| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <20070502061640.7066.55802.stgit@dwillia2-linux.ch.intel.com>
Date: Tue, 01 May 2007 23:16:41 -0700
From: Dan Williams <dan.j.williams@...el.com>
To: neilb@...e.de, akpm@...ux-foundation.org,
christopher.leech@...el.com
Cc: linux-kernel@...r.kernel.org, linux-raid@...r.kernel.org
Subject: [PATCH 04/16] dmaengine: add the async_tx api
The async_tx api provides methods for describing a chain of asynchronous
bulk memory transfers/transforms with support for inter-transactional
dependencies. It is implemented as a dmaengine client that smooths over
the details of different hardware offload engine implementations. Code
that is written to the api can optimize for asynchrnous operation and
the api will fit the chain of operations to the available offload
resources.
Currently the raid5 implementation in the MD raid456 driver has been
converted to the async_tx api. A driver for the offload engines on the
Intel Xscale series of I/O processors, iop-adma, is provided. With the
iop-adma driver and async_tx, raid456 is able to offload copy, xor, and
xor-zero-sum operations to hardware engines.
On iop342 tiobench showed higher throughput for sequential writes (20 -
30% improvement) and sequential reads to a degraded array (40 - 55%
improvement). For the other cases performance was roughly equal, +/- a
few percentage points. On a x86-smp platform the performance of the
async_tx implementation (in synchronous mode) was also +/- a few
percentage points of the original implementation. According to 'top'
CPU utilization was positively affected in the offload case, but exact
measurements have yet to be taken.
The tiobench command line used for testing was: tiobench --size 2048
--block 4096 --block 131072 --dir /mnt/raid --numruns 5 * iop342 had 1GB
of memory available
Xor operations are handled by async_tx, to this end xor.c is moved into
drivers/dma and is changed to take an explicit destination address and a
series of sources to match the hardware engine implementation.
When CONFIG_DMA_ENGINE is not set the asynchrounous path is compiled
away.
Changelog:
* fixed a leftover debug print
* don't allow callbacks in async_interrupt_cond
* fixed xor_block changes
* fixed usage of ASYNC_TX_XOR_DROP_DEST
* drop dma mapping methods, suggested by Chris Leech
* printk warning fixups from Andrew Morton
* don't use inline in C files, Adrian Bunk
* select the API when MD is enabled
* BUG_ON xor source counts <= 1
* implicitly handle hardware concerns like channel switching and
interrupts, Neil Brown
* remove the per operation type list, and distribute operation capabilities
evenly amongst the available channels
* simplify async_tx_find_channel to optimize the fast path
Signed-off-by: Dan Williams <dan.j.williams@...el.com>
---
drivers/Makefile | 1
drivers/dma/Kconfig | 15 +
drivers/dma/Makefile | 1
drivers/dma/async_tx.c | 889 ++++++++++++++++++++++++++++++++++++++++++++++
drivers/dma/xor.c | 153 ++++++++
drivers/md/Kconfig | 3
drivers/md/Makefile | 6
drivers/md/raid5.c | 52 +--
drivers/md/xor.c | 154 --------
include/linux/async_tx.h | 173 +++++++++
include/linux/raid/xor.h | 5
11 files changed, 1263 insertions(+), 189 deletions(-)
diff --git a/drivers/Makefile b/drivers/Makefile
index 3a718f5..2e8de9e 100644
--- a/drivers/Makefile
+++ b/drivers/Makefile
@@ -62,6 +62,7 @@ obj-$(CONFIG_I2C) += i2c/
obj-$(CONFIG_W1) += w1/
obj-$(CONFIG_HWMON) += hwmon/
obj-$(CONFIG_PHONE) += telephony/
+obj-$(CONFIG_ASYNC_TX_DMA) += dma/
obj-$(CONFIG_MD) += md/
obj-$(CONFIG_BT) += bluetooth/
obj-$(CONFIG_ISDN) += isdn/
diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index 30d021d..292ddad 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -7,8 +7,8 @@ menu "DMA Engine support"
config DMA_ENGINE
bool "Support for DMA engines"
---help---
- DMA engines offload copy operations from the CPU to dedicated
- hardware, allowing the copies to happen asynchronously.
+ DMA engines offload bulk memory operations from the CPU to dedicated
+ hardware, allowing the operations to happen asynchronously.
comment "DMA Clients"
@@ -22,6 +22,16 @@ config NET_DMA
Since this is the main user of the DMA engine, it should be enabled;
say Y here.
+config ASYNC_TX_DMA
+ tristate "Asynchronous Bulk Memory Transfers/Transforms API"
+ ---help---
+ This enables the async_tx management layer for dma engines.
+ Subsystems coded to this API will use offload engines for bulk
+ memory operations where present. Software implementations are
+ called when a dma engine is not present or fails to allocate
+ memory to carry out the transaction.
+ Current subsystems ported to async_tx: MD_RAID4,5
+
comment "DMA Devices"
config INTEL_IOATDMA
@@ -30,5 +40,4 @@ config INTEL_IOATDMA
default m
---help---
Enable support for the Intel(R) I/OAT DMA engine.
-
endmenu
diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile
index bdcfdbd..6a99341 100644
--- a/drivers/dma/Makefile
+++ b/drivers/dma/Makefile
@@ -1,3 +1,4 @@
obj-$(CONFIG_DMA_ENGINE) += dmaengine.o
obj-$(CONFIG_NET_DMA) += iovlock.o
obj-$(CONFIG_INTEL_IOATDMA) += ioatdma.o
+obj-$(CONFIG_ASYNC_TX_DMA) += async_tx.o xor.o
diff --git a/drivers/dma/async_tx.c b/drivers/dma/async_tx.c
new file mode 100644
index 0000000..6bbbacc
--- /dev/null
+++ b/drivers/dma/async_tx.c
@@ -0,0 +1,889 @@
+/*
+ * Copyright(c) 2006 Intel Corporation. All rights reserved.
+ *
+ * Dan Williams <dan.j.williams@...el.com>
+ *
+ * with architecture considerations by:
+ * Neil Brown <neilb@...e.de>
+ * Jeff Garzik <jeff@...zik.org>
+ *
+ * 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.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc., 59
+ * Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * The full GNU General Public License is included in this distribution in the
+ * file called COPYING.
+ */
+#include <linux/kernel.h>
+#include <linux/interrupt.h>
+#include <linux/dma-mapping.h>
+#include <linux/raid/xor.h>
+#include <linux/async_tx.h>
+
+#ifdef CONFIG_DMA_ENGINE
+static int
+dma_channel_add_remove(struct dma_client *client,
+ struct dma_chan *chan, enum dma_state state);
+
+static struct dma_client async_tx_dma = {
+ .event_callback = dma_channel_add_remove,
+ /* .cap_mask == 0 defaults to all channels */
+};
+
+/**
+ * dma_cap_mask_all - enable iteration over all operation types
+ */
+dma_cap_mask_t dma_cap_mask_all;
+
+/**
+ * chan_ref_percpu - tracks channel allocations per core/opertion
+ */
+struct chan_ref_percpu {
+ struct dma_chan_ref *ref;
+};
+
+/* guarantee that calls that happen before init will not find a channel */
+static struct chan_ref_percpu __initdata null_chan_ref[NR_CPUS];
+
+static struct chan_ref_percpu *channel_table[] = {
+ DMA_TX_ARRAY_INIT(null_chan_ref)
+};
+
+/**
+ * async_tx_lock - protect modification of async_tx_master_list and serialize
+ * rebalance operations
+ */
+static spinlock_t async_tx_lock;
+
+struct list_head async_tx_master_list = LIST_HEAD_INIT(async_tx_master_list);
+EXPORT_SYMBOL_GPL(async_tx_master_list);
+
+static void
+free_dma_chan_ref(struct rcu_head *rcu)
+{
+ struct dma_chan_ref *ref;
+ ref = container_of(rcu, struct dma_chan_ref, rcu);
+ dma_chan_put(ref->chan);
+ kfree(ref);
+}
+
+static void
+init_dma_chan_ref(struct dma_chan_ref *ref, struct dma_chan *chan)
+{
+ INIT_LIST_HEAD(&ref->node);
+ INIT_RCU_HEAD(&ref->rcu);
+ ref->chan = chan;
+ atomic_set(&ref->count, 0);
+}
+
+/**
+ * get_chan_ref_by_cap - returns the nth channel of the given capability
+ * defaults to returning the channel with the desired capability and the
+ * lowest reference count if the index can not be satisfied
+ * @cap: capability to match
+ * @index: nth channel desired, passing -1 has the effect of forcing the
+ * default return value
+ */
+static struct dma_chan_ref *
+get_chan_ref_by_cap(enum dma_transaction_type cap, int index)
+{
+ struct dma_chan_ref *ret_ref = NULL, *min_ref = NULL, *ref;
+
+ list_for_each_entry_rcu(ref, &async_tx_master_list, node)
+ if (dma_has_cap(cap, ref->chan->device->cap_mask)) {
+ if (!min_ref)
+ min_ref = ref;
+ else if (atomic_read(&ref->count) <
+ atomic_read(&min_ref->count))
+ min_ref = ref;
+
+ if (index-- == 0) {
+ ret_ref = ref;
+ break;
+ }
+ }
+
+ if (!ret_ref)
+ ret_ref = min_ref;
+
+ if (ret_ref)
+ atomic_inc(&ret_ref->count);
+
+ return ret_ref;
+}
+
+/**
+ * async_tx_rebalance - redistribute the available channels, optimize
+ * for cpu isolation in the SMP case, and opertaion isolation in the
+ * uniprocessor case
+ */
+static void async_tx_rebalance(void)
+{
+ int cpu, cap, cpu_idx = 0;
+ unsigned long flags;
+
+ spin_lock_irqsave(&async_tx_lock, flags);
+
+ /* undo the last distribution */
+ for_each_dma_cap_mask(cap, dma_cap_mask_all)
+ for_each_possible_cpu(cpu) {
+ struct dma_chan_ref *ref =
+ per_cpu_ptr(channel_table[cap], cpu)->ref;
+ if (ref) {
+ atomic_set(&ref->count, 0);
+ per_cpu_ptr(channel_table[cap], cpu)->ref = NULL;
+ }
+ }
+
+ for_each_dma_cap_mask(cap, dma_cap_mask_all)
+ for_each_online_cpu(cpu) {
+ struct dma_chan_ref *new;
+ if (NR_CPUS > 1)
+ new = get_chan_ref_by_cap(cap, cpu_idx++);
+ else
+ new = get_chan_ref_by_cap(cap, -1);
+
+ per_cpu_ptr(channel_table[cap], cpu)->ref = new;
+ }
+
+ spin_unlock_irqrestore(&async_tx_lock, flags);
+}
+
+static int
+dma_channel_add_remove(struct dma_client *client,
+ struct dma_chan *chan, enum dma_state state)
+{
+ unsigned long found, ack = 0, flags;
+ struct dma_chan_ref *master_ref, *ref;
+
+ switch (state) {
+ case DMA_RESOURCE_AVAILABLE:
+ found = 0;
+ rcu_read_lock();
+ list_for_each_entry_rcu(ref, &async_tx_master_list, node)
+ if (ref->chan == chan) {
+ found = 1;
+ break;
+ }
+ rcu_read_unlock();
+
+ pr_debug("async_tx: dma resource available [%s]\n", found ? "old" : "new");
+
+ if (!found)
+ ack = 1;
+ else
+ break;
+
+ /* add the channel to the generic management list */
+ master_ref = kmalloc(sizeof(*master_ref), GFP_KERNEL);
+ if (master_ref) {
+ /* keep a reference until async_tx is unloaded */
+ dma_chan_get(chan);
+ init_dma_chan_ref(master_ref, chan);
+ spin_lock_irqsave(&async_tx_lock, flags);
+ list_add_tail_rcu(&master_ref->node,
+ &async_tx_master_list);
+ spin_unlock_irqrestore(&async_tx_lock,
+ flags);
+ } else {
+ printk(KERN_WARNING "async_tx: unable to create"
+ " new master entry in response to"
+ " a DMA_RESOURCE_ADDED event"
+ " (-ENOMEM)\n");
+ return 0;
+ }
+
+ async_tx_rebalance();
+ break;
+ case DMA_RESOURCE_REMOVED:
+ found = 0;
+ spin_lock_irqsave(&async_tx_lock, flags);
+ list_for_each_entry_rcu(ref, &async_tx_master_list, node)
+ if (ref->chan == chan) {
+ dma_chan_put(ref->chan); /* permit backing devices to go away */
+ list_del_rcu(&ref->node);
+ call_rcu(&ref->rcu, free_dma_chan_ref);
+ found = 1;
+ break;
+ }
+ spin_unlock_irqrestore(&async_tx_lock, flags);
+
+ pr_debug("async_tx: dma resource removed [%s]\n", found ? "ours" : "not ours");
+
+ if (found)
+ ack = 1;
+ else
+ break;
+
+ async_tx_rebalance();
+ break;
+ case DMA_RESOURCE_SUSPEND:
+ case DMA_RESOURCE_RESUME:
+ printk(KERN_WARNING "async_tx: does not support dma channel"
+ " suspend/resume\n");
+ break;
+ default:
+ BUG();
+ }
+
+ return ack;
+}
+
+static int __init
+async_tx_init(void)
+{
+ enum dma_transaction_type cap;
+
+ spin_lock_init(&async_tx_lock);
+ bitmap_fill(dma_cap_mask_all.bits, DMA_TX_TYPE_END);
+
+ /* an interrupt will never be an explicit operation type.
+ * clearing this bit prevents allocation to a slot in 'channel_table'
+ */
+ clear_bit(DMA_INTERRUPT, dma_cap_mask_all.bits);
+
+ for_each_dma_cap_mask(cap, dma_cap_mask_all) {
+ channel_table[cap] = alloc_percpu(struct chan_ref_percpu);
+ if (!channel_table[cap])
+ goto err;
+ }
+
+ dma_async_client_register(&async_tx_dma);
+ dma_async_client_chan_request(&async_tx_dma);
+
+ printk("async_tx: api initialized (async)\n");
+
+ return 0;
+err:
+ printk("async_tx: initialization failure\n");
+
+ while (--cap >= 0)
+ free_percpu(channel_table[cap]);
+
+ return 1;
+}
+
+static void __exit async_tx_exit(void)
+{
+ unsigned long flags;
+ struct dma_chan_ref *ref;
+ enum dma_transaction_type cap;
+
+ dma_async_client_unregister(&async_tx_dma);
+
+ for_each_dma_cap_mask(cap, dma_cap_mask_all)
+ if (channel_table[cap])
+ free_percpu(channel_table[cap]);
+
+ /* free all the channels on the master list */
+ spin_lock_irqsave(&async_tx_lock, flags);
+ list_for_each_entry_rcu(ref, &async_tx_master_list, node) {
+ dma_chan_put(ref->chan); /* permit backing devices to go away */
+ list_del_rcu(&ref->node);
+ call_rcu(&ref->rcu, free_dma_chan_ref);
+ }
+ spin_unlock_irqrestore(&async_tx_lock, flags);
+}
+
+/**
+ * async_tx_find_channel - find a channel to carry out the operation or let
+ * the transaction execute synchronously
+ * @depend_tx: transaction dependency
+ * @tx_type: transaction type
+ */
+static struct dma_chan *
+async_tx_find_channel(struct dma_async_tx_descriptor *depend_tx,
+ enum dma_transaction_type tx_type)
+{
+ /* see if we can keep the chain on one channel */
+ if (depend_tx &&
+ dma_has_cap(tx_type, depend_tx->chan->device->cap_mask))
+ return depend_tx->chan;
+ else {
+ struct dma_chan_ref *ref;
+ int cpu = get_cpu();
+ ref = per_cpu_ptr(channel_table[tx_type], cpu)->ref;
+ put_cpu();
+ return ref ? ref->chan : NULL;
+ }
+}
+#else
+static int __init async_tx_init(void)
+{
+ printk("async_tx: api initialized (sync-only)\n");
+ return 0;
+}
+
+static void __exit async_tx_exit(void)
+{
+ do { } while (0);
+}
+
+static struct dma_chan *
+async_tx_find_channel(struct dma_async_tx_descriptor *depend_tx,
+ enum dma_transaction_type tx_type)
+{
+ return NULL;
+}
+#endif
+
+static void
+async_tx_submit(struct dma_chan *chan, struct dma_async_tx_descriptor *tx,
+ enum async_tx_flags flags, struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback callback, void *callback_param)
+{
+ tx->callback = callback;
+ tx->callback_param = callback_param;
+
+ /* set this new tx to run after depend_tx if:
+ * 1/ a dependency exists (depend_tx is !NULL)
+ * 2/ the tx can not be submitted to the current channel
+ */
+ if (depend_tx && depend_tx->chan != chan) {
+ /* if ack is already set then we cannot be sure
+ * we are referring to the correct operation
+ */
+ BUG_ON(depend_tx->ack);
+
+ tx->parent = depend_tx;
+ spin_lock_bh(&depend_tx->lock);
+ list_add_tail(&tx->depend_node, &depend_tx->depend_list);
+ if (depend_tx->cookie == 0) {
+ struct dma_chan *dep_chan = depend_tx->chan;
+ struct dma_device *dep_dev = dep_chan->device;
+ dep_dev->device_dependency_added(dep_chan);
+ }
+ spin_unlock_bh(&depend_tx->lock);
+
+ /* schedule an interrupt to trigger the channel switch */
+ async_trigger_callback(ASYNC_TX_ACK, depend_tx, NULL, NULL);
+ } else {
+ tx->parent = NULL;
+ chan->device->device_tx_submit(tx);
+ }
+
+ if (flags & ASYNC_TX_ACK)
+ async_tx_ack(tx);
+
+ if (depend_tx && (flags & ASYNC_TX_DEP_ACK))
+ async_tx_ack(depend_tx);
+}
+
+/**
+ * sync_epilog - actions to take if an operation is run synchronously
+ * @flags: async_tx flags
+ * @depend_tx: transaction depends on depend_tx
+ * @callback: function to call when the transaction completes
+ * @callback_param: parameter to pass to the callback routine
+ */
+static void
+sync_epilog(unsigned long flags, struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback callback, void *callback_param)
+{
+ if (callback)
+ callback(callback_param);
+
+ if (depend_tx && (flags & ASYNC_TX_DEP_ACK))
+ async_tx_ack(depend_tx);
+}
+
+static void
+do_async_xor(struct dma_async_tx_descriptor *tx, struct dma_device *device,
+ struct dma_chan *chan, struct page *dest, struct page **src_list,
+ unsigned int offset, unsigned int src_cnt, size_t len,
+ enum async_tx_flags flags, struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback callback, void *callback_param)
+{
+ dma_addr_t dma_addr;
+ enum dma_data_direction dir;
+ int i;
+
+ pr_debug("%s: len: %zu\n", __FUNCTION__, len);
+
+ dir = (flags & ASYNC_TX_ASSUME_COHERENT) ?
+ DMA_NONE : DMA_FROM_DEVICE;
+
+ dma_addr = dma_map_page(device->dev, dest, offset, len, dir);
+ device->device_set_dest(dma_addr, tx, 0);
+
+ dir = (flags & ASYNC_TX_ASSUME_COHERENT) ?
+ DMA_NONE : DMA_TO_DEVICE;
+
+ for (i = 0; i < src_cnt; i++) {
+ dma_addr = dma_map_page(device->dev, src_list[i],
+ offset, len, dir);
+ device->device_set_src(dma_addr, tx, i);
+ }
+
+ async_tx_submit(chan, tx, flags, depend_tx, callback,
+ callback_param);
+}
+
+static void
+do_sync_xor(struct page *dest, struct page **src_list, unsigned int offset,
+ unsigned int src_cnt, size_t len, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback callback, void *callback_param)
+{
+ void *_dest;
+ int i;
+
+ pr_debug("%s: len: %zu\n", __FUNCTION__, len);
+
+ /* reuse the 'src_list' array to convert to buffer pointers */
+ for (i = 0; i < src_cnt; i++)
+ src_list[i] = (struct page *)
+ (page_address(src_list[i]) + offset);
+
+ /* set destination address */
+ _dest = page_address(dest) + offset;
+
+ if (flags & ASYNC_TX_XOR_ZERO_DST)
+ memset(_dest, 0, len);
+
+ xor_block(src_cnt, len, _dest,
+ (void **) src_list);
+
+ sync_epilog(flags, depend_tx, callback, callback_param);
+}
+
+/**
+ * async_xor - attempt to xor a set of blocks with a dma engine.
+ * xor_block always uses the dest as a source so the ASYNC_TX_XOR_ZERO_DST
+ * flag must be set to not include dest data in the calculation. The
+ * assumption with dma eninges is that they only use the destination
+ * buffer as a source when it is explicity specified in the source list.
+ * @dest: destination page
+ * @src_list: array of source pages (if the dest is also a source it must be
+ * at index zero). The contents of this array may be overwritten.
+ * @offset: offset in pages to start transaction
+ * @src_cnt: number of source pages
+ * @len: length in bytes
+ * @flags: ASYNC_TX_XOR_ZERO_DST, ASYNC_TX_XOR_DROP_DEST,
+ ASYNC_TX_ASSUME_COHERENT, ASYNC_TX_ACK, ASYNC_TX_DEP_ACK
+ * @depend_tx: xor depends on the result of this transaction.
+ * @callback: function to call when the xor completes
+ * @callback_param: parameter to pass to the callback routine
+ */
+struct dma_async_tx_descriptor *
+async_xor(struct page *dest, struct page **src_list, unsigned int offset,
+ int src_cnt, size_t len, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback callback, void *callback_param)
+{
+ struct dma_chan *chan = async_tx_find_channel(depend_tx, DMA_XOR);
+ struct dma_device *device = chan ? chan->device : NULL;
+ struct dma_async_tx_descriptor *tx = NULL;
+ dma_async_tx_callback _callback;
+ void *_callback_param;
+ unsigned long local_flags;
+ int xor_src_cnt;
+ int i = 0, src_off = 0, int_en;
+
+ BUG_ON(src_cnt <= 1);
+
+ while (src_cnt) {
+ local_flags = flags;
+ if (device) { /* run the xor asynchronously */
+ xor_src_cnt = min(src_cnt, device->max_xor);
+ /* if we are submitting additional xors
+ * only set the callback on the last transaction
+ */
+ if (src_cnt > xor_src_cnt) {
+ local_flags &= ~ASYNC_TX_ACK;
+ _callback = NULL;
+ _callback_param = NULL;
+ } else {
+ _callback = callback;
+ _callback_param = callback_param;
+ }
+
+ int_en = _callback ? 1 : 0;
+
+ tx = device->device_prep_dma_xor(
+ chan, xor_src_cnt, len, int_en);
+
+ if (tx) {
+ do_async_xor(tx, device, chan, dest,
+ &src_list[src_off], offset, xor_src_cnt, len,
+ local_flags, depend_tx, _callback,
+ _callback_param);
+ } else /* fall through */
+ goto xor_sync;
+ } else { /* run the xor synchronously */
+xor_sync:
+ /* in the sync case the dest is an implied source
+ * (assumes the dest is at the src_off index)
+ */
+ if (flags & ASYNC_TX_XOR_DROP_DST) {
+ src_cnt--;
+ src_off++;
+ }
+
+ /* process up to 'MAX_XOR_BLOCKS' sources */
+ xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS);
+
+ /* if we are submitting additional xors
+ * only set the callback on the last transaction
+ */
+ if (src_cnt > xor_src_cnt) {
+ local_flags &= ~ASYNC_TX_ACK;
+ _callback = NULL;
+ _callback_param = NULL;
+ } else {
+ _callback = callback;
+ _callback_param = callback_param;
+ }
+
+ /* wait for any prerequisite operations */
+ if (depend_tx) {
+ /* if ack is already set then we cannot be sure
+ * we are referring to the correct operation
+ */
+ BUG_ON(depend_tx->ack);
+ if (dma_wait_for_async_tx(depend_tx) == DMA_ERROR)
+ panic("%s: DMA_ERROR waiting for depend_tx\n",
+ __FUNCTION__);
+ }
+
+ do_sync_xor(dest, &src_list[src_off], offset,
+ xor_src_cnt, len, local_flags, depend_tx,
+ _callback, _callback_param);
+ }
+
+ /* the previous tx is hidden from the client,
+ * so ack it
+ */
+ if (i && depend_tx)
+ async_tx_ack(depend_tx);
+
+ depend_tx = tx;
+
+ if (src_cnt > xor_src_cnt) {
+ /* drop completed sources */
+ src_cnt -= xor_src_cnt;
+ src_off += xor_src_cnt;
+
+ /* unconditionally preserve the destination */
+ flags &= ~ASYNC_TX_XOR_ZERO_DST;
+
+ /* use the intermediate result a source, but remember
+ * it's dropped, because it's implied, in the sync case
+ */
+ src_list[--src_off] = dest;
+ src_cnt++;
+ flags |= ASYNC_TX_XOR_DROP_DST;
+ } else
+ src_cnt = 0;
+ i++;
+ }
+
+ return tx;
+}
+EXPORT_SYMBOL_GPL(async_xor);
+
+static int page_is_zero(struct page *p, size_t len)
+{
+ char *a = page_address(p);
+ return ((*(u32*)a) == 0 &&
+ memcmp(a, a+4, len-4)==0);
+}
+
+/**
+ * async_xor_zero_sum - attempt a xor parity check with a dma engine.
+ * @dest: destination page used if the xor is performed synchronously
+ * @src_list: array of source pages. The dest page must be listed as a source
+ * at index zero. The contents of this array may be overwritten.
+ * @offset: offset in pages to start transaction
+ * @src_cnt: number of source pages
+ * @len: length in bytes
+ * @result: 0 if sum == 0 else non-zero
+ * @flags: ASYNC_TX_ASSUME_COHERENT, ASYNC_TX_ACK, ASYNC_TX_DEP_ACK
+ * @depend_tx: xor depends on the result of this transaction.
+ * @callback: function to call when the xor completes
+ * @callback_param: parameter to pass to the callback routine
+ */
+struct dma_async_tx_descriptor *
+async_xor_zero_sum(struct page *dest, struct page **src_list,
+ unsigned int offset, int src_cnt, size_t len,
+ u32 *result, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback callback, void *callback_param)
+{
+ struct dma_chan *chan = async_tx_find_channel(depend_tx, DMA_ZERO_SUM);
+ struct dma_device *device = chan ? chan->device : NULL;
+ int int_en = callback ? 1 : 0;
+ struct dma_async_tx_descriptor *tx = device ?
+ device->device_prep_dma_zero_sum(chan, src_cnt, len, result,
+ int_en) : NULL;
+ int i;
+
+ BUG_ON(src_cnt <= 1);
+
+ if (tx) {
+ dma_addr_t dma_addr;
+ enum dma_data_direction dir;
+
+ pr_debug("%s: (async) len: %zu\n", __FUNCTION__, len);
+
+ dir = (flags & ASYNC_TX_ASSUME_COHERENT) ?
+ DMA_NONE : DMA_TO_DEVICE;
+
+ for (i = 0; i < src_cnt; i++) {
+ dma_addr = dma_map_page(device->dev, src_list[i],
+ offset, len, dir);
+ device->device_set_src(dma_addr, tx, i);
+ }
+
+ async_tx_submit(chan, tx, flags, depend_tx, callback,
+ callback_param);
+ } else {
+ unsigned long xor_flags = flags;
+
+ pr_debug("%s: (sync) len: %zu\n", __FUNCTION__, len);
+
+ xor_flags |= ASYNC_TX_XOR_DROP_DST;
+ xor_flags &= ~ASYNC_TX_ACK;
+
+ tx = async_xor(dest, src_list, offset, src_cnt, len, xor_flags,
+ depend_tx, NULL, NULL);
+
+ if (tx) {
+ if (dma_wait_for_async_tx(tx) == DMA_ERROR)
+ panic("%s: DMA_ERROR waiting for tx\n",
+ __FUNCTION__);
+ async_tx_ack(tx);
+ }
+
+ *result = page_is_zero(dest, len) ? 0 : 1;
+
+ tx = NULL;
+
+ sync_epilog(flags, depend_tx, callback, callback_param);
+ }
+
+ return tx;
+}
+EXPORT_SYMBOL_GPL(async_xor_zero_sum);
+
+/**
+ * async_memcpy - attempt to copy memory with a dma engine.
+ * @dest: destination page
+ * @src: src page
+ * @offset: offset in pages to start transaction
+ * @len: length in bytes
+ * @flags: ASYNC_TX_ASSUME_COHERENT, ASYNC_TX_ACK, ASYNC_TX_DEP_ACK,
+ * ASYNC_TX_KMAP_SRC, ASYNC_TX_KMAP_DST
+ * @depend_tx: memcpy depends on the result of this transaction
+ * @callback: function to call when the memcpy completes
+ * @callback_param: parameter to pass to the callback routine
+ */
+struct dma_async_tx_descriptor *
+async_memcpy(struct page *dest, struct page *src, unsigned int dest_offset,
+ unsigned int src_offset, size_t len, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback callback, void *callback_param)
+{
+ struct dma_chan *chan = async_tx_find_channel(depend_tx, DMA_MEMCPY);
+ struct dma_device *device = chan ? chan->device : NULL;
+ int int_en = callback ? 1 : 0;
+ struct dma_async_tx_descriptor *tx = device ?
+ device->device_prep_dma_memcpy(chan, len,
+ int_en) : NULL;
+
+ if (tx) { /* run the memcpy asynchronously */
+ dma_addr_t dma_addr;
+ enum dma_data_direction dir;
+
+ pr_debug("%s: (async) len: %zu\n", __FUNCTION__, len);
+
+ dir = (flags & ASYNC_TX_ASSUME_COHERENT) ?
+ DMA_NONE : DMA_FROM_DEVICE;
+
+ dma_addr = dma_map_page(device->dev, dest, dest_offset, len, dir);
+ device->device_set_dest(dma_addr, tx, 0);
+
+ dir = (flags & ASYNC_TX_ASSUME_COHERENT) ?
+ DMA_NONE : DMA_TO_DEVICE;
+
+ dma_addr = dma_map_page(device->dev, src, src_offset, len, dir);
+ device->device_set_src(dma_addr, tx, 0);
+
+ async_tx_submit(chan, tx, flags, depend_tx, callback,
+ callback_param);
+ } else { /* run the memcpy synchronously */
+ void *dest_buf, *src_buf;
+ pr_debug("%s: (sync) len: %zu\n", __FUNCTION__, len);
+
+ /* wait for any prerequisite operations */
+ if (depend_tx) {
+ /* if ack is already set then we cannot be sure
+ * we are referring to the correct operation
+ */
+ BUG_ON(depend_tx->ack);
+ if (dma_wait_for_async_tx(depend_tx) == DMA_ERROR)
+ panic("%s: DMA_ERROR waiting for depend_tx\n",
+ __FUNCTION__);
+ }
+
+ if (flags & ASYNC_TX_KMAP_DST)
+ dest_buf = kmap_atomic(dest, KM_USER0) + dest_offset;
+ else
+ dest_buf = page_address(dest) + dest_offset;
+
+ if (flags & ASYNC_TX_KMAP_SRC)
+ src_buf = kmap_atomic(src, KM_USER0) + src_offset;
+ else
+ src_buf = page_address(src) + src_offset;
+
+ memcpy(dest_buf, src_buf, len);
+
+ if (flags & ASYNC_TX_KMAP_DST)
+ kunmap_atomic(dest_buf, KM_USER0);
+
+ if (flags & ASYNC_TX_KMAP_SRC)
+ kunmap_atomic(src_buf, KM_USER0);
+
+ sync_epilog(flags, depend_tx, callback, callback_param);
+ }
+
+ return tx;
+}
+EXPORT_SYMBOL_GPL(async_memcpy);
+
+/**
+ * async_memset - attempt to fill memory with a dma engine.
+ * @dest: destination page
+ * @val: fill value
+ * @offset: offset in pages to start transaction
+ * @len: length in bytes
+ * @flags: ASYNC_TX_ASSUME_COHERENT, ASYNC_TX_ACK, ASYNC_TX_DEP_ACK
+ * @depend_tx: memset depends on the result of this transaction
+ * @callback: function to call when the memcpy completes
+ * @callback_param: parameter to pass to the callback routine
+ */
+struct dma_async_tx_descriptor *
+async_memset(struct page *dest, int val, unsigned int offset,
+ size_t len, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback callback, void *callback_param)
+{
+ struct dma_chan *chan = async_tx_find_channel(depend_tx, DMA_MEMSET);
+ struct dma_device *device = chan ? chan->device : NULL;
+ int int_en = callback ? 1 : 0;
+ struct dma_async_tx_descriptor *tx = device ?
+ device->device_prep_dma_memset(chan, val, len,
+ int_en) : NULL;
+
+ if (tx) { /* run the memset asynchronously */
+ dma_addr_t dma_addr;
+ enum dma_data_direction dir;
+
+ pr_debug("%s: (async) len: %zu\n", __FUNCTION__, len);
+ dir = (flags & ASYNC_TX_ASSUME_COHERENT) ?
+ DMA_NONE : DMA_FROM_DEVICE;
+
+ dma_addr = dma_map_page(device->dev, dest, offset, len, dir);
+ device->device_set_dest(dma_addr, tx, 0);
+
+ async_tx_submit(chan, tx, flags, depend_tx, callback,
+ callback_param);
+ } else { /* run the memset synchronously */
+ void *dest_buf;
+ pr_debug("%s: (sync) len: %zu\n", __FUNCTION__, len);
+
+ dest_buf = (void *) (((char *) page_address(dest)) + offset);
+
+ /* wait for any prerequisite operations */
+ if (depend_tx) {
+ /* if ack is already set then we cannot be sure
+ * we are referring to the correct operation
+ */
+ BUG_ON(depend_tx->ack);
+ if (dma_wait_for_async_tx(depend_tx) == DMA_ERROR)
+ panic("%s: DMA_ERROR waiting for depend_tx\n",
+ __FUNCTION__);
+ }
+
+ memset(dest_buf, val, len);
+
+ sync_epilog(flags, depend_tx, callback, callback_param);
+ }
+
+ return tx;
+}
+EXPORT_SYMBOL_GPL(async_memset);
+
+/**
+ * async_trigger_callback - schedules the callback function to be run after
+ * any dependent operations have been completed.
+ * @flags: ASYNC_TX_ACK, ASYNC_TX_DEP_ACK
+ * @depend_tx: 'callback' requires the completion of this transaction
+ * @callback: function to call after depend_tx completes
+ * @callback_param: parameter to pass to the callback routine
+ */
+struct dma_async_tx_descriptor *
+async_trigger_callback(enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback callback, void *callback_param)
+{
+ struct dma_chan *chan;
+ struct dma_device *device;
+ struct dma_async_tx_descriptor *tx;
+
+ if (depend_tx) {
+ chan = depend_tx->chan;
+ device = chan->device;
+
+ /* see if we can schedule an interrupt
+ * otherwise poll for completion
+ */
+ if (device && !dma_has_cap(DMA_INTERRUPT, device->cap_mask))
+ device = NULL;
+
+ tx = device ? device->device_prep_dma_interrupt(chan) : NULL;
+ } else
+ tx = NULL;
+
+ if (tx) {
+ pr_debug("%s: (async)\n", __FUNCTION__);
+
+ async_tx_submit(chan, tx, flags, depend_tx, callback,
+ callback_param);
+ } else {
+ pr_debug("%s: (sync)\n", __FUNCTION__);
+
+ /* wait for any prerequisite operations */
+ if (depend_tx) {
+ /* if ack is already set then we cannot be sure
+ * we are referring to the correct operation
+ */
+ BUG_ON(depend_tx->ack);
+ if (dma_wait_for_async_tx(depend_tx) == DMA_ERROR)
+ panic("%s: DMA_ERROR waiting for depend_tx\n",
+ __FUNCTION__);
+ }
+
+ sync_epilog(flags, depend_tx, callback, callback_param);
+ }
+
+ return tx;
+}
+EXPORT_SYMBOL_GPL(async_trigger_callback);
+
+module_init(async_tx_init);
+module_exit(async_tx_exit);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_DESCRIPTION("Asynchronous Bulk Memory Transactions API");
+MODULE_LICENSE("GPL");
diff --git a/drivers/dma/xor.c b/drivers/dma/xor.c
new file mode 100644
index 0000000..6eb3416
--- /dev/null
+++ b/drivers/dma/xor.c
@@ -0,0 +1,153 @@
+/*
+ * xor.c : Multiple Devices driver for Linux
+ *
+ * Copyright (C) 1996, 1997, 1998, 1999, 2000,
+ * Ingo Molnar, Matti Aarnio, Jakub Jelinek, Richard Henderson.
+ *
+ * Dispatch optimized RAID-5 checksumming functions.
+ *
+ * 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, or (at your option)
+ * any later version.
+ *
+ * You should have received a copy of the GNU General Public License
+ * (for example /usr/src/linux/COPYING); if not, write to the Free
+ * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#define BH_TRACE 0
+#include <linux/module.h>
+#include <linux/raid/md.h>
+#include <linux/raid/xor.h>
+#include <asm/xor.h>
+
+/* The xor routines to use. */
+static struct xor_block_template *active_template;
+
+void
+xor_block(unsigned int src_count, unsigned int bytes, void *dest, void **srcs)
+{
+ unsigned long *p1, *p2, *p3, *p4;
+
+ p1 = (unsigned long *) srcs[0];
+ if (src_count == 1) {
+ active_template->do_2(bytes, dest, p1);
+ return;
+ }
+
+ p2 = (unsigned long *) srcs[1];
+ if (src_count == 2) {
+ active_template->do_3(bytes, dest, p1, p2);
+ return;
+ }
+
+ p3 = (unsigned long *) srcs[2];
+ if (src_count == 3) {
+ active_template->do_4(bytes, dest, p1, p2, p3);
+ return;
+ }
+
+ p4 = (unsigned long *) srcs[3];
+ active_template->do_5(bytes, dest, p1, p2, p3, p4);
+}
+
+/* Set of all registered templates. */
+static struct xor_block_template *template_list;
+
+#define BENCH_SIZE (PAGE_SIZE)
+
+static void
+do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
+{
+ int speed;
+ unsigned long now;
+ int i, count, max;
+
+ tmpl->next = template_list;
+ template_list = tmpl;
+
+ /*
+ * Count the number of XORs done during a whole jiffy, and use
+ * this to calculate the speed of checksumming. We use a 2-page
+ * allocation to have guaranteed color L1-cache layout.
+ */
+ max = 0;
+ for (i = 0; i < 5; i++) {
+ now = jiffies;
+ count = 0;
+ while (jiffies == now) {
+ mb();
+ tmpl->do_2(BENCH_SIZE, b1, b2);
+ mb();
+ count++;
+ mb();
+ }
+ if (count > max)
+ max = count;
+ }
+
+ speed = max * (HZ * BENCH_SIZE / 1024);
+ tmpl->speed = speed;
+
+ printk(" %-10s: %5d.%03d MB/sec\n", tmpl->name,
+ speed / 1000, speed % 1000);
+}
+
+static int
+calibrate_xor_block(void)
+{
+ void *b1, *b2;
+ struct xor_block_template *f, *fastest;
+
+ b1 = (void *) __get_free_pages(GFP_KERNEL, 2);
+ if (! b1) {
+ printk("xor: Yikes! No memory available.\n");
+ return -ENOMEM;
+ }
+ b2 = b1 + 2*PAGE_SIZE + BENCH_SIZE;
+
+ /*
+ * If this arch/cpu has a short-circuited selection, don't loop through all
+ * the possible functions, just test the best one
+ */
+
+ fastest = NULL;
+
+#ifdef XOR_SELECT_TEMPLATE
+ fastest = XOR_SELECT_TEMPLATE(fastest);
+#endif
+
+#define xor_speed(templ) do_xor_speed((templ), b1, b2)
+
+ if (fastest) {
+ printk(KERN_INFO "xor: automatically using best checksumming function: %s\n",
+ fastest->name);
+ xor_speed(fastest);
+ } else {
+ printk(KERN_INFO "xor: measuring software checksumming speed\n");
+ XOR_TRY_TEMPLATES;
+ fastest = template_list;
+ for (f = fastest; f; f = f->next)
+ if (f->speed > fastest->speed)
+ fastest = f;
+ }
+
+ printk("xor: using function: %s (%d.%03d MB/sec)\n",
+ fastest->name, fastest->speed / 1000, fastest->speed % 1000);
+
+#undef xor_speed
+
+ free_pages((unsigned long)b1, 2);
+
+ active_template = fastest;
+ return 0;
+}
+
+static __exit void xor_exit(void) { }
+
+EXPORT_SYMBOL(xor_block);
+MODULE_LICENSE("GPL");
+
+module_init(calibrate_xor_block);
+module_exit(xor_exit);
diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
index 4540ade..79a361e 100644
--- a/drivers/md/Kconfig
+++ b/drivers/md/Kconfig
@@ -8,6 +8,7 @@ menu "Multi-device support (RAID and LVM)"
config MD
bool "Multiple devices driver support (RAID and LVM)"
+ select ASYNC_TX_DMA
help
Support multiple physical spindles through a single logical device.
Required for RAID and logical volume management.
@@ -108,7 +109,7 @@ config MD_RAID10
config MD_RAID456
tristate "RAID-4/RAID-5/RAID-6 mode"
- depends on BLK_DEV_MD
+ depends on BLK_DEV_MD && ASYNC_TX_DMA
---help---
A RAID-5 set of N drives with a capacity of C MB per drive provides
the capacity of C * (N - 1) MB, and protects against a failure
diff --git a/drivers/md/Makefile b/drivers/md/Makefile
index 34957a6..23c3049 100644
--- a/drivers/md/Makefile
+++ b/drivers/md/Makefile
@@ -17,15 +17,15 @@ raid456-objs := raid5.o raid6algos.o raid6recov.o raid6tables.o \
hostprogs-y := mktables
# Note: link order is important. All raid personalities
-# and xor.o must come before md.o, as they each initialise
-# themselves, and md.o may use the personalities when it
+# must come before md.o, as they each initialise
+# themselves, and md.o may use the personalities when it
# auto-initialised.
obj-$(CONFIG_MD_LINEAR) += linear.o
obj-$(CONFIG_MD_RAID0) += raid0.o
obj-$(CONFIG_MD_RAID1) += raid1.o
obj-$(CONFIG_MD_RAID10) += raid10.o
-obj-$(CONFIG_MD_RAID456) += raid456.o xor.o
+obj-$(CONFIG_MD_RAID456) += raid456.o
obj-$(CONFIG_MD_MULTIPATH) += multipath.o
obj-$(CONFIG_MD_FAULTY) += faulty.o
obj-$(CONFIG_BLK_DEV_MD) += md-mod.o
diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index 8d59914..ab8702d 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -916,25 +916,25 @@ static void copy_data(int frombio, struct bio *bio,
}
}
-#define check_xor() do { \
- if (count == MAX_XOR_BLOCKS) { \
- xor_block(count, STRIPE_SIZE, ptr); \
- count = 1; \
- } \
+#define check_xor() do { \
+ if (count == MAX_XOR_BLOCKS) { \
+ xor_block(count, STRIPE_SIZE, dest, ptr); \
+ count = 0; \
+ } \
} while(0)
static void compute_block(struct stripe_head *sh, int dd_idx)
{
int i, count, disks = sh->disks;
- void *ptr[MAX_XOR_BLOCKS], *p;
+ void *ptr[MAX_XOR_BLOCKS], *dest, *p;
PRINTK("compute_block, stripe %llu, idx %d\n",
(unsigned long long)sh->sector, dd_idx);
- ptr[0] = page_address(sh->dev[dd_idx].page);
- memset(ptr[0], 0, STRIPE_SIZE);
- count = 1;
+ dest = page_address(sh->dev[dd_idx].page);
+ memset(dest, 0, STRIPE_SIZE);
+ count = 0;
for (i = disks ; i--; ) {
if (i == dd_idx)
continue;
@@ -948,8 +948,8 @@ static void compute_block(struct stripe_head *sh, int dd_idx)
check_xor();
}
- if (count != 1)
- xor_block(count, STRIPE_SIZE, ptr);
+ if (count)
+ xor_block(count, STRIPE_SIZE, dest, ptr);
set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags);
}
@@ -957,14 +957,14 @@ static void compute_parity5(struct stripe_head *sh, int method)
{
raid5_conf_t *conf = sh->raid_conf;
int i, pd_idx = sh->pd_idx, disks = sh->disks, count;
- void *ptr[MAX_XOR_BLOCKS];
+ void *ptr[MAX_XOR_BLOCKS], *dest;
struct bio *chosen;
PRINTK("compute_parity5, stripe %llu, method %d\n",
(unsigned long long)sh->sector, method);
- count = 1;
- ptr[0] = page_address(sh->dev[pd_idx].page);
+ count = 0;
+ dest = page_address(sh->dev[pd_idx].page);
switch(method) {
case READ_MODIFY_WRITE:
BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags));
@@ -987,7 +987,7 @@ static void compute_parity5(struct stripe_head *sh, int method)
}
break;
case RECONSTRUCT_WRITE:
- memset(ptr[0], 0, STRIPE_SIZE);
+ memset(dest, 0, STRIPE_SIZE);
for (i= disks; i-- ;)
if (i!=pd_idx && sh->dev[i].towrite) {
chosen = sh->dev[i].towrite;
@@ -1003,9 +1003,9 @@ static void compute_parity5(struct stripe_head *sh, int method)
case CHECK_PARITY:
break;
}
- if (count>1) {
- xor_block(count, STRIPE_SIZE, ptr);
- count = 1;
+ if (count) {
+ xor_block(count, STRIPE_SIZE, dest, ptr);
+ count = 0;
}
for (i = disks; i--;)
@@ -1037,8 +1037,8 @@ static void compute_parity5(struct stripe_head *sh, int method)
check_xor();
}
}
- if (count != 1)
- xor_block(count, STRIPE_SIZE, ptr);
+ if (count)
+ xor_block(count, STRIPE_SIZE, dest, ptr);
if (method != CHECK_PARITY) {
set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
@@ -1132,7 +1132,7 @@ static void compute_parity6(struct stripe_head *sh, int method)
static void compute_block_1(struct stripe_head *sh, int dd_idx, int nozero)
{
int i, count, disks = sh->disks;
- void *ptr[MAX_XOR_BLOCKS], *p;
+ void *ptr[MAX_XOR_BLOCKS], *dest, *p;
int pd_idx = sh->pd_idx;
int qd_idx = raid6_next_disk(pd_idx, disks);
@@ -1143,9 +1143,9 @@ static void compute_block_1(struct stripe_head *sh, int dd_idx, int nozero)
/* We're actually computing the Q drive */
compute_parity6(sh, UPDATE_PARITY);
} else {
- ptr[0] = page_address(sh->dev[dd_idx].page);
- if (!nozero) memset(ptr[0], 0, STRIPE_SIZE);
- count = 1;
+ dest = page_address(sh->dev[dd_idx].page);
+ if (!nozero) memset(dest, 0, STRIPE_SIZE);
+ count = 0;
for (i = disks ; i--; ) {
if (i == dd_idx || i == qd_idx)
continue;
@@ -1159,8 +1159,8 @@ static void compute_block_1(struct stripe_head *sh, int dd_idx, int nozero)
check_xor();
}
- if (count != 1)
- xor_block(count, STRIPE_SIZE, ptr);
+ if (count)
+ xor_block(count, STRIPE_SIZE, dest, ptr);
if (!nozero) set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags);
else clear_bit(R5_UPTODATE, &sh->dev[dd_idx].flags);
}
diff --git a/drivers/md/xor.c b/drivers/md/xor.c
deleted file mode 100644
index 324897c..0000000
--- a/drivers/md/xor.c
+++ /dev/null
@@ -1,154 +0,0 @@
-/*
- * xor.c : Multiple Devices driver for Linux
- *
- * Copyright (C) 1996, 1997, 1998, 1999, 2000,
- * Ingo Molnar, Matti Aarnio, Jakub Jelinek, Richard Henderson.
- *
- * Dispatch optimized RAID-5 checksumming functions.
- *
- * 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, or (at your option)
- * any later version.
- *
- * You should have received a copy of the GNU General Public License
- * (for example /usr/src/linux/COPYING); if not, write to the Free
- * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
- */
-
-#define BH_TRACE 0
-#include <linux/module.h>
-#include <linux/raid/md.h>
-#include <linux/raid/xor.h>
-#include <asm/xor.h>
-
-/* The xor routines to use. */
-static struct xor_block_template *active_template;
-
-void
-xor_block(unsigned int count, unsigned int bytes, void **ptr)
-{
- unsigned long *p0, *p1, *p2, *p3, *p4;
-
- p0 = (unsigned long *) ptr[0];
- p1 = (unsigned long *) ptr[1];
- if (count == 2) {
- active_template->do_2(bytes, p0, p1);
- return;
- }
-
- p2 = (unsigned long *) ptr[2];
- if (count == 3) {
- active_template->do_3(bytes, p0, p1, p2);
- return;
- }
-
- p3 = (unsigned long *) ptr[3];
- if (count == 4) {
- active_template->do_4(bytes, p0, p1, p2, p3);
- return;
- }
-
- p4 = (unsigned long *) ptr[4];
- active_template->do_5(bytes, p0, p1, p2, p3, p4);
-}
-
-/* Set of all registered templates. */
-static struct xor_block_template *template_list;
-
-#define BENCH_SIZE (PAGE_SIZE)
-
-static void
-do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
-{
- int speed;
- unsigned long now;
- int i, count, max;
-
- tmpl->next = template_list;
- template_list = tmpl;
-
- /*
- * Count the number of XORs done during a whole jiffy, and use
- * this to calculate the speed of checksumming. We use a 2-page
- * allocation to have guaranteed color L1-cache layout.
- */
- max = 0;
- for (i = 0; i < 5; i++) {
- now = jiffies;
- count = 0;
- while (jiffies == now) {
- mb();
- tmpl->do_2(BENCH_SIZE, b1, b2);
- mb();
- count++;
- mb();
- }
- if (count > max)
- max = count;
- }
-
- speed = max * (HZ * BENCH_SIZE / 1024);
- tmpl->speed = speed;
-
- printk(" %-10s: %5d.%03d MB/sec\n", tmpl->name,
- speed / 1000, speed % 1000);
-}
-
-static int
-calibrate_xor_block(void)
-{
- void *b1, *b2;
- struct xor_block_template *f, *fastest;
-
- b1 = (void *) __get_free_pages(GFP_KERNEL, 2);
- if (! b1) {
- printk("raid5: Yikes! No memory available.\n");
- return -ENOMEM;
- }
- b2 = b1 + 2*PAGE_SIZE + BENCH_SIZE;
-
- /*
- * If this arch/cpu has a short-circuited selection, don't loop through all
- * the possible functions, just test the best one
- */
-
- fastest = NULL;
-
-#ifdef XOR_SELECT_TEMPLATE
- fastest = XOR_SELECT_TEMPLATE(fastest);
-#endif
-
-#define xor_speed(templ) do_xor_speed((templ), b1, b2)
-
- if (fastest) {
- printk(KERN_INFO "raid5: automatically using best checksumming function: %s\n",
- fastest->name);
- xor_speed(fastest);
- } else {
- printk(KERN_INFO "raid5: measuring checksumming speed\n");
- XOR_TRY_TEMPLATES;
- fastest = template_list;
- for (f = fastest; f; f = f->next)
- if (f->speed > fastest->speed)
- fastest = f;
- }
-
- printk("raid5: using function: %s (%d.%03d MB/sec)\n",
- fastest->name, fastest->speed / 1000, fastest->speed % 1000);
-
-#undef xor_speed
-
- free_pages((unsigned long)b1, 2);
-
- active_template = fastest;
- return 0;
-}
-
-static __exit void xor_exit(void) { }
-
-EXPORT_SYMBOL(xor_block);
-MODULE_LICENSE("GPL");
-
-module_init(calibrate_xor_block);
-module_exit(xor_exit);
diff --git a/include/linux/async_tx.h b/include/linux/async_tx.h
new file mode 100644
index 0000000..f91e294
--- /dev/null
+++ b/include/linux/async_tx.h
@@ -0,0 +1,173 @@
+/*
+ * Copyright(c) 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.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc., 59
+ * Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * The full GNU General Public License is included in this distribution in the
+ * file called COPYING.
+ */
+#ifndef _ASYNC_TX_H_
+#define _ASYNC_TX_H_
+#include <linux/dmaengine.h>
+#include <linux/spinlock.h>
+#include <linux/interrupt.h>
+
+/**
+ * dma_chan_ref - object used to manage dma channels received from the
+ * dmaengine core.
+ * @chan - the channel being tracked
+ * @node - node for the channel to be placed on async_tx_master_list
+ * @rcu - for list_del_rcu
+ * @count - number of times this channel is listed in the pool
+ * (for channels with multiple capabiities)
+ */
+struct dma_chan_ref {
+ struct dma_chan *chan;
+ struct list_head node;
+ struct rcu_head rcu;
+ atomic_t count;
+};
+
+/**
+ * async_tx_flags - modifiers for the async_* calls
+ * @ASYNC_TX_XOR_ZERO_DST: this flag must be used for xor operations where the
+ * the destination address is not a source. The asynchronous case handles this
+ * implicitly, the synchronous case needs to zero the destination block.
+ * @ASYNC_TX_XOR_DROP_DST: this flag must be used if the destination address is
+ * also one of the source addresses. In the synchronous case the destination
+ * address is an implied source, whereas the asynchronous case it must be listed
+ * as a source. The destination address must be the first address in the source
+ * array.
+ * @ASYNC_TX_ASSUME_COHERENT: skip cache maintenance operations
+ * @ASYNC_TX_ACK: immediately ack the descriptor, precludes setting up a
+ * dependency chain
+ * @ASYNC_TX_DEP_ACK: ack the dependency descriptor. Useful for chaining.
+ * @ASYNC_TX_KMAP_SRC: if the transaction is to be performed synchronously
+ * take an atomic mapping (KM_USER0) on the source page(s)
+ * @ASYNC_TX_KMAP_DST: if the transaction is to be performed synchronously
+ * take an atomic mapping (KM_USER0) on the dest page(s)
+ */
+enum async_tx_flags {
+ ASYNC_TX_XOR_ZERO_DST = (1 << 0),
+ ASYNC_TX_XOR_DROP_DST = (1 << 1),
+ ASYNC_TX_ASSUME_COHERENT = (1 << 2),
+ ASYNC_TX_ACK = (1 << 3),
+ ASYNC_TX_DEP_ACK = (1 << 4),
+ ASYNC_TX_KMAP_SRC = (1 << 5),
+ ASYNC_TX_KMAP_DST = (1 << 6),
+};
+
+#ifdef CONFIG_DMA_ENGINE
+static inline enum dma_status
+dma_wait_for_async_tx(struct dma_async_tx_descriptor *tx)
+{
+ enum dma_status status;
+ struct dma_async_tx_descriptor *iter;
+
+ if (!tx)
+ return DMA_SUCCESS;
+
+ /* poll through the dependency chain, return when tx is complete */
+ do {
+ iter = tx;
+ while (iter->cookie == -EBUSY)
+ iter = iter->parent;
+
+ status = dma_sync_wait(iter->chan, iter->cookie);
+ } while (status == DMA_IN_PROGRESS || (iter != tx));
+
+ return status;
+}
+
+extern struct list_head async_tx_master_list;
+static inline void async_tx_issue_pending_all(void)
+{
+ struct dma_chan_ref *ref;
+
+ rcu_read_lock();
+ list_for_each_entry_rcu(ref, &async_tx_master_list, node)
+ ref->chan->device->device_issue_pending(ref->chan);
+ rcu_read_unlock();
+}
+
+static inline void
+async_tx_run_dependencies(struct dma_async_tx_descriptor *tx,
+ struct dma_chan *host_chan)
+{
+ struct dma_async_tx_descriptor *dep_tx, *_dep_tx;
+ struct dma_device *dev;
+ struct dma_chan *chan;
+
+ list_for_each_entry_safe(dep_tx, _dep_tx, &tx->depend_list,
+ depend_node) {
+ chan = dep_tx->chan;
+ dev = chan->device;
+ /* we can't depend on ourselves */
+ BUG_ON(chan == host_chan);
+ list_del(&dep_tx->depend_node);
+ dev->device_tx_submit(dep_tx);
+
+ /* we need to poke the engine as client code does not
+ * know about dependency submission events
+ */
+ dev->device_issue_pending(chan);
+ }
+}
+#else
+static inline void
+async_tx_run_dependencies(struct dma_async_tx_descriptor *tx,
+ struct dma_chan *host_chan)
+{
+ do { } while (0);
+}
+
+static inline enum dma_status
+dma_wait_for_async_tx(struct dma_async_tx_descriptor *tx)
+{
+ return DMA_SUCCESS;
+}
+
+static inline void async_tx_issue_pending_all(void)
+{
+ do { } while (0);
+}
+#endif
+
+struct dma_async_tx_descriptor *
+async_xor(struct page *dest, struct page **src_list, unsigned int offset,
+ int src_cnt, size_t len, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback callback, void *callback_param);
+struct dma_async_tx_descriptor *
+async_xor_zero_sum(struct page *dest, struct page **src_list,
+ unsigned int offset, int src_cnt, size_t len,
+ u32 *result, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback callback, void *callback_param);
+struct dma_async_tx_descriptor *
+async_memcpy(struct page *dest, struct page *src, unsigned int dest_offset,
+ unsigned int src_offset, size_t len, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback callback, void *callback_param);
+struct dma_async_tx_descriptor *
+async_memset(struct page *dest, int val, unsigned int offset,
+ size_t len, enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback callback, void *callback_param);
+struct dma_async_tx_descriptor *
+async_trigger_callback(enum async_tx_flags flags,
+ struct dma_async_tx_descriptor *depend_tx,
+ dma_async_tx_callback callback, void *callback_param);
+#endif
diff --git a/include/linux/raid/xor.h b/include/linux/raid/xor.h
index f0d67cb..d151f16 100644
--- a/include/linux/raid/xor.h
+++ b/include/linux/raid/xor.h
@@ -3,9 +3,10 @@
#include <linux/raid/md.h>
-#define MAX_XOR_BLOCKS 5
+#define MAX_XOR_BLOCKS 4
-extern void xor_block(unsigned int count, unsigned int bytes, void **ptr);
+extern void xor_block(unsigned int count, unsigned int bytes,
+ void *dest, void **srcs);
struct xor_block_template {
struct xor_block_template *next;
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists