| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <4e777ed10706150859r3b6e1617o3b31e5e24a30e6c9@mail.gmail.com>
Date: Fri, 15 Jun 2007 11:59:43 -0400
From: "Zhu Han" <schumi.han@...il.com>
To: "Kieran Mansley" <kmansley@...arflare.com>
Cc: xen-devel@...ts.xensource.com, netdev@...r.kernel.org,
herbert@...dor.apana.org.au
Subject: Re: [Xen-devel] [PATCH 4/4] [Net] Support accelerated network plugin modules
Hi, Kieran,
I'm just wonder why you try to acquire the lock and increase the
hooks_usecount each time when you use the hook routine. Is there any
generic ways to synchronze the code path using hook routines and
netfront_accelerator_unloaded, considering you can synchronize the
tx/rx data path easily.
On 6/15/07, Kieran Mansley <kmansley@...arflare.com> wrote:
> Frontend net driver acceleration
>
> diff -r cd3ade350f3f drivers/xen/netfront/netfront.c
> --- a/drivers/xen/netfront/netfront.c Thu Jun 14 15:04:32 2007 +0100
> +++ b/drivers/xen/netfront/netfront.c Fri Jun 15 09:34:41 2007 +0100
> @@ -3,6 +3,7 @@
> *
> * Copyright (c) 2002-2005, K A Fraser
> * Copyright (c) 2005, XenSource Ltd
> + * Copyright (C) 2007 Solarflare Communications, Inc.
> *
> * This program is free software; you can redistribute it and/or
> * modify it under the terms of the GNU General Public License version 2
> @@ -47,6 +48,7 @@
> #include <linux/if_ether.h>
> #include <linux/io.h>
> #include <linux/moduleparam.h>
> +#include <linux/list.h>
> #include <net/sock.h>
> #include <net/pkt_sched.h>
> #include <net/arp.h>
> @@ -73,6 +75,8 @@ struct netfront_cb {
> };
>
> #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
> +
> +#include "netfront.h"
>
> /*
> * Mutually-exclusive module options to select receive data path:
> @@ -144,57 +148,6 @@ static inline int netif_needs_gso(struct
>
> #define GRANT_INVALID_REF 0
>
> -#define NET_TX_RING_SIZE __RING_SIZE((struct netif_tx_sring *)0, PAGE_SIZE)
> -#define NET_RX_RING_SIZE __RING_SIZE((struct netif_rx_sring *)0, PAGE_SIZE)
> -
> -struct netfront_info {
> - struct list_head list;
> - struct net_device *netdev;
> -
> - struct net_device_stats stats;
> -
> - struct netif_tx_front_ring tx;
> - struct netif_rx_front_ring rx;
> -
> - spinlock_t tx_lock;
> - spinlock_t rx_lock;
> -
> - unsigned int irq;
> - unsigned int copying_receiver;
> - unsigned int carrier;
> -
> - /* Receive-ring batched refills. */
> -#define RX_MIN_TARGET 8
> -#define RX_DFL_MIN_TARGET 64
> -#define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
> - unsigned rx_min_target, rx_max_target, rx_target;
> - struct sk_buff_head rx_batch;
> -
> - struct timer_list rx_refill_timer;
> -
> - /*
> - * {tx,rx}_skbs store outstanding skbuffs. The first entry in tx_skbs
> - * is an index into a chain of free entries.
> - */
> - struct sk_buff *tx_skbs[NET_TX_RING_SIZE+1];
> - struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
> -
> -#define TX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
> - grant_ref_t gref_tx_head;
> - grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1];
> - grant_ref_t gref_rx_head;
> - grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
> -
> - struct xenbus_device *xbdev;
> - int tx_ring_ref;
> - int rx_ring_ref;
> - u8 mac[ETH_ALEN];
> -
> - unsigned long rx_pfn_array[NET_RX_RING_SIZE];
> - struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
> - struct mmu_update rx_mmu[NET_RX_RING_SIZE];
> -};
> -
> struct netfront_rx_info {
> struct netif_rx_response rx;
> struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
> @@ -278,6 +231,369 @@ static void xennet_sysfs_delif(struct ne
> #define xennet_sysfs_delif(dev) do { } while(0)
> #endif
>
> +/*
> + * List of all netfront accelerator plugin modules available. Each
> + * list entry is of type struct netfront_accelerator.
> + */
> +static struct list_head accelerators_list;
> +/*
> + * Lock to protect access to accelerators_list, and also used to
> + * protect the hooks_usecount field in struct netfront_accelerator
> + * against concurrent access
> + */
> +static spinlock_t accelerators_lock;
> +
> +/*
> + * Safely remove the accelerator function hooks from a netfront state.
> + * Must only be called when there are no current users of the hooks.
> + */
> +static void accelerator_remove_hooks(struct netfront_accelerator *accelerator)
> +{
> + struct netfront_accel_vif_state *vif_state;
> +
> + list_for_each_entry( vif_state,
> + &accelerator->vif_states,
> + link ) {
> + /* Make sure there are no data path operations going on */
> + netif_poll_disable(vif_state->np->netdev);
> + netif_tx_lock_bh(vif_state->np->netdev);
> +
> + /*
> + * Remove the hooks, but leave the vif_state on the
> + * accelerator's list as that signifies this vif is
> + * interested in using that accelerator if it becomes
> + * available again
> + */
> + vif_state->hooks = NULL;
> +
> + netif_tx_unlock_bh(vif_state->np->netdev);
> + netif_poll_enable(vif_state->np->netdev);
> + }
> +
> + accelerator->hooks = NULL;
> +
> + /* Signal that all users of hooks are done */
> + up(&accelerator->exit_semaphore);
> +}
> +
> +
> +/*
> + * Compare a frontend description string against an accelerator to see
> + * if they match. Would ultimately be nice to replace the string with
> + * a unique numeric identifier for each accelerator.
> + */
> +static int match_accelerator(const char *frontend,
> + struct netfront_accelerator *accelerator)
> +{
> + return strcmp(frontend, accelerator->frontend) == 0;
> +}
> +
> +
> +/*
> + * Add a frontend vif to the list of vifs that is using a netfront
> + * accelerator plugin module.
> + */
> +static void add_accelerator_vif(struct netfront_accelerator *accelerator,
> + struct netfront_info *np,
> + struct xenbus_device *dev)
> +{
> + np->accelerator = accelerator;
> + np->accel_vif_state.np = np;
> + np->accel_vif_state.dev = dev;
> +
> + list_add(&np->accel_vif_state.link, &accelerator->vif_states);
> +}
> +
> +/*
> + * Initialise the netfront state of an accelerator plugin module.
> + */
> +static int init_accelerator(const char *frontend,
> + struct netfront_accelerator **result)
> +{
> + struct netfront_accelerator *accelerator =
> + kmalloc(sizeof(struct netfront_accelerator), GFP_KERNEL);
> + int frontend_len;
> +
> + if ( !accelerator ) {
> + DPRINTK("%s: no memory for accelerator", __FUNCTION__);
> + return -ENOMEM;
> + }
> +
> + frontend_len = strlen(frontend) + 1;
> + accelerator->frontend = kmalloc(frontend_len, GFP_KERNEL);
> + if ( !accelerator->frontend ) {
> + DPRINTK("%s: no memory for accelerator", __FUNCTION__);
> + kfree(accelerator);
> + return -ENOMEM;
> + }
> + strlcpy(accelerator->frontend, frontend, frontend_len);
> +
> + INIT_LIST_HEAD(&accelerator->vif_states);
> +
> + accelerator->hooks = NULL;
> + accelerator->hooks_usecount = 0;
> +
> + list_add(&accelerator->link, &accelerators_list);
> +
> + *result = accelerator;
> +
> + return 0;
> +}
> +
> +/*
> + * Modify the hooks stored in the per-vif state to match that in the
> + * netfront accelerator's state.
> + */
> +static void
> +accelerator_set_vif_state_hooks(struct netfront_accel_vif_state *vif_state)
> +{
> + /* Make sure there are no data path operations going on */
> + netif_poll_disable(vif_state->np->netdev);
> + netif_tx_lock_bh(vif_state->np->netdev);
> +
> + vif_state->hooks = vif_state->np->accelerator->hooks;
> +
> + netif_tx_unlock_bh(vif_state->np->netdev);
> + netif_poll_enable(vif_state->np->netdev);
> +}
> +
> +
> +/*
> + * Request that a particular netfront accelerator plugin is loaded.
> + * Usually called as a result of the vif configuration specifying
> + * which one to use.
> + */
> +static int netfront_load_accelerator(struct netfront_info *np,
> + struct xenbus_device *dev,
> + const char *frontend)
> +{
> + struct netfront_accelerator *accelerator;
> + int rc;
> + unsigned flags;
> +
> + spin_lock_irqsave(&accelerators_lock, flags);
> +
> + /*
> + * Look at list of loaded accelerators to see if the requested
> + * one is already there
> + */
> + list_for_each_entry( accelerator, &accelerators_list, link ) {
> + if ( match_accelerator(frontend, accelerator) ) {
> + /*
> + * Include this frontend device on the
> + * accelerator's list
> + */
> + add_accelerator_vif(accelerator, np, dev);
> +
> + ++accelerator->hooks_usecount;
> +
> + if ( accelerator->hooks == NULL )
> + DPRINTK("%s: no hooks set", __FUNCTION__);
> + else {
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> + accelerator->hooks->new_device(np->netdev, dev);
> + spin_lock_irqsave(&accelerators_lock, flags);
> + }
> +
> + if ( (--accelerator->hooks_usecount) == 0 )
> + accelerator_remove_hooks(accelerator);
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +
> + /*
> + * Hooks will get linked into vif_state by a
> + * future call by the accelerator to
> + * netfront_accelerator_ready()
> + */
> +
> + return 0;
> + }
> + }
> +
> + /* Couldn't find it, so create a new one and load the module */
> + if ( ( rc = init_accelerator(frontend, &accelerator) ) < 0 ) {
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> + return rc;
> + }
> +
> + /* Include this frontend device on the accelerator's list */
> + add_accelerator_vif(accelerator, np, dev);
> +
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +
> + DPRINTK("%s: loading module %s\n", __FUNCTION__, frontend);
> +
> + /* load module */
> + request_module("%s", frontend);
> +
> + /*
> + * Module should now call netfront_accelerator_loaded() once
> + * it's up and running, and we can continue from there
> + */
> +
> + return 0;
> +}
> +
> +/*
> + * Go through all the netfront vifs and see if they have requested
> + * this accelerator. Notify the accelerator plugin of the relevant
> + * device if so. Called when an accelerator plugin module is first
> + * loaded and connects to netfront.
> + */
> +static void
> +accelerator_probe_vifs_on_load(struct netfront_accelerator *accelerator)
> +{
> + struct netfront_accel_vif_state *accel_vif_state;
> +
> + DPRINTK("%s: %p\n", __FUNCTION__, accelerator);
> +
> + list_for_each_entry( accel_vif_state,
> + &accelerator->vif_states, link ) {
> + struct netfront_info *np = accel_vif_state->np;
> +
> + accelerator->hooks->new_device(np->netdev,
> + accel_vif_state->dev);
> +
> + /*
> + * Hooks will get linked into vif_state by a call to
> + * netfront_accelerator_ready() once accelerator
> + * plugin is ready for action
> + */
> + }
> +}
> +
> +
> +/*
> + * Called by the netfront accelerator plugin module when it has loaded
> + */
> +int netfront_accelerator_loaded(const char *frontend,
> + struct netfront_accel_hooks *hooks)
> +{
> + struct netfront_accelerator *accelerator;
> + unsigned flags;
> +
> + spin_lock_irqsave(&accelerators_lock, flags);
> +
> + /* Look through list of accelerators to see if it has already
> + been requested */
> + list_for_each_entry( accelerator, &accelerators_list, link ) {
> + if ( match_accelerator(frontend, accelerator) ) {
> + /*
> + * Deliberate double inc of usecount here -
> + * one to initialise it to 1 now hooks is
> + * being set (which persists until unloaded),
> + * and one for the use of hooks in this
> + * function (we don't want an unload to
> + * succeed in clearing hooks in the middle)
> + */
> + BUG_ON(accelerator->hooks != NULL ||
> + accelerator->hooks_usecount != 0);
> + accelerator->hooks_usecount = 2;
> +
> + accelerator->hooks = hooks;
> +
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +
> + accelerator_probe_vifs_on_load(accelerator);
> +
> + spin_lock_irqsave(&accelerators_lock, flags);
> + if ( (--accelerator->hooks_usecount) == 0 )
> + accelerator_remove_hooks(accelerator);
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +
> + return 0;
> + }
> + }
> +
> + /* If it wasn't in the list, add it now so that when it is
> + requested the caller will find it */
> + DPRINTK("%s: Couldn't find matching accelerator (%s)\n",
> + __FUNCTION__, frontend);
> +
> + init_accelerator(frontend, &accelerator);
> +
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +
> + return 0;
> +}
> +EXPORT_SYMBOL_GPL(netfront_accelerator_loaded);
> +
> +
> +/*
> + * Called by the accelerator module after it has been probed with a
> + * network device to say that it is ready to start accelerating
> + * traffic on that device
> + */
> +void netfront_accelerator_ready(const char *frontend,
> + struct xenbus_device *dev)
> +{
> + struct netfront_accelerator *accelerator;
> + struct netfront_accel_vif_state *accel_vif_state;
> + unsigned flags;
> +
> + spin_lock_irqsave(&accelerators_lock, flags);
> +
> + list_for_each_entry( accelerator, &accelerators_list, link ) {
> + if ( match_accelerator(frontend, accelerator) ) {
> + ++accelerator->hooks_usecount;
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +
> + list_for_each_entry( accel_vif_state,
> + &accelerator->vif_states, link ) {
> + if ( accel_vif_state->dev == dev )
> + accelerator_set_vif_state_hooks
> + (accel_vif_state);
> + }
> +
> + spin_lock_irqsave(&accelerators_lock, flags);
> + if ( (--accelerator->hooks_usecount) == 0 )
> + accelerator_remove_hooks(accelerator);
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +
> + return;
> + }
> + }
> +
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +}
> +EXPORT_SYMBOL_GPL(netfront_accelerator_ready);
> +
> +
> +/*
> + * Called by a netfront accelerator when it is unloaded. This safely
> + * removes the hooks into the plugin and blocks until all devices have
> + * finished using it, so on return it is safe to unload.
> + */
> +void netfront_accelerator_unloaded(const char *frontend)
> +{
> + struct netfront_accelerator *accelerator;
> + unsigned flags;
> +
> + spin_lock_irqsave(&accelerators_lock, flags);
> +
> + list_for_each_entry( accelerator, &accelerators_list, link ) {
> + if ( match_accelerator(frontend, accelerator) ) {
> + /*
> + * Use semaphore to ensure we know when all
> + * uses of hooks are complete
> + */
> + sema_init(&accelerator->exit_semaphore, 0);
> +
> + if ( (--accelerator->hooks_usecount) == 0 )
> + accelerator_remove_hooks(accelerator);
> +
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +
> + /* Wait for hooks to be unused, then return */
> + down(&accelerator->exit_semaphore);
> +
> + return;
> + }
> + }
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +}
> +EXPORT_SYMBOL_GPL(netfront_accelerator_unloaded);
> +
> +
> static inline int xennet_can_sg(struct net_device *dev)
> {
> return dev->features & NETIF_F_SG;
> @@ -331,8 +647,29 @@ static int __devexit netfront_remove(str
> static int __devexit netfront_remove(struct xenbus_device *dev)
> {
> struct netfront_info *info = dev->dev.driver_data;
> + unsigned flags;
>
> DPRINTK("%s\n", dev->nodename);
> +
> + /*
> + * Call the remove accelerator hook. The use count for the
> + * accelerator's hooks is incremented for the duration of the
> + * call to prevent the accelerator being able to modify the
> + * hooks in the middle (by, for example, unloading)
> + */
> + if ( info->accel_vif_state.hooks ) {
> + spin_lock_irqsave(&accelerators_lock, flags);
> + ++info->accelerator->hooks_usecount;
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +
> + if ( info->accel_vif_state.hooks )
> + info->accel_vif_state.hooks->remove(dev);
> +
> + spin_lock_irqsave(&accelerators_lock, flags);
> + if ( (--info->accelerator->hooks_usecount) == 0 )
> + accelerator_remove_hooks(info->accelerator);
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> + }
>
> netif_disconnect_backend(info);
>
> @@ -356,8 +693,29 @@ static int netfront_resume(struct xenbus
> static int netfront_resume(struct xenbus_device *dev)
> {
> struct netfront_info *info = dev->dev.driver_data;
> + unsigned flags;
>
> DPRINTK("%s\n", dev->nodename);
> +
> + /*
> + * Call the resume accelerator hook. The use count for the
> + * accelerator's hooks is incremented for the duration of
> + * the call to prevent the accelerator being able to modify
> + * the * hooks in the middle (by, for example, unloading)
> + */
> + if ( info->accel_vif_state.hooks ) {
> + spin_lock_irqsave(&accelerators_lock, flags);
> + ++info->accelerator->hooks_usecount;
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +
> + if ( info->accel_vif_state.hooks )
> + info->accel_vif_state.hooks->resume(dev);
> +
> + spin_lock_irqsave(&accelerators_lock, flags);
> + if ( (--info->accelerator->hooks_usecount) == 0 )
> + accelerator_remove_hooks(info->accelerator);
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> + }
>
> netif_disconnect_backend(info);
> return 0;
> @@ -553,6 +911,7 @@ static void backend_changed(struct xenbu
> {
> struct netfront_info *np = dev->dev.driver_data;
> struct net_device *netdev = np->netdev;
> + unsigned flags;
>
> DPRINTK("%s\n", xenbus_strstate(backend_state));
>
> @@ -577,6 +936,27 @@ static void backend_changed(struct xenbu
> xenbus_frontend_closed(dev);
> break;
> }
> +
> + /*
> + * Call the backend_changed accelerator hook. The use count
> + * for the accelerator's hooks is incremented for the duration
> + * of the call to prevent the accelerator being able to modify
> + * the hooks in the middle (by, for example, unloading)
> + */
> + if ( np->accel_vif_state.hooks ) {
> + spin_lock_irqsave(&accelerators_lock, flags);
> + ++np->accelerator->hooks_usecount;
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +
> + if ( np->accel_vif_state.hooks )
> + np->accel_vif_state.hooks->backend_changed
> + (dev, backend_state);
> +
> + spin_lock_irqsave(&accelerators_lock, flags);
> + if ( (--np->accelerator->hooks_usecount) == 0 )
> + accelerator_remove_hooks(np->accelerator);
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> + }
> }
>
> /** Send a packet on a net device to encourage switches to learn the
> @@ -613,19 +993,65 @@ static inline int netfront_tx_slot_avail
> (TX_MAX_TARGET - MAX_SKB_FRAGS - 2));
> }
>
> +
> +static inline
> +int netfront_check_accelerator_queue_busy(struct net_device *dev,
> + struct netfront_info *np)
> +{
> + int rc = 1;
> + unsigned flags;
> +
> + /*
> + * Call the check busy accelerator hook. The use count for the
> + * accelerator's hooks is incremented for the duration of the
> + * call to prevent the accelerator being able to modify the
> + * hooks in the middle (by, for example, unloading)
> + */
> + if ( np->accel_vif_state.hooks ) {
> + spin_lock_irqsave(&accelerators_lock, flags);
> + ++(np->accelerator->hooks_usecount);
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +
> + if ( np->accel_vif_state.hooks )
> + rc = np->accel_vif_state.hooks->check_busy(dev);
> +
> + spin_lock_irqsave(&accelerators_lock, flags);
> + if ( (--(np->accelerator->hooks_usecount) == 0 ) )
> + accelerator_remove_hooks(np->accelerator);
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> + }
> +
> + return rc;
> +}
> +
> +
> static inline void network_maybe_wake_tx(struct net_device *dev)
> {
> struct netfront_info *np = netdev_priv(dev);
>
> if (unlikely(netif_queue_stopped(dev)) &&
> netfront_tx_slot_available(np) &&
> - likely(netif_running(dev)))
> + likely(netif_running(dev)) &&
> + netfront_check_accelerator_queue_busy(dev, np))
> netif_wake_queue(dev);
> }
>
> +
> +int netfront_check_queue_busy(struct net_device *dev)
> +{
> + struct netfront_info *np = netdev_priv(dev);
> +
> + return unlikely(netif_queue_stopped(dev)) &&
> + netfront_tx_slot_available(np) &&
> + likely(netif_running(dev));
> +}
> +EXPORT_SYMBOL(netfront_check_queue_busy);
> +
> +
> static int network_open(struct net_device *dev)
> {
> struct netfront_info *np = netdev_priv(dev);
> + unsigned flags;
>
> memset(&np->stats, 0, sizeof(np->stats));
>
> @@ -633,8 +1059,34 @@ static int network_open(struct net_devic
> if (netfront_carrier_ok(np)) {
> network_alloc_rx_buffers(dev);
> np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
> - if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
> + if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx)){
> + /*
> + * Call the stop_napi_interrupts accelerator
> + * hook. The use count for the accelerator's
> + * hooks is incremented for the duration of
> + * the call to prevent the accelerator being
> + * able to modify the hooks in the middle (by,
> + * for example, unloading)
> + */
> + if ( np->accel_vif_state.hooks ) {
> + spin_lock_irqsave(&accelerators_lock, flags);
> + ++np->accelerator->hooks_usecount;
> + spin_unlock_irqrestore(&accelerators_lock,
> + flags);
> +
> + if ( np->accel_vif_state.hooks )
> + np->accel_vif_state.hooks->stop_napi_interrupts(dev);
> +
> + spin_lock_irqsave(&accelerators_lock, flags);
> + if ( (--np->accelerator->hooks_usecount)==0 )
> + accelerator_remove_hooks
> + (np->accelerator);
> + spin_unlock_irqrestore(&accelerators_lock,
> + flags);
> + }
> +
> netif_rx_schedule(dev);
> + }
> }
> spin_unlock_bh(&np->rx_lock);
>
> @@ -702,6 +1154,30 @@ static void rx_refill_timeout(unsigned l
> static void rx_refill_timeout(unsigned long data)
> {
> struct net_device *dev = (struct net_device *)data;
> + struct netfront_info *np = netdev_priv(dev);
> + unsigned flags;
> +
> + /*
> + * Call the stop_napi_interrupts accelerator hook. The use
> + * count for the accelerator's hooks is incremented for the
> + * duration of the call to prevent the accelerator being able
> + * to modify the hooks in the middle (by, for example,
> + * unloading)
> + */
> + if ( np->accel_vif_state.hooks ) {
> + spin_lock_irqsave(&accelerators_lock, flags);
> + ++np->accelerator->hooks_usecount;
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> +
> + if ( np->accel_vif_state.hooks )
> + np->accel_vif_state.hooks->stop_napi_interrupts(dev);
> +
> + spin_lock_irqsave(&accelerators_lock, flags);
> + if ( (--np->accelerator->hooks_usecount) == 0 )
> + accelerator_remove_hooks(np->accelerator);
> + spin_unlock_irqrestore(&accelerators_lock, flags);
> + }
> +
> netif_rx_schedule(dev);
> }
>
> @@ -941,6 +1417,12 @@ static int network_start_xmit(struct sk_
> unsigned int offset = offset_in_page(data);
> unsigned int len = skb_headlen(skb);
>
> + /* Check the fast path, if hooks are available */
> + if ( np->accel_vif_state.hooks &&
> + np->accel_vif_state.hooks->start_xmit(skb, dev) ) {
> + return 0;
> + }
> +
> frags += (offset + len + PAGE_SIZE - 1) / PAGE_SIZE;
> if (unlikely(frags > MAX_SKB_FRAGS + 1)) {
> printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n",
> @@ -1037,15 +1519,41 @@ static irqreturn_t netif_int(int irq, vo
> {
> struct net_device *dev = dev_id;
> struct netfront_info *np = netdev_priv(dev);
> - unsigned long flags;
> + unsigned long flags, flags1;
>
> spin_lock_irqsave(&np->tx_lock, flags);
>
> if (likely(netfront_carrier_ok(np))) {
> network_tx_buf_gc(dev);
> /* Under tx_lock: protects access to rx shared-ring indexes. */
> - if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
> + if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx)) {
> + /*
> + * Call the stop_napi_interrupts accelerator
> + * hook. The use count for the accelerator's
> + * hooks is incremented for the duration of
> + * the call to prevent the accelerator being
> + * able to modify the hooks in the middle (by,
> + * for example, unloading)
> + */
> + if ( np->accel_vif_state.hooks ) {
> + spin_lock_irqsave(&accelerators_lock, flags1);
> + ++np->accelerator->hooks_usecount;
> + spin_unlock_irqrestore(&accelerators_lock,
> + flags1);
> +
> + if ( np->accel_vif_state.hooks )
> + np->accel_vif_state.hooks->stop_napi_interrupts(dev);
> +
> + spin_lock_irqsave(&accelerators_lock, flags1);
> + if ( (--np->accelerator->hooks_usecount)==0 )
> + accelerator_remove_hooks
> + (np->accelerator);
> + spin_unlock_irqrestore(&accelerators_lock,
> + flags1);
> + }
> +
> netif_rx_schedule(dev);
> + }
> }
>
> spin_unlock_irqrestore(&np->tx_lock, flags);
> @@ -1305,7 +1813,7 @@ static int netif_poll(struct net_device
> struct netif_extra_info *extras = rinfo.extras;
> RING_IDX i, rp;
> struct multicall_entry *mcl;
> - int work_done, budget, more_to_do = 1;
> + int work_done, budget, more_to_do = 1, accel_more_to_do = 1;
> struct sk_buff_head rxq;
> struct sk_buff_head errq;
> struct sk_buff_head tmpq;
> @@ -1472,6 +1980,20 @@ err:
>
> network_alloc_rx_buffers(dev);
>
> + if (work_done < budget) {
> + /* there's some spare capacity, try the accelerated path */
> + int accel_budget = budget - work_done;
> + int accel_budget_start = accel_budget;
> +
> + if ( np->accel_vif_state.hooks ) {
> + accel_more_to_do =
> + np->accel_vif_state.hooks->netdev_poll
> + (dev, &accel_budget);
> + work_done += (accel_budget_start - accel_budget);
> + } else
> + accel_more_to_do = 0;
> + }
> +
> *pbudget -= work_done;
> dev->quota -= work_done;
>
> @@ -1479,15 +2001,28 @@ err:
> local_irq_save(flags);
>
> RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
> - if (!more_to_do)
> +
> + if (!more_to_do && !accel_more_to_do) {
> + /*
> + * Slow path has nothing more to do, see if
> + * fast path is likewise
> + */
> + if ( np->accel_vif_state.hooks ) {
> + accel_more_to_do =
> + np->accel_vif_state.hooks->start_napi_interrupts(dev);
> + }
> + }
> +
> + if (!more_to_do && !accel_more_to_do) {
> __netif_rx_complete(dev);
> + }
>
> local_irq_restore(flags);
> }
>
> spin_unlock(&np->rx_lock);
> -
> - return more_to_do;
> +
> + return more_to_do | accel_more_to_do;
> }
>
> static void netif_release_tx_bufs(struct netfront_info *np)
> @@ -1687,7 +2222,9 @@ static int network_connect(struct net_de
> struct sk_buff *skb;
> grant_ref_t ref;
> netif_rx_request_t *req;
> - unsigned int feature_rx_copy, feature_rx_flip;
> + unsigned int feature_rx_copy, feature_rx_flip, feature_accel;
> + char *accel_frontend;
> + int accel_len;
>
> err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
> "feature-rx-copy", "%u", &feature_rx_copy);
> @@ -1698,6 +2235,13 @@ static int network_connect(struct net_de
> if (err != 1)
> feature_rx_flip = 1;
>
> + feature_accel = 1;
> + accel_frontend = xenbus_read(XBT_NIL, np->xbdev->otherend,
> + "accel", &accel_len);
> + if ( IS_ERR(accel_frontend) ) {
> + feature_accel = 0;
> + }
> +
> /*
> * Copy packets on receive path if:
> * (a) This was requested by user, and the backend supports it; or
> @@ -1709,6 +2253,11 @@ static int network_connect(struct net_de
> err = talk_to_backend(np->xbdev, np);
> if (err)
> return err;
> +
> + if ( feature_accel ) {
> + netfront_load_accelerator(np, np->xbdev, accel_frontend);
> + kfree(accel_frontend);
> + }
>
> xennet_set_features(dev);
>
> @@ -1955,6 +2504,7 @@ static struct net_device * __devinit cre
>
> spin_lock_init(&np->tx_lock);
> spin_lock_init(&np->rx_lock);
> + np->accel_vif_state.hooks = NULL;
>
> skb_queue_head_init(&np->rx_batch);
> np->rx_target = RX_DFL_MIN_TARGET;
> @@ -2110,6 +2660,9 @@ static int __init netif_init(void)
> if (is_initial_xendomain())
> return 0;
>
> + INIT_LIST_HEAD(&accelerators_list);
> + spin_lock_init(&accelerators_lock);
> +
> IPRINTK("Initialising virtual ethernet driver.\n");
>
> (void)register_inetaddr_notifier(¬ifier_inetdev);
> diff -r cd3ade350f3f drivers/xen/netfront/netfront.h
> --- /dev/null Thu Jan 01 00:00:00 1970 +0000
> +++ b/drivers/xen/netfront/netfront.h Thu Jun 14 14:57:34 2007 +0100
> @@ -0,0 +1,198 @@
> +/******************************************************************************
> + * Virtual network driver for conversing with remote driver backends.
> + *
> + * Copyright (c) 2002-2005, K A Fraser
> + * Copyright (c) 2005, XenSource Ltd
> + * Copyright (C) 2007 Solarflare Communications, Inc.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License version 2
> + * as published by the Free Software Foundation; or, when distributed
> + * separately from the Linux kernel or incorporated into other
> + * software packages, subject to the following license:
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a copy
> + * of this source file (the "Software"), to deal in the Software without
> + * restriction, including without limitation the rights to use, copy, modify,
> + * merge, publish, distribute, sublicense, and/or sell copies of the Software,
> + * and to permit persons to whom the Software is furnished to do so, subject to
> + * the following conditions:
> + *
> + * The above copyright notice and this permission notice shall be included in
> + * all copies or substantial portions of the Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
> + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
> + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
> + * IN THE SOFTWARE.
> + */
> +
> +#ifndef NETFRONT_H
> +#define NETFRONT_H
> +
> +#include <xen/interface/io/netif.h>
> +#include <linux/netdevice.h>
> +#include <linux/skbuff.h>
> +#include <linux/list.h>
> +
> +#define NET_TX_RING_SIZE __RING_SIZE((struct netif_tx_sring *)0, PAGE_SIZE)
> +#define NET_RX_RING_SIZE __RING_SIZE((struct netif_rx_sring *)0, PAGE_SIZE)
> +
> +#include <xen/xenbus.h>
> +/* Function pointer table for hooks into a network acceleration
> + plugin. These are called at appropriate points from the netfront
> + driver */
> +struct netfront_accel_hooks {
> + /* new_device: The plugin is asked to support a new network interface */
> + int (*new_device)(struct net_device *net_dev, struct xenbus_device *dev);
> + /* suspend, resume, remove: Equivalent to the normal xenbus_* callbacks */
> + int (*suspend)(struct xenbus_device *dev);
> + int (*resume)(struct xenbus_device *dev);
> + int (*remove)(struct xenbus_device *dev);
> + /* backend_changed: Callback from watch based on backend's
> + xenbus state changing */
> + void (*backend_changed)(struct xenbus_device *dev,
> + enum xenbus_state backend_state);
> + /* The net_device is being polled, check the accelerated
> + hardware for any pending packets */
> + int (*netdev_poll)(struct net_device *dev, int *pbudget);
> + /* start_xmit: Used to give the accelerated plugin the option
> + of sending a packet. Returns non-zero if has done so, or
> + zero to decline and force the packet onto normal send path */
> + int (*start_xmit)(struct sk_buff *skb, struct net_device *dev);
> + /* start/stop_napi_interrupts Used by netfront to indicate
> + when napi interrupts should be enabled or disabled */
> + int (*start_napi_interrupts)(struct net_device *dev);
> + void (*stop_napi_interrupts)(struct net_device *dev);
> + /* Called before re-enabling the TX queue to check the fast
> + path has slots too */
> + int (*check_busy)(struct net_device *dev);
> +};
> +
> +/* Per-netfront device state for the accelerator. This is used to
> + allow efficient per-netfront device access to the accelerator hooks */
> +struct netfront_accel_vif_state {
> + struct list_head link;
> +
> + struct xenbus_device *dev;
> + struct netfront_info *np;
> + struct netfront_accel_hooks *hooks;
> +};
> +
> +/* Per-accelerator state stored in netfront. These form a list that
> + is used to track which devices are accelerated by which plugins,
> + and what plugins are available/have been requested */
> +struct netfront_accelerator {
> + /* Used to make a list */
> + struct list_head link;
> + /* ID of the accelerator */
> + int id;
> + /* String describing the accelerator. Currently this is the
> + name of the accelerator module. This is provided by the
> + backend accelerator through xenstore */
> + char *frontend;
> + /* The hooks into the accelerator plugin module */
> + struct netfront_accel_hooks *hooks;
> + /* Protect against removal of hooks while in use, must hold
> + accelerators_lock to change */
> + unsigned hooks_usecount;
> + /* List of per-netfront device state (struct netfront_accel_vif_state)
> + for each netfront device that is using this accelerator */
> + struct list_head vif_states;
> + /* Semaphore to signal that all users of this accelerator have
> + finished using it before module is unloaded */
> + struct semaphore exit_semaphore;
> +};
> +
> +
> +struct netfront_info {
> + struct list_head list;
> + struct net_device *netdev;
> +
> + struct net_device_stats stats;
> +
> + struct netif_tx_front_ring tx;
> + struct netif_rx_front_ring rx;
> +
> + spinlock_t tx_lock;
> + spinlock_t rx_lock;
> +
> + unsigned int irq;
> + unsigned int copying_receiver;
> + unsigned int carrier;
> +
> + /* Receive-ring batched refills. */
> +#define RX_MIN_TARGET 8
> +#define RX_DFL_MIN_TARGET 64
> +#define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
> + unsigned rx_min_target, rx_max_target, rx_target;
> + struct sk_buff_head rx_batch;
> +
> + struct timer_list rx_refill_timer;
> +
> + /*
> + * {tx,rx}_skbs store outstanding skbuffs. The first entry in tx_skbs
> + * is an index into a chain of free entries.
> + */
> + struct sk_buff *tx_skbs[NET_TX_RING_SIZE+1];
> + struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
> +
> +#define TX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
> + grant_ref_t gref_tx_head;
> + grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1];
> + grant_ref_t gref_rx_head;
> + grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
> +
> + struct xenbus_device *xbdev;
> + int tx_ring_ref;
> + int rx_ring_ref;
> + u8 mac[ETH_ALEN];
> +
> + unsigned long rx_pfn_array[NET_RX_RING_SIZE];
> + struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
> + struct mmu_update rx_mmu[NET_RX_RING_SIZE];
> +
> + /* Private pointer to state internal to accelerator module */
> + void *accel_priv;
> + /* The (list of) accelerator(s) used by this netfront device */
> + struct netfront_accelerator *accelerator;
> + /* The accelerator state for this netfront device */
> + struct netfront_accel_vif_state accel_vif_state;
> +};
> +
> +
> +/* Called by an accelerator plugin module when it has loaded.
> + *
> + * frontend: the string describing the accelerator, currently the module name
> + * hooks: the hooks for netfront to use to call into the accelerator
> + */
> +extern int netfront_accelerator_loaded(const char *frontend,
> + struct netfront_accel_hooks *hooks);
> +
> +/* Called when an accelerator plugin is ready to accelerate a device *
> + * that has been passed to it from netfront using the "new_device"
> + * hook.
> + *
> + * frontend: the string describing the accelerator. Must match the
> + * one passed to netfront_accelerator_loaded()
> + * dev: the xenbus device the plugin was asked to accelerate
> + */
> +extern void netfront_accelerator_ready(const char *frontend,
> + struct xenbus_device *dev);
> +
> +/* Called by an accelerator plugin module when it is about to unload.
> + *
> + * frontend: the string describing the accelerator. Must match the
> + * one passed to netfront_accelerator_loaded()
> + */
> +extern void netfront_accelerator_unloaded(const char *frontend);
> +
> +/* Called by an accelerator before waking the net device's TX queue to
> + * ensure the slow path has available slots. Returns true if OK to
> + * wake, false if still busy
> + */
> +extern int netfront_check_queue_busy(struct net_device *net_dev);
> +
> +#endif /* NETFRONT_H */
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@...ts.xensource.com
> http://lists.xensource.com/xen-devel
>
>
>
--
best regards,
hanzhu
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists