Re: [Qemu-devel] [RFC PATCH] PIIX: reset the VM when the Reset Control Register's RCPU bit gets set

[Andreas Färber]

Hi,

Am 11.01.2013 10:30, schrieb Laszlo Ersek:
> On 01/09/13 22:01, Blue Swirl wrote:
>> On Tue, Jan 8, 2013 at 9:44 PM, Laszlo Ersek <address@hidden> wrote:
> 
>>> +static void i440fx_host_config_write(void *opaque, hwaddr addr,
>>> +                                     uint64_t val, unsigned len)
>>> +{
>>> +    if (addr == 1 && len == 1) {
>>> +        if (val & 4) {
>>> +            qemu_system_reset_request();
>>> +        }
>>> +        return;
>>> +    }
>>> +    pci_host_conf_le_ops.write(opaque, addr, val, len);
>>> +}
>>> +
>>> +static MemoryRegionOps i440fx_host_conf_ops = {
>>> +    .read       = NULL,
>>> +    .write      = i440fx_host_config_write,
>>> +    .endianness = DEVICE_LITTLE_ENDIAN
>>> +};
>>> +
>>>  static int i440fx_pcihost_initfn(SysBusDevice *dev)
>>>  {
>>>      PCIHostState *s = PCI_HOST_BRIDGE(dev);
>>>
>>> -    memory_region_init_io(&s->conf_mem, &pci_host_conf_le_ops, s,
>>> +    i440fx_host_conf_ops.read = pci_host_conf_le_ops.read;
>>
>> It would be cleaner to introduce a new memory region (without this
>> copying) which passes 0xcf8 and 0xcfc to standard PCI host but catches
>> accesses to 0xcf9. This may mean that pci_host_config_{read,write}
>> will need to be exposed.
> 
> Do you mean:
> 
> (1) introducing the new "i440fx_host_conf_ops" struct-of-funcptrs with
> detached functions (that is, duplicating the guts of
> pci_host_config_{read,write} and modifying them, and then registering
> s->conf_mem with this "i440fx_host_conf_ops"; or
> 
> (2) leaving s->conf_mem as-is, and introducing a sub-region just for
> port 0xcf9, with higher visibility priority?
> 
> (I don't feel confident about (2), and based on "docs/memory.txt" I
> thought that overlapping regions had not been invented for this purpose.)
> 
> IOW, are you OK with the explicit offset + access-width based check,
> just organized differently, or are you proposing a one-byte-wide subregion?

Another option:

(3) leaving s->conf_mem as-is but implementing your own read function as
well that forwards to pci_host_conf_le_ops.read() to avoid this unusual
non-const MemoryRegionOps construct

But I guess Blue meant (2), which should be slightly more performant.

Regards,
Andreas

-- 
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg