[Qemu-devel] [PATCH] [0.12] Map BIOS f-segment as RAM, not as ROM

[Qemu-devel] [PATCH] [0.12] Map BIOS f-segment as RAM, not as ROM

[Alexander Graf]

SeaBIOS needs to write to the f-segment. So it needs to have some way to set
that from read-only to read-write, write in it and when it's done set it to
read-only again.

On PCI we have a mechanism for that. The ISA machine does not though. To stay
regression free and happily enable users to continue using the -M isapc machine
let's just map it as RAM on the ISA PC.

Signed-off-by: Alexander Graf <agraf@suse.de>
---
 hw/pc.c |   22 ++++++++++++++++++++--
 1 files changed, 20 insertions(+), 2 deletions(-)

diff --git a/hw/pc.c b/hw/pc.c
index 8c1b7ea..90373cf 100644
--- a/hw/pc.c
+++ b/hw/pc.c
@@ -982,7 +982,7 @@ static void pc_init1(ram_addr_t ram_size,
     int ret, linux_boot, i;
     ram_addr_t ram_addr, bios_offset, option_rom_offset;
     ram_addr_t below_4g_mem_size, above_4g_mem_size = 0;
-    int bios_size, isa_bios_size;
+    int bios_size, isa_bios_size, f_seg_type;
     PCIBus *pci_bus;
     ISADevice *isa_dev;
     int piix3_devfn = -1;
@@ -1073,9 +1073,27 @@ static void pc_init1(ram_addr_t ram_size,
     isa_bios_size = bios_size;
     if (isa_bios_size > (128 * 1024))
         isa_bios_size = 128 * 1024;
+
+    /* XXX
+     *
+     * Usually we would map this region read only, have the BIOS set it r/w
+     * when it needs to and have the BIOS also set it r/o when it's done
+     * messing with it.
+     *
+     * Unfortunately we don't export any device to the ISA PC machine that
+     * could do the r/w <-> r/o transition, so let's just always map it r/w
+     * there for now.
+     */
+
+    if (pci_enabled) {
+        f_seg_type = IO_MEM_ROM;
+    } else {
+        f_seg_type = 0;
+    }
+
     cpu_register_physical_memory(0x100000 - isa_bios_size,
                                  isa_bios_size,
-                                 (bios_offset + bios_size - isa_bios_size) | IO_MEM_ROM);
+                                 (bios_offset + bios_size - isa_bios_size) | f_seg_type);
 
 
 
-- 
1.6.0.2

[Qemu-devel] Re: [PATCH] [0.12] Map BIOS f-segment as RAM, not as ROM

[Kevin O'Connor]

On Fri, Dec 04, 2009 at 01:24:52AM +0100, Alexander Graf wrote:
> SeaBIOS needs to write to the f-segment. So it needs to have some way to set
> that from read-only to read-write, write in it and when it's done set it to
> read-only again.
> 
> On PCI we have a mechanism for that. The ISA machine does not though. To stay
> regression free and happily enable users to continue using the -M isapc machine
> let's just map it as RAM on the ISA PC.
> 
> Signed-off-by: Alexander Graf <agraf@suse.de>

Thanks Alexander.

Why not always map the f-segment as ram?  On a PCI machine, both bochs
bios and seabios will make it read-only at the end of POST regardless
of how it started.

-Kevin

[Qemu-devel] Re: [PATCH] [0.12] Map BIOS f-segment as RAM, not as ROM

[Alexander Graf]

On 04.12.2009, at 01:44, Kevin O'Connor wrote:

> On Fri, Dec 04, 2009 at 01:24:52AM +0100, Alexander Graf wrote:
>> SeaBIOS needs to write to the f-segment. So it needs to have some way to set
>> that from read-only to read-write, write in it and when it's done set it to
>> read-only again.
>> 
>> On PCI we have a mechanism for that. The ISA machine does not though. To stay
>> regression free and happily enable users to continue using the -M isapc machine
>> let's just map it as RAM on the ISA PC.
>> 
>> Signed-off-by: Alexander Graf <agraf@suse.de>
> 
> Thanks Alexander.
> 
> Why not always map the f-segment as ram?  On a PCI machine, both bochs
> bios and seabios will make it read-only at the end of POST regardless
> of how it started.

I merely wanted to fix the regression. If you think it's a good idea to always map it R/W, I'll gladly change the patch.

Alex