linux/drivers/dma/dmaengine.c

1356 lines
34 KiB
C
Raw Normal View History

/*
* Copyright(c) 2004 - 2006 Intel Corporation. All rights reserved.
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the Free
* Software Foundation; either version 2 of the License, or (at your option)
* any later version.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* The full GNU General Public License is included in this distribution in the
* file called COPYING.
*/
/*
* This code implements the DMA subsystem. It provides a HW-neutral interface
* for other kernel code to use asynchronous memory copy capabilities,
* if present, and allows different HW DMA drivers to register as providing
* this capability.
*
* Due to the fact we are accelerating what is already a relatively fast
* operation, the code goes to great lengths to avoid additional overhead,
* such as locking.
*
* LOCKING:
*
* The subsystem keeps a global list of dma_device structs it is protected by a
* mutex, dma_list_mutex.
*
* A subsystem can get access to a channel by calling dmaengine_get() followed
* by dma_find_channel(), or if it has need for an exclusive channel it can call
* dma_request_channel(). Once a channel is allocated a reference is taken
* against its corresponding driver to disable removal.
*
* Each device has a channels list, which runs unlocked but is never modified
* once the device is registered, it's just setup by the driver.
*
* See Documentation/driver-api/dmaengine for more details
*/
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
dmaengine: core: Introduce new, universal API to request a channel The two API function can cover most, if not all current APIs used to request a channel. With minimal effort dmaengine drivers, platforms and dmaengine user drivers can be converted to use the two function. struct dma_chan *dma_request_chan_by_mask(const dma_cap_mask_t *mask); To request any channel matching with the requested capabilities, can be used to request channel for memcpy, memset, xor, etc where no hardware synchronization is needed. struct dma_chan *dma_request_chan(struct device *dev, const char *name); To request a slave channel. The dma_request_chan() will try to find the channel via DT, ACPI or in case if the kernel booted in non DT/ACPI mode it will use a filter lookup table and retrieves the needed information from the dma_slave_map provided by the DMA drivers. This legacy mode needs changes in platform code, in dmaengine drivers and finally the dmaengine user drivers can be converted: For each dmaengine driver an array of DMA device, slave and the parameter for the filter function needs to be added: static const struct dma_slave_map da830_edma_map[] = { { "davinci-mcasp.0", "rx", EDMA_FILTER_PARAM(0, 0) }, { "davinci-mcasp.0", "tx", EDMA_FILTER_PARAM(0, 1) }, { "davinci-mcasp.1", "rx", EDMA_FILTER_PARAM(0, 2) }, { "davinci-mcasp.1", "tx", EDMA_FILTER_PARAM(0, 3) }, { "davinci-mcasp.2", "rx", EDMA_FILTER_PARAM(0, 4) }, { "davinci-mcasp.2", "tx", EDMA_FILTER_PARAM(0, 5) }, { "spi_davinci.0", "rx", EDMA_FILTER_PARAM(0, 14) }, { "spi_davinci.0", "tx", EDMA_FILTER_PARAM(0, 15) }, { "da830-mmc.0", "rx", EDMA_FILTER_PARAM(0, 16) }, { "da830-mmc.0", "tx", EDMA_FILTER_PARAM(0, 17) }, { "spi_davinci.1", "rx", EDMA_FILTER_PARAM(0, 18) }, { "spi_davinci.1", "tx", EDMA_FILTER_PARAM(0, 19) }, }; This information is going to be needed by the dmaengine driver, so modification to the platform_data is needed, and the driver map should be added to the pdata of the DMA driver: da8xx_edma0_pdata.slave_map = da830_edma_map; da8xx_edma0_pdata.slavecnt = ARRAY_SIZE(da830_edma_map); The DMA driver then needs to configure the needed device -> filter_fn mapping before it registers with dma_async_device_register() : ecc->dma_slave.filter_map.map = info->slave_map; ecc->dma_slave.filter_map.mapcnt = info->slavecnt; ecc->dma_slave.filter_map.fn = edma_filter_fn; When neither DT or ACPI lookup is available the dma_request_chan() will try to match the requester's device name with the filter_map's list of device names, when a match found it will use the information from the dma_slave_map to get the channel with the dma_get_channel() internal function. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com> Reviewed-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-12-14 20:47:40 +00:00
#include <linux/platform_device.h>
#include <linux/dma-mapping.h>
#include <linux/init.h>
#include <linux/module.h>
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
#include <linux/mm.h>
#include <linux/device.h>
#include <linux/dmaengine.h>
#include <linux/hardirq.h>
#include <linux/spinlock.h>
#include <linux/percpu.h>
#include <linux/rcupdate.h>
#include <linux/mutex.h>
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
#include <linux/jiffies.h>
#include <linux/rculist.h>
#include <linux/idr.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 08:04:11 +00:00
#include <linux/slab.h>
#include <linux/acpi.h>
#include <linux/acpi_dma.h>
#include <linux/of_dma.h>
#include <linux/mempool.h>
static DEFINE_MUTEX(dma_list_mutex);
static DEFINE_IDA(dma_ida);
static LIST_HEAD(dma_device_list);
static long dmaengine_ref_count;
/* --- sysfs implementation --- */
/**
* dev_to_dma_chan - convert a device pointer to the its sysfs container object
* @dev - device node
*
* Must be called under dma_list_mutex
*/
static struct dma_chan *dev_to_dma_chan(struct device *dev)
{
struct dma_chan_dev *chan_dev;
chan_dev = container_of(dev, typeof(*chan_dev), device);
return chan_dev->chan;
}
static ssize_t memcpy_count_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct dma_chan *chan;
unsigned long count = 0;
int i;
int err;
mutex_lock(&dma_list_mutex);
chan = dev_to_dma_chan(dev);
if (chan) {
for_each_possible_cpu(i)
count += per_cpu_ptr(chan->local, i)->memcpy_count;
err = sprintf(buf, "%lu\n", count);
} else
err = -ENODEV;
mutex_unlock(&dma_list_mutex);
return err;
}
static DEVICE_ATTR_RO(memcpy_count);
static ssize_t bytes_transferred_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct dma_chan *chan;
unsigned long count = 0;
int i;
int err;
mutex_lock(&dma_list_mutex);
chan = dev_to_dma_chan(dev);
if (chan) {
for_each_possible_cpu(i)
count += per_cpu_ptr(chan->local, i)->bytes_transferred;
err = sprintf(buf, "%lu\n", count);
} else
err = -ENODEV;
mutex_unlock(&dma_list_mutex);
return err;
}
static DEVICE_ATTR_RO(bytes_transferred);
static ssize_t in_use_show(struct device *dev, struct device_attribute *attr,
char *buf)
{
struct dma_chan *chan;
int err;
mutex_lock(&dma_list_mutex);
chan = dev_to_dma_chan(dev);
if (chan)
err = sprintf(buf, "%d\n", chan->client_count);
else
err = -ENODEV;
mutex_unlock(&dma_list_mutex);
return err;
}
static DEVICE_ATTR_RO(in_use);
static struct attribute *dma_dev_attrs[] = {
&dev_attr_memcpy_count.attr,
&dev_attr_bytes_transferred.attr,
&dev_attr_in_use.attr,
NULL,
};
ATTRIBUTE_GROUPS(dma_dev);
static void chan_dev_release(struct device *dev)
{
struct dma_chan_dev *chan_dev;
chan_dev = container_of(dev, typeof(*chan_dev), device);
if (atomic_dec_and_test(chan_dev->idr_ref)) {
mutex_lock(&dma_list_mutex);
ida_remove(&dma_ida, chan_dev->dev_id);
mutex_unlock(&dma_list_mutex);
kfree(chan_dev->idr_ref);
}
kfree(chan_dev);
}
static struct class dma_devclass = {
.name = "dma",
.dev_groups = dma_dev_groups,
.dev_release = chan_dev_release,
};
/* --- client and device registration --- */
#define dma_device_satisfies_mask(device, mask) \
__dma_device_satisfies_mask((device), &(mask))
static int
__dma_device_satisfies_mask(struct dma_device *device,
const dma_cap_mask_t *want)
{
dma_cap_mask_t has;
bitmap_and(has.bits, want->bits, device->cap_mask.bits,
DMA_TX_TYPE_END);
return bitmap_equal(want->bits, has.bits, DMA_TX_TYPE_END);
}
static struct module *dma_chan_to_owner(struct dma_chan *chan)
{
return chan->device->dev->driver->owner;
}
/**
* balance_ref_count - catch up the channel reference count
* @chan - channel to balance ->client_count versus dmaengine_ref_count
*
* balance_ref_count must be called under dma_list_mutex
*/
static void balance_ref_count(struct dma_chan *chan)
{
struct module *owner = dma_chan_to_owner(chan);
while (chan->client_count < dmaengine_ref_count) {
__module_get(owner);
chan->client_count++;
}
}
/**
* dma_chan_get - try to grab a dma channel's parent driver module
* @chan - channel to grab
*
* Must be called under dma_list_mutex
*/
static int dma_chan_get(struct dma_chan *chan)
{
struct module *owner = dma_chan_to_owner(chan);
int ret;
/* The channel is already in use, update client count */
if (chan->client_count) {
__module_get(owner);
goto out;
}
if (!try_module_get(owner))
return -ENODEV;
/* allocate upon first client reference */
if (chan->device->device_alloc_chan_resources) {
ret = chan->device->device_alloc_chan_resources(chan);
if (ret < 0)
goto err_out;
}
if (!dma_has_cap(DMA_PRIVATE, chan->device->cap_mask))
balance_ref_count(chan);
out:
chan->client_count++;
return 0;
err_out:
module_put(owner);
return ret;
}
/**
* dma_chan_put - drop a reference to a dma channel's parent driver module
* @chan - channel to release
*
* Must be called under dma_list_mutex
*/
static void dma_chan_put(struct dma_chan *chan)
{
/* This channel is not in use, bail out */
if (!chan->client_count)
return;
chan->client_count--;
module_put(dma_chan_to_owner(chan));
/* This channel is not in use anymore, free it */
dmaengine: Add transfer termination synchronization support The DMAengine API has a long standing race condition that is inherent to the API itself. Calling dmaengine_terminate_all() is supposed to stop and abort any pending or active transfers that have previously been submitted. Unfortunately it is possible that this operation races against a currently running (or with some drivers also scheduled) completion callback. Since the API allows dmaengine_terminate_all() to be called from atomic context as well as from within a completion callback it is not possible to synchronize to the execution of the completion callback from within dmaengine_terminate_all() itself. This means that a user of the DMAengine API does not know when it is safe to free resources used in the completion callback, which can result in a use-after-free race condition. This patch addresses the issue by introducing an explicit synchronization primitive to the DMAengine API called dmaengine_synchronize(). The existing dmaengine_terminate_all() is deprecated in favor of dmaengine_terminate_sync() and dmaengine_terminate_async(). The former aborts all pending and active transfers and synchronizes to the current context, meaning it will wait until all running completion callbacks have finished. This means it is only possible to call this function from non-atomic context. The later function does not synchronize, but can still be used in atomic context or from within a complete callback. It has to be followed up by dmaengine_synchronize() before a client can free the resources used in a completion callback. In addition to this the semantics of the device_terminate_all() callback are slightly relaxed by this patch. It is now OK for a driver to only schedule the termination of the active transfer, but does not necessarily have to wait until the DMA controller has completely stopped. The driver must ensure though that the controller has stopped and no longer accesses any memory when the device_synchronize() callback returns. This was in part done since most drivers do not pay attention to this anyway at the moment and to emphasize that this needs to be done when the device_synchronize() callback is implemented. But it also helps with implementing support for devices where stopping the controller can require operations that may sleep. Signed-off-by: Lars-Peter Clausen <lars@metafoo.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-10-20 09:46:28 +00:00
if (!chan->client_count && chan->device->device_free_chan_resources) {
/* Make sure all operations have completed */
dmaengine_synchronize(chan);
chan->device->device_free_chan_resources(chan);
dmaengine: Add transfer termination synchronization support The DMAengine API has a long standing race condition that is inherent to the API itself. Calling dmaengine_terminate_all() is supposed to stop and abort any pending or active transfers that have previously been submitted. Unfortunately it is possible that this operation races against a currently running (or with some drivers also scheduled) completion callback. Since the API allows dmaengine_terminate_all() to be called from atomic context as well as from within a completion callback it is not possible to synchronize to the execution of the completion callback from within dmaengine_terminate_all() itself. This means that a user of the DMAengine API does not know when it is safe to free resources used in the completion callback, which can result in a use-after-free race condition. This patch addresses the issue by introducing an explicit synchronization primitive to the DMAengine API called dmaengine_synchronize(). The existing dmaengine_terminate_all() is deprecated in favor of dmaengine_terminate_sync() and dmaengine_terminate_async(). The former aborts all pending and active transfers and synchronizes to the current context, meaning it will wait until all running completion callbacks have finished. This means it is only possible to call this function from non-atomic context. The later function does not synchronize, but can still be used in atomic context or from within a complete callback. It has to be followed up by dmaengine_synchronize() before a client can free the resources used in a completion callback. In addition to this the semantics of the device_terminate_all() callback are slightly relaxed by this patch. It is now OK for a driver to only schedule the termination of the active transfer, but does not necessarily have to wait until the DMA controller has completely stopped. The driver must ensure though that the controller has stopped and no longer accesses any memory when the device_synchronize() callback returns. This was in part done since most drivers do not pay attention to this anyway at the moment and to emphasize that this needs to be done when the device_synchronize() callback is implemented. But it also helps with implementing support for devices where stopping the controller can require operations that may sleep. Signed-off-by: Lars-Peter Clausen <lars@metafoo.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-10-20 09:46:28 +00:00
}
/* If the channel is used via a DMA request router, free the mapping */
if (chan->router && chan->router->route_free) {
chan->router->route_free(chan->router->dev, chan->route_data);
chan->router = NULL;
chan->route_data = NULL;
}
}
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
enum dma_status dma_sync_wait(struct dma_chan *chan, dma_cookie_t cookie)
{
enum dma_status status;
unsigned long dma_sync_wait_timeout = jiffies + msecs_to_jiffies(5000);
dma_async_issue_pending(chan);
do {
status = dma_async_is_tx_complete(chan, cookie, NULL, NULL);
if (time_after_eq(jiffies, dma_sync_wait_timeout)) {
dev_err(chan->device->dev, "%s: timeout!\n", __func__);
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
return DMA_ERROR;
}
if (status != DMA_IN_PROGRESS)
break;
cpu_relax();
} while (1);
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
return status;
}
EXPORT_SYMBOL(dma_sync_wait);
/**
* dma_cap_mask_all - enable iteration over all operation types
*/
static dma_cap_mask_t dma_cap_mask_all;
/**
* dma_chan_tbl_ent - tracks channel allocations per core/operation
* @chan - associated channel for this entry
*/
struct dma_chan_tbl_ent {
struct dma_chan *chan;
};
/**
* channel_table - percpu lookup table for memory-to-memory offload providers
*/
static struct dma_chan_tbl_ent __percpu *channel_table[DMA_TX_TYPE_END];
static int __init dma_channel_table_init(void)
{
enum dma_transaction_type cap;
int err = 0;
bitmap_fill(dma_cap_mask_all.bits, DMA_TX_TYPE_END);
/* 'interrupt', 'private', and 'slave' are channel capabilities,
* but are not associated with an operation so they do not need
* an entry in the channel_table
*/
clear_bit(DMA_INTERRUPT, dma_cap_mask_all.bits);
clear_bit(DMA_PRIVATE, dma_cap_mask_all.bits);
clear_bit(DMA_SLAVE, dma_cap_mask_all.bits);
for_each_dma_cap_mask(cap, dma_cap_mask_all) {
channel_table[cap] = alloc_percpu(struct dma_chan_tbl_ent);
if (!channel_table[cap]) {
err = -ENOMEM;
break;
}
}
if (err) {
pr_err("initialization failure\n");
for_each_dma_cap_mask(cap, dma_cap_mask_all)
free_percpu(channel_table[cap]);
}
return err;
}
arch_initcall(dma_channel_table_init);
/**
* dma_find_channel - find a channel to carry out the operation
* @tx_type: transaction type
*/
struct dma_chan *dma_find_channel(enum dma_transaction_type tx_type)
{
return this_cpu_read(channel_table[tx_type]->chan);
}
EXPORT_SYMBOL(dma_find_channel);
/**
* dma_issue_pending_all - flush all pending operations across all channels
*/
void dma_issue_pending_all(void)
{
struct dma_device *device;
struct dma_chan *chan;
rcu_read_lock();
list_for_each_entry_rcu(device, &dma_device_list, global_node) {
if (dma_has_cap(DMA_PRIVATE, device->cap_mask))
continue;
list_for_each_entry(chan, &device->channels, device_node)
if (chan->client_count)
device->device_issue_pending(chan);
}
rcu_read_unlock();
}
EXPORT_SYMBOL(dma_issue_pending_all);
/**
dmaengine: make dma_channel_rebalance() NUMA aware dma_channel_rebalance() currently distributes channels by processor ID. These IDs often change with the BIOS, and the order isn't related to the DMA channel list (related to PCI bus ids). * On my SuperMicro dual E5 machine, first socket has processor IDs [0-7] (and [16-23] for hyperthreads), second socket has [8-15]+[24-31] => channels are properly allocated to local CPUs. * On Dells R720 with same processors, first socket has even processor IDs, second socket has odd numbers => half the processors get channels on the remote socket, causing cross-NUMA traffic and lower DMA performance. Change nth_chan() to return the channel with min table_count and in the NUMA node of the given CPU, if any. If none, the (non-local) channel with min table_count is returned. nth_chan() is therefore renamed into min_chan() since we don't iterate until the nth channel anymore. In practice, the behavior is the same because first channels are taken first and are then ignored because they got an additional reference. The new code has a slightly higher complexity since we always scan the entire list of channels for finding the minimal table_count (instead of stopping after N chans), and because we check whether the CPU is in the DMA device locality mask. Overall we still have time complexity = number of chans x number of processors. This rebalance is rarely used, so this won't hurt. On the above SuperMicro machine, channels are still allocated the same. On the Dells, there are no locality issue anymore (MEMCPY channel X goes to processor X and to its hyperthread sibling). Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr> Signed-off-by: Dan Williams <djbw@fb.com>
2013-08-19 09:43:35 +00:00
* dma_chan_is_local - returns true if the channel is in the same numa-node as the cpu
*/
static bool dma_chan_is_local(struct dma_chan *chan, int cpu)
{
int node = dev_to_node(chan->device->dev);
return node == -1 || cpumask_test_cpu(cpu, cpumask_of_node(node));
}
/**
* min_chan - returns the channel with min count and in the same numa-node as the cpu
* @cap: capability to match
dmaengine: make dma_channel_rebalance() NUMA aware dma_channel_rebalance() currently distributes channels by processor ID. These IDs often change with the BIOS, and the order isn't related to the DMA channel list (related to PCI bus ids). * On my SuperMicro dual E5 machine, first socket has processor IDs [0-7] (and [16-23] for hyperthreads), second socket has [8-15]+[24-31] => channels are properly allocated to local CPUs. * On Dells R720 with same processors, first socket has even processor IDs, second socket has odd numbers => half the processors get channels on the remote socket, causing cross-NUMA traffic and lower DMA performance. Change nth_chan() to return the channel with min table_count and in the NUMA node of the given CPU, if any. If none, the (non-local) channel with min table_count is returned. nth_chan() is therefore renamed into min_chan() since we don't iterate until the nth channel anymore. In practice, the behavior is the same because first channels are taken first and are then ignored because they got an additional reference. The new code has a slightly higher complexity since we always scan the entire list of channels for finding the minimal table_count (instead of stopping after N chans), and because we check whether the CPU is in the DMA device locality mask. Overall we still have time complexity = number of chans x number of processors. This rebalance is rarely used, so this won't hurt. On the above SuperMicro machine, channels are still allocated the same. On the Dells, there are no locality issue anymore (MEMCPY channel X goes to processor X and to its hyperthread sibling). Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr> Signed-off-by: Dan Williams <djbw@fb.com>
2013-08-19 09:43:35 +00:00
* @cpu: cpu index which the channel should be close to
*
dmaengine: make dma_channel_rebalance() NUMA aware dma_channel_rebalance() currently distributes channels by processor ID. These IDs often change with the BIOS, and the order isn't related to the DMA channel list (related to PCI bus ids). * On my SuperMicro dual E5 machine, first socket has processor IDs [0-7] (and [16-23] for hyperthreads), second socket has [8-15]+[24-31] => channels are properly allocated to local CPUs. * On Dells R720 with same processors, first socket has even processor IDs, second socket has odd numbers => half the processors get channels on the remote socket, causing cross-NUMA traffic and lower DMA performance. Change nth_chan() to return the channel with min table_count and in the NUMA node of the given CPU, if any. If none, the (non-local) channel with min table_count is returned. nth_chan() is therefore renamed into min_chan() since we don't iterate until the nth channel anymore. In practice, the behavior is the same because first channels are taken first and are then ignored because they got an additional reference. The new code has a slightly higher complexity since we always scan the entire list of channels for finding the minimal table_count (instead of stopping after N chans), and because we check whether the CPU is in the DMA device locality mask. Overall we still have time complexity = number of chans x number of processors. This rebalance is rarely used, so this won't hurt. On the above SuperMicro machine, channels are still allocated the same. On the Dells, there are no locality issue anymore (MEMCPY channel X goes to processor X and to its hyperthread sibling). Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr> Signed-off-by: Dan Williams <djbw@fb.com>
2013-08-19 09:43:35 +00:00
* If some channels are close to the given cpu, the one with the lowest
* reference count is returned. Otherwise, cpu is ignored and only the
* reference count is taken into account.
* Must be called under dma_list_mutex.
*/
dmaengine: make dma_channel_rebalance() NUMA aware dma_channel_rebalance() currently distributes channels by processor ID. These IDs often change with the BIOS, and the order isn't related to the DMA channel list (related to PCI bus ids). * On my SuperMicro dual E5 machine, first socket has processor IDs [0-7] (and [16-23] for hyperthreads), second socket has [8-15]+[24-31] => channels are properly allocated to local CPUs. * On Dells R720 with same processors, first socket has even processor IDs, second socket has odd numbers => half the processors get channels on the remote socket, causing cross-NUMA traffic and lower DMA performance. Change nth_chan() to return the channel with min table_count and in the NUMA node of the given CPU, if any. If none, the (non-local) channel with min table_count is returned. nth_chan() is therefore renamed into min_chan() since we don't iterate until the nth channel anymore. In practice, the behavior is the same because first channels are taken first and are then ignored because they got an additional reference. The new code has a slightly higher complexity since we always scan the entire list of channels for finding the minimal table_count (instead of stopping after N chans), and because we check whether the CPU is in the DMA device locality mask. Overall we still have time complexity = number of chans x number of processors. This rebalance is rarely used, so this won't hurt. On the above SuperMicro machine, channels are still allocated the same. On the Dells, there are no locality issue anymore (MEMCPY channel X goes to processor X and to its hyperthread sibling). Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr> Signed-off-by: Dan Williams <djbw@fb.com>
2013-08-19 09:43:35 +00:00
static struct dma_chan *min_chan(enum dma_transaction_type cap, int cpu)
{
struct dma_device *device;
struct dma_chan *chan;
struct dma_chan *min = NULL;
dmaengine: make dma_channel_rebalance() NUMA aware dma_channel_rebalance() currently distributes channels by processor ID. These IDs often change with the BIOS, and the order isn't related to the DMA channel list (related to PCI bus ids). * On my SuperMicro dual E5 machine, first socket has processor IDs [0-7] (and [16-23] for hyperthreads), second socket has [8-15]+[24-31] => channels are properly allocated to local CPUs. * On Dells R720 with same processors, first socket has even processor IDs, second socket has odd numbers => half the processors get channels on the remote socket, causing cross-NUMA traffic and lower DMA performance. Change nth_chan() to return the channel with min table_count and in the NUMA node of the given CPU, if any. If none, the (non-local) channel with min table_count is returned. nth_chan() is therefore renamed into min_chan() since we don't iterate until the nth channel anymore. In practice, the behavior is the same because first channels are taken first and are then ignored because they got an additional reference. The new code has a slightly higher complexity since we always scan the entire list of channels for finding the minimal table_count (instead of stopping after N chans), and because we check whether the CPU is in the DMA device locality mask. Overall we still have time complexity = number of chans x number of processors. This rebalance is rarely used, so this won't hurt. On the above SuperMicro machine, channels are still allocated the same. On the Dells, there are no locality issue anymore (MEMCPY channel X goes to processor X and to its hyperthread sibling). Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr> Signed-off-by: Dan Williams <djbw@fb.com>
2013-08-19 09:43:35 +00:00
struct dma_chan *localmin = NULL;
list_for_each_entry(device, &dma_device_list, global_node) {
if (!dma_has_cap(cap, device->cap_mask) ||
dma_has_cap(DMA_PRIVATE, device->cap_mask))
continue;
list_for_each_entry(chan, &device->channels, device_node) {
if (!chan->client_count)
continue;
dmaengine: make dma_channel_rebalance() NUMA aware dma_channel_rebalance() currently distributes channels by processor ID. These IDs often change with the BIOS, and the order isn't related to the DMA channel list (related to PCI bus ids). * On my SuperMicro dual E5 machine, first socket has processor IDs [0-7] (and [16-23] for hyperthreads), second socket has [8-15]+[24-31] => channels are properly allocated to local CPUs. * On Dells R720 with same processors, first socket has even processor IDs, second socket has odd numbers => half the processors get channels on the remote socket, causing cross-NUMA traffic and lower DMA performance. Change nth_chan() to return the channel with min table_count and in the NUMA node of the given CPU, if any. If none, the (non-local) channel with min table_count is returned. nth_chan() is therefore renamed into min_chan() since we don't iterate until the nth channel anymore. In practice, the behavior is the same because first channels are taken first and are then ignored because they got an additional reference. The new code has a slightly higher complexity since we always scan the entire list of channels for finding the minimal table_count (instead of stopping after N chans), and because we check whether the CPU is in the DMA device locality mask. Overall we still have time complexity = number of chans x number of processors. This rebalance is rarely used, so this won't hurt. On the above SuperMicro machine, channels are still allocated the same. On the Dells, there are no locality issue anymore (MEMCPY channel X goes to processor X and to its hyperthread sibling). Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr> Signed-off-by: Dan Williams <djbw@fb.com>
2013-08-19 09:43:35 +00:00
if (!min || chan->table_count < min->table_count)
min = chan;
dmaengine: make dma_channel_rebalance() NUMA aware dma_channel_rebalance() currently distributes channels by processor ID. These IDs often change with the BIOS, and the order isn't related to the DMA channel list (related to PCI bus ids). * On my SuperMicro dual E5 machine, first socket has processor IDs [0-7] (and [16-23] for hyperthreads), second socket has [8-15]+[24-31] => channels are properly allocated to local CPUs. * On Dells R720 with same processors, first socket has even processor IDs, second socket has odd numbers => half the processors get channels on the remote socket, causing cross-NUMA traffic and lower DMA performance. Change nth_chan() to return the channel with min table_count and in the NUMA node of the given CPU, if any. If none, the (non-local) channel with min table_count is returned. nth_chan() is therefore renamed into min_chan() since we don't iterate until the nth channel anymore. In practice, the behavior is the same because first channels are taken first and are then ignored because they got an additional reference. The new code has a slightly higher complexity since we always scan the entire list of channels for finding the minimal table_count (instead of stopping after N chans), and because we check whether the CPU is in the DMA device locality mask. Overall we still have time complexity = number of chans x number of processors. This rebalance is rarely used, so this won't hurt. On the above SuperMicro machine, channels are still allocated the same. On the Dells, there are no locality issue anymore (MEMCPY channel X goes to processor X and to its hyperthread sibling). Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr> Signed-off-by: Dan Williams <djbw@fb.com>
2013-08-19 09:43:35 +00:00
if (dma_chan_is_local(chan, cpu))
if (!localmin ||
chan->table_count < localmin->table_count)
localmin = chan;
}
}
dmaengine: make dma_channel_rebalance() NUMA aware dma_channel_rebalance() currently distributes channels by processor ID. These IDs often change with the BIOS, and the order isn't related to the DMA channel list (related to PCI bus ids). * On my SuperMicro dual E5 machine, first socket has processor IDs [0-7] (and [16-23] for hyperthreads), second socket has [8-15]+[24-31] => channels are properly allocated to local CPUs. * On Dells R720 with same processors, first socket has even processor IDs, second socket has odd numbers => half the processors get channels on the remote socket, causing cross-NUMA traffic and lower DMA performance. Change nth_chan() to return the channel with min table_count and in the NUMA node of the given CPU, if any. If none, the (non-local) channel with min table_count is returned. nth_chan() is therefore renamed into min_chan() since we don't iterate until the nth channel anymore. In practice, the behavior is the same because first channels are taken first and are then ignored because they got an additional reference. The new code has a slightly higher complexity since we always scan the entire list of channels for finding the minimal table_count (instead of stopping after N chans), and because we check whether the CPU is in the DMA device locality mask. Overall we still have time complexity = number of chans x number of processors. This rebalance is rarely used, so this won't hurt. On the above SuperMicro machine, channels are still allocated the same. On the Dells, there are no locality issue anymore (MEMCPY channel X goes to processor X and to its hyperthread sibling). Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr> Signed-off-by: Dan Williams <djbw@fb.com>
2013-08-19 09:43:35 +00:00
chan = localmin ? localmin : min;
dmaengine: make dma_channel_rebalance() NUMA aware dma_channel_rebalance() currently distributes channels by processor ID. These IDs often change with the BIOS, and the order isn't related to the DMA channel list (related to PCI bus ids). * On my SuperMicro dual E5 machine, first socket has processor IDs [0-7] (and [16-23] for hyperthreads), second socket has [8-15]+[24-31] => channels are properly allocated to local CPUs. * On Dells R720 with same processors, first socket has even processor IDs, second socket has odd numbers => half the processors get channels on the remote socket, causing cross-NUMA traffic and lower DMA performance. Change nth_chan() to return the channel with min table_count and in the NUMA node of the given CPU, if any. If none, the (non-local) channel with min table_count is returned. nth_chan() is therefore renamed into min_chan() since we don't iterate until the nth channel anymore. In practice, the behavior is the same because first channels are taken first and are then ignored because they got an additional reference. The new code has a slightly higher complexity since we always scan the entire list of channels for finding the minimal table_count (instead of stopping after N chans), and because we check whether the CPU is in the DMA device locality mask. Overall we still have time complexity = number of chans x number of processors. This rebalance is rarely used, so this won't hurt. On the above SuperMicro machine, channels are still allocated the same. On the Dells, there are no locality issue anymore (MEMCPY channel X goes to processor X and to its hyperthread sibling). Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr> Signed-off-by: Dan Williams <djbw@fb.com>
2013-08-19 09:43:35 +00:00
if (chan)
chan->table_count++;
dmaengine: make dma_channel_rebalance() NUMA aware dma_channel_rebalance() currently distributes channels by processor ID. These IDs often change with the BIOS, and the order isn't related to the DMA channel list (related to PCI bus ids). * On my SuperMicro dual E5 machine, first socket has processor IDs [0-7] (and [16-23] for hyperthreads), second socket has [8-15]+[24-31] => channels are properly allocated to local CPUs. * On Dells R720 with same processors, first socket has even processor IDs, second socket has odd numbers => half the processors get channels on the remote socket, causing cross-NUMA traffic and lower DMA performance. Change nth_chan() to return the channel with min table_count and in the NUMA node of the given CPU, if any. If none, the (non-local) channel with min table_count is returned. nth_chan() is therefore renamed into min_chan() since we don't iterate until the nth channel anymore. In practice, the behavior is the same because first channels are taken first and are then ignored because they got an additional reference. The new code has a slightly higher complexity since we always scan the entire list of channels for finding the minimal table_count (instead of stopping after N chans), and because we check whether the CPU is in the DMA device locality mask. Overall we still have time complexity = number of chans x number of processors. This rebalance is rarely used, so this won't hurt. On the above SuperMicro machine, channels are still allocated the same. On the Dells, there are no locality issue anymore (MEMCPY channel X goes to processor X and to its hyperthread sibling). Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr> Signed-off-by: Dan Williams <djbw@fb.com>
2013-08-19 09:43:35 +00:00
return chan;
}
/**
* dma_channel_rebalance - redistribute the available channels
*
* Optimize for cpu isolation (each cpu gets a dedicated channel for an
* operation type) in the SMP case, and operation isolation (avoid
* multi-tasking channels) in the non-SMP case. Must be called under
* dma_list_mutex.
*/
static void dma_channel_rebalance(void)
{
struct dma_chan *chan;
struct dma_device *device;
int cpu;
int cap;
/* undo the last distribution */
for_each_dma_cap_mask(cap, dma_cap_mask_all)
for_each_possible_cpu(cpu)
per_cpu_ptr(channel_table[cap], cpu)->chan = NULL;
list_for_each_entry(device, &dma_device_list, global_node) {
if (dma_has_cap(DMA_PRIVATE, device->cap_mask))
continue;
list_for_each_entry(chan, &device->channels, device_node)
chan->table_count = 0;
}
/* don't populate the channel_table if no clients are available */
if (!dmaengine_ref_count)
return;
/* redistribute available channels */
for_each_dma_cap_mask(cap, dma_cap_mask_all)
for_each_online_cpu(cpu) {
dmaengine: make dma_channel_rebalance() NUMA aware dma_channel_rebalance() currently distributes channels by processor ID. These IDs often change with the BIOS, and the order isn't related to the DMA channel list (related to PCI bus ids). * On my SuperMicro dual E5 machine, first socket has processor IDs [0-7] (and [16-23] for hyperthreads), second socket has [8-15]+[24-31] => channels are properly allocated to local CPUs. * On Dells R720 with same processors, first socket has even processor IDs, second socket has odd numbers => half the processors get channels on the remote socket, causing cross-NUMA traffic and lower DMA performance. Change nth_chan() to return the channel with min table_count and in the NUMA node of the given CPU, if any. If none, the (non-local) channel with min table_count is returned. nth_chan() is therefore renamed into min_chan() since we don't iterate until the nth channel anymore. In practice, the behavior is the same because first channels are taken first and are then ignored because they got an additional reference. The new code has a slightly higher complexity since we always scan the entire list of channels for finding the minimal table_count (instead of stopping after N chans), and because we check whether the CPU is in the DMA device locality mask. Overall we still have time complexity = number of chans x number of processors. This rebalance is rarely used, so this won't hurt. On the above SuperMicro machine, channels are still allocated the same. On the Dells, there are no locality issue anymore (MEMCPY channel X goes to processor X and to its hyperthread sibling). Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr> Signed-off-by: Dan Williams <djbw@fb.com>
2013-08-19 09:43:35 +00:00
chan = min_chan(cap, cpu);
per_cpu_ptr(channel_table[cap], cpu)->chan = chan;
}
}
int dma_get_slave_caps(struct dma_chan *chan, struct dma_slave_caps *caps)
{
struct dma_device *device;
if (!chan || !caps)
return -EINVAL;
device = chan->device;
/* check if the channel supports slave transactions */
if (!(test_bit(DMA_SLAVE, device->cap_mask.bits) ||
test_bit(DMA_CYCLIC, device->cap_mask.bits)))
return -ENXIO;
/*
* Check whether it reports it uses the generic slave
* capabilities, if not, that means it doesn't support any
* kind of slave capabilities reporting.
*/
if (!device->directions)
return -ENXIO;
caps->src_addr_widths = device->src_addr_widths;
caps->dst_addr_widths = device->dst_addr_widths;
caps->directions = device->directions;
caps->max_burst = device->max_burst;
caps->residue_granularity = device->residue_granularity;
caps->descriptor_reuse = device->descriptor_reuse;
caps->cmd_pause = !!device->device_pause;
caps->cmd_resume = !!device->device_resume;
caps->cmd_terminate = !!device->device_terminate_all;
return 0;
}
EXPORT_SYMBOL_GPL(dma_get_slave_caps);
static struct dma_chan *private_candidate(const dma_cap_mask_t *mask,
struct dma_device *dev,
dma_filter_fn fn, void *fn_param)
{
struct dma_chan *chan;
if (mask && !__dma_device_satisfies_mask(dev, mask)) {
dev_dbg(dev->dev, "%s: wrong capabilities\n", __func__);
return NULL;
}
/* devices with multiple channels need special handling as we need to
* ensure that all channels are either private or public.
*/
if (dev->chancnt > 1 && !dma_has_cap(DMA_PRIVATE, dev->cap_mask))
list_for_each_entry(chan, &dev->channels, device_node) {
/* some channels are already publicly allocated */
if (chan->client_count)
return NULL;
}
list_for_each_entry(chan, &dev->channels, device_node) {
if (chan->client_count) {
dev_dbg(dev->dev, "%s: %s busy\n",
__func__, dma_chan_name(chan));
continue;
}
if (fn && !fn(chan, fn_param)) {
dev_dbg(dev->dev, "%s: %s filter said false\n",
__func__, dma_chan_name(chan));
continue;
}
return chan;
}
return NULL;
}
static struct dma_chan *find_candidate(struct dma_device *device,
const dma_cap_mask_t *mask,
dma_filter_fn fn, void *fn_param)
{
struct dma_chan *chan = private_candidate(mask, device, fn, fn_param);
int err;
if (chan) {
/* Found a suitable channel, try to grab, prep, and return it.
* We first set DMA_PRIVATE to disable balance_ref_count as this
* channel will not be published in the general-purpose
* allocator
*/
dma_cap_set(DMA_PRIVATE, device->cap_mask);
device->privatecnt++;
err = dma_chan_get(chan);
if (err) {
if (err == -ENODEV) {
dev_dbg(device->dev, "%s: %s module removed\n",
__func__, dma_chan_name(chan));
list_del_rcu(&device->global_node);
} else
dev_dbg(device->dev,
"%s: failed to get %s: (%d)\n",
__func__, dma_chan_name(chan), err);
if (--device->privatecnt == 0)
dma_cap_clear(DMA_PRIVATE, device->cap_mask);
chan = ERR_PTR(err);
}
}
return chan ? chan : ERR_PTR(-EPROBE_DEFER);
}
/**
* dma_get_slave_channel - try to get specific channel exclusively
* @chan: target channel
*/
struct dma_chan *dma_get_slave_channel(struct dma_chan *chan)
{
int err = -EBUSY;
/* lock against __dma_request_channel */
mutex_lock(&dma_list_mutex);
if (chan->client_count == 0) {
struct dma_device *device = chan->device;
dma_cap_set(DMA_PRIVATE, device->cap_mask);
device->privatecnt++;
err = dma_chan_get(chan);
if (err) {
dev_dbg(chan->device->dev,
"%s: failed to get %s: (%d)\n",
__func__, dma_chan_name(chan), err);
chan = NULL;
if (--device->privatecnt == 0)
dma_cap_clear(DMA_PRIVATE, device->cap_mask);
}
} else
chan = NULL;
mutex_unlock(&dma_list_mutex);
return chan;
}
EXPORT_SYMBOL_GPL(dma_get_slave_channel);
struct dma_chan *dma_get_any_slave_channel(struct dma_device *device)
{
dma_cap_mask_t mask;
struct dma_chan *chan;
dma_cap_zero(mask);
dma_cap_set(DMA_SLAVE, mask);
/* lock against __dma_request_channel */
mutex_lock(&dma_list_mutex);
chan = find_candidate(device, &mask, NULL, NULL);
mutex_unlock(&dma_list_mutex);
return IS_ERR(chan) ? NULL : chan;
}
EXPORT_SYMBOL_GPL(dma_get_any_slave_channel);
/**
* __dma_request_channel - try to allocate an exclusive channel
* @mask: capabilities that the channel must satisfy
* @fn: optional callback to disposition available channels
* @fn_param: opaque parameter to pass to dma_filter_fn
dma: add channel request API that supports deferred probe dma_request_slave_channel() simply returns NULL whenever DMA channel lookup fails. Lookup could fail for two distinct reasons: a) No DMA specification exists for the channel name. This includes situations where no DMA specifications exist at all, or other general lookup problems. b) A DMA specification does exist, yet the driver for that channel is not yet registered. Case (b) should trigger deferred probe in client drivers. However, since they have no way to differentiate the two situations, it cannot. Implement new function dma_request_slave_channel_reason(), which performs identically to dma_request_slave_channel(), except that it returns an error-pointer rather than NULL, which allows callers to detect when deferred probe should occur. Eventually, all drivers should be converted to this new API, the old API removed, and the new API renamed to the more desirable name. This patch doesn't convert the existing API and all drivers in one go, since some drivers call dma_request_slave_channel() then dma_request_channel() if that fails. That would require either modifying dma_request_channel() in the same way, or adding extra error-handling code to all affected drivers, and there are close to 100 drivers using the other API, rather than just the 15-20 or so that use dma_request_slave_channel(), which might be tenable in a single patch. acpi_dma_request_slave_chan_by_name() doesn't currently implement deferred probe. It should, but this will be addressed later. Acked-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Stephen Warren <swarren@nvidia.com> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2013-11-26 17:04:22 +00:00
*
* Returns pointer to appropriate DMA channel on success or NULL.
*/
struct dma_chan *__dma_request_channel(const dma_cap_mask_t *mask,
dma_filter_fn fn, void *fn_param)
{
struct dma_device *device, *_d;
struct dma_chan *chan = NULL;
/* Find a channel */
mutex_lock(&dma_list_mutex);
list_for_each_entry_safe(device, _d, &dma_device_list, global_node) {
chan = find_candidate(device, mask, fn, fn_param);
if (!IS_ERR(chan))
break;
chan = NULL;
}
mutex_unlock(&dma_list_mutex);
pr_debug("%s: %s (%s)\n",
__func__,
chan ? "success" : "fail",
chan ? dma_chan_name(chan) : NULL);
return chan;
}
EXPORT_SYMBOL_GPL(__dma_request_channel);
dmaengine: core: Introduce new, universal API to request a channel The two API function can cover most, if not all current APIs used to request a channel. With minimal effort dmaengine drivers, platforms and dmaengine user drivers can be converted to use the two function. struct dma_chan *dma_request_chan_by_mask(const dma_cap_mask_t *mask); To request any channel matching with the requested capabilities, can be used to request channel for memcpy, memset, xor, etc where no hardware synchronization is needed. struct dma_chan *dma_request_chan(struct device *dev, const char *name); To request a slave channel. The dma_request_chan() will try to find the channel via DT, ACPI or in case if the kernel booted in non DT/ACPI mode it will use a filter lookup table and retrieves the needed information from the dma_slave_map provided by the DMA drivers. This legacy mode needs changes in platform code, in dmaengine drivers and finally the dmaengine user drivers can be converted: For each dmaengine driver an array of DMA device, slave and the parameter for the filter function needs to be added: static const struct dma_slave_map da830_edma_map[] = { { "davinci-mcasp.0", "rx", EDMA_FILTER_PARAM(0, 0) }, { "davinci-mcasp.0", "tx", EDMA_FILTER_PARAM(0, 1) }, { "davinci-mcasp.1", "rx", EDMA_FILTER_PARAM(0, 2) }, { "davinci-mcasp.1", "tx", EDMA_FILTER_PARAM(0, 3) }, { "davinci-mcasp.2", "rx", EDMA_FILTER_PARAM(0, 4) }, { "davinci-mcasp.2", "tx", EDMA_FILTER_PARAM(0, 5) }, { "spi_davinci.0", "rx", EDMA_FILTER_PARAM(0, 14) }, { "spi_davinci.0", "tx", EDMA_FILTER_PARAM(0, 15) }, { "da830-mmc.0", "rx", EDMA_FILTER_PARAM(0, 16) }, { "da830-mmc.0", "tx", EDMA_FILTER_PARAM(0, 17) }, { "spi_davinci.1", "rx", EDMA_FILTER_PARAM(0, 18) }, { "spi_davinci.1", "tx", EDMA_FILTER_PARAM(0, 19) }, }; This information is going to be needed by the dmaengine driver, so modification to the platform_data is needed, and the driver map should be added to the pdata of the DMA driver: da8xx_edma0_pdata.slave_map = da830_edma_map; da8xx_edma0_pdata.slavecnt = ARRAY_SIZE(da830_edma_map); The DMA driver then needs to configure the needed device -> filter_fn mapping before it registers with dma_async_device_register() : ecc->dma_slave.filter_map.map = info->slave_map; ecc->dma_slave.filter_map.mapcnt = info->slavecnt; ecc->dma_slave.filter_map.fn = edma_filter_fn; When neither DT or ACPI lookup is available the dma_request_chan() will try to match the requester's device name with the filter_map's list of device names, when a match found it will use the information from the dma_slave_map to get the channel with the dma_get_channel() internal function. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com> Reviewed-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-12-14 20:47:40 +00:00
static const struct dma_slave_map *dma_filter_match(struct dma_device *device,
const char *name,
struct device *dev)
{
int i;
if (!device->filter.mapcnt)
return NULL;
for (i = 0; i < device->filter.mapcnt; i++) {
const struct dma_slave_map *map = &device->filter.map[i];
if (!strcmp(map->devname, dev_name(dev)) &&
!strcmp(map->slave, name))
return map;
}
return NULL;
}
/**
dmaengine: core: Introduce new, universal API to request a channel The two API function can cover most, if not all current APIs used to request a channel. With minimal effort dmaengine drivers, platforms and dmaengine user drivers can be converted to use the two function. struct dma_chan *dma_request_chan_by_mask(const dma_cap_mask_t *mask); To request any channel matching with the requested capabilities, can be used to request channel for memcpy, memset, xor, etc where no hardware synchronization is needed. struct dma_chan *dma_request_chan(struct device *dev, const char *name); To request a slave channel. The dma_request_chan() will try to find the channel via DT, ACPI or in case if the kernel booted in non DT/ACPI mode it will use a filter lookup table and retrieves the needed information from the dma_slave_map provided by the DMA drivers. This legacy mode needs changes in platform code, in dmaengine drivers and finally the dmaengine user drivers can be converted: For each dmaengine driver an array of DMA device, slave and the parameter for the filter function needs to be added: static const struct dma_slave_map da830_edma_map[] = { { "davinci-mcasp.0", "rx", EDMA_FILTER_PARAM(0, 0) }, { "davinci-mcasp.0", "tx", EDMA_FILTER_PARAM(0, 1) }, { "davinci-mcasp.1", "rx", EDMA_FILTER_PARAM(0, 2) }, { "davinci-mcasp.1", "tx", EDMA_FILTER_PARAM(0, 3) }, { "davinci-mcasp.2", "rx", EDMA_FILTER_PARAM(0, 4) }, { "davinci-mcasp.2", "tx", EDMA_FILTER_PARAM(0, 5) }, { "spi_davinci.0", "rx", EDMA_FILTER_PARAM(0, 14) }, { "spi_davinci.0", "tx", EDMA_FILTER_PARAM(0, 15) }, { "da830-mmc.0", "rx", EDMA_FILTER_PARAM(0, 16) }, { "da830-mmc.0", "tx", EDMA_FILTER_PARAM(0, 17) }, { "spi_davinci.1", "rx", EDMA_FILTER_PARAM(0, 18) }, { "spi_davinci.1", "tx", EDMA_FILTER_PARAM(0, 19) }, }; This information is going to be needed by the dmaengine driver, so modification to the platform_data is needed, and the driver map should be added to the pdata of the DMA driver: da8xx_edma0_pdata.slave_map = da830_edma_map; da8xx_edma0_pdata.slavecnt = ARRAY_SIZE(da830_edma_map); The DMA driver then needs to configure the needed device -> filter_fn mapping before it registers with dma_async_device_register() : ecc->dma_slave.filter_map.map = info->slave_map; ecc->dma_slave.filter_map.mapcnt = info->slavecnt; ecc->dma_slave.filter_map.fn = edma_filter_fn; When neither DT or ACPI lookup is available the dma_request_chan() will try to match the requester's device name with the filter_map's list of device names, when a match found it will use the information from the dma_slave_map to get the channel with the dma_get_channel() internal function. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com> Reviewed-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-12-14 20:47:40 +00:00
* dma_request_chan - try to allocate an exclusive slave channel
* @dev: pointer to client device structure
* @name: slave channel name
dma: add channel request API that supports deferred probe dma_request_slave_channel() simply returns NULL whenever DMA channel lookup fails. Lookup could fail for two distinct reasons: a) No DMA specification exists for the channel name. This includes situations where no DMA specifications exist at all, or other general lookup problems. b) A DMA specification does exist, yet the driver for that channel is not yet registered. Case (b) should trigger deferred probe in client drivers. However, since they have no way to differentiate the two situations, it cannot. Implement new function dma_request_slave_channel_reason(), which performs identically to dma_request_slave_channel(), except that it returns an error-pointer rather than NULL, which allows callers to detect when deferred probe should occur. Eventually, all drivers should be converted to this new API, the old API removed, and the new API renamed to the more desirable name. This patch doesn't convert the existing API and all drivers in one go, since some drivers call dma_request_slave_channel() then dma_request_channel() if that fails. That would require either modifying dma_request_channel() in the same way, or adding extra error-handling code to all affected drivers, and there are close to 100 drivers using the other API, rather than just the 15-20 or so that use dma_request_slave_channel(), which might be tenable in a single patch. acpi_dma_request_slave_chan_by_name() doesn't currently implement deferred probe. It should, but this will be addressed later. Acked-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Stephen Warren <swarren@nvidia.com> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2013-11-26 17:04:22 +00:00
*
* Returns pointer to appropriate DMA channel on success or an error pointer.
*/
dmaengine: core: Introduce new, universal API to request a channel The two API function can cover most, if not all current APIs used to request a channel. With minimal effort dmaengine drivers, platforms and dmaengine user drivers can be converted to use the two function. struct dma_chan *dma_request_chan_by_mask(const dma_cap_mask_t *mask); To request any channel matching with the requested capabilities, can be used to request channel for memcpy, memset, xor, etc where no hardware synchronization is needed. struct dma_chan *dma_request_chan(struct device *dev, const char *name); To request a slave channel. The dma_request_chan() will try to find the channel via DT, ACPI or in case if the kernel booted in non DT/ACPI mode it will use a filter lookup table and retrieves the needed information from the dma_slave_map provided by the DMA drivers. This legacy mode needs changes in platform code, in dmaengine drivers and finally the dmaengine user drivers can be converted: For each dmaengine driver an array of DMA device, slave and the parameter for the filter function needs to be added: static const struct dma_slave_map da830_edma_map[] = { { "davinci-mcasp.0", "rx", EDMA_FILTER_PARAM(0, 0) }, { "davinci-mcasp.0", "tx", EDMA_FILTER_PARAM(0, 1) }, { "davinci-mcasp.1", "rx", EDMA_FILTER_PARAM(0, 2) }, { "davinci-mcasp.1", "tx", EDMA_FILTER_PARAM(0, 3) }, { "davinci-mcasp.2", "rx", EDMA_FILTER_PARAM(0, 4) }, { "davinci-mcasp.2", "tx", EDMA_FILTER_PARAM(0, 5) }, { "spi_davinci.0", "rx", EDMA_FILTER_PARAM(0, 14) }, { "spi_davinci.0", "tx", EDMA_FILTER_PARAM(0, 15) }, { "da830-mmc.0", "rx", EDMA_FILTER_PARAM(0, 16) }, { "da830-mmc.0", "tx", EDMA_FILTER_PARAM(0, 17) }, { "spi_davinci.1", "rx", EDMA_FILTER_PARAM(0, 18) }, { "spi_davinci.1", "tx", EDMA_FILTER_PARAM(0, 19) }, }; This information is going to be needed by the dmaengine driver, so modification to the platform_data is needed, and the driver map should be added to the pdata of the DMA driver: da8xx_edma0_pdata.slave_map = da830_edma_map; da8xx_edma0_pdata.slavecnt = ARRAY_SIZE(da830_edma_map); The DMA driver then needs to configure the needed device -> filter_fn mapping before it registers with dma_async_device_register() : ecc->dma_slave.filter_map.map = info->slave_map; ecc->dma_slave.filter_map.mapcnt = info->slavecnt; ecc->dma_slave.filter_map.fn = edma_filter_fn; When neither DT or ACPI lookup is available the dma_request_chan() will try to match the requester's device name with the filter_map's list of device names, when a match found it will use the information from the dma_slave_map to get the channel with the dma_get_channel() internal function. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com> Reviewed-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-12-14 20:47:40 +00:00
struct dma_chan *dma_request_chan(struct device *dev, const char *name)
{
dmaengine: core: Introduce new, universal API to request a channel The two API function can cover most, if not all current APIs used to request a channel. With minimal effort dmaengine drivers, platforms and dmaengine user drivers can be converted to use the two function. struct dma_chan *dma_request_chan_by_mask(const dma_cap_mask_t *mask); To request any channel matching with the requested capabilities, can be used to request channel for memcpy, memset, xor, etc where no hardware synchronization is needed. struct dma_chan *dma_request_chan(struct device *dev, const char *name); To request a slave channel. The dma_request_chan() will try to find the channel via DT, ACPI or in case if the kernel booted in non DT/ACPI mode it will use a filter lookup table and retrieves the needed information from the dma_slave_map provided by the DMA drivers. This legacy mode needs changes in platform code, in dmaengine drivers and finally the dmaengine user drivers can be converted: For each dmaengine driver an array of DMA device, slave and the parameter for the filter function needs to be added: static const struct dma_slave_map da830_edma_map[] = { { "davinci-mcasp.0", "rx", EDMA_FILTER_PARAM(0, 0) }, { "davinci-mcasp.0", "tx", EDMA_FILTER_PARAM(0, 1) }, { "davinci-mcasp.1", "rx", EDMA_FILTER_PARAM(0, 2) }, { "davinci-mcasp.1", "tx", EDMA_FILTER_PARAM(0, 3) }, { "davinci-mcasp.2", "rx", EDMA_FILTER_PARAM(0, 4) }, { "davinci-mcasp.2", "tx", EDMA_FILTER_PARAM(0, 5) }, { "spi_davinci.0", "rx", EDMA_FILTER_PARAM(0, 14) }, { "spi_davinci.0", "tx", EDMA_FILTER_PARAM(0, 15) }, { "da830-mmc.0", "rx", EDMA_FILTER_PARAM(0, 16) }, { "da830-mmc.0", "tx", EDMA_FILTER_PARAM(0, 17) }, { "spi_davinci.1", "rx", EDMA_FILTER_PARAM(0, 18) }, { "spi_davinci.1", "tx", EDMA_FILTER_PARAM(0, 19) }, }; This information is going to be needed by the dmaengine driver, so modification to the platform_data is needed, and the driver map should be added to the pdata of the DMA driver: da8xx_edma0_pdata.slave_map = da830_edma_map; da8xx_edma0_pdata.slavecnt = ARRAY_SIZE(da830_edma_map); The DMA driver then needs to configure the needed device -> filter_fn mapping before it registers with dma_async_device_register() : ecc->dma_slave.filter_map.map = info->slave_map; ecc->dma_slave.filter_map.mapcnt = info->slavecnt; ecc->dma_slave.filter_map.fn = edma_filter_fn; When neither DT or ACPI lookup is available the dma_request_chan() will try to match the requester's device name with the filter_map's list of device names, when a match found it will use the information from the dma_slave_map to get the channel with the dma_get_channel() internal function. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com> Reviewed-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-12-14 20:47:40 +00:00
struct dma_device *d, *_d;
struct dma_chan *chan = NULL;
/* If device-tree is present get slave info from here */
if (dev->of_node)
dmaengine: core: Introduce new, universal API to request a channel The two API function can cover most, if not all current APIs used to request a channel. With minimal effort dmaengine drivers, platforms and dmaengine user drivers can be converted to use the two function. struct dma_chan *dma_request_chan_by_mask(const dma_cap_mask_t *mask); To request any channel matching with the requested capabilities, can be used to request channel for memcpy, memset, xor, etc where no hardware synchronization is needed. struct dma_chan *dma_request_chan(struct device *dev, const char *name); To request a slave channel. The dma_request_chan() will try to find the channel via DT, ACPI or in case if the kernel booted in non DT/ACPI mode it will use a filter lookup table and retrieves the needed information from the dma_slave_map provided by the DMA drivers. This legacy mode needs changes in platform code, in dmaengine drivers and finally the dmaengine user drivers can be converted: For each dmaengine driver an array of DMA device, slave and the parameter for the filter function needs to be added: static const struct dma_slave_map da830_edma_map[] = { { "davinci-mcasp.0", "rx", EDMA_FILTER_PARAM(0, 0) }, { "davinci-mcasp.0", "tx", EDMA_FILTER_PARAM(0, 1) }, { "davinci-mcasp.1", "rx", EDMA_FILTER_PARAM(0, 2) }, { "davinci-mcasp.1", "tx", EDMA_FILTER_PARAM(0, 3) }, { "davinci-mcasp.2", "rx", EDMA_FILTER_PARAM(0, 4) }, { "davinci-mcasp.2", "tx", EDMA_FILTER_PARAM(0, 5) }, { "spi_davinci.0", "rx", EDMA_FILTER_PARAM(0, 14) }, { "spi_davinci.0", "tx", EDMA_FILTER_PARAM(0, 15) }, { "da830-mmc.0", "rx", EDMA_FILTER_PARAM(0, 16) }, { "da830-mmc.0", "tx", EDMA_FILTER_PARAM(0, 17) }, { "spi_davinci.1", "rx", EDMA_FILTER_PARAM(0, 18) }, { "spi_davinci.1", "tx", EDMA_FILTER_PARAM(0, 19) }, }; This information is going to be needed by the dmaengine driver, so modification to the platform_data is needed, and the driver map should be added to the pdata of the DMA driver: da8xx_edma0_pdata.slave_map = da830_edma_map; da8xx_edma0_pdata.slavecnt = ARRAY_SIZE(da830_edma_map); The DMA driver then needs to configure the needed device -> filter_fn mapping before it registers with dma_async_device_register() : ecc->dma_slave.filter_map.map = info->slave_map; ecc->dma_slave.filter_map.mapcnt = info->slavecnt; ecc->dma_slave.filter_map.fn = edma_filter_fn; When neither DT or ACPI lookup is available the dma_request_chan() will try to match the requester's device name with the filter_map's list of device names, when a match found it will use the information from the dma_slave_map to get the channel with the dma_get_channel() internal function. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com> Reviewed-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-12-14 20:47:40 +00:00
chan = of_dma_request_slave_channel(dev->of_node, name);
/* If device was enumerated by ACPI get slave info from here */
dmaengine: core: Introduce new, universal API to request a channel The two API function can cover most, if not all current APIs used to request a channel. With minimal effort dmaengine drivers, platforms and dmaengine user drivers can be converted to use the two function. struct dma_chan *dma_request_chan_by_mask(const dma_cap_mask_t *mask); To request any channel matching with the requested capabilities, can be used to request channel for memcpy, memset, xor, etc where no hardware synchronization is needed. struct dma_chan *dma_request_chan(struct device *dev, const char *name); To request a slave channel. The dma_request_chan() will try to find the channel via DT, ACPI or in case if the kernel booted in non DT/ACPI mode it will use a filter lookup table and retrieves the needed information from the dma_slave_map provided by the DMA drivers. This legacy mode needs changes in platform code, in dmaengine drivers and finally the dmaengine user drivers can be converted: For each dmaengine driver an array of DMA device, slave and the parameter for the filter function needs to be added: static const struct dma_slave_map da830_edma_map[] = { { "davinci-mcasp.0", "rx", EDMA_FILTER_PARAM(0, 0) }, { "davinci-mcasp.0", "tx", EDMA_FILTER_PARAM(0, 1) }, { "davinci-mcasp.1", "rx", EDMA_FILTER_PARAM(0, 2) }, { "davinci-mcasp.1", "tx", EDMA_FILTER_PARAM(0, 3) }, { "davinci-mcasp.2", "rx", EDMA_FILTER_PARAM(0, 4) }, { "davinci-mcasp.2", "tx", EDMA_FILTER_PARAM(0, 5) }, { "spi_davinci.0", "rx", EDMA_FILTER_PARAM(0, 14) }, { "spi_davinci.0", "tx", EDMA_FILTER_PARAM(0, 15) }, { "da830-mmc.0", "rx", EDMA_FILTER_PARAM(0, 16) }, { "da830-mmc.0", "tx", EDMA_FILTER_PARAM(0, 17) }, { "spi_davinci.1", "rx", EDMA_FILTER_PARAM(0, 18) }, { "spi_davinci.1", "tx", EDMA_FILTER_PARAM(0, 19) }, }; This information is going to be needed by the dmaengine driver, so modification to the platform_data is needed, and the driver map should be added to the pdata of the DMA driver: da8xx_edma0_pdata.slave_map = da830_edma_map; da8xx_edma0_pdata.slavecnt = ARRAY_SIZE(da830_edma_map); The DMA driver then needs to configure the needed device -> filter_fn mapping before it registers with dma_async_device_register() : ecc->dma_slave.filter_map.map = info->slave_map; ecc->dma_slave.filter_map.mapcnt = info->slavecnt; ecc->dma_slave.filter_map.fn = edma_filter_fn; When neither DT or ACPI lookup is available the dma_request_chan() will try to match the requester's device name with the filter_map's list of device names, when a match found it will use the information from the dma_slave_map to get the channel with the dma_get_channel() internal function. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com> Reviewed-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-12-14 20:47:40 +00:00
if (has_acpi_companion(dev) && !chan)
chan = acpi_dma_request_slave_chan_by_name(dev, name);
if (chan) {
/* Valid channel found or requester need to be deferred */
if (!IS_ERR(chan) || PTR_ERR(chan) == -EPROBE_DEFER)
return chan;
}
/* Try to find the channel via the DMA filter map(s) */
mutex_lock(&dma_list_mutex);
list_for_each_entry_safe(d, _d, &dma_device_list, global_node) {
dma_cap_mask_t mask;
const struct dma_slave_map *map = dma_filter_match(d, name, dev);
dmaengine: core: Introduce new, universal API to request a channel The two API function can cover most, if not all current APIs used to request a channel. With minimal effort dmaengine drivers, platforms and dmaengine user drivers can be converted to use the two function. struct dma_chan *dma_request_chan_by_mask(const dma_cap_mask_t *mask); To request any channel matching with the requested capabilities, can be used to request channel for memcpy, memset, xor, etc where no hardware synchronization is needed. struct dma_chan *dma_request_chan(struct device *dev, const char *name); To request a slave channel. The dma_request_chan() will try to find the channel via DT, ACPI or in case if the kernel booted in non DT/ACPI mode it will use a filter lookup table and retrieves the needed information from the dma_slave_map provided by the DMA drivers. This legacy mode needs changes in platform code, in dmaengine drivers and finally the dmaengine user drivers can be converted: For each dmaengine driver an array of DMA device, slave and the parameter for the filter function needs to be added: static const struct dma_slave_map da830_edma_map[] = { { "davinci-mcasp.0", "rx", EDMA_FILTER_PARAM(0, 0) }, { "davinci-mcasp.0", "tx", EDMA_FILTER_PARAM(0, 1) }, { "davinci-mcasp.1", "rx", EDMA_FILTER_PARAM(0, 2) }, { "davinci-mcasp.1", "tx", EDMA_FILTER_PARAM(0, 3) }, { "davinci-mcasp.2", "rx", EDMA_FILTER_PARAM(0, 4) }, { "davinci-mcasp.2", "tx", EDMA_FILTER_PARAM(0, 5) }, { "spi_davinci.0", "rx", EDMA_FILTER_PARAM(0, 14) }, { "spi_davinci.0", "tx", EDMA_FILTER_PARAM(0, 15) }, { "da830-mmc.0", "rx", EDMA_FILTER_PARAM(0, 16) }, { "da830-mmc.0", "tx", EDMA_FILTER_PARAM(0, 17) }, { "spi_davinci.1", "rx", EDMA_FILTER_PARAM(0, 18) }, { "spi_davinci.1", "tx", EDMA_FILTER_PARAM(0, 19) }, }; This information is going to be needed by the dmaengine driver, so modification to the platform_data is needed, and the driver map should be added to the pdata of the DMA driver: da8xx_edma0_pdata.slave_map = da830_edma_map; da8xx_edma0_pdata.slavecnt = ARRAY_SIZE(da830_edma_map); The DMA driver then needs to configure the needed device -> filter_fn mapping before it registers with dma_async_device_register() : ecc->dma_slave.filter_map.map = info->slave_map; ecc->dma_slave.filter_map.mapcnt = info->slavecnt; ecc->dma_slave.filter_map.fn = edma_filter_fn; When neither DT or ACPI lookup is available the dma_request_chan() will try to match the requester's device name with the filter_map's list of device names, when a match found it will use the information from the dma_slave_map to get the channel with the dma_get_channel() internal function. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com> Reviewed-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-12-14 20:47:40 +00:00
if (!map)
continue;
dma_cap_zero(mask);
dma_cap_set(DMA_SLAVE, mask);
dmaengine: core: Introduce new, universal API to request a channel The two API function can cover most, if not all current APIs used to request a channel. With minimal effort dmaengine drivers, platforms and dmaengine user drivers can be converted to use the two function. struct dma_chan *dma_request_chan_by_mask(const dma_cap_mask_t *mask); To request any channel matching with the requested capabilities, can be used to request channel for memcpy, memset, xor, etc where no hardware synchronization is needed. struct dma_chan *dma_request_chan(struct device *dev, const char *name); To request a slave channel. The dma_request_chan() will try to find the channel via DT, ACPI or in case if the kernel booted in non DT/ACPI mode it will use a filter lookup table and retrieves the needed information from the dma_slave_map provided by the DMA drivers. This legacy mode needs changes in platform code, in dmaengine drivers and finally the dmaengine user drivers can be converted: For each dmaengine driver an array of DMA device, slave and the parameter for the filter function needs to be added: static const struct dma_slave_map da830_edma_map[] = { { "davinci-mcasp.0", "rx", EDMA_FILTER_PARAM(0, 0) }, { "davinci-mcasp.0", "tx", EDMA_FILTER_PARAM(0, 1) }, { "davinci-mcasp.1", "rx", EDMA_FILTER_PARAM(0, 2) }, { "davinci-mcasp.1", "tx", EDMA_FILTER_PARAM(0, 3) }, { "davinci-mcasp.2", "rx", EDMA_FILTER_PARAM(0, 4) }, { "davinci-mcasp.2", "tx", EDMA_FILTER_PARAM(0, 5) }, { "spi_davinci.0", "rx", EDMA_FILTER_PARAM(0, 14) }, { "spi_davinci.0", "tx", EDMA_FILTER_PARAM(0, 15) }, { "da830-mmc.0", "rx", EDMA_FILTER_PARAM(0, 16) }, { "da830-mmc.0", "tx", EDMA_FILTER_PARAM(0, 17) }, { "spi_davinci.1", "rx", EDMA_FILTER_PARAM(0, 18) }, { "spi_davinci.1", "tx", EDMA_FILTER_PARAM(0, 19) }, }; This information is going to be needed by the dmaengine driver, so modification to the platform_data is needed, and the driver map should be added to the pdata of the DMA driver: da8xx_edma0_pdata.slave_map = da830_edma_map; da8xx_edma0_pdata.slavecnt = ARRAY_SIZE(da830_edma_map); The DMA driver then needs to configure the needed device -> filter_fn mapping before it registers with dma_async_device_register() : ecc->dma_slave.filter_map.map = info->slave_map; ecc->dma_slave.filter_map.mapcnt = info->slavecnt; ecc->dma_slave.filter_map.fn = edma_filter_fn; When neither DT or ACPI lookup is available the dma_request_chan() will try to match the requester's device name with the filter_map's list of device names, when a match found it will use the information from the dma_slave_map to get the channel with the dma_get_channel() internal function. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com> Reviewed-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-12-14 20:47:40 +00:00
chan = find_candidate(d, &mask, d->filter.fn, map->param);
if (!IS_ERR(chan))
break;
}
mutex_unlock(&dma_list_mutex);
return chan ? chan : ERR_PTR(-EPROBE_DEFER);
dma: add channel request API that supports deferred probe dma_request_slave_channel() simply returns NULL whenever DMA channel lookup fails. Lookup could fail for two distinct reasons: a) No DMA specification exists for the channel name. This includes situations where no DMA specifications exist at all, or other general lookup problems. b) A DMA specification does exist, yet the driver for that channel is not yet registered. Case (b) should trigger deferred probe in client drivers. However, since they have no way to differentiate the two situations, it cannot. Implement new function dma_request_slave_channel_reason(), which performs identically to dma_request_slave_channel(), except that it returns an error-pointer rather than NULL, which allows callers to detect when deferred probe should occur. Eventually, all drivers should be converted to this new API, the old API removed, and the new API renamed to the more desirable name. This patch doesn't convert the existing API and all drivers in one go, since some drivers call dma_request_slave_channel() then dma_request_channel() if that fails. That would require either modifying dma_request_channel() in the same way, or adding extra error-handling code to all affected drivers, and there are close to 100 drivers using the other API, rather than just the 15-20 or so that use dma_request_slave_channel(), which might be tenable in a single patch. acpi_dma_request_slave_chan_by_name() doesn't currently implement deferred probe. It should, but this will be addressed later. Acked-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Stephen Warren <swarren@nvidia.com> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2013-11-26 17:04:22 +00:00
}
dmaengine: core: Introduce new, universal API to request a channel The two API function can cover most, if not all current APIs used to request a channel. With minimal effort dmaengine drivers, platforms and dmaengine user drivers can be converted to use the two function. struct dma_chan *dma_request_chan_by_mask(const dma_cap_mask_t *mask); To request any channel matching with the requested capabilities, can be used to request channel for memcpy, memset, xor, etc where no hardware synchronization is needed. struct dma_chan *dma_request_chan(struct device *dev, const char *name); To request a slave channel. The dma_request_chan() will try to find the channel via DT, ACPI or in case if the kernel booted in non DT/ACPI mode it will use a filter lookup table and retrieves the needed information from the dma_slave_map provided by the DMA drivers. This legacy mode needs changes in platform code, in dmaengine drivers and finally the dmaengine user drivers can be converted: For each dmaengine driver an array of DMA device, slave and the parameter for the filter function needs to be added: static const struct dma_slave_map da830_edma_map[] = { { "davinci-mcasp.0", "rx", EDMA_FILTER_PARAM(0, 0) }, { "davinci-mcasp.0", "tx", EDMA_FILTER_PARAM(0, 1) }, { "davinci-mcasp.1", "rx", EDMA_FILTER_PARAM(0, 2) }, { "davinci-mcasp.1", "tx", EDMA_FILTER_PARAM(0, 3) }, { "davinci-mcasp.2", "rx", EDMA_FILTER_PARAM(0, 4) }, { "davinci-mcasp.2", "tx", EDMA_FILTER_PARAM(0, 5) }, { "spi_davinci.0", "rx", EDMA_FILTER_PARAM(0, 14) }, { "spi_davinci.0", "tx", EDMA_FILTER_PARAM(0, 15) }, { "da830-mmc.0", "rx", EDMA_FILTER_PARAM(0, 16) }, { "da830-mmc.0", "tx", EDMA_FILTER_PARAM(0, 17) }, { "spi_davinci.1", "rx", EDMA_FILTER_PARAM(0, 18) }, { "spi_davinci.1", "tx", EDMA_FILTER_PARAM(0, 19) }, }; This information is going to be needed by the dmaengine driver, so modification to the platform_data is needed, and the driver map should be added to the pdata of the DMA driver: da8xx_edma0_pdata.slave_map = da830_edma_map; da8xx_edma0_pdata.slavecnt = ARRAY_SIZE(da830_edma_map); The DMA driver then needs to configure the needed device -> filter_fn mapping before it registers with dma_async_device_register() : ecc->dma_slave.filter_map.map = info->slave_map; ecc->dma_slave.filter_map.mapcnt = info->slavecnt; ecc->dma_slave.filter_map.fn = edma_filter_fn; When neither DT or ACPI lookup is available the dma_request_chan() will try to match the requester's device name with the filter_map's list of device names, when a match found it will use the information from the dma_slave_map to get the channel with the dma_get_channel() internal function. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com> Reviewed-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-12-14 20:47:40 +00:00
EXPORT_SYMBOL_GPL(dma_request_chan);
dma: add channel request API that supports deferred probe dma_request_slave_channel() simply returns NULL whenever DMA channel lookup fails. Lookup could fail for two distinct reasons: a) No DMA specification exists for the channel name. This includes situations where no DMA specifications exist at all, or other general lookup problems. b) A DMA specification does exist, yet the driver for that channel is not yet registered. Case (b) should trigger deferred probe in client drivers. However, since they have no way to differentiate the two situations, it cannot. Implement new function dma_request_slave_channel_reason(), which performs identically to dma_request_slave_channel(), except that it returns an error-pointer rather than NULL, which allows callers to detect when deferred probe should occur. Eventually, all drivers should be converted to this new API, the old API removed, and the new API renamed to the more desirable name. This patch doesn't convert the existing API and all drivers in one go, since some drivers call dma_request_slave_channel() then dma_request_channel() if that fails. That would require either modifying dma_request_channel() in the same way, or adding extra error-handling code to all affected drivers, and there are close to 100 drivers using the other API, rather than just the 15-20 or so that use dma_request_slave_channel(), which might be tenable in a single patch. acpi_dma_request_slave_chan_by_name() doesn't currently implement deferred probe. It should, but this will be addressed later. Acked-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Stephen Warren <swarren@nvidia.com> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2013-11-26 17:04:22 +00:00
/**
* dma_request_slave_channel - try to allocate an exclusive slave channel
* @dev: pointer to client device structure
* @name: slave channel name
*
* Returns pointer to appropriate DMA channel on success or NULL.
*/
struct dma_chan *dma_request_slave_channel(struct device *dev,
const char *name)
{
dmaengine: core: Introduce new, universal API to request a channel The two API function can cover most, if not all current APIs used to request a channel. With minimal effort dmaengine drivers, platforms and dmaengine user drivers can be converted to use the two function. struct dma_chan *dma_request_chan_by_mask(const dma_cap_mask_t *mask); To request any channel matching with the requested capabilities, can be used to request channel for memcpy, memset, xor, etc where no hardware synchronization is needed. struct dma_chan *dma_request_chan(struct device *dev, const char *name); To request a slave channel. The dma_request_chan() will try to find the channel via DT, ACPI or in case if the kernel booted in non DT/ACPI mode it will use a filter lookup table and retrieves the needed information from the dma_slave_map provided by the DMA drivers. This legacy mode needs changes in platform code, in dmaengine drivers and finally the dmaengine user drivers can be converted: For each dmaengine driver an array of DMA device, slave and the parameter for the filter function needs to be added: static const struct dma_slave_map da830_edma_map[] = { { "davinci-mcasp.0", "rx", EDMA_FILTER_PARAM(0, 0) }, { "davinci-mcasp.0", "tx", EDMA_FILTER_PARAM(0, 1) }, { "davinci-mcasp.1", "rx", EDMA_FILTER_PARAM(0, 2) }, { "davinci-mcasp.1", "tx", EDMA_FILTER_PARAM(0, 3) }, { "davinci-mcasp.2", "rx", EDMA_FILTER_PARAM(0, 4) }, { "davinci-mcasp.2", "tx", EDMA_FILTER_PARAM(0, 5) }, { "spi_davinci.0", "rx", EDMA_FILTER_PARAM(0, 14) }, { "spi_davinci.0", "tx", EDMA_FILTER_PARAM(0, 15) }, { "da830-mmc.0", "rx", EDMA_FILTER_PARAM(0, 16) }, { "da830-mmc.0", "tx", EDMA_FILTER_PARAM(0, 17) }, { "spi_davinci.1", "rx", EDMA_FILTER_PARAM(0, 18) }, { "spi_davinci.1", "tx", EDMA_FILTER_PARAM(0, 19) }, }; This information is going to be needed by the dmaengine driver, so modification to the platform_data is needed, and the driver map should be added to the pdata of the DMA driver: da8xx_edma0_pdata.slave_map = da830_edma_map; da8xx_edma0_pdata.slavecnt = ARRAY_SIZE(da830_edma_map); The DMA driver then needs to configure the needed device -> filter_fn mapping before it registers with dma_async_device_register() : ecc->dma_slave.filter_map.map = info->slave_map; ecc->dma_slave.filter_map.mapcnt = info->slavecnt; ecc->dma_slave.filter_map.fn = edma_filter_fn; When neither DT or ACPI lookup is available the dma_request_chan() will try to match the requester's device name with the filter_map's list of device names, when a match found it will use the information from the dma_slave_map to get the channel with the dma_get_channel() internal function. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com> Reviewed-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-12-14 20:47:40 +00:00
struct dma_chan *ch = dma_request_chan(dev, name);
dma: add channel request API that supports deferred probe dma_request_slave_channel() simply returns NULL whenever DMA channel lookup fails. Lookup could fail for two distinct reasons: a) No DMA specification exists for the channel name. This includes situations where no DMA specifications exist at all, or other general lookup problems. b) A DMA specification does exist, yet the driver for that channel is not yet registered. Case (b) should trigger deferred probe in client drivers. However, since they have no way to differentiate the two situations, it cannot. Implement new function dma_request_slave_channel_reason(), which performs identically to dma_request_slave_channel(), except that it returns an error-pointer rather than NULL, which allows callers to detect when deferred probe should occur. Eventually, all drivers should be converted to this new API, the old API removed, and the new API renamed to the more desirable name. This patch doesn't convert the existing API and all drivers in one go, since some drivers call dma_request_slave_channel() then dma_request_channel() if that fails. That would require either modifying dma_request_channel() in the same way, or adding extra error-handling code to all affected drivers, and there are close to 100 drivers using the other API, rather than just the 15-20 or so that use dma_request_slave_channel(), which might be tenable in a single patch. acpi_dma_request_slave_chan_by_name() doesn't currently implement deferred probe. It should, but this will be addressed later. Acked-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Stephen Warren <swarren@nvidia.com> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2013-11-26 17:04:22 +00:00
if (IS_ERR(ch))
return NULL;
dma: add channel request API that supports deferred probe dma_request_slave_channel() simply returns NULL whenever DMA channel lookup fails. Lookup could fail for two distinct reasons: a) No DMA specification exists for the channel name. This includes situations where no DMA specifications exist at all, or other general lookup problems. b) A DMA specification does exist, yet the driver for that channel is not yet registered. Case (b) should trigger deferred probe in client drivers. However, since they have no way to differentiate the two situations, it cannot. Implement new function dma_request_slave_channel_reason(), which performs identically to dma_request_slave_channel(), except that it returns an error-pointer rather than NULL, which allows callers to detect when deferred probe should occur. Eventually, all drivers should be converted to this new API, the old API removed, and the new API renamed to the more desirable name. This patch doesn't convert the existing API and all drivers in one go, since some drivers call dma_request_slave_channel() then dma_request_channel() if that fails. That would require either modifying dma_request_channel() in the same way, or adding extra error-handling code to all affected drivers, and there are close to 100 drivers using the other API, rather than just the 15-20 or so that use dma_request_slave_channel(), which might be tenable in a single patch. acpi_dma_request_slave_chan_by_name() doesn't currently implement deferred probe. It should, but this will be addressed later. Acked-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Stephen Warren <swarren@nvidia.com> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2013-11-26 17:04:22 +00:00
return ch;
}
EXPORT_SYMBOL_GPL(dma_request_slave_channel);
dmaengine: core: Introduce new, universal API to request a channel The two API function can cover most, if not all current APIs used to request a channel. With minimal effort dmaengine drivers, platforms and dmaengine user drivers can be converted to use the two function. struct dma_chan *dma_request_chan_by_mask(const dma_cap_mask_t *mask); To request any channel matching with the requested capabilities, can be used to request channel for memcpy, memset, xor, etc where no hardware synchronization is needed. struct dma_chan *dma_request_chan(struct device *dev, const char *name); To request a slave channel. The dma_request_chan() will try to find the channel via DT, ACPI or in case if the kernel booted in non DT/ACPI mode it will use a filter lookup table and retrieves the needed information from the dma_slave_map provided by the DMA drivers. This legacy mode needs changes in platform code, in dmaengine drivers and finally the dmaengine user drivers can be converted: For each dmaengine driver an array of DMA device, slave and the parameter for the filter function needs to be added: static const struct dma_slave_map da830_edma_map[] = { { "davinci-mcasp.0", "rx", EDMA_FILTER_PARAM(0, 0) }, { "davinci-mcasp.0", "tx", EDMA_FILTER_PARAM(0, 1) }, { "davinci-mcasp.1", "rx", EDMA_FILTER_PARAM(0, 2) }, { "davinci-mcasp.1", "tx", EDMA_FILTER_PARAM(0, 3) }, { "davinci-mcasp.2", "rx", EDMA_FILTER_PARAM(0, 4) }, { "davinci-mcasp.2", "tx", EDMA_FILTER_PARAM(0, 5) }, { "spi_davinci.0", "rx", EDMA_FILTER_PARAM(0, 14) }, { "spi_davinci.0", "tx", EDMA_FILTER_PARAM(0, 15) }, { "da830-mmc.0", "rx", EDMA_FILTER_PARAM(0, 16) }, { "da830-mmc.0", "tx", EDMA_FILTER_PARAM(0, 17) }, { "spi_davinci.1", "rx", EDMA_FILTER_PARAM(0, 18) }, { "spi_davinci.1", "tx", EDMA_FILTER_PARAM(0, 19) }, }; This information is going to be needed by the dmaengine driver, so modification to the platform_data is needed, and the driver map should be added to the pdata of the DMA driver: da8xx_edma0_pdata.slave_map = da830_edma_map; da8xx_edma0_pdata.slavecnt = ARRAY_SIZE(da830_edma_map); The DMA driver then needs to configure the needed device -> filter_fn mapping before it registers with dma_async_device_register() : ecc->dma_slave.filter_map.map = info->slave_map; ecc->dma_slave.filter_map.mapcnt = info->slavecnt; ecc->dma_slave.filter_map.fn = edma_filter_fn; When neither DT or ACPI lookup is available the dma_request_chan() will try to match the requester's device name with the filter_map's list of device names, when a match found it will use the information from the dma_slave_map to get the channel with the dma_get_channel() internal function. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com> Reviewed-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-12-14 20:47:40 +00:00
/**
* dma_request_chan_by_mask - allocate a channel satisfying certain capabilities
* @mask: capabilities that the channel must satisfy
*
* Returns pointer to appropriate DMA channel on success or an error pointer.
*/
struct dma_chan *dma_request_chan_by_mask(const dma_cap_mask_t *mask)
{
struct dma_chan *chan;
if (!mask)
return ERR_PTR(-ENODEV);
chan = __dma_request_channel(mask, NULL, NULL);
if (!chan) {
mutex_lock(&dma_list_mutex);
if (list_empty(&dma_device_list))
chan = ERR_PTR(-EPROBE_DEFER);
else
chan = ERR_PTR(-ENODEV);
mutex_unlock(&dma_list_mutex);
}
dmaengine: core: Introduce new, universal API to request a channel The two API function can cover most, if not all current APIs used to request a channel. With minimal effort dmaengine drivers, platforms and dmaengine user drivers can be converted to use the two function. struct dma_chan *dma_request_chan_by_mask(const dma_cap_mask_t *mask); To request any channel matching with the requested capabilities, can be used to request channel for memcpy, memset, xor, etc where no hardware synchronization is needed. struct dma_chan *dma_request_chan(struct device *dev, const char *name); To request a slave channel. The dma_request_chan() will try to find the channel via DT, ACPI or in case if the kernel booted in non DT/ACPI mode it will use a filter lookup table and retrieves the needed information from the dma_slave_map provided by the DMA drivers. This legacy mode needs changes in platform code, in dmaengine drivers and finally the dmaengine user drivers can be converted: For each dmaengine driver an array of DMA device, slave and the parameter for the filter function needs to be added: static const struct dma_slave_map da830_edma_map[] = { { "davinci-mcasp.0", "rx", EDMA_FILTER_PARAM(0, 0) }, { "davinci-mcasp.0", "tx", EDMA_FILTER_PARAM(0, 1) }, { "davinci-mcasp.1", "rx", EDMA_FILTER_PARAM(0, 2) }, { "davinci-mcasp.1", "tx", EDMA_FILTER_PARAM(0, 3) }, { "davinci-mcasp.2", "rx", EDMA_FILTER_PARAM(0, 4) }, { "davinci-mcasp.2", "tx", EDMA_FILTER_PARAM(0, 5) }, { "spi_davinci.0", "rx", EDMA_FILTER_PARAM(0, 14) }, { "spi_davinci.0", "tx", EDMA_FILTER_PARAM(0, 15) }, { "da830-mmc.0", "rx", EDMA_FILTER_PARAM(0, 16) }, { "da830-mmc.0", "tx", EDMA_FILTER_PARAM(0, 17) }, { "spi_davinci.1", "rx", EDMA_FILTER_PARAM(0, 18) }, { "spi_davinci.1", "tx", EDMA_FILTER_PARAM(0, 19) }, }; This information is going to be needed by the dmaengine driver, so modification to the platform_data is needed, and the driver map should be added to the pdata of the DMA driver: da8xx_edma0_pdata.slave_map = da830_edma_map; da8xx_edma0_pdata.slavecnt = ARRAY_SIZE(da830_edma_map); The DMA driver then needs to configure the needed device -> filter_fn mapping before it registers with dma_async_device_register() : ecc->dma_slave.filter_map.map = info->slave_map; ecc->dma_slave.filter_map.mapcnt = info->slavecnt; ecc->dma_slave.filter_map.fn = edma_filter_fn; When neither DT or ACPI lookup is available the dma_request_chan() will try to match the requester's device name with the filter_map's list of device names, when a match found it will use the information from the dma_slave_map to get the channel with the dma_get_channel() internal function. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com> Reviewed-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Vinod Koul <vinod.koul@intel.com>
2015-12-14 20:47:40 +00:00
return chan;
}
EXPORT_SYMBOL_GPL(dma_request_chan_by_mask);
void dma_release_channel(struct dma_chan *chan)
{
mutex_lock(&dma_list_mutex);
WARN_ONCE(chan->client_count != 1,
"chan reference count %d != 1\n", chan->client_count);
dma_chan_put(chan);
/* drop PRIVATE cap enabled by __dma_request_channel() */
if (--chan->device->privatecnt == 0)
dma_cap_clear(DMA_PRIVATE, chan->device->cap_mask);
mutex_unlock(&dma_list_mutex);
}
EXPORT_SYMBOL_GPL(dma_release_channel);
/**
* dmaengine_get - register interest in dma_channels
*/
void dmaengine_get(void)
{
struct dma_device *device, *_d;
struct dma_chan *chan;
int err;
mutex_lock(&dma_list_mutex);
dmaengine_ref_count++;
/* try to grab channels */
list_for_each_entry_safe(device, _d, &dma_device_list, global_node) {
if (dma_has_cap(DMA_PRIVATE, device->cap_mask))
continue;
list_for_each_entry(chan, &device->channels, device_node) {
err = dma_chan_get(chan);
if (err == -ENODEV) {
/* module removed before we could use it */
list_del_rcu(&device->global_node);
break;
} else if (err)
dev_dbg(chan->device->dev,
"%s: failed to get %s: (%d)\n",
__func__, dma_chan_name(chan), err);
}
}
/* if this is the first reference and there were channels
* waiting we need to rebalance to get those channels
* incorporated into the channel table
*/
if (dmaengine_ref_count == 1)
dma_channel_rebalance();
mutex_unlock(&dma_list_mutex);
}
EXPORT_SYMBOL(dmaengine_get);
/**
* dmaengine_put - let dma drivers be removed when ref_count == 0
*/
void dmaengine_put(void)
{
struct dma_device *device;
struct dma_chan *chan;
mutex_lock(&dma_list_mutex);
dmaengine_ref_count--;
BUG_ON(dmaengine_ref_count < 0);
/* drop channel references */
list_for_each_entry(device, &dma_device_list, global_node) {
if (dma_has_cap(DMA_PRIVATE, device->cap_mask))
continue;
list_for_each_entry(chan, &device->channels, device_node)
dma_chan_put(chan);
}
mutex_unlock(&dma_list_mutex);
}
EXPORT_SYMBOL(dmaengine_put);
static bool device_has_all_tx_types(struct dma_device *device)
{
/* A device that satisfies this test has channels that will never cause
* an async_tx channel switch event as all possible operation types can
* be handled.
*/
#ifdef CONFIG_ASYNC_TX_DMA
if (!dma_has_cap(DMA_INTERRUPT, device->cap_mask))
return false;
#endif
#if IS_ENABLED(CONFIG_ASYNC_MEMCPY)
if (!dma_has_cap(DMA_MEMCPY, device->cap_mask))
return false;
#endif
#if IS_ENABLED(CONFIG_ASYNC_XOR)
if (!dma_has_cap(DMA_XOR, device->cap_mask))
return false;
#ifndef CONFIG_ASYNC_TX_DISABLE_XOR_VAL_DMA
if (!dma_has_cap(DMA_XOR_VAL, device->cap_mask))
return false;
#endif
#endif
#if IS_ENABLED(CONFIG_ASYNC_PQ)
if (!dma_has_cap(DMA_PQ, device->cap_mask))
return false;
#ifndef CONFIG_ASYNC_TX_DISABLE_PQ_VAL_DMA
if (!dma_has_cap(DMA_PQ_VAL, device->cap_mask))
return false;
#endif
#endif
return true;
}
static int get_dma_id(struct dma_device *device)
{
int rc;
do {
if (!ida_pre_get(&dma_ida, GFP_KERNEL))
return -ENOMEM;
mutex_lock(&dma_list_mutex);
rc = ida_get_new(&dma_ida, &device->dev_id);
mutex_unlock(&dma_list_mutex);
} while (rc == -EAGAIN);
return rc;
}
/**
* dma_async_device_register - registers DMA devices found
* @device: &dma_device
*/
int dma_async_device_register(struct dma_device *device)
{
int chancnt = 0, rc;
struct dma_chan* chan;
atomic_t *idr_ref;
if (!device)
return -ENODEV;
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
/* validate device routines */
if (!device->dev) {
pr_err("DMAdevice must have dev\n");
return -EIO;
}
if (dma_has_cap(DMA_MEMCPY, device->cap_mask) && !device->device_prep_dma_memcpy) {
dev_err(device->dev,
"Device claims capability %s, but op is not defined\n",
"DMA_MEMCPY");
return -EIO;
}
if (dma_has_cap(DMA_XOR, device->cap_mask) && !device->device_prep_dma_xor) {
dev_err(device->dev,
"Device claims capability %s, but op is not defined\n",
"DMA_XOR");
return -EIO;
}
if (dma_has_cap(DMA_XOR_VAL, device->cap_mask) && !device->device_prep_dma_xor_val) {
dev_err(device->dev,
"Device claims capability %s, but op is not defined\n",
"DMA_XOR_VAL");
return -EIO;
}
if (dma_has_cap(DMA_PQ, device->cap_mask) && !device->device_prep_dma_pq) {
dev_err(device->dev,
"Device claims capability %s, but op is not defined\n",
"DMA_PQ");
return -EIO;
}
if (dma_has_cap(DMA_PQ_VAL, device->cap_mask) && !device->device_prep_dma_pq_val) {
dev_err(device->dev,
"Device claims capability %s, but op is not defined\n",
"DMA_PQ_VAL");
return -EIO;
}
if (dma_has_cap(DMA_MEMSET, device->cap_mask) && !device->device_prep_dma_memset) {
dev_err(device->dev,
"Device claims capability %s, but op is not defined\n",
"DMA_MEMSET");
return -EIO;
}
if (dma_has_cap(DMA_INTERRUPT, device->cap_mask) && !device->device_prep_dma_interrupt) {
dev_err(device->dev,
"Device claims capability %s, but op is not defined\n",
"DMA_INTERRUPT");
return -EIO;
}
if (dma_has_cap(DMA_CYCLIC, device->cap_mask) && !device->device_prep_dma_cyclic) {
dev_err(device->dev,
"Device claims capability %s, but op is not defined\n",
"DMA_CYCLIC");
return -EIO;
}
if (dma_has_cap(DMA_INTERLEAVE, device->cap_mask) && !device->device_prep_interleaved_dma) {
dev_err(device->dev,
"Device claims capability %s, but op is not defined\n",
"DMA_INTERLEAVE");
return -EIO;
}
if (!device->device_tx_status) {
dev_err(device->dev, "Device tx_status is not defined\n");
return -EIO;
}
if (!device->device_issue_pending) {
dev_err(device->dev, "Device issue_pending is not defined\n");
return -EIO;
}
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
/* note: this only matters in the
* CONFIG_ASYNC_TX_ENABLE_CHANNEL_SWITCH=n case
*/
if (device_has_all_tx_types(device))
dma_cap_set(DMA_ASYNC_TX, device->cap_mask);
idr_ref = kmalloc(sizeof(*idr_ref), GFP_KERNEL);
if (!idr_ref)
return -ENOMEM;
rc = get_dma_id(device);
if (rc != 0) {
kfree(idr_ref);
return rc;
}
atomic_set(idr_ref, 0);
/* represent channels in sysfs. Probably want devs too */
list_for_each_entry(chan, &device->channels, device_node) {
rc = -ENOMEM;
chan->local = alloc_percpu(typeof(*chan->local));
if (chan->local == NULL)
goto err_out;
chan->dev = kzalloc(sizeof(*chan->dev), GFP_KERNEL);
if (chan->dev == NULL) {
free_percpu(chan->local);
chan->local = NULL;
goto err_out;
}
chan->chan_id = chancnt++;
chan->dev->device.class = &dma_devclass;
chan->dev->device.parent = device->dev;
chan->dev->chan = chan;
chan->dev->idr_ref = idr_ref;
chan->dev->dev_id = device->dev_id;
atomic_inc(idr_ref);
dev_set_name(&chan->dev->device, "dma%dchan%d",
device->dev_id, chan->chan_id);
rc = device_register(&chan->dev->device);
if (rc) {
free_percpu(chan->local);
chan->local = NULL;
kfree(chan->dev);
atomic_dec(idr_ref);
goto err_out;
}
chan->client_count = 0;
}
2016-07-27 21:32:58 +00:00
if (!chancnt) {
dev_err(device->dev, "%s: device has no channels!\n", __func__);
rc = -ENODEV;
goto err_out;
}
device->chancnt = chancnt;
mutex_lock(&dma_list_mutex);
/* take references on public channels */
if (dmaengine_ref_count && !dma_has_cap(DMA_PRIVATE, device->cap_mask))
list_for_each_entry(chan, &device->channels, device_node) {
/* if clients are already waiting for channels we need
* to take references on their behalf
*/
if (dma_chan_get(chan) == -ENODEV) {
/* note we can only get here for the first
* channel as the remaining channels are
* guaranteed to get a reference
*/
rc = -ENODEV;
mutex_unlock(&dma_list_mutex);
goto err_out;
}
}
list_add_tail_rcu(&device->global_node, &dma_device_list);
if (dma_has_cap(DMA_PRIVATE, device->cap_mask))
device->privatecnt++; /* Always private */
dma_channel_rebalance();
mutex_unlock(&dma_list_mutex);
return 0;
err_out:
/* if we never registered a channel just release the idr */
if (atomic_read(idr_ref) == 0) {
mutex_lock(&dma_list_mutex);
ida_remove(&dma_ida, device->dev_id);
mutex_unlock(&dma_list_mutex);
kfree(idr_ref);
return rc;
}
list_for_each_entry(chan, &device->channels, device_node) {
if (chan->local == NULL)
continue;
mutex_lock(&dma_list_mutex);
chan->dev->chan = NULL;
mutex_unlock(&dma_list_mutex);
device_unregister(&chan->dev->device);
free_percpu(chan->local);
}
return rc;
}
EXPORT_SYMBOL(dma_async_device_register);
/**
* dma_async_device_unregister - unregister a DMA device
* @device: &dma_device
*
* This routine is called by dma driver exit routines, dmaengine holds module
* references to prevent it being called while channels are in use.
*/
void dma_async_device_unregister(struct dma_device *device)
{
struct dma_chan *chan;
mutex_lock(&dma_list_mutex);
list_del_rcu(&device->global_node);
dma_channel_rebalance();
mutex_unlock(&dma_list_mutex);
list_for_each_entry(chan, &device->channels, device_node) {
WARN_ONCE(chan->client_count,
"%s called while %d clients hold a reference\n",
__func__, chan->client_count);
mutex_lock(&dma_list_mutex);
chan->dev->chan = NULL;
mutex_unlock(&dma_list_mutex);
device_unregister(&chan->dev->device);
free_percpu(chan->local);
}
}
EXPORT_SYMBOL(dma_async_device_unregister);
struct dmaengine_unmap_pool {
struct kmem_cache *cache;
const char *name;
mempool_t *pool;
size_t size;
};
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
#define __UNMAP_POOL(x) { .size = x, .name = "dmaengine-unmap-" __stringify(x) }
static struct dmaengine_unmap_pool unmap_pool[] = {
__UNMAP_POOL(2),
#if IS_ENABLED(CONFIG_DMA_ENGINE_RAID)
__UNMAP_POOL(16),
__UNMAP_POOL(128),
__UNMAP_POOL(256),
#endif
};
static struct dmaengine_unmap_pool *__get_unmap_pool(int nr)
{
int order = get_count_order(nr);
switch (order) {
case 0 ... 1:
return &unmap_pool[0];
#if IS_ENABLED(CONFIG_DMA_ENGINE_RAID)
case 2 ... 4:
return &unmap_pool[1];
case 5 ... 7:
return &unmap_pool[2];
case 8:
return &unmap_pool[3];
#endif
default:
BUG();
return NULL;
}
}
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
static void dmaengine_unmap(struct kref *kref)
{
struct dmaengine_unmap_data *unmap = container_of(kref, typeof(*unmap), kref);
struct device *dev = unmap->dev;
int cnt, i;
cnt = unmap->to_cnt;
for (i = 0; i < cnt; i++)
dma_unmap_page(dev, unmap->addr[i], unmap->len,
DMA_TO_DEVICE);
cnt += unmap->from_cnt;
for (; i < cnt; i++)
dma_unmap_page(dev, unmap->addr[i], unmap->len,
DMA_FROM_DEVICE);
cnt += unmap->bidi_cnt;
for (; i < cnt; i++) {
if (unmap->addr[i] == 0)
continue;
dma_unmap_page(dev, unmap->addr[i], unmap->len,
DMA_BIDIRECTIONAL);
}
cnt = unmap->map_cnt;
mempool_free(unmap, __get_unmap_pool(cnt)->pool);
}
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
void dmaengine_unmap_put(struct dmaengine_unmap_data *unmap)
{
if (unmap)
kref_put(&unmap->kref, dmaengine_unmap);
}
EXPORT_SYMBOL_GPL(dmaengine_unmap_put);
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
static void dmaengine_destroy_unmap_pool(void)
{
int i;
for (i = 0; i < ARRAY_SIZE(unmap_pool); i++) {
struct dmaengine_unmap_pool *p = &unmap_pool[i];
mempool_destroy(p->pool);
p->pool = NULL;
kmem_cache_destroy(p->cache);
p->cache = NULL;
}
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
}
static int __init dmaengine_init_unmap_pool(void)
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
{
int i;
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
for (i = 0; i < ARRAY_SIZE(unmap_pool); i++) {
struct dmaengine_unmap_pool *p = &unmap_pool[i];
size_t size;
size = sizeof(struct dmaengine_unmap_data) +
sizeof(dma_addr_t) * p->size;
p->cache = kmem_cache_create(p->name, size, 0,
SLAB_HWCACHE_ALIGN, NULL);
if (!p->cache)
break;
p->pool = mempool_create_slab_pool(1, p->cache);
if (!p->pool)
break;
}
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
if (i == ARRAY_SIZE(unmap_pool))
return 0;
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
dmaengine_destroy_unmap_pool();
return -ENOMEM;
}
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
struct dmaengine_unmap_data *
dmaengine_get_unmap_data(struct device *dev, int nr, gfp_t flags)
{
struct dmaengine_unmap_data *unmap;
unmap = mempool_alloc(__get_unmap_pool(nr)->pool, flags);
if (!unmap)
return NULL;
memset(unmap, 0, sizeof(*unmap));
kref_init(&unmap->kref);
unmap->dev = dev;
unmap->map_cnt = nr;
return unmap;
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
}
EXPORT_SYMBOL(dmaengine_get_unmap_data);
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
void dma_async_tx_descriptor_init(struct dma_async_tx_descriptor *tx,
struct dma_chan *chan)
{
tx->chan = chan;
#ifdef CONFIG_ASYNC_TX_ENABLE_CHANNEL_SWITCH
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
spin_lock_init(&tx->lock);
#endif
dmaengine: refactor dmaengine around dma_async_tx_descriptor The current dmaengine interface defines mutliple routines per operation, i.e. dma_async_memcpy_buf_to_buf, dma_async_memcpy_buf_to_page etc. Adding more operation types (xor, crc, etc) to this model would result in an unmanageable number of method permutations. Are we really going to add a set of hooks for each DMA engine whizbang feature? - Jeff Garzik The descriptor creation process is refactored using the new common dma_async_tx_descriptor structure. Instead of per driver do_<operation>_<dest>_to_<src> methods, drivers integrate dma_async_tx_descriptor into their private software descriptor and then define a 'prep' routine per operation. The prep routine allocates a descriptor and ensures that the tx_set_src, tx_set_dest, tx_submit routines are valid. Descriptor creation and submission becomes: struct dma_device *dev; struct dma_chan *chan; struct dma_async_tx_descriptor *tx; tx = dev->device_prep_dma_<operation>(chan, len, int_flag) tx->tx_set_src(dma_addr_t, tx, index /* for multi-source ops */) tx->tx_set_dest(dma_addr_t, tx, index) tx->tx_submit(tx) In addition to the refactoring, dma_async_tx_descriptor also lays the groundwork for definining cross-channel-operation dependencies, and a callback facility for asynchronous notification of operation completion. Changelog: * drop dma mapping methods, suggested by Chris Leech * fix ioat_dma_dependency_added, also caught by Andrew Morton * fix dma_sync_wait, change from Andrew Morton * uninline large functions, change from Andrew Morton * add tx->callback = NULL to dmaengine calls to interoperate with async_tx calls * hookup ioat_tx_submit * convert channel capabilities to a 'cpumask_t like' bitmap * removed DMA_TX_ARRAY_INIT, no longer needed * checkpatch.pl fixes * make set_src, set_dest, and tx_submit descriptor specific methods * fixup git-ioat merge * move group_list and phys to dma_async_tx_descriptor Cc: Jeff Garzik <jeff@garzik.org> Cc: Chris Leech <christopher.leech@intel.com> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-by: David S. Miller <davem@davemloft.net>
2007-01-02 18:10:43 +00:00
}
EXPORT_SYMBOL(dma_async_tx_descriptor_init);
/* dma_wait_for_async_tx - spin wait for a transaction to complete
* @tx: in-flight transaction to wait on
*/
enum dma_status
dma_wait_for_async_tx(struct dma_async_tx_descriptor *tx)
{
unsigned long dma_sync_wait_timeout = jiffies + msecs_to_jiffies(5000);
if (!tx)
return DMA_COMPLETE;
while (tx->cookie == -EBUSY) {
if (time_after_eq(jiffies, dma_sync_wait_timeout)) {
dev_err(tx->chan->device->dev,
"%s timeout waiting for descriptor submission\n",
__func__);
return DMA_ERROR;
}
cpu_relax();
}
return dma_sync_wait(tx->chan, tx->cookie);
}
EXPORT_SYMBOL_GPL(dma_wait_for_async_tx);
/* dma_run_dependencies - helper routine for dma drivers to process
* (start) dependent operations on their target channel
* @tx: transaction with dependencies
*/
void dma_run_dependencies(struct dma_async_tx_descriptor *tx)
{
struct dma_async_tx_descriptor *dep = txd_next(tx);
struct dma_async_tx_descriptor *dep_next;
struct dma_chan *chan;
if (!dep)
return;
/* we'll submit tx->next now, so clear the link */
txd_clear_next(tx);
chan = dep->chan;
/* keep submitting up until a channel switch is detected
* in that case we will be called again as a result of
* processing the interrupt from async_tx_channel_switch
*/
for (; dep; dep = dep_next) {
txd_lock(dep);
txd_clear_parent(dep);
dep_next = txd_next(dep);
if (dep_next && dep_next->chan == chan)
txd_clear_next(dep); /* ->next will be submitted */
else
dep_next = NULL; /* submit current dep and terminate */
txd_unlock(dep);
dep->tx_submit(dep);
}
chan->device->device_issue_pending(chan);
}
EXPORT_SYMBOL_GPL(dma_run_dependencies);
static int __init dma_bus_init(void)
{
int err = dmaengine_init_unmap_pool();
if (err)
return err;
return class_register(&dma_devclass);
}
arch_initcall(dma_bus_init);