Re: [PATCH v5] vfio/type1: Adopt fast IOTLB flush interface when unmap IOVAs

[Alex Williamson <alex.williamson@redhat.com>]

On Thu,  1 Feb 2018 01:27:38 -0500
Suravee Suthikulpanit <suravee.suthikulpanit@amd.com> wrote:

> VFIO IOMMU type1 currently upmaps IOVA pages synchronously, which requires
> IOTLB flushing for every unmapping. This results in large IOTLB flushing
> overhead when handling pass-through devices has a large number of mapped
> IOVAs. This can be avoided by using the new IOTLB flushing interface.
> 
> Cc: Alex Williamson <alex.williamson@redhat.com>
> Cc: Joerg Roedel <joro@8bytes.org>
> Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
> ---
> 
> Changes from v4 (https://lkml.org/lkml/2018/1/31/153)
>  * Change return type from ssize_t back to size_t since we no longer
>    changing IOMMU API. Also update error handling logic accordingly.
>  * In unmap_unpin_fast(), also sync when failing to allocate entry.
>  * Some code restructuring and variable renaming.
> 
>  drivers/vfio/vfio_iommu_type1.c | 128 ++++++++++++++++++++++++++++++++++++----
>  1 file changed, 117 insertions(+), 11 deletions(-)
> 
> diff --git a/drivers/vfio/vfio_iommu_type1.c b/drivers/vfio/vfio_iommu_type1.c
> index e30e29a..6041530 100644
> --- a/drivers/vfio/vfio_iommu_type1.c
> +++ b/drivers/vfio/vfio_iommu_type1.c
> @@ -102,6 +102,13 @@ struct vfio_pfn {
>  	atomic_t		ref_count;
>  };
>  
> +struct vfio_regions {
> +	struct list_head list;
> +	dma_addr_t iova;
> +	phys_addr_t phys;
> +	size_t len;
> +};
> +
>  #define IS_IOMMU_CAP_DOMAIN_IN_CONTAINER(iommu)	\
>  					(!list_empty(&iommu->domain_list))
>  
> @@ -648,11 +655,102 @@ static int vfio_iommu_type1_unpin_pages(void *iommu_data,
>  	return i > npage ? npage : (i > 0 ? i : -EINVAL);
>  }
>  
> +static long vfio_sync_unpin(struct vfio_dma *dma, struct vfio_domain *domain,
> +				struct list_head *regions)
> +{
> +	long unlocked = 0;
> +	struct vfio_regions *entry, *next;
> +
> +	iommu_tlb_sync(domain->domain);
> +
> +	list_for_each_entry_safe(entry, next, regions, list) {
> +		unlocked += vfio_unpin_pages_remote(dma,
> +						    entry->iova,
> +						    entry->phys >> PAGE_SHIFT,
> +						    entry->len >> PAGE_SHIFT,
> +						    false);
> +		list_del(&entry->list);
> +		kfree(entry);
> +	}
> +
> +	cond_resched();
> +
> +	return unlocked;
> +}
> +
> +/*
> + * Generally, VFIO needs to unpin remote pages after each IOTLB flush.
> + * Therefore, when using IOTLB flush sync interface, VFIO need to keep track
> + * of these regions (currently using a list).
> + *
> + * This value specifies maximum number of regions for each IOTLB flush sync.
> + */
> +#define VFIO_IOMMU_TLB_SYNC_MAX		512
> +
> +static size_t unmap_unpin_fast(struct vfio_domain *domain,
> +			       struct vfio_dma *dma, dma_addr_t *iova,
> +			       size_t len, phys_addr_t phys, long *unlocked,
> +			       struct list_head *unmapped_list,
> +			       int *unmapped_cnt)
> +{
> +	size_t unmapped = 0;
> +	struct vfio_regions *entry = kzalloc(sizeof(*entry), GFP_KERNEL);
> +
> +	if (entry) {
> +		unmapped = iommu_unmap_fast(domain->domain, *iova, len);
> +
> +		if (!unmapped) {
> +			kfree(entry);
> +		} else {
> +			iommu_tlb_range_add(domain->domain, *iova, unmapped);
> +			entry->iova = *iova;
> +			entry->phys = phys;
> +			entry->len  = unmapped;
> +			list_add_tail(&entry->list, unmapped_list);
> +
> +			*iova += unmapped;
> +			(*unmapped_cnt)++;
> +		}
> +	}
> +
> +	/*
> +	 * Sync if the number of fast-unmap regions hits the limit
> +	 * or in case of errors.
> +	 */
> +	if (*unmapped_cnt >= VFIO_IOMMU_TLB_SYNC_MAX || !unmapped) {
> +		*unlocked += vfio_sync_unpin(dma, domain,
> +					     unmapped_list);
> +		*unmapped_cnt = 0;
> +	}
> +
> +	return unmapped;
> +}
> +
> +static size_t unmap_unpin_slow(struct vfio_domain *domain,
> +			       struct vfio_dma *dma, dma_addr_t *iova,
> +			       size_t len, phys_addr_t phys,
> +			       long *unlocked)
> +{
> +	size_t unmapped = iommu_unmap(domain->domain, *iova, len);
> +
> +	if (unmapped) {
> +		*unlocked += vfio_unpin_pages_remote(dma, *iova,
> +						     phys >> PAGE_SHIFT,
> +						     unmapped >> PAGE_SHIFT,
> +						     false);
> +		*iova += unmapped;
> +		cond_resched();
> +	}
> +	return unmapped;
> +}
> +
>  static long vfio_unmap_unpin(struct vfio_iommu *iommu, struct vfio_dma *dma,
>  			     bool do_accounting)
>  {
>  	dma_addr_t iova = dma->iova, end = dma->iova + dma->size;
>  	struct vfio_domain *domain, *d;
> +	struct list_head unmapped_region_list;
> +	int unmapped_region_cnt = 0;
>  	long unlocked = 0;
>  
>  	if (!dma->size)
> @@ -661,6 +759,8 @@ static long vfio_unmap_unpin(struct vfio_iommu *iommu, struct vfio_dma *dma,
>  	if (!IS_IOMMU_CAP_DOMAIN_IN_CONTAINER(iommu))
>  		return 0;
>  
> +	INIT_LIST_HEAD(&unmapped_region_list);

Since I harassed Shameer about using LIST_HEAD() for the iova list
extension, I feel obligated to note that it can also be used here.  If
you approve I'll just remove the above INIT_LIST_HEAD() and declare
unmapped_region_list as LIST_HEAD(unmapped_region_list);, no need to
re-send.  Otherwise looks fine to me.  Thanks,

Alex