IB/mthca: Always fill MTTs from CPU

Speed up memory registration by filling in MTTs directly when the CPU
can write directly to the whole table (all mem-free cards, and to
Tavor mode on 64-bit systems with the patch I posted earlier).  This
reduces the number of FW commands needed to register an MR by at least
a factor of 2 and speeds up memory registration significantly.

Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

3 files changed, 89 insertions, 9 deletions

diff --git a/drivers/infiniband/hw/mthca/mthca_dev.h b/drivers/infiniband/hw/mthca/mthca_dev.h
index fe5cecf70fed..b7e42efaf43d 100644
--- a/drivers/infiniband/hw/mthca/mthca_dev.h
+++ b/drivers/infiniband/hw/mthca/mthca_dev.h
@@ -464,6 +464,8 @@ void mthca_uar_free(struct mthca_dev *dev, struct mthca_uar *uar);
 int mthca_pd_alloc(struct mthca_dev *dev, int privileged, struct mthca_pd *pd);
 void mthca_pd_free(struct mthca_dev *dev, struct mthca_pd *pd);
 
+int mthca_write_mtt_size(struct mthca_dev *dev);
+
 struct mthca_mtt *mthca_alloc_mtt(struct mthca_dev *dev, int size);
 void mthca_free_mtt(struct mthca_dev *dev, struct mthca_mtt *mtt);
 int mthca_write_mtt(struct mthca_dev *dev, struct mthca_mtt *mtt,
diff --git a/drivers/infiniband/hw/mthca/mthca_mr.c b/drivers/infiniband/hw/mthca/mthca_mr.c
index 958c6d5b6bc0..6037dd3f87df 100644
--- a/drivers/infiniband/hw/mthca/mthca_mr.c
+++ b/drivers/infiniband/hw/mthca/mthca_mr.c
@@ -243,8 +243,8 @@ void mthca_free_mtt(struct mthca_dev *dev, struct mthca_mtt *mtt)
         kfree(mtt);
 }
 
-int mthca_write_mtt(struct mthca_dev *dev, struct mthca_mtt *mtt,
-                    int start_index, u64 *buffer_list, int list_len)
+static int __mthca_write_mtt(struct mthca_dev *dev, struct mthca_mtt *mtt,
+                             int start_index, u64 *buffer_list, int list_len)
 {
         struct mthca_mailbox *mailbox;
         __be64 *mtt_entry;
@@ -295,6 +295,84 @@ out:
         return err;
 }
 
+int mthca_write_mtt_size(struct mthca_dev *dev)
+{
+        if (dev->mr_table.fmr_mtt_buddy != &dev->mr_table.mtt_buddy)
+                /*
+                 * Be friendly to WRITE_MTT command
+                 * and leave two empty slots for the
+                 * index and reserved fields of the
+                 * mailbox.
+                 */
+                return PAGE_SIZE / sizeof (u64) - 2;
+
+        /* For Arbel, all MTTs must fit in the same page. */
+        return mthca_is_memfree(dev) ? (PAGE_SIZE / sizeof (u64)) : 0x7ffffff;
+}
+
+void mthca_tavor_write_mtt_seg(struct mthca_dev *dev, struct mthca_mtt *mtt,
+                              int start_index, u64 *buffer_list, int list_len)
+{
+        u64 __iomem *mtts;
+        int i;
+
+        mtts = dev->mr_table.tavor_fmr.mtt_base + mtt->first_seg * MTHCA_MTT_SEG_SIZE +
+                start_index * sizeof (u64);
+        for (i = 0; i < list_len; ++i)
+                mthca_write64_raw(cpu_to_be64(buffer_list[i] | MTHCA_MTT_FLAG_PRESENT),
+                                  mtts + i);
+}
+
+void mthca_arbel_write_mtt_seg(struct mthca_dev *dev, struct mthca_mtt *mtt,
+                              int start_index, u64 *buffer_list, int list_len)
+{
+        __be64 *mtts;
+        dma_addr_t dma_handle;
+        int i;
+        int s = start_index * sizeof (u64);
+
+        /* For Arbel, all MTTs must fit in the same page. */
+        BUG_ON(s / PAGE_SIZE != (s + list_len * sizeof(u64) - 1) / PAGE_SIZE);
+        /* Require full segments */
+        BUG_ON(s % MTHCA_MTT_SEG_SIZE);
+
+        mtts = mthca_table_find(dev->mr_table.mtt_table, mtt->first_seg +
+                                s / MTHCA_MTT_SEG_SIZE, &dma_handle);
+
+        BUG_ON(!mtts);
+
+        for (i = 0; i < list_len; ++i)
+                mtts[i] = cpu_to_be64(buffer_list[i] | MTHCA_MTT_FLAG_PRESENT);
+
+        dma_sync_single(&dev->pdev->dev, dma_handle, list_len * sizeof (u64), DMA_TO_DEVICE);
+}
+
+int mthca_write_mtt(struct mthca_dev *dev, struct mthca_mtt *mtt,
+                    int start_index, u64 *buffer_list, int list_len)
+{
+        int size = mthca_write_mtt_size(dev);
+        int chunk;
+
+        if (dev->mr_table.fmr_mtt_buddy != &dev->mr_table.mtt_buddy)
+                return __mthca_write_mtt(dev, mtt, start_index, buffer_list, list_len);
+
+        while (list_len > 0) {
+                chunk = min(size, list_len);
+                if (mthca_is_memfree(dev))
+                        mthca_arbel_write_mtt_seg(dev, mtt, start_index,
+                                                  buffer_list, chunk);
+                else
+                        mthca_tavor_write_mtt_seg(dev, mtt, start_index,
+                                                  buffer_list, chunk);
+
+                list_len    -= chunk;
+                start_index += chunk;
+                buffer_list += chunk;
+        }
+
+        return 0;
+}
+
 static inline u32 tavor_hw_index_to_key(u32 ind)
 {
         return ind;
diff --git a/drivers/infiniband/hw/mthca/mthca_provider.c b/drivers/infiniband/hw/mthca/mthca_provider.c
index 7b96751695ea..0725ad7ad9bf 100644
--- a/drivers/infiniband/hw/mthca/mthca_provider.c
+++ b/drivers/infiniband/hw/mthca/mthca_provider.c
@@ -1015,6 +1015,7 @@ static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, struct ib_umem *region,
         int shift, n, len;
         int i, j, k;
         int err = 0;
+        int write_mtt_size;
 
         shift = ffs(region->page_size) - 1;
 
@@ -1040,6 +1041,8 @@ static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, struct ib_umem *region,
 
         i = n = 0;
 
+        write_mtt_size = min(mthca_write_mtt_size(dev), (int) (PAGE_SIZE / sizeof *pages));
+
         list_for_each_entry(chunk, &region->chunk_list, list)
                 for (j = 0; j < chunk->nmap; ++j) {
                         len = sg_dma_len(&chunk->page_list[j]) >> shift;
@@ -1047,14 +1050,11 @@ static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, struct ib_umem *region,
                                 pages[i++] = sg_dma_address(&chunk->page_list[j]) +
                                         region->page_size * k;
                                 /*
-                                 * Be friendly to WRITE_MTT command
-                                 * and leave two empty slots for the
-                                 * index and reserved fields of the
-                                 * mailbox.
+                                 * Be friendly to write_mtt and pass it chunks
+                                 * of appropriate size.
                                  */
-                                if (i == PAGE_SIZE / sizeof (u64) - 2) {
-                                        err = mthca_write_mtt(dev, mr->mtt,
-                                                              n, pages, i);
+                                if (i == write_mtt_size) {
+                                        err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
                                         if (err)
                                                 goto mtt_done;
                                         n += i;