Re: [RFC 2/2] KVM: add initial support for ioregionfd blocking read/write operations

[Jason Wang <jasowang@redhat.com>]

On 2021/1/4 上午4:37, Elena Afanasova wrote:
> On Thu, 2020-12-31 at 11:46 +0800, Jason Wang wrote:
>> On 2020/12/29 下午6:02, Elena Afanasova wrote:
>>> Signed-off-by: Elena Afanasova<eafanasova@gmail.com>
>>> ---
>>>    virt/kvm/ioregion.c | 157
>>> ++++++++++++++++++++++++++++++++++++++++++++
>>>    1 file changed, 157 insertions(+)
>>>
>>> diff --git a/virt/kvm/ioregion.c b/virt/kvm/ioregion.c
>>> index a200c3761343..8523f4126337 100644
>>> --- a/virt/kvm/ioregion.c
>>> +++ b/virt/kvm/ioregion.c
>>> @@ -4,6 +4,33 @@
>>>    #include <kvm/iodev.h>
>>>    #include "eventfd.h"
>>>    
>>> +/* Wire protocol */
>>> +struct ioregionfd_cmd {
>>> +	__u32 info;
>>> +	__u32 padding;
>>> +	__u64 user_data;
>>> +	__u64 offset;
>>> +	__u64 data;
>>> +};
>>> +
>> I wonder do we need a seq in the protocol. It might be useful if we
>> allow a pair of file descriptors to be used for multiple different
>> ranges.
>>
> I think it might be helpful in the case of out-of-order requests.
> In the case of in order requests seq field seems not to be necessary
> since there will be cmds/replies serialization. I’ll include the
> synchronization code in a RFC v2 series.


See my reply to V1. It might be helpful for the case of using single 
ioregionfd for multiple ranges.

Thanks


>
>> Thanks
>>
>>
>>> +struct ioregionfd_resp {
>>> +	__u64 data;
>>> +	__u8 pad[24];
>>> +};