Re: [PATCH v1] KVM: s390: Fall back to short-term pinning in MAP ioctl

[Christian Borntraeger <borntraeger@linux.ibm.com>]

Am 22.07.26 um 17:51 schrieb Matthew Rosato:
> On 7/20/26 6:22 PM, Jaehoon Kim wrote:
>> FOLL_LONGTERM pinning fails for some memory types, such as file-backed
>> guest memory. As a result, kvm_s390_adapter_map() returns -EINVAL and
>> irqfd adapter registration fails even though interrupt delivery could
>> still work via the existing non-atomic path.
>>
>> When FOLL_LONGTERM pinning fails, verify that the page is accessible
>> using a short-term pin instead. If the short-term pin succeeds,
>> immediately unpin the page and return success. The non-atomic irqfd
>> path already performs short-term pinning for interrupt delivery, so
>> this restores the previous behavior for memory that cannot be pinned
>> long-term.
>>
>> Fixes: adcd5b3e758b ("KVM: s390: Add map/unmap ioctl and clean mappings post-guest")
>> Signed-off-by: Jaehoon Kim <jhkim@linux.ibm.com>
>> ---
>>   arch/s390/kvm/interrupt.c | 15 +++++++++++++--
>>   1 file changed, 13 insertions(+), 2 deletions(-)
>>
>> diff --git a/arch/s390/kvm/interrupt.c b/arch/s390/kvm/interrupt.c
>> index 9e3e6b0d72ad..4b68e0cb51f6 100644
>> --- a/arch/s390/kvm/interrupt.c
>> +++ b/arch/s390/kvm/interrupt.c
>> @@ -2520,8 +2520,19 @@ static int kvm_s390_adapter_map(struct kvm *kvm, unsigned int id, __u64 addr)
>>   	map->addr = host_addr;
>>   	map->page = pin_map_page(kvm, host_addr, FOLL_LONGTERM);
>>   	if (!map->page) {
>> -		ret = -EINVAL;
>> -		goto out;
>> +		/*
>> +		 * Long-term pinning may fail for memory types such as file-backed
>> +		 * memory. Check whether short-term pinning is possible so that the
>> +		 * non-atomic irqfd path can handle interrupt injection.
>> +		 */
>> +		map->page = pin_map_page(kvm, host_addr, 0);
>> +		if (!map->page) {
>> +			ret = -EINVAL;
>> +			goto out;
>> +		}
>> +		unpin_user_page(map->page);
>> +		kfree(map);
>> +		return 0;
> 
> This makes sense, but what happens later on the kvm_s390_adapter_unmap?
> 
> Won't we return -ENOENT because nothing was in the list?
> 
> Is it worth detecting/recognizing this case?  I could think of 2 ways...
> 
> 1) re-doing the above test during unmap
> 
> 2) stashing an entry in adapter->maps but with a new flag that indicates
> it was approved for short-term pinning only (e.g. return NULL
> immediately if a match is found during get_map_info() with this flag
> set), and during kvm_s390_adapter_unmap() and
> kvm_s390_unmap_all_adapters() just remove it from the list without doing
> any unpin/page dirtying.
> 
> #2 might be a little more code but sounds cleaner?
> 

When we are at it, can we also fix  Documentation/virt/kvm/devices/s390_flic.rst ?

To me it looks like qemu does not check the error for unmapping, but yes, maybe 2 is
the right thing to do.