xref: /illumos-gate/usr/src/uts/common/io/e1000g/README (revision d2b9c6763c7b3aad7a4c2609ca71e0b02114bca9)

#
# This file is provided under a CDDLv1 license.  When using or
# redistributing this file, you may do so under this license.
# In redistributing this file this license must be included
# and no other modification of this header file is permitted.
#
# CDDL LICENSE SUMMARY
#
# Copyright(c) 1999 - 2007 Intel Corporation. All rights reserved.
#
# The contents of this file are subject to the terms of Version
# 1.0 of the Common Development and Distribution License (the "License").
#
# You should have received a copy of the License with this software.
# You can obtain a copy of the License at
#	http://www.opensolaris.org/os/licensing.
# See the License for the specific language governing permissions
# and limitations under the License.
#
#
# Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
# Use is subject to license terms of the CDDLv1.
#
#
# ident	"%Z%%M%	%I%	%E% SMI"
#

4.0.4 code drop from Intel on 1/30/2003
=======================================
  This version is used as the base for putback into both s10 and
  s9u4.

  Following changes were done to this code base before putting
  back into solaris trees (S10 & S9U4):

  1) Changes to e1000_hw.c to remove build warnings

     + diff /home/naroori/grizzly/sources/e1000g/1_4.0.4/e1000_hw.c .
     2616c2616
     <  phy_info->downshift = hw->speed_downgraded;
     ---
     >  phy_info->downshift = (int)hw->speed_downgraded;
     2684c2684
     <  phy_info->downshift = hw->speed_downgraded;
     ---
     >  phy_info->downshift = (int)hw->speed_downgraded;

  2) Changes to e1000g_dlpi.c for ipv6 plumbing problem and
     problem with default MTU size of 2024.

     + diff /home/naroori/grizzly/sources/e1000g/1_4.0.4/e1000g_dlpi.c .
     313a314
     >  /* the following causes problem with ipv6 plumbing...
     314a316
     >  */
     628c630,632
     <    if (Adapter->Shared.max_frame_size == FRAME_SIZE_UPTO_16K)
     ---
     >    if (Adapter->Shared.max_frame_size == ETHERMTU) {
     >    }
     >    else if (Adapter->Shared.max_frame_size == FRAME_SIZE_UPTO_16K) {
     630c634,635
     <    else
     ---
     >    }
     >    else  {
     631a637
     >    }

  3) Changes to e1000g_main.c for version string modification,
     removing console messages on bootup and problems with default
     MTU size of 2024.

     + diff /home/naroori/grizzly/sources/e1000g/1_4.0.4/e1000g_main.c .
     59c59
     < static char e1000g_version[]="Driver Ver. 4.0.4-beta";
     ---
     > static char e1000g_version[]="Driver Ver. 4.0.4";
     937c937
     <        e1000g_log(Adapter,CE_CONT,"\n%s, %s\n",AdapterName,e1000g_version);
     ---
     >        cmn_err(CE_CONT,"!%s, %s\n",AdapterName,e1000g_version);
     3000c3000
     <        Adapter->Shared.max_frame_size = FRAME_SIZE_UPTO_2K;
     ---
     >        Adapter->Shared.max_frame_size = ETHERMTU;
     3013c3013
     <        Adapter->Shared.max_frame_size = FRAME_SIZE_UPTO_2K;
     ---
     >        Adapter->Shared.max_frame_size = ETHERMTU;



Bug fixes to 4.0.4:
===================
   1) bug#4829398 code cleanup for ethernet bootup messages


4.0.12 code drop from Intel on 3/18/2003
========================================

  This drop had extensive changes from 4.0.4. Many changes after
  4.0.4 were realted to new hardware (a.k.a. BayCity chipset),
  not relevant to Grizzly.After looking at the changelog from Intel,
  we decided not to do a putback but rather take only high priority
  fixes from this drop and generate a new version 4.0.4.1.

  Here is the changelog supplied by Intel for this drop:

    ________________________________________________________________________
    Component Name: Unix-Solaris_8254x_BayCity_Intel
    Description: Bay City software component for gigabit products
    ________________________________________________________________________


    ---Version: 4.0.13  Date: 3/14/2003 5:00:00 PM---
    Added transmit mutex locking to the 82547 workaround code.

    ---Version: 4.0.12  Date: 3/10/2003 8:01:00 PM---
    Added new shared code with LED link speed fix

    ---Version: 4.0.11  Date: 3/7/2003 5:57:00 PM---
    Updated the Tanacross FIFO hang workaround to reset FIFO pointers instead of sending autonomous packets.

    ---Version: 4.0.10  Date: 3/4/2003 2:39:00 PM---
    Removed an extra sdu adjustment that was left in by accident

    ---Version: 4.0.9   Date: 3/4/2003 11:13:00 AM---
    Removed the first call to adapter_stop.  This was called before the HW struct was properly initialized, and could cause hangs.

    Changed reported ethernet sdu from 2024 to 1500, as driver could be configured for invalid frame MTU sizes in Solaris 9.

    Added Tanacross workaround for FIFO hang.

    ---Version: 4.0.8   Date: 2/25/2003 2:27:00 PM---
    Shared code update including Phy init script.
    Smartspeed workaround added.  e1000g_smartspeed called from LocalTimer function every two seconds when link is down.

    ---Version: 4.0.7   Date: 2/18/2003 11:41:00 AM---
    Updated to include bug fixes from the shared code.

    ---Version: 4.0.5   Date: 1/31/2003 10:13:00 AM---
    Support for new hardware changes
    Removed the extra ack sent at the end of the attach_req routine.  This extra ack previously made IPv6 plumbing on Solaris 9 fail.


4.0.4.1
=======
   This version is not an Intel drop but 4.0.4 merged with two
   high priority fixes from 4.0.12. We will give back the 4.0.4.1
   driver back to Intel, per their request, so they have the exact
   code that we are releasing through Solaris.

   The following are the two fixes selected for inclusion from
   4.0.12 codebase:

   1) Removing the first call to adapter_stop as discussed in
      4.0.12 changelog.

   2) Corrections to MTU size calculations first introduced
      in our 4.0.4 putback to bring them into sync with 4.0.12
      MTU size calculations.

   We also added README file (current file you are reading) during
   this putback.

4.0.4.2
=======
  This version has the following fixes:

  1) bug#4857098: e1000g driver assumes another intel card as its own.
     The fix involved eliminating 1009, 1012, 1015, 1016, 1017
     device-ids from e1000g nodelist so that they do not conflict
     with iprb's subsystem-ids.

     This fix has to be reverted later once we have a proper fix for
     bug#4859426.

   2) bug#4853683: e1000g lacks a large number of expected kstats
      The fix involved three steps: a) Renaming some of the existing
      kstats to comply with WDD documentation b) Computing some of
      those which are not being tracked c) Supplying zero for the rest
      the error stats which are not being tracked.

      This fix has to be revisited to improvise the statistics in
      future.

   3) bug#4862529: System panics during configuring devices on a Dell 4-way
      servers. The fix involved removing all the devic-ids from nodelist
      except for 1010. This means that we support now only the grizzly
      on-board gigabit chipset and "Intel Pro/1000 MT Dual Server Adapter".

      This fix has to be reverted later once we have a proper fix for
      bug#4859426.

4.0.4.3
=======
  This version has the following fixes:

  1) bug#4874857: ipv6 ping fails on bootup for e1000g.
	 The fix involved implementing multicast tables both at per stream
	 level and per device level.

  2) bug#4875743: null pointer dereference in ddi_dma_addr_bind_handle.
	 The fix involved adding LastTxDescriptorBuf to struct e1000g and
	 handling consequent proper calculations for LastTxDescriptor.

  3) bug#4882541: setting up ethernet address causes machine to panic.
     The fix involved removing a spurious log message which was causing
	 the panic.

  4) bug#4883302: Reinstate the removed device-ids - phase 2.
	 The fix involved some chip initializations to handle 1000 id.

  5) bug#4919285: device id 8086,100e needs to be added
	 The fix needed for the Intel Precision 350 workstation

4.2.11
======

We need to sync up with the latest Intel code base for e1000g driver.

Intel is currently at version 4.2.11 while we are on 4.0.4 branch.
The exact version under 4.0.4 branch is 4.0.4.3. Since we are on
the branch while Intel is working on the main trunk, we have had
lot of friction in getting timely fixes from Intel. The current
syncing up with their main trunk lets us a smooth migration path.

The new code base  also would provide us with support for additional
cards including Quad port cards (See bugs: 4921433, 4914460).
In our old 4.0.4 branch, we had to drop support for a lot of cards
for various problems; with the current syncing up, we would be able
to support so many various cards except for the cards clashing with
iprb driver.

The new code base also has support for additional new chipsets. The code
base also benefits from any fixes to the shared code base with other
operating system drivers.

The new codebase also has lot of cleanups in terms of various #ifdefs
and should be much simpler to maintain.

So the list of actual ids take out are:
1009
1012
1013
1015
1016
1017

4.2.15
======
This intel code drop addresses the following bug:

    4909863: Bad trap panic in e1000g_rtn_buf

4.2.16
======
  This version has the following fix:
   4932422 interconnect remains faulted after adapter disconnected and then reconnected

Notes on packaging:
===================
   Intel names the package as INTCGigaE while we chose to rename
   it as SUNWintgige as per Sun packaging conventions.

4.2.17
======
  This version has the following fixes:
   4971416 e1000g seems to hang system when no network connected
   4997067 e1000gattach panicks when the going gets tough
   4990064 e1000g leaks memory when it runs out of transmit descriptors

4.2.17.1
======
  This version has the following fix:
   5014625 e1000g driver panicked when snoop(1M) exited

4.2.17.2
======
  This version has the following fix:
   4940870 e1000g: Dependence on instance initialization order disables .conf properties
   5033616 e1000g panicked after rem_drv while some service still binds to the driver
   5037062 strings have crept back in e1000g/adpu320 related mutex_init()s
   5050268 Some e1000g flag operations are not correct.

4.2.17.3
======
  This version has the following fix:
   5067162 e1000g cannot transmit heavy UDP

4.2.17.4
======
  This version has the following fix:
   5010052 The stress load/unload test of Intel PWLA8494MT Quad Port NIC causes system hung
   5050265 E1000g does not support IPV6 fast path. This will impact IPV6 performance.
   5050369 E1000g locks the system if plumb/unplumb under heavy workload and snoop.
   5050415 E1000g will accepts ethernet packets which are not addressed to it.
   5055132 E1000g crashes when injected with Jumbo Frame packets.

4.2.17.5
======
  This version has the following fix:
   6175307 system crash with memory size of 11GB
   6176031 the sequence of releasing resource error in e1000gattach
   6180180 PktAlloc - Could not bind rbuf packet as single frag. Count - 2
   6185022 Port hang when unplumb interface under heavy traffic

4.2.17.6
======
  This version has the following fix:
   6196426 e1000g should support NICs that use the internal serdes for external connectivity

4.2.17.7
======
  This version has the following fix:
   4921433 e1000g: Should not probe by using VID/DID table to identify adapters

4.2.17.8
======
  This version has the following fix:
   6213101 64bit DMA max address space should be supported by e1000g driver
   6215261 e1000g driver 211% slower in AMD64 than 32 bit e1000g driver (S10_73, Single CPU V20z)

4.2.17.9
======
  This version has the following fix:
   6256212 fix for 6213101 breaks e1000g for 32-bit kernels on machines with >4GB of physical memory

4.2.17.10
======
  This version has the following fix:
   6203602 e1000g needs to support the Intel PCI-E card
   6217046 we need to merge the Intel code (version 4.2.28) into e1000g

5.0.0
======
  This version has the following fix:
   PSARC/2005/470 E1000g Nemo Migration
   6221079 e1000g driver interface conversion from DLPI to GLD3.0(nemo)
   6224901 e1000g fails to configure VLANs on a Sun v65x
   6286414 snmp: ifOperStatus for e1000g driver doesn't change when cable is unplugged.
   6238486 e1000g: poor performance in web server benchmark, high CPU utilization noted
   6293265 e1000g_intr doesn't use mutex at all
   6304657 e1000g_PktAlloc_buffers() causes a panic when ddi_dma_addr_bind_handle() fails.
   4913292 e1000g: does not support PSARC 2003/264 Extended IP cksum_offload
   5071610 e1000g can't send any packets after a jumbo-packet forwarding test.
   5082975 in Jumbo enable mode, netperf indicates a lower throughput, compared to Jumbo of
   5085394 e1000g driver hangs system during 64byte stress testing
   6174349 netlb support needed for Solaris e1000g driver
   6207682 e1000g needs link up/down support
   6211068 e1000g has some input error packets during stress test
   6212810 e1000g caught sending unsolicited M_PROTO/M_PCPROTO messages
   6219298 warnings about "inconsistent getinfo(9E) implementation"
   6226217 e1000g should support trunking
   6239279 Network stess test fail on 82545GM chip on IBM Blade HS40
   6242612 e1000g needs to support SPARC
   6247936 In e1000g transmission, ddi_dma_addr_bind_handle() returns more cookies than speficied
   6273730 e1000g causes memory leaks within Freelist_alloc and PktAlloc_buffers
   5073739 find some memory leaks after load_unload test
   6287250 e1000g may need to use new Intel-recommended semaphore mechanism

5.0.1
======
  This version has the following fix:
   6313609 2-port Intel NIC (part# 370-6687) performs poorly compared to 1-port Intel NIC (part# 370-6685)
   6315002 gcc and e1000g don't get along

5.0.2
======
  This version has the following fix:
   6319100 Tier0 Ontario machine does not boot net with sparc e1000g driver substitution for sparc ipge driver

5.0.3
======
  This version has the following fix:
   6327693 e1000g should use MSIs when possible
   6333461 panic[cpu0]/thread=d63f4de0: assertion failed: !dma->dp_inuse, file: ../../i86pc/io/rootnex.c

5.0.4
======
  This version has the following fix:
   6351378 e1000g panic seen after bfu'ing latest onnv nightly

5.0.5
======
  This version has the following fix:
   6303973 System hard hangs when sharing interrupts (in PIC) mode
   6323732 e1000g reports incorrect packet statistics
   6333880 Customer is requesting driver for a newIntel 82573 NIC
   6336163 netlb support needed for Solaris (sparc) e1000g driver
   6339377 e1000g.conf needs to be simplified
   6350407 e1000 card in Gateway E2300 does not actually send packets in S10U1
   6364052 e1000g driver should provide interface to ndd utility
   6367490 e1000g has problem in deliver UDP fragment with addin card

5.0.6
======
  This version has the following fix:
   4967732 e1000g: high mutex contention in e1000g_start
   6316907 e1000g: bcopy() brings better performance in transmitting and receiving small packet
   6371753 aggregation of e1000g NICs fails to provide connectivity unless snoop is running
   6397163 Jumbo frame statistics always counts zero with e1000g driver
   6401903 Memory leak in e1000g driver

5.0.7
======
  This version has the following fix:
   6337450 e1000g interfaces take much much longer to plumb than bge interfaces on the V40.
   6392401 Intel Ophir (82571) e1000g driver fails to attach on sun4u Boston with snv_33
   6401659 vlan header getting corrupted due to memmove in e1000g sendPacket
   6407374 e1000g still uses some old ddi interrupt routines that need to be changed
   6411339 BAD TRAP: e1000g_fill_tx_ring NULL pointer dereference

5.0.8
======
  This version has the following fix:
   6399084 ipmp doesn't work with e1000g
   6412153 ndd force speed is not consistent with e1000g.conf and e1000g(7D)
   6415019 system panic when plumb e1000g interface up on SPARC
   6415416 netlbtest does not run on Intel(R) PRO/1000 Network controller

5.0.9
======
  This version has the following fix:
  6337987 Serial Over Lan issue when e1000g driver is used

5.1.0
======
  This version has the following fix:
  PSARC/2006/248 Nemo MAC-Type Plugin Architecture
  PSARC/2006/249 Nemo Changes for Binary Compatibility
  6384371 GLDv3 only supports Ethernet
  6242059 nemo drivers must not know the size of the mac_t structure
  6226635 MAC stats interface could cause problems with binary compatibility

5.1.1
======
  This version has the following fix:
   6384266 e1000g driver should provide external loopback support
   6428883 NEM e1000g drive netlbtest intermittent failed when running Sunvts FUNC test with cputest,vmemtest..
   6430731 e1000g.conf file removed on upgrade
   6431317 add support for intel Pro/1000 PT Adapter
   6433236 Update the e1000g driver with the latest shared code from Intel
   6439589 e1000g: whines about checksum capability, unclear what user should do
   6441386 The default value of some e1000g parameters may impact rx performance
   6445239 assertion failed: cookie_count == 1, file: ../../common/io/e1000g/e1000g_alloc.c, line: 422
   6447914 e1000g should support setting of multiple unicast addresses
   6449595 Ophir LAA driver workaround is not working for e1000g Driver

5.1.2
======
  This version has the following fix:
   6450683 Boston panics with e1000g under stress
   6463536 BAD TRAP panic for removing aggregation
   6463756 e1000g - link_speed / kstat output
   6464426 e1000g driver needs a workaround to solve an ESB2 hardware issue

5.1.3
======
  This version has the following fix:
   6436387 e1000g panic while using cfgadm to unconfigure the pcie slot
   6460403 disconnect of pcieX that has ophir with network if plumbed crashes the system
   6466962 S10u2 e1000g 5.0.9 adv_autoneg_cap is not reflected on the kstat -p
   6470233 e1000g_fill_tx_ring doesn't always update TDT properly

5.1.4
======
  This version has the following fix:
   6462893 HCTS network test failed on thumper machine due to e1000g issue
   6467860 e1000g configured/used as cluster interconnect goes up and down when connected back to back

5.1.5
======
  This version has the following fix:
   6479912 Interrupt throttling tunable needed
   6492025 northstar MMF e1000g external loopback fail.

5.1.6
======
  This version has the following fix:
   6482530 Need Ethernet driver for Intel G965 chipset
   6491179 link aggregation with e1000g does not work unless snoop is running
   6494743 e1000g 5.1.4 driver does not work with s11_52
   6502458 e1000g is open source, move the source from usr/closed to use/src
   6505360 e1000g Makefile should not include "-N drv/dld" in the LDFLAGS

5.1.7
======
  This version has the following fix:
   6454375 e1000g link flaps at initialization, triggering failovers
   6472255 e1000g can't restore to 1000M with ndd setting
   6496763 e1000g should free packets when link is down
   6501294 "eeprom checksum failed" with pci8086,108c device
   6504688 e1000g.conf settings are inconsistent with ndd output
   6505445 e1000g : when all advertised capabilities are set to 0, ndd puts all of them 1
   6519690 e1000g should not print the link up/down messages to console
   6531474 Fatal PCIe Fabric Error panics on T2000 when using jumbo frames on e1000g interfaces
   6535712 e1000g: the processing of the checksum flags should be protected by tx_lock

5.1.8
======
  This version has the following fix:
   6548711 e1000g: recursive mutex_enter in e1000g_link_check() on ESB2 platforms
   6550086 e1000g: detaching driver immediately after attach induces panic

5.1.9
======
  This version has the following fix:
   6507422 Dynamic Reconfiguration detach fails for e1000g

5.1.10
======
  This version has the following fix:
   6490623 Some networking problems with Solaris_b44_64 domU(using solaris_b44_64 dom0)
   6510396 system panicked in e1000g_82547_timeout
   6554976 e1000g driver does not support 10D5 device - Sun Pentwater PEM quad port

5.1.11
======
  This version has the following fix:
   6552853 system panics in e1000g_alloc_dvma_buffer during hotplug testing

5.2.0
======
  This version has the following fix:
   6535620 e1000g needs to support ICH9 devices
   6572330 e1000g: integrate the latest Intel refactored shared code
   6573381 e1000g receiving VLAN tagged frames does not do hardware checksumming

5.2.1
======
  This version has the following fix:
   6594676 e1000g should free private dips while no rx buffers are held by upper layer