[intel] Push new RX descriptors in batches

Inside a virtual machine, writing the RX ring tail pointer may incur a
substantial overhead of processing inside the hypervisor.  Minimise
this overhead by writing the tail pointer once per batch of
descriptors, rather than once per descriptor.

Profiling under qemu-kvm (version 1.6.2) shows that this reduces the
amount of time taken to refill the RX descriptor ring by around 90%.

Signed-off-by: Michael Brown <mcb30@ipxe.org>

diff --git a/src/drivers/net/intel.c b/src/drivers/net/intel.c
index 5abcdd7f82411e5858b8571564cdab2fe624b7d3..6684bdbd9c4896aa0c290287580ea37a6c6c8ba7 100644 (file)
--- a/src/drivers/net/intel.c
+++ b/src/drivers/net/intel.c
@@ -456,19 +456,20 @@ void intel_refill_rx ( struct intel_nic *intel ) {
        unsigned int rx_idx;
        unsigned int rx_tail;
        physaddr_t address;
+       unsigned int refilled = 0;
 
+       /* Refill ring */
        while ( ( intel->rx.prod - intel->rx.cons ) < INTEL_RX_FILL ) {
 
                /* Allocate I/O buffer */
                iobuf = alloc_iob ( INTEL_RX_MAX_LEN );
                if ( ! iobuf ) {
                        /* Wait for next refill */
-                       return;
+                       break;
                }
 
                /* Get next receive descriptor */
                rx_idx = ( intel->rx.prod++ % INTEL_NUM_RX_DESC );
-               rx_tail = ( intel->rx.prod % INTEL_NUM_RX_DESC );
                rx = &intel->rx.desc[rx_idx];
 
                /* Populate receive descriptor */
@@ -477,20 +478,24 @@ void intel_refill_rx ( struct intel_nic *intel ) {
                rx->length = 0;
                rx->status = 0;
                rx->errors = 0;
-               wmb();
 
                /* Record I/O buffer */
                assert ( intel->rx_iobuf[rx_idx] == NULL );
                intel->rx_iobuf[rx_idx] = iobuf;
 
-               /* Push descriptor to card */
-               profile_start ( &intel_vm_refill_profiler );
-               writel ( rx_tail, intel->regs + intel->rx.reg + INTEL_xDT );
-               profile_stop ( &intel_vm_refill_profiler );
-
                DBGC2 ( intel, "INTEL %p RX %d is [%llx,%llx)\n", intel, rx_idx,
                        ( ( unsigned long long ) address ),
                        ( ( unsigned long long ) address + INTEL_RX_MAX_LEN ) );
+               refilled++;
+       }
+
+       /* Push descriptors to card, if applicable */
+       if ( refilled ) {
+               wmb();
+               rx_tail = ( intel->rx.prod % INTEL_NUM_RX_DESC );
+               profile_start ( &intel_vm_refill_profiler );
+               writel ( rx_tail, intel->regs + intel->rx.reg + INTEL_xDT );
+               profile_stop ( &intel_vm_refill_profiler );
        }
 }