target/linux/rb532/patches/240-via_rhine_performance.patch

   1 --- linux-2.6.23.12/drivers/net/via-rhine.c     2007-12-18 22:55:57.000000000 +0100
   2 +++ linux-2.6.23.12.new/drivers/net/via-rhine.c 2008-01-11 08:58:58.000000000 +0100
   3 @@ -33,6 +33,8 @@
   4  #define DRV_VERSION    "1.4.3"
   5  #define DRV_RELDATE    "2007-03-06"
   6
   7 +#define PKT_ALIGN 1
   8 +
   9
  10  /* A few user-configurable values.
  11     These may be modified when a driver module is loaded. */
  12 @@ -40,6 +42,7 @@
  13  static int debug = 1;  /* 1 normal messages, 0 quiet .. 7 verbose. */
  14  static int max_interrupt_work = 20;
  15
  16 +#ifndef PKT_ALIGN
  17  /* Set the copy breakpoint for the copy-only-tiny-frames scheme.
  18     Setting to > 1518 effectively disables this feature. */
  19  #if defined(__alpha__) || defined(__arm__) || defined(__hppa__) \
  20 @@ -49,6 +52,7 @@
  21  #else
  22  static int rx_copybreak;
  23  #endif
  24 +#endif /* PKT_ALIGN */
  25
  26  /* Work-around for broken BIOSes: they are unable to get the chip back out of
  27     power state D3 so PXE booting fails. bootparam(7): via-rhine.avoid_D3=1 */
  28 @@ -111,6 +115,7 @@
  29  #include <asm/io.h>
  30  #include <asm/irq.h>
  31  #include <asm/uaccess.h>
  32 +#include <asm/unaligned.h>
  33  #include <linux/dmi.h>
  34
  35  /* These identify the driver base version and may not be removed. */
  36 @@ -130,12 +135,14 @@
  37
  38  module_param(max_interrupt_work, int, 0);
  39  module_param(debug, int, 0);
  40 -module_param(rx_copybreak, int, 0);
  41  module_param(avoid_D3, bool, 0);
  42  MODULE_PARM_DESC(max_interrupt_work, "VIA Rhine maximum events handled per interrupt");
  43  MODULE_PARM_DESC(debug, "VIA Rhine debug level (0-7)");
  44 -MODULE_PARM_DESC(rx_copybreak, "VIA Rhine copy breakpoint for copy-only-tiny-frames");
  45  MODULE_PARM_DESC(avoid_D3, "Avoid power state D3 (work-around for broken BIOSes)");
  46 +#ifndef PKT_ALIGN
  47 +module_param(rx_copybreak, int, 0);
  48 +MODULE_PARM_DESC(rx_copybreak, "VIA Rhine copy breakpoint for copy-only-tiny-frames");
  49 +#endif
  50
  51  /*
  52                 Theory of Operation
  53 @@ -927,7 +934,7 @@
  54
  55         /* Fill in the Rx buffers.  Handle allocation failure gracefully. */
  56         for (i = 0; i < RX_RING_SIZE; i++) {
  57 -               struct sk_buff *skb = dev_alloc_skb(rp->rx_buf_sz);
  58 +               struct sk_buff *skb = dev_alloc_skb(rp->rx_buf_sz + 4);
  59                 rp->rx_skbuff[i] = skb;
  60                 if (skb == NULL)
  61                         break;
  62 @@ -1484,7 +1491,9 @@
  63                         struct sk_buff *skb;
  64                         /* Length should omit the CRC */
  65                         int pkt_len = data_size - 4;
  66 -
  67 +#ifdef PKT_ALIGN
  68 +                       int i;
  69 +#else
  70                         /* Check if the packet is long enough to accept without
  71                            copying to a minimally-sized skbuff. */
  72                         if (pkt_len < rx_copybreak &&
  73 @@ -1503,7 +1512,9 @@
  74                                                                rp->rx_skbuff_dma[entry],
  75                                                                rp->rx_buf_sz,
  76                                                                PCI_DMA_FROMDEVICE);
  77 -                       } else {
  78 +                       } else
  79 +#endif
  80 +                       {
  81                                 skb = rp->rx_skbuff[entry];
  82                                 if (skb == NULL) {
  83                                         printk(KERN_ERR "%s: Inconsistent Rx "
  84 @@ -1517,6 +1528,14 @@
  85                                                  rp->rx_skbuff_dma[entry],
  86                                                  rp->rx_buf_sz,
  87                                                  PCI_DMA_FROMDEVICE);
  88 +#ifdef PKT_ALIGN
  89 +                               /* align the data to the ip header - should be faster than copying the entire packet */
  90 +                               for (i = pkt_len - (pkt_len % 4); i >= 0; i -= 4) {
  91 +                                       put_unaligned(*((u32 *) (skb->data + i)), (u32 *) (skb->data + i + 2));
  92 +                               }
  93 +                               skb->data += 2;
  94 +                               skb->tail += 2;
  95 +#endif
  96                         }
  97                         skb->protocol = eth_type_trans(skb, dev);
  98  #ifdef CONFIG_VIA_RHINE_NAPI