Re: [Xen-devel] [PATCH for-4.10] libxc: load acpi RSDP table at correct address

[Juergen Gross <jgross@xxxxxxxx>]

On 20/11/17 11:57, Andrew Cooper wrote:
> On 20/11/17 10:43, Juergen Gross wrote:
>> On 20/11/17 11:21, Andrew Cooper wrote:
>>> On 20/11/17 10:04, Juergen Gross wrote:
>>>> On 20/11/17 10:58, Andrew Cooper wrote:
>>>>> On 20/11/2017 09:55, Juergen Gross wrote:
>>>>>> On 20/11/17 10:51, Roger Pau Monné wrote:
>>>>>>> Adding xen-devel, dropped it on my reply. 
>>>>>>>
>>>>>>>     Replying from my phone, sorry for the formatting. 
>>>>>>>
>>>>>>>
>>>>>>>     El 20 nov. 2017 9:35, "Juergen Gross" <jgross@xxxxxxxx
>>>>>>>     <mailto:jgross@xxxxxxxx>> escribió:
>>>>>>>
>>>>>>>         For PVH domains loading of the ACPI RSDP table is done via
>>>>>>>         allocating
>>>>>>>         a domain loader segment after having loaded the kernel. This
>>>>>>>         leads to
>>>>>>>         the RSDP table being loaded at an arbitrary guest address 
>>>>>>> instead of
>>>>>>>         the architectural correct address just below 1MB. 
>>>>>>>
>>>>>>>
>>>>>>>     AFAIK this is only true for legacy BIOS boot, when using UEFI the
>>>>>>>     RSDP can be anywhere in memory, hence grub2 must already have an
>>>>>>>     alternative way of finding the RSDP apart from scanning the low 1MB.
>>>>>> The problem isn't grub2, but the loaded linux kernel. Without this
>>>>>> patch Linux won't find the RSDP when booted in a PVH domain via grub2.
>>>>>>
>>>>>> I could modify grub2 even further to move the RSDP to the correct
>>>>>> address, but I think doing it correctly on Xen side is the better
>>>>>> option.
>>>>> Why?  The PVH info block contains a pointer directly to the RSDP, and
>>>>> Linux should be following this rather than scanning for it using the
>>>>> legacy method.
>>>> Oh no, please not this discussion again.
>>>>
>>>> We already had a very long discussion how to do PVH support in grub2,
>>>> and the outcome was to try to use the standard boot entry of the kernel
>>>> instead the PVH sepcific one.
>>>>
>>>> The Linux kernel right now doesn't make use of the RSDP pointer in the
>>>> PVH info block, so I think we shouldn't change this when using grub2.
>>> I clearly missed the previous discussion, and I don't advocate using yet
>>> another PVH-specific entry point, but how does Linux cope in other
>>> non-BIOS environments?  Does it genuinely rely exclusively on the legacy
>>> mechanism?
>> Looking at the code I think so, yes. Maybe there are cases where no RSDP
>> is needed, but in the grub2/PVH case we need it to distinguish PVH from
>> HVM.
> 
> In which case, being a Linux limitation, I think it is wrong to
> unilaterally apply this restriction to all other PVH guests.

Which restriction? I'm loading the RSDP table to its architectural
correct addres if possible, otherwise it will be loaded to the same
address as without my patch. So I'm not adding a restriction, but
removing one.

> Doing this in grub seems like the more appropriate place IMO.

I don't think so.


Juergen

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