Re: [Qemu-devel] Memory read/write issues

[Andreas Färber]

Am 27.11.2011 09:32, schrieb Avi Kivity:
> On 11/25/2011 06:25 PM, Andreas Färber wrote:
>> Am 24.11.2011 00:59, schrieb Andreas Färber:
>>> Unlike PowerPC, an architecture that I'm trying to emulate does not
>>> store branch instructions in the reset vector but a memory address. I'm
>>> therefore trying to read physical address 0x00000 and store its value
>>> into my env->pc.
>>>
>>> I've verified by running with -S that xp /xh 0x00000 shows the expected
>>> value.
>>>
>>> When doing lduw_phys(0x00000) or cpu_read_physical_memory() in the CPU
>>> reset function though, I just seem to read from uninitialized memory
>>> (0xbaba). I've taken care to reorder CPU initialization to after the
>>> BIOS file is loaded in the machine initialization function.
>>
>> Another weird memory issue is that tcg_gen_qemu_st16() succeeds but the
>> value stored doesn't show up with xp or x on the monitor but 0x0000.
>>
>> MOVW 0xf8,#0xfee0 (at 0x02010) is trying to write 0xfee0 to 0xFFFf8.
>>
>> Similarly, a subsequent tcg_gen_qemu_ld16u() reads 0x0000 from there.
>>
>> TCG snippet:
>> https://github.com/afaerber/qemu-rl78/commit/d3880bd53a26d224c56d16a6ea5950019d411cf0
>>
>>             uint8_t sfrp = ldub_code(s->pc + 1) & ~0x1;
>>             uint16_t data = lduw_code(s->pc + 2);
>>             LOG_DISAS("MOVW 0x%" PRIx8 ",#0x%04" PRIx16 "\n", sfrp, data);
>>             TCGv addr = tcg_const_tl(0xFFF00 | sfrp);
>>             TCGv_i32 val = tcg_const_i32(data);
>>             tcg_gen_qemu_st16(val, addr, 0);
>>
>> /* for testing: */
>>         tcg_gen_qemu_ld16u(env_sp, addr, 0);
>>
>>             tcg_temp_free(addr);
>>             tcg_temp_free(val);
>>
>> (qemu) info mtree
>> memory
>> 00000000-fffffffe (prio 0): system
>>   00000000-0000ffff (prio 0): rl78g13_pb.code_flash
>>   000fef00-000ffeff (prio 0): rl78g13_pb.ram
>>     000ffee0-000ffeff (prio 0): rl78g13_pb.gpr
>>   000fff00-000fffff (prio 0): rl78g13_pb.sfr
>>
>> https://github.com/afaerber/qemu-rl78/blob/d3880bd53a26d224c56d16a6ea5950019d411cf0/hw/rl78g13_pb.c#L88
>>
>>     memory_region_init_ram(sfr, NULL, "rl78g13_pb.sfr", 256);
>>     memory_region_add_subregion(get_system_memory(), 0xFFF00, sfr);
>>
>> Anything obvious that I'm missing?
> 
> This region is a little special in that it is a subpage RAM region, so
> it doesn't follow the normal paths where RAM is mapped directly to the
> guest.  Maybe there's some bug in that area.

Thanks a lot! You were right, setting TARGET_PAGE_BITS to 8 solves this
issue.

I'd still like to fix this subpage case for others' benefit. Do you have
any pointer where I should set breakpoints / review code?

>  If the target is big
> endian that could further complicate things.

Target and host are Little Endian, target is 20 bits using
TARGET_LONG_BITS 32.

Andreas