Modified mii_init to support boards with PHYs that are not set to
autonegotiate, but still want to use u-boot's mii commands to probe
the smi bus. Such PHYs will not set the Autonegotiate-done bit.
Signed-off-by: Richard Retanubun <RichardRetanubun@RuggedCom.com>
Redo how pointers are managed to get rid of ugly casts and strict pointer
aliasing issues that are highlighted by gcc 4.3.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Acked-by: Ben Warren <biggerbadderben@gmail.com>
No point in having a Blackfin-specific define "CONFIG_BFIN_MAC_RMII" that
does exactly the same thing as common "CONFIG_RMII".
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Acked-by: Ben Warren <biggerbadderben@gmail.com>
Rather then defining our own DEBUGF(), just use the common debug().
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Acked-by: Ben Warren <biggerbadderben@gmail.com>
Rather than having the on-chip MAC hardcoded to phy address 1 and a speed
of 2.5mhz, use these as defaults if the board doesn't specify otherwise.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Acked-by: Ben Warren <biggerbadderben@gmail.com>
Cleanup and rewrite the MII/PHY related functions so that we can reuse the
existing common linux/miiphy.h code and hook into the `mii` command.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Acked-by: Ben Warren <biggerbadderben@gmail.com>
Rather than hardcoding MDCDIV to 24 (which is correct for ~125mhz SCLK),
use the real algorithm so it gets set correctly regardless of SCLK.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Acked-by: Ben Warren <biggerbadderben@gmail.com>
This patch adds support for searching through available PHY-addresses in
the macb-driver. This is needed for the ATEVK1100 evaluation board,
where the PHY-address will be initialized to either 1 or 7.
This patch adds a config option, CONFIG_MACB_SEARCH_PHY, which when
enabled tells the driver to search for the PHY address.
Signed-off-by: Gunnar Rangoy <gunnar@rangoy.com>
Signed-off-by: Paul Driveklepp <pauldriveklepp@gmail.com>
Signed-off-by: Olav Morken <olavmrk@gmail.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
This patch adjusts the LED control so that interrupt lines are not reading LEDs
and effectively causing indefinite interrupts to the controller.
Signed-off-by: Ron Madrid <ron_madrid@sbcglobal.net>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
sh_eth used old network API. This patch changed new API.
Signed-off-by: Nobuhiro Iwamatsu <iwamatsu.nobuhiro@renesas.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
This patch changes the reg_read/_write to smc911x_reg_read/_write
and defines then as weak so that they can be overridden by board
specific version.
This will be used by the upcoming VCTH board support.
Signed-off-by: Stefan Roese <sr@denx.de>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
This patch tries to ensure that phy interrupt pin
won't be asserted after booting. We experienced
following issues with current 88E1121R phy init:
Marvell 88E1121R phy can be hardware-configured
to share MDC/MDIO and interrupt pins for both ports
P0 and P1 (e.g. as configured on socrates board).
Port 0 interrupt pin will be shared by both ports
in such configuration. After booting Linux and
configuring eth0 interface, port 0 phy interrupts
are enabled. After rebooting without proper eth0
interface shutdown port 0 phy interrupts remain
enabled so any change on port 0 (link status, etc.)
cause assertion of the interrupt. Now booting Linux
and configuring eth1 interface will cause permanent
phy interrupt storm as the registered phy 1 interrupt
handler doesn't acknowledge phy 0 interrupts. This
of course should be fixed in Linux driver too.
Signed-off-by: Anatolij Gustschin <agust@denx.de>
Acked-by: Andy Fleming <afleming@freescale.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
All CF platforms' mii.c are consolidated into one
Signed-off-by: TsiChung Liew <Tsi-Chung.Liew@freescale.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Also changed path in all linker scripts that reference this driver
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Acked-by: Stefan Roese <sr@denx.de>
Patch to fix buffer allocation size and alignment. Buffer needs to be u32 aligned and
PKTSIZE_ALIGN bytes long.
Acked-by: Michal Simek <monstr@monstr.eu>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Currently VSC8601 doesn't link with 10/100M partners if the
EEPROM/Strapping is not set up.
Setting the auto-neg register fixes this.
Signed-off-by: Andre Schwarz <andre.schwarz@matrix-vision.de>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
get_prom function was used __attriute__ , but it is not enable.
ax88796.o does not do link besides ne2000.o. When ld is carried
out, get_prom function of ax88796.c is ignored.
This problem is a thing by specifications of ld.
I checked and test this patch on SuperH and MIPS.
Signed-off-by: Nobuhiro Iwamatsu <iwamatsu.nobuhiro@renesas.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Fix TBI PHY accesses to use the proper offset in CPU register space. The
previous code would incorrectly access the TBI PHY by reading/writing to CPU
register space at the same location as would be used to access external PHYs.
Signed-off-by: Peter Tyser <ptyser@xes-inc.com>
Acked-by: Andy Fleming <afleming@freescale.com>
The RSK7203 board has the SMSC9118 wired up 'incorrectly'.
Byte-swapping is necessary, and so poor performance is inevitable.
This problem cannot evade by the swap function of CHIP, this can
evade by software Byte-swapping.
And this has problem by FIFO access only. pkt_data_pull/pkt_data_push
functions necessary to solve this problem.
Signed-off-by: Nobuhiro Iwamatsu <iwamatsu.nobuhiro@renesas.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Modified board_eth_init() functions of boards that have this FEC in addition
to other Ethernet controllers.
Affected boards:
bc3450
icecube
mvbc_p
o2dnt
pm520
total5200
tq5200
Removed initialization of controller from net/eth.c
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Affected boards:
hidden_dragon
MPC8544DS
MPC8610HPCN
R2DPLUS
TB0229
Removed initialization of the driver from net/eth.c
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
This addresses all drivers whose initializers have already
been moved to board_eth_init()/cpu_eth_init().
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Adds support for configuring the TBI to talk properly with the SerDes.
Signed-off-by: Andy Fleming <afleming@freescale.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
The tsec driver contains a hard-coded array of configuration information
for the tsec ethernet controllers. We create a default function that works
for most tsecs, and allow that to be overridden by board code. It creates
an array of tsec_info structures, which are then parsed by the corresponding
driver instance to determine configuration. Also, add regs, miiregs, and
devname fields to the tsec_info structure, so that we don't need the kludgy
"index" parameter.
Signed-off-by: Andy Fleming <afleming@freescale.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
This is to prepare the way for board code passing in the tsec_info structure
Signed-off-by: Andy Fleming <afleming@freescale.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
dm9000 has code to detect and initialize external phy parts, but later
on in the code the part is forced to use the internal phy
unconditionally. Remove the unused/untested code.
change the GPIO initialization so that only the GPIO used as an
internal phy reset (hardwired in the chip) is set as an output. The
remaining GPIO need to be handled by board specific code to prevent
possible drive conflicts. Set as inputs for safety.
replace a few magic numbers with defines
Signed-off-by: Andrew Dyer <adyer@righthandtech.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Add FEC Buffer descriptors and data buffer in SRAM for
faster execution and access.
Signed-off-by: TsiChung Liew <Tsi-Chung.Liew@freescale.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
The only board using this driver is the SL8245 board.
Removed initialization for the driver from net/eth.c
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Added board_eth_init() function to bf537-stamp board.
Removed initialization for the Blackin EMAC driver from net/eth.c
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
This patch fixes a problem that RTL8110SCL started transfer
with an incorrect memory address.
Signed-off-by: Yoshihiro Shimoda <shimoda.yoshihiro@renesas.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
The received status and len was in little endian
format and caused the ethernet unable to proceed
further. Add __le16_to_cpu() in dm9000_rx_status_8/16/32bit().
Signed-off-by: TsiChung Liew <Tsi-Chung.Liew@freescale.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
This PCI-X e1000 variant works by just adding in the correct
PCI IDs in the appropriate places.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Remove duplicate definitions in include/lxt971a.h.
Remove duplicate registers and bits definitions in
include/lxt971a.h for standard MII registers, and
use values in include/miiphy.h instead.
Signed-off-by: Hugo Villeneuve <hugo.villeneuve@lyrtech.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Renesas SH7763 has 2 channel Ethernet device.
This is 10/100/1000 Base support.
But this patch check 10/100 Base only.
Signed-off-by: Nobuhiro Iwamatsu <iwamatsu.nobuhiro@renesas.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
If enable DEBUG, can not compile ne2000 driver.
Signed-off-by: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
According to the Application Notes of the DM9000, only the 2 bits 0:1 of
the status byte need to be checked to identify a valid packet in the fifo
But, The several different Application Notes do not all speak the same
language on these bits. They do not disagree, but only 1 Application Note
noted explicitly that only these 2 bits need to be checked.
Even the datasheets do not mention anything about these 2 bits.
Because the old code, and the kernel check the whole byte, I left this piece
untouched.
However, I tested all board/DM9000[A|E|EP] devices with this 2 bit check, so
it should work.
Notice, that the 2nd iteration through this receive loop (when a 2nd packet is
in the fifo) is much shorter now, compared to the older U-boot driver code,
so that we can maybe run into a hardware condition now that was never seen
before, or maybe was seen very unfrequently.
Additionaly added a cleanup of a stack variable.
Signed-off-by: Remy Bohmer <linux@bohmer.net>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
A last minute cleanup before submitting the DM9000A patch series yesterday introduced
a bug in reading the rx-status registers in 32bit mode only.
This patch repairs this.
Signed-off-by: Remy Bohmer <linux@bohmer.net>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Some lines of the U-boot DM9000x driver are longer than 80 characters, or
need some other minor cleanup.
Signed-off-by: Remy Bohmer <linux@bohmer.net>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
The DM9000A network controller does not work with the U-boot DM9000x driver.
Analysis showed that many incoming packets are lost.
The DM9000A Application Notes V1.20 (section 5.6.1) recommend that the poll to
check for a valid rx packet be done on the interrupt status register, not
directly by performing the dummy read and the rx status check as is currently
the case in the u-boot driver.
When the recommended poll is done as suggested the driver starts working
correctly on 10Mbit/HD, but on 100MBit/FD packets come in faster so that there
can be more than 1 package in the fifo at the same time.
The driver must perform the rx-status check in a loop and read and handle all
packages until there is no more left _after_ the interrupt RX flag is set.
This change has been tested with DM9000A, DM9000E, DM9000EP.
Signed-off-by: Remy Bohmer <linux@bohmer.net>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
According to the application notes of the DM9000 v1.22 chapter 5.2 bullet 2, the
reset procedure must be done twice to properly reset the DM9000 by means of software.
This errata is not needed anymore for the DM9000A, but it does not bother it.
This change has been tested with DM9000A, DM9000E, DM9000EP.
Signed-off-by: Remy Bohmer <linux@bohmer.net>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
The eth_send routine of the U-boot DM9000x driver does not match the
DM9000 or DM9000A application notes/programming guides.
This change improves the stability of the DM9000A network controller.
This change has been tested with DM9000A, DM9000E, DM9000EP.
Signed-off-by: Remy Bohmer <linux@bohmer.net>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
It seems that the debugging code of the DM9000x driver in U-boot has not been
compiled for a long time, because it cannot compile...
Also rearranged some loglines to get more useful info while debugging.
Signed-off-by: Remy Bohmer <linux@bohmer.net>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
The U-boot DM9000x driver contains a compile time bus-width definition for
the databus connected to the network controller.
This compile check makes the code unclear, inflexible and is unneccessary.
It can be asked to the network controller what its bus-width is by reading bits
6 and 7 of the interrupt status register.
The linux kernel already uses a runtime mechanism to determine this bus-width,
so the implementation below looks somewhat like that implementation.
This change has been tested with DM9000A, DM9000E, DM9000EP.
Signed-off-by: Remy Bohmer <linux@bohmer.net>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
This commit gets rid of a huge amount of silly white-space issues.
Especially, all sequences of SPACEs followed by TAB characters get
removed (unless they appear in print statements).
Also remove all embedded "vim:" and "vi:" statements which hide
indentation problems.
Signed-off-by: Wolfgang Denk <wd@denx.de>
This patch adds support for the AT91SAM9263 chip and the AT91SAM9263EK
board.
Signed-off-by: Stelian Pop <stelian@popies.net>
Signed-off-by: Jean-Christophe PLAGNIOL-VILLARD <plagnioj@jcrosoft.com>
This patch adds support for the AT91SAM9261 chip and the AT91SAM9261EK
board.
Signed-off-by: Stelian Pop <stelian@popies.net>
Signed-off-by: Jean-Christophe PLAGNIOL-VILLARD <plagnioj@jcrosoft.com>
When applying the AT91CAP9 patches upstream, something transformed
the '@' character into the ' <at> ' sequence.
The patch below restores the original form in all the places where
it has been modified (the AT91CAP9 files, the AT91SAM9260 files which
were copied from AT91CAP9, and a couple of other files where the
' <at> ' sequence was present).
Signed-off-by: Stelian Pop <stelian@popies.net>
When applying the AT91CAP9 patches upstream, something transformed
the '@' character into the ' <at> ' sequence.
The patch below restores the original form in all the places where
it has been modified (the AT91CAP9 files, the AT91SAM9260 files which
were copied from AT91CAP9, and a couple of other files where the
' <at> ' sequence was present).
Signed-off-by: Stelian Pop <stelian@popies.net>
Signed-off-by: Jean-Christophe PLAGNIOL-VILLARD <plagnioj@jcrosoft.com>
The Vitesse VSC8601 RGMII PHY has internal delay for both Rx
and Tx clock lines. They are configured using 2 bits in extended
register 0x17.
Therefore CFG_VSC8601_SKEW_TX and CFG_VSC8601_SKEW_RX have
been introduced with valid values 0-3 giving 0.0, 1.4,1.7 and 2.0ns delay.
Signed-off-by: Andre Schwarz <andre.schwarz@matrix-vision.de>
Acked-by: Andy Fleming <afleming@freescale.com>
Acked-by: Ben Warren <biggerbadderben@gmail.com>
--
drivers/net/tsec.c | 6 ++++++
drivers/net/tsec.h | 3 +++
2 files changed, 9 insertions(+), 0 deletions(-)
The ethernet hang is caused by receiving buffer in DRAM is not
yet ready due to access cycles require longer time in DRAM.
Relocate DMA buffer descriptors from DRAM to internal SRAM.
Signed-off-by: TsiChung Liew <Tsi-Chung.Liew@freescale.com>
Move non-inlied functions into specific drivers file
Set get_prom as weak
Signed-off-by: Jean-Christophe PLAGNIOL-VILLARD <plagnioj@jcrosoft.com>
Signed-off-by: Vlad Lungu <vlad@comsys.ro>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Remove a redundant register definition, clean up some coding style
violations.
Signed-off-by: Guennadi Liakhovetski <lg@denx.de>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
This patch makes SPARC/LEON processors able to read and write
to the SMC91111 chip using the chip external I/O bus of the memory
controller. This patchs defines the standard in and out macros
expected by the SMC9111 driver.
To access that I/O bus one must set up the memory controller
(MCTRL or FTMCTRL) correctly. It is assumed that the user sets
up this correctly when the other MCTRL parameters are set up. It
can be set up from the board configuration header file.
Signed-off-by: Daniel Hellstrom <daniel@gaisler.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
This patch adds a driver for the following smsc network controllers:
LAN9115
LAN9116
LAN9117
LAN9215
LAN9216
LAN9217
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: Guennadi Liakhovetski <lg@denx.de>
Add 82541ER device with latest integrated IGP2 PHY.
Introduced CONFIG_E1000_FALLBACK_MAC for NIC bring-up with empty eeprom.
Signed-off-by: Andre Schwarz <andre.schwarz@matrix-vision.de>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
GRETH is an Ethernet 10/100 or 10/100/1000 MAC with out without
a debug link (EDCL). The GRETH core is documented in GRIP.pdf
available at www.gaisler.com.
If the GRETH has GigaBit support (GBIT, Scatter gather, checksum
offloading etc.) can be determined by a bit in the control register.
The GBIT MAC is supported by operating in GRTEH 10/100 legacy mode.
Signed-off-by: Daniel Hellstrom <daniel@gaisler.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
RTL8211B sets link state register after autonegotiation complete,
so with bootdelay=0 RTL8211B will report lack of the link.
To fix this, we should wait for aneg to complete, even if the
link is currently down.
Signed-off-by: Anton Vorontsov <avorontsov@ru.mvista.com>
Signed-off-by: Kim Phillips <kim.phillips@freescale.com>
The Vitesse VSC7385 is a 5-port switch found on the Freescale MPC8349E-mITX
and other boards. A small firwmare must be uploaded to its on-board memory
before it can be enabled. This patch adds the code which uploads firmware
(but not the firmware itself).
Previously, this feature was provided by a U-Boot application that was
made available only on Freescale BSPs. The VSC7385 firmware must still
be obtained separately, but at least there is no longer a need for a separate
application.
Signed-off-by: Timur Tabi <timur@freescale.com>
Acked-by: Ben Warren <biggerbadderben@gmail.com>
AX88796L is device of NE2000 compatible.
This patch support AX88796L ethernet device.
Signed-off-by: Yusuke Goda <goda.yusuke@renesas.com>
Acked-by: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
There are more devices of the NE2000 base.
A present code is difficult for us to support more devices.
To support more NE2000 clone devices, separated the function.
Signed-off-by: Yusuke Goda <goda.yusuke@renesas.com>
Acked-by: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
The Blackfin on-chip MAC driver was being managed in the BF537-STAMP board
directory, but it is not board specific, so relocate it to the drivers dir
so that other Blackfin ports can utilize it.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Since the "ssync" instruction may have hardware anomalies associated with
it, have the smc91111 driver use the SSYNC macro rather than invoking it
directly. We workaround all the anomalies via this macro.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
When rtl_recv() of rtl8169 is called, OWNbit of status register
is not enable occasionally.
rtl_recv() doesn't work normally when the driver doesn't do
appropriate processing.
This patch fix this problem.
Signed-off-by: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
rx_status on the memory is basically in LE, but needs to be handled in CPU
endian. le32_to_cpu() takes up this mission. Even if on the sane hardware,
it'll work fine.
Signed-off-by: Shinya Kuribayashi <shinya.kuribayashi@necel.com>
Cc: Masami Komiya <mkomiya@sonare.it>
Cc: Lucas Jin <lucasjin@gmail.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Current driver is meant for cache coherent systems. This patch adds
flush_cache() routines to support cache non-coherent systems.
Signed-off-by: Shinya Kuribayashi <shinya.kuribayashi@necel.com>
Cc: Masami Komiya <mkomiya@sonare.it>
Cc: Lucas Jin <lucasjin@gmail.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
"to = (currticks() + RTL_TIMEOUT)" has possibilities to wrap around. If it
does, the condition "(currticks() < to)" becomes invalid and immediately
leads to tx timeout error. This patch introduces the fine-graded udely(10)
loops to ease the impact of wrapping around.
Signed-off-by: Shinya Kuribayashi <shinya.kuribayashi@necel.com>
Cc: Masami Komiya <mkomiya@sonare.it>
Cc: Lucas Jin <lucasjin@gmail.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Add the support of RealTek RTL8211B PHY, the RTL8211B
PHY only supports RGMII and MII mode.
Signed-off-by: Dave Liu <daveliu@freescale.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
The current TSEC driver limits MDIO access to the devices that have been configured as attached
to a TSEC MAC. This patch allows access to any PHY device on the MDIO bus through the 'mii' commands.
Signed-off-by: Michael Firth <michael.firth@bt.com>
Acked-by: Andy Fleming <afleming@freescale.com>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Change return values of init() functions in all Ethernet drivers to conform
to the following:
>=0: Success
<0: Failure
All drivers going forward should return 0 on success. Current drivers that
return 1 on success were left as-is to minimize changes.
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>
Acked-by: Stefan Roese <sr@denx.de>
Acked-by: Jean-Christophe PLAGNIOL-VILLARD <plagnioj@jcrosoft.com>
Acked-by: Kim Phillips <kim.phillips@freescale.com>
Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com>
Acked-By: Timur Tabi <timur@freescale.com>
It is unclear on what platforms this driver has been tested, since
noone up to now defines CONFIG_RTL8169 in the board configuration
header. Now it has been fixed for a big-endian mpc8241 based
linkstation platform. This patch presents the necessary endianness
conversion fixes.
Signed-off-by: Guennadi Liakhovetski <lg@denx.de>
The MPC837xEMDS board support:
* DDR2 400MHz hardcoded and SPD init
* Local bus NOR Flash
* I2C, UART, MII and RTC
* eTSEC RGMII
* PCI host
Signed-off-by: Dave Liu <daveliu@freescale.com>
Enabled cache in cpu_init_f() for faster flash to mem allocation. Updated cache handling in start.S. Applied cache invalidate in fec_send() and fec_recv(). Added CFG_UNIFY_CACHE for CF V3 only.
Signed-off-by: TsiChung <tcliew@Goku.(none)>
