e1000e: use hardware writeback batching
Most e1000e parts support batching writebacks. The problem with this is
that when some of the TADV or TIDV timers are not set, Tx can sit forever.
This is solved in this patch with write flushes using the Flush Partial
Descriptors (FPD) bit in TIDV and RDTR.
This improves bus utilization and removes partial writes on e1000e,
particularly from 82571 parts in S5500 chipset based machines.
Only ES2LAN and 82571/2 parts are included in this optimization, to reduce
testing load.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Emil Tantilov <emil.s.tantilov@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/drivers/net/e1000e/netdev.c b/drivers/net/e1000e/netdev.c
index c69563c..1aa4228 100644
--- a/drivers/net/e1000e/netdev.c
+++ b/drivers/net/e1000e/netdev.c
@@ -2650,6 +2650,26 @@
/* Tx irq moderation */
ew32(TADV, adapter->tx_abs_int_delay);
+ if (adapter->flags2 & FLAG2_DMA_BURST) {
+ u32 txdctl = er32(TXDCTL(0));
+ txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
+ E1000_TXDCTL_WTHRESH);
+ /*
+ * set up some performance related parameters to encourage the
+ * hardware to use the bus more efficiently in bursts, depends
+ * on the tx_int_delay to be enabled,
+ * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
+ * hthresh = 1 ==> prefetch when one or more available
+ * pthresh = 0x1f ==> prefetch if internal cache 31 or less
+ * BEWARE: this seems to work but should be considered first if
+ * there are tx hangs or other tx related bugs
+ */
+ txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
+ ew32(TXDCTL(0), txdctl);
+ /* erratum work around: set txdctl the same for both queues */
+ ew32(TXDCTL(1), txdctl);
+ }
+
/* Program the Transmit Control Register */
tctl = er32(TCTL);
tctl &= ~E1000_TCTL_CT;
@@ -2872,6 +2892,29 @@
e1e_flush();
msleep(10);
+ if (adapter->flags2 & FLAG2_DMA_BURST) {
+ /*
+ * set the writeback threshold (only takes effect if the RDTR
+ * is set). set GRAN=1 and write back up to 0x4 worth, and
+ * enable prefetching of 0x20 rx descriptors
+ * granularity = 01
+ * wthresh = 04,
+ * hthresh = 04,
+ * pthresh = 0x20
+ */
+ ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
+ ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
+
+ /*
+ * override the delay timers for enabling bursting, only if
+ * the value was not set by the user via module options
+ */
+ if (adapter->rx_int_delay == DEFAULT_RDTR)
+ adapter->rx_int_delay = BURST_RDTR;
+ if (adapter->rx_abs_int_delay == DEFAULT_RADV)
+ adapter->rx_abs_int_delay = BURST_RADV;
+ }
+
/* set the Receive Delay Timer Register */
ew32(RDTR, adapter->rx_int_delay);
@@ -4235,6 +4278,16 @@
/* Force detection of hung controller every watchdog period */
adapter->detect_tx_hung = 1;
+ /* flush partial descriptors to memory before detecting tx hang */
+ if (adapter->flags2 & FLAG2_DMA_BURST) {
+ ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
+ ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
+ /*
+ * no need to flush the writes because the timeout code does
+ * an er32 first thing
+ */
+ }
+
/*
* With 82571 controllers, LAA may be overwritten due to controller
* reset from the other port. Set the appropriate LAA in RAR[0]