Rao Shoaib wrote:
James Carlson wrote:
Rao Shoaib writes:
Yes it's a driver bug. I am not sure why this was not pursued. This
must not be the case anymore or else sendfile would not work.
Plain old TCP retransmit would be clobbered by such a thing. If it
happens anywhere, it seems like a pretty substantial bug to me.
Jim I would have to dig through my emails to find the reason. I coded
NFS to use desballoc but Peter Staubach pointed out that he had done
this in the past (he is the one who proposed desballoc) and ran into
this NIC bug and hence I was not allowed to putback.
I found some of the emails and Jim is correct. The issue was not
corruption as it was resolved by the use of desballoca. The issue was
when the buffer is freed. There are drivers that do lazy free i.e. they
do not free the buffer right after it has been transmitted. This delay
causes problems with other filesystem operations. Other siginificant
delay in releasing the buffer can be due to tcp retransmission.
Rao.
Rao.
_______________________________________________
networking-discuss mailing list
[email protected]
_______________________________________________
networking-discuss mailing list
[email protected]
