tg3: Prepare for larger rx ring sizes
authorMatt Carlson <mcarlson@broadcom.com>
Thu, 30 Sep 2010 10:34:35 +0000 (10:34 +0000)
committerDavid S. Miller <davem@davemloft.net>
Fri, 1 Oct 2010 07:24:44 +0000 (00:24 -0700)
This patch adds two new variables to track the size of the standard and
jumbo rx producer ring sizes.  The code is then pivoted to these
variables from preprocessor constants.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/tg3.c
drivers/net/tg3.h

index 3c9f95dd61e0172b1318fffb2c9a65a1ba00b7ff..af59a293b5baf3fb50cef8886b9c8624a96371cd 100644 (file)
  * You can't change the ring sizes, but you can change where you place
  * them in the NIC onboard memory.
  */
-#define TG3_RX_RING_SIZE               512
+#define TG3_RX_STD_RING_SIZE(tp)       512
 #define TG3_DEF_RX_RING_PENDING                200
-#define TG3_RX_JUMBO_RING_SIZE         256
+#define TG3_RX_JMB_RING_SIZE(tp)       256
 #define TG3_DEF_RX_JUMBO_RING_PENDING  100
 #define TG3_RSS_INDIR_TBL_SIZE         128
 
 #define TG3_TX_RING_SIZE               512
 #define TG3_DEF_TX_RING_PENDING                (TG3_TX_RING_SIZE - 1)
 
-#define TG3_RX_RING_BYTES      (sizeof(struct tg3_rx_buffer_desc) * \
-                                TG3_RX_RING_SIZE)
-#define TG3_RX_JUMBO_RING_BYTES        (sizeof(struct tg3_ext_rx_buffer_desc) * \
-                                TG3_RX_JUMBO_RING_SIZE)
-#define TG3_RX_RCB_RING_BYTES(tp) (sizeof(struct tg3_rx_buffer_desc) * \
-                                TG3_RX_RCB_RING_SIZE(tp))
+#define TG3_RX_STD_RING_BYTES(tp) \
+       (sizeof(struct tg3_rx_buffer_desc) * TG3_RX_STD_RING_SIZE(tp))
+#define TG3_RX_JMB_RING_BYTES(tp) \
+       (sizeof(struct tg3_ext_rx_buffer_desc) * TG3_RX_JMB_RING_SIZE(tp))
+#define TG3_RX_RCB_RING_BYTES(tp) \
+       (sizeof(struct tg3_rx_buffer_desc) * TG3_RX_RCB_RING_SIZE(tp))
 #define TG3_TX_RING_BYTES      (sizeof(struct tg3_tx_buffer_desc) * \
                                 TG3_TX_RING_SIZE)
 #define NEXT_TX(N)             (((N) + 1) & (TG3_TX_RING_SIZE - 1))
 #define TG3_RX_STD_MAP_SZ              TG3_RX_DMA_TO_MAP_SZ(TG3_RX_STD_DMA_SZ)
 #define TG3_RX_JMB_MAP_SZ              TG3_RX_DMA_TO_MAP_SZ(TG3_RX_JMB_DMA_SZ)
 
-#define TG3_RX_STD_BUFF_RING_SIZE \
-       (sizeof(struct ring_info) * TG3_RX_RING_SIZE)
+#define TG3_RX_STD_BUFF_RING_SIZE(tp) \
+       (sizeof(struct ring_info) * TG3_RX_STD_RING_SIZE(tp))
 
-#define TG3_RX_JMB_BUFF_RING_SIZE \
-       (sizeof(struct ring_info) * TG3_RX_JUMBO_RING_SIZE)
+#define TG3_RX_JMB_BUFF_RING_SIZE(tp) \
+       (sizeof(struct ring_info) * TG3_RX_JMB_RING_SIZE(tp))
 
 /* Due to a hardware bug, the 5701 can only DMA to memory addresses
  * that are at least dword aligned when used in PCIX mode.  The driver
@@ -4445,14 +4445,14 @@ static int tg3_alloc_rx_skb(struct tg3 *tp, struct tg3_rx_prodring_set *tpr,
        src_map = NULL;
        switch (opaque_key) {
        case RXD_OPAQUE_RING_STD:
-               dest_idx = dest_idx_unmasked % TG3_RX_RING_SIZE;
+               dest_idx = dest_idx_unmasked & tp->rx_std_ring_mask;
                desc = &tpr->rx_std[dest_idx];
                map = &tpr->rx_std_buffers[dest_idx];
                skb_size = tp->rx_pkt_map_sz;
                break;
 
        case RXD_OPAQUE_RING_JUMBO:
-               dest_idx = dest_idx_unmasked % TG3_RX_JUMBO_RING_SIZE;
+               dest_idx = dest_idx_unmasked & tp->rx_jmb_ring_mask;
                desc = &tpr->rx_jmb[dest_idx].std;
                map = &tpr->rx_jmb_buffers[dest_idx];
                skb_size = TG3_RX_JMB_MAP_SZ;
@@ -4507,7 +4507,7 @@ static void tg3_recycle_rx(struct tg3_napi *tnapi,
 
        switch (opaque_key) {
        case RXD_OPAQUE_RING_STD:
-               dest_idx = dest_idx_unmasked % TG3_RX_RING_SIZE;
+               dest_idx = dest_idx_unmasked & tp->rx_std_ring_mask;
                dest_desc = &dpr->rx_std[dest_idx];
                dest_map = &dpr->rx_std_buffers[dest_idx];
                src_desc = &spr->rx_std[src_idx];
@@ -4515,7 +4515,7 @@ static void tg3_recycle_rx(struct tg3_napi *tnapi,
                break;
 
        case RXD_OPAQUE_RING_JUMBO:
-               dest_idx = dest_idx_unmasked % TG3_RX_JUMBO_RING_SIZE;
+               dest_idx = dest_idx_unmasked & tp->rx_jmb_ring_mask;
                dest_desc = &dpr->rx_jmb[dest_idx].std;
                dest_map = &dpr->rx_jmb_buffers[dest_idx];
                src_desc = &spr->rx_jmb[src_idx].std;
@@ -4715,7 +4715,8 @@ next_pkt:
                (*post_ptr)++;
 
                if (unlikely(rx_std_posted >= tp->rx_std_max_post)) {
-                       tpr->rx_std_prod_idx = std_prod_idx % TG3_RX_RING_SIZE;
+                       tpr->rx_std_prod_idx = std_prod_idx &
+                                              tp->rx_std_ring_mask;
                        tw32_rx_mbox(TG3_RX_STD_PROD_IDX_REG,
                                     tpr->rx_std_prod_idx);
                        work_mask &= ~RXD_OPAQUE_RING_STD;
@@ -4739,13 +4740,14 @@ next_pkt_nopost:
        /* Refill RX ring(s). */
        if (!(tp->tg3_flags3 & TG3_FLG3_ENABLE_RSS)) {
                if (work_mask & RXD_OPAQUE_RING_STD) {
-                       tpr->rx_std_prod_idx = std_prod_idx % TG3_RX_RING_SIZE;
+                       tpr->rx_std_prod_idx = std_prod_idx &
+                                              tp->rx_std_ring_mask;
                        tw32_rx_mbox(TG3_RX_STD_PROD_IDX_REG,
                                     tpr->rx_std_prod_idx);
                }
                if (work_mask & RXD_OPAQUE_RING_JUMBO) {
-                       tpr->rx_jmb_prod_idx = jmb_prod_idx %
-                                              TG3_RX_JUMBO_RING_SIZE;
+                       tpr->rx_jmb_prod_idx = jmb_prod_idx &
+                                              tp->rx_jmb_ring_mask;
                        tw32_rx_mbox(TG3_RX_JMB_PROD_IDX_REG,
                                     tpr->rx_jmb_prod_idx);
                }
@@ -4756,8 +4758,8 @@ next_pkt_nopost:
                 */
                smp_wmb();
 
-               tpr->rx_std_prod_idx = std_prod_idx % TG3_RX_RING_SIZE;
-               tpr->rx_jmb_prod_idx = jmb_prod_idx % TG3_RX_JUMBO_RING_SIZE;
+               tpr->rx_std_prod_idx = std_prod_idx & tp->rx_std_ring_mask;
+               tpr->rx_jmb_prod_idx = jmb_prod_idx & tp->rx_jmb_ring_mask;
 
                if (tnapi != &tp->napi[1])
                        napi_schedule(&tp->napi[1].napi);
@@ -4813,9 +4815,11 @@ static int tg3_rx_prodring_xfer(struct tg3 *tp,
                if (spr->rx_std_cons_idx < src_prod_idx)
                        cpycnt = src_prod_idx - spr->rx_std_cons_idx;
                else
-                       cpycnt = TG3_RX_RING_SIZE - spr->rx_std_cons_idx;
+                       cpycnt = tp->rx_std_ring_mask + 1 -
+                                spr->rx_std_cons_idx;
 
-               cpycnt = min(cpycnt, TG3_RX_RING_SIZE - dpr->rx_std_prod_idx);
+               cpycnt = min(cpycnt,
+                            tp->rx_std_ring_mask + 1 - dpr->rx_std_prod_idx);
 
                si = spr->rx_std_cons_idx;
                di = dpr->rx_std_prod_idx;
@@ -4849,10 +4853,10 @@ static int tg3_rx_prodring_xfer(struct tg3 *tp,
                        dbd->addr_lo = sbd->addr_lo;
                }
 
-               spr->rx_std_cons_idx = (spr->rx_std_cons_idx + cpycnt) %
-                                      TG3_RX_RING_SIZE;
-               dpr->rx_std_prod_idx = (dpr->rx_std_prod_idx + cpycnt) %
-                                      TG3_RX_RING_SIZE;
+               spr->rx_std_cons_idx = (spr->rx_std_cons_idx + cpycnt) &
+                                      tp->rx_std_ring_mask;
+               dpr->rx_std_prod_idx = (dpr->rx_std_prod_idx + cpycnt) &
+                                      tp->rx_std_ring_mask;
        }
 
        while (1) {
@@ -4869,10 +4873,11 @@ static int tg3_rx_prodring_xfer(struct tg3 *tp,
                if (spr->rx_jmb_cons_idx < src_prod_idx)
                        cpycnt = src_prod_idx - spr->rx_jmb_cons_idx;
                else
-                       cpycnt = TG3_RX_JUMBO_RING_SIZE - spr->rx_jmb_cons_idx;
+                       cpycnt = tp->rx_jmb_ring_mask + 1 -
+                                spr->rx_jmb_cons_idx;
 
                cpycnt = min(cpycnt,
-                            TG3_RX_JUMBO_RING_SIZE - dpr->rx_jmb_prod_idx);
+                            tp->rx_jmb_ring_mask + 1 - dpr->rx_jmb_prod_idx);
 
                si = spr->rx_jmb_cons_idx;
                di = dpr->rx_jmb_prod_idx;
@@ -4906,10 +4911,10 @@ static int tg3_rx_prodring_xfer(struct tg3 *tp,
                        dbd->addr_lo = sbd->addr_lo;
                }
 
-               spr->rx_jmb_cons_idx = (spr->rx_jmb_cons_idx + cpycnt) %
-                                      TG3_RX_JUMBO_RING_SIZE;
-               dpr->rx_jmb_prod_idx = (dpr->rx_jmb_prod_idx + cpycnt) %
-                                      TG3_RX_JUMBO_RING_SIZE;
+               spr->rx_jmb_cons_idx = (spr->rx_jmb_cons_idx + cpycnt) &
+                                      tp->rx_jmb_ring_mask;
+               dpr->rx_jmb_prod_idx = (dpr->rx_jmb_prod_idx + cpycnt) &
+                                      tp->rx_jmb_ring_mask;
        }
 
        return err;
@@ -6059,14 +6064,14 @@ static void tg3_rx_prodring_free(struct tg3 *tp,
 
        if (tpr != &tp->napi[0].prodring) {
                for (i = tpr->rx_std_cons_idx; i != tpr->rx_std_prod_idx;
-                    i = (i + 1) % TG3_RX_RING_SIZE)
+                    i = (i + 1) & tp->rx_std_ring_mask)
                        tg3_rx_skb_free(tp, &tpr->rx_std_buffers[i],
                                        tp->rx_pkt_map_sz);
 
                if (tp->tg3_flags & TG3_FLAG_JUMBO_CAPABLE) {
                        for (i = tpr->rx_jmb_cons_idx;
                             i != tpr->rx_jmb_prod_idx;
-                            i = (i + 1) % TG3_RX_JUMBO_RING_SIZE) {
+                            i = (i + 1) & tp->rx_jmb_ring_mask) {
                                tg3_rx_skb_free(tp, &tpr->rx_jmb_buffers[i],
                                                TG3_RX_JMB_MAP_SZ);
                        }
@@ -6075,12 +6080,12 @@ static void tg3_rx_prodring_free(struct tg3 *tp,
                return;
        }
 
-       for (i = 0; i < TG3_RX_RING_SIZE; i++)
+       for (i = 0; i <= tp->rx_std_ring_mask; i++)
                tg3_rx_skb_free(tp, &tpr->rx_std_buffers[i],
                                tp->rx_pkt_map_sz);
 
        if (tp->tg3_flags & TG3_FLAG_JUMBO_CAPABLE) {
-               for (i = 0; i < TG3_RX_JUMBO_RING_SIZE; i++)
+               for (i = 0; i <= tp->rx_jmb_ring_mask; i++)
                        tg3_rx_skb_free(tp, &tpr->rx_jmb_buffers[i],
                                        TG3_RX_JMB_MAP_SZ);
        }
@@ -6104,15 +6109,16 @@ static int tg3_rx_prodring_alloc(struct tg3 *tp,
        tpr->rx_jmb_prod_idx = 0;
 
        if (tpr != &tp->napi[0].prodring) {
-               memset(&tpr->rx_std_buffers[0], 0, TG3_RX_STD_BUFF_RING_SIZE);
+               memset(&tpr->rx_std_buffers[0], 0,
+                      TG3_RX_STD_BUFF_RING_SIZE(tp));
                if (tp->tg3_flags & TG3_FLAG_JUMBO_CAPABLE)
                        memset(&tpr->rx_jmb_buffers[0], 0,
-                              TG3_RX_JMB_BUFF_RING_SIZE);
+                              TG3_RX_JMB_BUFF_RING_SIZE(tp));
                goto done;
        }
 
        /* Zero out all descriptors. */
-       memset(tpr->rx_std, 0, TG3_RX_RING_BYTES);
+       memset(tpr->rx_std, 0, TG3_RX_STD_RING_BYTES(tp));
 
        rx_pkt_dma_sz = TG3_RX_STD_DMA_SZ;
        if ((tp->tg3_flags2 & TG3_FLG2_5780_CLASS) &&
@@ -6124,7 +6130,7 @@ static int tg3_rx_prodring_alloc(struct tg3 *tp,
         * stuff once.  This works because the card does not
         * write into the rx buffer posting rings.
         */
-       for (i = 0; i < TG3_RX_RING_SIZE; i++) {
+       for (i = 0; i <= tp->rx_std_ring_mask; i++) {
                struct tg3_rx_buffer_desc *rxd;
 
                rxd = &tpr->rx_std[i];
@@ -6151,12 +6157,12 @@ static int tg3_rx_prodring_alloc(struct tg3 *tp,
        if (!(tp->tg3_flags & TG3_FLAG_JUMBO_CAPABLE))
                goto done;
 
-       memset(tpr->rx_jmb, 0, TG3_RX_JUMBO_RING_BYTES);
+       memset(tpr->rx_jmb, 0, TG3_RX_JMB_RING_BYTES(tp));
 
        if (!(tp->tg3_flags & TG3_FLAG_JUMBO_RING_ENABLE))
                goto done;
 
-       for (i = 0; i < TG3_RX_JUMBO_RING_SIZE; i++) {
+       for (i = 0; i <= tp->rx_jmb_ring_mask; i++) {
                struct tg3_rx_buffer_desc *rxd;
 
                rxd = &tpr->rx_jmb[i].std;
@@ -6196,12 +6202,12 @@ static void tg3_rx_prodring_fini(struct tg3 *tp,
        kfree(tpr->rx_jmb_buffers);
        tpr->rx_jmb_buffers = NULL;
        if (tpr->rx_std) {
-               pci_free_consistent(tp->pdev, TG3_RX_RING_BYTES,
+               pci_free_consistent(tp->pdev, TG3_RX_STD_RING_BYTES(tp),
                                    tpr->rx_std, tpr->rx_std_mapping);
                tpr->rx_std = NULL;
        }
        if (tpr->rx_jmb) {
-               pci_free_consistent(tp->pdev, TG3_RX_JUMBO_RING_BYTES,
+               pci_free_consistent(tp->pdev, TG3_RX_JMB_RING_BYTES(tp),
                                    tpr->rx_jmb, tpr->rx_jmb_mapping);
                tpr->rx_jmb = NULL;
        }
@@ -6210,23 +6216,24 @@ static void tg3_rx_prodring_fini(struct tg3 *tp,
 static int tg3_rx_prodring_init(struct tg3 *tp,
                                struct tg3_rx_prodring_set *tpr)
 {
-       tpr->rx_std_buffers = kzalloc(TG3_RX_STD_BUFF_RING_SIZE, GFP_KERNEL);
+       tpr->rx_std_buffers = kzalloc(TG3_RX_STD_BUFF_RING_SIZE(tp),
+                                     GFP_KERNEL);
        if (!tpr->rx_std_buffers)
                return -ENOMEM;
 
-       tpr->rx_std = pci_alloc_consistent(tp->pdev, TG3_RX_RING_BYTES,
+       tpr->rx_std = pci_alloc_consistent(tp->pdev, TG3_RX_STD_RING_BYTES(tp),
                                           &tpr->rx_std_mapping);
        if (!tpr->rx_std)
                goto err_out;
 
        if (tp->tg3_flags & TG3_FLAG_JUMBO_CAPABLE) {
-               tpr->rx_jmb_buffers = kzalloc(TG3_RX_JMB_BUFF_RING_SIZE,
+               tpr->rx_jmb_buffers = kzalloc(TG3_RX_JMB_BUFF_RING_SIZE(tp),
                                              GFP_KERNEL);
                if (!tpr->rx_jmb_buffers)
                        goto err_out;
 
                tpr->rx_jmb = pci_alloc_consistent(tp->pdev,
-                                                  TG3_RX_JUMBO_RING_BYTES,
+                                                  TG3_RX_JMB_RING_BYTES(tp),
                                                   &tpr->rx_jmb_mapping);
                if (!tpr->rx_jmb)
                        goto err_out;
@@ -9854,10 +9861,10 @@ static void tg3_get_ringparam(struct net_device *dev, struct ethtool_ringparam *
 {
        struct tg3 *tp = netdev_priv(dev);
 
-       ering->rx_max_pending = TG3_RX_RING_SIZE - 1;
+       ering->rx_max_pending = tp->rx_std_ring_mask;
        ering->rx_mini_max_pending = 0;
        if (tp->tg3_flags & TG3_FLAG_JUMBO_RING_ENABLE)
-               ering->rx_jumbo_max_pending = TG3_RX_JUMBO_RING_SIZE - 1;
+               ering->rx_jumbo_max_pending = tp->rx_jmb_ring_mask;
        else
                ering->rx_jumbo_max_pending = 0;
 
@@ -9878,8 +9885,8 @@ static int tg3_set_ringparam(struct net_device *dev, struct ethtool_ringparam *e
        struct tg3 *tp = netdev_priv(dev);
        int i, irq_sync = 0, err = 0;
 
-       if ((ering->rx_pending > TG3_RX_RING_SIZE - 1) ||
-           (ering->rx_jumbo_pending > TG3_RX_JUMBO_RING_SIZE - 1) ||
+       if ((ering->rx_pending > tp->rx_std_ring_mask) ||
+           (ering->rx_jumbo_pending > tp->rx_jmb_ring_mask) ||
            (ering->tx_pending > TG3_TX_RING_SIZE - 1) ||
            (ering->tx_pending <= MAX_SKB_FRAGS) ||
            ((tp->tg3_flags2 & TG3_FLG2_TSO_BUG) &&
@@ -13597,7 +13604,9 @@ static int __devinit tg3_get_invariants(struct tg3 *tp)
 #endif
        }
 
-       tp->rx_std_max_post = TG3_RX_RING_SIZE;
+       tp->rx_std_ring_mask = TG3_RX_STD_RING_SIZE(tp) - 1;
+       tp->rx_jmb_ring_mask = TG3_RX_JMB_RING_SIZE(tp) - 1;
+       tp->rx_std_max_post = tp->rx_std_ring_mask + 1;
 
        /* Increment the rx prod index on the rx std ring by at most
         * 8 for these chips to workaround hw errata.
index 241e314871663cee0e1572f3b024daac96304df6..9763298c22c61b8a6b661369855ecf0457c4dbca 100644 (file)
@@ -2762,6 +2762,8 @@ struct tg3 {
        void                            (*write32_rx_mbox) (struct tg3 *, u32,
                                                            u32);
        u32                             rx_copy_thresh;
+       u32                             rx_std_ring_mask;
+       u32                             rx_jmb_ring_mask;
        u32                             rx_pending;
        u32                             rx_jumbo_pending;
        u32                             rx_std_max_post;