Re: [Qemu-devel] [RFC] hw/net: add DEC Tulip NIC emulation

[Stefan Hajnoczi]

On Fri, Nov 22, 2013 at 06:47:49PM +0400, Dmitry Smagin wrote:
> From: Antony Pavlov <address@hidden>
> 
> This patch adds emulation of DEC/Intel Tulip 21143 with some external chips:
>   * Intel LXT971A 10/100 Mbps PHY MII Tranceiver;
>   * Microchip 93LC46B 1K Microwire Compatible Serial EEPROM.
> 
> Restrictions and TODOs:
>   - Tulip always work in promisc-mode with no packet filtering
>     (TODO: check qemu packet filtering interfaces)
> 
>   - Address errors in packet descriptors are not checked
> 
>   - Tulip is bound to PCI, no other bus is possible for now
>     (TODO: Rewrite code to permit Tulip to live on buses other than PCI)
> 
>   - Internal transceiver is not emulated
>     The reason: internal transceiver works with 10 Mbps but we need 100Mbps
>     which only external transceivers can give.
> 
>   - Link status is not emulated, it's always ON
>     (TODO: use standard qemu interface)
> 
>   - Incorrect behavior of the Tulip driver is not checked and there's no error
>     signaling via register CSR5
> 
>   - Only transfer/receive (TI/RI) interrupts are emulated
> 
>   - Subsystem IDs in EEPROM image are not good
>     (TODO: find proper IDs)

The NIC needs to tell QEMU when it transitions from a rx descriptors
exhausted state to a rx refilled state.  For example, the e1000 NIC
calls qemu_flush_queued_packets(qemu_get_queue(s->nic)) when the RDT
hardware register is written to by the guest.

This is necessary since the netdev (e.g. tap) disables receive when the
NIC runs out of rx descriptors.

> +static const VMStateDescription vmstate_tulip = {
> +    .name = "tulip",
> +    .version_id = 2,
> +    .minimum_version_id = 1,
> +    .minimum_version_id_old = 1,
> +    .fields      = (VMStateField[]) {
> +        VMSTATE_PCI_DEVICE(dev, TulipState),
> +        VMSTATE_MACADDR(conf.macaddr, TulipState),

This looks a little sparse :).  Other hardware registers probably need
to be saved/loaded too.