* [PATCH 0/8] Intel I/O Acceleration Technology (I/OAT)
From: Chris Leech @ 2006-03-03 21:40 UTC (permalink / raw)
To: linux-kernel, netdev
This patch series is the first full release of the Intel(R) I/O
Acceleration Technology (I/OAT) for Linux. It includes an in kernel API
for offloading memory copies to hardware, a driver for the I/OAT DMA memcpy
engine, and changes to the TCP stack to offload copies of received
networking data to application space.
These changes apply to DaveM's net-2.6.17 tree as of commit
2bd84a93d8bb7192ad8c23ef41008502be1cb603 ([IRDA]: TOIM3232 dongle support)
They are available to pull from
git://198.78.49.142/~cleech/linux-2.6 ioat-2.6.17
There are 8 patches in the series:
1) The memcpy offload APIs and class code
2) The Intel I/OAT DMA driver (ioatdma)
3) Core networking code to setup networking as a DMA memcpy client
4) Utility functions for sk_buff to iovec offloaded copy
5) Structure changes needed for TCP receive offload
6) Rename cleanup_rbuf to tcp_cleanup_rbuf
7) Add a sysctl to tune the minimum offloaded I/O size for TCP
8) The main TCP receive offload changes
--
Chris Leech <christopher.leech@intel.com>
I/O Acceleration Technology Software Development
LAN Access Division / Digital Enterprise Group
^ permalink raw reply
* [PATCH 1/8] [I/OAT] DMA memcpy subsystem
From: Chris Leech @ 2006-03-03 21:42 UTC (permalink / raw)
To: linux-kernel, netdev
In-Reply-To: <20060303214036.11908.10499.stgit@gitlost.site>
Provides an API for offloading memory copies to DMA devices
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
drivers/Kconfig | 2
drivers/Makefile | 1
drivers/dma/Kconfig | 13 ++
drivers/dma/Makefile | 1
drivers/dma/dmaengine.c | 361 +++++++++++++++++++++++++++++++++++++++++++++
include/linux/dmaengine.h | 322 ++++++++++++++++++++++++++++++++++++++++
6 files changed, 700 insertions(+), 0 deletions(-)
diff --git a/drivers/Kconfig b/drivers/Kconfig
index bddf431..ce7ffa7 100644
--- a/drivers/Kconfig
+++ b/drivers/Kconfig
@@ -70,4 +70,6 @@ source "drivers/sn/Kconfig"
source "drivers/edac/Kconfig"
+source "drivers/dma/Kconfig"
+
endmenu
diff --git a/drivers/Makefile b/drivers/Makefile
index 5c69b86..516ba5e 100644
--- a/drivers/Makefile
+++ b/drivers/Makefile
@@ -73,3 +73,4 @@ obj-$(CONFIG_SGI_SN) += sn/
obj-y += firmware/
obj-$(CONFIG_CRYPTO) += crypto/
obj-$(CONFIG_SUPERH) += sh/
+obj-$(CONFIG_DMA_ENGINE) += dma/
diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
new file mode 100644
index 0000000..f9ac4bc
--- /dev/null
+++ b/drivers/dma/Kconfig
@@ -0,0 +1,13 @@
+#
+# DMA engine configuration
+#
+
+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.
+
+endmenu
diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile
new file mode 100644
index 0000000..10b7391
--- /dev/null
+++ b/drivers/dma/Makefile
@@ -0,0 +1 @@
+obj-y += dmaengine.o
diff --git a/drivers/dma/dmaengine.c b/drivers/dma/dmaengine.c
new file mode 100644
index 0000000..77cfcb3
--- /dev/null
+++ b/drivers/dma/dmaengine.c
@@ -0,0 +1,361 @@
+/*****************************************************************************
+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.
+
+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 LICENSE.
+*****************************************************************************/
+#include <linux/init.h>
+#include <linux/module.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>
+
+static spinlock_t dma_list_lock;
+static LIST_HEAD(dma_device_list);
+static LIST_HEAD(dma_client_list);
+
+/* --- sysfs implementation --- */
+
+static ssize_t show_memcpy_count(struct class_device *cd, char *buf)
+{
+ struct dma_chan *chan = container_of(cd, struct dma_chan, class_dev);
+ unsigned long count = 0;
+ int i;
+
+ for_each_cpu(i)
+ count += per_cpu_ptr(chan->local, i)->memcpy_count;
+
+ sprintf(buf, "%lu\n", count);
+ return strlen(buf) + 1;
+}
+
+static ssize_t show_bytes_transferred(struct class_device *cd, char *buf)
+{
+ struct dma_chan *chan = container_of(cd, struct dma_chan, class_dev);
+ unsigned long count = 0;
+ int i;
+
+ for_each_cpu(i)
+ count += per_cpu_ptr(chan->local, i)->bytes_transferred;
+
+ sprintf(buf, "%lu\n", count);
+ return strlen(buf) + 1;
+}
+
+static ssize_t show_in_use(struct class_device *cd, char *buf)
+{
+ struct dma_chan *chan = container_of(cd, struct dma_chan, class_dev);
+
+ sprintf(buf, "%d\n", (chan->client ? 1 : 0));
+ return strlen(buf) + 1;
+}
+
+static struct class_device_attribute dma_class_attrs[] = {
+ __ATTR(memcpy_count, S_IRUGO, show_memcpy_count, NULL),
+ __ATTR(bytes_transferred, S_IRUGO, show_bytes_transferred, NULL),
+ __ATTR(in_use, S_IRUGO, show_in_use, NULL),
+ __ATTR_NULL
+};
+
+static void dma_async_device_cleanup(struct kref *kref);
+
+static void dma_class_dev_release(struct class_device *cd)
+{
+ struct dma_chan *chan = container_of(cd, struct dma_chan, class_dev);
+ kref_put(&chan->device->refcount, dma_async_device_cleanup);
+}
+
+static struct class dma_devclass = {
+ .name = "dma",
+ .class_dev_attrs = dma_class_attrs,
+ .release = dma_class_dev_release,
+};
+
+/* --- client and device registration --- */
+
+/**
+ * dma_client_chan_alloc - try to allocate a channel to a client
+ * @client: &dma_client
+ *
+ * Called with dma_list_lock held.
+ */
+static struct dma_chan * dma_client_chan_alloc(struct dma_client *client)
+{
+ struct dma_device *device;
+ struct dma_chan *chan;
+ unsigned long flags;
+
+ /* Find a channel, any DMA engine will do */
+ list_for_each_entry(device, &dma_device_list, global_node) {
+ list_for_each_entry(chan, &device->channels, device_node) {
+ if (chan->client)
+ continue;
+
+ if (chan->device->device_alloc_chan_resources(chan) >= 0) {
+ kref_get(&device->refcount);
+ kref_init(&chan->refcount);
+ chan->slow_ref = 0;
+ INIT_RCU_HEAD(&chan->rcu);
+ chan->client = client;
+ spin_lock_irqsave(&client->lock, flags);
+ list_add_tail_rcu(&chan->client_node, &client->channels);
+ spin_unlock_irqrestore(&client->lock, flags);
+ return chan;
+ }
+ }
+ }
+
+ return NULL;
+}
+
+/**
+ * dma_client_chan_free - release a DMA channel
+ * @chan: &dma_chan
+ */
+void dma_async_device_cleanup(struct kref *kref);
+void dma_chan_cleanup(struct kref *kref)
+{
+ struct dma_chan *chan = container_of(kref, struct dma_chan, refcount);
+ chan->device->device_free_chan_resources(chan);
+ chan->client = NULL;
+ kref_put(&chan->device->refcount, dma_async_device_cleanup);
+}
+
+static void dma_chan_free_rcu(struct rcu_head *rcu) {
+ struct dma_chan *chan = container_of(rcu, struct dma_chan, rcu);
+ int bias = 0x7FFFFFFF;
+ int i;
+ for_each_cpu(i)
+ bias -= local_read(&per_cpu_ptr(chan->local, i)->refcount);
+ atomic_sub(bias, &chan->refcount.refcount);
+ kref_put(&chan->refcount, dma_chan_cleanup);
+}
+
+static void dma_client_chan_free(struct dma_chan *chan)
+{
+ atomic_add(0x7FFFFFFF, &chan->refcount.refcount);
+ chan->slow_ref = 1;
+ call_rcu(&chan->rcu, dma_chan_free_rcu);
+}
+
+/**
+ * dma_chans_rebalance - reallocate channels to clients
+ *
+ * When the number of DMA channel in the system changes,
+ * channels need to be rebalanced among clients
+ */
+static void dma_chans_rebalance(void)
+{
+ struct dma_client *client;
+ struct dma_chan *chan;
+ unsigned long flags;
+
+ spin_lock(&dma_list_lock);
+ list_for_each_entry(client, &dma_client_list, global_node) {
+
+ while (client->chans_desired > client->chan_count) {
+ chan = dma_client_chan_alloc(client);
+ if (!chan)
+ break;
+
+ client->chan_count++;
+ client->event_callback(client, chan, DMA_RESOURCE_ADDED);
+ }
+
+ while (client->chans_desired < client->chan_count) {
+ spin_lock_irqsave(&client->lock, flags);
+ chan = list_entry(client->channels.next, struct dma_chan, client_node);
+ list_del_rcu(&chan->client_node);
+ spin_unlock_irqrestore(&client->lock, flags);
+ client->chan_count--;
+ client->event_callback(client, chan, DMA_RESOURCE_REMOVED);
+ dma_client_chan_free(chan);
+ }
+ }
+ spin_unlock(&dma_list_lock);
+}
+
+/**
+ * dma_async_client_register - allocate and register a &dma_client
+ * @event_callback: callback for notification of channel addition/removal
+ */
+struct dma_client * dma_async_client_register(dma_event_callback event_callback)
+{
+ struct dma_client *client;
+
+ client = kzalloc(sizeof(*client), GFP_KERNEL);
+ if (!client)
+ return NULL;
+
+ INIT_LIST_HEAD(&client->channels);
+ spin_lock_init(&client->lock);
+
+ client->chans_desired = 0;
+ client->chan_count = 0;
+ client->event_callback = event_callback;
+
+ spin_lock(&dma_list_lock);
+ list_add_tail(&client->global_node, &dma_client_list);
+ spin_unlock(&dma_list_lock);
+
+ return client;
+}
+
+/**
+ * dma_async_client_unregister - unregister a client and free the &dma_client
+ * @client:
+ *
+ * Force frees any allocated DMA channels, frees the &dma_client memory
+ */
+void dma_async_client_unregister(struct dma_client *client)
+{
+ struct dma_chan *chan;
+
+ if (!client)
+ return;
+
+ rcu_read_lock();
+ list_for_each_entry_rcu(chan, &client->channels, client_node) {
+ dma_client_chan_free(chan);
+ }
+ rcu_read_unlock();
+
+ spin_lock(&dma_list_lock);
+ list_del(&client->global_node);
+ spin_unlock(&dma_list_lock);
+
+ kfree(client);
+ dma_chans_rebalance();
+}
+
+/**
+ * dma_async_client_chan_request - request DMA channels
+ * @client: &dma_client
+ * @number: count of DMA channels requested
+ *
+ * Clients call dma_async_client_chan_request() to specify how many
+ * DMA channels they need, 0 to free all currently allocated.
+ * The resulting allocations/frees are indicated to the client via the
+ * event callback.
+ */
+void dma_async_client_chan_request(struct dma_client *client,
+ unsigned int number)
+{
+ client->chans_desired = number;
+ dma_chans_rebalance();
+}
+
+/**
+ * dma_async_device_register -
+ * @device: &dma_device
+ */
+int dma_async_device_register(struct dma_device *device)
+{
+ static int id;
+ int chancnt = 0;
+ struct dma_chan* chan;
+
+ if (!device)
+ return -ENODEV;
+
+ init_completion(&device->done);
+ kref_init(&device->refcount);
+ device->dev_id = id++;
+
+ /* represent channels in sysfs. Probably want devs too */
+ list_for_each_entry(chan, &device->channels, device_node) {
+ chan->local = alloc_percpu(typeof(*chan->local));
+ if (chan->local == NULL)
+ continue;
+
+ chan->chan_id = chancnt++;
+ chan->class_dev.class = &dma_devclass;
+ chan->class_dev.dev = NULL;
+ snprintf(chan->class_dev.class_id, BUS_ID_SIZE, "dma%dchan%d",
+ device->dev_id, chan->chan_id);
+
+ kref_get(&device->refcount);
+ class_device_register(&chan->class_dev);
+ }
+
+ spin_lock(&dma_list_lock);
+ list_add_tail(&device->global_node, &dma_device_list);
+ spin_unlock(&dma_list_lock);
+
+ dma_chans_rebalance();
+
+ return 0;
+}
+
+/**
+ * dma_async_device_unregister -
+ * @device: &dma_device
+ */
+static void dma_async_device_cleanup(struct kref *kref) {
+ struct dma_device *device = container_of(kref, struct dma_device, refcount);
+ complete(&device->done);
+}
+
+void dma_async_device_unregister(struct dma_device* device)
+{
+ struct dma_chan *chan;
+ unsigned long flags;
+
+ spin_lock(&dma_list_lock);
+ list_del(&device->global_node);
+ spin_unlock(&dma_list_lock);
+
+ list_for_each_entry(chan, &device->channels, device_node) {
+ if (chan->client) {
+ spin_lock_irqsave(&chan->client->lock, flags);
+ list_del(&chan->client_node);
+ chan->client->chan_count--;
+ spin_unlock_irqrestore(&chan->client->lock, flags);
+ chan->client->event_callback(chan->client, chan, DMA_RESOURCE_REMOVED);
+ dma_client_chan_free(chan);
+ }
+ class_device_unregister(&chan->class_dev);
+ }
+
+ dma_chans_rebalance();
+
+ kref_put(&device->refcount, dma_async_device_cleanup);
+ wait_for_completion(&device->done);
+}
+
+static int __init dma_bus_init(void)
+{
+ spin_lock_init(&dma_list_lock);
+
+ return class_register(&dma_devclass);
+}
+
+subsys_initcall(dma_bus_init);
+
+EXPORT_SYMBOL(dma_async_client_register);
+EXPORT_SYMBOL(dma_async_client_unregister);
+EXPORT_SYMBOL(dma_async_client_chan_request);
+EXPORT_SYMBOL(dma_async_memcpy_buf_to_buf);
+EXPORT_SYMBOL(dma_async_memcpy_buf_to_pg);
+EXPORT_SYMBOL(dma_async_memcpy_pg_to_pg);
+EXPORT_SYMBOL(dma_async_memcpy_complete);
+EXPORT_SYMBOL(dma_async_memcpy_issue_pending);
+EXPORT_SYMBOL(dma_async_device_register);
+EXPORT_SYMBOL(dma_async_device_unregister);
diff --git a/include/linux/dmaengine.h b/include/linux/dmaengine.h
new file mode 100644
index 0000000..f8a77ab
--- /dev/null
+++ b/include/linux/dmaengine.h
@@ -0,0 +1,322 @@
+/*****************************************************************************
+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.
+
+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 LICENSE.
+*****************************************************************************/
+#ifndef DMAENGINE_H
+#define DMAENGINE_H
+
+#include <linux/device.h>
+#include <linux/uio.h>
+#include <linux/kref.h>
+#include <linux/completion.h>
+#include <linux/rcupdate.h>
+
+/**
+ * enum dma_event - resource PNP/power managment events
+ * @DMA_RESOURCE_SUSPEND: DMA device going into low power state
+ * @DMA_RESOURCE_RESUME: DMA device returning to full power
+ * @DMA_RESOURCE_ADDED: DMA device added to the system
+ * @DMA_RESOURCE_REMOVED: DMA device removed from the system
+ */
+enum dma_event {
+ DMA_RESOURCE_SUSPEND,
+ DMA_RESOURCE_RESUME,
+ DMA_RESOURCE_ADDED,
+ DMA_RESOURCE_REMOVED,
+};
+
+/**
+ * typedef dma_cookie_t
+ *
+ * if dma_cookie_t is >0 it's a DMA request cookie, <0 it's an error code
+ */
+typedef s32 dma_cookie_t;
+
+#define dma_submit_error(cookie) ((cookie) < 0 ? 1 : 0)
+
+/**
+ * enum dma_status - DMA transaction status
+ * @DMA_SUCCESS: transaction completed successfully
+ * @DMA_IN_PROGRESS: transaction not yet processed
+ * @DMA_ERROR: transaction failed
+ */
+enum dma_status {
+ DMA_SUCCESS,
+ DMA_IN_PROGRESS,
+ DMA_ERROR,
+};
+
+struct dma_chan_percpu
+{
+ local_t refcount;
+ /* stats */
+ unsigned long memcpy_count;
+ unsigned long bytes_transferred;
+};
+
+/**
+ * struct dma_chan - devices supply DMA channels, clients use them
+ * @client: ptr to the client user of this chan, will be NULL when unused
+ * @device: ptr to the dma device who supplies this channel, always !NULL
+ * @cookie: last cookie value returned to client
+ * @chan_id:
+ * @class_dev:
+ * @client_node: used to add this to the client chan list
+ * @device_node: used to add this to the device chan list
+ */
+struct dma_chan
+{
+ struct dma_client *client;
+ struct dma_device *device;
+ dma_cookie_t cookie;
+
+ /* sysfs */
+ int chan_id;
+ struct class_device class_dev;
+
+ struct kref refcount;
+ int slow_ref;
+ struct rcu_head rcu;
+
+ struct list_head client_node;
+ struct list_head device_node;
+ struct dma_chan_percpu *local;
+};
+
+void dma_chan_cleanup(struct kref *kref);
+
+static inline void dma_chan_get(struct dma_chan *chan)
+{
+ if (unlikely(chan->slow_ref))
+ kref_get(&chan->refcount);
+ else {
+ local_inc(&(per_cpu_ptr(chan->local, get_cpu())->refcount));
+ put_cpu();
+ }
+}
+
+static inline void dma_chan_put(struct dma_chan *chan)
+{
+ if (unlikely(chan->slow_ref))
+ kref_put(&chan->refcount, dma_chan_cleanup);
+ else {
+ local_dec(&(per_cpu_ptr(chan->local, get_cpu())->refcount));
+ put_cpu();
+ }
+}
+
+/*
+ * typedef dma_event_callback - function pointer to a DMA event callback
+ */
+typedef void (*dma_event_callback) (struct dma_client *client,
+ struct dma_chan *chan, enum dma_event event);
+
+/**
+ * struct dma_client - info on the entity making use of DMA services
+ * @event_callback: func ptr to call when something happens
+ * @chan_count: number of chans allocated
+ * @chans_desired: number of chans requested. Can be +/- chan_count
+ * @lock: protects access to the channels list
+ * @channels: the list of DMA channels allocated
+ * @global_node: list_head for global dma_client_list
+ */
+struct dma_client {
+ dma_event_callback event_callback;
+ unsigned int chan_count;
+ unsigned int chans_desired;
+
+ spinlock_t lock;
+ struct list_head channels;
+ struct list_head global_node;
+};
+
+/**
+ * struct dma_device - info on the entity supplying DMA services
+ * @chancnt: how many DMA channels are supported
+ * @channels: the list of struct dma_chan
+ * @global_node: list_head for global dma_device_list
+ * @dev_id:
+ * Other func ptrs: used to make use of this device's capabilities
+ */
+struct dma_device {
+
+ unsigned int chancnt;
+ struct list_head channels;
+ struct list_head global_node;
+
+ struct kref refcount;
+ struct completion done;
+
+ int dev_id;
+
+ int (*device_alloc_chan_resources)(struct dma_chan *chan);
+ void (*device_free_chan_resources)(struct dma_chan *chan);
+ dma_cookie_t (*device_memcpy_buf_to_buf)(struct dma_chan *chan,
+ void *dest, void *src, size_t len);
+ dma_cookie_t (*device_memcpy_buf_to_pg)(struct dma_chan *chan,
+ struct page *page, unsigned int offset, void *kdata,
+ size_t len);
+ dma_cookie_t (*device_memcpy_pg_to_pg)(struct dma_chan *chan,
+ struct page *dest_pg, unsigned int dest_off,
+ struct page *src_pg, unsigned int src_off, size_t len);
+ enum dma_status (*device_memcpy_complete)(struct dma_chan *chan,
+ dma_cookie_t cookie, dma_cookie_t *last,
+ dma_cookie_t *used);
+ void (*device_memcpy_issue_pending)(struct dma_chan *chan);
+};
+
+/* --- public DMA engine API --- */
+
+struct dma_client *dma_async_client_register(dma_event_callback event_callback);
+void dma_async_client_unregister(struct dma_client *client);
+void dma_async_client_chan_request(struct dma_client *client,
+ unsigned int number);
+
+/**
+ * dma_async_memcpy_buf_to_buf - offloaded copy between virtual addresses
+ * @chan: DMA channel to offload copy to
+ * @dest: destination address (virtual)
+ * @src: source address (virtual)
+ * @len: length
+ *
+ * Both @dest and @src must be mappable to a bus address according to the
+ * DMA mapping API rules for streaming mappings.
+ * Both @dest and @src must stay memory resident (kernel memory or locked
+ * user space pages)
+ */
+static inline dma_cookie_t dma_async_memcpy_buf_to_buf(struct dma_chan *chan,
+ void *dest, void *src, size_t len)
+{
+ int cpu = get_cpu();
+ per_cpu_ptr(chan->local, cpu)->bytes_transferred += len;
+ per_cpu_ptr(chan->local, cpu)->memcpy_count++;
+ put_cpu();
+
+ return chan->device->device_memcpy_buf_to_buf(chan, dest, src, len);
+}
+
+/**
+ * dma_async_memcpy_buf_to_pg - offloaded copy
+ * @chan: DMA channel to offload copy to
+ * @page: destination page
+ * @offset: offset in page to copy to
+ * @kdata: source address (virtual)
+ * @len: length
+ *
+ * Both @page/@offset and @kdata must be mappable to a bus address according
+ * to the DMA mapping API rules for streaming mappings.
+ * Both @page/@offset and @kdata must stay memory resident (kernel memory or
+ * locked user space pages)
+ */
+static inline dma_cookie_t dma_async_memcpy_buf_to_pg(struct dma_chan *chan,
+ struct page *page, unsigned int offset, void *kdata, size_t len)
+{
+ int cpu = get_cpu();
+ per_cpu_ptr(chan->local, cpu)->bytes_transferred += len;
+ per_cpu_ptr(chan->local, cpu)->memcpy_count++;
+ put_cpu();
+
+ return chan->device->device_memcpy_buf_to_pg(chan, page, offset,
+ kdata, len);
+}
+
+/**
+ * dma_async_memcpy_buf_to_pg - offloaded copy
+ * @chan: DMA channel to offload copy to
+ * @dest_page: destination page
+ * @dest_off: offset in page to copy to
+ * @src_page: source page
+ * @src_off: offset in page to copy from
+ * @len: length
+ *
+ * Both @dest_page/@dest_off and @src_page/@src_off must be mappable to a bus
+ * address according to the DMA mapping API rules for streaming mappings.
+ * Both @dest_page/@dest_off and @src_page/@src_off must stay memory resident
+ * (kernel memory or locked user space pages)
+ */
+static inline dma_cookie_t dma_async_memcpy_pg_to_pg(struct dma_chan *chan,
+ struct page *dest_pg, unsigned int dest_off, struct page *src_pg,
+ unsigned int src_off, size_t len)
+{
+ int cpu = get_cpu();
+ per_cpu_ptr(chan->local, cpu)->bytes_transferred += len;
+ per_cpu_ptr(chan->local, cpu)->memcpy_count++;
+ put_cpu();
+
+ return chan->device->device_memcpy_pg_to_pg(chan, dest_pg, dest_off,
+ src_pg, src_off, len);
+}
+
+/**
+ * dma_async_memcpy_issue_pending - flush pending copies to HW
+ * @chan:
+ *
+ * This allows drivers to push copies to HW in batches,
+ * reducing MMIO writes where possible.
+ */
+static inline void dma_async_memcpy_issue_pending(struct dma_chan *chan)
+{
+ return chan->device->device_memcpy_issue_pending(chan);
+}
+
+/**
+ * dma_async_memcpy_complete - poll for transaction completion
+ * @chan: DMA channel
+ * @cookie: transaction identifier to check status of
+ * @last: returns last completed cookie, can be NULL
+ * @used: returns last issued cookie, can be NULL
+ *
+ * If @last and @used are passed in, upon return they reflect the driver
+ * internal state and can be used with dma_async_is_complete() to check
+ * the status of multiple cookies without re-checking hardware state.
+ */
+static inline enum dma_status dma_async_memcpy_complete(struct dma_chan *chan,
+ dma_cookie_t cookie, dma_cookie_t *last, dma_cookie_t *used)
+{
+ return chan->device->device_memcpy_complete(chan, cookie, last, used);
+}
+
+/**
+ * dma_async_is_complete - test a cookie against chan state
+ * @cookie: transaction identifier to test status of
+ * @last_complete: last know completed transaction
+ * @last_used: last cookie value handed out
+ *
+ * dma_async_is_complete() is used in dma_async_memcpy_complete()
+ * the test logic is seperated for lightweight testing of multiple cookies
+ */
+static inline enum dma_status dma_async_is_complete(dma_cookie_t cookie,
+ dma_cookie_t last_complete, dma_cookie_t last_used) {
+ if (last_complete <= last_used) {
+ if ((cookie <= last_complete) || (cookie > last_used))
+ return DMA_SUCCESS;
+ } else {
+ if ((cookie <= last_complete) && (cookie > last_used))
+ return DMA_SUCCESS;
+ }
+ return DMA_IN_PROGRESS;
+}
+
+
+/* --- DMA device --- */
+
+int dma_async_device_register(struct dma_device *device);
+void dma_async_device_unregister(struct dma_device *device);
+
+#endif /* DMAENGINE_H */
^ permalink raw reply related
* [PATCH 3/8] [I/OAT] Setup the networking subsystem as a DMA client
From: Chris Leech @ 2006-03-03 21:42 UTC (permalink / raw)
To: linux-kernel, netdev
In-Reply-To: <20060303214036.11908.10499.stgit@gitlost.site>
Attempts to allocate per-CPU DMA channels
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
drivers/dma/Kconfig | 12 +++++
include/linux/netdevice.h | 6 +++
include/net/netdma.h | 36 ++++++++++++++++
net/core/dev.c | 102 +++++++++++++++++++++++++++++++++++++++++++++
4 files changed, 156 insertions(+), 0 deletions(-)
diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index 0f15e76..30d021d 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -10,6 +10,18 @@ config DMA_ENGINE
DMA engines offload copy operations from the CPU to dedicated
hardware, allowing the copies to happen asynchronously.
+comment "DMA Clients"
+
+config NET_DMA
+ bool "Network: TCP receive copy offload"
+ depends on DMA_ENGINE && NET
+ default y
+ ---help---
+ This enables the use of DMA engines in the network stack to
+ offload receive copy-to-user operations, freeing CPU cycles.
+ Since this is the main user of the DMA engine, it should be enabled;
+ say Y here.
+
comment "DMA Devices"
config INTEL_IOATDMA
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index b825be2..ecbde56 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -37,6 +37,9 @@
#include <linux/config.h>
#include <linux/device.h>
#include <linux/percpu.h>
+#ifdef CONFIG_NET_DMA
+#include <linux/dmaengine.h>
+#endif
struct divert_blk;
struct vlan_group;
@@ -592,6 +595,9 @@ struct softnet_data
struct sk_buff *completion_queue;
struct net_device backlog_dev; /* Sorry. 8) */
+#ifdef CONFIG_NET_DMA
+ struct dma_chan *net_dma;
+#endif
};
DECLARE_PER_CPU(struct softnet_data,softnet_data);
diff --git a/include/net/netdma.h b/include/net/netdma.h
new file mode 100644
index 0000000..3cd9e6d
--- /dev/null
+++ b/include/net/netdma.h
@@ -0,0 +1,36 @@
+/*****************************************************************************
+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.
+
+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 LICENSE.
+*****************************************************************************/
+#ifndef NETDMA_H
+#define NETDMA_H
+#include <linux/dmaengine.h>
+
+static inline struct dma_chan *get_softnet_dma(void)
+{
+ struct dma_chan *chan;
+ rcu_read_lock();
+ chan = rcu_dereference(__get_cpu_var(softnet_data.net_dma));
+ if (chan)
+ dma_chan_get(chan);
+ rcu_read_unlock();
+ return chan;
+}
+
+#endif /* NETDMA_H */
diff --git a/net/core/dev.c b/net/core/dev.c
index 7ca47bf..b54e5f3 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -114,6 +114,7 @@
#include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
#include <net/iw_handler.h>
#endif /* CONFIG_NET_RADIO */
+#include <linux/dmaengine.h>
#include <asm/current.h>
/*
@@ -148,6 +149,11 @@ static DEFINE_SPINLOCK(ptype_lock);
static struct list_head ptype_base[16]; /* 16 way hashed list */
static struct list_head ptype_all; /* Taps */
+#ifdef CONFIG_NET_DMA
+static struct dma_client *net_dma_client;
+static unsigned int net_dma_count;
+#endif
+
/*
* The @dev_base list is protected by @dev_base_lock and the rtln
* semaphore.
@@ -1719,6 +1725,9 @@ static void net_rx_action(struct softirq
unsigned long start_time = jiffies;
int budget = netdev_budget;
void *have;
+#ifdef CONFIG_NET_DMA
+ struct dma_chan *chan;
+#endif
local_irq_disable();
@@ -1750,6 +1759,18 @@ static void net_rx_action(struct softirq
}
}
out:
+#ifdef CONFIG_NET_DMA
+ /*
+ * There may not be any more sk_buffs coming right now, so push
+ * any pending DMA copies to hardware
+ */
+ if (net_dma_client) {
+ rcu_read_lock();
+ list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
+ dma_async_memcpy_issue_pending(chan);
+ rcu_read_unlock();
+ }
+#endif
local_irq_enable();
return;
@@ -3205,6 +3226,85 @@ static int dev_cpu_callback(struct notif
}
#endif /* CONFIG_HOTPLUG_CPU */
+#ifdef CONFIG_NET_DMA
+/**
+ * net_dma_rebalance -
+ * This is called when the number of channels allocated to the net_dma_client
+ * changes. The net_dma_client tries to have one DMA channel per CPU.
+ */
+static void net_dma_rebalance(void)
+{
+ unsigned int cpu, i, n;
+ struct dma_chan *chan;
+
+ lock_cpu_hotplug();
+
+ if (net_dma_count == 0) {
+ for_each_online_cpu(cpu)
+ rcu_assign_pointer(per_cpu(softnet_data.net_dma, cpu), NULL);
+ unlock_cpu_hotplug();
+ return;
+ }
+
+ i = 0;
+ cpu = first_cpu(cpu_online_map);
+
+ rcu_read_lock();
+ list_for_each_entry(chan, &net_dma_client->channels, client_node) {
+ n = ((num_online_cpus() / net_dma_count)
+ + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
+
+ while(n) {
+ per_cpu(softnet_data.net_dma, cpu) = chan;
+ cpu = next_cpu(cpu, cpu_online_map);
+ n--;
+ }
+ i++;
+ }
+ rcu_read_unlock();
+
+ unlock_cpu_hotplug();
+}
+
+/**
+ * netdev_dma_event - event callback for the net_dma_client
+ * @client: should always be net_dma_client
+ * @chan:
+ * @event:
+ */
+static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
+ enum dma_event event)
+{
+ switch (event) {
+ case DMA_RESOURCE_ADDED:
+ net_dma_count++;
+ net_dma_rebalance();
+ break;
+ case DMA_RESOURCE_REMOVED:
+ net_dma_count--;
+ net_dma_rebalance();
+ break;
+ default:
+ break;
+ }
+}
+
+/**
+ * netdev_dma_regiser - register the networking subsystem as a DMA client
+ */
+static int __init netdev_dma_register(void)
+{
+ net_dma_client = dma_async_client_register(netdev_dma_event);
+ if (net_dma_client == NULL)
+ return -ENOMEM;
+
+ dma_async_client_chan_request(net_dma_client, num_online_cpus());
+ return 0;
+}
+
+#else
+static int __init netdev_dma_register(void) { return -ENODEV; }
+#endif /* CONFIG_NET_DMA */
/*
* Initialize the DEV module. At boot time this walks the device list and
@@ -3258,6 +3358,8 @@ static int __init net_dev_init(void)
atomic_set(&queue->backlog_dev.refcnt, 1);
}
+ netdev_dma_register();
+
dev_boot_phase = 0;
open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
^ permalink raw reply related
* [PATCH 4/8] [I/OAT] Utility functions for offloading sk_buff to iovec copies
From: Chris Leech @ 2006-03-03 21:42 UTC (permalink / raw)
To: linux-kernel, netdev
In-Reply-To: <20060303214036.11908.10499.stgit@gitlost.site>
Provides for pinning user space pages in memory, copying to iovecs,
and copying from sk_buffs including fragmented and chained sk_buffs.
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
drivers/dma/Makefile | 1
drivers/dma/iovlock.c | 320 +++++++++++++++++++++++++++++++++++++++++++++
include/linux/dmaengine.h | 24 +++
include/net/netdma.h | 5 +
net/core/Makefile | 3
net/core/user_dma.c | 133 +++++++++++++++++++
6 files changed, 485 insertions(+), 1 deletions(-)
diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile
index c8a5f56..ea2f110 100644
--- a/drivers/dma/Makefile
+++ b/drivers/dma/Makefile
@@ -1,2 +1,3 @@
obj-y += dmaengine.o
+obj-$(CONFIG_NET_DMA) += iovlock.o
obj-$(CONFIG_INTEL_IOATDMA) += ioatdma.o
diff --git a/drivers/dma/iovlock.c b/drivers/dma/iovlock.c
new file mode 100644
index 0000000..edbf581
--- /dev/null
+++ b/drivers/dma/iovlock.c
@@ -0,0 +1,320 @@
+/*****************************************************************************
+Copyright(c) 2004 - 2006 Intel Corporation. All rights reserved.
+Portions based on net/core/datagram.c and copyrighted by their authors.
+
+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 LICENSE.
+*****************************************************************************/
+
+/*
+ * This code allows the net stack to make use of a DMA engine for
+ * skb to iovec copies.
+ */
+
+#include <linux/dmaengine.h>
+#include <linux/pagemap.h>
+#include <net/tcp.h> /* for memcpy_toiovec */
+#include <asm/io.h>
+#include <asm/uaccess.h>
+
+#ifdef CONFIG_DMA_ENGINE
+
+#define NUM_PAGES_SPANNED(start, length) \
+ ((PAGE_ALIGN((unsigned long)start + length) - \
+ ((unsigned long)start & PAGE_MASK)) >> PAGE_SHIFT)
+
+/*
+ * Lock down all the iovec pages needed for len bytes.
+ * Return a struct dma_locked_list to keep track of pages locked down.
+ *
+ * We are allocating a single chunk of memory, and then carving it up into
+ * 3 sections, the latter 2 whose size depends on the number of iovecs and the
+ * total number of pages, respectively.
+ */
+int dma_lock_iovec_pages(struct iovec *iov, size_t len, struct dma_locked_list
+ **locked_list)
+{
+ struct dma_locked_list *local_list;
+ struct page **pages;
+ int i;
+ int ret;
+
+ int nr_iovecs = 0;
+ int iovec_len_used = 0;
+ int iovec_pages_used = 0;
+
+ /* don't lock down non-user-based iovecs */
+ if (segment_eq(get_fs(), KERNEL_DS)) {
+ *locked_list = NULL;
+ return 0;
+ }
+
+ /* determine how many iovecs/pages there are, up front */
+ do {
+ iovec_len_used += iov[nr_iovecs].iov_len;
+ iovec_pages_used += NUM_PAGES_SPANNED(iov[nr_iovecs].iov_base,
+ iov[nr_iovecs].iov_len);
+ nr_iovecs++;
+ } while (iovec_len_used < len);
+
+ /* single kmalloc for locked list, page_list[], and the page arrays */
+ local_list = kmalloc(sizeof(*local_list)
+ + (nr_iovecs * sizeof (struct dma_page_list))
+ + (iovec_pages_used * sizeof (struct page*)), GFP_KERNEL);
+ if (!local_list)
+ return -ENOMEM;
+
+ /* list of pages starts right after the page list array */
+ pages = (struct page **) &local_list->page_list[nr_iovecs];
+
+ /* it's a userspace pointer */
+ might_sleep();
+
+ for (i = 0; i < nr_iovecs; i++) {
+ struct dma_page_list *page_list = &local_list->page_list[i];
+
+ len -= iov[i].iov_len;
+
+ if (!access_ok(VERIFY_WRITE, iov[i].iov_base, iov[i].iov_len)) {
+ dma_unlock_iovec_pages(local_list);
+ return -EFAULT;
+ }
+
+ page_list->nr_pages = NUM_PAGES_SPANNED(iov[i].iov_base,
+ iov[i].iov_len);
+ page_list->base_address = iov[i].iov_base;
+
+ page_list->pages = pages;
+ pages += page_list->nr_pages;
+
+ /* lock pages down */
+ down_read(¤t->mm->mmap_sem);
+ ret = get_user_pages(
+ current,
+ current->mm,
+ (unsigned long) iov[i].iov_base,
+ page_list->nr_pages,
+ 1,
+ 0,
+ page_list->pages,
+ NULL);
+ up_read(¤t->mm->mmap_sem);
+
+ if (ret != page_list->nr_pages) {
+ goto mem_error;
+ }
+
+ local_list->nr_iovecs = i + 1;
+ }
+
+ *locked_list = local_list;
+ return 0;
+
+mem_error:
+ dma_unlock_iovec_pages(local_list);
+ return -ENOMEM;
+}
+
+void dma_unlock_iovec_pages(struct dma_locked_list *locked_list)
+{
+ int i, j;
+
+ if (!locked_list)
+ return;
+
+ for (i = 0; i < locked_list->nr_iovecs; i++) {
+ struct dma_page_list *page_list = &locked_list->page_list[i];
+ for (j = 0; j < page_list->nr_pages; j++) {
+ SetPageDirty(page_list->pages[j]);
+ page_cache_release(page_list->pages[j]);
+ }
+ }
+
+ kfree(locked_list);
+}
+
+static dma_cookie_t dma_memcpy_tokerneliovec(struct dma_chan *chan, struct
+ iovec *iov, unsigned char *kdata, size_t len)
+{
+ dma_cookie_t dma_cookie = 0;
+
+ while (len > 0) {
+ if (iov->iov_len) {
+ int copy = min_t(unsigned int, iov->iov_len, len);
+ dma_cookie = dma_async_memcpy_buf_to_buf(
+ chan,
+ iov->iov_base,
+ kdata,
+ copy);
+ kdata += copy;
+ len -= copy;
+ iov->iov_len -= copy;
+ iov->iov_base += copy;
+ }
+ iov++;
+ }
+
+ return dma_cookie;
+}
+
+/*
+ * We have already locked down the pages we will be using in the iovecs.
+ * Each entry in iov array has corresponding entry in locked_list->page_list.
+ * Using array indexing to keep iov[] and page_list[] in sync.
+ * Initial elements in iov array's iov->iov_len will be 0 if already copied into
+ * by another call.
+ * iov array length remaining guaranteed to be bigger than len.
+ */
+dma_cookie_t dma_memcpy_toiovec(struct dma_chan *chan, struct iovec *iov,
+ struct dma_locked_list *locked_list, unsigned char *kdata, size_t len)
+{
+ int iov_byte_offset;
+ int copy;
+ dma_cookie_t dma_cookie = 0;
+ int iovec_idx;
+ int page_idx;
+
+ if (!chan)
+ return memcpy_toiovec(iov, kdata, len);
+
+ /* -> kernel copies (e.g. smbfs) */
+ if (!locked_list)
+ return dma_memcpy_tokerneliovec(chan, iov, kdata, len);
+
+ iovec_idx = 0;
+ while (iovec_idx < locked_list->nr_iovecs) {
+ struct dma_page_list *page_list;
+
+ /* skip already used-up iovecs */
+ while (!iov[iovec_idx].iov_len)
+ iovec_idx++;
+
+ page_list = &locked_list->page_list[iovec_idx];
+
+ iov_byte_offset = ((unsigned long)iov[iovec_idx].iov_base & ~PAGE_MASK);
+ page_idx = (((unsigned long)iov[iovec_idx].iov_base & PAGE_MASK)
+ - ((unsigned long)page_list->base_address & PAGE_MASK)) >> PAGE_SHIFT;
+
+ /* break up copies to not cross page boundary */
+ while (iov[iovec_idx].iov_len) {
+ copy = min_t(int, PAGE_SIZE - iov_byte_offset, len);
+ copy = min_t(int, copy, iov[iovec_idx].iov_len);
+
+ dma_cookie = dma_async_memcpy_buf_to_pg(chan,
+ page_list->pages[page_idx],
+ iov_byte_offset,
+ kdata,
+ copy);
+
+ len -= copy;
+ iov[iovec_idx].iov_len -= copy;
+ iov[iovec_idx].iov_base += copy;
+
+ if (!len)
+ return dma_cookie;
+
+ kdata += copy;
+ iov_byte_offset = 0;
+ page_idx++;
+ }
+ iovec_idx++;
+ }
+
+ /* really bad if we ever run out of iovecs */
+ BUG();
+ return -EFAULT;
+}
+
+dma_cookie_t dma_memcpy_pg_toiovec(struct dma_chan *chan, struct iovec *iov,
+ struct dma_locked_list *locked_list, struct page *page,
+ unsigned int offset, size_t len)
+{
+ int iov_byte_offset;
+ int copy;
+ dma_cookie_t dma_cookie = 0;
+ int iovec_idx;
+ int page_idx;
+ int err;
+
+ /* this needs as-yet-unimplemented buf-to-buff, so punt. */
+ /* TODO: use dma for this */
+ if (!chan || !locked_list) {
+ u8 *vaddr = kmap(page);
+ err = memcpy_toiovec(iov, vaddr + offset, len);
+ kunmap(page);
+ return err;
+ }
+
+ iovec_idx = 0;
+ while (iovec_idx < locked_list->nr_iovecs) {
+ struct dma_page_list *page_list;
+
+ /* skip already used-up iovecs */
+ while (!iov[iovec_idx].iov_len)
+ iovec_idx++;
+
+ page_list = &locked_list->page_list[iovec_idx];
+
+ iov_byte_offset = ((unsigned long)iov[iovec_idx].iov_base & ~PAGE_MASK);
+ page_idx = (((unsigned long)iov[iovec_idx].iov_base & PAGE_MASK)
+ - ((unsigned long)page_list->base_address & PAGE_MASK)) >> PAGE_SHIFT;
+
+ /* break up copies to not cross page boundary */
+ while (iov[iovec_idx].iov_len) {
+ copy = min_t(int, PAGE_SIZE - iov_byte_offset, len);
+ copy = min_t(int, copy, iov[iovec_idx].iov_len);
+
+ dma_cookie = dma_async_memcpy_pg_to_pg(chan,
+ page_list->pages[page_idx],
+ iov_byte_offset,
+ page,
+ offset,
+ copy);
+
+ len -= copy;
+ iov[iovec_idx].iov_len -= copy;
+ iov[iovec_idx].iov_base += copy;
+
+ if (!len)
+ return dma_cookie;
+
+ offset += copy;
+ iov_byte_offset = 0;
+ page_idx++;
+ }
+ iovec_idx++;
+ }
+
+ /* really bad if we ever run out of iovecs */
+ BUG();
+ return -EFAULT;
+}
+
+#else
+
+int dma_lock_iovec_pages(struct iovec *iov, size_t len, struct dma_locked_list
+ **locked_list)
+{
+ *locked_list = NULL;
+
+ return 0;
+}
+
+void dma_unlock_iovec_pages(struct dma_locked_list* locked_list)
+{ }
+
+#endif
diff --git a/include/linux/dmaengine.h b/include/linux/dmaengine.h
index f8a77ab..e198712 100644
--- a/include/linux/dmaengine.h
+++ b/include/linux/dmaengine.h
@@ -319,4 +319,28 @@ static inline enum dma_status dma_async_
int dma_async_device_register(struct dma_device *device);
void dma_async_device_unregister(struct dma_device *device);
+/* --- Helper iov-locking functions --- */
+
+struct dma_page_list
+{
+ char *base_address;
+ int nr_pages;
+ struct page **pages;
+};
+
+struct dma_locked_list
+{
+ int nr_iovecs;
+ struct dma_page_list page_list[0];
+};
+
+int dma_lock_iovec_pages(struct iovec *iov, size_t len,
+ struct dma_locked_list **locked_list);
+void dma_unlock_iovec_pages(struct dma_locked_list* locked_list);
+dma_cookie_t dma_memcpy_toiovec(struct dma_chan *chan, struct iovec *iov,
+ struct dma_locked_list *locked_list, unsigned char *kdata, size_t len);
+dma_cookie_t dma_memcpy_pg_toiovec(struct dma_chan *chan, struct iovec *iov,
+ struct dma_locked_list *locked_list, struct page *page,
+ unsigned int offset, size_t len);
+
#endif /* DMAENGINE_H */
diff --git a/include/net/netdma.h b/include/net/netdma.h
index 3cd9e6d..415d74c 100644
--- a/include/net/netdma.h
+++ b/include/net/netdma.h
@@ -21,6 +21,7 @@ file called LICENSE.
#ifndef NETDMA_H
#define NETDMA_H
#include <linux/dmaengine.h>
+#include <linux/skbuff.h>
static inline struct dma_chan *get_softnet_dma(void)
{
@@ -33,4 +34,8 @@ static inline struct dma_chan *get_softn
return chan;
}
+int dma_skb_copy_datagram_iovec(struct dma_chan* chan,
+ const struct sk_buff *skb, int offset, struct iovec *to,
+ size_t len, struct dma_locked_list *locked_list);
+
#endif /* NETDMA_H */
diff --git a/net/core/Makefile b/net/core/Makefile
index 630da0f..d02132b 100644
--- a/net/core/Makefile
+++ b/net/core/Makefile
@@ -8,7 +8,8 @@ obj-y := sock.o request_sock.o skbuff.o
obj-$(CONFIG_SYSCTL) += sysctl_net_core.o
obj-y += dev.o ethtool.o dev_mcast.o dst.o \
- neighbour.o rtnetlink.o utils.o link_watch.o filter.o
+ neighbour.o rtnetlink.o utils.o link_watch.o filter.o \
+ user_dma.o
obj-$(CONFIG_XFRM) += flow.o
obj-$(CONFIG_SYSFS) += net-sysfs.o
diff --git a/net/core/user_dma.c b/net/core/user_dma.c
new file mode 100644
index 0000000..1e1aae5
--- /dev/null
+++ b/net/core/user_dma.c
@@ -0,0 +1,133 @@
+/*****************************************************************************
+Copyright(c) 2004 - 2006 Intel Corporation. All rights reserved.
+Portions based on net/core/datagram.c and copyrighted by their authors.
+
+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 LICENSE.
+*****************************************************************************/
+
+/*
+ * This code allows the net stack to make use of a DMA engine for
+ * skb to iovec copies.
+ */
+
+#include <linux/dmaengine.h>
+#include <linux/socket.h>
+#include <linux/rtnetlink.h> /* for BUG_TRAP */
+#include <net/tcp.h>
+
+
+#ifdef CONFIG_NET_DMA
+
+/**
+ * dma_skb_copy_datagram_iovec - Copy a datagram to an iovec.
+ * @skb - buffer to copy
+ * @offset - offset in the buffer to start copying from
+ * @iovec - io vector to copy to
+ * @len - amount of data to copy from buffer to iovec
+ * @locked_list - locked iovec buffer data
+ *
+ * Note: the iovec is modified during the copy.
+ */
+int dma_skb_copy_datagram_iovec(struct dma_chan *chan,
+ struct sk_buff *skb, int offset, struct iovec *to,
+ size_t len, struct dma_locked_list *locked_list)
+{
+ int start = skb_headlen(skb);
+ int i, copy = start - offset;
+ dma_cookie_t cookie = 0;
+
+ /* Copy header. */
+ if (copy > 0) {
+ if (copy > len)
+ copy = len;
+ if ((cookie = dma_memcpy_toiovec(chan, to, locked_list,
+ skb->data + offset, copy)) < 0)
+ goto fault;
+ if ((len -= copy) == 0)
+ goto end;
+ offset += copy;
+ }
+
+ /* Copy paged appendix. Hmm... why does this look so complicated? */
+ for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+ int end;
+
+ BUG_TRAP(start <= offset + len);
+
+ end = start + skb_shinfo(skb)->frags[i].size;
+ if ((copy = end - offset) > 0) {
+ skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
+ struct page *page = frag->page;
+
+ if (copy > len)
+ copy = len;
+
+ cookie = dma_memcpy_pg_toiovec(chan, to, locked_list, page,
+ frag->page_offset + offset - start, copy);
+ if (cookie < 0)
+ goto fault;
+ if (!(len -= copy))
+ goto end;
+ offset += copy;
+ }
+ start = end;
+ }
+
+ if (skb_shinfo(skb)->frag_list) {
+ struct sk_buff *list = skb_shinfo(skb)->frag_list;
+
+ for (; list; list = list->next) {
+ int end;
+
+ BUG_TRAP(start <= offset + len);
+
+ end = start + list->len;
+ if ((copy = end - offset) > 0) {
+ if (copy > len)
+ copy = len;
+ if ((cookie = dma_skb_copy_datagram_iovec(chan, list,
+ offset - start, to, copy, locked_list)) < 0)
+ goto fault;
+ if ((len -= copy) == 0)
+ goto end;
+ offset += copy;
+ }
+ start = end;
+ }
+ }
+
+end:
+ if (!len) {
+ skb->dma_cookie = cookie;
+ return cookie;
+ }
+
+fault:
+ return -EFAULT;
+}
+
+#else
+
+int dma_skb_copy_datagram_iovec(struct dma_chan *chan,
+ const struct sk_buff *skb, int offset, struct iovec *to,
+ size_t len, struct dma_locked_list *locked_list)
+{
+ return skb_copy_datagram_iovec(skb, offset, to, len);
+}
+
+#endif
^ permalink raw reply related
* [PATCH 5/8] [I/OAT] Structure changes for TCP recv offload to I/OAT
From: Chris Leech @ 2006-03-03 21:42 UTC (permalink / raw)
To: linux-kernel, netdev
In-Reply-To: <20060303214036.11908.10499.stgit@gitlost.site>
Adds an async_wait_queue and some additional fields to tcp_sock, and a
dma_cookie_t to sk_buff.
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
include/linux/skbuff.h | 6 ++++++
include/linux/tcp.h | 10 ++++++++++
include/net/sock.h | 2 ++
include/net/tcp.h | 9 +++++++++
net/core/sock.c | 6 ++++++
5 files changed, 33 insertions(+), 0 deletions(-)
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 75c9631..572b7ae 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -29,6 +29,9 @@
#include <linux/net.h>
#include <linux/textsearch.h>
#include <net/checksum.h>
+#ifdef CONFIG_NET_DMA
+#include <linux/dmaengine.h>
+#endif
#define HAVE_ALLOC_SKB /* For the drivers to know */
#define HAVE_ALIGNABLE_SKB /* Ditto 8) */
@@ -285,6 +288,9 @@ struct sk_buff {
__u16 tc_verd; /* traffic control verdict */
#endif
#endif
+#ifdef CONFIG_NET_DMA
+ dma_cookie_t dma_cookie;
+#endif
/* These elements must be at the end, see alloc_skb() for details. */
diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index 542d395..6d7dc19 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -18,6 +18,9 @@
#define _LINUX_TCP_H
#include <linux/types.h>
+#ifdef CONFIG_NET_DMA
+#include <linux/dmaengine.h>
+#endif
#include <asm/byteorder.h>
struct tcphdr {
@@ -233,6 +236,13 @@ struct tcp_sock {
struct iovec *iov;
int memory;
int len;
+#ifdef CONFIG_NET_DMA
+ /* members for async copy */
+ struct dma_chan *dma_chan;
+ int wakeup;
+ struct dma_locked_list *locked_list;
+ dma_cookie_t dma_cookie;
+#endif
} ucopy;
__u32 snd_wl1; /* Sequence for window update */
diff --git a/include/net/sock.h b/include/net/sock.h
index 3075803..5d1b895 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -132,6 +132,7 @@ struct sock_common {
* @sk_receive_queue: incoming packets
* @sk_wmem_alloc: transmit queue bytes committed
* @sk_write_queue: Packet sending queue
+ * @sk_async_wait_queue: DMA copied packets
* @sk_omem_alloc: "o" is "option" or "other"
* @sk_wmem_queued: persistent queue size
* @sk_forward_alloc: space allocated forward
@@ -205,6 +206,7 @@ struct sock {
atomic_t sk_omem_alloc;
struct sk_buff_head sk_receive_queue;
struct sk_buff_head sk_write_queue;
+ struct sk_buff_head sk_async_wait_queue;
int sk_wmem_queued;
int sk_forward_alloc;
gfp_t sk_allocation;
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 16879fa..fd7c3e4 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -28,6 +28,9 @@
#include <linux/cache.h>
#include <linux/percpu.h>
#include <linux/skbuff.h>
+#ifdef CONFIG_NET_DMA
+#include <linux/dmaengine.h>
+#endif
#include <net/inet_connection_sock.h>
#include <net/inet_timewait_sock.h>
@@ -813,6 +816,12 @@ static inline void tcp_prequeue_init(str
tp->ucopy.len = 0;
tp->ucopy.memory = 0;
skb_queue_head_init(&tp->ucopy.prequeue);
+#ifdef CONFIG_NET_DMA
+ tp->ucopy.dma_chan = NULL;
+ tp->ucopy.wakeup = 0;
+ tp->ucopy.locked_list = NULL;
+ tp->ucopy.dma_cookie = 0;
+#endif
}
/* Packet is added to VJ-style prequeue for processing in process
diff --git a/net/core/sock.c b/net/core/sock.c
index 6e00811..90275ec 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -724,6 +724,9 @@ struct sock *sk_clone(const struct sock
atomic_set(&newsk->sk_omem_alloc, 0);
skb_queue_head_init(&newsk->sk_receive_queue);
skb_queue_head_init(&newsk->sk_write_queue);
+#ifdef CONFIG_NET_DMA
+ skb_queue_head_init(&newsk->sk_async_wait_queue);
+#endif
rwlock_init(&newsk->sk_dst_lock);
rwlock_init(&newsk->sk_callback_lock);
@@ -1275,6 +1278,9 @@ void sock_init_data(struct socket *sock,
skb_queue_head_init(&sk->sk_receive_queue);
skb_queue_head_init(&sk->sk_write_queue);
skb_queue_head_init(&sk->sk_error_queue);
+#ifdef CONFIG_NET_DMA
+ skb_queue_head_init(&sk->sk_async_wait_queue);
+#endif
sk->sk_send_head = NULL;
^ permalink raw reply related
* [PATCH 6/8] [I/OAT] Rename cleanup_rbuf to tcp_cleanup_rbuf and make non-static
From: Chris Leech @ 2006-03-03 21:42 UTC (permalink / raw)
To: linux-kernel, netdev
In-Reply-To: <20060303214036.11908.10499.stgit@gitlost.site>
Needed to be able to call tcp_cleanup_rbuf in tcp_input.c for I/OAT
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
include/net/tcp.h | 2 ++
net/ipv4/tcp.c | 10 +++++-----
2 files changed, 7 insertions(+), 5 deletions(-)
diff --git a/include/net/tcp.h b/include/net/tcp.h
index fd7c3e4..2fc7f05 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -295,6 +295,8 @@ extern int tcp_rcv_established(struct
extern void tcp_rcv_space_adjust(struct sock *sk);
+extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
+
extern int tcp_twsk_unique(struct sock *sk,
struct sock *sktw, void *twp);
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 00aa80e..13abfa2 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -936,7 +936,7 @@ static int tcp_recv_urg(struct sock *sk,
* calculation of whether or not we must ACK for the sake of
* a window update.
*/
-static void cleanup_rbuf(struct sock *sk, int copied)
+void tcp_cleanup_rbuf(struct sock *sk, int copied)
{
struct tcp_sock *tp = tcp_sk(sk);
int time_to_ack = 0;
@@ -1085,7 +1085,7 @@ int tcp_read_sock(struct sock *sk, read_
/* Clean up data we have read: This will do ACK frames. */
if (copied)
- cleanup_rbuf(sk, copied);
+ tcp_cleanup_rbuf(sk, copied);
return copied;
}
@@ -1219,7 +1219,7 @@ int tcp_recvmsg(struct kiocb *iocb, stru
}
}
- cleanup_rbuf(sk, copied);
+ tcp_cleanup_rbuf(sk, copied);
if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) {
/* Install new reader */
@@ -1390,7 +1390,7 @@ skip_copy:
*/
/* Clean up data we have read: This will do ACK frames. */
- cleanup_rbuf(sk, copied);
+ tcp_cleanup_rbuf(sk, copied);
TCP_CHECK_TIMER(sk);
release_sock(sk);
@@ -1856,7 +1856,7 @@ int tcp_setsockopt(struct sock *sk, int
(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
inet_csk_ack_scheduled(sk)) {
icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
- cleanup_rbuf(sk, 1);
+ tcp_cleanup_rbuf(sk, 1);
if (!(val & 1))
icsk->icsk_ack.pingpong = 1;
}
^ permalink raw reply related
* [PATCH 7/8] [I/OAT] Add a sysctl for tuning the I/OAT offloaded I/O threshold
From: Chris Leech @ 2006-03-03 21:42 UTC (permalink / raw)
To: linux-kernel, netdev
In-Reply-To: <20060303214036.11908.10499.stgit@gitlost.site>
Any socket recv of less than this ammount will not be offloaded
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
include/linux/sysctl.h | 1 +
include/net/tcp.h | 1 +
net/core/user_dma.c | 4 ++++
net/ipv4/sysctl_net_ipv4.c | 10 ++++++++++
4 files changed, 16 insertions(+), 0 deletions(-)
diff --git a/include/linux/sysctl.h b/include/linux/sysctl.h
index dfcf449..f532f1e 100644
--- a/include/linux/sysctl.h
+++ b/include/linux/sysctl.h
@@ -402,6 +402,7 @@ enum
NET_IPV4_IPFRAG_MAX_DIST=112,
NET_TCP_MTU_PROBING=113,
NET_TCP_BASE_MSS=114,
+ NET_TCP_DMA_COPYBREAK=115,
};
enum {
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 2fc7f05..0740f32 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -221,6 +221,7 @@ extern int sysctl_tcp_adv_win_scale;
extern int sysctl_tcp_tw_reuse;
extern int sysctl_tcp_frto;
extern int sysctl_tcp_low_latency;
+extern int sysctl_tcp_dma_copybreak;
extern int sysctl_tcp_nometrics_save;
extern int sysctl_tcp_moderate_rcvbuf;
extern int sysctl_tcp_tso_win_divisor;
diff --git a/net/core/user_dma.c b/net/core/user_dma.c
index 1e1aae5..dd259f0 100644
--- a/net/core/user_dma.c
+++ b/net/core/user_dma.c
@@ -33,6 +33,10 @@ file called LICENSE.
#ifdef CONFIG_NET_DMA
+#define NET_DMA_DEFAULT_COPYBREAK 1024
+
+int sysctl_tcp_dma_copybreak = NET_DMA_DEFAULT_COPYBREAK;
+
/**
* dma_skb_copy_datagram_iovec - Copy a datagram to an iovec.
* @skb - buffer to copy
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index ebf2e0b..f7bd9c2 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -680,6 +680,16 @@ ctl_table ipv4_table[] = {
.mode = 0644,
.proc_handler = &proc_dointvec,
},
+#ifdef CONFIG_NET_DMA
+ {
+ .ctl_name = NET_TCP_DMA_COPYBREAK,
+ .procname = "tcp_dma_copybreak",
+ .data = &sysctl_tcp_dma_copybreak,
+ .maxlen = sizeof(int),
+ .mode = 0644,
+ .proc_handler = &proc_dointvec
+ },
+#endif
{ .ctl_name = 0 }
};
^ permalink raw reply related
* [PATCH 8/8] [I/OAT] TCP recv offload to I/OAT
From: Chris Leech @ 2006-03-03 21:42 UTC (permalink / raw)
To: linux-kernel, netdev
In-Reply-To: <20060303214036.11908.10499.stgit@gitlost.site>
Locks down user pages and sets up for DMA in tcp_recvmsg, then calls
dma_async_try_early_copy in tcp_v4_do_rcv
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
include/net/netdma.h | 1
net/ipv4/tcp.c | 111 +++++++++++++++++++++++++++++++++++++++++++++-----
net/ipv4/tcp_input.c | 78 ++++++++++++++++++++++++++++++++---
net/ipv4/tcp_ipv4.c | 20 +++++++++
net/ipv6/tcp_ipv6.c | 12 +++++
5 files changed, 201 insertions(+), 21 deletions(-)
diff --git a/include/net/netdma.h b/include/net/netdma.h
index 415d74c..2d829e1 100644
--- a/include/net/netdma.h
+++ b/include/net/netdma.h
@@ -37,5 +37,6 @@ static inline struct dma_chan *get_softn
int dma_skb_copy_datagram_iovec(struct dma_chan* chan,
const struct sk_buff *skb, int offset, struct iovec *to,
size_t len, struct dma_locked_list *locked_list);
+int dma_async_try_early_copy(struct sock *sk, struct sk_buff *skb, int hlen);
#endif /* NETDMA_H */
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 13abfa2..b792048 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -262,6 +262,9 @@
#include <net/tcp.h>
#include <net/xfrm.h>
#include <net/ip.h>
+#ifdef CONFIG_NET_DMA
+#include <net/netdma.h>
+#endif
#include <asm/uaccess.h>
@@ -1109,6 +1112,7 @@ int tcp_recvmsg(struct kiocb *iocb, stru
int target; /* Read at least this many bytes */
long timeo;
struct task_struct *user_recv = NULL;
+ int copied_early = 0;
lock_sock(sk);
@@ -1132,6 +1136,12 @@ int tcp_recvmsg(struct kiocb *iocb, stru
target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
+#ifdef CONFIG_NET_DMA
+ tp->ucopy.dma_chan = NULL;
+ if ((len > sysctl_tcp_dma_copybreak) && !(flags & MSG_PEEK) && !sysctl_tcp_low_latency && __get_cpu_var(softnet_data.net_dma))
+ dma_lock_iovec_pages(msg->msg_iov, len, &tp->ucopy.locked_list);
+#endif
+
do {
struct sk_buff *skb;
u32 offset;
@@ -1273,6 +1283,10 @@ int tcp_recvmsg(struct kiocb *iocb, stru
} else
sk_wait_data(sk, &timeo);
+#ifdef CONFIG_NET_DMA
+ tp->ucopy.wakeup = 0;
+#endif
+
if (user_recv) {
int chunk;
@@ -1328,13 +1342,39 @@ do_prequeue:
}
if (!(flags & MSG_TRUNC)) {
- err = skb_copy_datagram_iovec(skb, offset,
- msg->msg_iov, used);
- if (err) {
- /* Exception. Bailout! */
- if (!copied)
- copied = -EFAULT;
- break;
+#ifdef CONFIG_NET_DMA
+ if (!tp->ucopy.dma_chan && tp->ucopy.locked_list)
+ tp->ucopy.dma_chan = get_softnet_dma();
+
+ if (tp->ucopy.dma_chan) {
+ tp->ucopy.dma_cookie = dma_skb_copy_datagram_iovec(
+ tp->ucopy.dma_chan, skb, offset,
+ msg->msg_iov, used,
+ tp->ucopy.locked_list);
+
+ if (tp->ucopy.dma_cookie < 0) {
+
+ printk(KERN_ALERT "dma_cookie < 0\n");
+
+ /* Exception. Bailout! */
+ if (!copied)
+ copied = -EFAULT;
+ break;
+ }
+ if ((offset + used) == skb->len)
+ copied_early = 1;
+
+ } else
+#endif
+ {
+ err = skb_copy_datagram_iovec(skb, offset,
+ msg->msg_iov, used);
+ if (err) {
+ /* Exception. Bailout! */
+ if (!copied)
+ copied = -EFAULT;
+ break;
+ }
}
}
@@ -1354,15 +1394,33 @@ skip_copy:
if (skb->h.th->fin)
goto found_fin_ok;
- if (!(flags & MSG_PEEK))
- sk_eat_skb(sk, skb);
+ if (!(flags & MSG_PEEK)) {
+ if (!copied_early)
+ sk_eat_skb(sk, skb);
+#ifdef CONFIG_NET_DMA
+ else {
+ __skb_unlink(skb, &sk->sk_receive_queue);
+ __skb_queue_tail(&sk->sk_async_wait_queue, skb);
+ copied_early = 0;
+ }
+#endif
+ }
continue;
found_fin_ok:
/* Process the FIN. */
++*seq;
- if (!(flags & MSG_PEEK))
- sk_eat_skb(sk, skb);
+ if (!(flags & MSG_PEEK)) {
+ if (!copied_early)
+ sk_eat_skb(sk, skb);
+#ifdef CONFIG_NET_DMA
+ else {
+ __skb_unlink(skb, &sk->sk_receive_queue);
+ __skb_queue_tail(&sk->sk_async_wait_queue, skb);
+ copied_early = 0;
+ }
+#endif
+ }
break;
} while (len > 0);
@@ -1385,6 +1443,34 @@ skip_copy:
tp->ucopy.len = 0;
}
+#ifdef CONFIG_NET_DMA
+ if (tp->ucopy.dma_chan) {
+ struct sk_buff *skb;
+ dma_cookie_t done, used;
+
+ dma_async_memcpy_issue_pending(tp->ucopy.dma_chan);
+
+ while (dma_async_memcpy_complete(tp->ucopy.dma_chan,
+ tp->ucopy.dma_cookie, &done,
+ &used) == DMA_IN_PROGRESS) {
+ /* do partial cleanup of sk_async_wait_queue */
+ while ((skb = skb_peek(&sk->sk_async_wait_queue)) &&
+ (dma_async_is_complete(skb->dma_cookie, done,
+ used) == DMA_SUCCESS)) {
+ __skb_dequeue(&sk->sk_async_wait_queue);
+ kfree_skb(skb);
+ }
+ }
+
+ /* Safe to free early-copied skbs now */
+ __skb_queue_purge(&sk->sk_async_wait_queue);
+ dma_unlock_iovec_pages(tp->ucopy.locked_list);
+ dma_chan_put(tp->ucopy.dma_chan);
+ tp->ucopy.dma_chan = NULL;
+ tp->ucopy.locked_list = NULL;
+ }
+#endif
+
/* According to UNIX98, msg_name/msg_namelen are ignored
* on connected socket. I was just happy when found this 8) --ANK
*/
@@ -1652,6 +1738,9 @@ int tcp_disconnect(struct sock *sk, int
__skb_queue_purge(&sk->sk_receive_queue);
sk_stream_writequeue_purge(sk);
__skb_queue_purge(&tp->out_of_order_queue);
+#ifdef CONFIG_NET_DMA
+ __skb_queue_purge(&sk->sk_async_wait_queue);
+#endif
inet->dport = 0;
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 7625eaf..9b6290d 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -71,6 +71,9 @@
#include <net/inet_common.h>
#include <linux/ipsec.h>
#include <asm/unaligned.h>
+#ifdef CONFIG_NET_DMA
+#include <net/netdma.h>
+#endif
int sysctl_tcp_timestamps = 1;
int sysctl_tcp_window_scaling = 1;
@@ -3901,14 +3904,23 @@ int tcp_rcv_established(struct sock *sk,
}
} else {
int eaten = 0;
+ int copied_early = 0;
- if (tp->ucopy.task == current &&
- tp->copied_seq == tp->rcv_nxt &&
- len - tcp_header_len <= tp->ucopy.len &&
- sock_owned_by_user(sk)) {
- __set_current_state(TASK_RUNNING);
+ if (tp->copied_seq == tp->rcv_nxt &&
+ len - tcp_header_len <= tp->ucopy.len) {
+#ifdef CONFIG_NET_DMA
+ if (dma_async_try_early_copy(sk, skb, tcp_header_len)) {
+ copied_early = 1;
+ eaten = 1;
+ }
+#endif
+ if (tp->ucopy.task == current && sock_owned_by_user(sk) && !copied_early) {
+ __set_current_state(TASK_RUNNING);
- if (!tcp_copy_to_iovec(sk, skb, tcp_header_len)) {
+ if (!tcp_copy_to_iovec(sk, skb, tcp_header_len))
+ eaten = 1;
+ }
+ if (eaten) {
/* Predicted packet is in window by definition.
* seq == rcv_nxt and rcv_wup <= rcv_nxt.
* Hence, check seq<=rcv_wup reduces to:
@@ -3924,8 +3936,9 @@ int tcp_rcv_established(struct sock *sk,
__skb_pull(skb, tcp_header_len);
tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
NET_INC_STATS_BH(LINUX_MIB_TCPHPHITSTOUSER);
- eaten = 1;
}
+ if (copied_early)
+ tcp_cleanup_rbuf(sk, skb->len);
}
if (!eaten) {
if (tcp_checksum_complete_user(sk, skb))
@@ -3966,6 +3979,11 @@ int tcp_rcv_established(struct sock *sk,
__tcp_ack_snd_check(sk, 0);
no_ack:
+#ifdef CONFIG_NET_DMA
+ if (copied_early)
+ __skb_queue_tail(&sk->sk_async_wait_queue, skb);
+ else
+#endif
if (eaten)
__kfree_skb(skb);
else
@@ -4049,6 +4067,52 @@ discard:
return 0;
}
+#ifdef CONFIG_NET_DMA
+int dma_async_try_early_copy(struct sock *sk, struct sk_buff *skb, int hlen)
+{
+ struct tcp_sock *tp = tcp_sk(sk);
+ int chunk = skb->len - hlen;
+ int dma_cookie;
+ int copied_early = 0;
+
+ if (tp->ucopy.wakeup)
+ goto out;
+
+ if (!tp->ucopy.dma_chan && tp->ucopy.locked_list)
+ tp->ucopy.dma_chan = get_softnet_dma();
+
+ if (tp->ucopy.dma_chan && skb->ip_summed == CHECKSUM_UNNECESSARY) {
+
+ dma_cookie = dma_skb_copy_datagram_iovec(tp->ucopy.dma_chan,
+ skb, hlen, tp->ucopy.iov, chunk, tp->ucopy.locked_list);
+
+ if (dma_cookie < 0)
+ goto out;
+
+ tp->ucopy.dma_cookie = dma_cookie;
+ copied_early = 1;
+
+ tp->ucopy.len -= chunk;
+ tp->copied_seq += chunk;
+ tcp_rcv_space_adjust(sk);
+
+ if ((tp->ucopy.len == 0) ||
+ (tcp_flag_word(skb->h.th) & TCP_FLAG_PSH) ||
+ (atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1))) {
+ tp->ucopy.wakeup = 1;
+ sk->sk_data_ready(sk, 0);
+ }
+ } else if (chunk > 0) {
+ tp->ucopy.wakeup = 1;
+ sk->sk_data_ready(sk, 0);
+ }
+out:
+ return copied_early;
+}
+
+EXPORT_SYMBOL(dma_async_try_early_copy);
+#endif /* CONFIG_NET_DMA */
+
static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb,
struct tcphdr *th, unsigned len)
{
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index 4eb903d..fecc022 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -71,6 +71,9 @@
#include <net/inet_common.h>
#include <net/timewait_sock.h>
#include <net/xfrm.h>
+#ifdef CONFIG_NET_DMA
+#include <net/netdma.h>
+#endif
#include <linux/inet.h>
#include <linux/ipv6.h>
@@ -1091,8 +1094,18 @@ process:
bh_lock_sock(sk);
ret = 0;
if (!sock_owned_by_user(sk)) {
- if (!tcp_prequeue(sk, skb))
+#ifdef CONFIG_NET_DMA
+ struct tcp_sock *tp = tcp_sk(sk);
+ if (!tp->ucopy.dma_chan && tp->ucopy.locked_list)
+ tp->ucopy.dma_chan = get_softnet_dma();
+ if (tp->ucopy.dma_chan)
+ ret = tcp_v4_do_rcv(sk, skb);
+ else
+#endif
+ {
+ if (!tcp_prequeue(sk, skb))
ret = tcp_v4_do_rcv(sk, skb);
+ }
} else
sk_add_backlog(sk, skb);
bh_unlock_sock(sk);
@@ -1292,6 +1305,11 @@ int tcp_v4_destroy_sock(struct sock *sk)
/* Cleans up our, hopefully empty, out_of_order_queue. */
__skb_queue_purge(&tp->out_of_order_queue);
+#ifdef CONFIG_NET_DMA
+ /* Cleans up our sk_async_wait_queue */
+ __skb_queue_purge(&sk->sk_async_wait_queue);
+#endif
+
/* Clean prequeue, it must be empty really */
__skb_queue_purge(&tp->ucopy.prequeue);
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index af6a0c6..acf798c 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -1218,8 +1218,16 @@ process:
bh_lock_sock(sk);
ret = 0;
if (!sock_owned_by_user(sk)) {
- if (!tcp_prequeue(sk, skb))
- ret = tcp_v6_do_rcv(sk, skb);
+#ifdef CONFIG_NET_DMA
+ struct tcp_sock *tp = tcp_sk(sk);
+ if (tp->ucopy.dma_chan)
+ ret = tcp_v6_do_rcv(sk, skb);
+ else
+#endif
+ {
+ if (!tcp_prequeue(sk, skb))
+ ret = tcp_v6_do_rcv(sk, skb);
+ }
} else
sk_add_backlog(sk, skb);
bh_unlock_sock(sk);
^ permalink raw reply related
* Re: [PATCH 0/8] Intel I/O Acceleration Technology (I/OAT)
From: Jeff Garzik @ 2006-03-03 22:27 UTC (permalink / raw)
To: Chris Leech; +Cc: linux-kernel, netdev
In-Reply-To: <20060303214036.11908.10499.stgit@gitlost.site>
Chris Leech wrote:
> This patch series is the first full release of the Intel(R) I/O
> Acceleration Technology (I/OAT) for Linux. It includes an in kernel API
> for offloading memory copies to hardware, a driver for the I/OAT DMA memcpy
> engine, and changes to the TCP stack to offload copies of received
> networking data to application space.
>
> These changes apply to DaveM's net-2.6.17 tree as of commit
> 2bd84a93d8bb7192ad8c23ef41008502be1cb603 ([IRDA]: TOIM3232 dongle support)
>
> They are available to pull from
> git://198.78.49.142/~cleech/linux-2.6 ioat-2.6.17
>
> There are 8 patches in the series:
> 1) The memcpy offload APIs and class code
> 2) The Intel I/OAT DMA driver (ioatdma)
Patch #2 didn't make it. Too big for the list?
Jeff
^ permalink raw reply
* Re: [PATCH 0/8] Intel I/O Acceleration Technology (I/OAT)
From: Chris Leech @ 2006-03-03 22:39 UTC (permalink / raw)
To: Jeff Garzik; +Cc: linux-kernel, netdev
In-Reply-To: <4408C2CA.5010909@garzik.org>
[-- Attachment #1: Type: text/plain, Size: 186 bytes --]
> Patch #2 didn't make it. Too big for the list?
Could be, it's the largest of the series. I've attached the gziped
patch. I can try and split this up for the future.
- Chris
[-- Attachment #2: 02-ioatdma_driver.diff.gz --]
[-- Type: application/x-gzip, Size: 10272 bytes --]
^ permalink raw reply
* Re: [PATCH 0/8] Intel I/O Acceleration Technology (I/OAT)
From: Jeff Garzik @ 2006-03-03 22:45 UTC (permalink / raw)
To: chris.leech; +Cc: linux-kernel, netdev
In-Reply-To: <41b516cb0603031439n13e4df4cg8e5b21b606d2b4b8@mail.gmail.com>
Chris Leech wrote:
>>Patch #2 didn't make it. Too big for the list?
>
>
> Could be, it's the largest of the series. I've attached the gziped
> patch. I can try and split this up for the future.
Well, for huge hunks of new code, it sometimes gets silly to split it up.
Once its not in a "reply to email" reviewable form, gzip or URL-to-patch
work just fine.
Jeff
^ permalink raw reply
* Re: [PATCH 0/5] Infiniband: connection abstraction
From: Roland Dreier @ 2006-03-03 22:53 UTC (permalink / raw)
To: Sean Hefty; +Cc: netdev, linux-kernel, openib-general
In-Reply-To: <ORSMSX4011XvpFVjCRG00000015@orsmsx401.amr.corp.intel.com>
Sean> I can resubmit the patches if necessary.
Yes, can you please send out the patches again, with descriptive
subjects for each patch and with the ip_dev_find() re-export split
into its own patch? That would make it easier for me to pull into my
git tree.
Thanks,
Roland
^ permalink raw reply
* Re: [PATCH 0/8] Intel I/O Acceleration Technology (I/OAT)
From: Kumar Gala @ 2006-03-03 22:58 UTC (permalink / raw)
To: Chris Leech; +Cc: linux-kernel, netdev
In-Reply-To: <20060303214036.11908.10499.stgit@gitlost.site>
On Mar 3, 2006, at 3:40 PM, Chris Leech wrote:
> This patch series is the first full release of the Intel(R) I/O
> Acceleration Technology (I/OAT) for Linux. It includes an in
> kernel API
> for offloading memory copies to hardware, a driver for the I/OAT
> DMA memcpy
> engine, and changes to the TCP stack to offload copies of received
> networking data to application space.
>
> These changes apply to DaveM's net-2.6.17 tree as of commit
> 2bd84a93d8bb7192ad8c23ef41008502be1cb603 ([IRDA]: TOIM3232 dongle
> support)
>
> They are available to pull from
> git://198.78.49.142/~cleech/linux-2.6 ioat-2.6.17
>
> There are 8 patches in the series:
> 1) The memcpy offload APIs and class code
> 2) The Intel I/OAT DMA driver (ioatdma)
> 3) Core networking code to setup networking as a DMA memcpy client
> 4) Utility functions for sk_buff to iovec offloaded copy
> 5) Structure changes needed for TCP receive offload
> 6) Rename cleanup_rbuf to tcp_cleanup_rbuf
> 7) Add a sysctl to tune the minimum offloaded I/O size for TCP
> 8) The main TCP receive offload changes
How does this relate to Dan William's ADMA work?
http://marc.theaimsgroup.com/?t=113892936300001&r=1&w=2
- kumar
^ permalink raw reply
* Re: [PATCH 0/8] Intel I/O Acceleration Technology (I/OAT)
From: Chris Leech @ 2006-03-03 23:32 UTC (permalink / raw)
To: Kumar Gala; +Cc: linux-kernel, netdev
In-Reply-To: <54FF0817-23ED-47F1-8234-FD3079B3E403@kernel.crashing.org>
On 3/3/06, Kumar Gala <galak@kernel.crashing.org> wrote:
>
> How does this relate to Dan William's ADMA work?
I only became aware of Dan's ADMA work when he posted it last month,
and so far have not made any attempts to merge the I/OAT code with it.
Moving forward, combining these interfaces certainly seems like the
right way to go. I particularly like ADMA's handling of operations
other than just a copy (memset, compare, XOR, CRC).
Chris
^ permalink raw reply
* Re: [PATCH 1/8] [I/OAT] DMA memcpy subsystem
From: David S. Miller @ 2006-03-04 1:40 UTC (permalink / raw)
To: christopher.leech; +Cc: linux-kernel, netdev
In-Reply-To: <20060303214220.11908.75517.stgit@gitlost.site>
From: Chris Leech <christopher.leech@intel.com>
Date: Fri, 03 Mar 2006 13:42:20 -0800
> +static spinlock_t dma_list_lock;
Please use DEFINE_SPINLOCK().
> +static void dma_chan_free_rcu(struct rcu_head *rcu) {
Newline before the brace please.
> +static void dma_async_device_cleanup(struct kref *kref) {
Newline before the brace please.
> +struct dma_chan_percpu
> +{
Left brace on the same line as "struct dma_chan_percpu" please.
> +struct dma_chan
> +{
Similarly.
Otherwise this patch looks mostly ok.
^ permalink raw reply
* Re: [PATCH RESEND] net: Move destructor from neigh->ops to neigh_params
From: David S. Miller @ 2006-03-04 1:58 UTC (permalink / raw)
To: rdreier; +Cc: netdev, openib-general
In-Reply-To: <adabqxq1rdh.fsf@cisco.com>
From: Roland Dreier <rdreier@cisco.com>
Date: Wed, 01 Feb 2006 17:28:10 -0800
> Dave, if you want to merge this directly, that's fine. Or I'm fine
> with merging this through the IB tree if you'd prefer (if you want me
> to do that, let me know if you think it's 2.6.16 material).
Applied to net-2.6.17, thanks for being so incredibly patient :)
^ permalink raw reply
* Re: [PATCH] RTPROT addition in rtnetlink.h
From: David S. Miller @ 2006-03-04 2:01 UTC (permalink / raw)
To: alpt; +Cc: linux-net, netdev
In-Reply-To: <20060304015852.GA26183@nihil>
From: Alpt <alpt@freaknet.org>
Date: Sat, 4 Mar 2006 02:58:52 +0100
> This patch defines the Netsukuku protocol (RTPROT_NTK) by the number
> 15 in rtnetlink.h.
>
> The Netsukuku daemon is using the same number to mark its routes,
> you can see it here:
> http://hinezumilabs.org/cgi-bin/viewcvs.cgi/netsukuku/src/krnl_route.h?rev=HEAD&content-type=text/vnd.viewcvs-markup
Applied.
Please, next time, post patches to netdev@vger.kernel.org and provide
proper "Signed-off-by: " lines.
Thanks.
^ permalink raw reply
* Re: [2.6 patch] schedule eepro100.c for removal
From: Jeff Garzik @ 2006-03-04 2:24 UTC (permalink / raw)
To: Adrian Bunk; +Cc: linux-kernel, netdev
In-Reply-To: <20060214152226.GK10701@stusta.de>
Adrian Bunk wrote:
> Signed-off-by: Adrian Bunk <bunk@stusta.de>
>
> ---
>
> This patch was already sent on:
> - 4 Feb 2006
>
> Documentation/feature-removal-schedule.txt | 6 ++++++
> drivers/net/eepro100.c | 1 +
> 2 files changed, 7 insertions(+)
>
> --- linux-2.6.15-mm4-full/Documentation/feature-removal-schedule.txt.old 2006-01-18 08:38:57.000000000 +0100
> +++ linux-2.6.15-mm4-full/Documentation/feature-removal-schedule.txt 2006-01-18 08:39:59.000000000 +0100
> @@ -164,0 +165,6 @@
> +---------------------------
> +
> +What: eepro100 network driver
> +When: August 2006
> +Why: replaced by the e100 driver
> +Who: Adrian Bunk <bunk@stusta.de>
ACK, provided you change the year to 2007
> --- linux-2.6.16-rc1-mm5-full/drivers/net/eepro100.c.old 2006-02-03 23:37:55.000000000 +0100
> +++ linux-2.6.16-rc1-mm5-full/drivers/net/eepro100.c 2006-02-03 23:39:10.000000000 +0100
> @@ -2391,6 +2391,7 @@ static int __init eepro100_init_module(v
> #ifdef MODULE
> printk(version);
> #endif
> + printk(KERN_WARNING "eepro100 will be removed soon, please use the e100 driver\n");
> return pci_module_init(&eepro100_driver);
NAK
^ permalink raw reply
* Re: [2.6 patch] schedule eepro100.c for removal
From: Adrian Bunk @ 2006-03-04 11:04 UTC (permalink / raw)
To: Jeff Garzik; +Cc: linux-kernel, netdev
In-Reply-To: <4408FA6E.3040808@pobox.com>
On Fri, Mar 03, 2006 at 09:24:46PM -0500, Jeff Garzik wrote:
> Adrian Bunk wrote:
> >Signed-off-by: Adrian Bunk <bunk@stusta.de>
> >
> >---
> >
> >This patch was already sent on:
> >- 4 Feb 2006
> >
> > Documentation/feature-removal-schedule.txt | 6 ++++++
> > drivers/net/eepro100.c | 1 +
> > 2 files changed, 7 insertions(+)
> >
> >--- linux-2.6.15-mm4-full/Documentation/feature-removal-schedule.txt.old
> >2006-01-18 08:38:57.000000000 +0100
> >+++ linux-2.6.15-mm4-full/Documentation/feature-removal-schedule.txt
> >2006-01-18 08:39:59.000000000 +0100
> >@@ -164,0 +165,6 @@
> >+---------------------------
> >+
> >+What: eepro100 network driver
> >+When: August 2006
> >+Why: replaced by the e100 driver
> >+Who: Adrian Bunk <bunk@stusta.de>
>
> ACK, provided you change the year to 2007
>
> >--- linux-2.6.16-rc1-mm5-full/drivers/net/eepro100.c.old 2006-02-03
> >23:37:55.000000000 +0100
> >+++ linux-2.6.16-rc1-mm5-full/drivers/net/eepro100.c 2006-02-03
> >23:39:10.000000000 +0100
> >@@ -2391,6 +2391,7 @@ static int __init eepro100_init_module(v
> > #ifdef MODULE
> > printk(version);
> > #endif
> >+ printk(KERN_WARNING "eepro100 will be removed soon, please use the
> >e100 driver\n");
> > return pci_module_init(&eepro100_driver);
>
> NAK
Other people wanted this part of the patch...
It seems the feature-removal-schedule.txt part with a changed date is
already part of git-netdev-all.patch in recent -mm kernels.
Do you still want me to resend it?
cu
Adrian
--
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon Seed
^ permalink raw reply
* Re: [PATCH 7/8] [I/OAT] Add a sysctl for tuning the I/OAT offloaded I/O threshold
From: Alexey Dobriyan @ 2006-03-04 11:22 UTC (permalink / raw)
To: Chris Leech; +Cc: linux-kernel, netdev
In-Reply-To: <20060303214234.11908.99495.stgit@gitlost.site>
On Fri, Mar 03, 2006 at 01:42:34PM -0800, Chris Leech wrote:
> Any socket recv of less than this ammount will not be offloaded
There is no documentation update.
^ permalink raw reply
* Re: [PATCH 0/8] Intel I/O Acceleration Technology (I/OAT)
From: Evgeniy Polyakov @ 2006-03-04 11:35 UTC (permalink / raw)
To: chris.leech; +Cc: Jeff Garzik, linux-kernel, netdev
In-Reply-To: <41b516cb0603031439n13e4df4cg8e5b21b606d2b4b8@mail.gmail.com>
On Fri, Mar 03, 2006 at 02:39:22PM -0800, Chris Leech (christopher.leech@intel.com) wrote:
> > Patch #2 didn't make it. Too big for the list?
>
> Could be, it's the largest of the series. I've attached the gziped
> patch. I can try and split this up for the future.
How can owner of cb_chan->common.device_node be removed?
It looks like that channels are only allocated (without proper error path)
and queued into device->common.channels list in
enumerate_dma_channels() in PCI probe callback and no removing at all, only lockless access.
PCI remove callback only calls dma_async_device_unregister() where only
channel's clients are removed.
> - Chris
--
Evgeniy Polyakov
^ permalink raw reply
* [2.6 patch] drivers/net/wireless/ipw2200.c: make ipw_qos_current_mode() static
From: Adrian Bunk @ 2006-03-04 12:14 UTC (permalink / raw)
To: yi.zhu, jketreno; +Cc: linux-kernel, linville, netdev
This patch makes the needlessly global function ipw_qos_current_mode()
static.
Signed-off-by: Adrian Bunk <bunk@stusta.de>
--- linux-2.6.16-rc5-mm2-full/drivers/net/wireless/ipw2200.c.old 2006-03-03 17:49:37.000000000 +0100
+++ linux-2.6.16-rc5-mm2-full/drivers/net/wireless/ipw2200.c 2006-03-03 17:50:00.000000000 +0100
@@ -6566,7 +6566,7 @@
* get the modulation type of the current network or
* the card current mode
*/
-u8 ipw_qos_current_mode(struct ipw_priv * priv)
+static u8 ipw_qos_current_mode(struct ipw_priv * priv)
{
u8 mode = 0;
^ permalink raw reply
* GigE on PowerMac G5
From: Andreas Schwab @ 2006-03-04 14:53 UTC (permalink / raw)
To: netdev; +Cc: linuxppc64-dev
[Sorry for duplicate posting, I've used the wrong list address.]
I suppose the NIC in the PowerMac G5 can do GigE, yet when plugged into a
GB switch it is only willing to talk 100MB with it. Any idea why? Kernel
is 2.6.16-rc5-git2.
# lsprop /proc/device-tree/ht@0,f2000000/pci@6/ethernet@f
name "ethernet"
linux,phandle ff9c53d8
interrupt-parent ff9779b0
gbit-phy
assigned-addresses 82047810 00000000 80400000 00000000 00200000
82047830 00000000 80300000 00000000 00100000
local-mac-address 00 0a 95 ba b8 70 .....p
stats 00000000 00000000 00000000 00000000 00000000
reg 00047800 00000000 00000000 00000000 00000000
02047810 00000000 00000000 00000000 00020000
02047830 00000000 00000000 00000000 00010000
max-frame-size 000005ee (1518)
address-bits 00000030 (48)
built-in
compatible "K2-GMAC"
category "net"
removable "network"
network-type "ethernet"
device_type "network"
fast-back-to-back
devsel-speed 00000002
max-latency 00000040 (64)
min-grant 00000040 (64)
interrupts 00000029 00000001
class-code 00020000 (131072)
revision-id 00000000
device-id 0000004c (76)
vendor-id 0000106b (4203)
# ethtool eth0
Settings for eth0:
Supported ports: [ TP MII ]
Supported link modes: 10baseT/Half 10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Half 1000baseT/Full
Supports auto-negotiation: Yes
Advertised link modes: 10baseT/Half 10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Half 1000baseT/Full
Advertised auto-negotiation: No
Speed: 100Mb/s
Duplex: Full
Port: MII
PHYAD: 0
Transceiver: external
Auto-negotiation: on
Supports Wake-on: g
Wake-on: d
Current message level: 0x00000007 (7)
Link detected: yes
Andreas.
--
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux Products GmbH, Maxfeldstraße 5, 90409 Nürnberg, Germany
PGP key fingerprint = 58CA 54C7 6D53 942B 1756 01D3 44D5 214B 8276 4ED5
"And now for something completely different."
^ permalink raw reply
* [RFC: 2.6 patch] let NET_CLS_ACT no longer depend on EXPERIMENTAL
From: Adrian Bunk @ 2006-03-04 16:07 UTC (permalink / raw)
To: netdev; +Cc: linux-kernel
This option should IMHO no longer depend on EXPERIMENTAL.
Signed-off-by: Adrian Bunk <bunk@stusta.de>
---
This patch was already sent on:
- 12 Feb 2006
--- linux-2.6.16-rc2-mm1-full/net/sched/Kconfig.old 2006-02-12 02:21:30.000000000 +0100
+++ linux-2.6.16-rc2-mm1-full/net/sched/Kconfig 2006-02-12 02:21:58.000000000 +0100
@@ -434,7 +434,6 @@
config NET_CLS_ACT
bool "Actions"
- depends on EXPERIMENTAL
select NET_ESTIMATOR
---help---
Say Y here if you want to use traffic control actions. Actions
^ permalink raw reply
* Re: [PATCH 8/8] [I/OAT] TCP recv offload to I/OAT
From: Pavel Machek @ 2006-03-04 16:39 UTC (permalink / raw)
To: Chris Leech; +Cc: linux-kernel, netdev
In-Reply-To: <20060303214236.11908.98881.stgit@gitlost.site>
Hi!
> --- a/net/ipv4/tcp.c
> +++ b/net/ipv4/tcp.c
> @@ -262,6 +262,9 @@
> #include <net/tcp.h>
> #include <net/xfrm.h>
> #include <net/ip.h>
> +#ifdef CONFIG_NET_DMA
> +#include <net/netdma.h>
> +#endif
>
Remove the ifdefs, move them inside .h if needed.
> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
> index 7625eaf..9b6290d 100644
> --- a/net/ipv4/tcp_input.c
> +++ b/net/ipv4/tcp_input.c
> @@ -71,6 +71,9 @@
> #include <net/inet_common.h>
> #include <linux/ipsec.h>
> #include <asm/unaligned.h>
> +#ifdef CONFIG_NET_DMA
> +#include <net/netdma.h>
> +#endif
Here, too.
> +#ifdef CONFIG_NET_DMA
> + if (copied_early)
> + __skb_queue_tail(&sk->sk_async_wait_queue, skb);
> + else
> +#endif
> if (eaten)
> __kfree_skb(skb);
> else
Could you #define copied_early to 0 and avoid ifdefs?
> @@ -1091,8 +1094,18 @@ process:
> bh_lock_sock(sk);
> ret = 0;
> if (!sock_owned_by_user(sk)) {
> - if (!tcp_prequeue(sk, skb))
> +#ifdef CONFIG_NET_DMA
> + struct tcp_sock *tp = tcp_sk(sk);
> + if (!tp->ucopy.dma_chan && tp->ucopy.locked_list)
> + tp->ucopy.dma_chan = get_softnet_dma();
> + if (tp->ucopy.dma_chan)
> + ret = tcp_v4_do_rcv(sk, skb);
> + else
> +#endif
> + {
> + if (!tcp_prequeue(sk, skb))
> ret = tcp_v4_do_rcv(sk, skb);
> + }
> } else
Wrong indentation...
Pavel
--
Thanks, Sharp!
^ permalink raw reply
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox