* Re: [Qemu-devel] Standard vga breakage
2009-11-15 20:16 ` Stefan Weil
@ 2009-11-15 20:37 ` Blue Swirl
2009-11-15 21:21 ` malc
2009-11-16 19:12 ` Anthony Liguori
2 siblings, 0 replies; 12+ messages in thread
From: Blue Swirl @ 2009-11-15 20:37 UTC (permalink / raw)
To: Stefan Weil; +Cc: qemu-devel, Gleb Natapov
On Sun, Nov 15, 2009 at 10:16 PM, Stefan Weil <weil@mail.berlios.de> wrote:
> malc schrieb:
>> On Sun, 15 Nov 2009, Gleb Natapov wrote:
>>
>>> On Sun, Nov 15, 2009 at 06:22:40PM +0300, malc wrote:
>>>> On Sun, 15 Nov 2009, Blue Swirl wrote:
>>>>
>>>>> On Sun, Nov 15, 2009 at 3:25 AM, malc <av1474@comtv.ru> wrote:
>>>>>> debian-503-i386-netinst.iso when booted on a machine with `-vga
>>>>>> std' just
>>>>>> shows black screen (after bios' setup screen)
>>>>> Just some observations:
>>>>> - if you press enter, installer starts and Linux is able to fix the
>>>>> display
>>>>> - the screen is fine with -bios pcbios.bin, so the issue is related
>>>>> to SeaBIOS
>>>> Yes i forgot to mention that -bios works, also worth noting is that
>>>> before
>>>> the commit that fixed cirrus -bios hack also made it work.
>
>
> Here is a summary of my analysis of this problem:
>
> Graphics mode memory (ram 0xe0000000 ...) is not mapped
> in QEMU address space (which results in unassigned
> memory writes) for standard pci vga (-vga std), so users
> don't get vga display output with grub (in graphical mode),
> debian installer (in graphical mode) or similar boot software.
>
> This happens for PC system emulation using SeaBIOS.
> It won't happen when using Bochs BIOS.
>
> Other system emulations which can run standard pci vga
> seem to suffer from the same problem: when I run mips
> malta using -vga std, it won't display anything on the vga
> screen (it works with cirrus pci vga).
>
> Malta has no vga bios, so the error might be in QEMU's
> vga code (hw/vga-pci.c). Maybe wrong initial PCI mappings?
> Maybe this has always be wrong (which would explain why
> git bisect does not work).
>
> pci_vga_init() is called by sun4u, ppc_prep, ppc_oldworld,
> ppc_newworld, pc and mips, so all these emulations
> share the same potential problem.
At least sun4u, ppc_oldworld and ppc_newworld are fine with -vga std.
There is no VGA BIOS.
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [Qemu-devel] Standard vga breakage
2009-11-15 20:16 ` Stefan Weil
2009-11-15 20:37 ` Blue Swirl
@ 2009-11-15 21:21 ` malc
2009-11-15 21:35 ` Blue Swirl
2009-11-16 18:21 ` Stefan Weil
2009-11-16 19:12 ` Anthony Liguori
2 siblings, 2 replies; 12+ messages in thread
From: malc @ 2009-11-15 21:21 UTC (permalink / raw)
To: Stefan Weil; +Cc: Blue Swirl, qemu-devel, Gleb Natapov
On Sun, 15 Nov 2009, Stefan Weil wrote:
> malc schrieb:
> > On Sun, 15 Nov 2009, Gleb Natapov wrote:
[..snip..]
>
> Here is a summary of my analysis of this problem:
>
> Graphics mode memory (ram 0xe0000000 ...) is not mapped
> in QEMU address space (which results in unassigned
> memory writes) for standard pci vga (-vga std), so users
> don't get vga display output with grub (in graphical mode),
> debian installer (in graphical mode) or similar boot software.
>
> This happens for PC system emulation using SeaBIOS.
> It won't happen when using Bochs BIOS.
>
> Other system emulations which can run standard pci vga
> seem to suffer from the same problem: when I run mips
> malta using -vga std, it won't display anything on the vga
> screen (it works with cirrus pci vga).
>
> Malta has no vga bios, so the error might be in QEMU's
> vga code (hw/vga-pci.c). Maybe wrong initial PCI mappings?
> Maybe this has always be wrong (which would explain why
> git bisect does not work).
git bisect doesn't work because most of the time:
a. Compilation fails
b. Compilation succeeds but qemu segfaults before even trying to
execute any guest code
>
> pci_vga_init() is called by sun4u, ppc_prep, ppc_oldworld,
> ppc_newworld, pc and mips, so all these emulations
> share the same potential problem.
>
> Bochs BIOS and PC Linux obviously fix the problem.
>
> Regards,
> Stefan
>
--
mailto:av1474@comtv.ru
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [Qemu-devel] Standard vga breakage
2009-11-15 21:21 ` malc
@ 2009-11-15 21:35 ` Blue Swirl
2009-11-16 18:21 ` Stefan Weil
1 sibling, 0 replies; 12+ messages in thread
From: Blue Swirl @ 2009-11-15 21:35 UTC (permalink / raw)
To: malc; +Cc: qemu-devel, Gleb Natapov
On Sun, Nov 15, 2009 at 11:21 PM, malc <av1474@comtv.ru> wrote:
> On Sun, 15 Nov 2009, Stefan Weil wrote:
>
>> malc schrieb:
>> > On Sun, 15 Nov 2009, Gleb Natapov wrote:
>
> [..snip..]
>
>>
>> Here is a summary of my analysis of this problem:
>>
>> Graphics mode memory (ram 0xe0000000 ...) is not mapped
>> in QEMU address space (which results in unassigned
>> memory writes) for standard pci vga (-vga std), so users
>> don't get vga display output with grub (in graphical mode),
>> debian installer (in graphical mode) or similar boot software.
>>
>> This happens for PC system emulation using SeaBIOS.
>> It won't happen when using Bochs BIOS.
>>
>> Other system emulations which can run standard pci vga
>> seem to suffer from the same problem: when I run mips
>> malta using -vga std, it won't display anything on the vga
>> screen (it works with cirrus pci vga).
>>
>> Malta has no vga bios, so the error might be in QEMU's
>> vga code (hw/vga-pci.c). Maybe wrong initial PCI mappings?
>> Maybe this has always be wrong (which would explain why
>> git bisect does not work).
>
> git bisect doesn't work because most of the time:
>
> a. Compilation fails
> b. Compilation succeeds but qemu segfaults before even trying to
> execute any guest code
With infinite time and resources, we could construct a fully
bisectable tree with the following properties:
- buggy commits (especially commits that break build or require manual
reconfiguration) are replaced with corrected commits by changing the
history
- no merges, linear history
- compound commits are broken down to simple, atomic versions
- documentation is added by the first commit to reduce the number of commits
- spelling fixes applied by changing history, not as a separate commit
- each commit should still have some relation to some real tree commit
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [Qemu-devel] Standard vga breakage
2009-11-15 21:21 ` malc
2009-11-15 21:35 ` Blue Swirl
@ 2009-11-16 18:21 ` Stefan Weil
1 sibling, 0 replies; 12+ messages in thread
From: Stefan Weil @ 2009-11-16 18:21 UTC (permalink / raw)
To: malc; +Cc: Blue Swirl, qemu-devel, Gleb Natapov
malc schrieb:
> On Sun, 15 Nov 2009, Stefan Weil wrote:
>
>
>> malc schrieb:
>>
>>> On Sun, 15 Nov 2009, Gleb Natapov wrote:
>>>
>
> [..snip..]
>
>
>> Here is a summary of my analysis of this problem:
>>
>> Graphics mode memory (ram 0xe0000000 ...) is not mapped
>> in QEMU address space (which results in unassigned
>> memory writes) for standard pci vga (-vga std), so users
>> don't get vga display output with grub (in graphical mode),
>> debian installer (in graphical mode) or similar boot software.
>>
>> This happens for PC system emulation using SeaBIOS.
>> It won't happen when using Bochs BIOS.
>>
>> Other system emulations which can run standard pci vga
>> seem to suffer from the same problem: when I run mips
>> malta using -vga std, it won't display anything on the vga
>> screen (it works with cirrus pci vga).
>>
>> Malta has no vga bios, so the error might be in QEMU's
>> vga code (hw/vga-pci.c). Maybe wrong initial PCI mappings?
>> Maybe this has always be wrong (which would explain why
>> git bisect does not work).
>>
>
> git bisect doesn't work because most of the time:
>
> a. Compilation fails
> b. Compilation succeeds but qemu segfaults before even trying to
> execute any guest code
>
I did git bisect: even very early versions (from 2006) of std vga had
the same kind of problem (no access to frame buffer at 0xe0000000)
when booted with SeaBIOS). It was not really a bisection, because
I never reached status "good".
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [Qemu-devel] Standard vga breakage
2009-11-15 20:16 ` Stefan Weil
2009-11-15 20:37 ` Blue Swirl
2009-11-15 21:21 ` malc
@ 2009-11-16 19:12 ` Anthony Liguori
2009-11-16 20:11 ` [Qemu-devel] [PATCH] vga-pci: Fix access to linear framebuffer Stefan Weil
2 siblings, 1 reply; 12+ messages in thread
From: Anthony Liguori @ 2009-11-16 19:12 UTC (permalink / raw)
To: Stefan Weil; +Cc: Blue Swirl, qemu-devel, Gleb Natapov
Stefan Weil wrote:
> malc schrieb:
>
>> On Sun, 15 Nov 2009, Gleb Natapov wrote:
>>
>>
>>> On Sun, Nov 15, 2009 at 06:22:40PM +0300, malc wrote:
>>>
>>>> On Sun, 15 Nov 2009, Blue Swirl wrote:
>>>>
>>>>
>>>>> On Sun, Nov 15, 2009 at 3:25 AM, malc <av1474@comtv.ru> wrote:
>>>>>
>>>>>> debian-503-i386-netinst.iso when booted on a machine with `-vga
>>>>>> std' just
>>>>>> shows black screen (after bios' setup screen)
>>>>>>
>>>>> Just some observations:
>>>>> - if you press enter, installer starts and Linux is able to fix the
>>>>> display
>>>>> - the screen is fine with -bios pcbios.bin, so the issue is related
>>>>> to SeaBIOS
>>>>>
>>>> Yes i forgot to mention that -bios works, also worth noting is that
>>>> before
>>>> the commit that fixed cirrus -bios hack also made it work.
>>>>
>
>
> Here is a summary of my analysis of this problem:
>
> Graphics mode memory (ram 0xe0000000 ...) is not mapped
> in QEMU address space (which results in unassigned
> memory writes) for standard pci vga (-vga std), so users
> don't get vga display output with grub (in graphical mode),
> debian installer (in graphical mode) or similar boot software.
>
0xe00... is the VBE mapping. The VGABIOS assumes this mapping exists
when using VBE. Cirrus doesn't support VBE so it wouldn't be exposed there.
Does the following help:
diff --git a/hw/vga-pci.c b/hw/vga-pci.c
index 479ffb1..e273b8b 100644
--- a/hw/vga-pci.c
+++ b/hw/vga-pci.c
@@ -107,6 +107,12 @@ static int pci_vga_initfn(PCIDevice *dev)
PCI_BASE_ADDRESS_MEM_PREFETCH, vga_map);
}
+#ifdef CONFIG_BOCHS_VBE
+ /* XXX: use optimized standard vga accesses */
+ cpu_register_physical_memory(VBE_DISPI_LFB_PHYSICAL_ADDRESS,
+ vga_ram_size, s->vga.vram_offset);
+#endif
+
/* ROM BIOS */
rom_add_vga(VGABIOS_FILENAME);
return 0;
Regards,
Anthony Liguori
^ permalink raw reply related [flat|nested] 12+ messages in thread* [Qemu-devel] [PATCH] vga-pci: Fix access to linear framebuffer
2009-11-16 19:12 ` Anthony Liguori
@ 2009-11-16 20:11 ` Stefan Weil
0 siblings, 0 replies; 12+ messages in thread
From: Stefan Weil @ 2009-11-16 20:11 UTC (permalink / raw)
To: anthony, qemu-devel
Anthony Liguori's patch fixes the problems with
vga display in graphical mode and SeaBIOS.
I only adapted some values for vga-pci.
Signed-off-by: Stefan Weil <weil@mail.berlios.de>
---
hw/vga-pci.c | 6 ++++++
1 files changed, 6 insertions(+), 0 deletions(-)
diff --git a/hw/vga-pci.c b/hw/vga-pci.c
index 234d581..a6ad46b 100644
--- a/hw/vga-pci.c
+++ b/hw/vga-pci.c
@@ -107,6 +107,12 @@ static int pci_vga_initfn(PCIDevice *dev)
PCI_BASE_ADDRESS_MEM_PREFETCH, vga_map);
}
+#ifdef CONFIG_BOCHS_VBE
+ /* XXX: use optimized standard vga accesses */
+ cpu_register_physical_memory(VBE_DISPI_LFB_PHYSICAL_ADDRESS,
+ VGA_RAM_SIZE, s->vram_offset);
+#endif
+
/* ROM BIOS */
rom_add_vga(VGABIOS_FILENAME);
return 0;
--
1.5.6.5
^ permalink raw reply related [flat|nested] 12+ messages in thread