nds32: toolchains for building ?

where can we actually get a toolchain to build nds32 ? support doesn't appear to be in the mainline FSF projects ...
-mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
Url : http://lists.denx.de/pipermail/u-boot/attachments/20111103/79abadb5/attachment.pgp

Hi Mike,

2011/11/4 Mike Frysinger <[email protected]>:

Unfortunately both the toolchain and the Linux kernel haven't commit back to mainline yet.
You can only found it in autotools which is in FSF projects. :-( The official explanation is that the company has put the code commit activity in schedule.
But I and my colledgue are not sure when will it be started. :-( We don't know if the CEO has signed the agreement with FSF already or not. :-(

Current toolchain we're using is gcc-4.5, binutils-2.19, the newer version is still under developing.

It's pity that only you can found the toolchain with full source from public is very old.
I can give you the url as follows.
http://osdk.andestech.com/

Currently, only the customers and universities could be able to get the newest toolchain and BSP right now.

U-boot is the first project which Andestech to commit the source code into mainline.
Hope this could encourage the company to commit the source code into mainline more active.

Thanks.

Best regards,
Macpaul Lin

yes, i d/l-ed that monster of a .rar (which contains .tar.gz which themselves contain .tar.gz .....), but the linker in there segfaults for me :(

for people who want to try building u-boot but skip d/lin a 100MB+ rar, i've ripped things apart and posted it here as 3MiB:

	http://dev.gentoo.org/~vapier/u-boot/nds32.tar.xz

Configuring for adp-ag101 board...
board.c:107: warning: 'pmu_init' defined but not used
cmd_bdinfo.c:489: warning: 'print_lnum' defined but not used
cmd_mmc.c: In function `do_mmcops':
cmd_mmc.c:230: warning: comparison is always false due to limited range of data type
cmd_mmc.c:245: warning: comparison is always false due to limited range of data type
ns16550.c: In function `NS16550_init':
ns16550.c:44: warning: passing arg 2 of `writeb' makes pointer from integer without a cast
/bin/sh: line 1: 28408 Segmentation fault (core dumped) nds32le-linux-ld
--gc-sections --relax -T u-boot.lds -Bstatic -Ttext 0x03200000 $UNDEF_SYM arch/nds32/cpu/n1213/start.o --start-group api/libapi.o arch/nds32/cpu/n1213/ag101/libag101.o arch/nds32/cpu/n1213/libn1213.o arch/nds32/lib/libnds32.o common/libcommon.o disk/libdisk.o drivers/bios_emulator/libatibiosemu.o drivers/block/libblock.o drivers/dma/libdma.o drivers/fpga/libfpga.o drivers/gpio/libgpio.o drivers/hwmon/libhwmon.o drivers/i2c/libi2c.o drivers/input/libinput.o drivers/misc/libmisc.o drivers/mmc/libmmc.o drivers/mtd/libmtd.o drivers/mtd/nand/libnand.o drivers/mtd/onenand/libonenand.o drivers/mtd/spi/libspi_flash.o drivers/mtd/ubi/libubi.o drivers/net/libnet.o drivers/net/phy/libphy.o drivers/pci/libpci.o drivers/pcmcia/libpcmcia.o drivers/power/libpower.o drivers/rtc/librtc.o drivers/serial/libserial.o drivers/spi/libspi.o drivers/twserial/libtws.o drivers/usb/eth/libusb_eth.o drivers/usb/gadget/libusb_gadget.o drivers/usb/host/libusb_host.o drivers/usb/musb/libusb_musb.o drivers/usb/phy/libusb_phy.o drivers/video/libvideo.o drivers/watchdog/libwatchdog.o fs/cramfs/libcramfs.o fs/ext2/libext2fs.o fs/fat/libfat.o fs/fdos/libfdos.o fs/jffs2/libjffs2.o fs/reiserfs/libreiserfs.o fs/ubifs/libubifs.o fs/yaffs2/libyaffs2.o lib/libfdt/libfdt.o lib/libgeneric.o lib/lzma/liblzma.o lib/lzo/liblzo.o lib/zlib/libz.o net/libnet.o post/libpost.o board/AndesTech/adp-ag101/libadp- ag101.o --end-group -L /usr/local/src/u-boot/toolchains/nds32le- linux/bin/../lib/gcc/nds32le-linux/3.4.4 -lgcc -Map u-boot.map -o u-boot
make: *** [u-boot] Error 139
-mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
Url : http://lists.denx.de/pipermail/u-boot/attachments/20111104/9685fef5/attachment.pgp

seems to partially be due to the nds32 u-boot.lds not handling .data.rel sections which a few objects have, so the linker just arbitrarily appends them. you should probably tweak the linker script, perhaps like so:

@@ -41,7 +41,7 @@ SECTIONS
 	.rodata : { *(SORT_BY_ALIGNMENT(SORT_BY_NAME(.rodata*))) }
 
 	. = ALIGN(4);
-	.data : { *(.data) }
+	.data : { *(.data*) }
 
 	. = ALIGN(4);
 

fixing that though just gets me a diff bug:
nds32le-linux-ld: BFD 2.16.91 internal error, aborting at ../../bfd/section.c line 1229 in bfd_map_over_sections
nds32le-linux-ld: Please report this bug.
-mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
Url : http://lists.denx.de/pipermail/u-boot/attachments/20111104/a33bc1c3/attachment.pgp

Hi Mike,

2011/11/5 Mike Frysinger <[email protected]>:

The old toolchain contains a lot of bugs, even the implementation of rpath has problems.
Please do not waste your time!
I'll try to get one for you next week. :)

Best regards,
Macpaul Lin

thanks. i think you should still fix that linker script issue i pointed out as it seems like the linker is putting stuff after the _end/.bss in your u-boot image ... not that i can double check as i can't link the final ELF and analyze it :).
-mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
Url : http://lists.denx.de/pipermail/u-boot/attachments/20111105/b6b3c715/attachment.pgp

Hi Mike,

2011/11/6 Mike Frysinger <[email protected]>:

I'm very thanks for your great help.
I've updated the linker script, and...

The toolchain binaries are available at the following url.
http://osdk.andestech.com/pre-release.html

The latest toolchain version is gcc 4.5 built on 2011.07. Which you can downloaded is gcc 4.5 built on 2011.02

Thanks!

Best regards,
Macpaul Lin

awesome !

what's the difference between v0 v1 v1f ?
-mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
Url : http://lists.denx.de/pipermail/u-boot/attachments/20111110/45d58d43/attachment.pgp

Hi Mike,

2011/11/11 Mike Frysinger <[email protected]>:

It should be described in detail in the README, however, the v0 toolchain will be phased out in recent. So it becomes not so important. XD

The version numbers v0, v1, v1f, v2, v2f, v3, and v3m are meant to instruction set (ISA) versions.
Currently, the first SoC ag101 can be adapted with v0 toolchain only. The second SoC is soft core ag101p, which can run the most compatible v1 instruction sets.
While v1f means v1 ISA + floating point ISA.
Other details of v2 and v3 ISA should be able to be found though url
http://www.andestech.com/

However, I don't know where exactly url of the document is. :p Please use v0 toolchain to play with adp-ag101 board. Although you can use v1 toolchain to build the binary for ag101 SoC, it couldn't be booted up.
For the backward compatibility, of course, you can use v0 and v1 toolchain to build binary for ag101p SoC.
Both of the binaries could run correctly.

Best regards,
Macpaul Lin

so the ISA version only affects the generated libraries right ? i can still use any of the compilers to output any ISA version ?

i grabbed the v1f and it seems to build the one nds32 board in the tree fine.

i ripped it down to the min needed for u-boot, so people can grab the 4.3MiB:

	http://dev.gentoo.org/~vapier/u-boot/nds32.tar.xz

-mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
Url : http://lists.denx.de/pipermail/u-boot/attachments/20111110/49b28397/attachment.pgp

HI Mike,

2011/11/11 Mike Frysinger <[email protected]>:

Unfortunately, wrong.

ISA will affect the assembly and binary patterns, also some basic stack operation.
You cannot run v1 v1f binary on the adp-ag101 board.
You also cannot run v1f on the chip which didn't have floating point unit. So, you cannot run v1f binary on adp-ag101p board.
There will decode floating related ISA as non-floating point then lead fault.

The capability is listed as follows.
adp-ag101: v0 binary only.
adp-ag101p: v0 and v1 binary.
adp-ag102: v0, v1, v1f

Best regards,
Macpaul Lin

Hi Mike

In much more detail, v0 has stack conflict will all other toolchain since the program counter
register was designed forbidden to access directly in 3 years ago.

V1 and later ISA has a common behavior on stack operation since the program counter is
available directly.

However, I don't know if there are parameters to turn-off the floating binary generation in v1f
toolchain. Even I don't know if there are parameters to generation v0's binary with v1 toolchain.
I think they are defined in compiling time when the toolchain is built.

Best regards,
Macpaul Lin

right, which is why gcc ports have options like -march=, -mcpu=, -mtune=, etc... this way you only need 1 gcc to support all these different ISAs.

nds32le-linux-gcc -march=v0 -c ......
nds32le-linux-gcc -march=v1 -c ......
nds32le-linux-gcc -march=v1f -c ......

what i meant was that you don't have to configure gcc for a specific ISA. -mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
Url : http://lists.denx.de/pipermail/u-boot/attachments/20111110/c96126ab/attachment.pgp

Hi Mike,

2011/11/11 Mike Frysinger <[email protected]>:

I just checked it out with my toolchain collegues, they said the option is -mbaseline.
Please use 'nds32le-linux-gcc --target-help' to check what machine related options could be used.
But, they remineded me, if you don't use the default baseline setting of the toolchain, the behavior might
lead crash or some unknown problem.
I'll check it later.

However, you cannot generate v0 baseline with any other toolchains, only v0 toolchain can do.
Sorry. :-(

Best regards,
Macpaul Lin