Re: Unable to build a BEAGLEBOARD kernel

[Matt Thomas <matt%3am-software.com@localhost>]

On Sep 17, 2012, at 3:11 AM, Pierre Pronchery wrote:

>                       Hi,
> 
> On Sun, 16 Sep 2012 21:47:34 +0000, Pierre Pronchery wrote:
>> 
>> I am currently unable to build an evbarm kernel for the Beagleboard; it
>> fails with the following error:
>> 
>> === BEGIN PASTE ===
>>    compile  BEAGLEBOARD/arm32_kvminit.o
>> cc1: warnings being treated as errors
>> /home/khorben/Projects/netbsd/src/sys/arch/arm/arm32/arm32_kvminit.c: In
>> function 'arm32_kernel_vm_init':
>> /home/khorben/Projects/netbsd/src/sys/arch/arm/arm32/
>> arm32_kvminit.c:359:1: error: not protecting local variables: variable
>> length buffer
>> 
>> *** Failed target:  arm32_kvminit.o
>> === END PASTE ===
> 
> Apparently the real error is here:
> 
> 401         pv_addr_t chunks[KERNEL_L2PT_KERNEL_NUM+KERNEL_L2PT_VMDATA_NUM
> +11];
> 
> KERNEL_L2PT_VMDATA_NUM is fine, since it is a constant value defined in 
> the same file. However KERNEL_L2PT_KERNEL_NUM is not constant:
> 
> 387         const size_t KERNEL_L2PT_KERNEL_NUM =
> 388             (kernel_size + L2_S_SEGSIZE - 1) / L2_S_SEGSIZE;
> 
> kernel_size being itself:
> 373         size_t kernel_size = bmi->bmi_kernelend;
> (and then additional run-time computation)
> 
> Any recommended way to allocate this instead? I have no idea what's 
> allowed at this stage.


There is no way to allocate things at this stage.  If SSP can't deal
with standard C99 usages, then SSP is broken.