Patchwork [30/39] i2400m/USB: read transactions from the USB device

login
register
mail settings
Submitter Inaky Perez-Gonzalez
Date Nov. 24, 2008, 9:50 p.m.
Message ID <25f917a5fc866ea66ca164df80f56aee691c59a7.1227562829.git.inaky@linux.intel.com>
Download mbox | patch
Permalink /patch/10515/
State Changes Requested
Delegated to: David Miller
Headers show

Comments

Inaky Perez-Gonzalez - Nov. 24, 2008, 9:50 p.m.
This code implements reading from the USB device; it uses a simple
adaptive algorithm so it doesn't preallocate huge buffers that then
are underused. When a transaction is read, it is passed to the generic
driver for processing and delivery.

All is done in a kthread sitting on a loop. We need a thread context
to run USB autopm functions and to process the received data (can get
to be heavy in processing time). It is dedicated to a single task, so
it has less overhead than a workqueue.

Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
---
 drivers/net/wimax/i2400m/usb-rx.c |  417 +++++++++++++++++++++++++++++++++++++
 1 files changed, 417 insertions(+), 0 deletions(-)
 create mode 100644 drivers/net/wimax/i2400m/usb-rx.c

Patch

diff --git a/drivers/net/wimax/i2400m/usb-rx.c b/drivers/net/wimax/i2400m/usb-rx.c
new file mode 100644
index 0000000..074cc1f
--- /dev/null
+++ b/drivers/net/wimax/i2400m/usb-rx.c
@@ -0,0 +1,417 @@ 
+/*
+ * Intel Wireless WiMAX Connection 2400m
+ * USB RX handling
+ *
+ *
+ * Copyright (C) 2007-2008 Intel Corporation. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ *   * Redistributions of source code must retain the above copyright
+ *     notice, this list of conditions and the following disclaimer.
+ *   * Redistributions in binary form must reproduce the above copyright
+ *     notice, this list of conditions and the following disclaimer in
+ *     the documentation and/or other materials provided with the
+ *     distribution.
+ *   * Neither the name of Intel Corporation nor the names of its
+ *     contributors may be used to endorse or promote products derived
+ *     from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ *
+ * Intel Corporation <linux-wimax@intel.com>
+ * Yanir Lubetkin <yanirx.lubetkin@intel.com>
+ *  - Initial implementation
+ * Inaky Perez-Gonzalez <inaky.perez-gonzalez@intel.com>
+ *  - Use skb_clone(), break up processing in chunks
+ *  - Split transport/device specific
+ *  - Make buffer size dynamic to exert less memory pressure
+ *
+ *
+ * This handles the RX path on USB.
+ *
+ * When a notification is received that says 'there is RX data ready',
+ * we call i2400mu_rx_kick(); that wakes up the RX kthread, which
+ * reads a buffer from USB and passes it to i2400m_rx() in the generic
+ * handling code. The RX buffer has an specific format that is
+ * described in rx.c.
+ *
+ * We use a kernel thread in a loop because:
+ *
+ *  - we want to be able to call the USB power management get/put
+ *    functions (blocking) before each transaction.
+ *
+ *  - We might get a lot of notifications and we don't want to submit
+ *    a zillion reads; by serializing, we are throttling.
+ *
+ *  - RX data processing can get heavy enough so that it is not
+ *    appropiate for doing it in the USB callback; thus we run it in a
+ *    process context.
+ *
+ * We provide a read buffer of an arbitrary size (short of a page); if
+ * the callback reports -EOVERFLOW, it means it was too small, so we
+ * just double the size and retry (being careful to append, as
+ * sometimes the device provided some data). Every now and then we
+ * check if the average packet size is smaller than the current packet
+ * size and if so, we halve it. At the end, the size of the
+ * preallocated buffer should be following the average received
+ * transaction size, adapting dynamically to it.
+ *
+ * ROADMAP
+ *
+ * i2400mu_rx_kick()		   Called from notif.c when we get a
+ *   			           'data ready' notification
+ * i2400mu_rxd()                   Kernel RX daemon
+ *   i2400mu_rx()                  Receive USB data
+ *   i2400m_rx()                   Send data to generic i2400m RX handling
+ *
+ * i2400mu_rx_setup()              called from i2400mu_bus_dev_start()
+ *
+ * i2400mu_rx_release()            called from i2400mu_bus_dev_stop()
+ */
+#include <linux/workqueue.h>
+#include <linux/usb.h>
+#include "i2400m-usb.h"
+
+
+#define D_SUBMODULE rx
+#include "usb-debug-levels.h"
+
+/*
+ * Dynamic RX size
+ *
+ * We can't let the rx_size be a multiple of 512 bytes (the RX
+ * endpoint's max packet size). On some USB host controllers (we
+ * haven't been able to fully characterize which), if the device is
+ * about to send (for example) X bytes and we only post a buffer to
+ * receive n*512, it will fail to mark that as babble (so that
+ * i2400mu_rx() [case -EOVERFLOW] can resize the buffer and get the
+ * rest).
+ *
+ * So on growing or shrinking, if it is a multiple of the
+ * maxpacketsize, we remove some (instead of incresing some, so in a
+ * buddy allocator we try to waste less space).
+ *
+ * Note we also need a hook for this on i2400mu_rx() -- when we do the
+ * first read, we are sure we won't hit this spot because
+ * i240mm->rx_size has been set properly. However, if we have to
+ * double because of -EOVERFLOW, when we launch the read to get the
+ * rest of the data, we *have* to make sure that also is not a
+ * multiple of the max_pkt_size.
+ */
+
+static
+size_t i2400mu_rx_size_grow(struct i2400mu *i2400mu)
+{
+	struct device *dev = &i2400mu->usb_iface->dev;
+	size_t rx_size;
+	const size_t max_pkt_size = 512;
+
+	rx_size = 2 * i2400mu->rx_size;
+	if (rx_size % max_pkt_size == 0) {
+		rx_size -= 8;
+		d_printf(1, dev,
+			 "RX: expected size grew to %zu [adjusted -8] "
+			 "from %zu\n",
+			 rx_size, i2400mu->rx_size);
+	} else
+		d_printf(1, dev,
+			 "RX: expected size grew to %zu from %zu\n",
+			 rx_size, i2400mu->rx_size);
+	return rx_size;
+}
+
+
+static
+void i2400mu_rx_size_maybe_shrink(struct i2400mu *i2400mu)
+{
+	const size_t max_pkt_size = 512;
+	struct device *dev = &i2400mu->usb_iface->dev;
+
+	if (unlikely(i2400mu->rx_size_cnt >= 100
+		     && i2400mu->rx_size_auto_shrink)) {
+		size_t avg_rx_size =
+			i2400mu->rx_size_acc / i2400mu->rx_size_cnt;
+		size_t new_rx_size = i2400mu->rx_size / 2;
+		if (avg_rx_size < new_rx_size) {
+			if (new_rx_size % max_pkt_size == 0) {
+				new_rx_size -= 8;
+				d_printf(1, dev,
+					 "RX: expected size shrank to %zu "
+					 "[adjusted -8] from %zu\n",
+					 new_rx_size, i2400mu->rx_size);
+			} else
+				d_printf(1, dev,
+					 "RX: expected size shrank to %zu "
+					 "from %zu\n",
+					 new_rx_size, i2400mu->rx_size);
+			i2400mu->rx_size = new_rx_size;
+			i2400mu->rx_size_cnt = 0;
+			i2400mu->rx_size_acc = i2400mu->rx_size;
+		}
+	}
+}
+
+/*
+ * Receive a message with payloads from the USB bus into an skb
+ *
+ * @i2400mu: USB device descriptor
+ * @rx_skb: skb where to place the received message
+ *
+ * Deals with all the USB-specifics of receiving, dynamically
+ * increasing the buffer size if so needed. Returns the payload in the
+ * skb, ready to process. On a zero-length packet, we retry.
+ *
+ * On soft USB errors, we retry (until they become too frequent and
+ * then are promoted to hard); on hard USB errors, we reset the
+ * device. On other errors (skb realloacation, we just drop it and
+ * hope for the next invocation to solve it).
+ *
+ * Returns: pointer to the skb if ok, ERR_PTR on error.
+ *   NOTE: this function might realloc the skb (if it is too small),
+ *   so always update with the one returned.
+ *   ERR_PTR() is < 0 on error.
+ */
+static
+struct sk_buff *i2400mu_rx(struct i2400mu *i2400mu, struct sk_buff *rx_skb)
+{
+	int result = 0;
+	struct device *dev = &i2400mu->usb_iface->dev;
+	int usb_pipe, read_size, rx_size, do_autopm;
+	struct usb_endpoint_descriptor *epd;
+	const size_t max_pkt_size = 512;
+
+	d_fnstart(4, dev, "(i2400mu %p)\n", i2400mu);
+	do_autopm = atomic_read(&i2400mu->do_autopm);
+	result = do_autopm ?
+		usb_autopm_get_interface(i2400mu->usb_iface) : 0;
+	if (result < 0) {
+		dev_err(dev, "RX: can't get autopm: %d\n", result);
+		do_autopm = 0;
+	}
+	epd = usb_get_epd(i2400mu->usb_iface, I2400MU_EP_BULK_IN);
+	usb_pipe = usb_rcvbulkpipe(i2400mu->usb_dev, epd->bEndpointAddress);
+retry:
+	rx_size = skb_end_pointer(rx_skb) - rx_skb->data - rx_skb->len;
+	if (unlikely(rx_size % max_pkt_size == 0)) {
+		rx_size -= 8;
+		d_printf(1, dev, "RX: rx_size adapted to %d [-8]\n", rx_size);
+	}
+	result = usb_bulk_msg(
+		i2400mu->usb_dev, usb_pipe, rx_skb->data + rx_skb->len,
+		rx_size, &read_size, HZ);
+	usb_mark_last_busy(i2400mu->usb_dev);
+	switch (result) {
+	case 0:
+		if (read_size == 0)
+			goto retry;	/* ZLP, just resubmit */
+		skb_put(rx_skb, read_size);
+		break;
+	case -EINVAL:			/* while removing driver */
+	case -ENODEV:			/* dev disconnect ... */
+	case -ENOENT:			/* just ignore it */
+	case -ESHUTDOWN:
+	case -ECONNRESET:
+		break;
+	case -EOVERFLOW: {		/* too small, reallocate */
+		struct sk_buff *new_skb;
+		rx_size = i2400mu_rx_size_grow(i2400mu);
+		if (rx_size <= (1 << 16))	/* cap it */
+			i2400mu->rx_size = rx_size;
+		else if (printk_ratelimit()) {
+			dev_err(dev, "BUG? rx_size up to %d\n", rx_size);
+			result = -EINVAL;
+			goto out;
+		}
+		skb_put(rx_skb, read_size);
+		new_skb = skb_copy_expand(rx_skb, 0, rx_size - rx_skb->len,
+					  GFP_KERNEL);
+		if (new_skb == NULL) {
+			if (printk_ratelimit())
+				dev_err(dev, "RX: Can't reallocate skb to %d; "
+					"RX dropped\n", rx_size);
+			kfree(rx_skb);
+			result = 0;
+			goto out;	/* drop it...*/
+		}
+		kfree_skb(rx_skb);
+		rx_skb = new_skb;
+		i2400mu->rx_size_cnt = 0;
+		i2400mu->rx_size_acc = i2400mu->rx_size;
+		d_printf(1, dev, "RX: size changed to %d, received %d, "
+			 "copied %d, capacity %ld\n",
+			 rx_size, read_size, rx_skb->len,
+			 (long) (skb_end_pointer(new_skb) - new_skb->head));
+		goto retry;
+	}
+		/* In most cases, it happens due to the hardware scheduling a
+		 * read when there was no data - unfortunately, we have no way
+		 * to tell this timeout from a USB timeout. So we just ignore
+		 * it. */
+	case -ETIMEDOUT:
+		dev_err(dev, "RX: timeout: %d\n", result);
+		result = 0;
+		break;
+	default:			/* Any error */
+		if (edc_inc(&i2400mu->urb_edc,
+			    EDC_MAX_ERRORS, EDC_ERROR_TIMEFRAME))
+			goto error_reset;
+		dev_err(dev, "RX: error receiving URB: %d, retrying\n", result);
+		goto retry;
+	}
+out:
+	if (do_autopm)
+		usb_autopm_put_interface(i2400mu->usb_iface);
+	d_fnend(4, dev, "(i2400mu %p) = %p\n", i2400mu, rx_skb);
+	return rx_skb;
+
+error_reset:
+	dev_err(dev, "RX: maximum errors in URB exceeded; "
+		"resetting device\n");
+	usb_queue_reset_device(i2400mu->usb_iface);
+	rx_skb = ERR_PTR(result);
+	goto out;
+}
+
+
+/*
+ * Kernel thread for USB reception of data
+ *
+ * This thread waits for a kick; once kicked, it will allocate an skb
+ * and receive a single message to it from USB (using
+ * i2400mu_rx()). Once received, it is passed to the generic i2400m RX
+ * code for processing.
+ *
+ * When done processing, it runs some dirty statistics to verify if
+ * the last 100 messages received were smaller than half of the
+ * current RX buffer size. In that case, the RX buffer size is
+ * halved. This will helps lowering the pressure on the memory
+ * allocator.
+ *
+ * Hard errors force the thread to exit.
+ */
+static
+int i2400mu_rxd(void *_i2400mu)
+{
+	int result = 0;
+	struct i2400mu *i2400mu = _i2400mu;
+	struct i2400m *i2400m = &i2400mu->i2400m;
+	struct device *dev = &i2400mu->usb_iface->dev;
+	struct net_device *net_dev = i2400m->wimax_dev.net_dev;
+	size_t pending;
+	int rx_size;
+	struct sk_buff *rx_skb;
+
+	d_fnstart(4, dev, "(i2400mu %p)\n", i2400mu);
+	while (1) {
+		d_printf(2, dev, "TX: waiting for messages\n");
+		pending = 0;
+		wait_event_interruptible(
+			i2400mu->rx_wq,
+			(kthread_should_stop()	/* check this first! */
+			 || (pending = atomic_read(&i2400mu->rx_pending_count)))
+			);
+		if (kthread_should_stop())
+			break;
+		if (pending == 0)
+			continue;
+		rx_size = i2400mu->rx_size;
+		d_printf(2, dev, "RX: reading up to %d bytes\n", rx_size);
+		rx_skb = __netdev_alloc_skb(net_dev, rx_size, GFP_KERNEL);
+		if (rx_skb == NULL) {
+			dev_err(dev, "RX: can't allocate skb [%d bytes]\n",
+				rx_size);
+			msleep(50);	/* give it some time? */
+			continue;
+		}
+
+		/* Receive the message with the payloads */
+		rx_skb = i2400mu_rx(i2400mu, rx_skb);
+		result = PTR_ERR(rx_skb);
+		if (IS_ERR(rx_skb))
+			goto out;
+		atomic_dec(&i2400mu->rx_pending_count);
+		if (rx_skb->len == 0) {	/* some ignorable condition */
+			kfree_skb(rx_skb);
+			continue;
+		}
+
+		/* Deliver the message to the generic i2400m code */
+		i2400mu->rx_size_cnt++;
+		i2400mu->rx_size_acc += rx_skb->len;
+		result = i2400m_rx(i2400m, rx_skb);
+		if (result == -EIO
+		    && edc_inc(&i2400mu->urb_edc,
+			       EDC_MAX_ERRORS, EDC_ERROR_TIMEFRAME)) {
+			goto error_reset;
+		}
+
+		/* Maybe adjust RX buffer size */
+		i2400mu_rx_size_maybe_shrink(i2400mu);
+	}
+	result = 0;
+out:
+	d_fnend(4, dev, "(i2400mu %p) = %d\n", i2400mu, result);
+	return result;
+
+error_reset:
+	dev_err(dev, "RX: maximum errors in received buffer exceeded; "
+		"resetting device\n");
+	usb_queue_reset_device(i2400mu->usb_iface);
+	goto out;
+}
+
+
+/*
+ * Start reading from the device
+ *
+ * @i2400m: device instance
+ *
+ * Notify the RX thread that there is data pending.
+ */
+void i2400mu_rx_kick(struct i2400mu *i2400mu)
+{
+	struct i2400m *i2400m = &i2400mu->i2400m;
+	struct device *dev = &i2400mu->usb_iface->dev;
+
+	d_fnstart(3, dev, "(i2400mu %p)\n", i2400m);
+	atomic_inc(&i2400mu->rx_pending_count);
+	wake_up_all(&i2400mu->rx_wq);
+	d_fnend(3, dev, "(i2400m %p) = void\n", i2400m);
+}
+
+
+int i2400mu_rx_setup(struct i2400mu *i2400mu)
+{
+	int result = 0;
+	struct i2400m *i2400m = &i2400mu->i2400m;
+	struct device *dev = &i2400mu->usb_iface->dev;
+	struct wimax_dev *wimax_dev = &i2400m->wimax_dev;
+
+	i2400mu->rx_kthread = kthread_run(i2400mu_rxd, i2400mu, "%s-rx",
+					  wimax_dev->name);
+	if (IS_ERR(i2400mu->rx_kthread)) {
+		result = PTR_ERR(i2400mu->rx_kthread);
+		dev_err(dev, "RX: cannot start thread: %d\n", result);
+	}
+	return result;
+}
+
+void i2400mu_rx_release(struct i2400mu *i2400mu)
+{
+	kthread_stop(i2400mu->rx_kthread);
+}
+