Re: [Xen-devel] [PATCH v2] xenfb: remove xen_init_display "temporary" hack

2017-07-06 Thread Stefano Stabellini 
On Thu, 29 Jun 2017, Paul Durrant wrote:
> > -Original Message-
> > From: Stefano Stabellini [mailto:sstabell...@kernel.org]
> > Sent: 28 June 2017 19:37
> > To: xen-de...@lists.xenproject.org; qemu-de...@nongnu.org
> > Cc: sstabell...@kernel.org; peter.mayd...@linaro.org; Anthony Perard
> > ; kra...@redhat.com; Paul Durrant
> > 
> > Subject: [PATCH v2] xenfb: remove xen_init_display "temporary" hack
> > 
> > Initialize xenfb properly, as all other backends, from its own
> > "initialise" function.
> > 
> > Signed-off-by: Stefano Stabellini 
> > 
> > ---
> > Changes in v2:
> > - remove xen_init_display from xen_backend.h
> > - handle cases where vkbd is missing
> > 
> > diff --git a/hw/display/xenfb.c b/hw/display/xenfb.c
> > index e76c0d8..3b0168b 100644
> > --- a/hw/display/xenfb.c
> > +++ b/hw/display/xenfb.c
> > @@ -71,7 +71,6 @@ struct XenFB {
> >  int   fbpages;
> >  int   feature_update;
> >  int   bug_trigger;
> > -int   have_console;
> >  int   do_resize;
> > 
> >  struct {
> > @@ -80,6 +79,7 @@ struct XenFB {
> >  int   up_count;
> >  int   up_fullscreen;
> >  };
> > +static const GraphicHwOps xenfb_ops;
> > 
> >  /*  */
> > 
> > @@ -855,6 +855,8 @@ static int fb_init(struct XenDevice *xendev)
> >  static int fb_initialise(struct XenDevice *xendev)
> >  {
> >  struct XenFB *fb = container_of(xendev, struct XenFB, c.xendev);
> > +struct XenDevice *xin;
> > +struct XenInput *in;
> 
> I think the scope of 'in' can be limited to the 'else' clause where it is 
> used below.
> 
> >  struct xenfb_page *fb_page;
> >  int videoram;
> >  int rc;
> > @@ -877,16 +879,16 @@ static int fb_initialise(struct XenDevice *xendev)
> >  if (rc != 0)
> > return rc;
> > 
> > -#if 0  /* handled in xen_init_display() for now */
> > -if (!fb->have_console) {
> > -fb->c.ds = graphic_console_init(xenfb_update,
> > -xenfb_invalidate,
> > -NULL,
> > -NULL,
> > -fb);
> > -fb->have_console = 1;
> > +fb->c.con = graphic_console_init(NULL, 0, _ops, fb);
> > +
> > +xin = xen_pv_find_xendev("vkbd", xen_domid, 0);
> > +if (xin == NULL) {
> > +xen_pv_printf(xendev, 1, "xenfb is up, but vkbd is not present\n");
> > +} else {
> > +in = container_of(xin, struct XenInput, c.xendev);
> > +in->c.con = fb->c.con;
> 
> I notice that in->c.con is tested in input_initialise() and it will fail if 
> NULL, so there is an inherent race between the frontends and also processing 
> of xenstore watches. That seems a little fragile at best, but the dependency 
> of input_initialise() on in->c.con is also somewhat bogus as the vkbd backend 
> is created for xenfv machine types whereas xenfb is only there for xenpv 
> machine types. I think this needs to be fixed before xenfb can be relied on 
> to initialize correctly in all circumstances.

Yes, you are right that it is fragile. Also, there is no need for this
dependency: I'll remove it, and use qemu_console_lookup_by_index
instead.

Owen's patch will take care of making the qemu_console_lookup_by_index
call conditional on xenfb->raw_pointer_wanted != 1. 


> > +xen_be_check_state(xin);
> >  }
> > -#endif
> > 
> >  if (xenstore_read_fe_int(xendev, "feature-update", 
> > >feature_update) == -1)
> > fb->feature_update = 0;
> > @@ -972,42 +974,3 @@ static const GraphicHwOps xenfb_ops = {
> >  .gfx_update  = xenfb_update,
> >  .update_interval = xenfb_update_interval,
> >  };
> > -
> > -/*
> > - * FIXME/TODO: Kill this.
> > - * Temporary needed while DisplayState reorganization is in flight.
> > - */
> > -void xen_init_display(int domid)
> > -{
> > -struct XenDevice *xfb, *xin;
> > -struct XenFB *fb;
> > -struct XenInput *in;
> > -int i = 0;
> > -
> > -wait_more:
> > -i++;
> > -main_loop_wait(true);
> > -xfb = xen_pv_find_xendev("vfb", domid, 0);
> > -xin = xen_pv_find_xendev("vkbd", domid, 0);
> > -if (!xfb || !xin) {
> > -if (i < 256) {
> > -usleep(1);
> > -goto wait_more;
> > -}
> > -xen_pv_printf(NULL, 1, "displaystate setup failed\n");
> > -return;
> > -}
> > -
> > -/* vfb */
> > -fb = container_of(xfb, struct XenFB, c.xendev);
> > -fb->c.con = graphic_console_init(NULL, 0, _ops, fb);
> > -fb->have_console = 1;
> > -
> > -/* vkbd */
> > -in = container_of(xin, struct XenInput, c.xendev);
> > -in->c.con = fb->c.con;
> > -
> > -/* retry ->init() */
> > -xen_be_check_state(xin);
> > -xen_be_check_state(xfb);
> > -}
> > diff

Re: [Xen-devel] [PATCH v2] xenfb: remove xen_init_display "temporary" hack

2017-06-29 Thread Paul Durrant 
> -Original Message-
> From: Stefano Stabellini [mailto:sstabell...@kernel.org]
> Sent: 28 June 2017 19:37
> To: xen-de...@lists.xenproject.org; qemu-de...@nongnu.org
> Cc: sstabell...@kernel.org; peter.mayd...@linaro.org; Anthony Perard
> ; kra...@redhat.com; Paul Durrant
> 
> Subject: [PATCH v2] xenfb: remove xen_init_display "temporary" hack
> 
> Initialize xenfb properly, as all other backends, from its own
> "initialise" function.
> 
> Signed-off-by: Stefano Stabellini 
> 
> ---
> Changes in v2:
> - remove xen_init_display from xen_backend.h
> - handle cases where vkbd is missing
> 
> diff --git a/hw/display/xenfb.c b/hw/display/xenfb.c
> index e76c0d8..3b0168b 100644
> --- a/hw/display/xenfb.c
> +++ b/hw/display/xenfb.c
> @@ -71,7 +71,6 @@ struct XenFB {
>  int   fbpages;
>  int   feature_update;
>  int   bug_trigger;
> -int   have_console;
>  int   do_resize;
> 
>  struct {
> @@ -80,6 +79,7 @@ struct XenFB {
>  int   up_count;
>  int   up_fullscreen;
>  };
> +static const GraphicHwOps xenfb_ops;
> 
>  /*  */
> 
> @@ -855,6 +855,8 @@ static int fb_init(struct XenDevice *xendev)
>  static int fb_initialise(struct XenDevice *xendev)
>  {
>  struct XenFB *fb = container_of(xendev, struct XenFB, c.xendev);
> +struct XenDevice *xin;
> +struct XenInput *in;

I think the scope of 'in' can be limited to the 'else' clause where it is used 
below.

>  struct xenfb_page *fb_page;
>  int videoram;
>  int rc;
> @@ -877,16 +879,16 @@ static int fb_initialise(struct XenDevice *xendev)
>  if (rc != 0)
>   return rc;
> 
> -#if 0  /* handled in xen_init_display() for now */
> -if (!fb->have_console) {
> -fb->c.ds = graphic_console_init(xenfb_update,
> -xenfb_invalidate,
> -NULL,
> -NULL,
> -fb);
> -fb->have_console = 1;
> +fb->c.con = graphic_console_init(NULL, 0, _ops, fb);
> +
> +xin = xen_pv_find_xendev("vkbd", xen_domid, 0);
> +if (xin == NULL) {
> +xen_pv_printf(xendev, 1, "xenfb is up, but vkbd is not present\n");
> +} else {
> +in = container_of(xin, struct XenInput, c.xendev);
> +in->c.con = fb->c.con;

I notice that in->c.con is tested in input_initialise() and it will fail if 
NULL, so there is an inherent race between the frontends and also processing of 
xenstore watches. That seems a little fragile at best, but the dependency of 
input_initialise() on in->c.con is also somewhat bogus as the vkbd backend is 
created for xenfv machine types whereas xenfb is only there for xenpv machine 
types. I think this needs to be fixed before xenfb can be relied on to 
initialize correctly in all circumstances.

  Paul

> +xen_be_check_state(xin);
>  }
> -#endif
> 
>  if (xenstore_read_fe_int(xendev, "feature-update", 
> >feature_update) == -1)
>   fb->feature_update = 0;
> @@ -972,42 +974,3 @@ static const GraphicHwOps xenfb_ops = {
>  .gfx_update  = xenfb_update,
>  .update_interval = xenfb_update_interval,
>  };
> -
> -/*
> - * FIXME/TODO: Kill this.
> - * Temporary needed while DisplayState reorganization is in flight.
> - */
> -void xen_init_display(int domid)
> -{
> -struct XenDevice *xfb, *xin;
> -struct XenFB *fb;
> -struct XenInput *in;
> -int i = 0;
> -
> -wait_more:
> -i++;
> -main_loop_wait(true);
> -xfb = xen_pv_find_xendev("vfb", domid, 0);
> -xin = xen_pv_find_xendev("vkbd", domid, 0);
> -if (!xfb || !xin) {
> -if (i < 256) {
> -usleep(1);
> -goto wait_more;
> -}
> -xen_pv_printf(NULL, 1, "displaystate setup failed\n");
> -return;
> -}
> -
> -/* vfb */
> -fb = container_of(xfb, struct XenFB, c.xendev);
> -fb->c.con = graphic_console_init(NULL, 0, _ops, fb);
> -fb->have_console = 1;
> -
> -/* vkbd */
> -in = container_of(xin, struct XenInput, c.xendev);
> -in->c.con = fb->c.con;
> -
> -/* retry ->init() */
> -xen_be_check_state(xin);
> -xen_be_check_state(xfb);
> -}
> diff --git a/hw/xenpv/xen_machine_pv.c b/hw/xenpv/xen_machine_pv.c
> index 79aef4e..31d2f25 100644
> --- a/hw/xenpv/xen_machine_pv.c
> +++ b/hw/xenpv/xen_machine_pv.c
> @@ -94,9 +94,6 @@ static void xen_init_pv(MachineState *machine)
> 
>  /* config cleanup hook */
>  atexit(xen_config_cleanup);
> -
> -/* setup framebuffer */
> -xen_init_display(xen_domid);
>  }
> 
>  static void xenpv_machine_init(MachineClass *mc)
> diff --git a/include/hw/xen/xen_backend.h
> b/include/hw/xen/xen_backend.h
> index 852c2ea..8a6fbcb 100644
> ---