Boot CPU0 always handles DMA interrupts and under some rare circumstances
it could stuck in uninterruptible state for a significant time (like in a
case of KASAN + NFS root). In this case sibling CPU, which waits for DMA
transfer completion, will get a DMA transfer timeout. In order to handle
this rare condition, interrupt status needs to be polled until interrupt
is handled.
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Link: https://lore.kernel.org/r/20200319212321.3297-2-digetx@gmail.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
When CONFIG_PM is disabled, gcc warning this:
drivers/dma/tegra20-apb-dma.c:1587:12: warning: 'tegra_dma_runtime_resume' defined but not used [-Wunused-function]
drivers/dma/tegra20-apb-dma.c:1578:12: warning: 'tegra_dma_runtime_suspend' defined but not used [-Wunused-function]
Make it as __maybe_unused to fix the warnings,
also remove unneeded function declarations.
Fixes: ec8a158678 ("dma: tegra: add dmaengine based dma driver")
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Thierry Reding <treding@nvidia.com>
Link: https://lore.kernel.org/r/20200320071337.59756-1-yuehaibing@huawei.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
On new Spreadtrum platforms, when the CPU enters idle, it will close
the DMA controllers' clock to save power if the DMA controller is not
busy. Moreover the DMA controller's busy signal depends on the DMA
enable flag and the request pending flag.
When DMA controller starts to transfer data, which means we already
set the DMA enable flag, but now we should also set the request pending
flag, in case the DMA clock will be closed accidentally if the CPU
can not detect the DMA controller's busy signal.
Signed-off-by: Zhenfang Wang <zhenfang.wang@unisoc.com>
Signed-off-by: Baolin Wang <baolin.wang7@gmail.com>
Link: https://lore.kernel.org/r/02adbe4364ec436ec2c5bc8fd2386bab98edd884.1584019223.git.baolin.wang7@gmail.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Via the /sys/kernel/debug/dmaengine/summary users can get information
about the DMA devices and the used channels.
Example output on am654-evm with audio using two channels and after running
dmatest on 4 channels:
dma0 (285c0000.dma-controller): number of channels: 96
dma1 (31150000.dma-controller): number of channels: 267
dma1chan0 | 2b00000.mcasp:tx
dma1chan1 | 2b00000.mcasp:rx
dma1chan2 | in-use
dma1chan3 | in-use
dma1chan4 | in-use
dma1chan5 | in-use
For slave channels we can show the device and the channel name a given
channel is requested.
For non slave devices the only information we know is that the channel is
in use.
DMA drivers can implement the optional dbg_summary_show callback to
provide controller specific information instead of the generic one.
It is easy to extend the generic dmaengine_summary_show() to print
additional information about the used channels.
I have taken the idea from gpiolib and clk subsystems.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Link: https://lore.kernel.org/r/20200306142839.17910-2-peter.ujfalusi@ti.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
This adds external DMA controller driver implemented in Socionext
UniPhier SoCs. This driver supports DMA_MEMCPY and DMA_SLAVE modes.
Since this driver does not support the the way to transfer size
unaligned to burst width, 'src_maxburst' or 'dst_maxburst' of
dma_slave_config must be 1 to transfer arbitrary size. If transfer
size is unaligned to burst size, the transfer isn't started and
the driver displays an error message.
Signed-off-by: Kunihiko Hayashi <hayashi.kunihiko@socionext.com>
Link: https://lore.kernel.org/r/1582271550-3403-3-git-send-email-hayashi.kunihiko@socionext.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
The DT for virtualized hosts have dma-cells == 2 where the second parameter
is the ATYPE for the channel.
In case of dma-cells == 1 we can configure the ATYPE as 0 (reset value).
The ATYPE defined for j721e are:
0: pointers are physical addresses (no translation)
1: pointers are intermediate addresses (PVU)
2: pointers are virtual addresses (SMMU)
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Link: https://lore.kernel.org/r/20200218143126.11361-3-peter.ujfalusi@ti.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
In UDMA each channel can use different ATYPE value which tells UDMA how
the addresses in the descriptors should be treated:
0: pointers are physical addresses (no translation)
1: pointers are intermediate addresses (PVU)
2: pointers are virtual addresses (SMMU)
When virtualized environment is used then the dma binding should use
additional cell to configure the desired ATYPE for the channel.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Link: https://lore.kernel.org/r/20200218143126.11361-2-peter.ujfalusi@ti.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
It is enough to check whether hardware is busy on suspend and to reset
it across of suspend-resume because:
1. Channel's configuration is fully re-programmed on each DMA
transfer anyways.
2. Context save-restore of an active channel won't end up well without
pausing transfer prior to the context's saving, but note that every
channel shall be idling at the time of suspend, so save-restore is
not needed at all.
3. The only case where context save-restore may be useful is when
channel is in a paused state during suspend. But channel's pausing
could be supported only on Tegra114+ and this functionality wasn't
implemented by the driver for years now because there is no need for
it in upstream kernel.
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Jon Hunter <jonathanh@nvidia.com>
Link: https://lore.kernel.org/r/20200209163356.6439-14-digetx@gmail.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
It's a bit impractical to enable hardware's clock at the time of DMA
channel's allocation because most of DMA client drivers allocate DMA
channel at the time of the driver's probing, and thus, DMA clock is kept
always-enabled in practice, defeating the whole purpose of runtime PM.
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Jon Hunter <jonathanh@nvidia.com>
Link: https://lore.kernel.org/r/20200209163356.6439-13-digetx@gmail.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
There is no need to kill tasklet when driver's probe fails because tasklet
can't be scheduled at this time. It is also cleaner to kill tasklet on
channel's freeing rather than to kill it on driver's removal, otherwise
tasklet could perform a dummy execution after channel's releasing, which
isn't very nice.
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
Acked-by: Jon Hunter <jonathanh@nvidia.com>
Link: https://lore.kernel.org/r/20200209163356.6439-6-digetx@gmail.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Tasklets run with all the interrupts enabled. This means that we should
replace all the (already present) spin_lock_irqsave() uses in the tasklet
with spin_lock_irq() to protect being interrupted by a IRQ which tries
to get the same lock (via calls to device_prep_dma_* for example).
spin_lock and spin_lock_bh in tasklets are not enough to protect from IRQs,
update these to spin_lock_irq().
at_xdmac_advance_work() can be called with all the interrupts enabled (when
called from tasklet), or with interrupts disabled (when called from
at_xdmac_issue_pending). Move the locking in the callers to be able to use
spin_lock_irq() and spin_lock_irqsave() for these cases.
Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Acked-by: Ludovic Desroches <ludovic.desroches@microchip.com>
Link: https://lore.kernel.org/r/20200123140237.125799-10-tudor.ambarus@microchip.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Fix the following deadlocks:
1/ atc_handle_cyclic() and atc_chain_complete() called
dmaengine_desc_get_callback_invoke() while wrongly holding the
atchan->lock. Clients can set the callback to dmaengine_terminate_sync()
which will end up trying to get the same lock, thus a deadlock occurred.
2/ dma_run_dependencies() was called with the atchan->lock held, but the
method calls device_issue_pending() which tries to get the same lock,
and so a deadlock occurred.
The driver must not hold the lock when invoking the callback or when
running dependencies. Releasing the spinlock within a called function
before calling the callback is not a nice thing to do -> called functions
become non-atomic when called within an atomic region. Thus the lock is
now taken in the child routines whereever is needed.
Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Acked-by: Ludovic Desroches <ludovic.desroches@microchip.com>
Link: https://lore.kernel.org/r/20200123140237.125799-6-tudor.ambarus@microchip.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
