Re: [PATCH v2 08/12] macfb: add common monitor modes supported by the MacOS toolbox ROM

[Laurent Vivier]

Le 06/10/2021 à 18:09, Mark Cave-Ayland a écrit :
> On 06/10/2021 16:46, Laurent Vivier wrote:
>> Le 06/10/2021 à 15:54, Mark Cave-Ayland a écrit :
>>> On 06/10/2021 13:24, Laurent Vivier wrote:
>>>
>>>>> This is where it becomes a bit trickier, since technically booting Linux 
>>>>> with -kernel you can use
>>>>> any supported values as long as everything fits in the video RAM which is 
>>>>> why there isn't
>>>>> currently
>>>>> a hard-coded list :)
>>>>>
>>>>
>>>> We need the list of "supported values". I don't want to read the code or 
>>>> try values combination
>>>> until it works.
>>>>
>>>> In a perfect world, I would like to be able to use any value I want with 
>>>> "-kernel".
>>>>
>>>> For instance I was using "-g 1200x800x24" and it was working fine.
>>>>
>>>> Now I have:
>>>>
>>>> qemu-system-m68k: unknown display mode: width 1200, height 800, depth 24
>>>>
>>>> If it's not possible (because the original hardware cannot provide it, and 
>>>> we don't know the base
>>>> address or things like that), we don't need the list of the display id, 
>>>> but the list of available
>>>> modes: (width,height,depth).
>>>>
>>>> Rougly, something like:
>>>>
>>>> qemu-system-m68k: unknown display mode: width 1200, height 800, depth 24
>>>> Available modes:
>>>>       1152x870x8
>>>>       1152x870x4
>>>>       1152x870x2
>>>>       1152x870x1
>>>>       800x600x24
>>>>       800x600x8
>>>>       800x600x4
>>>>       800x600x2
>>>>       800x600x1
>>>>       640x480x24
>>>>       640x480x8
>>>>       640x480x4
>>>>       640x480x2
>>>>       640x480x1
>>>>
>>>> diff --git a/hw/display/macfb.c b/hw/display/macfb.c
>>>> index 5b8812e9e7d8..4b352eb89c3f 100644
>>>> --- a/hw/display/macfb.c
>>>> +++ b/hw/display/macfb.c
>>>> @@ -438,6 +438,26 @@ static MacFbMode *macfb_find_mode(MacfbDisplayType 
>>>> display_type,
>>>>        return NULL;
>>>>    }
>>>>
>>>> +static gchar *macfb_mode_list(void)
>>>> +{
>>>> +    gchar *list = NULL;
>>>> +    gchar *mode;
>>>> +    MacFbMode *macfb_mode;
>>>> +    int i;
>>>> +
>>>> +    for (i = 0; i < ARRAY_SIZE(macfb_mode_table); i++) {
>>>> +        macfb_mode = &macfb_mode_table[i];
>>>> +
>>>> +        mode = g_strdup_printf("    %dx%dx%d\n", macfb_mode->width,
>>>> +                               macfb_mode->height, macfb_mode->depth);
>>>> +        list = g_strconcat(mode, list, NULL);
>>>> +        g_free(mode);
>>>> +    }
>>>> +
>>>> +    return list;
>>>> +}
>>>> +
>>>> +
>>>>    static void macfb_update_display(void *opaque)
>>>>    {
>>>>        MacfbState *s = opaque;
>>>> @@ -620,8 +640,13 @@ static bool macfb_common_realize(DeviceState *dev, 
>>>> MacfbState *s, Error
>>>> **errp)
>>>>
>>>>        s->mode = macfb_find_mode(s->type, s->width, s->height, s->depth);
>>>>        if (!s->mode) {
>>>> +        gchar *list;
>>>>            error_setg(errp, "unknown display mode: width %d, height %d, 
>>>> depth %d",
>>>>                       s->width, s->height, s->depth);
>>>> +        list =  macfb_mode_list();
>>>> +        error_append_hint(errp, "Available modes:\n%s", list);
>>>> +        g_free(list);
>>>> +
>>>>            return false;
>>>>        }
>>>
>>> Hi Laurent,
>>>
>>> Thanks for the example - I can certainly squash this into patch 8.
>>
>> yes, please.
> 
> Okay I'll do that for a v3 (and also split the 1st patch that Phil suggested).
> 
>>> As for allowing extra resolutions via -kernel, since the check is being 
>>> done in
>>> macfb_common_realize() then it would be possible to add a qdev property 
>>> that only gets set when
>>> -kernel is used on the command line which bypasses the mode check if you 
>>> prefer?
>>>
>>
>> I think it can wait and be done by a patch later. For the moment we can 
>> focus on MacOS.
>>
>>> I'm not sure that your existing example of "-g 1200x800x24" (or indeed any 
>>> resolution with 24-bit
>>> depth) with -kernel will still work after this patchset given that the 
>>> 24-bit encoding scheme has
>>> changed. Presumably this would also need a corresponding change in the 
>>> bootinfo/kernel framebuffer/X
>>> configuration somewhere?
>>
>> The kernel framebuffer should be easy to fix, if needed, normally we pass 
>> the needed information via
>> the bootinfo structure.
>>
>> My X configuration is broken for a while. With debian/sid I've never been 
>> able to start X (even on a
>> real q800, I think), and with debian/etch we need a special kernel as the 
>> ADB stack has been broken
>> with old kernel. I was not able to start X for a while now...
> 
> FWIW I found that the last set of ADB fixes in mac_via.c actually fixed ADB 
> on old kernels again (I
> was able to use keyboard and mouse on the 4.15 kernel you used for the 
> original patches), so you may
> be able to get debian/etch working in QEMU. I'd expect forcing EMILE into a 
> 24-bit depth on a real
> Quadra 800 would also show the same issue here.
> 

In fact, I was trying the etch kernel, 2.6.32 that was never able to boot 
before because VBL
interrupt was not implemented. Now, with your series it boots (congratulation) 
but there is some
other issues in ADB but I don't think it's related to the changes in QEMU.

I've found another kernel on my disk (4.1.39) and I'm able to start a 
gnome-session with 1152x870x8
mode. ADB works well. There is an issue with gdm that doesn't take characters 
from the keyboard but
if I start X manually and then gnome-sessions I'm able to play with it.

Thanks,
Laurent