Re: [Qemu-devel] subpages with memory region aliases

[Jan Kiszka <jan.kiszka@web.de>]

[-- Attachment #1: Type: text/plain, Size: 1731 bytes --]

On 2012-02-09 10:32, Avi Kivity wrote:
> On 02/09/2012 10:35 AM, Jan Kiszka wrote:
>> Avi,
>>
>> Before I forget:
>>
>> On 2012-02-05 13:39, Jan Kiszka wrote:
>>> +static void vapic_map_rom_writable(VAPICROMState *s)
>>> +{
>>> +    target_phys_addr_t rom_paddr = s->rom_state_paddr & ROM_BLOCK_MASK;
>>> +    MemoryRegionSection section;
>>> +    MemoryRegion *as;
>>> +    size_t rom_size;
>>> +    uint8_t *ram;
>>> +
>>> +    as = sysbus_address_space(&s->busdev);
>>> +
>>> +    if (s->rom_mapped_writable) {
>>> +        memory_region_del_subregion(as, &s->rom);
>>> +        memory_region_destroy(&s->rom);
>>> +    }
>>> +
>>> +    /* grab RAM memory region (region @rom_paddr may still be pc.rom) */
>>> +    section = memory_region_find(as, 0, 1);
>>> +
>>> +    /* read ROM size from RAM region */
>>> +    ram = memory_region_get_ram_ptr(section.mr);
>>> +    rom_size = ram[rom_paddr + 2] * ROM_BLOCK_SIZE;
>>> +    s->rom_size = rom_size;
>>> +
>>> +    /* FIXME: round up as everything underneath would fall apart otherwise
>>> +     * (subpages are broken) */
>>> +    rom_size = TARGET_PAGE_ALIGN(rom_size);
>>
>> Removing this alignment triggers an interesting bug in the memory layer.
>> Haven't understood the details yet. Is subpage support supposed to work?
>>
>>
> 
> There are plenty of restrictions wrt page size in the memory API.  Some
> will be removed, some are enforced by hardware (for example the low bits
> of the guest address and host address must match).
> 
> Subpage of course works, but it can't be direct mapped.
> 

Does this mean you can't use them with RAM backing? Should probably be
documented and caught gracefully in that case.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]