diff mbox series

[2/2] warn about two vga cards

Message ID 20180627115836.25092-2-kraxel@redhat.com
State New
Headers show
Series [1/2] vga: disable global_vmstate for 3.0+ machine types | expand

Commit Message

Gerd Hoffmann June 27, 2018, 11:58 a.m. UTC
Two vga cards will listen on the same legacy (isa) ioports.  Due to this
conflict only one of the two cards will work correctly in vga mode.
Print a warning message in that case.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
---
 hw/display/vga.c | 6 ++++++
 1 file changed, 6 insertions(+)

Comments

Philippe Mathieu-Daudé June 27, 2018, 1:10 p.m. UTC | #1
Hi Gerd,

On 06/27/2018 08:58 AM, Gerd Hoffmann wrote:
> Two vga cards will listen on the same legacy (isa) ioports.  Due to this
> conflict only one of the two cards will work correctly in vga mode.

The last one registered...

> Print a warning message in that case.
> 
> Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
> ---
>  hw/display/vga.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/hw/display/vga.c b/hw/display/vga.c
> index c82e6d240a..a6c246f76b 100644
> --- a/hw/display/vga.c
> +++ b/hw/display/vga.c
> @@ -2273,6 +2273,7 @@ MemoryRegion *vga_init_io(VGACommonState *s, Object *obj,
>  void vga_init(VGACommonState *s, Object *obj, MemoryRegion *address_space,
>                MemoryRegion *address_space_io, bool init_vga_ports)
>  {
> +    static bool vgaports_registered;
>      MemoryRegion *vga_io_memory;
>      const MemoryRegionPortio *vga_ports, *vbe_ports;
>  
> @@ -2289,6 +2290,11 @@ void vga_init(VGACommonState *s, Object *obj, MemoryRegion *address_space,
>                                          1);
>      memory_region_set_coalescing(vga_io_memory);
>      if (init_vga_ports) {
> +        if (vgaports_registered) {
> +            warn_report("multiple vga cards may not work as expected");
> +        } else {
> +            vgaports_registered = true;
> +        }
>          portio_list_init(&s->vga_port_list, obj, vga_ports, s, "vga");

Here we leak s->vga_port_list->regions from the 1st card.
Can we call portio_list_destroy() in the if (vgaports_registered) block?

>          portio_list_set_flush_coalesced(&s->vga_port_list);
>          portio_list_add(&s->vga_port_list, address_space_io, 0x3b0);
> 

Regards,

Phil.
Gerd Hoffmann June 27, 2018, 1:51 p.m. UTC | #2
On Wed, Jun 27, 2018 at 10:10:17AM -0300, Philippe Mathieu-Daudé wrote:
> Hi Gerd,
> 
> On 06/27/2018 08:58 AM, Gerd Hoffmann wrote:
> > Two vga cards will listen on the same legacy (isa) ioports.  Due to this
> > conflict only one of the two cards will work correctly in vga mode.
> 
> The last one registered...
> 
> > Print a warning message in that case.
> > 
> > Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
> > ---
> >  hw/display/vga.c | 6 ++++++
> >  1 file changed, 6 insertions(+)
> > 
> > diff --git a/hw/display/vga.c b/hw/display/vga.c
> > index c82e6d240a..a6c246f76b 100644
> > --- a/hw/display/vga.c
> > +++ b/hw/display/vga.c
> > @@ -2273,6 +2273,7 @@ MemoryRegion *vga_init_io(VGACommonState *s, Object *obj,
> >  void vga_init(VGACommonState *s, Object *obj, MemoryRegion *address_space,
> >                MemoryRegion *address_space_io, bool init_vga_ports)
> >  {
> > +    static bool vgaports_registered;
> >      MemoryRegion *vga_io_memory;
> >      const MemoryRegionPortio *vga_ports, *vbe_ports;
> >  
> > @@ -2289,6 +2290,11 @@ void vga_init(VGACommonState *s, Object *obj, MemoryRegion *address_space,
> >                                          1);
> >      memory_region_set_coalescing(vga_io_memory);
> >      if (init_vga_ports) {
> > +        if (vgaports_registered) {
> > +            warn_report("multiple vga cards may not work as expected");
> > +        } else {
> > +            vgaports_registered = true;
> > +        }
> >          portio_list_init(&s->vga_port_list, obj, vga_ports, s, "vga");
> 
> Here we leak s->vga_port_list->regions from the 1st card.
> Can we call portio_list_destroy() in the if (vgaports_registered) block?

We don't have a pointer at hand.  Easier would be to only register
the ports in the else block.

Or throw an error in the if (vgaports_registered) block instead of
printing a warning only.

Hmm ...

cheers,
  Gerd
Philippe Mathieu-Daudé June 27, 2018, 2:15 p.m. UTC | #3
On 06/27/2018 10:51 AM, Gerd Hoffmann wrote:
> On Wed, Jun 27, 2018 at 10:10:17AM -0300, Philippe Mathieu-Daudé wrote:
>> Hi Gerd,
>>
>> On 06/27/2018 08:58 AM, Gerd Hoffmann wrote:
>>> Two vga cards will listen on the same legacy (isa) ioports.  Due to this
>>> conflict only one of the two cards will work correctly in vga mode.
>>
>> The last one registered...
>>
>>> Print a warning message in that case.
>>>
>>> Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
>>> ---
>>>  hw/display/vga.c | 6 ++++++
>>>  1 file changed, 6 insertions(+)
>>>
>>> diff --git a/hw/display/vga.c b/hw/display/vga.c
>>> index c82e6d240a..a6c246f76b 100644
>>> --- a/hw/display/vga.c
>>> +++ b/hw/display/vga.c
>>> @@ -2273,6 +2273,7 @@ MemoryRegion *vga_init_io(VGACommonState *s, Object *obj,
>>>  void vga_init(VGACommonState *s, Object *obj, MemoryRegion *address_space,
>>>                MemoryRegion *address_space_io, bool init_vga_ports)
>>>  {
>>> +    static bool vgaports_registered;
>>>      MemoryRegion *vga_io_memory;
>>>      const MemoryRegionPortio *vga_ports, *vbe_ports;
>>>  
>>> @@ -2289,6 +2290,11 @@ void vga_init(VGACommonState *s, Object *obj, MemoryRegion *address_space,
>>>                                          1);
>>>      memory_region_set_coalescing(vga_io_memory);
>>>      if (init_vga_ports) {
>>> +        if (vgaports_registered) {
>>> +            warn_report("multiple vga cards may not work as expected");
>>> +        } else {
>>> +            vgaports_registered = true;
>>> +        }
>>>          portio_list_init(&s->vga_port_list, obj, vga_ports, s, "vga");
>>
>> Here we leak s->vga_port_list->regions from the 1st card.
>> Can we call portio_list_destroy() in the if (vgaports_registered) block?
> 
> We don't have a pointer at hand.  Easier would be to only register
> the ports in the else block.

I was thinking of:

        if (vgaports_registered) {
            warn_report("multiple vga cards may not work as expected");
            portio_list_destroy(&s->vga_port_list);
        } else {
            vgaports_registered = true;
        }

But "only register the ports in the else block" is easier indeed.

> 
> Or throw an error in the if (vgaports_registered) block instead of
> printing a warning only.
> 
> Hmm ...

Let's see if there interested user replying to this thread...
diff mbox series

Patch

diff --git a/hw/display/vga.c b/hw/display/vga.c
index c82e6d240a..a6c246f76b 100644
--- a/hw/display/vga.c
+++ b/hw/display/vga.c
@@ -2273,6 +2273,7 @@  MemoryRegion *vga_init_io(VGACommonState *s, Object *obj,
 void vga_init(VGACommonState *s, Object *obj, MemoryRegion *address_space,
               MemoryRegion *address_space_io, bool init_vga_ports)
 {
+    static bool vgaports_registered;
     MemoryRegion *vga_io_memory;
     const MemoryRegionPortio *vga_ports, *vbe_ports;
 
@@ -2289,6 +2290,11 @@  void vga_init(VGACommonState *s, Object *obj, MemoryRegion *address_space,
                                         1);
     memory_region_set_coalescing(vga_io_memory);
     if (init_vga_ports) {
+        if (vgaports_registered) {
+            warn_report("multiple vga cards may not work as expected");
+        } else {
+            vgaports_registered = true;
+        }
         portio_list_init(&s->vga_port_list, obj, vga_ports, s, "vga");
         portio_list_set_flush_coalesced(&s->vga_port_list);
         portio_list_add(&s->vga_port_list, address_space_io, 0x3b0);