vfio/ccw: don't group contiguous pages on 2K IDAWs

The vfio_pin_pages() interface allows contiguous pages to be
pinned as a single request, which is great for the 4K pages
that are normally processed. Old IDA formats operate on 2K
chunks, which makes this logic more difficult.

Since these formats are rare, let's just invoke the page
pinning one-at-a-time, instead of trying to group them.
We can rework this code at a later date if needed.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
Reviewed-by: Matthew Rosato <mjrosato@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>

1 files changed, 20 insertions, 10 deletions

diff --git a/drivers/s390/cio/vfio_ccw_cp.c b/drivers/s390/cio/vfio_ccw_cp.c
index a7415d440a81..1500058dbe1f 100644
--- a/drivers/s390/cio/vfio_ccw_cp.c
+++ b/drivers/s390/cio/vfio_ccw_cp.c
@@ -83,12 +83,13 @@ static int page_array_alloc(struct page_array *pa, unsigned int len)
  * @pa: page_array on which to perform the operation
  * @vdev: the vfio device to perform the operation
  * @pa_nr: number of user pages to unpin
+ * @unaligned: were pages unaligned on the pin request
  *
  * Only unpin if any pages were pinned to begin with, i.e. pa_nr > 0,
  * otherwise only clear pa->pa_nr
  */
 static void page_array_unpin(struct page_array *pa,
-			     struct vfio_device *vdev, int pa_nr)
+			     struct vfio_device *vdev, int pa_nr, bool unaligned)
 {
 	int unpinned = 0, npage = 1;
 
@@ -97,7 +98,8 @@ static void page_array_unpin(struct page_array *pa,
 		dma_addr_t *last = &first[npage];
 
 		if (unpinned + npage < pa_nr &&
-		    *first + npage * PAGE_SIZE == *last) {
+		    *first + npage * PAGE_SIZE == *last &&
+		    !unaligned) {
 			npage++;
 			continue;
 		}
@@ -114,12 +116,19 @@ static void page_array_unpin(struct page_array *pa,
  * page_array_pin() - Pin user pages in memory
  * @pa: page_array on which to perform the operation
  * @vdev: the vfio device to perform pin operations
+ * @unaligned: are pages aligned to 4K boundary?
  *
  * Returns number of pages pinned upon success.
  * If the pin request partially succeeds, or fails completely,
  * all pages are left unpinned and a negative error value is returned.
+ *
+ * Requests to pin "aligned" pages can be coalesced into a single
+ * vfio_pin_pages request for the sake of efficiency, based on the
+ * expectation of 4K page requests. Unaligned requests are probably
+ * dealing with 2K "pages", and cannot be coalesced without
+ * reworking this logic to incorporate that math.
  */
-static int page_array_pin(struct page_array *pa, struct vfio_device *vdev)
+static int page_array_pin(struct page_array *pa, struct vfio_device *vdev, bool unaligned)
 {
 	int pinned = 0, npage = 1;
 	int ret = 0;
@@ -129,7 +138,8 @@ static int page_array_pin(struct page_array *pa, struct vfio_device *vdev)
 		dma_addr_t *last = &first[npage];
 
 		if (pinned + npage < pa->pa_nr &&
-		    *first + npage * PAGE_SIZE == *last) {
+		    *first + npage * PAGE_SIZE == *last &&
+		    !unaligned) {
 			npage++;
 			continue;
 		}
@@ -151,14 +161,14 @@ static int page_array_pin(struct page_array *pa, struct vfio_device *vdev)
 	return ret;
 
 err_out:
-	page_array_unpin(pa, vdev, pinned);
+	page_array_unpin(pa, vdev, pinned, unaligned);
 	return ret;
 }
 
 /* Unpin the pages before releasing the memory. */
-static void page_array_unpin_free(struct page_array *pa, struct vfio_device *vdev)
+static void page_array_unpin_free(struct page_array *pa, struct vfio_device *vdev, bool unaligned)
 {
-	page_array_unpin(pa, vdev, pa->pa_nr);
+	page_array_unpin(pa, vdev, pa->pa_nr, unaligned);
 	kfree(pa->pa_page);
 	kfree(pa->pa_iova);
 }
@@ -643,7 +653,7 @@ static int ccwchain_fetch_ccw(struct ccw1 *ccw,
 	}
 
 	if (ccw_does_data_transfer(ccw)) {
-		ret = page_array_pin(pa, vdev);
+		ret = page_array_pin(pa, vdev, idal_is_2k(cp));
 		if (ret < 0)
 			goto out_unpin;
 	} else {
@@ -659,7 +669,7 @@ static int ccwchain_fetch_ccw(struct ccw1 *ccw,
 	return 0;
 
 out_unpin:
-	page_array_unpin_free(pa, vdev);
+	page_array_unpin_free(pa, vdev, idal_is_2k(cp));
 out_free_idaws:
 	kfree(idaws);
 out_init:
@@ -757,7 +767,7 @@ void cp_free(struct channel_program *cp)
 	cp->initialized = false;
 	list_for_each_entry_safe(chain, temp, &cp->ccwchain_list, next) {
 		for (i = 0; i < chain->ch_len; i++) {
-			page_array_unpin_free(&chain->ch_pa[i], vdev);
+			page_array_unpin_free(&chain->ch_pa[i], vdev, idal_is_2k(cp));
 			ccwchain_cda_free(chain, i);
 		}
 		ccwchain_free(chain);