Re: Freescale i.mx31 on NetBSD

[Chris Gilbert <chris%dokein.co.uk@localhost>]

Doug Brewer wrote:
> Hello,
>
> I'm trying to running imx31ads on netbsd with matt's imx31 source,
> but it seems to not work:
>
> RedBoot> go
>
> NetBSD/evbarm (imx31lk) booting ...
> initarm: Configuring system ...
> physmemory: 32768 pages at 0x80000000 -> 0x87ffffff
> Allocating page tables
> freestart = 0x800c0000, free_pages = 320 (0x00000140)
> IRQ stack: p0x801f1000 v0x801f1000
> ABT stack: p0x801f0000 v0x801f0000
> UND stack: p0x801ef000 v0x801ef000
> SVC stack: p0x801ed000 v0x801ed000
> Creating L1 page table at p0x801fc000 v0x801fc000
> Mapping kernel
> initarm: etext 803d959b, _end 80474244
> initarm: textsize 0x1e0000, totalsize 0x280000
> pmap_map_chunk: pa=0x80200000 va=0x80200000 size=0x1e0000 resid=0x1e0000 prot=01
> SLLLLLLLLLLLLLL
> pmap_map_chunk: pa=0x803e0000 va=0x803e0000 size=0xa0000 resid=0xa0000 prot=0x31
> LLLLLLLLLL
> Constructing L2 page tables
> pmap_map_chunk: pa=0x801f1000 va=0x801f1000 size=0x1000 resid=0x1000 prot=0x3 c1
> P
> pmap_map_chunk: pa=0x801f0000 va=0x801f0000 size=0x1000 resid=0x1000 prot=0x3 c1
> ...
> pmap_map_chunk: pa=0x801f6000 va=0x801f6000 size=0x1000 resid=0x1000 prot=0x3 c2
> P
> pmap_map_chunk: pa=0x801f5000 va=0x801f5000 size=0x1000 resid=0x1000 prot=0x3 c2
> P
> pmap_map_chunk: pa=0x801f4000 va=0x801f4000 size=0x1000 resid=0x1000 prot=0x3 c2
> P
> pmap_map_chunk: pa=0x801f3000 va=0x801f3000 size=0x1000 resid=0x1000 prot=0x3 c2
> P
> devmap: 43f00000 -> 43ffffff @ fd000000
> pmap_map_chunk: pa=0x43f00000 va=0xfd000000 size=0x100000 resid=0x100000 prot=00
> S
> freestart = 0x80475000, free_pages = 31627 (0x7b8b)
> switching to new L1 page table  @0x801fc000...
>
> I also noticed that imx uart has not implemented yet or did I miss something?
>
> Thank you,
> Doug.
>   

It's worth checking you're enabling domain access before hitting the new 
L1 table, eg from cats/cats_machdep.c
       /*
        * enable the system bit in the control register, otherwise we can't
        * access the kernel after the switch to the new L1 table
        * I suspect cyclone hid this problem, by enabling the ROM bit
        * Note can not have both SYST and ROM enabled together, the results
        * are "undefined"
        */
       cpu_control(CPU_CONTROL_SYST_ENABLE | CPU_CONTROL_ROM_ENABLE, 
CPU_CONTROL_SYST_ENABLE);
#ifdef VERBOSE_INIT_ARM
       printf("switching domains\n");
#endif
       /* be a client to all domains */
       cpu_domains(0x55555555);
       /* Switch tables */
#ifdef VERBOSE_INIT_ARM
       printf("switching to new L1 page table\n");
#endif

       setttb(kernel_l1pt.pv_pa);
       cpu_tlb_flushID();
       cpu_domains(DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2));

You may need to be a client of all domains, before the switch in case 
the bootloader is dropping you into a domain (although you're getting a 
printf, so probably isn't) but you definetly need to be a client of the 
kernel domain when you switch the L1 table.

Chris