Commit 3a8b36f378 ("Btrfs: fix data loss in the fast fsync path") added
a performance regression for that causes an unnecessary sync of the log
trees (fs/subvol and root log trees) when 2 consecutive fsyncs are done
against a file, without no writes or any metadata updates to the inode in
between them and if a transaction is committed before the second fsync is
called.
Huang Ying reported this to lkml (https://lkml.org/lkml/2015/3/18/99)
after a test sysbench test that measured a -62% decrease of file io
requests per second for that tests' workload.
The test is:
echo performance > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
echo performance > /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor
echo performance > /sys/devices/system/cpu/cpu2/cpufreq/scaling_governor
echo performance > /sys/devices/system/cpu/cpu3/cpufreq/scaling_governor
mkfs -t btrfs /dev/sda2
mount -t btrfs /dev/sda2 /fs/sda2
cd /fs/sda2
for ((i = 0; i < 1024; i++)); do fallocate -l 67108864 testfile.$i; done
sysbench --test=fileio --max-requests=0 --num-threads=4 --max-time=600 \
--file-test-mode=rndwr --file-total-size=68719476736 --file-io-mode=sync \
--file-num=1024 run
A test on kvm guest, running a debug kernel gave me the following results:
Without 3a8b36f378: 16.01 reqs/sec
With 3a8b36f378: 3.39 reqs/sec
With 3a8b36f378 and this patch: 16.04 reqs/sec
Reported-by: Huang Ying <ying.huang@intel.com>
Tested-by: Huang, Ying <ying.huang@intel.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
A previous commit wanted to make CFQ default to IOPS mode on
non-rotational storage, however it did so when the queue was
initialized and the non-rotational flag is only set later on
in the probe.
Add an elevator hook that gets called off the add_disk() path,
at that point we know that feature probing has finished, and
we can reliably check for the various flags that drivers can
set.
Fixes: 41c0126b ("block: Make CFQ default to IOPS mode on SSDs")
Tested-by: Romain Francoise <romain@orebokech.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
ts2020_attach() allocates a variable pdata on the stack and then passes a
pointer to it to i2c_new_device() which stashes the pointer in persistent
structures.
Add a comment to the effect that this isn't actually an error because the
contents of the variable are only used in ts2020_probe() and this is only
called ts2020_attach()'s stack frame exists.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Antti Palosaari <crope@iki.fi>
Signed-off-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>
Register driver using I2C bindings internally when legacy media
attach is used. That is done by registering driver using I2C binding
from legacy attach. That way we can get valid I2C client, which is
needed for proper dev_() logging and regmap for example even legacy
binding is used.
Signed-off-by: Antti Palosaari <crope@iki.fi>
Signed-off-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>
* We don't need calculate channel bandwidth from symbol rate as it
is calculated by DVB core.
* Use clamp() to force upper/lower limit of filter 3dB frequency.
Upper limit should never exceeded 40MHz (80MHz BW) in any case,
though...
Signed-off-by: Antti Palosaari <crope@iki.fi>
Signed-off-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>
Used frequency synthesizer is simple Integer-N PLL, with configurable
reference divider, output divider and of course N itself. Old
calculations were working fine, but not so easy to understand.
Signed-off-by: Antti Palosaari <crope@iki.fi>
Signed-off-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>
smu_init allocates buffers and initializes them. It does not
touch the hw. There is no need to do it again on resume. It
should really be part of sw_init (and smu_fini should be part
of sw_fini), but we need the firmware sizes from the other IPs
for firmware loading so we have to wait until sw init is done
for all other IPs.
Reviewed-by: Sonny Jiang <Sonny.Jiang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
smu_init allocates buffers and initializes them. It does not
touch the hw. There is no need to do it again on resume. It
should really be part of sw_init (and smu_fini should be part
of sw_fini), but we need the firmware sizes from the other IPs
for firmware loading so we have to wait until sw init is done
for all other IPs.
Reviewed-by: Sonny Jiang <Sonny.Jiang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
smu_init allocates buffers and initializes them. It does not
touch the hw. There is no need to do it again on resume. It
should really be part of sw_init (and smu_fini should be part
of sw_fini), but we need the firmware sizes from the other IPs
for firmware loading so we have to wait until sw init is done
for all other IPs.
Reviewed-by: Sonny Jiang <Sonny.Jiang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
acpi_gpiochip_request(free)_interrupts can be used for modules,
so export them. This also fixs a compile error when xgene-sb
configured as kernel module.
Fixes: 733cf014f0 "gpio: xgene: add ACPI support for APM X-Gene GPIO standby driver"
Reviewed-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Hanjun Guo <hanjun.guo@linaro.org>
Fix section mismatch error during kernel build for xgene_slimpro_i2c_probe
function. It was incorrectly defined with __init declaration.
Signed-off-by: Feng Kan <fkan@apm.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Some leftover copy and pastes from radeon that never
got updated.
Reviewed-by: Christian König <christian.koenig@amd.com>
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Sending a message to own address locks the controller up in very bizarre state,
it behaves as slave even if MDR register clearly states master. The controller
remains in this state until reset. To avoid unnecessary timeouts simply avoid
sending to own address. The controller cannot do this any way. Also, do not
enable AAS IRQ, as the slave mode is not supported by the driver and the only
possibility to trigger this IRQ is to send to own address.
Signed-off-by: Alexander Sverdlin <alexander.sverdlin@nokia.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
There are several problems in the function:
- "to_cnt" variable does nothing
- schedule_timeout() call without setting current state does nothing
- "allow_sleep" parameter is not really used
Refactor the function so that it really tries to wait. In case of timeout try
to recover the bus.
Signed-off-by: Alexander Sverdlin <alexander.sverdlin@nokia.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Adding support for i2c controller driver for Broadcom settop
SoCs.
Signed-off-by: Kamal Dasu <kdasu.kdev@gmail.com>
[wsa: removed superfluous owner in platform_driver]
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
There are still some 64-bit division problems in the cobalt code.
Replace it by div_u64.
[mchehab@osg.samsung.com: folded with an additional diff sent by
Hans via a priv e-mail]
Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reported-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>
When FIFOs are available and enabled, the driver now configures the Atmel
eXtended DMA Controller to perform word accesses instead of byte accesses
when possible.
The actual access width depends on the size of the buffer to transmit.
To enable FIFO support the "atmel,fifo-size" property must be set properly
in the I2C controller node of the device tree.
Signed-off-by: Cyrille Pitchen <cyrille.pitchen@atmel.com>
Acked-by: Ludovic Desroches <ludovic.desroches@atmel.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
The probe() function now prints the hardware version of the I2C
controller.
Signed-off-by: Cyrille Pitchen <cyrille.pitchen@atmel.com>
[wsa: s/version/hw version/] for clarity]
Acked-by: Ludovic Desroches <ludovic.desroches@atmel.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
The alternative command mode was introduced to simplify the transmission
of STOP conditions and to solve timing and latency issues around them.
This mode relies on a new register, the Alternative Command Register,
which must be set at the same time as the Master Mode Register. This new
register was designed to allow simple setup of basic combined transactions
built from up to two unitary transactions.
Indeed, the ACR is split into two areas, which describe one unitary
transaction each. Each area is filled with Data Length 8bit counter, a
Direction and a PEC Request bit. The PEC bit is only used in SMBus mode
and is not supported by this driver yet. Also when using alternative
command mode, the MREAD bit from the Master Mode Register is ignored.
Instead the Direction bits from ACR are used to setup the direction, read
or write, of each unitary transaction. Finally the 8bit counters must
filled with the data length of their respective transaction. Then if only
one transaction is to be used, the data length of the second one must be
set to zero. At the moment, this driver uses only the first transaction.
In addition to MMR and ACR, the Control Register also need to be written
to enable the alternative command mode. That's the purpose of its ACMEN
bit, which stands for Alternative Command Mode Enable.
Note that the alternative command mode is compatible with the use of the
Internal Address Register. So combined transactions for eeprom read are
actually implemented with the Internal Address Register. This register is
written with up to 3 bytes, which are the internal address sent to the
slave through the first write transaction. Then the first area of the ACR
describe the write transaction to follow, which carries the data to be
read from the eeprom. The second area of the ACR is not used so its Data
Length 8bit counter is cleared.
For each byte sent or received by the device, the Data Length 8bit counter
is decremented. When it reaches 0, a STOP condition is automatically sent.
Signed-off-by: Cyrille Pitchen <cyrille.pitchen@atmel.com>
Acked-by: Ludovic Desroches <ludovic.desroches@atmel.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
add a new value "atmel,sama5d2-i2c" for the "compatible" property.
add a new optional property "atmel,fifo-size" to enable FIFO support when
available.
add missing optional properties "dmas" and "dma-names".
Signed-off-by: Cyrille Pitchen <cyrille.pitchen@atmel.com>
Acked-by: Ludovic Desroches <ludovic.desroches@atmel.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
This patch just fixes typo before applying later patches which will use
register bits with index above 16.
Signed-off-by: Cyrille Pitchen <cyrille.pitchen@atmel.com>
Acked-by: Ludovic Desroches <ludovic.desroches@atmel.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
For TX transactions, the TXCOMP bit in the Status Register is cleared
when the first data is written into the Transmit Holding Register.
In the lines from at91_do_twi_transfer():
at91_twi_write_data_dma(dev);
at91_twi_write(dev, AT91_TWI_IER, AT91_TWI_TXCOMP);
the TXCOMP interrupt may be enabled before the DMA controller has
actually started to write into the THR. In such a case, the TXCOMP bit
is still set into the Status Register so the interrupt is triggered
immediately. The driver understands that a transaction completion has
occurred but this transaction hasn't started yet. Hence the TXCOMP
interrupt is no longer enabled by at91_do_twi_transfer() but instead
by at91_twi_write_data_dma_callback().
Also, the TXCOMP bit in the Status Register in not a clear on read flag
but a snapshot of the transmission state at the time the Status
Register is read.
When a NACK error is dectected by the I2C controller, the TXCOMP, NACK
and TXRDY bits are set together to 1 in the SR. If enabled, the TXCOMP
interrupt is triggered at the same time. Also setting the TXRDY to 1
triggers the DMA controller to write the next data into the THR. Such
a write resets the TXCOMP bit to 0 in the SR. So depending on when the
interrupt handler reads the SR, it may fail to detect the NACK error
if it relies on the TXCOMP bit. The NACK bit and its interrupt should
be used instead.
For RX transactions, the TXCOMP bit in the Status Register is cleared
when the START bit is set into the Control Register. However to unify
the management of the TXCOMP bit when the DMA controller is used, the
TXCOMP interrupt is now enabled by the DMA callbacks for both TX and
RX transfers.
Signed-off-by: Cyrille Pitchen <cyrille.pitchen@atmel.com>
Cc: stable@vger.kernel.org #3.10 and later
Acked-by: Ludovic Desroches <ludovic.desroches@atmel.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Currently, pinctrl_register() just returns NULL on error, so the
callers can not know the exact reason of the failure.
Some of the pinctrl drivers return -EINVAL, some -ENODEV, and some
-ENOMEM on error of pinctrl_register(), although the error code
might be different from the real cause of the error.
This commit reworks pinctrl_register() to return the appropriate
error code and modifies all of the pinctrl drivers to use IS_ERR()
for the error checking and PTR_ERR() for getting the error code.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Acked-by: Patrice Chotard <patrice.chotard@st.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Acked-by: Heiko Stuebner <heiko@sntech.de>
Tested-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Acked-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Acked-by: Lee Jones <lee@kernel.org>
Acked-by: Sören Brinkmann <soren.brinkmann@xilinx.com>
Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: Ray Jui <rjui@broadcom.com>
Acked-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Acked-by: Hongzhou Yang <hongzhou.yang@mediatek.com>
Acked-by: Wei Chen <Wei.Chen@csr.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
The Armada 39x SoC family has grown a new variant, the Armada 395,
which sits between the Armada 390 and Armada 398 in terms of
features. This commit adds support for this additional variant to the
Armada 39x pinctrl driver.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
The latest version of the Armada 39x datasheet documents several new
SATA related functions on various MPP pins. This commit adds the
description of these new functions to the Armada 39x pinctrl driver as
well as to its DT binding documentation.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
The latest version of the Armada 39x datasheet documents several new
PCIe related functions on various MPP pins. This commit adds the
description of these new functions to the Armada 39x pinctrl driver as
well as to its DT binding documentation.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
The latest version of the Armada 38x datasheet documents several new
PTP related functions on various MPP pins. This commit adds the
description of these new functions to the Armada 38x pinctrl driver as
well as to its DT binding documentation.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
The latest version of the Armada 38x datasheet documents several new
UART1 related functions on various MPP pins. This commit adds the
description of these new functions to the Armada 38x pinctrl driver as
well as to its DT binding documentation.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
The latest version of the Armada 38x datasheet documents several new
NAND related functions on various MPP pins. This commit adds the
description of these new functions to the Armada 38x pinctrl driver as
well as to its DT binding documentation.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
The latest version of the Armada 38x datasheet documents several new
SATA related functions on various MPP pins. This commit adds the
description of these new functions to the Armada 38x pinctrl driver as
well as to its DT binding documentation.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
The latest Armada XP datasheet documents several new DRAM related
functions on various MPPs. This commit adds the description of these
new functions in the Armada XP pinctrl driver and its DT binding
documentation.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
The latest version of the Armada XP datasheet documents a new
NAND-related MPP function on MPP48, for which this commit adds
support.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
The latest Armada XP datasheet documents that some of the MPP pins can
be used to access the second SPI bus, labelled 'spi1'. This commit
adds the corresponding pins in the pinctrl driver and its DT binding
documentation.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
This commit normalizes the subnames of the reference clock MPP pins in
the Armada 39x pinctrl driver to match with the name used on other
SoCs.
Since only the subnames are changed, DT backward compatibility is not
affected.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
After updating to the latest Armada XP datasheet, we discovered that
there is a second SPI bus accessible from the MPP pins, called 'spi1'.
In order to be consistent with other SoCs having two SPI busses, this
commit renames the functions of the first SPI bus to 'spi0' instead of
just 'spi'.
This commit obviously breaks the DT backward compatibility for the
people using the "spi" function name in their Device Tree.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Across all SoCs, even on Armada 370 for SPI0, the clock pin uses the
'sck' subname and not 'clk', so this commit adjusts the code and
documentation accordingly.
Since this commit only changes the subname, DT backward compatibility
is not affected.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
For consistency with the datasheet, this commit renames the VDD
function of the MPP4 pin.
While this changes the DT compatibility, it is not considered to be a
problem since this pin is unlikely to be used for anything but
debugging purposes.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
There was an off-by-one in the documentation of the ge1(txd[0-3])
pins, which is fixed by this commit. Since the driver was correct, and
the subnames are anyway not used in the DT binding itself, there is no
need to push this documentation fix for stable.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
This commit normalizes the naming of the Ethernet txclkout pin to be
the same accross Marvell SoCs. It is worth mentioning that the DT
binding documentation of the Armada XP was wrong for MPP12: it said
the function was ge1(txd0), while it is in fact ge1(txclkout). It is
however not really a fix worth sending to stable since it does not
change the behavior, and the driver itself was correct.
Since only the subnames are changed, DT backward compatibility is not
affected.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
This commit aligns the naming of the audio 'lrclk' pin accross Marvell
SoCs.
Since only the subname is changed, the DT backward compatibility is
not affected.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
This commit normalizes the naming of PCIe pins to use 'rstout' instead
of 'rstoutn' or 'rst-out'.
Since only the subnames are changed, DT compatibility is not affected.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
This commit normalizes the naming of the TDM pins accross the
different Marvell SoCs. Mainly it consists in:
* Removing the 'n' from signal names: 'intn' becomes 'int' and 'rstn'
becomes 'rst'
* Renaming the main name 'tdm2c' to 'tdm' on Armada 38x.
* Change the main name 'tdm-1' to 'tdm' for one of the pins of the
Armada XP
The last two changes affect DT compatibility, but since the TDM
interface is nowhere near being supported in mainline, it should not
be considered to be a serious problem at this point.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
