As of this patchset, the loader builds (under NetBSD/macppc), boots, interacts
and talks to BOOTP/NFS servers.
(main.c was moved from boot/ofw/libofw to boot/ofw/common but has no revision
history)
Reviewed by: obrien
This brings the loader up to the point where I can compile it under
NetBSD/macppc and have it boot, interact and talk to NFS servers.
sys/boot/ofw/libofw/main.c has been deleted (it has no revision history) and
replaced with sys/boot/ofw/common/main.c
Reviewed by: obrien
to reinstall boot1 after a 'make world'.
Unfortunately this means that people who have already installed a new
boot1 from a 'make world' after 2000/09/18 *must* reinstall it after
their next build using something like:
# disklabel -B /dev/da0c
expands beyond the limit we will extend the address space before trying
to zero the BSS. This should give us plenty of headroom for modest
expansion of the loader.
Previous revision of this file changed the "boot" commands to take
no arguments from the stack. This is only valid in the case where
a kernel has not been loaded. In that case, load_kernel_and_modules
will be called, which takes a list of arguments from the stack.
When a kernel is presently loaded, though, the list of arguments must
be passed to the boot command, which was the behaviour before the last
revision.
Fix things for both cases.
Noticed by: S-Max and others on that chat room
Taking over the sector following the MBR causes problems on some
machines, and the actual gains are fairly small in terms of how
the space is presently used.
Since we need a number of further features (eg. handling extended
partitions) that can't be readily accommodated in the basic boot0
design anyway, rather choose to implement the additional stuff
separately and concentrate on compatibility rather than features
here.
used by start to find the kernel. Fix this.
Also, boot would proceed immediately in the absence of a path as
argument. Check first if a kernel has already been loaded, and, if
not, fall back to load kernel&modules behavior.
Some further factorizing. I deem this code to be mostly readable by
now! :-)
Many thanks to: Makoto MATSUSHITA <matusita@jp.FreeBSD.org>
The boot-conf and boot code had various bugs, and some of it was big,
ugly, unwieldy, and, sometimes, plain incorrect. I'm just about
completely replaced these ugly parts with something much more manageable.
Minor changes were made to the well-factorized parts of it, to accomodate
the new code.
Of note:
* make sure boot-conf has the exact same behavior wrt boot order
as start.
* Correct both boot and boot-conf so they'll work correctly when
compiled in, as they both had some bugs, minor and major.
* Remove all the crud from loader.4th back into support.4th, for
the first time since boot-conf was first improved. Hurray!
I'm fairly satisfied with the code at this time. Time to see about those
man pages...
to make things more interchangeable between it and the FORTH case.
Perhaps requiring the space is a bit too much, but...
Nothing in the tree seems to produce loader.rc files with comment
line, at this time.
as the kernel name. The one very unfortunate consequence is that kernel
as an absolute path loses the priority. It will only be tried after
/boot/${kernel}/${bootfile}. I'll see what can be done about it later.
Load the first of the following kernels to be found:
${kernel} if ${kernel} is an absolute path
/boot/${kernel}/${kernel}
/boot/${kernel}/${bootfile}
${kernel}/${kernel}
${kernel}/${bootfile}
${kernel}
${bootfile}
The last instance of ${kernel} and ${bootfile} will be treated as a
list of semicolon separated file names, and each will be tried in turn,
from left to right.
Also, for each filename loader(8) will try filename, filename.ko,
filename.gz, filename.ko.gz, in that order, but that's not related
to this code.
This resulted in a major reorganization of the code, and much of what
was accumulating on loader.4th was rightly transfered to support.4th.
The semantics of boot-conf and boot also changed. Both will try to load
a kernel the same as above.
After a kernel was loaded, the variable module_path may get changed. Such
change will happen if the kernel was found with a directory prefix. In
that case, the module path will be set to ${directory};${module_path}.
Next, the modules are loaded as usual.
This is intended so kernel="xyzzy" in /boot/loader.conf will load
/boot/xyzzy/kernel.ko, load system modules from /boot/xyzzy/, and
load third party modules from /boot/modules or /modules. If that doesn't
work, it's a bug.
Also, fix a breakage of "boot" which was recently introduced. Boot without
any arguments would fail. No longer. Also, boot will only unload/reload
if the first argument is a path. If no argument exists or the first
argument is a flag, boot will use whatever is already loaded. I hope this
is POLA. That behavior is markedly different from that of boot-conf, which
will always unload/reload.
The semantics introduced here are experimental. Even if the code works,
we might decide this is not the prefered behavior. If you feel so, send
your feedback. (Yeah, this belongs in a HEADS UP or something, but I've
been working for the past 16 hours on this stuff, so gimme a break.)
Now boot-conf can also receive parameters to be passed to the kernel
being booted. The syntax is the same as in the boot command, so one
boots /kernel.OLD in single-user mode by typing:
boot-conf /kernel.OLD -s instead of
boot-conf -s /kernel.OLD
The syntax still supports use of directory instead of file name, so
boot-conf kernel.OLD -s
may be used to boot /boot/kernel.OLD/kernel.ko in single-user mode.
Notice that if one passes a flag to boot-conf, it will override the
flags set in .conf files, but only for that invocation. If the user
aborts the countdown and tries again without passing any flags, the
flags set in .conf files will be used.
Some factorization was done in the process of enhancing boot-conf,
as it has been growing steadly as features are getting added, becoming
too big for a Forth word. It still could do with more factorization,
as a matter of fact.
Override the builtin "boot" with something based on boot-conf. It will
behave exactly like boot-conf, but booting directly instead of going
through autoboot.
Since we are now pairing kernel and module set in the same directory,
this change to boot makes sense.
load the modules needed according to a file relating module names
(actually, _file_ names, not really modules -- the dependency
stuff is not exported to loader's UI) to PnP IDs.
But it still lacks a number of desired features, and it's too crude
for my tastes. But since I don't have time to work on it, it might
be preferable to make it available to those who might. It's not
installed by default, much less loaded. In fact, it wouldn't even
had a copyright message (who? me? assume responsibility for _this_?),
if the cvs commit hadn't aborted for lack of $FreeBSD$, and I decided
to just cut&paste the stuff from elsewhere.
Also, export the file_findfile() function. Again, this is taken from
work in progress but frozen for the time being. Since it works, I'd
rather commit and remove any uglyness later than hide it on my tree.
code into a more modular interface, with hidden vocabularies and
such. Remove the need to a lot of ugly initialization.
Also, add a few structure definitions, from stuff used on the C
part of loader. Some of this will disappear, and the crude structure
support will most likely be replaced by full-blown OOP support
already present on FICL, but not installed by default. But it was
getting increasingly inconvenient to keep this separate on my tree,
and I already lost lots of work once because of the hurdles, so
commit this.
Anyway, it makes support.4th more structured, and I'm not proceeding
with the work on it any time soon, unfortunately.
gets the name from the environment variable kernelname, which is set
when a kernel is loaded. For this reason, autoboot will _first_ try
to load a kernel, and only proceed with the wait prompt after that
succeeds. If it fails, it will abort immediately.
While I understand some may think this behavior undesirable, I think
it is, overall, the best thing to do, even if we do not consider the
aesthetic issue. Notice that anyone using the default loader.rc
already has the kernel loaded before autoboot.
On unload, unset kernelname.
Separate the code that tries to load a kernel from the list of options
to the function loadakernel(). It is used by both boot() and
autoboot().
to allow commonality between varying platforms. This is a step
towards parsing the diskless configuration information with MI code
inside the kernel.
Export the interface hardware address to the kernel, so that it is possible
to determine the boot interface with certainty.
Export the NFS filehandle for the root mount to the kernel, so that the
kernel does not need to perform a mount RPC call.
drives. Some IDE cards don't set propler information into BIOS work
are and their sector size were always recognized as 256 bytes/sector.
Pointed out by: jagarl@creator.club.ne.jp
BOOT_BTX_NOHANG, then BTX will be compiled with the appropriate flags so
that it reboots after a fault instead of hanging forever.
Requested by: ps
Approved by: rnordier
identifier to the DHCP server. Now you can check for this string
in your dhcp configuration to decide whether you will hand out a
lease to the client or not.
to 0x600 via a 'rep movsw'. Once that was done, %cx was zero, so we could
simply use 'movb' to update the lower byte of %cx in preparation for
zeroing out the fake partition entry used to boot to other drives via F5.
Well, in the new boot0, we don't actually relocate ourselves, instead it
is easier to create the fake partition entry first and then just use it to
get the BIOS to load all of boot0 into memory at 0x600. However, since we
aren't doing the relocate code anymore, we don't know that %cx == 0 when
we hit the 'movb' to setup %cx for clearning the fake partition entry.
Thus, if %ch != 0 when the BIOS started boot0, then it would end up zeroing
a lot more memory than just 8 words. The solution is to do a word move of
$8 into %cx.
Debugging help from: David Wolfskill <dhw@whistle.com>
<stand.h>. Also, since bcache_strategy() used to not have a prototype,
arcdisk happily called bcache_strategy() with 6 parameters instead of 7,
leaving out the disk unit number, which is the 2nd parameter. Add in the
unit number to the bcache_strategy() call to fix this.
sector 0 of a disk and boot0.5 is the `boot selector' which starts
from address 0x400. The IPL loads boot0.5 and boot0.5 loads bootblock
of a slice.
The boot manager stuff was developed by me (kato) with Borland C++,
and then, translated into bcc in the ports collection by Nokubi-san.
After that, boot0 has been translated into gas with the .code16
directive by Takahashi-san (nyan) and boot0.5 has been rewritten in
gas by me.
Linux kernel image, and is designed to be dropped into a Linux system
and booted via LILO. Once booted, the user is greeted by the FreeBSD
loader. This still isn't quite complete, as the the root= specification
from LILO isn't currently passed to the loader yet.
- Autodetection and support of the BIOS EDD extensions to work around the
1024 cylinder limit on all but really ancient BIOS's.
- To work around some BIOS's which break when EDD is used with older drives,
we only attempt to use EDD if the cylinder is > 1023.
- Since this new code required more space than we had left, expand boot0 to
2 sectors (1024 bytes) in length.
- Add support for boot0 being multiple sectors using predefined constants.
If boot0 needs to be extended in the future, all that is required is
bumping the NUM_SECTORS constant.
- Now that we have more room to work with, add a few more fs type
descriptions while making others more verbose.
only doing so if loader.rc does not exist. This fixes the problem where
installworld doesn't update /boot/loader.4th, resulting in device.hints not
being loaded after updating past the config(8) changes, which resulted in
mcclock0 not being probed, and a nice kernel panic during boot.
use the BIOS Equipment List to determine how many hard drives are
installed and if the drive number we received in %dl is valid.
- Don't bother to disable interrupts when setting up the stack. The 8086
and beyond implicitly disable interrupts after an instruction that sets
%ss (for example, a pop or a mov) so that you can safely set %ss and %sp
in two consecutive instructions. An exception to this is the lss
instruction, which can set both registers simultaneously and thus doesn't
need this hack.
- Add support for EDD BIOS extensions to support booting off of hard drives
of nearly arbitrary length.
- Add in support for the EDD (Enhanced Disk Drive) BIOS extensions to
use LBA mode for accessing drives past cylinder 1024. This should allow
us to load a kernel from anywhere on a newer drive up to 2 TB. Part
of this came from the PR below.
PR: i386/13847
Submitted by: Tor Egge <Tor.Egge@fast.no>
theory, this should allow the K7V Athlon motherboard to boot ok with boot
virus protection enabled. However, I have no hardware to test this. It
shouldn't break anything though. :)
Prodded by: Kelly Yancey <kbyanc@posi.net>
no clue.
Set sourceid to 0 when booting, which is the correct setting for stdin.
Set sourceid to an arbitrary fd when include'ing, preserving and restoring
the previous sourceid. This is possibly broken(), as 0 is a valid fd. Maybe
we should +1 to this value.
This fixes the version problem widely reported.
is failing for everybody that I have spoken with that has tried it.
FreeBSD/i386 bootstrap loader, Revision 0.8
(root@outback.netplex.com.au, Tue Jun 13 23:26:49 PDT 2000)
Loader version 0.3+ required
Aborted!
start not found
Note that the 0.3+ message is from inside the arch-alpha block, not the
i386 block of code. And even then, 0.8 is higher than 0.3.
This prevents the rest of the loader.conf stuff working. :-/
Use Warner Losh's "hint" driver to decode ascii strings to fill the
resource table at boot time.
config(8) no longer generates an ioconf.c table - ie: the configuration
no longer has to be compiled into the kernel. You can reconfigure your
isa devices with the likes of this at loader(8) time:
set hint.ed.0.port=0x320
userconfig will be rewritten to use this style interface one day and will
move to /boot/userconfig.4th or something like that.
It is still possible to statically compile in a set of hints into a kernel
if you do not wish to use loader(8). See the "hints" directive in GENERIC
as an example.
All device wiring has been moved out of config(8). There is a set of
helper scripts (see i386/conf/gethints.pl, and the same for alpha and pc98)
that extract the 'at isa? port foo irq bar' from the old files and produces
a hints file. If you install this file as /boot/device.hints (and update
/boot/defaults/loader.conf - You can do a build/install in sys/boot) then
loader will load it automatically for you. You can also compile in the
hints directly with: hints "device.hints" as well.
There are a few things that I'm not too happy with yet. Under this scheme,
things like LINT would no longer be useful as "documentation" of settings.
I have renamed this file to 'NOTES' and stored the example hints strings
in it. However... this is not something that config(8) understands, so
there is a script that extracts the build-specific data from the
documentation file (NOTES) to produce a LINT that can be config'ed and
built. A stack of man4 pages will need updating. :-/
Also, since there is no longer a difference between 'device' and
'pseudo-device' I collapsed the two together, and the resulting 'device'
takes a 'number of units' for devices that still have it statically
allocated. eg: 'device fe 4' will compile the fe driver with NFE set
to 4. You can then set hints for 4 units (0 - 3). Also note that
'device fe0' will be interpreted as "zero units of 'fe'" which would be
bad, so there is a config warning for this. This is only needed for
old drivers that still have static limits on numbers of units.
All the statically limited drivers that I could find were marked.
Please exercise EXTREME CAUTION when transitioning!
Moral support by: phk, msmith, dfr, asmodai, imp, and others
was found or not. Fix it's usage. Alas, it caused no problem before,
besides leaving garbage in the stack, because refill, used by [if]
[else] [then], was broken.
a parameter and dtrt.
Also, make boot-conf always unload first. There wasn't really any
point in not doing this, as the kernel _has_ to be loaded before
any other modules.
Tested by: dwhite
with the new binutils. Now that we have a decent assembler, all the old
m4 macros are no longer needed. Instead, straight assembly can be used
since as(1) now understands 16-bit addressing, branches, etc. Also,
several bugs have been fixed in as(1), allowing boot0.s to be further
cleaned up.
be booted. Due to a bug, this wasn't happening.
There is still a lesser bug in that the loader decides which file to boot
after the 10sec count down. This means the bootfile listed in the count
down in is wrong in the case where the loader will boot /kernel.old.
FICL. bootforth is now live on the Alpha!
**BEWARE** - you *MUST* build and install a current libstand or you will
most likely get zfree() panics at loader startup.
We should now be able to set up the loader.conf stuff on the Alpha too.
/boot/loader (even though it is 100% dormant in the Alpha version),
then the loader panics with a zfree error:Loading /boot/loader.test
*** keyboard not plugged in...
Console: SRM firmware console
panic: zfree(0x2003cb58,4096): wild pointer
versus the exact same code but without FICL linked in:
Loading /boot/loader
Console: SRM firmware console
VMS PAL rev: 0x1000600010114
OSF PAL rev: 0x1000600020116
Switch to OSF PAL code succeeded.
FreeBSD/alpha SRM disk boot, Revision 0.1
This is almost certainly an alpha infrastructure bug, not a FICL
problem. It's probably the same thing that made FICL fail for no
apparent reason on the Alpha.
code instead of using 32-bit code and having to just "know" that it's
really 16-bit instructions when things run. This also allows the code
to use fewer macros and more actual assembly statements, which eases
maintenance. Unfortunately, due to as(1) brokenness, we still use m4
macros for all 16-bit addresses, and all short jumps (i.e., 8-bit
relative addresses in the jump instruction) must be wrapped in .code32
directives to avoid useless bloat by as(1). This also fixes a few
problems that were preventing boot0 from compiling with the latest
and greatest version of as(1).
below). This did not work previously because interrupts were
disabled when PXE calls were being made, and they must be enabled.
This should also allow us to be compliant with all newer PXE rom's
from Intel.
For PXE 0.99, this has been tested using the Intel N440BX motherboard
and I am confident it will work on the Intel L440GX motherboard.
Lots of help/information from: jhb, peter
I would like to thank Michael Johnston <michael.johnston@intel.com>,
Mike Henry <mike.henry@intel.com>, and all the other PXE developers
at Intel for their help, and information in helping solve this
problem.
from user mode. Don't disable interrupts when returning from vm86 mode
to user mode either. Now, we only disable interrupts before calling a
hardware interrupt handler, which is the only time we _should_ be
disabling interrupts.
Because of this, err, feature, any routine that one called in vm86 mode
had to re-enable interrupts by setting the interrupt flag or interrupts
would remain disabled even after the routine returned. For example, I
have a simple debugging routine that uses a vm86 mode function to dump
any arbitrary memory word that I use to read the BIOS timer or any other
memory location. This function does 1 load instruction from memory and
then returns. Since it didn't re-enable interrupts, the first time I
called it to read the BIOS timer, it disabled interrupts. This also
affected the PXE bootstrap as it needs interrupts enabled while it is
processing. This patch fixes both of those situations so that those
functions do not worry about having to enable interrupts. Hardware
interrupt handlers worked fine with the old code because they always
enable interrupts as part of their routine.
If you have any problems with the loader after this commit, please
let me know. I'd like to MFC it in a week or two since PXE support
needs it.
Noticed by: ps, Michael Johnston <michael.johnston@intel.com>
- Add support for using the PCI BIOS functions for configuration space
accesses, and make this the default.
- Make PNPBIOS the default (obsoletes the PNPBIOS config option).
- Add two new boot-time tunables to disable each of the above.
You may specify TFTP or NFS via compile time options in the loader,
but not both at this time.
Also, remove a warning about not knowing how to boot from network
devices. We can obviously do that now.
- Don't hard code 0x10000 as the entry point for the loader. Instead add
src/sys/boot/i386/Makefile.inc which defines a make variable with the
entry point for the loader. Move the loader's entry point up to
0x20000, which makes PXE happy.
- Don't try to use cpp to parse btxldr for the optional BTXLDR_VERBOSE,
instead use m4 to achieve this. Also, add a BTXLDR_VERBOSE knob in the
btxldr Makefile to turn this option on.
- Redo parts of cdldr's Makefile so that it now builds and installs cdboot
instead of having i386/loader/Makefile do that. Also, add in some more
variables to make the pxeldr Makefile almost identical and thus to ease
maintainability.
- Teach cdldr about the a.out format. Cdldr now parsers the a.out header
of the loader binary and relocates it based on that. The entry point of
the loader no longer has to be hardcoded into cdldr. Also, the boot
info table from mkisofs is no longer required to get a useful cdboot.
- Update the lsdev function for BIOS disks to parse other file systems
(such as DOS FAT) that we currently support. This is still buggy as
it assumes that a floppy with a DOS boot sector actually has a MBR and
parses it as such. I'll be fixing this in the future.
- The biggie: Add in support for booting off of PXE-enabled network
adapters. Currently, we use the TFTP API provided by the PXE BIOS.
Eventually we will switch to using the low-level NIC driver thus
allowing both TFTP and NFS to be used, but for now it's just TFTP.
Submitted by: ps, alfred
Testing by: Benno Rice <benno@netizen.com.au>
necessary. Pass an absolute block number too, instead of receiving a
relative one in realstrategy(), as bcache_strategy() requires this.
The fix is sligthly different from the one in the PR.
PR: 17098
Submitted by: John Hood <jhood@sitaranetworks.com>
was the last unit number received. If it changes, it flushes the cache.
Add bcache_flash().
The actual fix is sligthly different from the one in the PR.
PR: 17098
Submitted by: John Hood <jhood@sitaranetworks.com>
flushed if the unit changes. Compute the absolute offset before
bcache_strategy() instead of after.
The actual fix is sligthly different for the one in the PR.
PR: 17098
Submitted by: John Hood <jhood@sitaranetworks.com>
1) Fix a bug in the int15 function 87 emulation where we only copied half
of what the BIOS asked for. This caused the Mylex RAID adapter to go
haywire and start trashing memory when you tried to boot from it.
2) Don't use interrupt 19 to reboot. Instead, set the reboot flag to a warm
boot and jump to the BIOS's reboot handler. int 19 doesn't clear memory
or restore the interrupt vector table, and thus really isn't safe. For
example, when booting off of PXE, the PXE BIOS eats up a chunk of memory
for its internal data and structures. Since we rebooted via int 19,
using the 'reboot' command in the loader resulted in that memory not
being reclaimed by the BIOS. Thus, after a few PXE boots, the system
was out of lower memory.
3) Catch any int 19 calls made by a BTX client or a user pressing
Ctrl-Alt-Delete and shutdown BTX and reboot the machine cleanly. This
fixes Ctrl-Alt-Delete in the loader and in boot2 instead of presenting
the user with a BTX fault.
Approved by: jkh
Found by: 1) by msmith
- Fix btxldr to preserve a NULL bootinfo pointer when it copies the kernel
arguments.
- Add the cdldr bootstrap program. This program is tacked onto the
beginning of the standard 3rd stage boot loader (/boot/loader) to form
the CD boot loader (/boot/cdboot). When a CD is booted, the cdboot file
is copied into memory instead and executed. The cdldr stub emulates the
environment normally provided by boot2 and then starts the loader. This
booting method does not emulate a floppy drive, but boots directly off of
the CD. This should fix the problems some BIOS's have with emulating a
2.88 MB floppy image.
- Add support to the loader to recognize that it has been booted by cdldr
instead of boot2 and use a simpler method of extracting the BIOS boot
device.
of %cr0 wasn't reloaded into %eax before being modified to turn protected
mode off if PAGING was not defined. The result was that the processor did
not exit protected mode, so when it tried to jump to segment 0x0 in the
next instruction to clear the prefetch cache like one should when leaving
protected mode, it actually tried to jump to a null selector, causing a
GPF.
NICs. (Finally!) The PCMCIA, ISA and PCI varieties are all supported,
though only the ISA and PCI ones will work on the alpha for now.
PCCARD, ISA and PCI attachments are all provided. Also provided an
ancontrol(8) utility for configuring the NIC, man pages, and updated
pccard.conf.sample. ISA cards are supported in both ISA PnP and hard-wired
mode, although you must configure the kernel explicitly to support the
hardwired mode since you have to know the I/O address and port ahead
of time.
Special thanks to Doug Ambrisko for doing the initial newbus hackery
and getting it to work in infrastructure mode.
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
for our use. Use the same search order for BIOS memory size functions
as the kernel will later use.
Allow the loader to use all of the detected physical memory (this will
greatly help people trying to load enormous memory disk images).
More correctly handle running out of memory when loading an object.
Use the end of base memory for the top of the heap, rather than
blindly hoping that there is 384k left.
Add copyrights to a couple of files I forgot.
is an application space macro and the applications are supposed to be free
to use it as they please (but cannot). This is consistant with the other
BSD's who made this change quite some time ago. More commits to come.
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
