Re: [Madwifi-devel] ioremap on IXP425

[Sam L. <sa...@er...>]

On Monday 01 March 2004 11:50 pm, Dale Whitfield wrote:
> Hi,
>
> Problem: oops in ath_hal.o
>
> =======================
> Unable to handle kernel paging request at virtual address 4bff000c
> pgd = c117c000
> [4bff000c] *pgd=00000000, *pmd = 00000000
> Internal error: Oops: f5
> CPU: 0
> pc : [<c2db3f00>]    lr : [<c2db3be0>]    Not tainted
> =======================
> pc is in ath_hal.o nowhere near any useful symbols
> =======================
> 00:01.0 Class 0200: 168c:0013 (rev 01)
>         Subsystem: 168c:2026
>         Flags: bus master, fast Back2Back, medium devsel, latency 168, IRQ
> 28 Memory at 4bff0000 (32-bit, non-prefetchable) [size=64K]
>         Capabilities: [44] Power Management version 2
> =======================
>
> It is my understanding that on IXP425 the cookie returned from ioremap
> should not be derefenced and treated as a virtual address. It appears that
> this is happening in ath_hal.o.
>
> Kernel is 2.4.21-rmk1-ds3 (with ARM and IXP425 patches).
> madwifi configured with xscale, big-endian.
>
> Tried building modules with AH_DEBUG and AH_REGOPS but the oops is in the
> binary portion and it didn't show up anything useful.
>
> How to proceed? Would a binary hal with REGOPS defined fix this?
>
> Some additional info - had to update Makefiles to add LDOPTS in order to
> pick up the big-endian (-EB) linker flag. Makefiles in ath_hal and wlan
> needed this.

The armv4-be-elf HAL build assumes the pointer you give it as the HAL_SOFTC 
parameter at attach time is suitable for direct reference.  If it is not then 
you need a version built with AH_REGOPS and you need to supply register 
access rouines to do the right thing.

If you can't provide a suitable pointer then you'll need to wait until someone 
has time to build a HAL to your specifications.

	Sam