caprevoke: Fix up an optimization

Suppose the revoker encounters a non-resident page in a swap-backed object.  We
use swap_pager_cheri_revoke_next() to see if we can simply skip over the page,
and furthermore to see how far we can skip (by finding the smaller of the pindex
of the next resident page and the pindex of the swap block that contains at
least one CHERI tag).  The value returned by swap_pager_cheri_revoke_next() is
the next pindex we should scan.

However, the handling of this return value was wrong.  If the current pindex
corresponds to a swapped-out page, we would end up retrying endlessly.  We
should only return early if the next pindex to scan is not the one that we are
currently being asked to scan.

Fixes:	68928f0 ("caprevoke: Make the skip_fault optimization smarter")

Author: markjdb

sys/vm/vm_cheri_revoke.c

@@ -677,12 +677,13 @@ vm_cheri_revoke_object_at(const struct vm_cheri_revoke_cookie *crc,
 				CHERI_REVOKE_STATS_INC(crst, pages_skip_fast,
 				    (entry->end - addr) >> PAGE_SHIFT);
 				*ooff = lastoff;
-			} else {
+				return (VM_CHERI_REVOKE_AT_OK);
+			} else if (nextpindex != ipi) {
 				CHERI_REVOKE_STATS_INC(crst, pages_skip_fast,
 				    nextpindex - ipi);
 				*ooff = IDX_TO_OFF(nextpindex);
+				return (VM_CHERI_REVOKE_AT_OK);
 			}
-			return (VM_CHERI_REVOKE_AT_OK);
 		}
 
 		CHERI_REVOKE_STATS_BUMP(crst, pages_faulted_ro);