pfSense

It doesn't happen on any Intel 10G cards with SFP+ that we have around. The XG-C2758 has no such problem.

XG-C2758 uses a 82599.

I don't know what other "intel 10G cards with SFP+" you tried, so I can't comment.

The device on the 1540-D identifies as:
[2.3-BETA][root@xg-1540-1.localdomain]/root: pciconf -l | grep ix
ix0@pci0:3:0:0: class=0x020000 card=0x15ad15d9 chip=0x15ad8086 rev=0x00 hdr=0x00
ix1@pci0:3:0:1: class=0x020000 card=0x15ad15d9 chip=0x15ad8086 rev=0x00 hdr=0x00

This is known to the driver as: IXGBE_DEV_ID_X550EM_X_10G_T

ixgbe_type.h:#define IXGBE_DEV_ID_X550EM_X_10G_T 0x15AD

(this is on the machine that you apparently upgraded to the 3.1.14 driver).

The driver in -HEAD is 3.1.13-k, (FreeBSD-10.2 is apparently the same base driver from Intel).

https://github.com/pfsense/FreeBSD-src/blob/devel/sys/dev/ixgbe/if_ix.c#L51 says:
char ixgbe_driver_version[] = "3.1.13-k";

There are a number of differences specific to IXGBE_DEV_ID_X550EM_X_10G_T between the two revisions of the base driver, especially this:

        if (hw->device_id == IXGBE_DEV_ID_X550EM_X_10G_T) {
                /* Config MDIO clock speed before the first MDIO PHY access */
                hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
                hlreg0 &= ~IXGBE_HLREG0_MDCSPD;
                IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
        }

found in ixgbe_reset_hw_X550em() in 3.1.14 driver, but not in the 3.1.13k driver. There are also a lot of minor (but could be significant) changes to the way link speeds are negotiated.

Assigned to Luiz.

all the circumstances that failed before are good on the latest 2.3 snapshot