Re: [Qemu-devel] [RFC v3 PATCH 19/45] multi-process: Add LSI device proxy object

[Jag Raman <jag.raman@oracle.com>]

On 9/5/2019 6:22 AM, Gerd Hoffmann wrote:
>    Hi,
> 
>> +static uint64_t proxy_lsi_io_read(void *opaque, hwaddr addr, unsigned size)
>> +{
>> +    ProxyLSIState *s = opaque;
>> +
>> +    return proxy_default_bar_read(PCI_PROXY_DEV(s), &s->io_io, addr, size,
>> +                                  false);
>> +}
>> +
>> +static void proxy_lsi_io_write(void *opaque, hwaddr addr, uint64_t val,
>> +                               unsigned size)
>> +{
>> +    ProxyLSIState *s = opaque;
>> +
>> +    proxy_default_bar_write(PCI_PROXY_DEV(s), &s->io_io, addr, val, size,
>> +                            false);
>> +}
>> +
>> +static const MemoryRegionOps proxy_lsi_io_ops = {
>> +    .read = proxy_lsi_io_read,
>> +    .write = proxy_lsi_io_write,
>> +    .endianness = DEVICE_NATIVE_ENDIAN,
>> +    .impl = {
>> +        .min_access_size = 1,
>> +        .max_access_size = 1,
>> +    },
>> +};
> 
> Hmm, as more devices get proxy support there will be alot of simliar
> boilerplate.
> 
> I think it would be useful to have a
> 
> struct pci_device_description {
> 	u16 vendor_id,
> 	u16 device_id,
> 	[ ... ]
> 	struct {
> 		[ ... ]
> 	} bar[6];
> };
> 
>> +    proxy_class->realize = proxy_lsi_realize;
>> +    proxy_class->command = g_strdup("qemu-scsi-dev");
> 
> Hook that up here (proxy_class->description = &lsi_description),
> then have the pci proxy realize function setup everything.  All pci
> bar access is just forwarded to the proxy, that should be doable
> without duplicating the code for each proxied pci device ...
> 
> At least parts of the pci_device_description can probably also used for
> non-proxy device setup (should work for pci config space, probably
> would not work for memory regions as they are very device specific).

Sounds great. We'll move similar boilerplate code to the pci proxy
class in the next rev.

Thanks!
--
Jag

> 
> cheers,
>    Gerd
> 
>