In the tegra UART driver there are three places where the RX DMA buffer
is handled and pushed up to the tty layer. In all three instances the
same functions are called and so instead of duplicating the code in three
places, move this code to a new helper function and use this new function.
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The serial-tegra driver always uses DMA and hence the driver always
allocates DMA channels. Therefore, the test to see if the RX DMA channel
is initialised in tegra_uart_stop_rx() is unnecessary and so remove
the test and the code that corresponds to the case where the RX DMA
channel is not initialised. Please note that the call to
tegra_uart_stop_rx() should always be before the call to
tegra_uart_shutdown() which will uninitialise the RX DMA channel.
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Some functions in the serial-tegra driver have unnecessary return
statements at the end of a void function and so remove them.
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 853a699739 ("serial: tegra: handle race condition on uart rx
side") attempted to fix a race condition between the RX end of
transmission interrupt and RX DMA completion callback. Despite this
fix there is still another case where these two paths can race and
result in duplicated data. The race condition is as follows:
1. DMA completion interrupt occurs and schedules tasklet to call DMA
callback.
2. DMA callback for the UART driver starts to execute. This will copy
the data from the DMA buffer and restart the DMA. This is done under
uart port spinlock.
3. During the callback, UART interrupt is raised for end of receive. The
UART ISR runs and waits to acquire port spinlock held by the DMA
callback.
4. DMA callback gives up spinlock after copying the data, but before
restarting DMA.
5. UART ISR acquires the spin lock and reads the same DMA buffer because
DMA has not been restarted yet.
The release of the spinlock during the DMA callback was introduced by
commit 9b88748b36 ("tty: serial: tegra: drop uart_port->lock before
calling tty_flip_buffer_push()") to fix a spinlock lock-up issue when
calling tty_flip_buffer_push(). However, since then commit a9c3f68f3c
("tty: Fix low_latency BUG") migrated tty_flip_buffer_push() to always
use a workqueue, allowing tty_flip_buffer_push() to be called from
within atomic sections. Therefore, we can remove the unlocking of the
spinlock from the DMA callback and UART ISR and this will ensure that
the race condition no longer occurs.
Reported-by: Christopher Freeman <cfreeman@nvidia.com>
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Clearing UART_MCR_RTS or UART_MCR_XONANY is unnecessary; these bits
are never set in the shadow mcr. The RTS clear is especially confusing.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The same serial hardware is present on LS2080A which is arm64, and
LS1021A which is arm32, so don't limit the workaround to PPC.
Unlike PPC which uses arch/powerpc/kernel/legacy_serial.c, the ARM
targets use drivers/tty/serial/of_serial.c, so add the handle_irq
override check there as well.
Signed-off-by: Scott Wood <scottwood@freescale.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
There is no need to hardcode the MEN_Z135_MEM_SIZE. The MCB subsystem
already knowns the size which is located in the chameleon table.
MCB parse the chameleon table to get the resources of each IP and provide
the mcb_request_mem function to get those resources.
Use mcb_request_mem to get the resources. This function also takes care of
the memory region naming allocated by the driver for each of the instances.
Signed-off-by: Andreas Werner <andy@wernerandy.de>
Acked-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Turning on KVM and LPAE support on top of a multi_v7_defconfig will
produce a compiler warning in the Atmel serial driver:
drivers/tty/serial/atmel_serial.c: In function 'atmel_verify_port':
drivers/tty/serial/atmel_serial.c:2299:6: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
if ((void *)port->mapbase != ser->iomem_base)
^
Fix that by using the cast on the right hand side instead, as similar
code already does in other drivers.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We want the tty fixes and reverts in here as well so that people can
properly test and use it.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This is mostly a hardening fix, given that write-only access to other
users' ttys is usually only given through setgid tty executables.
Signed-off-by: Jann Horn <jann@thejh.net>
Cc: stable@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This adds an entry to the uart_config table for PORT_RT2880
enabling rx/tx FIFOs. The UART is actually a Palmchip BK-3103
which is found in several devices from Alchemy/RMI, Ralink, and
Sigma Designs.
Signed-off-by: Mans Rullgard <mans@mansr.com>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
UART controller is capable to perform transfers up to 4 Mbps.
Remove artificial 115.2 Kbps limitation.
Signed-off-by: Ivan T. Ivanov <ivan.ivanov@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Add receive DMA support for UARTDM type of controllers.
Tested on APQ8064, which have UARTDM v1.3 and ADM DMA engine
and APQ8016, which have UARTDM v1.4 and BAM DMA engine.
Signed-off-by: Ivan T. Ivanov <ivan.ivanov@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Add transmit DMA support for UARTDM type of controllers.
Tested on APQ8064, which have UARTDM v1.3 and ADM DMA engine
and APQ8016, which have UARTDM v1.4 and BAM DMA engine.
Signed-off-by: Ivan T. Ivanov <ivan.ivanov@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Make function naming consistent across this driver.
Also rename msm_irq to msm_uart_irq. No functional changes.
Signed-off-by: Ivan T. Ivanov <ivan.ivanov@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The bit masks for RFR_LEVEL1 and STALE_TIMEOUT_MSB values in MR1 and
IPR registers respectively are different for UART and UART_DM hardware
cores. We have been using UART core mask values for these. Add the same
for UART_DM core.
There is no bit setting as UART_IPR_RXSTALE_LAST for UART_DM core so do
it only for UART core.
Signed-off-by: Pramod Gurav <gpramod@codeaurora.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Ivan T. Ivanov <ivan.ivanov@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If a wrong compatible flag in specified, the of_match_device
returning null.
Implemented check and if NULL then returning -ENODEV.
Signed-off-by: Sean Nyekjaer <sean.nyekjaer@prevas.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
They're used by only if CONFIG_PM is enabled, so enclose them
with proper ifdefs. Removes these warnings:
drivers/tty/serial/omap-serial.c:202:12: warning: 'serial_omap_get_context_loss_count' defined but not used [-Wunused-function]
drivers/tty/serial/omap-serial.c:213:13: warning: 'serial_omap_enable_wakeup' defined but not used [-Wunused-function]
Signed-off-by: Guido Martínez <guido@vanguardiasur.com.ar>
Signed-off-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This ensures the dma mask that is supported by the driver is recorded
in the device structure.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
RX buffer is allocated from DMA coherent memory. Thus there is no need to call
DMA sync API for it.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Move h/w reinit of serial console restore-from-suspend into
standalone helper function.
No functional change.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When the UART clock is set slightly under 1.8432MHz, the 8250 driver
core doesn't permit the 115200 baud rate since it calculates the maximum
frequency to pass to uart_get_baud_rate by simply dividing the uart
clock by 16 which yields a value slightly under 115200, even though the
frequency is close enough for the UART to operate reliably.
Therefore add some tolerance in the calculation of the maximum baud
rate. 1% tolerance allows for marginally slower uart clk than nominal
without introducing transmission errors.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
[pjh: Forward-port & refactor original patch; change tolerance to 1%]
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
There are three natural ways in which devices may be wired to the system:
little endian (device receives correctly ordered bits of a word written
by little-endian CPU to its register, but big-endian CPU needs to swap
bytes of a word before writing it), big endian (same, but with big-endian
CPU in more favourable position) and native endian (CPU of either
endianness may do word-sized I/O without need for byteswapping).
Adding an option for native endianness allows using single kernel command
line for boards with native-endian serial ports on bi-endian
architectures. This goes in parallel with 'native-endian' DTS attribute.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
These drivers doesn't claim the serial device to be wakeup source. Even
if it is, it needs to use enable_irq_wake or other related PM wakeup
APIs to enable it.
This patch removes yet another misuse of IRQF_NO_SUSPEND.
Cc: Srinivas Kandagatla <srinivas.kandagatla@gmail.com>
Cc: Maxime Coquelin <maxime.coquelin@st.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: linux-arm-kernel@lists.infradead.org
Cc: kernel@stlinux.com
Cc: linux-serial@vger.kernel.org
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Since the FD interrupt handler can discern spurious IRQs and it is
shared with timer interrupt, use IRQF_COND_SUSPEND instead of
IRQF_NO_SUSPEND.
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Jiri Slaby <jslaby@suse.com>
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The function can return negative values.
The problem has been detected using proposed semantic patch
scripts/coccinelle/tests/unsigned_lesser_than_zero.cocci [1].
[1]: http://permalink.gmane.org/gmane.linux.kernel/2038576
Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This platform driver has a OF device ID table but the OF module
alias information is not created so module autoloading won't work.
Signed-off-by: Luis de Bethencourt <luisbg@osg.samsung.com>
Acked-by: Timur Tabi <timur@tabi.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This platform driver has a OF device ID table but the OF module
alias information is not created so module autoloading won't work.
Signed-off-by: Luis de Bethencourt <luisbg@osg.samsung.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This platform driver has a OF device ID table but the OF module
alias information is not created so module autoloading won't work.
Signed-off-by: Luis de Bethencourt <luisbg@osg.samsung.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This platform driver has a OF device ID table but the OF module
alias information is not created so module autoloading won't work.
Signed-off-by: Luis de Bethencourt <luisbg@osg.samsung.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Some SoCs, including Ralink/Mediatek and Alchemy Au1xxx, have a
16550-like UART with a non-standard register layout. These are
supported by a simple mapping table in 8250_port.c Rather than
list every SoC type using this access mode in the ifdefs there,
allow selecting the SERIAL_8250_RT288X Kconfig option with any
system and default it to y for the known cases needing it. The
help text is reworded accordingly.
This change simplifies adding support for other SoCs also using
the same UART.
The name of the option is a little misleading, but not knowing
the true origin of this UART, it is as good a choice as any.
Signed-off-by: Mans Rullgard <mans@mansr.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Race on buffer data happens when newly committed data is
picked up by an old flush work in the following scenario:
__tty_buffer_request_room does a plain write of tail->commit,
no barriers were executed before that.
At this point flush_to_ldisc reads this new value of commit,
and reads buffer data, no barriers in between.
The committed buffer data is not necessary visible to flush_to_ldisc.
Similar bug happens when tty_schedule_flip commits data.
Update commit with smp_store_release and read commit with
smp_load_acquire, as it is commit that signals data readiness.
This is orthogonal to the existing synchronization on tty_buffer.next,
which is required to not dismiss a buffer with unconsumed data.
The data race was found with KernelThreadSanitizer (KTSAN).
Signed-off-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
tty_buffer_flush frees not acquired buffers.
As the result, for example, read of b->size in tty_buffer_free
can return garbage value which will lead to a huge buffer
hanging in the freelist. This is just the benignest
manifestation of freeing of a not acquired object.
If the object is passed to kfree, heap can be corrupted.
Acquire visibility over the buffer before freeing it.
The data race was found with KernelThreadSanitizer (KTSAN).
Signed-off-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
flush_to_ldisc reads port->itty and checks that it is not NULL,
concurrently release_tty sets port->itty to NULL. It is possible
that flush_to_ldisc loads port->itty once, ensures that it is
not NULL, but then reloads it again and uses. The second load
can already return NULL, which will cause a crash.
Use READ_ONCE to read port->itty.
The data race was found with KernelThreadSanitizer (KTSAN).
Signed-off-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
My colleague ran into a program stall on a x86_64 server, where
n_tty_read() was waiting for data even if there was data in the buffer
in the pty. kernel stack for the stuck process looks like below.
#0 [ffff88303d107b58] __schedule at ffffffff815c4b20
#1 [ffff88303d107bd0] schedule at ffffffff815c513e
#2 [ffff88303d107bf0] schedule_timeout at ffffffff815c7818
#3 [ffff88303d107ca0] wait_woken at ffffffff81096bd2
#4 [ffff88303d107ce0] n_tty_read at ffffffff8136fa23
#5 [ffff88303d107dd0] tty_read at ffffffff81368013
#6 [ffff88303d107e20] __vfs_read at ffffffff811a3704
#7 [ffff88303d107ec0] vfs_read at ffffffff811a3a57
#8 [ffff88303d107f00] sys_read at ffffffff811a4306
#9 [ffff88303d107f50] entry_SYSCALL_64_fastpath at ffffffff815c86d7
There seems to be two problems causing this issue.
First, in drivers/tty/n_tty.c, __receive_buf() stores the data and
updates ldata->commit_head using smp_store_release() and then checks
the wait queue using waitqueue_active(). However, since there is no
memory barrier, __receive_buf() could return without calling
wake_up_interactive_poll(), and at the same time, n_tty_read() could
start to wait in wait_woken() as in the following chart.
__receive_buf() n_tty_read()
------------------------------------------------------------------------
if (waitqueue_active(&tty->read_wait))
/* Memory operations issued after the
RELEASE may be completed before the
RELEASE operation has completed */
add_wait_queue(&tty->read_wait, &wait);
...
if (!input_available_p(tty, 0)) {
smp_store_release(&ldata->commit_head,
ldata->read_head);
...
timeout = wait_woken(&wait,
TASK_INTERRUPTIBLE, timeout);
------------------------------------------------------------------------
The second problem is that n_tty_read() also lacks a memory barrier
call and could also cause __receive_buf() to return without calling
wake_up_interactive_poll(), and n_tty_read() to wait in wait_woken()
as in the chart below.
__receive_buf() n_tty_read()
------------------------------------------------------------------------
spin_lock_irqsave(&q->lock, flags);
/* from add_wait_queue() */
...
if (!input_available_p(tty, 0)) {
/* Memory operations issued after the
RELEASE may be completed before the
RELEASE operation has completed */
smp_store_release(&ldata->commit_head,
ldata->read_head);
if (waitqueue_active(&tty->read_wait))
__add_wait_queue(q, wait);
spin_unlock_irqrestore(&q->lock,flags);
/* from add_wait_queue() */
...
timeout = wait_woken(&wait,
TASK_INTERRUPTIBLE, timeout);
------------------------------------------------------------------------
There are also other places in drivers/tty/n_tty.c which have similar
calls to waitqueue_active(), so instead of adding many memory barrier
calls, this patch simply removes the call to waitqueue_active(),
leaving just wake_up*() behind.
This fixes both problems because, even though the memory access before
or after the spinlocks in both wake_up*() and add_wait_queue() can
sneak into the critical section, it cannot go past it and the critical
section assures that they will be serialized (please see "INTER-CPU
ACQUIRING BARRIER EFFECTS" in Documentation/memory-barriers.txt for a
better explanation). Moreover, the resulting code is much simpler.
Latency measurement using a ping-pong test over a pty doesn't show any
visible performance drop.
Signed-off-by: Kosuke Tatsukawa <tatsu@ab.jp.nec.com>
Cc: stable@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If atmel_init_gpios fails the port has already been marked as busy (in
line 2629), so this must be undone in the error path.
This bug was introduced because I created the patch that finally
became 722ccf416a ("serial: atmel: fix error handling when
mctrl_gpio_init fails") on top of 3.19 which didn't have commit
6fbb9bdf0f ("tty/serial: at91: fix error handling in
atmel_serial_probe()") yet.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Fixes: 722ccf416a ("serial: atmel: fix error handling when mctrl_gpio_init fails")
Acked-by: Nicolas Ferre <nicolas.ferre@atmel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit a3a10ce342 ("Avoid usb reset crashes by making tty_io cdevs truly
dynamic") which mixes using cdev_alloc() and cdev_init() is problematic.
Subsequent call to cdev_init() after cdev_alloc() sets kobj release method
from cdev_dynamic_release() to cdev_default_release() and thus makes it
impossible to free allocated cdev.
This patch also consolidates error path of cdev_add() as cdev can also leak
here if things went wrong.
Signed-off-by: Leon Yu <chianglungyu@gmail.com>
Fixes: a3a10ce342 ("Avoid usb reset crashes by making tty_io cdevs truly dynamic")
Acked-by: Richard Watts <rrw@kynesim.co.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This reverts commit 9e7b399d65.
Commit ("9e7b399d6528ea") causes the following warning and sometimes
also hangs the system:
------------[ cut here ]------------
WARNING: CPU: 0 PID: 0 at kernel/locking/mutex.c:868 mutex_trylock+0x20c/0x22c()
DEBUG_LOCKS_WARN_ON(in_interrupt())
Modules linked in:
CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.2.0-rc7-next-20150818-00001-g14418a6 #4
Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree)
Backtrace:
[<80012f08>] (dump_backtrace) from [<800130a4>] (show_stack+0x18/0x1c)
r6:00000364 r5:00000000 r4:00000000 r3:00000000
[<8001308c>] (show_stack) from [<807902b8>] (dump_stack+0x88/0xa4)
[<80790230>] (dump_stack) from [<8002a604>] (warn_slowpath_common+0x80/0xbc)
r5:807945c4 r4:80ab3b50
[<8002a584>] (warn_slowpath_common) from [<8002a6e4>] (warn_slowpath_fmt+0x38/0x40)
r8:00000000 r7:8131100c r6:8054c3cc r5:8131300c r4:80b0a570
[<8002a6b0>] (warn_slowpath_fmt) from [<807945c4>] (mutex_trylock+0x20c/0x22c)
r3:8095d0d8 r2:8095ab28
[<807943b8>] (mutex_trylock) from [<8054c3cc>] (clk_prepare_lock+0x14/0xf4)
r7:8131100c r6:be3f0c80 r5:00000037 r4:be3f0c80
[<8054c3b8>] (clk_prepare_lock) from [<8054dbfc>] (clk_prepare+0x18/0x30)
r5:00000037 r4:be3f0c80
[<8054dbe4>] (clk_prepare) from [<8036a600>] (imx_console_write+0x30/0x244)
r4:812d0bc8 r3:8132b9a4
To reproduce the problem we only need to let the board idle for something
like 30 seconds.
Tested on a imx6q-sabresd.
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Reviewed-by: Eduardo Valentin <edubezval@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This is done before adding more functionality to the init function with
the existing name. As this new functionality conflicts with stuff
drivers are required to implement themselves up to I want to convert
them one by one to make reviewing and reverting more easy in case I
broke something.
Once mctrl_gpio_init is there and all drivers are converted
mctrl_gpio_init_noauto can be removed again.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This commit also fixes compiler warnings and errors seen when building
on x86_64.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Acked-by: Nicolas Ferre <nicolas.ferre@atmel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The RX bytecount was only updated in the PIO path and thus
the device erroneously reported a value of 0 if DMA is in
use.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Jiada Wang <jiada_wang@mentor.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The commit enabling DMA support even if no flow control is present
was reverted on the grounds that it uncovered a number of bugs in
the code that lead to hanging tty devices and/or missing characters.
After tracking down the issues it is clear that those were generic
bugs and had nothing to do with flow control being present or not,
only that allowing DMA without hardware flow control increased
the exposure of that code a lot.
Now that those bugs are fixed, it should be safe to re-enable DMA
support.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Jiada Wang <jiada_wang@mentor.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The reference manual states that idle condition detect should not be used
with DMA transfers, as the ROM SDMA scripts don't check those conditions.
The RAM SDMA scripts worked around this, but the change broke compatibility
with the ROM scripts.
The previous commits fixed the DMA burst sizes, so that the aging timer is
now working as described in the reference manual. With this fixed we can
remove the hack of using the idle condition detect to stop the DMA transfer
if there are no new characters incoming.
This should work with both the ROM and RAM SDMA scripts.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Jiada Wang <jiada_wang@mentor.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Triggering the DMA engine for every byte is horribly inefficient.
Also it doesn't allow to use the aging timer for the RX FIFO as this
requires the DMA engine to leave one byte remaining in the FIFO when
doing a normal burst transfer.
Adjust watermark levels so that the DMA engine can do at least 8 byte
burst transfers. This is a conservative value, as the both TX and RX
FIFOs are able to contain 32 bytes.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Jiada Wang <jiada_wang@mentor.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Simplify the DMA restart logic to always queue up the next transfer
immediately if there is at least one more byte available in the FIFO,
so that the transfer will finish in a limited time.
This way the driver stops to rely on zero length transfers to signal
transfers ends. Those will go away when the idle detect DMA requests
are disabled.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Jiada Wang <jiada_wang@mentor.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The DMA transfer is only started once we are sure it will finish
in a limited time, i.e. only after we received a RRDY interrupt.
In order to allow the watermark level to be raised the aging
timer and the corresponding interrupt need to be set up as an
additional trigger, so that the transfer is also started if the
incoming amount of bytes never reach the watermark.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Jiada Wang <jiada_wang@mentor.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This function currently doesn't use its parameter.
Change prototype to pass in watermark levels, so we can reuse this
function in the DMA setup paths. Also relocate to be near the calling
functions.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Jiada Wang <jiada_wang@mentor.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Jon Hunter