Re: [Xen-devel] [PATCH] xen-blkback: use bigger array for batch gnt operations

["Roger Pau Monné" <roger.pau@citrix.com>]

On 01/08/13 16:18, Roger Pau Monné wrote:
> On 01/08/13 14:30, David Vrabel wrote:
>> On 01/08/13 13:08, Roger Pau Monne wrote:
>>> Right now the maximum number of grant operations that can be batched
>>> in a single request is BLKIF_MAX_SEGMENTS_PER_REQUEST (11). This was
>>> OK before indirect descriptors because the maximum number of segments
>>> in a request was 11, but with the introduction of indirect
>>> descriptors the maximum number of segments in a request has been
>>> increased past 11.
>>>
>>> The memory used by the structures that are passed in the hypercall was
>>> allocated from the stack, but if we have to increase the size of the
>>> array we can no longer use stack memory, so we have to pre-allocate
>>> it.
>>>
>>> This patch increases the maximum size of batch grant operations and
>>> replaces the use of stack memory with pre-allocated memory, that is
>>> reserved when the blkback instance is initialized.
>> [...]
>>> --- a/drivers/block/xen-blkback/xenbus.c
>>> +++ b/drivers/block/xen-blkback/xenbus.c
>> [...]
>>> @@ -148,6 +155,16 @@ static struct xen_blkif *xen_blkif_alloc(domid_t domid)
>>>  			if (!req->indirect_pages[j])
>>>  				goto fail;
>>>  		}
>>> +		req->map = kcalloc(GNT_OPERATIONS_SIZE, sizeof(req->map[0]), GFP_KERNEL);
>>> +		if (!req->map)
>>> +			goto fail;
>>> +		req->unmap = kcalloc(GNT_OPERATIONS_SIZE, sizeof(req->unmap[0]), GFP_KERNEL);
>>> +		if (!req->unmap)
>>> +			goto fail;
>>> +		req->pages_to_gnt = kcalloc(GNT_OPERATIONS_SIZE, sizeof(req->pages_to_gnt[0]),
>>> +		                            GFP_KERNEL);
>>> +		if (!req->pages_to_gnt)
>>> +			goto fail;
>>
>> Do these need to be per-request? Or can they all share a common set of
>> arrays?
> 
> No, we cannot share them unless we serialize the unmap of grants using a
> spinlock (like we do when writing the reponse on the ring).

Any other comments on this one? Should I resubmit it?