Commit graph

195 commits

Author SHA1 Message Date
Ali Saidi
3ef950abba Fixes for bigendian platforms
arch/alpha/vtophys.cc:
    PGOFSET -> ALPHA_PGOFSET to avoid include file problems
base/callback.hh:
    Added  a class to create a callback from a function
base/intmath.hh:
    make FloorLog2 inlined
dev/pcidev.cc:
    more work in getting pciconfig space happy with different endiannesses
dev/uart.cc:
    used an incorrect size for write uint64_t instead of uint8_t
sim/system.cc:
    when writing things into system data structures we need to pay
    attention to endianness

--HG--
extra : convert_revision : 52f441b5789c45db30ef2f6fd4975cbc7323a381
2004-10-16 19:10:51 -05:00
Nathan Binkert
1d02345a24 a bit more cleaning of the network header wrappers.
base/inet.hh:
    add functions to the various headers to grab the most common
    encapsulated protocols.  This could easily get out of hand, but
    we're just worrying about tcp, udp, and ip for now.

    add common functions size(), bytes(), and payload() to all wrappers.
    size() gets the header size
    bytes() returns a uint8_t * to the beginning of the header
    payload() returns a uint8_t * to the beginning of the payload.
dev/etherpkt.cc:
dev/etherpkt.hh:
    don't cache pointers to headers.  It's probably not worth the
    hassle.

--HG--
extra : convert_revision : ba9df85ac019b8a48233042dde79fb9da9546410
2004-09-21 01:41:55 -04:00
Nathan Binkert
a58b834c8e Clean up network header stuff and make it more generic. Use
libdnet when we can instead of our own home grown stuff.

SConscript:
    separate the crc code into its own file
base/inet.cc:
    move the crc stuff to crc.cc
    add generic code for calculating ip/tcp/udp checksums
base/inet.hh:
    -  move crc stuff to crc.hh
    -  #include all of the libdnet stuff.  (this makes base/inet.hh the
    only file you need to include if you want to use this kind of stuff.)
    -  Wrap some of the libdnet structs to get easier access to structure
    members.  These wrappers will automatically deal with masking/shifting/
    byte-swapping.
base/refcnt.hh:
    If one derives from RefCountingPtr, they should have access to
    the internal data pointer.
build/SConstruct:
    make #include of dnet stuff work
dev/etherlink.cc:
dev/ethertap.cc:
dev/ethertap.hh:
    EtherPacket -> PacketData
dev/etherpkt.cc:
    EtherPacket -> PacketData
    add a function for populating extra info about a packet.
    Basically just gives pointers to ethernet/ip/tcp/udp headers
    if they exist.
dev/etherpkt.hh:
    EtherPacket -> PacketData
    remove most of the packet header stuff from teh PacketData
    class and just add a few generic functions for grabbing various
    headers that may exist in the packet.  The old functionality is
    contained in the headers.
dev/ns_gige.cc:
    -  IP -> Ip, UDP -> Udp, TCP ->Tcp when used in variable names
    -  get rid of our own byte swapping functions.
    -  whack checksum code and use libdnet version.
    -  Get pointers to the various packet headers and grab info from
    those headers.  (The byte swapping in the headers now.)
    -  Add stats for Udp Checksums
dev/ns_gige.hh:
    use libdnet for checksum code.
    IP -> Ip, TCP -> Tcp in variable names
    add stats for UDP checksums

--HG--
extra : convert_revision : 96c4160e1967b7c0090acd456df4a76e1f3aab53
2004-09-20 10:43:53 -04:00
Ali Saidi
23e0643c69 Updates to make SMP work.
dev/tsunami_cchip.cc:
    updates to ipi handling chipset code
sim/system.cc:
    debugSymbolTable, now has symbols from pal, console, and linux

--HG--
extra : convert_revision : c981d857f7e3d75f4c46172809e6d14e5f0a1238
2004-08-31 23:47:57 -04:00
Nathan Binkert
0b0a6778c9 Fixup interrupting of the ethernet device.
dev/ns_gige.cc:
    clean up the interrupt handling code so that it is simpler and
    less prone to bugs.  I hope I removed the bug where the:
    assert(intrTick >= curTick || intrTick == 0); would get triggered.
    I'm pretty sure that was due to intrTick not being cleared when
    the interrupt was cleared.

    This code probably still needs to be looked at more closely to
    make sure interrupts are not missed.

--HG--
extra : convert_revision : 61e2eb043540f2534a80c9b633006a71e7d6e282
2004-08-20 11:40:57 -04:00
Nathan Binkert
40174f119d cleanup interrupt code for gigabit device
dev/ns_gige.cc:
    Clean up the interrupt code.  Do a better job determining when
    coalescing should happen.
    Remove some bogus comments
    Stop using magic numbers in initialization and comment what
    the various numbers do
dev/ns_gige_reg.h:
    #define describing which interrupts cannot be delayed and
    which interrrupts we don't implement

--HG--
extra : convert_revision : eb196afa0bf448e1849dd2bd30dde32134effee7
2004-08-11 10:14:26 -04:00
Nathan Binkert
256afbbe26 std::
--HG--
extra : convert_revision : 8383048265ef5b4d7c995dbf44f7d85e349ddbde
2004-08-11 09:05:13 -04:00
Nathan Binkert
a3a1b4dbae better debugging
dev/ns_gige.cc:
    Make all DPRINTF statements take one line. If they need two lines,
    break them up into separate statements.  This makes grep much more
    effective since *every* line will be prefixed by the cycle that the
    trace message is from and the object that caused the message.
    normalize some debugging statements so that searching is easier
    (e.g. always say rxState, not rx state or receive state)
    break into the debugger when a packet is dropped since we don't really
    like dropping packets.

--HG--
extra : convert_revision : 9cf309ca2143a6b8c8215ac5dd6a31aae60173cd
2004-07-30 11:33:45 -04:00
Nathan Binkert
e60a4c58ee Fix a few bugs in the receive state machine. In doing back to tracking
whether or not the state machine is enabled rather than tracking the
specific instance of trying to halt the state machine.

dev/ns_gige.cc:
    change back to tracking the state machine's enableness instead of
    whether or not it is trying to halt.  Also fix a major bug that
    would cause the NIC to drop packets when the rx state machine was
    idle, but enabled.
    Fix a couple other bugs in the state machine where the idle interrupt
    would happen at the wrong time.
    Add a warning to deal with improper values of intrTick
dev/ns_gige.hh:
    We need to keep track of whether the state machine is enabled
    or not separately from the control register since the bits don't
    always reflect the truth.

--HG--
extra : convert_revision : 20056b225fa62a0744473babfd693506aa5f29b2
2004-07-30 11:29:45 -04:00
Nathan Binkert
8f87a57e70 make all ethernet dma use the new interface
--HG--
extra : convert_revision : 1cab883fac2bffd83cb035ebf28958c15118dfdc
2004-07-30 11:17:20 -04:00
Nathan Binkert
f1f85c5470 better debugging of DMA operations
dev/ns_gige.cc:
    use the new PhysicalMemory dma_read and dma_write functions

--HG--
extra : convert_revision : 427049d43355f02ac8bdfe2e60c24825dd734006
2004-07-30 11:09:03 -04:00
Nathan Binkert
da2566f014 style
--HG--
extra : convert_revision : 7489726b6c7bd11af603a448e4ff56c1e46139fe
2004-07-26 23:10:20 -04:00
Nathan Binkert
0650e4a43f When a packet has finished transmitting, kick the state
machine too.  The fifo may have been full and we want
to get another packet into it if we can.

--HG--
extra : convert_revision : 8f57294bd18f5eccb900118f2de83a4ccfd20dbb
2004-07-23 12:19:27 -04:00
Nathan Binkert
8844eadcdf notify the transmitter when the packet has finished transmitting.
--HG--
extra : convert_revision : 66f6f80945c1f621cdc3e743f601d2a73361bacd
2004-07-23 00:05:29 -04:00
Nathan Binkert
c9e6a15196 Add support to store less than the full packet in an etherdump
and actually default to only storing a max of 96 bytes per
packet since that should be plenty to fit all of the headers in.

--HG--
extra : convert_revision : 0c4a6571d80536477ed166e695d957e39da0334e
2004-07-22 23:54:24 -04:00
Nathan Binkert
5b1da825d8 Wrap tracing stuff in #if TRACING_ON to avoid compile error
in m5.fast

--HG--
extra : convert_revision : 329d810bfe282e88133ca35da62ee6dcc73a38f5
2004-07-21 15:44:57 -04:00
Erik Hallnor
a6939573ce Fix serialization when a tx interrupt isn't scheduled
dev/uart.cc:
    Fix serialization when a tx interrupt isn't scheduled.

--HG--
extra : convert_revision : 3029e284cbba417e3f453d83fdf2ef307d684916
2004-07-14 21:34:51 -04:00
Nathan Binkert
6f43d2b8d8 Fix infinite loop in ns_gige stat machine
argh!

dev/ns_gige.cc:
    Exit the state machine so that we don't come right back where
    we started and enter an infinite loop.

--HG--
extra : convert_revision : a5f2b5b5a692de6c80e4b02d7f9bc5d27fe17252
2004-07-14 09:02:15 -04:00
Nathan Binkert
13f8dc981f make the cache access latency a parameter that is based on bus
ticks for the most commonly accessed devices.

dev/baddev.cc:
    Get rid of the constant cache access latency.
    For unimportant devices, don't add any latency.
dev/ide_ctrl.cc:
dev/ide_ctrl.hh:
dev/ns_gige.cc:
dev/pciconfigall.cc:
dev/pciconfigall.hh:
dev/tsunami_cchip.cc:
dev/tsunami_cchip.hh:
dev/tsunami_io.cc:
dev/tsunami_io.hh:
dev/tsunami_pchip.cc:
dev/tsunami_pchip.hh:
dev/uart.cc:
dev/uart.hh:
    make the cache access latency a parameter that is based on bus
    ticks.
dev/io_device.cc:
dev/io_device.hh:
    add an io latency variable
dev/ns_gige.hh:
    this moved to io_device.hh

--HG--
extra : convert_revision : 4883130feeaef48abee492eddf0b8eb40eb94789
2004-07-12 22:58:22 -04:00
Lisa Hsu
f857bd9dff make tx/rx fifo size a param, also fix the empty dma problem by adjusting the state machine.
dev/ns_gige.cc:
    make tx/rx fifo sizes a param. the default is 128K for each.  also, make the state machine not move onto txFragRead if there is no room in the txfifo for data.
dev/ns_gige.hh:
    make tx/rx fifo size a param

--HG--
extra : convert_revision : ed91eb31c2b21d4cdc6de87d8641df6197be5209
2004-07-12 16:09:52 -04:00
Andrew Schultz
4f4252fa89 Remove second pciToDma translation on current PRD address pointer b/c
second translation (which is superfluous) doesn't work properly when
system memory is higher than 128MB

--HG--
extra : convert_revision : 9cdf6925689d376953b1aa071bcd1e2f06419202
2004-07-09 22:32:27 -04:00
Lisa Hsu
bbbfebd885 Merge zizzer.eecs.umich.edu:/bk/m5
into shizzle.(none):/home/hsul/work/m5clean

--HG--
extra : convert_revision : 78beb2735c6536ce3e639d59c99e755e9139f4f1
2004-07-09 11:50:40 -04:00
Lisa Hsu
7b2691d537 when you add caches, dma commands within the state machine are delayed and cause the state machine to exit until the dma comes back. thus, all relevant code must be executed BEFORE going to do the dma code.
dev/ns_gige.cc:
    rearrange code so nothing gets skipped when "doing dma."

--HG--
extra : convert_revision : cca66885f45e7df1831e2d8ccaddf5ece7600b13
2004-07-09 11:50:27 -04:00
Nathan Binkert
7c2084dcfe store the checkpoint to the proper directory
dev/disk_image.cc:
    store the disk checkpoint to the proper directory

--HG--
extra : convert_revision : f76347576691560f00c60b86ac2044960d5a193c
2004-07-08 15:59:46 -04:00
Lisa Hsu
19fd3439c7 tag each mem_req as coming from the nic if it is DMA'd from the NIC. the NIC tells the DMA interface, which in turn sets a new nic_req flag in the MemReq it makes.
dev/ns_gige.cc:
    tell all outgoing dma events that this request is from the NIC

--HG--
extra : convert_revision : 62af17a2728a0ff729e7723dc29bd0d130ca5fe3
2004-07-06 17:55:16 -04:00
Lisa Hsu
5bbd57917f these changes are an undo from my last changeset. these are meant for later.
--HG--
extra : convert_revision : 0126918ef293cba02aaaa61dff3c471eb4743116
2004-07-05 11:20:53 -04:00
Lisa Hsu
7ca7b343ff almost forgot to do this - hope it doesn't mess up schedule.
dev/ns_gige.cc:
dev/ns_gige.hh:
    add the stats nate wanted

--HG--
extra : convert_revision : b59d586def7df31741b53cdb59cf3b19253caf26
2004-07-04 16:47:07 -04:00
Lisa Hsu
31c52ac0a7 Merge zizzer.eecs.umich.edu:/bk/m5
into shizzle.(none):/home/hsul/work/linux-clean

--HG--
extra : convert_revision : a072c1560e1f8b5cac378ed5f047a8a9c86b97b9
2004-06-30 00:51:11 -04:00
Lisa Hsu
167c57f265 fix transmit side checksum offloading to not generate a pseudo header.
dev/ns_gige.cc:
    transmit side checksum offloading doesn't need pseudo header generation, it just takes the pseudo header checksum and uses it.

--HG--
extra : convert_revision : 9741bd650415c18ed37b06a453b23610d028135b
2004-06-30 00:50:56 -04:00
Ali Saidi
5a1340d046 Another fix for the too much work problem in 2.6. This should do it.
Both rx/tx interrupts are now scheduled for the future to give the
linux kernel time to get out of its loop.

--HG--
extra : convert_revision : 8fee0a25fde0ce0545c924f8547bed460602e006
2004-06-29 16:08:26 -04:00
Ali Saidi
38fe4d9a86 With the new uart code 300 cycles isn't quite enough, 350 seems to
work. When everything gets changed to seconds this should be updated.

--HG--
extra : convert_revision : 9f1064ff6fec5deceb591904f4571c9129ecc998
2004-06-28 21:23:10 -04:00
Nathan Binkert
19d5789db1 Don't serialize the absolute path so we can move checkpoints around.
sim/serialize.cc:
sim/serialize.hh:
    export the checkpoint directory

--HG--
extra : convert_revision : ba5b796fd930ab0487a502beefcf76bd6bb7c366
2004-06-28 15:54:05 -04:00
Ali Saidi
c75eadc2b2 allow the use of old console code and update elf_object not to rely on EM_ALPHA value.
base/loader/elf_object.cc:
    EM_ALPHA value isn't official, so perhaps we shouldn't use it
dev/alpha_console.cc:
dev/alpha_console.hh:
    this change allows the use of old console code

--HG--
extra : convert_revision : cfacd64ae7fd2595158ca1a83ebcdb66ee7e119b
2004-06-27 14:33:55 -04:00
Ali Saidi
54b49f933a rewrote uart and renamed console.cc to simconsole to reduce confusion
base/traceflags.py:
    removed TsunamiUart/TlaserUart and added a plain Uart
dev/alpha_console.cc:
    updated for new simconsole
dev/platform.hh:
    added a uart member to platform
dev/simconsole.cc:
dev/simconsole.hh:
    removed lots of legacy code, it should all be ours now.
    converted tabs to 8 spaces
    added our copyright
dev/tsunami.cc:
    uses simconsole.hh rather than console.hh
dev/tsunami_cchip.cc:
dev/tsunami_io.cc:
    never needed console.hh
dev/tsunami_io.hh:
    this does need eventq.hh and it just happend to be working whenn console.hh was
    included everywhere
dev/tsunamireg.h:
    added a couple more 8250/16550 uart defines
dev/uart.cc:
    new uart code, rewritten to support both tlaser and tsunami (both a 8250 and 8530
    uart).
dev/uart.hh:
    updated for new uart, legacy code removed

--HG--
rename : dev/console.cc => dev/simconsole.cc
rename : dev/console.hh => dev/simconsole.hh
rename : dev/tsunami_uart.cc => dev/uart.cc
rename : dev/tsunami_uart.hh => dev/uart.hh
extra : convert_revision : e663352d49d4c2d3c95643030cf73c0e85ba2f08
2004-06-26 21:26:28 -04:00
Ali Saidi
4799a7b874 minor changes to endian code so that it runs on x86 linux
base/intmath.hh:
    only need FloorLog2(size_t) on a mac, so ifdefed for this
dev/alpha_console.cc:
    Actually allocate the alphaAccess struct.

--HG--
extra : convert_revision : 1f50b1a025c8ee728a9f3d2c603ea38347234f54
2004-06-23 18:39:11 -04:00
Ali Saidi
17bfb1540e Merge zizzer:/bk/linux into zeep.eecs.umich.edu:/z/saidi/work/m5-endian
--HG--
extra : convert_revision : d4938c6011173d3017f47fd592c4b5e4c8d543a3
2004-06-23 17:55:54 -04:00
Andrew Schultz
06d8f0af5d Fix to properly shadow the DEV bit in the Drive/Head register so other
disks are properly detected and handled

--HG--
extra : convert_revision : ffc3046deb68458ee2ef6fa5263dc471488abc45
2004-06-23 15:37:05 -04:00
Ali Saidi
231fac0a2b more modifications for cross-endian support. linux now gets to pciconfig
dev/alpha_console.cc:
    rather than acessing a byte array for alpha access, access the members
    **this requires an updated console**
dev/pcidev.cc:
    correctly type all the pci data and store in in little endian no
    matter what system we are on
dev/tsunami_uart.cc:
    correct a bug with the data type.
kern/linux/linux_system.cc:
    system type in hwprb needs to be endian happy as well.

--HG--
extra : convert_revision : 8de9bb69365b5d30fceaf4fa342a1639f92d7a83
2004-06-23 15:07:09 -04:00
Lisa Hsu
e05dbb5116 l
base/traceflags.py:
    added some more traceflags for ethernet to break it up better
dev/etherpkt.hh:
    since we are not network host order, must reverse bytes for these typechecks.

    also, overload isTcp/UdpPkt to take an argument so you don't have to reget the ip header if you've already got one.
dev/ns_gige.cc:
    1) add some functions that reverse Endianness so we can generate adn evaluate checksum adn dprintf data accurately/more understandably

    2) forget about the implementation of fifo fill/drain thresholds, it's not used by the driver much, nor does it matter with use sending/receiving in whole packets anyway.
    get rid of teh associated variables.

    3) get rid of txFifoCnt the variable, it's redundant and unnecessary, just use txFifoAvail.

    4) change io_enable to ioEnable, just to be picky.

    5) modify some DPRINTF's to be clearer, also added a lot, and spread them into better traceflag categories

    6) fix the device bug!  it's the intrTick = 0 at teh beginning of cpuInterrupt().

    7) clear some bools in regsReset() so they don't holdover wrong state

    8) fix pseudo header generation for Tcp checksumming to account for network order
dev/ns_gige.hh:
    change io_enable to ioEnable, get rid of fill/drain thresh related variables and txFifoCnt, which is redundant

--HG--
extra : convert_revision : c538b75731f3c9e04354f57e6df9a40aeca5096d
2004-06-21 17:25:18 -04:00
Ali Saidi
3f7b780af5 Merge saidi@zizzer.eecs.umich.edu:/bk/linux
into ali-saidis-computer.local:/research/linux

--HG--
extra : convert_revision : 446dc5fad11a6b29954dc5b1e974430853aa2969
2004-06-21 01:04:48 -04:00
Ali Saidi
c27139c701 start towards getting m5 endian compliant
base/inifile.cc:
    Added mac os support and fixed a bug, on error we need to exit the
    child process not return
base/intmath.hh:
    gcc on macos wanted a seperate function for the size_t type
base/loader/elf_object.cc:
    I'm not sure why this works under linux because it seems to return
    the wrong value.
base/stats/text.cc:
    added define/include for mac os x
cpu/exec_context.hh:
cpu/simple_cpu/simple_cpu.cc:
    added endian conversion code
dev/alpha_console.cc:
    rather than accessing a charecter array of varying size depending on
    the access, lets actually do this properly.
dev/alpha_console.hh:
    get rid of now nolonger used consoleData
dev/disk_image.cc:
    We have to byte swap the data is some cases, added function to do that
dev/ethertap.cc:
    added preproc directive for mac os

--HG--
extra : convert_revision : 2b5685765cfa2844926d7397f363d2788e3d640a
2004-06-21 00:58:30 -04:00
Andrew Schultz
f3a7930fa6 Fixes to IDE disk to fix serialization. Now passes simple CPU serialize
tests

dev/ide_ctrl.cc:
    Formatting
dev/ide_disk.cc:
    Remove some junk, add an assert to serialize, and add missing serialize
    for command register.

--HG--
extra : convert_revision : 8f99857e32f278dd4e6f23deffc8047c6411d5b2
2004-06-17 11:24:14 -04:00
Andrew Schultz
d1256a2f2c Fix serialize/unserialize of the timers and RTC events
--HG--
extra : convert_revision : aecf09b3b13a23ffef852a1539e8d4eec32008ad
2004-06-16 19:47:07 -04:00
Ali Saidi
e937b38e2c Updated serialization code and added #if tracing so that make fast
builds

dev/ide_ctrl.cc:
    added #if to remove variables that are optimized out.
dev/tsunami_io.cc:
dev/tsunami_io.hh:
    Updated serialization code

--HG--
extra : convert_revision : b322a3299097cbd05b9b5bb8b0a80e9fa33bdc20
2004-06-16 18:20:10 -04:00
Lisa Hsu
bfcb088281 minor mods for mimicking NS83820 functionality
dev/ide_ctrl.cc:
    generalize these #defs
dev/ide_ctrl.hh:
    put these in pcireg.h
dev/ns_gige.cc:
    do i need io_enable?  and assert will fail if i actually need to implement it, which may give clue as to wehtehr i need to implmeent the mem_enable and bm_enable stuff.
dev/ns_gige.hh:
    implement this in case it's needed
dev/pcireg.h:
    put these defs in pcireg instead

--HG--
extra : convert_revision : 5e3581b5da17410f943907139bd479f15d2231e8
2004-06-12 14:24:20 -04:00
Lisa Hsu
e5dba1642b fix serialization
dev/ns_gige.cc:
    fix serialization and move regsReset into the cc file
dev/ns_gige.hh:
    put regsReset into cc instead of here in hh

--HG--
extra : convert_revision : 3a8796fa583e0765503104a9dbe28cc69f1a8fa9
2004-06-11 15:26:21 -04:00
Ali Saidi
ca0dcd048c after unserialization in the ide driver we need to reprogram the
busbridges. Also small fix for gcc 3.3.3.

dev/etherpkt.hh:
    Gcc 3.3.3 wants assert.h to be include in this file.
dev/ide_ctrl.cc:
    after unserialization we need to tell the busbridges what addresses to
    respond to.

--HG--
extra : convert_revision : a421197a5be07761bdef571d0a9406d77788e270
2004-06-11 01:55:20 -04:00
Lisa Hsu
c78d847c01 Merge zizzer.eecs.umich.edu:/bk/linux
into lush.(none):/z/hsul/work/linux

--HG--
extra : convert_revision : 30b93e7a4548a00ba18e86aad85df4c790f51d43
2004-06-10 14:40:51 -04:00
Lisa Hsu
3c7f961751 fix some things on linux todo list. 1) the config.ini. 2) the etherpkt struct defs into ined_hdrs.hh
3) add some new functions to etherpkt.  4) checkin the nice ping_linux.ini that i've been using.

dev/etherpkt.hh:
    remove the packet headers nate hated so much.
    also add some new functionality regarding packets, like isIpHdr(), etc.
dev/ns_gige.cc:
    improve the code given the nice new functions offered in etherpkt

--HG--
extra : convert_revision : 2e27f5a8dca5323c0fa22d3c51af44a35f6be1a2
2004-06-10 14:27:44 -04:00
Ali Saidi
02f69b94c5 Fixes for detailed boot, made cttz and ctlz instructions more compact,
and started cleaning up config files.

arch/alpha/isa_desc:
    Made implementation of cttz and ctlz more compact
base/remote_gdb.cc:
    Added comment about PALcode debugger accesses
dev/baddev.cc:
dev/baddev.hh:
dev/ide_ctrl.cc:
dev/ide_ctrl.hh:
dev/pciconfigall.cc:
dev/pciconfigall.hh:
dev/tsunami_cchip.cc:
dev/tsunami_cchip.hh:
dev/tsunami_io.cc:
dev/tsunami_io.hh:
dev/tsunami_pchip.cc:
dev/tsunami_pchip.hh:
dev/tsunami_uart.cc:
dev/tsunami_uart.hh:
    Cleaned up includes and changed device from FunctionalMemory to
    PioDevice for detailed boot
dev/ns_gige.cc:
    The ethernet dev uses two BARs, and the first bars size was being set
    incorrectly.
dev/tsunamireg.h:
    I don't know why we were using the superpage as the PCI memory addr.
    Changed and works correctly with detailed boot.

--HG--
extra : convert_revision : b535e76612cb90b544305dc1aa8c5e0e774564bd
2004-06-10 13:30:58 -04:00