You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
openwrt/package/madwifi/patches/300-napi_polling.patch

537 lines
17 KiB
Diff

--- a/ath/if_ath.c
+++ b/ath/if_ath.c
@@ -184,7 +184,11 @@ static void ath_recv_mgmt(struct ieee802
struct sk_buff *, int, int, u_int64_t);
static void ath_setdefantenna(struct ath_softc *, u_int);
static struct ath_txq *ath_txq_setup(struct ath_softc *, int, int);
-static void ath_rx_tasklet(TQUEUE_ARG);
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,24)
+static int ath_rx_poll(struct napi_struct *napi, int budget);
+#else
+static int ath_rx_poll(struct net_device *dev, int *budget);
+#endif
static int ath_hardstart(struct sk_buff *, struct net_device *);
static int ath_mgtstart(struct ieee80211com *, struct sk_buff *);
#ifdef ATH_SUPERG_COMP
@@ -376,6 +380,9 @@ static u_int32_t ath_set_clamped_maxtxpo
u_int32_t new_clamped_maxtxpower);
static u_int32_t ath_get_real_maxtxpower(struct ath_softc *sc);
+static void ath_poll_disable(struct net_device *dev);
+static void ath_poll_enable(struct net_device *dev);
+
/* calibrate every 30 secs in steady state but check every second at first. */
static int ath_calinterval = ATH_SHORT_CALINTERVAL;
static int ath_countrycode = CTRY_DEFAULT; /* country code */
@@ -547,7 +554,6 @@ ath_attach(u_int16_t devid, struct net_d
atomic_set(&sc->sc_txbuf_counter, 0);
- ATH_INIT_TQUEUE(&sc->sc_rxtq, ath_rx_tasklet, dev);
ATH_INIT_TQUEUE(&sc->sc_txtq, ath_tx_tasklet, dev);
ATH_INIT_TQUEUE(&sc->sc_bmisstq, ath_bmiss_tasklet, dev);
ATH_INIT_TQUEUE(&sc->sc_bstucktq, ath_bstuck_tasklet, dev);
@@ -821,6 +827,12 @@ ath_attach(u_int16_t devid, struct net_d
dev->set_mac_address = ath_set_mac_address;
dev->change_mtu = ath_change_mtu;
dev->tx_queue_len = ATH_TXBUF - ATH_TXBUF_MGT_RESERVED;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,24)
+ netif_napi_add(dev, &sc->sc_napi, ath_rx_poll, 64);
+#else
+ dev->poll = ath_rx_poll;
+ dev->weight = 64;
+#endif
#ifdef USE_HEADERLEN_RESV
dev->hard_header_len += sizeof(struct ieee80211_qosframe) +
sizeof(struct llc) +
@@ -2220,6 +2232,7 @@ ath_intr(int irq, void *dev_id, struct p
(status & HAL_INT_GLOBAL) ? " HAL_INT_GLOBAL" : ""
);
+ sc->sc_isr = status;
status &= sc->sc_imask; /* discard unasked for bits */
/* As soon as we know we have a real interrupt we intend to service,
* we will check to see if we need an initial hardware TSF reading.
@@ -2277,7 +2290,21 @@ ath_intr(int irq, void *dev_id, struct p
}
if (status & (HAL_INT_RX | HAL_INT_RXPHY)) {
ath_uapsd_processtriggers(sc, hw_tsf);
- ATH_SCHEDULE_TQUEUE(&sc->sc_rxtq, &needmark);
+ sc->sc_isr &= ~HAL_INT_RX;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,24)
+ if (netif_rx_schedule_prep(dev, &sc->sc_napi))
+#else
+ if (netif_rx_schedule_prep(dev))
+#endif
+ {
+ sc->sc_imask &= ~HAL_INT_RX;
+ ath_hal_intrset(ah, sc->sc_imask);
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,24)
+ __netif_rx_schedule(dev, &sc->sc_napi);
+#else
+ __netif_rx_schedule(dev);
+#endif
+ }
}
if (status & HAL_INT_TX) {
#ifdef ATH_SUPERG_DYNTURBO
@@ -2303,6 +2330,11 @@ ath_intr(int irq, void *dev_id, struct p
}
}
#endif
+ /* disable transmit interrupt */
+ sc->sc_isr &= ~HAL_INT_TX;
+ ath_hal_intrset(ah, sc->sc_imask & ~HAL_INT_TX);
+ sc->sc_imask &= ~HAL_INT_TX;
+
ATH_SCHEDULE_TQUEUE(&sc->sc_txtq, &needmark);
}
if (status & HAL_INT_BMISS) {
@@ -2515,6 +2547,7 @@ ath_init(struct net_device *dev)
if (sc->sc_tx99 != NULL)
sc->sc_tx99->start(sc->sc_tx99);
#endif
+ ath_poll_enable(dev);
done:
ATH_UNLOCK(sc);
@@ -2555,6 +2588,9 @@ ath_stop_locked(struct net_device *dev)
if (sc->sc_tx99 != NULL)
sc->sc_tx99->stop(sc->sc_tx99);
#endif
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,24)
+ ath_poll_disable(dev);
+#endif
netif_stop_queue(dev); /* XXX re-enabled by ath_newstate */
dev->flags &= ~IFF_RUNNING; /* NB: avoid recursion */
ieee80211_stop_running(ic); /* stop all VAPs */
@@ -4013,12 +4049,47 @@ ath_key_set(struct ieee80211vap *vap, co
return ath_keyset(sc, k, mac, vap->iv_bss);
}
+static void ath_poll_disable(struct net_device *dev)
+{
+ struct ath_softc *sc = dev->priv;
+
+ /*
+ * XXX Using in_softirq is not right since we might
+ * be called from other soft irq contexts than
+ * ath_rx_poll
+ */
+ if (!in_softirq()) {
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,24)
+ napi_disable(&sc->sc_napi);
+#else
+ netif_poll_disable(dev);
+#endif
+ }
+}
+
+static void ath_poll_enable(struct net_device *dev)
+{
+ struct ath_softc *sc = dev->priv;
+
+ /* NB: see above */
+ if (!in_softirq()) {
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,24)
+ napi_enable(&sc->sc_napi);
+#else
+ netif_poll_enable(dev);
+#endif
+ }
+}
+
+
/*
* Block/unblock tx+rx processing while a key change is done.
* We assume the caller serializes key management operations
* so we only need to worry about synchronization with other
* uses that originate in the driver.
*/
+#define IS_UP(_dev) \
+ (((_dev)->flags & (IFF_RUNNING|IFF_UP)) == (IFF_RUNNING|IFF_UP))
static void
ath_key_update_begin(struct ieee80211vap *vap)
{
@@ -4032,14 +4103,9 @@ ath_key_update_begin(struct ieee80211vap
* When called from the rx tasklet we cannot use
* tasklet_disable because it will block waiting
* for us to complete execution.
- *
- * XXX Using in_softirq is not right since we might
- * be called from other soft irq contexts than
- * ath_rx_tasklet.
*/
- if (!in_softirq())
- tasklet_disable(&sc->sc_rxtq);
- netif_stop_queue(dev);
+ if (IS_UP(vap->iv_dev))
+ netif_stop_queue(dev);
}
static void
@@ -4051,9 +4117,9 @@ ath_key_update_end(struct ieee80211vap *
#endif
DPRINTF(sc, ATH_DEBUG_KEYCACHE, "End\n");
- netif_wake_queue(dev);
- if (!in_softirq()) /* NB: see above */
- tasklet_enable(&sc->sc_rxtq);
+
+ if (IS_UP(vap->iv_dev))
+ netif_wake_queue(dev);
}
/*
@@ -6360,15 +6426,25 @@ ath_setdefantenna(struct ath_softc *sc,
sc->sc_rxotherant = 0;
}
-static void
-ath_rx_tasklet(TQUEUE_ARG data)
+static int
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,24)
+ath_rx_poll(struct napi_struct *napi, int budget)
+#else
+ath_rx_poll(struct net_device *dev, int *budget)
+#endif
{
#define PA2DESC(_sc, _pa) \
((struct ath_desc *)((caddr_t)(_sc)->sc_rxdma.dd_desc + \
((_pa) - (_sc)->sc_rxdma.dd_desc_paddr)))
- struct net_device *dev = (struct net_device *)data;
- struct ath_buf *bf;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,24)
+ struct ath_softc *sc = container_of(napi, struct ath_softc, sc_napi);
+ struct net_device *dev = sc->sc_dev;
+ u_int rx_limit = budget;
+#else
struct ath_softc *sc = dev->priv;
+ u_int rx_limit = min(dev->quota, *budget);
+#endif
+ struct ath_buf *bf;
struct ieee80211com *ic = &sc->sc_ic;
struct ath_hal *ah = sc ? sc->sc_ah : NULL;
struct ath_desc *ds;
@@ -6378,8 +6454,10 @@ ath_rx_tasklet(TQUEUE_ARG data)
unsigned int len;
int type;
u_int phyerr;
+ u_int processed = 0, early_stop = 0;
DPRINTF(sc, ATH_DEBUG_RX_PROC, "invoked\n");
+process_rx_again:
do {
bf = STAILQ_FIRST(&sc->sc_rxbuf);
if (bf == NULL) { /* XXX ??? can this happen */
@@ -6403,6 +6481,15 @@ ath_rx_tasklet(TQUEUE_ARG data)
/* NB: never process the self-linked entry at the end */
break;
}
+
+ if (rx_limit-- < 2) {
+ early_stop = 1;
+ break;
+ }
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,24)
+ processed++;
+#endif
+
skb = bf->bf_skb;
if (skb == NULL) {
EPRINTF(sc, "Dropping; buffer contains NULL skbuff.\n");
@@ -6450,6 +6537,7 @@ ath_rx_tasklet(TQUEUE_ARG data)
sc->sc_stats.ast_rx_phyerr++;
phyerr = rs->rs_phyerr & 0x1f;
sc->sc_stats.ast_rx_phy[phyerr]++;
+ goto rx_next;
}
if (rs->rs_status & HAL_RXERR_DECRYPT) {
/*
@@ -6645,9 +6733,39 @@ rx_next:
STAILQ_INSERT_TAIL(&sc->sc_rxbuf, bf, bf_list);
ATH_RXBUF_UNLOCK_IRQ(sc);
} while (ath_rxbuf_init(sc, bf) == 0);
+ if (!early_stop) {
+ unsigned long flags;
+ /* Check if more data is received while we were
+ * processing the descriptor chain.
+ */
+ local_irq_save(flags);
+ if (sc->sc_isr & HAL_INT_RX) {
+ u_int64_t hw_tsf = ath_hal_gettsf64(ah);
+ sc->sc_isr &= ~HAL_INT_RX;
+ local_irq_restore(flags);
+ ath_uapsd_processtriggers(sc, hw_tsf);
+ goto process_rx_again;
+ }
+ local_irq_restore(flags);
+ }
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,24)
+ netif_rx_complete(dev, napi);
+#else
+ netif_rx_complete(dev);
+ *budget -= processed;
+ dev->quota -= processed;
+#endif
+ sc->sc_imask |= HAL_INT_RX;
+ ath_hal_intrset(ah, sc->sc_imask);
/* rx signal state monitoring */
ath_hal_rxmonitor(ah, &sc->sc_halstats, &sc->sc_curchan);
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,24)
+ return processed;
+#else
+ return early_stop;
+#endif
#undef PA2DESC
}
@@ -8298,12 +8416,24 @@ ath_tx_tasklet_q0(TQUEUE_ARG data)
{
struct net_device *dev = (struct net_device *)data;
struct ath_softc *sc = dev->priv;
+ unsigned long flags;
+process_tx_again:
if (txqactive(sc->sc_ah, 0))
ath_tx_processq(sc, &sc->sc_txq[0]);
if (txqactive(sc->sc_ah, sc->sc_cabq->axq_qnum))
ath_tx_processq(sc, sc->sc_cabq);
+ local_irq_save(flags);
+ if (sc->sc_isr & HAL_INT_TX) {
+ sc->sc_isr &= ~HAL_INT_TX;
+ local_irq_restore(flags);
+ goto process_tx_again;
+ }
+ sc->sc_imask |= HAL_INT_TX;
+ ath_hal_intrset(sc->sc_ah, sc->sc_imask);
+ local_irq_restore(flags);
+
netif_wake_queue(dev);
if (sc->sc_softled)
@@ -8319,7 +8449,9 @@ ath_tx_tasklet_q0123(TQUEUE_ARG data)
{
struct net_device *dev = (struct net_device *)data;
struct ath_softc *sc = dev->priv;
+ unsigned long flags;
+process_tx_again:
/*
* Process each active queue.
*/
@@ -8340,6 +8472,16 @@ ath_tx_tasklet_q0123(TQUEUE_ARG data)
if (sc->sc_uapsdq && txqactive(sc->sc_ah, sc->sc_uapsdq->axq_qnum))
ath_tx_processq(sc, sc->sc_uapsdq);
+ local_irq_save(flags);
+ if (sc->sc_isr & HAL_INT_TX) {
+ sc->sc_isr &= ~HAL_INT_TX;
+ local_irq_restore(flags);
+ goto process_tx_again;
+ }
+ sc->sc_imask |= HAL_INT_TX;
+ ath_hal_intrset(sc->sc_ah, sc->sc_imask);
+ local_irq_restore(flags);
+
netif_wake_queue(dev);
if (sc->sc_softled)
@@ -8355,13 +8497,25 @@ ath_tx_tasklet(TQUEUE_ARG data)
struct net_device *dev = (struct net_device *)data;
struct ath_softc *sc = dev->priv;
unsigned int i;
+ unsigned long flags;
/* Process each active queue. This includes sc_cabq, sc_xrtq and
* sc_uapsdq */
+process_tx_again:
for (i = 0; i < HAL_NUM_TX_QUEUES; i++)
if (ATH_TXQ_SETUP(sc, i) && txqactive(sc->sc_ah, i))
ath_tx_processq(sc, &sc->sc_txq[i]);
+ local_irq_save(flags);
+ if (sc->sc_isr & HAL_INT_TX) {
+ sc->sc_isr &= ~HAL_INT_TX;
+ local_irq_restore(flags);
+ goto process_tx_again;
+ }
+ sc->sc_imask |= HAL_INT_TX;
+ ath_hal_intrset(sc->sc_ah, sc->sc_imask);
+ local_irq_restore(flags);
+
netif_wake_queue(dev);
if (sc->sc_softled)
@@ -10296,9 +10450,9 @@ ath_change_mtu(struct net_device *dev, i
dev->mtu = mtu;
if ((dev->flags & IFF_RUNNING) && !sc->sc_invalid) {
/* NB: the rx buffers may need to be reallocated */
- tasklet_disable(&sc->sc_rxtq);
+ ath_poll_disable(dev);
error = ath_reset(dev);
- tasklet_enable(&sc->sc_rxtq);
+ ath_poll_enable(dev);
}
ATH_UNLOCK(sc);
--- a/ath/if_athvar.h
+++ b/ath/if_athvar.h
@@ -53,6 +53,10 @@
# include <asm/bitops.h>
#endif
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)
+#define irqs_disabled() 0
+#endif
+
/*
* Deduce if tasklets are available. If not then
* fall back to using the immediate work queue.
@@ -616,6 +620,9 @@ struct ath_rp {
struct ath_softc {
struct ieee80211com sc_ic; /* NB: must be first */
struct net_device *sc_dev;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,24)
+ struct napi_struct sc_napi;
+#endif
void __iomem *sc_iobase; /* address of the device */
struct semaphore sc_lock; /* dev-level lock */
struct net_device_stats sc_devstats; /* device statistics */
@@ -730,7 +737,6 @@ struct ath_softc {
struct ath_buf *sc_rxbufcur; /* current rx buffer */
u_int32_t *sc_rxlink; /* link ptr in last RX desc */
spinlock_t sc_rxbuflock;
- struct ATH_TQ_STRUCT sc_rxtq; /* rx intr tasklet */
struct ATH_TQ_STRUCT sc_rxorntq; /* rxorn intr tasklet */
u_int8_t sc_defant; /* current default antenna */
u_int8_t sc_rxotherant; /* RXs on non-default antenna */
@@ -745,6 +751,7 @@ struct ath_softc {
u_int sc_txintrperiod; /* tx interrupt batching */
struct ath_txq sc_txq[HAL_NUM_TX_QUEUES];
struct ath_txq *sc_ac2q[WME_NUM_AC]; /* WME AC -> h/w qnum */
+ HAL_INT sc_isr; /* unmasked ISR state */
struct ATH_TQ_STRUCT sc_txtq; /* tx intr tasklet */
u_int8_t sc_grppoll_str[GRPPOLL_RATE_STR_LEN];
struct ath_descdma sc_bdma; /* beacon descriptors */
@@ -858,6 +865,8 @@ typedef void (*ath_callback) (struct ath
#define ATH_TXBUF_LOCK_CHECK(_sc)
#endif
+#define ATH_DISABLE_INTR local_irq_disable
+#define ATH_ENABLE_INTR local_irq_enable
#define ATH_RXBUF_LOCK_INIT(_sc) spin_lock_init(&(_sc)->sc_rxbuflock)
#define ATH_RXBUF_LOCK_DESTROY(_sc)
--- a/net80211/ieee80211_input.c
+++ b/net80211/ieee80211_input.c
@@ -1198,7 +1198,7 @@ ieee80211_deliver_data(struct ieee80211_
/* attach vlan tag */
struct ieee80211_node *ni_tmp = SKB_CB(skb)->ni;
if (vlan_hwaccel_receive_skb(skb, vap->iv_vlgrp, ni->ni_vlan) == NET_RX_DROP) {
- /* If netif_rx dropped the packet because
+ /* If netif_receive_skb dropped the packet because
* device was too busy */
if (ni_tmp != NULL) {
/* node reference was leaked */
@@ -1209,8 +1209,8 @@ ieee80211_deliver_data(struct ieee80211_
skb = NULL; /* SKB is no longer ours */
} else {
struct ieee80211_node *ni_tmp = SKB_CB(skb)->ni;
- if (netif_rx(skb) == NET_RX_DROP) {
- /* If netif_rx dropped the packet because
+ if (netif_receive_skb(skb) == NET_RX_DROP) {
+ /* If netif_receive_skb dropped the packet because
* device was too busy */
if (ni_tmp != NULL) {
/* node reference was leaked */
@@ -2322,8 +2322,8 @@ forward_mgmt_to_app(struct ieee80211vap
skb1->protocol = __constant_htons(0x0019); /* ETH_P_80211_RAW */
ni_tmp = SKB_CB(skb1)->ni;
- if (netif_rx(skb1) == NET_RX_DROP) {
- /* If netif_rx dropped the packet because
+ if (netif_receive_skb(skb1) == NET_RX_DROP) {
+ /* If netif_receive_skb dropped the packet because
* device was too busy */
if (ni_tmp != NULL) {
/* node reference was leaked */
--- a/net80211/ieee80211_monitor.c
+++ b/net80211/ieee80211_monitor.c
@@ -584,8 +584,8 @@ ieee80211_input_monitor(struct ieee80211
skb1->protocol =
__constant_htons(0x0019); /* ETH_P_80211_RAW */
- if (netif_rx(skb1) == NET_RX_DROP) {
- /* If netif_rx dropped the packet because
+ if (netif_receive_skb(skb1) == NET_RX_DROP) {
+ /* If netif_receive_skb dropped the packet because
* device was too busy, reclaim the ref. in
* the skb. */
if (SKB_CB(skb1)->ni != NULL)
--- a/net80211/ieee80211_skb.c
+++ b/net80211/ieee80211_skb.c
@@ -73,7 +73,7 @@
#undef dev_queue_xmit
#undef kfree_skb
#undef kfree_skb_fast
-#undef netif_rx
+#undef netif_receive_skb
#undef pskb_copy
#undef skb_clone
#undef skb_copy
@@ -638,8 +638,8 @@ int vlan_hwaccel_receive_skb_debug(stru
grp, vlan_tag);
}
-int netif_rx_debug(struct sk_buff *skb, const char* func, int line) {
- return netif_rx(untrack_skb(skb, 0, func, line, __func__, __LINE__));
+int netif_receive_skb_debug(struct sk_buff *skb, const char* func, int line) {
+ return netif_receive_skb(untrack_skb(skb, 0, func, line, __func__, __LINE__));
}
struct sk_buff * alloc_skb_debug(unsigned int length, gfp_t gfp_mask,
@@ -760,7 +760,7 @@ struct sk_buff * skb_copy_expand_debug(c
}
EXPORT_SYMBOL(vlan_hwaccel_receive_skb_debug);
-EXPORT_SYMBOL(netif_rx_debug);
+EXPORT_SYMBOL(netif_receive_skb_debug);
EXPORT_SYMBOL(alloc_skb_debug);
EXPORT_SYMBOL(dev_alloc_skb_debug);
EXPORT_SYMBOL(skb_clone_debug);
--- a/net80211/ieee80211_skb.h
+++ b/net80211/ieee80211_skb.h
@@ -116,7 +116,7 @@ int ieee80211_skb_references(void);
int vlan_hwaccel_receive_skb_debug(struct sk_buff *skb,
struct vlan_group *grp, unsigned short vlan_tag,
const char* func, int line);
-int netif_rx_debug(struct sk_buff *skb, const char* func, int line);
+int netif_receive_skb_debug(struct sk_buff *skb, const char* func, int line);
struct sk_buff * alloc_skb_debug(unsigned int length, gfp_t gfp_mask,
const char *func, int line);
struct sk_buff * dev_alloc_skb_debug(unsigned int length,
@@ -151,7 +151,7 @@ struct sk_buff * skb_copy_expand_debug(c
#undef dev_queue_xmit
#undef kfree_skb
#undef kfree_skb_fast
-#undef netif_rx
+#undef netif_receive_skb
#undef pskb_copy
#undef skb_clone
#undef skb_copy
@@ -168,8 +168,8 @@ struct sk_buff * skb_copy_expand_debug(c
skb_copy_expand_debug(_skb, _newheadroom, _newtailroom, _gfp_mask, __func__, __LINE__)
#define vlan_hwaccel_receive_skb(_skb, _grp, _tag) \
vlan_hwaccel_receive_skb_debug(_skb, _grp, _tag, __func__, __LINE__)
-#define netif_rx(_skb) \
- netif_rx_debug(_skb, __func__, __LINE__)
+#define netif_receive_skb(_skb) \
+ netif_receive_skb_debug(_skb, __func__, __LINE__)
#define alloc_skb(_length, _gfp_mask) \
alloc_skb_debug(_length, _gfp_mask, __func__, __LINE__)
#define dev_alloc_skb(_length) \