On Mon 11.Jan'21 at 13:52:19 -0800, Davidlohr Bueso wrote:
On Wed, 06 Jan 2021, shuo.a....@intel.com wrote:
The processing flow of I/O requests are listed as following:
a) The I/O handler of the hypervisor will fill an I/O request with
PENDING state when a trapped I/O access happens in a User VM.
b) The hypervisor makes an upcall, which is a notification interrupt, to
the Service VM.
c) The upcall handler schedules a tasklet to dispatch I/O requests.
d) The tasklet looks for the PENDING I/O requests, assigns them to
different registered clients based on the address of the I/O accesses,
updates their state to PROCESSING, and notifies the corresponding
client to handle.
Hmm so tasklets are deprecated (and have been for a while) and it's sad
to see incoming new users in modern Linux. This wouldn't be the first one,
however. We should be _removing_ users, not adding... In addition, this
expands the whole tasklet_disable/enable() hacks.
Sorry, i had not noticed that.
Could this not be done in process context instead?
It could be. The original consideration with tasklet was more about
performance as the I/O requests dispatching is a hot code path. I think
irq thread has little performance impact? I can have a try to convert
the tasklet to irq thread.
Thanks
shuo
