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>
diff --git a/drivers/infiniband/hw/mthca/mthca_provider.c b/drivers/infiniband/hw/mthca/mthca_provider.c
index 7b96751..0725ad7 100644
--- a/drivers/infiniband/hw/mthca/mthca_provider.c
+++ b/drivers/infiniband/hw/mthca/mthca_provider.c
@@ -1015,6 +1015,7 @@
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 @@
i = n = 0;
+ write_mtt_size = min(mthca_write_mtt_size(dev), (int) (PAGE_SIZE / sizeof *pages));
+
list_for_each_entry(chunk, ®ion->chunk_list, list)
for (j = 0; j < chunk->nmap; ++j) {
len = sg_dma_len(&chunk->page_list[j]) >> shift;
@@ -1047,14 +1050,11 @@
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;