MFC ixgbe(4) 3.1.13-k

[sbruno]

Summary

MFC ixgbe(4) 3.1.13-k

• r292674

Add support for X552 and X550T

• 292751 and 292697 added here to fix the build

Fix SFP module insertion post boot

• 293334

Fix VF handling of VLANS for Amazon Cloud

• 293338

Test Plan

-Basic Traffic tests (ftp, NFS, TCP, UDP)
-Cable plug/unplug and module detect
-Stress Traffic w/ Netperf (TCP/UDP IPv4/6)

[smh]

We have a version of just this but back ported to 10.2 running in production.

Comparing our diffs highlights the following:

1. Looks like your missing MFC of r285592, was that intentional?
2. We have one additional block, looking at head this looks like it should be present in your changes too wanted to double check it wasn't a miss-merge on either your or our side?

sys/dev/ixgbe/if_ixv.c
@@ -1703,17 +1694,11 @@ ixv_initialize_receive_units(struct adap
        reg |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
        IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(i), reg);
 
-       /* Setup the HW Rx Head and Tail Descriptor Pointers */
-       IXGBE_WRITE_REG(hw, IXGBE_VFRDH(rxr->me), 0);
-       IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rxr->me),
-           adapter->num_rx_desc - 1);
-
-       /* Set Rx Tail register */

[sbruno]

In D5117#108830, @smh wrote:

We have a version of just this but back ported to 10.2 running in production.

Comparing our diffs highlights the following:

1. Looks like your missing MFC of r285592, was that intentional?
2. We have one additional block, looking at head this looks like it should be present in your changes too wanted to double check it wasn't a miss-merge on either your or our side?

sys/dev/ixgbe/if_ixv.c
@@ -1703,17 +1694,11 @@ ixv_initialize_receive_units(struct adap
        reg |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
        IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(i), reg);
 
-       /* Setup the HW Rx Head and Tail Descriptor Pointers */
-       IXGBE_WRITE_REG(hw, IXGBE_VFRDH(rxr->me), 0);
-       IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rxr->me),
-           adapter->num_rx_desc - 1);
-
-       /* Set Rx Tail register */

I should have captured r285592 in the MFC of driver version 3.1.0 to stable 10. https://svnweb.freebsd.org/base?view=revision&revision=294061

The second point, looks like an svn mismerge. I'll fix that directly.

[sbruno]

Fix mismerge is ixv_initialize_receive_units as pointed out by smh.

[smh]

In D5117#108850, @sbruno wrote:

I should have captured r285592 in the MFC of driver version 3.1.0 to stable 10. https://svnweb.freebsd.org/base?view=revision&revision=294061

In that case you may be missing a few netmap bits according to our patch:

sys/dev/ixgbe/if_ix.c
@@ -446,9 +463,14 @@ ixgbe_attach(device_t dev)
 
    /* Allocate, clear, and link in our adapter structure */
    adapter = device_get_softc(dev);
-   adapter->dev = adapter->osdep.dev = dev;
+   adapter->dev = dev;
    hw = &adapter->hw;
 
+#ifdef DEV_NETMAP
+   adapter->init_locked = ixgbe_init_locked;
+   adapter->stop_locked = ixgbe_stop;
+#endif
+
    /* Core Lock Init*/
    IXGBE_CORE_LOCK_INIT(adapter, device_get_nameunit(dev));


sys/dev/ixgbe/ixgbe.h
@@ -555,6 +562,10 @@ struct adapter {
 #ifdef PCI_IOV
    struct ixgbe_vf     *vfs;
 #endif
+#ifdef DEV_NETMAP
+   void            (*init_locked)(struct adapter *);
+   void            (*stop_locked)(void *);
+#endif
 
    /* Misc stats maintained by the driver */
    unsigned long       dropped_pkts;


sys/dev/ixgbe/if_ixv.c
@@ -292,9 +292,14 @@ ixv_attach(device_t dev)
 
    /* Allocate, clear, and link in our adapter structure */
    adapter = device_get_softc(dev);
-   adapter->dev = adapter->osdep.dev = dev;
+   adapter->dev = dev;
    hw = &adapter->hw;
 
+#ifdef DEV_NETMAP
+   adapter->init_locked = ixv_init_locked;
+   adapter->stop_locked = ixv_stop;
+#endif
+
    /* Core Lock Init*/
    IXGBE_CORE_LOCK_INIT(adapter, device_get_nameunit(dev));

[sbruno]

In D5117#108861, @smh wrote:

In D5117#108850, @sbruno wrote:

I should have captured r285592 in the MFC of driver version 3.1.0 to stable 10. https://svnweb.freebsd.org/base?view=revision&revision=294061

In that case you may be missing a few netmap bits according to our patch:

sys/dev/ixgbe/if_ix.c
@@ -446,9 +463,14 @@ ixgbe_attach(device_t dev)
 
    /* Allocate, clear, and link in our adapter structure */
    adapter = device_get_softc(dev);
-   adapter->dev = adapter->osdep.dev = dev;
+   adapter->dev = dev;
    hw = &adapter->hw;
 
+#ifdef DEV_NETMAP
+   adapter->init_locked = ixgbe_init_locked;
+   adapter->stop_locked = ixgbe_stop;
+#endif
+
    /* Core Lock Init*/
    IXGBE_CORE_LOCK_INIT(adapter, device_get_nameunit(dev));


sys/dev/ixgbe/ixgbe.h
@@ -555,6 +562,10 @@ struct adapter {
 #ifdef PCI_IOV
    struct ixgbe_vf     *vfs;
 #endif
+#ifdef DEV_NETMAP
+   void            (*init_locked)(struct adapter *);
+   void            (*stop_locked)(void *);
+#endif
 
    /* Misc stats maintained by the driver */
    unsigned long       dropped_pkts;


sys/dev/ixgbe/if_ixv.c
@@ -292,9 +292,14 @@ ixv_attach(device_t dev)
 
    /* Allocate, clear, and link in our adapter structure */
    adapter = device_get_softc(dev);
-   adapter->dev = adapter->osdep.dev = dev;
+   adapter->dev = dev;
    hw = &adapter->hw;
 
+#ifdef DEV_NETMAP
+   adapter->init_locked = ixv_init_locked;
+   adapter->stop_locked = ixv_stop;
+#endif
+
    /* Core Lock Init*/
    IXGBE_CORE_LOCK_INIT(adapter, device_get_nameunit(dev));

the changes for osdep.dev seem to be deprecated completely by this driver update and the dev element of struct ixgbe_osdep has been removed. The NETMAP changes may be relevant, I'll poke at them.

[sbruno]

https://svnweb.freebsd.org/base/head/sys/dev/ixgbe/if_ix.c?r1=289238&r2=292674

Relevant svn revision.

[sbruno]

Address missing #ifdef DEV_NETMAP code pointed out by smh.

Compile tested amd64/i386 +netmap.

[smh]

Almost there. recomparing with our diff the only remaining missing block is the following which comes from between r283882 and r283883:

sys/dev/ixgbe/if_ix.c
@@ -3146,11 +3285,11 @@ ixgbe_initialize_receive_units(struct ad
 			srrctl &= ~IXGBE_SRRCTL_DROP_EN;
 		}
 
-		IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(i), srrctl);
+		IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(j), srrctl);
 
 		/* Setup the HW Rx Head and Tail Descriptor Pointers */
-		IXGBE_WRITE_REG(hw, IXGBE_RDH(i), 0);
-		IXGBE_WRITE_REG(hw, IXGBE_RDT(i), 0);
+		IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
+		IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
 
 		/* Set the driver rx tail address */
 		rxr->tail =  IXGBE_RDT(rxr->me);

[sbruno]

Fix mismerge of r283883 noted by smh to repair queue descriptor setup.

[jeffrey.e.pieper_intel.com]

I have this running on top of 10-STABLE (10.3-PRERELEASE) on X520, X540, X550, X552, and X552/X557. Everything seems to behaving as expected and doing "stuff". Will continue to run tests.

[ngie]

Other than the slight nitpicky spot with sys/conf/files that might make future merges a bit weird, LGTM!

sys/conf/files

1731–1732 ↗

(On Diff #12959)

This is in a different spot on head:

284767        erj dev/ixgbe/ix_txrx.c           optional ix inet | ixv inet \
215911        jfv       compile-with "${NORMAL_C} -I$S/dev/ixgbe"
292674     sbruno dev/ixgbe/ixgbe_osdep.c               optional ix inet | ixv inet \
292674     sbruno       compile-with "${NORMAL_C} -I$S/dev/ixgbe"
284767        erj dev/ixgbe/ixgbe_phy.c         optional ix inet | ixv inet \
171383        jfv       compile-with "${NORMAL_C} -I$S/dev/ixgbe"