[Qemu-devel] Fix for bochs vbe issues

[Qemu-devel] Fix for bochs vbe issues

[Matthew Mastracci]

[-- Attachment #1: Type: text/plain, Size: 2332 bytes --]

I found a way to fix the bochs VBE issues in Windows 2000.  It looks 
really good in 16+ bit color.  The changes in this patch are:

 - Use linear (chain-4) mode for all VBE_DISPI_ENABLED modes
 - Add ff80 and ff81 as valid ioport registers (these are the old ports, 
still used by the driver)

Some of the issues with the patch:

 - There are black lines that appear on the screen at what appear to be 
lines that span a display bank.  This only seems to happen in 8- or 
24-bit.  Puzzling!
 - The shutdown screen in windows 2000 is corrupted.

A better patch might tweak some of the sequence controller and bank 
select registers to set the chain-4 mode, but I haven't tried that yet.

diff -u -r ./hw/vga.c ../../qemu-0.5.4/hw/vga.c
--- ./hw/vga.c  2004-04-26 15:17:28.000000000 -0600
+++ ../../qemu-0.5.4/hw/vga.c   2004-04-27 21:46:20.828796000 -0600
@@ -678,7 +678,10 @@
         break;
     }

-    if (s->sr[4] & 0x08) {
+       if (s->vbe_regs[VBE_DISPI_INDEX_ENABLE] & VBE_DISPI_ENABLED) {
+        /* VBE mode : simplest access */
+        ret = s->vram_ptr[addr];
+       } else if (s->sr[4] & 0x08) {
         /* chain 4 mode : simplest access */
         ret = s->vram_ptr[addr];
     } else if (s->gr[5] & 0x10) {
@@ -756,7 +759,17 @@
         break;
     }

-    if (s->sr[4] & 0x08) {
+       if (s->vbe_regs[VBE_DISPI_INDEX_ENABLE] & VBE_DISPI_ENABLED) {
+        /* VBE mode : simplest access */
+        plane = (s->gr[4] & 2) | (addr & 1);
+        if (s->sr[2] & (1 << plane)) {
+            s->vram_ptr[addr] = val;
+#ifdef DEBUG_VGA_MEM
+            printf("vga: VBE: [0x%x]\n", addr);
+#endif
+            cpu_physical_memory_set_dirty(s->vram_offset + addr);
+        }
+       } else if (s->sr[4] & 0x08) {
         /* chain 4 mode : simplest access */
         plane = addr & 3;
         if (s->sr[2] & (1 << plane)) {
@@ -1768,6 +1781,11 @@

     register_ioport_write(0x1ce, 1, 2, vbe_ioport_write, s);
     register_ioport_write(0x1cf, 1, 2, vbe_ioport_write, s);
+    register_ioport_read(0xff80, 1, 2, vbe_ioport_read, s);
+    register_ioport_read(0xff81, 1, 2, vbe_ioport_read, s);
+
+    register_ioport_write(0xff80, 1, 2, vbe_ioport_write, s);
+    register_ioport_write(0xff81, 1, 2, vbe_ioport_write, s);
 #endif

     vga_io_memory = cpu_register_io_memory(0, vga_mem_read, vga_mem_write);


[-- Attachment #2: matt.vcf --]
[-- Type: text/x-vcard, Size: 286 bytes --]

begin:vcard
fn:Matthew Mastracci
n:Mastracci;Matthew
org:aclaro Softworks, inc.
adr:;;1900 a - 11 St. SE;Calgary;Alberta;T2H 3G2;Canada
email;internet:matt@aclaro.com
title:Software Developer
tel;work:(403) 299-6612
x-mozilla-html:FALSE
url:http://www.aclaro.com
version:2.1
end:vcard

Re: [Qemu-devel] Fix for bochs vbe issues

[Fabrice Bellard]

Hi,

I just commited a slightly different patch. Tell me if it works.

Fabrice

Matthew Mastracci wrote:
> I found a way to fix the bochs VBE issues in Windows 2000.  It looks 
> really good in 16+ bit color.  The changes in this patch are:
> 
> - Use linear (chain-4) mode for all VBE_DISPI_ENABLED modes
> - Add ff80 and ff81 as valid ioport registers (these are the old ports, 
> still used by the driver)
> 
> Some of the issues with the patch:
> 
> - There are black lines that appear on the screen at what appear to be 
> lines that span a display bank.  This only seems to happen in 8- or 
> 24-bit.  Puzzling!
> - The shutdown screen in windows 2000 is corrupted.
> 
> A better patch might tweak some of the sequence controller and bank 
> select registers to set the chain-4 mode, but I haven't tried that yet.
> 
> diff -u -r ./hw/vga.c ../../qemu-0.5.4/hw/vga.c
> --- ./hw/vga.c  2004-04-26 15:17:28.000000000 -0600
> +++ ../../qemu-0.5.4/hw/vga.c   2004-04-27 21:46:20.828796000 -0600
> @@ -678,7 +678,10 @@
>         break;
>     }
> 
> -    if (s->sr[4] & 0x08) {
> +       if (s->vbe_regs[VBE_DISPI_INDEX_ENABLE] & VBE_DISPI_ENABLED) {
> +        /* VBE mode : simplest access */
> +        ret = s->vram_ptr[addr];
> +       } else if (s->sr[4] & 0x08) {
>         /* chain 4 mode : simplest access */
>         ret = s->vram_ptr[addr];
>     } else if (s->gr[5] & 0x10) {
> @@ -756,7 +759,17 @@
>         break;
>     }
> 
> -    if (s->sr[4] & 0x08) {
> +       if (s->vbe_regs[VBE_DISPI_INDEX_ENABLE] & VBE_DISPI_ENABLED) {
> +        /* VBE mode : simplest access */
> +        plane = (s->gr[4] & 2) | (addr & 1);
> +        if (s->sr[2] & (1 << plane)) {
> +            s->vram_ptr[addr] = val;
> +#ifdef DEBUG_VGA_MEM
> +            printf("vga: VBE: [0x%x]\n", addr);
> +#endif
> +            cpu_physical_memory_set_dirty(s->vram_offset + addr);
> +        }
> +       } else if (s->sr[4] & 0x08) {
>         /* chain 4 mode : simplest access */
>         plane = addr & 3;
>         if (s->sr[2] & (1 << plane)) {
> @@ -1768,6 +1781,11 @@
> 
>     register_ioport_write(0x1ce, 1, 2, vbe_ioport_write, s);
>     register_ioport_write(0x1cf, 1, 2, vbe_ioport_write, s);
> +    register_ioport_read(0xff80, 1, 2, vbe_ioport_read, s);
> +    register_ioport_read(0xff81, 1, 2, vbe_ioport_read, s);
> +
> +    register_ioport_write(0xff80, 1, 2, vbe_ioport_write, s);
> +    register_ioport_write(0xff81, 1, 2, vbe_ioport_write, s);
> #endif
> 
>     vga_io_memory = cpu_register_io_memory(0, vga_mem_read, vga_mem_write);
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Qemu-devel mailing list
> Qemu-devel@nongnu.org
> http://mail.nongnu.org/mailman/listinfo/qemu-devel

Re: [Qemu-devel] Fix for bochs vbe issues

[Matthew Mastracci]

Looks good.  The shutdown screen is still corrupted in Windows, but
that's not a huge issue.  I think that it's expecting to be back in
latched mode, but isn't setting it itself.  

One last thing to add - we need to ensure that the chain-4 mask is set
when we start up VBE mode.  This will ensure that all of the planes are
written (inserted right after your chain-4 mode change):

sr[2] |= 0x0f;

-- 
Matthew Mastracci <matt@aclaro.com>

Re: [Qemu-devel] Fix for bochs vbe issues

[Matthew Mastracci]

And right after I sent this message, I discovered why the shutdown
screen is corrupted:

The VBE_DISPI_ENABLED flag wasn't getting cleared internally when VBE
was disabled by the driver, causing some issues in other parts of the
code that check it:

            } else {
                /* XXX: the bios should do that */
                s->bank_offset = -0xa0000;
+               s->vbe_regs[s->vbe_index] = val;
            }

The above makes the shutdown screen appear as expected.

Matt.

On Wed, 2004-04-28 at 20:17, Matthew Mastracci wrote:
> Looks good.  The shutdown screen is still corrupted in Windows, but
> that's not a huge issue.  I think that it's expecting to be back in
> latched mode, but isn't setting it itself.  
> 
> One last thing to add - we need to ensure that the chain-4 mask is set
> when we start up VBE mode.  This will ensure that all of the planes are
> written (inserted right after your chain-4 mode change):
> 
> sr[2] |= 0x0f;
-- 
Matthew Mastracci <matt@aclaro.com>

Re: [Qemu-devel] Fix for bochs vbe issues

[Michael L Torrie]

[-- Attachment #1: Type: text/plain, Size: 3517 bytes --]

On Wed, 2004-04-28 at 16:42, Fabrice Bellard wrote:
> Hi,
> 
> I just commited a slightly different patch. Tell me if it works.

I updated from CVS this morning, but unfortunately it doens't completely
work.  The screen looks all garbled.  The interesting thing is that if I
install VNC server, the exported screen is all garbled also.

Attached is a screenshot.

It could be the whole install is scrambled.  Or I have the wrong driver.

Michael

> 
> Fabrice
> 
> Matthew Mastracci wrote:
> > I found a way to fix the bochs VBE issues in Windows 2000.  It looks 
> > really good in 16+ bit color.  The changes in this patch are:
> > 
> > - Use linear (chain-4) mode for all VBE_DISPI_ENABLED modes
> > - Add ff80 and ff81 as valid ioport registers (these are the old ports, 
> > still used by the driver)
> > 
> > Some of the issues with the patch:
> > 
> > - There are black lines that appear on the screen at what appear to be 
> > lines that span a display bank.  This only seems to happen in 8- or 
> > 24-bit.  Puzzling!
> > - The shutdown screen in windows 2000 is corrupted.
> > 
> > A better patch might tweak some of the sequence controller and bank 
> > select registers to set the chain-4 mode, but I haven't tried that yet.
> > 
> > diff -u -r ./hw/vga.c ../../qemu-0.5.4/hw/vga.c
> > --- ./hw/vga.c  2004-04-26 15:17:28.000000000 -0600
> > +++ ../../qemu-0.5.4/hw/vga.c   2004-04-27 21:46:20.828796000 -0600
> > @@ -678,7 +678,10 @@
> >         break;
> >     }
> > 
> > -    if (s->sr[4] & 0x08) {
> > +       if (s->vbe_regs[VBE_DISPI_INDEX_ENABLE] & VBE_DISPI_ENABLED) {
> > +        /* VBE mode : simplest access */
> > +        ret = s->vram_ptr[addr];
> > +       } else if (s->sr[4] & 0x08) {
> >         /* chain 4 mode : simplest access */
> >         ret = s->vram_ptr[addr];
> >     } else if (s->gr[5] & 0x10) {
> > @@ -756,7 +759,17 @@
> >         break;
> >     }
> > 
> > -    if (s->sr[4] & 0x08) {
> > +       if (s->vbe_regs[VBE_DISPI_INDEX_ENABLE] & VBE_DISPI_ENABLED) {
> > +        /* VBE mode : simplest access */
> > +        plane = (s->gr[4] & 2) | (addr & 1);
> > +        if (s->sr[2] & (1 << plane)) {
> > +            s->vram_ptr[addr] = val;
> > +#ifdef DEBUG_VGA_MEM
> > +            printf("vga: VBE: [0x%x]\n", addr);
> > +#endif
> > +            cpu_physical_memory_set_dirty(s->vram_offset + addr);
> > +        }
> > +       } else if (s->sr[4] & 0x08) {
> >         /* chain 4 mode : simplest access */
> >         plane = addr & 3;
> >         if (s->sr[2] & (1 << plane)) {
> > @@ -1768,6 +1781,11 @@
> > 
> >     register_ioport_write(0x1ce, 1, 2, vbe_ioport_write, s);
> >     register_ioport_write(0x1cf, 1, 2, vbe_ioport_write, s);
> > +    register_ioport_read(0xff80, 1, 2, vbe_ioport_read, s);
> > +    register_ioport_read(0xff81, 1, 2, vbe_ioport_read, s);
> > +
> > +    register_ioport_write(0xff80, 1, 2, vbe_ioport_write, s);
> > +    register_ioport_write(0xff81, 1, 2, vbe_ioport_write, s);
> > #endif
> > 
> >     vga_io_memory = cpu_register_io_memory(0, vga_mem_read, vga_mem_write);
> > 
> > 
> > ------------------------------------------------------------------------
> > 
> > _______________________________________________
> > Qemu-devel mailing list
> > Qemu-devel@nongnu.org
> > http://mail.nongnu.org/mailman/listinfo/qemu-devel
> 
> 
> 
> 
> _______________________________________________
> Qemu-devel mailing list
> Qemu-devel@nongnu.org
> http://mail.nongnu.org/mailman/listinfo/qemu-devel
-- 
Michael L Torrie <torriem@chem.byu.edu>

[-- Attachment #2: Screenshot-qemu.png --]
[-- Type: image/png, Size: 22204 bytes --]

[Qemu-devel] Re: Fix for bochs vbe issues

[Matthew Mastracci]

[-- Attachment #1: Type: text/plain, Size: 504 bytes --]


>I updated from CVS this morning, but unfortunately it doens't completely
>work.  The screen looks all garbled.  The interesting thing is that if I
>install VNC server, the exported screen is all garbled also.
>
>Attached is a screenshot.
>
>It could be the whole install is scrambled.  Or I have the wrong driver.
>
>  
>
You have the right driver, but you'll need to grab the latest CVS stuff 
for it all to work.  I noticed that Fabrice just recently committed the 
other two fixes remaining.

Matt.

[-- Attachment #2: matt.vcf --]
[-- Type: text/x-vcard, Size: 286 bytes --]

begin:vcard
fn:Matthew Mastracci
n:Mastracci;Matthew
org:aclaro Softworks, inc.
adr:;;1900 a - 11 St. SE;Calgary;Alberta;T2H 3G2;Canada
email;internet:matt@aclaro.com
title:Software Developer
tel;work:(403) 299-6612
x-mozilla-html:FALSE
url:http://www.aclaro.com
version:2.1
end:vcard

Re: [Qemu-devel] Re: Fix for bochs vbe issues

[Lionel Ulmer]

> You have the right driver, but you'll need to grab the latest CVS stuff 
> for it all to work.  I noticed that Fabrice just recently committed the 
> other two fixes remaining.

Well, just tried with the latest CVS and I still have the 'black horizontal
bars' in 32 bpp mode (but not the disappearing Red color as before)..

But, more problematic, is a nice crash when switching to 16 bpp mode
(segmentation fault).

             Lionel

-- 
		 Lionel Ulmer - http://www.bbrox.org/

[Qemu-devel] Re: Fix for bochs vbe issues

[Matthew Mastracci]

[-- Attachment #1: Type: text/plain, Size: 339 bytes --]

Lionel Ulmer wrote:

>Well, just tried with the latest CVS and I still have the 'black horizontal
>bars' in 32 bpp mode (but not the disappearing Red color as before)..
>  
>
I've found that the horizontal bars will disappear if you use 
1024x768x16bpp or 1024x768x32bpp.  I'm not certain what the problem is 
that causes the bars.

Matt.

[-- Attachment #2: matt.vcf --]
[-- Type: text/x-vcard, Size: 286 bytes --]

begin:vcard
fn:Matthew Mastracci
n:Mastracci;Matthew
org:aclaro Softworks, inc.
adr:;;1900 a - 11 St. SE;Calgary;Alberta;T2H 3G2;Canada
email;internet:matt@aclaro.com
title:Software Developer
tel;work:(403) 299-6612
x-mozilla-html:FALSE
url:http://www.aclaro.com
version:2.1
end:vcard

Re: [Qemu-devel] Re: Fix for bochs vbe issues

[Jean-Michel POURE]

Le jeudi 29 Avril 2004 22:12, Lionel Ulmer a écrit :
> Well, just tried with the latest CVS and I still have the 'black horizontal
> bars' in 32 bpp mode (but not the disappearing Red color as before)..

I remember reading on the Bochs mailing list that the VBE drivers did not 
support 32 but 24 bits. Can you try 24 bits depth? Just a guess.
Cheers, Jean-Michel