On 1/14/2016 4:21 PM, Yuanhan Liu wrote:
> On Thu, Jan 14, 2016 at 08:08:28AM +0000, Xie, Huawei wrote:
>> On 1/14/2016 3:58 PM, Yuanhan Liu wrote:
>>> On Thu, Jan 14, 2016 at 07:51:08AM +0000, Xie, Huawei wrote:
>>>> On 1/14/2016 3:49 PM, Yuanhan Liu wrote:
>>>>> On Thu, Jan 14, 2016 at 07:47:17AM +0000, Xie, Huawei wrote:
>>>>>> On 1/12/2016 2:58 PM, Yuanhan Liu wrote:
>>>>>>> Modern (v1.0) virtio pci device defines several pci capabilities.
>>>>>> [snip]
>>>>>>> +static void
>>>>>>> +modern_notify_queue(struct virtio_hw *hw __rte_unused, struct
>>>>>>> virtqueue *vq)
>>>>>>> +{
>>>>>>> + modern_write16(1, vq->notify_addr);
>>>>>>> +}
>>>>>> Does virtio 1.0 only supports MMIO? MMIO has long VMEXIT latency than
>>>>>> PORT IO.
>>>>> Virtio 1.0 supports three transport layer, including MMIO and PCI. And
>>>>> we use PCI only in our pmd driver.
>>>> I don't mean that MMIO but use memory mapped IO for configuration.
>>> Then, yes.
>> 00:03.0 Ethernet controller: Red Hat, Inc Virtio network device
>> Subsystem: Red Hat, Inc Device 0001
>> Physical Slot: 3
>> Flags: bus master, fast devsel, latency 0, IRQ 10
>> I/O ports at c100 [size=32]
>> Memory at febd1000 (32-bit, non-prefetchable) [size=4K]
>> Memory at fe000000 (64-bit, prefetchable) [size=8M]
>> Expansion ROM at feb40000 [disabled] [size=256K]
>> Capabilities: [98] MSI-X: Enable+ Count=3 Masked-
>> Capabilities: [84] Vendor Specific Information: Len=14 <?>
>> Capabilities: [70] Vendor Specific Information: Len=14 <?>
>> Capabilities: [60] Vendor Specific Information: Len=10 <?>
>> Capabilities: [50] Vendor Specific Information: Len=10 <?>
>> Capabilities: [40] Vendor Specific Information: Len=10 <?>
>> Kernel driver in use: igb_uio
>> Kernel modules: virtio_pci
>>
>> c100 is still there.
> Yes,
>
>> For the notification, try PORT IO if possible.
> But it doesn't seem right to me to mix legacy registers in modern pci device.
On TLB and cache miss, this could cause plenty of cycles. Considering
that our current focus is dpdkvhost which doesn't need notification, let
us revisit this later.
>
> --yliu
>
