Re: [Qemu-devel] [PATCH 0/2] ignore bus master for e1000

[Jason Wang <jasowang@redhat.com>]

On Thu, Dec 18, 2014 at 7:11 PM, Denis V. Lunev 
<den-lists@parallels.com> wrote:
> On 18/12/14 14:01, Denis V. Lunev wrote:
>> On 18/12/14 13:05, Jason Wang wrote:
>>> 
>>> 
>>> On Thu, Dec 18, 2014 at 5:22 PM, Amos Kong <akong@redhat.com> wrote:
>>>> Win2012-64r2 guest doesn't set bus mastering correctly,
>>>> it caused guest network down, this patch ignored it for
>>>> e1000 nic for workarounding the guest issue.
>>>> 
>>>> Patch 1 is an update version of:
>>>> http://lists.gnu.org/archive/html/qemu-devel/2014-12/msg00048.html
>>>> transmit packets are also defered in latest version.
>>>> 
>>>> Amos Kong (1):
>>>>   e1000: unconditionally enable bus mastering
>>>> 
>>>> Michael S. Tsirkin (1):
>>>>   e1000: defer packets until BM enabled
>>>> 
>>>>  hw/net/e1000.c | 34 +++++++++++++++++++++++++++++++++-
>>>>  1 file changed, 33 insertions(+), 1 deletion(-)
>>>> 
>>>> --
>>>> 2.1.0
>>> 
>>> Another question unrelated to this series:
>>> Does 82540EM still supported by Win2k12?
>>> The following link shows that 82540EM were not supported in Win2k12.
>>> https://downloadcenter.intel.com/SearchResult.aspx?lang=eng&ProdId=983
>>> 
>>> The oldest 825x card supported by 2k12 seems 82574.
>>> https://downloadcenter.intel.com/Detail_Desc.aspx?DwnldID=21694&lang=eng&ProdId=3189
>>> 
>>> 
>>> 
>>> If yes, workaround a buggy driver may be endless work in the future.
>>> 
>>> 
>> 
>> w2k12r2 datacenter edition founds QEMU e1000 card just fine
>> (checked by observation, very basic browser test).
>> 
>> Though I know some troubles with Parallels emulation of 82574
>> in w2k12 on driver size. The driver is really old and not
>> supported.
>> 
>> Thus from my opinion switching to this device might not be
>> a good choice.
> 
> correction. Mistaken a bit. Sorry for this.
> 
> 1) 82540EM and 82545EM are at least working. There are some
>    troubles in driver under the load. Driver is the same.

If you mean the troubles happen on real hardware.
We'd really not encourage the user to use e1000 in 2k12 guest.
> 
> 2) 82574 is known to be slower in Parallels/VMware.