Description:
For DLink Fiber NIC, Linux 2.4.22 ships with driver version 1.19,
whereas, Linux 2.6.x ship with driver version 1.17.
The following patch upgrades the 2.6.x driver to include changes (and
bug fixes done until 1.19b).
These fixes are (copied from the driver):
1.18 2002/11/07 New tx scheme, adaptive tx_coalesce.
Remove tx_coalesce option.
1.19 2003/12/16 Fix problem parsing the eeprom on big endian
systems. ([email protected])
Disclaimer:
Since I returned my DLink NIC to the store I couldn't test it
thoroughly. It seemed to work just as well as v1.17. However, both
version made the NIC hang after a few minutes.
--- drivers/net/dl2k.c.orig 2006-09-07 11:34:20.000000000 +0300
+++ drivers/net/dl2k.c 2006-09-07 11:34:24.000000000 +0300
@@ -1,7 +1,8 @@
/* D-Link DL2000-based Gigabit Ethernet Adapter Linux driver */
/*
Copyright (c) 2001, 2002 by D-Link Corporation
- Written by Edward Peng.<[email protected]>
+ Copyright (c) 2003 by Alpha Networks
+ Written by Edward Peng.<[email protected]>
Created 03-May-2001, base on Linux' sundance.c.
This program is free software; you can redistribute it and/or modify
@@ -9,10 +10,54 @@
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
*/
+/*
+ Rev Date Description
+ ==========================================================================
+ 0.01 2001/05/03 Created DL2000-based linux driver
+ 0.02 2001/05/21 Added VLAN and hardware checksum support.
+ 1.00 2001/06/26 Added jumbo frame support.
+ 1.01 2001/08/21 Added two parameters, rx_coalesce and rx_timeout.
+ 1.02 2001/10/08 Supported fiber media.
+ Added flow control parameters.
+ 1.03 2001/10/12 Changed the default media to 1000mbps_fd for
+ the fiber devices.
+ 1.04 2001/11/08 Fixed Tx stopped when tx very busy.
+ 1.05 2001/11/22 Fixed Tx stopped when unidirectional tx busy.
+ 1.06 2001/12/13 Fixed disconnect bug at 10Mbps mode.
+ Fixed tx_full flag incorrect.
+ Added tx_coalesce paramter.
+ 1.07 2002/01/03 Fixed miscount of RX frame error.
+ 1.08 2002/01/17 Fixed the multicast bug.
+ 1.09 2002/03/07 Move rx-poll-now to re-fill loop.
+ Added rio_timer() to watch rx buffers.
+ 1.10 2002/04/16 Fixed miscount of carrier error.
+ 1.11 2002/05/23 Added ISR schedule scheme
+ Fixed miscount of rx frame error for DGE-550SX.
+ Fixed VLAN bug.
+ 1.12 2002/06/13 Lock tx_coalesce=1 on 10/100Mbps mode.
+ 1.13 2002/08/13 1. Fix disconnection (many tx:carrier/rx:frame
+ errs) with some mainboards.
+ 2. Use definition "DRV_NAME" "DRV_VERSION"
+ "DRV_RELDATE" for flexibility.
+ 1.14 2002/08/14 Support ethtool.
+ 1.15 2002/08/27 Changed the default media to Auto-Negotiation
+ for the fiber devices.
+ 1.16 2002/09/04 More power down time for fiber devices auto-
+ negotiation.
+ Fix disconnect bug after ifup and ifdown.
+ 1.17 2002/10/03 Fix RMON statistics overflow.
+ Always use I/O mapping to access eeprom,
+ avoid system freezing with some chipsets.
+ 1.18 2002/11/07 New tx scheme, adaptive tx_coalesce.
+ Remove tx_coalesce option.
+ 1.19 2003/12/16 Fix problem parsing the eeprom on big endian
+ systems. ([email protected])
+ 1.19b 2006/07/01 made compile on kernel 2.6.15. ([email protected])
+*/
#define DRV_NAME "D-Link DL2000-based linux driver"
-#define DRV_VERSION "v1.18"
-#define DRV_RELDATE "2006/06/27"
+#define DRV_VERSION "v1.19b"
+#define DRV_RELDATE "2006/07/01"
#include "dl2k.h"
#include <linux/dma-mapping.h>
@@ -26,9 +71,8 @@
static int tx_flow=-1;
static int rx_flow=-1;
static int copy_thresh;
-static int rx_coalesce=10; /* Rx frame count each interrupt */
+static int rx_coalesce=0; /* Rx frame count each interrupt */
static int rx_timeout=200; /* Rx DMA wait time in 640ns increments */
-static int tx_coalesce=16; /* HW xmit count each TxDMAComplete */
MODULE_AUTHOR ("Edward Peng");
@@ -43,12 +87,11 @@
module_param(copy_thresh, int, 0);
module_param(rx_coalesce, int, 0); /* Rx frame count each interrupt */
module_param(rx_timeout, int, 0); /* Rx DMA wait time in 64ns increments */
-module_param(tx_coalesce, int, 0); /* HW xmit count each TxDMAComplete */
/* Enable the default interrupts */
#define DEFAULT_INTR (RxDMAComplete | HostError | IntRequested | TxDMAComplete| \
- UpdateStats | LinkEvent)
+ UpdateStats | LinkEvent|TxComplete)
#define EnableInt() \
writew(DEFAULT_INTR, ioaddr + IntEnable)
@@ -59,11 +102,13 @@
static void rio_timer (unsigned long data);
static void rio_tx_timeout (struct net_device *dev);
static void alloc_list (struct net_device *dev);
+static void tx_poll (unsigned long data);
static int start_xmit (struct sk_buff *skb, struct net_device *dev);
static irqreturn_t rio_interrupt (int irq, void *dev_instance, struct pt_regs *regs);
-static void rio_free_tx (struct net_device *dev, int irq);
+static void rio_free_tx (struct net_device *dev);
static void tx_error (struct net_device *dev, int tx_status);
-static int receive_packet (struct net_device *dev);
+static void rx_poll (unsigned long data);
+static void refill_rx (struct net_device *dev);
static void rio_error (struct net_device *dev, int int_status);
static int change_mtu (struct net_device *dev, int new_mtu);
static void set_multicast (struct net_device *dev);
@@ -136,10 +181,11 @@
np->pdev = pdev;
spin_lock_init (&np->tx_lock);
spin_lock_init (&np->rx_lock);
+ tasklet_init(&np->tx_tasklet, tx_poll, (unsigned long) dev);
+ tasklet_init(&np->rx_tasklet, rx_poll, (unsigned long) dev);
/* Parse manual configuration */
np->an_enable = 1;
- np->tx_coalesce = 1;
if (card_idx < MAX_UNITS) {
if (media[card_idx] != NULL) {
np->an_enable = 0;
@@ -193,10 +239,6 @@
np->tx_flow = (tx_flow == 0) ? 0 : 1;
np->rx_flow = (rx_flow == 0) ? 0 : 1;
- if (tx_coalesce < 1)
- tx_coalesce = 1;
- else if (tx_coalesce > TX_RING_SIZE-1)
- tx_coalesce = TX_RING_SIZE - 1;
}
dev->open = &rio_open;
dev->hard_start_xmit = &start_xmit;
@@ -262,9 +304,6 @@
dev->name, np->name,
dev->dev_addr[0], dev->dev_addr[1], dev->dev_addr[2],
dev->dev_addr[3], dev->dev_addr[4], dev->dev_addr[5], irq);
- if (tx_coalesce > 1)
- printk(KERN_INFO "tx_coalesce:\t%d packets\n",
- tx_coalesce);
if (np->coalesce)
printk(KERN_INFO "rx_coalesce:\t%d packets\n"
KERN_INFO "rx_timeout: \t%d ns\n",
@@ -335,8 +374,9 @@
#endif
/* Read eeprom */
for (i = 0; i < 128; i++) {
- ((u16 *) sromdata)[i] = le16_to_cpu (read_eeprom (ioaddr, i));
+ ((u16 *) sromdata)[i] = cpu_to_le16 (read_eeprom (ioaddr, i));
}
+ psrom->crc = le32_to_cpu(psrom->crc);
#ifdef MEM_MAPPING
ioaddr = dev->base_addr;
#endif
@@ -351,7 +391,7 @@
for (i = 0; i < 6; i++)
dev->dev_addr[i] = psrom->mac_addr[i];
- /* Parse Software Information Block */
+ /* Parse Software Infomation Block */
i = 0x30;
psib = (u8 *) sromdata;
do {
@@ -401,7 +441,7 @@
int i;
u16 macctrl;
- i = request_irq (dev->irq, &rio_interrupt, IRQF_SHARED, dev->name, dev);
+ i = request_irq (dev->irq, &rio_interrupt, SA_SHIRQ, dev->name, dev);
if (i)
return i;
@@ -434,9 +474,12 @@
writeb (0x30, ioaddr + RxDMABurstThresh);
writeb (0x30, ioaddr + RxDMAUrgentThresh);
writel (0x0007ffff, ioaddr + RmonStatMask);
+
/* clear statistics */
clear_stats (dev);
+ atomic_set(&np->tx_desc_lock, 0);
+
/* VLAN supported */
if (np->vlan) {
/* priority field in RxDMAIntCtrl */
@@ -521,17 +564,31 @@
np->timer.expires = jiffies + next_tick;
add_timer(&np->timer);
}
-
+
static void
rio_tx_timeout (struct net_device *dev)
{
+ struct netdev_private *np = dev->priv;
long ioaddr = dev->base_addr;
- printk (KERN_INFO "%s: Tx timed out (%4.4x), is buffer full?\n",
+ printk (KERN_INFO "%s: Tx timed out (%4.4x), "
+ "clean up tx queue to restart..\n",
dev->name, readl (ioaddr + TxStatus));
- rio_free_tx(dev, 0);
+ printk (KERN_INFO "%s: cur_tx=%lx cur_task=%lx old_tx=%lx "
+ "TFDListPtr=%08x tx_full=%d",
+ dev->name, np->cur_tx, np->cur_task, np->old_tx,
+ readl(ioaddr + TFDListPtr0),
+ netif_queue_stopped(dev));
+ writew(0, ioaddr + IntEnable);
+ writew(0, ioaddr + TFDListPtr0);
+ writew(0, ioaddr + TFDListPtr1);
+ tasklet_disable(&np->tx_tasklet);
+ rio_free_tx(dev);
dev->if_port = 0;
dev->trans_start = jiffies;
+ tasklet_enable(&np->tx_tasklet);
+ writew(DEFAULT_INTR, ioaddr + IntEnable);
+ return;
}
/* allocate and initialize Tx and Rx descriptors */
@@ -543,6 +600,7 @@
np->cur_rx = np->cur_tx = 0;
np->old_rx = np->old_tx = 0;
+ np->cur_task = 0;
np->rx_buf_sz = (dev->mtu <= 1500 ? PACKET_SIZE : dev->mtu + 32);
/* Initialize Tx descriptors, TFDListPtr leaves in start_xmit(). */
@@ -592,6 +650,31 @@
return;
}
+static void tx_poll (unsigned long data)
+{
+ struct net_device *dev = (struct net_device*) data;
+ struct netdev_private *np = dev->priv;
+ unsigned head = np->cur_task % TX_RING_SIZE;
+ struct netdev_desc *txdesc =
+ &np->tx_ring[(np->cur_tx + TX_RING_SIZE - 1) % TX_RING_SIZE];
+
+ /* Indicate the latest descriptor of tx ring */
+ txdesc->status |= cpu_to_le64 (TxDMAIndicate);
+
+ while ((np->cur_task + TX_RING_SIZE - np->cur_tx) % TX_RING_SIZE > 0) {
+ txdesc = &np->tx_ring[np->cur_task];
+ txdesc->status &= cpu_to_le64 (~TFDDone);
+ np->cur_task = (np->cur_task + 1) % TX_RING_SIZE;
+ }
+
+ if (readl (dev->base_addr + TFDListPtr0) == 0) {
+ writel (np->tx_ring_dma + head * sizeof (struct netdev_desc),
+ dev->base_addr + TFDListPtr0);
+ writel (0, dev->base_addr + TFDListPtr1);
+ }
+
+ return;
+}
static int
start_xmit (struct sk_buff *skb, struct net_device *dev)
{
@@ -630,35 +713,40 @@
/* DL2K bug: DMA fails to get next descriptor ptr in 10Mbps mode
* Work around: Always use 1 descriptor in 10Mbps mode */
- if (entry % np->tx_coalesce == 0 || np->speed == 10)
- txdesc->status = cpu_to_le64 (entry | tfc_vlan_tag |
+ if (np->speed == 10) {
+ /* Note the order! Stop queue before hardware processing
+ * this descriptor */
+ netif_stop_queue (dev);
+ txdesc->status = cpu_to_le64 (entry | tfc_vlan_tag |
WordAlignDisable |
- TxDMAIndicate |
+ TxDMAIndicate | TxComplete |
(1 << FragCountShift));
- else
- txdesc->status = cpu_to_le64 (entry | tfc_vlan_tag |
+ /* TxDMAPollNow */
+ writel(readl(ioaddr + DMACtrl) | 0x00001000, ioaddr + DMACtrl);
+ if (readl (dev->base_addr + TFDListPtr0) == 0) {
+ writel (np->tx_ring_dma +
+ entry * sizeof (struct netdev_desc),
+ dev->base_addr + TFDListPtr0);
+ writel (0, dev->base_addr + TFDListPtr1);
+ }
+ np->cur_tx = np->cur_task = (np->cur_tx + 1) % TX_RING_SIZE;
+ } else {
+ np->cur_tx = (np->cur_tx + 1) % TX_RING_SIZE;
+ mb();
+ txdesc->status = cpu_to_le64 (entry | tfc_vlan_tag | TFDDone |
WordAlignDisable |
(1 << FragCountShift));
-
- /* TxDMAPollNow */
- writel (readl (ioaddr + DMACtrl) | 0x00001000, ioaddr + DMACtrl);
- /* Schedule ISR */
- writel(10000, ioaddr + CountDown);
- np->cur_tx = (np->cur_tx + 1) % TX_RING_SIZE;
- if ((np->cur_tx - np->old_tx + TX_RING_SIZE) % TX_RING_SIZE
- < TX_QUEUE_LEN - 1 && np->speed != 10) {
- /* do nothing */
- } else if (!netif_queue_stopped(dev)) {
- netif_stop_queue (dev);
+ tasklet_schedule (&np->tx_tasklet);
+ /* Schedule ISR */
+ if ((np->cur_tx - np->old_tx + TX_RING_SIZE) % TX_RING_SIZE
+ < TX_QUEUE_LEN - 1) {
+ /* do nothing */
+ } else if (!netif_queue_stopped(dev)) {
+ netif_stop_queue (dev);
+ }
}
- /* The first TFDListPtr */
- if (readl (dev->base_addr + TFDListPtr0) == 0) {
- writel (np->tx_ring_dma + entry * sizeof (struct netdev_desc),
- dev->base_addr + TFDListPtr0);
- writel (0, dev->base_addr + TFDListPtr1);
- }
-
+
/* NETDEV WATCHDOG timer */
dev->trans_start = jiffies;
return 0;
@@ -684,16 +772,22 @@
break;
handled = 1;
/* Processing received packets */
- if (int_status & RxDMAComplete)
- receive_packet (dev);
+ if (int_status & RxDMAComplete) {
+ writew(DEFAULT_INTR & ~(RxDMAComplete | RxComplete),
+ ioaddr + IntEnable);
+ if (np->budget < 0) {
+ np->budget = RX_BUDGET;
+ }
+ tasklet_schedule (&np->rx_tasklet);
+ }
/* TxDMAComplete interrupt */
- if ((int_status & (TxDMAComplete|IntRequested))) {
+ if ((int_status & (TxDMAComplete|IntRequested|TxComplete))) {
int tx_status;
tx_status = readl (ioaddr + TxStatus);
if (tx_status & 0x01)
tx_error (dev, tx_status);
/* Free used tx skbuffs */
- rio_free_tx (dev, 1);
+ rio_free_tx (dev);
}
/* Handle uncommon events */
@@ -701,26 +795,28 @@
(HostError | LinkEvent | UpdateStats))
rio_error (dev, int_status);
}
- if (np->cur_tx != np->old_tx)
- writel (100, ioaddr + CountDown);
+ writel(5000, ioaddr + CountDown);
return IRQ_RETVAL(handled);
}
static void
-rio_free_tx (struct net_device *dev, int irq)
+rio_free_tx (struct net_device *dev)
{
struct netdev_private *np = netdev_priv(dev);
int entry = np->old_tx % TX_RING_SIZE;
- int tx_use = 0;
unsigned long flag = 0;
+ int irq = in_interrupt();
+
+ if (atomic_read(&np->tx_desc_lock))
+ return;
+ atomic_inc(&np->tx_desc_lock);
if (irq)
spin_lock(&np->tx_lock);
else
spin_lock_irqsave(&np->tx_lock, flag);
-
/* Free used tx skbuffs */
- while (entry != np->cur_tx) {
+ while (entry != np->cur_task) {
struct sk_buff *skb;
if (!(np->tx_ring[entry].status & TFDDone))
@@ -736,8 +832,8 @@
np->tx_skbuff[entry] = NULL;
entry = (entry + 1) % TX_RING_SIZE;
- tx_use++;
}
+
if (irq)
spin_unlock(&np->tx_lock);
else
@@ -746,12 +842,19 @@
/* If the ring is no longer full, clear tx_full and
call netif_wake_queue() */
-
- if (netif_queue_stopped(dev) &&
+ if (np->speed != 10) {
+ if (netif_queue_stopped(dev) &&
((np->cur_tx - np->old_tx + TX_RING_SIZE) % TX_RING_SIZE
- < TX_QUEUE_LEN - 1 || np->speed == 10)) {
+ < TX_QUEUE_LEN - 1)) {
netif_wake_queue (dev);
+ }
+ }
+ else {
+ if (netif_queue_stopped(dev) &&
+ np->cur_tx == np->old_tx)
+ netif_wake_queue(dev);
}
+ atomic_dec(&np->tx_desc_lock);
}
static void
@@ -782,7 +885,7 @@
break;
mdelay (1);
}
- rio_free_tx (dev, 1);
+ rio_free_tx (dev);
/* Reset TFDListPtr */
writel (np->tx_ring_dma +
np->old_tx * sizeof (struct netdev_desc),
@@ -816,28 +919,36 @@
writel (readw (dev->base_addr + MACCtrl) | TxEnable, ioaddr + MACCtrl);
}
-static int
-receive_packet (struct net_device *dev)
+static void rx_poll (unsigned long data)
{
- struct netdev_private *np = netdev_priv(dev);
+ struct net_device *dev = (struct net_device*) data;
+ struct netdev_private *np = (struct netdev_private *) dev->priv;
int entry = np->cur_rx % RX_RING_SIZE;
- int cnt = 30;
-
+ int cnt = np->budget;
+ long ioaddr = dev->base_addr;
+ int received = 0;
/* If RFDDone, FrameStart and FrameEnd set, there is a new packet in. */
while (1) {
struct netdev_desc *desc = &np->rx_ring[entry];
int pkt_len;
u64 frame_status;
- if (!(desc->status & RFDDone) ||
- !(desc->status & FrameStart) || !(desc->status & FrameEnd))
- break;
+ if (--cnt < 0) {
+ np->cur_rx = entry;
+ goto not_done;
+ }
/* Chip omits the CRC. */
pkt_len = le64_to_cpu (desc->status & 0xffff);
frame_status = le64_to_cpu (desc->status);
- if (--cnt < 0)
+
+ if (!(desc->status & RFDDone) ||
+ !(desc->status & FrameStart) || !(desc->status & FrameEnd))
break;
+
+
+ pci_dma_sync_single_for_device (np->pdev, desc->fraginfo, np->rx_buf_sz,
+ PCI_DMA_FROMDEVICE);
/* Update rx error statistics, drop packet. */
if (frame_status & RFS_Errors) {
np->stats.rx_errors++;
@@ -891,9 +1002,37 @@
dev->last_rx = jiffies;
}
entry = (entry + 1) % RX_RING_SIZE;
+ received++;
}
- spin_lock(&np->rx_lock);
np->cur_rx = entry;
+ refill_rx (dev);
+ np->budget -= received;
+ writew (DEFAULT_INTR, ioaddr + IntEnable);
+ return;
+
+not_done:
+ refill_rx (dev);
+ if (!received)
+ received = 1;
+ np->budget -= received;
+ if (np->budget <= 0)
+ np->budget = RX_BUDGET;
+ tasklet_schedule (&np->rx_tasklet);
+ return;
+}
+
+static void refill_rx (struct net_device *dev)
+{
+ struct netdev_private *np = dev->priv;
+ int entry;
+ int irq = in_interrupt();
+ unsigned long flag = 0;
+
+ if (irq)
+ spin_lock(&np->rx_lock);
+ else
+ spin_lock_irqsave(&np->rx_lock, flag);
+
/* Re-allocate skbuffs to fill the descriptor ring */
entry = np->old_rx;
while (entry != np->cur_rx) {
@@ -924,8 +1063,12 @@
entry = (entry + 1) % RX_RING_SIZE;
}
np->old_rx = entry;
- spin_unlock(&np->rx_lock);
- return 0;
+
+ if (irq)
+ spin_unlock(&np->rx_lock);
+ else
+ spin_unlock_irqrestore(&np->rx_lock, flag);
+ return;
}
static void
@@ -943,10 +1086,6 @@
mii_get_media_pcs (dev);
else
mii_get_media (dev);
- if (np->speed == 1000)
- np->tx_coalesce = tx_coalesce;
- else
- np->tx_coalesce = 1;
macctrl = 0;
macctrl |= (np->vlan) ? AutoVLANuntagging : 0;
macctrl |= (np->full_duplex) ? DuplexSelect : 0;
@@ -984,10 +1123,10 @@
{
long ioaddr = dev->base_addr;
struct netdev_private *np = netdev_priv(dev);
+ unsigned int stat_reg;
#ifdef MEM_MAPPING
int i;
#endif
- unsigned int stat_reg;
/* All statistics registers need to be acknowledged,
else statistic overflow could cause problems */
@@ -1028,7 +1167,6 @@
readw (ioaddr + BcstFramesXmtdOk);
readw (ioaddr + MacControlFramesXmtd);
readw (ioaddr + FramesWEXDeferal);
-
#ifdef MEM_MAPPING
for (i = 0x100; i <= 0x150; i += 4)
readl (ioaddr + i);
@@ -1047,7 +1185,7 @@
long ioaddr = dev->base_addr;
#ifdef MEM_MAPPING
int i;
-#endif
+#endif
/* All statistics registers need to be acknowledged,
else statistic overflow could cause problems */
@@ -1086,7 +1224,7 @@
#ifdef MEM_MAPPING
for (i = 0x100; i <= 0x150; i += 4)
readl (ioaddr + i);
-#endif
+#endif
readw (ioaddr + TxJumboFrames);
readw (ioaddr + RxJumboFrames);
readw (ioaddr + TCPCheckSumErrors);
@@ -1752,6 +1890,8 @@
/* Stop Tx and Rx logics */
writel (TxDisable | RxDisable | StatsDisable, ioaddr + MACCtrl);
synchronize_irq (dev->irq);
+ tasklet_kill (&np->tx_tasklet);
+ tasklet_kill (&np->rx_tasklet);
free_irq (dev->irq, dev);
del_timer_sync (&np->timer);
> @@ -335,8 +374,9 @@
> #endif
> /* Read eeprom */
> for (i = 0; i < 128; i++) {
> - ((u16 *) sromdata)[i] = le16_to_cpu (read_eeprom (ioaddr, i));
> + ((u16 *) sromdata)[i] = cpu_to_le16 (read_eeprom (ioaddr, i));
> }
> + psrom->crc = le32_to_cpu(psrom->crc);
this looks wrong, the data comes from the hw as le, so le*_to_cpu()
sounds the right direction
> @@ -401,7 +441,7 @@
> int i;
> u16 macctrl;
>
> - i = request_irq (dev->irq, &rio_interrupt, IRQF_SHARED, dev->name, dev);
> + i = request_irq (dev->irq, &rio_interrupt, SA_SHIRQ, dev->name, dev);
> if (i)
> return i;
this is backing out a fix/conversion to the new API. Bad.
>
> @@ -434,9 +474,12 @@
> writeb (0x30, ioaddr + RxDMABurstThresh);
> writeb (0x30, ioaddr + RxDMAUrgentThresh);
> writel (0x0007ffff, ioaddr + RmonStatMask);
> +
> /* clear statistics */
> clear_stats (dev);
>
> + atomic_set(&np->tx_desc_lock, 0);
I'm quite scared by this naming; it suggests home-brew locking
> dev->trans_start = jiffies;
> + tasklet_enable(&np->tx_tasklet);
> + writew(DEFAULT_INTR, ioaddr + IntEnable);
> + return;
this looks like a PCI posting bug
> -rio_free_tx (struct net_device *dev, int irq)
> +rio_free_tx (struct net_device *dev)
> {
> struct netdev_private *np = netdev_priv(dev);
> int entry = np->old_tx % TX_RING_SIZE;
> - int tx_use = 0;
> unsigned long flag = 0;
> + int irq = in_interrupt();
eeeeep
> +
> + if (atomic_read(&np->tx_desc_lock))
> + return;
> + atomic_inc(&np->tx_desc_lock);
and yes.. it is broken self made locking....
there is a nice race between the _read and the _inc here.
>
> if (irq)
> spin_lock(&np->tx_lock);
> else
> spin_lock_irqsave(&np->tx_lock, flag);
double eeeep
this is wrong to do with in_interrupt() as gating factor!
Always doing the irqsave() is fine btw
Arjan van de Ven <[email protected]> :
[...]
> > dev->trans_start = jiffies;
> > + tasklet_enable(&np->tx_tasklet);
> > + writew(DEFAULT_INTR, ioaddr + IntEnable);
> > + return;
>
> this looks like a PCI posting bug
Btw tx_tasklet is introduced in the relevant struct in patch #2.
Patch #1 will not compile.
--
Ueimor
I'm not really an expert, and I didn't understand all your remarks
but I can tell you this:
The driver supplied with 2.6.15 looks like dlink's driver version 1.17.
I had a dlink NIC that got stuck once in a while running that driver.
dlink's version 1.19 is written for 2.4 kernels, so all I did was
convert it to 2.6 kernels.
The new version still got stuck once in a while.
Maybe because the bugs you pointed out.
I don't have the dlink NIC anymore so I don't see how I can help here.
Maybe Edward can answer to that.
Sorry.
On Thu, 2006-09-07 at 11:19 +0200, Arjan van de Ven wrote:
> > @@ -335,8 +374,9 @@
> > #endif
> > /* Read eeprom */
> > for (i = 0; i < 128; i++) {
> > - ((u16 *) sromdata)[i] = le16_to_cpu (read_eeprom (ioaddr, i));
> > + ((u16 *) sromdata)[i] = cpu_to_le16 (read_eeprom (ioaddr, i));
> > }
> > + psrom->crc = le32_to_cpu(psrom->crc);
>
> this looks wrong, the data comes from the hw as le, so le*_to_cpu()
> sounds the right direction
>
> > @@ -401,7 +441,7 @@
> > int i;
> > u16 macctrl;
> >
> > - i = request_irq (dev->irq, &rio_interrupt, IRQF_SHARED, dev->name, dev);
> > + i = request_irq (dev->irq, &rio_interrupt, SA_SHIRQ, dev->name, dev);
> > if (i)
> > return i;
>
> this is backing out a fix/conversion to the new API. Bad.
>
>
> >
> > @@ -434,9 +474,12 @@
> > writeb (0x30, ioaddr + RxDMABurstThresh);
> > writeb (0x30, ioaddr + RxDMAUrgentThresh);
> > writel (0x0007ffff, ioaddr + RmonStatMask);
> > +
> > /* clear statistics */
> > clear_stats (dev);
> >
> > + atomic_set(&np->tx_desc_lock, 0);
>
> I'm quite scared by this naming; it suggests home-brew locking
>
> > dev->trans_start = jiffies;
> > + tasklet_enable(&np->tx_tasklet);
> > + writew(DEFAULT_INTR, ioaddr + IntEnable);
> > + return;
>
> this looks like a PCI posting bug
>
>
> > -rio_free_tx (struct net_device *dev, int irq)
> > +rio_free_tx (struct net_device *dev)
> > {
> > struct netdev_private *np = netdev_priv(dev);
> > int entry = np->old_tx % TX_RING_SIZE;
> > - int tx_use = 0;
> > unsigned long flag = 0;
> > + int irq = in_interrupt();
>
> eeeeep
>
> > +
> > + if (atomic_read(&np->tx_desc_lock))
> > + return;
> > + atomic_inc(&np->tx_desc_lock);
>
> and yes.. it is broken self made locking....
> there is a nice race between the _read and the _inc here.
>
>
> >
> > if (irq)
> > spin_lock(&np->tx_lock);
> > else
> > spin_lock_irqsave(&np->tx_lock, flag);
>
> double eeeep
>
> this is wrong to do with in_interrupt() as gating factor!
> Always doing the irqsave() is fine btw
>
>
>
* Arjan van de Ven:
>> if (irq)
>> spin_lock(&np->tx_lock);
>> else
>> spin_lock_irqsave(&np->tx_lock, flag);
>
> double eeeep
>
> this is wrong to do with in_interrupt() as gating factor!
Well, this is not Hayim fault. It's in 2.6.18 as is.
Unfortunately, the driver does not work very well. Transmitting
frames hangs after a while (quickly to reproduce with ping -f
-s40000). Any suggestions how we could this get working? The cards
used work quite well in older systems despite all the driver bugs, but
current systems with different timings expose them.
--
Florian Weimer <[email protected]>
BFK edv-consulting GmbH http://www.bfk.de/
Durlacher Allee 47 tel: +49-721-96201-1
D-76131 Karlsruhe fax: +49-721-96201-99
Hayim Shaul wrote:
> Description:
> For DLink Fiber NIC, Linux 2.4.22 ships with driver version 1.19,
> whereas, Linux 2.6.x ship with driver version 1.17.
>
> The following patch upgrades the 2.6.x driver to include changes (and
> bug fixes done until 1.19b).
>
> These fixes are (copied from the driver):
> 1.18 2002/11/07 New tx scheme, adaptive tx_coalesce.
> Remove tx_coalesce option.
> 1.19 2003/12/16 Fix problem parsing the eeprom on big endian
> systems. ([email protected])
>
> Disclaimer:
> Since I returned my DLink NIC to the store I couldn't test it
> thoroughly. It seemed to work just as well as v1.17. However, both
> version made the NIC hang after a few minutes.
Please update to the latest upstream kernel (2.6.22 or 2.6.23-rc).
Also, please make sure you do not back out changes included in the
upstream kernel since 2.6.17.