Re: [PATCH net-next v3 1/3] selftests/ptr_ring: add benchmark application for ptr_ring

[Michael S. Tsirkin]

On Fri, Jul 02, 2021 at 05:04:44PM +0800, Jason Wang wrote:
> 
> 在 2021/7/2 下午4:46, Yunsheng Lin 写道:
> > On 2021/7/2 16:30, Michael S. Tsirkin wrote:
> > > On Fri, Jul 02, 2021 at 04:17:17PM +0800, Yunsheng Lin wrote:
> > > > > Let's reuse ptr_ring.c in tools/virtio/ringtest. Nothing virt specific there.
> > > > It *does* have some virtio specific at the end of ptr_ring.c.
> 
> 
> They are just wrappers to make ptr ring works like a virtio ring. We can
> split them out into another file if necessary.
> 
> 
> > > > It can be argued that the ptr_ring.c in tools/virtio/ringtest
> > > > could be refactored to remove the function related to virtio.
> > > > 
> > > > But as mentioned in the previous disscusion [1], the tools/virtio/
> > > > seems to have compile error in the latest kernel, it does not seems
> > > > right to reuse that.
> > > > And most of testcase in tools/virtio/ seems
> > > > better be in tools/virtio/ringtest instead，so until the testcase
> > > > in tools/virtio/ is compile-error-free and moved to tools/testing/
> > > > selftests/, it seems better not to reuse it for now.
> > > 
> > > That's a great reason to reuse - so tools/virtio/ stays working.
> > > Please just fix that.
> 
> 
> +1
> 
> 
> > I understand that you guys like to see a working testcase of virtio.
> > I would love to do that if I have the time and knowledge of virtio,
> > But I do not think I have the time and I am familiar enough with
> > virtio to fix that now.
> 
> 
> So ringtest is used for bench-marking the ring performance for different
> format. Virtio is only one of the supported ring format, ptr ring is
> another. Wrappers were used to reuse the same test logic.
> 
> Though you may see host/guest in the test, it's in fact done via two
> processes.
> 
> We need figure out:
> 
> 1) why the current ringtest.c does not fit for your requirement (it has SPSC
> test)
> 2) why can't we tweak the ptr_ring.c to be used by both ring_test and your
> benchmark
> 
> If neither of the above work, we can invent new ptr_ring infrastructure
> under tests/
> 
> Thanks

For me 1) is not a question.

All the available/used terminology is not an ideal fit for ptr ring.
With virtio buffers are always owned by driver (producer) so producer
has a way to find out if a buffer has been consumed.  With ptr ring
there's no way for producer to know a buffer has been consumed.
The test hacks around that but it is very reasonable
not to want to rely on that.

However 2) is very much a question. We can split ptr_ring
to the preamble and virtio related hacks.
So all the portability infrastructure for building
kernel code from userspace, command line parsing,
run-on-all.sh to figure out affinity effects,
all that can and should IMHO be reused and not copy-pasted.

> 
> > 
> >