Jenkins@mentor.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
When hci_register_dev() fails, hu->hdev should be set to NULL before
freeing hdev. This avoids potential use of hu->hdev after it has been
freed.
This commit sets hu->hdev to NULL before calling hci_free_dev() in error
handling scenarios in hci_uart_init_work() and hci_uart_register_dev().
Signed-off-by: Dean Jenkins <Dean_Jenkins@mentor.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
If hci_register_dev() returns an error in hci_uart_init_work()
then the HCI_UART_REGISTERED bit gets erroneously set due to
a missing return statement. Therefore, add the missing return
statement.
The consequence of the missing return is that the HCI UART is not
registered but HCI_UART_REGISTERED is set which allows the code
to think that hu->hdev is safe to access but hu->hdev has been
freed so could lead to a crash.
Signed-off-by: Dean Jenkins <Dean_Jenkins@mentor.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
There are no users of hci_uart_init_tty, so remove it.
Signed-off-by: Rob Herring <robh@kernel.org>
Cc: Marcel Holtmann <marcel@holtmann.org>
Cc: Gustavo Padovan <gustavo@padovan.org>
Cc: Johan Hedberg <johan.hedberg@gmail.com>
Cc: linux-bluetooth@vger.kernel.org
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
For bluetooth protocol driver only supporting serdev it makes
sense to follow common practice and built them into their own
module.
Such modules need access to hci_uart_register_device and
hci_uart_tx_wakeup for using the common protocol helpers.
Signed-off-by: Sebastian Reichel <sre@kernel.org>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
This will be used by Nokia's H4+ protocol, which
uses 2-byte aligned packets.
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Sebastian Reichel <sre@kernel.org>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Remove multiple return statements in hci_uart_tty_ioctl() call and
added a single return statement.
This code re-organisation allows subsequent locking to be easily
added.
Signed-off-by: Vignesh Raman <Vignesh_Raman@mentor.com>
Signed-off-by: Dean Jenkins <Dean_Jenkins@mentor.com>
Signed-off-by: Rajeev Kumar <rajeev_kumar@mentor.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
This patch introduces support for Marvell Bluetooth controller over
UART (8897 for now). In order to send the final firmware at full speed,
a helper firmware is firstly sent. Firmware download is driven by the
controller which sends request firmware packets (including expected
size).
This driver is a global rework of the one proposed by
Amitkumar Karwar <akarwar@marvell.com>.
Signed-off-by: Loic Poulain <loic.poulain@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
The HCI_BREDR naming is confusing since it actually stands for Primary
Bluetooth Controller. Which is a term that has been used in the latest
standard. However from a legacy point of view there only really have
been Basic Rate (BR) and Enhanced Data Rate (EDR). Recent versions of
Bluetooth introduced Low Energy (LE) and made this terminology a little
bit confused since Dual Mode Controllers include BR/EDR and LE. To
simplify this the name HCI_PRIMARY stands for the Primary Controller
which can be a single mode or dual mode controller.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
HCI_UART_PROTO_SET flag is set before hci_uart_set_proto call. If we
receive data from tty layer during this procedure, proto pointer may
not be assigned yet, leading to null pointer dereference in rx method
hci_uart_tty_receive.
This patch fixes this issue by introducing HCI_UART_PROTO_READY flag in
order to avoid any proto operation before proto opening and assignment.
Signed-off-by: Loic Poulain <loic.poulain@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
This driver implements support for iBT2.1 Bluetooth controller embedded
in the AG620 communication combo. The controller needs to be configured
with bddata and can be patched with a binary patch file (pbn).
These operations are performed in manufacturing mode.
Signed-off-by: Loic Poulain <loic.poulain@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
The N_HCI ldisc does not define a flush_buffer() ldisc method, so
the check when opening the ldisc is always false.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
The new hci_skb_pkt_* wrappers are mainly intented for drivers to
require less knowledge about bt_cb(sbk) handling. So after converting
the core packet handling, convert all drivers.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Provide an early indication about the manufacturer information so that
it can be forwarded into monitor channel.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Setting and clearing of HCI_RUNNING flag in each and every driver is
just duplicating the same code all over the place. So instead of having
the driver do it in their hdev->open and hdev->close callbacks, set it
globally in the core transport handling.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
In all callbacks for hdev->send the status of HCI_RUNNING is checked. So
instead of repeating that code in every driver, move the check into the
hci_send_frame function before calling hdev->send.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
rx_lock spinlock is only used in hci_uart_tty_receive() which is the
receive_buf ldisc callback.
hci_uart_tty_receive() is protected from re-entrance by its only
caller (flush_to_ldisc() in drivers/tty/tty_buffer.c) which held a
mutex (buf->lock) for this section.
This lock allows "safe use of the line discipline's receive_buf()
method by excluding the buffer work and any pending flush from using
the flip buffer." (comments from tty_buffer_lock_exclusive() in
drivers/tty/tty_buffer.c)
So, no need to double protect this resource with rx_lock.
Signed-off-by: Frederic Danis <frederic.danis@linux.intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
QCA61x4 chips have supported sleep feature using In-Band-Sleep commands
to enable sleep feature based on H4 protocol. After sending
patch/nvm configuration is done, IBS mode will be up and running
Signed-off-by: Ben Young Tae Kim <ytkim@qca.qualcomm.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Fix for Cleanpatch warning:
Space prohibited between function name and open parenthesis '('
Signed-off-by: Prasanna Karthik <mkarthi3@visteon.com>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The Intel Lightning Peak devices do not come with Bluetooth firmware
loaded and thus require a full download of the operational Bluetooth
firmware when the device is attached via the Bluetooth line discipline.
Lightning Peak devices start with a bootloader mode that only accepts
a very limited set of HCI commands. The supported commands are enough
to identify the hardware and select the right firmware to load.
Signed-off-by: Loic Poulain <loic.poulain@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
hu->proto->*speed will always be used for all device where it is set.
But hu->*speed should be used if exist, so the test should be swapped.
The equivalent change is needed in bcm_setup() of hci_bcm.c.
Signed-off-by: Frederic Danis <frederic.danis@linux.intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Added the ability to flow control the UART, improved the UART baud
rate setting, transferred the speeds into line discipline from the
protocol and introduced the tty init function.
Signed-off-by: Ilya Faenson <ifaenson@broadcom.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Add initial and operational speeds.
Change to operational speed as soon as possible. If controller
set_baudrate() fails, continue at initial speed.
Signed-off-by: Frederic Danis <frederic.danis@linux.intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Move request/release_firmware() out of btbcm_patchram().
This allows a better error management, if request_firmware() returns an
error then the controller will be used without firmware loading and 0 is
returned.
This will imply to change btbcm_patchram() to accept a firmware instead
of firmware name.
Signed-off-by: Frederic Danis <frederic.danis@linux.intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
The Bluetooth address setting for Intel devices is provided by a generic
module now. Start using that module instead of relying it being included
in the driver.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
This adds the protocol support for Broadcom based UART devices to
enable firmware and patchram download procedure. It is a pretty
straight forward H:4 driver with the exception of actually having
its own setup and address configuration callbacks.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The new Broadcom Bluetooth support module provides generic functionality
for changing and checking the Bluetooth device address. Use these new
features instead of keeping a duplicate in the driver.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
This version number is more cosmetic and for debugging purposes, but
since there has been a few changes lately, increase it now.
Two left-over and not used version constants that were never exposed
anywhere are removed since they have no actual value.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The init function for each HCI UART protocol prints the same on success
and failure. This information is so generic, remove it and let the main
HCI UART handling print it instead.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The usage of struct hci_uart_proto should always be const. Change the
function headers and individual protocol drivers.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The TTY layer provides its data pointers as const, but the HCI UART
callbacks expect them as general data pointers. This is of course
wrong and instead of casting them, just fix the individual drivers
to actually take a const data pointer.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Casting the tty->disc_data to (void *) is not needed at all. So just
remove this cast and assign the object without a cast.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
When using vendor detection, this adds support for the Broadcom
specific address configuration command.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
When using vendor detection, this adds support for the Intel specific
address configuration command.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
This adds a new HCI_UART_VND_DETECT flag to allow automatic vendor
detection. This allows to enable known vendor commands (for example
for setting the public device address) when using a standard H:4
UART protocol or when running in virtual machines.
When this new flag is configured and no vendor specific setup
routine is provided, then the local version information are read
and the provided manufacturer information can be evaluated to
configure extra vendor callbacks.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Some Bluetooth controllers require initialization before being
used (vendor config, firmware download). Add possibility for a
HCI UART proto to implement this early init via the setup callback.
Signed-off-by: Loic Poulain <loic.poulain@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
The protocol for the UART might be configured, but that does not
mean the HCI device is registered. Return an error in that case
and only return the index number when HCI_UART_REGISTERED is set.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The quirk for enabling external configuration with UART needs to be
provided via the HCI UART flags. Add a new flag for it and declare
it as valid.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Providing unknown or invalid flags to the HCI UART driver should
result in an error. So check which flags are valid and otherwise
return an error.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
LDISCs shouldn't call tty->ops->write() from within
->write_wakeup().
->write_wakeup() is called with port lock taken and
IRQs disabled, tty->ops->write() will try to acquire
the same port lock and we will deadlock.
Acked-by: Marcel Holtmann <marcel@holtmann.org>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Reported-by: Huang Shijie <b32955@freescale.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Tested-by: Andreas Bießmann <andreas@biessmann.de>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
