* Re: Patch to add M3Pnet Driver
[not found] ` <20080523023623.GA17651@linux-mips.org>
@ 2008-05-23 16:42 ` Kevin D. Kissell
2008-06-17 13:40 ` Kevin D. Kissell
0 siblings, 1 reply; 2+ messages in thread
From: Kevin D. Kissell @ 2008-05-23 16:42 UTC (permalink / raw)
To: jeff, netdev; +Cc: Ralf Baechle
[-- Attachment #1: Type: text/plain, Size: 1845 bytes --]
Guysm
Yet another network driver is probably not what the world needs most,
but I had to write one for MIPS Linux this month that was a little
different from any of the others. It's a driver that impersonates
an ethernet interface whilst actually passing packets in memory between
different processors running different OS images in a shared memory
environment, something that actually happens in embedded systems.
I call it M3P (pronounced "mep"), for MIPS Memory Message Passing,
but if I didn't think it could be adapted to non-MIPS based and indeed
heterogeneous systems, I wouldn't trouble you with it. I note that
the "mipsnet" driver, which is limited in its function to running
on a MIPS simulator, made it in, so I figured I'd at least ask if
it's something that could be absorbed into the network driver mass,
the better for other folks to improve on it - there are a lot of
enhancements I'd like to make, some of which are described in the
comments, but it's not clear that I'll have time to make them any
time soon. A familiar refrain, I suppose. Anyway, the patch is
attached. See also Ralf's ack below.
Regards,
Kevin K.
Ralf Baechle wrote:
> On Fri, May 23, 2008 at 12:45:45AM +0200, Kevin D. Kissell wrote:
>> From: "Kevin D. Kissell" <kevink@mips.com>
>> Date: Fri, 23 May 2008 00:45:45 +0200
>> To: Linux MIPS Org <linux-mips@linux-mips.org>
>> Subject: Patch to add M3Pnet Driver
>> Content-Type: multipart/mixed;
>> boundary="------------050707080700070300010401"
>>
>> As per previous messages, here's 2 of 2
>
> This is largely a network driver patch with minor arch bits thrown in so
> this one will have to go via jeff@garzik.org and netdev@vger.kernel.org.
> The MIPS bits are looking good though so feel free to add my
>
> Acked-by: Ralf Baechle <ralf@linux-mips.org>
>
> line for them.
>
> Ralf
[-- Attachment #2: 0002-Add-support-for-M3P-MIPS-Memory-Message-Passing-ps.patch --]
[-- Type: text/x-patch, Size: 15837 bytes --]
>From f98ebc82b0273d30e0a29943a1553a60c7684200 Mon Sep 17 00:00:00 2001
From: Kevin D. Kissell <kevink@mips.com>
Date: Thu, 22 May 2008 16:34:42 +0200
Subject: [PATCH] Add support for M3P (MIPS Memory Message Passing) pseudo-ethernet
device for communication between e.g. AP and RP in APRP Linux.
Signed-off-by: Kevin D. Kissell <kevink@mips.com>
---
arch/mips/mips-boards/generic/Makefile | 3 +-
arch/mips/mips-boards/generic/platform.c | 34 +++
drivers/net/Kconfig | 8 +
drivers/net/Makefile | 1 +
drivers/net/m3pnet.c | 431 ++++++++++++++++++++++++++++++
drivers/net/m3pnet.h | 47 ++++
6 files changed, 523 insertions(+), 1 deletions(-)
create mode 100644 arch/mips/mips-boards/generic/platform.c
create mode 100644 drivers/net/m3pnet.c
create mode 100644 drivers/net/m3pnet.h
diff --git a/arch/mips/mips-boards/generic/Makefile b/arch/mips/mips-boards/generic/Makefile
index b31d8df..6f0d02b 100644
--- a/arch/mips/mips-boards/generic/Makefile
+++ b/arch/mips/mips-boards/generic/Makefile
@@ -19,10 +19,11 @@
#
obj-y := reset.o display.o init.o memory.o \
- cmdline.o time.o
+ cmdline.o time.o platform.o
obj-$(CONFIG_EARLY_PRINTK) += console.o
obj-$(CONFIG_PCI) += pci.o
obj-$(CONFIG_KGDB) += gdb_hook.o
+obj-$(CONFIG_MIPS_M3P_NET) += platform.o
EXTRA_CFLAGS += -Werror
diff --git a/arch/mips/mips-boards/generic/platform.c b/arch/mips/mips-boards/generic/platform.c
new file mode 100644
index 0000000..1d56094
--- /dev/null
+++ b/arch/mips/mips-boards/generic/platform.c
@@ -0,0 +1,34 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License. See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 2008 by Kevin D. Kissell, MIPS Technologies Inc.
+ */
+#include <linux/init.h>
+#include <linux/if_ether.h>
+#include <linux/kernel.h>
+#include <linux/platform_device.h>
+
+static char m3pnet_string[] = "m3pnet";
+
+static struct platform_device eth3_device = {
+ .name = m3pnet_string,
+ .id = 0,
+};
+
+/*
+ * Create a platform device for the M3P driver
+ */
+static int __init m3pnet_devinit(void)
+{
+ int err;
+
+ err = platform_device_register(ð3_device);
+ if (err)
+ printk(KERN_ERR "%s: registration failed\n", m3pnet_string);
+
+ return err;
+}
+
+device_initcall(m3pnet_devinit);
diff --git a/drivers/net/Kconfig b/drivers/net/Kconfig
index 28d284a..b38e2c5 100644
--- a/drivers/net/Kconfig
+++ b/drivers/net/Kconfig
@@ -496,6 +496,14 @@ config MIPS_SIM_NET
emulated by the MIPS Simulator.
If you are not using a MIPSsim or are unsure, say N.
+config MIPS_M3P_NET
+ tristate "MIPS Memory Message Passing device"
+ depends on MIPS
+ help
+ The M3PNET device is a pseudo-network device intended
+ for communication between real or virtual processors
+ running distinct OS images.
+ If you are unsure, say N.
config SGI_O2MACE_ETH
tristate "SGI O2 MACE Fast Ethernet support"
depends on SGI_IP32=y
diff --git a/drivers/net/Makefile b/drivers/net/Makefile
index 931264b..140be5b 100644
--- a/drivers/net/Makefile
+++ b/drivers/net/Makefile
@@ -192,6 +192,7 @@ obj-$(CONFIG_EQUALIZER) += eql.o
obj-$(CONFIG_MIPS_JAZZ_SONIC) += jazzsonic.o
obj-$(CONFIG_MIPS_AU1X00_ENET) += au1000_eth.o
obj-$(CONFIG_MIPS_SIM_NET) += mipsnet.o
+obj-$(CONFIG_MIPS_M3P_NET) += m3pnet.o
obj-$(CONFIG_SGI_IOC3_ETH) += ioc3-eth.o
obj-$(CONFIG_DECLANCE) += declance.o
obj-$(CONFIG_ATARILANCE) += atarilance.o
diff --git a/drivers/net/m3pnet.c b/drivers/net/m3pnet.c
new file mode 100644
index 0000000..ab01d42
--- /dev/null
+++ b/drivers/net/m3pnet.c
@@ -0,0 +1,431 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License. See the file "COPYING" in the main directory of this archive
+ * for more details.
+ */
+
+/*
+ * Memory based message passing network "device". Written for
+ * MIPS systems, but probably more broadly usable. Initial prototype
+ * supports only a single point-to-point instance, with arbitrary
+ * "ethernet addresses" at either end. It would probably be good
+ * to evolve M3PNet to support multipoint, with data structure
+ * indices trivially computed based on bits of the "NIC address".
+ */
+
+#include <linux/init.h>
+#include <linux/io.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/platform_device.h>
+#include <asm/spinlock.h>
+
+#include "m3pnet.h"
+
+#define M3PNET_VERSION "2008-05-22"
+
+static char m3pnet_string[] = "m3pnet";
+
+#define N_XBUF 8
+#define N_RBUF 8
+#define M3P_MTU 1500
+
+#define M3PSIZE (((N_XBUF+N_RBUF) * M3P_MTU)\
+ + ((N_XBUF+N_RBUF)*sizeof(struct m3p_buffer))\
+ + sizeof(struct m3p_channel))
+
+#define M3P_SEND 1
+#define M3P_RECV 2
+
+/* Arbitrary unique type for VPE shared area */
+
+#define M3P_AREA_TYPE 0xEC0BA8EA
+
+struct m3p_private {
+ int xblocked;
+ int rblocked;
+};
+
+
+#ifdef M3P_DEBUG
+void pktdump(unsigned char *pkt, int len)
+{
+ printk("m3p pkt: ");
+ while (len-- > 1)
+ printk("%02x", *pkt++);
+ printk("%02x\n", *pkt);
+}
+#endif /* M3P_DEBUG */
+
+static unsigned char *m3pnet_ringinit(struct m3p_buffer *pbuf,
+ unsigned char *pool, int nbufs)
+{
+ int i;
+
+ for (i = 0; i < nbufs; i++) {
+ if (i == 0)
+ pbuf[i].prev = &pbuf[nbufs - 1];
+ else
+ pbuf[i].prev = &pbuf[i - 1];
+ if (i == (nbufs - 1))
+ pbuf[i].next = &pbuf[0];
+ else
+ pbuf[i].next = &pbuf[i+1];
+ pbuf[i].size = M3P_MTU;
+ pbuf[i].length = 0;
+ pbuf[i].content = pool;
+ pool += M3P_MTU;
+ }
+ return(pool);
+}
+
+static void m3pnet_bufsetup(struct net_device *netdev)
+{
+ struct m3p_channel *pchan = (struct m3p_channel *)netdev->base_addr;
+ struct m3p_buffer *pbuf = (struct m3p_buffer *)(netdev->base_addr
+ + sizeof(struct m3p_channel));
+ unsigned char *pcontent = (unsigned char *)(netdev->base_addr
+ + sizeof(struct m3p_channel)
+ + ((N_XBUF + N_RBUF) * sizeof(struct m3p_buffer)));
+
+ spin_lock_init(&pchan->lock);
+ pchan->send_head = pchan->send_tail = &pbuf[0];
+ pcontent = m3pnet_ringinit(pbuf, pcontent, N_XBUF);
+ /* Advance pointer to base of receive buffer array */
+ pbuf = &pbuf[N_XBUF];
+ pchan->recv_head = pchan->recv_tail = &pbuf[0];
+ pcontent = m3pnet_ringinit(pbuf, pcontent, N_RBUF);
+}
+
+
+#define DEFAULT_VPE 1
+
+static void m3p_ipi(struct net_device *dev, int type)
+{
+ /*
+ * This needs to be abstracted through platform code
+ * to use whatever hardware IPI mechanism is available.
+ * Here we're assuming a MIPS APRP environment with
+ * cross-VPE interrupt support.
+ */
+ int vpe_send_interrupt(int v, int t);
+
+ /*
+ * Only a single M3P device is supported in this version
+ * of the driver. With multiple devices, we would
+ * store a VPE ID as part of the per-net_device
+ * private data and use it here.
+ */
+
+ (void)vpe_send_interrupt(DEFAULT_VPE, type);
+}
+
+static inline ssize_t m3pnet_send(struct net_device *dev,
+ struct sk_buff *skb)
+{
+ struct m3p_channel *pchan = (struct m3p_channel *)dev->base_addr;
+ unsigned int flags;
+ int kickstart = 0;
+
+ memcpy(pchan->send_tail->content, skb->data, skb->len);
+#ifdef M3P_DEBUG
+ printk("m3pnet_send (%d)", skb->len);
+ pktdump(pchan->send_tail->content, skb->len);
+#endif /* M3P_DEBUG */
+ spin_lock_irqsave(&pchan->lock, flags);
+ pchan->send_tail->length = skb->len;
+ if (pchan->send_tail->next->length == 0) {
+ /*
+ * If the next buffer is free, advance in ring
+ * and interrupt the other guy if necessary.
+ */
+ if (pchan->send_head == pchan->send_tail)
+ kickstart = 1;
+ pchan->send_tail = pchan->send_tail->next;
+ }
+ spin_unlock_irqrestore(&pchan->lock, flags);
+ if (kickstart)
+ m3p_ipi(dev, M3P_SEND);
+
+ dev->stats.tx_packets++;
+ dev->stats.tx_bytes += skb->len;
+ dev_kfree_skb(skb);
+
+ return 0;
+}
+
+static int m3pnet_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+ struct m3p_channel *pchan = (struct m3p_channel *)dev->base_addr;
+ struct m3p_private *private = (struct m3p_private *)dev->priv;
+
+ if (!m3p_can_send(pchan)) {
+ printk(KERN_ERR "M3P send channel blockage, skb discarded\n");
+ dev_kfree_skb(skb);
+ return 1;
+ }
+
+ m3pnet_send(dev, skb);
+
+ if (!m3p_can_send(pchan)) {
+ netif_stop_queue(dev);
+ private->xblocked = 1;
+ }
+
+ return 0;
+}
+
+static inline ssize_t m3pnet_receive(struct net_device *dev)
+{
+ struct sk_buff *skb;
+ struct m3p_channel *pchan = (struct m3p_channel *)dev->base_addr;
+ size_t len = pchan->recv_head->length;
+ unsigned char *cp;
+ unsigned int flags;
+ int kickstart = 0;
+
+ skb = dev_alloc_skb(len + NET_IP_ALIGN);
+ if (!skb) {
+ dev->stats.rx_dropped++;
+ return -ENOMEM;
+ }
+
+ skb_reserve(skb, NET_IP_ALIGN);
+ cp = skb_put(skb, len);
+ memcpy(cp, pchan->recv_head->content, len);
+#ifdef M3P_DEBUG
+ printk("m3pnet_recieve (%d)", len);
+ pktdump(cp, len);
+#endif /* M3P_DEBUG */
+ spin_lock_irqsave(&pchan->lock, flags);
+ /* It may be worth considering a "low water mark" for this */
+ if (pchan->recv_head->prev == pchan->recv_tail
+ && pchan->recv_tail->length != 0)
+ kickstart = 1;
+ pchan->recv_head = pchan->recv_head->next;
+ pchan->recv_head->prev->length = 0;
+ spin_unlock_irqrestore(&pchan->lock, flags);
+ if (kickstart)
+ m3p_ipi(dev, M3P_RECV);
+ skb->protocol = eth_type_trans(skb, dev);
+ skb->ip_summed = CHECKSUM_UNNECESSARY;
+
+ netif_rx(skb);
+
+ dev->stats.rx_packets++;
+ dev->stats.rx_bytes += len;
+
+ return len;
+}
+
+static irqreturn_t m3pnet_interrupt(int irq, void *dev_id)
+{
+ struct net_device *dev = dev_id;
+ struct m3p_channel *pchan = (struct m3p_channel *)dev->base_addr;
+ struct m3p_private *private = (struct m3p_private *)dev->priv;
+
+ irqreturn_t retval = IRQ_NONE;
+
+ if (irq == dev->irq) {
+
+ if (m3p_can_send(pchan) && private->xblocked) {
+ private->xblocked = 0;
+ netif_wake_queue(dev);
+ retval = IRQ_HANDLED;
+ }
+
+ while (m3p_can_receive(pchan)) {
+ m3pnet_receive(dev);
+ retval = IRQ_HANDLED;
+ }
+
+ } else {
+ printk(KERN_INFO "%s: %s(): irq %d for unknown device\n",
+ dev->name, __FUNCTION__, irq);
+ }
+ return retval;
+}
+
+static int m3pnet_open(struct net_device *dev)
+{
+ int err;
+ struct m3p_channel *pchan = (struct m3p_channel *)dev->base_addr;
+ struct m3p_private *private = (struct m3p_private *)dev->priv;
+
+ err = request_irq(dev->irq, &m3pnet_interrupt,
+ IRQF_SHARED, dev->name, (void *) dev);
+
+ if (err) {
+ kfree((char *)dev->base_addr);
+ return err;
+ }
+
+ /*
+ * Initial design did buffer initialization
+ * exclusively in probe routine, after memory
+ * allocation. In practice, if an APRP application
+ * fails or terminates with the buffers in an
+ * invalid state, we want have easy assurance
+ * that the next run will see a sensible array.
+ */
+
+ m3pnet_bufsetup(dev);
+
+ if (m3p_can_send(pchan))
+ netif_start_queue(dev);
+ else
+ private->xblocked = 1;
+
+ return 0;
+}
+
+static int m3pnet_close(struct net_device *dev)
+{
+ netif_stop_queue(dev);
+ free_irq(dev->irq, dev);
+ return 0;
+}
+
+static void m3pnet_set_mclist(struct net_device *dev)
+{
+}
+
+static int __init m3pnet_probe(struct platform_device *dev)
+{
+ struct net_device *netdev;
+ struct m3p_private *private;
+ int err;
+ int irq;
+ int arch_get_xcpu_irq(void);
+
+ netdev = alloc_etherdev(sizeof(struct m3p_private));
+ if (!netdev) {
+ err = -ENOMEM;
+ goto out;
+ }
+
+ SET_NETDEV_DEV(netdev, &dev->dev);
+
+ netdev->open = m3pnet_open;
+ netdev->stop = m3pnet_close;
+ netdev->hard_start_xmit = m3pnet_xmit;
+ netdev->set_multicast_list = m3pnet_set_mclist;
+
+ private = netdev->priv;
+ private->xblocked = 0;
+ private->rblocked = 0;
+
+ /*
+ * Allocate and format send and receive buffer rings on channel
+ */
+ netdev->base_addr = (unsigned long)kmalloc(M3PSIZE, GFP_KERNEL);
+ if (netdev->base_addr == 0) {
+ err = -ENOMEM;
+ goto out_free_netdev;
+ }
+
+ m3pnet_bufsetup(netdev);
+
+ /*
+ * Initial prototype is cross-VPE on pre-GIC Malta FPGA,
+ * but future versions should support CMP. Push responsibility
+ * for identifying IRQs to the generic architecture support code,
+ * which will eventually get it from platform code.
+ */
+
+ irq = arch_get_xcpu_irq();
+ if (irq >= 0)
+ netdev->irq = irq;
+ else {
+ err = irq;
+ goto out_free_buffers;
+ }
+
+ /*
+ * Lacking any better mechanism to allocate a MAC address we use a
+ * random one ...
+ */
+ random_ether_addr(netdev->dev_addr);
+
+ err = register_netdev(netdev);
+ if (err) {
+ printk(KERN_ERR "M3PNet: failed to register netdev.\n");
+ goto out_free_buffers;
+ }
+ /*
+ * Not clear how this driver would ever be used in a non-APRP
+ * configuration, but let's be sparing in our assumptions.
+ */
+#ifdef CONFIG_MIPS_VPE_LOADER
+ else {
+ void mips_publish_vpe_area(int type, void *ptr);
+
+ mips_publish_vpe_area(M3P_AREA_TYPE, (void *)netdev->base_addr);
+ }
+#endif /* CONFIG_MIPS_VPE_LOADER */
+
+#ifdef M3P_DEBUG
+ printk(KERN_INFO "M3PNet driver probe returning 0 (success)\n");
+#endif /* M3P_DEBUG */
+ return 0;
+
+out_free_buffers:
+ kfree((char *)netdev->base_addr);
+
+out_free_netdev:
+ free_netdev(netdev);
+
+out:
+#ifdef M3P_DEBUG
+ printk(KERN_INFO "M3PNet driver probe returning %d\n", err);
+#endif /* M3P_DEBUG */
+ return err;
+}
+
+static int __exit m3pnet_device_remove(struct platform_device *device)
+{
+ struct net_device *dev = platform_get_drvdata(device);
+
+ unregister_netdev(dev);
+ kfree((char *)dev->base_addr);
+ free_netdev(dev);
+ platform_set_drvdata(device, NULL);
+
+ return 0;
+}
+
+static struct platform_driver m3pnet_driver = {
+ .probe = m3pnet_probe,
+ .remove = __devexit_p(m3pnet_device_remove),
+ .driver = {
+ .name = m3pnet_string,
+ .owner = THIS_MODULE,
+ }
+};
+
+static int __init m3pnet_init_module(void)
+{
+ int err;
+
+ printk(KERN_INFO "M3PNet Network driver. Version: %s. "
+ "(c)2008 MIPS Technologies, Inc.\n", M3PNET_VERSION);
+
+ err = platform_driver_register(&m3pnet_driver);
+ if (err)
+ printk(KERN_ERR "Driver registration failed\n");
+
+ return err;
+}
+
+static void __exit m3pnet_exit_module(void)
+{
+ platform_driver_unregister(&m3pnet_driver);
+}
+
+module_init(m3pnet_init_module);
+module_exit(m3pnet_exit_module);
+MODULE_AUTHOR("Kevin D. Kissell");
+MODULE_DESCRIPTION("M3P Memory Message Pasing Driver");
+MODULE_LICENSE("GPL");
diff --git a/drivers/net/m3pnet.h b/drivers/net/m3pnet.h
new file mode 100644
index 0000000..de2e861
--- /dev/null
+++ b/drivers/net/m3pnet.h
@@ -0,0 +1,47 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License. See the file "COPYING" in the main directory of this archive
+ * for more details.
+ */
+#ifndef __M3PNET_H
+#define __M3PNET_H
+
+/*
+ * M3PNET passes messages in shared non-necessarily-hardware-coherent
+ * memory between real or virtual processors. Messages are copied from
+ * Linux (or other OS) network buffers into low-level M3P buffers,
+ * which are managed as a ring.
+ */
+
+struct m3p_buffer {
+ struct m3p_buffer *prev;
+ struct m3p_buffer *next;
+ int size;
+ void *content;
+ int length;
+};
+
+/*
+ * Channel structure here is implicitly point-to-point.
+ * The "RP"side of the prototype simply reverses the
+ * senses of "send" and "recv". A more general and
+ * extensible scheme would be preferable, with bits
+ * of the "NIC" address used to select the queues.
+ */
+
+struct m3p_channel {
+ unsigned int status;
+ spinlock_t lock;
+ struct m3p_buffer *send_head;
+ struct m3p_buffer *send_tail;
+ struct m3p_buffer *recv_head;
+ struct m3p_buffer *recv_tail;
+};
+
+#define m3p_can_receive(c) \
+ (c->recv_head->length != 0)
+
+#define m3p_can_send(c) \
+ (c->send_tail->length == 0)
+
+#endif /* __MIPSNET_H */
--
1.5.3.3
^ permalink raw reply related [flat|nested] 2+ messages in thread