Re: [Xen-devel] [PATCH v1 10/20] acpi/hvmloader: Replace mem_alloc() and virt_to_phys() with memory ops

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 08.07.16 at 17:23, <boris.ostrovsky@xxxxxxxxxx> wrote:
> On 07/08/2016 09:58 AM, Jan Beulich wrote:
>>>>> On 05.07.16 at 21:05, <boris.ostrovsky@xxxxxxxxxx> wrote:
>>> Components that wish to use ACPI builder will need to provide their own
>>> mem_alloc() and virt_to_phys() routines. Pointers to these routines will
>>> be passed to the builder as memory ops.
>>>
>>> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>
>>> ---
>>>
>>> Changes in v1:
>>> * Keep memory ops seprate from acpi_config, in struct acpi_context.
>>>
>>> Jan requested adding a free() op to struct acpi_mem_ops. I couldn't see who 
>>> might want to
>>> use those. The builder uses (should use) mem_alloc() to allocate memory for 
>>> tables, not as a
>>> general-purpose allocator.
>> In addition to what I said back then, did you think of error cleanup
>> paths here? Not all errors mean the guest has to die.
> 
> If there is an error and the builder decides to free up memory needed
> for a table, how do we communicate to the caller which table has been
> failed?

I don't understand - that's what the proposed free hook would be for.

> Is it up to the builder to decide which tables are important and which
> are not?

I'm afraid that's not so easy to tell. If for example we can't fit the
HPET table, the guest could be run without HPET unless a HPET
was specifically requested (rather than just defaulted to).

Jan


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