Menu
▾
▴
[Gc-linux-cvs] linux/drivers/net gc-net.c,1.19,1.20
|
From: <ham...@us...> - 2004-02-29 08:46:37
|
Update of /cvsroot/gc-linux/linux/drivers/net In directory sc8-pr-cvs1.sourceforge.net:/tmp/cvs-serv13964/linux/drivers/net Modified Files: gc-net.c Log Message: improofed internal structure, read comment in file Index: gc-net.c =================================================================== RCS file: /cvsroot/gc-linux/linux/drivers/net/gc-net.c,v retrieving revision 1.19 retrieving revision 1.20 diff -u -d -r1.19 -r1.20 --- gc-net.c 27 Feb 2004 00:06:33 -0000 1.19 +++ gc-net.c 29 Feb 2004 08:36:29 -0000 1.20 @@ -22,6 +22,9 @@ * $Id$ * * $Log$ + * Revision 1.20 2004/02/29 08:36:29 hamtitampti + * improofed internal structure, read comment in file + * * Revision 1.19 2004/02/27 00:06:33 hamtitampti * changed IRQ timing, but .. yes, the bus is too slow it seems * @@ -34,6 +37,84 @@ * */ +/* + Important Notes for Developers: + + currently speeds: + 1008 bytes from 192.168.0.2: icmp_seq=1 ttl=64 time=3.3 ms + 2008 bytes from 192.168.0.2: icmp_seq=1 ttl=64 time=6.0 ms + 5008 bytes from 192.168.0.2: icmp_seq=2 ttl=64 time=14.1 ms + 10008 bytes from 192.168.0.2: icmp_seq=0 ttl=64 time=27.0 ms + 20008 bytes from 192.168.0.2: icmp_seq=0 ttl=64 time=53.2 ms + 65000 bytes about 172ms (with Win2000) + + As you see, the Delay increases "linear" + this comes due to a Bandwith Problem with the EXI bus. + + note: + All ! i repeat ALL + packets are received with the + static void gc_input(struct net_device *dev) + and sent to the linux kernel then. + + BUT ! + the Transmit Back , the + static int gc_bba_start_xmit(struct sk_buff *skb, struct net_device *dev) + + As you see inside this function, i check, if the Transmit is currently in progress + if "yes" (priv->tx_lock) + i wait then a small time, and return with 1 ! + (otherwise, the linux kernel try's to resume the packet .. about 1mill / sec or so) + the Return 1; indicates to the linux kernel to resume the packet tramsission. + + if (priv->tx_lock) { + priv->stats.tx_dropped++; + udelay(1000); + netif_wake_queue(dev); + return 1; // This return will Resume the packet try + } + + So it happens, that the packet wich the linux Stack wants to send , enters multiple times the Xmit thing. + Until it time-out in Linux Packet buffer, logically. + + due to the tx_dropped, you see in ifconfig the dropped TX packets increasing. + Clear, there are more dropped packets then (as kernel resumes automaticall) + but i had no better idea. + + RX packets:582 errors:0 dropped:0 overruns:0 frame:0 + TX packets:475 errors:0 dropped:775 overruns:0 carrier:0 + + + Flood Performance: + 1K: 2531 packets transmitted, 2103 packets received, 16% packet loss + 2K: 1095 packets transmitted, 736 packets received, 32% packet loss + 4K: 886 packets transmitted, 472 packets received, 46% packet loss + 8K: 773 packets transmitted, 276 packets received, 64% packet loss + + ok + AT last: + + Even at flood, the kernel gets all RX (received frames) + but fails due to send back. + + + 100Mbit issue: + eth_outb(0x30, 0x2); + + This activates 100Mbit + Testet and operational. (market out in source) + + BAD: the ping / flood gets more worse. + as the packets arrive now faster, the bus almost crashes. + + BOYS, Write a Faster EXI driver, this is the maximum we can do now. + + hamtitampti + + + +*/ + #include <linux/module.h> #include <linux/kernel.h> #include <linux/types.h> @@ -71,6 +152,7 @@ spinlock_t lock; spinlock_t phylock; /* phy lock */ unsigned long lockflags; /* Lock flags */ + int tx_lock; // Statistical Data struct net_device_stats stats; @@ -157,8 +239,7 @@ while (len) { int tc = len; - if (tc > 4) - tc = 4; + if (tc > 4) tc = 4; exi_imm(channel, d, tc, mode, 0); exi_sync(channel); len-=tc; @@ -404,30 +485,6 @@ /* - -static inline u8 de600_read_status(struct net_device *dev) -{ - u8 status; - - outb_p(STATUS, DATA_PORT); - status = inb(STATUS_PORT); - outb_p(NULL_COMMAND | HI_NIBBLE, DATA_PORT); - - return status; -} - -static inline u8 de600_read_byte(unsigned char type, struct net_device *dev) -{ - // dev used by macros - u8 lo; - outb_p((type), DATA_PORT); - lo = ((unsigned char)inb(STATUS_PORT)) >> 4; - outb_p((type) | HI_NIBBLE, DATA_PORT); - return ((unsigned char)inb(STATUS_PORT) & (unsigned char)0xf0) | lo; -} -*/ - -/* * Open/initialize the board. This is called (in the current kernel) * after booting when 'ifconfig <dev->name> $IP_ADDR' is run (in rc.inet1). * @@ -463,12 +520,6 @@ static int gc_bba_close(struct net_device *dev) { BBA_DBG("gc_bba_close\n"); - -// //select_nic(); -// de600_put_command(RESET); -// de600_put_command(STOP_RESET); -// de600_put_command(0); -// //select_prn(); free_irq(dev->irq, dev); return 0; } @@ -480,15 +531,6 @@ return &priv->stats; } -static inline void trigger_interrupt(struct net_device *dev) -{ -// de600_put_command(FLIP_IRQ); -// //select_prn(); -// DE600_SLOW_DOWN; -// //select_nic(); -// de600_put_command(0); -} - /* * Copy a buffer to the adapter transmit page memory. * Start sending. @@ -497,17 +539,29 @@ static int gc_bba_start_xmit(struct sk_buff *skb, struct net_device *dev) { struct gc_private *priv = (struct gc_private *)dev->priv; - unsigned char Reg0; + + netif_stop_queue(dev); + + if (priv->tx_lock) { + priv->stats.tx_dropped++; + udelay(1000); + netif_wake_queue(dev); + return 1; // This return will Resume the packet try + } + + priv->tx_lock = 1; - spin_lock_irqsave(&priv->lock, priv->lockflags); - netif_stop_queue(dev); + netif_wake_queue(dev); + spin_lock_irqsave(&priv->lock, priv->lockflags); + //while(eth_inb(0x3a)&0x1); // TX Fifo Page to 0 + eth_outb(0xf, 0); eth_outb(0xe, 0); - + //printk("XMIT packet : len %d\n",skb->len); /* We send using the FIFO mode @@ -516,13 +570,16 @@ BBA_DBG("gc_bba_start_xmit:len=%d, page %d/%d\n", skb->len, tx_fifo_in, free_tx_pages); // We set the packetbuffer to beginning... - + + eth_outb(0x3e, skb->len&0xff); + eth_outb(0x3f, (skb->len&0x0f00)>>8); + unsigned int val=0xC0004800; // register 0x48 is the output queue exi_select(0, 2, 5); exi_imm(0, &val, 4, EXI_WRITE, 0); exi_sync(0); - + // Send Data from skb buffer to Network Driver now exi_imm_ex(0, skb->data, skb->len, EXI_WRITE); @@ -544,13 +601,13 @@ val |=4; eth_outb(0, val); - spin_unlock_irqrestore(&priv->lock, priv->lockflags); + dev->trans_start = jiffies; - dev_kfree_skb(skb); - - priv->stats.tx_bytes++; + priv->stats.tx_bytes += skb->len; priv->stats.tx_packets++; + + dev_kfree_skb(skb); return 0; } @@ -570,7 +627,7 @@ /* Input Flow concept: Take a look to Hardware spec page 35 */ - +/* eth_outb(0x3a, 2); if (eth_inb(0x3a) & 2) { @@ -578,7 +635,7 @@ priv->stats.rx_missed_errors++; return ; } - +*/ /* Receive Buffer Write Page Pointer: Current receive write page pointer. The MSB is the Reg17h.3 bit. The LSB is the Reg16h.0 bit. This register is controlled by @@ -597,7 +654,6 @@ p_read = eth_inb(0x18); p_read |= (eth_inb(0x19)&0x0f) << 8; - if (p_read == p_write) return; //printk("Received a packet .. (start)\n"); @@ -686,13 +742,22 @@ dev->last_rx = jiffies; priv->stats.rx_packets ++; /* count all receives */ - priv->stats.rx_bytes ++; /* count all received bytes */ + priv->stats.rx_bytes += skb->len; /* count all received bytes */ gc_input(dev); return ; } +#define BBA_IRQ_FRAGI 0x01 +#define BBA_IRQ_RI 0x02 +#define BBA_IRQ_TI 0x04 +#define BBA_IRQ_REI 0x08 +#define BBA_IRQ_TEI 0x10 +#define BBA_IRQ_FIFOEI 0x20 +#define BBA_IRQ_BUSEI 0x40 +#define BBA_IRQ_RBFI 0x80 + static void inline gcif_service(struct net_device *dev) { @@ -705,61 +770,73 @@ int status = inb9 & inb8; BBA_DBG("gcif_service: %08x %08x status %08x\n", inb8, inb9, status); - + if (!status) { eth_outb(9, 0xff); BBA_DBG("?? GC irq but no irq ??\n"); } - - if (status & 4) + + if (status & BBA_IRQ_FRAGI) // 0x1 { - eth_outb(9, 4); - // We clear the IRQ - netif_wake_queue(dev); /* allow more packets into adapter */ - - - // TX Transmisstion compleated ... - - /* - int s = eth_inb(4); - if (s) // should not occur, since 4 == TX OK - BBA_DBG("tx error %02x\n", s); - */ + eth_outb(9, BBA_IRQ_FRAGI); + BBA_DBG("Fragmentet Interrupt\n"); } - if (status & 2) + if (status & BBA_IRQ_RI) // 0x2 { // We clear the IRQ - eth_outb(9, 2); + eth_outb(9, BBA_IRQ_RI); spin_lock_irqsave(&priv->lock, priv->lockflags); gc_input(dev); spin_unlock_irqrestore(&priv->lock, priv->lockflags); } + + if (status & BBA_IRQ_TI) // 0x4 + { + eth_outb(9, BBA_IRQ_TI); + // We clear the IRQ + netif_wake_queue(dev); /* allow more packets into adapter */ + priv->tx_lock = 0; + // TX Transmisstion compleated ... + } - if (status & 8) + if (status & BBA_IRQ_REI) // 0x8 { - eth_outb(9, 8); - //printk("receive error :(\n"); + eth_outb(9, BBA_IRQ_REI); + BBA_DBG("receive error :(\n"); + priv->stats.rx_frame_errors++; } - if (status & 0x10) + if (status & BBA_IRQ_TEI) // 0x10 { - eth_outb(9, 0x10); + eth_outb(9, BBA_IRQ_TEI); netif_wake_queue(dev); /* allow more packets into adapter */ - //printk("tx error\n"); + priv->tx_lock = 0; + BBA_DBG("tx error\n"); + priv->stats.tx_errors++; } - if (status & 0x20) + + if (status & BBA_IRQ_FIFOEI) // 0x20 { - eth_outb(9, 0x20); - BBA_DBG("rx fifo error\n"); + eth_outb(9, BBA_IRQ_FIFOEI); + BBA_DBG("rx/tx fifo error\n"); + adapter_init(dev); + priv->stats.rx_errors++; } - if (status & 0x80) + + if (status & BBA_IRQ_BUSEI) // 0x40 { - eth_outb(9, 0x80); - //printk("rx overflow!\n"); - gc_input(dev); - + eth_outb(9, BBA_IRQ_BUSEI); + BBA_DBG("Bus Error\n"); + } + + if (status & BBA_IRQ_RBFI) // 0x80 + { + eth_outb(9, BBA_IRQ_RBFI); + BBA_DBG("rx overflow!\n"); + //gc_input(dev); + priv->stats.rx_over_errors++; // RWP eth_outb(0x16, GBA_RX_RWP); eth_outb(0x17, 0x0); @@ -771,16 +848,9 @@ // RHBP eth_outb(0x1a, GBA_RX_RHBP); eth_outb(0x1b, 0); - - - } - if (status & ~(0xBE)) - { - eth_outb(9, status & ~0xBE); - BBA_DBG("status %02x\n", status & ~0xBE); } -// eth_outb(9, status); + } @@ -795,8 +865,6 @@ struct net_device *dev = (struct net_device *)dev_id; struct gc_private *priv = (struct gc_private *)dev->priv; - int retrig = 0; - /* This might just as well be deleted now, no crummy drivers present :-) */ if ((dev == NULL) || (dev->irq != irq)) { BBA_DBG(KERN_ERR "%s: bogus interrupt %d\n", dev?dev->name:"GC_BBA", irq); @@ -820,9 +888,6 @@ have_irq(ch * 3 + EXI_EVENT_IRQ); } - if (retrig) - trigger_interrupt(dev); - spin_unlock(&priv->lock); return IRQ_HANDLED; @@ -958,7 +1023,7 @@ BBA_DBG("initializing BBA...\n"); - + priv->tx_lock = 0; eth_outb(0x60, 0); // unknown udelay(10000); @@ -982,7 +1047,7 @@ eth_outb(1, 0x10|PACKETS_PER_IRQ | BBA_PROMISC); eth_outb(0x14, 0x0); - eth_outb(0x15, 0x8); + eth_outb(0x15, 0x6); eth_outb(0x50, 0x80); @@ -1037,7 +1102,7 @@ eth_outb(8, 0xFF); // enable all IRQs eth_outb(9, 0xFF); // clear all irqs - //eth_outb(0x30, 0x2); // 100 Mbit ? + //eth_outb(0x30, 0x2); // 100 Mbit ? -- FATAL , IF speed so high, EXI totally overloaded BBA_DBG("after all: irq mask %x %x\n", eth_inb(8), eth_inb(9)); @@ -1077,24 +1142,22 @@ eth_exi_outb(2, 0); s = eth_exi_inb(3); - if (s & 0x80) + if (s & 0x08) { -// BBA_DBG("GC_IRQ service.\n"); - eth_exi_outb(3, 0x80); - gcif_service(dev); + BBA_DBG("GCIF - EXI - HASH function\n"); + eth_exi_outb(3, 0x08); eth_exi_outb(2, 0xF8); return; } - if (s & 0x40) + + if (s & 0x10) { - eth_exi_outb(3, 0x40); + BBA_DBG("GCIF - EXI - patchtru!\n"); + eth_exi_outb(3, 0x10); eth_exi_outb(2, 0xF8); -// BBA_DBG("GCIF - EXI - 0x40!\n"); - adapter_init(dev); return; - -// return; } + if (s & 0x20) { BBA_DBG("GCIF - EXI - CMDERR!\n"); @@ -1102,20 +1165,25 @@ eth_exi_outb(2, 0xF8); return; } - if (s & 0x10) + + if (s & 0x40) { - BBA_DBG("GCIF - EXI - patchtru!\n"); - eth_exi_outb(3, 0x10); + eth_exi_outb(3, 0x40); eth_exi_outb(2, 0xF8); + BBA_DBG("GCIF - EXI - 0x40!\n"); + adapter_init(dev); return; - } - if (s & 0x08) + } + + if (s & 0x80) { - BBA_DBG("GCIF - EXI - HASH function\n"); - eth_exi_outb(3, 0x08); + BBA_DBG("GC_IRQ service.\n"); + eth_exi_outb(3, 0x80); + gcif_service(dev); eth_exi_outb(2, 0xF8); return; } + // printk("GCIF - EXI - ?? %02x\n", s); eth_exi_outb(2, 0xF8); } |
