[PATCH v9 13/27] gunyah: vm_mgr: Add/remove user memory regions

[Elliot Berman]

On 1/25/2023 5:34 AM, Srivatsa Vaddagiri wrote:
> * Elliot Berman <quic_eberman at quicinc.com> [2023-01-20 14:46:12]:
> 
>> +	/* Check for overlap */
>> +	list_for_each_entry(tmp_mapping, &ghvm->memory_mappings, list) {
>> +		if (!((mapping->guest_phys_addr + (mapping->npages << PAGE_SHIFT) <=
>> +			tmp_mapping->guest_phys_addr) ||
>> +			(mapping->guest_phys_addr >=
>> +			tmp_mapping->guest_phys_addr + (tmp_mapping->npages << PAGE_SHIFT)))) {
>> +			ret = -EEXIST;
>> +			goto unlock;
>> +		}
>> +	}
>> +
>> +	list_add(&mapping->list, &ghvm->memory_mappings);
> 
> I think the potential race condition described last time is still possible. Pls
> check.
> 
>> +unlock:
>> +	mutex_unlock(&ghvm->mm_lock);
>> +	if (ret)
>> +		goto free_mapping;
>> +
>> +	mapping->pages = kcalloc(mapping->npages, sizeof(*mapping->pages), GFP_KERNEL);
>> +	if (!mapping->pages) {
>> +		ret = -ENOMEM;
>> +		goto reclaim;
> 
> Same comment as last time. Can you check this error path? We seem to call
> unpin_user_page() in this path, which is not correct.
> 
>> +	}
>> +
>> +	pinned = pin_user_pages_fast(region->userspace_addr, mapping->npages,
>> +					FOLL_WRITE | FOLL_LONGTERM, mapping->pages);
>> +	if (pinned < 0) {
>> +		ret = pinned;
>> +		goto reclaim;
> 
> Same comment as above.
> 

Caught the race condition you are describing between adding and removing 
the memory parcels. Should be fixed for v10.

- Elliot