Re: [PATCH v2] tools: Mark ACPI SDTs as NVS in the PVH build path

[Jan Beulich <jbeulich@xxxxxxxx>]

On 13.03.2025 15:30, Alejandro Vallejo wrote:
> On Thu Mar 13, 2025 at 1:14 PM GMT, Jan Beulich wrote:
>> On 11.03.2025 10:29, Alejandro Vallejo wrote:
>>> Commit cefeffc7e583 marked ACPI tables as NVS in the hvmloader path
>>> because SeaBIOS may otherwise just mark it as RAM. There is, however,
>>> yet another reason to do it even in the PVH path. Xen's incarnation of
>>> AML relies on having access to some ACPI tables (e.g: _STA of Processor
>>> objects relies on reading the processor online bit in its MADT entry)
>>>
>>> This is problematic if the OS tries to reclaim ACPI memory for page
>>> tables as it's needed for runtime and can't be reclaimed after the OSPM
>>> is up and running.
>>>
>>> Fixes: de6d188a519f("hvmloader: flip "ACPI data" to "ACPI NVS" type for 
>>> ACPI table region)"
>>> Signed-off-by: Alejandro Vallejo <alejandro.vallejo@xxxxxxxxx>
>>> ---
>>> v1->v2:
>>>   * Copy explanatory comment in hvmloader/e820.c to its libxl_x86.c 
>>> counterpart
>>>
>>> ---
>>>  tools/firmware/hvmloader/e820.c |  4 ++++
>>>  tools/libs/light/libxl_x86.c    | 17 ++++++++++++++++-
>>>  2 files changed, 20 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/tools/firmware/hvmloader/e820.c 
>>> b/tools/firmware/hvmloader/e820.c
>>> index c490a0bc790c..86d39544e887 100644
>>> --- a/tools/firmware/hvmloader/e820.c
>>> +++ b/tools/firmware/hvmloader/e820.c
>>> @@ -210,6 +210,10 @@ int build_e820_table(struct e820entry *e820,
>>>       * space reuse by an ACPI unaware / buggy bootloader, option ROM, etc.
>>>       * before an ACPI OS takes control. This is possible due to the fact 
>>> that
>>>       * ACPI NVS memory is explicitly described as non-reclaimable in ACPI 
>>> spec.
>>> +     *
>>> +     * Furthermore, Xen relies on accessing ACPI tables from within the AML
>>> +     * code exposed to guests. So Xen's ACPI tables are not, in general,
>>> +     * reclaimable.
>>>       */
>>>  
>>>      if ( acpi_enabled )
>>> diff --git a/tools/libs/light/libxl_x86.c b/tools/libs/light/libxl_x86.c
>>> index a3164a3077fe..2ba96d12e595 100644
>>> --- a/tools/libs/light/libxl_x86.c
>>> +++ b/tools/libs/light/libxl_x86.c
>>> @@ -737,12 +737,27 @@ static int domain_construct_memmap(libxl__gc *gc,
>>>          nr++;
>>>      }
>>>  
>>> +    /*
>>> +     * Mark populated reserved memory that contains ACPI tables as ACPI 
>>> NVS.
>>> +     * That should help the guest to treat it correctly later: e.g. pass to
>>> +     * the next kernel on kexec.
>>> +     *
>>> +     * Using NVS type instead of a regular one helps to prevent potential
>>> +     * space reuse by an ACPI unaware / buggy bootloader, option ROM, etc.
>>> +     * before an ACPI OS takes control. This is possible due to the fact 
>>> that
>>> +     * ACPI NVS memory is explicitly described as non-reclaimable in ACPI 
>>> spec.
>>> +     *
>>> +     * Furthermore, Xen relies on accessing ACPI tables from within the AML
>>> +     * code exposed to guests. So Xen's ACPI tables are not, in general,
>>> +     * reclaimable.
>>> +     */
>>
>> When asking for a comment here I really only was after what the last 
>> paragraph says.
>> Especially the middle paragraph seems questionable to me: It would not only 
>> be ACPI-
>> unawareness, but also E820-unawareness, for the range to be prematurely 
>> re-used. And
>> buggy bootloaders really would need fixing, I think - they'd put OSes into 
>> trouble on
>> real hardware as well.
>>
>> In short - I'd like to ask that the middle paragraph be dropped from here 
>> (which
>> surely could be done while committing).
> 
> I feel the rationale is the same on both paths, so the comment blocks ought to
> be aligned (whichever way). But I have no strong motivations and would be fine
> dropping the middle paragraph here.
> 
> If that's your only remark, I'm happy for it to be dropped on commit.
> 
>>
>> However, there's a second concern: You say "PVH" in the title, yet this 
>> function is
>> in use also for HVM, and ...
>>
>>>      for (i = 0; i < MAX_ACPI_MODULES; i++) {
>>>          if (dom->acpi_modules[i].length) {
>>>              e820[nr].addr = dom->acpi_modules[i].guest_addr_out & 
>>> ~(page_size - 1);
>>>              e820[nr].size = dom->acpi_modules[i].length +
>>>                  (dom->acpi_modules[i].guest_addr_out & (page_size - 1));
>>> -            e820[nr].type = E820_ACPI;
>>> +            e820[nr].type = E820_NVS;
>>>              nr++;
>>>          }
>>>      }
>>
>> ... this code is outside of any conditionals. This imo needs sorting one way 
>> or
>> another.
> 
> ACPI tables are populated by hvmloader, while libxl generates those of PVH.
> 
> dom->acpi_modules are populated by libxl__dom_load_acpi(), which is gated on
> the type being PVH (see the caller of this function). So this loop should be
> effectively skipped.
> 
> I called it the PVH path because it happens to be at the moment. Nothing
> prevents this path from being the HVM path too, but that involves rewiring
> hvmloader.

Oh, okay - what I was missing then is that ->acpi_modules[] is populated by
libxl__dom_load_acpi(), and that bails early for non-PVH. So the loop here
will have only no-op iterations for HVM.

Jan