GitHub/LineageOS/android_kernel_motorola_exynos9610.git
10 years agotty: serial: 8250_omap: add custom DMA-TX callback
Sebastian Andrzej Siewior [Mon, 29 Sep 2014 18:06:43 +0000 (20:06 +0200)]
tty: serial: 8250_omap: add custom DMA-TX callback

This patch provides mostly a copy of serial8250_tx_dma() +
__dma_tx_complete() with the following extensions:

- DMA bug
  At least on AM335x the following problem exists: Even if the TX FIFO is
  empty and a TX transfer is programmed (and started) the UART does not
  trigger the DMA transfer.
  After $TRESHOLD number of bytes have been written to the FIFO manually the
  UART reevaluates the whole situation and decides that now there is enough
  room in the FIFO and so the transfer begins.
  This problem has not been seen on DRA7 or beagle board xm (OMAP3). I am not
  sure if this is UART-IP core specific or DMA engine.

  The workaround is to use a threshold of one byte, program the DMA
  transfer minus one byte and then to put the first byte into the FIFO to
  kick start the transfer.

- support for runtime PM
  RPM is enabled on start_tx(). We can't disable RPM on DMA complete callback
  because there is still data in the FIFO which is being sent. We have to wait
  until the FIFO is empty before we disable it.
  For this to happen we fake a TX sent error and enable THRI. Once the
  FIFO is empty we receive an interrupt and since the TTY-buffer is still
  empty we "put RPM" via __stop_tx(). Should it been filed then in the
  start_tx() path we should program the DMA transfer and remove the error
  flag and the THRI bit.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: serial: 8250: allow to use custom DMA implementation
Sebastian Andrzej Siewior [Mon, 29 Sep 2014 18:06:42 +0000 (20:06 +0200)]
tty: serial: 8250: allow to use custom DMA implementation

The OMAP has a few corner cases where it needs a share of kindness of
affection to do the right thing. Heikki Krogerus suggested that instead
adding the quirks into the default DMA implementation, OMAP could get
its own copy of the function. And Alan suggested the same thing so here
we go.

This patch provides callbacks for custom TX/RX DMA implementation. If
there are not setup / used, then the default (current) implementation is
used.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: serial: 8250_dma: keep own book keeping about RX transfers
Sebastian Andrzej Siewior [Mon, 29 Sep 2014 18:06:41 +0000 (20:06 +0200)]
tty: serial: 8250_dma: keep own book keeping about RX transfers

After dmaengine_terminate_all() has been invoked then both DMA drivers
(edma and omap-dma) do not invoke dma_cookie_complete() to mark the
transfer as complete. This dma_cookie_complete() is performed by the
Synopsys DesignWare driver which is probably the only one that is used
by omap8250-dma and hence don't see following problem…
…which is that once a RX transfer has been terminated then following
query of channel status reports DMA_IN_PROGRESS (again: the actual
transfer has been canceled, there is nothing going on anymore).

This means that serial8250_rx_dma() never enqueues another DMA transfer
because it (wrongly) assumes that there is a transer already pending.

Vinod Koul refuses to accept a patch which adds this
dma_cookie_complete() to both drivers and so dmaengine_tx_status() would
report DMA_COMPLETE instead (and behave like the Synopsys DesignWare
driver already does). He argues that I am not allowed to use the cookie
to query the status and that the driver already cleaned everything up after
the invokation of dmaengine_terminate_all().

To end this I add a bookkeeping whether or not a RX-transfer has been
started to the 8250-dma code. It has already been done for the TX side.
*Now* we learn about the RX status based on our bookkeeping and don't
need dmaengine_tx_status() for this anymore.

Cc: vinod.koul@intel.com
Reviewed-by: Tony Lindgren <tony@atomide.com>
Tested-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: serial: 8250_dma: handle error on TX submit
Sebastian Andrzej Siewior [Mon, 29 Sep 2014 18:06:40 +0000 (20:06 +0200)]
tty: serial: 8250_dma: handle error on TX submit

Right now it is possible that serial8250_tx_dma() fails and returns
-EBUSY. The caller (serial8250_start_tx()) will then enable
UART_IER_THRI which will generate an interrupt once the TX FIFO is
empty.
In serial8250_handle_irq() nothing will happen because up->dma is set
and so serial8250_tx_chars() won't be invoked. We end up with plenty of
interrupts and some "too much work for irq" output.

This patch introduces dma_tx_err in struct uart_8250_port to signal that
the last invocation of serial8250_tx_dma() failed so we can fill the TX
FIFO manually. Should the next invocation of serial8250_start_tx()
succeed then the dma_tx_err flag along with the THRI bit is removed and
DMA only usage may continue.

Reviewed-by: Tony Lindgren <tony@atomide.com>
Tested-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: serial: Add 8250-core based omap driver
Sebastian Andrzej Siewior [Mon, 29 Sep 2014 18:06:39 +0000 (20:06 +0200)]
tty: serial: Add 8250-core based omap driver

This patch provides a 8250-core based UART driver for the internal OMAP
UART. The long term goal is to provide the same functionality as the
current OMAP uart driver and DMA support.
I tried to merge omap-serial code together with the 8250-core code.
There should should be hardly a noticable difference. The trigger levels
are different compared to omap-serial:
- omap serial
  TX: Interrupt comes after TX FIFO has room for 16 bytes.
      TX of 4096 bytes in one go results in 256 interrupts

  RX: Interrupt comes after there is on byte in the FIFO.
      RX of 4096 bytes results in 4096 interrupts.

- this driver
  TX: Interrupt comes once the TX FIFO is empty.
      TX of 4096 bytes results in 65 interrupts. That means there will
      be gaps on the line while the driver reloads the FIFO.

  RX: Interrupt comes once there are 48 bytes in the FIFO or less over
      "longer" time frame. We have
          1 / 11520 * 10^3 * 16 => 1.38… ms
      1.38ms to react and purge the FIFO on 115200,8N1. Since the other
      driver fired after each byte it had ~5.47ms time to react. This
      _may_ cause problems if one relies on no missing bytes and has no
      flow control. On the other hand we get only 85 interrupts for the
      same amount of data.

It has been only tested as console UART on am335x-evm, dra7-evm and
beagle bone. I also did some longer raw-transfers to meassure the load.

The device name is ttyS based instead of ttyO. If a ttyO based node name
is required please ask udev for it. If both driver are activated (this
and omap-serial) then this serial driver will take control over the
device due to the link order

v9…v10:
- Tony noticed that omap3 won't show anything after waking up
  from core off. In v9 I reworked the register restore and set
  IER to 0 by accident. This went unnoticed because start_tx
  usually sets ier (either due to DMA bug or due to TX-complete
  IRQ).
- dropped EFR and SLEEP from capabilities. We do have both but
  nobody should touch it. We already handle SLEEP ourself.
- make the private copy of the registers (like EFR) u8 instead
  u32
- drop MDR1 & DL[ML] reset in restore registers. Does not look
  required it is set to the required value later.
- update MDR1 & SCR only if changed.
- set MDR1 as the last thing. The errata says that we should
  setup everything before MDR1 set.
- avoid div by 0 in omap_8250_get_divisor() if baud rate gets
  very large (Frans Klaver fixed the same thing omap-serial)
- drop "is in early stage" from Kconfig.
v8…v9:
- less on a file seems to hang the am335x after a while. I
  believe I introduce this bug a while ago since I can reproduce
  this prior to v8. Fixed by redoing the omap8250_restore_regs()
v7…v8:
- redo the register write. There is now one function for that
  which is used from set_termios() and runtime-resume.
- drop PORT_OMAP_16750 and move the setup to the omap file. We
  have our own set termios function anyway (Heikki Krogerus)
- use MEM instead of MEM32. TRM of AM/DM37x says that 32bit
  access on THR might result in data abort. We only need 32bit
  access in the errata function which is before we use 8250's
  read function so it doesn't matter.
v4…v7:
- change trigger levels after some tests with raw transfers.
v3…v4:
- drop RS485 support
- wire up ->throttle / ->unthrottle
v2…v3:
- wire up startup & shutdown for wakeup-irq handling.
- RS485 handling (well the core does).

v1…v2:
- added runtime PM. Could somebody could please double check
  this?
- added omap_8250_set_termios()

Reviewed-by: Tony Lindgren <tony@atomide.com>
Tested-by: Tony Lindgren <tony@atomide.com>
Tested-by: Frans Klaver <frans.klaver@xsens.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: serial: 8250: make serial8250_console_setup() non _init
Sebastian Andrzej Siewior [Mon, 29 Sep 2014 18:06:38 +0000 (20:06 +0200)]
tty: serial: 8250: make serial8250_console_setup() non _init

if I boot with console=ttyS0 and the omap driver is module I end up with

| console [ttyS0] disabled
| omap8250 44e09000.serial: ttyS0 at MMIO 0x44e09000 (irq = 88, base_baud = 3000000) is a 8250
| Unable to handle kernel paging request at virtual address c07a9de0
| Modules linked in: 8250_omap(+)
| CPU: 0 PID: 908 Comm: modprobe Not tainted 3.17.0-rc5+ #1593
| PC is at serial8250_console_setup+0x0/0xc8
| LR is at register_console+0x13c/0x3a4
| [<c0078788>] (register_console) from [<c02d0340>] (uart_add_one_port+0x3cc/0x420)
| [<c02d0340>] (uart_add_one_port) from [<c02d38a4>] (serial8250_register_8250_port+0x298/0x39c)
| [<c02d38a4>] (serial8250_register_8250_port) from [<bf006274>] (omap8250_probe+0x218/0x3dc [8250_omap])
| [<bf006274>] (omap8250_probe [8250_omap]) from [<c02e3424>] (platform_drv_probe+0x2c/0x5c)
| [<c02e3424>] (platform_drv_probe) from [<c02e1eac>] (driver_probe_device+0x104/0x228)

| [<c009fa48>] (SyS_init_module) from [<c000e6e0>] (ret_fast_syscall+0x0/0x30)
| Code: 7823603b f8314620 051b3013 491ed416 (44792204)

because serial8250_console_setup() is already gone.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: serial: 8250: Fix wording in runtime-PM comments
Sebastian Andrzej Siewior [Mon, 29 Sep 2014 18:06:37 +0000 (20:06 +0200)]
tty: serial: 8250: Fix wording in runtime-PM comments

Frans reworded the two comments with better English for better
understanding. His review hit the mailing list after the patch got
applied so here is an incremental update.

Reported-by: Frans Klaver <frans.klaver@xsens.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agoserial: mxs-auart: add sysrq support
Janusz Uzycki [Fri, 10 Oct 2014 11:13:28 +0000 (13:13 +0200)]
serial: mxs-auart: add sysrq support

When using mxs-auart based console, sometime we need the sysrq function
to help debugging kernel. The sysrq code is basically there,
this patch just simply enable it.

Signed-off-by: Janusz Uzycki <j.uzycki@elproma.com.pl>
Reviewed-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: serial: omap: Remove probe error message
Markus Pargmann [Wed, 5 Nov 2014 12:20:50 +0000 (13:20 +0100)]
tty: serial: omap: Remove probe error message

This error message is not necessary. The driver core code will print all
probe error messages. It also resolves some error codes to proper error
messages. For example -EPROBE_DEFER will only be printed as an info message.

This patch removes the error message as the core prints the same
information.

Signed-off-by: Markus Pargmann <mpa@pengutronix.de>
Acked-by: Felipe Balbi <balbi@ti.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agoserial: core: Fix port count when uart_open() errors
Peter Hurley [Thu, 16 Oct 2014 20:54:27 +0000 (16:54 -0400)]
serial: core: Fix port count when uart_open() errors

A port count mismatch occurs if mutex_lock_interruptible()
exits uart_open() and the port has already been opened. This may
prematurely close a port on an open tty. Since uart_close() is _always_
called if uart_open() fails, the port count must be corrected if errors
occur.

Always increment the port count in uart_open(), regardless of errors;
always decrement the port count in uart_close(). Note that
tty_port_close_start() decrements the port count when uart_open()
was successful.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agoserial: core: Remove extra locking in uart_write()
Peter Hurley [Thu, 16 Oct 2014 20:54:26 +0000 (16:54 -0400)]
serial: core: Remove extra locking in uart_write()

uart_start() only claims the port->lock to call __uart_start(),
which does the actual processing. Eliminate the extra acquire/release
in uart_write(); call __uart_start() directly with port->lock already
held.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agoserial: core: Colocate crucial structure linkage
Peter Hurley [Thu, 16 Oct 2014 20:54:25 +0000 (16:54 -0400)]
serial: core: Colocate crucial structure linkage

The key function of uart_add_one_port() is to cross-reference the
UART driver's port structure with the serial core's state table;
keep the assignments together and document this crucial association.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agoserial: core: Remove redundant timeout assignments
Peter Hurley [Thu, 16 Oct 2014 20:54:24 +0000 (16:54 -0400)]
serial: core: Remove redundant timeout assignments

tty_port_init() initializes close_delay and closing_wait to these
same values; remove.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agoserial: core: Unwrap >80 char line in uart_close()
Peter Hurley [Thu, 16 Oct 2014 20:54:23 +0000 (16:54 -0400)]
serial: core: Unwrap >80 char line in uart_close()

The wrapped line looks wrong and out-of-place; leave it as
>80 char line.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Document defunct ASYNC_* bits in uapi header
Peter Hurley [Thu, 16 Oct 2014 20:54:22 +0000 (16:54 -0400)]
tty: Document defunct ASYNC_* bits in uapi header

Note the serial_struct flags for which the kernel ignores and performs
no action. The flags cannot be removed since they form part of the
userspace interface via the TIOCSSERIAL/TIOCGSERIAL ioctls.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty,serial: Unify UPF_* and ASYNC_* flag definitions
Peter Hurley [Thu, 16 Oct 2014 20:54:21 +0000 (16:54 -0400)]
tty,serial: Unify UPF_* and ASYNC_* flag definitions

The userspace-defined ASYNC_* flags in include/uapi/linux/tty_flags.h
are the authoritative bit definitions for the serial_struct flags,
and thus for any derivative values or fields.

Although the serial core provides the TIOCSSERIAL and TIOCGSERIAL
ioctls to set and retrieve these flags from userspace, it defines these
bits independently, as UPF_* macros.

Define the UPF_* macros which are userspace-modifiable directly from
the ASYNC_* symbolic constants. Add compile-time test to ensure the
bits changeable by TIOCSSERIAL match the defined range in the uapi
header.

Add ASYNCB_MAGIC_MULTIPLIER to the uapi header since this bit is
programmable by userspace.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agoserial: Fix locking for uart driver set_termios() method
Peter Hurley [Thu, 16 Oct 2014 20:54:20 +0000 (16:54 -0400)]
serial: Fix locking for uart driver set_termios() method

The low-level uart driver may modify termios settings to override
settings that are not compatible with the uart, such as CRTSCTS.
Thus, callers of the low-level uart driver's set_termios() method must
hold termios_rwsem write lock to prevent concurrent access to termios,
in case such override occurs.

The termios_rwsem lock requirement does not extend to console setup
(ie., uart_set_options), as console setup cannot race with tty
operations. Nor does this lock requirement extend to functions which
cannot be concurrent with tty ioctls (ie., uart_port_startup() and
uart_resume_port()).

Further, always claim the port mutex to protect hardware
re-reprogramming in the set_termios() uart driver method. Note this
is unnecessary for console initialization in uart_set_options()
which cannot be concurrent with other uart operations.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agoserial: core: Flush ldisc after dropping port mutex in uart_close()
Peter Hurley [Thu, 16 Oct 2014 20:54:19 +0000 (16:54 -0400)]
serial: core: Flush ldisc after dropping port mutex in uart_close()

The tty buffers (and any line discipline buffers) must be flushed after
the UART hardware has shutdown; otherwise, a racing open on the same
tty may receive data from the previous session, which is a security
hazard. However, holding the port mutex while flushing the line
discipline buffers creates a lock inversion if the set_termios()
handler takes the port mutex (as it does in the followup patch,
'serial: Fix locking for uart driver set_termios method'.

Flush the ldisc buffers after dropping the port mutex; the tty lock
is still held which prevents a concurrent open() from advancing while
flushing. Since no new rx data is possible after uart_shutdown() until
a new open reinitializes the port, the later flush has no impact on
what data is being discarded.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agoserial: Refactor uart_flush_buffer() from uart_close()
Peter Hurley [Thu, 16 Oct 2014 20:54:18 +0000 (16:54 -0400)]
serial: Refactor uart_flush_buffer() from uart_close()

In the context of the final tty & port close, flushing the tx
ring buffer after the hardware has already been shutdown and
the ring buffer freed is neither required nor desirable.

uart_flush_buffer() performs 3 operations:
1. Resets tx ring buffer indices, but the tx ring buffer has
   already been freed and the indices are reset if the port is
   re-opened.
2. Calls uart driver's flush_buffer() method
   5 in-tree uart drivers define flush_buffer() methods:
     amba-pl011, atmel-serial, imx, serial-tegra, timbuart
   These have been refactored into the shutdown() method, if
   required.
3. Kicks the ldisc for more writing, but this is undesirable.
   The file handle is being released; any waiting writer will
   will be kicked out by tty_release() with a warning. Further,
   the N_TTY ldisc may generate SIGIO for a file handle which
   is no longer valid.

Cc: Nicolas Ferre <nicolas.ferre@atmel.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Laxman Dewangan <ldewangan@nvidia.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Flush ldisc buffer atomically with tty flip buffers
Peter Hurley [Wed, 5 Nov 2014 17:13:09 +0000 (12:13 -0500)]
tty: Flush ldisc buffer atomically with tty flip buffers

tty_ldisc_flush() first clears the line discipline input buffer,
then clears the tty flip buffers. However, this allows for existing
data in the tty flip buffers to be added after the ldisc input
buffer has been cleared, but before the flip buffers have been cleared.

Add an optional ldisc parameter to tty_buffer_flush() to allow
tty_ldisc_flush() to pass the ldisc to clear.

NB: Initially, the plan was to do this automatically in
tty_buffer_flush(). However, an audit of the behavior of existing
line disciplines showed that performing a ldisc buffer flush on
ioctl(TCFLSH) was not always the outcome. For example, some line
disciplines have flush_buffer() methods but not ioctl() methods,
so a ->flush_buffer() command would be unexpected.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Fix timeout on pty set ldisc
Peter Hurley [Wed, 5 Nov 2014 17:13:08 +0000 (12:13 -0500)]
tty: Fix timeout on pty set ldisc

When changing the ldisc on one end of a pty pair, there may be
waiting readers/writers on the other end which may not exit from
the ldisc i/o loop, preventing tty_ldisc_lock_pair_timeout() from
acquiring the other side's ldisc lock.

Only acquire this side's ldisc lock; although this will no longer
prevent the other side from writing new input, that input will not
be processed until after the ldisc change completes. This has no
effect on normal ttys; new input from the driver was never disabled.

Remove tty_ldisc_enable_pair().

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Fix hung task on pty hangup
Peter Hurley [Wed, 5 Nov 2014 17:13:07 +0000 (12:13 -0500)]
tty: Fix hung task on pty hangup

When hanging up one end of a pty pair, there may be waiting
readers/writers on the other end which may not exit, preventing
tty_ldisc_lock_pair() from acquiring the other side's ldisc lock.

Only acquire this side's ldisc lock; although this will no longer
prevent the other side from writing new input, that input will not
be processing until after the ldisc hangup is complete.

Reported-by: Sasha Levin <sasha.levin@oracle.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Prefix tty_ldisc_{lock,lock_nested,unlock} functions
Peter Hurley [Wed, 5 Nov 2014 17:13:06 +0000 (12:13 -0500)]
tty: Prefix tty_ldisc_{lock,lock_nested,unlock} functions

tty_ldisc_lock(), tty_ldisc_unlock(), and tty_ldisc_lock_nested()
are low-level aliases for the underlying lock mechanism. Rename
with double underscore to allow for new, higher level functions
with those names.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agopty, n_tty: Simplify input processing on final close
Peter Hurley [Wed, 5 Nov 2014 17:13:05 +0000 (12:13 -0500)]
pty, n_tty: Simplify input processing on final close

When releasing one end of a pty pair, that end may just have written
to the other, which the input processing worker, flush_to_ldisc(), is
still working on but has not completed the copy to the other end's
read buffer. So input may not appear to be available to a waiting
reader but yet TTY_OTHER_CLOSED is now observed. The n_tty line
discipline has worked around this by waiting for input processing
to complete and then re-checking if input is available before
exiting with -EIO.

Since the tty/ldisc lock reordering, the wait for input processing
to complete can now occur during final close before setting
TTY_OTHER_CLOSED. In this way, a waiting reader is guaranteed to
see input available (if any) before observing TTY_OTHER_CLOSED.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agopty: Don't drop pty master tty lock to hangup slave
Peter Hurley [Wed, 5 Nov 2014 17:13:04 +0000 (12:13 -0500)]
pty: Don't drop pty master tty lock to hangup slave

With the revised tty lock order and lockdep annotation, claiming
the pty slave lock is now safe while still holding the pty master lock.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Remove tty_unhangup() declaration
Peter Hurley [Wed, 5 Nov 2014 17:13:03 +0000 (12:13 -0500)]
tty: Remove tty_unhangup() declaration

The tty_unhangup() function is not defined.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Preset lock subclass for nested tty locks
Peter Hurley [Wed, 5 Nov 2014 17:13:02 +0000 (12:13 -0500)]
tty: Preset lock subclass for nested tty locks

Eliminate the requirement of specifying the tty lock nesting at
lock time; instead, set the lock subclass for slave ptys at pty
install (normal ttys and master ptys use subclass 0).

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Change tty lock order to master->slave
Peter Hurley [Wed, 5 Nov 2014 17:13:01 +0000 (12:13 -0500)]
tty: Change tty lock order to master->slave

When releasing the master pty, the slave pty also needs to be locked
to prevent concurrent tty count changes for the slave pty and to
ensure that only one parallel master and slave release observe the
final close, and proceed to destruct the pty pair. Conversely, when
releasing the slave pty, locking the master pty is not necessary
(since the master's state can be inferred by the slave tty count).

Introduce tty_lock_slave()/tty_unlock_slave() which acquires/releases
the tty lock of the slave pty. Remove tty_lock_pair()/tty_unlock_pair().

Dropping the tty_lock is no longer required to re-establish a stable
lock order.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Simplify tty_release() state checks
Peter Hurley [Wed, 5 Nov 2014 17:13:00 +0000 (12:13 -0500)]
tty: Simplify tty_release() state checks

The local o_tty variable in tty_release() is now accessed only
when closing the pty master.

Set o_tty to slave pty when closing pty master, otherwise NULL;
use o_tty != NULL as replacement for pty_master.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Simplify tty_release_checks() interface
Peter Hurley [Wed, 5 Nov 2014 17:12:59 +0000 (12:12 -0500)]
tty: Simplify tty_release_checks() interface

Passing the 'other' tty to tty_release_checks() only makes sense
for a pty pair; make o_tty scope local instead.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Simplify tty_ldisc_release() interface
Peter Hurley [Wed, 5 Nov 2014 17:12:58 +0000 (12:12 -0500)]
tty: Simplify tty_ldisc_release() interface

Passing the 'other' tty to tty_ldisc_release() only makes sense
for a pty pair; make o_tty function local instead.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Fold pty pair handling into tty_flush_works()
Peter Hurley [Wed, 5 Nov 2014 17:12:57 +0000 (12:12 -0500)]
tty: Fold pty pair handling into tty_flush_works()

Perform work flush for both ends of a pty pair within tty_flush_works(),
rather than calling twice.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Simplify pty pair teardown logic
Peter Hurley [Wed, 5 Nov 2014 17:12:56 +0000 (12:12 -0500)]
tty: Simplify pty pair teardown logic

When the slave side closes and its tty count is 0, the pty
pair can be destroyed; the master side must have already
closed for the slave side tty count to be 0. Thus, only the
pty master close must check if the slave side has closed by
checking the slave tty count.

Remove the pre-computed closing flags and check the actual count(s).
Regular ttys are unaffected by this change.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Document check_tty_count() requires tty_lock held
Peter Hurley [Wed, 5 Nov 2014 17:12:55 +0000 (12:12 -0500)]
tty: Document check_tty_count() requires tty_lock held

Holding the tty_lock() is necessary to prevent concurrent changes
to the tty count that may cause it to differ from the open file
list count. The tty_lock() is already held at all call sites.

NB: Note that the check for the pty master tty count is safe because
the slave's tty_lock() is held while decrementing the pty master
tty count.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Don't release tty locks for wait queue sanity check
Peter Hurley [Wed, 5 Nov 2014 17:12:54 +0000 (12:12 -0500)]
tty: Don't release tty locks for wait queue sanity check

Releasing the tty locks while waiting for the tty wait queues to
be empty is no longer necessary nor desirable. Prior to
"tty: Don't take tty_mutex for tty count changes", dropping the
tty locks was necessary to reestablish the correct lock order between
tty_mutex and the tty locks. Dropping the global tty_mutex was necessary;
otherwise new ttys could not have been opened while waiting.

However, without needing the global tty_mutex held, the tty locks for
the releasing tty can now be held through the sleep. The sanity check
is for abnormal conditions caused by kernel bugs, not for recoverable
errors caused by misbehaving userspace; dropping the tty locks only
allows the tty state to get more sideways.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Don't take tty_mutex for tty count changes
Peter Hurley [Wed, 5 Nov 2014 17:12:53 +0000 (12:12 -0500)]
tty: Don't take tty_mutex for tty count changes

Holding tty_mutex is no longer required to serialize changes to
the tty_count or to prevent concurrent opens of closing ttys;
tty_lock() is sufficient.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Remove TTY_CLOSING
Peter Hurley [Wed, 5 Nov 2014 17:12:52 +0000 (12:12 -0500)]
tty: Remove TTY_CLOSING

Now that re-open is not permitted for a legacy BSD pty master,
using TTY_CLOSING to indicate when a tty can be torn-down is
no longer necessary.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Drop tty_mutex before tty reopen
Peter Hurley [Wed, 5 Nov 2014 17:12:51 +0000 (12:12 -0500)]
tty: Drop tty_mutex before tty reopen

Holding tty_mutex for a tty re-open is no longer necessary since
"tty: Clarify re-open behavior of master ptys". Because the
slave tty count is no longer accessed by tty_reopen(), holding
tty_mutex to prevent concurrent final tty_release() of the slave
pty is not required.

As with "tty: Re-open /dev/tty without tty_mutex", holding a
tty kref until the tty_lock is acquired is sufficient to ensure
the tty has not been freed, which, in turn, is sufficient to
ensure the tty_lock can be safely acquired and the tty count
can be safely retrieved. A non-zero tty count with the tty lock
held guarantees that release_tty() has not run and cannot
run concurrently with tty_reopen().

Change tty_driver_lookup_tty() to acquire the tty kref, which
allows the tty_mutex to be dropped before acquiring the tty lock.
Dropping the tty_mutex before attempting the tty_lock allows
other ttys to be opened and released, without needing this
tty_reopen() to complete.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Re-open /dev/tty without tty_mutex
Peter Hurley [Wed, 5 Nov 2014 17:12:50 +0000 (12:12 -0500)]
tty: Re-open /dev/tty without tty_mutex

Opening /dev/tty (ie., the controlling tty for the current task)
is always a re-open of the underlying tty. Because holding the
tty_lock is sufficient for safely re-opening a tty, and because
having a tty kref is sufficient for safely acquiring the tty_lock [1],
tty_open_current_tty() does not require holding tty_mutex.

Repurpose tty_open_current_tty() to perform the re-open itself and
refactor tty_open().

[1] Analysis of safely re-opening the current tty w/o tty_mutex

get_current_tty() gets a tty kref from the already kref'ed tty value of
current->signal->tty while holding the sighand lock for the current
task. This guarantees that the tty pointer returned from
get_current_tty() points to a tty which remains referenceable
while holding the kref.

Although release_tty() may run concurrently, and thus the driver
reference may be removed, release_one_tty() cannot have run, and
won't while holding the tty kref.

This, in turn, guarantees the tty_lock() can safely be acquired
(since tty->magic and tty->legacy_mutex are still a valid dereferences).
The tty_lock() also gets a tty kref to prevent the tty_unlock() from
dereferencing a released tty. Thus, the kref returned from
get_current_tty() can be released.

Lastly, the first operation of tty_reopen() is to check the tty count.
If non-zero, this ensures release_tty() is not running concurrently,
and the driver references have not been removed.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agopty: Always return -EIO if slave BSD pty opened first
Peter Hurley [Wed, 5 Nov 2014 17:12:49 +0000 (12:12 -0500)]
pty: Always return -EIO if slave BSD pty opened first

Opening the slave BSD pty first already returns -EIO from the slave
pty_open(), which in turn causes the newly installed tty pair to be
released before returning from tty_open(). However, this can also
cause a parallel master BSD pty open to fail because the pty pair
destruction may already been taking place in tty_release().

Failing at driver->install() if the slave pty is opened first ensures
that a pty master open cannot fail, because the driver tables will
not have been updated so tty_driver_lookup_tty() won't find the
master pty (and attempt to "re-open" it).

In turn, this guarantees that any tty with a tty->count == 0 is
in final close (rather than never opened).

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Check tty->count instead of TTY_CLOSING in tty_reopen()
Peter Hurley [Wed, 5 Nov 2014 17:12:48 +0000 (12:12 -0500)]
tty: Check tty->count instead of TTY_CLOSING in tty_reopen()

Although perhaps not obvious, the TTY_CLOSING bit is set when the
tty count has been decremented to 0 (which occurs while holding
tty_lock). The only other case when tty count is 0 during a re-open
is when a legacy BSD pty master has been opened in parallel but
after the pty slave, which is unsupported and returns an error.

Thus !tty->count contains the complete set of degenerate conditions
under which a tty open fails.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Clarify re-open behavior of master ptys
Peter Hurley [Wed, 5 Nov 2014 17:12:47 +0000 (12:12 -0500)]
tty: Clarify re-open behavior of master ptys

Re-opening master ptys is not allowed. Once opened and for the remaining
lifetime of the master pty, its tty count is 1. If its tty count has
dropped to 0, then the master pty was closed and TTY_CLOSING was set,
and destruction may begin imminently.

Besides the normal case of a legacy BSD pty master being re-opened
(which always returns -EIO), this code is only reachable in 2 degenerate
cases:
1. The pty master is the controlling terminal (this is possible through
   the TIOCSCTTY ioctl). pty masters are not designed to be controlling
   terminals and it's an oversight that tiocsctty() ever let that happen.
   The attempted open of /dev/tty will always fail. No known program does
   this.
2. The legacy BSD pty slave was opened first. The slave open will fail
   in pty_open() and tty_release() will commence. But before tty_release()
   claims the tty_mutex, there is a very small window where a parallel
   master open might succeed. In a test of racing legacy BSD slave and
   master parallel opens, where:
      slave open attempts:  10000   success:4527  failure:5473
      master open attempts: 11728   success:5789  failure:5939
   only 8 master open attempts would have succeeded reaching this code and
   successfully opened the master pty. This case is not possible with
   SysV ptys.

Always return -EIO if a master pty is re-opened or the slave is opened
first and the master opened in parallel (for legacy BSD ptys).

Furthermore, now that changing the slave's count is not required,
the tty_lock is sufficient for preventing concurrent changes to the
tty being re-opened (or failing re-opening).

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Remove TTY_HUPPING
Peter Hurley [Wed, 5 Nov 2014 17:12:46 +0000 (12:12 -0500)]
tty: Remove TTY_HUPPING

Now that tty_ldisc_hangup() does not drop the tty lock, it is no
longer possible to observe TTY_HUPPING while holding the tty lock
on another cpu.

Remove TTY_HUPPING bit definition.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Invert tty_lock/ldisc_sem lock order
Peter Hurley [Wed, 5 Nov 2014 17:12:45 +0000 (12:12 -0500)]
tty: Invert tty_lock/ldisc_sem lock order

Dropping the tty lock to acquire the tty->ldisc_sem allows several
race conditions (such as hangup while changing the ldisc) which requires
extra states and testing. The ldisc_sem->tty_lock lock order has
not been required since tty buffer ownership was moved to tty_port.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Don't hold tty_lock for ldisc release
Peter Hurley [Wed, 5 Nov 2014 17:12:44 +0000 (12:12 -0500)]
tty: Don't hold tty_lock for ldisc release

The tty->ldisc_sem write lock is sufficient for serializing changes
to tty->ldisc; holding the tty lock is not required.

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years ago8250_pci: Added driver for Endrun Technologies PTP PCIe card.
Mike Skoog [Thu, 16 Oct 2014 20:10:01 +0000 (13:10 -0700)]
8250_pci: Added driver for Endrun Technologies PTP PCIe card.

Added recognition of EndRun Technologies PCIe PTP slave card
and setup two ttySx ports for communication with the card for
retrieval of PTP based time and to communicate with the card's
Linux OS.

Signed-off-by: Mike Skoog <mskoog@endruntechnologies.com>
Signed-off-by: Mike Korreng <mkorreng@endruntechnologies.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agon_tty: Remove stale read lock comment
Peter Hurley [Thu, 16 Oct 2014 19:36:38 +0000 (15:36 -0400)]
n_tty: Remove stale read lock comment

The stale comment refers to lock behavior which was eliminated in
commit 6d76bd2618535c581f1673047b8341fd291abc67,
n_tty: Make N_TTY ldisc receive path lockless.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agon_tty: Only process packet mode data in raw mode
Peter Hurley [Thu, 16 Oct 2014 19:33:30 +0000 (15:33 -0400)]
n_tty: Only process packet mode data in raw mode

Packet mode can only be set for a pty master, and a pty master is
always in raw mode since its termios cannot be changed.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Fix missed wakeup from packet mode status update
Peter Hurley [Thu, 16 Oct 2014 19:33:29 +0000 (15:33 -0400)]
tty: Fix missed wakeup from packet mode status update

The pty master read() can miss the wake up for a packet mode
status change. For example,

CPU 0                                   | CPU 1
n_tty_read()                            | n_tty_packet_mode_flush()
  ...                                   |   .
  if (packet & link->ctrl_status) {     |   .
    /* no new ctrl_status ATM */        |   .
                                        |   spin_lock
                                        |     ctrl_status |= TIOCPKT_FLUSHREAD
                                        |   spin_unlock
                                        |   wake_up(link->read_wait)
  }                                     |
  set_current_state(TASK_INTERRUPTIBLE) |
  ...                                   |

The pty master read() will now sleep (assuming there is no input) having
missed the read_wait wakeup.

Set the task state before the condition test.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agopty: Hold ctrl_lock for packet mode updates
Peter Hurley [Thu, 16 Oct 2014 19:33:28 +0000 (15:33 -0400)]
pty: Hold ctrl_lock for packet mode updates

Updates to the packet mode enable require holding the ctrl_lock;
the serialization prevents corruption of adjacent fields.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agopty: Fix packet mode setting race
Peter Hurley [Thu, 16 Oct 2014 19:33:27 +0000 (15:33 -0400)]
pty: Fix packet mode setting race

Because pty_set_pktmode() does not claim the slave's ctrl_lock
to clear ->ctrl_status (to avoid unnecessary lock nesting),
pty_set_pktmode() may accidentally erase new ->ctrl_status updates.
For example,

CPU 0                             | CPU 1
pty_set_pktmode()                 | pty_start()
  spin_lock(master's ctrl_lock)   |
  tty->packet = 1                 |
                                  |   if (tty->link->packet)
                                  |     spin_lock(slave's ctrl_lock)
                                  |     tty->ctrl_status = TIOCPKT_START
  tty->link->ctrl_status = 0      |

Ensure the clear of ->ctrl_status occurs before packet mode is set
(and observable on another cpu).

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agopty: Don't claim slave's ctrl_lock for master's packet mode
Peter Hurley [Thu, 16 Oct 2014 19:33:26 +0000 (15:33 -0400)]
pty: Don't claim slave's ctrl_lock for master's packet mode

The slave's ctrl_lock serializes updates to the ctrl_status field
only, whereas the master's ctrl_lock serializes updates to the
packet mode enable (ie., the master does not have ctrl_status and
the slave does not have packet mode). Thus, claiming the slave's
ctrl_lock to access ->packet is useless.

Unlocked reads of ->packet are already smp-safe.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Use spin_lock_irq() for ctrl_lock when interrupts enabled
Peter Hurley [Thu, 16 Oct 2014 19:33:25 +0000 (15:33 -0400)]
tty: Use spin_lock_irq() for ctrl_lock when interrupts enabled

Interrupts are enabled in the n_tty_read() loop, ioctl(TIOCPKT)
and pty driver flush_buffer() routine; no need to save and restore
local interrupt state.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agopty: Use spin_lock_irq() for pty_set_termios()
Peter Hurley [Thu, 16 Oct 2014 19:33:24 +0000 (15:33 -0400)]
pty: Use spin_lock_irq() for pty_set_termios()

The tty driver's set_termios() method is called with interrupts
enabled; there is no need to save and restore the local interrupt state.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Move pty-specific set_termios() handling to pty driver
Peter Hurley [Thu, 16 Oct 2014 19:33:23 +0000 (15:33 -0400)]
tty: Move pty-specific set_termios() handling to pty driver

Packet mode is unique to the pty driver; move the packet mode state
change code from the generic tty ioctl handler to the pty driver.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: WARN for attempted set_termios() of pty master
Peter Hurley [Thu, 16 Oct 2014 19:33:22 +0000 (15:33 -0400)]
tty: WARN for attempted set_termios() of pty master

The pty master's termios should never be set; currently, all code
paths which call the driver's set_termios() method ensure that the
pty slave's termios is being set.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Update code comment in __proc_set_tty()
Peter Hurley [Thu, 16 Oct 2014 18:59:50 +0000 (14:59 -0400)]
tty: Update code comment in __proc_set_tty()

The session and foreground process group pid references will be
non-NULL if tiocsctty() is stealing the controlling tty from another
session (ie., arg == 1 in tiocsctty()).

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Serialize proc_set_tty() with tty_lock
Peter Hurley [Thu, 16 Oct 2014 18:59:49 +0000 (14:59 -0400)]
tty: Serialize proc_set_tty() with tty_lock

Setting the controlling terminal for a session occurs with either
the first open of a non-pty master tty or with ioctl(TIOCSCTTY).
Since only the session leader can set the controlling terminal for
a session (and the session leader cannot change), it is not
necessary to prevent a process from attempting to set different
ttys as the controlling terminal concurrently.

So it's only necessary to prevent the same tty from becoming the
controlling terminal for different session leaders. The tty_lock()
is sufficient to prevent concurrent proc_set_tty() for the same
tty.

Remove the tty_mutex lock region; add tty_lock() to tiocsctty().

While this may appear to allow a race condition between opening
the controlling tty via tty_open_current_tty() and stealing the
controlling tty via ioctl(TIOCSCTTY, 1), that race condition already
existed. Even if the tty_mutex prevented stealing the controlling tty
while tty_open_current_tty() returned the original controlling tty,
it cannot prevent stealing the controlling tty before tty_open() returns.
Thus, tty_open() could already return a no-longer-controlling tty when
opening /dev/tty.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Move session_of_pgrp() and make static
Peter Hurley [Thu, 16 Oct 2014 18:59:48 +0000 (14:59 -0400)]
tty: Move session_of_pgrp() and make static

tiocspgrp() is the lone caller of session_of_pgrp(); relocate and
limit to file scope.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Fix multiple races when setting the controlling terminal
Peter Hurley [Thu, 16 Oct 2014 18:59:47 +0000 (14:59 -0400)]
tty: Fix multiple races when setting the controlling terminal

Claim a read lock on the tasklist_lock while setting the controlling
terminal for the session leader. This fixes multiple races:
1. task_pgrp() and task_session() cannot be safely dereferenced, such
   as passing to get_pid(), without holding either rcu_read_lock() or
   tasklist_lock
2. setsid() unwisely allows any thread in the thread group to
   make the thread group leader the session leader; this makes the
   unlocked reads of ->signal->leader and signal->tty potentially
   unordered, stale or even have spurious values.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Remove !tty condition from __proc_set_tty()
Peter Hurley [Thu, 16 Oct 2014 18:59:46 +0000 (14:59 -0400)]
tty: Remove !tty condition from __proc_set_tty()

The tty parameter to __proc_set_tty() cannot be NULL; all
call sites have already dereferenced tty.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Replace open-coded tty_get_pgrp()
Peter Hurley [Thu, 16 Oct 2014 18:59:45 +0000 (14:59 -0400)]
tty: Replace open-coded tty_get_pgrp()

Replace open-coded instances of tty_get_pgrp().

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agouml: Fix unsafe pid reference to foreground process group
Peter Hurley [Thu, 16 Oct 2014 18:59:44 +0000 (14:59 -0400)]
uml: Fix unsafe pid reference to foreground process group

Although the tty core maintains a pid reference for the foreground
process group, if the foreground process group is changed that
pid reference is dropped. Thus, the pid reference used for signalling
could become stale.

Safely obtain a pid reference to the foreground process group and
release the reference after signalling is complete.

cc: Jeff Dike <jdike@addtoit.com>
Acked-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Remove tsk parameter from proc_set_tty()
Peter Hurley [Thu, 16 Oct 2014 18:59:43 +0000 (14:59 -0400)]
tty: Remove tsk parameter from proc_set_tty()

Only the current task itself can set its controlling tty (other
than before the task has been forked). Equivalent to existing usage.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Reorder proc_set_tty() and related fns
Peter Hurley [Thu, 16 Oct 2014 18:59:42 +0000 (14:59 -0400)]
tty: Reorder proc_set_tty() and related fns

Move the controlling tty-related functions and remove forward
declarations for __proc_set_tty() and proc_set_tty().

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Remove tty_pair_get_tty()/tty_pair_get_pty() api
Peter Hurley [Thu, 16 Oct 2014 18:59:41 +0000 (14:59 -0400)]
tty: Remove tty_pair_get_tty()/tty_pair_get_pty() api

tty_pair_get_pty() has no in-tree users and tty_pair_get_tty()
has only one file-local user. Remove the external declarations,
the export declarations, and declare tty_pair_get_tty() static.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: serial: 8250_core: restore the LCR register in set_sleep
Sudhir Sreedharan [Fri, 17 Oct 2014 12:39:18 +0000 (18:09 +0530)]
tty: serial: 8250_core: restore the LCR register in set_sleep

In ST16650V2 based serial uarts, while initalizing the PM state,
LCR registers are being initialized to 0 in serial8250_set_sleep().
If console port is already initialized and being used, this will
throws garbage in the console.

Signed-off-by: Sudhir Sreedharan <ssreedharan@mvista.com>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Tested-by: Kevin Hilman <khilman@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agoserial: of-serial: fix up PM ops on no_console_suspend and port type
Jingchang Lu [Wed, 15 Oct 2014 06:19:27 +0000 (14:19 +0800)]
serial: of-serial: fix up PM ops on no_console_suspend and port type

This patch fixes commit 2dea53bf57783f243c892e99c10c6921e956aa7e,
"serial: of-serial: add PM suspend/resume support", which disables
the uart clock on suspend, but also causes a hardware hang on register
access if no_console_suspend command line option is used.

Also, not every of_serial device is an 8250 port, so the serial8250
suspend/resume functions should only be applied to a real 8250 port.

Signed-off-by: Jingchang Lu <jingchang.lu@freescale.com>
Tested-by: Joseph Lo <josephl@nvidia.com>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agotty: Remove sparse lock annotations from tty_write_lock()/_unlock()
Peter Hurley [Thu, 16 Oct 2014 18:19:49 +0000 (14:19 -0400)]
tty: Remove sparse lock annotations from tty_write_lock()/_unlock()

sparse lock annotations cannot represent conditional acquire, such
as mutex_lock_interruptible() or mutex_trylock(), and produce sparse
warnings at _every_ correct call site.

Remove lock annotations from tty_write_lock() and tty_write_unlock().

Fixes sparse warnings:
drivers/tty/tty_io.c:1083:13: warning: context imbalance in 'tty_write_unlock' - wrong count at exit
drivers/tty/tty_io.c:1090:12: warning: context imbalance in 'tty_write_lock' - wrong count at exit
drivers/tty/tty_io.c:1211:17: warning: context imbalance in 'tty_write_message' - unexpected unlock
drivers/tty/tty_io.c:1233:16: warning: context imbalance in 'tty_write' - different lock contexts for basic block
drivers/tty/tty_io.c:1285:5: warning: context imbalance in 'tty_send_xchar' - different lock contexts for basic block
drivers/tty/tty_io.c:2653:12: warning: context imbalance in 'send_break' - different lock contexts for basic block

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agoserial: Fix sparse warnings in uart_throttle()/uart_unthrottle()
Peter Hurley [Thu, 16 Oct 2014 18:19:48 +0000 (14:19 -0400)]
serial: Fix sparse warnings in uart_throttle()/uart_unthrottle()

The struct uart_port.flags field is type upf_t, as are the matching
bit definitions. Change local mask variable to type upf_t.

Fixes sparse warnings:
drivers/tty/serial/serial_core.c:620:22: warning: invalid assignment: |=
drivers/tty/serial/serial_core.c:620:22:    left side has type unsigned int
drivers/tty/serial/serial_core.c:620:22:    right side has type restricted upf_t
drivers/tty/serial/serial_core.c:622:22: warning: invalid assignment: |=
drivers/tty/serial/serial_core.c:622:22:    left side has type unsigned int
drivers/tty/serial/serial_core.c:622:22:    right side has type restricted upf_t
drivers/tty/serial/serial_core.c:624:17: warning: restricted upf_t degrades to integer
drivers/tty/serial/serial_core.c:626:22: warning: invalid assignment: &=
drivers/tty/serial/serial_core.c:626:22:    left side has type unsigned int
drivers/tty/serial/serial_core.c:626:22:    right side has type restricted upf_t
drivers/tty/serial/serial_core.c:629:20: warning: restricted upf_t degrades to integer
drivers/tty/serial/serial_core.c:632:20: warning: restricted upf_t degrades to integer
drivers/tty/serial/serial_core.c:643:22: warning: invalid assignment: |=
drivers/tty/serial/serial_core.c:643:22:    left side has type unsigned int
drivers/tty/serial/serial_core.c:643:22:    right side has type restricted upf_t
drivers/tty/serial/serial_core.c:645:22: warning: invalid assignment: |=
drivers/tty/serial/serial_core.c:645:22:    left side has type unsigned int
drivers/tty/serial/serial_core.c:645:22:    right side has type restricted upf_t
drivers/tty/serial/serial_core.c:647:17: warning: restricted upf_t degrades to integer
drivers/tty/serial/serial_core.c:649:22: warning: invalid assignment: &=
drivers/tty/serial/serial_core.c:649:22:    left side has type unsigned int
drivers/tty/serial/serial_core.c:649:22:    right side has type restricted upf_t
drivers/tty/serial/serial_core.c:652:20: warning: restricted upf_t degrades to integer
drivers/tty/serial/serial_core.c:655:20: warning: restricted upf_t degrades to integer

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agoserial: Fix upstat_t sparse warnings
Peter Hurley [Thu, 16 Oct 2014 18:19:47 +0000 (14:19 -0400)]
serial: Fix upstat_t sparse warnings

Commit 299245a145b2ad4cfb4c5432eb1264299f55e7e0,
serial: core: Privatize modem status enable flags, introduced
the upstat_t type and matching bit definitions. The purpose is to
produce sparse warnings if the wrong bit definitions are used
(by warning of implicit integer conversions).

Fix implicit conversion to integer return type from uart_cts_enabled()
and uart_dcd_enabled().

Fixes the following sparse warnings:
drivers/tty/serial/serial_core.c:63:30: warning: incorrect type in return expression (different base types)
drivers/tty/serial/serial_core.c:63:30:    expected int
drivers/tty/serial/serial_core.c:63:30:    got restricted upstat_t
include/linux/serial_core.h:364:30: warning: incorrect type in return expression (different base types)
include/linux/serial_core.h:364:30:    expected bool
include/linux/serial_core.h:364:30:    got restricted upstat_t
include/linux/serial_core.h:364:30: warning: incorrect type in return expression (different base types)
include/linux/serial_core.h:364:30:    expected bool
include/linux/serial_core.h:364:30:    got restricted upstat_t

Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agogoldfish: off by one in goldfish_tty_console_setup()
Dan Carpenter [Wed, 29 Oct 2014 08:43:25 +0000 (11:43 +0300)]
goldfish: off by one in goldfish_tty_console_setup()

The goldfish_ttys[] array has "goldfish_tty_line_count" number of
elements.  It's allocated in goldfish_tty_create_driver().  This test
should be >= instead of >.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agogoldfish: fix sparse warnings
Fabian Frederick [Sun, 28 Sep 2014 18:10:17 +0000 (20:10 +0200)]
goldfish: fix sparse warnings

drivers/tty/goldfish.c:160:46: warning: Using plain integer as NULL pointer
drivers/tty/goldfish.c:320:22: warning: Using plain integer as NULL pointer

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
10 years agoLinux 3.18-rc3
Linus Torvalds [Sun, 2 Nov 2014 23:01:51 +0000 (15:01 -0800)]
Linux 3.18-rc3

10 years agoMerge tag 'for-linus-20141102' of git://git.infradead.org/linux-mtd
Linus Torvalds [Sun, 2 Nov 2014 22:45:52 +0000 (14:45 -0800)]
Merge tag 'for-linus-20141102' of git://git.infradead.org/linux-mtd

Pull MTD fixes from Brian Norris:
 "Three main MTD fixes for 3.18:

   - A regression from 3.16 which was noticed in 3.17.  With the
     restructuring of the m25p80.c driver and the SPI NOR library
     framework, we omitted proper listing of the SPI device IDs.  This
     means m25p80.c wouldn't auto-load (modprobe) properly when built as
     a module.  For now, we duplicate the device IDs into both modules.

   - The OMAP / ELM modules were depending on an implicit link ordering.
     Use deferred probing so that the new link order (in 3.18-rc) can
     still allow for successful probing.

   - Fix suspend/resume support for LH28F640BF NOR flash"

* tag 'for-linus-20141102' of git://git.infradead.org/linux-mtd:
  mtd: cfi_cmdset_0001.c: fix resume for LH28F640BF chips
  mtd: omap: fix mtd devices not showing up
  mtd: m25p80,spi-nor: Fix module aliases for m25p80
  mtd: spi-nor: make spi_nor_scan() take a chip type name, not spi_device_id
  mtd: m25p80: get rid of spi_get_device_id

10 years agoMerge tag 'scsi-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb...
Linus Torvalds [Sun, 2 Nov 2014 22:39:35 +0000 (14:39 -0800)]
Merge tag 'scsi-for-linus' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "This is a set of six patches consisting of:
   - two MAINTAINER updates
   - two scsi-mq fixs for the old parallel interface (not every request
     is tagged and we need to set the right flags to populate the SPI
     tag message)
   - a fix for a memory leak in scatterlist traversal caused by a
     preallocation update in 3.17
   - an ipv6 fix for cxgbi"

[ The scatterlist fix also came in separately through the block layer tree ]

* tag 'scsi-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  MAINTAINERS: ufs - remove self
  MAINTAINERS: change hpsa and cciss maintainer
  libcxgbi : support ipv6 address host_param
  scsi: set REQ_QUEUE for the blk-mq case
  Revert "block: all blk-mq requests are tagged"
  lib/scatterlist: fix memory leak with scsi-mq

10 years agoMerge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Sun, 2 Nov 2014 22:27:30 +0000 (14:27 -0800)]
Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux

Pull drm fixes from Dave Airlie:
 "Nothing too astounding or major: radeon, i915, vmwgfx, armada and
  exynos.

  Biggest ones:
   - vmwgfx has one big locking regression fix
   - i915 has come displayport fixes
   - radeon has some stability and a memory alloc failure
   - armada and exynos have some vblank fixes"

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux: (24 commits)
  drm/exynos: correct connector->dpms field before resuming
  drm/exynos: enable vblank after DPMS on
  drm/exynos: init kms poll at the end of initialization
  drm/exynos: propagate plane initialization errors
  drm/exynos: vidi: fix build warning
  drm/exynos: remove explicit encoder/connector de-initialization
  drm/exynos: init vblank with real number of crtcs
  drm/vmwgfx: Filter out modes those cannot be supported by the current VRAM size.
  drm/vmwgfx: Fix hash key computation
  drm/vmwgfx: fix lock breakage
  drm/i915/dp: only use training pattern 3 on platforms that support it
  drm/radeon: remove some buggy dead code
  drm/i915: Ignore VBT backlight check on Macbook 2, 1
  drm/radeon: remove invalid pci id
  drm/radeon: dpm fixes for asrock systems
  radeon: clean up coding style differences in radeon_get_bios()
  drm/radeon: Use drm_malloc_ab instead of kmalloc_array
  drm/radeon/dpm: disable ulv support on SI
  drm/i915: Fix GMBUSFREQ on vlv/chv
  drm/i915: Ignore long hpds on eDP ports
  ...

10 years agoMerge branch 'fixes' of git://ftp.arm.linux.org.uk/~rmk/linux-arm
Linus Torvalds [Sun, 2 Nov 2014 20:56:20 +0000 (12:56 -0800)]
Merge branch 'fixes' of git://ftp.arm.linux.org.uk/~rmk/linux-arm

Pull ARM fixes from Russell King:
 - add the new bpf syscall to ARM.
 - drop a redundant return statement in __iommu_alloc_remap()
 - fix a performance issue noticed by Thomas Petazzoni with
   kmap_atomic().
 - fix an issue with the L2 cache OF parsing code which caused it to
   incorrectly print warnings on each boot, and make the warning text
   more consistent with the rest of the code

* 'fixes' of git://ftp.arm.linux.org.uk/~rmk/linux-arm:
  ARM: 8180/1: mm: implement no-highmem fast path in kmap_atomic_pfn()
  ARM: 8183/1: l2c: Improve l2c310_of_parse() error message
  ARM: 8181/1: Drop extra return statement
  ARM: 8182/1: l2c: Make l2x0_cache_size_of_parse() return 'int'
  ARM: enable bpf syscall

10 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Sun, 2 Nov 2014 20:31:02 +0000 (12:31 -0800)]
Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull kvm fixes from Paolo Bonzini:
 "A small set of x86 fixes.  The most serious is an SRCU lockdep fix.

  A bit late - needed some time to test the SRCU fix, which only came in
  on Friday"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: vmx: defer load of APIC access page address during reset
  KVM: nVMX: Disable preemption while reading from shadow VMCS
  KVM: x86: Fix far-jump to non-canonical check
  KVM: emulator: fix execution close to the segment limit
  KVM: emulator: fix error code for __linearize

10 years agoMerge branch 'exynos-drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git...
Dave Airlie [Sun, 2 Nov 2014 19:23:17 +0000 (05:23 +1000)]
Merge branch 'exynos-drm-fixes' of git://git./linux/kernel/git/daeinki/drm-exynos into drm-fixes

This pull-request includes some bug fixes and code cleanups.
Especially, this fixes the bind failure issue occurred when it tries
to re-bind Exynos drm driver after unbound, and the modetest failure
issue incurred by not having a pair to vblank on and off requests.

* 'exynos-drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/daeinki/drm-exynos:
  drm/exynos: correct connector->dpms field before resuming
  drm/exynos: enable vblank after DPMS on
  drm/exynos: init kms poll at the end of initialization
  drm/exynos: propagate plane initialization errors
  drm/exynos: vidi: fix build warning
  drm/exynos: remove explicit encoder/connector de-initialization
  drm/exynos: init vblank with real number of crtcs

10 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Linus Torvalds [Sun, 2 Nov 2014 18:28:43 +0000 (10:28 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/viro/vfs

Pull VFS fixes from Al Viro:
 "A bunch of assorted fixes, most of them followups to overlayfs merge"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  ovl: initialize ->is_cursor
  Return short read or 0 at end of a raw device, not EIO
  isofs: don't bother with ->d_op for normal case
  isofs_cmp(): we'll never see a dentry for . or ..
  overlayfs: fix lockdep misannotation
  ovl: fix check for cursor
  overlayfs: barriers for opening upper-layer directory
  rcu: Provide counterpart to rcu_dereference() for non-RCU situations
  staging: android: logger: Fix log corruption regression

10 years agoirda: stop calling sk_prot->disconnect() on connection failure
Linus Torvalds [Sun, 2 Nov 2014 18:20:26 +0000 (10:20 -0800)]
irda: stop calling sk_prot->disconnect() on connection failure

The sk_prot is irda's own set of protocol handlers, so irda should
statically know what that function is anyway, without using an indirect
pointer.  And as it happens, we know *exactly* what that pointer is
statically: it's NULL, because irda doesn't define a disconnect
operation.

So calling that function is doubly wrong, and will just cause an oops.

Reported-by: Martin Lang <mlg.hessigheim@gmail.com>
Cc: Samuel Ortiz <samuel@sortiz.org>
Cc: David Miller <davem@davemloft.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agodrm/exynos: correct connector->dpms field before resuming
Andrzej Hajda [Fri, 10 Oct 2014 12:31:56 +0000 (14:31 +0200)]
drm/exynos: correct connector->dpms field before resuming

During system suspend after connector switch off its dpms field
is set to connector previous dpms state. To properly resume dpms field
should be set to its actual state (off) before resuming to previous dpms state.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
10 years agodrm/exynos: enable vblank after DPMS on
Andrzej Hajda [Fri, 10 Oct 2014 12:31:55 +0000 (14:31 +0200)]
drm/exynos: enable vblank after DPMS on

Before DPMS off driver disables vblank.
It should be balanced by vblank enable after DPMS on.
The patch fixes issue with page_flip ioctl not being able
to acquire vblank counter introduced by patch:
drm: Always reject drm_vblank_get() after drm_vblank_off()

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
10 years agodrm/exynos: init kms poll at the end of initialization
Andrzej Hajda [Fri, 10 Oct 2014 12:31:54 +0000 (14:31 +0200)]
drm/exynos: init kms poll at the end of initialization

HPD events can be generated by components even if drm_dev is not fully
initialized, to skip such events kms poll initialization should
be performed at the end of load callback followed directly by forced
connection detection.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
10 years agodrm/exynos: propagate plane initialization errors
Andrzej Hajda [Fri, 10 Oct 2014 12:31:53 +0000 (14:31 +0200)]
drm/exynos: propagate plane initialization errors

In case of error during plane initialization load callback
incorrectly return success, this patch fixes it.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
10 years agodrm/exynos: vidi: fix build warning
Inki Dae [Tue, 7 Oct 2014 15:16:34 +0000 (00:16 +0900)]
drm/exynos: vidi: fix build warning

encoder object isn't used anymore so remove it.

Signed-off-by: Inki Dae <inki.dae@samsung.com>
10 years agodrm/exynos: remove explicit encoder/connector de-initialization
Andrzej Hajda [Mon, 22 Sep 2014 09:30:48 +0000 (11:30 +0200)]
drm/exynos: remove explicit encoder/connector de-initialization

All KMS objects are destroyed by drm_mode_config_cleanup in proper order
so component drivers should not care about it.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
10 years agodrm/exynos: init vblank with real number of crtcs
Andrzej Hajda [Tue, 7 Oct 2014 13:09:14 +0000 (22:09 +0900)]
drm/exynos: init vblank with real number of crtcs

Initialization of vblank with MAX_CRTC caused attempts
to disabling vblanks for non-existing crtcs in case
drm used fewer crtcs. The patch fixes it.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
10 years agoKVM: vmx: defer load of APIC access page address during reset
Paolo Bonzini [Sun, 2 Nov 2014 06:54:30 +0000 (07:54 +0100)]
KVM: vmx: defer load of APIC access page address during reset

Most call paths to vmx_vcpu_reset do not hold the SRCU lock.  Defer loading
the APIC access page to the next vmentry.

This avoids the following lockdep splat:

[ INFO: suspicious RCU usage. ]
3.18.0-rc2-test2+ #70 Not tainted
-------------------------------
include/linux/kvm_host.h:474 suspicious rcu_dereference_check() usage!

other info that might help us debug this:

rcu_scheduler_active = 1, debug_locks = 0
1 lock held by qemu-system-x86/2371:
 #0:  (&vcpu->mutex){+.+...}, at: [<ffffffffa037d800>] vcpu_load+0x20/0xd0 [kvm]

stack backtrace:
CPU: 4 PID: 2371 Comm: qemu-system-x86 Not tainted 3.18.0-rc2-test2+ #70
Hardware name: Dell Inc. OptiPlex 9010/0M9KCM, BIOS A12 01/10/2013
 0000000000000001 ffff880209983ca8 ffffffff816f514f 0000000000000000
 ffff8802099b8990 ffff880209983cd8 ffffffff810bd687 00000000000fee00
 ffff880208a2c000 ffff880208a10000 ffff88020ef50040 ffff880209983d08
Call Trace:
 [<ffffffff816f514f>] dump_stack+0x4e/0x71
 [<ffffffff810bd687>] lockdep_rcu_suspicious+0xe7/0x120
 [<ffffffffa037d055>] gfn_to_memslot+0xd5/0xe0 [kvm]
 [<ffffffffa03807d3>] __gfn_to_pfn+0x33/0x60 [kvm]
 [<ffffffffa0380885>] gfn_to_page+0x25/0x90 [kvm]
 [<ffffffffa038aeec>] kvm_vcpu_reload_apic_access_page+0x3c/0x80 [kvm]
 [<ffffffffa08f0a9c>] vmx_vcpu_reset+0x20c/0x460 [kvm_intel]
 [<ffffffffa039ab8e>] kvm_vcpu_reset+0x15e/0x1b0 [kvm]
 [<ffffffffa039ac0c>] kvm_arch_vcpu_setup+0x2c/0x50 [kvm]
 [<ffffffffa037f7e0>] kvm_vm_ioctl+0x1d0/0x780 [kvm]
 [<ffffffff810bc664>] ? __lock_is_held+0x54/0x80
 [<ffffffff812231f0>] do_vfs_ioctl+0x300/0x520
 [<ffffffff8122ee45>] ? __fget+0x5/0x250
 [<ffffffff8122f0fa>] ? __fget_light+0x2a/0xe0
 [<ffffffff81223491>] SyS_ioctl+0x81/0xa0
 [<ffffffff816fed6d>] system_call_fastpath+0x16/0x1b

Reported-by: Takashi Iwai <tiwai@suse.de>
Reported-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Reviewed-by: Wanpeng Li <wanpeng.li@linux.intel.com>
Tested-by: Wanpeng Li <wanpeng.li@linux.intel.com>
Fixes: 38b9917350cb2946e368ba684cfc33d1672f104e
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agoKVM: nVMX: Disable preemption while reading from shadow VMCS
Jan Kiszka [Wed, 8 Oct 2014 16:05:39 +0000 (18:05 +0200)]
KVM: nVMX: Disable preemption while reading from shadow VMCS

In order to access the shadow VMCS, we need to load it. At this point,
vmx->loaded_vmcs->vmcs and the actually loaded one start to differ. If
we now get preempted by Linux, vmx_vcpu_put and, on return, the
vmx_vcpu_load will work against the wrong vmcs. That can cause
copy_shadow_to_vmcs12 to corrupt the vmcs12 state.

Fix the issue by disabling preemption during the copy operation.
copy_vmcs12_to_shadow is safe from this issue as it is executed by
vmx_vcpu_run when preemption is already disabled before vmentry.

This bug is exposed by running Jailhouse within KVM on CPUs with
shadow VMCS support.  Jailhouse never expects an interrupt pending
vmexit, but the bug can cause it if, after copy_shadow_to_vmcs12
is preempted, the active VMCS happens to have the virtual interrupt
pending flag set in the CPU-based execution controls.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agoKVM: x86: Fix far-jump to non-canonical check
Nadav Amit [Mon, 27 Oct 2014 22:03:43 +0000 (00:03 +0200)]
KVM: x86: Fix far-jump to non-canonical check

Commit d1442d85cc30 ("KVM: x86: Handle errors when RIP is set during far
jumps") introduced a bug that caused the fix to be incomplete.  Due to
incorrect evaluation, far jump to segment with L bit cleared (i.e., 32-bit
segment) and RIP with any of the high bits set (i.e, RIP[63:32] != 0) set may
not trigger #GP.  As we know, this imposes a security problem.

In addition, the condition for two warnings was incorrect.

Fixes: d1442d85cc30ea75f7d399474ca738e0bc96f715
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
[Add #ifdef CONFIG_X86_64 to avoid complaints of undefined behavior. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agoMerge branch 'vmwgfx-fixes-3.18' of git://people.freedesktop.org/~thomash/linux
Dave Airlie [Sat, 1 Nov 2014 23:23:31 +0000 (09:23 +1000)]
Merge branch 'vmwgfx-fixes-3.18' of git://people.freedesktop.org/~thomash/linux

A critical 3.18 regression fix from Rob, (thanks!)
A fix to avoid advertizing modes we can't support from Sinclair
  (welcome Sinclair!)
and a fix for an incorrect  hash key computation from me that is
  completely harmless, but can wait 'til the next merge window if necessary.
  (I can't really bother stable with this one).

* 'vmwgfx-fixes-3.18' of git://people.freedesktop.org/~thomash/linux:
  drm/vmwgfx: Filter out modes those cannot be supported by the current VRAM size.
  drm/vmwgfx: Fix hash key computation
  drm/vmwgfx: fix lock breakage

10 years agoMerge tag 'staging-3.18-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Sat, 1 Nov 2014 22:11:27 +0000 (15:11 -0700)]
Merge tag 'staging-3.18-rc3' of git://git./linux/kernel/git/gregkh/staging

Pull staging fixes from Greg KH:
 "Here are some staging driver fixes for 3.18-rc3.  Mostly iio and
  comedi driver fixes for issues reported by people.

  All of these have been in linux-next for a while with no reported
  issues"

* tag 'staging-3.18-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
  staging: comedi: fix memory leak / bad pointer freeing for chanlist
  staging: comedi: Kconfig: fix config COMEDI_ADDI_APCI_3120 dependants
  staging: comedi: widen subdevice number argument in ioctl handlers
  staging: rtl8723au: Fix alignment of mac_addr for ether_addr_copy() usage
  drivers/staging/comedi/Kconfig: Let COMEDI_II_PCI20KC depend on HAS_IOMEM
  staging: comedi: (regression) channel list must be set for COMEDI_CMD ioctl
  iio: adc: mxs-lradc: Disable the clock on probe failure
  iio: st_sensors: Fix buffer copy
  staging:iio:ad5933: Drop "raw" from channel names
  staging:iio:ad5933: Fix NULL pointer deref when enabling buffer

10 years agoMerge tag 'usb-3.18-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
Linus Torvalds [Sat, 1 Nov 2014 22:08:04 +0000 (15:08 -0700)]
Merge tag 'usb-3.18-rc3' of git://git./linux/kernel/git/gregkh/usb

Pull USB fixes from Greg KH:
 "Here are a bunch of USB fixes for 3.18-rc3.

  Mostly usb-serial device ids and gadget fixes for issues that have
  been reported.  Full details are in the shortlog.

  All of these have been in linux-next for a while"

* tag 'usb-3.18-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb: (42 commits)
  usb: chipidea: Fix oops when removing the ci_hdrc module
  usb: gadget: function: Fixed the return value on error path
  usb: dwc2: gadget: disable phy before turning off power regulators
  usb: gadget: function: Remove redundant usb_free_all_descriptors
  usb: dwc3: gadget: Properly initialize LINK TRB
  usb: dwc2: gadget: fix gadget unregistration in udc_stop() function
  usb: dwc2: Bits in bitfield should add up to 32
  usb: dwc2: gadget: sparse warning of context imbalance
  usb: gadget: udc: core: fix kernel oops with soft-connect
  usb: musb: musb_dsps: fix NULL pointer in suspend
  usb: musb: dsps: start OTG timer on resume again
  usb: gadget: loopback: don't queue requests to bogus endpoints
  usb: ffs: fix regression when quirk_ep_out_aligned_size flag is set
  usb: gadget: f_fs: remove redundant ffs_data_get()
  usb: gadget: udc: USB_GADGET_XILINX should depend on HAS_DMA
  Revert "usb: dwc3: dwc3-omap: Disable/Enable only wrapper interrupts in prepare/complete"
  usb: gadget: composite: enable BESL support
  usb: musb: cppi41: restart hrtimer only if not yet done
  usb: dwc3: ep0: fix Data Phase for transfer sizes aligned to wMaxPacketSize
  usb: serial: ftdi_sio: add "bricked" FTDI device PID
  ...

10 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux...
Linus Torvalds [Sat, 1 Nov 2014 17:41:26 +0000 (10:41 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/mason/linux-btrfs

Pull btrfs fixes from Chris Mason:
 "Filipe is nailing down some problems with our skinny extent variation,
  and Dave's patch fixes endian problems in the new super block checks"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs:
  Btrfs: fix race that makes btrfs_lookup_extent_info miss skinny extent items
  Btrfs: properly clean up btrfs_end_io_wq_cache
  Btrfs: fix invalid leaf slot access in btrfs_lookup_extent()
  btrfs: use macro accessors in superblock validation checks

10 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
Linus Torvalds [Sat, 1 Nov 2014 02:51:11 +0000 (19:51 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/dtor/input

Pull input updates from Dmitry Torokhov:
 "A bunch of fixes for minor defects reported by Coverity, a few driver
  fixups and revert of i8042.nomux change so that we are once again
  enable active MUX mode if box claims to support it"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Revert "Input: i8042 - disable active multiplexing by default"
  Input: altera_ps2 - use correct type for irq return value
  Input: altera_ps2 - write to correct register when disabling interrupts
  Input: max77693-haptic - fix potential overflow
  Input: psmouse - remove unneeded check in psmouse_reconnect()
  Input: vsxxxaa - fix code dropping bytes from queue
  Input: ims-pcu - fix dead code in ims_pcu_ofn_reg_addr_store()
  Input: opencores-kbd - fix error handling
  Input: wm97xx - adapt parameters to tosa touchscreen.
  Input: i8042 - quirks for Fujitsu Lifebook A544 and Lifebook AH544
  Input: stmpe-keypad - fix valid key line bitmask
  Input: soc_button_array - update calls to gpiod_get*()

10 years agoMerge tag 'pm+acpi-3.18-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
Linus Torvalds [Sat, 1 Nov 2014 02:08:25 +0000 (19:08 -0700)]
Merge tag 'pm+acpi-3.18-rc3' of git://git./linux/kernel/git/rafael/linux-pm

Pull ACPI and power management fixes from Rafael Wysocki:
 "These are fixes received after my previous pull request plus one that
  has been in the works for quite a while, but its previous version
  caused problems to happen, so it's been deferred till now.

  Fixed are two recent regressions (MFD enumeration and cpufreq-dt),
  ACPI EC regression introduced in 3.17, system suspend error code path
  regression introduced in 3.15, an older bug related to recovery from
  failing resume from hibernation and a cpufreq-dt driver issue related
  to operation performance points.

  Specifics:

   - Fix a crash on r8a7791/koelsch during resume from system suspend
     caused by a recent cpufreq-dt commit (Geert Uytterhoeven).

   - Fix an MFD enumeration problem introduced by a recent commit adding
     ACPI support to the MFD subsystem that exposed a weakness in the
     ACPI core causing ACPI enumeration to be applied to all devices
     associated with one ACPI companion object, although it should be
     used for one of them only (Mika Westerberg).

   - Fix an ACPI EC regression introduced during the 3.17 cycle causing
     some Samsung laptops to misbehave as a result of a workaround
     targeted at some Acer machines.  That includes a revert of a commit
     that went too far and a quirk for the Acer machines in question.
     From Lv Zheng.

   - Fix a regression in the system suspend error code path introduced
     during the 3.15 cycle that causes it to fail to take errors from
     asychronous execution of "late" suspend callbacks into account
     (Imre Deak).

   - Fix a long-standing bug in the hibernation resume error code path
     that fails to roll back everything correcty on "freeze" callback
     errors and leaves some devices in a "suspended" state causing more
     breakage to happen subsequently (Imre Deak).

   - Make the cpufreq-dt driver disable operation performance points
     that are not supported by the VR connected to the CPU voltage plane
     with acceptable tolerance instead of constantly failing voltage
     scaling later on (Lucas Stach)"

* tag 'pm+acpi-3.18-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  ACPI / EC: Fix regression due to conflicting firmware behavior between Samsung and Acer.
  Revert "ACPI / EC: Add support to disallow QR_EC to be issued before completing previous QR_EC"
  cpufreq: cpufreq-dt: Restore default cpumask_setall(policy->cpus)
  PM / Sleep: fix recovery during resuming from hibernation
  PM / Sleep: fix async suspend_late/freeze_late error handling
  ACPI: Use ACPI companion to match only the first physical device
  cpufreq: cpufreq-dt: disable unsupported OPPs

10 years agoMerge tag 'pci-v3.18-fixes-1' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaa...
Linus Torvalds [Sat, 1 Nov 2014 01:48:29 +0000 (18:48 -0700)]
Merge tag 'pci-v3.18-fixes-1' of git://git./linux/kernel/git/helgaas/pci

Pull PCI fixes from Bjorn Helgaas:
 "These changes, intended for v3.18, fix:

  Sysfs
    - Fix "enable" filename change (Greg Kroah-Hartman)

      An unintentional sysfs filename change in commit 5136b2da770d
      ("PCI: convert bus code to use dev_groups"), which appeared in
      v3.13, changed "enable" to "enabled", and this changes it back.

      Old users of "enable" are currently broken and will be helped by
      this change.  Anything that started to use "enabled" after v3.13
      will be broken by this change.  If necessary, we can add a symlink
      to make both work, but this patch doesn't do that.

  PCI device hotplug
    - Revert duplicate merge (Kamal Mostafa)

      A mistaken duplicate merge that added a check twice.  Nothing's
      broken; this just removes the unnecessary code.

  Freescale i.MX6
    - Wait for clocks to stabilize after ref_en (Richard Zhu)

      An i.MX6 clock problem that prevents mx6 nitrogen boards from booting"

* tag 'pci-v3.18-fixes-1' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci:
  PCI: Rename sysfs 'enabled' file back to 'enable'
  PCI: imx6: Wait for clocks to stabilize after ref_en
  Revert duplicate "PCI: pciehp: Prevent NULL dereference during probe"

10 years agox86_64, entry: Fix out of bounds read on sysenter
Andy Lutomirski [Sat, 1 Nov 2014 01:08:45 +0000 (18:08 -0700)]
x86_64, entry: Fix out of bounds read on sysenter

Rusty noticed a Really Bad Bug (tm) in my NT fix.  The entry code
reads out of bounds, causing the NT fix to be unreliable.  But, and
this is much, much worse, if your stack is somehow just below the
top of the direct map (or a hole), you read out of bounds and crash.

Excerpt from the crash:

[    1.129513] RSP: 0018:ffff88001da4bf88  EFLAGS: 00010296

  2b:*    f7 84 24 90 00 00 00     testl  $0x4000,0x90(%rsp)

That read is deterministically above the top of the stack.  I
thought I even single-stepped through this code when I wrote it to
check the offset, but I clearly screwed it up.

Fixes: 8c7aa698baca ("x86_64, entry: Filter RFLAGS.NT on entry from userspace")
Reported-by: Rusty Russell <rusty@ozlabs.org>
Cc: stable@vger.kernel.org
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>