#define SGE_FREEL_REFILL_THRESH 16
#define SGE_RESPQ_E_N 1024
#define SGE_INTRTIMER_NRES 1000
-#define SGE_RX_COPY_THRES 256
#define SGE_RX_SM_BUF_SIZE 1536
#define SGE_TX_DESC_MAX_PLEN 16384
-# define SGE_RX_DROP_THRES 2
-
#define SGE_RESPQ_REPLENISH_THRES (SGE_RESPQ_E_N / 4)
/*
*/
#define TX_RECLAIM_PERIOD (HZ / 4)
-#ifndef NET_IP_ALIGN
-# define NET_IP_ALIGN 2
-#endif
-
#define M_CMD_LEN 0x7fffffff
#define V_CMD_LEN(v) (v)
#define G_CMD_LEN(v) ((v) & M_CMD_LEN)
q->size = p->freelQ_size[i];
q->dma_offset = sge->rx_pkt_pad ? 0 : NET_IP_ALIGN;
size = sizeof(struct freelQ_e) * q->size;
- q->entries = (struct freelQ_e *)
- pci_alloc_consistent(pdev, size, &q->dma_addr);
+ q->entries = pci_alloc_consistent(pdev, size, &q->dma_addr);
if (!q->entries)
goto err_no_mem;
- memset(q->entries, 0, size);
+
size = sizeof(struct freelQ_ce) * q->size;
q->centries = kzalloc(size, GFP_KERNEL);
if (!q->centries)
sge->respQ.size = SGE_RESPQ_E_N;
sge->respQ.credits = 0;
size = sizeof(struct respQ_e) * sge->respQ.size;
- sge->respQ.entries = (struct respQ_e *)
+ sge->respQ.entries =
pci_alloc_consistent(pdev, size, &sge->respQ.dma_addr);
if (!sge->respQ.entries)
goto err_no_mem;
- memset(sge->respQ.entries, 0, size);
return 0;
err_no_mem:
q->in_use -= n;
ce = &q->centries[cidx];
while (n--) {
- if (q->sop) {
- if (likely(pci_unmap_len(ce, dma_len))) {
- pci_unmap_single(pdev,
- pci_unmap_addr(ce, dma_addr),
- pci_unmap_len(ce, dma_len),
- PCI_DMA_TODEVICE);
+ if (likely(pci_unmap_len(ce, dma_len))) {
+ pci_unmap_single(pdev, pci_unmap_addr(ce, dma_addr),
+ pci_unmap_len(ce, dma_len),
+ PCI_DMA_TODEVICE);
+ if (q->sop)
q->sop = 0;
- }
- } else {
- if (likely(pci_unmap_len(ce, dma_len))) {
- pci_unmap_page(pdev, pci_unmap_addr(ce, dma_addr),
- pci_unmap_len(ce, dma_len),
- PCI_DMA_TODEVICE);
- }
}
if (ce->skb) {
dev_kfree_skb_any(ce->skb);
q->stop_thres = 0;
spin_lock_init(&q->lock);
size = sizeof(struct cmdQ_e) * q->size;
- q->entries = (struct cmdQ_e *)
- pci_alloc_consistent(pdev, size, &q->dma_addr);
+ q->entries = pci_alloc_consistent(pdev, size, &q->dma_addr);
if (!q->entries)
goto err_no_mem;
- memset(q->entries, 0, size);
+
size = sizeof(struct cmdQ_ce) * q->size;
q->centries = kzalloc(size, GFP_KERNEL);
if (!q->centries)
skb_reserve(skb, q->dma_offset);
mapping = pci_map_single(pdev, skb->data, dma_len,
PCI_DMA_FROMDEVICE);
+ skb_reserve(skb, sge->rx_pkt_pad);
+
ce->skb = skb;
pci_unmap_addr_set(ce, dma_addr, mapping);
pci_unmap_len_set(ce, dma_len, dma_len);
}
}
+static int copybreak __read_mostly = 256;
+module_param(copybreak, int, 0);
+MODULE_PARM_DESC(copybreak, "Receive copy threshold");
+
/**
* get_packet - return the next ingress packet buffer
* @pdev: the PCI device that received the packet
* be copied but there is no memory for the copy.
*/
static inline struct sk_buff *get_packet(struct pci_dev *pdev,
- struct freelQ *fl, unsigned int len,
- int dma_pad, int skb_pad,
- unsigned int copy_thres,
- unsigned int drop_thres)
+ struct freelQ *fl, unsigned int len)
{
struct sk_buff *skb;
- struct freelQ_ce *ce = &fl->centries[fl->cidx];
+ const struct freelQ_ce *ce = &fl->centries[fl->cidx];
- if (len < copy_thres) {
- skb = alloc_skb(len + skb_pad, GFP_ATOMIC);
- if (likely(skb != NULL)) {
- skb_reserve(skb, skb_pad);
- skb_put(skb, len);
- pci_dma_sync_single_for_cpu(pdev,
- pci_unmap_addr(ce, dma_addr),
- pci_unmap_len(ce, dma_len),
- PCI_DMA_FROMDEVICE);
- memcpy(skb->data, ce->skb->data + dma_pad, len);
- pci_dma_sync_single_for_device(pdev,
+ if (len < copybreak) {
+ skb = alloc_skb(len + 2, GFP_ATOMIC);
+ if (!skb)
+ goto use_orig_buf;
+
+ skb_reserve(skb, 2); /* align IP header */
+ skb_put(skb, len);
+ pci_dma_sync_single_for_cpu(pdev,
pci_unmap_addr(ce, dma_addr),
pci_unmap_len(ce, dma_len),
PCI_DMA_FROMDEVICE);
- } else if (!drop_thres)
- goto use_orig_buf;
-
+ skb_copy_from_linear_data(ce->skb, skb->data, len);
+ pci_dma_sync_single_for_device(pdev,
+ pci_unmap_addr(ce, dma_addr),
+ pci_unmap_len(ce, dma_len),
+ PCI_DMA_FROMDEVICE);
recycle_fl_buf(fl, fl->cidx);
return skb;
}
- if (fl->credits < drop_thres) {
+use_orig_buf:
+ if (fl->credits < 2) {
recycle_fl_buf(fl, fl->cidx);
return NULL;
}
-use_orig_buf:
pci_unmap_single(pdev, pci_unmap_addr(ce, dma_addr),
pci_unmap_len(ce, dma_len), PCI_DMA_FROMDEVICE);
skb = ce->skb;
- skb_reserve(skb, dma_pad);
+ prefetch(skb->data);
+
skb_put(skb, len);
return skb;
}
*
* Process an ingress ethernet pakcet and deliver it to the stack.
*/
-static int sge_rx(struct sge *sge, struct freelQ *fl, unsigned int len)
+static void sge_rx(struct sge *sge, struct freelQ *fl, unsigned int len)
{
struct sk_buff *skb;
- struct cpl_rx_pkt *p;
+ const struct cpl_rx_pkt *p;
struct adapter *adapter = sge->adapter;
struct sge_port_stats *st;
- skb = get_packet(adapter->pdev, fl, len - sge->rx_pkt_pad,
- sge->rx_pkt_pad, 2, SGE_RX_COPY_THRES,
- SGE_RX_DROP_THRES);
+ skb = get_packet(adapter->pdev, fl, len - sge->rx_pkt_pad);
if (unlikely(!skb)) {
sge->stats.rx_drops++;
- return 0;
+ return;
}
- p = (struct cpl_rx_pkt *)skb->data;
- skb_pull(skb, sizeof(*p));
+ p = (const struct cpl_rx_pkt *) skb->data;
if (p->iff >= adapter->params.nports) {
kfree_skb(skb);
- return 0;
+ return;
}
+ __skb_pull(skb, sizeof(*p));
- skb->dev = adapter->port[p->iff].dev;
skb->dev->last_rx = jiffies;
st = per_cpu_ptr(sge->port_stats[p->iff], smp_processor_id());
st->rx_packets++;
- skb->protocol = eth_type_trans(skb, skb->dev);
+ skb->protocol = eth_type_trans(skb, adapter->port[p->iff].dev);
if ((adapter->flags & RX_CSUM_ENABLED) && p->csum == 0xffff &&
skb->protocol == htons(ETH_P_IP) &&
(skb->data[9] == IPPROTO_TCP || skb->data[9] == IPPROTO_UDP)) {
netif_rx(skb);
#endif
}
- return 0;
}
/*
static void restart_tx_queues(struct sge *sge)
{
struct adapter *adap = sge->adapter;
+ int i;
- if (enough_free_Tx_descs(&sge->cmdQ[0])) {
- int i;
+ if (!enough_free_Tx_descs(&sge->cmdQ[0]))
+ return;
- for_each_port(adap, i) {
- struct net_device *nd = adap->port[i].dev;
+ for_each_port(adap, i) {
+ struct net_device *nd = adap->port[i].dev;
- if (test_and_clear_bit(nd->if_port,
- &sge->stopped_tx_queues) &&
- netif_running(nd)) {
- sge->stats.cmdQ_restarted[2]++;
- netif_wake_queue(nd);
- }
+ if (test_and_clear_bit(nd->if_port, &sge->stopped_tx_queues) &&
+ netif_running(nd)) {
+ sge->stats.cmdQ_restarted[2]++;
+ netif_wake_queue(nd);
}
}
}
struct sge *sge = adapter->sge;
struct respQ *q = &sge->respQ;
struct respQ_e *e = &q->entries[q->cidx];
- int budget_left = budget;
+ int done = 0;
unsigned int flags = 0;
unsigned int cmdq_processed[SGE_CMDQ_N] = {0, 0};
-
- while (likely(budget_left && e->GenerationBit == q->genbit)) {
+ while (done < budget && e->GenerationBit == q->genbit) {
flags |= e->Qsleeping;
cmdq_processed[0] += e->Cmdq0CreditReturn;
* ping-pong of TX state information on MP where the sender
* might run on a different CPU than this function...
*/
- if (unlikely(flags & F_CMDQ0_ENABLE || cmdq_processed[0] > 64)) {
+ if (unlikely((flags & F_CMDQ0_ENABLE) || cmdq_processed[0] > 64)) {
flags = update_tx_info(adapter, flags, cmdq_processed[0]);
cmdq_processed[0] = 0;
}
+
if (unlikely(cmdq_processed[1] > 16)) {
sge->cmdQ[1].processed += cmdq_processed[1];
cmdq_processed[1] = 0;
}
+
if (likely(e->DataValid)) {
struct freelQ *fl = &sge->freelQ[e->FreelistQid];
else
sge_rx(sge, fl, e->BufferLength);
+ ++done;
+
/*
* Note: this depends on each packet consuming a
* single free-list buffer; cf. the BUG above.
*/
if (++fl->cidx == fl->size)
fl->cidx = 0;
+ prefetch(fl->centries[fl->cidx].skb);
+
if (unlikely(--fl->credits <
fl->size - SGE_FREEL_REFILL_THRESH))
refill_free_list(sge, fl);
writel(q->credits, adapter->regs + A_SG_RSPQUEUECREDIT);
q->credits = 0;
}
- --budget_left;
}
flags = update_tx_info(adapter, flags, cmdq_processed[0]);
sge->cmdQ[1].processed += cmdq_processed[1];
- budget -= budget_left;
- return budget;
+ return done;
+}
+
+static inline int responses_pending(const struct adapter *adapter)
+{
+ const struct respQ *Q = &adapter->sge->respQ;
+ const struct respQ_e *e = &Q->entries[Q->cidx];
+
+ return (e->GenerationBit == Q->genbit);
}
#ifdef CONFIG_CHELSIO_T1_NAPI
* which the caller must ensure is a valid pure response. Returns 1 if it
* encounters a valid data-carrying response, 0 otherwise.
*/
-static int process_pure_responses(struct adapter *adapter, struct respQ_e *e)
+static int process_pure_responses(struct adapter *adapter)
{
struct sge *sge = adapter->sge;
struct respQ *q = &sge->respQ;
+ struct respQ_e *e = &q->entries[q->cidx];
+ const struct freelQ *fl = &sge->freelQ[e->FreelistQid];
unsigned int flags = 0;
unsigned int cmdq_processed[SGE_CMDQ_N] = {0, 0};
+ prefetch(fl->centries[fl->cidx].skb);
+ if (e->DataValid)
+ return 1;
+
do {
flags |= e->Qsleeping;
int t1_poll(struct net_device *dev, int *budget)
{
struct adapter *adapter = dev->priv;
- int effective_budget = min(*budget, dev->quota);
- int work_done = process_responses(adapter, effective_budget);
+ int work_done;
+ work_done = process_responses(adapter, min(*budget, dev->quota));
*budget -= work_done;
dev->quota -= work_done;
- if (work_done >= effective_budget)
+ if (unlikely(responses_pending(adapter)))
return 1;
- spin_lock_irq(&adapter->async_lock);
- __netif_rx_complete(dev);
+ netif_rx_complete(dev);
writel(adapter->sge->respQ.cidx, adapter->regs + A_SG_SLEEPING);
- writel(adapter->slow_intr_mask | F_PL_INTR_SGE_DATA,
- adapter->regs + A_PL_ENABLE);
- spin_unlock_irq(&adapter->async_lock);
return 0;
+
}
/*
irqreturn_t t1_interrupt(int irq, void *data)
{
struct adapter *adapter = data;
- struct net_device *dev = adapter->sge->netdev;
struct sge *sge = adapter->sge;
- u32 cause;
- int handled = 0;
-
- cause = readl(adapter->regs + A_PL_CAUSE);
- if (cause == 0 || cause == ~0)
- return IRQ_NONE;
+ int handled;
- spin_lock(&adapter->async_lock);
- if (cause & F_PL_INTR_SGE_DATA) {
- struct respQ *q = &adapter->sge->respQ;
- struct respQ_e *e = &q->entries[q->cidx];
+ if (likely(responses_pending(adapter))) {
+ struct net_device *dev = sge->netdev;
- handled = 1;
writel(F_PL_INTR_SGE_DATA, adapter->regs + A_PL_CAUSE);
- if (e->GenerationBit == q->genbit &&
- __netif_rx_schedule_prep(dev)) {
- if (e->DataValid || process_pure_responses(adapter, e)) {
- /* mask off data IRQ */
- writel(adapter->slow_intr_mask,
- adapter->regs + A_PL_ENABLE);
- __netif_rx_schedule(sge->netdev);
- goto unlock;
+ if (__netif_rx_schedule_prep(dev)) {
+ if (process_pure_responses(adapter))
+ __netif_rx_schedule(dev);
+ else {
+ /* no data, no NAPI needed */
+ writel(sge->respQ.cidx, adapter->regs + A_SG_SLEEPING);
+ netif_poll_enable(dev); /* undo schedule_prep */
}
- /* no data, no NAPI needed */
- netif_poll_enable(dev);
-
}
- writel(q->cidx, adapter->regs + A_SG_SLEEPING);
- } else
- handled = t1_slow_intr_handler(adapter);
+ return IRQ_HANDLED;
+ }
+
+ spin_lock(&adapter->async_lock);
+ handled = t1_slow_intr_handler(adapter);
+ spin_unlock(&adapter->async_lock);
if (!handled)
sge->stats.unhandled_irqs++;
-unlock:
- spin_unlock(&adapter->async_lock);
+
return IRQ_RETVAL(handled != 0);
}
irqreturn_t t1_interrupt(int irq, void *cookie)
{
int work_done;
- struct respQ_e *e;
struct adapter *adapter = cookie;
struct respQ *Q = &adapter->sge->respQ;
spin_lock(&adapter->async_lock);
- e = &Q->entries[Q->cidx];
- prefetch(e);
writel(F_PL_INTR_SGE_DATA, adapter->regs + A_PL_CAUSE);
- if (likely(e->GenerationBit == Q->genbit))
+ if (likely(responses_pending(adapter)))
work_done = process_responses(adapter, -1);
else
work_done = t1_slow_intr_handler(adapter);
++st->tx_tso;
- eth_type = skb->nh.raw - skb->data == ETH_HLEN ?
+ eth_type = skb_network_offset(skb) == ETH_HLEN ?
CPL_ETH_II : CPL_ETH_II_VLAN;
hdr = (struct cpl_tx_pkt_lso *)skb_push(skb, sizeof(*hdr));
hdr->opcode = CPL_TX_PKT_LSO;
hdr->ip_csum_dis = hdr->l4_csum_dis = 0;
- hdr->ip_hdr_words = skb->nh.iph->ihl;
- hdr->tcp_hdr_words = skb->h.th->doff;
+ hdr->ip_hdr_words = ip_hdr(skb)->ihl;
+ hdr->tcp_hdr_words = tcp_hdr(skb)->doff;
hdr->eth_type_mss = htons(MK_ETH_TYPE_MSS(eth_type,
skb_shinfo(skb)->gso_size));
hdr->len = htonl(skb->len - sizeof(*hdr));
if (!(adapter->flags & UDP_CSUM_CAPABLE) &&
skb->ip_summed == CHECKSUM_PARTIAL &&
- skb->nh.iph->protocol == IPPROTO_UDP) {
+ ip_hdr(skb)->protocol == IPPROTO_UDP) {
if (unlikely(skb_checksum_help(skb))) {
pr_debug("%s: unable to do udp checksum\n", dev->name);
dev_kfree_skb_any(skb);
*/
if ((unlikely(!adapter->sge->espibug_skb[dev->if_port]))) {
if (skb->protocol == htons(ETH_P_ARP) &&
- skb->nh.arph->ar_op == htons(ARPOP_REQUEST)) {
+ arp_hdr(skb)->ar_op == htons(ARPOP_REQUEST)) {
adapter->sge->espibug_skb[dev->if_port] = skb;
/* We want to re-use this skb later. We
* simply bump the reference count and it
0x0, 0x7, 0x43, 0x0, 0x0, 0x0
};
- memcpy(skb->data + sizeof(struct cpl_tx_pkt),
- ch_mac_addr, ETH_ALEN);
- memcpy(skb->data + skb->len - 10,
- ch_mac_addr, ETH_ALEN);
+ skb_copy_to_linear_data_offset(skb,
+ sizeof(struct cpl_tx_pkt),
+ ch_mac_addr,
+ ETH_ALEN);
+ skb_copy_to_linear_data_offset(skb,
+ skb->len - 10,
+ ch_mac_addr,
+ ETH_ALEN);
skb->cb[0] = 0xff;
}
if (!skb->cb[0]) {
u8 ch_mac_addr[ETH_ALEN] =
{0x0, 0x7, 0x43, 0x0, 0x0, 0x0};
- memcpy(skb->data + sizeof(struct cpl_tx_pkt),
- ch_mac_addr, ETH_ALEN);
- memcpy(skb->data + skb->len - 10, ch_mac_addr,
- ETH_ALEN);
+ skb_copy_to_linear_data_offset(skb,
+ sizeof(struct cpl_tx_pkt),
+ ch_mac_addr,
+ ETH_ALEN);
+ skb_copy_to_linear_data_offset(skb,
+ skb->len - 10,
+ ch_mac_addr,
+ ETH_ALEN);
skb->cb[0] = 0xff;
}
if (adapter->params.nports > 1) {
tx_sched_init(sge);
sge->espibug_timer.function = espibug_workaround_t204;
- } else {
+ } else
sge->espibug_timer.function = espibug_workaround;
- }
sge->espibug_timer.data = (unsigned long)sge->adapter;
sge->espibug_timeout = 1;