From: Shradha Shah <sshah@solarflare.com>
To: David Miller <davem@davemloft.net>
Cc: <netdev@vger.kernel.org>, <linux-net-drivers@solarflare.com>
Subject: [PATCH net-next] sfc: Add support for busy polling
Date: Tue, 22 Jul 2014 14:03:25 +0100 [thread overview]
Message-ID: <53CE611D.7080109@solarflare.com> (raw)
From: Alexandre Rames <arames@solarflare.com>
This patch adds the sfc driver code for implementing busy polling.
It adds ndo_busy_poll method and locking between it and napi poll.
It also adds each napi to the napi_hash right after netif_napi_add().
Uses efx_start_eventq and efx_stop_eventq in the self tests.
Signed-off-by: Shradha Shah <sshah@solarflare.com>
---
drivers/net/ethernet/sfc/efx.c | 54 ++++++++++++-
drivers/net/ethernet/sfc/efx.h | 2 +
drivers/net/ethernet/sfc/net_driver.h | 148 ++++++++++++++++++++++++++++++++++
drivers/net/ethernet/sfc/rx.c | 6 +-
drivers/net/ethernet/sfc/selftest.c | 5 +-
5 files changed, 208 insertions(+), 7 deletions(-)
diff --git a/drivers/net/ethernet/sfc/efx.c b/drivers/net/ethernet/sfc/efx.c
index 4b80c0b..4cebe9d 100644
--- a/drivers/net/ethernet/sfc/efx.c
+++ b/drivers/net/ethernet/sfc/efx.c
@@ -272,6 +272,9 @@ static int efx_poll(struct napi_struct *napi, int budget)
struct efx_nic *efx = channel->efx;
int spent;
+ if (!efx_channel_lock_napi(channel))
+ return budget;
+
netif_vdbg(efx, intr, efx->net_dev,
"channel %d NAPI poll executing on CPU %d\n",
channel->channel, raw_smp_processor_id());
@@ -311,6 +314,7 @@ static int efx_poll(struct napi_struct *napi, int budget)
efx_nic_eventq_read_ack(channel);
}
+ efx_channel_unlock_napi(channel);
return spent;
}
@@ -357,7 +361,7 @@ static int efx_init_eventq(struct efx_channel *channel)
}
/* Enable event queue processing and NAPI */
-static void efx_start_eventq(struct efx_channel *channel)
+void efx_start_eventq(struct efx_channel *channel)
{
netif_dbg(channel->efx, ifup, channel->efx->net_dev,
"chan %d start event queue\n", channel->channel);
@@ -366,17 +370,20 @@ static void efx_start_eventq(struct efx_channel *channel)
channel->enabled = true;
smp_wmb();
+ efx_channel_enable(channel);
napi_enable(&channel->napi_str);
efx_nic_eventq_read_ack(channel);
}
/* Disable event queue processing and NAPI */
-static void efx_stop_eventq(struct efx_channel *channel)
+void efx_stop_eventq(struct efx_channel *channel)
{
if (!channel->enabled)
return;
napi_disable(&channel->napi_str);
+ while (!efx_channel_disable(channel))
+ usleep_range(1000, 20000);
channel->enabled = false;
}
@@ -1960,6 +1967,8 @@ static void efx_init_napi_channel(struct efx_channel *channel)
channel->napi_dev = efx->net_dev;
netif_napi_add(channel->napi_dev, &channel->napi_str,
efx_poll, napi_weight);
+ napi_hash_add(&channel->napi_str);
+ efx_channel_init_lock(channel);
}
static void efx_init_napi(struct efx_nic *efx)
@@ -1972,8 +1981,10 @@ static void efx_init_napi(struct efx_nic *efx)
static void efx_fini_napi_channel(struct efx_channel *channel)
{
- if (channel->napi_dev)
+ if (channel->napi_dev) {
netif_napi_del(&channel->napi_str);
+ napi_hash_del(&channel->napi_str);
+ }
channel->napi_dev = NULL;
}
@@ -2008,6 +2019,37 @@ static void efx_netpoll(struct net_device *net_dev)
#endif
+#ifdef CONFIG_NET_RX_BUSY_POLL
+static int efx_busy_poll(struct napi_struct *napi)
+{
+ struct efx_channel *channel =
+ container_of(napi, struct efx_channel, napi_str);
+ struct efx_nic *efx = channel->efx;
+ int budget = 4;
+ int old_rx_packets, rx_packets;
+
+ if (!netif_running(efx->net_dev))
+ return LL_FLUSH_FAILED;
+
+ if (!efx_channel_lock_poll(channel))
+ return LL_FLUSH_BUSY;
+
+ old_rx_packets = channel->rx_queue.rx_packets;
+ efx_process_channel(channel, budget);
+
+ rx_packets = channel->rx_queue.rx_packets - old_rx_packets;
+
+ /* There is no race condition with NAPI here.
+ * NAPI will automatically be rescheduled if it yielded during busy
+ * polling, because it was not able to take the lock and thus returned
+ * the full budget.
+ */
+ efx_channel_unlock_poll(channel);
+
+ return rx_packets;
+}
+#endif
+
/**************************************************************************
*
* Kernel net device interface
@@ -2177,6 +2219,9 @@ static const struct net_device_ops efx_farch_netdev_ops = {
.ndo_poll_controller = efx_netpoll,
#endif
.ndo_setup_tc = efx_setup_tc,
+#ifdef CONFIG_NET_RX_BUSY_POLL
+ .ndo_busy_poll = efx_busy_poll,
+#endif
#ifdef CONFIG_RFS_ACCEL
.ndo_rx_flow_steer = efx_filter_rfs,
#endif
@@ -2197,6 +2242,9 @@ static const struct net_device_ops efx_ef10_netdev_ops = {
#ifdef CONFIG_NET_POLL_CONTROLLER
.ndo_poll_controller = efx_netpoll,
#endif
+#ifdef CONFIG_NET_RX_BUSY_POLL
+ .ndo_busy_poll = efx_busy_poll,
+#endif
#ifdef CONFIG_RFS_ACCEL
.ndo_rx_flow_steer = efx_filter_rfs,
#endif
diff --git a/drivers/net/ethernet/sfc/efx.h b/drivers/net/ethernet/sfc/efx.h
index b41601e..2587c58 100644
--- a/drivers/net/ethernet/sfc/efx.h
+++ b/drivers/net/ethernet/sfc/efx.h
@@ -194,6 +194,8 @@ int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
bool rx_may_override_tx);
void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
unsigned int *rx_usecs, bool *rx_adaptive);
+void efx_stop_eventq(struct efx_channel *channel);
+void efx_start_eventq(struct efx_channel *channel);
/* Dummy PHY ops for PHY drivers */
int efx_port_dummy_op_int(struct efx_nic *efx);
diff --git a/drivers/net/ethernet/sfc/net_driver.h b/drivers/net/ethernet/sfc/net_driver.h
index fb2e3bf..9ede320 100644
--- a/drivers/net/ethernet/sfc/net_driver.h
+++ b/drivers/net/ethernet/sfc/net_driver.h
@@ -28,6 +28,7 @@
#include <linux/vmalloc.h>
#include <linux/i2c.h>
#include <linux/mtd/mtd.h>
+#include <net/busy_poll.h>
#include "enum.h"
#include "bitfield.h"
@@ -387,6 +388,8 @@ enum efx_sync_events_state {
* @irq_moderation: IRQ moderation value (in hardware ticks)
* @napi_dev: Net device used with NAPI
* @napi_str: NAPI control structure
+ * @state: state for NAPI vs busy polling
+ * @state_lock: lock protecting @state
* @eventq: Event queue buffer
* @eventq_mask: Event queue pointer mask
* @eventq_read_ptr: Event queue read pointer
@@ -424,6 +427,22 @@ struct efx_channel {
unsigned int irq_moderation;
struct net_device *napi_dev;
struct napi_struct napi_str;
+#ifdef CONFIG_NET_RX_BUSY_POLL
+ unsigned int state;
+ spinlock_t state_lock;
+#define EFX_CHANNEL_STATE_IDLE 0
+#define EFX_CHANNEL_STATE_NAPI (1 << 0) /* NAPI owns this channel */
+#define EFX_CHANNEL_STATE_POLL (1 << 1) /* poll owns this channel */
+#define EFX_CHANNEL_STATE_DISABLED (1 << 2) /* channel is disabled */
+#define EFX_CHANNEL_STATE_NAPI_YIELD (1 << 3) /* NAPI yielded this channel */
+#define EFX_CHANNEL_STATE_POLL_YIELD (1 << 4) /* poll yielded this channel */
+#define EFX_CHANNEL_OWNED \
+ (EFX_CHANNEL_STATE_NAPI | EFX_CHANNEL_STATE_POLL)
+#define EFX_CHANNEL_LOCKED \
+ (EFX_CHANNEL_OWNED | EFX_CHANNEL_STATE_DISABLED)
+#define EFX_CHANNEL_USER_PEND \
+ (EFX_CHANNEL_STATE_POLL | EFX_CHANNEL_STATE_POLL_YIELD)
+#endif /* CONFIG_NET_RX_BUSY_POLL */
struct efx_special_buffer eventq;
unsigned int eventq_mask;
unsigned int eventq_read_ptr;
@@ -457,6 +476,135 @@ struct efx_channel {
u32 sync_timestamp_minor;
};
+#ifdef CONFIG_NET_RX_BUSY_POLL
+static inline void efx_channel_init_lock(struct efx_channel *channel)
+{
+ spin_lock_init(&channel->state_lock);
+}
+
+/* Called from the device poll routine to get ownership of a channel. */
+static inline bool efx_channel_lock_napi(struct efx_channel *channel)
+{
+ bool rc = true;
+
+ spin_lock_bh(&channel->state_lock);
+ if (channel->state & EFX_CHANNEL_LOCKED) {
+ WARN_ON(channel->state & EFX_CHANNEL_STATE_NAPI);
+ channel->state |= EFX_CHANNEL_STATE_NAPI_YIELD;
+ rc = false;
+ } else {
+ /* we don't care if someone yielded */
+ channel->state = EFX_CHANNEL_STATE_NAPI;
+ }
+ spin_unlock_bh(&channel->state_lock);
+ return rc;
+}
+
+static inline void efx_channel_unlock_napi(struct efx_channel *channel)
+{
+ spin_lock_bh(&channel->state_lock);
+ WARN_ON(channel->state &
+ (EFX_CHANNEL_STATE_POLL | EFX_CHANNEL_STATE_NAPI_YIELD));
+
+ channel->state &= EFX_CHANNEL_STATE_DISABLED;
+ spin_unlock_bh(&channel->state_lock);
+}
+
+/* Called from efx_busy_poll(). */
+static inline bool efx_channel_lock_poll(struct efx_channel *channel)
+{
+ bool rc = true;
+
+ spin_lock_bh(&channel->state_lock);
+ if ((channel->state & EFX_CHANNEL_LOCKED)) {
+ channel->state |= EFX_CHANNEL_STATE_POLL_YIELD;
+ rc = false;
+ } else {
+ /* preserve yield marks */
+ channel->state |= EFX_CHANNEL_STATE_POLL;
+ }
+ spin_unlock_bh(&channel->state_lock);
+ return rc;
+}
+
+/* Returns true if NAPI tried to get the channel while it was locked. */
+static inline void efx_channel_unlock_poll(struct efx_channel *channel)
+{
+ spin_lock_bh(&channel->state_lock);
+ WARN_ON(channel->state & EFX_CHANNEL_STATE_NAPI);
+
+ /* will reset state to idle, unless channel is disabled */
+ channel->state &= EFX_CHANNEL_STATE_DISABLED;
+ spin_unlock_bh(&channel->state_lock);
+}
+
+/* True if a socket is polling, even if it did not get the lock. */
+static inline bool efx_channel_busy_polling(struct efx_channel *channel)
+{
+ WARN_ON(!(channel->state & EFX_CHANNEL_OWNED));
+ return channel->state & EFX_CHANNEL_USER_PEND;
+}
+
+static inline void efx_channel_enable(struct efx_channel *channel)
+{
+ spin_lock_bh(&channel->state_lock);
+ channel->state = EFX_CHANNEL_STATE_IDLE;
+ spin_unlock_bh(&channel->state_lock);
+}
+
+/* False if the channel is currently owned. */
+static inline bool efx_channel_disable(struct efx_channel *channel)
+{
+ bool rc = true;
+
+ spin_lock_bh(&channel->state_lock);
+ if (channel->state & EFX_CHANNEL_OWNED)
+ rc = false;
+ channel->state |= EFX_CHANNEL_STATE_DISABLED;
+ spin_unlock_bh(&channel->state_lock);
+
+ return rc;
+}
+
+#else /* CONFIG_NET_RX_BUSY_POLL */
+
+static inline void efx_channel_init_lock(struct efx_channel *channel)
+{
+}
+
+static inline bool efx_channel_lock_napi(struct efx_channel *channel)
+{
+ return true;
+}
+
+static inline void efx_channel_unlock_napi(struct efx_channel *channel)
+{
+}
+
+static inline bool efx_channel_lock_poll(struct efx_channel *channel)
+{
+ return false;
+}
+
+static inline void efx_channel_unlock_poll(struct efx_channel *channel)
+{
+}
+
+static inline bool efx_channel_busy_polling(struct efx_channel *channel)
+{
+ return false;
+}
+
+static inline void efx_channel_enable(struct efx_channel *channel)
+{
+}
+
+static inline bool efx_channel_disable(struct efx_channel *channel)
+{
+ return true;
+}
+#endif /* CONFIG_NET_RX_BUSY_POLL */
+
/**
* struct efx_msi_context - Context for each MSI
* @efx: The associated NIC
diff --git a/drivers/net/ethernet/sfc/rx.c b/drivers/net/ethernet/sfc/rx.c
index a7bb63a..c0ad95d 100644
--- a/drivers/net/ethernet/sfc/rx.c
+++ b/drivers/net/ethernet/sfc/rx.c
@@ -462,6 +462,7 @@ efx_rx_packet_gro(struct efx_channel *channel, struct efx_rx_buffer *rx_buf,
skb_record_rx_queue(skb, channel->rx_queue.core_index);
+ skb_mark_napi_id(skb, &channel->napi_str);
gro_result = napi_gro_frags(napi);
if (gro_result != GRO_DROP)
channel->irq_mod_score += 2;
@@ -520,6 +521,8 @@ static struct sk_buff *efx_rx_mk_skb(struct efx_channel *channel,
/* Move past the ethernet header */
skb->protocol = eth_type_trans(skb, efx->net_dev);
+ skb_mark_napi_id(skb, &channel->napi_str);
+
return skb;
}
@@ -666,7 +669,8 @@ void __efx_rx_packet(struct efx_channel *channel)
if (unlikely(!(efx->net_dev->features & NETIF_F_RXCSUM)))
rx_buf->flags &= ~EFX_RX_PKT_CSUMMED;
- if ((rx_buf->flags & EFX_RX_PKT_TCP) && !channel->type->receive_skb)
+ if ((rx_buf->flags & EFX_RX_PKT_TCP) && !channel->type->receive_skb &&
+ !efx_channel_busy_polling(channel))
efx_rx_packet_gro(channel, rx_buf, channel->rx_pkt_n_frags, eh);
else
efx_rx_deliver(channel, eh, rx_buf, channel->rx_pkt_n_frags);
diff --git a/drivers/net/ethernet/sfc/selftest.c b/drivers/net/ethernet/sfc/selftest.c
index 0fc5bae..10b6173 100644
--- a/drivers/net/ethernet/sfc/selftest.c
+++ b/drivers/net/ethernet/sfc/selftest.c
@@ -188,7 +188,7 @@ static int efx_test_eventq_irq(struct efx_nic *efx,
schedule_timeout_uninterruptible(wait);
efx_for_each_channel(channel, efx) {
- napi_disable(&channel->napi_str);
+ efx_stop_eventq(channel);
if (channel->eventq_read_ptr !=
read_ptr[channel->channel]) {
set_bit(channel->channel, &napi_ran);
@@ -200,8 +200,7 @@ static int efx_test_eventq_irq(struct efx_nic *efx,
if (efx_nic_event_test_irq_cpu(channel) >= 0)
clear_bit(channel->channel, &int_pend);
}
- napi_enable(&channel->napi_str);
- efx_nic_eventq_read_ack(channel);
+ efx_start_eventq(channel);
}
wait *= 2;
next reply other threads:[~2014-07-22 13:03 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2014-07-22 13:03 Shradha Shah [this message]
2014-07-23 2:55 ` [PATCH net-next] sfc: Add support for busy polling David Miller
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=53CE611D.7080109@solarflare.com \
--to=sshah@solarflare.com \
--cc=davem@davemloft.net \
--cc=linux-net-drivers@solarflare.com \
--cc=netdev@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.