Re: [PATCH] virtio_ring: switch to dma_XX barriers for rpmsg

[Bjorn Andersson]

On Thu 19 Apr 10:48 PDT 2018, Paolo Bonzini wrote:

> On 19/04/2018 19:46, Michael S. Tsirkin wrote:
> >> This should be okay, but I wonder if there should be a virtio_wmb(...)
> >> or an "if (weak_barriers) wmb()" before the "writel" in vm_notify
> >> (drivers/virtio/virtio_mmio.c).
> >>
> >> Thanks,
> >>
> >> Paolo
> > That one uses weak barriers AFAIK.
> > 
> > IIUC you mean rproc_virtio_notify.
> > 
> > I suspect it works because specific kick callbacks have a barrier internally.
> 
> Yes, that one.  At least keystone_rproc_kick doesn't seem to have a barrier.
> 

Afaict you're correct. My expectation is that the kick ensures write
ordering internally and if I read this correct keystone_rproc_kick()
results in a writel_relaxed() in the gpio driver.

@Suman, can you please have a look at this.

Thanks Paolo,
Bjorn