[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH] xen-netfront: correct MAX_TX_TARGET calculation.

On 01/27/12 11:36, Wei Liu wrote:
On Thu, 2012-01-26 at 18:19 +0000, Konrad Rzeszutek Wilk wrote:
On Thu, Jan 26, 2012 at 05:23:23PM +0000, Wei Liu wrote:

Can you give some more details please? What is the impact of
not having this? Should it be backported to stable?

I think it will not cause crash, only the scratch space size is
affected, thus impacting tx batching a bit.

As the tx structure is bigger than rx structure. I think scratch space
size is likely to shrink after correction.

It also seems to affect the netfront_tx_slot_available() function, making it stricter (likely). Before the patch, the function may have reported available slots when there were none, causing spurious(?) queue wakeups in xennet_maybe_wake_tx(), and not stopping the queue in xennet_start_xmit() when it should have(?).

It seems there are no further uses of TX_MAX_TARGET, and for bounds checking NET_TX_RING_SIZE was used (which was always correct). So I guess the typo may have caused some performance degradation.

I can't either prove or disprove a DoS-like busy loop in the pre-patch form.


Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.