If the transfer is shorted then 64K we can complete it with one ACNT burst
by configuring ACNT to the length of the copy, this require one paRAM slot.
Otherwise we use two paRAM slots for the copy:
slot1: will copy (length / 32767) number of 32767 byte long blocks
slot2: will be configured to copy the remaining data.
According to tests this patch increases the throughput of memcpy from
~3MB/s to 15MB/s
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Despite the claim by the original commit adding the memcpy
support, eDMA does not have constraint on the alignment of src, dst
or length in increment mode.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
HSU (High Speed UART) DMA engine, like the name suggests, is
an integrated DMA engine for UART and UART alone. Therefore,
making the UART drivers responsible of selecting it and
removing the user selectable option for it. The UARTs with
this DMA engine can always select HSU_DMA when
SERIAL_8250_DMA option is enabled.
Suggested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Heikki Krogerus <heikki.krogerus@linux.intel.com>
Acked-by: Vinod Koul <vinod.koul@intel.com>
Acked-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The DMA engine supports memory copy, RAID5 XOR, RAID6 PQ, and other
computations. But the bandwidth of the entire DMA engine is shared
among all channels. This patch re-configures operations availability
such that one can achieve maximum performance for XOR and PQ
computation by removing the memory offload operations.
Signed-off-by: Rameshwar Prasad Sahu <rsahu@apm.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
If the eDMA3 has support for channel paRAM slot mapping we can utilize it
to allocate slots on demand and save precious slots for real transfers.
On am335x the eDMA has 64 channels which means we can unlock 64 paRAM
slots out from the available 256.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
The names chosen for the bitfields were quite confusing and given no real
information on what they are used for...
edma_inuse -> slot_inuse: tracks the slot usage/availability
edma_unused -> channel_unused: tracks the channel usage/availability
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Instead of directly reading it from CCCFG register take the information out
once when we set up the configuration from the HW.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
No need to run through the bits in QEMR and CCERR events since they will
not trigger any action, so just clearing the errors there is fine.
In case of the missed event the loop can be optimized so we spend less time
to handle the event.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
In the ccerr interrupt handler the code checks for pending errors in the
error status registers in two different places.
Move the check out to a helper function.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
With the merger of the arch/arm/common/edma.c code into the dmaengine
driver, there is no longer need to have per channel callback/data storage
for interrupt events.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Warning message in case of linking between paRAM slots in different eDMA
controllers.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
We have access to dev, so it is better to use the dev_dbg for debug prints.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Be consistent and do not mix the use of dev, &pdev->dev, etc in the
functions.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
When allocating a memory for number of items it is better (looks better)
to use devm_kcalloc.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Instead of using defines to specify the size of different arrays and
bitmaps, allocate the memory for them based on the information we get from
the HW itself.
Since these defines are set based on the worst case, there are devices
where they are not valid.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Move the code out from arch/arm/common and merge it inside of the dmaengine
driver.
This change is done with as minimal (if eny) functional change to the code
as possible to avoid introducing regression.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Acked-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Since the driver stack no longer depends on lookup with id number in a
global array of pointers, the limitation for the number of eDMAs are no
longer needed. We can handle as many eDMAs in legacy and DT boot as we have
memory for them to allocate the needed structures.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Instead of relying on indexes pointing to edma private date in the global
pointer array, pass the private data pointer via the public API.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
If the of_dma_controller is registered in the non dmaengine driver we could
have race condition:
the of_dma_controller has been registered, but the dmaengine driver is not
yet probed. Drivers requesting DMA channels during this window will fail
since we do not yet have dmaengine drivers registered.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
The code path in edma_execute() and edma_callback() can be simplified
and make it more optimal.
There is not need to call in to edma_execute() when the transfer
has been finished for example.
Also the handling of missed/first or next batch of paRAMs can
be done in a more optimal way.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
There is no need to print that the driver has been initialized
or removed, so remove such messages.
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Since commit d078cd1b41 ("dmaengine: imx-sdma: Add imx6sx platform
support") we get this message on every boot on mx6q:
imx-sdma 20ec000.sdma: no event needs to be remapped
, which is not very helpful.
Move the message to debug level instead.
Cc: Zidan Wang <zidan.wang@freescale.com>
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
This is needed due to the duplicated iommu stuff to help with the merge
and to prevent future issues.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The L3 throughput can be higher than expected when packed access
is not enabled. The ratio depends on the number of bytes in a
transaction and the EMIF interface width.
The throughput was measured for the following settings/cases:
* Case 1: Burst size of 64 bytes, packed access disabled
* Case 2: Burst size of 64 bytes, packed access enabled
* Case 3: Burst disabled, packed access disabled
Throughput measurements were done during McASP-based audio
playback on the Jacinto6 EVM using the omapconf tool [1]:
$ omapconf trace bw -m sdma_rd
---------------------------------------------------------
Throughput (MB/s)
Audio parameters Case 1 Case 2 Case 3
---------------------------------------------------------
44.1kHz, 16-bits, stereo 1.41 0.18 1.41
44.1kHz, 32-bits, stereo 1.41 0.35 1.41
44.1kHz, 16-bits, 4-chan 2.82 0.35 2.82
44.1kHz, 16-bits, 6-chan 4.23 0.53 4.23
44.1kHz, 16-bits, 8-chan 5.64 0.71 5.64
---------------------------------------------------------
From above measurements, case 2 is the only one that delivers
the expected throughput for the given audio parameters. For
that reason, the packed accesses are now enabled.
It's worth to mention that packed accesses cannot be enabled
for all addressing modes. In cyclic transfers, it can be
enabled in the source for MEM_TO_DEV and in dest for DEV_TO_MEM,
as they use post-increment mode which supports packed accesses.
Peter Ujfalusi:
From the TRM regarding to this:
"NOTE: Except in the constant addressing mode, the source or
destination must be specified as packed for burst transactions
to occur."
So w/o the packed setting the burst on the MEM side was not
enabled, this explains the numbers.
[1] https://github.com/omapconf/omapconf
Signed-off-by: Misael Lopez Cruz <misael.lopez@ti.com>
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
The MIC X100 DMA engine has a special status descriptor which writes
an 8 byte value to a destination location. This is used to signal
completion of all DMA descriptors prior to the status descriptor.
This patch add a new DMA engine API which enables updating a
destination address with an 8 byte immediate data value.
Reviewed-by: Nikhil Rao <nikhil.rao@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Lawrynowicz, Jacek <jacek.lawrynowicz@intel.com>
Signed-off-by: Sudeep Dutt <sudeep.dutt@intel.com>
Signed-off-by: Siva Yerramreddy <yshivakrishna@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Pull dmaengine fixes from Vinod Koul:
"This contains fixes spread throughout the drivers, and also fixes one
more instance of privatecnt in dmaengine.
Driver fixes summary:
- bunch of pxa_dma fixes for reuse of descriptor issue, residue and
no-requestor
- odd fixes in xgene, idma, sun4i and zxdma
- at_xdmac fixes for cleaning descriptor and block addr mode"
* tag 'dmaengine-fix-4.3-rc4' of git://git.infradead.org/users/vkoul/slave-dma:
dmaengine: pxa_dma: fix residue corner case
dmaengine: pxa_dma: fix the no-requestor case
dmaengine: zxdma: Fix off-by-one for testing valid pchan request
dmaengine: at_xdmac: clean used descriptor
dmaengine: at_xdmac: change block increment addressing mode
dmaengine: dw: properly read DWC_PARAMS register
dmaengine: xgene-dma: Fix overwritting DMA tx ring
dmaengine: fix balance of privatecnt
dmaengine: sun4i: fix unsafe list iteration
dmaengine: idma64: improve residue estimation
dmaengine: xgene-dma: fix handling xgene_dma_get_ring_size result
dmaengine: pxa_dma: fix initial list move
A very tiny temporal window exists in the residue calculation where :
- upon entering residue calculation, the transfer is ongoing
- when reading the current transfer pointer, it just changed to
the "finisher/linker" descriptor
In this case, the residue returned is the whole transfer length instead
of 0. Fix it.
This appears almost in one extreme case, where the driver is used
by older clients which inquire for residue in interrupt context, such
as the smsc91x ethernet driver, in a tight loop :
interrupt_handler()
dmaengine_submit()
do {
dmaengine_tx_status()
} while (residue > 0 || status != DMA_ERROR)
Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
A very small number of devices don't use the flow control offered by
requestor lines. In these specific cases, the pxa dma driver should be
aware of that and not try to use a requestor line.
Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
This platform driver has a OF device ID table but the OF module
alias information is not created so module autoloading won't work.
Signed-off-by: Luis de Bethencourt <luisbg@osg.samsung.com>
Acked-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
This platform driver has a OF device ID table but the OF module
alias information is not created so module autoloading won't work.
Signed-off-by: Luis de Bethencourt <luisbg@osg.samsung.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
This platform driver has a OF device ID table but the OF module
alias information is not created so module autoloading won't work.
Signed-off-by: Luis de Bethencourt <luisbg@osg.samsung.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
This platform driver has a OF device ID table but the OF module
alias information is not created so module autoloading won't work.
Signed-off-by: Luis de Bethencourt <luisbg@osg.samsung.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
This platform driver has a OF device ID table but the OF module
alias information is not created so module autoloading won't work.
Signed-off-by: Luis de Bethencourt <luisbg@osg.samsung.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
In interleaved mode, when numf > 1, we have only one descriptor for the
transfer but this descriptor has to be added to the descs_list. If not,
when doing remove_xfer, the descriptor won't be put back in the
free_descs_list.
Signed-off-by: Ludovic Desroches <ludovic.desroches@atmel.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
When putting back a descriptor to the free descs list, some fields are
not set to 0, it can cause bugs if someone uses it without having this
in mind.
Descriptor are not put back one by one so it is easier to clean
descriptors when we request them.
Signed-off-by: Ludovic Desroches <ludovic.desroches@atmel.com>
Cc: stable@vger.kernel.org #4.2
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
The addressing mode we were using was not only incrementing the address at
each microblock, but also at each data boundary, which was severely slowing
the transfer, without any benefit since we were not using the data stride.
Switch to the micro block increment only in order to get back to an
acceptable performance level.
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Signed-off-by: Ludovic Desroches <ludovic.desroches@atmel.com>
Fixes: 6007ccb577 ("dmaengine: xdmac: Add interleaved transfer support")
Cc: stable@vger.kernel.org #4.2
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Fix the call to memset in this driver
[linux-4.2-next-20150911/drivers/dma/zx296702_dma.c:444]: (warning)
memset() called to fill 0 bytes of 'ds'.
Reported-by: David Binderman <dcb314@hotmail.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Instead of hardconding a platform data for dw_dmac let's use it's own
autoconfiguration feature. Thus, remove hardcoded values.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
We replace __fls() by __ffs() since we have to find a *minimum* data width that
satisfies both source and destination.
While here, rename dwc_fast_fls() to dwc_fast_ffs() which it really is.
Fixes: 4c2d56c574 (dw_dmac: introduce dwc_fast_fls())
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
In case we have less than maximum allowed channels (8) and autoconfiguration is
enabled the DWC_PARAMS read is wrong because it uses different arithmetic to
what is needed for channel priority setup.
Re-do the caclulations properly. This now works on AVR32 board well.
Fixes: fed2574b3c (dw_dmac: introduce software emulation of LLP transfers)
Cc: yitian.bu@tangramtek.com
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
This patch fixes an over flow issue with the TX ring descriptor. Each
descriptor is 32B in size and an operation requires 2 of these
descriptors.
Signed-off-by: Rameshwar Prasad Sahu <rsahu@apm.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
dma_release_channel() decrements privatecnt counter and almost all dma_get*
function increments it with the exception of dma_get_slave_channel().
In most cases this does not cause issue since normally the channel is not
requested and released, but if a driver requests DMA channel via
dma_get_slave_channel() and releases the channel the privatecnt will be
unbalanced and this will prevent for example getting channel for memcpy.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
