* are those of the authors and should not be interpreted as representing
* official policies, either expressed or implied, of Alacritech, Inc.
*
+ * Parts developed by LinSysSoft Sahara team
+ *
**************************************************************************/
/*
#include <linux/types.h>
#include <linux/dma-mapping.h>
#include <linux/mii.h>
+#include <linux/ip.h>
+#include <linux/in.h>
+#include <linux/tcp.h>
+#include <linux/ipv6.h>
#define SLIC_GET_STATS_ENABLED 0
#define LINUX_FREES_ADAPTER_RESOURCES 1
static int sxg_allocate_buffer_memory(struct adapter_t *adapter, u32 Size,
enum sxg_buffer_type BufferType);
-static void sxg_allocate_rcvblock_complete(struct adapter_t *adapter,
+static int sxg_allocate_rcvblock_complete(struct adapter_t *adapter,
void *RcvBlock,
dma_addr_t PhysicalAddress,
u32 Length);
static void sxg_mcast_init_crc32(void);
static int sxg_entry_open(struct net_device *dev);
+static int sxg_second_open(struct net_device * dev);
static int sxg_entry_halt(struct net_device *dev);
static int sxg_ioctl(struct net_device *dev, struct ifreq *rq, int cmd);
static int sxg_send_packets(struct sk_buff *skb, struct net_device *dev);
*/
for (i = 0; i < SXG_INITIAL_RCV_DATA_BUFFERS;
i += SXG_RCV_DESCRIPTORS_PER_BLOCK) {
- sxg_allocate_buffer_memory(adapter,
+ status = sxg_allocate_buffer_memory(adapter,
SXG_RCV_BLOCK_SIZE(SXG_RCV_DATA_HDR_SIZE),
SXG_BUFFER_TYPE_RCV);
+ if (status != STATUS_SUCCESS)
+ return status;
}
/*
* NBL resource allocation can fail in the 'AllocateComplete' routine,
SXG_TRACE(TRACE_SXG, SxgTraceBuffer, TRACE_NOISY, "XAlcResS",
adapter, SXG_MAX_ENTRIES, 0, 0);
- DBG_ERROR("%s EXIT\n", __func__);
- return (STATUS_SUCCESS);
+ return status;
}
/*
if (!memmapped_ioaddr) {
DBG_ERROR("%s cannot remap MMIO region %lx @ %lx\n",
__func__, mmio_len, mmio_start);
- goto err_out_free_mmio_region;
+ goto err_out_free_mmio_region_0;
}
DBG_ERROR("sxg: %s found Alacritech SXG PCI, MMIO at %p, start[%lx] \
if (!memmapped_ioaddr) {
DBG_ERROR("%s cannot remap MMIO region %lx @ %lx\n",
__func__, mmio_len, mmio_start);
- goto err_out_free_mmio_region;
+ goto err_out_free_mmio_region_2;
}
DBG_ERROR("sxg: %s found Alacritech SXG PCI, MMIO at %p, "
return status;
err_out_unmap:
- iounmap((void *)memmapped_ioaddr);
+ sxg_free_resources(adapter);
+
+ err_out_free_mmio_region_2:
+
+ mmio_start = pci_resource_start(pcidev, 2);
+ mmio_len = pci_resource_len(pcidev, 2);
+ release_mem_region(mmio_start, mmio_len);
+
+ err_out_free_mmio_region_0:
+
+ mmio_start = pci_resource_start(pcidev, 0);
+ mmio_len = pci_resource_len(pcidev, 0);
- err_out_free_mmio_region:
release_mem_region(mmio_start, mmio_len);
err_out_exit_sxg_probe:
DBG_ERROR("%s EXIT jiffies[%lx] cpu %d\n", __func__, jiffies,
smp_processor_id());
+ pci_disable_device(pcidev);
+ DBG_ERROR("sxg: %s deallocate device\n", __FUNCTION__);
+ kfree(netdev);
+ printk("Exit %s, Sxg driver loading failed..\n", __FUNCTION__);
+
return -ENODEV;
}
struct sxg_event_ring *EventRing = &adapter->EventRings[RssId];
struct sxg_event *Event = &EventRing->Ring[adapter->NextEvent[RssId]];
u32 EventsProcessed = 0, Batches = 0;
- u32 num_skbs = 0;
struct sk_buff *skb;
#ifdef LINUX_HANDLES_RCV_INDICATION_LISTS
struct sk_buff *prev_skb = NULL;
{
struct adapter_t *adapter = (struct adapter_t *) netdev_priv(dev);
int status;
+ static int turn;
+
+ if (turn) {
+ sxg_second_open(adapter->netdev);
+
+ return STATUS_SUCCESS;
+ }
+
+ turn++;
ASSERT(adapter);
DBG_ERROR("sxg: %s adapter->activated[%d]\n", __func__,
return STATUS_SUCCESS;
}
+int sxg_second_open(struct net_device * dev)
+{
+ struct adapter_t *adapter = (struct adapter_t*) netdev_priv(dev);
+
+ spin_lock_irqsave(&sxg_global.driver_lock, sxg_global.flags);
+ netif_start_queue(adapter->netdev);
+ adapter->state = ADAPT_UP;
+ adapter->linkstate = LINK_UP;
+
+ /* Re-enable interrupts */
+ SXG_ENABLE_ALL_INTERRUPTS(adapter);
+
+ netif_carrier_on(dev);
+ spin_unlock_irqrestore(&sxg_global.driver_lock, sxg_global.flags);
+ sxg_register_interrupt(adapter);
+ return (STATUS_SUCCESS);
+
+}
+
static void __devexit sxg_entry_remove(struct pci_dev *pcidev)
{
+ u32 mmio_start = 0;
+ u32 mmio_len = 0;
+
struct net_device *dev = pci_get_drvdata(pcidev);
- u32 mmio_start = 0;
- unsigned int mmio_len = 0;
struct adapter_t *adapter = (struct adapter_t *) netdev_priv(dev);
flush_scheduled_work();
sxg_free_resources(adapter);
ASSERT(adapter);
- DBG_ERROR("sxg: %s ENTER dev[%p] adapter[%p]\n", __func__, dev,
- adapter);
- mmio_start = pci_resource_start(pcidev, 0);
- mmio_len = pci_resource_len(pcidev, 0);
+ mmio_start = pci_resource_start(pcidev, 0);
+ mmio_len = pci_resource_len(pcidev, 0);
- DBG_ERROR("sxg: %s rel_region(0) start[%x] len[%x]\n", __func__,
- mmio_start, mmio_len);
- release_mem_region(mmio_start, mmio_len);
+ DBG_ERROR("sxg: %s rel_region(0) start[%x] len[%x]\n", __FUNCTION__,
+ mmio_start, mmio_len);
+ release_mem_region(mmio_start, mmio_len);
mmio_start = pci_resource_start(pcidev, 2);
mmio_len = pci_resource_len(pcidev, 2);
/* Disable interrupts */
SXG_DISABLE_ALL_INTERRUPTS(adapter);
+ netif_carrier_off(dev);
spin_unlock_irqrestore(&sxg_global.driver_lock, sxg_global.flags);
sxg_deregister_interrupt(adapter);
/* u32 SglOffset; */
u64 phys_addr;
unsigned long flags;
+ unsigned long queue_id=0;
SXG_TRACE(TRACE_SXG, SxgTraceBuffer, TRACE_NOISY, "DumbSgl",
pSgl, SxgSgl, 0, 0);
*/
SxgSgl->Sgl.NumberOfElements = 1;
+ /*
+ * Set ucode Queue ID based on bottom bits of destination TCP port.
+ * This Queue ID splits slowpath/dumb-nic packet processing across
+ * multiple threads on the card to improve performance. It is split
+ * using the TCP port to avoid out-of-order packets that can result
+ * from multithreaded processing. We use the destination port because
+ * we expect to be run on a server, so in nearly all cases the local
+ * port is likely to be constant (well-known server port) and the
+ * remote port is likely to be random. The exception to this is iSCSI,
+ * in which case we use the sport instead. Note
+ * that original attempt at XOR'ing source and dest port resulted in
+ * poor balance on NTTTCP/iometer applications since they tend to
+ * line up (even-even, odd-odd..).
+ */
+
+ if (skb->protocol == htons(ETH_P_IP)) {
+ struct iphdr *ip;
+
+ ip = ip_hdr(skb);
+ if ((ip->protocol == IPPROTO_TCP)&&(DataLength >= sizeof(
+ struct tcphdr))){
+ queue_id = ((ntohs(tcp_hdr(skb)->dest) == ISCSI_PORT) ?
+ (ntohs (tcp_hdr(skb)->source) &
+ SXG_LARGE_SEND_QUEUE_MASK):
+ (ntohs(tcp_hdr(skb)->dest) &
+ SXG_LARGE_SEND_QUEUE_MASK));
+ }
+ } else if (skb->protocol == htons(ETH_P_IPV6)) {
+ if ( (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP) && (DataLength >=
+ sizeof(struct tcphdr)) ) {
+ queue_id = ((ntohs(tcp_hdr(skb)->dest) == ISCSI_PORT) ?
+ (ntohs (tcp_hdr(skb)->source) &
+ SXG_LARGE_SEND_QUEUE_MASK):
+ (ntohs(tcp_hdr(skb)->dest) &
+ SXG_LARGE_SEND_QUEUE_MASK));
+ }
+ }
/* Grab the spinlock and acquire a command */
spin_lock_irqsave(&adapter->XmtZeroLock, flags);
* NOTE - See comments in SxgTcpOutput where we write
* to the XmtCmd register regarding CPU ID values and/or
* multiple commands.
+ * Top 16 bits specify queue_id. See comments about queue_id above
*/
- WRITE_REG(adapter->UcodeRegs[0].XmtCmd, 1, TRUE);
+ /* Four queues at the moment */
+ ASSERT((queue_id & ~SXG_LARGE_SEND_QUEUE_MASK) == 0);
+ WRITE_REG(adapter->UcodeRegs[0].XmtCmd, ((queue_id << 16) | 1), TRUE);
adapter->Stats.XmtQLen++; /* Stats within lock */
spin_unlock_irqrestore(&adapter->XmtZeroLock, flags);
SXG_TRACE(TRACE_SXG, SxgTraceBuffer, TRACE_NOISY, "XDumSgl2",
static void sxg_link_event(struct adapter_t *adapter)
{
struct sxg_hw_regs *HwRegs = adapter->HwRegs;
+ struct net_device *netdev = adapter->netdev;
enum SXG_LINK_STATE LinkState;
int status;
u32 Value;
sxg_link_state(adapter, LinkState);
DBG_ERROR("SXG: Link Alarm occurred. Link is %s\n",
((LinkState == SXG_LINK_UP) ? "UP" : "DOWN"));
+ if (LinkState == SXG_LINK_UP)
+ netif_carrier_on(netdev);
+ else
+ netif_carrier_off(netdev);
} else {
/*
* XXXTODO - Assuming Link Attention is only being generated
* Return
* None.
*/
-static void sxg_allocate_complete(struct adapter_t *adapter,
+static int sxg_allocate_complete(struct adapter_t *adapter,
void *VirtualAddress,
dma_addr_t PhysicalAddress,
u32 Length, enum sxg_buffer_type Context)
{
+ int status = 0;
SXG_TRACE(TRACE_SXG, SxgTraceBuffer, TRACE_NOISY, "AllocCmp",
adapter, VirtualAddress, Length, Context);
ASSERT(atomic_read(&adapter->pending_allocations));
switch (Context) {
case SXG_BUFFER_TYPE_RCV:
- sxg_allocate_rcvblock_complete(adapter,
+ status = sxg_allocate_rcvblock_complete(adapter,
VirtualAddress,
PhysicalAddress, Length);
break;
}
SXG_TRACE(TRACE_SXG, SxgTraceBuffer, TRACE_NOISY, "XAlocCmp",
adapter, VirtualAddress, Length, Context);
+
+ return status;
}
/*
adapter, Size, BufferType, 0);
return (STATUS_RESOURCES);
}
- sxg_allocate_complete(adapter, Buffer, pBuffer, Size, BufferType);
- status = STATUS_SUCCESS;
+ status = sxg_allocate_complete(adapter, Buffer, pBuffer, Size, BufferType);
SXG_TRACE(TRACE_SXG, SxgTraceBuffer, TRACE_NOISY, "XAlocMem",
adapter, Size, BufferType, status);
- return (status);
+ return status;
}
/*
*
* Return
*/
-static void sxg_allocate_rcvblock_complete(struct adapter_t *adapter,
+static int sxg_allocate_rcvblock_complete(struct adapter_t *adapter,
void *RcvBlock,
dma_addr_t PhysicalAddress,
u32 Length)
spin_unlock(&adapter->RcvQLock);
SXG_TRACE(TRACE_SXG, SxgTraceBuffer, TRACE_NOISY, "XAlRBlk",
adapter, RcvBlock, Length, 0);
- return;
+ return STATUS_SUCCESS;
fail:
/* Free any allocated resources */
if (RcvBlock) {
adapter, adapter->FreeRcvBufferCount,
adapter->FreeRcvBlockCount, adapter->AllRcvBlockCount);
adapter->Stats.NoMem++;
+ /* As allocation failed, free all previously allocated blocks..*/
+ //sxg_free_rcvblocks(adapter);
+
+ return STATUS_RESOURCES;
}
/*
projects / linux-beck.git / commitdiff