* [PATCH net-next 0/4] New CAIF features
@ 2012-04-12 18:27 Sjur Brændeland
2012-04-12 18:27 ` [PATCH net-next 1/4] caif: set traffic class for caif packets Sjur Brændeland
` (4 more replies)
0 siblings, 5 replies; 6+ messages in thread
From: Sjur Brændeland @ 2012-04-12 18:27 UTC (permalink / raw)
To: netdev, davem; +Cc: sjurbren, Sjur Brændeland
Add feature to aggregate CAIF HSI frames into one large HSI-transfer.
The triggers for starting a HSI transfer are:
Max transfer size reached, timeout or reception of low-latency traffic.
As long as none of these triggers are met, CAIF-frames will be aggregated.
In order to be able to support aggregation and detect low-latency traffic,
CAIF packets are assigned a TC class.
In addition we are now allowing other kernel modules to use the HSI interface
while the CAIF interface is closed. The CAIF-HSI interface is changed so that
reference to the HSI interface is held only when CAIF-HSI is in state UP.
Initialization/de-initialization of the HSI HW is performed when the interface
is opened/closed instead of at module load time.
NOTE: This patch-set depends on the patch "caif-hsi: Free flip_buffer at shutdown",
sent in the Bug-fix patch-set.
Regards,
Sjur
--
Dmitry Tarnyagin (2):
caif: set traffic class for caif packets
caif-hsi: robust frame aggregation for HSI
Sjur Brændeland (2):
caif-hsi: Remove stop/start of queue.
caif-hsi: Postpone init of HSI until open()
drivers/net/caif/caif_hsi.c | 350 ++++++++++++++++++++++++++++--------------
include/net/caif/caif_hsi.h | 19 ++-
include/net/caif/cfpkt.h | 9 +-
net/caif/caif_socket.c | 16 ++-
net/caif/cfctrl.c | 4 +
net/caif/cfpkt_skbuff.c | 7 +
net/caif/cfsrvl.c | 3 +
7 files changed, 286 insertions(+), 122 deletions(-)
--
1.7.5.4
^ permalink raw reply [flat|nested] 6+ messages in thread
* [PATCH net-next 1/4] caif: set traffic class for caif packets
2012-04-12 18:27 [PATCH net-next 0/4] New CAIF features Sjur Brændeland
@ 2012-04-12 18:27 ` Sjur Brændeland
2012-04-12 18:27 ` [PATCH net-next 2/4] caif-hsi: robust frame aggregation for HSI Sjur Brændeland
` (3 subsequent siblings)
4 siblings, 0 replies; 6+ messages in thread
From: Sjur Brændeland @ 2012-04-12 18:27 UTC (permalink / raw)
To: netdev, davem; +Cc: sjurbren, Dmitry Tarnyagin
From: Dmitry Tarnyagin <dmitry.tarnyagin@stericsson.com>
Set traffic class for CAIF packets, based on socket
priority, CAIF protocol type, or type of message.
Traffic class mapping for different packet types:
- control: TC_PRIO_CONTROL;
- flow control: TC_PRIO_CONTROL;
- at: TC_PRIO_CONTROL;
- rfm: TC_PRIO_INTERACTIVE_BULK;
- other sockets: equals to socket's TC;
- network data: no change.
Signed-off-by: Dmitry Tarnyagin <dmitry.tarnyagin@stericsson.com>
---
include/net/caif/cfpkt.h | 9 ++++++++-
net/caif/caif_socket.c | 16 ++++++++++++++--
net/caif/cfctrl.c | 4 ++++
net/caif/cfpkt_skbuff.c | 7 +++++++
net/caif/cfsrvl.c | 3 +++
5 files changed, 36 insertions(+), 3 deletions(-)
diff --git a/include/net/caif/cfpkt.h b/include/net/caif/cfpkt.h
index 6bd200a..83a89ba 100644
--- a/include/net/caif/cfpkt.h
+++ b/include/net/caif/cfpkt.h
@@ -188,11 +188,18 @@ struct cfpkt *cfpkt_fromnative(enum caif_direction dir, void *nativepkt);
*/
void *cfpkt_tonative(struct cfpkt *pkt);
-
/*
* Returns packet information for a packet.
* pkt Packet to get info from;
* @return Packet information
*/
struct caif_payload_info *cfpkt_info(struct cfpkt *pkt);
+
+/** cfpkt_set_prio - set priority for a CAIF packet.
+ *
+ * @pkt: The CAIF packet to be adjusted.
+ * @prio: one of TC_PRIO_ constants.
+ */
+void cfpkt_set_prio(struct cfpkt *pkt, int prio);
+
#endif /* CFPKT_H_ */
diff --git a/net/caif/caif_socket.c b/net/caif/caif_socket.c
index 5016fa5..ce47ee9 100644
--- a/net/caif/caif_socket.c
+++ b/net/caif/caif_socket.c
@@ -19,7 +19,7 @@
#include <linux/uaccess.h>
#include <linux/debugfs.h>
#include <linux/caif/caif_socket.h>
-#include <linux/atomic.h>
+#include <linux/pkt_sched.h>
#include <net/sock.h>
#include <net/tcp_states.h>
#include <net/caif/caif_layer.h>
@@ -505,6 +505,7 @@ static int transmit_skb(struct sk_buff *skb, struct caifsock *cf_sk,
pkt = cfpkt_fromnative(CAIF_DIR_OUT, skb);
memset(skb->cb, 0, sizeof(struct caif_payload_info));
+ cfpkt_set_prio(pkt, cf_sk->sk.sk_priority);
if (cf_sk->layer.dn == NULL) {
kfree_skb(skb);
@@ -1062,6 +1063,18 @@ static int caif_create(struct net *net, struct socket *sock, int protocol,
/* Store the protocol */
sk->sk_protocol = (unsigned char) protocol;
+ /* Initialize default priority for well-known cases */
+ switch (protocol) {
+ case CAIFPROTO_AT:
+ sk->sk_priority = TC_PRIO_CONTROL;
+ break;
+ case CAIFPROTO_RFM:
+ sk->sk_priority = TC_PRIO_INTERACTIVE_BULK;
+ break;
+ default:
+ sk->sk_priority = TC_PRIO_BESTEFFORT;
+ }
+
/*
* Lock in order to try to stop someone from opening the socket
* too early.
@@ -1081,7 +1094,6 @@ static int caif_create(struct net *net, struct socket *sock, int protocol,
set_rx_flow_on(cf_sk);
/* Set default options on configuration */
- cf_sk->sk.sk_priority = CAIF_PRIO_NORMAL;
cf_sk->conn_req.link_selector = CAIF_LINK_LOW_LATENCY;
cf_sk->conn_req.protocol = protocol;
release_sock(&cf_sk->sk);
diff --git a/net/caif/cfctrl.c b/net/caif/cfctrl.c
index 5cf5222..047cd0e 100644
--- a/net/caif/cfctrl.c
+++ b/net/caif/cfctrl.c
@@ -9,6 +9,7 @@
#include <linux/stddef.h>
#include <linux/spinlock.h>
#include <linux/slab.h>
+#include <linux/pkt_sched.h>
#include <net/caif/caif_layer.h>
#include <net/caif/cfpkt.h>
#include <net/caif/cfctrl.h>
@@ -189,6 +190,7 @@ void cfctrl_enum_req(struct cflayer *layer, u8 physlinkid)
cfctrl->serv.dev_info.id = physlinkid;
cfpkt_addbdy(pkt, CFCTRL_CMD_ENUM);
cfpkt_addbdy(pkt, physlinkid);
+ cfpkt_set_prio(pkt, TC_PRIO_CONTROL);
dn->transmit(dn, pkt);
}
@@ -281,6 +283,7 @@ int cfctrl_linkup_request(struct cflayer *layer,
* might arrive with the newly allocated channel ID.
*/
cfpkt_info(pkt)->dev_info->id = param->phyid;
+ cfpkt_set_prio(pkt, TC_PRIO_CONTROL);
ret =
dn->transmit(dn, pkt);
if (ret < 0) {
@@ -314,6 +317,7 @@ int cfctrl_linkdown_req(struct cflayer *layer, u8 channelid,
cfpkt_addbdy(pkt, CFCTRL_CMD_LINK_DESTROY);
cfpkt_addbdy(pkt, channelid);
init_info(cfpkt_info(pkt), cfctrl);
+ cfpkt_set_prio(pkt, TC_PRIO_CONTROL);
ret =
dn->transmit(dn, pkt);
#ifndef CAIF_NO_LOOP
diff --git a/net/caif/cfpkt_skbuff.c b/net/caif/cfpkt_skbuff.c
index e335ba8..863dedd 100644
--- a/net/caif/cfpkt_skbuff.c
+++ b/net/caif/cfpkt_skbuff.c
@@ -381,6 +381,7 @@ struct cfpkt *cfpkt_split(struct cfpkt *pkt, u16 pos)
memcpy(skb2->data, split, len2nd);
skb2->tail += len2nd;
skb2->len += len2nd;
+ skb2->priority = skb->priority;
return skb_to_pkt(skb2);
}
@@ -394,3 +395,9 @@ struct caif_payload_info *cfpkt_info(struct cfpkt *pkt)
return (struct caif_payload_info *)&pkt_to_skb(pkt)->cb;
}
EXPORT_SYMBOL(cfpkt_info);
+
+void cfpkt_set_prio(struct cfpkt *pkt, int prio)
+{
+ pkt_to_skb(pkt)->priority = prio;
+}
+EXPORT_SYMBOL(cfpkt_set_prio);
diff --git a/net/caif/cfsrvl.c b/net/caif/cfsrvl.c
index 4aa33d4..dd485f6 100644
--- a/net/caif/cfsrvl.c
+++ b/net/caif/cfsrvl.c
@@ -11,6 +11,7 @@
#include <linux/errno.h>
#include <linux/slab.h>
#include <linux/module.h>
+#include <linux/pkt_sched.h>
#include <net/caif/caif_layer.h>
#include <net/caif/cfsrvl.h>
#include <net/caif/cfpkt.h>
@@ -120,6 +121,7 @@ static int cfservl_modemcmd(struct cflayer *layr, enum caif_modemcmd ctrl)
info->channel_id = service->layer.id;
info->hdr_len = 1;
info->dev_info = &service->dev_info;
+ cfpkt_set_prio(pkt, TC_PRIO_CONTROL);
return layr->dn->transmit(layr->dn, pkt);
}
case CAIF_MODEMCMD_FLOW_OFF_REQ:
@@ -140,6 +142,7 @@ static int cfservl_modemcmd(struct cflayer *layr, enum caif_modemcmd ctrl)
info->channel_id = service->layer.id;
info->hdr_len = 1;
info->dev_info = &service->dev_info;
+ cfpkt_set_prio(pkt, TC_PRIO_CONTROL);
return layr->dn->transmit(layr->dn, pkt);
}
default:
--
1.7.5.4
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH net-next 2/4] caif-hsi: robust frame aggregation for HSI
2012-04-12 18:27 [PATCH net-next 0/4] New CAIF features Sjur Brændeland
2012-04-12 18:27 ` [PATCH net-next 1/4] caif: set traffic class for caif packets Sjur Brændeland
@ 2012-04-12 18:27 ` Sjur Brændeland
2012-04-12 18:27 ` [PATCH net-next 3/4] caif-hsi: Remove stop/start of queue Sjur Brændeland
` (2 subsequent siblings)
4 siblings, 0 replies; 6+ messages in thread
From: Sjur Brændeland @ 2012-04-12 18:27 UTC (permalink / raw)
To: netdev, davem; +Cc: sjurbren, Dmitry Tarnyagin
From: Dmitry Tarnyagin <dmitry.tarnyagin@stericsson.com>
Implement aggregation algorithm, combining more data into a single
HSI transfer. 4 different traffic categories are supported:
1. TC_PRIO_CONTROL .. TC_PRIO_MAX (CTL)
2. TC_PRIO_INTERACTIVE (VO)
3. TC_PRIO_INTERACTIVE_BULK (VI)
4. TC_PRIO_BESTEFFORT, TC_PRIO_BULK, TC_PRIO_FILLER (BEBK)
Signed-off-by: Dmitry Tarnyagin <dmitry.tarnyagin@stericsson.com>
---
drivers/net/caif/caif_hsi.c | 243 +++++++++++++++++++++++++++++++++----------
include/net/caif/caif_hsi.h | 19 +++-
2 files changed, 205 insertions(+), 57 deletions(-)
diff --git a/drivers/net/caif/caif_hsi.c b/drivers/net/caif/caif_hsi.c
index 9c1c8cd..9849a23 100644
--- a/drivers/net/caif/caif_hsi.c
+++ b/drivers/net/caif/caif_hsi.c
@@ -19,6 +19,7 @@
#include <linux/if_arp.h>
#include <linux/timer.h>
#include <linux/rtnetlink.h>
+#include <linux/pkt_sched.h>
#include <net/caif/caif_layer.h>
#include <net/caif/caif_hsi.h>
@@ -34,6 +35,10 @@ static int inactivity_timeout = 1000;
module_param(inactivity_timeout, int, S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(inactivity_timeout, "Inactivity timeout on HSI, ms.");
+static int aggregation_timeout = 1;
+module_param(aggregation_timeout, int, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(aggregation_timeout, "Aggregation timeout on HSI, ms.");
+
/*
* HSI padding options.
* Warning: must be a base of 2 (& operation used) and can not be zero !
@@ -86,24 +91,84 @@ static void cfhsi_inactivity_tout(unsigned long arg)
queue_work(cfhsi->wq, &cfhsi->wake_down_work);
}
+static void cfhsi_update_aggregation_stats(struct cfhsi *cfhsi,
+ const struct sk_buff *skb,
+ int direction)
+{
+ struct caif_payload_info *info;
+ int hpad, tpad, len;
+
+ info = (struct caif_payload_info *)&skb->cb;
+ hpad = 1 + PAD_POW2((info->hdr_len + 1), hsi_head_align);
+ tpad = PAD_POW2((skb->len + hpad), hsi_tail_align);
+ len = skb->len + hpad + tpad;
+
+ if (direction > 0)
+ cfhsi->aggregation_len += len;
+ else if (direction < 0)
+ cfhsi->aggregation_len -= len;
+}
+
+static bool cfhsi_can_send_aggregate(struct cfhsi *cfhsi)
+{
+ int i;
+
+ if (cfhsi->aggregation_timeout < 0)
+ return true;
+
+ for (i = 0; i < CFHSI_PRIO_BEBK; ++i) {
+ if (cfhsi->qhead[i].qlen)
+ return true;
+ }
+
+ /* TODO: Use aggregation_len instead */
+ if (cfhsi->qhead[CFHSI_PRIO_BEBK].qlen >= CFHSI_MAX_PKTS)
+ return true;
+
+ return false;
+}
+
+static struct sk_buff *cfhsi_dequeue(struct cfhsi *cfhsi)
+{
+ struct sk_buff *skb;
+ int i;
+
+ for (i = 0; i < CFHSI_PRIO_LAST; ++i) {
+ skb = skb_dequeue(&cfhsi->qhead[i]);
+ if (skb)
+ break;
+ }
+
+ return skb;
+}
+
+static int cfhsi_tx_queue_len(struct cfhsi *cfhsi)
+{
+ int i, len = 0;
+ for (i = 0; i < CFHSI_PRIO_LAST; ++i)
+ len += skb_queue_len(&cfhsi->qhead[i]);
+ return len;
+}
+
static void cfhsi_abort_tx(struct cfhsi *cfhsi)
{
struct sk_buff *skb;
for (;;) {
spin_lock_bh(&cfhsi->lock);
- skb = skb_dequeue(&cfhsi->qhead);
+ skb = cfhsi_dequeue(cfhsi);
if (!skb)
break;
cfhsi->ndev->stats.tx_errors++;
cfhsi->ndev->stats.tx_dropped++;
+ cfhsi_update_aggregation_stats(cfhsi, skb, -1);
spin_unlock_bh(&cfhsi->lock);
kfree_skb(skb);
}
cfhsi->tx_state = CFHSI_TX_STATE_IDLE;
if (!test_bit(CFHSI_SHUTDOWN, &cfhsi->bits))
- mod_timer(&cfhsi->timer,
+ mod_timer(&cfhsi->inactivity_timer,
jiffies + cfhsi->inactivity_timeout);
spin_unlock_bh(&cfhsi->lock);
}
@@ -169,7 +234,7 @@ static int cfhsi_tx_frm(struct cfhsi_desc *desc, struct cfhsi *cfhsi)
struct sk_buff *skb;
u8 *pfrm = desc->emb_frm + CFHSI_MAX_EMB_FRM_SZ;
- skb = skb_dequeue(&cfhsi->qhead);
+ skb = cfhsi_dequeue(cfhsi);
if (!skb)
return 0;
@@ -196,11 +261,16 @@ static int cfhsi_tx_frm(struct cfhsi_desc *desc, struct cfhsi *cfhsi)
pemb += hpad;
/* Update network statistics. */
+ spin_lock_bh(&cfhsi->lock);
cfhsi->ndev->stats.tx_packets++;
cfhsi->ndev->stats.tx_bytes += skb->len;
+ cfhsi_update_aggregation_stats(cfhsi, skb, -1);
+ spin_unlock_bh(&cfhsi->lock);
/* Copy in embedded CAIF frame. */
skb_copy_bits(skb, 0, pemb, skb->len);
+
+ /* Consume the SKB */
consume_skb(skb);
skb = NULL;
}
@@ -214,7 +284,7 @@ static int cfhsi_tx_frm(struct cfhsi_desc *desc, struct cfhsi *cfhsi)
int tpad = 0;
if (!skb)
- skb = skb_dequeue(&cfhsi->qhead);
+ skb = cfhsi_dequeue(cfhsi);
if (!skb)
break;
@@ -233,8 +303,11 @@ static int cfhsi_tx_frm(struct cfhsi_desc *desc, struct cfhsi *cfhsi)
pfrm += hpad;
/* Update network statistics. */
+ spin_lock_bh(&cfhsi->lock);
cfhsi->ndev->stats.tx_packets++;
cfhsi->ndev->stats.tx_bytes += skb->len;
+ cfhsi_update_aggregation_stats(cfhsi, skb, -1);
+ spin_unlock_bh(&cfhsi->lock);
/* Copy in CAIF frame. */
skb_copy_bits(skb, 0, pfrm, skb->len);
@@ -244,6 +317,8 @@ static int cfhsi_tx_frm(struct cfhsi_desc *desc, struct cfhsi *cfhsi)
/* Update frame pointer. */
pfrm += skb->len + tpad;
+
+ /* Consume the SKB */
consume_skb(skb);
skb = NULL;
@@ -258,8 +333,7 @@ static int cfhsi_tx_frm(struct cfhsi_desc *desc, struct cfhsi *cfhsi)
}
/* Check if we can piggy-back another descriptor. */
- skb = skb_peek(&cfhsi->qhead);
- if (skb)
+ if (cfhsi_can_send_aggregate(cfhsi))
desc->header |= CFHSI_PIGGY_DESC;
else
desc->header &= ~CFHSI_PIGGY_DESC;
@@ -267,61 +341,71 @@ static int cfhsi_tx_frm(struct cfhsi_desc *desc, struct cfhsi *cfhsi)
return CFHSI_DESC_SZ + pld_len;
}
-static void cfhsi_tx_done(struct cfhsi *cfhsi)
+static void cfhsi_start_tx(struct cfhsi *cfhsi)
{
- struct cfhsi_desc *desc = NULL;
- int len = 0;
- int res;
+ struct cfhsi_desc *desc = (struct cfhsi_desc *)cfhsi->tx_buf;
+ int len, res;
dev_dbg(&cfhsi->ndev->dev, "%s.\n", __func__);
if (test_bit(CFHSI_SHUTDOWN, &cfhsi->bits))
return;
- desc = (struct cfhsi_desc *)cfhsi->tx_buf;
-
do {
- /*
- * Send flow on if flow off has been previously signalled
- * and number of packets is below low water mark.
- */
- spin_lock_bh(&cfhsi->lock);
- if (cfhsi->flow_off_sent &&
- cfhsi->qhead.qlen <= cfhsi->q_low_mark &&
- cfhsi->cfdev.flowctrl) {
-
- cfhsi->flow_off_sent = 0;
- cfhsi->cfdev.flowctrl(cfhsi->ndev, ON);
- }
- spin_unlock_bh(&cfhsi->lock);
-
/* Create HSI frame. */
- do {
- len = cfhsi_tx_frm(desc, cfhsi);
- if (!len) {
- spin_lock_bh(&cfhsi->lock);
- if (unlikely(skb_peek(&cfhsi->qhead))) {
- spin_unlock_bh(&cfhsi->lock);
- continue;
- }
- cfhsi->tx_state = CFHSI_TX_STATE_IDLE;
- /* Start inactivity timer. */
- mod_timer(&cfhsi->timer,
- jiffies + cfhsi->inactivity_timeout);
+ len = cfhsi_tx_frm(desc, cfhsi);
+ if (!len) {
+ spin_lock_bh(&cfhsi->lock);
+ if (unlikely(cfhsi_tx_queue_len(cfhsi))) {
spin_unlock_bh(&cfhsi->lock);
- goto done;
+ res = -EAGAIN;
+ continue;
}
- } while (!len);
+ cfhsi->tx_state = CFHSI_TX_STATE_IDLE;
+ /* Start inactivity timer. */
+ mod_timer(&cfhsi->inactivity_timer,
+ jiffies + cfhsi->inactivity_timeout);
+ spin_unlock_bh(&cfhsi->lock);
+ break;
+ }
/* Set up new transfer. */
res = cfhsi->dev->cfhsi_tx(cfhsi->tx_buf, len, cfhsi->dev);
- if (WARN_ON(res < 0)) {
+ if (WARN_ON(res < 0))
dev_err(&cfhsi->ndev->dev, "%s: TX error %d.\n",
__func__, res);
- }
} while (res < 0);
+}
+
+static void cfhsi_tx_done(struct cfhsi *cfhsi)
+{
+ dev_dbg(&cfhsi->ndev->dev, "%s.\n", __func__);
+
+ if (test_bit(CFHSI_SHUTDOWN, &cfhsi->bits))
+ return;
+
+ /*
+ * Send flow on if flow off has been previously signalled
+ * and number of packets is below low water mark.
+ */
+ spin_lock_bh(&cfhsi->lock);
+ if (cfhsi->flow_off_sent &&
+ cfhsi_tx_queue_len(cfhsi) <= cfhsi->q_low_mark &&
+ cfhsi->cfdev.flowctrl) {
+
+ cfhsi->flow_off_sent = 0;
+ cfhsi->cfdev.flowctrl(cfhsi->ndev, ON);
+ }
+
+ if (cfhsi_can_send_aggregate(cfhsi)) {
+ spin_unlock_bh(&cfhsi->lock);
+ cfhsi_start_tx(cfhsi);
+ } else {
+ mod_timer(&cfhsi->aggregation_timer,
+ jiffies + cfhsi->aggregation_timeout);
+ spin_unlock_bh(&cfhsi->lock);
+ }
-done:
return;
}
@@ -560,7 +644,7 @@ static void cfhsi_rx_done(struct cfhsi *cfhsi)
/* Update inactivity timer if pending. */
spin_lock_bh(&cfhsi->lock);
- mod_timer_pending(&cfhsi->timer,
+ mod_timer_pending(&cfhsi->inactivity_timer,
jiffies + cfhsi->inactivity_timeout);
spin_unlock_bh(&cfhsi->lock);
@@ -793,12 +877,12 @@ wake_ack:
spin_lock_bh(&cfhsi->lock);
- /* Resume transmit if queue is not empty. */
- if (!skb_peek(&cfhsi->qhead)) {
+ /* Resume transmit if queues are not empty. */
+ if (!cfhsi_tx_queue_len(cfhsi)) {
dev_dbg(&cfhsi->ndev->dev, "%s: Peer wake, start timer.\n",
__func__);
/* Start inactivity timer. */
- mod_timer(&cfhsi->timer,
+ mod_timer(&cfhsi->inactivity_timer,
jiffies + cfhsi->inactivity_timeout);
spin_unlock_bh(&cfhsi->lock);
return;
@@ -934,20 +1018,53 @@ static void cfhsi_wake_down_cb(struct cfhsi_drv *drv)
wake_up_interruptible(&cfhsi->wake_down_wait);
}
+static void cfhsi_aggregation_tout(unsigned long arg)
+{
+ struct cfhsi *cfhsi = (struct cfhsi *)arg;
+
+ dev_dbg(&cfhsi->ndev->dev, "%s.\n",
+ __func__);
+
+ cfhsi_start_tx(cfhsi);
+}
+
static int cfhsi_xmit(struct sk_buff *skb, struct net_device *dev)
{
struct cfhsi *cfhsi = NULL;
int start_xfer = 0;
int timer_active;
+ int prio;
if (!dev)
return -EINVAL;
cfhsi = netdev_priv(dev);
+ switch (skb->priority) {
+ case TC_PRIO_BESTEFFORT:
+ case TC_PRIO_FILLER:
+ case TC_PRIO_BULK:
+ prio = CFHSI_PRIO_BEBK;
+ break;
+ case TC_PRIO_INTERACTIVE_BULK:
+ prio = CFHSI_PRIO_VI;
+ break;
+ case TC_PRIO_INTERACTIVE:
+ prio = CFHSI_PRIO_VO;
+ break;
+ case TC_PRIO_CONTROL:
+ default:
+ prio = CFHSI_PRIO_CTL;
+ break;
+ }
+
spin_lock_bh(&cfhsi->lock);
- skb_queue_tail(&cfhsi->qhead, skb);
+ /* Update aggregation statistics */
+ cfhsi_update_aggregation_stats(cfhsi, skb, 1);
+
+ /* Queue the SKB */
+ skb_queue_tail(&cfhsi->qhead[prio], skb);
/* Sanity check; xmit should not be called after unregister_netdev */
if (WARN_ON(test_bit(CFHSI_SHUTDOWN, &cfhsi->bits))) {
@@ -958,7 +1075,7 @@ static int cfhsi_xmit(struct sk_buff *skb, struct net_device *dev)
/* Send flow off if number of packets is above high water mark. */
if (!cfhsi->flow_off_sent &&
- cfhsi->qhead.qlen > cfhsi->q_high_mark &&
+ cfhsi_tx_queue_len(cfhsi) > cfhsi->q_high_mark &&
cfhsi->cfdev.flowctrl) {
cfhsi->flow_off_sent = 1;
cfhsi->cfdev.flowctrl(cfhsi->ndev, OFF);
@@ -970,12 +1087,18 @@ static int cfhsi_xmit(struct sk_buff *skb, struct net_device *dev)
}
if (!start_xfer) {
+ /* Send aggregate if it is possible */
+ bool aggregate_ready =
+ cfhsi_can_send_aggregate(cfhsi) &&
+ del_timer(&cfhsi->aggregation_timer) > 0;
spin_unlock_bh(&cfhsi->lock);
+ if (aggregate_ready)
+ cfhsi_start_tx(cfhsi);
return 0;
}
/* Delete inactivity timer if started. */
- timer_active = del_timer_sync(&cfhsi->timer);
+ timer_active = del_timer_sync(&cfhsi->inactivity_timer);
spin_unlock_bh(&cfhsi->lock);
@@ -1026,6 +1149,7 @@ static const struct net_device_ops cfhsi_ops = {
static void cfhsi_setup(struct net_device *dev)
{
+ int i;
struct cfhsi *cfhsi = netdev_priv(dev);
dev->features = 0;
dev->netdev_ops = &cfhsi_ops;
@@ -1034,7 +1158,8 @@ static void cfhsi_setup(struct net_device *dev)
dev->mtu = CFHSI_MAX_CAIF_FRAME_SZ;
dev->tx_queue_len = 0;
dev->destructor = free_netdev;
- skb_queue_head_init(&cfhsi->qhead);
+ for (i = 0; i < CFHSI_PRIO_LAST; ++i)
+ skb_queue_head_init(&cfhsi->qhead[i]);
cfhsi->cfdev.link_select = CAIF_LINK_HIGH_BANDW;
cfhsi->cfdev.use_frag = false;
cfhsi->cfdev.use_stx = false;
@@ -1111,6 +1236,9 @@ int cfhsi_probe(struct platform_device *pdev)
cfhsi->inactivity_timeout = NEXT_TIMER_MAX_DELTA;
}
+ /* Initialize aggregation timeout */
+ cfhsi->aggregation_timeout = aggregation_timeout;
+
/* Initialize recieve vaiables. */
cfhsi->rx_ptr = cfhsi->rx_buf;
cfhsi->rx_len = CFHSI_DESC_SZ;
@@ -1150,13 +1278,17 @@ int cfhsi_probe(struct platform_device *pdev)
init_waitqueue_head(&cfhsi->flush_fifo_wait);
/* Setup the inactivity timer. */
- init_timer(&cfhsi->timer);
- cfhsi->timer.data = (unsigned long)cfhsi;
- cfhsi->timer.function = cfhsi_inactivity_tout;
+ init_timer(&cfhsi->inactivity_timer);
+ cfhsi->inactivity_timer.data = (unsigned long)cfhsi;
+ cfhsi->inactivity_timer.function = cfhsi_inactivity_tout;
/* Setup the slowpath RX timer. */
init_timer(&cfhsi->rx_slowpath_timer);
cfhsi->rx_slowpath_timer.data = (unsigned long)cfhsi;
cfhsi->rx_slowpath_timer.function = cfhsi_rx_slowpath;
+ /* Setup the aggregation timer. */
+ init_timer(&cfhsi->aggregation_timer);
+ cfhsi->aggregation_timer.data = (unsigned long)cfhsi;
+ cfhsi->aggregation_timer.function = cfhsi_aggregation_tout;
/* Add CAIF HSI device to list. */
spin_lock(&cfhsi_list_lock);
@@ -1222,8 +1354,9 @@ static void cfhsi_shutdown(struct cfhsi *cfhsi)
flush_workqueue(cfhsi->wq);
/* Delete timers if pending */
- del_timer_sync(&cfhsi->timer);
+ del_timer_sync(&cfhsi->inactivity_timer);
del_timer_sync(&cfhsi->rx_slowpath_timer);
+ del_timer_sync(&cfhsi->aggregation_timer);
/* Cancel pending RX request (if any) */
cfhsi->dev->cfhsi_rx_cancel(cfhsi->dev);
diff --git a/include/net/caif/caif_hsi.h b/include/net/caif/caif_hsi.h
index 6db8ecf..439dadc 100644
--- a/include/net/caif/caif_hsi.h
+++ b/include/net/caif/caif_hsi.h
@@ -123,12 +123,21 @@ struct cfhsi_rx_state {
bool piggy_desc;
};
+/* Priority mapping */
+enum {
+ CFHSI_PRIO_CTL = 0,
+ CFHSI_PRIO_VI,
+ CFHSI_PRIO_VO,
+ CFHSI_PRIO_BEBK,
+ CFHSI_PRIO_LAST,
+};
+
/* Structure implemented by CAIF HSI drivers. */
struct cfhsi {
struct caif_dev_common cfdev;
struct net_device *ndev;
struct platform_device *pdev;
- struct sk_buff_head qhead;
+ struct sk_buff_head qhead[CFHSI_PRIO_LAST];
struct cfhsi_drv drv;
struct cfhsi_dev *dev;
int tx_state;
@@ -151,8 +160,14 @@ struct cfhsi {
wait_queue_head_t wake_up_wait;
wait_queue_head_t wake_down_wait;
wait_queue_head_t flush_fifo_wait;
- struct timer_list timer;
+ struct timer_list inactivity_timer;
struct timer_list rx_slowpath_timer;
+
+ /* TX aggregation */
+ unsigned long aggregation_timeout;
+ int aggregation_len;
+ struct timer_list aggregation_timer;
+
unsigned long bits;
};
--
1.7.5.4
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH net-next 3/4] caif-hsi: Remove stop/start of queue.
2012-04-12 18:27 [PATCH net-next 0/4] New CAIF features Sjur Brændeland
2012-04-12 18:27 ` [PATCH net-next 1/4] caif: set traffic class for caif packets Sjur Brændeland
2012-04-12 18:27 ` [PATCH net-next 2/4] caif-hsi: robust frame aggregation for HSI Sjur Brændeland
@ 2012-04-12 18:27 ` Sjur Brændeland
2012-04-12 18:27 ` [PATCH net-next 4/4] caif-hsi: Postpone init of HSI until open() Sjur Brændeland
2012-04-13 15:37 ` [PATCH net-next 0/4] New CAIF features David Miller
4 siblings, 0 replies; 6+ messages in thread
From: Sjur Brændeland @ 2012-04-12 18:27 UTC (permalink / raw)
To: netdev, davem; +Cc: sjurbren, Sjur Brændeland
CAIF HSI is currently a virtual device. Stopping/starting the
queues is wrong on a virtual device.
Signed-off-by: Sjur Brændeland <sjur.brandeland@stericsson.com>
---
drivers/net/caif/caif_hsi.c | 10 ----------
1 files changed, 0 insertions(+), 10 deletions(-)
diff --git a/drivers/net/caif/caif_hsi.c b/drivers/net/caif/caif_hsi.c
index 9849a23..4253991 100644
--- a/drivers/net/caif/caif_hsi.c
+++ b/drivers/net/caif/caif_hsi.c
@@ -1129,15 +1129,11 @@ static int cfhsi_xmit(struct sk_buff *skb, struct net_device *dev)
static int cfhsi_open(struct net_device *dev)
{
- netif_wake_queue(dev);
-
return 0;
}
static int cfhsi_close(struct net_device *dev)
{
- netif_stop_queue(dev);
-
return 0;
}
@@ -1319,9 +1315,6 @@ int cfhsi_probe(struct platform_device *pdev)
__func__, res);
goto err_net_reg;
}
-
- netif_stop_queue(ndev);
-
return res;
err_net_reg:
@@ -1344,9 +1337,6 @@ static void cfhsi_shutdown(struct cfhsi *cfhsi)
{
u8 *tx_buf, *rx_buf, *flip_buf;
- /* Stop TXing */
- netif_tx_stop_all_queues(cfhsi->ndev);
-
/* going to shutdown driver */
set_bit(CFHSI_SHUTDOWN, &cfhsi->bits);
--
1.7.5.4
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH net-next 4/4] caif-hsi: Postpone init of HSI until open()
2012-04-12 18:27 [PATCH net-next 0/4] New CAIF features Sjur Brændeland
` (2 preceding siblings ...)
2012-04-12 18:27 ` [PATCH net-next 3/4] caif-hsi: Remove stop/start of queue Sjur Brændeland
@ 2012-04-12 18:27 ` Sjur Brændeland
2012-04-13 15:37 ` [PATCH net-next 0/4] New CAIF features David Miller
4 siblings, 0 replies; 6+ messages in thread
From: Sjur Brændeland @ 2012-04-12 18:27 UTC (permalink / raw)
To: netdev, davem; +Cc: sjurbren, Sjur Brændeland
Do the initialization of the HSI interface when the
interface is opened, instead of upon registration.
When the interface is closed the HSI interface is
de-initialized, allowing other modules to use the
HSI interface.
Signed-off-by: Sjur Brændeland <sjur.brandeland@stericsson.com>
---
Hi Dave,
This patch does not apply cleanly on net-next,
but depends on the bug-fix patch intended for 3.4:
caif-hsi: Free flip_buffer at shutdown
I hope this don't cause trouble for you.
Thanks,
Sjur
drivers/net/caif/caif_hsi.c | 97 ++++++++++++++++++++-----------------------
1 files changed, 45 insertions(+), 52 deletions(-)
diff --git a/drivers/net/caif/caif_hsi.c b/drivers/net/caif/caif_hsi.c
index 4253991..1520814 100644
--- a/drivers/net/caif/caif_hsi.c
+++ b/drivers/net/caif/caif_hsi.c
@@ -6,6 +6,8 @@
* License terms: GNU General Public License (GPL) version 2.
*/
+#define pr_fmt(fmt) KBUILD_MODNAME fmt
+
#include <linux/init.h>
#include <linux/module.h>
#include <linux/device.h>
@@ -1127,21 +1129,7 @@ static int cfhsi_xmit(struct sk_buff *skb, struct net_device *dev)
return 0;
}
-static int cfhsi_open(struct net_device *dev)
-{
- return 0;
-}
-
-static int cfhsi_close(struct net_device *dev)
-{
- return 0;
-}
-
-static const struct net_device_ops cfhsi_ops = {
- .ndo_open = cfhsi_open,
- .ndo_stop = cfhsi_close,
- .ndo_start_xmit = cfhsi_xmit
-};
+static const struct net_device_ops cfhsi_ops;
static void cfhsi_setup(struct net_device *dev)
{
@@ -1167,7 +1155,7 @@ int cfhsi_probe(struct platform_device *pdev)
{
struct cfhsi *cfhsi = NULL;
struct net_device *ndev;
- struct cfhsi_dev *dev;
+
int res;
ndev = alloc_netdev(sizeof(struct cfhsi), "cfhsi%d", cfhsi_setup);
@@ -1178,6 +1166,34 @@ int cfhsi_probe(struct platform_device *pdev)
cfhsi->ndev = ndev;
cfhsi->pdev = pdev;
+ /* Assign the HSI device. */
+ cfhsi->dev = pdev->dev.platform_data;
+
+ /* Assign the driver to this HSI device. */
+ cfhsi->dev->drv = &cfhsi->drv;
+
+ /* Register network device. */
+ res = register_netdev(ndev);
+ if (res) {
+ dev_err(&ndev->dev, "%s: Registration error: %d.\n",
+ __func__, res);
+ free_netdev(ndev);
+ }
+ /* Add CAIF HSI device to list. */
+ spin_lock(&cfhsi_list_lock);
+ list_add_tail(&cfhsi->list, &cfhsi_list);
+ spin_unlock(&cfhsi_list_lock);
+
+ return res;
+}
+
+static int cfhsi_open(struct net_device *ndev)
+{
+ struct cfhsi *cfhsi = netdev_priv(ndev);
+ int res;
+
+ clear_bit(CFHSI_SHUTDOWN, &cfhsi->bits);
+
/* Initialize state vaiables. */
cfhsi->tx_state = CFHSI_TX_STATE_IDLE;
cfhsi->rx_state.state = CFHSI_RX_STATE_DESC;
@@ -1187,12 +1203,6 @@ int cfhsi_probe(struct platform_device *pdev)
cfhsi->q_low_mark = LOW_WATER_MARK;
cfhsi->q_high_mark = HIGH_WATER_MARK;
- /* Assign the HSI device. */
- dev = (struct cfhsi_dev *)pdev->dev.platform_data;
- cfhsi->dev = dev;
-
- /* Assign the driver to this HSI device. */
- dev->drv = &cfhsi->drv;
/*
* Allocate a TX buffer with the size of a HSI packet descriptors
@@ -1260,9 +1270,9 @@ int cfhsi_probe(struct platform_device *pdev)
clear_bit(CFHSI_AWAKE, &cfhsi->bits);
/* Create work thread. */
- cfhsi->wq = create_singlethread_workqueue(pdev->name);
+ cfhsi->wq = create_singlethread_workqueue(cfhsi->pdev->name);
if (!cfhsi->wq) {
- dev_err(&ndev->dev, "%s: Failed to create work queue.\n",
+ dev_err(&cfhsi->ndev->dev, "%s: Failed to create work queue.\n",
__func__);
res = -ENODEV;
goto err_create_wq;
@@ -1286,11 +1296,6 @@ int cfhsi_probe(struct platform_device *pdev)
cfhsi->aggregation_timer.data = (unsigned long)cfhsi;
cfhsi->aggregation_timer.function = cfhsi_aggregation_tout;
- /* Add CAIF HSI device to list. */
- spin_lock(&cfhsi_list_lock);
- list_add_tail(&cfhsi->list, &cfhsi_list);
- spin_unlock(&cfhsi_list_lock);
-
/* Activate HSI interface. */
res = cfhsi->dev->cfhsi_up(cfhsi->dev);
if (res) {
@@ -1303,15 +1308,7 @@ int cfhsi_probe(struct platform_device *pdev)
/* Flush FIFO */
res = cfhsi_flush_fifo(cfhsi);
if (res) {
- dev_err(&ndev->dev, "%s: Can't flush FIFO: %d.\n",
- __func__, res);
- goto err_net_reg;
- }
-
- /* Register network device. */
- res = register_netdev(ndev);
- if (res) {
- dev_err(&ndev->dev, "%s: Registration error: %d.\n",
+ dev_err(&cfhsi->ndev->dev, "%s: Can't flush FIFO: %d.\n",
__func__, res);
goto err_net_reg;
}
@@ -1328,13 +1325,12 @@ int cfhsi_probe(struct platform_device *pdev)
err_alloc_rx:
kfree(cfhsi->tx_buf);
err_alloc_tx:
- free_netdev(ndev);
-
return res;
}
-static void cfhsi_shutdown(struct cfhsi *cfhsi)
+static int cfhsi_close(struct net_device *ndev)
{
+ struct cfhsi *cfhsi = netdev_priv(ndev);
u8 *tx_buf, *rx_buf, *flip_buf;
/* going to shutdown driver */
@@ -1364,15 +1360,19 @@ static void cfhsi_shutdown(struct cfhsi *cfhsi)
/* Deactivate interface */
cfhsi->dev->cfhsi_down(cfhsi->dev);
- /* Finally unregister the network device. */
- unregister_netdev(cfhsi->ndev);
-
/* Free buffers. */
kfree(tx_buf);
kfree(rx_buf);
kfree(flip_buf);
+ return 0;
}
+static const struct net_device_ops cfhsi_ops = {
+ .ndo_open = cfhsi_open,
+ .ndo_stop = cfhsi_close,
+ .ndo_start_xmit = cfhsi_xmit
+};
+
int cfhsi_remove(struct platform_device *pdev)
{
struct list_head *list_node;
@@ -1389,10 +1389,6 @@ int cfhsi_remove(struct platform_device *pdev)
/* Remove from list. */
list_del(list_node);
spin_unlock(&cfhsi_list_lock);
-
- /* Shutdown driver. */
- cfhsi_shutdown(cfhsi);
-
return 0;
}
}
@@ -1423,8 +1419,7 @@ static void __exit cfhsi_exit_module(void)
list_del(list_node);
spin_unlock(&cfhsi_list_lock);
- /* Shutdown driver. */
- cfhsi_shutdown(cfhsi);
+ unregister_netdevice(cfhsi->ndev);
spin_lock(&cfhsi_list_lock);
}
@@ -1449,8 +1444,6 @@ static int __init cfhsi_init_module(void)
goto err_dev_register;
}
- return result;
-
err_dev_register:
return result;
}
--
1.7.5.4
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH net-next 0/4] New CAIF features
2012-04-12 18:27 [PATCH net-next 0/4] New CAIF features Sjur Brændeland
` (3 preceding siblings ...)
2012-04-12 18:27 ` [PATCH net-next 4/4] caif-hsi: Postpone init of HSI until open() Sjur Brændeland
@ 2012-04-13 15:37 ` David Miller
4 siblings, 0 replies; 6+ messages in thread
From: David Miller @ 2012-04-13 15:37 UTC (permalink / raw)
To: sjur.brandeland; +Cc: netdev, sjurbren
From: Sjur Brændeland <sjur.brandeland@stericsson.com>
Date: Thu, 12 Apr 2012 20:27:23 +0200
> Add feature to aggregate CAIF HSI frames into one large HSI-transfer.
> The triggers for starting a HSI transfer are:
> Max transfer size reached, timeout or reception of low-latency traffic.
> As long as none of these triggers are met, CAIF-frames will be aggregated.
>
> In order to be able to support aggregation and detect low-latency traffic,
> CAIF packets are assigned a TC class.
>
> In addition we are now allowing other kernel modules to use the HSI interface
> while the CAIF interface is closed. The CAIF-HSI interface is changed so that
> reference to the HSI interface is held only when CAIF-HSI is in state UP.
> Initialization/de-initialization of the HSI HW is performed when the interface
> is opened/closed instead of at module load time.
>
> NOTE: This patch-set depends on the patch "caif-hsi: Free flip_buffer at shutdown",
> sent in the Bug-fix patch-set.
All applied, thanks for letting me know explicitly about the dependency.
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2012-04-13 15:38 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2012-04-12 18:27 [PATCH net-next 0/4] New CAIF features Sjur Brændeland
2012-04-12 18:27 ` [PATCH net-next 1/4] caif: set traffic class for caif packets Sjur Brændeland
2012-04-12 18:27 ` [PATCH net-next 2/4] caif-hsi: robust frame aggregation for HSI Sjur Brændeland
2012-04-12 18:27 ` [PATCH net-next 3/4] caif-hsi: Remove stop/start of queue Sjur Brændeland
2012-04-12 18:27 ` [PATCH net-next 4/4] caif-hsi: Postpone init of HSI until open() Sjur Brændeland
2012-04-13 15:37 ` [PATCH net-next 0/4] New CAIF features David Miller
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).