Add facilities for tuning the "em" driver's interrupt delays without

recompiling the driver.  See the comments near the top of "if_em.h"
for descriptions of these delays.  Four new loader tunables control
the system-wide default values:

    hw.em.tx_int_delay
    hw.em.rx_int_delay
    hw.em.tx_abs_int_delay
    hw.em.rx_abs_int_delay

The tunables are specified in microseconds.  The valid range is
0-67108 usec., and 0 means that the timer is disabled.

There are also four new sysctls (actually, a set of four for each
"em" device in the system) to query and change the interrupt delays
after the system is up:

    hw.em0.tx_int_delay
    hw.em0.rx_int_delay
    hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters)
    hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters)

It seems to be OK to change these values even while the adapter is
passing traffic.

Approved by:	Prafulla Deuskar <pdeuskar@FreeBSD.ORG>
MFC after:	4 weeks

3 files changed, 148 insertions, 44 deletions

diff --git a/sys/dev/em/if_em.c b/sys/dev/em/if_em.c
index ceb4e1f9bda5..dd895894f9fd 100644
--- a/sys/dev/em/if_em.c
+++ b/sys/dev/em/if_em.c
@@ -163,6 +163,10 @@ static void em_print_debug_info(struct adapter *);
 static int  em_is_valid_ether_addr(u_int8_t *);
 static int  em_sysctl_stats(SYSCTL_HANDLER_ARGS);
 static int  em_sysctl_debug_info(SYSCTL_HANDLER_ARGS);
+static int  em_sysctl_int_delay(SYSCTL_HANDLER_ARGS);
+static void em_add_int_delay_sysctl(struct adapter *, const char *,
+				    const char *, struct em_int_delay_info *,
+				    int, int);
 
 /*********************************************************************
  *  FreeBSD Device Interface Entry Points                    
@@ -187,6 +191,23 @@ MODULE_DEPEND(em, pci, 1, 1, 1);
 MODULE_DEPEND(em, ether, 1, 1, 1);
 
 /*********************************************************************
+ *  Tunable default values.
+ *********************************************************************/
+
+#define E1000_TICKS_TO_USECS(ticks)	((1024 * (ticks) + 500) / 1000)
+#define E1000_USECS_TO_TICKS(usecs)	((1000 * (usecs) + 512) / 1024)
+
+static int em_tx_int_delay_dflt = E1000_TICKS_TO_USECS(EM_TIDV);
+static int em_rx_int_delay_dflt = E1000_TICKS_TO_USECS(EM_RDTR);
+static int em_tx_abs_int_delay_dflt = E1000_TICKS_TO_USECS(EM_TADV);
+static int em_rx_abs_int_delay_dflt = E1000_TICKS_TO_USECS(EM_RADV);
+
+TUNABLE_INT("hw.em.tx_int_delay", &em_tx_int_delay_dflt);
+TUNABLE_INT("hw.em.rx_int_delay", &em_rx_int_delay_dflt);
+TUNABLE_INT("hw.em.tx_abs_int_delay", &em_tx_abs_int_delay_dflt);
+TUNABLE_INT("hw.em.rx_abs_int_delay", &em_rx_abs_int_delay_dflt);
+
+/*********************************************************************
  *  Device identification routine
  *
  *  em_probe determines if the driver should be loaded on
@@ -305,14 +326,30 @@ em_attach(device_t dev)
 
 	/* Determine hardware revision */
 	em_identify_hardware(adapter);
+
+	/* Set up some sysctls for the tunable interrupt delays */
+	em_add_int_delay_sysctl(adapter, "rx_int_delay",
+	    "receive interrupt delay in usecs", &adapter->rx_int_delay,
+	    E1000_REG_OFFSET(&adapter->hw, RDTR), em_rx_int_delay_dflt);
+	em_add_int_delay_sysctl(adapter, "tx_int_delay",
+	    "transmit interrupt delay in usecs", &adapter->tx_int_delay,
+	    E1000_REG_OFFSET(&adapter->hw, TIDV), em_tx_int_delay_dflt);
+	if (adapter->hw.mac_type >= em_82540) {
+		em_add_int_delay_sysctl(adapter, "rx_abs_int_delay",
+		    "receive interrupt delay limit in usecs",
+		    &adapter->rx_abs_int_delay,
+		    E1000_REG_OFFSET(&adapter->hw, RADV),
+		    em_rx_abs_int_delay_dflt);
+		em_add_int_delay_sysctl(adapter, "tx_abs_int_delay",
+		    "transmit interrupt delay limit in usecs",
+		    &adapter->tx_abs_int_delay,
+		    E1000_REG_OFFSET(&adapter->hw, TADV),
+		    em_tx_abs_int_delay_dflt);
+	}
       
 	/* Parameters (to be read from user) */   
         adapter->num_tx_desc = EM_MAX_TXD;
         adapter->num_rx_desc = EM_MAX_RXD;
-        adapter->tx_int_delay = EM_TIDV;
-        adapter->tx_abs_int_delay = EM_TADV;
-        adapter->rx_int_delay = EM_RDTR;
-        adapter->rx_abs_int_delay = EM_RADV;
         adapter->hw.autoneg = DO_AUTO_NEG;
         adapter->hw.wait_autoneg_complete = WAIT_FOR_AUTO_NEG_DEFAULT;
         adapter->hw.autoneg_advertised = AUTONEG_ADV_DEFAULT;
@@ -1982,9 +2019,10 @@ em_initialize_transmit_unit(struct adapter * adapter)
         }
 
 	E1000_WRITE_REG(&adapter->hw, TIPG, reg_tipg);
-	E1000_WRITE_REG(&adapter->hw, TIDV, adapter->tx_int_delay);
+	E1000_WRITE_REG(&adapter->hw, TIDV, adapter->tx_int_delay.value);
 	if(adapter->hw.mac_type >= em_82540)
-		E1000_WRITE_REG(&adapter->hw, TADV, adapter->tx_abs_int_delay);
+		E1000_WRITE_REG(&adapter->hw, TADV,
+		    adapter->tx_abs_int_delay.value);
 
 	/* Program the Transmit Control Register */
 	reg_tctl = E1000_TCTL_PSP | E1000_TCTL_EN |
@@ -1999,7 +2037,7 @@ em_initialize_transmit_unit(struct adapter * adapter)
 	/* Setup Transmit Descriptor Settings for this adapter */   
 	adapter->txd_cmd = E1000_TXD_CMD_IFCS | E1000_TXD_CMD_RS;
 
-	if (adapter->tx_int_delay > 0)
+	if (adapter->tx_int_delay.value > 0)
 		adapter->txd_cmd |= E1000_TXD_CMD_IDE;
 
 	return;
@@ -2368,10 +2406,11 @@ em_initialize_receive_unit(struct adapter * adapter)
 
 	/* Set the Receive Delay Timer Register */
 	E1000_WRITE_REG(&adapter->hw, RDTR, 
-			adapter->rx_int_delay | E1000_RDT_FPDB);
+			adapter->rx_int_delay.value | E1000_RDT_FPDB);
 
 	if(adapter->hw.mac_type >= em_82540) {
-		E1000_WRITE_REG(&adapter->hw, RADV, adapter->rx_abs_int_delay);
+		E1000_WRITE_REG(&adapter->hw, RADV,
+		    adapter->rx_abs_int_delay.value);
 
                 /* Set the interrupt throttling rate.  Value is calculated
                  * as DEFAULT_ITR = 1/(MAX_INTS_PER_SEC * 256ns) */
@@ -2999,3 +3038,63 @@ em_sysctl_stats(SYSCTL_HANDLER_ARGS)
 
         return error;
 }
+
+static int
+em_sysctl_int_delay(SYSCTL_HANDLER_ARGS)
+{
+	struct em_int_delay_info *info;
+	struct adapter *adapter;
+	u_int32_t regval;
+	int error;
+	int usecs;
+	int ticks;
+	int s;
+
+	info = (struct em_int_delay_info *)arg1;
+	adapter = info->adapter;
+	usecs = info->value;
+	error = sysctl_handle_int(oidp, &usecs, 0, req);
+	if (error != 0 || req->newptr == NULL)
+		return error;
+	if (usecs < 0 || usecs > E1000_TICKS_TO_USECS(65535))
+		return EINVAL;
+	info->value = usecs;
+	ticks = E1000_USECS_TO_TICKS(usecs);
+	
+	s = splimp();
+	regval = E1000_READ_OFFSET(&adapter->hw, info->offset);
+	regval = (regval & ~0xffff) | (ticks & 0xffff);
+	/* Handle a few special cases. */
+	switch (info->offset) {
+	case E1000_RDTR:
+	case E1000_82542_RDTR:
+		regval |= E1000_RDT_FPDB;
+		break;
+	case E1000_TIDV:
+	case E1000_82542_TIDV:
+		if (ticks == 0) {
+			adapter->txd_cmd &= ~E1000_TXD_CMD_IDE;
+			/* Don't write 0 into the TIDV register. */
+			regval++;
+		} else
+			adapter->txd_cmd |= E1000_TXD_CMD_IDE;
+		break;
+	}
+	E1000_WRITE_OFFSET(&adapter->hw, info->offset, regval);
+	splx(s);
+	return 0;
+}
+
+static void
+em_add_int_delay_sysctl(struct adapter *adapter, const char *name,
+    const char *description, struct em_int_delay_info *info,
+    int offset, int value)
+{
+	info->adapter = adapter;
+	info->offset = offset;
+	info->value = value;
+	SYSCTL_ADD_PROC(&adapter->sysctl_ctx,
+	    SYSCTL_CHILDREN(adapter->sysctl_tree),
+	    OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW,
+	    info, 0, em_sysctl_int_delay, "I", description);
+}
diff --git a/sys/dev/em/if_em.h b/sys/dev/em/if_em.h
index 25327409a8d5..ca85fbb30777 100644
--- a/sys/dev/em/if_em.h
+++ b/sys/dev/em/if_em.h
@@ -294,6 +294,13 @@ typedef enum _XSUM_CONTEXT_T {
 	OFFLOAD_UDP_IP
 } XSUM_CONTEXT_T;
 
+struct adapter;
+struct em_int_delay_info {
+	struct adapter *adapter;	/* Back-pointer to the adapter struct */
+	int offset;			/* Register offset to read/write */
+	int value;			/* Current value in usecs */
+};
+
 /* Our adapter structure */
 struct adapter {
 	struct arpcom   interface_data;
@@ -320,10 +327,10 @@ struct adapter {
 	u_int16_t       link_speed;
 	u_int16_t       link_duplex;
 	u_int32_t       smartspeed;
-	u_int32_t       tx_int_delay;
-	u_int32_t       tx_abs_int_delay;
-	u_int32_t       rx_int_delay;
-	u_int32_t       rx_abs_int_delay;
+	struct em_int_delay_info tx_int_delay;
+	struct em_int_delay_info tx_abs_int_delay;
+	struct em_int_delay_info rx_int_delay;
+	struct em_int_delay_info rx_abs_int_delay;
 
 	XSUM_CONTEXT_T  active_checksum_context;
 
diff --git a/sys/dev/em/if_em_osdep.h b/sys/dev/em/if_em_osdep.h
index f18f446f0080..0bd806efd507 100644
--- a/sys/dev/em/if_em_osdep.h
+++ b/sys/dev/em/if_em_osdep.h
@@ -89,38 +89,36 @@ struct em_osdep
 	struct device     *dev;
 };
 
-#define E1000_WRITE_FLUSH(a) E1000_READ_REG(a, STATUS)
-
-#define E1000_READ_REG(a, reg) (\
-   bus_space_read_4( ((struct em_osdep *)(a)->back)->mem_bus_space_tag, \
-                     ((struct em_osdep *)(a)->back)->mem_bus_space_handle, \
-                     ((a)->mac_type >= em_82543) ? E1000_##reg : E1000_82542_##reg))
-
-#define E1000_WRITE_REG(a, reg, value) (\
-   bus_space_write_4( ((struct em_osdep *)(a)->back)->mem_bus_space_tag, \
-                     ((struct em_osdep *)(a)->back)->mem_bus_space_handle, \
-		     ((a)->mac_type >= em_82543) ? E1000_##reg : E1000_82542_##reg, \
-                     value))
-
-#define E1000_READ_REG_ARRAY(a, reg, offset) (\
- ((a)->mac_type >= em_82543) ? \
-   bus_space_read_4( ((struct em_osdep *)(a)->back)->mem_bus_space_tag, \
-                     ((struct em_osdep *)(a)->back)->mem_bus_space_handle, \
-                     (E1000_##reg + ((offset) << 2))): \
-   bus_space_read_4( ((struct em_osdep *)(a)->back)->mem_bus_space_tag, \
-                      ((struct em_osdep *)(a)->back)->mem_bus_space_handle, \
-                       (E1000_82542_##reg + ((offset) << 2))))
-
-
-#define E1000_WRITE_REG_ARRAY(a, reg, offset, value) (\
-  ((a)->mac_type >= em_82543) ? \
-      bus_space_write_4( ((struct em_osdep *)(a)->back)->mem_bus_space_tag, \
-                      ((struct em_osdep *)(a)->back)->mem_bus_space_handle, \
-                      (E1000_##reg + ((offset) << 2)), value): \
-      bus_space_write_4( ((struct em_osdep *)(a)->back)->mem_bus_space_tag, \
-                      ((struct em_osdep *)(a)->back)->mem_bus_space_handle, \
-                      (E1000_82542_##reg + ((offset) << 2)), value))
+#define E1000_WRITE_FLUSH(hw) E1000_READ_REG(hw, STATUS)
 
+/* Read from an absolute offset in the adapter's memory space */
+#define E1000_READ_OFFSET(hw, offset) \
+    bus_space_read_4( ((struct em_osdep *)(hw)->back)->mem_bus_space_tag, \
+                      ((struct em_osdep *)(hw)->back)->mem_bus_space_handle, \
+                      offset)
+
+/* Write to an absolute offset in the adapter's memory space */
+#define E1000_WRITE_OFFSET(hw, offset, value) \
+    bus_space_write_4( ((struct em_osdep *)(hw)->back)->mem_bus_space_tag, \
+                       ((struct em_osdep *)(hw)->back)->mem_bus_space_handle, \
+                       offset, \
+                       value)
+
+/* Convert a register name to its offset in the adapter's memory space */
+#define E1000_REG_OFFSET(hw, reg) \
+    ((hw)->mac_type >= em_82543 ? E1000_##reg : E1000_82542_##reg)
+
+#define E1000_READ_REG(hw, reg) \
+    E1000_READ_OFFSET(hw, E1000_REG_OFFSET(hw, reg))
+
+#define E1000_WRITE_REG(hw, reg, value) \
+    E1000_WRITE_OFFSET(hw, E1000_REG_OFFSET(hw, reg), value)
+
+#define E1000_READ_REG_ARRAY(hw, reg, index) \
+    E1000_READ_OFFSET(hw, E1000_REG_OFFSET(hw, reg) + ((index) << 2))
+
+#define E1000_WRITE_REG_ARRAY(hw, reg, index, value) \
+    E1000_WRITE_OFFSET(hw, E1000_REG_OFFSET(hw, reg) + ((index) << 2), value)
 
 #endif  /* _FREEBSD_OS_H_ */