Hi Paolo,
On 24.06.2016 14:53, Paolo Bonzini wrote:
>
>
> On 24/06/2016 08:41, Claudio Fontana wrote:
>>> I use qboot for similar goals, you mention that PAM is necessary because of
>>> how qboot probes parallel flash,
>>> however in my custom platform I removed PAM completely from QEMU, and
>>> everything seems to work without any problems..
>>
>> Btw before you ask: yes I am booting with pflash.
>
> By default low memory points to PCI address space
>
> 00000000000f0000-00000000000fffff (prio 1, RW): alias pam-ram @pc.ram
> 00000000000f0000-00000000000fffff [disabled]
> 00000000000f0000-00000000000fffff (prio 1, RW): alias pam-pci @pc.ram
> 00000000000f0000-00000000000fffff [disabled]
> 00000000000f0000-00000000000fffff (prio 1, R-): alias pam-rom @pc.ram
> 00000000000f0000-00000000000fffff [disabled]
> 00000000000f0000-00000000000fffff (prio 1, RW): alias pam-pci @pci
> 00000000000f0000-00000000000fffff
>
> All that qboot does is enabling pam-ram:
>
> // Make ram from 0xc0000-0xf0000 read-write
> int i;
> for (i=0; i<6; i++) {
> int pam = pambase + 1 + i;
> pci_config_writeb(bdf, pam, 0x33);
> }
>
> // Make ram from 0xf0000-0x100000 read-write and shadow BIOS
> // We're still running from 0xffff0000
> pci_config_writeb(bdf, pambase, 0x30);
> memcpy(low_start, bios_start, 0x10000);
>
> So if you remove PAM but you are leaving 0xC000-0x10000 pointing to
> RAM, you are effectively moving qboot's PAM configuration to QEMU. :)
>
> Of these writes, only the last write is strictly necessary. qboot
> currently uses 0xe0000-0xf0000 for the ACPI tables but we could move
> them to the EBDA instead and save the initial loop. But I'd like to
> see a trace saying how much time is spent configuring PAM exactly.
>
> Paolo
>
In my case the boot times are satisfactory including the PAM configuration loop
in qboot.
The reason I removed the PAM backend in QEMU (or rather, made them configurable
via existing CONFIG_PAM),
is as part of memory saving patches, not because of boottime issues.
Ciao,
Claudio
