Re: [Minios-devel] [UNIKRAFT PATCH 1/1] plat/xen: Add flush for xen console

[Julien Grall <julien.grall@xxxxxxx>]

Hi,

You want to add a commit message here explaining why the flush. You 
cover letter seems a good candidate for that.
On 10/25/18 5:05 PM, Birlea Costin wrote:
> Signed-off-by: Birlea Costin <costin.birlea@xxxxxxxxx>
> ---
>   plat/xen/console.c                | 14 ++++++++++++++
>   plat/xen/include/common/console.h |  3 +++
>   plat/xen/shutdown.c               |  3 +++
>   3 files changed, 20 insertions(+)
>
> diff --git a/plat/xen/console.c b/plat/xen/console.c
> index c0b35c7..3df06d4 100644
> --- a/plat/xen/console.c
> +++ b/plat/xen/console.c
> @@ -203,6 +203,20 @@ static void hvconsole_input(evtchn_port_t port __unused,
>         /* NOT IMPLEMENTED YET */
>   }
>   
> +void hvconsole_flush()
> +{
> +       struct xencons_interface *intf;
> +
> +       if (!console_ready)
> +               return;
> +
> +       intf = console_ring;
> +       if (unlikely(!intf))
> +               return ;
> +
> +       while (intf->out_cons < intf->out_prod)
The compiler is free to assume that the value of out_const and out_prod 
will never change. So the following optimization is valid:
a = intf->out_cons;
b = intf->out_prod;
while (a  < b);

And will result to potentially an infinite loop. You can request the 
compiler to always read out_cons and out_prod by adding a barrier in the 
loop. I.e:
while (intf->out_cons < int->out_prod)
{
    barrier();
}

Cheers,

--
Julien Grall

_______________________________________________
Minios-devel mailing list
Minios-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/minios-devel