ethernet controllers. This adds support for the 3Com 3c996-T, the
SysKonnect SK-9D21 and SK-9D41, and the built-in gigE NICs on
Dell PowerEdge 2550 servers. The latter configuration hauls ass:
preliminary measurements show TCP speeds of over 900Mbps using
only normal size frames.
TCP/IP checksum offload, jumbo frames and VLAN tag insertion/stripping
are supported, as well as interrupt moderation.
Still need to fix autonegotiation support for 1000baseSX NICs, but
beyond that, driver is pretty solid.
survive a sysinstall Ctrl-C -> 'Restart'. This fixes another annoying
bug where restarting sysinstall will try to reload kernel modules and
do other external things that have already been done. For now, use
these persistent variables to keep track of module, usbd, and pccardd
initialization.
Bug found by: rwatson
MFC after: 1 week
environment. This fixes an annoying bug where hitting Ctrl-C and
telling sysinstall to 'restart' will do no such thing since many of
the options are still set and so you won't be prompted for them
again.
MFC after: 1 week
variable to check for debug functionality. Previously, you had to set
both 'debug' and 'SYSINSTALL_DEBUG' to get a log of sysinstall's
activities. Now, only 'debug' is necessary.
useful for post install configuration or other cases that might not be
handled by usb.c. (usb.c already sets usbd_enable iff sysinstall
detects usb during install).
PR: bin/18946
Submitted by: Peter van Heusden <pvh@egenetics.com>
Reviewed by: jhb
and RTSOL in sysinstall. If the respective TRY_FOO variable is set to
"YES" then it will be tried without prompting the user.
However, if the TRY_FOO variable is set to "NO" then the user will not
be prompted for a choice. This is the correct behavior, since we want
people to be able to script sysinstall in either case.
However, the default TRY_FOO variable has been "NO" since 1999. This
is incorrect, and when the logic was corrected in tcpip.c this has the
effect of never giving the user a choice to use DHCP or IPv6. The
value should be undefined until it is set by a script or by the user.
Submitted by: Randy Pratt, Chern Lee, many others.
post-configuration "Startup" menu. In the event that diskcheckd is
removed, this can easily be trimmed also; in the mean time, it allows
diskcheckd to be easily disabled using our documented management
tool
the name for the moderate security profile is "moderate", not
"medium", so update this one reference to it as "medium".
This is a 4.4-RELEASE MFC candidate.
MFC after: 2 days
Add a timestamp to the comment so that it's possible to see when
changes were made.
e.g.:
# -- sysinstall generated deltas -- # Wed Aug 15 18:10:20 2001
conservative default, and actually prompt specifically for inetd rather
than handling it as a side effect of the security profile. Update the
help file to reflect this change.
o Rename "Fascist" to "Extreme" in the source code, to match the names
presented to the user.
o Remove portmap and inetd from profile management. Portmap is now
disabled by default, but automatically turned on if a feature requires
it (such as NFS, etc).
This is an MFC candidate for 4.4-RELEASE.
Reviewed by: freebsd-arch@FreeBSD.org
Approved by: re@FreeBSD.org
MFC after: 2 days
post-install config, reduce the potential confusion from the existence
of both configTTYs and configTtys by renaming configTTYs to
configEtcTtys. While this is not a C naming conflict, it was probably
a poor choice of names on my part.
system installation process. This allows users installing via serial
console to enable serial console login during the installation
process using an un-customized install. The user is not prompted to
modify /etc/ttys during a normal install, but is offered the
opportunity during post-install configuration.
- Introduce configTTYs(), which describes the benefits of editing
/etc/ttys, and asks for confirmation before spawning the editor.
- add configTTYs to the post-install configuration, as well as to
the global configuration index.
by providing the opportunity to edit inetd.conf during the system
installation process. The following modifications were made:
(1) Expand the Anonymous FTP description dialog to indicate that inetd
and ftpd must be enabled before it can be used.
(2) Introduce a new configInetd() pair of dialogs, the first describing
inetd, giving a couple of examples of services that require it, and
hinting at potential risk, then asking the user if they wish to
enable it. The second indicates that inetd.conf must be configured
to enabled specific services, and asks if the user would like to
load inetd.conf into the editor to modify it. Add this
configuration action to the index.
There are some further improvements that might be considered:
(1) Provide a more inetd.conf-specific configuration tool that speaks
inetd.conf(5). However, this is made difficult by the "yet another
configuration format" nature of inetd.conf, as well as its use of
commenting to disable services, rather than an in-syntax way to
disable a service without commenting it out. Submissions here
would probably be welcome.
(2) There's some overlap between settings in the somewhat obtuse
Security Profile mechanism and other settings, including the inetd
setting, and NFS server configuration. As features become
individually tunable, they should probably be removed from the
security profile mechanism. Otherwise, somewhat counter-intuitively,
sysinstall (in practice) queries multiple times whether inetd, nfsd,
etc, should be enabled/disabled. A possible future direction might
be to drive profiles not by degree of paranoia, rather, the set
of services desired. Or simply to remove the Security Profile
mechanism and resort to feature-driven configuration.
Reviewed by: imp, chris, jake, nate, -arch, -stable
blown over by the Hurricane and had a house dropped on you by the Tornado.
Now it's time to have your parade rained on by... the Typhoon!
This commit adds driver support for 3Com 3cR990 10/100 ethernet
adapters based on the Typhoon I and Typhoon II chipsets. This is actually
a port of the OpenBSD driver with many hacks by me.
No Virginia, there isn't any support for the hardware crypto yet. However
there is support for TCP/IP checksum offload and VLANs.
Special thanks go to Jason Wright, Aaron Campbell and Theo de Raadt for
squeezing enough info out of 3Com to get this written, and for doing
most of the hard work.
Manual page is included. Compiled as a module and included in GENERIC.
names suggest, they perform methods on Device's. In addition, they
check that the pointer passed to them is valid; if it isn't, they
pretend that the action failed. This fixes some crashes due to NULL
dereferences (e.g., PR 26509).
Approved by: jkh (some time ago)
/usr/share/syscons/keymaps. This should prevent word breakage when new keymaps
have been added.
Prompted by: Matthew D. Fuller <fullermd@over-yonder.net>
and DP83821 gigabit ethernet MAC chips and the NatSemi DP83861 10/100/1000
copper PHY. There are a whole bunch of very low cost cards available with
this chipset selling for $150USD or less. This includes the SMC9462TX,
D-Link DGE-500T, Asante GigaNIX 1000TA and 1000TPC, and a couple cards
from Addtron.
This chip supports TCP/IP checksum offload, VLAN tagging/insertion.
2048-bit multicast filter, jumbograms and has 8K TX and 32K RX FIFOs.
I have not done serious performance testing with this driver. I know
it works, and I want it under CVS control so I can keep tabs on it.
Note that there's no serious mutex stuff in here yet either: I need
to talk more with jhb to figure out the right way to do this. That
said, I don't think there will be any problems.
This driver should also work on the alpha. It's not turned on in
GENERIC.
/usr/src/usr.sbin/sysinstall/menus.c:1323: initializer element is not computable at load time
/usr/src/usr.sbin/sysinstall/menus.c:1323: (near initialization for `MenuNetworking.items[9].aux')
Eliminate an old warning brought about by insufficient foresight when creating
the Menu structure. Have I ever mentioned that sysinstall really needs to
be rewritten?
1. Has a time-stamp to show when it was created
2. Sorts and uniq's the output to only contain single instances of a
given setting. This doesn't mean you still can't have settings which
override one another, that's still possible since it's too much
trouble to do the redundancy checking here.
Requested by: lots of people
a few cosmetic problems:
o Allow it to work with scripts (see man page or install.cfg file).
o Preserve old softupdates flag across newfs toggles
o Clean up partitioned/labelled flag handling
o Don't ask for MBR choice again if you've already written it out.
o Actually document the new features.
and also obey most of the rules of english in their construction.
Add a help screen for the security menu which gives the user a rough idea
just what the various security profiles do.
need to manually force the network_interfaces variable in /etc/rc.conf,
and it only ever gets in the way. rc.network and rc.network6 DTRT with
the default of 'auto'. This should have died over a year ago.
not right because rtermcap would be reading the *host* termcap, not
from the termcap in the src tree. Besides, /usr/sbin/sysinstall
(not the crunched one in /stand) should use the runtime termcap
not the precompiled set.
static version that installs in /stand. Also, don't use an extra
before-install target to create /stand.
- Add missing $FreeBSD$.
- Fix dependencies to handle keymap.h. (*)
Submitted by: obrien (*)
non-advertised option (F = "FreeBSD only"), and leave the A key with
standard partitioning. It seems people still want a runtime backdoo
to get to dangerously dedicated mode.
at people. This has been sitting in my tree for a few months now. I
have spoken with quite a few folks about this and the support for doing
this was pretty strong. I dont remember names though, so I cannot share
the blame :-(. Note that this does not *remove* DD mode, it just stops
waving it at new users. You can still set it via config files etc, and
the bootblocks and kernel still support it. You can still use disklabel
to make true DD disks.
no as a default. Sysinstall should be both less dangerous and less
annoying as a result of this change, though that's just my opinion
(since they're the defaults which annoy ME the least :).
standard or serial. This change needs to be done to the entire system that
depends on this. This way we don't have some code using OnVTY checks
and other doing
strcmp(variable_get(VAR_FIXIT_TTY), "standard") == 0
checks. Also we need to set VAR_FIXIT_TTY to "serial" if we come up on
a serial console.
Also fixed a dialog problem in that dialog was used when dialog was
disabled causing some troubles such as not letting the cursor keys
work when exiting the fixit mode on media (ie. not the fixit shell but
for example fixit on a floppy).
Submitted by: Doug Ambrisko <ambrisko@whistle.com>
PR: 22352
rename the previous one to indicate that it's not just high, it's
extreme (everything off, secure level raised).
Submitted mostly by: Tony Finch <dot@dotat.at>
Replace all in-tree uses with necessary subset of <sys/{fb,kb,cons}io.h>.
This is also the appropriate fix for exo-tree sources.
Put warnings in <machine/console.h> to discourage use.
November 15th 2000 the warnings will be converted to errors.
January 15th 2001 the <machine/console.h> files will be removed.
Approved by: jkh
Write kern_securelevel_enable variable to rc.conf if user selects
medium or low security in sysinstall. This overrides the case where a
user selects fascist security and then tries to go back to a lower
setting.
a default. This should prevent people from whacking return at
the Distributions menu and getting nothing selected as a result
(a minimal "standard" system will at least install).
Flagged as big tech support headache by: Chris Shumway <cshumway@osd.bsdi.com>
support which use National Semiconductor DP8393X (SONIC) as ethernet
controller. Currently, this driver is used on only PC-98.
Submitted by: Motomichi Matsuzaki <mzaki@e-mail.ne.jp>
Obtained from: NetBSD/pc98
appropriate(?) defaults for "low", "medium" and "high" security
environments. Medium is basically what we currently have with a little
seat-belt tightening where it made sense. Low is the same as medium but
without the tightening. High is positively fascist with nothing turned
on by default and an automatic call to 911 if it can find a modem.
Previously, these cards were supported by the lnc driver (and they
still are, but the pcn driver will claim them first), which is fine
except the lnc driver runs them in 16-bit LANCE compatibility mode.
The pcn driver runs these chips in 32-bit mode and uses the RX alignment
feature to achieve zero-copy receive. (Which puts it in the same
class as the xl, fxp and tl chipsets.) This driver is also MI, so it
will work on the x86 and alpha platforms. (The lnc driver is still
needed to support non-PCI cards. At some point, I'll need to newbusify
it so that it too will me MI.)
The Am79c978 HomePNA adapter is also supported.
Make sysinstall override this on install, so the effective behavioural
change for a newly installed system is null. Overall, this makes a system
with an empty /etc/rc.conf not run any network services, and makes the
FreeBSD-provided network services that are running visible in /etc/rc.conf
(instead of making people look through /etc/defaults/rc.conf to find the
things they need to disable to secure the system.)
Reviewed by: jhb
Discussed with: The usual cabal
Now, if a release is specified, instead of just looking for a directory
with the same name as the release, try several possible directories (each
suffixed with the release name) relative to the base directory including
".", "releases/MACHINE", "snapshots/MACHINE", and each of those prefixed
with "pub/FreeBSD/". This will allow us to remove the evil symlinks under
pub/FreeBSD/releases/MACHINE/ to the snapshots on the ftp site.
does bad things to /etc/make.conf in certain situations. Also
soften the "don't install crypto from the USA!" messages since,
except for RSA (which is still noted), that's not so true anymore.
IPv6 configuration is only done by rtsol. Does someone really
need manual configuration? :-)
You can specify IPv6 DNS server as well.
We have only one server ftp7.jp.freebsd.org that speaks IPv6
in this time. ftp7.jp speaks IPv4 as well and also listed as
Japan #7.
Approved by: jkh
boot.flp and plain boot.flp.
- Clean up crunchgen related routine.
- Add PC-98 support.
TODO:
o Documentation
o Fix some messages for PC-98
o Decrease the size of fixit.flp to 1.2MB
o I18N (See: http://www.jp.FreeBSD.org/BootAsia/index.html)
No response from jkh
setting 'usbd_enable' in rc.conf during nwe installs if USB is detected.
Also, since usbd already handles USB mice automatically, note that the
mouse setup section in sysinstall only applies to non-USB mice.
Approved by: jkh
You can't enable 'emulate 3 button' option for moused in sysinstall.
This adds a menu option to set moused_flags and the help text explains
that entering "-3" will enable this feature.
be detected by netscape and such.
PR: bin/17659
Submitted by: Murray Stokelay <murray@cdrom.com>
Approved by: jkh
jkh made updates that conflict with the submitters patch, so I updated
accordingly, any mistakes are mine, not the submitters.
for generating /boot/kernel.conf. Since this structure is shared, move
its definition out to a header file, just as struct isa_device was defined
in a header file. This fixes the sysinstall breakage in -current.
on locale.
o Allow use of "G" in label editor to stand for gigabytes. This
is actually an unrelated patch which I meant to commit separately
but what the heck, it's late.
Partially submitted by: phk
straight into debug mode if you boot -v. Also conditionalize some
annoying debugging output now that we have this ability.
Partially submitted by: msmith
Approved by: jkh [to make certain wise-acres happy ;)]
USB-EL1202A chipset. Between this and the other two drivers, we should
have support for pretty much every USB ethernet adapter on the market.
The only other USB chip that I know of is the SMC USB97C196, and right
now I don't know of any adapters that use it (including the ones made
by SMC :/ ).
Note that the CATC chip supports a nifty feature: read and write combining.
This allows multiple ethernet packets to be transfered in a single USB
bulk in/out transaction. However I'm again having trouble with large
bulk in transfers like I did with the ADMtek chip, which leads me to
believe that our USB stack needs some work before we can really make
use of this feature. When/if things improve, I intend to revisit the
aue and cue drivers. For now, I've lost enough sanity points.
Kawasaki LSI KL5KUSB101B chip, including the LinkSys USB10T, the
Entrega NET-USB-E45, the Peracom USB Ethernet Adapter, the 3Com
3c19250 and the ADS Technologies USB-10BT. This device is 10mbs
half-duplex only, so there's miibus or ifmedia support. This device
also requires firmware to be loaded into it, however KLSI allows
redistribution of the firmware images (I specifically asked about
this; they said it was ok).
Special thanks to Annelise Anderson for getting me in touch with
KLSI (eventually) and thanks to KLSI for providing the necessary
programming info.
Highlights:
- Add driver files to /sys/dev/usb
- update usbdevs and regenerate attendate files
- update usb_quirks.c
- Update HARDWARE.TXT and RELNOTES.TXT for i386 and alpha
- Update LINT, GENERIC and others for i386, alpha and pc98
- Add man page
- Add module
- Update sysinstall and userconfig.c
USB ethernet chip. Adapters that use this chip include the LinkSys
USB100TX. There are a few others, but I'm not certain of their
availability in the U.S. I used an ADMtek eval board for development.
Note that while the ADMtek chip is a 100Mbps device, you can't really
get 100Mbps speeds over USB. Regardless, this driver uses miibus to
allow speed and duplex mode selection as well as autonegotiation.
Building and kldloading the driver as a module is also supported.
Note that in order to make this driver work, I had to make what some
may consider an ugly hack to sys/dev/usb/usbdi.c. The usbd_transfer()
function will use tsleep() for synchronous transfers that don't complete
right away. This is a problem since there are times when we need to
do sync transfers from an interrupt context (i.e. when reading registers
from the MAC via the control endpoint), where tsleep() us a no-no.
My hack allows the driver to have the code poll for transfer completion
subject to the xfer->timeout timeout rather that calling tsleep().
This hack is controlled by a quirk entry and is only enabled for the
ADMtek device.
Now, I'm sure there are a few of you out there ready to jump on me
and suggest some other approach that doesn't involve a busy wait. The
only solution that might work is to handle the interrupts in a kernel
thread, where you may have something resembling a process context that
makes it okay to tsleep(). This is lovely, except we don't have any
mechanism like that now, and I'm not about to implement such a thing
myself since it's beyond the scope of driver development. (Translation:
I'll be damned if I know how to do it.) If FreeBSD ever aquires such
a mechanism, I'll be glad to revisit the driver to take advantage of
it. In the meantime, I settled for what I perceived to be the solution
that involved the least amount of code changes. In general, the hit
is pretty light.
Also note that my only USB test box has a UHCI controller: I haven't
I don't have a machine with an OHCI controller available.
Highlights:
- Updated usb_quirks.* to add UQ_NO_TSLEEP quirk for ADMtek part.
- Updated usbdevs and regenerated generated files
- Updated HARDWARE.TXT and RELNOTES.TXT files
- Updated sysinstall/device.c and userconfig.c
- Updated kernel configs -- device aue0 is commented out by default
- Updated /sys/conf/files
- Added new kld module directory
working. It was, as I predicted, a stupid bug and thanks to the
submitter for spotting it. I'll also re-roll some 3.4-RELEASE install
floppies for this.
as redoing all the menus to have proper, or at least non-hallucinogenic,
keyboard accelerators.
This requires my recent update to libdialog to work properly and will
probably also exhibit some other "interesting" behavior while the last
few missing screen clears are found (which is why I'm not going to MFC
immediately). At least now, however, sysinstall does not gratuitously
redraw random screens at the drop of a hat and drive serial console
installers out of their minds.
which it replaces. The new driver supports all of the chips supported
by the ones it replaces, as well as many DEC/Intel 21143 10/100 cards.
This also completes my quest to convert things to miibus and add
Alpha support.
for the AN985 "Centaur" chip, which is apparently the next genetation
of the "Comet." The AN985 is also a tulip clone and is similar to the
AL981 except that it uses a 99C66 EEPROM and a serial MII interface
(instead of direct access to the PHY registers).
Also updated various documentation to mention the AN985 and created
a loadable module.
I don't think there are any cards that use this chip on the market yet:
the datasheet I got from ADMtek has boxes with big X's in them where the
diagrams should be, and the sample boards I got have chips without any
artwork on them.
the Davicom DM9100 and DM9102 chipsets, including the Jaton Corporation
XPressNet. Datasheet is available from www.davicom8.com.
The DM910x chips are still more tulip clones. The API is reproduced
pretty faithfully, unfortunately the performance is pretty bad. The
transmitter seems to have a lot of problems DMAing multi-fragment
packets. The only way to make it work reliably is to coalesce transmitted
packets into a single contiguous buffer. The Linux driver (written by
Davicom) actually does something similar to this. I can't recomment this
NIC as anything more than a "connectivity solution."
This driver uses newbus and miibus and is supported on both i386
and alpha platforms.
SiS 900 and SiS 7016 PCI fast ethernet chipsets. Full manuals for the
SiS chips can be found at www.sis.com.tw.
This is a fairly simple chipset. The receiver uses a 128-bit multicast
hash table and single perfect entry for the station address. Transmit and
receive DMA and FIFO thresholds are easily tuneable. Documentation is
pretty decent and performance is not bad, even on my crufty 486. This
driver uses newbus and miibus and is supported on both the i386 and
alpha architectures.
I backed-out the changes in -current and didn't touch stable at all (I
thought I had my patch order reversed, not what actually happened).
AIEEE! I can't even blame the crack for this one since I broke my
crack pipe a few weeks ago. I think sleep deprivation gets the blame
for this one.
Medal for noticing this one goes to: Jim Bloom <bloom@acm.org>
bringing in DHCP support. The only thing I left out were Poul-Henning's
newfs changes since I'm not sure if he's brought the rest of that support
into -stable yet. If it turns out that this is the case, I'll MFC those
changes too.
PCI fast ethernet controller. Currently, the only card I know that uses
this chip is the D-Link DFE-550TX. (Don't ask me where to buy these: the
only cards I have are samples sent to me by D-Link.)
This driver is the first to make use of the miibus code once I'm sure
it all works together nicely, I'll start converting the other drivers.
The Sundance chip is a clone of the 3Com 3c90x Etherlink XL design
only with its own register layout. Support is provided for ifmedia,
hardware multicast filtering, bridging and promiscuous mode.
in a previous FreeBSD version. That never happened. Document that
it is due to be replaced, but leave it open-ended as to when.
Also do some mdoc cleanup.
PR: docs/13148
PR: docs/13144
Submitted by: Lee Cremeans <lcremeans@erols.com>
Alex M. Zelkim <phantom@cris.net>
Discussed with: jkh
Originally submitted by: Wayne Self <wself@cdrom.com>
Allow a ppp startup option in rc.conf.
Adjust sysinstall so that it appends to the end of ppp.conf
and uses the generated profile to start ppp in auto mode on
boot.
Submitted by: Josef L. Karthauser <joe@uk.FreeBSD.org>
ethernet controllers based on the AIC-6915 "Starfire" controller chip.
There are single port, dual port and quad port cards, plus one 100baseFX
card. All are 64-bit PCI devices, except one single port model.
The Starfire would be a very nice chip were it not for the fact that
receive buffers have to be longword aligned. This requires buffer
copying in order to achieve proper payload alignment on the alpha.
Payload alignment is enforced on both the alpha and x86 platforms.
The Starfire has several different DMA descriptor formats and transfer
mechanisms. This driver uses frame descriptors for transmission which
can address up to 14 packet fragments, and a single fragment descriptor
for receive. It also uses the producer/consumer model and completion
queues for both transmit and receive. The transmit ring has 128
descriptors and the receive ring has 256.
This driver supports both FreeBSD/i386 and FreeBSD/alpha, and uses newbus
so that it can be compiled as a loadable kernel module. Support for BPF
and hardware multicast filtering is included.
in some code from C. Stone to parse the lease information. This is still
a WIP and this commit is largely intended to allow others to sync up; the
dhclient code still only works when doing dhcp configuration post-install
and requires a bit more work on the boot floppy before it will truly
work in the minimal bootstrapping role.
gigabit ethernet adapters. This includes two single port cards
(single mode and multimode fiber) and two dual port cards (also single
mode and multimode fiber). SysKonnect is currently the only
vendor with a dual port gigabit ethernet NIC.
The ports on dual port adapters are treated as separate network
interfaces. Thus, if you have an SK-9844 dual port SX card, you
should have both sk0 and sk1 interfaces attached. Dual port cards
are implemented using two XMAC II chips connected to a single
SysKonnect GEnesis controller. Hence, dual port cards are really
one PCI device, as opposed to two separate PCI devices connected
through a PCI to PCI bridge. Note that SysKonnect's drivers use
the two ports for failover purposes rather that as two separate
interfaces, plus they don't support jumbo frames. This applies to
their Linux driver too. :)
Support is provided for hardware multicast filtering, BPF and
jumbo frames. The SysKonnect cards support TCP checksum offload
however this feature is not currently enabled (hopefully it will
be once we get checksum offload support).
There are still a few things that need to be implemeted, like
the ability to communicate with the on-board LM80 voltage/temperature
monitor, but I wanted to get the driver under CVS control and into
-current so people could bang on it.
A big thanks for SysKonnect for making all their programming info
for these cards (and for their FDDI and token ring cards) available
without NDA (see www.syskonnect.com).
ADMtek AL981 "Comet" chipset. The AL981 is yet another DEC tulip clone,
except with simpler receive filter options. The AL981 has a built-in
transceiver, power management support, wake on LAN and flow control.
This chip performs extremely well; it's on par with the ASIX chipset
in terms of speed, which is pretty good (it can do 11.5MB/sec with TCP
easily).
I would have committed this driver sooner, except I ran into one problem
with the AL981 that required a workaround. When the chip is transmitting
at full speed, it will sometimes wedge if you queue a series of packets
that wrap from the end of the transmit descriptor list back to the
beginning. I can't explain why this happens, and none of the other tulip
clones behave this way. The workaround this is to just watch for the end
of the transmit ring and make sure that al_start() breaks out of its
packet queuing loop and waiting until the current batch of transmissions
completes before wrapping back to the start of the ring. Fortunately, this
does not significantly impact transmit performance.
This is one of those things that takes weeks of analysis just to come
up with two or three lines of code changes.
