Re: [Xen-devel] [Patch V2 2/2] xen: before ballooning hotplugged memory, set frames to invalid

[David Vrabel <david.vrabel@citrix.com>]

On 20/03/15 13:44, Boris Ostrovsky wrote:
> On 03/20/2015 08:55 AM, Juergen Gross wrote:
>> Commit 25b884a83d487fd62c3de7ac1ab5549979188482 ("x86/xen: set
>> regions above the end of RAM as 1:1") introduced a regression.
>>
>> To be able to add memory pages which were added via memory hotplug to
>> a pv domain, the pages must be "invalid" instead of "identity" in the
>> p2m list before they can be added.
>>
>> Suggested-by: David Vrabel <david.vrabel@citrix.com>
>> Signed-off-by: Juergen Gross <jgross@suse.com>
>> ---
>>   drivers/xen/balloon.c | 13 +++++++++++++
>>   1 file changed, 13 insertions(+)
>>
>> diff --git a/drivers/xen/balloon.c b/drivers/xen/balloon.c
>> index 0b52d92..65fedb8 100644
>> --- a/drivers/xen/balloon.c
>> +++ b/drivers/xen/balloon.c
>> @@ -229,6 +229,19 @@ static enum bp_state
>> reserve_additional_memory(long credit)
>>       balloon_hotplug = round_up(balloon_hotplug, PAGES_PER_SECTION);
>>       nid = memory_add_physaddr_to_nid(hotplug_start_paddr);
>>   +#ifdef CONFIG_XEN_HAVE_PVMMU

I've added this comment:

        /*
         * add_memory() will build page tables for the new memory so
         * the p2m must contain invalidate entries so the correct
         * non-present PTEs will be written.
         *
         * If a failure occurs, the original (identity) p2m entries
         * are not restored since this region is now known not to
         * conflict with any devices.
         */

>> +    if (!xen_feature(XENFEAT_auto_translated_physmap)) {
>> +        unsigned long pfn, i;
>> +
>> +        pfn = PFN_DOWN(hotplug_start_paddr);
>> +        for (i = 0; i < balloon_hotplug; i++)
>> +            if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
>> +                pr_warn("set_phys_to_machine() failed, no memory
>> added\n");
>> +                return BP_ECANCELED;
> 
> I should have asked last time --- should we restore [0..i-1] mapping
> back to identity on error here? I wonder whether leaving those frames as
> invalid (when they are expected to be 'identity') could cause problems
> in the future.

See new comment above.

David