Re: [Xen-devel] placement of emulated flash [was: seabios Improved multi-platform support]

[Ian Campbell <ijc@xxxxxxxxxxxxxx>]

(dropping Attilio he has left Citrix).

On Wed, 2013-02-20 at 23:50 +0100, Laszlo Ersek wrote:
> On 02/20/13 18:18, Ian Campbell wrote:
> > On Wed, 2013-02-20 at 17:55 +0100, Laszlo Ersek wrote:
> >> However in OVMF the RESERVED_MEMBASE range is not parsed from this
> >> Xen-exported table, it is added manually in InitializeXen()
> >> [OvmfPkg/PlatformPei/Xen.c]:
> >>
> >>   //
> >>   // Reserve away HVMLOADER reserved memory [0xFC000000,0xFD000000).
> >>   // This needs to match HVMLOADER RESERVED_MEMBASE/RESERVED_MEMSIZE.
> >>   //
> >>   AddReservedMemoryBaseSizeHob (0xFC000000, 0x1000000);
> > 
> > ICK, it would be far preferable for OVMF to do what SeaBIOS does and
> > actually "communicate" with hvmloader IMHO.
> 
> The table with "XenHVMSeaBIOS" signature is not created for OVMF I
> think; comparing the "bios_config" structure of function pointers
> between "seabios.c" and "ovmf.c" in tools/firmware/hvmloader/, the
> ovmf_config.bios_info_setup field is NULL.

Yes. I think this should be changed then to provide a proper way for
hvmloader to communicate to OVMF. It would seem sensible to follow a
similar mechanism to SeaBIOS, or if possible just reuse the exact same
thing (perhaps with a different signature).

> However there seem to be at least two "info" tables. Referring back,
> "seabios_info" at BIOS_INFO_PHYSICAL_ADDRESS (0x1000) is for SeaBIOS
> only, but "hvm_info" just below the end of conventional memory (at
> HVM_INFO_PADDR, 0x9F800) looks guest firmware independent.

hvm_info is used by the toolstack to pass parameters to hvmloader itself
when it starts, it is not intended to pass options from hvmloader to the
bios. It is not a stable ABI struct (since hvmloader and the toolstack
are a matched pair it doesn't have to be) and we hope to eventually get
rid of it. Possibly hvmloader should scrub it before calling the BIOS to
discourage its use.

> 
> I guess the quoted range would be available from
> "hvm_info.reserved_mem_pgstart"? OvmfPkg/PlatformPei/Xen.c has a comment
> in XenConnect() saying
> 
>   /* TBD: Locate hvm_info and reserve it away. */
>   mXenInfo.HvmInfo = NULL;

Oh dear.

> Is the generic approach "see if you can find all what you need in
> hvm_info, ask for the rest in a dedicated table"? (Out of pure curiosity.)

As I hope is clear from the above, Nope ;-)

Ian.
-- 
Ian Campbell

"Everyone is entitled to an *informed* opinion."
                -- Harlan Ellison


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel