src/patches/suse-2.6.27.31/patches.suse/SoN-17-net-ps_rx.patch

   1 From: Peter Zijlstra <a.p.zijlstra@chello.nl>
   2 Subject: net: packet split receive api
   3 Patch-mainline: No
   4 References: FATE#303834, bnc#484306
   5
   6 Add some packet-split receive hooks.
   7
   8 For one this allows to do NUMA node affine page allocs. Later on these hooks
   9 will be extended to do emergency reserve allocations for fragments.
  10
  11 Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
  12 Acked-by: Neil Brown <neilb@suse.de>
  13 Acked-by: Suresh Jayaraman <sjayaraman@suse.de>
  14
  15 bnx2.c part fixed by Jiri Bohac <jbohac@suse.cz> (bnc#484306)
  16
  17
  18
  19 ---
  20  drivers/net/bnx2.c             |    9 +++------
  21  drivers/net/e1000/e1000_main.c |    8 ++------
  22  drivers/net/e1000e/netdev.c    |    7 ++-----
  23  drivers/net/igb/igb_main.c     |    9 ++-------
  24  drivers/net/ixgbe/ixgbe_main.c |   14 ++++++--------
  25  drivers/net/sky2.c             |   16 ++++++----------
  26  include/linux/skbuff.h         |   23 +++++++++++++++++++++++
  27  net/core/skbuff.c              |   20 ++++++++++++++++++++
  28  8 files changed, 64 insertions(+), 42 deletions(-)
  29
  30 --- a/drivers/net/bnx2.c
  31 +++ b/drivers/net/bnx2.c
  32 @@ -2476,7 +2476,7 @@ bnx2_alloc_rx_page(struct bnx2 *bp, stru
  33         struct sw_pg *rx_pg = &rxr->rx_pg_ring[index];
  34         struct rx_bd *rxbd =
  35                 &rxr->rx_pg_desc_ring[RX_RING(index)][RX_IDX(index)];
  36 -       struct page *page = alloc_page(GFP_ATOMIC);
  37 +       struct page *page = netdev_alloc_page(bp->dev);
  38
  39         if (!page)
  40                 return -ENOMEM;
  41 @@ -2501,7 +2501,7 @@ bnx2_free_rx_page(struct bnx2 *bp, struc
  42         pci_unmap_page(bp->pdev, pci_unmap_addr(rx_pg, mapping), PAGE_SIZE,
  43                        PCI_DMA_FROMDEVICE);
  44
  45 -       __free_page(page);
  46 +       netdev_free_page(bp->dev, page);
  47         rx_pg->page = NULL;
  48  }
  49
  50 @@ -2819,7 +2819,7 @@ bnx2_rx_skb(struct bnx2 *bp, struct bnx2
  51                         if (i == pages - 1)
  52                                 frag_len -= 4;
  53
  54 -                       skb_fill_page_desc(skb, i, rx_pg->page, 0, frag_len);
  55 +                       skb_add_rx_frag(skb, i, rx_pg->page, 0, frag_len);
  56                         rx_pg->page = NULL;
  57
  58                         err = bnx2_alloc_rx_page(bp, rxr,
  59 @@ -2833,9 +2833,6 @@ bnx2_rx_skb(struct bnx2 *bp, struct bnx2
  60                         }
  61
  62                         frag_size -= frag_len;
  63 -                       skb->data_len += frag_len;
  64 -                       skb->truesize += frag_len;
  65 -                       skb->len += frag_len;
  66
  67                         pg_prod = NEXT_RX_BD(pg_prod);
  68                         pg_cons = RX_PG_RING_IDX(NEXT_RX_BD(pg_cons));
  69 --- a/drivers/net/e1000/e1000_main.c
  70 +++ b/drivers/net/e1000/e1000_main.c
  71 @@ -4349,12 +4349,8 @@ static bool e1000_clean_rx_irq_ps(struct
  72                         pci_unmap_page(pdev, ps_page_dma->ps_page_dma[j],
  73                                         PAGE_SIZE, PCI_DMA_FROMDEVICE);
  74                         ps_page_dma->ps_page_dma[j] = 0;
  75 -                       skb_fill_page_desc(skb, j, ps_page->ps_page[j], 0,
  76 -                                          length);
  77 +                       skb_add_rx_frag(skb, j, ps_page->ps_page[j], 0, length);
  78                         ps_page->ps_page[j] = NULL;
  79 -                       skb->len += length;
  80 -                       skb->data_len += length;
  81 -                       skb->truesize += length;
  82                 }
  83
  84                 /* strip the ethernet crc, problem is we're using pages now so
  85 @@ -4553,7 +4549,7 @@ static void e1000_alloc_rx_buffers_ps(st
  86                         if (j < adapter->rx_ps_pages) {
  87                                 if (likely(!ps_page->ps_page[j])) {
  88                                         ps_page->ps_page[j] =
  89 -                                               alloc_page(GFP_ATOMIC);
  90 +                                               netdev_alloc_page(netdev);
  91                                         if (unlikely(!ps_page->ps_page[j])) {
  92                                                 adapter->alloc_rx_buff_failed++;
  93                                                 goto no_buffers;
  94 --- a/drivers/net/e1000e/netdev.c
  95 +++ b/drivers/net/e1000e/netdev.c
  96 @@ -258,7 +258,7 @@ static void e1000_alloc_rx_buffers_ps(st
  97                                 continue;
  98                         }
  99                         if (!ps_page->page) {
 100 -                               ps_page->page = alloc_page(GFP_ATOMIC);
 101 +                               ps_page->page = netdev_alloc_page(netdev);
 102                                 if (!ps_page->page) {
 103                                         adapter->alloc_rx_buff_failed++;
 104                                         goto no_buffers;
 105 @@ -826,11 +826,8 @@ static bool e1000_clean_rx_irq_ps(struct
 106                         pci_unmap_page(pdev, ps_page->dma, PAGE_SIZE,
 107                                        PCI_DMA_FROMDEVICE);
 108                         ps_page->dma = 0;
 109 -                       skb_fill_page_desc(skb, j, ps_page->page, 0, length);
 110 +                       skb_add_rx_frag(skb, j, ps_page->page, 0, length);
 111                         ps_page->page = NULL;
 112 -                       skb->len += length;
 113 -                       skb->data_len += length;
 114 -                       skb->truesize += length;
 115                 }
 116
 117                 /* strip the ethernet crc, problem is we're using pages now so
 118 --- a/drivers/net/igb/igb_main.c
 119 +++ b/drivers/net/igb/igb_main.c
 120 @@ -3885,7 +3885,7 @@ static bool igb_clean_rx_irq_adv(struct
 121                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
 122                         buffer_info->page_dma = 0;
 123
 124 -                       skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
 125 +                       skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags++,
 126                                                 buffer_info->page,
 127                                                 buffer_info->page_offset,
 128                                                 length);
 129 @@ -3895,11 +3895,6 @@ static bool igb_clean_rx_irq_adv(struct
 130                                 buffer_info->page = NULL;
 131                         else
 132                                 get_page(buffer_info->page);
 133 -
 134 -                       skb->len += length;
 135 -                       skb->data_len += length;
 136 -
 137 -                       skb->truesize += length;
 138                 }
 139  send_up:
 140                 i++;
 141 @@ -3993,7 +3988,7 @@ static void igb_alloc_rx_buffers_adv(str
 142
 143                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
 144                         if (!buffer_info->page) {
 145 -                               buffer_info->page = alloc_page(GFP_ATOMIC);
 146 +                               buffer_info->page = netdev_alloc_page(netdev);
 147                                 if (!buffer_info->page) {
 148                                         adapter->alloc_rx_buff_failed++;
 149                                         goto no_buffers;
 150 --- a/drivers/net/ixgbe/ixgbe_main.c
 151 +++ b/drivers/net/ixgbe/ixgbe_main.c
 152 @@ -480,6 +480,7 @@ static void ixgbe_alloc_rx_buffers(struc
 153                                     int cleaned_count)
 154  {
 155         struct pci_dev *pdev = adapter->pdev;
 156 +       struct net_device *netdev = adapter->netdev;
 157         union ixgbe_adv_rx_desc *rx_desc;
 158         struct ixgbe_rx_buffer *bi;
 159         unsigned int i;
 160 @@ -494,7 +495,7 @@ static void ixgbe_alloc_rx_buffers(struc
 161                 if (!bi->page_dma &&
 162                     (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)) {
 163                         if (!bi->page) {
 164 -                               bi->page = alloc_page(GFP_ATOMIC);
 165 +                               bi->page = netdev_alloc_page(netdev);
 166                                 if (!bi->page) {
 167                                         adapter->alloc_rx_page_failed++;
 168                                         goto no_buffers;
 169 @@ -628,10 +629,10 @@ static bool ixgbe_clean_rx_irq(struct ix
 170                         pci_unmap_page(pdev, rx_buffer_info->page_dma,
 171                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
 172                         rx_buffer_info->page_dma = 0;
 173 -                       skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
 174 -                                          rx_buffer_info->page,
 175 -                                          rx_buffer_info->page_offset,
 176 -                                          upper_len);
 177 +                       skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
 178 +                                       rx_buffer_info->page,
 179 +                                       rx_buffer_info->page_offset,
 180 +                                       upper_len);
 181
 182                         if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
 183                             (page_count(rx_buffer_info->page) != 1))
 184 @@ -639,9 +640,6 @@ static bool ixgbe_clean_rx_irq(struct ix
 185                         else
 186                                 get_page(rx_buffer_info->page);
 187
 188 -                       skb->len += upper_len;
 189 -                       skb->data_len += upper_len;
 190 -                       skb->truesize += upper_len;
 191                 }
 192
 193                 i++;
 194 --- a/drivers/net/sky2.c
 195 +++ b/drivers/net/sky2.c
 196 @@ -1272,7 +1272,7 @@ static struct sk_buff *sky2_rx_alloc(str
 197         }
 198
 199         for (i = 0; i < sky2->rx_nfrags; i++) {
 200 -               struct page *page = alloc_page(GFP_ATOMIC);
 201 +               struct page *page = netdev_alloc_page(sky2->netdev);
 202
 203                 if (!page)
 204                         goto free_partial;
 205 @@ -2141,8 +2141,8 @@ static struct sk_buff *receive_copy(stru
 206  }
 207
 208  /* Adjust length of skb with fragments to match received data */
 209 -static void skb_put_frags(struct sk_buff *skb, unsigned int hdr_space,
 210 -                         unsigned int length)
 211 +static void skb_put_frags(struct sky2_port *sky2, struct sk_buff *skb,
 212 +                         unsigned int hdr_space, unsigned int length)
 213  {
 214         int i, num_frags;
 215         unsigned int size;
 216 @@ -2159,15 +2159,11 @@ static void skb_put_frags(struct sk_buff
 217
 218                 if (length == 0) {
 219                         /* don't need this page */
 220 -                       __free_page(frag->page);
 221 +                       netdev_free_page(sky2->netdev, frag->page);
 222                         --skb_shinfo(skb)->nr_frags;
 223                 } else {
 224                         size = min(length, (unsigned) PAGE_SIZE);
 225 -
 226 -                       frag->size = size;
 227 -                       skb->data_len += size;
 228 -                       skb->truesize += size;
 229 -                       skb->len += size;
 230 +                       skb_add_rx_frag(skb, i, frag->page, 0, size);
 231                         length -= size;
 232                 }
 233         }
 234 @@ -2194,7 +2190,7 @@ static struct sk_buff *receive_new(struc
 235         sky2_rx_map_skb(sky2->hw->pdev, re, hdr_space);
 236
 237         if (skb_shinfo(skb)->nr_frags)
 238 -               skb_put_frags(skb, hdr_space, length);
 239 +               skb_put_frags(sky2, skb, hdr_space, length);
 240         else
 241                 skb_put(skb, length);
 242         return skb;
 243 --- a/include/linux/skbuff.h
 244 +++ b/include/linux/skbuff.h
 245 @@ -829,6 +829,9 @@ static inline void skb_fill_page_desc(st
 246         skb_shinfo(skb)->nr_frags = i + 1;
 247  }
 248
 249 +extern void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page,
 250 +                           int off, int size);
 251 +
 252  #define SKB_PAGE_ASSERT(skb)   BUG_ON(skb_shinfo(skb)->nr_frags)
 253  #define SKB_FRAG_ASSERT(skb)   BUG_ON(skb_shinfo(skb)->frag_list)
 254  #define SKB_LINEAR_ASSERT(skb)  BUG_ON(skb_is_nonlinear(skb))
 255 @@ -1243,6 +1246,26 @@ static inline struct sk_buff *netdev_all
 256         return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
 257  }
 258
 259 +extern struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask);
 260 +
 261 +/**
 262 + *     netdev_alloc_page - allocate a page for ps-rx on a specific device
 263 + *     @dev: network device to receive on
 264 + *
 265 + *     Allocate a new page node local to the specified device.
 266 + *
 267 + *     %NULL is returned if there is no free memory.
 268 + */
 269 +static inline struct page *netdev_alloc_page(struct net_device *dev)
 270 +{
 271 +       return __netdev_alloc_page(dev, GFP_ATOMIC);
 272 +}
 273 +
 274 +static inline void netdev_free_page(struct net_device *dev, struct page *page)
 275 +{
 276 +       __free_page(page);
 277 +}
 278 +
 279  /**
 280   *     skb_clone_writable - is the header of a clone writable
 281   *     @skb: buffer to check
 282 --- a/net/core/skbuff.c
 283 +++ b/net/core/skbuff.c
 284 @@ -259,6 +259,26 @@ struct sk_buff *__netdev_alloc_skb(struc
 285         return skb;
 286  }
 287
 288 +struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask)
 289 +{
 290 +       int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
 291 +       struct page *page;
 292 +
 293 +       page = alloc_pages_node(node, gfp_mask, 0);
 294 +       return page;
 295 +}
 296 +EXPORT_SYMBOL(__netdev_alloc_page);
 297 +
 298 +void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
 299 +               int size)
 300 +{
 301 +       skb_fill_page_desc(skb, i, page, off, size);
 302 +       skb->len += size;
 303 +       skb->data_len += size;
 304 +       skb->truesize += size;
 305 +}
 306 +EXPORT_SYMBOL(skb_add_rx_frag);
 307 +
 308  /**
 309   *     dev_alloc_skb - allocate an skbuff for receiving
 310   *     @length: length to allocate