xref: /titanic_52/usr/src/uts/common/io/e1000api/README.illumos (revision 42cc51e07cdbcad3b9aca8d9d991fc09b251feb7)

#
# This file and its contents are supplied under the terms of the
# Common Development and Distribution License ("CDDL"), version 1.0.
# You may only use this file in accordance with the terms of version
# 1.0 of the CDDL.
#
# A full copy of the text of the CDDL should have accompanied this
# source.  A copy of the CDDL is also available via the Internet at
# http://www.illumos.org/license/CDDL.
#

#
# Copyright (c) 2013 Joyent, Inc.  All rights reserved.
#

Historically e1000g and igb were maintained by two different teams at Sun and
thus while they used identical common code from Intel, they each only ever used
portions of it and were not kept in sync with one another. To help make
maintenance and the adding of new devices easier in illumos, we have gone
through and made it so that igb and e1000g rely on the same set of common code;
however, this code is not in its own module, each has its own copy of the code
compiled into it for various reasons which will be laid out below.

As part of the interface with the common code, the driver is required to define
an e1000_osdep.h. Currently each version of the driver defines its *own* version
of this header file in their own driver specific directory. However, the code
that implements this is different in each directory, specifically e1000g_osdep.c
and igb_osdep.c. It's important that they have different names and not be called
the same thing due to how the uts makefiles work.

Deviations from the common FreeBSD code:

We have a few differences from the common version of the FreeBSD code that exist
so that we can both gather firmware information and that have workarounds for
older chipsets. While, we would like to get that to be synced up and common, it
is not currently.

Energy Efficient Ethernet (EEE) is not enabled by default. This technology was
introduced with the I350 family of parts in the igb driver. However, there have
been issues seen with it in the wild and thus we opt to disable it by default
until tests have proven that there are no longer problems.

To help make that easier, we've documented here what these extra definitions
are. DO NOT just blindly copy over new common code. There is more work that
needs to be done in terms of changed interfaces and expectations for the
drivers.

# e1000_defines.h

In e1000_defines.h we add the following three definitions which are not
currently present. These definitions allow us to attach firware revisions and
other information to the devinfo tree.

#define		NVM_VERSION			0x0005
#define		NVM_OEM_OFFSET_0		6
#define		NVM_OEM_OFFSET_1		7

# Workarounds for the 82541 and 82547

There are various workarounds in place for the 82541 and 82547 due to errata
that exist for these devices. This has traditionally been a part of the common
code. Until this can get merged into the common code completely, we've spearted
out the changes that are the actual C functions into
uts/common/io/e1000g/e1000g_workarounds.c. However, this alone is not
sufficient. You must make sure that in e1000_hw.h that the struc
e1000_dev_spec_82541 actually looks like the following:

struct e1000_dev_spec_82541 {
	enum e1000_dsp_config dsp_config;
	enum e1000_ffe_config ffe_config;
	u32 tx_fifo_head;
	u32 tx_fifo_start;
	u32 tx_fifo_size;
	u16 spd_default;
	bool phy_init_script;
	bool ttl_workaround;
};

# EEE

By default we disable all support for EEE. To cause this to happen you must
make the following change in e1000_82575.c's init_mac_params.

From:
 394         /* Enable EEE default settings for EEE supported devices */
 395         if (mac->type >= e1000_i350)
 396                 dev_spec->eee_disable = FALSE;
To:
 394         /* Enable EEE default settings for EEE supported devices */
 395         if (mac->type >= e1000_i350)
 396                 dev_spec->eee_disable = TRUE;

Future work:

The next step here is to take the osdep portions and merge them. That would
allow us to build one common misc module e1000api that both igb and e1000g
depend on rather than building separate copies of the common code into each
driver. Another potential option which may prove to have less value is to take
all of the gld and ddi logic and have one driver export that leaving e1000g and
igb as small stubs which depend on that. Note however, that the latter is not
how our upstream is currently structuring their igb and em (FreeBSD's e1000g)
drivers.