Subversion Repositories openrisc

2002-06-14  Gary Thomas  

        * src/if_upd985xx.c:

        Need to include  for proper configuration

        of stand-alone (polled) vs. system (interrupt driven) mode.

2001-09-13  Hugo Tyson  

        * src/if_upd985xx.c (upd985xx_eth_upd985xx_init): Fake an ESA if

        we see all ones from the EEPROM as well as all zeros.

2001-08-30  Hugo Tyson  

        * cdl/upd985xx_eth_drivers.cdl: Make the "..._E2ONLY" workaround

        option on by default since this is how it will be used - the CPU

        load does suffer somewhat if promisc mode is set in the hardware.

        This should be unset to allow the workaround for E1 at 100Mbit.

2001-08-30  Hugo Tyson  

        * src/if_upd985xx.c (PacketRxReady): Re-write the manual

        implementation of ESA matching for workaround E1E2 when the device

        is in promiscuous mode.  It was having problems with the previous

        version; this works better.

2001-08-24  Hugo Tyson  

        * cdl/upd985xx_eth_drivers.cdl: Configury for an alternate case

        where we workaround hardware bug E2 only, as a subset of the E1E2

        complete fix.  Added more description to the options too.

        * src/if_upd985xx.c (eth_upd985xx_configure): Handle an alternate

        case where we workaround hardware bug E2 only.  This means leaving

        the device in normal mode (unless set to promisc) and doing MAC

        address filtering by hand anyway.

2001-08-20  Hugo Tyson  

        * src/if_upd985xx.c: Guard all entrypoints that can mess with

        hardware state by "active" check.  If the net is included in an

        app, but not used, it is init'd but not started - this can leave a

        pending interrupt from RedBoot's use of the network to take us

        completely by surprise.  So init() acks and masks the interrupt,

        can_send(), recv() and deliver() now demur if not active.

        Also some additional STATIC's on entrypoint functions.

2001-08-16  Hugo Tyson  

        * src/if_upd985xx.c (upd985xx_eth_upd985xx_init): If the EEPROM

        contains nothing (or isn't fitted?) fake an ESA so we can get

        RedBoot going on the board without special configury.

2001-08-16  Hugo Tyson  

        * src/if_upd985xx.c (eth_upd985xx_configure): Use smaller numbers

        for the Tx Fill Threshold [TX_FLTH] and DMA Tx Burst Size [DTBS]

        because the hardware is even more broken than first throught -

        this is new information on fault E4.  I also tagged this with the

        name of the option we would use if this were cdl controlled - but

        since it's just setup I see no need to change it, so no CDL.

2001-08-16  Hugo Tyson  

        * src/if_upd985xx.c (TxDone): Since it still wedged occasionally,

        with an "out of rx buffers" status but nothing else, this is a

        much simplified workaround for bug E8.  If we see the suspect

        transmit status, simply reset the whole subsystem there and then.

        This leaves it in far more of a known state.  It's neater anyway.

2001-08-15  Hugo Tyson  

        * src/if_upd985xx.c (eth_upd985xx_send): Small hacks to recover

        from various wedged states with bogus or unexpected ETH_ISR

        values...  0x80000000: We detect this in the deliver routine and

        totally reset the system.  "out of rx buffers" with no "good rx":

        we unmask and check for all these RX interrupts, not just "good

        rx".  Also PacketRxReady() shortcuts to resetting the receive

        engine when it sees the problem.  I suspect these might be caused

        by the E8 workaround below, perhaps introducing some race

        condition with turning off the receiver just when it rx'd - and of

        course E1E2 means it receives far more packets.

2001-08-07  Hugo Tyson  

        * src/if_upd985xx.c: Workaround various bugs in the hardware;

        these workarounds are conditionally compiled via CDL options named

        CYGOPT_DEVS_ETH_MIPS_UPD985XX_HARDWARE_BUGS_xxx in general; 'xxx'

        is the reference for the mis-feature.  All are enabled by default.

        To summarize:

        (eth_upd985xx_reset): xxx=S1: insert reads between writes to the

        device to avoid a possible deadlock; macro FLUSH_WRITES().

        (PacketRxReady): xxx=E1E2: we set the device in promiscuous mode

        always, and implement ESA matching in code.  The cost is small.

        If promisc mode is set by the stack, we pass all packets.

        (eth_upd985xx_send): xxx=E3: we copy any transmit that uses 3 or

        more SGs into a static contiguous buffer and transmit from that

        thus using only one buffer descriptor.

        (eth_upd985xx_send):

        (TxDone): xxx=E8: we make a note that a tx ended badly and when

        starting the next tx, we disable and reset the transmitter.

        * cdl/upd985xx_eth_drivers.cdl: New subcomponent for controlling

        these workarounds: CYGPKG_DEVS_ETH_MIPS_UPD985XX_HARDWARE_BUGS.

        "Workarounds for Ethernet Hardware bugs"

2001-07-16  Hugo Tyson  

        * src/if_upd985xx.c (PacketRxReady): Test for, rather than assert,

        packet size in range.  The hardware can report a tiny packet as

        AOK, with no bad in the status, despite the doc's reassurances.

2001-07-13  Hugo Tyson  

        * cdl/upd985xx_eth_drivers.cdl: Turn off the startup chatter.

2001-07-13  Hugo Tyson  

        * src/if_upd985xx.c (TxDone): Test a few more bits for tx

        complete; it turns out you can get tx underruns when the CPU us

        heavily loaded, as in the tcp_echo tests with high load.

2001-07-13  Hugo Tyson  

        * src/if_upd985xx.c (eth_upd985xx_send): Use HAL_DCACHE_STORE()

        rather than syncing the whole of cache every sglist entry(=mbuf).

        Turns out the cache op must be cache line aligned to work on the

        4120, boo, unlike other MIPS and unlike the doc, even.

2001-07-12  Hugo Tyson  

        * cdl/upd985xx_eth_drivers.cdl (CYGPKG_DEVS_ETH_MIPS_UPD985XX_ETH0):

        Whole new section to address configuring the source of the MAC

        address.  Also allows configury of the device's name ("eth0") for

        cohabitation with additional devices.

        * src/if_upd985xx.c (upd985xx_eth_upd985xx_init): Pick up the ESA

        from EEPROM if it's available, also support a fixed ESA from CDL

        configuration land.  A few minor changes to the structure

        initialization to accommodate this; also pick up the interrupt

        vector from struct init.

        (eth_set_mac_address): New routine available via the ioctl()

        entry, for use when neither a fixed nor EEPROM address is

        available.

2001-07-12  Hugo Tyson  

        * src/if_upd985xx.c (NUM_RXBUFS): Reduce NUM_RXBUFS to 8; IME

        fewer results in lost rx's in typical systems.  Enlarge rx buffers

        slightly, to accommodate oversize VLAN packets.  128 bytes extra

        should be enough.  Implemented eth_upd985xx_configure() selection

        of promiscuous mode and allow oversize packets - up to the allowed

        oversize.  Otherwise we would get confused if a packet ate more

        than 1 rx buffer.

2001-07-12  Hugo Tyson  

        * src/if_upd985xx.c: Tidy up debug print defaults; make functions

        static; add a few extra statistics to the device object; pass a

        p_eth_upd985xx around more consistently for if we switch to

        multiple devices in future; comment out mii_write(); handle

        stopping the device with a tx pending; remove some commented-out

        templates copied from another driver; and fill in SNMP statistics.

        In other words, many minor changes.

2001-07-11  Hugo Tyson  

        * src/if_upd985xx.c (PacketRxReady): Recover from running out of

        receive buffers.  All very dodgy, but it seems to work.

        Additional efforts are also made to reset the device, having

        realized how hard it is to re-initialize the receive engine once

        it has been awakened.

2001-07-11  Hugo Tyson  

        * src/if_upd985xx.c: Efficient Rx now essentially working, with a

        simple circular buffer, always linked into a ring, and one entry

        of which is always NULL,NULL to bring the rx machinery to a halt.

        If it reaches thus far the rx mechanism seems to jam; will deal

        with that next.

2001-07-09  Hugo Tyson  

        * src/if_upd985xx.c: Rejigged version of the rx buffer system;

        still not working properly, still not a good match for the

        hardware's elusive semantics.  Committed anyway, to keep it around

        for reference.

        (eth_upd985xx_status): Also removed all the cruft about

        renegotiating line status; it's not needed.

2001-07-06  Hugo Tyson  

        * include/upd985xx_eth.h (ETH_MADR_PHY_DEVICE_PHYS_ADDRESS):

        Change name of PHY address symbol to generic not SEEQ.

        Comment out the non-standard symbols for useful bits that the

        previous PHY device supported.

        * src/if_upd985xx.c (eth_upd985xx_reset): If there is a valid ESA

        in the MAC already, run with it - it would have come from the

        not-fitted serial EEPROM, via some different registers.

        (upd985xx_eth_upd985xx_init): Moved the call to reset about to

        accommodate this.

        (eth_upd985xx_status): Omit renegotiation of link properties and

        use the intersection of the capabilities bits to report what

        speed, duplex, we are running at.  More portable.

        (mii_write):

        (mii_read): Change name of PHY address symbol to generic not SEEQ

        'cos the board has changed.

2001-07-06  Hugo Tyson  

        * ChangeLog:

        * cdl/upd985xx_eth_drivers.cdl:

        * include/upd985xx_eth.h:

        * src/if_upd985xx.c:

        New files.  Initial checkin of limping along version of

        NEC upd985xx ethernet driver.

        Limitations:

        ESA is hard coded.

        It talks to the PHY just to make sure - helped with debug anyway.

        No SNMP data exported.

        No ioctl() for promiscuous mode or VLAN mode.

        Only one TX at once.

        Only one RX buffer, so no RX until serviced.

        It seems to loose interrupts - inevitably, for an eth device - and

        there's no "catchup" defense against this yet.

        It's oriented to the "old" (already) board - so the particular PHY

        and GPIO layout.

//===========================================================================

// ####GPLCOPYRIGHTBEGIN####

// -------------------------------------------

// This file is part of eCos, the Embedded Configurable Operating System.

// Copyright (C) 1998, 1999, 2000, 2001, 2002 Free Software Foundation, Inc.

//

// This program is free software; you can redistribute it and/or modify

// it under the terms of the GNU General Public License as published by

// the Free Software Foundation; either version 2 or (at your option) any

// later version.

//

// This program is distributed in the hope that it will be useful, but

// WITHOUT ANY WARRANTY; without even the implied warranty of

// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

// General Public License for more details.

//

// You should have received a copy of the GNU General Public License

// along with this program; if not, write to the

// Free Software Foundation, Inc., 51 Franklin Street,

// Fifth Floor, Boston, MA  02110-1301, USA.

// -------------------------------------------

// ####GPLCOPYRIGHTEND####

//===========================================================================