Re: Adding IO memory region to mipssim

[Philippe Mathieu-Daudé]

On 9/10/21 3:21 PM, Hinko Kocevar wrote:
> I'm trying to add an I/O memory region to mipssim machine to emulate a
> MMIO region used by the u-boot loaded as BIOS image. I can confirm that
> the machine starts and loads the BIOS, starts execution but hangs due to
> unhandled IO access as described below.
> 
> The region should be at 0xB8810000, of size 0x10000.
> 
> I've added these lines of code to mispsim.c mips_mipssim_init():
> 
>     my_state *s = g_malloc0(sizeof(my_state));
>     memory_region_init_io(&s->mmio, NULL, &my_ops, s,
>                          "mips_mipssim.foo", 0x10000);
>     memory_region_add_subregion(address_space_mem, 0xB8810000LL, &s->mmio);

You need to map your device at its physical address, not the virtual
one.

> All goes well, the machine starts, and I can see the newly added region
> in qemu monitor info mtree output like so:
> 
>     00000000b8810000-00000000b881ffff (prio 0, i/o): mips_mipssim.foo
> 
> With some tracing enabled I see this error:
> 
>  Invalid access at addr 0x18810104, size 4, region '(null)', reason:
> rejected
> 
> I know the u-boot is making request to 0xB8810104 and not 0x18810104. I

U-boot accessed the virtual address which is resolved into the physical
one (where your device should be mapped).

> also can see 0xB8810104 address being handed to io_writex(), but
> mr_offset becomes 0x18810104 here:
> 
>   mr_offset = (iotlbentry->addr & TARGET_PAGE_MASK) + addr;
> 
> What is going on?
> 
> FWIW, I can place my emulated memory region at 0x18810104, but would
> like to understand the behavior above.

Yes, this is the correct address to map it.

Maybe this helps:
https://training.mips.com/basic_mips/PDF/Memory_Map.pdf

Regards,

Phil.