[PATCH v3 03/18] KVM: PPC: Account TCE pages in locked_vm

[Alexey Kardashevskiy]

On 07/28/2014 10:43 AM, Benjamin Herrenschmidt wrote:
> On Thu, 2014-07-24 at 18:47 +1000, Alexey Kardashevskiy wrote:
>> Signed-off-by: Alexey Kardashevskiy <aik at ozlabs.ru>
>> ---
> 
> You need a description.
> 
>>  arch/powerpc/kvm/book3s_64_vio.c | 35 ++++++++++++++++++++++++++++++++++-
>>  1 file changed, 34 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/powerpc/kvm/book3s_64_vio.c b/arch/powerpc/kvm/book3s_64_vio.c
>> index 516f2ee..48b7ed4 100644
>> --- a/arch/powerpc/kvm/book3s_64_vio.c
>> +++ b/arch/powerpc/kvm/book3s_64_vio.c
>> @@ -45,18 +45,48 @@ static long kvmppc_stt_npages(unsigned long window_size)
>>  		     * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
>>  }
>>  
>> +/*
>> + * Checks ulimit in order not to let the user space to pin all
>> + * available memory for TCE tables.
>> + */
>> +static long kvmppc_account_memlimit(long npages)
>> +{
>> +	unsigned long ret = 0, locked, lock_limit;
>> +
>> +	if (!current->mm)
>> +		return -ESRCH; /* process exited */
>> +
>> +	down_write(&current->mm->mmap_sem);
>> +	locked = current->mm->locked_vm + npages;
>> +	lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
>> +	if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
>> +		pr_warn("RLIMIT_MEMLOCK (%ld) exceeded\n",
>> +				rlimit(RLIMIT_MEMLOCK));
>> +		ret = -ENOMEM;
>> +	} else {
>> +		current->mm->locked_vm += npages;
>> +	}
>> +	up_write(&current->mm->mmap_sem);
>> +
>> +	return ret;
>> +}
>> +
>>  static void release_spapr_tce_table(struct kvmppc_spapr_tce_table *stt)
>>  {
>>  	struct kvm *kvm = stt->kvm;
>>  	int i;
>> +	long npages = kvmppc_stt_npages(stt->window_size);
>>  
>>  	mutex_lock(&kvm->lock);
>>  	list_del(&stt->list);
>> -	for (i = 0; i < kvmppc_stt_npages(stt->window_size); i++)
>> +	for (i = 0; i < npages; i++)
>>  		__free_page(stt->pages[i]);
>> +
>>  	kfree(stt);
>>  	mutex_unlock(&kvm->lock);
>>  
>> +	kvmppc_account_memlimit(-(npages + 1));
>> +
>>  	kvm_put_kvm(kvm);
>>  }
>>  
>> @@ -112,6 +142,9 @@ long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
>>  	}
>>  
>>  	npages = kvmppc_stt_npages(args->window_size);
>> +	ret = kvmppc_account_memlimit(npages + 1);
>> +	if (ret)
>> +		goto fail;
> 
> This is called for VFIO only or is it also called when creating TCE
> tables for emulated devices ? Because in the latter case, you don't
> want to account the pages as locked, do you ?

At the moment TCE-containing pages (for emulated TCE) are allocated with
alloc_page() which is kernel memory and therefore always locked, no?


> Also, you need to explain what +1
> 
> Finally, do I correctly deduce that creating 10 TCE tables of 2G
> each will end up accounting 20G as locked even if the guest for
> example only has 4G of RAM ? 


The user is required to set the limit to 20G, correct. But this does not
mean all 20G will be pinned. Ugly but better than nothing. As I remember
from you explanations, even if we give up real/virtual mode handlers for
H_PUT_TCE&Co, we cannot rely of existing counters in page struct in order
to understand whether we need to account a page again or not so we are
stuck with this code till we have a "clone DDW window" API.


But this patch is not about guest pages, it is about pages with TCEs, there
was no counting for this at all.



> 
>>  	stt = kzalloc(sizeof(*stt) + npages * sizeof(struct page *),
>>  		      GFP_KERNEL);
> 
> Ben.
> 
> 


-- 
Alexey