Hostap For Rc12 Anyone?

[pgas]

I've downloaded the pdaXrom-builder tarball, but I need to get to a faster pipe to download the sources and the pdaX86 image.

You don't need to download everything, but I' m not sure exactly what you'll need

Hmm, according to the docs I need that pdaX86 live image as the host system to use pdaXrom-builder. Is that true? I have a couple of Linux systems here (mostly SuSE 9.3), couldn't I use one of these as the build host instead?

I think you could use your buildhost. When I tried the builder, i had extracted the fs of the iso and build in a chrooted pdaxromx86. The advantage of the iso is that you have what you need ready

Anyway, that looks like an awful lot of extra work just to compile a few kernel modules. sad.gif

Yes, though I think you only need to build the toolchain (perhaps the two toolchains 3.3 and 2,95, because the kernel requires 2.95 I'm not sure if it is still the case) and then the kernel. You don't need to build say gtk.

[agraef]

I think you could use your buildhost. When I tried the builder, i had extracted the fs of the iso and build in a chrooted pdaxromx86. The advantage of the iso is that you have what you need ready

Yup, excellent idea, I'll try that. And thanks for the quick reply.

[kopsis]

If you're just building modules, you may not actually need the pdaXrom version of the kernel to build against. I've built the uinput modules for pdaXrom using just the stock Sharp kernel source for the C1000. If you don't want to thrash one of your existing boxen, just grab my Sharp ROM development DSL image and run it in QEMU or via a DSL LiveCD. That image has all the right toolchain binaries for kernel building pre-installed and ready to go. Still a sizeable download, but smaller (and faster) than building the whole toolchain.

[agraef]

Yes, though I think you only need to build the toolchain (perhaps the two toolchains 3.3 and 2,95, because the kernel requires 2.95 I'm not sure if it is still the case) and then the kernel. You don't need to build say gtk.

Looks like the 2.95.3 toolchain is still required to build the kernel, at least it's set up in the Akita config file that way.

Ok, here is how far I got. I set up a chrooted environment for pdaX86 and used the RC12 builder to build the 3.4.4 toolchain. That went smoothly.

But gcc 2.95.3 wouldn't compile with the RC12 builder -- whatever I tried, the compile would always end with an unrecognized -fno-unit-at-a-time option. If I removed that option, the compile would proceed but eventually abort with a bunch of linker error messages.

So I built gcc 2.95.3 using the RC10 builder (couldn't find an RC11 builder anywhere). That worked. But now when I try to use the RC12 builder with that gcc 2.95.3 version to build the kernel, it always bails out with an asm error (unrecognized -miwmmxt option):

Code: [Select]

----------------------
target: kernel.compile
----------------------

PATH="/opt/cross/arm/2.95.3/bin":$PATH make -C /root/pdaXrom-builder/build/"linux" HOSTCC=gcc GENKSYMS="armv5tel-cacko-linux"-genksyms DEPMOD=true ARCH=arm CROSS_COMPILE=arm-linux- \
        zImage modules
make[1]: Entering directory `/root/pdaXrom-builder/build/linux'
gcc -Wall -Wstrict-prototypes -O2 -fomit-frame-pointer -o scripts/split-include scripts/split-include.c
scripts/split-include include/linux/autoconf.h include/config
arm-linux-gcc -D__KERNEL__ -I/root/pdaXrom-builder/build/linux/include -Wall -Wstrict-prototypes -Wno-trigraphs -O2 -fno-strict-aliasing -fno-common -Uarm -fno-common -pipe -mapcs-32 -D__LINUX_ARM_ARCH__=4 -march=armv4 -Wa,-miwmmxt -mtune=strongarm -mshort-load-bytes -msoft-float -Uarm   -DKBUILD_BASENAME=main -c -o init/main.o init/main.c
Assembler messages:
Error: Invalid processor variant -miwmmxt
In file included from /root/pdaXrom-builder/build/linux/include/asm/pgalloc.h:23,
                 from /root/pdaXrom-builder/build/linux/include/linux/highmem.h:5,
                 from /root/pdaXrom-builder/build/linux/include/linux/pagemap.h:16,
                 from /root/pdaXrom-builder/build/linux/include/linux/locks.h:8,
                 from /root/pdaXrom-builder/build/linux/include/linux/blk.h:5,
                 from init/main.c:25:
/root/pdaXrom-builder/build/linux/include/asm/proc/cache.h: In function `flush_page_to_ram':
/root/pdaXrom-builder/build/linux/include/asm/proc/cache.h:118: warning: passing arg 1 of `cpu_pid_va_to_mva' makes integer from pointer without a cast
/root/pdaXrom-builder/build/linux/include/asm/proc/cache.h:118: warning: passing arg 1 of `cpu_xscale_flush_ram_page' makes pointer from integer without a cast
/root/pdaXrom-builder/build/linux/include/asm/proc/cache.h: In function `flush_dcache_page':
/root/pdaXrom-builder/build/linux/include/asm/proc/cache.h:159: warning: passing arg 1 of `cpu_pid_va_to_mva' makes integer from pointer without a cast
In file included from /root/pdaXrom-builder/build/linux/include/linux/highmem.h:5,
                 from /root/pdaXrom-builder/build/linux/include/linux/pagemap.h:16,
                 from /root/pdaXrom-builder/build/linux/include/linux/locks.h:8,
                 from /root/pdaXrom-builder/build/linux/include/linux/blk.h:5,
                 from init/main.c:25:
/root/pdaXrom-builder/build/linux/include/asm/pgalloc.h: In function `get_pgd_fast':
/root/pdaXrom-builder/build/linux/include/asm/pgalloc.h:67: warning: passing arg 1 of `cpu_pid_va_to_mva' makes integer from pointer without a cast
/root/pdaXrom-builder/build/linux/include/asm/pgalloc.h: In function `pte_alloc_one_fast':
/root/pdaXrom-builder/build/linux/include/asm/pgalloc.h:91: warning: passing arg 1 of `cpu_pid_va_to_mva' makes integer from pointer without a cast
init/main.c: In function `debug_kernel':
init/main.c:228: output pipe has been closed
make[1]: *** [init/main.o] Error 1
make[1]: Leaving directory `/root/pdaXrom-builder/build/linux'
make: *** [/root/pdaXrom-builder/state/kernel.compile] Error 2
Sure enough, the asm I built with the RC10 builder doesn't understand that option.

So what now? Can someone please post a tarball with the right 2.95.3 cross toolchain to make the RC12 kernel build go through? (I'd presume that at least the pdaXrom devs should have this.) Or should I just get rid of that -Wa,-miwmmxt compiler option? But then the generated code will be slower, no?

UPDATE: Just replaced that -Wa,-miwmmxt in build/linux/arch/arm/Makefile with -Wa,-mxscale, and now the compile dies at entry-armv.S:

Code: [Select]

entry-armv.S: Assembler messages:
entry-armv.S:980: Error: bad instruction `tmrc r4,wCon'
entry-armv.S:980: Error: bad instruction `wstrw wCSSF,[r5,#(0x00)]'
entry-armv.S:980: Error: bad instruction `wstrw wCASF,[r5,#(0x04)]'
...
I'm stuck now.

[agraef]

If you're just building modules, you may not actually need the pdaXrom version of the kernel to build against. I've built the uinput modules for pdaXrom using just the stock Sharp kernel source for the C1000. If you don't want to thrash one of your existing boxen, just grab my Sharp ROM development DSL image and run it in QEMU or via a DSL LiveCD. That image has all the right toolchain binaries for kernel building pre-installed and ready to go. Still a sizeable download, but smaller (and faster) than building the whole toolchain.
[div align=\"right\"][a href=\"index.php?act=findpost&pid=100781\"][{POST_SNAPBACK}][/a][/div]

Dave, I tried your DSL image in qemu and the sample qpe application compiles. But I didn't have any success compiling the C1K kernel (linux-c1000-20050228-rom1_01.tar.bz2) with that setup so far. For one thing, libncurses seems to be missing, so make menuconfig doesn't work. Then if I just copy the arch/arm/def-configs/akita-j config to .config and run make dep, I see error messages from find, and make zImage then fails with some mysterious broken dependencies. The culprit apparently is that broken find command from Busybox, so a real GNU find would be needed but that doesn't seem to be available in your image.

Do you have a working setup which allows you to compile the C1K kernel? Then it would be very kind if you could just zip up that disk image and upload it to your website.

Thanks,
Albert

[agraef]

All right, with the latest RC12 pdaXrom-builder released today I'm now able to build the 2.95.3 toolchain and the Akita kernel, very nice.

But the wlan modules (hermes, orinoco etc.) are still missing in this release for the Akita although they get built during make modules all right. It seems that they just don't get installed during make modules_install or maybe they are installed in the wrong place so that they're missed when the kernel-modules ipk is built. Will have to investigate that.

UPDATE: Well, the modules are not in the kernel-modules package quite simply because they are not built as modules, but compiled into the kernel. That was easy to fix. Now I just need to compile the hostap stuff...

[agraef]

Ok, hostap driver 0.3.9 compiled fine, and my D-Link DCF-660W works great now, no more hangups. I've attached a tarball with the binaries of the kernel modules to this post for those who need them. Might also work on RC11, but I've only tested them on RC12 so far.

Thanks to all who helped me get this up and running!

EDIT: Original tarball had files with wrong user/group info, that is fixed now.

[pgas]

I'm glad to see that you finally manage to compile hostap. well done!

[agraef]

I'm glad to see that you finally manage to compile hostap. well done!
[div align=\"right\"][a href=\"index.php?act=findpost&pid=101328\"][{POST_SNAPBACK}][/a][/div]

Thanks. I think I'm starting to get the hang of it ... The latest builder really works very well. My earlier compile probs stemmed from not being able to build the right 2.95 toolchain with the builder from the RC12-test release. This has all been resolved with the final RC12.

[gryning]

trying to use this hostap driver with Planex GW-CF11X i get the following.
Anyone able to help with this?

hostap_cs: 0.3.9 - 2005-06-10 (Jouni Malinen <jkmaline@cc.hut.fi>)
hostap_cs: setting Vcc=33 (constant)
hostap_cs: CS_EVENT_CARD_INSERTION
hostap_cs: setting Vcc=33 (from config)
Checking CFTABLE_ENTRY 0x01 (default 0x01)
IO window settings: cfg->io.nwin=1 dflt.io.nwin=1
io->flags = 0x0047, io.base=0x0000, len=128
hostap_cs: Registered netdevice wifi0
hostap_cs: index 0x01: Vcc 3.3, irq 135, io 0xf6000000-0xf600007f
wifi0: __hfa384x_cmd_no_wait(6) - timeout - reg=0xf072
hostap_cs: first command failed - assuming card does not have primary firmware
hostap_cs: assuming no Primary image in flash - card initialization not completed
wifi0: test Genesis mode with HCR 0x1f
prism2_pccard_cor_sreset: original COR 41
prism2_pccard_genesis_sreset: original COR 41
Readback test failed, HCR 0x1f write 00 e1 a1 ff read 00 ce a1 ce
wifi0: test Genesis mode with HCR 0x0f
prism2_pccard_cor_sreset: original COR 41
prism2_pccard_genesis_sreset: original COR 41
Readback test succeeded, HCR 0x0f
prism2_pccard_genesis_sreset: original COR 41
wifi0: registered netdevice wlan0
prism2: wlan0: operating mode changed 3 -> 2
wlan0: cannot set RID fc00 (len=2) - no PRI f/w
wlan0: cannot set RID fc0e (len=34) - no PRI f/w
wifi0: cannot get RID fc28 (len=2) - no PRI f/w
Could not read current WEP flags.
wifi0: encryption setup failed
wlan0: set_encryption failed
wlan0: cannot set RID fc02 (len=34) - no PRI f/w
wlan0: could not set interface UP - no PRI f/w
wlan0: could not set interface UP - no PRI f/w

[magicbb]

I'm glad to see that you finally manage to compile hostap. well done!
[div align=\"right\"][a href=\"index.php?act=findpost&pid=101328\"][{POST_SNAPBACK}][/a][/div]

Thanks. I think I'm starting to get the hang of it ... The latest builder really works very well. My earlier compile probs stemmed from not being able to build the right 2.95 toolchain with the builder from the RC12-test release. This has all been resolved with the final RC12.
[div align=\"right\"][a href=\"index.php?act=findpost&pid=101330\"][{POST_SNAPBACK}][/a][/div]

thank you so much for compiling this to us

[agraef]

...
hostap_cs: first command failed - assuming card does not have primary firmware

hostap_cs: assuming no Primary image in flash - card initialization not completed
...
[div align=\"right\"][{POST_SNAPBACK}][/a][/div]

Well, obviously the card is recognized, that's good. However, it looks like this card needs a firmware, so you probably have to install hostap-utils in addition to the driver (my tarball only contains the driver itself, as this is all I needed). The hostap-utils should be straightforward to build with the native or cross devkit, I think you don't need the kernel sources for that. You also need to figure out which firmware you need, and maybe you want to write a little hotplug script to automate the firmware loading process. See this thread: "[a href=\"https://www.oesf.org/forums/index.php?showtopic=12567]Sandisk Connect Plus Wi-fi + 128mb[/url]" for more information (that thread is about the SanDisk Connect plus, so you'll have to adapt this for your card, of course). Taking a look at the hostap* and prism* packages from OpenZaurus might also help to figure out how to get the card to work. It seems that OZ has much better out-of-the-box support for different wlan cards right now.

If that sounds like too much work, then it's probably easiest to buy another card which has the firmware on board.

HTH,
Albert

[gryning]

ok...
got planex gw-cf11x working with this driver and details from above post.

will write up and post what i did

[Laze]

I "ordered" sash to look into this and we will probally be using all the fixes and solutions from the cacko rom.

[agraef]

Laze, that's good news, would really be nice if the newer wlan cards would work out of the box on pdaXrom.

gryning, it would still be nice if you could describe the steps needed to make the firmware loader work. I might give it a try myself (still have a Sandisk Connect Plus lying around here), and this information would probably be useful for the pdaXrom developers, too.