[edk2-devel] The different I/O ports space distribution in Ovmf and seaBIOS , can I change the default I/O space in Ovmf?

[edk2-devel] The different I/O ports space distribution in Ovmf and seaBIOS , can I change the default I/O space in Ovmf?

[huaminxu3 via groups.io]

[-- Attachment #1.1: Type: text/plain, Size: 3051 bytes --]

In OVMF PCI I/O space address distribution , according to the OvmfPkg/Library/PlatformInitLib/Platform.c
VOID
EFIAPI
PlatformMemMapInitialization (
IN OUT EFI_HOB_PLATFORM_INFO * PlatformInfoHob
)
{
UINT64 PciIoBase ;
UINT64 PciIoSize ;
UINT64 PciExBarBase ;
UINT32 PciBase ;
UINT32 PciSize ;

PciIoBase = 0xC000 ;
PciIoSize = 0x4000 ;

The PciIoBase and PciIoSize determine the I/O address space spare for PCI devices. In Ovmf, the two variables are fixed. The PciIoBase = 0xC000 and the PciIoSize = 0x4000 , which means the I/O space for PCI devices is 0x4000. In some circumstances, for example, when I attach many devices to the virtual machine , the 0x4000 is not enough. When I attach 8 PCI bridges and 100 netcards and 100 virtio disks, many devices are unusable. The I/O address of these devices become all 1, just as the following picture shows:

I compared the ovmf with seaBIOS and found that the virtual machine using seaBIOS can work properly. I looked into the I/O space  distribution implement of seaBIOS and found that in qemu\roms\seabios\src\fw\pciinit.c, it says :
/*
* QEMU I/O address space usage:
*   0000 - 0fff    legacy isa, pci config, pci root bus, ...
*   1000 - 9fff    free
*   a000 - afff    hotplug (cpu, pci via acpi, i440fx/piix only)
*   b000 - bfff    power management (PORT_ACPI_PM_BASE)
*                  [ qemu 1.4+ implements pci config registers
*                    properly so guests can place the registers
*                    where they want, on older versions its fixed ]
*   c000 - ffff    free, traditionally used for pci io
*/
It seems that the 0x1000-0x9fff space can be used for PCI I/O space. In the seaBIOS, it calculates the total I/O space usage by PCI devices, if the total usage exceeds 0x4000, it use 0x1000-0x9fff for PCI I/O space, otherwise it uses 0xC000-0xFFFF:
struct pci_region * r_io = & bus -> r [ PCI_REGION_TYPE_IO ];
u64 sum = pci_region_sum ( r_io );
if ( sum < 0x4000 ) {
/* traditional region is big enougth, use it */
r_io -> base = 0xc000 ;
} else if ( sum < pci_io_low_end - 0x1000 ) {
/* use the larger region at 0x1000 */
r_io -> base = 0x1000 ;
} else {
/* not enouth io address space -> error out */
return - 1 ;
}

Due to the different implement of OVMF and seaBIOS, the OVMF can not know how much PCI I/O space all the devices use. So I am wondering that is it proper to change the default PCI I/O space to 0x1000-0x9FFF to fit more devices (for example , 100 netcards and 100 virtio disks).  In that case, the PciIoBase =0x1000 and PciIoSize = 0x9000.
I am hoping that I can get a reply. Thank for you help.


-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#121027): https://edk2.groups.io/g/devel/message/121027
Mute This Topic: https://groups.io/mt/110712298/7686176
Group Owner: devel+owner@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [rebecca@openfw.io]
-=-=-=-=-=-=-=-=-=-=-=-



[-- Attachment #1.2: Type: text/html, Size: 10939 bytes --]

[-- Attachment #2: image.png --]
[-- Type: image/png, Size: 150527 bytes --]